xref: /freebsd/sys/geom/raid3/g_raid3.c (revision e5283a9da86a0080ab0267bcbff4febb049d323d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2004-2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/bio.h>
32 #include <sys/eventhandler.h>
33 #include <sys/kernel.h>
34 #include <sys/kthread.h>
35 #include <sys/limits.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/mutex.h>
40 #include <sys/proc.h>
41 #include <sys/reboot.h>
42 #include <sys/sbuf.h>
43 #include <sys/sched.h>
44 #include <sys/sysctl.h>
45 
46 #include <vm/uma.h>
47 
48 #include <geom/geom.h>
49 #include <geom/geom_dbg.h>
50 #include <geom/geom_disk.h>
51 #include <geom/raid3/g_raid3.h>
52 
53 FEATURE(geom_raid3, "GEOM RAID-3 functionality");
54 
55 static MALLOC_DEFINE(M_RAID3, "raid3_data", "GEOM_RAID3 Data");
56 
57 SYSCTL_DECL(_kern_geom);
58 static SYSCTL_NODE(_kern_geom, OID_AUTO, raid3, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
59     "GEOM_RAID3 stuff");
60 u_int g_raid3_debug = 0;
61 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, debug, CTLFLAG_RWTUN, &g_raid3_debug, 0,
62     "Debug level");
63 static u_int g_raid3_timeout = 4;
64 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, timeout, CTLFLAG_RWTUN, &g_raid3_timeout,
65     0, "Time to wait on all raid3 components");
66 static u_int g_raid3_idletime = 5;
67 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, idletime, CTLFLAG_RWTUN,
68     &g_raid3_idletime, 0, "Mark components as clean when idling");
69 static u_int g_raid3_disconnect_on_failure = 1;
70 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, disconnect_on_failure, CTLFLAG_RWTUN,
71     &g_raid3_disconnect_on_failure, 0, "Disconnect component on I/O failure.");
72 static u_int g_raid3_syncreqs = 2;
73 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, sync_requests, CTLFLAG_RDTUN,
74     &g_raid3_syncreqs, 0, "Parallel synchronization I/O requests.");
75 static u_int g_raid3_use_malloc = 0;
76 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, use_malloc, CTLFLAG_RDTUN,
77     &g_raid3_use_malloc, 0, "Use malloc(9) instead of uma(9).");
78 
79 static u_int g_raid3_n64k = 50;
80 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n64k, CTLFLAG_RDTUN, &g_raid3_n64k, 0,
81     "Maximum number of 64kB allocations");
82 static u_int g_raid3_n16k = 200;
83 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n16k, CTLFLAG_RDTUN, &g_raid3_n16k, 0,
84     "Maximum number of 16kB allocations");
85 static u_int g_raid3_n4k = 1200;
86 SYSCTL_UINT(_kern_geom_raid3, OID_AUTO, n4k, CTLFLAG_RDTUN, &g_raid3_n4k, 0,
87     "Maximum number of 4kB allocations");
88 
89 static SYSCTL_NODE(_kern_geom_raid3, OID_AUTO, stat,
90     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
91     "GEOM_RAID3 statistics");
92 static u_int g_raid3_parity_mismatch = 0;
93 SYSCTL_UINT(_kern_geom_raid3_stat, OID_AUTO, parity_mismatch, CTLFLAG_RD,
94     &g_raid3_parity_mismatch, 0, "Number of failures in VERIFY mode");
95 
96 #define	MSLEEP(ident, mtx, priority, wmesg, timeout)	do {		\
97 	G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, (ident));	\
98 	msleep((ident), (mtx), (priority), (wmesg), (timeout));		\
99 	G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, (ident));	\
100 } while (0)
101 
102 static eventhandler_tag g_raid3_post_sync = NULL;
103 static int g_raid3_shutdown = 0;
104 
105 static int g_raid3_destroy_geom(struct gctl_req *req, struct g_class *mp,
106     struct g_geom *gp);
107 static g_taste_t g_raid3_taste;
108 static void g_raid3_init(struct g_class *mp);
109 static void g_raid3_fini(struct g_class *mp);
110 static void g_raid3_providergone(struct g_provider *pp);
111 
112 struct g_class g_raid3_class = {
113 	.name = G_RAID3_CLASS_NAME,
114 	.version = G_VERSION,
115 	.ctlreq = g_raid3_config,
116 	.taste = g_raid3_taste,
117 	.destroy_geom = g_raid3_destroy_geom,
118 	.init = g_raid3_init,
119 	.fini = g_raid3_fini,
120 	.providergone = g_raid3_providergone,
121 };
122 
123 static void g_raid3_destroy_provider(struct g_raid3_softc *sc);
124 static int g_raid3_update_disk(struct g_raid3_disk *disk, u_int state);
125 static void g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force);
126 static void g_raid3_dumpconf(struct sbuf *sb, const char *indent,
127     struct g_geom *gp, struct g_consumer *cp, struct g_provider *pp);
128 static void g_raid3_sync_stop(struct g_raid3_softc *sc, int type);
129 static int g_raid3_register_request(struct bio *pbp);
130 static void g_raid3_sync_release(struct g_raid3_softc *sc);
131 static void g_raid3_timeout_drain(struct g_raid3_softc *sc);
132 
133 static const char *
g_raid3_disk_state2str(int state)134 g_raid3_disk_state2str(int state)
135 {
136 
137 	switch (state) {
138 	case G_RAID3_DISK_STATE_NODISK:
139 		return ("NODISK");
140 	case G_RAID3_DISK_STATE_NONE:
141 		return ("NONE");
142 	case G_RAID3_DISK_STATE_NEW:
143 		return ("NEW");
144 	case G_RAID3_DISK_STATE_ACTIVE:
145 		return ("ACTIVE");
146 	case G_RAID3_DISK_STATE_STALE:
147 		return ("STALE");
148 	case G_RAID3_DISK_STATE_SYNCHRONIZING:
149 		return ("SYNCHRONIZING");
150 	case G_RAID3_DISK_STATE_DISCONNECTED:
151 		return ("DISCONNECTED");
152 	default:
153 		return ("INVALID");
154 	}
155 }
156 
157 static const char *
g_raid3_device_state2str(int state)158 g_raid3_device_state2str(int state)
159 {
160 
161 	switch (state) {
162 	case G_RAID3_DEVICE_STATE_STARTING:
163 		return ("STARTING");
164 	case G_RAID3_DEVICE_STATE_DEGRADED:
165 		return ("DEGRADED");
166 	case G_RAID3_DEVICE_STATE_COMPLETE:
167 		return ("COMPLETE");
168 	default:
169 		return ("INVALID");
170 	}
171 }
172 
173 const char *
g_raid3_get_diskname(struct g_raid3_disk * disk)174 g_raid3_get_diskname(struct g_raid3_disk *disk)
175 {
176 
177 	if (disk->d_consumer == NULL || disk->d_consumer->provider == NULL)
178 		return ("[unknown]");
179 	return (disk->d_name);
180 }
181 
182 static void *
g_raid3_alloc(struct g_raid3_softc * sc,size_t size,int flags)183 g_raid3_alloc(struct g_raid3_softc *sc, size_t size, int flags)
184 {
185 	void *ptr;
186 	enum g_raid3_zones zone;
187 
188 	if (g_raid3_use_malloc ||
189 	    (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES)
190 		ptr = malloc(size, M_RAID3, flags);
191 	else {
192 		ptr = uma_zalloc_arg(sc->sc_zones[zone].sz_zone,
193 		   &sc->sc_zones[zone], flags);
194 		sc->sc_zones[zone].sz_requested++;
195 		if (ptr == NULL)
196 			sc->sc_zones[zone].sz_failed++;
197 	}
198 	return (ptr);
199 }
200 
201 static void
g_raid3_free(struct g_raid3_softc * sc,void * ptr,size_t size)202 g_raid3_free(struct g_raid3_softc *sc, void *ptr, size_t size)
203 {
204 	enum g_raid3_zones zone;
205 
206 	if (g_raid3_use_malloc ||
207 	    (zone = g_raid3_zone(size)) == G_RAID3_NUM_ZONES)
208 		free(ptr, M_RAID3);
209 	else {
210 		uma_zfree_arg(sc->sc_zones[zone].sz_zone,
211 		    ptr, &sc->sc_zones[zone]);
212 	}
213 }
214 
215 static int
g_raid3_uma_ctor(void * mem,int size,void * arg,int flags)216 g_raid3_uma_ctor(void *mem, int size, void *arg, int flags)
217 {
218 	struct g_raid3_zone *sz = arg;
219 
220 	if (sz->sz_max > 0 && sz->sz_inuse == sz->sz_max)
221 		return (ENOMEM);
222 	sz->sz_inuse++;
223 	return (0);
224 }
225 
226 static void
g_raid3_uma_dtor(void * mem,int size,void * arg)227 g_raid3_uma_dtor(void *mem, int size, void *arg)
228 {
229 	struct g_raid3_zone *sz = arg;
230 
231 	sz->sz_inuse--;
232 }
233 
234 #define	g_raid3_xor(src, dst, size)					\
235 	_g_raid3_xor((uint64_t *)(src),					\
236 	    (uint64_t *)(dst), (size_t)size)
237 static void
_g_raid3_xor(uint64_t * src,uint64_t * dst,size_t size)238 _g_raid3_xor(uint64_t *src, uint64_t *dst, size_t size)
239 {
240 
241 	KASSERT((size % 128) == 0, ("Invalid size: %zu.", size));
242 	for (; size > 0; size -= 128) {
243 		*dst++ ^= (*src++);
244 		*dst++ ^= (*src++);
245 		*dst++ ^= (*src++);
246 		*dst++ ^= (*src++);
247 		*dst++ ^= (*src++);
248 		*dst++ ^= (*src++);
249 		*dst++ ^= (*src++);
250 		*dst++ ^= (*src++);
251 		*dst++ ^= (*src++);
252 		*dst++ ^= (*src++);
253 		*dst++ ^= (*src++);
254 		*dst++ ^= (*src++);
255 		*dst++ ^= (*src++);
256 		*dst++ ^= (*src++);
257 		*dst++ ^= (*src++);
258 		*dst++ ^= (*src++);
259 	}
260 }
261 
262 static int
g_raid3_is_zero(struct bio * bp)263 g_raid3_is_zero(struct bio *bp)
264 {
265 	static const uint64_t zeros[] = {
266 	    0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
267 	};
268 	u_char *addr;
269 	ssize_t size;
270 
271 	size = bp->bio_length;
272 	addr = (u_char *)bp->bio_data;
273 	for (; size > 0; size -= sizeof(zeros), addr += sizeof(zeros)) {
274 		if (bcmp(addr, zeros, sizeof(zeros)) != 0)
275 			return (0);
276 	}
277 	return (1);
278 }
279 
280 /*
281  * --- Events handling functions ---
282  * Events in geom_raid3 are used to maintain disks and device status
283  * from one thread to simplify locking.
284  */
285 static void
g_raid3_event_free(struct g_raid3_event * ep)286 g_raid3_event_free(struct g_raid3_event *ep)
287 {
288 
289 	free(ep, M_RAID3);
290 }
291 
292 static int
g_raid3_event_dispatch(struct g_raid3_event * ep,void * arg,int state,int flags)293 g_raid3_event_dispatch(struct g_raid3_event *ep, void *arg, int state,
294     int flags)
295 {
296 	struct g_raid3_softc *sc;
297 	struct g_raid3_disk *disk;
298 	int error;
299 
300 	G_RAID3_DEBUG(4, "%s: Sending event %p.", __func__, ep);
301 	if ((flags & G_RAID3_EVENT_DEVICE) != 0) {
302 		disk = NULL;
303 		sc = arg;
304 	} else {
305 		disk = arg;
306 		sc = disk->d_softc;
307 	}
308 	ep->e_disk = disk;
309 	ep->e_state = state;
310 	ep->e_flags = flags;
311 	ep->e_error = 0;
312 	mtx_lock(&sc->sc_events_mtx);
313 	TAILQ_INSERT_TAIL(&sc->sc_events, ep, e_next);
314 	mtx_unlock(&sc->sc_events_mtx);
315 	G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
316 	mtx_lock(&sc->sc_queue_mtx);
317 	wakeup(sc);
318 	wakeup(&sc->sc_queue);
319 	mtx_unlock(&sc->sc_queue_mtx);
320 	if ((flags & G_RAID3_EVENT_DONTWAIT) != 0)
321 		return (0);
322 	sx_assert(&sc->sc_lock, SX_XLOCKED);
323 	G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, ep);
324 	sx_xunlock(&sc->sc_lock);
325 	while ((ep->e_flags & G_RAID3_EVENT_DONE) == 0) {
326 		mtx_lock(&sc->sc_events_mtx);
327 		MSLEEP(ep, &sc->sc_events_mtx, PRIBIO | PDROP, "r3:event",
328 		    hz * 5);
329 	}
330 	error = ep->e_error;
331 	g_raid3_event_free(ep);
332 	sx_xlock(&sc->sc_lock);
333 	return (error);
334 }
335 
336 int
g_raid3_event_send(void * arg,int state,int flags)337 g_raid3_event_send(void *arg, int state, int flags)
338 {
339 	struct g_raid3_event *ep;
340 
341 	ep = malloc(sizeof(*ep), M_RAID3, M_WAITOK);
342 	return (g_raid3_event_dispatch(ep, arg, state, flags));
343 }
344 
345 static struct g_raid3_event *
g_raid3_event_get(struct g_raid3_softc * sc)346 g_raid3_event_get(struct g_raid3_softc *sc)
347 {
348 	struct g_raid3_event *ep;
349 
350 	mtx_lock(&sc->sc_events_mtx);
351 	ep = TAILQ_FIRST(&sc->sc_events);
352 	mtx_unlock(&sc->sc_events_mtx);
353 	return (ep);
354 }
355 
356 static void
g_raid3_event_remove(struct g_raid3_softc * sc,struct g_raid3_event * ep)357 g_raid3_event_remove(struct g_raid3_softc *sc, struct g_raid3_event *ep)
358 {
359 
360 	mtx_lock(&sc->sc_events_mtx);
361 	TAILQ_REMOVE(&sc->sc_events, ep, e_next);
362 	mtx_unlock(&sc->sc_events_mtx);
363 }
364 
365 static void
g_raid3_event_cancel(struct g_raid3_disk * disk)366 g_raid3_event_cancel(struct g_raid3_disk *disk)
367 {
368 	struct g_raid3_softc *sc;
369 	struct g_raid3_event *ep, *tmpep;
370 
371 	sc = disk->d_softc;
372 	sx_assert(&sc->sc_lock, SX_XLOCKED);
373 
374 	mtx_lock(&sc->sc_events_mtx);
375 	TAILQ_FOREACH_SAFE(ep, &sc->sc_events, e_next, tmpep) {
376 		if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0)
377 			continue;
378 		if (ep->e_disk != disk)
379 			continue;
380 		TAILQ_REMOVE(&sc->sc_events, ep, e_next);
381 		if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0)
382 			g_raid3_event_free(ep);
383 		else {
384 			ep->e_error = ECANCELED;
385 			wakeup(ep);
386 		}
387 	}
388 	mtx_unlock(&sc->sc_events_mtx);
389 }
390 
391 /*
392  * Return the number of disks in the given state.
393  * If state is equal to -1, count all connected disks.
394  */
395 u_int
g_raid3_ndisks(struct g_raid3_softc * sc,int state)396 g_raid3_ndisks(struct g_raid3_softc *sc, int state)
397 {
398 	struct g_raid3_disk *disk;
399 	u_int n, ndisks;
400 
401 	sx_assert(&sc->sc_lock, SX_LOCKED);
402 
403 	for (n = ndisks = 0; n < sc->sc_ndisks; n++) {
404 		disk = &sc->sc_disks[n];
405 		if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
406 			continue;
407 		if (state == -1 || disk->d_state == state)
408 			ndisks++;
409 	}
410 	return (ndisks);
411 }
412 
413 static u_int
g_raid3_nrequests(struct g_raid3_softc * sc,struct g_consumer * cp)414 g_raid3_nrequests(struct g_raid3_softc *sc, struct g_consumer *cp)
415 {
416 	struct bio *bp;
417 	u_int nreqs = 0;
418 
419 	mtx_lock(&sc->sc_queue_mtx);
420 	TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
421 		if (bp->bio_from == cp)
422 			nreqs++;
423 	}
424 	mtx_unlock(&sc->sc_queue_mtx);
425 	return (nreqs);
426 }
427 
428 static int
g_raid3_is_busy(struct g_raid3_softc * sc,struct g_consumer * cp)429 g_raid3_is_busy(struct g_raid3_softc *sc, struct g_consumer *cp)
430 {
431 
432 	if (cp->index > 0) {
433 		G_RAID3_DEBUG(2,
434 		    "I/O requests for %s exist, can't destroy it now.",
435 		    cp->provider->name);
436 		return (1);
437 	}
438 	if (g_raid3_nrequests(sc, cp) > 0) {
439 		G_RAID3_DEBUG(2,
440 		    "I/O requests for %s in queue, can't destroy it now.",
441 		    cp->provider->name);
442 		return (1);
443 	}
444 	return (0);
445 }
446 
447 static void
g_raid3_destroy_consumer(void * arg,int flags __unused)448 g_raid3_destroy_consumer(void *arg, int flags __unused)
449 {
450 	struct g_consumer *cp;
451 
452 	g_topology_assert();
453 
454 	cp = arg;
455 	G_RAID3_DEBUG(1, "Consumer %s destroyed.", cp->provider->name);
456 	g_detach(cp);
457 	g_destroy_consumer(cp);
458 }
459 
460 static void
g_raid3_kill_consumer(struct g_raid3_softc * sc,struct g_consumer * cp)461 g_raid3_kill_consumer(struct g_raid3_softc *sc, struct g_consumer *cp)
462 {
463 	struct g_provider *pp;
464 	int retaste_wait;
465 
466 	g_topology_assert();
467 
468 	cp->private = NULL;
469 	if (g_raid3_is_busy(sc, cp))
470 		return;
471 	G_RAID3_DEBUG(2, "Consumer %s destroyed.", cp->provider->name);
472 	pp = cp->provider;
473 	retaste_wait = 0;
474 	if (cp->acw == 1) {
475 		if ((pp->geom->flags & G_GEOM_WITHER) == 0)
476 			retaste_wait = 1;
477 	}
478 	G_RAID3_DEBUG(2, "Access %s r%dw%de%d = %d", pp->name, -cp->acr,
479 	    -cp->acw, -cp->ace, 0);
480 	if (cp->acr > 0 || cp->acw > 0 || cp->ace > 0)
481 		g_access(cp, -cp->acr, -cp->acw, -cp->ace);
482 	if (retaste_wait) {
483 		/*
484 		 * After retaste event was send (inside g_access()), we can send
485 		 * event to detach and destroy consumer.
486 		 * A class, which has consumer to the given provider connected
487 		 * will not receive retaste event for the provider.
488 		 * This is the way how I ignore retaste events when I close
489 		 * consumers opened for write: I detach and destroy consumer
490 		 * after retaste event is sent.
491 		 */
492 		g_post_event(g_raid3_destroy_consumer, cp, M_WAITOK, NULL);
493 		return;
494 	}
495 	G_RAID3_DEBUG(1, "Consumer %s destroyed.", pp->name);
496 	g_detach(cp);
497 	g_destroy_consumer(cp);
498 }
499 
500 static int
g_raid3_connect_disk(struct g_raid3_disk * disk,struct g_provider * pp)501 g_raid3_connect_disk(struct g_raid3_disk *disk, struct g_provider *pp)
502 {
503 	struct g_consumer *cp;
504 	int error;
505 
506 	g_topology_assert_not();
507 	KASSERT(disk->d_consumer == NULL,
508 	    ("Disk already connected (device %s).", disk->d_softc->sc_name));
509 
510 	g_topology_lock();
511 	cp = g_new_consumer(disk->d_softc->sc_geom);
512 	error = g_attach(cp, pp);
513 	if (error != 0) {
514 		g_destroy_consumer(cp);
515 		g_topology_unlock();
516 		return (error);
517 	}
518 	error = g_access(cp, 1, 1, 1);
519 		g_topology_unlock();
520 	if (error != 0) {
521 		g_detach(cp);
522 		g_destroy_consumer(cp);
523 		G_RAID3_DEBUG(0, "Cannot open consumer %s (error=%d).",
524 		    pp->name, error);
525 		return (error);
526 	}
527 	disk->d_consumer = cp;
528 	disk->d_consumer->private = disk;
529 	disk->d_consumer->index = 0;
530 	G_RAID3_DEBUG(2, "Disk %s connected.", g_raid3_get_diskname(disk));
531 	return (0);
532 }
533 
534 static void
g_raid3_disconnect_consumer(struct g_raid3_softc * sc,struct g_consumer * cp)535 g_raid3_disconnect_consumer(struct g_raid3_softc *sc, struct g_consumer *cp)
536 {
537 
538 	g_topology_assert();
539 
540 	if (cp == NULL)
541 		return;
542 	if (cp->provider != NULL)
543 		g_raid3_kill_consumer(sc, cp);
544 	else
545 		g_destroy_consumer(cp);
546 }
547 
548 /*
549  * Initialize disk. This means allocate memory, create consumer, attach it
550  * to the provider and open access (r1w1e1) to it.
551  */
552 static struct g_raid3_disk *
g_raid3_init_disk(struct g_raid3_softc * sc,struct g_provider * pp,struct g_raid3_metadata * md,int * errorp)553 g_raid3_init_disk(struct g_raid3_softc *sc, struct g_provider *pp,
554     struct g_raid3_metadata *md, int *errorp)
555 {
556 	struct g_raid3_disk *disk;
557 	int error;
558 
559 	disk = &sc->sc_disks[md->md_no];
560 	error = g_raid3_connect_disk(disk, pp);
561 	if (error != 0) {
562 		if (errorp != NULL)
563 			*errorp = error;
564 		return (NULL);
565 	}
566 	disk->d_state = G_RAID3_DISK_STATE_NONE;
567 	disk->d_flags = md->md_dflags;
568 	error = g_getattr("GEOM::rotation_rate", disk->d_consumer,
569 	    &disk->d_rotation_rate);
570 	if (error != 0)
571 		disk->d_rotation_rate = DISK_RR_UNKNOWN;
572 	if (md->md_provider[0] != '\0')
573 		disk->d_flags |= G_RAID3_DISK_FLAG_HARDCODED;
574 	disk->d_sync.ds_consumer = NULL;
575 	disk->d_sync.ds_offset = md->md_sync_offset;
576 	disk->d_sync.ds_offset_done = md->md_sync_offset;
577 	disk->d_genid = md->md_genid;
578 	disk->d_sync.ds_syncid = md->md_syncid;
579 	if (errorp != NULL)
580 		*errorp = 0;
581 	return (disk);
582 }
583 
584 static void
g_raid3_destroy_disk(struct g_raid3_disk * disk)585 g_raid3_destroy_disk(struct g_raid3_disk *disk)
586 {
587 	struct g_raid3_softc *sc;
588 
589 	g_topology_assert_not();
590 	sc = disk->d_softc;
591 	sx_assert(&sc->sc_lock, SX_XLOCKED);
592 
593 	if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
594 		return;
595 	g_raid3_event_cancel(disk);
596 	switch (disk->d_state) {
597 	case G_RAID3_DISK_STATE_SYNCHRONIZING:
598 		if (sc->sc_syncdisk != NULL)
599 			g_raid3_sync_stop(sc, 1);
600 		/* FALLTHROUGH */
601 	case G_RAID3_DISK_STATE_NEW:
602 	case G_RAID3_DISK_STATE_STALE:
603 	case G_RAID3_DISK_STATE_ACTIVE:
604 		g_topology_lock();
605 		g_raid3_disconnect_consumer(sc, disk->d_consumer);
606 		g_topology_unlock();
607 		disk->d_consumer = NULL;
608 		break;
609 	default:
610 		KASSERT(0 == 1, ("Wrong disk state (%s, %s).",
611 		    g_raid3_get_diskname(disk),
612 		    g_raid3_disk_state2str(disk->d_state)));
613 	}
614 	disk->d_state = G_RAID3_DISK_STATE_NODISK;
615 }
616 
617 static void
g_raid3_free_device(struct g_raid3_softc * sc)618 g_raid3_free_device(struct g_raid3_softc *sc)
619 {
620 	KASSERT(sc->sc_refcnt == 0,
621 	    ("%s: non-zero refcount %u", __func__, sc->sc_refcnt));
622 
623 	if (!g_raid3_use_malloc) {
624 		uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_64K].sz_zone);
625 		uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_16K].sz_zone);
626 		uma_zdestroy(sc->sc_zones[G_RAID3_ZONE_4K].sz_zone);
627 	}
628 	mtx_destroy(&sc->sc_queue_mtx);
629 	mtx_destroy(&sc->sc_events_mtx);
630 	sx_xunlock(&sc->sc_lock);
631 	sx_destroy(&sc->sc_lock);
632 	free(sc->sc_disks, M_RAID3);
633 	free(sc, M_RAID3);
634 }
635 
636 static void
g_raid3_providergone(struct g_provider * pp)637 g_raid3_providergone(struct g_provider *pp)
638 {
639 	struct g_raid3_softc *sc = pp->private;
640 
641 	if (--sc->sc_refcnt == 0)
642 		g_raid3_free_device(sc);
643 }
644 
645 static void
g_raid3_destroy_device(struct g_raid3_softc * sc)646 g_raid3_destroy_device(struct g_raid3_softc *sc)
647 {
648 	struct g_raid3_event *ep;
649 	struct g_raid3_disk *disk;
650 	struct g_geom *gp;
651 	struct g_consumer *cp;
652 	u_int n;
653 
654 	g_topology_assert_not();
655 	sx_assert(&sc->sc_lock, SX_XLOCKED);
656 
657 	gp = sc->sc_geom;
658 	if (sc->sc_provider != NULL)
659 		g_raid3_destroy_provider(sc);
660 	for (n = 0; n < sc->sc_ndisks; n++) {
661 		disk = &sc->sc_disks[n];
662 		if (disk->d_state != G_RAID3_DISK_STATE_NODISK) {
663 			disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
664 			g_raid3_update_metadata(disk);
665 			g_raid3_destroy_disk(disk);
666 		}
667 	}
668 	while ((ep = g_raid3_event_get(sc)) != NULL) {
669 		g_raid3_event_remove(sc, ep);
670 		if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0)
671 			g_raid3_event_free(ep);
672 		else {
673 			ep->e_error = ECANCELED;
674 			ep->e_flags |= G_RAID3_EVENT_DONE;
675 			G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, ep);
676 			mtx_lock(&sc->sc_events_mtx);
677 			wakeup(ep);
678 			mtx_unlock(&sc->sc_events_mtx);
679 		}
680 	}
681 	g_raid3_timeout_drain(sc);
682 	cp = LIST_FIRST(&sc->sc_sync.ds_geom->consumer);
683 	g_topology_lock();
684 	if (cp != NULL)
685 		g_raid3_disconnect_consumer(sc, cp);
686 	g_wither_geom(sc->sc_sync.ds_geom, ENXIO);
687 	G_RAID3_DEBUG(0, "Device %s destroyed.", gp->name);
688 	g_wither_geom(gp, ENXIO);
689 	if (--sc->sc_refcnt == 0)
690 		g_raid3_free_device(sc);
691 	g_topology_unlock();
692 }
693 
694 static void
g_raid3_orphan(struct g_consumer * cp)695 g_raid3_orphan(struct g_consumer *cp)
696 {
697 	struct g_raid3_disk *disk;
698 
699 	g_topology_assert();
700 
701 	disk = cp->private;
702 	if (disk == NULL)
703 		return;
704 	disk->d_softc->sc_bump_id = G_RAID3_BUMP_SYNCID;
705 	g_raid3_event_send(disk, G_RAID3_DISK_STATE_DISCONNECTED,
706 	    G_RAID3_EVENT_DONTWAIT);
707 }
708 
709 static int
g_raid3_write_metadata(struct g_raid3_disk * disk,struct g_raid3_metadata * md)710 g_raid3_write_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md)
711 {
712 	struct g_raid3_softc *sc;
713 	struct g_consumer *cp;
714 	off_t offset, length;
715 	u_char *sector;
716 	int error = 0;
717 
718 	g_topology_assert_not();
719 	sc = disk->d_softc;
720 	sx_assert(&sc->sc_lock, SX_LOCKED);
721 
722 	cp = disk->d_consumer;
723 	KASSERT(cp != NULL, ("NULL consumer (%s).", sc->sc_name));
724 	KASSERT(cp->provider != NULL, ("NULL provider (%s).", sc->sc_name));
725 	KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
726 	    ("Consumer %s closed? (r%dw%de%d).", cp->provider->name, cp->acr,
727 	    cp->acw, cp->ace));
728 	length = cp->provider->sectorsize;
729 	offset = cp->provider->mediasize - length;
730 	sector = malloc((size_t)length, M_RAID3, M_WAITOK | M_ZERO);
731 	if (md != NULL)
732 		raid3_metadata_encode(md, sector);
733 	error = g_write_data(cp, offset, sector, length);
734 	free(sector, M_RAID3);
735 	if (error != 0) {
736 		if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
737 			G_RAID3_DEBUG(0, "Cannot write metadata on %s "
738 			    "(device=%s, error=%d).",
739 			    g_raid3_get_diskname(disk), sc->sc_name, error);
740 			disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
741 		} else {
742 			G_RAID3_DEBUG(1, "Cannot write metadata on %s "
743 			    "(device=%s, error=%d).",
744 			    g_raid3_get_diskname(disk), sc->sc_name, error);
745 		}
746 		if (g_raid3_disconnect_on_failure &&
747 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
748 			sc->sc_bump_id |= G_RAID3_BUMP_GENID;
749 			g_raid3_event_send(disk,
750 			    G_RAID3_DISK_STATE_DISCONNECTED,
751 			    G_RAID3_EVENT_DONTWAIT);
752 		}
753 	}
754 	return (error);
755 }
756 
757 int
g_raid3_clear_metadata(struct g_raid3_disk * disk)758 g_raid3_clear_metadata(struct g_raid3_disk *disk)
759 {
760 	int error;
761 
762 	g_topology_assert_not();
763 	sx_assert(&disk->d_softc->sc_lock, SX_LOCKED);
764 
765 	error = g_raid3_write_metadata(disk, NULL);
766 	if (error == 0) {
767 		G_RAID3_DEBUG(2, "Metadata on %s cleared.",
768 		    g_raid3_get_diskname(disk));
769 	} else {
770 		G_RAID3_DEBUG(0,
771 		    "Cannot clear metadata on disk %s (error=%d).",
772 		    g_raid3_get_diskname(disk), error);
773 	}
774 	return (error);
775 }
776 
777 void
g_raid3_fill_metadata(struct g_raid3_disk * disk,struct g_raid3_metadata * md)778 g_raid3_fill_metadata(struct g_raid3_disk *disk, struct g_raid3_metadata *md)
779 {
780 	struct g_raid3_softc *sc;
781 	struct g_provider *pp;
782 
783 	bzero(md, sizeof(*md));
784 	sc = disk->d_softc;
785 	strlcpy(md->md_magic, G_RAID3_MAGIC, sizeof(md->md_magic));
786 	md->md_version = G_RAID3_VERSION;
787 	strlcpy(md->md_name, sc->sc_name, sizeof(md->md_name));
788 	md->md_id = sc->sc_id;
789 	md->md_all = sc->sc_ndisks;
790 	md->md_genid = sc->sc_genid;
791 	md->md_mediasize = sc->sc_mediasize;
792 	md->md_sectorsize = sc->sc_sectorsize;
793 	md->md_mflags = (sc->sc_flags & G_RAID3_DEVICE_FLAG_MASK);
794 	md->md_no = disk->d_no;
795 	md->md_syncid = disk->d_sync.ds_syncid;
796 	md->md_dflags = (disk->d_flags & G_RAID3_DISK_FLAG_MASK);
797 	if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
798 		md->md_sync_offset =
799 		    disk->d_sync.ds_offset_done / (sc->sc_ndisks - 1);
800 	}
801 	if (disk->d_consumer != NULL && disk->d_consumer->provider != NULL)
802 		pp = disk->d_consumer->provider;
803 	else
804 		pp = NULL;
805 	if ((disk->d_flags & G_RAID3_DISK_FLAG_HARDCODED) != 0 && pp != NULL)
806 		strlcpy(md->md_provider, pp->name, sizeof(md->md_provider));
807 	if (pp != NULL)
808 		md->md_provsize = pp->mediasize;
809 }
810 
811 void
g_raid3_update_metadata(struct g_raid3_disk * disk)812 g_raid3_update_metadata(struct g_raid3_disk *disk)
813 {
814 	struct g_raid3_softc *sc __diagused;
815 	struct g_raid3_metadata md;
816 	int error;
817 
818 	g_topology_assert_not();
819 	sc = disk->d_softc;
820 	sx_assert(&sc->sc_lock, SX_LOCKED);
821 
822 	g_raid3_fill_metadata(disk, &md);
823 	error = g_raid3_write_metadata(disk, &md);
824 	if (error == 0) {
825 		G_RAID3_DEBUG(2, "Metadata on %s updated.",
826 		    g_raid3_get_diskname(disk));
827 	} else {
828 		G_RAID3_DEBUG(0,
829 		    "Cannot update metadata on disk %s (error=%d).",
830 		    g_raid3_get_diskname(disk), error);
831 	}
832 }
833 
834 static void
g_raid3_bump_syncid(struct g_raid3_softc * sc)835 g_raid3_bump_syncid(struct g_raid3_softc *sc)
836 {
837 	struct g_raid3_disk *disk;
838 	u_int n;
839 
840 	g_topology_assert_not();
841 	sx_assert(&sc->sc_lock, SX_XLOCKED);
842 	KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0,
843 	    ("%s called with no active disks (device=%s).", __func__,
844 	    sc->sc_name));
845 
846 	sc->sc_syncid++;
847 	G_RAID3_DEBUG(1, "Device %s: syncid bumped to %u.", sc->sc_name,
848 	    sc->sc_syncid);
849 	for (n = 0; n < sc->sc_ndisks; n++) {
850 		disk = &sc->sc_disks[n];
851 		if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
852 		    disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
853 			disk->d_sync.ds_syncid = sc->sc_syncid;
854 			g_raid3_update_metadata(disk);
855 		}
856 	}
857 }
858 
859 static void
g_raid3_bump_genid(struct g_raid3_softc * sc)860 g_raid3_bump_genid(struct g_raid3_softc *sc)
861 {
862 	struct g_raid3_disk *disk;
863 	u_int n;
864 
865 	g_topology_assert_not();
866 	sx_assert(&sc->sc_lock, SX_XLOCKED);
867 	KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) > 0,
868 	    ("%s called with no active disks (device=%s).", __func__,
869 	    sc->sc_name));
870 
871 	sc->sc_genid++;
872 	G_RAID3_DEBUG(1, "Device %s: genid bumped to %u.", sc->sc_name,
873 	    sc->sc_genid);
874 	for (n = 0; n < sc->sc_ndisks; n++) {
875 		disk = &sc->sc_disks[n];
876 		if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
877 		    disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
878 			disk->d_genid = sc->sc_genid;
879 			g_raid3_update_metadata(disk);
880 		}
881 	}
882 }
883 
884 static int
g_raid3_idle(struct g_raid3_softc * sc,int acw)885 g_raid3_idle(struct g_raid3_softc *sc, int acw)
886 {
887 	struct g_raid3_disk *disk;
888 	u_int i;
889 	int timeout;
890 
891 	g_topology_assert_not();
892 	sx_assert(&sc->sc_lock, SX_XLOCKED);
893 
894 	if (sc->sc_provider == NULL)
895 		return (0);
896 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
897 		return (0);
898 	if (sc->sc_idle)
899 		return (0);
900 	if (sc->sc_writes > 0)
901 		return (0);
902 	if (acw > 0 || (acw == -1 && sc->sc_provider->acw > 0)) {
903 		timeout = g_raid3_idletime - (time_uptime - sc->sc_last_write);
904 		if (!g_raid3_shutdown && timeout > 0)
905 			return (timeout);
906 	}
907 	sc->sc_idle = 1;
908 	for (i = 0; i < sc->sc_ndisks; i++) {
909 		disk = &sc->sc_disks[i];
910 		if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
911 			continue;
912 		G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.",
913 		    g_raid3_get_diskname(disk), sc->sc_name);
914 		disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
915 		g_raid3_update_metadata(disk);
916 	}
917 	return (0);
918 }
919 
920 static void
g_raid3_unidle(struct g_raid3_softc * sc)921 g_raid3_unidle(struct g_raid3_softc *sc)
922 {
923 	struct g_raid3_disk *disk;
924 	u_int i;
925 
926 	g_topology_assert_not();
927 	sx_assert(&sc->sc_lock, SX_XLOCKED);
928 
929 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
930 		return;
931 	sc->sc_idle = 0;
932 	sc->sc_last_write = time_uptime;
933 	for (i = 0; i < sc->sc_ndisks; i++) {
934 		disk = &sc->sc_disks[i];
935 		if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
936 			continue;
937 		G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.",
938 		    g_raid3_get_diskname(disk), sc->sc_name);
939 		disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
940 		g_raid3_update_metadata(disk);
941 	}
942 }
943 
944 /*
945  * Treat bio_driver1 field in parent bio as list head and field bio_caller1
946  * in child bio as pointer to the next element on the list.
947  */
948 #define	G_RAID3_HEAD_BIO(pbp)	(pbp)->bio_driver1
949 
950 #define	G_RAID3_NEXT_BIO(cbp)	(cbp)->bio_caller1
951 
952 #define	G_RAID3_FOREACH_BIO(pbp, bp)					\
953 	for ((bp) = G_RAID3_HEAD_BIO(pbp); (bp) != NULL;		\
954 	    (bp) = G_RAID3_NEXT_BIO(bp))
955 
956 #define	G_RAID3_FOREACH_SAFE_BIO(pbp, bp, tmpbp)			\
957 	for ((bp) = G_RAID3_HEAD_BIO(pbp);				\
958 	    (bp) != NULL && ((tmpbp) = G_RAID3_NEXT_BIO(bp), 1);	\
959 	    (bp) = (tmpbp))
960 
961 static void
g_raid3_init_bio(struct bio * pbp)962 g_raid3_init_bio(struct bio *pbp)
963 {
964 
965 	G_RAID3_HEAD_BIO(pbp) = NULL;
966 }
967 
968 static void
g_raid3_remove_bio(struct bio * cbp)969 g_raid3_remove_bio(struct bio *cbp)
970 {
971 	struct bio *pbp, *bp;
972 
973 	pbp = cbp->bio_parent;
974 	if (G_RAID3_HEAD_BIO(pbp) == cbp)
975 		G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp);
976 	else {
977 		G_RAID3_FOREACH_BIO(pbp, bp) {
978 			if (G_RAID3_NEXT_BIO(bp) == cbp) {
979 				G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp);
980 				break;
981 			}
982 		}
983 	}
984 	G_RAID3_NEXT_BIO(cbp) = NULL;
985 }
986 
987 static void
g_raid3_replace_bio(struct bio * sbp,struct bio * dbp)988 g_raid3_replace_bio(struct bio *sbp, struct bio *dbp)
989 {
990 	struct bio *pbp, *bp;
991 
992 	g_raid3_remove_bio(sbp);
993 	pbp = dbp->bio_parent;
994 	G_RAID3_NEXT_BIO(sbp) = G_RAID3_NEXT_BIO(dbp);
995 	if (G_RAID3_HEAD_BIO(pbp) == dbp)
996 		G_RAID3_HEAD_BIO(pbp) = sbp;
997 	else {
998 		G_RAID3_FOREACH_BIO(pbp, bp) {
999 			if (G_RAID3_NEXT_BIO(bp) == dbp) {
1000 				G_RAID3_NEXT_BIO(bp) = sbp;
1001 				break;
1002 			}
1003 		}
1004 	}
1005 	G_RAID3_NEXT_BIO(dbp) = NULL;
1006 }
1007 
1008 static void
g_raid3_destroy_bio(struct g_raid3_softc * sc,struct bio * cbp)1009 g_raid3_destroy_bio(struct g_raid3_softc *sc, struct bio *cbp)
1010 {
1011 	struct bio *bp, *pbp;
1012 	size_t size;
1013 
1014 	pbp = cbp->bio_parent;
1015 	pbp->bio_children--;
1016 	KASSERT(cbp->bio_data != NULL, ("NULL bio_data"));
1017 	size = pbp->bio_length / (sc->sc_ndisks - 1);
1018 	g_raid3_free(sc, cbp->bio_data, size);
1019 	if (G_RAID3_HEAD_BIO(pbp) == cbp) {
1020 		G_RAID3_HEAD_BIO(pbp) = G_RAID3_NEXT_BIO(cbp);
1021 		G_RAID3_NEXT_BIO(cbp) = NULL;
1022 		g_destroy_bio(cbp);
1023 	} else {
1024 		G_RAID3_FOREACH_BIO(pbp, bp) {
1025 			if (G_RAID3_NEXT_BIO(bp) == cbp)
1026 				break;
1027 		}
1028 		if (bp != NULL) {
1029 			KASSERT(G_RAID3_NEXT_BIO(bp) != NULL,
1030 			    ("NULL bp->bio_driver1"));
1031 			G_RAID3_NEXT_BIO(bp) = G_RAID3_NEXT_BIO(cbp);
1032 			G_RAID3_NEXT_BIO(cbp) = NULL;
1033 		}
1034 		g_destroy_bio(cbp);
1035 	}
1036 }
1037 
1038 static struct bio *
g_raid3_clone_bio(struct g_raid3_softc * sc,struct bio * pbp)1039 g_raid3_clone_bio(struct g_raid3_softc *sc, struct bio *pbp)
1040 {
1041 	struct bio *bp, *cbp;
1042 	size_t size;
1043 	int memflag;
1044 
1045 	cbp = g_clone_bio(pbp);
1046 	if (cbp == NULL)
1047 		return (NULL);
1048 	size = pbp->bio_length / (sc->sc_ndisks - 1);
1049 	if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0)
1050 		memflag = M_WAITOK;
1051 	else
1052 		memflag = M_NOWAIT;
1053 	cbp->bio_data = g_raid3_alloc(sc, size, memflag);
1054 	if (cbp->bio_data == NULL) {
1055 		pbp->bio_children--;
1056 		g_destroy_bio(cbp);
1057 		return (NULL);
1058 	}
1059 	G_RAID3_NEXT_BIO(cbp) = NULL;
1060 	if (G_RAID3_HEAD_BIO(pbp) == NULL)
1061 		G_RAID3_HEAD_BIO(pbp) = cbp;
1062 	else {
1063 		G_RAID3_FOREACH_BIO(pbp, bp) {
1064 			if (G_RAID3_NEXT_BIO(bp) == NULL) {
1065 				G_RAID3_NEXT_BIO(bp) = cbp;
1066 				break;
1067 			}
1068 		}
1069 	}
1070 	return (cbp);
1071 }
1072 
1073 static void
g_raid3_scatter(struct bio * pbp)1074 g_raid3_scatter(struct bio *pbp)
1075 {
1076 	struct g_raid3_softc *sc;
1077 	struct g_raid3_disk *disk;
1078 	struct bio *bp, *cbp, *tmpbp;
1079 	off_t atom, cadd, padd, left;
1080 	int first;
1081 
1082 	sc = pbp->bio_to->private;
1083 	bp = NULL;
1084 	if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) {
1085 		/*
1086 		 * Find bio for which we should calculate data.
1087 		 */
1088 		G_RAID3_FOREACH_BIO(pbp, cbp) {
1089 			if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) {
1090 				bp = cbp;
1091 				break;
1092 			}
1093 		}
1094 		KASSERT(bp != NULL, ("NULL parity bio."));
1095 	}
1096 	atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1097 	cadd = padd = 0;
1098 	for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) {
1099 		G_RAID3_FOREACH_BIO(pbp, cbp) {
1100 			if (cbp == bp)
1101 				continue;
1102 			bcopy(pbp->bio_data + padd, cbp->bio_data + cadd, atom);
1103 			padd += atom;
1104 		}
1105 		cadd += atom;
1106 	}
1107 	if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_NOPARITY) == 0) {
1108 		/*
1109 		 * Calculate parity.
1110 		 */
1111 		first = 1;
1112 		G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1113 			if (cbp == bp)
1114 				continue;
1115 			if (first) {
1116 				bcopy(cbp->bio_data, bp->bio_data,
1117 				    bp->bio_length);
1118 				first = 0;
1119 			} else {
1120 				g_raid3_xor(cbp->bio_data, bp->bio_data,
1121 				    bp->bio_length);
1122 			}
1123 			if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_NODISK) != 0)
1124 				g_raid3_destroy_bio(sc, cbp);
1125 		}
1126 	}
1127 	G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1128 		struct g_consumer *cp;
1129 
1130 		disk = cbp->bio_caller2;
1131 		cp = disk->d_consumer;
1132 		cbp->bio_to = cp->provider;
1133 		G_RAID3_LOGREQ(3, cbp, "Sending request.");
1134 		KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1135 		    ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1136 		    cp->acr, cp->acw, cp->ace));
1137 		cp->index++;
1138 		sc->sc_writes++;
1139 		g_io_request(cbp, cp);
1140 	}
1141 }
1142 
1143 static void
g_raid3_gather(struct bio * pbp)1144 g_raid3_gather(struct bio *pbp)
1145 {
1146 	struct g_raid3_softc *sc;
1147 	struct g_raid3_disk *disk;
1148 	struct bio *xbp, *fbp, *cbp;
1149 	off_t atom, cadd, padd, left;
1150 
1151 	sc = pbp->bio_to->private;
1152 	/*
1153 	 * Find bio for which we have to calculate data.
1154 	 * While going through this path, check if all requests
1155 	 * succeeded, if not, deny whole request.
1156 	 * If we're in COMPLETE mode, we allow one request to fail,
1157 	 * so if we find one, we're sending it to the parity consumer.
1158 	 * If there are more failed requests, we deny whole request.
1159 	 */
1160 	xbp = fbp = NULL;
1161 	G_RAID3_FOREACH_BIO(pbp, cbp) {
1162 		if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0) {
1163 			KASSERT(xbp == NULL, ("More than one parity bio."));
1164 			xbp = cbp;
1165 		}
1166 		if (cbp->bio_error == 0)
1167 			continue;
1168 		/*
1169 		 * Found failed request.
1170 		 */
1171 		if (fbp == NULL) {
1172 			if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_DEGRADED) != 0) {
1173 				/*
1174 				 * We are already in degraded mode, so we can't
1175 				 * accept any failures.
1176 				 */
1177 				if (pbp->bio_error == 0)
1178 					pbp->bio_error = cbp->bio_error;
1179 			} else {
1180 				fbp = cbp;
1181 			}
1182 		} else {
1183 			/*
1184 			 * Next failed request, that's too many.
1185 			 */
1186 			if (pbp->bio_error == 0)
1187 				pbp->bio_error = fbp->bio_error;
1188 		}
1189 		disk = cbp->bio_caller2;
1190 		if (disk == NULL)
1191 			continue;
1192 		if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
1193 			disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
1194 			G_RAID3_LOGREQ(0, cbp, "Request failed (error=%d).",
1195 			    cbp->bio_error);
1196 		} else {
1197 			G_RAID3_LOGREQ(1, cbp, "Request failed (error=%d).",
1198 			    cbp->bio_error);
1199 		}
1200 		if (g_raid3_disconnect_on_failure &&
1201 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1202 			sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1203 			g_raid3_event_send(disk,
1204 			    G_RAID3_DISK_STATE_DISCONNECTED,
1205 			    G_RAID3_EVENT_DONTWAIT);
1206 		}
1207 	}
1208 	if (pbp->bio_error != 0)
1209 		goto finish;
1210 	if (fbp != NULL && (pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) {
1211 		pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_VERIFY;
1212 		if (xbp != fbp)
1213 			g_raid3_replace_bio(xbp, fbp);
1214 		g_raid3_destroy_bio(sc, fbp);
1215 	} else if (fbp != NULL) {
1216 		struct g_consumer *cp;
1217 
1218 		/*
1219 		 * One request failed, so send the same request to
1220 		 * the parity consumer.
1221 		 */
1222 		disk = pbp->bio_driver2;
1223 		if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) {
1224 			pbp->bio_error = fbp->bio_error;
1225 			goto finish;
1226 		}
1227 		pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1228 		pbp->bio_inbed--;
1229 		fbp->bio_flags &= ~(BIO_DONE | BIO_ERROR);
1230 		if (disk->d_no == sc->sc_ndisks - 1)
1231 			fbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1232 		fbp->bio_error = 0;
1233 		fbp->bio_completed = 0;
1234 		fbp->bio_children = 0;
1235 		fbp->bio_inbed = 0;
1236 		cp = disk->d_consumer;
1237 		fbp->bio_caller2 = disk;
1238 		fbp->bio_to = cp->provider;
1239 		G_RAID3_LOGREQ(3, fbp, "Sending request (recover).");
1240 		KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1241 		    ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1242 		    cp->acr, cp->acw, cp->ace));
1243 		cp->index++;
1244 		g_io_request(fbp, cp);
1245 		return;
1246 	}
1247 	if (xbp != NULL) {
1248 		/*
1249 		 * Calculate parity.
1250 		 */
1251 		G_RAID3_FOREACH_BIO(pbp, cbp) {
1252 			if ((cbp->bio_cflags & G_RAID3_BIO_CFLAG_PARITY) != 0)
1253 				continue;
1254 			g_raid3_xor(cbp->bio_data, xbp->bio_data,
1255 			    xbp->bio_length);
1256 		}
1257 		xbp->bio_cflags &= ~G_RAID3_BIO_CFLAG_PARITY;
1258 		if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0) {
1259 			if (!g_raid3_is_zero(xbp)) {
1260 				g_raid3_parity_mismatch++;
1261 				pbp->bio_error = EIO;
1262 				goto finish;
1263 			}
1264 			g_raid3_destroy_bio(sc, xbp);
1265 		}
1266 	}
1267 	atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1268 	cadd = padd = 0;
1269 	for (left = pbp->bio_length; left > 0; left -= sc->sc_sectorsize) {
1270 		G_RAID3_FOREACH_BIO(pbp, cbp) {
1271 			bcopy(cbp->bio_data + cadd, pbp->bio_data + padd, atom);
1272 			pbp->bio_completed += atom;
1273 			padd += atom;
1274 		}
1275 		cadd += atom;
1276 	}
1277 finish:
1278 	if (pbp->bio_error == 0)
1279 		G_RAID3_LOGREQ(3, pbp, "Request finished.");
1280 	else {
1281 		if ((pbp->bio_pflags & G_RAID3_BIO_PFLAG_VERIFY) != 0)
1282 			G_RAID3_LOGREQ(1, pbp, "Verification error.");
1283 		else
1284 			G_RAID3_LOGREQ(0, pbp, "Request failed.");
1285 	}
1286 	pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_MASK;
1287 	while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL)
1288 		g_raid3_destroy_bio(sc, cbp);
1289 	g_io_deliver(pbp, pbp->bio_error);
1290 }
1291 
1292 static void
g_raid3_done(struct bio * bp)1293 g_raid3_done(struct bio *bp)
1294 {
1295 	struct g_raid3_softc *sc;
1296 
1297 	sc = bp->bio_from->geom->softc;
1298 	bp->bio_cflags |= G_RAID3_BIO_CFLAG_REGULAR;
1299 	G_RAID3_LOGREQ(3, bp, "Regular request done (error=%d).", bp->bio_error);
1300 	mtx_lock(&sc->sc_queue_mtx);
1301 	bioq_insert_head(&sc->sc_queue, bp);
1302 	mtx_unlock(&sc->sc_queue_mtx);
1303 	wakeup(sc);
1304 	wakeup(&sc->sc_queue);
1305 }
1306 
1307 static void
g_raid3_regular_request(struct bio * cbp)1308 g_raid3_regular_request(struct bio *cbp)
1309 {
1310 	struct g_raid3_softc *sc;
1311 	struct g_raid3_disk *disk;
1312 	struct bio *pbp;
1313 
1314 	g_topology_assert_not();
1315 
1316 	pbp = cbp->bio_parent;
1317 	sc = pbp->bio_to->private;
1318 	cbp->bio_from->index--;
1319 	if (cbp->bio_cmd == BIO_WRITE)
1320 		sc->sc_writes--;
1321 	disk = cbp->bio_from->private;
1322 	if (disk == NULL) {
1323 		g_topology_lock();
1324 		g_raid3_kill_consumer(sc, cbp->bio_from);
1325 		g_topology_unlock();
1326 	}
1327 
1328 	G_RAID3_LOGREQ(3, cbp, "Request finished.");
1329 	pbp->bio_inbed++;
1330 	KASSERT(pbp->bio_inbed <= pbp->bio_children,
1331 	    ("bio_inbed (%u) is bigger than bio_children (%u).", pbp->bio_inbed,
1332 	    pbp->bio_children));
1333 	if (pbp->bio_inbed != pbp->bio_children)
1334 		return;
1335 	switch (pbp->bio_cmd) {
1336 	case BIO_READ:
1337 		g_raid3_gather(pbp);
1338 		break;
1339 	case BIO_WRITE:
1340 	case BIO_DELETE:
1341 	    {
1342 		int error = 0;
1343 
1344 		pbp->bio_completed = pbp->bio_length;
1345 		while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL) {
1346 			if (cbp->bio_error == 0) {
1347 				g_raid3_destroy_bio(sc, cbp);
1348 				continue;
1349 			}
1350 
1351 			if (error == 0)
1352 				error = cbp->bio_error;
1353 			else if (pbp->bio_error == 0) {
1354 				/*
1355 				 * Next failed request, that's too many.
1356 				 */
1357 				pbp->bio_error = error;
1358 			}
1359 
1360 			disk = cbp->bio_caller2;
1361 			if (disk == NULL) {
1362 				g_raid3_destroy_bio(sc, cbp);
1363 				continue;
1364 			}
1365 
1366 			if ((disk->d_flags & G_RAID3_DISK_FLAG_BROKEN) == 0) {
1367 				disk->d_flags |= G_RAID3_DISK_FLAG_BROKEN;
1368 				G_RAID3_LOGREQ(0, cbp,
1369 				    "Request failed (error=%d).",
1370 				    cbp->bio_error);
1371 			} else {
1372 				G_RAID3_LOGREQ(1, cbp,
1373 				    "Request failed (error=%d).",
1374 				    cbp->bio_error);
1375 			}
1376 			if (g_raid3_disconnect_on_failure &&
1377 			    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1378 				sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1379 				g_raid3_event_send(disk,
1380 				    G_RAID3_DISK_STATE_DISCONNECTED,
1381 				    G_RAID3_EVENT_DONTWAIT);
1382 			}
1383 			g_raid3_destroy_bio(sc, cbp);
1384 		}
1385 		if (pbp->bio_error == 0)
1386 			G_RAID3_LOGREQ(3, pbp, "Request finished.");
1387 		else
1388 			G_RAID3_LOGREQ(0, pbp, "Request failed.");
1389 		pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_DEGRADED;
1390 		pbp->bio_pflags &= ~G_RAID3_BIO_PFLAG_NOPARITY;
1391 		bioq_remove(&sc->sc_inflight, pbp);
1392 		/* Release delayed sync requests if possible. */
1393 		g_raid3_sync_release(sc);
1394 		g_io_deliver(pbp, pbp->bio_error);
1395 		break;
1396 	    }
1397 	}
1398 }
1399 
1400 static void
g_raid3_sync_done(struct bio * bp)1401 g_raid3_sync_done(struct bio *bp)
1402 {
1403 	struct g_raid3_softc *sc;
1404 
1405 	G_RAID3_LOGREQ(3, bp, "Synchronization request delivered.");
1406 	sc = bp->bio_from->geom->softc;
1407 	bp->bio_cflags |= G_RAID3_BIO_CFLAG_SYNC;
1408 	mtx_lock(&sc->sc_queue_mtx);
1409 	bioq_insert_head(&sc->sc_queue, bp);
1410 	mtx_unlock(&sc->sc_queue_mtx);
1411 	wakeup(sc);
1412 	wakeup(&sc->sc_queue);
1413 }
1414 
1415 static void
g_raid3_flush(struct g_raid3_softc * sc,struct bio * bp)1416 g_raid3_flush(struct g_raid3_softc *sc, struct bio *bp)
1417 {
1418 	struct bio_queue_head queue;
1419 	struct g_raid3_disk *disk;
1420 	struct g_consumer *cp __diagused;
1421 	struct bio *cbp;
1422 	u_int i;
1423 
1424 	bioq_init(&queue);
1425 	for (i = 0; i < sc->sc_ndisks; i++) {
1426 		disk = &sc->sc_disks[i];
1427 		if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE)
1428 			continue;
1429 		cbp = g_clone_bio(bp);
1430 		if (cbp == NULL) {
1431 			for (cbp = bioq_first(&queue); cbp != NULL;
1432 			    cbp = bioq_first(&queue)) {
1433 				bioq_remove(&queue, cbp);
1434 				g_destroy_bio(cbp);
1435 			}
1436 			if (bp->bio_error == 0)
1437 				bp->bio_error = ENOMEM;
1438 			g_io_deliver(bp, bp->bio_error);
1439 			return;
1440 		}
1441 		bioq_insert_tail(&queue, cbp);
1442 		cbp->bio_done = g_std_done;
1443 		cbp->bio_caller1 = disk;
1444 		cbp->bio_to = disk->d_consumer->provider;
1445 	}
1446 	for (cbp = bioq_first(&queue); cbp != NULL; cbp = bioq_first(&queue)) {
1447 		bioq_remove(&queue, cbp);
1448 		G_RAID3_LOGREQ(3, cbp, "Sending request.");
1449 		disk = cbp->bio_caller1;
1450 		cbp->bio_caller1 = NULL;
1451 		cp = disk->d_consumer;
1452 		KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1453 		    ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1454 		    cp->acr, cp->acw, cp->ace));
1455 		g_io_request(cbp, disk->d_consumer);
1456 	}
1457 }
1458 
1459 static void
g_raid3_rotation_rate(struct g_raid3_softc * sc,struct bio * bp)1460 g_raid3_rotation_rate(struct g_raid3_softc *sc, struct bio *bp)
1461 {
1462 	struct g_raid3_disk *disk;
1463 	bool first = true;
1464 	uint16_t rr = DISK_RR_UNKNOWN;
1465 	u_int n;
1466 
1467 	for (n = 0; n < sc->sc_ndisks; n++) {
1468 		disk = &sc->sc_disks[n];
1469 		if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
1470 			continue;
1471 		if (first)
1472 			rr = disk->d_rotation_rate;
1473 		else if (rr != disk->d_rotation_rate) {
1474 			rr = DISK_RR_UNKNOWN;
1475 			break;
1476 		}
1477 		first = false;
1478 	}
1479 	g_handleattr(bp, "GEOM::rotation_rate", &rr, sizeof(rr));
1480 }
1481 
1482 static void
g_raid3_start(struct bio * bp)1483 g_raid3_start(struct bio *bp)
1484 {
1485 	struct g_raid3_softc *sc;
1486 
1487 	sc = bp->bio_to->private;
1488 	/*
1489 	 * If sc == NULL or there are no valid disks, provider's error
1490 	 * should be set and g_raid3_start() should not be called at all.
1491 	 */
1492 	KASSERT(sc != NULL && (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
1493 	    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE),
1494 	    ("Provider's error should be set (error=%d)(device=%s).",
1495 	    bp->bio_to->error, bp->bio_to->name));
1496 	G_RAID3_LOGREQ(3, bp, "Request received.");
1497 
1498 	switch (bp->bio_cmd) {
1499 	case BIO_READ:
1500 	case BIO_WRITE:
1501 	case BIO_DELETE:
1502 		break;
1503 	case BIO_SPEEDUP:
1504 	case BIO_FLUSH:
1505 		g_raid3_flush(sc, bp);
1506 		return;
1507 	case BIO_GETATTR:
1508 		if (!strcmp(bp->bio_attribute, "GEOM::rotation_rate")) {
1509 			g_raid3_rotation_rate(sc, bp);
1510 			return;
1511 		}
1512 		/* FALLTHROUGH */
1513 	default:
1514 		g_io_deliver(bp, EOPNOTSUPP);
1515 		return;
1516 	}
1517 	mtx_lock(&sc->sc_queue_mtx);
1518 	bioq_insert_tail(&sc->sc_queue, bp);
1519 	mtx_unlock(&sc->sc_queue_mtx);
1520 	G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
1521 	wakeup(sc);
1522 }
1523 
1524 /*
1525  * Return TRUE if the given request is colliding with a in-progress
1526  * synchronization request.
1527  */
1528 static int
g_raid3_sync_collision(struct g_raid3_softc * sc,struct bio * bp)1529 g_raid3_sync_collision(struct g_raid3_softc *sc, struct bio *bp)
1530 {
1531 	struct g_raid3_disk *disk;
1532 	struct bio *sbp;
1533 	off_t rstart, rend, sstart, send;
1534 	int i;
1535 
1536 	disk = sc->sc_syncdisk;
1537 	if (disk == NULL)
1538 		return (0);
1539 	rstart = bp->bio_offset;
1540 	rend = bp->bio_offset + bp->bio_length;
1541 	for (i = 0; i < g_raid3_syncreqs; i++) {
1542 		sbp = disk->d_sync.ds_bios[i];
1543 		if (sbp == NULL)
1544 			continue;
1545 		sstart = sbp->bio_offset;
1546 		send = sbp->bio_length;
1547 		if (sbp->bio_cmd == BIO_WRITE) {
1548 			sstart *= sc->sc_ndisks - 1;
1549 			send *= sc->sc_ndisks - 1;
1550 		}
1551 		send += sstart;
1552 		if (rend > sstart && rstart < send)
1553 			return (1);
1554 	}
1555 	return (0);
1556 }
1557 
1558 /*
1559  * Return TRUE if the given sync request is colliding with a in-progress regular
1560  * request.
1561  */
1562 static int
g_raid3_regular_collision(struct g_raid3_softc * sc,struct bio * sbp)1563 g_raid3_regular_collision(struct g_raid3_softc *sc, struct bio *sbp)
1564 {
1565 	off_t rstart, rend, sstart, send;
1566 	struct bio *bp;
1567 
1568 	if (sc->sc_syncdisk == NULL)
1569 		return (0);
1570 	sstart = sbp->bio_offset;
1571 	send = sstart + sbp->bio_length;
1572 	TAILQ_FOREACH(bp, &sc->sc_inflight.queue, bio_queue) {
1573 		rstart = bp->bio_offset;
1574 		rend = bp->bio_offset + bp->bio_length;
1575 		if (rend > sstart && rstart < send)
1576 			return (1);
1577 	}
1578 	return (0);
1579 }
1580 
1581 /*
1582  * Puts request onto delayed queue.
1583  */
1584 static void
g_raid3_regular_delay(struct g_raid3_softc * sc,struct bio * bp)1585 g_raid3_regular_delay(struct g_raid3_softc *sc, struct bio *bp)
1586 {
1587 
1588 	G_RAID3_LOGREQ(2, bp, "Delaying request.");
1589 	bioq_insert_head(&sc->sc_regular_delayed, bp);
1590 }
1591 
1592 /*
1593  * Puts synchronization request onto delayed queue.
1594  */
1595 static void
g_raid3_sync_delay(struct g_raid3_softc * sc,struct bio * bp)1596 g_raid3_sync_delay(struct g_raid3_softc *sc, struct bio *bp)
1597 {
1598 
1599 	G_RAID3_LOGREQ(2, bp, "Delaying synchronization request.");
1600 	bioq_insert_tail(&sc->sc_sync_delayed, bp);
1601 }
1602 
1603 /*
1604  * Releases delayed regular requests which don't collide anymore with sync
1605  * requests.
1606  */
1607 static void
g_raid3_regular_release(struct g_raid3_softc * sc)1608 g_raid3_regular_release(struct g_raid3_softc *sc)
1609 {
1610 	struct bio *bp, *bp2;
1611 
1612 	TAILQ_FOREACH_SAFE(bp, &sc->sc_regular_delayed.queue, bio_queue, bp2) {
1613 		if (g_raid3_sync_collision(sc, bp))
1614 			continue;
1615 		bioq_remove(&sc->sc_regular_delayed, bp);
1616 		G_RAID3_LOGREQ(2, bp, "Releasing delayed request (%p).", bp);
1617 		mtx_lock(&sc->sc_queue_mtx);
1618 		bioq_insert_head(&sc->sc_queue, bp);
1619 #if 0
1620 		/*
1621 		 * wakeup() is not needed, because this function is called from
1622 		 * the worker thread.
1623 		 */
1624 		wakeup(&sc->sc_queue);
1625 #endif
1626 		mtx_unlock(&sc->sc_queue_mtx);
1627 	}
1628 }
1629 
1630 /*
1631  * Releases delayed sync requests which don't collide anymore with regular
1632  * requests.
1633  */
1634 static void
g_raid3_sync_release(struct g_raid3_softc * sc)1635 g_raid3_sync_release(struct g_raid3_softc *sc)
1636 {
1637 	struct bio *bp, *bp2;
1638 
1639 	TAILQ_FOREACH_SAFE(bp, &sc->sc_sync_delayed.queue, bio_queue, bp2) {
1640 		if (g_raid3_regular_collision(sc, bp))
1641 			continue;
1642 		bioq_remove(&sc->sc_sync_delayed, bp);
1643 		G_RAID3_LOGREQ(2, bp,
1644 		    "Releasing delayed synchronization request.");
1645 		g_io_request(bp, bp->bio_from);
1646 	}
1647 }
1648 
1649 /*
1650  * Handle synchronization requests.
1651  * Every synchronization request is two-steps process: first, READ request is
1652  * send to active provider and then WRITE request (with read data) to the provider
1653  * being synchronized. When WRITE is finished, new synchronization request is
1654  * send.
1655  */
1656 static void
g_raid3_sync_request(struct bio * bp)1657 g_raid3_sync_request(struct bio *bp)
1658 {
1659 	struct g_raid3_softc *sc;
1660 	struct g_raid3_disk *disk;
1661 
1662 	bp->bio_from->index--;
1663 	sc = bp->bio_from->geom->softc;
1664 	disk = bp->bio_from->private;
1665 	if (disk == NULL) {
1666 		sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
1667 		g_topology_lock();
1668 		g_raid3_kill_consumer(sc, bp->bio_from);
1669 		g_topology_unlock();
1670 		free(bp->bio_data, M_RAID3);
1671 		g_destroy_bio(bp);
1672 		sx_xlock(&sc->sc_lock);
1673 		return;
1674 	}
1675 
1676 	/*
1677 	 * Synchronization request.
1678 	 */
1679 	switch (bp->bio_cmd) {
1680 	case BIO_READ:
1681 	    {
1682 		struct g_consumer *cp;
1683 		u_char *dst, *src;
1684 		off_t left;
1685 		u_int atom;
1686 
1687 		if (bp->bio_error != 0) {
1688 			G_RAID3_LOGREQ(0, bp,
1689 			    "Synchronization request failed (error=%d).",
1690 			    bp->bio_error);
1691 			g_destroy_bio(bp);
1692 			return;
1693 		}
1694 		G_RAID3_LOGREQ(3, bp, "Synchronization request finished.");
1695 		atom = sc->sc_sectorsize / (sc->sc_ndisks - 1);
1696 		dst = src = bp->bio_data;
1697 		if (disk->d_no == sc->sc_ndisks - 1) {
1698 			u_int n;
1699 
1700 			/* Parity component. */
1701 			for (left = bp->bio_length; left > 0;
1702 			    left -= sc->sc_sectorsize) {
1703 				bcopy(src, dst, atom);
1704 				src += atom;
1705 				for (n = 1; n < sc->sc_ndisks - 1; n++) {
1706 					g_raid3_xor(src, dst, atom);
1707 					src += atom;
1708 				}
1709 				dst += atom;
1710 			}
1711 		} else {
1712 			/* Regular component. */
1713 			src += atom * disk->d_no;
1714 			for (left = bp->bio_length; left > 0;
1715 			    left -= sc->sc_sectorsize) {
1716 				bcopy(src, dst, atom);
1717 				src += sc->sc_sectorsize;
1718 				dst += atom;
1719 			}
1720 		}
1721 		bp->bio_driver1 = bp->bio_driver2 = NULL;
1722 		bp->bio_pflags = 0;
1723 		bp->bio_offset /= sc->sc_ndisks - 1;
1724 		bp->bio_length /= sc->sc_ndisks - 1;
1725 		bp->bio_cmd = BIO_WRITE;
1726 		bp->bio_cflags = 0;
1727 		bp->bio_children = bp->bio_inbed = 0;
1728 		cp = disk->d_consumer;
1729 		KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1730 		    ("Consumer %s not opened (r%dw%de%d).", cp->provider->name,
1731 		    cp->acr, cp->acw, cp->ace));
1732 		cp->index++;
1733 		g_io_request(bp, cp);
1734 		return;
1735 	    }
1736 	case BIO_WRITE:
1737 	    {
1738 		struct g_raid3_disk_sync *sync;
1739 		off_t boffset, moffset;
1740 		void *data;
1741 		int i;
1742 
1743 		if (bp->bio_error != 0) {
1744 			G_RAID3_LOGREQ(0, bp,
1745 			    "Synchronization request failed (error=%d).",
1746 			    bp->bio_error);
1747 			g_destroy_bio(bp);
1748 			sc->sc_bump_id |= G_RAID3_BUMP_GENID;
1749 			g_raid3_event_send(disk,
1750 			    G_RAID3_DISK_STATE_DISCONNECTED,
1751 			    G_RAID3_EVENT_DONTWAIT);
1752 			return;
1753 		}
1754 		G_RAID3_LOGREQ(3, bp, "Synchronization request finished.");
1755 		sync = &disk->d_sync;
1756 		if (sync->ds_offset == sc->sc_mediasize / (sc->sc_ndisks - 1) ||
1757 		    sync->ds_consumer == NULL ||
1758 		    (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
1759 			/* Don't send more synchronization requests. */
1760 			sync->ds_inflight--;
1761 			if (sync->ds_bios != NULL) {
1762 				i = (int)(uintptr_t)bp->bio_caller1;
1763 				sync->ds_bios[i] = NULL;
1764 			}
1765 			free(bp->bio_data, M_RAID3);
1766 			g_destroy_bio(bp);
1767 			if (sync->ds_inflight > 0)
1768 				return;
1769 			if (sync->ds_consumer == NULL ||
1770 			    (sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
1771 				return;
1772 			}
1773 			/*
1774 			 * Disk up-to-date, activate it.
1775 			 */
1776 			g_raid3_event_send(disk, G_RAID3_DISK_STATE_ACTIVE,
1777 			    G_RAID3_EVENT_DONTWAIT);
1778 			return;
1779 		}
1780 
1781 		/* Send next synchronization request. */
1782 		data = bp->bio_data;
1783 		g_reset_bio(bp);
1784 		bp->bio_cmd = BIO_READ;
1785 		bp->bio_offset = sync->ds_offset * (sc->sc_ndisks - 1);
1786 		bp->bio_length = MIN(maxphys, sc->sc_mediasize - bp->bio_offset);
1787 		sync->ds_offset += bp->bio_length / (sc->sc_ndisks - 1);
1788 		bp->bio_done = g_raid3_sync_done;
1789 		bp->bio_data = data;
1790 		bp->bio_from = sync->ds_consumer;
1791 		bp->bio_to = sc->sc_provider;
1792 		G_RAID3_LOGREQ(3, bp, "Sending synchronization request.");
1793 		sync->ds_consumer->index++;
1794 		/*
1795 		 * Delay the request if it is colliding with a regular request.
1796 		 */
1797 		if (g_raid3_regular_collision(sc, bp))
1798 			g_raid3_sync_delay(sc, bp);
1799 		else
1800 			g_io_request(bp, sync->ds_consumer);
1801 
1802 		/* Release delayed requests if possible. */
1803 		g_raid3_regular_release(sc);
1804 
1805 		/* Find the smallest offset. */
1806 		moffset = sc->sc_mediasize;
1807 		for (i = 0; i < g_raid3_syncreqs; i++) {
1808 			bp = sync->ds_bios[i];
1809 			boffset = bp->bio_offset;
1810 			if (bp->bio_cmd == BIO_WRITE)
1811 				boffset *= sc->sc_ndisks - 1;
1812 			if (boffset < moffset)
1813 				moffset = boffset;
1814 		}
1815 		if (sync->ds_offset_done + maxphys * 100 < moffset) {
1816 			/* Update offset_done on every 100 blocks. */
1817 			sync->ds_offset_done = moffset;
1818 			g_raid3_update_metadata(disk);
1819 		}
1820 		return;
1821 	    }
1822 	default:
1823 		KASSERT(1 == 0, ("Invalid command here: %u (device=%s)",
1824 		    bp->bio_cmd, sc->sc_name));
1825 		break;
1826 	}
1827 }
1828 
1829 static int
g_raid3_register_request(struct bio * pbp)1830 g_raid3_register_request(struct bio *pbp)
1831 {
1832 	struct g_raid3_softc *sc;
1833 	struct g_raid3_disk *disk;
1834 	struct g_consumer *cp;
1835 	struct bio *cbp, *tmpbp;
1836 	off_t offset, length;
1837 	u_int n, ndisks;
1838 	int round_robin, verify;
1839 
1840 	ndisks = 0;
1841 	sc = pbp->bio_to->private;
1842 	if ((pbp->bio_cflags & G_RAID3_BIO_CFLAG_REGSYNC) != 0 &&
1843 	    sc->sc_syncdisk == NULL) {
1844 		g_io_deliver(pbp, EIO);
1845 		return (0);
1846 	}
1847 	g_raid3_init_bio(pbp);
1848 	length = pbp->bio_length / (sc->sc_ndisks - 1);
1849 	offset = pbp->bio_offset / (sc->sc_ndisks - 1);
1850 	round_robin = verify = 0;
1851 	switch (pbp->bio_cmd) {
1852 	case BIO_READ:
1853 		if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 &&
1854 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1855 			pbp->bio_pflags |= G_RAID3_BIO_PFLAG_VERIFY;
1856 			verify = 1;
1857 			ndisks = sc->sc_ndisks;
1858 		} else {
1859 			verify = 0;
1860 			ndisks = sc->sc_ndisks - 1;
1861 		}
1862 		if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0 &&
1863 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
1864 			round_robin = 1;
1865 		} else {
1866 			round_robin = 0;
1867 		}
1868 		KASSERT(!round_robin || !verify,
1869 		    ("ROUND-ROBIN and VERIFY are mutually exclusive."));
1870 		pbp->bio_driver2 = &sc->sc_disks[sc->sc_ndisks - 1];
1871 		break;
1872 	case BIO_WRITE:
1873 	case BIO_DELETE:
1874 		/*
1875 		 * Delay the request if it is colliding with a synchronization
1876 		 * request.
1877 		 */
1878 		if (g_raid3_sync_collision(sc, pbp)) {
1879 			g_raid3_regular_delay(sc, pbp);
1880 			return (0);
1881 		}
1882 
1883 		if (sc->sc_idle)
1884 			g_raid3_unidle(sc);
1885 		else
1886 			sc->sc_last_write = time_uptime;
1887 
1888 		ndisks = sc->sc_ndisks;
1889 		break;
1890 	}
1891 	for (n = 0; n < ndisks; n++) {
1892 		disk = &sc->sc_disks[n];
1893 		cbp = g_raid3_clone_bio(sc, pbp);
1894 		if (cbp == NULL) {
1895 			while ((cbp = G_RAID3_HEAD_BIO(pbp)) != NULL)
1896 				g_raid3_destroy_bio(sc, cbp);
1897 			/*
1898 			 * To prevent deadlock, we must run back up
1899 			 * with the ENOMEM for failed requests of any
1900 			 * of our consumers.  Our own sync requests
1901 			 * can stick around, as they are finite.
1902 			 */
1903 			if ((pbp->bio_cflags &
1904 			    G_RAID3_BIO_CFLAG_REGULAR) != 0) {
1905 				g_io_deliver(pbp, ENOMEM);
1906 				return (0);
1907 			}
1908 			return (ENOMEM);
1909 		}
1910 		cbp->bio_offset = offset;
1911 		cbp->bio_length = length;
1912 		cbp->bio_done = g_raid3_done;
1913 		switch (pbp->bio_cmd) {
1914 		case BIO_READ:
1915 			if (disk->d_state != G_RAID3_DISK_STATE_ACTIVE) {
1916 				/*
1917 				 * Replace invalid component with the parity
1918 				 * component.
1919 				 */
1920 				disk = &sc->sc_disks[sc->sc_ndisks - 1];
1921 				cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1922 				pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1923 			} else if (round_robin &&
1924 			    disk->d_no == sc->sc_round_robin) {
1925 				/*
1926 				 * In round-robin mode skip one data component
1927 				 * and use parity component when reading.
1928 				 */
1929 				pbp->bio_driver2 = disk;
1930 				disk = &sc->sc_disks[sc->sc_ndisks - 1];
1931 				cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1932 				sc->sc_round_robin++;
1933 				round_robin = 0;
1934 			} else if (verify && disk->d_no == sc->sc_ndisks - 1) {
1935 				cbp->bio_cflags |= G_RAID3_BIO_CFLAG_PARITY;
1936 			}
1937 			break;
1938 		case BIO_WRITE:
1939 		case BIO_DELETE:
1940 			if (disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
1941 			    disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
1942 				if (n == ndisks - 1) {
1943 					/*
1944 					 * Active parity component, mark it as such.
1945 					 */
1946 					cbp->bio_cflags |=
1947 					    G_RAID3_BIO_CFLAG_PARITY;
1948 				}
1949 			} else {
1950 				pbp->bio_pflags |= G_RAID3_BIO_PFLAG_DEGRADED;
1951 				if (n == ndisks - 1) {
1952 					/*
1953 					 * Parity component is not connected,
1954 					 * so destroy its request.
1955 					 */
1956 					pbp->bio_pflags |=
1957 					    G_RAID3_BIO_PFLAG_NOPARITY;
1958 					g_raid3_destroy_bio(sc, cbp);
1959 					cbp = NULL;
1960 				} else {
1961 					cbp->bio_cflags |=
1962 					    G_RAID3_BIO_CFLAG_NODISK;
1963 					disk = NULL;
1964 				}
1965 			}
1966 			break;
1967 		}
1968 		if (cbp != NULL)
1969 			cbp->bio_caller2 = disk;
1970 	}
1971 	switch (pbp->bio_cmd) {
1972 	case BIO_READ:
1973 		if (round_robin) {
1974 			/*
1975 			 * If we are in round-robin mode and 'round_robin' is
1976 			 * still 1, it means, that we skipped parity component
1977 			 * for this read and must reset sc_round_robin field.
1978 			 */
1979 			sc->sc_round_robin = 0;
1980 		}
1981 		G_RAID3_FOREACH_SAFE_BIO(pbp, cbp, tmpbp) {
1982 			disk = cbp->bio_caller2;
1983 			cp = disk->d_consumer;
1984 			cbp->bio_to = cp->provider;
1985 			G_RAID3_LOGREQ(3, cbp, "Sending request.");
1986 			KASSERT(cp->acr >= 1 && cp->acw >= 1 && cp->ace >= 1,
1987 			    ("Consumer %s not opened (r%dw%de%d).",
1988 			    cp->provider->name, cp->acr, cp->acw, cp->ace));
1989 			cp->index++;
1990 			g_io_request(cbp, cp);
1991 		}
1992 		break;
1993 	case BIO_WRITE:
1994 	case BIO_DELETE:
1995 		/*
1996 		 * Put request onto inflight queue, so we can check if new
1997 		 * synchronization requests don't collide with it.
1998 		 */
1999 		bioq_insert_tail(&sc->sc_inflight, pbp);
2000 
2001 		/*
2002 		 * Bump syncid on first write.
2003 		 */
2004 		if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0) {
2005 			sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID;
2006 			g_raid3_bump_syncid(sc);
2007 		}
2008 		g_raid3_scatter(pbp);
2009 		break;
2010 	}
2011 	return (0);
2012 }
2013 
2014 static int
g_raid3_can_destroy(struct g_raid3_softc * sc)2015 g_raid3_can_destroy(struct g_raid3_softc *sc)
2016 {
2017 	struct g_geom *gp;
2018 	struct g_consumer *cp;
2019 
2020 	g_topology_assert();
2021 	gp = sc->sc_geom;
2022 	if (gp->softc == NULL)
2023 		return (1);
2024 	LIST_FOREACH(cp, &gp->consumer, consumer) {
2025 		if (g_raid3_is_busy(sc, cp))
2026 			return (0);
2027 	}
2028 	gp = sc->sc_sync.ds_geom;
2029 	LIST_FOREACH(cp, &gp->consumer, consumer) {
2030 		if (g_raid3_is_busy(sc, cp))
2031 			return (0);
2032 	}
2033 	G_RAID3_DEBUG(2, "No I/O requests for %s, it can be destroyed.",
2034 	    sc->sc_name);
2035 	return (1);
2036 }
2037 
2038 static int
g_raid3_try_destroy(struct g_raid3_softc * sc)2039 g_raid3_try_destroy(struct g_raid3_softc *sc)
2040 {
2041 
2042 	g_topology_assert_not();
2043 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2044 
2045 	if (sc->sc_rootmount != NULL) {
2046 		G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2047 		    sc->sc_rootmount);
2048 		root_mount_rel(sc->sc_rootmount);
2049 		sc->sc_rootmount = NULL;
2050 	}
2051 
2052 	g_topology_lock();
2053 	if (!g_raid3_can_destroy(sc)) {
2054 		g_topology_unlock();
2055 		return (0);
2056 	}
2057 	sc->sc_geom->softc = NULL;
2058 	sc->sc_sync.ds_geom->softc = NULL;
2059 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_WAIT) != 0) {
2060 		g_topology_unlock();
2061 		G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__,
2062 		    &sc->sc_worker);
2063 		/* Unlock sc_lock here, as it can be destroyed after wakeup. */
2064 		sx_xunlock(&sc->sc_lock);
2065 		wakeup(&sc->sc_worker);
2066 		sc->sc_worker = NULL;
2067 	} else {
2068 		g_topology_unlock();
2069 		g_raid3_destroy_device(sc);
2070 	}
2071 	return (1);
2072 }
2073 
2074 /*
2075  * Worker thread.
2076  */
2077 static void
g_raid3_worker(void * arg)2078 g_raid3_worker(void *arg)
2079 {
2080 	struct g_raid3_softc *sc;
2081 	struct g_raid3_event *ep;
2082 	struct bio *bp;
2083 	int timeout;
2084 
2085 	sc = arg;
2086 	thread_lock(curthread);
2087 	sched_prio(curthread, PRIBIO);
2088 	thread_unlock(curthread);
2089 
2090 	sx_xlock(&sc->sc_lock);
2091 	for (;;) {
2092 		G_RAID3_DEBUG(5, "%s: Let's see...", __func__);
2093 		/*
2094 		 * First take a look at events.
2095 		 * This is important to handle events before any I/O requests.
2096 		 */
2097 		ep = g_raid3_event_get(sc);
2098 		if (ep != NULL) {
2099 			g_raid3_event_remove(sc, ep);
2100 			if ((ep->e_flags & G_RAID3_EVENT_DEVICE) != 0) {
2101 				/* Update only device status. */
2102 				G_RAID3_DEBUG(3,
2103 				    "Running event for device %s.",
2104 				    sc->sc_name);
2105 				ep->e_error = 0;
2106 				g_raid3_update_device(sc, 1);
2107 			} else {
2108 				/* Update disk status. */
2109 				G_RAID3_DEBUG(3, "Running event for disk %s.",
2110 				     g_raid3_get_diskname(ep->e_disk));
2111 				ep->e_error = g_raid3_update_disk(ep->e_disk,
2112 				    ep->e_state);
2113 				if (ep->e_error == 0)
2114 					g_raid3_update_device(sc, 0);
2115 			}
2116 			if ((ep->e_flags & G_RAID3_EVENT_DONTWAIT) != 0) {
2117 				KASSERT(ep->e_error == 0,
2118 				    ("Error cannot be handled."));
2119 				g_raid3_event_free(ep);
2120 			} else {
2121 				ep->e_flags |= G_RAID3_EVENT_DONE;
2122 				G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__,
2123 				    ep);
2124 				mtx_lock(&sc->sc_events_mtx);
2125 				wakeup(ep);
2126 				mtx_unlock(&sc->sc_events_mtx);
2127 			}
2128 			if ((sc->sc_flags &
2129 			    G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
2130 				if (g_raid3_try_destroy(sc)) {
2131 					curthread->td_pflags &= ~TDP_GEOM;
2132 					G_RAID3_DEBUG(1, "Thread exiting.");
2133 					kproc_exit(0);
2134 				}
2135 			}
2136 			G_RAID3_DEBUG(5, "%s: I'm here 1.", __func__);
2137 			continue;
2138 		}
2139 		/*
2140 		 * Check if we can mark array as CLEAN and if we can't take
2141 		 * how much seconds should we wait.
2142 		 */
2143 		timeout = g_raid3_idle(sc, -1);
2144 		/*
2145 		 * Now I/O requests.
2146 		 */
2147 		/* Get first request from the queue. */
2148 		mtx_lock(&sc->sc_queue_mtx);
2149 		bp = bioq_first(&sc->sc_queue);
2150 		if (bp == NULL) {
2151 			if ((sc->sc_flags &
2152 			    G_RAID3_DEVICE_FLAG_DESTROY) != 0) {
2153 				mtx_unlock(&sc->sc_queue_mtx);
2154 				if (g_raid3_try_destroy(sc)) {
2155 					curthread->td_pflags &= ~TDP_GEOM;
2156 					G_RAID3_DEBUG(1, "Thread exiting.");
2157 					kproc_exit(0);
2158 				}
2159 				mtx_lock(&sc->sc_queue_mtx);
2160 			}
2161 			sx_xunlock(&sc->sc_lock);
2162 			/*
2163 			 * XXX: We can miss an event here, because an event
2164 			 *      can be added without sx-device-lock and without
2165 			 *      mtx-queue-lock. Maybe I should just stop using
2166 			 *      dedicated mutex for events synchronization and
2167 			 *      stick with the queue lock?
2168 			 *      The event will hang here until next I/O request
2169 			 *      or next event is received.
2170 			 */
2171 			MSLEEP(sc, &sc->sc_queue_mtx, PRIBIO | PDROP, "r3:w1",
2172 			    timeout * hz);
2173 			sx_xlock(&sc->sc_lock);
2174 			G_RAID3_DEBUG(5, "%s: I'm here 4.", __func__);
2175 			continue;
2176 		}
2177 process:
2178 		bioq_remove(&sc->sc_queue, bp);
2179 		mtx_unlock(&sc->sc_queue_mtx);
2180 
2181 		if (bp->bio_from->geom == sc->sc_sync.ds_geom &&
2182 		    (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0) {
2183 			g_raid3_sync_request(bp);	/* READ */
2184 		} else if (bp->bio_to != sc->sc_provider) {
2185 			if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR) != 0)
2186 				g_raid3_regular_request(bp);
2187 			else if ((bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC) != 0)
2188 				g_raid3_sync_request(bp);	/* WRITE */
2189 			else {
2190 				KASSERT(0,
2191 				    ("Invalid request cflags=0x%hx to=%s.",
2192 				    bp->bio_cflags, bp->bio_to->name));
2193 			}
2194 		} else if (g_raid3_register_request(bp) != 0) {
2195 			mtx_lock(&sc->sc_queue_mtx);
2196 			bioq_insert_head(&sc->sc_queue, bp);
2197 			/*
2198 			 * We are short in memory, let see if there are finished
2199 			 * request we can free.
2200 			 */
2201 			TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
2202 				if (bp->bio_cflags & G_RAID3_BIO_CFLAG_REGULAR)
2203 					goto process;
2204 			}
2205 			/*
2206 			 * No finished regular request, so at least keep
2207 			 * synchronization running.
2208 			 */
2209 			TAILQ_FOREACH(bp, &sc->sc_queue.queue, bio_queue) {
2210 				if (bp->bio_cflags & G_RAID3_BIO_CFLAG_SYNC)
2211 					goto process;
2212 			}
2213 			sx_xunlock(&sc->sc_lock);
2214 			MSLEEP(&sc->sc_queue, &sc->sc_queue_mtx, PRIBIO | PDROP,
2215 			    "r3:lowmem", hz / 10);
2216 			sx_xlock(&sc->sc_lock);
2217 		}
2218 		G_RAID3_DEBUG(5, "%s: I'm here 9.", __func__);
2219 	}
2220 }
2221 
2222 static void
g_raid3_update_idle(struct g_raid3_softc * sc,struct g_raid3_disk * disk)2223 g_raid3_update_idle(struct g_raid3_softc *sc, struct g_raid3_disk *disk)
2224 {
2225 
2226 	sx_assert(&sc->sc_lock, SX_LOCKED);
2227 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) != 0)
2228 		return;
2229 	if (!sc->sc_idle && (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) == 0) {
2230 		G_RAID3_DEBUG(1, "Disk %s (device %s) marked as dirty.",
2231 		    g_raid3_get_diskname(disk), sc->sc_name);
2232 		disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
2233 	} else if (sc->sc_idle &&
2234 	    (disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0) {
2235 		G_RAID3_DEBUG(1, "Disk %s (device %s) marked as clean.",
2236 		    g_raid3_get_diskname(disk), sc->sc_name);
2237 		disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2238 	}
2239 }
2240 
2241 static void
g_raid3_sync_start(struct g_raid3_softc * sc)2242 g_raid3_sync_start(struct g_raid3_softc *sc)
2243 {
2244 	struct g_raid3_disk *disk;
2245 	struct g_consumer *cp;
2246 	struct bio *bp;
2247 	int error __diagused;
2248 	u_int n;
2249 
2250 	g_topology_assert_not();
2251 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2252 
2253 	KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED,
2254 	    ("Device not in DEGRADED state (%s, %u).", sc->sc_name,
2255 	    sc->sc_state));
2256 	KASSERT(sc->sc_syncdisk == NULL, ("Syncdisk is not NULL (%s, %u).",
2257 	    sc->sc_name, sc->sc_state));
2258 	disk = NULL;
2259 	for (n = 0; n < sc->sc_ndisks; n++) {
2260 		if (sc->sc_disks[n].d_state != G_RAID3_DISK_STATE_SYNCHRONIZING)
2261 			continue;
2262 		disk = &sc->sc_disks[n];
2263 		break;
2264 	}
2265 	if (disk == NULL)
2266 		return;
2267 
2268 	sx_xunlock(&sc->sc_lock);
2269 	g_topology_lock();
2270 	cp = g_new_consumer(sc->sc_sync.ds_geom);
2271 	error = g_attach(cp, sc->sc_provider);
2272 	KASSERT(error == 0,
2273 	    ("Cannot attach to %s (error=%d).", sc->sc_name, error));
2274 	error = g_access(cp, 1, 0, 0);
2275 	KASSERT(error == 0, ("Cannot open %s (error=%d).", sc->sc_name, error));
2276 	g_topology_unlock();
2277 	sx_xlock(&sc->sc_lock);
2278 
2279 	G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s.", sc->sc_name,
2280 	    g_raid3_get_diskname(disk));
2281 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOFAILSYNC) == 0)
2282 		disk->d_flags |= G_RAID3_DISK_FLAG_DIRTY;
2283 	KASSERT(disk->d_sync.ds_consumer == NULL,
2284 	    ("Sync consumer already exists (device=%s, disk=%s).",
2285 	    sc->sc_name, g_raid3_get_diskname(disk)));
2286 
2287 	disk->d_sync.ds_consumer = cp;
2288 	disk->d_sync.ds_consumer->private = disk;
2289 	disk->d_sync.ds_consumer->index = 0;
2290 	sc->sc_syncdisk = disk;
2291 
2292 	/*
2293 	 * Allocate memory for synchronization bios and initialize them.
2294 	 */
2295 	disk->d_sync.ds_bios = malloc(sizeof(struct bio *) * g_raid3_syncreqs,
2296 	    M_RAID3, M_WAITOK);
2297 	for (n = 0; n < g_raid3_syncreqs; n++) {
2298 		bp = g_alloc_bio();
2299 		disk->d_sync.ds_bios[n] = bp;
2300 		bp->bio_parent = NULL;
2301 		bp->bio_cmd = BIO_READ;
2302 		bp->bio_data = malloc(maxphys, M_RAID3, M_WAITOK);
2303 		bp->bio_cflags = 0;
2304 		bp->bio_offset = disk->d_sync.ds_offset * (sc->sc_ndisks - 1);
2305 		bp->bio_length = MIN(maxphys, sc->sc_mediasize - bp->bio_offset);
2306 		disk->d_sync.ds_offset += bp->bio_length / (sc->sc_ndisks - 1);
2307 		bp->bio_done = g_raid3_sync_done;
2308 		bp->bio_from = disk->d_sync.ds_consumer;
2309 		bp->bio_to = sc->sc_provider;
2310 		bp->bio_caller1 = (void *)(uintptr_t)n;
2311 	}
2312 
2313 	/* Set the number of in-flight synchronization requests. */
2314 	disk->d_sync.ds_inflight = g_raid3_syncreqs;
2315 
2316 	/*
2317 	 * Fire off first synchronization requests.
2318 	 */
2319 	for (n = 0; n < g_raid3_syncreqs; n++) {
2320 		bp = disk->d_sync.ds_bios[n];
2321 		G_RAID3_LOGREQ(3, bp, "Sending synchronization request.");
2322 		disk->d_sync.ds_consumer->index++;
2323 		/*
2324 		 * Delay the request if it is colliding with a regular request.
2325 		 */
2326 		if (g_raid3_regular_collision(sc, bp))
2327 			g_raid3_sync_delay(sc, bp);
2328 		else
2329 			g_io_request(bp, disk->d_sync.ds_consumer);
2330 	}
2331 }
2332 
2333 /*
2334  * Stop synchronization process.
2335  * type: 0 - synchronization finished
2336  *       1 - synchronization stopped
2337  */
2338 static void
g_raid3_sync_stop(struct g_raid3_softc * sc,int type)2339 g_raid3_sync_stop(struct g_raid3_softc *sc, int type)
2340 {
2341 	struct g_raid3_disk *disk;
2342 	struct g_consumer *cp;
2343 
2344 	g_topology_assert_not();
2345 	sx_assert(&sc->sc_lock, SX_LOCKED);
2346 
2347 	KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED,
2348 	    ("Device not in DEGRADED state (%s, %u).", sc->sc_name,
2349 	    sc->sc_state));
2350 	disk = sc->sc_syncdisk;
2351 	sc->sc_syncdisk = NULL;
2352 	KASSERT(disk != NULL, ("No disk was synchronized (%s).", sc->sc_name));
2353 	KASSERT(disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2354 	    ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2355 	    g_raid3_disk_state2str(disk->d_state)));
2356 	if (disk->d_sync.ds_consumer == NULL)
2357 		return;
2358 
2359 	if (type == 0) {
2360 		G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s finished.",
2361 		    sc->sc_name, g_raid3_get_diskname(disk));
2362 	} else /* if (type == 1) */ {
2363 		G_RAID3_DEBUG(0, "Device %s: rebuilding provider %s stopped.",
2364 		    sc->sc_name, g_raid3_get_diskname(disk));
2365 	}
2366 	free(disk->d_sync.ds_bios, M_RAID3);
2367 	disk->d_sync.ds_bios = NULL;
2368 	cp = disk->d_sync.ds_consumer;
2369 	disk->d_sync.ds_consumer = NULL;
2370 	disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2371 	sx_xunlock(&sc->sc_lock); /* Avoid recursion on sc_lock. */
2372 	g_topology_lock();
2373 	g_raid3_kill_consumer(sc, cp);
2374 	g_topology_unlock();
2375 	sx_xlock(&sc->sc_lock);
2376 }
2377 
2378 static void
g_raid3_launch_provider(struct g_raid3_softc * sc)2379 g_raid3_launch_provider(struct g_raid3_softc *sc)
2380 {
2381 	struct g_provider *pp;
2382 	struct g_raid3_disk *disk;
2383 	int n;
2384 
2385 	sx_assert(&sc->sc_lock, SX_LOCKED);
2386 
2387 	g_topology_lock();
2388 	pp = g_new_providerf(sc->sc_geom, "raid3/%s", sc->sc_name);
2389 	pp->mediasize = sc->sc_mediasize;
2390 	pp->sectorsize = sc->sc_sectorsize;
2391 	pp->stripesize = 0;
2392 	pp->stripeoffset = 0;
2393 	for (n = 0; n < sc->sc_ndisks; n++) {
2394 		disk = &sc->sc_disks[n];
2395 		if (disk->d_consumer && disk->d_consumer->provider &&
2396 		    disk->d_consumer->provider->stripesize > pp->stripesize) {
2397 			pp->stripesize = disk->d_consumer->provider->stripesize;
2398 			pp->stripeoffset = disk->d_consumer->provider->stripeoffset;
2399 		}
2400 	}
2401 	pp->stripesize *= sc->sc_ndisks - 1;
2402 	pp->stripeoffset *= sc->sc_ndisks - 1;
2403 	pp->private = sc;
2404 	sc->sc_refcnt++;
2405 	sc->sc_provider = pp;
2406 	g_error_provider(pp, 0);
2407 	g_topology_unlock();
2408 	G_RAID3_DEBUG(0, "Device %s launched (%u/%u).", pp->name,
2409 	    g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE), sc->sc_ndisks);
2410 
2411 	if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED)
2412 		g_raid3_sync_start(sc);
2413 }
2414 
2415 static void
g_raid3_destroy_provider(struct g_raid3_softc * sc)2416 g_raid3_destroy_provider(struct g_raid3_softc *sc)
2417 {
2418 	struct bio *bp;
2419 
2420 	g_topology_assert_not();
2421 	KASSERT(sc->sc_provider != NULL, ("NULL provider (device=%s).",
2422 	    sc->sc_name));
2423 
2424 	g_topology_lock();
2425 	g_error_provider(sc->sc_provider, ENXIO);
2426 	mtx_lock(&sc->sc_queue_mtx);
2427 	while ((bp = bioq_first(&sc->sc_queue)) != NULL) {
2428 		bioq_remove(&sc->sc_queue, bp);
2429 		g_io_deliver(bp, ENXIO);
2430 	}
2431 	mtx_unlock(&sc->sc_queue_mtx);
2432 	G_RAID3_DEBUG(0, "Device %s: provider %s destroyed.", sc->sc_name,
2433 	    sc->sc_provider->name);
2434 	g_wither_provider(sc->sc_provider, ENXIO);
2435 	g_topology_unlock();
2436 	sc->sc_provider = NULL;
2437 	if (sc->sc_syncdisk != NULL)
2438 		g_raid3_sync_stop(sc, 1);
2439 }
2440 
2441 static void
g_raid3_go(void * arg)2442 g_raid3_go(void *arg)
2443 {
2444 	struct g_raid3_softc *sc;
2445 	struct g_raid3_event *ep;
2446 
2447 	sc = arg;
2448 	G_RAID3_DEBUG(0, "Force device %s start due to timeout.", sc->sc_name);
2449 	ep = sc->sc_timeout_event;
2450 	sc->sc_timeout_event = NULL;
2451 	g_raid3_event_dispatch(ep, sc, 0,
2452 	    G_RAID3_EVENT_DONTWAIT | G_RAID3_EVENT_DEVICE);
2453 }
2454 
2455 static void
g_raid3_timeout_drain(struct g_raid3_softc * sc)2456 g_raid3_timeout_drain(struct g_raid3_softc *sc)
2457 {
2458 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2459 
2460 	callout_drain(&sc->sc_callout);
2461 	g_raid3_event_free(sc->sc_timeout_event);
2462 	sc->sc_timeout_event = NULL;
2463 }
2464 
2465 static u_int
g_raid3_determine_state(struct g_raid3_disk * disk)2466 g_raid3_determine_state(struct g_raid3_disk *disk)
2467 {
2468 	struct g_raid3_softc *sc;
2469 	u_int state;
2470 
2471 	sc = disk->d_softc;
2472 	if (sc->sc_syncid == disk->d_sync.ds_syncid) {
2473 		if ((disk->d_flags &
2474 		    G_RAID3_DISK_FLAG_SYNCHRONIZING) == 0) {
2475 			/* Disk does not need synchronization. */
2476 			state = G_RAID3_DISK_STATE_ACTIVE;
2477 		} else {
2478 			if ((sc->sc_flags &
2479 			     G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2480 			    (disk->d_flags &
2481 			     G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) {
2482 				/*
2483 				 * We can start synchronization from
2484 				 * the stored offset.
2485 				 */
2486 				state = G_RAID3_DISK_STATE_SYNCHRONIZING;
2487 			} else {
2488 				state = G_RAID3_DISK_STATE_STALE;
2489 			}
2490 		}
2491 	} else if (disk->d_sync.ds_syncid < sc->sc_syncid) {
2492 		/*
2493 		 * Reset all synchronization data for this disk,
2494 		 * because if it even was synchronized, it was
2495 		 * synchronized to disks with different syncid.
2496 		 */
2497 		disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING;
2498 		disk->d_sync.ds_offset = 0;
2499 		disk->d_sync.ds_offset_done = 0;
2500 		disk->d_sync.ds_syncid = sc->sc_syncid;
2501 		if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) == 0 ||
2502 		    (disk->d_flags & G_RAID3_DISK_FLAG_FORCE_SYNC) != 0) {
2503 			state = G_RAID3_DISK_STATE_SYNCHRONIZING;
2504 		} else {
2505 			state = G_RAID3_DISK_STATE_STALE;
2506 		}
2507 	} else /* if (sc->sc_syncid < disk->d_sync.ds_syncid) */ {
2508 		/*
2509 		 * Not good, NOT GOOD!
2510 		 * It means that device was started on stale disks
2511 		 * and more fresh disk just arrive.
2512 		 * If there were writes, device is broken, sorry.
2513 		 * I think the best choice here is don't touch
2514 		 * this disk and inform the user loudly.
2515 		 */
2516 		G_RAID3_DEBUG(0, "Device %s was started before the freshest "
2517 		    "disk (%s) arrives!! It will not be connected to the "
2518 		    "running device.", sc->sc_name,
2519 		    g_raid3_get_diskname(disk));
2520 		g_raid3_destroy_disk(disk);
2521 		state = G_RAID3_DISK_STATE_NONE;
2522 		/* Return immediately, because disk was destroyed. */
2523 		return (state);
2524 	}
2525 	G_RAID3_DEBUG(3, "State for %s disk: %s.",
2526 	    g_raid3_get_diskname(disk), g_raid3_disk_state2str(state));
2527 	return (state);
2528 }
2529 
2530 /*
2531  * Update device state.
2532  */
2533 static void
g_raid3_update_device(struct g_raid3_softc * sc,boolean_t force)2534 g_raid3_update_device(struct g_raid3_softc *sc, boolean_t force)
2535 {
2536 	struct g_raid3_disk *disk;
2537 	u_int state;
2538 
2539 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2540 
2541 	switch (sc->sc_state) {
2542 	case G_RAID3_DEVICE_STATE_STARTING:
2543 	    {
2544 		u_int n, ndirty, ndisks, genid, syncid;
2545 
2546 		KASSERT(sc->sc_provider == NULL,
2547 		    ("Non-NULL provider in STARTING state (%s).", sc->sc_name));
2548 		/*
2549 		 * Are we ready? We are, if all disks are connected or
2550 		 * one disk is missing and 'force' is true.
2551 		 */
2552 		if (g_raid3_ndisks(sc, -1) + force == sc->sc_ndisks) {
2553 			if (!force)
2554 				g_raid3_timeout_drain(sc);
2555 		} else {
2556 			if (force) {
2557 				/*
2558 				 * Timeout expired, so destroy device.
2559 				 */
2560 				sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2561 				G_RAID3_DEBUG(1, "root_mount_rel[%u] %p",
2562 				    __LINE__, sc->sc_rootmount);
2563 				root_mount_rel(sc->sc_rootmount);
2564 				sc->sc_rootmount = NULL;
2565 			}
2566 			return;
2567 		}
2568 
2569 		/*
2570 		 * Find the biggest genid.
2571 		 */
2572 		genid = 0;
2573 		for (n = 0; n < sc->sc_ndisks; n++) {
2574 			disk = &sc->sc_disks[n];
2575 			if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2576 				continue;
2577 			if (disk->d_genid > genid)
2578 				genid = disk->d_genid;
2579 		}
2580 		sc->sc_genid = genid;
2581 		/*
2582 		 * Remove all disks without the biggest genid.
2583 		 */
2584 		for (n = 0; n < sc->sc_ndisks; n++) {
2585 			disk = &sc->sc_disks[n];
2586 			if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2587 				continue;
2588 			if (disk->d_genid < genid) {
2589 				G_RAID3_DEBUG(0,
2590 				    "Component %s (device %s) broken, skipping.",
2591 				    g_raid3_get_diskname(disk), sc->sc_name);
2592 				g_raid3_destroy_disk(disk);
2593 			}
2594 		}
2595 
2596 		/*
2597 		 * There must be at least 'sc->sc_ndisks - 1' components
2598 		 * with the same syncid and without SYNCHRONIZING flag.
2599 		 */
2600 
2601 		/*
2602 		 * Find the biggest syncid, number of valid components and
2603 		 * number of dirty components.
2604 		 */
2605 		ndirty = ndisks = syncid = 0;
2606 		for (n = 0; n < sc->sc_ndisks; n++) {
2607 			disk = &sc->sc_disks[n];
2608 			if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2609 				continue;
2610 			if ((disk->d_flags & G_RAID3_DISK_FLAG_DIRTY) != 0)
2611 				ndirty++;
2612 			if (disk->d_sync.ds_syncid > syncid) {
2613 				syncid = disk->d_sync.ds_syncid;
2614 				ndisks = 0;
2615 			} else if (disk->d_sync.ds_syncid < syncid) {
2616 				continue;
2617 			}
2618 			if ((disk->d_flags &
2619 			    G_RAID3_DISK_FLAG_SYNCHRONIZING) != 0) {
2620 				continue;
2621 			}
2622 			ndisks++;
2623 		}
2624 		/*
2625 		 * Do we have enough valid components?
2626 		 */
2627 		if (ndisks + 1 < sc->sc_ndisks) {
2628 			G_RAID3_DEBUG(0,
2629 			    "Device %s is broken, too few valid components.",
2630 			    sc->sc_name);
2631 			sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2632 			return;
2633 		}
2634 		/*
2635 		 * If there is one DIRTY component and all disks are present,
2636 		 * mark it for synchronization. If there is more than one DIRTY
2637 		 * component, mark parity component for synchronization.
2638 		 */
2639 		if (ndisks == sc->sc_ndisks && ndirty == 1) {
2640 			for (n = 0; n < sc->sc_ndisks; n++) {
2641 				disk = &sc->sc_disks[n];
2642 				if ((disk->d_flags &
2643 				    G_RAID3_DISK_FLAG_DIRTY) == 0) {
2644 					continue;
2645 				}
2646 				disk->d_flags |=
2647 				    G_RAID3_DISK_FLAG_SYNCHRONIZING;
2648 			}
2649 		} else if (ndisks == sc->sc_ndisks && ndirty > 1) {
2650 			disk = &sc->sc_disks[sc->sc_ndisks - 1];
2651 			disk->d_flags |= G_RAID3_DISK_FLAG_SYNCHRONIZING;
2652 		}
2653 
2654 		sc->sc_syncid = syncid;
2655 		if (force) {
2656 			/* Remember to bump syncid on first write. */
2657 			sc->sc_bump_id |= G_RAID3_BUMP_SYNCID;
2658 		}
2659 		if (ndisks == sc->sc_ndisks)
2660 			state = G_RAID3_DEVICE_STATE_COMPLETE;
2661 		else /* if (ndisks == sc->sc_ndisks - 1) */
2662 			state = G_RAID3_DEVICE_STATE_DEGRADED;
2663 		G_RAID3_DEBUG(1, "Device %s state changed from %s to %s.",
2664 		    sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2665 		    g_raid3_device_state2str(state));
2666 		sc->sc_state = state;
2667 		for (n = 0; n < sc->sc_ndisks; n++) {
2668 			disk = &sc->sc_disks[n];
2669 			if (disk->d_state == G_RAID3_DISK_STATE_NODISK)
2670 				continue;
2671 			state = g_raid3_determine_state(disk);
2672 			g_raid3_event_send(disk, state, G_RAID3_EVENT_DONTWAIT);
2673 			if (state == G_RAID3_DISK_STATE_STALE)
2674 				sc->sc_bump_id |= G_RAID3_BUMP_SYNCID;
2675 		}
2676 		break;
2677 	    }
2678 	case G_RAID3_DEVICE_STATE_DEGRADED:
2679 		/*
2680 		 * Genid need to be bumped immediately, so do it here.
2681 		 */
2682 		if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) {
2683 			sc->sc_bump_id &= ~G_RAID3_BUMP_GENID;
2684 			g_raid3_bump_genid(sc);
2685 		}
2686 
2687 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0)
2688 			return;
2689 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) <
2690 		    sc->sc_ndisks - 1) {
2691 			if (sc->sc_provider != NULL)
2692 				g_raid3_destroy_provider(sc);
2693 			sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
2694 			return;
2695 		}
2696 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) ==
2697 		    sc->sc_ndisks) {
2698 			state = G_RAID3_DEVICE_STATE_COMPLETE;
2699 			G_RAID3_DEBUG(1,
2700 			    "Device %s state changed from %s to %s.",
2701 			    sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2702 			    g_raid3_device_state2str(state));
2703 			sc->sc_state = state;
2704 		}
2705 		if (sc->sc_provider == NULL)
2706 			g_raid3_launch_provider(sc);
2707 		if (sc->sc_rootmount != NULL) {
2708 			G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2709 			    sc->sc_rootmount);
2710 			root_mount_rel(sc->sc_rootmount);
2711 			sc->sc_rootmount = NULL;
2712 		}
2713 		break;
2714 	case G_RAID3_DEVICE_STATE_COMPLETE:
2715 		/*
2716 		 * Genid need to be bumped immediately, so do it here.
2717 		 */
2718 		if ((sc->sc_bump_id & G_RAID3_BUMP_GENID) != 0) {
2719 			sc->sc_bump_id &= ~G_RAID3_BUMP_GENID;
2720 			g_raid3_bump_genid(sc);
2721 		}
2722 
2723 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NEW) > 0)
2724 			return;
2725 		KASSERT(g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) >=
2726 		    sc->sc_ndisks - 1,
2727 		    ("Too few ACTIVE components in COMPLETE state (device %s).",
2728 		    sc->sc_name));
2729 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) ==
2730 		    sc->sc_ndisks - 1) {
2731 			state = G_RAID3_DEVICE_STATE_DEGRADED;
2732 			G_RAID3_DEBUG(1,
2733 			    "Device %s state changed from %s to %s.",
2734 			    sc->sc_name, g_raid3_device_state2str(sc->sc_state),
2735 			    g_raid3_device_state2str(state));
2736 			sc->sc_state = state;
2737 		}
2738 		if (sc->sc_provider == NULL)
2739 			g_raid3_launch_provider(sc);
2740 		if (sc->sc_rootmount != NULL) {
2741 			G_RAID3_DEBUG(1, "root_mount_rel[%u] %p", __LINE__,
2742 			    sc->sc_rootmount);
2743 			root_mount_rel(sc->sc_rootmount);
2744 			sc->sc_rootmount = NULL;
2745 		}
2746 		break;
2747 	default:
2748 		KASSERT(1 == 0, ("Wrong device state (%s, %s).", sc->sc_name,
2749 		    g_raid3_device_state2str(sc->sc_state)));
2750 		break;
2751 	}
2752 }
2753 
2754 /*
2755  * Update disk state and device state if needed.
2756  */
2757 #define	DISK_STATE_CHANGED()	G_RAID3_DEBUG(1,			\
2758 	"Disk %s state changed from %s to %s (device %s).",		\
2759 	g_raid3_get_diskname(disk),					\
2760 	g_raid3_disk_state2str(disk->d_state),				\
2761 	g_raid3_disk_state2str(state), sc->sc_name)
2762 static int
g_raid3_update_disk(struct g_raid3_disk * disk,u_int state)2763 g_raid3_update_disk(struct g_raid3_disk *disk, u_int state)
2764 {
2765 	struct g_raid3_softc *sc;
2766 
2767 	sc = disk->d_softc;
2768 	sx_assert(&sc->sc_lock, SX_XLOCKED);
2769 
2770 again:
2771 	G_RAID3_DEBUG(3, "Changing disk %s state from %s to %s.",
2772 	    g_raid3_get_diskname(disk), g_raid3_disk_state2str(disk->d_state),
2773 	    g_raid3_disk_state2str(state));
2774 	switch (state) {
2775 	case G_RAID3_DISK_STATE_NEW:
2776 		/*
2777 		 * Possible scenarios:
2778 		 * 1. New disk arrive.
2779 		 */
2780 		/* Previous state should be NONE. */
2781 		KASSERT(disk->d_state == G_RAID3_DISK_STATE_NONE,
2782 		    ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2783 		    g_raid3_disk_state2str(disk->d_state)));
2784 		DISK_STATE_CHANGED();
2785 
2786 		disk->d_state = state;
2787 		G_RAID3_DEBUG(1, "Device %s: provider %s detected.",
2788 		    sc->sc_name, g_raid3_get_diskname(disk));
2789 		if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING)
2790 			break;
2791 		KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2792 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2793 		    ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2794 		    g_raid3_device_state2str(sc->sc_state),
2795 		    g_raid3_get_diskname(disk),
2796 		    g_raid3_disk_state2str(disk->d_state)));
2797 		state = g_raid3_determine_state(disk);
2798 		if (state != G_RAID3_DISK_STATE_NONE)
2799 			goto again;
2800 		break;
2801 	case G_RAID3_DISK_STATE_ACTIVE:
2802 		/*
2803 		 * Possible scenarios:
2804 		 * 1. New disk does not need synchronization.
2805 		 * 2. Synchronization process finished successfully.
2806 		 */
2807 		KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2808 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2809 		    ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2810 		    g_raid3_device_state2str(sc->sc_state),
2811 		    g_raid3_get_diskname(disk),
2812 		    g_raid3_disk_state2str(disk->d_state)));
2813 		/* Previous state should be NEW or SYNCHRONIZING. */
2814 		KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW ||
2815 		    disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2816 		    ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2817 		    g_raid3_disk_state2str(disk->d_state)));
2818 		DISK_STATE_CHANGED();
2819 
2820 		if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
2821 			disk->d_flags &= ~G_RAID3_DISK_FLAG_SYNCHRONIZING;
2822 			disk->d_flags &= ~G_RAID3_DISK_FLAG_FORCE_SYNC;
2823 			g_raid3_sync_stop(sc, 0);
2824 		}
2825 		disk->d_state = state;
2826 		disk->d_sync.ds_offset = 0;
2827 		disk->d_sync.ds_offset_done = 0;
2828 		g_raid3_update_idle(sc, disk);
2829 		g_raid3_update_metadata(disk);
2830 		G_RAID3_DEBUG(1, "Device %s: provider %s activated.",
2831 		    sc->sc_name, g_raid3_get_diskname(disk));
2832 		break;
2833 	case G_RAID3_DISK_STATE_STALE:
2834 		/*
2835 		 * Possible scenarios:
2836 		 * 1. Stale disk was connected.
2837 		 */
2838 		/* Previous state should be NEW. */
2839 		KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2840 		    ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2841 		    g_raid3_disk_state2str(disk->d_state)));
2842 		KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2843 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2844 		    ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2845 		    g_raid3_device_state2str(sc->sc_state),
2846 		    g_raid3_get_diskname(disk),
2847 		    g_raid3_disk_state2str(disk->d_state)));
2848 		/*
2849 		 * STALE state is only possible if device is marked
2850 		 * NOAUTOSYNC.
2851 		 */
2852 		KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_NOAUTOSYNC) != 0,
2853 		    ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2854 		    g_raid3_device_state2str(sc->sc_state),
2855 		    g_raid3_get_diskname(disk),
2856 		    g_raid3_disk_state2str(disk->d_state)));
2857 		DISK_STATE_CHANGED();
2858 
2859 		disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2860 		disk->d_state = state;
2861 		g_raid3_update_metadata(disk);
2862 		G_RAID3_DEBUG(0, "Device %s: provider %s is stale.",
2863 		    sc->sc_name, g_raid3_get_diskname(disk));
2864 		break;
2865 	case G_RAID3_DISK_STATE_SYNCHRONIZING:
2866 		/*
2867 		 * Possible scenarios:
2868 		 * 1. Disk which needs synchronization was connected.
2869 		 */
2870 		/* Previous state should be NEW. */
2871 		KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2872 		    ("Wrong disk state (%s, %s).", g_raid3_get_diskname(disk),
2873 		    g_raid3_disk_state2str(disk->d_state)));
2874 		KASSERT(sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2875 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE,
2876 		    ("Wrong device state (%s, %s, %s, %s).", sc->sc_name,
2877 		    g_raid3_device_state2str(sc->sc_state),
2878 		    g_raid3_get_diskname(disk),
2879 		    g_raid3_disk_state2str(disk->d_state)));
2880 		DISK_STATE_CHANGED();
2881 
2882 		if (disk->d_state == G_RAID3_DISK_STATE_NEW)
2883 			disk->d_flags &= ~G_RAID3_DISK_FLAG_DIRTY;
2884 		disk->d_state = state;
2885 		if (sc->sc_provider != NULL) {
2886 			g_raid3_sync_start(sc);
2887 			g_raid3_update_metadata(disk);
2888 		}
2889 		break;
2890 	case G_RAID3_DISK_STATE_DISCONNECTED:
2891 		/*
2892 		 * Possible scenarios:
2893 		 * 1. Device wasn't running yet, but disk disappear.
2894 		 * 2. Disk was active and disapppear.
2895 		 * 3. Disk disappear during synchronization process.
2896 		 */
2897 		if (sc->sc_state == G_RAID3_DEVICE_STATE_DEGRADED ||
2898 		    sc->sc_state == G_RAID3_DEVICE_STATE_COMPLETE) {
2899 			/*
2900 			 * Previous state should be ACTIVE, STALE or
2901 			 * SYNCHRONIZING.
2902 			 */
2903 			KASSERT(disk->d_state == G_RAID3_DISK_STATE_ACTIVE ||
2904 			    disk->d_state == G_RAID3_DISK_STATE_STALE ||
2905 			    disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING,
2906 			    ("Wrong disk state (%s, %s).",
2907 			    g_raid3_get_diskname(disk),
2908 			    g_raid3_disk_state2str(disk->d_state)));
2909 		} else if (sc->sc_state == G_RAID3_DEVICE_STATE_STARTING) {
2910 			/* Previous state should be NEW. */
2911 			KASSERT(disk->d_state == G_RAID3_DISK_STATE_NEW,
2912 			    ("Wrong disk state (%s, %s).",
2913 			    g_raid3_get_diskname(disk),
2914 			    g_raid3_disk_state2str(disk->d_state)));
2915 			/*
2916 			 * Reset bumping syncid if disk disappeared in STARTING
2917 			 * state.
2918 			 */
2919 			if ((sc->sc_bump_id & G_RAID3_BUMP_SYNCID) != 0)
2920 				sc->sc_bump_id &= ~G_RAID3_BUMP_SYNCID;
2921 #ifdef	INVARIANTS
2922 		} else {
2923 			KASSERT(1 == 0, ("Wrong device state (%s, %s, %s, %s).",
2924 			    sc->sc_name,
2925 			    g_raid3_device_state2str(sc->sc_state),
2926 			    g_raid3_get_diskname(disk),
2927 			    g_raid3_disk_state2str(disk->d_state)));
2928 #endif
2929 		}
2930 		DISK_STATE_CHANGED();
2931 		G_RAID3_DEBUG(0, "Device %s: provider %s disconnected.",
2932 		    sc->sc_name, g_raid3_get_diskname(disk));
2933 
2934 		g_raid3_destroy_disk(disk);
2935 		break;
2936 	default:
2937 		KASSERT(1 == 0, ("Unknown state (%u).", state));
2938 		break;
2939 	}
2940 	return (0);
2941 }
2942 #undef	DISK_STATE_CHANGED
2943 
2944 int
g_raid3_read_metadata(struct g_consumer * cp,struct g_raid3_metadata * md)2945 g_raid3_read_metadata(struct g_consumer *cp, struct g_raid3_metadata *md)
2946 {
2947 	struct g_provider *pp;
2948 	u_char *buf;
2949 	int error;
2950 
2951 	g_topology_assert();
2952 
2953 	error = g_access(cp, 1, 0, 0);
2954 	if (error != 0)
2955 		return (error);
2956 	pp = cp->provider;
2957 	g_topology_unlock();
2958 	/* Metadata are stored on last sector. */
2959 	buf = g_read_data(cp, pp->mediasize - pp->sectorsize, pp->sectorsize,
2960 	    &error);
2961 	g_topology_lock();
2962 	g_access(cp, -1, 0, 0);
2963 	if (buf == NULL) {
2964 		G_RAID3_DEBUG(1, "Cannot read metadata from %s (error=%d).",
2965 		    cp->provider->name, error);
2966 		return (error);
2967 	}
2968 
2969 	/* Decode metadata. */
2970 	error = raid3_metadata_decode(buf, md);
2971 	g_free(buf);
2972 	if (strcmp(md->md_magic, G_RAID3_MAGIC) != 0)
2973 		return (EINVAL);
2974 	if (md->md_version > G_RAID3_VERSION) {
2975 		G_RAID3_DEBUG(0,
2976 		    "Kernel module is too old to handle metadata from %s.",
2977 		    cp->provider->name);
2978 		return (EINVAL);
2979 	}
2980 	if (error != 0) {
2981 		G_RAID3_DEBUG(1, "MD5 metadata hash mismatch for provider %s.",
2982 		    cp->provider->name);
2983 		return (error);
2984 	}
2985 	if (md->md_sectorsize > maxphys) {
2986 		G_RAID3_DEBUG(0, "The blocksize is too big.");
2987 		return (EINVAL);
2988 	}
2989 
2990 	return (0);
2991 }
2992 
2993 static int
g_raid3_check_metadata(struct g_raid3_softc * sc,struct g_provider * pp,struct g_raid3_metadata * md)2994 g_raid3_check_metadata(struct g_raid3_softc *sc, struct g_provider *pp,
2995     struct g_raid3_metadata *md)
2996 {
2997 
2998 	if (md->md_no >= sc->sc_ndisks) {
2999 		G_RAID3_DEBUG(1, "Invalid disk %s number (no=%u), skipping.",
3000 		    pp->name, md->md_no);
3001 		return (EINVAL);
3002 	}
3003 	if (sc->sc_disks[md->md_no].d_state != G_RAID3_DISK_STATE_NODISK) {
3004 		G_RAID3_DEBUG(1, "Disk %s (no=%u) already exists, skipping.",
3005 		    pp->name, md->md_no);
3006 		return (EEXIST);
3007 	}
3008 	if (md->md_all != sc->sc_ndisks) {
3009 		G_RAID3_DEBUG(1,
3010 		    "Invalid '%s' field on disk %s (device %s), skipping.",
3011 		    "md_all", pp->name, sc->sc_name);
3012 		return (EINVAL);
3013 	}
3014 	if ((md->md_mediasize % md->md_sectorsize) != 0) {
3015 		G_RAID3_DEBUG(1, "Invalid metadata (mediasize %% sectorsize != "
3016 		    "0) on disk %s (device %s), skipping.", pp->name,
3017 		    sc->sc_name);
3018 		return (EINVAL);
3019 	}
3020 	if (md->md_mediasize != sc->sc_mediasize) {
3021 		G_RAID3_DEBUG(1,
3022 		    "Invalid '%s' field on disk %s (device %s), skipping.",
3023 		    "md_mediasize", pp->name, sc->sc_name);
3024 		return (EINVAL);
3025 	}
3026 	if ((md->md_mediasize % (sc->sc_ndisks - 1)) != 0) {
3027 		G_RAID3_DEBUG(1,
3028 		    "Invalid '%s' field on disk %s (device %s), skipping.",
3029 		    "md_mediasize", pp->name, sc->sc_name);
3030 		return (EINVAL);
3031 	}
3032 	if ((sc->sc_mediasize / (sc->sc_ndisks - 1)) > pp->mediasize) {
3033 		G_RAID3_DEBUG(1,
3034 		    "Invalid size of disk %s (device %s), skipping.", pp->name,
3035 		    sc->sc_name);
3036 		return (EINVAL);
3037 	}
3038 	if ((md->md_sectorsize / pp->sectorsize) < sc->sc_ndisks - 1) {
3039 		G_RAID3_DEBUG(1,
3040 		    "Invalid '%s' field on disk %s (device %s), skipping.",
3041 		    "md_sectorsize", pp->name, sc->sc_name);
3042 		return (EINVAL);
3043 	}
3044 	if (md->md_sectorsize != sc->sc_sectorsize) {
3045 		G_RAID3_DEBUG(1,
3046 		    "Invalid '%s' field on disk %s (device %s), skipping.",
3047 		    "md_sectorsize", pp->name, sc->sc_name);
3048 		return (EINVAL);
3049 	}
3050 	if ((sc->sc_sectorsize % pp->sectorsize) != 0) {
3051 		G_RAID3_DEBUG(1,
3052 		    "Invalid sector size of disk %s (device %s), skipping.",
3053 		    pp->name, sc->sc_name);
3054 		return (EINVAL);
3055 	}
3056 	if ((md->md_mflags & ~G_RAID3_DEVICE_FLAG_MASK) != 0) {
3057 		G_RAID3_DEBUG(1,
3058 		    "Invalid device flags on disk %s (device %s), skipping.",
3059 		    pp->name, sc->sc_name);
3060 		return (EINVAL);
3061 	}
3062 	if ((md->md_mflags & G_RAID3_DEVICE_FLAG_VERIFY) != 0 &&
3063 	    (md->md_mflags & G_RAID3_DEVICE_FLAG_ROUND_ROBIN) != 0) {
3064 		/*
3065 		 * VERIFY and ROUND-ROBIN options are mutally exclusive.
3066 		 */
3067 		G_RAID3_DEBUG(1, "Both VERIFY and ROUND-ROBIN flags exist on "
3068 		    "disk %s (device %s), skipping.", pp->name, sc->sc_name);
3069 		return (EINVAL);
3070 	}
3071 	if ((md->md_dflags & ~G_RAID3_DISK_FLAG_MASK) != 0) {
3072 		G_RAID3_DEBUG(1,
3073 		    "Invalid disk flags on disk %s (device %s), skipping.",
3074 		    pp->name, sc->sc_name);
3075 		return (EINVAL);
3076 	}
3077 	return (0);
3078 }
3079 
3080 int
g_raid3_add_disk(struct g_raid3_softc * sc,struct g_provider * pp,struct g_raid3_metadata * md)3081 g_raid3_add_disk(struct g_raid3_softc *sc, struct g_provider *pp,
3082     struct g_raid3_metadata *md)
3083 {
3084 	struct g_raid3_disk *disk;
3085 	int error;
3086 
3087 	g_topology_assert_not();
3088 	G_RAID3_DEBUG(2, "Adding disk %s.", pp->name);
3089 
3090 	error = g_raid3_check_metadata(sc, pp, md);
3091 	if (error != 0)
3092 		return (error);
3093 	if (sc->sc_state != G_RAID3_DEVICE_STATE_STARTING &&
3094 	    md->md_genid < sc->sc_genid) {
3095 		G_RAID3_DEBUG(0, "Component %s (device %s) broken, skipping.",
3096 		    pp->name, sc->sc_name);
3097 		return (EINVAL);
3098 	}
3099 	disk = g_raid3_init_disk(sc, pp, md, &error);
3100 	if (disk == NULL)
3101 		return (error);
3102 	error = g_raid3_event_send(disk, G_RAID3_DISK_STATE_NEW,
3103 	    G_RAID3_EVENT_WAIT);
3104 	if (error != 0)
3105 		return (error);
3106 	if (md->md_version < G_RAID3_VERSION) {
3107 		G_RAID3_DEBUG(0, "Upgrading metadata on %s (v%d->v%d).",
3108 		    pp->name, md->md_version, G_RAID3_VERSION);
3109 		g_raid3_update_metadata(disk);
3110 	}
3111 	return (0);
3112 }
3113 
3114 static void
g_raid3_destroy_delayed(void * arg,int flag)3115 g_raid3_destroy_delayed(void *arg, int flag)
3116 {
3117 	struct g_raid3_softc *sc;
3118 	int error;
3119 
3120 	if (flag == EV_CANCEL) {
3121 		G_RAID3_DEBUG(1, "Destroying canceled.");
3122 		return;
3123 	}
3124 	sc = arg;
3125 	g_topology_unlock();
3126 	sx_xlock(&sc->sc_lock);
3127 	KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) == 0,
3128 	    ("DESTROY flag set on %s.", sc->sc_name));
3129 	KASSERT((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0,
3130 	    ("DESTROYING flag not set on %s.", sc->sc_name));
3131 	G_RAID3_DEBUG(0, "Destroying %s (delayed).", sc->sc_name);
3132 	error = g_raid3_destroy(sc, G_RAID3_DESTROY_SOFT);
3133 	if (error != 0) {
3134 		G_RAID3_DEBUG(0, "Cannot destroy %s.", sc->sc_name);
3135 		sx_xunlock(&sc->sc_lock);
3136 	}
3137 	g_topology_lock();
3138 }
3139 
3140 static int
g_raid3_access(struct g_provider * pp,int acr,int acw,int ace)3141 g_raid3_access(struct g_provider *pp, int acr, int acw, int ace)
3142 {
3143 	struct g_raid3_softc *sc;
3144 	int dcr, dcw, dce, error = 0;
3145 
3146 	g_topology_assert();
3147 	G_RAID3_DEBUG(2, "Access request for %s: r%dw%de%d.", pp->name, acr,
3148 	    acw, ace);
3149 
3150 	sc = pp->private;
3151 	KASSERT(sc != NULL, ("NULL softc (provider=%s).", pp->name));
3152 
3153 	dcr = pp->acr + acr;
3154 	dcw = pp->acw + acw;
3155 	dce = pp->ace + ace;
3156 
3157 	g_topology_unlock();
3158 	sx_xlock(&sc->sc_lock);
3159 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROY) != 0 ||
3160 	    g_raid3_ndisks(sc, G_RAID3_DISK_STATE_ACTIVE) < sc->sc_ndisks - 1) {
3161 		if (acr > 0 || acw > 0 || ace > 0)
3162 			error = ENXIO;
3163 		goto end;
3164 	}
3165 	if (dcw == 0)
3166 		g_raid3_idle(sc, dcw);
3167 	if ((sc->sc_flags & G_RAID3_DEVICE_FLAG_DESTROYING) != 0) {
3168 		if (acr > 0 || acw > 0 || ace > 0) {
3169 			error = ENXIO;
3170 			goto end;
3171 		}
3172 		if (dcr == 0 && dcw == 0 && dce == 0) {
3173 			g_post_event(g_raid3_destroy_delayed, sc, M_WAITOK,
3174 			    sc, NULL);
3175 		}
3176 	}
3177 end:
3178 	sx_xunlock(&sc->sc_lock);
3179 	g_topology_lock();
3180 	return (error);
3181 }
3182 
3183 static struct g_geom *
g_raid3_create(struct g_class * mp,const struct g_raid3_metadata * md)3184 g_raid3_create(struct g_class *mp, const struct g_raid3_metadata *md)
3185 {
3186 	struct g_raid3_softc *sc;
3187 	struct g_geom *gp;
3188 	int error, timeout;
3189 	u_int n;
3190 
3191 	g_topology_assert();
3192 	G_RAID3_DEBUG(1, "Creating device %s (id=%u).", md->md_name, md->md_id);
3193 
3194 	/* One disk is minimum. */
3195 	if (md->md_all < 1)
3196 		return (NULL);
3197 	/*
3198 	 * Action geom.
3199 	 */
3200 	gp = g_new_geom(mp, md->md_name);
3201 	sc = malloc(sizeof(*sc), M_RAID3, M_WAITOK | M_ZERO);
3202 	sc->sc_disks = malloc(sizeof(struct g_raid3_disk) * md->md_all, M_RAID3,
3203 	    M_WAITOK | M_ZERO);
3204 	gp->start = g_raid3_start;
3205 	gp->orphan = g_raid3_orphan;
3206 	gp->access = g_raid3_access;
3207 	gp->dumpconf = g_raid3_dumpconf;
3208 
3209 	sc->sc_id = md->md_id;
3210 	sc->sc_mediasize = md->md_mediasize;
3211 	sc->sc_sectorsize = md->md_sectorsize;
3212 	sc->sc_ndisks = md->md_all;
3213 	sc->sc_round_robin = 0;
3214 	sc->sc_flags = md->md_mflags;
3215 	sc->sc_bump_id = 0;
3216 	sc->sc_idle = 1;
3217 	sc->sc_last_write = time_uptime;
3218 	sc->sc_writes = 0;
3219 	sc->sc_refcnt = 1;
3220 	for (n = 0; n < sc->sc_ndisks; n++) {
3221 		sc->sc_disks[n].d_softc = sc;
3222 		sc->sc_disks[n].d_no = n;
3223 		sc->sc_disks[n].d_state = G_RAID3_DISK_STATE_NODISK;
3224 	}
3225 	sx_init(&sc->sc_lock, "graid3:lock");
3226 	bioq_init(&sc->sc_queue);
3227 	mtx_init(&sc->sc_queue_mtx, "graid3:queue", NULL, MTX_DEF);
3228 	bioq_init(&sc->sc_regular_delayed);
3229 	bioq_init(&sc->sc_inflight);
3230 	bioq_init(&sc->sc_sync_delayed);
3231 	TAILQ_INIT(&sc->sc_events);
3232 	mtx_init(&sc->sc_events_mtx, "graid3:events", NULL, MTX_DEF);
3233 	callout_init(&sc->sc_callout, 1);
3234 	sc->sc_state = G_RAID3_DEVICE_STATE_STARTING;
3235 	gp->softc = sc;
3236 	sc->sc_geom = gp;
3237 	sc->sc_provider = NULL;
3238 	/*
3239 	 * Synchronization geom.
3240 	 */
3241 	gp = g_new_geomf(mp, "%s.sync", md->md_name);
3242 	gp->softc = sc;
3243 	gp->orphan = g_raid3_orphan;
3244 	sc->sc_sync.ds_geom = gp;
3245 
3246 	if (!g_raid3_use_malloc) {
3247 		sc->sc_zones[G_RAID3_ZONE_64K].sz_zone = uma_zcreate("gr3:64k",
3248 		    65536, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3249 		    UMA_ALIGN_PTR, 0);
3250 		sc->sc_zones[G_RAID3_ZONE_64K].sz_inuse = 0;
3251 		sc->sc_zones[G_RAID3_ZONE_64K].sz_max = g_raid3_n64k;
3252 		sc->sc_zones[G_RAID3_ZONE_64K].sz_requested =
3253 		    sc->sc_zones[G_RAID3_ZONE_64K].sz_failed = 0;
3254 		sc->sc_zones[G_RAID3_ZONE_16K].sz_zone = uma_zcreate("gr3:16k",
3255 		    16384, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3256 		    UMA_ALIGN_PTR, 0);
3257 		sc->sc_zones[G_RAID3_ZONE_16K].sz_inuse = 0;
3258 		sc->sc_zones[G_RAID3_ZONE_16K].sz_max = g_raid3_n16k;
3259 		sc->sc_zones[G_RAID3_ZONE_16K].sz_requested =
3260 		    sc->sc_zones[G_RAID3_ZONE_16K].sz_failed = 0;
3261 		sc->sc_zones[G_RAID3_ZONE_4K].sz_zone = uma_zcreate("gr3:4k",
3262 		    4096, g_raid3_uma_ctor, g_raid3_uma_dtor, NULL, NULL,
3263 		    UMA_ALIGN_PTR, 0);
3264 		sc->sc_zones[G_RAID3_ZONE_4K].sz_inuse = 0;
3265 		sc->sc_zones[G_RAID3_ZONE_4K].sz_max = g_raid3_n4k;
3266 		sc->sc_zones[G_RAID3_ZONE_4K].sz_requested =
3267 		    sc->sc_zones[G_RAID3_ZONE_4K].sz_failed = 0;
3268 	}
3269 
3270 	error = kproc_create(g_raid3_worker, sc, &sc->sc_worker, 0, 0,
3271 	    "g_raid3 %s", md->md_name);
3272 	if (error != 0) {
3273 		G_RAID3_DEBUG(1, "Cannot create kernel thread for %s.",
3274 		    sc->sc_name);
3275 		g_destroy_geom(sc->sc_geom);
3276 		g_raid3_free_device(sc);
3277 		return (NULL);
3278 	}
3279 
3280 	G_RAID3_DEBUG(1, "Device %s created (%u components, id=%u).",
3281 	    sc->sc_name, sc->sc_ndisks, sc->sc_id);
3282 
3283 	sc->sc_rootmount = root_mount_hold("GRAID3");
3284 	G_RAID3_DEBUG(1, "root_mount_hold %p", sc->sc_rootmount);
3285 
3286 	/*
3287 	 * Schedule startup timeout.
3288 	 */
3289 	timeout = atomic_load_acq_int(&g_raid3_timeout);
3290 	sc->sc_timeout_event = malloc(sizeof(struct g_raid3_event), M_RAID3,
3291 	    M_WAITOK);
3292 	callout_reset(&sc->sc_callout, timeout * hz, g_raid3_go, sc);
3293 	return (sc->sc_geom);
3294 }
3295 
3296 int
g_raid3_destroy(struct g_raid3_softc * sc,int how)3297 g_raid3_destroy(struct g_raid3_softc *sc, int how)
3298 {
3299 	struct g_provider *pp;
3300 
3301 	g_topology_assert_not();
3302 	sx_assert(&sc->sc_lock, SX_XLOCKED);
3303 
3304 	pp = sc->sc_provider;
3305 	if (pp != NULL && (pp->acr != 0 || pp->acw != 0 || pp->ace != 0)) {
3306 		switch (how) {
3307 		case G_RAID3_DESTROY_SOFT:
3308 			G_RAID3_DEBUG(1,
3309 			    "Device %s is still open (r%dw%de%d).", pp->name,
3310 			    pp->acr, pp->acw, pp->ace);
3311 			return (EBUSY);
3312 		case G_RAID3_DESTROY_DELAYED:
3313 			G_RAID3_DEBUG(1,
3314 			    "Device %s will be destroyed on last close.",
3315 			    pp->name);
3316 			if (sc->sc_syncdisk != NULL)
3317 				g_raid3_sync_stop(sc, 1);
3318 			sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROYING;
3319 			return (EBUSY);
3320 		case G_RAID3_DESTROY_HARD:
3321 			G_RAID3_DEBUG(1, "Device %s is still open, so it "
3322 			    "can't be definitely removed.", pp->name);
3323 			break;
3324 		}
3325 	}
3326 
3327 	g_topology_lock();
3328 	if (sc->sc_geom->softc == NULL) {
3329 		g_topology_unlock();
3330 		return (0);
3331 	}
3332 	sc->sc_geom->softc = NULL;
3333 	sc->sc_sync.ds_geom->softc = NULL;
3334 	g_topology_unlock();
3335 
3336 	sc->sc_flags |= G_RAID3_DEVICE_FLAG_DESTROY;
3337 	sc->sc_flags |= G_RAID3_DEVICE_FLAG_WAIT;
3338 	G_RAID3_DEBUG(4, "%s: Waking up %p.", __func__, sc);
3339 	sx_xunlock(&sc->sc_lock);
3340 	mtx_lock(&sc->sc_queue_mtx);
3341 	wakeup(sc);
3342 	wakeup(&sc->sc_queue);
3343 	mtx_unlock(&sc->sc_queue_mtx);
3344 	G_RAID3_DEBUG(4, "%s: Sleeping %p.", __func__, &sc->sc_worker);
3345 	while (sc->sc_worker != NULL)
3346 		tsleep(&sc->sc_worker, PRIBIO, "r3:destroy", hz / 5);
3347 	G_RAID3_DEBUG(4, "%s: Woken up %p.", __func__, &sc->sc_worker);
3348 	sx_xlock(&sc->sc_lock);
3349 	g_raid3_destroy_device(sc);
3350 	return (0);
3351 }
3352 
3353 static void
g_raid3_taste_orphan(struct g_consumer * cp)3354 g_raid3_taste_orphan(struct g_consumer *cp)
3355 {
3356 
3357 	KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
3358 	    cp->provider->name));
3359 }
3360 
3361 static struct g_geom *
g_raid3_taste(struct g_class * mp,struct g_provider * pp,int flags __unused)3362 g_raid3_taste(struct g_class *mp, struct g_provider *pp, int flags __unused)
3363 {
3364 	struct g_raid3_metadata md;
3365 	struct g_raid3_softc *sc;
3366 	struct g_consumer *cp;
3367 	struct g_geom *gp;
3368 	int error;
3369 
3370 	g_topology_assert();
3371 	g_trace(G_T_TOPOLOGY, "%s(%s, %s)", __func__, mp->name, pp->name);
3372 	G_RAID3_DEBUG(2, "Tasting %s.", pp->name);
3373 
3374 	gp = g_new_geom(mp, "raid3:taste");
3375 	/* This orphan function should be never called. */
3376 	gp->orphan = g_raid3_taste_orphan;
3377 	cp = g_new_consumer(gp);
3378 	cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
3379 	error = g_attach(cp, pp);
3380 	if (error == 0) {
3381 		error = g_raid3_read_metadata(cp, &md);
3382 		g_detach(cp);
3383 	}
3384 	g_destroy_consumer(cp);
3385 	g_destroy_geom(gp);
3386 	if (error != 0)
3387 		return (NULL);
3388 	gp = NULL;
3389 
3390 	if (md.md_provider[0] != '\0' &&
3391 	    !g_compare_names(md.md_provider, pp->name))
3392 		return (NULL);
3393 	if (md.md_provsize != 0 && md.md_provsize != pp->mediasize)
3394 		return (NULL);
3395 	if (g_raid3_debug >= 2)
3396 		raid3_metadata_dump(&md);
3397 
3398 	/*
3399 	 * Let's check if device already exists.
3400 	 */
3401 	sc = NULL;
3402 	LIST_FOREACH(gp, &mp->geom, geom) {
3403 		sc = gp->softc;
3404 		if (sc == NULL)
3405 			continue;
3406 		if (sc->sc_sync.ds_geom == gp)
3407 			continue;
3408 		if (strcmp(md.md_name, sc->sc_name) != 0)
3409 			continue;
3410 		if (md.md_id != sc->sc_id) {
3411 			G_RAID3_DEBUG(0, "Device %s already configured.",
3412 			    sc->sc_name);
3413 			return (NULL);
3414 		}
3415 		break;
3416 	}
3417 	if (gp == NULL) {
3418 		gp = g_raid3_create(mp, &md);
3419 		if (gp == NULL) {
3420 			G_RAID3_DEBUG(0, "Cannot create device %s.",
3421 			    md.md_name);
3422 			return (NULL);
3423 		}
3424 		sc = gp->softc;
3425 	}
3426 	G_RAID3_DEBUG(1, "Adding disk %s to %s.", pp->name, gp->name);
3427 	g_topology_unlock();
3428 	sx_xlock(&sc->sc_lock);
3429 	error = g_raid3_add_disk(sc, pp, &md);
3430 	if (error != 0) {
3431 		G_RAID3_DEBUG(0, "Cannot add disk %s to %s (error=%d).",
3432 		    pp->name, gp->name, error);
3433 		if (g_raid3_ndisks(sc, G_RAID3_DISK_STATE_NODISK) ==
3434 		    sc->sc_ndisks) {
3435 			g_cancel_event(sc);
3436 			g_raid3_destroy(sc, G_RAID3_DESTROY_HARD);
3437 			g_topology_lock();
3438 			return (NULL);
3439 		}
3440 		gp = NULL;
3441 	}
3442 	sx_xunlock(&sc->sc_lock);
3443 	g_topology_lock();
3444 	return (gp);
3445 }
3446 
3447 static int
g_raid3_destroy_geom(struct gctl_req * req __unused,struct g_class * mp __unused,struct g_geom * gp)3448 g_raid3_destroy_geom(struct gctl_req *req __unused, struct g_class *mp __unused,
3449     struct g_geom *gp)
3450 {
3451 	struct g_raid3_softc *sc;
3452 	int error;
3453 
3454 	g_topology_unlock();
3455 	sc = gp->softc;
3456 	sx_xlock(&sc->sc_lock);
3457 	g_cancel_event(sc);
3458 	error = g_raid3_destroy(gp->softc, G_RAID3_DESTROY_SOFT);
3459 	if (error != 0)
3460 		sx_xunlock(&sc->sc_lock);
3461 	g_topology_lock();
3462 	return (error);
3463 }
3464 
3465 static void
g_raid3_dumpconf(struct sbuf * sb,const char * indent,struct g_geom * gp,struct g_consumer * cp,struct g_provider * pp)3466 g_raid3_dumpconf(struct sbuf *sb, const char *indent, struct g_geom *gp,
3467     struct g_consumer *cp, struct g_provider *pp)
3468 {
3469 	struct g_raid3_softc *sc;
3470 
3471 	g_topology_assert();
3472 
3473 	sc = gp->softc;
3474 	if (sc == NULL)
3475 		return;
3476 	/* Skip synchronization geom. */
3477 	if (gp == sc->sc_sync.ds_geom)
3478 		return;
3479 	if (pp != NULL) {
3480 		/* Nothing here. */
3481 	} else if (cp != NULL) {
3482 		struct g_raid3_disk *disk;
3483 
3484 		disk = cp->private;
3485 		if (disk == NULL)
3486 			return;
3487 		g_topology_unlock();
3488 		sx_xlock(&sc->sc_lock);
3489 		sbuf_printf(sb, "%s<Type>", indent);
3490 		if (disk->d_no == sc->sc_ndisks - 1)
3491 			sbuf_cat(sb, "PARITY");
3492 		else
3493 			sbuf_cat(sb, "DATA");
3494 		sbuf_cat(sb, "</Type>\n");
3495 		sbuf_printf(sb, "%s<Number>%u</Number>\n", indent,
3496 		    (u_int)disk->d_no);
3497 		if (disk->d_state == G_RAID3_DISK_STATE_SYNCHRONIZING) {
3498 			sbuf_printf(sb, "%s<Synchronized>", indent);
3499 			if (disk->d_sync.ds_offset == 0)
3500 				sbuf_cat(sb, "0%");
3501 			else {
3502 				sbuf_printf(sb, "%u%%",
3503 				    (u_int)((disk->d_sync.ds_offset * 100) /
3504 				    (sc->sc_mediasize / (sc->sc_ndisks - 1))));
3505 			}
3506 			sbuf_cat(sb, "</Synchronized>\n");
3507 			if (disk->d_sync.ds_offset > 0) {
3508 				sbuf_printf(sb, "%s<BytesSynced>%jd"
3509 				    "</BytesSynced>\n", indent,
3510 				    (intmax_t)disk->d_sync.ds_offset);
3511 			}
3512 		}
3513 		sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent,
3514 		    disk->d_sync.ds_syncid);
3515 		sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, disk->d_genid);
3516 		sbuf_printf(sb, "%s<Flags>", indent);
3517 		if (disk->d_flags == 0)
3518 			sbuf_cat(sb, "NONE");
3519 		else {
3520 			int first = 1;
3521 
3522 #define	ADD_FLAG(flag, name)	do {					\
3523 	if ((disk->d_flags & (flag)) != 0) {				\
3524 		if (!first)						\
3525 			sbuf_cat(sb, ", ");				\
3526 		else							\
3527 			first = 0;					\
3528 		sbuf_cat(sb, name);					\
3529 	}								\
3530 } while (0)
3531 			ADD_FLAG(G_RAID3_DISK_FLAG_DIRTY, "DIRTY");
3532 			ADD_FLAG(G_RAID3_DISK_FLAG_HARDCODED, "HARDCODED");
3533 			ADD_FLAG(G_RAID3_DISK_FLAG_SYNCHRONIZING,
3534 			    "SYNCHRONIZING");
3535 			ADD_FLAG(G_RAID3_DISK_FLAG_FORCE_SYNC, "FORCE_SYNC");
3536 			ADD_FLAG(G_RAID3_DISK_FLAG_BROKEN, "BROKEN");
3537 #undef	ADD_FLAG
3538 		}
3539 		sbuf_cat(sb, "</Flags>\n");
3540 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
3541 		    g_raid3_disk_state2str(disk->d_state));
3542 		sx_xunlock(&sc->sc_lock);
3543 		g_topology_lock();
3544 	} else {
3545 		g_topology_unlock();
3546 		sx_xlock(&sc->sc_lock);
3547 		if (!g_raid3_use_malloc) {
3548 			sbuf_printf(sb,
3549 			    "%s<Zone4kRequested>%u</Zone4kRequested>\n", indent,
3550 			    sc->sc_zones[G_RAID3_ZONE_4K].sz_requested);
3551 			sbuf_printf(sb,
3552 			    "%s<Zone4kFailed>%u</Zone4kFailed>\n", indent,
3553 			    sc->sc_zones[G_RAID3_ZONE_4K].sz_failed);
3554 			sbuf_printf(sb,
3555 			    "%s<Zone16kRequested>%u</Zone16kRequested>\n", indent,
3556 			    sc->sc_zones[G_RAID3_ZONE_16K].sz_requested);
3557 			sbuf_printf(sb,
3558 			    "%s<Zone16kFailed>%u</Zone16kFailed>\n", indent,
3559 			    sc->sc_zones[G_RAID3_ZONE_16K].sz_failed);
3560 			sbuf_printf(sb,
3561 			    "%s<Zone64kRequested>%u</Zone64kRequested>\n", indent,
3562 			    sc->sc_zones[G_RAID3_ZONE_64K].sz_requested);
3563 			sbuf_printf(sb,
3564 			    "%s<Zone64kFailed>%u</Zone64kFailed>\n", indent,
3565 			    sc->sc_zones[G_RAID3_ZONE_64K].sz_failed);
3566 		}
3567 		sbuf_printf(sb, "%s<ID>%u</ID>\n", indent, (u_int)sc->sc_id);
3568 		sbuf_printf(sb, "%s<SyncID>%u</SyncID>\n", indent, sc->sc_syncid);
3569 		sbuf_printf(sb, "%s<GenID>%u</GenID>\n", indent, sc->sc_genid);
3570 		sbuf_printf(sb, "%s<Flags>", indent);
3571 		if (sc->sc_flags == 0)
3572 			sbuf_cat(sb, "NONE");
3573 		else {
3574 			int first = 1;
3575 
3576 #define	ADD_FLAG(flag, name)	do {					\
3577 	if ((sc->sc_flags & (flag)) != 0) {				\
3578 		if (!first)						\
3579 			sbuf_cat(sb, ", ");				\
3580 		else							\
3581 			first = 0;					\
3582 		sbuf_cat(sb, name);					\
3583 	}								\
3584 } while (0)
3585 			ADD_FLAG(G_RAID3_DEVICE_FLAG_NOFAILSYNC, "NOFAILSYNC");
3586 			ADD_FLAG(G_RAID3_DEVICE_FLAG_NOAUTOSYNC, "NOAUTOSYNC");
3587 			ADD_FLAG(G_RAID3_DEVICE_FLAG_ROUND_ROBIN,
3588 			    "ROUND-ROBIN");
3589 			ADD_FLAG(G_RAID3_DEVICE_FLAG_VERIFY, "VERIFY");
3590 #undef	ADD_FLAG
3591 		}
3592 		sbuf_cat(sb, "</Flags>\n");
3593 		sbuf_printf(sb, "%s<Components>%u</Components>\n", indent,
3594 		    sc->sc_ndisks);
3595 		sbuf_printf(sb, "%s<State>%s</State>\n", indent,
3596 		    g_raid3_device_state2str(sc->sc_state));
3597 		sx_xunlock(&sc->sc_lock);
3598 		g_topology_lock();
3599 	}
3600 }
3601 
3602 static void
g_raid3_shutdown_post_sync(void * arg,int howto)3603 g_raid3_shutdown_post_sync(void *arg, int howto)
3604 {
3605 	struct g_class *mp;
3606 	struct g_geom *gp, *gp2;
3607 	struct g_raid3_softc *sc;
3608 	int error;
3609 
3610 	if ((howto & RB_NOSYNC) != 0)
3611 		return;
3612 
3613 	mp = arg;
3614 	g_topology_lock();
3615 	g_raid3_shutdown = 1;
3616 	LIST_FOREACH_SAFE(gp, &mp->geom, geom, gp2) {
3617 		if ((sc = gp->softc) == NULL)
3618 			continue;
3619 		/* Skip synchronization geom. */
3620 		if (gp == sc->sc_sync.ds_geom)
3621 			continue;
3622 		g_topology_unlock();
3623 		sx_xlock(&sc->sc_lock);
3624 		g_raid3_idle(sc, -1);
3625 		g_cancel_event(sc);
3626 		error = g_raid3_destroy(sc, G_RAID3_DESTROY_DELAYED);
3627 		if (error != 0)
3628 			sx_xunlock(&sc->sc_lock);
3629 		g_topology_lock();
3630 	}
3631 	g_topology_unlock();
3632 }
3633 
3634 static void
g_raid3_init(struct g_class * mp)3635 g_raid3_init(struct g_class *mp)
3636 {
3637 
3638 	g_raid3_post_sync = EVENTHANDLER_REGISTER(shutdown_post_sync,
3639 	    g_raid3_shutdown_post_sync, mp, SHUTDOWN_PRI_FIRST);
3640 	if (g_raid3_post_sync == NULL)
3641 		G_RAID3_DEBUG(0, "Warning! Cannot register shutdown event.");
3642 }
3643 
3644 static void
g_raid3_fini(struct g_class * mp)3645 g_raid3_fini(struct g_class *mp)
3646 {
3647 
3648 	if (g_raid3_post_sync != NULL)
3649 		EVENTHANDLER_DEREGISTER(shutdown_post_sync, g_raid3_post_sync);
3650 }
3651 
3652 DECLARE_GEOM_CLASS(g_raid3_class, g_raid3);
3653 MODULE_VERSION(geom_raid3, 0);
3654