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.
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 * We want to be loud in DEBUG kernels when DKIOCGMEDIAINFOEXT fails, or when
140 * even a fallback to DKIOCGMEDIAINFO fails.
143 #define VDEV_DEBUG(...) cmn_err(CE_NOTE, __VA_ARGS__)
145 #define VDEV_DEBUG(...) /* Nothing... */
149 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
152 spa_t *spa = vd->vdev_spa;
155 struct dk_minfo_ext ude;
158 struct dk_minfo_ext *dkmext = &dks.ude;
159 struct dk_minfo *dkm = &dks.ud;
163 boolean_t validate_devid = B_FALSE;
165 uint64_t capacity = 0, blksz = 0, pbsize;
168 * We must have a pathname, and it must be absolute.
170 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
171 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
172 return (SET_ERROR(EINVAL));
176 * Reopen the device if it's not currently open. Otherwise,
177 * just update the physical size of the device.
179 if (vd->vdev_tsd != NULL) {
180 ASSERT(vd->vdev_reopening);
185 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
188 * When opening a disk device, we want to preserve the user's original
189 * intent. We always want to open the device by the path the user gave
190 * us, even if it is one of multiple paths to the save device. But we
191 * also want to be able to survive disks being removed/recabled.
192 * Therefore the sequence of opening devices is:
194 * 1. Try opening the device by path. For legacy pools without the
195 * 'whole_disk' property, attempt to fix the path by appending 's0'.
197 * 2. If the devid of the device matches the stored value, return
200 * 3. Otherwise, the device may have moved. Try opening the device
201 * by the devid instead.
203 if (vd->vdev_devid != NULL) {
204 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
205 &dvd->vd_minor) != 0) {
206 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
207 return (SET_ERROR(EINVAL));
211 error = EINVAL; /* presume failure */
213 if (vd->vdev_path != NULL) {
215 if (vd->vdev_wholedisk == -1ULL) {
216 size_t len = strlen(vd->vdev_path) + 3;
217 char *buf = kmem_alloc(len, KM_SLEEP);
220 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
222 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
224 spa_strfree(vd->vdev_path);
226 vd->vdev_wholedisk = 1ULL;
227 (void) ldi_close(lh, spa_mode(spa), kcred);
233 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
234 &dvd->vd_lh, zfs_li);
237 * Compare the devid to the stored value.
239 if (error == 0 && vd->vdev_devid != NULL &&
240 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
241 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
242 error = SET_ERROR(EINVAL);
243 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
247 ddi_devid_free(devid);
251 * If we succeeded in opening the device, but 'vdev_wholedisk'
252 * is not yet set, then this must be a slice.
254 if (error == 0 && vd->vdev_wholedisk == -1ULL)
255 vd->vdev_wholedisk = 0;
259 * If we were unable to open by path, or the devid check fails, open by
262 if (error != 0 && vd->vdev_devid != NULL) {
263 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
264 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
268 * If all else fails, then try opening by physical path (if available)
269 * or the logical path (if we failed due to the devid check). While not
270 * as reliable as the devid, this will give us something, and the higher
271 * level vdev validation will prevent us from opening the wrong device.
274 if (vd->vdev_devid != NULL)
275 validate_devid = B_TRUE;
277 if (vd->vdev_physpath != NULL &&
278 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
279 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
280 kcred, &dvd->vd_lh, zfs_li);
283 * Note that we don't support the legacy auto-wholedisk support
284 * as above. This hasn't been used in a very long time and we
285 * don't need to propagate its oddities to this edge condition.
287 if (error && vd->vdev_path != NULL)
288 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
289 kcred, &dvd->vd_lh, zfs_li);
293 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
298 * Now that the device has been successfully opened, update the devid
301 if (validate_devid && spa_writeable(spa) &&
302 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
303 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
306 vd_devid = ddi_devid_str_encode(devid, dvd->vd_minor);
307 zfs_dbgmsg("vdev %s: update devid from %s, "
308 "to %s", vd->vdev_path, vd->vdev_devid, vd_devid);
309 spa_strfree(vd->vdev_devid);
310 vd->vdev_devid = spa_strdup(vd_devid);
311 ddi_devid_str_free(vd_devid);
313 ddi_devid_free(devid);
317 * Once a device is opened, verify that the physical device path (if
318 * available) is up to date.
320 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
321 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
322 char *physpath, *minorname;
324 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
326 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
327 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
328 (vd->vdev_physpath == NULL ||
329 strcmp(vd->vdev_physpath, physpath) != 0)) {
330 if (vd->vdev_physpath)
331 spa_strfree(vd->vdev_physpath);
332 (void) strlcat(physpath, ":", MAXPATHLEN);
333 (void) strlcat(physpath, minorname, MAXPATHLEN);
334 vd->vdev_physpath = spa_strdup(physpath);
337 kmem_free(minorname, strlen(minorname) + 1);
338 kmem_free(physpath, MAXPATHLEN);
343 * Determine the actual size of the device.
345 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
346 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
347 return (SET_ERROR(EINVAL));
353 * Determine the device's minimum transfer size.
354 * If the ioctl isn't supported, assume DEV_BSIZE.
356 if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT,
357 (intptr_t)dkmext, FKIOCTL, kcred, NULL)) == 0) {
358 capacity = dkmext->dki_capacity - 1;
359 blksz = dkmext->dki_lbsize;
360 pbsize = dkmext->dki_pbsize;
361 } else if ((error = ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO,
362 (intptr_t)dkm, FKIOCTL, kcred, NULL)) == 0) {
364 "vdev_disk_open(\"%s\"): fallback to DKIOCGMEDIAINFO\n",
366 capacity = dkm->dki_capacity - 1;
367 blksz = dkm->dki_lbsize;
370 VDEV_DEBUG("vdev_disk_open(\"%s\"): "
371 "both DKIOCGMEDIAINFO{,EXT} calls failed, %d\n",
372 vd->vdev_path, error);
376 *ashift = highbit(MAX(pbsize, SPA_MINBLOCKSIZE)) - 1;
378 if (vd->vdev_wholedisk == 1) {
383 * If we have the capability to expand, we'd have
384 * found out via success from DKIOCGMEDIAINFO{,EXT}.
385 * Adjust max_psize upward accordingly since we know
386 * we own the whole disk now.
388 *max_psize += vdev_disk_get_space(vd, capacity, blksz);
389 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
390 "max_psize %llu", vd->vdev_path, *psize,
395 * Since we own the whole disk, try to enable disk write
396 * caching. We ignore errors because it's OK if we can't do it.
398 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
399 FKIOCTL, kcred, NULL);
403 * Clear the nowritecache bit, so that on a vdev_reopen() we will
406 vd->vdev_nowritecache = B_FALSE;
412 vdev_disk_close(vdev_t *vd)
414 vdev_disk_t *dvd = vd->vdev_tsd;
416 if (vd->vdev_reopening || dvd == NULL)
419 if (dvd->vd_minor != NULL)
420 ddi_devid_str_free(dvd->vd_minor);
422 if (dvd->vd_devid != NULL)
423 ddi_devid_free(dvd->vd_devid);
425 if (dvd->vd_lh != NULL)
426 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
428 vd->vdev_delayed_close = B_FALSE;
429 kmem_free(dvd, sizeof (vdev_disk_t));
434 vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size,
435 uint64_t offset, int flags)
441 return (SET_ERROR(EINVAL));
443 ASSERT(flags & B_READ || flags & B_WRITE);
445 bp = getrbuf(KM_SLEEP);
446 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
448 bp->b_un.b_addr = (void *)data;
449 bp->b_lblkno = lbtodb(offset);
450 bp->b_bufsize = size;
452 error = ldi_strategy(vd_lh, bp);
454 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
455 error = SET_ERROR(EIO);
462 vdev_disk_io_intr(buf_t *bp)
464 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
465 zio_t *zio = vdb->vdb_io;
468 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
469 * Rather than teach the rest of the stack about other error
470 * possibilities (EFAULT, etc), we normalize the error value here.
472 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
474 if (zio->io_error == 0 && bp->b_resid != 0)
475 zio->io_error = SET_ERROR(EIO);
477 kmem_free(vdb, sizeof (vdev_disk_buf_t));
483 vdev_disk_ioctl_free(zio_t *zio)
485 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
488 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
489 vdev_disk_ioctl_free,
490 zio_vsd_default_cksum_report
494 vdev_disk_ioctl_done(void *zio_arg, int error)
496 zio_t *zio = zio_arg;
498 zio->io_error = error;
504 vdev_disk_io_start(zio_t *zio)
506 vdev_t *vd = zio->io_vd;
507 vdev_disk_t *dvd = vd->vdev_tsd;
508 vdev_disk_buf_t *vdb;
509 struct dk_callback *dkc;
513 if (zio->io_type == ZIO_TYPE_IOCTL) {
515 if (!vdev_readable(vd)) {
516 zio->io_error = SET_ERROR(ENXIO);
517 return (ZIO_PIPELINE_CONTINUE);
520 switch (zio->io_cmd) {
522 case DKIOCFLUSHWRITECACHE:
524 if (zfs_nocacheflush)
527 if (vd->vdev_nowritecache) {
528 zio->io_error = SET_ERROR(ENOTSUP);
532 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
533 zio->io_vsd_ops = &vdev_disk_vsd_ops;
535 dkc->dkc_callback = vdev_disk_ioctl_done;
536 dkc->dkc_flag = FLUSH_VOLATILE;
537 dkc->dkc_cookie = zio;
539 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
540 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
544 * The ioctl will be done asychronously,
545 * and will call vdev_disk_ioctl_done()
548 return (ZIO_PIPELINE_STOP);
551 if (error == ENOTSUP || error == ENOTTY) {
553 * If we get ENOTSUP or ENOTTY, we know that
554 * no future attempts will ever succeed.
555 * In this case we set a persistent bit so
556 * that we don't bother with the ioctl in the
559 vd->vdev_nowritecache = B_TRUE;
561 zio->io_error = error;
566 zio->io_error = SET_ERROR(ENOTSUP);
569 return (ZIO_PIPELINE_CONTINUE);
572 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
578 bp->b_flags = B_BUSY | B_NOCACHE |
579 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
580 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
581 bp->b_flags |= B_FAILFAST;
582 bp->b_bcount = zio->io_size;
583 bp->b_un.b_addr = zio->io_data;
584 bp->b_lblkno = lbtodb(zio->io_offset);
585 bp->b_bufsize = zio->io_size;
586 bp->b_iodone = (int (*)())vdev_disk_io_intr;
588 /* ldi_strategy() will return non-zero only on programming errors */
589 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
591 return (ZIO_PIPELINE_STOP);
595 vdev_disk_io_done(zio_t *zio)
597 vdev_t *vd = zio->io_vd;
600 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
601 * the device has been removed. If this is the case, then we trigger an
602 * asynchronous removal of the device. Otherwise, probe the device and
603 * make sure it's still accessible.
605 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
606 vdev_disk_t *dvd = vd->vdev_tsd;
607 int state = DKIO_NONE;
609 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
610 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
612 * We post the resource as soon as possible, instead of
613 * when the async removal actually happens, because the
614 * DE is using this information to discard previous I/O
617 zfs_post_remove(zio->io_spa, vd);
618 vd->vdev_remove_wanted = B_TRUE;
619 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
620 } else if (!vd->vdev_delayed_close) {
621 vd->vdev_delayed_close = B_TRUE;
626 vdev_ops_t vdev_disk_ops = {
635 VDEV_TYPE_DISK, /* name of this vdev type */
636 B_TRUE /* leaf vdev */
640 * Given the root disk device devid or pathname, read the label from
641 * the device, and construct a configuration nvlist.
644 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
650 ddi_devid_t tmpdevid;
655 * Read the device label and build the nvlist.
657 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
659 error = ldi_open_by_devid(tmpdevid, minor_name,
660 FREAD, kcred, &vd_lh, zfs_li);
661 ddi_devid_free(tmpdevid);
662 ddi_devid_str_free(minor_name);
665 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
669 if (ldi_get_size(vd_lh, &s)) {
670 (void) ldi_close(vd_lh, FREAD, kcred);
671 return (SET_ERROR(EIO));
674 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
675 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
678 for (l = 0; l < VDEV_LABELS; l++) {
679 uint64_t offset, state, txg = 0;
681 /* read vdev label */
682 offset = vdev_label_offset(size, l, 0);
683 if (vdev_disk_physio(vd_lh, (caddr_t)label,
684 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
687 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
688 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
693 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
694 &state) != 0 || state >= POOL_STATE_DESTROYED) {
695 nvlist_free(*config);
700 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
701 &txg) != 0 || txg == 0) {
702 nvlist_free(*config);
710 kmem_free(label, sizeof (vdev_label_t));
711 (void) ldi_close(vd_lh, FREAD, kcred);
713 error = SET_ERROR(EIDRM);