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) 2008-2010 Lawrence Livermore National Security, LLC.
23 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
24 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
28 #include <sys/zfs_context.h>
30 #include <sys/vdev_disk.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/fs/zfs.h>
34 #include <sys/sunldi.h>
37 * Virtual device vector for disks.
39 typedef struct dio_request {
40 struct completion dr_comp; /* Completion for sync IO */
41 atomic_t dr_ref; /* References */
42 zio_t *dr_zio; /* Parent ZIO */
43 int dr_rw; /* Read/Write */
44 int dr_error; /* Bio error */
45 int dr_bio_count; /* Count of bio's */
46 struct bio *dr_bio[0]; /* Attached bio's */
50 #ifdef HAVE_OPEN_BDEV_EXCLUSIVE
52 vdev_bdev_mode(int smode)
56 ASSERT3S(smode & (FREAD | FWRITE), !=, 0);
68 vdev_bdev_mode(int smode)
72 ASSERT3S(smode & (FREAD | FWRITE), !=, 0);
74 if ((smode & FREAD) && !(smode & FWRITE))
79 #endif /* HAVE_OPEN_BDEV_EXCLUSIVE */
82 bdev_capacity(struct block_device *bdev)
84 struct hd_struct *part = bdev->bd_part;
86 /* The partition capacity referenced by the block device */
88 return part->nr_sects;
90 /* Otherwise assume the full device capacity */
91 return get_capacity(bdev->bd_disk);
95 vdev_disk_error(zio_t *zio)
98 printk("ZFS: zio error=%d type=%d offset=%llu size=%llu "
99 "flags=%x delay=%llu\n", zio->io_error, zio->io_type,
100 (u_longlong_t)zio->io_offset, (u_longlong_t)zio->io_size,
101 zio->io_flags, (u_longlong_t)zio->io_delay);
106 vdev_disk_open(vdev_t *v, uint64_t *psize, uint64_t *ashift)
108 struct block_device *bdev;
110 int mode, block_size;
112 /* Must have a pathname and it must be absolute. */
113 if (v->vdev_path == NULL || v->vdev_path[0] != '/') {
114 v->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
118 vd = kmem_zalloc(sizeof(vdev_disk_t), KM_SLEEP);
123 * Devices are always opened by the path provided at configuration
124 * time. This means that if the provided path is a udev by-id path
125 * then drives may be recabled without an issue. If the provided
126 * path is a udev by-path path then the physical location information
127 * will be preserved. This can be critical for more complicated
128 * configurations where drives are located in specific physical
129 * locations to maximize the systems tolerence to component failure.
130 * Alternately you can provide your own udev rule to flexibly map
131 * the drives as you see fit. It is not advised that you use the
132 * /dev/[hd]d devices which may be reorder due to probing order.
133 * Devices in the wrong locations will be detected by the higher
134 * level vdev validation.
136 mode = spa_mode(v->vdev_spa);
137 bdev = vdev_bdev_open(v->vdev_path, vdev_bdev_mode(mode), vd);
139 kmem_free(vd, sizeof(vdev_disk_t));
140 return -PTR_ERR(bdev);
145 block_size = vdev_bdev_block_size(bdev);
147 /* We think the wholedisk property should always be set when this
148 * function is called. ASSERT here so if any legitimate cases exist
149 * where it's not set, we'll find them during debugging. If we never
150 * hit the ASSERT, this and the following conditional statement can be
152 ASSERT3S(v->vdev_wholedisk, !=, -1ULL);
154 /* The wholedisk property was initialized to -1 in vdev_alloc() if it
155 * was unspecified. In that case, check if this is a whole device.
156 * When bdev->bd_contains == bdev we have a whole device and not simply
158 if (v->vdev_wholedisk == -1ULL)
159 v->vdev_wholedisk = (bdev->bd_contains == bdev);
161 /* Clear the nowritecache bit, causes vdev_reopen() to try again. */
162 v->vdev_nowritecache = B_FALSE;
164 /* Physical volume size in bytes */
165 *psize = bdev_capacity(bdev) * block_size;
167 /* Based on the minimum sector size set the block size */
168 *ashift = highbit(MAX(block_size, SPA_MINBLOCKSIZE)) - 1;
174 vdev_disk_close(vdev_t *v)
176 vdev_disk_t *vd = v->vdev_tsd;
181 if (vd->vd_bdev != NULL)
182 vdev_bdev_close(vd->vd_bdev,
183 vdev_bdev_mode(spa_mode(v->vdev_spa)));
185 kmem_free(vd, sizeof(vdev_disk_t));
189 static dio_request_t *
190 vdev_disk_dio_alloc(int bio_count)
195 dr = kmem_zalloc(sizeof(dio_request_t) +
196 sizeof(struct bio *) * bio_count, KM_SLEEP);
198 init_completion(&dr->dr_comp);
199 atomic_set(&dr->dr_ref, 0);
200 dr->dr_bio_count = bio_count;
203 for (i = 0; i < dr->dr_bio_count; i++)
204 dr->dr_bio[i] = NULL;
211 vdev_disk_dio_free(dio_request_t *dr)
215 for (i = 0; i < dr->dr_bio_count; i++)
217 bio_put(dr->dr_bio[i]);
219 kmem_free(dr, sizeof(dio_request_t) +
220 sizeof(struct bio *) * dr->dr_bio_count);
224 vdev_disk_dio_is_sync(dio_request_t *dr)
226 #ifdef HAVE_BIO_RW_SYNC
227 /* BIO_RW_SYNC preferred interface from 2.6.12-2.6.29 */
228 return (dr->dr_rw & (1 << BIO_RW_SYNC));
230 # ifdef HAVE_BIO_RW_SYNCIO
231 /* BIO_RW_SYNCIO preferred interface from 2.6.30-2.6.35 */
232 return (dr->dr_rw & (1 << BIO_RW_SYNCIO));
234 # ifdef HAVE_REQ_SYNC
235 /* REQ_SYNC preferred interface from 2.6.36-2.6.xx */
236 return (dr->dr_rw & REQ_SYNC);
238 # error "Unable to determine bio sync flag"
239 # endif /* HAVE_REQ_SYNC */
240 # endif /* HAVE_BIO_RW_SYNC */
241 #endif /* HAVE_BIO_RW_SYNCIO */
245 vdev_disk_dio_get(dio_request_t *dr)
247 atomic_inc(&dr->dr_ref);
251 vdev_disk_dio_put(dio_request_t *dr)
253 int rc = atomic_dec_return(&dr->dr_ref);
256 * Free the dio_request when the last reference is dropped and
257 * ensure zio_interpret is called only once with the correct zio
260 zio_t *zio = dr->dr_zio;
261 int error = dr->dr_error;
263 vdev_disk_dio_free(dr);
266 zio->io_delay = jiffies_to_msecs(
267 jiffies_64 - zio->io_delay);
268 zio->io_error = error;
269 ASSERT3S(zio->io_error, >=, 0);
271 vdev_disk_error(zio);
279 BIO_END_IO_PROTO(vdev_disk_physio_completion, bio, size, error)
281 dio_request_t *dr = bio->bi_private;
284 /* Fatal error but print some useful debugging before asserting */
286 PANIC("dr == NULL, bio->bi_private == NULL\n"
287 "bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d\n"
288 "bi_idx: %d, bi_size: %d, bi_end_io: %p, bi_cnt: %d\n",
289 bio->bi_next, bio->bi_flags, bio->bi_rw, bio->bi_vcnt,
290 bio->bi_idx, bio->bi_size, bio->bi_end_io,
291 atomic_read(&bio->bi_cnt));
293 #ifndef HAVE_2ARGS_BIO_END_IO_T
296 #endif /* HAVE_2ARGS_BIO_END_IO_T */
298 if (error == 0 && !test_bit(BIO_UPTODATE, &bio->bi_flags))
301 if (dr->dr_error == 0)
302 dr->dr_error = -error;
304 /* Drop reference aquired by __vdev_disk_physio */
305 rc = vdev_disk_dio_put(dr);
307 /* Wake up synchronous waiter this is the last outstanding bio */
308 if ((rc == 1) && vdev_disk_dio_is_sync(dr))
309 complete(&dr->dr_comp);
311 BIO_END_IO_RETURN(0);
314 static inline unsigned long
315 bio_nr_pages(void *bio_ptr, unsigned int bio_size)
317 return ((((unsigned long)bio_ptr + bio_size + PAGE_SIZE - 1) >>
318 PAGE_SHIFT) - ((unsigned long)bio_ptr >> PAGE_SHIFT));
322 bio_map(struct bio *bio, void *bio_ptr, unsigned int bio_size)
324 unsigned int offset, size, i;
327 offset = offset_in_page(bio_ptr);
328 for (i = 0; i < bio->bi_max_vecs; i++) {
329 size = PAGE_SIZE - offset;
337 if (kmem_virt(bio_ptr))
338 page = vmalloc_to_page(bio_ptr);
340 page = virt_to_page(bio_ptr);
342 if (bio_add_page(bio, page, size, offset) != size)
354 __vdev_disk_physio(struct block_device *bdev, zio_t *zio, caddr_t kbuf_ptr,
355 size_t kbuf_size, uint64_t kbuf_offset, int flags)
360 int bio_size, bio_count = 16;
361 int i = 0, error = 0, block_size;
364 dr = vdev_disk_dio_alloc(bio_count);
368 if (zio && !(zio->io_flags & (ZIO_FLAG_IO_RETRY | ZIO_FLAG_TRYHARD)))
369 bio_set_flags_failfast(bdev, &flags);
373 block_size = vdev_bdev_block_size(bdev);
376 * When the IO size exceeds the maximum bio size for the request
377 * queue we are forced to break the IO in multiple bio's and wait
378 * for them all to complete. Ideally, all pool users will set
379 * their volume block size to match the maximum request size and
380 * the common case will be one bio per vdev IO request.
383 bio_offset = kbuf_offset;
384 bio_size = kbuf_size;
385 for (i = 0; i <= dr->dr_bio_count; i++) {
387 /* Finished constructing bio's for given buffer */
392 * By default only 'bio_count' bio's per dio are allowed.
393 * However, if we find ourselves in a situation where more
394 * are needed we allocate a larger dio and warn the user.
396 if (dr->dr_bio_count == i) {
397 vdev_disk_dio_free(dr);
399 printk("WARNING: Resized bio's/dio to %d\n",bio_count);
403 dr->dr_bio[i] = bio_alloc(GFP_NOIO,
404 bio_nr_pages(bio_ptr, bio_size));
405 if (dr->dr_bio[i] == NULL) {
406 vdev_disk_dio_free(dr);
410 /* Matching put called by vdev_disk_physio_completion */
411 vdev_disk_dio_get(dr);
413 dr->dr_bio[i]->bi_bdev = bdev;
414 dr->dr_bio[i]->bi_sector = bio_offset / block_size;
415 dr->dr_bio[i]->bi_rw = dr->dr_rw;
416 dr->dr_bio[i]->bi_end_io = vdev_disk_physio_completion;
417 dr->dr_bio[i]->bi_private = dr;
419 /* Remaining size is returned to become the new size */
420 bio_size = bio_map(dr->dr_bio[i], bio_ptr, bio_size);
422 /* Advance in buffer and construct another bio if needed */
423 bio_ptr += dr->dr_bio[i]->bi_size;
424 bio_offset += dr->dr_bio[i]->bi_size;
427 /* Extra reference to protect dio_request during submit_bio */
428 vdev_disk_dio_get(dr);
430 zio->io_delay = jiffies_64;
432 /* Submit all bio's associated with this dio */
433 for (i = 0; i < dr->dr_bio_count; i++)
435 submit_bio(dr->dr_rw, dr->dr_bio[i]);
438 * On synchronous blocking requests we wait for all bio the completion
439 * callbacks to run. We will be woken when the last callback runs
440 * for this dio. We are responsible for putting the last dio_request
441 * reference will in turn put back the last bio references. The
442 * only synchronous consumer is vdev_disk_read_rootlabel() all other
443 * IO originating from vdev_disk_io_start() is asynchronous.
445 if (vdev_disk_dio_is_sync(dr)) {
446 wait_for_completion(&dr->dr_comp);
447 error = dr->dr_error;
448 ASSERT3S(atomic_read(&dr->dr_ref), ==, 1);
451 (void)vdev_disk_dio_put(dr);
457 vdev_disk_physio(struct block_device *bdev, caddr_t kbuf,
458 size_t size, uint64_t offset, int flags)
460 bio_set_flags_failfast(bdev, &flags);
461 return __vdev_disk_physio(bdev, NULL, kbuf, size, offset, flags);
464 /* 2.6.24 API change */
465 #ifdef HAVE_BIO_EMPTY_BARRIER
466 BIO_END_IO_PROTO(vdev_disk_io_flush_completion, bio, size, rc)
468 zio_t *zio = bio->bi_private;
470 zio->io_delay = jiffies_to_msecs(jiffies_64 - zio->io_delay);
472 if (rc && (rc == -EOPNOTSUPP))
473 zio->io_vd->vdev_nowritecache = B_TRUE;
476 ASSERT3S(zio->io_error, >=, 0);
478 vdev_disk_error(zio);
481 BIO_END_IO_RETURN(0);
485 vdev_disk_io_flush(struct block_device *bdev, zio_t *zio)
487 struct request_queue *q;
490 q = bdev_get_queue(bdev);
494 bio = bio_alloc(GFP_KERNEL, 0);
498 bio->bi_end_io = vdev_disk_io_flush_completion;
499 bio->bi_private = zio;
501 zio->io_delay = jiffies_64;
502 submit_bio(WRITE_BARRIER, bio);
508 vdev_disk_io_flush(struct block_device *bdev, zio_t *zio)
512 #endif /* HAVE_BIO_EMPTY_BARRIER */
515 vdev_disk_io_start(zio_t *zio)
517 vdev_t *v = zio->io_vd;
518 vdev_disk_t *vd = v->vdev_tsd;
521 switch (zio->io_type) {
524 if (!vdev_readable(v)) {
525 zio->io_error = ENXIO;
526 return ZIO_PIPELINE_CONTINUE;
529 switch (zio->io_cmd) {
530 case DKIOCFLUSHWRITECACHE:
532 if (zfs_nocacheflush)
535 if (v->vdev_nowritecache) {
536 zio->io_error = ENOTSUP;
540 error = vdev_disk_io_flush(vd->vd_bdev, zio);
542 return ZIO_PIPELINE_STOP;
544 zio->io_error = error;
545 if (error == ENOTSUP)
546 v->vdev_nowritecache = B_TRUE;
551 zio->io_error = ENOTSUP;
554 return ZIO_PIPELINE_CONTINUE;
565 zio->io_error = ENOTSUP;
566 return ZIO_PIPELINE_CONTINUE;
569 error = __vdev_disk_physio(vd->vd_bdev, zio, zio->io_data,
570 zio->io_size, zio->io_offset, flags);
572 zio->io_error = error;
573 return ZIO_PIPELINE_CONTINUE;
576 return ZIO_PIPELINE_STOP;
580 vdev_disk_io_done(zio_t *zio)
583 * If the device returned EIO, we revalidate the media. If it is
584 * determined the media has changed this triggers the asynchronous
585 * removal of the device from the configuration.
587 if (zio->io_error == EIO) {
588 vdev_t *v = zio->io_vd;
589 vdev_disk_t *vd = v->vdev_tsd;
591 if (check_disk_change(vd->vd_bdev)) {
592 vdev_bdev_invalidate(vd->vd_bdev);
593 v->vdev_remove_wanted = B_TRUE;
594 spa_async_request(zio->io_spa, SPA_ASYNC_REMOVE);
600 vdev_disk_hold(vdev_t *vd)
602 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
604 /* We must have a pathname, and it must be absolute. */
605 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/')
609 * Only prefetch path and devid info if the device has
612 if (vd->vdev_tsd != NULL)
615 /* XXX: Implement me as a vnode lookup for the device */
616 vd->vdev_name_vp = NULL;
617 vd->vdev_devid_vp = NULL;
621 vdev_disk_rele(vdev_t *vd)
623 ASSERT(spa_config_held(vd->vdev_spa, SCL_STATE, RW_WRITER));
625 /* XXX: Implement me as a vnode rele for the device */
628 vdev_ops_t vdev_disk_ops = {
637 VDEV_TYPE_DISK, /* name of this vdev type */
638 B_TRUE /* leaf vdev */
642 * Given the root disk device devid or pathname, read the label from
643 * the device, and construct a configuration nvlist.
646 vdev_disk_read_rootlabel(char *devpath, char *devid, nvlist_t **config)
648 struct block_device *bdev;
653 bdev = vdev_bdev_open(devpath, vdev_bdev_mode(FREAD), NULL);
655 return -PTR_ERR(bdev);
657 s = bdev_capacity(bdev) * vdev_bdev_block_size(bdev);
659 vdev_bdev_close(bdev, vdev_bdev_mode(FREAD));
663 size = P2ALIGN_TYPED(s, sizeof(vdev_label_t), uint64_t);
664 label = vmem_alloc(sizeof(vdev_label_t), KM_SLEEP);
666 for (i = 0; i < VDEV_LABELS; i++) {
667 uint64_t offset, state, txg = 0;
669 /* read vdev label */
670 offset = vdev_label_offset(size, i, 0);
671 if (vdev_disk_physio(bdev, (caddr_t)label,
672 VDEV_SKIP_SIZE + VDEV_PHYS_SIZE, offset, READ_SYNC) != 0)
675 if (nvlist_unpack(label->vl_vdev_phys.vp_nvlist,
676 sizeof (label->vl_vdev_phys.vp_nvlist), config, 0) != 0) {
681 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_STATE,
682 &state) != 0 || state >= POOL_STATE_DESTROYED) {
683 nvlist_free(*config);
688 if (nvlist_lookup_uint64(*config, ZPOOL_CONFIG_POOL_TXG,
689 &txg) != 0 || txg == 0) {
690 nvlist_free(*config);
698 vmem_free(label, sizeof(vdev_label_t));
699 vdev_bdev_close(bdev, vdev_bdev_mode(FREAD));