xref: /freebsd/sys/geom/raid/g_raid.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
1 /*-
2  * Copyright (c) 2010 Alexander Motin <mav@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
13  *
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
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
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>
35 #include <sys/lock.h>
36 #include <sys/mutex.h>
37 #include <sys/bio.h>
38 #include <sys/sysctl.h>
39 #include <sys/malloc.h>
40 #include <sys/eventhandler.h>
41 #include <vm/uma.h>
42 #include <geom/geom.h>
43 #include <sys/proc.h>
44 #include <sys/kthread.h>
45 #include <sys/sched.h>
46 #include <geom/raid/g_raid.h>
47 #include "g_raid_md_if.h"
48 #include "g_raid_tr_if.h"
49 
50 static MALLOC_DEFINE(M_RAID, "raid_data", "GEOM_RAID Data");
51 
52 SYSCTL_DECL(_kern_geom);
53 SYSCTL_NODE(_kern_geom, OID_AUTO, raid, CTLFLAG_RW, 0, "GEOM_RAID stuff");
54 u_int g_raid_aggressive_spare = 0;
55 TUNABLE_INT("kern.geom.raid.aggressive_spare", &g_raid_aggressive_spare);
56 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, aggressive_spare, CTLFLAG_RW,
57     &g_raid_aggressive_spare, 0, "Use disks without metadata as spare");
58 u_int g_raid_debug = 0;
59 TUNABLE_INT("kern.geom.raid.debug", &g_raid_debug);
60 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, debug, CTLFLAG_RW, &g_raid_debug, 0,
61     "Debug level");
62 int g_raid_read_err_thresh = 10;
63 TUNABLE_INT("kern.geom.raid.read_err_thresh", &g_raid_read_err_thresh);
64 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, read_err_thresh, CTLFLAG_RW,
65     &g_raid_read_err_thresh, 0,
66     "Number of read errors equated to disk failure");
67 u_int g_raid_start_timeout = 30;
68 TUNABLE_INT("kern.geom.raid.start_timeout", &g_raid_start_timeout);
69 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, start_timeout, CTLFLAG_RW,
70     &g_raid_start_timeout, 0,
71     "Time to wait for all array components");
72 static u_int g_raid_clean_time = 5;
73 TUNABLE_INT("kern.geom.raid.clean_time", &g_raid_clean_time);
74 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, clean_time, CTLFLAG_RW,
75     &g_raid_clean_time, 0, "Mark volume as clean when idling");
76 static u_int g_raid_disconnect_on_failure = 1;
77 TUNABLE_INT("kern.geom.raid.disconnect_on_failure",
78     &g_raid_disconnect_on_failure);
79 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, disconnect_on_failure, CTLFLAG_RW,
80     &g_raid_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
81 static u_int g_raid_name_format = 0;
82 TUNABLE_INT("kern.geom.raid.name_format", &g_raid_name_format);
83 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, name_format, CTLFLAG_RW,
84     &g_raid_name_format, 0, "Providers name format.");
85 static u_int g_raid_idle_threshold = 1000000;
86 TUNABLE_INT("kern.geom.raid.idle_threshold", &g_raid_idle_threshold);
87 SYSCTL_UINT(_kern_geom_raid, OID_AUTO, idle_threshold, CTLFLAG_RW,
88     &g_raid_idle_threshold, 1000000,
89     "Time in microseconds to consider a volume idle.");
90 
91 #define	MSLEEP(rv, ident, mtx, priority, wmesg, timeout)	do {	\
92 	G_RAID_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));		\
93 	rv = msleep((ident), (mtx), (priority), (wmesg), (timeout));	\
94 	G_RAID_DEBUG(4, "%s: Woken up %p.", __func__, (ident));		\
95 } while (0)
96 
97 LIST_HEAD(, g_raid_md_class) g_raid_md_classes =
98     LIST_HEAD_INITIALIZER(g_raid_md_classes);
99 
100 LIST_HEAD(, g_raid_tr_class) g_raid_tr_classes =
101     LIST_HEAD_INITIALIZER(g_raid_tr_classes);
102 
103 LIST_HEAD(, g_raid_volume) g_raid_volumes =
104     LIST_HEAD_INITIALIZER(g_raid_volumes);
105 
106 static eventhandler_tag g_raid_pre_sync = NULL;
107 static int g_raid_started = 0;
108 
109 static int g_raid_destroy_geom(struct gctl_req *req, struct g_class *mp,
110     struct g_geom *gp);
111 static g_taste_t g_raid_taste;
112 static void g_raid_init(struct g_class *mp);
113 static void g_raid_fini(struct g_class *mp);
114 
115 struct g_class g_raid_class = {
116 	.name = G_RAID_CLASS_NAME,
117 	.version = G_VERSION,
118 	.ctlreq = g_raid_ctl,
119 	.taste = g_raid_taste,
120 	.destroy_geom = g_raid_destroy_geom,
121 	.init = g_raid_init,
122 	.fini = g_raid_fini
123 };
124 
125 static void g_raid_destroy_provider(struct g_raid_volume *vol);
126 static int g_raid_update_disk(struct g_raid_disk *disk, u_int event);
127 static int g_raid_update_subdisk(struct g_raid_subdisk *subdisk, u_int event);
128 static int g_raid_update_volume(struct g_raid_volume *vol, u_int event);
129 static int g_raid_update_node(struct g_raid_softc *sc, u_int event);
130 static void g_raid_dumpconf(struct sbuf *sb, const char *indent,
131     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
132 static void g_raid_start(struct bio *bp);
133 static void g_raid_start_request(struct bio *bp);
134 static void g_raid_disk_done(struct bio *bp);
135 static void g_raid_poll(struct g_raid_softc *sc);
136 
137 static const char *
138 g_raid_node_event2str(int event)
139 {
140 
141 	switch (event) {
142 	case G_RAID_NODE_E_WAKE:
143 		return ("WAKE");
144 	case G_RAID_NODE_E_START:
145 		return ("START");
146 	default:
147 		return ("INVALID");
148 	}
149 }
150 
151 const char *
152 g_raid_disk_state2str(int state)
153 {
154 
155 	switch (state) {
156 	case G_RAID_DISK_S_NONE:
157 		return ("NONE");
158 	case G_RAID_DISK_S_OFFLINE:
159 		return ("OFFLINE");
160 	case G_RAID_DISK_S_FAILED:
161 		return ("FAILED");
162 	case G_RAID_DISK_S_STALE_FAILED:
163 		return ("STALE_FAILED");
164 	case G_RAID_DISK_S_SPARE:
165 		return ("SPARE");
166 	case G_RAID_DISK_S_STALE:
167 		return ("STALE");
168 	case G_RAID_DISK_S_ACTIVE:
169 		return ("ACTIVE");
170 	default:
171 		return ("INVALID");
172 	}
173 }
174 
175 static const char *
176 g_raid_disk_event2str(int event)
177 {
178 
179 	switch (event) {
180 	case G_RAID_DISK_E_DISCONNECTED:
181 		return ("DISCONNECTED");
182 	default:
183 		return ("INVALID");
184 	}
185 }
186 
187 const char *
188 g_raid_subdisk_state2str(int state)
189 {
190 
191 	switch (state) {
192 	case G_RAID_SUBDISK_S_NONE:
193 		return ("NONE");
194 	case G_RAID_SUBDISK_S_FAILED:
195 		return ("FAILED");
196 	case G_RAID_SUBDISK_S_NEW:
197 		return ("NEW");
198 	case G_RAID_SUBDISK_S_REBUILD:
199 		return ("REBUILD");
200 	case G_RAID_SUBDISK_S_UNINITIALIZED:
201 		return ("UNINITIALIZED");
202 	case G_RAID_SUBDISK_S_STALE:
203 		return ("STALE");
204 	case G_RAID_SUBDISK_S_RESYNC:
205 		return ("RESYNC");
206 	case G_RAID_SUBDISK_S_ACTIVE:
207 		return ("ACTIVE");
208 	default:
209 		return ("INVALID");
210 	}
211 }
212 
213 static const char *
214 g_raid_subdisk_event2str(int event)
215 {
216 
217 	switch (event) {
218 	case G_RAID_SUBDISK_E_NEW:
219 		return ("NEW");
220 	case G_RAID_SUBDISK_E_DISCONNECTED:
221 		return ("DISCONNECTED");
222 	default:
223 		return ("INVALID");
224 	}
225 }
226 
227 const char *
228 g_raid_volume_state2str(int state)
229 {
230 
231 	switch (state) {
232 	case G_RAID_VOLUME_S_STARTING:
233 		return ("STARTING");
234 	case G_RAID_VOLUME_S_BROKEN:
235 		return ("BROKEN");
236 	case G_RAID_VOLUME_S_DEGRADED:
237 		return ("DEGRADED");
238 	case G_RAID_VOLUME_S_SUBOPTIMAL:
239 		return ("SUBOPTIMAL");
240 	case G_RAID_VOLUME_S_OPTIMAL:
241 		return ("OPTIMAL");
242 	case G_RAID_VOLUME_S_UNSUPPORTED:
243 		return ("UNSUPPORTED");
244 	case G_RAID_VOLUME_S_STOPPED:
245 		return ("STOPPED");
246 	default:
247 		return ("INVALID");
248 	}
249 }
250 
251 static const char *
252 g_raid_volume_event2str(int event)
253 {
254 
255 	switch (event) {
256 	case G_RAID_VOLUME_E_UP:
257 		return ("UP");
258 	case G_RAID_VOLUME_E_DOWN:
259 		return ("DOWN");
260 	case G_RAID_VOLUME_E_START:
261 		return ("START");
262 	case G_RAID_VOLUME_E_STARTMD:
263 		return ("STARTMD");
264 	default:
265 		return ("INVALID");
266 	}
267 }
268 
269 const char *
270 g_raid_volume_level2str(int level, int qual)
271 {
272 
273 	switch (level) {
274 	case G_RAID_VOLUME_RL_RAID0:
275 		return ("RAID0");
276 	case G_RAID_VOLUME_RL_RAID1:
277 		return ("RAID1");
278 	case G_RAID_VOLUME_RL_RAID3:
279 		return ("RAID3");
280 	case G_RAID_VOLUME_RL_RAID4:
281 		return ("RAID4");
282 	case G_RAID_VOLUME_RL_RAID5:
283 		return ("RAID5");
284 	case G_RAID_VOLUME_RL_RAID6:
285 		return ("RAID6");
286 	case G_RAID_VOLUME_RL_RAID1E:
287 		return ("RAID1E");
288 	case G_RAID_VOLUME_RL_SINGLE:
289 		return ("SINGLE");
290 	case G_RAID_VOLUME_RL_CONCAT:
291 		return ("CONCAT");
292 	case G_RAID_VOLUME_RL_RAID5E:
293 		return ("RAID5E");
294 	case G_RAID_VOLUME_RL_RAID5EE:
295 		return ("RAID5EE");
296 	default:
297 		return ("UNKNOWN");
298 	}
299 }
300 
301 int
302 g_raid_volume_str2level(const char *str, int *level, int *qual)
303 {
304 
305 	*level = G_RAID_VOLUME_RL_UNKNOWN;
306 	*qual = G_RAID_VOLUME_RLQ_NONE;
307 	if (strcasecmp(str, "RAID0") == 0)
308 		*level = G_RAID_VOLUME_RL_RAID0;
309 	else if (strcasecmp(str, "RAID1") == 0)
310 		*level = G_RAID_VOLUME_RL_RAID1;
311 	else if (strcasecmp(str, "RAID3") == 0)
312 		*level = G_RAID_VOLUME_RL_RAID3;
313 	else if (strcasecmp(str, "RAID4") == 0)
314 		*level = G_RAID_VOLUME_RL_RAID4;
315 	else if (strcasecmp(str, "RAID5") == 0)
316 		*level = G_RAID_VOLUME_RL_RAID5;
317 	else if (strcasecmp(str, "RAID6") == 0)
318 		*level = G_RAID_VOLUME_RL_RAID6;
319 	else if (strcasecmp(str, "RAID10") == 0 ||
320 		 strcasecmp(str, "RAID1E") == 0)
321 		*level = G_RAID_VOLUME_RL_RAID1E;
322 	else if (strcasecmp(str, "SINGLE") == 0)
323 		*level = G_RAID_VOLUME_RL_SINGLE;
324 	else if (strcasecmp(str, "CONCAT") == 0)
325 		*level = G_RAID_VOLUME_RL_CONCAT;
326 	else if (strcasecmp(str, "RAID5E") == 0)
327 		*level = G_RAID_VOLUME_RL_RAID5E;
328 	else if (strcasecmp(str, "RAID5EE") == 0)
329 		*level = G_RAID_VOLUME_RL_RAID5EE;
330 	else
331 		return (-1);
332 	return (0);
333 }
334 
335 const char *
336 g_raid_get_diskname(struct g_raid_disk *disk)
337 {
338 
339 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
340 		return ("[unknown]");
341 	return (disk->d_consumer->provider->name);
342 }
343 
344 void
345 g_raid_report_disk_state(struct g_raid_disk *disk)
346 {
347 	struct g_raid_subdisk *sd;
348 	int len, state;
349 	uint32_t s;
350 
351 	if (disk->d_consumer == NULL)
352 		return;
353 	if (disk->d_state == G_RAID_DISK_S_FAILED ||
354 	    disk->d_state == G_RAID_DISK_S_STALE_FAILED) {
355 		s = G_STATE_FAILED;
356 	} else {
357 		state = G_RAID_SUBDISK_S_ACTIVE;
358 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
359 			if (sd->sd_state < state)
360 				state = sd->sd_state;
361 		}
362 		if (state == G_RAID_SUBDISK_S_FAILED)
363 			s = G_STATE_FAILED;
364 		else if (state == G_RAID_SUBDISK_S_NEW ||
365 		    state == G_RAID_SUBDISK_S_REBUILD)
366 			s = G_STATE_REBUILD;
367 		else if (state == G_RAID_SUBDISK_S_STALE ||
368 		    state == G_RAID_SUBDISK_S_RESYNC)
369 			s = G_STATE_RESYNC;
370 		else
371 			s = G_STATE_ACTIVE;
372 	}
373 	len = sizeof(s);
374 	g_io_getattr("GEOM::setstate", disk->d_consumer, &len, &s);
375 	G_RAID_DEBUG1(2, disk->d_softc, "Disk %s state reported as %d.",
376 	    g_raid_get_diskname(disk), s);
377 }
378 
379 void
380 g_raid_change_disk_state(struct g_raid_disk *disk, int state)
381 {
382 
383 	G_RAID_DEBUG1(0, disk->d_softc, "Disk %s state changed from %s to %s.",
384 	    g_raid_get_diskname(disk),
385 	    g_raid_disk_state2str(disk->d_state),
386 	    g_raid_disk_state2str(state));
387 	disk->d_state = state;
388 	g_raid_report_disk_state(disk);
389 }
390 
391 void
392 g_raid_change_subdisk_state(struct g_raid_subdisk *sd, int state)
393 {
394 
395 	G_RAID_DEBUG1(0, sd->sd_softc,
396 	    "Subdisk %s:%d-%s state changed from %s to %s.",
397 	    sd->sd_volume->v_name, sd->sd_pos,
398 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]",
399 	    g_raid_subdisk_state2str(sd->sd_state),
400 	    g_raid_subdisk_state2str(state));
401 	sd->sd_state = state;
402 	if (sd->sd_disk)
403 		g_raid_report_disk_state(sd->sd_disk);
404 }
405 
406 void
407 g_raid_change_volume_state(struct g_raid_volume *vol, int state)
408 {
409 
410 	G_RAID_DEBUG1(0, vol->v_softc,
411 	    "Volume %s state changed from %s to %s.",
412 	    vol->v_name,
413 	    g_raid_volume_state2str(vol->v_state),
414 	    g_raid_volume_state2str(state));
415 	vol->v_state = state;
416 }
417 
418 /*
419  * --- Events handling functions ---
420  * Events in geom_raid are used to maintain subdisks and volumes status
421  * from one thread to simplify locking.
422  */
423 static void
424 g_raid_event_free(struct g_raid_event *ep)
425 {
426 
427 	free(ep, M_RAID);
428 }
429 
430 int
431 g_raid_event_send(void *arg, int event, int flags)
432 {
433 	struct g_raid_softc *sc;
434 	struct g_raid_event *ep;
435 	int error;
436 
437 	if ((flags & G_RAID_EVENT_VOLUME) != 0) {
438 		sc = ((struct g_raid_volume *)arg)->v_softc;
439 	} else if ((flags & G_RAID_EVENT_DISK) != 0) {
440 		sc = ((struct g_raid_disk *)arg)->d_softc;
441 	} else if ((flags & G_RAID_EVENT_SUBDISK) != 0) {
442 		sc = ((struct g_raid_subdisk *)arg)->sd_softc;
443 	} else {
444 		sc = arg;
445 	}
446 	ep = malloc(sizeof(*ep), M_RAID,
447 	    sx_xlocked(&sc->sc_lock) ? M_WAITOK : M_NOWAIT);
448 	if (ep == NULL)
449 		return (ENOMEM);
450 	ep->e_tgt = arg;
451 	ep->e_event = event;
452 	ep->e_flags = flags;
453 	ep->e_error = 0;
454 	G_RAID_DEBUG1(4, sc, "Sending event %p. Waking up %p.", ep, sc);
455 	mtx_lock(&sc->sc_queue_mtx);
456 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
457 	mtx_unlock(&sc->sc_queue_mtx);
458 	wakeup(sc);
459 
460 	if ((flags & G_RAID_EVENT_WAIT) == 0)
461 		return (0);
462 
463 	sx_assert(&sc->sc_lock, SX_XLOCKED);
464 	G_RAID_DEBUG1(4, sc, "Sleeping on %p.", ep);
465 	sx_xunlock(&sc->sc_lock);
466 	while ((ep->e_flags & G_RAID_EVENT_DONE) == 0) {
467 		mtx_lock(&sc->sc_queue_mtx);
468 		MSLEEP(error, ep, &sc->sc_queue_mtx, PRIBIO | PDROP, "m:event",
469 		    hz * 5);
470 	}
471 	error = ep->e_error;
472 	g_raid_event_free(ep);
473 	sx_xlock(&sc->sc_lock);
474 	return (error);
475 }
476 
477 static void
478 g_raid_event_cancel(struct g_raid_softc *sc, void *tgt)
479 {
480 	struct g_raid_event *ep, *tmpep;
481 
482 	sx_assert(&sc->sc_lock, SX_XLOCKED);
483 
484 	mtx_lock(&sc->sc_queue_mtx);
485 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
486 		if (ep->e_tgt != tgt)
487 			continue;
488 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
489 		if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0)
490 			g_raid_event_free(ep);
491 		else {
492 			ep->e_error = ECANCELED;
493 			wakeup(ep);
494 		}
495 	}
496 	mtx_unlock(&sc->sc_queue_mtx);
497 }
498 
499 static int
500 g_raid_event_check(struct g_raid_softc *sc, void *tgt)
501 {
502 	struct g_raid_event *ep;
503 	int	res = 0;
504 
505 	sx_assert(&sc->sc_lock, SX_XLOCKED);
506 
507 	mtx_lock(&sc->sc_queue_mtx);
508 	TAILQ_FOREACH(ep, &sc->sc_events, e_next) {
509 		if (ep->e_tgt != tgt)
510 			continue;
511 		res = 1;
512 		break;
513 	}
514 	mtx_unlock(&sc->sc_queue_mtx);
515 	return (res);
516 }
517 
518 /*
519  * Return the number of disks in given state.
520  * If state is equal to -1, count all connected disks.
521  */
522 u_int
523 g_raid_ndisks(struct g_raid_softc *sc, int state)
524 {
525 	struct g_raid_disk *disk;
526 	u_int n;
527 
528 	sx_assert(&sc->sc_lock, SX_LOCKED);
529 
530 	n = 0;
531 	TAILQ_FOREACH(disk, &sc->sc_disks, d_next) {
532 		if (disk->d_state == state || state == -1)
533 			n++;
534 	}
535 	return (n);
536 }
537 
538 /*
539  * Return the number of subdisks in given state.
540  * If state is equal to -1, count all connected disks.
541  */
542 u_int
543 g_raid_nsubdisks(struct g_raid_volume *vol, int state)
544 {
545 	struct g_raid_subdisk *subdisk;
546 	struct g_raid_softc *sc;
547 	u_int i, n ;
548 
549 	sc = vol->v_softc;
550 	sx_assert(&sc->sc_lock, SX_LOCKED);
551 
552 	n = 0;
553 	for (i = 0; i < vol->v_disks_count; i++) {
554 		subdisk = &vol->v_subdisks[i];
555 		if ((state == -1 &&
556 		     subdisk->sd_state != G_RAID_SUBDISK_S_NONE) ||
557 		    subdisk->sd_state == state)
558 			n++;
559 	}
560 	return (n);
561 }
562 
563 /*
564  * Return the first subdisk in given state.
565  * If state is equal to -1, then the first connected disks.
566  */
567 struct g_raid_subdisk *
568 g_raid_get_subdisk(struct g_raid_volume *vol, int state)
569 {
570 	struct g_raid_subdisk *sd;
571 	struct g_raid_softc *sc;
572 	u_int i;
573 
574 	sc = vol->v_softc;
575 	sx_assert(&sc->sc_lock, SX_LOCKED);
576 
577 	for (i = 0; i < vol->v_disks_count; i++) {
578 		sd = &vol->v_subdisks[i];
579 		if ((state == -1 &&
580 		     sd->sd_state != G_RAID_SUBDISK_S_NONE) ||
581 		    sd->sd_state == state)
582 			return (sd);
583 	}
584 	return (NULL);
585 }
586 
587 struct g_consumer *
588 g_raid_open_consumer(struct g_raid_softc *sc, const char *name)
589 {
590 	struct g_consumer *cp;
591 	struct g_provider *pp;
592 
593 	g_topology_assert();
594 
595 	if (strncmp(name, "/dev/", 5) == 0)
596 		name += 5;
597 	pp = g_provider_by_name(name);
598 	if (pp == NULL)
599 		return (NULL);
600 	cp = g_new_consumer(sc->sc_geom);
601 	if (g_attach(cp, pp) != 0) {
602 		g_destroy_consumer(cp);
603 		return (NULL);
604 	}
605 	if (g_access(cp, 1, 1, 1) != 0) {
606 		g_detach(cp);
607 		g_destroy_consumer(cp);
608 		return (NULL);
609 	}
610 	return (cp);
611 }
612 
613 static u_int
614 g_raid_nrequests(struct g_raid_softc *sc, struct g_consumer *cp)
615 {
616 	struct bio *bp;
617 	u_int nreqs = 0;
618 
619 	mtx_lock(&sc->sc_queue_mtx);
620 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
621 		if (bp->bio_from == cp)
622 			nreqs++;
623 	}
624 	mtx_unlock(&sc->sc_queue_mtx);
625 	return (nreqs);
626 }
627 
628 u_int
629 g_raid_nopens(struct g_raid_softc *sc)
630 {
631 	struct g_raid_volume *vol;
632 	u_int opens;
633 
634 	opens = 0;
635 	TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
636 		if (vol->v_provider_open != 0)
637 			opens++;
638 	}
639 	return (opens);
640 }
641 
642 static int
643 g_raid_consumer_is_busy(struct g_raid_softc *sc, struct g_consumer *cp)
644 {
645 
646 	if (cp->index > 0) {
647 		G_RAID_DEBUG1(2, sc,
648 		    "I/O requests for %s exist, can't destroy it now.",
649 		    cp->provider->name);
650 		return (1);
651 	}
652 	if (g_raid_nrequests(sc, cp) > 0) {
653 		G_RAID_DEBUG1(2, sc,
654 		    "I/O requests for %s in queue, can't destroy it now.",
655 		    cp->provider->name);
656 		return (1);
657 	}
658 	return (0);
659 }
660 
661 static void
662 g_raid_destroy_consumer(void *arg, int flags __unused)
663 {
664 	struct g_consumer *cp;
665 
666 	g_topology_assert();
667 
668 	cp = arg;
669 	G_RAID_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
670 	g_detach(cp);
671 	g_destroy_consumer(cp);
672 }
673 
674 void
675 g_raid_kill_consumer(struct g_raid_softc *sc, struct g_consumer *cp)
676 {
677 	struct g_provider *pp;
678 	int retaste_wait;
679 
680 	g_topology_assert_not();
681 
682 	g_topology_lock();
683 	cp->private = NULL;
684 	if (g_raid_consumer_is_busy(sc, cp))
685 		goto out;
686 	pp = cp->provider;
687 	retaste_wait = 0;
688 	if (cp->acw == 1) {
689 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
690 			retaste_wait = 1;
691 	}
692 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
693 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
694 	if (retaste_wait) {
695 		/*
696 		 * After retaste event was send (inside g_access()), we can send
697 		 * event to detach and destroy consumer.
698 		 * A class, which has consumer to the given provider connected
699 		 * will not receive retaste event for the provider.
700 		 * This is the way how I ignore retaste events when I close
701 		 * consumers opened for write: I detach and destroy consumer
702 		 * after retaste event is sent.
703 		 */
704 		g_post_event(g_raid_destroy_consumer, cp, M_WAITOK, NULL);
705 		goto out;
706 	}
707 	G_RAID_DEBUG(1, "Consumer %s destroyed.", pp->name);
708 	g_detach(cp);
709 	g_destroy_consumer(cp);
710 out:
711 	g_topology_unlock();
712 }
713 
714 static void
715 g_raid_orphan(struct g_consumer *cp)
716 {
717 	struct g_raid_disk *disk;
718 
719 	g_topology_assert();
720 
721 	disk = cp->private;
722 	if (disk == NULL)
723 		return;
724 	g_raid_event_send(disk, G_RAID_DISK_E_DISCONNECTED,
725 	    G_RAID_EVENT_DISK);
726 }
727 
728 static int
729 g_raid_clean(struct g_raid_volume *vol, int acw)
730 {
731 	struct g_raid_softc *sc;
732 	int timeout;
733 
734 	sc = vol->v_softc;
735 	g_topology_assert_not();
736 	sx_assert(&sc->sc_lock, SX_XLOCKED);
737 
738 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
739 //		return (0);
740 	if (!vol->v_dirty)
741 		return (0);
742 	if (vol->v_writes > 0)
743 		return (0);
744 	if (acw > 0 || (acw == -1 &&
745 	    vol->v_provider != NULL && vol->v_provider->acw > 0)) {
746 		timeout = g_raid_clean_time - (time_uptime - vol->v_last_write);
747 		if (timeout > 0)
748 			return (timeout);
749 	}
750 	vol->v_dirty = 0;
751 	G_RAID_DEBUG1(1, sc, "Volume %s marked as clean.",
752 	    vol->v_name);
753 	g_raid_write_metadata(sc, vol, NULL, NULL);
754 	return (0);
755 }
756 
757 static void
758 g_raid_dirty(struct g_raid_volume *vol)
759 {
760 	struct g_raid_softc *sc;
761 
762 	sc = vol->v_softc;
763 	g_topology_assert_not();
764 	sx_assert(&sc->sc_lock, SX_XLOCKED);
765 
766 //	if ((sc->sc_flags & G_RAID_DEVICE_FLAG_NOFAILSYNC) != 0)
767 //		return;
768 	vol->v_dirty = 1;
769 	G_RAID_DEBUG1(1, sc, "Volume %s marked as dirty.",
770 	    vol->v_name);
771 	g_raid_write_metadata(sc, vol, NULL, NULL);
772 }
773 
774 void
775 g_raid_tr_flush_common(struct g_raid_tr_object *tr, struct bio *bp)
776 {
777 	struct g_raid_softc *sc;
778 	struct g_raid_volume *vol;
779 	struct g_raid_subdisk *sd;
780 	struct bio_queue_head queue;
781 	struct bio *cbp;
782 	int i;
783 
784 	vol = tr->tro_volume;
785 	sc = vol->v_softc;
786 
787 	/*
788 	 * Allocate all bios before sending any request, so we can return
789 	 * ENOMEM in nice and clean way.
790 	 */
791 	bioq_init(&queue);
792 	for (i = 0; i < vol->v_disks_count; i++) {
793 		sd = &vol->v_subdisks[i];
794 		if (sd->sd_state == G_RAID_SUBDISK_S_NONE ||
795 		    sd->sd_state == G_RAID_SUBDISK_S_FAILED)
796 			continue;
797 		cbp = g_clone_bio(bp);
798 		if (cbp == NULL)
799 			goto failure;
800 		cbp->bio_caller1 = sd;
801 		bioq_insert_tail(&queue, cbp);
802 	}
803 	for (cbp = bioq_first(&queue); cbp != NULL;
804 	    cbp = bioq_first(&queue)) {
805 		bioq_remove(&queue, cbp);
806 		sd = cbp->bio_caller1;
807 		cbp->bio_caller1 = NULL;
808 		g_raid_subdisk_iostart(sd, cbp);
809 	}
810 	return;
811 failure:
812 	for (cbp = bioq_first(&queue); cbp != NULL;
813 	    cbp = bioq_first(&queue)) {
814 		bioq_remove(&queue, cbp);
815 		g_destroy_bio(cbp);
816 	}
817 	if (bp->bio_error == 0)
818 		bp->bio_error = ENOMEM;
819 	g_raid_iodone(bp, bp->bio_error);
820 }
821 
822 static void
823 g_raid_tr_kerneldump_common_done(struct bio *bp)
824 {
825 
826 	bp->bio_flags |= BIO_DONE;
827 }
828 
829 int
830 g_raid_tr_kerneldump_common(struct g_raid_tr_object *tr,
831     void *virtual, vm_offset_t physical, off_t offset, size_t length)
832 {
833 	struct g_raid_softc *sc;
834 	struct g_raid_volume *vol;
835 	struct bio bp;
836 
837 	vol = tr->tro_volume;
838 	sc = vol->v_softc;
839 
840 	bzero(&bp, sizeof(bp));
841 	bp.bio_cmd = BIO_WRITE;
842 	bp.bio_done = g_raid_tr_kerneldump_common_done;
843 	bp.bio_attribute = NULL;
844 	bp.bio_offset = offset;
845 	bp.bio_length = length;
846 	bp.bio_data = virtual;
847 	bp.bio_to = vol->v_provider;
848 
849 	g_raid_start(&bp);
850 	while (!(bp.bio_flags & BIO_DONE)) {
851 		G_RAID_DEBUG1(4, sc, "Poll...");
852 		g_raid_poll(sc);
853 		DELAY(10);
854 	}
855 
856 	return (bp.bio_error != 0 ? EIO : 0);
857 }
858 
859 static int
860 g_raid_dump(void *arg,
861     void *virtual, vm_offset_t physical, off_t offset, size_t length)
862 {
863 	struct g_raid_volume *vol;
864 	int error;
865 
866 	vol = (struct g_raid_volume *)arg;
867 	G_RAID_DEBUG1(3, vol->v_softc, "Dumping at off %llu len %llu.",
868 	    (long long unsigned)offset, (long long unsigned)length);
869 
870 	error = G_RAID_TR_KERNELDUMP(vol->v_tr,
871 	    virtual, physical, offset, length);
872 	return (error);
873 }
874 
875 static void
876 g_raid_kerneldump(struct g_raid_softc *sc, struct bio *bp)
877 {
878 	struct g_kerneldump *gkd;
879 	struct g_provider *pp;
880 	struct g_raid_volume *vol;
881 
882 	gkd = (struct g_kerneldump*)bp->bio_data;
883 	pp = bp->bio_to;
884 	vol = pp->private;
885 	g_trace(G_T_TOPOLOGY, "g_raid_kerneldump(%s, %jd, %jd)",
886 		pp->name, (intmax_t)gkd->offset, (intmax_t)gkd->length);
887 	gkd->di.dumper = g_raid_dump;
888 	gkd->di.priv = vol;
889 	gkd->di.blocksize = vol->v_sectorsize;
890 	gkd->di.maxiosize = DFLTPHYS;
891 	gkd->di.mediaoffset = gkd->offset;
892 	if ((gkd->offset + gkd->length) > vol->v_mediasize)
893 		gkd->length = vol->v_mediasize - gkd->offset;
894 	gkd->di.mediasize = gkd->length;
895 	g_io_deliver(bp, 0);
896 }
897 
898 static void
899 g_raid_start(struct bio *bp)
900 {
901 	struct g_raid_softc *sc;
902 
903 	sc = bp->bio_to->geom->softc;
904 	/*
905 	 * If sc == NULL or there are no valid disks, provider's error
906 	 * should be set and g_raid_start() should not be called at all.
907 	 */
908 //	KASSERT(sc != NULL && sc->sc_state == G_RAID_VOLUME_S_RUNNING,
909 //	    ("Provider's error should be set (error=%d)(mirror=%s).",
910 //	    bp->bio_to->error, bp->bio_to->name));
911 	G_RAID_LOGREQ(3, bp, "Request received.");
912 
913 	switch (bp->bio_cmd) {
914 	case BIO_READ:
915 	case BIO_WRITE:
916 	case BIO_DELETE:
917 	case BIO_FLUSH:
918 		break;
919 	case BIO_GETATTR:
920 		if (!strcmp(bp->bio_attribute, "GEOM::kerneldump"))
921 			g_raid_kerneldump(sc, bp);
922 		else
923 			g_io_deliver(bp, EOPNOTSUPP);
924 		return;
925 	default:
926 		g_io_deliver(bp, EOPNOTSUPP);
927 		return;
928 	}
929 	mtx_lock(&sc->sc_queue_mtx);
930 	bioq_disksort(&sc->sc_queue, bp);
931 	mtx_unlock(&sc->sc_queue_mtx);
932 	if (!dumping) {
933 		G_RAID_DEBUG1(4, sc, "Waking up %p.", sc);
934 		wakeup(sc);
935 	}
936 }
937 
938 static int
939 g_raid_bio_overlaps(const struct bio *bp, off_t lstart, off_t len)
940 {
941 	/*
942 	 * 5 cases:
943 	 * (1) bp entirely below NO
944 	 * (2) bp entirely above NO
945 	 * (3) bp start below, but end in range YES
946 	 * (4) bp entirely within YES
947 	 * (5) bp starts within, ends above YES
948 	 *
949 	 * lock range 10-19 (offset 10 length 10)
950 	 * (1) 1-5: first if kicks it out
951 	 * (2) 30-35: second if kicks it out
952 	 * (3) 5-15: passes both ifs
953 	 * (4) 12-14: passes both ifs
954 	 * (5) 19-20: passes both
955 	 */
956 	off_t lend = lstart + len - 1;
957 	off_t bstart = bp->bio_offset;
958 	off_t bend = bp->bio_offset + bp->bio_length - 1;
959 
960 	if (bend < lstart)
961 		return (0);
962 	if (lend < bstart)
963 		return (0);
964 	return (1);
965 }
966 
967 static int
968 g_raid_is_in_locked_range(struct g_raid_volume *vol, const struct bio *bp)
969 {
970 	struct g_raid_lock *lp;
971 
972 	sx_assert(&vol->v_softc->sc_lock, SX_LOCKED);
973 
974 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
975 		if (g_raid_bio_overlaps(bp, lp->l_offset, lp->l_length))
976 			return (1);
977 	}
978 	return (0);
979 }
980 
981 static void
982 g_raid_start_request(struct bio *bp)
983 {
984 	struct g_raid_softc *sc;
985 	struct g_raid_volume *vol;
986 
987 	sc = bp->bio_to->geom->softc;
988 	sx_assert(&sc->sc_lock, SX_LOCKED);
989 	vol = bp->bio_to->private;
990 
991 	/*
992 	 * Check to see if this item is in a locked range.  If so,
993 	 * queue it to our locked queue and return.  We'll requeue
994 	 * it when the range is unlocked.  Internal I/O for the
995 	 * rebuild/rescan/recovery process is excluded from this
996 	 * check so we can actually do the recovery.
997 	 */
998 	if (!(bp->bio_cflags & G_RAID_BIO_FLAG_SPECIAL) &&
999 	    g_raid_is_in_locked_range(vol, bp)) {
1000 		G_RAID_LOGREQ(3, bp, "Defer request.");
1001 		bioq_insert_tail(&vol->v_locked, bp);
1002 		return;
1003 	}
1004 
1005 	/*
1006 	 * If we're actually going to do the write/delete, then
1007 	 * update the idle stats for the volume.
1008 	 */
1009 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1010 		if (!vol->v_dirty)
1011 			g_raid_dirty(vol);
1012 		vol->v_writes++;
1013 	}
1014 
1015 	/*
1016 	 * Put request onto inflight queue, so we can check if new
1017 	 * synchronization requests don't collide with it.  Then tell
1018 	 * the transformation layer to start the I/O.
1019 	 */
1020 	bioq_insert_tail(&vol->v_inflight, bp);
1021 	G_RAID_LOGREQ(4, bp, "Request started");
1022 	G_RAID_TR_IOSTART(vol->v_tr, bp);
1023 }
1024 
1025 static void
1026 g_raid_finish_with_locked_ranges(struct g_raid_volume *vol, struct bio *bp)
1027 {
1028 	off_t off, len;
1029 	struct bio *nbp;
1030 	struct g_raid_lock *lp;
1031 
1032 	vol->v_pending_lock = 0;
1033 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1034 		if (lp->l_pending) {
1035 			off = lp->l_offset;
1036 			len = lp->l_length;
1037 			lp->l_pending = 0;
1038 			TAILQ_FOREACH(nbp, &vol->v_inflight.queue, bio_queue) {
1039 				if (g_raid_bio_overlaps(nbp, off, len))
1040 					lp->l_pending++;
1041 			}
1042 			if (lp->l_pending) {
1043 				vol->v_pending_lock = 1;
1044 				G_RAID_DEBUG1(4, vol->v_softc,
1045 				    "Deferred lock(%jd, %jd) has %d pending",
1046 				    (intmax_t)off, (intmax_t)(off + len),
1047 				    lp->l_pending);
1048 				continue;
1049 			}
1050 			G_RAID_DEBUG1(4, vol->v_softc,
1051 			    "Deferred lock of %jd to %jd completed",
1052 			    (intmax_t)off, (intmax_t)(off + len));
1053 			G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1054 		}
1055 	}
1056 }
1057 
1058 void
1059 g_raid_iodone(struct bio *bp, int error)
1060 {
1061 	struct g_raid_softc *sc;
1062 	struct g_raid_volume *vol;
1063 
1064 	sc = bp->bio_to->geom->softc;
1065 	sx_assert(&sc->sc_lock, SX_LOCKED);
1066 	vol = bp->bio_to->private;
1067 	G_RAID_LOGREQ(3, bp, "Request done: %d.", error);
1068 
1069 	/* Update stats if we done write/delete. */
1070 	if (bp->bio_cmd == BIO_WRITE || bp->bio_cmd == BIO_DELETE) {
1071 		vol->v_writes--;
1072 		vol->v_last_write = time_uptime;
1073 	}
1074 
1075 	bioq_remove(&vol->v_inflight, bp);
1076 	if (vol->v_pending_lock && g_raid_is_in_locked_range(vol, bp))
1077 		g_raid_finish_with_locked_ranges(vol, bp);
1078 	getmicrouptime(&vol->v_last_done);
1079 	g_io_deliver(bp, error);
1080 }
1081 
1082 int
1083 g_raid_lock_range(struct g_raid_volume *vol, off_t off, off_t len,
1084     struct bio *ignore, void *argp)
1085 {
1086 	struct g_raid_softc *sc;
1087 	struct g_raid_lock *lp;
1088 	struct bio *bp;
1089 
1090 	sc = vol->v_softc;
1091 	lp = malloc(sizeof(*lp), M_RAID, M_WAITOK | M_ZERO);
1092 	LIST_INSERT_HEAD(&vol->v_locks, lp, l_next);
1093 	lp->l_offset = off;
1094 	lp->l_length = len;
1095 	lp->l_callback_arg = argp;
1096 
1097 	lp->l_pending = 0;
1098 	TAILQ_FOREACH(bp, &vol->v_inflight.queue, bio_queue) {
1099 		if (bp != ignore && g_raid_bio_overlaps(bp, off, len))
1100 			lp->l_pending++;
1101 	}
1102 
1103 	/*
1104 	 * If there are any writes that are pending, we return EBUSY.  All
1105 	 * callers will have to wait until all pending writes clear.
1106 	 */
1107 	if (lp->l_pending > 0) {
1108 		vol->v_pending_lock = 1;
1109 		G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd deferred %d pend",
1110 		    (intmax_t)off, (intmax_t)(off+len), lp->l_pending);
1111 		return (EBUSY);
1112 	}
1113 	G_RAID_DEBUG1(4, sc, "Locking range %jd to %jd",
1114 	    (intmax_t)off, (intmax_t)(off+len));
1115 	G_RAID_TR_LOCKED(vol->v_tr, lp->l_callback_arg);
1116 	return (0);
1117 }
1118 
1119 int
1120 g_raid_unlock_range(struct g_raid_volume *vol, off_t off, off_t len)
1121 {
1122 	struct g_raid_lock *lp;
1123 	struct g_raid_softc *sc;
1124 	struct bio *bp;
1125 
1126 	sc = vol->v_softc;
1127 	LIST_FOREACH(lp, &vol->v_locks, l_next) {
1128 		if (lp->l_offset == off && lp->l_length == len) {
1129 			LIST_REMOVE(lp, l_next);
1130 			/* XXX
1131 			 * Right now we just put them all back on the queue
1132 			 * and hope for the best.  We hope this because any
1133 			 * locked ranges will go right back on this list
1134 			 * when the worker thread runs.
1135 			 * XXX
1136 			 */
1137 			G_RAID_DEBUG1(4, sc, "Unlocked %jd to %jd",
1138 			    (intmax_t)lp->l_offset,
1139 			    (intmax_t)(lp->l_offset+lp->l_length));
1140 			mtx_lock(&sc->sc_queue_mtx);
1141 			while ((bp = bioq_takefirst(&vol->v_locked)) != NULL)
1142 				bioq_disksort(&sc->sc_queue, bp);
1143 			mtx_unlock(&sc->sc_queue_mtx);
1144 			free(lp, M_RAID);
1145 			return (0);
1146 		}
1147 	}
1148 	return (EINVAL);
1149 }
1150 
1151 void
1152 g_raid_subdisk_iostart(struct g_raid_subdisk *sd, struct bio *bp)
1153 {
1154 	struct g_consumer *cp;
1155 	struct g_raid_disk *disk, *tdisk;
1156 
1157 	bp->bio_caller1 = sd;
1158 
1159 	/*
1160 	 * Make sure that the disk is present. Generally it is a task of
1161 	 * transformation layers to not send requests to absent disks, but
1162 	 * it is better to be safe and report situation then sorry.
1163 	 */
1164 	if (sd->sd_disk == NULL) {
1165 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to an absent disk!");
1166 nodisk:
1167 		bp->bio_from = NULL;
1168 		bp->bio_to = NULL;
1169 		bp->bio_error = ENXIO;
1170 		g_raid_disk_done(bp);
1171 		return;
1172 	}
1173 	disk = sd->sd_disk;
1174 	if (disk->d_state != G_RAID_DISK_S_ACTIVE &&
1175 	    disk->d_state != G_RAID_DISK_S_FAILED) {
1176 		G_RAID_LOGREQ(0, bp, "Warning! I/O request to a disk in a "
1177 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
1178 		goto nodisk;
1179 	}
1180 
1181 	cp = disk->d_consumer;
1182 	bp->bio_from = cp;
1183 	bp->bio_to = cp->provider;
1184 	cp->index++;
1185 
1186 	/* Update average disks load. */
1187 	TAILQ_FOREACH(tdisk, &sd->sd_softc->sc_disks, d_next) {
1188 		if (tdisk->d_consumer == NULL)
1189 			tdisk->d_load = 0;
1190 		else
1191 			tdisk->d_load = (tdisk->d_consumer->index *
1192 			    G_RAID_SUBDISK_LOAD_SCALE + tdisk->d_load * 7) / 8;
1193 	}
1194 
1195 	disk->d_last_offset = bp->bio_offset + bp->bio_length;
1196 	if (dumping) {
1197 		G_RAID_LOGREQ(3, bp, "Sending dumping request.");
1198 		if (bp->bio_cmd == BIO_WRITE) {
1199 			bp->bio_error = g_raid_subdisk_kerneldump(sd,
1200 			    bp->bio_data, 0, bp->bio_offset, bp->bio_length);
1201 		} else
1202 			bp->bio_error = EOPNOTSUPP;
1203 		g_raid_disk_done(bp);
1204 	} else {
1205 		bp->bio_done = g_raid_disk_done;
1206 		bp->bio_offset += sd->sd_offset;
1207 		G_RAID_LOGREQ(3, bp, "Sending request.");
1208 		g_io_request(bp, cp);
1209 	}
1210 }
1211 
1212 int
1213 g_raid_subdisk_kerneldump(struct g_raid_subdisk *sd,
1214     void *virtual, vm_offset_t physical, off_t offset, size_t length)
1215 {
1216 
1217 	if (sd->sd_disk == NULL)
1218 		return (ENXIO);
1219 	if (sd->sd_disk->d_kd.di.dumper == NULL)
1220 		return (EOPNOTSUPP);
1221 	return (dump_write(&sd->sd_disk->d_kd.di,
1222 	    virtual, physical,
1223 	    sd->sd_disk->d_kd.di.mediaoffset + sd->sd_offset + offset,
1224 	    length));
1225 }
1226 
1227 static void
1228 g_raid_disk_done(struct bio *bp)
1229 {
1230 	struct g_raid_softc *sc;
1231 	struct g_raid_subdisk *sd;
1232 
1233 	sd = bp->bio_caller1;
1234 	sc = sd->sd_softc;
1235 	mtx_lock(&sc->sc_queue_mtx);
1236 	bioq_disksort(&sc->sc_queue, bp);
1237 	mtx_unlock(&sc->sc_queue_mtx);
1238 	if (!dumping)
1239 		wakeup(sc);
1240 }
1241 
1242 static void
1243 g_raid_disk_done_request(struct bio *bp)
1244 {
1245 	struct g_raid_softc *sc;
1246 	struct g_raid_disk *disk;
1247 	struct g_raid_subdisk *sd;
1248 	struct g_raid_volume *vol;
1249 
1250 	g_topology_assert_not();
1251 
1252 	G_RAID_LOGREQ(3, bp, "Disk request done: %d.", bp->bio_error);
1253 	sd = bp->bio_caller1;
1254 	sc = sd->sd_softc;
1255 	vol = sd->sd_volume;
1256 	if (bp->bio_from != NULL) {
1257 		bp->bio_from->index--;
1258 		disk = bp->bio_from->private;
1259 		if (disk == NULL)
1260 			g_raid_kill_consumer(sc, bp->bio_from);
1261 	}
1262 	bp->bio_offset -= sd->sd_offset;
1263 
1264 	G_RAID_TR_IODONE(vol->v_tr, sd, bp);
1265 }
1266 
1267 static void
1268 g_raid_handle_event(struct g_raid_softc *sc, struct g_raid_event *ep)
1269 {
1270 
1271 	if ((ep->e_flags & G_RAID_EVENT_VOLUME) != 0)
1272 		ep->e_error = g_raid_update_volume(ep->e_tgt, ep->e_event);
1273 	else if ((ep->e_flags & G_RAID_EVENT_DISK) != 0)
1274 		ep->e_error = g_raid_update_disk(ep->e_tgt, ep->e_event);
1275 	else if ((ep->e_flags & G_RAID_EVENT_SUBDISK) != 0)
1276 		ep->e_error = g_raid_update_subdisk(ep->e_tgt, ep->e_event);
1277 	else
1278 		ep->e_error = g_raid_update_node(ep->e_tgt, ep->e_event);
1279 	if ((ep->e_flags & G_RAID_EVENT_WAIT) == 0) {
1280 		KASSERT(ep->e_error == 0,
1281 		    ("Error cannot be handled."));
1282 		g_raid_event_free(ep);
1283 	} else {
1284 		ep->e_flags |= G_RAID_EVENT_DONE;
1285 		G_RAID_DEBUG1(4, sc, "Waking up %p.", ep);
1286 		mtx_lock(&sc->sc_queue_mtx);
1287 		wakeup(ep);
1288 		mtx_unlock(&sc->sc_queue_mtx);
1289 	}
1290 }
1291 
1292 /*
1293  * Worker thread.
1294  */
1295 static void
1296 g_raid_worker(void *arg)
1297 {
1298 	struct g_raid_softc *sc;
1299 	struct g_raid_event *ep;
1300 	struct g_raid_volume *vol;
1301 	struct bio *bp;
1302 	struct timeval now, t;
1303 	int timeout, rv;
1304 
1305 	sc = arg;
1306 	thread_lock(curthread);
1307 	sched_prio(curthread, PRIBIO);
1308 	thread_unlock(curthread);
1309 
1310 	sx_xlock(&sc->sc_lock);
1311 	for (;;) {
1312 		mtx_lock(&sc->sc_queue_mtx);
1313 		/*
1314 		 * First take a look at events.
1315 		 * This is important to handle events before any I/O requests.
1316 		 */
1317 		bp = NULL;
1318 		vol = NULL;
1319 		rv = 0;
1320 		ep = TAILQ_FIRST(&sc->sc_events);
1321 		if (ep != NULL)
1322 			TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1323 		else if ((bp = bioq_takefirst(&sc->sc_queue)) != NULL)
1324 			;
1325 		else {
1326 			getmicrouptime(&now);
1327 			t = now;
1328 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1329 				if (bioq_first(&vol->v_inflight) == NULL &&
1330 				    vol->v_tr &&
1331 				    timevalcmp(&vol->v_last_done, &t, < ))
1332 					t = vol->v_last_done;
1333 			}
1334 			timevalsub(&t, &now);
1335 			timeout = g_raid_idle_threshold +
1336 			    t.tv_sec * 1000000 + t.tv_usec;
1337 			if (timeout > 0) {
1338 				/*
1339 				 * Two steps to avoid overflows at HZ=1000
1340 				 * and idle timeouts > 2.1s.  Some rounding
1341 				 * errors can occur, but they are < 1tick,
1342 				 * which is deemed to be close enough for
1343 				 * this purpose.
1344 				 */
1345 				int micpertic = 1000000 / hz;
1346 				timeout = (timeout + micpertic - 1) / micpertic;
1347 				sx_xunlock(&sc->sc_lock);
1348 				MSLEEP(rv, sc, &sc->sc_queue_mtx,
1349 				    PRIBIO | PDROP, "-", timeout);
1350 				sx_xlock(&sc->sc_lock);
1351 				goto process;
1352 			} else
1353 				rv = EWOULDBLOCK;
1354 		}
1355 		mtx_unlock(&sc->sc_queue_mtx);
1356 process:
1357 		if (ep != NULL) {
1358 			g_raid_handle_event(sc, ep);
1359 		} else if (bp != NULL) {
1360 			if (bp->bio_to != NULL &&
1361 			    bp->bio_to->geom == sc->sc_geom)
1362 				g_raid_start_request(bp);
1363 			else
1364 				g_raid_disk_done_request(bp);
1365 		} else if (rv == EWOULDBLOCK) {
1366 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
1367 				if (vol->v_writes == 0 && vol->v_dirty)
1368 					g_raid_clean(vol, -1);
1369 				if (bioq_first(&vol->v_inflight) == NULL &&
1370 				    vol->v_tr) {
1371 					t.tv_sec = g_raid_idle_threshold / 1000000;
1372 					t.tv_usec = g_raid_idle_threshold % 1000000;
1373 					timevaladd(&t, &vol->v_last_done);
1374 					getmicrouptime(&now);
1375 					if (timevalcmp(&t, &now, <= )) {
1376 						G_RAID_TR_IDLE(vol->v_tr);
1377 						vol->v_last_done = now;
1378 					}
1379 				}
1380 			}
1381 		}
1382 		if (sc->sc_stopping == G_RAID_DESTROY_HARD)
1383 			g_raid_destroy_node(sc, 1);	/* May not return. */
1384 	}
1385 }
1386 
1387 static void
1388 g_raid_poll(struct g_raid_softc *sc)
1389 {
1390 	struct g_raid_event *ep;
1391 	struct bio *bp;
1392 
1393 	sx_xlock(&sc->sc_lock);
1394 	mtx_lock(&sc->sc_queue_mtx);
1395 	/*
1396 	 * First take a look at events.
1397 	 * This is important to handle events before any I/O requests.
1398 	 */
1399 	ep = TAILQ_FIRST(&sc->sc_events);
1400 	if (ep != NULL) {
1401 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
1402 		mtx_unlock(&sc->sc_queue_mtx);
1403 		g_raid_handle_event(sc, ep);
1404 		goto out;
1405 	}
1406 	bp = bioq_takefirst(&sc->sc_queue);
1407 	if (bp != NULL) {
1408 		mtx_unlock(&sc->sc_queue_mtx);
1409 		if (bp->bio_from == NULL ||
1410 		    bp->bio_from->geom != sc->sc_geom)
1411 			g_raid_start_request(bp);
1412 		else
1413 			g_raid_disk_done_request(bp);
1414 	}
1415 out:
1416 	sx_xunlock(&sc->sc_lock);
1417 }
1418 
1419 static void
1420 g_raid_launch_provider(struct g_raid_volume *vol)
1421 {
1422 	struct g_raid_disk *disk;
1423 	struct g_raid_softc *sc;
1424 	struct g_provider *pp;
1425 	char name[G_RAID_MAX_VOLUMENAME];
1426 	off_t off;
1427 
1428 	sc = vol->v_softc;
1429 	sx_assert(&sc->sc_lock, SX_LOCKED);
1430 
1431 	g_topology_lock();
1432 	/* Try to name provider with volume name. */
1433 	snprintf(name, sizeof(name), "raid/%s", vol->v_name);
1434 	if (g_raid_name_format == 0 || vol->v_name[0] == 0 ||
1435 	    g_provider_by_name(name) != NULL) {
1436 		/* Otherwise use sequential volume number. */
1437 		snprintf(name, sizeof(name), "raid/r%d", vol->v_global_id);
1438 	}
1439 	pp = g_new_providerf(sc->sc_geom, "%s", name);
1440 	pp->private = vol;
1441 	pp->mediasize = vol->v_mediasize;
1442 	pp->sectorsize = vol->v_sectorsize;
1443 	pp->stripesize = 0;
1444 	pp->stripeoffset = 0;
1445 	if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID1 ||
1446 	    vol->v_raid_level == G_RAID_VOLUME_RL_RAID3 ||
1447 	    vol->v_raid_level == G_RAID_VOLUME_RL_SINGLE ||
1448 	    vol->v_raid_level == G_RAID_VOLUME_RL_CONCAT) {
1449 		if ((disk = vol->v_subdisks[0].sd_disk) != NULL &&
1450 		    disk->d_consumer != NULL &&
1451 		    disk->d_consumer->provider != NULL) {
1452 			pp->stripesize = disk->d_consumer->provider->stripesize;
1453 			off = disk->d_consumer->provider->stripeoffset;
1454 			pp->stripeoffset = off + vol->v_subdisks[0].sd_offset;
1455 			if (off > 0)
1456 				pp->stripeoffset %= off;
1457 		}
1458 		if (vol->v_raid_level == G_RAID_VOLUME_RL_RAID3) {
1459 			pp->stripesize *= (vol->v_disks_count - 1);
1460 			pp->stripeoffset *= (vol->v_disks_count - 1);
1461 		}
1462 	} else
1463 		pp->stripesize = vol->v_strip_size;
1464 	vol->v_provider = pp;
1465 	g_error_provider(pp, 0);
1466 	g_topology_unlock();
1467 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s created.",
1468 	    pp->name, vol->v_name);
1469 }
1470 
1471 static void
1472 g_raid_destroy_provider(struct g_raid_volume *vol)
1473 {
1474 	struct g_raid_softc *sc;
1475 	struct g_provider *pp;
1476 	struct bio *bp, *tmp;
1477 
1478 	g_topology_assert_not();
1479 	sc = vol->v_softc;
1480 	pp = vol->v_provider;
1481 	KASSERT(pp != NULL, ("NULL provider (volume=%s).", vol->v_name));
1482 
1483 	g_topology_lock();
1484 	g_error_provider(pp, ENXIO);
1485 	mtx_lock(&sc->sc_queue_mtx);
1486 	TAILQ_FOREACH_SAFE(bp, &sc->sc_queue.queue, bio_queue, tmp) {
1487 		if (bp->bio_to != pp)
1488 			continue;
1489 		bioq_remove(&sc->sc_queue, bp);
1490 		g_io_deliver(bp, ENXIO);
1491 	}
1492 	mtx_unlock(&sc->sc_queue_mtx);
1493 	G_RAID_DEBUG1(0, sc, "Provider %s for volume %s destroyed.",
1494 	    pp->name, vol->v_name);
1495 	g_wither_provider(pp, ENXIO);
1496 	g_topology_unlock();
1497 	vol->v_provider = NULL;
1498 }
1499 
1500 /*
1501  * Update device state.
1502  */
1503 static int
1504 g_raid_update_volume(struct g_raid_volume *vol, u_int event)
1505 {
1506 	struct g_raid_softc *sc;
1507 
1508 	sc = vol->v_softc;
1509 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1510 
1511 	G_RAID_DEBUG1(2, sc, "Event %s for volume %s.",
1512 	    g_raid_volume_event2str(event),
1513 	    vol->v_name);
1514 	switch (event) {
1515 	case G_RAID_VOLUME_E_DOWN:
1516 		if (vol->v_provider != NULL)
1517 			g_raid_destroy_provider(vol);
1518 		break;
1519 	case G_RAID_VOLUME_E_UP:
1520 		if (vol->v_provider == NULL)
1521 			g_raid_launch_provider(vol);
1522 		break;
1523 	case G_RAID_VOLUME_E_START:
1524 		if (vol->v_tr)
1525 			G_RAID_TR_START(vol->v_tr);
1526 		return (0);
1527 	default:
1528 		if (sc->sc_md)
1529 			G_RAID_MD_VOLUME_EVENT(sc->sc_md, vol, event);
1530 		return (0);
1531 	}
1532 
1533 	/* Manage root mount release. */
1534 	if (vol->v_starting) {
1535 		vol->v_starting = 0;
1536 		G_RAID_DEBUG1(1, sc, "root_mount_rel %p", vol->v_rootmount);
1537 		root_mount_rel(vol->v_rootmount);
1538 		vol->v_rootmount = NULL;
1539 	}
1540 	if (vol->v_stopping && vol->v_provider_open == 0)
1541 		g_raid_destroy_volume(vol);
1542 	return (0);
1543 }
1544 
1545 /*
1546  * Update subdisk state.
1547  */
1548 static int
1549 g_raid_update_subdisk(struct g_raid_subdisk *sd, u_int event)
1550 {
1551 	struct g_raid_softc *sc;
1552 	struct g_raid_volume *vol;
1553 
1554 	sc = sd->sd_softc;
1555 	vol = sd->sd_volume;
1556 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1557 
1558 	G_RAID_DEBUG1(2, sc, "Event %s for subdisk %s:%d-%s.",
1559 	    g_raid_subdisk_event2str(event),
1560 	    vol->v_name, sd->sd_pos,
1561 	    sd->sd_disk ? g_raid_get_diskname(sd->sd_disk) : "[none]");
1562 	if (vol->v_tr)
1563 		G_RAID_TR_EVENT(vol->v_tr, sd, event);
1564 
1565 	return (0);
1566 }
1567 
1568 /*
1569  * Update disk state.
1570  */
1571 static int
1572 g_raid_update_disk(struct g_raid_disk *disk, u_int event)
1573 {
1574 	struct g_raid_softc *sc;
1575 
1576 	sc = disk->d_softc;
1577 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1578 
1579 	G_RAID_DEBUG1(2, sc, "Event %s for disk %s.",
1580 	    g_raid_disk_event2str(event),
1581 	    g_raid_get_diskname(disk));
1582 
1583 	if (sc->sc_md)
1584 		G_RAID_MD_EVENT(sc->sc_md, disk, event);
1585 	return (0);
1586 }
1587 
1588 /*
1589  * Node event.
1590  */
1591 static int
1592 g_raid_update_node(struct g_raid_softc *sc, u_int event)
1593 {
1594 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1595 
1596 	G_RAID_DEBUG1(2, sc, "Event %s for the array.",
1597 	    g_raid_node_event2str(event));
1598 
1599 	if (event == G_RAID_NODE_E_WAKE)
1600 		return (0);
1601 	if (sc->sc_md)
1602 		G_RAID_MD_EVENT(sc->sc_md, NULL, event);
1603 	return (0);
1604 }
1605 
1606 static int
1607 g_raid_access(struct g_provider *pp, int acr, int acw, int ace)
1608 {
1609 	struct g_raid_volume *vol;
1610 	struct g_raid_softc *sc;
1611 	int dcw, opens, error = 0;
1612 
1613 	g_topology_assert();
1614 	sc = pp->geom->softc;
1615 	vol = pp->private;
1616 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
1617 	KASSERT(vol != NULL, ("NULL volume (provider=%s).", pp->name));
1618 
1619 	G_RAID_DEBUG1(2, sc, "Access request for %s: r%dw%de%d.", pp->name,
1620 	    acr, acw, ace);
1621 	dcw = pp->acw + acw;
1622 
1623 	g_topology_unlock();
1624 	sx_xlock(&sc->sc_lock);
1625 	/* Deny new opens while dying. */
1626 	if (sc->sc_stopping != 0 && (acr > 0 || acw > 0 || ace > 0)) {
1627 		error = ENXIO;
1628 		goto out;
1629 	}
1630 	if (dcw == 0 && vol->v_dirty)
1631 		g_raid_clean(vol, dcw);
1632 	vol->v_provider_open += acr + acw + ace;
1633 	/* Handle delayed node destruction. */
1634 	if (sc->sc_stopping == G_RAID_DESTROY_DELAYED &&
1635 	    vol->v_provider_open == 0) {
1636 		/* Count open volumes. */
1637 		opens = g_raid_nopens(sc);
1638 		if (opens == 0) {
1639 			sc->sc_stopping = G_RAID_DESTROY_HARD;
1640 			/* Wake up worker to make it selfdestruct. */
1641 			g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1642 		}
1643 	}
1644 	/* Handle open volume destruction. */
1645 	if (vol->v_stopping && vol->v_provider_open == 0)
1646 		g_raid_destroy_volume(vol);
1647 out:
1648 	sx_xunlock(&sc->sc_lock);
1649 	g_topology_lock();
1650 	return (error);
1651 }
1652 
1653 struct g_raid_softc *
1654 g_raid_create_node(struct g_class *mp,
1655     const char *name, struct g_raid_md_object *md)
1656 {
1657 	struct g_raid_softc *sc;
1658 	struct g_geom *gp;
1659 	int error;
1660 
1661 	g_topology_assert();
1662 	G_RAID_DEBUG(1, "Creating array %s.", name);
1663 
1664 	gp = g_new_geomf(mp, "%s", name);
1665 	sc = malloc(sizeof(*sc), M_RAID, M_WAITOK | M_ZERO);
1666 	gp->start = g_raid_start;
1667 	gp->orphan = g_raid_orphan;
1668 	gp->access = g_raid_access;
1669 	gp->dumpconf = g_raid_dumpconf;
1670 
1671 	sc->sc_md = md;
1672 	sc->sc_geom = gp;
1673 	sc->sc_flags = 0;
1674 	TAILQ_INIT(&sc->sc_volumes);
1675 	TAILQ_INIT(&sc->sc_disks);
1676 	sx_init(&sc->sc_lock, "gmirror:lock");
1677 	mtx_init(&sc->sc_queue_mtx, "gmirror:queue", NULL, MTX_DEF);
1678 	TAILQ_INIT(&sc->sc_events);
1679 	bioq_init(&sc->sc_queue);
1680 	gp->softc = sc;
1681 	error = kproc_create(g_raid_worker, sc, &sc->sc_worker, 0, 0,
1682 	    "g_raid %s", name);
1683 	if (error != 0) {
1684 		G_RAID_DEBUG(0, "Cannot create kernel thread for %s.", name);
1685 		mtx_destroy(&sc->sc_queue_mtx);
1686 		sx_destroy(&sc->sc_lock);
1687 		g_destroy_geom(sc->sc_geom);
1688 		free(sc, M_RAID);
1689 		return (NULL);
1690 	}
1691 
1692 	G_RAID_DEBUG1(0, sc, "Array %s created.", name);
1693 	return (sc);
1694 }
1695 
1696 struct g_raid_volume *
1697 g_raid_create_volume(struct g_raid_softc *sc, const char *name, int id)
1698 {
1699 	struct g_raid_volume	*vol, *vol1;
1700 	int i;
1701 
1702 	G_RAID_DEBUG1(1, sc, "Creating volume %s.", name);
1703 	vol = malloc(sizeof(*vol), M_RAID, M_WAITOK | M_ZERO);
1704 	vol->v_softc = sc;
1705 	strlcpy(vol->v_name, name, G_RAID_MAX_VOLUMENAME);
1706 	vol->v_state = G_RAID_VOLUME_S_STARTING;
1707 	vol->v_raid_level = G_RAID_VOLUME_RL_UNKNOWN;
1708 	vol->v_raid_level_qualifier = G_RAID_VOLUME_RLQ_UNKNOWN;
1709 	bioq_init(&vol->v_inflight);
1710 	bioq_init(&vol->v_locked);
1711 	LIST_INIT(&vol->v_locks);
1712 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1713 		vol->v_subdisks[i].sd_softc = sc;
1714 		vol->v_subdisks[i].sd_volume = vol;
1715 		vol->v_subdisks[i].sd_pos = i;
1716 		vol->v_subdisks[i].sd_state = G_RAID_DISK_S_NONE;
1717 	}
1718 
1719 	/* Find free ID for this volume. */
1720 	g_topology_lock();
1721 	vol1 = vol;
1722 	if (id >= 0) {
1723 		LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1724 			if (vol1->v_global_id == id)
1725 				break;
1726 		}
1727 	}
1728 	if (vol1 != NULL) {
1729 		for (id = 0; ; id++) {
1730 			LIST_FOREACH(vol1, &g_raid_volumes, v_global_next) {
1731 				if (vol1->v_global_id == id)
1732 					break;
1733 			}
1734 			if (vol1 == NULL)
1735 				break;
1736 		}
1737 	}
1738 	vol->v_global_id = id;
1739 	LIST_INSERT_HEAD(&g_raid_volumes, vol, v_global_next);
1740 	g_topology_unlock();
1741 
1742 	/* Delay root mounting. */
1743 	vol->v_rootmount = root_mount_hold("GRAID");
1744 	G_RAID_DEBUG1(1, sc, "root_mount_hold %p", vol->v_rootmount);
1745 	vol->v_starting = 1;
1746 	TAILQ_INSERT_TAIL(&sc->sc_volumes, vol, v_next);
1747 	return (vol);
1748 }
1749 
1750 struct g_raid_disk *
1751 g_raid_create_disk(struct g_raid_softc *sc)
1752 {
1753 	struct g_raid_disk	*disk;
1754 
1755 	G_RAID_DEBUG1(1, sc, "Creating disk.");
1756 	disk = malloc(sizeof(*disk), M_RAID, M_WAITOK | M_ZERO);
1757 	disk->d_softc = sc;
1758 	disk->d_state = G_RAID_DISK_S_NONE;
1759 	TAILQ_INIT(&disk->d_subdisks);
1760 	TAILQ_INSERT_TAIL(&sc->sc_disks, disk, d_next);
1761 	return (disk);
1762 }
1763 
1764 int g_raid_start_volume(struct g_raid_volume *vol)
1765 {
1766 	struct g_raid_tr_class *class;
1767 	struct g_raid_tr_object *obj;
1768 	int status;
1769 
1770 	G_RAID_DEBUG1(2, vol->v_softc, "Starting volume %s.", vol->v_name);
1771 	LIST_FOREACH(class, &g_raid_tr_classes, trc_list) {
1772 		G_RAID_DEBUG1(2, vol->v_softc,
1773 		    "Tasting volume %s for %s transformation.",
1774 		    vol->v_name, class->name);
1775 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
1776 		    M_WAITOK);
1777 		obj->tro_class = class;
1778 		obj->tro_volume = vol;
1779 		status = G_RAID_TR_TASTE(obj, vol);
1780 		if (status != G_RAID_TR_TASTE_FAIL)
1781 			break;
1782 		kobj_delete((kobj_t)obj, M_RAID);
1783 	}
1784 	if (class == NULL) {
1785 		G_RAID_DEBUG1(0, vol->v_softc,
1786 		    "No transformation module found for %s.",
1787 		    vol->v_name);
1788 		vol->v_tr = NULL;
1789 		g_raid_change_volume_state(vol, G_RAID_VOLUME_S_UNSUPPORTED);
1790 		g_raid_event_send(vol, G_RAID_VOLUME_E_DOWN,
1791 		    G_RAID_EVENT_VOLUME);
1792 		return (-1);
1793 	}
1794 	G_RAID_DEBUG1(2, vol->v_softc,
1795 	    "Transformation module %s chosen for %s.",
1796 	    class->name, vol->v_name);
1797 	vol->v_tr = obj;
1798 	return (0);
1799 }
1800 
1801 int
1802 g_raid_destroy_node(struct g_raid_softc *sc, int worker)
1803 {
1804 	struct g_raid_volume *vol, *tmpv;
1805 	struct g_raid_disk *disk, *tmpd;
1806 	int error = 0;
1807 
1808 	sc->sc_stopping = G_RAID_DESTROY_HARD;
1809 	TAILQ_FOREACH_SAFE(vol, &sc->sc_volumes, v_next, tmpv) {
1810 		if (g_raid_destroy_volume(vol))
1811 			error = EBUSY;
1812 	}
1813 	if (error)
1814 		return (error);
1815 	TAILQ_FOREACH_SAFE(disk, &sc->sc_disks, d_next, tmpd) {
1816 		if (g_raid_destroy_disk(disk))
1817 			error = EBUSY;
1818 	}
1819 	if (error)
1820 		return (error);
1821 	if (sc->sc_md) {
1822 		G_RAID_MD_FREE(sc->sc_md);
1823 		kobj_delete((kobj_t)sc->sc_md, M_RAID);
1824 		sc->sc_md = NULL;
1825 	}
1826 	if (sc->sc_geom != NULL) {
1827 		G_RAID_DEBUG1(0, sc, "Array %s destroyed.", sc->sc_name);
1828 		g_topology_lock();
1829 		sc->sc_geom->softc = NULL;
1830 		g_wither_geom(sc->sc_geom, ENXIO);
1831 		g_topology_unlock();
1832 		sc->sc_geom = NULL;
1833 	} else
1834 		G_RAID_DEBUG(1, "Array destroyed.");
1835 	if (worker) {
1836 		g_raid_event_cancel(sc, sc);
1837 		mtx_destroy(&sc->sc_queue_mtx);
1838 		sx_xunlock(&sc->sc_lock);
1839 		sx_destroy(&sc->sc_lock);
1840 		wakeup(&sc->sc_stopping);
1841 		free(sc, M_RAID);
1842 		curthread->td_pflags &= ~TDP_GEOM;
1843 		G_RAID_DEBUG(1, "Thread exiting.");
1844 		kproc_exit(0);
1845 	} else {
1846 		/* Wake up worker to make it selfdestruct. */
1847 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1848 	}
1849 	return (0);
1850 }
1851 
1852 int
1853 g_raid_destroy_volume(struct g_raid_volume *vol)
1854 {
1855 	struct g_raid_softc *sc;
1856 	struct g_raid_disk *disk;
1857 	int i;
1858 
1859 	sc = vol->v_softc;
1860 	G_RAID_DEBUG1(2, sc, "Destroying volume %s.", vol->v_name);
1861 	vol->v_stopping = 1;
1862 	if (vol->v_state != G_RAID_VOLUME_S_STOPPED) {
1863 		if (vol->v_tr) {
1864 			G_RAID_TR_STOP(vol->v_tr);
1865 			return (EBUSY);
1866 		} else
1867 			vol->v_state = G_RAID_VOLUME_S_STOPPED;
1868 	}
1869 	if (g_raid_event_check(sc, vol) != 0)
1870 		return (EBUSY);
1871 	if (vol->v_provider != NULL)
1872 		return (EBUSY);
1873 	if (vol->v_provider_open != 0)
1874 		return (EBUSY);
1875 	if (vol->v_tr) {
1876 		G_RAID_TR_FREE(vol->v_tr);
1877 		kobj_delete((kobj_t)vol->v_tr, M_RAID);
1878 		vol->v_tr = NULL;
1879 	}
1880 	if (vol->v_rootmount)
1881 		root_mount_rel(vol->v_rootmount);
1882 	g_topology_lock();
1883 	LIST_REMOVE(vol, v_global_next);
1884 	g_topology_unlock();
1885 	TAILQ_REMOVE(&sc->sc_volumes, vol, v_next);
1886 	for (i = 0; i < G_RAID_MAX_SUBDISKS; i++) {
1887 		g_raid_event_cancel(sc, &vol->v_subdisks[i]);
1888 		disk = vol->v_subdisks[i].sd_disk;
1889 		if (disk == NULL)
1890 			continue;
1891 		TAILQ_REMOVE(&disk->d_subdisks, &vol->v_subdisks[i], sd_next);
1892 	}
1893 	G_RAID_DEBUG1(2, sc, "Volume %s destroyed.", vol->v_name);
1894 	if (sc->sc_md)
1895 		G_RAID_MD_FREE_VOLUME(sc->sc_md, vol);
1896 	g_raid_event_cancel(sc, vol);
1897 	free(vol, M_RAID);
1898 	if (sc->sc_stopping == G_RAID_DESTROY_HARD) {
1899 		/* Wake up worker to let it selfdestruct. */
1900 		g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1901 	}
1902 	return (0);
1903 }
1904 
1905 int
1906 g_raid_destroy_disk(struct g_raid_disk *disk)
1907 {
1908 	struct g_raid_softc *sc;
1909 	struct g_raid_subdisk *sd, *tmp;
1910 
1911 	sc = disk->d_softc;
1912 	G_RAID_DEBUG1(2, sc, "Destroying disk.");
1913 	if (disk->d_consumer) {
1914 		g_raid_kill_consumer(sc, disk->d_consumer);
1915 		disk->d_consumer = NULL;
1916 	}
1917 	TAILQ_FOREACH_SAFE(sd, &disk->d_subdisks, sd_next, tmp) {
1918 		g_raid_change_subdisk_state(sd, G_RAID_SUBDISK_S_NONE);
1919 		g_raid_event_send(sd, G_RAID_SUBDISK_E_DISCONNECTED,
1920 		    G_RAID_EVENT_SUBDISK);
1921 		TAILQ_REMOVE(&disk->d_subdisks, sd, sd_next);
1922 		sd->sd_disk = NULL;
1923 	}
1924 	TAILQ_REMOVE(&sc->sc_disks, disk, d_next);
1925 	if (sc->sc_md)
1926 		G_RAID_MD_FREE_DISK(sc->sc_md, disk);
1927 	g_raid_event_cancel(sc, disk);
1928 	free(disk, M_RAID);
1929 	return (0);
1930 }
1931 
1932 int
1933 g_raid_destroy(struct g_raid_softc *sc, int how)
1934 {
1935 	int opens;
1936 
1937 	g_topology_assert_not();
1938 	if (sc == NULL)
1939 		return (ENXIO);
1940 	sx_assert(&sc->sc_lock, SX_XLOCKED);
1941 
1942 	/* Count open volumes. */
1943 	opens = g_raid_nopens(sc);
1944 
1945 	/* React on some opened volumes. */
1946 	if (opens > 0) {
1947 		switch (how) {
1948 		case G_RAID_DESTROY_SOFT:
1949 			G_RAID_DEBUG1(1, sc,
1950 			    "%d volumes are still open.",
1951 			    opens);
1952 			return (EBUSY);
1953 		case G_RAID_DESTROY_DELAYED:
1954 			G_RAID_DEBUG1(1, sc,
1955 			    "Array will be destroyed on last close.");
1956 			sc->sc_stopping = G_RAID_DESTROY_DELAYED;
1957 			return (EBUSY);
1958 		case G_RAID_DESTROY_HARD:
1959 			G_RAID_DEBUG1(1, sc,
1960 			    "%d volumes are still open.",
1961 			    opens);
1962 		}
1963 	}
1964 
1965 	/* Mark node for destruction. */
1966 	sc->sc_stopping = G_RAID_DESTROY_HARD;
1967 	/* Wake up worker to let it selfdestruct. */
1968 	g_raid_event_send(sc, G_RAID_NODE_E_WAKE, 0);
1969 	/* Sleep until node destroyed. */
1970 	sx_sleep(&sc->sc_stopping, &sc->sc_lock,
1971 	    PRIBIO | PDROP, "r:destroy", 0);
1972 	return (0);
1973 }
1974 
1975 static void
1976 g_raid_taste_orphan(struct g_consumer *cp)
1977 {
1978 
1979 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
1980 	    cp->provider->name));
1981 }
1982 
1983 static struct g_geom *
1984 g_raid_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
1985 {
1986 	struct g_consumer *cp;
1987 	struct g_geom *gp, *geom;
1988 	struct g_raid_md_class *class;
1989 	struct g_raid_md_object *obj;
1990 	int status;
1991 
1992 	g_topology_assert();
1993 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
1994 	G_RAID_DEBUG(2, "Tasting provider %s.", pp->name);
1995 
1996 	gp = g_new_geomf(mp, "mirror:taste");
1997 	/*
1998 	 * This orphan function should be never called.
1999 	 */
2000 	gp->orphan = g_raid_taste_orphan;
2001 	cp = g_new_consumer(gp);
2002 	g_attach(cp, pp);
2003 
2004 	geom = NULL;
2005 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2006 		G_RAID_DEBUG(2, "Tasting provider %s for %s metadata.",
2007 		    pp->name, class->name);
2008 		obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2009 		    M_WAITOK);
2010 		obj->mdo_class = class;
2011 		status = G_RAID_MD_TASTE(obj, mp, cp, &geom);
2012 		if (status != G_RAID_MD_TASTE_NEW)
2013 			kobj_delete((kobj_t)obj, M_RAID);
2014 		if (status != G_RAID_MD_TASTE_FAIL)
2015 			break;
2016 	}
2017 
2018 	g_detach(cp);
2019 	g_destroy_consumer(cp);
2020 	g_destroy_geom(gp);
2021 	G_RAID_DEBUG(2, "Tasting provider %s done.", pp->name);
2022 	return (geom);
2023 }
2024 
2025 int
2026 g_raid_create_node_format(const char *format, struct g_geom **gp)
2027 {
2028 	struct g_raid_md_class *class;
2029 	struct g_raid_md_object *obj;
2030 	int status;
2031 
2032 	G_RAID_DEBUG(2, "Creating array for %s metadata.", format);
2033 	LIST_FOREACH(class, &g_raid_md_classes, mdc_list) {
2034 		if (strcasecmp(class->name, format) == 0)
2035 			break;
2036 	}
2037 	if (class == NULL) {
2038 		G_RAID_DEBUG(1, "No support for %s metadata.", format);
2039 		return (G_RAID_MD_TASTE_FAIL);
2040 	}
2041 	obj = (void *)kobj_create((kobj_class_t)class, M_RAID,
2042 	    M_WAITOK);
2043 	obj->mdo_class = class;
2044 	status = G_RAID_MD_CREATE(obj, &g_raid_class, gp);
2045 	if (status != G_RAID_MD_TASTE_NEW)
2046 		kobj_delete((kobj_t)obj, M_RAID);
2047 	return (status);
2048 }
2049 
2050 static int
2051 g_raid_destroy_geom(struct gctl_req *req __unused,
2052     struct g_class *mp __unused, struct g_geom *gp)
2053 {
2054 	struct g_raid_softc *sc;
2055 	int error;
2056 
2057 	g_topology_unlock();
2058 	sc = gp->softc;
2059 	sx_xlock(&sc->sc_lock);
2060 	g_cancel_event(sc);
2061 	error = g_raid_destroy(gp->softc, G_RAID_DESTROY_SOFT);
2062 	if (error != 0)
2063 		sx_xunlock(&sc->sc_lock);
2064 	g_topology_lock();
2065 	return (error);
2066 }
2067 
2068 void g_raid_write_metadata(struct g_raid_softc *sc, struct g_raid_volume *vol,
2069     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2070 {
2071 
2072 	if (sc->sc_stopping == G_RAID_DESTROY_HARD)
2073 		return;
2074 	if (sc->sc_md)
2075 		G_RAID_MD_WRITE(sc->sc_md, vol, sd, disk);
2076 }
2077 
2078 void g_raid_fail_disk(struct g_raid_softc *sc,
2079     struct g_raid_subdisk *sd, struct g_raid_disk *disk)
2080 {
2081 
2082 	if (disk == NULL)
2083 		disk = sd->sd_disk;
2084 	if (disk == NULL) {
2085 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to an absent disk!");
2086 		return;
2087 	}
2088 	if (disk->d_state != G_RAID_DISK_S_ACTIVE) {
2089 		G_RAID_DEBUG1(0, sc, "Warning! Fail request to a disk in a "
2090 		    "wrong state (%s)!", g_raid_disk_state2str(disk->d_state));
2091 		return;
2092 	}
2093 	if (sc->sc_md)
2094 		G_RAID_MD_FAIL_DISK(sc->sc_md, sd, disk);
2095 }
2096 
2097 static void
2098 g_raid_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
2099     struct g_consumer *cp, struct g_provider *pp)
2100 {
2101 	struct g_raid_softc *sc;
2102 	struct g_raid_volume *vol;
2103 	struct g_raid_subdisk *sd;
2104 	struct g_raid_disk *disk;
2105 	int i, s;
2106 
2107 	g_topology_assert();
2108 
2109 	sc = gp->softc;
2110 	if (sc == NULL)
2111 		return;
2112 	if (pp != NULL) {
2113 		vol = pp->private;
2114 		g_topology_unlock();
2115 		sx_xlock(&sc->sc_lock);
2116 		sbuf_printf(sb, "%s<Label>%s</Label>\n", indent,
2117 		    vol->v_name);
2118 		sbuf_printf(sb, "%s<RAIDLevel>%s</RAIDLevel>\n", indent,
2119 		    g_raid_volume_level2str(vol->v_raid_level,
2120 		    vol->v_raid_level_qualifier));
2121 		sbuf_printf(sb,
2122 		    "%s<Transformation>%s</Transformation>\n", indent,
2123 		    vol->v_tr ? vol->v_tr->tro_class->name : "NONE");
2124 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
2125 		    vol->v_disks_count);
2126 		sbuf_printf(sb, "%s<Strip>%u</Strip>\n", indent,
2127 		    vol->v_strip_size);
2128 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2129 		    g_raid_volume_state2str(vol->v_state));
2130 		sbuf_printf(sb, "%s<Dirty>%s</Dirty>\n", indent,
2131 		    vol->v_dirty ? "Yes" : "No");
2132 		sbuf_printf(sb, "%s<Subdisks>", indent);
2133 		for (i = 0; i < vol->v_disks_count; i++) {
2134 			sd = &vol->v_subdisks[i];
2135 			if (sd->sd_disk != NULL &&
2136 			    sd->sd_disk->d_consumer != NULL) {
2137 				sbuf_printf(sb, "%s ",
2138 				    g_raid_get_diskname(sd->sd_disk));
2139 			} else {
2140 				sbuf_printf(sb, "NONE ");
2141 			}
2142 			sbuf_printf(sb, "(%s",
2143 			    g_raid_subdisk_state2str(sd->sd_state));
2144 			if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2145 			    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2146 				sbuf_printf(sb, " %d%%",
2147 				    (int)(sd->sd_rebuild_pos * 100 /
2148 				     sd->sd_size));
2149 			}
2150 			sbuf_printf(sb, ")");
2151 			if (i + 1 < vol->v_disks_count)
2152 				sbuf_printf(sb, ", ");
2153 		}
2154 		sbuf_printf(sb, "</Subdisks>\n");
2155 		sx_xunlock(&sc->sc_lock);
2156 		g_topology_lock();
2157 	} else if (cp != NULL) {
2158 		disk = cp->private;
2159 		if (disk == NULL)
2160 			return;
2161 		g_topology_unlock();
2162 		sx_xlock(&sc->sc_lock);
2163 		sbuf_printf(sb, "%s<State>%s", indent,
2164 		    g_raid_disk_state2str(disk->d_state));
2165 		if (!TAILQ_EMPTY(&disk->d_subdisks)) {
2166 			sbuf_printf(sb, " (");
2167 			TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2168 				sbuf_printf(sb, "%s",
2169 				    g_raid_subdisk_state2str(sd->sd_state));
2170 				if (sd->sd_state == G_RAID_SUBDISK_S_REBUILD ||
2171 				    sd->sd_state == G_RAID_SUBDISK_S_RESYNC) {
2172 					sbuf_printf(sb, " %d%%",
2173 					    (int)(sd->sd_rebuild_pos * 100 /
2174 					     sd->sd_size));
2175 				}
2176 				if (TAILQ_NEXT(sd, sd_next))
2177 					sbuf_printf(sb, ", ");
2178 			}
2179 			sbuf_printf(sb, ")");
2180 		}
2181 		sbuf_printf(sb, "</State>\n");
2182 		sbuf_printf(sb, "%s<Subdisks>", indent);
2183 		TAILQ_FOREACH(sd, &disk->d_subdisks, sd_next) {
2184 			sbuf_printf(sb, "r%d(%s):%d@%ju",
2185 			    sd->sd_volume->v_global_id,
2186 			    sd->sd_volume->v_name,
2187 			    sd->sd_pos, sd->sd_offset);
2188 			if (TAILQ_NEXT(sd, sd_next))
2189 				sbuf_printf(sb, ", ");
2190 		}
2191 		sbuf_printf(sb, "</Subdisks>\n");
2192 		sbuf_printf(sb, "%s<ReadErrors>%d</ReadErrors>\n", indent,
2193 		    disk->d_read_errs);
2194 		sx_xunlock(&sc->sc_lock);
2195 		g_topology_lock();
2196 	} else {
2197 		g_topology_unlock();
2198 		sx_xlock(&sc->sc_lock);
2199 		if (sc->sc_md) {
2200 			sbuf_printf(sb, "%s<Metadata>%s</Metadata>\n", indent,
2201 			    sc->sc_md->mdo_class->name);
2202 		}
2203 		if (!TAILQ_EMPTY(&sc->sc_volumes)) {
2204 			s = 0xff;
2205 			TAILQ_FOREACH(vol, &sc->sc_volumes, v_next) {
2206 				if (vol->v_state < s)
2207 					s = vol->v_state;
2208 			}
2209 			sbuf_printf(sb, "%s<State>%s</State>\n", indent,
2210 			    g_raid_volume_state2str(s));
2211 		}
2212 		sx_xunlock(&sc->sc_lock);
2213 		g_topology_lock();
2214 	}
2215 }
2216 
2217 static void
2218 g_raid_shutdown_pre_sync(void *arg, int howto)
2219 {
2220 	struct g_class *mp;
2221 	struct g_geom *gp, *gp2;
2222 	struct g_raid_softc *sc;
2223 	int error;
2224 
2225 	mp = arg;
2226 	DROP_GIANT();
2227 	g_topology_lock();
2228 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
2229 		if ((sc = gp->softc) == NULL)
2230 			continue;
2231 		g_topology_unlock();
2232 		sx_xlock(&sc->sc_lock);
2233 		g_cancel_event(sc);
2234 		error = g_raid_destroy(sc, G_RAID_DESTROY_DELAYED);
2235 		if (error != 0)
2236 			sx_xunlock(&sc->sc_lock);
2237 		g_topology_lock();
2238 	}
2239 	g_topology_unlock();
2240 	PICKUP_GIANT();
2241 }
2242 
2243 static void
2244 g_raid_init(struct g_class *mp)
2245 {
2246 
2247 	g_raid_pre_sync = EVENTHANDLER_REGISTER(shutdown_pre_sync,
2248 	    g_raid_shutdown_pre_sync, mp, SHUTDOWN_PRI_FIRST);
2249 	if (g_raid_pre_sync == NULL)
2250 		G_RAID_DEBUG(0, "Warning! Cannot register shutdown event.");
2251 	g_raid_started = 1;
2252 }
2253 
2254 static void
2255 g_raid_fini(struct g_class *mp)
2256 {
2257 
2258 	if (g_raid_pre_sync != NULL)
2259 		EVENTHANDLER_DEREGISTER(shutdown_pre_sync, g_raid_pre_sync);
2260 	g_raid_started = 0;
2261 }
2262 
2263 int
2264 g_raid_md_modevent(module_t mod, int type, void *arg)
2265 {
2266 	struct g_raid_md_class *class, *c, *nc;
2267 	int error;
2268 
2269 	error = 0;
2270 	class = arg;
2271 	switch (type) {
2272 	case MOD_LOAD:
2273 		c = LIST_FIRST(&g_raid_md_classes);
2274 		if (c == NULL || c->mdc_priority > class->mdc_priority)
2275 			LIST_INSERT_HEAD(&g_raid_md_classes, class, mdc_list);
2276 		else {
2277 			while ((nc = LIST_NEXT(c, mdc_list)) != NULL &&
2278 			    nc->mdc_priority < class->mdc_priority)
2279 				c = nc;
2280 			LIST_INSERT_AFTER(c, class, mdc_list);
2281 		}
2282 		if (g_raid_started)
2283 			g_retaste(&g_raid_class);
2284 		break;
2285 	case MOD_UNLOAD:
2286 		LIST_REMOVE(class, mdc_list);
2287 		break;
2288 	default:
2289 		error = EOPNOTSUPP;
2290 		break;
2291 	}
2292 
2293 	return (error);
2294 }
2295 
2296 int
2297 g_raid_tr_modevent(module_t mod, int type, void *arg)
2298 {
2299 	struct g_raid_tr_class *class, *c, *nc;
2300 	int error;
2301 
2302 	error = 0;
2303 	class = arg;
2304 	switch (type) {
2305 	case MOD_LOAD:
2306 		c = LIST_FIRST(&g_raid_tr_classes);
2307 		if (c == NULL || c->trc_priority > class->trc_priority)
2308 			LIST_INSERT_HEAD(&g_raid_tr_classes, class, trc_list);
2309 		else {
2310 			while ((nc = LIST_NEXT(c, trc_list)) != NULL &&
2311 			    nc->trc_priority < class->trc_priority)
2312 				c = nc;
2313 			LIST_INSERT_AFTER(c, class, trc_list);
2314 		}
2315 		break;
2316 	case MOD_UNLOAD:
2317 		LIST_REMOVE(class, trc_list);
2318 		break;
2319 	default:
2320 		error = EOPNOTSUPP;
2321 		break;
2322 	}
2323 
2324 	return (error);
2325 }
2326 
2327 /*
2328  * Use local implementation of DECLARE_GEOM_CLASS(g_raid_class, g_raid)
2329  * to reduce module priority, allowing submodules to register them first.
2330  */
2331 static moduledata_t g_raid_mod = {
2332 	"g_raid",
2333 	g_modevent,
2334 	&g_raid_class
2335 };
2336 DECLARE_MODULE(g_raid, g_raid_mod, SI_SUB_DRIVERS, SI_ORDER_THIRD);
2337 MODULE_VERSION(geom_raid, 0);
2338