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) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright 2013 Joyent, Inc. All rights reserved.
27 #include <sys/zfs_context.h>
28 #include <sys/spa_impl.h>
29 #include <sys/refcount.h>
30 #include <sys/vdev_disk.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/fs/zfs.h>
34 #include <sys/sunldi.h>
35 #include <sys/efi_partition.h>
36 #include <sys/fm/fs/zfs.h>
39 * Virtual device vector for disks.
42 extern ldi_ident_t zfs_li;
44 typedef struct vdev_disk_buf {
50 vdev_disk_hold(vdev_t *vd)
55 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
58 * We must have a pathname, and it must be absolute.
60 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
64 * Only prefetch path and devid info if the device has
67 if (vd->vdev_tsd != NULL)
70 if (vd->vdev_wholedisk == -1ULL) {
71 size_t len = strlen(vd->vdev_path) + 3;
72 char *buf = kmem_alloc(len, KM_SLEEP);
74 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
76 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
80 if (vd->vdev_name_vp == NULL)
81 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
83 if (vd->vdev_devid != NULL &&
84 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
85 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
86 ddi_devid_str_free(minor);
87 ddi_devid_free(devid);
92 vdev_disk_rele(vdev_t *vd)
94 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
96 if (vd->vdev_name_vp) {
97 VN_RELE_ASYNC(vd->vdev_name_vp,
98 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
99 vd->vdev_name_vp = NULL;
101 if (vd->vdev_devid_vp) {
102 VN_RELE_ASYNC(vd->vdev_devid_vp,
103 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
104 vd->vdev_devid_vp = NULL;
109 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
111 ASSERT(vd->vdev_wholedisk);
113 vdev_disk_t *dvd = vd->vdev_tsd;
116 uint64_t avail_space = 0;
117 int efisize = EFI_LABEL_SIZE * 2;
119 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
121 dk_ioc.dki_length = efisize;
122 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
123 efi = dk_ioc.dki_data;
125 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
126 FKIOCTL, kcred, NULL) == 0) {
127 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
129 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
130 vd->vdev_path, capacity, efi_altern_lba);
131 if (capacity > efi_altern_lba)
132 avail_space = (capacity - efi_altern_lba) * blksz;
134 kmem_free(dk_ioc.dki_data, efisize);
135 return (avail_space);
139 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
142 spa_t *spa = vd->vdev_spa;
144 struct dk_minfo_ext dkmext;
148 boolean_t validate_devid = B_FALSE;
152 * We must have a pathname, and it must be absolute.
154 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
155 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
156 return (SET_ERROR(EINVAL));
160 * Reopen the device if it's not currently open. Otherwise,
161 * just update the physical size of the device.
163 if (vd->vdev_tsd != NULL) {
164 ASSERT(vd->vdev_reopening);
169 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
172 * When opening a disk device, we want to preserve the user's original
173 * intent. We always want to open the device by the path the user gave
174 * us, even if it is one of multiple paths to the save device. But we
175 * also want to be able to survive disks being removed/recabled.
176 * Therefore the sequence of opening devices is:
178 * 1. Try opening the device by path. For legacy pools without the
179 * 'whole_disk' property, attempt to fix the path by appending 's0'.
181 * 2. If the devid of the device matches the stored value, return
184 * 3. Otherwise, the device may have moved. Try opening the device
185 * by the devid instead.
187 if (vd->vdev_devid != NULL) {
188 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
189 &dvd->vd_minor) != 0) {
190 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
191 return (SET_ERROR(EINVAL));
195 error = EINVAL; /* presume failure */
197 if (vd->vdev_path != NULL) {
199 if (vd->vdev_wholedisk == -1ULL) {
200 size_t len = strlen(vd->vdev_path) + 3;
201 char *buf = kmem_alloc(len, KM_SLEEP);
204 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
206 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
208 spa_strfree(vd->vdev_path);
210 vd->vdev_wholedisk = 1ULL;
211 (void) ldi_close(lh, spa_mode(spa), kcred);
217 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
218 &dvd->vd_lh, zfs_li);
221 * Compare the devid to the stored value.
223 if (error == 0 && vd->vdev_devid != NULL &&
224 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
225 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
226 error = SET_ERROR(EINVAL);
227 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
231 ddi_devid_free(devid);
235 * If we succeeded in opening the device, but 'vdev_wholedisk'
236 * is not yet set, then this must be a slice.
238 if (error == 0 && vd->vdev_wholedisk == -1ULL)
239 vd->vdev_wholedisk = 0;
243 * If we were unable to open by path, or the devid check fails, open by
246 if (error != 0 && vd->vdev_devid != NULL) {
247 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
248 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
252 * If all else fails, then try opening by physical path (if available)
253 * or the logical path (if we failed due to the devid check). While not
254 * as reliable as the devid, this will give us something, and the higher
255 * level vdev validation will prevent us from opening the wrong device.
258 if (vd->vdev_devid != NULL)
259 validate_devid = B_TRUE;
261 if (vd->vdev_physpath != NULL &&
262 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
263 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
264 kcred, &dvd->vd_lh, zfs_li);
267 * Note that we don't support the legacy auto-wholedisk support
268 * as above. This hasn't been used in a very long time and we
269 * don't need to propagate its oddities to this edge condition.
271 if (error && vd->vdev_path != NULL)
272 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
273 kcred, &dvd->vd_lh, zfs_li);
277 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
282 * Now that the device has been successfully opened, update the devid
285 if (validate_devid && spa_writeable(spa) &&
286 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
287 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
290 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
291 zfs_dbgmsg("vdev %s: update devid from %s, "
292 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
293 spa_strfree(vd->vdev_devid);
294 vd->vdev_devid = spa_strdup(vd_devid);
295 ddi_devid_str_free(vd_devid);
297 ddi_devid_free(devid);
301 * Once a device is opened, verify that the physical device path (if
302 * available) is up to date.
304 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
305 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
306 char *physpath, *minorname;
308 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
310 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
311 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
312 (vd->vdev_physpath == NULL ||
313 strcmp(vd->vdev_physpath, physpath) != 0)) {
314 if (vd->vdev_physpath)
315 spa_strfree(vd->vdev_physpath);
316 (void) strlcat(physpath, ":", MAXPATHLEN);
317 (void) strlcat(physpath, minorname, MAXPATHLEN);
318 vd->vdev_physpath = spa_strdup(physpath);
321 kmem_free(minorname, strlen(minorname) + 1);
322 kmem_free(physpath, MAXPATHLEN);
327 * Determine the actual size of the device.
329 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
330 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
331 return (SET_ERROR(EINVAL));
335 * Determine the device's minimum transfer size.
336 * If the ioctl isn't supported, assume DEV_BSIZE.
338 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
339 FKIOCTL, kcred, NULL) != 0)
340 dkmext.dki_pbsize = DEV_BSIZE;
342 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
344 if (vd->vdev_wholedisk == 1) {
345 uint64_t capacity = dkmext.dki_capacity - 1;
346 uint64_t blksz = dkmext.dki_lbsize;
350 * If we own the whole disk, try to enable disk write caching.
351 * We ignore errors because it's OK if we can't do it.
353 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
354 FKIOCTL, kcred, NULL);
356 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
357 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
358 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
364 * Clear the nowritecache bit, so that on a vdev_reopen() we will
367 vd->vdev_nowritecache = B_FALSE;
373 vdev_disk_close(vdev_t *vd)
375 vdev_disk_t *dvd = vd->vdev_tsd;
377 if (vd->vdev_reopening || dvd == NULL)
380 if (dvd->vd_minor != NULL)
381 ddi_devid_str_free(dvd->vd_minor);
383 if (dvd->vd_devid != NULL)
384 ddi_devid_free(dvd->vd_devid);
386 if (dvd->vd_lh != NULL)
387 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
389 vd->vdev_delayed_close = B_FALSE;
390 kmem_free(dvd, sizeof (vdev_disk_t));
395 vdev_disk_physio(vdev_t *vd, caddr_t data,
396 size_t size, uint64_t offset, int flags, boolean_t isdump)
398 vdev_disk_t *dvd = vd->vdev_tsd;
400 ASSERT(vd->vdev_ops == &vdev_disk_ops);
403 * If in the context of an active crash dump, use the ldi_dump(9F)
404 * call instead of ldi_strategy(9F) as usual.
407 ASSERT3P(dvd, !=, NULL);
408 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
412 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
416 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
417 size_t size, uint64_t offset, int flags)
423 return (SET_ERROR(EINVAL));
425 ASSERT(flags & B_READ || flags & B_WRITE);
427 bp = getrbuf(KM_SLEEP);
428 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
430 bp->b_un.b_addr = (void *)data;
431 bp->b_lblkno = lbtodb(offset);
432 bp->b_bufsize = size;
434 error = ldi_strategy(vd_lh, bp);
436 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
437 error = SET_ERROR(EIO);
444 vdev_disk_io_intr(buf_t *bp)
446 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
447 zio_t *zio = vdb->vdb_io;
450 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
451 * Rather than teach the rest of the stack about other error
452 * possibilities (EFAULT, etc), we normalize the error value here.
454 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
456 if (zio->io_error == 0 && bp->b_resid != 0)
457 zio->io_error = SET_ERROR(EIO);
459 kmem_free(vdb, sizeof (vdev_disk_buf_t));
465 vdev_disk_ioctl_free(zio_t *zio)
467 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
470 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
471 vdev_disk_ioctl_free,
472 zio_vsd_default_cksum_report
476 vdev_disk_ioctl_done(void *zio_arg, int error)
478 zio_t *zio = zio_arg;
480 zio->io_error = error;
486 vdev_disk_io_start(zio_t *zio)
488 vdev_t *vd = zio->io_vd;
489 vdev_disk_t *dvd = vd->vdev_tsd;
490 vdev_disk_buf_t *vdb;
491 struct dk_callback *dkc;
495 if (zio->io_type == ZIO_TYPE_IOCTL) {
497 if (!vdev_readable(vd)) {
498 zio->io_error = SET_ERROR(ENXIO);
499 return (ZIO_PIPELINE_CONTINUE);
502 switch (zio->io_cmd) {
504 case DKIOCFLUSHWRITECACHE:
506 if (zfs_nocacheflush)
509 if (vd->vdev_nowritecache) {
510 zio->io_error = SET_ERROR(ENOTSUP);
514 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
515 zio->io_vsd_ops = &vdev_disk_vsd_ops;
517 dkc->dkc_callback = vdev_disk_ioctl_done;
518 dkc->dkc_flag = FLUSH_VOLATILE;
519 dkc->dkc_cookie = zio;
521 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
522 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
526 * The ioctl will be done asychronously,
527 * and will call vdev_disk_ioctl_done()
530 return (ZIO_PIPELINE_STOP);
533 if (error == ENOTSUP || error == ENOTTY) {
535 * If we get ENOTSUP or ENOTTY, we know that
536 * no future attempts will ever succeed.
537 * In this case we set a persistent bit so
538 * that we don't bother with the ioctl in the
541 vd->vdev_nowritecache = B_TRUE;
543 zio->io_error = error;
548 zio->io_error = SET_ERROR(ENOTSUP);
551 return (ZIO_PIPELINE_CONTINUE);
554 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
560 bp->b_flags = B_BUSY | B_NOCACHE |
561 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
562 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
563 bp->b_flags |= B_FAILFAST;
564 bp->b_bcount = zio->io_size;
565 bp->b_un.b_addr = zio->io_data;
566 bp->b_lblkno = lbtodb(zio->io_offset);
567 bp->b_bufsize = zio->io_size;
568 bp->b_iodone = (int (*)())vdev_disk_io_intr;
570 /* ldi_strategy() will return non-zero only on programming errors */
571 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
573 return (ZIO_PIPELINE_STOP);
577 vdev_disk_io_done(zio_t *zio)
579 vdev_t *vd = zio->io_vd;
582 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
583 * the device has been removed. If this is the case, then we trigger an
584 * asynchronous removal of the device. Otherwise, probe the device and
585 * make sure it's still accessible.
587 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
588 vdev_disk_t *dvd = vd->vdev_tsd;
589 int state = DKIO_NONE;
591 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
592 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
594 * We post the resource as soon as possible, instead of
595 * when the async removal actually happens, because the
596 * DE is using this information to discard previous I/O
599 zfs_post_remove(zio->io_spa, vd);
600 vd->vdev_remove_wanted = B_TRUE;
601 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
602 } else if (!vd->vdev_delayed_close) {
603 vd->vdev_delayed_close = B_TRUE;
608 vdev_ops_t vdev_disk_ops = {
617 VDEV_TYPE_DISK, /* name of this vdev type */
618 B_TRUE /* leaf vdev */
622 * Given the root disk device devid or pathname, read the label from
623 * the device, and construct a configuration nvlist.
626 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
632 ddi_devid_t tmpdevid;
637 * Read the device label and build the nvlist.
639 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
641 error = ldi_open_by_devid(tmpdevid, minor_name,
642 FREAD, kcred, &vd_lh, zfs_li);
643 ddi_devid_free(tmpdevid);
644 ddi_devid_str_free(minor_name);
647 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
651 if (ldi_get_size(vd_lh, &s)) {
652 (void) ldi_close(vd_lh, FREAD, kcred);
653 return (SET_ERROR(EIO));
656 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
657 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
660 for (l = 0; l < VDEV_LABELS; l++) {
661 uint64_t offset, state, txg = 0;
663 /* read vdev label */
664 offset = vdev_label_offset(size, l, 0);
665 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
666 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
669 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
670 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
675 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
676 &state) != 0 || state >= POOL_STATE_DESTROYED) {
677 nvlist_free(*config);
682 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
683 &txg) != 0 || txg == 0) {
684 nvlist_free(*config);
692 kmem_free(label, sizeof (vdev_label_t));
693 (void) ldi_close(vd_lh, FREAD, kcred);
695 error = SET_ERROR(EIDRM);