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/kthread.h>
27 #include <asm/div64.h>
30 #include "extent_map.h"
32 #include "transaction.h"
33 #include "print-tree.h"
35 #include "async-thread.h"
36 #include "check-integrity.h"
38 static int init_first_rw_device(struct btrfs_trans_handle *trans,
39 struct btrfs_root *root,
40 struct btrfs_device *device);
41 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
43 static DEFINE_MUTEX(uuid_mutex);
44 static LIST_HEAD(fs_uuids);
46 static void lock_chunks(struct btrfs_root *root)
48 mutex_lock(&root->fs_info->chunk_mutex);
51 static void unlock_chunks(struct btrfs_root *root)
53 mutex_unlock(&root->fs_info->chunk_mutex);
56 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
58 struct btrfs_device *device;
59 WARN_ON(fs_devices->opened);
60 while (!list_empty(&fs_devices->devices)) {
61 device = list_entry(fs_devices->devices.next,
62 struct btrfs_device, dev_list);
63 list_del(&device->dev_list);
70 void btrfs_cleanup_fs_uuids(void)
72 struct btrfs_fs_devices *fs_devices;
74 while (!list_empty(&fs_uuids)) {
75 fs_devices = list_entry(fs_uuids.next,
76 struct btrfs_fs_devices, list);
77 list_del(&fs_devices->list);
78 free_fs_devices(fs_devices);
82 static noinline struct btrfs_device *__find_device(struct list_head *head,
85 struct btrfs_device *dev;
87 list_for_each_entry(dev, head, dev_list) {
88 if (dev->devid == devid &&
89 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
96 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
98 struct btrfs_fs_devices *fs_devices;
100 list_for_each_entry(fs_devices, &fs_uuids, list) {
101 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
107 static void requeue_list(struct btrfs_pending_bios *pending_bios,
108 struct bio *head, struct bio *tail)
111 struct bio *old_head;
113 old_head = pending_bios->head;
114 pending_bios->head = head;
115 if (pending_bios->tail)
116 tail->bi_next = old_head;
118 pending_bios->tail = tail;
122 * we try to collect pending bios for a device so we don't get a large
123 * number of procs sending bios down to the same device. This greatly
124 * improves the schedulers ability to collect and merge the bios.
126 * But, it also turns into a long list of bios to process and that is sure
127 * to eventually make the worker thread block. The solution here is to
128 * make some progress and then put this work struct back at the end of
129 * the list if the block device is congested. This way, multiple devices
130 * can make progress from a single worker thread.
132 static noinline void run_scheduled_bios(struct btrfs_device *device)
135 struct backing_dev_info *bdi;
136 struct btrfs_fs_info *fs_info;
137 struct btrfs_pending_bios *pending_bios;
141 unsigned long num_run;
142 unsigned long batch_run = 0;
144 unsigned long last_waited = 0;
146 int sync_pending = 0;
147 struct blk_plug plug;
150 * this function runs all the bios we've collected for
151 * a particular device. We don't want to wander off to
152 * another device without first sending all of these down.
153 * So, setup a plug here and finish it off before we return
155 blk_start_plug(&plug);
157 bdi = blk_get_backing_dev_info(device->bdev);
158 fs_info = device->dev_root->fs_info;
159 limit = btrfs_async_submit_limit(fs_info);
160 limit = limit * 2 / 3;
163 spin_lock(&device->io_lock);
168 /* take all the bios off the list at once and process them
169 * later on (without the lock held). But, remember the
170 * tail and other pointers so the bios can be properly reinserted
171 * into the list if we hit congestion
173 if (!force_reg && device->pending_sync_bios.head) {
174 pending_bios = &device->pending_sync_bios;
177 pending_bios = &device->pending_bios;
181 pending = pending_bios->head;
182 tail = pending_bios->tail;
183 WARN_ON(pending && !tail);
186 * if pending was null this time around, no bios need processing
187 * at all and we can stop. Otherwise it'll loop back up again
188 * and do an additional check so no bios are missed.
190 * device->running_pending is used to synchronize with the
193 if (device->pending_sync_bios.head == NULL &&
194 device->pending_bios.head == NULL) {
196 device->running_pending = 0;
199 device->running_pending = 1;
202 pending_bios->head = NULL;
203 pending_bios->tail = NULL;
205 spin_unlock(&device->io_lock);
210 /* we want to work on both lists, but do more bios on the
211 * sync list than the regular list
214 pending_bios != &device->pending_sync_bios &&
215 device->pending_sync_bios.head) ||
216 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
217 device->pending_bios.head)) {
218 spin_lock(&device->io_lock);
219 requeue_list(pending_bios, pending, tail);
224 pending = pending->bi_next;
226 atomic_dec(&fs_info->nr_async_bios);
228 if (atomic_read(&fs_info->nr_async_bios) < limit &&
229 waitqueue_active(&fs_info->async_submit_wait))
230 wake_up(&fs_info->async_submit_wait);
232 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
235 * if we're doing the sync list, record that our
236 * plug has some sync requests on it
238 * If we're doing the regular list and there are
239 * sync requests sitting around, unplug before
242 if (pending_bios == &device->pending_sync_bios) {
244 } else if (sync_pending) {
245 blk_finish_plug(&plug);
246 blk_start_plug(&plug);
250 btrfsic_submit_bio(cur->bi_rw, cur);
257 * we made progress, there is more work to do and the bdi
258 * is now congested. Back off and let other work structs
261 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
262 fs_info->fs_devices->open_devices > 1) {
263 struct io_context *ioc;
265 ioc = current->io_context;
268 * the main goal here is that we don't want to
269 * block if we're going to be able to submit
270 * more requests without blocking.
272 * This code does two great things, it pokes into
273 * the elevator code from a filesystem _and_
274 * it makes assumptions about how batching works.
276 if (ioc && ioc->nr_batch_requests > 0 &&
277 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
279 ioc->last_waited == last_waited)) {
281 * we want to go through our batch of
282 * requests and stop. So, we copy out
283 * the ioc->last_waited time and test
284 * against it before looping
286 last_waited = ioc->last_waited;
291 spin_lock(&device->io_lock);
292 requeue_list(pending_bios, pending, tail);
293 device->running_pending = 1;
295 spin_unlock(&device->io_lock);
296 btrfs_requeue_work(&device->work);
299 /* unplug every 64 requests just for good measure */
300 if (batch_run % 64 == 0) {
301 blk_finish_plug(&plug);
302 blk_start_plug(&plug);
311 spin_lock(&device->io_lock);
312 if (device->pending_bios.head || device->pending_sync_bios.head)
314 spin_unlock(&device->io_lock);
317 blk_finish_plug(&plug);
320 static void pending_bios_fn(struct btrfs_work *work)
322 struct btrfs_device *device;
324 device = container_of(work, struct btrfs_device, work);
325 run_scheduled_bios(device);
328 static noinline int device_list_add(const char *path,
329 struct btrfs_super_block *disk_super,
330 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
332 struct btrfs_device *device;
333 struct btrfs_fs_devices *fs_devices;
334 u64 found_transid = btrfs_super_generation(disk_super);
337 fs_devices = find_fsid(disk_super->fsid);
339 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
342 INIT_LIST_HEAD(&fs_devices->devices);
343 INIT_LIST_HEAD(&fs_devices->alloc_list);
344 list_add(&fs_devices->list, &fs_uuids);
345 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
346 fs_devices->latest_devid = devid;
347 fs_devices->latest_trans = found_transid;
348 mutex_init(&fs_devices->device_list_mutex);
351 device = __find_device(&fs_devices->devices, devid,
352 disk_super->dev_item.uuid);
355 if (fs_devices->opened)
358 device = kzalloc(sizeof(*device), GFP_NOFS);
360 /* we can safely leave the fs_devices entry around */
363 device->devid = devid;
364 device->work.func = pending_bios_fn;
365 memcpy(device->uuid, disk_super->dev_item.uuid,
367 spin_lock_init(&device->io_lock);
368 device->name = kstrdup(path, GFP_NOFS);
373 INIT_LIST_HEAD(&device->dev_alloc_list);
375 /* init readahead state */
376 spin_lock_init(&device->reada_lock);
377 device->reada_curr_zone = NULL;
378 atomic_set(&device->reada_in_flight, 0);
379 device->reada_next = 0;
380 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
381 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
383 mutex_lock(&fs_devices->device_list_mutex);
384 list_add_rcu(&device->dev_list, &fs_devices->devices);
385 mutex_unlock(&fs_devices->device_list_mutex);
387 device->fs_devices = fs_devices;
388 fs_devices->num_devices++;
389 } else if (!device->name || strcmp(device->name, path)) {
390 name = kstrdup(path, GFP_NOFS);
395 if (device->missing) {
396 fs_devices->missing_devices--;
401 if (found_transid > fs_devices->latest_trans) {
402 fs_devices->latest_devid = devid;
403 fs_devices->latest_trans = found_transid;
405 *fs_devices_ret = fs_devices;
409 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
411 struct btrfs_fs_devices *fs_devices;
412 struct btrfs_device *device;
413 struct btrfs_device *orig_dev;
415 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
417 return ERR_PTR(-ENOMEM);
419 INIT_LIST_HEAD(&fs_devices->devices);
420 INIT_LIST_HEAD(&fs_devices->alloc_list);
421 INIT_LIST_HEAD(&fs_devices->list);
422 mutex_init(&fs_devices->device_list_mutex);
423 fs_devices->latest_devid = orig->latest_devid;
424 fs_devices->latest_trans = orig->latest_trans;
425 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
427 /* We have held the volume lock, it is safe to get the devices. */
428 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
429 device = kzalloc(sizeof(*device), GFP_NOFS);
433 device->name = kstrdup(orig_dev->name, GFP_NOFS);
439 device->devid = orig_dev->devid;
440 device->work.func = pending_bios_fn;
441 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
442 spin_lock_init(&device->io_lock);
443 INIT_LIST_HEAD(&device->dev_list);
444 INIT_LIST_HEAD(&device->dev_alloc_list);
446 list_add(&device->dev_list, &fs_devices->devices);
447 device->fs_devices = fs_devices;
448 fs_devices->num_devices++;
452 free_fs_devices(fs_devices);
453 return ERR_PTR(-ENOMEM);
456 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
458 struct btrfs_device *device, *next;
460 struct block_device *latest_bdev = NULL;
461 u64 latest_devid = 0;
462 u64 latest_transid = 0;
464 mutex_lock(&uuid_mutex);
466 /* This is the initialized path, it is safe to release the devices. */
467 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
468 if (device->in_fs_metadata) {
469 if (!latest_transid ||
470 device->generation > latest_transid) {
471 latest_devid = device->devid;
472 latest_transid = device->generation;
473 latest_bdev = device->bdev;
479 blkdev_put(device->bdev, device->mode);
481 fs_devices->open_devices--;
483 if (device->writeable) {
484 list_del_init(&device->dev_alloc_list);
485 device->writeable = 0;
486 fs_devices->rw_devices--;
488 list_del_init(&device->dev_list);
489 fs_devices->num_devices--;
494 if (fs_devices->seed) {
495 fs_devices = fs_devices->seed;
499 fs_devices->latest_bdev = latest_bdev;
500 fs_devices->latest_devid = latest_devid;
501 fs_devices->latest_trans = latest_transid;
503 mutex_unlock(&uuid_mutex);
506 static void __free_device(struct work_struct *work)
508 struct btrfs_device *device;
510 device = container_of(work, struct btrfs_device, rcu_work);
513 blkdev_put(device->bdev, device->mode);
519 static void free_device(struct rcu_head *head)
521 struct btrfs_device *device;
523 device = container_of(head, struct btrfs_device, rcu);
525 INIT_WORK(&device->rcu_work, __free_device);
526 schedule_work(&device->rcu_work);
529 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
531 struct btrfs_device *device;
533 if (--fs_devices->opened > 0)
536 mutex_lock(&fs_devices->device_list_mutex);
537 list_for_each_entry(device, &fs_devices->devices, dev_list) {
538 struct btrfs_device *new_device;
541 fs_devices->open_devices--;
543 if (device->writeable) {
544 list_del_init(&device->dev_alloc_list);
545 fs_devices->rw_devices--;
548 if (device->can_discard)
549 fs_devices->num_can_discard--;
551 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
552 BUG_ON(!new_device); /* -ENOMEM */
553 memcpy(new_device, device, sizeof(*new_device));
554 new_device->name = kstrdup(device->name, GFP_NOFS);
555 BUG_ON(device->name && !new_device->name); /* -ENOMEM */
556 new_device->bdev = NULL;
557 new_device->writeable = 0;
558 new_device->in_fs_metadata = 0;
559 new_device->can_discard = 0;
560 list_replace_rcu(&device->dev_list, &new_device->dev_list);
562 call_rcu(&device->rcu, free_device);
564 mutex_unlock(&fs_devices->device_list_mutex);
566 WARN_ON(fs_devices->open_devices);
567 WARN_ON(fs_devices->rw_devices);
568 fs_devices->opened = 0;
569 fs_devices->seeding = 0;
574 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
576 struct btrfs_fs_devices *seed_devices = NULL;
579 mutex_lock(&uuid_mutex);
580 ret = __btrfs_close_devices(fs_devices);
581 if (!fs_devices->opened) {
582 seed_devices = fs_devices->seed;
583 fs_devices->seed = NULL;
585 mutex_unlock(&uuid_mutex);
587 while (seed_devices) {
588 fs_devices = seed_devices;
589 seed_devices = fs_devices->seed;
590 __btrfs_close_devices(fs_devices);
591 free_fs_devices(fs_devices);
596 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
597 fmode_t flags, void *holder)
599 struct request_queue *q;
600 struct block_device *bdev;
601 struct list_head *head = &fs_devices->devices;
602 struct btrfs_device *device;
603 struct block_device *latest_bdev = NULL;
604 struct buffer_head *bh;
605 struct btrfs_super_block *disk_super;
606 u64 latest_devid = 0;
607 u64 latest_transid = 0;
614 list_for_each_entry(device, head, dev_list) {
620 bdev = blkdev_get_by_path(device->name, flags, holder);
622 printk(KERN_INFO "open %s failed\n", device->name);
625 filemap_write_and_wait(bdev->bd_inode->i_mapping);
626 invalidate_bdev(bdev);
627 set_blocksize(bdev, 4096);
629 bh = btrfs_read_dev_super(bdev);
633 disk_super = (struct btrfs_super_block *)bh->b_data;
634 devid = btrfs_stack_device_id(&disk_super->dev_item);
635 if (devid != device->devid)
638 if (memcmp(device->uuid, disk_super->dev_item.uuid,
642 device->generation = btrfs_super_generation(disk_super);
643 if (!latest_transid || device->generation > latest_transid) {
644 latest_devid = devid;
645 latest_transid = device->generation;
649 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
650 device->writeable = 0;
652 device->writeable = !bdev_read_only(bdev);
656 q = bdev_get_queue(bdev);
657 if (blk_queue_discard(q)) {
658 device->can_discard = 1;
659 fs_devices->num_can_discard++;
663 device->in_fs_metadata = 0;
664 device->mode = flags;
666 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
667 fs_devices->rotating = 1;
669 fs_devices->open_devices++;
670 if (device->writeable) {
671 fs_devices->rw_devices++;
672 list_add(&device->dev_alloc_list,
673 &fs_devices->alloc_list);
681 blkdev_put(bdev, flags);
685 if (fs_devices->open_devices == 0) {
689 fs_devices->seeding = seeding;
690 fs_devices->opened = 1;
691 fs_devices->latest_bdev = latest_bdev;
692 fs_devices->latest_devid = latest_devid;
693 fs_devices->latest_trans = latest_transid;
694 fs_devices->total_rw_bytes = 0;
699 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
700 fmode_t flags, void *holder)
704 mutex_lock(&uuid_mutex);
705 if (fs_devices->opened) {
706 fs_devices->opened++;
709 ret = __btrfs_open_devices(fs_devices, flags, holder);
711 mutex_unlock(&uuid_mutex);
715 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
716 struct btrfs_fs_devices **fs_devices_ret)
718 struct btrfs_super_block *disk_super;
719 struct block_device *bdev;
720 struct buffer_head *bh;
726 bdev = blkdev_get_by_path(path, flags, holder);
733 mutex_lock(&uuid_mutex);
734 ret = set_blocksize(bdev, 4096);
737 bh = btrfs_read_dev_super(bdev);
742 disk_super = (struct btrfs_super_block *)bh->b_data;
743 devid = btrfs_stack_device_id(&disk_super->dev_item);
744 transid = btrfs_super_generation(disk_super);
745 if (disk_super->label[0])
746 printk(KERN_INFO "device label %s ", disk_super->label);
748 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
749 printk(KERN_CONT "devid %llu transid %llu %s\n",
750 (unsigned long long)devid, (unsigned long long)transid, path);
751 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
755 mutex_unlock(&uuid_mutex);
756 blkdev_put(bdev, flags);
761 /* helper to account the used device space in the range */
762 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
763 u64 end, u64 *length)
765 struct btrfs_key key;
766 struct btrfs_root *root = device->dev_root;
767 struct btrfs_dev_extent *dev_extent;
768 struct btrfs_path *path;
772 struct extent_buffer *l;
776 if (start >= device->total_bytes)
779 path = btrfs_alloc_path();
784 key.objectid = device->devid;
786 key.type = BTRFS_DEV_EXTENT_KEY;
788 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
792 ret = btrfs_previous_item(root, path, key.objectid, key.type);
799 slot = path->slots[0];
800 if (slot >= btrfs_header_nritems(l)) {
801 ret = btrfs_next_leaf(root, path);
809 btrfs_item_key_to_cpu(l, &key, slot);
811 if (key.objectid < device->devid)
814 if (key.objectid > device->devid)
817 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
820 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
821 extent_end = key.offset + btrfs_dev_extent_length(l,
823 if (key.offset <= start && extent_end > end) {
824 *length = end - start + 1;
826 } else if (key.offset <= start && extent_end > start)
827 *length += extent_end - start;
828 else if (key.offset > start && extent_end <= end)
829 *length += extent_end - key.offset;
830 else if (key.offset > start && key.offset <= end) {
831 *length += end - key.offset + 1;
833 } else if (key.offset > end)
841 btrfs_free_path(path);
846 * find_free_dev_extent - find free space in the specified device
847 * @device: the device which we search the free space in
848 * @num_bytes: the size of the free space that we need
849 * @start: store the start of the free space.
850 * @len: the size of the free space. that we find, or the size of the max
851 * free space if we don't find suitable free space
853 * this uses a pretty simple search, the expectation is that it is
854 * called very infrequently and that a given device has a small number
857 * @start is used to store the start of the free space if we find. But if we
858 * don't find suitable free space, it will be used to store the start position
859 * of the max free space.
861 * @len is used to store the size of the free space that we find.
862 * But if we don't find suitable free space, it is used to store the size of
863 * the max free space.
865 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
866 u64 *start, u64 *len)
868 struct btrfs_key key;
869 struct btrfs_root *root = device->dev_root;
870 struct btrfs_dev_extent *dev_extent;
871 struct btrfs_path *path;
877 u64 search_end = device->total_bytes;
880 struct extent_buffer *l;
882 /* FIXME use last free of some kind */
884 /* we don't want to overwrite the superblock on the drive,
885 * so we make sure to start at an offset of at least 1MB
887 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
889 max_hole_start = search_start;
893 if (search_start >= search_end) {
898 path = btrfs_alloc_path();
905 key.objectid = device->devid;
906 key.offset = search_start;
907 key.type = BTRFS_DEV_EXTENT_KEY;
909 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
913 ret = btrfs_previous_item(root, path, key.objectid, key.type);
920 slot = path->slots[0];
921 if (slot >= btrfs_header_nritems(l)) {
922 ret = btrfs_next_leaf(root, path);
930 btrfs_item_key_to_cpu(l, &key, slot);
932 if (key.objectid < device->devid)
935 if (key.objectid > device->devid)
938 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
941 if (key.offset > search_start) {
942 hole_size = key.offset - search_start;
944 if (hole_size > max_hole_size) {
945 max_hole_start = search_start;
946 max_hole_size = hole_size;
950 * If this free space is greater than which we need,
951 * it must be the max free space that we have found
952 * until now, so max_hole_start must point to the start
953 * of this free space and the length of this free space
954 * is stored in max_hole_size. Thus, we return
955 * max_hole_start and max_hole_size and go back to the
958 if (hole_size >= num_bytes) {
964 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
965 extent_end = key.offset + btrfs_dev_extent_length(l,
967 if (extent_end > search_start)
968 search_start = extent_end;
975 * At this point, search_start should be the end of
976 * allocated dev extents, and when shrinking the device,
977 * search_end may be smaller than search_start.
979 if (search_end > search_start)
980 hole_size = search_end - search_start;
982 if (hole_size > max_hole_size) {
983 max_hole_start = search_start;
984 max_hole_size = hole_size;
988 if (hole_size < num_bytes)
994 btrfs_free_path(path);
996 *start = max_hole_start;
998 *len = max_hole_size;
1002 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1003 struct btrfs_device *device,
1007 struct btrfs_path *path;
1008 struct btrfs_root *root = device->dev_root;
1009 struct btrfs_key key;
1010 struct btrfs_key found_key;
1011 struct extent_buffer *leaf = NULL;
1012 struct btrfs_dev_extent *extent = NULL;
1014 path = btrfs_alloc_path();
1018 key.objectid = device->devid;
1020 key.type = BTRFS_DEV_EXTENT_KEY;
1022 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1024 ret = btrfs_previous_item(root, path, key.objectid,
1025 BTRFS_DEV_EXTENT_KEY);
1028 leaf = path->nodes[0];
1029 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1030 extent = btrfs_item_ptr(leaf, path->slots[0],
1031 struct btrfs_dev_extent);
1032 BUG_ON(found_key.offset > start || found_key.offset +
1033 btrfs_dev_extent_length(leaf, extent) < start);
1035 btrfs_release_path(path);
1037 } else if (ret == 0) {
1038 leaf = path->nodes[0];
1039 extent = btrfs_item_ptr(leaf, path->slots[0],
1040 struct btrfs_dev_extent);
1042 btrfs_error(root->fs_info, ret, "Slot search failed");
1046 if (device->bytes_used > 0) {
1047 u64 len = btrfs_dev_extent_length(leaf, extent);
1048 device->bytes_used -= len;
1049 spin_lock(&root->fs_info->free_chunk_lock);
1050 root->fs_info->free_chunk_space += len;
1051 spin_unlock(&root->fs_info->free_chunk_lock);
1053 ret = btrfs_del_item(trans, root, path);
1055 btrfs_error(root->fs_info, ret,
1056 "Failed to remove dev extent item");
1059 btrfs_free_path(path);
1063 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1064 struct btrfs_device *device,
1065 u64 chunk_tree, u64 chunk_objectid,
1066 u64 chunk_offset, u64 start, u64 num_bytes)
1069 struct btrfs_path *path;
1070 struct btrfs_root *root = device->dev_root;
1071 struct btrfs_dev_extent *extent;
1072 struct extent_buffer *leaf;
1073 struct btrfs_key key;
1075 WARN_ON(!device->in_fs_metadata);
1076 path = btrfs_alloc_path();
1080 key.objectid = device->devid;
1082 key.type = BTRFS_DEV_EXTENT_KEY;
1083 ret = btrfs_insert_empty_item(trans, root, path, &key,
1088 leaf = path->nodes[0];
1089 extent = btrfs_item_ptr(leaf, path->slots[0],
1090 struct btrfs_dev_extent);
1091 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1092 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1093 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1095 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1096 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1099 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1100 btrfs_mark_buffer_dirty(leaf);
1102 btrfs_free_path(path);
1106 static noinline int find_next_chunk(struct btrfs_root *root,
1107 u64 objectid, u64 *offset)
1109 struct btrfs_path *path;
1111 struct btrfs_key key;
1112 struct btrfs_chunk *chunk;
1113 struct btrfs_key found_key;
1115 path = btrfs_alloc_path();
1119 key.objectid = objectid;
1120 key.offset = (u64)-1;
1121 key.type = BTRFS_CHUNK_ITEM_KEY;
1123 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1127 BUG_ON(ret == 0); /* Corruption */
1129 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1133 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1135 if (found_key.objectid != objectid)
1138 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1139 struct btrfs_chunk);
1140 *offset = found_key.offset +
1141 btrfs_chunk_length(path->nodes[0], chunk);
1146 btrfs_free_path(path);
1150 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1153 struct btrfs_key key;
1154 struct btrfs_key found_key;
1155 struct btrfs_path *path;
1157 root = root->fs_info->chunk_root;
1159 path = btrfs_alloc_path();
1163 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1164 key.type = BTRFS_DEV_ITEM_KEY;
1165 key.offset = (u64)-1;
1167 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1171 BUG_ON(ret == 0); /* Corruption */
1173 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1174 BTRFS_DEV_ITEM_KEY);
1178 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1180 *objectid = found_key.offset + 1;
1184 btrfs_free_path(path);
1189 * the device information is stored in the chunk root
1190 * the btrfs_device struct should be fully filled in
1192 int btrfs_add_device(struct btrfs_trans_handle *trans,
1193 struct btrfs_root *root,
1194 struct btrfs_device *device)
1197 struct btrfs_path *path;
1198 struct btrfs_dev_item *dev_item;
1199 struct extent_buffer *leaf;
1200 struct btrfs_key key;
1203 root = root->fs_info->chunk_root;
1205 path = btrfs_alloc_path();
1209 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1210 key.type = BTRFS_DEV_ITEM_KEY;
1211 key.offset = device->devid;
1213 ret = btrfs_insert_empty_item(trans, root, path, &key,
1218 leaf = path->nodes[0];
1219 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1221 btrfs_set_device_id(leaf, dev_item, device->devid);
1222 btrfs_set_device_generation(leaf, dev_item, 0);
1223 btrfs_set_device_type(leaf, dev_item, device->type);
1224 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1225 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1226 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1227 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1228 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1229 btrfs_set_device_group(leaf, dev_item, 0);
1230 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1231 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1232 btrfs_set_device_start_offset(leaf, dev_item, 0);
1234 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1235 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1236 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1237 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1238 btrfs_mark_buffer_dirty(leaf);
1242 btrfs_free_path(path);
1246 static int btrfs_rm_dev_item(struct btrfs_root *root,
1247 struct btrfs_device *device)
1250 struct btrfs_path *path;
1251 struct btrfs_key key;
1252 struct btrfs_trans_handle *trans;
1254 root = root->fs_info->chunk_root;
1256 path = btrfs_alloc_path();
1260 trans = btrfs_start_transaction(root, 0);
1261 if (IS_ERR(trans)) {
1262 btrfs_free_path(path);
1263 return PTR_ERR(trans);
1265 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1266 key.type = BTRFS_DEV_ITEM_KEY;
1267 key.offset = device->devid;
1270 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1279 ret = btrfs_del_item(trans, root, path);
1283 btrfs_free_path(path);
1284 unlock_chunks(root);
1285 btrfs_commit_transaction(trans, root);
1289 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1291 struct btrfs_device *device;
1292 struct btrfs_device *next_device;
1293 struct block_device *bdev;
1294 struct buffer_head *bh = NULL;
1295 struct btrfs_super_block *disk_super;
1296 struct btrfs_fs_devices *cur_devices;
1302 bool clear_super = false;
1304 mutex_lock(&uuid_mutex);
1306 all_avail = root->fs_info->avail_data_alloc_bits |
1307 root->fs_info->avail_system_alloc_bits |
1308 root->fs_info->avail_metadata_alloc_bits;
1310 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1311 root->fs_info->fs_devices->num_devices <= 4) {
1312 printk(KERN_ERR "btrfs: unable to go below four devices "
1318 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1319 root->fs_info->fs_devices->num_devices <= 2) {
1320 printk(KERN_ERR "btrfs: unable to go below two "
1321 "devices on raid1\n");
1326 if (strcmp(device_path, "missing") == 0) {
1327 struct list_head *devices;
1328 struct btrfs_device *tmp;
1331 devices = &root->fs_info->fs_devices->devices;
1333 * It is safe to read the devices since the volume_mutex
1336 list_for_each_entry(tmp, devices, dev_list) {
1337 if (tmp->in_fs_metadata && !tmp->bdev) {
1346 printk(KERN_ERR "btrfs: no missing devices found to "
1351 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1352 root->fs_info->bdev_holder);
1354 ret = PTR_ERR(bdev);
1358 set_blocksize(bdev, 4096);
1359 invalidate_bdev(bdev);
1360 bh = btrfs_read_dev_super(bdev);
1365 disk_super = (struct btrfs_super_block *)bh->b_data;
1366 devid = btrfs_stack_device_id(&disk_super->dev_item);
1367 dev_uuid = disk_super->dev_item.uuid;
1368 device = btrfs_find_device(root, devid, dev_uuid,
1376 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1377 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1383 if (device->writeable) {
1385 list_del_init(&device->dev_alloc_list);
1386 unlock_chunks(root);
1387 root->fs_info->fs_devices->rw_devices--;
1391 ret = btrfs_shrink_device(device, 0);
1395 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1399 spin_lock(&root->fs_info->free_chunk_lock);
1400 root->fs_info->free_chunk_space = device->total_bytes -
1402 spin_unlock(&root->fs_info->free_chunk_lock);
1404 device->in_fs_metadata = 0;
1405 btrfs_scrub_cancel_dev(root, device);
1408 * the device list mutex makes sure that we don't change
1409 * the device list while someone else is writing out all
1410 * the device supers.
1413 cur_devices = device->fs_devices;
1414 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1415 list_del_rcu(&device->dev_list);
1417 device->fs_devices->num_devices--;
1419 if (device->missing)
1420 root->fs_info->fs_devices->missing_devices--;
1422 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1423 struct btrfs_device, dev_list);
1424 if (device->bdev == root->fs_info->sb->s_bdev)
1425 root->fs_info->sb->s_bdev = next_device->bdev;
1426 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1427 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1430 device->fs_devices->open_devices--;
1432 call_rcu(&device->rcu, free_device);
1433 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1435 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1436 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1438 if (cur_devices->open_devices == 0) {
1439 struct btrfs_fs_devices *fs_devices;
1440 fs_devices = root->fs_info->fs_devices;
1441 while (fs_devices) {
1442 if (fs_devices->seed == cur_devices)
1444 fs_devices = fs_devices->seed;
1446 fs_devices->seed = cur_devices->seed;
1447 cur_devices->seed = NULL;
1449 __btrfs_close_devices(cur_devices);
1450 unlock_chunks(root);
1451 free_fs_devices(cur_devices);
1455 * at this point, the device is zero sized. We want to
1456 * remove it from the devices list and zero out the old super
1459 /* make sure this device isn't detected as part of
1462 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1463 set_buffer_dirty(bh);
1464 sync_dirty_buffer(bh);
1473 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1475 mutex_unlock(&uuid_mutex);
1478 if (device->writeable) {
1480 list_add(&device->dev_alloc_list,
1481 &root->fs_info->fs_devices->alloc_list);
1482 unlock_chunks(root);
1483 root->fs_info->fs_devices->rw_devices++;
1489 * does all the dirty work required for changing file system's UUID.
1491 static int btrfs_prepare_sprout(struct btrfs_root *root)
1493 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1494 struct btrfs_fs_devices *old_devices;
1495 struct btrfs_fs_devices *seed_devices;
1496 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1497 struct btrfs_device *device;
1500 BUG_ON(!mutex_is_locked(&uuid_mutex));
1501 if (!fs_devices->seeding)
1504 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1508 old_devices = clone_fs_devices(fs_devices);
1509 if (IS_ERR(old_devices)) {
1510 kfree(seed_devices);
1511 return PTR_ERR(old_devices);
1514 list_add(&old_devices->list, &fs_uuids);
1516 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1517 seed_devices->opened = 1;
1518 INIT_LIST_HEAD(&seed_devices->devices);
1519 INIT_LIST_HEAD(&seed_devices->alloc_list);
1520 mutex_init(&seed_devices->device_list_mutex);
1522 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1523 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1525 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1527 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1528 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1529 device->fs_devices = seed_devices;
1532 fs_devices->seeding = 0;
1533 fs_devices->num_devices = 0;
1534 fs_devices->open_devices = 0;
1535 fs_devices->seed = seed_devices;
1537 generate_random_uuid(fs_devices->fsid);
1538 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1539 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1540 super_flags = btrfs_super_flags(disk_super) &
1541 ~BTRFS_SUPER_FLAG_SEEDING;
1542 btrfs_set_super_flags(disk_super, super_flags);
1548 * strore the expected generation for seed devices in device items.
1550 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1551 struct btrfs_root *root)
1553 struct btrfs_path *path;
1554 struct extent_buffer *leaf;
1555 struct btrfs_dev_item *dev_item;
1556 struct btrfs_device *device;
1557 struct btrfs_key key;
1558 u8 fs_uuid[BTRFS_UUID_SIZE];
1559 u8 dev_uuid[BTRFS_UUID_SIZE];
1563 path = btrfs_alloc_path();
1567 root = root->fs_info->chunk_root;
1568 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1570 key.type = BTRFS_DEV_ITEM_KEY;
1573 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1577 leaf = path->nodes[0];
1579 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1580 ret = btrfs_next_leaf(root, path);
1585 leaf = path->nodes[0];
1586 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1587 btrfs_release_path(path);
1591 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1592 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1593 key.type != BTRFS_DEV_ITEM_KEY)
1596 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1597 struct btrfs_dev_item);
1598 devid = btrfs_device_id(leaf, dev_item);
1599 read_extent_buffer(leaf, dev_uuid,
1600 (unsigned long)btrfs_device_uuid(dev_item),
1602 read_extent_buffer(leaf, fs_uuid,
1603 (unsigned long)btrfs_device_fsid(dev_item),
1605 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1606 BUG_ON(!device); /* Logic error */
1608 if (device->fs_devices->seeding) {
1609 btrfs_set_device_generation(leaf, dev_item,
1610 device->generation);
1611 btrfs_mark_buffer_dirty(leaf);
1619 btrfs_free_path(path);
1623 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1625 struct request_queue *q;
1626 struct btrfs_trans_handle *trans;
1627 struct btrfs_device *device;
1628 struct block_device *bdev;
1629 struct list_head *devices;
1630 struct super_block *sb = root->fs_info->sb;
1632 int seeding_dev = 0;
1635 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1638 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1639 root->fs_info->bdev_holder);
1641 return PTR_ERR(bdev);
1643 if (root->fs_info->fs_devices->seeding) {
1645 down_write(&sb->s_umount);
1646 mutex_lock(&uuid_mutex);
1649 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1651 devices = &root->fs_info->fs_devices->devices;
1653 * we have the volume lock, so we don't need the extra
1654 * device list mutex while reading the list here.
1656 list_for_each_entry(device, devices, dev_list) {
1657 if (device->bdev == bdev) {
1663 device = kzalloc(sizeof(*device), GFP_NOFS);
1665 /* we can safely leave the fs_devices entry around */
1670 device->name = kstrdup(device_path, GFP_NOFS);
1671 if (!device->name) {
1677 ret = find_next_devid(root, &device->devid);
1679 kfree(device->name);
1684 trans = btrfs_start_transaction(root, 0);
1685 if (IS_ERR(trans)) {
1686 kfree(device->name);
1688 ret = PTR_ERR(trans);
1694 q = bdev_get_queue(bdev);
1695 if (blk_queue_discard(q))
1696 device->can_discard = 1;
1697 device->writeable = 1;
1698 device->work.func = pending_bios_fn;
1699 generate_random_uuid(device->uuid);
1700 spin_lock_init(&device->io_lock);
1701 device->generation = trans->transid;
1702 device->io_width = root->sectorsize;
1703 device->io_align = root->sectorsize;
1704 device->sector_size = root->sectorsize;
1705 device->total_bytes = i_size_read(bdev->bd_inode);
1706 device->disk_total_bytes = device->total_bytes;
1707 device->dev_root = root->fs_info->dev_root;
1708 device->bdev = bdev;
1709 device->in_fs_metadata = 1;
1710 device->mode = FMODE_EXCL;
1711 set_blocksize(device->bdev, 4096);
1714 sb->s_flags &= ~MS_RDONLY;
1715 ret = btrfs_prepare_sprout(root);
1716 BUG_ON(ret); /* -ENOMEM */
1719 device->fs_devices = root->fs_info->fs_devices;
1722 * we don't want write_supers to jump in here with our device
1725 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1726 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1727 list_add(&device->dev_alloc_list,
1728 &root->fs_info->fs_devices->alloc_list);
1729 root->fs_info->fs_devices->num_devices++;
1730 root->fs_info->fs_devices->open_devices++;
1731 root->fs_info->fs_devices->rw_devices++;
1732 if (device->can_discard)
1733 root->fs_info->fs_devices->num_can_discard++;
1734 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1736 spin_lock(&root->fs_info->free_chunk_lock);
1737 root->fs_info->free_chunk_space += device->total_bytes;
1738 spin_unlock(&root->fs_info->free_chunk_lock);
1740 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1741 root->fs_info->fs_devices->rotating = 1;
1743 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1744 btrfs_set_super_total_bytes(root->fs_info->super_copy,
1745 total_bytes + device->total_bytes);
1747 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1748 btrfs_set_super_num_devices(root->fs_info->super_copy,
1750 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1753 ret = init_first_rw_device(trans, root, device);
1756 ret = btrfs_finish_sprout(trans, root);
1760 ret = btrfs_add_device(trans, root, device);
1766 * we've got more storage, clear any full flags on the space
1769 btrfs_clear_space_info_full(root->fs_info);
1771 unlock_chunks(root);
1772 ret = btrfs_commit_transaction(trans, root);
1775 mutex_unlock(&uuid_mutex);
1776 up_write(&sb->s_umount);
1778 if (ret) /* transaction commit */
1781 ret = btrfs_relocate_sys_chunks(root);
1783 btrfs_error(root->fs_info, ret,
1784 "Failed to relocate sys chunks after "
1785 "device initialization. This can be fixed "
1786 "using the \"btrfs balance\" command.");
1792 unlock_chunks(root);
1793 btrfs_abort_transaction(trans, root, ret);
1794 btrfs_end_transaction(trans, root);
1795 kfree(device->name);
1798 blkdev_put(bdev, FMODE_EXCL);
1800 mutex_unlock(&uuid_mutex);
1801 up_write(&sb->s_umount);
1806 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1807 struct btrfs_device *device)
1810 struct btrfs_path *path;
1811 struct btrfs_root *root;
1812 struct btrfs_dev_item *dev_item;
1813 struct extent_buffer *leaf;
1814 struct btrfs_key key;
1816 root = device->dev_root->fs_info->chunk_root;
1818 path = btrfs_alloc_path();
1822 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1823 key.type = BTRFS_DEV_ITEM_KEY;
1824 key.offset = device->devid;
1826 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1835 leaf = path->nodes[0];
1836 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1838 btrfs_set_device_id(leaf, dev_item, device->devid);
1839 btrfs_set_device_type(leaf, dev_item, device->type);
1840 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1841 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1842 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1843 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1844 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1845 btrfs_mark_buffer_dirty(leaf);
1848 btrfs_free_path(path);
1852 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1853 struct btrfs_device *device, u64 new_size)
1855 struct btrfs_super_block *super_copy =
1856 device->dev_root->fs_info->super_copy;
1857 u64 old_total = btrfs_super_total_bytes(super_copy);
1858 u64 diff = new_size - device->total_bytes;
1860 if (!device->writeable)
1862 if (new_size <= device->total_bytes)
1865 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1866 device->fs_devices->total_rw_bytes += diff;
1868 device->total_bytes = new_size;
1869 device->disk_total_bytes = new_size;
1870 btrfs_clear_space_info_full(device->dev_root->fs_info);
1872 return btrfs_update_device(trans, device);
1875 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1876 struct btrfs_device *device, u64 new_size)
1879 lock_chunks(device->dev_root);
1880 ret = __btrfs_grow_device(trans, device, new_size);
1881 unlock_chunks(device->dev_root);
1885 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1886 struct btrfs_root *root,
1887 u64 chunk_tree, u64 chunk_objectid,
1891 struct btrfs_path *path;
1892 struct btrfs_key key;
1894 root = root->fs_info->chunk_root;
1895 path = btrfs_alloc_path();
1899 key.objectid = chunk_objectid;
1900 key.offset = chunk_offset;
1901 key.type = BTRFS_CHUNK_ITEM_KEY;
1903 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1906 else if (ret > 0) { /* Logic error or corruption */
1907 btrfs_error(root->fs_info, -ENOENT,
1908 "Failed lookup while freeing chunk.");
1913 ret = btrfs_del_item(trans, root, path);
1915 btrfs_error(root->fs_info, ret,
1916 "Failed to delete chunk item.");
1918 btrfs_free_path(path);
1922 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1925 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1926 struct btrfs_disk_key *disk_key;
1927 struct btrfs_chunk *chunk;
1934 struct btrfs_key key;
1936 array_size = btrfs_super_sys_array_size(super_copy);
1938 ptr = super_copy->sys_chunk_array;
1941 while (cur < array_size) {
1942 disk_key = (struct btrfs_disk_key *)ptr;
1943 btrfs_disk_key_to_cpu(&key, disk_key);
1945 len = sizeof(*disk_key);
1947 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1948 chunk = (struct btrfs_chunk *)(ptr + len);
1949 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1950 len += btrfs_chunk_item_size(num_stripes);
1955 if (key.objectid == chunk_objectid &&
1956 key.offset == chunk_offset) {
1957 memmove(ptr, ptr + len, array_size - (cur + len));
1959 btrfs_set_super_sys_array_size(super_copy, array_size);
1968 static int btrfs_relocate_chunk(struct btrfs_root *root,
1969 u64 chunk_tree, u64 chunk_objectid,
1972 struct extent_map_tree *em_tree;
1973 struct btrfs_root *extent_root;
1974 struct btrfs_trans_handle *trans;
1975 struct extent_map *em;
1976 struct map_lookup *map;
1980 root = root->fs_info->chunk_root;
1981 extent_root = root->fs_info->extent_root;
1982 em_tree = &root->fs_info->mapping_tree.map_tree;
1984 ret = btrfs_can_relocate(extent_root, chunk_offset);
1988 /* step one, relocate all the extents inside this chunk */
1989 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1993 trans = btrfs_start_transaction(root, 0);
1994 BUG_ON(IS_ERR(trans));
1999 * step two, delete the device extents and the
2000 * chunk tree entries
2002 read_lock(&em_tree->lock);
2003 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2004 read_unlock(&em_tree->lock);
2006 BUG_ON(!em || em->start > chunk_offset ||
2007 em->start + em->len < chunk_offset);
2008 map = (struct map_lookup *)em->bdev;
2010 for (i = 0; i < map->num_stripes; i++) {
2011 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2012 map->stripes[i].physical);
2015 if (map->stripes[i].dev) {
2016 ret = btrfs_update_device(trans, map->stripes[i].dev);
2020 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2025 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2027 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2028 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2032 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2035 write_lock(&em_tree->lock);
2036 remove_extent_mapping(em_tree, em);
2037 write_unlock(&em_tree->lock);
2042 /* once for the tree */
2043 free_extent_map(em);
2045 free_extent_map(em);
2047 unlock_chunks(root);
2048 btrfs_end_transaction(trans, root);
2052 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2054 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2055 struct btrfs_path *path;
2056 struct extent_buffer *leaf;
2057 struct btrfs_chunk *chunk;
2058 struct btrfs_key key;
2059 struct btrfs_key found_key;
2060 u64 chunk_tree = chunk_root->root_key.objectid;
2062 bool retried = false;
2066 path = btrfs_alloc_path();
2071 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2072 key.offset = (u64)-1;
2073 key.type = BTRFS_CHUNK_ITEM_KEY;
2076 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2079 BUG_ON(ret == 0); /* Corruption */
2081 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2088 leaf = path->nodes[0];
2089 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2091 chunk = btrfs_item_ptr(leaf, path->slots[0],
2092 struct btrfs_chunk);
2093 chunk_type = btrfs_chunk_type(leaf, chunk);
2094 btrfs_release_path(path);
2096 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2097 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2106 if (found_key.offset == 0)
2108 key.offset = found_key.offset - 1;
2111 if (failed && !retried) {
2115 } else if (failed && retried) {
2120 btrfs_free_path(path);
2124 static int insert_balance_item(struct btrfs_root *root,
2125 struct btrfs_balance_control *bctl)
2127 struct btrfs_trans_handle *trans;
2128 struct btrfs_balance_item *item;
2129 struct btrfs_disk_balance_args disk_bargs;
2130 struct btrfs_path *path;
2131 struct extent_buffer *leaf;
2132 struct btrfs_key key;
2135 path = btrfs_alloc_path();
2139 trans = btrfs_start_transaction(root, 0);
2140 if (IS_ERR(trans)) {
2141 btrfs_free_path(path);
2142 return PTR_ERR(trans);
2145 key.objectid = BTRFS_BALANCE_OBJECTID;
2146 key.type = BTRFS_BALANCE_ITEM_KEY;
2149 ret = btrfs_insert_empty_item(trans, root, path, &key,
2154 leaf = path->nodes[0];
2155 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2157 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2159 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2160 btrfs_set_balance_data(leaf, item, &disk_bargs);
2161 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2162 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2163 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2164 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2166 btrfs_set_balance_flags(leaf, item, bctl->flags);
2168 btrfs_mark_buffer_dirty(leaf);
2170 btrfs_free_path(path);
2171 err = btrfs_commit_transaction(trans, root);
2177 static int del_balance_item(struct btrfs_root *root)
2179 struct btrfs_trans_handle *trans;
2180 struct btrfs_path *path;
2181 struct btrfs_key key;
2184 path = btrfs_alloc_path();
2188 trans = btrfs_start_transaction(root, 0);
2189 if (IS_ERR(trans)) {
2190 btrfs_free_path(path);
2191 return PTR_ERR(trans);
2194 key.objectid = BTRFS_BALANCE_OBJECTID;
2195 key.type = BTRFS_BALANCE_ITEM_KEY;
2198 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2206 ret = btrfs_del_item(trans, root, path);
2208 btrfs_free_path(path);
2209 err = btrfs_commit_transaction(trans, root);
2216 * This is a heuristic used to reduce the number of chunks balanced on
2217 * resume after balance was interrupted.
2219 static void update_balance_args(struct btrfs_balance_control *bctl)
2222 * Turn on soft mode for chunk types that were being converted.
2224 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2225 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2226 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2227 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2228 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2229 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2232 * Turn on usage filter if is not already used. The idea is
2233 * that chunks that we have already balanced should be
2234 * reasonably full. Don't do it for chunks that are being
2235 * converted - that will keep us from relocating unconverted
2236 * (albeit full) chunks.
2238 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2239 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2240 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2241 bctl->data.usage = 90;
2243 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2244 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2245 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2246 bctl->sys.usage = 90;
2248 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2249 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2250 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2251 bctl->meta.usage = 90;
2256 * Should be called with both balance and volume mutexes held to
2257 * serialize other volume operations (add_dev/rm_dev/resize) with
2258 * restriper. Same goes for unset_balance_control.
2260 static void set_balance_control(struct btrfs_balance_control *bctl)
2262 struct btrfs_fs_info *fs_info = bctl->fs_info;
2264 BUG_ON(fs_info->balance_ctl);
2266 spin_lock(&fs_info->balance_lock);
2267 fs_info->balance_ctl = bctl;
2268 spin_unlock(&fs_info->balance_lock);
2271 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2273 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2275 BUG_ON(!fs_info->balance_ctl);
2277 spin_lock(&fs_info->balance_lock);
2278 fs_info->balance_ctl = NULL;
2279 spin_unlock(&fs_info->balance_lock);
2285 * Balance filters. Return 1 if chunk should be filtered out
2286 * (should not be balanced).
2288 static int chunk_profiles_filter(u64 chunk_type,
2289 struct btrfs_balance_args *bargs)
2291 chunk_type = chunk_to_extended(chunk_type) &
2292 BTRFS_EXTENDED_PROFILE_MASK;
2294 if (bargs->profiles & chunk_type)
2300 static u64 div_factor_fine(u64 num, int factor)
2312 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2313 struct btrfs_balance_args *bargs)
2315 struct btrfs_block_group_cache *cache;
2316 u64 chunk_used, user_thresh;
2319 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2320 chunk_used = btrfs_block_group_used(&cache->item);
2322 user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2323 if (chunk_used < user_thresh)
2326 btrfs_put_block_group(cache);
2330 static int chunk_devid_filter(struct extent_buffer *leaf,
2331 struct btrfs_chunk *chunk,
2332 struct btrfs_balance_args *bargs)
2334 struct btrfs_stripe *stripe;
2335 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2338 for (i = 0; i < num_stripes; i++) {
2339 stripe = btrfs_stripe_nr(chunk, i);
2340 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2347 /* [pstart, pend) */
2348 static int chunk_drange_filter(struct extent_buffer *leaf,
2349 struct btrfs_chunk *chunk,
2351 struct btrfs_balance_args *bargs)
2353 struct btrfs_stripe *stripe;
2354 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2360 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2363 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2364 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2368 factor = num_stripes / factor;
2370 for (i = 0; i < num_stripes; i++) {
2371 stripe = btrfs_stripe_nr(chunk, i);
2372 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2375 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2376 stripe_length = btrfs_chunk_length(leaf, chunk);
2377 do_div(stripe_length, factor);
2379 if (stripe_offset < bargs->pend &&
2380 stripe_offset + stripe_length > bargs->pstart)
2387 /* [vstart, vend) */
2388 static int chunk_vrange_filter(struct extent_buffer *leaf,
2389 struct btrfs_chunk *chunk,
2391 struct btrfs_balance_args *bargs)
2393 if (chunk_offset < bargs->vend &&
2394 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2395 /* at least part of the chunk is inside this vrange */
2401 static int chunk_soft_convert_filter(u64 chunk_type,
2402 struct btrfs_balance_args *bargs)
2404 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2407 chunk_type = chunk_to_extended(chunk_type) &
2408 BTRFS_EXTENDED_PROFILE_MASK;
2410 if (bargs->target == chunk_type)
2416 static int should_balance_chunk(struct btrfs_root *root,
2417 struct extent_buffer *leaf,
2418 struct btrfs_chunk *chunk, u64 chunk_offset)
2420 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2421 struct btrfs_balance_args *bargs = NULL;
2422 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2425 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2426 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2430 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2431 bargs = &bctl->data;
2432 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2434 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2435 bargs = &bctl->meta;
2437 /* profiles filter */
2438 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2439 chunk_profiles_filter(chunk_type, bargs)) {
2444 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2445 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2450 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2451 chunk_devid_filter(leaf, chunk, bargs)) {
2455 /* drange filter, makes sense only with devid filter */
2456 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2457 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2462 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2463 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2467 /* soft profile changing mode */
2468 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2469 chunk_soft_convert_filter(chunk_type, bargs)) {
2476 static u64 div_factor(u64 num, int factor)
2485 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2487 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2488 struct btrfs_root *chunk_root = fs_info->chunk_root;
2489 struct btrfs_root *dev_root = fs_info->dev_root;
2490 struct list_head *devices;
2491 struct btrfs_device *device;
2494 struct btrfs_chunk *chunk;
2495 struct btrfs_path *path;
2496 struct btrfs_key key;
2497 struct btrfs_key found_key;
2498 struct btrfs_trans_handle *trans;
2499 struct extent_buffer *leaf;
2502 int enospc_errors = 0;
2503 bool counting = true;
2505 /* step one make some room on all the devices */
2506 devices = &fs_info->fs_devices->devices;
2507 list_for_each_entry(device, devices, dev_list) {
2508 old_size = device->total_bytes;
2509 size_to_free = div_factor(old_size, 1);
2510 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2511 if (!device->writeable ||
2512 device->total_bytes - device->bytes_used > size_to_free)
2515 ret = btrfs_shrink_device(device, old_size - size_to_free);
2520 trans = btrfs_start_transaction(dev_root, 0);
2521 BUG_ON(IS_ERR(trans));
2523 ret = btrfs_grow_device(trans, device, old_size);
2526 btrfs_end_transaction(trans, dev_root);
2529 /* step two, relocate all the chunks */
2530 path = btrfs_alloc_path();
2536 /* zero out stat counters */
2537 spin_lock(&fs_info->balance_lock);
2538 memset(&bctl->stat, 0, sizeof(bctl->stat));
2539 spin_unlock(&fs_info->balance_lock);
2541 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2542 key.offset = (u64)-1;
2543 key.type = BTRFS_CHUNK_ITEM_KEY;
2546 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2547 atomic_read(&fs_info->balance_cancel_req)) {
2552 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2557 * this shouldn't happen, it means the last relocate
2561 BUG(); /* FIXME break ? */
2563 ret = btrfs_previous_item(chunk_root, path, 0,
2564 BTRFS_CHUNK_ITEM_KEY);
2570 leaf = path->nodes[0];
2571 slot = path->slots[0];
2572 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2574 if (found_key.objectid != key.objectid)
2577 /* chunk zero is special */
2578 if (found_key.offset == 0)
2581 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2584 spin_lock(&fs_info->balance_lock);
2585 bctl->stat.considered++;
2586 spin_unlock(&fs_info->balance_lock);
2589 ret = should_balance_chunk(chunk_root, leaf, chunk,
2591 btrfs_release_path(path);
2596 spin_lock(&fs_info->balance_lock);
2597 bctl->stat.expected++;
2598 spin_unlock(&fs_info->balance_lock);
2602 ret = btrfs_relocate_chunk(chunk_root,
2603 chunk_root->root_key.objectid,
2606 if (ret && ret != -ENOSPC)
2608 if (ret == -ENOSPC) {
2611 spin_lock(&fs_info->balance_lock);
2612 bctl->stat.completed++;
2613 spin_unlock(&fs_info->balance_lock);
2616 key.offset = found_key.offset - 1;
2620 btrfs_release_path(path);
2625 btrfs_free_path(path);
2626 if (enospc_errors) {
2627 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2637 * alloc_profile_is_valid - see if a given profile is valid and reduced
2638 * @flags: profile to validate
2639 * @extended: if true @flags is treated as an extended profile
2641 static int alloc_profile_is_valid(u64 flags, int extended)
2643 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2644 BTRFS_BLOCK_GROUP_PROFILE_MASK);
2646 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2648 /* 1) check that all other bits are zeroed */
2652 /* 2) see if profile is reduced */
2654 return !extended; /* "0" is valid for usual profiles */
2656 /* true if exactly one bit set */
2657 return (flags & (flags - 1)) == 0;
2660 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2662 /* cancel requested || normal exit path */
2663 return atomic_read(&fs_info->balance_cancel_req) ||
2664 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2665 atomic_read(&fs_info->balance_cancel_req) == 0);
2668 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2672 unset_balance_control(fs_info);
2673 ret = del_balance_item(fs_info->tree_root);
2677 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2678 struct btrfs_ioctl_balance_args *bargs);
2681 * Should be called with both balance and volume mutexes held
2683 int btrfs_balance(struct btrfs_balance_control *bctl,
2684 struct btrfs_ioctl_balance_args *bargs)
2686 struct btrfs_fs_info *fs_info = bctl->fs_info;
2691 if (btrfs_fs_closing(fs_info) ||
2692 atomic_read(&fs_info->balance_pause_req) ||
2693 atomic_read(&fs_info->balance_cancel_req)) {
2698 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2699 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2703 * In case of mixed groups both data and meta should be picked,
2704 * and identical options should be given for both of them.
2706 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2707 if (mixed && (bctl->flags & allowed)) {
2708 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2709 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2710 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2711 printk(KERN_ERR "btrfs: with mixed groups data and "
2712 "metadata balance options must be the same\n");
2718 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2719 if (fs_info->fs_devices->num_devices == 1)
2720 allowed |= BTRFS_BLOCK_GROUP_DUP;
2721 else if (fs_info->fs_devices->num_devices < 4)
2722 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2724 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2725 BTRFS_BLOCK_GROUP_RAID10);
2727 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2728 (!alloc_profile_is_valid(bctl->data.target, 1) ||
2729 (bctl->data.target & ~allowed))) {
2730 printk(KERN_ERR "btrfs: unable to start balance with target "
2731 "data profile %llu\n",
2732 (unsigned long long)bctl->data.target);
2736 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2737 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2738 (bctl->meta.target & ~allowed))) {
2739 printk(KERN_ERR "btrfs: unable to start balance with target "
2740 "metadata profile %llu\n",
2741 (unsigned long long)bctl->meta.target);
2745 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2746 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2747 (bctl->sys.target & ~allowed))) {
2748 printk(KERN_ERR "btrfs: unable to start balance with target "
2749 "system profile %llu\n",
2750 (unsigned long long)bctl->sys.target);
2755 /* allow dup'ed data chunks only in mixed mode */
2756 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2757 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2758 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2763 /* allow to reduce meta or sys integrity only if force set */
2764 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2765 BTRFS_BLOCK_GROUP_RAID10;
2766 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2767 (fs_info->avail_system_alloc_bits & allowed) &&
2768 !(bctl->sys.target & allowed)) ||
2769 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2770 (fs_info->avail_metadata_alloc_bits & allowed) &&
2771 !(bctl->meta.target & allowed))) {
2772 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2773 printk(KERN_INFO "btrfs: force reducing metadata "
2776 printk(KERN_ERR "btrfs: balance will reduce metadata "
2777 "integrity, use force if you want this\n");
2783 ret = insert_balance_item(fs_info->tree_root, bctl);
2784 if (ret && ret != -EEXIST)
2787 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2788 BUG_ON(ret == -EEXIST);
2789 set_balance_control(bctl);
2791 BUG_ON(ret != -EEXIST);
2792 spin_lock(&fs_info->balance_lock);
2793 update_balance_args(bctl);
2794 spin_unlock(&fs_info->balance_lock);
2797 atomic_inc(&fs_info->balance_running);
2798 mutex_unlock(&fs_info->balance_mutex);
2800 ret = __btrfs_balance(fs_info);
2802 mutex_lock(&fs_info->balance_mutex);
2803 atomic_dec(&fs_info->balance_running);
2806 memset(bargs, 0, sizeof(*bargs));
2807 update_ioctl_balance_args(fs_info, 0, bargs);
2810 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2811 balance_need_close(fs_info)) {
2812 __cancel_balance(fs_info);
2815 wake_up(&fs_info->balance_wait_q);
2819 if (bctl->flags & BTRFS_BALANCE_RESUME)
2820 __cancel_balance(fs_info);
2826 static int balance_kthread(void *data)
2828 struct btrfs_balance_control *bctl =
2829 (struct btrfs_balance_control *)data;
2830 struct btrfs_fs_info *fs_info = bctl->fs_info;
2833 mutex_lock(&fs_info->volume_mutex);
2834 mutex_lock(&fs_info->balance_mutex);
2836 set_balance_control(bctl);
2838 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2839 printk(KERN_INFO "btrfs: force skipping balance\n");
2841 printk(KERN_INFO "btrfs: continuing balance\n");
2842 ret = btrfs_balance(bctl, NULL);
2845 mutex_unlock(&fs_info->balance_mutex);
2846 mutex_unlock(&fs_info->volume_mutex);
2850 int btrfs_recover_balance(struct btrfs_root *tree_root)
2852 struct task_struct *tsk;
2853 struct btrfs_balance_control *bctl;
2854 struct btrfs_balance_item *item;
2855 struct btrfs_disk_balance_args disk_bargs;
2856 struct btrfs_path *path;
2857 struct extent_buffer *leaf;
2858 struct btrfs_key key;
2861 path = btrfs_alloc_path();
2865 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2871 key.objectid = BTRFS_BALANCE_OBJECTID;
2872 key.type = BTRFS_BALANCE_ITEM_KEY;
2875 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2878 if (ret > 0) { /* ret = -ENOENT; */
2883 leaf = path->nodes[0];
2884 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2886 bctl->fs_info = tree_root->fs_info;
2887 bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME;
2889 btrfs_balance_data(leaf, item, &disk_bargs);
2890 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2891 btrfs_balance_meta(leaf, item, &disk_bargs);
2892 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2893 btrfs_balance_sys(leaf, item, &disk_bargs);
2894 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2896 tsk = kthread_run(balance_kthread, bctl, "btrfs-balance");
2905 btrfs_free_path(path);
2909 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2913 mutex_lock(&fs_info->balance_mutex);
2914 if (!fs_info->balance_ctl) {
2915 mutex_unlock(&fs_info->balance_mutex);
2919 if (atomic_read(&fs_info->balance_running)) {
2920 atomic_inc(&fs_info->balance_pause_req);
2921 mutex_unlock(&fs_info->balance_mutex);
2923 wait_event(fs_info->balance_wait_q,
2924 atomic_read(&fs_info->balance_running) == 0);
2926 mutex_lock(&fs_info->balance_mutex);
2927 /* we are good with balance_ctl ripped off from under us */
2928 BUG_ON(atomic_read(&fs_info->balance_running));
2929 atomic_dec(&fs_info->balance_pause_req);
2934 mutex_unlock(&fs_info->balance_mutex);
2938 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2940 mutex_lock(&fs_info->balance_mutex);
2941 if (!fs_info->balance_ctl) {
2942 mutex_unlock(&fs_info->balance_mutex);
2946 atomic_inc(&fs_info->balance_cancel_req);
2948 * if we are running just wait and return, balance item is
2949 * deleted in btrfs_balance in this case
2951 if (atomic_read(&fs_info->balance_running)) {
2952 mutex_unlock(&fs_info->balance_mutex);
2953 wait_event(fs_info->balance_wait_q,
2954 atomic_read(&fs_info->balance_running) == 0);
2955 mutex_lock(&fs_info->balance_mutex);
2957 /* __cancel_balance needs volume_mutex */
2958 mutex_unlock(&fs_info->balance_mutex);
2959 mutex_lock(&fs_info->volume_mutex);
2960 mutex_lock(&fs_info->balance_mutex);
2962 if (fs_info->balance_ctl)
2963 __cancel_balance(fs_info);
2965 mutex_unlock(&fs_info->volume_mutex);
2968 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
2969 atomic_dec(&fs_info->balance_cancel_req);
2970 mutex_unlock(&fs_info->balance_mutex);
2975 * shrinking a device means finding all of the device extents past
2976 * the new size, and then following the back refs to the chunks.
2977 * The chunk relocation code actually frees the device extent
2979 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2981 struct btrfs_trans_handle *trans;
2982 struct btrfs_root *root = device->dev_root;
2983 struct btrfs_dev_extent *dev_extent = NULL;
2984 struct btrfs_path *path;
2992 bool retried = false;
2993 struct extent_buffer *l;
2994 struct btrfs_key key;
2995 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2996 u64 old_total = btrfs_super_total_bytes(super_copy);
2997 u64 old_size = device->total_bytes;
2998 u64 diff = device->total_bytes - new_size;
3000 if (new_size >= device->total_bytes)
3003 path = btrfs_alloc_path();
3011 device->total_bytes = new_size;
3012 if (device->writeable) {
3013 device->fs_devices->total_rw_bytes -= diff;
3014 spin_lock(&root->fs_info->free_chunk_lock);
3015 root->fs_info->free_chunk_space -= diff;
3016 spin_unlock(&root->fs_info->free_chunk_lock);
3018 unlock_chunks(root);
3021 key.objectid = device->devid;
3022 key.offset = (u64)-1;
3023 key.type = BTRFS_DEV_EXTENT_KEY;
3026 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3030 ret = btrfs_previous_item(root, path, 0, key.type);
3035 btrfs_release_path(path);
3040 slot = path->slots[0];
3041 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3043 if (key.objectid != device->devid) {
3044 btrfs_release_path(path);
3048 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3049 length = btrfs_dev_extent_length(l, dev_extent);
3051 if (key.offset + length <= new_size) {
3052 btrfs_release_path(path);
3056 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3057 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3058 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3059 btrfs_release_path(path);
3061 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3063 if (ret && ret != -ENOSPC)
3067 } while (key.offset-- > 0);
3069 if (failed && !retried) {
3073 } else if (failed && retried) {
3077 device->total_bytes = old_size;
3078 if (device->writeable)
3079 device->fs_devices->total_rw_bytes += diff;
3080 spin_lock(&root->fs_info->free_chunk_lock);
3081 root->fs_info->free_chunk_space += diff;
3082 spin_unlock(&root->fs_info->free_chunk_lock);
3083 unlock_chunks(root);
3087 /* Shrinking succeeded, else we would be at "done". */
3088 trans = btrfs_start_transaction(root, 0);
3089 if (IS_ERR(trans)) {
3090 ret = PTR_ERR(trans);
3096 device->disk_total_bytes = new_size;
3097 /* Now btrfs_update_device() will change the on-disk size. */
3098 ret = btrfs_update_device(trans, device);
3100 unlock_chunks(root);
3101 btrfs_end_transaction(trans, root);
3104 WARN_ON(diff > old_total);
3105 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3106 unlock_chunks(root);
3107 btrfs_end_transaction(trans, root);
3109 btrfs_free_path(path);
3113 static int btrfs_add_system_chunk(struct btrfs_root *root,
3114 struct btrfs_key *key,
3115 struct btrfs_chunk *chunk, int item_size)
3117 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3118 struct btrfs_disk_key disk_key;
3122 array_size = btrfs_super_sys_array_size(super_copy);
3123 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3126 ptr = super_copy->sys_chunk_array + array_size;
3127 btrfs_cpu_key_to_disk(&disk_key, key);
3128 memcpy(ptr, &disk_key, sizeof(disk_key));
3129 ptr += sizeof(disk_key);
3130 memcpy(ptr, chunk, item_size);
3131 item_size += sizeof(disk_key);
3132 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3137 * sort the devices in descending order by max_avail, total_avail
3139 static int btrfs_cmp_device_info(const void *a, const void *b)
3141 const struct btrfs_device_info *di_a = a;
3142 const struct btrfs_device_info *di_b = b;
3144 if (di_a->max_avail > di_b->max_avail)
3146 if (di_a->max_avail < di_b->max_avail)
3148 if (di_a->total_avail > di_b->total_avail)
3150 if (di_a->total_avail < di_b->total_avail)
3155 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3156 struct btrfs_root *extent_root,
3157 struct map_lookup **map_ret,
3158 u64 *num_bytes_out, u64 *stripe_size_out,
3159 u64 start, u64 type)
3161 struct btrfs_fs_info *info = extent_root->fs_info;
3162 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3163 struct list_head *cur;
3164 struct map_lookup *map = NULL;
3165 struct extent_map_tree *em_tree;
3166 struct extent_map *em;
3167 struct btrfs_device_info *devices_info = NULL;
3169 int num_stripes; /* total number of stripes to allocate */
3170 int sub_stripes; /* sub_stripes info for map */
3171 int dev_stripes; /* stripes per dev */
3172 int devs_max; /* max devs to use */
3173 int devs_min; /* min devs needed */
3174 int devs_increment; /* ndevs has to be a multiple of this */
3175 int ncopies; /* how many copies to data has */
3177 u64 max_stripe_size;
3185 BUG_ON(!alloc_profile_is_valid(type, 0));
3187 if (list_empty(&fs_devices->alloc_list))
3194 devs_max = 0; /* 0 == as many as possible */
3198 * define the properties of each RAID type.
3199 * FIXME: move this to a global table and use it in all RAID
3202 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3206 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3208 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3213 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3222 if (type & BTRFS_BLOCK_GROUP_DATA) {
3223 max_stripe_size = 1024 * 1024 * 1024;
3224 max_chunk_size = 10 * max_stripe_size;
3225 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3226 /* for larger filesystems, use larger metadata chunks */
3227 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3228 max_stripe_size = 1024 * 1024 * 1024;
3230 max_stripe_size = 256 * 1024 * 1024;
3231 max_chunk_size = max_stripe_size;
3232 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3233 max_stripe_size = 32 * 1024 * 1024;
3234 max_chunk_size = 2 * max_stripe_size;
3236 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3241 /* we don't want a chunk larger than 10% of writeable space */
3242 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3245 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3250 cur = fs_devices->alloc_list.next;
3253 * in the first pass through the devices list, we gather information
3254 * about the available holes on each device.
3257 while (cur != &fs_devices->alloc_list) {
3258 struct btrfs_device *device;
3262 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3266 if (!device->writeable) {
3268 "btrfs: read-only device in alloc_list\n");
3273 if (!device->in_fs_metadata)
3276 if (device->total_bytes > device->bytes_used)
3277 total_avail = device->total_bytes - device->bytes_used;
3281 /* If there is no space on this device, skip it. */
3282 if (total_avail == 0)
3285 ret = find_free_dev_extent(device,
3286 max_stripe_size * dev_stripes,
3287 &dev_offset, &max_avail);
3288 if (ret && ret != -ENOSPC)
3292 max_avail = max_stripe_size * dev_stripes;
3294 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3297 devices_info[ndevs].dev_offset = dev_offset;
3298 devices_info[ndevs].max_avail = max_avail;
3299 devices_info[ndevs].total_avail = total_avail;
3300 devices_info[ndevs].dev = device;
3305 * now sort the devices by hole size / available space
3307 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3308 btrfs_cmp_device_info, NULL);
3310 /* round down to number of usable stripes */
3311 ndevs -= ndevs % devs_increment;
3313 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3318 if (devs_max && ndevs > devs_max)
3321 * the primary goal is to maximize the number of stripes, so use as many
3322 * devices as possible, even if the stripes are not maximum sized.
3324 stripe_size = devices_info[ndevs-1].max_avail;
3325 num_stripes = ndevs * dev_stripes;
3327 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
3328 stripe_size = max_chunk_size * ncopies;
3329 do_div(stripe_size, num_stripes);
3332 do_div(stripe_size, dev_stripes);
3333 do_div(stripe_size, BTRFS_STRIPE_LEN);
3334 stripe_size *= BTRFS_STRIPE_LEN;
3336 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3341 map->num_stripes = num_stripes;
3343 for (i = 0; i < ndevs; ++i) {
3344 for (j = 0; j < dev_stripes; ++j) {
3345 int s = i * dev_stripes + j;
3346 map->stripes[s].dev = devices_info[i].dev;
3347 map->stripes[s].physical = devices_info[i].dev_offset +
3351 map->sector_size = extent_root->sectorsize;
3352 map->stripe_len = BTRFS_STRIPE_LEN;
3353 map->io_align = BTRFS_STRIPE_LEN;
3354 map->io_width = BTRFS_STRIPE_LEN;
3356 map->sub_stripes = sub_stripes;
3359 num_bytes = stripe_size * (num_stripes / ncopies);
3361 *stripe_size_out = stripe_size;
3362 *num_bytes_out = num_bytes;
3364 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3366 em = alloc_extent_map();
3371 em->bdev = (struct block_device *)map;
3373 em->len = num_bytes;
3374 em->block_start = 0;
3375 em->block_len = em->len;
3377 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3378 write_lock(&em_tree->lock);
3379 ret = add_extent_mapping(em_tree, em);
3380 write_unlock(&em_tree->lock);
3381 free_extent_map(em);
3385 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3386 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3391 for (i = 0; i < map->num_stripes; ++i) {
3392 struct btrfs_device *device;
3395 device = map->stripes[i].dev;
3396 dev_offset = map->stripes[i].physical;
3398 ret = btrfs_alloc_dev_extent(trans, device,
3399 info->chunk_root->root_key.objectid,
3400 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3401 start, dev_offset, stripe_size);
3403 btrfs_abort_transaction(trans, extent_root, ret);
3408 kfree(devices_info);
3413 kfree(devices_info);
3417 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3418 struct btrfs_root *extent_root,
3419 struct map_lookup *map, u64 chunk_offset,
3420 u64 chunk_size, u64 stripe_size)
3423 struct btrfs_key key;
3424 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3425 struct btrfs_device *device;
3426 struct btrfs_chunk *chunk;
3427 struct btrfs_stripe *stripe;
3428 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3432 chunk = kzalloc(item_size, GFP_NOFS);
3437 while (index < map->num_stripes) {
3438 device = map->stripes[index].dev;
3439 device->bytes_used += stripe_size;
3440 ret = btrfs_update_device(trans, device);
3446 spin_lock(&extent_root->fs_info->free_chunk_lock);
3447 extent_root->fs_info->free_chunk_space -= (stripe_size *
3449 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3452 stripe = &chunk->stripe;
3453 while (index < map->num_stripes) {
3454 device = map->stripes[index].dev;
3455 dev_offset = map->stripes[index].physical;
3457 btrfs_set_stack_stripe_devid(stripe, device->devid);
3458 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3459 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3464 btrfs_set_stack_chunk_length(chunk, chunk_size);
3465 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3466 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3467 btrfs_set_stack_chunk_type(chunk, map->type);
3468 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3469 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3470 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3471 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3472 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3474 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3475 key.type = BTRFS_CHUNK_ITEM_KEY;
3476 key.offset = chunk_offset;
3478 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3480 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3482 * TODO: Cleanup of inserted chunk root in case of
3485 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3495 * Chunk allocation falls into two parts. The first part does works
3496 * that make the new allocated chunk useable, but not do any operation
3497 * that modifies the chunk tree. The second part does the works that
3498 * require modifying the chunk tree. This division is important for the
3499 * bootstrap process of adding storage to a seed btrfs.
3501 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3502 struct btrfs_root *extent_root, u64 type)
3507 struct map_lookup *map;
3508 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3511 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3516 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3517 &stripe_size, chunk_offset, type);
3521 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3522 chunk_size, stripe_size);
3528 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3529 struct btrfs_root *root,
3530 struct btrfs_device *device)
3533 u64 sys_chunk_offset;
3537 u64 sys_stripe_size;
3539 struct map_lookup *map;
3540 struct map_lookup *sys_map;
3541 struct btrfs_fs_info *fs_info = root->fs_info;
3542 struct btrfs_root *extent_root = fs_info->extent_root;
3545 ret = find_next_chunk(fs_info->chunk_root,
3546 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3550 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3551 fs_info->avail_metadata_alloc_bits;
3552 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3554 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3555 &stripe_size, chunk_offset, alloc_profile);
3559 sys_chunk_offset = chunk_offset + chunk_size;
3561 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3562 fs_info->avail_system_alloc_bits;
3563 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3565 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3566 &sys_chunk_size, &sys_stripe_size,
3567 sys_chunk_offset, alloc_profile);
3571 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3576 * Modifying chunk tree needs allocating new blocks from both
3577 * system block group and metadata block group. So we only can
3578 * do operations require modifying the chunk tree after both
3579 * block groups were created.
3581 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3582 chunk_size, stripe_size);
3586 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3587 sys_chunk_offset, sys_chunk_size,
3595 btrfs_abort_transaction(trans, root, ret);
3599 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3601 struct extent_map *em;
3602 struct map_lookup *map;
3603 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3607 read_lock(&map_tree->map_tree.lock);
3608 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3609 read_unlock(&map_tree->map_tree.lock);
3613 if (btrfs_test_opt(root, DEGRADED)) {
3614 free_extent_map(em);
3618 map = (struct map_lookup *)em->bdev;
3619 for (i = 0; i < map->num_stripes; i++) {
3620 if (!map->stripes[i].dev->writeable) {
3625 free_extent_map(em);
3629 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3631 extent_map_tree_init(&tree->map_tree);
3634 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3636 struct extent_map *em;
3639 write_lock(&tree->map_tree.lock);
3640 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3642 remove_extent_mapping(&tree->map_tree, em);
3643 write_unlock(&tree->map_tree.lock);
3648 free_extent_map(em);
3649 /* once for the tree */
3650 free_extent_map(em);
3654 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3656 struct extent_map *em;
3657 struct map_lookup *map;
3658 struct extent_map_tree *em_tree = &map_tree->map_tree;
3661 read_lock(&em_tree->lock);
3662 em = lookup_extent_mapping(em_tree, logical, len);
3663 read_unlock(&em_tree->lock);
3666 BUG_ON(em->start > logical || em->start + em->len < logical);
3667 map = (struct map_lookup *)em->bdev;
3668 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3669 ret = map->num_stripes;
3670 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3671 ret = map->sub_stripes;
3674 free_extent_map(em);
3678 static int find_live_mirror(struct map_lookup *map, int first, int num,
3682 if (map->stripes[optimal].dev->bdev)
3684 for (i = first; i < first + num; i++) {
3685 if (map->stripes[i].dev->bdev)
3688 /* we couldn't find one that doesn't fail. Just return something
3689 * and the io error handling code will clean up eventually
3694 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3695 u64 logical, u64 *length,
3696 struct btrfs_bio **bbio_ret,
3699 struct extent_map *em;
3700 struct map_lookup *map;
3701 struct extent_map_tree *em_tree = &map_tree->map_tree;
3704 u64 stripe_end_offset;
3713 struct btrfs_bio *bbio = NULL;
3715 read_lock(&em_tree->lock);
3716 em = lookup_extent_mapping(em_tree, logical, *length);
3717 read_unlock(&em_tree->lock);
3720 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3721 (unsigned long long)logical,
3722 (unsigned long long)*length);
3726 BUG_ON(em->start > logical || em->start + em->len < logical);
3727 map = (struct map_lookup *)em->bdev;
3728 offset = logical - em->start;
3730 if (mirror_num > map->num_stripes)
3735 * stripe_nr counts the total number of stripes we have to stride
3736 * to get to this block
3738 do_div(stripe_nr, map->stripe_len);
3740 stripe_offset = stripe_nr * map->stripe_len;
3741 BUG_ON(offset < stripe_offset);
3743 /* stripe_offset is the offset of this block in its stripe*/
3744 stripe_offset = offset - stripe_offset;
3746 if (rw & REQ_DISCARD)
3747 *length = min_t(u64, em->len - offset, *length);
3748 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3749 /* we limit the length of each bio to what fits in a stripe */
3750 *length = min_t(u64, em->len - offset,
3751 map->stripe_len - stripe_offset);
3753 *length = em->len - offset;
3761 stripe_nr_orig = stripe_nr;
3762 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3763 (~(map->stripe_len - 1));
3764 do_div(stripe_nr_end, map->stripe_len);
3765 stripe_end_offset = stripe_nr_end * map->stripe_len -
3767 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3768 if (rw & REQ_DISCARD)
3769 num_stripes = min_t(u64, map->num_stripes,
3770 stripe_nr_end - stripe_nr_orig);
3771 stripe_index = do_div(stripe_nr, map->num_stripes);
3772 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3773 if (rw & (REQ_WRITE | REQ_DISCARD))
3774 num_stripes = map->num_stripes;
3775 else if (mirror_num)
3776 stripe_index = mirror_num - 1;
3778 stripe_index = find_live_mirror(map, 0,
3780 current->pid % map->num_stripes);
3781 mirror_num = stripe_index + 1;
3784 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3785 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3786 num_stripes = map->num_stripes;
3787 } else if (mirror_num) {
3788 stripe_index = mirror_num - 1;
3793 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3794 int factor = map->num_stripes / map->sub_stripes;
3796 stripe_index = do_div(stripe_nr, factor);
3797 stripe_index *= map->sub_stripes;
3800 num_stripes = map->sub_stripes;
3801 else if (rw & REQ_DISCARD)
3802 num_stripes = min_t(u64, map->sub_stripes *
3803 (stripe_nr_end - stripe_nr_orig),
3805 else if (mirror_num)
3806 stripe_index += mirror_num - 1;
3808 stripe_index = find_live_mirror(map, stripe_index,
3809 map->sub_stripes, stripe_index +
3810 current->pid % map->sub_stripes);
3811 mirror_num = stripe_index + 1;
3815 * after this do_div call, stripe_nr is the number of stripes
3816 * on this device we have to walk to find the data, and
3817 * stripe_index is the number of our device in the stripe array
3819 stripe_index = do_div(stripe_nr, map->num_stripes);
3820 mirror_num = stripe_index + 1;
3822 BUG_ON(stripe_index >= map->num_stripes);
3824 bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3829 atomic_set(&bbio->error, 0);
3831 if (rw & REQ_DISCARD) {
3833 int sub_stripes = 0;
3834 u64 stripes_per_dev = 0;
3835 u32 remaining_stripes = 0;
3838 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3839 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3842 sub_stripes = map->sub_stripes;
3844 factor = map->num_stripes / sub_stripes;
3845 stripes_per_dev = div_u64_rem(stripe_nr_end -
3848 &remaining_stripes);
3851 for (i = 0; i < num_stripes; i++) {
3852 bbio->stripes[i].physical =
3853 map->stripes[stripe_index].physical +
3854 stripe_offset + stripe_nr * map->stripe_len;
3855 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3857 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3858 BTRFS_BLOCK_GROUP_RAID10)) {
3859 bbio->stripes[i].length = stripes_per_dev *
3861 if (i / sub_stripes < remaining_stripes)
3862 bbio->stripes[i].length +=
3864 if (i < sub_stripes)
3865 bbio->stripes[i].length -=
3867 if ((i / sub_stripes + 1) %
3868 sub_stripes == remaining_stripes)
3869 bbio->stripes[i].length -=
3871 if (i == sub_stripes - 1)
3874 bbio->stripes[i].length = *length;
3877 if (stripe_index == map->num_stripes) {
3878 /* This could only happen for RAID0/10 */
3884 for (i = 0; i < num_stripes; i++) {
3885 bbio->stripes[i].physical =
3886 map->stripes[stripe_index].physical +
3888 stripe_nr * map->stripe_len;
3889 bbio->stripes[i].dev =
3890 map->stripes[stripe_index].dev;
3895 if (rw & REQ_WRITE) {
3896 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3897 BTRFS_BLOCK_GROUP_RAID10 |
3898 BTRFS_BLOCK_GROUP_DUP)) {
3904 bbio->num_stripes = num_stripes;
3905 bbio->max_errors = max_errors;
3906 bbio->mirror_num = mirror_num;
3908 free_extent_map(em);
3912 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3913 u64 logical, u64 *length,
3914 struct btrfs_bio **bbio_ret, int mirror_num)
3916 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3920 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3921 u64 chunk_start, u64 physical, u64 devid,
3922 u64 **logical, int *naddrs, int *stripe_len)
3924 struct extent_map_tree *em_tree = &map_tree->map_tree;
3925 struct extent_map *em;
3926 struct map_lookup *map;
3933 read_lock(&em_tree->lock);
3934 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3935 read_unlock(&em_tree->lock);
3937 BUG_ON(!em || em->start != chunk_start);
3938 map = (struct map_lookup *)em->bdev;
3941 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3942 do_div(length, map->num_stripes / map->sub_stripes);
3943 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3944 do_div(length, map->num_stripes);
3946 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3947 BUG_ON(!buf); /* -ENOMEM */
3949 for (i = 0; i < map->num_stripes; i++) {
3950 if (devid && map->stripes[i].dev->devid != devid)
3952 if (map->stripes[i].physical > physical ||
3953 map->stripes[i].physical + length <= physical)
3956 stripe_nr = physical - map->stripes[i].physical;
3957 do_div(stripe_nr, map->stripe_len);
3959 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3960 stripe_nr = stripe_nr * map->num_stripes + i;
3961 do_div(stripe_nr, map->sub_stripes);
3962 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3963 stripe_nr = stripe_nr * map->num_stripes + i;
3965 bytenr = chunk_start + stripe_nr * map->stripe_len;
3966 WARN_ON(nr >= map->num_stripes);
3967 for (j = 0; j < nr; j++) {
3968 if (buf[j] == bytenr)
3972 WARN_ON(nr >= map->num_stripes);
3979 *stripe_len = map->stripe_len;
3981 free_extent_map(em);
3985 static void btrfs_end_bio(struct bio *bio, int err)
3987 struct btrfs_bio *bbio = bio->bi_private;
3988 int is_orig_bio = 0;
3991 atomic_inc(&bbio->error);
3993 if (bio == bbio->orig_bio)
3996 if (atomic_dec_and_test(&bbio->stripes_pending)) {
3999 bio = bbio->orig_bio;
4001 bio->bi_private = bbio->private;
4002 bio->bi_end_io = bbio->end_io;
4003 bio->bi_bdev = (struct block_device *)
4004 (unsigned long)bbio->mirror_num;
4005 /* only send an error to the higher layers if it is
4006 * beyond the tolerance of the multi-bio
4008 if (atomic_read(&bbio->error) > bbio->max_errors) {
4012 * this bio is actually up to date, we didn't
4013 * go over the max number of errors
4015 set_bit(BIO_UPTODATE, &bio->bi_flags);
4020 bio_endio(bio, err);
4021 } else if (!is_orig_bio) {
4026 struct async_sched {
4029 struct btrfs_fs_info *info;
4030 struct btrfs_work work;
4034 * see run_scheduled_bios for a description of why bios are collected for
4037 * This will add one bio to the pending list for a device and make sure
4038 * the work struct is scheduled.
4040 static noinline void schedule_bio(struct btrfs_root *root,
4041 struct btrfs_device *device,
4042 int rw, struct bio *bio)
4044 int should_queue = 1;
4045 struct btrfs_pending_bios *pending_bios;
4047 /* don't bother with additional async steps for reads, right now */
4048 if (!(rw & REQ_WRITE)) {
4050 btrfsic_submit_bio(rw, bio);
4056 * nr_async_bios allows us to reliably return congestion to the
4057 * higher layers. Otherwise, the async bio makes it appear we have
4058 * made progress against dirty pages when we've really just put it
4059 * on a queue for later
4061 atomic_inc(&root->fs_info->nr_async_bios);
4062 WARN_ON(bio->bi_next);
4063 bio->bi_next = NULL;
4066 spin_lock(&device->io_lock);
4067 if (bio->bi_rw & REQ_SYNC)
4068 pending_bios = &device->pending_sync_bios;
4070 pending_bios = &device->pending_bios;
4072 if (pending_bios->tail)
4073 pending_bios->tail->bi_next = bio;
4075 pending_bios->tail = bio;
4076 if (!pending_bios->head)
4077 pending_bios->head = bio;
4078 if (device->running_pending)
4081 spin_unlock(&device->io_lock);
4084 btrfs_queue_worker(&root->fs_info->submit_workers,
4088 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4089 int mirror_num, int async_submit)
4091 struct btrfs_mapping_tree *map_tree;
4092 struct btrfs_device *dev;
4093 struct bio *first_bio = bio;
4094 u64 logical = (u64)bio->bi_sector << 9;
4100 struct btrfs_bio *bbio = NULL;
4102 length = bio->bi_size;
4103 map_tree = &root->fs_info->mapping_tree;
4104 map_length = length;
4106 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4108 if (ret) /* -ENOMEM */
4111 total_devs = bbio->num_stripes;
4112 if (map_length < length) {
4113 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4114 "len %llu\n", (unsigned long long)logical,
4115 (unsigned long long)length,
4116 (unsigned long long)map_length);
4120 bbio->orig_bio = first_bio;
4121 bbio->private = first_bio->bi_private;
4122 bbio->end_io = first_bio->bi_end_io;
4123 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4125 while (dev_nr < total_devs) {
4126 if (dev_nr < total_devs - 1) {
4127 bio = bio_clone(first_bio, GFP_NOFS);
4128 BUG_ON(!bio); /* -ENOMEM */
4132 bio->bi_private = bbio;
4133 bio->bi_end_io = btrfs_end_bio;
4134 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4135 dev = bbio->stripes[dev_nr].dev;
4136 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4137 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4138 "(%s id %llu), size=%u\n", rw,
4139 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4140 dev->name, dev->devid, bio->bi_size);
4141 bio->bi_bdev = dev->bdev;
4143 schedule_bio(root, dev, rw, bio);
4145 btrfsic_submit_bio(rw, bio);
4147 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4148 bio->bi_sector = logical >> 9;
4149 bio_endio(bio, -EIO);
4156 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4159 struct btrfs_device *device;
4160 struct btrfs_fs_devices *cur_devices;
4162 cur_devices = root->fs_info->fs_devices;
4163 while (cur_devices) {
4165 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4166 device = __find_device(&cur_devices->devices,
4171 cur_devices = cur_devices->seed;
4176 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4177 u64 devid, u8 *dev_uuid)
4179 struct btrfs_device *device;
4180 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4182 device = kzalloc(sizeof(*device), GFP_NOFS);
4185 list_add(&device->dev_list,
4186 &fs_devices->devices);
4187 device->dev_root = root->fs_info->dev_root;
4188 device->devid = devid;
4189 device->work.func = pending_bios_fn;
4190 device->fs_devices = fs_devices;
4191 device->missing = 1;
4192 fs_devices->num_devices++;
4193 fs_devices->missing_devices++;
4194 spin_lock_init(&device->io_lock);
4195 INIT_LIST_HEAD(&device->dev_alloc_list);
4196 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4200 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4201 struct extent_buffer *leaf,
4202 struct btrfs_chunk *chunk)
4204 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4205 struct map_lookup *map;
4206 struct extent_map *em;
4210 u8 uuid[BTRFS_UUID_SIZE];
4215 logical = key->offset;
4216 length = btrfs_chunk_length(leaf, chunk);
4218 read_lock(&map_tree->map_tree.lock);
4219 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4220 read_unlock(&map_tree->map_tree.lock);
4222 /* already mapped? */
4223 if (em && em->start <= logical && em->start + em->len > logical) {
4224 free_extent_map(em);
4227 free_extent_map(em);
4230 em = alloc_extent_map();
4233 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4234 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4236 free_extent_map(em);
4240 em->bdev = (struct block_device *)map;
4241 em->start = logical;
4243 em->block_start = 0;
4244 em->block_len = em->len;
4246 map->num_stripes = num_stripes;
4247 map->io_width = btrfs_chunk_io_width(leaf, chunk);
4248 map->io_align = btrfs_chunk_io_align(leaf, chunk);
4249 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4250 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4251 map->type = btrfs_chunk_type(leaf, chunk);
4252 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4253 for (i = 0; i < num_stripes; i++) {
4254 map->stripes[i].physical =
4255 btrfs_stripe_offset_nr(leaf, chunk, i);
4256 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4257 read_extent_buffer(leaf, uuid, (unsigned long)
4258 btrfs_stripe_dev_uuid_nr(chunk, i),
4260 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4262 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4264 free_extent_map(em);
4267 if (!map->stripes[i].dev) {
4268 map->stripes[i].dev =
4269 add_missing_dev(root, devid, uuid);
4270 if (!map->stripes[i].dev) {
4272 free_extent_map(em);
4276 map->stripes[i].dev->in_fs_metadata = 1;
4279 write_lock(&map_tree->map_tree.lock);
4280 ret = add_extent_mapping(&map_tree->map_tree, em);
4281 write_unlock(&map_tree->map_tree.lock);
4282 BUG_ON(ret); /* Tree corruption */
4283 free_extent_map(em);
4288 static void fill_device_from_item(struct extent_buffer *leaf,
4289 struct btrfs_dev_item *dev_item,
4290 struct btrfs_device *device)
4294 device->devid = btrfs_device_id(leaf, dev_item);
4295 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4296 device->total_bytes = device->disk_total_bytes;
4297 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4298 device->type = btrfs_device_type(leaf, dev_item);
4299 device->io_align = btrfs_device_io_align(leaf, dev_item);
4300 device->io_width = btrfs_device_io_width(leaf, dev_item);
4301 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4303 ptr = (unsigned long)btrfs_device_uuid(dev_item);
4304 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4307 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4309 struct btrfs_fs_devices *fs_devices;
4312 BUG_ON(!mutex_is_locked(&uuid_mutex));
4314 fs_devices = root->fs_info->fs_devices->seed;
4315 while (fs_devices) {
4316 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4320 fs_devices = fs_devices->seed;
4323 fs_devices = find_fsid(fsid);
4329 fs_devices = clone_fs_devices(fs_devices);
4330 if (IS_ERR(fs_devices)) {
4331 ret = PTR_ERR(fs_devices);
4335 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4336 root->fs_info->bdev_holder);
4340 if (!fs_devices->seeding) {
4341 __btrfs_close_devices(fs_devices);
4342 free_fs_devices(fs_devices);
4347 fs_devices->seed = root->fs_info->fs_devices->seed;
4348 root->fs_info->fs_devices->seed = fs_devices;
4353 static int read_one_dev(struct btrfs_root *root,
4354 struct extent_buffer *leaf,
4355 struct btrfs_dev_item *dev_item)
4357 struct btrfs_device *device;
4360 u8 fs_uuid[BTRFS_UUID_SIZE];
4361 u8 dev_uuid[BTRFS_UUID_SIZE];
4363 devid = btrfs_device_id(leaf, dev_item);
4364 read_extent_buffer(leaf, dev_uuid,
4365 (unsigned long)btrfs_device_uuid(dev_item),
4367 read_extent_buffer(leaf, fs_uuid,
4368 (unsigned long)btrfs_device_fsid(dev_item),
4371 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4372 ret = open_seed_devices(root, fs_uuid);
4373 if (ret && !btrfs_test_opt(root, DEGRADED))
4377 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4378 if (!device || !device->bdev) {
4379 if (!btrfs_test_opt(root, DEGRADED))
4383 printk(KERN_WARNING "warning devid %llu missing\n",
4384 (unsigned long long)devid);
4385 device = add_missing_dev(root, devid, dev_uuid);
4388 } else if (!device->missing) {
4390 * this happens when a device that was properly setup
4391 * in the device info lists suddenly goes bad.
4392 * device->bdev is NULL, and so we have to set
4393 * device->missing to one here
4395 root->fs_info->fs_devices->missing_devices++;
4396 device->missing = 1;
4400 if (device->fs_devices != root->fs_info->fs_devices) {
4401 BUG_ON(device->writeable);
4402 if (device->generation !=
4403 btrfs_device_generation(leaf, dev_item))
4407 fill_device_from_item(leaf, dev_item, device);
4408 device->dev_root = root->fs_info->dev_root;
4409 device->in_fs_metadata = 1;
4410 if (device->writeable) {
4411 device->fs_devices->total_rw_bytes += device->total_bytes;
4412 spin_lock(&root->fs_info->free_chunk_lock);
4413 root->fs_info->free_chunk_space += device->total_bytes -
4415 spin_unlock(&root->fs_info->free_chunk_lock);
4421 int btrfs_read_sys_array(struct btrfs_root *root)
4423 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4424 struct extent_buffer *sb;
4425 struct btrfs_disk_key *disk_key;
4426 struct btrfs_chunk *chunk;
4428 unsigned long sb_ptr;
4434 struct btrfs_key key;
4436 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4437 BTRFS_SUPER_INFO_SIZE);
4440 btrfs_set_buffer_uptodate(sb);
4441 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4443 * The sb extent buffer is artifical and just used to read the system array.
4444 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4445 * pages up-to-date when the page is larger: extent does not cover the
4446 * whole page and consequently check_page_uptodate does not find all
4447 * the page's extents up-to-date (the hole beyond sb),
4448 * write_extent_buffer then triggers a WARN_ON.
4450 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4451 * but sb spans only this function. Add an explicit SetPageUptodate call
4452 * to silence the warning eg. on PowerPC 64.
4454 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4455 SetPageUptodate(sb->pages[0]);
4457 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4458 array_size = btrfs_super_sys_array_size(super_copy);
4460 ptr = super_copy->sys_chunk_array;
4461 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4464 while (cur < array_size) {
4465 disk_key = (struct btrfs_disk_key *)ptr;
4466 btrfs_disk_key_to_cpu(&key, disk_key);
4468 len = sizeof(*disk_key); ptr += len;
4472 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4473 chunk = (struct btrfs_chunk *)sb_ptr;
4474 ret = read_one_chunk(root, &key, sb, chunk);
4477 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4478 len = btrfs_chunk_item_size(num_stripes);
4487 free_extent_buffer(sb);
4491 int btrfs_read_chunk_tree(struct btrfs_root *root)
4493 struct btrfs_path *path;
4494 struct extent_buffer *leaf;
4495 struct btrfs_key key;
4496 struct btrfs_key found_key;
4500 root = root->fs_info->chunk_root;
4502 path = btrfs_alloc_path();
4506 mutex_lock(&uuid_mutex);
4509 /* first we search for all of the device items, and then we
4510 * read in all of the chunk items. This way we can create chunk
4511 * mappings that reference all of the devices that are afound
4513 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4517 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4521 leaf = path->nodes[0];
4522 slot = path->slots[0];
4523 if (slot >= btrfs_header_nritems(leaf)) {
4524 ret = btrfs_next_leaf(root, path);
4531 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4532 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4533 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4535 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4536 struct btrfs_dev_item *dev_item;
4537 dev_item = btrfs_item_ptr(leaf, slot,
4538 struct btrfs_dev_item);
4539 ret = read_one_dev(root, leaf, dev_item);
4543 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4544 struct btrfs_chunk *chunk;
4545 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4546 ret = read_one_chunk(root, &found_key, leaf, chunk);
4552 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4554 btrfs_release_path(path);
4559 unlock_chunks(root);
4560 mutex_unlock(&uuid_mutex);
4562 btrfs_free_path(path);