2 * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/kernel.h>
33 #include <sys/module.h>
34 #include <sys/limits.h>
36 #include <sys/mutex.h>
39 #include <sys/sysctl.h>
40 #include <sys/malloc.h>
41 #include <sys/eventhandler.h>
43 #include <geom/geom.h>
45 #include <sys/kthread.h>
46 #include <sys/sched.h>
47 #include <geom/raid/g_raid.h>
48 #include "g_raid_md_if.h"
49 #include "g_raid_tr_if.h"
51 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
53 SYSCTL_DECL(_kern_geom);
54 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
55 int g_raid_enable = 1;
56 TUNABLE_INT("kern.geom.raid.enable", &g_raid_enable);
57 SYSCTL_INT(_kern_geom_raid, OID_AUTO, enable, CTLFLAG_RW,
58 &g_raid_enable, 0, "Enable on-disk metadata taste");
59 u_int g_raid_aggressive_spare = 0;
60 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
61 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
62 &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
63 u_int g_raid_debug = 0;
64 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
65 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
67 int g_raid_read_err_thresh = 10;
68 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
70 &g_raid_read_err_thresh, 0,
71 "Number of read errors equated to disk failure");
72 u_int g_raid_start_timeout = 30;
73 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
75 &g_raid_start_timeout, 0,
76 "Time to wait for all array components");
77 static u_int g_raid_clean_time = 5;
78 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
80 &g_raid_clean_time, 0, "Mark volume as clean when idling");
81 static u_int g_raid_disconnect_on_failure = 1;
82 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
83 &g_raid_disconnect_on_failure);
84 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
85 &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
86 static u_int g_raid_name_format = 0;
87 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
88 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
89 &g_raid_name_format, 0, "Providers name format.");
90 static u_int g_raid_idle_threshold = 1000000;
91 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
92 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
93 &g_raid_idle_threshold, 1000000,
94 "Time in microseconds to consider a volume idle.");
95 static u_int ar_legacy_aliases = 1;
96 SYSCTL_INT(_kern_geom_raid, OID_AUTO, legacy_aliases, CTLFLAG_RW,
97 &ar_legacy_aliases, 0, "Create aliases named as the legacy ataraid style.");
98 TUNABLE_INT("kern.geom_raid.legacy_aliases", &ar_legacy_aliases);
101 #define MSLEEP(rv, ident, mtx, priority, wmesg, timeout) do { \
102 G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident)); \
103 rv = msleep((ident), (mtx), (priority), (wmesg), (timeout)); \
104 G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident)); \
107 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
108 LIST_HEAD_INITIALIZER(g_raid_md_classes);
110 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
111 LIST_HEAD_INITIALIZER(g_raid_tr_classes);
113 LIST_HEAD(, g_raid_volume) g_raid_volumes =
114 LIST_HEAD_INITIALIZER(g_raid_volumes);
116 static eventhandler_tag g_raid_post_sync = NULL;
117 static int g_raid_started = 0;
118 static int g_raid_shutdown = 0;
120 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
122 static g_taste_t g_raid_taste;
123 static void g_raid_init(struct g_class *mp);
124 static void g_raid_fini(struct g_class *mp);
126 struct g_class g_raid_class = {
127 .name = G_RAID_CLASS_NAME,
128 .version = G_VERSION,
129 .ctlreq = g_raid_ctl,
130 .taste = g_raid_taste,
131 .destroy_geom = g_raid_destroy_geom,
136 static void g_raid_destroy_provider(struct g_raid_volume *vol);
137 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
138 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
139 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
140 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
141 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
142 struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
143 static void g_raid_start(struct bio *bp);
144 static void g_raid_start_request(struct bio *bp);
145 static void g_raid_disk_done(struct bio *bp);
146 static void g_raid_poll(struct g_raid_softc *sc);
149 g_raid_node_event2str(int event)
153 case G_RAID_NODE_E_WAKE:
155 case G_RAID_NODE_E_START:
163 g_raid_disk_state2str(int state)
167 case G_RAID_DISK_S_NONE:
169 case G_RAID_DISK_S_OFFLINE:
171 case G_RAID_DISK_S_DISABLED:
173 case G_RAID_DISK_S_FAILED:
175 case G_RAID_DISK_S_STALE_FAILED:
176 return ("STALE_FAILED");
177 case G_RAID_DISK_S_SPARE:
179 case G_RAID_DISK_S_STALE:
181 case G_RAID_DISK_S_ACTIVE:
189 g_raid_disk_event2str(int event)
193 case G_RAID_DISK_E_DISCONNECTED:
194 return ("DISCONNECTED");
201 g_raid_subdisk_state2str(int state)
205 case G_RAID_SUBDISK_S_NONE:
207 case G_RAID_SUBDISK_S_FAILED:
209 case G_RAID_SUBDISK_S_NEW:
211 case G_RAID_SUBDISK_S_REBUILD:
213 case G_RAID_SUBDISK_S_UNINITIALIZED:
214 return ("UNINITIALIZED");
215 case G_RAID_SUBDISK_S_STALE:
217 case G_RAID_SUBDISK_S_RESYNC:
219 case G_RAID_SUBDISK_S_ACTIVE:
227 g_raid_subdisk_event2str(int event)
231 case G_RAID_SUBDISK_E_NEW:
233 case G_RAID_SUBDISK_E_FAILED:
235 case G_RAID_SUBDISK_E_DISCONNECTED:
236 return ("DISCONNECTED");
243 g_raid_volume_state2str(int state)
247 case G_RAID_VOLUME_S_STARTING:
249 case G_RAID_VOLUME_S_BROKEN:
251 case G_RAID_VOLUME_S_DEGRADED:
253 case G_RAID_VOLUME_S_SUBOPTIMAL:
254 return ("SUBOPTIMAL");
255 case G_RAID_VOLUME_S_OPTIMAL:
257 case G_RAID_VOLUME_S_UNSUPPORTED:
258 return ("UNSUPPORTED");
259 case G_RAID_VOLUME_S_STOPPED:
267 g_raid_volume_event2str(int event)
271 case G_RAID_VOLUME_E_UP:
273 case G_RAID_VOLUME_E_DOWN:
275 case G_RAID_VOLUME_E_START:
277 case G_RAID_VOLUME_E_STARTMD:
285 g_raid_volume_level2str(int level, int qual)
289 case G_RAID_VOLUME_RL_RAID0:
291 case G_RAID_VOLUME_RL_RAID1:
293 case G_RAID_VOLUME_RL_RAID3:
294 if (qual == G_RAID_VOLUME_RLQ_R3P0)
296 if (qual == G_RAID_VOLUME_RLQ_R3PN)
299 case G_RAID_VOLUME_RL_RAID4:
300 if (qual == G_RAID_VOLUME_RLQ_R4P0)
302 if (qual == G_RAID_VOLUME_RLQ_R4PN)
305 case G_RAID_VOLUME_RL_RAID5:
306 if (qual == G_RAID_VOLUME_RLQ_R5RA)
308 if (qual == G_RAID_VOLUME_RLQ_R5RS)
310 if (qual == G_RAID_VOLUME_RLQ_R5LA)
312 if (qual == G_RAID_VOLUME_RLQ_R5LS)
315 case G_RAID_VOLUME_RL_RAID6:
316 if (qual == G_RAID_VOLUME_RLQ_R6RA)
318 if (qual == G_RAID_VOLUME_RLQ_R6RS)
320 if (qual == G_RAID_VOLUME_RLQ_R6LA)
322 if (qual == G_RAID_VOLUME_RLQ_R6LS)
325 case G_RAID_VOLUME_RL_RAIDMDF:
326 if (qual == G_RAID_VOLUME_RLQ_RMDFRA)
327 return ("RAIDMDF-RA");
328 if (qual == G_RAID_VOLUME_RLQ_RMDFRS)
329 return ("RAIDMDF-RS");
330 if (qual == G_RAID_VOLUME_RLQ_RMDFLA)
331 return ("RAIDMDF-LA");
332 if (qual == G_RAID_VOLUME_RLQ_RMDFLS)
333 return ("RAIDMDF-LS");
335 case G_RAID_VOLUME_RL_RAID1E:
336 if (qual == G_RAID_VOLUME_RLQ_R1EA)
338 if (qual == G_RAID_VOLUME_RLQ_R1EO)
341 case G_RAID_VOLUME_RL_SINGLE:
343 case G_RAID_VOLUME_RL_CONCAT:
345 case G_RAID_VOLUME_RL_RAID5E:
346 if (qual == G_RAID_VOLUME_RLQ_R5ERA)
347 return ("RAID5E-RA");
348 if (qual == G_RAID_VOLUME_RLQ_R5ERS)
349 return ("RAID5E-RS");
350 if (qual == G_RAID_VOLUME_RLQ_R5ELA)
351 return ("RAID5E-LA");
352 if (qual == G_RAID_VOLUME_RLQ_R5ELS)
353 return ("RAID5E-LS");
355 case G_RAID_VOLUME_RL_RAID5EE:
356 if (qual == G_RAID_VOLUME_RLQ_R5EERA)
357 return ("RAID5EE-RA");
358 if (qual == G_RAID_VOLUME_RLQ_R5EERS)
359 return ("RAID5EE-RS");
360 if (qual == G_RAID_VOLUME_RLQ_R5EELA)
361 return ("RAID5EE-LA");
362 if (qual == G_RAID_VOLUME_RLQ_R5EELS)
363 return ("RAID5EE-LS");
365 case G_RAID_VOLUME_RL_RAID5R:
366 if (qual == G_RAID_VOLUME_RLQ_R5RRA)
367 return ("RAID5R-RA");
368 if (qual == G_RAID_VOLUME_RLQ_R5RRS)
369 return ("RAID5R-RS");
370 if (qual == G_RAID_VOLUME_RLQ_R5RLA)
371 return ("RAID5R-LA");
372 if (qual == G_RAID_VOLUME_RLQ_R5RLS)
373 return ("RAID5R-LS");
381 g_raid_volume_str2level(const char *str, int *level, int *qual)
384 *level = G_RAID_VOLUME_RL_UNKNOWN;
385 *qual = G_RAID_VOLUME_RLQ_NONE;
386 if (strcasecmp(str, "RAID0") == 0)
387 *level = G_RAID_VOLUME_RL_RAID0;
388 else if (strcasecmp(str, "RAID1") == 0)
389 *level = G_RAID_VOLUME_RL_RAID1;
390 else if (strcasecmp(str, "RAID3-P0") == 0) {
391 *level = G_RAID_VOLUME_RL_RAID3;
392 *qual = G_RAID_VOLUME_RLQ_R3P0;
393 } else if (strcasecmp(str, "RAID3-PN") == 0 ||
394 strcasecmp(str, "RAID3") == 0) {
395 *level = G_RAID_VOLUME_RL_RAID3;
396 *qual = G_RAID_VOLUME_RLQ_R3PN;
397 } else if (strcasecmp(str, "RAID4-P0") == 0) {
398 *level = G_RAID_VOLUME_RL_RAID4;
399 *qual = G_RAID_VOLUME_RLQ_R4P0;
400 } else if (strcasecmp(str, "RAID4-PN") == 0 ||
401 strcasecmp(str, "RAID4") == 0) {
402 *level = G_RAID_VOLUME_RL_RAID4;
403 *qual = G_RAID_VOLUME_RLQ_R4PN;
404 } else if (strcasecmp(str, "RAID5-RA") == 0) {
405 *level = G_RAID_VOLUME_RL_RAID5;
406 *qual = G_RAID_VOLUME_RLQ_R5RA;
407 } else if (strcasecmp(str, "RAID5-RS") == 0) {
408 *level = G_RAID_VOLUME_RL_RAID5;
409 *qual = G_RAID_VOLUME_RLQ_R5RS;
410 } else if (strcasecmp(str, "RAID5") == 0 ||
411 strcasecmp(str, "RAID5-LA") == 0) {
412 *level = G_RAID_VOLUME_RL_RAID5;
413 *qual = G_RAID_VOLUME_RLQ_R5LA;
414 } else if (strcasecmp(str, "RAID5-LS") == 0) {
415 *level = G_RAID_VOLUME_RL_RAID5;
416 *qual = G_RAID_VOLUME_RLQ_R5LS;
417 } else if (strcasecmp(str, "RAID6-RA") == 0) {
418 *level = G_RAID_VOLUME_RL_RAID6;
419 *qual = G_RAID_VOLUME_RLQ_R6RA;
420 } else if (strcasecmp(str, "RAID6-RS") == 0) {
421 *level = G_RAID_VOLUME_RL_RAID6;
422 *qual = G_RAID_VOLUME_RLQ_R6RS;
423 } else if (strcasecmp(str, "RAID6") == 0 ||
424 strcasecmp(str, "RAID6-LA") == 0) {
425 *level = G_RAID_VOLUME_RL_RAID6;
426 *qual = G_RAID_VOLUME_RLQ_R6LA;
427 } else if (strcasecmp(str, "RAID6-LS") == 0) {
428 *level = G_RAID_VOLUME_RL_RAID6;
429 *qual = G_RAID_VOLUME_RLQ_R6LS;
430 } else if (strcasecmp(str, "RAIDMDF-RA") == 0) {
431 *level = G_RAID_VOLUME_RL_RAIDMDF;
432 *qual = G_RAID_VOLUME_RLQ_RMDFRA;
433 } else if (strcasecmp(str, "RAIDMDF-RS") == 0) {
434 *level = G_RAID_VOLUME_RL_RAIDMDF;
435 *qual = G_RAID_VOLUME_RLQ_RMDFRS;
436 } else if (strcasecmp(str, "RAIDMDF") == 0 ||
437 strcasecmp(str, "RAIDMDF-LA") == 0) {
438 *level = G_RAID_VOLUME_RL_RAIDMDF;
439 *qual = G_RAID_VOLUME_RLQ_RMDFLA;
440 } else if (strcasecmp(str, "RAIDMDF-LS") == 0) {
441 *level = G_RAID_VOLUME_RL_RAIDMDF;
442 *qual = G_RAID_VOLUME_RLQ_RMDFLS;
443 } else if (strcasecmp(str, "RAID10") == 0 ||
444 strcasecmp(str, "RAID1E") == 0 ||
445 strcasecmp(str, "RAID1E-A") == 0) {
446 *level = G_RAID_VOLUME_RL_RAID1E;
447 *qual = G_RAID_VOLUME_RLQ_R1EA;
448 } else if (strcasecmp(str, "RAID1E-O") == 0) {
449 *level = G_RAID_VOLUME_RL_RAID1E;
450 *qual = G_RAID_VOLUME_RLQ_R1EO;
451 } else if (strcasecmp(str, "SINGLE") == 0)
452 *level = G_RAID_VOLUME_RL_SINGLE;
453 else if (strcasecmp(str, "CONCAT") == 0)
454 *level = G_RAID_VOLUME_RL_CONCAT;
455 else if (strcasecmp(str, "RAID5E-RA") == 0) {
456 *level = G_RAID_VOLUME_RL_RAID5E;
457 *qual = G_RAID_VOLUME_RLQ_R5ERA;
458 } else if (strcasecmp(str, "RAID5E-RS") == 0) {
459 *level = G_RAID_VOLUME_RL_RAID5E;
460 *qual = G_RAID_VOLUME_RLQ_R5ERS;
461 } else if (strcasecmp(str, "RAID5E") == 0 ||
462 strcasecmp(str, "RAID5E-LA") == 0) {
463 *level = G_RAID_VOLUME_RL_RAID5E;
464 *qual = G_RAID_VOLUME_RLQ_R5ELA;
465 } else if (strcasecmp(str, "RAID5E-LS") == 0) {
466 *level = G_RAID_VOLUME_RL_RAID5E;
467 *qual = G_RAID_VOLUME_RLQ_R5ELS;
468 } else if (strcasecmp(str, "RAID5EE-RA") == 0) {
469 *level = G_RAID_VOLUME_RL_RAID5EE;
470 *qual = G_RAID_VOLUME_RLQ_R5EERA;
471 } else if (strcasecmp(str, "RAID5EE-RS") == 0) {
472 *level = G_RAID_VOLUME_RL_RAID5EE;
473 *qual = G_RAID_VOLUME_RLQ_R5EERS;
474 } else if (strcasecmp(str, "RAID5EE") == 0 ||
475 strcasecmp(str, "RAID5EE-LA") == 0) {
476 *level = G_RAID_VOLUME_RL_RAID5EE;
477 *qual = G_RAID_VOLUME_RLQ_R5EELA;
478 } else if (strcasecmp(str, "RAID5EE-LS") == 0) {
479 *level = G_RAID_VOLUME_RL_RAID5EE;
480 *qual = G_RAID_VOLUME_RLQ_R5EELS;
481 } else if (strcasecmp(str, "RAID5R-RA") == 0) {
482 *level = G_RAID_VOLUME_RL_RAID5R;
483 *qual = G_RAID_VOLUME_RLQ_R5RRA;
484 } else if (strcasecmp(str, "RAID5R-RS") == 0) {
485 *level = G_RAID_VOLUME_RL_RAID5R;
486 *qual = G_RAID_VOLUME_RLQ_R5RRS;
487 } else if (strcasecmp(str, "RAID5R") == 0 ||
488 strcasecmp(str, "RAID5R-LA") == 0) {
489 *level = G_RAID_VOLUME_RL_RAID5R;
490 *qual = G_RAID_VOLUME_RLQ_R5RLA;
491 } else if (strcasecmp(str, "RAID5R-LS") == 0) {
492 *level = G_RAID_VOLUME_RL_RAID5R;
493 *qual = G_RAID_VOLUME_RLQ_R5RLS;
500 g_raid_get_diskname(struct g_raid_disk *disk)
503 if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
504 return ("[unknown]");
505 return (disk->d_consumer->provider->name);
509 g_raid_get_disk_info(struct g_raid_disk *disk)
511 struct g_consumer *cp = disk->d_consumer;
514 /* Read kernel dumping information. */
515 disk->d_kd.offset = 0;
516 disk->d_kd.length = OFF_MAX;
517 len = sizeof(disk->d_kd);
518 error = g_io_getattr("GEOM::kerneldump", cp, &len, &disk->d_kd);
520 disk->d_kd.di.dumper = NULL;
521 if (disk->d_kd.di.dumper == NULL)
522 G_RAID_DEBUG1(2, disk->d_softc,
523 "Dumping not supported by %s: %d.",
524 cp->provider->name, error);
526 /* Read BIO_DELETE support. */
527 error = g_getattr("GEOM::candelete", cp, &disk->d_candelete);
529 disk->d_candelete = 0;
530 if (!disk->d_candelete)
531 G_RAID_DEBUG1(2, disk->d_softc,
532 "BIO_DELETE not supported by %s: %d.",
533 cp->provider->name, error);
537 g_raid_report_disk_state(struct g_raid_disk *disk)
539 struct g_raid_subdisk *sd;
543 if (disk->d_consumer == NULL)
545 if (disk->d_state == G_RAID_DISK_S_DISABLED) {
546 s = G_STATE_ACTIVE; /* XXX */
547 } else if (disk->d_state == G_RAID_DISK_S_FAILED ||
548 disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
551 state = G_RAID_SUBDISK_S_ACTIVE;
552 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
553 if (sd->sd_state < state)
554 state = sd->sd_state;
556 if (state == G_RAID_SUBDISK_S_FAILED)
558 else if (state == G_RAID_SUBDISK_S_NEW ||
559 state == G_RAID_SUBDISK_S_REBUILD)
561 else if (state == G_RAID_SUBDISK_S_STALE ||
562 state == G_RAID_SUBDISK_S_RESYNC)
568 g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
569 G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
570 g_raid_get_diskname(disk), s);
574 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
577 G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
578 g_raid_get_diskname(disk),
579 g_raid_disk_state2str(disk->d_state),
580 g_raid_disk_state2str(state));
581 disk->d_state = state;
582 g_raid_report_disk_state(disk);
586 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
589 G_RAID_DEBUG1(0, sd->sd_softc,
590 "Subdisk %s:%d-%s state changed from %s to %s.",
591 sd->sd_volume->v_name, sd->sd_pos,
592 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
593 g_raid_subdisk_state2str(sd->sd_state),
594 g_raid_subdisk_state2str(state));
595 sd->sd_state = state;
597 g_raid_report_disk_state(sd->sd_disk);
601 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
604 G_RAID_DEBUG1(0, vol->v_softc,
605 "Volume %s state changed from %s to %s.",
607 g_raid_volume_state2str(vol->v_state),
608 g_raid_volume_state2str(state));
609 vol->v_state = state;
613 * --- Events handling functions ---
614 * Events in geom_raid are used to maintain subdisks and volumes status
615 * from one thread to simplify locking.
618 g_raid_event_free(struct g_raid_event *ep)
625 g_raid_event_send(void *arg, int event, int flags)
627 struct g_raid_softc *sc;
628 struct g_raid_event *ep;
631 if ((flags & G_RAID_EVENT_VOLUME) != 0) {
632 sc = ((struct g_raid_volume *)arg)->v_softc;
633 } else if ((flags & G_RAID_EVENT_DISK) != 0) {
634 sc = ((struct g_raid_disk *)arg)->d_softc;
635 } else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
636 sc = ((struct g_raid_subdisk *)arg)->sd_softc;
640 ep = malloc(sizeof(*ep), M_RAID,
641 sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
648 G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
649 mtx_lock(&sc->sc_queue_mtx);
650 TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
651 mtx_unlock(&sc->sc_queue_mtx);
654 if ((flags & G_RAID_EVENT_WAIT) == 0)
657 sx_assert(&sc->sc_lock, SX_XLOCKED);
658 G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
659 sx_xunlock(&sc->sc_lock);
660 while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
661 mtx_lock(&sc->sc_queue_mtx);
662 MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
666 g_raid_event_free(ep);
667 sx_xlock(&sc->sc_lock);
672 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
674 struct g_raid_event *ep, *tmpep;
676 sx_assert(&sc->sc_lock, SX_XLOCKED);
678 mtx_lock(&sc->sc_queue_mtx);
679 TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
680 if (ep->e_tgt != tgt)
682 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
683 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
684 g_raid_event_free(ep);
686 ep->e_error = ECANCELED;
690 mtx_unlock(&sc->sc_queue_mtx);
694 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
696 struct g_raid_event *ep;
699 sx_assert(&sc->sc_lock, SX_XLOCKED);
701 mtx_lock(&sc->sc_queue_mtx);
702 TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
703 if (ep->e_tgt != tgt)
708 mtx_unlock(&sc->sc_queue_mtx);
713 * Return the number of disks in given state.
714 * If state is equal to -1, count all connected disks.
717 g_raid_ndisks(struct g_raid_softc *sc, int state)
719 struct g_raid_disk *disk;
722 sx_assert(&sc->sc_lock, SX_LOCKED);
725 TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
726 if (disk->d_state == state || state == -1)
733 * Return the number of subdisks in given state.
734 * If state is equal to -1, count all connected disks.
737 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
739 struct g_raid_subdisk *subdisk;
740 struct g_raid_softc *sc;
744 sx_assert(&sc->sc_lock, SX_LOCKED);
747 for (i = 0; i < vol->v_disks_count; i++) {
748 subdisk = &vol->v_subdisks[i];
750 subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
751 subdisk->sd_state == state)
758 * Return the first subdisk in given state.
759 * If state is equal to -1, then the first connected disks.
761 struct g_raid_subdisk *
762 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
764 struct g_raid_subdisk *sd;
765 struct g_raid_softc *sc;
769 sx_assert(&sc->sc_lock, SX_LOCKED);
771 for (i = 0; i < vol->v_disks_count; i++) {
772 sd = &vol->v_subdisks[i];
774 sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
775 sd->sd_state == state)
782 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
784 struct g_consumer *cp;
785 struct g_provider *pp;
789 if (strncmp(name, "/dev/", 5) == 0)
791 pp = g_provider_by_name(name);
794 cp = g_new_consumer(sc->sc_geom);
795 if (g_attach(cp, pp) != 0) {
796 g_destroy_consumer(cp);
799 if (g_access(cp, 1, 1, 1) != 0) {
801 g_destroy_consumer(cp);
808 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
813 mtx_lock(&sc->sc_queue_mtx);
814 TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
815 if (bp->bio_from == cp)
818 mtx_unlock(&sc->sc_queue_mtx);
823 g_raid_nopens(struct g_raid_softc *sc)
825 struct g_raid_volume *vol;
829 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
830 if (vol->v_provider_open != 0)
837 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
842 "I/O requests for %s exist, can't destroy it now.",
846 if (g_raid_nrequests(sc, cp) > 0) {
848 "I/O requests for %s in queue, can't destroy it now.",
856 g_raid_destroy_consumer(void *arg, int flags __unused)
858 struct g_consumer *cp;
863 G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
865 g_destroy_consumer(cp);
869 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
871 struct g_provider *pp;
874 g_topology_assert_not();
878 if (g_raid_consumer_is_busy(sc, cp))
883 if ((pp->geom->flags & G_GEOM_WITHER) == 0)
886 if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
887 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
890 * After retaste event was send (inside g_access()), we can send
891 * event to detach and destroy consumer.
892 * A class, which has consumer to the given provider connected
893 * will not receive retaste event for the provider.
894 * This is the way how I ignore retaste events when I close
895 * consumers opened for write: I detach and destroy consumer
896 * after retaste event is sent.
898 g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
901 G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
903 g_destroy_consumer(cp);
909 g_raid_orphan(struct g_consumer *cp)
911 struct g_raid_disk *disk;
918 g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
923 g_raid_clean(struct g_raid_volume *vol, int acw)
925 struct g_raid_softc *sc;
929 g_topology_assert_not();
930 sx_assert(&sc->sc_lock, SX_XLOCKED);
932 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
936 if (vol->v_writes > 0)
938 if (acw > 0 || (acw == -1 &&
939 vol->v_provider != NULL && vol->v_provider->acw > 0)) {
940 timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
941 if (!g_raid_shutdown && timeout > 0)
945 G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
947 g_raid_write_metadata(sc, vol, NULL, NULL);
951 g_raid_dirty(struct g_raid_volume *vol)
953 struct g_raid_softc *sc;
956 g_topology_assert_not();
957 sx_assert(&sc->sc_lock, SX_XLOCKED);
959 // if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
962 G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
964 g_raid_write_metadata(sc, vol, NULL, NULL);
968 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
970 struct g_raid_softc *sc;
971 struct g_raid_volume *vol;
972 struct g_raid_subdisk *sd;
973 struct bio_queue_head queue;
977 vol = tr->tro_volume;
981 * Allocate all bios before sending any request, so we can return
982 * ENOMEM in nice and clean way.
985 for (i = 0; i < vol->v_disks_count; i++) {
986 sd = &vol->v_subdisks[i];
987 if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
988 sd->sd_state == G_RAID_SUBDISK_S_FAILED)
990 cbp = g_clone_bio(bp);
993 cbp->bio_caller1 = sd;
994 bioq_insert_tail(&queue, cbp);
996 for (cbp = bioq_first(&queue); cbp != NULL;
997 cbp = bioq_first(&queue)) {
998 bioq_remove(&queue, cbp);
999 sd = cbp->bio_caller1;
1000 cbp->bio_caller1 = NULL;
1001 g_raid_subdisk_iostart(sd, cbp);
1005 for (cbp = bioq_first(&queue); cbp != NULL;
1006 cbp = bioq_first(&queue)) {
1007 bioq_remove(&queue, cbp);
1010 if (bp->bio_error == 0)
1011 bp->bio_error = ENOMEM;
1012 g_raid_iodone(bp, bp->bio_error);
1016 g_raid_tr_kerneldump_common_done(struct bio *bp)
1019 bp->bio_flags |= BIO_DONE;
1023 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
1024 void *virtual, vm_offset_t physical, off_t offset, size_t length)
1026 struct g_raid_softc *sc;
1027 struct g_raid_volume *vol;
1030 vol = tr->tro_volume;
1033 bzero(&bp, sizeof(bp));
1034 bp.bio_cmd = BIO_WRITE;
1035 bp.bio_done = g_raid_tr_kerneldump_common_done;
1036 bp.bio_attribute = NULL;
1037 bp.bio_offset = offset;
1038 bp.bio_length = length;
1039 bp.bio_data = virtual;
1040 bp.bio_to = vol->v_provider;
1043 while (!(bp.bio_flags & BIO_DONE)) {
1044 G_RAID_DEBUG1(4, sc, "Poll...");
1049 return (bp.bio_error != 0 ? EIO : 0);
1053 g_raid_dump(void *arg,
1054 void *virtual, vm_offset_t physical, off_t offset, size_t length)
1056 struct g_raid_volume *vol;
1059 vol = (struct g_raid_volume *)arg;
1060 G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
1061 (long long unsigned)offset, (long long unsigned)length);
1063 error = G_RAID_TR_KERNELDUMP(vol->v_tr,
1064 virtual, physical, offset, length);
1069 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
1071 struct g_kerneldump *gkd;
1072 struct g_provider *pp;
1073 struct g_raid_volume *vol;
1075 gkd = (struct g_kerneldump*)bp->bio_data;
1078 g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
1079 pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
1080 gkd->di.dumper = g_raid_dump;
1082 gkd->di.blocksize = vol->v_sectorsize;
1083 gkd->di.maxiosize = DFLTPHYS;
1084 gkd->di.mediaoffset = gkd->offset;
1085 if ((gkd->offset + gkd->length) > vol->v_mediasize)
1086 gkd->length = vol->v_mediasize - gkd->offset;
1087 gkd->di.mediasize = gkd->length;
1088 g_io_deliver(bp, 0);
1092 g_raid_candelete(struct g_raid_softc *sc, struct bio *bp)
1094 struct g_provider *pp;
1095 struct g_raid_volume *vol;
1096 struct g_raid_subdisk *sd;
1100 val = (int *)bp->bio_data;
1104 for (i = 0; i < vol->v_disks_count; i++) {
1105 sd = &vol->v_subdisks[i];
1106 if (sd->sd_state == G_RAID_SUBDISK_S_NONE)
1108 if (sd->sd_disk->d_candelete) {
1113 g_io_deliver(bp, 0);
1117 g_raid_start(struct bio *bp)
1119 struct g_raid_softc *sc;
1121 sc = bp->bio_to->geom->softc;
1123 * If sc == NULL or there are no valid disks, provider's error
1124 * should be set and g_raid_start() should not be called at all.
1126 // KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
1127 // ("Provider's error should be set (error=%d)(mirror=%s).",
1128 // bp->bio_to->error, bp->bio_to->name));
1129 G_RAID_LOGREQ(3, bp, "Request received.");
1131 switch (bp->bio_cmd) {
1138 if (!strcmp(bp->bio_attribute, "GEOM::candelete"))
1139 g_raid_candelete(sc, bp);
1140 else if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
1141 g_raid_kerneldump(sc, bp);
1143 g_io_deliver(bp, EOPNOTSUPP);
1146 g_io_deliver(bp, EOPNOTSUPP);
1149 mtx_lock(&sc->sc_queue_mtx);
1150 bioq_disksort(&sc->sc_queue, bp);
1151 mtx_unlock(&sc->sc_queue_mtx);
1153 G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
1159 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
1163 * (1) bp entirely below NO
1164 * (2) bp entirely above NO
1165 * (3) bp start below, but end in range YES
1166 * (4) bp entirely within YES
1167 * (5) bp starts within, ends above YES
1169 * lock range 10-19 (offset 10 length 10)
1170 * (1) 1-5: first if kicks it out
1171 * (2) 30-35: second if kicks it out
1172 * (3) 5-15: passes both ifs
1173 * (4) 12-14: passes both ifs
1174 * (5) 19-20: passes both
1176 off_t lend = lstart + len - 1;
1177 off_t bstart = bp->bio_offset;
1178 off_t bend = bp->bio_offset + bp->bio_length - 1;
1188 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
1190 struct g_raid_lock *lp;
1192 sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
1194 LIST_FOREACH(lp, &vol->v_locks, l_next) {
1195 if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
1202 g_raid_start_request(struct bio *bp)
1204 struct g_raid_softc *sc;
1205 struct g_raid_volume *vol;
1207 sc = bp->bio_to->geom->softc;
1208 sx_assert(&sc->sc_lock, SX_LOCKED);
1209 vol = bp->bio_to->private;
1212 * Check to see if this item is in a locked range. If so,
1213 * queue it to our locked queue and return. We'll requeue
1214 * it when the range is unlocked. Internal I/O for the
1215 * rebuild/rescan/recovery process is excluded from this
1216 * check so we can actually do the recovery.
1218 if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
1219 g_raid_is_in_locked_range(vol, bp)) {
1220 G_RAID_LOGREQ(3, bp, "Defer request.");
1221 bioq_insert_tail(&vol->v_locked, bp);
1226 * If we're actually going to do the write/delete, then
1227 * update the idle stats for the volume.
1229 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1236 * Put request onto inflight queue, so we can check if new
1237 * synchronization requests don't collide with it. Then tell
1238 * the transformation layer to start the I/O.
1240 bioq_insert_tail(&vol->v_inflight, bp);
1241 G_RAID_LOGREQ(4, bp, "Request started");
1242 G_RAID_TR_IOSTART(vol->v_tr, bp);
1246 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1250 struct g_raid_lock *lp;
1252 vol->v_pending_lock = 0;
1253 LIST_FOREACH(lp, &vol->v_locks, l_next) {
1254 if (lp->l_pending) {
1258 TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1259 if (g_raid_bio_overlaps(nbp, off, len))
1262 if (lp->l_pending) {
1263 vol->v_pending_lock = 1;
1264 G_RAID_DEBUG1(4, vol->v_softc,
1265 "Deferred lock(%jd, %jd) has %d pending",
1266 (intmax_t)off, (intmax_t)(off + len),
1270 G_RAID_DEBUG1(4, vol->v_softc,
1271 "Deferred lock of %jd to %jd completed",
1272 (intmax_t)off, (intmax_t)(off + len));
1273 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1279 g_raid_iodone(struct bio *bp, int error)
1281 struct g_raid_softc *sc;
1282 struct g_raid_volume *vol;
1284 sc = bp->bio_to->geom->softc;
1285 sx_assert(&sc->sc_lock, SX_LOCKED);
1286 vol = bp->bio_to->private;
1287 G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1289 /* Update stats if we done write/delete. */
1290 if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1292 vol->v_last_write = time_uptime;
1295 bioq_remove(&vol->v_inflight, bp);
1296 if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1297 g_raid_finish_with_locked_ranges(vol, bp);
1298 getmicrouptime(&vol->v_last_done);
1299 g_io_deliver(bp, error);
1303 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1304 struct bio *ignore, void *argp)
1306 struct g_raid_softc *sc;
1307 struct g_raid_lock *lp;
1311 lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1312 LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1315 lp->l_callback_arg = argp;
1318 TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1319 if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1324 * If there are any writes that are pending, we return EBUSY. All
1325 * callers will have to wait until all pending writes clear.
1327 if (lp->l_pending > 0) {
1328 vol->v_pending_lock = 1;
1329 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1330 (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1333 G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1334 (intmax_t)off, (intmax_t)(off+len));
1335 G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1340 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1342 struct g_raid_lock *lp;
1343 struct g_raid_softc *sc;
1347 LIST_FOREACH(lp, &vol->v_locks, l_next) {
1348 if (lp->l_offset == off && lp->l_length == len) {
1349 LIST_REMOVE(lp, l_next);
1351 * Right now we just put them all back on the queue
1352 * and hope for the best. We hope this because any
1353 * locked ranges will go right back on this list
1354 * when the worker thread runs.
1357 G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1358 (intmax_t)lp->l_offset,
1359 (intmax_t)(lp->l_offset+lp->l_length));
1360 mtx_lock(&sc->sc_queue_mtx);
1361 while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1362 bioq_disksort(&sc->sc_queue, bp);
1363 mtx_unlock(&sc->sc_queue_mtx);
1372 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1374 struct g_consumer *cp;
1375 struct g_raid_disk *disk, *tdisk;
1377 bp->bio_caller1 = sd;
1380 * Make sure that the disk is present. Generally it is a task of
1381 * transformation layers to not send requests to absent disks, but
1382 * it is better to be safe and report situation then sorry.
1384 if (sd->sd_disk == NULL) {
1385 G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1387 bp->bio_from = NULL;
1389 bp->bio_error = ENXIO;
1390 g_raid_disk_done(bp);
1394 if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1395 disk->d_state != G_RAID_DISK_S_FAILED) {
1396 G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1397 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1401 cp = disk->d_consumer;
1403 bp->bio_to = cp->provider;
1406 /* Update average disks load. */
1407 TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1408 if (tdisk->d_consumer == NULL)
1411 tdisk->d_load = (tdisk->d_consumer->index *
1412 G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1415 disk->d_last_offset = bp->bio_offset + bp->bio_length;
1417 G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1418 if (bp->bio_cmd == BIO_WRITE) {
1419 bp->bio_error = g_raid_subdisk_kerneldump(sd,
1420 bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1422 bp->bio_error = EOPNOTSUPP;
1423 g_raid_disk_done(bp);
1425 bp->bio_done = g_raid_disk_done;
1426 bp->bio_offset += sd->sd_offset;
1427 G_RAID_LOGREQ(3, bp, "Sending request.");
1428 g_io_request(bp, cp);
1433 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1434 void *virtual, vm_offset_t physical, off_t offset, size_t length)
1437 if (sd->sd_disk == NULL)
1439 if (sd->sd_disk->d_kd.di.dumper == NULL)
1440 return (EOPNOTSUPP);
1441 return (dump_write(&sd->sd_disk->d_kd.di,
1443 sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1448 g_raid_disk_done(struct bio *bp)
1450 struct g_raid_softc *sc;
1451 struct g_raid_subdisk *sd;
1453 sd = bp->bio_caller1;
1455 mtx_lock(&sc->sc_queue_mtx);
1456 bioq_disksort(&sc->sc_queue, bp);
1457 mtx_unlock(&sc->sc_queue_mtx);
1463 g_raid_disk_done_request(struct bio *bp)
1465 struct g_raid_softc *sc;
1466 struct g_raid_disk *disk;
1467 struct g_raid_subdisk *sd;
1468 struct g_raid_volume *vol;
1470 g_topology_assert_not();
1472 G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1473 sd = bp->bio_caller1;
1475 vol = sd->sd_volume;
1476 if (bp->bio_from != NULL) {
1477 bp->bio_from->index--;
1478 disk = bp->bio_from->private;
1480 g_raid_kill_consumer(sc, bp->bio_from);
1482 bp->bio_offset -= sd->sd_offset;
1484 G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1488 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1491 if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1492 ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1493 else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1494 ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1495 else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1496 ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1498 ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1499 if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1500 KASSERT(ep->e_error == 0,
1501 ("Error cannot be handled."));
1502 g_raid_event_free(ep);
1504 ep->e_flags |= G_RAID_EVENT_DONE;
1505 G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1506 mtx_lock(&sc->sc_queue_mtx);
1508 mtx_unlock(&sc->sc_queue_mtx);
1516 g_raid_worker(void *arg)
1518 struct g_raid_softc *sc;
1519 struct g_raid_event *ep;
1520 struct g_raid_volume *vol;
1522 struct timeval now, t;
1526 thread_lock(curthread);
1527 sched_prio(curthread, PRIBIO);
1528 thread_unlock(curthread);
1530 sx_xlock(&sc->sc_lock);
1532 mtx_lock(&sc->sc_queue_mtx);
1534 * First take a look at events.
1535 * This is important to handle events before any I/O requests.
1540 ep = TAILQ_FIRST(&sc->sc_events);
1542 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1543 else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1546 getmicrouptime(&now);
1548 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1549 if (bioq_first(&vol->v_inflight) == NULL &&
1551 timevalcmp(&vol->v_last_done, &t, < ))
1552 t = vol->v_last_done;
1554 timevalsub(&t, &now);
1555 timeout = g_raid_idle_threshold +
1556 t.tv_sec * 1000000 + t.tv_usec;
1559 * Two steps to avoid overflows at HZ=1000
1560 * and idle timeouts > 2.1s. Some rounding
1561 * errors can occur, but they are < 1tick,
1562 * which is deemed to be close enough for
1565 int micpertic = 1000000 / hz;
1566 timeout = (timeout + micpertic - 1) / micpertic;
1567 sx_xunlock(&sc->sc_lock);
1568 MSLEEP(rv, sc, &sc->sc_queue_mtx,
1569 PRIBIO | PDROP, "-", timeout);
1570 sx_xlock(&sc->sc_lock);
1575 mtx_unlock(&sc->sc_queue_mtx);
1578 g_raid_handle_event(sc, ep);
1579 } else if (bp != NULL) {
1580 if (bp->bio_to != NULL &&
1581 bp->bio_to->geom == sc->sc_geom)
1582 g_raid_start_request(bp);
1584 g_raid_disk_done_request(bp);
1585 } else if (rv == EWOULDBLOCK) {
1586 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1587 g_raid_clean(vol, -1);
1588 if (bioq_first(&vol->v_inflight) == NULL &&
1590 t.tv_sec = g_raid_idle_threshold / 1000000;
1591 t.tv_usec = g_raid_idle_threshold % 1000000;
1592 timevaladd(&t, &vol->v_last_done);
1593 getmicrouptime(&now);
1594 if (timevalcmp(&t, &now, <= )) {
1595 G_RAID_TR_IDLE(vol->v_tr);
1596 vol->v_last_done = now;
1601 if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1602 g_raid_destroy_node(sc, 1); /* May not return. */
1607 g_raid_poll(struct g_raid_softc *sc)
1609 struct g_raid_event *ep;
1612 sx_xlock(&sc->sc_lock);
1613 mtx_lock(&sc->sc_queue_mtx);
1615 * First take a look at events.
1616 * This is important to handle events before any I/O requests.
1618 ep = TAILQ_FIRST(&sc->sc_events);
1620 TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1621 mtx_unlock(&sc->sc_queue_mtx);
1622 g_raid_handle_event(sc, ep);
1625 bp = bioq_takefirst(&sc->sc_queue);
1627 mtx_unlock(&sc->sc_queue_mtx);
1628 if (bp->bio_from == NULL ||
1629 bp->bio_from->geom != sc->sc_geom)
1630 g_raid_start_request(bp);
1632 g_raid_disk_done_request(bp);
1635 sx_xunlock(&sc->sc_lock);
1639 g_raid_launch_provider(struct g_raid_volume *vol)
1641 struct g_raid_disk *disk;
1642 struct g_raid_softc *sc;
1643 struct g_provider *pp;
1644 char name[G_RAID_MAX_VOLUMENAME];
1645 char announce_buf[80], buf1[32];
1649 sx_assert(&sc->sc_lock, SX_LOCKED);
1652 /* Try to name provider with volume name. */
1653 snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1654 if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1655 g_provider_by_name(name) != NULL) {
1656 /* Otherwise use sequential volume number. */
1657 snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1661 * Create a /dev/ar%d that the old ataraid(4) stack once
1662 * created as an alias for /dev/raid/r%d if requested.
1663 * This helps going from stable/7 ataraid devices to newer
1664 * FreeBSD releases. sbruno 07 MAY 2013
1667 if (ar_legacy_aliases) {
1668 snprintf(announce_buf, sizeof(announce_buf),
1669 "kern.devalias.%s", name);
1670 snprintf(buf1, sizeof(buf1),
1671 "ar%d", vol->v_global_id);
1672 setenv(announce_buf, buf1);
1675 pp = g_new_providerf(sc->sc_geom, "%s", name);
1677 pp->mediasize = vol->v_mediasize;
1678 pp->sectorsize = vol->v_sectorsize;
1680 pp->stripeoffset = 0;
1681 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1682 vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1683 vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1684 vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1685 if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1686 disk->d_consumer != NULL &&
1687 disk->d_consumer->provider != NULL) {
1688 pp->stripesize = disk->d_consumer->provider->stripesize;
1689 off = disk->d_consumer->provider->stripeoffset;
1690 pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1692 pp->stripeoffset %= off;
1694 if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1695 pp->stripesize *= (vol->v_disks_count - 1);
1696 pp->stripeoffset *= (vol->v_disks_count - 1);
1699 pp->stripesize = vol->v_strip_size;
1700 vol->v_provider = pp;
1701 g_error_provider(pp, 0);
1702 g_topology_unlock();
1703 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1704 pp->name, vol->v_name);
1708 g_raid_destroy_provider(struct g_raid_volume *vol)
1710 struct g_raid_softc *sc;
1711 struct g_provider *pp;
1712 struct bio *bp, *tmp;
1714 g_topology_assert_not();
1716 pp = vol->v_provider;
1717 KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1720 g_error_provider(pp, ENXIO);
1721 mtx_lock(&sc->sc_queue_mtx);
1722 TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1723 if (bp->bio_to != pp)
1725 bioq_remove(&sc->sc_queue, bp);
1726 g_io_deliver(bp, ENXIO);
1728 mtx_unlock(&sc->sc_queue_mtx);
1729 G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1730 pp->name, vol->v_name);
1731 g_wither_provider(pp, ENXIO);
1732 g_topology_unlock();
1733 vol->v_provider = NULL;
1737 * Update device state.
1740 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1742 struct g_raid_softc *sc;
1745 sx_assert(&sc->sc_lock, SX_XLOCKED);
1747 G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1748 g_raid_volume_event2str(event),
1751 case G_RAID_VOLUME_E_DOWN:
1752 if (vol->v_provider != NULL)
1753 g_raid_destroy_provider(vol);
1755 case G_RAID_VOLUME_E_UP:
1756 if (vol->v_provider == NULL)
1757 g_raid_launch_provider(vol);
1759 case G_RAID_VOLUME_E_START:
1761 G_RAID_TR_START(vol->v_tr);
1765 G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1769 /* Manage root mount release. */
1770 if (vol->v_starting) {
1771 vol->v_starting = 0;
1772 G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1773 root_mount_rel(vol->v_rootmount);
1774 vol->v_rootmount = NULL;
1776 if (vol->v_stopping && vol->v_provider_open == 0)
1777 g_raid_destroy_volume(vol);
1782 * Update subdisk state.
1785 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1787 struct g_raid_softc *sc;
1788 struct g_raid_volume *vol;
1791 vol = sd->sd_volume;
1792 sx_assert(&sc->sc_lock, SX_XLOCKED);
1794 G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1795 g_raid_subdisk_event2str(event),
1796 vol->v_name, sd->sd_pos,
1797 sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1799 G_RAID_TR_EVENT(vol->v_tr, sd, event);
1805 * Update disk state.
1808 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1810 struct g_raid_softc *sc;
1813 sx_assert(&sc->sc_lock, SX_XLOCKED);
1815 G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1816 g_raid_disk_event2str(event),
1817 g_raid_get_diskname(disk));
1820 G_RAID_MD_EVENT(sc->sc_md, disk, event);
1828 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1830 sx_assert(&sc->sc_lock, SX_XLOCKED);
1832 G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1833 g_raid_node_event2str(event));
1835 if (event == G_RAID_NODE_E_WAKE)
1838 G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1843 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1845 struct g_raid_volume *vol;
1846 struct g_raid_softc *sc;
1847 int dcw, opens, error = 0;
1849 g_topology_assert();
1850 sc = pp->geom->softc;
1852 KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1853 KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1855 G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1857 dcw = pp->acw + acw;
1859 g_topology_unlock();
1860 sx_xlock(&sc->sc_lock);
1861 /* Deny new opens while dying. */
1862 if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1867 g_raid_clean(vol, dcw);
1868 vol->v_provider_open += acr + acw + ace;
1869 /* Handle delayed node destruction. */
1870 if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1871 vol->v_provider_open == 0) {
1872 /* Count open volumes. */
1873 opens = g_raid_nopens(sc);
1875 sc->sc_stopping = G_RAID_DESTROY_HARD;
1876 /* Wake up worker to make it selfdestruct. */
1877 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1880 /* Handle open volume destruction. */
1881 if (vol->v_stopping && vol->v_provider_open == 0)
1882 g_raid_destroy_volume(vol);
1884 sx_xunlock(&sc->sc_lock);
1889 struct g_raid_softc *
1890 g_raid_create_node(struct g_class *mp,
1891 const char *name, struct g_raid_md_object *md)
1893 struct g_raid_softc *sc;
1897 g_topology_assert();
1898 G_RAID_DEBUG(1, "Creating array %s.", name);
1900 gp = g_new_geomf(mp, "%s", name);
1901 sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1902 gp->start = g_raid_start;
1903 gp->orphan = g_raid_orphan;
1904 gp->access = g_raid_access;
1905 gp->dumpconf = g_raid_dumpconf;
1910 TAILQ_INIT(&sc->sc_volumes);
1911 TAILQ_INIT(&sc->sc_disks);
1912 sx_init(&sc->sc_lock, "graid:lock");
1913 mtx_init(&sc->sc_queue_mtx, "graid:queue", NULL, MTX_DEF);
1914 TAILQ_INIT(&sc->sc_events);
1915 bioq_init(&sc->sc_queue);
1917 error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1920 G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1921 mtx_destroy(&sc->sc_queue_mtx);
1922 sx_destroy(&sc->sc_lock);
1923 g_destroy_geom(sc->sc_geom);
1928 G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1932 struct g_raid_volume *
1933 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1935 struct g_raid_volume *vol, *vol1;
1938 G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1939 vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1941 strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1942 vol->v_state = G_RAID_VOLUME_S_STARTING;
1943 vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1944 vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1945 vol->v_rotate_parity = 1;
1946 bioq_init(&vol->v_inflight);
1947 bioq_init(&vol->v_locked);
1948 LIST_INIT(&vol->v_locks);
1949 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1950 vol->v_subdisks[i].sd_softc = sc;
1951 vol->v_subdisks[i].sd_volume = vol;
1952 vol->v_subdisks[i].sd_pos = i;
1953 vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1956 /* Find free ID for this volume. */
1960 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1961 if (vol1->v_global_id == id)
1966 for (id = 0; ; id++) {
1967 LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1968 if (vol1->v_global_id == id)
1975 vol->v_global_id = id;
1976 LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1977 g_topology_unlock();
1979 /* Delay root mounting. */
1980 vol->v_rootmount = root_mount_hold("GRAID");
1981 G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1982 vol->v_starting = 1;
1983 TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1987 struct g_raid_disk *
1988 g_raid_create_disk(struct g_raid_softc *sc)
1990 struct g_raid_disk *disk;
1992 G_RAID_DEBUG1(1, sc, "Creating disk.");
1993 disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1995 disk->d_state = G_RAID_DISK_S_NONE;
1996 TAILQ_INIT(&disk->d_subdisks);
1997 TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
2001 int g_raid_start_volume(struct g_raid_volume *vol)
2003 struct g_raid_tr_class *class;
2004 struct g_raid_tr_object *obj;
2007 G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
2008 LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
2009 if (!class->trc_enable)
2011 G_RAID_DEBUG1(2, vol->v_softc,
2012 "Tasting volume %s for %s transformation.",
2013 vol->v_name, class->name);
2014 obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2016 obj->tro_class = class;
2017 obj->tro_volume = vol;
2018 status = G_RAID_TR_TASTE(obj, vol);
2019 if (status != G_RAID_TR_TASTE_FAIL)
2021 kobj_delete((kobj_t)obj, M_RAID);
2023 if (class == NULL) {
2024 G_RAID_DEBUG1(0, vol->v_softc,
2025 "No transformation module found for %s.",
2028 g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
2029 g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
2030 G_RAID_EVENT_VOLUME);
2033 G_RAID_DEBUG1(2, vol->v_softc,
2034 "Transformation module %s chosen for %s.",
2035 class->name, vol->v_name);
2041 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
2043 struct g_raid_volume *vol, *tmpv;
2044 struct g_raid_disk *disk, *tmpd;
2047 sc->sc_stopping = G_RAID_DESTROY_HARD;
2048 TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
2049 if (g_raid_destroy_volume(vol))
2054 TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
2055 if (g_raid_destroy_disk(disk))
2061 G_RAID_MD_FREE(sc->sc_md);
2062 kobj_delete((kobj_t)sc->sc_md, M_RAID);
2065 if (sc->sc_geom != NULL) {
2066 G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
2068 sc->sc_geom->softc = NULL;
2069 g_wither_geom(sc->sc_geom, ENXIO);
2070 g_topology_unlock();
2073 G_RAID_DEBUG(1, "Array destroyed.");
2075 g_raid_event_cancel(sc, sc);
2076 mtx_destroy(&sc->sc_queue_mtx);
2077 sx_xunlock(&sc->sc_lock);
2078 sx_destroy(&sc->sc_lock);
2079 wakeup(&sc->sc_stopping);
2081 curthread->td_pflags &= ~TDP_GEOM;
2082 G_RAID_DEBUG(1, "Thread exiting.");
2085 /* Wake up worker to make it selfdestruct. */
2086 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2092 g_raid_destroy_volume(struct g_raid_volume *vol)
2094 struct g_raid_softc *sc;
2095 struct g_raid_disk *disk;
2099 G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
2100 vol->v_stopping = 1;
2101 if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
2103 G_RAID_TR_STOP(vol->v_tr);
2106 vol->v_state = G_RAID_VOLUME_S_STOPPED;
2108 if (g_raid_event_check(sc, vol) != 0)
2110 if (vol->v_provider != NULL)
2112 if (vol->v_provider_open != 0)
2115 G_RAID_TR_FREE(vol->v_tr);
2116 kobj_delete((kobj_t)vol->v_tr, M_RAID);
2119 if (vol->v_rootmount)
2120 root_mount_rel(vol->v_rootmount);
2122 LIST_REMOVE(vol, v_global_next);
2123 g_topology_unlock();
2124 TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
2125 for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
2126 g_raid_event_cancel(sc, &vol->v_subdisks[i]);
2127 disk = vol->v_subdisks[i].sd_disk;
2130 TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
2132 G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
2134 G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
2135 g_raid_event_cancel(sc, vol);
2137 if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
2138 /* Wake up worker to let it selfdestruct. */
2139 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2145 g_raid_destroy_disk(struct g_raid_disk *disk)
2147 struct g_raid_softc *sc;
2148 struct g_raid_subdisk *sd, *tmp;
2151 G_RAID_DEBUG1(2, sc, "Destroying disk.");
2152 if (disk->d_consumer) {
2153 g_raid_kill_consumer(sc, disk->d_consumer);
2154 disk->d_consumer = NULL;
2156 TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
2157 g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
2158 g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
2159 G_RAID_EVENT_SUBDISK);
2160 TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
2163 TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
2165 G_RAID_MD_FREE_DISK(sc->sc_md, disk);
2166 g_raid_event_cancel(sc, disk);
2172 g_raid_destroy(struct g_raid_softc *sc, int how)
2176 g_topology_assert_not();
2179 sx_assert(&sc->sc_lock, SX_XLOCKED);
2181 /* Count open volumes. */
2182 opens = g_raid_nopens(sc);
2184 /* React on some opened volumes. */
2187 case G_RAID_DESTROY_SOFT:
2188 G_RAID_DEBUG1(1, sc,
2189 "%d volumes are still open.",
2192 case G_RAID_DESTROY_DELAYED:
2193 G_RAID_DEBUG1(1, sc,
2194 "Array will be destroyed on last close.");
2195 sc->sc_stopping = G_RAID_DESTROY_DELAYED;
2197 case G_RAID_DESTROY_HARD:
2198 G_RAID_DEBUG1(1, sc,
2199 "%d volumes are still open.",
2204 /* Mark node for destruction. */
2205 sc->sc_stopping = G_RAID_DESTROY_HARD;
2206 /* Wake up worker to let it selfdestruct. */
2207 g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
2208 /* Sleep until node destroyed. */
2209 sx_sleep(&sc->sc_stopping, &sc->sc_lock,
2210 PRIBIO | PDROP, "r:destroy", 0);
2215 g_raid_taste_orphan(struct g_consumer *cp)
2218 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
2219 cp->provider->name));
2222 static struct g_geom *
2223 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
2225 struct g_consumer *cp;
2226 struct g_geom *gp, *geom;
2227 struct g_raid_md_class *class;
2228 struct g_raid_md_object *obj;
2231 g_topology_assert();
2232 g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
2235 G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
2237 gp = g_new_geomf(mp, "raid:taste");
2239 * This orphan function should be never called.
2241 gp->orphan = g_raid_taste_orphan;
2242 cp = g_new_consumer(gp);
2246 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2247 if (!class->mdc_enable)
2249 G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2250 pp->name, class->name);
2251 obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2253 obj->mdo_class = class;
2254 status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2255 if (status != G_RAID_MD_TASTE_NEW)
2256 kobj_delete((kobj_t)obj, M_RAID);
2257 if (status != G_RAID_MD_TASTE_FAIL)
2262 g_destroy_consumer(cp);
2264 G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2269 g_raid_create_node_format(const char *format, struct gctl_req *req,
2272 struct g_raid_md_class *class;
2273 struct g_raid_md_object *obj;
2276 G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2277 LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2278 if (strcasecmp(class->name, format) == 0)
2281 if (class == NULL) {
2282 G_RAID_DEBUG(1, "No support for %s metadata.", format);
2283 return (G_RAID_MD_TASTE_FAIL);
2285 obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2287 obj->mdo_class = class;
2288 status = G_RAID_MD_CREATE_REQ(obj, &g_raid_class, req, gp);
2289 if (status != G_RAID_MD_TASTE_NEW)
2290 kobj_delete((kobj_t)obj, M_RAID);
2295 g_raid_destroy_geom(struct gctl_req *req __unused,
2296 struct g_class *mp __unused, struct g_geom *gp)
2298 struct g_raid_softc *sc;
2301 g_topology_unlock();
2303 sx_xlock(&sc->sc_lock);
2305 error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2307 sx_xunlock(&sc->sc_lock);
2312 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2313 struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2316 if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2319 G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2322 void g_raid_fail_disk(struct g_raid_softc *sc,
2323 struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2329 G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2332 if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2333 G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2334 "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2338 G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2342 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2343 struct g_consumer *cp, struct g_provider *pp)
2345 struct g_raid_softc *sc;
2346 struct g_raid_volume *vol;
2347 struct g_raid_subdisk *sd;
2348 struct g_raid_disk *disk;
2351 g_topology_assert();
2358 g_topology_unlock();
2359 sx_xlock(&sc->sc_lock);
2360 sbuf_printf(sb, "%s<descr>%s %s volume</descr>\n", indent,
2361 sc->sc_md->mdo_class->name,
2362 g_raid_volume_level2str(vol->v_raid_level,
2363 vol->v_raid_level_qualifier));
2364 sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2366 sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2367 g_raid_volume_level2str(vol->v_raid_level,
2368 vol->v_raid_level_qualifier));
2370 "%s<Transformation>%s</Transformation>\n", indent,
2371 vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2372 sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2373 vol->v_disks_count);
2374 sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2376 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2377 g_raid_volume_state2str(vol->v_state));
2378 sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2379 vol->v_dirty ? "Yes" : "No");
2380 sbuf_printf(sb, "%s<Subdisks>", indent);
2381 for (i = 0; i < vol->v_disks_count; i++) {
2382 sd = &vol->v_subdisks[i];
2383 if (sd->sd_disk != NULL &&
2384 sd->sd_disk->d_consumer != NULL) {
2385 sbuf_printf(sb, "%s ",
2386 g_raid_get_diskname(sd->sd_disk));
2388 sbuf_printf(sb, "NONE ");
2390 sbuf_printf(sb, "(%s",
2391 g_raid_subdisk_state2str(sd->sd_state));
2392 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2393 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2394 sbuf_printf(sb, " %d%%",
2395 (int)(sd->sd_rebuild_pos * 100 /
2398 sbuf_printf(sb, ")");
2399 if (i + 1 < vol->v_disks_count)
2400 sbuf_printf(sb, ", ");
2402 sbuf_printf(sb, "</Subdisks>\n");
2403 sx_xunlock(&sc->sc_lock);
2405 } else if (cp != NULL) {
2409 g_topology_unlock();
2410 sx_xlock(&sc->sc_lock);
2411 sbuf_printf(sb, "%s<State>%s", indent,
2412 g_raid_disk_state2str(disk->d_state));
2413 if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2414 sbuf_printf(sb, " (");
2415 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2416 sbuf_printf(sb, "%s",
2417 g_raid_subdisk_state2str(sd->sd_state));
2418 if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2419 sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2420 sbuf_printf(sb, " %d%%",
2421 (int)(sd->sd_rebuild_pos * 100 /
2424 if (TAILQ_NEXT(sd, sd_next))
2425 sbuf_printf(sb, ", ");
2427 sbuf_printf(sb, ")");
2429 sbuf_printf(sb, "</State>\n");
2430 sbuf_printf(sb, "%s<Subdisks>", indent);
2431 TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2432 sbuf_printf(sb, "r%d(%s):%d@%ju",
2433 sd->sd_volume->v_global_id,
2434 sd->sd_volume->v_name,
2435 sd->sd_pos, sd->sd_offset);
2436 if (TAILQ_NEXT(sd, sd_next))
2437 sbuf_printf(sb, ", ");
2439 sbuf_printf(sb, "</Subdisks>\n");
2440 sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2442 sx_xunlock(&sc->sc_lock);
2445 g_topology_unlock();
2446 sx_xlock(&sc->sc_lock);
2448 sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2449 sc->sc_md->mdo_class->name);
2451 if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2453 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2454 if (vol->v_state < s)
2457 sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2458 g_raid_volume_state2str(s));
2460 sx_xunlock(&sc->sc_lock);
2466 g_raid_shutdown_post_sync(void *arg, int howto)
2469 struct g_geom *gp, *gp2;
2470 struct g_raid_softc *sc;
2471 struct g_raid_volume *vol;
2477 g_raid_shutdown = 1;
2478 LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2479 if ((sc = gp->softc) == NULL)
2481 g_topology_unlock();
2482 sx_xlock(&sc->sc_lock);
2483 TAILQ_FOREACH(vol, &sc->sc_volumes, v_next)
2484 g_raid_clean(vol, -1);
2486 error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2488 sx_xunlock(&sc->sc_lock);
2491 g_topology_unlock();
2496 g_raid_init(struct g_class *mp)
2499 g_raid_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
2500 g_raid_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
2501 if (g_raid_post_sync == NULL)
2502 G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2507 g_raid_fini(struct g_class *mp)
2510 if (g_raid_post_sync != NULL)
2511 EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid_post_sync);
2516 g_raid_md_modevent(module_t mod, int type, void *arg)
2518 struct g_raid_md_class *class, *c, *nc;
2525 c = LIST_FIRST(&g_raid_md_classes);
2526 if (c == NULL || c->mdc_priority > class->mdc_priority)
2527 LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2529 while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2530 nc->mdc_priority < class->mdc_priority)
2532 LIST_INSERT_AFTER(c, class, mdc_list);
2535 g_retaste(&g_raid_class);
2538 LIST_REMOVE(class, mdc_list);
2549 g_raid_tr_modevent(module_t mod, int type, void *arg)
2551 struct g_raid_tr_class *class, *c, *nc;
2558 c = LIST_FIRST(&g_raid_tr_classes);
2559 if (c == NULL || c->trc_priority > class->trc_priority)
2560 LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2562 while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2563 nc->trc_priority < class->trc_priority)
2565 LIST_INSERT_AFTER(c, class, trc_list);
2569 LIST_REMOVE(class, trc_list);
2580 * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2581 * to reduce module priority, allowing submodules to register them first.
2583 static moduledata_t g_raid_mod = {
2588 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2589 MODULE_VERSION(geom_raid, 0);