]> CyberLeo.Net >> Repos - FreeBSD/stable/10.git/blob - sys/cam/ctl/ctl.c
MFC r310259:
[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,
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*/0
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*/0,
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                 c = str[i];
4366                 if (isdigit(c))
4367                         c -= '0';
4368                 else if (isalpha(c))
4369                         c -= isupper(c) ? 'A' - 10 : 'a' - 10;
4370                 else
4371                         break;
4372                 if (c >= 16)
4373                         break;
4374                 if ((i & 1) == 0)
4375                         buf[i / 2] |= (c << 4);
4376                 else
4377                         buf[i / 2] |= c;
4378         }
4379         return ((i + 1) / 2);
4380 }
4381
4382 /*
4383  * LUN allocation.
4384  *
4385  * Requirements:
4386  * - caller allocates and zeros LUN storage, or passes in a NULL LUN if he
4387  *   wants us to allocate the LUN and he can block.
4388  * - ctl_softc is always set
4389  * - be_lun is set if the LUN has a backend (needed for disk LUNs)
4390  *
4391  * Returns 0 for success, non-zero (errno) for failure.
4392  */
4393 static int
4394 ctl_alloc_lun(struct ctl_softc *ctl_softc, struct ctl_lun *ctl_lun,
4395               struct ctl_be_lun *const be_lun)
4396 {
4397         struct ctl_lun *nlun, *lun;
4398         struct scsi_vpd_id_descriptor *desc;
4399         struct scsi_vpd_id_t10 *t10id;
4400         const char *eui, *naa, *scsiname, *vendor, *value;
4401         int lun_number, i, lun_malloced;
4402         int devidlen, idlen1, idlen2 = 0, len;
4403
4404         if (be_lun == NULL)
4405                 return (EINVAL);
4406
4407         /*
4408          * We currently only support Direct Access or Processor LUN types.
4409          */
4410         switch (be_lun->lun_type) {
4411         case T_DIRECT:
4412         case T_PROCESSOR:
4413         case T_CDROM:
4414                 break;
4415         case T_SEQUENTIAL:
4416         case T_CHANGER:
4417         default:
4418                 be_lun->lun_config_status(be_lun->be_lun,
4419                                           CTL_LUN_CONFIG_FAILURE);
4420                 break;
4421         }
4422         if (ctl_lun == NULL) {
4423                 lun = malloc(sizeof(*lun), M_CTL, M_WAITOK);
4424                 lun_malloced = 1;
4425         } else {
4426                 lun_malloced = 0;
4427                 lun = ctl_lun;
4428         }
4429
4430         memset(lun, 0, sizeof(*lun));
4431         if (lun_malloced)
4432                 lun->flags = CTL_LUN_MALLOCED;
4433
4434         /* Generate LUN ID. */
4435         devidlen = max(CTL_DEVID_MIN_LEN,
4436             strnlen(be_lun->device_id, CTL_DEVID_LEN));
4437         idlen1 = sizeof(*t10id) + devidlen;
4438         len = sizeof(struct scsi_vpd_id_descriptor) + idlen1;
4439         scsiname = ctl_get_opt(&be_lun->options, "scsiname");
4440         if (scsiname != NULL) {
4441                 idlen2 = roundup2(strlen(scsiname) + 1, 4);
4442                 len += sizeof(struct scsi_vpd_id_descriptor) + idlen2;
4443         }
4444         eui = ctl_get_opt(&be_lun->options, "eui");
4445         if (eui != NULL) {
4446                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4447         }
4448         naa = ctl_get_opt(&be_lun->options, "naa");
4449         if (naa != NULL) {
4450                 len += sizeof(struct scsi_vpd_id_descriptor) + 16;
4451         }
4452         lun->lun_devid = malloc(sizeof(struct ctl_devid) + len,
4453             M_CTL, M_WAITOK | M_ZERO);
4454         desc = (struct scsi_vpd_id_descriptor *)lun->lun_devid->data;
4455         desc->proto_codeset = SVPD_ID_CODESET_ASCII;
4456         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN | SVPD_ID_TYPE_T10;
4457         desc->length = idlen1;
4458         t10id = (struct scsi_vpd_id_t10 *)&desc->identifier[0];
4459         memset(t10id->vendor, ' ', sizeof(t10id->vendor));
4460         if ((vendor = ctl_get_opt(&be_lun->options, "vendor")) == NULL) {
4461                 strncpy((char *)t10id->vendor, CTL_VENDOR, sizeof(t10id->vendor));
4462         } else {
4463                 strncpy(t10id->vendor, vendor,
4464                     min(sizeof(t10id->vendor), strlen(vendor)));
4465         }
4466         strncpy((char *)t10id->vendor_spec_id,
4467             (char *)be_lun->device_id, devidlen);
4468         if (scsiname != NULL) {
4469                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4470                     desc->length);
4471                 desc->proto_codeset = SVPD_ID_CODESET_UTF8;
4472                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4473                     SVPD_ID_TYPE_SCSI_NAME;
4474                 desc->length = idlen2;
4475                 strlcpy(desc->identifier, scsiname, idlen2);
4476         }
4477         if (eui != NULL) {
4478                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4479                     desc->length);
4480                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4481                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4482                     SVPD_ID_TYPE_EUI64;
4483                 desc->length = hex2bin(eui, desc->identifier, 16);
4484                 desc->length = desc->length > 12 ? 16 :
4485                     (desc->length > 8 ? 12 : 8);
4486                 len -= 16 - desc->length;
4487         }
4488         if (naa != NULL) {
4489                 desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
4490                     desc->length);
4491                 desc->proto_codeset = SVPD_ID_CODESET_BINARY;
4492                 desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_LUN |
4493                     SVPD_ID_TYPE_NAA;
4494                 desc->length = hex2bin(naa, desc->identifier, 16);
4495                 desc->length = desc->length > 8 ? 16 : 8;
4496                 len -= 16 - desc->length;
4497         }
4498         lun->lun_devid->len = len;
4499
4500         mtx_lock(&ctl_softc->ctl_lock);
4501         /*
4502          * See if the caller requested a particular LUN number.  If so, see
4503          * if it is available.  Otherwise, allocate the first available LUN.
4504          */
4505         if (be_lun->flags & CTL_LUN_FLAG_ID_REQ) {
4506                 if ((be_lun->req_lun_id > (CTL_MAX_LUNS - 1))
4507                  || (ctl_is_set(ctl_softc->ctl_lun_mask, be_lun->req_lun_id))) {
4508                         mtx_unlock(&ctl_softc->ctl_lock);
4509                         if (be_lun->req_lun_id > (CTL_MAX_LUNS - 1)) {
4510                                 printf("ctl: requested LUN ID %d is higher "
4511                                        "than CTL_MAX_LUNS - 1 (%d)\n",
4512                                        be_lun->req_lun_id, CTL_MAX_LUNS - 1);
4513                         } else {
4514                                 /*
4515                                  * XXX KDM return an error, or just assign
4516                                  * another LUN ID in this case??
4517                                  */
4518                                 printf("ctl: requested LUN ID %d is already "
4519                                        "in use\n", be_lun->req_lun_id);
4520                         }
4521                         if (lun->flags & CTL_LUN_MALLOCED)
4522                                 free(lun, M_CTL);
4523                         be_lun->lun_config_status(be_lun->be_lun,
4524                                                   CTL_LUN_CONFIG_FAILURE);
4525                         return (ENOSPC);
4526                 }
4527                 lun_number = be_lun->req_lun_id;
4528         } else {
4529                 lun_number = ctl_ffz(ctl_softc->ctl_lun_mask, 0, CTL_MAX_LUNS);
4530                 if (lun_number == -1) {
4531                         mtx_unlock(&ctl_softc->ctl_lock);
4532                         printf("ctl: can't allocate LUN, out of LUNs\n");
4533                         if (lun->flags & CTL_LUN_MALLOCED)
4534                                 free(lun, M_CTL);
4535                         be_lun->lun_config_status(be_lun->be_lun,
4536                                                   CTL_LUN_CONFIG_FAILURE);
4537                         return (ENOSPC);
4538                 }
4539         }
4540         ctl_set_mask(ctl_softc->ctl_lun_mask, lun_number);
4541
4542         mtx_init(&lun->lun_lock, "CTL LUN", NULL, MTX_DEF);
4543         lun->lun = lun_number;
4544         lun->be_lun = be_lun;
4545         /*
4546          * The processor LUN is always enabled.  Disk LUNs come on line
4547          * disabled, and must be enabled by the backend.
4548          */
4549         lun->flags |= CTL_LUN_DISABLED;
4550         lun->backend = be_lun->be;
4551         be_lun->ctl_lun = lun;
4552         be_lun->lun_id = lun_number;
4553         atomic_add_int(&be_lun->be->num_luns, 1);
4554         if (be_lun->flags & CTL_LUN_FLAG_EJECTED)
4555                 lun->flags |= CTL_LUN_EJECTED;
4556         if (be_lun->flags & CTL_LUN_FLAG_NO_MEDIA)
4557                 lun->flags |= CTL_LUN_NO_MEDIA;
4558         if (be_lun->flags & CTL_LUN_FLAG_STOPPED)
4559                 lun->flags |= CTL_LUN_STOPPED;
4560
4561         if (be_lun->flags & CTL_LUN_FLAG_PRIMARY)
4562                 lun->flags |= CTL_LUN_PRIMARY_SC;
4563
4564         value = ctl_get_opt(&be_lun->options, "removable");
4565         if (value != NULL) {
4566                 if (strcmp(value, "on") == 0)
4567                         lun->flags |= CTL_LUN_REMOVABLE;
4568         } else if (be_lun->lun_type == T_CDROM)
4569                 lun->flags |= CTL_LUN_REMOVABLE;
4570
4571         lun->ctl_softc = ctl_softc;
4572 #ifdef CTL_TIME_IO
4573         lun->last_busy = getsbinuptime();
4574 #endif
4575         TAILQ_INIT(&lun->ooa_queue);
4576         TAILQ_INIT(&lun->blocked_queue);
4577         STAILQ_INIT(&lun->error_list);
4578         lun->ie_reported = 1;
4579         callout_init_mtx(&lun->ie_callout, &lun->lun_lock, 0);
4580         ctl_tpc_lun_init(lun);
4581
4582         /*
4583          * Initialize the mode and log page index.
4584          */
4585         ctl_init_page_index(lun);
4586         ctl_init_log_page_index(lun);
4587
4588         /*
4589          * Now, before we insert this lun on the lun list, set the lun
4590          * inventory changed UA for all other luns.
4591          */
4592         STAILQ_FOREACH(nlun, &ctl_softc->lun_list, links) {
4593                 mtx_lock(&nlun->lun_lock);
4594                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4595                 mtx_unlock(&nlun->lun_lock);
4596         }
4597
4598         STAILQ_INSERT_TAIL(&ctl_softc->lun_list, lun, links);
4599
4600         ctl_softc->ctl_luns[lun_number] = lun;
4601
4602         ctl_softc->num_luns++;
4603
4604         /* Setup statistics gathering */
4605         lun->stats.device_type = be_lun->lun_type;
4606         lun->stats.lun_number = lun_number;
4607         lun->stats.blocksize = be_lun->blocksize;
4608         if (be_lun->blocksize == 0)
4609                 lun->stats.flags = CTL_LUN_STATS_NO_BLOCKSIZE;
4610         for (i = 0;i < CTL_MAX_PORTS;i++)
4611                 lun->stats.ports[i].targ_port = i;
4612
4613         mtx_unlock(&ctl_softc->ctl_lock);
4614
4615         lun->be_lun->lun_config_status(lun->be_lun->be_lun, CTL_LUN_CONFIG_OK);
4616         return (0);
4617 }
4618
4619 /*
4620  * Delete a LUN.
4621  * Assumptions:
4622  * - LUN has already been marked invalid and any pending I/O has been taken
4623  *   care of.
4624  */
4625 static int
4626 ctl_free_lun(struct ctl_lun *lun)
4627 {
4628         struct ctl_softc *softc;
4629         struct ctl_lun *nlun;
4630         int i;
4631
4632         softc = lun->ctl_softc;
4633
4634         mtx_assert(&softc->ctl_lock, MA_OWNED);
4635
4636         STAILQ_REMOVE(&softc->lun_list, lun, ctl_lun, links);
4637
4638         ctl_clear_mask(softc->ctl_lun_mask, lun->lun);
4639
4640         softc->ctl_luns[lun->lun] = NULL;
4641
4642         if (!TAILQ_EMPTY(&lun->ooa_queue))
4643                 panic("Freeing a LUN %p with outstanding I/O!!\n", lun);
4644
4645         softc->num_luns--;
4646
4647         /*
4648          * Tell the backend to free resources, if this LUN has a backend.
4649          */
4650         atomic_subtract_int(&lun->be_lun->be->num_luns, 1);
4651         lun->be_lun->lun_shutdown(lun->be_lun->be_lun);
4652
4653         lun->ie_reportcnt = UINT32_MAX;
4654         callout_drain(&lun->ie_callout);
4655
4656         ctl_tpc_lun_shutdown(lun);
4657         mtx_destroy(&lun->lun_lock);
4658         free(lun->lun_devid, M_CTL);
4659         for (i = 0; i < CTL_MAX_PORTS; i++)
4660                 free(lun->pending_ua[i], M_CTL);
4661         for (i = 0; i < CTL_MAX_PORTS; i++)
4662                 free(lun->pr_keys[i], M_CTL);
4663         free(lun->write_buffer, M_CTL);
4664         if (lun->flags & CTL_LUN_MALLOCED)
4665                 free(lun, M_CTL);
4666
4667         STAILQ_FOREACH(nlun, &softc->lun_list, links) {
4668                 mtx_lock(&nlun->lun_lock);
4669                 ctl_est_ua_all(nlun, -1, CTL_UA_LUN_CHANGE);
4670                 mtx_unlock(&nlun->lun_lock);
4671         }
4672
4673         return (0);
4674 }
4675
4676 static void
4677 ctl_create_lun(struct ctl_be_lun *be_lun)
4678 {
4679
4680         /*
4681          * ctl_alloc_lun() should handle all potential failure cases.
4682          */
4683         ctl_alloc_lun(control_softc, NULL, be_lun);
4684 }
4685
4686 int
4687 ctl_add_lun(struct ctl_be_lun *be_lun)
4688 {
4689         struct ctl_softc *softc = control_softc;
4690
4691         mtx_lock(&softc->ctl_lock);
4692         STAILQ_INSERT_TAIL(&softc->pending_lun_queue, be_lun, links);
4693         mtx_unlock(&softc->ctl_lock);
4694         wakeup(&softc->pending_lun_queue);
4695
4696         return (0);
4697 }
4698
4699 int
4700 ctl_enable_lun(struct ctl_be_lun *be_lun)
4701 {
4702         struct ctl_softc *softc;
4703         struct ctl_port *port, *nport;
4704         struct ctl_lun *lun;
4705         int retval;
4706
4707         lun = (struct ctl_lun *)be_lun->ctl_lun;
4708         softc = lun->ctl_softc;
4709
4710         mtx_lock(&softc->ctl_lock);
4711         mtx_lock(&lun->lun_lock);
4712         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4713                 /*
4714                  * eh?  Why did we get called if the LUN is already
4715                  * enabled?
4716                  */
4717                 mtx_unlock(&lun->lun_lock);
4718                 mtx_unlock(&softc->ctl_lock);
4719                 return (0);
4720         }
4721         lun->flags &= ~CTL_LUN_DISABLED;
4722         mtx_unlock(&lun->lun_lock);
4723
4724         STAILQ_FOREACH_SAFE(port, &softc->port_list, links, nport) {
4725                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4726                     port->lun_map != NULL || port->lun_enable == NULL)
4727                         continue;
4728
4729                 /*
4730                  * Drop the lock while we call the FETD's enable routine.
4731                  * This can lead to a callback into CTL (at least in the
4732                  * case of the internal initiator frontend.
4733                  */
4734                 mtx_unlock(&softc->ctl_lock);
4735                 retval = port->lun_enable(port->targ_lun_arg, lun->lun);
4736                 mtx_lock(&softc->ctl_lock);
4737                 if (retval != 0) {
4738                         printf("%s: FETD %s port %d returned error "
4739                                "%d for lun_enable on lun %jd\n",
4740                                __func__, port->port_name, port->targ_port,
4741                                retval, (intmax_t)lun->lun);
4742                 }
4743         }
4744
4745         mtx_unlock(&softc->ctl_lock);
4746         ctl_isc_announce_lun(lun);
4747
4748         return (0);
4749 }
4750
4751 int
4752 ctl_disable_lun(struct ctl_be_lun *be_lun)
4753 {
4754         struct ctl_softc *softc;
4755         struct ctl_port *port;
4756         struct ctl_lun *lun;
4757         int retval;
4758
4759         lun = (struct ctl_lun *)be_lun->ctl_lun;
4760         softc = lun->ctl_softc;
4761
4762         mtx_lock(&softc->ctl_lock);
4763         mtx_lock(&lun->lun_lock);
4764         if (lun->flags & CTL_LUN_DISABLED) {
4765                 mtx_unlock(&lun->lun_lock);
4766                 mtx_unlock(&softc->ctl_lock);
4767                 return (0);
4768         }
4769         lun->flags |= CTL_LUN_DISABLED;
4770         mtx_unlock(&lun->lun_lock);
4771
4772         STAILQ_FOREACH(port, &softc->port_list, links) {
4773                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0 ||
4774                     port->lun_map != NULL || port->lun_disable == NULL)
4775                         continue;
4776
4777                 /*
4778                  * Drop the lock before we call the frontend's disable
4779                  * routine, to avoid lock order reversals.
4780                  *
4781                  * XXX KDM what happens if the frontend list changes while
4782                  * we're traversing it?  It's unlikely, but should be handled.
4783                  */
4784                 mtx_unlock(&softc->ctl_lock);
4785                 retval = port->lun_disable(port->targ_lun_arg, lun->lun);
4786                 mtx_lock(&softc->ctl_lock);
4787                 if (retval != 0) {
4788                         printf("%s: FETD %s port %d returned error "
4789                                "%d for lun_disable on lun %jd\n",
4790                                __func__, port->port_name, port->targ_port,
4791                                retval, (intmax_t)lun->lun);
4792                 }
4793         }
4794
4795         mtx_unlock(&softc->ctl_lock);
4796         ctl_isc_announce_lun(lun);
4797
4798         return (0);
4799 }
4800
4801 int
4802 ctl_start_lun(struct ctl_be_lun *be_lun)
4803 {
4804         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4805
4806         mtx_lock(&lun->lun_lock);
4807         lun->flags &= ~CTL_LUN_STOPPED;
4808         mtx_unlock(&lun->lun_lock);
4809         return (0);
4810 }
4811
4812 int
4813 ctl_stop_lun(struct ctl_be_lun *be_lun)
4814 {
4815         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4816
4817         mtx_lock(&lun->lun_lock);
4818         lun->flags |= CTL_LUN_STOPPED;
4819         mtx_unlock(&lun->lun_lock);
4820         return (0);
4821 }
4822
4823 int
4824 ctl_lun_no_media(struct ctl_be_lun *be_lun)
4825 {
4826         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4827
4828         mtx_lock(&lun->lun_lock);
4829         lun->flags |= CTL_LUN_NO_MEDIA;
4830         mtx_unlock(&lun->lun_lock);
4831         return (0);
4832 }
4833
4834 int
4835 ctl_lun_has_media(struct ctl_be_lun *be_lun)
4836 {
4837         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4838         union ctl_ha_msg msg;
4839
4840         mtx_lock(&lun->lun_lock);
4841         lun->flags &= ~(CTL_LUN_NO_MEDIA | CTL_LUN_EJECTED);
4842         if (lun->flags & CTL_LUN_REMOVABLE)
4843                 ctl_est_ua_all(lun, -1, CTL_UA_MEDIUM_CHANGE);
4844         mtx_unlock(&lun->lun_lock);
4845         if ((lun->flags & CTL_LUN_REMOVABLE) &&
4846             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4847                 bzero(&msg.ua, sizeof(msg.ua));
4848                 msg.hdr.msg_type = CTL_MSG_UA;
4849                 msg.hdr.nexus.initid = -1;
4850                 msg.hdr.nexus.targ_port = -1;
4851                 msg.hdr.nexus.targ_lun = lun->lun;
4852                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4853                 msg.ua.ua_all = 1;
4854                 msg.ua.ua_set = 1;
4855                 msg.ua.ua_type = CTL_UA_MEDIUM_CHANGE;
4856                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4857                     M_WAITOK);
4858         }
4859         return (0);
4860 }
4861
4862 int
4863 ctl_lun_ejected(struct ctl_be_lun *be_lun)
4864 {
4865         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4866
4867         mtx_lock(&lun->lun_lock);
4868         lun->flags |= CTL_LUN_EJECTED;
4869         mtx_unlock(&lun->lun_lock);
4870         return (0);
4871 }
4872
4873 int
4874 ctl_lun_primary(struct ctl_be_lun *be_lun)
4875 {
4876         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4877
4878         mtx_lock(&lun->lun_lock);
4879         lun->flags |= CTL_LUN_PRIMARY_SC;
4880         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4881         mtx_unlock(&lun->lun_lock);
4882         ctl_isc_announce_lun(lun);
4883         return (0);
4884 }
4885
4886 int
4887 ctl_lun_secondary(struct ctl_be_lun *be_lun)
4888 {
4889         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4890
4891         mtx_lock(&lun->lun_lock);
4892         lun->flags &= ~CTL_LUN_PRIMARY_SC;
4893         ctl_est_ua_all(lun, -1, CTL_UA_ASYM_ACC_CHANGE);
4894         mtx_unlock(&lun->lun_lock);
4895         ctl_isc_announce_lun(lun);
4896         return (0);
4897 }
4898
4899 int
4900 ctl_invalidate_lun(struct ctl_be_lun *be_lun)
4901 {
4902         struct ctl_softc *softc;
4903         struct ctl_lun *lun;
4904
4905         lun = (struct ctl_lun *)be_lun->ctl_lun;
4906         softc = lun->ctl_softc;
4907
4908         mtx_lock(&lun->lun_lock);
4909
4910         /*
4911          * The LUN needs to be disabled before it can be marked invalid.
4912          */
4913         if ((lun->flags & CTL_LUN_DISABLED) == 0) {
4914                 mtx_unlock(&lun->lun_lock);
4915                 return (-1);
4916         }
4917         /*
4918          * Mark the LUN invalid.
4919          */
4920         lun->flags |= CTL_LUN_INVALID;
4921
4922         /*
4923          * If there is nothing in the OOA queue, go ahead and free the LUN.
4924          * If we have something in the OOA queue, we'll free it when the
4925          * last I/O completes.
4926          */
4927         if (TAILQ_EMPTY(&lun->ooa_queue)) {
4928                 mtx_unlock(&lun->lun_lock);
4929                 mtx_lock(&softc->ctl_lock);
4930                 ctl_free_lun(lun);
4931                 mtx_unlock(&softc->ctl_lock);
4932         } else
4933                 mtx_unlock(&lun->lun_lock);
4934
4935         return (0);
4936 }
4937
4938 void
4939 ctl_lun_capacity_changed(struct ctl_be_lun *be_lun)
4940 {
4941         struct ctl_lun *lun = (struct ctl_lun *)be_lun->ctl_lun;
4942         union ctl_ha_msg msg;
4943
4944         mtx_lock(&lun->lun_lock);
4945         ctl_est_ua_all(lun, -1, CTL_UA_CAPACITY_CHANGE);
4946         mtx_unlock(&lun->lun_lock);
4947         if (lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
4948                 /* Send msg to other side. */
4949                 bzero(&msg.ua, sizeof(msg.ua));
4950                 msg.hdr.msg_type = CTL_MSG_UA;
4951                 msg.hdr.nexus.initid = -1;
4952                 msg.hdr.nexus.targ_port = -1;
4953                 msg.hdr.nexus.targ_lun = lun->lun;
4954                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
4955                 msg.ua.ua_all = 1;
4956                 msg.ua.ua_set = 1;
4957                 msg.ua.ua_type = CTL_UA_CAPACITY_CHANGE;
4958                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg, sizeof(msg.ua),
4959                     M_WAITOK);
4960         }
4961 }
4962
4963 /*
4964  * Backend "memory move is complete" callback for requests that never
4965  * make it down to say RAIDCore's configuration code.
4966  */
4967 int
4968 ctl_config_move_done(union ctl_io *io)
4969 {
4970         int retval;
4971
4972         CTL_DEBUG_PRINT(("ctl_config_move_done\n"));
4973         KASSERT(io->io_hdr.io_type == CTL_IO_SCSI,
4974             ("Config I/O type isn't CTL_IO_SCSI (%d)!", io->io_hdr.io_type));
4975
4976         if ((io->io_hdr.port_status != 0) &&
4977             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
4978              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
4979                 /*
4980                  * For hardware error sense keys, the sense key
4981                  * specific value is defined to be a retry count,
4982                  * but we use it to pass back an internal FETD
4983                  * error code.  XXX KDM  Hopefully the FETD is only
4984                  * using 16 bits for an error code, since that's
4985                  * all the space we have in the sks field.
4986                  */
4987                 ctl_set_internal_failure(&io->scsiio,
4988                                          /*sks_valid*/ 1,
4989                                          /*retry_count*/
4990                                          io->io_hdr.port_status);
4991         }
4992
4993         if (ctl_debug & CTL_DEBUG_CDB_DATA)
4994                 ctl_data_print(io);
4995         if (((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN) ||
4996             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
4997              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) ||
4998             ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0)) {
4999                 /*
5000                  * XXX KDM just assuming a single pointer here, and not a
5001                  * S/G list.  If we start using S/G lists for config data,
5002                  * we'll need to know how to clean them up here as well.
5003                  */
5004                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5005                         free(io->scsiio.kern_data_ptr, M_CTL);
5006                 ctl_done(io);
5007                 retval = CTL_RETVAL_COMPLETE;
5008         } else {
5009                 /*
5010                  * XXX KDM now we need to continue data movement.  Some
5011                  * options:
5012                  * - call ctl_scsiio() again?  We don't do this for data
5013                  *   writes, because for those at least we know ahead of
5014                  *   time where the write will go and how long it is.  For
5015                  *   config writes, though, that information is largely
5016                  *   contained within the write itself, thus we need to
5017                  *   parse out the data again.
5018                  *
5019                  * - Call some other function once the data is in?
5020                  */
5021
5022                 /*
5023                  * XXX KDM call ctl_scsiio() again for now, and check flag
5024                  * bits to see whether we're allocated or not.
5025                  */
5026                 retval = ctl_scsiio(&io->scsiio);
5027         }
5028         return (retval);
5029 }
5030
5031 /*
5032  * This gets called by a backend driver when it is done with a
5033  * data_submit method.
5034  */
5035 void
5036 ctl_data_submit_done(union ctl_io *io)
5037 {
5038         /*
5039          * If the IO_CONT flag is set, we need to call the supplied
5040          * function to continue processing the I/O, instead of completing
5041          * the I/O just yet.
5042          *
5043          * If there is an error, though, we don't want to keep processing.
5044          * Instead, just send status back to the initiator.
5045          */
5046         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5047             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5048             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5049              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5050                 io->scsiio.io_cont(io);
5051                 return;
5052         }
5053         ctl_done(io);
5054 }
5055
5056 /*
5057  * This gets called by a backend driver when it is done with a
5058  * configuration write.
5059  */
5060 void
5061 ctl_config_write_done(union ctl_io *io)
5062 {
5063         uint8_t *buf;
5064
5065         /*
5066          * If the IO_CONT flag is set, we need to call the supplied
5067          * function to continue processing the I/O, instead of completing
5068          * the I/O just yet.
5069          *
5070          * If there is an error, though, we don't want to keep processing.
5071          * Instead, just send status back to the initiator.
5072          */
5073         if ((io->io_hdr.flags & CTL_FLAG_IO_CONT) &&
5074             (io->io_hdr.flags & CTL_FLAG_ABORT) == 0 &&
5075             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_STATUS_NONE ||
5076              (io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS)) {
5077                 io->scsiio.io_cont(io);
5078                 return;
5079         }
5080         /*
5081          * Since a configuration write can be done for commands that actually
5082          * have data allocated, like write buffer, and commands that have
5083          * no data, like start/stop unit, we need to check here.
5084          */
5085         if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5086                 buf = io->scsiio.kern_data_ptr;
5087         else
5088                 buf = NULL;
5089         ctl_done(io);
5090         if (buf)
5091                 free(buf, M_CTL);
5092 }
5093
5094 void
5095 ctl_config_read_done(union ctl_io *io)
5096 {
5097         uint8_t *buf;
5098
5099         /*
5100          * If there is some error -- we are done, skip data transfer.
5101          */
5102         if ((io->io_hdr.flags & CTL_FLAG_ABORT) != 0 ||
5103             ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
5104              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS)) {
5105                 if (io->io_hdr.flags & CTL_FLAG_ALLOCATED)
5106                         buf = io->scsiio.kern_data_ptr;
5107                 else
5108                         buf = NULL;
5109                 ctl_done(io);
5110                 if (buf)
5111                         free(buf, M_CTL);
5112                 return;
5113         }
5114
5115         /*
5116          * If the IO_CONT flag is set, we need to call the supplied
5117          * function to continue processing the I/O, instead of completing
5118          * the I/O just yet.
5119          */
5120         if (io->io_hdr.flags & CTL_FLAG_IO_CONT) {
5121                 io->scsiio.io_cont(io);
5122                 return;
5123         }
5124
5125         ctl_datamove(io);
5126 }
5127
5128 /*
5129  * SCSI release command.
5130  */
5131 int
5132 ctl_scsi_release(struct ctl_scsiio *ctsio)
5133 {
5134         struct ctl_lun *lun;
5135         uint32_t residx;
5136
5137         CTL_DEBUG_PRINT(("ctl_scsi_release\n"));
5138
5139         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5140         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5141
5142         /*
5143          * XXX KDM right now, we only support LUN reservation.  We don't
5144          * support 3rd party reservations, or extent reservations, which
5145          * might actually need the parameter list.  If we've gotten this
5146          * far, we've got a LUN reservation.  Anything else got kicked out
5147          * above.  So, according to SPC, ignore the length.
5148          */
5149
5150         mtx_lock(&lun->lun_lock);
5151
5152         /*
5153          * According to SPC, it is not an error for an intiator to attempt
5154          * to release a reservation on a LUN that isn't reserved, or that
5155          * is reserved by another initiator.  The reservation can only be
5156          * released, though, by the initiator who made it or by one of
5157          * several reset type events.
5158          */
5159         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == residx))
5160                         lun->flags &= ~CTL_LUN_RESERVED;
5161
5162         mtx_unlock(&lun->lun_lock);
5163
5164         ctl_set_success(ctsio);
5165         ctl_done((union ctl_io *)ctsio);
5166         return (CTL_RETVAL_COMPLETE);
5167 }
5168
5169 int
5170 ctl_scsi_reserve(struct ctl_scsiio *ctsio)
5171 {
5172         struct ctl_lun *lun;
5173         uint32_t residx;
5174
5175         CTL_DEBUG_PRINT(("ctl_reserve\n"));
5176
5177         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5178         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5179
5180         /*
5181          * XXX KDM right now, we only support LUN reservation.  We don't
5182          * support 3rd party reservations, or extent reservations, which
5183          * might actually need the parameter list.  If we've gotten this
5184          * far, we've got a LUN reservation.  Anything else got kicked out
5185          * above.  So, according to SPC, ignore the length.
5186          */
5187
5188         mtx_lock(&lun->lun_lock);
5189         if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx != residx)) {
5190                 ctl_set_reservation_conflict(ctsio);
5191                 goto bailout;
5192         }
5193
5194         /* SPC-3 exceptions to SPC-2 RESERVE and RELEASE behavior. */
5195         if (lun->flags & CTL_LUN_PR_RESERVED) {
5196                 ctl_set_success(ctsio);
5197                 goto bailout;
5198         }
5199
5200         lun->flags |= CTL_LUN_RESERVED;
5201         lun->res_idx = residx;
5202         ctl_set_success(ctsio);
5203
5204 bailout:
5205         mtx_unlock(&lun->lun_lock);
5206         ctl_done((union ctl_io *)ctsio);
5207         return (CTL_RETVAL_COMPLETE);
5208 }
5209
5210 int
5211 ctl_start_stop(struct ctl_scsiio *ctsio)
5212 {
5213         struct scsi_start_stop_unit *cdb;
5214         struct ctl_lun *lun;
5215         int retval;
5216
5217         CTL_DEBUG_PRINT(("ctl_start_stop\n"));
5218
5219         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5220         cdb = (struct scsi_start_stop_unit *)ctsio->cdb;
5221
5222         if ((cdb->how & SSS_PC_MASK) == 0) {
5223                 if ((lun->flags & CTL_LUN_PR_RESERVED) &&
5224                     (cdb->how & SSS_START) == 0) {
5225                         uint32_t residx;
5226
5227                         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5228                         if (ctl_get_prkey(lun, residx) == 0 ||
5229                             (lun->pr_res_idx != residx && lun->pr_res_type < 4)) {
5230
5231                                 ctl_set_reservation_conflict(ctsio);
5232                                 ctl_done((union ctl_io *)ctsio);
5233                                 return (CTL_RETVAL_COMPLETE);
5234                         }
5235                 }
5236
5237                 if ((cdb->how & SSS_LOEJ) &&
5238                     (lun->flags & CTL_LUN_REMOVABLE) == 0) {
5239                         ctl_set_invalid_field(ctsio,
5240                                               /*sks_valid*/ 1,
5241                                               /*command*/ 1,
5242                                               /*field*/ 4,
5243                                               /*bit_valid*/ 1,
5244                                               /*bit*/ 1);
5245                         ctl_done((union ctl_io *)ctsio);
5246                         return (CTL_RETVAL_COMPLETE);
5247                 }
5248
5249                 if ((cdb->how & SSS_START) == 0 && (cdb->how & SSS_LOEJ) &&
5250                     lun->prevent_count > 0) {
5251                         /* "Medium removal prevented" */
5252                         ctl_set_sense(ctsio, /*current_error*/ 1,
5253                             /*sense_key*/(lun->flags & CTL_LUN_NO_MEDIA) ?
5254                              SSD_KEY_NOT_READY : SSD_KEY_ILLEGAL_REQUEST,
5255                             /*asc*/ 0x53, /*ascq*/ 0x02, SSD_ELEM_NONE);
5256                         ctl_done((union ctl_io *)ctsio);
5257                         return (CTL_RETVAL_COMPLETE);
5258                 }
5259         }
5260
5261         retval = lun->backend->config_write((union ctl_io *)ctsio);
5262         return (retval);
5263 }
5264
5265 int
5266 ctl_prevent_allow(struct ctl_scsiio *ctsio)
5267 {
5268         struct ctl_lun *lun;
5269         struct scsi_prevent *cdb;
5270         int retval;
5271         uint32_t initidx;
5272
5273         CTL_DEBUG_PRINT(("ctl_prevent_allow\n"));
5274
5275         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5276         cdb = (struct scsi_prevent *)ctsio->cdb;
5277
5278         if ((lun->flags & CTL_LUN_REMOVABLE) == 0) {
5279                 ctl_set_invalid_opcode(ctsio);
5280                 ctl_done((union ctl_io *)ctsio);
5281                 return (CTL_RETVAL_COMPLETE);
5282         }
5283
5284         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5285         mtx_lock(&lun->lun_lock);
5286         if ((cdb->how & PR_PREVENT) &&
5287             ctl_is_set(lun->prevent, initidx) == 0) {
5288                 ctl_set_mask(lun->prevent, initidx);
5289                 lun->prevent_count++;
5290         } else if ((cdb->how & PR_PREVENT) == 0 &&
5291             ctl_is_set(lun->prevent, initidx)) {
5292                 ctl_clear_mask(lun->prevent, initidx);
5293                 lun->prevent_count--;
5294         }
5295         mtx_unlock(&lun->lun_lock);
5296         retval = lun->backend->config_write((union ctl_io *)ctsio);
5297         return (retval);
5298 }
5299
5300 /*
5301  * We support the SYNCHRONIZE CACHE command (10 and 16 byte versions), but
5302  * we don't really do anything with the LBA and length fields if the user
5303  * passes them in.  Instead we'll just flush out the cache for the entire
5304  * LUN.
5305  */
5306 int
5307 ctl_sync_cache(struct ctl_scsiio *ctsio)
5308 {
5309         struct ctl_lun *lun;
5310         struct ctl_softc *softc;
5311         struct ctl_lba_len_flags *lbalen;
5312         uint64_t starting_lba;
5313         uint32_t block_count;
5314         int retval;
5315         uint8_t byte2;
5316
5317         CTL_DEBUG_PRINT(("ctl_sync_cache\n"));
5318
5319         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5320         softc = lun->ctl_softc;
5321         retval = 0;
5322
5323         switch (ctsio->cdb[0]) {
5324         case SYNCHRONIZE_CACHE: {
5325                 struct scsi_sync_cache *cdb;
5326                 cdb = (struct scsi_sync_cache *)ctsio->cdb;
5327
5328                 starting_lba = scsi_4btoul(cdb->begin_lba);
5329                 block_count = scsi_2btoul(cdb->lb_count);
5330                 byte2 = cdb->byte2;
5331                 break;
5332         }
5333         case SYNCHRONIZE_CACHE_16: {
5334                 struct scsi_sync_cache_16 *cdb;
5335                 cdb = (struct scsi_sync_cache_16 *)ctsio->cdb;
5336
5337                 starting_lba = scsi_8btou64(cdb->begin_lba);
5338                 block_count = scsi_4btoul(cdb->lb_count);
5339                 byte2 = cdb->byte2;
5340                 break;
5341         }
5342         default:
5343                 ctl_set_invalid_opcode(ctsio);
5344                 ctl_done((union ctl_io *)ctsio);
5345                 goto bailout;
5346                 break; /* NOTREACHED */
5347         }
5348
5349         /*
5350          * We check the LBA and length, but don't do anything with them.
5351          * A SYNCHRONIZE CACHE will cause the entire cache for this lun to
5352          * get flushed.  This check will just help satisfy anyone who wants
5353          * to see an error for an out of range LBA.
5354          */
5355         if ((starting_lba + block_count) > (lun->be_lun->maxlba + 1)) {
5356                 ctl_set_lba_out_of_range(ctsio);
5357                 ctl_done((union ctl_io *)ctsio);
5358                 goto bailout;
5359         }
5360
5361         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5362         lbalen->lba = starting_lba;
5363         lbalen->len = block_count;
5364         lbalen->flags = byte2;
5365         retval = lun->backend->config_write((union ctl_io *)ctsio);
5366
5367 bailout:
5368         return (retval);
5369 }
5370
5371 int
5372 ctl_format(struct ctl_scsiio *ctsio)
5373 {
5374         struct scsi_format *cdb;
5375         struct ctl_lun *lun;
5376         int length, defect_list_len;
5377
5378         CTL_DEBUG_PRINT(("ctl_format\n"));
5379
5380         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5381
5382         cdb = (struct scsi_format *)ctsio->cdb;
5383
5384         length = 0;
5385         if (cdb->byte2 & SF_FMTDATA) {
5386                 if (cdb->byte2 & SF_LONGLIST)
5387                         length = sizeof(struct scsi_format_header_long);
5388                 else
5389                         length = sizeof(struct scsi_format_header_short);
5390         }
5391
5392         if (((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0)
5393          && (length > 0)) {
5394                 ctsio->kern_data_ptr = malloc(length, M_CTL, M_WAITOK);
5395                 ctsio->kern_data_len = length;
5396                 ctsio->kern_total_len = length;
5397                 ctsio->kern_data_resid = 0;
5398                 ctsio->kern_rel_offset = 0;
5399                 ctsio->kern_sg_entries = 0;
5400                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5401                 ctsio->be_move_done = ctl_config_move_done;
5402                 ctl_datamove((union ctl_io *)ctsio);
5403
5404                 return (CTL_RETVAL_COMPLETE);
5405         }
5406
5407         defect_list_len = 0;
5408
5409         if (cdb->byte2 & SF_FMTDATA) {
5410                 if (cdb->byte2 & SF_LONGLIST) {
5411                         struct scsi_format_header_long *header;
5412
5413                         header = (struct scsi_format_header_long *)
5414                                 ctsio->kern_data_ptr;
5415
5416                         defect_list_len = scsi_4btoul(header->defect_list_len);
5417                         if (defect_list_len != 0) {
5418                                 ctl_set_invalid_field(ctsio,
5419                                                       /*sks_valid*/ 1,
5420                                                       /*command*/ 0,
5421                                                       /*field*/ 2,
5422                                                       /*bit_valid*/ 0,
5423                                                       /*bit*/ 0);
5424                                 goto bailout;
5425                         }
5426                 } else {
5427                         struct scsi_format_header_short *header;
5428
5429                         header = (struct scsi_format_header_short *)
5430                                 ctsio->kern_data_ptr;
5431
5432                         defect_list_len = scsi_2btoul(header->defect_list_len);
5433                         if (defect_list_len != 0) {
5434                                 ctl_set_invalid_field(ctsio,
5435                                                       /*sks_valid*/ 1,
5436                                                       /*command*/ 0,
5437                                                       /*field*/ 2,
5438                                                       /*bit_valid*/ 0,
5439                                                       /*bit*/ 0);
5440                                 goto bailout;
5441                         }
5442                 }
5443         }
5444
5445         ctl_set_success(ctsio);
5446 bailout:
5447
5448         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5449                 free(ctsio->kern_data_ptr, M_CTL);
5450                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5451         }
5452
5453         ctl_done((union ctl_io *)ctsio);
5454         return (CTL_RETVAL_COMPLETE);
5455 }
5456
5457 int
5458 ctl_read_buffer(struct ctl_scsiio *ctsio)
5459 {
5460         struct ctl_lun *lun;
5461         uint64_t buffer_offset;
5462         uint32_t len;
5463         uint8_t byte2;
5464         static uint8_t descr[4];
5465         static uint8_t echo_descr[4] = { 0 };
5466
5467         CTL_DEBUG_PRINT(("ctl_read_buffer\n"));
5468         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5469         switch (ctsio->cdb[0]) {
5470         case READ_BUFFER: {
5471                 struct scsi_read_buffer *cdb;
5472
5473                 cdb = (struct scsi_read_buffer *)ctsio->cdb;
5474                 buffer_offset = scsi_3btoul(cdb->offset);
5475                 len = scsi_3btoul(cdb->length);
5476                 byte2 = cdb->byte2;
5477                 break;
5478         }
5479         case READ_BUFFER_16: {
5480                 struct scsi_read_buffer_16 *cdb;
5481
5482                 cdb = (struct scsi_read_buffer_16 *)ctsio->cdb;
5483                 buffer_offset = scsi_8btou64(cdb->offset);
5484                 len = scsi_4btoul(cdb->length);
5485                 byte2 = cdb->byte2;
5486                 break;
5487         }
5488         default: /* This shouldn't happen. */
5489                 ctl_set_invalid_opcode(ctsio);
5490                 ctl_done((union ctl_io *)ctsio);
5491                 return (CTL_RETVAL_COMPLETE);
5492         }
5493
5494         if ((byte2 & RWB_MODE) != RWB_MODE_DATA &&
5495             (byte2 & RWB_MODE) != RWB_MODE_ECHO_DESCR &&
5496             (byte2 & RWB_MODE) != RWB_MODE_DESCR) {
5497                 ctl_set_invalid_field(ctsio,
5498                                       /*sks_valid*/ 1,
5499                                       /*command*/ 1,
5500                                       /*field*/ 1,
5501                                       /*bit_valid*/ 1,
5502                                       /*bit*/ 4);
5503                 ctl_done((union ctl_io *)ctsio);
5504                 return (CTL_RETVAL_COMPLETE);
5505         }
5506
5507         if (buffer_offset > CTL_WRITE_BUFFER_SIZE ||
5508             buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5509                 ctl_set_invalid_field(ctsio,
5510                                       /*sks_valid*/ 1,
5511                                       /*command*/ 1,
5512                                       /*field*/ 6,
5513                                       /*bit_valid*/ 0,
5514                                       /*bit*/ 0);
5515                 ctl_done((union ctl_io *)ctsio);
5516                 return (CTL_RETVAL_COMPLETE);
5517         }
5518
5519         if ((byte2 & RWB_MODE) == RWB_MODE_DESCR) {
5520                 descr[0] = 0;
5521                 scsi_ulto3b(CTL_WRITE_BUFFER_SIZE, &descr[1]);
5522                 ctsio->kern_data_ptr = descr;
5523                 len = min(len, sizeof(descr));
5524         } else if ((byte2 & RWB_MODE) == RWB_MODE_ECHO_DESCR) {
5525                 ctsio->kern_data_ptr = echo_descr;
5526                 len = min(len, sizeof(echo_descr));
5527         } else {
5528                 if (lun->write_buffer == NULL) {
5529                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5530                             M_CTL, M_WAITOK);
5531                 }
5532                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5533         }
5534         ctsio->kern_data_len = len;
5535         ctsio->kern_total_len = len;
5536         ctsio->kern_data_resid = 0;
5537         ctsio->kern_rel_offset = 0;
5538         ctsio->kern_sg_entries = 0;
5539         ctl_set_success(ctsio);
5540         ctsio->be_move_done = ctl_config_move_done;
5541         ctl_datamove((union ctl_io *)ctsio);
5542         return (CTL_RETVAL_COMPLETE);
5543 }
5544
5545 int
5546 ctl_write_buffer(struct ctl_scsiio *ctsio)
5547 {
5548         struct scsi_write_buffer *cdb;
5549         struct ctl_lun *lun;
5550         int buffer_offset, len;
5551
5552         CTL_DEBUG_PRINT(("ctl_write_buffer\n"));
5553
5554         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5555         cdb = (struct scsi_write_buffer *)ctsio->cdb;
5556
5557         if ((cdb->byte2 & RWB_MODE) != RWB_MODE_DATA) {
5558                 ctl_set_invalid_field(ctsio,
5559                                       /*sks_valid*/ 1,
5560                                       /*command*/ 1,
5561                                       /*field*/ 1,
5562                                       /*bit_valid*/ 1,
5563                                       /*bit*/ 4);
5564                 ctl_done((union ctl_io *)ctsio);
5565                 return (CTL_RETVAL_COMPLETE);
5566         }
5567
5568         len = scsi_3btoul(cdb->length);
5569         buffer_offset = scsi_3btoul(cdb->offset);
5570
5571         if (buffer_offset + len > CTL_WRITE_BUFFER_SIZE) {
5572                 ctl_set_invalid_field(ctsio,
5573                                       /*sks_valid*/ 1,
5574                                       /*command*/ 1,
5575                                       /*field*/ 6,
5576                                       /*bit_valid*/ 0,
5577                                       /*bit*/ 0);
5578                 ctl_done((union ctl_io *)ctsio);
5579                 return (CTL_RETVAL_COMPLETE);
5580         }
5581
5582         /*
5583          * If we've got a kernel request that hasn't been malloced yet,
5584          * malloc it and tell the caller the data buffer is here.
5585          */
5586         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5587                 if (lun->write_buffer == NULL) {
5588                         lun->write_buffer = malloc(CTL_WRITE_BUFFER_SIZE,
5589                             M_CTL, M_WAITOK);
5590                 }
5591                 ctsio->kern_data_ptr = lun->write_buffer + buffer_offset;
5592                 ctsio->kern_data_len = len;
5593                 ctsio->kern_total_len = len;
5594                 ctsio->kern_data_resid = 0;
5595                 ctsio->kern_rel_offset = 0;
5596                 ctsio->kern_sg_entries = 0;
5597                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5598                 ctsio->be_move_done = ctl_config_move_done;
5599                 ctl_datamove((union ctl_io *)ctsio);
5600
5601                 return (CTL_RETVAL_COMPLETE);
5602         }
5603
5604         ctl_set_success(ctsio);
5605         ctl_done((union ctl_io *)ctsio);
5606         return (CTL_RETVAL_COMPLETE);
5607 }
5608
5609 int
5610 ctl_write_same(struct ctl_scsiio *ctsio)
5611 {
5612         struct ctl_lun *lun;
5613         struct ctl_lba_len_flags *lbalen;
5614         uint64_t lba;
5615         uint32_t num_blocks;
5616         int len, retval;
5617         uint8_t byte2;
5618
5619         CTL_DEBUG_PRINT(("ctl_write_same\n"));
5620
5621         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5622
5623         switch (ctsio->cdb[0]) {
5624         case WRITE_SAME_10: {
5625                 struct scsi_write_same_10 *cdb;
5626
5627                 cdb = (struct scsi_write_same_10 *)ctsio->cdb;
5628
5629                 lba = scsi_4btoul(cdb->addr);
5630                 num_blocks = scsi_2btoul(cdb->length);
5631                 byte2 = cdb->byte2;
5632                 break;
5633         }
5634         case WRITE_SAME_16: {
5635                 struct scsi_write_same_16 *cdb;
5636
5637                 cdb = (struct scsi_write_same_16 *)ctsio->cdb;
5638
5639                 lba = scsi_8btou64(cdb->addr);
5640                 num_blocks = scsi_4btoul(cdb->length);
5641                 byte2 = cdb->byte2;
5642                 break;
5643         }
5644         default:
5645                 /*
5646                  * We got a command we don't support.  This shouldn't
5647                  * happen, commands should be filtered out above us.
5648                  */
5649                 ctl_set_invalid_opcode(ctsio);
5650                 ctl_done((union ctl_io *)ctsio);
5651
5652                 return (CTL_RETVAL_COMPLETE);
5653                 break; /* NOTREACHED */
5654         }
5655
5656         /* ANCHOR flag can be used only together with UNMAP */
5657         if ((byte2 & SWS_UNMAP) == 0 && (byte2 & SWS_ANCHOR) != 0) {
5658                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
5659                     /*command*/ 1, /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
5660                 ctl_done((union ctl_io *)ctsio);
5661                 return (CTL_RETVAL_COMPLETE);
5662         }
5663
5664         /*
5665          * The first check is to make sure we're in bounds, the second
5666          * check is to catch wrap-around problems.  If the lba + num blocks
5667          * is less than the lba, then we've wrapped around and the block
5668          * range is invalid anyway.
5669          */
5670         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5671          || ((lba + num_blocks) < lba)) {
5672                 ctl_set_lba_out_of_range(ctsio);
5673                 ctl_done((union ctl_io *)ctsio);
5674                 return (CTL_RETVAL_COMPLETE);
5675         }
5676
5677         /* Zero number of blocks means "to the last logical block" */
5678         if (num_blocks == 0) {
5679                 if ((lun->be_lun->maxlba + 1) - lba > UINT32_MAX) {
5680                         ctl_set_invalid_field(ctsio,
5681                                               /*sks_valid*/ 0,
5682                                               /*command*/ 1,
5683                                               /*field*/ 0,
5684                                               /*bit_valid*/ 0,
5685                                               /*bit*/ 0);
5686                         ctl_done((union ctl_io *)ctsio);
5687                         return (CTL_RETVAL_COMPLETE);
5688                 }
5689                 num_blocks = (lun->be_lun->maxlba + 1) - lba;
5690         }
5691
5692         len = lun->be_lun->blocksize;
5693
5694         /*
5695          * If we've got a kernel request that hasn't been malloced yet,
5696          * malloc it and tell the caller the data buffer is here.
5697          */
5698         if ((byte2 & SWS_NDOB) == 0 &&
5699             (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5700                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5701                 ctsio->kern_data_len = len;
5702                 ctsio->kern_total_len = len;
5703                 ctsio->kern_data_resid = 0;
5704                 ctsio->kern_rel_offset = 0;
5705                 ctsio->kern_sg_entries = 0;
5706                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5707                 ctsio->be_move_done = ctl_config_move_done;
5708                 ctl_datamove((union ctl_io *)ctsio);
5709
5710                 return (CTL_RETVAL_COMPLETE);
5711         }
5712
5713         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5714         lbalen->lba = lba;
5715         lbalen->len = num_blocks;
5716         lbalen->flags = byte2;
5717         retval = lun->backend->config_write((union ctl_io *)ctsio);
5718
5719         return (retval);
5720 }
5721
5722 int
5723 ctl_unmap(struct ctl_scsiio *ctsio)
5724 {
5725         struct ctl_lun *lun;
5726         struct scsi_unmap *cdb;
5727         struct ctl_ptr_len_flags *ptrlen;
5728         struct scsi_unmap_header *hdr;
5729         struct scsi_unmap_desc *buf, *end, *endnz, *range;
5730         uint64_t lba;
5731         uint32_t num_blocks;
5732         int len, retval;
5733         uint8_t byte2;
5734
5735         CTL_DEBUG_PRINT(("ctl_unmap\n"));
5736
5737         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5738         cdb = (struct scsi_unmap *)ctsio->cdb;
5739
5740         len = scsi_2btoul(cdb->length);
5741         byte2 = cdb->byte2;
5742
5743         /*
5744          * If we've got a kernel request that hasn't been malloced yet,
5745          * malloc it and tell the caller the data buffer is here.
5746          */
5747         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
5748                 ctsio->kern_data_ptr = malloc(len, M_CTL, M_WAITOK);
5749                 ctsio->kern_data_len = len;
5750                 ctsio->kern_total_len = len;
5751                 ctsio->kern_data_resid = 0;
5752                 ctsio->kern_rel_offset = 0;
5753                 ctsio->kern_sg_entries = 0;
5754                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
5755                 ctsio->be_move_done = ctl_config_move_done;
5756                 ctl_datamove((union ctl_io *)ctsio);
5757
5758                 return (CTL_RETVAL_COMPLETE);
5759         }
5760
5761         len = ctsio->kern_total_len - ctsio->kern_data_resid;
5762         hdr = (struct scsi_unmap_header *)ctsio->kern_data_ptr;
5763         if (len < sizeof (*hdr) ||
5764             len < (scsi_2btoul(hdr->length) + sizeof(hdr->length)) ||
5765             len < (scsi_2btoul(hdr->desc_length) + sizeof (*hdr)) ||
5766             scsi_2btoul(hdr->desc_length) % sizeof(*buf) != 0) {
5767                 ctl_set_invalid_field(ctsio,
5768                                       /*sks_valid*/ 0,
5769                                       /*command*/ 0,
5770                                       /*field*/ 0,
5771                                       /*bit_valid*/ 0,
5772                                       /*bit*/ 0);
5773                 goto done;
5774         }
5775         len = scsi_2btoul(hdr->desc_length);
5776         buf = (struct scsi_unmap_desc *)(hdr + 1);
5777         end = buf + len / sizeof(*buf);
5778
5779         endnz = buf;
5780         for (range = buf; range < end; range++) {
5781                 lba = scsi_8btou64(range->lba);
5782                 num_blocks = scsi_4btoul(range->length);
5783                 if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
5784                  || ((lba + num_blocks) < lba)) {
5785                         ctl_set_lba_out_of_range(ctsio);
5786                         ctl_done((union ctl_io *)ctsio);
5787                         return (CTL_RETVAL_COMPLETE);
5788                 }
5789                 if (num_blocks != 0)
5790                         endnz = range + 1;
5791         }
5792
5793         /*
5794          * Block backend can not handle zero last range.
5795          * Filter it out and return if there is nothing left.
5796          */
5797         len = (uint8_t *)endnz - (uint8_t *)buf;
5798         if (len == 0) {
5799                 ctl_set_success(ctsio);
5800                 goto done;
5801         }
5802
5803         mtx_lock(&lun->lun_lock);
5804         ptrlen = (struct ctl_ptr_len_flags *)
5805             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
5806         ptrlen->ptr = (void *)buf;
5807         ptrlen->len = len;
5808         ptrlen->flags = byte2;
5809         ctl_check_blocked(lun);
5810         mtx_unlock(&lun->lun_lock);
5811
5812         retval = lun->backend->config_write((union ctl_io *)ctsio);
5813         return (retval);
5814
5815 done:
5816         if (ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) {
5817                 free(ctsio->kern_data_ptr, M_CTL);
5818                 ctsio->io_hdr.flags &= ~CTL_FLAG_ALLOCATED;
5819         }
5820         ctl_done((union ctl_io *)ctsio);
5821         return (CTL_RETVAL_COMPLETE);
5822 }
5823
5824 int
5825 ctl_default_page_handler(struct ctl_scsiio *ctsio,
5826                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5827 {
5828         struct ctl_lun *lun;
5829         uint8_t *current_cp, *saved_cp;
5830         int set_ua;
5831         uint32_t initidx;
5832
5833         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5834         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
5835         set_ua = 0;
5836
5837         current_cp = (page_index->page_data + (page_index->page_len *
5838             CTL_PAGE_CURRENT));
5839         saved_cp = (page_index->page_data + (page_index->page_len *
5840             CTL_PAGE_SAVED));
5841
5842         mtx_lock(&lun->lun_lock);
5843         if (memcmp(current_cp, page_ptr, page_index->page_len)) {
5844                 memcpy(current_cp, page_ptr, page_index->page_len);
5845                 memcpy(saved_cp, page_ptr, page_index->page_len);
5846                 set_ua = 1;
5847         }
5848         if (set_ua != 0)
5849                 ctl_est_ua_all(lun, initidx, CTL_UA_MODE_CHANGE);
5850         mtx_unlock(&lun->lun_lock);
5851         if (set_ua) {
5852                 ctl_isc_announce_mode(lun,
5853                     ctl_get_initindex(&ctsio->io_hdr.nexus),
5854                     page_index->page_code, page_index->subpage);
5855         }
5856         return (CTL_RETVAL_COMPLETE);
5857 }
5858
5859 static void
5860 ctl_ie_timer(void *arg)
5861 {
5862         struct ctl_lun *lun = arg;
5863         struct scsi_info_exceptions_page *pg;
5864         uint64_t t;
5865
5866         if (lun->ie_asc == 0)
5867                 return;
5868
5869         pg = &lun->mode_pages.ie_page[CTL_PAGE_CURRENT];
5870         if (pg->mrie == SIEP_MRIE_UA)
5871                 ctl_est_ua_all(lun, -1, CTL_UA_IE);
5872         else
5873                 lun->ie_reported = 0;
5874
5875         if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5876                 lun->ie_reportcnt++;
5877                 t = scsi_4btoul(pg->interval_timer);
5878                 if (t == 0 || t == UINT32_MAX)
5879                         t = 3000;  /* 5 min */
5880                 callout_schedule(&lun->ie_callout, t * hz / 10);
5881         }
5882 }
5883
5884 int
5885 ctl_ie_page_handler(struct ctl_scsiio *ctsio,
5886                          struct ctl_page_index *page_index, uint8_t *page_ptr)
5887 {
5888         struct scsi_info_exceptions_page *pg;
5889         struct ctl_lun *lun;
5890         uint64_t t;
5891
5892         (void)ctl_default_page_handler(ctsio, page_index, page_ptr);
5893
5894         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5895         pg = (struct scsi_info_exceptions_page *)page_ptr;
5896         mtx_lock(&lun->lun_lock);
5897         if (pg->info_flags & SIEP_FLAGS_TEST) {
5898                 lun->ie_asc = 0x5d;
5899                 lun->ie_ascq = 0xff;
5900                 if (pg->mrie == SIEP_MRIE_UA) {
5901                         ctl_est_ua_all(lun, -1, CTL_UA_IE);
5902                         lun->ie_reported = 1;
5903                 } else {
5904                         ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5905                         lun->ie_reported = -1;
5906                 }
5907                 lun->ie_reportcnt = 1;
5908                 if (lun->ie_reportcnt < scsi_4btoul(pg->report_count)) {
5909                         lun->ie_reportcnt++;
5910                         t = scsi_4btoul(pg->interval_timer);
5911                         if (t == 0 || t == UINT32_MAX)
5912                                 t = 3000;  /* 5 min */
5913                         callout_reset(&lun->ie_callout, t * hz / 10,
5914                             ctl_ie_timer, lun);
5915                 }
5916         } else {
5917                 lun->ie_asc = 0;
5918                 lun->ie_ascq = 0;
5919                 lun->ie_reported = 1;
5920                 ctl_clr_ua_all(lun, -1, CTL_UA_IE);
5921                 lun->ie_reportcnt = UINT32_MAX;
5922                 callout_stop(&lun->ie_callout);
5923         }
5924         mtx_unlock(&lun->lun_lock);
5925         return (CTL_RETVAL_COMPLETE);
5926 }
5927
5928 static int
5929 ctl_do_mode_select(union ctl_io *io)
5930 {
5931         struct scsi_mode_page_header *page_header;
5932         struct ctl_page_index *page_index;
5933         struct ctl_scsiio *ctsio;
5934         int page_len, page_len_offset, page_len_size;
5935         union ctl_modepage_info *modepage_info;
5936         struct ctl_lun *lun;
5937         int *len_left, *len_used;
5938         int retval, i;
5939
5940         ctsio = &io->scsiio;
5941         page_index = NULL;
5942         page_len = 0;
5943         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
5944
5945         modepage_info = (union ctl_modepage_info *)
5946                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
5947         len_left = &modepage_info->header.len_left;
5948         len_used = &modepage_info->header.len_used;
5949
5950 do_next_page:
5951
5952         page_header = (struct scsi_mode_page_header *)
5953                 (ctsio->kern_data_ptr + *len_used);
5954
5955         if (*len_left == 0) {
5956                 free(ctsio->kern_data_ptr, M_CTL);
5957                 ctl_set_success(ctsio);
5958                 ctl_done((union ctl_io *)ctsio);
5959                 return (CTL_RETVAL_COMPLETE);
5960         } else if (*len_left < sizeof(struct scsi_mode_page_header)) {
5961
5962                 free(ctsio->kern_data_ptr, M_CTL);
5963                 ctl_set_param_len_error(ctsio);
5964                 ctl_done((union ctl_io *)ctsio);
5965                 return (CTL_RETVAL_COMPLETE);
5966
5967         } else if ((page_header->page_code & SMPH_SPF)
5968                 && (*len_left < sizeof(struct scsi_mode_page_header_sp))) {
5969
5970                 free(ctsio->kern_data_ptr, M_CTL);
5971                 ctl_set_param_len_error(ctsio);
5972                 ctl_done((union ctl_io *)ctsio);
5973                 return (CTL_RETVAL_COMPLETE);
5974         }
5975
5976
5977         /*
5978          * XXX KDM should we do something with the block descriptor?
5979          */
5980         for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
5981                 page_index = &lun->mode_pages.index[i];
5982                 if (lun->be_lun->lun_type == T_DIRECT &&
5983                     (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
5984                         continue;
5985                 if (lun->be_lun->lun_type == T_PROCESSOR &&
5986                     (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
5987                         continue;
5988                 if (lun->be_lun->lun_type == T_CDROM &&
5989                     (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
5990                         continue;
5991
5992                 if ((page_index->page_code & SMPH_PC_MASK) !=
5993                     (page_header->page_code & SMPH_PC_MASK))
5994                         continue;
5995
5996                 /*
5997                  * If neither page has a subpage code, then we've got a
5998                  * match.
5999                  */
6000                 if (((page_index->page_code & SMPH_SPF) == 0)
6001                  && ((page_header->page_code & SMPH_SPF) == 0)) {
6002                         page_len = page_header->page_length;
6003                         break;
6004                 }
6005
6006                 /*
6007                  * If both pages have subpages, then the subpage numbers
6008                  * have to match.
6009                  */
6010                 if ((page_index->page_code & SMPH_SPF)
6011                   && (page_header->page_code & SMPH_SPF)) {
6012                         struct scsi_mode_page_header_sp *sph;
6013
6014                         sph = (struct scsi_mode_page_header_sp *)page_header;
6015                         if (page_index->subpage == sph->subpage) {
6016                                 page_len = scsi_2btoul(sph->page_length);
6017                                 break;
6018                         }
6019                 }
6020         }
6021
6022         /*
6023          * If we couldn't find the page, or if we don't have a mode select
6024          * handler for it, send back an error to the user.
6025          */
6026         if ((i >= CTL_NUM_MODE_PAGES)
6027          || (page_index->select_handler == NULL)) {
6028                 ctl_set_invalid_field(ctsio,
6029                                       /*sks_valid*/ 1,
6030                                       /*command*/ 0,
6031                                       /*field*/ *len_used,
6032                                       /*bit_valid*/ 0,
6033                                       /*bit*/ 0);
6034                 free(ctsio->kern_data_ptr, M_CTL);
6035                 ctl_done((union ctl_io *)ctsio);
6036                 return (CTL_RETVAL_COMPLETE);
6037         }
6038
6039         if (page_index->page_code & SMPH_SPF) {
6040                 page_len_offset = 2;
6041                 page_len_size = 2;
6042         } else {
6043                 page_len_size = 1;
6044                 page_len_offset = 1;
6045         }
6046
6047         /*
6048          * If the length the initiator gives us isn't the one we specify in
6049          * the mode page header, or if they didn't specify enough data in
6050          * the CDB to avoid truncating this page, kick out the request.
6051          */
6052         if ((page_len != (page_index->page_len - page_len_offset -
6053                           page_len_size))
6054          || (*len_left < page_index->page_len)) {
6055
6056
6057                 ctl_set_invalid_field(ctsio,
6058                                       /*sks_valid*/ 1,
6059                                       /*command*/ 0,
6060                                       /*field*/ *len_used + page_len_offset,
6061                                       /*bit_valid*/ 0,
6062                                       /*bit*/ 0);
6063                 free(ctsio->kern_data_ptr, M_CTL);
6064                 ctl_done((union ctl_io *)ctsio);
6065                 return (CTL_RETVAL_COMPLETE);
6066         }
6067
6068         /*
6069          * Run through the mode page, checking to make sure that the bits
6070          * the user changed are actually legal for him to change.
6071          */
6072         for (i = 0; i < page_index->page_len; i++) {
6073                 uint8_t *user_byte, *change_mask, *current_byte;
6074                 int bad_bit;
6075                 int j;
6076
6077                 user_byte = (uint8_t *)page_header + i;
6078                 change_mask = page_index->page_data +
6079                               (page_index->page_len * CTL_PAGE_CHANGEABLE) + i;
6080                 current_byte = page_index->page_data +
6081                                (page_index->page_len * CTL_PAGE_CURRENT) + i;
6082
6083                 /*
6084                  * Check to see whether the user set any bits in this byte
6085                  * that he is not allowed to set.
6086                  */
6087                 if ((*user_byte & ~(*change_mask)) ==
6088                     (*current_byte & ~(*change_mask)))
6089                         continue;
6090
6091                 /*
6092                  * Go through bit by bit to determine which one is illegal.
6093                  */
6094                 bad_bit = 0;
6095                 for (j = 7; j >= 0; j--) {
6096                         if ((((1 << i) & ~(*change_mask)) & *user_byte) !=
6097                             (((1 << i) & ~(*change_mask)) & *current_byte)) {
6098                                 bad_bit = i;
6099                                 break;
6100                         }
6101                 }
6102                 ctl_set_invalid_field(ctsio,
6103                                       /*sks_valid*/ 1,
6104                                       /*command*/ 0,
6105                                       /*field*/ *len_used + i,
6106                                       /*bit_valid*/ 1,
6107                                       /*bit*/ bad_bit);
6108                 free(ctsio->kern_data_ptr, M_CTL);
6109                 ctl_done((union ctl_io *)ctsio);
6110                 return (CTL_RETVAL_COMPLETE);
6111         }
6112
6113         /*
6114          * Decrement these before we call the page handler, since we may
6115          * end up getting called back one way or another before the handler
6116          * returns to this context.
6117          */
6118         *len_left -= page_index->page_len;
6119         *len_used += page_index->page_len;
6120
6121         retval = page_index->select_handler(ctsio, page_index,
6122                                             (uint8_t *)page_header);
6123
6124         /*
6125          * If the page handler returns CTL_RETVAL_QUEUED, then we need to
6126          * wait until this queued command completes to finish processing
6127          * the mode page.  If it returns anything other than
6128          * CTL_RETVAL_COMPLETE (e.g. CTL_RETVAL_ERROR), then it should have
6129          * already set the sense information, freed the data pointer, and
6130          * completed the io for us.
6131          */
6132         if (retval != CTL_RETVAL_COMPLETE)
6133                 goto bailout_no_done;
6134
6135         /*
6136          * If the initiator sent us more than one page, parse the next one.
6137          */
6138         if (*len_left > 0)
6139                 goto do_next_page;
6140
6141         ctl_set_success(ctsio);
6142         free(ctsio->kern_data_ptr, M_CTL);
6143         ctl_done((union ctl_io *)ctsio);
6144
6145 bailout_no_done:
6146
6147         return (CTL_RETVAL_COMPLETE);
6148
6149 }
6150
6151 int
6152 ctl_mode_select(struct ctl_scsiio *ctsio)
6153 {
6154         int param_len, pf, sp;
6155         int header_size, bd_len;
6156         union ctl_modepage_info *modepage_info;
6157
6158         switch (ctsio->cdb[0]) {
6159         case MODE_SELECT_6: {
6160                 struct scsi_mode_select_6 *cdb;
6161
6162                 cdb = (struct scsi_mode_select_6 *)ctsio->cdb;
6163
6164                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6165                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6166                 param_len = cdb->length;
6167                 header_size = sizeof(struct scsi_mode_header_6);
6168                 break;
6169         }
6170         case MODE_SELECT_10: {
6171                 struct scsi_mode_select_10 *cdb;
6172
6173                 cdb = (struct scsi_mode_select_10 *)ctsio->cdb;
6174
6175                 pf = (cdb->byte2 & SMS_PF) ? 1 : 0;
6176                 sp = (cdb->byte2 & SMS_SP) ? 1 : 0;
6177                 param_len = scsi_2btoul(cdb->length);
6178                 header_size = sizeof(struct scsi_mode_header_10);
6179                 break;
6180         }
6181         default:
6182                 ctl_set_invalid_opcode(ctsio);
6183                 ctl_done((union ctl_io *)ctsio);
6184                 return (CTL_RETVAL_COMPLETE);
6185         }
6186
6187         /*
6188          * From SPC-3:
6189          * "A parameter list length of zero indicates that the Data-Out Buffer
6190          * shall be empty. This condition shall not be considered as an error."
6191          */
6192         if (param_len == 0) {
6193                 ctl_set_success(ctsio);
6194                 ctl_done((union ctl_io *)ctsio);
6195                 return (CTL_RETVAL_COMPLETE);
6196         }
6197
6198         /*
6199          * Since we'll hit this the first time through, prior to
6200          * allocation, we don't need to free a data buffer here.
6201          */
6202         if (param_len < header_size) {
6203                 ctl_set_param_len_error(ctsio);
6204                 ctl_done((union ctl_io *)ctsio);
6205                 return (CTL_RETVAL_COMPLETE);
6206         }
6207
6208         /*
6209          * Allocate the data buffer and grab the user's data.  In theory,
6210          * we shouldn't have to sanity check the parameter list length here
6211          * because the maximum size is 64K.  We should be able to malloc
6212          * that much without too many problems.
6213          */
6214         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
6215                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
6216                 ctsio->kern_data_len = param_len;
6217                 ctsio->kern_total_len = param_len;
6218                 ctsio->kern_data_resid = 0;
6219                 ctsio->kern_rel_offset = 0;
6220                 ctsio->kern_sg_entries = 0;
6221                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6222                 ctsio->be_move_done = ctl_config_move_done;
6223                 ctl_datamove((union ctl_io *)ctsio);
6224
6225                 return (CTL_RETVAL_COMPLETE);
6226         }
6227
6228         switch (ctsio->cdb[0]) {
6229         case MODE_SELECT_6: {
6230                 struct scsi_mode_header_6 *mh6;
6231
6232                 mh6 = (struct scsi_mode_header_6 *)ctsio->kern_data_ptr;
6233                 bd_len = mh6->blk_desc_len;
6234                 break;
6235         }
6236         case MODE_SELECT_10: {
6237                 struct scsi_mode_header_10 *mh10;
6238
6239                 mh10 = (struct scsi_mode_header_10 *)ctsio->kern_data_ptr;
6240                 bd_len = scsi_2btoul(mh10->blk_desc_len);
6241                 break;
6242         }
6243         default:
6244                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6245         }
6246
6247         if (param_len < (header_size + bd_len)) {
6248                 free(ctsio->kern_data_ptr, M_CTL);
6249                 ctl_set_param_len_error(ctsio);
6250                 ctl_done((union ctl_io *)ctsio);
6251                 return (CTL_RETVAL_COMPLETE);
6252         }
6253
6254         /*
6255          * Set the IO_CONT flag, so that if this I/O gets passed to
6256          * ctl_config_write_done(), it'll get passed back to
6257          * ctl_do_mode_select() for further processing, or completion if
6258          * we're all done.
6259          */
6260         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
6261         ctsio->io_cont = ctl_do_mode_select;
6262
6263         modepage_info = (union ctl_modepage_info *)
6264                 ctsio->io_hdr.ctl_private[CTL_PRIV_MODEPAGE].bytes;
6265         memset(modepage_info, 0, sizeof(*modepage_info));
6266         modepage_info->header.len_left = param_len - header_size - bd_len;
6267         modepage_info->header.len_used = header_size + bd_len;
6268
6269         return (ctl_do_mode_select((union ctl_io *)ctsio));
6270 }
6271
6272 int
6273 ctl_mode_sense(struct ctl_scsiio *ctsio)
6274 {
6275         struct ctl_lun *lun;
6276         int pc, page_code, dbd, llba, subpage;
6277         int alloc_len, page_len, header_len, total_len;
6278         struct scsi_mode_block_descr *block_desc;
6279         struct ctl_page_index *page_index;
6280
6281         dbd = 0;
6282         llba = 0;
6283         block_desc = NULL;
6284
6285         CTL_DEBUG_PRINT(("ctl_mode_sense\n"));
6286
6287         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6288         switch (ctsio->cdb[0]) {
6289         case MODE_SENSE_6: {
6290                 struct scsi_mode_sense_6 *cdb;
6291
6292                 cdb = (struct scsi_mode_sense_6 *)ctsio->cdb;
6293
6294                 header_len = sizeof(struct scsi_mode_hdr_6);
6295                 if (cdb->byte2 & SMS_DBD)
6296                         dbd = 1;
6297                 else
6298                         header_len += sizeof(struct scsi_mode_block_descr);
6299
6300                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6301                 page_code = cdb->page & SMS_PAGE_CODE;
6302                 subpage = cdb->subpage;
6303                 alloc_len = cdb->length;
6304                 break;
6305         }
6306         case MODE_SENSE_10: {
6307                 struct scsi_mode_sense_10 *cdb;
6308
6309                 cdb = (struct scsi_mode_sense_10 *)ctsio->cdb;
6310
6311                 header_len = sizeof(struct scsi_mode_hdr_10);
6312
6313                 if (cdb->byte2 & SMS_DBD)
6314                         dbd = 1;
6315                 else
6316                         header_len += sizeof(struct scsi_mode_block_descr);
6317                 if (cdb->byte2 & SMS10_LLBAA)
6318                         llba = 1;
6319                 pc = (cdb->page & SMS_PAGE_CTRL_MASK) >> 6;
6320                 page_code = cdb->page & SMS_PAGE_CODE;
6321                 subpage = cdb->subpage;
6322                 alloc_len = scsi_2btoul(cdb->length);
6323                 break;
6324         }
6325         default:
6326                 ctl_set_invalid_opcode(ctsio);
6327                 ctl_done((union ctl_io *)ctsio);
6328                 return (CTL_RETVAL_COMPLETE);
6329                 break; /* NOTREACHED */
6330         }
6331
6332         /*
6333          * We have to make a first pass through to calculate the size of
6334          * the pages that match the user's query.  Then we allocate enough
6335          * memory to hold it, and actually copy the data into the buffer.
6336          */
6337         switch (page_code) {
6338         case SMS_ALL_PAGES_PAGE: {
6339                 u_int i;
6340
6341                 page_len = 0;
6342
6343                 /*
6344                  * At the moment, values other than 0 and 0xff here are
6345                  * reserved according to SPC-3.
6346                  */
6347                 if ((subpage != SMS_SUBPAGE_PAGE_0)
6348                  && (subpage != SMS_SUBPAGE_ALL)) {
6349                         ctl_set_invalid_field(ctsio,
6350                                               /*sks_valid*/ 1,
6351                                               /*command*/ 1,
6352                                               /*field*/ 3,
6353                                               /*bit_valid*/ 0,
6354                                               /*bit*/ 0);
6355                         ctl_done((union ctl_io *)ctsio);
6356                         return (CTL_RETVAL_COMPLETE);
6357                 }
6358
6359                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6360                         page_index = &lun->mode_pages.index[i];
6361
6362                         /* Make sure the page is supported for this dev type */
6363                         if (lun->be_lun->lun_type == T_DIRECT &&
6364                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6365                                 continue;
6366                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6367                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6368                                 continue;
6369                         if (lun->be_lun->lun_type == T_CDROM &&
6370                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6371                                 continue;
6372
6373                         /*
6374                          * We don't use this subpage if the user didn't
6375                          * request all subpages.
6376                          */
6377                         if ((page_index->subpage != 0)
6378                          && (subpage == SMS_SUBPAGE_PAGE_0))
6379                                 continue;
6380
6381 #if 0
6382                         printf("found page %#x len %d\n",
6383                                page_index->page_code & SMPH_PC_MASK,
6384                                page_index->page_len);
6385 #endif
6386                         page_len += page_index->page_len;
6387                 }
6388                 break;
6389         }
6390         default: {
6391                 u_int i;
6392
6393                 page_len = 0;
6394
6395                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6396                         page_index = &lun->mode_pages.index[i];
6397
6398                         /* Make sure the page is supported for this dev type */
6399                         if (lun->be_lun->lun_type == T_DIRECT &&
6400                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6401                                 continue;
6402                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6403                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6404                                 continue;
6405                         if (lun->be_lun->lun_type == T_CDROM &&
6406                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6407                                 continue;
6408
6409                         /* Look for the right page code */
6410                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6411                                 continue;
6412
6413                         /* Look for the right subpage or the subpage wildcard*/
6414                         if ((page_index->subpage != subpage)
6415                          && (subpage != SMS_SUBPAGE_ALL))
6416                                 continue;
6417
6418 #if 0
6419                         printf("found page %#x len %d\n",
6420                                page_index->page_code & SMPH_PC_MASK,
6421                                page_index->page_len);
6422 #endif
6423
6424                         page_len += page_index->page_len;
6425                 }
6426
6427                 if (page_len == 0) {
6428                         ctl_set_invalid_field(ctsio,
6429                                               /*sks_valid*/ 1,
6430                                               /*command*/ 1,
6431                                               /*field*/ 2,
6432                                               /*bit_valid*/ 1,
6433                                               /*bit*/ 5);
6434                         ctl_done((union ctl_io *)ctsio);
6435                         return (CTL_RETVAL_COMPLETE);
6436                 }
6437                 break;
6438         }
6439         }
6440
6441         total_len = header_len + page_len;
6442 #if 0
6443         printf("header_len = %d, page_len = %d, total_len = %d\n",
6444                header_len, page_len, total_len);
6445 #endif
6446
6447         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6448         ctsio->kern_sg_entries = 0;
6449         ctsio->kern_data_resid = 0;
6450         ctsio->kern_rel_offset = 0;
6451         if (total_len < alloc_len) {
6452                 ctsio->residual = alloc_len - total_len;
6453                 ctsio->kern_data_len = total_len;
6454                 ctsio->kern_total_len = total_len;
6455         } else {
6456                 ctsio->residual = 0;
6457                 ctsio->kern_data_len = alloc_len;
6458                 ctsio->kern_total_len = alloc_len;
6459         }
6460
6461         switch (ctsio->cdb[0]) {
6462         case MODE_SENSE_6: {
6463                 struct scsi_mode_hdr_6 *header;
6464
6465                 header = (struct scsi_mode_hdr_6 *)ctsio->kern_data_ptr;
6466
6467                 header->datalen = MIN(total_len - 1, 254);
6468                 if (lun->be_lun->lun_type == T_DIRECT) {
6469                         header->dev_specific = 0x10; /* DPOFUA */
6470                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6471                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6472                             .eca_and_aen & SCP_SWP) != 0)
6473                                     header->dev_specific |= 0x80; /* WP */
6474                 }
6475                 if (dbd)
6476                         header->block_descr_len = 0;
6477                 else
6478                         header->block_descr_len =
6479                                 sizeof(struct scsi_mode_block_descr);
6480                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6481                 break;
6482         }
6483         case MODE_SENSE_10: {
6484                 struct scsi_mode_hdr_10 *header;
6485                 int datalen;
6486
6487                 header = (struct scsi_mode_hdr_10 *)ctsio->kern_data_ptr;
6488
6489                 datalen = MIN(total_len - 2, 65533);
6490                 scsi_ulto2b(datalen, header->datalen);
6491                 if (lun->be_lun->lun_type == T_DIRECT) {
6492                         header->dev_specific = 0x10; /* DPOFUA */
6493                         if ((lun->be_lun->flags & CTL_LUN_FLAG_READONLY) ||
6494                             (lun->mode_pages.control_page[CTL_PAGE_CURRENT]
6495                             .eca_and_aen & SCP_SWP) != 0)
6496                                     header->dev_specific |= 0x80; /* WP */
6497                 }
6498                 if (dbd)
6499                         scsi_ulto2b(0, header->block_descr_len);
6500                 else
6501                         scsi_ulto2b(sizeof(struct scsi_mode_block_descr),
6502                                     header->block_descr_len);
6503                 block_desc = (struct scsi_mode_block_descr *)&header[1];
6504                 break;
6505         }
6506         default:
6507                 panic("%s: Invalid CDB type %#x", __func__, ctsio->cdb[0]);
6508         }
6509
6510         /*
6511          * If we've got a disk, use its blocksize in the block
6512          * descriptor.  Otherwise, just set it to 0.
6513          */
6514         if (dbd == 0) {
6515                 if (lun->be_lun->lun_type == T_DIRECT)
6516                         scsi_ulto3b(lun->be_lun->blocksize,
6517                                     block_desc->block_len);
6518                 else
6519                         scsi_ulto3b(0, block_desc->block_len);
6520         }
6521
6522         switch (page_code) {
6523         case SMS_ALL_PAGES_PAGE: {
6524                 int i, data_used;
6525
6526                 data_used = header_len;
6527                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6528                         struct ctl_page_index *page_index;
6529
6530                         page_index = &lun->mode_pages.index[i];
6531                         if (lun->be_lun->lun_type == T_DIRECT &&
6532                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6533                                 continue;
6534                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6535                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6536                                 continue;
6537                         if (lun->be_lun->lun_type == T_CDROM &&
6538                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6539                                 continue;
6540
6541                         /*
6542                          * We don't use this subpage if the user didn't
6543                          * request all subpages.  We already checked (above)
6544                          * to make sure the user only specified a subpage
6545                          * of 0 or 0xff in the SMS_ALL_PAGES_PAGE case.
6546                          */
6547                         if ((page_index->subpage != 0)
6548                          && (subpage == SMS_SUBPAGE_PAGE_0))
6549                                 continue;
6550
6551                         /*
6552                          * Call the handler, if it exists, to update the
6553                          * page to the latest values.
6554                          */
6555                         if (page_index->sense_handler != NULL)
6556                                 page_index->sense_handler(ctsio, page_index,pc);
6557
6558                         memcpy(ctsio->kern_data_ptr + data_used,
6559                                page_index->page_data +
6560                                (page_index->page_len * pc),
6561                                page_index->page_len);
6562                         data_used += page_index->page_len;
6563                 }
6564                 break;
6565         }
6566         default: {
6567                 int i, data_used;
6568
6569                 data_used = header_len;
6570
6571                 for (i = 0; i < CTL_NUM_MODE_PAGES; i++) {
6572                         struct ctl_page_index *page_index;
6573
6574                         page_index = &lun->mode_pages.index[i];
6575
6576                         /* Look for the right page code */
6577                         if ((page_index->page_code & SMPH_PC_MASK) != page_code)
6578                                 continue;
6579
6580                         /* Look for the right subpage or the subpage wildcard*/
6581                         if ((page_index->subpage != subpage)
6582                          && (subpage != SMS_SUBPAGE_ALL))
6583                                 continue;
6584
6585                         /* Make sure the page is supported for this dev type */
6586                         if (lun->be_lun->lun_type == T_DIRECT &&
6587                             (page_index->page_flags & CTL_PAGE_FLAG_DIRECT) == 0)
6588                                 continue;
6589                         if (lun->be_lun->lun_type == T_PROCESSOR &&
6590                             (page_index->page_flags & CTL_PAGE_FLAG_PROC) == 0)
6591                                 continue;
6592                         if (lun->be_lun->lun_type == T_CDROM &&
6593                             (page_index->page_flags & CTL_PAGE_FLAG_CDROM) == 0)
6594                                 continue;
6595
6596                         /*
6597                          * Call the handler, if it exists, to update the
6598                          * page to the latest values.
6599                          */
6600                         if (page_index->sense_handler != NULL)
6601                                 page_index->sense_handler(ctsio, page_index,pc);
6602
6603                         memcpy(ctsio->kern_data_ptr + data_used,
6604                                page_index->page_data +
6605                                (page_index->page_len * pc),
6606                                page_index->page_len);
6607                         data_used += page_index->page_len;
6608                 }
6609                 break;
6610         }
6611         }
6612
6613         ctl_set_success(ctsio);
6614         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6615         ctsio->be_move_done = ctl_config_move_done;
6616         ctl_datamove((union ctl_io *)ctsio);
6617         return (CTL_RETVAL_COMPLETE);
6618 }
6619
6620 int
6621 ctl_lbp_log_sense_handler(struct ctl_scsiio *ctsio,
6622                                struct ctl_page_index *page_index,
6623                                int pc)
6624 {
6625         struct ctl_lun *lun;
6626         struct scsi_log_param_header *phdr;
6627         uint8_t *data;
6628         uint64_t val;
6629
6630         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6631         data = page_index->page_data;
6632
6633         if (lun->backend->lun_attr != NULL &&
6634             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksavail"))
6635              != UINT64_MAX) {
6636                 phdr = (struct scsi_log_param_header *)data;
6637                 scsi_ulto2b(0x0001, phdr->param_code);
6638                 phdr->param_control = SLP_LBIN | SLP_LP;
6639                 phdr->param_len = 8;
6640                 data = (uint8_t *)(phdr + 1);
6641                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6642                 data[4] = 0x02; /* per-pool */
6643                 data += phdr->param_len;
6644         }
6645
6646         if (lun->backend->lun_attr != NULL &&
6647             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "blocksused"))
6648              != UINT64_MAX) {
6649                 phdr = (struct scsi_log_param_header *)data;
6650                 scsi_ulto2b(0x0002, phdr->param_code);
6651                 phdr->param_control = SLP_LBIN | SLP_LP;
6652                 phdr->param_len = 8;
6653                 data = (uint8_t *)(phdr + 1);
6654                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6655                 data[4] = 0x01; /* per-LUN */
6656                 data += phdr->param_len;
6657         }
6658
6659         if (lun->backend->lun_attr != NULL &&
6660             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksavail"))
6661              != UINT64_MAX) {
6662                 phdr = (struct scsi_log_param_header *)data;
6663                 scsi_ulto2b(0x00f1, phdr->param_code);
6664                 phdr->param_control = SLP_LBIN | SLP_LP;
6665                 phdr->param_len = 8;
6666                 data = (uint8_t *)(phdr + 1);
6667                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6668                 data[4] = 0x02; /* per-pool */
6669                 data += phdr->param_len;
6670         }
6671
6672         if (lun->backend->lun_attr != NULL &&
6673             (val = lun->backend->lun_attr(lun->be_lun->be_lun, "poolblocksused"))
6674              != UINT64_MAX) {
6675                 phdr = (struct scsi_log_param_header *)data;
6676                 scsi_ulto2b(0x00f2, phdr->param_code);
6677                 phdr->param_control = SLP_LBIN | SLP_LP;
6678                 phdr->param_len = 8;
6679                 data = (uint8_t *)(phdr + 1);
6680                 scsi_ulto4b(val >> CTL_LBP_EXPONENT, data);
6681                 data[4] = 0x02; /* per-pool */
6682                 data += phdr->param_len;
6683         }
6684
6685         page_index->page_len = data - page_index->page_data;
6686         return (0);
6687 }
6688
6689 int
6690 ctl_sap_log_sense_handler(struct ctl_scsiio *ctsio,
6691                                struct ctl_page_index *page_index,
6692                                int pc)
6693 {
6694         struct ctl_lun *lun;
6695         struct stat_page *data;
6696         uint64_t rn, wn, rb, wb;
6697         struct bintime rt, wt;
6698         int i;
6699
6700         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6701         data = (struct stat_page *)page_index->page_data;
6702
6703         scsi_ulto2b(SLP_SAP, data->sap.hdr.param_code);
6704         data->sap.hdr.param_control = SLP_LBIN;
6705         data->sap.hdr.param_len = sizeof(struct scsi_log_stat_and_perf) -
6706             sizeof(struct scsi_log_param_header);
6707         rn = wn = rb = wb = 0;
6708         bintime_clear(&rt);
6709         bintime_clear(&wt);
6710         for (i = 0; i < CTL_MAX_PORTS; i++) {
6711                 rn += lun->stats.ports[i].operations[CTL_STATS_READ];
6712                 wn += lun->stats.ports[i].operations[CTL_STATS_WRITE];
6713                 rb += lun->stats.ports[i].bytes[CTL_STATS_READ];
6714                 wb += lun->stats.ports[i].bytes[CTL_STATS_WRITE];
6715                 bintime_add(&rt, &lun->stats.ports[i].time[CTL_STATS_READ]);
6716                 bintime_add(&wt, &lun->stats.ports[i].time[CTL_STATS_WRITE]);
6717         }
6718         scsi_u64to8b(rn, data->sap.read_num);
6719         scsi_u64to8b(wn, data->sap.write_num);
6720         if (lun->stats.blocksize > 0) {
6721                 scsi_u64to8b(wb / lun->stats.blocksize,
6722                     data->sap.recvieved_lba);
6723                 scsi_u64to8b(rb / lun->stats.blocksize,
6724                     data->sap.transmitted_lba);
6725         }
6726         scsi_u64to8b((uint64_t)rt.sec * 1000 + rt.frac / (UINT64_MAX / 1000),
6727             data->sap.read_int);
6728         scsi_u64to8b((uint64_t)wt.sec * 1000 + wt.frac / (UINT64_MAX / 1000),
6729             data->sap.write_int);
6730         scsi_u64to8b(0, data->sap.weighted_num);
6731         scsi_u64to8b(0, data->sap.weighted_int);
6732         scsi_ulto2b(SLP_IT, data->it.hdr.param_code);
6733         data->it.hdr.param_control = SLP_LBIN;
6734         data->it.hdr.param_len = sizeof(struct scsi_log_idle_time) -
6735             sizeof(struct scsi_log_param_header);
6736 #ifdef CTL_TIME_IO
6737         scsi_u64to8b(lun->idle_time / SBT_1MS, data->it.idle_int);
6738 #endif
6739         scsi_ulto2b(SLP_TI, data->ti.hdr.param_code);
6740         data->it.hdr.param_control = SLP_LBIN;
6741         data->ti.hdr.param_len = sizeof(struct scsi_log_time_interval) -
6742             sizeof(struct scsi_log_param_header);
6743         scsi_ulto4b(3, data->ti.exponent);
6744         scsi_ulto4b(1, data->ti.integer);
6745         return (0);
6746 }
6747
6748 int
6749 ctl_ie_log_sense_handler(struct ctl_scsiio *ctsio,
6750                                struct ctl_page_index *page_index,
6751                                int pc)
6752 {
6753         struct ctl_lun *lun;
6754         struct scsi_log_informational_exceptions *data;
6755
6756         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6757         data = (struct scsi_log_informational_exceptions *)page_index->page_data;
6758
6759         scsi_ulto2b(SLP_IE_GEN, data->hdr.param_code);
6760         data->hdr.param_control = SLP_LBIN;
6761         data->hdr.param_len = sizeof(struct scsi_log_informational_exceptions) -
6762             sizeof(struct scsi_log_param_header);
6763         data->ie_asc = lun->ie_asc;
6764         data->ie_ascq = lun->ie_ascq;
6765         data->temperature = 0xff;
6766         return (0);
6767 }
6768
6769 int
6770 ctl_log_sense(struct ctl_scsiio *ctsio)
6771 {
6772         struct ctl_lun *lun;
6773         int i, pc, page_code, subpage;
6774         int alloc_len, total_len;
6775         struct ctl_page_index *page_index;
6776         struct scsi_log_sense *cdb;
6777         struct scsi_log_header *header;
6778
6779         CTL_DEBUG_PRINT(("ctl_log_sense\n"));
6780
6781         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6782         cdb = (struct scsi_log_sense *)ctsio->cdb;
6783         pc = (cdb->page & SLS_PAGE_CTRL_MASK) >> 6;
6784         page_code = cdb->page & SLS_PAGE_CODE;
6785         subpage = cdb->subpage;
6786         alloc_len = scsi_2btoul(cdb->length);
6787
6788         page_index = NULL;
6789         for (i = 0; i < CTL_NUM_LOG_PAGES; i++) {
6790                 page_index = &lun->log_pages.index[i];
6791
6792                 /* Look for the right page code */
6793                 if ((page_index->page_code & SL_PAGE_CODE) != page_code)
6794                         continue;
6795
6796                 /* Look for the right subpage or the subpage wildcard*/
6797                 if (page_index->subpage != subpage)
6798                         continue;
6799
6800                 break;
6801         }
6802         if (i >= CTL_NUM_LOG_PAGES) {
6803                 ctl_set_invalid_field(ctsio,
6804                                       /*sks_valid*/ 1,
6805                                       /*command*/ 1,
6806                                       /*field*/ 2,
6807                                       /*bit_valid*/ 0,
6808                                       /*bit*/ 0);
6809                 ctl_done((union ctl_io *)ctsio);
6810                 return (CTL_RETVAL_COMPLETE);
6811         }
6812
6813         total_len = sizeof(struct scsi_log_header) + page_index->page_len;
6814
6815         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
6816         ctsio->kern_sg_entries = 0;
6817         ctsio->kern_data_resid = 0;
6818         ctsio->kern_rel_offset = 0;
6819         if (total_len < alloc_len) {
6820                 ctsio->residual = alloc_len - total_len;
6821                 ctsio->kern_data_len = total_len;
6822                 ctsio->kern_total_len = total_len;
6823         } else {
6824                 ctsio->residual = 0;
6825                 ctsio->kern_data_len = alloc_len;
6826                 ctsio->kern_total_len = alloc_len;
6827         }
6828
6829         header = (struct scsi_log_header *)ctsio->kern_data_ptr;
6830         header->page = page_index->page_code;
6831         if (page_index->page_code == SLS_LOGICAL_BLOCK_PROVISIONING)
6832                 header->page |= SL_DS;
6833         if (page_index->subpage) {
6834                 header->page |= SL_SPF;
6835                 header->subpage = page_index->subpage;
6836         }
6837         scsi_ulto2b(page_index->page_len, header->datalen);
6838
6839         /*
6840          * Call the handler, if it exists, to update the
6841          * page to the latest values.
6842          */
6843         if (page_index->sense_handler != NULL)
6844                 page_index->sense_handler(ctsio, page_index, pc);
6845
6846         memcpy(header + 1, page_index->page_data, page_index->page_len);
6847
6848         ctl_set_success(ctsio);
6849         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6850         ctsio->be_move_done = ctl_config_move_done;
6851         ctl_datamove((union ctl_io *)ctsio);
6852         return (CTL_RETVAL_COMPLETE);
6853 }
6854
6855 int
6856 ctl_read_capacity(struct ctl_scsiio *ctsio)
6857 {
6858         struct scsi_read_capacity *cdb;
6859         struct scsi_read_capacity_data *data;
6860         struct ctl_lun *lun;
6861         uint32_t lba;
6862
6863         CTL_DEBUG_PRINT(("ctl_read_capacity\n"));
6864
6865         cdb = (struct scsi_read_capacity *)ctsio->cdb;
6866
6867         lba = scsi_4btoul(cdb->addr);
6868         if (((cdb->pmi & SRC_PMI) == 0)
6869          && (lba != 0)) {
6870                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6871                                       /*sks_valid*/ 1,
6872                                       /*command*/ 1,
6873                                       /*field*/ 2,
6874                                       /*bit_valid*/ 0,
6875                                       /*bit*/ 0);
6876                 ctl_done((union ctl_io *)ctsio);
6877                 return (CTL_RETVAL_COMPLETE);
6878         }
6879
6880         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6881
6882         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6883         data = (struct scsi_read_capacity_data *)ctsio->kern_data_ptr;
6884         ctsio->residual = 0;
6885         ctsio->kern_data_len = sizeof(*data);
6886         ctsio->kern_total_len = sizeof(*data);
6887         ctsio->kern_data_resid = 0;
6888         ctsio->kern_rel_offset = 0;
6889         ctsio->kern_sg_entries = 0;
6890
6891         /*
6892          * If the maximum LBA is greater than 0xfffffffe, the user must
6893          * issue a SERVICE ACTION IN (16) command, with the read capacity
6894          * serivce action set.
6895          */
6896         if (lun->be_lun->maxlba > 0xfffffffe)
6897                 scsi_ulto4b(0xffffffff, data->addr);
6898         else
6899                 scsi_ulto4b(lun->be_lun->maxlba, data->addr);
6900
6901         /*
6902          * XXX KDM this may not be 512 bytes...
6903          */
6904         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6905
6906         ctl_set_success(ctsio);
6907         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6908         ctsio->be_move_done = ctl_config_move_done;
6909         ctl_datamove((union ctl_io *)ctsio);
6910         return (CTL_RETVAL_COMPLETE);
6911 }
6912
6913 int
6914 ctl_read_capacity_16(struct ctl_scsiio *ctsio)
6915 {
6916         struct scsi_read_capacity_16 *cdb;
6917         struct scsi_read_capacity_data_long *data;
6918         struct ctl_lun *lun;
6919         uint64_t lba;
6920         uint32_t alloc_len;
6921
6922         CTL_DEBUG_PRINT(("ctl_read_capacity_16\n"));
6923
6924         cdb = (struct scsi_read_capacity_16 *)ctsio->cdb;
6925
6926         alloc_len = scsi_4btoul(cdb->alloc_len);
6927         lba = scsi_8btou64(cdb->addr);
6928
6929         if ((cdb->reladr & SRC16_PMI)
6930          && (lba != 0)) {
6931                 ctl_set_invalid_field(/*ctsio*/ ctsio,
6932                                       /*sks_valid*/ 1,
6933                                       /*command*/ 1,
6934                                       /*field*/ 2,
6935                                       /*bit_valid*/ 0,
6936                                       /*bit*/ 0);
6937                 ctl_done((union ctl_io *)ctsio);
6938                 return (CTL_RETVAL_COMPLETE);
6939         }
6940
6941         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6942
6943         ctsio->kern_data_ptr = malloc(sizeof(*data), M_CTL, M_WAITOK | M_ZERO);
6944         data = (struct scsi_read_capacity_data_long *)ctsio->kern_data_ptr;
6945
6946         if (sizeof(*data) < alloc_len) {
6947                 ctsio->residual = alloc_len - sizeof(*data);
6948                 ctsio->kern_data_len = sizeof(*data);
6949                 ctsio->kern_total_len = sizeof(*data);
6950         } else {
6951                 ctsio->residual = 0;
6952                 ctsio->kern_data_len = alloc_len;
6953                 ctsio->kern_total_len = alloc_len;
6954         }
6955         ctsio->kern_data_resid = 0;
6956         ctsio->kern_rel_offset = 0;
6957         ctsio->kern_sg_entries = 0;
6958
6959         scsi_u64to8b(lun->be_lun->maxlba, data->addr);
6960         /* XXX KDM this may not be 512 bytes... */
6961         scsi_ulto4b(lun->be_lun->blocksize, data->length);
6962         data->prot_lbppbe = lun->be_lun->pblockexp & SRC16_LBPPBE;
6963         scsi_ulto2b(lun->be_lun->pblockoff & SRC16_LALBA_A, data->lalba_lbp);
6964         if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP)
6965                 data->lalba_lbp[0] |= SRC16_LBPME | SRC16_LBPRZ;
6966
6967         ctl_set_success(ctsio);
6968         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
6969         ctsio->be_move_done = ctl_config_move_done;
6970         ctl_datamove((union ctl_io *)ctsio);
6971         return (CTL_RETVAL_COMPLETE);
6972 }
6973
6974 int
6975 ctl_get_lba_status(struct ctl_scsiio *ctsio)
6976 {
6977         struct scsi_get_lba_status *cdb;
6978         struct scsi_get_lba_status_data *data;
6979         struct ctl_lun *lun;
6980         struct ctl_lba_len_flags *lbalen;
6981         uint64_t lba;
6982         uint32_t alloc_len, total_len;
6983         int retval;
6984
6985         CTL_DEBUG_PRINT(("ctl_get_lba_status\n"));
6986
6987         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
6988         cdb = (struct scsi_get_lba_status *)ctsio->cdb;
6989         lba = scsi_8btou64(cdb->addr);
6990         alloc_len = scsi_4btoul(cdb->alloc_len);
6991
6992         if (lba > lun->be_lun->maxlba) {
6993                 ctl_set_lba_out_of_range(ctsio);
6994                 ctl_done((union ctl_io *)ctsio);
6995                 return (CTL_RETVAL_COMPLETE);
6996         }
6997
6998         total_len = sizeof(*data) + sizeof(data->descr[0]);
6999         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7000         data = (struct scsi_get_lba_status_data *)ctsio->kern_data_ptr;
7001
7002         if (total_len < alloc_len) {
7003                 ctsio->residual = alloc_len - total_len;
7004                 ctsio->kern_data_len = total_len;
7005                 ctsio->kern_total_len = total_len;
7006         } else {
7007                 ctsio->residual = 0;
7008                 ctsio->kern_data_len = alloc_len;
7009                 ctsio->kern_total_len = alloc_len;
7010         }
7011         ctsio->kern_data_resid = 0;
7012         ctsio->kern_rel_offset = 0;
7013         ctsio->kern_sg_entries = 0;
7014
7015         /* Fill dummy data in case backend can't tell anything. */
7016         scsi_ulto4b(4 + sizeof(data->descr[0]), data->length);
7017         scsi_u64to8b(lba, data->descr[0].addr);
7018         scsi_ulto4b(MIN(UINT32_MAX, lun->be_lun->maxlba + 1 - lba),
7019             data->descr[0].length);
7020         data->descr[0].status = 0; /* Mapped or unknown. */
7021
7022         ctl_set_success(ctsio);
7023         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7024         ctsio->be_move_done = ctl_config_move_done;
7025
7026         lbalen = (struct ctl_lba_len_flags *)&ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
7027         lbalen->lba = lba;
7028         lbalen->len = total_len;
7029         lbalen->flags = 0;
7030         retval = lun->backend->config_read((union ctl_io *)ctsio);
7031         return (CTL_RETVAL_COMPLETE);
7032 }
7033
7034 int
7035 ctl_read_defect(struct ctl_scsiio *ctsio)
7036 {
7037         struct scsi_read_defect_data_10 *ccb10;
7038         struct scsi_read_defect_data_12 *ccb12;
7039         struct scsi_read_defect_data_hdr_10 *data10;
7040         struct scsi_read_defect_data_hdr_12 *data12;
7041         uint32_t alloc_len, data_len;
7042         uint8_t format;
7043
7044         CTL_DEBUG_PRINT(("ctl_read_defect\n"));
7045
7046         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7047                 ccb10 = (struct scsi_read_defect_data_10 *)&ctsio->cdb;
7048                 format = ccb10->format;
7049                 alloc_len = scsi_2btoul(ccb10->alloc_length);
7050                 data_len = sizeof(*data10);
7051         } else {
7052                 ccb12 = (struct scsi_read_defect_data_12 *)&ctsio->cdb;
7053                 format = ccb12->format;
7054                 alloc_len = scsi_4btoul(ccb12->alloc_length);
7055                 data_len = sizeof(*data12);
7056         }
7057         if (alloc_len == 0) {
7058                 ctl_set_success(ctsio);
7059                 ctl_done((union ctl_io *)ctsio);
7060                 return (CTL_RETVAL_COMPLETE);
7061         }
7062
7063         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
7064         if (data_len < alloc_len) {
7065                 ctsio->residual = alloc_len - data_len;
7066                 ctsio->kern_data_len = data_len;
7067                 ctsio->kern_total_len = data_len;
7068         } else {
7069                 ctsio->residual = 0;
7070                 ctsio->kern_data_len = alloc_len;
7071                 ctsio->kern_total_len = alloc_len;
7072         }
7073         ctsio->kern_data_resid = 0;
7074         ctsio->kern_rel_offset = 0;
7075         ctsio->kern_sg_entries = 0;
7076
7077         if (ctsio->cdb[0] == READ_DEFECT_DATA_10) {
7078                 data10 = (struct scsi_read_defect_data_hdr_10 *)
7079                     ctsio->kern_data_ptr;
7080                 data10->format = format;
7081                 scsi_ulto2b(0, data10->length);
7082         } else {
7083                 data12 = (struct scsi_read_defect_data_hdr_12 *)
7084                     ctsio->kern_data_ptr;
7085                 data12->format = format;
7086                 scsi_ulto2b(0, data12->generation);
7087                 scsi_ulto4b(0, data12->length);
7088         }
7089
7090         ctl_set_success(ctsio);
7091         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7092         ctsio->be_move_done = ctl_config_move_done;
7093         ctl_datamove((union ctl_io *)ctsio);
7094         return (CTL_RETVAL_COMPLETE);
7095 }
7096
7097 int
7098 ctl_report_tagret_port_groups(struct ctl_scsiio *ctsio)
7099 {
7100         struct scsi_maintenance_in *cdb;
7101         int retval;
7102         int alloc_len, ext, total_len = 0, g, pc, pg, ts, os;
7103         int num_ha_groups, num_target_ports, shared_group;
7104         struct ctl_lun *lun;
7105         struct ctl_softc *softc;
7106         struct ctl_port *port;
7107         struct scsi_target_group_data *rtg_ptr;
7108         struct scsi_target_group_data_extended *rtg_ext_ptr;
7109         struct scsi_target_port_group_descriptor *tpg_desc;
7110
7111         CTL_DEBUG_PRINT(("ctl_report_tagret_port_groups\n"));
7112
7113         cdb = (struct scsi_maintenance_in *)ctsio->cdb;
7114         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7115         softc = lun->ctl_softc;
7116
7117         retval = CTL_RETVAL_COMPLETE;
7118
7119         switch (cdb->byte2 & STG_PDF_MASK) {
7120         case STG_PDF_LENGTH:
7121                 ext = 0;
7122                 break;
7123         case STG_PDF_EXTENDED:
7124                 ext = 1;
7125                 break;
7126         default:
7127                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7128                                       /*sks_valid*/ 1,
7129                                       /*command*/ 1,
7130                                       /*field*/ 2,
7131                                       /*bit_valid*/ 1,
7132                                       /*bit*/ 5);
7133                 ctl_done((union ctl_io *)ctsio);
7134                 return(retval);
7135         }
7136
7137         num_target_ports = 0;
7138         shared_group = (softc->is_single != 0);
7139         mtx_lock(&softc->ctl_lock);
7140         STAILQ_FOREACH(port, &softc->port_list, links) {
7141                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7142                         continue;
7143                 if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7144                         continue;
7145                 num_target_ports++;
7146                 if (port->status & CTL_PORT_STATUS_HA_SHARED)
7147                         shared_group = 1;
7148         }
7149         mtx_unlock(&softc->ctl_lock);
7150         num_ha_groups = (softc->is_single) ? 0 : NUM_HA_SHELVES;
7151
7152         if (ext)
7153                 total_len = sizeof(struct scsi_target_group_data_extended);
7154         else
7155                 total_len = sizeof(struct scsi_target_group_data);
7156         total_len += sizeof(struct scsi_target_port_group_descriptor) *
7157                 (shared_group + num_ha_groups) +
7158             sizeof(struct scsi_target_port_descriptor) * num_target_ports;
7159
7160         alloc_len = scsi_4btoul(cdb->length);
7161
7162         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7163
7164         ctsio->kern_sg_entries = 0;
7165
7166         if (total_len < alloc_len) {
7167                 ctsio->residual = alloc_len - total_len;
7168                 ctsio->kern_data_len = total_len;
7169                 ctsio->kern_total_len = total_len;
7170         } else {
7171                 ctsio->residual = 0;
7172                 ctsio->kern_data_len = alloc_len;
7173                 ctsio->kern_total_len = alloc_len;
7174         }
7175         ctsio->kern_data_resid = 0;
7176         ctsio->kern_rel_offset = 0;
7177
7178         if (ext) {
7179                 rtg_ext_ptr = (struct scsi_target_group_data_extended *)
7180                     ctsio->kern_data_ptr;
7181                 scsi_ulto4b(total_len - 4, rtg_ext_ptr->length);
7182                 rtg_ext_ptr->format_type = 0x10;
7183                 rtg_ext_ptr->implicit_transition_time = 0;
7184                 tpg_desc = &rtg_ext_ptr->groups[0];
7185         } else {
7186                 rtg_ptr = (struct scsi_target_group_data *)
7187                     ctsio->kern_data_ptr;
7188                 scsi_ulto4b(total_len - 4, rtg_ptr->length);
7189                 tpg_desc = &rtg_ptr->groups[0];
7190         }
7191
7192         mtx_lock(&softc->ctl_lock);
7193         pg = softc->port_min / softc->port_cnt;
7194         if (lun->flags & (CTL_LUN_PRIMARY_SC | CTL_LUN_PEER_SC_PRIMARY)) {
7195                 /* Some shelf is known to be primary. */
7196                 if (softc->ha_link == CTL_HA_LINK_OFFLINE)
7197                         os = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7198                 else if (softc->ha_link == CTL_HA_LINK_UNKNOWN)
7199                         os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7200                 else if (softc->ha_mode == CTL_HA_MODE_ACT_STBY)
7201                         os = TPG_ASYMMETRIC_ACCESS_STANDBY;
7202                 else
7203                         os = TPG_ASYMMETRIC_ACCESS_NONOPTIMIZED;
7204                 if (lun->flags & CTL_LUN_PRIMARY_SC) {
7205                         ts = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7206                 } else {
7207                         ts = os;
7208                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7209                 }
7210         } else {
7211                 /* No known primary shelf. */
7212                 if (softc->ha_link == CTL_HA_LINK_OFFLINE) {
7213                         ts = TPG_ASYMMETRIC_ACCESS_UNAVAILABLE;
7214                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7215                 } else if (softc->ha_link == CTL_HA_LINK_UNKNOWN) {
7216                         ts = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7217                         os = TPG_ASYMMETRIC_ACCESS_OPTIMIZED;
7218                 } else {
7219                         ts = os = TPG_ASYMMETRIC_ACCESS_TRANSITIONING;
7220                 }
7221         }
7222         if (shared_group) {
7223                 tpg_desc->pref_state = ts;
7224                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7225                     TPG_U_SUP | TPG_T_SUP;
7226                 scsi_ulto2b(1, tpg_desc->target_port_group);
7227                 tpg_desc->status = TPG_IMPLICIT;
7228                 pc = 0;
7229                 STAILQ_FOREACH(port, &softc->port_list, links) {
7230                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7231                                 continue;
7232                         if (!softc->is_single &&
7233                             (port->status & CTL_PORT_STATUS_HA_SHARED) == 0)
7234                                 continue;
7235                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7236                                 continue;
7237                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7238                             relative_target_port_identifier);
7239                         pc++;
7240                 }
7241                 tpg_desc->target_port_count = pc;
7242                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7243                     &tpg_desc->descriptors[pc];
7244         }
7245         for (g = 0; g < num_ha_groups; g++) {
7246                 tpg_desc->pref_state = (g == pg) ? ts : os;
7247                 tpg_desc->support = TPG_AO_SUP | TPG_AN_SUP | TPG_S_SUP |
7248                     TPG_U_SUP | TPG_T_SUP;
7249                 scsi_ulto2b(2 + g, tpg_desc->target_port_group);
7250                 tpg_desc->status = TPG_IMPLICIT;
7251                 pc = 0;
7252                 STAILQ_FOREACH(port, &softc->port_list, links) {
7253                         if (port->targ_port < g * softc->port_cnt ||
7254                             port->targ_port >= (g + 1) * softc->port_cnt)
7255                                 continue;
7256                         if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
7257                                 continue;
7258                         if (port->status & CTL_PORT_STATUS_HA_SHARED)
7259                                 continue;
7260                         if (ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
7261                                 continue;
7262                         scsi_ulto2b(port->targ_port, tpg_desc->descriptors[pc].
7263                             relative_target_port_identifier);
7264                         pc++;
7265                 }
7266                 tpg_desc->target_port_count = pc;
7267                 tpg_desc = (struct scsi_target_port_group_descriptor *)
7268                     &tpg_desc->descriptors[pc];
7269         }
7270         mtx_unlock(&softc->ctl_lock);
7271
7272         ctl_set_success(ctsio);
7273         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7274         ctsio->be_move_done = ctl_config_move_done;
7275         ctl_datamove((union ctl_io *)ctsio);
7276         return(retval);
7277 }
7278
7279 int
7280 ctl_report_supported_opcodes(struct ctl_scsiio *ctsio)
7281 {
7282         struct ctl_lun *lun;
7283         struct scsi_report_supported_opcodes *cdb;
7284         const struct ctl_cmd_entry *entry, *sentry;
7285         struct scsi_report_supported_opcodes_all *all;
7286         struct scsi_report_supported_opcodes_descr *descr;
7287         struct scsi_report_supported_opcodes_one *one;
7288         int retval;
7289         int alloc_len, total_len;
7290         int opcode, service_action, i, j, num;
7291
7292         CTL_DEBUG_PRINT(("ctl_report_supported_opcodes\n"));
7293
7294         cdb = (struct scsi_report_supported_opcodes *)ctsio->cdb;
7295         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7296
7297         retval = CTL_RETVAL_COMPLETE;
7298
7299         opcode = cdb->requested_opcode;
7300         service_action = scsi_2btoul(cdb->requested_service_action);
7301         switch (cdb->options & RSO_OPTIONS_MASK) {
7302         case RSO_OPTIONS_ALL:
7303                 num = 0;
7304                 for (i = 0; i < 256; i++) {
7305                         entry = &ctl_cmd_table[i];
7306                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7307                                 for (j = 0; j < 32; j++) {
7308                                         sentry = &((const struct ctl_cmd_entry *)
7309                                             entry->execute)[j];
7310                                         if (ctl_cmd_applicable(
7311                                             lun->be_lun->lun_type, sentry))
7312                                                 num++;
7313                                 }
7314                         } else {
7315                                 if (ctl_cmd_applicable(lun->be_lun->lun_type,
7316                                     entry))
7317                                         num++;
7318                         }
7319                 }
7320                 total_len = sizeof(struct scsi_report_supported_opcodes_all) +
7321                     num * sizeof(struct scsi_report_supported_opcodes_descr);
7322                 break;
7323         case RSO_OPTIONS_OC:
7324                 if (ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) {
7325                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7326                                               /*sks_valid*/ 1,
7327                                               /*command*/ 1,
7328                                               /*field*/ 2,
7329                                               /*bit_valid*/ 1,
7330                                               /*bit*/ 2);
7331                         ctl_done((union ctl_io *)ctsio);
7332                         return (CTL_RETVAL_COMPLETE);
7333                 }
7334                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7335                 break;
7336         case RSO_OPTIONS_OC_SA:
7337                 if ((ctl_cmd_table[opcode].flags & CTL_CMD_FLAG_SA5) == 0 ||
7338                     service_action >= 32) {
7339                         ctl_set_invalid_field(/*ctsio*/ ctsio,
7340                                               /*sks_valid*/ 1,
7341                                               /*command*/ 1,
7342                                               /*field*/ 2,
7343                                               /*bit_valid*/ 1,
7344                                               /*bit*/ 2);
7345                         ctl_done((union ctl_io *)ctsio);
7346                         return (CTL_RETVAL_COMPLETE);
7347                 }
7348                 total_len = sizeof(struct scsi_report_supported_opcodes_one) + 32;
7349                 break;
7350         default:
7351                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7352                                       /*sks_valid*/ 1,
7353                                       /*command*/ 1,
7354                                       /*field*/ 2,
7355                                       /*bit_valid*/ 1,
7356                                       /*bit*/ 2);
7357                 ctl_done((union ctl_io *)ctsio);
7358                 return (CTL_RETVAL_COMPLETE);
7359         }
7360
7361         alloc_len = scsi_4btoul(cdb->length);
7362
7363         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7364
7365         ctsio->kern_sg_entries = 0;
7366
7367         if (total_len < alloc_len) {
7368                 ctsio->residual = alloc_len - total_len;
7369                 ctsio->kern_data_len = total_len;
7370                 ctsio->kern_total_len = total_len;
7371         } else {
7372                 ctsio->residual = 0;
7373                 ctsio->kern_data_len = alloc_len;
7374                 ctsio->kern_total_len = alloc_len;
7375         }
7376         ctsio->kern_data_resid = 0;
7377         ctsio->kern_rel_offset = 0;
7378
7379         switch (cdb->options & RSO_OPTIONS_MASK) {
7380         case RSO_OPTIONS_ALL:
7381                 all = (struct scsi_report_supported_opcodes_all *)
7382                     ctsio->kern_data_ptr;
7383                 num = 0;
7384                 for (i = 0; i < 256; i++) {
7385                         entry = &ctl_cmd_table[i];
7386                         if (entry->flags & CTL_CMD_FLAG_SA5) {
7387                                 for (j = 0; j < 32; j++) {
7388                                         sentry = &((const struct ctl_cmd_entry *)
7389                                             entry->execute)[j];
7390                                         if (!ctl_cmd_applicable(
7391                                             lun->be_lun->lun_type, sentry))
7392                                                 continue;
7393                                         descr = &all->descr[num++];
7394                                         descr->opcode = i;
7395                                         scsi_ulto2b(j, descr->service_action);
7396                                         descr->flags = RSO_SERVACTV;
7397                                         scsi_ulto2b(sentry->length,
7398                                             descr->cdb_length);
7399                                 }
7400                         } else {
7401                                 if (!ctl_cmd_applicable(lun->be_lun->lun_type,
7402                                     entry))
7403                                         continue;
7404                                 descr = &all->descr[num++];
7405                                 descr->opcode = i;
7406                                 scsi_ulto2b(0, descr->service_action);
7407                                 descr->flags = 0;
7408                                 scsi_ulto2b(entry->length, descr->cdb_length);
7409                         }
7410                 }
7411                 scsi_ulto4b(
7412                     num * sizeof(struct scsi_report_supported_opcodes_descr),
7413                     all->length);
7414                 break;
7415         case RSO_OPTIONS_OC:
7416                 one = (struct scsi_report_supported_opcodes_one *)
7417                     ctsio->kern_data_ptr;
7418                 entry = &ctl_cmd_table[opcode];
7419                 goto fill_one;
7420         case RSO_OPTIONS_OC_SA:
7421                 one = (struct scsi_report_supported_opcodes_one *)
7422                     ctsio->kern_data_ptr;
7423                 entry = &ctl_cmd_table[opcode];
7424                 entry = &((const struct ctl_cmd_entry *)
7425                     entry->execute)[service_action];
7426 fill_one:
7427                 if (ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
7428                         one->support = 3;
7429                         scsi_ulto2b(entry->length, one->cdb_length);
7430                         one->cdb_usage[0] = opcode;
7431                         memcpy(&one->cdb_usage[1], entry->usage,
7432                             entry->length - 1);
7433                 } else
7434                         one->support = 1;
7435                 break;
7436         }
7437
7438         ctl_set_success(ctsio);
7439         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7440         ctsio->be_move_done = ctl_config_move_done;
7441         ctl_datamove((union ctl_io *)ctsio);
7442         return(retval);
7443 }
7444
7445 int
7446 ctl_report_supported_tmf(struct ctl_scsiio *ctsio)
7447 {
7448         struct scsi_report_supported_tmf *cdb;
7449         struct scsi_report_supported_tmf_data *data;
7450         int retval;
7451         int alloc_len, total_len;
7452
7453         CTL_DEBUG_PRINT(("ctl_report_supported_tmf\n"));
7454
7455         cdb = (struct scsi_report_supported_tmf *)ctsio->cdb;
7456
7457         retval = CTL_RETVAL_COMPLETE;
7458
7459         total_len = sizeof(struct scsi_report_supported_tmf_data);
7460         alloc_len = scsi_4btoul(cdb->length);
7461
7462         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7463
7464         ctsio->kern_sg_entries = 0;
7465
7466         if (total_len < alloc_len) {
7467                 ctsio->residual = alloc_len - total_len;
7468                 ctsio->kern_data_len = total_len;
7469                 ctsio->kern_total_len = total_len;
7470         } else {
7471                 ctsio->residual = 0;
7472                 ctsio->kern_data_len = alloc_len;
7473                 ctsio->kern_total_len = alloc_len;
7474         }
7475         ctsio->kern_data_resid = 0;
7476         ctsio->kern_rel_offset = 0;
7477
7478         data = (struct scsi_report_supported_tmf_data *)ctsio->kern_data_ptr;
7479         data->byte1 |= RST_ATS | RST_ATSS | RST_CTSS | RST_LURS | RST_QTS |
7480             RST_TRS;
7481         data->byte2 |= RST_QAES | RST_QTSS | RST_ITNRS;
7482
7483         ctl_set_success(ctsio);
7484         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7485         ctsio->be_move_done = ctl_config_move_done;
7486         ctl_datamove((union ctl_io *)ctsio);
7487         return (retval);
7488 }
7489
7490 int
7491 ctl_report_timestamp(struct ctl_scsiio *ctsio)
7492 {
7493         struct scsi_report_timestamp *cdb;
7494         struct scsi_report_timestamp_data *data;
7495         struct timeval tv;
7496         int64_t timestamp;
7497         int retval;
7498         int alloc_len, total_len;
7499
7500         CTL_DEBUG_PRINT(("ctl_report_timestamp\n"));
7501
7502         cdb = (struct scsi_report_timestamp *)ctsio->cdb;
7503
7504         retval = CTL_RETVAL_COMPLETE;
7505
7506         total_len = sizeof(struct scsi_report_timestamp_data);
7507         alloc_len = scsi_4btoul(cdb->length);
7508
7509         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7510
7511         ctsio->kern_sg_entries = 0;
7512
7513         if (total_len < alloc_len) {
7514                 ctsio->residual = alloc_len - total_len;
7515                 ctsio->kern_data_len = total_len;
7516                 ctsio->kern_total_len = total_len;
7517         } else {
7518                 ctsio->residual = 0;
7519                 ctsio->kern_data_len = alloc_len;
7520                 ctsio->kern_total_len = alloc_len;
7521         }
7522         ctsio->kern_data_resid = 0;
7523         ctsio->kern_rel_offset = 0;
7524
7525         data = (struct scsi_report_timestamp_data *)ctsio->kern_data_ptr;
7526         scsi_ulto2b(sizeof(*data) - 2, data->length);
7527         data->origin = RTS_ORIG_OUTSIDE;
7528         getmicrotime(&tv);
7529         timestamp = (int64_t)tv.tv_sec * 1000 + tv.tv_usec / 1000;
7530         scsi_ulto4b(timestamp >> 16, data->timestamp);
7531         scsi_ulto2b(timestamp & 0xffff, &data->timestamp[4]);
7532
7533         ctl_set_success(ctsio);
7534         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7535         ctsio->be_move_done = ctl_config_move_done;
7536         ctl_datamove((union ctl_io *)ctsio);
7537         return (retval);
7538 }
7539
7540 int
7541 ctl_persistent_reserve_in(struct ctl_scsiio *ctsio)
7542 {
7543         struct scsi_per_res_in *cdb;
7544         int alloc_len, total_len = 0;
7545         /* struct scsi_per_res_in_rsrv in_data; */
7546         struct ctl_lun *lun;
7547         struct ctl_softc *softc;
7548         uint64_t key;
7549
7550         CTL_DEBUG_PRINT(("ctl_persistent_reserve_in\n"));
7551
7552         cdb = (struct scsi_per_res_in *)ctsio->cdb;
7553
7554         alloc_len = scsi_2btoul(cdb->length);
7555
7556         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
7557         softc = lun->ctl_softc;
7558
7559 retry:
7560         mtx_lock(&lun->lun_lock);
7561         switch (cdb->action) {
7562         case SPRI_RK: /* read keys */
7563                 total_len = sizeof(struct scsi_per_res_in_keys) +
7564                         lun->pr_key_count *
7565                         sizeof(struct scsi_per_res_key);
7566                 break;
7567         case SPRI_RR: /* read reservation */
7568                 if (lun->flags & CTL_LUN_PR_RESERVED)
7569                         total_len = sizeof(struct scsi_per_res_in_rsrv);
7570                 else
7571                         total_len = sizeof(struct scsi_per_res_in_header);
7572                 break;
7573         case SPRI_RC: /* report capabilities */
7574                 total_len = sizeof(struct scsi_per_res_cap);
7575                 break;
7576         case SPRI_RS: /* read full status */
7577                 total_len = sizeof(struct scsi_per_res_in_header) +
7578                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7579                     lun->pr_key_count;
7580                 break;
7581         default:
7582                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7583         }
7584         mtx_unlock(&lun->lun_lock);
7585
7586         ctsio->kern_data_ptr = malloc(total_len, M_CTL, M_WAITOK | M_ZERO);
7587
7588         if (total_len < alloc_len) {
7589                 ctsio->residual = alloc_len - total_len;
7590                 ctsio->kern_data_len = total_len;
7591                 ctsio->kern_total_len = total_len;
7592         } else {
7593                 ctsio->residual = 0;
7594                 ctsio->kern_data_len = alloc_len;
7595                 ctsio->kern_total_len = alloc_len;
7596         }
7597
7598         ctsio->kern_data_resid = 0;
7599         ctsio->kern_rel_offset = 0;
7600         ctsio->kern_sg_entries = 0;
7601
7602         mtx_lock(&lun->lun_lock);
7603         switch (cdb->action) {
7604         case SPRI_RK: { // read keys
7605         struct scsi_per_res_in_keys *res_keys;
7606                 int i, key_count;
7607
7608                 res_keys = (struct scsi_per_res_in_keys*)ctsio->kern_data_ptr;
7609
7610                 /*
7611                  * We had to drop the lock to allocate our buffer, which
7612                  * leaves time for someone to come in with another
7613                  * persistent reservation.  (That is unlikely, though,
7614                  * since this should be the only persistent reservation
7615                  * command active right now.)
7616                  */
7617                 if (total_len != (sizeof(struct scsi_per_res_in_keys) +
7618                     (lun->pr_key_count *
7619                      sizeof(struct scsi_per_res_key)))){
7620                         mtx_unlock(&lun->lun_lock);
7621                         free(ctsio->kern_data_ptr, M_CTL);
7622                         printf("%s: reservation length changed, retrying\n",
7623                                __func__);
7624                         goto retry;
7625                 }
7626
7627                 scsi_ulto4b(lun->pr_generation, res_keys->header.generation);
7628
7629                 scsi_ulto4b(sizeof(struct scsi_per_res_key) *
7630                              lun->pr_key_count, res_keys->header.length);
7631
7632                 for (i = 0, key_count = 0; i < CTL_MAX_INITIATORS; i++) {
7633                         if ((key = ctl_get_prkey(lun, i)) == 0)
7634                                 continue;
7635
7636                         /*
7637                          * We used lun->pr_key_count to calculate the
7638                          * size to allocate.  If it turns out the number of
7639                          * initiators with the registered flag set is
7640                          * larger than that (i.e. they haven't been kept in
7641                          * sync), we've got a problem.
7642                          */
7643                         if (key_count >= lun->pr_key_count) {
7644                                 key_count++;
7645                                 continue;
7646                         }
7647                         scsi_u64to8b(key, res_keys->keys[key_count].key);
7648                         key_count++;
7649                 }
7650                 break;
7651         }
7652         case SPRI_RR: { // read reservation
7653                 struct scsi_per_res_in_rsrv *res;
7654                 int tmp_len, header_only;
7655
7656                 res = (struct scsi_per_res_in_rsrv *)ctsio->kern_data_ptr;
7657
7658                 scsi_ulto4b(lun->pr_generation, res->header.generation);
7659
7660                 if (lun->flags & CTL_LUN_PR_RESERVED)
7661                 {
7662                         tmp_len = sizeof(struct scsi_per_res_in_rsrv);
7663                         scsi_ulto4b(sizeof(struct scsi_per_res_in_rsrv_data),
7664                                     res->header.length);
7665                         header_only = 0;
7666                 } else {
7667                         tmp_len = sizeof(struct scsi_per_res_in_header);
7668                         scsi_ulto4b(0, res->header.length);
7669                         header_only = 1;
7670                 }
7671
7672                 /*
7673                  * We had to drop the lock to allocate our buffer, which
7674                  * leaves time for someone to come in with another
7675                  * persistent reservation.  (That is unlikely, though,
7676                  * since this should be the only persistent reservation
7677                  * command active right now.)
7678                  */
7679                 if (tmp_len != total_len) {
7680                         mtx_unlock(&lun->lun_lock);
7681                         free(ctsio->kern_data_ptr, M_CTL);
7682                         printf("%s: reservation status changed, retrying\n",
7683                                __func__);
7684                         goto retry;
7685                 }
7686
7687                 /*
7688                  * No reservation held, so we're done.
7689                  */
7690                 if (header_only != 0)
7691                         break;
7692
7693                 /*
7694                  * If the registration is an All Registrants type, the key
7695                  * is 0, since it doesn't really matter.
7696                  */
7697                 if (lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
7698                         scsi_u64to8b(ctl_get_prkey(lun, lun->pr_res_idx),
7699                             res->data.reservation);
7700                 }
7701                 res->data.scopetype = lun->pr_res_type;
7702                 break;
7703         }
7704         case SPRI_RC:     //report capabilities
7705         {
7706                 struct scsi_per_res_cap *res_cap;
7707                 uint16_t type_mask;
7708
7709                 res_cap = (struct scsi_per_res_cap *)ctsio->kern_data_ptr;
7710                 scsi_ulto2b(sizeof(*res_cap), res_cap->length);
7711                 res_cap->flags1 = SPRI_CRH;
7712                 res_cap->flags2 = SPRI_TMV | SPRI_ALLOW_5;
7713                 type_mask = SPRI_TM_WR_EX_AR |
7714                             SPRI_TM_EX_AC_RO |
7715                             SPRI_TM_WR_EX_RO |
7716                             SPRI_TM_EX_AC |
7717                             SPRI_TM_WR_EX |
7718                             SPRI_TM_EX_AC_AR;
7719                 scsi_ulto2b(type_mask, res_cap->type_mask);
7720                 break;
7721         }
7722         case SPRI_RS: { // read full status
7723                 struct scsi_per_res_in_full *res_status;
7724                 struct scsi_per_res_in_full_desc *res_desc;
7725                 struct ctl_port *port;
7726                 int i, len;
7727
7728                 res_status = (struct scsi_per_res_in_full*)ctsio->kern_data_ptr;
7729
7730                 /*
7731                  * We had to drop the lock to allocate our buffer, which
7732                  * leaves time for someone to come in with another
7733                  * persistent reservation.  (That is unlikely, though,
7734                  * since this should be the only persistent reservation
7735                  * command active right now.)
7736                  */
7737                 if (total_len < (sizeof(struct scsi_per_res_in_header) +
7738                     (sizeof(struct scsi_per_res_in_full_desc) + 256) *
7739                      lun->pr_key_count)){
7740                         mtx_unlock(&lun->lun_lock);
7741                         free(ctsio->kern_data_ptr, M_CTL);
7742                         printf("%s: reservation length changed, retrying\n",
7743                                __func__);
7744                         goto retry;
7745                 }
7746
7747                 scsi_ulto4b(lun->pr_generation, res_status->header.generation);
7748
7749                 res_desc = &res_status->desc[0];
7750                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7751                         if ((key = ctl_get_prkey(lun, i)) == 0)
7752                                 continue;
7753
7754                         scsi_u64to8b(key, res_desc->res_key.key);
7755                         if ((lun->flags & CTL_LUN_PR_RESERVED) &&
7756                             (lun->pr_res_idx == i ||
7757                              lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS)) {
7758                                 res_desc->flags = SPRI_FULL_R_HOLDER;
7759                                 res_desc->scopetype = lun->pr_res_type;
7760                         }
7761                         scsi_ulto2b(i / CTL_MAX_INIT_PER_PORT,
7762                             res_desc->rel_trgt_port_id);
7763                         len = 0;
7764                         port = softc->ctl_ports[i / CTL_MAX_INIT_PER_PORT];
7765                         if (port != NULL)
7766                                 len = ctl_create_iid(port,
7767                                     i % CTL_MAX_INIT_PER_PORT,
7768                                     res_desc->transport_id);
7769                         scsi_ulto4b(len, res_desc->additional_length);
7770                         res_desc = (struct scsi_per_res_in_full_desc *)
7771                             &res_desc->transport_id[len];
7772                 }
7773                 scsi_ulto4b((uint8_t *)res_desc - (uint8_t *)&res_status->desc[0],
7774                     res_status->header.length);
7775                 break;
7776         }
7777         default:
7778                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
7779         }
7780         mtx_unlock(&lun->lun_lock);
7781
7782         ctl_set_success(ctsio);
7783         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
7784         ctsio->be_move_done = ctl_config_move_done;
7785         ctl_datamove((union ctl_io *)ctsio);
7786         return (CTL_RETVAL_COMPLETE);
7787 }
7788
7789 /*
7790  * Returns 0 if ctl_persistent_reserve_out() should continue, non-zero if
7791  * it should return.
7792  */
7793 static int
7794 ctl_pro_preempt(struct ctl_softc *softc, struct ctl_lun *lun, uint64_t res_key,
7795                 uint64_t sa_res_key, uint8_t type, uint32_t residx,
7796                 struct ctl_scsiio *ctsio, struct scsi_per_res_out *cdb,
7797                 struct scsi_per_res_out_parms* param)
7798 {
7799         union ctl_ha_msg persis_io;
7800         int i;
7801
7802         mtx_lock(&lun->lun_lock);
7803         if (sa_res_key == 0) {
7804                 if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
7805                         /* validate scope and type */
7806                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7807                              SPR_LU_SCOPE) {
7808                                 mtx_unlock(&lun->lun_lock);
7809                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7810                                                       /*sks_valid*/ 1,
7811                                                       /*command*/ 1,
7812                                                       /*field*/ 2,
7813                                                       /*bit_valid*/ 1,
7814                                                       /*bit*/ 4);
7815                                 ctl_done((union ctl_io *)ctsio);
7816                                 return (1);
7817                         }
7818
7819                         if (type>8 || type==2 || type==4 || type==0) {
7820                                 mtx_unlock(&lun->lun_lock);
7821                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7822                                                       /*sks_valid*/ 1,
7823                                                       /*command*/ 1,
7824                                                       /*field*/ 2,
7825                                                       /*bit_valid*/ 1,
7826                                                       /*bit*/ 0);
7827                                 ctl_done((union ctl_io *)ctsio);
7828                                 return (1);
7829                         }
7830
7831                         /*
7832                          * Unregister everybody else and build UA for
7833                          * them
7834                          */
7835                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7836                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7837                                         continue;
7838
7839                                 ctl_clr_prkey(lun, i);
7840                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7841                         }
7842                         lun->pr_key_count = 1;
7843                         lun->pr_res_type = type;
7844                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7845                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7846                                 lun->pr_res_idx = residx;
7847                         lun->pr_generation++;
7848                         mtx_unlock(&lun->lun_lock);
7849
7850                         /* send msg to other side */
7851                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7852                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7853                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7854                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
7855                         persis_io.pr.pr_info.res_type = type;
7856                         memcpy(persis_io.pr.pr_info.sa_res_key,
7857                                param->serv_act_res_key,
7858                                sizeof(param->serv_act_res_key));
7859                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7860                             sizeof(persis_io.pr), M_WAITOK);
7861                 } else {
7862                         /* not all registrants */
7863                         mtx_unlock(&lun->lun_lock);
7864                         free(ctsio->kern_data_ptr, M_CTL);
7865                         ctl_set_invalid_field(ctsio,
7866                                               /*sks_valid*/ 1,
7867                                               /*command*/ 0,
7868                                               /*field*/ 8,
7869                                               /*bit_valid*/ 0,
7870                                               /*bit*/ 0);
7871                         ctl_done((union ctl_io *)ctsio);
7872                         return (1);
7873                 }
7874         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
7875                 || !(lun->flags & CTL_LUN_PR_RESERVED)) {
7876                 int found = 0;
7877
7878                 if (res_key == sa_res_key) {
7879                         /* special case */
7880                         /*
7881                          * The spec implies this is not good but doesn't
7882                          * say what to do. There are two choices either
7883                          * generate a res conflict or check condition
7884                          * with illegal field in parameter data. Since
7885                          * that is what is done when the sa_res_key is
7886                          * zero I'll take that approach since this has
7887                          * to do with the sa_res_key.
7888                          */
7889                         mtx_unlock(&lun->lun_lock);
7890                         free(ctsio->kern_data_ptr, M_CTL);
7891                         ctl_set_invalid_field(ctsio,
7892                                               /*sks_valid*/ 1,
7893                                               /*command*/ 0,
7894                                               /*field*/ 8,
7895                                               /*bit_valid*/ 0,
7896                                               /*bit*/ 0);
7897                         ctl_done((union ctl_io *)ctsio);
7898                         return (1);
7899                 }
7900
7901                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
7902                         if (ctl_get_prkey(lun, i) != sa_res_key)
7903                                 continue;
7904
7905                         found = 1;
7906                         ctl_clr_prkey(lun, i);
7907                         lun->pr_key_count--;
7908                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7909                 }
7910                 if (!found) {
7911                         mtx_unlock(&lun->lun_lock);
7912                         free(ctsio->kern_data_ptr, M_CTL);
7913                         ctl_set_reservation_conflict(ctsio);
7914                         ctl_done((union ctl_io *)ctsio);
7915                         return (CTL_RETVAL_COMPLETE);
7916                 }
7917                 lun->pr_generation++;
7918                 mtx_unlock(&lun->lun_lock);
7919
7920                 /* send msg to other side */
7921                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7922                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
7923                 persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
7924                 persis_io.pr.pr_info.residx = lun->pr_res_idx;
7925                 persis_io.pr.pr_info.res_type = type;
7926                 memcpy(persis_io.pr.pr_info.sa_res_key,
7927                        param->serv_act_res_key,
7928                        sizeof(param->serv_act_res_key));
7929                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
7930                     sizeof(persis_io.pr), M_WAITOK);
7931         } else {
7932                 /* Reserved but not all registrants */
7933                 /* sa_res_key is res holder */
7934                 if (sa_res_key == ctl_get_prkey(lun, lun->pr_res_idx)) {
7935                         /* validate scope and type */
7936                         if ((cdb->scope_type & SPR_SCOPE_MASK) !=
7937                              SPR_LU_SCOPE) {
7938                                 mtx_unlock(&lun->lun_lock);
7939                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7940                                                       /*sks_valid*/ 1,
7941                                                       /*command*/ 1,
7942                                                       /*field*/ 2,
7943                                                       /*bit_valid*/ 1,
7944                                                       /*bit*/ 4);
7945                                 ctl_done((union ctl_io *)ctsio);
7946                                 return (1);
7947                         }
7948
7949                         if (type>8 || type==2 || type==4 || type==0) {
7950                                 mtx_unlock(&lun->lun_lock);
7951                                 ctl_set_invalid_field(/*ctsio*/ ctsio,
7952                                                       /*sks_valid*/ 1,
7953                                                       /*command*/ 1,
7954                                                       /*field*/ 2,
7955                                                       /*bit_valid*/ 1,
7956                                                       /*bit*/ 0);
7957                                 ctl_done((union ctl_io *)ctsio);
7958                                 return (1);
7959                         }
7960
7961                         /*
7962                          * Do the following:
7963                          * if sa_res_key != res_key remove all
7964                          * registrants w/sa_res_key and generate UA
7965                          * for these registrants(Registrations
7966                          * Preempted) if it wasn't an exclusive
7967                          * reservation generate UA(Reservations
7968                          * Preempted) for all other registered nexuses
7969                          * if the type has changed. Establish the new
7970                          * reservation and holder. If res_key and
7971                          * sa_res_key are the same do the above
7972                          * except don't unregister the res holder.
7973                          */
7974
7975                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
7976                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
7977                                         continue;
7978
7979                                 if (sa_res_key == ctl_get_prkey(lun, i)) {
7980                                         ctl_clr_prkey(lun, i);
7981                                         lun->pr_key_count--;
7982                                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
7983                                 } else if (type != lun->pr_res_type &&
7984                                     (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
7985                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
7986                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
7987                                 }
7988                         }
7989                         lun->pr_res_type = type;
7990                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
7991                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
7992                                 lun->pr_res_idx = residx;
7993                         else
7994                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
7995                         lun->pr_generation++;
7996                         mtx_unlock(&lun->lun_lock);
7997
7998                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
7999                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8000                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8001                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8002                         persis_io.pr.pr_info.res_type = type;
8003                         memcpy(persis_io.pr.pr_info.sa_res_key,
8004                                param->serv_act_res_key,
8005                                sizeof(param->serv_act_res_key));
8006                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8007                             sizeof(persis_io.pr), M_WAITOK);
8008                 } else {
8009                         /*
8010                          * sa_res_key is not the res holder just
8011                          * remove registrants
8012                          */
8013                         int found=0;
8014
8015                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8016                                 if (sa_res_key != ctl_get_prkey(lun, i))
8017                                         continue;
8018
8019                                 found = 1;
8020                                 ctl_clr_prkey(lun, i);
8021                                 lun->pr_key_count--;
8022                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8023                         }
8024
8025                         if (!found) {
8026                                 mtx_unlock(&lun->lun_lock);
8027                                 free(ctsio->kern_data_ptr, M_CTL);
8028                                 ctl_set_reservation_conflict(ctsio);
8029                                 ctl_done((union ctl_io *)ctsio);
8030                                 return (1);
8031                         }
8032                         lun->pr_generation++;
8033                         mtx_unlock(&lun->lun_lock);
8034
8035                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8036                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8037                         persis_io.pr.pr_info.action = CTL_PR_PREEMPT;
8038                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8039                         persis_io.pr.pr_info.res_type = type;
8040                         memcpy(persis_io.pr.pr_info.sa_res_key,
8041                                param->serv_act_res_key,
8042                                sizeof(param->serv_act_res_key));
8043                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8044                             sizeof(persis_io.pr), M_WAITOK);
8045                 }
8046         }
8047         return (0);
8048 }
8049
8050 static void
8051 ctl_pro_preempt_other(struct ctl_lun *lun, union ctl_ha_msg *msg)
8052 {
8053         uint64_t sa_res_key;
8054         int i;
8055
8056         sa_res_key = scsi_8btou64(msg->pr.pr_info.sa_res_key);
8057
8058         if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS
8059          || lun->pr_res_idx == CTL_PR_NO_RESERVATION
8060          || sa_res_key != ctl_get_prkey(lun, lun->pr_res_idx)) {
8061                 if (sa_res_key == 0) {
8062                         /*
8063                          * Unregister everybody else and build UA for
8064                          * them
8065                          */
8066                         for(i = 0; i < CTL_MAX_INITIATORS; i++) {
8067                                 if (i == msg->pr.pr_info.residx ||
8068                                     ctl_get_prkey(lun, i) == 0)
8069                                         continue;
8070
8071                                 ctl_clr_prkey(lun, i);
8072                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8073                         }
8074
8075                         lun->pr_key_count = 1;
8076                         lun->pr_res_type = msg->pr.pr_info.res_type;
8077                         if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8078                             lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8079                                 lun->pr_res_idx = msg->pr.pr_info.residx;
8080                 } else {
8081                         for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8082                                 if (sa_res_key == ctl_get_prkey(lun, i))
8083                                         continue;
8084
8085                                 ctl_clr_prkey(lun, i);
8086                                 lun->pr_key_count--;
8087                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8088                         }
8089                 }
8090         } else {
8091                 for (i = 0; i < CTL_MAX_INITIATORS; i++) {
8092                         if (i == msg->pr.pr_info.residx ||
8093                             ctl_get_prkey(lun, i) == 0)
8094                                 continue;
8095
8096                         if (sa_res_key == ctl_get_prkey(lun, i)) {
8097                                 ctl_clr_prkey(lun, i);
8098                                 lun->pr_key_count--;
8099                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8100                         } else if (msg->pr.pr_info.res_type != lun->pr_res_type
8101                             && (lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8102                              lun->pr_res_type == SPR_TYPE_EX_AC_RO)) {
8103                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8104                         }
8105                 }
8106                 lun->pr_res_type = msg->pr.pr_info.res_type;
8107                 if (lun->pr_res_type != SPR_TYPE_WR_EX_AR &&
8108                     lun->pr_res_type != SPR_TYPE_EX_AC_AR)
8109                         lun->pr_res_idx = msg->pr.pr_info.residx;
8110                 else
8111                         lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8112         }
8113         lun->pr_generation++;
8114
8115 }
8116
8117
8118 int
8119 ctl_persistent_reserve_out(struct ctl_scsiio *ctsio)
8120 {
8121         int retval;
8122         u_int32_t param_len;
8123         struct scsi_per_res_out *cdb;
8124         struct ctl_lun *lun;
8125         struct scsi_per_res_out_parms* param;
8126         struct ctl_softc *softc;
8127         uint32_t residx;
8128         uint64_t res_key, sa_res_key, key;
8129         uint8_t type;
8130         union ctl_ha_msg persis_io;
8131         int    i;
8132
8133         CTL_DEBUG_PRINT(("ctl_persistent_reserve_out\n"));
8134
8135         retval = CTL_RETVAL_COMPLETE;
8136
8137         cdb = (struct scsi_per_res_out *)ctsio->cdb;
8138         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8139         softc = lun->ctl_softc;
8140
8141         /*
8142          * We only support whole-LUN scope.  The scope & type are ignored for
8143          * register, register and ignore existing key and clear.
8144          * We sometimes ignore scope and type on preempts too!!
8145          * Verify reservation type here as well.
8146          */
8147         type = cdb->scope_type & SPR_TYPE_MASK;
8148         if ((cdb->action == SPRO_RESERVE)
8149          || (cdb->action == SPRO_RELEASE)) {
8150                 if ((cdb->scope_type & SPR_SCOPE_MASK) != SPR_LU_SCOPE) {
8151                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8152                                               /*sks_valid*/ 1,
8153                                               /*command*/ 1,
8154                                               /*field*/ 2,
8155                                               /*bit_valid*/ 1,
8156                                               /*bit*/ 4);
8157                         ctl_done((union ctl_io *)ctsio);
8158                         return (CTL_RETVAL_COMPLETE);
8159                 }
8160
8161                 if (type>8 || type==2 || type==4 || type==0) {
8162                         ctl_set_invalid_field(/*ctsio*/ ctsio,
8163                                               /*sks_valid*/ 1,
8164                                               /*command*/ 1,
8165                                               /*field*/ 2,
8166                                               /*bit_valid*/ 1,
8167                                               /*bit*/ 0);
8168                         ctl_done((union ctl_io *)ctsio);
8169                         return (CTL_RETVAL_COMPLETE);
8170                 }
8171         }
8172
8173         param_len = scsi_4btoul(cdb->length);
8174
8175         if ((ctsio->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0) {
8176                 ctsio->kern_data_ptr = malloc(param_len, M_CTL, M_WAITOK);
8177                 ctsio->kern_data_len = param_len;
8178                 ctsio->kern_total_len = param_len;
8179                 ctsio->kern_data_resid = 0;
8180                 ctsio->kern_rel_offset = 0;
8181                 ctsio->kern_sg_entries = 0;
8182                 ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
8183                 ctsio->be_move_done = ctl_config_move_done;
8184                 ctl_datamove((union ctl_io *)ctsio);
8185
8186                 return (CTL_RETVAL_COMPLETE);
8187         }
8188
8189         param = (struct scsi_per_res_out_parms *)ctsio->kern_data_ptr;
8190
8191         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
8192         res_key = scsi_8btou64(param->res_key.key);
8193         sa_res_key = scsi_8btou64(param->serv_act_res_key);
8194
8195         /*
8196          * Validate the reservation key here except for SPRO_REG_IGNO
8197          * This must be done for all other service actions
8198          */
8199         if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REG_IGNO) {
8200                 mtx_lock(&lun->lun_lock);
8201                 if ((key = ctl_get_prkey(lun, residx)) != 0) {
8202                         if (res_key != key) {
8203                                 /*
8204                                  * The current key passed in doesn't match
8205                                  * the one the initiator previously
8206                                  * registered.
8207                                  */
8208                                 mtx_unlock(&lun->lun_lock);
8209                                 free(ctsio->kern_data_ptr, M_CTL);
8210                                 ctl_set_reservation_conflict(ctsio);
8211                                 ctl_done((union ctl_io *)ctsio);
8212                                 return (CTL_RETVAL_COMPLETE);
8213                         }
8214                 } else if ((cdb->action & SPRO_ACTION_MASK) != SPRO_REGISTER) {
8215                         /*
8216                          * We are not registered
8217                          */
8218                         mtx_unlock(&lun->lun_lock);
8219                         free(ctsio->kern_data_ptr, M_CTL);
8220                         ctl_set_reservation_conflict(ctsio);
8221                         ctl_done((union ctl_io *)ctsio);
8222                         return (CTL_RETVAL_COMPLETE);
8223                 } else if (res_key != 0) {
8224                         /*
8225                          * We are not registered and trying to register but
8226                          * the register key isn't zero.
8227                          */
8228                         mtx_unlock(&lun->lun_lock);
8229                         free(ctsio->kern_data_ptr, M_CTL);
8230                         ctl_set_reservation_conflict(ctsio);
8231                         ctl_done((union ctl_io *)ctsio);
8232                         return (CTL_RETVAL_COMPLETE);
8233                 }
8234                 mtx_unlock(&lun->lun_lock);
8235         }
8236
8237         switch (cdb->action & SPRO_ACTION_MASK) {
8238         case SPRO_REGISTER:
8239         case SPRO_REG_IGNO: {
8240
8241 #if 0
8242                 printf("Registration received\n");
8243 #endif
8244
8245                 /*
8246                  * We don't support any of these options, as we report in
8247                  * the read capabilities request (see
8248                  * ctl_persistent_reserve_in(), above).
8249                  */
8250                 if ((param->flags & SPR_SPEC_I_PT)
8251                  || (param->flags & SPR_ALL_TG_PT)
8252                  || (param->flags & SPR_APTPL)) {
8253                         int bit_ptr;
8254
8255                         if (param->flags & SPR_APTPL)
8256                                 bit_ptr = 0;
8257                         else if (param->flags & SPR_ALL_TG_PT)
8258                                 bit_ptr = 2;
8259                         else /* SPR_SPEC_I_PT */
8260                                 bit_ptr = 3;
8261
8262                         free(ctsio->kern_data_ptr, M_CTL);
8263                         ctl_set_invalid_field(ctsio,
8264                                               /*sks_valid*/ 1,
8265                                               /*command*/ 0,
8266                                               /*field*/ 20,
8267                                               /*bit_valid*/ 1,
8268                                               /*bit*/ bit_ptr);
8269                         ctl_done((union ctl_io *)ctsio);
8270                         return (CTL_RETVAL_COMPLETE);
8271                 }
8272
8273                 mtx_lock(&lun->lun_lock);
8274
8275                 /*
8276                  * The initiator wants to clear the
8277                  * key/unregister.
8278                  */
8279                 if (sa_res_key == 0) {
8280                         if ((res_key == 0
8281                           && (cdb->action & SPRO_ACTION_MASK) == SPRO_REGISTER)
8282                          || ((cdb->action & SPRO_ACTION_MASK) == SPRO_REG_IGNO
8283                           && ctl_get_prkey(lun, residx) == 0)) {
8284                                 mtx_unlock(&lun->lun_lock);
8285                                 goto done;
8286                         }
8287
8288                         ctl_clr_prkey(lun, residx);
8289                         lun->pr_key_count--;
8290
8291                         if (residx == lun->pr_res_idx) {
8292                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8293                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8294
8295                                 if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8296                                      lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8297                                     lun->pr_key_count) {
8298                                         /*
8299                                          * If the reservation is a registrants
8300                                          * only type we need to generate a UA
8301                                          * for other registered inits.  The
8302                                          * sense code should be RESERVATIONS
8303                                          * RELEASED
8304                                          */
8305
8306                                         for (i = softc->init_min; i < softc->init_max; i++){
8307                                                 if (ctl_get_prkey(lun, i) == 0)
8308                                                         continue;
8309                                                 ctl_est_ua(lun, i,
8310                                                     CTL_UA_RES_RELEASE);
8311                                         }
8312                                 }
8313                                 lun->pr_res_type = 0;
8314                         } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8315                                 if (lun->pr_key_count==0) {
8316                                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8317                                         lun->pr_res_type = 0;
8318                                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8319                                 }
8320                         }
8321                         lun->pr_generation++;
8322                         mtx_unlock(&lun->lun_lock);
8323
8324                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8325                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8326                         persis_io.pr.pr_info.action = CTL_PR_UNREG_KEY;
8327                         persis_io.pr.pr_info.residx = residx;
8328                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8329                             sizeof(persis_io.pr), M_WAITOK);
8330                 } else /* sa_res_key != 0 */ {
8331
8332                         /*
8333                          * If we aren't registered currently then increment
8334                          * the key count and set the registered flag.
8335                          */
8336                         ctl_alloc_prkey(lun, residx);
8337                         if (ctl_get_prkey(lun, residx) == 0)
8338                                 lun->pr_key_count++;
8339                         ctl_set_prkey(lun, residx, sa_res_key);
8340                         lun->pr_generation++;
8341                         mtx_unlock(&lun->lun_lock);
8342
8343                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8344                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8345                         persis_io.pr.pr_info.action = CTL_PR_REG_KEY;
8346                         persis_io.pr.pr_info.residx = residx;
8347                         memcpy(persis_io.pr.pr_info.sa_res_key,
8348                                param->serv_act_res_key,
8349                                sizeof(param->serv_act_res_key));
8350                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8351                             sizeof(persis_io.pr), M_WAITOK);
8352                 }
8353
8354                 break;
8355         }
8356         case SPRO_RESERVE:
8357 #if 0
8358                 printf("Reserve executed type %d\n", type);
8359 #endif
8360                 mtx_lock(&lun->lun_lock);
8361                 if (lun->flags & CTL_LUN_PR_RESERVED) {
8362                         /*
8363                          * if this isn't the reservation holder and it's
8364                          * not a "all registrants" type or if the type is
8365                          * different then we have a conflict
8366                          */
8367                         if ((lun->pr_res_idx != residx
8368                           && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS)
8369                          || lun->pr_res_type != type) {
8370                                 mtx_unlock(&lun->lun_lock);
8371                                 free(ctsio->kern_data_ptr, M_CTL);
8372                                 ctl_set_reservation_conflict(ctsio);
8373                                 ctl_done((union ctl_io *)ctsio);
8374                                 return (CTL_RETVAL_COMPLETE);
8375                         }
8376                         mtx_unlock(&lun->lun_lock);
8377                 } else /* create a reservation */ {
8378                         /*
8379                          * If it's not an "all registrants" type record
8380                          * reservation holder
8381                          */
8382                         if (type != SPR_TYPE_WR_EX_AR
8383                          && type != SPR_TYPE_EX_AC_AR)
8384                                 lun->pr_res_idx = residx; /* Res holder */
8385                         else
8386                                 lun->pr_res_idx = CTL_PR_ALL_REGISTRANTS;
8387
8388                         lun->flags |= CTL_LUN_PR_RESERVED;
8389                         lun->pr_res_type = type;
8390
8391                         mtx_unlock(&lun->lun_lock);
8392
8393                         /* send msg to other side */
8394                         persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8395                         persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8396                         persis_io.pr.pr_info.action = CTL_PR_RESERVE;
8397                         persis_io.pr.pr_info.residx = lun->pr_res_idx;
8398                         persis_io.pr.pr_info.res_type = type;
8399                         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8400                             sizeof(persis_io.pr), M_WAITOK);
8401                 }
8402                 break;
8403
8404         case SPRO_RELEASE:
8405                 mtx_lock(&lun->lun_lock);
8406                 if ((lun->flags & CTL_LUN_PR_RESERVED) == 0) {
8407                         /* No reservation exists return good status */
8408                         mtx_unlock(&lun->lun_lock);
8409                         goto done;
8410                 }
8411                 /*
8412                  * Is this nexus a reservation holder?
8413                  */
8414                 if (lun->pr_res_idx != residx
8415                  && lun->pr_res_idx != CTL_PR_ALL_REGISTRANTS) {
8416                         /*
8417                          * not a res holder return good status but
8418                          * do nothing
8419                          */
8420                         mtx_unlock(&lun->lun_lock);
8421                         goto done;
8422                 }
8423
8424                 if (lun->pr_res_type != type) {
8425                         mtx_unlock(&lun->lun_lock);
8426                         free(ctsio->kern_data_ptr, M_CTL);
8427                         ctl_set_illegal_pr_release(ctsio);
8428                         ctl_done((union ctl_io *)ctsio);
8429                         return (CTL_RETVAL_COMPLETE);
8430                 }
8431
8432                 /* okay to release */
8433                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8434                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8435                 lun->pr_res_type = 0;
8436
8437                 /*
8438                  * if this isn't an exclusive access
8439                  * res generate UA for all other
8440                  * registrants.
8441                  */
8442                 if (type != SPR_TYPE_EX_AC
8443                  && type != SPR_TYPE_WR_EX) {
8444                         for (i = softc->init_min; i < softc->init_max; i++) {
8445                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8446                                         continue;
8447                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8448                         }
8449                 }
8450                 mtx_unlock(&lun->lun_lock);
8451
8452                 /* Send msg to other side */
8453                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8454                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8455                 persis_io.pr.pr_info.action = CTL_PR_RELEASE;
8456                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8457                      sizeof(persis_io.pr), M_WAITOK);
8458                 break;
8459
8460         case SPRO_CLEAR:
8461                 /* send msg to other side */
8462
8463                 mtx_lock(&lun->lun_lock);
8464                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8465                 lun->pr_res_type = 0;
8466                 lun->pr_key_count = 0;
8467                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8468
8469                 ctl_clr_prkey(lun, residx);
8470                 for (i = 0; i < CTL_MAX_INITIATORS; i++)
8471                         if (ctl_get_prkey(lun, i) != 0) {
8472                                 ctl_clr_prkey(lun, i);
8473                                 ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8474                         }
8475                 lun->pr_generation++;
8476                 mtx_unlock(&lun->lun_lock);
8477
8478                 persis_io.hdr.nexus = ctsio->io_hdr.nexus;
8479                 persis_io.hdr.msg_type = CTL_MSG_PERS_ACTION;
8480                 persis_io.pr.pr_info.action = CTL_PR_CLEAR;
8481                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &persis_io,
8482                      sizeof(persis_io.pr), M_WAITOK);
8483                 break;
8484
8485         case SPRO_PREEMPT:
8486         case SPRO_PRE_ABO: {
8487                 int nretval;
8488
8489                 nretval = ctl_pro_preempt(softc, lun, res_key, sa_res_key, type,
8490                                           residx, ctsio, cdb, param);
8491                 if (nretval != 0)
8492                         return (CTL_RETVAL_COMPLETE);
8493                 break;
8494         }
8495         default:
8496                 panic("%s: Invalid PR type %#x", __func__, cdb->action);
8497         }
8498
8499 done:
8500         free(ctsio->kern_data_ptr, M_CTL);
8501         ctl_set_success(ctsio);
8502         ctl_done((union ctl_io *)ctsio);
8503
8504         return (retval);
8505 }
8506
8507 /*
8508  * This routine is for handling a message from the other SC pertaining to
8509  * persistent reserve out. All the error checking will have been done
8510  * so only perorming the action need be done here to keep the two
8511  * in sync.
8512  */
8513 static void
8514 ctl_hndl_per_res_out_on_other_sc(union ctl_ha_msg *msg)
8515 {
8516         struct ctl_softc *softc = control_softc;
8517         struct ctl_lun *lun;
8518         int i;
8519         uint32_t residx, targ_lun;
8520
8521         targ_lun = msg->hdr.nexus.targ_mapped_lun;
8522         mtx_lock(&softc->ctl_lock);
8523         if ((targ_lun >= CTL_MAX_LUNS) ||
8524             ((lun = softc->ctl_luns[targ_lun]) == NULL)) {
8525                 mtx_unlock(&softc->ctl_lock);
8526                 return;
8527         }
8528         mtx_lock(&lun->lun_lock);
8529         mtx_unlock(&softc->ctl_lock);
8530         if (lun->flags & CTL_LUN_DISABLED) {
8531                 mtx_unlock(&lun->lun_lock);
8532                 return;
8533         }
8534         residx = ctl_get_initindex(&msg->hdr.nexus);
8535         switch(msg->pr.pr_info.action) {
8536         case CTL_PR_REG_KEY:
8537                 ctl_alloc_prkey(lun, msg->pr.pr_info.residx);
8538                 if (ctl_get_prkey(lun, msg->pr.pr_info.residx) == 0)
8539                         lun->pr_key_count++;
8540                 ctl_set_prkey(lun, msg->pr.pr_info.residx,
8541                     scsi_8btou64(msg->pr.pr_info.sa_res_key));
8542                 lun->pr_generation++;
8543                 break;
8544
8545         case CTL_PR_UNREG_KEY:
8546                 ctl_clr_prkey(lun, msg->pr.pr_info.residx);
8547                 lun->pr_key_count--;
8548
8549                 /* XXX Need to see if the reservation has been released */
8550                 /* if so do we need to generate UA? */
8551                 if (msg->pr.pr_info.residx == lun->pr_res_idx) {
8552                         lun->flags &= ~CTL_LUN_PR_RESERVED;
8553                         lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8554
8555                         if ((lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
8556                              lun->pr_res_type == SPR_TYPE_EX_AC_RO) &&
8557                             lun->pr_key_count) {
8558                                 /*
8559                                  * If the reservation is a registrants
8560                                  * only type we need to generate a UA
8561                                  * for other registered inits.  The
8562                                  * sense code should be RESERVATIONS
8563                                  * RELEASED
8564                                  */
8565
8566                                 for (i = softc->init_min; i < softc->init_max; i++) {
8567                                         if (ctl_get_prkey(lun, i) == 0)
8568                                                 continue;
8569
8570                                         ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8571                                 }
8572                         }
8573                         lun->pr_res_type = 0;
8574                 } else if (lun->pr_res_idx == CTL_PR_ALL_REGISTRANTS) {
8575                         if (lun->pr_key_count==0) {
8576                                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8577                                 lun->pr_res_type = 0;
8578                                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8579                         }
8580                 }
8581                 lun->pr_generation++;
8582                 break;
8583
8584         case CTL_PR_RESERVE:
8585                 lun->flags |= CTL_LUN_PR_RESERVED;
8586                 lun->pr_res_type = msg->pr.pr_info.res_type;
8587                 lun->pr_res_idx = msg->pr.pr_info.residx;
8588
8589                 break;
8590
8591         case CTL_PR_RELEASE:
8592                 /*
8593                  * if this isn't an exclusive access res generate UA for all
8594                  * other registrants.
8595                  */
8596                 if (lun->pr_res_type != SPR_TYPE_EX_AC &&
8597                     lun->pr_res_type != SPR_TYPE_WR_EX) {
8598                         for (i = softc->init_min; i < softc->init_max; i++)
8599                                 if (i == residx || ctl_get_prkey(lun, i) == 0)
8600                                         continue;
8601                                 ctl_est_ua(lun, i, CTL_UA_RES_RELEASE);
8602                 }
8603
8604                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8605                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8606                 lun->pr_res_type = 0;
8607                 break;
8608
8609         case CTL_PR_PREEMPT:
8610                 ctl_pro_preempt_other(lun, msg);
8611                 break;
8612         case CTL_PR_CLEAR:
8613                 lun->flags &= ~CTL_LUN_PR_RESERVED;
8614                 lun->pr_res_type = 0;
8615                 lun->pr_key_count = 0;
8616                 lun->pr_res_idx = CTL_PR_NO_RESERVATION;
8617
8618                 for (i=0; i < CTL_MAX_INITIATORS; i++) {
8619                         if (ctl_get_prkey(lun, i) == 0)
8620                                 continue;
8621                         ctl_clr_prkey(lun, i);
8622                         ctl_est_ua(lun, i, CTL_UA_REG_PREEMPT);
8623                 }
8624                 lun->pr_generation++;
8625                 break;
8626         }
8627
8628         mtx_unlock(&lun->lun_lock);
8629 }
8630
8631 int
8632 ctl_read_write(struct ctl_scsiio *ctsio)
8633 {
8634         struct ctl_lun *lun;
8635         struct ctl_lba_len_flags *lbalen;
8636         uint64_t lba;
8637         uint32_t num_blocks;
8638         int flags, retval;
8639         int isread;
8640
8641         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8642
8643         CTL_DEBUG_PRINT(("ctl_read_write: command: %#x\n", ctsio->cdb[0]));
8644
8645         flags = 0;
8646         isread = ctsio->cdb[0] == READ_6  || ctsio->cdb[0] == READ_10
8647               || ctsio->cdb[0] == READ_12 || ctsio->cdb[0] == READ_16;
8648         switch (ctsio->cdb[0]) {
8649         case READ_6:
8650         case WRITE_6: {
8651                 struct scsi_rw_6 *cdb;
8652
8653                 cdb = (struct scsi_rw_6 *)ctsio->cdb;
8654
8655                 lba = scsi_3btoul(cdb->addr);
8656                 /* only 5 bits are valid in the most significant address byte */
8657                 lba &= 0x1fffff;
8658                 num_blocks = cdb->length;
8659                 /*
8660                  * This is correct according to SBC-2.
8661                  */
8662                 if (num_blocks == 0)
8663                         num_blocks = 256;
8664                 break;
8665         }
8666         case READ_10:
8667         case WRITE_10: {
8668                 struct scsi_rw_10 *cdb;
8669
8670                 cdb = (struct scsi_rw_10 *)ctsio->cdb;
8671                 if (cdb->byte2 & SRW10_FUA)
8672                         flags |= CTL_LLF_FUA;
8673                 if (cdb->byte2 & SRW10_DPO)
8674                         flags |= CTL_LLF_DPO;
8675                 lba = scsi_4btoul(cdb->addr);
8676                 num_blocks = scsi_2btoul(cdb->length);
8677                 break;
8678         }
8679         case WRITE_VERIFY_10: {
8680                 struct scsi_write_verify_10 *cdb;
8681
8682                 cdb = (struct scsi_write_verify_10 *)ctsio->cdb;
8683                 flags |= CTL_LLF_FUA;
8684                 if (cdb->byte2 & SWV_DPO)
8685                         flags |= CTL_LLF_DPO;
8686                 lba = scsi_4btoul(cdb->addr);
8687                 num_blocks = scsi_2btoul(cdb->length);
8688                 break;
8689         }
8690         case READ_12:
8691         case WRITE_12: {
8692                 struct scsi_rw_12 *cdb;
8693
8694                 cdb = (struct scsi_rw_12 *)ctsio->cdb;
8695                 if (cdb->byte2 & SRW12_FUA)
8696                         flags |= CTL_LLF_FUA;
8697                 if (cdb->byte2 & SRW12_DPO)
8698                         flags |= CTL_LLF_DPO;
8699                 lba = scsi_4btoul(cdb->addr);
8700                 num_blocks = scsi_4btoul(cdb->length);
8701                 break;
8702         }
8703         case WRITE_VERIFY_12: {
8704                 struct scsi_write_verify_12 *cdb;
8705
8706                 cdb = (struct scsi_write_verify_12 *)ctsio->cdb;
8707                 flags |= CTL_LLF_FUA;
8708                 if (cdb->byte2 & SWV_DPO)
8709                         flags |= CTL_LLF_DPO;
8710                 lba = scsi_4btoul(cdb->addr);
8711                 num_blocks = scsi_4btoul(cdb->length);
8712                 break;
8713         }
8714         case READ_16:
8715         case WRITE_16: {
8716                 struct scsi_rw_16 *cdb;
8717
8718                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8719                 if (cdb->byte2 & SRW12_FUA)
8720                         flags |= CTL_LLF_FUA;
8721                 if (cdb->byte2 & SRW12_DPO)
8722                         flags |= CTL_LLF_DPO;
8723                 lba = scsi_8btou64(cdb->addr);
8724                 num_blocks = scsi_4btoul(cdb->length);
8725                 break;
8726         }
8727         case WRITE_ATOMIC_16: {
8728                 struct scsi_write_atomic_16 *cdb;
8729
8730                 if (lun->be_lun->atomicblock == 0) {
8731                         ctl_set_invalid_opcode(ctsio);
8732                         ctl_done((union ctl_io *)ctsio);
8733                         return (CTL_RETVAL_COMPLETE);
8734                 }
8735
8736                 cdb = (struct scsi_write_atomic_16 *)ctsio->cdb;
8737                 if (cdb->byte2 & SRW12_FUA)
8738                         flags |= CTL_LLF_FUA;
8739                 if (cdb->byte2 & SRW12_DPO)
8740                         flags |= CTL_LLF_DPO;
8741                 lba = scsi_8btou64(cdb->addr);
8742                 num_blocks = scsi_2btoul(cdb->length);
8743                 if (num_blocks > lun->be_lun->atomicblock) {
8744                         ctl_set_invalid_field(ctsio, /*sks_valid*/ 1,
8745                             /*command*/ 1, /*field*/ 12, /*bit_valid*/ 0,
8746                             /*bit*/ 0);
8747                         ctl_done((union ctl_io *)ctsio);
8748                         return (CTL_RETVAL_COMPLETE);
8749                 }
8750                 break;
8751         }
8752         case WRITE_VERIFY_16: {
8753                 struct scsi_write_verify_16 *cdb;
8754
8755                 cdb = (struct scsi_write_verify_16 *)ctsio->cdb;
8756                 flags |= CTL_LLF_FUA;
8757                 if (cdb->byte2 & SWV_DPO)
8758                         flags |= CTL_LLF_DPO;
8759                 lba = scsi_8btou64(cdb->addr);
8760                 num_blocks = scsi_4btoul(cdb->length);
8761                 break;
8762         }
8763         default:
8764                 /*
8765                  * We got a command we don't support.  This shouldn't
8766                  * happen, commands should be filtered out above us.
8767                  */
8768                 ctl_set_invalid_opcode(ctsio);
8769                 ctl_done((union ctl_io *)ctsio);
8770
8771                 return (CTL_RETVAL_COMPLETE);
8772                 break; /* NOTREACHED */
8773         }
8774
8775         /*
8776          * The first check is to make sure we're in bounds, the second
8777          * check is to catch wrap-around problems.  If the lba + num blocks
8778          * is less than the lba, then we've wrapped around and the block
8779          * range is invalid anyway.
8780          */
8781         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8782          || ((lba + num_blocks) < lba)) {
8783                 ctl_set_lba_out_of_range(ctsio);
8784                 ctl_done((union ctl_io *)ctsio);
8785                 return (CTL_RETVAL_COMPLETE);
8786         }
8787
8788         /*
8789          * According to SBC-3, a transfer length of 0 is not an error.
8790          * Note that this cannot happen with WRITE(6) or READ(6), since 0
8791          * translates to 256 blocks for those commands.
8792          */
8793         if (num_blocks == 0) {
8794                 ctl_set_success(ctsio);
8795                 ctl_done((union ctl_io *)ctsio);
8796                 return (CTL_RETVAL_COMPLETE);
8797         }
8798
8799         /* Set FUA and/or DPO if caches are disabled. */
8800         if (isread) {
8801                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8802                     SCP_RCD) != 0)
8803                         flags |= CTL_LLF_FUA | CTL_LLF_DPO;
8804         } else {
8805                 if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8806                     SCP_WCE) == 0)
8807                         flags |= CTL_LLF_FUA;
8808         }
8809
8810         lbalen = (struct ctl_lba_len_flags *)
8811             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8812         lbalen->lba = lba;
8813         lbalen->len = num_blocks;
8814         lbalen->flags = (isread ? CTL_LLF_READ : CTL_LLF_WRITE) | flags;
8815
8816         ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
8817         ctsio->kern_rel_offset = 0;
8818
8819         CTL_DEBUG_PRINT(("ctl_read_write: calling data_submit()\n"));
8820
8821         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8822         return (retval);
8823 }
8824
8825 static int
8826 ctl_cnw_cont(union ctl_io *io)
8827 {
8828         struct ctl_scsiio *ctsio;
8829         struct ctl_lun *lun;
8830         struct ctl_lba_len_flags *lbalen;
8831         int retval;
8832
8833         ctsio = &io->scsiio;
8834         ctsio->io_hdr.status = CTL_STATUS_NONE;
8835         ctsio->io_hdr.flags &= ~CTL_FLAG_IO_CONT;
8836         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8837         lbalen = (struct ctl_lba_len_flags *)
8838             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8839         lbalen->flags &= ~CTL_LLF_COMPARE;
8840         lbalen->flags |= CTL_LLF_WRITE;
8841
8842         CTL_DEBUG_PRINT(("ctl_cnw_cont: calling data_submit()\n"));
8843         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8844         return (retval);
8845 }
8846
8847 int
8848 ctl_cnw(struct ctl_scsiio *ctsio)
8849 {
8850         struct ctl_lun *lun;
8851         struct ctl_lba_len_flags *lbalen;
8852         uint64_t lba;
8853         uint32_t num_blocks;
8854         int flags, retval;
8855
8856         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8857
8858         CTL_DEBUG_PRINT(("ctl_cnw: command: %#x\n", ctsio->cdb[0]));
8859
8860         flags = 0;
8861         switch (ctsio->cdb[0]) {
8862         case COMPARE_AND_WRITE: {
8863                 struct scsi_compare_and_write *cdb;
8864
8865                 cdb = (struct scsi_compare_and_write *)ctsio->cdb;
8866                 if (cdb->byte2 & SRW10_FUA)
8867                         flags |= CTL_LLF_FUA;
8868                 if (cdb->byte2 & SRW10_DPO)
8869                         flags |= CTL_LLF_DPO;
8870                 lba = scsi_8btou64(cdb->addr);
8871                 num_blocks = cdb->length;
8872                 break;
8873         }
8874         default:
8875                 /*
8876                  * We got a command we don't support.  This shouldn't
8877                  * happen, commands should be filtered out above us.
8878                  */
8879                 ctl_set_invalid_opcode(ctsio);
8880                 ctl_done((union ctl_io *)ctsio);
8881
8882                 return (CTL_RETVAL_COMPLETE);
8883                 break; /* NOTREACHED */
8884         }
8885
8886         /*
8887          * The first check is to make sure we're in bounds, the second
8888          * check is to catch wrap-around problems.  If the lba + num blocks
8889          * is less than the lba, then we've wrapped around and the block
8890          * range is invalid anyway.
8891          */
8892         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
8893          || ((lba + num_blocks) < lba)) {
8894                 ctl_set_lba_out_of_range(ctsio);
8895                 ctl_done((union ctl_io *)ctsio);
8896                 return (CTL_RETVAL_COMPLETE);
8897         }
8898
8899         /*
8900          * According to SBC-3, a transfer length of 0 is not an error.
8901          */
8902         if (num_blocks == 0) {
8903                 ctl_set_success(ctsio);
8904                 ctl_done((union ctl_io *)ctsio);
8905                 return (CTL_RETVAL_COMPLETE);
8906         }
8907
8908         /* Set FUA if write cache is disabled. */
8909         if ((lun->mode_pages.caching_page[CTL_PAGE_CURRENT].flags1 &
8910             SCP_WCE) == 0)
8911                 flags |= CTL_LLF_FUA;
8912
8913         ctsio->kern_total_len = 2 * num_blocks * lun->be_lun->blocksize;
8914         ctsio->kern_rel_offset = 0;
8915
8916         /*
8917          * Set the IO_CONT flag, so that if this I/O gets passed to
8918          * ctl_data_submit_done(), it'll get passed back to
8919          * ctl_ctl_cnw_cont() for further processing.
8920          */
8921         ctsio->io_hdr.flags |= CTL_FLAG_IO_CONT;
8922         ctsio->io_cont = ctl_cnw_cont;
8923
8924         lbalen = (struct ctl_lba_len_flags *)
8925             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
8926         lbalen->lba = lba;
8927         lbalen->len = num_blocks;
8928         lbalen->flags = CTL_LLF_COMPARE | flags;
8929
8930         CTL_DEBUG_PRINT(("ctl_cnw: calling data_submit()\n"));
8931         retval = lun->backend->data_submit((union ctl_io *)ctsio);
8932         return (retval);
8933 }
8934
8935 int
8936 ctl_verify(struct ctl_scsiio *ctsio)
8937 {
8938         struct ctl_lun *lun;
8939         struct ctl_lba_len_flags *lbalen;
8940         uint64_t lba;
8941         uint32_t num_blocks;
8942         int bytchk, flags;
8943         int retval;
8944
8945         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
8946
8947         CTL_DEBUG_PRINT(("ctl_verify: command: %#x\n", ctsio->cdb[0]));
8948
8949         bytchk = 0;
8950         flags = CTL_LLF_FUA;
8951         switch (ctsio->cdb[0]) {
8952         case VERIFY_10: {
8953                 struct scsi_verify_10 *cdb;
8954
8955                 cdb = (struct scsi_verify_10 *)ctsio->cdb;
8956                 if (cdb->byte2 & SVFY_BYTCHK)
8957                         bytchk = 1;
8958                 if (cdb->byte2 & SVFY_DPO)
8959                         flags |= CTL_LLF_DPO;
8960                 lba = scsi_4btoul(cdb->addr);
8961                 num_blocks = scsi_2btoul(cdb->length);
8962                 break;
8963         }
8964         case VERIFY_12: {
8965                 struct scsi_verify_12 *cdb;
8966
8967                 cdb = (struct scsi_verify_12 *)ctsio->cdb;
8968                 if (cdb->byte2 & SVFY_BYTCHK)
8969                         bytchk = 1;
8970                 if (cdb->byte2 & SVFY_DPO)
8971                         flags |= CTL_LLF_DPO;
8972                 lba = scsi_4btoul(cdb->addr);
8973                 num_blocks = scsi_4btoul(cdb->length);
8974                 break;
8975         }
8976         case VERIFY_16: {
8977                 struct scsi_rw_16 *cdb;
8978
8979                 cdb = (struct scsi_rw_16 *)ctsio->cdb;
8980                 if (cdb->byte2 & SVFY_BYTCHK)
8981                         bytchk = 1;
8982                 if (cdb->byte2 & SVFY_DPO)
8983                         flags |= CTL_LLF_DPO;
8984                 lba = scsi_8btou64(cdb->addr);
8985                 num_blocks = scsi_4btoul(cdb->length);
8986                 break;
8987         }
8988         default:
8989                 /*
8990                  * We got a command we don't support.  This shouldn't
8991                  * happen, commands should be filtered out above us.
8992                  */
8993                 ctl_set_invalid_opcode(ctsio);
8994                 ctl_done((union ctl_io *)ctsio);
8995                 return (CTL_RETVAL_COMPLETE);
8996         }
8997
8998         /*
8999          * The first check is to make sure we're in bounds, the second
9000          * check is to catch wrap-around problems.  If the lba + num blocks
9001          * is less than the lba, then we've wrapped around and the block
9002          * range is invalid anyway.
9003          */
9004         if (((lba + num_blocks) > (lun->be_lun->maxlba + 1))
9005          || ((lba + num_blocks) < lba)) {
9006                 ctl_set_lba_out_of_range(ctsio);
9007                 ctl_done((union ctl_io *)ctsio);
9008                 return (CTL_RETVAL_COMPLETE);
9009         }
9010
9011         /*
9012          * According to SBC-3, a transfer length of 0 is not an error.
9013          */
9014         if (num_blocks == 0) {
9015                 ctl_set_success(ctsio);
9016                 ctl_done((union ctl_io *)ctsio);
9017                 return (CTL_RETVAL_COMPLETE);
9018         }
9019
9020         lbalen = (struct ctl_lba_len_flags *)
9021             &ctsio->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
9022         lbalen->lba = lba;
9023         lbalen->len = num_blocks;
9024         if (bytchk) {
9025                 lbalen->flags = CTL_LLF_COMPARE | flags;
9026                 ctsio->kern_total_len = num_blocks * lun->be_lun->blocksize;
9027         } else {
9028                 lbalen->flags = CTL_LLF_VERIFY | flags;
9029                 ctsio->kern_total_len = 0;
9030         }
9031         ctsio->kern_rel_offset = 0;
9032
9033         CTL_DEBUG_PRINT(("ctl_verify: calling data_submit()\n"));
9034         retval = lun->backend->data_submit((union ctl_io *)ctsio);
9035         return (retval);
9036 }
9037
9038 int
9039 ctl_report_luns(struct ctl_scsiio *ctsio)
9040 {
9041         struct ctl_softc *softc;
9042         struct scsi_report_luns *cdb;
9043         struct scsi_report_luns_data *lun_data;
9044         struct ctl_lun *lun, *request_lun;
9045         struct ctl_port *port;
9046         int num_luns, retval;
9047         uint32_t alloc_len, lun_datalen;
9048         int num_filled;
9049         uint32_t initidx, targ_lun_id, lun_id;
9050
9051         retval = CTL_RETVAL_COMPLETE;
9052         cdb = (struct scsi_report_luns *)ctsio->cdb;
9053         port = ctl_io_port(&ctsio->io_hdr);
9054         softc = port->ctl_softc;
9055
9056         CTL_DEBUG_PRINT(("ctl_report_luns\n"));
9057
9058         mtx_lock(&softc->ctl_lock);
9059         num_luns = 0;
9060         for (targ_lun_id = 0; targ_lun_id < CTL_MAX_LUNS; targ_lun_id++) {
9061                 if (ctl_lun_map_from_port(port, targ_lun_id) < CTL_MAX_LUNS)
9062                         num_luns++;
9063         }
9064         mtx_unlock(&softc->ctl_lock);
9065
9066         switch (cdb->select_report) {
9067         case RPL_REPORT_DEFAULT:
9068         case RPL_REPORT_ALL:
9069         case RPL_REPORT_NONSUBSID:
9070                 break;
9071         case RPL_REPORT_WELLKNOWN:
9072         case RPL_REPORT_ADMIN:
9073         case RPL_REPORT_CONGLOM:
9074                 num_luns = 0;
9075                 break;
9076         default:
9077                 ctl_set_invalid_field(ctsio,
9078                                       /*sks_valid*/ 1,
9079                                       /*command*/ 1,
9080                                       /*field*/ 2,
9081                                       /*bit_valid*/ 0,
9082                                       /*bit*/ 0);
9083                 ctl_done((union ctl_io *)ctsio);
9084                 return (retval);
9085                 break; /* NOTREACHED */
9086         }
9087
9088         alloc_len = scsi_4btoul(cdb->length);
9089         /*
9090          * The initiator has to allocate at least 16 bytes for this request,
9091          * so he can at least get the header and the first LUN.  Otherwise
9092          * we reject the request (per SPC-3 rev 14, section 6.21).
9093          */
9094         if (alloc_len < (sizeof(struct scsi_report_luns_data) +
9095             sizeof(struct scsi_report_luns_lundata))) {
9096                 ctl_set_invalid_field(ctsio,
9097                                       /*sks_valid*/ 1,
9098                                       /*command*/ 1,
9099                                       /*field*/ 6,
9100                                       /*bit_valid*/ 0,
9101                                       /*bit*/ 0);
9102                 ctl_done((union ctl_io *)ctsio);
9103                 return (retval);
9104         }
9105
9106         request_lun = (struct ctl_lun *)
9107                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9108
9109         lun_datalen = sizeof(*lun_data) +
9110                 (num_luns * sizeof(struct scsi_report_luns_lundata));
9111
9112         ctsio->kern_data_ptr = malloc(lun_datalen, M_CTL, M_WAITOK | M_ZERO);
9113         lun_data = (struct scsi_report_luns_data *)ctsio->kern_data_ptr;
9114         ctsio->kern_sg_entries = 0;
9115
9116         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9117
9118         mtx_lock(&softc->ctl_lock);
9119         for (targ_lun_id = 0, num_filled = 0; targ_lun_id < CTL_MAX_LUNS && num_filled < num_luns; targ_lun_id++) {
9120                 lun_id = ctl_lun_map_from_port(port, targ_lun_id);
9121                 if (lun_id >= CTL_MAX_LUNS)
9122                         continue;
9123                 lun = softc->ctl_luns[lun_id];
9124                 if (lun == NULL)
9125                         continue;
9126
9127                 be64enc(lun_data->luns[num_filled++].lundata,
9128                     ctl_encode_lun(targ_lun_id));
9129
9130                 /*
9131                  * According to SPC-3, rev 14 section 6.21:
9132                  *
9133                  * "The execution of a REPORT LUNS command to any valid and
9134                  * installed logical unit shall clear the REPORTED LUNS DATA
9135                  * HAS CHANGED unit attention condition for all logical
9136                  * units of that target with respect to the requesting
9137                  * initiator. A valid and installed logical unit is one
9138                  * having a PERIPHERAL QUALIFIER of 000b in the standard
9139                  * INQUIRY data (see 6.4.2)."
9140                  *
9141                  * If request_lun is NULL, the LUN this report luns command
9142                  * was issued to is either disabled or doesn't exist. In that
9143                  * case, we shouldn't clear any pending lun change unit
9144                  * attention.
9145                  */
9146                 if (request_lun != NULL) {
9147                         mtx_lock(&lun->lun_lock);
9148                         ctl_clr_ua(lun, initidx, CTL_UA_LUN_CHANGE);
9149                         mtx_unlock(&lun->lun_lock);
9150                 }
9151         }
9152         mtx_unlock(&softc->ctl_lock);
9153
9154         /*
9155          * It's quite possible that we've returned fewer LUNs than we allocated
9156          * space for.  Trim it.
9157          */
9158         lun_datalen = sizeof(*lun_data) +
9159                 (num_filled * sizeof(struct scsi_report_luns_lundata));
9160
9161         if (lun_datalen < alloc_len) {
9162                 ctsio->residual = alloc_len - lun_datalen;
9163                 ctsio->kern_data_len = lun_datalen;
9164                 ctsio->kern_total_len = lun_datalen;
9165         } else {
9166                 ctsio->residual = 0;
9167                 ctsio->kern_data_len = alloc_len;
9168                 ctsio->kern_total_len = alloc_len;
9169         }
9170         ctsio->kern_data_resid = 0;
9171         ctsio->kern_rel_offset = 0;
9172         ctsio->kern_sg_entries = 0;
9173
9174         /*
9175          * We set this to the actual data length, regardless of how much
9176          * space we actually have to return results.  If the user looks at
9177          * this value, he'll know whether or not he allocated enough space
9178          * and reissue the command if necessary.  We don't support well
9179          * known logical units, so if the user asks for that, return none.
9180          */
9181         scsi_ulto4b(lun_datalen - 8, lun_data->length);
9182
9183         /*
9184          * We can only return SCSI_STATUS_CHECK_COND when we can't satisfy
9185          * this request.
9186          */
9187         ctl_set_success(ctsio);
9188         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9189         ctsio->be_move_done = ctl_config_move_done;
9190         ctl_datamove((union ctl_io *)ctsio);
9191         return (retval);
9192 }
9193
9194 int
9195 ctl_request_sense(struct ctl_scsiio *ctsio)
9196 {
9197         struct scsi_request_sense *cdb;
9198         struct scsi_sense_data *sense_ptr;
9199         struct ctl_softc *softc;
9200         struct ctl_lun *lun;
9201         uint32_t initidx;
9202         int have_error;
9203         scsi_sense_data_type sense_format;
9204         ctl_ua_type ua_type;
9205         uint8_t asc = 0, ascq = 0;
9206
9207         cdb = (struct scsi_request_sense *)ctsio->cdb;
9208
9209         softc = control_softc;
9210         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9211
9212         CTL_DEBUG_PRINT(("ctl_request_sense\n"));
9213
9214         /*
9215          * Determine which sense format the user wants.
9216          */
9217         if (cdb->byte2 & SRS_DESC)
9218                 sense_format = SSD_TYPE_DESC;
9219         else
9220                 sense_format = SSD_TYPE_FIXED;
9221
9222         ctsio->kern_data_ptr = malloc(sizeof(*sense_ptr), M_CTL, M_WAITOK);
9223         sense_ptr = (struct scsi_sense_data *)ctsio->kern_data_ptr;
9224         ctsio->kern_sg_entries = 0;
9225
9226         /*
9227          * struct scsi_sense_data, which is currently set to 256 bytes, is
9228          * larger than the largest allowed value for the length field in the
9229          * REQUEST SENSE CDB, which is 252 bytes as of SPC-4.
9230          */
9231         ctsio->residual = 0;
9232         ctsio->kern_data_len = cdb->length;
9233         ctsio->kern_total_len = cdb->length;
9234
9235         ctsio->kern_data_resid = 0;
9236         ctsio->kern_rel_offset = 0;
9237         ctsio->kern_sg_entries = 0;
9238
9239         /*
9240          * If we don't have a LUN, we don't have any pending sense.
9241          */
9242         if (lun == NULL ||
9243             ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
9244              softc->ha_link < CTL_HA_LINK_UNKNOWN)) {
9245                 /* "Logical unit not supported" */
9246                 ctl_set_sense_data(sense_ptr, NULL, sense_format,
9247                     /*current_error*/ 1,
9248                     /*sense_key*/ SSD_KEY_ILLEGAL_REQUEST,
9249                     /*asc*/ 0x25,
9250                     /*ascq*/ 0x00,
9251                     SSD_ELEM_NONE);
9252                 goto send;
9253         }
9254
9255         have_error = 0;
9256         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
9257         /*
9258          * Check for pending sense, and then for pending unit attentions.
9259          * Pending sense gets returned first, then pending unit attentions.
9260          */
9261         mtx_lock(&lun->lun_lock);
9262 #ifdef CTL_WITH_CA
9263         if (ctl_is_set(lun->have_ca, initidx)) {
9264                 scsi_sense_data_type stored_format;
9265
9266                 /*
9267                  * Check to see which sense format was used for the stored
9268                  * sense data.
9269                  */
9270                 stored_format = scsi_sense_type(&lun->pending_sense[initidx]);
9271
9272                 /*
9273                  * If the user requested a different sense format than the
9274                  * one we stored, then we need to convert it to the other
9275                  * format.  If we're going from descriptor to fixed format
9276                  * sense data, we may lose things in translation, depending
9277                  * on what options were used.
9278                  *
9279                  * If the stored format is SSD_TYPE_NONE (i.e. invalid),
9280                  * for some reason we'll just copy it out as-is.
9281                  */
9282                 if ((stored_format == SSD_TYPE_FIXED)
9283                  && (sense_format == SSD_TYPE_DESC))
9284                         ctl_sense_to_desc((struct scsi_sense_data_fixed *)
9285                             &lun->pending_sense[initidx],
9286                             (struct scsi_sense_data_desc *)sense_ptr);
9287                 else if ((stored_format == SSD_TYPE_DESC)
9288                       && (sense_format == SSD_TYPE_FIXED))
9289                         ctl_sense_to_fixed((struct scsi_sense_data_desc *)
9290                             &lun->pending_sense[initidx],
9291                             (struct scsi_sense_data_fixed *)sense_ptr);
9292                 else
9293                         memcpy(sense_ptr, &lun->pending_sense[initidx],
9294                                MIN(sizeof(*sense_ptr),
9295                                sizeof(lun->pending_sense[initidx])));
9296
9297                 ctl_clear_mask(lun->have_ca, initidx);
9298                 have_error = 1;
9299         } else
9300 #endif
9301         if (have_error == 0) {
9302                 ua_type = ctl_build_ua(lun, initidx, sense_ptr, sense_format);
9303                 if (ua_type != CTL_UA_NONE)
9304                         have_error = 1;
9305                 if (ua_type == CTL_UA_LUN_CHANGE) {
9306                         mtx_unlock(&lun->lun_lock);
9307                         mtx_lock(&softc->ctl_lock);
9308                         ctl_clr_ua_allluns(softc, initidx, ua_type);
9309                         mtx_unlock(&softc->ctl_lock);
9310                         mtx_lock(&lun->lun_lock);
9311                 }
9312         }
9313         if (have_error == 0) {
9314                 /*
9315                  * Report informational exception if have one and allowed.
9316                  */
9317                 if (lun->mode_pages.ie_page[CTL_PAGE_CURRENT].mrie != SIEP_MRIE_NO) {
9318                         asc = lun->ie_asc;
9319                         ascq = lun->ie_ascq;
9320                 }
9321                 ctl_set_sense_data(sense_ptr, lun, sense_format,
9322                     /*current_error*/ 1,
9323                     /*sense_key*/ SSD_KEY_NO_SENSE,
9324                     /*asc*/ asc,
9325                     /*ascq*/ ascq,
9326                     SSD_ELEM_NONE);
9327         }
9328         mtx_unlock(&lun->lun_lock);
9329
9330 send:
9331         /*
9332          * We report the SCSI status as OK, since the status of the command
9333          * itself is OK.  We're reporting sense as parameter data.
9334          */
9335         ctl_set_success(ctsio);
9336         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9337         ctsio->be_move_done = ctl_config_move_done;
9338         ctl_datamove((union ctl_io *)ctsio);
9339         return (CTL_RETVAL_COMPLETE);
9340 }
9341
9342 int
9343 ctl_tur(struct ctl_scsiio *ctsio)
9344 {
9345
9346         CTL_DEBUG_PRINT(("ctl_tur\n"));
9347
9348         ctl_set_success(ctsio);
9349         ctl_done((union ctl_io *)ctsio);
9350
9351         return (CTL_RETVAL_COMPLETE);
9352 }
9353
9354 /*
9355  * SCSI VPD page 0x00, the Supported VPD Pages page.
9356  */
9357 static int
9358 ctl_inquiry_evpd_supported(struct ctl_scsiio *ctsio, int alloc_len)
9359 {
9360         struct scsi_vpd_supported_pages *pages;
9361         int sup_page_size;
9362         struct ctl_lun *lun;
9363         int p;
9364
9365         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9366
9367         sup_page_size = sizeof(struct scsi_vpd_supported_pages) *
9368             SCSI_EVPD_NUM_SUPPORTED_PAGES;
9369         ctsio->kern_data_ptr = malloc(sup_page_size, M_CTL, M_WAITOK | M_ZERO);
9370         pages = (struct scsi_vpd_supported_pages *)ctsio->kern_data_ptr;
9371         ctsio->kern_sg_entries = 0;
9372
9373         if (sup_page_size < alloc_len) {
9374                 ctsio->residual = alloc_len - sup_page_size;
9375                 ctsio->kern_data_len = sup_page_size;
9376                 ctsio->kern_total_len = sup_page_size;
9377         } else {
9378                 ctsio->residual = 0;
9379                 ctsio->kern_data_len = alloc_len;
9380                 ctsio->kern_total_len = alloc_len;
9381         }
9382         ctsio->kern_data_resid = 0;
9383         ctsio->kern_rel_offset = 0;
9384         ctsio->kern_sg_entries = 0;
9385
9386         /*
9387          * The control device is always connected.  The disk device, on the
9388          * other hand, may not be online all the time.  Need to change this
9389          * to figure out whether the disk device is actually online or not.
9390          */
9391         if (lun != NULL)
9392                 pages->device = (SID_QUAL_LU_CONNECTED << 5) |
9393                                 lun->be_lun->lun_type;
9394         else
9395                 pages->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9396
9397         p = 0;
9398         /* Supported VPD pages */
9399         pages->page_list[p++] = SVPD_SUPPORTED_PAGES;
9400         /* Serial Number */
9401         pages->page_list[p++] = SVPD_UNIT_SERIAL_NUMBER;
9402         /* Device Identification */
9403         pages->page_list[p++] = SVPD_DEVICE_ID;
9404         /* Extended INQUIRY Data */
9405         pages->page_list[p++] = SVPD_EXTENDED_INQUIRY_DATA;
9406         /* Mode Page Policy */
9407         pages->page_list[p++] = SVPD_MODE_PAGE_POLICY;
9408         /* SCSI Ports */
9409         pages->page_list[p++] = SVPD_SCSI_PORTS;
9410         /* Third-party Copy */
9411         pages->page_list[p++] = SVPD_SCSI_TPC;
9412         if (lun != NULL && lun->be_lun->lun_type == T_DIRECT) {
9413                 /* Block limits */
9414                 pages->page_list[p++] = SVPD_BLOCK_LIMITS;
9415                 /* Block Device Characteristics */
9416                 pages->page_list[p++] = SVPD_BDC;
9417                 /* Logical Block Provisioning */
9418                 pages->page_list[p++] = SVPD_LBP;
9419         }
9420         pages->length = p;
9421
9422         ctl_set_success(ctsio);
9423         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9424         ctsio->be_move_done = ctl_config_move_done;
9425         ctl_datamove((union ctl_io *)ctsio);
9426         return (CTL_RETVAL_COMPLETE);
9427 }
9428
9429 /*
9430  * SCSI VPD page 0x80, the Unit Serial Number page.
9431  */
9432 static int
9433 ctl_inquiry_evpd_serial(struct ctl_scsiio *ctsio, int alloc_len)
9434 {
9435         struct scsi_vpd_unit_serial_number *sn_ptr;
9436         struct ctl_lun *lun;
9437         int data_len;
9438
9439         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9440
9441         data_len = 4 + CTL_SN_LEN;
9442         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9443         sn_ptr = (struct scsi_vpd_unit_serial_number *)ctsio->kern_data_ptr;
9444         if (data_len < alloc_len) {
9445                 ctsio->residual = alloc_len - data_len;
9446                 ctsio->kern_data_len = data_len;
9447                 ctsio->kern_total_len = data_len;
9448         } else {
9449                 ctsio->residual = 0;
9450                 ctsio->kern_data_len = alloc_len;
9451                 ctsio->kern_total_len = alloc_len;
9452         }
9453         ctsio->kern_data_resid = 0;
9454         ctsio->kern_rel_offset = 0;
9455         ctsio->kern_sg_entries = 0;
9456
9457         /*
9458          * The control device is always connected.  The disk device, on the
9459          * other hand, may not be online all the time.  Need to change this
9460          * to figure out whether the disk device is actually online or not.
9461          */
9462         if (lun != NULL)
9463                 sn_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9464                                   lun->be_lun->lun_type;
9465         else
9466                 sn_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9467
9468         sn_ptr->page_code = SVPD_UNIT_SERIAL_NUMBER;
9469         sn_ptr->length = CTL_SN_LEN;
9470         /*
9471          * If we don't have a LUN, we just leave the serial number as
9472          * all spaces.
9473          */
9474         if (lun != NULL) {
9475                 strncpy((char *)sn_ptr->serial_num,
9476                         (char *)lun->be_lun->serial_num, CTL_SN_LEN);
9477         } else
9478                 memset(sn_ptr->serial_num, 0x20, CTL_SN_LEN);
9479
9480         ctl_set_success(ctsio);
9481         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9482         ctsio->be_move_done = ctl_config_move_done;
9483         ctl_datamove((union ctl_io *)ctsio);
9484         return (CTL_RETVAL_COMPLETE);
9485 }
9486
9487
9488 /*
9489  * SCSI VPD page 0x86, the Extended INQUIRY Data page.
9490  */
9491 static int
9492 ctl_inquiry_evpd_eid(struct ctl_scsiio *ctsio, int alloc_len)
9493 {
9494         struct scsi_vpd_extended_inquiry_data *eid_ptr;
9495         struct ctl_lun *lun;
9496         int data_len;
9497
9498         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9499
9500         data_len = sizeof(struct scsi_vpd_extended_inquiry_data);
9501         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9502         eid_ptr = (struct scsi_vpd_extended_inquiry_data *)ctsio->kern_data_ptr;
9503         ctsio->kern_sg_entries = 0;
9504
9505         if (data_len < alloc_len) {
9506                 ctsio->residual = alloc_len - data_len;
9507                 ctsio->kern_data_len = data_len;
9508                 ctsio->kern_total_len = data_len;
9509         } else {
9510                 ctsio->residual = 0;
9511                 ctsio->kern_data_len = alloc_len;
9512                 ctsio->kern_total_len = alloc_len;
9513         }
9514         ctsio->kern_data_resid = 0;
9515         ctsio->kern_rel_offset = 0;
9516         ctsio->kern_sg_entries = 0;
9517
9518         /*
9519          * The control device is always connected.  The disk device, on the
9520          * other hand, may not be online all the time.
9521          */
9522         if (lun != NULL)
9523                 eid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9524                                      lun->be_lun->lun_type;
9525         else
9526                 eid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9527         eid_ptr->page_code = SVPD_EXTENDED_INQUIRY_DATA;
9528         scsi_ulto2b(data_len - 4, eid_ptr->page_length);
9529         /*
9530          * We support head of queue, ordered and simple tags.
9531          */
9532         eid_ptr->flags2 = SVPD_EID_HEADSUP | SVPD_EID_ORDSUP | SVPD_EID_SIMPSUP;
9533         /*
9534          * Volatile cache supported.
9535          */
9536         eid_ptr->flags3 = SVPD_EID_V_SUP;
9537
9538         /*
9539          * This means that we clear the REPORTED LUNS DATA HAS CHANGED unit
9540          * attention for a particular IT nexus on all LUNs once we report
9541          * it to that nexus once.  This bit is required as of SPC-4.
9542          */
9543         eid_ptr->flags4 = SVPD_EID_LUICLT;
9544
9545         /*
9546          * XXX KDM in order to correctly answer this, we would need
9547          * information from the SIM to determine how much sense data it
9548          * can send.  So this would really be a path inquiry field, most
9549          * likely.  This can be set to a maximum of 252 according to SPC-4,
9550          * but the hardware may or may not be able to support that much.
9551          * 0 just means that the maximum sense data length is not reported.
9552          */
9553         eid_ptr->max_sense_length = 0;
9554
9555         ctl_set_success(ctsio);
9556         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9557         ctsio->be_move_done = ctl_config_move_done;
9558         ctl_datamove((union ctl_io *)ctsio);
9559         return (CTL_RETVAL_COMPLETE);
9560 }
9561
9562 static int
9563 ctl_inquiry_evpd_mpp(struct ctl_scsiio *ctsio, int alloc_len)
9564 {
9565         struct scsi_vpd_mode_page_policy *mpp_ptr;
9566         struct ctl_lun *lun;
9567         int data_len;
9568
9569         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9570
9571         data_len = sizeof(struct scsi_vpd_mode_page_policy) +
9572             sizeof(struct scsi_vpd_mode_page_policy_descr);
9573
9574         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9575         mpp_ptr = (struct scsi_vpd_mode_page_policy *)ctsio->kern_data_ptr;
9576         ctsio->kern_sg_entries = 0;
9577
9578         if (data_len < alloc_len) {
9579                 ctsio->residual = alloc_len - data_len;
9580                 ctsio->kern_data_len = data_len;
9581                 ctsio->kern_total_len = data_len;
9582         } else {
9583                 ctsio->residual = 0;
9584                 ctsio->kern_data_len = alloc_len;
9585                 ctsio->kern_total_len = alloc_len;
9586         }
9587         ctsio->kern_data_resid = 0;
9588         ctsio->kern_rel_offset = 0;
9589         ctsio->kern_sg_entries = 0;
9590
9591         /*
9592          * The control device is always connected.  The disk device, on the
9593          * other hand, may not be online all the time.
9594          */
9595         if (lun != NULL)
9596                 mpp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9597                                      lun->be_lun->lun_type;
9598         else
9599                 mpp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9600         mpp_ptr->page_code = SVPD_MODE_PAGE_POLICY;
9601         scsi_ulto2b(data_len - 4, mpp_ptr->page_length);
9602         mpp_ptr->descr[0].page_code = 0x3f;
9603         mpp_ptr->descr[0].subpage_code = 0xff;
9604         mpp_ptr->descr[0].policy = SVPD_MPP_SHARED;
9605
9606         ctl_set_success(ctsio);
9607         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9608         ctsio->be_move_done = ctl_config_move_done;
9609         ctl_datamove((union ctl_io *)ctsio);
9610         return (CTL_RETVAL_COMPLETE);
9611 }
9612
9613 /*
9614  * SCSI VPD page 0x83, the Device Identification page.
9615  */
9616 static int
9617 ctl_inquiry_evpd_devid(struct ctl_scsiio *ctsio, int alloc_len)
9618 {
9619         struct scsi_vpd_device_id *devid_ptr;
9620         struct scsi_vpd_id_descriptor *desc;
9621         struct ctl_softc *softc;
9622         struct ctl_lun *lun;
9623         struct ctl_port *port;
9624         int data_len, g;
9625         uint8_t proto;
9626
9627         softc = control_softc;
9628
9629         port = ctl_io_port(&ctsio->io_hdr);
9630         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9631
9632         data_len = sizeof(struct scsi_vpd_device_id) +
9633             sizeof(struct scsi_vpd_id_descriptor) +
9634                 sizeof(struct scsi_vpd_id_rel_trgt_port_id) +
9635             sizeof(struct scsi_vpd_id_descriptor) +
9636                 sizeof(struct scsi_vpd_id_trgt_port_grp_id);
9637         if (lun && lun->lun_devid)
9638                 data_len += lun->lun_devid->len;
9639         if (port && port->port_devid)
9640                 data_len += port->port_devid->len;
9641         if (port && port->target_devid)
9642                 data_len += port->target_devid->len;
9643
9644         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9645         devid_ptr = (struct scsi_vpd_device_id *)ctsio->kern_data_ptr;
9646         ctsio->kern_sg_entries = 0;
9647
9648         if (data_len < alloc_len) {
9649                 ctsio->residual = alloc_len - data_len;
9650                 ctsio->kern_data_len = data_len;
9651                 ctsio->kern_total_len = data_len;
9652         } else {
9653                 ctsio->residual = 0;
9654                 ctsio->kern_data_len = alloc_len;
9655                 ctsio->kern_total_len = alloc_len;
9656         }
9657         ctsio->kern_data_resid = 0;
9658         ctsio->kern_rel_offset = 0;
9659         ctsio->kern_sg_entries = 0;
9660
9661         /*
9662          * The control device is always connected.  The disk device, on the
9663          * other hand, may not be online all the time.
9664          */
9665         if (lun != NULL)
9666                 devid_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9667                                      lun->be_lun->lun_type;
9668         else
9669                 devid_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9670         devid_ptr->page_code = SVPD_DEVICE_ID;
9671         scsi_ulto2b(data_len - 4, devid_ptr->length);
9672
9673         if (port && port->port_type == CTL_PORT_FC)
9674                 proto = SCSI_PROTO_FC << 4;
9675         else if (port && port->port_type == CTL_PORT_ISCSI)
9676                 proto = SCSI_PROTO_ISCSI << 4;
9677         else
9678                 proto = SCSI_PROTO_SPI << 4;
9679         desc = (struct scsi_vpd_id_descriptor *)devid_ptr->desc_list;
9680
9681         /*
9682          * We're using a LUN association here.  i.e., this device ID is a
9683          * per-LUN identifier.
9684          */
9685         if (lun && lun->lun_devid) {
9686                 memcpy(desc, lun->lun_devid->data, lun->lun_devid->len);
9687                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9688                     lun->lun_devid->len);
9689         }
9690
9691         /*
9692          * This is for the WWPN which is a port association.
9693          */
9694         if (port && port->port_devid) {
9695                 memcpy(desc, port->port_devid->data, port->port_devid->len);
9696                 desc = (struct scsi_vpd_id_descriptor *)((uint8_t *)desc +
9697                     port->port_devid->len);
9698         }
9699
9700         /*
9701          * This is for the Relative Target Port(type 4h) identifier
9702          */
9703         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9704         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9705             SVPD_ID_TYPE_RELTARG;
9706         desc->length = 4;
9707         scsi_ulto2b(ctsio->io_hdr.nexus.targ_port, &desc->identifier[2]);
9708         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9709             sizeof(struct scsi_vpd_id_rel_trgt_port_id));
9710
9711         /*
9712          * This is for the Target Port Group(type 5h) identifier
9713          */
9714         desc->proto_codeset = proto | SVPD_ID_CODESET_BINARY;
9715         desc->id_type = SVPD_ID_PIV | SVPD_ID_ASSOC_PORT |
9716             SVPD_ID_TYPE_TPORTGRP;
9717         desc->length = 4;
9718         if (softc->is_single ||
9719             (port && port->status & CTL_PORT_STATUS_HA_SHARED))
9720                 g = 1;
9721         else
9722                 g = 2 + ctsio->io_hdr.nexus.targ_port / softc->port_cnt;
9723         scsi_ulto2b(g, &desc->identifier[2]);
9724         desc = (struct scsi_vpd_id_descriptor *)(&desc->identifier[0] +
9725             sizeof(struct scsi_vpd_id_trgt_port_grp_id));
9726
9727         /*
9728          * This is for the Target identifier
9729          */
9730         if (port && port->target_devid) {
9731                 memcpy(desc, port->target_devid->data, port->target_devid->len);
9732         }
9733
9734         ctl_set_success(ctsio);
9735         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9736         ctsio->be_move_done = ctl_config_move_done;
9737         ctl_datamove((union ctl_io *)ctsio);
9738         return (CTL_RETVAL_COMPLETE);
9739 }
9740
9741 static int
9742 ctl_inquiry_evpd_scsi_ports(struct ctl_scsiio *ctsio, int alloc_len)
9743 {
9744         struct ctl_softc *softc = control_softc;
9745         struct scsi_vpd_scsi_ports *sp;
9746         struct scsi_vpd_port_designation *pd;
9747         struct scsi_vpd_port_designation_cont *pdc;
9748         struct ctl_lun *lun;
9749         struct ctl_port *port;
9750         int data_len, num_target_ports, iid_len, id_len;
9751
9752         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9753
9754         num_target_ports = 0;
9755         iid_len = 0;
9756         id_len = 0;
9757         mtx_lock(&softc->ctl_lock);
9758         STAILQ_FOREACH(port, &softc->port_list, links) {
9759                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9760                         continue;
9761                 if (lun != NULL &&
9762                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9763                         continue;
9764                 num_target_ports++;
9765                 if (port->init_devid)
9766                         iid_len += port->init_devid->len;
9767                 if (port->port_devid)
9768                         id_len += port->port_devid->len;
9769         }
9770         mtx_unlock(&softc->ctl_lock);
9771
9772         data_len = sizeof(struct scsi_vpd_scsi_ports) +
9773             num_target_ports * (sizeof(struct scsi_vpd_port_designation) +
9774              sizeof(struct scsi_vpd_port_designation_cont)) + iid_len + id_len;
9775         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
9776         sp = (struct scsi_vpd_scsi_ports *)ctsio->kern_data_ptr;
9777         ctsio->kern_sg_entries = 0;
9778
9779         if (data_len < alloc_len) {
9780                 ctsio->residual = alloc_len - data_len;
9781                 ctsio->kern_data_len = data_len;
9782                 ctsio->kern_total_len = data_len;
9783         } else {
9784                 ctsio->residual = 0;
9785                 ctsio->kern_data_len = alloc_len;
9786                 ctsio->kern_total_len = alloc_len;
9787         }
9788         ctsio->kern_data_resid = 0;
9789         ctsio->kern_rel_offset = 0;
9790         ctsio->kern_sg_entries = 0;
9791
9792         /*
9793          * The control device is always connected.  The disk device, on the
9794          * other hand, may not be online all the time.  Need to change this
9795          * to figure out whether the disk device is actually online or not.
9796          */
9797         if (lun != NULL)
9798                 sp->device = (SID_QUAL_LU_CONNECTED << 5) |
9799                                   lun->be_lun->lun_type;
9800         else
9801                 sp->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9802
9803         sp->page_code = SVPD_SCSI_PORTS;
9804         scsi_ulto2b(data_len - sizeof(struct scsi_vpd_scsi_ports),
9805             sp->page_length);
9806         pd = &sp->design[0];
9807
9808         mtx_lock(&softc->ctl_lock);
9809         STAILQ_FOREACH(port, &softc->port_list, links) {
9810                 if ((port->status & CTL_PORT_STATUS_ONLINE) == 0)
9811                         continue;
9812                 if (lun != NULL &&
9813                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
9814                         continue;
9815                 scsi_ulto2b(port->targ_port, pd->relative_port_id);
9816                 if (port->init_devid) {
9817                         iid_len = port->init_devid->len;
9818                         memcpy(pd->initiator_transportid,
9819                             port->init_devid->data, port->init_devid->len);
9820                 } else
9821                         iid_len = 0;
9822                 scsi_ulto2b(iid_len, pd->initiator_transportid_length);
9823                 pdc = (struct scsi_vpd_port_designation_cont *)
9824                     (&pd->initiator_transportid[iid_len]);
9825                 if (port->port_devid) {
9826                         id_len = port->port_devid->len;
9827                         memcpy(pdc->target_port_descriptors,
9828                             port->port_devid->data, port->port_devid->len);
9829                 } else
9830                         id_len = 0;
9831                 scsi_ulto2b(id_len, pdc->target_port_descriptors_length);
9832                 pd = (struct scsi_vpd_port_designation *)
9833                     ((uint8_t *)pdc->target_port_descriptors + id_len);
9834         }
9835         mtx_unlock(&softc->ctl_lock);
9836
9837         ctl_set_success(ctsio);
9838         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9839         ctsio->be_move_done = ctl_config_move_done;
9840         ctl_datamove((union ctl_io *)ctsio);
9841         return (CTL_RETVAL_COMPLETE);
9842 }
9843
9844 static int
9845 ctl_inquiry_evpd_block_limits(struct ctl_scsiio *ctsio, int alloc_len)
9846 {
9847         struct scsi_vpd_block_limits *bl_ptr;
9848         struct ctl_lun *lun;
9849         uint64_t ival;
9850
9851         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9852
9853         ctsio->kern_data_ptr = malloc(sizeof(*bl_ptr), M_CTL, M_WAITOK | M_ZERO);
9854         bl_ptr = (struct scsi_vpd_block_limits *)ctsio->kern_data_ptr;
9855         ctsio->kern_sg_entries = 0;
9856
9857         if (sizeof(*bl_ptr) < alloc_len) {
9858                 ctsio->residual = alloc_len - sizeof(*bl_ptr);
9859                 ctsio->kern_data_len = sizeof(*bl_ptr);
9860                 ctsio->kern_total_len = sizeof(*bl_ptr);
9861         } else {
9862                 ctsio->residual = 0;
9863                 ctsio->kern_data_len = alloc_len;
9864                 ctsio->kern_total_len = alloc_len;
9865         }
9866         ctsio->kern_data_resid = 0;
9867         ctsio->kern_rel_offset = 0;
9868         ctsio->kern_sg_entries = 0;
9869
9870         /*
9871          * The control device is always connected.  The disk device, on the
9872          * other hand, may not be online all the time.  Need to change this
9873          * to figure out whether the disk device is actually online or not.
9874          */
9875         if (lun != NULL)
9876                 bl_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9877                                   lun->be_lun->lun_type;
9878         else
9879                 bl_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9880
9881         bl_ptr->page_code = SVPD_BLOCK_LIMITS;
9882         scsi_ulto2b(sizeof(*bl_ptr) - 4, bl_ptr->page_length);
9883         bl_ptr->max_cmp_write_len = 0xff;
9884         scsi_ulto4b(0xffffffff, bl_ptr->max_txfer_len);
9885         if (lun != NULL) {
9886                 scsi_ulto4b(lun->be_lun->opttxferlen, bl_ptr->opt_txfer_len);
9887                 if (lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
9888                         ival = 0xffffffff;
9889                         ctl_get_opt_number(&lun->be_lun->options,
9890                             "unmap_max_lba", &ival);
9891                         scsi_ulto4b(ival, bl_ptr->max_unmap_lba_cnt);
9892                         ival = 0xffffffff;
9893                         ctl_get_opt_number(&lun->be_lun->options,
9894                             "unmap_max_descr", &ival);
9895                         scsi_ulto4b(ival, bl_ptr->max_unmap_blk_cnt);
9896                         if (lun->be_lun->ublockexp != 0) {
9897                                 scsi_ulto4b((1 << lun->be_lun->ublockexp),
9898                                     bl_ptr->opt_unmap_grain);
9899                                 scsi_ulto4b(0x80000000 | lun->be_lun->ublockoff,
9900                                     bl_ptr->unmap_grain_align);
9901                         }
9902                 }
9903                 scsi_ulto4b(lun->be_lun->atomicblock,
9904                     bl_ptr->max_atomic_transfer_length);
9905                 scsi_ulto4b(0, bl_ptr->atomic_alignment);
9906                 scsi_ulto4b(0, bl_ptr->atomic_transfer_length_granularity);
9907                 scsi_ulto4b(0, bl_ptr->max_atomic_transfer_length_with_atomic_boundary);
9908                 scsi_ulto4b(0, bl_ptr->max_atomic_boundary_size);
9909                 ival = UINT64_MAX;
9910                 ctl_get_opt_number(&lun->be_lun->options, "write_same_max_lba", &ival);
9911                 scsi_u64to8b(ival, bl_ptr->max_write_same_length);
9912         }
9913
9914         ctl_set_success(ctsio);
9915         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9916         ctsio->be_move_done = ctl_config_move_done;
9917         ctl_datamove((union ctl_io *)ctsio);
9918         return (CTL_RETVAL_COMPLETE);
9919 }
9920
9921 static int
9922 ctl_inquiry_evpd_bdc(struct ctl_scsiio *ctsio, int alloc_len)
9923 {
9924         struct scsi_vpd_block_device_characteristics *bdc_ptr;
9925         struct ctl_lun *lun;
9926         const char *value;
9927         u_int i;
9928
9929         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9930
9931         ctsio->kern_data_ptr = malloc(sizeof(*bdc_ptr), M_CTL, M_WAITOK | M_ZERO);
9932         bdc_ptr = (struct scsi_vpd_block_device_characteristics *)ctsio->kern_data_ptr;
9933         ctsio->kern_sg_entries = 0;
9934
9935         if (sizeof(*bdc_ptr) < alloc_len) {
9936                 ctsio->residual = alloc_len - sizeof(*bdc_ptr);
9937                 ctsio->kern_data_len = sizeof(*bdc_ptr);
9938                 ctsio->kern_total_len = sizeof(*bdc_ptr);
9939         } else {
9940                 ctsio->residual = 0;
9941                 ctsio->kern_data_len = alloc_len;
9942                 ctsio->kern_total_len = alloc_len;
9943         }
9944         ctsio->kern_data_resid = 0;
9945         ctsio->kern_rel_offset = 0;
9946         ctsio->kern_sg_entries = 0;
9947
9948         /*
9949          * The control device is always connected.  The disk device, on the
9950          * other hand, may not be online all the time.  Need to change this
9951          * to figure out whether the disk device is actually online or not.
9952          */
9953         if (lun != NULL)
9954                 bdc_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
9955                                   lun->be_lun->lun_type;
9956         else
9957                 bdc_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
9958         bdc_ptr->page_code = SVPD_BDC;
9959         scsi_ulto2b(sizeof(*bdc_ptr) - 4, bdc_ptr->page_length);
9960         if (lun != NULL &&
9961             (value = ctl_get_opt(&lun->be_lun->options, "rpm")) != NULL)
9962                 i = strtol(value, NULL, 0);
9963         else
9964                 i = CTL_DEFAULT_ROTATION_RATE;
9965         scsi_ulto2b(i, bdc_ptr->medium_rotation_rate);
9966         if (lun != NULL &&
9967             (value = ctl_get_opt(&lun->be_lun->options, "formfactor")) != NULL)
9968                 i = strtol(value, NULL, 0);
9969         else
9970                 i = 0;
9971         bdc_ptr->wab_wac_ff = (i & 0x0f);
9972         bdc_ptr->flags = SVPD_FUAB | SVPD_VBULS;
9973
9974         ctl_set_success(ctsio);
9975         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
9976         ctsio->be_move_done = ctl_config_move_done;
9977         ctl_datamove((union ctl_io *)ctsio);
9978         return (CTL_RETVAL_COMPLETE);
9979 }
9980
9981 static int
9982 ctl_inquiry_evpd_lbp(struct ctl_scsiio *ctsio, int alloc_len)
9983 {
9984         struct scsi_vpd_logical_block_prov *lbp_ptr;
9985         struct ctl_lun *lun;
9986         const char *value;
9987
9988         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
9989
9990         ctsio->kern_data_ptr = malloc(sizeof(*lbp_ptr), M_CTL, M_WAITOK | M_ZERO);
9991         lbp_ptr = (struct scsi_vpd_logical_block_prov *)ctsio->kern_data_ptr;
9992         ctsio->kern_sg_entries = 0;
9993
9994         if (sizeof(*lbp_ptr) < alloc_len) {
9995                 ctsio->residual = alloc_len - sizeof(*lbp_ptr);
9996                 ctsio->kern_data_len = sizeof(*lbp_ptr);
9997                 ctsio->kern_total_len = sizeof(*lbp_ptr);
9998         } else {
9999                 ctsio->residual = 0;
10000                 ctsio->kern_data_len = alloc_len;
10001                 ctsio->kern_total_len = alloc_len;
10002         }
10003         ctsio->kern_data_resid = 0;
10004         ctsio->kern_rel_offset = 0;
10005         ctsio->kern_sg_entries = 0;
10006
10007         /*
10008          * The control device is always connected.  The disk device, on the
10009          * other hand, may not be online all the time.  Need to change this
10010          * to figure out whether the disk device is actually online or not.
10011          */
10012         if (lun != NULL)
10013                 lbp_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10014                                   lun->be_lun->lun_type;
10015         else
10016                 lbp_ptr->device = (SID_QUAL_LU_OFFLINE << 5) | T_DIRECT;
10017
10018         lbp_ptr->page_code = SVPD_LBP;
10019         scsi_ulto2b(sizeof(*lbp_ptr) - 4, lbp_ptr->page_length);
10020         lbp_ptr->threshold_exponent = CTL_LBP_EXPONENT;
10021         if (lun != NULL && lun->be_lun->flags & CTL_LUN_FLAG_UNMAP) {
10022                 lbp_ptr->flags = SVPD_LBP_UNMAP | SVPD_LBP_WS16 |
10023                     SVPD_LBP_WS10 | SVPD_LBP_RZ | SVPD_LBP_ANC_SUP;
10024                 value = ctl_get_opt(&lun->be_lun->options, "provisioning_type");
10025                 if (value != NULL) {
10026                         if (strcmp(value, "resource") == 0)
10027                                 lbp_ptr->prov_type = SVPD_LBP_RESOURCE;
10028                         else if (strcmp(value, "thin") == 0)
10029                                 lbp_ptr->prov_type = SVPD_LBP_THIN;
10030                 } else
10031                         lbp_ptr->prov_type = SVPD_LBP_THIN;
10032         }
10033
10034         ctl_set_success(ctsio);
10035         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10036         ctsio->be_move_done = ctl_config_move_done;
10037         ctl_datamove((union ctl_io *)ctsio);
10038         return (CTL_RETVAL_COMPLETE);
10039 }
10040
10041 /*
10042  * INQUIRY with the EVPD bit set.
10043  */
10044 static int
10045 ctl_inquiry_evpd(struct ctl_scsiio *ctsio)
10046 {
10047         struct ctl_lun *lun;
10048         struct scsi_inquiry *cdb;
10049         int alloc_len, retval;
10050
10051         lun = (struct ctl_lun *)ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10052         cdb = (struct scsi_inquiry *)ctsio->cdb;
10053         alloc_len = scsi_2btoul(cdb->length);
10054
10055         switch (cdb->page_code) {
10056         case SVPD_SUPPORTED_PAGES:
10057                 retval = ctl_inquiry_evpd_supported(ctsio, alloc_len);
10058                 break;
10059         case SVPD_UNIT_SERIAL_NUMBER:
10060                 retval = ctl_inquiry_evpd_serial(ctsio, alloc_len);
10061                 break;
10062         case SVPD_DEVICE_ID:
10063                 retval = ctl_inquiry_evpd_devid(ctsio, alloc_len);
10064                 break;
10065         case SVPD_EXTENDED_INQUIRY_DATA:
10066                 retval = ctl_inquiry_evpd_eid(ctsio, alloc_len);
10067                 break;
10068         case SVPD_MODE_PAGE_POLICY:
10069                 retval = ctl_inquiry_evpd_mpp(ctsio, alloc_len);
10070                 break;
10071         case SVPD_SCSI_PORTS:
10072                 retval = ctl_inquiry_evpd_scsi_ports(ctsio, alloc_len);
10073                 break;
10074         case SVPD_SCSI_TPC:
10075                 retval = ctl_inquiry_evpd_tpc(ctsio, alloc_len);
10076                 break;
10077         case SVPD_BLOCK_LIMITS:
10078                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10079                         goto err;
10080                 retval = ctl_inquiry_evpd_block_limits(ctsio, alloc_len);
10081                 break;
10082         case SVPD_BDC:
10083                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10084                         goto err;
10085                 retval = ctl_inquiry_evpd_bdc(ctsio, alloc_len);
10086                 break;
10087         case SVPD_LBP:
10088                 if (lun == NULL || lun->be_lun->lun_type != T_DIRECT)
10089                         goto err;
10090                 retval = ctl_inquiry_evpd_lbp(ctsio, alloc_len);
10091                 break;
10092         default:
10093 err:
10094                 ctl_set_invalid_field(ctsio,
10095                                       /*sks_valid*/ 1,
10096                                       /*command*/ 1,
10097                                       /*field*/ 2,
10098                                       /*bit_valid*/ 0,
10099                                       /*bit*/ 0);
10100                 ctl_done((union ctl_io *)ctsio);
10101                 retval = CTL_RETVAL_COMPLETE;
10102                 break;
10103         }
10104
10105         return (retval);
10106 }
10107
10108 /*
10109  * Standard INQUIRY data.
10110  */
10111 static int
10112 ctl_inquiry_std(struct ctl_scsiio *ctsio)
10113 {
10114         struct scsi_inquiry_data *inq_ptr;
10115         struct scsi_inquiry *cdb;
10116         struct ctl_softc *softc = control_softc;
10117         struct ctl_port *port;
10118         struct ctl_lun *lun;
10119         char *val;
10120         uint32_t alloc_len, data_len;
10121         ctl_port_type port_type;
10122
10123         port = ctl_io_port(&ctsio->io_hdr);
10124         port_type = port->port_type;
10125         if (port_type == CTL_PORT_IOCTL || port_type == CTL_PORT_INTERNAL)
10126                 port_type = CTL_PORT_SCSI;
10127
10128         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10129         cdb = (struct scsi_inquiry *)ctsio->cdb;
10130         alloc_len = scsi_2btoul(cdb->length);
10131
10132         /*
10133          * We malloc the full inquiry data size here and fill it
10134          * in.  If the user only asks for less, we'll give him
10135          * that much.
10136          */
10137         data_len = offsetof(struct scsi_inquiry_data, vendor_specific1);
10138         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10139         inq_ptr = (struct scsi_inquiry_data *)ctsio->kern_data_ptr;
10140         ctsio->kern_sg_entries = 0;
10141         ctsio->kern_data_resid = 0;
10142         ctsio->kern_rel_offset = 0;
10143
10144         if (data_len < alloc_len) {
10145                 ctsio->residual = alloc_len - data_len;
10146                 ctsio->kern_data_len = data_len;
10147                 ctsio->kern_total_len = data_len;
10148         } else {
10149                 ctsio->residual = 0;
10150                 ctsio->kern_data_len = alloc_len;
10151                 ctsio->kern_total_len = alloc_len;
10152         }
10153
10154         if (lun != NULL) {
10155                 if ((lun->flags & CTL_LUN_PRIMARY_SC) ||
10156                     softc->ha_link >= CTL_HA_LINK_UNKNOWN) {
10157                         inq_ptr->device = (SID_QUAL_LU_CONNECTED << 5) |
10158                             lun->be_lun->lun_type;
10159                 } else {
10160                         inq_ptr->device = (SID_QUAL_LU_OFFLINE << 5) |
10161                             lun->be_lun->lun_type;
10162                 }
10163                 if (lun->flags & CTL_LUN_REMOVABLE)
10164                         inq_ptr->dev_qual2 |= SID_RMB;
10165         } else
10166                 inq_ptr->device = (SID_QUAL_BAD_LU << 5) | T_NODEVICE;
10167
10168         /* RMB in byte 2 is 0 */
10169         inq_ptr->version = SCSI_REV_SPC4;
10170
10171         /*
10172          * According to SAM-3, even if a device only supports a single
10173          * level of LUN addressing, it should still set the HISUP bit:
10174          *
10175          * 4.9.1 Logical unit numbers overview
10176          *
10177          * All logical unit number formats described in this standard are
10178          * hierarchical in structure even when only a single level in that
10179          * hierarchy is used. The HISUP bit shall be set to one in the
10180          * standard INQUIRY data (see SPC-2) when any logical unit number
10181          * format described in this standard is used.  Non-hierarchical
10182          * formats are outside the scope of this standard.
10183          *
10184          * Therefore we set the HiSup bit here.
10185          *
10186          * The response format is 2, per SPC-3.
10187          */
10188         inq_ptr->response_format = SID_HiSup | 2;
10189
10190         inq_ptr->additional_length = data_len -
10191             (offsetof(struct scsi_inquiry_data, additional_length) + 1);
10192         CTL_DEBUG_PRINT(("additional_length = %d\n",
10193                          inq_ptr->additional_length));
10194
10195         inq_ptr->spc3_flags = SPC3_SID_3PC | SPC3_SID_TPGS_IMPLICIT;
10196         if (port_type == CTL_PORT_SCSI)
10197                 inq_ptr->spc2_flags = SPC2_SID_ADDR16;
10198         inq_ptr->spc2_flags |= SPC2_SID_MultiP;
10199         inq_ptr->flags = SID_CmdQue;
10200         if (port_type == CTL_PORT_SCSI)
10201                 inq_ptr->flags |= SID_WBus16 | SID_Sync;
10202
10203         /*
10204          * Per SPC-3, unused bytes in ASCII strings are filled with spaces.
10205          * We have 8 bytes for the vendor name, and 16 bytes for the device
10206          * name and 4 bytes for the revision.
10207          */
10208         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10209             "vendor")) == NULL) {
10210                 strncpy(inq_ptr->vendor, CTL_VENDOR, sizeof(inq_ptr->vendor));
10211         } else {
10212                 memset(inq_ptr->vendor, ' ', sizeof(inq_ptr->vendor));
10213                 strncpy(inq_ptr->vendor, val,
10214                     min(sizeof(inq_ptr->vendor), strlen(val)));
10215         }
10216         if (lun == NULL) {
10217                 strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10218                     sizeof(inq_ptr->product));
10219         } else if ((val = ctl_get_opt(&lun->be_lun->options, "product")) == NULL) {
10220                 switch (lun->be_lun->lun_type) {
10221                 case T_DIRECT:
10222                         strncpy(inq_ptr->product, CTL_DIRECT_PRODUCT,
10223                             sizeof(inq_ptr->product));
10224                         break;
10225                 case T_PROCESSOR:
10226                         strncpy(inq_ptr->product, CTL_PROCESSOR_PRODUCT,
10227                             sizeof(inq_ptr->product));
10228                         break;
10229                 case T_CDROM:
10230                         strncpy(inq_ptr->product, CTL_CDROM_PRODUCT,
10231                             sizeof(inq_ptr->product));
10232                         break;
10233                 default:
10234                         strncpy(inq_ptr->product, CTL_UNKNOWN_PRODUCT,
10235                             sizeof(inq_ptr->product));
10236                         break;
10237                 }
10238         } else {
10239                 memset(inq_ptr->product, ' ', sizeof(inq_ptr->product));
10240                 strncpy(inq_ptr->product, val,
10241                     min(sizeof(inq_ptr->product), strlen(val)));
10242         }
10243
10244         /*
10245          * XXX make this a macro somewhere so it automatically gets
10246          * incremented when we make changes.
10247          */
10248         if (lun == NULL || (val = ctl_get_opt(&lun->be_lun->options,
10249             "revision")) == NULL) {
10250                 strncpy(inq_ptr->revision, "0001", sizeof(inq_ptr->revision));
10251         } else {
10252                 memset(inq_ptr->revision, ' ', sizeof(inq_ptr->revision));
10253                 strncpy(inq_ptr->revision, val,
10254                     min(sizeof(inq_ptr->revision), strlen(val)));
10255         }
10256
10257         /*
10258          * For parallel SCSI, we support double transition and single
10259          * transition clocking.  We also support QAS (Quick Arbitration
10260          * and Selection) and Information Unit transfers on both the
10261          * control and array devices.
10262          */
10263         if (port_type == CTL_PORT_SCSI)
10264                 inq_ptr->spi3data = SID_SPI_CLOCK_DT_ST | SID_SPI_QAS |
10265                                     SID_SPI_IUS;
10266
10267         /* SAM-5 (no version claimed) */
10268         scsi_ulto2b(0x00A0, inq_ptr->version1);
10269         /* SPC-4 (no version claimed) */
10270         scsi_ulto2b(0x0460, inq_ptr->version2);
10271         if (port_type == CTL_PORT_FC) {
10272                 /* FCP-2 ANSI INCITS.350:2003 */
10273                 scsi_ulto2b(0x0917, inq_ptr->version3);
10274         } else if (port_type == CTL_PORT_SCSI) {
10275                 /* SPI-4 ANSI INCITS.362:200x */
10276                 scsi_ulto2b(0x0B56, inq_ptr->version3);
10277         } else if (port_type == CTL_PORT_ISCSI) {
10278                 /* iSCSI (no version claimed) */
10279                 scsi_ulto2b(0x0960, inq_ptr->version3);
10280         } else if (port_type == CTL_PORT_SAS) {
10281                 /* SAS (no version claimed) */
10282                 scsi_ulto2b(0x0BE0, inq_ptr->version3);
10283         }
10284
10285         if (lun == NULL) {
10286                 /* SBC-4 (no version claimed) */
10287                 scsi_ulto2b(0x0600, inq_ptr->version4);
10288         } else {
10289                 switch (lun->be_lun->lun_type) {
10290                 case T_DIRECT:
10291                         /* SBC-4 (no version claimed) */
10292                         scsi_ulto2b(0x0600, inq_ptr->version4);
10293                         break;
10294                 case T_PROCESSOR:
10295                         break;
10296                 case T_CDROM:
10297                         /* MMC-6 (no version claimed) */
10298                         scsi_ulto2b(0x04E0, inq_ptr->version4);
10299                         break;
10300                 default:
10301                         break;
10302                 }
10303         }
10304
10305         ctl_set_success(ctsio);
10306         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10307         ctsio->be_move_done = ctl_config_move_done;
10308         ctl_datamove((union ctl_io *)ctsio);
10309         return (CTL_RETVAL_COMPLETE);
10310 }
10311
10312 int
10313 ctl_inquiry(struct ctl_scsiio *ctsio)
10314 {
10315         struct scsi_inquiry *cdb;
10316         int retval;
10317
10318         CTL_DEBUG_PRINT(("ctl_inquiry\n"));
10319
10320         cdb = (struct scsi_inquiry *)ctsio->cdb;
10321         if (cdb->byte2 & SI_EVPD)
10322                 retval = ctl_inquiry_evpd(ctsio);
10323         else if (cdb->page_code == 0)
10324                 retval = ctl_inquiry_std(ctsio);
10325         else {
10326                 ctl_set_invalid_field(ctsio,
10327                                       /*sks_valid*/ 1,
10328                                       /*command*/ 1,
10329                                       /*field*/ 2,
10330                                       /*bit_valid*/ 0,
10331                                       /*bit*/ 0);
10332                 ctl_done((union ctl_io *)ctsio);
10333                 return (CTL_RETVAL_COMPLETE);
10334         }
10335
10336         return (retval);
10337 }
10338
10339 int
10340 ctl_get_config(struct ctl_scsiio *ctsio)
10341 {
10342         struct scsi_get_config_header *hdr;
10343         struct scsi_get_config_feature *feature;
10344         struct scsi_get_config *cdb;
10345         struct ctl_lun *lun;
10346         uint32_t alloc_len, data_len;
10347         int rt, starting;
10348
10349         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10350         cdb = (struct scsi_get_config *)ctsio->cdb;
10351         rt = (cdb->rt & SGC_RT_MASK);
10352         starting = scsi_2btoul(cdb->starting_feature);
10353         alloc_len = scsi_2btoul(cdb->length);
10354
10355         data_len = sizeof(struct scsi_get_config_header) +
10356             sizeof(struct scsi_get_config_feature) + 8 +
10357             sizeof(struct scsi_get_config_feature) + 8 +
10358             sizeof(struct scsi_get_config_feature) + 4 +
10359             sizeof(struct scsi_get_config_feature) + 4 +
10360             sizeof(struct scsi_get_config_feature) + 8 +
10361             sizeof(struct scsi_get_config_feature) +
10362             sizeof(struct scsi_get_config_feature) + 4 +
10363             sizeof(struct scsi_get_config_feature) + 4 +
10364             sizeof(struct scsi_get_config_feature) + 4 +
10365             sizeof(struct scsi_get_config_feature) + 4 +
10366             sizeof(struct scsi_get_config_feature) + 4 +
10367             sizeof(struct scsi_get_config_feature) + 4;
10368         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10369         ctsio->kern_sg_entries = 0;
10370         ctsio->kern_data_resid = 0;
10371         ctsio->kern_rel_offset = 0;
10372
10373         hdr = (struct scsi_get_config_header *)ctsio->kern_data_ptr;
10374         if (lun->flags & CTL_LUN_NO_MEDIA)
10375                 scsi_ulto2b(0x0000, hdr->current_profile);
10376         else
10377                 scsi_ulto2b(0x0010, hdr->current_profile);
10378         feature = (struct scsi_get_config_feature *)(hdr + 1);
10379
10380         if (starting > 0x003b)
10381                 goto done;
10382         if (starting > 0x003a)
10383                 goto f3b;
10384         if (starting > 0x002b)
10385                 goto f3a;
10386         if (starting > 0x002a)
10387                 goto f2b;
10388         if (starting > 0x001f)
10389                 goto f2a;
10390         if (starting > 0x001e)
10391                 goto f1f;
10392         if (starting > 0x001d)
10393                 goto f1e;
10394         if (starting > 0x0010)
10395                 goto f1d;
10396         if (starting > 0x0003)
10397                 goto f10;
10398         if (starting > 0x0002)
10399                 goto f3;
10400         if (starting > 0x0001)
10401                 goto f2;
10402         if (starting > 0x0000)
10403                 goto f1;
10404
10405         /* Profile List */
10406         scsi_ulto2b(0x0000, feature->feature_code);
10407         feature->flags = SGC_F_PERSISTENT | SGC_F_CURRENT;
10408         feature->add_length = 8;
10409         scsi_ulto2b(0x0008, &feature->feature_data[0]); /* CD-ROM */
10410         feature->feature_data[2] = 0x00;
10411         scsi_ulto2b(0x0010, &feature->feature_data[4]); /* DVD-ROM */
10412         feature->feature_data[6] = 0x01;
10413         feature = (struct scsi_get_config_feature *)
10414             &feature->feature_data[feature->add_length];
10415
10416 f1:     /* Core */
10417         scsi_ulto2b(0x0001, feature->feature_code);
10418         feature->flags = 0x08 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10419         feature->add_length = 8;
10420         scsi_ulto4b(0x00000000, &feature->feature_data[0]);
10421         feature->feature_data[4] = 0x03;
10422         feature = (struct scsi_get_config_feature *)
10423             &feature->feature_data[feature->add_length];
10424
10425 f2:     /* Morphing */
10426         scsi_ulto2b(0x0002, feature->feature_code);
10427         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10428         feature->add_length = 4;
10429         feature->feature_data[0] = 0x02;
10430         feature = (struct scsi_get_config_feature *)
10431             &feature->feature_data[feature->add_length];
10432
10433 f3:     /* Removable Medium */
10434         scsi_ulto2b(0x0003, feature->feature_code);
10435         feature->flags = 0x04 | SGC_F_PERSISTENT | SGC_F_CURRENT;
10436         feature->add_length = 4;
10437         feature->feature_data[0] = 0x39;
10438         feature = (struct scsi_get_config_feature *)
10439             &feature->feature_data[feature->add_length];
10440
10441         if (rt == SGC_RT_CURRENT && (lun->flags & CTL_LUN_NO_MEDIA))
10442                 goto done;
10443
10444 f10:    /* Random Read */
10445         scsi_ulto2b(0x0010, feature->feature_code);
10446         feature->flags = 0x00;
10447         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10448                 feature->flags |= SGC_F_CURRENT;
10449         feature->add_length = 8;
10450         scsi_ulto4b(lun->be_lun->blocksize, &feature->feature_data[0]);
10451         scsi_ulto2b(1, &feature->feature_data[4]);
10452         feature->feature_data[6] = 0x00;
10453         feature = (struct scsi_get_config_feature *)
10454             &feature->feature_data[feature->add_length];
10455
10456 f1d:    /* Multi-Read */
10457         scsi_ulto2b(0x001D, feature->feature_code);
10458         feature->flags = 0x00;
10459         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10460                 feature->flags |= SGC_F_CURRENT;
10461         feature->add_length = 0;
10462         feature = (struct scsi_get_config_feature *)
10463             &feature->feature_data[feature->add_length];
10464
10465 f1e:    /* CD Read */
10466         scsi_ulto2b(0x001E, feature->feature_code);
10467         feature->flags = 0x00;
10468         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10469                 feature->flags |= SGC_F_CURRENT;
10470         feature->add_length = 4;
10471         feature->feature_data[0] = 0x00;
10472         feature = (struct scsi_get_config_feature *)
10473             &feature->feature_data[feature->add_length];
10474
10475 f1f:    /* DVD Read */
10476         scsi_ulto2b(0x001F, feature->feature_code);
10477         feature->flags = 0x08;
10478         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10479                 feature->flags |= SGC_F_CURRENT;
10480         feature->add_length = 4;
10481         feature->feature_data[0] = 0x01;
10482         feature->feature_data[2] = 0x03;
10483         feature = (struct scsi_get_config_feature *)
10484             &feature->feature_data[feature->add_length];
10485
10486 f2a:    /* DVD+RW */
10487         scsi_ulto2b(0x002A, feature->feature_code);
10488         feature->flags = 0x04;
10489         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10490                 feature->flags |= SGC_F_CURRENT;
10491         feature->add_length = 4;
10492         feature->feature_data[0] = 0x00;
10493         feature->feature_data[1] = 0x00;
10494         feature = (struct scsi_get_config_feature *)
10495             &feature->feature_data[feature->add_length];
10496
10497 f2b:    /* DVD+R */
10498         scsi_ulto2b(0x002B, feature->feature_code);
10499         feature->flags = 0x00;
10500         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10501                 feature->flags |= SGC_F_CURRENT;
10502         feature->add_length = 4;
10503         feature->feature_data[0] = 0x00;
10504         feature = (struct scsi_get_config_feature *)
10505             &feature->feature_data[feature->add_length];
10506
10507 f3a:    /* DVD+RW Dual Layer */
10508         scsi_ulto2b(0x003A, feature->feature_code);
10509         feature->flags = 0x00;
10510         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10511                 feature->flags |= SGC_F_CURRENT;
10512         feature->add_length = 4;
10513         feature->feature_data[0] = 0x00;
10514         feature->feature_data[1] = 0x00;
10515         feature = (struct scsi_get_config_feature *)
10516             &feature->feature_data[feature->add_length];
10517
10518 f3b:    /* DVD+R Dual Layer */
10519         scsi_ulto2b(0x003B, feature->feature_code);
10520         feature->flags = 0x00;
10521         if ((lun->flags & CTL_LUN_NO_MEDIA) == 0)
10522                 feature->flags |= SGC_F_CURRENT;
10523         feature->add_length = 4;
10524         feature->feature_data[0] = 0x00;
10525         feature = (struct scsi_get_config_feature *)
10526             &feature->feature_data[feature->add_length];
10527
10528 done:
10529         data_len = (uint8_t *)feature - (uint8_t *)hdr;
10530         if (rt == SGC_RT_SPECIFIC && data_len > 4) {
10531                 feature = (struct scsi_get_config_feature *)(hdr + 1);
10532                 if (scsi_2btoul(feature->feature_code) == starting)
10533                         feature = (struct scsi_get_config_feature *)
10534                             &feature->feature_data[feature->add_length];
10535                 data_len = (uint8_t *)feature - (uint8_t *)hdr;
10536         }
10537         scsi_ulto4b(data_len - 4, hdr->data_length);
10538         if (data_len < alloc_len) {
10539                 ctsio->residual = alloc_len - data_len;
10540                 ctsio->kern_data_len = data_len;
10541                 ctsio->kern_total_len = data_len;
10542         } else {
10543                 ctsio->residual = 0;
10544                 ctsio->kern_data_len = alloc_len;
10545                 ctsio->kern_total_len = alloc_len;
10546         }
10547
10548         ctl_set_success(ctsio);
10549         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10550         ctsio->be_move_done = ctl_config_move_done;
10551         ctl_datamove((union ctl_io *)ctsio);
10552         return (CTL_RETVAL_COMPLETE);
10553 }
10554
10555 int
10556 ctl_get_event_status(struct ctl_scsiio *ctsio)
10557 {
10558         struct scsi_get_event_status_header *hdr;
10559         struct scsi_get_event_status *cdb;
10560         struct ctl_lun *lun;
10561         uint32_t alloc_len, data_len;
10562         int notif_class;
10563
10564         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10565         cdb = (struct scsi_get_event_status *)ctsio->cdb;
10566         if ((cdb->byte2 & SGESN_POLLED) == 0) {
10567                 ctl_set_invalid_field(ctsio, /*sks_valid*/ 1, /*command*/ 1,
10568                     /*field*/ 1, /*bit_valid*/ 1, /*bit*/ 0);
10569                 ctl_done((union ctl_io *)ctsio);
10570                 return (CTL_RETVAL_COMPLETE);
10571         }
10572         notif_class = cdb->notif_class;
10573         alloc_len = scsi_2btoul(cdb->length);
10574
10575         data_len = sizeof(struct scsi_get_event_status_header);
10576         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10577         ctsio->kern_sg_entries = 0;
10578         ctsio->kern_data_resid = 0;
10579         ctsio->kern_rel_offset = 0;
10580
10581         if (data_len < alloc_len) {
10582                 ctsio->residual = alloc_len - data_len;
10583                 ctsio->kern_data_len = data_len;
10584                 ctsio->kern_total_len = data_len;
10585         } else {
10586                 ctsio->residual = 0;
10587                 ctsio->kern_data_len = alloc_len;
10588                 ctsio->kern_total_len = alloc_len;
10589         }
10590
10591         hdr = (struct scsi_get_event_status_header *)ctsio->kern_data_ptr;
10592         scsi_ulto2b(0, hdr->descr_length);
10593         hdr->nea_class = SGESN_NEA;
10594         hdr->supported_class = 0;
10595
10596         ctl_set_success(ctsio);
10597         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10598         ctsio->be_move_done = ctl_config_move_done;
10599         ctl_datamove((union ctl_io *)ctsio);
10600         return (CTL_RETVAL_COMPLETE);
10601 }
10602
10603 int
10604 ctl_mechanism_status(struct ctl_scsiio *ctsio)
10605 {
10606         struct scsi_mechanism_status_header *hdr;
10607         struct scsi_mechanism_status *cdb;
10608         struct ctl_lun *lun;
10609         uint32_t alloc_len, data_len;
10610
10611         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10612         cdb = (struct scsi_mechanism_status *)ctsio->cdb;
10613         alloc_len = scsi_2btoul(cdb->length);
10614
10615         data_len = sizeof(struct scsi_mechanism_status_header);
10616         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10617         ctsio->kern_sg_entries = 0;
10618         ctsio->kern_data_resid = 0;
10619         ctsio->kern_rel_offset = 0;
10620
10621         if (data_len < alloc_len) {
10622                 ctsio->residual = alloc_len - data_len;
10623                 ctsio->kern_data_len = data_len;
10624                 ctsio->kern_total_len = data_len;
10625         } else {
10626                 ctsio->residual = 0;
10627                 ctsio->kern_data_len = alloc_len;
10628                 ctsio->kern_total_len = alloc_len;
10629         }
10630
10631         hdr = (struct scsi_mechanism_status_header *)ctsio->kern_data_ptr;
10632         hdr->state1 = 0x00;
10633         hdr->state2 = 0xe0;
10634         scsi_ulto3b(0, hdr->lba);
10635         hdr->slots_num = 0;
10636         scsi_ulto2b(0, hdr->slots_length);
10637
10638         ctl_set_success(ctsio);
10639         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10640         ctsio->be_move_done = ctl_config_move_done;
10641         ctl_datamove((union ctl_io *)ctsio);
10642         return (CTL_RETVAL_COMPLETE);
10643 }
10644
10645 static void
10646 ctl_ultomsf(uint32_t lba, uint8_t *buf)
10647 {
10648
10649         lba += 150;
10650         buf[0] = 0;
10651         buf[1] = bin2bcd((lba / 75) / 60);
10652         buf[2] = bin2bcd((lba / 75) % 60);
10653         buf[3] = bin2bcd(lba % 75);
10654 }
10655
10656 int
10657 ctl_read_toc(struct ctl_scsiio *ctsio)
10658 {
10659         struct scsi_read_toc_hdr *hdr;
10660         struct scsi_read_toc_type01_descr *descr;
10661         struct scsi_read_toc *cdb;
10662         struct ctl_lun *lun;
10663         uint32_t alloc_len, data_len;
10664         int format, msf;
10665
10666         lun = ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
10667         cdb = (struct scsi_read_toc *)ctsio->cdb;
10668         msf = (cdb->byte2 & CD_MSF) != 0;
10669         format = cdb->format;
10670         alloc_len = scsi_2btoul(cdb->data_len);
10671
10672         data_len = sizeof(struct scsi_read_toc_hdr);
10673         if (format == 0)
10674                 data_len += 2 * sizeof(struct scsi_read_toc_type01_descr);
10675         else
10676                 data_len += sizeof(struct scsi_read_toc_type01_descr);
10677         ctsio->kern_data_ptr = malloc(data_len, M_CTL, M_WAITOK | M_ZERO);
10678         ctsio->kern_sg_entries = 0;
10679         ctsio->kern_data_resid = 0;
10680         ctsio->kern_rel_offset = 0;
10681
10682         if (data_len < alloc_len) {
10683                 ctsio->residual = alloc_len - data_len;
10684                 ctsio->kern_data_len = data_len;
10685                 ctsio->kern_total_len = data_len;
10686         } else {
10687                 ctsio->residual = 0;
10688                 ctsio->kern_data_len = alloc_len;
10689                 ctsio->kern_total_len = alloc_len;
10690         }
10691
10692         hdr = (struct scsi_read_toc_hdr *)ctsio->kern_data_ptr;
10693         if (format == 0) {
10694                 scsi_ulto2b(0x12, hdr->data_length);
10695                 hdr->first = 1;
10696                 hdr->last = 1;
10697                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10698                 descr->addr_ctl = 0x14;
10699                 descr->track_number = 1;
10700                 if (msf)
10701                         ctl_ultomsf(0, descr->track_start);
10702                 else
10703                         scsi_ulto4b(0, descr->track_start);
10704                 descr++;
10705                 descr->addr_ctl = 0x14;
10706                 descr->track_number = 0xaa;
10707                 if (msf)
10708                         ctl_ultomsf(lun->be_lun->maxlba+1, descr->track_start);
10709                 else
10710                         scsi_ulto4b(lun->be_lun->maxlba+1, descr->track_start);
10711         } else {
10712                 scsi_ulto2b(0x0a, hdr->data_length);
10713                 hdr->first = 1;
10714                 hdr->last = 1;
10715                 descr = (struct scsi_read_toc_type01_descr *)(hdr + 1);
10716                 descr->addr_ctl = 0x14;
10717                 descr->track_number = 1;
10718                 if (msf)
10719                         ctl_ultomsf(0, descr->track_start);
10720                 else
10721                         scsi_ulto4b(0, descr->track_start);
10722         }
10723
10724         ctl_set_success(ctsio);
10725         ctsio->io_hdr.flags |= CTL_FLAG_ALLOCATED;
10726         ctsio->be_move_done = ctl_config_move_done;
10727         ctl_datamove((union ctl_io *)ctsio);
10728         return (CTL_RETVAL_COMPLETE);
10729 }
10730
10731 /*
10732  * For known CDB types, parse the LBA and length.
10733  */
10734 static int
10735 ctl_get_lba_len(union ctl_io *io, uint64_t *lba, uint64_t *len)
10736 {
10737         if (io->io_hdr.io_type != CTL_IO_SCSI)
10738                 return (1);
10739
10740         switch (io->scsiio.cdb[0]) {
10741         case COMPARE_AND_WRITE: {
10742                 struct scsi_compare_and_write *cdb;
10743
10744                 cdb = (struct scsi_compare_and_write *)io->scsiio.cdb;
10745
10746                 *lba = scsi_8btou64(cdb->addr);
10747                 *len = cdb->length;
10748                 break;
10749         }
10750         case READ_6:
10751         case WRITE_6: {
10752                 struct scsi_rw_6 *cdb;
10753
10754                 cdb = (struct scsi_rw_6 *)io->scsiio.cdb;
10755
10756                 *lba = scsi_3btoul(cdb->addr);
10757                 /* only 5 bits are valid in the most significant address byte */
10758                 *lba &= 0x1fffff;
10759                 *len = cdb->length;
10760                 break;
10761         }
10762         case READ_10:
10763         case WRITE_10: {
10764                 struct scsi_rw_10 *cdb;
10765
10766                 cdb = (struct scsi_rw_10 *)io->scsiio.cdb;
10767
10768                 *lba = scsi_4btoul(cdb->addr);
10769                 *len = scsi_2btoul(cdb->length);
10770                 break;
10771         }
10772         case WRITE_VERIFY_10: {
10773                 struct scsi_write_verify_10 *cdb;
10774
10775                 cdb = (struct scsi_write_verify_10 *)io->scsiio.cdb;
10776
10777                 *lba = scsi_4btoul(cdb->addr);
10778                 *len = scsi_2btoul(cdb->length);
10779                 break;
10780         }
10781         case READ_12:
10782         case WRITE_12: {
10783                 struct scsi_rw_12 *cdb;
10784
10785                 cdb = (struct scsi_rw_12 *)io->scsiio.cdb;
10786
10787                 *lba = scsi_4btoul(cdb->addr);
10788                 *len = scsi_4btoul(cdb->length);
10789                 break;
10790         }
10791         case WRITE_VERIFY_12: {
10792                 struct scsi_write_verify_12 *cdb;
10793
10794                 cdb = (struct scsi_write_verify_12 *)io->scsiio.cdb;
10795
10796                 *lba = scsi_4btoul(cdb->addr);
10797                 *len = scsi_4btoul(cdb->length);
10798                 break;
10799         }
10800         case READ_16:
10801         case WRITE_16: {
10802                 struct scsi_rw_16 *cdb;
10803
10804                 cdb = (struct scsi_rw_16 *)io->scsiio.cdb;
10805
10806                 *lba = scsi_8btou64(cdb->addr);
10807                 *len = scsi_4btoul(cdb->length);
10808                 break;
10809         }
10810         case WRITE_ATOMIC_16: {
10811                 struct scsi_write_atomic_16 *cdb;
10812
10813                 cdb = (struct scsi_write_atomic_16 *)io->scsiio.cdb;
10814
10815                 *lba = scsi_8btou64(cdb->addr);
10816                 *len = scsi_2btoul(cdb->length);
10817                 break;
10818         }
10819         case WRITE_VERIFY_16: {
10820                 struct scsi_write_verify_16 *cdb;
10821
10822                 cdb = (struct scsi_write_verify_16 *)io->scsiio.cdb;
10823
10824                 *lba = scsi_8btou64(cdb->addr);
10825                 *len = scsi_4btoul(cdb->length);
10826                 break;
10827         }
10828         case WRITE_SAME_10: {
10829                 struct scsi_write_same_10 *cdb;
10830
10831                 cdb = (struct scsi_write_same_10 *)io->scsiio.cdb;
10832
10833                 *lba = scsi_4btoul(cdb->addr);
10834                 *len = scsi_2btoul(cdb->length);
10835                 break;
10836         }
10837         case WRITE_SAME_16: {
10838                 struct scsi_write_same_16 *cdb;
10839
10840                 cdb = (struct scsi_write_same_16 *)io->scsiio.cdb;
10841
10842                 *lba = scsi_8btou64(cdb->addr);
10843                 *len = scsi_4btoul(cdb->length);
10844                 break;
10845         }
10846         case VERIFY_10: {
10847                 struct scsi_verify_10 *cdb;
10848
10849                 cdb = (struct scsi_verify_10 *)io->scsiio.cdb;
10850
10851                 *lba = scsi_4btoul(cdb->addr);
10852                 *len = scsi_2btoul(cdb->length);
10853                 break;
10854         }
10855         case VERIFY_12: {
10856                 struct scsi_verify_12 *cdb;
10857
10858                 cdb = (struct scsi_verify_12 *)io->scsiio.cdb;
10859
10860                 *lba = scsi_4btoul(cdb->addr);
10861                 *len = scsi_4btoul(cdb->length);
10862                 break;
10863         }
10864         case VERIFY_16: {
10865                 struct scsi_verify_16 *cdb;
10866
10867                 cdb = (struct scsi_verify_16 *)io->scsiio.cdb;
10868
10869                 *lba = scsi_8btou64(cdb->addr);
10870                 *len = scsi_4btoul(cdb->length);
10871                 break;
10872         }
10873         case UNMAP: {
10874                 *lba = 0;
10875                 *len = UINT64_MAX;
10876                 break;
10877         }
10878         case SERVICE_ACTION_IN: {       /* GET LBA STATUS */
10879                 struct scsi_get_lba_status *cdb;
10880
10881                 cdb = (struct scsi_get_lba_status *)io->scsiio.cdb;
10882                 *lba = scsi_8btou64(cdb->addr);
10883                 *len = UINT32_MAX;
10884                 break;
10885         }
10886         default:
10887                 return (1);
10888                 break; /* NOTREACHED */
10889         }
10890
10891         return (0);
10892 }
10893
10894 static ctl_action
10895 ctl_extent_check_lba(uint64_t lba1, uint64_t len1, uint64_t lba2, uint64_t len2,
10896     bool seq)
10897 {
10898         uint64_t endlba1, endlba2;
10899
10900         endlba1 = lba1 + len1 - (seq ? 0 : 1);
10901         endlba2 = lba2 + len2 - 1;
10902
10903         if ((endlba1 < lba2) || (endlba2 < lba1))
10904                 return (CTL_ACTION_PASS);
10905         else
10906                 return (CTL_ACTION_BLOCK);
10907 }
10908
10909 static int
10910 ctl_extent_check_unmap(union ctl_io *io, uint64_t lba2, uint64_t len2)
10911 {
10912         struct ctl_ptr_len_flags *ptrlen;
10913         struct scsi_unmap_desc *buf, *end, *range;
10914         uint64_t lba;
10915         uint32_t len;
10916
10917         /* If not UNMAP -- go other way. */
10918         if (io->io_hdr.io_type != CTL_IO_SCSI ||
10919             io->scsiio.cdb[0] != UNMAP)
10920                 return (CTL_ACTION_ERROR);
10921
10922         /* If UNMAP without data -- block and wait for data. */
10923         ptrlen = (struct ctl_ptr_len_flags *)
10924             &io->io_hdr.ctl_private[CTL_PRIV_LBA_LEN];
10925         if ((io->io_hdr.flags & CTL_FLAG_ALLOCATED) == 0 ||
10926             ptrlen->ptr == NULL)
10927                 return (CTL_ACTION_BLOCK);
10928
10929         /* UNMAP with data -- check for collision. */
10930         buf = (struct scsi_unmap_desc *)ptrlen->ptr;
10931         end = buf + ptrlen->len / sizeof(*buf);
10932         for (range = buf; range < end; range++) {
10933                 lba = scsi_8btou64(range->lba);
10934                 len = scsi_4btoul(range->length);
10935                 if ((lba < lba2 + len2) && (lba + len > lba2))
10936                         return (CTL_ACTION_BLOCK);
10937         }
10938         return (CTL_ACTION_PASS);
10939 }
10940
10941 static ctl_action
10942 ctl_extent_check(union ctl_io *io1, union ctl_io *io2, bool seq)
10943 {
10944         uint64_t lba1, lba2;
10945         uint64_t len1, len2;
10946         int retval;
10947
10948         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10949                 return (CTL_ACTION_ERROR);
10950
10951         retval = ctl_extent_check_unmap(io1, lba2, len2);
10952         if (retval != CTL_ACTION_ERROR)
10953                 return (retval);
10954
10955         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10956                 return (CTL_ACTION_ERROR);
10957
10958         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10959                 seq = FALSE;
10960         return (ctl_extent_check_lba(lba1, len1, lba2, len2, seq));
10961 }
10962
10963 static ctl_action
10964 ctl_extent_check_seq(union ctl_io *io1, union ctl_io *io2)
10965 {
10966         uint64_t lba1, lba2;
10967         uint64_t len1, len2;
10968
10969         if (io1->io_hdr.flags & CTL_FLAG_SERSEQ_DONE)
10970                 return (CTL_ACTION_PASS);
10971         if (ctl_get_lba_len(io1, &lba1, &len1) != 0)
10972                 return (CTL_ACTION_ERROR);
10973         if (ctl_get_lba_len(io2, &lba2, &len2) != 0)
10974                 return (CTL_ACTION_ERROR);
10975
10976         if (lba1 + len1 == lba2)
10977                 return (CTL_ACTION_BLOCK);
10978         return (CTL_ACTION_PASS);
10979 }
10980
10981 static ctl_action
10982 ctl_check_for_blockage(struct ctl_lun *lun, union ctl_io *pending_io,
10983     union ctl_io *ooa_io)
10984 {
10985         const struct ctl_cmd_entry *pending_entry, *ooa_entry;
10986         const ctl_serialize_action *serialize_row;
10987
10988         /*
10989          * The initiator attempted multiple untagged commands at the same
10990          * time.  Can't do that.
10991          */
10992         if ((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10993          && (ooa_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
10994          && ((pending_io->io_hdr.nexus.targ_port ==
10995               ooa_io->io_hdr.nexus.targ_port)
10996           && (pending_io->io_hdr.nexus.initid ==
10997               ooa_io->io_hdr.nexus.initid))
10998          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
10999               CTL_FLAG_STATUS_SENT)) == 0))
11000                 return (CTL_ACTION_OVERLAP);
11001
11002         /*
11003          * The initiator attempted to send multiple tagged commands with
11004          * the same ID.  (It's fine if different initiators have the same
11005          * tag ID.)
11006          *
11007          * Even if all of those conditions are true, we don't kill the I/O
11008          * if the command ahead of us has been aborted.  We won't end up
11009          * sending it to the FETD, and it's perfectly legal to resend a
11010          * command with the same tag number as long as the previous
11011          * instance of this tag number has been aborted somehow.
11012          */
11013         if ((pending_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11014          && (ooa_io->scsiio.tag_type != CTL_TAG_UNTAGGED)
11015          && (pending_io->scsiio.tag_num == ooa_io->scsiio.tag_num)
11016          && ((pending_io->io_hdr.nexus.targ_port ==
11017               ooa_io->io_hdr.nexus.targ_port)
11018           && (pending_io->io_hdr.nexus.initid ==
11019               ooa_io->io_hdr.nexus.initid))
11020          && ((ooa_io->io_hdr.flags & (CTL_FLAG_ABORT |
11021               CTL_FLAG_STATUS_SENT)) == 0))
11022                 return (CTL_ACTION_OVERLAP_TAG);
11023
11024         /*
11025          * If we get a head of queue tag, SAM-3 says that we should
11026          * immediately execute it.
11027          *
11028          * What happens if this command would normally block for some other
11029          * reason?  e.g. a request sense with a head of queue tag
11030          * immediately after a write.  Normally that would block, but this
11031          * will result in its getting executed immediately...
11032          *
11033          * We currently return "pass" instead of "skip", so we'll end up
11034          * going through the rest of the queue to check for overlapped tags.
11035          *
11036          * XXX KDM check for other types of blockage first??
11037          */
11038         if (pending_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11039                 return (CTL_ACTION_PASS);
11040
11041         /*
11042          * Ordered tags have to block until all items ahead of them
11043          * have completed.  If we get called with an ordered tag, we always
11044          * block, if something else is ahead of us in the queue.
11045          */
11046         if (pending_io->scsiio.tag_type == CTL_TAG_ORDERED)
11047                 return (CTL_ACTION_BLOCK);
11048
11049         /*
11050          * Simple tags get blocked until all head of queue and ordered tags
11051          * ahead of them have completed.  I'm lumping untagged commands in
11052          * with simple tags here.  XXX KDM is that the right thing to do?
11053          */
11054         if (((pending_io->scsiio.tag_type == CTL_TAG_UNTAGGED)
11055           || (pending_io->scsiio.tag_type == CTL_TAG_SIMPLE))
11056          && ((ooa_io->scsiio.tag_type == CTL_TAG_HEAD_OF_QUEUE)
11057           || (ooa_io->scsiio.tag_type == CTL_TAG_ORDERED)))
11058                 return (CTL_ACTION_BLOCK);
11059
11060         pending_entry = ctl_get_cmd_entry(&pending_io->scsiio, NULL);
11061         KASSERT(pending_entry->seridx < CTL_SERIDX_COUNT,
11062             ("%s: Invalid seridx %d for pending CDB %02x %02x @ %p",
11063              __func__, pending_entry->seridx, pending_io->scsiio.cdb[0],
11064              pending_io->scsiio.cdb[1], pending_io));
11065         ooa_entry = ctl_get_cmd_entry(&ooa_io->scsiio, NULL);
11066         if (ooa_entry->seridx == CTL_SERIDX_INVLD)
11067                 return (CTL_ACTION_PASS); /* Unsupported command in OOA queue */
11068         KASSERT(ooa_entry->seridx < CTL_SERIDX_COUNT,
11069             ("%s: Invalid seridx %d for ooa CDB %02x %02x @ %p",
11070              __func__, ooa_entry->seridx, ooa_io->scsiio.cdb[0],
11071              ooa_io->scsiio.cdb[1], ooa_io));
11072
11073         serialize_row = ctl_serialize_table[ooa_entry->seridx];
11074
11075         switch (serialize_row[pending_entry->seridx]) {
11076         case CTL_SER_BLOCK:
11077                 return (CTL_ACTION_BLOCK);
11078         case CTL_SER_EXTENT:
11079                 return (ctl_extent_check(ooa_io, pending_io,
11080                     (lun->be_lun && lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11081         case CTL_SER_EXTENTOPT:
11082                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11083                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11084                         return (ctl_extent_check(ooa_io, pending_io,
11085                             (lun->be_lun &&
11086                              lun->be_lun->serseq == CTL_LUN_SERSEQ_ON)));
11087                 return (CTL_ACTION_PASS);
11088         case CTL_SER_EXTENTSEQ:
11089                 if (lun->be_lun && lun->be_lun->serseq != CTL_LUN_SERSEQ_OFF)
11090                         return (ctl_extent_check_seq(ooa_io, pending_io));
11091                 return (CTL_ACTION_PASS);
11092         case CTL_SER_PASS:
11093                 return (CTL_ACTION_PASS);
11094         case CTL_SER_BLOCKOPT:
11095                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT].queue_flags
11096                     & SCP_QUEUE_ALG_MASK) != SCP_QUEUE_ALG_UNRESTRICTED)
11097                         return (CTL_ACTION_BLOCK);
11098                 return (CTL_ACTION_PASS);
11099         case CTL_SER_SKIP:
11100                 return (CTL_ACTION_SKIP);
11101         default:
11102                 panic("%s: Invalid serialization value %d for %d => %d",
11103                     __func__, serialize_row[pending_entry->seridx],
11104                     pending_entry->seridx, ooa_entry->seridx);
11105         }
11106
11107         return (CTL_ACTION_ERROR);
11108 }
11109
11110 /*
11111  * Check for blockage or overlaps against the OOA (Order Of Arrival) queue.
11112  * Assumptions:
11113  * - pending_io is generally either incoming, or on the blocked queue
11114  * - starting I/O is the I/O we want to start the check with.
11115  */
11116 static ctl_action
11117 ctl_check_ooa(struct ctl_lun *lun, union ctl_io *pending_io,
11118               union ctl_io *starting_io)
11119 {
11120         union ctl_io *ooa_io;
11121         ctl_action action;
11122
11123         mtx_assert(&lun->lun_lock, MA_OWNED);
11124
11125         /*
11126          * Run back along the OOA queue, starting with the current
11127          * blocked I/O and going through every I/O before it on the
11128          * queue.  If starting_io is NULL, we'll just end up returning
11129          * CTL_ACTION_PASS.
11130          */
11131         for (ooa_io = starting_io; ooa_io != NULL;
11132              ooa_io = (union ctl_io *)TAILQ_PREV(&ooa_io->io_hdr, ctl_ooaq,
11133              ooa_links)){
11134
11135                 /*
11136                  * This routine just checks to see whether
11137                  * cur_blocked is blocked by ooa_io, which is ahead
11138                  * of it in the queue.  It doesn't queue/dequeue
11139                  * cur_blocked.
11140                  */
11141                 action = ctl_check_for_blockage(lun, pending_io, ooa_io);
11142                 switch (action) {
11143                 case CTL_ACTION_BLOCK:
11144                 case CTL_ACTION_OVERLAP:
11145                 case CTL_ACTION_OVERLAP_TAG:
11146                 case CTL_ACTION_SKIP:
11147                 case CTL_ACTION_ERROR:
11148                         return (action);
11149                         break; /* NOTREACHED */
11150                 case CTL_ACTION_PASS:
11151                         break;
11152                 default:
11153                         panic("%s: Invalid action %d\n", __func__, action);
11154                 }
11155         }
11156
11157         return (CTL_ACTION_PASS);
11158 }
11159
11160 /*
11161  * Assumptions:
11162  * - An I/O has just completed, and has been removed from the per-LUN OOA
11163  *   queue, so some items on the blocked queue may now be unblocked.
11164  */
11165 static int
11166 ctl_check_blocked(struct ctl_lun *lun)
11167 {
11168         struct ctl_softc *softc = lun->ctl_softc;
11169         union ctl_io *cur_blocked, *next_blocked;
11170
11171         mtx_assert(&lun->lun_lock, MA_OWNED);
11172
11173         /*
11174          * Run forward from the head of the blocked queue, checking each
11175          * entry against the I/Os prior to it on the OOA queue to see if
11176          * there is still any blockage.
11177          *
11178          * We cannot use the TAILQ_FOREACH() macro, because it can't deal
11179          * with our removing a variable on it while it is traversing the
11180          * list.
11181          */
11182         for (cur_blocked = (union ctl_io *)TAILQ_FIRST(&lun->blocked_queue);
11183              cur_blocked != NULL; cur_blocked = next_blocked) {
11184                 union ctl_io *prev_ooa;
11185                 ctl_action action;
11186
11187                 next_blocked = (union ctl_io *)TAILQ_NEXT(&cur_blocked->io_hdr,
11188                                                           blocked_links);
11189
11190                 prev_ooa = (union ctl_io *)TAILQ_PREV(&cur_blocked->io_hdr,
11191                                                       ctl_ooaq, ooa_links);
11192
11193                 /*
11194                  * If cur_blocked happens to be the first item in the OOA
11195                  * queue now, prev_ooa will be NULL, and the action
11196                  * returned will just be CTL_ACTION_PASS.
11197                  */
11198                 action = ctl_check_ooa(lun, cur_blocked, prev_ooa);
11199
11200                 switch (action) {
11201                 case CTL_ACTION_BLOCK:
11202                         /* Nothing to do here, still blocked */
11203                         break;
11204                 case CTL_ACTION_OVERLAP:
11205                 case CTL_ACTION_OVERLAP_TAG:
11206                         /*
11207                          * This shouldn't happen!  In theory we've already
11208                          * checked this command for overlap...
11209                          */
11210                         break;
11211                 case CTL_ACTION_PASS:
11212                 case CTL_ACTION_SKIP: {
11213                         const struct ctl_cmd_entry *entry;
11214
11215                         /*
11216                          * The skip case shouldn't happen, this transaction
11217                          * should have never made it onto the blocked queue.
11218                          */
11219                         /*
11220                          * This I/O is no longer blocked, we can remove it
11221                          * from the blocked queue.  Since this is a TAILQ
11222                          * (doubly linked list), we can do O(1) removals
11223                          * from any place on the list.
11224                          */
11225                         TAILQ_REMOVE(&lun->blocked_queue, &cur_blocked->io_hdr,
11226                                      blocked_links);
11227                         cur_blocked->io_hdr.flags &= ~CTL_FLAG_BLOCKED;
11228
11229                         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
11230                             (cur_blocked->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)){
11231                                 /*
11232                                  * Need to send IO back to original side to
11233                                  * run
11234                                  */
11235                                 union ctl_ha_msg msg_info;
11236
11237                                 cur_blocked->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11238                                 msg_info.hdr.original_sc =
11239                                         cur_blocked->io_hdr.original_sc;
11240                                 msg_info.hdr.serializing_sc = cur_blocked;
11241                                 msg_info.hdr.msg_type = CTL_MSG_R2R;
11242                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11243                                     sizeof(msg_info.hdr), M_NOWAIT);
11244                                 break;
11245                         }
11246                         entry = ctl_get_cmd_entry(&cur_blocked->scsiio, NULL);
11247
11248                         /*
11249                          * Check this I/O for LUN state changes that may
11250                          * have happened while this command was blocked.
11251                          * The LUN state may have been changed by a command
11252                          * ahead of us in the queue, so we need to re-check
11253                          * for any states that can be caused by SCSI
11254                          * commands.
11255                          */
11256                         if (ctl_scsiio_lun_check(lun, entry,
11257                                                  &cur_blocked->scsiio) == 0) {
11258                                 cur_blocked->io_hdr.flags |=
11259                                                       CTL_FLAG_IS_WAS_ON_RTR;
11260                                 ctl_enqueue_rtr(cur_blocked);
11261                         } else
11262                                 ctl_done(cur_blocked);
11263                         break;
11264                 }
11265                 default:
11266                         /*
11267                          * This probably shouldn't happen -- we shouldn't
11268                          * get CTL_ACTION_ERROR, or anything else.
11269                          */
11270                         break;
11271                 }
11272         }
11273
11274         return (CTL_RETVAL_COMPLETE);
11275 }
11276
11277 /*
11278  * This routine (with one exception) checks LUN flags that can be set by
11279  * commands ahead of us in the OOA queue.  These flags have to be checked
11280  * when a command initially comes in, and when we pull a command off the
11281  * blocked queue and are preparing to execute it.  The reason we have to
11282  * check these flags for commands on the blocked queue is that the LUN
11283  * state may have been changed by a command ahead of us while we're on the
11284  * blocked queue.
11285  *
11286  * Ordering is somewhat important with these checks, so please pay
11287  * careful attention to the placement of any new checks.
11288  */
11289 static int
11290 ctl_scsiio_lun_check(struct ctl_lun *lun,
11291     const struct ctl_cmd_entry *entry, struct ctl_scsiio *ctsio)
11292 {
11293         struct ctl_softc *softc = lun->ctl_softc;
11294         int retval;
11295         uint32_t residx;
11296
11297         retval = 0;
11298
11299         mtx_assert(&lun->lun_lock, MA_OWNED);
11300
11301         /*
11302          * If this shelf is a secondary shelf controller, we may have to
11303          * reject some commands disallowed by HA mode and link state.
11304          */
11305         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0) {
11306                 if (softc->ha_link == CTL_HA_LINK_OFFLINE &&
11307                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11308                         ctl_set_lun_unavail(ctsio);
11309                         retval = 1;
11310                         goto bailout;
11311                 }
11312                 if ((lun->flags & CTL_LUN_PEER_SC_PRIMARY) == 0 &&
11313                     (entry->flags & CTL_CMD_FLAG_OK_ON_UNAVAIL) == 0) {
11314                         ctl_set_lun_transit(ctsio);
11315                         retval = 1;
11316                         goto bailout;
11317                 }
11318                 if (softc->ha_mode == CTL_HA_MODE_ACT_STBY &&
11319                     (entry->flags & CTL_CMD_FLAG_OK_ON_STANDBY) == 0) {
11320                         ctl_set_lun_standby(ctsio);
11321                         retval = 1;
11322                         goto bailout;
11323                 }
11324
11325                 /* The rest of checks are only done on executing side */
11326                 if (softc->ha_mode == CTL_HA_MODE_XFER)
11327                         goto bailout;
11328         }
11329
11330         if (entry->pattern & CTL_LUN_PAT_WRITE) {
11331                 if (lun->be_lun &&
11332                     lun->be_lun->flags & CTL_LUN_FLAG_READONLY) {
11333                         ctl_set_hw_write_protected(ctsio);
11334                         retval = 1;
11335                         goto bailout;
11336                 }
11337                 if ((lun->mode_pages.control_page[CTL_PAGE_CURRENT]
11338                     .eca_and_aen & SCP_SWP) != 0) {
11339                         ctl_set_sense(ctsio, /*current_error*/ 1,
11340                             /*sense_key*/ SSD_KEY_DATA_PROTECT,
11341                             /*asc*/ 0x27, /*ascq*/ 0x02, SSD_ELEM_NONE);
11342                         retval = 1;
11343                         goto bailout;
11344                 }
11345         }
11346
11347         /*
11348          * Check for a reservation conflict.  If this command isn't allowed
11349          * even on reserved LUNs, and if this initiator isn't the one who
11350          * reserved us, reject the command with a reservation conflict.
11351          */
11352         residx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11353         if ((lun->flags & CTL_LUN_RESERVED)
11354          && ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_RESV) == 0)) {
11355                 if (lun->res_idx != residx) {
11356                         ctl_set_reservation_conflict(ctsio);
11357                         retval = 1;
11358                         goto bailout;
11359                 }
11360         }
11361
11362         if ((lun->flags & CTL_LUN_PR_RESERVED) == 0 ||
11363             (entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_RESV)) {
11364                 /* No reservation or command is allowed. */;
11365         } else if ((entry->flags & CTL_CMD_FLAG_ALLOW_ON_PR_WRESV) &&
11366             (lun->pr_res_type == SPR_TYPE_WR_EX ||
11367              lun->pr_res_type == SPR_TYPE_WR_EX_RO ||
11368              lun->pr_res_type == SPR_TYPE_WR_EX_AR)) {
11369                 /* The command is allowed for Write Exclusive resv. */;
11370         } else {
11371                 /*
11372                  * if we aren't registered or it's a res holder type
11373                  * reservation and this isn't the res holder then set a
11374                  * conflict.
11375                  */
11376                 if (ctl_get_prkey(lun, residx) == 0 ||
11377                     (residx != lun->pr_res_idx && lun->pr_res_type < 4)) {
11378                         ctl_set_reservation_conflict(ctsio);
11379                         retval = 1;
11380                         goto bailout;
11381                 }
11382         }
11383
11384         if ((entry->flags & CTL_CMD_FLAG_OK_ON_NO_MEDIA) == 0) {
11385                 if (lun->flags & CTL_LUN_EJECTED)
11386                         ctl_set_lun_ejected(ctsio);
11387                 else if (lun->flags & CTL_LUN_NO_MEDIA) {
11388                         if (lun->flags & CTL_LUN_REMOVABLE)
11389                                 ctl_set_lun_no_media(ctsio);
11390                         else
11391                                 ctl_set_lun_int_reqd(ctsio);
11392                 } else if (lun->flags & CTL_LUN_STOPPED)
11393                         ctl_set_lun_stopped(ctsio);
11394                 else
11395                         goto bailout;
11396                 retval = 1;
11397                 goto bailout;
11398         }
11399
11400 bailout:
11401         return (retval);
11402 }
11403
11404 static void
11405 ctl_failover_io(union ctl_io *io, int have_lock)
11406 {
11407         ctl_set_busy(&io->scsiio);
11408         ctl_done(io);
11409 }
11410
11411 static void
11412 ctl_failover_lun(union ctl_io *rio)
11413 {
11414         struct ctl_softc *softc = control_softc;
11415         struct ctl_lun *lun;
11416         struct ctl_io_hdr *io, *next_io;
11417         uint32_t targ_lun;
11418
11419         targ_lun = rio->io_hdr.nexus.targ_mapped_lun;
11420         CTL_DEBUG_PRINT(("FAILOVER for lun %ju\n", targ_lun));
11421
11422         /* Find and lock the LUN. */
11423         mtx_lock(&softc->ctl_lock);
11424         if ((targ_lun < CTL_MAX_LUNS) &&
11425             ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11426                 mtx_lock(&lun->lun_lock);
11427                 mtx_unlock(&softc->ctl_lock);
11428                 if (lun->flags & CTL_LUN_DISABLED) {
11429                         mtx_unlock(&lun->lun_lock);
11430                         return;
11431                 }
11432         } else {
11433                 mtx_unlock(&softc->ctl_lock);
11434                 return;
11435         }
11436
11437         if (softc->ha_mode == CTL_HA_MODE_XFER) {
11438                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11439                         /* We are master */
11440                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11441                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11442                                         io->flags |= CTL_FLAG_ABORT;
11443                                         io->flags |= CTL_FLAG_FAILOVER;
11444                                 } else { /* This can be only due to DATAMOVE */
11445                                         io->msg_type = CTL_MSG_DATAMOVE_DONE;
11446                                         io->flags &= ~CTL_FLAG_DMA_INPROG;
11447                                         io->flags |= CTL_FLAG_IO_ACTIVE;
11448                                         io->port_status = 31340;
11449                                         ctl_enqueue_isc((union ctl_io *)io);
11450                                 }
11451                         }
11452                         /* We are slave */
11453                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11454                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11455                                 if (io->flags & CTL_FLAG_IO_ACTIVE) {
11456                                         io->flags |= CTL_FLAG_FAILOVER;
11457                                 } else {
11458                                         ctl_set_busy(&((union ctl_io *)io)->
11459                                             scsiio);
11460                                         ctl_done((union ctl_io *)io);
11461                                 }
11462                         }
11463                 }
11464         } else { /* SERIALIZE modes */
11465                 TAILQ_FOREACH_SAFE(io, &lun->blocked_queue, blocked_links,
11466                     next_io) {
11467                         /* We are master */
11468                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11469                                 TAILQ_REMOVE(&lun->blocked_queue, io,
11470                                     blocked_links);
11471                                 io->flags &= ~CTL_FLAG_BLOCKED;
11472                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11473                                 ctl_free_io((union ctl_io *)io);
11474                         }
11475                 }
11476                 TAILQ_FOREACH_SAFE(io, &lun->ooa_queue, ooa_links, next_io) {
11477                         /* We are master */
11478                         if (io->flags & CTL_FLAG_FROM_OTHER_SC) {
11479                                 TAILQ_REMOVE(&lun->ooa_queue, io, ooa_links);
11480                                 ctl_free_io((union ctl_io *)io);
11481                         }
11482                         /* We are slave */
11483                         if (io->flags & CTL_FLAG_SENT_2OTHER_SC) {
11484                                 io->flags &= ~CTL_FLAG_SENT_2OTHER_SC;
11485                                 if (!(io->flags & CTL_FLAG_IO_ACTIVE)) {
11486                                         ctl_set_busy(&((union ctl_io *)io)->
11487                                             scsiio);
11488                                         ctl_done((union ctl_io *)io);
11489                                 }
11490                         }
11491                 }
11492                 ctl_check_blocked(lun);
11493         }
11494         mtx_unlock(&lun->lun_lock);
11495 }
11496
11497 static int
11498 ctl_scsiio_precheck(struct ctl_softc *softc, struct ctl_scsiio *ctsio)
11499 {
11500         struct ctl_lun *lun;
11501         const struct ctl_cmd_entry *entry;
11502         uint32_t initidx, targ_lun;
11503         int retval;
11504
11505         retval = 0;
11506
11507         lun = NULL;
11508
11509         targ_lun = ctsio->io_hdr.nexus.targ_mapped_lun;
11510         if ((targ_lun < CTL_MAX_LUNS)
11511          && ((lun = softc->ctl_luns[targ_lun]) != NULL)) {
11512                 /*
11513                  * If the LUN is invalid, pretend that it doesn't exist.
11514                  * It will go away as soon as all pending I/O has been
11515                  * completed.
11516                  */
11517                 mtx_lock(&lun->lun_lock);
11518                 if (lun->flags & CTL_LUN_DISABLED) {
11519                         mtx_unlock(&lun->lun_lock);
11520                         lun = NULL;
11521                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11522                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11523                 } else {
11524                         ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = lun;
11525                         ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr =
11526                                 lun->be_lun;
11527
11528                         /*
11529                          * Every I/O goes into the OOA queue for a
11530                          * particular LUN, and stays there until completion.
11531                          */
11532 #ifdef CTL_TIME_IO
11533                         if (TAILQ_EMPTY(&lun->ooa_queue)) {
11534                                 lun->idle_time += getsbinuptime() -
11535                                     lun->last_busy;
11536                         }
11537 #endif
11538                         TAILQ_INSERT_TAIL(&lun->ooa_queue, &ctsio->io_hdr,
11539                             ooa_links);
11540                 }
11541         } else {
11542                 ctsio->io_hdr.ctl_private[CTL_PRIV_LUN].ptr = NULL;
11543                 ctsio->io_hdr.ctl_private[CTL_PRIV_BACKEND_LUN].ptr = NULL;
11544         }
11545
11546         /* Get command entry and return error if it is unsuppotyed. */
11547         entry = ctl_validate_command(ctsio);
11548         if (entry == NULL) {
11549                 if (lun)
11550                         mtx_unlock(&lun->lun_lock);
11551                 return (retval);
11552         }
11553
11554         ctsio->io_hdr.flags &= ~CTL_FLAG_DATA_MASK;
11555         ctsio->io_hdr.flags |= entry->flags & CTL_FLAG_DATA_MASK;
11556
11557         /*
11558          * Check to see whether we can send this command to LUNs that don't
11559          * exist.  This should pretty much only be the case for inquiry
11560          * and request sense.  Further checks, below, really require having
11561          * a LUN, so we can't really check the command anymore.  Just put
11562          * it on the rtr queue.
11563          */
11564         if (lun == NULL) {
11565                 if (entry->flags & CTL_CMD_FLAG_OK_ON_NO_LUN) {
11566                         ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11567                         ctl_enqueue_rtr((union ctl_io *)ctsio);
11568                         return (retval);
11569                 }
11570
11571                 ctl_set_unsupported_lun(ctsio);
11572                 ctl_done((union ctl_io *)ctsio);
11573                 CTL_DEBUG_PRINT(("ctl_scsiio_precheck: bailing out due to invalid LUN\n"));
11574                 return (retval);
11575         } else {
11576                 /*
11577                  * Make sure we support this particular command on this LUN.
11578                  * e.g., we don't support writes to the control LUN.
11579                  */
11580                 if (!ctl_cmd_applicable(lun->be_lun->lun_type, entry)) {
11581                         mtx_unlock(&lun->lun_lock);
11582                         ctl_set_invalid_opcode(ctsio);
11583                         ctl_done((union ctl_io *)ctsio);
11584                         return (retval);
11585                 }
11586         }
11587
11588         initidx = ctl_get_initindex(&ctsio->io_hdr.nexus);
11589
11590 #ifdef CTL_WITH_CA
11591         /*
11592          * If we've got a request sense, it'll clear the contingent
11593          * allegiance condition.  Otherwise, if we have a CA condition for
11594          * this initiator, clear it, because it sent down a command other
11595          * than request sense.
11596          */
11597         if ((ctsio->cdb[0] != REQUEST_SENSE)
11598          && (ctl_is_set(lun->have_ca, initidx)))
11599                 ctl_clear_mask(lun->have_ca, initidx);
11600 #endif
11601
11602         /*
11603          * If the command has this flag set, it handles its own unit
11604          * attention reporting, we shouldn't do anything.  Otherwise we
11605          * check for any pending unit attentions, and send them back to the
11606          * initiator.  We only do this when a command initially comes in,
11607          * not when we pull it off the blocked queue.
11608          *
11609          * According to SAM-3, section 5.3.2, the order that things get
11610          * presented back to the host is basically unit attentions caused
11611          * by some sort of reset event, busy status, reservation conflicts
11612          * or task set full, and finally any other status.
11613          *
11614          * One issue here is that some of the unit attentions we report
11615          * don't fall into the "reset" category (e.g. "reported luns data
11616          * has changed").  So reporting it here, before the reservation
11617          * check, may be technically wrong.  I guess the only thing to do
11618          * would be to check for and report the reset events here, and then
11619          * check for the other unit attention types after we check for a
11620          * reservation conflict.
11621          *
11622          * XXX KDM need to fix this
11623          */
11624         if ((entry->flags & CTL_CMD_FLAG_NO_SENSE) == 0) {
11625                 ctl_ua_type ua_type;
11626
11627                 ua_type = ctl_build_ua(lun, initidx, &ctsio->sense_data,
11628                     SSD_TYPE_NONE);
11629                 if (ua_type != CTL_UA_NONE) {
11630                         mtx_unlock(&lun->lun_lock);
11631                         ctsio->scsi_status = SCSI_STATUS_CHECK_COND;
11632                         ctsio->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
11633                         ctsio->sense_len = SSD_FULL_SIZE;
11634                         ctl_done((union ctl_io *)ctsio);
11635                         return (retval);
11636                 }
11637         }
11638
11639
11640         if (ctl_scsiio_lun_check(lun, entry, ctsio) != 0) {
11641                 mtx_unlock(&lun->lun_lock);
11642                 ctl_done((union ctl_io *)ctsio);
11643                 return (retval);
11644         }
11645
11646         /*
11647          * XXX CHD this is where we want to send IO to other side if
11648          * this LUN is secondary on this SC. We will need to make a copy
11649          * of the IO and flag the IO on this side as SENT_2OTHER and the flag
11650          * the copy we send as FROM_OTHER.
11651          * We also need to stuff the address of the original IO so we can
11652          * find it easily. Something similar will need be done on the other
11653          * side so when we are done we can find the copy.
11654          */
11655         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
11656             (lun->flags & CTL_LUN_PEER_SC_PRIMARY) != 0 &&
11657             (entry->flags & CTL_CMD_FLAG_RUN_HERE) == 0) {
11658                 union ctl_ha_msg msg_info;
11659                 int isc_retval;
11660
11661                 ctsio->io_hdr.flags |= CTL_FLAG_SENT_2OTHER_SC;
11662                 ctsio->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
11663                 mtx_unlock(&lun->lun_lock);
11664
11665                 msg_info.hdr.msg_type = CTL_MSG_SERIALIZE;
11666                 msg_info.hdr.original_sc = (union ctl_io *)ctsio;
11667                 msg_info.hdr.serializing_sc = NULL;
11668                 msg_info.hdr.nexus = ctsio->io_hdr.nexus;
11669                 msg_info.scsi.tag_num = ctsio->tag_num;
11670                 msg_info.scsi.tag_type = ctsio->tag_type;
11671                 msg_info.scsi.cdb_len = ctsio->cdb_len;
11672                 memcpy(msg_info.scsi.cdb, ctsio->cdb, CTL_MAX_CDBLEN);
11673
11674                 if ((isc_retval = ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11675                     sizeof(msg_info.scsi) - sizeof(msg_info.scsi.sense_data),
11676                     M_WAITOK)) > CTL_HA_STATUS_SUCCESS) {
11677                         ctl_set_busy(ctsio);
11678                         ctl_done((union ctl_io *)ctsio);
11679                         return (retval);
11680                 }
11681                 return (retval);
11682         }
11683
11684         switch (ctl_check_ooa(lun, (union ctl_io *)ctsio,
11685                               (union ctl_io *)TAILQ_PREV(&ctsio->io_hdr,
11686                               ctl_ooaq, ooa_links))) {
11687         case CTL_ACTION_BLOCK:
11688                 ctsio->io_hdr.flags |= CTL_FLAG_BLOCKED;
11689                 TAILQ_INSERT_TAIL(&lun->blocked_queue, &ctsio->io_hdr,
11690                                   blocked_links);
11691                 mtx_unlock(&lun->lun_lock);
11692                 return (retval);
11693         case CTL_ACTION_PASS:
11694         case CTL_ACTION_SKIP:
11695                 ctsio->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
11696                 mtx_unlock(&lun->lun_lock);
11697                 ctl_enqueue_rtr((union ctl_io *)ctsio);
11698                 break;
11699         case CTL_ACTION_OVERLAP:
11700                 mtx_unlock(&lun->lun_lock);
11701                 ctl_set_overlapped_cmd(ctsio);
11702                 ctl_done((union ctl_io *)ctsio);
11703                 break;
11704         case CTL_ACTION_OVERLAP_TAG:
11705                 mtx_unlock(&lun->lun_lock);
11706                 ctl_set_overlapped_tag(ctsio, ctsio->tag_num & 0xff);
11707                 ctl_done((union ctl_io *)ctsio);
11708                 break;
11709         case CTL_ACTION_ERROR:
11710         default:
11711                 mtx_unlock(&lun->lun_lock);
11712                 ctl_set_internal_failure(ctsio,
11713                                          /*sks_valid*/ 0,
11714                                          /*retry_count*/ 0);
11715                 ctl_done((union ctl_io *)ctsio);
11716                 break;
11717         }
11718         return (retval);
11719 }
11720
11721 const struct ctl_cmd_entry *
11722 ctl_get_cmd_entry(struct ctl_scsiio *ctsio, int *sa)
11723 {
11724         const struct ctl_cmd_entry *entry;
11725         int service_action;
11726
11727         entry = &ctl_cmd_table[ctsio->cdb[0]];
11728         if (sa)
11729                 *sa = ((entry->flags & CTL_CMD_FLAG_SA5) != 0);
11730         if (entry->flags & CTL_CMD_FLAG_SA5) {
11731                 service_action = ctsio->cdb[1] & SERVICE_ACTION_MASK;
11732                 entry = &((const struct ctl_cmd_entry *)
11733                     entry->execute)[service_action];
11734         }
11735         return (entry);
11736 }
11737
11738 const struct ctl_cmd_entry *
11739 ctl_validate_command(struct ctl_scsiio *ctsio)
11740 {
11741         const struct ctl_cmd_entry *entry;
11742         int i, sa;
11743         uint8_t diff;
11744
11745         entry = ctl_get_cmd_entry(ctsio, &sa);
11746         if (entry->execute == NULL) {
11747                 if (sa)
11748                         ctl_set_invalid_field(ctsio,
11749                                               /*sks_valid*/ 1,
11750                                               /*command*/ 1,
11751                                               /*field*/ 1,
11752                                               /*bit_valid*/ 1,
11753                                               /*bit*/ 4);
11754                 else
11755                         ctl_set_invalid_opcode(ctsio);
11756                 ctl_done((union ctl_io *)ctsio);
11757                 return (NULL);
11758         }
11759         KASSERT(entry->length > 0,
11760             ("Not defined length for command 0x%02x/0x%02x",
11761              ctsio->cdb[0], ctsio->cdb[1]));
11762         for (i = 1; i < entry->length; i++) {
11763                 diff = ctsio->cdb[i] & ~entry->usage[i - 1];
11764                 if (diff == 0)
11765                         continue;
11766                 ctl_set_invalid_field(ctsio,
11767                                       /*sks_valid*/ 1,
11768                                       /*command*/ 1,
11769                                       /*field*/ i,
11770                                       /*bit_valid*/ 1,
11771                                       /*bit*/ fls(diff) - 1);
11772                 ctl_done((union ctl_io *)ctsio);
11773                 return (NULL);
11774         }
11775         return (entry);
11776 }
11777
11778 static int
11779 ctl_cmd_applicable(uint8_t lun_type, const struct ctl_cmd_entry *entry)
11780 {
11781
11782         switch (lun_type) {
11783         case T_DIRECT:
11784                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_DIRECT) == 0)
11785                         return (0);
11786                 break;
11787         case T_PROCESSOR:
11788                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_PROC) == 0)
11789                         return (0);
11790                 break;
11791         case T_CDROM:
11792                 if ((entry->flags & CTL_CMD_FLAG_OK_ON_CDROM) == 0)
11793                         return (0);
11794                 break;
11795         default:
11796                 return (0);
11797         }
11798         return (1);
11799 }
11800
11801 static int
11802 ctl_scsiio(struct ctl_scsiio *ctsio)
11803 {
11804         int retval;
11805         const struct ctl_cmd_entry *entry;
11806
11807         retval = CTL_RETVAL_COMPLETE;
11808
11809         CTL_DEBUG_PRINT(("ctl_scsiio cdb[0]=%02X\n", ctsio->cdb[0]));
11810
11811         entry = ctl_get_cmd_entry(ctsio, NULL);
11812
11813         /*
11814          * If this I/O has been aborted, just send it straight to
11815          * ctl_done() without executing it.
11816          */
11817         if (ctsio->io_hdr.flags & CTL_FLAG_ABORT) {
11818                 ctl_done((union ctl_io *)ctsio);
11819                 goto bailout;
11820         }
11821
11822         /*
11823          * All the checks should have been handled by ctl_scsiio_precheck().
11824          * We should be clear now to just execute the I/O.
11825          */
11826         retval = entry->execute(ctsio);
11827
11828 bailout:
11829         return (retval);
11830 }
11831
11832 /*
11833  * Since we only implement one target right now, a bus reset simply resets
11834  * our single target.
11835  */
11836 static int
11837 ctl_bus_reset(struct ctl_softc *softc, union ctl_io *io)
11838 {
11839         return(ctl_target_reset(softc, io, CTL_UA_BUS_RESET));
11840 }
11841
11842 static int
11843 ctl_target_reset(struct ctl_softc *softc, union ctl_io *io,
11844                  ctl_ua_type ua_type)
11845 {
11846         struct ctl_port *port;
11847         struct ctl_lun *lun;
11848         int retval;
11849
11850         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
11851                 union ctl_ha_msg msg_info;
11852
11853                 msg_info.hdr.nexus = io->io_hdr.nexus;
11854                 if (ua_type==CTL_UA_TARG_RESET)
11855                         msg_info.task.task_action = CTL_TASK_TARGET_RESET;
11856                 else
11857                         msg_info.task.task_action = CTL_TASK_BUS_RESET;
11858                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11859                 msg_info.hdr.original_sc = NULL;
11860                 msg_info.hdr.serializing_sc = NULL;
11861                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11862                     sizeof(msg_info.task), M_WAITOK);
11863         }
11864         retval = 0;
11865
11866         mtx_lock(&softc->ctl_lock);
11867         port = ctl_io_port(&io->io_hdr);
11868         STAILQ_FOREACH(lun, &softc->lun_list, links) {
11869                 if (port != NULL &&
11870                     ctl_lun_map_to_port(port, lun->lun) >= CTL_MAX_LUNS)
11871                         continue;
11872                 retval += ctl_do_lun_reset(lun, io, ua_type);
11873         }
11874         mtx_unlock(&softc->ctl_lock);
11875         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11876         return (retval);
11877 }
11878
11879 /*
11880  * The LUN should always be set.  The I/O is optional, and is used to
11881  * distinguish between I/Os sent by this initiator, and by other
11882  * initiators.  We set unit attention for initiators other than this one.
11883  * SAM-3 is vague on this point.  It does say that a unit attention should
11884  * be established for other initiators when a LUN is reset (see section
11885  * 5.7.3), but it doesn't specifically say that the unit attention should
11886  * be established for this particular initiator when a LUN is reset.  Here
11887  * is the relevant text, from SAM-3 rev 8:
11888  *
11889  * 5.7.2 When a SCSI initiator port aborts its own tasks
11890  *
11891  * When a SCSI initiator port causes its own task(s) to be aborted, no
11892  * notification that the task(s) have been aborted shall be returned to
11893  * the SCSI initiator port other than the completion response for the
11894  * command or task management function action that caused the task(s) to
11895  * be aborted and notification(s) associated with related effects of the
11896  * action (e.g., a reset unit attention condition).
11897  *
11898  * XXX KDM for now, we're setting unit attention for all initiators.
11899  */
11900 static int
11901 ctl_do_lun_reset(struct ctl_lun *lun, union ctl_io *io, ctl_ua_type ua_type)
11902 {
11903         union ctl_io *xio;
11904 #if 0
11905         uint32_t initidx;
11906 #endif
11907         int i;
11908
11909         mtx_lock(&lun->lun_lock);
11910         /*
11911          * Run through the OOA queue and abort each I/O.
11912          */
11913         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11914              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
11915                 xio->io_hdr.flags |= CTL_FLAG_ABORT | CTL_FLAG_ABORT_STATUS;
11916         }
11917
11918         /*
11919          * This version sets unit attention for every
11920          */
11921 #if 0
11922         initidx = ctl_get_initindex(&io->io_hdr.nexus);
11923         ctl_est_ua_all(lun, initidx, ua_type);
11924 #else
11925         ctl_est_ua_all(lun, -1, ua_type);
11926 #endif
11927
11928         /*
11929          * A reset (any kind, really) clears reservations established with
11930          * RESERVE/RELEASE.  It does not clear reservations established
11931          * with PERSISTENT RESERVE OUT, but we don't support that at the
11932          * moment anyway.  See SPC-2, section 5.6.  SPC-3 doesn't address
11933          * reservations made with the RESERVE/RELEASE commands, because
11934          * those commands are obsolete in SPC-3.
11935          */
11936         lun->flags &= ~CTL_LUN_RESERVED;
11937
11938 #ifdef CTL_WITH_CA
11939         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11940                 ctl_clear_mask(lun->have_ca, i);
11941 #endif
11942         lun->prevent_count = 0;
11943         for (i = 0; i < CTL_MAX_INITIATORS; i++)
11944                 ctl_clear_mask(lun->prevent, i);
11945         mtx_unlock(&lun->lun_lock);
11946
11947         return (0);
11948 }
11949
11950 static int
11951 ctl_lun_reset(struct ctl_softc *softc, union ctl_io *io)
11952 {
11953         struct ctl_lun *lun;
11954         uint32_t targ_lun;
11955         int retval;
11956
11957         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
11958         mtx_lock(&softc->ctl_lock);
11959         if ((targ_lun >= CTL_MAX_LUNS) ||
11960             (lun = softc->ctl_luns[targ_lun]) == NULL) {
11961                 mtx_unlock(&softc->ctl_lock);
11962                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
11963                 return (1);
11964         }
11965         retval = ctl_do_lun_reset(lun, io, CTL_UA_LUN_RESET);
11966         mtx_unlock(&softc->ctl_lock);
11967         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
11968
11969         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0) {
11970                 union ctl_ha_msg msg_info;
11971
11972                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
11973                 msg_info.hdr.nexus = io->io_hdr.nexus;
11974                 msg_info.task.task_action = CTL_TASK_LUN_RESET;
11975                 msg_info.hdr.original_sc = NULL;
11976                 msg_info.hdr.serializing_sc = NULL;
11977                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
11978                     sizeof(msg_info.task), M_WAITOK);
11979         }
11980         return (retval);
11981 }
11982
11983 static void
11984 ctl_abort_tasks_lun(struct ctl_lun *lun, uint32_t targ_port, uint32_t init_id,
11985     int other_sc)
11986 {
11987         union ctl_io *xio;
11988
11989         mtx_assert(&lun->lun_lock, MA_OWNED);
11990
11991         /*
11992          * Run through the OOA queue and attempt to find the given I/O.
11993          * The target port, initiator ID, tag type and tag number have to
11994          * match the values that we got from the initiator.  If we have an
11995          * untagged command to abort, simply abort the first untagged command
11996          * we come to.  We only allow one untagged command at a time of course.
11997          */
11998         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
11999              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12000
12001                 if ((targ_port == UINT32_MAX ||
12002                      targ_port == xio->io_hdr.nexus.targ_port) &&
12003                     (init_id == UINT32_MAX ||
12004                      init_id == xio->io_hdr.nexus.initid)) {
12005                         if (targ_port != xio->io_hdr.nexus.targ_port ||
12006                             init_id != xio->io_hdr.nexus.initid)
12007                                 xio->io_hdr.flags |= CTL_FLAG_ABORT_STATUS;
12008                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12009                         if (!other_sc && !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12010                                 union ctl_ha_msg msg_info;
12011
12012                                 msg_info.hdr.nexus = xio->io_hdr.nexus;
12013                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12014                                 msg_info.task.tag_num = xio->scsiio.tag_num;
12015                                 msg_info.task.tag_type = xio->scsiio.tag_type;
12016                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12017                                 msg_info.hdr.original_sc = NULL;
12018                                 msg_info.hdr.serializing_sc = NULL;
12019                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12020                                     sizeof(msg_info.task), M_NOWAIT);
12021                         }
12022                 }
12023         }
12024 }
12025
12026 static int
12027 ctl_abort_task_set(union ctl_io *io)
12028 {
12029         struct ctl_softc *softc = control_softc;
12030         struct ctl_lun *lun;
12031         uint32_t targ_lun;
12032
12033         /*
12034          * Look up the LUN.
12035          */
12036         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12037         mtx_lock(&softc->ctl_lock);
12038         if ((targ_lun >= CTL_MAX_LUNS) ||
12039             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12040                 mtx_unlock(&softc->ctl_lock);
12041                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12042                 return (1);
12043         }
12044
12045         mtx_lock(&lun->lun_lock);
12046         mtx_unlock(&softc->ctl_lock);
12047         if (io->taskio.task_action == CTL_TASK_ABORT_TASK_SET) {
12048                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12049                     io->io_hdr.nexus.initid,
12050                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12051         } else { /* CTL_TASK_CLEAR_TASK_SET */
12052                 ctl_abort_tasks_lun(lun, UINT32_MAX, UINT32_MAX,
12053                     (io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) != 0);
12054         }
12055         mtx_unlock(&lun->lun_lock);
12056         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12057         return (0);
12058 }
12059
12060 static int
12061 ctl_i_t_nexus_reset(union ctl_io *io)
12062 {
12063         struct ctl_softc *softc = control_softc;
12064         struct ctl_lun *lun;
12065         uint32_t initidx;
12066
12067         if (!(io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC)) {
12068                 union ctl_ha_msg msg_info;
12069
12070                 msg_info.hdr.nexus = io->io_hdr.nexus;
12071                 msg_info.task.task_action = CTL_TASK_I_T_NEXUS_RESET;
12072                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12073                 msg_info.hdr.original_sc = NULL;
12074                 msg_info.hdr.serializing_sc = NULL;
12075                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12076                     sizeof(msg_info.task), M_WAITOK);
12077         }
12078
12079         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12080         mtx_lock(&softc->ctl_lock);
12081         STAILQ_FOREACH(lun, &softc->lun_list, links) {
12082                 mtx_lock(&lun->lun_lock);
12083                 ctl_abort_tasks_lun(lun, io->io_hdr.nexus.targ_port,
12084                     io->io_hdr.nexus.initid, 1);
12085 #ifdef CTL_WITH_CA
12086                 ctl_clear_mask(lun->have_ca, initidx);
12087 #endif
12088                 if ((lun->flags & CTL_LUN_RESERVED) && (lun->res_idx == initidx))
12089                         lun->flags &= ~CTL_LUN_RESERVED;
12090                 if (ctl_is_set(lun->prevent, initidx)) {
12091                         ctl_clear_mask(lun->prevent, initidx);
12092                         lun->prevent_count--;
12093                 }
12094                 ctl_est_ua(lun, initidx, CTL_UA_I_T_NEXUS_LOSS);
12095                 mtx_unlock(&lun->lun_lock);
12096         }
12097         mtx_unlock(&softc->ctl_lock);
12098         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12099         return (0);
12100 }
12101
12102 static int
12103 ctl_abort_task(union ctl_io *io)
12104 {
12105         union ctl_io *xio;
12106         struct ctl_lun *lun;
12107         struct ctl_softc *softc;
12108 #if 0
12109         struct sbuf sb;
12110         char printbuf[128];
12111 #endif
12112         int found;
12113         uint32_t targ_lun;
12114
12115         softc = control_softc;
12116         found = 0;
12117
12118         /*
12119          * Look up the LUN.
12120          */
12121         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12122         mtx_lock(&softc->ctl_lock);
12123         if ((targ_lun >= CTL_MAX_LUNS) ||
12124             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12125                 mtx_unlock(&softc->ctl_lock);
12126                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12127                 return (1);
12128         }
12129
12130 #if 0
12131         printf("ctl_abort_task: called for lun %lld, tag %d type %d\n",
12132                lun->lun, io->taskio.tag_num, io->taskio.tag_type);
12133 #endif
12134
12135         mtx_lock(&lun->lun_lock);
12136         mtx_unlock(&softc->ctl_lock);
12137         /*
12138          * Run through the OOA queue and attempt to find the given I/O.
12139          * The target port, initiator ID, tag type and tag number have to
12140          * match the values that we got from the initiator.  If we have an
12141          * untagged command to abort, simply abort the first untagged command
12142          * we come to.  We only allow one untagged command at a time of course.
12143          */
12144         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12145              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12146 #if 0
12147                 sbuf_new(&sb, printbuf, sizeof(printbuf), SBUF_FIXEDLEN);
12148
12149                 sbuf_printf(&sb, "LUN %lld tag %d type %d%s%s%s%s: ",
12150                             lun->lun, xio->scsiio.tag_num,
12151                             xio->scsiio.tag_type,
12152                             (xio->io_hdr.blocked_links.tqe_prev
12153                             == NULL) ? "" : " BLOCKED",
12154                             (xio->io_hdr.flags &
12155                             CTL_FLAG_DMA_INPROG) ? " DMA" : "",
12156                             (xio->io_hdr.flags &
12157                             CTL_FLAG_ABORT) ? " ABORT" : "",
12158                             (xio->io_hdr.flags &
12159                             CTL_FLAG_IS_WAS_ON_RTR ? " RTR" : ""));
12160                 ctl_scsi_command_string(&xio->scsiio, NULL, &sb);
12161                 sbuf_finish(&sb);
12162                 printf("%s\n", sbuf_data(&sb));
12163 #endif
12164
12165                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12166                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12167                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12168                         continue;
12169
12170                 /*
12171                  * If the abort says that the task is untagged, the
12172                  * task in the queue must be untagged.  Otherwise,
12173                  * we just check to see whether the tag numbers
12174                  * match.  This is because the QLogic firmware
12175                  * doesn't pass back the tag type in an abort
12176                  * request.
12177                  */
12178 #if 0
12179                 if (((xio->scsiio.tag_type == CTL_TAG_UNTAGGED)
12180                   && (io->taskio.tag_type == CTL_TAG_UNTAGGED))
12181                  || (xio->scsiio.tag_num == io->taskio.tag_num))
12182 #endif
12183                 /*
12184                  * XXX KDM we've got problems with FC, because it
12185                  * doesn't send down a tag type with aborts.  So we
12186                  * can only really go by the tag number...
12187                  * This may cause problems with parallel SCSI.
12188                  * Need to figure that out!!
12189                  */
12190                 if (xio->scsiio.tag_num == io->taskio.tag_num) {
12191                         xio->io_hdr.flags |= CTL_FLAG_ABORT;
12192                         found = 1;
12193                         if ((io->io_hdr.flags & CTL_FLAG_FROM_OTHER_SC) == 0 &&
12194                             !(lun->flags & CTL_LUN_PRIMARY_SC)) {
12195                                 union ctl_ha_msg msg_info;
12196
12197                                 msg_info.hdr.nexus = io->io_hdr.nexus;
12198                                 msg_info.task.task_action = CTL_TASK_ABORT_TASK;
12199                                 msg_info.task.tag_num = io->taskio.tag_num;
12200                                 msg_info.task.tag_type = io->taskio.tag_type;
12201                                 msg_info.hdr.msg_type = CTL_MSG_MANAGE_TASKS;
12202                                 msg_info.hdr.original_sc = NULL;
12203                                 msg_info.hdr.serializing_sc = NULL;
12204 #if 0
12205                                 printf("Sent Abort to other side\n");
12206 #endif
12207                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg_info,
12208                                     sizeof(msg_info.task), M_NOWAIT);
12209                         }
12210 #if 0
12211                         printf("ctl_abort_task: found I/O to abort\n");
12212 #endif
12213                 }
12214         }
12215         mtx_unlock(&lun->lun_lock);
12216
12217         if (found == 0) {
12218                 /*
12219                  * This isn't really an error.  It's entirely possible for
12220                  * the abort and command completion to cross on the wire.
12221                  * This is more of an informative/diagnostic error.
12222                  */
12223 #if 0
12224                 printf("ctl_abort_task: ABORT sent for nonexistent I/O: "
12225                        "%u:%u:%u tag %d type %d\n",
12226                        io->io_hdr.nexus.initid,
12227                        io->io_hdr.nexus.targ_port,
12228                        io->io_hdr.nexus.targ_lun, io->taskio.tag_num,
12229                        io->taskio.tag_type);
12230 #endif
12231         }
12232         io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12233         return (0);
12234 }
12235
12236 static int
12237 ctl_query_task(union ctl_io *io, int task_set)
12238 {
12239         union ctl_io *xio;
12240         struct ctl_lun *lun;
12241         struct ctl_softc *softc;
12242         int found = 0;
12243         uint32_t targ_lun;
12244
12245         softc = control_softc;
12246         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12247         mtx_lock(&softc->ctl_lock);
12248         if ((targ_lun >= CTL_MAX_LUNS) ||
12249             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12250                 mtx_unlock(&softc->ctl_lock);
12251                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12252                 return (1);
12253         }
12254         mtx_lock(&lun->lun_lock);
12255         mtx_unlock(&softc->ctl_lock);
12256         for (xio = (union ctl_io *)TAILQ_FIRST(&lun->ooa_queue); xio != NULL;
12257              xio = (union ctl_io *)TAILQ_NEXT(&xio->io_hdr, ooa_links)) {
12258
12259                 if ((xio->io_hdr.nexus.targ_port != io->io_hdr.nexus.targ_port)
12260                  || (xio->io_hdr.nexus.initid != io->io_hdr.nexus.initid)
12261                  || (xio->io_hdr.flags & CTL_FLAG_ABORT))
12262                         continue;
12263
12264                 if (task_set || xio->scsiio.tag_num == io->taskio.tag_num) {
12265                         found = 1;
12266                         break;
12267                 }
12268         }
12269         mtx_unlock(&lun->lun_lock);
12270         if (found)
12271                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12272         else
12273                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12274         return (0);
12275 }
12276
12277 static int
12278 ctl_query_async_event(union ctl_io *io)
12279 {
12280         struct ctl_lun *lun;
12281         struct ctl_softc *softc;
12282         ctl_ua_type ua;
12283         uint32_t targ_lun, initidx;
12284
12285         softc = control_softc;
12286         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12287         mtx_lock(&softc->ctl_lock);
12288         if ((targ_lun >= CTL_MAX_LUNS) ||
12289             (lun = softc->ctl_luns[targ_lun]) == NULL) {
12290                 mtx_unlock(&softc->ctl_lock);
12291                 io->taskio.task_status = CTL_TASK_LUN_DOES_NOT_EXIST;
12292                 return (1);
12293         }
12294         mtx_lock(&lun->lun_lock);
12295         mtx_unlock(&softc->ctl_lock);
12296         initidx = ctl_get_initindex(&io->io_hdr.nexus);
12297         ua = ctl_build_qae(lun, initidx, io->taskio.task_resp);
12298         mtx_unlock(&lun->lun_lock);
12299         if (ua != CTL_UA_NONE)
12300                 io->taskio.task_status = CTL_TASK_FUNCTION_SUCCEEDED;
12301         else
12302                 io->taskio.task_status = CTL_TASK_FUNCTION_COMPLETE;
12303         return (0);
12304 }
12305
12306 static void
12307 ctl_run_task(union ctl_io *io)
12308 {
12309         struct ctl_softc *softc = control_softc;
12310         int retval = 1;
12311
12312         CTL_DEBUG_PRINT(("ctl_run_task\n"));
12313         KASSERT(io->io_hdr.io_type == CTL_IO_TASK,
12314             ("ctl_run_task: Unextected io_type %d\n", io->io_hdr.io_type));
12315         io->taskio.task_status = CTL_TASK_FUNCTION_NOT_SUPPORTED;
12316         bzero(io->taskio.task_resp, sizeof(io->taskio.task_resp));
12317         switch (io->taskio.task_action) {
12318         case CTL_TASK_ABORT_TASK:
12319                 retval = ctl_abort_task(io);
12320                 break;
12321         case CTL_TASK_ABORT_TASK_SET:
12322         case CTL_TASK_CLEAR_TASK_SET:
12323                 retval = ctl_abort_task_set(io);
12324                 break;
12325         case CTL_TASK_CLEAR_ACA:
12326                 break;
12327         case CTL_TASK_I_T_NEXUS_RESET:
12328                 retval = ctl_i_t_nexus_reset(io);
12329                 break;
12330         case CTL_TASK_LUN_RESET:
12331                 retval = ctl_lun_reset(softc, io);
12332                 break;
12333         case CTL_TASK_TARGET_RESET:
12334                 retval = ctl_target_reset(softc, io, CTL_UA_TARG_RESET);
12335                 break;
12336         case CTL_TASK_BUS_RESET:
12337                 retval = ctl_bus_reset(softc, io);
12338                 break;
12339         case CTL_TASK_PORT_LOGIN:
12340                 break;
12341         case CTL_TASK_PORT_LOGOUT:
12342                 break;
12343         case CTL_TASK_QUERY_TASK:
12344                 retval = ctl_query_task(io, 0);
12345                 break;
12346         case CTL_TASK_QUERY_TASK_SET:
12347                 retval = ctl_query_task(io, 1);
12348                 break;
12349         case CTL_TASK_QUERY_ASYNC_EVENT:
12350                 retval = ctl_query_async_event(io);
12351                 break;
12352         default:
12353                 printf("%s: got unknown task management event %d\n",
12354                        __func__, io->taskio.task_action);
12355                 break;
12356         }
12357         if (retval == 0)
12358                 io->io_hdr.status = CTL_SUCCESS;
12359         else
12360                 io->io_hdr.status = CTL_ERROR;
12361         ctl_done(io);
12362 }
12363
12364 /*
12365  * For HA operation.  Handle commands that come in from the other
12366  * controller.
12367  */
12368 static void
12369 ctl_handle_isc(union ctl_io *io)
12370 {
12371         int free_io;
12372         struct ctl_lun *lun;
12373         struct ctl_softc *softc = control_softc;
12374         uint32_t targ_lun;
12375
12376         targ_lun = io->io_hdr.nexus.targ_mapped_lun;
12377         lun = softc->ctl_luns[targ_lun];
12378
12379         switch (io->io_hdr.msg_type) {
12380         case CTL_MSG_SERIALIZE:
12381                 free_io = ctl_serialize_other_sc_cmd(&io->scsiio);
12382                 break;
12383         case CTL_MSG_R2R: {
12384                 const struct ctl_cmd_entry *entry;
12385
12386                 /*
12387                  * This is only used in SER_ONLY mode.
12388                  */
12389                 free_io = 0;
12390                 entry = ctl_get_cmd_entry(&io->scsiio, NULL);
12391                 mtx_lock(&lun->lun_lock);
12392                 if (ctl_scsiio_lun_check(lun,
12393                     entry, (struct ctl_scsiio *)io) != 0) {
12394                         mtx_unlock(&lun->lun_lock);
12395                         ctl_done(io);
12396                         break;
12397                 }
12398                 io->io_hdr.flags |= CTL_FLAG_IS_WAS_ON_RTR;
12399                 mtx_unlock(&lun->lun_lock);
12400                 ctl_enqueue_rtr(io);
12401                 break;
12402         }
12403         case CTL_MSG_FINISH_IO:
12404                 if (softc->ha_mode == CTL_HA_MODE_XFER) {
12405                         free_io = 0;
12406                         ctl_done(io);
12407                 } else {
12408                         free_io = 1;
12409                         mtx_lock(&lun->lun_lock);
12410                         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr,
12411                                      ooa_links);
12412                         ctl_check_blocked(lun);
12413                         mtx_unlock(&lun->lun_lock);
12414                 }
12415                 break;
12416         case CTL_MSG_PERS_ACTION:
12417                 ctl_hndl_per_res_out_on_other_sc(
12418                         (union ctl_ha_msg *)&io->presio.pr_msg);
12419                 free_io = 1;
12420                 break;
12421         case CTL_MSG_BAD_JUJU:
12422                 free_io = 0;
12423                 ctl_done(io);
12424                 break;
12425         case CTL_MSG_DATAMOVE:
12426                 /* Only used in XFER mode */
12427                 free_io = 0;
12428                 ctl_datamove_remote(io);
12429                 break;
12430         case CTL_MSG_DATAMOVE_DONE:
12431                 /* Only used in XFER mode */
12432                 free_io = 0;
12433                 io->scsiio.be_move_done(io);
12434                 break;
12435         case CTL_MSG_FAILOVER:
12436                 ctl_failover_lun(io);
12437                 free_io = 1;
12438                 break;
12439         default:
12440                 free_io = 1;
12441                 printf("%s: Invalid message type %d\n",
12442                        __func__, io->io_hdr.msg_type);
12443                 break;
12444         }
12445         if (free_io)
12446                 ctl_free_io(io);
12447
12448 }
12449
12450
12451 /*
12452  * Returns the match type in the case of a match, or CTL_LUN_PAT_NONE if
12453  * there is no match.
12454  */
12455 static ctl_lun_error_pattern
12456 ctl_cmd_pattern_match(struct ctl_scsiio *ctsio, struct ctl_error_desc *desc)
12457 {
12458         const struct ctl_cmd_entry *entry;
12459         ctl_lun_error_pattern filtered_pattern, pattern;
12460
12461         pattern = desc->error_pattern;
12462
12463         /*
12464          * XXX KDM we need more data passed into this function to match a
12465          * custom pattern, and we actually need to implement custom pattern
12466          * matching.
12467          */
12468         if (pattern & CTL_LUN_PAT_CMD)
12469                 return (CTL_LUN_PAT_CMD);
12470
12471         if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_ANY)
12472                 return (CTL_LUN_PAT_ANY);
12473
12474         entry = ctl_get_cmd_entry(ctsio, NULL);
12475
12476         filtered_pattern = entry->pattern & pattern;
12477
12478         /*
12479          * If the user requested specific flags in the pattern (e.g.
12480          * CTL_LUN_PAT_RANGE), make sure the command supports all of those
12481          * flags.
12482          *
12483          * If the user did not specify any flags, it doesn't matter whether
12484          * or not the command supports the flags.
12485          */
12486         if ((filtered_pattern & ~CTL_LUN_PAT_MASK) !=
12487              (pattern & ~CTL_LUN_PAT_MASK))
12488                 return (CTL_LUN_PAT_NONE);
12489
12490         /*
12491          * If the user asked for a range check, see if the requested LBA
12492          * range overlaps with this command's LBA range.
12493          */
12494         if (filtered_pattern & CTL_LUN_PAT_RANGE) {
12495                 uint64_t lba1;
12496                 uint64_t len1;
12497                 ctl_action action;
12498                 int retval;
12499
12500                 retval = ctl_get_lba_len((union ctl_io *)ctsio, &lba1, &len1);
12501                 if (retval != 0)
12502                         return (CTL_LUN_PAT_NONE);
12503
12504                 action = ctl_extent_check_lba(lba1, len1, desc->lba_range.lba,
12505                                               desc->lba_range.len, FALSE);
12506                 /*
12507                  * A "pass" means that the LBA ranges don't overlap, so
12508                  * this doesn't match the user's range criteria.
12509                  */
12510                 if (action == CTL_ACTION_PASS)
12511                         return (CTL_LUN_PAT_NONE);
12512         }
12513
12514         return (filtered_pattern);
12515 }
12516
12517 static void
12518 ctl_inject_error(struct ctl_lun *lun, union ctl_io *io)
12519 {
12520         struct ctl_error_desc *desc, *desc2;
12521
12522         mtx_assert(&lun->lun_lock, MA_OWNED);
12523
12524         STAILQ_FOREACH_SAFE(desc, &lun->error_list, links, desc2) {
12525                 ctl_lun_error_pattern pattern;
12526                 /*
12527                  * Check to see whether this particular command matches
12528                  * the pattern in the descriptor.
12529                  */
12530                 pattern = ctl_cmd_pattern_match(&io->scsiio, desc);
12531                 if ((pattern & CTL_LUN_PAT_MASK) == CTL_LUN_PAT_NONE)
12532                         continue;
12533
12534                 switch (desc->lun_error & CTL_LUN_INJ_TYPE) {
12535                 case CTL_LUN_INJ_ABORTED:
12536                         ctl_set_aborted(&io->scsiio);
12537                         break;
12538                 case CTL_LUN_INJ_MEDIUM_ERR:
12539                         ctl_set_medium_error(&io->scsiio,
12540                             (io->io_hdr.flags & CTL_FLAG_DATA_MASK) !=
12541                              CTL_FLAG_DATA_OUT);
12542                         break;
12543                 case CTL_LUN_INJ_UA:
12544                         /* 29h/00h  POWER ON, RESET, OR BUS DEVICE RESET
12545                          * OCCURRED */
12546                         ctl_set_ua(&io->scsiio, 0x29, 0x00);
12547                         break;
12548                 case CTL_LUN_INJ_CUSTOM:
12549                         /*
12550                          * We're assuming the user knows what he is doing.
12551                          * Just copy the sense information without doing
12552                          * checks.
12553                          */
12554                         bcopy(&desc->custom_sense, &io->scsiio.sense_data,
12555                               MIN(sizeof(desc->custom_sense),
12556                                   sizeof(io->scsiio.sense_data)));
12557                         io->scsiio.scsi_status = SCSI_STATUS_CHECK_COND;
12558                         io->scsiio.sense_len = SSD_FULL_SIZE;
12559                         io->io_hdr.status = CTL_SCSI_ERROR | CTL_AUTOSENSE;
12560                         break;
12561                 case CTL_LUN_INJ_NONE:
12562                 default:
12563                         /*
12564                          * If this is an error injection type we don't know
12565                          * about, clear the continuous flag (if it is set)
12566                          * so it will get deleted below.
12567                          */
12568                         desc->lun_error &= ~CTL_LUN_INJ_CONTINUOUS;
12569                         break;
12570                 }
12571                 /*
12572                  * By default, each error injection action is a one-shot
12573                  */
12574                 if (desc->lun_error & CTL_LUN_INJ_CONTINUOUS)
12575                         continue;
12576
12577                 STAILQ_REMOVE(&lun->error_list, desc, ctl_error_desc, links);
12578
12579                 free(desc, M_CTL);
12580         }
12581 }
12582
12583 #ifdef CTL_IO_DELAY
12584 static void
12585 ctl_datamove_timer_wakeup(void *arg)
12586 {
12587         union ctl_io *io;
12588
12589         io = (union ctl_io *)arg;
12590
12591         ctl_datamove(io);
12592 }
12593 #endif /* CTL_IO_DELAY */
12594
12595 void
12596 ctl_datamove(union ctl_io *io)
12597 {
12598         struct ctl_lun *lun;
12599         void (*fe_datamove)(union ctl_io *io);
12600
12601         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
12602
12603         CTL_DEBUG_PRINT(("ctl_datamove\n"));
12604
12605         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
12606 #ifdef CTL_TIME_IO
12607         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
12608                 char str[256];
12609                 char path_str[64];
12610                 struct sbuf sb;
12611
12612                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
12613                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12614
12615                 sbuf_cat(&sb, path_str);
12616                 switch (io->io_hdr.io_type) {
12617                 case CTL_IO_SCSI:
12618                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
12619                         sbuf_printf(&sb, "\n");
12620                         sbuf_cat(&sb, path_str);
12621                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12622                                     io->scsiio.tag_num, io->scsiio.tag_type);
12623                         break;
12624                 case CTL_IO_TASK:
12625                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
12626                                     "Tag Type: %d\n", io->taskio.task_action,
12627                                     io->taskio.tag_num, io->taskio.tag_type);
12628                         break;
12629                 default:
12630                         panic("%s: Invalid CTL I/O type %d\n",
12631                             __func__, io->io_hdr.io_type);
12632                 }
12633                 sbuf_cat(&sb, path_str);
12634                 sbuf_printf(&sb, "ctl_datamove: %jd seconds\n",
12635                             (intmax_t)time_uptime - io->io_hdr.start_time);
12636                 sbuf_finish(&sb);
12637                 printf("%s", sbuf_data(&sb));
12638         }
12639 #endif /* CTL_TIME_IO */
12640
12641 #ifdef CTL_IO_DELAY
12642         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
12643                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
12644         } else {
12645                 if ((lun != NULL)
12646                  && (lun->delay_info.datamove_delay > 0)) {
12647
12648                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
12649                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
12650                         callout_reset(&io->io_hdr.delay_callout,
12651                                       lun->delay_info.datamove_delay * hz,
12652                                       ctl_datamove_timer_wakeup, io);
12653                         if (lun->delay_info.datamove_type ==
12654                             CTL_DELAY_TYPE_ONESHOT)
12655                                 lun->delay_info.datamove_delay = 0;
12656                         return;
12657                 }
12658         }
12659 #endif
12660
12661         /*
12662          * This command has been aborted.  Set the port status, so we fail
12663          * the data move.
12664          */
12665         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
12666                 printf("ctl_datamove: tag 0x%04x on (%u:%u:%u) aborted\n",
12667                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
12668                        io->io_hdr.nexus.targ_port,
12669                        io->io_hdr.nexus.targ_lun);
12670                 io->io_hdr.port_status = 31337;
12671                 /*
12672                  * Note that the backend, in this case, will get the
12673                  * callback in its context.  In other cases it may get
12674                  * called in the frontend's interrupt thread context.
12675                  */
12676                 io->scsiio.be_move_done(io);
12677                 return;
12678         }
12679
12680         /* Don't confuse frontend with zero length data move. */
12681         if (io->scsiio.kern_data_len == 0) {
12682                 io->scsiio.be_move_done(io);
12683                 return;
12684         }
12685
12686         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12687         fe_datamove(io);
12688 }
12689
12690 static void
12691 ctl_send_datamove_done(union ctl_io *io, int have_lock)
12692 {
12693         union ctl_ha_msg msg;
12694 #ifdef CTL_TIME_IO
12695         struct bintime cur_bt;
12696 #endif
12697
12698         memset(&msg, 0, sizeof(msg));
12699         msg.hdr.msg_type = CTL_MSG_DATAMOVE_DONE;
12700         msg.hdr.original_sc = io;
12701         msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
12702         msg.hdr.nexus = io->io_hdr.nexus;
12703         msg.hdr.status = io->io_hdr.status;
12704         msg.scsi.tag_num = io->scsiio.tag_num;
12705         msg.scsi.tag_type = io->scsiio.tag_type;
12706         msg.scsi.scsi_status = io->scsiio.scsi_status;
12707         memcpy(&msg.scsi.sense_data, &io->scsiio.sense_data,
12708                io->scsiio.sense_len);
12709         msg.scsi.sense_len = io->scsiio.sense_len;
12710         msg.scsi.sense_residual = io->scsiio.sense_residual;
12711         msg.scsi.fetd_status = io->io_hdr.port_status;
12712         msg.scsi.residual = io->scsiio.residual;
12713         io->io_hdr.flags &= ~CTL_FLAG_IO_ACTIVE;
12714         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
12715                 ctl_failover_io(io, /*have_lock*/ have_lock);
12716                 return;
12717         }
12718         ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
12719             sizeof(msg.scsi) - sizeof(msg.scsi.sense_data) +
12720             msg.scsi.sense_len, M_WAITOK);
12721
12722 #ifdef CTL_TIME_IO
12723         getbinuptime(&cur_bt);
12724         bintime_sub(&cur_bt, &io->io_hdr.dma_start_bt);
12725         bintime_add(&io->io_hdr.dma_bt, &cur_bt);
12726 #endif
12727         io->io_hdr.num_dmas++;
12728 }
12729
12730 /*
12731  * The DMA to the remote side is done, now we need to tell the other side
12732  * we're done so it can continue with its data movement.
12733  */
12734 static void
12735 ctl_datamove_remote_write_cb(struct ctl_ha_dt_req *rq)
12736 {
12737         union ctl_io *io;
12738         uint32_t i;
12739
12740         io = rq->context;
12741
12742         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12743                 printf("%s: ISC DMA write failed with error %d", __func__,
12744                        rq->ret);
12745                 ctl_set_internal_failure(&io->scsiio,
12746                                          /*sks_valid*/ 1,
12747                                          /*retry_count*/ rq->ret);
12748         }
12749
12750         ctl_dt_req_free(rq);
12751
12752         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12753                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12754         free(io->io_hdr.remote_sglist, M_CTL);
12755         io->io_hdr.remote_sglist = NULL;
12756         io->io_hdr.local_sglist = NULL;
12757
12758         /*
12759          * The data is in local and remote memory, so now we need to send
12760          * status (good or back) back to the other side.
12761          */
12762         ctl_send_datamove_done(io, /*have_lock*/ 0);
12763 }
12764
12765 /*
12766  * We've moved the data from the host/controller into local memory.  Now we
12767  * need to push it over to the remote controller's memory.
12768  */
12769 static int
12770 ctl_datamove_remote_dm_write_cb(union ctl_io *io)
12771 {
12772         int retval;
12773
12774         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_WRITE,
12775                                           ctl_datamove_remote_write_cb);
12776         return (retval);
12777 }
12778
12779 static void
12780 ctl_datamove_remote_write(union ctl_io *io)
12781 {
12782         int retval;
12783         void (*fe_datamove)(union ctl_io *io);
12784
12785         /*
12786          * - Get the data from the host/HBA into local memory.
12787          * - DMA memory from the local controller to the remote controller.
12788          * - Send status back to the remote controller.
12789          */
12790
12791         retval = ctl_datamove_remote_sgl_setup(io);
12792         if (retval != 0)
12793                 return;
12794
12795         /* Switch the pointer over so the FETD knows what to do */
12796         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12797
12798         /*
12799          * Use a custom move done callback, since we need to send completion
12800          * back to the other controller, not to the backend on this side.
12801          */
12802         io->scsiio.be_move_done = ctl_datamove_remote_dm_write_cb;
12803
12804         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12805         fe_datamove(io);
12806 }
12807
12808 static int
12809 ctl_datamove_remote_dm_read_cb(union ctl_io *io)
12810 {
12811 #if 0
12812         char str[256];
12813         char path_str[64];
12814         struct sbuf sb;
12815 #endif
12816         uint32_t i;
12817
12818         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12819                 free(io->io_hdr.local_sglist[i].addr, M_CTL);
12820         free(io->io_hdr.remote_sglist, M_CTL);
12821         io->io_hdr.remote_sglist = NULL;
12822         io->io_hdr.local_sglist = NULL;
12823
12824 #if 0
12825         scsi_path_string(io, path_str, sizeof(path_str));
12826         sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
12827         sbuf_cat(&sb, path_str);
12828         scsi_command_string(&io->scsiio, NULL, &sb);
12829         sbuf_printf(&sb, "\n");
12830         sbuf_cat(&sb, path_str);
12831         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
12832                     io->scsiio.tag_num, io->scsiio.tag_type);
12833         sbuf_cat(&sb, path_str);
12834         sbuf_printf(&sb, "%s: flags %#x, status %#x\n", __func__,
12835                     io->io_hdr.flags, io->io_hdr.status);
12836         sbuf_finish(&sb);
12837         printk("%s", sbuf_data(&sb));
12838 #endif
12839
12840
12841         /*
12842          * The read is done, now we need to send status (good or bad) back
12843          * to the other side.
12844          */
12845         ctl_send_datamove_done(io, /*have_lock*/ 0);
12846
12847         return (0);
12848 }
12849
12850 static void
12851 ctl_datamove_remote_read_cb(struct ctl_ha_dt_req *rq)
12852 {
12853         union ctl_io *io;
12854         void (*fe_datamove)(union ctl_io *io);
12855
12856         io = rq->context;
12857
12858         if (rq->ret != CTL_HA_STATUS_SUCCESS) {
12859                 printf("%s: ISC DMA read failed with error %d\n", __func__,
12860                        rq->ret);
12861                 ctl_set_internal_failure(&io->scsiio,
12862                                          /*sks_valid*/ 1,
12863                                          /*retry_count*/ rq->ret);
12864         }
12865
12866         ctl_dt_req_free(rq);
12867
12868         /* Switch the pointer over so the FETD knows what to do */
12869         io->scsiio.kern_data_ptr = (uint8_t *)io->io_hdr.local_sglist;
12870
12871         /*
12872          * Use a custom move done callback, since we need to send completion
12873          * back to the other controller, not to the backend on this side.
12874          */
12875         io->scsiio.be_move_done = ctl_datamove_remote_dm_read_cb;
12876
12877         /* XXX KDM add checks like the ones in ctl_datamove? */
12878
12879         fe_datamove = ctl_io_port(&io->io_hdr)->fe_datamove;
12880         fe_datamove(io);
12881 }
12882
12883 static int
12884 ctl_datamove_remote_sgl_setup(union ctl_io *io)
12885 {
12886         struct ctl_sg_entry *local_sglist;
12887         uint32_t len_to_go;
12888         int retval;
12889         int i;
12890
12891         retval = 0;
12892         local_sglist = io->io_hdr.local_sglist;
12893         len_to_go = io->scsiio.kern_data_len;
12894
12895         /*
12896          * The difficult thing here is that the size of the various
12897          * S/G segments may be different than the size from the
12898          * remote controller.  That'll make it harder when DMAing
12899          * the data back to the other side.
12900          */
12901         for (i = 0; len_to_go > 0; i++) {
12902                 local_sglist[i].len = MIN(len_to_go, CTL_HA_DATAMOVE_SEGMENT);
12903                 local_sglist[i].addr =
12904                     malloc(local_sglist[i].len, M_CTL, M_WAITOK);
12905
12906                 len_to_go -= local_sglist[i].len;
12907         }
12908         /*
12909          * Reset the number of S/G entries accordingly.  The original
12910          * number of S/G entries is available in rem_sg_entries.
12911          */
12912         io->scsiio.kern_sg_entries = i;
12913
12914 #if 0
12915         printf("%s: kern_sg_entries = %d\n", __func__,
12916                io->scsiio.kern_sg_entries);
12917         for (i = 0; i < io->scsiio.kern_sg_entries; i++)
12918                 printf("%s: sg[%d] = %p, %lu\n", __func__, i,
12919                        local_sglist[i].addr, local_sglist[i].len);
12920 #endif
12921
12922         return (retval);
12923 }
12924
12925 static int
12926 ctl_datamove_remote_xfer(union ctl_io *io, unsigned command,
12927                          ctl_ha_dt_cb callback)
12928 {
12929         struct ctl_ha_dt_req *rq;
12930         struct ctl_sg_entry *remote_sglist, *local_sglist;
12931         uint32_t local_used, remote_used, total_used;
12932         int i, j, isc_ret;
12933
12934         rq = ctl_dt_req_alloc();
12935
12936         /*
12937          * If we failed to allocate the request, and if the DMA didn't fail
12938          * anyway, set busy status.  This is just a resource allocation
12939          * failure.
12940          */
12941         if ((rq == NULL)
12942          && ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12943              (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS))
12944                 ctl_set_busy(&io->scsiio);
12945
12946         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_STATUS_NONE &&
12947             (io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS) {
12948
12949                 if (rq != NULL)
12950                         ctl_dt_req_free(rq);
12951
12952                 /*
12953                  * The data move failed.  We need to return status back
12954                  * to the other controller.  No point in trying to DMA
12955                  * data to the remote controller.
12956                  */
12957
12958                 ctl_send_datamove_done(io, /*have_lock*/ 0);
12959
12960                 return (1);
12961         }
12962
12963         local_sglist = io->io_hdr.local_sglist;
12964         remote_sglist = io->io_hdr.remote_sglist;
12965         local_used = 0;
12966         remote_used = 0;
12967         total_used = 0;
12968
12969         /*
12970          * Pull/push the data over the wire from/to the other controller.
12971          * This takes into account the possibility that the local and
12972          * remote sglists may not be identical in terms of the size of
12973          * the elements and the number of elements.
12974          *
12975          * One fundamental assumption here is that the length allocated for
12976          * both the local and remote sglists is identical.  Otherwise, we've
12977          * essentially got a coding error of some sort.
12978          */
12979         isc_ret = CTL_HA_STATUS_SUCCESS;
12980         for (i = 0, j = 0; total_used < io->scsiio.kern_data_len; ) {
12981                 uint32_t cur_len;
12982                 uint8_t *tmp_ptr;
12983
12984                 rq->command = command;
12985                 rq->context = io;
12986
12987                 /*
12988                  * Both pointers should be aligned.  But it is possible
12989                  * that the allocation length is not.  They should both
12990                  * also have enough slack left over at the end, though,
12991                  * to round up to the next 8 byte boundary.
12992                  */
12993                 cur_len = MIN(local_sglist[i].len - local_used,
12994                               remote_sglist[j].len - remote_used);
12995                 rq->size = cur_len;
12996
12997                 tmp_ptr = (uint8_t *)local_sglist[i].addr;
12998                 tmp_ptr += local_used;
12999
13000 #if 0
13001                 /* Use physical addresses when talking to ISC hardware */
13002                 if ((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0) {
13003                         /* XXX KDM use busdma */
13004                         rq->local = vtophys(tmp_ptr);
13005                 } else
13006                         rq->local = tmp_ptr;
13007 #else
13008                 KASSERT((io->io_hdr.flags & CTL_FLAG_BUS_ADDR) == 0,
13009                     ("HA does not support BUS_ADDR"));
13010                 rq->local = tmp_ptr;
13011 #endif
13012
13013                 tmp_ptr = (uint8_t *)remote_sglist[j].addr;
13014                 tmp_ptr += remote_used;
13015                 rq->remote = tmp_ptr;
13016
13017                 rq->callback = NULL;
13018
13019                 local_used += cur_len;
13020                 if (local_used >= local_sglist[i].len) {
13021                         i++;
13022                         local_used = 0;
13023                 }
13024
13025                 remote_used += cur_len;
13026                 if (remote_used >= remote_sglist[j].len) {
13027                         j++;
13028                         remote_used = 0;
13029                 }
13030                 total_used += cur_len;
13031
13032                 if (total_used >= io->scsiio.kern_data_len)
13033                         rq->callback = callback;
13034
13035 #if 0
13036                 printf("%s: %s: local %p remote %p size %d\n", __func__,
13037                        (command == CTL_HA_DT_CMD_WRITE) ? "WRITE" : "READ",
13038                        rq->local, rq->remote, rq->size);
13039 #endif
13040
13041                 isc_ret = ctl_dt_single(rq);
13042                 if (isc_ret > CTL_HA_STATUS_SUCCESS)
13043                         break;
13044         }
13045         if (isc_ret != CTL_HA_STATUS_WAIT) {
13046                 rq->ret = isc_ret;
13047                 callback(rq);
13048         }
13049
13050         return (0);
13051 }
13052
13053 static void
13054 ctl_datamove_remote_read(union ctl_io *io)
13055 {
13056         int retval;
13057         uint32_t i;
13058
13059         /*
13060          * This will send an error to the other controller in the case of a
13061          * failure.
13062          */
13063         retval = ctl_datamove_remote_sgl_setup(io);
13064         if (retval != 0)
13065                 return;
13066
13067         retval = ctl_datamove_remote_xfer(io, CTL_HA_DT_CMD_READ,
13068                                           ctl_datamove_remote_read_cb);
13069         if (retval != 0) {
13070                 /*
13071                  * Make sure we free memory if there was an error..  The
13072                  * ctl_datamove_remote_xfer() function will send the
13073                  * datamove done message, or call the callback with an
13074                  * error if there is a problem.
13075                  */
13076                 for (i = 0; i < io->scsiio.kern_sg_entries; i++)
13077                         free(io->io_hdr.local_sglist[i].addr, M_CTL);
13078                 free(io->io_hdr.remote_sglist, M_CTL);
13079                 io->io_hdr.remote_sglist = NULL;
13080                 io->io_hdr.local_sglist = NULL;
13081         }
13082 }
13083
13084 /*
13085  * Process a datamove request from the other controller.  This is used for
13086  * XFER mode only, not SER_ONLY mode.  For writes, we DMA into local memory
13087  * first.  Once that is complete, the data gets DMAed into the remote
13088  * controller's memory.  For reads, we DMA from the remote controller's
13089  * memory into our memory first, and then move it out to the FETD.
13090  */
13091 static void
13092 ctl_datamove_remote(union ctl_io *io)
13093 {
13094
13095         mtx_assert(&control_softc->ctl_lock, MA_NOTOWNED);
13096
13097         if (io->io_hdr.flags & CTL_FLAG_FAILOVER) {
13098                 ctl_failover_io(io, /*have_lock*/ 0);
13099                 return;
13100         }
13101
13102         /*
13103          * Note that we look for an aborted I/O here, but don't do some of
13104          * the other checks that ctl_datamove() normally does.
13105          * We don't need to run the datamove delay code, since that should
13106          * have been done if need be on the other controller.
13107          */
13108         if (io->io_hdr.flags & CTL_FLAG_ABORT) {
13109                 printf("%s: tag 0x%04x on (%u:%u:%u) aborted\n", __func__,
13110                        io->scsiio.tag_num, io->io_hdr.nexus.initid,
13111                        io->io_hdr.nexus.targ_port,
13112                        io->io_hdr.nexus.targ_lun);
13113                 io->io_hdr.port_status = 31338;
13114                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13115                 return;
13116         }
13117
13118         if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_OUT)
13119                 ctl_datamove_remote_write(io);
13120         else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) == CTL_FLAG_DATA_IN)
13121                 ctl_datamove_remote_read(io);
13122         else {
13123                 io->io_hdr.port_status = 31339;
13124                 ctl_send_datamove_done(io, /*have_lock*/ 0);
13125         }
13126 }
13127
13128 static void
13129 ctl_process_done(union ctl_io *io)
13130 {
13131         struct ctl_lun *lun;
13132         struct ctl_softc *softc = control_softc;
13133         void (*fe_done)(union ctl_io *io);
13134         union ctl_ha_msg msg;
13135         uint32_t targ_port = io->io_hdr.nexus.targ_port;
13136
13137         CTL_DEBUG_PRINT(("ctl_process_done\n"));
13138         fe_done = softc->ctl_ports[targ_port]->fe_done;
13139
13140 #ifdef CTL_TIME_IO
13141         if ((time_uptime - io->io_hdr.start_time) > ctl_time_io_secs) {
13142                 char str[256];
13143                 char path_str[64];
13144                 struct sbuf sb;
13145
13146                 ctl_scsi_path_string(io, path_str, sizeof(path_str));
13147                 sbuf_new(&sb, str, sizeof(str), SBUF_FIXEDLEN);
13148
13149                 sbuf_cat(&sb, path_str);
13150                 switch (io->io_hdr.io_type) {
13151                 case CTL_IO_SCSI:
13152                         ctl_scsi_command_string(&io->scsiio, NULL, &sb);
13153                         sbuf_printf(&sb, "\n");
13154                         sbuf_cat(&sb, path_str);
13155                         sbuf_printf(&sb, "Tag: 0x%04x, type %d\n",
13156                                     io->scsiio.tag_num, io->scsiio.tag_type);
13157                         break;
13158                 case CTL_IO_TASK:
13159                         sbuf_printf(&sb, "Task I/O type: %d, Tag: 0x%04x, "
13160                                     "Tag Type: %d\n", io->taskio.task_action,
13161                                     io->taskio.tag_num, io->taskio.tag_type);
13162                         break;
13163                 default:
13164                         panic("%s: Invalid CTL I/O type %d\n",
13165                             __func__, io->io_hdr.io_type);
13166                 }
13167                 sbuf_cat(&sb, path_str);
13168                 sbuf_printf(&sb, "ctl_process_done: %jd seconds\n",
13169                             (intmax_t)time_uptime - io->io_hdr.start_time);
13170                 sbuf_finish(&sb);
13171                 printf("%s", sbuf_data(&sb));
13172         }
13173 #endif /* CTL_TIME_IO */
13174
13175         switch (io->io_hdr.io_type) {
13176         case CTL_IO_SCSI:
13177                 break;
13178         case CTL_IO_TASK:
13179                 if (ctl_debug & CTL_DEBUG_INFO)
13180                         ctl_io_error_print(io, NULL);
13181                 fe_done(io);
13182                 return;
13183         default:
13184                 panic("%s: Invalid CTL I/O type %d\n",
13185                     __func__, io->io_hdr.io_type);
13186         }
13187
13188         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13189         if (lun == NULL) {
13190                 CTL_DEBUG_PRINT(("NULL LUN for lun %d\n",
13191                                  io->io_hdr.nexus.targ_mapped_lun));
13192                 goto bailout;
13193         }
13194
13195         mtx_lock(&lun->lun_lock);
13196
13197         /*
13198          * Check to see if we have any informational exception and status
13199          * of this command can be modified to report it in form of either
13200          * RECOVERED ERROR or NO SENSE, depending on MRIE mode page field.
13201          */
13202         if (lun->ie_reported == 0 && lun->ie_asc != 0 &&
13203             io->io_hdr.status == CTL_SUCCESS &&
13204             (io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0) {
13205                 uint8_t mrie = lun->mode_pages.ie_page[CTL_PAGE_CURRENT].mrie;
13206                 uint8_t per =
13207                     ((lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT].byte3 &
13208                       SMS_RWER_PER) ||
13209                      (lun->mode_pages.verify_er_page[CTL_PAGE_CURRENT].byte3 &
13210                       SMS_VER_PER));
13211                 if (((mrie == SIEP_MRIE_REC_COND && per) ||
13212                      mrie == SIEP_MRIE_REC_UNCOND ||
13213                      mrie == SIEP_MRIE_NO_SENSE) &&
13214                     (ctl_get_cmd_entry(&io->scsiio, NULL)->flags &
13215                      CTL_CMD_FLAG_NO_SENSE) == 0) {
13216                         ctl_set_sense(&io->scsiio,
13217                               /*current_error*/ 1,
13218                               /*sense_key*/ (mrie == SIEP_MRIE_NO_SENSE) ?
13219                                 SSD_KEY_NO_SENSE : SSD_KEY_RECOVERED_ERROR,
13220                               /*asc*/ lun->ie_asc,
13221                               /*ascq*/ lun->ie_ascq,
13222                               SSD_ELEM_NONE);
13223                         lun->ie_reported = 1;
13224                 }
13225         } else if (lun->ie_reported < 0)
13226                 lun->ie_reported = 0;
13227
13228         /*
13229          * Check to see if we have any errors to inject here.  We only
13230          * inject errors for commands that don't already have errors set.
13231          */
13232         if (!STAILQ_EMPTY(&lun->error_list) &&
13233             ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS) &&
13234             ((io->io_hdr.flags & CTL_FLAG_STATUS_SENT) == 0))
13235                 ctl_inject_error(lun, io);
13236
13237         /*
13238          * XXX KDM how do we treat commands that aren't completed
13239          * successfully?
13240          *
13241          * XXX KDM should we also track I/O latency?
13242          */
13243         if ((io->io_hdr.status & CTL_STATUS_MASK) == CTL_SUCCESS &&
13244             io->io_hdr.io_type == CTL_IO_SCSI) {
13245 #ifdef CTL_TIME_IO
13246                 struct bintime cur_bt;
13247 #endif
13248                 int type;
13249
13250                 if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13251                     CTL_FLAG_DATA_IN)
13252                         type = CTL_STATS_READ;
13253                 else if ((io->io_hdr.flags & CTL_FLAG_DATA_MASK) ==
13254                     CTL_FLAG_DATA_OUT)
13255                         type = CTL_STATS_WRITE;
13256                 else
13257                         type = CTL_STATS_NO_IO;
13258
13259                 lun->stats.ports[targ_port].bytes[type] +=
13260                     io->scsiio.kern_total_len;
13261                 lun->stats.ports[targ_port].operations[type]++;
13262 #ifdef CTL_TIME_IO
13263                 bintime_add(&lun->stats.ports[targ_port].dma_time[type],
13264                    &io->io_hdr.dma_bt);
13265                 getbinuptime(&cur_bt);
13266                 bintime_sub(&cur_bt, &io->io_hdr.start_bt);
13267                 bintime_add(&lun->stats.ports[targ_port].time[type], &cur_bt);
13268 #endif
13269                 lun->stats.ports[targ_port].num_dmas[type] +=
13270                     io->io_hdr.num_dmas;
13271         }
13272
13273         /*
13274          * Remove this from the OOA queue.
13275          */
13276         TAILQ_REMOVE(&lun->ooa_queue, &io->io_hdr, ooa_links);
13277 #ifdef CTL_TIME_IO
13278         if (TAILQ_EMPTY(&lun->ooa_queue))
13279                 lun->last_busy = getsbinuptime();
13280 #endif
13281
13282         /*
13283          * Run through the blocked queue on this LUN and see if anything
13284          * has become unblocked, now that this transaction is done.
13285          */
13286         ctl_check_blocked(lun);
13287
13288         /*
13289          * If the LUN has been invalidated, free it if there is nothing
13290          * left on its OOA queue.
13291          */
13292         if ((lun->flags & CTL_LUN_INVALID)
13293          && TAILQ_EMPTY(&lun->ooa_queue)) {
13294                 mtx_unlock(&lun->lun_lock);
13295                 mtx_lock(&softc->ctl_lock);
13296                 ctl_free_lun(lun);
13297                 mtx_unlock(&softc->ctl_lock);
13298         } else
13299                 mtx_unlock(&lun->lun_lock);
13300
13301 bailout:
13302
13303         /*
13304          * If this command has been aborted, make sure we set the status
13305          * properly.  The FETD is responsible for freeing the I/O and doing
13306          * whatever it needs to do to clean up its state.
13307          */
13308         if (io->io_hdr.flags & CTL_FLAG_ABORT)
13309                 ctl_set_task_aborted(&io->scsiio);
13310
13311         /*
13312          * If enabled, print command error status.
13313          */
13314         if ((io->io_hdr.status & CTL_STATUS_MASK) != CTL_SUCCESS &&
13315             (ctl_debug & CTL_DEBUG_INFO) != 0)
13316                 ctl_io_error_print(io, NULL);
13317
13318         /*
13319          * Tell the FETD or the other shelf controller we're done with this
13320          * command.  Note that only SCSI commands get to this point.  Task
13321          * management commands are completed above.
13322          */
13323         if ((softc->ha_mode != CTL_HA_MODE_XFER) &&
13324             (io->io_hdr.flags & CTL_FLAG_SENT_2OTHER_SC)) {
13325                 memset(&msg, 0, sizeof(msg));
13326                 msg.hdr.msg_type = CTL_MSG_FINISH_IO;
13327                 msg.hdr.serializing_sc = io->io_hdr.serializing_sc;
13328                 msg.hdr.nexus = io->io_hdr.nexus;
13329                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13330                     sizeof(msg.scsi) - sizeof(msg.scsi.sense_data),
13331                     M_WAITOK);
13332         }
13333
13334         fe_done(io);
13335 }
13336
13337 #ifdef CTL_WITH_CA
13338 /*
13339  * Front end should call this if it doesn't do autosense.  When the request
13340  * sense comes back in from the initiator, we'll dequeue this and send it.
13341  */
13342 int
13343 ctl_queue_sense(union ctl_io *io)
13344 {
13345         struct ctl_lun *lun;
13346         struct ctl_port *port;
13347         struct ctl_softc *softc;
13348         uint32_t initidx, targ_lun;
13349
13350         softc = control_softc;
13351
13352         CTL_DEBUG_PRINT(("ctl_queue_sense\n"));
13353
13354         /*
13355          * LUN lookup will likely move to the ctl_work_thread() once we
13356          * have our new queueing infrastructure (that doesn't put things on
13357          * a per-LUN queue initially).  That is so that we can handle
13358          * things like an INQUIRY to a LUN that we don't have enabled.  We
13359          * can't deal with that right now.
13360          */
13361         mtx_lock(&softc->ctl_lock);
13362
13363         /*
13364          * If we don't have a LUN for this, just toss the sense
13365          * information.
13366          */
13367         port = ctl_io_port(&ctsio->io_hdr);
13368         targ_lun = ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13369         if ((targ_lun < CTL_MAX_LUNS)
13370          && (softc->ctl_luns[targ_lun] != NULL))
13371                 lun = softc->ctl_luns[targ_lun];
13372         else
13373                 goto bailout;
13374
13375         initidx = ctl_get_initindex(&io->io_hdr.nexus);
13376
13377         mtx_lock(&lun->lun_lock);
13378         /*
13379          * Already have CA set for this LUN...toss the sense information.
13380          */
13381         if (ctl_is_set(lun->have_ca, initidx)) {
13382                 mtx_unlock(&lun->lun_lock);
13383                 goto bailout;
13384         }
13385
13386         memcpy(&lun->pending_sense[initidx], &io->scsiio.sense_data,
13387                MIN(sizeof(lun->pending_sense[initidx]),
13388                sizeof(io->scsiio.sense_data)));
13389         ctl_set_mask(lun->have_ca, initidx);
13390         mtx_unlock(&lun->lun_lock);
13391
13392 bailout:
13393         mtx_unlock(&softc->ctl_lock);
13394
13395         ctl_free_io(io);
13396
13397         return (CTL_RETVAL_COMPLETE);
13398 }
13399 #endif
13400
13401 /*
13402  * Primary command inlet from frontend ports.  All SCSI and task I/O
13403  * requests must go through this function.
13404  */
13405 int
13406 ctl_queue(union ctl_io *io)
13407 {
13408         struct ctl_port *port;
13409
13410         CTL_DEBUG_PRINT(("ctl_queue cdb[0]=%02X\n", io->scsiio.cdb[0]));
13411
13412 #ifdef CTL_TIME_IO
13413         io->io_hdr.start_time = time_uptime;
13414         getbinuptime(&io->io_hdr.start_bt);
13415 #endif /* CTL_TIME_IO */
13416
13417         /* Map FE-specific LUN ID into global one. */
13418         port = ctl_io_port(&io->io_hdr);
13419         io->io_hdr.nexus.targ_mapped_lun =
13420             ctl_lun_map_from_port(port, io->io_hdr.nexus.targ_lun);
13421
13422         switch (io->io_hdr.io_type) {
13423         case CTL_IO_SCSI:
13424         case CTL_IO_TASK:
13425                 if (ctl_debug & CTL_DEBUG_CDB)
13426                         ctl_io_print(io);
13427                 ctl_enqueue_incoming(io);
13428                 break;
13429         default:
13430                 printf("ctl_queue: unknown I/O type %d\n", io->io_hdr.io_type);
13431                 return (EINVAL);
13432         }
13433
13434         return (CTL_RETVAL_COMPLETE);
13435 }
13436
13437 #ifdef CTL_IO_DELAY
13438 static void
13439 ctl_done_timer_wakeup(void *arg)
13440 {
13441         union ctl_io *io;
13442
13443         io = (union ctl_io *)arg;
13444         ctl_done(io);
13445 }
13446 #endif /* CTL_IO_DELAY */
13447
13448 void
13449 ctl_serseq_done(union ctl_io *io)
13450 {
13451         struct ctl_lun *lun;
13452
13453         lun = (struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13454         if (lun->be_lun == NULL ||
13455             lun->be_lun->serseq == CTL_LUN_SERSEQ_OFF)
13456                 return;
13457         mtx_lock(&lun->lun_lock);
13458         io->io_hdr.flags |= CTL_FLAG_SERSEQ_DONE;
13459         ctl_check_blocked(lun);
13460         mtx_unlock(&lun->lun_lock);
13461 }
13462
13463 void
13464 ctl_done(union ctl_io *io)
13465 {
13466
13467         /*
13468          * Enable this to catch duplicate completion issues.
13469          */
13470 #if 0
13471         if (io->io_hdr.flags & CTL_FLAG_ALREADY_DONE) {
13472                 printf("%s: type %d msg %d cdb %x iptl: "
13473                        "%u:%u:%u tag 0x%04x "
13474                        "flag %#x status %x\n",
13475                         __func__,
13476                         io->io_hdr.io_type,
13477                         io->io_hdr.msg_type,
13478                         io->scsiio.cdb[0],
13479                         io->io_hdr.nexus.initid,
13480                         io->io_hdr.nexus.targ_port,
13481                         io->io_hdr.nexus.targ_lun,
13482                         (io->io_hdr.io_type ==
13483                         CTL_IO_TASK) ?
13484                         io->taskio.tag_num :
13485                         io->scsiio.tag_num,
13486                         io->io_hdr.flags,
13487                         io->io_hdr.status);
13488         } else
13489                 io->io_hdr.flags |= CTL_FLAG_ALREADY_DONE;
13490 #endif
13491
13492         /*
13493          * This is an internal copy of an I/O, and should not go through
13494          * the normal done processing logic.
13495          */
13496         if (io->io_hdr.flags & CTL_FLAG_INT_COPY)
13497                 return;
13498
13499 #ifdef CTL_IO_DELAY
13500         if (io->io_hdr.flags & CTL_FLAG_DELAY_DONE) {
13501                 io->io_hdr.flags &= ~CTL_FLAG_DELAY_DONE;
13502         } else {
13503                 struct ctl_lun *lun;
13504
13505                 lun =(struct ctl_lun *)io->io_hdr.ctl_private[CTL_PRIV_LUN].ptr;
13506
13507                 if ((lun != NULL)
13508                  && (lun->delay_info.done_delay > 0)) {
13509
13510                         callout_init(&io->io_hdr.delay_callout, /*mpsafe*/ 1);
13511                         io->io_hdr.flags |= CTL_FLAG_DELAY_DONE;
13512                         callout_reset(&io->io_hdr.delay_callout,
13513                                       lun->delay_info.done_delay * hz,
13514                                       ctl_done_timer_wakeup, io);
13515                         if (lun->delay_info.done_type == CTL_DELAY_TYPE_ONESHOT)
13516                                 lun->delay_info.done_delay = 0;
13517                         return;
13518                 }
13519         }
13520 #endif /* CTL_IO_DELAY */
13521
13522         ctl_enqueue_done(io);
13523 }
13524
13525 static void
13526 ctl_work_thread(void *arg)
13527 {
13528         struct ctl_thread *thr = (struct ctl_thread *)arg;
13529         struct ctl_softc *softc = thr->ctl_softc;
13530         union ctl_io *io;
13531         int retval;
13532
13533         CTL_DEBUG_PRINT(("ctl_work_thread starting\n"));
13534
13535         for (;;) {
13536                 /*
13537                  * We handle the queues in this order:
13538                  * - ISC
13539                  * - done queue (to free up resources, unblock other commands)
13540                  * - RtR queue
13541                  * - incoming queue
13542                  *
13543                  * If those queues are empty, we break out of the loop and
13544                  * go to sleep.
13545                  */
13546                 mtx_lock(&thr->queue_lock);
13547                 io = (union ctl_io *)STAILQ_FIRST(&thr->isc_queue);
13548                 if (io != NULL) {
13549                         STAILQ_REMOVE_HEAD(&thr->isc_queue, links);
13550                         mtx_unlock(&thr->queue_lock);
13551                         ctl_handle_isc(io);
13552                         continue;
13553                 }
13554                 io = (union ctl_io *)STAILQ_FIRST(&thr->done_queue);
13555                 if (io != NULL) {
13556                         STAILQ_REMOVE_HEAD(&thr->done_queue, links);
13557                         /* clear any blocked commands, call fe_done */
13558                         mtx_unlock(&thr->queue_lock);
13559                         ctl_process_done(io);
13560                         continue;
13561                 }
13562                 io = (union ctl_io *)STAILQ_FIRST(&thr->incoming_queue);
13563                 if (io != NULL) {
13564                         STAILQ_REMOVE_HEAD(&thr->incoming_queue, links);
13565                         mtx_unlock(&thr->queue_lock);
13566                         if (io->io_hdr.io_type == CTL_IO_TASK)
13567                                 ctl_run_task(io);
13568                         else
13569                                 ctl_scsiio_precheck(softc, &io->scsiio);
13570                         continue;
13571                 }
13572                 io = (union ctl_io *)STAILQ_FIRST(&thr->rtr_queue);
13573                 if (io != NULL) {
13574                         STAILQ_REMOVE_HEAD(&thr->rtr_queue, links);
13575                         mtx_unlock(&thr->queue_lock);
13576                         retval = ctl_scsiio(&io->scsiio);
13577                         if (retval != CTL_RETVAL_COMPLETE)
13578                                 CTL_DEBUG_PRINT(("ctl_scsiio failed\n"));
13579                         continue;
13580                 }
13581
13582                 /* Sleep until we have something to do. */
13583                 mtx_sleep(thr, &thr->queue_lock, PDROP | PRIBIO, "-", 0);
13584         }
13585 }
13586
13587 static void
13588 ctl_lun_thread(void *arg)
13589 {
13590         struct ctl_softc *softc = (struct ctl_softc *)arg;
13591         struct ctl_be_lun *be_lun;
13592
13593         CTL_DEBUG_PRINT(("ctl_lun_thread starting\n"));
13594
13595         for (;;) {
13596                 mtx_lock(&softc->ctl_lock);
13597                 be_lun = STAILQ_FIRST(&softc->pending_lun_queue);
13598                 if (be_lun != NULL) {
13599                         STAILQ_REMOVE_HEAD(&softc->pending_lun_queue, links);
13600                         mtx_unlock(&softc->ctl_lock);
13601                         ctl_create_lun(be_lun);
13602                         continue;
13603                 }
13604
13605                 /* Sleep until we have something to do. */
13606                 mtx_sleep(&softc->pending_lun_queue, &softc->ctl_lock,
13607                     PDROP | PRIBIO, "-", 0);
13608         }
13609 }
13610
13611 static void
13612 ctl_thresh_thread(void *arg)
13613 {
13614         struct ctl_softc *softc = (struct ctl_softc *)arg;
13615         struct ctl_lun *lun;
13616         struct scsi_da_rw_recovery_page *rwpage;
13617         struct ctl_logical_block_provisioning_page *page;
13618         const char *attr;
13619         union ctl_ha_msg msg;
13620         uint64_t thres, val;
13621         int i, e, set;
13622
13623         CTL_DEBUG_PRINT(("ctl_thresh_thread starting\n"));
13624
13625         for (;;) {
13626                 mtx_lock(&softc->ctl_lock);
13627                 STAILQ_FOREACH(lun, &softc->lun_list, links) {
13628                         if ((lun->flags & CTL_LUN_DISABLED) ||
13629                             (lun->flags & CTL_LUN_NO_MEDIA) ||
13630                             lun->backend->lun_attr == NULL)
13631                                 continue;
13632                         if ((lun->flags & CTL_LUN_PRIMARY_SC) == 0 &&
13633                             softc->ha_mode == CTL_HA_MODE_XFER)
13634                                 continue;
13635                         rwpage = &lun->mode_pages.rw_er_page[CTL_PAGE_CURRENT];
13636                         if ((rwpage->byte8 & SMS_RWER_LBPERE) == 0)
13637                                 continue;
13638                         e = 0;
13639                         page = &lun->mode_pages.lbp_page[CTL_PAGE_CURRENT];
13640                         for (i = 0; i < CTL_NUM_LBP_THRESH; i++) {
13641                                 if ((page->descr[i].flags & SLBPPD_ENABLED) == 0)
13642                                         continue;
13643                                 thres = scsi_4btoul(page->descr[i].count);
13644                                 thres <<= CTL_LBP_EXPONENT;
13645                                 switch (page->descr[i].resource) {
13646                                 case 0x01:
13647                                         attr = "blocksavail";
13648                                         break;
13649                                 case 0x02:
13650                                         attr = "blocksused";
13651                                         break;
13652                                 case 0xf1:
13653                                         attr = "poolblocksavail";
13654                                         break;
13655                                 case 0xf2:
13656                                         attr = "poolblocksused";
13657                                         break;
13658                                 default:
13659                                         continue;
13660                                 }
13661                                 mtx_unlock(&softc->ctl_lock); // XXX
13662                                 val = lun->backend->lun_attr(
13663                                     lun->be_lun->be_lun, attr);
13664                                 mtx_lock(&softc->ctl_lock);
13665                                 if (val == UINT64_MAX)
13666                                         continue;
13667                                 if ((page->descr[i].flags & SLBPPD_ARMING_MASK)
13668                                     == SLBPPD_ARMING_INC)
13669                                         e = (val >= thres);
13670                                 else
13671                                         e = (val <= thres);
13672                                 if (e)
13673                                         break;
13674                         }
13675                         mtx_lock(&lun->lun_lock);
13676                         if (e) {
13677                                 scsi_u64to8b((uint8_t *)&page->descr[i] -
13678                                     (uint8_t *)page, lun->ua_tpt_info);
13679                                 if (lun->lasttpt == 0 ||
13680                                     time_uptime - lun->lasttpt >= CTL_LBP_UA_PERIOD) {
13681                                         lun->lasttpt = time_uptime;
13682                                         ctl_est_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13683                                         set = 1;
13684                                 } else
13685                                         set = 0;
13686                         } else {
13687                                 lun->lasttpt = 0;
13688                                 ctl_clr_ua_all(lun, -1, CTL_UA_THIN_PROV_THRES);
13689                                 set = -1;
13690                         }
13691                         mtx_unlock(&lun->lun_lock);
13692                         if (set != 0 &&
13693                             lun->ctl_softc->ha_mode == CTL_HA_MODE_XFER) {
13694                                 /* Send msg to other side. */
13695                                 bzero(&msg.ua, sizeof(msg.ua));
13696                                 msg.hdr.msg_type = CTL_MSG_UA;
13697                                 msg.hdr.nexus.initid = -1;
13698                                 msg.hdr.nexus.targ_port = -1;
13699                                 msg.hdr.nexus.targ_lun = lun->lun;
13700                                 msg.hdr.nexus.targ_mapped_lun = lun->lun;
13701                                 msg.ua.ua_all = 1;
13702                                 msg.ua.ua_set = (set > 0);
13703                                 msg.ua.ua_type = CTL_UA_THIN_PROV_THRES;
13704                                 memcpy(msg.ua.ua_info, lun->ua_tpt_info, 8);
13705                                 mtx_unlock(&softc->ctl_lock); // XXX
13706                                 ctl_ha_msg_send(CTL_HA_CHAN_CTL, &msg,
13707                                     sizeof(msg.ua), M_WAITOK);
13708                                 mtx_lock(&softc->ctl_lock);
13709                         }
13710                 }
13711                 mtx_unlock(&softc->ctl_lock);
13712                 pause("-", CTL_LBP_PERIOD * hz);
13713         }
13714 }
13715
13716 static void
13717 ctl_enqueue_incoming(union ctl_io *io)
13718 {
13719         struct ctl_softc *softc = control_softc;
13720         struct ctl_thread *thr;
13721         u_int idx;
13722
13723         idx = (io->io_hdr.nexus.targ_port * 127 +
13724                io->io_hdr.nexus.initid) % worker_threads;
13725         thr = &softc->threads[idx];
13726         mtx_lock(&thr->queue_lock);
13727         STAILQ_INSERT_TAIL(&thr->incoming_queue, &io->io_hdr, links);
13728         mtx_unlock(&thr->queue_lock);
13729         wakeup(thr);
13730 }
13731
13732 static void
13733 ctl_enqueue_rtr(union ctl_io *io)
13734 {
13735         struct ctl_softc *softc = control_softc;
13736         struct ctl_thread *thr;
13737
13738         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13739         mtx_lock(&thr->queue_lock);
13740         STAILQ_INSERT_TAIL(&thr->rtr_queue, &io->io_hdr, links);
13741         mtx_unlock(&thr->queue_lock);
13742         wakeup(thr);
13743 }
13744
13745 static void
13746 ctl_enqueue_done(union ctl_io *io)
13747 {
13748         struct ctl_softc *softc = control_softc;
13749         struct ctl_thread *thr;
13750
13751         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13752         mtx_lock(&thr->queue_lock);
13753         STAILQ_INSERT_TAIL(&thr->done_queue, &io->io_hdr, links);
13754         mtx_unlock(&thr->queue_lock);
13755         wakeup(thr);
13756 }
13757
13758 static void
13759 ctl_enqueue_isc(union ctl_io *io)
13760 {
13761         struct ctl_softc *softc = control_softc;
13762         struct ctl_thread *thr;
13763
13764         thr = &softc->threads[io->io_hdr.nexus.targ_mapped_lun % worker_threads];
13765         mtx_lock(&thr->queue_lock);
13766         STAILQ_INSERT_TAIL(&thr->isc_queue, &io->io_hdr, links);
13767         mtx_unlock(&thr->queue_lock);
13768         wakeup(thr);
13769 }
13770
13771 /*
13772  *  vim: ts=8
13773  */