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) 2012 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
27 #include <sys/spa_impl.h>
28 #include <sys/refcount.h>
29 #include <sys/vdev_disk.h>
30 #include <sys/vdev_impl.h>
31 #include <sys/fs/zfs.h>
33 #include <sys/sunldi.h>
34 #include <sys/efi_partition.h>
35 #include <sys/fm/fs/zfs.h>
38 * Virtual device vector for disks.
41 extern ldi_ident_t zfs_li;
43 typedef struct vdev_disk_buf {
49 vdev_disk_hold(vdev_t *vd)
54 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
57 * We must have a pathname, and it must be absolute.
59 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
63 * Only prefetch path and devid info if the device has
66 if (vd->vdev_tsd != NULL)
69 if (vd->vdev_wholedisk == -1ULL) {
70 size_t len = strlen(vd->vdev_path) + 3;
71 char *buf = kmem_alloc(len, KM_SLEEP);
73 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
75 (void) ldi_vp_from_name(buf, &vd->vdev_name_vp);
79 if (vd->vdev_name_vp == NULL)
80 (void) ldi_vp_from_name(vd->vdev_path, &vd->vdev_name_vp);
82 if (vd->vdev_devid != NULL &&
83 ddi_devid_str_decode(vd->vdev_devid, &devid, &minor) == 0) {
84 (void) ldi_vp_from_devid(devid, minor, &vd->vdev_devid_vp);
85 ddi_devid_str_free(minor);
86 ddi_devid_free(devid);
91 vdev_disk_rele(vdev_t *vd)
93 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
95 if (vd->vdev_name_vp) {
96 VN_RELE_ASYNC(vd->vdev_name_vp,
97 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
98 vd->vdev_name_vp = NULL;
100 if (vd->vdev_devid_vp) {
101 VN_RELE_ASYNC(vd->vdev_devid_vp,
102 dsl_pool_vnrele_taskq(vd->vdev_spa->spa_dsl_pool));
103 vd->vdev_devid_vp = NULL;
108 vdev_disk_get_space(vdev_t *vd, uint64_t capacity, uint_t blksz)
110 ASSERT(vd->vdev_wholedisk);
112 vdev_disk_t *dvd = vd->vdev_tsd;
115 uint64_t avail_space = 0;
116 int efisize = EFI_LABEL_SIZE * 2;
118 dk_ioc.dki_data = kmem_alloc(efisize, KM_SLEEP);
120 dk_ioc.dki_length = efisize;
121 dk_ioc.dki_data_64 = (uint64_t)(uintptr_t)dk_ioc.dki_data;
122 efi = dk_ioc.dki_data;
124 if (ldi_ioctl(dvd->vd_lh, DKIOCGETEFI, (intptr_t)&dk_ioc,
125 FKIOCTL, kcred, NULL) == 0) {
126 uint64_t efi_altern_lba = LE_64(efi->efi_gpt_AlternateLBA);
128 zfs_dbgmsg("vdev %s, capacity %llu, altern lba %llu",
129 vd->vdev_path, capacity, efi_altern_lba);
130 if (capacity > efi_altern_lba)
131 avail_space = (capacity - efi_altern_lba) * blksz;
133 kmem_free(dk_ioc.dki_data, efisize);
134 return (avail_space);
138 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
141 spa_t *spa = vd->vdev_spa;
143 struct dk_minfo_ext dkmext;
149 * We must have a pathname, and it must be absolute.
151 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
152 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
157 * Reopen the device if it's not currently open. Otherwise,
158 * just update the physical size of the device.
160 if (vd->vdev_tsd != NULL) {
161 ASSERT(vd->vdev_reopening);
166 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
169 * When opening a disk device, we want to preserve the user's original
170 * intent. We always want to open the device by the path the user gave
171 * us, even if it is one of multiple paths to the save device. But we
172 * also want to be able to survive disks being removed/recabled.
173 * Therefore the sequence of opening devices is:
175 * 1. Try opening the device by path. For legacy pools without the
176 * 'whole_disk' property, attempt to fix the path by appending 's0'.
178 * 2. If the devid of the device matches the stored value, return
181 * 3. Otherwise, the device may have moved. Try opening the device
182 * by the devid instead.
184 if (vd->vdev_devid != NULL) {
185 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
186 &dvd->vd_minor) != 0) {
187 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
192 error = EINVAL; /* presume failure */
194 if (vd->vdev_path != NULL) {
197 if (vd->vdev_wholedisk == -1ULL) {
198 size_t len = strlen(vd->vdev_path) + 3;
199 char *buf = kmem_alloc(len, KM_SLEEP);
202 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
204 if (ldi_open_by_name(buf, spa_mode(spa), kcred,
206 spa_strfree(vd->vdev_path);
208 vd->vdev_wholedisk = 1ULL;
209 (void) ldi_close(lh, spa_mode(spa), kcred);
215 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa), kcred,
216 &dvd->vd_lh, zfs_li);
219 * Compare the devid to the stored value.
221 if (error == 0 && vd->vdev_devid != NULL &&
222 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
223 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
225 (void) ldi_close(dvd->vd_lh, spa_mode(spa),
229 ddi_devid_free(devid);
233 * If we succeeded in opening the device, but 'vdev_wholedisk'
234 * is not yet set, then this must be a slice.
236 if (error == 0 && vd->vdev_wholedisk == -1ULL)
237 vd->vdev_wholedisk = 0;
241 * If we were unable to open by path, or the devid check fails, open by
244 if (error != 0 && vd->vdev_devid != NULL)
245 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
246 spa_mode(spa), kcred, &dvd->vd_lh, zfs_li);
249 * If all else fails, then try opening by physical path (if available)
250 * or the logical path (if we failed due to the devid check). While not
251 * as reliable as the devid, this will give us something, and the higher
252 * level vdev validation will prevent us from opening the wrong device.
255 if (vd->vdev_physpath != NULL &&
256 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != NODEV)
257 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode(spa),
258 kcred, &dvd->vd_lh, zfs_li);
261 * Note that we don't support the legacy auto-wholedisk support
262 * as above. This hasn't been used in a very long time and we
263 * don't need to propagate its oddities to this edge condition.
265 if (error && vd->vdev_path != NULL)
266 error = ldi_open_by_name(vd->vdev_path, spa_mode(spa),
267 kcred, &dvd->vd_lh, zfs_li);
271 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
276 * Once a device is opened, verify that the physical device path (if
277 * available) is up to date.
279 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
280 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
281 char *physpath, *minorname;
283 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
285 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
286 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
287 (vd->vdev_physpath == NULL ||
288 strcmp(vd->vdev_physpath, physpath) != 0)) {
289 if (vd->vdev_physpath)
290 spa_strfree(vd->vdev_physpath);
291 (void) strlcat(physpath, ":", MAXPATHLEN);
292 (void) strlcat(physpath, minorname, MAXPATHLEN);
293 vd->vdev_physpath = spa_strdup(physpath);
296 kmem_free(minorname, strlen(minorname) + 1);
297 kmem_free(physpath, MAXPATHLEN);
302 * Determine the actual size of the device.
304 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
305 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
310 * Determine the device's minimum transfer size.
311 * If the ioctl isn't supported, assume DEV_BSIZE.
313 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFOEXT, (intptr_t)&dkmext,
314 FKIOCTL, kcred, NULL) != 0)
315 dkmext.dki_pbsize = DEV_BSIZE;
317 *ashift = highbit(MAX(dkmext.dki_pbsize, SPA_MINBLOCKSIZE)) - 1;
319 if (vd->vdev_wholedisk == 1) {
320 uint64_t capacity = dkmext.dki_capacity - 1;
321 uint64_t blksz = dkmext.dki_lbsize;
325 * If we own the whole disk, try to enable disk write caching.
326 * We ignore errors because it's OK if we can't do it.
328 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
329 FKIOCTL, kcred, NULL);
331 *max_psize = *psize + vdev_disk_get_space(vd, capacity, blksz);
332 zfs_dbgmsg("capacity change: vdev %s, psize %llu, "
333 "max_psize %llu", vd->vdev_path, *psize, *max_psize);
339 * Clear the nowritecache bit, so that on a vdev_reopen() we will
342 vd->vdev_nowritecache = B_FALSE;
348 vdev_disk_close(vdev_t *vd)
350 vdev_disk_t *dvd = vd->vdev_tsd;
352 if (vd->vdev_reopening || dvd == NULL)
355 if (dvd->vd_minor != NULL)
356 ddi_devid_str_free(dvd->vd_minor);
358 if (dvd->vd_devid != NULL)
359 ddi_devid_free(dvd->vd_devid);
361 if (dvd->vd_lh != NULL)
362 (void) ldi_close(dvd->vd_lh, spa_mode(vd->vdev_spa), kcred);
364 vd->vdev_delayed_close = B_FALSE;
365 kmem_free(dvd, sizeof (vdev_disk_t));
370 vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size,
371 uint64_t offset, int flags)
379 ASSERT(flags & B_READ || flags & B_WRITE);
381 bp = getrbuf(KM_SLEEP);
382 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
384 bp->b_un.b_addr = (void *)data;
385 bp->b_lblkno = lbtodb(offset);
386 bp->b_bufsize = size;
388 error = ldi_strategy(vd_lh, bp);
390 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
398 vdev_disk_io_intr(buf_t *bp)
400 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
401 zio_t *zio = vdb->vdb_io;
404 * The rest of the zio stack only deals with EIO, ECKSUM, and ENXIO.
405 * Rather than teach the rest of the stack about other error
406 * possibilities (EFAULT, etc), we normalize the error value here.
408 zio->io_error = (geterror(bp) != 0 ? EIO : 0);
410 if (zio->io_error == 0 && bp->b_resid != 0)
413 kmem_free(vdb, sizeof (vdev_disk_buf_t));
419 vdev_disk_ioctl_free(zio_t *zio)
421 kmem_free(zio->io_vsd, sizeof (struct dk_callback));
424 static const zio_vsd_ops_t vdev_disk_vsd_ops = {
425 vdev_disk_ioctl_free,
426 zio_vsd_default_cksum_report
430 vdev_disk_ioctl_done(void *zio_arg, int error)
432 zio_t *zio = zio_arg;
434 zio->io_error = error;
440 vdev_disk_io_start(zio_t *zio)
442 vdev_t *vd = zio->io_vd;
443 vdev_disk_t *dvd = vd->vdev_tsd;
444 vdev_disk_buf_t *vdb;
445 struct dk_callback *dkc;
449 if (zio->io_type == ZIO_TYPE_IOCTL) {
451 if (!vdev_readable(vd)) {
452 zio->io_error = ENXIO;
453 return (ZIO_PIPELINE_CONTINUE);
456 switch (zio->io_cmd) {
458 case DKIOCFLUSHWRITECACHE:
460 if (zfs_nocacheflush)
463 if (vd->vdev_nowritecache) {
464 zio->io_error = ENOTSUP;
468 zio->io_vsd = dkc = kmem_alloc(sizeof (*dkc), KM_SLEEP);
469 zio->io_vsd_ops = &vdev_disk_vsd_ops;
471 dkc->dkc_callback = vdev_disk_ioctl_done;
472 dkc->dkc_flag = FLUSH_VOLATILE;
473 dkc->dkc_cookie = zio;
475 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
476 (uintptr_t)dkc, FKIOCTL, kcred, NULL);
480 * The ioctl will be done asychronously,
481 * and will call vdev_disk_ioctl_done()
484 return (ZIO_PIPELINE_STOP);
487 if (error == ENOTSUP || error == ENOTTY) {
489 * If we get ENOTSUP or ENOTTY, we know that
490 * no future attempts will ever succeed.
491 * In this case we set a persistent bit so
492 * that we don't bother with the ioctl in the
495 vd->vdev_nowritecache = B_TRUE;
497 zio->io_error = error;
502 zio->io_error = ENOTSUP;
505 return (ZIO_PIPELINE_CONTINUE);
508 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
514 bp->b_flags = B_BUSY | B_NOCACHE |
515 (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
516 if (!(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
517 bp->b_flags |= B_FAILFAST;
518 bp->b_bcount = zio->io_size;
519 bp->b_un.b_addr = zio->io_data;
520 bp->b_lblkno = lbtodb(zio->io_offset);
521 bp->b_bufsize = zio->io_size;
522 bp->b_iodone = (int (*)())vdev_disk_io_intr;
524 /* ldi_strategy() will return non-zero only on programming errors */
525 VERIFY(ldi_strategy(dvd->vd_lh, bp) == 0);
527 return (ZIO_PIPELINE_STOP);
531 vdev_disk_io_done(zio_t *zio)
533 vdev_t *vd = zio->io_vd;
536 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
537 * the device has been removed. If this is the case, then we trigger an
538 * asynchronous removal of the device. Otherwise, probe the device and
539 * make sure it's still accessible.
541 if (zio->io_error == EIO && !vd->vdev_remove_wanted) {
542 vdev_disk_t *dvd = vd->vdev_tsd;
543 int state = DKIO_NONE;
545 if (ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
546 FKIOCTL, kcred, NULL) == 0 && state != DKIO_INSERTED) {
548 * We post the resource as soon as possible, instead of
549 * when the async removal actually happens, because the
550 * DE is using this information to discard previous I/O
553 zfs_post_remove(zio->io_spa, vd);
554 vd->vdev_remove_wanted = B_TRUE;
555 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
556 } else if (!vd->vdev_delayed_close) {
557 vd->vdev_delayed_close = B_TRUE;
562 vdev_ops_t vdev_disk_ops = {
571 VDEV_TYPE_DISK, /* name of this vdev type */
572 B_TRUE /* leaf vdev */
576 * Given the root disk device devid or pathname, read the label from
577 * the device, and construct a configuration nvlist.
580 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
586 ddi_devid_t tmpdevid;
591 * Read the device label and build the nvlist.
593 if (devid != NULL && ddi_devid_str_decode(devid, &tmpdevid,
595 error = ldi_open_by_devid(tmpdevid, minor_name,
596 FREAD, kcred, &vd_lh, zfs_li);
597 ddi_devid_free(tmpdevid);
598 ddi_devid_str_free(minor_name);
601 if (error && (error = ldi_open_by_name(devpath, FREAD, kcred, &vd_lh,
605 if (ldi_get_size(vd_lh, &s)) {
606 (void) ldi_close(vd_lh, FREAD, kcred);
610 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
611 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
614 for (l = 0; l < VDEV_LABELS; l++) {
615 uint64_t offset, state, txg = 0;
617 /* read vdev label */
618 offset = vdev_label_offset(size, l, 0);
619 if (vdev_disk_physio(vd_lh, (caddr_t)label,
620 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, B_READ) != 0)
623 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
624 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
629 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
630 &state) != 0 || state >= POOL_STATE_DESTROYED) {
631 nvlist_free(*config);
636 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
637 &txg) != 0 || txg == 0) {
638 nvlist_free(*config);
646 kmem_free(label, sizeof (vdev_label_t));
647 (void) ldi_close(vd_lh, FREAD, kcred);