]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/cam/ctl/ctl.c
MFC r310524: Improve length handling when writing sense data.
[FreeBSD/stable/10.git] / sys / cam / ctl / ctl.c
1 /*-
2  * Copyright (c) 2003-2009 Silicon Graphics International Corp.
3  * Copyright (c) 2012 The FreeBSD Foundation
4  * Copyright (c) 2015 Alexander Motin <mav@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Portions of this software were developed by Edward Tomasz Napierala
8  * under sponsorship from the FreeBSD Foundation.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions, and the following disclaimer,
15  *    without modification.
16  * 2. Redistributions in binary form must reproduce at minimum a disclaimer
17  *    substantially similar to the "NO WARRANTY" disclaimer below
18  *    ("Disclaimer") and any redistribution must be conditioned upon
19  *    including a substantially similar Disclaimer requirement for further
20  *    binary redistribution.
21  *
22  * NO WARRANTY
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTIBILITY AND FITNESS FOR
26  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
27  * HOLDERS OR CONTRIBUTORS BE LIABLE FOR SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
31  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
32  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
33  * POSSIBILITY OF SUCH DAMAGES.
34  *
35  * $Id$
36  */
37 /*
38  * CAM Target Layer, a SCSI device emulation subsystem.
39  *
40  * Author: Ken Merry <ken@FreeBSD.org>
41  */
42
43 #define _CTL_C
44
45 #include <sys/cdefs.h>
46 __FBSDID("$FreeBSD$");
47
48 #include <sys/param.h>
49 #include <sys/systm.h>
50 #include <sys/ctype.h>
51 #include <sys/kernel.h>
52 #include <sys/types.h>
53 #include <sys/kthread.h>
54 #include <sys/bio.h>
55 #include <sys/fcntl.h>
56 #include <sys/lock.h>
57 #include <sys/module.h>
58 #include <sys/mutex.h>
59 #include <sys/condvar.h>
60 #include <sys/malloc.h>
61 #include <sys/conf.h>
62 #include <sys/ioccom.h>
63 #include <sys/queue.h>
64 #include <sys/sbuf.h>
65 #include <sys/smp.h>
66 #include <sys/endian.h>
67 #include <sys/sysctl.h>
68 #include <vm/uma.h>
69
70 #include <cam/cam.h>
71 #include <cam/scsi/scsi_all.h>
72 #include <cam/scsi/scsi_cd.h>
73 #include <cam/scsi/scsi_da.h>
74 #include <cam/ctl/ctl_io.h>
75 #include <cam/ctl/ctl.h>
76 #include <cam/ctl/ctl_frontend.h>
77 #include <cam/ctl/ctl_util.h>
78 #include <cam/ctl/ctl_backend.h>
79 #include <cam/ctl/ctl_ioctl.h>
80 #include <cam/ctl/ctl_ha.h>
81 #include <cam/ctl/ctl_private.h>
82 #include <cam/ctl/ctl_debug.h>
83 #include <cam/ctl/ctl_scsi_all.h>
84 #include <cam/ctl/ctl_error.h>
85
86 struct ctl_softc *control_softc = NULL;
87
88 /*
89  * Template mode pages.
90  */
91
92 /*
93  * Note that these are default values only.  The actual values will be
94  * filled in when the user does a mode sense.
95  */
96 const static struct scsi_da_rw_recovery_page rw_er_page_default = {
97         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
98         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
99         /*byte3*/SMS_RWER_AWRE|SMS_RWER_ARRE,
100         /*read_retry_count*/0,
101         /*correction_span*/0,
102         /*head_offset_count*/0,
103         /*data_strobe_offset_cnt*/0,
104         /*byte8*/SMS_RWER_LBPERE,
105         /*write_retry_count*/0,
106         /*reserved2*/0,
107         /*recovery_time_limit*/{0, 0},
108 };
109
110 const static struct scsi_da_rw_recovery_page rw_er_page_changeable = {
111         /*page_code*/SMS_RW_ERROR_RECOVERY_PAGE,
112         /*page_length*/sizeof(struct scsi_da_rw_recovery_page) - 2,
113         /*byte3*/SMS_RWER_PER,
114         /*read_retry_count*/0,
115         /*correction_span*/0,
116         /*head_offset_count*/0,
117         /*data_strobe_offset_cnt*/0,
118         /*byte8*/SMS_RWER_LBPERE,
119         /*write_retry_count*/0,
120         /*reserved2*/0,
121         /*recovery_time_limit*/{0, 0},
122 };
123
124 const static struct scsi_format_page format_page_default = {
125         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
126         /*page_length*/sizeof(struct scsi_format_page) - 2,
127         /*tracks_per_zone*/ {0, 0},
128         /*alt_sectors_per_zone*/ {0, 0},
129         /*alt_tracks_per_zone*/ {0, 0},
130         /*alt_tracks_per_lun*/ {0, 0},
131         /*sectors_per_track*/ {(CTL_DEFAULT_SECTORS_PER_TRACK >> 8) & 0xff,
132                                 CTL_DEFAULT_SECTORS_PER_TRACK & 0xff},
133         /*bytes_per_sector*/ {0, 0},
134         /*interleave*/ {0, 0},
135         /*track_skew*/ {0, 0},
136         /*cylinder_skew*/ {0, 0},
137         /*flags*/ SFP_HSEC,
138         /*reserved*/ {0, 0, 0}
139 };
140
141 const static struct scsi_format_page format_page_changeable = {
142         /*page_code*/SMS_FORMAT_DEVICE_PAGE,
143         /*page_length*/sizeof(struct scsi_format_page) - 2,
144         /*tracks_per_zone*/ {0, 0},
145         /*alt_sectors_per_zone*/ {0, 0},
146         /*alt_tracks_per_zone*/ {0, 0},
147         /*alt_tracks_per_lun*/ {0, 0},
148         /*sectors_per_track*/ {0, 0},
149         /*bytes_per_sector*/ {0, 0},
150         /*interleave*/ {0, 0},
151         /*track_skew*/ {0, 0},
152         /*cylinder_skew*/ {0, 0},
153         /*flags*/ 0,
154         /*reserved*/ {0, 0, 0}
155 };
156
157 const static struct scsi_rigid_disk_page rigid_disk_page_default = {
158         /*page_code*/SMS_RIGID_DISK_PAGE,
159         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
160         /*cylinders*/ {0, 0, 0},
161         /*heads*/ CTL_DEFAULT_HEADS,
162         /*start_write_precomp*/ {0, 0, 0},
163         /*start_reduced_current*/ {0, 0, 0},
164         /*step_rate*/ {0, 0},
165         /*landing_zone_cylinder*/ {0, 0, 0},
166         /*rpl*/ SRDP_RPL_DISABLED,
167         /*rotational_offset*/ 0,
168         /*reserved1*/ 0,
169         /*rotation_rate*/ {(CTL_DEFAULT_ROTATION_RATE >> 8) & 0xff,
170                            CTL_DEFAULT_ROTATION_RATE & 0xff},
171         /*reserved2*/ {0, 0}
172 };
173
174 const static struct scsi_rigid_disk_page rigid_disk_page_changeable = {
175         /*page_code*/SMS_RIGID_DISK_PAGE,
176         /*page_length*/sizeof(struct scsi_rigid_disk_page) - 2,
177         /*cylinders*/ {0, 0, 0},
178         /*heads*/ 0,
179         /*start_write_precomp*/ {0, 0, 0},
180         /*start_reduced_current*/ {0, 0, 0},
181         /*step_rate*/ {0, 0},
182         /*landing_zone_cylinder*/ {0, 0, 0},
183         /*rpl*/ 0,
184         /*rotational_offset*/ 0,
185         /*reserved1*/ 0,
186         /*rotation_rate*/ {0, 0},
187         /*reserved2*/ {0, 0}
188 };
189
190 const static struct scsi_da_verify_recovery_page verify_er_page_default = {
191         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
192         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
193         /*byte3*/0,
194         /*read_retry_count*/0,
195         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
196         /*recovery_time_limit*/{0, 0},
197 };
198
199 const static struct scsi_da_verify_recovery_page verify_er_page_changeable = {
200         /*page_code*/SMS_VERIFY_ERROR_RECOVERY_PAGE,
201         /*page_length*/sizeof(struct scsi_da_verify_recovery_page) - 2,
202         /*byte3*/SMS_VER_PER,
203         /*read_retry_count*/0,
204         /*reserved*/{ 0, 0, 0, 0, 0, 0 },
205         /*recovery_time_limit*/{0, 0},
206 };
207
208 const static struct scsi_caching_page caching_page_default = {
209         /*page_code*/SMS_CACHING_PAGE,
210         /*page_length*/sizeof(struct scsi_caching_page) - 2,
211         /*flags1*/ SCP_DISC | SCP_WCE,
212         /*ret_priority*/ 0,
213         /*disable_pf_transfer_len*/ {0xff, 0xff},
214         /*min_prefetch*/ {0, 0},
215         /*max_prefetch*/ {0xff, 0xff},
216         /*max_pf_ceiling*/ {0xff, 0xff},
217         /*flags2*/ 0,
218         /*cache_segments*/ 0,
219         /*cache_seg_size*/ {0, 0},
220         /*reserved*/ 0,
221         /*non_cache_seg_size*/ {0, 0, 0}
222 };
223
224 const static struct scsi_caching_page caching_page_changeable = {
225         /*page_code*/SMS_CACHING_PAGE,
226         /*page_length*/sizeof(struct scsi_caching_page) - 2,
227         /*flags1*/ SCP_WCE | SCP_RCD,
228         /*ret_priority*/ 0,
229         /*disable_pf_transfer_len*/ {0, 0},
230         /*min_prefetch*/ {0, 0},
231         /*max_prefetch*/ {0, 0},
232         /*max_pf_ceiling*/ {0, 0},
233         /*flags2*/ 0,
234         /*cache_segments*/ 0,
235         /*cache_seg_size*/ {0, 0},
236         /*reserved*/ 0,
237         /*non_cache_seg_size*/ {0, 0, 0}
238 };
239
240 const static struct scsi_control_page control_page_default = {
241         /*page_code*/SMS_CONTROL_MODE_PAGE,
242         /*page_length*/sizeof(struct scsi_control_page) - 2,
243         /*rlec*/0,
244         /*queue_flags*/SCP_QUEUE_ALG_RESTRICTED,
245         /*eca_and_aen*/0,
246         /*flags4*/SCP_TAS,
247         /*aen_holdoff_period*/{0, 0},
248         /*busy_timeout_period*/{0, 0},
249         /*extended_selftest_completion_time*/{0, 0}
250 };
251
252 const static struct scsi_control_page control_page_changeable = {
253         /*page_code*/SMS_CONTROL_MODE_PAGE,
254         /*page_length*/sizeof(struct scsi_control_page) - 2,
255         /*rlec*/SCP_DSENSE,
256         /*queue_flags*/SCP_QUEUE_ALG_MASK | SCP_NUAR,
257         /*eca_and_aen*/SCP_SWP,
258         /*flags4*/0,
259         /*aen_holdoff_period*/{0, 0},
260         /*busy_timeout_period*/{0, 0},
261         /*extended_selftest_completion_time*/{0, 0}
262 };
263
264 #define CTL_CEM_LEN     (sizeof(struct scsi_control_ext_page) - 4)
265
266 const static struct scsi_control_ext_page control_ext_page_default = {
267         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
268         /*subpage_code*/0x01,
269         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
270         /*flags*/0,
271         /*prio*/0,
272         /*max_sense*/0
273 };
274
275 const static struct scsi_control_ext_page control_ext_page_changeable = {
276         /*page_code*/SMS_CONTROL_MODE_PAGE | SMPH_SPF,
277         /*subpage_code*/0x01,
278         /*page_length*/{CTL_CEM_LEN >> 8, CTL_CEM_LEN},
279         /*flags*/0,
280         /*prio*/0,
281         /*max_sense*/0xff
282 };
283
284 const static struct scsi_info_exceptions_page ie_page_default = {
285         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
286         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
287         /*info_flags*/SIEP_FLAGS_EWASC,
288         /*mrie*/SIEP_MRIE_NO,
289         /*interval_timer*/{0, 0, 0, 0},
290         /*report_count*/{0, 0, 0, 1}
291 };
292
293 const static struct scsi_info_exceptions_page ie_page_changeable = {
294         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE,
295         /*page_length*/sizeof(struct scsi_info_exceptions_page) - 2,
296         /*info_flags*/SIEP_FLAGS_EWASC | SIEP_FLAGS_DEXCPT | SIEP_FLAGS_TEST |
297             SIEP_FLAGS_LOGERR,
298         /*mrie*/0x0f,
299         /*interval_timer*/{0xff, 0xff, 0xff, 0xff},
300         /*report_count*/{0xff, 0xff, 0xff, 0xff}
301 };
302
303 #define CTL_LBPM_LEN    (sizeof(struct ctl_logical_block_provisioning_page) - 4)
304
305 const static struct ctl_logical_block_provisioning_page lbp_page_default = {{
306         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
307         /*subpage_code*/0x02,
308         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
309         /*flags*/0,
310         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
311         /*descr*/{}},
312         {{/*flags*/0,
313           /*resource*/0x01,
314           /*reserved*/{0, 0},
315           /*count*/{0, 0, 0, 0}},
316          {/*flags*/0,
317           /*resource*/0x02,
318           /*reserved*/{0, 0},
319           /*count*/{0, 0, 0, 0}},
320          {/*flags*/0,
321           /*resource*/0xf1,
322           /*reserved*/{0, 0},
323           /*count*/{0, 0, 0, 0}},
324          {/*flags*/0,
325           /*resource*/0xf2,
326           /*reserved*/{0, 0},
327           /*count*/{0, 0, 0, 0}}
328         }
329 };
330
331 const static struct ctl_logical_block_provisioning_page lbp_page_changeable = {{
332         /*page_code*/SMS_INFO_EXCEPTIONS_PAGE | SMPH_SPF,
333         /*subpage_code*/0x02,
334         /*page_length*/{CTL_LBPM_LEN >> 8, CTL_LBPM_LEN},
335         /*flags*/SLBPP_SITUA,
336         /*reserved*/{0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0},
337         /*descr*/{}},
338         {{/*flags*/0,
339           /*resource*/0,
340           /*reserved*/{0, 0},
341           /*count*/{0, 0, 0, 0}},
342          {/*flags*/0,
343           /*resource*/0,
344           /*reserved*/{0, 0},
345           /*count*/{0, 0, 0, 0}},
346          {/*flags*/0,
347           /*resource*/0,
348           /*reserved*/{0, 0},
349           /*count*/{0, 0, 0, 0}},
350          {/*flags*/0,
351           /*resource*/0,
352           /*reserved*/{0, 0},
353           /*count*/{0, 0, 0, 0}}
354         }
355 };
356
357 const static struct scsi_cddvd_capabilities_page cddvd_page_default = {
358         /*page_code*/SMS_CDDVD_CAPS_PAGE,
359         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
360         /*caps1*/0x3f,
361         /*caps2*/0x00,
362         /*caps3*/0xf0,
363         /*caps4*/0x00,
364         /*caps5*/0x29,
365         /*caps6*/0x00,
366         /*obsolete*/{0, 0},
367         /*nvol_levels*/{0, 0},
368         /*buffer_size*/{8, 0},
369         /*obsolete2*/{0, 0},
370         /*reserved*/0,
371         /*digital*/0,
372         /*obsolete3*/0,
373         /*copy_management*/0,
374         /*reserved2*/0,
375         /*rotation_control*/0,
376         /*cur_write_speed*/0,
377         /*num_speed_descr*/0,
378 };
379
380 const static struct scsi_cddvd_capabilities_page cddvd_page_changeable = {
381         /*page_code*/SMS_CDDVD_CAPS_PAGE,
382         /*page_length*/sizeof(struct scsi_cddvd_capabilities_page) - 2,
383         /*caps1*/0,
384         /*caps2*/0,
385         /*caps3*/0,
386         /*caps4*/0,
387         /*caps5*/0,
388         /*caps6*/0,
389         /*obsolete*/{0, 0},
390         /*nvol_levels*/{0, 0},
391         /*buffer_size*/{0, 0},
392         /*obsolete2*/{0, 0},
393         /*reserved*/0,
394         /*digital*/0,
395         /*obsolete3*/0,
396         /*copy_management*/0,
397         /*reserved2*/0,
398         /*rotation_control*/0,
399         /*cur_write_speed*/0,
400         /*num_speed_descr*/0,
401 };
402
403 SYSCTL_NODE(_kern_cam, OID_AUTO, ctl, CTLFLAG_RD, 0, "CAM Target Layer");
404 static int worker_threads = -1;
405 TUNABLE_INT("kern.cam.ctl.worker_threads", &worker_threads);
406 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, worker_threads, CTLFLAG_RDTUN,
407     &worker_threads, 1, "Number of worker threads");
408 static int ctl_debug = CTL_DEBUG_NONE;
409 TUNABLE_INT("kern.cam.ctl.debug", &ctl_debug);
410 SYSCTL_INT(_kern_cam_ctl, OID_AUTO, debug, CTLFLAG_RWTUN,
411     &ctl_debug, 0, "Enabled debug flags");
412
413 /*
414  * Supported pages (0x00), Serial number (0x80), Device ID (0x83),
415  * Extended INQUIRY Data (0x86), Mode Page Policy (0x87),
416  * SCSI Ports (0x88), Third-party Copy (0x8F), Block limits (0xB0),
417  * Block Device Characteristics (0xB1) and Logical Block Provisioning (0xB2)
418  */
419 #define SCSI_EVPD_NUM_SUPPORTED_PAGES   10
420
421 static void ctl_isc_event_handler(ctl_ha_channel chanel, ctl_ha_event event,
422                                   int param);
423 static void ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest);
424 static void ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest);
425 static int ctl_init(void);
426 void ctl_shutdown(void);
427 static int ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td);
428 static int ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td);
429 static int ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio);
430 static void ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
431                               struct ctl_ooa *ooa_hdr,
432                               struct ctl_ooa_entry *kern_entries);
433 static int ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
434                      struct thread *td);
435 static int ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *lun,
436                          struct ctl_be_lun *be_lun);
437 static int ctl_free_lun(struct ctl_lun *lun);
438 static void ctl_create_lun(struct ctl_be_lun *be_lun);
439 static struct ctl_port * ctl_io_port(struct ctl_io_hdr *io_hdr);
440
441 static int ctl_do_mode_select(union ctl_io *io);
442 static int ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun,
443                            uint64_t res_key, uint64_t sa_res_key,
444                            uint8_t type, uint32_t residx,
445                            struct ctl_scsiio *ctsio,
446                            struct scsi_per_res_out *cdb,
447                            struct scsi_per_res_out_parms* param);
448 static void ctl_pro_preempt_other(struct ctl_lun *lun,
449                                   union ctl_ha_msg *msg);
450 static void ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg);
451 static int ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len);
452 static int ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len);
453 static int ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len);
454 static int ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len);
455 static int ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len);
456 static int ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio,
457                                          int alloc_len);
458 static int ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio,
459                                          int alloc_len);
460 static int ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len);
461 static int ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len);
462 static int ctl_inquiry_evpd(struct ctl_scsiio *ctsio);
463 static int ctl_inquiry_std(struct ctl_scsiio *ctsio);
464 static int ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len);
465 static ctl_action ctl_extent_check(union ctl_io *io1, union ctl_io *io2,
466     bool seq);
467 static ctl_action ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2);
468 static ctl_action ctl_check_for_blockage(struct ctl_lun *lun,
469     union ctl_io *pending_io, union ctl_io *ooa_io);
470 static ctl_action ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
471                                 union ctl_io *starting_io);
472 static int ctl_check_blocked(struct ctl_lun *lun);
473 static int ctl_scsiio_lun_check(struct ctl_lun *lun,
474                                 const struct ctl_cmd_entry *entry,
475                                 struct ctl_scsiio *ctsio);
476 static void ctl_failover_lun(union ctl_io *io);
477 static int ctl_scsiio_precheck(struct ctl_softc *ctl_softc,
478                                struct ctl_scsiio *ctsio);
479 static int ctl_scsiio(struct ctl_scsiio *ctsio);
480
481 static int ctl_bus_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
482 static int ctl_target_reset(struct ctl_softc *ctl_softc, union ctl_io *io,
483                             ctl_ua_type ua_type);
484 static int ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io,
485                          ctl_ua_type ua_type);
486 static int ctl_lun_reset(struct ctl_softc *ctl_softc, union ctl_io *io);
487 static int ctl_abort_task(union ctl_io *io);
488 static int ctl_abort_task_set(union ctl_io *io);
489 static int ctl_query_task(union ctl_io *io, int task_set);
490 static int ctl_i_t_nexus_reset(union ctl_io *io);
491 static int ctl_query_async_event(union ctl_io *io);
492 static void ctl_run_task(union ctl_io *io);
493 #ifdef CTL_IO_DELAY
494 static void ctl_datamove_timer_wakeup(void *arg);
495 static void ctl_done_timer_wakeup(void *arg);
496 #endif /* CTL_IO_DELAY */
497
498 static void ctl_send_datamove_done(union ctl_io *io, int have_lock);
499 static void ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq);
500 static int ctl_datamove_remote_dm_write_cb(union ctl_io *io);
501 static void ctl_datamove_remote_write(union ctl_io *io);
502 static int ctl_datamove_remote_dm_read_cb(union ctl_io *io);
503 static void ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq);
504 static int ctl_datamove_remote_sgl_setup(union ctl_io *io);
505 static int ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
506                                     ctl_ha_dt_cb callback);
507 static void ctl_datamove_remote_read(union ctl_io *io);
508 static void ctl_datamove_remote(union ctl_io *io);
509 static void ctl_process_done(union ctl_io *io);
510 static void ctl_lun_thread(void *arg);
511 static void ctl_thresh_thread(void *arg);
512 static void ctl_work_thread(void *arg);
513 static void ctl_enqueue_incoming(union ctl_io *io);
514 static void ctl_enqueue_rtr(union ctl_io *io);
515 static void ctl_enqueue_done(union ctl_io *io);
516 static void ctl_enqueue_isc(union ctl_io *io);
517 static const struct ctl_cmd_entry *
518     ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa);
519 static const struct ctl_cmd_entry *
520     ctl_validate_command(struct ctl_scsiio *ctsio);
521 static int ctl_cmd_applicable(uint8_t lun_type,
522     const struct ctl_cmd_entry *entry);
523
524 static uint64_t ctl_get_prkey(struct ctl_lun *lun, uint32_t residx);
525 static void ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx);
526 static void ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx);
527 static void ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key);
528
529 /*
530  * Load the serialization table.  This isn't very pretty, but is probably
531  * the easiest way to do it.
532  */
533 #include "ctl_ser_table.c"
534
535 /*
536  * We only need to define open, close and ioctl routines for this driver.
537  */
538 static struct cdevsw ctl_cdevsw = {
539         .d_version =    D_VERSION,
540         .d_flags =      0,
541         .d_open =       ctl_open,
542         .d_close =      ctl_close,
543         .d_ioctl =      ctl_ioctl,
544         .d_name =       "ctl",
545 };
546
547
548 MALLOC_DEFINE(M_CTL, "ctlmem", "Memory used for CTL");
549
550 static int ctl_module_event_handler(module_t, int /*modeventtype_t*/, void *);
551
552 static moduledata_t ctl_moduledata = {
553         "ctl",
554         ctl_module_event_handler,
555         NULL
556 };
557
558 DECLARE_MODULE(ctl, ctl_moduledata, SI_SUB_CONFIGURE, SI_ORDER_THIRD);
559 MODULE_VERSION(ctl, 1);
560
561 static struct ctl_frontend ha_frontend =
562 {
563         .name = "ha",
564 };
565
566 static void
567 ctl_ha_datamove(union ctl_io *io)
568 {
569         struct ctl_lun *lun;
570         struct ctl_sg_entry *sgl;
571         union ctl_ha_msg msg;
572         uint32_t sg_entries_sent;
573         int do_sg_copy, i, j;
574
575         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
576         memset(&msg.dt, 0, sizeof(msg.dt));
577         msg.hdr.msg_type = CTL_MSG_DATAMOVE;
578         msg.hdr.original_sc = io->io_hdr.original_sc;
579         msg.hdr.serializing_sc = io;
580         msg.hdr.nexus = io->io_hdr.nexus;
581         msg.hdr.status = io->io_hdr.status;
582         msg.dt.flags = io->io_hdr.flags;
583
584         /*
585          * We convert everything into a S/G list here.  We can't
586          * pass by reference, only by value between controllers.
587          * So we can't pass a pointer to the S/G list, only as many
588          * S/G entries as we can fit in here.  If it's possible for
589          * us to get more than CTL_HA_MAX_SG_ENTRIES S/G entries,
590          * then we need to break this up into multiple transfers.
591          */
592         if (io->scsiio.kern_sg_entries == 0) {
593                 msg.dt.kern_sg_entries = 1;
594 #if 0
595                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
596                         msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
597                 } else {
598                         /* XXX KDM use busdma here! */
599                         msg.dt.sg_list[0].addr =
600                             (void *)vtophys(io->scsiio.kern_data_ptr);
601                 }
602 #else
603                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
604                     ("HA does not support BUS_ADDR"));
605                 msg.dt.sg_list[0].addr = io->scsiio.kern_data_ptr;
606 #endif
607                 msg.dt.sg_list[0].len = io->scsiio.kern_data_len;
608                 do_sg_copy = 0;
609         } else {
610                 msg.dt.kern_sg_entries = io->scsiio.kern_sg_entries;
611                 do_sg_copy = 1;
612         }
613
614         msg.dt.kern_data_len = io->scsiio.kern_data_len;
615         msg.dt.kern_total_len = io->scsiio.kern_total_len;
616         msg.dt.kern_data_resid = io->scsiio.kern_data_resid;
617         msg.dt.kern_rel_offset = io->scsiio.kern_rel_offset;
618         msg.dt.sg_sequence = 0;
619
620         /*
621          * Loop until we've sent all of the S/G entries.  On the
622          * other end, we'll recompose these S/G entries into one
623          * contiguous list before processing.
624          */
625         for (sg_entries_sent = 0; sg_entries_sent < msg.dt.kern_sg_entries;
626             msg.dt.sg_sequence++) {
627                 msg.dt.cur_sg_entries = MIN((sizeof(msg.dt.sg_list) /
628                     sizeof(msg.dt.sg_list[0])),
629                     msg.dt.kern_sg_entries - sg_entries_sent);
630                 if (do_sg_copy != 0) {
631                         sgl = (struct ctl_sg_entry *)io->scsiio.kern_data_ptr;
632                         for (i = sg_entries_sent, j = 0;
633                              i < msg.dt.cur_sg_entries; i++, j++) {
634 #if 0
635                                 if (io->io_hdr.flags & CTL_FLAG_BUS_ADDR) {
636                                         msg.dt.sg_list[j].addr = sgl[i].addr;
637                                 } else {
638                                         /* XXX KDM use busdma here! */
639                                         msg.dt.sg_list[j].addr =
640                                             (void *)vtophys(sgl[i].addr);
641                                 }
642 #else
643                                 KASSERT((io->io_hdr.flags &
644                                     CTL_FLAG_BUS_ADDR) == 0,
645                                     ("HA does not support BUS_ADDR"));
646                                 msg.dt.sg_list[j].addr = sgl[i].addr;
647 #endif
648                                 msg.dt.sg_list[j].len = sgl[i].len;
649                         }
650                 }
651
652                 sg_entries_sent += msg.dt.cur_sg_entries;
653                 msg.dt.sg_last = (sg_entries_sent >= msg.dt.kern_sg_entries);
654                 if (ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
655                     sizeof(msg.dt) - sizeof(msg.dt.sg_list) +
656                     sizeof(struct ctl_sg_entry) * msg.dt.cur_sg_entries,
657                     M_WAITOK) > CTL_HA_STATUS_SUCCESS) {
658                         io->io_hdr.port_status = 31341;
659                         io->scsiio.be_move_done(io);
660                         return;
661                 }
662                 msg.dt.sent_sg_entries = sg_entries_sent;
663         }
664
665         /*
666          * Officially handover the request from us to peer.
667          * If failover has just happened, then we must return error.
668          * If failover happen just after, then it is not our problem.
669          */
670         if (lun)
671                 mtx_lock(&lun->lun_lock);
672         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
673                 if (lun)
674                         mtx_unlock(&lun->lun_lock);
675                 io->io_hdr.port_status = 31342;
676                 io->scsiio.be_move_done(io);
677                 return;
678         }
679         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
680         io->io_hdr.flags |= CTL_FLAG_DMA_INPROG;
681         if (lun)
682                 mtx_unlock(&lun->lun_lock);
683 }
684
685 static void
686 ctl_ha_done(union ctl_io *io)
687 {
688         union ctl_ha_msg msg;
689
690         if (io->io_hdr.io_type == CTL_IO_SCSI) {
691                 memset(&msg, 0, sizeof(msg));
692                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
693                 msg.hdr.original_sc = io->io_hdr.original_sc;
694                 msg.hdr.nexus = io->io_hdr.nexus;
695                 msg.hdr.status = io->io_hdr.status;
696                 msg.scsi.scsi_status = io->scsiio.scsi_status;
697                 msg.scsi.tag_num = io->scsiio.tag_num;
698                 msg.scsi.tag_type = io->scsiio.tag_type;
699                 msg.scsi.sense_len = io->scsiio.sense_len;
700                 msg.scsi.sense_residual = io->scsiio.sense_residual;
701                 msg.scsi.residual = io->scsiio.residual;
702                 memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
703                     io->scsiio.sense_len);
704                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
705                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
706                     msg.scsi.sense_len, M_WAITOK);
707         }
708         ctl_free_io(io);
709 }
710
711 static void
712 ctl_isc_handler_finish_xfer(struct ctl_softc *ctl_softc,
713                             union ctl_ha_msg *msg_info)
714 {
715         struct ctl_scsiio *ctsio;
716
717         if (msg_info->hdr.original_sc == NULL) {
718                 printf("%s: original_sc == NULL!\n", __func__);
719                 /* XXX KDM now what? */
720                 return;
721         }
722
723         ctsio = &msg_info->hdr.original_sc->scsiio;
724         ctsio->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
725         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
726         ctsio->io_hdr.status = msg_info->hdr.status;
727         ctsio->scsi_status = msg_info->scsi.scsi_status;
728         ctsio->sense_len = msg_info->scsi.sense_len;
729         ctsio->sense_residual = msg_info->scsi.sense_residual;
730         ctsio->residual = msg_info->scsi.residual;
731         memcpy(&ctsio->sense_data, &msg_info->scsi.sense_data,
732                msg_info->scsi.sense_len);
733         ctl_enqueue_isc((union ctl_io *)ctsio);
734 }
735
736 static void
737 ctl_isc_handler_finish_ser_only(struct ctl_softc *ctl_softc,
738                                 union ctl_ha_msg *msg_info)
739 {
740         struct ctl_scsiio *ctsio;
741
742         if (msg_info->hdr.serializing_sc == NULL) {
743                 printf("%s: serializing_sc == NULL!\n", __func__);
744                 /* XXX KDM now what? */
745                 return;
746         }
747
748         ctsio = &msg_info->hdr.serializing_sc->scsiio;
749         ctsio->io_hdr.msg_type = CTL_MSG_FINISH_IO;
750         ctl_enqueue_isc((union ctl_io *)ctsio);
751 }
752
753 void
754 ctl_isc_announce_lun(struct ctl_lun *lun)
755 {
756         struct ctl_softc *softc = lun->ctl_softc;
757         union ctl_ha_msg *msg;
758         struct ctl_ha_msg_lun_pr_key pr_key;
759         int i, k;
760
761         if (softc->ha_link != CTL_HA_LINK_ONLINE)
762                 return;
763         mtx_lock(&lun->lun_lock);
764         i = sizeof(msg->lun);
765         if (lun->lun_devid)
766                 i += lun->lun_devid->len;
767         i += sizeof(pr_key) * lun->pr_key_count;
768 alloc:
769         mtx_unlock(&lun->lun_lock);
770         msg = malloc(i, M_CTL, M_WAITOK);
771         mtx_lock(&lun->lun_lock);
772         k = sizeof(msg->lun);
773         if (lun->lun_devid)
774                 k += lun->lun_devid->len;
775         k += sizeof(pr_key) * lun->pr_key_count;
776         if (i < k) {
777                 free(msg, M_CTL);
778                 i = k;
779                 goto alloc;
780         }
781         bzero(&msg->lun, sizeof(msg->lun));
782         msg->hdr.msg_type = CTL_MSG_LUN_SYNC;
783         msg->hdr.nexus.targ_lun = lun->lun;
784         msg->hdr.nexus.targ_mapped_lun = lun->lun;
785         msg->lun.flags = lun->flags;
786         msg->lun.pr_generation = lun->pr_generation;
787         msg->lun.pr_res_idx = lun->pr_res_idx;
788         msg->lun.pr_res_type = lun->pr_res_type;
789         msg->lun.pr_key_count = lun->pr_key_count;
790         i = 0;
791         if (lun->lun_devid) {
792                 msg->lun.lun_devid_len = lun->lun_devid->len;
793                 memcpy(&msg->lun.data[i], lun->lun_devid->data,
794                     msg->lun.lun_devid_len);
795                 i += msg->lun.lun_devid_len;
796         }
797         for (k = 0; k < CTL_MAX_INITIATORS; k++) {
798                 if ((pr_key.pr_key = ctl_get_prkey(lun, k)) == 0)
799                         continue;
800                 pr_key.pr_iid = k;
801                 memcpy(&msg->lun.data[i], &pr_key, sizeof(pr_key));
802                 i += sizeof(pr_key);
803         }
804         mtx_unlock(&lun->lun_lock);
805         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
806             M_WAITOK);
807         free(msg, M_CTL);
808
809         if (lun->flags & CTL_LUN_PRIMARY_SC) {
810                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
811                         ctl_isc_announce_mode(lun, -1,
812                             lun->mode_pages.index[i].page_code & SMPH_PC_MASK,
813                             lun->mode_pages.index[i].subpage);
814                 }
815         }
816 }
817
818 void
819 ctl_isc_announce_port(struct ctl_port *port)
820 {
821         struct ctl_softc *softc = port->ctl_softc;
822         union ctl_ha_msg *msg;
823         int i;
824
825         if (port->targ_port < softc->port_min ||
826             port->targ_port >= softc->port_max ||
827             softc->ha_link != CTL_HA_LINK_ONLINE)
828                 return;
829         i = sizeof(msg->port) + strlen(port->port_name) + 1;
830         if (port->lun_map)
831                 i += sizeof(uint32_t) * CTL_MAX_LUNS;
832         if (port->port_devid)
833                 i += port->port_devid->len;
834         if (port->target_devid)
835                 i += port->target_devid->len;
836         if (port->init_devid)
837                 i += port->init_devid->len;
838         msg = malloc(i, M_CTL, M_WAITOK);
839         bzero(&msg->port, sizeof(msg->port));
840         msg->hdr.msg_type = CTL_MSG_PORT_SYNC;
841         msg->hdr.nexus.targ_port = port->targ_port;
842         msg->port.port_type = port->port_type;
843         msg->port.physical_port = port->physical_port;
844         msg->port.virtual_port = port->virtual_port;
845         msg->port.status = port->status;
846         i = 0;
847         msg->port.name_len = sprintf(&msg->port.data[i],
848             "%d:%s", softc->ha_id, port->port_name) + 1;
849         i += msg->port.name_len;
850         if (port->lun_map) {
851                 msg->port.lun_map_len = sizeof(uint32_t) * CTL_MAX_LUNS;
852                 memcpy(&msg->port.data[i], port->lun_map,
853                     msg->port.lun_map_len);
854                 i += msg->port.lun_map_len;
855         }
856         if (port->port_devid) {
857                 msg->port.port_devid_len = port->port_devid->len;
858                 memcpy(&msg->port.data[i], port->port_devid->data,
859                     msg->port.port_devid_len);
860                 i += msg->port.port_devid_len;
861         }
862         if (port->target_devid) {
863                 msg->port.target_devid_len = port->target_devid->len;
864                 memcpy(&msg->port.data[i], port->target_devid->data,
865                     msg->port.target_devid_len);
866                 i += msg->port.target_devid_len;
867         }
868         if (port->init_devid) {
869                 msg->port.init_devid_len = port->init_devid->len;
870                 memcpy(&msg->port.data[i], port->init_devid->data,
871                     msg->port.init_devid_len);
872                 i += msg->port.init_devid_len;
873         }
874         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->port, sizeof(msg->port) + i,
875             M_WAITOK);
876         free(msg, M_CTL);
877 }
878
879 void
880 ctl_isc_announce_iid(struct ctl_port *port, int iid)
881 {
882         struct ctl_softc *softc = port->ctl_softc;
883         union ctl_ha_msg *msg;
884         int i, l;
885
886         if (port->targ_port < softc->port_min ||
887             port->targ_port >= softc->port_max ||
888             softc->ha_link != CTL_HA_LINK_ONLINE)
889                 return;
890         mtx_lock(&softc->ctl_lock);
891         i = sizeof(msg->iid);
892         l = 0;
893         if (port->wwpn_iid[iid].name)
894                 l = strlen(port->wwpn_iid[iid].name) + 1;
895         i += l;
896         msg = malloc(i, M_CTL, M_NOWAIT);
897         if (msg == NULL) {
898                 mtx_unlock(&softc->ctl_lock);
899                 return;
900         }
901         bzero(&msg->iid, sizeof(msg->iid));
902         msg->hdr.msg_type = CTL_MSG_IID_SYNC;
903         msg->hdr.nexus.targ_port = port->targ_port;
904         msg->hdr.nexus.initid = iid;
905         msg->iid.in_use = port->wwpn_iid[iid].in_use;
906         msg->iid.name_len = l;
907         msg->iid.wwpn = port->wwpn_iid[iid].wwpn;
908         if (port->wwpn_iid[iid].name)
909                 strlcpy(msg->iid.data, port->wwpn_iid[iid].name, l);
910         mtx_unlock(&softc->ctl_lock);
911         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg->iid, i, M_NOWAIT);
912         free(msg, M_CTL);
913 }
914
915 void
916 ctl_isc_announce_mode(struct ctl_lun *lun, uint32_t initidx,
917     uint8_t page, uint8_t subpage)
918 {
919         struct ctl_softc *softc = lun->ctl_softc;
920         union ctl_ha_msg msg;
921         u_int i;
922
923         if (softc->ha_link != CTL_HA_LINK_ONLINE)
924                 return;
925         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
926                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
927                     page && lun->mode_pages.index[i].subpage == subpage)
928                         break;
929         }
930         if (i == CTL_NUM_MODE_PAGES)
931                 return;
932
933         /* Don't try to replicate pages not present on this device. */
934         if (lun->mode_pages.index[i].page_data == NULL)
935                 return;
936
937         bzero(&msg.mode, sizeof(msg.mode));
938         msg.hdr.msg_type = CTL_MSG_MODE_SYNC;
939         msg.hdr.nexus.targ_port = initidx / CTL_MAX_INIT_PER_PORT;
940         msg.hdr.nexus.initid = initidx % CTL_MAX_INIT_PER_PORT;
941         msg.hdr.nexus.targ_lun = lun->lun;
942         msg.hdr.nexus.targ_mapped_lun = lun->lun;
943         msg.mode.page_code = page;
944         msg.mode.subpage = subpage;
945         msg.mode.page_len = lun->mode_pages.index[i].page_len;
946         memcpy(msg.mode.data, lun->mode_pages.index[i].page_data,
947             msg.mode.page_len);
948         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.mode, sizeof(msg.mode),
949             M_WAITOK);
950 }
951
952 static void
953 ctl_isc_ha_link_up(struct ctl_softc *softc)
954 {
955         struct ctl_port *port;
956         struct ctl_lun *lun;
957         union ctl_ha_msg msg;
958         int i;
959
960         /* Announce this node parameters to peer for validation. */
961         msg.login.msg_type = CTL_MSG_LOGIN;
962         msg.login.version = CTL_HA_VERSION;
963         msg.login.ha_mode = softc->ha_mode;
964         msg.login.ha_id = softc->ha_id;
965         msg.login.max_luns = CTL_MAX_LUNS;
966         msg.login.max_ports = CTL_MAX_PORTS;
967         msg.login.max_init_per_port = CTL_MAX_INIT_PER_PORT;
968         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg.login, sizeof(msg.login),
969             M_WAITOK);
970
971         STAILQ_FOREACH(port, &softc->port_list, links) {
972                 ctl_isc_announce_port(port);
973                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
974                         if (port->wwpn_iid[i].in_use)
975                                 ctl_isc_announce_iid(port, i);
976                 }
977         }
978         STAILQ_FOREACH(lun, &softc->lun_list, links)
979                 ctl_isc_announce_lun(lun);
980 }
981
982 static void
983 ctl_isc_ha_link_down(struct ctl_softc *softc)
984 {
985         struct ctl_port *port;
986         struct ctl_lun *lun;
987         union ctl_io *io;
988         int i;
989
990         mtx_lock(&softc->ctl_lock);
991         STAILQ_FOREACH(lun, &softc->lun_list, links) {
992                 mtx_lock(&lun->lun_lock);
993                 if (lun->flags & CTL_LUN_PEER_SC_PRIMARY) {
994                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
995                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
996                 }
997                 mtx_unlock(&lun->lun_lock);
998
999                 mtx_unlock(&softc->ctl_lock);
1000                 io = ctl_alloc_io(softc->othersc_pool);
1001                 mtx_lock(&softc->ctl_lock);
1002                 ctl_zero_io(io);
1003                 io->io_hdr.msg_type = CTL_MSG_FAILOVER;
1004                 io->io_hdr.nexus.targ_mapped_lun = lun->lun;
1005                 ctl_enqueue_isc(io);
1006         }
1007
1008         STAILQ_FOREACH(port, &softc->port_list, links) {
1009                 if (port->targ_port >= softc->port_min &&
1010                     port->targ_port < softc->port_max)
1011                         continue;
1012                 port->status &= ~CTL_PORT_STATUS_ONLINE;
1013                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1014                         port->wwpn_iid[i].in_use = 0;
1015                         free(port->wwpn_iid[i].name, M_CTL);
1016                         port->wwpn_iid[i].name = NULL;
1017                 }
1018         }
1019         mtx_unlock(&softc->ctl_lock);
1020 }
1021
1022 static void
1023 ctl_isc_ua(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1024 {
1025         struct ctl_lun *lun;
1026         uint32_t iid = ctl_get_initindex(&msg->hdr.nexus);
1027
1028         mtx_lock(&softc->ctl_lock);
1029         if (msg->hdr.nexus.targ_lun < CTL_MAX_LUNS &&
1030             (lun = softc->ctl_luns[msg->hdr.nexus.targ_mapped_lun]) != NULL) {
1031                 mtx_lock(&lun->lun_lock);
1032                 mtx_unlock(&softc->ctl_lock);
1033                 if (msg->ua.ua_type == CTL_UA_THIN_PROV_THRES &&
1034                     msg->ua.ua_set)
1035                         memcpy(lun->ua_tpt_info, msg->ua.ua_info, 8);
1036                 if (msg->ua.ua_all) {
1037                         if (msg->ua.ua_set)
1038                                 ctl_est_ua_all(lun, iid, msg->ua.ua_type);
1039                         else
1040                                 ctl_clr_ua_all(lun, iid, msg->ua.ua_type);
1041                 } else {
1042                         if (msg->ua.ua_set)
1043                                 ctl_est_ua(lun, iid, msg->ua.ua_type);
1044                         else
1045                                 ctl_clr_ua(lun, iid, msg->ua.ua_type);
1046                 }
1047                 mtx_unlock(&lun->lun_lock);
1048         } else
1049                 mtx_unlock(&softc->ctl_lock);
1050 }
1051
1052 static void
1053 ctl_isc_lun_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1054 {
1055         struct ctl_lun *lun;
1056         struct ctl_ha_msg_lun_pr_key pr_key;
1057         int i, k;
1058         ctl_lun_flags oflags;
1059         uint32_t targ_lun;
1060
1061         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1062         mtx_lock(&softc->ctl_lock);
1063         if ((targ_lun >= CTL_MAX_LUNS) ||
1064             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1065                 mtx_unlock(&softc->ctl_lock);
1066                 return;
1067         }
1068         mtx_lock(&lun->lun_lock);
1069         mtx_unlock(&softc->ctl_lock);
1070         if (lun->flags & CTL_LUN_DISABLED) {
1071                 mtx_unlock(&lun->lun_lock);
1072                 return;
1073         }
1074         i = (lun->lun_devid != NULL) ? lun->lun_devid->len : 0;
1075         if (msg->lun.lun_devid_len != i || (i > 0 &&
1076             memcmp(&msg->lun.data[0], lun->lun_devid->data, i) != 0)) {
1077                 mtx_unlock(&lun->lun_lock);
1078                 printf("%s: Received conflicting HA LUN %d\n",
1079                     __func__, msg->hdr.nexus.targ_lun);
1080                 return;
1081         } else {
1082                 /* Record whether peer is primary. */
1083                 oflags = lun->flags;
1084                 if ((msg->lun.flags & CTL_LUN_PRIMARY_SC) &&
1085                     (msg->lun.flags & CTL_LUN_DISABLED) == 0)
1086                         lun->flags |= CTL_LUN_PEER_SC_PRIMARY;
1087                 else
1088                         lun->flags &= ~CTL_LUN_PEER_SC_PRIMARY;
1089                 if (oflags != lun->flags)
1090                         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
1091
1092                 /* If peer is primary and we are not -- use data */
1093                 if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
1094                     (lun->flags & CTL_LUN_PEER_SC_PRIMARY)) {
1095                         lun->pr_generation = msg->lun.pr_generation;
1096                         lun->pr_res_idx = msg->lun.pr_res_idx;
1097                         lun->pr_res_type = msg->lun.pr_res_type;
1098                         lun->pr_key_count = msg->lun.pr_key_count;
1099                         for (k = 0; k < CTL_MAX_INITIATORS; k++)
1100                                 ctl_clr_prkey(lun, k);
1101                         for (k = 0; k < msg->lun.pr_key_count; k++) {
1102                                 memcpy(&pr_key, &msg->lun.data[i],
1103                                     sizeof(pr_key));
1104                                 ctl_alloc_prkey(lun, pr_key.pr_iid);
1105                                 ctl_set_prkey(lun, pr_key.pr_iid,
1106                                     pr_key.pr_key);
1107                                 i += sizeof(pr_key);
1108                         }
1109                 }
1110
1111                 mtx_unlock(&lun->lun_lock);
1112                 CTL_DEBUG_PRINT(("%s: Known LUN %d, peer is %s\n",
1113                     __func__, msg->hdr.nexus.targ_lun,
1114                     (msg->lun.flags & CTL_LUN_PRIMARY_SC) ?
1115                     "primary" : "secondary"));
1116
1117                 /* If we are primary but peer doesn't know -- notify */
1118                 if ((lun->flags & CTL_LUN_PRIMARY_SC) &&
1119                     (msg->lun.flags & CTL_LUN_PEER_SC_PRIMARY) == 0)
1120                         ctl_isc_announce_lun(lun);
1121         }
1122 }
1123
1124 static void
1125 ctl_isc_port_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1126 {
1127         struct ctl_port *port;
1128         struct ctl_lun *lun;
1129         int i, new;
1130
1131         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1132         if (port == NULL) {
1133                 CTL_DEBUG_PRINT(("%s: New port %d\n", __func__,
1134                     msg->hdr.nexus.targ_port));
1135                 new = 1;
1136                 port = malloc(sizeof(*port), M_CTL, M_WAITOK | M_ZERO);
1137                 port->frontend = &ha_frontend;
1138                 port->targ_port = msg->hdr.nexus.targ_port;
1139                 port->fe_datamove = ctl_ha_datamove;
1140                 port->fe_done = ctl_ha_done;
1141         } else if (port->frontend == &ha_frontend) {
1142                 CTL_DEBUG_PRINT(("%s: Updated port %d\n", __func__,
1143                     msg->hdr.nexus.targ_port));
1144                 new = 0;
1145         } else {
1146                 printf("%s: Received conflicting HA port %d\n",
1147                     __func__, msg->hdr.nexus.targ_port);
1148                 return;
1149         }
1150         port->port_type = msg->port.port_type;
1151         port->physical_port = msg->port.physical_port;
1152         port->virtual_port = msg->port.virtual_port;
1153         port->status = msg->port.status;
1154         i = 0;
1155         free(port->port_name, M_CTL);
1156         port->port_name = strndup(&msg->port.data[i], msg->port.name_len,
1157             M_CTL);
1158         i += msg->port.name_len;
1159         if (msg->port.lun_map_len != 0) {
1160                 if (port->lun_map == NULL)
1161                         port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
1162                             M_CTL, M_WAITOK);
1163                 memcpy(port->lun_map, &msg->port.data[i],
1164                     sizeof(uint32_t) * CTL_MAX_LUNS);
1165                 i += msg->port.lun_map_len;
1166         } else {
1167                 free(port->lun_map, M_CTL);
1168                 port->lun_map = NULL;
1169         }
1170         if (msg->port.port_devid_len != 0) {
1171                 if (port->port_devid == NULL ||
1172                     port->port_devid->len != msg->port.port_devid_len) {
1173                         free(port->port_devid, M_CTL);
1174                         port->port_devid = malloc(sizeof(struct ctl_devid) +
1175                             msg->port.port_devid_len, M_CTL, M_WAITOK);
1176                 }
1177                 memcpy(port->port_devid->data, &msg->port.data[i],
1178                     msg->port.port_devid_len);
1179                 port->port_devid->len = msg->port.port_devid_len;
1180                 i += msg->port.port_devid_len;
1181         } else {
1182                 free(port->port_devid, M_CTL);
1183                 port->port_devid = NULL;
1184         }
1185         if (msg->port.target_devid_len != 0) {
1186                 if (port->target_devid == NULL ||
1187                     port->target_devid->len != msg->port.target_devid_len) {
1188                         free(port->target_devid, M_CTL);
1189                         port->target_devid = malloc(sizeof(struct ctl_devid) +
1190                             msg->port.target_devid_len, M_CTL, M_WAITOK);
1191                 }
1192                 memcpy(port->target_devid->data, &msg->port.data[i],
1193                     msg->port.target_devid_len);
1194                 port->target_devid->len = msg->port.target_devid_len;
1195                 i += msg->port.target_devid_len;
1196         } else {
1197                 free(port->target_devid, M_CTL);
1198                 port->target_devid = NULL;
1199         }
1200         if (msg->port.init_devid_len != 0) {
1201                 if (port->init_devid == NULL ||
1202                     port->init_devid->len != msg->port.init_devid_len) {
1203                         free(port->init_devid, M_CTL);
1204                         port->init_devid = malloc(sizeof(struct ctl_devid) +
1205                             msg->port.init_devid_len, M_CTL, M_WAITOK);
1206                 }
1207                 memcpy(port->init_devid->data, &msg->port.data[i],
1208                     msg->port.init_devid_len);
1209                 port->init_devid->len = msg->port.init_devid_len;
1210                 i += msg->port.init_devid_len;
1211         } else {
1212                 free(port->init_devid, M_CTL);
1213                 port->init_devid = NULL;
1214         }
1215         if (new) {
1216                 if (ctl_port_register(port) != 0) {
1217                         printf("%s: ctl_port_register() failed with error\n",
1218                             __func__);
1219                 }
1220         }
1221         mtx_lock(&softc->ctl_lock);
1222         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1223                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
1224                         continue;
1225                 mtx_lock(&lun->lun_lock);
1226                 ctl_est_ua_all(lun, -1, CTL_UA_INQ_CHANGE);
1227                 mtx_unlock(&lun->lun_lock);
1228         }
1229         mtx_unlock(&softc->ctl_lock);
1230 }
1231
1232 static void
1233 ctl_isc_iid_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1234 {
1235         struct ctl_port *port;
1236         int iid;
1237
1238         port = softc->ctl_ports[msg->hdr.nexus.targ_port];
1239         if (port == NULL) {
1240                 printf("%s: Received IID for unknown port %d\n",
1241                     __func__, msg->hdr.nexus.targ_port);
1242                 return;
1243         }
1244         iid = msg->hdr.nexus.initid;
1245         port->wwpn_iid[iid].in_use = msg->iid.in_use;
1246         port->wwpn_iid[iid].wwpn = msg->iid.wwpn;
1247         free(port->wwpn_iid[iid].name, M_CTL);
1248         if (msg->iid.name_len) {
1249                 port->wwpn_iid[iid].name = strndup(&msg->iid.data[0],
1250                     msg->iid.name_len, M_CTL);
1251         } else
1252                 port->wwpn_iid[iid].name = NULL;
1253 }
1254
1255 static void
1256 ctl_isc_login(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1257 {
1258
1259         if (msg->login.version != CTL_HA_VERSION) {
1260                 printf("CTL HA peers have different versions %d != %d\n",
1261                     msg->login.version, CTL_HA_VERSION);
1262                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1263                 return;
1264         }
1265         if (msg->login.ha_mode != softc->ha_mode) {
1266                 printf("CTL HA peers have different ha_mode %d != %d\n",
1267                     msg->login.ha_mode, softc->ha_mode);
1268                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1269                 return;
1270         }
1271         if (msg->login.ha_id == softc->ha_id) {
1272                 printf("CTL HA peers have same ha_id %d\n", msg->login.ha_id);
1273                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1274                 return;
1275         }
1276         if (msg->login.max_luns != CTL_MAX_LUNS ||
1277             msg->login.max_ports != CTL_MAX_PORTS ||
1278             msg->login.max_init_per_port != CTL_MAX_INIT_PER_PORT) {
1279                 printf("CTL HA peers have different limits\n");
1280                 ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1281                 return;
1282         }
1283 }
1284
1285 static void
1286 ctl_isc_mode_sync(struct ctl_softc *softc, union ctl_ha_msg *msg, int len)
1287 {
1288         struct ctl_lun *lun;
1289         u_int i;
1290         uint32_t initidx, targ_lun;
1291
1292         targ_lun = msg->hdr.nexus.targ_mapped_lun;
1293         mtx_lock(&softc->ctl_lock);
1294         if ((targ_lun >= CTL_MAX_LUNS) ||
1295             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
1296                 mtx_unlock(&softc->ctl_lock);
1297                 return;
1298         }
1299         mtx_lock(&lun->lun_lock);
1300         mtx_unlock(&softc->ctl_lock);
1301         if (lun->flags & CTL_LUN_DISABLED) {
1302                 mtx_unlock(&lun->lun_lock);
1303                 return;
1304         }
1305         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
1306                 if ((lun->mode_pages.index[i].page_code & SMPH_PC_MASK) ==
1307                     msg->mode.page_code &&
1308                     lun->mode_pages.index[i].subpage == msg->mode.subpage)
1309                         break;
1310         }
1311         if (i == CTL_NUM_MODE_PAGES) {
1312                 mtx_unlock(&lun->lun_lock);
1313                 return;
1314         }
1315         memcpy(lun->mode_pages.index[i].page_data, msg->mode.data,
1316             lun->mode_pages.index[i].page_len);
1317         initidx = ctl_get_initindex(&msg->hdr.nexus);
1318         if (initidx != -1)
1319                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
1320         mtx_unlock(&lun->lun_lock);
1321 }
1322
1323 /*
1324  * ISC (Inter Shelf Communication) event handler.  Events from the HA
1325  * subsystem come in here.
1326  */
1327 static void
1328 ctl_isc_event_handler(ctl_ha_channel channel, ctl_ha_event event, int param)
1329 {
1330         struct ctl_softc *softc = control_softc;
1331         union ctl_io *io;
1332         struct ctl_prio *presio;
1333         ctl_ha_status isc_status;
1334
1335         CTL_DEBUG_PRINT(("CTL: Isc Msg event %d\n", event));
1336         if (event == CTL_HA_EVT_MSG_RECV) {
1337                 union ctl_ha_msg *msg, msgbuf;
1338
1339                 if (param > sizeof(msgbuf))
1340                         msg = malloc(param, M_CTL, M_WAITOK);
1341                 else
1342                         msg = &msgbuf;
1343                 isc_status = ctl_ha_msg_recv(CTL_HA_CHAN_CTL, msg, param,
1344                     M_WAITOK);
1345                 if (isc_status != CTL_HA_STATUS_SUCCESS) {
1346                         printf("%s: Error receiving message: %d\n",
1347                             __func__, isc_status);
1348                         if (msg != &msgbuf)
1349                                 free(msg, M_CTL);
1350                         return;
1351                 }
1352
1353                 CTL_DEBUG_PRINT(("CTL: msg_type %d\n", msg->msg_type));
1354                 switch (msg->hdr.msg_type) {
1355                 case CTL_MSG_SERIALIZE:
1356                         io = ctl_alloc_io(softc->othersc_pool);
1357                         ctl_zero_io(io);
1358                         // populate ctsio from msg
1359                         io->io_hdr.io_type = CTL_IO_SCSI;
1360                         io->io_hdr.msg_type = CTL_MSG_SERIALIZE;
1361                         io->io_hdr.original_sc = msg->hdr.original_sc;
1362                         io->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC |
1363                                             CTL_FLAG_IO_ACTIVE;
1364                         /*
1365                          * If we're in serialization-only mode, we don't
1366                          * want to go through full done processing.  Thus
1367                          * the COPY flag.
1368                          *
1369                          * XXX KDM add another flag that is more specific.
1370                          */
1371                         if (softc->ha_mode != CTL_HA_MODE_XFER)
1372                                 io->io_hdr.flags |= CTL_FLAG_INT_COPY;
1373                         io->io_hdr.nexus = msg->hdr.nexus;
1374 #if 0
1375                         printf("port %u, iid %u, lun %u\n",
1376                                io->io_hdr.nexus.targ_port,
1377                                io->io_hdr.nexus.initid,
1378                                io->io_hdr.nexus.targ_lun);
1379 #endif
1380                         io->scsiio.tag_num = msg->scsi.tag_num;
1381                         io->scsiio.tag_type = msg->scsi.tag_type;
1382 #ifdef CTL_TIME_IO
1383                         io->io_hdr.start_time = time_uptime;
1384                         getbinuptime(&io->io_hdr.start_bt);
1385 #endif /* CTL_TIME_IO */
1386                         io->scsiio.cdb_len = msg->scsi.cdb_len;
1387                         memcpy(io->scsiio.cdb, msg->scsi.cdb,
1388                                CTL_MAX_CDBLEN);
1389                         if (softc->ha_mode == CTL_HA_MODE_XFER) {
1390                                 const struct ctl_cmd_entry *entry;
1391
1392                                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
1393                                 io->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
1394                                 io->io_hdr.flags |=
1395                                         entry->flags & CTL_FLAG_DATA_MASK;
1396                         }
1397                         ctl_enqueue_isc(io);
1398                         break;
1399
1400                 /* Performed on the Originating SC, XFER mode only */
1401                 case CTL_MSG_DATAMOVE: {
1402                         struct ctl_sg_entry *sgl;
1403                         int i, j;
1404
1405                         io = msg->hdr.original_sc;
1406                         if (io == NULL) {
1407                                 printf("%s: original_sc == NULL!\n", __func__);
1408                                 /* XXX KDM do something here */
1409                                 break;
1410                         }
1411                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE;
1412                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1413                         /*
1414                          * Keep track of this, we need to send it back over
1415                          * when the datamove is complete.
1416                          */
1417                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1418                         if (msg->hdr.status == CTL_SUCCESS)
1419                                 io->io_hdr.status = msg->hdr.status;
1420
1421                         if (msg->dt.sg_sequence == 0) {
1422 #ifdef CTL_TIME_IO
1423                                 getbinuptime(&io->io_hdr.dma_start_bt);
1424 #endif
1425                                 i = msg->dt.kern_sg_entries +
1426                                     msg->dt.kern_data_len /
1427                                     CTL_HA_DATAMOVE_SEGMENT + 1;
1428                                 sgl = malloc(sizeof(*sgl) * i, M_CTL,
1429                                     M_WAITOK | M_ZERO);
1430                                 io->io_hdr.remote_sglist = sgl;
1431                                 io->io_hdr.local_sglist =
1432                                     &sgl[msg->dt.kern_sg_entries];
1433
1434                                 io->scsiio.kern_data_ptr = (uint8_t *)sgl;
1435
1436                                 io->scsiio.kern_sg_entries =
1437                                         msg->dt.kern_sg_entries;
1438                                 io->scsiio.rem_sg_entries =
1439                                         msg->dt.kern_sg_entries;
1440                                 io->scsiio.kern_data_len =
1441                                         msg->dt.kern_data_len;
1442                                 io->scsiio.kern_total_len =
1443                                         msg->dt.kern_total_len;
1444                                 io->scsiio.kern_data_resid =
1445                                         msg->dt.kern_data_resid;
1446                                 io->scsiio.kern_rel_offset =
1447                                         msg->dt.kern_rel_offset;
1448                                 io->io_hdr.flags &= ~CTL_FLAG_BUS_ADDR;
1449                                 io->io_hdr.flags |= msg->dt.flags &
1450                                     CTL_FLAG_BUS_ADDR;
1451                         } else
1452                                 sgl = (struct ctl_sg_entry *)
1453                                         io->scsiio.kern_data_ptr;
1454
1455                         for (i = msg->dt.sent_sg_entries, j = 0;
1456                              i < (msg->dt.sent_sg_entries +
1457                              msg->dt.cur_sg_entries); i++, j++) {
1458                                 sgl[i].addr = msg->dt.sg_list[j].addr;
1459                                 sgl[i].len = msg->dt.sg_list[j].len;
1460
1461 #if 0
1462                                 printf("%s: DATAMOVE: %p,%lu j=%d, i=%d\n",
1463                                     __func__, sgl[i].addr, sgl[i].len, j, i);
1464 #endif
1465                         }
1466
1467                         /*
1468                          * If this is the last piece of the I/O, we've got
1469                          * the full S/G list.  Queue processing in the thread.
1470                          * Otherwise wait for the next piece.
1471                          */
1472                         if (msg->dt.sg_last != 0)
1473                                 ctl_enqueue_isc(io);
1474                         break;
1475                 }
1476                 /* Performed on the Serializing (primary) SC, XFER mode only */
1477                 case CTL_MSG_DATAMOVE_DONE: {
1478                         if (msg->hdr.serializing_sc == NULL) {
1479                                 printf("%s: serializing_sc == NULL!\n",
1480                                        __func__);
1481                                 /* XXX KDM now what? */
1482                                 break;
1483                         }
1484                         /*
1485                          * We grab the sense information here in case
1486                          * there was a failure, so we can return status
1487                          * back to the initiator.
1488                          */
1489                         io = msg->hdr.serializing_sc;
1490                         io->io_hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
1491                         io->io_hdr.flags &= ~CTL_FLAG_DMA_INPROG;
1492                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1493                         io->io_hdr.port_status = msg->scsi.fetd_status;
1494                         io->scsiio.residual = msg->scsi.residual;
1495                         if (msg->hdr.status != CTL_STATUS_NONE) {
1496                                 io->io_hdr.status = msg->hdr.status;
1497                                 io->scsiio.scsi_status = msg->scsi.scsi_status;
1498                                 io->scsiio.sense_len = msg->scsi.sense_len;
1499                                 io->scsiio.sense_residual =msg->scsi.sense_residual;
1500                                 memcpy(&io->scsiio.sense_data,
1501                                     &msg->scsi.sense_data,
1502                                     msg->scsi.sense_len);
1503                                 if (msg->hdr.status == CTL_SUCCESS)
1504                                         io->io_hdr.flags |= CTL_FLAG_STATUS_SENT;
1505                         }
1506                         ctl_enqueue_isc(io);
1507                         break;
1508                 }
1509
1510                 /* Preformed on Originating SC, SER_ONLY mode */
1511                 case CTL_MSG_R2R:
1512                         io = msg->hdr.original_sc;
1513                         if (io == NULL) {
1514                                 printf("%s: original_sc == NULL!\n",
1515                                     __func__);
1516                                 break;
1517                         }
1518                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1519                         io->io_hdr.msg_type = CTL_MSG_R2R;
1520                         io->io_hdr.serializing_sc = msg->hdr.serializing_sc;
1521                         ctl_enqueue_isc(io);
1522                         break;
1523
1524                 /*
1525                  * Performed on Serializing(i.e. primary SC) SC in SER_ONLY
1526                  * mode.
1527                  * Performed on the Originating (i.e. secondary) SC in XFER
1528                  * mode
1529                  */
1530                 case CTL_MSG_FINISH_IO:
1531                         if (softc->ha_mode == CTL_HA_MODE_XFER)
1532                                 ctl_isc_handler_finish_xfer(softc, msg);
1533                         else
1534                                 ctl_isc_handler_finish_ser_only(softc, msg);
1535                         break;
1536
1537                 /* Preformed on Originating SC */
1538                 case CTL_MSG_BAD_JUJU:
1539                         io = msg->hdr.original_sc;
1540                         if (io == NULL) {
1541                                 printf("%s: Bad JUJU!, original_sc is NULL!\n",
1542                                        __func__);
1543                                 break;
1544                         }
1545                         ctl_copy_sense_data(msg, io);
1546                         /*
1547                          * IO should have already been cleaned up on other
1548                          * SC so clear this flag so we won't send a message
1549                          * back to finish the IO there.
1550                          */
1551                         io->io_hdr.flags &= ~CTL_FLAG_SENT_2OTHER_SC;
1552                         io->io_hdr.flags |= CTL_FLAG_IO_ACTIVE;
1553
1554                         /* io = msg->hdr.serializing_sc; */
1555                         io->io_hdr.msg_type = CTL_MSG_BAD_JUJU;
1556                         ctl_enqueue_isc(io);
1557                         break;
1558
1559                 /* Handle resets sent from the other side */
1560                 case CTL_MSG_MANAGE_TASKS: {
1561                         struct ctl_taskio *taskio;
1562                         taskio = (struct ctl_taskio *)ctl_alloc_io(
1563                             softc->othersc_pool);
1564                         ctl_zero_io((union ctl_io *)taskio);
1565                         taskio->io_hdr.io_type = CTL_IO_TASK;
1566                         taskio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1567                         taskio->io_hdr.nexus = msg->hdr.nexus;
1568                         taskio->task_action = msg->task.task_action;
1569                         taskio->tag_num = msg->task.tag_num;
1570                         taskio->tag_type = msg->task.tag_type;
1571 #ifdef CTL_TIME_IO
1572                         taskio->io_hdr.start_time = time_uptime;
1573                         getbinuptime(&taskio->io_hdr.start_bt);
1574 #endif /* CTL_TIME_IO */
1575                         ctl_run_task((union ctl_io *)taskio);
1576                         break;
1577                 }
1578                 /* Persistent Reserve action which needs attention */
1579                 case CTL_MSG_PERS_ACTION:
1580                         presio = (struct ctl_prio *)ctl_alloc_io(
1581                             softc->othersc_pool);
1582                         ctl_zero_io((union ctl_io *)presio);
1583                         presio->io_hdr.msg_type = CTL_MSG_PERS_ACTION;
1584                         presio->io_hdr.flags |= CTL_FLAG_FROM_OTHER_SC;
1585                         presio->io_hdr.nexus = msg->hdr.nexus;
1586                         presio->pr_msg = msg->pr;
1587                         ctl_enqueue_isc((union ctl_io *)presio);
1588                         break;
1589                 case CTL_MSG_UA:
1590                         ctl_isc_ua(softc, msg, param);
1591                         break;
1592                 case CTL_MSG_PORT_SYNC:
1593                         ctl_isc_port_sync(softc, msg, param);
1594                         break;
1595                 case CTL_MSG_LUN_SYNC:
1596                         ctl_isc_lun_sync(softc, msg, param);
1597                         break;
1598                 case CTL_MSG_IID_SYNC:
1599                         ctl_isc_iid_sync(softc, msg, param);
1600                         break;
1601                 case CTL_MSG_LOGIN:
1602                         ctl_isc_login(softc, msg, param);
1603                         break;
1604                 case CTL_MSG_MODE_SYNC:
1605                         ctl_isc_mode_sync(softc, msg, param);
1606                         break;
1607                 default:
1608                         printf("Received HA message of unknown type %d\n",
1609                             msg->hdr.msg_type);
1610                         ctl_ha_msg_abort(CTL_HA_CHAN_CTL);
1611                         break;
1612                 }
1613                 if (msg != &msgbuf)
1614                         free(msg, M_CTL);
1615         } else if (event == CTL_HA_EVT_LINK_CHANGE) {
1616                 printf("CTL: HA link status changed from %d to %d\n",
1617                     softc->ha_link, param);
1618                 if (param == softc->ha_link)
1619                         return;
1620                 if (softc->ha_link == CTL_HA_LINK_ONLINE) {
1621                         softc->ha_link = param;
1622                         ctl_isc_ha_link_down(softc);
1623                 } else {
1624                         softc->ha_link = param;
1625                         if (softc->ha_link == CTL_HA_LINK_ONLINE)
1626                                 ctl_isc_ha_link_up(softc);
1627                 }
1628                 return;
1629         } else {
1630                 printf("ctl_isc_event_handler: Unknown event %d\n", event);
1631                 return;
1632         }
1633 }
1634
1635 static void
1636 ctl_copy_sense_data(union ctl_ha_msg *src, union ctl_io *dest)
1637 {
1638
1639         memcpy(&dest->scsiio.sense_data, &src->scsi.sense_data,
1640             src->scsi.sense_len);
1641         dest->scsiio.scsi_status = src->scsi.scsi_status;
1642         dest->scsiio.sense_len = src->scsi.sense_len;
1643         dest->io_hdr.status = src->hdr.status;
1644 }
1645
1646 static void
1647 ctl_copy_sense_data_back(union ctl_io *src, union ctl_ha_msg *dest)
1648 {
1649
1650         memcpy(&dest->scsi.sense_data, &src->scsiio.sense_data,
1651             src->scsiio.sense_len);
1652         dest->scsi.scsi_status = src->scsiio.scsi_status;
1653         dest->scsi.sense_len = src->scsiio.sense_len;
1654         dest->hdr.status = src->io_hdr.status;
1655 }
1656
1657 void
1658 ctl_est_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1659 {
1660         struct ctl_softc *softc = lun->ctl_softc;
1661         ctl_ua_type *pu;
1662
1663         if (initidx < softc->init_min || initidx >= softc->init_max)
1664                 return;
1665         mtx_assert(&lun->lun_lock, MA_OWNED);
1666         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1667         if (pu == NULL)
1668                 return;
1669         pu[initidx % CTL_MAX_INIT_PER_PORT] |= ua;
1670 }
1671
1672 void
1673 ctl_est_ua_port(struct ctl_lun *lun, int port, uint32_t except, ctl_ua_type ua)
1674 {
1675         int i;
1676
1677         mtx_assert(&lun->lun_lock, MA_OWNED);
1678         if (lun->pending_ua[port] == NULL)
1679                 return;
1680         for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
1681                 if (port * CTL_MAX_INIT_PER_PORT + i == except)
1682                         continue;
1683                 lun->pending_ua[port][i] |= ua;
1684         }
1685 }
1686
1687 void
1688 ctl_est_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1689 {
1690         struct ctl_softc *softc = lun->ctl_softc;
1691         int i;
1692
1693         mtx_assert(&lun->lun_lock, MA_OWNED);
1694         for (i = softc->port_min; i < softc->port_max; i++)
1695                 ctl_est_ua_port(lun, i, except, ua);
1696 }
1697
1698 void
1699 ctl_clr_ua(struct ctl_lun *lun, uint32_t initidx, ctl_ua_type ua)
1700 {
1701         struct ctl_softc *softc = lun->ctl_softc;
1702         ctl_ua_type *pu;
1703
1704         if (initidx < softc->init_min || initidx >= softc->init_max)
1705                 return;
1706         mtx_assert(&lun->lun_lock, MA_OWNED);
1707         pu = lun->pending_ua[initidx / CTL_MAX_INIT_PER_PORT];
1708         if (pu == NULL)
1709                 return;
1710         pu[initidx % CTL_MAX_INIT_PER_PORT] &= ~ua;
1711 }
1712
1713 void
1714 ctl_clr_ua_all(struct ctl_lun *lun, uint32_t except, ctl_ua_type ua)
1715 {
1716         struct ctl_softc *softc = lun->ctl_softc;
1717         int i, j;
1718
1719         mtx_assert(&lun->lun_lock, MA_OWNED);
1720         for (i = softc->port_min; i < softc->port_max; i++) {
1721                 if (lun->pending_ua[i] == NULL)
1722                         continue;
1723                 for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
1724                         if (i * CTL_MAX_INIT_PER_PORT + j == except)
1725                                 continue;
1726                         lun->pending_ua[i][j] &= ~ua;
1727                 }
1728         }
1729 }
1730
1731 void
1732 ctl_clr_ua_allluns(struct ctl_softc *ctl_softc, uint32_t initidx,
1733     ctl_ua_type ua_type)
1734 {
1735         struct ctl_lun *lun;
1736
1737         mtx_assert(&ctl_softc->ctl_lock, MA_OWNED);
1738         STAILQ_FOREACH(lun, &ctl_softc->lun_list, links) {
1739                 mtx_lock(&lun->lun_lock);
1740                 ctl_clr_ua(lun, initidx, ua_type);
1741                 mtx_unlock(&lun->lun_lock);
1742         }
1743 }
1744
1745 static int
1746 ctl_ha_role_sysctl(SYSCTL_HANDLER_ARGS)
1747 {
1748         struct ctl_softc *softc = (struct ctl_softc *)arg1;
1749         struct ctl_lun *lun;
1750         struct ctl_lun_req ireq;
1751         int error, value;
1752
1753         value = (softc->flags & CTL_FLAG_ACTIVE_SHELF) ? 0 : 1;
1754         error = sysctl_handle_int(oidp, &value, 0, req);
1755         if ((error != 0) || (req->newptr == NULL))
1756                 return (error);
1757
1758         mtx_lock(&softc->ctl_lock);
1759         if (value == 0)
1760                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1761         else
1762                 softc->flags &= ~CTL_FLAG_ACTIVE_SHELF;
1763         STAILQ_FOREACH(lun, &softc->lun_list, links) {
1764                 mtx_unlock(&softc->ctl_lock);
1765                 bzero(&ireq, sizeof(ireq));
1766                 ireq.reqtype = CTL_LUNREQ_MODIFY;
1767                 ireq.reqdata.modify.lun_id = lun->lun;
1768                 lun->backend->ioctl(NULL, CTL_LUN_REQ, (caddr_t)&ireq, 0,
1769                     curthread);
1770                 if (ireq.status != CTL_LUN_OK) {
1771                         printf("%s: CTL_LUNREQ_MODIFY returned %d '%s'\n",
1772                             __func__, ireq.status, ireq.error_str);
1773                 }
1774                 mtx_lock(&softc->ctl_lock);
1775         }
1776         mtx_unlock(&softc->ctl_lock);
1777         return (0);
1778 }
1779
1780 static int
1781 ctl_init(void)
1782 {
1783         struct make_dev_args args;
1784         struct ctl_softc *softc;
1785         void *other_pool;
1786         int i, error;
1787
1788         softc = control_softc = malloc(sizeof(*control_softc), M_DEVBUF,
1789                                M_WAITOK | M_ZERO);
1790
1791         make_dev_args_init(&args);
1792         args.mda_devsw = &ctl_cdevsw;
1793         args.mda_uid = UID_ROOT;
1794         args.mda_gid = GID_OPERATOR;
1795         args.mda_mode = 0600;
1796         args.mda_si_drv1 = softc;
1797         error = make_dev_s(&args, &softc->dev, "cam/ctl");
1798         if (error != 0) {
1799                 free(control_softc, M_DEVBUF);
1800                 return (error);
1801         }
1802
1803         sysctl_ctx_init(&softc->sysctl_ctx);
1804         softc->sysctl_tree = SYSCTL_ADD_NODE(&softc->sysctl_ctx,
1805                 SYSCTL_STATIC_CHILDREN(_kern_cam), OID_AUTO, "ctl",
1806                 CTLFLAG_RD, 0, "CAM Target Layer");
1807
1808         if (softc->sysctl_tree == NULL) {
1809                 printf("%s: unable to allocate sysctl tree\n", __func__);
1810                 destroy_dev(softc->dev);
1811                 free(control_softc, M_DEVBUF);
1812                 control_softc = NULL;
1813                 return (ENOMEM);
1814         }
1815
1816         mtx_init(&softc->ctl_lock, "CTL mutex", NULL, MTX_DEF);
1817         softc->io_zone = uma_zcreate("CTL IO", sizeof(union ctl_io),
1818             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
1819         softc->flags = 0;
1820
1821         TUNABLE_INT_FETCH("kern.cam.ctl.ha_mode", (int *)&softc->ha_mode);
1822         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1823             OID_AUTO, "ha_mode", CTLFLAG_RDTUN, (int *)&softc->ha_mode, 0,
1824             "HA mode (0 - act/stby, 1 - serialize only, 2 - xfer)");
1825
1826         /*
1827          * In Copan's HA scheme, the "master" and "slave" roles are
1828          * figured out through the slot the controller is in.  Although it
1829          * is an active/active system, someone has to be in charge.
1830          */
1831         TUNABLE_INT_FETCH("kern.cam.ctl.ha_id", &softc->ha_id);
1832         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1833             OID_AUTO, "ha_id", CTLFLAG_RDTUN, &softc->ha_id, 0,
1834             "HA head ID (0 - no HA)");
1835         if (softc->ha_id == 0 || softc->ha_id > NUM_HA_SHELVES) {
1836                 softc->flags |= CTL_FLAG_ACTIVE_SHELF;
1837                 softc->is_single = 1;
1838                 softc->port_cnt = CTL_MAX_PORTS;
1839                 softc->port_min = 0;
1840         } else {
1841                 softc->port_cnt = CTL_MAX_PORTS / NUM_HA_SHELVES;
1842                 softc->port_min = (softc->ha_id - 1) * softc->port_cnt;
1843         }
1844         softc->port_max = softc->port_min + softc->port_cnt;
1845         softc->init_min = softc->port_min * CTL_MAX_INIT_PER_PORT;
1846         softc->init_max = softc->port_max * CTL_MAX_INIT_PER_PORT;
1847
1848         SYSCTL_ADD_INT(&softc->sysctl_ctx, SYSCTL_CHILDREN(softc->sysctl_tree),
1849             OID_AUTO, "ha_link", CTLFLAG_RD, (int *)&softc->ha_link, 0,
1850             "HA link state (0 - offline, 1 - unknown, 2 - online)");
1851
1852         STAILQ_INIT(&softc->lun_list);
1853         STAILQ_INIT(&softc->pending_lun_queue);
1854         STAILQ_INIT(&softc->fe_list);
1855         STAILQ_INIT(&softc->port_list);
1856         STAILQ_INIT(&softc->be_list);
1857         ctl_tpc_init(softc);
1858
1859         if (ctl_pool_create(softc, "othersc", CTL_POOL_ENTRIES_OTHER_SC,
1860                             &other_pool) != 0)
1861         {
1862                 printf("ctl: can't allocate %d entry other SC pool, "
1863                        "exiting\n", CTL_POOL_ENTRIES_OTHER_SC);
1864                 return (ENOMEM);
1865         }
1866         softc->othersc_pool = other_pool;
1867
1868         if (worker_threads <= 0)
1869                 worker_threads = max(1, mp_ncpus / 4);
1870         if (worker_threads > CTL_MAX_THREADS)
1871                 worker_threads = CTL_MAX_THREADS;
1872
1873         for (i = 0; i < worker_threads; i++) {
1874                 struct ctl_thread *thr = &softc->threads[i];
1875
1876                 mtx_init(&thr->queue_lock, "CTL queue mutex", NULL, MTX_DEF);
1877                 thr->ctl_softc = softc;
1878                 STAILQ_INIT(&thr->incoming_queue);
1879                 STAILQ_INIT(&thr->rtr_queue);
1880                 STAILQ_INIT(&thr->done_queue);
1881                 STAILQ_INIT(&thr->isc_queue);
1882
1883                 error = kproc_kthread_add(ctl_work_thread, thr,
1884                     &softc->ctl_proc, &thr->thread, 0, 0, "ctl", "work%d", i);
1885                 if (error != 0) {
1886                         printf("error creating CTL work thread!\n");
1887                         ctl_pool_free(other_pool);
1888                         return (error);
1889                 }
1890         }
1891         error = kproc_kthread_add(ctl_lun_thread, softc,
1892             &softc->ctl_proc, NULL, 0, 0, "ctl", "lun");
1893         if (error != 0) {
1894                 printf("error creating CTL lun thread!\n");
1895                 ctl_pool_free(other_pool);
1896                 return (error);
1897         }
1898         error = kproc_kthread_add(ctl_thresh_thread, softc,
1899             &softc->ctl_proc, NULL, 0, 0, "ctl", "thresh");
1900         if (error != 0) {
1901                 printf("error creating CTL threshold thread!\n");
1902                 ctl_pool_free(other_pool);
1903                 return (error);
1904         }
1905
1906         SYSCTL_ADD_PROC(&softc->sysctl_ctx,SYSCTL_CHILDREN(softc->sysctl_tree),
1907             OID_AUTO, "ha_role", CTLTYPE_INT | CTLFLAG_RWTUN,
1908             softc, 0, ctl_ha_role_sysctl, "I", "HA role for this head");
1909
1910         if (softc->is_single == 0) {
1911                 ctl_frontend_register(&ha_frontend);
1912                 if (ctl_ha_msg_init(softc) != CTL_HA_STATUS_SUCCESS) {
1913                         printf("ctl_init: ctl_ha_msg_init failed.\n");
1914                         softc->is_single = 1;
1915                 } else
1916                 if (ctl_ha_msg_register(CTL_HA_CHAN_CTL, ctl_isc_event_handler)
1917                     != CTL_HA_STATUS_SUCCESS) {
1918                         printf("ctl_init: ctl_ha_msg_register failed.\n");
1919                         softc->is_single = 1;
1920                 }
1921         }
1922         return (0);
1923 }
1924
1925 void
1926 ctl_shutdown(void)
1927 {
1928         struct ctl_softc *softc = control_softc;
1929         struct ctl_lun *lun, *next_lun;
1930
1931         if (softc->is_single == 0) {
1932                 ctl_ha_msg_shutdown(softc);
1933                 if (ctl_ha_msg_deregister(CTL_HA_CHAN_CTL)
1934                     != CTL_HA_STATUS_SUCCESS)
1935                         printf("%s: ctl_ha_msg_deregister failed.\n", __func__);
1936                 if (ctl_ha_msg_destroy(softc) != CTL_HA_STATUS_SUCCESS)
1937                         printf("%s: ctl_ha_msg_destroy failed.\n", __func__);
1938                 ctl_frontend_deregister(&ha_frontend);
1939         }
1940
1941         mtx_lock(&softc->ctl_lock);
1942
1943         STAILQ_FOREACH_SAFE(lun, &softc->lun_list, links, next_lun)
1944                 ctl_free_lun(lun);
1945
1946         mtx_unlock(&softc->ctl_lock);
1947
1948 #if 0
1949         ctl_shutdown_thread(softc->work_thread);
1950         mtx_destroy(&softc->queue_lock);
1951 #endif
1952
1953         ctl_tpc_shutdown(softc);
1954         uma_zdestroy(softc->io_zone);
1955         mtx_destroy(&softc->ctl_lock);
1956
1957         destroy_dev(softc->dev);
1958
1959         sysctl_ctx_free(&softc->sysctl_ctx);
1960
1961         free(control_softc, M_DEVBUF);
1962         control_softc = NULL;
1963 }
1964
1965 static int
1966 ctl_module_event_handler(module_t mod, int what, void *arg)
1967 {
1968
1969         switch (what) {
1970         case MOD_LOAD:
1971                 return (ctl_init());
1972         case MOD_UNLOAD:
1973                 return (EBUSY);
1974         default:
1975                 return (EOPNOTSUPP);
1976         }
1977 }
1978
1979 /*
1980  * XXX KDM should we do some access checks here?  Bump a reference count to
1981  * prevent a CTL module from being unloaded while someone has it open?
1982  */
1983 static int
1984 ctl_open(struct cdev *dev, int flags, int fmt, struct thread *td)
1985 {
1986         return (0);
1987 }
1988
1989 static int
1990 ctl_close(struct cdev *dev, int flags, int fmt, struct thread *td)
1991 {
1992         return (0);
1993 }
1994
1995 /*
1996  * Remove an initiator by port number and initiator ID.
1997  * Returns 0 for success, -1 for failure.
1998  */
1999 int
2000 ctl_remove_initiator(struct ctl_port *port, int iid)
2001 {
2002         struct ctl_softc *softc = port->ctl_softc;
2003
2004         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2005
2006         if (iid > CTL_MAX_INIT_PER_PORT) {
2007                 printf("%s: initiator ID %u > maximun %u!\n",
2008                        __func__, iid, CTL_MAX_INIT_PER_PORT);
2009                 return (-1);
2010         }
2011
2012         mtx_lock(&softc->ctl_lock);
2013         port->wwpn_iid[iid].in_use--;
2014         port->wwpn_iid[iid].last_use = time_uptime;
2015         mtx_unlock(&softc->ctl_lock);
2016         ctl_isc_announce_iid(port, iid);
2017
2018         return (0);
2019 }
2020
2021 /*
2022  * Add an initiator to the initiator map.
2023  * Returns iid for success, < 0 for failure.
2024  */
2025 int
2026 ctl_add_initiator(struct ctl_port *port, int iid, uint64_t wwpn, char *name)
2027 {
2028         struct ctl_softc *softc = port->ctl_softc;
2029         time_t best_time;
2030         int i, best;
2031
2032         mtx_assert(&softc->ctl_lock, MA_NOTOWNED);
2033
2034         if (iid >= CTL_MAX_INIT_PER_PORT) {
2035                 printf("%s: WWPN %#jx initiator ID %u > maximum %u!\n",
2036                        __func__, wwpn, iid, CTL_MAX_INIT_PER_PORT);
2037                 free(name, M_CTL);
2038                 return (-1);
2039         }
2040
2041         mtx_lock(&softc->ctl_lock);
2042
2043         if (iid < 0 && (wwpn != 0 || name != NULL)) {
2044                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2045                         if (wwpn != 0 && wwpn == port->wwpn_iid[i].wwpn) {
2046                                 iid = i;
2047                                 break;
2048                         }
2049                         if (name != NULL && port->wwpn_iid[i].name != NULL &&
2050                             strcmp(name, port->wwpn_iid[i].name) == 0) {
2051                                 iid = i;
2052                                 break;
2053                         }
2054                 }
2055         }
2056
2057         if (iid < 0) {
2058                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2059                         if (port->wwpn_iid[i].in_use == 0 &&
2060                             port->wwpn_iid[i].wwpn == 0 &&
2061                             port->wwpn_iid[i].name == NULL) {
2062                                 iid = i;
2063                                 break;
2064                         }
2065                 }
2066         }
2067
2068         if (iid < 0) {
2069                 best = -1;
2070                 best_time = INT32_MAX;
2071                 for (i = 0; i < CTL_MAX_INIT_PER_PORT; i++) {
2072                         if (port->wwpn_iid[i].in_use == 0) {
2073                                 if (port->wwpn_iid[i].last_use < best_time) {
2074                                         best = i;
2075                                         best_time = port->wwpn_iid[i].last_use;
2076                                 }
2077                         }
2078                 }
2079                 iid = best;
2080         }
2081
2082         if (iid < 0) {
2083                 mtx_unlock(&softc->ctl_lock);
2084                 free(name, M_CTL);
2085                 return (-2);
2086         }
2087
2088         if (port->wwpn_iid[iid].in_use > 0 && (wwpn != 0 || name != NULL)) {
2089                 /*
2090                  * This is not an error yet.
2091                  */
2092                 if (wwpn != 0 && wwpn == port->wwpn_iid[iid].wwpn) {
2093 #if 0
2094                         printf("%s: port %d iid %u WWPN %#jx arrived"
2095                             " again\n", __func__, port->targ_port,
2096                             iid, (uintmax_t)wwpn);
2097 #endif
2098                         goto take;
2099                 }
2100                 if (name != NULL && port->wwpn_iid[iid].name != NULL &&
2101                     strcmp(name, port->wwpn_iid[iid].name) == 0) {
2102 #if 0
2103                         printf("%s: port %d iid %u name '%s' arrived"
2104                             " again\n", __func__, port->targ_port,
2105                             iid, name);
2106 #endif
2107                         goto take;
2108                 }
2109
2110                 /*
2111                  * This is an error, but what do we do about it?  The
2112                  * driver is telling us we have a new WWPN for this
2113                  * initiator ID, so we pretty much need to use it.
2114                  */
2115                 printf("%s: port %d iid %u WWPN %#jx '%s' arrived,"
2116                     " but WWPN %#jx '%s' is still at that address\n",
2117                     __func__, port->targ_port, iid, wwpn, name,
2118                     (uintmax_t)port->wwpn_iid[iid].wwpn,
2119                     port->wwpn_iid[iid].name);
2120
2121                 /*
2122                  * XXX KDM clear have_ca and ua_pending on each LUN for
2123                  * this initiator.
2124                  */
2125         }
2126 take:
2127         free(port->wwpn_iid[iid].name, M_CTL);
2128         port->wwpn_iid[iid].name = name;
2129         port->wwpn_iid[iid].wwpn = wwpn;
2130         port->wwpn_iid[iid].in_use++;
2131         mtx_unlock(&softc->ctl_lock);
2132         ctl_isc_announce_iid(port, iid);
2133
2134         return (iid);
2135 }
2136
2137 static int
2138 ctl_create_iid(struct ctl_port *port, int iid, uint8_t *buf)
2139 {
2140         int len;
2141
2142         switch (port->port_type) {
2143         case CTL_PORT_FC:
2144         {
2145                 struct scsi_transportid_fcp *id =
2146                     (struct scsi_transportid_fcp *)buf;
2147                 if (port->wwpn_iid[iid].wwpn == 0)
2148                         return (0);
2149                 memset(id, 0, sizeof(*id));
2150                 id->format_protocol = SCSI_PROTO_FC;
2151                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->n_port_name);
2152                 return (sizeof(*id));
2153         }
2154         case CTL_PORT_ISCSI:
2155         {
2156                 struct scsi_transportid_iscsi_port *id =
2157                     (struct scsi_transportid_iscsi_port *)buf;
2158                 if (port->wwpn_iid[iid].name == NULL)
2159                         return (0);
2160                 memset(id, 0, 256);
2161                 id->format_protocol = SCSI_TRN_ISCSI_FORMAT_PORT |
2162                     SCSI_PROTO_ISCSI;
2163                 len = strlcpy(id->iscsi_name, port->wwpn_iid[iid].name, 252) + 1;
2164                 len = roundup2(min(len, 252), 4);
2165                 scsi_ulto2b(len, id->additional_length);
2166                 return (sizeof(*id) + len);
2167         }
2168         case CTL_PORT_SAS:
2169         {
2170                 struct scsi_transportid_sas *id =
2171                     (struct scsi_transportid_sas *)buf;
2172                 if (port->wwpn_iid[iid].wwpn == 0)
2173                         return (0);
2174                 memset(id, 0, sizeof(*id));
2175                 id->format_protocol = SCSI_PROTO_SAS;
2176                 scsi_u64to8b(port->wwpn_iid[iid].wwpn, id->sas_address);
2177                 return (sizeof(*id));
2178         }
2179         default:
2180         {
2181                 struct scsi_transportid_spi *id =
2182                     (struct scsi_transportid_spi *)buf;
2183                 memset(id, 0, sizeof(*id));
2184                 id->format_protocol = SCSI_PROTO_SPI;
2185                 scsi_ulto2b(iid, id->scsi_addr);
2186                 scsi_ulto2b(port->targ_port, id->rel_trgt_port_id);
2187                 return (sizeof(*id));
2188         }
2189         }
2190 }
2191
2192 /*
2193  * Serialize a command that went down the "wrong" side, and so was sent to
2194  * this controller for execution.  The logic is a little different than the
2195  * standard case in ctl_scsiio_precheck().  Errors in this case need to get
2196  * sent back to the other side, but in the success case, we execute the
2197  * command on this side (XFER mode) or tell the other side to execute it
2198  * (SER_ONLY mode).
2199  */
2200 static int
2201 ctl_serialize_other_sc_cmd(struct ctl_scsiio *ctsio)
2202 {
2203         struct ctl_softc *softc = control_softc;
2204         union ctl_ha_msg msg_info;
2205         struct ctl_port *port;
2206         struct ctl_lun *lun;
2207         const struct ctl_cmd_entry *entry;
2208         int retval = 0;
2209         uint32_t targ_lun;
2210
2211         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
2212         mtx_lock(&softc->ctl_lock);
2213
2214         /* Make sure that we know about this port. */
2215         port = ctl_io_port(&ctsio->io_hdr);
2216         if (port == NULL || (port->status & CTL_PORT_STATUS_ONLINE) == 0) {
2217                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2218                                          /*retry_count*/ 1);
2219                 goto badjuju;
2220         }
2221
2222         /* Make sure that we know about this LUN. */
2223         if ((targ_lun < CTL_MAX_LUNS) &&
2224             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
2225                 mtx_lock(&lun->lun_lock);
2226                 mtx_unlock(&softc->ctl_lock);
2227                 /*
2228                  * If the LUN is invalid, pretend that it doesn't exist.
2229                  * It will go away as soon as all pending I/O has been
2230                  * completed.
2231                  */
2232                 if (lun->flags & CTL_LUN_DISABLED) {
2233                         mtx_unlock(&lun->lun_lock);
2234                         lun = NULL;
2235                 }
2236         } else {
2237                 mtx_unlock(&softc->ctl_lock);
2238                 lun = NULL;
2239         }
2240         if (lun == NULL) {
2241                 /*
2242                  * The other node would not send this request to us unless
2243                  * received announce that we are primary node for this LUN.
2244                  * If this LUN does not exist now, it is probably result of
2245                  * a race, so respond to initiator in the most opaque way.
2246                  */
2247                 ctl_set_busy(ctsio);
2248                 goto badjuju;
2249         }
2250
2251         entry = ctl_get_cmd_entry(ctsio, NULL);
2252         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
2253                 mtx_unlock(&lun->lun_lock);
2254                 goto badjuju;
2255         }
2256
2257         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
2258         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = lun->be_lun;
2259
2260         /*
2261          * Every I/O goes into the OOA queue for a
2262          * particular LUN, and stays there until completion.
2263          */
2264 #ifdef CTL_TIME_IO
2265         if (TAILQ_EMPTY(&lun->ooa_queue))
2266                 lun->idle_time += getsbinuptime() - lun->last_busy;
2267 #endif
2268         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2269
2270         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
2271                 (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr, ctl_ooaq,
2272                  ooa_links))) {
2273         case CTL_ACTION_BLOCK:
2274                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
2275                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
2276                                   blocked_links);
2277                 mtx_unlock(&lun->lun_lock);
2278                 break;
2279         case CTL_ACTION_PASS:
2280         case CTL_ACTION_SKIP:
2281                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
2282                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
2283                         ctl_enqueue_rtr((union ctl_io *)ctsio);
2284                         mtx_unlock(&lun->lun_lock);
2285                 } else {
2286                         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
2287                         mtx_unlock(&lun->lun_lock);
2288
2289                         /* send msg back to other side */
2290                         msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2291                         msg_info.hdr.serializing_sc = (union ctl_io *)ctsio;
2292                         msg_info.hdr.msg_type = CTL_MSG_R2R;
2293                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2294                             sizeof(msg_info.hdr), M_WAITOK);
2295                 }
2296                 break;
2297         case CTL_ACTION_OVERLAP:
2298                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2299                 mtx_unlock(&lun->lun_lock);
2300                 ctl_set_overlapped_cmd(ctsio);
2301                 goto badjuju;
2302         case CTL_ACTION_OVERLAP_TAG:
2303                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2304                 mtx_unlock(&lun->lun_lock);
2305                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num);
2306                 goto badjuju;
2307         case CTL_ACTION_ERROR:
2308         default:
2309                 TAILQ_REMOVE(&lun->ooa_queue, &ctsio->io_hdr, ooa_links);
2310                 mtx_unlock(&lun->lun_lock);
2311
2312                 ctl_set_internal_failure(ctsio, /*sks_valid*/ 0,
2313                                          /*retry_count*/ 0);
2314 badjuju:
2315                 ctl_copy_sense_data_back((union ctl_io *)ctsio, &msg_info);
2316                 msg_info.hdr.original_sc = ctsio->io_hdr.original_sc;
2317                 msg_info.hdr.serializing_sc = NULL;
2318                 msg_info.hdr.msg_type = CTL_MSG_BAD_JUJU;
2319                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
2320                     sizeof(msg_info.scsi), M_WAITOK);
2321                 retval = 1;
2322                 break;
2323         }
2324         return (retval);
2325 }
2326
2327 /*
2328  * Returns 0 for success, errno for failure.
2329  */
2330 static void
2331 ctl_ioctl_fill_ooa(struct ctl_lun *lun, uint32_t *cur_fill_num,
2332                    struct ctl_ooa *ooa_hdr, struct ctl_ooa_entry *kern_entries)
2333 {
2334         union ctl_io *io;
2335
2336         mtx_lock(&lun->lun_lock);
2337         for (io = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); (io != NULL);
2338              (*cur_fill_num)++, io = (union ctl_io *)TAILQ_NEXT(&io->io_hdr,
2339              ooa_links)) {
2340                 struct ctl_ooa_entry *entry;
2341
2342                 /*
2343                  * If we've got more than we can fit, just count the
2344                  * remaining entries.
2345                  */
2346                 if (*cur_fill_num >= ooa_hdr->alloc_num)
2347                         continue;
2348
2349                 entry = &kern_entries[*cur_fill_num];
2350
2351                 entry->tag_num = io->scsiio.tag_num;
2352                 entry->lun_num = lun->lun;
2353 #ifdef CTL_TIME_IO
2354                 entry->start_bt = io->io_hdr.start_bt;
2355 #endif
2356                 bcopy(io->scsiio.cdb, entry->cdb, io->scsiio.cdb_len);
2357                 entry->cdb_len = io->scsiio.cdb_len;
2358                 if (io->io_hdr.flags & CTL_FLAG_BLOCKED)
2359                         entry->cmd_flags |= CTL_OOACMD_FLAG_BLOCKED;
2360
2361                 if (io->io_hdr.flags & CTL_FLAG_DMA_INPROG)
2362                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA;
2363
2364                 if (io->io_hdr.flags & CTL_FLAG_ABORT)
2365                         entry->cmd_flags |= CTL_OOACMD_FLAG_ABORT;
2366
2367                 if (io->io_hdr.flags & CTL_FLAG_IS_WAS_ON_RTR)
2368                         entry->cmd_flags |= CTL_OOACMD_FLAG_RTR;
2369
2370                 if (io->io_hdr.flags & CTL_FLAG_DMA_QUEUED)
2371                         entry->cmd_flags |= CTL_OOACMD_FLAG_DMA_QUEUED;
2372         }
2373         mtx_unlock(&lun->lun_lock);
2374 }
2375
2376 static void *
2377 ctl_copyin_alloc(void *user_addr, int len, char *error_str,
2378                  size_t error_str_len)
2379 {
2380         void *kptr;
2381
2382         kptr = malloc(len, M_CTL, M_WAITOK | M_ZERO);
2383
2384         if (copyin(user_addr, kptr, len) != 0) {
2385                 snprintf(error_str, error_str_len, "Error copying %d bytes "
2386                          "from user address %p to kernel address %p", len,
2387                          user_addr, kptr);
2388                 free(kptr, M_CTL);
2389                 return (NULL);
2390         }
2391
2392         return (kptr);
2393 }
2394
2395 static void
2396 ctl_free_args(int num_args, struct ctl_be_arg *args)
2397 {
2398         int i;
2399
2400         if (args == NULL)
2401                 return;
2402
2403         for (i = 0; i < num_args; i++) {
2404                 free(args[i].kname, M_CTL);
2405                 free(args[i].kvalue, M_CTL);
2406         }
2407
2408         free(args, M_CTL);
2409 }
2410
2411 static struct ctl_be_arg *
2412 ctl_copyin_args(int num_args, struct ctl_be_arg *uargs,
2413                 char *error_str, size_t error_str_len)
2414 {
2415         struct ctl_be_arg *args;
2416         int i;
2417
2418         args = ctl_copyin_alloc(uargs, num_args * sizeof(*args),
2419                                 error_str, error_str_len);
2420
2421         if (args == NULL)
2422                 goto bailout;
2423
2424         for (i = 0; i < num_args; i++) {
2425                 args[i].kname = NULL;
2426                 args[i].kvalue = NULL;
2427         }
2428
2429         for (i = 0; i < num_args; i++) {
2430                 uint8_t *tmpptr;
2431
2432                 args[i].kname = ctl_copyin_alloc(args[i].name,
2433                         args[i].namelen, error_str, error_str_len);
2434                 if (args[i].kname == NULL)
2435                         goto bailout;
2436
2437                 if (args[i].kname[args[i].namelen - 1] != '\0') {
2438                         snprintf(error_str, error_str_len, "Argument %d "
2439                                  "name is not NUL-terminated", i);
2440                         goto bailout;
2441                 }
2442
2443                 if (args[i].flags & CTL_BEARG_RD) {
2444                         tmpptr = ctl_copyin_alloc(args[i].value,
2445                                 args[i].vallen, error_str, error_str_len);
2446                         if (tmpptr == NULL)
2447                                 goto bailout;
2448                         if ((args[i].flags & CTL_BEARG_ASCII)
2449                          && (tmpptr[args[i].vallen - 1] != '\0')) {
2450                                 snprintf(error_str, error_str_len, "Argument "
2451                                     "%d value is not NUL-terminated", i);
2452                                 free(tmpptr, M_CTL);
2453                                 goto bailout;
2454                         }
2455                         args[i].kvalue = tmpptr;
2456                 } else {
2457                         args[i].kvalue = malloc(args[i].vallen,
2458                             M_CTL, M_WAITOK | M_ZERO);
2459                 }
2460         }
2461
2462         return (args);
2463 bailout:
2464
2465         ctl_free_args(num_args, args);
2466
2467         return (NULL);
2468 }
2469
2470 static void
2471 ctl_copyout_args(int num_args, struct ctl_be_arg *args)
2472 {
2473         int i;
2474
2475         for (i = 0; i < num_args; i++) {
2476                 if (args[i].flags & CTL_BEARG_WR)
2477                         copyout(args[i].kvalue, args[i].value, args[i].vallen);
2478         }
2479 }
2480
2481 /*
2482  * Escape characters that are illegal or not recommended in XML.
2483  */
2484 int
2485 ctl_sbuf_printf_esc(struct sbuf *sb, char *str, int size)
2486 {
2487         char *end = str + size;
2488         int retval;
2489
2490         retval = 0;
2491
2492         for (; *str && str < end; str++) {
2493                 switch (*str) {
2494                 case '&':
2495                         retval = sbuf_printf(sb, "&amp;");
2496                         break;
2497                 case '>':
2498                         retval = sbuf_printf(sb, "&gt;");
2499                         break;
2500                 case '<':
2501                         retval = sbuf_printf(sb, "&lt;");
2502                         break;
2503                 default:
2504                         retval = sbuf_putc(sb, *str);
2505                         break;
2506                 }
2507
2508                 if (retval != 0)
2509                         break;
2510
2511         }
2512
2513         return (retval);
2514 }
2515
2516 static void
2517 ctl_id_sbuf(struct ctl_devid *id, struct sbuf *sb)
2518 {
2519         struct scsi_vpd_id_descriptor *desc;
2520         int i;
2521
2522         if (id == NULL || id->len < 4)
2523                 return;
2524         desc = (struct scsi_vpd_id_descriptor *)id->data;
2525         switch (desc->id_type & SVPD_ID_TYPE_MASK) {
2526         case SVPD_ID_TYPE_T10:
2527                 sbuf_printf(sb, "t10.");
2528                 break;
2529         case SVPD_ID_TYPE_EUI64:
2530                 sbuf_printf(sb, "eui.");
2531                 break;
2532         case SVPD_ID_TYPE_NAA:
2533                 sbuf_printf(sb, "naa.");
2534                 break;
2535         case SVPD_ID_TYPE_SCSI_NAME:
2536                 break;
2537         }
2538         switch (desc->proto_codeset & SVPD_ID_CODESET_MASK) {
2539         case SVPD_ID_CODESET_BINARY:
2540                 for (i = 0; i < desc->length; i++)
2541                         sbuf_printf(sb, "%02x", desc->identifier[i]);
2542                 break;
2543         case SVPD_ID_CODESET_ASCII:
2544                 sbuf_printf(sb, "%.*s", (int)desc->length,
2545                     (char *)desc->identifier);
2546                 break;
2547         case SVPD_ID_CODESET_UTF8:
2548                 sbuf_printf(sb, "%s", (char *)desc->identifier);
2549                 break;
2550         }
2551 }
2552
2553 static int
2554 ctl_ioctl(struct cdev *dev, u_long cmd, caddr_t addr, int flag,
2555           struct thread *td)
2556 {
2557         struct ctl_softc *softc = dev->si_drv1;
2558         struct ctl_lun *lun;
2559         int retval;
2560
2561         retval = 0;
2562
2563         switch (cmd) {
2564         case CTL_IO:
2565                 retval = ctl_ioctl_io(dev, cmd, addr, flag, td);
2566                 break;
2567         case CTL_ENABLE_PORT:
2568         case CTL_DISABLE_PORT:
2569         case CTL_SET_PORT_WWNS: {
2570                 struct ctl_port *port;
2571                 struct ctl_port_entry *entry;
2572
2573                 entry = (struct ctl_port_entry *)addr;
2574                 
2575                 mtx_lock(&softc->ctl_lock);
2576                 STAILQ_FOREACH(port, &softc->port_list, links) {
2577                         int action, done;
2578
2579                         if (port->targ_port < softc->port_min ||
2580                             port->targ_port >= softc->port_max)
2581                                 continue;
2582
2583                         action = 0;
2584                         done = 0;
2585                         if ((entry->port_type == CTL_PORT_NONE)
2586                          && (entry->targ_port == port->targ_port)) {
2587                                 /*
2588                                  * If the user only wants to enable or
2589                                  * disable or set WWNs on a specific port,
2590                                  * do the operation and we're done.
2591                                  */
2592                                 action = 1;
2593                                 done = 1;
2594                         } else if (entry->port_type & port->port_type) {
2595                                 /*
2596                                  * Compare the user's type mask with the
2597                                  * particular frontend type to see if we
2598                                  * have a match.
2599                                  */
2600                                 action = 1;
2601                                 done = 0;
2602
2603                                 /*
2604                                  * Make sure the user isn't trying to set
2605                                  * WWNs on multiple ports at the same time.
2606                                  */
2607                                 if (cmd == CTL_SET_PORT_WWNS) {
2608                                         printf("%s: Can't set WWNs on "
2609                                                "multiple ports\n", __func__);
2610                                         retval = EINVAL;
2611                                         break;
2612                                 }
2613                         }
2614                         if (action == 0)
2615                                 continue;
2616
2617                         /*
2618                          * XXX KDM we have to drop the lock here, because
2619                          * the online/offline operations can potentially
2620                          * block.  We need to reference count the frontends
2621                          * so they can't go away,
2622                          */
2623                         if (cmd == CTL_ENABLE_PORT) {
2624                                 mtx_unlock(&softc->ctl_lock);
2625                                 ctl_port_online(port);
2626                                 mtx_lock(&softc->ctl_lock);
2627                         } else if (cmd == CTL_DISABLE_PORT) {
2628                                 mtx_unlock(&softc->ctl_lock);
2629                                 ctl_port_offline(port);
2630                                 mtx_lock(&softc->ctl_lock);
2631                         } else if (cmd == CTL_SET_PORT_WWNS) {
2632                                 ctl_port_set_wwns(port,
2633                                     (entry->flags & CTL_PORT_WWNN_VALID) ?
2634                                     1 : 0, entry->wwnn,
2635                                     (entry->flags & CTL_PORT_WWPN_VALID) ?
2636                                     1 : 0, entry->wwpn);
2637                         }
2638                         if (done != 0)
2639                                 break;
2640                 }
2641                 mtx_unlock(&softc->ctl_lock);
2642                 break;
2643         }
2644         case CTL_GET_OOA: {
2645                 struct ctl_ooa *ooa_hdr;
2646                 struct ctl_ooa_entry *entries;
2647                 uint32_t cur_fill_num;
2648
2649                 ooa_hdr = (struct ctl_ooa *)addr;
2650
2651                 if ((ooa_hdr->alloc_len == 0)
2652                  || (ooa_hdr->alloc_num == 0)) {
2653                         printf("%s: CTL_GET_OOA: alloc len %u and alloc num %u "
2654                                "must be non-zero\n", __func__,
2655                                ooa_hdr->alloc_len, ooa_hdr->alloc_num);
2656                         retval = EINVAL;
2657                         break;
2658                 }
2659
2660                 if (ooa_hdr->alloc_len != (ooa_hdr->alloc_num *
2661                     sizeof(struct ctl_ooa_entry))) {
2662                         printf("%s: CTL_GET_OOA: alloc len %u must be alloc "
2663                                "num %d * sizeof(struct ctl_ooa_entry) %zd\n",
2664                                __func__, ooa_hdr->alloc_len,
2665                                ooa_hdr->alloc_num,sizeof(struct ctl_ooa_entry));
2666                         retval = EINVAL;
2667                         break;
2668                 }
2669
2670                 entries = malloc(ooa_hdr->alloc_len, M_CTL, M_WAITOK | M_ZERO);
2671                 if (entries == NULL) {
2672                         printf("%s: could not allocate %d bytes for OOA "
2673                                "dump\n", __func__, ooa_hdr->alloc_len);
2674                         retval = ENOMEM;
2675                         break;
2676                 }
2677
2678                 mtx_lock(&softc->ctl_lock);
2679                 if (((ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) == 0)
2680                  && ((ooa_hdr->lun_num >= CTL_MAX_LUNS)
2681                   || (softc->ctl_luns[ooa_hdr->lun_num] == NULL))) {
2682                         mtx_unlock(&softc->ctl_lock);
2683                         free(entries, M_CTL);
2684                         printf("%s: CTL_GET_OOA: invalid LUN %ju\n",
2685                                __func__, (uintmax_t)ooa_hdr->lun_num);
2686                         retval = EINVAL;
2687                         break;
2688                 }
2689
2690                 cur_fill_num = 0;
2691
2692                 if (ooa_hdr->flags & CTL_OOA_FLAG_ALL_LUNS) {
2693                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
2694                                 ctl_ioctl_fill_ooa(lun, &cur_fill_num,
2695                                     ooa_hdr, entries);
2696                         }
2697                 } else {
2698                         lun = softc->ctl_luns[ooa_hdr->lun_num];
2699                         ctl_ioctl_fill_ooa(lun, &cur_fill_num, ooa_hdr,
2700                             entries);
2701                 }
2702                 mtx_unlock(&softc->ctl_lock);
2703
2704                 ooa_hdr->fill_num = min(cur_fill_num, ooa_hdr->alloc_num);
2705                 ooa_hdr->fill_len = ooa_hdr->fill_num *
2706                         sizeof(struct ctl_ooa_entry);
2707                 retval = copyout(entries, ooa_hdr->entries, ooa_hdr->fill_len);
2708                 if (retval != 0) {
2709                         printf("%s: error copying out %d bytes for OOA dump\n", 
2710                                __func__, ooa_hdr->fill_len);
2711                 }
2712
2713                 getbinuptime(&ooa_hdr->cur_bt);
2714
2715                 if (cur_fill_num > ooa_hdr->alloc_num) {
2716                         ooa_hdr->dropped_num = cur_fill_num -ooa_hdr->alloc_num;
2717                         ooa_hdr->status = CTL_OOA_NEED_MORE_SPACE;
2718                 } else {
2719                         ooa_hdr->dropped_num = 0;
2720                         ooa_hdr->status = CTL_OOA_OK;
2721                 }
2722
2723                 free(entries, M_CTL);
2724                 break;
2725         }
2726         case CTL_DELAY_IO: {
2727                 struct ctl_io_delay_info *delay_info;
2728
2729                 delay_info = (struct ctl_io_delay_info *)addr;
2730
2731 #ifdef CTL_IO_DELAY
2732                 mtx_lock(&softc->ctl_lock);
2733
2734                 if ((delay_info->lun_id >= CTL_MAX_LUNS)
2735                  || (softc->ctl_luns[delay_info->lun_id] == NULL)) {
2736                         delay_info->status = CTL_DELAY_STATUS_INVALID_LUN;
2737                 } else {
2738                         lun = softc->ctl_luns[delay_info->lun_id];
2739                         mtx_lock(&lun->lun_lock);
2740
2741                         delay_info->status = CTL_DELAY_STATUS_OK;
2742
2743                         switch (delay_info->delay_type) {
2744                         case CTL_DELAY_TYPE_CONT:
2745                                 break;
2746                         case CTL_DELAY_TYPE_ONESHOT:
2747                                 break;
2748                         default:
2749                                 delay_info->status =
2750                                         CTL_DELAY_STATUS_INVALID_TYPE;
2751                                 break;
2752                         }
2753
2754                         switch (delay_info->delay_loc) {
2755                         case CTL_DELAY_LOC_DATAMOVE:
2756                                 lun->delay_info.datamove_type =
2757                                         delay_info->delay_type;
2758                                 lun->delay_info.datamove_delay =
2759                                         delay_info->delay_secs;
2760                                 break;
2761                         case CTL_DELAY_LOC_DONE:
2762                                 lun->delay_info.done_type =
2763                                         delay_info->delay_type;
2764                                 lun->delay_info.done_delay =
2765                                         delay_info->delay_secs;
2766                                 break;
2767                         default:
2768                                 delay_info->status =
2769                                         CTL_DELAY_STATUS_INVALID_LOC;
2770                                 break;
2771                         }
2772                         mtx_unlock(&lun->lun_lock);
2773                 }
2774
2775                 mtx_unlock(&softc->ctl_lock);
2776 #else
2777                 delay_info->status = CTL_DELAY_STATUS_NOT_IMPLEMENTED;
2778 #endif /* CTL_IO_DELAY */
2779                 break;
2780         }
2781         case CTL_GETSTATS: {
2782                 struct ctl_stats *stats;
2783                 int i;
2784
2785                 stats = (struct ctl_stats *)addr;
2786
2787                 if ((sizeof(struct ctl_lun_io_stats) * softc->num_luns) >
2788                      stats->alloc_len) {
2789                         stats->status = CTL_SS_NEED_MORE_SPACE;
2790                         stats->num_luns = softc->num_luns;
2791                         break;
2792                 }
2793                 /*
2794                  * XXX KDM no locking here.  If the LUN list changes,
2795                  * things can blow up.
2796                  */
2797                 i = 0;
2798                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
2799                         retval = copyout(&lun->stats, &stats->lun_stats[i++],
2800                                          sizeof(lun->stats));
2801                         if (retval != 0)
2802                                 break;
2803                 }
2804                 stats->num_luns = softc->num_luns;
2805                 stats->fill_len = sizeof(struct ctl_lun_io_stats) *
2806                                  softc->num_luns;
2807                 stats->status = CTL_SS_OK;
2808 #ifdef CTL_TIME_IO
2809                 stats->flags = CTL_STATS_FLAG_TIME_VALID;
2810 #else
2811                 stats->flags = CTL_STATS_FLAG_NONE;
2812 #endif
2813                 getnanouptime(&stats->timestamp);
2814                 break;
2815         }
2816         case CTL_ERROR_INJECT: {
2817                 struct ctl_error_desc *err_desc, *new_err_desc;
2818
2819                 err_desc = (struct ctl_error_desc *)addr;
2820
2821                 new_err_desc = malloc(sizeof(*new_err_desc), M_CTL,
2822                                       M_WAITOK | M_ZERO);
2823                 bcopy(err_desc, new_err_desc, sizeof(*new_err_desc));
2824
2825                 mtx_lock(&softc->ctl_lock);
2826                 lun = softc->ctl_luns[err_desc->lun_id];
2827                 if (lun == NULL) {
2828                         mtx_unlock(&softc->ctl_lock);
2829                         free(new_err_desc, M_CTL);
2830                         printf("%s: CTL_ERROR_INJECT: invalid LUN %ju\n",
2831                                __func__, (uintmax_t)err_desc->lun_id);
2832                         retval = EINVAL;
2833                         break;
2834                 }
2835                 mtx_lock(&lun->lun_lock);
2836                 mtx_unlock(&softc->ctl_lock);
2837
2838                 /*
2839                  * We could do some checking here to verify the validity
2840                  * of the request, but given the complexity of error
2841                  * injection requests, the checking logic would be fairly
2842                  * complex.
2843                  *
2844                  * For now, if the request is invalid, it just won't get
2845                  * executed and might get deleted.
2846                  */
2847                 STAILQ_INSERT_TAIL(&lun->error_list, new_err_desc, links);
2848
2849                 /*
2850                  * XXX KDM check to make sure the serial number is unique,
2851                  * in case we somehow manage to wrap.  That shouldn't
2852                  * happen for a very long time, but it's the right thing to
2853                  * do.
2854                  */
2855                 new_err_desc->serial = lun->error_serial;
2856                 err_desc->serial = lun->error_serial;
2857                 lun->error_serial++;
2858
2859                 mtx_unlock(&lun->lun_lock);
2860                 break;
2861         }
2862         case CTL_ERROR_INJECT_DELETE: {
2863                 struct ctl_error_desc *delete_desc, *desc, *desc2;
2864                 int delete_done;
2865
2866                 delete_desc = (struct ctl_error_desc *)addr;
2867                 delete_done = 0;
2868
2869                 mtx_lock(&softc->ctl_lock);
2870                 lun = softc->ctl_luns[delete_desc->lun_id];
2871                 if (lun == NULL) {
2872                         mtx_unlock(&softc->ctl_lock);
2873                         printf("%s: CTL_ERROR_INJECT_DELETE: invalid LUN %ju\n",
2874                                __func__, (uintmax_t)delete_desc->lun_id);
2875                         retval = EINVAL;
2876                         break;
2877                 }
2878                 mtx_lock(&lun->lun_lock);
2879                 mtx_unlock(&softc->ctl_lock);
2880                 STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
2881                         if (desc->serial != delete_desc->serial)
2882                                 continue;
2883
2884                         STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc,
2885                                       links);
2886                         free(desc, M_CTL);
2887                         delete_done = 1;
2888                 }
2889                 mtx_unlock(&lun->lun_lock);
2890                 if (delete_done == 0) {
2891                         printf("%s: CTL_ERROR_INJECT_DELETE: can't find "
2892                                "error serial %ju on LUN %u\n", __func__, 
2893                                delete_desc->serial, delete_desc->lun_id);
2894                         retval = EINVAL;
2895                         break;
2896                 }
2897                 break;
2898         }
2899         case CTL_DUMP_STRUCTS: {
2900                 int i, j, k;
2901                 struct ctl_port *port;
2902                 struct ctl_frontend *fe;
2903
2904                 mtx_lock(&softc->ctl_lock);
2905                 printf("CTL Persistent Reservation information start:\n");
2906                 for (i = 0; i < CTL_MAX_LUNS; i++) {
2907                         lun = softc->ctl_luns[i];
2908
2909                         if ((lun == NULL)
2910                          || ((lun->flags & CTL_LUN_DISABLED) != 0))
2911                                 continue;
2912
2913                         for (j = 0; j < CTL_MAX_PORTS; j++) {
2914                                 if (lun->pr_keys[j] == NULL)
2915                                         continue;
2916                                 for (k = 0; k < CTL_MAX_INIT_PER_PORT; k++){
2917                                         if (lun->pr_keys[j][k] == 0)
2918                                                 continue;
2919                                         printf("  LUN %d port %d iid %d key "
2920                                                "%#jx\n", i, j, k,
2921                                                (uintmax_t)lun->pr_keys[j][k]);
2922                                 }
2923                         }
2924                 }
2925                 printf("CTL Persistent Reservation information end\n");
2926                 printf("CTL Ports:\n");
2927                 STAILQ_FOREACH(port, &softc->port_list, links) {
2928                         printf("  Port %d '%s' Frontend '%s' Type %u pp %d vp %d WWNN "
2929                                "%#jx WWPN %#jx\n", port->targ_port, port->port_name,
2930                                port->frontend->name, port->port_type,
2931                                port->physical_port, port->virtual_port,
2932                                (uintmax_t)port->wwnn, (uintmax_t)port->wwpn);
2933                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
2934                                 if (port->wwpn_iid[j].in_use == 0 &&
2935                                     port->wwpn_iid[j].wwpn == 0 &&
2936                                     port->wwpn_iid[j].name == NULL)
2937                                         continue;
2938
2939                                 printf("    iid %u use %d WWPN %#jx '%s'\n",
2940                                     j, port->wwpn_iid[j].in_use,
2941                                     (uintmax_t)port->wwpn_iid[j].wwpn,
2942                                     port->wwpn_iid[j].name);
2943                         }
2944                 }
2945                 printf("CTL Port information end\n");
2946                 mtx_unlock(&softc->ctl_lock);
2947                 /*
2948                  * XXX KDM calling this without a lock.  We'd likely want
2949                  * to drop the lock before calling the frontend's dump
2950                  * routine anyway.
2951                  */
2952                 printf("CTL Frontends:\n");
2953                 STAILQ_FOREACH(fe, &softc->fe_list, links) {
2954                         printf("  Frontend '%s'\n", fe->name);
2955                         if (fe->fe_dump != NULL)
2956                                 fe->fe_dump();
2957                 }
2958                 printf("CTL Frontend information end\n");
2959                 break;
2960         }
2961         case CTL_LUN_REQ: {
2962                 struct ctl_lun_req *lun_req;
2963                 struct ctl_backend_driver *backend;
2964
2965                 lun_req = (struct ctl_lun_req *)addr;
2966
2967                 backend = ctl_backend_find(lun_req->backend);
2968                 if (backend == NULL) {
2969                         lun_req->status = CTL_LUN_ERROR;
2970                         snprintf(lun_req->error_str,
2971                                  sizeof(lun_req->error_str),
2972                                  "Backend \"%s\" not found.",
2973                                  lun_req->backend);
2974                         break;
2975                 }
2976                 if (lun_req->num_be_args > 0) {
2977                         lun_req->kern_be_args = ctl_copyin_args(
2978                                 lun_req->num_be_args,
2979                                 lun_req->be_args,
2980                                 lun_req->error_str,
2981                                 sizeof(lun_req->error_str));
2982                         if (lun_req->kern_be_args == NULL) {
2983                                 lun_req->status = CTL_LUN_ERROR;
2984                                 break;
2985                         }
2986                 }
2987
2988                 retval = backend->ioctl(dev, cmd, addr, flag, td);
2989
2990                 if (lun_req->num_be_args > 0) {
2991                         ctl_copyout_args(lun_req->num_be_args,
2992                                       lun_req->kern_be_args);
2993                         ctl_free_args(lun_req->num_be_args,
2994                                       lun_req->kern_be_args);
2995                 }
2996                 break;
2997         }
2998         case CTL_LUN_LIST: {
2999                 struct sbuf *sb;
3000                 struct ctl_lun_list *list;
3001                 struct ctl_option *opt;
3002
3003                 list = (struct ctl_lun_list *)addr;
3004
3005                 /*
3006                  * Allocate a fixed length sbuf here, based on the length
3007                  * of the user's buffer.  We could allocate an auto-extending
3008                  * buffer, and then tell the user how much larger our
3009                  * amount of data is than his buffer, but that presents
3010                  * some problems:
3011                  *
3012                  * 1.  The sbuf(9) routines use a blocking malloc, and so
3013                  *     we can't hold a lock while calling them with an
3014                  *     auto-extending buffer.
3015                  *
3016                  * 2.  There is not currently a LUN reference counting
3017                  *     mechanism, outside of outstanding transactions on
3018                  *     the LUN's OOA queue.  So a LUN could go away on us
3019                  *     while we're getting the LUN number, backend-specific
3020                  *     information, etc.  Thus, given the way things
3021                  *     currently work, we need to hold the CTL lock while
3022                  *     grabbing LUN information.
3023                  *
3024                  * So, from the user's standpoint, the best thing to do is
3025                  * allocate what he thinks is a reasonable buffer length,
3026                  * and then if he gets a CTL_LUN_LIST_NEED_MORE_SPACE error,
3027                  * double the buffer length and try again.  (And repeat
3028                  * that until he succeeds.)
3029                  */
3030                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3031                 if (sb == NULL) {
3032                         list->status = CTL_LUN_LIST_ERROR;
3033                         snprintf(list->error_str, sizeof(list->error_str),
3034                                  "Unable to allocate %d bytes for LUN list",
3035                                  list->alloc_len);
3036                         break;
3037                 }
3038
3039                 sbuf_printf(sb, "<ctllunlist>\n");
3040
3041                 mtx_lock(&softc->ctl_lock);
3042                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
3043                         mtx_lock(&lun->lun_lock);
3044                         retval = sbuf_printf(sb, "<lun id=\"%ju\">\n",
3045                                              (uintmax_t)lun->lun);
3046
3047                         /*
3048                          * Bail out as soon as we see that we've overfilled
3049                          * the buffer.
3050                          */
3051                         if (retval != 0)
3052                                 break;
3053
3054                         retval = sbuf_printf(sb, "\t<backend_type>%s"
3055                                              "</backend_type>\n",
3056                                              (lun->backend == NULL) ?  "none" :
3057                                              lun->backend->name);
3058
3059                         if (retval != 0)
3060                                 break;
3061
3062                         retval = sbuf_printf(sb, "\t<lun_type>%d</lun_type>\n",
3063                                              lun->be_lun->lun_type);
3064
3065                         if (retval != 0)
3066                                 break;
3067
3068                         if (lun->backend == NULL) {
3069                                 retval = sbuf_printf(sb, "</lun>\n");
3070                                 if (retval != 0)
3071                                         break;
3072                                 continue;
3073                         }
3074
3075                         retval = sbuf_printf(sb, "\t<size>%ju</size>\n",
3076                                              (lun->be_lun->maxlba > 0) ?
3077                                              lun->be_lun->maxlba + 1 : 0);
3078
3079                         if (retval != 0)
3080                                 break;
3081
3082                         retval = sbuf_printf(sb, "\t<blocksize>%u</blocksize>\n",
3083                                              lun->be_lun->blocksize);
3084
3085                         if (retval != 0)
3086                                 break;
3087
3088                         retval = sbuf_printf(sb, "\t<serial_number>");
3089
3090                         if (retval != 0)
3091                                 break;
3092
3093                         retval = ctl_sbuf_printf_esc(sb,
3094                             lun->be_lun->serial_num,
3095                             sizeof(lun->be_lun->serial_num));
3096
3097                         if (retval != 0)
3098                                 break;
3099
3100                         retval = sbuf_printf(sb, "</serial_number>\n");
3101                 
3102                         if (retval != 0)
3103                                 break;
3104
3105                         retval = sbuf_printf(sb, "\t<device_id>");
3106
3107                         if (retval != 0)
3108                                 break;
3109
3110                         retval = ctl_sbuf_printf_esc(sb,
3111                             lun->be_lun->device_id,
3112                             sizeof(lun->be_lun->device_id));
3113
3114                         if (retval != 0)
3115                                 break;
3116
3117                         retval = sbuf_printf(sb, "</device_id>\n");
3118
3119                         if (retval != 0)
3120                                 break;
3121
3122                         if (lun->backend->lun_info != NULL) {
3123                                 retval = lun->backend->lun_info(lun->be_lun->be_lun, sb);
3124                                 if (retval != 0)
3125                                         break;
3126                         }
3127                         STAILQ_FOREACH(opt, &lun->be_lun->options, links) {
3128                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3129                                     opt->name, opt->value, opt->name);
3130                                 if (retval != 0)
3131                                         break;
3132                         }
3133
3134                         retval = sbuf_printf(sb, "</lun>\n");
3135
3136                         if (retval != 0)
3137                                 break;
3138                         mtx_unlock(&lun->lun_lock);
3139                 }
3140                 if (lun != NULL)
3141                         mtx_unlock(&lun->lun_lock);
3142                 mtx_unlock(&softc->ctl_lock);
3143
3144                 if ((retval != 0)
3145                  || ((retval = sbuf_printf(sb, "</ctllunlist>\n")) != 0)) {
3146                         retval = 0;
3147                         sbuf_delete(sb);
3148                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3149                         snprintf(list->error_str, sizeof(list->error_str),
3150                                  "Out of space, %d bytes is too small",
3151                                  list->alloc_len);
3152                         break;
3153                 }
3154
3155                 sbuf_finish(sb);
3156
3157                 retval = copyout(sbuf_data(sb), list->lun_xml,
3158                                  sbuf_len(sb) + 1);
3159
3160                 list->fill_len = sbuf_len(sb) + 1;
3161                 list->status = CTL_LUN_LIST_OK;
3162                 sbuf_delete(sb);
3163                 break;
3164         }
3165         case CTL_ISCSI: {
3166                 struct ctl_iscsi *ci;
3167                 struct ctl_frontend *fe;
3168
3169                 ci = (struct ctl_iscsi *)addr;
3170
3171                 fe = ctl_frontend_find("iscsi");
3172                 if (fe == NULL) {
3173                         ci->status = CTL_ISCSI_ERROR;
3174                         snprintf(ci->error_str, sizeof(ci->error_str),
3175                             "Frontend \"iscsi\" not found.");
3176                         break;
3177                 }
3178
3179                 retval = fe->ioctl(dev, cmd, addr, flag, td);
3180                 break;
3181         }
3182         case CTL_PORT_REQ: {
3183                 struct ctl_req *req;
3184                 struct ctl_frontend *fe;
3185
3186                 req = (struct ctl_req *)addr;
3187
3188                 fe = ctl_frontend_find(req->driver);
3189                 if (fe == NULL) {
3190                         req->status = CTL_LUN_ERROR;
3191                         snprintf(req->error_str, sizeof(req->error_str),
3192                             "Frontend \"%s\" not found.", req->driver);
3193                         break;
3194                 }
3195                 if (req->num_args > 0) {
3196                         req->kern_args = ctl_copyin_args(req->num_args,
3197                             req->args, req->error_str, sizeof(req->error_str));
3198                         if (req->kern_args == NULL) {
3199                                 req->status = CTL_LUN_ERROR;
3200                                 break;
3201                         }
3202                 }
3203
3204                 if (fe->ioctl)
3205                         retval = fe->ioctl(dev, cmd, addr, flag, td);
3206                 else
3207                         retval = ENODEV;
3208
3209                 if (req->num_args > 0) {
3210                         ctl_copyout_args(req->num_args, req->kern_args);
3211                         ctl_free_args(req->num_args, req->kern_args);
3212                 }
3213                 break;
3214         }
3215         case CTL_PORT_LIST: {
3216                 struct sbuf *sb;
3217                 struct ctl_port *port;
3218                 struct ctl_lun_list *list;
3219                 struct ctl_option *opt;
3220                 int j;
3221                 uint32_t plun;
3222
3223                 list = (struct ctl_lun_list *)addr;
3224
3225                 sb = sbuf_new(NULL, NULL, list->alloc_len, SBUF_FIXEDLEN);
3226                 if (sb == NULL) {
3227                         list->status = CTL_LUN_LIST_ERROR;
3228                         snprintf(list->error_str, sizeof(list->error_str),
3229                                  "Unable to allocate %d bytes for LUN list",
3230                                  list->alloc_len);
3231                         break;
3232                 }
3233
3234                 sbuf_printf(sb, "<ctlportlist>\n");
3235
3236                 mtx_lock(&softc->ctl_lock);
3237                 STAILQ_FOREACH(port, &softc->port_list, links) {
3238                         retval = sbuf_printf(sb, "<targ_port id=\"%ju\">\n",
3239                                              (uintmax_t)port->targ_port);
3240
3241                         /*
3242                          * Bail out as soon as we see that we've overfilled
3243                          * the buffer.
3244                          */
3245                         if (retval != 0)
3246                                 break;
3247
3248                         retval = sbuf_printf(sb, "\t<frontend_type>%s"
3249                             "</frontend_type>\n", port->frontend->name);
3250                         if (retval != 0)
3251                                 break;
3252
3253                         retval = sbuf_printf(sb, "\t<port_type>%d</port_type>\n",
3254                                              port->port_type);
3255                         if (retval != 0)
3256                                 break;
3257
3258                         retval = sbuf_printf(sb, "\t<online>%s</online>\n",
3259                             (port->status & CTL_PORT_STATUS_ONLINE) ? "YES" : "NO");
3260                         if (retval != 0)
3261                                 break;
3262
3263                         retval = sbuf_printf(sb, "\t<port_name>%s</port_name>\n",
3264                             port->port_name);
3265                         if (retval != 0)
3266                                 break;
3267
3268                         retval = sbuf_printf(sb, "\t<physical_port>%d</physical_port>\n",
3269                             port->physical_port);
3270                         if (retval != 0)
3271                                 break;
3272
3273                         retval = sbuf_printf(sb, "\t<virtual_port>%d</virtual_port>\n",
3274                             port->virtual_port);
3275                         if (retval != 0)
3276                                 break;
3277
3278                         if (port->target_devid != NULL) {
3279                                 sbuf_printf(sb, "\t<target>");
3280                                 ctl_id_sbuf(port->target_devid, sb);
3281                                 sbuf_printf(sb, "</target>\n");
3282                         }
3283
3284                         if (port->port_devid != NULL) {
3285                                 sbuf_printf(sb, "\t<port>");
3286                                 ctl_id_sbuf(port->port_devid, sb);
3287                                 sbuf_printf(sb, "</port>\n");
3288                         }
3289
3290                         if (port->port_info != NULL) {
3291                                 retval = port->port_info(port->onoff_arg, sb);
3292                                 if (retval != 0)
3293                                         break;
3294                         }
3295                         STAILQ_FOREACH(opt, &port->options, links) {
3296                                 retval = sbuf_printf(sb, "\t<%s>%s</%s>\n",
3297                                     opt->name, opt->value, opt->name);
3298                                 if (retval != 0)
3299                                         break;
3300                         }
3301
3302                         if (port->lun_map != NULL) {
3303                                 sbuf_printf(sb, "\t<lun_map>on</lun_map>\n");
3304                                 for (j = 0; j < CTL_MAX_LUNS; j++) {
3305                                         plun = ctl_lun_map_from_port(port, j);
3306                                         if (plun >= CTL_MAX_LUNS)
3307                                                 continue;
3308                                         sbuf_printf(sb,
3309                                             "\t<lun id=\"%u\">%u</lun>\n",
3310                                             j, plun);
3311                                 }
3312                         }
3313
3314                         for (j = 0; j < CTL_MAX_INIT_PER_PORT; j++) {
3315                                 if (port->wwpn_iid[j].in_use == 0 ||
3316                                     (port->wwpn_iid[j].wwpn == 0 &&
3317                                      port->wwpn_iid[j].name == NULL))
3318                                         continue;
3319
3320                                 if (port->wwpn_iid[j].name != NULL)
3321                                         retval = sbuf_printf(sb,
3322                                             "\t<initiator id=\"%u\">%s</initiator>\n",
3323                                             j, port->wwpn_iid[j].name);
3324                                 else
3325                                         retval = sbuf_printf(sb,
3326                                             "\t<initiator id=\"%u\">naa.%08jx</initiator>\n",
3327                                             j, port->wwpn_iid[j].wwpn);
3328                                 if (retval != 0)
3329                                         break;
3330                         }
3331                         if (retval != 0)
3332                                 break;
3333
3334                         retval = sbuf_printf(sb, "</targ_port>\n");
3335                         if (retval != 0)
3336                                 break;
3337                 }
3338                 mtx_unlock(&softc->ctl_lock);
3339
3340                 if ((retval != 0)
3341                  || ((retval = sbuf_printf(sb, "</ctlportlist>\n")) != 0)) {
3342                         retval = 0;
3343                         sbuf_delete(sb);
3344                         list->status = CTL_LUN_LIST_NEED_MORE_SPACE;
3345                         snprintf(list->error_str, sizeof(list->error_str),
3346                                  "Out of space, %d bytes is too small",
3347                                  list->alloc_len);
3348                         break;
3349                 }
3350
3351                 sbuf_finish(sb);
3352
3353                 retval = copyout(sbuf_data(sb), list->lun_xml,
3354                                  sbuf_len(sb) + 1);
3355
3356                 list->fill_len = sbuf_len(sb) + 1;
3357                 list->status = CTL_LUN_LIST_OK;
3358                 sbuf_delete(sb);
3359                 break;
3360         }
3361         case CTL_LUN_MAP: {
3362                 struct ctl_lun_map *lm  = (struct ctl_lun_map *)addr;
3363                 struct ctl_port *port;
3364
3365                 mtx_lock(&softc->ctl_lock);
3366                 if (lm->port < softc->port_min ||
3367                     lm->port >= softc->port_max ||
3368                     (port = softc->ctl_ports[lm->port]) == NULL) {
3369                         mtx_unlock(&softc->ctl_lock);
3370                         return (ENXIO);
3371                 }
3372                 if (port->status & CTL_PORT_STATUS_ONLINE) {
3373                         STAILQ_FOREACH(lun, &softc->lun_list, links) {
3374                                 if (ctl_lun_map_to_port(port, lun->lun) >=
3375                                     CTL_MAX_LUNS)
3376                                         continue;
3377                                 mtx_lock(&lun->lun_lock);
3378                                 ctl_est_ua_port(lun, lm->port, -1,
3379                                     CTL_UA_LUN_CHANGE);
3380                                 mtx_unlock(&lun->lun_lock);
3381                         }
3382                 }
3383                 mtx_unlock(&softc->ctl_lock); // XXX: port_enable sleeps
3384                 if (lm->plun < CTL_MAX_LUNS) {
3385                         if (lm->lun == UINT32_MAX)
3386                                 retval = ctl_lun_map_unset(port, lm->plun);
3387                         else if (lm->lun < CTL_MAX_LUNS &&
3388                             softc->ctl_luns[lm->lun] != NULL)
3389                                 retval = ctl_lun_map_set(port, lm->plun, lm->lun);
3390                         else
3391                                 return (ENXIO);
3392                 } else if (lm->plun == UINT32_MAX) {
3393                         if (lm->lun == UINT32_MAX)
3394                                 retval = ctl_lun_map_deinit(port);
3395                         else
3396                                 retval = ctl_lun_map_init(port);
3397                 } else
3398                         return (ENXIO);
3399                 if (port->status & CTL_PORT_STATUS_ONLINE)
3400                         ctl_isc_announce_port(port);
3401                 break;
3402         }
3403         default: {
3404                 /* XXX KDM should we fix this? */
3405 #if 0
3406                 struct ctl_backend_driver *backend;
3407                 unsigned int type;
3408                 int found;
3409
3410                 found = 0;
3411
3412                 /*
3413                  * We encode the backend type as the ioctl type for backend
3414                  * ioctls.  So parse it out here, and then search for a
3415                  * backend of this type.
3416                  */
3417                 type = _IOC_TYPE(cmd);
3418
3419                 STAILQ_FOREACH(backend, &softc->be_list, links) {
3420                         if (backend->type == type) {
3421                                 found = 1;
3422                                 break;
3423                         }
3424                 }
3425                 if (found == 0) {
3426                         printf("ctl: unknown ioctl command %#lx or backend "
3427                                "%d\n", cmd, type);
3428                         retval = EINVAL;
3429                         break;
3430                 }
3431                 retval = backend->ioctl(dev, cmd, addr, flag, td);
3432 #endif
3433                 retval = ENOTTY;
3434                 break;
3435         }
3436         }
3437         return (retval);
3438 }
3439
3440 uint32_t
3441 ctl_get_initindex(struct ctl_nexus *nexus)
3442 {
3443         return (nexus->initid + (nexus->targ_port * CTL_MAX_INIT_PER_PORT));
3444 }
3445
3446 int
3447 ctl_lun_map_init(struct ctl_port *port)
3448 {
3449         struct ctl_softc *softc = port->ctl_softc;
3450         struct ctl_lun *lun;
3451         uint32_t i;
3452
3453         if (port->lun_map == NULL)
3454                 port->lun_map = malloc(sizeof(uint32_t) * CTL_MAX_LUNS,
3455                     M_CTL, M_NOWAIT);
3456         if (port->lun_map == NULL)
3457                 return (ENOMEM);
3458         for (i = 0; i < CTL_MAX_LUNS; i++)
3459                 port->lun_map[i] = UINT32_MAX;
3460         if (port->status & CTL_PORT_STATUS_ONLINE) {
3461                 if (port->lun_disable != NULL) {
3462                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3463                                 port->lun_disable(port->targ_lun_arg, lun->lun);
3464                 }
3465                 ctl_isc_announce_port(port);
3466         }
3467         return (0);
3468 }
3469
3470 int
3471 ctl_lun_map_deinit(struct ctl_port *port)
3472 {
3473         struct ctl_softc *softc = port->ctl_softc;
3474         struct ctl_lun *lun;
3475
3476         if (port->lun_map == NULL)
3477                 return (0);
3478         free(port->lun_map, M_CTL);
3479         port->lun_map = NULL;
3480         if (port->status & CTL_PORT_STATUS_ONLINE) {
3481                 if (port->lun_enable != NULL) {
3482                         STAILQ_FOREACH(lun, &softc->lun_list, links)
3483                                 port->lun_enable(port->targ_lun_arg, lun->lun);
3484                 }
3485                 ctl_isc_announce_port(port);
3486         }
3487         return (0);
3488 }
3489
3490 int
3491 ctl_lun_map_set(struct ctl_port *port, uint32_t plun, uint32_t glun)
3492 {
3493         int status;
3494         uint32_t old;
3495
3496         if (port->lun_map == NULL) {
3497                 status = ctl_lun_map_init(port);
3498                 if (status != 0)
3499                         return (status);
3500         }
3501         old = port->lun_map[plun];
3502         port->lun_map[plun] = glun;
3503         if ((port->status & CTL_PORT_STATUS_ONLINE) && old >= CTL_MAX_LUNS) {
3504                 if (port->lun_enable != NULL)
3505                         port->lun_enable(port->targ_lun_arg, plun);
3506                 ctl_isc_announce_port(port);
3507         }
3508         return (0);
3509 }
3510
3511 int
3512 ctl_lun_map_unset(struct ctl_port *port, uint32_t plun)
3513 {
3514         uint32_t old;
3515
3516         if (port->lun_map == NULL)
3517                 return (0);
3518         old = port->lun_map[plun];
3519         port->lun_map[plun] = UINT32_MAX;
3520         if ((port->status & CTL_PORT_STATUS_ONLINE) && old < CTL_MAX_LUNS) {
3521                 if (port->lun_disable != NULL)
3522                         port->lun_disable(port->targ_lun_arg, plun);
3523                 ctl_isc_announce_port(port);
3524         }
3525         return (0);
3526 }
3527
3528 uint32_t
3529 ctl_lun_map_from_port(struct ctl_port *port, uint32_t lun_id)
3530 {
3531
3532         if (port == NULL)
3533                 return (UINT32_MAX);
3534         if (port->lun_map == NULL || lun_id >= CTL_MAX_LUNS)
3535                 return (lun_id);
3536         return (port->lun_map[lun_id]);
3537 }
3538
3539 uint32_t
3540 ctl_lun_map_to_port(struct ctl_port *port, uint32_t lun_id)
3541 {
3542         uint32_t i;
3543
3544         if (port == NULL)
3545                 return (UINT32_MAX);
3546         if (port->lun_map == NULL)
3547                 return (lun_id);
3548         for (i = 0; i < CTL_MAX_LUNS; i++) {
3549                 if (port->lun_map[i] == lun_id)
3550                         return (i);
3551         }
3552         return (UINT32_MAX);
3553 }
3554
3555 uint32_t
3556 ctl_decode_lun(uint64_t encoded)
3557 {
3558         uint8_t lun[8];
3559         uint32_t result = 0xffffffff;
3560
3561         be64enc(lun, encoded);
3562         switch (lun[0] & RPL_LUNDATA_ATYP_MASK) {
3563         case RPL_LUNDATA_ATYP_PERIPH:
3564                 if ((lun[0] & 0x3f) == 0 && lun[2] == 0 && lun[3] == 0 &&
3565                     lun[4] == 0 && lun[5] == 0 && lun[6] == 0 && lun[7] == 0)
3566                         result = lun[1];
3567                 break;
3568         case RPL_LUNDATA_ATYP_FLAT:
3569                 if (lun[2] == 0 && lun[3] == 0 && lun[4] == 0 && lun[5] == 0 &&
3570                     lun[6] == 0 && lun[7] == 0)
3571                         result = ((lun[0] & 0x3f) << 8) + lun[1];
3572                 break;
3573         case RPL_LUNDATA_ATYP_EXTLUN:
3574                 switch (lun[0] & RPL_LUNDATA_EXT_EAM_MASK) {
3575                 case 0x02:
3576                         switch (lun[0] & RPL_LUNDATA_EXT_LEN_MASK) {
3577                         case 0x00:
3578                                 result = lun[1];
3579                                 break;
3580                         case 0x10:
3581                                 result = (lun[1] << 16) + (lun[2] << 8) +
3582                                     lun[3];
3583                                 break;
3584                         case 0x20:
3585                                 if (lun[1] == 0 && lun[6] == 0 && lun[7] == 0)
3586                                         result = (lun[2] << 24) +
3587                                             (lun[3] << 16) + (lun[4] << 8) +
3588                                             lun[5];
3589                                 break;
3590                         }
3591                         break;
3592                 case RPL_LUNDATA_EXT_EAM_NOT_SPEC:
3593                         result = 0xffffffff;
3594                         break;
3595                 }
3596                 break;
3597         }
3598         return (result);
3599 }
3600
3601 uint64_t
3602 ctl_encode_lun(uint32_t decoded)
3603 {
3604         uint64_t l = decoded;
3605
3606         if (l <= 0xff)
3607                 return (((uint64_t)RPL_LUNDATA_ATYP_PERIPH << 56) | (l << 48));
3608         if (l <= 0x3fff)
3609                 return (((uint64_t)RPL_LUNDATA_ATYP_FLAT << 56) | (l << 48));
3610         if (l <= 0xffffff)
3611                 return (((uint64_t)(RPL_LUNDATA_ATYP_EXTLUN | 0x12) << 56) |
3612                     (l << 32));
3613         return ((((uint64_t)RPL_LUNDATA_ATYP_EXTLUN | 0x22) << 56) | (l << 16));
3614 }
3615
3616 static struct ctl_port *
3617 ctl_io_port(struct ctl_io_hdr *io_hdr)
3618 {
3619
3620         return (control_softc->ctl_ports[io_hdr->nexus.targ_port]);
3621 }
3622
3623 int
3624 ctl_ffz(uint32_t *mask, uint32_t first, uint32_t last)
3625 {
3626         int i;
3627
3628         for (i = first; i < last; i++) {
3629                 if ((mask[i / 32] & (1 << (i % 32))) == 0)
3630                         return (i);
3631         }
3632         return (-1);
3633 }
3634
3635 int
3636 ctl_set_mask(uint32_t *mask, uint32_t bit)
3637 {
3638         uint32_t chunk, piece;
3639
3640         chunk = bit >> 5;
3641         piece = bit % (sizeof(uint32_t) * 8);
3642
3643         if ((mask[chunk] & (1 << piece)) != 0)
3644                 return (-1);
3645         else
3646                 mask[chunk] |= (1 << piece);
3647
3648         return (0);
3649 }
3650
3651 int
3652 ctl_clear_mask(uint32_t *mask, uint32_t bit)
3653 {
3654         uint32_t chunk, piece;
3655
3656         chunk = bit >> 5;
3657         piece = bit % (sizeof(uint32_t) * 8);
3658
3659         if ((mask[chunk] & (1 << piece)) == 0)
3660                 return (-1);
3661         else
3662                 mask[chunk] &= ~(1 << piece);
3663
3664         return (0);
3665 }
3666
3667 int
3668 ctl_is_set(uint32_t *mask, uint32_t bit)
3669 {
3670         uint32_t chunk, piece;
3671
3672         chunk = bit >> 5;
3673         piece = bit % (sizeof(uint32_t) * 8);
3674
3675         if ((mask[chunk] & (1 << piece)) == 0)
3676                 return (0);
3677         else
3678                 return (1);
3679 }
3680
3681 static uint64_t
3682 ctl_get_prkey(struct ctl_lun *lun, uint32_t residx)
3683 {
3684         uint64_t *t;
3685
3686         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3687         if (t == NULL)
3688                 return (0);
3689         return (t[residx % CTL_MAX_INIT_PER_PORT]);
3690 }
3691
3692 static void
3693 ctl_clr_prkey(struct ctl_lun *lun, uint32_t residx)
3694 {
3695         uint64_t *t;
3696
3697         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3698         if (t == NULL)
3699                 return;
3700         t[residx % CTL_MAX_INIT_PER_PORT] = 0;
3701 }
3702
3703 static void
3704 ctl_alloc_prkey(struct ctl_lun *lun, uint32_t residx)
3705 {
3706         uint64_t *p;
3707         u_int i;
3708
3709         i = residx/CTL_MAX_INIT_PER_PORT;
3710         if (lun->pr_keys[i] != NULL)
3711                 return;
3712         mtx_unlock(&lun->lun_lock);
3713         p = malloc(sizeof(uint64_t) * CTL_MAX_INIT_PER_PORT, M_CTL,
3714             M_WAITOK | M_ZERO);
3715         mtx_lock(&lun->lun_lock);
3716         if (lun->pr_keys[i] == NULL)
3717                 lun->pr_keys[i] = p;
3718         else
3719                 free(p, M_CTL);
3720 }
3721
3722 static void
3723 ctl_set_prkey(struct ctl_lun *lun, uint32_t residx, uint64_t key)
3724 {
3725         uint64_t *t;
3726
3727         t = lun->pr_keys[residx/CTL_MAX_INIT_PER_PORT];
3728         KASSERT(t != NULL, ("prkey %d is not allocated", residx));
3729         t[residx % CTL_MAX_INIT_PER_PORT] = key;
3730 }
3731
3732 /*
3733  * ctl_softc, pool_name, total_ctl_io are passed in.
3734  * npool is passed out.
3735  */
3736 int
3737 ctl_pool_create(struct ctl_softc *ctl_softc, const char *pool_name,
3738                 uint32_t total_ctl_io, void **npool)
3739 {
3740 #ifdef IO_POOLS
3741         struct ctl_io_pool *pool;
3742
3743         pool = (struct ctl_io_pool *)malloc(sizeof(*pool), M_CTL,
3744                                             M_NOWAIT | M_ZERO);
3745         if (pool == NULL)
3746                 return (ENOMEM);
3747
3748         snprintf(pool->name, sizeof(pool->name), "CTL IO %s", pool_name);
3749         pool->ctl_softc = ctl_softc;
3750         pool->zone = uma_zsecond_create(pool->name, NULL,
3751             NULL, NULL, NULL, ctl_softc->io_zone);
3752         /* uma_prealloc(pool->zone, total_ctl_io); */
3753
3754         *npool = pool;
3755 #else
3756         *npool = ctl_softc->io_zone;
3757 #endif
3758         return (0);
3759 }
3760
3761 void
3762 ctl_pool_free(struct ctl_io_pool *pool)
3763 {
3764
3765         if (pool == NULL)
3766                 return;
3767
3768 #ifdef IO_POOLS
3769         uma_zdestroy(pool->zone);
3770         free(pool, M_CTL);
3771 #endif
3772 }
3773
3774 union ctl_io *
3775 ctl_alloc_io(void *pool_ref)
3776 {
3777         union ctl_io *io;
3778 #ifdef IO_POOLS
3779         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3780
3781         io = uma_zalloc(pool->zone, M_WAITOK);
3782 #else
3783         io = uma_zalloc((uma_zone_t)pool_ref, M_WAITOK);
3784 #endif
3785         if (io != NULL)
3786                 io->io_hdr.pool = pool_ref;
3787         return (io);
3788 }
3789
3790 union ctl_io *
3791 ctl_alloc_io_nowait(void *pool_ref)
3792 {
3793         union ctl_io *io;
3794 #ifdef IO_POOLS
3795         struct ctl_io_pool *pool = (struct ctl_io_pool *)pool_ref;
3796
3797         io = uma_zalloc(pool->zone, M_NOWAIT);
3798 #else
3799         io = uma_zalloc((uma_zone_t)pool_ref, M_NOWAIT);
3800 #endif
3801         if (io != NULL)
3802                 io->io_hdr.pool = pool_ref;
3803         return (io);
3804 }
3805
3806 void
3807 ctl_free_io(union ctl_io *io)
3808 {
3809 #ifdef IO_POOLS
3810         struct ctl_io_pool *pool;
3811 #endif
3812
3813         if (io == NULL)
3814                 return;
3815
3816 #ifdef IO_POOLS
3817         pool = (struct ctl_io_pool *)io->io_hdr.pool;
3818         uma_zfree(pool->zone, io);
3819 #else
3820         uma_zfree((uma_zone_t)io->io_hdr.pool, io);
3821 #endif
3822 }
3823
3824 void
3825 ctl_zero_io(union ctl_io *io)
3826 {
3827         void *pool_ref;
3828
3829         if (io == NULL)
3830                 return;
3831
3832         /*
3833          * May need to preserve linked list pointers at some point too.
3834          */
3835         pool_ref = io->io_hdr.pool;
3836         memset(io, 0, sizeof(*io));
3837         io->io_hdr.pool = pool_ref;
3838 }
3839
3840 int
3841 ctl_expand_number(const char *buf, uint64_t *num)
3842 {
3843         char *endptr;
3844         uint64_t number;
3845         unsigned shift;
3846
3847         number = strtoq(buf, &endptr, 0);
3848
3849         switch (tolower((unsigned char)*endptr)) {
3850         case 'e':
3851                 shift = 60;
3852                 break;
3853         case 'p':
3854                 shift = 50;
3855                 break;
3856         case 't':
3857                 shift = 40;
3858                 break;
3859         case 'g':
3860                 shift = 30;
3861                 break;
3862         case 'm':
3863                 shift = 20;
3864                 break;
3865         case 'k':
3866                 shift = 10;
3867                 break;
3868         case 'b':
3869         case '\0': /* No unit. */
3870                 *num = number;
3871                 return (0);
3872         default:
3873                 /* Unrecognized unit. */
3874                 return (-1);
3875         }
3876
3877         if ((number << shift) >> shift != number) {
3878                 /* Overflow */
3879                 return (-1);
3880         }
3881         *num = number << shift;
3882         return (0);
3883 }
3884
3885
3886 /*
3887  * This routine could be used in the future to load default and/or saved
3888  * mode page parameters for a particuar lun.
3889  */
3890 static int
3891 ctl_init_page_index(struct ctl_lun *lun)
3892 {
3893         int i, page_code;
3894         struct ctl_page_index *page_index;
3895         const char *value;
3896         uint64_t ival;
3897
3898         memcpy(&lun->mode_pages.index, page_index_template,
3899                sizeof(page_index_template));
3900
3901         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
3902
3903                 page_index = &lun->mode_pages.index[i];
3904                 if (lun->be_lun->lun_type == T_DIRECT &&
3905                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
3906                         continue;
3907                 if (lun->be_lun->lun_type == T_PROCESSOR &&
3908                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
3909                         continue;
3910                 if (lun->be_lun->lun_type == T_CDROM &&
3911                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
3912                         continue;
3913
3914                 page_code = page_index->page_code & SMPH_PC_MASK;
3915                 switch (page_code) {
3916                 case SMS_RW_ERROR_RECOVERY_PAGE: {
3917                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3918                             ("subpage %#x for page %#x is incorrect!",
3919                             page_index->subpage, page_code));
3920                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT],
3921                                &rw_er_page_default,
3922                                sizeof(rw_er_page_default));
3923                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_CHANGEABLE],
3924                                &rw_er_page_changeable,
3925                                sizeof(rw_er_page_changeable));
3926                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_DEFAULT],
3927                                &rw_er_page_default,
3928                                sizeof(rw_er_page_default));
3929                         memcpy(&lun->mode_pages.rw_er_page[CTL_PAGE_SAVED],
3930                                &rw_er_page_default,
3931                                sizeof(rw_er_page_default));
3932                         page_index->page_data =
3933                                 (uint8_t *)lun->mode_pages.rw_er_page;
3934                         break;
3935                 }
3936                 case SMS_FORMAT_DEVICE_PAGE: {
3937                         struct scsi_format_page *format_page;
3938
3939                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3940                             ("subpage %#x for page %#x is incorrect!",
3941                             page_index->subpage, page_code));
3942
3943                         /*
3944                          * Sectors per track are set above.  Bytes per
3945                          * sector need to be set here on a per-LUN basis.
3946                          */
3947                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_CURRENT],
3948                                &format_page_default,
3949                                sizeof(format_page_default));
3950                         memcpy(&lun->mode_pages.format_page[
3951                                CTL_PAGE_CHANGEABLE], &format_page_changeable,
3952                                sizeof(format_page_changeable));
3953                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_DEFAULT],
3954                                &format_page_default,
3955                                sizeof(format_page_default));
3956                         memcpy(&lun->mode_pages.format_page[CTL_PAGE_SAVED],
3957                                &format_page_default,
3958                                sizeof(format_page_default));
3959
3960                         format_page = &lun->mode_pages.format_page[
3961                                 CTL_PAGE_CURRENT];
3962                         scsi_ulto2b(lun->be_lun->blocksize,
3963                                     format_page->bytes_per_sector);
3964
3965                         format_page = &lun->mode_pages.format_page[
3966                                 CTL_PAGE_DEFAULT];
3967                         scsi_ulto2b(lun->be_lun->blocksize,
3968                                     format_page->bytes_per_sector);
3969
3970                         format_page = &lun->mode_pages.format_page[
3971                                 CTL_PAGE_SAVED];
3972                         scsi_ulto2b(lun->be_lun->blocksize,
3973                                     format_page->bytes_per_sector);
3974
3975                         page_index->page_data =
3976                                 (uint8_t *)lun->mode_pages.format_page;
3977                         break;
3978                 }
3979                 case SMS_RIGID_DISK_PAGE: {
3980                         struct scsi_rigid_disk_page *rigid_disk_page;
3981                         uint32_t sectors_per_cylinder;
3982                         uint64_t cylinders;
3983 #ifndef __XSCALE__
3984                         int shift;
3985 #endif /* !__XSCALE__ */
3986
3987                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
3988                             ("subpage %#x for page %#x is incorrect!",
3989                             page_index->subpage, page_code));
3990
3991                         /*
3992                          * Rotation rate and sectors per track are set
3993                          * above.  We calculate the cylinders here based on
3994                          * capacity.  Due to the number of heads and
3995                          * sectors per track we're using, smaller arrays
3996                          * may turn out to have 0 cylinders.  Linux and
3997                          * FreeBSD don't pay attention to these mode pages
3998                          * to figure out capacity, but Solaris does.  It
3999                          * seems to deal with 0 cylinders just fine, and
4000                          * works out a fake geometry based on the capacity.
4001                          */
4002                         memcpy(&lun->mode_pages.rigid_disk_page[
4003                                CTL_PAGE_DEFAULT], &rigid_disk_page_default,
4004                                sizeof(rigid_disk_page_default));
4005                         memcpy(&lun->mode_pages.rigid_disk_page[
4006                                CTL_PAGE_CHANGEABLE],&rigid_disk_page_changeable,
4007                                sizeof(rigid_disk_page_changeable));
4008
4009                         sectors_per_cylinder = CTL_DEFAULT_SECTORS_PER_TRACK *
4010                                 CTL_DEFAULT_HEADS;
4011
4012                         /*
4013                          * The divide method here will be more accurate,
4014                          * probably, but results in floating point being
4015                          * used in the kernel on i386 (__udivdi3()).  On the
4016                          * XScale, though, __udivdi3() is implemented in
4017                          * software.
4018                          *
4019                          * The shift method for cylinder calculation is
4020                          * accurate if sectors_per_cylinder is a power of
4021                          * 2.  Otherwise it might be slightly off -- you
4022                          * might have a bit of a truncation problem.
4023                          */
4024 #ifdef  __XSCALE__
4025                         cylinders = (lun->be_lun->maxlba + 1) /
4026                                 sectors_per_cylinder;
4027 #else
4028                         for (shift = 31; shift > 0; shift--) {
4029                                 if (sectors_per_cylinder & (1 << shift))
4030                                         break;
4031                         }
4032                         cylinders = (lun->be_lun->maxlba + 1) >> shift;
4033 #endif
4034
4035                         /*
4036                          * We've basically got 3 bytes, or 24 bits for the
4037                          * cylinder size in the mode page.  If we're over,
4038                          * just round down to 2^24.
4039                          */
4040                         if (cylinders > 0xffffff)
4041                                 cylinders = 0xffffff;
4042
4043                         rigid_disk_page = &lun->mode_pages.rigid_disk_page[
4044                                 CTL_PAGE_DEFAULT];
4045                         scsi_ulto3b(cylinders, rigid_disk_page->cylinders);
4046
4047                         if ((value = ctl_get_opt(&lun->be_lun->options,
4048                             "rpm")) != NULL) {
4049                                 scsi_ulto2b(strtol(value, NULL, 0),
4050                                      rigid_disk_page->rotation_rate);
4051                         }
4052
4053                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_CURRENT],
4054                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4055                                sizeof(rigid_disk_page_default));
4056                         memcpy(&lun->mode_pages.rigid_disk_page[CTL_PAGE_SAVED],
4057                                &lun->mode_pages.rigid_disk_page[CTL_PAGE_DEFAULT],
4058                                sizeof(rigid_disk_page_default));
4059
4060                         page_index->page_data =
4061                                 (uint8_t *)lun->mode_pages.rigid_disk_page;
4062                         break;
4063                 }
4064                 case SMS_VERIFY_ERROR_RECOVERY_PAGE: {
4065                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4066                             ("subpage %#x for page %#x is incorrect!",
4067                             page_index->subpage, page_code));
4068                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT],
4069                                &verify_er_page_default,
4070                                sizeof(verify_er_page_default));
4071                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_CHANGEABLE],
4072                                &verify_er_page_changeable,
4073                                sizeof(verify_er_page_changeable));
4074                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_DEFAULT],
4075                                &verify_er_page_default,
4076                                sizeof(verify_er_page_default));
4077                         memcpy(&lun->mode_pages.verify_er_page[CTL_PAGE_SAVED],
4078                                &verify_er_page_default,
4079                                sizeof(verify_er_page_default));
4080                         page_index->page_data =
4081                                 (uint8_t *)lun->mode_pages.verify_er_page;
4082                         break;
4083                 }
4084                 case SMS_CACHING_PAGE: {
4085                         struct scsi_caching_page *caching_page;
4086
4087                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4088                             ("subpage %#x for page %#x is incorrect!",
4089                             page_index->subpage, page_code));
4090                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_DEFAULT],
4091                                &caching_page_default,
4092                                sizeof(caching_page_default));
4093                         memcpy(&lun->mode_pages.caching_page[
4094                                CTL_PAGE_CHANGEABLE], &caching_page_changeable,
4095                                sizeof(caching_page_changeable));
4096                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4097                                &caching_page_default,
4098                                sizeof(caching_page_default));
4099                         caching_page = &lun->mode_pages.caching_page[
4100                             CTL_PAGE_SAVED];
4101                         value = ctl_get_opt(&lun->be_lun->options, "writecache");
4102                         if (value != NULL && strcmp(value, "off") == 0)
4103                                 caching_page->flags1 &= ~SCP_WCE;
4104                         value = ctl_get_opt(&lun->be_lun->options, "readcache");
4105                         if (value != NULL && strcmp(value, "off") == 0)
4106                                 caching_page->flags1 |= SCP_RCD;
4107                         memcpy(&lun->mode_pages.caching_page[CTL_PAGE_CURRENT],
4108                                &lun->mode_pages.caching_page[CTL_PAGE_SAVED],
4109                                sizeof(caching_page_default));
4110                         page_index->page_data =
4111                                 (uint8_t *)lun->mode_pages.caching_page;
4112                         break;
4113                 }
4114                 case SMS_CONTROL_MODE_PAGE: {
4115                         switch (page_index->subpage) {
4116                         case SMS_SUBPAGE_PAGE_0: {
4117                                 struct scsi_control_page *control_page;
4118
4119                                 memcpy(&lun->mode_pages.control_page[
4120                                     CTL_PAGE_DEFAULT],
4121                                        &control_page_default,
4122                                        sizeof(control_page_default));
4123                                 memcpy(&lun->mode_pages.control_page[
4124                                     CTL_PAGE_CHANGEABLE],
4125                                        &control_page_changeable,
4126                                        sizeof(control_page_changeable));
4127                                 memcpy(&lun->mode_pages.control_page[
4128                                     CTL_PAGE_SAVED],
4129                                        &control_page_default,
4130                                        sizeof(control_page_default));
4131                                 control_page = &lun->mode_pages.control_page[
4132                                     CTL_PAGE_SAVED];
4133                                 value = ctl_get_opt(&lun->be_lun->options,
4134                                     "reordering");
4135                                 if (value != NULL &&
4136                                     strcmp(value, "unrestricted") == 0) {
4137                                         control_page->queue_flags &=
4138                                             ~SCP_QUEUE_ALG_MASK;
4139                                         control_page->queue_flags |=
4140                                             SCP_QUEUE_ALG_UNRESTRICTED;
4141                                 }
4142                                 memcpy(&lun->mode_pages.control_page[
4143                                     CTL_PAGE_CURRENT],
4144                                        &lun->mode_pages.control_page[
4145                                     CTL_PAGE_SAVED],
4146                                        sizeof(control_page_default));
4147                                 page_index->page_data =
4148                                     (uint8_t *)lun->mode_pages.control_page;
4149                                 break;
4150                         }
4151                         case 0x01:
4152                                 memcpy(&lun->mode_pages.control_ext_page[
4153                                     CTL_PAGE_DEFAULT],
4154                                        &control_ext_page_default,
4155                                        sizeof(control_ext_page_default));
4156                                 memcpy(&lun->mode_pages.control_ext_page[
4157                                     CTL_PAGE_CHANGEABLE],
4158                                        &control_ext_page_changeable,
4159                                        sizeof(control_ext_page_changeable));
4160                                 memcpy(&lun->mode_pages.control_ext_page[
4161                                     CTL_PAGE_SAVED],
4162                                        &control_ext_page_default,
4163                                        sizeof(control_ext_page_default));
4164                                 memcpy(&lun->mode_pages.control_ext_page[
4165                                     CTL_PAGE_CURRENT],
4166                                        &lun->mode_pages.control_ext_page[
4167                                     CTL_PAGE_SAVED],
4168                                        sizeof(control_ext_page_default));
4169                                 page_index->page_data =
4170                                     (uint8_t *)lun->mode_pages.control_ext_page;
4171                                 break;
4172                         default:
4173                                 panic("subpage %#x for page %#x is incorrect!",
4174                                       page_index->subpage, page_code);
4175                         }
4176                         break;
4177                 }
4178                 case SMS_INFO_EXCEPTIONS_PAGE: {
4179                         switch (page_index->subpage) {
4180                         case SMS_SUBPAGE_PAGE_0:
4181                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_CURRENT],
4182                                        &ie_page_default,
4183                                        sizeof(ie_page_default));
4184                                 memcpy(&lun->mode_pages.ie_page[
4185                                        CTL_PAGE_CHANGEABLE], &ie_page_changeable,
4186                                        sizeof(ie_page_changeable));
4187                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_DEFAULT],
4188                                        &ie_page_default,
4189                                        sizeof(ie_page_default));
4190                                 memcpy(&lun->mode_pages.ie_page[CTL_PAGE_SAVED],
4191                                        &ie_page_default,
4192                                        sizeof(ie_page_default));
4193                                 page_index->page_data =
4194                                         (uint8_t *)lun->mode_pages.ie_page;
4195                                 break;
4196                         case 0x02: {
4197                                 struct ctl_logical_block_provisioning_page *page;
4198
4199                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_DEFAULT],
4200                                        &lbp_page_default,
4201                                        sizeof(lbp_page_default));
4202                                 memcpy(&lun->mode_pages.lbp_page[
4203                                        CTL_PAGE_CHANGEABLE], &lbp_page_changeable,
4204                                        sizeof(lbp_page_changeable));
4205                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4206                                        &lbp_page_default,
4207                                        sizeof(lbp_page_default));
4208                                 page = &lun->mode_pages.lbp_page[CTL_PAGE_SAVED];
4209                                 value = ctl_get_opt(&lun->be_lun->options,
4210                                     "avail-threshold");
4211                                 if (value != NULL &&
4212                                     ctl_expand_number(value, &ival) == 0) {
4213                                         page->descr[0].flags |= SLBPPD_ENABLED |
4214                                             SLBPPD_ARMING_DEC;
4215                                         if (lun->be_lun->blocksize)
4216                                                 ival /= lun->be_lun->blocksize;
4217                                         else
4218                                                 ival /= 512;
4219                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4220                                             page->descr[0].count);
4221                                 }
4222                                 value = ctl_get_opt(&lun->be_lun->options,
4223                                     "used-threshold");
4224                                 if (value != NULL &&
4225                                     ctl_expand_number(value, &ival) == 0) {
4226                                         page->descr[1].flags |= SLBPPD_ENABLED |
4227                                             SLBPPD_ARMING_INC;
4228                                         if (lun->be_lun->blocksize)
4229                                                 ival /= lun->be_lun->blocksize;
4230                                         else
4231                                                 ival /= 512;
4232                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4233                                             page->descr[1].count);
4234                                 }
4235                                 value = ctl_get_opt(&lun->be_lun->options,
4236                                     "pool-avail-threshold");
4237                                 if (value != NULL &&
4238                                     ctl_expand_number(value, &ival) == 0) {
4239                                         page->descr[2].flags |= SLBPPD_ENABLED |
4240                                             SLBPPD_ARMING_DEC;
4241                                         if (lun->be_lun->blocksize)
4242                                                 ival /= lun->be_lun->blocksize;
4243                                         else
4244                                                 ival /= 512;
4245                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4246                                             page->descr[2].count);
4247                                 }
4248                                 value = ctl_get_opt(&lun->be_lun->options,
4249                                     "pool-used-threshold");
4250                                 if (value != NULL &&
4251                                     ctl_expand_number(value, &ival) == 0) {
4252                                         page->descr[3].flags |= SLBPPD_ENABLED |
4253                                             SLBPPD_ARMING_INC;
4254                                         if (lun->be_lun->blocksize)
4255                                                 ival /= lun->be_lun->blocksize;
4256                                         else
4257                                                 ival /= 512;
4258                                         scsi_ulto4b(ival >> CTL_LBP_EXPONENT,
4259                                             page->descr[3].count);
4260                                 }
4261                                 memcpy(&lun->mode_pages.lbp_page[CTL_PAGE_CURRENT],
4262                                        &lun->mode_pages.lbp_page[CTL_PAGE_SAVED],
4263                                        sizeof(lbp_page_default));
4264                                 page_index->page_data =
4265                                         (uint8_t *)lun->mode_pages.lbp_page;
4266                                 break;
4267                         }
4268                         default:
4269                                 panic("subpage %#x for page %#x is incorrect!",
4270                                       page_index->subpage, page_code);
4271                         }
4272                         break;
4273                 }
4274                 case SMS_CDDVD_CAPS_PAGE:{
4275                         KASSERT(page_index->subpage == SMS_SUBPAGE_PAGE_0,
4276                             ("subpage %#x for page %#x is incorrect!",
4277                             page_index->subpage, page_code));
4278                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_DEFAULT],
4279                                &cddvd_page_default,
4280                                sizeof(cddvd_page_default));
4281                         memcpy(&lun->mode_pages.cddvd_page[
4282                                CTL_PAGE_CHANGEABLE], &cddvd_page_changeable,
4283                                sizeof(cddvd_page_changeable));
4284                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4285                                &cddvd_page_default,
4286                                sizeof(cddvd_page_default));
4287                         memcpy(&lun->mode_pages.cddvd_page[CTL_PAGE_CURRENT],
4288                                &lun->mode_pages.cddvd_page[CTL_PAGE_SAVED],
4289                                sizeof(cddvd_page_default));
4290                         page_index->page_data =
4291                                 (uint8_t *)lun->mode_pages.cddvd_page;
4292                         break;
4293                 }
4294                 default:
4295                         panic("invalid page code value %#x", page_code);
4296                 }
4297         }
4298
4299         return (CTL_RETVAL_COMPLETE);
4300 }
4301
4302 static int
4303 ctl_init_log_page_index(struct ctl_lun *lun)
4304 {
4305         struct ctl_page_index *page_index;
4306         int i, j, k, prev;
4307
4308         memcpy(&lun->log_pages.index, log_page_index_template,
4309                sizeof(log_page_index_template));
4310
4311         prev = -1;
4312         for (i = 0, j = 0, k = 0; i < CTL_NUM_LOG_PAGES; i++) {
4313
4314                 page_index = &lun->log_pages.index[i];
4315                 if (lun->be_lun->lun_type == T_DIRECT &&
4316                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
4317                         continue;
4318                 if (lun->be_lun->lun_type == T_PROCESSOR &&
4319                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
4320                         continue;
4321                 if (lun->be_lun->lun_type == T_CDROM &&
4322                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
4323                         continue;
4324
4325                 if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING &&
4326                     lun->backend->lun_attr == NULL)
4327                         continue;
4328
4329                 if (page_index->page_code != prev) {
4330                         lun->log_pages.pages_page[j] = page_index->page_code;
4331                         prev = page_index->page_code;
4332                         j++;
4333                 }
4334                 lun->log_pages.subpages_page[k*2] = page_index->page_code;
4335                 lun->log_pages.subpages_page[k*2+1] = page_index->subpage;
4336                 k++;
4337         }
4338         lun->log_pages.index[0].page_data = &lun->log_pages.pages_page[0];
4339         lun->log_pages.index[0].page_len = j;
4340         lun->log_pages.index[1].page_data = &lun->log_pages.subpages_page[0];
4341         lun->log_pages.index[1].page_len = k * 2;
4342         lun->log_pages.index[2].page_data = &lun->log_pages.lbp_page[0];
4343         lun->log_pages.index[2].page_len = 12*CTL_NUM_LBP_PARAMS;
4344         lun->log_pages.index[3].page_data = (uint8_t *)&lun->log_pages.stat_page;
4345         lun->log_pages.index[3].page_len = sizeof(lun->log_pages.stat_page);
4346         lun->log_pages.index[4].page_data = (uint8_t *)&lun->log_pages.ie_page;
4347         lun->log_pages.index[4].page_len = sizeof(lun->log_pages.ie_page);
4348
4349         return (CTL_RETVAL_COMPLETE);
4350 }
4351
4352 static int
4353 hex2bin(const char *str, uint8_t *buf, int buf_size)
4354 {
4355         int i;
4356         u_char c;
4357
4358         memset(buf, 0, buf_size);
4359         while (isspace(str[0]))
4360                 str++;
4361         if (str[0] == '0' && (str[1] == 'x' || str[1] == 'X'))
4362                 str += 2;
4363         buf_size *= 2;
4364         for (i = 0; str[i] != 0 && i < buf_size; i++) {
4365                 while (str[i] == '-')   /* Skip dashes in UUIDs. */
4366                         str++;
4367                 c = str[i];
4368                 if (isdigit(c))
4369                         c -= '0';
4370                 else if (isalpha(c))
4371                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4372                 else
4373                         break;
4374                 if (c >= 16)
4375                         break;
4376                 if ((i & 1) == 0)
4377                         buf[i / 2] |= (c << 4);
4378                 else
4379                         buf[i / 2] |= c;
4380         }
4381         return ((i + 1) / 2);
4382 }
4383
4384 /*
4385  * LUN allocation.
4386  *
4387  * Requirements:
4388  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4389  *   wants us to allocate the LUN and he can block.
4390  * - ctl_softc is always set
4391  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4392  *
4393  * Returns 0 for success, non-zero (errno) for failure.
4394  */
4395 static int
4396 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4397               struct ctl_be_lun *const be_lun)
4398 {
4399         struct ctl_lun *nlun, *lun;
4400         struct scsi_vpd_id_descriptor *desc;
4401         struct scsi_vpd_id_t10 *t10id;
4402         const char *eui, *naa, *scsiname, *uuid, *vendor, *value;
4403         int lun_number, i, lun_malloced;
4404         int devidlen, idlen1, idlen2 = 0, len;
4405
4406         if (be_lun == NULL)
4407                 return (EINVAL);
4408
4409         /*
4410          * We currently only support Direct Access or Processor LUN types.
4411          */
4412         switch (be_lun->lun_type) {
4413         case T_DIRECT:
4414         case T_PROCESSOR:
4415         case T_CDROM:
4416                 break;
4417         case T_SEQUENTIAL:
4418         case T_CHANGER:
4419         default:
4420                 be_lun->lun_config_status(be_lun->be_lun,
4421                                           CTL_LUN_CONFIG_FAILURE);
4422                 break;
4423         }
4424         if (ctl_lun == NULL) {
4425                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4426                 lun_malloced = 1;
4427         } else {
4428                 lun_malloced = 0;
4429                 lun = ctl_lun;
4430         }
4431
4432         memset(lun, 0, sizeof(*lun));
4433         if (lun_malloced)
4434                 lun->flags = CTL_LUN_MALLOCED;
4435
4436         /* Generate LUN ID. */
4437         devidlen = max(CTL_DEVID_MIN_LEN,
4438             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4439         idlen1 = sizeof(*t10id) + devidlen;
4440         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4441         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4442         if (scsiname != NULL) {
4443                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4444                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4445         }
4446         eui = ctl_get_opt(&be_lun->options, "eui");
4447         if (eui != NULL) {
4448                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4449         }
4450         naa = ctl_get_opt(&be_lun->options, "naa");
4451         if (naa != NULL) {
4452                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4453         }
4454         uuid = ctl_get_opt(&be_lun->options, "uuid");
4455         if (uuid != NULL) {
4456                 len += sizeof(struct scsi_vpd_id_descriptor) + 18;
4457         }
4458         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4459             M_CTL, M_WAITOK | M_ZERO);
4460         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4461         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4462         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4463         desc->length = idlen1;
4464         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4465         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4466         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4467                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4468         } else {
4469                 strncpy(t10id->vendor, vendor,
4470                     min(sizeof(t10id->vendor), strlen(vendor)));
4471         }
4472         strncpy((char *)t10id->vendor_spec_id,
4473             (char *)be_lun->device_id, devidlen);
4474         if (scsiname != NULL) {
4475                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4476                     desc->length);
4477                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4478                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4479                     SVPD_ID_TYPE_SCSI_NAME;
4480                 desc->length = idlen2;
4481                 strlcpy(desc->identifier, scsiname, idlen2);
4482         }
4483         if (eui != NULL) {
4484                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4485                     desc->length);
4486                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4487                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4488                     SVPD_ID_TYPE_EUI64;
4489                 desc->length = hex2bin(eui, desc->identifier, 16);
4490                 desc->length = desc->length > 12 ? 16 :
4491                     (desc->length > 8 ? 12 : 8);
4492                 len -= 16 - desc->length;
4493         }
4494         if (naa != NULL) {
4495                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4496                     desc->length);
4497                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4498                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4499                     SVPD_ID_TYPE_NAA;
4500                 desc->length = hex2bin(naa, desc->identifier, 16);
4501                 desc->length = desc->length > 8 ? 16 : 8;
4502                 len -= 16 - desc->length;
4503         }
4504         if (uuid != NULL) {
4505                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4506                     desc->length);
4507                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4508                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4509                     SVPD_ID_TYPE_UUID;
4510                 desc->identifier[0] = 0x10;
4511                 hex2bin(uuid, &desc->identifier[2], 16);
4512                 desc->length = 18;
4513         }
4514         lun->lun_devid->len = len;
4515
4516         mtx_lock(&ctl_softc->ctl_lock);
4517         /*
4518          * See if the caller requested a particular LUN number.  If so, see
4519          * if it is available.  Otherwise, allocate the first available LUN.
4520          */
4521         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4522                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4523                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4524                         mtx_unlock(&ctl_softc->ctl_lock);
4525                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4526                                 printf("ctl: requested LUN ID %d is higher "
4527                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4528                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4529                         } else {
4530                                 /*
4531                                  * XXX KDM return an error, or just assign
4532                                  * another LUN ID in this case??
4533                                  */
4534                                 printf("ctl: requested LUN ID %d is already "
4535                                        "in use\n", be_lun->req_lun_id);
4536                         }
4537                         if (lun->flags & CTL_LUN_MALLOCED)
4538                                 free(lun, M_CTL);
4539                         be_lun->lun_config_status(be_lun->be_lun,
4540                                                   CTL_LUN_CONFIG_FAILURE);
4541                         return (ENOSPC);
4542                 }
4543                 lun_number = be_lun->req_lun_id;
4544         } else {
4545                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4546                 if (lun_number == -1) {
4547                         mtx_unlock(&ctl_softc->ctl_lock);
4548                         printf("ctl: can't allocate LUN, out of LUNs\n");
4549                         if (lun->flags & CTL_LUN_MALLOCED)
4550                                 free(lun, M_CTL);
4551                         be_lun->lun_config_status(be_lun->be_lun,
4552                                                   CTL_LUN_CONFIG_FAILURE);
4553                         return (ENOSPC);
4554                 }
4555         }
4556         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4557
4558         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4559         lun->lun = lun_number;
4560         lun->be_lun = be_lun;
4561         /*
4562          * The processor LUN is always enabled.  Disk LUNs come on line
4563          * disabled, and must be enabled by the backend.
4564          */
4565         lun->flags |= CTL_LUN_DISABLED;
4566         lun->backend = be_lun->be;
4567         be_lun->ctl_lun = lun;
4568         be_lun->lun_id = lun_number;
4569         atomic_add_int(&be_lun->be->num_luns, 1);
4570         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4571                 lun->flags |= CTL_LUN_EJECTED;
4572         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4573                 lun->flags |= CTL_LUN_NO_MEDIA;
4574         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4575                 lun->flags |= CTL_LUN_STOPPED;
4576
4577         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4578                 lun->flags |= CTL_LUN_PRIMARY_SC;
4579
4580         value = ctl_get_opt(&be_lun->options, "removable");
4581         if (value != NULL) {
4582                 if (strcmp(value, "on") == 0)
4583                         lun->flags |= CTL_LUN_REMOVABLE;
4584         } else if (be_lun->lun_type == T_CDROM)
4585                 lun->flags |= CTL_LUN_REMOVABLE;
4586
4587         lun->ctl_softc = ctl_softc;
4588 #ifdef CTL_TIME_IO
4589         lun->last_busy = getsbinuptime();
4590 #endif
4591         TAILQ_INIT(&lun->ooa_queue);
4592         TAILQ_INIT(&lun->blocked_queue);
4593         STAILQ_INIT(&lun->error_list);
4594         lun->ie_reported = 1;
4595         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4596         ctl_tpc_lun_init(lun);
4597
4598         /*
4599          * Initialize the mode and log page index.
4600          */
4601         ctl_init_page_index(lun);
4602         ctl_init_log_page_index(lun);
4603
4604         /*
4605          * Now, before we insert this lun on the lun list, set the lun
4606          * inventory changed UA for all other luns.
4607          */
4608         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4609                 mtx_lock(&nlun->lun_lock);
4610                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4611                 mtx_unlock(&nlun->lun_lock);
4612         }
4613
4614         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4615
4616         ctl_softc->ctl_luns[lun_number] = lun;
4617
4618         ctl_softc->num_luns++;
4619
4620         /* Setup statistics gathering */
4621         lun->stats.device_type = be_lun->lun_type;
4622         lun->stats.lun_number = lun_number;
4623         lun->stats.blocksize = be_lun->blocksize;
4624         if (be_lun->blocksize == 0)
4625                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4626         for (i = 0;i < CTL_MAX_PORTS;i++)
4627                 lun->stats.ports[i].targ_port = i;
4628
4629         mtx_unlock(&ctl_softc->ctl_lock);
4630
4631         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4632         return (0);
4633 }
4634
4635 /*
4636  * Delete a LUN.
4637  * Assumptions:
4638  * - LUN has already been marked invalid and any pending I/O has been taken
4639  *   care of.
4640  */
4641 static int
4642 ctl_free_lun(struct ctl_lun *lun)
4643 {
4644         struct ctl_softc *softc;
4645         struct ctl_lun *nlun;
4646         int i;
4647
4648         softc = lun->ctl_softc;
4649
4650         mtx_assert(&softc->ctl_lock, MA_OWNED);
4651
4652         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4653
4654         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4655
4656         softc->ctl_luns[lun->lun] = NULL;
4657
4658         if (!TAILQ_EMPTY(&lun->ooa_queue))
4659                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4660
4661         softc->num_luns--;
4662
4663         /*
4664          * Tell the backend to free resources, if this LUN has a backend.
4665          */
4666         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4667         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4668
4669         lun->ie_reportcnt = UINT32_MAX;
4670         callout_drain(&lun->ie_callout);
4671
4672         ctl_tpc_lun_shutdown(lun);
4673         mtx_destroy(&lun->lun_lock);
4674         free(lun->lun_devid, M_CTL);
4675         for (i = 0; i < CTL_MAX_PORTS; i++)
4676                 free(lun->pending_ua[i], M_CTL);
4677         for (i = 0; i < CTL_MAX_PORTS; i++)
4678                 free(lun->pr_keys[i], M_CTL);
4679         free(lun->write_buffer, M_CTL);
4680         if (lun->flags & CTL_LUN_MALLOCED)
4681                 free(lun, M_CTL);
4682
4683         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4684                 mtx_lock(&nlun->lun_lock);
4685                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4686                 mtx_unlock(&nlun->lun_lock);
4687         }
4688
4689         return (0);
4690 }
4691
4692 static void
4693 ctl_create_lun(struct ctl_be_lun *be_lun)
4694 {
4695
4696         /*
4697          * ctl_alloc_lun() should handle all potential failure cases.
4698          */
4699         ctl_alloc_lun(control_softc, NULL, be_lun);
4700 }
4701
4702 int
4703 ctl_add_lun(struct ctl_be_lun *be_lun)
4704 {
4705         struct ctl_softc *softc = control_softc;
4706
4707         mtx_lock(&softc->ctl_lock);
4708         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4709         mtx_unlock(&softc->ctl_lock);
4710         wakeup(&softc->pending_lun_queue);
4711
4712         return (0);
4713 }
4714
4715 int
4716 ctl_enable_lun(struct ctl_be_lun *be_lun)
4717 {
4718         struct ctl_softc *softc;
4719         struct ctl_port *port, *nport;
4720         struct ctl_lun *lun;
4721         int retval;
4722
4723         lun = (struct ctl_lun *)be_lun->ctl_lun;
4724         softc = lun->ctl_softc;
4725
4726         mtx_lock(&softc->ctl_lock);
4727         mtx_lock(&lun->lun_lock);
4728         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4729                 /*
4730                  * eh?  Why did we get called if the LUN is already
4731                  * enabled?
4732                  */
4733                 mtx_unlock(&lun->lun_lock);
4734                 mtx_unlock(&softc->ctl_lock);
4735                 return (0);
4736         }
4737         lun->flags &= ~CTL_LUN_DISABLED;
4738         mtx_unlock(&lun->lun_lock);
4739
4740         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4741                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4742                     port->lun_map != NULL || port->lun_enable == NULL)
4743                         continue;
4744
4745                 /*
4746                  * Drop the lock while we call the FETD's enable routine.
4747                  * This can lead to a callback into CTL (at least in the
4748                  * case of the internal initiator frontend.
4749                  */
4750                 mtx_unlock(&softc->ctl_lock);
4751                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4752                 mtx_lock(&softc->ctl_lock);
4753                 if (retval != 0) {
4754                         printf("%s: FETD %s port %d returned error "
4755                                "%d for lun_enable on lun %jd\n",
4756                                __func__, port->port_name, port->targ_port,
4757                                retval, (intmax_t)lun->lun);
4758                 }
4759         }
4760
4761         mtx_unlock(&softc->ctl_lock);
4762         ctl_isc_announce_lun(lun);
4763
4764         return (0);
4765 }
4766
4767 int
4768 ctl_disable_lun(struct ctl_be_lun *be_lun)
4769 {
4770         struct ctl_softc *softc;
4771         struct ctl_port *port;
4772         struct ctl_lun *lun;
4773         int retval;
4774
4775         lun = (struct ctl_lun *)be_lun->ctl_lun;
4776         softc = lun->ctl_softc;
4777
4778         mtx_lock(&softc->ctl_lock);
4779         mtx_lock(&lun->lun_lock);
4780         if (lun->flags & CTL_LUN_DISABLED) {
4781                 mtx_unlock(&lun->lun_lock);
4782                 mtx_unlock(&softc->ctl_lock);
4783                 return (0);
4784         }
4785         lun->flags |= CTL_LUN_DISABLED;
4786         mtx_unlock(&lun->lun_lock);
4787
4788         STAILQ_FOREACH(port, &softc->port_list, links) {
4789                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4790                     port->lun_map != NULL || port->lun_disable == NULL)
4791                         continue;
4792
4793                 /*
4794                  * Drop the lock before we call the frontend's disable
4795                  * routine, to avoid lock order reversals.
4796                  *
4797                  * XXX KDM what happens if the frontend list changes while
4798                  * we're traversing it?  It's unlikely, but should be handled.
4799                  */
4800                 mtx_unlock(&softc->ctl_lock);
4801                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4802                 mtx_lock(&softc->ctl_lock);
4803                 if (retval != 0) {
4804                         printf("%s: FETD %s port %d returned error "
4805                                "%d for lun_disable on lun %jd\n",
4806                                __func__, port->port_name, port->targ_port,
4807                                retval, (intmax_t)lun->lun);
4808                 }
4809         }
4810
4811         mtx_unlock(&softc->ctl_lock);
4812         ctl_isc_announce_lun(lun);
4813
4814         return (0);
4815 }
4816
4817 int
4818 ctl_start_lun(struct ctl_be_lun *be_lun)
4819 {
4820         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4821
4822         mtx_lock(&lun->lun_lock);
4823         lun->flags &= ~CTL_LUN_STOPPED;
4824         mtx_unlock(&lun->lun_lock);
4825         return (0);
4826 }
4827
4828 int
4829 ctl_stop_lun(struct ctl_be_lun *be_lun)
4830 {
4831         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4832
4833         mtx_lock(&lun->lun_lock);
4834         lun->flags |= CTL_LUN_STOPPED;
4835         mtx_unlock(&lun->lun_lock);
4836         return (0);
4837 }
4838
4839 int
4840 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4841 {
4842         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4843
4844         mtx_lock(&lun->lun_lock);
4845         lun->flags |= CTL_LUN_NO_MEDIA;
4846         mtx_unlock(&lun->lun_lock);
4847         return (0);
4848 }
4849
4850 int
4851 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4852 {
4853         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4854         union ctl_ha_msg msg;
4855
4856         mtx_lock(&lun->lun_lock);
4857         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4858         if (lun->flags & CTL_LUN_REMOVABLE)
4859                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4860         mtx_unlock(&lun->lun_lock);
4861         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4862             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4863                 bzero(&msg.ua, sizeof(msg.ua));
4864                 msg.hdr.msg_type = CTL_MSG_UA;
4865                 msg.hdr.nexus.initid = -1;
4866                 msg.hdr.nexus.targ_port = -1;
4867                 msg.hdr.nexus.targ_lun = lun->lun;
4868                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4869                 msg.ua.ua_all = 1;
4870                 msg.ua.ua_set = 1;
4871                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4872                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4873                     M_WAITOK);
4874         }
4875         return (0);
4876 }
4877
4878 int
4879 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4880 {
4881         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4882
4883         mtx_lock(&lun->lun_lock);
4884         lun->flags |= CTL_LUN_EJECTED;
4885         mtx_unlock(&lun->lun_lock);
4886         return (0);
4887 }
4888
4889 int
4890 ctl_lun_primary(struct ctl_be_lun *be_lun)
4891 {
4892         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4893
4894         mtx_lock(&lun->lun_lock);
4895         lun->flags |= CTL_LUN_PRIMARY_SC;
4896         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4897         mtx_unlock(&lun->lun_lock);
4898         ctl_isc_announce_lun(lun);
4899         return (0);
4900 }
4901
4902 int
4903 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4904 {
4905         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4906
4907         mtx_lock(&lun->lun_lock);
4908         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4909         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4910         mtx_unlock(&lun->lun_lock);
4911         ctl_isc_announce_lun(lun);
4912         return (0);
4913 }
4914
4915 int
4916 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4917 {
4918         struct ctl_softc *softc;
4919         struct ctl_lun *lun;
4920
4921         lun = (struct ctl_lun *)be_lun->ctl_lun;
4922         softc = lun->ctl_softc;
4923
4924         mtx_lock(&lun->lun_lock);
4925
4926         /*
4927          * The LUN needs to be disabled before it can be marked invalid.
4928          */
4929         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4930                 mtx_unlock(&lun->lun_lock);
4931                 return (-1);
4932         }
4933         /*
4934          * Mark the LUN invalid.
4935          */
4936         lun->flags |= CTL_LUN_INVALID;
4937
4938         /*
4939          * If there is nothing in the OOA queue, go ahead and free the LUN.
4940          * If we have something in the OOA queue, we'll free it when the
4941          * last I/O completes.
4942          */
4943         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4944                 mtx_unlock(&lun->lun_lock);
4945                 mtx_lock(&softc->ctl_lock);
4946                 ctl_free_lun(lun);
4947                 mtx_unlock(&softc->ctl_lock);
4948         } else
4949                 mtx_unlock(&lun->lun_lock);
4950
4951         return (0);
4952 }
4953
4954 void
4955 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4956 {
4957         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4958         union ctl_ha_msg msg;
4959
4960         mtx_lock(&lun->lun_lock);
4961         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4962         mtx_unlock(&lun->lun_lock);
4963         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4964                 /* Send msg to other side. */
4965                 bzero(&msg.ua, sizeof(msg.ua));
4966                 msg.hdr.msg_type = CTL_MSG_UA;
4967                 msg.hdr.nexus.initid = -1;
4968                 msg.hdr.nexus.targ_port = -1;
4969                 msg.hdr.nexus.targ_lun = lun->lun;
4970                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4971                 msg.ua.ua_all = 1;
4972                 msg.ua.ua_set = 1;
4973                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4974                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4975                     M_WAITOK);
4976         }
4977 }
4978
4979 /*
4980  * Backend "memory move is complete" callback for requests that never
4981  * make it down to say RAIDCore's configuration code.
4982  */
4983 int
4984 ctl_config_move_done(union ctl_io *io)
4985 {
4986         int retval;
4987
4988         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4989         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4990             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4991
4992         if ((io->io_hdr.port_status != 0) &&
4993             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4994              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4995                 /*
4996                  * For hardware error sense keys, the sense key
4997                  * specific value is defined to be a retry count,
4998                  * but we use it to pass back an internal FETD
4999                  * error code.  XXX KDM  Hopefully the FETD is only
5000                  * using 16 bits for an error code, since that's
5001                  * all the space we have in the sks field.
5002                  */
5003                 ctl_set_internal_failure(&io->scsiio,
5004                                          /*sks_valid*/ 1,
5005                                          /*retry_count*/
5006                                          io->io_hdr.port_status);
5007         }
5008
5009         if (ctl_debug & CTL_DEBUG_CDB_DATA)
5010                 ctl_data_print(io);
5011         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
5012             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5013              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
5014             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
5015                 /*
5016                  * XXX KDM just assuming a single pointer here, and not a
5017                  * S/G list.  If we start using S/G lists for config data,
5018                  * we'll need to know how to clean them up here as well.
5019                  */
5020                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5021                         free(io->scsiio.kern_data_ptr, M_CTL);
5022                 ctl_done(io);
5023                 retval = CTL_RETVAL_COMPLETE;
5024         } else {
5025                 /*
5026                  * XXX KDM now we need to continue data movement.  Some
5027                  * options:
5028                  * - call ctl_scsiio() again?  We don't do this for data
5029                  *   writes, because for those at least we know ahead of
5030                  *   time where the write will go and how long it is.  For
5031                  *   config writes, though, that information is largely
5032                  *   contained within the write itself, thus we need to
5033                  *   parse out the data again.
5034                  *
5035                  * - Call some other function once the data is in?
5036                  */
5037
5038                 /*
5039                  * XXX KDM call ctl_scsiio() again for now, and check flag
5040                  * bits to see whether we're allocated or not.
5041                  */
5042                 retval = ctl_scsiio(&io->scsiio);
5043         }
5044         return (retval);
5045 }
5046
5047 /*
5048  * This gets called by a backend driver when it is done with a
5049  * data_submit method.
5050  */
5051 void
5052 ctl_data_submit_done(union ctl_io *io)
5053 {
5054         /*
5055          * If the IO_CONT flag is set, we need to call the supplied
5056          * function to continue processing the I/O, instead of completing
5057          * the I/O just yet.
5058          *
5059          * If there is an error, though, we don't want to keep processing.
5060          * Instead, just send status back to the initiator.
5061          */
5062         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5063             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5064             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5065              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5066                 io->scsiio.io_cont(io);
5067                 return;
5068         }
5069         ctl_done(io);
5070 }
5071
5072 /*
5073  * This gets called by a backend driver when it is done with a
5074  * configuration write.
5075  */
5076 void
5077 ctl_config_write_done(union ctl_io *io)
5078 {
5079         uint8_t *buf;
5080
5081         /*
5082          * If the IO_CONT flag is set, we need to call the supplied
5083          * function to continue processing the I/O, instead of completing
5084          * the I/O just yet.
5085          *
5086          * If there is an error, though, we don't want to keep processing.
5087          * Instead, just send status back to the initiator.
5088          */
5089         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5090             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5091             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5092              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5093                 io->scsiio.io_cont(io);
5094                 return;
5095         }
5096         /*
5097          * Since a configuration write can be done for commands that actually
5098          * have data allocated, like write buffer, and commands that have
5099          * no data, like start/stop unit, we need to check here.
5100          */
5101         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5102                 buf = io->scsiio.kern_data_ptr;
5103         else
5104                 buf = NULL;
5105         ctl_done(io);
5106         if (buf)
5107                 free(buf, M_CTL);
5108 }
5109
5110 void
5111 ctl_config_read_done(union ctl_io *io)
5112 {
5113         uint8_t *buf;
5114
5115         /*
5116          * If there is some error -- we are done, skip data transfer.
5117          */
5118         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5119             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5120              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5121                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5122                         buf = io->scsiio.kern_data_ptr;
5123                 else
5124                         buf = NULL;
5125                 ctl_done(io);
5126                 if (buf)
5127                         free(buf, M_CTL);
5128                 return;
5129         }
5130
5131         /*
5132          * If the IO_CONT flag is set, we need to call the supplied
5133          * function to continue processing the I/O, instead of completing
5134          * the I/O just yet.
5135          */
5136         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5137                 io->scsiio.io_cont(io);
5138                 return;
5139         }
5140
5141         ctl_datamove(io);
5142 }
5143
5144 /*
5145  * SCSI release command.
5146  */
5147 int
5148 ctl_scsi_release(struct ctl_scsiio *ctsio)
5149 {
5150         struct ctl_lun *lun;
5151         uint32_t residx;
5152
5153         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5154
5155         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5156         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5157
5158         /*
5159          * XXX KDM right now, we only support LUN reservation.  We don't
5160          * support 3rd party reservations, or extent reservations, which
5161          * might actually need the parameter list.  If we've gotten this
5162          * far, we've got a LUN reservation.  Anything else got kicked out
5163          * above.  So, according to SPC, ignore the length.
5164          */
5165
5166         mtx_lock(&lun->lun_lock);
5167
5168         /*
5169          * According to SPC, it is not an error for an intiator to attempt
5170          * to release a reservation on a LUN that isn't reserved, or that
5171          * is reserved by another initiator.  The reservation can only be
5172          * released, though, by the initiator who made it or by one of
5173          * several reset type events.
5174          */
5175         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5176                         lun->flags &= ~CTL_LUN_RESERVED;
5177
5178         mtx_unlock(&lun->lun_lock);
5179
5180         ctl_set_success(ctsio);
5181         ctl_done((union ctl_io *)ctsio);
5182         return (CTL_RETVAL_COMPLETE);
5183 }
5184
5185 int
5186 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5187 {
5188         struct ctl_lun *lun;
5189         uint32_t residx;
5190
5191         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5192
5193         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5194         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5195
5196         /*
5197          * XXX KDM right now, we only support LUN reservation.  We don't
5198          * support 3rd party reservations, or extent reservations, which
5199          * might actually need the parameter list.  If we've gotten this
5200          * far, we've got a LUN reservation.  Anything else got kicked out
5201          * above.  So, according to SPC, ignore the length.
5202          */
5203
5204         mtx_lock(&lun->lun_lock);
5205         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5206                 ctl_set_reservation_conflict(ctsio);
5207                 goto bailout;
5208         }
5209
5210         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5211         if (lun->flags & CTL_LUN_PR_RESERVED) {
5212                 ctl_set_success(ctsio);
5213                 goto bailout;
5214         }
5215
5216         lun->flags |= CTL_LUN_RESERVED;
5217         lun->res_idx = residx;
5218         ctl_set_success(ctsio);
5219
5220 bailout:
5221         mtx_unlock(&lun->lun_lock);
5222         ctl_done((union ctl_io *)ctsio);
5223         return (CTL_RETVAL_COMPLETE);
5224 }
5225
5226 int
5227 ctl_start_stop(struct ctl_scsiio *ctsio)
5228 {
5229         struct scsi_start_stop_unit *cdb;
5230         struct ctl_lun *lun;
5231         int retval;
5232
5233         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5234
5235         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5236         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5237
5238         if ((cdb->how & SSS_PC_MASK) == 0) {
5239                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5240                     (cdb->how & SSS_START) == 0) {
5241                         uint32_t residx;
5242
5243                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5244                         if (ctl_get_prkey(lun, residx) == 0 ||
5245                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5246
5247                                 ctl_set_reservation_conflict(ctsio);
5248                                 ctl_done((union ctl_io *)ctsio);
5249                                 return (CTL_RETVAL_COMPLETE);
5250                         }
5251                 }
5252
5253                 if ((cdb->how & SSS_LOEJ) &&
5254                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5255                         ctl_set_invalid_field(ctsio,
5256                                               /*sks_valid*/ 1,
5257                                               /*command*/ 1,
5258                                               /*field*/ 4,
5259                                               /*bit_valid*/ 1,
5260                                               /*bit*/ 1);
5261                         ctl_done((union ctl_io *)ctsio);
5262                         return (CTL_RETVAL_COMPLETE);
5263                 }
5264
5265                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5266                     lun->prevent_count > 0) {
5267                         /* "Medium removal prevented" */
5268                         ctl_set_sense(ctsio, /*current_error*/ 1,
5269                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5270                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5271                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5272                         ctl_done((union ctl_io *)ctsio);
5273                         return (CTL_RETVAL_COMPLETE);
5274                 }
5275         }
5276
5277         retval = lun->backend->config_write((union ctl_io *)ctsio);
5278         return (retval);
5279 }
5280
5281 int
5282 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5283 {
5284         struct ctl_lun *lun;
5285         struct scsi_prevent *cdb;
5286         int retval;
5287         uint32_t initidx;
5288
5289         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5290
5291         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5292         cdb = (struct scsi_prevent *)ctsio->cdb;
5293
5294         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5295                 ctl_set_invalid_opcode(ctsio);
5296                 ctl_done((union ctl_io *)ctsio);
5297                 return (CTL_RETVAL_COMPLETE);
5298         }
5299
5300         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5301         mtx_lock(&lun->lun_lock);
5302         if ((cdb->how & PR_PREVENT) &&
5303             ctl_is_set(lun->prevent, initidx) == 0) {
5304                 ctl_set_mask(lun->prevent, initidx);
5305                 lun->prevent_count++;
5306         } else if ((cdb->how & PR_PREVENT) == 0 &&
5307             ctl_is_set(lun->prevent, initidx)) {
5308                 ctl_clear_mask(lun->prevent, initidx);
5309                 lun->prevent_count--;
5310         }
5311         mtx_unlock(&lun->lun_lock);
5312         retval = lun->backend->config_write((union ctl_io *)ctsio);
5313         return (retval);
5314 }
5315
5316 /*
5317  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5318  * we don't really do anything with the LBA and length fields if the user
5319  * passes them in.  Instead we'll just flush out the cache for the entire
5320  * LUN.
5321  */
5322 int
5323 ctl_sync_cache(struct ctl_scsiio *ctsio)
5324 {
5325         struct ctl_lun *lun;
5326         struct ctl_softc *softc;
5327         struct ctl_lba_len_flags *lbalen;
5328         uint64_t starting_lba;
5329         uint32_t block_count;
5330         int retval;
5331         uint8_t byte2;
5332
5333         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5334
5335         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5336         softc = lun->ctl_softc;
5337         retval = 0;
5338
5339         switch (ctsio->cdb[0]) {
5340         case SYNCHRONIZE_CACHE: {
5341                 struct scsi_sync_cache *cdb;
5342                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5343
5344                 starting_lba = scsi_4btoul(cdb->begin_lba);
5345                 block_count = scsi_2btoul(cdb->lb_count);
5346                 byte2 = cdb->byte2;
5347                 break;
5348         }
5349         case SYNCHRONIZE_CACHE_16: {
5350                 struct scsi_sync_cache_16 *cdb;
5351                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5352
5353                 starting_lba = scsi_8btou64(cdb->begin_lba);
5354                 block_count = scsi_4btoul(cdb->lb_count);
5355                 byte2 = cdb->byte2;
5356                 break;
5357         }
5358         default:
5359                 ctl_set_invalid_opcode(ctsio);
5360                 ctl_done((union ctl_io *)ctsio);
5361                 goto bailout;
5362                 break; /* NOTREACHED */
5363         }
5364
5365         /*
5366          * We check the LBA and length, but don't do anything with them.
5367          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5368          * get flushed.  This check will just help satisfy anyone who wants
5369          * to see an error for an out of range LBA.
5370          */
5371         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5372                 ctl_set_lba_out_of_range(ctsio,
5373                     MAX(starting_lba, lun->be_lun->maxlba + 1));
5374                 ctl_done((union ctl_io *)ctsio);
5375                 goto bailout;
5376         }
5377
5378         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5379         lbalen->lba = starting_lba;
5380         lbalen->len = block_count;
5381         lbalen->flags = byte2;
5382         retval = lun->backend->config_write((union ctl_io *)ctsio);
5383
5384 bailout:
5385         return (retval);
5386 }
5387
5388 int
5389 ctl_format(struct ctl_scsiio *ctsio)
5390 {
5391         struct scsi_format *cdb;
5392         struct ctl_lun *lun;
5393         int length, defect_list_len;
5394
5395         CTL_DEBUG_PRINT(("ctl_format\n"));
5396
5397         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5398
5399         cdb = (struct scsi_format *)ctsio->cdb;
5400
5401         length = 0;
5402         if (cdb->byte2 & SF_FMTDATA) {
5403                 if (cdb->byte2 & SF_LONGLIST)
5404                         length = sizeof(struct scsi_format_header_long);
5405                 else
5406                         length = sizeof(struct scsi_format_header_short);
5407         }
5408
5409         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5410          && (length > 0)) {
5411                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5412                 ctsio->kern_data_len = length;
5413                 ctsio->kern_total_len = length;
5414                 ctsio->kern_data_resid = 0;
5415                 ctsio->kern_rel_offset = 0;
5416                 ctsio->kern_sg_entries = 0;
5417                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5418                 ctsio->be_move_done = ctl_config_move_done;
5419                 ctl_datamove((union ctl_io *)ctsio);
5420
5421                 return (CTL_RETVAL_COMPLETE);
5422         }
5423
5424         defect_list_len = 0;
5425
5426         if (cdb->byte2 & SF_FMTDATA) {
5427                 if (cdb->byte2 & SF_LONGLIST) {
5428                         struct scsi_format_header_long *header;
5429
5430                         header = (struct scsi_format_header_long *)
5431                                 ctsio->kern_data_ptr;
5432
5433                         defect_list_len = scsi_4btoul(header->defect_list_len);
5434                         if (defect_list_len != 0) {
5435                                 ctl_set_invalid_field(ctsio,
5436                                                       /*sks_valid*/ 1,
5437                                                       /*command*/ 0,
5438                                                       /*field*/ 2,
5439                                                       /*bit_valid*/ 0,
5440                                                       /*bit*/ 0);
5441                                 goto bailout;
5442                         }
5443                 } else {
5444                         struct scsi_format_header_short *header;
5445
5446                         header = (struct scsi_format_header_short *)
5447                                 ctsio->kern_data_ptr;
5448
5449                         defect_list_len = scsi_2btoul(header->defect_list_len);
5450                         if (defect_list_len != 0) {
5451                                 ctl_set_invalid_field(ctsio,
5452                                                       /*sks_valid*/ 1,
5453                                                       /*command*/ 0,
5454                                                       /*field*/ 2,
5455                                                       /*bit_valid*/ 0,
5456                                                       /*bit*/ 0);
5457                                 goto bailout;
5458                         }
5459                 }
5460         }
5461
5462         ctl_set_success(ctsio);
5463 bailout:
5464
5465         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5466                 free(ctsio->kern_data_ptr, M_CTL);
5467                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5468         }
5469
5470         ctl_done((union ctl_io *)ctsio);
5471         return (CTL_RETVAL_COMPLETE);
5472 }
5473
5474 int
5475 ctl_read_buffer(struct ctl_scsiio *ctsio)
5476 {
5477         struct ctl_lun *lun;
5478         uint64_t buffer_offset;
5479         uint32_t len;
5480         uint8_t byte2;
5481         static uint8_t descr[4];
5482         static uint8_t echo_descr[4] = { 0 };
5483
5484         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5485         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5486         switch (ctsio->cdb[0]) {
5487         case READ_BUFFER: {
5488                 struct scsi_read_buffer *cdb;
5489
5490                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5491                 buffer_offset = scsi_3btoul(cdb->offset);
5492                 len = scsi_3btoul(cdb->length);
5493                 byte2 = cdb->byte2;
5494                 break;
5495         }
5496         case READ_BUFFER_16: {
5497                 struct scsi_read_buffer_16 *cdb;
5498
5499                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5500                 buffer_offset = scsi_8btou64(cdb->offset);
5501                 len = scsi_4btoul(cdb->length);
5502                 byte2 = cdb->byte2;
5503                 break;
5504         }
5505         default: /* This shouldn't happen. */
5506                 ctl_set_invalid_opcode(ctsio);
5507                 ctl_done((union ctl_io *)ctsio);
5508                 return (CTL_RETVAL_COMPLETE);
5509         }
5510
5511         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5512             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5513                 ctl_set_invalid_field(ctsio,
5514                                       /*sks_valid*/ 1,
5515                                       /*command*/ 1,
5516                                       /*field*/ 6,
5517                                       /*bit_valid*/ 0,
5518                                       /*bit*/ 0);
5519                 ctl_done((union ctl_io *)ctsio);
5520                 return (CTL_RETVAL_COMPLETE);
5521         }
5522
5523         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5524                 descr[0] = 0;
5525                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5526                 ctsio->kern_data_ptr = descr;
5527                 len = min(len, sizeof(descr));
5528         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5529                 ctsio->kern_data_ptr = echo_descr;
5530                 len = min(len, sizeof(echo_descr));
5531         } else {
5532                 if (lun->write_buffer == NULL) {
5533                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5534                             M_CTL, M_WAITOK);
5535                 }
5536                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5537         }
5538         ctsio->kern_data_len = len;
5539         ctsio->kern_total_len = len;
5540         ctsio->kern_data_resid = 0;
5541         ctsio->kern_rel_offset = 0;
5542         ctsio->kern_sg_entries = 0;
5543         ctl_set_success(ctsio);
5544         ctsio->be_move_done = ctl_config_move_done;
5545         ctl_datamove((union ctl_io *)ctsio);
5546         return (CTL_RETVAL_COMPLETE);
5547 }
5548
5549 int
5550 ctl_write_buffer(struct ctl_scsiio *ctsio)
5551 {
5552         struct scsi_write_buffer *cdb;
5553         struct ctl_lun *lun;
5554         int buffer_offset, len;
5555
5556         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5557
5558         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5559         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5560
5561         len = scsi_3btoul(cdb->length);
5562         buffer_offset = scsi_3btoul(cdb->offset);
5563
5564         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5565                 ctl_set_invalid_field(ctsio,
5566                                       /*sks_valid*/ 1,
5567                                       /*command*/ 1,
5568                                       /*field*/ 6,
5569                                       /*bit_valid*/ 0,
5570                                       /*bit*/ 0);
5571                 ctl_done((union ctl_io *)ctsio);
5572                 return (CTL_RETVAL_COMPLETE);
5573         }
5574
5575         /*
5576          * If we've got a kernel request that hasn't been malloced yet,
5577          * malloc it and tell the caller the data buffer is here.
5578          */
5579         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5580                 if (lun->write_buffer == NULL) {
5581                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5582                             M_CTL, M_WAITOK);
5583                 }
5584                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5585                 ctsio->kern_data_len = len;
5586                 ctsio->kern_total_len = len;
5587                 ctsio->kern_data_resid = 0;
5588                 ctsio->kern_rel_offset = 0;
5589                 ctsio->kern_sg_entries = 0;
5590                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5591                 ctsio->be_move_done = ctl_config_move_done;
5592                 ctl_datamove((union ctl_io *)ctsio);
5593
5594                 return (CTL_RETVAL_COMPLETE);
5595         }
5596
5597         ctl_set_success(ctsio);
5598         ctl_done((union ctl_io *)ctsio);
5599         return (CTL_RETVAL_COMPLETE);
5600 }
5601
5602 int
5603 ctl_write_same(struct ctl_scsiio *ctsio)
5604 {
5605         struct ctl_lun *lun;
5606         struct ctl_lba_len_flags *lbalen;
5607         uint64_t lba;
5608         uint32_t num_blocks;
5609         int len, retval;
5610         uint8_t byte2;
5611
5612         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5613
5614         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5615
5616         switch (ctsio->cdb[0]) {
5617         case WRITE_SAME_10: {
5618                 struct scsi_write_same_10 *cdb;
5619
5620                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5621
5622                 lba = scsi_4btoul(cdb->addr);
5623                 num_blocks = scsi_2btoul(cdb->length);
5624                 byte2 = cdb->byte2;
5625                 break;
5626         }
5627         case WRITE_SAME_16: {
5628                 struct scsi_write_same_16 *cdb;
5629
5630                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5631
5632                 lba = scsi_8btou64(cdb->addr);
5633                 num_blocks = scsi_4btoul(cdb->length);
5634                 byte2 = cdb->byte2;
5635                 break;
5636         }
5637         default:
5638                 /*
5639                  * We got a command we don't support.  This shouldn't
5640                  * happen, commands should be filtered out above us.
5641                  */
5642                 ctl_set_invalid_opcode(ctsio);
5643                 ctl_done((union ctl_io *)ctsio);
5644
5645                 return (CTL_RETVAL_COMPLETE);
5646                 break; /* NOTREACHED */
5647         }
5648
5649         /* ANCHOR flag can be used only together with UNMAP */
5650         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5651                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5652                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5653                 ctl_done((union ctl_io *)ctsio);
5654                 return (CTL_RETVAL_COMPLETE);
5655         }
5656
5657         /*
5658          * The first check is to make sure we're in bounds, the second
5659          * check is to catch wrap-around problems.  If the lba + num blocks
5660          * is less than the lba, then we've wrapped around and the block
5661          * range is invalid anyway.
5662          */
5663         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5664          || ((lba + num_blocks) < lba)) {
5665                 ctl_set_lba_out_of_range(ctsio,
5666                     MAX(lba, lun->be_lun->maxlba + 1));
5667                 ctl_done((union ctl_io *)ctsio);
5668                 return (CTL_RETVAL_COMPLETE);
5669         }
5670
5671         /* Zero number of blocks means "to the last logical block" */
5672         if (num_blocks == 0) {
5673                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5674                         ctl_set_invalid_field(ctsio,
5675                                               /*sks_valid*/ 0,
5676                                               /*command*/ 1,
5677                                               /*field*/ 0,
5678                                               /*bit_valid*/ 0,
5679                                               /*bit*/ 0);
5680                         ctl_done((union ctl_io *)ctsio);
5681                         return (CTL_RETVAL_COMPLETE);
5682                 }
5683                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5684         }
5685
5686         len = lun->be_lun->blocksize;
5687
5688         /*
5689          * If we've got a kernel request that hasn't been malloced yet,
5690          * malloc it and tell the caller the data buffer is here.
5691          */
5692         if ((byte2 & SWS_NDOB) == 0 &&
5693             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5694                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5695                 ctsio->kern_data_len = len;
5696                 ctsio->kern_total_len = len;
5697                 ctsio->kern_data_resid = 0;
5698                 ctsio->kern_rel_offset = 0;
5699                 ctsio->kern_sg_entries = 0;
5700                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5701                 ctsio->be_move_done = ctl_config_move_done;
5702                 ctl_datamove((union ctl_io *)ctsio);
5703
5704                 return (CTL_RETVAL_COMPLETE);
5705         }
5706
5707         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5708         lbalen->lba = lba;
5709         lbalen->len = num_blocks;
5710         lbalen->flags = byte2;
5711         retval = lun->backend->config_write((union ctl_io *)ctsio);
5712
5713         return (retval);
5714 }
5715
5716 int
5717 ctl_unmap(struct ctl_scsiio *ctsio)
5718 {
5719         struct ctl_lun *lun;
5720         struct scsi_unmap *cdb;
5721         struct ctl_ptr_len_flags *ptrlen;
5722         struct scsi_unmap_header *hdr;
5723         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5724         uint64_t lba;
5725         uint32_t num_blocks;
5726         int len, retval;
5727         uint8_t byte2;
5728
5729         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5730
5731         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5732         cdb = (struct scsi_unmap *)ctsio->cdb;
5733
5734         len = scsi_2btoul(cdb->length);
5735         byte2 = cdb->byte2;
5736
5737         /*
5738          * If we've got a kernel request that hasn't been malloced yet,
5739          * malloc it and tell the caller the data buffer is here.
5740          */
5741         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5742                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5743                 ctsio->kern_data_len = len;
5744                 ctsio->kern_total_len = len;
5745                 ctsio->kern_data_resid = 0;
5746                 ctsio->kern_rel_offset = 0;
5747                 ctsio->kern_sg_entries = 0;
5748                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5749                 ctsio->be_move_done = ctl_config_move_done;
5750                 ctl_datamove((union ctl_io *)ctsio);
5751
5752                 return (CTL_RETVAL_COMPLETE);
5753         }
5754
5755         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5756         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5757         if (len < sizeof (*hdr) ||
5758             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5759             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5760             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5761                 ctl_set_invalid_field(ctsio,
5762                                       /*sks_valid*/ 0,
5763                                       /*command*/ 0,
5764                                       /*field*/ 0,
5765                                       /*bit_valid*/ 0,
5766                                       /*bit*/ 0);
5767                 goto done;
5768         }
5769         len = scsi_2btoul(hdr->desc_length);
5770         buf = (struct scsi_unmap_desc *)(hdr + 1);
5771         end = buf + len / sizeof(*buf);
5772
5773         endnz = buf;
5774         for (range = buf; range < end; range++) {
5775                 lba = scsi_8btou64(range->lba);
5776                 num_blocks = scsi_4btoul(range->length);
5777                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5778                  || ((lba + num_blocks) < lba)) {
5779                         ctl_set_lba_out_of_range(ctsio,
5780                             MAX(lba, lun->be_lun->maxlba + 1));
5781                         ctl_done((union ctl_io *)ctsio);
5782                         return (CTL_RETVAL_COMPLETE);
5783                 }
5784                 if (num_blocks != 0)
5785                         endnz = range + 1;
5786         }
5787
5788         /*
5789          * Block backend can not handle zero last range.
5790          * Filter it out and return if there is nothing left.
5791          */
5792         len = (uint8_t *)endnz - (uint8_t *)buf;
5793         if (len == 0) {
5794                 ctl_set_success(ctsio);
5795                 goto done;
5796         }
5797
5798         mtx_lock(&lun->lun_lock);
5799         ptrlen = (struct ctl_ptr_len_flags *)
5800             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5801         ptrlen->ptr = (void *)buf;
5802         ptrlen->len = len;
5803         ptrlen->flags = byte2;
5804         ctl_check_blocked(lun);
5805         mtx_unlock(&lun->lun_lock);
5806
5807         retval = lun->backend->config_write((union ctl_io *)ctsio);
5808         return (retval);
5809
5810 done:
5811         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5812                 free(ctsio->kern_data_ptr, M_CTL);
5813                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5814         }
5815         ctl_done((union ctl_io *)ctsio);
5816         return (CTL_RETVAL_COMPLETE);
5817 }
5818
5819 int
5820 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5821                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5822 {
5823         struct ctl_lun *lun;
5824         uint8_t *current_cp, *saved_cp;
5825         int set_ua;
5826         uint32_t initidx;
5827
5828         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5829         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5830         set_ua = 0;
5831
5832         current_cp = (page_index->page_data + (page_index->page_len *
5833             CTL_PAGE_CURRENT));
5834         saved_cp = (page_index->page_data + (page_index->page_len *
5835             CTL_PAGE_SAVED));
5836
5837         mtx_lock(&lun->lun_lock);
5838         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5839                 memcpy(current_cp, page_ptr, page_index->page_len);
5840                 memcpy(saved_cp, page_ptr, page_index->page_len);
5841                 set_ua = 1;
5842         }
5843         if (set_ua != 0)
5844                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5845         mtx_unlock(&lun->lun_lock);
5846         if (set_ua) {
5847                 ctl_isc_announce_mode(lun,
5848                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5849                     page_index->page_code, page_index->subpage);
5850         }
5851         return (CTL_RETVAL_COMPLETE);
5852 }
5853
5854 static void
5855 ctl_ie_timer(void *arg)
5856 {
5857         struct ctl_lun *lun = arg;
5858         uint64_t t;
5859
5860         if (lun->ie_asc == 0)
5861                 return;
5862
5863         if (lun->MODE_IE.mrie == SIEP_MRIE_UA)
5864                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5865         else
5866                 lun->ie_reported = 0;
5867
5868         if (lun->ie_reportcnt < scsi_4btoul(lun->MODE_IE.report_count)) {
5869                 lun->ie_reportcnt++;
5870                 t = scsi_4btoul(lun->MODE_IE.interval_timer);
5871                 if (t == 0 || t == UINT32_MAX)
5872                         t = 3000;  /* 5 min */
5873                 callout_schedule(&lun->ie_callout, t * hz / 10);
5874         }
5875 }
5876
5877 int
5878 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5879                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5880 {
5881         struct scsi_info_exceptions_page *pg;
5882         struct ctl_lun *lun;
5883         uint64_t t;
5884
5885         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5886
5887         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5888         pg = (struct scsi_info_exceptions_page *)page_ptr;
5889         mtx_lock(&lun->lun_lock);
5890         if (pg->info_flags & SIEP_FLAGS_TEST) {
5891                 lun->ie_asc = 0x5d;
5892                 lun->ie_ascq = 0xff;
5893                 if (pg->mrie == SIEP_MRIE_UA) {
5894                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5895                         lun->ie_reported = 1;
5896                 } else {
5897                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5898                         lun->ie_reported = -1;
5899                 }
5900                 lun->ie_reportcnt = 1;
5901                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5902                         lun->ie_reportcnt++;
5903                         t = scsi_4btoul(pg->interval_timer);
5904                         if (t == 0 || t == UINT32_MAX)
5905                                 t = 3000;  /* 5 min */
5906                         callout_reset(&lun->ie_callout, t * hz / 10,
5907                             ctl_ie_timer, lun);
5908                 }
5909         } else {
5910                 lun->ie_asc = 0;
5911                 lun->ie_ascq = 0;
5912                 lun->ie_reported = 1;
5913                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5914                 lun->ie_reportcnt = UINT32_MAX;
5915                 callout_stop(&lun->ie_callout);
5916         }
5917         mtx_unlock(&lun->lun_lock);
5918         return (CTL_RETVAL_COMPLETE);
5919 }
5920
5921 static int
5922 ctl_do_mode_select(union ctl_io *io)
5923 {
5924         struct scsi_mode_page_header *page_header;
5925         struct ctl_page_index *page_index;
5926         struct ctl_scsiio *ctsio;
5927         int page_len, page_len_offset, page_len_size;
5928         union ctl_modepage_info *modepage_info;
5929         struct ctl_lun *lun;
5930         int *len_left, *len_used;
5931         int retval, i;
5932
5933         ctsio = &io->scsiio;
5934         page_index = NULL;
5935         page_len = 0;
5936         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5937
5938         modepage_info = (union ctl_modepage_info *)
5939                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5940         len_left = &modepage_info->header.len_left;
5941         len_used = &modepage_info->header.len_used;
5942
5943 do_next_page:
5944
5945         page_header = (struct scsi_mode_page_header *)
5946                 (ctsio->kern_data_ptr + *len_used);
5947
5948         if (*len_left == 0) {
5949                 free(ctsio->kern_data_ptr, M_CTL);
5950                 ctl_set_success(ctsio);
5951                 ctl_done((union ctl_io *)ctsio);
5952                 return (CTL_RETVAL_COMPLETE);
5953         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5954
5955                 free(ctsio->kern_data_ptr, M_CTL);
5956                 ctl_set_param_len_error(ctsio);
5957                 ctl_done((union ctl_io *)ctsio);
5958                 return (CTL_RETVAL_COMPLETE);
5959
5960         } else if ((page_header->page_code & SMPH_SPF)
5961                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5962
5963                 free(ctsio->kern_data_ptr, M_CTL);
5964                 ctl_set_param_len_error(ctsio);
5965                 ctl_done((union ctl_io *)ctsio);
5966                 return (CTL_RETVAL_COMPLETE);
5967         }
5968
5969
5970         /*
5971          * XXX KDM should we do something with the block descriptor?
5972          */
5973         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5974                 page_index = &lun->mode_pages.index[i];
5975                 if (lun->be_lun->lun_type == T_DIRECT &&
5976                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
5977                         continue;
5978                 if (lun->be_lun->lun_type == T_PROCESSOR &&
5979                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
5980                         continue;
5981                 if (lun->be_lun->lun_type == T_CDROM &&
5982                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
5983                         continue;
5984
5985                 if ((page_index->page_code & SMPH_PC_MASK) !=
5986                     (page_header->page_code & SMPH_PC_MASK))
5987                         continue;
5988
5989                 /*
5990                  * If neither page has a subpage code, then we've got a
5991                  * match.
5992                  */
5993                 if (((page_index->page_code & SMPH_SPF) == 0)
5994                  && ((page_header->page_code & SMPH_SPF) == 0)) {
5995                         page_len = page_header->page_length;
5996                         break;
5997                 }
5998
5999                 /*
6000                  * If both pages have subpages, then the subpage numbers
6001                  * have to match.
6002                  */
6003                 if ((page_index->page_code & SMPH_SPF)
6004                   && (page_header->page_code & SMPH_SPF)) {
6005                         struct scsi_mode_page_header_sp *sph;
6006
6007                         sph = (struct scsi_mode_page_header_sp *)page_header;
6008                         if (page_index->subpage == sph->subpage) {
6009                                 page_len = scsi_2btoul(sph->page_length);
6010                                 break;
6011                         }
6012                 }
6013         }
6014
6015         /*
6016          * If we couldn't find the page, or if we don't have a mode select
6017          * handler for it, send back an error to the user.
6018          */
6019         if ((i >= CTL_NUM_MODE_PAGES)
6020          || (page_index->select_handler == NULL)) {
6021                 ctl_set_invalid_field(ctsio,
6022                                       /*sks_valid*/ 1,
6023                                       /*command*/ 0,
6024                                       /*field*/ *len_used,
6025                                       /*bit_valid*/ 0,
6026                                       /*bit*/ 0);
6027                 free(ctsio->kern_data_ptr, M_CTL);
6028                 ctl_done((union ctl_io *)ctsio);
6029                 return (CTL_RETVAL_COMPLETE);
6030         }
6031
6032         if (page_index->page_code & SMPH_SPF) {
6033                 page_len_offset = 2;
6034                 page_len_size = 2;
6035         } else {
6036                 page_len_size = 1;
6037                 page_len_offset = 1;
6038         }
6039
6040         /*
6041          * If the length the initiator gives us isn't the one we specify in
6042          * the mode page header, or if they didn't specify enough data in
6043          * the CDB to avoid truncating this page, kick out the request.
6044          */
6045         if (page_len != page_index->page_len - page_len_offset - page_len_size) {
6046                 ctl_set_invalid_field(ctsio,
6047                                       /*sks_valid*/ 1,
6048                                       /*command*/ 0,
6049                                       /*field*/ *len_used + page_len_offset,
6050                                       /*bit_valid*/ 0,
6051                                       /*bit*/ 0);
6052                 free(ctsio->kern_data_ptr, M_CTL);
6053                 ctl_done((union ctl_io *)ctsio);
6054                 return (CTL_RETVAL_COMPLETE);
6055         }
6056         if (*len_left < page_index->page_len) {
6057                 free(ctsio->kern_data_ptr, M_CTL);
6058                 ctl_set_param_len_error(ctsio);
6059                 ctl_done((union ctl_io *)ctsio);
6060                 return (CTL_RETVAL_COMPLETE);
6061         }
6062
6063         /*
6064          * Run through the mode page, checking to make sure that the bits
6065          * the user changed are actually legal for him to change.
6066          */
6067         for (i = 0; i < page_index->page_len; i++) {
6068                 uint8_t *user_byte, *change_mask, *current_byte;
6069                 int bad_bit;
6070                 int j;
6071
6072                 user_byte = (uint8_t *)page_header + i;
6073                 change_mask = page_index->page_data +
6074                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6075                 current_byte = page_index->page_data +
6076                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6077
6078                 /*
6079                  * Check to see whether the user set any bits in this byte
6080                  * that he is not allowed to set.
6081                  */
6082                 if ((*user_byte & ~(*change_mask)) ==
6083                     (*current_byte & ~(*change_mask)))
6084                         continue;
6085
6086                 /*
6087                  * Go through bit by bit to determine which one is illegal.
6088                  */
6089                 bad_bit = 0;
6090                 for (j = 7; j >= 0; j--) {
6091                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6092                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6093                                 bad_bit = i;
6094                                 break;
6095                         }
6096                 }
6097                 ctl_set_invalid_field(ctsio,
6098                                       /*sks_valid*/ 1,
6099                                       /*command*/ 0,
6100                                       /*field*/ *len_used + i,
6101                                       /*bit_valid*/ 1,
6102                                       /*bit*/ bad_bit);
6103                 free(ctsio->kern_data_ptr, M_CTL);
6104                 ctl_done((union ctl_io *)ctsio);
6105                 return (CTL_RETVAL_COMPLETE);
6106         }
6107
6108         /*
6109          * Decrement these before we call the page handler, since we may
6110          * end up getting called back one way or another before the handler
6111          * returns to this context.
6112          */
6113         *len_left -= page_index->page_len;
6114         *len_used += page_index->page_len;
6115
6116         retval = page_index->select_handler(ctsio, page_index,
6117                                             (uint8_t *)page_header);
6118
6119         /*
6120          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6121          * wait until this queued command completes to finish processing
6122          * the mode page.  If it returns anything other than
6123          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6124          * already set the sense information, freed the data pointer, and
6125          * completed the io for us.
6126          */
6127         if (retval != CTL_RETVAL_COMPLETE)
6128                 goto bailout_no_done;
6129
6130         /*
6131          * If the initiator sent us more than one page, parse the next one.
6132          */
6133         if (*len_left > 0)
6134                 goto do_next_page;
6135
6136         ctl_set_success(ctsio);
6137         free(ctsio->kern_data_ptr, M_CTL);
6138         ctl_done((union ctl_io *)ctsio);
6139
6140 bailout_no_done:
6141
6142         return (CTL_RETVAL_COMPLETE);
6143
6144 }
6145
6146 int
6147 ctl_mode_select(struct ctl_scsiio *ctsio)
6148 {
6149         int param_len, pf, sp;
6150         int header_size, bd_len;
6151         union ctl_modepage_info *modepage_info;
6152
6153         switch (ctsio->cdb[0]) {
6154         case MODE_SELECT_6: {
6155                 struct scsi_mode_select_6 *cdb;
6156
6157                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6158
6159                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6160                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6161                 param_len = cdb->length;
6162                 header_size = sizeof(struct scsi_mode_header_6);
6163                 break;
6164         }
6165         case MODE_SELECT_10: {
6166                 struct scsi_mode_select_10 *cdb;
6167
6168                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6169
6170                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6171                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6172                 param_len = scsi_2btoul(cdb->length);
6173                 header_size = sizeof(struct scsi_mode_header_10);
6174                 break;
6175         }
6176         default:
6177                 ctl_set_invalid_opcode(ctsio);
6178                 ctl_done((union ctl_io *)ctsio);
6179                 return (CTL_RETVAL_COMPLETE);
6180         }
6181
6182         /*
6183          * From SPC-3:
6184          * "A parameter list length of zero indicates that the Data-Out Buffer
6185          * shall be empty. This condition shall not be considered as an error."
6186          */
6187         if (param_len == 0) {
6188                 ctl_set_success(ctsio);
6189                 ctl_done((union ctl_io *)ctsio);
6190                 return (CTL_RETVAL_COMPLETE);
6191         }
6192
6193         /*
6194          * Since we'll hit this the first time through, prior to
6195          * allocation, we don't need to free a data buffer here.
6196          */
6197         if (param_len < header_size) {
6198                 ctl_set_param_len_error(ctsio);
6199                 ctl_done((union ctl_io *)ctsio);
6200                 return (CTL_RETVAL_COMPLETE);
6201         }
6202
6203         /*
6204          * Allocate the data buffer and grab the user's data.  In theory,
6205          * we shouldn't have to sanity check the parameter list length here
6206          * because the maximum size is 64K.  We should be able to malloc
6207          * that much without too many problems.
6208          */
6209         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6210                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6211                 ctsio->kern_data_len = param_len;
6212                 ctsio->kern_total_len = param_len;
6213                 ctsio->kern_data_resid = 0;
6214                 ctsio->kern_rel_offset = 0;
6215                 ctsio->kern_sg_entries = 0;
6216                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6217                 ctsio->be_move_done = ctl_config_move_done;
6218                 ctl_datamove((union ctl_io *)ctsio);
6219
6220                 return (CTL_RETVAL_COMPLETE);
6221         }
6222
6223         switch (ctsio->cdb[0]) {
6224         case MODE_SELECT_6: {
6225                 struct scsi_mode_header_6 *mh6;
6226
6227                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6228                 bd_len = mh6->blk_desc_len;
6229                 break;
6230         }
6231         case MODE_SELECT_10: {
6232                 struct scsi_mode_header_10 *mh10;
6233
6234                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6235                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6236                 break;
6237         }
6238         default:
6239                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6240         }
6241
6242         if (param_len < (header_size + bd_len)) {
6243                 free(ctsio->kern_data_ptr, M_CTL);
6244                 ctl_set_param_len_error(ctsio);
6245                 ctl_done((union ctl_io *)ctsio);
6246                 return (CTL_RETVAL_COMPLETE);
6247         }
6248
6249         /*
6250          * Set the IO_CONT flag, so that if this I/O gets passed to
6251          * ctl_config_write_done(), it'll get passed back to
6252          * ctl_do_mode_select() for further processing, or completion if
6253          * we're all done.
6254          */
6255         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6256         ctsio->io_cont = ctl_do_mode_select;
6257
6258         modepage_info = (union ctl_modepage_info *)
6259                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6260         memset(modepage_info, 0, sizeof(*modepage_info));
6261         modepage_info->header.len_left = param_len - header_size - bd_len;
6262         modepage_info->header.len_used = header_size + bd_len;
6263
6264         return (ctl_do_mode_select((union ctl_io *)ctsio));
6265 }
6266
6267 int
6268 ctl_mode_sense(struct ctl_scsiio *ctsio)
6269 {
6270         struct ctl_lun *lun;
6271         int pc, page_code, dbd, llba, subpage;
6272         int alloc_len, page_len, header_len, total_len;
6273         struct scsi_mode_block_descr *block_desc;
6274         struct ctl_page_index *page_index;
6275
6276         dbd = 0;
6277         llba = 0;
6278         block_desc = NULL;
6279
6280         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6281
6282         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6283         switch (ctsio->cdb[0]) {
6284         case MODE_SENSE_6: {
6285                 struct scsi_mode_sense_6 *cdb;
6286
6287                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6288
6289                 header_len = sizeof(struct scsi_mode_hdr_6);
6290                 if (cdb->byte2 & SMS_DBD)
6291                         dbd = 1;
6292                 else
6293                         header_len += sizeof(struct scsi_mode_block_descr);
6294
6295                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6296                 page_code = cdb->page & SMS_PAGE_CODE;
6297                 subpage = cdb->subpage;
6298                 alloc_len = cdb->length;
6299                 break;
6300         }
6301         case MODE_SENSE_10: {
6302                 struct scsi_mode_sense_10 *cdb;
6303
6304                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6305
6306                 header_len = sizeof(struct scsi_mode_hdr_10);
6307
6308                 if (cdb->byte2 & SMS_DBD)
6309                         dbd = 1;
6310                 else
6311                         header_len += sizeof(struct scsi_mode_block_descr);
6312                 if (cdb->byte2 & SMS10_LLBAA)
6313                         llba = 1;
6314                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6315                 page_code = cdb->page & SMS_PAGE_CODE;
6316                 subpage = cdb->subpage;
6317                 alloc_len = scsi_2btoul(cdb->length);
6318                 break;
6319         }
6320         default:
6321                 ctl_set_invalid_opcode(ctsio);
6322                 ctl_done((union ctl_io *)ctsio);
6323                 return (CTL_RETVAL_COMPLETE);
6324                 break; /* NOTREACHED */
6325         }
6326
6327         /*
6328          * We have to make a first pass through to calculate the size of
6329          * the pages that match the user's query.  Then we allocate enough
6330          * memory to hold it, and actually copy the data into the buffer.
6331          */
6332         switch (page_code) {
6333         case SMS_ALL_PAGES_PAGE: {
6334                 u_int i;
6335
6336                 page_len = 0;
6337
6338                 /*
6339                  * At the moment, values other than 0 and 0xff here are
6340                  * reserved according to SPC-3.
6341                  */
6342                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6343                  && (subpage != SMS_SUBPAGE_ALL)) {
6344                         ctl_set_invalid_field(ctsio,
6345                                               /*sks_valid*/ 1,
6346                                               /*command*/ 1,
6347                                               /*field*/ 3,
6348                                               /*bit_valid*/ 0,
6349                                               /*bit*/ 0);
6350                         ctl_done((union ctl_io *)ctsio);
6351                         return (CTL_RETVAL_COMPLETE);
6352                 }
6353
6354                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6355                         page_index = &lun->mode_pages.index[i];
6356
6357                         /* Make sure the page is supported for this dev type */
6358                         if (lun->be_lun->lun_type == T_DIRECT &&
6359                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6360                                 continue;
6361                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6362                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6363                                 continue;
6364                         if (lun->be_lun->lun_type == T_CDROM &&
6365                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6366                                 continue;
6367
6368                         /*
6369                          * We don't use this subpage if the user didn't
6370                          * request all subpages.
6371                          */
6372                         if ((page_index->subpage != 0)
6373                          && (subpage == SMS_SUBPAGE_PAGE_0))
6374                                 continue;
6375
6376 #if 0
6377                         printf("found page %#x len %d\n",
6378                                page_index->page_code & SMPH_PC_MASK,
6379                                page_index->page_len);
6380 #endif
6381                         page_len += page_index->page_len;
6382                 }
6383                 break;
6384         }
6385         default: {
6386                 u_int i;
6387
6388                 page_len = 0;
6389
6390                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6391                         page_index = &lun->mode_pages.index[i];
6392
6393                         /* Make sure the page is supported for this dev type */
6394                         if (lun->be_lun->lun_type == T_DIRECT &&
6395                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6396                                 continue;
6397                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6398                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6399                                 continue;
6400                         if (lun->be_lun->lun_type == T_CDROM &&
6401                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6402                                 continue;
6403
6404                         /* Look for the right page code */
6405                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6406                                 continue;
6407
6408                         /* Look for the right subpage or the subpage wildcard*/
6409                         if ((page_index->subpage != subpage)
6410                          && (subpage != SMS_SUBPAGE_ALL))
6411                                 continue;
6412
6413 #if 0
6414                         printf("found page %#x len %d\n",
6415                                page_index->page_code & SMPH_PC_MASK,
6416                                page_index->page_len);
6417 #endif
6418
6419                         page_len += page_index->page_len;
6420                 }
6421
6422                 if (page_len == 0) {
6423                         ctl_set_invalid_field(ctsio,
6424                                               /*sks_valid*/ 1,
6425                                               /*command*/ 1,
6426                                               /*field*/ 2,
6427                                               /*bit_valid*/ 1,
6428                                               /*bit*/ 5);
6429                         ctl_done((union ctl_io *)ctsio);
6430                         return (CTL_RETVAL_COMPLETE);
6431                 }
6432                 break;
6433         }
6434         }
6435
6436         total_len = header_len + page_len;
6437 #if 0
6438         printf("header_len = %d, page_len = %d, total_len = %d\n",
6439                header_len, page_len, total_len);
6440 #endif
6441
6442         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6443         ctsio->kern_sg_entries = 0;
6444         ctsio->kern_data_resid = 0;
6445         ctsio->kern_rel_offset = 0;
6446         if (total_len < alloc_len) {
6447                 ctsio->residual = alloc_len - total_len;
6448                 ctsio->kern_data_len = total_len;
6449                 ctsio->kern_total_len = total_len;
6450         } else {
6451                 ctsio->residual = 0;
6452                 ctsio->kern_data_len = alloc_len;
6453                 ctsio->kern_total_len = alloc_len;
6454         }
6455
6456         switch (ctsio->cdb[0]) {
6457         case MODE_SENSE_6: {
6458                 struct scsi_mode_hdr_6 *header;
6459
6460                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6461
6462                 header->datalen = MIN(total_len - 1, 254);
6463                 if (lun->be_lun->lun_type == T_DIRECT) {
6464                         header->dev_specific = 0x10; /* DPOFUA */
6465                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6466                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6467                                 header->dev_specific |= 0x80; /* WP */
6468                 }
6469                 if (dbd)
6470                         header->block_descr_len = 0;
6471                 else
6472                         header->block_descr_len =
6473                                 sizeof(struct scsi_mode_block_descr);
6474                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6475                 break;
6476         }
6477         case MODE_SENSE_10: {
6478                 struct scsi_mode_hdr_10 *header;
6479                 int datalen;
6480
6481                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6482
6483                 datalen = MIN(total_len - 2, 65533);
6484                 scsi_ulto2b(datalen, header->datalen);
6485                 if (lun->be_lun->lun_type == T_DIRECT) {
6486                         header->dev_specific = 0x10; /* DPOFUA */
6487                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6488                             (lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0)
6489                                 header->dev_specific |= 0x80; /* WP */
6490                 }
6491                 if (dbd)
6492                         scsi_ulto2b(0, header->block_descr_len);
6493                 else
6494                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6495                                     header->block_descr_len);
6496                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6497                 break;
6498         }
6499         default:
6500                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6501         }
6502
6503         /*
6504          * If we've got a disk, use its blocksize in the block
6505          * descriptor.  Otherwise, just set it to 0.
6506          */
6507         if (dbd == 0) {
6508                 if (lun->be_lun->lun_type == T_DIRECT)
6509                         scsi_ulto3b(lun->be_lun->blocksize,
6510                                     block_desc->block_len);
6511                 else
6512                         scsi_ulto3b(0, block_desc->block_len);
6513         }
6514
6515         switch (page_code) {
6516         case SMS_ALL_PAGES_PAGE: {
6517                 int i, data_used;
6518
6519                 data_used = header_len;
6520                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6521                         struct ctl_page_index *page_index;
6522
6523                         page_index = &lun->mode_pages.index[i];
6524                         if (lun->be_lun->lun_type == T_DIRECT &&
6525                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6526                                 continue;
6527                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6528                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6529                                 continue;
6530                         if (lun->be_lun->lun_type == T_CDROM &&
6531                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6532                                 continue;
6533
6534                         /*
6535                          * We don't use this subpage if the user didn't
6536                          * request all subpages.  We already checked (above)
6537                          * to make sure the user only specified a subpage
6538                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6539                          */
6540                         if ((page_index->subpage != 0)
6541                          && (subpage == SMS_SUBPAGE_PAGE_0))
6542                                 continue;
6543
6544                         /*
6545                          * Call the handler, if it exists, to update the
6546                          * page to the latest values.
6547                          */
6548                         if (page_index->sense_handler != NULL)
6549                                 page_index->sense_handler(ctsio, page_index,pc);
6550
6551                         memcpy(ctsio->kern_data_ptr + data_used,
6552                                page_index->page_data +
6553                                (page_index->page_len * pc),
6554                                page_index->page_len);
6555                         data_used += page_index->page_len;
6556                 }
6557                 break;
6558         }
6559         default: {
6560                 int i, data_used;
6561
6562                 data_used = header_len;
6563
6564                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6565                         struct ctl_page_index *page_index;
6566
6567                         page_index = &lun->mode_pages.index[i];
6568
6569                         /* Look for the right page code */
6570                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6571                                 continue;
6572
6573                         /* Look for the right subpage or the subpage wildcard*/
6574                         if ((page_index->subpage != subpage)
6575                          && (subpage != SMS_SUBPAGE_ALL))
6576                                 continue;
6577
6578                         /* Make sure the page is supported for this dev type */
6579                         if (lun->be_lun->lun_type == T_DIRECT &&
6580                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6581                                 continue;
6582                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6583                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6584                                 continue;
6585                         if (lun->be_lun->lun_type == T_CDROM &&
6586                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6587                                 continue;
6588
6589                         /*
6590                          * Call the handler, if it exists, to update the
6591                          * page to the latest values.
6592                          */
6593                         if (page_index->sense_handler != NULL)
6594                                 page_index->sense_handler(ctsio, page_index,pc);
6595
6596                         memcpy(ctsio->kern_data_ptr + data_used,
6597                                page_index->page_data +
6598                                (page_index->page_len * pc),
6599                                page_index->page_len);
6600                         data_used += page_index->page_len;
6601                 }
6602                 break;
6603         }
6604         }
6605
6606         ctl_set_success(ctsio);
6607         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6608         ctsio->be_move_done = ctl_config_move_done;
6609         ctl_datamove((union ctl_io *)ctsio);
6610         return (CTL_RETVAL_COMPLETE);
6611 }
6612
6613 int
6614 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6615                                struct ctl_page_index *page_index,
6616                                int pc)
6617 {
6618         struct ctl_lun *lun;
6619         struct scsi_log_param_header *phdr;
6620         uint8_t *data;
6621         uint64_t val;
6622
6623         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6624         data = page_index->page_data;
6625
6626         if (lun->backend->lun_attr != NULL &&
6627             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6628              != UINT64_MAX) {
6629                 phdr = (struct scsi_log_param_header *)data;
6630                 scsi_ulto2b(0x0001, phdr->param_code);
6631                 phdr->param_control = SLP_LBIN | SLP_LP;
6632                 phdr->param_len = 8;
6633                 data = (uint8_t *)(phdr + 1);
6634                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6635                 data[4] = 0x02; /* per-pool */
6636                 data += phdr->param_len;
6637         }
6638
6639         if (lun->backend->lun_attr != NULL &&
6640             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6641              != UINT64_MAX) {
6642                 phdr = (struct scsi_log_param_header *)data;
6643                 scsi_ulto2b(0x0002, phdr->param_code);
6644                 phdr->param_control = SLP_LBIN | SLP_LP;
6645                 phdr->param_len = 8;
6646                 data = (uint8_t *)(phdr + 1);
6647                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6648                 data[4] = 0x01; /* per-LUN */
6649                 data += phdr->param_len;
6650         }
6651
6652         if (lun->backend->lun_attr != NULL &&
6653             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6654              != UINT64_MAX) {
6655                 phdr = (struct scsi_log_param_header *)data;
6656                 scsi_ulto2b(0x00f1, phdr->param_code);
6657                 phdr->param_control = SLP_LBIN | SLP_LP;
6658                 phdr->param_len = 8;
6659                 data = (uint8_t *)(phdr + 1);
6660                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6661                 data[4] = 0x02; /* per-pool */
6662                 data += phdr->param_len;
6663         }
6664
6665         if (lun->backend->lun_attr != NULL &&
6666             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6667              != UINT64_MAX) {
6668                 phdr = (struct scsi_log_param_header *)data;
6669                 scsi_ulto2b(0x00f2, phdr->param_code);
6670                 phdr->param_control = SLP_LBIN | SLP_LP;
6671                 phdr->param_len = 8;
6672                 data = (uint8_t *)(phdr + 1);
6673                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6674                 data[4] = 0x02; /* per-pool */
6675                 data += phdr->param_len;
6676         }
6677
6678         page_index->page_len = data - page_index->page_data;
6679         return (0);
6680 }
6681
6682 int
6683 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6684                                struct ctl_page_index *page_index,
6685                                int pc)
6686 {
6687         struct ctl_lun *lun;
6688         struct stat_page *data;
6689         uint64_t rn, wn, rb, wb;
6690         struct bintime rt, wt;
6691         int i;
6692
6693         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6694         data = (struct stat_page *)page_index->page_data;
6695
6696         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6697         data->sap.hdr.param_control = SLP_LBIN;
6698         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6699             sizeof(struct scsi_log_param_header);
6700         rn = wn = rb = wb = 0;
6701         bintime_clear(&rt);
6702         bintime_clear(&wt);
6703         for (i = 0; i < CTL_MAX_PORTS; i++) {
6704                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6705                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6706                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6707                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6708                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6709                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6710         }
6711         scsi_u64to8b(rn, data->sap.read_num);
6712         scsi_u64to8b(wn, data->sap.write_num);
6713         if (lun->stats.blocksize > 0) {
6714                 scsi_u64to8b(wb / lun->stats.blocksize,
6715                     data->sap.recvieved_lba);
6716                 scsi_u64to8b(rb / lun->stats.blocksize,
6717                     data->sap.transmitted_lba);
6718         }
6719         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6720             data->sap.read_int);
6721         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6722             data->sap.write_int);
6723         scsi_u64to8b(0, data->sap.weighted_num);
6724         scsi_u64to8b(0, data->sap.weighted_int);
6725         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6726         data->it.hdr.param_control = SLP_LBIN;
6727         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6728             sizeof(struct scsi_log_param_header);
6729 #ifdef CTL_TIME_IO
6730         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6731 #endif
6732         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6733         data->it.hdr.param_control = SLP_LBIN;
6734         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6735             sizeof(struct scsi_log_param_header);
6736         scsi_ulto4b(3, data->ti.exponent);
6737         scsi_ulto4b(1, data->ti.integer);
6738         return (0);
6739 }
6740
6741 int
6742 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6743                                struct ctl_page_index *page_index,
6744                                int pc)
6745 {
6746         struct ctl_lun *lun;
6747         struct scsi_log_informational_exceptions *data;
6748
6749         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6750         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6751
6752         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6753         data->hdr.param_control = SLP_LBIN;
6754         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6755             sizeof(struct scsi_log_param_header);
6756         data->ie_asc = lun->ie_asc;
6757         data->ie_ascq = lun->ie_ascq;
6758         data->temperature = 0xff;
6759         return (0);
6760 }
6761
6762 int
6763 ctl_log_sense(struct ctl_scsiio *ctsio)
6764 {
6765         struct ctl_lun *lun;
6766         int i, pc, page_code, subpage;
6767         int alloc_len, total_len;
6768         struct ctl_page_index *page_index;
6769         struct scsi_log_sense *cdb;
6770         struct scsi_log_header *header;
6771
6772         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6773
6774         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6775         cdb = (struct scsi_log_sense *)ctsio->cdb;
6776         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6777         page_code = cdb->page & SLS_PAGE_CODE;
6778         subpage = cdb->subpage;
6779         alloc_len = scsi_2btoul(cdb->length);
6780
6781         page_index = NULL;
6782         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6783                 page_index = &lun->log_pages.index[i];
6784
6785                 /* Look for the right page code */
6786                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6787                         continue;
6788
6789                 /* Look for the right subpage or the subpage wildcard*/
6790                 if (page_index->subpage != subpage)
6791                         continue;
6792
6793                 break;
6794         }
6795         if (i >= CTL_NUM_LOG_PAGES) {
6796                 ctl_set_invalid_field(ctsio,
6797                                       /*sks_valid*/ 1,
6798                                       /*command*/ 1,
6799                                       /*field*/ 2,
6800                                       /*bit_valid*/ 0,
6801                                       /*bit*/ 0);
6802                 ctl_done((union ctl_io *)ctsio);
6803                 return (CTL_RETVAL_COMPLETE);
6804         }
6805
6806         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6807
6808         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6809         ctsio->kern_sg_entries = 0;
6810         ctsio->kern_data_resid = 0;
6811         ctsio->kern_rel_offset = 0;
6812         if (total_len < alloc_len) {
6813                 ctsio->residual = alloc_len - total_len;
6814                 ctsio->kern_data_len = total_len;
6815                 ctsio->kern_total_len = total_len;
6816         } else {
6817                 ctsio->residual = 0;
6818                 ctsio->kern_data_len = alloc_len;
6819                 ctsio->kern_total_len = alloc_len;
6820         }
6821
6822         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6823         header->page = page_index->page_code;
6824         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6825                 header->page |= SL_DS;
6826         if (page_index->subpage) {
6827                 header->page |= SL_SPF;
6828                 header->subpage = page_index->subpage;
6829         }
6830         scsi_ulto2b(page_index->page_len, header->datalen);
6831
6832         /*
6833          * Call the handler, if it exists, to update the
6834          * page to the latest values.
6835          */
6836         if (page_index->sense_handler != NULL)
6837                 page_index->sense_handler(ctsio, page_index, pc);
6838
6839         memcpy(header + 1, page_index->page_data, page_index->page_len);
6840
6841         ctl_set_success(ctsio);
6842         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6843         ctsio->be_move_done = ctl_config_move_done;
6844         ctl_datamove((union ctl_io *)ctsio);
6845         return (CTL_RETVAL_COMPLETE);
6846 }
6847
6848 int
6849 ctl_read_capacity(struct ctl_scsiio *ctsio)
6850 {
6851         struct scsi_read_capacity *cdb;
6852         struct scsi_read_capacity_data *data;
6853         struct ctl_lun *lun;
6854         uint32_t lba;
6855
6856         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6857
6858         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6859
6860         lba = scsi_4btoul(cdb->addr);
6861         if (((cdb->pmi & SRC_PMI) == 0)
6862          && (lba != 0)) {
6863                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6864                                       /*sks_valid*/ 1,
6865                                       /*command*/ 1,
6866                                       /*field*/ 2,
6867                                       /*bit_valid*/ 0,
6868                                       /*bit*/ 0);
6869                 ctl_done((union ctl_io *)ctsio);
6870                 return (CTL_RETVAL_COMPLETE);
6871         }
6872
6873         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6874
6875         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6876         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6877         ctsio->residual = 0;
6878         ctsio->kern_data_len = sizeof(*data);
6879         ctsio->kern_total_len = sizeof(*data);
6880         ctsio->kern_data_resid = 0;
6881         ctsio->kern_rel_offset = 0;
6882         ctsio->kern_sg_entries = 0;
6883
6884         /*
6885          * If the maximum LBA is greater than 0xfffffffe, the user must
6886          * issue a SERVICE ACTION IN (16) command, with the read capacity
6887          * serivce action set.
6888          */
6889         if (lun->be_lun->maxlba > 0xfffffffe)
6890                 scsi_ulto4b(0xffffffff, data->addr);
6891         else
6892                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6893
6894         /*
6895          * XXX KDM this may not be 512 bytes...
6896          */
6897         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6898
6899         ctl_set_success(ctsio);
6900         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6901         ctsio->be_move_done = ctl_config_move_done;
6902         ctl_datamove((union ctl_io *)ctsio);
6903         return (CTL_RETVAL_COMPLETE);
6904 }
6905
6906 int
6907 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6908 {
6909         struct scsi_read_capacity_16 *cdb;
6910         struct scsi_read_capacity_data_long *data;
6911         struct ctl_lun *lun;
6912         uint64_t lba;
6913         uint32_t alloc_len;
6914
6915         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6916
6917         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6918
6919         alloc_len = scsi_4btoul(cdb->alloc_len);
6920         lba = scsi_8btou64(cdb->addr);
6921
6922         if ((cdb->reladr & SRC16_PMI)
6923          && (lba != 0)) {
6924                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6925                                       /*sks_valid*/ 1,
6926                                       /*command*/ 1,
6927                                       /*field*/ 2,
6928                                       /*bit_valid*/ 0,
6929                                       /*bit*/ 0);
6930                 ctl_done((union ctl_io *)ctsio);
6931                 return (CTL_RETVAL_COMPLETE);
6932         }
6933
6934         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6935
6936         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6937         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6938
6939         if (sizeof(*data) < alloc_len) {
6940                 ctsio->residual = alloc_len - sizeof(*data);
6941                 ctsio->kern_data_len = sizeof(*data);
6942                 ctsio->kern_total_len = sizeof(*data);
6943         } else {
6944                 ctsio->residual = 0;
6945                 ctsio->kern_data_len = alloc_len;
6946                 ctsio->kern_total_len = alloc_len;
6947         }
6948         ctsio->kern_data_resid = 0;
6949         ctsio->kern_rel_offset = 0;
6950         ctsio->kern_sg_entries = 0;
6951
6952         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6953         /* XXX KDM this may not be 512 bytes... */
6954         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6955         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6956         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6957         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6958                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6959
6960         ctl_set_success(ctsio);
6961         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6962         ctsio->be_move_done = ctl_config_move_done;
6963         ctl_datamove((union ctl_io *)ctsio);
6964         return (CTL_RETVAL_COMPLETE);
6965 }
6966
6967 int
6968 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6969 {
6970         struct scsi_get_lba_status *cdb;
6971         struct scsi_get_lba_status_data *data;
6972         struct ctl_lun *lun;
6973         struct ctl_lba_len_flags *lbalen;
6974         uint64_t lba;
6975         uint32_t alloc_len, total_len;
6976         int retval;
6977
6978         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6979
6980         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6981         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6982         lba = scsi_8btou64(cdb->addr);
6983         alloc_len = scsi_4btoul(cdb->alloc_len);
6984
6985         if (lba > lun->be_lun->maxlba) {
6986                 ctl_set_lba_out_of_range(ctsio, lba);
6987                 ctl_done((union ctl_io *)ctsio);
6988                 return (CTL_RETVAL_COMPLETE);
6989         }
6990
6991         total_len = sizeof(*data) + sizeof(data->descr[0]);
6992         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6993         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
6994
6995         if (total_len < alloc_len) {
6996                 ctsio->residual = alloc_len - total_len;
6997                 ctsio->kern_data_len = total_len;
6998                 ctsio->kern_total_len = total_len;
6999         } else {
7000                 ctsio->residual = 0;
7001                 ctsio->kern_data_len = alloc_len;
7002                 ctsio->kern_total_len = alloc_len;
7003         }
7004         ctsio->kern_data_resid = 0;
7005         ctsio->kern_rel_offset = 0;
7006         ctsio->kern_sg_entries = 0;
7007
7008         /* Fill dummy data in case backend can't tell anything. */
7009         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7010         scsi_u64to8b(lba, data->descr[0].addr);
7011         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7012             data->descr[0].length);
7013         data->descr[0].status = 0; /* Mapped or unknown. */
7014
7015         ctl_set_success(ctsio);
7016         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7017         ctsio->be_move_done = ctl_config_move_done;
7018
7019         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7020         lbalen->lba = lba;
7021         lbalen->len = total_len;
7022         lbalen->flags = 0;
7023         retval = lun->backend->config_read((union ctl_io *)ctsio);
7024         return (CTL_RETVAL_COMPLETE);
7025 }
7026
7027 int
7028 ctl_read_defect(struct ctl_scsiio *ctsio)
7029 {
7030         struct scsi_read_defect_data_10 *ccb10;
7031         struct scsi_read_defect_data_12 *ccb12;
7032         struct scsi_read_defect_data_hdr_10 *data10;
7033         struct scsi_read_defect_data_hdr_12 *data12;
7034         uint32_t alloc_len, data_len;
7035         uint8_t format;
7036
7037         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7038
7039         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7040                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7041                 format = ccb10->format;
7042                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7043                 data_len = sizeof(*data10);
7044         } else {
7045                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7046                 format = ccb12->format;
7047                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7048                 data_len = sizeof(*data12);
7049         }
7050         if (alloc_len == 0) {
7051                 ctl_set_success(ctsio);
7052                 ctl_done((union ctl_io *)ctsio);
7053                 return (CTL_RETVAL_COMPLETE);
7054         }
7055
7056         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7057         if (data_len < alloc_len) {
7058                 ctsio->residual = alloc_len - data_len;
7059                 ctsio->kern_data_len = data_len;
7060                 ctsio->kern_total_len = data_len;
7061         } else {
7062                 ctsio->residual = 0;
7063                 ctsio->kern_data_len = alloc_len;
7064                 ctsio->kern_total_len = alloc_len;
7065         }
7066         ctsio->kern_data_resid = 0;
7067         ctsio->kern_rel_offset = 0;
7068         ctsio->kern_sg_entries = 0;
7069
7070         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7071                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7072                     ctsio->kern_data_ptr;
7073                 data10->format = format;
7074                 scsi_ulto2b(0, data10->length);
7075         } else {
7076                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7077                     ctsio->kern_data_ptr;
7078                 data12->format = format;
7079                 scsi_ulto2b(0, data12->generation);
7080                 scsi_ulto4b(0, data12->length);
7081         }
7082
7083         ctl_set_success(ctsio);
7084         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7085         ctsio->be_move_done = ctl_config_move_done;
7086         ctl_datamove((union ctl_io *)ctsio);
7087         return (CTL_RETVAL_COMPLETE);
7088 }
7089
7090 int
7091 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7092 {
7093         struct scsi_maintenance_in *cdb;
7094         int retval;
7095         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7096         int num_ha_groups, num_target_ports, shared_group;
7097         struct ctl_lun *lun;
7098         struct ctl_softc *softc;
7099         struct ctl_port *port;
7100         struct scsi_target_group_data *rtg_ptr;
7101         struct scsi_target_group_data_extended *rtg_ext_ptr;
7102         struct scsi_target_port_group_descriptor *tpg_desc;
7103
7104         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7105
7106         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7107         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7108         softc = lun->ctl_softc;
7109
7110         retval = CTL_RETVAL_COMPLETE;
7111
7112         switch (cdb->byte2 & STG_PDF_MASK) {
7113         case STG_PDF_LENGTH:
7114                 ext = 0;
7115                 break;
7116         case STG_PDF_EXTENDED:
7117                 ext = 1;
7118                 break;
7119         default:
7120                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7121                                       /*sks_valid*/ 1,
7122                                       /*command*/ 1,
7123                                       /*field*/ 2,
7124                                       /*bit_valid*/ 1,
7125                                       /*bit*/ 5);
7126                 ctl_done((union ctl_io *)ctsio);
7127                 return(retval);
7128         }
7129
7130         num_target_ports = 0;
7131         shared_group = (softc->is_single != 0);
7132         mtx_lock(&softc->ctl_lock);
7133         STAILQ_FOREACH(port, &softc->port_list, links) {
7134                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7135                         continue;
7136                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7137                         continue;
7138                 num_target_ports++;
7139                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7140                         shared_group = 1;
7141         }
7142         mtx_unlock(&softc->ctl_lock);
7143         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7144
7145         if (ext)
7146                 total_len = sizeof(struct scsi_target_group_data_extended);
7147         else
7148                 total_len = sizeof(struct scsi_target_group_data);
7149         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7150                 (shared_group + num_ha_groups) +
7151             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7152
7153         alloc_len = scsi_4btoul(cdb->length);
7154
7155         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7156
7157         ctsio->kern_sg_entries = 0;
7158
7159         if (total_len < alloc_len) {
7160                 ctsio->residual = alloc_len - total_len;
7161                 ctsio->kern_data_len = total_len;
7162                 ctsio->kern_total_len = total_len;
7163         } else {
7164                 ctsio->residual = 0;
7165                 ctsio->kern_data_len = alloc_len;
7166                 ctsio->kern_total_len = alloc_len;
7167         }
7168         ctsio->kern_data_resid = 0;
7169         ctsio->kern_rel_offset = 0;
7170
7171         if (ext) {
7172                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7173                     ctsio->kern_data_ptr;
7174                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7175                 rtg_ext_ptr->format_type = 0x10;
7176                 rtg_ext_ptr->implicit_transition_time = 0;
7177                 tpg_desc = &rtg_ext_ptr->groups[0];
7178         } else {
7179                 rtg_ptr = (struct scsi_target_group_data *)
7180                     ctsio->kern_data_ptr;
7181                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7182                 tpg_desc = &rtg_ptr->groups[0];
7183         }
7184
7185         mtx_lock(&softc->ctl_lock);
7186         pg = softc->port_min / softc->port_cnt;
7187         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7188                 /* Some shelf is known to be primary. */
7189                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7190                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7191                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7192                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7193                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7194                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7195                 else
7196                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7197                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7198                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7199                 } else {
7200                         ts = os;
7201                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7202                 }
7203         } else {
7204                 /* No known primary shelf. */
7205                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7206                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7207                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7208                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7209                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7210                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7211                 } else {
7212                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7213                 }
7214         }
7215         if (shared_group) {
7216                 tpg_desc->pref_state = ts;
7217                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7218                     TPG_U_SUP | TPG_T_SUP;
7219                 scsi_ulto2b(1, tpg_desc->target_port_group);
7220                 tpg_desc->status = TPG_IMPLICIT;
7221                 pc = 0;
7222                 STAILQ_FOREACH(port, &softc->port_list, links) {
7223                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7224                                 continue;
7225                         if (!softc->is_single &&
7226                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7227                                 continue;
7228                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7229                                 continue;
7230                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7231                             relative_target_port_identifier);
7232                         pc++;
7233                 }
7234                 tpg_desc->target_port_count = pc;
7235                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7236                     &tpg_desc->descriptors[pc];
7237         }
7238         for (g = 0; g < num_ha_groups; g++) {
7239                 tpg_desc->pref_state = (g == pg) ? ts : os;
7240                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7241                     TPG_U_SUP | TPG_T_SUP;
7242                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7243                 tpg_desc->status = TPG_IMPLICIT;
7244                 pc = 0;
7245                 STAILQ_FOREACH(port, &softc->port_list, links) {
7246                         if (port->targ_port < g * softc->port_cnt ||
7247                             port->targ_port >= (g + 1) * softc->port_cnt)
7248                                 continue;
7249                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7250                                 continue;
7251                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7252                                 continue;
7253                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7254                                 continue;
7255                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7256                             relative_target_port_identifier);
7257                         pc++;
7258                 }
7259                 tpg_desc->target_port_count = pc;
7260                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7261                     &tpg_desc->descriptors[pc];
7262         }
7263         mtx_unlock(&softc->ctl_lock);
7264
7265         ctl_set_success(ctsio);
7266         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7267         ctsio->be_move_done = ctl_config_move_done;
7268         ctl_datamove((union ctl_io *)ctsio);
7269         return(retval);
7270 }
7271
7272 int
7273 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7274 {
7275         struct ctl_lun *lun;
7276         struct scsi_report_supported_opcodes *cdb;
7277         const struct ctl_cmd_entry *entry, *sentry;
7278         struct scsi_report_supported_opcodes_all *all;
7279         struct scsi_report_supported_opcodes_descr *descr;
7280         struct scsi_report_supported_opcodes_one *one;
7281         int retval;
7282         int alloc_len, total_len;
7283         int opcode, service_action, i, j, num;
7284
7285         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7286
7287         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7288         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7289
7290         retval = CTL_RETVAL_COMPLETE;
7291
7292         opcode = cdb->requested_opcode;
7293         service_action = scsi_2btoul(cdb->requested_service_action);
7294         switch (cdb->options & RSO_OPTIONS_MASK) {
7295         case RSO_OPTIONS_ALL:
7296                 num = 0;
7297                 for (i = 0; i < 256; i++) {
7298                         entry = &ctl_cmd_table[i];
7299                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7300                                 for (j = 0; j < 32; j++) {
7301                                         sentry = &((const struct ctl_cmd_entry *)
7302                                             entry->execute)[j];
7303                                         if (ctl_cmd_applicable(
7304                                             lun->be_lun->lun_type, sentry))
7305                                                 num++;
7306                                 }
7307                         } else {
7308                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7309                                     entry))
7310                                         num++;
7311                         }
7312                 }
7313                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7314                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7315                 break;
7316         case RSO_OPTIONS_OC:
7317                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7318                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7319                                               /*sks_valid*/ 1,
7320                                               /*command*/ 1,
7321                                               /*field*/ 2,
7322                                               /*bit_valid*/ 1,
7323                                               /*bit*/ 2);
7324                         ctl_done((union ctl_io *)ctsio);
7325                         return (CTL_RETVAL_COMPLETE);
7326                 }
7327                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7328                 break;
7329         case RSO_OPTIONS_OC_SA:
7330                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7331                     service_action >= 32) {
7332                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7333                                               /*sks_valid*/ 1,
7334                                               /*command*/ 1,
7335                                               /*field*/ 2,
7336                                               /*bit_valid*/ 1,
7337                                               /*bit*/ 2);
7338                         ctl_done((union ctl_io *)ctsio);
7339                         return (CTL_RETVAL_COMPLETE);
7340                 }
7341                 /* FALLTHROUGH */
7342         case RSO_OPTIONS_OC_ASA:
7343                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7344                 break;
7345         default:
7346                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7347                                       /*sks_valid*/ 1,
7348                                       /*command*/ 1,
7349                                       /*field*/ 2,
7350                                       /*bit_valid*/ 1,
7351                                       /*bit*/ 2);
7352                 ctl_done((union ctl_io *)ctsio);
7353                 return (CTL_RETVAL_COMPLETE);
7354         }
7355
7356         alloc_len = scsi_4btoul(cdb->length);
7357
7358         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7359
7360         ctsio->kern_sg_entries = 0;
7361
7362         if (total_len < alloc_len) {
7363                 ctsio->residual = alloc_len - total_len;
7364                 ctsio->kern_data_len = total_len;
7365                 ctsio->kern_total_len = total_len;
7366         } else {
7367                 ctsio->residual = 0;
7368                 ctsio->kern_data_len = alloc_len;
7369                 ctsio->kern_total_len = alloc_len;
7370         }
7371         ctsio->kern_data_resid = 0;
7372         ctsio->kern_rel_offset = 0;
7373
7374         switch (cdb->options & RSO_OPTIONS_MASK) {
7375         case RSO_OPTIONS_ALL:
7376                 all = (struct scsi_report_supported_opcodes_all *)
7377                     ctsio->kern_data_ptr;
7378                 num = 0;
7379                 for (i = 0; i < 256; i++) {
7380                         entry = &ctl_cmd_table[i];
7381                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7382                                 for (j = 0; j < 32; j++) {
7383                                         sentry = &((const struct ctl_cmd_entry *)
7384                                             entry->execute)[j];
7385                                         if (!ctl_cmd_applicable(
7386                                             lun->be_lun->lun_type, sentry))
7387                                                 continue;
7388                                         descr = &all->descr[num++];
7389                                         descr->opcode = i;
7390                                         scsi_ulto2b(j, descr->service_action);
7391                                         descr->flags = RSO_SERVACTV;
7392                                         scsi_ulto2b(sentry->length,
7393                                             descr->cdb_length);
7394                                 }
7395                         } else {
7396                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7397                                     entry))
7398                                         continue;
7399                                 descr = &all->descr[num++];
7400                                 descr->opcode = i;
7401                                 scsi_ulto2b(0, descr->service_action);
7402                                 descr->flags = 0;
7403                                 scsi_ulto2b(entry->length, descr->cdb_length);
7404                         }
7405                 }
7406                 scsi_ulto4b(
7407                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7408                     all->length);
7409                 break;
7410         case RSO_OPTIONS_OC:
7411                 one = (struct scsi_report_supported_opcodes_one *)
7412                     ctsio->kern_data_ptr;
7413                 entry = &ctl_cmd_table[opcode];
7414                 goto fill_one;
7415         case RSO_OPTIONS_OC_SA:
7416                 one = (struct scsi_report_supported_opcodes_one *)
7417                     ctsio->kern_data_ptr;
7418                 entry = &ctl_cmd_table[opcode];
7419                 entry = &((const struct ctl_cmd_entry *)
7420                     entry->execute)[service_action];
7421 fill_one:
7422                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7423                         one->support = 3;
7424                         scsi_ulto2b(entry->length, one->cdb_length);
7425                         one->cdb_usage[0] = opcode;
7426                         memcpy(&one->cdb_usage[1], entry->usage,
7427                             entry->length - 1);
7428                 } else
7429                         one->support = 1;
7430                 break;
7431         case RSO_OPTIONS_OC_ASA:
7432                 one = (struct scsi_report_supported_opcodes_one *)
7433                     ctsio->kern_data_ptr;
7434                 entry = &ctl_cmd_table[opcode];
7435                 if (entry->flags & CTL_CMD_FLAG_SA5) {
7436                         entry = &((const struct ctl_cmd_entry *)
7437                             entry->execute)[service_action];
7438                 } else if (service_action != 0) {
7439                         one->support = 1;
7440                         break;
7441                 }
7442                 goto fill_one;
7443         }
7444
7445         ctl_set_success(ctsio);
7446         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7447         ctsio->be_move_done = ctl_config_move_done;
7448         ctl_datamove((union ctl_io *)ctsio);
7449         return(retval);
7450 }
7451
7452 int
7453 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7454 {
7455         struct scsi_report_supported_tmf *cdb;
7456         struct scsi_report_supported_tmf_ext_data *data;
7457         int retval;
7458         int alloc_len, total_len;
7459
7460         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7461
7462         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7463
7464         retval = CTL_RETVAL_COMPLETE;
7465
7466         if (cdb->options & RST_REPD)
7467                 total_len = sizeof(struct scsi_report_supported_tmf_ext_data);
7468         else
7469                 total_len = sizeof(struct scsi_report_supported_tmf_data);
7470         alloc_len = scsi_4btoul(cdb->length);
7471
7472         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7473
7474         ctsio->kern_sg_entries = 0;
7475
7476         if (total_len < alloc_len) {
7477                 ctsio->residual = alloc_len - total_len;
7478                 ctsio->kern_data_len = total_len;
7479                 ctsio->kern_total_len = total_len;
7480         } else {
7481                 ctsio->residual = 0;
7482                 ctsio->kern_data_len = alloc_len;
7483                 ctsio->kern_total_len = alloc_len;
7484         }
7485         ctsio->kern_data_resid = 0;
7486         ctsio->kern_rel_offset = 0;
7487
7488         data = (struct scsi_report_supported_tmf_ext_data *)ctsio->kern_data_ptr;
7489         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7490             RST_TRS;
7491         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7492         data->length = total_len - 4;
7493
7494         ctl_set_success(ctsio);
7495         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7496         ctsio->be_move_done = ctl_config_move_done;
7497         ctl_datamove((union ctl_io *)ctsio);
7498         return (retval);
7499 }
7500
7501 int
7502 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7503 {
7504         struct scsi_report_timestamp *cdb;
7505         struct scsi_report_timestamp_data *data;
7506         struct timeval tv;
7507         int64_t timestamp;
7508         int retval;
7509         int alloc_len, total_len;
7510
7511         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7512
7513         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7514
7515         retval = CTL_RETVAL_COMPLETE;
7516
7517         total_len = sizeof(struct scsi_report_timestamp_data);
7518         alloc_len = scsi_4btoul(cdb->length);
7519
7520         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7521
7522         ctsio->kern_sg_entries = 0;
7523
7524         if (total_len < alloc_len) {
7525                 ctsio->residual = alloc_len - total_len;
7526                 ctsio->kern_data_len = total_len;
7527                 ctsio->kern_total_len = total_len;
7528         } else {
7529                 ctsio->residual = 0;
7530                 ctsio->kern_data_len = alloc_len;
7531                 ctsio->kern_total_len = alloc_len;
7532         }
7533         ctsio->kern_data_resid = 0;
7534         ctsio->kern_rel_offset = 0;
7535
7536         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7537         scsi_ulto2b(sizeof(*data) - 2, data->length);
7538         data->origin = RTS_ORIG_OUTSIDE;
7539         getmicrotime(&tv);
7540         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7541         scsi_ulto4b(timestamp >> 16, data->timestamp);
7542         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7543
7544         ctl_set_success(ctsio);
7545         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7546         ctsio->be_move_done = ctl_config_move_done;
7547         ctl_datamove((union ctl_io *)ctsio);
7548         return (retval);
7549 }
7550
7551 int
7552 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7553 {
7554         struct scsi_per_res_in *cdb;
7555         int alloc_len, total_len = 0;
7556         /* struct scsi_per_res_in_rsrv in_data; */
7557         struct ctl_lun *lun;
7558         struct ctl_softc *softc;
7559         uint64_t key;
7560
7561         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7562
7563         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7564
7565         alloc_len = scsi_2btoul(cdb->length);
7566
7567         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7568         softc = lun->ctl_softc;
7569
7570 retry:
7571         mtx_lock(&lun->lun_lock);
7572         switch (cdb->action) {
7573         case SPRI_RK: /* read keys */
7574                 total_len = sizeof(struct scsi_per_res_in_keys) +
7575                         lun->pr_key_count *
7576                         sizeof(struct scsi_per_res_key);
7577                 break;
7578         case SPRI_RR: /* read reservation */
7579                 if (lun->flags & CTL_LUN_PR_RESERVED)
7580                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7581                 else
7582                         total_len = sizeof(struct scsi_per_res_in_header);
7583                 break;
7584         case SPRI_RC: /* report capabilities */
7585                 total_len = sizeof(struct scsi_per_res_cap);
7586                 break;
7587         case SPRI_RS: /* read full status */
7588                 total_len = sizeof(struct scsi_per_res_in_header) +
7589                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7590                     lun->pr_key_count;
7591                 break;
7592         default:
7593                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7594         }
7595         mtx_unlock(&lun->lun_lock);
7596
7597         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7598
7599         if (total_len < alloc_len) {
7600                 ctsio->residual = alloc_len - total_len;
7601                 ctsio->kern_data_len = total_len;
7602                 ctsio->kern_total_len = total_len;
7603         } else {
7604                 ctsio->residual = 0;
7605                 ctsio->kern_data_len = alloc_len;
7606                 ctsio->kern_total_len = alloc_len;
7607         }
7608
7609         ctsio->kern_data_resid = 0;
7610         ctsio->kern_rel_offset = 0;
7611         ctsio->kern_sg_entries = 0;
7612
7613         mtx_lock(&lun->lun_lock);
7614         switch (cdb->action) {
7615         case SPRI_RK: { // read keys
7616         struct scsi_per_res_in_keys *res_keys;
7617                 int i, key_count;
7618
7619                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7620
7621                 /*
7622                  * We had to drop the lock to allocate our buffer, which
7623                  * leaves time for someone to come in with another
7624                  * persistent reservation.  (That is unlikely, though,
7625                  * since this should be the only persistent reservation
7626                  * command active right now.)
7627                  */
7628                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7629                     (lun->pr_key_count *
7630                      sizeof(struct scsi_per_res_key)))){
7631                         mtx_unlock(&lun->lun_lock);
7632                         free(ctsio->kern_data_ptr, M_CTL);
7633                         printf("%s: reservation length changed, retrying\n",
7634                                __func__);
7635                         goto retry;
7636                 }
7637
7638                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7639
7640                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7641                              lun->pr_key_count, res_keys->header.length);
7642
7643                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7644                         if ((key = ctl_get_prkey(lun, i)) == 0)
7645                                 continue;
7646
7647                         /*
7648                          * We used lun->pr_key_count to calculate the
7649                          * size to allocate.  If it turns out the number of
7650                          * initiators with the registered flag set is
7651                          * larger than that (i.e. they haven't been kept in
7652                          * sync), we've got a problem.
7653                          */
7654                         if (key_count >= lun->pr_key_count) {
7655                                 key_count++;
7656                                 continue;
7657                         }
7658                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7659                         key_count++;
7660                 }
7661                 break;
7662         }
7663         case SPRI_RR: { // read reservation
7664                 struct scsi_per_res_in_rsrv *res;
7665                 int tmp_len, header_only;
7666
7667                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7668
7669                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7670
7671                 if (lun->flags & CTL_LUN_PR_RESERVED)
7672                 {
7673                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7674                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7675                                     res->header.length);
7676                         header_only = 0;
7677                 } else {
7678                         tmp_len = sizeof(struct scsi_per_res_in_header);
7679                         scsi_ulto4b(0, res->header.length);
7680                         header_only = 1;
7681                 }
7682
7683                 /*
7684                  * We had to drop the lock to allocate our buffer, which
7685                  * leaves time for someone to come in with another
7686                  * persistent reservation.  (That is unlikely, though,
7687                  * since this should be the only persistent reservation
7688                  * command active right now.)
7689                  */
7690                 if (tmp_len != total_len) {
7691                         mtx_unlock(&lun->lun_lock);
7692                         free(ctsio->kern_data_ptr, M_CTL);
7693                         printf("%s: reservation status changed, retrying\n",
7694                                __func__);
7695                         goto retry;
7696                 }
7697
7698                 /*
7699                  * No reservation held, so we're done.
7700                  */
7701                 if (header_only != 0)
7702                         break;
7703
7704                 /*
7705                  * If the registration is an All Registrants type, the key
7706                  * is 0, since it doesn't really matter.
7707                  */
7708                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7709                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7710                             res->data.reservation);
7711                 }
7712                 res->data.scopetype = lun->pr_res_type;
7713                 break;
7714         }
7715         case SPRI_RC:     //report capabilities
7716         {
7717                 struct scsi_per_res_cap *res_cap;
7718                 uint16_t type_mask;
7719
7720                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7721                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7722                 res_cap->flags1 = SPRI_CRH;
7723                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7724                 type_mask = SPRI_TM_WR_EX_AR |
7725                             SPRI_TM_EX_AC_RO |
7726                             SPRI_TM_WR_EX_RO |
7727                             SPRI_TM_EX_AC |
7728                             SPRI_TM_WR_EX |
7729                             SPRI_TM_EX_AC_AR;
7730                 scsi_ulto2b(type_mask, res_cap->type_mask);
7731                 break;
7732         }
7733         case SPRI_RS: { // read full status
7734                 struct scsi_per_res_in_full *res_status;
7735                 struct scsi_per_res_in_full_desc *res_desc;
7736                 struct ctl_port *port;
7737                 int i, len;
7738
7739                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7740
7741                 /*
7742                  * We had to drop the lock to allocate our buffer, which
7743                  * leaves time for someone to come in with another
7744                  * persistent reservation.  (That is unlikely, though,
7745                  * since this should be the only persistent reservation
7746                  * command active right now.)
7747                  */
7748                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7749                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7750                      lun->pr_key_count)){
7751                         mtx_unlock(&lun->lun_lock);
7752                         free(ctsio->kern_data_ptr, M_CTL);
7753                         printf("%s: reservation length changed, retrying\n",
7754                                __func__);
7755                         goto retry;
7756                 }
7757
7758                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7759
7760                 res_desc = &res_status->desc[0];
7761                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7762                         if ((key = ctl_get_prkey(lun, i)) == 0)
7763                                 continue;
7764
7765                         scsi_u64to8b(key, res_desc->res_key.key);
7766                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7767                             (lun->pr_res_idx == i ||
7768                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7769                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7770                                 res_desc->scopetype = lun->pr_res_type;
7771                         }
7772                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7773                             res_desc->rel_trgt_port_id);
7774                         len = 0;
7775                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7776                         if (port != NULL)
7777                                 len = ctl_create_iid(port,
7778                                     i % CTL_MAX_INIT_PER_PORT,
7779                                     res_desc->transport_id);
7780                         scsi_ulto4b(len, res_desc->additional_length);
7781                         res_desc = (struct scsi_per_res_in_full_desc *)
7782                             &res_desc->transport_id[len];
7783                 }
7784                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7785                     res_status->header.length);
7786                 break;
7787         }
7788         default:
7789                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7790         }
7791         mtx_unlock(&lun->lun_lock);
7792
7793         ctl_set_success(ctsio);
7794         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7795         ctsio->be_move_done = ctl_config_move_done;
7796         ctl_datamove((union ctl_io *)ctsio);
7797         return (CTL_RETVAL_COMPLETE);
7798 }
7799
7800 /*
7801  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7802  * it should return.
7803  */
7804 static int
7805 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7806                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7807                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7808                 struct scsi_per_res_out_parms* param)
7809 {
7810         union ctl_ha_msg persis_io;
7811         int i;
7812
7813         mtx_lock(&lun->lun_lock);
7814         if (sa_res_key == 0) {
7815                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7816                         /* validate scope and type */
7817                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7818                              SPR_LU_SCOPE) {
7819                                 mtx_unlock(&lun->lun_lock);
7820                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7821                                                       /*sks_valid*/ 1,
7822                                                       /*command*/ 1,
7823                                                       /*field*/ 2,
7824                                                       /*bit_valid*/ 1,
7825                                                       /*bit*/ 4);
7826                                 ctl_done((union ctl_io *)ctsio);
7827                                 return (1);
7828                         }
7829
7830                         if (type>8 || type==2 || type==4 || type==0) {
7831                                 mtx_unlock(&lun->lun_lock);
7832                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7833                                                       /*sks_valid*/ 1,
7834                                                       /*command*/ 1,
7835                                                       /*field*/ 2,
7836                                                       /*bit_valid*/ 1,
7837                                                       /*bit*/ 0);
7838                                 ctl_done((union ctl_io *)ctsio);
7839                                 return (1);
7840                         }
7841
7842                         /*
7843                          * Unregister everybody else and build UA for
7844                          * them
7845                          */
7846                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7847                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7848                                         continue;
7849
7850                                 ctl_clr_prkey(lun, i);
7851                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7852                         }
7853                         lun->pr_key_count = 1;
7854                         lun->pr_res_type = type;
7855                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7856                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7857                                 lun->pr_res_idx = residx;
7858                         lun->pr_generation++;
7859                         mtx_unlock(&lun->lun_lock);
7860
7861                         /* send msg to other side */
7862                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7863                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7864                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7865                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7866                         persis_io.pr.pr_info.res_type = type;
7867                         memcpy(persis_io.pr.pr_info.sa_res_key,
7868                                param->serv_act_res_key,
7869                                sizeof(param->serv_act_res_key));
7870                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7871                             sizeof(persis_io.pr), M_WAITOK);
7872                 } else {
7873                         /* not all registrants */
7874                         mtx_unlock(&lun->lun_lock);
7875                         free(ctsio->kern_data_ptr, M_CTL);
7876                         ctl_set_invalid_field(ctsio,
7877                                               /*sks_valid*/ 1,
7878                                               /*command*/ 0,
7879                                               /*field*/ 8,
7880                                               /*bit_valid*/ 0,
7881                                               /*bit*/ 0);
7882                         ctl_done((union ctl_io *)ctsio);
7883                         return (1);
7884                 }
7885         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7886                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7887                 int found = 0;
7888
7889                 if (res_key == sa_res_key) {
7890                         /* special case */
7891                         /*
7892                          * The spec implies this is not good but doesn't
7893                          * say what to do. There are two choices either
7894                          * generate a res conflict or check condition
7895                          * with illegal field in parameter data. Since
7896                          * that is what is done when the sa_res_key is
7897                          * zero I'll take that approach since this has
7898                          * to do with the sa_res_key.
7899                          */
7900                         mtx_unlock(&lun->lun_lock);
7901                         free(ctsio->kern_data_ptr, M_CTL);
7902                         ctl_set_invalid_field(ctsio,
7903                                               /*sks_valid*/ 1,
7904                                               /*command*/ 0,
7905                                               /*field*/ 8,
7906                                               /*bit_valid*/ 0,
7907                                               /*bit*/ 0);
7908                         ctl_done((union ctl_io *)ctsio);
7909                         return (1);
7910                 }
7911
7912                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7913                         if (ctl_get_prkey(lun, i) != sa_res_key)
7914                                 continue;
7915
7916                         found = 1;
7917                         ctl_clr_prkey(lun, i);
7918                         lun->pr_key_count--;
7919                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7920                 }
7921                 if (!found) {
7922                         mtx_unlock(&lun->lun_lock);
7923                         free(ctsio->kern_data_ptr, M_CTL);
7924                         ctl_set_reservation_conflict(ctsio);
7925                         ctl_done((union ctl_io *)ctsio);
7926                         return (CTL_RETVAL_COMPLETE);
7927                 }
7928                 lun->pr_generation++;
7929                 mtx_unlock(&lun->lun_lock);
7930
7931                 /* send msg to other side */
7932                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7933                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7934                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7935                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7936                 persis_io.pr.pr_info.res_type = type;
7937                 memcpy(persis_io.pr.pr_info.sa_res_key,
7938                        param->serv_act_res_key,
7939                        sizeof(param->serv_act_res_key));
7940                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7941                     sizeof(persis_io.pr), M_WAITOK);
7942         } else {
7943                 /* Reserved but not all registrants */
7944                 /* sa_res_key is res holder */
7945                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7946                         /* validate scope and type */
7947                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7948                              SPR_LU_SCOPE) {
7949                                 mtx_unlock(&lun->lun_lock);
7950                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7951                                                       /*sks_valid*/ 1,
7952                                                       /*command*/ 1,
7953                                                       /*field*/ 2,
7954                                                       /*bit_valid*/ 1,
7955                                                       /*bit*/ 4);
7956                                 ctl_done((union ctl_io *)ctsio);
7957                                 return (1);
7958                         }
7959
7960                         if (type>8 || type==2 || type==4 || type==0) {
7961                                 mtx_unlock(&lun->lun_lock);
7962                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7963                                                       /*sks_valid*/ 1,
7964                                                       /*command*/ 1,
7965                                                       /*field*/ 2,
7966                                                       /*bit_valid*/ 1,
7967                                                       /*bit*/ 0);
7968                                 ctl_done((union ctl_io *)ctsio);
7969                                 return (1);
7970                         }
7971
7972                         /*
7973                          * Do the following:
7974                          * if sa_res_key != res_key remove all
7975                          * registrants w/sa_res_key and generate UA
7976                          * for these registrants(Registrations
7977                          * Preempted) if it wasn't an exclusive
7978                          * reservation generate UA(Reservations
7979                          * Preempted) for all other registered nexuses
7980                          * if the type has changed. Establish the new
7981                          * reservation and holder. If res_key and
7982                          * sa_res_key are the same do the above
7983                          * except don't unregister the res holder.
7984                          */
7985
7986                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7987                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7988                                         continue;
7989
7990                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7991                                         ctl_clr_prkey(lun, i);
7992                                         lun->pr_key_count--;
7993                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7994                                 } else if (type != lun->pr_res_type &&
7995                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7996                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7997                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7998                                 }
7999                         }
8000                         lun->pr_res_type = type;
8001                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8002                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8003                                 lun->pr_res_idx = residx;
8004                         else
8005                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8006                         lun->pr_generation++;
8007                         mtx_unlock(&lun->lun_lock);
8008
8009                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8010                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8011                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8012                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8013                         persis_io.pr.pr_info.res_type = type;
8014                         memcpy(persis_io.pr.pr_info.sa_res_key,
8015                                param->serv_act_res_key,
8016                                sizeof(param->serv_act_res_key));
8017                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8018                             sizeof(persis_io.pr), M_WAITOK);
8019                 } else {
8020                         /*
8021                          * sa_res_key is not the res holder just
8022                          * remove registrants
8023                          */
8024                         int found=0;
8025
8026                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8027                                 if (sa_res_key != ctl_get_prkey(lun, i))
8028                                         continue;
8029
8030                                 found = 1;
8031                                 ctl_clr_prkey(lun, i);
8032                                 lun->pr_key_count--;
8033                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8034                         }
8035
8036                         if (!found) {
8037                                 mtx_unlock(&lun->lun_lock);
8038                                 free(ctsio->kern_data_ptr, M_CTL);
8039                                 ctl_set_reservation_conflict(ctsio);
8040                                 ctl_done((union ctl_io *)ctsio);
8041                                 return (1);
8042                         }
8043                         lun->pr_generation++;
8044                         mtx_unlock(&lun->lun_lock);
8045
8046                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8047                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8048                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8049                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8050                         persis_io.pr.pr_info.res_type = type;
8051                         memcpy(persis_io.pr.pr_info.sa_res_key,
8052                                param->serv_act_res_key,
8053                                sizeof(param->serv_act_res_key));
8054                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8055                             sizeof(persis_io.pr), M_WAITOK);
8056                 }
8057         }
8058         return (0);
8059 }
8060
8061 static void
8062 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8063 {
8064         uint64_t sa_res_key;
8065         int i;
8066
8067         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8068
8069         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8070          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8071          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8072                 if (sa_res_key == 0) {
8073                         /*
8074                          * Unregister everybody else and build UA for
8075                          * them
8076                          */
8077                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8078                                 if (i == msg->pr.pr_info.residx ||
8079                                     ctl_get_prkey(lun, i) == 0)
8080                                         continue;
8081
8082                                 ctl_clr_prkey(lun, i);
8083                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8084                         }
8085
8086                         lun->pr_key_count = 1;
8087                         lun->pr_res_type = msg->pr.pr_info.res_type;
8088                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8089                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8090                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8091                 } else {
8092                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8093                                 if (sa_res_key == ctl_get_prkey(lun, i))
8094                                         continue;
8095
8096                                 ctl_clr_prkey(lun, i);
8097                                 lun->pr_key_count--;
8098                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8099                         }
8100                 }
8101         } else {
8102                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8103                         if (i == msg->pr.pr_info.residx ||
8104                             ctl_get_prkey(lun, i) == 0)
8105                                 continue;
8106
8107                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8108                                 ctl_clr_prkey(lun, i);
8109                                 lun->pr_key_count--;
8110                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8111                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8112                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8113                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8114                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8115                         }
8116                 }
8117                 lun->pr_res_type = msg->pr.pr_info.res_type;
8118                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8119                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8120                         lun->pr_res_idx = msg->pr.pr_info.residx;
8121                 else
8122                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8123         }
8124         lun->pr_generation++;
8125
8126 }
8127
8128
8129 int
8130 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8131 {
8132         int retval;
8133         u_int32_t param_len;
8134         struct scsi_per_res_out *cdb;
8135         struct ctl_lun *lun;
8136         struct scsi_per_res_out_parms* param;
8137         struct ctl_softc *softc;
8138         uint32_t residx;
8139         uint64_t res_key, sa_res_key, key;
8140         uint8_t type;
8141         union ctl_ha_msg persis_io;
8142         int    i;
8143
8144         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8145
8146         retval = CTL_RETVAL_COMPLETE;
8147
8148         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8149         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8150         softc = lun->ctl_softc;
8151
8152         /*
8153          * We only support whole-LUN scope.  The scope & type are ignored for
8154          * register, register and ignore existing key and clear.
8155          * We sometimes ignore scope and type on preempts too!!
8156          * Verify reservation type here as well.
8157          */
8158         type = cdb->scope_type & SPR_TYPE_MASK;
8159         if ((cdb->action == SPRO_RESERVE)
8160          || (cdb->action == SPRO_RELEASE)) {
8161                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8162                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8163                                               /*sks_valid*/ 1,
8164                                               /*command*/ 1,
8165                                               /*field*/ 2,
8166                                               /*bit_valid*/ 1,
8167                                               /*bit*/ 4);
8168                         ctl_done((union ctl_io *)ctsio);
8169                         return (CTL_RETVAL_COMPLETE);
8170                 }
8171
8172                 if (type>8 || type==2 || type==4 || type==0) {
8173                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8174                                               /*sks_valid*/ 1,
8175                                               /*command*/ 1,
8176                                               /*field*/ 2,
8177                                               /*bit_valid*/ 1,
8178                                               /*bit*/ 0);
8179                         ctl_done((union ctl_io *)ctsio);
8180                         return (CTL_RETVAL_COMPLETE);
8181                 }
8182         }
8183
8184         param_len = scsi_4btoul(cdb->length);
8185
8186         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8187                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8188                 ctsio->kern_data_len = param_len;
8189                 ctsio->kern_total_len = param_len;
8190                 ctsio->kern_data_resid = 0;
8191                 ctsio->kern_rel_offset = 0;
8192                 ctsio->kern_sg_entries = 0;
8193                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8194                 ctsio->be_move_done = ctl_config_move_done;
8195                 ctl_datamove((union ctl_io *)ctsio);
8196
8197                 return (CTL_RETVAL_COMPLETE);
8198         }
8199
8200         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8201
8202         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8203         res_key = scsi_8btou64(param->res_key.key);
8204         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8205
8206         /*
8207          * Validate the reservation key here except for SPRO_REG_IGNO
8208          * This must be done for all other service actions
8209          */
8210         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8211                 mtx_lock(&lun->lun_lock);
8212                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8213                         if (res_key != key) {
8214                                 /*
8215                                  * The current key passed in doesn't match
8216                                  * the one the initiator previously
8217                                  * registered.
8218                                  */
8219                                 mtx_unlock(&lun->lun_lock);
8220                                 free(ctsio->kern_data_ptr, M_CTL);
8221                                 ctl_set_reservation_conflict(ctsio);
8222                                 ctl_done((union ctl_io *)ctsio);
8223                                 return (CTL_RETVAL_COMPLETE);
8224                         }
8225                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8226                         /*
8227                          * We are not registered
8228                          */
8229                         mtx_unlock(&lun->lun_lock);
8230                         free(ctsio->kern_data_ptr, M_CTL);
8231                         ctl_set_reservation_conflict(ctsio);
8232                         ctl_done((union ctl_io *)ctsio);
8233                         return (CTL_RETVAL_COMPLETE);
8234                 } else if (res_key != 0) {
8235                         /*
8236                          * We are not registered and trying to register but
8237                          * the register key isn't zero.
8238                          */
8239                         mtx_unlock(&lun->lun_lock);
8240                         free(ctsio->kern_data_ptr, M_CTL);
8241                         ctl_set_reservation_conflict(ctsio);
8242                         ctl_done((union ctl_io *)ctsio);
8243                         return (CTL_RETVAL_COMPLETE);
8244                 }
8245                 mtx_unlock(&lun->lun_lock);
8246         }
8247
8248         switch (cdb->action & SPRO_ACTION_MASK) {
8249         case SPRO_REGISTER:
8250         case SPRO_REG_IGNO: {
8251
8252 #if 0
8253                 printf("Registration received\n");
8254 #endif
8255
8256                 /*
8257                  * We don't support any of these options, as we report in
8258                  * the read capabilities request (see
8259                  * ctl_persistent_reserve_in(), above).
8260                  */
8261                 if ((param->flags & SPR_SPEC_I_PT)
8262                  || (param->flags & SPR_ALL_TG_PT)
8263                  || (param->flags & SPR_APTPL)) {
8264                         int bit_ptr;
8265
8266                         if (param->flags & SPR_APTPL)
8267                                 bit_ptr = 0;
8268                         else if (param->flags & SPR_ALL_TG_PT)
8269                                 bit_ptr = 2;
8270                         else /* SPR_SPEC_I_PT */
8271                                 bit_ptr = 3;
8272
8273                         free(ctsio->kern_data_ptr, M_CTL);
8274                         ctl_set_invalid_field(ctsio,
8275                                               /*sks_valid*/ 1,
8276                                               /*command*/ 0,
8277                                               /*field*/ 20,
8278                                               /*bit_valid*/ 1,
8279                                               /*bit*/ bit_ptr);
8280                         ctl_done((union ctl_io *)ctsio);
8281                         return (CTL_RETVAL_COMPLETE);
8282                 }
8283
8284                 mtx_lock(&lun->lun_lock);
8285
8286                 /*
8287                  * The initiator wants to clear the
8288                  * key/unregister.
8289                  */
8290                 if (sa_res_key == 0) {
8291                         if ((res_key == 0
8292                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8293                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8294                           && ctl_get_prkey(lun, residx) == 0)) {
8295                                 mtx_unlock(&lun->lun_lock);
8296                                 goto done;
8297                         }
8298
8299                         ctl_clr_prkey(lun, residx);
8300                         lun->pr_key_count--;
8301
8302                         if (residx == lun->pr_res_idx) {
8303                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8304                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8305
8306                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8307                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8308                                     lun->pr_key_count) {
8309                                         /*
8310                                          * If the reservation is a registrants
8311                                          * only type we need to generate a UA
8312                                          * for other registered inits.  The
8313                                          * sense code should be RESERVATIONS
8314                                          * RELEASED
8315                                          */
8316
8317                                         for (i = softc->init_min; i < softc->init_max; i++){
8318                                                 if (ctl_get_prkey(lun, i) == 0)
8319                                                         continue;
8320                                                 ctl_est_ua(lun, i,
8321                                                     CTL_UA_RES_RELEASE);
8322                                         }
8323                                 }
8324                                 lun->pr_res_type = 0;
8325                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8326                                 if (lun->pr_key_count==0) {
8327                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8328                                         lun->pr_res_type = 0;
8329                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8330                                 }
8331                         }
8332                         lun->pr_generation++;
8333                         mtx_unlock(&lun->lun_lock);
8334
8335                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8336                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8337                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8338                         persis_io.pr.pr_info.residx = residx;
8339                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8340                             sizeof(persis_io.pr), M_WAITOK);
8341                 } else /* sa_res_key != 0 */ {
8342
8343                         /*
8344                          * If we aren't registered currently then increment
8345                          * the key count and set the registered flag.
8346                          */
8347                         ctl_alloc_prkey(lun, residx);
8348                         if (ctl_get_prkey(lun, residx) == 0)
8349                                 lun->pr_key_count++;
8350                         ctl_set_prkey(lun, residx, sa_res_key);
8351                         lun->pr_generation++;
8352                         mtx_unlock(&lun->lun_lock);
8353
8354                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8355                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8356                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8357                         persis_io.pr.pr_info.residx = residx;
8358                         memcpy(persis_io.pr.pr_info.sa_res_key,
8359                                param->serv_act_res_key,
8360                                sizeof(param->serv_act_res_key));
8361                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8362                             sizeof(persis_io.pr), M_WAITOK);
8363                 }
8364
8365                 break;
8366         }
8367         case SPRO_RESERVE:
8368 #if 0
8369                 printf("Reserve executed type %d\n", type);
8370 #endif
8371                 mtx_lock(&lun->lun_lock);
8372                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8373                         /*
8374                          * if this isn't the reservation holder and it's
8375                          * not a "all registrants" type or if the type is
8376                          * different then we have a conflict
8377                          */
8378                         if ((lun->pr_res_idx != residx
8379                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8380                          || lun->pr_res_type != type) {
8381                                 mtx_unlock(&lun->lun_lock);
8382                                 free(ctsio->kern_data_ptr, M_CTL);
8383                                 ctl_set_reservation_conflict(ctsio);
8384                                 ctl_done((union ctl_io *)ctsio);
8385                                 return (CTL_RETVAL_COMPLETE);
8386                         }
8387                         mtx_unlock(&lun->lun_lock);
8388                 } else /* create a reservation */ {
8389                         /*
8390                          * If it's not an "all registrants" type record
8391                          * reservation holder
8392                          */
8393                         if (type != SPR_TYPE_WR_EX_AR
8394                          && type != SPR_TYPE_EX_AC_AR)
8395                                 lun->pr_res_idx = residx; /* Res holder */
8396                         else
8397                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8398
8399                         lun->flags |= CTL_LUN_PR_RESERVED;
8400                         lun->pr_res_type = type;
8401
8402                         mtx_unlock(&lun->lun_lock);
8403
8404                         /* send msg to other side */
8405                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8406                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8407                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8408                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8409                         persis_io.pr.pr_info.res_type = type;
8410                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8411                             sizeof(persis_io.pr), M_WAITOK);
8412                 }
8413                 break;
8414
8415         case SPRO_RELEASE:
8416                 mtx_lock(&lun->lun_lock);
8417                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8418                         /* No reservation exists return good status */
8419                         mtx_unlock(&lun->lun_lock);
8420                         goto done;
8421                 }
8422                 /*
8423                  * Is this nexus a reservation holder?
8424                  */
8425                 if (lun->pr_res_idx != residx
8426                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8427                         /*
8428                          * not a res holder return good status but
8429                          * do nothing
8430                          */
8431                         mtx_unlock(&lun->lun_lock);
8432                         goto done;
8433                 }
8434
8435                 if (lun->pr_res_type != type) {
8436                         mtx_unlock(&lun->lun_lock);
8437                         free(ctsio->kern_data_ptr, M_CTL);
8438                         ctl_set_illegal_pr_release(ctsio);
8439                         ctl_done((union ctl_io *)ctsio);
8440                         return (CTL_RETVAL_COMPLETE);
8441                 }
8442
8443                 /* okay to release */
8444                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8445                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8446                 lun->pr_res_type = 0;
8447
8448                 /*
8449                  * If this isn't an exclusive access reservation and NUAR
8450                  * is not set, generate UA for all other registrants.
8451                  */
8452                 if (type != SPR_TYPE_EX_AC && type != SPR_TYPE_WR_EX &&
8453                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8454                         for (i = softc->init_min; i < softc->init_max; i++) {
8455                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8456                                         continue;
8457                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8458                         }
8459                 }
8460                 mtx_unlock(&lun->lun_lock);
8461
8462                 /* Send msg to other side */
8463                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8464                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8465                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8466                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8467                      sizeof(persis_io.pr), M_WAITOK);
8468                 break;
8469
8470         case SPRO_CLEAR:
8471                 /* send msg to other side */
8472
8473                 mtx_lock(&lun->lun_lock);
8474                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8475                 lun->pr_res_type = 0;
8476                 lun->pr_key_count = 0;
8477                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8478
8479                 ctl_clr_prkey(lun, residx);
8480                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8481                         if (ctl_get_prkey(lun, i) != 0) {
8482                                 ctl_clr_prkey(lun, i);
8483                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8484                         }
8485                 lun->pr_generation++;
8486                 mtx_unlock(&lun->lun_lock);
8487
8488                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8489                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8490                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8491                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8492                      sizeof(persis_io.pr), M_WAITOK);
8493                 break;
8494
8495         case SPRO_PREEMPT:
8496         case SPRO_PRE_ABO: {
8497                 int nretval;
8498
8499                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8500                                           residx, ctsio, cdb, param);
8501                 if (nretval != 0)
8502                         return (CTL_RETVAL_COMPLETE);
8503                 break;
8504         }
8505         default:
8506                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8507         }
8508
8509 done:
8510         free(ctsio->kern_data_ptr, M_CTL);
8511         ctl_set_success(ctsio);
8512         ctl_done((union ctl_io *)ctsio);
8513
8514         return (retval);
8515 }
8516
8517 /*
8518  * This routine is for handling a message from the other SC pertaining to
8519  * persistent reserve out. All the error checking will have been done
8520  * so only perorming the action need be done here to keep the two
8521  * in sync.
8522  */
8523 static void
8524 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8525 {
8526         struct ctl_softc *softc = control_softc;
8527         struct ctl_lun *lun;
8528         int i;
8529         uint32_t residx, targ_lun;
8530
8531         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8532         mtx_lock(&softc->ctl_lock);
8533         if ((targ_lun >= CTL_MAX_LUNS) ||
8534             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8535                 mtx_unlock(&softc->ctl_lock);
8536                 return;
8537         }
8538         mtx_lock(&lun->lun_lock);
8539         mtx_unlock(&softc->ctl_lock);
8540         if (lun->flags & CTL_LUN_DISABLED) {
8541                 mtx_unlock(&lun->lun_lock);
8542                 return;
8543         }
8544         residx = ctl_get_initindex(&msg->hdr.nexus);
8545         switch(msg->pr.pr_info.action) {
8546         case CTL_PR_REG_KEY:
8547                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8548                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8549                         lun->pr_key_count++;
8550                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8551                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8552                 lun->pr_generation++;
8553                 break;
8554
8555         case CTL_PR_UNREG_KEY:
8556                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8557                 lun->pr_key_count--;
8558
8559                 /* XXX Need to see if the reservation has been released */
8560                 /* if so do we need to generate UA? */
8561                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8562                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8563                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8564
8565                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8566                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8567                             lun->pr_key_count) {
8568                                 /*
8569                                  * If the reservation is a registrants
8570                                  * only type we need to generate a UA
8571                                  * for other registered inits.  The
8572                                  * sense code should be RESERVATIONS
8573                                  * RELEASED
8574                                  */
8575
8576                                 for (i = softc->init_min; i < softc->init_max; i++) {
8577                                         if (ctl_get_prkey(lun, i) == 0)
8578                                                 continue;
8579
8580                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8581                                 }
8582                         }
8583                         lun->pr_res_type = 0;
8584                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8585                         if (lun->pr_key_count==0) {
8586                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8587                                 lun->pr_res_type = 0;
8588                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8589                         }
8590                 }
8591                 lun->pr_generation++;
8592                 break;
8593
8594         case CTL_PR_RESERVE:
8595                 lun->flags |= CTL_LUN_PR_RESERVED;
8596                 lun->pr_res_type = msg->pr.pr_info.res_type;
8597                 lun->pr_res_idx = msg->pr.pr_info.residx;
8598
8599                 break;
8600
8601         case CTL_PR_RELEASE:
8602                 /*
8603                  * If this isn't an exclusive access reservation and NUAR
8604                  * is not set, generate UA for all other registrants.
8605                  */
8606                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8607                     lun->pr_res_type != SPR_TYPE_WR_EX &&
8608                     (lun->MODE_CTRL.queue_flags & SCP_NUAR) == 0) {
8609                         for (i = softc->init_min; i < softc->init_max; i++)
8610                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8611                                         continue;
8612                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8613                 }
8614
8615                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8616                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8617                 lun->pr_res_type = 0;
8618                 break;
8619
8620         case CTL_PR_PREEMPT:
8621                 ctl_pro_preempt_other(lun, msg);
8622                 break;
8623         case CTL_PR_CLEAR:
8624                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8625                 lun->pr_res_type = 0;
8626                 lun->pr_key_count = 0;
8627                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8628
8629                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8630                         if (ctl_get_prkey(lun, i) == 0)
8631                                 continue;
8632                         ctl_clr_prkey(lun, i);
8633                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8634                 }
8635                 lun->pr_generation++;
8636                 break;
8637         }
8638
8639         mtx_unlock(&lun->lun_lock);
8640 }
8641
8642 int
8643 ctl_read_write(struct ctl_scsiio *ctsio)
8644 {
8645         struct ctl_lun *lun;
8646         struct ctl_lba_len_flags *lbalen;
8647         uint64_t lba;
8648         uint32_t num_blocks;
8649         int flags, retval;
8650         int isread;
8651
8652         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8653
8654         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8655
8656         flags = 0;
8657         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8658               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8659         switch (ctsio->cdb[0]) {
8660         case READ_6:
8661         case WRITE_6: {
8662                 struct scsi_rw_6 *cdb;
8663
8664                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8665
8666                 lba = scsi_3btoul(cdb->addr);
8667                 /* only 5 bits are valid in the most significant address byte */
8668                 lba &= 0x1fffff;
8669                 num_blocks = cdb->length;
8670                 /*
8671                  * This is correct according to SBC-2.
8672                  */
8673                 if (num_blocks == 0)
8674                         num_blocks = 256;
8675                 break;
8676         }
8677         case READ_10:
8678         case WRITE_10: {
8679                 struct scsi_rw_10 *cdb;
8680
8681                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8682                 if (cdb->byte2 & SRW10_FUA)
8683                         flags |= CTL_LLF_FUA;
8684                 if (cdb->byte2 & SRW10_DPO)
8685                         flags |= CTL_LLF_DPO;
8686                 lba = scsi_4btoul(cdb->addr);
8687                 num_blocks = scsi_2btoul(cdb->length);
8688                 break;
8689         }
8690         case WRITE_VERIFY_10: {
8691                 struct scsi_write_verify_10 *cdb;
8692
8693                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8694                 flags |= CTL_LLF_FUA;
8695                 if (cdb->byte2 & SWV_DPO)
8696                         flags |= CTL_LLF_DPO;
8697                 lba = scsi_4btoul(cdb->addr);
8698                 num_blocks = scsi_2btoul(cdb->length);
8699                 break;
8700         }
8701         case READ_12:
8702         case WRITE_12: {
8703                 struct scsi_rw_12 *cdb;
8704
8705                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8706                 if (cdb->byte2 & SRW12_FUA)
8707                         flags |= CTL_LLF_FUA;
8708                 if (cdb->byte2 & SRW12_DPO)
8709                         flags |= CTL_LLF_DPO;
8710                 lba = scsi_4btoul(cdb->addr);
8711                 num_blocks = scsi_4btoul(cdb->length);
8712                 break;
8713         }
8714         case WRITE_VERIFY_12: {
8715                 struct scsi_write_verify_12 *cdb;
8716
8717                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8718                 flags |= CTL_LLF_FUA;
8719                 if (cdb->byte2 & SWV_DPO)
8720                         flags |= CTL_LLF_DPO;
8721                 lba = scsi_4btoul(cdb->addr);
8722                 num_blocks = scsi_4btoul(cdb->length);
8723                 break;
8724         }
8725         case READ_16:
8726         case WRITE_16: {
8727                 struct scsi_rw_16 *cdb;
8728
8729                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8730                 if (cdb->byte2 & SRW12_FUA)
8731                         flags |= CTL_LLF_FUA;
8732                 if (cdb->byte2 & SRW12_DPO)
8733                         flags |= CTL_LLF_DPO;
8734                 lba = scsi_8btou64(cdb->addr);
8735                 num_blocks = scsi_4btoul(cdb->length);
8736                 break;
8737         }
8738         case WRITE_ATOMIC_16: {
8739                 struct scsi_write_atomic_16 *cdb;
8740
8741                 if (lun->be_lun->atomicblock == 0) {
8742                         ctl_set_invalid_opcode(ctsio);
8743                         ctl_done((union ctl_io *)ctsio);
8744                         return (CTL_RETVAL_COMPLETE);
8745                 }
8746
8747                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8748                 if (cdb->byte2 & SRW12_FUA)
8749                         flags |= CTL_LLF_FUA;
8750                 if (cdb->byte2 & SRW12_DPO)
8751                         flags |= CTL_LLF_DPO;
8752                 lba = scsi_8btou64(cdb->addr);
8753                 num_blocks = scsi_2btoul(cdb->length);
8754                 if (num_blocks > lun->be_lun->atomicblock) {
8755                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8756                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8757                             /*bit*/ 0);
8758                         ctl_done((union ctl_io *)ctsio);
8759                         return (CTL_RETVAL_COMPLETE);
8760                 }
8761                 break;
8762         }
8763         case WRITE_VERIFY_16: {
8764                 struct scsi_write_verify_16 *cdb;
8765
8766                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8767                 flags |= CTL_LLF_FUA;
8768                 if (cdb->byte2 & SWV_DPO)
8769                         flags |= CTL_LLF_DPO;
8770                 lba = scsi_8btou64(cdb->addr);
8771                 num_blocks = scsi_4btoul(cdb->length);
8772                 break;
8773         }
8774         default:
8775                 /*
8776                  * We got a command we don't support.  This shouldn't
8777                  * happen, commands should be filtered out above us.
8778                  */
8779                 ctl_set_invalid_opcode(ctsio);
8780                 ctl_done((union ctl_io *)ctsio);
8781
8782                 return (CTL_RETVAL_COMPLETE);
8783                 break; /* NOTREACHED */
8784         }
8785
8786         /*
8787          * The first check is to make sure we're in bounds, the second
8788          * check is to catch wrap-around problems.  If the lba + num blocks
8789          * is less than the lba, then we've wrapped around and the block
8790          * range is invalid anyway.
8791          */
8792         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8793          || ((lba + num_blocks) < lba)) {
8794                 ctl_set_lba_out_of_range(ctsio,
8795                     MAX(lba, lun->be_lun->maxlba + 1));
8796                 ctl_done((union ctl_io *)ctsio);
8797                 return (CTL_RETVAL_COMPLETE);
8798         }
8799
8800         /*
8801          * According to SBC-3, a transfer length of 0 is not an error.
8802          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8803          * translates to 256 blocks for those commands.
8804          */
8805         if (num_blocks == 0) {
8806                 ctl_set_success(ctsio);
8807                 ctl_done((union ctl_io *)ctsio);
8808                 return (CTL_RETVAL_COMPLETE);
8809         }
8810
8811         /* Set FUA and/or DPO if caches are disabled. */
8812         if (isread) {
8813                 if ((lun->MODE_CACHING.flags1 & SCP_RCD) != 0)
8814                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8815         } else {
8816                 if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8817                         flags |= CTL_LLF_FUA;
8818         }
8819
8820         lbalen = (struct ctl_lba_len_flags *)
8821             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8822         lbalen->lba = lba;
8823         lbalen->len = num_blocks;
8824         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8825
8826         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8827         ctsio->kern_rel_offset = 0;
8828
8829         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8830
8831         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8832         return (retval);
8833 }
8834
8835 static int
8836 ctl_cnw_cont(union ctl_io *io)
8837 {
8838         struct ctl_scsiio *ctsio;
8839         struct ctl_lun *lun;
8840         struct ctl_lba_len_flags *lbalen;
8841         int retval;
8842
8843         ctsio = &io->scsiio;
8844         ctsio->io_hdr.status = CTL_STATUS_NONE;
8845         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8846         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8847         lbalen = (struct ctl_lba_len_flags *)
8848             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8849         lbalen->flags &= ~CTL_LLF_COMPARE;
8850         lbalen->flags |= CTL_LLF_WRITE;
8851
8852         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8853         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8854         return (retval);
8855 }
8856
8857 int
8858 ctl_cnw(struct ctl_scsiio *ctsio)
8859 {
8860         struct ctl_lun *lun;
8861         struct ctl_lba_len_flags *lbalen;
8862         uint64_t lba;
8863         uint32_t num_blocks;
8864         int flags, retval;
8865
8866         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8867
8868         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8869
8870         flags = 0;
8871         switch (ctsio->cdb[0]) {
8872         case COMPARE_AND_WRITE: {
8873                 struct scsi_compare_and_write *cdb;
8874
8875                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8876                 if (cdb->byte2 & SRW10_FUA)
8877                         flags |= CTL_LLF_FUA;
8878                 if (cdb->byte2 & SRW10_DPO)
8879                         flags |= CTL_LLF_DPO;
8880                 lba = scsi_8btou64(cdb->addr);
8881                 num_blocks = cdb->length;
8882                 break;
8883         }
8884         default:
8885                 /*
8886                  * We got a command we don't support.  This shouldn't
8887                  * happen, commands should be filtered out above us.
8888                  */
8889                 ctl_set_invalid_opcode(ctsio);
8890                 ctl_done((union ctl_io *)ctsio);
8891
8892                 return (CTL_RETVAL_COMPLETE);
8893                 break; /* NOTREACHED */
8894         }
8895
8896         /*
8897          * The first check is to make sure we're in bounds, the second
8898          * check is to catch wrap-around problems.  If the lba + num blocks
8899          * is less than the lba, then we've wrapped around and the block
8900          * range is invalid anyway.
8901          */
8902         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8903          || ((lba + num_blocks) < lba)) {
8904                 ctl_set_lba_out_of_range(ctsio,
8905                     MAX(lba, lun->be_lun->maxlba + 1));
8906                 ctl_done((union ctl_io *)ctsio);
8907                 return (CTL_RETVAL_COMPLETE);
8908         }
8909
8910         /*
8911          * According to SBC-3, a transfer length of 0 is not an error.
8912          */
8913         if (num_blocks == 0) {
8914                 ctl_set_success(ctsio);
8915                 ctl_done((union ctl_io *)ctsio);
8916                 return (CTL_RETVAL_COMPLETE);
8917         }
8918
8919         /* Set FUA if write cache is disabled. */
8920         if ((lun->MODE_CACHING.flags1 & SCP_WCE) == 0)
8921                 flags |= CTL_LLF_FUA;
8922
8923         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8924         ctsio->kern_rel_offset = 0;
8925
8926         /*
8927          * Set the IO_CONT flag, so that if this I/O gets passed to
8928          * ctl_data_submit_done(), it'll get passed back to
8929          * ctl_ctl_cnw_cont() for further processing.
8930          */
8931         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8932         ctsio->io_cont = ctl_cnw_cont;
8933
8934         lbalen = (struct ctl_lba_len_flags *)
8935             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8936         lbalen->lba = lba;
8937         lbalen->len = num_blocks;
8938         lbalen->flags = CTL_LLF_COMPARE | flags;
8939
8940         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8941         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8942         return (retval);
8943 }
8944
8945 int
8946 ctl_verify(struct ctl_scsiio *ctsio)
8947 {
8948         struct ctl_lun *lun;
8949         struct ctl_lba_len_flags *lbalen;
8950         uint64_t lba;
8951         uint32_t num_blocks;
8952         int bytchk, flags;
8953         int retval;
8954
8955         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8956
8957         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8958
8959         bytchk = 0;
8960         flags = CTL_LLF_FUA;
8961         switch (ctsio->cdb[0]) {
8962         case VERIFY_10: {
8963                 struct scsi_verify_10 *cdb;
8964
8965                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8966                 if (cdb->byte2 & SVFY_BYTCHK)
8967                         bytchk = 1;
8968                 if (cdb->byte2 & SVFY_DPO)
8969                         flags |= CTL_LLF_DPO;
8970                 lba = scsi_4btoul(cdb->addr);
8971                 num_blocks = scsi_2btoul(cdb->length);
8972                 break;
8973         }
8974         case VERIFY_12: {
8975                 struct scsi_verify_12 *cdb;
8976
8977                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8978                 if (cdb->byte2 & SVFY_BYTCHK)
8979                         bytchk = 1;
8980                 if (cdb->byte2 & SVFY_DPO)
8981                         flags |= CTL_LLF_DPO;
8982                 lba = scsi_4btoul(cdb->addr);
8983                 num_blocks = scsi_4btoul(cdb->length);
8984                 break;
8985         }
8986         case VERIFY_16: {
8987                 struct scsi_rw_16 *cdb;
8988
8989                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8990                 if (cdb->byte2 & SVFY_BYTCHK)
8991                         bytchk = 1;
8992                 if (cdb->byte2 & SVFY_DPO)
8993                         flags |= CTL_LLF_DPO;
8994                 lba = scsi_8btou64(cdb->addr);
8995                 num_blocks = scsi_4btoul(cdb->length);
8996                 break;
8997         }
8998         default:
8999                 /*
9000                  * We got a command we don't support.  This shouldn't
9001                  * happen, commands should be filtered out above us.
9002                  */
9003                 ctl_set_invalid_opcode(ctsio);
9004                 ctl_done((union ctl_io *)ctsio);
9005                 return (CTL_RETVAL_COMPLETE);
9006         }
9007
9008         /*
9009          * The first check is to make sure we're in bounds, the second
9010          * check is to catch wrap-around problems.  If the lba + num blocks
9011          * is less than the lba, then we've wrapped around and the block
9012          * range is invalid anyway.
9013          */
9014         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9015          || ((lba + num_blocks) < lba)) {
9016                 ctl_set_lba_out_of_range(ctsio,
9017                     MAX(lba, lun->be_lun->maxlba + 1));
9018                 ctl_done((union ctl_io *)ctsio);
9019                 return (CTL_RETVAL_COMPLETE);
9020         }
9021
9022         /*
9023          * According to SBC-3, a transfer length of 0 is not an error.
9024          */
9025         if (num_blocks == 0) {
9026                 ctl_set_success(ctsio);
9027                 ctl_done((union ctl_io *)ctsio);
9028                 return (CTL_RETVAL_COMPLETE);
9029         }
9030
9031         lbalen = (struct ctl_lba_len_flags *)
9032             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9033         lbalen->lba = lba;
9034         lbalen->len = num_blocks;
9035         if (bytchk) {
9036                 lbalen->flags = CTL_LLF_COMPARE | flags;
9037                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9038         } else {
9039                 lbalen->flags = CTL_LLF_VERIFY | flags;
9040                 ctsio->kern_total_len = 0;
9041         }
9042         ctsio->kern_rel_offset = 0;
9043
9044         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9045         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9046         return (retval);
9047 }
9048
9049 int
9050 ctl_report_luns(struct ctl_scsiio *ctsio)
9051 {
9052         struct ctl_softc *softc;
9053         struct scsi_report_luns *cdb;
9054         struct scsi_report_luns_data *lun_data;
9055         struct ctl_lun *lun, *request_lun;
9056         struct ctl_port *port;
9057         int num_luns, retval;
9058         uint32_t alloc_len, lun_datalen;
9059         int num_filled;
9060         uint32_t initidx, targ_lun_id, lun_id;
9061
9062         retval = CTL_RETVAL_COMPLETE;
9063         cdb = (struct scsi_report_luns *)ctsio->cdb;
9064         port = ctl_io_port(&ctsio->io_hdr);
9065         softc = port->ctl_softc;
9066
9067         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9068
9069         mtx_lock(&softc->ctl_lock);
9070         num_luns = 0;
9071         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9072                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9073                         num_luns++;
9074         }
9075         mtx_unlock(&softc->ctl_lock);
9076
9077         switch (cdb->select_report) {
9078         case RPL_REPORT_DEFAULT:
9079         case RPL_REPORT_ALL:
9080         case RPL_REPORT_NONSUBSID:
9081                 break;
9082         case RPL_REPORT_WELLKNOWN:
9083         case RPL_REPORT_ADMIN:
9084         case RPL_REPORT_CONGLOM:
9085                 num_luns = 0;
9086                 break;
9087         default:
9088                 ctl_set_invalid_field(ctsio,
9089                                       /*sks_valid*/ 1,
9090                                       /*command*/ 1,
9091                                       /*field*/ 2,
9092                                       /*bit_valid*/ 0,
9093                                       /*bit*/ 0);
9094                 ctl_done((union ctl_io *)ctsio);
9095                 return (retval);
9096                 break; /* NOTREACHED */
9097         }
9098
9099         alloc_len = scsi_4btoul(cdb->length);
9100         /*
9101          * The initiator has to allocate at least 16 bytes for this request,
9102          * so he can at least get the header and the first LUN.  Otherwise
9103          * we reject the request (per SPC-3 rev 14, section 6.21).
9104          */
9105         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9106             sizeof(struct scsi_report_luns_lundata))) {
9107                 ctl_set_invalid_field(ctsio,
9108                                       /*sks_valid*/ 1,
9109                                       /*command*/ 1,
9110                                       /*field*/ 6,
9111                                       /*bit_valid*/ 0,
9112                                       /*bit*/ 0);
9113                 ctl_done((union ctl_io *)ctsio);
9114                 return (retval);
9115         }
9116
9117         request_lun = (struct ctl_lun *)
9118                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9119
9120         lun_datalen = sizeof(*lun_data) +
9121                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9122
9123         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9124         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9125         ctsio->kern_sg_entries = 0;
9126
9127         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9128
9129         mtx_lock(&softc->ctl_lock);
9130         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9131                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9132                 if (lun_id >= CTL_MAX_LUNS)
9133                         continue;
9134                 lun = softc->ctl_luns[lun_id];
9135                 if (lun == NULL)
9136                         continue;
9137
9138                 be64enc(lun_data->luns[num_filled++].lundata,
9139                     ctl_encode_lun(targ_lun_id));
9140
9141                 /*
9142                  * According to SPC-3, rev 14 section 6.21:
9143                  *
9144                  * "The execution of a REPORT LUNS command to any valid and
9145                  * installed logical unit shall clear the REPORTED LUNS DATA
9146                  * HAS CHANGED unit attention condition for all logical
9147                  * units of that target with respect to the requesting
9148                  * initiator. A valid and installed logical unit is one
9149                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9150                  * INQUIRY data (see 6.4.2)."
9151                  *
9152                  * If request_lun is NULL, the LUN this report luns command
9153                  * was issued to is either disabled or doesn't exist. In that
9154                  * case, we shouldn't clear any pending lun change unit
9155                  * attention.
9156                  */
9157                 if (request_lun != NULL) {
9158                         mtx_lock(&lun->lun_lock);
9159                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9160                         mtx_unlock(&lun->lun_lock);
9161                 }
9162         }
9163         mtx_unlock(&softc->ctl_lock);
9164
9165         /*
9166          * It's quite possible that we've returned fewer LUNs than we allocated
9167          * space for.  Trim it.
9168          */
9169         lun_datalen = sizeof(*lun_data) +
9170                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9171
9172         if (lun_datalen < alloc_len) {
9173                 ctsio->residual = alloc_len - lun_datalen;
9174                 ctsio->kern_data_len = lun_datalen;
9175                 ctsio->kern_total_len = lun_datalen;
9176         } else {
9177                 ctsio->residual = 0;
9178                 ctsio->kern_data_len = alloc_len;
9179                 ctsio->kern_total_len = alloc_len;
9180         }
9181         ctsio->kern_data_resid = 0;
9182         ctsio->kern_rel_offset = 0;
9183         ctsio->kern_sg_entries = 0;
9184
9185         /*
9186          * We set this to the actual data length, regardless of how much
9187          * space we actually have to return results.  If the user looks at
9188          * this value, he'll know whether or not he allocated enough space
9189          * and reissue the command if necessary.  We don't support well
9190          * known logical units, so if the user asks for that, return none.
9191          */
9192         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9193
9194         /*
9195          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9196          * this request.
9197          */
9198         ctl_set_success(ctsio);
9199         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9200         ctsio->be_move_done = ctl_config_move_done;
9201         ctl_datamove((union ctl_io *)ctsio);
9202         return (retval);
9203 }
9204
9205 int
9206 ctl_request_sense(struct ctl_scsiio *ctsio)
9207 {
9208         struct scsi_request_sense *cdb;
9209         struct scsi_sense_data *sense_ptr;
9210         struct ctl_softc *softc;
9211         struct ctl_lun *lun;
9212         uint32_t initidx;
9213         int have_error;
9214         u_int sense_len = SSD_FULL_SIZE;
9215         scsi_sense_data_type sense_format;
9216         ctl_ua_type ua_type;
9217         uint8_t asc = 0, ascq = 0;
9218
9219         cdb = (struct scsi_request_sense *)ctsio->cdb;
9220
9221         softc = control_softc;
9222         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9223
9224         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9225
9226         /*
9227          * Determine which sense format the user wants.
9228          */
9229         if (cdb->byte2 & SRS_DESC)
9230                 sense_format = SSD_TYPE_DESC;
9231         else
9232                 sense_format = SSD_TYPE_FIXED;
9233
9234         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9235         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9236         ctsio->kern_sg_entries = 0;
9237
9238         /*
9239          * struct scsi_sense_data, which is currently set to 256 bytes, is
9240          * larger than the largest allowed value for the length field in the
9241          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9242          */
9243         ctsio->residual = 0;
9244         ctsio->kern_data_len = cdb->length;
9245         ctsio->kern_total_len = cdb->length;
9246
9247         ctsio->kern_data_resid = 0;
9248         ctsio->kern_rel_offset = 0;
9249         ctsio->kern_sg_entries = 0;
9250
9251         /*
9252          * If we don't have a LUN, we don't have any pending sense.
9253          */
9254         if (lun == NULL ||
9255             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9256              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9257                 /* "Logical unit not supported" */
9258                 ctl_set_sense_data(sense_ptr, &sense_len, NULL, sense_format,
9259                     /*current_error*/ 1,
9260                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9261                     /*asc*/ 0x25,
9262                     /*ascq*/ 0x00,
9263                     SSD_ELEM_NONE);
9264                 goto send;
9265         }
9266
9267         have_error = 0;
9268         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9269         /*
9270          * Check for pending sense, and then for pending unit attentions.
9271          * Pending sense gets returned first, then pending unit attentions.
9272          */
9273         mtx_lock(&lun->lun_lock);
9274 #ifdef CTL_WITH_CA
9275         if (ctl_is_set(lun->have_ca, initidx)) {
9276                 scsi_sense_data_type stored_format;
9277
9278                 /*
9279                  * Check to see which sense format was used for the stored
9280                  * sense data.
9281                  */
9282                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9283
9284                 /*
9285                  * If the user requested a different sense format than the
9286                  * one we stored, then we need to convert it to the other
9287                  * format.  If we're going from descriptor to fixed format
9288                  * sense data, we may lose things in translation, depending
9289                  * on what options were used.
9290                  *
9291                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9292                  * for some reason we'll just copy it out as-is.
9293                  */
9294                 if ((stored_format == SSD_TYPE_FIXED)
9295                  && (sense_format == SSD_TYPE_DESC))
9296                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9297                             &lun->pending_sense[initidx],
9298                             (struct scsi_sense_data_desc *)sense_ptr);
9299                 else if ((stored_format == SSD_TYPE_DESC)
9300                       && (sense_format == SSD_TYPE_FIXED))
9301                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9302                             &lun->pending_sense[initidx],
9303                             (struct scsi_sense_data_fixed *)sense_ptr);
9304                 else
9305                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9306                                MIN(sizeof(*sense_ptr),
9307                                sizeof(lun->pending_sense[initidx])));
9308
9309                 ctl_clear_mask(lun->have_ca, initidx);
9310                 have_error = 1;
9311         } else
9312 #endif
9313         if (have_error == 0) {
9314                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, &sense_len,
9315                     sense_format);
9316                 if (ua_type != CTL_UA_NONE)
9317                         have_error = 1;
9318         }
9319         if (have_error == 0) {
9320                 /*
9321                  * Report informational exception if have one and allowed.
9322                  */
9323                 if (lun->MODE_IE.mrie != SIEP_MRIE_NO) {
9324                         asc = lun->ie_asc;
9325                         ascq = lun->ie_ascq;
9326                 }
9327                 ctl_set_sense_data(sense_ptr, &sense_len, lun, sense_format,
9328                     /*current_error*/ 1,
9329                     /*sense_key*/ SSD_KEY_NO_SENSE,
9330                     /*asc*/ asc,
9331                     /*ascq*/ ascq,
9332                     SSD_ELEM_NONE);
9333         }
9334         mtx_unlock(&lun->lun_lock);
9335
9336 send:
9337         /*
9338          * We report the SCSI status as OK, since the status of the command
9339          * itself is OK.  We're reporting sense as parameter data.
9340          */
9341         ctl_set_success(ctsio);
9342         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9343         ctsio->be_move_done = ctl_config_move_done;
9344         ctl_datamove((union ctl_io *)ctsio);
9345         return (CTL_RETVAL_COMPLETE);
9346 }
9347
9348 int
9349 ctl_tur(struct ctl_scsiio *ctsio)
9350 {
9351
9352         CTL_DEBUG_PRINT(("ctl_tur\n"));
9353
9354         ctl_set_success(ctsio);
9355         ctl_done((union ctl_io *)ctsio);
9356
9357         return (CTL_RETVAL_COMPLETE);
9358 }
9359
9360 /*
9361  * SCSI VPD page 0x00, the Supported VPD Pages page.
9362  */
9363 static int
9364 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9365 {
9366         struct scsi_vpd_supported_pages *pages;
9367         int sup_page_size;
9368         struct ctl_lun *lun;
9369         int p;
9370
9371         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9372
9373         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9374             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9375         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9376         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9377         ctsio->kern_sg_entries = 0;
9378
9379         if (sup_page_size < alloc_len) {
9380                 ctsio->residual = alloc_len - sup_page_size;
9381                 ctsio->kern_data_len = sup_page_size;
9382                 ctsio->kern_total_len = sup_page_size;
9383         } else {
9384                 ctsio->residual = 0;
9385                 ctsio->kern_data_len = alloc_len;
9386                 ctsio->kern_total_len = alloc_len;
9387         }
9388         ctsio->kern_data_resid = 0;
9389         ctsio->kern_rel_offset = 0;
9390         ctsio->kern_sg_entries = 0;
9391
9392         /*
9393          * The control device is always connected.  The disk device, on the
9394          * other hand, may not be online all the time.  Need to change this
9395          * to figure out whether the disk device is actually online or not.
9396          */
9397         if (lun != NULL)
9398                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9399                                 lun->be_lun->lun_type;
9400         else
9401                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9402
9403         p = 0;
9404         /* Supported VPD pages */
9405         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9406         /* Serial Number */
9407         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9408         /* Device Identification */
9409         pages->page_list[p++] = SVPD_DEVICE_ID;
9410         /* Extended INQUIRY Data */
9411         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9412         /* Mode Page Policy */
9413         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9414         /* SCSI Ports */
9415         pages->page_list[p++] = SVPD_SCSI_PORTS;
9416         /* Third-party Copy */
9417         pages->page_list[p++] = SVPD_SCSI_TPC;
9418         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9419                 /* Block limits */
9420                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9421                 /* Block Device Characteristics */
9422                 pages->page_list[p++] = SVPD_BDC;
9423                 /* Logical Block Provisioning */
9424                 pages->page_list[p++] = SVPD_LBP;
9425         }
9426         pages->length = p;
9427
9428         ctl_set_success(ctsio);
9429         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9430         ctsio->be_move_done = ctl_config_move_done;
9431         ctl_datamove((union ctl_io *)ctsio);
9432         return (CTL_RETVAL_COMPLETE);
9433 }
9434
9435 /*
9436  * SCSI VPD page 0x80, the Unit Serial Number page.
9437  */
9438 static int
9439 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9440 {
9441         struct scsi_vpd_unit_serial_number *sn_ptr;
9442         struct ctl_lun *lun;
9443         int data_len;
9444
9445         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9446
9447         data_len = 4 + CTL_SN_LEN;
9448         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9449         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9450         if (data_len < alloc_len) {
9451                 ctsio->residual = alloc_len - data_len;
9452                 ctsio->kern_data_len = data_len;
9453                 ctsio->kern_total_len = data_len;
9454         } else {
9455                 ctsio->residual = 0;
9456                 ctsio->kern_data_len = alloc_len;
9457                 ctsio->kern_total_len = alloc_len;
9458         }
9459         ctsio->kern_data_resid = 0;
9460         ctsio->kern_rel_offset = 0;
9461         ctsio->kern_sg_entries = 0;
9462
9463         /*
9464          * The control device is always connected.  The disk device, on the
9465          * other hand, may not be online all the time.  Need to change this
9466          * to figure out whether the disk device is actually online or not.
9467          */
9468         if (lun != NULL)
9469                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9470                                   lun->be_lun->lun_type;
9471         else
9472                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9473
9474         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9475         sn_ptr->length = CTL_SN_LEN;
9476         /*
9477          * If we don't have a LUN, we just leave the serial number as
9478          * all spaces.
9479          */
9480         if (lun != NULL) {
9481                 strncpy((char *)sn_ptr->serial_num,
9482                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9483         } else
9484                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9485
9486         ctl_set_success(ctsio);
9487         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9488         ctsio->be_move_done = ctl_config_move_done;
9489         ctl_datamove((union ctl_io *)ctsio);
9490         return (CTL_RETVAL_COMPLETE);
9491 }
9492
9493
9494 /*
9495  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9496  */
9497 static int
9498 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9499 {
9500         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9501         struct ctl_lun *lun;
9502         int data_len;
9503
9504         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9505
9506         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9507         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9508         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9509         ctsio->kern_sg_entries = 0;
9510
9511         if (data_len < alloc_len) {
9512                 ctsio->residual = alloc_len - data_len;
9513                 ctsio->kern_data_len = data_len;
9514                 ctsio->kern_total_len = data_len;
9515         } else {
9516                 ctsio->residual = 0;
9517                 ctsio->kern_data_len = alloc_len;
9518                 ctsio->kern_total_len = alloc_len;
9519         }
9520         ctsio->kern_data_resid = 0;
9521         ctsio->kern_rel_offset = 0;
9522         ctsio->kern_sg_entries = 0;
9523
9524         /*
9525          * The control device is always connected.  The disk device, on the
9526          * other hand, may not be online all the time.
9527          */
9528         if (lun != NULL)
9529                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9530                                      lun->be_lun->lun_type;
9531         else
9532                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9533         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9534         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9535         /*
9536          * We support head of queue, ordered and simple tags.
9537          */
9538         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9539         /*
9540          * Volatile cache supported.
9541          */
9542         eid_ptr->flags3 = SVPD_EID_V_SUP;
9543
9544         /*
9545          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9546          * attention for a particular IT nexus on all LUNs once we report
9547          * it to that nexus once.  This bit is required as of SPC-4.
9548          */
9549         eid_ptr->flags4 = SVPD_EID_LUICLR;
9550
9551         /*
9552          * XXX KDM in order to correctly answer this, we would need
9553          * information from the SIM to determine how much sense data it
9554          * can send.  So this would really be a path inquiry field, most
9555          * likely.  This can be set to a maximum of 252 according to SPC-4,
9556          * but the hardware may or may not be able to support that much.
9557          * 0 just means that the maximum sense data length is not reported.
9558          */
9559         eid_ptr->max_sense_length = 0;
9560
9561         ctl_set_success(ctsio);
9562         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9563         ctsio->be_move_done = ctl_config_move_done;
9564         ctl_datamove((union ctl_io *)ctsio);
9565         return (CTL_RETVAL_COMPLETE);
9566 }
9567
9568 static int
9569 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9570 {
9571         struct scsi_vpd_mode_page_policy *mpp_ptr;
9572         struct ctl_lun *lun;
9573         int data_len;
9574
9575         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9576
9577         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9578             sizeof(struct scsi_vpd_mode_page_policy_descr);
9579
9580         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9581         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9582         ctsio->kern_sg_entries = 0;
9583
9584         if (data_len < alloc_len) {
9585                 ctsio->residual = alloc_len - data_len;
9586                 ctsio->kern_data_len = data_len;
9587                 ctsio->kern_total_len = data_len;
9588         } else {
9589                 ctsio->residual = 0;
9590                 ctsio->kern_data_len = alloc_len;
9591                 ctsio->kern_total_len = alloc_len;
9592         }
9593         ctsio->kern_data_resid = 0;
9594         ctsio->kern_rel_offset = 0;
9595         ctsio->kern_sg_entries = 0;
9596
9597         /*
9598          * The control device is always connected.  The disk device, on the
9599          * other hand, may not be online all the time.
9600          */
9601         if (lun != NULL)
9602                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9603                                      lun->be_lun->lun_type;
9604         else
9605                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9606         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9607         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9608         mpp_ptr->descr[0].page_code = 0x3f;
9609         mpp_ptr->descr[0].subpage_code = 0xff;
9610         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9611
9612         ctl_set_success(ctsio);
9613         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9614         ctsio->be_move_done = ctl_config_move_done;
9615         ctl_datamove((union ctl_io *)ctsio);
9616         return (CTL_RETVAL_COMPLETE);
9617 }
9618
9619 /*
9620  * SCSI VPD page 0x83, the Device Identification page.
9621  */
9622 static int
9623 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9624 {
9625         struct scsi_vpd_device_id *devid_ptr;
9626         struct scsi_vpd_id_descriptor *desc;
9627         struct ctl_softc *softc;
9628         struct ctl_lun *lun;
9629         struct ctl_port *port;
9630         int data_len, g;
9631         uint8_t proto;
9632
9633         softc = control_softc;
9634
9635         port = ctl_io_port(&ctsio->io_hdr);
9636         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9637
9638         data_len = sizeof(struct scsi_vpd_device_id) +
9639             sizeof(struct scsi_vpd_id_descriptor) +
9640                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9641             sizeof(struct scsi_vpd_id_descriptor) +
9642                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9643         if (lun && lun->lun_devid)
9644                 data_len += lun->lun_devid->len;
9645         if (port && port->port_devid)
9646                 data_len += port->port_devid->len;
9647         if (port && port->target_devid)
9648                 data_len += port->target_devid->len;
9649
9650         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9651         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9652         ctsio->kern_sg_entries = 0;
9653
9654         if (data_len < alloc_len) {
9655                 ctsio->residual = alloc_len - data_len;
9656                 ctsio->kern_data_len = data_len;
9657                 ctsio->kern_total_len = data_len;
9658         } else {
9659                 ctsio->residual = 0;
9660                 ctsio->kern_data_len = alloc_len;
9661                 ctsio->kern_total_len = alloc_len;
9662         }
9663         ctsio->kern_data_resid = 0;
9664         ctsio->kern_rel_offset = 0;
9665         ctsio->kern_sg_entries = 0;
9666
9667         /*
9668          * The control device is always connected.  The disk device, on the
9669          * other hand, may not be online all the time.
9670          */
9671         if (lun != NULL)
9672                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9673                                      lun->be_lun->lun_type;
9674         else
9675                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9676         devid_ptr->page_code = SVPD_DEVICE_ID;
9677         scsi_ulto2b(data_len - 4, devid_ptr->length);
9678
9679         if (port && port->port_type == CTL_PORT_FC)
9680                 proto = SCSI_PROTO_FC << 4;
9681         else if (port && port->port_type == CTL_PORT_ISCSI)
9682                 proto = SCSI_PROTO_ISCSI << 4;
9683         else
9684                 proto = SCSI_PROTO_SPI << 4;
9685         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9686
9687         /*
9688          * We're using a LUN association here.  i.e., this device ID is a
9689          * per-LUN identifier.
9690          */
9691         if (lun && lun->lun_devid) {
9692                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9693                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9694                     lun->lun_devid->len);
9695         }
9696
9697         /*
9698          * This is for the WWPN which is a port association.
9699          */
9700         if (port && port->port_devid) {
9701                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9702                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9703                     port->port_devid->len);
9704         }
9705
9706         /*
9707          * This is for the Relative Target Port(type 4h) identifier
9708          */
9709         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9710         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9711             SVPD_ID_TYPE_RELTARG;
9712         desc->length = 4;
9713         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9714         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9715             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9716
9717         /*
9718          * This is for the Target Port Group(type 5h) identifier
9719          */
9720         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9721         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9722             SVPD_ID_TYPE_TPORTGRP;
9723         desc->length = 4;
9724         if (softc->is_single ||
9725             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9726                 g = 1;
9727         else
9728                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9729         scsi_ulto2b(g, &desc->identifier[2]);
9730         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9731             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9732
9733         /*
9734          * This is for the Target identifier
9735          */
9736         if (port && port->target_devid) {
9737                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9738         }
9739
9740         ctl_set_success(ctsio);
9741         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9742         ctsio->be_move_done = ctl_config_move_done;
9743         ctl_datamove((union ctl_io *)ctsio);
9744         return (CTL_RETVAL_COMPLETE);
9745 }
9746
9747 static int
9748 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9749 {
9750         struct ctl_softc *softc = control_softc;
9751         struct scsi_vpd_scsi_ports *sp;
9752         struct scsi_vpd_port_designation *pd;
9753         struct scsi_vpd_port_designation_cont *pdc;
9754         struct ctl_lun *lun;
9755         struct ctl_port *port;
9756         int data_len, num_target_ports, iid_len, id_len;
9757
9758         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9759
9760         num_target_ports = 0;
9761         iid_len = 0;
9762         id_len = 0;
9763         mtx_lock(&softc->ctl_lock);
9764         STAILQ_FOREACH(port, &softc->port_list, links) {
9765                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9766                         continue;
9767                 if (lun != NULL &&
9768                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9769                         continue;
9770                 num_target_ports++;
9771                 if (port->init_devid)
9772                         iid_len += port->init_devid->len;
9773                 if (port->port_devid)
9774                         id_len += port->port_devid->len;
9775         }
9776         mtx_unlock(&softc->ctl_lock);
9777
9778         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9779             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9780              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9781         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9782         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9783         ctsio->kern_sg_entries = 0;
9784
9785         if (data_len < alloc_len) {
9786                 ctsio->residual = alloc_len - data_len;
9787                 ctsio->kern_data_len = data_len;
9788                 ctsio->kern_total_len = data_len;
9789         } else {
9790                 ctsio->residual = 0;
9791                 ctsio->kern_data_len = alloc_len;
9792                 ctsio->kern_total_len = alloc_len;
9793         }
9794         ctsio->kern_data_resid = 0;
9795         ctsio->kern_rel_offset = 0;
9796         ctsio->kern_sg_entries = 0;
9797
9798         /*
9799          * The control device is always connected.  The disk device, on the
9800          * other hand, may not be online all the time.  Need to change this
9801          * to figure out whether the disk device is actually online or not.
9802          */
9803         if (lun != NULL)
9804                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9805                                   lun->be_lun->lun_type;
9806         else
9807                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9808
9809         sp->page_code = SVPD_SCSI_PORTS;
9810         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9811             sp->page_length);
9812         pd = &sp->design[0];
9813
9814         mtx_lock(&softc->ctl_lock);
9815         STAILQ_FOREACH(port, &softc->port_list, links) {
9816                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9817                         continue;
9818                 if (lun != NULL &&
9819                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9820                         continue;
9821                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9822                 if (port->init_devid) {
9823                         iid_len = port->init_devid->len;
9824                         memcpy(pd->initiator_transportid,
9825                             port->init_devid->data, port->init_devid->len);
9826                 } else
9827                         iid_len = 0;
9828                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9829                 pdc = (struct scsi_vpd_port_designation_cont *)
9830                     (&pd->initiator_transportid[iid_len]);
9831                 if (port->port_devid) {
9832                         id_len = port->port_devid->len;
9833                         memcpy(pdc->target_port_descriptors,
9834                             port->port_devid->data, port->port_devid->len);
9835                 } else
9836                         id_len = 0;
9837                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9838                 pd = (struct scsi_vpd_port_designation *)
9839                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9840         }
9841         mtx_unlock(&softc->ctl_lock);
9842
9843         ctl_set_success(ctsio);
9844         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9845         ctsio->be_move_done = ctl_config_move_done;
9846         ctl_datamove((union ctl_io *)ctsio);
9847         return (CTL_RETVAL_COMPLETE);
9848 }
9849
9850 static int
9851 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9852 {
9853         struct scsi_vpd_block_limits *bl_ptr;
9854         struct ctl_lun *lun;
9855         uint64_t ival;
9856
9857         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9858
9859         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9860         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9861         ctsio->kern_sg_entries = 0;
9862
9863         if (sizeof(*bl_ptr) < alloc_len) {
9864                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9865                 ctsio->kern_data_len = sizeof(*bl_ptr);
9866                 ctsio->kern_total_len = sizeof(*bl_ptr);
9867         } else {
9868                 ctsio->residual = 0;
9869                 ctsio->kern_data_len = alloc_len;
9870                 ctsio->kern_total_len = alloc_len;
9871         }
9872         ctsio->kern_data_resid = 0;
9873         ctsio->kern_rel_offset = 0;
9874         ctsio->kern_sg_entries = 0;
9875
9876         /*
9877          * The control device is always connected.  The disk device, on the
9878          * other hand, may not be online all the time.  Need to change this
9879          * to figure out whether the disk device is actually online or not.
9880          */
9881         if (lun != NULL)
9882                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9883                                   lun->be_lun->lun_type;
9884         else
9885                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9886
9887         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9888         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9889         bl_ptr->max_cmp_write_len = 0xff;
9890         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9891         if (lun != NULL) {
9892                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9893                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9894                         ival = 0xffffffff;
9895                         ctl_get_opt_number(&lun->be_lun->options,
9896                             "unmap_max_lba", &ival);
9897                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9898                         ival = 0xffffffff;
9899                         ctl_get_opt_number(&lun->be_lun->options,
9900                             "unmap_max_descr", &ival);
9901                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9902                         if (lun->be_lun->ublockexp != 0) {
9903                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9904                                     bl_ptr->opt_unmap_grain);
9905                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9906                                     bl_ptr->unmap_grain_align);
9907                         }
9908                 }
9909                 scsi_ulto4b(lun->be_lun->atomicblock,
9910                     bl_ptr->max_atomic_transfer_length);
9911                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9912                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9913                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9914                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9915                 ival = UINT64_MAX;
9916                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9917                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9918         }
9919
9920         ctl_set_success(ctsio);
9921         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9922         ctsio->be_move_done = ctl_config_move_done;
9923         ctl_datamove((union ctl_io *)ctsio);
9924         return (CTL_RETVAL_COMPLETE);
9925 }
9926
9927 static int
9928 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9929 {
9930         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9931         struct ctl_lun *lun;
9932         const char *value;
9933         u_int i;
9934
9935         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9936
9937         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9938         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9939         ctsio->kern_sg_entries = 0;
9940
9941         if (sizeof(*bdc_ptr) < alloc_len) {
9942                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9943                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9944                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9945         } else {
9946                 ctsio->residual = 0;
9947                 ctsio->kern_data_len = alloc_len;
9948                 ctsio->kern_total_len = alloc_len;
9949         }
9950         ctsio->kern_data_resid = 0;
9951         ctsio->kern_rel_offset = 0;
9952         ctsio->kern_sg_entries = 0;
9953
9954         /*
9955          * The control device is always connected.  The disk device, on the
9956          * other hand, may not be online all the time.  Need to change this
9957          * to figure out whether the disk device is actually online or not.
9958          */
9959         if (lun != NULL)
9960                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9961                                   lun->be_lun->lun_type;
9962         else
9963                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9964         bdc_ptr->page_code = SVPD_BDC;
9965         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9966         if (lun != NULL &&
9967             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9968                 i = strtol(value, NULL, 0);
9969         else
9970                 i = CTL_DEFAULT_ROTATION_RATE;
9971         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9972         if (lun != NULL &&
9973             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9974                 i = strtol(value, NULL, 0);
9975         else
9976                 i = 0;
9977         bdc_ptr->wab_wac_ff = (i & 0x0f);
9978         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9979
9980         ctl_set_success(ctsio);
9981         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9982         ctsio->be_move_done = ctl_config_move_done;
9983         ctl_datamove((union ctl_io *)ctsio);
9984         return (CTL_RETVAL_COMPLETE);
9985 }
9986
9987 static int
9988 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9989 {
9990         struct scsi_vpd_logical_block_prov *lbp_ptr;
9991         struct ctl_lun *lun;
9992         const char *value;
9993
9994         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9995
9996         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9997         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9998         ctsio->kern_sg_entries = 0;
9999
10000         if (sizeof(*lbp_ptr) < alloc_len) {
10001                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
10002                 ctsio->kern_data_len = sizeof(*lbp_ptr);
10003                 ctsio->kern_total_len = sizeof(*lbp_ptr);
10004         } else {
10005                 ctsio->residual = 0;
10006                 ctsio->kern_data_len = alloc_len;
10007                 ctsio->kern_total_len = alloc_len;
10008         }
10009         ctsio->kern_data_resid = 0;
10010         ctsio->kern_rel_offset = 0;
10011         ctsio->kern_sg_entries = 0;
10012
10013         /*
10014          * The control device is always connected.  The disk device, on the
10015          * other hand, may not be online all the time.  Need to change this
10016          * to figure out whether the disk device is actually online or not.
10017          */
10018         if (lun != NULL)
10019                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10020                                   lun->be_lun->lun_type;
10021         else
10022                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10023
10024         lbp_ptr->page_code = SVPD_LBP;
10025         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10026         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10027         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10028                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10029                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10030                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10031                 if (value != NULL) {
10032                         if (strcmp(value, "resource") == 0)
10033                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10034                         else if (strcmp(value, "thin") == 0)
10035                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10036                 } else
10037                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10038         }
10039
10040         ctl_set_success(ctsio);
10041         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10042         ctsio->be_move_done = ctl_config_move_done;
10043         ctl_datamove((union ctl_io *)ctsio);
10044         return (CTL_RETVAL_COMPLETE);
10045 }
10046
10047 /*
10048  * INQUIRY with the EVPD bit set.
10049  */
10050 static int
10051 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10052 {
10053         struct ctl_lun *lun;
10054         struct scsi_inquiry *cdb;
10055         int alloc_len, retval;
10056
10057         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10058         cdb = (struct scsi_inquiry *)ctsio->cdb;
10059         alloc_len = scsi_2btoul(cdb->length);
10060
10061         switch (cdb->page_code) {
10062         case SVPD_SUPPORTED_PAGES:
10063                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10064                 break;
10065         case SVPD_UNIT_SERIAL_NUMBER:
10066                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10067                 break;
10068         case SVPD_DEVICE_ID:
10069                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10070                 break;
10071         case SVPD_EXTENDED_INQUIRY_DATA:
10072                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10073                 break;
10074         case SVPD_MODE_PAGE_POLICY:
10075                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10076                 break;
10077         case SVPD_SCSI_PORTS:
10078                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10079                 break;
10080         case SVPD_SCSI_TPC:
10081                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10082                 break;
10083         case SVPD_BLOCK_LIMITS:
10084                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10085                         goto err;
10086                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10087                 break;
10088         case SVPD_BDC:
10089                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10090                         goto err;
10091                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10092                 break;
10093         case SVPD_LBP:
10094                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10095                         goto err;
10096                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10097                 break;
10098         default:
10099 err:
10100                 ctl_set_invalid_field(ctsio,
10101                                       /*sks_valid*/ 1,
10102                                       /*command*/ 1,
10103                                       /*field*/ 2,
10104                                       /*bit_valid*/ 0,
10105                                       /*bit*/ 0);
10106                 ctl_done((union ctl_io *)ctsio);
10107                 retval = CTL_RETVAL_COMPLETE;
10108                 break;
10109         }
10110
10111         return (retval);
10112 }
10113
10114 /*
10115  * Standard INQUIRY data.
10116  */
10117 static int
10118 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10119 {
10120         struct scsi_inquiry_data *inq_ptr;
10121         struct scsi_inquiry *cdb;
10122         struct ctl_softc *softc = control_softc;
10123         struct ctl_port *port;
10124         struct ctl_lun *lun;
10125         char *val;
10126         uint32_t alloc_len, data_len;
10127         ctl_port_type port_type;
10128
10129         port = ctl_io_port(&ctsio->io_hdr);
10130         port_type = port->port_type;
10131         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10132                 port_type = CTL_PORT_SCSI;
10133
10134         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10135         cdb = (struct scsi_inquiry *)ctsio->cdb;
10136         alloc_len = scsi_2btoul(cdb->length);
10137
10138         /*
10139          * We malloc the full inquiry data size here and fill it
10140          * in.  If the user only asks for less, we'll give him
10141          * that much.
10142          */
10143         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10144         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10145         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10146         ctsio->kern_sg_entries = 0;
10147         ctsio->kern_data_resid = 0;
10148         ctsio->kern_rel_offset = 0;
10149
10150         if (data_len < alloc_len) {
10151                 ctsio->residual = alloc_len - data_len;
10152                 ctsio->kern_data_len = data_len;
10153                 ctsio->kern_total_len = data_len;
10154         } else {
10155                 ctsio->residual = 0;
10156                 ctsio->kern_data_len = alloc_len;
10157                 ctsio->kern_total_len = alloc_len;
10158         }
10159
10160         if (lun != NULL) {
10161                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10162                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10163                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10164                             lun->be_lun->lun_type;
10165                 } else {
10166                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10167                             lun->be_lun->lun_type;
10168                 }
10169                 if (lun->flags & CTL_LUN_REMOVABLE)
10170                         inq_ptr->dev_qual2 |= SID_RMB;
10171         } else
10172                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10173
10174         /* RMB in byte 2 is 0 */
10175         inq_ptr->version = SCSI_REV_SPC5;
10176
10177         /*
10178          * According to SAM-3, even if a device only supports a single
10179          * level of LUN addressing, it should still set the HISUP bit:
10180          *
10181          * 4.9.1 Logical unit numbers overview
10182          *
10183          * All logical unit number formats described in this standard are
10184          * hierarchical in structure even when only a single level in that
10185          * hierarchy is used. The HISUP bit shall be set to one in the
10186          * standard INQUIRY data (see SPC-2) when any logical unit number
10187          * format described in this standard is used.  Non-hierarchical
10188          * formats are outside the scope of this standard.
10189          *
10190          * Therefore we set the HiSup bit here.
10191          *
10192          * The response format is 2, per SPC-3.
10193          */
10194         inq_ptr->response_format = SID_HiSup | 2;
10195
10196         inq_ptr->additional_length = data_len -
10197             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10198         CTL_DEBUG_PRINT(("additional_length = %d\n",
10199                          inq_ptr->additional_length));
10200
10201         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10202         if (port_type == CTL_PORT_SCSI)
10203                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10204         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10205         inq_ptr->flags = SID_CmdQue;
10206         if (port_type == CTL_PORT_SCSI)
10207                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10208
10209         /*
10210          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10211          * We have 8 bytes for the vendor name, and 16 bytes for the device
10212          * name and 4 bytes for the revision.
10213          */
10214         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10215             "vendor")) == NULL) {
10216                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10217         } else {
10218                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10219                 strncpy(inq_ptr->vendor, val,
10220                     min(sizeof(inq_ptr->vendor), strlen(val)));
10221         }
10222         if (lun == NULL) {
10223                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10224                     sizeof(inq_ptr->product));
10225         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10226                 switch (lun->be_lun->lun_type) {
10227                 case T_DIRECT:
10228                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10229                             sizeof(inq_ptr->product));
10230                         break;
10231                 case T_PROCESSOR:
10232                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10233                             sizeof(inq_ptr->product));
10234                         break;
10235                 case T_CDROM:
10236                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10237                             sizeof(inq_ptr->product));
10238                         break;
10239                 default:
10240                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10241                             sizeof(inq_ptr->product));
10242                         break;
10243                 }
10244         } else {
10245                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10246                 strncpy(inq_ptr->product, val,
10247                     min(sizeof(inq_ptr->product), strlen(val)));
10248         }
10249
10250         /*
10251          * XXX make this a macro somewhere so it automatically gets
10252          * incremented when we make changes.
10253          */
10254         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10255             "revision")) == NULL) {
10256                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10257         } else {
10258                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10259                 strncpy(inq_ptr->revision, val,
10260                     min(sizeof(inq_ptr->revision), strlen(val)));
10261         }
10262
10263         /*
10264          * For parallel SCSI, we support double transition and single
10265          * transition clocking.  We also support QAS (Quick Arbitration
10266          * and Selection) and Information Unit transfers on both the
10267          * control and array devices.
10268          */
10269         if (port_type == CTL_PORT_SCSI)
10270                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10271                                     SID_SPI_IUS;
10272
10273         /* SAM-6 (no version claimed) */
10274         scsi_ulto2b(0x00C0, inq_ptr->version1);
10275         /* SPC-5 (no version claimed) */
10276         scsi_ulto2b(0x05C0, inq_ptr->version2);
10277         if (port_type == CTL_PORT_FC) {
10278                 /* FCP-2 ANSI INCITS.350:2003 */
10279                 scsi_ulto2b(0x0917, inq_ptr->version3);
10280         } else if (port_type == CTL_PORT_SCSI) {
10281                 /* SPI-4 ANSI INCITS.362:200x */
10282                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10283         } else if (port_type == CTL_PORT_ISCSI) {
10284                 /* iSCSI (no version claimed) */
10285                 scsi_ulto2b(0x0960, inq_ptr->version3);
10286         } else if (port_type == CTL_PORT_SAS) {
10287                 /* SAS (no version claimed) */
10288                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10289         }
10290
10291         if (lun == NULL) {
10292                 /* SBC-4 (no version claimed) */
10293                 scsi_ulto2b(0x0600, inq_ptr->version4);
10294         } else {
10295                 switch (lun->be_lun->lun_type) {
10296                 case T_DIRECT:
10297                         /* SBC-4 (no version claimed) */
10298                         scsi_ulto2b(0x0600, inq_ptr->version4);
10299                         break;
10300                 case T_PROCESSOR:
10301                         break;
10302                 case T_CDROM:
10303                         /* MMC-6 (no version claimed) */
10304                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10305                         break;
10306                 default:
10307                         break;
10308                 }
10309         }
10310
10311         ctl_set_success(ctsio);
10312         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10313         ctsio->be_move_done = ctl_config_move_done;
10314         ctl_datamove((union ctl_io *)ctsio);
10315         return (CTL_RETVAL_COMPLETE);
10316 }
10317
10318 int
10319 ctl_inquiry(struct ctl_scsiio *ctsio)
10320 {
10321         struct scsi_inquiry *cdb;
10322         int retval;
10323
10324         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10325
10326         cdb = (struct scsi_inquiry *)ctsio->cdb;
10327         if (cdb->byte2 & SI_EVPD)
10328                 retval = ctl_inquiry_evpd(ctsio);
10329         else if (cdb->page_code == 0)
10330                 retval = ctl_inquiry_std(ctsio);
10331         else {
10332                 ctl_set_invalid_field(ctsio,
10333                                       /*sks_valid*/ 1,
10334                                       /*command*/ 1,
10335                                       /*field*/ 2,
10336                                       /*bit_valid*/ 0,
10337                                       /*bit*/ 0);
10338                 ctl_done((union ctl_io *)ctsio);
10339                 return (CTL_RETVAL_COMPLETE);
10340         }
10341
10342         return (retval);
10343 }
10344
10345 int
10346 ctl_get_config(struct ctl_scsiio *ctsio)
10347 {
10348         struct scsi_get_config_header *hdr;
10349         struct scsi_get_config_feature *feature;
10350         struct scsi_get_config *cdb;
10351         struct ctl_lun *lun;
10352         uint32_t alloc_len, data_len;
10353         int rt, starting;
10354
10355         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10356         cdb = (struct scsi_get_config *)ctsio->cdb;
10357         rt = (cdb->rt & SGC_RT_MASK);
10358         starting = scsi_2btoul(cdb->starting_feature);
10359         alloc_len = scsi_2btoul(cdb->length);
10360
10361         data_len = sizeof(struct scsi_get_config_header) +
10362             sizeof(struct scsi_get_config_feature) + 8 +
10363             sizeof(struct scsi_get_config_feature) + 8 +
10364             sizeof(struct scsi_get_config_feature) + 4 +
10365             sizeof(struct scsi_get_config_feature) + 4 +
10366             sizeof(struct scsi_get_config_feature) + 8 +
10367             sizeof(struct scsi_get_config_feature) +
10368             sizeof(struct scsi_get_config_feature) + 4 +
10369             sizeof(struct scsi_get_config_feature) + 4 +
10370             sizeof(struct scsi_get_config_feature) + 4 +
10371             sizeof(struct scsi_get_config_feature) + 4 +
10372             sizeof(struct scsi_get_config_feature) + 4 +
10373             sizeof(struct scsi_get_config_feature) + 4;
10374         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10375         ctsio->kern_sg_entries = 0;
10376         ctsio->kern_data_resid = 0;
10377         ctsio->kern_rel_offset = 0;
10378
10379         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10380         if (lun->flags & CTL_LUN_NO_MEDIA)
10381                 scsi_ulto2b(0x0000, hdr->current_profile);
10382         else
10383                 scsi_ulto2b(0x0010, hdr->current_profile);
10384         feature = (struct scsi_get_config_feature *)(hdr + 1);
10385
10386         if (starting > 0x003b)
10387                 goto done;
10388         if (starting > 0x003a)
10389                 goto f3b;
10390         if (starting > 0x002b)
10391                 goto f3a;
10392         if (starting > 0x002a)
10393                 goto f2b;
10394         if (starting > 0x001f)
10395                 goto f2a;
10396         if (starting > 0x001e)
10397                 goto f1f;
10398         if (starting > 0x001d)
10399                 goto f1e;
10400         if (starting > 0x0010)
10401                 goto f1d;
10402         if (starting > 0x0003)
10403                 goto f10;
10404         if (starting > 0x0002)
10405                 goto f3;
10406         if (starting > 0x0001)
10407                 goto f2;
10408         if (starting > 0x0000)
10409                 goto f1;
10410
10411         /* Profile List */
10412         scsi_ulto2b(0x0000, feature->feature_code);
10413         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10414         feature->add_length = 8;
10415         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10416         feature->feature_data[2] = 0x00;
10417         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10418         feature->feature_data[6] = 0x01;
10419         feature = (struct scsi_get_config_feature *)
10420             &feature->feature_data[feature->add_length];
10421
10422 f1:     /* Core */
10423         scsi_ulto2b(0x0001, feature->feature_code);
10424         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10425         feature->add_length = 8;
10426         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10427         feature->feature_data[4] = 0x03;
10428         feature = (struct scsi_get_config_feature *)
10429             &feature->feature_data[feature->add_length];
10430
10431 f2:     /* Morphing */
10432         scsi_ulto2b(0x0002, feature->feature_code);
10433         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10434         feature->add_length = 4;
10435         feature->feature_data[0] = 0x02;
10436         feature = (struct scsi_get_config_feature *)
10437             &feature->feature_data[feature->add_length];
10438
10439 f3:     /* Removable Medium */
10440         scsi_ulto2b(0x0003, feature->feature_code);
10441         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10442         feature->add_length = 4;
10443         feature->feature_data[0] = 0x39;
10444         feature = (struct scsi_get_config_feature *)
10445             &feature->feature_data[feature->add_length];
10446
10447         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10448                 goto done;
10449
10450 f10:    /* Random Read */
10451         scsi_ulto2b(0x0010, feature->feature_code);
10452         feature->flags = 0x00;
10453         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10454                 feature->flags |= SGC_F_CURRENT;
10455         feature->add_length = 8;
10456         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10457         scsi_ulto2b(1, &feature->feature_data[4]);
10458         feature->feature_data[6] = 0x00;
10459         feature = (struct scsi_get_config_feature *)
10460             &feature->feature_data[feature->add_length];
10461
10462 f1d:    /* Multi-Read */
10463         scsi_ulto2b(0x001D, feature->feature_code);
10464         feature->flags = 0x00;
10465         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10466                 feature->flags |= SGC_F_CURRENT;
10467         feature->add_length = 0;
10468         feature = (struct scsi_get_config_feature *)
10469             &feature->feature_data[feature->add_length];
10470
10471 f1e:    /* CD Read */
10472         scsi_ulto2b(0x001E, feature->feature_code);
10473         feature->flags = 0x00;
10474         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10475                 feature->flags |= SGC_F_CURRENT;
10476         feature->add_length = 4;
10477         feature->feature_data[0] = 0x00;
10478         feature = (struct scsi_get_config_feature *)
10479             &feature->feature_data[feature->add_length];
10480
10481 f1f:    /* DVD Read */
10482         scsi_ulto2b(0x001F, feature->feature_code);
10483         feature->flags = 0x08;
10484         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10485                 feature->flags |= SGC_F_CURRENT;
10486         feature->add_length = 4;
10487         feature->feature_data[0] = 0x01;
10488         feature->feature_data[2] = 0x03;
10489         feature = (struct scsi_get_config_feature *)
10490             &feature->feature_data[feature->add_length];
10491
10492 f2a:    /* DVD+RW */
10493         scsi_ulto2b(0x002A, feature->feature_code);
10494         feature->flags = 0x04;
10495         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10496                 feature->flags |= SGC_F_CURRENT;
10497         feature->add_length = 4;
10498         feature->feature_data[0] = 0x00;
10499         feature->feature_data[1] = 0x00;
10500         feature = (struct scsi_get_config_feature *)
10501             &feature->feature_data[feature->add_length];
10502
10503 f2b:    /* DVD+R */
10504         scsi_ulto2b(0x002B, feature->feature_code);
10505         feature->flags = 0x00;
10506         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10507                 feature->flags |= SGC_F_CURRENT;
10508         feature->add_length = 4;
10509         feature->feature_data[0] = 0x00;
10510         feature = (struct scsi_get_config_feature *)
10511             &feature->feature_data[feature->add_length];
10512
10513 f3a:    /* DVD+RW Dual Layer */
10514         scsi_ulto2b(0x003A, feature->feature_code);
10515         feature->flags = 0x00;
10516         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10517                 feature->flags |= SGC_F_CURRENT;
10518         feature->add_length = 4;
10519         feature->feature_data[0] = 0x00;
10520         feature->feature_data[1] = 0x00;
10521         feature = (struct scsi_get_config_feature *)
10522             &feature->feature_data[feature->add_length];
10523
10524 f3b:    /* DVD+R Dual Layer */
10525         scsi_ulto2b(0x003B, feature->feature_code);
10526         feature->flags = 0x00;
10527         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10528                 feature->flags |= SGC_F_CURRENT;
10529         feature->add_length = 4;
10530         feature->feature_data[0] = 0x00;
10531         feature = (struct scsi_get_config_feature *)
10532             &feature->feature_data[feature->add_length];
10533
10534 done:
10535         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10536         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10537                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10538                 if (scsi_2btoul(feature->feature_code) == starting)
10539                         feature = (struct scsi_get_config_feature *)
10540                             &feature->feature_data[feature->add_length];
10541                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10542         }
10543         scsi_ulto4b(data_len - 4, hdr->data_length);
10544         if (data_len < alloc_len) {
10545                 ctsio->residual = alloc_len - data_len;
10546                 ctsio->kern_data_len = data_len;
10547                 ctsio->kern_total_len = data_len;
10548         } else {
10549                 ctsio->residual = 0;
10550                 ctsio->kern_data_len = alloc_len;
10551                 ctsio->kern_total_len = alloc_len;
10552         }
10553
10554         ctl_set_success(ctsio);
10555         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10556         ctsio->be_move_done = ctl_config_move_done;
10557         ctl_datamove((union ctl_io *)ctsio);
10558         return (CTL_RETVAL_COMPLETE);
10559 }
10560
10561 int
10562 ctl_get_event_status(struct ctl_scsiio *ctsio)
10563 {
10564         struct scsi_get_event_status_header *hdr;
10565         struct scsi_get_event_status *cdb;
10566         struct ctl_lun *lun;
10567         uint32_t alloc_len, data_len;
10568         int notif_class;
10569
10570         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10571         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10572         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10573                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10574                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10575                 ctl_done((union ctl_io *)ctsio);
10576                 return (CTL_RETVAL_COMPLETE);
10577         }
10578         notif_class = cdb->notif_class;
10579         alloc_len = scsi_2btoul(cdb->length);
10580
10581         data_len = sizeof(struct scsi_get_event_status_header);
10582         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10583         ctsio->kern_sg_entries = 0;
10584         ctsio->kern_data_resid = 0;
10585         ctsio->kern_rel_offset = 0;
10586
10587         if (data_len < alloc_len) {
10588                 ctsio->residual = alloc_len - data_len;
10589                 ctsio->kern_data_len = data_len;
10590                 ctsio->kern_total_len = data_len;
10591         } else {
10592                 ctsio->residual = 0;
10593                 ctsio->kern_data_len = alloc_len;
10594                 ctsio->kern_total_len = alloc_len;
10595         }
10596
10597         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10598         scsi_ulto2b(0, hdr->descr_length);
10599         hdr->nea_class = SGESN_NEA;
10600         hdr->supported_class = 0;
10601
10602         ctl_set_success(ctsio);
10603         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10604         ctsio->be_move_done = ctl_config_move_done;
10605         ctl_datamove((union ctl_io *)ctsio);
10606         return (CTL_RETVAL_COMPLETE);
10607 }
10608
10609 int
10610 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10611 {
10612         struct scsi_mechanism_status_header *hdr;
10613         struct scsi_mechanism_status *cdb;
10614         struct ctl_lun *lun;
10615         uint32_t alloc_len, data_len;
10616
10617         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10618         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10619         alloc_len = scsi_2btoul(cdb->length);
10620
10621         data_len = sizeof(struct scsi_mechanism_status_header);
10622         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10623         ctsio->kern_sg_entries = 0;
10624         ctsio->kern_data_resid = 0;
10625         ctsio->kern_rel_offset = 0;
10626
10627         if (data_len < alloc_len) {
10628                 ctsio->residual = alloc_len - data_len;
10629                 ctsio->kern_data_len = data_len;
10630                 ctsio->kern_total_len = data_len;
10631         } else {
10632                 ctsio->residual = 0;
10633                 ctsio->kern_data_len = alloc_len;
10634                 ctsio->kern_total_len = alloc_len;
10635         }
10636
10637         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10638         hdr->state1 = 0x00;
10639         hdr->state2 = 0xe0;
10640         scsi_ulto3b(0, hdr->lba);
10641         hdr->slots_num = 0;
10642         scsi_ulto2b(0, hdr->slots_length);
10643
10644         ctl_set_success(ctsio);
10645         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10646         ctsio->be_move_done = ctl_config_move_done;
10647         ctl_datamove((union ctl_io *)ctsio);
10648         return (CTL_RETVAL_COMPLETE);
10649 }
10650
10651 static void
10652 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10653 {
10654
10655         lba += 150;
10656         buf[0] = 0;
10657         buf[1] = bin2bcd((lba / 75) / 60);
10658         buf[2] = bin2bcd((lba / 75) % 60);
10659         buf[3] = bin2bcd(lba % 75);
10660 }
10661
10662 int
10663 ctl_read_toc(struct ctl_scsiio *ctsio)
10664 {
10665         struct scsi_read_toc_hdr *hdr;
10666         struct scsi_read_toc_type01_descr *descr;
10667         struct scsi_read_toc *cdb;
10668         struct ctl_lun *lun;
10669         uint32_t alloc_len, data_len;
10670         int format, msf;
10671
10672         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10673         cdb = (struct scsi_read_toc *)ctsio->cdb;
10674         msf = (cdb->byte2 & CD_MSF) != 0;
10675         format = cdb->format;
10676         alloc_len = scsi_2btoul(cdb->data_len);
10677
10678         data_len = sizeof(struct scsi_read_toc_hdr);
10679         if (format == 0)
10680                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10681         else
10682                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10683         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10684         ctsio->kern_sg_entries = 0;
10685         ctsio->kern_data_resid = 0;
10686         ctsio->kern_rel_offset = 0;
10687
10688         if (data_len < alloc_len) {
10689                 ctsio->residual = alloc_len - data_len;
10690                 ctsio->kern_data_len = data_len;
10691                 ctsio->kern_total_len = data_len;
10692         } else {
10693                 ctsio->residual = 0;
10694                 ctsio->kern_data_len = alloc_len;
10695                 ctsio->kern_total_len = alloc_len;
10696         }
10697
10698         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10699         if (format == 0) {
10700                 scsi_ulto2b(0x12, hdr->data_length);
10701                 hdr->first = 1;
10702                 hdr->last = 1;
10703                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10704                 descr->addr_ctl = 0x14;
10705                 descr->track_number = 1;
10706                 if (msf)
10707                         ctl_ultomsf(0, descr->track_start);
10708                 else
10709                         scsi_ulto4b(0, descr->track_start);
10710                 descr++;
10711                 descr->addr_ctl = 0x14;
10712                 descr->track_number = 0xaa;
10713                 if (msf)
10714                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10715                 else
10716                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10717         } else {
10718                 scsi_ulto2b(0x0a, hdr->data_length);
10719                 hdr->first = 1;
10720                 hdr->last = 1;
10721                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10722                 descr->addr_ctl = 0x14;
10723                 descr->track_number = 1;
10724                 if (msf)
10725                         ctl_ultomsf(0, descr->track_start);
10726                 else
10727                         scsi_ulto4b(0, descr->track_start);
10728         }
10729
10730         ctl_set_success(ctsio);
10731         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10732         ctsio->be_move_done = ctl_config_move_done;
10733         ctl_datamove((union ctl_io *)ctsio);
10734         return (CTL_RETVAL_COMPLETE);
10735 }
10736
10737 /*
10738  * For known CDB types, parse the LBA and length.
10739  */
10740 static int
10741 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10742 {
10743         if (io->io_hdr.io_type != CTL_IO_SCSI)
10744                 return (1);
10745
10746         switch (io->scsiio.cdb[0]) {
10747         case COMPARE_AND_WRITE: {
10748                 struct scsi_compare_and_write *cdb;
10749
10750                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10751
10752                 *lba = scsi_8btou64(cdb->addr);
10753                 *len = cdb->length;
10754                 break;
10755         }
10756         case READ_6:
10757         case WRITE_6: {
10758                 struct scsi_rw_6 *cdb;
10759
10760                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10761
10762                 *lba = scsi_3btoul(cdb->addr);
10763                 /* only 5 bits are valid in the most significant address byte */
10764                 *lba &= 0x1fffff;
10765                 *len = cdb->length;
10766                 break;
10767         }
10768         case READ_10:
10769         case WRITE_10: {
10770                 struct scsi_rw_10 *cdb;
10771
10772                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10773
10774                 *lba = scsi_4btoul(cdb->addr);
10775                 *len = scsi_2btoul(cdb->length);
10776                 break;
10777         }
10778         case WRITE_VERIFY_10: {
10779                 struct scsi_write_verify_10 *cdb;
10780
10781                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10782
10783                 *lba = scsi_4btoul(cdb->addr);
10784                 *len = scsi_2btoul(cdb->length);
10785                 break;
10786         }
10787         case READ_12:
10788         case WRITE_12: {
10789                 struct scsi_rw_12 *cdb;
10790
10791                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10792
10793                 *lba = scsi_4btoul(cdb->addr);
10794                 *len = scsi_4btoul(cdb->length);
10795                 break;
10796         }
10797         case WRITE_VERIFY_12: {
10798                 struct scsi_write_verify_12 *cdb;
10799
10800                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10801
10802                 *lba = scsi_4btoul(cdb->addr);
10803                 *len = scsi_4btoul(cdb->length);
10804                 break;
10805         }
10806         case READ_16:
10807         case WRITE_16: {
10808                 struct scsi_rw_16 *cdb;
10809
10810                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10811
10812                 *lba = scsi_8btou64(cdb->addr);
10813                 *len = scsi_4btoul(cdb->length);
10814                 break;
10815         }
10816         case WRITE_ATOMIC_16: {
10817                 struct scsi_write_atomic_16 *cdb;
10818
10819                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10820
10821                 *lba = scsi_8btou64(cdb->addr);
10822                 *len = scsi_2btoul(cdb->length);
10823                 break;
10824         }
10825         case WRITE_VERIFY_16: {
10826                 struct scsi_write_verify_16 *cdb;
10827
10828                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10829
10830                 *lba = scsi_8btou64(cdb->addr);
10831                 *len = scsi_4btoul(cdb->length);
10832                 break;
10833         }
10834         case WRITE_SAME_10: {
10835                 struct scsi_write_same_10 *cdb;
10836
10837                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10838
10839                 *lba = scsi_4btoul(cdb->addr);
10840                 *len = scsi_2btoul(cdb->length);
10841                 break;
10842         }
10843         case WRITE_SAME_16: {
10844                 struct scsi_write_same_16 *cdb;
10845
10846                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10847
10848                 *lba = scsi_8btou64(cdb->addr);
10849                 *len = scsi_4btoul(cdb->length);
10850                 break;
10851         }
10852         case VERIFY_10: {
10853                 struct scsi_verify_10 *cdb;
10854
10855                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10856
10857                 *lba = scsi_4btoul(cdb->addr);
10858                 *len = scsi_2btoul(cdb->length);
10859                 break;
10860         }
10861         case VERIFY_12: {
10862                 struct scsi_verify_12 *cdb;
10863
10864                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10865
10866                 *lba = scsi_4btoul(cdb->addr);
10867                 *len = scsi_4btoul(cdb->length);
10868                 break;
10869         }
10870         case VERIFY_16: {
10871                 struct scsi_verify_16 *cdb;
10872
10873                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10874
10875                 *lba = scsi_8btou64(cdb->addr);
10876                 *len = scsi_4btoul(cdb->length);
10877                 break;
10878         }
10879         case UNMAP: {
10880                 *lba = 0;
10881                 *len = UINT64_MAX;
10882                 break;
10883         }
10884         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10885                 struct scsi_get_lba_status *cdb;
10886
10887                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10888                 *lba = scsi_8btou64(cdb->addr);
10889                 *len = UINT32_MAX;
10890                 break;
10891         }
10892         default:
10893                 return (1);
10894                 break; /* NOTREACHED */
10895         }
10896
10897         return (0);
10898 }
10899
10900 static ctl_action
10901 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10902     bool seq)
10903 {
10904         uint64_t endlba1, endlba2;
10905
10906         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10907         endlba2 = lba2 + len2 - 1;
10908
10909         if ((endlba1 < lba2) || (endlba2 < lba1))
10910                 return (CTL_ACTION_PASS);
10911         else
10912                 return (CTL_ACTION_BLOCK);
10913 }
10914
10915 static int
10916 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10917 {
10918         struct ctl_ptr_len_flags *ptrlen;
10919         struct scsi_unmap_desc *buf, *end, *range;
10920         uint64_t lba;
10921         uint32_t len;
10922
10923         /* If not UNMAP -- go other way. */
10924         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10925             io->scsiio.cdb[0] != UNMAP)
10926                 return (CTL_ACTION_ERROR);
10927
10928         /* If UNMAP without data -- block and wait for data. */
10929         ptrlen = (struct ctl_ptr_len_flags *)
10930             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10931         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10932             ptrlen->ptr == NULL)
10933                 return (CTL_ACTION_BLOCK);
10934
10935         /* UNMAP with data -- check for collision. */
10936         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10937         end = buf + ptrlen->len / sizeof(*buf);
10938         for (range = buf; range < end; range++) {
10939                 lba = scsi_8btou64(range->lba);
10940                 len = scsi_4btoul(range->length);
10941                 if ((lba < lba2 + len2) && (lba + len > lba2))
10942                         return (CTL_ACTION_BLOCK);
10943         }
10944         return (CTL_ACTION_PASS);
10945 }
10946
10947 static ctl_action
10948 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10949 {
10950         uint64_t lba1, lba2;
10951         uint64_t len1, len2;
10952         int retval;
10953
10954         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10955                 return (CTL_ACTION_ERROR);
10956
10957         retval = ctl_extent_check_unmap(io1, lba2, len2);
10958         if (retval != CTL_ACTION_ERROR)
10959                 return (retval);
10960
10961         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10962                 return (CTL_ACTION_ERROR);
10963
10964         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10965                 seq = FALSE;
10966         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10967 }
10968
10969 static ctl_action
10970 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10971 {
10972         uint64_t lba1, lba2;
10973         uint64_t len1, len2;
10974
10975         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10976                 return (CTL_ACTION_PASS);
10977         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10978                 return (CTL_ACTION_ERROR);
10979         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10980                 return (CTL_ACTION_ERROR);
10981
10982         if (lba1 + len1 == lba2)
10983                 return (CTL_ACTION_BLOCK);
10984         return (CTL_ACTION_PASS);
10985 }
10986
10987 static ctl_action
10988 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10989     union ctl_io *ooa_io)
10990 {
10991         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10992         const ctl_serialize_action *serialize_row;
10993
10994         /*
10995          * The initiator attempted multiple untagged commands at the same
10996          * time.  Can't do that.
10997          */
10998         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10999          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11000          && ((pending_io->io_hdr.nexus.targ_port ==
11001               ooa_io->io_hdr.nexus.targ_port)
11002           && (pending_io->io_hdr.nexus.initid ==
11003               ooa_io->io_hdr.nexus.initid))
11004          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11005               CTL_FLAG_STATUS_SENT)) == 0))
11006                 return (CTL_ACTION_OVERLAP);
11007
11008         /*
11009          * The initiator attempted to send multiple tagged commands with
11010          * the same ID.  (It's fine if different initiators have the same
11011          * tag ID.)
11012          *
11013          * Even if all of those conditions are true, we don't kill the I/O
11014          * if the command ahead of us has been aborted.  We won't end up
11015          * sending it to the FETD, and it's perfectly legal to resend a
11016          * command with the same tag number as long as the previous
11017          * instance of this tag number has been aborted somehow.
11018          */
11019         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11020          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11021          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11022          && ((pending_io->io_hdr.nexus.targ_port ==
11023               ooa_io->io_hdr.nexus.targ_port)
11024           && (pending_io->io_hdr.nexus.initid ==
11025               ooa_io->io_hdr.nexus.initid))
11026          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11027               CTL_FLAG_STATUS_SENT)) == 0))
11028                 return (CTL_ACTION_OVERLAP_TAG);
11029
11030         /*
11031          * If we get a head of queue tag, SAM-3 says that we should
11032          * immediately execute it.
11033          *
11034          * What happens if this command would normally block for some other
11035          * reason?  e.g. a request sense with a head of queue tag
11036          * immediately after a write.  Normally that would block, but this
11037          * will result in its getting executed immediately...
11038          *
11039          * We currently return "pass" instead of "skip", so we'll end up
11040          * going through the rest of the queue to check for overlapped tags.
11041          *
11042          * XXX KDM check for other types of blockage first??
11043          */
11044         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11045                 return (CTL_ACTION_PASS);
11046
11047         /*
11048          * Ordered tags have to block until all items ahead of them
11049          * have completed.  If we get called with an ordered tag, we always
11050          * block, if something else is ahead of us in the queue.
11051          */
11052         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11053                 return (CTL_ACTION_BLOCK);
11054
11055         /*
11056          * Simple tags get blocked until all head of queue and ordered tags
11057          * ahead of them have completed.  I'm lumping untagged commands in
11058          * with simple tags here.  XXX KDM is that the right thing to do?
11059          */
11060         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11061           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11062          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11063           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11064                 return (CTL_ACTION_BLOCK);
11065
11066         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11067         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11068             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11069              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11070              pending_io->scsiio.cdb[1], pending_io));
11071         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11072         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11073                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11074         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11075             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11076              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11077              ooa_io->scsiio.cdb[1], ooa_io));
11078
11079         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11080
11081         switch (serialize_row[pending_entry->seridx]) {
11082         case CTL_SER_BLOCK:
11083                 return (CTL_ACTION_BLOCK);
11084         case CTL_SER_EXTENT:
11085                 return (ctl_extent_check(ooa_io, pending_io,
11086                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11087         case CTL_SER_EXTENTOPT:
11088                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11089                     SCP_QUEUE_ALG_UNRESTRICTED)
11090                         return (ctl_extent_check(ooa_io, pending_io,
11091                             (lun->be_lun &&
11092                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11093                 return (CTL_ACTION_PASS);
11094         case CTL_SER_EXTENTSEQ:
11095                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11096                         return (ctl_extent_check_seq(ooa_io, pending_io));
11097                 return (CTL_ACTION_PASS);
11098         case CTL_SER_PASS:
11099                 return (CTL_ACTION_PASS);
11100         case CTL_SER_BLOCKOPT:
11101                 if ((lun->MODE_CTRL.queue_flags & SCP_QUEUE_ALG_MASK) !=
11102                     SCP_QUEUE_ALG_UNRESTRICTED)
11103                         return (CTL_ACTION_BLOCK);
11104                 return (CTL_ACTION_PASS);
11105         case CTL_SER_SKIP:
11106                 return (CTL_ACTION_SKIP);
11107         default:
11108                 panic("%s: Invalid serialization value %d for %d => %d",
11109                     __func__, serialize_row[pending_entry->seridx],
11110                     pending_entry->seridx, ooa_entry->seridx);
11111         }
11112
11113         return (CTL_ACTION_ERROR);
11114 }
11115
11116 /*
11117  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11118  * Assumptions:
11119  * - pending_io is generally either incoming, or on the blocked queue
11120  * - starting I/O is the I/O we want to start the check with.
11121  */
11122 static ctl_action
11123 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11124               union ctl_io *starting_io)
11125 {
11126         union ctl_io *ooa_io;
11127         ctl_action action;
11128
11129         mtx_assert(&lun->lun_lock, MA_OWNED);
11130
11131         /*
11132          * Run back along the OOA queue, starting with the current
11133          * blocked I/O and going through every I/O before it on the
11134          * queue.  If starting_io is NULL, we'll just end up returning
11135          * CTL_ACTION_PASS.
11136          */
11137         for (ooa_io = starting_io; ooa_io != NULL;
11138              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11139              ooa_links)){
11140
11141                 /*
11142                  * This routine just checks to see whether
11143                  * cur_blocked is blocked by ooa_io, which is ahead
11144                  * of it in the queue.  It doesn't queue/dequeue
11145                  * cur_blocked.
11146                  */
11147                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11148                 switch (action) {
11149                 case CTL_ACTION_BLOCK:
11150                 case CTL_ACTION_OVERLAP:
11151                 case CTL_ACTION_OVERLAP_TAG:
11152                 case CTL_ACTION_SKIP:
11153                 case CTL_ACTION_ERROR:
11154                         return (action);
11155                         break; /* NOTREACHED */
11156                 case CTL_ACTION_PASS:
11157                         break;
11158                 default:
11159                         panic("%s: Invalid action %d\n", __func__, action);
11160                 }
11161         }
11162
11163         return (CTL_ACTION_PASS);
11164 }
11165
11166 /*
11167  * Assumptions:
11168  * - An I/O has just completed, and has been removed from the per-LUN OOA
11169  *   queue, so some items on the blocked queue may now be unblocked.
11170  */
11171 static int
11172 ctl_check_blocked(struct ctl_lun *lun)
11173 {
11174         struct ctl_softc *softc = lun->ctl_softc;
11175         union ctl_io *cur_blocked, *next_blocked;
11176
11177         mtx_assert(&lun->lun_lock, MA_OWNED);
11178
11179         /*
11180          * Run forward from the head of the blocked queue, checking each
11181          * entry against the I/Os prior to it on the OOA queue to see if
11182          * there is still any blockage.
11183          *
11184          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11185          * with our removing a variable on it while it is traversing the
11186          * list.
11187          */
11188         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11189              cur_blocked != NULL; cur_blocked = next_blocked) {
11190                 union ctl_io *prev_ooa;
11191                 ctl_action action;
11192
11193                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11194                                                           blocked_links);
11195
11196                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11197                                                       ctl_ooaq, ooa_links);
11198
11199                 /*
11200                  * If cur_blocked happens to be the first item in the OOA
11201                  * queue now, prev_ooa will be NULL, and the action
11202                  * returned will just be CTL_ACTION_PASS.
11203                  */
11204                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11205
11206                 switch (action) {
11207                 case CTL_ACTION_BLOCK:
11208                         /* Nothing to do here, still blocked */
11209                         break;
11210                 case CTL_ACTION_OVERLAP:
11211                 case CTL_ACTION_OVERLAP_TAG:
11212                         /*
11213                          * This shouldn't happen!  In theory we've already
11214                          * checked this command for overlap...
11215                          */
11216                         break;
11217                 case CTL_ACTION_PASS:
11218                 case CTL_ACTION_SKIP: {
11219                         const struct ctl_cmd_entry *entry;
11220
11221                         /*
11222                          * The skip case shouldn't happen, this transaction
11223                          * should have never made it onto the blocked queue.
11224                          */
11225                         /*
11226                          * This I/O is no longer blocked, we can remove it
11227                          * from the blocked queue.  Since this is a TAILQ
11228                          * (doubly linked list), we can do O(1) removals
11229                          * from any place on the list.
11230                          */
11231                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11232                                      blocked_links);
11233                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11234
11235                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11236                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11237                                 /*
11238                                  * Need to send IO back to original side to
11239                                  * run
11240                                  */
11241                                 union ctl_ha_msg msg_info;
11242
11243                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11244                                 msg_info.hdr.original_sc =
11245                                         cur_blocked->io_hdr.original_sc;
11246                                 msg_info.hdr.serializing_sc = cur_blocked;
11247                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11248                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11249                                     sizeof(msg_info.hdr), M_NOWAIT);
11250                                 break;
11251                         }
11252                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11253
11254                         /*
11255                          * Check this I/O for LUN state changes that may
11256                          * have happened while this command was blocked.
11257                          * The LUN state may have been changed by a command
11258                          * ahead of us in the queue, so we need to re-check
11259                          * for any states that can be caused by SCSI
11260                          * commands.
11261                          */
11262                         if (ctl_scsiio_lun_check(lun, entry,
11263                                                  &cur_blocked->scsiio) == 0) {
11264                                 cur_blocked->io_hdr.flags |=
11265                                                       CTL_FLAG_IS_WAS_ON_RTR;
11266                                 ctl_enqueue_rtr(cur_blocked);
11267                         } else
11268                                 ctl_done(cur_blocked);
11269                         break;
11270                 }
11271                 default:
11272                         /*
11273                          * This probably shouldn't happen -- we shouldn't
11274                          * get CTL_ACTION_ERROR, or anything else.
11275                          */
11276                         break;
11277                 }
11278         }
11279
11280         return (CTL_RETVAL_COMPLETE);
11281 }
11282
11283 /*
11284  * This routine (with one exception) checks LUN flags that can be set by
11285  * commands ahead of us in the OOA queue.  These flags have to be checked
11286  * when a command initially comes in, and when we pull a command off the
11287  * blocked queue and are preparing to execute it.  The reason we have to
11288  * check these flags for commands on the blocked queue is that the LUN
11289  * state may have been changed by a command ahead of us while we're on the
11290  * blocked queue.
11291  *
11292  * Ordering is somewhat important with these checks, so please pay
11293  * careful attention to the placement of any new checks.
11294  */
11295 static int
11296 ctl_scsiio_lun_check(struct ctl_lun *lun,
11297     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11298 {
11299         struct ctl_softc *softc = lun->ctl_softc;
11300         int retval;
11301         uint32_t residx;
11302
11303         retval = 0;
11304
11305         mtx_assert(&lun->lun_lock, MA_OWNED);
11306
11307         /*
11308          * If this shelf is a secondary shelf controller, we may have to
11309          * reject some commands disallowed by HA mode and link state.
11310          */
11311         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11312                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11313                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11314                         ctl_set_lun_unavail(ctsio);
11315                         retval = 1;
11316                         goto bailout;
11317                 }
11318                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11319                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11320                         ctl_set_lun_transit(ctsio);
11321                         retval = 1;
11322                         goto bailout;
11323                 }
11324                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11325                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11326                         ctl_set_lun_standby(ctsio);
11327                         retval = 1;
11328                         goto bailout;
11329                 }
11330
11331                 /* The rest of checks are only done on executing side */
11332                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11333                         goto bailout;
11334         }
11335
11336         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11337                 if (lun->be_lun &&
11338                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11339                         ctl_set_hw_write_protected(ctsio);
11340                         retval = 1;
11341                         goto bailout;
11342                 }
11343                 if ((lun->MODE_CTRL.eca_and_aen & SCP_SWP) != 0) {
11344                         ctl_set_sense(ctsio, /*current_error*/ 1,
11345                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11346                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11347                         retval = 1;
11348                         goto bailout;
11349                 }
11350         }
11351
11352         /*
11353          * Check for a reservation conflict.  If this command isn't allowed
11354          * even on reserved LUNs, and if this initiator isn't the one who
11355          * reserved us, reject the command with a reservation conflict.
11356          */
11357         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11358         if ((lun->flags & CTL_LUN_RESERVED)
11359          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11360                 if (lun->res_idx != residx) {
11361                         ctl_set_reservation_conflict(ctsio);
11362                         retval = 1;
11363                         goto bailout;
11364                 }
11365         }
11366
11367         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11368             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11369                 /* No reservation or command is allowed. */;
11370         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11371             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11372              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11373              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11374                 /* The command is allowed for Write Exclusive resv. */;
11375         } else {
11376                 /*
11377                  * if we aren't registered or it's a res holder type
11378                  * reservation and this isn't the res holder then set a
11379                  * conflict.
11380                  */
11381                 if (ctl_get_prkey(lun, residx) == 0 ||
11382                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11383                         ctl_set_reservation_conflict(ctsio);
11384                         retval = 1;
11385                         goto bailout;
11386                 }
11387         }
11388
11389         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11390                 if (lun->flags & CTL_LUN_EJECTED)
11391                         ctl_set_lun_ejected(ctsio);
11392                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11393                         if (lun->flags & CTL_LUN_REMOVABLE)
11394                                 ctl_set_lun_no_media(ctsio);
11395                         else
11396                                 ctl_set_lun_int_reqd(ctsio);
11397                 } else if (lun->flags & CTL_LUN_STOPPED)
11398                         ctl_set_lun_stopped(ctsio);
11399                 else
11400                         goto bailout;
11401                 retval = 1;
11402                 goto bailout;
11403         }
11404
11405 bailout:
11406         return (retval);
11407 }
11408
11409 static void
11410 ctl_failover_io(union ctl_io *io, int have_lock)
11411 {
11412         ctl_set_busy(&io->scsiio);
11413         ctl_done(io);
11414 }
11415
11416 static void
11417 ctl_failover_lun(union ctl_io *rio)
11418 {
11419         struct ctl_softc *softc = control_softc;
11420         struct ctl_lun *lun;
11421         struct ctl_io_hdr *io, *next_io;
11422         uint32_t targ_lun;
11423
11424         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11425         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11426
11427         /* Find and lock the LUN. */
11428         mtx_lock(&softc->ctl_lock);
11429         if ((targ_lun < CTL_MAX_LUNS) &&
11430             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11431                 mtx_lock(&lun->lun_lock);
11432                 mtx_unlock(&softc->ctl_lock);
11433                 if (lun->flags & CTL_LUN_DISABLED) {
11434                         mtx_unlock(&lun->lun_lock);
11435                         return;
11436                 }
11437         } else {
11438                 mtx_unlock(&softc->ctl_lock);
11439                 return;
11440         }
11441
11442         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11443                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11444                         /* We are master */
11445                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11446                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11447                                         io->flags |= CTL_FLAG_ABORT;
11448                                         io->flags |= CTL_FLAG_FAILOVER;
11449                                 } else { /* This can be only due to DATAMOVE */
11450                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11451                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11452                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11453                                         io->port_status = 31340;
11454                                         ctl_enqueue_isc((union ctl_io *)io);
11455                                 }
11456                         }
11457                         /* We are slave */
11458                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11459                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11460                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11461                                         io->flags |= CTL_FLAG_FAILOVER;
11462                                 } else {
11463                                         ctl_set_busy(&((union ctl_io *)io)->
11464                                             scsiio);
11465                                         ctl_done((union ctl_io *)io);
11466                                 }
11467                         }
11468                 }
11469         } else { /* SERIALIZE modes */
11470                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11471                     next_io) {
11472                         /* We are master */
11473                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11474                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11475                                     blocked_links);
11476                                 io->flags &= ~CTL_FLAG_BLOCKED;
11477                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11478                                 ctl_free_io((union ctl_io *)io);
11479                         }
11480                 }
11481                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11482                         /* We are master */
11483                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11484                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11485                                 ctl_free_io((union ctl_io *)io);
11486                         }
11487                         /* We are slave */
11488                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11489                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11490                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11491                                         ctl_set_busy(&((union ctl_io *)io)->
11492                                             scsiio);
11493                                         ctl_done((union ctl_io *)io);
11494                                 }
11495                         }
11496                 }
11497                 ctl_check_blocked(lun);
11498         }
11499         mtx_unlock(&lun->lun_lock);
11500 }
11501
11502 static int
11503 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11504 {
11505         struct ctl_lun *lun;
11506         const struct ctl_cmd_entry *entry;
11507         uint32_t initidx, targ_lun;
11508         int retval;
11509
11510         retval = 0;
11511
11512         lun = NULL;
11513
11514         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11515         if ((targ_lun < CTL_MAX_LUNS)
11516          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11517                 /*
11518                  * If the LUN is invalid, pretend that it doesn't exist.
11519                  * It will go away as soon as all pending I/O has been
11520                  * completed.
11521                  */
11522                 mtx_lock(&lun->lun_lock);
11523                 if (lun->flags & CTL_LUN_DISABLED) {
11524                         mtx_unlock(&lun->lun_lock);
11525                         lun = NULL;
11526                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11527                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11528                 } else {
11529                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11530                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11531                                 lun->be_lun;
11532
11533                         /*
11534                          * Every I/O goes into the OOA queue for a
11535                          * particular LUN, and stays there until completion.
11536                          */
11537 #ifdef CTL_TIME_IO
11538                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11539                                 lun->idle_time += getsbinuptime() -
11540                                     lun->last_busy;
11541                         }
11542 #endif
11543                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11544                             ooa_links);
11545                 }
11546         } else {
11547                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11548                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11549         }
11550
11551         /* Get command entry and return error if it is unsuppotyed. */
11552         entry = ctl_validate_command(ctsio);
11553         if (entry == NULL) {
11554                 if (lun)
11555                         mtx_unlock(&lun->lun_lock);
11556                 return (retval);
11557         }
11558
11559         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11560         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11561
11562         /*
11563          * Check to see whether we can send this command to LUNs that don't
11564          * exist.  This should pretty much only be the case for inquiry
11565          * and request sense.  Further checks, below, really require having
11566          * a LUN, so we can't really check the command anymore.  Just put
11567          * it on the rtr queue.
11568          */
11569         if (lun == NULL) {
11570                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11571                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11572                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11573                         return (retval);
11574                 }
11575
11576                 ctl_set_unsupported_lun(ctsio);
11577                 ctl_done((union ctl_io *)ctsio);
11578                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11579                 return (retval);
11580         } else {
11581                 /*
11582                  * Make sure we support this particular command on this LUN.
11583                  * e.g., we don't support writes to the control LUN.
11584                  */
11585                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11586                         mtx_unlock(&lun->lun_lock);
11587                         ctl_set_invalid_opcode(ctsio);
11588                         ctl_done((union ctl_io *)ctsio);
11589                         return (retval);
11590                 }
11591         }
11592
11593         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11594
11595 #ifdef CTL_WITH_CA
11596         /*
11597          * If we've got a request sense, it'll clear the contingent
11598          * allegiance condition.  Otherwise, if we have a CA condition for
11599          * this initiator, clear it, because it sent down a command other
11600          * than request sense.
11601          */
11602         if ((ctsio->cdb[0] != REQUEST_SENSE)
11603          && (ctl_is_set(lun->have_ca, initidx)))
11604                 ctl_clear_mask(lun->have_ca, initidx);
11605 #endif
11606
11607         /*
11608          * If the command has this flag set, it handles its own unit
11609          * attention reporting, we shouldn't do anything.  Otherwise we
11610          * check for any pending unit attentions, and send them back to the
11611          * initiator.  We only do this when a command initially comes in,
11612          * not when we pull it off the blocked queue.
11613          *
11614          * According to SAM-3, section 5.3.2, the order that things get
11615          * presented back to the host is basically unit attentions caused
11616          * by some sort of reset event, busy status, reservation conflicts
11617          * or task set full, and finally any other status.
11618          *
11619          * One issue here is that some of the unit attentions we report
11620          * don't fall into the "reset" category (e.g. "reported luns data
11621          * has changed").  So reporting it here, before the reservation
11622          * check, may be technically wrong.  I guess the only thing to do
11623          * would be to check for and report the reset events here, and then
11624          * check for the other unit attention types after we check for a
11625          * reservation conflict.
11626          *
11627          * XXX KDM need to fix this
11628          */
11629         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11630                 ctl_ua_type ua_type;
11631                 u_int sense_len = 0;
11632
11633                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11634                     &sense_len, SSD_TYPE_NONE);
11635                 if (ua_type != CTL_UA_NONE) {
11636                         mtx_unlock(&lun->lun_lock);
11637                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11638                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11639                         ctsio->sense_len = sense_len;
11640                         ctl_done((union ctl_io *)ctsio);
11641                         return (retval);
11642                 }
11643         }
11644
11645
11646         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11647                 mtx_unlock(&lun->lun_lock);
11648                 ctl_done((union ctl_io *)ctsio);
11649                 return (retval);
11650         }
11651
11652         /*
11653          * XXX CHD this is where we want to send IO to other side if
11654          * this LUN is secondary on this SC. We will need to make a copy
11655          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11656          * the copy we send as FROM_OTHER.
11657          * We also need to stuff the address of the original IO so we can
11658          * find it easily. Something similar will need be done on the other
11659          * side so when we are done we can find the copy.
11660          */
11661         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11662             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11663             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11664                 union ctl_ha_msg msg_info;
11665                 int isc_retval;
11666
11667                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11668                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11669                 mtx_unlock(&lun->lun_lock);
11670
11671                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11672                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11673                 msg_info.hdr.serializing_sc = NULL;
11674                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11675                 msg_info.scsi.tag_num = ctsio->tag_num;
11676                 msg_info.scsi.tag_type = ctsio->tag_type;
11677                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11678                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11679
11680                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11681                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11682                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11683                         ctl_set_busy(ctsio);
11684                         ctl_done((union ctl_io *)ctsio);
11685                         return (retval);
11686                 }
11687                 return (retval);
11688         }
11689
11690         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11691                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11692                               ctl_ooaq, ooa_links))) {
11693         case CTL_ACTION_BLOCK:
11694                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11695                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11696                                   blocked_links);
11697                 mtx_unlock(&lun->lun_lock);
11698                 return (retval);
11699         case CTL_ACTION_PASS:
11700         case CTL_ACTION_SKIP:
11701                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11702                 mtx_unlock(&lun->lun_lock);
11703                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11704                 break;
11705         case CTL_ACTION_OVERLAP:
11706                 mtx_unlock(&lun->lun_lock);
11707                 ctl_set_overlapped_cmd(ctsio);
11708                 ctl_done((union ctl_io *)ctsio);
11709                 break;
11710         case CTL_ACTION_OVERLAP_TAG:
11711                 mtx_unlock(&lun->lun_lock);
11712                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11713                 ctl_done((union ctl_io *)ctsio);
11714                 break;
11715         case CTL_ACTION_ERROR:
11716         default:
11717                 mtx_unlock(&lun->lun_lock);
11718                 ctl_set_internal_failure(ctsio,
11719                                          /*sks_valid*/ 0,
11720                                          /*retry_count*/ 0);
11721                 ctl_done((union ctl_io *)ctsio);
11722                 break;
11723         }
11724         return (retval);
11725 }
11726
11727 const struct ctl_cmd_entry *
11728 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11729 {
11730         const struct ctl_cmd_entry *entry;
11731         int service_action;
11732
11733         entry = &ctl_cmd_table[ctsio->cdb[0]];
11734         if (sa)
11735                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11736         if (entry->flags & CTL_CMD_FLAG_SA5) {
11737                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11738                 entry = &((const struct ctl_cmd_entry *)
11739                     entry->execute)[service_action];
11740         }
11741         return (entry);
11742 }
11743
11744 const struct ctl_cmd_entry *
11745 ctl_validate_command(struct ctl_scsiio *ctsio)
11746 {
11747         const struct ctl_cmd_entry *entry;
11748         int i, sa;
11749         uint8_t diff;
11750
11751         entry = ctl_get_cmd_entry(ctsio, &sa);
11752         if (entry->execute == NULL) {
11753                 if (sa)
11754                         ctl_set_invalid_field(ctsio,
11755                                               /*sks_valid*/ 1,
11756                                               /*command*/ 1,
11757                                               /*field*/ 1,
11758                                               /*bit_valid*/ 1,
11759                                               /*bit*/ 4);
11760                 else
11761                         ctl_set_invalid_opcode(ctsio);
11762                 ctl_done((union ctl_io *)ctsio);
11763                 return (NULL);
11764         }
11765         KASSERT(entry->length > 0,
11766             ("Not defined length for command 0x%02x/0x%02x",
11767              ctsio->cdb[0], ctsio->cdb[1]));
11768         for (i = 1; i < entry->length; i++) {
11769                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11770                 if (diff == 0)
11771                         continue;
11772                 ctl_set_invalid_field(ctsio,
11773                                       /*sks_valid*/ 1,
11774                                       /*command*/ 1,
11775                                       /*field*/ i,
11776                                       /*bit_valid*/ 1,
11777                                       /*bit*/ fls(diff) - 1);
11778                 ctl_done((union ctl_io *)ctsio);
11779                 return (NULL);
11780         }
11781         return (entry);
11782 }
11783
11784 static int
11785 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11786 {
11787
11788         switch (lun_type) {
11789         case T_DIRECT:
11790                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11791                         return (0);
11792                 break;
11793         case T_PROCESSOR:
11794                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11795                         return (0);
11796                 break;
11797         case T_CDROM:
11798                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11799                         return (0);
11800                 break;
11801         default:
11802                 return (0);
11803         }
11804         return (1);
11805 }
11806
11807 static int
11808 ctl_scsiio(struct ctl_scsiio *ctsio)
11809 {
11810         int retval;
11811         const struct ctl_cmd_entry *entry;
11812
11813         retval = CTL_RETVAL_COMPLETE;
11814
11815         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11816
11817         entry = ctl_get_cmd_entry(ctsio, NULL);
11818
11819         /*
11820          * If this I/O has been aborted, just send it straight to
11821          * ctl_done() without executing it.
11822          */
11823         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11824                 ctl_done((union ctl_io *)ctsio);
11825                 goto bailout;
11826         }
11827
11828         /*
11829          * All the checks should have been handled by ctl_scsiio_precheck().
11830          * We should be clear now to just execute the I/O.
11831          */
11832         retval = entry->execute(ctsio);
11833
11834 bailout:
11835         return (retval);
11836 }
11837
11838 /*
11839  * Since we only implement one target right now, a bus reset simply resets
11840  * our single target.
11841  */
11842 static int
11843 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11844 {
11845         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11846 }
11847
11848 static int
11849 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11850                  ctl_ua_type ua_type)
11851 {
11852         struct ctl_port *port;
11853         struct ctl_lun *lun;
11854         int retval;
11855
11856         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11857                 union ctl_ha_msg msg_info;
11858
11859                 msg_info.hdr.nexus = io->io_hdr.nexus;
11860                 if (ua_type==CTL_UA_TARG_RESET)
11861                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11862                 else
11863                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11864                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11865                 msg_info.hdr.original_sc = NULL;
11866                 msg_info.hdr.serializing_sc = NULL;
11867                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11868                     sizeof(msg_info.task), M_WAITOK);
11869         }
11870         retval = 0;
11871
11872         mtx_lock(&softc->ctl_lock);
11873         port = ctl_io_port(&io->io_hdr);
11874         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11875                 if (port != NULL &&
11876                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11877                         continue;
11878                 retval += ctl_do_lun_reset(lun, io, ua_type);
11879         }
11880         mtx_unlock(&softc->ctl_lock);
11881         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11882         return (retval);
11883 }
11884
11885 /*
11886  * The LUN should always be set.  The I/O is optional, and is used to
11887  * distinguish between I/Os sent by this initiator, and by other
11888  * initiators.  We set unit attention for initiators other than this one.
11889  * SAM-3 is vague on this point.  It does say that a unit attention should
11890  * be established for other initiators when a LUN is reset (see section
11891  * 5.7.3), but it doesn't specifically say that the unit attention should
11892  * be established for this particular initiator when a LUN is reset.  Here
11893  * is the relevant text, from SAM-3 rev 8:
11894  *
11895  * 5.7.2 When a SCSI initiator port aborts its own tasks
11896  *
11897  * When a SCSI initiator port causes its own task(s) to be aborted, no
11898  * notification that the task(s) have been aborted shall be returned to
11899  * the SCSI initiator port other than the completion response for the
11900  * command or task management function action that caused the task(s) to
11901  * be aborted and notification(s) associated with related effects of the
11902  * action (e.g., a reset unit attention condition).
11903  *
11904  * XXX KDM for now, we're setting unit attention for all initiators.
11905  */
11906 static int
11907 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11908 {
11909         union ctl_io *xio;
11910 #if 0
11911         uint32_t initidx;
11912 #endif
11913         int i;
11914
11915         mtx_lock(&lun->lun_lock);
11916         /*
11917          * Run through the OOA queue and abort each I/O.
11918          */
11919         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11920              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11921                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11922         }
11923
11924         /*
11925          * This version sets unit attention for every
11926          */
11927 #if 0
11928         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11929         ctl_est_ua_all(lun, initidx, ua_type);
11930 #else
11931         ctl_est_ua_all(lun, -1, ua_type);
11932 #endif
11933
11934         /*
11935          * A reset (any kind, really) clears reservations established with
11936          * RESERVE/RELEASE.  It does not clear reservations established
11937          * with PERSISTENT RESERVE OUT, but we don't support that at the
11938          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11939          * reservations made with the RESERVE/RELEASE commands, because
11940          * those commands are obsolete in SPC-3.
11941          */
11942         lun->flags &= ~CTL_LUN_RESERVED;
11943
11944 #ifdef CTL_WITH_CA
11945         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11946                 ctl_clear_mask(lun->have_ca, i);
11947 #endif
11948         lun->prevent_count = 0;
11949         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11950                 ctl_clear_mask(lun->prevent, i);
11951         mtx_unlock(&lun->lun_lock);
11952
11953         return (0);
11954 }
11955
11956 static int
11957 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11958 {
11959         struct ctl_lun *lun;
11960         uint32_t targ_lun;
11961         int retval;
11962
11963         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11964         mtx_lock(&softc->ctl_lock);
11965         if ((targ_lun >= CTL_MAX_LUNS) ||
11966             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11967                 mtx_unlock(&softc->ctl_lock);
11968                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11969                 return (1);
11970         }
11971         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11972         mtx_unlock(&softc->ctl_lock);
11973         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11974
11975         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11976                 union ctl_ha_msg msg_info;
11977
11978                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11979                 msg_info.hdr.nexus = io->io_hdr.nexus;
11980                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11981                 msg_info.hdr.original_sc = NULL;
11982                 msg_info.hdr.serializing_sc = NULL;
11983                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11984                     sizeof(msg_info.task), M_WAITOK);
11985         }
11986         return (retval);
11987 }
11988
11989 static void
11990 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11991     int other_sc)
11992 {
11993         union ctl_io *xio;
11994
11995         mtx_assert(&lun->lun_lock, MA_OWNED);
11996
11997         /*
11998          * Run through the OOA queue and attempt to find the given I/O.
11999          * The target port, initiator ID, tag type and tag number have to
12000          * match the values that we got from the initiator.  If we have an
12001          * untagged command to abort, simply abort the first untagged command
12002          * we come to.  We only allow one untagged command at a time of course.
12003          */
12004         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12005              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12006
12007                 if ((targ_port == UINT32_MAX ||
12008                      targ_port == xio->io_hdr.nexus.targ_port) &&
12009                     (init_id == UINT32_MAX ||
12010                      init_id == xio->io_hdr.nexus.initid)) {
12011                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12012                             init_id != xio->io_hdr.nexus.initid)
12013                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12014                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12015                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12016                                 union ctl_ha_msg msg_info;
12017
12018                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12019                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12020                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12021                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12022                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12023                                 msg_info.hdr.original_sc = NULL;
12024                                 msg_info.hdr.serializing_sc = NULL;
12025                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12026                                     sizeof(msg_info.task), M_NOWAIT);
12027                         }
12028                 }
12029         }
12030 }
12031
12032 static int
12033 ctl_abort_task_set(union ctl_io *io)
12034 {
12035         struct ctl_softc *softc = control_softc;
12036         struct ctl_lun *lun;
12037         uint32_t targ_lun;
12038
12039         /*
12040          * Look up the LUN.
12041          */
12042         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12043         mtx_lock(&softc->ctl_lock);
12044         if ((targ_lun >= CTL_MAX_LUNS) ||
12045             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12046                 mtx_unlock(&softc->ctl_lock);
12047                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12048                 return (1);
12049         }
12050
12051         mtx_lock(&lun->lun_lock);
12052         mtx_unlock(&softc->ctl_lock);
12053         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12054                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12055                     io->io_hdr.nexus.initid,
12056                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12057         } else { /* CTL_TASK_CLEAR_TASK_SET */
12058                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12059                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12060         }
12061         mtx_unlock(&lun->lun_lock);
12062         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12063         return (0);
12064 }
12065
12066 static int
12067 ctl_i_t_nexus_reset(union ctl_io *io)
12068 {
12069         struct ctl_softc *softc = control_softc;
12070         struct ctl_lun *lun;
12071         uint32_t initidx;
12072
12073         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12074                 union ctl_ha_msg msg_info;
12075
12076                 msg_info.hdr.nexus = io->io_hdr.nexus;
12077                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12078                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12079                 msg_info.hdr.original_sc = NULL;
12080                 msg_info.hdr.serializing_sc = NULL;
12081                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12082                     sizeof(msg_info.task), M_WAITOK);
12083         }
12084
12085         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12086         mtx_lock(&softc->ctl_lock);
12087         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12088                 mtx_lock(&lun->lun_lock);
12089                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12090                     io->io_hdr.nexus.initid, 1);
12091 #ifdef CTL_WITH_CA
12092                 ctl_clear_mask(lun->have_ca, initidx);
12093 #endif
12094                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12095                         lun->flags &= ~CTL_LUN_RESERVED;
12096                 if (ctl_is_set(lun->prevent, initidx)) {
12097                         ctl_clear_mask(lun->prevent, initidx);
12098                         lun->prevent_count--;
12099                 }
12100                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12101                 mtx_unlock(&lun->lun_lock);
12102         }
12103         mtx_unlock(&softc->ctl_lock);
12104         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12105         return (0);
12106 }
12107
12108 static int
12109 ctl_abort_task(union ctl_io *io)
12110 {
12111         union ctl_io *xio;
12112         struct ctl_lun *lun;
12113         struct ctl_softc *softc;
12114 #if 0
12115         struct sbuf sb;
12116         char printbuf[128];
12117 #endif
12118         int found;
12119         uint32_t targ_lun;
12120
12121         softc = control_softc;
12122         found = 0;
12123
12124         /*
12125          * Look up the LUN.
12126          */
12127         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12128         mtx_lock(&softc->ctl_lock);
12129         if ((targ_lun >= CTL_MAX_LUNS) ||
12130             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12131                 mtx_unlock(&softc->ctl_lock);
12132                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12133                 return (1);
12134         }
12135
12136 #if 0
12137         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12138                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12139 #endif
12140
12141         mtx_lock(&lun->lun_lock);
12142         mtx_unlock(&softc->ctl_lock);
12143         /*
12144          * Run through the OOA queue and attempt to find the given I/O.
12145          * The target port, initiator ID, tag type and tag number have to
12146          * match the values that we got from the initiator.  If we have an
12147          * untagged command to abort, simply abort the first untagged command
12148          * we come to.  We only allow one untagged command at a time of course.
12149          */
12150         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12151              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12152 #if 0
12153                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12154
12155                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12156                             lun->lun, xio->scsiio.tag_num,
12157                             xio->scsiio.tag_type,
12158                             (xio->io_hdr.blocked_links.tqe_prev
12159                             == NULL) ? "" : " BLOCKED",
12160                             (xio->io_hdr.flags &
12161                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12162                             (xio->io_hdr.flags &
12163                             CTL_FLAG_ABORT) ? " ABORT" : "",
12164                             (xio->io_hdr.flags &
12165                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12166                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12167                 sbuf_finish(&sb);
12168                 printf("%s\n", sbuf_data(&sb));
12169 #endif
12170
12171                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12172                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12173                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12174                         continue;
12175
12176                 /*
12177                  * If the abort says that the task is untagged, the
12178                  * task in the queue must be untagged.  Otherwise,
12179                  * we just check to see whether the tag numbers
12180                  * match.  This is because the QLogic firmware
12181                  * doesn't pass back the tag type in an abort
12182                  * request.
12183                  */
12184 #if 0
12185                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12186                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12187                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12188 #endif
12189                 /*
12190                  * XXX KDM we've got problems with FC, because it
12191                  * doesn't send down a tag type with aborts.  So we
12192                  * can only really go by the tag number...
12193                  * This may cause problems with parallel SCSI.
12194                  * Need to figure that out!!
12195                  */
12196                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12197                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12198                         found = 1;
12199                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12200                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12201                                 union ctl_ha_msg msg_info;
12202
12203                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12204                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12205                                 msg_info.task.tag_num = io->taskio.tag_num;
12206                                 msg_info.task.tag_type = io->taskio.tag_type;
12207                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12208                                 msg_info.hdr.original_sc = NULL;
12209                                 msg_info.hdr.serializing_sc = NULL;
12210 #if 0
12211                                 printf("Sent Abort to other side\n");
12212 #endif
12213                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12214                                     sizeof(msg_info.task), M_NOWAIT);
12215                         }
12216 #if 0
12217                         printf("ctl_abort_task: found I/O to abort\n");
12218 #endif
12219                 }
12220         }
12221         mtx_unlock(&lun->lun_lock);
12222
12223         if (found == 0) {
12224                 /*
12225                  * This isn't really an error.  It's entirely possible for
12226                  * the abort and command completion to cross on the wire.
12227                  * This is more of an informative/diagnostic error.
12228                  */
12229 #if 0
12230                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12231                        "%u:%u:%u tag %d type %d\n",
12232                        io->io_hdr.nexus.initid,
12233                        io->io_hdr.nexus.targ_port,
12234                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12235                        io->taskio.tag_type);
12236 #endif
12237         }
12238         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12239         return (0);
12240 }
12241
12242 static int
12243 ctl_query_task(union ctl_io *io, int task_set)
12244 {
12245         union ctl_io *xio;
12246         struct ctl_lun *lun;
12247         struct ctl_softc *softc;
12248         int found = 0;
12249         uint32_t targ_lun;
12250
12251         softc = control_softc;
12252         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12253         mtx_lock(&softc->ctl_lock);
12254         if ((targ_lun >= CTL_MAX_LUNS) ||
12255             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12256                 mtx_unlock(&softc->ctl_lock);
12257                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12258                 return (1);
12259         }
12260         mtx_lock(&lun->lun_lock);
12261         mtx_unlock(&softc->ctl_lock);
12262         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12263              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12264
12265                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12266                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12267                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12268                         continue;
12269
12270                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12271                         found = 1;
12272                         break;
12273                 }
12274         }
12275         mtx_unlock(&lun->lun_lock);
12276         if (found)
12277                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12278         else
12279                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12280         return (0);
12281 }
12282
12283 static int
12284 ctl_query_async_event(union ctl_io *io)
12285 {
12286         struct ctl_lun *lun;
12287         struct ctl_softc *softc;
12288         ctl_ua_type ua;
12289         uint32_t targ_lun, initidx;
12290
12291         softc = control_softc;
12292         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12293         mtx_lock(&softc->ctl_lock);
12294         if ((targ_lun >= CTL_MAX_LUNS) ||
12295             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12296                 mtx_unlock(&softc->ctl_lock);
12297                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12298                 return (1);
12299         }
12300         mtx_lock(&lun->lun_lock);
12301         mtx_unlock(&softc->ctl_lock);
12302         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12303         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12304         mtx_unlock(&lun->lun_lock);
12305         if (ua != CTL_UA_NONE)
12306                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12307         else
12308                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12309         return (0);
12310 }
12311
12312 static void
12313 ctl_run_task(union ctl_io *io)
12314 {
12315         struct ctl_softc *softc = control_softc;
12316         int retval = 1;
12317
12318         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12319         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12320             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12321         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12322         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12323         switch (io->taskio.task_action) {
12324         case CTL_TASK_ABORT_TASK:
12325                 retval = ctl_abort_task(io);
12326                 break;
12327         case CTL_TASK_ABORT_TASK_SET:
12328         case CTL_TASK_CLEAR_TASK_SET:
12329                 retval = ctl_abort_task_set(io);
12330                 break;
12331         case CTL_TASK_CLEAR_ACA:
12332                 break;
12333         case CTL_TASK_I_T_NEXUS_RESET:
12334                 retval = ctl_i_t_nexus_reset(io);
12335                 break;
12336         case CTL_TASK_LUN_RESET:
12337                 retval = ctl_lun_reset(softc, io);
12338                 break;
12339         case CTL_TASK_TARGET_RESET:
12340                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12341                 break;
12342         case CTL_TASK_BUS_RESET:
12343                 retval = ctl_bus_reset(softc, io);
12344                 break;
12345         case CTL_TASK_PORT_LOGIN:
12346                 break;
12347         case CTL_TASK_PORT_LOGOUT:
12348                 break;
12349         case CTL_TASK_QUERY_TASK:
12350                 retval = ctl_query_task(io, 0);
12351                 break;
12352         case CTL_TASK_QUERY_TASK_SET:
12353                 retval = ctl_query_task(io, 1);
12354                 break;
12355         case CTL_TASK_QUERY_ASYNC_EVENT:
12356                 retval = ctl_query_async_event(io);
12357                 break;
12358         default:
12359                 printf("%s: got unknown task management event %d\n",
12360                        __func__, io->taskio.task_action);
12361                 break;
12362         }
12363         if (retval == 0)
12364                 io->io_hdr.status = CTL_SUCCESS;
12365         else
12366                 io->io_hdr.status = CTL_ERROR;
12367         ctl_done(io);
12368 }
12369
12370 /*
12371  * For HA operation.  Handle commands that come in from the other
12372  * controller.
12373  */
12374 static void
12375 ctl_handle_isc(union ctl_io *io)
12376 {
12377         int free_io;
12378         struct ctl_lun *lun;
12379         struct ctl_softc *softc = control_softc;
12380         uint32_t targ_lun;
12381
12382         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12383         lun = softc->ctl_luns[targ_lun];
12384
12385         switch (io->io_hdr.msg_type) {
12386         case CTL_MSG_SERIALIZE:
12387                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12388                 break;
12389         case CTL_MSG_R2R: {
12390                 const struct ctl_cmd_entry *entry;
12391
12392                 /*
12393                  * This is only used in SER_ONLY mode.
12394                  */
12395                 free_io = 0;
12396                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12397                 mtx_lock(&lun->lun_lock);
12398                 if (ctl_scsiio_lun_check(lun,
12399                     entry, (struct ctl_scsiio *)io) != 0) {
12400                         mtx_unlock(&lun->lun_lock);
12401                         ctl_done(io);
12402                         break;
12403                 }
12404                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12405                 mtx_unlock(&lun->lun_lock);
12406                 ctl_enqueue_rtr(io);
12407                 break;
12408         }
12409         case CTL_MSG_FINISH_IO:
12410                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12411                         free_io = 0;
12412                         ctl_done(io);
12413                 } else {
12414                         free_io = 1;
12415                         mtx_lock(&lun->lun_lock);
12416                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12417                                      ooa_links);
12418                         ctl_check_blocked(lun);
12419                         mtx_unlock(&lun->lun_lock);
12420                 }
12421                 break;
12422         case CTL_MSG_PERS_ACTION:
12423                 ctl_hndl_per_res_out_on_other_sc(
12424                         (union ctl_ha_msg *)&io->presio.pr_msg);
12425                 free_io = 1;
12426                 break;
12427         case CTL_MSG_BAD_JUJU:
12428                 free_io = 0;
12429                 ctl_done(io);
12430                 break;
12431         case CTL_MSG_DATAMOVE:
12432                 /* Only used in XFER mode */
12433                 free_io = 0;
12434                 ctl_datamove_remote(io);
12435                 break;
12436         case CTL_MSG_DATAMOVE_DONE:
12437                 /* Only used in XFER mode */
12438                 free_io = 0;
12439                 io->scsiio.be_move_done(io);
12440                 break;
12441         case CTL_MSG_FAILOVER:
12442                 ctl_failover_lun(io);
12443                 free_io = 1;
12444                 break;
12445         default:
12446                 free_io = 1;
12447                 printf("%s: Invalid message type %d\n",
12448                        __func__, io->io_hdr.msg_type);
12449                 break;
12450         }
12451         if (free_io)
12452                 ctl_free_io(io);
12453
12454 }
12455
12456
12457 /*
12458  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12459  * there is no match.
12460  */
12461 static ctl_lun_error_pattern
12462 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12463 {
12464         const struct ctl_cmd_entry *entry;
12465         ctl_lun_error_pattern filtered_pattern, pattern;
12466
12467         pattern = desc->error_pattern;
12468
12469         /*
12470          * XXX KDM we need more data passed into this function to match a
12471          * custom pattern, and we actually need to implement custom pattern
12472          * matching.
12473          */
12474         if (pattern & CTL_LUN_PAT_CMD)
12475                 return (CTL_LUN_PAT_CMD);
12476
12477         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12478                 return (CTL_LUN_PAT_ANY);
12479
12480         entry = ctl_get_cmd_entry(ctsio, NULL);
12481
12482         filtered_pattern = entry->pattern & pattern;
12483
12484         /*
12485          * If the user requested specific flags in the pattern (e.g.
12486          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12487          * flags.
12488          *
12489          * If the user did not specify any flags, it doesn't matter whether
12490          * or not the command supports the flags.
12491          */
12492         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12493              (pattern & ~CTL_LUN_PAT_MASK))
12494                 return (CTL_LUN_PAT_NONE);
12495
12496         /*
12497          * If the user asked for a range check, see if the requested LBA
12498          * range overlaps with this command's LBA range.
12499          */
12500         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12501                 uint64_t lba1;
12502                 uint64_t len1;
12503                 ctl_action action;
12504                 int retval;
12505
12506                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12507                 if (retval != 0)
12508                         return (CTL_LUN_PAT_NONE);
12509
12510                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12511                                               desc->lba_range.len, FALSE);
12512                 /*
12513                  * A "pass" means that the LBA ranges don't overlap, so
12514                  * this doesn't match the user's range criteria.
12515                  */
12516                 if (action == CTL_ACTION_PASS)
12517                         return (CTL_LUN_PAT_NONE);
12518         }
12519
12520         return (filtered_pattern);
12521 }
12522
12523 static void
12524 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12525 {
12526         struct ctl_error_desc *desc, *desc2;
12527
12528         mtx_assert(&lun->lun_lock, MA_OWNED);
12529
12530         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12531                 ctl_lun_error_pattern pattern;
12532                 /*
12533                  * Check to see whether this particular command matches
12534                  * the pattern in the descriptor.
12535                  */
12536                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12537                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12538                         continue;
12539
12540                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12541                 case CTL_LUN_INJ_ABORTED:
12542                         ctl_set_aborted(&io->scsiio);
12543                         break;
12544                 case CTL_LUN_INJ_MEDIUM_ERR:
12545                         ctl_set_medium_error(&io->scsiio,
12546                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12547                              CTL_FLAG_DATA_OUT);
12548                         break;
12549                 case CTL_LUN_INJ_UA:
12550                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12551                          * OCCURRED */
12552                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12553                         break;
12554                 case CTL_LUN_INJ_CUSTOM:
12555                         /*
12556                          * We're assuming the user knows what he is doing.
12557                          * Just copy the sense information without doing
12558                          * checks.
12559                          */
12560                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12561                               MIN(sizeof(desc->custom_sense),
12562                                   sizeof(io->scsiio.sense_data)));
12563                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12564                         io->scsiio.sense_len = SSD_FULL_SIZE;
12565                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12566                         break;
12567                 case CTL_LUN_INJ_NONE:
12568                 default:
12569                         /*
12570                          * If this is an error injection type we don't know
12571                          * about, clear the continuous flag (if it is set)
12572                          * so it will get deleted below.
12573                          */
12574                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12575                         break;
12576                 }
12577                 /*
12578                  * By default, each error injection action is a one-shot
12579                  */
12580                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12581                         continue;
12582
12583                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12584
12585                 free(desc, M_CTL);
12586         }
12587 }
12588
12589 #ifdef CTL_IO_DELAY
12590 static void
12591 ctl_datamove_timer_wakeup(void *arg)
12592 {
12593         union ctl_io *io;
12594
12595         io = (union ctl_io *)arg;
12596
12597         ctl_datamove(io);
12598 }
12599 #endif /* CTL_IO_DELAY */
12600
12601 void
12602 ctl_datamove(union ctl_io *io)
12603 {
12604         struct ctl_lun *lun;
12605         void (*fe_datamove)(union ctl_io *io);
12606
12607         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12608
12609         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12610
12611         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12612 #ifdef CTL_TIME_IO
12613         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12614                 char str[256];
12615                 char path_str[64];
12616                 struct sbuf sb;
12617
12618                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12619                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12620
12621                 sbuf_cat(&sb, path_str);
12622                 switch (io->io_hdr.io_type) {
12623                 case CTL_IO_SCSI:
12624                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12625                         sbuf_printf(&sb, "\n");
12626                         sbuf_cat(&sb, path_str);
12627                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12628                                     io->scsiio.tag_num, io->scsiio.tag_type);
12629                         break;
12630                 case CTL_IO_TASK:
12631                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12632                                     "Tag Type: %d\n", io->taskio.task_action,
12633                                     io->taskio.tag_num, io->taskio.tag_type);
12634                         break;
12635                 default:
12636                         panic("%s: Invalid CTL I/O type %d\n",
12637                             __func__, io->io_hdr.io_type);
12638                 }
12639                 sbuf_cat(&sb, path_str);
12640                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12641                             (intmax_t)time_uptime - io->io_hdr.start_time);
12642                 sbuf_finish(&sb);
12643                 printf("%s", sbuf_data(&sb));
12644         }
12645 #endif /* CTL_TIME_IO */
12646
12647 #ifdef CTL_IO_DELAY
12648         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12649                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12650         } else {
12651                 if ((lun != NULL)
12652                  && (lun->delay_info.datamove_delay > 0)) {
12653
12654                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12655                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12656                         callout_reset(&io->io_hdr.delay_callout,
12657                                       lun->delay_info.datamove_delay * hz,
12658                                       ctl_datamove_timer_wakeup, io);
12659                         if (lun->delay_info.datamove_type ==
12660                             CTL_DELAY_TYPE_ONESHOT)
12661                                 lun->delay_info.datamove_delay = 0;
12662                         return;
12663                 }
12664         }
12665 #endif
12666
12667         /*
12668          * This command has been aborted.  Set the port status, so we fail
12669          * the data move.
12670          */
12671         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12672                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12673                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12674                        io->io_hdr.nexus.targ_port,
12675                        io->io_hdr.nexus.targ_lun);
12676                 io->io_hdr.port_status = 31337;
12677                 /*
12678                  * Note that the backend, in this case, will get the
12679                  * callback in its context.  In other cases it may get
12680                  * called in the frontend's interrupt thread context.
12681                  */
12682                 io->scsiio.be_move_done(io);
12683                 return;
12684         }
12685
12686         /* Don't confuse frontend with zero length data move. */
12687         if (io->scsiio.kern_data_len == 0) {
12688                 io->scsiio.be_move_done(io);
12689                 return;
12690         }
12691
12692         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12693         fe_datamove(io);
12694 }
12695
12696 static void
12697 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12698 {
12699         union ctl_ha_msg msg;
12700 #ifdef CTL_TIME_IO
12701         struct bintime cur_bt;
12702 #endif
12703
12704         memset(&msg, 0, sizeof(msg));
12705         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12706         msg.hdr.original_sc = io;
12707         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12708         msg.hdr.nexus = io->io_hdr.nexus;
12709         msg.hdr.status = io->io_hdr.status;
12710         msg.scsi.tag_num = io->scsiio.tag_num;
12711         msg.scsi.tag_type = io->scsiio.tag_type;
12712         msg.scsi.scsi_status = io->scsiio.scsi_status;
12713         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12714                io->scsiio.sense_len);
12715         msg.scsi.sense_len = io->scsiio.sense_len;
12716         msg.scsi.sense_residual = io->scsiio.sense_residual;
12717         msg.scsi.fetd_status = io->io_hdr.port_status;
12718         msg.scsi.residual = io->scsiio.residual;
12719         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12720         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12721                 ctl_failover_io(io, /*have_lock*/ have_lock);
12722                 return;
12723         }
12724         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12725             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12726             msg.scsi.sense_len, M_WAITOK);
12727
12728 #ifdef CTL_TIME_IO
12729         getbinuptime(&cur_bt);
12730         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12731         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12732 #endif
12733         io->io_hdr.num_dmas++;
12734 }
12735
12736 /*
12737  * The DMA to the remote side is done, now we need to tell the other side
12738  * we're done so it can continue with its data movement.
12739  */
12740 static void
12741 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12742 {
12743         union ctl_io *io;
12744         uint32_t i;
12745
12746         io = rq->context;
12747
12748         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12749                 printf("%s: ISC DMA write failed with error %d", __func__,
12750                        rq->ret);
12751                 ctl_set_internal_failure(&io->scsiio,
12752                                          /*sks_valid*/ 1,
12753                                          /*retry_count*/ rq->ret);
12754         }
12755
12756         ctl_dt_req_free(rq);
12757
12758         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12759                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12760         free(io->io_hdr.remote_sglist, M_CTL);
12761         io->io_hdr.remote_sglist = NULL;
12762         io->io_hdr.local_sglist = NULL;
12763
12764         /*
12765          * The data is in local and remote memory, so now we need to send
12766          * status (good or back) back to the other side.
12767          */
12768         ctl_send_datamove_done(io, /*have_lock*/ 0);
12769 }
12770
12771 /*
12772  * We've moved the data from the host/controller into local memory.  Now we
12773  * need to push it over to the remote controller's memory.
12774  */
12775 static int
12776 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12777 {
12778         int retval;
12779
12780         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12781                                           ctl_datamove_remote_write_cb);
12782         return (retval);
12783 }
12784
12785 static void
12786 ctl_datamove_remote_write(union ctl_io *io)
12787 {
12788         int retval;
12789         void (*fe_datamove)(union ctl_io *io);
12790
12791         /*
12792          * - Get the data from the host/HBA into local memory.
12793          * - DMA memory from the local controller to the remote controller.
12794          * - Send status back to the remote controller.
12795          */
12796
12797         retval = ctl_datamove_remote_sgl_setup(io);
12798         if (retval != 0)
12799                 return;
12800
12801         /* Switch the pointer over so the FETD knows what to do */
12802         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12803
12804         /*
12805          * Use a custom move done callback, since we need to send completion
12806          * back to the other controller, not to the backend on this side.
12807          */
12808         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12809
12810         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12811         fe_datamove(io);
12812 }
12813
12814 static int
12815 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12816 {
12817 #if 0
12818         char str[256];
12819         char path_str[64];
12820         struct sbuf sb;
12821 #endif
12822         uint32_t i;
12823
12824         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12825                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12826         free(io->io_hdr.remote_sglist, M_CTL);
12827         io->io_hdr.remote_sglist = NULL;
12828         io->io_hdr.local_sglist = NULL;
12829
12830 #if 0
12831         scsi_path_string(io, path_str, sizeof(path_str));
12832         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12833         sbuf_cat(&sb, path_str);
12834         scsi_command_string(&io->scsiio, NULL, &sb);
12835         sbuf_printf(&sb, "\n");
12836         sbuf_cat(&sb, path_str);
12837         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12838                     io->scsiio.tag_num, io->scsiio.tag_type);
12839         sbuf_cat(&sb, path_str);
12840         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12841                     io->io_hdr.flags, io->io_hdr.status);
12842         sbuf_finish(&sb);
12843         printk("%s", sbuf_data(&sb));
12844 #endif
12845
12846
12847         /*
12848          * The read is done, now we need to send status (good or bad) back
12849          * to the other side.
12850          */
12851         ctl_send_datamove_done(io, /*have_lock*/ 0);
12852
12853         return (0);
12854 }
12855
12856 static void
12857 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12858 {
12859         union ctl_io *io;
12860         void (*fe_datamove)(union ctl_io *io);
12861
12862         io = rq->context;
12863
12864         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12865                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12866                        rq->ret);
12867                 ctl_set_internal_failure(&io->scsiio,
12868                                          /*sks_valid*/ 1,
12869                                          /*retry_count*/ rq->ret);
12870         }
12871
12872         ctl_dt_req_free(rq);
12873
12874         /* Switch the pointer over so the FETD knows what to do */
12875         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12876
12877         /*
12878          * Use a custom move done callback, since we need to send completion
12879          * back to the other controller, not to the backend on this side.
12880          */
12881         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12882
12883         /* XXX KDM add checks like the ones in ctl_datamove? */
12884
12885         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12886         fe_datamove(io);
12887 }
12888
12889 static int
12890 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12891 {
12892         struct ctl_sg_entry *local_sglist;
12893         uint32_t len_to_go;
12894         int retval;
12895         int i;
12896
12897         retval = 0;
12898         local_sglist = io->io_hdr.local_sglist;
12899         len_to_go = io->scsiio.kern_data_len;
12900
12901         /*
12902          * The difficult thing here is that the size of the various
12903          * S/G segments may be different than the size from the
12904          * remote controller.  That'll make it harder when DMAing
12905          * the data back to the other side.
12906          */
12907         for (i = 0; len_to_go > 0; i++) {
12908                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12909                 local_sglist[i].addr =
12910                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12911
12912                 len_to_go -= local_sglist[i].len;
12913         }
12914         /*
12915          * Reset the number of S/G entries accordingly.  The original
12916          * number of S/G entries is available in rem_sg_entries.
12917          */
12918         io->scsiio.kern_sg_entries = i;
12919
12920 #if 0
12921         printf("%s: kern_sg_entries = %d\n", __func__,
12922                io->scsiio.kern_sg_entries);
12923         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12924                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12925                        local_sglist[i].addr, local_sglist[i].len);
12926 #endif
12927
12928         return (retval);
12929 }
12930
12931 static int
12932 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12933                          ctl_ha_dt_cb callback)
12934 {
12935         struct ctl_ha_dt_req *rq;
12936         struct ctl_sg_entry *remote_sglist, *local_sglist;
12937         uint32_t local_used, remote_used, total_used;
12938         int i, j, isc_ret;
12939
12940         rq = ctl_dt_req_alloc();
12941
12942         /*
12943          * If we failed to allocate the request, and if the DMA didn't fail
12944          * anyway, set busy status.  This is just a resource allocation
12945          * failure.
12946          */
12947         if ((rq == NULL)
12948          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12949              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12950                 ctl_set_busy(&io->scsiio);
12951
12952         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12953             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12954
12955                 if (rq != NULL)
12956                         ctl_dt_req_free(rq);
12957
12958                 /*
12959                  * The data move failed.  We need to return status back
12960                  * to the other controller.  No point in trying to DMA
12961                  * data to the remote controller.
12962                  */
12963
12964                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12965
12966                 return (1);
12967         }
12968
12969         local_sglist = io->io_hdr.local_sglist;
12970         remote_sglist = io->io_hdr.remote_sglist;
12971         local_used = 0;
12972         remote_used = 0;
12973         total_used = 0;
12974
12975         /*
12976          * Pull/push the data over the wire from/to the other controller.
12977          * This takes into account the possibility that the local and
12978          * remote sglists may not be identical in terms of the size of
12979          * the elements and the number of elements.
12980          *
12981          * One fundamental assumption here is that the length allocated for
12982          * both the local and remote sglists is identical.  Otherwise, we've
12983          * essentially got a coding error of some sort.
12984          */
12985         isc_ret = CTL_HA_STATUS_SUCCESS;
12986         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12987                 uint32_t cur_len;
12988                 uint8_t *tmp_ptr;
12989
12990                 rq->command = command;
12991                 rq->context = io;
12992
12993                 /*
12994                  * Both pointers should be aligned.  But it is possible
12995                  * that the allocation length is not.  They should both
12996                  * also have enough slack left over at the end, though,
12997                  * to round up to the next 8 byte boundary.
12998                  */
12999                 cur_len = MIN(local_sglist[i].len - local_used,
13000                               remote_sglist[j].len - remote_used);
13001                 rq->size = cur_len;
13002
13003                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
13004                 tmp_ptr += local_used;
13005
13006 #if 0
13007                 /* Use physical addresses when talking to ISC hardware */
13008                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13009                         /* XXX KDM use busdma */
13010                         rq->local = vtophys(tmp_ptr);
13011                 } else
13012                         rq->local = tmp_ptr;
13013 #else
13014                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13015                     ("HA does not support BUS_ADDR"));
13016                 rq->local = tmp_ptr;
13017 #endif
13018
13019                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13020                 tmp_ptr += remote_used;
13021                 rq->remote = tmp_ptr;
13022
13023                 rq->callback = NULL;
13024
13025                 local_used += cur_len;
13026                 if (local_used >= local_sglist[i].len) {
13027                         i++;
13028                         local_used = 0;
13029                 }
13030
13031                 remote_used += cur_len;
13032                 if (remote_used >= remote_sglist[j].len) {
13033                         j++;
13034                         remote_used = 0;
13035                 }
13036                 total_used += cur_len;
13037
13038                 if (total_used >= io->scsiio.kern_data_len)
13039                         rq->callback = callback;
13040
13041 #if 0
13042                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13043                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13044                        rq->local, rq->remote, rq->size);
13045 #endif
13046
13047                 isc_ret = ctl_dt_single(rq);
13048                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13049                         break;
13050         }
13051         if (isc_ret != CTL_HA_STATUS_WAIT) {
13052                 rq->ret = isc_ret;
13053                 callback(rq);
13054         }
13055
13056         return (0);
13057 }
13058
13059 static void
13060 ctl_datamove_remote_read(union ctl_io *io)
13061 {
13062         int retval;
13063         uint32_t i;
13064
13065         /*
13066          * This will send an error to the other controller in the case of a
13067          * failure.
13068          */
13069         retval = ctl_datamove_remote_sgl_setup(io);
13070         if (retval != 0)
13071                 return;
13072
13073         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13074                                           ctl_datamove_remote_read_cb);
13075         if (retval != 0) {
13076                 /*
13077                  * Make sure we free memory if there was an error..  The
13078                  * ctl_datamove_remote_xfer() function will send the
13079                  * datamove done message, or call the callback with an
13080                  * error if there is a problem.
13081                  */
13082                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13083                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13084                 free(io->io_hdr.remote_sglist, M_CTL);
13085                 io->io_hdr.remote_sglist = NULL;
13086                 io->io_hdr.local_sglist = NULL;
13087         }
13088 }
13089
13090 /*
13091  * Process a datamove request from the other controller.  This is used for
13092  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13093  * first.  Once that is complete, the data gets DMAed into the remote
13094  * controller's memory.  For reads, we DMA from the remote controller's
13095  * memory into our memory first, and then move it out to the FETD.
13096  */
13097 static void
13098 ctl_datamove_remote(union ctl_io *io)
13099 {
13100
13101         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13102
13103         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13104                 ctl_failover_io(io, /*have_lock*/ 0);
13105                 return;
13106         }
13107
13108         /*
13109          * Note that we look for an aborted I/O here, but don't do some of
13110          * the other checks that ctl_datamove() normally does.
13111          * We don't need to run the datamove delay code, since that should
13112          * have been done if need be on the other controller.
13113          */
13114         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13115                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13116                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13117                        io->io_hdr.nexus.targ_port,
13118                        io->io_hdr.nexus.targ_lun);
13119                 io->io_hdr.port_status = 31338;
13120                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13121                 return;
13122         }
13123
13124         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13125                 ctl_datamove_remote_write(io);
13126         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13127                 ctl_datamove_remote_read(io);
13128         else {
13129                 io->io_hdr.port_status = 31339;
13130                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13131         }
13132 }
13133
13134 static void
13135 ctl_process_done(union ctl_io *io)
13136 {
13137         struct ctl_lun *lun;
13138         struct ctl_softc *softc = control_softc;
13139         void (*fe_done)(union ctl_io *io);
13140         union ctl_ha_msg msg;
13141         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13142
13143         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13144         fe_done = softc->ctl_ports[targ_port]->fe_done;
13145
13146 #ifdef CTL_TIME_IO
13147         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13148                 char str[256];
13149                 char path_str[64];
13150                 struct sbuf sb;
13151
13152                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13153                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13154
13155                 sbuf_cat(&sb, path_str);
13156                 switch (io->io_hdr.io_type) {
13157                 case CTL_IO_SCSI:
13158                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13159                         sbuf_printf(&sb, "\n");
13160                         sbuf_cat(&sb, path_str);
13161                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13162                                     io->scsiio.tag_num, io->scsiio.tag_type);
13163                         break;
13164                 case CTL_IO_TASK:
13165                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13166                                     "Tag Type: %d\n", io->taskio.task_action,
13167                                     io->taskio.tag_num, io->taskio.tag_type);
13168                         break;
13169                 default:
13170                         panic("%s: Invalid CTL I/O type %d\n",
13171                             __func__, io->io_hdr.io_type);
13172                 }
13173                 sbuf_cat(&sb, path_str);
13174                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13175                             (intmax_t)time_uptime - io->io_hdr.start_time);
13176                 sbuf_finish(&sb);
13177                 printf("%s", sbuf_data(&sb));
13178         }
13179 #endif /* CTL_TIME_IO */
13180
13181         switch (io->io_hdr.io_type) {
13182         case CTL_IO_SCSI:
13183                 break;
13184         case CTL_IO_TASK:
13185                 if (ctl_debug & CTL_DEBUG_INFO)
13186                         ctl_io_error_print(io, NULL);
13187                 fe_done(io);
13188                 return;
13189         default:
13190                 panic("%s: Invalid CTL I/O type %d\n",
13191                     __func__, io->io_hdr.io_type);
13192         }
13193
13194         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13195         if (lun == NULL) {
13196                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13197                                  io->io_hdr.nexus.targ_mapped_lun));
13198                 goto bailout;
13199         }
13200
13201         mtx_lock(&lun->lun_lock);
13202
13203         /*
13204          * Check to see if we have any informational exception and status
13205          * of this command can be modified to report it in form of either
13206          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13207          */
13208         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13209             io->io_hdr.status == CTL_SUCCESS &&
13210             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13211                 uint8_t mrie = lun->MODE_IE.mrie;
13212                 uint8_t per = ((lun->MODE_RWER.byte3 & SMS_RWER_PER) ||
13213                     (lun->MODE_VER.byte3 & SMS_VER_PER));
13214                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13215                      mrie == SIEP_MRIE_REC_UNCOND ||
13216                      mrie == SIEP_MRIE_NO_SENSE) &&
13217                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13218                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13219                         ctl_set_sense(&io->scsiio,
13220                               /*current_error*/ 1,
13221                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13222                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13223                               /*asc*/ lun->ie_asc,
13224                               /*ascq*/ lun->ie_ascq,
13225                               SSD_ELEM_NONE);
13226                         lun->ie_reported = 1;
13227                 }
13228         } else if (lun->ie_reported < 0)
13229                 lun->ie_reported = 0;
13230
13231         /*
13232          * Check to see if we have any errors to inject here.  We only
13233          * inject errors for commands that don't already have errors set.
13234          */
13235         if (!STAILQ_EMPTY(&lun->error_list) &&
13236             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13237             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13238                 ctl_inject_error(lun, io);
13239
13240         /*
13241          * XXX KDM how do we treat commands that aren't completed
13242          * successfully?
13243          *
13244          * XXX KDM should we also track I/O latency?
13245          */
13246         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13247             io->io_hdr.io_type == CTL_IO_SCSI) {
13248 #ifdef CTL_TIME_IO
13249                 struct bintime cur_bt;
13250 #endif
13251                 int type;
13252
13253                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13254                     CTL_FLAG_DATA_IN)
13255                         type = CTL_STATS_READ;
13256                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13257                     CTL_FLAG_DATA_OUT)
13258                         type = CTL_STATS_WRITE;
13259                 else
13260                         type = CTL_STATS_NO_IO;
13261
13262                 lun->stats.ports[targ_port].bytes[type] +=
13263                     io->scsiio.kern_total_len;
13264                 lun->stats.ports[targ_port].operations[type]++;
13265 #ifdef CTL_TIME_IO
13266                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13267                    &io->io_hdr.dma_bt);
13268                 getbinuptime(&cur_bt);
13269                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13270                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13271 #endif
13272                 lun->stats.ports[targ_port].num_dmas[type] +=
13273                     io->io_hdr.num_dmas;
13274         }
13275
13276         /*
13277          * Remove this from the OOA queue.
13278          */
13279         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13280 #ifdef CTL_TIME_IO
13281         if (TAILQ_EMPTY(&lun->ooa_queue))
13282                 lun->last_busy = getsbinuptime();
13283 #endif
13284
13285         /*
13286          * Run through the blocked queue on this LUN and see if anything
13287          * has become unblocked, now that this transaction is done.
13288          */
13289         ctl_check_blocked(lun);
13290
13291         /*
13292          * If the LUN has been invalidated, free it if there is nothing
13293          * left on its OOA queue.
13294          */
13295         if ((lun->flags & CTL_LUN_INVALID)
13296          && TAILQ_EMPTY(&lun->ooa_queue)) {
13297                 mtx_unlock(&lun->lun_lock);
13298                 mtx_lock(&softc->ctl_lock);
13299                 ctl_free_lun(lun);
13300                 mtx_unlock(&softc->ctl_lock);
13301         } else
13302                 mtx_unlock(&lun->lun_lock);
13303
13304 bailout:
13305
13306         /*
13307          * If this command has been aborted, make sure we set the status
13308          * properly.  The FETD is responsible for freeing the I/O and doing
13309          * whatever it needs to do to clean up its state.
13310          */
13311         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13312                 ctl_set_task_aborted(&io->scsiio);
13313
13314         /*
13315          * If enabled, print command error status.
13316          */
13317         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13318             (ctl_debug & CTL_DEBUG_INFO) != 0)
13319                 ctl_io_error_print(io, NULL);
13320
13321         /*
13322          * Tell the FETD or the other shelf controller we're done with this
13323          * command.  Note that only SCSI commands get to this point.  Task
13324          * management commands are completed above.
13325          */
13326         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13327             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13328                 memset(&msg, 0, sizeof(msg));
13329                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13330                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13331                 msg.hdr.nexus = io->io_hdr.nexus;
13332                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13333                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13334                     M_WAITOK);
13335         }
13336
13337         fe_done(io);
13338 }
13339
13340 #ifdef CTL_WITH_CA
13341 /*
13342  * Front end should call this if it doesn't do autosense.  When the request
13343  * sense comes back in from the initiator, we'll dequeue this and send it.
13344  */
13345 int
13346 ctl_queue_sense(union ctl_io *io)
13347 {
13348         struct ctl_lun *lun;
13349         struct ctl_port *port;
13350         struct ctl_softc *softc;
13351         uint32_t initidx, targ_lun;
13352
13353         softc = control_softc;
13354
13355         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13356
13357         /*
13358          * LUN lookup will likely move to the ctl_work_thread() once we
13359          * have our new queueing infrastructure (that doesn't put things on
13360          * a per-LUN queue initially).  That is so that we can handle
13361          * things like an INQUIRY to a LUN that we don't have enabled.  We
13362          * can't deal with that right now.
13363          */
13364         mtx_lock(&softc->ctl_lock);
13365
13366         /*
13367          * If we don't have a LUN for this, just toss the sense
13368          * information.
13369          */
13370         port = ctl_io_port(&ctsio->io_hdr);
13371         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13372         if ((targ_lun < CTL_MAX_LUNS)
13373          && (softc->ctl_luns[targ_lun] != NULL))
13374                 lun = softc->ctl_luns[targ_lun];
13375         else
13376                 goto bailout;
13377
13378         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13379
13380         mtx_lock(&lun->lun_lock);
13381         /*
13382          * Already have CA set for this LUN...toss the sense information.
13383          */
13384         if (ctl_is_set(lun->have_ca, initidx)) {
13385                 mtx_unlock(&lun->lun_lock);
13386                 goto bailout;
13387         }
13388
13389         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13390                MIN(sizeof(lun->pending_sense[initidx]),
13391                sizeof(io->scsiio.sense_data)));
13392         ctl_set_mask(lun->have_ca, initidx);
13393         mtx_unlock(&lun->lun_lock);
13394
13395 bailout:
13396         mtx_unlock(&softc->ctl_lock);
13397
13398         ctl_free_io(io);
13399
13400         return (CTL_RETVAL_COMPLETE);
13401 }
13402 #endif
13403
13404 /*
13405  * Primary command inlet from frontend ports.  All SCSI and task I/O
13406  * requests must go through this function.
13407  */
13408 int
13409 ctl_queue(union ctl_io *io)
13410 {
13411         struct ctl_port *port;
13412
13413         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13414
13415 #ifdef CTL_TIME_IO
13416         io->io_hdr.start_time = time_uptime;
13417         getbinuptime(&io->io_hdr.start_bt);
13418 #endif /* CTL_TIME_IO */
13419
13420         /* Map FE-specific LUN ID into global one. */
13421         port = ctl_io_port(&io->io_hdr);
13422         io->io_hdr.nexus.targ_mapped_lun =
13423             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13424
13425         switch (io->io_hdr.io_type) {
13426         case CTL_IO_SCSI:
13427         case CTL_IO_TASK:
13428                 if (ctl_debug & CTL_DEBUG_CDB)
13429                         ctl_io_print(io);
13430                 ctl_enqueue_incoming(io);
13431                 break;
13432         default:
13433                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13434                 return (EINVAL);
13435         }
13436
13437         return (CTL_RETVAL_COMPLETE);
13438 }
13439
13440 #ifdef CTL_IO_DELAY
13441 static void
13442 ctl_done_timer_wakeup(void *arg)
13443 {
13444         union ctl_io *io;
13445
13446         io = (union ctl_io *)arg;
13447         ctl_done(io);
13448 }
13449 #endif /* CTL_IO_DELAY */
13450
13451 void
13452 ctl_serseq_done(union ctl_io *io)
13453 {
13454         struct ctl_lun *lun;
13455
13456         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13457         if (lun->be_lun == NULL ||
13458             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13459                 return;
13460         mtx_lock(&lun->lun_lock);
13461         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13462         ctl_check_blocked(lun);
13463         mtx_unlock(&lun->lun_lock);
13464 }
13465
13466 void
13467 ctl_done(union ctl_io *io)
13468 {
13469
13470         /*
13471          * Enable this to catch duplicate completion issues.
13472          */
13473 #if 0
13474         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13475                 printf("%s: type %d msg %d cdb %x iptl: "
13476                        "%u:%u:%u tag 0x%04x "
13477                        "flag %#x status %x\n",
13478                         __func__,
13479                         io->io_hdr.io_type,
13480                         io->io_hdr.msg_type,
13481                         io->scsiio.cdb[0],
13482                         io->io_hdr.nexus.initid,
13483                         io->io_hdr.nexus.targ_port,
13484                         io->io_hdr.nexus.targ_lun,
13485                         (io->io_hdr.io_type ==
13486                         CTL_IO_TASK) ?
13487                         io->taskio.tag_num :
13488                         io->scsiio.tag_num,
13489                         io->io_hdr.flags,
13490                         io->io_hdr.status);
13491         } else
13492                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13493 #endif
13494
13495         /*
13496          * This is an internal copy of an I/O, and should not go through
13497          * the normal done processing logic.
13498          */
13499         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13500                 return;
13501
13502 #ifdef CTL_IO_DELAY
13503         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13504                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13505         } else {
13506                 struct ctl_lun *lun;
13507
13508                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13509
13510                 if ((lun != NULL)
13511                  && (lun->delay_info.done_delay > 0)) {
13512
13513                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13514                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13515                         callout_reset(&io->io_hdr.delay_callout,
13516                                       lun->delay_info.done_delay * hz,
13517                                       ctl_done_timer_wakeup, io);
13518                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13519                                 lun->delay_info.done_delay = 0;
13520                         return;
13521                 }
13522         }
13523 #endif /* CTL_IO_DELAY */
13524
13525         ctl_enqueue_done(io);
13526 }
13527
13528 static void
13529 ctl_work_thread(void *arg)
13530 {
13531         struct ctl_thread *thr = (struct ctl_thread *)arg;
13532         struct ctl_softc *softc = thr->ctl_softc;
13533         union ctl_io *io;
13534         int retval;
13535
13536         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13537
13538         for (;;) {
13539                 /*
13540                  * We handle the queues in this order:
13541                  * - ISC
13542                  * - done queue (to free up resources, unblock other commands)
13543                  * - RtR queue
13544                  * - incoming queue
13545                  *
13546                  * If those queues are empty, we break out of the loop and
13547                  * go to sleep.
13548                  */
13549                 mtx_lock(&thr->queue_lock);
13550                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13551                 if (io != NULL) {
13552                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13553                         mtx_unlock(&thr->queue_lock);
13554                         ctl_handle_isc(io);
13555                         continue;
13556                 }
13557                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13558                 if (io != NULL) {
13559                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13560                         /* clear any blocked commands, call fe_done */
13561                         mtx_unlock(&thr->queue_lock);
13562                         ctl_process_done(io);
13563                         continue;
13564                 }
13565                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13566                 if (io != NULL) {
13567                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13568                         mtx_unlock(&thr->queue_lock);
13569                         if (io->io_hdr.io_type == CTL_IO_TASK)
13570                                 ctl_run_task(io);
13571                         else
13572                                 ctl_scsiio_precheck(softc, &io->scsiio);
13573                         continue;
13574                 }
13575                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13576                 if (io != NULL) {
13577                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13578                         mtx_unlock(&thr->queue_lock);
13579                         retval = ctl_scsiio(&io->scsiio);
13580                         if (retval != CTL_RETVAL_COMPLETE)
13581                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13582                         continue;
13583                 }
13584
13585                 /* Sleep until we have something to do. */
13586                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13587         }
13588 }
13589
13590 static void
13591 ctl_lun_thread(void *arg)
13592 {
13593         struct ctl_softc *softc = (struct ctl_softc *)arg;
13594         struct ctl_be_lun *be_lun;
13595
13596         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13597
13598         for (;;) {
13599                 mtx_lock(&softc->ctl_lock);
13600                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13601                 if (be_lun != NULL) {
13602                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13603                         mtx_unlock(&softc->ctl_lock);
13604                         ctl_create_lun(be_lun);
13605                         continue;
13606                 }
13607
13608                 /* Sleep until we have something to do. */
13609                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13610                     PDROP | PRIBIO, "-", 0);
13611         }
13612 }
13613
13614 static void
13615 ctl_thresh_thread(void *arg)
13616 {
13617         struct ctl_softc *softc = (struct ctl_softc *)arg;
13618         struct ctl_lun *lun;
13619         struct ctl_logical_block_provisioning_page *page;
13620         const char *attr;
13621         union ctl_ha_msg msg;
13622         uint64_t thres, val;
13623         int i, e, set;
13624
13625         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13626
13627         for (;;) {
13628                 mtx_lock(&softc->ctl_lock);
13629                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13630                         if ((lun->flags & CTL_LUN_DISABLED) ||
13631                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13632                             lun->backend->lun_attr == NULL)
13633                                 continue;
13634                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13635                             softc->ha_mode == CTL_HA_MODE_XFER)
13636                                 continue;
13637                         if ((lun->MODE_RWER.byte8 & SMS_RWER_LBPERE) == 0)
13638                                 continue;
13639                         e = 0;
13640                         page = &lun->MODE_LBP;
13641                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13642                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13643                                         continue;
13644                                 thres = scsi_4btoul(page->descr[i].count);
13645                                 thres <<= CTL_LBP_EXPONENT;
13646                                 switch (page->descr[i].resource) {
13647                                 case 0x01:
13648                                         attr = "blocksavail";
13649                                         break;
13650                                 case 0x02:
13651                                         attr = "blocksused";
13652                                         break;
13653                                 case 0xf1:
13654                                         attr = "poolblocksavail";
13655                                         break;
13656                                 case 0xf2:
13657                                         attr = "poolblocksused";
13658                                         break;
13659                                 default:
13660                                         continue;
13661                                 }
13662                                 mtx_unlock(&softc->ctl_lock); // XXX
13663                                 val = lun->backend->lun_attr(
13664                                     lun->be_lun->be_lun, attr);
13665                                 mtx_lock(&softc->ctl_lock);
13666                                 if (val == UINT64_MAX)
13667                                         continue;
13668                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13669                                     == SLBPPD_ARMING_INC)
13670                                         e = (val >= thres);
13671                                 else
13672                                         e = (val <= thres);
13673                                 if (e)
13674                                         break;
13675                         }
13676                         mtx_lock(&lun->lun_lock);
13677                         if (e) {
13678                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13679                                     (uint8_t *)page, lun->ua_tpt_info);
13680                                 if (lun->lasttpt == 0 ||
13681                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13682                                         lun->lasttpt = time_uptime;
13683                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13684                                         set = 1;
13685                                 } else
13686                                         set = 0;
13687                         } else {
13688                                 lun->lasttpt = 0;
13689                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13690                                 set = -1;
13691                         }
13692                         mtx_unlock(&lun->lun_lock);
13693                         if (set != 0 &&
13694                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13695                                 /* Send msg to other side. */
13696                                 bzero(&msg.ua, sizeof(msg.ua));
13697                                 msg.hdr.msg_type = CTL_MSG_UA;
13698                                 msg.hdr.nexus.initid = -1;
13699                                 msg.hdr.nexus.targ_port = -1;
13700                                 msg.hdr.nexus.targ_lun = lun->lun;
13701                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13702                                 msg.ua.ua_all = 1;
13703                                 msg.ua.ua_set = (set > 0);
13704                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13705                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13706                                 mtx_unlock(&softc->ctl_lock); // XXX
13707                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13708                                     sizeof(msg.ua), M_WAITOK);
13709                                 mtx_lock(&softc->ctl_lock);
13710                         }
13711                 }
13712                 mtx_unlock(&softc->ctl_lock);
13713                 pause("-", CTL_LBP_PERIOD * hz);
13714         }
13715 }
13716
13717 static void
13718 ctl_enqueue_incoming(union ctl_io *io)
13719 {
13720         struct ctl_softc *softc = control_softc;
13721         struct ctl_thread *thr;
13722         u_int idx;
13723
13724         idx = (io->io_hdr.nexus.targ_port * 127 +
13725                io->io_hdr.nexus.initid) % worker_threads;
13726         thr = &softc->threads[idx];
13727         mtx_lock(&thr->queue_lock);
13728         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13729         mtx_unlock(&thr->queue_lock);
13730         wakeup(thr);
13731 }
13732
13733 static void
13734 ctl_enqueue_rtr(union ctl_io *io)
13735 {
13736         struct ctl_softc *softc = control_softc;
13737         struct ctl_thread *thr;
13738
13739         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13740         mtx_lock(&thr->queue_lock);
13741         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13742         mtx_unlock(&thr->queue_lock);
13743         wakeup(thr);
13744 }
13745
13746 static void
13747 ctl_enqueue_done(union ctl_io *io)
13748 {
13749         struct ctl_softc *softc = control_softc;
13750         struct ctl_thread *thr;
13751
13752         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13753         mtx_lock(&thr->queue_lock);
13754         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13755         mtx_unlock(&thr->queue_lock);
13756         wakeup(thr);
13757 }
13758
13759 static void
13760 ctl_enqueue_isc(union ctl_io *io)
13761 {
13762         struct ctl_softc *softc = control_softc;
13763         struct ctl_thread *thr;
13764
13765         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13766         mtx_lock(&thr->queue_lock);
13767         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13768         mtx_unlock(&thr->queue_lock);
13769         wakeup(thr);
13770 }
13771
13772 /*
13773  *  vim: ts=8
13774  */