2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <linux/raid/pq.h>
29 #include <linux/semaphore.h>
30 #include <asm/div64.h>
32 #include "extent_map.h"
34 #include "transaction.h"
35 #include "print-tree.h"
38 #include "async-thread.h"
39 #include "check-integrity.h"
40 #include "rcu-string.h"
42 #include "dev-replace.h"
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
49 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
50 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev);
51 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
53 DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 static struct btrfs_fs_devices *__alloc_fs_devices(void)
58 struct btrfs_fs_devices *fs_devs;
60 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
62 return ERR_PTR(-ENOMEM);
64 mutex_init(&fs_devs->device_list_mutex);
66 INIT_LIST_HEAD(&fs_devs->devices);
67 INIT_LIST_HEAD(&fs_devs->resized_devices);
68 INIT_LIST_HEAD(&fs_devs->alloc_list);
69 INIT_LIST_HEAD(&fs_devs->list);
75 * alloc_fs_devices - allocate struct btrfs_fs_devices
76 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
79 * Return: a pointer to a new &struct btrfs_fs_devices on success;
80 * ERR_PTR() on error. Returned struct is not linked onto any lists and
81 * can be destroyed with kfree() right away.
83 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
85 struct btrfs_fs_devices *fs_devs;
87 fs_devs = __alloc_fs_devices();
92 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
94 generate_random_uuid(fs_devs->fsid);
99 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
101 struct btrfs_device *device;
102 WARN_ON(fs_devices->opened);
103 while (!list_empty(&fs_devices->devices)) {
104 device = list_entry(fs_devices->devices.next,
105 struct btrfs_device, dev_list);
106 list_del(&device->dev_list);
107 rcu_string_free(device->name);
113 static void btrfs_kobject_uevent(struct block_device *bdev,
114 enum kobject_action action)
118 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
120 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
122 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
123 &disk_to_dev(bdev->bd_disk)->kobj);
126 void btrfs_cleanup_fs_uuids(void)
128 struct btrfs_fs_devices *fs_devices;
130 while (!list_empty(&fs_uuids)) {
131 fs_devices = list_entry(fs_uuids.next,
132 struct btrfs_fs_devices, list);
133 list_del(&fs_devices->list);
134 free_fs_devices(fs_devices);
138 static struct btrfs_device *__alloc_device(void)
140 struct btrfs_device *dev;
142 dev = kzalloc(sizeof(*dev), GFP_NOFS);
144 return ERR_PTR(-ENOMEM);
146 INIT_LIST_HEAD(&dev->dev_list);
147 INIT_LIST_HEAD(&dev->dev_alloc_list);
148 INIT_LIST_HEAD(&dev->resized_list);
150 spin_lock_init(&dev->io_lock);
152 spin_lock_init(&dev->reada_lock);
153 atomic_set(&dev->reada_in_flight, 0);
154 atomic_set(&dev->dev_stats_ccnt, 0);
155 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
156 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
161 static noinline struct btrfs_device *__find_device(struct list_head *head,
164 struct btrfs_device *dev;
166 list_for_each_entry(dev, head, dev_list) {
167 if (dev->devid == devid &&
168 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
175 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
177 struct btrfs_fs_devices *fs_devices;
179 list_for_each_entry(fs_devices, &fs_uuids, list) {
180 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
187 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
188 int flush, struct block_device **bdev,
189 struct buffer_head **bh)
193 *bdev = blkdev_get_by_path(device_path, flags, holder);
196 ret = PTR_ERR(*bdev);
197 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
202 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
203 ret = set_blocksize(*bdev, 4096);
205 blkdev_put(*bdev, flags);
208 invalidate_bdev(*bdev);
209 *bh = btrfs_read_dev_super(*bdev);
212 blkdev_put(*bdev, flags);
224 static void requeue_list(struct btrfs_pending_bios *pending_bios,
225 struct bio *head, struct bio *tail)
228 struct bio *old_head;
230 old_head = pending_bios->head;
231 pending_bios->head = head;
232 if (pending_bios->tail)
233 tail->bi_next = old_head;
235 pending_bios->tail = tail;
239 * we try to collect pending bios for a device so we don't get a large
240 * number of procs sending bios down to the same device. This greatly
241 * improves the schedulers ability to collect and merge the bios.
243 * But, it also turns into a long list of bios to process and that is sure
244 * to eventually make the worker thread block. The solution here is to
245 * make some progress and then put this work struct back at the end of
246 * the list if the block device is congested. This way, multiple devices
247 * can make progress from a single worker thread.
249 static noinline void run_scheduled_bios(struct btrfs_device *device)
252 struct backing_dev_info *bdi;
253 struct btrfs_fs_info *fs_info;
254 struct btrfs_pending_bios *pending_bios;
258 unsigned long num_run;
259 unsigned long batch_run = 0;
261 unsigned long last_waited = 0;
263 int sync_pending = 0;
264 struct blk_plug plug;
267 * this function runs all the bios we've collected for
268 * a particular device. We don't want to wander off to
269 * another device without first sending all of these down.
270 * So, setup a plug here and finish it off before we return
272 blk_start_plug(&plug);
274 bdi = blk_get_backing_dev_info(device->bdev);
275 fs_info = device->dev_root->fs_info;
276 limit = btrfs_async_submit_limit(fs_info);
277 limit = limit * 2 / 3;
280 spin_lock(&device->io_lock);
285 /* take all the bios off the list at once and process them
286 * later on (without the lock held). But, remember the
287 * tail and other pointers so the bios can be properly reinserted
288 * into the list if we hit congestion
290 if (!force_reg && device->pending_sync_bios.head) {
291 pending_bios = &device->pending_sync_bios;
294 pending_bios = &device->pending_bios;
298 pending = pending_bios->head;
299 tail = pending_bios->tail;
300 WARN_ON(pending && !tail);
303 * if pending was null this time around, no bios need processing
304 * at all and we can stop. Otherwise it'll loop back up again
305 * and do an additional check so no bios are missed.
307 * device->running_pending is used to synchronize with the
310 if (device->pending_sync_bios.head == NULL &&
311 device->pending_bios.head == NULL) {
313 device->running_pending = 0;
316 device->running_pending = 1;
319 pending_bios->head = NULL;
320 pending_bios->tail = NULL;
322 spin_unlock(&device->io_lock);
327 /* we want to work on both lists, but do more bios on the
328 * sync list than the regular list
331 pending_bios != &device->pending_sync_bios &&
332 device->pending_sync_bios.head) ||
333 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
334 device->pending_bios.head)) {
335 spin_lock(&device->io_lock);
336 requeue_list(pending_bios, pending, tail);
341 pending = pending->bi_next;
344 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
345 waitqueue_active(&fs_info->async_submit_wait))
346 wake_up(&fs_info->async_submit_wait);
348 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
351 * if we're doing the sync list, record that our
352 * plug has some sync requests on it
354 * If we're doing the regular list and there are
355 * sync requests sitting around, unplug before
358 if (pending_bios == &device->pending_sync_bios) {
360 } else if (sync_pending) {
361 blk_finish_plug(&plug);
362 blk_start_plug(&plug);
366 btrfsic_submit_bio(cur->bi_rw, cur);
373 * we made progress, there is more work to do and the bdi
374 * is now congested. Back off and let other work structs
377 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
378 fs_info->fs_devices->open_devices > 1) {
379 struct io_context *ioc;
381 ioc = current->io_context;
384 * the main goal here is that we don't want to
385 * block if we're going to be able to submit
386 * more requests without blocking.
388 * This code does two great things, it pokes into
389 * the elevator code from a filesystem _and_
390 * it makes assumptions about how batching works.
392 if (ioc && ioc->nr_batch_requests > 0 &&
393 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
395 ioc->last_waited == last_waited)) {
397 * we want to go through our batch of
398 * requests and stop. So, we copy out
399 * the ioc->last_waited time and test
400 * against it before looping
402 last_waited = ioc->last_waited;
407 spin_lock(&device->io_lock);
408 requeue_list(pending_bios, pending, tail);
409 device->running_pending = 1;
411 spin_unlock(&device->io_lock);
412 btrfs_queue_work(fs_info->submit_workers,
416 /* unplug every 64 requests just for good measure */
417 if (batch_run % 64 == 0) {
418 blk_finish_plug(&plug);
419 blk_start_plug(&plug);
428 spin_lock(&device->io_lock);
429 if (device->pending_bios.head || device->pending_sync_bios.head)
431 spin_unlock(&device->io_lock);
434 blk_finish_plug(&plug);
437 static void pending_bios_fn(struct btrfs_work *work)
439 struct btrfs_device *device;
441 device = container_of(work, struct btrfs_device, work);
442 run_scheduled_bios(device);
446 * Add new device to list of registered devices
449 * 1 - first time device is seen
450 * 0 - device already known
453 static noinline int device_list_add(const char *path,
454 struct btrfs_super_block *disk_super,
455 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
457 struct btrfs_device *device;
458 struct btrfs_fs_devices *fs_devices;
459 struct rcu_string *name;
461 u64 found_transid = btrfs_super_generation(disk_super);
463 fs_devices = find_fsid(disk_super->fsid);
465 fs_devices = alloc_fs_devices(disk_super->fsid);
466 if (IS_ERR(fs_devices))
467 return PTR_ERR(fs_devices);
469 list_add(&fs_devices->list, &fs_uuids);
473 device = __find_device(&fs_devices->devices, devid,
474 disk_super->dev_item.uuid);
478 if (fs_devices->opened)
481 device = btrfs_alloc_device(NULL, &devid,
482 disk_super->dev_item.uuid);
483 if (IS_ERR(device)) {
484 /* we can safely leave the fs_devices entry around */
485 return PTR_ERR(device);
488 name = rcu_string_strdup(path, GFP_NOFS);
493 rcu_assign_pointer(device->name, name);
495 mutex_lock(&fs_devices->device_list_mutex);
496 list_add_rcu(&device->dev_list, &fs_devices->devices);
497 fs_devices->num_devices++;
498 mutex_unlock(&fs_devices->device_list_mutex);
501 device->fs_devices = fs_devices;
502 } else if (!device->name || strcmp(device->name->str, path)) {
504 * When FS is already mounted.
505 * 1. If you are here and if the device->name is NULL that
506 * means this device was missing at time of FS mount.
507 * 2. If you are here and if the device->name is different
508 * from 'path' that means either
509 * a. The same device disappeared and reappeared with
511 * b. The missing-disk-which-was-replaced, has
514 * We must allow 1 and 2a above. But 2b would be a spurious
517 * Further in case of 1 and 2a above, the disk at 'path'
518 * would have missed some transaction when it was away and
519 * in case of 2a the stale bdev has to be updated as well.
520 * 2b must not be allowed at all time.
524 * For now, we do allow update to btrfs_fs_device through the
525 * btrfs dev scan cli after FS has been mounted. We're still
526 * tracking a problem where systems fail mount by subvolume id
527 * when we reject replacement on a mounted FS.
529 if (!fs_devices->opened && found_transid < device->generation) {
531 * That is if the FS is _not_ mounted and if you
532 * are here, that means there is more than one
533 * disk with same uuid and devid.We keep the one
534 * with larger generation number or the last-in if
535 * generation are equal.
540 name = rcu_string_strdup(path, GFP_NOFS);
543 rcu_string_free(device->name);
544 rcu_assign_pointer(device->name, name);
545 if (device->missing) {
546 fs_devices->missing_devices--;
552 * Unmount does not free the btrfs_device struct but would zero
553 * generation along with most of the other members. So just update
554 * it back. We need it to pick the disk with largest generation
557 if (!fs_devices->opened)
558 device->generation = found_transid;
560 *fs_devices_ret = fs_devices;
565 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
567 struct btrfs_fs_devices *fs_devices;
568 struct btrfs_device *device;
569 struct btrfs_device *orig_dev;
571 fs_devices = alloc_fs_devices(orig->fsid);
572 if (IS_ERR(fs_devices))
575 mutex_lock(&orig->device_list_mutex);
576 fs_devices->total_devices = orig->total_devices;
578 /* We have held the volume lock, it is safe to get the devices. */
579 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
580 struct rcu_string *name;
582 device = btrfs_alloc_device(NULL, &orig_dev->devid,
588 * This is ok to do without rcu read locked because we hold the
589 * uuid mutex so nothing we touch in here is going to disappear.
591 if (orig_dev->name) {
592 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
597 rcu_assign_pointer(device->name, name);
600 list_add(&device->dev_list, &fs_devices->devices);
601 device->fs_devices = fs_devices;
602 fs_devices->num_devices++;
604 mutex_unlock(&orig->device_list_mutex);
607 mutex_unlock(&orig->device_list_mutex);
608 free_fs_devices(fs_devices);
609 return ERR_PTR(-ENOMEM);
612 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
613 struct btrfs_fs_devices *fs_devices, int step)
615 struct btrfs_device *device, *next;
616 struct btrfs_device *latest_dev = NULL;
618 mutex_lock(&uuid_mutex);
620 /* This is the initialized path, it is safe to release the devices. */
621 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
622 if (device->in_fs_metadata) {
623 if (!device->is_tgtdev_for_dev_replace &&
625 device->generation > latest_dev->generation)) {
631 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
633 * In the first step, keep the device which has
634 * the correct fsid and the devid that is used
635 * for the dev_replace procedure.
636 * In the second step, the dev_replace state is
637 * read from the device tree and it is known
638 * whether the procedure is really active or
639 * not, which means whether this device is
640 * used or whether it should be removed.
642 if (step == 0 || device->is_tgtdev_for_dev_replace) {
647 blkdev_put(device->bdev, device->mode);
649 fs_devices->open_devices--;
651 if (device->writeable) {
652 list_del_init(&device->dev_alloc_list);
653 device->writeable = 0;
654 if (!device->is_tgtdev_for_dev_replace)
655 fs_devices->rw_devices--;
657 list_del_init(&device->dev_list);
658 fs_devices->num_devices--;
659 rcu_string_free(device->name);
663 if (fs_devices->seed) {
664 fs_devices = fs_devices->seed;
668 fs_devices->latest_bdev = latest_dev->bdev;
670 mutex_unlock(&uuid_mutex);
673 static void __free_device(struct work_struct *work)
675 struct btrfs_device *device;
677 device = container_of(work, struct btrfs_device, rcu_work);
680 blkdev_put(device->bdev, device->mode);
682 rcu_string_free(device->name);
686 static void free_device(struct rcu_head *head)
688 struct btrfs_device *device;
690 device = container_of(head, struct btrfs_device, rcu);
692 INIT_WORK(&device->rcu_work, __free_device);
693 schedule_work(&device->rcu_work);
696 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
698 struct btrfs_device *device;
700 if (--fs_devices->opened > 0)
703 mutex_lock(&fs_devices->device_list_mutex);
704 list_for_each_entry(device, &fs_devices->devices, dev_list) {
705 struct btrfs_device *new_device;
706 struct rcu_string *name;
709 fs_devices->open_devices--;
711 if (device->writeable &&
712 device->devid != BTRFS_DEV_REPLACE_DEVID) {
713 list_del_init(&device->dev_alloc_list);
714 fs_devices->rw_devices--;
718 fs_devices->missing_devices--;
720 new_device = btrfs_alloc_device(NULL, &device->devid,
722 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
724 /* Safe because we are under uuid_mutex */
726 name = rcu_string_strdup(device->name->str, GFP_NOFS);
727 BUG_ON(!name); /* -ENOMEM */
728 rcu_assign_pointer(new_device->name, name);
731 list_replace_rcu(&device->dev_list, &new_device->dev_list);
732 new_device->fs_devices = device->fs_devices;
734 call_rcu(&device->rcu, free_device);
736 mutex_unlock(&fs_devices->device_list_mutex);
738 WARN_ON(fs_devices->open_devices);
739 WARN_ON(fs_devices->rw_devices);
740 fs_devices->opened = 0;
741 fs_devices->seeding = 0;
746 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
748 struct btrfs_fs_devices *seed_devices = NULL;
751 mutex_lock(&uuid_mutex);
752 ret = __btrfs_close_devices(fs_devices);
753 if (!fs_devices->opened) {
754 seed_devices = fs_devices->seed;
755 fs_devices->seed = NULL;
757 mutex_unlock(&uuid_mutex);
759 while (seed_devices) {
760 fs_devices = seed_devices;
761 seed_devices = fs_devices->seed;
762 __btrfs_close_devices(fs_devices);
763 free_fs_devices(fs_devices);
766 * Wait for rcu kworkers under __btrfs_close_devices
767 * to finish all blkdev_puts so device is really
768 * free when umount is done.
774 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
775 fmode_t flags, void *holder)
777 struct request_queue *q;
778 struct block_device *bdev;
779 struct list_head *head = &fs_devices->devices;
780 struct btrfs_device *device;
781 struct btrfs_device *latest_dev = NULL;
782 struct buffer_head *bh;
783 struct btrfs_super_block *disk_super;
790 list_for_each_entry(device, head, dev_list) {
796 /* Just open everything we can; ignore failures here */
797 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
801 disk_super = (struct btrfs_super_block *)bh->b_data;
802 devid = btrfs_stack_device_id(&disk_super->dev_item);
803 if (devid != device->devid)
806 if (memcmp(device->uuid, disk_super->dev_item.uuid,
810 device->generation = btrfs_super_generation(disk_super);
812 device->generation > latest_dev->generation)
815 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
816 device->writeable = 0;
818 device->writeable = !bdev_read_only(bdev);
822 q = bdev_get_queue(bdev);
823 if (blk_queue_discard(q))
824 device->can_discard = 1;
827 device->in_fs_metadata = 0;
828 device->mode = flags;
830 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
831 fs_devices->rotating = 1;
833 fs_devices->open_devices++;
834 if (device->writeable &&
835 device->devid != BTRFS_DEV_REPLACE_DEVID) {
836 fs_devices->rw_devices++;
837 list_add(&device->dev_alloc_list,
838 &fs_devices->alloc_list);
845 blkdev_put(bdev, flags);
848 if (fs_devices->open_devices == 0) {
852 fs_devices->seeding = seeding;
853 fs_devices->opened = 1;
854 fs_devices->latest_bdev = latest_dev->bdev;
855 fs_devices->total_rw_bytes = 0;
860 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
861 fmode_t flags, void *holder)
865 mutex_lock(&uuid_mutex);
866 if (fs_devices->opened) {
867 fs_devices->opened++;
870 ret = __btrfs_open_devices(fs_devices, flags, holder);
872 mutex_unlock(&uuid_mutex);
877 * Look for a btrfs signature on a device. This may be called out of the mount path
878 * and we are not allowed to call set_blocksize during the scan. The superblock
879 * is read via pagecache
881 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
882 struct btrfs_fs_devices **fs_devices_ret)
884 struct btrfs_super_block *disk_super;
885 struct block_device *bdev;
896 * we would like to check all the supers, but that would make
897 * a btrfs mount succeed after a mkfs from a different FS.
898 * So, we need to add a special mount option to scan for
899 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
901 bytenr = btrfs_sb_offset(0);
903 mutex_lock(&uuid_mutex);
905 bdev = blkdev_get_by_path(path, flags, holder);
912 /* make sure our super fits in the device */
913 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
916 /* make sure our super fits in the page */
917 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
920 /* make sure our super doesn't straddle pages on disk */
921 index = bytenr >> PAGE_CACHE_SHIFT;
922 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
925 /* pull in the page with our super */
926 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
929 if (IS_ERR_OR_NULL(page))
934 /* align our pointer to the offset of the super block */
935 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
937 if (btrfs_super_bytenr(disk_super) != bytenr ||
938 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
941 devid = btrfs_stack_device_id(&disk_super->dev_item);
942 transid = btrfs_super_generation(disk_super);
943 total_devices = btrfs_super_num_devices(disk_super);
945 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
947 if (disk_super->label[0]) {
948 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
949 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
950 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
952 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
955 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
958 if (!ret && fs_devices_ret)
959 (*fs_devices_ret)->total_devices = total_devices;
963 page_cache_release(page);
966 blkdev_put(bdev, flags);
968 mutex_unlock(&uuid_mutex);
972 /* helper to account the used device space in the range */
973 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
974 u64 end, u64 *length)
976 struct btrfs_key key;
977 struct btrfs_root *root = device->dev_root;
978 struct btrfs_dev_extent *dev_extent;
979 struct btrfs_path *path;
983 struct extent_buffer *l;
987 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
990 path = btrfs_alloc_path();
995 key.objectid = device->devid;
997 key.type = BTRFS_DEV_EXTENT_KEY;
999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1003 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1010 slot = path->slots[0];
1011 if (slot >= btrfs_header_nritems(l)) {
1012 ret = btrfs_next_leaf(root, path);
1020 btrfs_item_key_to_cpu(l, &key, slot);
1022 if (key.objectid < device->devid)
1025 if (key.objectid > device->devid)
1028 if (key.type != BTRFS_DEV_EXTENT_KEY)
1031 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1032 extent_end = key.offset + btrfs_dev_extent_length(l,
1034 if (key.offset <= start && extent_end > end) {
1035 *length = end - start + 1;
1037 } else if (key.offset <= start && extent_end > start)
1038 *length += extent_end - start;
1039 else if (key.offset > start && extent_end <= end)
1040 *length += extent_end - key.offset;
1041 else if (key.offset > start && key.offset <= end) {
1042 *length += end - key.offset + 1;
1044 } else if (key.offset > end)
1052 btrfs_free_path(path);
1056 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1057 struct btrfs_device *device,
1058 u64 *start, u64 len)
1060 struct extent_map *em;
1061 struct list_head *search_list = &trans->transaction->pending_chunks;
1065 list_for_each_entry(em, search_list, list) {
1066 struct map_lookup *map;
1069 map = (struct map_lookup *)em->bdev;
1070 for (i = 0; i < map->num_stripes; i++) {
1071 if (map->stripes[i].dev != device)
1073 if (map->stripes[i].physical >= *start + len ||
1074 map->stripes[i].physical + em->orig_block_len <=
1077 *start = map->stripes[i].physical +
1082 if (search_list == &trans->transaction->pending_chunks) {
1083 search_list = &trans->root->fs_info->pinned_chunks;
1092 * find_free_dev_extent - find free space in the specified device
1093 * @device: the device which we search the free space in
1094 * @num_bytes: the size of the free space that we need
1095 * @start: store the start of the free space.
1096 * @len: the size of the free space. that we find, or the size of the max
1097 * free space if we don't find suitable free space
1099 * this uses a pretty simple search, the expectation is that it is
1100 * called very infrequently and that a given device has a small number
1103 * @start is used to store the start of the free space if we find. But if we
1104 * don't find suitable free space, it will be used to store the start position
1105 * of the max free space.
1107 * @len is used to store the size of the free space that we find.
1108 * But if we don't find suitable free space, it is used to store the size of
1109 * the max free space.
1111 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1112 struct btrfs_device *device, u64 num_bytes,
1113 u64 *start, u64 *len)
1115 struct btrfs_key key;
1116 struct btrfs_root *root = device->dev_root;
1117 struct btrfs_dev_extent *dev_extent;
1118 struct btrfs_path *path;
1124 u64 search_end = device->total_bytes;
1127 struct extent_buffer *l;
1129 /* FIXME use last free of some kind */
1131 /* we don't want to overwrite the superblock on the drive,
1132 * so we make sure to start at an offset of at least 1MB
1134 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1136 path = btrfs_alloc_path();
1140 max_hole_start = search_start;
1144 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1150 path->search_commit_root = 1;
1151 path->skip_locking = 1;
1153 key.objectid = device->devid;
1154 key.offset = search_start;
1155 key.type = BTRFS_DEV_EXTENT_KEY;
1157 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1161 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1168 slot = path->slots[0];
1169 if (slot >= btrfs_header_nritems(l)) {
1170 ret = btrfs_next_leaf(root, path);
1178 btrfs_item_key_to_cpu(l, &key, slot);
1180 if (key.objectid < device->devid)
1183 if (key.objectid > device->devid)
1186 if (key.type != BTRFS_DEV_EXTENT_KEY)
1189 if (key.offset > search_start) {
1190 hole_size = key.offset - search_start;
1193 * Have to check before we set max_hole_start, otherwise
1194 * we could end up sending back this offset anyway.
1196 if (contains_pending_extent(trans, device,
1201 if (hole_size > max_hole_size) {
1202 max_hole_start = search_start;
1203 max_hole_size = hole_size;
1207 * If this free space is greater than which we need,
1208 * it must be the max free space that we have found
1209 * until now, so max_hole_start must point to the start
1210 * of this free space and the length of this free space
1211 * is stored in max_hole_size. Thus, we return
1212 * max_hole_start and max_hole_size and go back to the
1215 if (hole_size >= num_bytes) {
1221 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1222 extent_end = key.offset + btrfs_dev_extent_length(l,
1224 if (extent_end > search_start)
1225 search_start = extent_end;
1232 * At this point, search_start should be the end of
1233 * allocated dev extents, and when shrinking the device,
1234 * search_end may be smaller than search_start.
1236 if (search_end > search_start)
1237 hole_size = search_end - search_start;
1239 if (hole_size > max_hole_size) {
1240 max_hole_start = search_start;
1241 max_hole_size = hole_size;
1244 if (contains_pending_extent(trans, device, &search_start, hole_size)) {
1245 btrfs_release_path(path);
1250 if (hole_size < num_bytes)
1256 btrfs_free_path(path);
1257 *start = max_hole_start;
1259 *len = max_hole_size;
1263 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1264 struct btrfs_device *device,
1265 u64 start, u64 *dev_extent_len)
1268 struct btrfs_path *path;
1269 struct btrfs_root *root = device->dev_root;
1270 struct btrfs_key key;
1271 struct btrfs_key found_key;
1272 struct extent_buffer *leaf = NULL;
1273 struct btrfs_dev_extent *extent = NULL;
1275 path = btrfs_alloc_path();
1279 key.objectid = device->devid;
1281 key.type = BTRFS_DEV_EXTENT_KEY;
1283 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1285 ret = btrfs_previous_item(root, path, key.objectid,
1286 BTRFS_DEV_EXTENT_KEY);
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1291 extent = btrfs_item_ptr(leaf, path->slots[0],
1292 struct btrfs_dev_extent);
1293 BUG_ON(found_key.offset > start || found_key.offset +
1294 btrfs_dev_extent_length(leaf, extent) < start);
1296 btrfs_release_path(path);
1298 } else if (ret == 0) {
1299 leaf = path->nodes[0];
1300 extent = btrfs_item_ptr(leaf, path->slots[0],
1301 struct btrfs_dev_extent);
1303 btrfs_error(root->fs_info, ret, "Slot search failed");
1307 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1309 ret = btrfs_del_item(trans, root, path);
1311 btrfs_error(root->fs_info, ret,
1312 "Failed to remove dev extent item");
1315 btrfs_free_path(path);
1319 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1320 struct btrfs_device *device,
1321 u64 chunk_tree, u64 chunk_objectid,
1322 u64 chunk_offset, u64 start, u64 num_bytes)
1325 struct btrfs_path *path;
1326 struct btrfs_root *root = device->dev_root;
1327 struct btrfs_dev_extent *extent;
1328 struct extent_buffer *leaf;
1329 struct btrfs_key key;
1331 WARN_ON(!device->in_fs_metadata);
1332 WARN_ON(device->is_tgtdev_for_dev_replace);
1333 path = btrfs_alloc_path();
1337 key.objectid = device->devid;
1339 key.type = BTRFS_DEV_EXTENT_KEY;
1340 ret = btrfs_insert_empty_item(trans, root, path, &key,
1345 leaf = path->nodes[0];
1346 extent = btrfs_item_ptr(leaf, path->slots[0],
1347 struct btrfs_dev_extent);
1348 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1349 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1350 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1352 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1353 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1355 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1356 btrfs_mark_buffer_dirty(leaf);
1358 btrfs_free_path(path);
1362 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1364 struct extent_map_tree *em_tree;
1365 struct extent_map *em;
1369 em_tree = &fs_info->mapping_tree.map_tree;
1370 read_lock(&em_tree->lock);
1371 n = rb_last(&em_tree->map);
1373 em = rb_entry(n, struct extent_map, rb_node);
1374 ret = em->start + em->len;
1376 read_unlock(&em_tree->lock);
1381 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1385 struct btrfs_key key;
1386 struct btrfs_key found_key;
1387 struct btrfs_path *path;
1389 path = btrfs_alloc_path();
1393 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1394 key.type = BTRFS_DEV_ITEM_KEY;
1395 key.offset = (u64)-1;
1397 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1401 BUG_ON(ret == 0); /* Corruption */
1403 ret = btrfs_previous_item(fs_info->chunk_root, path,
1404 BTRFS_DEV_ITEMS_OBJECTID,
1405 BTRFS_DEV_ITEM_KEY);
1409 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1411 *devid_ret = found_key.offset + 1;
1415 btrfs_free_path(path);
1420 * the device information is stored in the chunk root
1421 * the btrfs_device struct should be fully filled in
1423 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *root,
1425 struct btrfs_device *device)
1428 struct btrfs_path *path;
1429 struct btrfs_dev_item *dev_item;
1430 struct extent_buffer *leaf;
1431 struct btrfs_key key;
1434 root = root->fs_info->chunk_root;
1436 path = btrfs_alloc_path();
1440 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1441 key.type = BTRFS_DEV_ITEM_KEY;
1442 key.offset = device->devid;
1444 ret = btrfs_insert_empty_item(trans, root, path, &key,
1449 leaf = path->nodes[0];
1450 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1452 btrfs_set_device_id(leaf, dev_item, device->devid);
1453 btrfs_set_device_generation(leaf, dev_item, 0);
1454 btrfs_set_device_type(leaf, dev_item, device->type);
1455 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1456 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1457 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1458 btrfs_set_device_total_bytes(leaf, dev_item,
1459 btrfs_device_get_disk_total_bytes(device));
1460 btrfs_set_device_bytes_used(leaf, dev_item,
1461 btrfs_device_get_bytes_used(device));
1462 btrfs_set_device_group(leaf, dev_item, 0);
1463 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1464 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1465 btrfs_set_device_start_offset(leaf, dev_item, 0);
1467 ptr = btrfs_device_uuid(dev_item);
1468 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1469 ptr = btrfs_device_fsid(dev_item);
1470 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1471 btrfs_mark_buffer_dirty(leaf);
1475 btrfs_free_path(path);
1480 * Function to update ctime/mtime for a given device path.
1481 * Mainly used for ctime/mtime based probe like libblkid.
1483 static void update_dev_time(char *path_name)
1487 filp = filp_open(path_name, O_RDWR, 0);
1490 file_update_time(filp);
1491 filp_close(filp, NULL);
1495 static int btrfs_rm_dev_item(struct btrfs_root *root,
1496 struct btrfs_device *device)
1499 struct btrfs_path *path;
1500 struct btrfs_key key;
1501 struct btrfs_trans_handle *trans;
1503 root = root->fs_info->chunk_root;
1505 path = btrfs_alloc_path();
1509 trans = btrfs_start_transaction(root, 0);
1510 if (IS_ERR(trans)) {
1511 btrfs_free_path(path);
1512 return PTR_ERR(trans);
1514 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1515 key.type = BTRFS_DEV_ITEM_KEY;
1516 key.offset = device->devid;
1518 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1527 ret = btrfs_del_item(trans, root, path);
1531 btrfs_free_path(path);
1532 btrfs_commit_transaction(trans, root);
1536 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1538 struct btrfs_device *device;
1539 struct btrfs_device *next_device;
1540 struct block_device *bdev;
1541 struct buffer_head *bh = NULL;
1542 struct btrfs_super_block *disk_super;
1543 struct btrfs_fs_devices *cur_devices;
1550 bool clear_super = false;
1552 mutex_lock(&uuid_mutex);
1555 seq = read_seqbegin(&root->fs_info->profiles_lock);
1557 all_avail = root->fs_info->avail_data_alloc_bits |
1558 root->fs_info->avail_system_alloc_bits |
1559 root->fs_info->avail_metadata_alloc_bits;
1560 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1562 num_devices = root->fs_info->fs_devices->num_devices;
1563 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1564 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1565 WARN_ON(num_devices < 1);
1568 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1570 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1571 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1575 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1576 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1580 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1581 root->fs_info->fs_devices->rw_devices <= 2) {
1582 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1585 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1586 root->fs_info->fs_devices->rw_devices <= 3) {
1587 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1591 if (strcmp(device_path, "missing") == 0) {
1592 struct list_head *devices;
1593 struct btrfs_device *tmp;
1596 devices = &root->fs_info->fs_devices->devices;
1598 * It is safe to read the devices since the volume_mutex
1601 list_for_each_entry(tmp, devices, dev_list) {
1602 if (tmp->in_fs_metadata &&
1603 !tmp->is_tgtdev_for_dev_replace &&
1613 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1617 ret = btrfs_get_bdev_and_sb(device_path,
1618 FMODE_WRITE | FMODE_EXCL,
1619 root->fs_info->bdev_holder, 0,
1623 disk_super = (struct btrfs_super_block *)bh->b_data;
1624 devid = btrfs_stack_device_id(&disk_super->dev_item);
1625 dev_uuid = disk_super->dev_item.uuid;
1626 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1634 if (device->is_tgtdev_for_dev_replace) {
1635 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1639 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1640 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1644 if (device->writeable) {
1646 list_del_init(&device->dev_alloc_list);
1647 device->fs_devices->rw_devices--;
1648 unlock_chunks(root);
1652 mutex_unlock(&uuid_mutex);
1653 ret = btrfs_shrink_device(device, 0);
1654 mutex_lock(&uuid_mutex);
1659 * TODO: the superblock still includes this device in its num_devices
1660 * counter although write_all_supers() is not locked out. This
1661 * could give a filesystem state which requires a degraded mount.
1663 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1667 device->in_fs_metadata = 0;
1668 btrfs_scrub_cancel_dev(root->fs_info, device);
1671 * the device list mutex makes sure that we don't change
1672 * the device list while someone else is writing out all
1673 * the device supers. Whoever is writing all supers, should
1674 * lock the device list mutex before getting the number of
1675 * devices in the super block (super_copy). Conversely,
1676 * whoever updates the number of devices in the super block
1677 * (super_copy) should hold the device list mutex.
1680 cur_devices = device->fs_devices;
1681 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1682 list_del_rcu(&device->dev_list);
1684 device->fs_devices->num_devices--;
1685 device->fs_devices->total_devices--;
1687 if (device->missing)
1688 device->fs_devices->missing_devices--;
1690 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1691 struct btrfs_device, dev_list);
1692 if (device->bdev == root->fs_info->sb->s_bdev)
1693 root->fs_info->sb->s_bdev = next_device->bdev;
1694 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1695 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1698 device->fs_devices->open_devices--;
1699 /* remove sysfs entry */
1700 btrfs_kobj_rm_device(root->fs_info, device);
1703 call_rcu(&device->rcu, free_device);
1705 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1706 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1707 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1709 if (cur_devices->open_devices == 0) {
1710 struct btrfs_fs_devices *fs_devices;
1711 fs_devices = root->fs_info->fs_devices;
1712 while (fs_devices) {
1713 if (fs_devices->seed == cur_devices) {
1714 fs_devices->seed = cur_devices->seed;
1717 fs_devices = fs_devices->seed;
1719 cur_devices->seed = NULL;
1720 __btrfs_close_devices(cur_devices);
1721 free_fs_devices(cur_devices);
1724 root->fs_info->num_tolerated_disk_barrier_failures =
1725 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1728 * at this point, the device is zero sized. We want to
1729 * remove it from the devices list and zero out the old super
1731 if (clear_super && disk_super) {
1735 /* make sure this device isn't detected as part of
1738 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1739 set_buffer_dirty(bh);
1740 sync_dirty_buffer(bh);
1742 /* clear the mirror copies of super block on the disk
1743 * being removed, 0th copy is been taken care above and
1744 * the below would take of the rest
1746 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1747 bytenr = btrfs_sb_offset(i);
1748 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1749 i_size_read(bdev->bd_inode))
1753 bh = __bread(bdev, bytenr / 4096,
1754 BTRFS_SUPER_INFO_SIZE);
1758 disk_super = (struct btrfs_super_block *)bh->b_data;
1760 if (btrfs_super_bytenr(disk_super) != bytenr ||
1761 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1764 memset(&disk_super->magic, 0,
1765 sizeof(disk_super->magic));
1766 set_buffer_dirty(bh);
1767 sync_dirty_buffer(bh);
1774 /* Notify udev that device has changed */
1775 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1777 /* Update ctime/mtime for device path for libblkid */
1778 update_dev_time(device_path);
1784 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1786 mutex_unlock(&uuid_mutex);
1789 if (device->writeable) {
1791 list_add(&device->dev_alloc_list,
1792 &root->fs_info->fs_devices->alloc_list);
1793 device->fs_devices->rw_devices++;
1794 unlock_chunks(root);
1799 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1800 struct btrfs_device *srcdev)
1802 struct btrfs_fs_devices *fs_devices;
1804 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1807 * in case of fs with no seed, srcdev->fs_devices will point
1808 * to fs_devices of fs_info. However when the dev being replaced is
1809 * a seed dev it will point to the seed's local fs_devices. In short
1810 * srcdev will have its correct fs_devices in both the cases.
1812 fs_devices = srcdev->fs_devices;
1814 list_del_rcu(&srcdev->dev_list);
1815 list_del_rcu(&srcdev->dev_alloc_list);
1816 fs_devices->num_devices--;
1817 if (srcdev->missing)
1818 fs_devices->missing_devices--;
1820 if (srcdev->writeable) {
1821 fs_devices->rw_devices--;
1822 /* zero out the old super if it is writable */
1823 btrfs_scratch_superblock(srcdev);
1827 fs_devices->open_devices--;
1830 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1831 struct btrfs_device *srcdev)
1833 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1835 call_rcu(&srcdev->rcu, free_device);
1838 * unless fs_devices is seed fs, num_devices shouldn't go
1841 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1843 /* if this is no devs we rather delete the fs_devices */
1844 if (!fs_devices->num_devices) {
1845 struct btrfs_fs_devices *tmp_fs_devices;
1847 tmp_fs_devices = fs_info->fs_devices;
1848 while (tmp_fs_devices) {
1849 if (tmp_fs_devices->seed == fs_devices) {
1850 tmp_fs_devices->seed = fs_devices->seed;
1853 tmp_fs_devices = tmp_fs_devices->seed;
1855 fs_devices->seed = NULL;
1856 __btrfs_close_devices(fs_devices);
1857 free_fs_devices(fs_devices);
1861 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1862 struct btrfs_device *tgtdev)
1864 struct btrfs_device *next_device;
1866 mutex_lock(&uuid_mutex);
1868 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1870 btrfs_scratch_superblock(tgtdev);
1871 fs_info->fs_devices->open_devices--;
1873 fs_info->fs_devices->num_devices--;
1875 next_device = list_entry(fs_info->fs_devices->devices.next,
1876 struct btrfs_device, dev_list);
1877 if (tgtdev->bdev == fs_info->sb->s_bdev)
1878 fs_info->sb->s_bdev = next_device->bdev;
1879 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1880 fs_info->fs_devices->latest_bdev = next_device->bdev;
1881 list_del_rcu(&tgtdev->dev_list);
1883 call_rcu(&tgtdev->rcu, free_device);
1885 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1886 mutex_unlock(&uuid_mutex);
1889 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1890 struct btrfs_device **device)
1893 struct btrfs_super_block *disk_super;
1896 struct block_device *bdev;
1897 struct buffer_head *bh;
1900 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1901 root->fs_info->bdev_holder, 0, &bdev, &bh);
1904 disk_super = (struct btrfs_super_block *)bh->b_data;
1905 devid = btrfs_stack_device_id(&disk_super->dev_item);
1906 dev_uuid = disk_super->dev_item.uuid;
1907 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1912 blkdev_put(bdev, FMODE_READ);
1916 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1918 struct btrfs_device **device)
1921 if (strcmp(device_path, "missing") == 0) {
1922 struct list_head *devices;
1923 struct btrfs_device *tmp;
1925 devices = &root->fs_info->fs_devices->devices;
1927 * It is safe to read the devices since the volume_mutex
1928 * is held by the caller.
1930 list_for_each_entry(tmp, devices, dev_list) {
1931 if (tmp->in_fs_metadata && !tmp->bdev) {
1938 btrfs_err(root->fs_info, "no missing device found");
1944 return btrfs_find_device_by_path(root, device_path, device);
1949 * does all the dirty work required for changing file system's UUID.
1951 static int btrfs_prepare_sprout(struct btrfs_root *root)
1953 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1954 struct btrfs_fs_devices *old_devices;
1955 struct btrfs_fs_devices *seed_devices;
1956 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1957 struct btrfs_device *device;
1960 BUG_ON(!mutex_is_locked(&uuid_mutex));
1961 if (!fs_devices->seeding)
1964 seed_devices = __alloc_fs_devices();
1965 if (IS_ERR(seed_devices))
1966 return PTR_ERR(seed_devices);
1968 old_devices = clone_fs_devices(fs_devices);
1969 if (IS_ERR(old_devices)) {
1970 kfree(seed_devices);
1971 return PTR_ERR(old_devices);
1974 list_add(&old_devices->list, &fs_uuids);
1976 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1977 seed_devices->opened = 1;
1978 INIT_LIST_HEAD(&seed_devices->devices);
1979 INIT_LIST_HEAD(&seed_devices->alloc_list);
1980 mutex_init(&seed_devices->device_list_mutex);
1982 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1983 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1985 list_for_each_entry(device, &seed_devices->devices, dev_list)
1986 device->fs_devices = seed_devices;
1989 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1990 unlock_chunks(root);
1992 fs_devices->seeding = 0;
1993 fs_devices->num_devices = 0;
1994 fs_devices->open_devices = 0;
1995 fs_devices->missing_devices = 0;
1996 fs_devices->rotating = 0;
1997 fs_devices->seed = seed_devices;
1999 generate_random_uuid(fs_devices->fsid);
2000 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2001 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2002 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2004 super_flags = btrfs_super_flags(disk_super) &
2005 ~BTRFS_SUPER_FLAG_SEEDING;
2006 btrfs_set_super_flags(disk_super, super_flags);
2012 * strore the expected generation for seed devices in device items.
2014 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2015 struct btrfs_root *root)
2017 struct btrfs_path *path;
2018 struct extent_buffer *leaf;
2019 struct btrfs_dev_item *dev_item;
2020 struct btrfs_device *device;
2021 struct btrfs_key key;
2022 u8 fs_uuid[BTRFS_UUID_SIZE];
2023 u8 dev_uuid[BTRFS_UUID_SIZE];
2027 path = btrfs_alloc_path();
2031 root = root->fs_info->chunk_root;
2032 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2034 key.type = BTRFS_DEV_ITEM_KEY;
2037 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2041 leaf = path->nodes[0];
2043 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2044 ret = btrfs_next_leaf(root, path);
2049 leaf = path->nodes[0];
2050 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2051 btrfs_release_path(path);
2055 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2056 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2057 key.type != BTRFS_DEV_ITEM_KEY)
2060 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2061 struct btrfs_dev_item);
2062 devid = btrfs_device_id(leaf, dev_item);
2063 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2065 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2067 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2069 BUG_ON(!device); /* Logic error */
2071 if (device->fs_devices->seeding) {
2072 btrfs_set_device_generation(leaf, dev_item,
2073 device->generation);
2074 btrfs_mark_buffer_dirty(leaf);
2082 btrfs_free_path(path);
2086 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2088 struct request_queue *q;
2089 struct btrfs_trans_handle *trans;
2090 struct btrfs_device *device;
2091 struct block_device *bdev;
2092 struct list_head *devices;
2093 struct super_block *sb = root->fs_info->sb;
2094 struct rcu_string *name;
2096 int seeding_dev = 0;
2099 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2102 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2103 root->fs_info->bdev_holder);
2105 return PTR_ERR(bdev);
2107 if (root->fs_info->fs_devices->seeding) {
2109 down_write(&sb->s_umount);
2110 mutex_lock(&uuid_mutex);
2113 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2115 devices = &root->fs_info->fs_devices->devices;
2117 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2118 list_for_each_entry(device, devices, dev_list) {
2119 if (device->bdev == bdev) {
2122 &root->fs_info->fs_devices->device_list_mutex);
2126 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2128 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2129 if (IS_ERR(device)) {
2130 /* we can safely leave the fs_devices entry around */
2131 ret = PTR_ERR(device);
2135 name = rcu_string_strdup(device_path, GFP_NOFS);
2141 rcu_assign_pointer(device->name, name);
2143 trans = btrfs_start_transaction(root, 0);
2144 if (IS_ERR(trans)) {
2145 rcu_string_free(device->name);
2147 ret = PTR_ERR(trans);
2151 q = bdev_get_queue(bdev);
2152 if (blk_queue_discard(q))
2153 device->can_discard = 1;
2154 device->writeable = 1;
2155 device->generation = trans->transid;
2156 device->io_width = root->sectorsize;
2157 device->io_align = root->sectorsize;
2158 device->sector_size = root->sectorsize;
2159 device->total_bytes = i_size_read(bdev->bd_inode);
2160 device->disk_total_bytes = device->total_bytes;
2161 device->commit_total_bytes = device->total_bytes;
2162 device->dev_root = root->fs_info->dev_root;
2163 device->bdev = bdev;
2164 device->in_fs_metadata = 1;
2165 device->is_tgtdev_for_dev_replace = 0;
2166 device->mode = FMODE_EXCL;
2167 device->dev_stats_valid = 1;
2168 set_blocksize(device->bdev, 4096);
2171 sb->s_flags &= ~MS_RDONLY;
2172 ret = btrfs_prepare_sprout(root);
2173 BUG_ON(ret); /* -ENOMEM */
2176 device->fs_devices = root->fs_info->fs_devices;
2178 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2180 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2181 list_add(&device->dev_alloc_list,
2182 &root->fs_info->fs_devices->alloc_list);
2183 root->fs_info->fs_devices->num_devices++;
2184 root->fs_info->fs_devices->open_devices++;
2185 root->fs_info->fs_devices->rw_devices++;
2186 root->fs_info->fs_devices->total_devices++;
2187 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2189 spin_lock(&root->fs_info->free_chunk_lock);
2190 root->fs_info->free_chunk_space += device->total_bytes;
2191 spin_unlock(&root->fs_info->free_chunk_lock);
2193 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2194 root->fs_info->fs_devices->rotating = 1;
2196 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2197 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2198 tmp + device->total_bytes);
2200 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2201 btrfs_set_super_num_devices(root->fs_info->super_copy,
2204 /* add sysfs device entry */
2205 btrfs_kobj_add_device(root->fs_info, device);
2208 * we've got more storage, clear any full flags on the space
2211 btrfs_clear_space_info_full(root->fs_info);
2213 unlock_chunks(root);
2214 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2218 ret = init_first_rw_device(trans, root, device);
2219 unlock_chunks(root);
2221 btrfs_abort_transaction(trans, root, ret);
2226 ret = btrfs_add_device(trans, root, device);
2228 btrfs_abort_transaction(trans, root, ret);
2233 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2235 ret = btrfs_finish_sprout(trans, root);
2237 btrfs_abort_transaction(trans, root, ret);
2241 /* Sprouting would change fsid of the mounted root,
2242 * so rename the fsid on the sysfs
2244 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2245 root->fs_info->fsid);
2246 if (kobject_rename(&root->fs_info->super_kobj, fsid_buf))
2250 root->fs_info->num_tolerated_disk_barrier_failures =
2251 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2252 ret = btrfs_commit_transaction(trans, root);
2255 mutex_unlock(&uuid_mutex);
2256 up_write(&sb->s_umount);
2258 if (ret) /* transaction commit */
2261 ret = btrfs_relocate_sys_chunks(root);
2263 btrfs_error(root->fs_info, ret,
2264 "Failed to relocate sys chunks after "
2265 "device initialization. This can be fixed "
2266 "using the \"btrfs balance\" command.");
2267 trans = btrfs_attach_transaction(root);
2268 if (IS_ERR(trans)) {
2269 if (PTR_ERR(trans) == -ENOENT)
2271 return PTR_ERR(trans);
2273 ret = btrfs_commit_transaction(trans, root);
2276 /* Update ctime/mtime for libblkid */
2277 update_dev_time(device_path);
2281 btrfs_end_transaction(trans, root);
2282 rcu_string_free(device->name);
2283 btrfs_kobj_rm_device(root->fs_info, device);
2286 blkdev_put(bdev, FMODE_EXCL);
2288 mutex_unlock(&uuid_mutex);
2289 up_write(&sb->s_umount);
2294 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2295 struct btrfs_device *srcdev,
2296 struct btrfs_device **device_out)
2298 struct request_queue *q;
2299 struct btrfs_device *device;
2300 struct block_device *bdev;
2301 struct btrfs_fs_info *fs_info = root->fs_info;
2302 struct list_head *devices;
2303 struct rcu_string *name;
2304 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2308 if (fs_info->fs_devices->seeding) {
2309 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2313 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2314 fs_info->bdev_holder);
2316 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2317 return PTR_ERR(bdev);
2320 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2322 devices = &fs_info->fs_devices->devices;
2323 list_for_each_entry(device, devices, dev_list) {
2324 if (device->bdev == bdev) {
2325 btrfs_err(fs_info, "target device is in the filesystem!");
2332 if (i_size_read(bdev->bd_inode) <
2333 btrfs_device_get_total_bytes(srcdev)) {
2334 btrfs_err(fs_info, "target device is smaller than source device!");
2340 device = btrfs_alloc_device(NULL, &devid, NULL);
2341 if (IS_ERR(device)) {
2342 ret = PTR_ERR(device);
2346 name = rcu_string_strdup(device_path, GFP_NOFS);
2352 rcu_assign_pointer(device->name, name);
2354 q = bdev_get_queue(bdev);
2355 if (blk_queue_discard(q))
2356 device->can_discard = 1;
2357 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2358 device->writeable = 1;
2359 device->generation = 0;
2360 device->io_width = root->sectorsize;
2361 device->io_align = root->sectorsize;
2362 device->sector_size = root->sectorsize;
2363 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2364 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2365 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2366 ASSERT(list_empty(&srcdev->resized_list));
2367 device->commit_total_bytes = srcdev->commit_total_bytes;
2368 device->commit_bytes_used = device->bytes_used;
2369 device->dev_root = fs_info->dev_root;
2370 device->bdev = bdev;
2371 device->in_fs_metadata = 1;
2372 device->is_tgtdev_for_dev_replace = 1;
2373 device->mode = FMODE_EXCL;
2374 device->dev_stats_valid = 1;
2375 set_blocksize(device->bdev, 4096);
2376 device->fs_devices = fs_info->fs_devices;
2377 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2378 fs_info->fs_devices->num_devices++;
2379 fs_info->fs_devices->open_devices++;
2380 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2382 *device_out = device;
2386 blkdev_put(bdev, FMODE_EXCL);
2390 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2391 struct btrfs_device *tgtdev)
2393 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2394 tgtdev->io_width = fs_info->dev_root->sectorsize;
2395 tgtdev->io_align = fs_info->dev_root->sectorsize;
2396 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2397 tgtdev->dev_root = fs_info->dev_root;
2398 tgtdev->in_fs_metadata = 1;
2401 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2402 struct btrfs_device *device)
2405 struct btrfs_path *path;
2406 struct btrfs_root *root;
2407 struct btrfs_dev_item *dev_item;
2408 struct extent_buffer *leaf;
2409 struct btrfs_key key;
2411 root = device->dev_root->fs_info->chunk_root;
2413 path = btrfs_alloc_path();
2417 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2418 key.type = BTRFS_DEV_ITEM_KEY;
2419 key.offset = device->devid;
2421 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2430 leaf = path->nodes[0];
2431 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2433 btrfs_set_device_id(leaf, dev_item, device->devid);
2434 btrfs_set_device_type(leaf, dev_item, device->type);
2435 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2436 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2437 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2438 btrfs_set_device_total_bytes(leaf, dev_item,
2439 btrfs_device_get_disk_total_bytes(device));
2440 btrfs_set_device_bytes_used(leaf, dev_item,
2441 btrfs_device_get_bytes_used(device));
2442 btrfs_mark_buffer_dirty(leaf);
2445 btrfs_free_path(path);
2449 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2450 struct btrfs_device *device, u64 new_size)
2452 struct btrfs_super_block *super_copy =
2453 device->dev_root->fs_info->super_copy;
2454 struct btrfs_fs_devices *fs_devices;
2458 if (!device->writeable)
2461 lock_chunks(device->dev_root);
2462 old_total = btrfs_super_total_bytes(super_copy);
2463 diff = new_size - device->total_bytes;
2465 if (new_size <= device->total_bytes ||
2466 device->is_tgtdev_for_dev_replace) {
2467 unlock_chunks(device->dev_root);
2471 fs_devices = device->dev_root->fs_info->fs_devices;
2473 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2474 device->fs_devices->total_rw_bytes += diff;
2476 btrfs_device_set_total_bytes(device, new_size);
2477 btrfs_device_set_disk_total_bytes(device, new_size);
2478 btrfs_clear_space_info_full(device->dev_root->fs_info);
2479 if (list_empty(&device->resized_list))
2480 list_add_tail(&device->resized_list,
2481 &fs_devices->resized_devices);
2482 unlock_chunks(device->dev_root);
2484 return btrfs_update_device(trans, device);
2487 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2488 struct btrfs_root *root,
2489 u64 chunk_tree, u64 chunk_objectid,
2493 struct btrfs_path *path;
2494 struct btrfs_key key;
2496 root = root->fs_info->chunk_root;
2497 path = btrfs_alloc_path();
2501 key.objectid = chunk_objectid;
2502 key.offset = chunk_offset;
2503 key.type = BTRFS_CHUNK_ITEM_KEY;
2505 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2508 else if (ret > 0) { /* Logic error or corruption */
2509 btrfs_error(root->fs_info, -ENOENT,
2510 "Failed lookup while freeing chunk.");
2515 ret = btrfs_del_item(trans, root, path);
2517 btrfs_error(root->fs_info, ret,
2518 "Failed to delete chunk item.");
2520 btrfs_free_path(path);
2524 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2527 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2528 struct btrfs_disk_key *disk_key;
2529 struct btrfs_chunk *chunk;
2536 struct btrfs_key key;
2539 array_size = btrfs_super_sys_array_size(super_copy);
2541 ptr = super_copy->sys_chunk_array;
2544 while (cur < array_size) {
2545 disk_key = (struct btrfs_disk_key *)ptr;
2546 btrfs_disk_key_to_cpu(&key, disk_key);
2548 len = sizeof(*disk_key);
2550 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2551 chunk = (struct btrfs_chunk *)(ptr + len);
2552 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2553 len += btrfs_chunk_item_size(num_stripes);
2558 if (key.objectid == chunk_objectid &&
2559 key.offset == chunk_offset) {
2560 memmove(ptr, ptr + len, array_size - (cur + len));
2562 btrfs_set_super_sys_array_size(super_copy, array_size);
2568 unlock_chunks(root);
2572 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2573 struct btrfs_root *root, u64 chunk_offset)
2575 struct extent_map_tree *em_tree;
2576 struct extent_map *em;
2577 struct btrfs_root *extent_root = root->fs_info->extent_root;
2578 struct map_lookup *map;
2579 u64 dev_extent_len = 0;
2580 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2581 u64 chunk_tree = root->fs_info->chunk_root->objectid;
2585 root = root->fs_info->chunk_root;
2586 em_tree = &root->fs_info->mapping_tree.map_tree;
2588 read_lock(&em_tree->lock);
2589 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2590 read_unlock(&em_tree->lock);
2592 if (!em || em->start > chunk_offset ||
2593 em->start + em->len < chunk_offset) {
2595 * This is a logic error, but we don't want to just rely on the
2596 * user having built with ASSERT enabled, so if ASSERT doens't
2597 * do anything we still error out.
2601 free_extent_map(em);
2604 map = (struct map_lookup *)em->bdev;
2606 for (i = 0; i < map->num_stripes; i++) {
2607 struct btrfs_device *device = map->stripes[i].dev;
2608 ret = btrfs_free_dev_extent(trans, device,
2609 map->stripes[i].physical,
2612 btrfs_abort_transaction(trans, root, ret);
2616 if (device->bytes_used > 0) {
2618 btrfs_device_set_bytes_used(device,
2619 device->bytes_used - dev_extent_len);
2620 spin_lock(&root->fs_info->free_chunk_lock);
2621 root->fs_info->free_chunk_space += dev_extent_len;
2622 spin_unlock(&root->fs_info->free_chunk_lock);
2623 btrfs_clear_space_info_full(root->fs_info);
2624 unlock_chunks(root);
2627 if (map->stripes[i].dev) {
2628 ret = btrfs_update_device(trans, map->stripes[i].dev);
2630 btrfs_abort_transaction(trans, root, ret);
2635 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2638 btrfs_abort_transaction(trans, root, ret);
2642 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2644 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2645 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2647 btrfs_abort_transaction(trans, root, ret);
2652 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2654 btrfs_abort_transaction(trans, extent_root, ret);
2660 free_extent_map(em);
2664 static int btrfs_relocate_chunk(struct btrfs_root *root,
2665 u64 chunk_tree, u64 chunk_objectid,
2668 struct btrfs_root *extent_root;
2669 struct btrfs_trans_handle *trans;
2672 root = root->fs_info->chunk_root;
2673 extent_root = root->fs_info->extent_root;
2675 ret = btrfs_can_relocate(extent_root, chunk_offset);
2679 /* step one, relocate all the extents inside this chunk */
2680 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2684 trans = btrfs_start_transaction(root, 0);
2685 if (IS_ERR(trans)) {
2686 ret = PTR_ERR(trans);
2687 btrfs_std_error(root->fs_info, ret);
2692 * step two, delete the device extents and the
2693 * chunk tree entries
2695 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2696 btrfs_end_transaction(trans, root);
2700 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2702 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2703 struct btrfs_path *path;
2704 struct extent_buffer *leaf;
2705 struct btrfs_chunk *chunk;
2706 struct btrfs_key key;
2707 struct btrfs_key found_key;
2708 u64 chunk_tree = chunk_root->root_key.objectid;
2710 bool retried = false;
2714 path = btrfs_alloc_path();
2719 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2720 key.offset = (u64)-1;
2721 key.type = BTRFS_CHUNK_ITEM_KEY;
2724 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2727 BUG_ON(ret == 0); /* Corruption */
2729 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2736 leaf = path->nodes[0];
2737 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2739 chunk = btrfs_item_ptr(leaf, path->slots[0],
2740 struct btrfs_chunk);
2741 chunk_type = btrfs_chunk_type(leaf, chunk);
2742 btrfs_release_path(path);
2744 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2745 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2754 if (found_key.offset == 0)
2756 key.offset = found_key.offset - 1;
2759 if (failed && !retried) {
2763 } else if (WARN_ON(failed && retried)) {
2767 btrfs_free_path(path);
2771 static int insert_balance_item(struct btrfs_root *root,
2772 struct btrfs_balance_control *bctl)
2774 struct btrfs_trans_handle *trans;
2775 struct btrfs_balance_item *item;
2776 struct btrfs_disk_balance_args disk_bargs;
2777 struct btrfs_path *path;
2778 struct extent_buffer *leaf;
2779 struct btrfs_key key;
2782 path = btrfs_alloc_path();
2786 trans = btrfs_start_transaction(root, 0);
2787 if (IS_ERR(trans)) {
2788 btrfs_free_path(path);
2789 return PTR_ERR(trans);
2792 key.objectid = BTRFS_BALANCE_OBJECTID;
2793 key.type = BTRFS_BALANCE_ITEM_KEY;
2796 ret = btrfs_insert_empty_item(trans, root, path, &key,
2801 leaf = path->nodes[0];
2802 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2804 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2806 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2807 btrfs_set_balance_data(leaf, item, &disk_bargs);
2808 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2809 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2810 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2811 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2813 btrfs_set_balance_flags(leaf, item, bctl->flags);
2815 btrfs_mark_buffer_dirty(leaf);
2817 btrfs_free_path(path);
2818 err = btrfs_commit_transaction(trans, root);
2824 static int del_balance_item(struct btrfs_root *root)
2826 struct btrfs_trans_handle *trans;
2827 struct btrfs_path *path;
2828 struct btrfs_key key;
2831 path = btrfs_alloc_path();
2835 trans = btrfs_start_transaction(root, 0);
2836 if (IS_ERR(trans)) {
2837 btrfs_free_path(path);
2838 return PTR_ERR(trans);
2841 key.objectid = BTRFS_BALANCE_OBJECTID;
2842 key.type = BTRFS_BALANCE_ITEM_KEY;
2845 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2853 ret = btrfs_del_item(trans, root, path);
2855 btrfs_free_path(path);
2856 err = btrfs_commit_transaction(trans, root);
2863 * This is a heuristic used to reduce the number of chunks balanced on
2864 * resume after balance was interrupted.
2866 static void update_balance_args(struct btrfs_balance_control *bctl)
2869 * Turn on soft mode for chunk types that were being converted.
2871 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2872 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2873 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2874 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2875 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2876 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2879 * Turn on usage filter if is not already used. The idea is
2880 * that chunks that we have already balanced should be
2881 * reasonably full. Don't do it for chunks that are being
2882 * converted - that will keep us from relocating unconverted
2883 * (albeit full) chunks.
2885 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2886 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2887 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2888 bctl->data.usage = 90;
2890 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2891 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2892 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2893 bctl->sys.usage = 90;
2895 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2896 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2897 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2898 bctl->meta.usage = 90;
2903 * Should be called with both balance and volume mutexes held to
2904 * serialize other volume operations (add_dev/rm_dev/resize) with
2905 * restriper. Same goes for unset_balance_control.
2907 static void set_balance_control(struct btrfs_balance_control *bctl)
2909 struct btrfs_fs_info *fs_info = bctl->fs_info;
2911 BUG_ON(fs_info->balance_ctl);
2913 spin_lock(&fs_info->balance_lock);
2914 fs_info->balance_ctl = bctl;
2915 spin_unlock(&fs_info->balance_lock);
2918 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2920 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2922 BUG_ON(!fs_info->balance_ctl);
2924 spin_lock(&fs_info->balance_lock);
2925 fs_info->balance_ctl = NULL;
2926 spin_unlock(&fs_info->balance_lock);
2932 * Balance filters. Return 1 if chunk should be filtered out
2933 * (should not be balanced).
2935 static int chunk_profiles_filter(u64 chunk_type,
2936 struct btrfs_balance_args *bargs)
2938 chunk_type = chunk_to_extended(chunk_type) &
2939 BTRFS_EXTENDED_PROFILE_MASK;
2941 if (bargs->profiles & chunk_type)
2947 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2948 struct btrfs_balance_args *bargs)
2950 struct btrfs_block_group_cache *cache;
2951 u64 chunk_used, user_thresh;
2954 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2955 chunk_used = btrfs_block_group_used(&cache->item);
2957 if (bargs->usage == 0)
2959 else if (bargs->usage > 100)
2960 user_thresh = cache->key.offset;
2962 user_thresh = div_factor_fine(cache->key.offset,
2965 if (chunk_used < user_thresh)
2968 btrfs_put_block_group(cache);
2972 static int chunk_devid_filter(struct extent_buffer *leaf,
2973 struct btrfs_chunk *chunk,
2974 struct btrfs_balance_args *bargs)
2976 struct btrfs_stripe *stripe;
2977 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2980 for (i = 0; i < num_stripes; i++) {
2981 stripe = btrfs_stripe_nr(chunk, i);
2982 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2989 /* [pstart, pend) */
2990 static int chunk_drange_filter(struct extent_buffer *leaf,
2991 struct btrfs_chunk *chunk,
2993 struct btrfs_balance_args *bargs)
2995 struct btrfs_stripe *stripe;
2996 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3002 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3005 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3006 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3007 factor = num_stripes / 2;
3008 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3009 factor = num_stripes - 1;
3010 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3011 factor = num_stripes - 2;
3013 factor = num_stripes;
3016 for (i = 0; i < num_stripes; i++) {
3017 stripe = btrfs_stripe_nr(chunk, i);
3018 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3021 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3022 stripe_length = btrfs_chunk_length(leaf, chunk);
3023 do_div(stripe_length, factor);
3025 if (stripe_offset < bargs->pend &&
3026 stripe_offset + stripe_length > bargs->pstart)
3033 /* [vstart, vend) */
3034 static int chunk_vrange_filter(struct extent_buffer *leaf,
3035 struct btrfs_chunk *chunk,
3037 struct btrfs_balance_args *bargs)
3039 if (chunk_offset < bargs->vend &&
3040 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3041 /* at least part of the chunk is inside this vrange */
3047 static int chunk_soft_convert_filter(u64 chunk_type,
3048 struct btrfs_balance_args *bargs)
3050 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3053 chunk_type = chunk_to_extended(chunk_type) &
3054 BTRFS_EXTENDED_PROFILE_MASK;
3056 if (bargs->target == chunk_type)
3062 static int should_balance_chunk(struct btrfs_root *root,
3063 struct extent_buffer *leaf,
3064 struct btrfs_chunk *chunk, u64 chunk_offset)
3066 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3067 struct btrfs_balance_args *bargs = NULL;
3068 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3071 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3072 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3076 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3077 bargs = &bctl->data;
3078 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3080 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3081 bargs = &bctl->meta;
3083 /* profiles filter */
3084 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3085 chunk_profiles_filter(chunk_type, bargs)) {
3090 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3091 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3096 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3097 chunk_devid_filter(leaf, chunk, bargs)) {
3101 /* drange filter, makes sense only with devid filter */
3102 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3103 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3108 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3109 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3113 /* soft profile changing mode */
3114 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3115 chunk_soft_convert_filter(chunk_type, bargs)) {
3120 * limited by count, must be the last filter
3122 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3123 if (bargs->limit == 0)
3132 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3134 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3135 struct btrfs_root *chunk_root = fs_info->chunk_root;
3136 struct btrfs_root *dev_root = fs_info->dev_root;
3137 struct list_head *devices;
3138 struct btrfs_device *device;
3141 struct btrfs_chunk *chunk;
3142 struct btrfs_path *path;
3143 struct btrfs_key key;
3144 struct btrfs_key found_key;
3145 struct btrfs_trans_handle *trans;
3146 struct extent_buffer *leaf;
3149 int enospc_errors = 0;
3150 bool counting = true;
3151 u64 limit_data = bctl->data.limit;
3152 u64 limit_meta = bctl->meta.limit;
3153 u64 limit_sys = bctl->sys.limit;
3155 /* step one make some room on all the devices */
3156 devices = &fs_info->fs_devices->devices;
3157 list_for_each_entry(device, devices, dev_list) {
3158 old_size = btrfs_device_get_total_bytes(device);
3159 size_to_free = div_factor(old_size, 1);
3160 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3161 if (!device->writeable ||
3162 btrfs_device_get_total_bytes(device) -
3163 btrfs_device_get_bytes_used(device) > size_to_free ||
3164 device->is_tgtdev_for_dev_replace)
3167 ret = btrfs_shrink_device(device, old_size - size_to_free);
3172 trans = btrfs_start_transaction(dev_root, 0);
3173 BUG_ON(IS_ERR(trans));
3175 ret = btrfs_grow_device(trans, device, old_size);
3178 btrfs_end_transaction(trans, dev_root);
3181 /* step two, relocate all the chunks */
3182 path = btrfs_alloc_path();
3188 /* zero out stat counters */
3189 spin_lock(&fs_info->balance_lock);
3190 memset(&bctl->stat, 0, sizeof(bctl->stat));
3191 spin_unlock(&fs_info->balance_lock);
3194 bctl->data.limit = limit_data;
3195 bctl->meta.limit = limit_meta;
3196 bctl->sys.limit = limit_sys;
3198 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3199 key.offset = (u64)-1;
3200 key.type = BTRFS_CHUNK_ITEM_KEY;
3203 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3204 atomic_read(&fs_info->balance_cancel_req)) {
3209 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3214 * this shouldn't happen, it means the last relocate
3218 BUG(); /* FIXME break ? */
3220 ret = btrfs_previous_item(chunk_root, path, 0,
3221 BTRFS_CHUNK_ITEM_KEY);
3227 leaf = path->nodes[0];
3228 slot = path->slots[0];
3229 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3231 if (found_key.objectid != key.objectid)
3234 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3237 spin_lock(&fs_info->balance_lock);
3238 bctl->stat.considered++;
3239 spin_unlock(&fs_info->balance_lock);
3242 ret = should_balance_chunk(chunk_root, leaf, chunk,
3244 btrfs_release_path(path);
3249 spin_lock(&fs_info->balance_lock);
3250 bctl->stat.expected++;
3251 spin_unlock(&fs_info->balance_lock);
3255 ret = btrfs_relocate_chunk(chunk_root,
3256 chunk_root->root_key.objectid,
3259 if (ret && ret != -ENOSPC)
3261 if (ret == -ENOSPC) {
3264 spin_lock(&fs_info->balance_lock);
3265 bctl->stat.completed++;
3266 spin_unlock(&fs_info->balance_lock);
3269 if (found_key.offset == 0)
3271 key.offset = found_key.offset - 1;
3275 btrfs_release_path(path);
3280 btrfs_free_path(path);
3281 if (enospc_errors) {
3282 btrfs_info(fs_info, "%d enospc errors during balance",
3292 * alloc_profile_is_valid - see if a given profile is valid and reduced
3293 * @flags: profile to validate
3294 * @extended: if true @flags is treated as an extended profile
3296 static int alloc_profile_is_valid(u64 flags, int extended)
3298 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3299 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3301 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3303 /* 1) check that all other bits are zeroed */
3307 /* 2) see if profile is reduced */
3309 return !extended; /* "0" is valid for usual profiles */
3311 /* true if exactly one bit set */
3312 return (flags & (flags - 1)) == 0;
3315 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3317 /* cancel requested || normal exit path */
3318 return atomic_read(&fs_info->balance_cancel_req) ||
3319 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3320 atomic_read(&fs_info->balance_cancel_req) == 0);
3323 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3327 unset_balance_control(fs_info);
3328 ret = del_balance_item(fs_info->tree_root);
3330 btrfs_std_error(fs_info, ret);
3332 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3336 * Should be called with both balance and volume mutexes held
3338 int btrfs_balance(struct btrfs_balance_control *bctl,
3339 struct btrfs_ioctl_balance_args *bargs)
3341 struct btrfs_fs_info *fs_info = bctl->fs_info;
3348 if (btrfs_fs_closing(fs_info) ||
3349 atomic_read(&fs_info->balance_pause_req) ||
3350 atomic_read(&fs_info->balance_cancel_req)) {
3355 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3356 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3360 * In case of mixed groups both data and meta should be picked,
3361 * and identical options should be given for both of them.
3363 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3364 if (mixed && (bctl->flags & allowed)) {
3365 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3366 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3367 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3368 btrfs_err(fs_info, "with mixed groups data and "
3369 "metadata balance options must be the same");
3375 num_devices = fs_info->fs_devices->num_devices;
3376 btrfs_dev_replace_lock(&fs_info->dev_replace);
3377 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3378 BUG_ON(num_devices < 1);
3381 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3382 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3383 if (num_devices == 1)
3384 allowed |= BTRFS_BLOCK_GROUP_DUP;
3385 else if (num_devices > 1)
3386 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3387 if (num_devices > 2)
3388 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3389 if (num_devices > 3)
3390 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3391 BTRFS_BLOCK_GROUP_RAID6);
3392 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3393 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3394 (bctl->data.target & ~allowed))) {
3395 btrfs_err(fs_info, "unable to start balance with target "
3396 "data profile %llu",
3401 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3402 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3403 (bctl->meta.target & ~allowed))) {
3405 "unable to start balance with target metadata profile %llu",
3410 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3411 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3412 (bctl->sys.target & ~allowed))) {
3414 "unable to start balance with target system profile %llu",
3420 /* allow dup'ed data chunks only in mixed mode */
3421 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3422 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3423 btrfs_err(fs_info, "dup for data is not allowed");
3428 /* allow to reduce meta or sys integrity only if force set */
3429 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3430 BTRFS_BLOCK_GROUP_RAID10 |
3431 BTRFS_BLOCK_GROUP_RAID5 |
3432 BTRFS_BLOCK_GROUP_RAID6;
3434 seq = read_seqbegin(&fs_info->profiles_lock);
3436 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3437 (fs_info->avail_system_alloc_bits & allowed) &&
3438 !(bctl->sys.target & allowed)) ||
3439 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3440 (fs_info->avail_metadata_alloc_bits & allowed) &&
3441 !(bctl->meta.target & allowed))) {
3442 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3443 btrfs_info(fs_info, "force reducing metadata integrity");
3445 btrfs_err(fs_info, "balance will reduce metadata "
3446 "integrity, use force if you want this");
3451 } while (read_seqretry(&fs_info->profiles_lock, seq));
3453 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3454 int num_tolerated_disk_barrier_failures;
3455 u64 target = bctl->sys.target;
3457 num_tolerated_disk_barrier_failures =
3458 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3459 if (num_tolerated_disk_barrier_failures > 0 &&
3461 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3462 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3463 num_tolerated_disk_barrier_failures = 0;
3464 else if (num_tolerated_disk_barrier_failures > 1 &&
3466 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3467 num_tolerated_disk_barrier_failures = 1;
3469 fs_info->num_tolerated_disk_barrier_failures =
3470 num_tolerated_disk_barrier_failures;
3473 ret = insert_balance_item(fs_info->tree_root, bctl);
3474 if (ret && ret != -EEXIST)
3477 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3478 BUG_ON(ret == -EEXIST);
3479 set_balance_control(bctl);
3481 BUG_ON(ret != -EEXIST);
3482 spin_lock(&fs_info->balance_lock);
3483 update_balance_args(bctl);
3484 spin_unlock(&fs_info->balance_lock);
3487 atomic_inc(&fs_info->balance_running);
3488 mutex_unlock(&fs_info->balance_mutex);
3490 ret = __btrfs_balance(fs_info);
3492 mutex_lock(&fs_info->balance_mutex);
3493 atomic_dec(&fs_info->balance_running);
3495 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3496 fs_info->num_tolerated_disk_barrier_failures =
3497 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3501 memset(bargs, 0, sizeof(*bargs));
3502 update_ioctl_balance_args(fs_info, 0, bargs);
3505 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3506 balance_need_close(fs_info)) {
3507 __cancel_balance(fs_info);
3510 wake_up(&fs_info->balance_wait_q);
3514 if (bctl->flags & BTRFS_BALANCE_RESUME)
3515 __cancel_balance(fs_info);
3518 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3523 static int balance_kthread(void *data)
3525 struct btrfs_fs_info *fs_info = data;
3528 mutex_lock(&fs_info->volume_mutex);
3529 mutex_lock(&fs_info->balance_mutex);
3531 if (fs_info->balance_ctl) {
3532 btrfs_info(fs_info, "continuing balance");
3533 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3536 mutex_unlock(&fs_info->balance_mutex);
3537 mutex_unlock(&fs_info->volume_mutex);
3542 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3544 struct task_struct *tsk;
3546 spin_lock(&fs_info->balance_lock);
3547 if (!fs_info->balance_ctl) {
3548 spin_unlock(&fs_info->balance_lock);
3551 spin_unlock(&fs_info->balance_lock);
3553 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3554 btrfs_info(fs_info, "force skipping balance");
3558 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3559 return PTR_ERR_OR_ZERO(tsk);
3562 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3564 struct btrfs_balance_control *bctl;
3565 struct btrfs_balance_item *item;
3566 struct btrfs_disk_balance_args disk_bargs;
3567 struct btrfs_path *path;
3568 struct extent_buffer *leaf;
3569 struct btrfs_key key;
3572 path = btrfs_alloc_path();
3576 key.objectid = BTRFS_BALANCE_OBJECTID;
3577 key.type = BTRFS_BALANCE_ITEM_KEY;
3580 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3583 if (ret > 0) { /* ret = -ENOENT; */
3588 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3594 leaf = path->nodes[0];
3595 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3597 bctl->fs_info = fs_info;
3598 bctl->flags = btrfs_balance_flags(leaf, item);
3599 bctl->flags |= BTRFS_BALANCE_RESUME;
3601 btrfs_balance_data(leaf, item, &disk_bargs);
3602 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3603 btrfs_balance_meta(leaf, item, &disk_bargs);
3604 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3605 btrfs_balance_sys(leaf, item, &disk_bargs);
3606 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3608 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3610 mutex_lock(&fs_info->volume_mutex);
3611 mutex_lock(&fs_info->balance_mutex);
3613 set_balance_control(bctl);
3615 mutex_unlock(&fs_info->balance_mutex);
3616 mutex_unlock(&fs_info->volume_mutex);
3618 btrfs_free_path(path);
3622 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3626 mutex_lock(&fs_info->balance_mutex);
3627 if (!fs_info->balance_ctl) {
3628 mutex_unlock(&fs_info->balance_mutex);
3632 if (atomic_read(&fs_info->balance_running)) {
3633 atomic_inc(&fs_info->balance_pause_req);
3634 mutex_unlock(&fs_info->balance_mutex);
3636 wait_event(fs_info->balance_wait_q,
3637 atomic_read(&fs_info->balance_running) == 0);
3639 mutex_lock(&fs_info->balance_mutex);
3640 /* we are good with balance_ctl ripped off from under us */
3641 BUG_ON(atomic_read(&fs_info->balance_running));
3642 atomic_dec(&fs_info->balance_pause_req);
3647 mutex_unlock(&fs_info->balance_mutex);
3651 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3653 if (fs_info->sb->s_flags & MS_RDONLY)
3656 mutex_lock(&fs_info->balance_mutex);
3657 if (!fs_info->balance_ctl) {
3658 mutex_unlock(&fs_info->balance_mutex);
3662 atomic_inc(&fs_info->balance_cancel_req);
3664 * if we are running just wait and return, balance item is
3665 * deleted in btrfs_balance in this case
3667 if (atomic_read(&fs_info->balance_running)) {
3668 mutex_unlock(&fs_info->balance_mutex);
3669 wait_event(fs_info->balance_wait_q,
3670 atomic_read(&fs_info->balance_running) == 0);
3671 mutex_lock(&fs_info->balance_mutex);
3673 /* __cancel_balance needs volume_mutex */
3674 mutex_unlock(&fs_info->balance_mutex);
3675 mutex_lock(&fs_info->volume_mutex);
3676 mutex_lock(&fs_info->balance_mutex);
3678 if (fs_info->balance_ctl)
3679 __cancel_balance(fs_info);
3681 mutex_unlock(&fs_info->volume_mutex);
3684 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3685 atomic_dec(&fs_info->balance_cancel_req);
3686 mutex_unlock(&fs_info->balance_mutex);
3690 static int btrfs_uuid_scan_kthread(void *data)
3692 struct btrfs_fs_info *fs_info = data;
3693 struct btrfs_root *root = fs_info->tree_root;
3694 struct btrfs_key key;
3695 struct btrfs_key max_key;
3696 struct btrfs_path *path = NULL;
3698 struct extent_buffer *eb;
3700 struct btrfs_root_item root_item;
3702 struct btrfs_trans_handle *trans = NULL;
3704 path = btrfs_alloc_path();
3711 key.type = BTRFS_ROOT_ITEM_KEY;
3714 max_key.objectid = (u64)-1;
3715 max_key.type = BTRFS_ROOT_ITEM_KEY;
3716 max_key.offset = (u64)-1;
3719 ret = btrfs_search_forward(root, &key, path, 0);
3726 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3727 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3728 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3729 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3732 eb = path->nodes[0];
3733 slot = path->slots[0];
3734 item_size = btrfs_item_size_nr(eb, slot);
3735 if (item_size < sizeof(root_item))
3738 read_extent_buffer(eb, &root_item,
3739 btrfs_item_ptr_offset(eb, slot),
3740 (int)sizeof(root_item));
3741 if (btrfs_root_refs(&root_item) == 0)
3744 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3745 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3749 btrfs_release_path(path);
3751 * 1 - subvol uuid item
3752 * 1 - received_subvol uuid item
3754 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3755 if (IS_ERR(trans)) {
3756 ret = PTR_ERR(trans);
3764 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3765 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3767 BTRFS_UUID_KEY_SUBVOL,
3770 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3776 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3777 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3778 root_item.received_uuid,
3779 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3782 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3790 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3796 btrfs_release_path(path);
3797 if (key.offset < (u64)-1) {
3799 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3801 key.type = BTRFS_ROOT_ITEM_KEY;
3802 } else if (key.objectid < (u64)-1) {
3804 key.type = BTRFS_ROOT_ITEM_KEY;
3813 btrfs_free_path(path);
3814 if (trans && !IS_ERR(trans))
3815 btrfs_end_transaction(trans, fs_info->uuid_root);
3817 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3819 fs_info->update_uuid_tree_gen = 1;
3820 up(&fs_info->uuid_tree_rescan_sem);
3825 * Callback for btrfs_uuid_tree_iterate().
3827 * 0 check succeeded, the entry is not outdated.
3828 * < 0 if an error occured.
3829 * > 0 if the check failed, which means the caller shall remove the entry.
3831 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3832 u8 *uuid, u8 type, u64 subid)
3834 struct btrfs_key key;
3836 struct btrfs_root *subvol_root;
3838 if (type != BTRFS_UUID_KEY_SUBVOL &&
3839 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3842 key.objectid = subid;
3843 key.type = BTRFS_ROOT_ITEM_KEY;
3844 key.offset = (u64)-1;
3845 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3846 if (IS_ERR(subvol_root)) {
3847 ret = PTR_ERR(subvol_root);
3854 case BTRFS_UUID_KEY_SUBVOL:
3855 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3858 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3859 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3869 static int btrfs_uuid_rescan_kthread(void *data)
3871 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3875 * 1st step is to iterate through the existing UUID tree and
3876 * to delete all entries that contain outdated data.
3877 * 2nd step is to add all missing entries to the UUID tree.
3879 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
3881 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
3882 up(&fs_info->uuid_tree_rescan_sem);
3885 return btrfs_uuid_scan_kthread(data);
3888 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
3890 struct btrfs_trans_handle *trans;
3891 struct btrfs_root *tree_root = fs_info->tree_root;
3892 struct btrfs_root *uuid_root;
3893 struct task_struct *task;
3900 trans = btrfs_start_transaction(tree_root, 2);
3902 return PTR_ERR(trans);
3904 uuid_root = btrfs_create_tree(trans, fs_info,
3905 BTRFS_UUID_TREE_OBJECTID);
3906 if (IS_ERR(uuid_root)) {
3907 btrfs_abort_transaction(trans, tree_root,
3908 PTR_ERR(uuid_root));
3909 return PTR_ERR(uuid_root);
3912 fs_info->uuid_root = uuid_root;
3914 ret = btrfs_commit_transaction(trans, tree_root);
3918 down(&fs_info->uuid_tree_rescan_sem);
3919 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
3921 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3922 btrfs_warn(fs_info, "failed to start uuid_scan task");
3923 up(&fs_info->uuid_tree_rescan_sem);
3924 return PTR_ERR(task);
3930 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
3932 struct task_struct *task;
3934 down(&fs_info->uuid_tree_rescan_sem);
3935 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
3937 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
3938 btrfs_warn(fs_info, "failed to start uuid_rescan task");
3939 up(&fs_info->uuid_tree_rescan_sem);
3940 return PTR_ERR(task);
3947 * shrinking a device means finding all of the device extents past
3948 * the new size, and then following the back refs to the chunks.
3949 * The chunk relocation code actually frees the device extent
3951 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3953 struct btrfs_trans_handle *trans;
3954 struct btrfs_root *root = device->dev_root;
3955 struct btrfs_dev_extent *dev_extent = NULL;
3956 struct btrfs_path *path;
3964 bool retried = false;
3965 struct extent_buffer *l;
3966 struct btrfs_key key;
3967 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3968 u64 old_total = btrfs_super_total_bytes(super_copy);
3969 u64 old_size = btrfs_device_get_total_bytes(device);
3970 u64 diff = old_size - new_size;
3972 if (device->is_tgtdev_for_dev_replace)
3975 path = btrfs_alloc_path();
3983 btrfs_device_set_total_bytes(device, new_size);
3984 if (device->writeable) {
3985 device->fs_devices->total_rw_bytes -= diff;
3986 spin_lock(&root->fs_info->free_chunk_lock);
3987 root->fs_info->free_chunk_space -= diff;
3988 spin_unlock(&root->fs_info->free_chunk_lock);
3990 unlock_chunks(root);
3993 key.objectid = device->devid;
3994 key.offset = (u64)-1;
3995 key.type = BTRFS_DEV_EXTENT_KEY;
3998 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4002 ret = btrfs_previous_item(root, path, 0, key.type);
4007 btrfs_release_path(path);
4012 slot = path->slots[0];
4013 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4015 if (key.objectid != device->devid) {
4016 btrfs_release_path(path);
4020 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4021 length = btrfs_dev_extent_length(l, dev_extent);
4023 if (key.offset + length <= new_size) {
4024 btrfs_release_path(path);
4028 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
4029 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
4030 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4031 btrfs_release_path(path);
4033 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
4035 if (ret && ret != -ENOSPC)
4039 } while (key.offset-- > 0);
4041 if (failed && !retried) {
4045 } else if (failed && retried) {
4049 btrfs_device_set_total_bytes(device, old_size);
4050 if (device->writeable)
4051 device->fs_devices->total_rw_bytes += diff;
4052 spin_lock(&root->fs_info->free_chunk_lock);
4053 root->fs_info->free_chunk_space += diff;
4054 spin_unlock(&root->fs_info->free_chunk_lock);
4055 unlock_chunks(root);
4059 /* Shrinking succeeded, else we would be at "done". */
4060 trans = btrfs_start_transaction(root, 0);
4061 if (IS_ERR(trans)) {
4062 ret = PTR_ERR(trans);
4067 btrfs_device_set_disk_total_bytes(device, new_size);
4068 if (list_empty(&device->resized_list))
4069 list_add_tail(&device->resized_list,
4070 &root->fs_info->fs_devices->resized_devices);
4072 WARN_ON(diff > old_total);
4073 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4074 unlock_chunks(root);
4076 /* Now btrfs_update_device() will change the on-disk size. */
4077 ret = btrfs_update_device(trans, device);
4078 btrfs_end_transaction(trans, root);
4080 btrfs_free_path(path);
4084 static int btrfs_add_system_chunk(struct btrfs_root *root,
4085 struct btrfs_key *key,
4086 struct btrfs_chunk *chunk, int item_size)
4088 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4089 struct btrfs_disk_key disk_key;
4094 array_size = btrfs_super_sys_array_size(super_copy);
4095 if (array_size + item_size + sizeof(disk_key)
4096 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4097 unlock_chunks(root);
4101 ptr = super_copy->sys_chunk_array + array_size;
4102 btrfs_cpu_key_to_disk(&disk_key, key);
4103 memcpy(ptr, &disk_key, sizeof(disk_key));
4104 ptr += sizeof(disk_key);
4105 memcpy(ptr, chunk, item_size);
4106 item_size += sizeof(disk_key);
4107 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4108 unlock_chunks(root);
4114 * sort the devices in descending order by max_avail, total_avail
4116 static int btrfs_cmp_device_info(const void *a, const void *b)
4118 const struct btrfs_device_info *di_a = a;
4119 const struct btrfs_device_info *di_b = b;
4121 if (di_a->max_avail > di_b->max_avail)
4123 if (di_a->max_avail < di_b->max_avail)
4125 if (di_a->total_avail > di_b->total_avail)
4127 if (di_a->total_avail < di_b->total_avail)
4132 static struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4133 [BTRFS_RAID_RAID10] = {
4136 .devs_max = 0, /* 0 == as many as possible */
4138 .devs_increment = 2,
4141 [BTRFS_RAID_RAID1] = {
4146 .devs_increment = 2,
4149 [BTRFS_RAID_DUP] = {
4154 .devs_increment = 1,
4157 [BTRFS_RAID_RAID0] = {
4162 .devs_increment = 1,
4165 [BTRFS_RAID_SINGLE] = {
4170 .devs_increment = 1,
4173 [BTRFS_RAID_RAID5] = {
4178 .devs_increment = 1,
4181 [BTRFS_RAID_RAID6] = {
4186 .devs_increment = 1,
4191 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4193 /* TODO allow them to set a preferred stripe size */
4197 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4199 if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
4202 btrfs_set_fs_incompat(info, RAID56);
4205 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4206 - sizeof(struct btrfs_item) \
4207 - sizeof(struct btrfs_chunk)) \
4208 / sizeof(struct btrfs_stripe) + 1)
4210 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4211 - 2 * sizeof(struct btrfs_disk_key) \
4212 - 2 * sizeof(struct btrfs_chunk)) \
4213 / sizeof(struct btrfs_stripe) + 1)
4215 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4216 struct btrfs_root *extent_root, u64 start,
4219 struct btrfs_fs_info *info = extent_root->fs_info;
4220 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4221 struct list_head *cur;
4222 struct map_lookup *map = NULL;
4223 struct extent_map_tree *em_tree;
4224 struct extent_map *em;
4225 struct btrfs_device_info *devices_info = NULL;
4227 int num_stripes; /* total number of stripes to allocate */
4228 int data_stripes; /* number of stripes that count for
4230 int sub_stripes; /* sub_stripes info for map */
4231 int dev_stripes; /* stripes per dev */
4232 int devs_max; /* max devs to use */
4233 int devs_min; /* min devs needed */
4234 int devs_increment; /* ndevs has to be a multiple of this */
4235 int ncopies; /* how many copies to data has */
4237 u64 max_stripe_size;
4241 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4247 BUG_ON(!alloc_profile_is_valid(type, 0));
4249 if (list_empty(&fs_devices->alloc_list))
4252 index = __get_raid_index(type);
4254 sub_stripes = btrfs_raid_array[index].sub_stripes;
4255 dev_stripes = btrfs_raid_array[index].dev_stripes;
4256 devs_max = btrfs_raid_array[index].devs_max;
4257 devs_min = btrfs_raid_array[index].devs_min;
4258 devs_increment = btrfs_raid_array[index].devs_increment;
4259 ncopies = btrfs_raid_array[index].ncopies;
4261 if (type & BTRFS_BLOCK_GROUP_DATA) {
4262 max_stripe_size = 1024 * 1024 * 1024;
4263 max_chunk_size = 10 * max_stripe_size;
4265 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4266 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4267 /* for larger filesystems, use larger metadata chunks */
4268 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4269 max_stripe_size = 1024 * 1024 * 1024;
4271 max_stripe_size = 256 * 1024 * 1024;
4272 max_chunk_size = max_stripe_size;
4274 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4275 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4276 max_stripe_size = 32 * 1024 * 1024;
4277 max_chunk_size = 2 * max_stripe_size;
4279 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4281 btrfs_err(info, "invalid chunk type 0x%llx requested",
4286 /* we don't want a chunk larger than 10% of writeable space */
4287 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4290 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
4295 cur = fs_devices->alloc_list.next;
4298 * in the first pass through the devices list, we gather information
4299 * about the available holes on each device.
4302 while (cur != &fs_devices->alloc_list) {
4303 struct btrfs_device *device;
4307 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4311 if (!device->writeable) {
4313 "BTRFS: read-only device in alloc_list\n");
4317 if (!device->in_fs_metadata ||
4318 device->is_tgtdev_for_dev_replace)
4321 if (device->total_bytes > device->bytes_used)
4322 total_avail = device->total_bytes - device->bytes_used;
4326 /* If there is no space on this device, skip it. */
4327 if (total_avail == 0)
4330 ret = find_free_dev_extent(trans, device,
4331 max_stripe_size * dev_stripes,
4332 &dev_offset, &max_avail);
4333 if (ret && ret != -ENOSPC)
4337 max_avail = max_stripe_size * dev_stripes;
4339 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4342 if (ndevs == fs_devices->rw_devices) {
4343 WARN(1, "%s: found more than %llu devices\n",
4344 __func__, fs_devices->rw_devices);
4347 devices_info[ndevs].dev_offset = dev_offset;
4348 devices_info[ndevs].max_avail = max_avail;
4349 devices_info[ndevs].total_avail = total_avail;
4350 devices_info[ndevs].dev = device;
4355 * now sort the devices by hole size / available space
4357 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4358 btrfs_cmp_device_info, NULL);
4360 /* round down to number of usable stripes */
4361 ndevs -= ndevs % devs_increment;
4363 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4368 if (devs_max && ndevs > devs_max)
4371 * the primary goal is to maximize the number of stripes, so use as many
4372 * devices as possible, even if the stripes are not maximum sized.
4374 stripe_size = devices_info[ndevs-1].max_avail;
4375 num_stripes = ndevs * dev_stripes;
4378 * this will have to be fixed for RAID1 and RAID10 over
4381 data_stripes = num_stripes / ncopies;
4383 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4384 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4385 btrfs_super_stripesize(info->super_copy));
4386 data_stripes = num_stripes - 1;
4388 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4389 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4390 btrfs_super_stripesize(info->super_copy));
4391 data_stripes = num_stripes - 2;
4395 * Use the number of data stripes to figure out how big this chunk
4396 * is really going to be in terms of logical address space,
4397 * and compare that answer with the max chunk size
4399 if (stripe_size * data_stripes > max_chunk_size) {
4400 u64 mask = (1ULL << 24) - 1;
4401 stripe_size = max_chunk_size;
4402 do_div(stripe_size, data_stripes);
4404 /* bump the answer up to a 16MB boundary */
4405 stripe_size = (stripe_size + mask) & ~mask;
4407 /* but don't go higher than the limits we found
4408 * while searching for free extents
4410 if (stripe_size > devices_info[ndevs-1].max_avail)
4411 stripe_size = devices_info[ndevs-1].max_avail;
4414 do_div(stripe_size, dev_stripes);
4416 /* align to BTRFS_STRIPE_LEN */
4417 do_div(stripe_size, raid_stripe_len);
4418 stripe_size *= raid_stripe_len;
4420 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4425 map->num_stripes = num_stripes;
4427 for (i = 0; i < ndevs; ++i) {
4428 for (j = 0; j < dev_stripes; ++j) {
4429 int s = i * dev_stripes + j;
4430 map->stripes[s].dev = devices_info[i].dev;
4431 map->stripes[s].physical = devices_info[i].dev_offset +
4435 map->sector_size = extent_root->sectorsize;
4436 map->stripe_len = raid_stripe_len;
4437 map->io_align = raid_stripe_len;
4438 map->io_width = raid_stripe_len;
4440 map->sub_stripes = sub_stripes;
4442 num_bytes = stripe_size * data_stripes;
4444 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4446 em = alloc_extent_map();
4452 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4453 em->bdev = (struct block_device *)map;
4455 em->len = num_bytes;
4456 em->block_start = 0;
4457 em->block_len = em->len;
4458 em->orig_block_len = stripe_size;
4460 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4461 write_lock(&em_tree->lock);
4462 ret = add_extent_mapping(em_tree, em, 0);
4464 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4465 atomic_inc(&em->refs);
4467 write_unlock(&em_tree->lock);
4469 free_extent_map(em);
4473 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4474 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4477 goto error_del_extent;
4479 for (i = 0; i < map->num_stripes; i++) {
4480 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4481 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4484 spin_lock(&extent_root->fs_info->free_chunk_lock);
4485 extent_root->fs_info->free_chunk_space -= (stripe_size *
4487 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4489 free_extent_map(em);
4490 check_raid56_incompat_flag(extent_root->fs_info, type);
4492 kfree(devices_info);
4496 write_lock(&em_tree->lock);
4497 remove_extent_mapping(em_tree, em);
4498 write_unlock(&em_tree->lock);
4500 /* One for our allocation */
4501 free_extent_map(em);
4502 /* One for the tree reference */
4503 free_extent_map(em);
4505 kfree(devices_info);
4509 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4510 struct btrfs_root *extent_root,
4511 u64 chunk_offset, u64 chunk_size)
4513 struct btrfs_key key;
4514 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4515 struct btrfs_device *device;
4516 struct btrfs_chunk *chunk;
4517 struct btrfs_stripe *stripe;
4518 struct extent_map_tree *em_tree;
4519 struct extent_map *em;
4520 struct map_lookup *map;
4527 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4528 read_lock(&em_tree->lock);
4529 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4530 read_unlock(&em_tree->lock);
4533 btrfs_crit(extent_root->fs_info, "unable to find logical "
4534 "%Lu len %Lu", chunk_offset, chunk_size);
4538 if (em->start != chunk_offset || em->len != chunk_size) {
4539 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4540 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4541 chunk_size, em->start, em->len);
4542 free_extent_map(em);
4546 map = (struct map_lookup *)em->bdev;
4547 item_size = btrfs_chunk_item_size(map->num_stripes);
4548 stripe_size = em->orig_block_len;
4550 chunk = kzalloc(item_size, GFP_NOFS);
4556 for (i = 0; i < map->num_stripes; i++) {
4557 device = map->stripes[i].dev;
4558 dev_offset = map->stripes[i].physical;
4560 ret = btrfs_update_device(trans, device);
4563 ret = btrfs_alloc_dev_extent(trans, device,
4564 chunk_root->root_key.objectid,
4565 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4566 chunk_offset, dev_offset,
4572 stripe = &chunk->stripe;
4573 for (i = 0; i < map->num_stripes; i++) {
4574 device = map->stripes[i].dev;
4575 dev_offset = map->stripes[i].physical;
4577 btrfs_set_stack_stripe_devid(stripe, device->devid);
4578 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4579 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4583 btrfs_set_stack_chunk_length(chunk, chunk_size);
4584 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4585 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4586 btrfs_set_stack_chunk_type(chunk, map->type);
4587 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4588 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4589 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4590 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4591 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4593 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4594 key.type = BTRFS_CHUNK_ITEM_KEY;
4595 key.offset = chunk_offset;
4597 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4598 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4600 * TODO: Cleanup of inserted chunk root in case of
4603 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4609 free_extent_map(em);
4614 * Chunk allocation falls into two parts. The first part does works
4615 * that make the new allocated chunk useable, but not do any operation
4616 * that modifies the chunk tree. The second part does the works that
4617 * require modifying the chunk tree. This division is important for the
4618 * bootstrap process of adding storage to a seed btrfs.
4620 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4621 struct btrfs_root *extent_root, u64 type)
4625 chunk_offset = find_next_chunk(extent_root->fs_info);
4626 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4629 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4630 struct btrfs_root *root,
4631 struct btrfs_device *device)
4634 u64 sys_chunk_offset;
4636 struct btrfs_fs_info *fs_info = root->fs_info;
4637 struct btrfs_root *extent_root = fs_info->extent_root;
4640 chunk_offset = find_next_chunk(fs_info);
4641 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4642 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4647 sys_chunk_offset = find_next_chunk(root->fs_info);
4648 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4649 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4654 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4658 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4659 BTRFS_BLOCK_GROUP_RAID10 |
4660 BTRFS_BLOCK_GROUP_RAID5 |
4661 BTRFS_BLOCK_GROUP_DUP)) {
4663 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4672 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4674 struct extent_map *em;
4675 struct map_lookup *map;
4676 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4681 read_lock(&map_tree->map_tree.lock);
4682 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4683 read_unlock(&map_tree->map_tree.lock);
4687 map = (struct map_lookup *)em->bdev;
4688 for (i = 0; i < map->num_stripes; i++) {
4689 if (map->stripes[i].dev->missing) {
4694 if (!map->stripes[i].dev->writeable) {
4701 * If the number of missing devices is larger than max errors,
4702 * we can not write the data into that chunk successfully, so
4705 if (miss_ndevs > btrfs_chunk_max_errors(map))
4708 free_extent_map(em);
4712 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4714 extent_map_tree_init(&tree->map_tree);
4717 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4719 struct extent_map *em;
4722 write_lock(&tree->map_tree.lock);
4723 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4725 remove_extent_mapping(&tree->map_tree, em);
4726 write_unlock(&tree->map_tree.lock);
4730 free_extent_map(em);
4731 /* once for the tree */
4732 free_extent_map(em);
4736 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4738 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4739 struct extent_map *em;
4740 struct map_lookup *map;
4741 struct extent_map_tree *em_tree = &map_tree->map_tree;
4744 read_lock(&em_tree->lock);
4745 em = lookup_extent_mapping(em_tree, logical, len);
4746 read_unlock(&em_tree->lock);
4749 * We could return errors for these cases, but that could get ugly and
4750 * we'd probably do the same thing which is just not do anything else
4751 * and exit, so return 1 so the callers don't try to use other copies.
4754 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4759 if (em->start > logical || em->start + em->len < logical) {
4760 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4761 "%Lu-%Lu", logical, logical+len, em->start,
4762 em->start + em->len);
4763 free_extent_map(em);
4767 map = (struct map_lookup *)em->bdev;
4768 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4769 ret = map->num_stripes;
4770 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4771 ret = map->sub_stripes;
4772 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4774 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4778 free_extent_map(em);
4780 btrfs_dev_replace_lock(&fs_info->dev_replace);
4781 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4783 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4788 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4789 struct btrfs_mapping_tree *map_tree,
4792 struct extent_map *em;
4793 struct map_lookup *map;
4794 struct extent_map_tree *em_tree = &map_tree->map_tree;
4795 unsigned long len = root->sectorsize;
4797 read_lock(&em_tree->lock);
4798 em = lookup_extent_mapping(em_tree, logical, len);
4799 read_unlock(&em_tree->lock);
4802 BUG_ON(em->start > logical || em->start + em->len < logical);
4803 map = (struct map_lookup *)em->bdev;
4804 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4805 BTRFS_BLOCK_GROUP_RAID6)) {
4806 len = map->stripe_len * nr_data_stripes(map);
4808 free_extent_map(em);
4812 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4813 u64 logical, u64 len, int mirror_num)
4815 struct extent_map *em;
4816 struct map_lookup *map;
4817 struct extent_map_tree *em_tree = &map_tree->map_tree;
4820 read_lock(&em_tree->lock);
4821 em = lookup_extent_mapping(em_tree, logical, len);
4822 read_unlock(&em_tree->lock);
4825 BUG_ON(em->start > logical || em->start + em->len < logical);
4826 map = (struct map_lookup *)em->bdev;
4827 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4828 BTRFS_BLOCK_GROUP_RAID6))
4830 free_extent_map(em);
4834 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4835 struct map_lookup *map, int first, int num,
4836 int optimal, int dev_replace_is_ongoing)
4840 struct btrfs_device *srcdev;
4842 if (dev_replace_is_ongoing &&
4843 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4844 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4845 srcdev = fs_info->dev_replace.srcdev;
4850 * try to avoid the drive that is the source drive for a
4851 * dev-replace procedure, only choose it if no other non-missing
4852 * mirror is available
4854 for (tolerance = 0; tolerance < 2; tolerance++) {
4855 if (map->stripes[optimal].dev->bdev &&
4856 (tolerance || map->stripes[optimal].dev != srcdev))
4858 for (i = first; i < first + num; i++) {
4859 if (map->stripes[i].dev->bdev &&
4860 (tolerance || map->stripes[i].dev != srcdev))
4865 /* we couldn't find one that doesn't fail. Just return something
4866 * and the io error handling code will clean up eventually
4871 static inline int parity_smaller(u64 a, u64 b)
4876 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4877 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4879 struct btrfs_bio_stripe s;
4886 for (i = 0; i < bbio->num_stripes - 1; i++) {
4887 if (parity_smaller(raid_map[i], raid_map[i+1])) {
4888 s = bbio->stripes[i];
4890 bbio->stripes[i] = bbio->stripes[i+1];
4891 raid_map[i] = raid_map[i+1];
4892 bbio->stripes[i+1] = s;
4900 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4901 u64 logical, u64 *length,
4902 struct btrfs_bio **bbio_ret,
4903 int mirror_num, u64 **raid_map_ret)
4905 struct extent_map *em;
4906 struct map_lookup *map;
4907 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4908 struct extent_map_tree *em_tree = &map_tree->map_tree;
4911 u64 stripe_end_offset;
4916 u64 *raid_map = NULL;
4922 struct btrfs_bio *bbio = NULL;
4923 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4924 int dev_replace_is_ongoing = 0;
4925 int num_alloc_stripes;
4926 int patch_the_first_stripe_for_dev_replace = 0;
4927 u64 physical_to_patch_in_first_stripe = 0;
4928 u64 raid56_full_stripe_start = (u64)-1;
4930 read_lock(&em_tree->lock);
4931 em = lookup_extent_mapping(em_tree, logical, *length);
4932 read_unlock(&em_tree->lock);
4935 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
4940 if (em->start > logical || em->start + em->len < logical) {
4941 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
4942 "found %Lu-%Lu", logical, em->start,
4943 em->start + em->len);
4944 free_extent_map(em);
4948 map = (struct map_lookup *)em->bdev;
4949 offset = logical - em->start;
4951 stripe_len = map->stripe_len;
4954 * stripe_nr counts the total number of stripes we have to stride
4955 * to get to this block
4957 do_div(stripe_nr, stripe_len);
4959 stripe_offset = stripe_nr * stripe_len;
4960 BUG_ON(offset < stripe_offset);
4962 /* stripe_offset is the offset of this block in its stripe*/
4963 stripe_offset = offset - stripe_offset;
4965 /* if we're here for raid56, we need to know the stripe aligned start */
4966 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4967 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4968 raid56_full_stripe_start = offset;
4970 /* allow a write of a full stripe, but make sure we don't
4971 * allow straddling of stripes
4973 do_div(raid56_full_stripe_start, full_stripe_len);
4974 raid56_full_stripe_start *= full_stripe_len;
4977 if (rw & REQ_DISCARD) {
4978 /* we don't discard raid56 yet */
4980 (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4984 *length = min_t(u64, em->len - offset, *length);
4985 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4987 /* For writes to RAID[56], allow a full stripeset across all disks.
4988 For other RAID types and for RAID[56] reads, just allow a single
4989 stripe (on a single disk). */
4990 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4992 max_len = stripe_len * nr_data_stripes(map) -
4993 (offset - raid56_full_stripe_start);
4995 /* we limit the length of each bio to what fits in a stripe */
4996 max_len = stripe_len - stripe_offset;
4998 *length = min_t(u64, em->len - offset, max_len);
5000 *length = em->len - offset;
5003 /* This is for when we're called from btrfs_merge_bio_hook() and all
5004 it cares about is the length */
5008 btrfs_dev_replace_lock(dev_replace);
5009 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5010 if (!dev_replace_is_ongoing)
5011 btrfs_dev_replace_unlock(dev_replace);
5013 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5014 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5015 dev_replace->tgtdev != NULL) {
5017 * in dev-replace case, for repair case (that's the only
5018 * case where the mirror is selected explicitly when
5019 * calling btrfs_map_block), blocks left of the left cursor
5020 * can also be read from the target drive.
5021 * For REQ_GET_READ_MIRRORS, the target drive is added as
5022 * the last one to the array of stripes. For READ, it also
5023 * needs to be supported using the same mirror number.
5024 * If the requested block is not left of the left cursor,
5025 * EIO is returned. This can happen because btrfs_num_copies()
5026 * returns one more in the dev-replace case.
5028 u64 tmp_length = *length;
5029 struct btrfs_bio *tmp_bbio = NULL;
5030 int tmp_num_stripes;
5031 u64 srcdev_devid = dev_replace->srcdev->devid;
5032 int index_srcdev = 0;
5034 u64 physical_of_found = 0;
5036 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5037 logical, &tmp_length, &tmp_bbio, 0, NULL);
5039 WARN_ON(tmp_bbio != NULL);
5043 tmp_num_stripes = tmp_bbio->num_stripes;
5044 if (mirror_num > tmp_num_stripes) {
5046 * REQ_GET_READ_MIRRORS does not contain this
5047 * mirror, that means that the requested area
5048 * is not left of the left cursor
5056 * process the rest of the function using the mirror_num
5057 * of the source drive. Therefore look it up first.
5058 * At the end, patch the device pointer to the one of the
5061 for (i = 0; i < tmp_num_stripes; i++) {
5062 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5064 * In case of DUP, in order to keep it
5065 * simple, only add the mirror with the
5066 * lowest physical address
5069 physical_of_found <=
5070 tmp_bbio->stripes[i].physical)
5075 tmp_bbio->stripes[i].physical;
5080 mirror_num = index_srcdev + 1;
5081 patch_the_first_stripe_for_dev_replace = 1;
5082 physical_to_patch_in_first_stripe = physical_of_found;
5091 } else if (mirror_num > map->num_stripes) {
5097 stripe_nr_orig = stripe_nr;
5098 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5099 do_div(stripe_nr_end, map->stripe_len);
5100 stripe_end_offset = stripe_nr_end * map->stripe_len -
5103 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5104 if (rw & REQ_DISCARD)
5105 num_stripes = min_t(u64, map->num_stripes,
5106 stripe_nr_end - stripe_nr_orig);
5107 stripe_index = do_div(stripe_nr, map->num_stripes);
5108 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5110 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5111 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5112 num_stripes = map->num_stripes;
5113 else if (mirror_num)
5114 stripe_index = mirror_num - 1;
5116 stripe_index = find_live_mirror(fs_info, map, 0,
5118 current->pid % map->num_stripes,
5119 dev_replace_is_ongoing);
5120 mirror_num = stripe_index + 1;
5123 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5124 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5125 num_stripes = map->num_stripes;
5126 } else if (mirror_num) {
5127 stripe_index = mirror_num - 1;
5132 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5133 int factor = map->num_stripes / map->sub_stripes;
5135 stripe_index = do_div(stripe_nr, factor);
5136 stripe_index *= map->sub_stripes;
5138 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5139 num_stripes = map->sub_stripes;
5140 else if (rw & REQ_DISCARD)
5141 num_stripes = min_t(u64, map->sub_stripes *
5142 (stripe_nr_end - stripe_nr_orig),
5144 else if (mirror_num)
5145 stripe_index += mirror_num - 1;
5147 int old_stripe_index = stripe_index;
5148 stripe_index = find_live_mirror(fs_info, map,
5150 map->sub_stripes, stripe_index +
5151 current->pid % map->sub_stripes,
5152 dev_replace_is_ongoing);
5153 mirror_num = stripe_index - old_stripe_index + 1;
5156 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5157 BTRFS_BLOCK_GROUP_RAID6)) {
5160 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
5164 /* push stripe_nr back to the start of the full stripe */
5165 stripe_nr = raid56_full_stripe_start;
5166 do_div(stripe_nr, stripe_len);
5168 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5170 /* RAID[56] write or recovery. Return all stripes */
5171 num_stripes = map->num_stripes;
5172 max_errors = nr_parity_stripes(map);
5174 raid_map = kmalloc_array(num_stripes, sizeof(u64),
5181 /* Work out the disk rotation on this stripe-set */
5183 rot = do_div(tmp, num_stripes);
5185 /* Fill in the logical address of each stripe */
5186 tmp = stripe_nr * nr_data_stripes(map);
5187 for (i = 0; i < nr_data_stripes(map); i++)
5188 raid_map[(i+rot) % num_stripes] =
5189 em->start + (tmp + i) * map->stripe_len;
5191 raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5192 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5193 raid_map[(i+rot+1) % num_stripes] =
5196 *length = map->stripe_len;
5201 * Mirror #0 or #1 means the original data block.
5202 * Mirror #2 is RAID5 parity block.
5203 * Mirror #3 is RAID6 Q block.
5205 stripe_index = do_div(stripe_nr, nr_data_stripes(map));
5207 stripe_index = nr_data_stripes(map) +
5210 /* We distribute the parity blocks across stripes */
5211 tmp = stripe_nr + stripe_index;
5212 stripe_index = do_div(tmp, map->num_stripes);
5213 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5214 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5219 * after this do_div call, stripe_nr is the number of stripes
5220 * on this device we have to walk to find the data, and
5221 * stripe_index is the number of our device in the stripe array
5223 stripe_index = do_div(stripe_nr, map->num_stripes);
5224 mirror_num = stripe_index + 1;
5226 BUG_ON(stripe_index >= map->num_stripes);
5228 num_alloc_stripes = num_stripes;
5229 if (dev_replace_is_ongoing) {
5230 if (rw & (REQ_WRITE | REQ_DISCARD))
5231 num_alloc_stripes <<= 1;
5232 if (rw & REQ_GET_READ_MIRRORS)
5233 num_alloc_stripes++;
5235 bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
5241 atomic_set(&bbio->error, 0);
5243 if (rw & REQ_DISCARD) {
5245 int sub_stripes = 0;
5246 u64 stripes_per_dev = 0;
5247 u32 remaining_stripes = 0;
5248 u32 last_stripe = 0;
5251 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5252 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5255 sub_stripes = map->sub_stripes;
5257 factor = map->num_stripes / sub_stripes;
5258 stripes_per_dev = div_u64_rem(stripe_nr_end -
5261 &remaining_stripes);
5262 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5263 last_stripe *= sub_stripes;
5266 for (i = 0; i < num_stripes; i++) {
5267 bbio->stripes[i].physical =
5268 map->stripes[stripe_index].physical +
5269 stripe_offset + stripe_nr * map->stripe_len;
5270 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5272 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5273 BTRFS_BLOCK_GROUP_RAID10)) {
5274 bbio->stripes[i].length = stripes_per_dev *
5277 if (i / sub_stripes < remaining_stripes)
5278 bbio->stripes[i].length +=
5282 * Special for the first stripe and
5285 * |-------|...|-------|
5289 if (i < sub_stripes)
5290 bbio->stripes[i].length -=
5293 if (stripe_index >= last_stripe &&
5294 stripe_index <= (last_stripe +
5296 bbio->stripes[i].length -=
5299 if (i == sub_stripes - 1)
5302 bbio->stripes[i].length = *length;
5305 if (stripe_index == map->num_stripes) {
5306 /* This could only happen for RAID0/10 */
5312 for (i = 0; i < num_stripes; i++) {
5313 bbio->stripes[i].physical =
5314 map->stripes[stripe_index].physical +
5316 stripe_nr * map->stripe_len;
5317 bbio->stripes[i].dev =
5318 map->stripes[stripe_index].dev;
5323 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5324 max_errors = btrfs_chunk_max_errors(map);
5326 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5327 dev_replace->tgtdev != NULL) {
5328 int index_where_to_add;
5329 u64 srcdev_devid = dev_replace->srcdev->devid;
5332 * duplicate the write operations while the dev replace
5333 * procedure is running. Since the copying of the old disk
5334 * to the new disk takes place at run time while the
5335 * filesystem is mounted writable, the regular write
5336 * operations to the old disk have to be duplicated to go
5337 * to the new disk as well.
5338 * Note that device->missing is handled by the caller, and
5339 * that the write to the old disk is already set up in the
5342 index_where_to_add = num_stripes;
5343 for (i = 0; i < num_stripes; i++) {
5344 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5345 /* write to new disk, too */
5346 struct btrfs_bio_stripe *new =
5347 bbio->stripes + index_where_to_add;
5348 struct btrfs_bio_stripe *old =
5351 new->physical = old->physical;
5352 new->length = old->length;
5353 new->dev = dev_replace->tgtdev;
5354 index_where_to_add++;
5358 num_stripes = index_where_to_add;
5359 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5360 dev_replace->tgtdev != NULL) {
5361 u64 srcdev_devid = dev_replace->srcdev->devid;
5362 int index_srcdev = 0;
5364 u64 physical_of_found = 0;
5367 * During the dev-replace procedure, the target drive can
5368 * also be used to read data in case it is needed to repair
5369 * a corrupt block elsewhere. This is possible if the
5370 * requested area is left of the left cursor. In this area,
5371 * the target drive is a full copy of the source drive.
5373 for (i = 0; i < num_stripes; i++) {
5374 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5376 * In case of DUP, in order to keep it
5377 * simple, only add the mirror with the
5378 * lowest physical address
5381 physical_of_found <=
5382 bbio->stripes[i].physical)
5386 physical_of_found = bbio->stripes[i].physical;
5390 u64 length = map->stripe_len;
5392 if (physical_of_found + length <=
5393 dev_replace->cursor_left) {
5394 struct btrfs_bio_stripe *tgtdev_stripe =
5395 bbio->stripes + num_stripes;
5397 tgtdev_stripe->physical = physical_of_found;
5398 tgtdev_stripe->length =
5399 bbio->stripes[index_srcdev].length;
5400 tgtdev_stripe->dev = dev_replace->tgtdev;
5408 bbio->num_stripes = num_stripes;
5409 bbio->max_errors = max_errors;
5410 bbio->mirror_num = mirror_num;
5413 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5414 * mirror_num == num_stripes + 1 && dev_replace target drive is
5415 * available as a mirror
5417 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5418 WARN_ON(num_stripes > 1);
5419 bbio->stripes[0].dev = dev_replace->tgtdev;
5420 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5421 bbio->mirror_num = map->num_stripes + 1;
5424 sort_parity_stripes(bbio, raid_map);
5425 *raid_map_ret = raid_map;
5428 if (dev_replace_is_ongoing)
5429 btrfs_dev_replace_unlock(dev_replace);
5430 free_extent_map(em);
5434 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5435 u64 logical, u64 *length,
5436 struct btrfs_bio **bbio_ret, int mirror_num)
5438 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5442 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5443 u64 chunk_start, u64 physical, u64 devid,
5444 u64 **logical, int *naddrs, int *stripe_len)
5446 struct extent_map_tree *em_tree = &map_tree->map_tree;
5447 struct extent_map *em;
5448 struct map_lookup *map;
5456 read_lock(&em_tree->lock);
5457 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5458 read_unlock(&em_tree->lock);
5461 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5466 if (em->start != chunk_start) {
5467 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5468 em->start, chunk_start);
5469 free_extent_map(em);
5472 map = (struct map_lookup *)em->bdev;
5475 rmap_len = map->stripe_len;
5477 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5478 do_div(length, map->num_stripes / map->sub_stripes);
5479 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5480 do_div(length, map->num_stripes);
5481 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
5482 BTRFS_BLOCK_GROUP_RAID6)) {
5483 do_div(length, nr_data_stripes(map));
5484 rmap_len = map->stripe_len * nr_data_stripes(map);
5487 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
5488 BUG_ON(!buf); /* -ENOMEM */
5490 for (i = 0; i < map->num_stripes; i++) {
5491 if (devid && map->stripes[i].dev->devid != devid)
5493 if (map->stripes[i].physical > physical ||
5494 map->stripes[i].physical + length <= physical)
5497 stripe_nr = physical - map->stripes[i].physical;
5498 do_div(stripe_nr, map->stripe_len);
5500 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5501 stripe_nr = stripe_nr * map->num_stripes + i;
5502 do_div(stripe_nr, map->sub_stripes);
5503 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5504 stripe_nr = stripe_nr * map->num_stripes + i;
5505 } /* else if RAID[56], multiply by nr_data_stripes().
5506 * Alternatively, just use rmap_len below instead of
5507 * map->stripe_len */
5509 bytenr = chunk_start + stripe_nr * rmap_len;
5510 WARN_ON(nr >= map->num_stripes);
5511 for (j = 0; j < nr; j++) {
5512 if (buf[j] == bytenr)
5516 WARN_ON(nr >= map->num_stripes);
5523 *stripe_len = rmap_len;
5525 free_extent_map(em);
5529 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5531 if (likely(bbio->flags & BTRFS_BIO_ORIG_BIO_SUBMITTED))
5532 bio_endio_nodec(bio, err);
5534 bio_endio(bio, err);
5538 static void btrfs_end_bio(struct bio *bio, int err)
5540 struct btrfs_bio *bbio = bio->bi_private;
5541 struct btrfs_device *dev = bbio->stripes[0].dev;
5542 int is_orig_bio = 0;
5545 atomic_inc(&bbio->error);
5546 if (err == -EIO || err == -EREMOTEIO) {
5547 unsigned int stripe_index =
5548 btrfs_io_bio(bio)->stripe_index;
5550 BUG_ON(stripe_index >= bbio->num_stripes);
5551 dev = bbio->stripes[stripe_index].dev;
5553 if (bio->bi_rw & WRITE)
5554 btrfs_dev_stat_inc(dev,
5555 BTRFS_DEV_STAT_WRITE_ERRS);
5557 btrfs_dev_stat_inc(dev,
5558 BTRFS_DEV_STAT_READ_ERRS);
5559 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5560 btrfs_dev_stat_inc(dev,
5561 BTRFS_DEV_STAT_FLUSH_ERRS);
5562 btrfs_dev_stat_print_on_error(dev);
5567 if (bio == bbio->orig_bio)
5570 btrfs_bio_counter_dec(bbio->fs_info);
5572 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5575 bio = bbio->orig_bio;
5578 bio->bi_private = bbio->private;
5579 bio->bi_end_io = bbio->end_io;
5580 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5581 /* only send an error to the higher layers if it is
5582 * beyond the tolerance of the btrfs bio
5584 if (atomic_read(&bbio->error) > bbio->max_errors) {
5588 * this bio is actually up to date, we didn't
5589 * go over the max number of errors
5591 set_bit(BIO_UPTODATE, &bio->bi_flags);
5595 btrfs_end_bbio(bbio, bio, err);
5596 } else if (!is_orig_bio) {
5602 * see run_scheduled_bios for a description of why bios are collected for
5605 * This will add one bio to the pending list for a device and make sure
5606 * the work struct is scheduled.
5608 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5609 struct btrfs_device *device,
5610 int rw, struct bio *bio)
5612 int should_queue = 1;
5613 struct btrfs_pending_bios *pending_bios;
5615 if (device->missing || !device->bdev) {
5616 bio_endio(bio, -EIO);
5620 /* don't bother with additional async steps for reads, right now */
5621 if (!(rw & REQ_WRITE)) {
5623 btrfsic_submit_bio(rw, bio);
5629 * nr_async_bios allows us to reliably return congestion to the
5630 * higher layers. Otherwise, the async bio makes it appear we have
5631 * made progress against dirty pages when we've really just put it
5632 * on a queue for later
5634 atomic_inc(&root->fs_info->nr_async_bios);
5635 WARN_ON(bio->bi_next);
5636 bio->bi_next = NULL;
5639 spin_lock(&device->io_lock);
5640 if (bio->bi_rw & REQ_SYNC)
5641 pending_bios = &device->pending_sync_bios;
5643 pending_bios = &device->pending_bios;
5645 if (pending_bios->tail)
5646 pending_bios->tail->bi_next = bio;
5648 pending_bios->tail = bio;
5649 if (!pending_bios->head)
5650 pending_bios->head = bio;
5651 if (device->running_pending)
5654 spin_unlock(&device->io_lock);
5657 btrfs_queue_work(root->fs_info->submit_workers,
5661 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5664 struct bio_vec *prev;
5665 struct request_queue *q = bdev_get_queue(bdev);
5666 unsigned int max_sectors = queue_max_sectors(q);
5667 struct bvec_merge_data bvm = {
5669 .bi_sector = sector,
5670 .bi_rw = bio->bi_rw,
5673 if (WARN_ON(bio->bi_vcnt == 0))
5676 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5677 if (bio_sectors(bio) > max_sectors)
5680 if (!q->merge_bvec_fn)
5683 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5684 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5689 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5690 struct bio *bio, u64 physical, int dev_nr,
5693 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5695 bio->bi_private = bbio;
5696 btrfs_io_bio(bio)->stripe_index = dev_nr;
5697 bio->bi_end_io = btrfs_end_bio;
5698 bio->bi_iter.bi_sector = physical >> 9;
5701 struct rcu_string *name;
5704 name = rcu_dereference(dev->name);
5705 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5706 "(%s id %llu), size=%u\n", rw,
5707 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5708 name->str, dev->devid, bio->bi_iter.bi_size);
5712 bio->bi_bdev = dev->bdev;
5714 btrfs_bio_counter_inc_noblocked(root->fs_info);
5717 btrfs_schedule_bio(root, dev, rw, bio);
5719 btrfsic_submit_bio(rw, bio);
5722 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5723 struct bio *first_bio, struct btrfs_device *dev,
5724 int dev_nr, int rw, int async)
5726 struct bio_vec *bvec = first_bio->bi_io_vec;
5728 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5729 u64 physical = bbio->stripes[dev_nr].physical;
5732 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5736 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5737 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5738 bvec->bv_offset) < bvec->bv_len) {
5739 u64 len = bio->bi_iter.bi_size;
5741 atomic_inc(&bbio->stripes_pending);
5742 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5750 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5754 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5756 atomic_inc(&bbio->error);
5757 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5758 /* Shoud be the original bio. */
5759 WARN_ON(bio != bbio->orig_bio);
5761 bio->bi_private = bbio->private;
5762 bio->bi_end_io = bbio->end_io;
5763 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5764 bio->bi_iter.bi_sector = logical >> 9;
5766 btrfs_end_bbio(bbio, bio, -EIO);
5770 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5771 int mirror_num, int async_submit)
5773 struct btrfs_device *dev;
5774 struct bio *first_bio = bio;
5775 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5778 u64 *raid_map = NULL;
5782 struct btrfs_bio *bbio = NULL;
5784 length = bio->bi_iter.bi_size;
5785 map_length = length;
5787 btrfs_bio_counter_inc_blocked(root->fs_info);
5788 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5789 mirror_num, &raid_map);
5791 btrfs_bio_counter_dec(root->fs_info);
5795 total_devs = bbio->num_stripes;
5796 bbio->orig_bio = first_bio;
5797 bbio->private = first_bio->bi_private;
5798 bbio->end_io = first_bio->bi_end_io;
5799 bbio->fs_info = root->fs_info;
5800 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5803 /* In this case, map_length has been set to the length of
5804 a single stripe; not the whole write */
5806 ret = raid56_parity_write(root, bio, bbio,
5807 raid_map, map_length);
5809 ret = raid56_parity_recover(root, bio, bbio,
5810 raid_map, map_length,
5814 * FIXME, replace dosen't support raid56 yet, please fix
5817 btrfs_bio_counter_dec(root->fs_info);
5821 if (map_length < length) {
5822 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
5823 logical, length, map_length);
5827 while (dev_nr < total_devs) {
5828 dev = bbio->stripes[dev_nr].dev;
5829 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5830 bbio_error(bbio, first_bio, logical);
5836 * Check and see if we're ok with this bio based on it's size
5837 * and offset with the given device.
5839 if (!bio_size_ok(dev->bdev, first_bio,
5840 bbio->stripes[dev_nr].physical >> 9)) {
5841 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5842 dev_nr, rw, async_submit);
5848 if (dev_nr < total_devs - 1) {
5849 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
5850 BUG_ON(!bio); /* -ENOMEM */
5853 bbio->flags |= BTRFS_BIO_ORIG_BIO_SUBMITTED;
5856 submit_stripe_bio(root, bbio, bio,
5857 bbio->stripes[dev_nr].physical, dev_nr, rw,
5861 btrfs_bio_counter_dec(root->fs_info);
5865 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5868 struct btrfs_device *device;
5869 struct btrfs_fs_devices *cur_devices;
5871 cur_devices = fs_info->fs_devices;
5872 while (cur_devices) {
5874 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5875 device = __find_device(&cur_devices->devices,
5880 cur_devices = cur_devices->seed;
5885 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5886 struct btrfs_fs_devices *fs_devices,
5887 u64 devid, u8 *dev_uuid)
5889 struct btrfs_device *device;
5891 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
5895 list_add(&device->dev_list, &fs_devices->devices);
5896 device->fs_devices = fs_devices;
5897 fs_devices->num_devices++;
5899 device->missing = 1;
5900 fs_devices->missing_devices++;
5906 * btrfs_alloc_device - allocate struct btrfs_device
5907 * @fs_info: used only for generating a new devid, can be NULL if
5908 * devid is provided (i.e. @devid != NULL).
5909 * @devid: a pointer to devid for this device. If NULL a new devid
5911 * @uuid: a pointer to UUID for this device. If NULL a new UUID
5914 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
5915 * on error. Returned struct is not linked onto any lists and can be
5916 * destroyed with kfree() right away.
5918 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
5922 struct btrfs_device *dev;
5925 if (WARN_ON(!devid && !fs_info))
5926 return ERR_PTR(-EINVAL);
5928 dev = __alloc_device();
5937 ret = find_next_devid(fs_info, &tmp);
5940 return ERR_PTR(ret);
5946 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
5948 generate_random_uuid(dev->uuid);
5950 btrfs_init_work(&dev->work, btrfs_submit_helper,
5951 pending_bios_fn, NULL, NULL);
5956 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5957 struct extent_buffer *leaf,
5958 struct btrfs_chunk *chunk)
5960 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5961 struct map_lookup *map;
5962 struct extent_map *em;
5966 u8 uuid[BTRFS_UUID_SIZE];
5971 logical = key->offset;
5972 length = btrfs_chunk_length(leaf, chunk);
5974 read_lock(&map_tree->map_tree.lock);
5975 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5976 read_unlock(&map_tree->map_tree.lock);
5978 /* already mapped? */
5979 if (em && em->start <= logical && em->start + em->len > logical) {
5980 free_extent_map(em);
5983 free_extent_map(em);
5986 em = alloc_extent_map();
5989 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5990 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5992 free_extent_map(em);
5996 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
5997 em->bdev = (struct block_device *)map;
5998 em->start = logical;
6001 em->block_start = 0;
6002 em->block_len = em->len;
6004 map->num_stripes = num_stripes;
6005 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6006 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6007 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6008 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6009 map->type = btrfs_chunk_type(leaf, chunk);
6010 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6011 for (i = 0; i < num_stripes; i++) {
6012 map->stripes[i].physical =
6013 btrfs_stripe_offset_nr(leaf, chunk, i);
6014 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6015 read_extent_buffer(leaf, uuid, (unsigned long)
6016 btrfs_stripe_dev_uuid_nr(chunk, i),
6018 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6020 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6021 free_extent_map(em);
6024 if (!map->stripes[i].dev) {
6025 map->stripes[i].dev =
6026 add_missing_dev(root, root->fs_info->fs_devices,
6028 if (!map->stripes[i].dev) {
6029 free_extent_map(em);
6033 map->stripes[i].dev->in_fs_metadata = 1;
6036 write_lock(&map_tree->map_tree.lock);
6037 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6038 write_unlock(&map_tree->map_tree.lock);
6039 BUG_ON(ret); /* Tree corruption */
6040 free_extent_map(em);
6045 static void fill_device_from_item(struct extent_buffer *leaf,
6046 struct btrfs_dev_item *dev_item,
6047 struct btrfs_device *device)
6051 device->devid = btrfs_device_id(leaf, dev_item);
6052 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6053 device->total_bytes = device->disk_total_bytes;
6054 device->commit_total_bytes = device->disk_total_bytes;
6055 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6056 device->commit_bytes_used = device->bytes_used;
6057 device->type = btrfs_device_type(leaf, dev_item);
6058 device->io_align = btrfs_device_io_align(leaf, dev_item);
6059 device->io_width = btrfs_device_io_width(leaf, dev_item);
6060 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6061 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6062 device->is_tgtdev_for_dev_replace = 0;
6064 ptr = btrfs_device_uuid(dev_item);
6065 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6068 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6071 struct btrfs_fs_devices *fs_devices;
6074 BUG_ON(!mutex_is_locked(&uuid_mutex));
6076 fs_devices = root->fs_info->fs_devices->seed;
6077 while (fs_devices) {
6078 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6081 fs_devices = fs_devices->seed;
6084 fs_devices = find_fsid(fsid);
6086 if (!btrfs_test_opt(root, DEGRADED))
6087 return ERR_PTR(-ENOENT);
6089 fs_devices = alloc_fs_devices(fsid);
6090 if (IS_ERR(fs_devices))
6093 fs_devices->seeding = 1;
6094 fs_devices->opened = 1;
6098 fs_devices = clone_fs_devices(fs_devices);
6099 if (IS_ERR(fs_devices))
6102 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6103 root->fs_info->bdev_holder);
6105 free_fs_devices(fs_devices);
6106 fs_devices = ERR_PTR(ret);
6110 if (!fs_devices->seeding) {
6111 __btrfs_close_devices(fs_devices);
6112 free_fs_devices(fs_devices);
6113 fs_devices = ERR_PTR(-EINVAL);
6117 fs_devices->seed = root->fs_info->fs_devices->seed;
6118 root->fs_info->fs_devices->seed = fs_devices;
6123 static int read_one_dev(struct btrfs_root *root,
6124 struct extent_buffer *leaf,
6125 struct btrfs_dev_item *dev_item)
6127 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6128 struct btrfs_device *device;
6131 u8 fs_uuid[BTRFS_UUID_SIZE];
6132 u8 dev_uuid[BTRFS_UUID_SIZE];
6134 devid = btrfs_device_id(leaf, dev_item);
6135 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6137 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6140 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6141 fs_devices = open_seed_devices(root, fs_uuid);
6142 if (IS_ERR(fs_devices))
6143 return PTR_ERR(fs_devices);
6146 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6148 if (!btrfs_test_opt(root, DEGRADED))
6151 btrfs_warn(root->fs_info, "devid %llu missing", devid);
6152 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6156 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6159 if(!device->bdev && !device->missing) {
6161 * this happens when a device that was properly setup
6162 * in the device info lists suddenly goes bad.
6163 * device->bdev is NULL, and so we have to set
6164 * device->missing to one here
6166 device->fs_devices->missing_devices++;
6167 device->missing = 1;
6170 /* Move the device to its own fs_devices */
6171 if (device->fs_devices != fs_devices) {
6172 ASSERT(device->missing);
6174 list_move(&device->dev_list, &fs_devices->devices);
6175 device->fs_devices->num_devices--;
6176 fs_devices->num_devices++;
6178 device->fs_devices->missing_devices--;
6179 fs_devices->missing_devices++;
6181 device->fs_devices = fs_devices;
6185 if (device->fs_devices != root->fs_info->fs_devices) {
6186 BUG_ON(device->writeable);
6187 if (device->generation !=
6188 btrfs_device_generation(leaf, dev_item))
6192 fill_device_from_item(leaf, dev_item, device);
6193 device->in_fs_metadata = 1;
6194 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6195 device->fs_devices->total_rw_bytes += device->total_bytes;
6196 spin_lock(&root->fs_info->free_chunk_lock);
6197 root->fs_info->free_chunk_space += device->total_bytes -
6199 spin_unlock(&root->fs_info->free_chunk_lock);
6205 int btrfs_read_sys_array(struct btrfs_root *root)
6207 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6208 struct extent_buffer *sb;
6209 struct btrfs_disk_key *disk_key;
6210 struct btrfs_chunk *chunk;
6212 unsigned long sb_ptr;
6218 struct btrfs_key key;
6220 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
6221 BTRFS_SUPER_INFO_SIZE);
6224 btrfs_set_buffer_uptodate(sb);
6225 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6227 * The sb extent buffer is artifical and just used to read the system array.
6228 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6229 * pages up-to-date when the page is larger: extent does not cover the
6230 * whole page and consequently check_page_uptodate does not find all
6231 * the page's extents up-to-date (the hole beyond sb),
6232 * write_extent_buffer then triggers a WARN_ON.
6234 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6235 * but sb spans only this function. Add an explicit SetPageUptodate call
6236 * to silence the warning eg. on PowerPC 64.
6238 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6239 SetPageUptodate(sb->pages[0]);
6241 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6242 array_size = btrfs_super_sys_array_size(super_copy);
6244 ptr = super_copy->sys_chunk_array;
6245 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
6248 while (cur < array_size) {
6249 disk_key = (struct btrfs_disk_key *)ptr;
6250 btrfs_disk_key_to_cpu(&key, disk_key);
6252 len = sizeof(*disk_key); ptr += len;
6256 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6257 chunk = (struct btrfs_chunk *)sb_ptr;
6258 ret = read_one_chunk(root, &key, sb, chunk);
6261 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6262 len = btrfs_chunk_item_size(num_stripes);
6271 free_extent_buffer(sb);
6275 int btrfs_read_chunk_tree(struct btrfs_root *root)
6277 struct btrfs_path *path;
6278 struct extent_buffer *leaf;
6279 struct btrfs_key key;
6280 struct btrfs_key found_key;
6284 root = root->fs_info->chunk_root;
6286 path = btrfs_alloc_path();
6290 mutex_lock(&uuid_mutex);
6294 * Read all device items, and then all the chunk items. All
6295 * device items are found before any chunk item (their object id
6296 * is smaller than the lowest possible object id for a chunk
6297 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6299 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6302 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6306 leaf = path->nodes[0];
6307 slot = path->slots[0];
6308 if (slot >= btrfs_header_nritems(leaf)) {
6309 ret = btrfs_next_leaf(root, path);
6316 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6317 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6318 struct btrfs_dev_item *dev_item;
6319 dev_item = btrfs_item_ptr(leaf, slot,
6320 struct btrfs_dev_item);
6321 ret = read_one_dev(root, leaf, dev_item);
6324 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6325 struct btrfs_chunk *chunk;
6326 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6327 ret = read_one_chunk(root, &found_key, leaf, chunk);
6335 unlock_chunks(root);
6336 mutex_unlock(&uuid_mutex);
6338 btrfs_free_path(path);
6342 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6344 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6345 struct btrfs_device *device;
6347 while (fs_devices) {
6348 mutex_lock(&fs_devices->device_list_mutex);
6349 list_for_each_entry(device, &fs_devices->devices, dev_list)
6350 device->dev_root = fs_info->dev_root;
6351 mutex_unlock(&fs_devices->device_list_mutex);
6353 fs_devices = fs_devices->seed;
6357 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6361 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6362 btrfs_dev_stat_reset(dev, i);
6365 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6367 struct btrfs_key key;
6368 struct btrfs_key found_key;
6369 struct btrfs_root *dev_root = fs_info->dev_root;
6370 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6371 struct extent_buffer *eb;
6374 struct btrfs_device *device;
6375 struct btrfs_path *path = NULL;
6378 path = btrfs_alloc_path();
6384 mutex_lock(&fs_devices->device_list_mutex);
6385 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6387 struct btrfs_dev_stats_item *ptr;
6390 key.type = BTRFS_DEV_STATS_KEY;
6391 key.offset = device->devid;
6392 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6394 __btrfs_reset_dev_stats(device);
6395 device->dev_stats_valid = 1;
6396 btrfs_release_path(path);
6399 slot = path->slots[0];
6400 eb = path->nodes[0];
6401 btrfs_item_key_to_cpu(eb, &found_key, slot);
6402 item_size = btrfs_item_size_nr(eb, slot);
6404 ptr = btrfs_item_ptr(eb, slot,
6405 struct btrfs_dev_stats_item);
6407 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6408 if (item_size >= (1 + i) * sizeof(__le64))
6409 btrfs_dev_stat_set(device, i,
6410 btrfs_dev_stats_value(eb, ptr, i));
6412 btrfs_dev_stat_reset(device, i);
6415 device->dev_stats_valid = 1;
6416 btrfs_dev_stat_print_on_load(device);
6417 btrfs_release_path(path);
6419 mutex_unlock(&fs_devices->device_list_mutex);
6422 btrfs_free_path(path);
6423 return ret < 0 ? ret : 0;
6426 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6427 struct btrfs_root *dev_root,
6428 struct btrfs_device *device)
6430 struct btrfs_path *path;
6431 struct btrfs_key key;
6432 struct extent_buffer *eb;
6433 struct btrfs_dev_stats_item *ptr;
6438 key.type = BTRFS_DEV_STATS_KEY;
6439 key.offset = device->devid;
6441 path = btrfs_alloc_path();
6443 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6445 printk_in_rcu(KERN_WARNING "BTRFS: "
6446 "error %d while searching for dev_stats item for device %s!\n",
6447 ret, rcu_str_deref(device->name));
6452 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6453 /* need to delete old one and insert a new one */
6454 ret = btrfs_del_item(trans, dev_root, path);
6456 printk_in_rcu(KERN_WARNING "BTRFS: "
6457 "delete too small dev_stats item for device %s failed %d!\n",
6458 rcu_str_deref(device->name), ret);
6465 /* need to insert a new item */
6466 btrfs_release_path(path);
6467 ret = btrfs_insert_empty_item(trans, dev_root, path,
6468 &key, sizeof(*ptr));
6470 printk_in_rcu(KERN_WARNING "BTRFS: "
6471 "insert dev_stats item for device %s failed %d!\n",
6472 rcu_str_deref(device->name), ret);
6477 eb = path->nodes[0];
6478 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6479 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6480 btrfs_set_dev_stats_value(eb, ptr, i,
6481 btrfs_dev_stat_read(device, i));
6482 btrfs_mark_buffer_dirty(eb);
6485 btrfs_free_path(path);
6490 * called from commit_transaction. Writes all changed device stats to disk.
6492 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6493 struct btrfs_fs_info *fs_info)
6495 struct btrfs_root *dev_root = fs_info->dev_root;
6496 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6497 struct btrfs_device *device;
6501 mutex_lock(&fs_devices->device_list_mutex);
6502 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6503 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6506 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6507 ret = update_dev_stat_item(trans, dev_root, device);
6509 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6511 mutex_unlock(&fs_devices->device_list_mutex);
6516 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6518 btrfs_dev_stat_inc(dev, index);
6519 btrfs_dev_stat_print_on_error(dev);
6522 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6524 if (!dev->dev_stats_valid)
6526 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6527 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6528 rcu_str_deref(dev->name),
6529 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6530 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6531 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6532 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6533 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6536 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6540 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6541 if (btrfs_dev_stat_read(dev, i) != 0)
6543 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6544 return; /* all values == 0, suppress message */
6546 printk_in_rcu(KERN_INFO "BTRFS: "
6547 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6548 rcu_str_deref(dev->name),
6549 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6550 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6551 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6552 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6553 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6556 int btrfs_get_dev_stats(struct btrfs_root *root,
6557 struct btrfs_ioctl_get_dev_stats *stats)
6559 struct btrfs_device *dev;
6560 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6563 mutex_lock(&fs_devices->device_list_mutex);
6564 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6565 mutex_unlock(&fs_devices->device_list_mutex);
6568 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6570 } else if (!dev->dev_stats_valid) {
6571 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6573 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6574 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6575 if (stats->nr_items > i)
6577 btrfs_dev_stat_read_and_reset(dev, i);
6579 btrfs_dev_stat_reset(dev, i);
6582 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6583 if (stats->nr_items > i)
6584 stats->values[i] = btrfs_dev_stat_read(dev, i);
6586 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6587 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6591 int btrfs_scratch_superblock(struct btrfs_device *device)
6593 struct buffer_head *bh;
6594 struct btrfs_super_block *disk_super;
6596 bh = btrfs_read_dev_super(device->bdev);
6599 disk_super = (struct btrfs_super_block *)bh->b_data;
6601 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6602 set_buffer_dirty(bh);
6603 sync_dirty_buffer(bh);
6610 * Update the size of all devices, which is used for writing out the
6613 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6615 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6616 struct btrfs_device *curr, *next;
6618 if (list_empty(&fs_devices->resized_devices))
6621 mutex_lock(&fs_devices->device_list_mutex);
6622 lock_chunks(fs_info->dev_root);
6623 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6625 list_del_init(&curr->resized_list);
6626 curr->commit_total_bytes = curr->disk_total_bytes;
6628 unlock_chunks(fs_info->dev_root);
6629 mutex_unlock(&fs_devices->device_list_mutex);
6632 /* Must be invoked during the transaction commit */
6633 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6634 struct btrfs_transaction *transaction)
6636 struct extent_map *em;
6637 struct map_lookup *map;
6638 struct btrfs_device *dev;
6641 if (list_empty(&transaction->pending_chunks))
6644 /* In order to kick the device replace finish process */
6646 list_for_each_entry(em, &transaction->pending_chunks, list) {
6647 map = (struct map_lookup *)em->bdev;
6649 for (i = 0; i < map->num_stripes; i++) {
6650 dev = map->stripes[i].dev;
6651 dev->commit_bytes_used = dev->bytes_used;
6654 unlock_chunks(root);