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)
181 nvpair_t *elem = NULL;
184 while ((elem = nvlist_next_nvpair(list, elem)) != NULL) {
185 if (nvpair_type(elem) == DATA_TYPE_UINT64 &&
186 strcmp(nvpair_name(elem), "guid") == 0) {
187 VERIFY(nvpair_value_uint64(elem, &value) == 0);
195 vdev_geom_io(struct g_consumer *cp, int cmd, void *data, off_t offset, off_t size)
202 ASSERT((offset % cp->provider->sectorsize) == 0);
203 ASSERT((size % cp->provider->sectorsize) == 0);
209 maxio = MAXPHYS - (MAXPHYS % cp->provider->sectorsize);
212 for (; off < offset; off += maxio, p += maxio, size -= maxio) {
213 bzero(bp, sizeof(*bp));
216 bp->bio_offset = off;
217 bp->bio_length = MIN(size, maxio);
219 g_io_request(bp, cp);
220 error = biowait(bp, "vdev_geom_io");
230 vdev_geom_read_guid(struct g_consumer *cp)
232 struct g_provider *pp;
241 g_topology_assert_not();
244 ZFS_LOG(1, "Reading guid from %s...", pp->name);
246 psize = pp->mediasize;
247 psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
249 size = sizeof(*label) + pp->sectorsize -
250 ((sizeof(*label) - 1) % pp->sectorsize) - 1;
253 label = kmem_alloc(size, KM_SLEEP);
254 buflen = sizeof(label->vl_vdev_phys.vp_nvlist);
256 for (l = 0; l < VDEV_LABELS; l++) {
257 nvlist_t *config = NULL;
259 offset = vdev_label_offset(psize, l, 0);
260 if ((offset % pp->sectorsize) != 0)
263 if (vdev_geom_io(cp, BIO_READ, label, offset, size) != 0)
265 buf = label->vl_vdev_phys.vp_nvlist;
267 if (nvlist_unpack(buf, buflen, &config, 0) != 0)
270 guid = nvlist_get_guid(config);
276 kmem_free(label, size);
278 ZFS_LOG(1, "guid for %s is %ju", pp->name, (uintmax_t)guid);
283 vdev_geom_taste_orphan(struct g_consumer *cp)
286 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
287 cp->provider->name));
290 static struct g_consumer *
291 vdev_geom_attach_by_guid(uint64_t guid)
294 struct g_geom *gp, *zgp;
295 struct g_provider *pp;
296 struct g_consumer *cp, *zcp;
301 zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
302 /* This orphan function should be never called. */
303 zgp->orphan = vdev_geom_taste_orphan;
304 zcp = g_new_consumer(zgp);
307 LIST_FOREACH(mp, &g_classes, class) {
308 if (mp == &zfs_vdev_class)
310 LIST_FOREACH(gp, &mp->geom, geom) {
311 if (gp->flags & G_GEOM_WITHER)
313 LIST_FOREACH(pp, &gp->provider, provider) {
314 if (pp->flags & G_PF_WITHER)
317 if (g_access(zcp, 1, 0, 0) != 0) {
322 pguid = vdev_geom_read_guid(zcp);
324 g_access(zcp, -1, 0, 0);
328 cp = vdev_geom_attach(pp);
330 printf("ZFS WARNING: Unable to attach to %s.\n",
343 g_destroy_consumer(zcp);
348 static struct g_consumer *
349 vdev_geom_open_by_guid(vdev_t *vd)
351 struct g_consumer *cp;
357 ZFS_LOG(1, "Searching by guid [%ju].", (uintmax_t)vd->vdev_guid);
358 cp = vdev_geom_attach_by_guid(vd->vdev_guid);
360 len = strlen(cp->provider->name) + strlen("/dev/") + 1;
361 buf = kmem_alloc(len, KM_SLEEP);
363 snprintf(buf, len, "/dev/%s", cp->provider->name);
364 spa_strfree(vd->vdev_path);
367 ZFS_LOG(1, "Attach by guid [%ju] succeeded, provider %s.",
368 (uintmax_t)vd->vdev_guid, vd->vdev_path);
370 ZFS_LOG(1, "Search by guid [%ju] failed.",
371 (uintmax_t)vd->vdev_guid);
377 static struct g_consumer *
378 vdev_geom_open_by_path(vdev_t *vd, int check_guid)
380 struct g_provider *pp;
381 struct g_consumer *cp;
387 pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
389 ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
390 cp = vdev_geom_attach(pp);
391 if (cp != NULL && check_guid && ISP2(pp->sectorsize) &&
392 pp->sectorsize <= VDEV_PAD_SIZE) {
394 guid = vdev_geom_read_guid(cp);
396 if (guid != vd->vdev_guid) {
397 vdev_geom_detach(cp, 0);
399 ZFS_LOG(1, "guid mismatch for provider %s: "
400 "%ju != %ju.", vd->vdev_path,
401 (uintmax_t)vd->vdev_guid, (uintmax_t)guid);
403 ZFS_LOG(1, "guid match for provider %s.",
413 vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
416 struct g_provider *pp;
417 struct g_consumer *cp;
422 * We must have a pathname, and it must be absolute.
424 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
425 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
436 * If we're creating or splitting a pool, just find the GEOM provider
437 * by its name and ignore GUID mismatches.
439 if (vd->vdev_spa->spa_load_state == SPA_LOAD_NONE ||
440 vd->vdev_spa->spa_splitting_newspa == B_TRUE)
441 cp = vdev_geom_open_by_path(vd, 0);
443 cp = vdev_geom_open_by_path(vd, 1);
446 * The device at vd->vdev_path doesn't have the
447 * expected guid. The disks might have merely
448 * moved around so try all other GEOM providers
449 * to find one with the right guid.
451 cp = vdev_geom_open_by_guid(vd);
456 ZFS_LOG(1, "Provider %s not found.", vd->vdev_path);
458 } else if (cp->provider->sectorsize > VDEV_PAD_SIZE ||
459 !ISP2(cp->provider->sectorsize)) {
460 ZFS_LOG(1, "Provider %s has unsupported sectorsize.",
462 vdev_geom_detach(cp, 0);
465 } else if (cp->acw == 0 && (spa_mode(vd->vdev_spa) & FWRITE) != 0) {
468 for (i = 0; i < 5; i++) {
469 error = g_access(cp, 0, 1, 0);
473 tsleep(vd, 0, "vdev", hz / 2);
477 printf("ZFS WARNING: Unable to open %s for writing (error=%d).\n",
478 vd->vdev_path, error);
479 vdev_geom_detach(cp, 0);
486 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
495 * Determine the actual size of the device.
497 *max_psize = *psize = pp->mediasize;
500 * Determine the device's minimum transfer size.
502 *ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
505 * Clear the nowritecache settings, so that on a vdev_reopen()
508 vd->vdev_nowritecache = B_FALSE;
510 if (vd->vdev_physpath != NULL)
511 spa_strfree(vd->vdev_physpath);
512 bufsize = sizeof("/dev/") + strlen(pp->name);
513 vd->vdev_physpath = kmem_alloc(bufsize, KM_SLEEP);
514 snprintf(vd->vdev_physpath, bufsize, "/dev/%s", pp->name);
520 vdev_geom_close(vdev_t *vd)
522 struct g_consumer *cp;
528 vd->vdev_delayed_close = B_FALSE;
529 g_post_event(vdev_geom_detach, cp, M_WAITOK, NULL);
533 vdev_geom_io_intr(struct bio *bp)
538 zio = bp->bio_caller1;
540 zio->io_error = bp->bio_error;
541 if (zio->io_error == 0 && bp->bio_resid != 0)
543 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == ENOTSUP) {
545 * If we get ENOTSUP, we know that no future
546 * attempts will ever succeed. In this case we
547 * set a persistent bit so that we don't bother
548 * with the ioctl in the future.
550 vd->vdev_nowritecache = B_TRUE;
552 if (bp->bio_cmd == BIO_DELETE && bp->bio_error == ENOTSUP) {
554 * If we get ENOTSUP, we know that no future
555 * attempts will ever succeed. In this case we
556 * set a persistent bit so that we don't bother
557 * with the ioctl in the future.
559 vd->vdev_notrim = B_TRUE;
561 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
563 * If provider's error is set we assume it is being
566 if (bp->bio_to->error != 0) {
568 * We post the resource as soon as possible, instead of
569 * when the async removal actually happens, because the
570 * DE is using this information to discard previous I/O
573 /* XXX: zfs_post_remove() can sleep. */
574 zfs_post_remove(zio->io_spa, vd);
575 vd->vdev_remove_wanted = B_TRUE;
576 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
577 } else if (!vd->vdev_delayed_close) {
578 vd->vdev_delayed_close = B_TRUE;
586 vdev_geom_io_start(zio_t *zio)
589 struct g_consumer *cp;
595 if (zio->io_type == ZIO_TYPE_IOCTL) {
597 if (!vdev_readable(vd)) {
598 zio->io_error = ENXIO;
599 return (ZIO_PIPELINE_CONTINUE);
602 switch (zio->io_cmd) {
603 case DKIOCFLUSHWRITECACHE:
604 if (zfs_nocacheflush || vdev_geom_bio_flush_disable)
606 if (vd->vdev_nowritecache) {
607 zio->io_error = ENOTSUP;
612 if (vdev_geom_bio_delete_disable)
614 if (vd->vdev_notrim) {
615 zio->io_error = ENOTSUP;
620 zio->io_error = ENOTSUP;
623 return (ZIO_PIPELINE_CONTINUE);
628 zio->io_error = ENXIO;
629 return (ZIO_PIPELINE_CONTINUE);
632 bp->bio_caller1 = zio;
633 switch (zio->io_type) {
636 bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
637 bp->bio_data = zio->io_data;
638 bp->bio_offset = zio->io_offset;
639 bp->bio_length = zio->io_size;
642 switch (zio->io_cmd) {
643 case DKIOCFLUSHWRITECACHE:
644 bp->bio_cmd = BIO_FLUSH;
645 bp->bio_flags |= BIO_ORDERED;
647 bp->bio_offset = cp->provider->mediasize;
651 bp->bio_cmd = BIO_DELETE;
653 bp->bio_offset = zio->io_offset;
654 bp->bio_length = zio->io_size;
659 bp->bio_done = vdev_geom_io_intr;
661 g_io_request(bp, cp);
663 return (ZIO_PIPELINE_STOP);
667 vdev_geom_io_done(zio_t *zio)
672 vdev_geom_hold(vdev_t *vd)
677 vdev_geom_rele(vdev_t *vd)
681 vdev_ops_t vdev_geom_ops = {
690 VDEV_TYPE_DISK, /* name of this vdev type */
691 B_TRUE /* leaf vdev */