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;
143 boolean_t validate_devid = B_FALSE;
147 * We must have a pathname, and it must be absolute.
149 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
150 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
151 return (SET_ERROR(EINVAL));
155 * Reopen the device if it's not currently open. Otherwise,
156 * just update the physical size of the device.
158 if (vd->vdev_tsd != NULL) {
159 ASSERT(vd->vdev_reopening);
164 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
167 * When opening a disk device, we want to preserve the user's original
168 * intent. We always want to open the device by the path the user gave
169 * us, even if it is one of multiple paths to the save device. But we
170 * also want to be able to survive disks being removed/recabled.
171 * Therefore the sequence of opening devices is:
173 * 1. Try opening the device by path. For legacy pools without the
174 * 'whole_disk' property, attempt to fix the path by appending 's0'.
176 * 2. If the devid of the device matches the stored value, return
179 * 3. Otherwise, the device may have moved. Try opening the device
180 * by the devid instead.
182 if (vd->vdev_devid != NULL) {
183 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
184 &dvd->vd_minor) != 0) {
185 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
186 return (SET_ERROR(EINVAL));
190 error = EINVAL; /* presume failure */
192 if (vd->vdev_path != NULL) {
194 if (vd->vdev_wholedisk == -1ULL) {
195 size_t len = strlen(vd->vdev_path) + 3;
196 char *buf = kmem_alloc(len, KM_SLEEP);
199 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
201 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
203 spa_strfree(vd->vdev_path);
205 vd->vdev_wholedisk = 1ULL;
206 (void) ldi_close(lh, spa_mode(spa), kcred);
212 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
213 &dvd->vd_lh, zfs_li);
216 * Compare the devid to the stored value.
218 if (error == 0 && vd->vdev_devid != NULL &&
219 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
220 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
221 error = SET_ERROR(EINVAL);
222 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
226 ddi_devid_free(devid);
230 * If we succeeded in opening the device, but 'vdev_wholedisk'
231 * is not yet set, then this must be a slice.
233 if (error == 0 && vd->vdev_wholedisk == -1ULL)
234 vd->vdev_wholedisk = 0;
238 * If we were unable to open by path, or the devid check fails, open by
241 if (error != 0 && vd->vdev_devid != NULL) {
242 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
243 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
247 * If all else fails, then try opening by physical path (if available)
248 * or the logical path (if we failed due to the devid check). While not
249 * as reliable as the devid, this will give us something, and the higher
250 * level vdev validation will prevent us from opening the wrong device.
253 if (vd->vdev_devid != NULL)
254 validate_devid = B_TRUE;
256 if (vd->vdev_physpath != NULL &&
257 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
258 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
259 kcred, &dvd->vd_lh, zfs_li);
262 * Note that we don't support the legacy auto-wholedisk support
263 * as above. This hasn't been used in a very long time and we
264 * don't need to propagate its oddities to this edge condition.
266 if (error && vd->vdev_path != NULL)
267 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
268 kcred, &dvd->vd_lh, zfs_li);
272 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
277 * Now that the device has been successfully opened, update the devid
280 if (validate_devid && spa_writeable(spa) &&
281 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
282 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
285 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
286 zfs_dbgmsg("vdev %s: update devid from %s, "
287 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
288 spa_strfree(vd->vdev_devid);
289 vd->vdev_devid = spa_strdup(vd_devid);
290 ddi_devid_str_free(vd_devid);
292 ddi_devid_free(devid);
296 * Once a device is opened, verify that the physical device path (if
297 * available) is up to date.
299 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
300 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
301 char *physpath, *minorname;
303 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
305 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
306 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
307 (vd->vdev_physpath == NULL ||
308 strcmp(vd->vdev_physpath, physpath) != 0)) {
309 if (vd->vdev_physpath)
310 spa_strfree(vd->vdev_physpath);
311 (void) strlcat(physpath, ":", MAXPATHLEN);
312 (void) strlcat(physpath, minorname, MAXPATHLEN);
313 vd->vdev_physpath = spa_strdup(physpath);
316 kmem_free(minorname, strlen(minorname) + 1);
317 kmem_free(physpath, MAXPATHLEN);
322 * Determine the actual size of the device.
324 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
325 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
326 return (SET_ERROR(EINVAL));
330 * Determine the device's minimum transfer size.
331 * If the ioctl isn't supported, assume DEV_BSIZE.
333 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
334 FKIOCTL, kcred, NULL) != 0)
335 dkmext.dki_pbsize = DEV_BSIZE;
337 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
339 if (vd->vdev_wholedisk == 1) {
340 uint64_t capacity = dkmext.dki_capacity - 1;
341 uint64_t blksz = dkmext.dki_lbsize;
345 * If we own the whole disk, try to enable disk write caching.
346 * We ignore errors because it's OK if we can't do it.
348 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
349 FKIOCTL, kcred, NULL);
351 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
352 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
353 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
359 * Clear the nowritecache bit, so that on a vdev_reopen() we will
362 vd->vdev_nowritecache = B_FALSE;
368 vdev_disk_close(vdev_t *vd)
370 vdev_disk_t *dvd = vd->vdev_tsd;
372 if (vd->vdev_reopening || dvd == NULL)
375 if (dvd->vd_minor != NULL)
376 ddi_devid_str_free(dvd->vd_minor);
378 if (dvd->vd_devid != NULL)
379 ddi_devid_free(dvd->vd_devid);
381 if (dvd->vd_lh != NULL)
382 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
384 vd->vdev_delayed_close = B_FALSE;
385 kmem_free(dvd, sizeof (vdev_disk_t));
390 vdev_disk_physio(vdev_t *vd, caddr_t data,
391 size_t size, uint64_t offset, int flags, boolean_t isdump)
393 vdev_disk_t *dvd = vd->vdev_tsd;
395 ASSERT(vd->vdev_ops == &vdev_disk_ops);
398 * If in the context of an active crash dump, use the ldi_dump(9F)
399 * call instead of ldi_strategy(9F) as usual.
402 ASSERT3P(dvd, !=, NULL);
403 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
407 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
411 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
412 size_t size, uint64_t offset, int flags)
418 return (SET_ERROR(EINVAL));
420 ASSERT(flags & B_READ || flags & B_WRITE);
422 bp = getrbuf(KM_SLEEP);
423 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
425 bp->b_un.b_addr = (void *)data;
426 bp->b_lblkno = lbtodb(offset);
427 bp->b_bufsize = size;
429 error = ldi_strategy(vd_lh, bp);
431 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
432 error = SET_ERROR(EIO);
439 vdev_disk_io_intr(buf_t *bp)
441 vdev_buf_t *vb = (vdev_buf_t *)bp;
442 zio_t *zio = vb->vb_io;
445 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
446 * Rather than teach the rest of the stack about other error
447 * possibilities (EFAULT, etc), we normalize the error value here.
449 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
451 if (zio->io_error == 0 && bp->b_resid != 0)
452 zio->io_error = SET_ERROR(EIO);
454 kmem_free(vb, sizeof (vdev_buf_t));
460 vdev_disk_ioctl_free(zio_t *zio)
462 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
465 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
466 vdev_disk_ioctl_free,
467 zio_vsd_default_cksum_report
471 vdev_disk_ioctl_done(void *zio_arg, int error)
473 zio_t *zio = zio_arg;
475 zio->io_error = error;
481 vdev_disk_io_start(zio_t *zio)
483 vdev_t *vd = zio->io_vd;
484 vdev_disk_t *dvd = vd->vdev_tsd;
486 struct dk_callback *dkc;
490 if (zio->io_type == ZIO_TYPE_IOCTL) {
492 if (!vdev_readable(vd)) {
493 zio->io_error = SET_ERROR(ENXIO);
494 return (ZIO_PIPELINE_CONTINUE);
497 switch (zio->io_cmd) {
499 case DKIOCFLUSHWRITECACHE:
501 if (zfs_nocacheflush)
504 if (vd->vdev_nowritecache) {
505 zio->io_error = SET_ERROR(ENOTSUP);
509 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
510 zio->io_vsd_ops = &vdev_disk_vsd_ops;
512 dkc->dkc_callback = vdev_disk_ioctl_done;
513 dkc->dkc_flag = FLUSH_VOLATILE;
514 dkc->dkc_cookie = zio;
516 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
517 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
521 * The ioctl will be done asychronously,
522 * and will call vdev_disk_ioctl_done()
525 return (ZIO_PIPELINE_STOP);
528 if (error == ENOTSUP || error == ENOTTY) {
530 * If we get ENOTSUP or ENOTTY, we know that
531 * no future attempts will ever succeed.
532 * In this case we set a persistent bit so
533 * that we don't bother with the ioctl in the
536 vd->vdev_nowritecache = B_TRUE;
538 zio->io_error = error;
543 zio->io_error = SET_ERROR(ENOTSUP);
546 return (ZIO_PIPELINE_CONTINUE);
549 vb = kmem_alloc(sizeof (vdev_buf_t), KM_SLEEP);
555 bp->b_flags = B_BUSY | B_NOCACHE |
556 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
557 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
558 bp->b_flags |= B_FAILFAST;
559 bp->b_bcount = zio->io_size;
560 bp->b_un.b_addr = zio->io_data;
561 bp->b_lblkno = lbtodb(zio->io_offset);
562 bp->b_bufsize = zio->io_size;
563 bp->b_iodone = (int (*)())vdev_disk_io_intr;
565 /* ldi_strategy() will return non-zero only on programming errors */
566 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
568 return (ZIO_PIPELINE_STOP);
572 vdev_disk_io_done(zio_t *zio)
574 vdev_t *vd = zio->io_vd;
577 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
578 * the device has been removed. If this is the case, then we trigger an
579 * asynchronous removal of the device. Otherwise, probe the device and
580 * make sure it's still accessible.
582 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
583 vdev_disk_t *dvd = vd->vdev_tsd;
584 int state = DKIO_NONE;
586 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
587 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
589 * We post the resource as soon as possible, instead of
590 * when the async removal actually happens, because the
591 * DE is using this information to discard previous I/O
594 zfs_post_remove(zio->io_spa, vd);
595 vd->vdev_remove_wanted = B_TRUE;
596 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
597 } else if (!vd->vdev_delayed_close) {
598 vd->vdev_delayed_close = B_TRUE;
603 vdev_ops_t vdev_disk_ops = {
612 VDEV_TYPE_DISK, /* name of this vdev type */
613 B_TRUE /* leaf vdev */
617 * Given the root disk device devid or pathname, read the label from
618 * the device, and construct a configuration nvlist.
621 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
627 ddi_devid_t tmpdevid;
632 * Read the device label and build the nvlist.
634 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
636 error = ldi_open_by_devid(tmpdevid, minor_name,
637 FREAD, kcred, &vd_lh, zfs_li);
638 ddi_devid_free(tmpdevid);
639 ddi_devid_str_free(minor_name);
642 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
646 if (ldi_get_size(vd_lh, &s)) {
647 (void) ldi_close(vd_lh, FREAD, kcred);
648 return (SET_ERROR(EIO));
651 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
652 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
655 for (l = 0; l < VDEV_LABELS; l++) {
656 uint64_t offset, state, txg = 0;
658 /* read vdev label */
659 offset = vdev_label_offset(size, l, 0);
660 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
661 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
664 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
665 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
670 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
671 &state) != 0 || state >= POOL_STATE_DESTROYED) {
672 nvlist_free(*config);
677 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
678 &txg) != 0 || txg == 0) {
679 nvlist_free(*config);
687 kmem_free(label, sizeof (vdev_label_t));
688 (void) ldi_close(vd_lh, FREAD, kcred);
690 error = SET_ERROR(EIDRM);