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 static g_attrchanged_t vdev_geom_attrchanged;
46 struct g_class zfs_vdev_class = {
49 .attrchanged = vdev_geom_attrchanged,
52 struct consumer_vdev_elem {
53 SLIST_ENTRY(consumer_vdev_elem) elems;
57 SLIST_HEAD(consumer_priv_t, consumer_vdev_elem);
58 _Static_assert(sizeof(((struct g_consumer*)NULL)->private)
59 == sizeof(struct consumer_priv_t*),
60 "consumer_priv_t* can't be stored in g_consumer.private");
62 DECLARE_GEOM_CLASS(zfs_vdev_class, zfs_vdev);
64 SYSCTL_DECL(_vfs_zfs_vdev);
65 /* Don't send BIO_FLUSH. */
66 static int vdev_geom_bio_flush_disable = 0;
67 TUNABLE_INT("vfs.zfs.vdev.bio_flush_disable", &vdev_geom_bio_flush_disable);
68 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, bio_flush_disable, CTLFLAG_RW,
69 &vdev_geom_bio_flush_disable, 0, "Disable BIO_FLUSH");
70 /* Don't send BIO_DELETE. */
71 static int vdev_geom_bio_delete_disable = 0;
72 TUNABLE_INT("vfs.zfs.vdev.bio_delete_disable", &vdev_geom_bio_delete_disable);
73 SYSCTL_INT(_vfs_zfs_vdev, OID_AUTO, bio_delete_disable, CTLFLAG_RW,
74 &vdev_geom_bio_delete_disable, 0, "Disable BIO_DELETE");
76 /* Declare local functions */
77 static void vdev_geom_detach(struct g_consumer *cp, boolean_t open_for_read);
80 * Thread local storage used to indicate when a thread is probing geoms
81 * for their guids. If NULL, this thread is not tasting geoms. If non NULL,
82 * it is looking for a replacement for the vdev_t* that is its value.
84 uint_t zfs_geom_probe_vdev_key;
87 vdev_geom_set_rotation_rate(vdev_t *vd, struct g_consumer *cp)
92 error = g_getattr("GEOM::rotation_rate", cp, &rate);
94 vd->vdev_rotation_rate = rate;
96 vd->vdev_rotation_rate = VDEV_RATE_UNKNOWN;
100 vdev_geom_set_physpath(vdev_t *vd, struct g_consumer *cp,
101 boolean_t do_null_update)
103 boolean_t needs_update = B_FALSE;
105 int error, physpath_len;
107 physpath_len = MAXPATHLEN;
108 physpath = g_malloc(physpath_len, M_WAITOK|M_ZERO);
109 error = g_io_getattr("GEOM::physpath", cp, &physpath_len, physpath);
113 /* g_topology lock ensures that vdev has not been closed */
115 old_physpath = vd->vdev_physpath;
116 vd->vdev_physpath = spa_strdup(physpath);
118 if (old_physpath != NULL) {
119 needs_update = (strcmp(old_physpath,
120 vd->vdev_physpath) != 0);
121 spa_strfree(old_physpath);
123 needs_update = do_null_update;
128 * If the physical path changed, update the config.
129 * Only request an update for previously unset physpaths if
130 * requested by the caller.
133 spa_async_request(vd->vdev_spa, SPA_ASYNC_CONFIG_UPDATE);
138 vdev_geom_attrchanged(struct g_consumer *cp, const char *attr)
141 struct consumer_priv_t *priv;
142 struct consumer_vdev_elem *elem;
145 priv = (struct consumer_priv_t*)&cp->private;
146 if (SLIST_EMPTY(priv))
149 SLIST_FOREACH(elem, priv, elems) {
150 vdev_t *vd = elem->vd;
151 if (strcmp(attr, "GEOM::rotation_rate") == 0) {
152 vdev_geom_set_rotation_rate(vd, cp);
155 if (strcmp(attr, "GEOM::physpath") == 0) {
156 vdev_geom_set_physpath(vd, cp, /*null_update*/B_TRUE);
163 vdev_geom_orphan(struct g_consumer *cp)
165 struct consumer_priv_t *priv;
166 struct consumer_vdev_elem *elem;
170 priv = (struct consumer_priv_t*)&cp->private;
171 if (SLIST_EMPTY(priv))
172 /* Vdev close in progress. Ignore the event. */
176 * Orphan callbacks occur from the GEOM event thread.
177 * Concurrent with this call, new I/O requests may be
178 * working their way through GEOM about to find out
179 * (only once executed by the g_down thread) that we've
180 * been orphaned from our disk provider. These I/Os
181 * must be retired before we can detach our consumer.
182 * This is most easily achieved by acquiring the
183 * SPA ZIO configuration lock as a writer, but doing
184 * so with the GEOM topology lock held would cause
185 * a lock order reversal. Instead, rely on the SPA's
186 * async removal support to invoke a close on this
187 * vdev once it is safe to do so.
189 SLIST_FOREACH(elem, priv, elems) {
190 vdev_t *vd = elem->vd;
192 vd->vdev_remove_wanted = B_TRUE;
193 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
197 static struct g_consumer *
198 vdev_geom_attach(struct g_provider *pp, vdev_t *vd, boolean_t sanity)
201 struct g_consumer *cp;
206 ZFS_LOG(1, "Attaching to %s.", pp->name);
209 if (pp->sectorsize > VDEV_PAD_SIZE || !ISP2(pp->sectorsize)) {
210 ZFS_LOG(1, "Failing attach of %s. "
211 "Incompatible sectorsize %d\n",
212 pp->name, pp->sectorsize);
214 } else if (pp->mediasize < SPA_MINDEVSIZE) {
215 ZFS_LOG(1, "Failing attach of %s. "
216 "Incompatible mediasize %ju\n",
217 pp->name, pp->mediasize);
222 /* Do we have geom already? No? Create one. */
223 LIST_FOREACH(gp, &zfs_vdev_class.geom, geom) {
224 if (gp->flags & G_GEOM_WITHER)
226 if (strcmp(gp->name, "zfs::vdev") != 0)
231 gp = g_new_geomf(&zfs_vdev_class, "zfs::vdev");
232 gp->orphan = vdev_geom_orphan;
233 gp->attrchanged = vdev_geom_attrchanged;
234 cp = g_new_consumer(gp);
235 error = g_attach(cp, pp);
237 ZFS_LOG(1, "%s(%d): g_attach failed: %d\n", __func__,
239 vdev_geom_detach(cp, B_FALSE);
242 error = g_access(cp, 1, 0, 1);
244 ZFS_LOG(1, "%s(%d): g_access failed: %d", __func__,
246 vdev_geom_detach(cp, B_FALSE);
249 ZFS_LOG(1, "Created geom and consumer for %s.", pp->name);
251 /* Check if we are already connected to this provider. */
252 LIST_FOREACH(cp, &gp->consumer, consumer) {
253 if (cp->provider == pp) {
254 ZFS_LOG(1, "Found consumer for %s.", pp->name);
259 cp = g_new_consumer(gp);
260 error = g_attach(cp, pp);
262 ZFS_LOG(1, "%s(%d): g_attach failed: %d\n",
263 __func__, __LINE__, error);
264 vdev_geom_detach(cp, B_FALSE);
267 error = g_access(cp, 1, 0, 1);
269 ZFS_LOG(1, "%s(%d): g_access failed: %d\n",
270 __func__, __LINE__, error);
271 vdev_geom_detach(cp, B_FALSE);
274 ZFS_LOG(1, "Created consumer for %s.", pp->name);
276 error = g_access(cp, 1, 0, 1);
278 ZFS_LOG(1, "%s(%d): g_access failed: %d\n",
279 __func__, __LINE__, error);
282 ZFS_LOG(1, "Used existing consumer for %s.", pp->name);
289 cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE;
294 vdev_geom_detach(struct g_consumer *cp, boolean_t open_for_read)
300 ZFS_LOG(1, "Detaching from %s.",
301 cp->provider && cp->provider->name ? cp->provider->name : "NULL");
305 g_access(cp, -1, 0, -1);
306 /* Destroy consumer on last close. */
307 if (cp->acr == 0 && cp->ace == 0) {
309 g_access(cp, 0, -cp->acw, 0);
310 if (cp->provider != NULL) {
311 ZFS_LOG(1, "Destroying consumer for %s.",
312 cp->provider->name ? cp->provider->name : "NULL");
315 g_destroy_consumer(cp);
317 /* Destroy geom if there are no consumers left. */
318 if (LIST_EMPTY(&gp->consumer)) {
319 ZFS_LOG(1, "Destroyed geom %s.", gp->name);
320 g_wither_geom(gp, ENXIO);
325 vdev_geom_close_locked(vdev_t *vd)
327 struct g_consumer *cp;
328 struct consumer_priv_t *priv;
329 struct consumer_vdev_elem *elem, *elem_temp;
334 vd->vdev_delayed_close = B_FALSE;
338 ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
339 KASSERT(cp->private != NULL, ("%s: cp->private is NULL", __func__));
340 priv = (struct consumer_priv_t*)&cp->private;
342 SLIST_FOREACH_SAFE(elem, priv, elems, elem_temp) {
343 if (elem->vd == vd) {
344 SLIST_REMOVE(priv, elem, consumer_vdev_elem, elems);
349 vdev_geom_detach(cp, B_TRUE);
353 * Issue one or more bios to the vdev in parallel
354 * cmds, datas, offsets, errors, and sizes are arrays of length ncmds. Each IO
355 * operation is described by parallel entries from each array. There may be
356 * more bios actually issued than entries in the array
359 vdev_geom_io(struct g_consumer *cp, int *cmds, void **datas, off_t *offsets,
360 off_t *sizes, int *errors, int ncmds)
364 off_t off, maxio, s, end;
368 maxio = MAXPHYS - (MAXPHYS % cp->provider->sectorsize);
371 /* How many bios are required for all commands ? */
372 for (i = 0; i < ncmds; i++)
373 n_bios += (sizes[i] + maxio - 1) / maxio;
375 /* Allocate memory for the bios */
376 bios_size = n_bios * sizeof(struct bio*);
377 bios = kmem_zalloc(bios_size, KM_SLEEP);
379 /* Prepare and issue all of the bios */
380 for (i = j = 0; i < ncmds; i++) {
385 ASSERT((off % cp->provider->sectorsize) == 0);
386 ASSERT((s % cp->provider->sectorsize) == 0);
388 for (; off < end; off += maxio, p += maxio, s -= maxio, j++) {
389 bios[j] = g_alloc_bio();
390 bios[j]->bio_cmd = cmds[i];
391 bios[j]->bio_done = NULL;
392 bios[j]->bio_offset = off;
393 bios[j]->bio_length = MIN(s, maxio);
394 bios[j]->bio_data = p;
395 g_io_request(bios[j], cp);
400 /* Wait for all of the bios to complete, and clean them up */
401 for (i = j = 0; i < ncmds; i++) {
406 for (; off < end; off += maxio, s -= maxio, j++) {
407 errors[i] = biowait(bios[j], "vdev_geom_io") || errors[i];
408 g_destroy_bio(bios[j]);
411 kmem_free(bios, bios_size);
415 * Read the vdev config from a device. Return the number of valid labels that
416 * were found. The vdev config will be returned in config if and only if at
417 * least one valid label was found.
420 vdev_geom_read_config(struct g_consumer *cp, nvlist_t **configp)
422 struct g_provider *pp;
424 vdev_phys_t *vdev_lists[VDEV_LABELS];
427 uint64_t psize, state, txg;
428 off_t offsets[VDEV_LABELS];
430 off_t sizes[VDEV_LABELS];
431 int cmds[VDEV_LABELS];
432 int errors[VDEV_LABELS];
435 g_topology_assert_not();
438 ZFS_LOG(1, "Reading config from %s...", pp->name);
440 psize = pp->mediasize;
441 psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
443 size = sizeof(*vdev_lists[0]) + pp->sectorsize -
444 ((sizeof(*vdev_lists[0]) - 1) % pp->sectorsize) - 1;
446 buflen = sizeof(vdev_lists[0]->vp_nvlist);
448 /* Create all of the IO requests */
449 for (l = 0; l < VDEV_LABELS; l++) {
451 vdev_lists[l] = kmem_alloc(size, KM_SLEEP);
452 offsets[l] = vdev_label_offset(psize, l, 0) + VDEV_SKIP_SIZE;
455 ASSERT(offsets[l] % pp->sectorsize == 0);
458 /* Issue the IO requests */
459 vdev_geom_io(cp, cmds, (void**)vdev_lists, offsets, sizes, errors,
462 /* Parse the labels */
463 config = *configp = NULL;
465 for (l = 0; l < VDEV_LABELS; l++) {
469 buf = vdev_lists[l]->vp_nvlist;
471 if (nvlist_unpack(buf, buflen, &config, 0) != 0)
474 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
475 &state) != 0 || state > POOL_STATE_L2CACHE) {
480 if (state != POOL_STATE_SPARE &&
481 state != POOL_STATE_L2CACHE &&
482 (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
483 &txg) != 0 || txg == 0)) {
488 if (*configp != NULL)
489 nvlist_free(*configp);
495 /* Free the label storage */
496 for (l = 0; l < VDEV_LABELS; l++)
497 kmem_free(vdev_lists[l], size);
503 resize_configs(nvlist_t ***configs, uint64_t *count, uint64_t id)
505 nvlist_t **new_configs;
510 new_configs = kmem_zalloc((id + 1) * sizeof(nvlist_t *),
512 for (i = 0; i < *count; i++)
513 new_configs[i] = (*configs)[i];
514 if (*configs != NULL)
515 kmem_free(*configs, *count * sizeof(void *));
516 *configs = new_configs;
521 process_vdev_config(nvlist_t ***configs, uint64_t *count, nvlist_t *cfg,
522 const char *name, uint64_t* known_pool_guid)
526 uint64_t vdev_guid, known_guid;
527 uint64_t id, txg, known_txg;
531 if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &pname) != 0 ||
532 strcmp(pname, name) != 0)
535 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &pool_guid) != 0)
538 if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_TOP_GUID, &vdev_guid) != 0)
541 if (nvlist_lookup_nvlist(cfg, ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0)
544 if (nvlist_lookup_uint64(vdev_tree, ZPOOL_CONFIG_ID, &id) != 0)
547 VERIFY(nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_TXG, &txg) == 0);
549 if (*known_pool_guid != 0) {
550 if (pool_guid != *known_pool_guid)
553 *known_pool_guid = pool_guid;
555 resize_configs(configs, count, id);
557 if ((*configs)[id] != NULL) {
558 VERIFY(nvlist_lookup_uint64((*configs)[id],
559 ZPOOL_CONFIG_POOL_TXG, &known_txg) == 0);
560 if (txg <= known_txg)
562 nvlist_free((*configs)[id]);
565 (*configs)[id] = cfg;
573 vdev_geom_read_pool_label(const char *name,
574 nvlist_t ***configs, uint64_t *count)
578 struct g_provider *pp;
579 struct g_consumer *zcp;
590 LIST_FOREACH(mp, &g_classes, class) {
591 if (mp == &zfs_vdev_class)
593 LIST_FOREACH(gp, &mp->geom, geom) {
594 if (gp->flags & G_GEOM_WITHER)
596 LIST_FOREACH(pp, &gp->provider, provider) {
597 if (pp->flags & G_PF_WITHER)
599 zcp = vdev_geom_attach(pp, NULL, B_TRUE);
603 nlabels = vdev_geom_read_config(zcp, &vdev_cfg);
605 vdev_geom_detach(zcp, B_TRUE);
608 ZFS_LOG(1, "successfully read vdev config");
610 process_vdev_config(configs, count,
611 vdev_cfg, name, &pool_guid);
618 return (*count > 0 ? 0 : ENOENT);
622 NO_MATCH = 0, /* No matching labels found */
623 TOPGUID_MATCH = 1, /* Labels match top guid, not vdev guid*/
624 ZERO_MATCH = 1, /* Should never be returned */
625 ONE_MATCH = 2, /* 1 label matching the vdev_guid */
626 TWO_MATCH = 3, /* 2 label matching the vdev_guid */
627 THREE_MATCH = 4, /* 3 label matching the vdev_guid */
628 FULL_MATCH = 5 /* all labels match the vdev_guid */
632 vdev_attach_ok(vdev_t *vd, struct g_provider *pp)
635 uint64_t pool_guid, top_guid, vdev_guid;
636 struct g_consumer *cp;
639 cp = vdev_geom_attach(pp, NULL, B_TRUE);
641 ZFS_LOG(1, "Unable to attach tasting instance to %s.",
646 nlabels = vdev_geom_read_config(cp, &config);
648 vdev_geom_detach(cp, B_TRUE);
650 ZFS_LOG(1, "Unable to read config from %s.", pp->name);
655 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID, &pool_guid);
657 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_TOP_GUID, &top_guid);
659 (void) nvlist_lookup_uint64(config, ZPOOL_CONFIG_GUID, &vdev_guid);
663 * Check that the label's pool guid matches the desired guid.
664 * Inactive spares and L2ARCs do not have any pool guid in the label.
666 if (pool_guid != 0 && pool_guid != spa_guid(vd->vdev_spa)) {
667 ZFS_LOG(1, "pool guid mismatch for provider %s: %ju != %ju.",
669 (uintmax_t)spa_guid(vd->vdev_spa), (uintmax_t)pool_guid);
674 * Check that the label's vdev guid matches the desired guid.
675 * The second condition handles possible race on vdev detach, when
676 * remaining vdev receives GUID of destroyed top level mirror vdev.
678 if (vdev_guid == vd->vdev_guid) {
679 ZFS_LOG(1, "guids match for provider %s.", pp->name);
680 return (ZERO_MATCH + nlabels);
681 } else if (top_guid == vd->vdev_guid && vd == vd->vdev_top) {
682 ZFS_LOG(1, "top vdev guid match for provider %s.", pp->name);
683 return (TOPGUID_MATCH);
685 ZFS_LOG(1, "vdev guid mismatch for provider %s: %ju != %ju.",
686 pp->name, (uintmax_t)vd->vdev_guid, (uintmax_t)vdev_guid);
690 static struct g_consumer *
691 vdev_geom_attach_by_guids(vdev_t *vd)
695 struct g_provider *pp, *best_pp;
696 struct g_consumer *cp;
697 enum match match, best_match;
703 best_match = NO_MATCH;
704 LIST_FOREACH(mp, &g_classes, class) {
705 if (mp == &zfs_vdev_class)
707 LIST_FOREACH(gp, &mp->geom, geom) {
708 if (gp->flags & G_GEOM_WITHER)
710 LIST_FOREACH(pp, &gp->provider, provider) {
711 match = vdev_attach_ok(vd, pp);
712 if (match > best_match) {
716 if (match == FULL_MATCH)
724 cp = vdev_geom_attach(best_pp, vd, B_TRUE);
726 printf("ZFS WARNING: Unable to attach to %s.\n",
733 static struct g_consumer *
734 vdev_geom_open_by_guids(vdev_t *vd)
736 struct g_consumer *cp;
742 ZFS_LOG(1, "Searching by guids [%ju:%ju].",
743 (uintmax_t)spa_guid(vd->vdev_spa), (uintmax_t)vd->vdev_guid);
744 cp = vdev_geom_attach_by_guids(vd);
746 len = strlen(cp->provider->name) + strlen("/dev/") + 1;
747 buf = kmem_alloc(len, KM_SLEEP);
749 snprintf(buf, len, "/dev/%s", cp->provider->name);
750 spa_strfree(vd->vdev_path);
753 ZFS_LOG(1, "Attach by guid [%ju:%ju] succeeded, provider %s.",
754 (uintmax_t)spa_guid(vd->vdev_spa),
755 (uintmax_t)vd->vdev_guid, cp->provider->name);
757 ZFS_LOG(1, "Search by guid [%ju:%ju] failed.",
758 (uintmax_t)spa_guid(vd->vdev_spa),
759 (uintmax_t)vd->vdev_guid);
765 static struct g_consumer *
766 vdev_geom_open_by_path(vdev_t *vd, int check_guid)
768 struct g_provider *pp;
769 struct g_consumer *cp;
774 pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
776 ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
777 if (!check_guid || vdev_attach_ok(vd, pp) == FULL_MATCH)
778 cp = vdev_geom_attach(pp, vd, B_FALSE);
785 vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
786 uint64_t *logical_ashift, uint64_t *physical_ashift)
788 struct g_provider *pp;
789 struct g_consumer *cp;
793 /* Set the TLS to indicate downstack that we should not access zvols*/
794 VERIFY(tsd_set(zfs_geom_probe_vdev_key, vd) == 0);
797 * We must have a pathname, and it must be absolute.
799 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
800 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
805 * Reopen the device if it's not currently open. Otherwise,
806 * just update the physical size of the device.
808 if ((cp = vd->vdev_tsd) != NULL) {
809 ASSERT(vd->vdev_reopening);
817 if (vd->vdev_spa->spa_splitting_newspa ||
818 (vd->vdev_prevstate == VDEV_STATE_UNKNOWN &&
819 vd->vdev_spa->spa_load_state == SPA_LOAD_NONE ||
820 vd->vdev_spa->spa_load_state == SPA_LOAD_CREATE)) {
822 * We are dealing with a vdev that hasn't been previously
823 * opened (since boot), and we are not loading an
824 * existing pool configuration. This looks like a
825 * vdev add operation to a new or existing pool.
826 * Assume the user knows what he/she is doing and find
827 * GEOM provider by its name, ignoring GUID mismatches.
829 * XXPOLICY: It would be safer to only allow a device
830 * that is unlabeled or labeled but missing
831 * GUID information to be opened in this fashion,
832 * unless we are doing a split, in which case we
833 * should allow any guid.
835 cp = vdev_geom_open_by_path(vd, 0);
838 * Try using the recorded path for this device, but only
839 * accept it if its label data contains the expected GUIDs.
841 cp = vdev_geom_open_by_path(vd, 1);
844 * The device at vd->vdev_path doesn't have the
845 * expected GUIDs. The disks might have merely
846 * moved around so try all other GEOM providers
847 * to find one with the right GUIDs.
849 cp = vdev_geom_open_by_guids(vd);
853 /* Clear the TLS now that tasting is done */
854 VERIFY(tsd_set(zfs_geom_probe_vdev_key, NULL) == 0);
857 ZFS_LOG(1, "Vdev %s not found.", vd->vdev_path);
860 struct consumer_priv_t *priv;
861 struct consumer_vdev_elem *elem;
864 priv = (struct consumer_priv_t*)&cp->private;
865 if (cp->private == NULL)
867 elem = g_malloc(sizeof(*elem), M_WAITOK|M_ZERO);
869 SLIST_INSERT_HEAD(priv, elem, elems);
871 spamode = spa_mode(vd->vdev_spa);
872 if (cp->provider->sectorsize > VDEV_PAD_SIZE ||
873 !ISP2(cp->provider->sectorsize)) {
874 ZFS_LOG(1, "Provider %s has unsupported sectorsize.",
877 vdev_geom_close_locked(vd);
880 } else if (cp->acw == 0 && (spamode & FWRITE) != 0) {
883 for (i = 0; i < 5; i++) {
884 error = g_access(cp, 0, 1, 0);
888 tsleep(vd, 0, "vdev", hz / 2);
892 printf("ZFS WARNING: Unable to open %s for writing (error=%d).\n",
893 cp->provider->name, error);
894 vdev_geom_close_locked(vd);
900 /* Fetch initial physical path information for this device. */
902 vdev_geom_attrchanged(cp, "GEOM::physpath");
904 /* Set other GEOM characteristics */
905 vdev_geom_set_physpath(vd, cp, /*do_null_update*/B_FALSE);
906 vdev_geom_set_rotation_rate(vd, cp);
912 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
919 * Determine the actual size of the device.
921 *max_psize = *psize = pp->mediasize;
924 * Determine the device's minimum transfer size and preferred
927 *logical_ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
928 *physical_ashift = 0;
929 if (pp->stripesize > (1 << *logical_ashift) && ISP2(pp->stripesize) &&
930 pp->stripesize <= (1 << SPA_MAXASHIFT) && pp->stripeoffset == 0)
931 *physical_ashift = highbit(pp->stripesize) - 1;
934 * Clear the nowritecache settings, so that on a vdev_reopen()
937 vd->vdev_nowritecache = B_FALSE;
943 vdev_geom_close(vdev_t *vd)
945 struct g_consumer *cp;
952 if (!vd->vdev_reopening ||
953 (cp != NULL && ((cp->flags & G_CF_ORPHAN) != 0 ||
954 (cp->provider != NULL && cp->provider->error != 0))))
955 vdev_geom_close_locked(vd);
962 vdev_geom_io_intr(struct bio *bp)
967 zio = bp->bio_caller1;
969 zio->io_error = bp->bio_error;
970 if (zio->io_error == 0 && bp->bio_resid != 0)
971 zio->io_error = SET_ERROR(EIO);
973 switch(zio->io_error) {
976 * If we get ENOTSUP for BIO_FLUSH or BIO_DELETE we know
977 * that future attempts will never succeed. In this case
978 * we set a persistent flag so that we don't bother with
979 * requests in the future.
981 switch(bp->bio_cmd) {
983 vd->vdev_nowritecache = B_TRUE;
986 vd->vdev_notrim = B_TRUE;
991 if (!vd->vdev_remove_wanted) {
993 * If provider's error is set we assume it is being
996 if (bp->bio_to->error != 0) {
997 vd->vdev_remove_wanted = B_TRUE;
998 spa_async_request(zio->io_spa,
1000 } else if (!vd->vdev_delayed_close) {
1001 vd->vdev_delayed_close = B_TRUE;
1007 zio_delay_interrupt(zio);
1011 vdev_geom_io_start(zio_t *zio)
1014 struct g_consumer *cp;
1020 switch (zio->io_type) {
1021 case ZIO_TYPE_IOCTL:
1023 if (!vdev_readable(vd)) {
1024 zio->io_error = SET_ERROR(ENXIO);
1028 switch (zio->io_cmd) {
1029 case DKIOCFLUSHWRITECACHE:
1030 if (zfs_nocacheflush || vdev_geom_bio_flush_disable)
1032 if (vd->vdev_nowritecache) {
1033 zio->io_error = SET_ERROR(ENOTSUP);
1038 zio->io_error = SET_ERROR(ENOTSUP);
1045 if (vd->vdev_notrim) {
1046 zio->io_error = SET_ERROR(ENOTSUP);
1047 } else if (!vdev_geom_bio_delete_disable) {
1054 ASSERT(zio->io_type == ZIO_TYPE_READ ||
1055 zio->io_type == ZIO_TYPE_WRITE ||
1056 zio->io_type == ZIO_TYPE_FREE ||
1057 zio->io_type == ZIO_TYPE_IOCTL);
1061 zio->io_error = SET_ERROR(ENXIO);
1066 bp->bio_caller1 = zio;
1067 switch (zio->io_type) {
1069 case ZIO_TYPE_WRITE:
1070 zio->io_target_timestamp = zio_handle_io_delay(zio);
1071 bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
1072 bp->bio_data = zio->io_data;
1073 bp->bio_offset = zio->io_offset;
1074 bp->bio_length = zio->io_size;
1077 bp->bio_cmd = BIO_DELETE;
1078 bp->bio_data = NULL;
1079 bp->bio_offset = zio->io_offset;
1080 bp->bio_length = zio->io_size;
1082 case ZIO_TYPE_IOCTL:
1083 bp->bio_cmd = BIO_FLUSH;
1084 bp->bio_flags |= BIO_ORDERED;
1085 bp->bio_data = NULL;
1086 bp->bio_offset = cp->provider->mediasize;
1090 bp->bio_done = vdev_geom_io_intr;
1092 g_io_request(bp, cp);
1096 vdev_geom_io_done(zio_t *zio)
1101 vdev_geom_hold(vdev_t *vd)
1106 vdev_geom_rele(vdev_t *vd)
1110 vdev_ops_t vdev_geom_ops = {
1119 VDEV_TYPE_DISK, /* name of this vdev type */
1120 B_TRUE /* leaf vdev */