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;
45 vdev_disk_hold(vdev_t *vd)
50 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
53 * We must have a pathname, and it must be absolute.
55 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
59 * Only prefetch path and devid info if the device has
62 if (vd->vdev_tsd != NULL)
65 if (vd->vdev_wholedisk == -1ULL) {
66 size_t len = strlen(vd->vdev_path) + 3;
67 char *buf = kmem_alloc(len, KM_SLEEP);
69 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
71 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
75 if (vd->vdev_name_vp == NULL)
76 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
78 if (vd->vdev_devid != NULL &&
79 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
80 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
81 ddi_devid_str_free(minor);
82 ddi_devid_free(devid);
87 vdev_disk_rele(vdev_t *vd)
89 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
91 if (vd->vdev_name_vp) {
92 VN_RELE_ASYNC(vd->vdev_name_vp,
93 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
94 vd->vdev_name_vp = NULL;
96 if (vd->vdev_devid_vp) {
97 VN_RELE_ASYNC(vd->vdev_devid_vp,
98 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
99 vd->vdev_devid_vp = NULL;
104 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
106 ASSERT(vd->vdev_wholedisk);
108 vdev_disk_t *dvd = vd->vdev_tsd;
111 uint64_t avail_space = 0;
112 int efisize = EFI_LABEL_SIZE * 2;
114 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
116 dk_ioc.dki_length = efisize;
117 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
118 efi = dk_ioc.dki_data;
120 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
121 FKIOCTL, kcred, NULL) == 0) {
122 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
124 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
125 vd->vdev_path, capacity, efi_altern_lba);
126 if (capacity > efi_altern_lba)
127 avail_space = (capacity - efi_altern_lba) * blksz;
129 kmem_free(dk_ioc.dki_data, efisize);
130 return (avail_space);
134 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
137 spa_t *spa = vd->vdev_spa;
139 struct dk_minfo_ext dkmext;
145 * We must have a pathname, and it must be absolute.
147 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
148 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
149 return (SET_ERROR(EINVAL));
153 * Reopen the device if it's not currently open. Otherwise,
154 * just update the physical size of the device.
156 if (vd->vdev_tsd != NULL) {
157 ASSERT(vd->vdev_reopening);
162 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
165 * When opening a disk device, we want to preserve the user's original
166 * intent. We always want to open the device by the path the user gave
167 * us, even if it is one of multiple paths to the save device. But we
168 * also want to be able to survive disks being removed/recabled.
169 * Therefore the sequence of opening devices is:
171 * 1. Try opening the device by path. For legacy pools without the
172 * 'whole_disk' property, attempt to fix the path by appending 's0'.
174 * 2. If the devid of the device matches the stored value, return
177 * 3. Otherwise, the device may have moved. Try opening the device
178 * by the devid instead.
180 if (vd->vdev_devid != NULL) {
181 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
182 &dvd->vd_minor) != 0) {
183 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
184 return (SET_ERROR(EINVAL));
188 error = EINVAL; /* presume failure */
190 if (vd->vdev_path != NULL) {
193 if (vd->vdev_wholedisk == -1ULL) {
194 size_t len = strlen(vd->vdev_path) + 3;
195 char *buf = kmem_alloc(len, KM_SLEEP);
198 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
200 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
202 spa_strfree(vd->vdev_path);
204 vd->vdev_wholedisk = 1ULL;
205 (void) ldi_close(lh, spa_mode(spa), kcred);
211 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
212 &dvd->vd_lh, zfs_li);
215 * Compare the devid to the stored value.
217 if (error == 0 && vd->vdev_devid != NULL &&
218 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
219 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
220 error = SET_ERROR(EINVAL);
221 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
225 ddi_devid_free(devid);
229 * If we succeeded in opening the device, but 'vdev_wholedisk'
230 * is not yet set, then this must be a slice.
232 if (error == 0 && vd->vdev_wholedisk == -1ULL)
233 vd->vdev_wholedisk = 0;
237 * If we were unable to open by path, or the devid check fails, open by
240 if (error != 0 && vd->vdev_devid != NULL)
241 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
242 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
245 * If all else fails, then try opening by physical path (if available)
246 * or the logical path (if we failed due to the devid check). While not
247 * as reliable as the devid, this will give us something, and the higher
248 * level vdev validation will prevent us from opening the wrong device.
251 if (vd->vdev_physpath != NULL &&
252 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
253 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
254 kcred, &dvd->vd_lh, zfs_li);
257 * Note that we don't support the legacy auto-wholedisk support
258 * as above. This hasn't been used in a very long time and we
259 * don't need to propagate its oddities to this edge condition.
261 if (error && vd->vdev_path != NULL)
262 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
263 kcred, &dvd->vd_lh, zfs_li);
267 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
272 * Once a device is opened, verify that the physical device path (if
273 * available) is up to date.
275 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
276 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
277 char *physpath, *minorname;
279 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
281 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
282 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
283 (vd->vdev_physpath == NULL ||
284 strcmp(vd->vdev_physpath, physpath) != 0)) {
285 if (vd->vdev_physpath)
286 spa_strfree(vd->vdev_physpath);
287 (void) strlcat(physpath, ":", MAXPATHLEN);
288 (void) strlcat(physpath, minorname, MAXPATHLEN);
289 vd->vdev_physpath = spa_strdup(physpath);
292 kmem_free(minorname, strlen(minorname) + 1);
293 kmem_free(physpath, MAXPATHLEN);
298 * Determine the actual size of the device.
300 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
301 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
302 return (SET_ERROR(EINVAL));
306 * Determine the device's minimum transfer size.
307 * If the ioctl isn't supported, assume DEV_BSIZE.
309 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
310 FKIOCTL, kcred, NULL) != 0)
311 dkmext.dki_pbsize = DEV_BSIZE;
313 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
315 if (vd->vdev_wholedisk == 1) {
316 uint64_t capacity = dkmext.dki_capacity - 1;
317 uint64_t blksz = dkmext.dki_lbsize;
321 * If we own the whole disk, try to enable disk write caching.
322 * We ignore errors because it's OK if we can't do it.
324 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
325 FKIOCTL, kcred, NULL);
327 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
328 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
329 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
335 * Clear the nowritecache bit, so that on a vdev_reopen() we will
338 vd->vdev_nowritecache = B_FALSE;
344 vdev_disk_close(vdev_t *vd)
346 vdev_disk_t *dvd = vd->vdev_tsd;
348 if (vd->vdev_reopening || dvd == NULL)
351 if (dvd->vd_minor != NULL)
352 ddi_devid_str_free(dvd->vd_minor);
354 if (dvd->vd_devid != NULL)
355 ddi_devid_free(dvd->vd_devid);
357 if (dvd->vd_lh != NULL)
358 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
360 vd->vdev_delayed_close = B_FALSE;
361 kmem_free(dvd, sizeof (vdev_disk_t));
366 vdev_disk_physio(vdev_t *vd, caddr_t data,
367 size_t size, uint64_t offset, int flags, boolean_t isdump)
369 vdev_disk_t *dvd = vd->vdev_tsd;
371 ASSERT(vd->vdev_ops == &vdev_disk_ops);
374 * If in the context of an active crash dump, use the ldi_dump(9F)
375 * call instead of ldi_strategy(9F) as usual.
378 ASSERT3P(dvd, !=, NULL);
379 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
383 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
387 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
388 size_t size, uint64_t offset, int flags)
394 return (SET_ERROR(EINVAL));
396 ASSERT(flags & B_READ || flags & B_WRITE);
398 bp = getrbuf(KM_SLEEP);
399 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
401 bp->b_un.b_addr = (void *)data;
402 bp->b_lblkno = lbtodb(offset);
403 bp->b_bufsize = size;
405 error = ldi_strategy(vd_lh, bp);
407 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
408 error = SET_ERROR(EIO);
415 vdev_disk_io_intr(buf_t *bp)
417 vdev_buf_t *vb = (vdev_buf_t *)bp;
418 zio_t *zio = vb->vb_io;
421 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
422 * Rather than teach the rest of the stack about other error
423 * possibilities (EFAULT, etc), we normalize the error value here.
425 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
427 if (zio->io_error == 0 && bp->b_resid != 0)
428 zio->io_error = SET_ERROR(EIO);
430 kmem_free(vb, sizeof (vdev_buf_t));
436 vdev_disk_ioctl_free(zio_t *zio)
438 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
441 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
442 vdev_disk_ioctl_free,
443 zio_vsd_default_cksum_report
447 vdev_disk_ioctl_done(void *zio_arg, int error)
449 zio_t *zio = zio_arg;
451 zio->io_error = error;
457 vdev_disk_io_start(zio_t *zio)
459 vdev_t *vd = zio->io_vd;
460 vdev_disk_t *dvd = vd->vdev_tsd;
462 struct dk_callback *dkc;
466 if (zio->io_type == ZIO_TYPE_IOCTL) {
468 if (!vdev_readable(vd)) {
469 zio->io_error = SET_ERROR(ENXIO);
470 return (ZIO_PIPELINE_CONTINUE);
473 switch (zio->io_cmd) {
475 case DKIOCFLUSHWRITECACHE:
477 if (zfs_nocacheflush)
480 if (vd->vdev_nowritecache) {
481 zio->io_error = SET_ERROR(ENOTSUP);
485 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
486 zio->io_vsd_ops = &vdev_disk_vsd_ops;
488 dkc->dkc_callback = vdev_disk_ioctl_done;
489 dkc->dkc_flag = FLUSH_VOLATILE;
490 dkc->dkc_cookie = zio;
492 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
493 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
497 * The ioctl will be done asychronously,
498 * and will call vdev_disk_ioctl_done()
501 return (ZIO_PIPELINE_STOP);
504 if (error == ENOTSUP || error == ENOTTY) {
506 * If we get ENOTSUP or ENOTTY, we know that
507 * no future attempts will ever succeed.
508 * In this case we set a persistent bit so
509 * that we don't bother with the ioctl in the
512 vd->vdev_nowritecache = B_TRUE;
514 zio->io_error = error;
519 zio->io_error = SET_ERROR(ENOTSUP);
522 return (ZIO_PIPELINE_CONTINUE);
525 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
531 bp->b_flags = B_BUSY | B_NOCACHE |
532 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
533 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
534 bp->b_flags |= B_FAILFAST;
535 bp->b_bcount = zio->io_size;
536 bp->b_un.b_addr = zio->io_data;
537 bp->b_lblkno = lbtodb(zio->io_offset);
538 bp->b_bufsize = zio->io_size;
539 bp->b_iodone = (int (*)())vdev_disk_io_intr;
541 /* ldi_strategy() will return non-zero only on programming errors */
542 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
544 return (ZIO_PIPELINE_STOP);
548 vdev_disk_io_done(zio_t *zio)
550 vdev_t *vd = zio->io_vd;
553 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
554 * the device has been removed. If this is the case, then we trigger an
555 * asynchronous removal of the device. Otherwise, probe the device and
556 * make sure it's still accessible.
558 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
559 vdev_disk_t *dvd = vd->vdev_tsd;
560 int state = DKIO_NONE;
562 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
563 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
565 * We post the resource as soon as possible, instead of
566 * when the async removal actually happens, because the
567 * DE is using this information to discard previous I/O
570 zfs_post_remove(zio->io_spa, vd);
571 vd->vdev_remove_wanted = B_TRUE;
572 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
573 } else if (!vd->vdev_delayed_close) {
574 vd->vdev_delayed_close = B_TRUE;
579 vdev_ops_t vdev_disk_ops = {
588 VDEV_TYPE_DISK, /* name of this vdev type */
589 B_TRUE /* leaf vdev */
593 * Given the root disk device devid or pathname, read the label from
594 * the device, and construct a configuration nvlist.
597 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
603 ddi_devid_t tmpdevid;
608 * Read the device label and build the nvlist.
610 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
612 error = ldi_open_by_devid(tmpdevid, minor_name,
613 FREAD, kcred, &vd_lh, zfs_li);
614 ddi_devid_free(tmpdevid);
615 ddi_devid_str_free(minor_name);
618 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
622 if (ldi_get_size(vd_lh, &s)) {
623 (void) ldi_close(vd_lh, FREAD, kcred);
624 return (SET_ERROR(EIO));
627 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
628 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
631 for (l = 0; l < VDEV_LABELS; l++) {
632 uint64_t offset, state, txg = 0;
634 /* read vdev label */
635 offset = vdev_label_offset(size, l, 0);
636 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
637 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
640 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
641 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
646 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
647 &state) != 0 || state >= POOL_STATE_DESTROYED) {
648 nvlist_free(*config);
653 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
654 &txg) != 0 || txg == 0) {
655 nvlist_free(*config);
663 kmem_free(label, sizeof (vdev_label_t));
664 (void) ldi_close(vd_lh, FREAD, kcred);
666 error = SET_ERROR(EIDRM);