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.
26 #include <sys/zfs_context.h>
27 #include <sys/param.h>
28 #include <sys/kernel.h>
32 #include <sys/spa_impl.h>
33 #include <sys/vdev_impl.h>
34 #include <sys/fs/zfs.h>
36 #include <geom/geom.h>
37 #include <geom/geom_int.h>
40 * Virtual device vector for GEOM.
43 struct g_class zfs_vdev_class = {
48 DECLARE_GEOM_CLASS(zfs_vdev_class, zfs_vdev);
51 vdev_geom_orphan(struct g_consumer *cp)
61 error = cp->provider->error;
63 ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
64 if (cp->acr + cp->acw + cp->ace > 0)
65 g_access(cp, -cp->acr, -cp->acw, -cp->ace);
66 ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
68 g_destroy_consumer(cp);
69 /* Destroy geom if there are no consumers left. */
70 if (LIST_EMPTY(&gp->consumer)) {
71 ZFS_LOG(1, "Destroyed geom %s.", gp->name);
72 g_wither_geom(gp, error);
75 vd->vdev_remove_wanted = B_TRUE;
76 spa_async_request(vd->vdev_spa, SPA_ASYNC_REMOVE);
79 static struct g_consumer *
80 vdev_geom_attach(struct g_provider *pp)
83 struct g_consumer *cp;
87 ZFS_LOG(1, "Attaching to %s.", pp->name);
88 /* Do we have geom already? No? Create one. */
89 LIST_FOREACH(gp, &zfs_vdev_class.geom, geom) {
90 if (gp->flags & G_GEOM_WITHER)
92 if (strcmp(gp->name, "zfs::vdev") != 0)
97 gp = g_new_geomf(&zfs_vdev_class, "zfs::vdev");
98 gp->orphan = vdev_geom_orphan;
99 cp = g_new_consumer(gp);
100 if (g_attach(cp, pp) != 0) {
101 g_wither_geom(gp, ENXIO);
104 if (g_access(cp, 1, 0, 1) != 0) {
105 g_wither_geom(gp, ENXIO);
108 ZFS_LOG(1, "Created geom and consumer for %s.", pp->name);
110 /* Check if we are already connected to this provider. */
111 LIST_FOREACH(cp, &gp->consumer, consumer) {
112 if (cp->provider == pp) {
113 ZFS_LOG(1, "Found consumer for %s.", pp->name);
118 cp = g_new_consumer(gp);
119 if (g_attach(cp, pp) != 0) {
120 g_destroy_consumer(cp);
123 if (g_access(cp, 1, 0, 1) != 0) {
125 g_destroy_consumer(cp);
128 ZFS_LOG(1, "Created consumer for %s.", pp->name);
130 if (g_access(cp, 1, 0, 1) != 0)
132 ZFS_LOG(1, "Used existing consumer for %s.", pp->name);
139 vdev_geom_detach(void *arg, int flag __unused)
142 struct g_consumer *cp;
148 ZFS_LOG(1, "Closing access to %s.", cp->provider->name);
149 g_access(cp, -1, 0, -1);
150 /* Destroy consumer on last close. */
151 if (cp->acr == 0 && cp->ace == 0) {
152 ZFS_LOG(1, "Destroyed consumer to %s.", cp->provider->name);
154 g_access(cp, 0, -cp->acw, 0);
156 g_destroy_consumer(cp);
158 /* Destroy geom if there are no consumers left. */
159 if (LIST_EMPTY(&gp->consumer)) {
160 ZFS_LOG(1, "Destroyed geom %s.", gp->name);
161 g_wither_geom(gp, ENXIO);
166 nvlist_get_guid(nvlist_t *list)
168 nvpair_t *elem = NULL;
171 while ((elem = nvlist_next_nvpair(list, elem)) != NULL) {
172 if (nvpair_type(elem) == DATA_TYPE_UINT64 &&
173 strcmp(nvpair_name(elem), "guid") == 0) {
174 VERIFY(nvpair_value_uint64(elem, &value) == 0);
182 vdev_geom_io(struct g_consumer *cp, int cmd, void *data, off_t offset, off_t size)
189 ASSERT((offset % cp->provider->sectorsize) == 0);
190 ASSERT((size % cp->provider->sectorsize) == 0);
198 for (; off < offset; off += MAXPHYS, p += MAXPHYS, size -= MAXPHYS) {
199 bzero(bp, sizeof(*bp));
202 bp->bio_offset = off;
203 bp->bio_length = MIN(size, MAXPHYS);
205 g_io_request(bp, cp);
206 error = biowait(bp, "vdev_geom_io");
216 vdev_geom_read_guid(struct g_consumer *cp)
218 struct g_provider *pp;
225 int error, l, len, iszvol;
227 g_topology_assert_not();
230 ZFS_LOG(1, "Reading guid from %s...", pp->name);
231 if (g_getattr("ZFS::iszvol", cp, &iszvol) == 0 && iszvol) {
232 ZFS_LOG(1, "Skipping ZVOL-based provider %s.", pp->name);
236 psize = pp->mediasize;
237 psize = P2ALIGN(psize, (uint64_t)sizeof(vdev_label_t));
239 size = sizeof(*label) + pp->sectorsize -
240 ((sizeof(*label) - 1) % pp->sectorsize) - 1;
243 label = kmem_alloc(size, KM_SLEEP);
244 buflen = sizeof(label->vl_vdev_phys.vp_nvlist);
246 for (l = 0; l < VDEV_LABELS; l++) {
247 nvlist_t *config = NULL;
249 offset = vdev_label_offset(psize, l, 0);
250 if ((offset % pp->sectorsize) != 0)
253 if (vdev_geom_io(cp, BIO_READ, label, offset, size) != 0)
255 buf = label->vl_vdev_phys.vp_nvlist;
257 if (nvlist_unpack(buf, buflen, &config, 0) != 0)
260 guid = nvlist_get_guid(config);
266 kmem_free(label, size);
268 ZFS_LOG(1, "guid for %s is %ju", pp->name, (uintmax_t)guid);
272 struct vdev_geom_find {
274 struct g_consumer *cp;
278 vdev_geom_taste_orphan(struct g_consumer *cp)
281 KASSERT(1 == 0, ("%s called while tasting %s.", __func__,
282 cp->provider->name));
286 vdev_geom_attach_by_guid_event(void *arg, int flags __unused)
288 struct vdev_geom_find *ap;
290 struct g_geom *gp, *zgp;
291 struct g_provider *pp;
292 struct g_consumer *zcp;
299 zgp = g_new_geomf(&zfs_vdev_class, "zfs::vdev::taste");
300 /* This orphan function should be never called. */
301 zgp->orphan = vdev_geom_taste_orphan;
302 zcp = g_new_consumer(zgp);
304 LIST_FOREACH(mp, &g_classes, class) {
305 if (mp == &zfs_vdev_class)
307 LIST_FOREACH(gp, &mp->geom, geom) {
308 if (gp->flags & G_GEOM_WITHER)
310 LIST_FOREACH(pp, &gp->provider, provider) {
311 if (pp->flags & G_PF_WITHER)
314 if (g_access(zcp, 1, 0, 0) != 0) {
319 guid = vdev_geom_read_guid(zcp);
321 g_access(zcp, -1, 0, 0);
323 if (guid != ap->guid)
325 ap->cp = vdev_geom_attach(pp);
326 if (ap->cp == NULL) {
327 printf("ZFS WARNING: Unable to attach to %s.\n",
337 g_destroy_consumer(zcp);
341 static struct g_consumer *
342 vdev_geom_attach_by_guid(uint64_t guid)
344 struct vdev_geom_find *ap;
345 struct g_consumer *cp;
347 ap = kmem_zalloc(sizeof(*ap), KM_SLEEP);
349 g_waitfor_event(vdev_geom_attach_by_guid_event, ap, M_WAITOK, NULL);
351 kmem_free(ap, sizeof(*ap));
355 static struct g_consumer *
356 vdev_geom_open_by_guid(vdev_t *vd)
358 struct g_consumer *cp;
362 ZFS_LOG(1, "Searching by guid [%ju].", (uintmax_t)vd->vdev_guid);
363 cp = vdev_geom_attach_by_guid(vd->vdev_guid);
365 len = strlen(cp->provider->name) + strlen("/dev/") + 1;
366 buf = kmem_alloc(len, KM_SLEEP);
368 snprintf(buf, len, "/dev/%s", cp->provider->name);
369 spa_strfree(vd->vdev_path);
372 ZFS_LOG(1, "Attach by guid [%ju] succeeded, provider %s.",
373 (uintmax_t)vd->vdev_guid, vd->vdev_path);
375 ZFS_LOG(1, "Search by guid [%ju] failed.",
376 (uintmax_t)vd->vdev_guid);
382 static struct g_consumer *
383 vdev_geom_open_by_path(vdev_t *vd, int check_guid)
385 struct g_provider *pp;
386 struct g_consumer *cp;
391 pp = g_provider_by_name(vd->vdev_path + sizeof("/dev/") - 1);
393 ZFS_LOG(1, "Found provider by name %s.", vd->vdev_path);
394 cp = vdev_geom_attach(pp);
395 if (cp != NULL && check_guid) {
397 guid = vdev_geom_read_guid(cp);
399 if (guid != vd->vdev_guid) {
400 vdev_geom_detach(cp, 0);
402 ZFS_LOG(1, "guid mismatch for provider %s: "
403 "%ju != %ju.", vd->vdev_path,
404 (uintmax_t)vd->vdev_guid, (uintmax_t)guid);
406 ZFS_LOG(1, "guid match for provider %s.",
417 vdev_geom_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
419 struct g_provider *pp;
420 struct g_consumer *cp;
424 * We must have a pathname, and it must be absolute.
426 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
427 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
433 if ((owned = mtx_owned(&Giant)))
438 * If we're creating pool, just find GEOM provider by its name
439 * and ignore GUID mismatches.
441 if (vd->vdev_spa->spa_load_state == SPA_LOAD_NONE)
442 cp = vdev_geom_open_by_path(vd, 0);
444 cp = vdev_geom_open_by_path(vd, 1);
447 * The device at vd->vdev_path doesn't have the
448 * expected guid. The disks might have merely
449 * moved around so try all other GEOM providers
450 * to find one with the right guid.
452 cp = vdev_geom_open_by_guid(vd);
457 ZFS_LOG(1, "Provider %s not found.", vd->vdev_path);
459 } else if (cp->acw == 0 && (spa_mode(vd->vdev_spa) & FWRITE) != 0) {
463 for (i = 0; i < 5; i++) {
464 error = g_access(cp, 0, 1, 0);
468 tsleep(vd, 0, "vdev", hz / 2);
472 printf("ZFS WARNING: Unable to open %s for writing (error=%d).\n",
473 vd->vdev_path, error);
474 vdev_geom_detach(cp, 0);
482 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
491 * Determine the actual size of the device.
493 *psize = pp->mediasize;
496 * Determine the device's minimum transfer size.
498 *ashift = highbit(MAX(pp->sectorsize, SPA_MINBLOCKSIZE)) - 1;
501 * Clear the nowritecache bit, so that on a vdev_reopen() we will
504 vd->vdev_nowritecache = B_FALSE;
510 vdev_geom_close(vdev_t *vd)
512 struct g_consumer *cp;
518 g_post_event(vdev_geom_detach, cp, M_WAITOK, NULL);
522 vdev_geom_io_intr(struct bio *bp)
526 zio = bp->bio_caller1;
527 zio->io_error = bp->bio_error;
528 if (zio->io_error == 0 && bp->bio_resid != 0)
530 if (bp->bio_cmd == BIO_FLUSH && bp->bio_error == ENOTSUP) {
534 * If we get ENOTSUP, we know that no future
535 * attempts will ever succeed. In this case we
536 * set a persistent bit so that we don't bother
537 * with the ioctl in the future.
540 vd->vdev_nowritecache = B_TRUE;
547 vdev_geom_io_start(zio_t *zio)
550 struct g_consumer *cp;
556 if (zio->io_type == ZIO_TYPE_IOCTL) {
558 if (!vdev_readable(vd)) {
559 zio->io_error = ENXIO;
560 return (ZIO_PIPELINE_CONTINUE);
563 switch (zio->io_cmd) {
565 case DKIOCFLUSHWRITECACHE:
567 if (zfs_nocacheflush)
570 if (vd->vdev_nowritecache) {
571 zio->io_error = ENOTSUP;
577 zio->io_error = ENOTSUP;
580 return (ZIO_PIPELINE_CONTINUE);
585 zio->io_error = ENXIO;
586 return (ZIO_PIPELINE_CONTINUE);
589 bp->bio_caller1 = zio;
590 switch (zio->io_type) {
593 bp->bio_cmd = zio->io_type == ZIO_TYPE_READ ? BIO_READ : BIO_WRITE;
594 bp->bio_data = zio->io_data;
595 bp->bio_offset = zio->io_offset;
596 bp->bio_length = zio->io_size;
599 bp->bio_cmd = BIO_FLUSH;
601 bp->bio_offset = cp->provider->mediasize;
605 bp->bio_done = vdev_geom_io_intr;
607 g_io_request(bp, cp);
609 return (ZIO_PIPELINE_STOP);
613 vdev_geom_io_done(zio_t *zio)
617 vdev_ops_t vdev_geom_ops = {
624 VDEV_TYPE_DISK, /* name of this vdev type */
625 B_TRUE /* leaf vdev */