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);
55 struct list_head *btrfs_get_fs_uuids(void)
60 static struct btrfs_fs_devices *__alloc_fs_devices(void)
62 struct btrfs_fs_devices *fs_devs;
64 fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS);
66 return ERR_PTR(-ENOMEM);
68 mutex_init(&fs_devs->device_list_mutex);
70 INIT_LIST_HEAD(&fs_devs->devices);
71 INIT_LIST_HEAD(&fs_devs->resized_devices);
72 INIT_LIST_HEAD(&fs_devs->alloc_list);
73 INIT_LIST_HEAD(&fs_devs->list);
79 * alloc_fs_devices - allocate struct btrfs_fs_devices
80 * @fsid: a pointer to UUID for this FS. If NULL a new UUID is
83 * Return: a pointer to a new &struct btrfs_fs_devices on success;
84 * ERR_PTR() on error. Returned struct is not linked onto any lists and
85 * can be destroyed with kfree() right away.
87 static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid)
89 struct btrfs_fs_devices *fs_devs;
91 fs_devs = __alloc_fs_devices();
96 memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE);
98 generate_random_uuid(fs_devs->fsid);
103 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
105 struct btrfs_device *device;
106 WARN_ON(fs_devices->opened);
107 while (!list_empty(&fs_devices->devices)) {
108 device = list_entry(fs_devices->devices.next,
109 struct btrfs_device, dev_list);
110 list_del(&device->dev_list);
111 rcu_string_free(device->name);
117 static void btrfs_kobject_uevent(struct block_device *bdev,
118 enum kobject_action action)
122 ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
124 pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n",
126 kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
127 &disk_to_dev(bdev->bd_disk)->kobj);
130 void btrfs_cleanup_fs_uuids(void)
132 struct btrfs_fs_devices *fs_devices;
134 while (!list_empty(&fs_uuids)) {
135 fs_devices = list_entry(fs_uuids.next,
136 struct btrfs_fs_devices, list);
137 list_del(&fs_devices->list);
138 free_fs_devices(fs_devices);
142 static struct btrfs_device *__alloc_device(void)
144 struct btrfs_device *dev;
146 dev = kzalloc(sizeof(*dev), GFP_NOFS);
148 return ERR_PTR(-ENOMEM);
150 INIT_LIST_HEAD(&dev->dev_list);
151 INIT_LIST_HEAD(&dev->dev_alloc_list);
152 INIT_LIST_HEAD(&dev->resized_list);
154 spin_lock_init(&dev->io_lock);
156 spin_lock_init(&dev->reada_lock);
157 atomic_set(&dev->reada_in_flight, 0);
158 atomic_set(&dev->dev_stats_ccnt, 0);
159 INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT);
160 INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT);
165 static noinline struct btrfs_device *__find_device(struct list_head *head,
168 struct btrfs_device *dev;
170 list_for_each_entry(dev, head, dev_list) {
171 if (dev->devid == devid &&
172 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
179 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
181 struct btrfs_fs_devices *fs_devices;
183 list_for_each_entry(fs_devices, &fs_uuids, list) {
184 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
191 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
192 int flush, struct block_device **bdev,
193 struct buffer_head **bh)
197 *bdev = blkdev_get_by_path(device_path, flags, holder);
200 ret = PTR_ERR(*bdev);
201 printk(KERN_INFO "BTRFS: open %s failed\n", device_path);
206 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
207 ret = set_blocksize(*bdev, 4096);
209 blkdev_put(*bdev, flags);
212 invalidate_bdev(*bdev);
213 *bh = btrfs_read_dev_super(*bdev);
216 blkdev_put(*bdev, flags);
228 static void requeue_list(struct btrfs_pending_bios *pending_bios,
229 struct bio *head, struct bio *tail)
232 struct bio *old_head;
234 old_head = pending_bios->head;
235 pending_bios->head = head;
236 if (pending_bios->tail)
237 tail->bi_next = old_head;
239 pending_bios->tail = tail;
243 * we try to collect pending bios for a device so we don't get a large
244 * number of procs sending bios down to the same device. This greatly
245 * improves the schedulers ability to collect and merge the bios.
247 * But, it also turns into a long list of bios to process and that is sure
248 * to eventually make the worker thread block. The solution here is to
249 * make some progress and then put this work struct back at the end of
250 * the list if the block device is congested. This way, multiple devices
251 * can make progress from a single worker thread.
253 static noinline void run_scheduled_bios(struct btrfs_device *device)
256 struct backing_dev_info *bdi;
257 struct btrfs_fs_info *fs_info;
258 struct btrfs_pending_bios *pending_bios;
262 unsigned long num_run;
263 unsigned long batch_run = 0;
265 unsigned long last_waited = 0;
267 int sync_pending = 0;
268 struct blk_plug plug;
271 * this function runs all the bios we've collected for
272 * a particular device. We don't want to wander off to
273 * another device without first sending all of these down.
274 * So, setup a plug here and finish it off before we return
276 blk_start_plug(&plug);
278 bdi = blk_get_backing_dev_info(device->bdev);
279 fs_info = device->dev_root->fs_info;
280 limit = btrfs_async_submit_limit(fs_info);
281 limit = limit * 2 / 3;
284 spin_lock(&device->io_lock);
289 /* take all the bios off the list at once and process them
290 * later on (without the lock held). But, remember the
291 * tail and other pointers so the bios can be properly reinserted
292 * into the list if we hit congestion
294 if (!force_reg && device->pending_sync_bios.head) {
295 pending_bios = &device->pending_sync_bios;
298 pending_bios = &device->pending_bios;
302 pending = pending_bios->head;
303 tail = pending_bios->tail;
304 WARN_ON(pending && !tail);
307 * if pending was null this time around, no bios need processing
308 * at all and we can stop. Otherwise it'll loop back up again
309 * and do an additional check so no bios are missed.
311 * device->running_pending is used to synchronize with the
314 if (device->pending_sync_bios.head == NULL &&
315 device->pending_bios.head == NULL) {
317 device->running_pending = 0;
320 device->running_pending = 1;
323 pending_bios->head = NULL;
324 pending_bios->tail = NULL;
326 spin_unlock(&device->io_lock);
331 /* we want to work on both lists, but do more bios on the
332 * sync list than the regular list
335 pending_bios != &device->pending_sync_bios &&
336 device->pending_sync_bios.head) ||
337 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
338 device->pending_bios.head)) {
339 spin_lock(&device->io_lock);
340 requeue_list(pending_bios, pending, tail);
345 pending = pending->bi_next;
348 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
349 waitqueue_active(&fs_info->async_submit_wait))
350 wake_up(&fs_info->async_submit_wait);
352 BUG_ON(atomic_read(&cur->__bi_cnt) == 0);
355 * if we're doing the sync list, record that our
356 * plug has some sync requests on it
358 * If we're doing the regular list and there are
359 * sync requests sitting around, unplug before
362 if (pending_bios == &device->pending_sync_bios) {
364 } else if (sync_pending) {
365 blk_finish_plug(&plug);
366 blk_start_plug(&plug);
370 btrfsic_submit_bio(cur->bi_rw, cur);
377 * we made progress, there is more work to do and the bdi
378 * is now congested. Back off and let other work structs
381 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
382 fs_info->fs_devices->open_devices > 1) {
383 struct io_context *ioc;
385 ioc = current->io_context;
388 * the main goal here is that we don't want to
389 * block if we're going to be able to submit
390 * more requests without blocking.
392 * This code does two great things, it pokes into
393 * the elevator code from a filesystem _and_
394 * it makes assumptions about how batching works.
396 if (ioc && ioc->nr_batch_requests > 0 &&
397 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
399 ioc->last_waited == last_waited)) {
401 * we want to go through our batch of
402 * requests and stop. So, we copy out
403 * the ioc->last_waited time and test
404 * against it before looping
406 last_waited = ioc->last_waited;
410 spin_lock(&device->io_lock);
411 requeue_list(pending_bios, pending, tail);
412 device->running_pending = 1;
414 spin_unlock(&device->io_lock);
415 btrfs_queue_work(fs_info->submit_workers,
419 /* unplug every 64 requests just for good measure */
420 if (batch_run % 64 == 0) {
421 blk_finish_plug(&plug);
422 blk_start_plug(&plug);
431 spin_lock(&device->io_lock);
432 if (device->pending_bios.head || device->pending_sync_bios.head)
434 spin_unlock(&device->io_lock);
437 blk_finish_plug(&plug);
440 static void pending_bios_fn(struct btrfs_work *work)
442 struct btrfs_device *device;
444 device = container_of(work, struct btrfs_device, work);
445 run_scheduled_bios(device);
449 void btrfs_free_stale_device(struct btrfs_device *cur_dev)
451 struct btrfs_fs_devices *fs_devs;
452 struct btrfs_device *dev;
457 list_for_each_entry(fs_devs, &fs_uuids, list) {
462 if (fs_devs->seeding)
465 list_for_each_entry(dev, &fs_devs->devices, dev_list) {
473 * Todo: This won't be enough. What if the same device
474 * comes back (with new uuid and) with its mapper path?
475 * But for now, this does help as mostly an admin will
476 * either use mapper or non mapper path throughout.
479 del = strcmp(rcu_str_deref(dev->name),
480 rcu_str_deref(cur_dev->name));
487 /* delete the stale device */
488 if (fs_devs->num_devices == 1) {
489 btrfs_sysfs_remove_fsid(fs_devs);
490 list_del(&fs_devs->list);
491 free_fs_devices(fs_devs);
493 fs_devs->num_devices--;
494 list_del(&dev->dev_list);
495 rcu_string_free(dev->name);
504 * Add new device to list of registered devices
507 * 1 - first time device is seen
508 * 0 - device already known
511 static noinline int device_list_add(const char *path,
512 struct btrfs_super_block *disk_super,
513 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
515 struct btrfs_device *device;
516 struct btrfs_fs_devices *fs_devices;
517 struct rcu_string *name;
519 u64 found_transid = btrfs_super_generation(disk_super);
521 fs_devices = find_fsid(disk_super->fsid);
523 fs_devices = alloc_fs_devices(disk_super->fsid);
524 if (IS_ERR(fs_devices))
525 return PTR_ERR(fs_devices);
527 list_add(&fs_devices->list, &fs_uuids);
531 device = __find_device(&fs_devices->devices, devid,
532 disk_super->dev_item.uuid);
536 if (fs_devices->opened)
539 device = btrfs_alloc_device(NULL, &devid,
540 disk_super->dev_item.uuid);
541 if (IS_ERR(device)) {
542 /* we can safely leave the fs_devices entry around */
543 return PTR_ERR(device);
546 name = rcu_string_strdup(path, GFP_NOFS);
551 rcu_assign_pointer(device->name, name);
553 mutex_lock(&fs_devices->device_list_mutex);
554 list_add_rcu(&device->dev_list, &fs_devices->devices);
555 fs_devices->num_devices++;
556 mutex_unlock(&fs_devices->device_list_mutex);
559 device->fs_devices = fs_devices;
560 } else if (!device->name || strcmp(device->name->str, path)) {
562 * When FS is already mounted.
563 * 1. If you are here and if the device->name is NULL that
564 * means this device was missing at time of FS mount.
565 * 2. If you are here and if the device->name is different
566 * from 'path' that means either
567 * a. The same device disappeared and reappeared with
569 * b. The missing-disk-which-was-replaced, has
572 * We must allow 1 and 2a above. But 2b would be a spurious
575 * Further in case of 1 and 2a above, the disk at 'path'
576 * would have missed some transaction when it was away and
577 * in case of 2a the stale bdev has to be updated as well.
578 * 2b must not be allowed at all time.
582 * For now, we do allow update to btrfs_fs_device through the
583 * btrfs dev scan cli after FS has been mounted. We're still
584 * tracking a problem where systems fail mount by subvolume id
585 * when we reject replacement on a mounted FS.
587 if (!fs_devices->opened && found_transid < device->generation) {
589 * That is if the FS is _not_ mounted and if you
590 * are here, that means there is more than one
591 * disk with same uuid and devid.We keep the one
592 * with larger generation number or the last-in if
593 * generation are equal.
598 name = rcu_string_strdup(path, GFP_NOFS);
601 rcu_string_free(device->name);
602 rcu_assign_pointer(device->name, name);
603 if (device->missing) {
604 fs_devices->missing_devices--;
610 * Unmount does not free the btrfs_device struct but would zero
611 * generation along with most of the other members. So just update
612 * it back. We need it to pick the disk with largest generation
615 if (!fs_devices->opened)
616 device->generation = found_transid;
619 * if there is new btrfs on an already registered device,
620 * then remove the stale device entry.
622 btrfs_free_stale_device(device);
624 *fs_devices_ret = fs_devices;
629 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
631 struct btrfs_fs_devices *fs_devices;
632 struct btrfs_device *device;
633 struct btrfs_device *orig_dev;
635 fs_devices = alloc_fs_devices(orig->fsid);
636 if (IS_ERR(fs_devices))
639 mutex_lock(&orig->device_list_mutex);
640 fs_devices->total_devices = orig->total_devices;
642 /* We have held the volume lock, it is safe to get the devices. */
643 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
644 struct rcu_string *name;
646 device = btrfs_alloc_device(NULL, &orig_dev->devid,
652 * This is ok to do without rcu read locked because we hold the
653 * uuid mutex so nothing we touch in here is going to disappear.
655 if (orig_dev->name) {
656 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
661 rcu_assign_pointer(device->name, name);
664 list_add(&device->dev_list, &fs_devices->devices);
665 device->fs_devices = fs_devices;
666 fs_devices->num_devices++;
668 mutex_unlock(&orig->device_list_mutex);
671 mutex_unlock(&orig->device_list_mutex);
672 free_fs_devices(fs_devices);
673 return ERR_PTR(-ENOMEM);
676 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step)
678 struct btrfs_device *device, *next;
679 struct btrfs_device *latest_dev = NULL;
681 mutex_lock(&uuid_mutex);
683 /* This is the initialized path, it is safe to release the devices. */
684 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
685 if (device->in_fs_metadata) {
686 if (!device->is_tgtdev_for_dev_replace &&
688 device->generation > latest_dev->generation)) {
694 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
696 * In the first step, keep the device which has
697 * the correct fsid and the devid that is used
698 * for the dev_replace procedure.
699 * In the second step, the dev_replace state is
700 * read from the device tree and it is known
701 * whether the procedure is really active or
702 * not, which means whether this device is
703 * used or whether it should be removed.
705 if (step == 0 || device->is_tgtdev_for_dev_replace) {
710 blkdev_put(device->bdev, device->mode);
712 fs_devices->open_devices--;
714 if (device->writeable) {
715 list_del_init(&device->dev_alloc_list);
716 device->writeable = 0;
717 if (!device->is_tgtdev_for_dev_replace)
718 fs_devices->rw_devices--;
720 list_del_init(&device->dev_list);
721 fs_devices->num_devices--;
722 rcu_string_free(device->name);
726 if (fs_devices->seed) {
727 fs_devices = fs_devices->seed;
731 fs_devices->latest_bdev = latest_dev->bdev;
733 mutex_unlock(&uuid_mutex);
736 static void __free_device(struct work_struct *work)
738 struct btrfs_device *device;
740 device = container_of(work, struct btrfs_device, rcu_work);
743 blkdev_put(device->bdev, device->mode);
745 rcu_string_free(device->name);
749 static void free_device(struct rcu_head *head)
751 struct btrfs_device *device;
753 device = container_of(head, struct btrfs_device, rcu);
755 INIT_WORK(&device->rcu_work, __free_device);
756 schedule_work(&device->rcu_work);
759 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
761 struct btrfs_device *device, *tmp;
763 if (--fs_devices->opened > 0)
766 mutex_lock(&fs_devices->device_list_mutex);
767 list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) {
768 struct btrfs_device *new_device;
769 struct rcu_string *name;
772 fs_devices->open_devices--;
774 if (device->writeable &&
775 device->devid != BTRFS_DEV_REPLACE_DEVID) {
776 list_del_init(&device->dev_alloc_list);
777 fs_devices->rw_devices--;
781 fs_devices->missing_devices--;
783 new_device = btrfs_alloc_device(NULL, &device->devid,
785 BUG_ON(IS_ERR(new_device)); /* -ENOMEM */
787 /* Safe because we are under uuid_mutex */
789 name = rcu_string_strdup(device->name->str, GFP_NOFS);
790 BUG_ON(!name); /* -ENOMEM */
791 rcu_assign_pointer(new_device->name, name);
794 list_replace_rcu(&device->dev_list, &new_device->dev_list);
795 new_device->fs_devices = device->fs_devices;
797 call_rcu(&device->rcu, free_device);
799 mutex_unlock(&fs_devices->device_list_mutex);
801 WARN_ON(fs_devices->open_devices);
802 WARN_ON(fs_devices->rw_devices);
803 fs_devices->opened = 0;
804 fs_devices->seeding = 0;
809 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
811 struct btrfs_fs_devices *seed_devices = NULL;
814 mutex_lock(&uuid_mutex);
815 ret = __btrfs_close_devices(fs_devices);
816 if (!fs_devices->opened) {
817 seed_devices = fs_devices->seed;
818 fs_devices->seed = NULL;
820 mutex_unlock(&uuid_mutex);
822 while (seed_devices) {
823 fs_devices = seed_devices;
824 seed_devices = fs_devices->seed;
825 __btrfs_close_devices(fs_devices);
826 free_fs_devices(fs_devices);
829 * Wait for rcu kworkers under __btrfs_close_devices
830 * to finish all blkdev_puts so device is really
831 * free when umount is done.
837 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
838 fmode_t flags, void *holder)
840 struct request_queue *q;
841 struct block_device *bdev;
842 struct list_head *head = &fs_devices->devices;
843 struct btrfs_device *device;
844 struct btrfs_device *latest_dev = NULL;
845 struct buffer_head *bh;
846 struct btrfs_super_block *disk_super;
853 list_for_each_entry(device, head, dev_list) {
859 /* Just open everything we can; ignore failures here */
860 if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
864 disk_super = (struct btrfs_super_block *)bh->b_data;
865 devid = btrfs_stack_device_id(&disk_super->dev_item);
866 if (devid != device->devid)
869 if (memcmp(device->uuid, disk_super->dev_item.uuid,
873 device->generation = btrfs_super_generation(disk_super);
875 device->generation > latest_dev->generation)
878 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
879 device->writeable = 0;
881 device->writeable = !bdev_read_only(bdev);
885 q = bdev_get_queue(bdev);
886 if (blk_queue_discard(q))
887 device->can_discard = 1;
890 device->in_fs_metadata = 0;
891 device->mode = flags;
893 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
894 fs_devices->rotating = 1;
896 fs_devices->open_devices++;
897 if (device->writeable &&
898 device->devid != BTRFS_DEV_REPLACE_DEVID) {
899 fs_devices->rw_devices++;
900 list_add(&device->dev_alloc_list,
901 &fs_devices->alloc_list);
908 blkdev_put(bdev, flags);
911 if (fs_devices->open_devices == 0) {
915 fs_devices->seeding = seeding;
916 fs_devices->opened = 1;
917 fs_devices->latest_bdev = latest_dev->bdev;
918 fs_devices->total_rw_bytes = 0;
923 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
924 fmode_t flags, void *holder)
928 mutex_lock(&uuid_mutex);
929 if (fs_devices->opened) {
930 fs_devices->opened++;
933 ret = __btrfs_open_devices(fs_devices, flags, holder);
935 mutex_unlock(&uuid_mutex);
940 * Look for a btrfs signature on a device. This may be called out of the mount path
941 * and we are not allowed to call set_blocksize during the scan. The superblock
942 * is read via pagecache
944 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
945 struct btrfs_fs_devices **fs_devices_ret)
947 struct btrfs_super_block *disk_super;
948 struct block_device *bdev;
959 * we would like to check all the supers, but that would make
960 * a btrfs mount succeed after a mkfs from a different FS.
961 * So, we need to add a special mount option to scan for
962 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
964 bytenr = btrfs_sb_offset(0);
966 mutex_lock(&uuid_mutex);
968 bdev = blkdev_get_by_path(path, flags, holder);
975 /* make sure our super fits in the device */
976 if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
979 /* make sure our super fits in the page */
980 if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
983 /* make sure our super doesn't straddle pages on disk */
984 index = bytenr >> PAGE_CACHE_SHIFT;
985 if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
988 /* pull in the page with our super */
989 page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
992 if (IS_ERR_OR_NULL(page))
997 /* align our pointer to the offset of the super block */
998 disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
1000 if (btrfs_super_bytenr(disk_super) != bytenr ||
1001 btrfs_super_magic(disk_super) != BTRFS_MAGIC)
1004 devid = btrfs_stack_device_id(&disk_super->dev_item);
1005 transid = btrfs_super_generation(disk_super);
1006 total_devices = btrfs_super_num_devices(disk_super);
1008 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
1010 if (disk_super->label[0]) {
1011 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
1012 disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
1013 printk(KERN_INFO "BTRFS: device label %s ", disk_super->label);
1015 printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid);
1018 printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path);
1021 if (!ret && fs_devices_ret)
1022 (*fs_devices_ret)->total_devices = total_devices;
1026 page_cache_release(page);
1029 blkdev_put(bdev, flags);
1031 mutex_unlock(&uuid_mutex);
1035 /* helper to account the used device space in the range */
1036 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
1037 u64 end, u64 *length)
1039 struct btrfs_key key;
1040 struct btrfs_root *root = device->dev_root;
1041 struct btrfs_dev_extent *dev_extent;
1042 struct btrfs_path *path;
1046 struct extent_buffer *l;
1050 if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
1053 path = btrfs_alloc_path();
1058 key.objectid = device->devid;
1060 key.type = BTRFS_DEV_EXTENT_KEY;
1062 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1066 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1073 slot = path->slots[0];
1074 if (slot >= btrfs_header_nritems(l)) {
1075 ret = btrfs_next_leaf(root, path);
1083 btrfs_item_key_to_cpu(l, &key, slot);
1085 if (key.objectid < device->devid)
1088 if (key.objectid > device->devid)
1091 if (key.type != BTRFS_DEV_EXTENT_KEY)
1094 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1095 extent_end = key.offset + btrfs_dev_extent_length(l,
1097 if (key.offset <= start && extent_end > end) {
1098 *length = end - start + 1;
1100 } else if (key.offset <= start && extent_end > start)
1101 *length += extent_end - start;
1102 else if (key.offset > start && extent_end <= end)
1103 *length += extent_end - key.offset;
1104 else if (key.offset > start && key.offset <= end) {
1105 *length += end - key.offset + 1;
1107 } else if (key.offset > end)
1115 btrfs_free_path(path);
1119 static int contains_pending_extent(struct btrfs_trans_handle *trans,
1120 struct btrfs_device *device,
1121 u64 *start, u64 len)
1123 struct extent_map *em;
1124 struct list_head *search_list = &trans->transaction->pending_chunks;
1126 u64 physical_start = *start;
1129 list_for_each_entry(em, search_list, list) {
1130 struct map_lookup *map;
1133 map = (struct map_lookup *)em->bdev;
1134 for (i = 0; i < map->num_stripes; i++) {
1137 if (map->stripes[i].dev != device)
1139 if (map->stripes[i].physical >= physical_start + len ||
1140 map->stripes[i].physical + em->orig_block_len <=
1144 * Make sure that while processing the pinned list we do
1145 * not override our *start with a lower value, because
1146 * we can have pinned chunks that fall within this
1147 * device hole and that have lower physical addresses
1148 * than the pending chunks we processed before. If we
1149 * do not take this special care we can end up getting
1150 * 2 pending chunks that start at the same physical
1151 * device offsets because the end offset of a pinned
1152 * chunk can be equal to the start offset of some
1155 end = map->stripes[i].physical + em->orig_block_len;
1162 if (search_list == &trans->transaction->pending_chunks) {
1163 search_list = &trans->root->fs_info->pinned_chunks;
1172 * find_free_dev_extent - find free space in the specified device
1173 * @device: the device which we search the free space in
1174 * @num_bytes: the size of the free space that we need
1175 * @start: store the start of the free space.
1176 * @len: the size of the free space. that we find, or the size of the max
1177 * free space if we don't find suitable free space
1179 * this uses a pretty simple search, the expectation is that it is
1180 * called very infrequently and that a given device has a small number
1183 * @start is used to store the start of the free space if we find. But if we
1184 * don't find suitable free space, it will be used to store the start position
1185 * of the max free space.
1187 * @len is used to store the size of the free space that we find.
1188 * But if we don't find suitable free space, it is used to store the size of
1189 * the max free space.
1191 int find_free_dev_extent(struct btrfs_trans_handle *trans,
1192 struct btrfs_device *device, u64 num_bytes,
1193 u64 *start, u64 *len)
1195 struct btrfs_key key;
1196 struct btrfs_root *root = device->dev_root;
1197 struct btrfs_dev_extent *dev_extent;
1198 struct btrfs_path *path;
1204 u64 search_end = device->total_bytes;
1207 struct extent_buffer *l;
1209 /* FIXME use last free of some kind */
1211 /* we don't want to overwrite the superblock on the drive,
1212 * so we make sure to start at an offset of at least 1MB
1214 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1216 path = btrfs_alloc_path();
1220 max_hole_start = search_start;
1224 if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1230 path->search_commit_root = 1;
1231 path->skip_locking = 1;
1233 key.objectid = device->devid;
1234 key.offset = search_start;
1235 key.type = BTRFS_DEV_EXTENT_KEY;
1237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1241 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1248 slot = path->slots[0];
1249 if (slot >= btrfs_header_nritems(l)) {
1250 ret = btrfs_next_leaf(root, path);
1258 btrfs_item_key_to_cpu(l, &key, slot);
1260 if (key.objectid < device->devid)
1263 if (key.objectid > device->devid)
1266 if (key.type != BTRFS_DEV_EXTENT_KEY)
1269 if (key.offset > search_start) {
1270 hole_size = key.offset - search_start;
1273 * Have to check before we set max_hole_start, otherwise
1274 * we could end up sending back this offset anyway.
1276 if (contains_pending_extent(trans, device,
1279 if (key.offset >= search_start) {
1280 hole_size = key.offset - search_start;
1287 if (hole_size > max_hole_size) {
1288 max_hole_start = search_start;
1289 max_hole_size = hole_size;
1293 * If this free space is greater than which we need,
1294 * it must be the max free space that we have found
1295 * until now, so max_hole_start must point to the start
1296 * of this free space and the length of this free space
1297 * is stored in max_hole_size. Thus, we return
1298 * max_hole_start and max_hole_size and go back to the
1301 if (hole_size >= num_bytes) {
1307 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1308 extent_end = key.offset + btrfs_dev_extent_length(l,
1310 if (extent_end > search_start)
1311 search_start = extent_end;
1318 * At this point, search_start should be the end of
1319 * allocated dev extents, and when shrinking the device,
1320 * search_end may be smaller than search_start.
1322 if (search_end > search_start) {
1323 hole_size = search_end - search_start;
1325 if (contains_pending_extent(trans, device, &search_start,
1327 btrfs_release_path(path);
1331 if (hole_size > max_hole_size) {
1332 max_hole_start = search_start;
1333 max_hole_size = hole_size;
1338 if (max_hole_size < num_bytes)
1344 btrfs_free_path(path);
1345 *start = max_hole_start;
1347 *len = max_hole_size;
1351 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1352 struct btrfs_device *device,
1353 u64 start, u64 *dev_extent_len)
1356 struct btrfs_path *path;
1357 struct btrfs_root *root = device->dev_root;
1358 struct btrfs_key key;
1359 struct btrfs_key found_key;
1360 struct extent_buffer *leaf = NULL;
1361 struct btrfs_dev_extent *extent = NULL;
1363 path = btrfs_alloc_path();
1367 key.objectid = device->devid;
1369 key.type = BTRFS_DEV_EXTENT_KEY;
1371 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1373 ret = btrfs_previous_item(root, path, key.objectid,
1374 BTRFS_DEV_EXTENT_KEY);
1377 leaf = path->nodes[0];
1378 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1379 extent = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_dev_extent);
1381 BUG_ON(found_key.offset > start || found_key.offset +
1382 btrfs_dev_extent_length(leaf, extent) < start);
1384 btrfs_release_path(path);
1386 } else if (ret == 0) {
1387 leaf = path->nodes[0];
1388 extent = btrfs_item_ptr(leaf, path->slots[0],
1389 struct btrfs_dev_extent);
1391 btrfs_error(root->fs_info, ret, "Slot search failed");
1395 *dev_extent_len = btrfs_dev_extent_length(leaf, extent);
1397 ret = btrfs_del_item(trans, root, path);
1399 btrfs_error(root->fs_info, ret,
1400 "Failed to remove dev extent item");
1402 trans->transaction->have_free_bgs = 1;
1405 btrfs_free_path(path);
1409 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1410 struct btrfs_device *device,
1411 u64 chunk_tree, u64 chunk_objectid,
1412 u64 chunk_offset, u64 start, u64 num_bytes)
1415 struct btrfs_path *path;
1416 struct btrfs_root *root = device->dev_root;
1417 struct btrfs_dev_extent *extent;
1418 struct extent_buffer *leaf;
1419 struct btrfs_key key;
1421 WARN_ON(!device->in_fs_metadata);
1422 WARN_ON(device->is_tgtdev_for_dev_replace);
1423 path = btrfs_alloc_path();
1427 key.objectid = device->devid;
1429 key.type = BTRFS_DEV_EXTENT_KEY;
1430 ret = btrfs_insert_empty_item(trans, root, path, &key,
1435 leaf = path->nodes[0];
1436 extent = btrfs_item_ptr(leaf, path->slots[0],
1437 struct btrfs_dev_extent);
1438 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1439 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1440 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1442 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1443 btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE);
1445 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1446 btrfs_mark_buffer_dirty(leaf);
1448 btrfs_free_path(path);
1452 static u64 find_next_chunk(struct btrfs_fs_info *fs_info)
1454 struct extent_map_tree *em_tree;
1455 struct extent_map *em;
1459 em_tree = &fs_info->mapping_tree.map_tree;
1460 read_lock(&em_tree->lock);
1461 n = rb_last(&em_tree->map);
1463 em = rb_entry(n, struct extent_map, rb_node);
1464 ret = em->start + em->len;
1466 read_unlock(&em_tree->lock);
1471 static noinline int find_next_devid(struct btrfs_fs_info *fs_info,
1475 struct btrfs_key key;
1476 struct btrfs_key found_key;
1477 struct btrfs_path *path;
1479 path = btrfs_alloc_path();
1483 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1484 key.type = BTRFS_DEV_ITEM_KEY;
1485 key.offset = (u64)-1;
1487 ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0);
1491 BUG_ON(ret == 0); /* Corruption */
1493 ret = btrfs_previous_item(fs_info->chunk_root, path,
1494 BTRFS_DEV_ITEMS_OBJECTID,
1495 BTRFS_DEV_ITEM_KEY);
1499 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1501 *devid_ret = found_key.offset + 1;
1505 btrfs_free_path(path);
1510 * the device information is stored in the chunk root
1511 * the btrfs_device struct should be fully filled in
1513 static int btrfs_add_device(struct btrfs_trans_handle *trans,
1514 struct btrfs_root *root,
1515 struct btrfs_device *device)
1518 struct btrfs_path *path;
1519 struct btrfs_dev_item *dev_item;
1520 struct extent_buffer *leaf;
1521 struct btrfs_key key;
1524 root = root->fs_info->chunk_root;
1526 path = btrfs_alloc_path();
1530 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1531 key.type = BTRFS_DEV_ITEM_KEY;
1532 key.offset = device->devid;
1534 ret = btrfs_insert_empty_item(trans, root, path, &key,
1539 leaf = path->nodes[0];
1540 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1542 btrfs_set_device_id(leaf, dev_item, device->devid);
1543 btrfs_set_device_generation(leaf, dev_item, 0);
1544 btrfs_set_device_type(leaf, dev_item, device->type);
1545 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1546 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1547 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1548 btrfs_set_device_total_bytes(leaf, dev_item,
1549 btrfs_device_get_disk_total_bytes(device));
1550 btrfs_set_device_bytes_used(leaf, dev_item,
1551 btrfs_device_get_bytes_used(device));
1552 btrfs_set_device_group(leaf, dev_item, 0);
1553 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1554 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1555 btrfs_set_device_start_offset(leaf, dev_item, 0);
1557 ptr = btrfs_device_uuid(dev_item);
1558 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1559 ptr = btrfs_device_fsid(dev_item);
1560 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1561 btrfs_mark_buffer_dirty(leaf);
1565 btrfs_free_path(path);
1570 * Function to update ctime/mtime for a given device path.
1571 * Mainly used for ctime/mtime based probe like libblkid.
1573 static void update_dev_time(char *path_name)
1577 filp = filp_open(path_name, O_RDWR, 0);
1580 file_update_time(filp);
1581 filp_close(filp, NULL);
1585 static int btrfs_rm_dev_item(struct btrfs_root *root,
1586 struct btrfs_device *device)
1589 struct btrfs_path *path;
1590 struct btrfs_key key;
1591 struct btrfs_trans_handle *trans;
1593 root = root->fs_info->chunk_root;
1595 path = btrfs_alloc_path();
1599 trans = btrfs_start_transaction(root, 0);
1600 if (IS_ERR(trans)) {
1601 btrfs_free_path(path);
1602 return PTR_ERR(trans);
1604 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1605 key.type = BTRFS_DEV_ITEM_KEY;
1606 key.offset = device->devid;
1608 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1617 ret = btrfs_del_item(trans, root, path);
1621 btrfs_free_path(path);
1622 btrfs_commit_transaction(trans, root);
1626 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1628 struct btrfs_device *device;
1629 struct btrfs_device *next_device;
1630 struct block_device *bdev;
1631 struct buffer_head *bh = NULL;
1632 struct btrfs_super_block *disk_super;
1633 struct btrfs_fs_devices *cur_devices;
1640 bool clear_super = false;
1642 mutex_lock(&uuid_mutex);
1645 seq = read_seqbegin(&root->fs_info->profiles_lock);
1647 all_avail = root->fs_info->avail_data_alloc_bits |
1648 root->fs_info->avail_system_alloc_bits |
1649 root->fs_info->avail_metadata_alloc_bits;
1650 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1652 num_devices = root->fs_info->fs_devices->num_devices;
1653 btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1654 if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1655 WARN_ON(num_devices < 1);
1658 btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1660 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1661 ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET;
1665 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1666 ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET;
1670 if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1671 root->fs_info->fs_devices->rw_devices <= 2) {
1672 ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET;
1675 if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1676 root->fs_info->fs_devices->rw_devices <= 3) {
1677 ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET;
1681 if (strcmp(device_path, "missing") == 0) {
1682 struct list_head *devices;
1683 struct btrfs_device *tmp;
1686 devices = &root->fs_info->fs_devices->devices;
1688 * It is safe to read the devices since the volume_mutex
1691 list_for_each_entry(tmp, devices, dev_list) {
1692 if (tmp->in_fs_metadata &&
1693 !tmp->is_tgtdev_for_dev_replace &&
1703 ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND;
1707 ret = btrfs_get_bdev_and_sb(device_path,
1708 FMODE_WRITE | FMODE_EXCL,
1709 root->fs_info->bdev_holder, 0,
1713 disk_super = (struct btrfs_super_block *)bh->b_data;
1714 devid = btrfs_stack_device_id(&disk_super->dev_item);
1715 dev_uuid = disk_super->dev_item.uuid;
1716 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1724 if (device->is_tgtdev_for_dev_replace) {
1725 ret = BTRFS_ERROR_DEV_TGT_REPLACE;
1729 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1730 ret = BTRFS_ERROR_DEV_ONLY_WRITABLE;
1734 if (device->writeable) {
1736 list_del_init(&device->dev_alloc_list);
1737 device->fs_devices->rw_devices--;
1738 unlock_chunks(root);
1742 mutex_unlock(&uuid_mutex);
1743 ret = btrfs_shrink_device(device, 0);
1744 mutex_lock(&uuid_mutex);
1749 * TODO: the superblock still includes this device in its num_devices
1750 * counter although write_all_supers() is not locked out. This
1751 * could give a filesystem state which requires a degraded mount.
1753 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1757 device->in_fs_metadata = 0;
1758 btrfs_scrub_cancel_dev(root->fs_info, device);
1761 * the device list mutex makes sure that we don't change
1762 * the device list while someone else is writing out all
1763 * the device supers. Whoever is writing all supers, should
1764 * lock the device list mutex before getting the number of
1765 * devices in the super block (super_copy). Conversely,
1766 * whoever updates the number of devices in the super block
1767 * (super_copy) should hold the device list mutex.
1770 cur_devices = device->fs_devices;
1771 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1772 list_del_rcu(&device->dev_list);
1774 device->fs_devices->num_devices--;
1775 device->fs_devices->total_devices--;
1777 if (device->missing)
1778 device->fs_devices->missing_devices--;
1780 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1781 struct btrfs_device, dev_list);
1782 if (device->bdev == root->fs_info->sb->s_bdev)
1783 root->fs_info->sb->s_bdev = next_device->bdev;
1784 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1785 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1788 device->fs_devices->open_devices--;
1789 /* remove sysfs entry */
1790 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
1793 call_rcu(&device->rcu, free_device);
1795 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1796 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1797 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1799 if (cur_devices->open_devices == 0) {
1800 struct btrfs_fs_devices *fs_devices;
1801 fs_devices = root->fs_info->fs_devices;
1802 while (fs_devices) {
1803 if (fs_devices->seed == cur_devices) {
1804 fs_devices->seed = cur_devices->seed;
1807 fs_devices = fs_devices->seed;
1809 cur_devices->seed = NULL;
1810 __btrfs_close_devices(cur_devices);
1811 free_fs_devices(cur_devices);
1814 root->fs_info->num_tolerated_disk_barrier_failures =
1815 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1818 * at this point, the device is zero sized. We want to
1819 * remove it from the devices list and zero out the old super
1821 if (clear_super && disk_super) {
1825 /* make sure this device isn't detected as part of
1828 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1829 set_buffer_dirty(bh);
1830 sync_dirty_buffer(bh);
1832 /* clear the mirror copies of super block on the disk
1833 * being removed, 0th copy is been taken care above and
1834 * the below would take of the rest
1836 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1837 bytenr = btrfs_sb_offset(i);
1838 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
1839 i_size_read(bdev->bd_inode))
1843 bh = __bread(bdev, bytenr / 4096,
1844 BTRFS_SUPER_INFO_SIZE);
1848 disk_super = (struct btrfs_super_block *)bh->b_data;
1850 if (btrfs_super_bytenr(disk_super) != bytenr ||
1851 btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
1854 memset(&disk_super->magic, 0,
1855 sizeof(disk_super->magic));
1856 set_buffer_dirty(bh);
1857 sync_dirty_buffer(bh);
1864 /* Notify udev that device has changed */
1865 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1867 /* Update ctime/mtime for device path for libblkid */
1868 update_dev_time(device_path);
1874 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1876 mutex_unlock(&uuid_mutex);
1879 if (device->writeable) {
1881 list_add(&device->dev_alloc_list,
1882 &root->fs_info->fs_devices->alloc_list);
1883 device->fs_devices->rw_devices++;
1884 unlock_chunks(root);
1889 void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info,
1890 struct btrfs_device *srcdev)
1892 struct btrfs_fs_devices *fs_devices;
1894 WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1897 * in case of fs with no seed, srcdev->fs_devices will point
1898 * to fs_devices of fs_info. However when the dev being replaced is
1899 * a seed dev it will point to the seed's local fs_devices. In short
1900 * srcdev will have its correct fs_devices in both the cases.
1902 fs_devices = srcdev->fs_devices;
1904 list_del_rcu(&srcdev->dev_list);
1905 list_del_rcu(&srcdev->dev_alloc_list);
1906 fs_devices->num_devices--;
1907 if (srcdev->missing)
1908 fs_devices->missing_devices--;
1910 if (srcdev->writeable) {
1911 fs_devices->rw_devices--;
1912 /* zero out the old super if it is writable */
1913 btrfs_scratch_superblock(srcdev);
1917 fs_devices->open_devices--;
1920 void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info,
1921 struct btrfs_device *srcdev)
1923 struct btrfs_fs_devices *fs_devices = srcdev->fs_devices;
1925 call_rcu(&srcdev->rcu, free_device);
1928 * unless fs_devices is seed fs, num_devices shouldn't go
1931 BUG_ON(!fs_devices->num_devices && !fs_devices->seeding);
1933 /* if this is no devs we rather delete the fs_devices */
1934 if (!fs_devices->num_devices) {
1935 struct btrfs_fs_devices *tmp_fs_devices;
1937 tmp_fs_devices = fs_info->fs_devices;
1938 while (tmp_fs_devices) {
1939 if (tmp_fs_devices->seed == fs_devices) {
1940 tmp_fs_devices->seed = fs_devices->seed;
1943 tmp_fs_devices = tmp_fs_devices->seed;
1945 fs_devices->seed = NULL;
1946 __btrfs_close_devices(fs_devices);
1947 free_fs_devices(fs_devices);
1951 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1952 struct btrfs_device *tgtdev)
1954 struct btrfs_device *next_device;
1956 mutex_lock(&uuid_mutex);
1958 mutex_lock(&fs_info->fs_devices->device_list_mutex);
1960 btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev);
1963 btrfs_scratch_superblock(tgtdev);
1964 fs_info->fs_devices->open_devices--;
1966 fs_info->fs_devices->num_devices--;
1968 next_device = list_entry(fs_info->fs_devices->devices.next,
1969 struct btrfs_device, dev_list);
1970 if (tgtdev->bdev == fs_info->sb->s_bdev)
1971 fs_info->sb->s_bdev = next_device->bdev;
1972 if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1973 fs_info->fs_devices->latest_bdev = next_device->bdev;
1974 list_del_rcu(&tgtdev->dev_list);
1976 call_rcu(&tgtdev->rcu, free_device);
1978 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1979 mutex_unlock(&uuid_mutex);
1982 static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1983 struct btrfs_device **device)
1986 struct btrfs_super_block *disk_super;
1989 struct block_device *bdev;
1990 struct buffer_head *bh;
1993 ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1994 root->fs_info->bdev_holder, 0, &bdev, &bh);
1997 disk_super = (struct btrfs_super_block *)bh->b_data;
1998 devid = btrfs_stack_device_id(&disk_super->dev_item);
1999 dev_uuid = disk_super->dev_item.uuid;
2000 *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2005 blkdev_put(bdev, FMODE_READ);
2009 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
2011 struct btrfs_device **device)
2014 if (strcmp(device_path, "missing") == 0) {
2015 struct list_head *devices;
2016 struct btrfs_device *tmp;
2018 devices = &root->fs_info->fs_devices->devices;
2020 * It is safe to read the devices since the volume_mutex
2021 * is held by the caller.
2023 list_for_each_entry(tmp, devices, dev_list) {
2024 if (tmp->in_fs_metadata && !tmp->bdev) {
2031 btrfs_err(root->fs_info, "no missing device found");
2037 return btrfs_find_device_by_path(root, device_path, device);
2042 * does all the dirty work required for changing file system's UUID.
2044 static int btrfs_prepare_sprout(struct btrfs_root *root)
2046 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2047 struct btrfs_fs_devices *old_devices;
2048 struct btrfs_fs_devices *seed_devices;
2049 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
2050 struct btrfs_device *device;
2053 BUG_ON(!mutex_is_locked(&uuid_mutex));
2054 if (!fs_devices->seeding)
2057 seed_devices = __alloc_fs_devices();
2058 if (IS_ERR(seed_devices))
2059 return PTR_ERR(seed_devices);
2061 old_devices = clone_fs_devices(fs_devices);
2062 if (IS_ERR(old_devices)) {
2063 kfree(seed_devices);
2064 return PTR_ERR(old_devices);
2067 list_add(&old_devices->list, &fs_uuids);
2069 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
2070 seed_devices->opened = 1;
2071 INIT_LIST_HEAD(&seed_devices->devices);
2072 INIT_LIST_HEAD(&seed_devices->alloc_list);
2073 mutex_init(&seed_devices->device_list_mutex);
2075 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2076 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
2078 list_for_each_entry(device, &seed_devices->devices, dev_list)
2079 device->fs_devices = seed_devices;
2082 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
2083 unlock_chunks(root);
2085 fs_devices->seeding = 0;
2086 fs_devices->num_devices = 0;
2087 fs_devices->open_devices = 0;
2088 fs_devices->missing_devices = 0;
2089 fs_devices->rotating = 0;
2090 fs_devices->seed = seed_devices;
2092 generate_random_uuid(fs_devices->fsid);
2093 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2094 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
2095 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2097 super_flags = btrfs_super_flags(disk_super) &
2098 ~BTRFS_SUPER_FLAG_SEEDING;
2099 btrfs_set_super_flags(disk_super, super_flags);
2105 * strore the expected generation for seed devices in device items.
2107 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *root)
2110 struct btrfs_path *path;
2111 struct extent_buffer *leaf;
2112 struct btrfs_dev_item *dev_item;
2113 struct btrfs_device *device;
2114 struct btrfs_key key;
2115 u8 fs_uuid[BTRFS_UUID_SIZE];
2116 u8 dev_uuid[BTRFS_UUID_SIZE];
2120 path = btrfs_alloc_path();
2124 root = root->fs_info->chunk_root;
2125 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2127 key.type = BTRFS_DEV_ITEM_KEY;
2130 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2134 leaf = path->nodes[0];
2136 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2137 ret = btrfs_next_leaf(root, path);
2142 leaf = path->nodes[0];
2143 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2144 btrfs_release_path(path);
2148 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2149 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
2150 key.type != BTRFS_DEV_ITEM_KEY)
2153 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2154 struct btrfs_dev_item);
2155 devid = btrfs_device_id(leaf, dev_item);
2156 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
2158 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
2160 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
2162 BUG_ON(!device); /* Logic error */
2164 if (device->fs_devices->seeding) {
2165 btrfs_set_device_generation(leaf, dev_item,
2166 device->generation);
2167 btrfs_mark_buffer_dirty(leaf);
2175 btrfs_free_path(path);
2179 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
2181 struct request_queue *q;
2182 struct btrfs_trans_handle *trans;
2183 struct btrfs_device *device;
2184 struct block_device *bdev;
2185 struct list_head *devices;
2186 struct super_block *sb = root->fs_info->sb;
2187 struct rcu_string *name;
2189 int seeding_dev = 0;
2192 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
2195 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2196 root->fs_info->bdev_holder);
2198 return PTR_ERR(bdev);
2200 if (root->fs_info->fs_devices->seeding) {
2202 down_write(&sb->s_umount);
2203 mutex_lock(&uuid_mutex);
2206 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2208 devices = &root->fs_info->fs_devices->devices;
2210 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2211 list_for_each_entry(device, devices, dev_list) {
2212 if (device->bdev == bdev) {
2215 &root->fs_info->fs_devices->device_list_mutex);
2219 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2221 device = btrfs_alloc_device(root->fs_info, NULL, NULL);
2222 if (IS_ERR(device)) {
2223 /* we can safely leave the fs_devices entry around */
2224 ret = PTR_ERR(device);
2228 name = rcu_string_strdup(device_path, GFP_NOFS);
2234 rcu_assign_pointer(device->name, name);
2236 trans = btrfs_start_transaction(root, 0);
2237 if (IS_ERR(trans)) {
2238 rcu_string_free(device->name);
2240 ret = PTR_ERR(trans);
2244 q = bdev_get_queue(bdev);
2245 if (blk_queue_discard(q))
2246 device->can_discard = 1;
2247 device->writeable = 1;
2248 device->generation = trans->transid;
2249 device->io_width = root->sectorsize;
2250 device->io_align = root->sectorsize;
2251 device->sector_size = root->sectorsize;
2252 device->total_bytes = i_size_read(bdev->bd_inode);
2253 device->disk_total_bytes = device->total_bytes;
2254 device->commit_total_bytes = device->total_bytes;
2255 device->dev_root = root->fs_info->dev_root;
2256 device->bdev = bdev;
2257 device->in_fs_metadata = 1;
2258 device->is_tgtdev_for_dev_replace = 0;
2259 device->mode = FMODE_EXCL;
2260 device->dev_stats_valid = 1;
2261 set_blocksize(device->bdev, 4096);
2264 sb->s_flags &= ~MS_RDONLY;
2265 ret = btrfs_prepare_sprout(root);
2266 BUG_ON(ret); /* -ENOMEM */
2269 device->fs_devices = root->fs_info->fs_devices;
2271 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2273 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2274 list_add(&device->dev_alloc_list,
2275 &root->fs_info->fs_devices->alloc_list);
2276 root->fs_info->fs_devices->num_devices++;
2277 root->fs_info->fs_devices->open_devices++;
2278 root->fs_info->fs_devices->rw_devices++;
2279 root->fs_info->fs_devices->total_devices++;
2280 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2282 spin_lock(&root->fs_info->free_chunk_lock);
2283 root->fs_info->free_chunk_space += device->total_bytes;
2284 spin_unlock(&root->fs_info->free_chunk_lock);
2286 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2287 root->fs_info->fs_devices->rotating = 1;
2289 tmp = btrfs_super_total_bytes(root->fs_info->super_copy);
2290 btrfs_set_super_total_bytes(root->fs_info->super_copy,
2291 tmp + device->total_bytes);
2293 tmp = btrfs_super_num_devices(root->fs_info->super_copy);
2294 btrfs_set_super_num_devices(root->fs_info->super_copy,
2297 /* add sysfs device entry */
2298 btrfs_kobj_add_device(root->fs_info->fs_devices, device);
2301 * we've got more storage, clear any full flags on the space
2304 btrfs_clear_space_info_full(root->fs_info);
2306 unlock_chunks(root);
2307 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2311 ret = init_first_rw_device(trans, root, device);
2312 unlock_chunks(root);
2314 btrfs_abort_transaction(trans, root, ret);
2319 ret = btrfs_add_device(trans, root, device);
2321 btrfs_abort_transaction(trans, root, ret);
2326 char fsid_buf[BTRFS_UUID_UNPARSED_SIZE];
2328 ret = btrfs_finish_sprout(trans, root);
2330 btrfs_abort_transaction(trans, root, ret);
2334 /* Sprouting would change fsid of the mounted root,
2335 * so rename the fsid on the sysfs
2337 snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU",
2338 root->fs_info->fsid);
2339 if (kobject_rename(&root->fs_info->fs_devices->super_kobj,
2341 pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n");
2344 root->fs_info->num_tolerated_disk_barrier_failures =
2345 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2346 ret = btrfs_commit_transaction(trans, root);
2349 mutex_unlock(&uuid_mutex);
2350 up_write(&sb->s_umount);
2352 if (ret) /* transaction commit */
2355 ret = btrfs_relocate_sys_chunks(root);
2357 btrfs_error(root->fs_info, ret,
2358 "Failed to relocate sys chunks after "
2359 "device initialization. This can be fixed "
2360 "using the \"btrfs balance\" command.");
2361 trans = btrfs_attach_transaction(root);
2362 if (IS_ERR(trans)) {
2363 if (PTR_ERR(trans) == -ENOENT)
2365 return PTR_ERR(trans);
2367 ret = btrfs_commit_transaction(trans, root);
2370 /* Update ctime/mtime for libblkid */
2371 update_dev_time(device_path);
2375 btrfs_end_transaction(trans, root);
2376 rcu_string_free(device->name);
2377 btrfs_kobj_rm_device(root->fs_info->fs_devices, device);
2380 blkdev_put(bdev, FMODE_EXCL);
2382 mutex_unlock(&uuid_mutex);
2383 up_write(&sb->s_umount);
2388 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2389 struct btrfs_device *srcdev,
2390 struct btrfs_device **device_out)
2392 struct request_queue *q;
2393 struct btrfs_device *device;
2394 struct block_device *bdev;
2395 struct btrfs_fs_info *fs_info = root->fs_info;
2396 struct list_head *devices;
2397 struct rcu_string *name;
2398 u64 devid = BTRFS_DEV_REPLACE_DEVID;
2402 if (fs_info->fs_devices->seeding) {
2403 btrfs_err(fs_info, "the filesystem is a seed filesystem!");
2407 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2408 fs_info->bdev_holder);
2410 btrfs_err(fs_info, "target device %s is invalid!", device_path);
2411 return PTR_ERR(bdev);
2414 filemap_write_and_wait(bdev->bd_inode->i_mapping);
2416 devices = &fs_info->fs_devices->devices;
2417 list_for_each_entry(device, devices, dev_list) {
2418 if (device->bdev == bdev) {
2419 btrfs_err(fs_info, "target device is in the filesystem!");
2426 if (i_size_read(bdev->bd_inode) <
2427 btrfs_device_get_total_bytes(srcdev)) {
2428 btrfs_err(fs_info, "target device is smaller than source device!");
2434 device = btrfs_alloc_device(NULL, &devid, NULL);
2435 if (IS_ERR(device)) {
2436 ret = PTR_ERR(device);
2440 name = rcu_string_strdup(device_path, GFP_NOFS);
2446 rcu_assign_pointer(device->name, name);
2448 q = bdev_get_queue(bdev);
2449 if (blk_queue_discard(q))
2450 device->can_discard = 1;
2451 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2452 device->writeable = 1;
2453 device->generation = 0;
2454 device->io_width = root->sectorsize;
2455 device->io_align = root->sectorsize;
2456 device->sector_size = root->sectorsize;
2457 device->total_bytes = btrfs_device_get_total_bytes(srcdev);
2458 device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev);
2459 device->bytes_used = btrfs_device_get_bytes_used(srcdev);
2460 ASSERT(list_empty(&srcdev->resized_list));
2461 device->commit_total_bytes = srcdev->commit_total_bytes;
2462 device->commit_bytes_used = device->bytes_used;
2463 device->dev_root = fs_info->dev_root;
2464 device->bdev = bdev;
2465 device->in_fs_metadata = 1;
2466 device->is_tgtdev_for_dev_replace = 1;
2467 device->mode = FMODE_EXCL;
2468 device->dev_stats_valid = 1;
2469 set_blocksize(device->bdev, 4096);
2470 device->fs_devices = fs_info->fs_devices;
2471 list_add(&device->dev_list, &fs_info->fs_devices->devices);
2472 fs_info->fs_devices->num_devices++;
2473 fs_info->fs_devices->open_devices++;
2474 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2476 *device_out = device;
2480 blkdev_put(bdev, FMODE_EXCL);
2484 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2485 struct btrfs_device *tgtdev)
2487 WARN_ON(fs_info->fs_devices->rw_devices == 0);
2488 tgtdev->io_width = fs_info->dev_root->sectorsize;
2489 tgtdev->io_align = fs_info->dev_root->sectorsize;
2490 tgtdev->sector_size = fs_info->dev_root->sectorsize;
2491 tgtdev->dev_root = fs_info->dev_root;
2492 tgtdev->in_fs_metadata = 1;
2495 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2496 struct btrfs_device *device)
2499 struct btrfs_path *path;
2500 struct btrfs_root *root;
2501 struct btrfs_dev_item *dev_item;
2502 struct extent_buffer *leaf;
2503 struct btrfs_key key;
2505 root = device->dev_root->fs_info->chunk_root;
2507 path = btrfs_alloc_path();
2511 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2512 key.type = BTRFS_DEV_ITEM_KEY;
2513 key.offset = device->devid;
2515 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2524 leaf = path->nodes[0];
2525 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2527 btrfs_set_device_id(leaf, dev_item, device->devid);
2528 btrfs_set_device_type(leaf, dev_item, device->type);
2529 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2530 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2531 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2532 btrfs_set_device_total_bytes(leaf, dev_item,
2533 btrfs_device_get_disk_total_bytes(device));
2534 btrfs_set_device_bytes_used(leaf, dev_item,
2535 btrfs_device_get_bytes_used(device));
2536 btrfs_mark_buffer_dirty(leaf);
2539 btrfs_free_path(path);
2543 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2544 struct btrfs_device *device, u64 new_size)
2546 struct btrfs_super_block *super_copy =
2547 device->dev_root->fs_info->super_copy;
2548 struct btrfs_fs_devices *fs_devices;
2552 if (!device->writeable)
2555 lock_chunks(device->dev_root);
2556 old_total = btrfs_super_total_bytes(super_copy);
2557 diff = new_size - device->total_bytes;
2559 if (new_size <= device->total_bytes ||
2560 device->is_tgtdev_for_dev_replace) {
2561 unlock_chunks(device->dev_root);
2565 fs_devices = device->dev_root->fs_info->fs_devices;
2567 btrfs_set_super_total_bytes(super_copy, old_total + diff);
2568 device->fs_devices->total_rw_bytes += diff;
2570 btrfs_device_set_total_bytes(device, new_size);
2571 btrfs_device_set_disk_total_bytes(device, new_size);
2572 btrfs_clear_space_info_full(device->dev_root->fs_info);
2573 if (list_empty(&device->resized_list))
2574 list_add_tail(&device->resized_list,
2575 &fs_devices->resized_devices);
2576 unlock_chunks(device->dev_root);
2578 return btrfs_update_device(trans, device);
2581 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2582 struct btrfs_root *root, u64 chunk_objectid,
2586 struct btrfs_path *path;
2587 struct btrfs_key key;
2589 root = root->fs_info->chunk_root;
2590 path = btrfs_alloc_path();
2594 key.objectid = chunk_objectid;
2595 key.offset = chunk_offset;
2596 key.type = BTRFS_CHUNK_ITEM_KEY;
2598 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2601 else if (ret > 0) { /* Logic error or corruption */
2602 btrfs_error(root->fs_info, -ENOENT,
2603 "Failed lookup while freeing chunk.");
2608 ret = btrfs_del_item(trans, root, path);
2610 btrfs_error(root->fs_info, ret,
2611 "Failed to delete chunk item.");
2613 btrfs_free_path(path);
2617 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2620 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2621 struct btrfs_disk_key *disk_key;
2622 struct btrfs_chunk *chunk;
2629 struct btrfs_key key;
2632 array_size = btrfs_super_sys_array_size(super_copy);
2634 ptr = super_copy->sys_chunk_array;
2637 while (cur < array_size) {
2638 disk_key = (struct btrfs_disk_key *)ptr;
2639 btrfs_disk_key_to_cpu(&key, disk_key);
2641 len = sizeof(*disk_key);
2643 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2644 chunk = (struct btrfs_chunk *)(ptr + len);
2645 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2646 len += btrfs_chunk_item_size(num_stripes);
2651 if (key.objectid == chunk_objectid &&
2652 key.offset == chunk_offset) {
2653 memmove(ptr, ptr + len, array_size - (cur + len));
2655 btrfs_set_super_sys_array_size(super_copy, array_size);
2661 unlock_chunks(root);
2665 int btrfs_remove_chunk(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *root, u64 chunk_offset)
2668 struct extent_map_tree *em_tree;
2669 struct extent_map *em;
2670 struct btrfs_root *extent_root = root->fs_info->extent_root;
2671 struct map_lookup *map;
2672 u64 dev_extent_len = 0;
2673 u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2677 root = root->fs_info->chunk_root;
2678 em_tree = &root->fs_info->mapping_tree.map_tree;
2680 read_lock(&em_tree->lock);
2681 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2682 read_unlock(&em_tree->lock);
2684 if (!em || em->start > chunk_offset ||
2685 em->start + em->len < chunk_offset) {
2687 * This is a logic error, but we don't want to just rely on the
2688 * user having built with ASSERT enabled, so if ASSERT doens't
2689 * do anything we still error out.
2693 free_extent_map(em);
2696 map = (struct map_lookup *)em->bdev;
2697 lock_chunks(root->fs_info->chunk_root);
2698 check_system_chunk(trans, extent_root, map->type);
2699 unlock_chunks(root->fs_info->chunk_root);
2701 for (i = 0; i < map->num_stripes; i++) {
2702 struct btrfs_device *device = map->stripes[i].dev;
2703 ret = btrfs_free_dev_extent(trans, device,
2704 map->stripes[i].physical,
2707 btrfs_abort_transaction(trans, root, ret);
2711 if (device->bytes_used > 0) {
2713 btrfs_device_set_bytes_used(device,
2714 device->bytes_used - dev_extent_len);
2715 spin_lock(&root->fs_info->free_chunk_lock);
2716 root->fs_info->free_chunk_space += dev_extent_len;
2717 spin_unlock(&root->fs_info->free_chunk_lock);
2718 btrfs_clear_space_info_full(root->fs_info);
2719 unlock_chunks(root);
2722 if (map->stripes[i].dev) {
2723 ret = btrfs_update_device(trans, map->stripes[i].dev);
2725 btrfs_abort_transaction(trans, root, ret);
2730 ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset);
2732 btrfs_abort_transaction(trans, root, ret);
2736 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2738 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2739 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2741 btrfs_abort_transaction(trans, root, ret);
2746 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em);
2748 btrfs_abort_transaction(trans, extent_root, ret);
2754 free_extent_map(em);
2758 static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset)
2760 struct btrfs_root *extent_root;
2761 struct btrfs_trans_handle *trans;
2764 root = root->fs_info->chunk_root;
2765 extent_root = root->fs_info->extent_root;
2768 * Prevent races with automatic removal of unused block groups.
2769 * After we relocate and before we remove the chunk with offset
2770 * chunk_offset, automatic removal of the block group can kick in,
2771 * resulting in a failure when calling btrfs_remove_chunk() below.
2773 * Make sure to acquire this mutex before doing a tree search (dev
2774 * or chunk trees) to find chunks. Otherwise the cleaner kthread might
2775 * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after
2776 * we release the path used to search the chunk/dev tree and before
2777 * the current task acquires this mutex and calls us.
2779 ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex));
2781 ret = btrfs_can_relocate(extent_root, chunk_offset);
2785 /* step one, relocate all the extents inside this chunk */
2786 btrfs_scrub_pause(root);
2787 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2788 btrfs_scrub_continue(root);
2792 trans = btrfs_start_transaction(root, 0);
2793 if (IS_ERR(trans)) {
2794 ret = PTR_ERR(trans);
2795 btrfs_std_error(root->fs_info, ret);
2800 * step two, delete the device extents and the
2801 * chunk tree entries
2803 ret = btrfs_remove_chunk(trans, root, chunk_offset);
2804 btrfs_end_transaction(trans, root);
2808 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2810 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2811 struct btrfs_path *path;
2812 struct extent_buffer *leaf;
2813 struct btrfs_chunk *chunk;
2814 struct btrfs_key key;
2815 struct btrfs_key found_key;
2817 bool retried = false;
2821 path = btrfs_alloc_path();
2826 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2827 key.offset = (u64)-1;
2828 key.type = BTRFS_CHUNK_ITEM_KEY;
2831 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
2832 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2834 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2837 BUG_ON(ret == 0); /* Corruption */
2839 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2842 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2848 leaf = path->nodes[0];
2849 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2851 chunk = btrfs_item_ptr(leaf, path->slots[0],
2852 struct btrfs_chunk);
2853 chunk_type = btrfs_chunk_type(leaf, chunk);
2854 btrfs_release_path(path);
2856 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2857 ret = btrfs_relocate_chunk(chunk_root,
2864 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
2866 if (found_key.offset == 0)
2868 key.offset = found_key.offset - 1;
2871 if (failed && !retried) {
2875 } else if (WARN_ON(failed && retried)) {
2879 btrfs_free_path(path);
2883 static int insert_balance_item(struct btrfs_root *root,
2884 struct btrfs_balance_control *bctl)
2886 struct btrfs_trans_handle *trans;
2887 struct btrfs_balance_item *item;
2888 struct btrfs_disk_balance_args disk_bargs;
2889 struct btrfs_path *path;
2890 struct extent_buffer *leaf;
2891 struct btrfs_key key;
2894 path = btrfs_alloc_path();
2898 trans = btrfs_start_transaction(root, 0);
2899 if (IS_ERR(trans)) {
2900 btrfs_free_path(path);
2901 return PTR_ERR(trans);
2904 key.objectid = BTRFS_BALANCE_OBJECTID;
2905 key.type = BTRFS_BALANCE_ITEM_KEY;
2908 ret = btrfs_insert_empty_item(trans, root, path, &key,
2913 leaf = path->nodes[0];
2914 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2916 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2918 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2919 btrfs_set_balance_data(leaf, item, &disk_bargs);
2920 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2921 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2922 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2923 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2925 btrfs_set_balance_flags(leaf, item, bctl->flags);
2927 btrfs_mark_buffer_dirty(leaf);
2929 btrfs_free_path(path);
2930 err = btrfs_commit_transaction(trans, root);
2936 static int del_balance_item(struct btrfs_root *root)
2938 struct btrfs_trans_handle *trans;
2939 struct btrfs_path *path;
2940 struct btrfs_key key;
2943 path = btrfs_alloc_path();
2947 trans = btrfs_start_transaction(root, 0);
2948 if (IS_ERR(trans)) {
2949 btrfs_free_path(path);
2950 return PTR_ERR(trans);
2953 key.objectid = BTRFS_BALANCE_OBJECTID;
2954 key.type = BTRFS_BALANCE_ITEM_KEY;
2957 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2965 ret = btrfs_del_item(trans, root, path);
2967 btrfs_free_path(path);
2968 err = btrfs_commit_transaction(trans, root);
2975 * This is a heuristic used to reduce the number of chunks balanced on
2976 * resume after balance was interrupted.
2978 static void update_balance_args(struct btrfs_balance_control *bctl)
2981 * Turn on soft mode for chunk types that were being converted.
2983 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2984 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2985 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2986 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2987 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2988 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2991 * Turn on usage filter if is not already used. The idea is
2992 * that chunks that we have already balanced should be
2993 * reasonably full. Don't do it for chunks that are being
2994 * converted - that will keep us from relocating unconverted
2995 * (albeit full) chunks.
2997 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2998 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2999 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
3000 bctl->data.usage = 90;
3002 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3003 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3004 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
3005 bctl->sys.usage = 90;
3007 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
3008 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
3009 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
3010 bctl->meta.usage = 90;
3015 * Should be called with both balance and volume mutexes held to
3016 * serialize other volume operations (add_dev/rm_dev/resize) with
3017 * restriper. Same goes for unset_balance_control.
3019 static void set_balance_control(struct btrfs_balance_control *bctl)
3021 struct btrfs_fs_info *fs_info = bctl->fs_info;
3023 BUG_ON(fs_info->balance_ctl);
3025 spin_lock(&fs_info->balance_lock);
3026 fs_info->balance_ctl = bctl;
3027 spin_unlock(&fs_info->balance_lock);
3030 static void unset_balance_control(struct btrfs_fs_info *fs_info)
3032 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3034 BUG_ON(!fs_info->balance_ctl);
3036 spin_lock(&fs_info->balance_lock);
3037 fs_info->balance_ctl = NULL;
3038 spin_unlock(&fs_info->balance_lock);
3044 * Balance filters. Return 1 if chunk should be filtered out
3045 * (should not be balanced).
3047 static int chunk_profiles_filter(u64 chunk_type,
3048 struct btrfs_balance_args *bargs)
3050 chunk_type = chunk_to_extended(chunk_type) &
3051 BTRFS_EXTENDED_PROFILE_MASK;
3053 if (bargs->profiles & chunk_type)
3059 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
3060 struct btrfs_balance_args *bargs)
3062 struct btrfs_block_group_cache *cache;
3063 u64 chunk_used, user_thresh;
3066 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
3067 chunk_used = btrfs_block_group_used(&cache->item);
3069 if (bargs->usage == 0)
3071 else if (bargs->usage > 100)
3072 user_thresh = cache->key.offset;
3074 user_thresh = div_factor_fine(cache->key.offset,
3077 if (chunk_used < user_thresh)
3080 btrfs_put_block_group(cache);
3084 static int chunk_devid_filter(struct extent_buffer *leaf,
3085 struct btrfs_chunk *chunk,
3086 struct btrfs_balance_args *bargs)
3088 struct btrfs_stripe *stripe;
3089 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3092 for (i = 0; i < num_stripes; i++) {
3093 stripe = btrfs_stripe_nr(chunk, i);
3094 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
3101 /* [pstart, pend) */
3102 static int chunk_drange_filter(struct extent_buffer *leaf,
3103 struct btrfs_chunk *chunk,
3105 struct btrfs_balance_args *bargs)
3107 struct btrfs_stripe *stripe;
3108 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3114 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
3117 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
3118 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
3119 factor = num_stripes / 2;
3120 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
3121 factor = num_stripes - 1;
3122 } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
3123 factor = num_stripes - 2;
3125 factor = num_stripes;
3128 for (i = 0; i < num_stripes; i++) {
3129 stripe = btrfs_stripe_nr(chunk, i);
3130 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
3133 stripe_offset = btrfs_stripe_offset(leaf, stripe);
3134 stripe_length = btrfs_chunk_length(leaf, chunk);
3135 stripe_length = div_u64(stripe_length, factor);
3137 if (stripe_offset < bargs->pend &&
3138 stripe_offset + stripe_length > bargs->pstart)
3145 /* [vstart, vend) */
3146 static int chunk_vrange_filter(struct extent_buffer *leaf,
3147 struct btrfs_chunk *chunk,
3149 struct btrfs_balance_args *bargs)
3151 if (chunk_offset < bargs->vend &&
3152 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
3153 /* at least part of the chunk is inside this vrange */
3159 static int chunk_soft_convert_filter(u64 chunk_type,
3160 struct btrfs_balance_args *bargs)
3162 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
3165 chunk_type = chunk_to_extended(chunk_type) &
3166 BTRFS_EXTENDED_PROFILE_MASK;
3168 if (bargs->target == chunk_type)
3174 static int should_balance_chunk(struct btrfs_root *root,
3175 struct extent_buffer *leaf,
3176 struct btrfs_chunk *chunk, u64 chunk_offset)
3178 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
3179 struct btrfs_balance_args *bargs = NULL;
3180 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
3183 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
3184 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
3188 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
3189 bargs = &bctl->data;
3190 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
3192 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
3193 bargs = &bctl->meta;
3195 /* profiles filter */
3196 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
3197 chunk_profiles_filter(chunk_type, bargs)) {
3202 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
3203 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
3208 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
3209 chunk_devid_filter(leaf, chunk, bargs)) {
3213 /* drange filter, makes sense only with devid filter */
3214 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
3215 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
3220 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
3221 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
3225 /* soft profile changing mode */
3226 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
3227 chunk_soft_convert_filter(chunk_type, bargs)) {
3232 * limited by count, must be the last filter
3234 if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) {
3235 if (bargs->limit == 0)
3244 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
3246 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3247 struct btrfs_root *chunk_root = fs_info->chunk_root;
3248 struct btrfs_root *dev_root = fs_info->dev_root;
3249 struct list_head *devices;
3250 struct btrfs_device *device;
3253 struct btrfs_chunk *chunk;
3254 struct btrfs_path *path;
3255 struct btrfs_key key;
3256 struct btrfs_key found_key;
3257 struct btrfs_trans_handle *trans;
3258 struct extent_buffer *leaf;
3261 int enospc_errors = 0;
3262 bool counting = true;
3263 u64 limit_data = bctl->data.limit;
3264 u64 limit_meta = bctl->meta.limit;
3265 u64 limit_sys = bctl->sys.limit;
3267 /* step one make some room on all the devices */
3268 devices = &fs_info->fs_devices->devices;
3269 list_for_each_entry(device, devices, dev_list) {
3270 old_size = btrfs_device_get_total_bytes(device);
3271 size_to_free = div_factor(old_size, 1);
3272 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
3273 if (!device->writeable ||
3274 btrfs_device_get_total_bytes(device) -
3275 btrfs_device_get_bytes_used(device) > size_to_free ||
3276 device->is_tgtdev_for_dev_replace)
3279 ret = btrfs_shrink_device(device, old_size - size_to_free);
3284 trans = btrfs_start_transaction(dev_root, 0);
3285 BUG_ON(IS_ERR(trans));
3287 ret = btrfs_grow_device(trans, device, old_size);
3290 btrfs_end_transaction(trans, dev_root);
3293 /* step two, relocate all the chunks */
3294 path = btrfs_alloc_path();
3300 /* zero out stat counters */
3301 spin_lock(&fs_info->balance_lock);
3302 memset(&bctl->stat, 0, sizeof(bctl->stat));
3303 spin_unlock(&fs_info->balance_lock);
3306 bctl->data.limit = limit_data;
3307 bctl->meta.limit = limit_meta;
3308 bctl->sys.limit = limit_sys;
3310 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3311 key.offset = (u64)-1;
3312 key.type = BTRFS_CHUNK_ITEM_KEY;
3315 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
3316 atomic_read(&fs_info->balance_cancel_req)) {
3321 mutex_lock(&fs_info->delete_unused_bgs_mutex);
3322 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
3324 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3329 * this shouldn't happen, it means the last relocate
3333 BUG(); /* FIXME break ? */
3335 ret = btrfs_previous_item(chunk_root, path, 0,
3336 BTRFS_CHUNK_ITEM_KEY);
3338 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3343 leaf = path->nodes[0];
3344 slot = path->slots[0];
3345 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3347 if (found_key.objectid != key.objectid) {
3348 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3352 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3355 spin_lock(&fs_info->balance_lock);
3356 bctl->stat.considered++;
3357 spin_unlock(&fs_info->balance_lock);
3360 ret = should_balance_chunk(chunk_root, leaf, chunk,
3362 btrfs_release_path(path);
3364 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3369 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3370 spin_lock(&fs_info->balance_lock);
3371 bctl->stat.expected++;
3372 spin_unlock(&fs_info->balance_lock);
3376 ret = btrfs_relocate_chunk(chunk_root,
3378 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
3379 if (ret && ret != -ENOSPC)
3381 if (ret == -ENOSPC) {
3384 spin_lock(&fs_info->balance_lock);
3385 bctl->stat.completed++;
3386 spin_unlock(&fs_info->balance_lock);
3389 if (found_key.offset == 0)
3391 key.offset = found_key.offset - 1;
3395 btrfs_release_path(path);
3400 btrfs_free_path(path);
3401 if (enospc_errors) {
3402 btrfs_info(fs_info, "%d enospc errors during balance",
3412 * alloc_profile_is_valid - see if a given profile is valid and reduced
3413 * @flags: profile to validate
3414 * @extended: if true @flags is treated as an extended profile
3416 static int alloc_profile_is_valid(u64 flags, int extended)
3418 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3419 BTRFS_BLOCK_GROUP_PROFILE_MASK);
3421 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3423 /* 1) check that all other bits are zeroed */
3427 /* 2) see if profile is reduced */
3429 return !extended; /* "0" is valid for usual profiles */
3431 /* true if exactly one bit set */
3432 return (flags & (flags - 1)) == 0;
3435 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3437 /* cancel requested || normal exit path */
3438 return atomic_read(&fs_info->balance_cancel_req) ||
3439 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3440 atomic_read(&fs_info->balance_cancel_req) == 0);
3443 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3447 unset_balance_control(fs_info);
3448 ret = del_balance_item(fs_info->tree_root);
3450 btrfs_std_error(fs_info, ret);
3452 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3456 * Should be called with both balance and volume mutexes held
3458 int btrfs_balance(struct btrfs_balance_control *bctl,
3459 struct btrfs_ioctl_balance_args *bargs)
3461 struct btrfs_fs_info *fs_info = bctl->fs_info;
3468 if (btrfs_fs_closing(fs_info) ||
3469 atomic_read(&fs_info->balance_pause_req) ||
3470 atomic_read(&fs_info->balance_cancel_req)) {
3475 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3476 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3480 * In case of mixed groups both data and meta should be picked,
3481 * and identical options should be given for both of them.
3483 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3484 if (mixed && (bctl->flags & allowed)) {
3485 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3486 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3487 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3488 btrfs_err(fs_info, "with mixed groups data and "
3489 "metadata balance options must be the same");
3495 num_devices = fs_info->fs_devices->num_devices;
3496 btrfs_dev_replace_lock(&fs_info->dev_replace);
3497 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3498 BUG_ON(num_devices < 1);
3501 btrfs_dev_replace_unlock(&fs_info->dev_replace);
3502 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3503 if (num_devices == 1)
3504 allowed |= BTRFS_BLOCK_GROUP_DUP;
3505 else if (num_devices > 1)
3506 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3507 if (num_devices > 2)
3508 allowed |= BTRFS_BLOCK_GROUP_RAID5;
3509 if (num_devices > 3)
3510 allowed |= (BTRFS_BLOCK_GROUP_RAID10 |
3511 BTRFS_BLOCK_GROUP_RAID6);
3512 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3513 (!alloc_profile_is_valid(bctl->data.target, 1) ||
3514 (bctl->data.target & ~allowed))) {
3515 btrfs_err(fs_info, "unable to start balance with target "
3516 "data profile %llu",
3521 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3522 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3523 (bctl->meta.target & ~allowed))) {
3525 "unable to start balance with target metadata profile %llu",
3530 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3531 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3532 (bctl->sys.target & ~allowed))) {
3534 "unable to start balance with target system profile %llu",
3540 /* allow dup'ed data chunks only in mixed mode */
3541 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3542 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3543 btrfs_err(fs_info, "dup for data is not allowed");
3548 /* allow to reduce meta or sys integrity only if force set */
3549 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3550 BTRFS_BLOCK_GROUP_RAID10 |
3551 BTRFS_BLOCK_GROUP_RAID5 |
3552 BTRFS_BLOCK_GROUP_RAID6;
3554 seq = read_seqbegin(&fs_info->profiles_lock);
3556 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3557 (fs_info->avail_system_alloc_bits & allowed) &&
3558 !(bctl->sys.target & allowed)) ||
3559 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3560 (fs_info->avail_metadata_alloc_bits & allowed) &&
3561 !(bctl->meta.target & allowed))) {
3562 if (bctl->flags & BTRFS_BALANCE_FORCE) {
3563 btrfs_info(fs_info, "force reducing metadata integrity");
3565 btrfs_err(fs_info, "balance will reduce metadata "
3566 "integrity, use force if you want this");
3571 } while (read_seqretry(&fs_info->profiles_lock, seq));
3573 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3574 int num_tolerated_disk_barrier_failures;
3575 u64 target = bctl->sys.target;
3577 num_tolerated_disk_barrier_failures =
3578 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3579 if (num_tolerated_disk_barrier_failures > 0 &&
3581 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3582 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3583 num_tolerated_disk_barrier_failures = 0;
3584 else if (num_tolerated_disk_barrier_failures > 1 &&
3586 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3587 num_tolerated_disk_barrier_failures = 1;
3589 fs_info->num_tolerated_disk_barrier_failures =
3590 num_tolerated_disk_barrier_failures;
3593 ret = insert_balance_item(fs_info->tree_root, bctl);
3594 if (ret && ret != -EEXIST)
3597 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3598 BUG_ON(ret == -EEXIST);
3599 set_balance_control(bctl);
3601 BUG_ON(ret != -EEXIST);
3602 spin_lock(&fs_info->balance_lock);
3603 update_balance_args(bctl);
3604 spin_unlock(&fs_info->balance_lock);
3607 atomic_inc(&fs_info->balance_running);
3608 mutex_unlock(&fs_info->balance_mutex);
3610 ret = __btrfs_balance(fs_info);
3612 mutex_lock(&fs_info->balance_mutex);
3613 atomic_dec(&fs_info->balance_running);
3615 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3616 fs_info->num_tolerated_disk_barrier_failures =
3617 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3621 memset(bargs, 0, sizeof(*bargs));
3622 update_ioctl_balance_args(fs_info, 0, bargs);
3625 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
3626 balance_need_close(fs_info)) {
3627 __cancel_balance(fs_info);
3630 wake_up(&fs_info->balance_wait_q);
3634 if (bctl->flags & BTRFS_BALANCE_RESUME)
3635 __cancel_balance(fs_info);
3638 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3643 static int balance_kthread(void *data)
3645 struct btrfs_fs_info *fs_info = data;
3648 mutex_lock(&fs_info->volume_mutex);
3649 mutex_lock(&fs_info->balance_mutex);
3651 if (fs_info->balance_ctl) {
3652 btrfs_info(fs_info, "continuing balance");
3653 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3656 mutex_unlock(&fs_info->balance_mutex);
3657 mutex_unlock(&fs_info->volume_mutex);
3662 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3664 struct task_struct *tsk;
3666 spin_lock(&fs_info->balance_lock);
3667 if (!fs_info->balance_ctl) {
3668 spin_unlock(&fs_info->balance_lock);
3671 spin_unlock(&fs_info->balance_lock);
3673 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3674 btrfs_info(fs_info, "force skipping balance");
3678 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3679 return PTR_ERR_OR_ZERO(tsk);
3682 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3684 struct btrfs_balance_control *bctl;
3685 struct btrfs_balance_item *item;
3686 struct btrfs_disk_balance_args disk_bargs;
3687 struct btrfs_path *path;
3688 struct extent_buffer *leaf;
3689 struct btrfs_key key;
3692 path = btrfs_alloc_path();
3696 key.objectid = BTRFS_BALANCE_OBJECTID;
3697 key.type = BTRFS_BALANCE_ITEM_KEY;
3700 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3703 if (ret > 0) { /* ret = -ENOENT; */
3708 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3714 leaf = path->nodes[0];
3715 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3717 bctl->fs_info = fs_info;
3718 bctl->flags = btrfs_balance_flags(leaf, item);
3719 bctl->flags |= BTRFS_BALANCE_RESUME;
3721 btrfs_balance_data(leaf, item, &disk_bargs);
3722 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3723 btrfs_balance_meta(leaf, item, &disk_bargs);
3724 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3725 btrfs_balance_sys(leaf, item, &disk_bargs);
3726 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3728 WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3730 mutex_lock(&fs_info->volume_mutex);
3731 mutex_lock(&fs_info->balance_mutex);
3733 set_balance_control(bctl);
3735 mutex_unlock(&fs_info->balance_mutex);
3736 mutex_unlock(&fs_info->volume_mutex);
3738 btrfs_free_path(path);
3742 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3746 mutex_lock(&fs_info->balance_mutex);
3747 if (!fs_info->balance_ctl) {
3748 mutex_unlock(&fs_info->balance_mutex);
3752 if (atomic_read(&fs_info->balance_running)) {
3753 atomic_inc(&fs_info->balance_pause_req);
3754 mutex_unlock(&fs_info->balance_mutex);
3756 wait_event(fs_info->balance_wait_q,
3757 atomic_read(&fs_info->balance_running) == 0);
3759 mutex_lock(&fs_info->balance_mutex);
3760 /* we are good with balance_ctl ripped off from under us */
3761 BUG_ON(atomic_read(&fs_info->balance_running));
3762 atomic_dec(&fs_info->balance_pause_req);
3767 mutex_unlock(&fs_info->balance_mutex);
3771 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3773 if (fs_info->sb->s_flags & MS_RDONLY)
3776 mutex_lock(&fs_info->balance_mutex);
3777 if (!fs_info->balance_ctl) {
3778 mutex_unlock(&fs_info->balance_mutex);
3782 atomic_inc(&fs_info->balance_cancel_req);
3784 * if we are running just wait and return, balance item is
3785 * deleted in btrfs_balance in this case
3787 if (atomic_read(&fs_info->balance_running)) {
3788 mutex_unlock(&fs_info->balance_mutex);
3789 wait_event(fs_info->balance_wait_q,
3790 atomic_read(&fs_info->balance_running) == 0);
3791 mutex_lock(&fs_info->balance_mutex);
3793 /* __cancel_balance needs volume_mutex */
3794 mutex_unlock(&fs_info->balance_mutex);
3795 mutex_lock(&fs_info->volume_mutex);
3796 mutex_lock(&fs_info->balance_mutex);
3798 if (fs_info->balance_ctl)
3799 __cancel_balance(fs_info);
3801 mutex_unlock(&fs_info->volume_mutex);
3804 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3805 atomic_dec(&fs_info->balance_cancel_req);
3806 mutex_unlock(&fs_info->balance_mutex);
3810 static int btrfs_uuid_scan_kthread(void *data)
3812 struct btrfs_fs_info *fs_info = data;
3813 struct btrfs_root *root = fs_info->tree_root;
3814 struct btrfs_key key;
3815 struct btrfs_key max_key;
3816 struct btrfs_path *path = NULL;
3818 struct extent_buffer *eb;
3820 struct btrfs_root_item root_item;
3822 struct btrfs_trans_handle *trans = NULL;
3824 path = btrfs_alloc_path();
3831 key.type = BTRFS_ROOT_ITEM_KEY;
3834 max_key.objectid = (u64)-1;
3835 max_key.type = BTRFS_ROOT_ITEM_KEY;
3836 max_key.offset = (u64)-1;
3839 ret = btrfs_search_forward(root, &key, path, 0);
3846 if (key.type != BTRFS_ROOT_ITEM_KEY ||
3847 (key.objectid < BTRFS_FIRST_FREE_OBJECTID &&
3848 key.objectid != BTRFS_FS_TREE_OBJECTID) ||
3849 key.objectid > BTRFS_LAST_FREE_OBJECTID)
3852 eb = path->nodes[0];
3853 slot = path->slots[0];
3854 item_size = btrfs_item_size_nr(eb, slot);
3855 if (item_size < sizeof(root_item))
3858 read_extent_buffer(eb, &root_item,
3859 btrfs_item_ptr_offset(eb, slot),
3860 (int)sizeof(root_item));
3861 if (btrfs_root_refs(&root_item) == 0)
3864 if (!btrfs_is_empty_uuid(root_item.uuid) ||
3865 !btrfs_is_empty_uuid(root_item.received_uuid)) {
3869 btrfs_release_path(path);
3871 * 1 - subvol uuid item
3872 * 1 - received_subvol uuid item
3874 trans = btrfs_start_transaction(fs_info->uuid_root, 2);
3875 if (IS_ERR(trans)) {
3876 ret = PTR_ERR(trans);
3884 if (!btrfs_is_empty_uuid(root_item.uuid)) {
3885 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3887 BTRFS_UUID_KEY_SUBVOL,
3890 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3896 if (!btrfs_is_empty_uuid(root_item.received_uuid)) {
3897 ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root,
3898 root_item.received_uuid,
3899 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
3902 btrfs_warn(fs_info, "uuid_tree_add failed %d",
3910 ret = btrfs_end_transaction(trans, fs_info->uuid_root);
3916 btrfs_release_path(path);
3917 if (key.offset < (u64)-1) {
3919 } else if (key.type < BTRFS_ROOT_ITEM_KEY) {
3921 key.type = BTRFS_ROOT_ITEM_KEY;
3922 } else if (key.objectid < (u64)-1) {
3924 key.type = BTRFS_ROOT_ITEM_KEY;
3933 btrfs_free_path(path);
3934 if (trans && !IS_ERR(trans))
3935 btrfs_end_transaction(trans, fs_info->uuid_root);
3937 btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret);
3939 fs_info->update_uuid_tree_gen = 1;
3940 up(&fs_info->uuid_tree_rescan_sem);
3945 * Callback for btrfs_uuid_tree_iterate().
3947 * 0 check succeeded, the entry is not outdated.
3948 * < 0 if an error occured.
3949 * > 0 if the check failed, which means the caller shall remove the entry.
3951 static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info,
3952 u8 *uuid, u8 type, u64 subid)
3954 struct btrfs_key key;
3956 struct btrfs_root *subvol_root;
3958 if (type != BTRFS_UUID_KEY_SUBVOL &&
3959 type != BTRFS_UUID_KEY_RECEIVED_SUBVOL)
3962 key.objectid = subid;
3963 key.type = BTRFS_ROOT_ITEM_KEY;
3964 key.offset = (u64)-1;
3965 subvol_root = btrfs_read_fs_root_no_name(fs_info, &key);
3966 if (IS_ERR(subvol_root)) {
3967 ret = PTR_ERR(subvol_root);
3974 case BTRFS_UUID_KEY_SUBVOL:
3975 if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE))
3978 case BTRFS_UUID_KEY_RECEIVED_SUBVOL:
3979 if (memcmp(uuid, subvol_root->root_item.received_uuid,
3989 static int btrfs_uuid_rescan_kthread(void *data)
3991 struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data;
3995 * 1st step is to iterate through the existing UUID tree and
3996 * to delete all entries that contain outdated data.
3997 * 2nd step is to add all missing entries to the UUID tree.
3999 ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry);
4001 btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret);
4002 up(&fs_info->uuid_tree_rescan_sem);
4005 return btrfs_uuid_scan_kthread(data);
4008 int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info)
4010 struct btrfs_trans_handle *trans;
4011 struct btrfs_root *tree_root = fs_info->tree_root;
4012 struct btrfs_root *uuid_root;
4013 struct task_struct *task;
4020 trans = btrfs_start_transaction(tree_root, 2);
4022 return PTR_ERR(trans);
4024 uuid_root = btrfs_create_tree(trans, fs_info,
4025 BTRFS_UUID_TREE_OBJECTID);
4026 if (IS_ERR(uuid_root)) {
4027 ret = PTR_ERR(uuid_root);
4028 btrfs_abort_transaction(trans, tree_root, ret);
4032 fs_info->uuid_root = uuid_root;
4034 ret = btrfs_commit_transaction(trans, tree_root);
4038 down(&fs_info->uuid_tree_rescan_sem);
4039 task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid");
4041 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4042 btrfs_warn(fs_info, "failed to start uuid_scan task");
4043 up(&fs_info->uuid_tree_rescan_sem);
4044 return PTR_ERR(task);
4050 int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info)
4052 struct task_struct *task;
4054 down(&fs_info->uuid_tree_rescan_sem);
4055 task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid");
4057 /* fs_info->update_uuid_tree_gen remains 0 in all error case */
4058 btrfs_warn(fs_info, "failed to start uuid_rescan task");
4059 up(&fs_info->uuid_tree_rescan_sem);
4060 return PTR_ERR(task);
4067 * shrinking a device means finding all of the device extents past
4068 * the new size, and then following the back refs to the chunks.
4069 * The chunk relocation code actually frees the device extent
4071 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
4073 struct btrfs_trans_handle *trans;
4074 struct btrfs_root *root = device->dev_root;
4075 struct btrfs_dev_extent *dev_extent = NULL;
4076 struct btrfs_path *path;
4082 bool retried = false;
4083 bool checked_pending_chunks = false;
4084 struct extent_buffer *l;
4085 struct btrfs_key key;
4086 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4087 u64 old_total = btrfs_super_total_bytes(super_copy);
4088 u64 old_size = btrfs_device_get_total_bytes(device);
4089 u64 diff = old_size - new_size;
4091 if (device->is_tgtdev_for_dev_replace)
4094 path = btrfs_alloc_path();
4102 btrfs_device_set_total_bytes(device, new_size);
4103 if (device->writeable) {
4104 device->fs_devices->total_rw_bytes -= diff;
4105 spin_lock(&root->fs_info->free_chunk_lock);
4106 root->fs_info->free_chunk_space -= diff;
4107 spin_unlock(&root->fs_info->free_chunk_lock);
4109 unlock_chunks(root);
4112 key.objectid = device->devid;
4113 key.offset = (u64)-1;
4114 key.type = BTRFS_DEV_EXTENT_KEY;
4117 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
4118 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4120 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4124 ret = btrfs_previous_item(root, path, 0, key.type);
4126 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4131 btrfs_release_path(path);
4136 slot = path->slots[0];
4137 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
4139 if (key.objectid != device->devid) {
4140 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4141 btrfs_release_path(path);
4145 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
4146 length = btrfs_dev_extent_length(l, dev_extent);
4148 if (key.offset + length <= new_size) {
4149 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4150 btrfs_release_path(path);
4154 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
4155 btrfs_release_path(path);
4157 ret = btrfs_relocate_chunk(root, chunk_offset);
4158 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
4159 if (ret && ret != -ENOSPC)
4163 } while (key.offset-- > 0);
4165 if (failed && !retried) {
4169 } else if (failed && retried) {
4174 /* Shrinking succeeded, else we would be at "done". */
4175 trans = btrfs_start_transaction(root, 0);
4176 if (IS_ERR(trans)) {
4177 ret = PTR_ERR(trans);
4184 * We checked in the above loop all device extents that were already in
4185 * the device tree. However before we have updated the device's
4186 * total_bytes to the new size, we might have had chunk allocations that
4187 * have not complete yet (new block groups attached to transaction
4188 * handles), and therefore their device extents were not yet in the
4189 * device tree and we missed them in the loop above. So if we have any
4190 * pending chunk using a device extent that overlaps the device range
4191 * that we can not use anymore, commit the current transaction and
4192 * repeat the search on the device tree - this way we guarantee we will
4193 * not have chunks using device extents that end beyond 'new_size'.
4195 if (!checked_pending_chunks) {
4196 u64 start = new_size;
4197 u64 len = old_size - new_size;
4199 if (contains_pending_extent(trans, device, &start, len)) {
4200 unlock_chunks(root);
4201 checked_pending_chunks = true;
4204 ret = btrfs_commit_transaction(trans, root);
4211 btrfs_device_set_disk_total_bytes(device, new_size);
4212 if (list_empty(&device->resized_list))
4213 list_add_tail(&device->resized_list,
4214 &root->fs_info->fs_devices->resized_devices);
4216 WARN_ON(diff > old_total);
4217 btrfs_set_super_total_bytes(super_copy, old_total - diff);
4218 unlock_chunks(root);
4220 /* Now btrfs_update_device() will change the on-disk size. */
4221 ret = btrfs_update_device(trans, device);
4222 btrfs_end_transaction(trans, root);
4224 btrfs_free_path(path);
4227 btrfs_device_set_total_bytes(device, old_size);
4228 if (device->writeable)
4229 device->fs_devices->total_rw_bytes += diff;
4230 spin_lock(&root->fs_info->free_chunk_lock);
4231 root->fs_info->free_chunk_space += diff;
4232 spin_unlock(&root->fs_info->free_chunk_lock);
4233 unlock_chunks(root);
4238 static int btrfs_add_system_chunk(struct btrfs_root *root,
4239 struct btrfs_key *key,
4240 struct btrfs_chunk *chunk, int item_size)
4242 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4243 struct btrfs_disk_key disk_key;
4248 array_size = btrfs_super_sys_array_size(super_copy);
4249 if (array_size + item_size + sizeof(disk_key)
4250 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4251 unlock_chunks(root);
4255 ptr = super_copy->sys_chunk_array + array_size;
4256 btrfs_cpu_key_to_disk(&disk_key, key);
4257 memcpy(ptr, &disk_key, sizeof(disk_key));
4258 ptr += sizeof(disk_key);
4259 memcpy(ptr, chunk, item_size);
4260 item_size += sizeof(disk_key);
4261 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
4262 unlock_chunks(root);
4268 * sort the devices in descending order by max_avail, total_avail
4270 static int btrfs_cmp_device_info(const void *a, const void *b)
4272 const struct btrfs_device_info *di_a = a;
4273 const struct btrfs_device_info *di_b = b;
4275 if (di_a->max_avail > di_b->max_avail)
4277 if (di_a->max_avail < di_b->max_avail)
4279 if (di_a->total_avail > di_b->total_avail)
4281 if (di_a->total_avail < di_b->total_avail)
4286 static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
4287 [BTRFS_RAID_RAID10] = {
4290 .devs_max = 0, /* 0 == as many as possible */
4292 .devs_increment = 2,
4295 [BTRFS_RAID_RAID1] = {
4300 .devs_increment = 2,
4303 [BTRFS_RAID_DUP] = {
4308 .devs_increment = 1,
4311 [BTRFS_RAID_RAID0] = {
4316 .devs_increment = 1,
4319 [BTRFS_RAID_SINGLE] = {
4324 .devs_increment = 1,
4327 [BTRFS_RAID_RAID5] = {
4332 .devs_increment = 1,
4335 [BTRFS_RAID_RAID6] = {
4340 .devs_increment = 1,
4345 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
4347 /* TODO allow them to set a preferred stripe size */
4351 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
4353 if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK))
4356 btrfs_set_fs_incompat(info, RAID56);
4359 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
4360 - sizeof(struct btrfs_item) \
4361 - sizeof(struct btrfs_chunk)) \
4362 / sizeof(struct btrfs_stripe) + 1)
4364 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
4365 - 2 * sizeof(struct btrfs_disk_key) \
4366 - 2 * sizeof(struct btrfs_chunk)) \
4367 / sizeof(struct btrfs_stripe) + 1)
4369 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4370 struct btrfs_root *extent_root, u64 start,
4373 struct btrfs_fs_info *info = extent_root->fs_info;
4374 struct btrfs_fs_devices *fs_devices = info->fs_devices;
4375 struct list_head *cur;
4376 struct map_lookup *map = NULL;
4377 struct extent_map_tree *em_tree;
4378 struct extent_map *em;
4379 struct btrfs_device_info *devices_info = NULL;
4381 int num_stripes; /* total number of stripes to allocate */
4382 int data_stripes; /* number of stripes that count for
4384 int sub_stripes; /* sub_stripes info for map */
4385 int dev_stripes; /* stripes per dev */
4386 int devs_max; /* max devs to use */
4387 int devs_min; /* min devs needed */
4388 int devs_increment; /* ndevs has to be a multiple of this */
4389 int ncopies; /* how many copies to data has */
4391 u64 max_stripe_size;
4395 u64 raid_stripe_len = BTRFS_STRIPE_LEN;
4401 BUG_ON(!alloc_profile_is_valid(type, 0));
4403 if (list_empty(&fs_devices->alloc_list))
4406 index = __get_raid_index(type);
4408 sub_stripes = btrfs_raid_array[index].sub_stripes;
4409 dev_stripes = btrfs_raid_array[index].dev_stripes;
4410 devs_max = btrfs_raid_array[index].devs_max;
4411 devs_min = btrfs_raid_array[index].devs_min;
4412 devs_increment = btrfs_raid_array[index].devs_increment;
4413 ncopies = btrfs_raid_array[index].ncopies;
4415 if (type & BTRFS_BLOCK_GROUP_DATA) {
4416 max_stripe_size = 1024 * 1024 * 1024;
4417 max_chunk_size = 10 * max_stripe_size;
4419 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4420 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
4421 /* for larger filesystems, use larger metadata chunks */
4422 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
4423 max_stripe_size = 1024 * 1024 * 1024;
4425 max_stripe_size = 256 * 1024 * 1024;
4426 max_chunk_size = max_stripe_size;
4428 devs_max = BTRFS_MAX_DEVS(info->chunk_root);
4429 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
4430 max_stripe_size = 32 * 1024 * 1024;
4431 max_chunk_size = 2 * max_stripe_size;
4433 devs_max = BTRFS_MAX_DEVS_SYS_CHUNK;
4435 btrfs_err(info, "invalid chunk type 0x%llx requested",
4440 /* we don't want a chunk larger than 10% of writeable space */
4441 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
4444 devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info),
4449 cur = fs_devices->alloc_list.next;
4452 * in the first pass through the devices list, we gather information
4453 * about the available holes on each device.
4456 while (cur != &fs_devices->alloc_list) {
4457 struct btrfs_device *device;
4461 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
4465 if (!device->writeable) {
4467 "BTRFS: read-only device in alloc_list\n");
4471 if (!device->in_fs_metadata ||
4472 device->is_tgtdev_for_dev_replace)
4475 if (device->total_bytes > device->bytes_used)
4476 total_avail = device->total_bytes - device->bytes_used;
4480 /* If there is no space on this device, skip it. */
4481 if (total_avail == 0)
4484 ret = find_free_dev_extent(trans, device,
4485 max_stripe_size * dev_stripes,
4486 &dev_offset, &max_avail);
4487 if (ret && ret != -ENOSPC)
4491 max_avail = max_stripe_size * dev_stripes;
4493 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
4496 if (ndevs == fs_devices->rw_devices) {
4497 WARN(1, "%s: found more than %llu devices\n",
4498 __func__, fs_devices->rw_devices);
4501 devices_info[ndevs].dev_offset = dev_offset;
4502 devices_info[ndevs].max_avail = max_avail;
4503 devices_info[ndevs].total_avail = total_avail;
4504 devices_info[ndevs].dev = device;
4509 * now sort the devices by hole size / available space
4511 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
4512 btrfs_cmp_device_info, NULL);
4514 /* round down to number of usable stripes */
4515 ndevs -= ndevs % devs_increment;
4517 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
4522 if (devs_max && ndevs > devs_max)
4525 * the primary goal is to maximize the number of stripes, so use as many
4526 * devices as possible, even if the stripes are not maximum sized.
4528 stripe_size = devices_info[ndevs-1].max_avail;
4529 num_stripes = ndevs * dev_stripes;
4532 * this will have to be fixed for RAID1 and RAID10 over
4535 data_stripes = num_stripes / ncopies;
4537 if (type & BTRFS_BLOCK_GROUP_RAID5) {
4538 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
4539 btrfs_super_stripesize(info->super_copy));
4540 data_stripes = num_stripes - 1;
4542 if (type & BTRFS_BLOCK_GROUP_RAID6) {
4543 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
4544 btrfs_super_stripesize(info->super_copy));
4545 data_stripes = num_stripes - 2;
4549 * Use the number of data stripes to figure out how big this chunk
4550 * is really going to be in terms of logical address space,
4551 * and compare that answer with the max chunk size
4553 if (stripe_size * data_stripes > max_chunk_size) {
4554 u64 mask = (1ULL << 24) - 1;
4556 stripe_size = div_u64(max_chunk_size, data_stripes);
4558 /* bump the answer up to a 16MB boundary */
4559 stripe_size = (stripe_size + mask) & ~mask;
4561 /* but don't go higher than the limits we found
4562 * while searching for free extents
4564 if (stripe_size > devices_info[ndevs-1].max_avail)
4565 stripe_size = devices_info[ndevs-1].max_avail;
4568 stripe_size = div_u64(stripe_size, dev_stripes);
4570 /* align to BTRFS_STRIPE_LEN */
4571 stripe_size = div_u64(stripe_size, raid_stripe_len);
4572 stripe_size *= raid_stripe_len;
4574 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4579 map->num_stripes = num_stripes;
4581 for (i = 0; i < ndevs; ++i) {
4582 for (j = 0; j < dev_stripes; ++j) {
4583 int s = i * dev_stripes + j;
4584 map->stripes[s].dev = devices_info[i].dev;
4585 map->stripes[s].physical = devices_info[i].dev_offset +
4589 map->sector_size = extent_root->sectorsize;
4590 map->stripe_len = raid_stripe_len;
4591 map->io_align = raid_stripe_len;
4592 map->io_width = raid_stripe_len;
4594 map->sub_stripes = sub_stripes;
4596 num_bytes = stripe_size * data_stripes;
4598 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
4600 em = alloc_extent_map();
4606 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
4607 em->bdev = (struct block_device *)map;
4609 em->len = num_bytes;
4610 em->block_start = 0;
4611 em->block_len = em->len;
4612 em->orig_block_len = stripe_size;
4614 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4615 write_lock(&em_tree->lock);
4616 ret = add_extent_mapping(em_tree, em, 0);
4618 list_add_tail(&em->list, &trans->transaction->pending_chunks);
4619 atomic_inc(&em->refs);
4621 write_unlock(&em_tree->lock);
4623 free_extent_map(em);
4627 ret = btrfs_make_block_group(trans, extent_root, 0, type,
4628 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4631 goto error_del_extent;
4633 for (i = 0; i < map->num_stripes; i++) {
4634 num_bytes = map->stripes[i].dev->bytes_used + stripe_size;
4635 btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes);
4638 spin_lock(&extent_root->fs_info->free_chunk_lock);
4639 extent_root->fs_info->free_chunk_space -= (stripe_size *
4641 spin_unlock(&extent_root->fs_info->free_chunk_lock);
4643 free_extent_map(em);
4644 check_raid56_incompat_flag(extent_root->fs_info, type);
4646 kfree(devices_info);
4650 write_lock(&em_tree->lock);
4651 remove_extent_mapping(em_tree, em);
4652 write_unlock(&em_tree->lock);
4654 /* One for our allocation */
4655 free_extent_map(em);
4656 /* One for the tree reference */
4657 free_extent_map(em);
4658 /* One for the pending_chunks list reference */
4659 free_extent_map(em);
4661 kfree(devices_info);
4665 int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans,
4666 struct btrfs_root *extent_root,
4667 u64 chunk_offset, u64 chunk_size)
4669 struct btrfs_key key;
4670 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4671 struct btrfs_device *device;
4672 struct btrfs_chunk *chunk;
4673 struct btrfs_stripe *stripe;
4674 struct extent_map_tree *em_tree;
4675 struct extent_map *em;
4676 struct map_lookup *map;
4683 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
4684 read_lock(&em_tree->lock);
4685 em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size);
4686 read_unlock(&em_tree->lock);
4689 btrfs_crit(extent_root->fs_info, "unable to find logical "
4690 "%Lu len %Lu", chunk_offset, chunk_size);
4694 if (em->start != chunk_offset || em->len != chunk_size) {
4695 btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted"
4696 " %Lu-%Lu, found %Lu-%Lu", chunk_offset,
4697 chunk_size, em->start, em->len);
4698 free_extent_map(em);
4702 map = (struct map_lookup *)em->bdev;
4703 item_size = btrfs_chunk_item_size(map->num_stripes);
4704 stripe_size = em->orig_block_len;
4706 chunk = kzalloc(item_size, GFP_NOFS);
4712 for (i = 0; i < map->num_stripes; i++) {
4713 device = map->stripes[i].dev;
4714 dev_offset = map->stripes[i].physical;
4716 ret = btrfs_update_device(trans, device);
4719 ret = btrfs_alloc_dev_extent(trans, device,
4720 chunk_root->root_key.objectid,
4721 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4722 chunk_offset, dev_offset,
4728 stripe = &chunk->stripe;
4729 for (i = 0; i < map->num_stripes; i++) {
4730 device = map->stripes[i].dev;
4731 dev_offset = map->stripes[i].physical;
4733 btrfs_set_stack_stripe_devid(stripe, device->devid);
4734 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4735 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4739 btrfs_set_stack_chunk_length(chunk, chunk_size);
4740 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4741 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4742 btrfs_set_stack_chunk_type(chunk, map->type);
4743 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4744 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4745 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4746 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4747 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4749 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4750 key.type = BTRFS_CHUNK_ITEM_KEY;
4751 key.offset = chunk_offset;
4753 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4754 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4756 * TODO: Cleanup of inserted chunk root in case of
4759 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4765 free_extent_map(em);
4770 * Chunk allocation falls into two parts. The first part does works
4771 * that make the new allocated chunk useable, but not do any operation
4772 * that modifies the chunk tree. The second part does the works that
4773 * require modifying the chunk tree. This division is important for the
4774 * bootstrap process of adding storage to a seed btrfs.
4776 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4777 struct btrfs_root *extent_root, u64 type)
4781 ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex));
4782 chunk_offset = find_next_chunk(extent_root->fs_info);
4783 return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type);
4786 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4787 struct btrfs_root *root,
4788 struct btrfs_device *device)
4791 u64 sys_chunk_offset;
4793 struct btrfs_fs_info *fs_info = root->fs_info;
4794 struct btrfs_root *extent_root = fs_info->extent_root;
4797 chunk_offset = find_next_chunk(fs_info);
4798 alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4799 ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset,
4804 sys_chunk_offset = find_next_chunk(root->fs_info);
4805 alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4806 ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset,
4811 static inline int btrfs_chunk_max_errors(struct map_lookup *map)
4815 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4816 BTRFS_BLOCK_GROUP_RAID10 |
4817 BTRFS_BLOCK_GROUP_RAID5 |
4818 BTRFS_BLOCK_GROUP_DUP)) {
4820 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4829 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4831 struct extent_map *em;
4832 struct map_lookup *map;
4833 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4838 read_lock(&map_tree->map_tree.lock);
4839 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4840 read_unlock(&map_tree->map_tree.lock);
4844 map = (struct map_lookup *)em->bdev;
4845 for (i = 0; i < map->num_stripes; i++) {
4846 if (map->stripes[i].dev->missing) {
4851 if (!map->stripes[i].dev->writeable) {
4858 * If the number of missing devices is larger than max errors,
4859 * we can not write the data into that chunk successfully, so
4862 if (miss_ndevs > btrfs_chunk_max_errors(map))
4865 free_extent_map(em);
4869 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4871 extent_map_tree_init(&tree->map_tree);
4874 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4876 struct extent_map *em;
4879 write_lock(&tree->map_tree.lock);
4880 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4882 remove_extent_mapping(&tree->map_tree, em);
4883 write_unlock(&tree->map_tree.lock);
4887 free_extent_map(em);
4888 /* once for the tree */
4889 free_extent_map(em);
4893 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4895 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4896 struct extent_map *em;
4897 struct map_lookup *map;
4898 struct extent_map_tree *em_tree = &map_tree->map_tree;
4901 read_lock(&em_tree->lock);
4902 em = lookup_extent_mapping(em_tree, logical, len);
4903 read_unlock(&em_tree->lock);
4906 * We could return errors for these cases, but that could get ugly and
4907 * we'd probably do the same thing which is just not do anything else
4908 * and exit, so return 1 so the callers don't try to use other copies.
4911 btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical,
4916 if (em->start > logical || em->start + em->len < logical) {
4917 btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got "
4918 "%Lu-%Lu", logical, logical+len, em->start,
4919 em->start + em->len);
4920 free_extent_map(em);
4924 map = (struct map_lookup *)em->bdev;
4925 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4926 ret = map->num_stripes;
4927 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4928 ret = map->sub_stripes;
4929 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4931 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4935 free_extent_map(em);
4937 btrfs_dev_replace_lock(&fs_info->dev_replace);
4938 if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4940 btrfs_dev_replace_unlock(&fs_info->dev_replace);
4945 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4946 struct btrfs_mapping_tree *map_tree,
4949 struct extent_map *em;
4950 struct map_lookup *map;
4951 struct extent_map_tree *em_tree = &map_tree->map_tree;
4952 unsigned long len = root->sectorsize;
4954 read_lock(&em_tree->lock);
4955 em = lookup_extent_mapping(em_tree, logical, len);
4956 read_unlock(&em_tree->lock);
4959 BUG_ON(em->start > logical || em->start + em->len < logical);
4960 map = (struct map_lookup *)em->bdev;
4961 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4962 len = map->stripe_len * nr_data_stripes(map);
4963 free_extent_map(em);
4967 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4968 u64 logical, u64 len, int mirror_num)
4970 struct extent_map *em;
4971 struct map_lookup *map;
4972 struct extent_map_tree *em_tree = &map_tree->map_tree;
4975 read_lock(&em_tree->lock);
4976 em = lookup_extent_mapping(em_tree, logical, len);
4977 read_unlock(&em_tree->lock);
4980 BUG_ON(em->start > logical || em->start + em->len < logical);
4981 map = (struct map_lookup *)em->bdev;
4982 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK)
4984 free_extent_map(em);
4988 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4989 struct map_lookup *map, int first, int num,
4990 int optimal, int dev_replace_is_ongoing)
4994 struct btrfs_device *srcdev;
4996 if (dev_replace_is_ongoing &&
4997 fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4998 BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4999 srcdev = fs_info->dev_replace.srcdev;
5004 * try to avoid the drive that is the source drive for a
5005 * dev-replace procedure, only choose it if no other non-missing
5006 * mirror is available
5008 for (tolerance = 0; tolerance < 2; tolerance++) {
5009 if (map->stripes[optimal].dev->bdev &&
5010 (tolerance || map->stripes[optimal].dev != srcdev))
5012 for (i = first; i < first + num; i++) {
5013 if (map->stripes[i].dev->bdev &&
5014 (tolerance || map->stripes[i].dev != srcdev))
5019 /* we couldn't find one that doesn't fail. Just return something
5020 * and the io error handling code will clean up eventually
5025 static inline int parity_smaller(u64 a, u64 b)
5030 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
5031 static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes)
5033 struct btrfs_bio_stripe s;
5040 for (i = 0; i < num_stripes - 1; i++) {
5041 if (parity_smaller(bbio->raid_map[i],
5042 bbio->raid_map[i+1])) {
5043 s = bbio->stripes[i];
5044 l = bbio->raid_map[i];
5045 bbio->stripes[i] = bbio->stripes[i+1];
5046 bbio->raid_map[i] = bbio->raid_map[i+1];
5047 bbio->stripes[i+1] = s;
5048 bbio->raid_map[i+1] = l;
5056 static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes)
5058 struct btrfs_bio *bbio = kzalloc(
5059 /* the size of the btrfs_bio */
5060 sizeof(struct btrfs_bio) +
5061 /* plus the variable array for the stripes */
5062 sizeof(struct btrfs_bio_stripe) * (total_stripes) +
5063 /* plus the variable array for the tgt dev */
5064 sizeof(int) * (real_stripes) +
5066 * plus the raid_map, which includes both the tgt dev
5069 sizeof(u64) * (total_stripes),
5074 atomic_set(&bbio->error, 0);
5075 atomic_set(&bbio->refs, 1);
5080 void btrfs_get_bbio(struct btrfs_bio *bbio)
5082 WARN_ON(!atomic_read(&bbio->refs));
5083 atomic_inc(&bbio->refs);
5086 void btrfs_put_bbio(struct btrfs_bio *bbio)
5090 if (atomic_dec_and_test(&bbio->refs))
5094 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5095 u64 logical, u64 *length,
5096 struct btrfs_bio **bbio_ret,
5097 int mirror_num, int need_raid_map)
5099 struct extent_map *em;
5100 struct map_lookup *map;
5101 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
5102 struct extent_map_tree *em_tree = &map_tree->map_tree;
5105 u64 stripe_end_offset;
5115 int tgtdev_indexes = 0;
5116 struct btrfs_bio *bbio = NULL;
5117 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
5118 int dev_replace_is_ongoing = 0;
5119 int num_alloc_stripes;
5120 int patch_the_first_stripe_for_dev_replace = 0;
5121 u64 physical_to_patch_in_first_stripe = 0;
5122 u64 raid56_full_stripe_start = (u64)-1;
5124 read_lock(&em_tree->lock);
5125 em = lookup_extent_mapping(em_tree, logical, *length);
5126 read_unlock(&em_tree->lock);
5129 btrfs_crit(fs_info, "unable to find logical %llu len %llu",
5134 if (em->start > logical || em->start + em->len < logical) {
5135 btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, "
5136 "found %Lu-%Lu", logical, em->start,
5137 em->start + em->len);
5138 free_extent_map(em);
5142 map = (struct map_lookup *)em->bdev;
5143 offset = logical - em->start;
5145 stripe_len = map->stripe_len;
5148 * stripe_nr counts the total number of stripes we have to stride
5149 * to get to this block
5151 stripe_nr = div64_u64(stripe_nr, stripe_len);
5153 stripe_offset = stripe_nr * stripe_len;
5154 BUG_ON(offset < stripe_offset);
5156 /* stripe_offset is the offset of this block in its stripe*/
5157 stripe_offset = offset - stripe_offset;
5159 /* if we're here for raid56, we need to know the stripe aligned start */
5160 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5161 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
5162 raid56_full_stripe_start = offset;
5164 /* allow a write of a full stripe, but make sure we don't
5165 * allow straddling of stripes
5167 raid56_full_stripe_start = div64_u64(raid56_full_stripe_start,
5169 raid56_full_stripe_start *= full_stripe_len;
5172 if (rw & REQ_DISCARD) {
5173 /* we don't discard raid56 yet */
5174 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5178 *length = min_t(u64, em->len - offset, *length);
5179 } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
5181 /* For writes to RAID[56], allow a full stripeset across all disks.
5182 For other RAID types and for RAID[56] reads, just allow a single
5183 stripe (on a single disk). */
5184 if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) &&
5186 max_len = stripe_len * nr_data_stripes(map) -
5187 (offset - raid56_full_stripe_start);
5189 /* we limit the length of each bio to what fits in a stripe */
5190 max_len = stripe_len - stripe_offset;
5192 *length = min_t(u64, em->len - offset, max_len);
5194 *length = em->len - offset;
5197 /* This is for when we're called from btrfs_merge_bio_hook() and all
5198 it cares about is the length */
5202 btrfs_dev_replace_lock(dev_replace);
5203 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
5204 if (!dev_replace_is_ongoing)
5205 btrfs_dev_replace_unlock(dev_replace);
5207 if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
5208 !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
5209 dev_replace->tgtdev != NULL) {
5211 * in dev-replace case, for repair case (that's the only
5212 * case where the mirror is selected explicitly when
5213 * calling btrfs_map_block), blocks left of the left cursor
5214 * can also be read from the target drive.
5215 * For REQ_GET_READ_MIRRORS, the target drive is added as
5216 * the last one to the array of stripes. For READ, it also
5217 * needs to be supported using the same mirror number.
5218 * If the requested block is not left of the left cursor,
5219 * EIO is returned. This can happen because btrfs_num_copies()
5220 * returns one more in the dev-replace case.
5222 u64 tmp_length = *length;
5223 struct btrfs_bio *tmp_bbio = NULL;
5224 int tmp_num_stripes;
5225 u64 srcdev_devid = dev_replace->srcdev->devid;
5226 int index_srcdev = 0;
5228 u64 physical_of_found = 0;
5230 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
5231 logical, &tmp_length, &tmp_bbio, 0, 0);
5233 WARN_ON(tmp_bbio != NULL);
5237 tmp_num_stripes = tmp_bbio->num_stripes;
5238 if (mirror_num > tmp_num_stripes) {
5240 * REQ_GET_READ_MIRRORS does not contain this
5241 * mirror, that means that the requested area
5242 * is not left of the left cursor
5245 btrfs_put_bbio(tmp_bbio);
5250 * process the rest of the function using the mirror_num
5251 * of the source drive. Therefore look it up first.
5252 * At the end, patch the device pointer to the one of the
5255 for (i = 0; i < tmp_num_stripes; i++) {
5256 if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
5258 * In case of DUP, in order to keep it
5259 * simple, only add the mirror with the
5260 * lowest physical address
5263 physical_of_found <=
5264 tmp_bbio->stripes[i].physical)
5269 tmp_bbio->stripes[i].physical;
5274 mirror_num = index_srcdev + 1;
5275 patch_the_first_stripe_for_dev_replace = 1;
5276 physical_to_patch_in_first_stripe = physical_of_found;
5280 btrfs_put_bbio(tmp_bbio);
5284 btrfs_put_bbio(tmp_bbio);
5285 } else if (mirror_num > map->num_stripes) {
5291 stripe_nr_orig = stripe_nr;
5292 stripe_nr_end = ALIGN(offset + *length, map->stripe_len);
5293 stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len);
5294 stripe_end_offset = stripe_nr_end * map->stripe_len -
5297 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5298 if (rw & REQ_DISCARD)
5299 num_stripes = min_t(u64, map->num_stripes,
5300 stripe_nr_end - stripe_nr_orig);
5301 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5303 if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)))
5305 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
5306 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
5307 num_stripes = map->num_stripes;
5308 else if (mirror_num)
5309 stripe_index = mirror_num - 1;
5311 stripe_index = find_live_mirror(fs_info, map, 0,
5313 current->pid % map->num_stripes,
5314 dev_replace_is_ongoing);
5315 mirror_num = stripe_index + 1;
5318 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
5319 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
5320 num_stripes = map->num_stripes;
5321 } else if (mirror_num) {
5322 stripe_index = mirror_num - 1;
5327 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5328 u32 factor = map->num_stripes / map->sub_stripes;
5330 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index);
5331 stripe_index *= map->sub_stripes;
5333 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5334 num_stripes = map->sub_stripes;
5335 else if (rw & REQ_DISCARD)
5336 num_stripes = min_t(u64, map->sub_stripes *
5337 (stripe_nr_end - stripe_nr_orig),
5339 else if (mirror_num)
5340 stripe_index += mirror_num - 1;
5342 int old_stripe_index = stripe_index;
5343 stripe_index = find_live_mirror(fs_info, map,
5345 map->sub_stripes, stripe_index +
5346 current->pid % map->sub_stripes,
5347 dev_replace_is_ongoing);
5348 mirror_num = stripe_index - old_stripe_index + 1;
5351 } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5352 if (need_raid_map &&
5353 ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5355 /* push stripe_nr back to the start of the full stripe */
5356 stripe_nr = div_u64(raid56_full_stripe_start,
5357 stripe_len * nr_data_stripes(map));
5359 /* RAID[56] write or recovery. Return all stripes */
5360 num_stripes = map->num_stripes;
5361 max_errors = nr_parity_stripes(map);
5363 *length = map->stripe_len;
5368 * Mirror #0 or #1 means the original data block.
5369 * Mirror #2 is RAID5 parity block.
5370 * Mirror #3 is RAID6 Q block.
5372 stripe_nr = div_u64_rem(stripe_nr,
5373 nr_data_stripes(map), &stripe_index);
5375 stripe_index = nr_data_stripes(map) +
5378 /* We distribute the parity blocks across stripes */
5379 div_u64_rem(stripe_nr + stripe_index, map->num_stripes,
5381 if (!(rw & (REQ_WRITE | REQ_DISCARD |
5382 REQ_GET_READ_MIRRORS)) && mirror_num <= 1)
5387 * after this, stripe_nr is the number of stripes on this
5388 * device we have to walk to find the data, and stripe_index is
5389 * the number of our device in the stripe array
5391 stripe_nr = div_u64_rem(stripe_nr, map->num_stripes,
5393 mirror_num = stripe_index + 1;
5395 BUG_ON(stripe_index >= map->num_stripes);
5397 num_alloc_stripes = num_stripes;
5398 if (dev_replace_is_ongoing) {
5399 if (rw & (REQ_WRITE | REQ_DISCARD))
5400 num_alloc_stripes <<= 1;
5401 if (rw & REQ_GET_READ_MIRRORS)
5402 num_alloc_stripes++;
5403 tgtdev_indexes = num_stripes;
5406 bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes);
5411 if (dev_replace_is_ongoing)
5412 bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes);
5414 /* build raid_map */
5415 if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK &&
5416 need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) ||
5421 bbio->raid_map = (u64 *)((void *)bbio->stripes +
5422 sizeof(struct btrfs_bio_stripe) *
5424 sizeof(int) * tgtdev_indexes);
5426 /* Work out the disk rotation on this stripe-set */
5427 div_u64_rem(stripe_nr, num_stripes, &rot);
5429 /* Fill in the logical address of each stripe */
5430 tmp = stripe_nr * nr_data_stripes(map);
5431 for (i = 0; i < nr_data_stripes(map); i++)
5432 bbio->raid_map[(i+rot) % num_stripes] =
5433 em->start + (tmp + i) * map->stripe_len;
5435 bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
5436 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
5437 bbio->raid_map[(i+rot+1) % num_stripes] =
5441 if (rw & REQ_DISCARD) {
5443 u32 sub_stripes = 0;
5444 u64 stripes_per_dev = 0;
5445 u32 remaining_stripes = 0;
5446 u32 last_stripe = 0;
5449 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
5450 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5453 sub_stripes = map->sub_stripes;
5455 factor = map->num_stripes / sub_stripes;
5456 stripes_per_dev = div_u64_rem(stripe_nr_end -
5459 &remaining_stripes);
5460 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
5461 last_stripe *= sub_stripes;
5464 for (i = 0; i < num_stripes; i++) {
5465 bbio->stripes[i].physical =
5466 map->stripes[stripe_index].physical +
5467 stripe_offset + stripe_nr * map->stripe_len;
5468 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
5470 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
5471 BTRFS_BLOCK_GROUP_RAID10)) {
5472 bbio->stripes[i].length = stripes_per_dev *
5475 if (i / sub_stripes < remaining_stripes)
5476 bbio->stripes[i].length +=
5480 * Special for the first stripe and
5483 * |-------|...|-------|
5487 if (i < sub_stripes)
5488 bbio->stripes[i].length -=
5491 if (stripe_index >= last_stripe &&
5492 stripe_index <= (last_stripe +
5494 bbio->stripes[i].length -=
5497 if (i == sub_stripes - 1)
5500 bbio->stripes[i].length = *length;
5503 if (stripe_index == map->num_stripes) {
5504 /* This could only happen for RAID0/10 */
5510 for (i = 0; i < num_stripes; i++) {
5511 bbio->stripes[i].physical =
5512 map->stripes[stripe_index].physical +
5514 stripe_nr * map->stripe_len;
5515 bbio->stripes[i].dev =
5516 map->stripes[stripe_index].dev;
5521 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
5522 max_errors = btrfs_chunk_max_errors(map);
5525 sort_parity_stripes(bbio, num_stripes);
5528 if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
5529 dev_replace->tgtdev != NULL) {
5530 int index_where_to_add;
5531 u64 srcdev_devid = dev_replace->srcdev->devid;
5534 * duplicate the write operations while the dev replace
5535 * procedure is running. Since the copying of the old disk
5536 * to the new disk takes place at run time while the
5537 * filesystem is mounted writable, the regular write
5538 * operations to the old disk have to be duplicated to go
5539 * to the new disk as well.
5540 * Note that device->missing is handled by the caller, and
5541 * that the write to the old disk is already set up in the
5544 index_where_to_add = num_stripes;
5545 for (i = 0; i < num_stripes; i++) {
5546 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5547 /* write to new disk, too */
5548 struct btrfs_bio_stripe *new =
5549 bbio->stripes + index_where_to_add;
5550 struct btrfs_bio_stripe *old =
5553 new->physical = old->physical;
5554 new->length = old->length;
5555 new->dev = dev_replace->tgtdev;
5556 bbio->tgtdev_map[i] = index_where_to_add;
5557 index_where_to_add++;
5562 num_stripes = index_where_to_add;
5563 } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
5564 dev_replace->tgtdev != NULL) {
5565 u64 srcdev_devid = dev_replace->srcdev->devid;
5566 int index_srcdev = 0;
5568 u64 physical_of_found = 0;
5571 * During the dev-replace procedure, the target drive can
5572 * also be used to read data in case it is needed to repair
5573 * a corrupt block elsewhere. This is possible if the
5574 * requested area is left of the left cursor. In this area,
5575 * the target drive is a full copy of the source drive.
5577 for (i = 0; i < num_stripes; i++) {
5578 if (bbio->stripes[i].dev->devid == srcdev_devid) {
5580 * In case of DUP, in order to keep it
5581 * simple, only add the mirror with the
5582 * lowest physical address
5585 physical_of_found <=
5586 bbio->stripes[i].physical)
5590 physical_of_found = bbio->stripes[i].physical;
5594 if (physical_of_found + map->stripe_len <=
5595 dev_replace->cursor_left) {
5596 struct btrfs_bio_stripe *tgtdev_stripe =
5597 bbio->stripes + num_stripes;
5599 tgtdev_stripe->physical = physical_of_found;
5600 tgtdev_stripe->length =
5601 bbio->stripes[index_srcdev].length;
5602 tgtdev_stripe->dev = dev_replace->tgtdev;
5603 bbio->tgtdev_map[index_srcdev] = num_stripes;
5612 bbio->map_type = map->type;
5613 bbio->num_stripes = num_stripes;
5614 bbio->max_errors = max_errors;
5615 bbio->mirror_num = mirror_num;
5616 bbio->num_tgtdevs = tgtdev_indexes;
5619 * this is the case that REQ_READ && dev_replace_is_ongoing &&
5620 * mirror_num == num_stripes + 1 && dev_replace target drive is
5621 * available as a mirror
5623 if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
5624 WARN_ON(num_stripes > 1);
5625 bbio->stripes[0].dev = dev_replace->tgtdev;
5626 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
5627 bbio->mirror_num = map->num_stripes + 1;
5630 if (dev_replace_is_ongoing)
5631 btrfs_dev_replace_unlock(dev_replace);
5632 free_extent_map(em);
5636 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
5637 u64 logical, u64 *length,
5638 struct btrfs_bio **bbio_ret, int mirror_num)
5640 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5644 /* For Scrub/replace */
5645 int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw,
5646 u64 logical, u64 *length,
5647 struct btrfs_bio **bbio_ret, int mirror_num,
5650 return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
5651 mirror_num, need_raid_map);
5654 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
5655 u64 chunk_start, u64 physical, u64 devid,
5656 u64 **logical, int *naddrs, int *stripe_len)
5658 struct extent_map_tree *em_tree = &map_tree->map_tree;
5659 struct extent_map *em;
5660 struct map_lookup *map;
5668 read_lock(&em_tree->lock);
5669 em = lookup_extent_mapping(em_tree, chunk_start, 1);
5670 read_unlock(&em_tree->lock);
5673 printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n",
5678 if (em->start != chunk_start) {
5679 printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n",
5680 em->start, chunk_start);
5681 free_extent_map(em);
5684 map = (struct map_lookup *)em->bdev;
5687 rmap_len = map->stripe_len;
5689 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
5690 length = div_u64(length, map->num_stripes / map->sub_stripes);
5691 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
5692 length = div_u64(length, map->num_stripes);
5693 else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
5694 length = div_u64(length, nr_data_stripes(map));
5695 rmap_len = map->stripe_len * nr_data_stripes(map);
5698 buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS);
5699 BUG_ON(!buf); /* -ENOMEM */
5701 for (i = 0; i < map->num_stripes; i++) {
5702 if (devid && map->stripes[i].dev->devid != devid)
5704 if (map->stripes[i].physical > physical ||
5705 map->stripes[i].physical + length <= physical)
5708 stripe_nr = physical - map->stripes[i].physical;
5709 stripe_nr = div_u64(stripe_nr, map->stripe_len);
5711 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
5712 stripe_nr = stripe_nr * map->num_stripes + i;
5713 stripe_nr = div_u64(stripe_nr, map->sub_stripes);
5714 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
5715 stripe_nr = stripe_nr * map->num_stripes + i;
5716 } /* else if RAID[56], multiply by nr_data_stripes().
5717 * Alternatively, just use rmap_len below instead of
5718 * map->stripe_len */
5720 bytenr = chunk_start + stripe_nr * rmap_len;
5721 WARN_ON(nr >= map->num_stripes);
5722 for (j = 0; j < nr; j++) {
5723 if (buf[j] == bytenr)
5727 WARN_ON(nr >= map->num_stripes);
5734 *stripe_len = rmap_len;
5736 free_extent_map(em);
5740 static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err)
5742 bio->bi_private = bbio->private;
5743 bio->bi_end_io = bbio->end_io;
5744 bio_endio(bio, err);
5746 btrfs_put_bbio(bbio);
5749 static void btrfs_end_bio(struct bio *bio, int err)
5751 struct btrfs_bio *bbio = bio->bi_private;
5752 int is_orig_bio = 0;
5755 atomic_inc(&bbio->error);
5756 if (err == -EIO || err == -EREMOTEIO) {
5757 unsigned int stripe_index =
5758 btrfs_io_bio(bio)->stripe_index;
5759 struct btrfs_device *dev;
5761 BUG_ON(stripe_index >= bbio->num_stripes);
5762 dev = bbio->stripes[stripe_index].dev;
5764 if (bio->bi_rw & WRITE)
5765 btrfs_dev_stat_inc(dev,
5766 BTRFS_DEV_STAT_WRITE_ERRS);
5768 btrfs_dev_stat_inc(dev,
5769 BTRFS_DEV_STAT_READ_ERRS);
5770 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5771 btrfs_dev_stat_inc(dev,
5772 BTRFS_DEV_STAT_FLUSH_ERRS);
5773 btrfs_dev_stat_print_on_error(dev);
5778 if (bio == bbio->orig_bio)
5781 btrfs_bio_counter_dec(bbio->fs_info);
5783 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5786 bio = bbio->orig_bio;
5789 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5790 /* only send an error to the higher layers if it is
5791 * beyond the tolerance of the btrfs bio
5793 if (atomic_read(&bbio->error) > bbio->max_errors) {
5797 * this bio is actually up to date, we didn't
5798 * go over the max number of errors
5800 set_bit(BIO_UPTODATE, &bio->bi_flags);
5804 btrfs_end_bbio(bbio, bio, err);
5805 } else if (!is_orig_bio) {
5811 * see run_scheduled_bios for a description of why bios are collected for
5814 * This will add one bio to the pending list for a device and make sure
5815 * the work struct is scheduled.
5817 static noinline void btrfs_schedule_bio(struct btrfs_root *root,
5818 struct btrfs_device *device,
5819 int rw, struct bio *bio)
5821 int should_queue = 1;
5822 struct btrfs_pending_bios *pending_bios;
5824 if (device->missing || !device->bdev) {
5825 bio_endio(bio, -EIO);
5829 /* don't bother with additional async steps for reads, right now */
5830 if (!(rw & REQ_WRITE)) {
5832 btrfsic_submit_bio(rw, bio);
5838 * nr_async_bios allows us to reliably return congestion to the
5839 * higher layers. Otherwise, the async bio makes it appear we have
5840 * made progress against dirty pages when we've really just put it
5841 * on a queue for later
5843 atomic_inc(&root->fs_info->nr_async_bios);
5844 WARN_ON(bio->bi_next);
5845 bio->bi_next = NULL;
5848 spin_lock(&device->io_lock);
5849 if (bio->bi_rw & REQ_SYNC)
5850 pending_bios = &device->pending_sync_bios;
5852 pending_bios = &device->pending_bios;
5854 if (pending_bios->tail)
5855 pending_bios->tail->bi_next = bio;
5857 pending_bios->tail = bio;
5858 if (!pending_bios->head)
5859 pending_bios->head = bio;
5860 if (device->running_pending)
5863 spin_unlock(&device->io_lock);
5866 btrfs_queue_work(root->fs_info->submit_workers,
5870 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5873 struct bio_vec *prev;
5874 struct request_queue *q = bdev_get_queue(bdev);
5875 unsigned int max_sectors = queue_max_sectors(q);
5876 struct bvec_merge_data bvm = {
5878 .bi_sector = sector,
5879 .bi_rw = bio->bi_rw,
5882 if (WARN_ON(bio->bi_vcnt == 0))
5885 prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5886 if (bio_sectors(bio) > max_sectors)
5889 if (!q->merge_bvec_fn)
5892 bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len;
5893 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5898 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5899 struct bio *bio, u64 physical, int dev_nr,
5902 struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5904 bio->bi_private = bbio;
5905 btrfs_io_bio(bio)->stripe_index = dev_nr;
5906 bio->bi_end_io = btrfs_end_bio;
5907 bio->bi_iter.bi_sector = physical >> 9;
5910 struct rcu_string *name;
5913 name = rcu_dereference(dev->name);
5914 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5915 "(%s id %llu), size=%u\n", rw,
5916 (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev,
5917 name->str, dev->devid, bio->bi_iter.bi_size);
5921 bio->bi_bdev = dev->bdev;
5923 btrfs_bio_counter_inc_noblocked(root->fs_info);
5926 btrfs_schedule_bio(root, dev, rw, bio);
5928 btrfsic_submit_bio(rw, bio);
5931 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5932 struct bio *first_bio, struct btrfs_device *dev,
5933 int dev_nr, int rw, int async)
5935 struct bio_vec *bvec = first_bio->bi_io_vec;
5937 int nr_vecs = bio_get_nr_vecs(dev->bdev);
5938 u64 physical = bbio->stripes[dev_nr].physical;
5941 bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5945 if (first_bio->bi_ioc) {
5946 get_io_context_active(first_bio->bi_ioc);
5947 bio->bi_ioc = first_bio->bi_ioc;
5949 if (first_bio->bi_css) {
5950 css_get(first_bio->bi_css);
5951 bio->bi_css = first_bio->bi_css;
5953 while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5954 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5955 bvec->bv_offset) < bvec->bv_len) {
5956 u64 len = bio->bi_iter.bi_size;
5958 atomic_inc(&bbio->stripes_pending);
5959 submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5967 submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5971 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5973 atomic_inc(&bbio->error);
5974 if (atomic_dec_and_test(&bbio->stripes_pending)) {
5975 /* Shoud be the original bio. */
5976 WARN_ON(bio != bbio->orig_bio);
5978 btrfs_io_bio(bio)->mirror_num = bbio->mirror_num;
5979 bio->bi_iter.bi_sector = logical >> 9;
5981 btrfs_end_bbio(bbio, bio, -EIO);
5985 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5986 int mirror_num, int async_submit)
5988 struct btrfs_device *dev;
5989 struct bio *first_bio = bio;
5990 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
5996 struct btrfs_bio *bbio = NULL;
5998 length = bio->bi_iter.bi_size;
5999 map_length = length;
6001 btrfs_bio_counter_inc_blocked(root->fs_info);
6002 ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
6005 btrfs_bio_counter_dec(root->fs_info);
6009 total_devs = bbio->num_stripes;
6010 bbio->orig_bio = first_bio;
6011 bbio->private = first_bio->bi_private;
6012 bbio->end_io = first_bio->bi_end_io;
6013 bbio->fs_info = root->fs_info;
6014 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
6016 if (bbio->raid_map) {
6017 /* In this case, map_length has been set to the length of
6018 a single stripe; not the whole write */
6020 ret = raid56_parity_write(root, bio, bbio, map_length);
6022 ret = raid56_parity_recover(root, bio, bbio, map_length,
6026 btrfs_bio_counter_dec(root->fs_info);
6030 if (map_length < length) {
6031 btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu",
6032 logical, length, map_length);
6036 for (dev_nr = 0; dev_nr < total_devs; dev_nr++) {
6037 dev = bbio->stripes[dev_nr].dev;
6038 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
6039 bbio_error(bbio, first_bio, logical);
6044 * Check and see if we're ok with this bio based on it's size
6045 * and offset with the given device.
6047 if (!bio_size_ok(dev->bdev, first_bio,
6048 bbio->stripes[dev_nr].physical >> 9)) {
6049 ret = breakup_stripe_bio(root, bbio, first_bio, dev,
6050 dev_nr, rw, async_submit);
6055 if (dev_nr < total_devs - 1) {
6056 bio = btrfs_bio_clone(first_bio, GFP_NOFS);
6057 BUG_ON(!bio); /* -ENOMEM */
6061 submit_stripe_bio(root, bbio, bio,
6062 bbio->stripes[dev_nr].physical, dev_nr, rw,
6065 btrfs_bio_counter_dec(root->fs_info);
6069 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
6072 struct btrfs_device *device;
6073 struct btrfs_fs_devices *cur_devices;
6075 cur_devices = fs_info->fs_devices;
6076 while (cur_devices) {
6078 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
6079 device = __find_device(&cur_devices->devices,
6084 cur_devices = cur_devices->seed;
6089 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
6090 struct btrfs_fs_devices *fs_devices,
6091 u64 devid, u8 *dev_uuid)
6093 struct btrfs_device *device;
6095 device = btrfs_alloc_device(NULL, &devid, dev_uuid);
6099 list_add(&device->dev_list, &fs_devices->devices);
6100 device->fs_devices = fs_devices;
6101 fs_devices->num_devices++;
6103 device->missing = 1;
6104 fs_devices->missing_devices++;
6110 * btrfs_alloc_device - allocate struct btrfs_device
6111 * @fs_info: used only for generating a new devid, can be NULL if
6112 * devid is provided (i.e. @devid != NULL).
6113 * @devid: a pointer to devid for this device. If NULL a new devid
6115 * @uuid: a pointer to UUID for this device. If NULL a new UUID
6118 * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR()
6119 * on error. Returned struct is not linked onto any lists and can be
6120 * destroyed with kfree() right away.
6122 struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info,
6126 struct btrfs_device *dev;
6129 if (WARN_ON(!devid && !fs_info))
6130 return ERR_PTR(-EINVAL);
6132 dev = __alloc_device();
6141 ret = find_next_devid(fs_info, &tmp);
6144 return ERR_PTR(ret);
6150 memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE);
6152 generate_random_uuid(dev->uuid);
6154 btrfs_init_work(&dev->work, btrfs_submit_helper,
6155 pending_bios_fn, NULL, NULL);
6160 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
6161 struct extent_buffer *leaf,
6162 struct btrfs_chunk *chunk)
6164 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6165 struct map_lookup *map;
6166 struct extent_map *em;
6170 u8 uuid[BTRFS_UUID_SIZE];
6175 logical = key->offset;
6176 length = btrfs_chunk_length(leaf, chunk);
6178 read_lock(&map_tree->map_tree.lock);
6179 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
6180 read_unlock(&map_tree->map_tree.lock);
6182 /* already mapped? */
6183 if (em && em->start <= logical && em->start + em->len > logical) {
6184 free_extent_map(em);
6187 free_extent_map(em);
6190 em = alloc_extent_map();
6193 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
6194 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
6196 free_extent_map(em);
6200 set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags);
6201 em->bdev = (struct block_device *)map;
6202 em->start = logical;
6205 em->block_start = 0;
6206 em->block_len = em->len;
6208 map->num_stripes = num_stripes;
6209 map->io_width = btrfs_chunk_io_width(leaf, chunk);
6210 map->io_align = btrfs_chunk_io_align(leaf, chunk);
6211 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
6212 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
6213 map->type = btrfs_chunk_type(leaf, chunk);
6214 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
6215 for (i = 0; i < num_stripes; i++) {
6216 map->stripes[i].physical =
6217 btrfs_stripe_offset_nr(leaf, chunk, i);
6218 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
6219 read_extent_buffer(leaf, uuid, (unsigned long)
6220 btrfs_stripe_dev_uuid_nr(chunk, i),
6222 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
6224 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
6225 free_extent_map(em);
6228 if (!map->stripes[i].dev) {
6229 map->stripes[i].dev =
6230 add_missing_dev(root, root->fs_info->fs_devices,
6232 if (!map->stripes[i].dev) {
6233 free_extent_map(em);
6236 btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing",
6239 map->stripes[i].dev->in_fs_metadata = 1;
6242 write_lock(&map_tree->map_tree.lock);
6243 ret = add_extent_mapping(&map_tree->map_tree, em, 0);
6244 write_unlock(&map_tree->map_tree.lock);
6245 BUG_ON(ret); /* Tree corruption */
6246 free_extent_map(em);
6251 static void fill_device_from_item(struct extent_buffer *leaf,
6252 struct btrfs_dev_item *dev_item,
6253 struct btrfs_device *device)
6257 device->devid = btrfs_device_id(leaf, dev_item);
6258 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
6259 device->total_bytes = device->disk_total_bytes;
6260 device->commit_total_bytes = device->disk_total_bytes;
6261 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
6262 device->commit_bytes_used = device->bytes_used;
6263 device->type = btrfs_device_type(leaf, dev_item);
6264 device->io_align = btrfs_device_io_align(leaf, dev_item);
6265 device->io_width = btrfs_device_io_width(leaf, dev_item);
6266 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
6267 WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
6268 device->is_tgtdev_for_dev_replace = 0;
6270 ptr = btrfs_device_uuid(dev_item);
6271 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
6274 static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root,
6277 struct btrfs_fs_devices *fs_devices;
6280 BUG_ON(!mutex_is_locked(&uuid_mutex));
6282 fs_devices = root->fs_info->fs_devices->seed;
6283 while (fs_devices) {
6284 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE))
6287 fs_devices = fs_devices->seed;
6290 fs_devices = find_fsid(fsid);
6292 if (!btrfs_test_opt(root, DEGRADED))
6293 return ERR_PTR(-ENOENT);
6295 fs_devices = alloc_fs_devices(fsid);
6296 if (IS_ERR(fs_devices))
6299 fs_devices->seeding = 1;
6300 fs_devices->opened = 1;
6304 fs_devices = clone_fs_devices(fs_devices);
6305 if (IS_ERR(fs_devices))
6308 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
6309 root->fs_info->bdev_holder);
6311 free_fs_devices(fs_devices);
6312 fs_devices = ERR_PTR(ret);
6316 if (!fs_devices->seeding) {
6317 __btrfs_close_devices(fs_devices);
6318 free_fs_devices(fs_devices);
6319 fs_devices = ERR_PTR(-EINVAL);
6323 fs_devices->seed = root->fs_info->fs_devices->seed;
6324 root->fs_info->fs_devices->seed = fs_devices;
6329 static int read_one_dev(struct btrfs_root *root,
6330 struct extent_buffer *leaf,
6331 struct btrfs_dev_item *dev_item)
6333 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6334 struct btrfs_device *device;
6337 u8 fs_uuid[BTRFS_UUID_SIZE];
6338 u8 dev_uuid[BTRFS_UUID_SIZE];
6340 devid = btrfs_device_id(leaf, dev_item);
6341 read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item),
6343 read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item),
6346 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
6347 fs_devices = open_seed_devices(root, fs_uuid);
6348 if (IS_ERR(fs_devices))
6349 return PTR_ERR(fs_devices);
6352 device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
6354 if (!btrfs_test_opt(root, DEGRADED))
6357 device = add_missing_dev(root, fs_devices, devid, dev_uuid);
6360 btrfs_warn(root->fs_info, "devid %llu uuid %pU missing",
6363 if (!device->bdev && !btrfs_test_opt(root, DEGRADED))
6366 if(!device->bdev && !device->missing) {
6368 * this happens when a device that was properly setup
6369 * in the device info lists suddenly goes bad.
6370 * device->bdev is NULL, and so we have to set
6371 * device->missing to one here
6373 device->fs_devices->missing_devices++;
6374 device->missing = 1;
6377 /* Move the device to its own fs_devices */
6378 if (device->fs_devices != fs_devices) {
6379 ASSERT(device->missing);
6381 list_move(&device->dev_list, &fs_devices->devices);
6382 device->fs_devices->num_devices--;
6383 fs_devices->num_devices++;
6385 device->fs_devices->missing_devices--;
6386 fs_devices->missing_devices++;
6388 device->fs_devices = fs_devices;
6392 if (device->fs_devices != root->fs_info->fs_devices) {
6393 BUG_ON(device->writeable);
6394 if (device->generation !=
6395 btrfs_device_generation(leaf, dev_item))
6399 fill_device_from_item(leaf, dev_item, device);
6400 device->in_fs_metadata = 1;
6401 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
6402 device->fs_devices->total_rw_bytes += device->total_bytes;
6403 spin_lock(&root->fs_info->free_chunk_lock);
6404 root->fs_info->free_chunk_space += device->total_bytes -
6406 spin_unlock(&root->fs_info->free_chunk_lock);
6412 int btrfs_read_sys_array(struct btrfs_root *root)
6414 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
6415 struct extent_buffer *sb;
6416 struct btrfs_disk_key *disk_key;
6417 struct btrfs_chunk *chunk;
6419 unsigned long sb_array_offset;
6425 struct btrfs_key key;
6427 ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize);
6429 * This will create extent buffer of nodesize, superblock size is
6430 * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will
6431 * overallocate but we can keep it as-is, only the first page is used.
6433 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET);
6436 btrfs_set_buffer_uptodate(sb);
6437 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
6439 * The sb extent buffer is artifical and just used to read the system array.
6440 * btrfs_set_buffer_uptodate() call does not properly mark all it's
6441 * pages up-to-date when the page is larger: extent does not cover the
6442 * whole page and consequently check_page_uptodate does not find all
6443 * the page's extents up-to-date (the hole beyond sb),
6444 * write_extent_buffer then triggers a WARN_ON.
6446 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
6447 * but sb spans only this function. Add an explicit SetPageUptodate call
6448 * to silence the warning eg. on PowerPC 64.
6450 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
6451 SetPageUptodate(sb->pages[0]);
6453 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
6454 array_size = btrfs_super_sys_array_size(super_copy);
6456 array_ptr = super_copy->sys_chunk_array;
6457 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
6460 while (cur_offset < array_size) {
6461 disk_key = (struct btrfs_disk_key *)array_ptr;
6462 len = sizeof(*disk_key);
6463 if (cur_offset + len > array_size)
6464 goto out_short_read;
6466 btrfs_disk_key_to_cpu(&key, disk_key);
6469 sb_array_offset += len;
6472 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
6473 chunk = (struct btrfs_chunk *)sb_array_offset;
6475 * At least one btrfs_chunk with one stripe must be
6476 * present, exact stripe count check comes afterwards
6478 len = btrfs_chunk_item_size(1);
6479 if (cur_offset + len > array_size)
6480 goto out_short_read;
6482 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
6483 len = btrfs_chunk_item_size(num_stripes);
6484 if (cur_offset + len > array_size)
6485 goto out_short_read;
6487 ret = read_one_chunk(root, &key, sb, chunk);
6495 sb_array_offset += len;
6498 free_extent_buffer(sb);
6502 printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n",
6504 free_extent_buffer(sb);
6508 int btrfs_read_chunk_tree(struct btrfs_root *root)
6510 struct btrfs_path *path;
6511 struct extent_buffer *leaf;
6512 struct btrfs_key key;
6513 struct btrfs_key found_key;
6517 root = root->fs_info->chunk_root;
6519 path = btrfs_alloc_path();
6523 mutex_lock(&uuid_mutex);
6527 * Read all device items, and then all the chunk items. All
6528 * device items are found before any chunk item (their object id
6529 * is smaller than the lowest possible object id for a chunk
6530 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
6532 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
6535 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6539 leaf = path->nodes[0];
6540 slot = path->slots[0];
6541 if (slot >= btrfs_header_nritems(leaf)) {
6542 ret = btrfs_next_leaf(root, path);
6549 btrfs_item_key_to_cpu(leaf, &found_key, slot);
6550 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
6551 struct btrfs_dev_item *dev_item;
6552 dev_item = btrfs_item_ptr(leaf, slot,
6553 struct btrfs_dev_item);
6554 ret = read_one_dev(root, leaf, dev_item);
6557 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
6558 struct btrfs_chunk *chunk;
6559 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
6560 ret = read_one_chunk(root, &found_key, leaf, chunk);
6568 unlock_chunks(root);
6569 mutex_unlock(&uuid_mutex);
6571 btrfs_free_path(path);
6575 void btrfs_init_devices_late(struct btrfs_fs_info *fs_info)
6577 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6578 struct btrfs_device *device;
6580 while (fs_devices) {
6581 mutex_lock(&fs_devices->device_list_mutex);
6582 list_for_each_entry(device, &fs_devices->devices, dev_list)
6583 device->dev_root = fs_info->dev_root;
6584 mutex_unlock(&fs_devices->device_list_mutex);
6586 fs_devices = fs_devices->seed;
6590 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
6594 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6595 btrfs_dev_stat_reset(dev, i);
6598 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
6600 struct btrfs_key key;
6601 struct btrfs_key found_key;
6602 struct btrfs_root *dev_root = fs_info->dev_root;
6603 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6604 struct extent_buffer *eb;
6607 struct btrfs_device *device;
6608 struct btrfs_path *path = NULL;
6611 path = btrfs_alloc_path();
6617 mutex_lock(&fs_devices->device_list_mutex);
6618 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6620 struct btrfs_dev_stats_item *ptr;
6623 key.type = BTRFS_DEV_STATS_KEY;
6624 key.offset = device->devid;
6625 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
6627 __btrfs_reset_dev_stats(device);
6628 device->dev_stats_valid = 1;
6629 btrfs_release_path(path);
6632 slot = path->slots[0];
6633 eb = path->nodes[0];
6634 btrfs_item_key_to_cpu(eb, &found_key, slot);
6635 item_size = btrfs_item_size_nr(eb, slot);
6637 ptr = btrfs_item_ptr(eb, slot,
6638 struct btrfs_dev_stats_item);
6640 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6641 if (item_size >= (1 + i) * sizeof(__le64))
6642 btrfs_dev_stat_set(device, i,
6643 btrfs_dev_stats_value(eb, ptr, i));
6645 btrfs_dev_stat_reset(device, i);
6648 device->dev_stats_valid = 1;
6649 btrfs_dev_stat_print_on_load(device);
6650 btrfs_release_path(path);
6652 mutex_unlock(&fs_devices->device_list_mutex);
6655 btrfs_free_path(path);
6656 return ret < 0 ? ret : 0;
6659 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
6660 struct btrfs_root *dev_root,
6661 struct btrfs_device *device)
6663 struct btrfs_path *path;
6664 struct btrfs_key key;
6665 struct extent_buffer *eb;
6666 struct btrfs_dev_stats_item *ptr;
6671 key.type = BTRFS_DEV_STATS_KEY;
6672 key.offset = device->devid;
6674 path = btrfs_alloc_path();
6676 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
6678 printk_in_rcu(KERN_WARNING "BTRFS: "
6679 "error %d while searching for dev_stats item for device %s!\n",
6680 ret, rcu_str_deref(device->name));
6685 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
6686 /* need to delete old one and insert a new one */
6687 ret = btrfs_del_item(trans, dev_root, path);
6689 printk_in_rcu(KERN_WARNING "BTRFS: "
6690 "delete too small dev_stats item for device %s failed %d!\n",
6691 rcu_str_deref(device->name), ret);
6698 /* need to insert a new item */
6699 btrfs_release_path(path);
6700 ret = btrfs_insert_empty_item(trans, dev_root, path,
6701 &key, sizeof(*ptr));
6703 printk_in_rcu(KERN_WARNING "BTRFS: "
6704 "insert dev_stats item for device %s failed %d!\n",
6705 rcu_str_deref(device->name), ret);
6710 eb = path->nodes[0];
6711 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
6712 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6713 btrfs_set_dev_stats_value(eb, ptr, i,
6714 btrfs_dev_stat_read(device, i));
6715 btrfs_mark_buffer_dirty(eb);
6718 btrfs_free_path(path);
6723 * called from commit_transaction. Writes all changed device stats to disk.
6725 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
6726 struct btrfs_fs_info *fs_info)
6728 struct btrfs_root *dev_root = fs_info->dev_root;
6729 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6730 struct btrfs_device *device;
6734 mutex_lock(&fs_devices->device_list_mutex);
6735 list_for_each_entry(device, &fs_devices->devices, dev_list) {
6736 if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device))
6739 stats_cnt = atomic_read(&device->dev_stats_ccnt);
6740 ret = update_dev_stat_item(trans, dev_root, device);
6742 atomic_sub(stats_cnt, &device->dev_stats_ccnt);
6744 mutex_unlock(&fs_devices->device_list_mutex);
6749 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
6751 btrfs_dev_stat_inc(dev, index);
6752 btrfs_dev_stat_print_on_error(dev);
6755 static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
6757 if (!dev->dev_stats_valid)
6759 printk_ratelimited_in_rcu(KERN_ERR "BTRFS: "
6760 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6761 rcu_str_deref(dev->name),
6762 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6763 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6764 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6765 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6766 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6769 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
6773 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6774 if (btrfs_dev_stat_read(dev, i) != 0)
6776 if (i == BTRFS_DEV_STAT_VALUES_MAX)
6777 return; /* all values == 0, suppress message */
6779 printk_in_rcu(KERN_INFO "BTRFS: "
6780 "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
6781 rcu_str_deref(dev->name),
6782 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
6783 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
6784 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
6785 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
6786 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
6789 int btrfs_get_dev_stats(struct btrfs_root *root,
6790 struct btrfs_ioctl_get_dev_stats *stats)
6792 struct btrfs_device *dev;
6793 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
6796 mutex_lock(&fs_devices->device_list_mutex);
6797 dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
6798 mutex_unlock(&fs_devices->device_list_mutex);
6801 btrfs_warn(root->fs_info, "get dev_stats failed, device not found");
6803 } else if (!dev->dev_stats_valid) {
6804 btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid");
6806 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
6807 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
6808 if (stats->nr_items > i)
6810 btrfs_dev_stat_read_and_reset(dev, i);
6812 btrfs_dev_stat_reset(dev, i);
6815 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
6816 if (stats->nr_items > i)
6817 stats->values[i] = btrfs_dev_stat_read(dev, i);
6819 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
6820 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
6824 int btrfs_scratch_superblock(struct btrfs_device *device)
6826 struct buffer_head *bh;
6827 struct btrfs_super_block *disk_super;
6829 bh = btrfs_read_dev_super(device->bdev);
6832 disk_super = (struct btrfs_super_block *)bh->b_data;
6834 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
6835 set_buffer_dirty(bh);
6836 sync_dirty_buffer(bh);
6843 * Update the size of all devices, which is used for writing out the
6846 void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info)
6848 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6849 struct btrfs_device *curr, *next;
6851 if (list_empty(&fs_devices->resized_devices))
6854 mutex_lock(&fs_devices->device_list_mutex);
6855 lock_chunks(fs_info->dev_root);
6856 list_for_each_entry_safe(curr, next, &fs_devices->resized_devices,
6858 list_del_init(&curr->resized_list);
6859 curr->commit_total_bytes = curr->disk_total_bytes;
6861 unlock_chunks(fs_info->dev_root);
6862 mutex_unlock(&fs_devices->device_list_mutex);
6865 /* Must be invoked during the transaction commit */
6866 void btrfs_update_commit_device_bytes_used(struct btrfs_root *root,
6867 struct btrfs_transaction *transaction)
6869 struct extent_map *em;
6870 struct map_lookup *map;
6871 struct btrfs_device *dev;
6874 if (list_empty(&transaction->pending_chunks))
6877 /* In order to kick the device replace finish process */
6879 list_for_each_entry(em, &transaction->pending_chunks, list) {
6880 map = (struct map_lookup *)em->bdev;
6882 for (i = 0; i < map->num_stripes; i++) {
6883 dev = map->stripes[i].dev;
6884 dev->commit_bytes_used = dev->bytes_used;
6887 unlock_chunks(root);
6890 void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info)
6892 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6893 while (fs_devices) {
6894 fs_devices->fs_info = fs_info;
6895 fs_devices = fs_devices->seed;
6899 void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info)
6901 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
6902 while (fs_devices) {
6903 fs_devices->fs_info = NULL;
6904 fs_devices = fs_devices->seed;