4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2006 Pawel Jakub Dawidek <pjd@FreeBSD.org>
23 * All rights reserved.
25 * Portions Copyright (c) 2012 Martin Matuska <mm@FreeBSD.org>
28 #include <sys/zfs_context.h>
29 #include <sys/param.h>
30 #include <sys/kernel.h>
34 #include <sys/spa_impl.h>
35 #include <sys/vdev_impl.h>
36 #include <sys/fs/zfs.h>
38 #include <geom/geom.h>
39 #include <geom/geom_int.h>
42 * Virtual device vector for GEOM.
45 struct g_class zfs_vdev_class = {
50 DECLARE_GEOM_CLASS(zfs_vdev_class, zfs_vdev);
52 SYSCTL_DECL(_vfs_zfs_vdev);
53 /* Don't send BIO_FLUSH. */
54 static int vdev_geom_bio_flush_disable = 0;
55 TUNABLE_INT("vfs.zfs.vdev.bio_flush_disable", &vdev_geom_bio_flush_disable);
56 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, bio_flush_disable, CTLFLAG_RW,
57 &vdev_geom_bio_flush_disable, 0, "Disable BIO_FLUSH");
58 /* Don't send BIO_DELETE. */
59 static int vdev_geom_bio_delete_disable = 0;
60 TUNABLE_INT("vfs.zfs.vdev.bio_delete_disable", &vdev_geom_bio_delete_disable);
61 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, bio_delete_disable, CTLFLAG_RW,
62 &vdev_geom_bio_delete_disable, 0, "Disable BIO_DELETE");
65 vdev_geom_orphan(struct g_consumer *cp)
74 * Orphan callbacks occur from the GEOM event thread.
75 * Concurrent with this call, new I/O requests may be
76 * working their way through GEOM about to find out
77 * (only once executed by the g_down thread) that we've
78 * been orphaned from our disk provider. These I/Os
79 * must be retired before we can detach our consumer.
80 * This is most easily achieved by acquiring the
81 * SPA ZIO configuration lock as a writer, but doing
82 * so with the GEOM topology lock held would cause
83 * a lock order reversal. Instead, rely on the SPA's
84 * async removal support to invoke a close on this
85 * vdev once it is safe to do so.
87 zfs_post_remove(vd->vdev_spa, vd);
88 vd->vdev_remove_wanted = B_TRUE;
89 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
92 static struct g_consumer *
93 vdev_geom_attach(struct g_provider *pp)
96 struct g_consumer *cp;
100 ZFS_LOG(1, "Attaching to %s.", pp->name);
101 /* Do we have geom already? No? Create one. */
102 LIST_FOREACH(gp, &zfs_vdev_class.geom, geom) {
103 if (gp->flags & G_GEOM_WITHER)
105 if (strcmp(gp->name, "zfs::vdev") != 0)
110 gp = g_new_geomf(&zfs_vdev_class, "zfs::vdev");
111 gp->orphan = vdev_geom_orphan;
112 cp = g_new_consumer(gp);
113 if (g_attach(cp, pp) != 0) {
114 g_wither_geom(gp, ENXIO);
117 if (g_access(cp, 1, 0, 1) != 0) {
118 g_wither_geom(gp, ENXIO);
121 ZFS_LOG(1, "Created geom and consumer for %s.", pp->name);
123 /* Check if we are already connected to this provider. */
124 LIST_FOREACH(cp, &gp->consumer, consumer) {
125 if (cp->provider == pp) {
126 ZFS_LOG(1, "Found consumer for %s.", pp->name);
131 cp = g_new_consumer(gp);
132 if (g_attach(cp, pp) != 0) {
133 g_destroy_consumer(cp);
136 if (g_access(cp, 1, 0, 1) != 0) {
138 g_destroy_consumer(cp);
141 ZFS_LOG(1, "Created consumer for %s.", pp->name);
143 if (g_access(cp, 1, 0, 1) != 0)
145 ZFS_LOG(1, "Used existing consumer for %s.", pp->name);
152 vdev_geom_detach(void *arg, int flag __unused)
155 struct g_consumer *cp;
161 ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
162 g_access(cp, -1, 0, -1);
163 /* Destroy consumer on last close. */
164 if (cp->acr == 0 && cp->ace == 0) {
165 ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
167 g_access(cp, 0, -cp->acw, 0);
169 g_destroy_consumer(cp);
171 /* Destroy geom if there are no consumers left. */
172 if (LIST_EMPTY(&gp->consumer)) {
173 ZFS_LOG(1, "Destroyed geom %s.", gp->name);
174 g_wither_geom(gp, ENXIO);
179 nvlist_get_guid(nvlist_t *list)
184 nvlist_lookup_uint64(list, ZPOOL_CONFIG_GUID, &value);
189 vdev_geom_io(struct g_consumer *cp, int cmd, void *data, off_t offset, off_t size)
196 ASSERT((offset % cp->provider->sectorsize) == 0);
197 ASSERT((size % cp->provider->sectorsize) == 0);
203 maxio = MAXPHYS - (MAXPHYS % cp->provider->sectorsize);
206 for (; off < offset; off += maxio, p += maxio, size -= maxio) {
207 bzero(bp, sizeof(*bp));
210 bp->bio_offset = off;
211 bp->bio_length = MIN(size, maxio);
213 g_io_request(bp, cp);
214 error = biowait(bp, "vdev_geom_io");
224 vdev_geom_taste_orphan(struct g_consumer *cp)
227 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
228 cp->provider->name));
232 vdev_geom_read_config(struct g_consumer *cp, nvlist_t **config)
234 struct g_provider *pp;
240 uint64_t guid, state, txg;
243 g_topology_assert_not();
246 ZFS_LOG(1, "Reading config from %s...", pp->name);
248 psize = pp->mediasize;
249 psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
251 size = sizeof(*label) + pp->sectorsize -
252 ((sizeof(*label) - 1) % pp->sectorsize) - 1;
255 label = kmem_alloc(size, KM_SLEEP);
256 buflen = sizeof(label->vl_vdev_phys.vp_nvlist);
259 for (l = 0; l < VDEV_LABELS; l++) {
261 offset = vdev_label_offset(psize, l, 0);
262 if ((offset % pp->sectorsize) != 0)
265 if (vdev_geom_io(cp, BIO_READ, label, offset, size) != 0)
267 buf = label->vl_vdev_phys.vp_nvlist;
269 if (nvlist_unpack(buf, buflen, config, 0) != 0)
272 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
273 &state) != 0 || state == POOL_STATE_DESTROYED ||
274 state > POOL_STATE_L2CACHE) {
275 nvlist_free(*config);
280 if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
281 (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
282 &txg) != 0 || txg == 0)) {
283 nvlist_free(*config);
291 kmem_free(label, size);
292 return (*config == NULL ? ENOENT : 0);
296 resize_configs(nvlist_t ***configs, uint64_t *count, uint64_t id)
298 nvlist_t **new_configs;
303 new_configs = kmem_alloc((id + 1) * sizeof(nvlist_t *),
305 for (i = 0; i < *count; i++)
306 new_configs[i] = (*configs)[i];
307 if (*configs != NULL)
308 kmem_free(*configs, *count * sizeof(void *));
309 *configs = new_configs;
314 process_vdev_config(nvlist_t ***configs, uint64_t *count, nvlist_t *cfg,
315 const char *name, uint64_t* known_pool_guid)
319 uint64_t vdev_guid, known_guid;
320 uint64_t id, txg, known_txg;
324 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &pname) != 0 ||
325 strcmp(pname, name) != 0)
328 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &pool_guid) != 0)
331 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_TOP_GUID, &vdev_guid) != 0)
334 if (nvlist_lookup_nvlist(cfg, ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0)
337 if (nvlist_lookup_uint64(vdev_tree, ZPOOL_CONFIG_ID, &id) != 0)
340 VERIFY(nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
342 if (*known_pool_guid != 0) {
343 if (pool_guid != *known_pool_guid)
346 *known_pool_guid = pool_guid;
348 resize_configs(configs, count, id);
350 if ((*configs)[id] != NULL) {
351 VERIFY(nvlist_lookup_uint64((*configs)[id],
352 ZPOOL_CONFIG_POOL_TXG, &known_txg) == 0);
353 if (txg <= known_txg)
355 nvlist_free((*configs)[id]);
358 (*configs)[id] = cfg;
366 vdev_geom_attach_taster(struct g_consumer *cp, struct g_provider *pp)
370 if (pp->flags & G_PF_WITHER)
372 if (pp->sectorsize > VDEV_PAD_SIZE || !ISP2(pp->sectorsize))
375 error = g_access(cp, 1, 0, 0);
382 vdev_geom_detach_taster(struct g_consumer *cp)
384 g_access(cp, -1, 0, 0);
389 vdev_geom_read_pool_label(const char *name,
390 nvlist_t ***configs, uint64_t *count)
393 struct g_geom *gp, *zgp;
394 struct g_provider *pp;
395 struct g_consumer *zcp;
403 zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
404 /* This orphan function should be never called. */
405 zgp->orphan = vdev_geom_taste_orphan;
406 zcp = g_new_consumer(zgp);
411 LIST_FOREACH(mp, &g_classes, class) {
412 if (mp == &zfs_vdev_class)
414 LIST_FOREACH(gp, &mp->geom, geom) {
415 if (gp->flags & G_GEOM_WITHER)
417 LIST_FOREACH(pp, &gp->provider, provider) {
418 if (pp->flags & G_PF_WITHER)
420 if (vdev_geom_attach_taster(zcp, pp) != 0)
423 error = vdev_geom_read_config(zcp, &vdev_cfg);
425 vdev_geom_detach_taster(zcp);
428 ZFS_LOG(1, "successfully read vdev config");
430 process_vdev_config(configs, count,
431 vdev_cfg, name, &pool_guid);
436 g_destroy_consumer(zcp);
441 return (*count > 0 ? 0 : ENOENT);
445 vdev_geom_read_guid(struct g_consumer *cp)
450 g_topology_assert_not();
453 if (vdev_geom_read_config(cp, &config) == 0) {
454 guid = nvlist_get_guid(config);
460 static struct g_consumer *
461 vdev_geom_attach_by_guid(uint64_t guid)
464 struct g_geom *gp, *zgp;
465 struct g_provider *pp;
466 struct g_consumer *cp, *zcp;
471 zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
472 /* This orphan function should be never called. */
473 zgp->orphan = vdev_geom_taste_orphan;
474 zcp = g_new_consumer(zgp);
477 LIST_FOREACH(mp, &g_classes, class) {
478 if (mp == &zfs_vdev_class)
480 LIST_FOREACH(gp, &mp->geom, geom) {
481 if (gp->flags & G_GEOM_WITHER)
483 LIST_FOREACH(pp, &gp->provider, provider) {
484 if (vdev_geom_attach_taster(zcp, pp) != 0)
487 pguid = vdev_geom_read_guid(zcp);
489 vdev_geom_detach_taster(zcp);
492 cp = vdev_geom_attach(pp);
494 printf("ZFS WARNING: Unable to attach to %s.\n",
507 g_destroy_consumer(zcp);
512 static struct g_consumer *
513 vdev_geom_open_by_guid(vdev_t *vd)
515 struct g_consumer *cp;
521 ZFS_LOG(1, "Searching by guid [%ju].", (uintmax_t)vd->vdev_guid);
522 cp = vdev_geom_attach_by_guid(vd->vdev_guid);
524 len = strlen(cp->provider->name) + strlen("/dev/") + 1;
525 buf = kmem_alloc(len, KM_SLEEP);
527 snprintf(buf, len, "/dev/%s", cp->provider->name);
528 spa_strfree(vd->vdev_path);
531 ZFS_LOG(1, "Attach by guid [%ju] succeeded, provider %s.",
532 (uintmax_t)vd->vdev_guid, vd->vdev_path);
534 ZFS_LOG(1, "Search by guid [%ju] failed.",
535 (uintmax_t)vd->vdev_guid);
541 static struct g_consumer *
542 vdev_geom_open_by_path(vdev_t *vd, int check_guid)
544 struct g_provider *pp;
545 struct g_consumer *cp;
551 pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
553 ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
554 cp = vdev_geom_attach(pp);
555 if (cp != NULL && check_guid && ISP2(pp->sectorsize) &&
556 pp->sectorsize <= VDEV_PAD_SIZE) {
558 guid = vdev_geom_read_guid(cp);
560 if (guid != vd->vdev_guid) {
561 vdev_geom_detach(cp, 0);
563 ZFS_LOG(1, "guid mismatch for provider %s: "
564 "%ju != %ju.", vd->vdev_path,
565 (uintmax_t)vd->vdev_guid, (uintmax_t)guid);
567 ZFS_LOG(1, "guid match for provider %s.",
577 vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
580 struct g_provider *pp;
581 struct g_consumer *cp;
586 * We must have a pathname, and it must be absolute.
588 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
589 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
600 * If we're creating or splitting a pool, just find the GEOM provider
601 * by its name and ignore GUID mismatches.
603 if (vd->vdev_spa->spa_load_state == SPA_LOAD_NONE ||
604 vd->vdev_spa->spa_splitting_newspa == B_TRUE)
605 cp = vdev_geom_open_by_path(vd, 0);
607 cp = vdev_geom_open_by_path(vd, 1);
610 * The device at vd->vdev_path doesn't have the
611 * expected guid. The disks might have merely
612 * moved around so try all other GEOM providers
613 * to find one with the right guid.
615 cp = vdev_geom_open_by_guid(vd);
620 ZFS_LOG(1, "Provider %s not found.", vd->vdev_path);
622 } else if (cp->provider->sectorsize > VDEV_PAD_SIZE ||
623 !ISP2(cp->provider->sectorsize)) {
624 ZFS_LOG(1, "Provider %s has unsupported sectorsize.",
626 vdev_geom_detach(cp, 0);
629 } else if (cp->acw == 0 && (spa_mode(vd->vdev_spa) & FWRITE) != 0) {
632 for (i = 0; i < 5; i++) {
633 error = g_access(cp, 0, 1, 0);
637 tsleep(vd, 0, "vdev", hz / 2);
641 printf("ZFS WARNING: Unable to open %s for writing (error=%d).\n",
642 vd->vdev_path, error);
643 vdev_geom_detach(cp, 0);
650 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
659 * Determine the actual size of the device.
661 *max_psize = *psize = pp->mediasize;
664 * Determine the device's minimum transfer size.
666 *ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
669 * Clear the nowritecache settings, so that on a vdev_reopen()
672 vd->vdev_nowritecache = B_FALSE;
674 if (vd->vdev_physpath != NULL)
675 spa_strfree(vd->vdev_physpath);
676 bufsize = sizeof("/dev/") + strlen(pp->name);
677 vd->vdev_physpath = kmem_alloc(bufsize, KM_SLEEP);
678 snprintf(vd->vdev_physpath, bufsize, "/dev/%s", pp->name);
684 vdev_geom_close(vdev_t *vd)
686 struct g_consumer *cp;
692 vd->vdev_delayed_close = B_FALSE;
693 g_post_event(vdev_geom_detach, cp, M_WAITOK, NULL);
697 vdev_geom_io_intr(struct bio *bp)
702 zio = bp->bio_caller1;
704 zio->io_error = bp->bio_error;
705 if (zio->io_error == 0 && bp->bio_resid != 0)
707 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == ENOTSUP) {
709 * If we get ENOTSUP, we know that no future
710 * attempts will ever succeed. In this case we
711 * set a persistent bit so that we don't bother
712 * with the ioctl in the future.
714 vd->vdev_nowritecache = B_TRUE;
716 if (bp->bio_cmd == BIO_DELETE && bp->bio_error == ENOTSUP) {
718 * If we get ENOTSUP, we know that no future
719 * attempts will ever succeed. In this case we
720 * set a persistent bit so that we don't bother
721 * with the ioctl in the future.
723 vd->vdev_notrim = B_TRUE;
725 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
727 * If provider's error is set we assume it is being
730 if (bp->bio_to->error != 0) {
732 * We post the resource as soon as possible, instead of
733 * when the async removal actually happens, because the
734 * DE is using this information to discard previous I/O
737 /* XXX: zfs_post_remove() can sleep. */
738 zfs_post_remove(zio->io_spa, vd);
739 vd->vdev_remove_wanted = B_TRUE;
740 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
741 } else if (!vd->vdev_delayed_close) {
742 vd->vdev_delayed_close = B_TRUE;
750 vdev_geom_io_start(zio_t *zio)
753 struct g_consumer *cp;
759 if (zio->io_type == ZIO_TYPE_IOCTL) {
761 if (!vdev_readable(vd)) {
762 zio->io_error = ENXIO;
763 return (ZIO_PIPELINE_CONTINUE);
766 switch (zio->io_cmd) {
767 case DKIOCFLUSHWRITECACHE:
768 if (zfs_nocacheflush || vdev_geom_bio_flush_disable)
770 if (vd->vdev_nowritecache) {
771 zio->io_error = ENOTSUP;
776 if (vdev_geom_bio_delete_disable)
778 if (vd->vdev_notrim) {
779 zio->io_error = ENOTSUP;
784 zio->io_error = ENOTSUP;
787 return (ZIO_PIPELINE_CONTINUE);
792 zio->io_error = ENXIO;
793 return (ZIO_PIPELINE_CONTINUE);
796 bp->bio_caller1 = zio;
797 switch (zio->io_type) {
800 bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
801 bp->bio_data = zio->io_data;
802 bp->bio_offset = zio->io_offset;
803 bp->bio_length = zio->io_size;
806 switch (zio->io_cmd) {
807 case DKIOCFLUSHWRITECACHE:
808 bp->bio_cmd = BIO_FLUSH;
809 bp->bio_flags |= BIO_ORDERED;
811 bp->bio_offset = cp->provider->mediasize;
815 bp->bio_cmd = BIO_DELETE;
817 bp->bio_offset = zio->io_offset;
818 bp->bio_length = zio->io_size;
823 bp->bio_done = vdev_geom_io_intr;
825 g_io_request(bp, cp);
827 return (ZIO_PIPELINE_STOP);
831 vdev_geom_io_done(zio_t *zio)
836 vdev_geom_hold(vdev_t *vd)
841 vdev_geom_rele(vdev_t *vd)
845 vdev_ops_t vdev_geom_ops = {
854 VDEV_TYPE_DISK, /* name of this vdev type */
855 B_TRUE /* leaf vdev */