2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2004, 2007 Lukas Ertl
5 * Copyright (c) 2007, 2009 Ulf Lilleengen
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
33 #include <sys/param.h>
36 #include <sys/malloc.h>
37 #include <sys/systm.h>
39 #include <geom/geom.h>
40 #include <geom/geom_dbg.h>
41 #include <geom/vinum/geom_vinum_var.h>
42 #include <geom/vinum/geom_vinum_raid5.h>
43 #include <geom/vinum/geom_vinum.h>
45 static int gv_check_parity(struct gv_plex *, struct bio *,
46 struct gv_raid5_packet *);
47 static int gv_normal_parity(struct gv_plex *, struct bio *,
48 struct gv_raid5_packet *);
49 static void gv_plex_flush(struct gv_plex *);
50 static int gv_plex_offset(struct gv_plex *, off_t, off_t, off_t *, off_t *,
52 static int gv_plex_normal_request(struct gv_plex *, struct bio *, off_t,
54 static void gv_post_bio(struct gv_softc *, struct bio *);
57 gv_plex_start(struct gv_plex *p, struct bio *bp)
61 struct gv_raid5_packet *wp;
63 off_t bcount, boff, len;
65 bcount = bp->bio_length;
67 boff = bp->bio_offset;
69 /* Walk over the whole length of the request, we might split it up. */
74 * RAID5 plexes need special treatment, as a single request
75 * might involve several read/write sub-requests.
77 if (p->org == GV_PLEX_RAID5) {
78 wp = gv_raid5_start(p, bp, addr, boff, bcount);
84 if (TAILQ_EMPTY(&wp->bits))
86 else if (wp->lockbase != -1)
87 TAILQ_INSERT_TAIL(&p->packets, wp, list);
90 * Requests to concatenated and striped plexes go straight
94 len = gv_plex_normal_request(p, bp, boff, bcount, addr);
105 * Fire off all sub-requests. We get the correct consumer (== drive)
106 * to send each request to via the subdisk that was stored in
109 cbp = bioq_takefirst(p->bqueue);
110 while (cbp != NULL) {
112 * RAID5 sub-requests need to come in correct order, otherwise
113 * we trip over the parity, as it might be overwritten by
114 * another sub-request. We abuse cbp->bio_caller2 to mark
115 * potential overlap situations.
117 if (cbp->bio_caller2 != NULL && gv_stripe_active(p, cbp)) {
118 /* Park the bio on the waiting queue. */
119 cbp->bio_pflags |= GV_BIO_ONHOLD;
120 bioq_disksort(p->wqueue, cbp);
122 s = cbp->bio_caller1;
123 g_io_request(cbp, s->drive_sc->consumer);
125 cbp = bioq_takefirst(p->bqueue);
130 gv_plex_offset(struct gv_plex *p, off_t boff, off_t bcount, off_t *real_off,
131 off_t *real_len, int *sdno, int growing)
135 off_t len_left, stripeend, stripeno, stripestart;
140 * Find the subdisk where this request starts. The subdisks in
141 * this list must be ordered by plex_offset.
144 LIST_FOREACH(s, &p->subdisks, in_plex) {
145 if (s->plex_offset <= boff &&
146 s->plex_offset + s->size > boff) {
152 if (s == NULL || s->drive_sc == NULL)
153 return (GV_ERR_NOTFOUND);
155 /* Calculate corresponding offsets on disk. */
156 *real_off = boff - s->plex_offset;
157 len_left = s->size - (*real_off);
158 KASSERT(len_left >= 0, ("gv_plex_offset: len_left < 0"));
159 *real_len = (bcount > len_left) ? len_left : bcount;
162 case GV_PLEX_STRIPED:
163 /* The number of the stripe where the request starts. */
164 stripeno = boff / p->stripesize;
165 KASSERT(stripeno >= 0, ("gv_plex_offset: stripeno < 0"));
167 /* Take growing subdisks into account when calculating. */
168 sdcount = gv_sdcount(p, (boff >= p->synced));
170 if (!(boff + bcount <= p->synced) &&
171 (p->flags & GV_PLEX_GROWING) &&
173 return (GV_ERR_ISBUSY);
174 *sdno = stripeno % sdcount;
176 KASSERT(sdno >= 0, ("gv_plex_offset: sdno < 0"));
177 stripestart = (stripeno / sdcount) *
179 KASSERT(stripestart >= 0, ("gv_plex_offset: stripestart < 0"));
180 stripeend = stripestart + p->stripesize;
181 *real_off = boff - (stripeno * p->stripesize) +
183 len_left = stripeend - *real_off;
184 KASSERT(len_left >= 0, ("gv_plex_offset: len_left < 0"));
186 *real_len = (bcount <= len_left) ? bcount : len_left;
190 return (GV_ERR_PLEXORG);
196 * Prepare a normal plex request.
199 gv_plex_normal_request(struct gv_plex *p, struct bio *bp, off_t boff,
200 off_t bcount, caddr_t addr)
204 off_t real_len, real_off;
209 real_len = real_off = 0;
213 if (p == NULL || LIST_EMPTY(&p->subdisks))
216 err = gv_plex_offset(p, boff, bcount, &real_off,
217 &real_len, &sdno, (bp->bio_pflags & GV_BIO_GROW));
218 /* If the request was blocked, put it into wait. */
219 if (err == GV_ERR_ISBUSY) {
220 bioq_disksort(p->rqueue, bp);
221 return (-1); /* "Fail", and delay request. */
229 /* Find the right subdisk. */
231 LIST_FOREACH(s, &p->subdisks, in_plex) {
237 /* Subdisk not found. */
238 if (s == NULL || s->drive_sc == NULL)
241 /* Now check if we can handle the request on this subdisk. */
244 /* If the subdisk is up, just continue. */
247 if (bp->bio_pflags & GV_BIO_INTERNAL)
248 G_VINUM_DEBUG(0, "subdisk must be in the stale state in"
249 " order to perform administrative requests");
252 if (!(bp->bio_pflags & GV_BIO_SYNCREQ)) {
253 G_VINUM_DEBUG(0, "subdisk stale, unable to perform "
258 G_VINUM_DEBUG(1, "sd %s is initializing", s->name);
259 gv_set_sd_state(s, GV_SD_INITIALIZING, GV_SETSTATE_FORCE);
261 case GV_SD_INITIALIZING:
262 if (bp->bio_cmd == BIO_READ)
266 /* All other subdisk states mean it's not accessible. */
270 /* Clone the bio and adjust the offsets and sizes. */
271 cbp = g_clone_bio(bp);
276 cbp->bio_offset = real_off + s->drive_offset;
277 cbp->bio_length = real_len;
278 cbp->bio_data = addr;
279 cbp->bio_done = gv_done;
280 cbp->bio_caller1 = s;
281 s->drive_sc->active++;
283 /* Store the sub-requests now and let others issue them. */
284 bioq_insert_tail(p->bqueue, cbp);
287 G_VINUM_LOGREQ(0, bp, "plex request failed.");
288 /* Building the sub-request failed. If internal BIO, do not deliver. */
289 if (bp->bio_pflags & GV_BIO_INTERNAL) {
290 if (bp->bio_pflags & GV_BIO_MALLOC)
291 g_free(bp->bio_data);
293 p->flags &= ~(GV_PLEX_SYNCING | GV_PLEX_REBUILDING |
297 g_io_deliver(bp, err);
302 * Handle a completed request to a striped or concatenated plex.
305 gv_plex_normal_done(struct gv_plex *p, struct bio *bp)
309 pbp = bp->bio_parent;
310 if (pbp->bio_error == 0)
311 pbp->bio_error = bp->bio_error;
314 if (pbp->bio_children == pbp->bio_inbed) {
315 /* Just set it to length since multiple plexes will
316 * screw things up. */
317 pbp->bio_completed = pbp->bio_length;
318 if (pbp->bio_pflags & GV_BIO_SYNCREQ)
319 gv_sync_complete(p, pbp);
320 else if (pbp->bio_pflags & GV_BIO_GROW)
321 gv_grow_complete(p, pbp);
323 g_io_deliver(pbp, pbp->bio_error);
328 * Handle a completed request to a RAID-5 plex.
331 gv_plex_raid5_done(struct gv_plex *p, struct bio *bp)
334 struct bio *cbp, *pbp;
335 struct gv_bioq *bq, *bq2;
336 struct gv_raid5_packet *wp;
342 wp = bp->bio_caller2;
344 switch (bp->bio_parent->bio_cmd) {
347 completed = bp->bio_completed;
351 TAILQ_FOREACH_SAFE(bq, &wp->bits, queue, bq2) {
354 TAILQ_REMOVE(&wp->bits, bq, queue);
356 for (i = 0; i < wp->length; i++)
357 wp->data[i] ^= bp->bio_data[i];
360 if (TAILQ_EMPTY(&wp->bits)) {
361 completed = wp->length;
362 if (wp->lockbase != -1) {
363 TAILQ_REMOVE(&p->packets, wp, list);
364 /* Bring the waiting bios back into the game. */
365 pbp = bioq_takefirst(p->wqueue);
366 while (pbp != NULL) {
367 gv_post_bio(sc, pbp);
368 pbp = bioq_takefirst(p->wqueue);
377 /* XXX can this ever happen? */
379 completed = bp->bio_completed;
383 /* Check if we need to handle parity data. */
384 TAILQ_FOREACH_SAFE(bq, &wp->bits, queue, bq2) {
387 TAILQ_REMOVE(&wp->bits, bq, queue);
391 for (i = 0; i < wp->length; i++)
392 cbp->bio_data[i] ^= bp->bio_data[i];
397 /* Handle parity data. */
398 if (TAILQ_EMPTY(&wp->bits)) {
399 if (bp->bio_parent->bio_pflags & GV_BIO_CHECK)
400 i = gv_check_parity(p, bp, wp);
402 i = gv_normal_parity(p, bp, wp);
404 /* All of our sub-requests have finished. */
406 completed = wp->length;
407 TAILQ_REMOVE(&p->packets, wp, list);
408 /* Bring the waiting bios back into the game. */
409 pbp = bioq_takefirst(p->wqueue);
410 while (pbp != NULL) {
411 gv_post_bio(sc, pbp);
412 pbp = bioq_takefirst(p->wqueue);
421 pbp = bp->bio_parent;
422 if (pbp->bio_error == 0)
423 pbp->bio_error = bp->bio_error;
424 pbp->bio_completed += completed;
426 /* When the original request is finished, we deliver it. */
428 if (pbp->bio_inbed == pbp->bio_children) {
429 /* Hand it over for checking or delivery. */
430 if (pbp->bio_cmd == BIO_WRITE &&
431 (pbp->bio_pflags & GV_BIO_CHECK)) {
432 gv_parity_complete(p, pbp);
433 } else if (pbp->bio_cmd == BIO_WRITE &&
434 (pbp->bio_pflags & GV_BIO_REBUILD)) {
435 gv_rebuild_complete(p, pbp);
436 } else if (pbp->bio_pflags & GV_BIO_INIT) {
437 gv_init_complete(p, pbp);
438 } else if (pbp->bio_pflags & GV_BIO_SYNCREQ) {
439 gv_sync_complete(p, pbp);
440 } else if (pbp->bio_pflags & GV_BIO_GROW) {
441 gv_grow_complete(p, pbp);
443 g_io_deliver(pbp, pbp->bio_error);
447 /* Clean up what we allocated. */
448 if (bp->bio_cflags & GV_BIO_MALLOC)
449 g_free(bp->bio_data);
454 gv_check_parity(struct gv_plex *p, struct bio *bp, struct gv_raid5_packet *wp)
458 int err, finished, i;
463 if (wp->waiting != NULL) {
466 s = pbp->bio_caller1;
467 g_io_request(pbp, s->drive_sc->consumer);
470 } else if (wp->parity != NULL) {
474 /* Check if the parity is correct. */
475 for (i = 0; i < wp->length; i++) {
476 if (bp->bio_data[i] != pbp->bio_data[i]) {
482 /* The parity is not correct... */
484 bp->bio_parent->bio_error = EAGAIN;
486 /* ... but we rebuild it. */
487 if (bp->bio_parent->bio_pflags & GV_BIO_PARITY) {
488 s = pbp->bio_caller1;
489 g_io_request(pbp, s->drive_sc->consumer);
495 * Clean up the BIO we would have used for rebuilding the
499 bp->bio_parent->bio_inbed++;
508 gv_normal_parity(struct gv_plex *p, struct bio *bp, struct gv_raid5_packet *wp)
510 struct bio *cbp, *pbp;
516 if (wp->waiting != NULL) {
520 for (i = 0; i < wp->length; i++)
521 cbp->bio_data[i] ^= pbp->bio_data[i];
522 s = pbp->bio_caller1;
523 g_io_request(pbp, s->drive_sc->consumer);
526 } else if (wp->parity != NULL) {
529 s = cbp->bio_caller1;
530 g_io_request(cbp, s->drive_sc->consumer);
537 /* Flush the queue with delayed requests. */
539 gv_plex_flush(struct gv_plex *p)
543 bp = bioq_takefirst(p->rqueue);
545 gv_plex_start(p, bp);
546 bp = bioq_takefirst(p->rqueue);
551 gv_post_bio(struct gv_softc *sc, struct bio *bp)
554 KASSERT(sc != NULL, ("NULL sc"));
555 KASSERT(bp != NULL, ("NULL bp"));
556 mtx_lock(&sc->bqueue_mtx);
557 bioq_disksort(sc->bqueue_down, bp);
559 mtx_unlock(&sc->bqueue_mtx);
563 gv_sync_request(struct gv_plex *from, struct gv_plex *to, off_t offset,
564 off_t length, int type, caddr_t data)
569 KASSERT(from != NULL, ("NULL from"));
570 KASSERT(to != NULL, ("NULL to"));
571 sc = from->vinumconf;
572 KASSERT(sc != NULL, ("NULL sc"));
576 G_VINUM_DEBUG(0, "sync from '%s' failed at offset "
577 " %jd; out of memory", from->name, offset);
580 bp->bio_length = length;
582 bp->bio_pflags |= GV_BIO_SYNCREQ;
583 bp->bio_offset = offset;
584 bp->bio_caller1 = from;
585 bp->bio_caller2 = to;
588 data = g_malloc(length, M_WAITOK);
589 bp->bio_pflags |= GV_BIO_MALLOC; /* Free on the next run. */
592 /* Send down next. */
594 //gv_plex_start(from, bp);
599 * Handle a finished plex sync bio.
602 gv_sync_complete(struct gv_plex *to, struct bio *bp)
604 struct gv_plex *from, *p;
611 g_topology_assert_not();
614 KASSERT(to != NULL, ("NULL to"));
615 KASSERT(bp != NULL, ("NULL bp"));
616 from = bp->bio_caller2;
617 KASSERT(from != NULL, ("NULL from"));
619 KASSERT(v != NULL, ("NULL v"));
621 KASSERT(sc != NULL, ("NULL sc"));
623 /* If it was a read, write it. */
624 if (bp->bio_cmd == BIO_READ) {
625 err = gv_sync_request(from, to, bp->bio_offset, bp->bio_length,
626 BIO_WRITE, bp->bio_data);
627 /* If it was a write, read the next one. */
628 } else if (bp->bio_cmd == BIO_WRITE) {
629 if (bp->bio_pflags & GV_BIO_MALLOC)
630 g_free(bp->bio_data);
631 to->synced += bp->bio_length;
632 /* If we're finished, clean up. */
633 if (bp->bio_offset + bp->bio_length >= from->size) {
634 G_VINUM_DEBUG(1, "syncing of %s from %s completed",
635 to->name, from->name);
636 /* Update our state. */
637 LIST_FOREACH(s, &to->subdisks, in_plex)
638 gv_set_sd_state(s, GV_SD_UP, 0);
639 gv_update_plex_state(to);
640 to->flags &= ~GV_PLEX_SYNCING;
642 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
644 offset = bp->bio_offset + bp->bio_length;
645 err = gv_sync_request(from, to, offset,
646 MIN(bp->bio_length, from->size - offset),
651 /* Clean up if there was an error. */
653 to->flags &= ~GV_PLEX_SYNCING;
654 G_VINUM_DEBUG(0, "error syncing plexes: error code %d", err);
657 /* Check if all plexes are synced, and lower refcounts. */
659 LIST_FOREACH(p, &v->plexes, in_volume) {
660 if (p->flags & GV_PLEX_SYNCING) {
665 /* If we came here, all plexes are synced, and we're free. */
666 gv_access(v->provider, -1, -1, 0);
668 G_VINUM_DEBUG(1, "plex sync completed");
674 * Create a new bio struct for the next grow request.
677 gv_grow_request(struct gv_plex *p, off_t offset, off_t length, int type,
683 KASSERT(p != NULL, ("gv_grow_request: NULL p"));
685 KASSERT(sc != NULL, ("gv_grow_request: NULL sc"));
689 G_VINUM_DEBUG(0, "grow of %s failed creating bio: "
690 "out of memory", p->name);
698 bp->bio_offset = offset;
699 bp->bio_length = length;
700 bp->bio_pflags |= GV_BIO_GROW;
702 data = g_malloc(length, M_WAITOK);
703 bp->bio_pflags |= GV_BIO_MALLOC;
707 //gv_plex_start(p, bp);
712 * Finish handling of a bio to a growing plex.
715 gv_grow_complete(struct gv_plex *p, struct bio *bp)
720 off_t origsize, offset;
724 KASSERT(v != NULL, ("gv_grow_complete: NULL v"));
726 KASSERT(sc != NULL, ("gv_grow_complete: NULL sc"));
729 /* If it was a read, write it. */
730 if (bp->bio_cmd == BIO_READ) {
731 p->synced += bp->bio_length;
732 err = gv_grow_request(p, bp->bio_offset, bp->bio_length,
733 BIO_WRITE, bp->bio_data);
734 /* If it was a write, read next. */
735 } else if (bp->bio_cmd == BIO_WRITE) {
736 if (bp->bio_pflags & GV_BIO_MALLOC)
737 g_free(bp->bio_data);
739 /* Find the real size of the plex. */
740 sdcount = gv_sdcount(p, 1);
741 s = LIST_FIRST(&p->subdisks);
742 KASSERT(s != NULL, ("NULL s"));
743 origsize = (s->size * (sdcount - 1));
744 if (bp->bio_offset + bp->bio_length >= origsize) {
745 G_VINUM_DEBUG(1, "growing of %s completed", p->name);
746 p->flags &= ~GV_PLEX_GROWING;
747 LIST_FOREACH(s, &p->subdisks, in_plex) {
748 s->flags &= ~GV_SD_GROW;
749 gv_set_sd_state(s, GV_SD_UP, 0);
751 p->size = gv_plex_size(p);
752 gv_update_vol_size(v, gv_vol_size(v));
753 gv_set_plex_state(p, GV_PLEX_UP, 0);
755 gv_access(v->provider, -1, -1, 0);
758 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
759 /* Issue delayed requests. */
762 offset = bp->bio_offset + bp->bio_length;
763 err = gv_grow_request(p, offset,
764 MIN(bp->bio_length, origsize - offset),
771 p->flags &= ~GV_PLEX_GROWING;
772 G_VINUM_DEBUG(0, "error growing plex: error code %d", err);
777 * Create an initialization BIO and send it off to the consumer. Assume that
778 * we're given initialization data as parameter.
781 gv_init_request(struct gv_sd *s, off_t start, caddr_t data, off_t length)
784 struct g_consumer *cp;
785 struct bio *bp, *cbp;
787 KASSERT(s != NULL, ("gv_init_request: NULL s"));
789 KASSERT(d != NULL, ("gv_init_request: NULL d"));
791 KASSERT(cp != NULL, ("gv_init_request: NULL cp"));
795 G_VINUM_DEBUG(0, "subdisk '%s' init: write failed at offset %jd"
796 " (drive offset %jd); out of memory", s->name,
797 (intmax_t)s->initialized, (intmax_t)start);
798 return; /* XXX: Error codes. */
800 bp->bio_cmd = BIO_WRITE;
804 bp->bio_length = length;
805 bp->bio_pflags |= GV_BIO_INIT;
806 bp->bio_offset = start;
809 /* Then ofcourse, we have to clone it. */
810 cbp = g_clone_bio(bp);
812 G_VINUM_DEBUG(0, "subdisk '%s' init: write failed at offset %jd"
813 " (drive offset %jd); out of memory", s->name,
814 (intmax_t)s->initialized, (intmax_t)start);
815 return; /* XXX: Error codes. */
817 cbp->bio_done = gv_done;
818 cbp->bio_caller1 = s;
820 /* Send it off to the consumer. */
821 g_io_request(cbp, cp);
825 * Handle a finished initialization BIO.
828 gv_init_complete(struct gv_plex *p, struct bio *bp)
832 struct g_consumer *cp;
839 start = bp->bio_offset;
840 length = bp->bio_length;
841 error = bp->bio_error;
844 KASSERT(s != NULL, ("gv_init_complete: NULL s"));
846 KASSERT(d != NULL, ("gv_init_complete: NULL d"));
848 KASSERT(cp != NULL, ("gv_init_complete: NULL cp"));
850 KASSERT(sc != NULL, ("gv_init_complete: NULL sc"));
855 * First we need to find out if it was okay, and abort if it's not.
856 * Then we need to free previous buffers, find out the correct subdisk,
857 * as well as getting the correct starting point and length of the BIO.
859 if (start >= s->drive_offset + s->size) {
860 /* Free the data we initialized. */
863 g_topology_assert_not();
865 g_access(cp, 0, -1, 0);
868 gv_set_sd_state(s, GV_SD_STALE, GV_SETSTATE_FORCE |
871 gv_set_sd_state(s, GV_SD_UP, GV_SETSTATE_CONFIG);
873 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
874 G_VINUM_DEBUG(1, "subdisk '%s' init: finished "
875 "successfully", s->name);
879 s->initialized += length;
881 gv_init_request(s, start, data, length);
885 * Create a new bio struct for the next parity rebuild. Used both by internal
886 * rebuild of degraded plexes as well as user initiated rebuilds/checks.
889 gv_parity_request(struct gv_plex *p, int flags, off_t offset)
894 KASSERT(p != NULL, ("gv_parity_request: NULL p"));
896 KASSERT(sc != NULL, ("gv_parity_request: NULL sc"));
900 G_VINUM_DEBUG(0, "rebuild of %s failed creating bio: "
901 "out of memory", p->name);
905 bp->bio_cmd = BIO_WRITE;
908 bp->bio_length = p->stripesize;
912 * Check if it's a rebuild of a degraded plex or a user request of
915 if (flags & GV_BIO_REBUILD)
916 bp->bio_data = g_malloc(GV_DFLT_SYNCSIZE, M_WAITOK);
917 else if (flags & GV_BIO_CHECK)
918 bp->bio_data = g_malloc(p->stripesize, M_WAITOK | M_ZERO);
920 G_VINUM_DEBUG(0, "invalid flags given in rebuild");
924 bp->bio_pflags = flags;
925 bp->bio_pflags |= GV_BIO_MALLOC;
927 /* We still have more parity to build. */
928 bp->bio_offset = offset;
930 //gv_plex_start(p, bp); /* Send it down to the plex. */
934 * Handle a finished parity write.
937 gv_parity_complete(struct gv_plex *p, struct bio *bp)
942 error = bp->bio_error;
943 flags = bp->bio_pflags;
944 flags &= ~GV_BIO_MALLOC;
947 KASSERT(sc != NULL, ("gv_parity_complete: NULL sc"));
949 /* Clean up what we allocated. */
950 if (bp->bio_pflags & GV_BIO_MALLOC)
951 g_free(bp->bio_data);
954 if (error == EAGAIN) {
955 G_VINUM_DEBUG(0, "parity incorrect at offset 0x%jx",
956 (intmax_t)p->synced);
959 /* Any error is fatal, except EAGAIN when we're rebuilding. */
960 if (error && !(error == EAGAIN && (flags & GV_BIO_PARITY))) {
961 /* Make sure we don't have the lock. */
962 g_topology_assert_not();
964 gv_access(p->vol_sc->provider, -1, -1, 0);
966 G_VINUM_DEBUG(0, "parity check on %s failed at 0x%jx "
967 "errno %d", p->name, (intmax_t)p->synced, error);
970 p->synced += p->stripesize;
973 if (p->synced >= p->size) {
974 /* Make sure we don't have the lock. */
975 g_topology_assert_not();
977 gv_access(p->vol_sc->provider, -1, -1, 0);
979 /* We're finished. */
980 G_VINUM_DEBUG(1, "parity operation on %s finished", p->name);
982 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
986 /* Send down next. It will determine if we need to itself. */
987 gv_parity_request(p, flags, p->synced);
991 * Handle a finished plex rebuild bio.
994 gv_rebuild_complete(struct gv_plex *p, struct bio *bp)
1001 error = bp->bio_error;
1002 flags = bp->bio_pflags;
1003 offset = bp->bio_offset;
1004 flags &= ~GV_BIO_MALLOC;
1006 KASSERT(sc != NULL, ("gv_rebuild_complete: NULL sc"));
1008 /* Clean up what we allocated. */
1009 if (bp->bio_pflags & GV_BIO_MALLOC)
1010 g_free(bp->bio_data);
1014 g_topology_assert_not();
1016 gv_access(p->vol_sc->provider, -1, -1, 0);
1017 g_topology_unlock();
1019 G_VINUM_DEBUG(0, "rebuild of %s failed at offset %jd errno: %d",
1020 p->name, (intmax_t)offset, error);
1021 p->flags &= ~GV_PLEX_REBUILDING;
1023 gv_plex_flush(p); /* Flush out remaining rebuild BIOs. */
1027 offset += (p->stripesize * (gv_sdcount(p, 1) - 1));
1028 if (offset >= p->size) {
1029 /* We're finished. */
1030 g_topology_assert_not();
1032 gv_access(p->vol_sc->provider, -1, -1, 0);
1033 g_topology_unlock();
1035 G_VINUM_DEBUG(1, "rebuild of %s finished", p->name);
1036 gv_save_config(p->vinumconf);
1037 p->flags &= ~GV_PLEX_REBUILDING;
1039 /* Try to up all subdisks. */
1040 LIST_FOREACH(s, &p->subdisks, in_plex)
1041 gv_update_sd_state(s);
1042 gv_post_event(sc, GV_EVENT_SAVE_CONFIG, sc, NULL, 0, 0);
1043 gv_plex_flush(p); /* Flush out remaining rebuild BIOs. */
1047 /* Send down next. It will determine if we need to itself. */
1048 gv_parity_request(p, flags, offset);