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 Nexenta Systems, Inc. All rights reserved.
25 * Copyright 2013 Joyent, Inc. All rights reserved.
28 #include <sys/zfs_context.h>
29 #include <sys/spa_impl.h>
30 #include <sys/refcount.h>
31 #include <sys/vdev_disk.h>
32 #include <sys/vdev_impl.h>
33 #include <sys/fs/zfs.h>
35 #include <sys/sunldi.h>
36 #include <sys/efi_partition.h>
37 #include <sys/fm/fs/zfs.h>
40 * Virtual device vector for disks.
43 extern ldi_ident_t zfs_li;
45 typedef struct vdev_disk_buf {
51 vdev_disk_hold(vdev_t *vd)
56 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
59 * We must have a pathname, and it must be absolute.
61 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
65 * Only prefetch path and devid info if the device has
68 if (vd->vdev_tsd != NULL)
71 if (vd->vdev_wholedisk == -1ULL) {
72 size_t len = strlen(vd->vdev_path) + 3;
73 char *buf = kmem_alloc(len, KM_SLEEP);
75 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
77 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
81 if (vd->vdev_name_vp == NULL)
82 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
84 if (vd->vdev_devid != NULL &&
85 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
86 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
87 ddi_devid_str_free(minor);
88 ddi_devid_free(devid);
93 vdev_disk_rele(vdev_t *vd)
95 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
97 if (vd->vdev_name_vp) {
98 VN_RELE_ASYNC(vd->vdev_name_vp,
99 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
100 vd->vdev_name_vp = NULL;
102 if (vd->vdev_devid_vp) {
103 VN_RELE_ASYNC(vd->vdev_devid_vp,
104 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
105 vd->vdev_devid_vp = NULL;
110 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
112 ASSERT(vd->vdev_wholedisk);
114 vdev_disk_t *dvd = vd->vdev_tsd;
117 uint64_t avail_space = 0;
118 int efisize = EFI_LABEL_SIZE * 2;
120 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
122 dk_ioc.dki_length = efisize;
123 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
124 efi = dk_ioc.dki_data;
126 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
127 FKIOCTL, kcred, NULL) == 0) {
128 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
130 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
131 vd->vdev_path, capacity, efi_altern_lba);
132 if (capacity > efi_altern_lba)
133 avail_space = (capacity - efi_altern_lba) * blksz;
135 kmem_free(dk_ioc.dki_data, efisize);
136 return (avail_space);
140 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
141 * even a fallback to DKIOCGMEDIAINFO fails.
144 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
146 #define VDEV_DEBUG(...) /* Nothing... */
150 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
153 spa_t *spa = vd->vdev_spa;
156 struct dk_minfo_ext ude;
159 struct dk_minfo_ext *dkmext = &dks.ude;
160 struct dk_minfo *dkm = &dks.ud;
164 boolean_t validate_devid = B_FALSE;
166 uint64_t capacity = 0, blksz = 0, pbsize;
169 * We must have a pathname, and it must be absolute.
171 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
172 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
173 return (SET_ERROR(EINVAL));
177 * Reopen the device if it's not currently open. Otherwise,
178 * just update the physical size of the device.
180 if (vd->vdev_tsd != NULL) {
181 ASSERT(vd->vdev_reopening);
186 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
189 * When opening a disk device, we want to preserve the user's original
190 * intent. We always want to open the device by the path the user gave
191 * us, even if it is one of multiple paths to the save device. But we
192 * also want to be able to survive disks being removed/recabled.
193 * Therefore the sequence of opening devices is:
195 * 1. Try opening the device by path. For legacy pools without the
196 * 'whole_disk' property, attempt to fix the path by appending 's0'.
198 * 2. If the devid of the device matches the stored value, return
201 * 3. Otherwise, the device may have moved. Try opening the device
202 * by the devid instead.
204 if (vd->vdev_devid != NULL) {
205 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
206 &dvd->vd_minor) != 0) {
207 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
208 return (SET_ERROR(EINVAL));
212 error = EINVAL; /* presume failure */
214 if (vd->vdev_path != NULL) {
216 if (vd->vdev_wholedisk == -1ULL) {
217 size_t len = strlen(vd->vdev_path) + 3;
218 char *buf = kmem_alloc(len, KM_SLEEP);
221 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
223 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
225 spa_strfree(vd->vdev_path);
227 vd->vdev_wholedisk = 1ULL;
228 (void) ldi_close(lh, spa_mode(spa), kcred);
234 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
235 &dvd->vd_lh, zfs_li);
238 * Compare the devid to the stored value.
240 if (error == 0 && vd->vdev_devid != NULL &&
241 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
242 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
243 error = SET_ERROR(EINVAL);
244 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
248 ddi_devid_free(devid);
252 * If we succeeded in opening the device, but 'vdev_wholedisk'
253 * is not yet set, then this must be a slice.
255 if (error == 0 && vd->vdev_wholedisk == -1ULL)
256 vd->vdev_wholedisk = 0;
260 * If we were unable to open by path, or the devid check fails, open by
263 if (error != 0 && vd->vdev_devid != NULL) {
264 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
265 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
269 * If all else fails, then try opening by physical path (if available)
270 * or the logical path (if we failed due to the devid check). While not
271 * as reliable as the devid, this will give us something, and the higher
272 * level vdev validation will prevent us from opening the wrong device.
275 if (vd->vdev_devid != NULL)
276 validate_devid = B_TRUE;
278 if (vd->vdev_physpath != NULL &&
279 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
280 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
281 kcred, &dvd->vd_lh, zfs_li);
284 * Note that we don't support the legacy auto-wholedisk support
285 * as above. This hasn't been used in a very long time and we
286 * don't need to propagate its oddities to this edge condition.
288 if (error && vd->vdev_path != NULL)
289 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
290 kcred, &dvd->vd_lh, zfs_li);
294 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
299 * Now that the device has been successfully opened, update the devid
302 if (validate_devid && spa_writeable(spa) &&
303 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
304 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
307 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
308 zfs_dbgmsg("vdev %s: update devid from %s, "
309 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
310 spa_strfree(vd->vdev_devid);
311 vd->vdev_devid = spa_strdup(vd_devid);
312 ddi_devid_str_free(vd_devid);
314 ddi_devid_free(devid);
318 * Once a device is opened, verify that the physical device path (if
319 * available) is up to date.
321 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
322 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
323 char *physpath, *minorname;
325 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
327 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
328 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
329 (vd->vdev_physpath == NULL ||
330 strcmp(vd->vdev_physpath, physpath) != 0)) {
331 if (vd->vdev_physpath)
332 spa_strfree(vd->vdev_physpath);
333 (void) strlcat(physpath, ":", MAXPATHLEN);
334 (void) strlcat(physpath, minorname, MAXPATHLEN);
335 vd->vdev_physpath = spa_strdup(physpath);
338 kmem_free(minorname, strlen(minorname) + 1);
339 kmem_free(physpath, MAXPATHLEN);
344 * Determine the actual size of the device.
346 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
347 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
348 return (SET_ERROR(EINVAL));
354 * Determine the device's minimum transfer size.
355 * If the ioctl isn't supported, assume DEV_BSIZE.
357 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
358 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
359 capacity = dkmext->dki_capacity - 1;
360 blksz = dkmext->dki_lbsize;
361 pbsize = dkmext->dki_pbsize;
362 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
363 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
365 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
367 capacity = dkm->dki_capacity - 1;
368 blksz = dkm->dki_lbsize;
371 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
372 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
373 vd->vdev_path, error);
377 *ashift = highbit(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
379 if (vd->vdev_wholedisk == 1) {
384 * If we have the capability to expand, we'd have
385 * found out via success from DKIOCGMEDIAINFO{,EXT}.
386 * Adjust max_psize upward accordingly since we know
387 * we own the whole disk now.
389 *max_psize += vdev_disk_get_space(vd, capacity, blksz);
390 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
391 "max_psize %llu", vd->vdev_path, *psize,
396 * Since we own the whole disk, try to enable disk write
397 * caching. We ignore errors because it's OK if we can't do it.
399 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
400 FKIOCTL, kcred, NULL);
404 * Clear the nowritecache bit, so that on a vdev_reopen() we will
407 vd->vdev_nowritecache = B_FALSE;
413 vdev_disk_close(vdev_t *vd)
415 vdev_disk_t *dvd = vd->vdev_tsd;
417 if (vd->vdev_reopening || dvd == NULL)
420 if (dvd->vd_minor != NULL)
421 ddi_devid_str_free(dvd->vd_minor);
423 if (dvd->vd_devid != NULL)
424 ddi_devid_free(dvd->vd_devid);
426 if (dvd->vd_lh != NULL)
427 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
429 vd->vdev_delayed_close = B_FALSE;
430 kmem_free(dvd, sizeof (vdev_disk_t));
435 vdev_disk_physio(vdev_t *vd, caddr_t data,
436 size_t size, uint64_t offset, int flags, boolean_t isdump)
438 vdev_disk_t *dvd = vd->vdev_tsd;
440 ASSERT(vd->vdev_ops == &vdev_disk_ops);
443 * If in the context of an active crash dump, use the ldi_dump(9F)
444 * call instead of ldi_strategy(9F) as usual.
447 ASSERT3P(dvd, !=, NULL);
448 return (ldi_dump(dvd->vd_lh, data, lbtodb(offset),
452 return (vdev_disk_ldi_physio(dvd->vd_lh, data, size, offset, flags));
456 vdev_disk_ldi_physio(ldi_handle_t vd_lh, caddr_t data,
457 size_t size, uint64_t offset, int flags)
463 return (SET_ERROR(EINVAL));
465 ASSERT(flags & B_READ || flags & B_WRITE);
467 bp = getrbuf(KM_SLEEP);
468 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
470 bp->b_un.b_addr = (void *)data;
471 bp->b_lblkno = lbtodb(offset);
472 bp->b_bufsize = size;
474 error = ldi_strategy(vd_lh, bp);
476 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
477 error = SET_ERROR(EIO);
484 vdev_disk_io_intr(buf_t *bp)
486 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
487 zio_t *zio = vdb->vdb_io;
490 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
491 * Rather than teach the rest of the stack about other error
492 * possibilities (EFAULT, etc), we normalize the error value here.
494 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
496 if (zio->io_error == 0 && bp->b_resid != 0)
497 zio->io_error = SET_ERROR(EIO);
499 kmem_free(vdb, sizeof (vdev_disk_buf_t));
505 vdev_disk_ioctl_free(zio_t *zio)
507 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
510 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
511 vdev_disk_ioctl_free,
512 zio_vsd_default_cksum_report
516 vdev_disk_ioctl_done(void *zio_arg, int error)
518 zio_t *zio = zio_arg;
520 zio->io_error = error;
526 vdev_disk_io_start(zio_t *zio)
528 vdev_t *vd = zio->io_vd;
529 vdev_disk_t *dvd = vd->vdev_tsd;
530 vdev_disk_buf_t *vdb;
531 struct dk_callback *dkc;
535 if (zio->io_type == ZIO_TYPE_IOCTL) {
537 if (!vdev_readable(vd)) {
538 zio->io_error = SET_ERROR(ENXIO);
539 return (ZIO_PIPELINE_CONTINUE);
542 switch (zio->io_cmd) {
544 case DKIOCFLUSHWRITECACHE:
546 if (zfs_nocacheflush)
549 if (vd->vdev_nowritecache) {
550 zio->io_error = SET_ERROR(ENOTSUP);
554 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
555 zio->io_vsd_ops = &vdev_disk_vsd_ops;
557 dkc->dkc_callback = vdev_disk_ioctl_done;
558 dkc->dkc_flag = FLUSH_VOLATILE;
559 dkc->dkc_cookie = zio;
561 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
562 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
566 * The ioctl will be done asychronously,
567 * and will call vdev_disk_ioctl_done()
570 return (ZIO_PIPELINE_STOP);
573 if (error == ENOTSUP || error == ENOTTY) {
575 * If we get ENOTSUP or ENOTTY, we know that
576 * no future attempts will ever succeed.
577 * In this case we set a persistent bit so
578 * that we don't bother with the ioctl in the
581 vd->vdev_nowritecache = B_TRUE;
583 zio->io_error = error;
588 zio->io_error = SET_ERROR(ENOTSUP);
591 return (ZIO_PIPELINE_CONTINUE);
594 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
600 bp->b_flags = B_BUSY | B_NOCACHE |
601 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
602 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
603 bp->b_flags |= B_FAILFAST;
604 bp->b_bcount = zio->io_size;
605 bp->b_un.b_addr = zio->io_data;
606 bp->b_lblkno = lbtodb(zio->io_offset);
607 bp->b_bufsize = zio->io_size;
608 bp->b_iodone = (int (*)())vdev_disk_io_intr;
610 /* ldi_strategy() will return non-zero only on programming errors */
611 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
613 return (ZIO_PIPELINE_STOP);
617 vdev_disk_io_done(zio_t *zio)
619 vdev_t *vd = zio->io_vd;
622 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
623 * the device has been removed. If this is the case, then we trigger an
624 * asynchronous removal of the device. Otherwise, probe the device and
625 * make sure it's still accessible.
627 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
628 vdev_disk_t *dvd = vd->vdev_tsd;
629 int state = DKIO_NONE;
631 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
632 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
634 * We post the resource as soon as possible, instead of
635 * when the async removal actually happens, because the
636 * DE is using this information to discard previous I/O
639 zfs_post_remove(zio->io_spa, vd);
640 vd->vdev_remove_wanted = B_TRUE;
641 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
642 } else if (!vd->vdev_delayed_close) {
643 vd->vdev_delayed_close = B_TRUE;
648 vdev_ops_t vdev_disk_ops = {
657 VDEV_TYPE_DISK, /* name of this vdev type */
658 B_TRUE /* leaf vdev */
662 * Given the root disk device devid or pathname, read the label from
663 * the device, and construct a configuration nvlist.
666 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
672 ddi_devid_t tmpdevid;
677 * Read the device label and build the nvlist.
679 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
681 error = ldi_open_by_devid(tmpdevid, minor_name,
682 FREAD, kcred, &vd_lh, zfs_li);
683 ddi_devid_free(tmpdevid);
684 ddi_devid_str_free(minor_name);
687 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
691 if (ldi_get_size(vd_lh, &s)) {
692 (void) ldi_close(vd_lh, FREAD, kcred);
693 return (SET_ERROR(EIO));
696 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
697 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
700 for (l = 0; l < VDEV_LABELS; l++) {
701 uint64_t offset, state, txg = 0;
703 /* read vdev label */
704 offset = vdev_label_offset(size, l, 0);
705 if (vdev_disk_ldi_physio(vd_lh, (caddr_t)label,
706 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
709 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
710 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
715 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
716 &state) != 0 || state >= POOL_STATE_DESTROYED) {
717 nvlist_free(*config);
722 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
723 &txg) != 0 || txg == 0) {
724 nvlist_free(*config);
732 kmem_free(label, sizeof (vdev_label_t));
733 (void) ldi_close(vd_lh, FREAD, kcred);
735 error = SET_ERROR(EIDRM);