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 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #pragma ident "@(#)vdev_disk.c 1.15 08/04/09 SMI"
28 #include <sys/zfs_context.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>
38 * Virtual device vector for disks.
41 extern ldi_ident_t zfs_li;
43 typedef struct vdev_disk_buf {
49 vdev_disk_open_common(vdev_t *vd)
56 * We must have a pathname, and it must be absolute.
58 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
59 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
63 dvd = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_disk_t), KM_SLEEP);
66 * When opening a disk device, we want to preserve the user's original
67 * intent. We always want to open the device by the path the user gave
68 * us, even if it is one of multiple paths to the save device. But we
69 * also want to be able to survive disks being removed/recabled.
70 * Therefore the sequence of opening devices is:
72 * 1. Try opening the device by path. For legacy pools without the
73 * 'whole_disk' property, attempt to fix the path by appending 's0'.
75 * 2. If the devid of the device matches the stored value, return
78 * 3. Otherwise, the device may have moved. Try opening the device
79 * by the devid instead.
82 if (vd->vdev_devid != NULL) {
83 if (ddi_devid_str_decode(vd->vdev_devid, &dvd->vd_devid,
84 &dvd->vd_minor) != 0) {
85 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
90 error = EINVAL; /* presume failure */
92 if (vd->vdev_path != NULL) {
95 if (vd->vdev_wholedisk == -1ULL) {
96 size_t len = strlen(vd->vdev_path) + 3;
97 char *buf = kmem_alloc(len, KM_SLEEP);
100 (void) snprintf(buf, len, "%ss0", vd->vdev_path);
102 if (ldi_open_by_name(buf, spa_mode, kcred,
104 spa_strfree(vd->vdev_path);
106 vd->vdev_wholedisk = 1ULL;
107 (void) ldi_close(lh, spa_mode, kcred);
113 error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred,
114 &dvd->vd_lh, zfs_li);
117 * Compare the devid to the stored value.
119 if (error == 0 && vd->vdev_devid != NULL &&
120 ldi_get_devid(dvd->vd_lh, &devid) == 0) {
121 if (ddi_devid_compare(devid, dvd->vd_devid) != 0) {
123 (void) ldi_close(dvd->vd_lh, spa_mode, kcred);
126 ddi_devid_free(devid);
130 * If we succeeded in opening the device, but 'vdev_wholedisk'
131 * is not yet set, then this must be a slice.
133 if (error == 0 && vd->vdev_wholedisk == -1ULL)
134 vd->vdev_wholedisk = 0;
138 * If we were unable to open by path, or the devid check fails, open by
141 if (error != 0 && vd->vdev_devid != NULL)
142 error = ldi_open_by_devid(dvd->vd_devid, dvd->vd_minor,
143 spa_mode, kcred, &dvd->vd_lh, zfs_li);
146 * If all else fails, then try opening by physical path (if available)
147 * or the logical path (if we failed due to the devid check). While not
148 * as reliable as the devid, this will give us something, and the higher
149 * level vdev validation will prevent us from opening the wrong device.
152 if (vd->vdev_physpath != NULL &&
153 (dev = ddi_pathname_to_dev_t(vd->vdev_physpath)) != ENODEV)
154 error = ldi_open_by_dev(&dev, OTYP_BLK, spa_mode,
155 kcred, &dvd->vd_lh, zfs_li);
158 * Note that we don't support the legacy auto-wholedisk support
159 * as above. This hasn't been used in a very long time and we
160 * don't need to propagate its oddities to this edge condition.
162 if (error && vd->vdev_path != NULL)
163 error = ldi_open_by_name(vd->vdev_path, spa_mode, kcred,
164 &dvd->vd_lh, zfs_li);
168 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
174 vdev_disk_open(vdev_t *vd, uint64_t *psize, uint64_t *ashift)
182 error = vdev_disk_open_common(vd);
188 * Once a device is opened, verify that the physical device path (if
189 * available) is up to date.
191 if (ldi_get_dev(dvd->vd_lh, &dev) == 0 &&
192 ldi_get_otyp(dvd->vd_lh, &otyp) == 0) {
193 char *physpath, *minorname;
195 physpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
197 if (ddi_dev_pathname(dev, otyp, physpath) == 0 &&
198 ldi_get_minor_name(dvd->vd_lh, &minorname) == 0 &&
199 (vd->vdev_physpath == NULL ||
200 strcmp(vd->vdev_physpath, physpath) != 0)) {
201 if (vd->vdev_physpath)
202 spa_strfree(vd->vdev_physpath);
203 (void) strlcat(physpath, ":", MAXPATHLEN);
204 (void) strlcat(physpath, minorname, MAXPATHLEN);
205 vd->vdev_physpath = spa_strdup(physpath);
208 kmem_free(minorname, strlen(minorname) + 1);
209 kmem_free(physpath, MAXPATHLEN);
213 * Determine the actual size of the device.
215 if (ldi_get_size(dvd->vd_lh, psize) != 0) {
216 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
221 * If we own the whole disk, try to enable disk write caching.
222 * We ignore errors because it's OK if we can't do it.
224 if (vd->vdev_wholedisk == 1) {
226 (void) ldi_ioctl(dvd->vd_lh, DKIOCSETWCE, (intptr_t)&wce,
227 FKIOCTL, kcred, NULL);
231 * Determine the device's minimum transfer size.
232 * If the ioctl isn't supported, assume DEV_BSIZE.
234 if (ldi_ioctl(dvd->vd_lh, DKIOCGMEDIAINFO, (intptr_t)&dkm,
235 FKIOCTL, kcred, NULL) != 0)
236 dkm.dki_lbsize = DEV_BSIZE;
238 *ashift = highbit(MAX(dkm.dki_lbsize, SPA_MINBLOCKSIZE)) - 1;
241 * Clear the nowritecache bit, so that on a vdev_reopen() we will
244 vd->vdev_nowritecache = B_FALSE;
250 vdev_disk_close(vdev_t *vd)
252 vdev_disk_t *dvd = vd->vdev_tsd;
257 if (dvd->vd_minor != NULL)
258 ddi_devid_str_free(dvd->vd_minor);
260 if (dvd->vd_devid != NULL)
261 ddi_devid_free(dvd->vd_devid);
263 if (dvd->vd_lh != NULL)
264 (void) ldi_close(dvd->vd_lh, spa_mode, kcred);
266 kmem_free(dvd, sizeof (vdev_disk_t));
271 vdev_disk_physio(ldi_handle_t vd_lh, caddr_t data, size_t size,
272 uint64_t offset, int flags)
280 ASSERT(flags & B_READ || flags & B_WRITE);
282 bp = getrbuf(KM_SLEEP);
283 bp->b_flags = flags | B_BUSY | B_NOCACHE | B_FAILFAST;
285 bp->b_un.b_addr = (void *)data;
286 bp->b_lblkno = lbtodb(offset);
287 bp->b_bufsize = size;
289 error = ldi_strategy(vd_lh, bp);
291 if ((error = biowait(bp)) == 0 && bp->b_resid != 0)
299 vdev_disk_probe_io(vdev_t *vd, caddr_t data, size_t size, uint64_t offset,
303 vdev_disk_t *dvd = vd->vdev_tsd;
305 if (vd == NULL || dvd == NULL || dvd->vd_lh == NULL)
308 error = vdev_disk_physio(dvd->vd_lh, data, size, offset, flags);
310 if (zio_injection_enabled && error == 0)
311 error = zio_handle_device_injection(vd, EIO);
317 * Determine if the underlying device is accessible by reading and writing
318 * to a known location. We must be able to do this during syncing context
319 * and thus we cannot set the vdev state directly.
322 vdev_disk_probe(vdev_t *vd)
326 int l, error = 0, retries = 0;
332 /* Hijack the current vdev */
336 * Pick a random label to rewrite.
338 l = spa_get_random(VDEV_LABELS);
339 ASSERT(l < VDEV_LABELS);
341 offset = vdev_label_offset(vd->vdev_psize, l,
342 offsetof(vdev_label_t, vl_pad));
344 vl_pad = kmem_alloc(VDEV_SKIP_SIZE, KM_SLEEP);
347 * Try to read and write to a special location on the
348 * label. We use the existing vdev initially and only
349 * try to create and reopen it if we encounter a failure.
351 while ((error = vdev_disk_probe_io(nvd, vl_pad, VDEV_SKIP_SIZE,
352 offset, B_READ)) != 0 && retries == 0) {
354 nvd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP);
356 nvd->vdev_path = spa_strdup(vd->vdev_path);
357 if (vd->vdev_physpath)
358 nvd->vdev_physpath = spa_strdup(vd->vdev_physpath);
360 nvd->vdev_devid = spa_strdup(vd->vdev_devid);
361 nvd->vdev_wholedisk = vd->vdev_wholedisk;
362 nvd->vdev_guid = vd->vdev_guid;
365 error = vdev_disk_open_common(nvd);
371 error = vdev_disk_probe_io(nvd, vl_pad, VDEV_SKIP_SIZE,
375 /* Clean up if we allocated a new vdev */
377 vdev_disk_close(nvd);
379 spa_strfree(nvd->vdev_path);
380 if (nvd->vdev_physpath)
381 spa_strfree(nvd->vdev_physpath);
383 spa_strfree(nvd->vdev_devid);
384 kmem_free(nvd, sizeof (vdev_t));
386 kmem_free(vl_pad, VDEV_SKIP_SIZE);
388 /* Reset the failing flag */
390 vd->vdev_is_failing = B_FALSE;
396 vdev_disk_io_intr(buf_t *bp)
398 vdev_disk_buf_t *vdb = (vdev_disk_buf_t *)bp;
399 zio_t *zio = vdb->vdb_io;
401 if ((zio->io_error = geterror(bp)) == 0 && bp->b_resid != 0)
404 kmem_free(vdb, sizeof (vdev_disk_buf_t));
410 vdev_disk_ioctl_done(void *zio_arg, int error)
412 zio_t *zio = zio_arg;
414 zio->io_error = error;
420 vdev_disk_io_start(zio_t *zio)
422 vdev_t *vd = zio->io_vd;
423 vdev_disk_t *dvd = vd->vdev_tsd;
424 vdev_disk_buf_t *vdb;
428 if (zio->io_type == ZIO_TYPE_IOCTL) {
429 zio_vdev_io_bypass(zio);
432 if (!vdev_readable(vd)) {
433 zio->io_error = ENXIO;
434 return (ZIO_PIPELINE_CONTINUE);
437 switch (zio->io_cmd) {
439 case DKIOCFLUSHWRITECACHE:
441 if (zfs_nocacheflush)
444 if (vd->vdev_nowritecache) {
445 zio->io_error = ENOTSUP;
449 zio->io_dk_callback.dkc_callback = vdev_disk_ioctl_done;
450 zio->io_dk_callback.dkc_flag = FLUSH_VOLATILE;
451 zio->io_dk_callback.dkc_cookie = zio;
453 error = ldi_ioctl(dvd->vd_lh, zio->io_cmd,
454 (uintptr_t)&zio->io_dk_callback,
455 FKIOCTL, kcred, NULL);
459 * The ioctl will be done asychronously,
460 * and will call vdev_disk_ioctl_done()
463 return (ZIO_PIPELINE_STOP);
466 if (error == ENOTSUP || error == ENOTTY) {
468 * If we get ENOTSUP or ENOTTY, we know that
469 * no future attempts will ever succeed.
470 * In this case we set a persistent bit so
471 * that we don't bother with the ioctl in the
474 vd->vdev_nowritecache = B_TRUE;
476 zio->io_error = error;
481 zio->io_error = ENOTSUP;
484 return (ZIO_PIPELINE_CONTINUE);
487 if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
488 return (ZIO_PIPELINE_STOP);
490 if ((zio = vdev_queue_io(zio)) == NULL)
491 return (ZIO_PIPELINE_STOP);
493 if (zio->io_type == ZIO_TYPE_WRITE)
494 error = vdev_writeable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
496 error = vdev_readable(vd) ? vdev_error_inject(vd, zio) : ENXIO;
497 error = (vd->vdev_remove_wanted || vd->vdev_is_failing) ? ENXIO : error;
500 zio->io_error = error;
502 return (ZIO_PIPELINE_STOP);
505 flags = (zio->io_type == ZIO_TYPE_READ ? B_READ : B_WRITE);
506 flags |= B_BUSY | B_NOCACHE;
507 if (zio->io_flags & ZIO_FLAG_FAILFAST)
510 vdb = kmem_alloc(sizeof (vdev_disk_buf_t), KM_SLEEP);
517 bp->b_bcount = zio->io_size;
518 bp->b_un.b_addr = zio->io_data;
519 bp->b_lblkno = lbtodb(zio->io_offset);
520 bp->b_bufsize = zio->io_size;
521 bp->b_iodone = (int (*)())vdev_disk_io_intr;
523 error = ldi_strategy(dvd->vd_lh, bp);
524 /* ldi_strategy() will return non-zero only on programming errors */
527 return (ZIO_PIPELINE_STOP);
531 vdev_disk_io_done(zio_t *zio)
533 vdev_queue_io_done(zio);
535 if (zio->io_type == ZIO_TYPE_WRITE)
536 vdev_cache_write(zio);
538 if (zio_injection_enabled && zio->io_error == 0)
539 zio->io_error = zio_handle_device_injection(zio->io_vd, EIO);
542 * If the device returned EIO, then attempt a DKIOCSTATE ioctl to see if
543 * the device has been removed. If this is the case, then we trigger an
544 * asynchronous removal of the device. Otherwise, probe the device and
545 * make sure it's still accessible.
547 if (zio->io_error == EIO) {
548 vdev_t *vd = zio->io_vd;
549 vdev_disk_t *dvd = vd->vdev_tsd;
553 if (dvd && ldi_ioctl(dvd->vd_lh, DKIOCSTATE, (intptr_t)&state,
554 FKIOCTL, kcred, NULL) == 0 &&
555 state != DKIO_INSERTED) {
556 vd->vdev_remove_wanted = B_TRUE;
557 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
558 } else if (vdev_probe(vd) != 0) {
559 ASSERT(vd->vdev_ops->vdev_op_leaf);
560 vd->vdev_is_failing = B_TRUE;
564 return (ZIO_PIPELINE_CONTINUE);
567 vdev_ops_t vdev_disk_ops = {
575 VDEV_TYPE_DISK, /* name of this vdev type */
576 B_TRUE /* leaf vdev */
580 * Given the root disk device pathname, read the label from the device,
581 * and construct a configuration nvlist.
584 vdev_disk_read_rootlabel(char *devpath)
586 nvlist_t *config = NULL;
593 * Read the device label and build the nvlist.
595 if (ldi_open_by_name(devpath, FREAD, kcred, &vd_lh, zfs_li))
598 if (ldi_get_size(vd_lh, &s))
601 size = P2ALIGN_TYPED(s, sizeof (vdev_label_t), uint64_t);
602 label = kmem_alloc(sizeof (vdev_label_t), KM_SLEEP);
604 for (l = 0; l < VDEV_LABELS; l++) {
605 uint64_t offset, state, txg = 0;
607 /* read vdev label */
608 offset = vdev_label_offset(size, l, 0);
609 if (vdev_disk_physio(vd_lh, (caddr_t)label,
610 VDEV_SKIP_SIZE + VDEV_BOOT_HEADER_SIZE +
611 VDEV_PHYS_SIZE, offset, B_READ) != 0)
614 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
615 sizeof (label->vl_vdev_phys.vp_nvlist), &config, 0) != 0) {
620 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_STATE,
621 &state) != 0 || state >= POOL_STATE_DESTROYED) {
627 if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_TXG,
628 &txg) != 0 || txg == 0) {
637 kmem_free(label, sizeof (vdev_label_t));