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 <asm/div64.h>
29 #include "extent_map.h"
31 #include "transaction.h"
32 #include "print-tree.h"
34 #include "async-thread.h"
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
41 static DEFINE_MUTEX(uuid_mutex);
42 static LIST_HEAD(fs_uuids);
44 static void lock_chunks(struct btrfs_root *root)
46 mutex_lock(&root->fs_info->chunk_mutex);
49 static void unlock_chunks(struct btrfs_root *root)
51 mutex_unlock(&root->fs_info->chunk_mutex);
54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
56 struct btrfs_device *device;
57 WARN_ON(fs_devices->opened);
58 while (!list_empty(&fs_devices->devices)) {
59 device = list_entry(fs_devices->devices.next,
60 struct btrfs_device, dev_list);
61 list_del(&device->dev_list);
68 int btrfs_cleanup_fs_uuids(void)
70 struct btrfs_fs_devices *fs_devices;
72 while (!list_empty(&fs_uuids)) {
73 fs_devices = list_entry(fs_uuids.next,
74 struct btrfs_fs_devices, list);
75 list_del(&fs_devices->list);
76 free_fs_devices(fs_devices);
81 static noinline struct btrfs_device *__find_device(struct list_head *head,
84 struct btrfs_device *dev;
86 list_for_each_entry(dev, head, dev_list) {
87 if (dev->devid == devid &&
88 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
97 struct btrfs_fs_devices *fs_devices;
99 list_for_each_entry(fs_devices, &fs_uuids, list) {
100 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 struct bio *head, struct bio *tail)
110 struct bio *old_head;
112 old_head = pending_bios->head;
113 pending_bios->head = head;
114 if (pending_bios->tail)
115 tail->bi_next = old_head;
117 pending_bios->tail = tail;
121 * we try to collect pending bios for a device so we don't get a large
122 * number of procs sending bios down to the same device. This greatly
123 * improves the schedulers ability to collect and merge the bios.
125 * But, it also turns into a long list of bios to process and that is sure
126 * to eventually make the worker thread block. The solution here is to
127 * make some progress and then put this work struct back at the end of
128 * the list if the block device is congested. This way, multiple devices
129 * can make progress from a single worker thread.
131 static noinline int run_scheduled_bios(struct btrfs_device *device)
134 struct backing_dev_info *bdi;
135 struct btrfs_fs_info *fs_info;
136 struct btrfs_pending_bios *pending_bios;
140 unsigned long num_run;
141 unsigned long batch_run = 0;
143 unsigned long last_waited = 0;
145 int sync_pending = 0;
146 struct blk_plug plug;
149 * this function runs all the bios we've collected for
150 * a particular device. We don't want to wander off to
151 * another device without first sending all of these down.
152 * So, setup a plug here and finish it off before we return
154 blk_start_plug(&plug);
156 bdi = blk_get_backing_dev_info(device->bdev);
157 fs_info = device->dev_root->fs_info;
158 limit = btrfs_async_submit_limit(fs_info);
159 limit = limit * 2 / 3;
162 spin_lock(&device->io_lock);
167 /* take all the bios off the list at once and process them
168 * later on (without the lock held). But, remember the
169 * tail and other pointers so the bios can be properly reinserted
170 * into the list if we hit congestion
172 if (!force_reg && device->pending_sync_bios.head) {
173 pending_bios = &device->pending_sync_bios;
176 pending_bios = &device->pending_bios;
180 pending = pending_bios->head;
181 tail = pending_bios->tail;
182 WARN_ON(pending && !tail);
185 * if pending was null this time around, no bios need processing
186 * at all and we can stop. Otherwise it'll loop back up again
187 * and do an additional check so no bios are missed.
189 * device->running_pending is used to synchronize with the
192 if (device->pending_sync_bios.head == NULL &&
193 device->pending_bios.head == NULL) {
195 device->running_pending = 0;
198 device->running_pending = 1;
201 pending_bios->head = NULL;
202 pending_bios->tail = NULL;
204 spin_unlock(&device->io_lock);
209 /* we want to work on both lists, but do more bios on the
210 * sync list than the regular list
213 pending_bios != &device->pending_sync_bios &&
214 device->pending_sync_bios.head) ||
215 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
216 device->pending_bios.head)) {
217 spin_lock(&device->io_lock);
218 requeue_list(pending_bios, pending, tail);
223 pending = pending->bi_next;
225 atomic_dec(&fs_info->nr_async_bios);
227 if (atomic_read(&fs_info->nr_async_bios) < limit &&
228 waitqueue_active(&fs_info->async_submit_wait))
229 wake_up(&fs_info->async_submit_wait);
231 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
234 * if we're doing the sync list, record that our
235 * plug has some sync requests on it
237 * If we're doing the regular list and there are
238 * sync requests sitting around, unplug before
241 if (pending_bios == &device->pending_sync_bios) {
243 } else if (sync_pending) {
244 blk_finish_plug(&plug);
245 blk_start_plug(&plug);
249 submit_bio(cur->bi_rw, cur);
256 * we made progress, there is more work to do and the bdi
257 * is now congested. Back off and let other work structs
260 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
261 fs_info->fs_devices->open_devices > 1) {
262 struct io_context *ioc;
264 ioc = current->io_context;
267 * the main goal here is that we don't want to
268 * block if we're going to be able to submit
269 * more requests without blocking.
271 * This code does two great things, it pokes into
272 * the elevator code from a filesystem _and_
273 * it makes assumptions about how batching works.
275 if (ioc && ioc->nr_batch_requests > 0 &&
276 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
278 ioc->last_waited == last_waited)) {
280 * we want to go through our batch of
281 * requests and stop. So, we copy out
282 * the ioc->last_waited time and test
283 * against it before looping
285 last_waited = ioc->last_waited;
290 spin_lock(&device->io_lock);
291 requeue_list(pending_bios, pending, tail);
292 device->running_pending = 1;
294 spin_unlock(&device->io_lock);
295 btrfs_requeue_work(&device->work);
304 spin_lock(&device->io_lock);
305 if (device->pending_bios.head || device->pending_sync_bios.head)
307 spin_unlock(&device->io_lock);
310 blk_finish_plug(&plug);
314 static void pending_bios_fn(struct btrfs_work *work)
316 struct btrfs_device *device;
318 device = container_of(work, struct btrfs_device, work);
319 run_scheduled_bios(device);
322 static noinline int device_list_add(const char *path,
323 struct btrfs_super_block *disk_super,
324 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
326 struct btrfs_device *device;
327 struct btrfs_fs_devices *fs_devices;
328 u64 found_transid = btrfs_super_generation(disk_super);
331 fs_devices = find_fsid(disk_super->fsid);
333 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
336 INIT_LIST_HEAD(&fs_devices->devices);
337 INIT_LIST_HEAD(&fs_devices->alloc_list);
338 list_add(&fs_devices->list, &fs_uuids);
339 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
340 fs_devices->latest_devid = devid;
341 fs_devices->latest_trans = found_transid;
342 mutex_init(&fs_devices->device_list_mutex);
345 device = __find_device(&fs_devices->devices, devid,
346 disk_super->dev_item.uuid);
349 if (fs_devices->opened)
352 device = kzalloc(sizeof(*device), GFP_NOFS);
354 /* we can safely leave the fs_devices entry around */
357 device->devid = devid;
358 device->work.func = pending_bios_fn;
359 memcpy(device->uuid, disk_super->dev_item.uuid,
361 spin_lock_init(&device->io_lock);
362 device->name = kstrdup(path, GFP_NOFS);
367 INIT_LIST_HEAD(&device->dev_alloc_list);
369 /* init readahead state */
370 spin_lock_init(&device->reada_lock);
371 device->reada_curr_zone = NULL;
372 atomic_set(&device->reada_in_flight, 0);
373 device->reada_next = 0;
374 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
375 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
377 mutex_lock(&fs_devices->device_list_mutex);
378 list_add_rcu(&device->dev_list, &fs_devices->devices);
379 mutex_unlock(&fs_devices->device_list_mutex);
381 device->fs_devices = fs_devices;
382 fs_devices->num_devices++;
383 } else if (!device->name || strcmp(device->name, path)) {
384 name = kstrdup(path, GFP_NOFS);
389 if (device->missing) {
390 fs_devices->missing_devices--;
395 if (found_transid > fs_devices->latest_trans) {
396 fs_devices->latest_devid = devid;
397 fs_devices->latest_trans = found_transid;
399 *fs_devices_ret = fs_devices;
403 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
405 struct btrfs_fs_devices *fs_devices;
406 struct btrfs_device *device;
407 struct btrfs_device *orig_dev;
409 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
411 return ERR_PTR(-ENOMEM);
413 INIT_LIST_HEAD(&fs_devices->devices);
414 INIT_LIST_HEAD(&fs_devices->alloc_list);
415 INIT_LIST_HEAD(&fs_devices->list);
416 mutex_init(&fs_devices->device_list_mutex);
417 fs_devices->latest_devid = orig->latest_devid;
418 fs_devices->latest_trans = orig->latest_trans;
419 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
421 /* We have held the volume lock, it is safe to get the devices. */
422 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
423 device = kzalloc(sizeof(*device), GFP_NOFS);
427 device->name = kstrdup(orig_dev->name, GFP_NOFS);
433 device->devid = orig_dev->devid;
434 device->work.func = pending_bios_fn;
435 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
436 spin_lock_init(&device->io_lock);
437 INIT_LIST_HEAD(&device->dev_list);
438 INIT_LIST_HEAD(&device->dev_alloc_list);
440 list_add(&device->dev_list, &fs_devices->devices);
441 device->fs_devices = fs_devices;
442 fs_devices->num_devices++;
446 free_fs_devices(fs_devices);
447 return ERR_PTR(-ENOMEM);
450 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
452 struct btrfs_device *device, *next;
454 mutex_lock(&uuid_mutex);
456 /* This is the initialized path, it is safe to release the devices. */
457 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
458 if (device->in_fs_metadata)
462 blkdev_put(device->bdev, device->mode);
464 fs_devices->open_devices--;
466 if (device->writeable) {
467 list_del_init(&device->dev_alloc_list);
468 device->writeable = 0;
469 fs_devices->rw_devices--;
471 list_del_init(&device->dev_list);
472 fs_devices->num_devices--;
477 if (fs_devices->seed) {
478 fs_devices = fs_devices->seed;
482 mutex_unlock(&uuid_mutex);
486 static void __free_device(struct work_struct *work)
488 struct btrfs_device *device;
490 device = container_of(work, struct btrfs_device, rcu_work);
493 blkdev_put(device->bdev, device->mode);
499 static void free_device(struct rcu_head *head)
501 struct btrfs_device *device;
503 device = container_of(head, struct btrfs_device, rcu);
505 INIT_WORK(&device->rcu_work, __free_device);
506 schedule_work(&device->rcu_work);
509 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
511 struct btrfs_device *device;
513 if (--fs_devices->opened > 0)
516 mutex_lock(&fs_devices->device_list_mutex);
517 list_for_each_entry(device, &fs_devices->devices, dev_list) {
518 struct btrfs_device *new_device;
521 fs_devices->open_devices--;
523 if (device->writeable) {
524 list_del_init(&device->dev_alloc_list);
525 fs_devices->rw_devices--;
528 if (device->can_discard)
529 fs_devices->num_can_discard--;
531 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
533 memcpy(new_device, device, sizeof(*new_device));
534 new_device->name = kstrdup(device->name, GFP_NOFS);
535 BUG_ON(device->name && !new_device->name);
536 new_device->bdev = NULL;
537 new_device->writeable = 0;
538 new_device->in_fs_metadata = 0;
539 new_device->can_discard = 0;
540 list_replace_rcu(&device->dev_list, &new_device->dev_list);
542 call_rcu(&device->rcu, free_device);
544 mutex_unlock(&fs_devices->device_list_mutex);
546 WARN_ON(fs_devices->open_devices);
547 WARN_ON(fs_devices->rw_devices);
548 fs_devices->opened = 0;
549 fs_devices->seeding = 0;
554 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
556 struct btrfs_fs_devices *seed_devices = NULL;
559 mutex_lock(&uuid_mutex);
560 ret = __btrfs_close_devices(fs_devices);
561 if (!fs_devices->opened) {
562 seed_devices = fs_devices->seed;
563 fs_devices->seed = NULL;
565 mutex_unlock(&uuid_mutex);
567 while (seed_devices) {
568 fs_devices = seed_devices;
569 seed_devices = fs_devices->seed;
570 __btrfs_close_devices(fs_devices);
571 free_fs_devices(fs_devices);
576 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
577 fmode_t flags, void *holder)
579 struct request_queue *q;
580 struct block_device *bdev;
581 struct list_head *head = &fs_devices->devices;
582 struct btrfs_device *device;
583 struct block_device *latest_bdev = NULL;
584 struct buffer_head *bh;
585 struct btrfs_super_block *disk_super;
586 u64 latest_devid = 0;
587 u64 latest_transid = 0;
594 list_for_each_entry(device, head, dev_list) {
600 bdev = blkdev_get_by_path(device->name, flags, holder);
602 printk(KERN_INFO "open %s failed\n", device->name);
605 set_blocksize(bdev, 4096);
607 bh = btrfs_read_dev_super(bdev);
611 disk_super = (struct btrfs_super_block *)bh->b_data;
612 devid = btrfs_stack_device_id(&disk_super->dev_item);
613 if (devid != device->devid)
616 if (memcmp(device->uuid, disk_super->dev_item.uuid,
620 device->generation = btrfs_super_generation(disk_super);
621 if (!latest_transid || device->generation > latest_transid) {
622 latest_devid = devid;
623 latest_transid = device->generation;
627 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
628 device->writeable = 0;
630 device->writeable = !bdev_read_only(bdev);
634 q = bdev_get_queue(bdev);
635 if (blk_queue_discard(q)) {
636 device->can_discard = 1;
637 fs_devices->num_can_discard++;
641 device->in_fs_metadata = 0;
642 device->mode = flags;
644 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
645 fs_devices->rotating = 1;
647 fs_devices->open_devices++;
648 if (device->writeable) {
649 fs_devices->rw_devices++;
650 list_add(&device->dev_alloc_list,
651 &fs_devices->alloc_list);
659 blkdev_put(bdev, flags);
663 if (fs_devices->open_devices == 0) {
667 fs_devices->seeding = seeding;
668 fs_devices->opened = 1;
669 fs_devices->latest_bdev = latest_bdev;
670 fs_devices->latest_devid = latest_devid;
671 fs_devices->latest_trans = latest_transid;
672 fs_devices->total_rw_bytes = 0;
677 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
678 fmode_t flags, void *holder)
682 mutex_lock(&uuid_mutex);
683 if (fs_devices->opened) {
684 fs_devices->opened++;
687 ret = __btrfs_open_devices(fs_devices, flags, holder);
689 mutex_unlock(&uuid_mutex);
693 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
694 struct btrfs_fs_devices **fs_devices_ret)
696 struct btrfs_super_block *disk_super;
697 struct block_device *bdev;
698 struct buffer_head *bh;
703 mutex_lock(&uuid_mutex);
706 bdev = blkdev_get_by_path(path, flags, holder);
713 ret = set_blocksize(bdev, 4096);
716 bh = btrfs_read_dev_super(bdev);
721 disk_super = (struct btrfs_super_block *)bh->b_data;
722 devid = btrfs_stack_device_id(&disk_super->dev_item);
723 transid = btrfs_super_generation(disk_super);
724 if (disk_super->label[0])
725 printk(KERN_INFO "device label %s ", disk_super->label);
727 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
728 printk(KERN_CONT "devid %llu transid %llu %s\n",
729 (unsigned long long)devid, (unsigned long long)transid, path);
730 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
734 blkdev_put(bdev, flags);
736 mutex_unlock(&uuid_mutex);
740 /* helper to account the used device space in the range */
741 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
742 u64 end, u64 *length)
744 struct btrfs_key key;
745 struct btrfs_root *root = device->dev_root;
746 struct btrfs_dev_extent *dev_extent;
747 struct btrfs_path *path;
751 struct extent_buffer *l;
755 if (start >= device->total_bytes)
758 path = btrfs_alloc_path();
763 key.objectid = device->devid;
765 key.type = BTRFS_DEV_EXTENT_KEY;
767 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
771 ret = btrfs_previous_item(root, path, key.objectid, key.type);
778 slot = path->slots[0];
779 if (slot >= btrfs_header_nritems(l)) {
780 ret = btrfs_next_leaf(root, path);
788 btrfs_item_key_to_cpu(l, &key, slot);
790 if (key.objectid < device->devid)
793 if (key.objectid > device->devid)
796 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
799 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
800 extent_end = key.offset + btrfs_dev_extent_length(l,
802 if (key.offset <= start && extent_end > end) {
803 *length = end - start + 1;
805 } else if (key.offset <= start && extent_end > start)
806 *length += extent_end - start;
807 else if (key.offset > start && extent_end <= end)
808 *length += extent_end - key.offset;
809 else if (key.offset > start && key.offset <= end) {
810 *length += end - key.offset + 1;
812 } else if (key.offset > end)
820 btrfs_free_path(path);
825 * find_free_dev_extent - find free space in the specified device
826 * @trans: transaction handler
827 * @device: the device which we search the free space in
828 * @num_bytes: the size of the free space that we need
829 * @start: store the start of the free space.
830 * @len: the size of the free space. that we find, or the size of the max
831 * free space if we don't find suitable free space
833 * this uses a pretty simple search, the expectation is that it is
834 * called very infrequently and that a given device has a small number
837 * @start is used to store the start of the free space if we find. But if we
838 * don't find suitable free space, it will be used to store the start position
839 * of the max free space.
841 * @len is used to store the size of the free space that we find.
842 * But if we don't find suitable free space, it is used to store the size of
843 * the max free space.
845 int find_free_dev_extent(struct btrfs_trans_handle *trans,
846 struct btrfs_device *device, u64 num_bytes,
847 u64 *start, u64 *len)
849 struct btrfs_key key;
850 struct btrfs_root *root = device->dev_root;
851 struct btrfs_dev_extent *dev_extent;
852 struct btrfs_path *path;
858 u64 search_end = device->total_bytes;
861 struct extent_buffer *l;
863 /* FIXME use last free of some kind */
865 /* we don't want to overwrite the superblock on the drive,
866 * so we make sure to start at an offset of at least 1MB
868 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
870 max_hole_start = search_start;
874 if (search_start >= search_end) {
879 path = btrfs_alloc_path();
886 key.objectid = device->devid;
887 key.offset = search_start;
888 key.type = BTRFS_DEV_EXTENT_KEY;
890 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
894 ret = btrfs_previous_item(root, path, key.objectid, key.type);
901 slot = path->slots[0];
902 if (slot >= btrfs_header_nritems(l)) {
903 ret = btrfs_next_leaf(root, path);
911 btrfs_item_key_to_cpu(l, &key, slot);
913 if (key.objectid < device->devid)
916 if (key.objectid > device->devid)
919 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
922 if (key.offset > search_start) {
923 hole_size = key.offset - search_start;
925 if (hole_size > max_hole_size) {
926 max_hole_start = search_start;
927 max_hole_size = hole_size;
931 * If this free space is greater than which we need,
932 * it must be the max free space that we have found
933 * until now, so max_hole_start must point to the start
934 * of this free space and the length of this free space
935 * is stored in max_hole_size. Thus, we return
936 * max_hole_start and max_hole_size and go back to the
939 if (hole_size >= num_bytes) {
945 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
946 extent_end = key.offset + btrfs_dev_extent_length(l,
948 if (extent_end > search_start)
949 search_start = extent_end;
956 * At this point, search_start should be the end of
957 * allocated dev extents, and when shrinking the device,
958 * search_end may be smaller than search_start.
960 if (search_end > search_start)
961 hole_size = search_end - search_start;
963 if (hole_size > max_hole_size) {
964 max_hole_start = search_start;
965 max_hole_size = hole_size;
969 if (hole_size < num_bytes)
975 btrfs_free_path(path);
977 *start = max_hole_start;
979 *len = max_hole_size;
983 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
984 struct btrfs_device *device,
988 struct btrfs_path *path;
989 struct btrfs_root *root = device->dev_root;
990 struct btrfs_key key;
991 struct btrfs_key found_key;
992 struct extent_buffer *leaf = NULL;
993 struct btrfs_dev_extent *extent = NULL;
995 path = btrfs_alloc_path();
999 key.objectid = device->devid;
1001 key.type = BTRFS_DEV_EXTENT_KEY;
1003 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1005 ret = btrfs_previous_item(root, path, key.objectid,
1006 BTRFS_DEV_EXTENT_KEY);
1009 leaf = path->nodes[0];
1010 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1011 extent = btrfs_item_ptr(leaf, path->slots[0],
1012 struct btrfs_dev_extent);
1013 BUG_ON(found_key.offset > start || found_key.offset +
1014 btrfs_dev_extent_length(leaf, extent) < start);
1016 btrfs_release_path(path);
1018 } else if (ret == 0) {
1019 leaf = path->nodes[0];
1020 extent = btrfs_item_ptr(leaf, path->slots[0],
1021 struct btrfs_dev_extent);
1025 if (device->bytes_used > 0) {
1026 u64 len = btrfs_dev_extent_length(leaf, extent);
1027 device->bytes_used -= len;
1028 spin_lock(&root->fs_info->free_chunk_lock);
1029 root->fs_info->free_chunk_space += len;
1030 spin_unlock(&root->fs_info->free_chunk_lock);
1032 ret = btrfs_del_item(trans, root, path);
1035 btrfs_free_path(path);
1039 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1040 struct btrfs_device *device,
1041 u64 chunk_tree, u64 chunk_objectid,
1042 u64 chunk_offset, u64 start, u64 num_bytes)
1045 struct btrfs_path *path;
1046 struct btrfs_root *root = device->dev_root;
1047 struct btrfs_dev_extent *extent;
1048 struct extent_buffer *leaf;
1049 struct btrfs_key key;
1051 WARN_ON(!device->in_fs_metadata);
1052 path = btrfs_alloc_path();
1056 key.objectid = device->devid;
1058 key.type = BTRFS_DEV_EXTENT_KEY;
1059 ret = btrfs_insert_empty_item(trans, root, path, &key,
1063 leaf = path->nodes[0];
1064 extent = btrfs_item_ptr(leaf, path->slots[0],
1065 struct btrfs_dev_extent);
1066 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1067 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1068 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1070 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1071 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1074 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1075 btrfs_mark_buffer_dirty(leaf);
1076 btrfs_free_path(path);
1080 static noinline int find_next_chunk(struct btrfs_root *root,
1081 u64 objectid, u64 *offset)
1083 struct btrfs_path *path;
1085 struct btrfs_key key;
1086 struct btrfs_chunk *chunk;
1087 struct btrfs_key found_key;
1089 path = btrfs_alloc_path();
1093 key.objectid = objectid;
1094 key.offset = (u64)-1;
1095 key.type = BTRFS_CHUNK_ITEM_KEY;
1097 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1103 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1107 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1109 if (found_key.objectid != objectid)
1112 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1113 struct btrfs_chunk);
1114 *offset = found_key.offset +
1115 btrfs_chunk_length(path->nodes[0], chunk);
1120 btrfs_free_path(path);
1124 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1127 struct btrfs_key key;
1128 struct btrfs_key found_key;
1129 struct btrfs_path *path;
1131 root = root->fs_info->chunk_root;
1133 path = btrfs_alloc_path();
1137 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1138 key.type = BTRFS_DEV_ITEM_KEY;
1139 key.offset = (u64)-1;
1141 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1147 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1148 BTRFS_DEV_ITEM_KEY);
1152 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1154 *objectid = found_key.offset + 1;
1158 btrfs_free_path(path);
1163 * the device information is stored in the chunk root
1164 * the btrfs_device struct should be fully filled in
1166 int btrfs_add_device(struct btrfs_trans_handle *trans,
1167 struct btrfs_root *root,
1168 struct btrfs_device *device)
1171 struct btrfs_path *path;
1172 struct btrfs_dev_item *dev_item;
1173 struct extent_buffer *leaf;
1174 struct btrfs_key key;
1177 root = root->fs_info->chunk_root;
1179 path = btrfs_alloc_path();
1183 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1184 key.type = BTRFS_DEV_ITEM_KEY;
1185 key.offset = device->devid;
1187 ret = btrfs_insert_empty_item(trans, root, path, &key,
1192 leaf = path->nodes[0];
1193 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1195 btrfs_set_device_id(leaf, dev_item, device->devid);
1196 btrfs_set_device_generation(leaf, dev_item, 0);
1197 btrfs_set_device_type(leaf, dev_item, device->type);
1198 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1199 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1200 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1201 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1202 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1203 btrfs_set_device_group(leaf, dev_item, 0);
1204 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1205 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1206 btrfs_set_device_start_offset(leaf, dev_item, 0);
1208 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1209 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1210 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1211 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1212 btrfs_mark_buffer_dirty(leaf);
1216 btrfs_free_path(path);
1220 static int btrfs_rm_dev_item(struct btrfs_root *root,
1221 struct btrfs_device *device)
1224 struct btrfs_path *path;
1225 struct btrfs_key key;
1226 struct btrfs_trans_handle *trans;
1228 root = root->fs_info->chunk_root;
1230 path = btrfs_alloc_path();
1234 trans = btrfs_start_transaction(root, 0);
1235 if (IS_ERR(trans)) {
1236 btrfs_free_path(path);
1237 return PTR_ERR(trans);
1239 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1240 key.type = BTRFS_DEV_ITEM_KEY;
1241 key.offset = device->devid;
1244 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1253 ret = btrfs_del_item(trans, root, path);
1257 btrfs_free_path(path);
1258 unlock_chunks(root);
1259 btrfs_commit_transaction(trans, root);
1263 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1265 struct btrfs_device *device;
1266 struct btrfs_device *next_device;
1267 struct block_device *bdev;
1268 struct buffer_head *bh = NULL;
1269 struct btrfs_super_block *disk_super;
1270 struct btrfs_fs_devices *cur_devices;
1276 bool clear_super = false;
1278 mutex_lock(&uuid_mutex);
1279 mutex_lock(&root->fs_info->volume_mutex);
1281 all_avail = root->fs_info->avail_data_alloc_bits |
1282 root->fs_info->avail_system_alloc_bits |
1283 root->fs_info->avail_metadata_alloc_bits;
1285 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1286 root->fs_info->fs_devices->num_devices <= 4) {
1287 printk(KERN_ERR "btrfs: unable to go below four devices "
1293 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1294 root->fs_info->fs_devices->num_devices <= 2) {
1295 printk(KERN_ERR "btrfs: unable to go below two "
1296 "devices on raid1\n");
1301 if (strcmp(device_path, "missing") == 0) {
1302 struct list_head *devices;
1303 struct btrfs_device *tmp;
1306 devices = &root->fs_info->fs_devices->devices;
1308 * It is safe to read the devices since the volume_mutex
1311 list_for_each_entry(tmp, devices, dev_list) {
1312 if (tmp->in_fs_metadata && !tmp->bdev) {
1321 printk(KERN_ERR "btrfs: no missing devices found to "
1326 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1327 root->fs_info->bdev_holder);
1329 ret = PTR_ERR(bdev);
1333 set_blocksize(bdev, 4096);
1334 bh = btrfs_read_dev_super(bdev);
1339 disk_super = (struct btrfs_super_block *)bh->b_data;
1340 devid = btrfs_stack_device_id(&disk_super->dev_item);
1341 dev_uuid = disk_super->dev_item.uuid;
1342 device = btrfs_find_device(root, devid, dev_uuid,
1350 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1351 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1357 if (device->writeable) {
1359 list_del_init(&device->dev_alloc_list);
1360 unlock_chunks(root);
1361 root->fs_info->fs_devices->rw_devices--;
1365 ret = btrfs_shrink_device(device, 0);
1369 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1373 spin_lock(&root->fs_info->free_chunk_lock);
1374 root->fs_info->free_chunk_space = device->total_bytes -
1376 spin_unlock(&root->fs_info->free_chunk_lock);
1378 device->in_fs_metadata = 0;
1379 btrfs_scrub_cancel_dev(root, device);
1382 * the device list mutex makes sure that we don't change
1383 * the device list while someone else is writing out all
1384 * the device supers.
1387 cur_devices = device->fs_devices;
1388 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1389 list_del_rcu(&device->dev_list);
1391 device->fs_devices->num_devices--;
1393 if (device->missing)
1394 root->fs_info->fs_devices->missing_devices--;
1396 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1397 struct btrfs_device, dev_list);
1398 if (device->bdev == root->fs_info->sb->s_bdev)
1399 root->fs_info->sb->s_bdev = next_device->bdev;
1400 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1401 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1404 device->fs_devices->open_devices--;
1406 call_rcu(&device->rcu, free_device);
1407 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1409 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1410 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1412 if (cur_devices->open_devices == 0) {
1413 struct btrfs_fs_devices *fs_devices;
1414 fs_devices = root->fs_info->fs_devices;
1415 while (fs_devices) {
1416 if (fs_devices->seed == cur_devices)
1418 fs_devices = fs_devices->seed;
1420 fs_devices->seed = cur_devices->seed;
1421 cur_devices->seed = NULL;
1423 __btrfs_close_devices(cur_devices);
1424 unlock_chunks(root);
1425 free_fs_devices(cur_devices);
1429 * at this point, the device is zero sized. We want to
1430 * remove it from the devices list and zero out the old super
1433 /* make sure this device isn't detected as part of
1436 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1437 set_buffer_dirty(bh);
1438 sync_dirty_buffer(bh);
1447 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1449 mutex_unlock(&root->fs_info->volume_mutex);
1450 mutex_unlock(&uuid_mutex);
1453 if (device->writeable) {
1455 list_add(&device->dev_alloc_list,
1456 &root->fs_info->fs_devices->alloc_list);
1457 unlock_chunks(root);
1458 root->fs_info->fs_devices->rw_devices++;
1464 * does all the dirty work required for changing file system's UUID.
1466 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1467 struct btrfs_root *root)
1469 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1470 struct btrfs_fs_devices *old_devices;
1471 struct btrfs_fs_devices *seed_devices;
1472 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1473 struct btrfs_device *device;
1476 BUG_ON(!mutex_is_locked(&uuid_mutex));
1477 if (!fs_devices->seeding)
1480 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1484 old_devices = clone_fs_devices(fs_devices);
1485 if (IS_ERR(old_devices)) {
1486 kfree(seed_devices);
1487 return PTR_ERR(old_devices);
1490 list_add(&old_devices->list, &fs_uuids);
1492 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1493 seed_devices->opened = 1;
1494 INIT_LIST_HEAD(&seed_devices->devices);
1495 INIT_LIST_HEAD(&seed_devices->alloc_list);
1496 mutex_init(&seed_devices->device_list_mutex);
1498 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1499 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1501 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1503 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1504 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1505 device->fs_devices = seed_devices;
1508 fs_devices->seeding = 0;
1509 fs_devices->num_devices = 0;
1510 fs_devices->open_devices = 0;
1511 fs_devices->seed = seed_devices;
1513 generate_random_uuid(fs_devices->fsid);
1514 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1515 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1516 super_flags = btrfs_super_flags(disk_super) &
1517 ~BTRFS_SUPER_FLAG_SEEDING;
1518 btrfs_set_super_flags(disk_super, super_flags);
1524 * strore the expected generation for seed devices in device items.
1526 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1527 struct btrfs_root *root)
1529 struct btrfs_path *path;
1530 struct extent_buffer *leaf;
1531 struct btrfs_dev_item *dev_item;
1532 struct btrfs_device *device;
1533 struct btrfs_key key;
1534 u8 fs_uuid[BTRFS_UUID_SIZE];
1535 u8 dev_uuid[BTRFS_UUID_SIZE];
1539 path = btrfs_alloc_path();
1543 root = root->fs_info->chunk_root;
1544 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1546 key.type = BTRFS_DEV_ITEM_KEY;
1549 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1553 leaf = path->nodes[0];
1555 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1556 ret = btrfs_next_leaf(root, path);
1561 leaf = path->nodes[0];
1562 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1563 btrfs_release_path(path);
1567 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1568 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1569 key.type != BTRFS_DEV_ITEM_KEY)
1572 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1573 struct btrfs_dev_item);
1574 devid = btrfs_device_id(leaf, dev_item);
1575 read_extent_buffer(leaf, dev_uuid,
1576 (unsigned long)btrfs_device_uuid(dev_item),
1578 read_extent_buffer(leaf, fs_uuid,
1579 (unsigned long)btrfs_device_fsid(dev_item),
1581 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1584 if (device->fs_devices->seeding) {
1585 btrfs_set_device_generation(leaf, dev_item,
1586 device->generation);
1587 btrfs_mark_buffer_dirty(leaf);
1595 btrfs_free_path(path);
1599 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1601 struct request_queue *q;
1602 struct btrfs_trans_handle *trans;
1603 struct btrfs_device *device;
1604 struct block_device *bdev;
1605 struct list_head *devices;
1606 struct super_block *sb = root->fs_info->sb;
1608 int seeding_dev = 0;
1611 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1614 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1615 root->fs_info->bdev_holder);
1617 return PTR_ERR(bdev);
1619 if (root->fs_info->fs_devices->seeding) {
1621 down_write(&sb->s_umount);
1622 mutex_lock(&uuid_mutex);
1625 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1626 mutex_lock(&root->fs_info->volume_mutex);
1628 devices = &root->fs_info->fs_devices->devices;
1630 * we have the volume lock, so we don't need the extra
1631 * device list mutex while reading the list here.
1633 list_for_each_entry(device, devices, dev_list) {
1634 if (device->bdev == bdev) {
1640 device = kzalloc(sizeof(*device), GFP_NOFS);
1642 /* we can safely leave the fs_devices entry around */
1647 device->name = kstrdup(device_path, GFP_NOFS);
1648 if (!device->name) {
1654 ret = find_next_devid(root, &device->devid);
1656 kfree(device->name);
1661 trans = btrfs_start_transaction(root, 0);
1662 if (IS_ERR(trans)) {
1663 kfree(device->name);
1665 ret = PTR_ERR(trans);
1671 q = bdev_get_queue(bdev);
1672 if (blk_queue_discard(q))
1673 device->can_discard = 1;
1674 device->writeable = 1;
1675 device->work.func = pending_bios_fn;
1676 generate_random_uuid(device->uuid);
1677 spin_lock_init(&device->io_lock);
1678 device->generation = trans->transid;
1679 device->io_width = root->sectorsize;
1680 device->io_align = root->sectorsize;
1681 device->sector_size = root->sectorsize;
1682 device->total_bytes = i_size_read(bdev->bd_inode);
1683 device->disk_total_bytes = device->total_bytes;
1684 device->dev_root = root->fs_info->dev_root;
1685 device->bdev = bdev;
1686 device->in_fs_metadata = 1;
1687 device->mode = FMODE_EXCL;
1688 set_blocksize(device->bdev, 4096);
1691 sb->s_flags &= ~MS_RDONLY;
1692 ret = btrfs_prepare_sprout(trans, root);
1696 device->fs_devices = root->fs_info->fs_devices;
1699 * we don't want write_supers to jump in here with our device
1702 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1703 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1704 list_add(&device->dev_alloc_list,
1705 &root->fs_info->fs_devices->alloc_list);
1706 root->fs_info->fs_devices->num_devices++;
1707 root->fs_info->fs_devices->open_devices++;
1708 root->fs_info->fs_devices->rw_devices++;
1709 if (device->can_discard)
1710 root->fs_info->fs_devices->num_can_discard++;
1711 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1713 spin_lock(&root->fs_info->free_chunk_lock);
1714 root->fs_info->free_chunk_space += device->total_bytes;
1715 spin_unlock(&root->fs_info->free_chunk_lock);
1717 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1718 root->fs_info->fs_devices->rotating = 1;
1720 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1721 btrfs_set_super_total_bytes(root->fs_info->super_copy,
1722 total_bytes + device->total_bytes);
1724 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1725 btrfs_set_super_num_devices(root->fs_info->super_copy,
1727 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1730 ret = init_first_rw_device(trans, root, device);
1732 ret = btrfs_finish_sprout(trans, root);
1735 ret = btrfs_add_device(trans, root, device);
1739 * we've got more storage, clear any full flags on the space
1742 btrfs_clear_space_info_full(root->fs_info);
1744 unlock_chunks(root);
1745 btrfs_commit_transaction(trans, root);
1748 mutex_unlock(&uuid_mutex);
1749 up_write(&sb->s_umount);
1751 ret = btrfs_relocate_sys_chunks(root);
1755 mutex_unlock(&root->fs_info->volume_mutex);
1758 blkdev_put(bdev, FMODE_EXCL);
1760 mutex_unlock(&uuid_mutex);
1761 up_write(&sb->s_umount);
1766 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1767 struct btrfs_device *device)
1770 struct btrfs_path *path;
1771 struct btrfs_root *root;
1772 struct btrfs_dev_item *dev_item;
1773 struct extent_buffer *leaf;
1774 struct btrfs_key key;
1776 root = device->dev_root->fs_info->chunk_root;
1778 path = btrfs_alloc_path();
1782 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1783 key.type = BTRFS_DEV_ITEM_KEY;
1784 key.offset = device->devid;
1786 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1795 leaf = path->nodes[0];
1796 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1798 btrfs_set_device_id(leaf, dev_item, device->devid);
1799 btrfs_set_device_type(leaf, dev_item, device->type);
1800 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1801 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1802 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1803 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1804 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1805 btrfs_mark_buffer_dirty(leaf);
1808 btrfs_free_path(path);
1812 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1813 struct btrfs_device *device, u64 new_size)
1815 struct btrfs_super_block *super_copy =
1816 device->dev_root->fs_info->super_copy;
1817 u64 old_total = btrfs_super_total_bytes(super_copy);
1818 u64 diff = new_size - device->total_bytes;
1820 if (!device->writeable)
1822 if (new_size <= device->total_bytes)
1825 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1826 device->fs_devices->total_rw_bytes += diff;
1828 device->total_bytes = new_size;
1829 device->disk_total_bytes = new_size;
1830 btrfs_clear_space_info_full(device->dev_root->fs_info);
1832 return btrfs_update_device(trans, device);
1835 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1836 struct btrfs_device *device, u64 new_size)
1839 lock_chunks(device->dev_root);
1840 ret = __btrfs_grow_device(trans, device, new_size);
1841 unlock_chunks(device->dev_root);
1845 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1846 struct btrfs_root *root,
1847 u64 chunk_tree, u64 chunk_objectid,
1851 struct btrfs_path *path;
1852 struct btrfs_key key;
1854 root = root->fs_info->chunk_root;
1855 path = btrfs_alloc_path();
1859 key.objectid = chunk_objectid;
1860 key.offset = chunk_offset;
1861 key.type = BTRFS_CHUNK_ITEM_KEY;
1863 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1866 ret = btrfs_del_item(trans, root, path);
1868 btrfs_free_path(path);
1872 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1875 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1876 struct btrfs_disk_key *disk_key;
1877 struct btrfs_chunk *chunk;
1884 struct btrfs_key key;
1886 array_size = btrfs_super_sys_array_size(super_copy);
1888 ptr = super_copy->sys_chunk_array;
1891 while (cur < array_size) {
1892 disk_key = (struct btrfs_disk_key *)ptr;
1893 btrfs_disk_key_to_cpu(&key, disk_key);
1895 len = sizeof(*disk_key);
1897 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1898 chunk = (struct btrfs_chunk *)(ptr + len);
1899 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1900 len += btrfs_chunk_item_size(num_stripes);
1905 if (key.objectid == chunk_objectid &&
1906 key.offset == chunk_offset) {
1907 memmove(ptr, ptr + len, array_size - (cur + len));
1909 btrfs_set_super_sys_array_size(super_copy, array_size);
1918 static int btrfs_relocate_chunk(struct btrfs_root *root,
1919 u64 chunk_tree, u64 chunk_objectid,
1922 struct extent_map_tree *em_tree;
1923 struct btrfs_root *extent_root;
1924 struct btrfs_trans_handle *trans;
1925 struct extent_map *em;
1926 struct map_lookup *map;
1930 root = root->fs_info->chunk_root;
1931 extent_root = root->fs_info->extent_root;
1932 em_tree = &root->fs_info->mapping_tree.map_tree;
1934 ret = btrfs_can_relocate(extent_root, chunk_offset);
1938 /* step one, relocate all the extents inside this chunk */
1939 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1943 trans = btrfs_start_transaction(root, 0);
1944 BUG_ON(IS_ERR(trans));
1949 * step two, delete the device extents and the
1950 * chunk tree entries
1952 read_lock(&em_tree->lock);
1953 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1954 read_unlock(&em_tree->lock);
1956 BUG_ON(em->start > chunk_offset ||
1957 em->start + em->len < chunk_offset);
1958 map = (struct map_lookup *)em->bdev;
1960 for (i = 0; i < map->num_stripes; i++) {
1961 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1962 map->stripes[i].physical);
1965 if (map->stripes[i].dev) {
1966 ret = btrfs_update_device(trans, map->stripes[i].dev);
1970 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1975 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1977 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1978 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1982 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1985 write_lock(&em_tree->lock);
1986 remove_extent_mapping(em_tree, em);
1987 write_unlock(&em_tree->lock);
1992 /* once for the tree */
1993 free_extent_map(em);
1995 free_extent_map(em);
1997 unlock_chunks(root);
1998 btrfs_end_transaction(trans, root);
2002 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2004 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2005 struct btrfs_path *path;
2006 struct extent_buffer *leaf;
2007 struct btrfs_chunk *chunk;
2008 struct btrfs_key key;
2009 struct btrfs_key found_key;
2010 u64 chunk_tree = chunk_root->root_key.objectid;
2012 bool retried = false;
2016 path = btrfs_alloc_path();
2021 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2022 key.offset = (u64)-1;
2023 key.type = BTRFS_CHUNK_ITEM_KEY;
2026 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2031 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2038 leaf = path->nodes[0];
2039 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2041 chunk = btrfs_item_ptr(leaf, path->slots[0],
2042 struct btrfs_chunk);
2043 chunk_type = btrfs_chunk_type(leaf, chunk);
2044 btrfs_release_path(path);
2046 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2047 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2056 if (found_key.offset == 0)
2058 key.offset = found_key.offset - 1;
2061 if (failed && !retried) {
2065 } else if (failed && retried) {
2070 btrfs_free_path(path);
2074 static u64 div_factor(u64 num, int factor)
2083 int btrfs_balance(struct btrfs_root *dev_root)
2086 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2087 struct btrfs_device *device;
2090 struct btrfs_path *path;
2091 struct btrfs_key key;
2092 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2093 struct btrfs_trans_handle *trans;
2094 struct btrfs_key found_key;
2096 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2099 if (!capable(CAP_SYS_ADMIN))
2102 mutex_lock(&dev_root->fs_info->volume_mutex);
2103 dev_root = dev_root->fs_info->dev_root;
2105 /* step one make some room on all the devices */
2106 list_for_each_entry(device, devices, dev_list) {
2107 old_size = device->total_bytes;
2108 size_to_free = div_factor(old_size, 1);
2109 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2110 if (!device->writeable ||
2111 device->total_bytes - device->bytes_used > size_to_free)
2114 ret = btrfs_shrink_device(device, old_size - size_to_free);
2119 trans = btrfs_start_transaction(dev_root, 0);
2120 BUG_ON(IS_ERR(trans));
2122 ret = btrfs_grow_device(trans, device, old_size);
2125 btrfs_end_transaction(trans, dev_root);
2128 /* step two, relocate all the chunks */
2129 path = btrfs_alloc_path();
2134 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2135 key.offset = (u64)-1;
2136 key.type = BTRFS_CHUNK_ITEM_KEY;
2139 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2144 * this shouldn't happen, it means the last relocate
2150 ret = btrfs_previous_item(chunk_root, path, 0,
2151 BTRFS_CHUNK_ITEM_KEY);
2155 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2157 if (found_key.objectid != key.objectid)
2160 /* chunk zero is special */
2161 if (found_key.offset == 0)
2164 btrfs_release_path(path);
2165 ret = btrfs_relocate_chunk(chunk_root,
2166 chunk_root->root_key.objectid,
2169 if (ret && ret != -ENOSPC)
2171 key.offset = found_key.offset - 1;
2175 btrfs_free_path(path);
2176 mutex_unlock(&dev_root->fs_info->volume_mutex);
2181 * shrinking a device means finding all of the device extents past
2182 * the new size, and then following the back refs to the chunks.
2183 * The chunk relocation code actually frees the device extent
2185 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2187 struct btrfs_trans_handle *trans;
2188 struct btrfs_root *root = device->dev_root;
2189 struct btrfs_dev_extent *dev_extent = NULL;
2190 struct btrfs_path *path;
2198 bool retried = false;
2199 struct extent_buffer *l;
2200 struct btrfs_key key;
2201 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2202 u64 old_total = btrfs_super_total_bytes(super_copy);
2203 u64 old_size = device->total_bytes;
2204 u64 diff = device->total_bytes - new_size;
2206 if (new_size >= device->total_bytes)
2209 path = btrfs_alloc_path();
2217 device->total_bytes = new_size;
2218 if (device->writeable) {
2219 device->fs_devices->total_rw_bytes -= diff;
2220 spin_lock(&root->fs_info->free_chunk_lock);
2221 root->fs_info->free_chunk_space -= diff;
2222 spin_unlock(&root->fs_info->free_chunk_lock);
2224 unlock_chunks(root);
2227 key.objectid = device->devid;
2228 key.offset = (u64)-1;
2229 key.type = BTRFS_DEV_EXTENT_KEY;
2232 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2236 ret = btrfs_previous_item(root, path, 0, key.type);
2241 btrfs_release_path(path);
2246 slot = path->slots[0];
2247 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2249 if (key.objectid != device->devid) {
2250 btrfs_release_path(path);
2254 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2255 length = btrfs_dev_extent_length(l, dev_extent);
2257 if (key.offset + length <= new_size) {
2258 btrfs_release_path(path);
2262 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2263 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2264 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2265 btrfs_release_path(path);
2267 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2269 if (ret && ret != -ENOSPC)
2276 if (failed && !retried) {
2280 } else if (failed && retried) {
2284 device->total_bytes = old_size;
2285 if (device->writeable)
2286 device->fs_devices->total_rw_bytes += diff;
2287 spin_lock(&root->fs_info->free_chunk_lock);
2288 root->fs_info->free_chunk_space += diff;
2289 spin_unlock(&root->fs_info->free_chunk_lock);
2290 unlock_chunks(root);
2294 /* Shrinking succeeded, else we would be at "done". */
2295 trans = btrfs_start_transaction(root, 0);
2296 if (IS_ERR(trans)) {
2297 ret = PTR_ERR(trans);
2303 device->disk_total_bytes = new_size;
2304 /* Now btrfs_update_device() will change the on-disk size. */
2305 ret = btrfs_update_device(trans, device);
2307 unlock_chunks(root);
2308 btrfs_end_transaction(trans, root);
2311 WARN_ON(diff > old_total);
2312 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2313 unlock_chunks(root);
2314 btrfs_end_transaction(trans, root);
2316 btrfs_free_path(path);
2320 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2321 struct btrfs_root *root,
2322 struct btrfs_key *key,
2323 struct btrfs_chunk *chunk, int item_size)
2325 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2326 struct btrfs_disk_key disk_key;
2330 array_size = btrfs_super_sys_array_size(super_copy);
2331 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2334 ptr = super_copy->sys_chunk_array + array_size;
2335 btrfs_cpu_key_to_disk(&disk_key, key);
2336 memcpy(ptr, &disk_key, sizeof(disk_key));
2337 ptr += sizeof(disk_key);
2338 memcpy(ptr, chunk, item_size);
2339 item_size += sizeof(disk_key);
2340 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2345 * sort the devices in descending order by max_avail, total_avail
2347 static int btrfs_cmp_device_info(const void *a, const void *b)
2349 const struct btrfs_device_info *di_a = a;
2350 const struct btrfs_device_info *di_b = b;
2352 if (di_a->max_avail > di_b->max_avail)
2354 if (di_a->max_avail < di_b->max_avail)
2356 if (di_a->total_avail > di_b->total_avail)
2358 if (di_a->total_avail < di_b->total_avail)
2363 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2364 struct btrfs_root *extent_root,
2365 struct map_lookup **map_ret,
2366 u64 *num_bytes_out, u64 *stripe_size_out,
2367 u64 start, u64 type)
2369 struct btrfs_fs_info *info = extent_root->fs_info;
2370 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2371 struct list_head *cur;
2372 struct map_lookup *map = NULL;
2373 struct extent_map_tree *em_tree;
2374 struct extent_map *em;
2375 struct btrfs_device_info *devices_info = NULL;
2377 int num_stripes; /* total number of stripes to allocate */
2378 int sub_stripes; /* sub_stripes info for map */
2379 int dev_stripes; /* stripes per dev */
2380 int devs_max; /* max devs to use */
2381 int devs_min; /* min devs needed */
2382 int devs_increment; /* ndevs has to be a multiple of this */
2383 int ncopies; /* how many copies to data has */
2385 u64 max_stripe_size;
2393 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2394 (type & BTRFS_BLOCK_GROUP_DUP)) {
2396 type &= ~BTRFS_BLOCK_GROUP_DUP;
2399 if (list_empty(&fs_devices->alloc_list))
2406 devs_max = 0; /* 0 == as many as possible */
2410 * define the properties of each RAID type.
2411 * FIXME: move this to a global table and use it in all RAID
2414 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2418 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2420 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2425 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2434 if (type & BTRFS_BLOCK_GROUP_DATA) {
2435 max_stripe_size = 1024 * 1024 * 1024;
2436 max_chunk_size = 10 * max_stripe_size;
2437 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2438 max_stripe_size = 256 * 1024 * 1024;
2439 max_chunk_size = max_stripe_size;
2440 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2441 max_stripe_size = 8 * 1024 * 1024;
2442 max_chunk_size = 2 * max_stripe_size;
2444 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2449 /* we don't want a chunk larger than 10% of writeable space */
2450 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2453 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2458 cur = fs_devices->alloc_list.next;
2461 * in the first pass through the devices list, we gather information
2462 * about the available holes on each device.
2465 while (cur != &fs_devices->alloc_list) {
2466 struct btrfs_device *device;
2470 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2474 if (!device->writeable) {
2476 "btrfs: read-only device in alloc_list\n");
2481 if (!device->in_fs_metadata)
2484 if (device->total_bytes > device->bytes_used)
2485 total_avail = device->total_bytes - device->bytes_used;
2489 /* If there is no space on this device, skip it. */
2490 if (total_avail == 0)
2493 ret = find_free_dev_extent(trans, device,
2494 max_stripe_size * dev_stripes,
2495 &dev_offset, &max_avail);
2496 if (ret && ret != -ENOSPC)
2500 max_avail = max_stripe_size * dev_stripes;
2502 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2505 devices_info[ndevs].dev_offset = dev_offset;
2506 devices_info[ndevs].max_avail = max_avail;
2507 devices_info[ndevs].total_avail = total_avail;
2508 devices_info[ndevs].dev = device;
2513 * now sort the devices by hole size / available space
2515 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2516 btrfs_cmp_device_info, NULL);
2518 /* round down to number of usable stripes */
2519 ndevs -= ndevs % devs_increment;
2521 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2526 if (devs_max && ndevs > devs_max)
2529 * the primary goal is to maximize the number of stripes, so use as many
2530 * devices as possible, even if the stripes are not maximum sized.
2532 stripe_size = devices_info[ndevs-1].max_avail;
2533 num_stripes = ndevs * dev_stripes;
2535 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2536 stripe_size = max_chunk_size * ncopies;
2537 do_div(stripe_size, num_stripes);
2540 do_div(stripe_size, dev_stripes);
2541 do_div(stripe_size, BTRFS_STRIPE_LEN);
2542 stripe_size *= BTRFS_STRIPE_LEN;
2544 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2549 map->num_stripes = num_stripes;
2551 for (i = 0; i < ndevs; ++i) {
2552 for (j = 0; j < dev_stripes; ++j) {
2553 int s = i * dev_stripes + j;
2554 map->stripes[s].dev = devices_info[i].dev;
2555 map->stripes[s].physical = devices_info[i].dev_offset +
2559 map->sector_size = extent_root->sectorsize;
2560 map->stripe_len = BTRFS_STRIPE_LEN;
2561 map->io_align = BTRFS_STRIPE_LEN;
2562 map->io_width = BTRFS_STRIPE_LEN;
2564 map->sub_stripes = sub_stripes;
2567 num_bytes = stripe_size * (num_stripes / ncopies);
2569 *stripe_size_out = stripe_size;
2570 *num_bytes_out = num_bytes;
2572 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2574 em = alloc_extent_map();
2579 em->bdev = (struct block_device *)map;
2581 em->len = num_bytes;
2582 em->block_start = 0;
2583 em->block_len = em->len;
2585 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2586 write_lock(&em_tree->lock);
2587 ret = add_extent_mapping(em_tree, em);
2588 write_unlock(&em_tree->lock);
2590 free_extent_map(em);
2592 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2593 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2597 for (i = 0; i < map->num_stripes; ++i) {
2598 struct btrfs_device *device;
2601 device = map->stripes[i].dev;
2602 dev_offset = map->stripes[i].physical;
2604 ret = btrfs_alloc_dev_extent(trans, device,
2605 info->chunk_root->root_key.objectid,
2606 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2607 start, dev_offset, stripe_size);
2611 kfree(devices_info);
2616 kfree(devices_info);
2620 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2621 struct btrfs_root *extent_root,
2622 struct map_lookup *map, u64 chunk_offset,
2623 u64 chunk_size, u64 stripe_size)
2626 struct btrfs_key key;
2627 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2628 struct btrfs_device *device;
2629 struct btrfs_chunk *chunk;
2630 struct btrfs_stripe *stripe;
2631 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2635 chunk = kzalloc(item_size, GFP_NOFS);
2640 while (index < map->num_stripes) {
2641 device = map->stripes[index].dev;
2642 device->bytes_used += stripe_size;
2643 ret = btrfs_update_device(trans, device);
2648 spin_lock(&extent_root->fs_info->free_chunk_lock);
2649 extent_root->fs_info->free_chunk_space -= (stripe_size *
2651 spin_unlock(&extent_root->fs_info->free_chunk_lock);
2654 stripe = &chunk->stripe;
2655 while (index < map->num_stripes) {
2656 device = map->stripes[index].dev;
2657 dev_offset = map->stripes[index].physical;
2659 btrfs_set_stack_stripe_devid(stripe, device->devid);
2660 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2661 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2666 btrfs_set_stack_chunk_length(chunk, chunk_size);
2667 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2668 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2669 btrfs_set_stack_chunk_type(chunk, map->type);
2670 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2671 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2672 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2673 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2674 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2676 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2677 key.type = BTRFS_CHUNK_ITEM_KEY;
2678 key.offset = chunk_offset;
2680 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2683 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2684 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2694 * Chunk allocation falls into two parts. The first part does works
2695 * that make the new allocated chunk useable, but not do any operation
2696 * that modifies the chunk tree. The second part does the works that
2697 * require modifying the chunk tree. This division is important for the
2698 * bootstrap process of adding storage to a seed btrfs.
2700 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2701 struct btrfs_root *extent_root, u64 type)
2706 struct map_lookup *map;
2707 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2710 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2715 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2716 &stripe_size, chunk_offset, type);
2720 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2721 chunk_size, stripe_size);
2726 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2727 struct btrfs_root *root,
2728 struct btrfs_device *device)
2731 u64 sys_chunk_offset;
2735 u64 sys_stripe_size;
2737 struct map_lookup *map;
2738 struct map_lookup *sys_map;
2739 struct btrfs_fs_info *fs_info = root->fs_info;
2740 struct btrfs_root *extent_root = fs_info->extent_root;
2743 ret = find_next_chunk(fs_info->chunk_root,
2744 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2748 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2749 (fs_info->metadata_alloc_profile &
2750 fs_info->avail_metadata_alloc_bits);
2751 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2753 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2754 &stripe_size, chunk_offset, alloc_profile);
2757 sys_chunk_offset = chunk_offset + chunk_size;
2759 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2760 (fs_info->system_alloc_profile &
2761 fs_info->avail_system_alloc_bits);
2762 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2764 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2765 &sys_chunk_size, &sys_stripe_size,
2766 sys_chunk_offset, alloc_profile);
2769 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2773 * Modifying chunk tree needs allocating new blocks from both
2774 * system block group and metadata block group. So we only can
2775 * do operations require modifying the chunk tree after both
2776 * block groups were created.
2778 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2779 chunk_size, stripe_size);
2782 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2783 sys_chunk_offset, sys_chunk_size,
2789 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2791 struct extent_map *em;
2792 struct map_lookup *map;
2793 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2797 read_lock(&map_tree->map_tree.lock);
2798 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2799 read_unlock(&map_tree->map_tree.lock);
2803 if (btrfs_test_opt(root, DEGRADED)) {
2804 free_extent_map(em);
2808 map = (struct map_lookup *)em->bdev;
2809 for (i = 0; i < map->num_stripes; i++) {
2810 if (!map->stripes[i].dev->writeable) {
2815 free_extent_map(em);
2819 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2821 extent_map_tree_init(&tree->map_tree);
2824 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2826 struct extent_map *em;
2829 write_lock(&tree->map_tree.lock);
2830 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2832 remove_extent_mapping(&tree->map_tree, em);
2833 write_unlock(&tree->map_tree.lock);
2838 free_extent_map(em);
2839 /* once for the tree */
2840 free_extent_map(em);
2844 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2846 struct extent_map *em;
2847 struct map_lookup *map;
2848 struct extent_map_tree *em_tree = &map_tree->map_tree;
2851 read_lock(&em_tree->lock);
2852 em = lookup_extent_mapping(em_tree, logical, len);
2853 read_unlock(&em_tree->lock);
2856 BUG_ON(em->start > logical || em->start + em->len < logical);
2857 map = (struct map_lookup *)em->bdev;
2858 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2859 ret = map->num_stripes;
2860 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2861 ret = map->sub_stripes;
2864 free_extent_map(em);
2868 static int find_live_mirror(struct map_lookup *map, int first, int num,
2872 if (map->stripes[optimal].dev->bdev)
2874 for (i = first; i < first + num; i++) {
2875 if (map->stripes[i].dev->bdev)
2878 /* we couldn't find one that doesn't fail. Just return something
2879 * and the io error handling code will clean up eventually
2884 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2885 u64 logical, u64 *length,
2886 struct btrfs_bio **bbio_ret,
2889 struct extent_map *em;
2890 struct map_lookup *map;
2891 struct extent_map_tree *em_tree = &map_tree->map_tree;
2894 u64 stripe_end_offset;
2898 int stripes_allocated = 8;
2899 int stripes_required = 1;
2904 struct btrfs_bio *bbio = NULL;
2906 if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2907 stripes_allocated = 1;
2910 bbio = kzalloc(btrfs_bio_size(stripes_allocated),
2915 atomic_set(&bbio->error, 0);
2918 read_lock(&em_tree->lock);
2919 em = lookup_extent_mapping(em_tree, logical, *length);
2920 read_unlock(&em_tree->lock);
2923 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2924 (unsigned long long)logical,
2925 (unsigned long long)*length);
2929 BUG_ON(em->start > logical || em->start + em->len < logical);
2930 map = (struct map_lookup *)em->bdev;
2931 offset = logical - em->start;
2933 if (mirror_num > map->num_stripes)
2936 /* if our btrfs_bio struct is too small, back off and try again */
2937 if (rw & REQ_WRITE) {
2938 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2939 BTRFS_BLOCK_GROUP_DUP)) {
2940 stripes_required = map->num_stripes;
2942 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2943 stripes_required = map->sub_stripes;
2947 if (rw & REQ_DISCARD) {
2948 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2949 BTRFS_BLOCK_GROUP_RAID1 |
2950 BTRFS_BLOCK_GROUP_DUP |
2951 BTRFS_BLOCK_GROUP_RAID10)) {
2952 stripes_required = map->num_stripes;
2955 if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2956 stripes_allocated < stripes_required) {
2957 stripes_allocated = map->num_stripes;
2958 free_extent_map(em);
2964 * stripe_nr counts the total number of stripes we have to stride
2965 * to get to this block
2967 do_div(stripe_nr, map->stripe_len);
2969 stripe_offset = stripe_nr * map->stripe_len;
2970 BUG_ON(offset < stripe_offset);
2972 /* stripe_offset is the offset of this block in its stripe*/
2973 stripe_offset = offset - stripe_offset;
2975 if (rw & REQ_DISCARD)
2976 *length = min_t(u64, em->len - offset, *length);
2977 else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2978 BTRFS_BLOCK_GROUP_RAID1 |
2979 BTRFS_BLOCK_GROUP_RAID10 |
2980 BTRFS_BLOCK_GROUP_DUP)) {
2981 /* we limit the length of each bio to what fits in a stripe */
2982 *length = min_t(u64, em->len - offset,
2983 map->stripe_len - stripe_offset);
2985 *length = em->len - offset;
2993 stripe_nr_orig = stripe_nr;
2994 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2995 (~(map->stripe_len - 1));
2996 do_div(stripe_nr_end, map->stripe_len);
2997 stripe_end_offset = stripe_nr_end * map->stripe_len -
2999 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3000 if (rw & REQ_DISCARD)
3001 num_stripes = min_t(u64, map->num_stripes,
3002 stripe_nr_end - stripe_nr_orig);
3003 stripe_index = do_div(stripe_nr, map->num_stripes);
3004 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3005 if (rw & (REQ_WRITE | REQ_DISCARD))
3006 num_stripes = map->num_stripes;
3007 else if (mirror_num)
3008 stripe_index = mirror_num - 1;
3010 stripe_index = find_live_mirror(map, 0,
3012 current->pid % map->num_stripes);
3013 mirror_num = stripe_index + 1;
3016 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3017 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3018 num_stripes = map->num_stripes;
3019 } else if (mirror_num) {
3020 stripe_index = mirror_num - 1;
3025 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3026 int factor = map->num_stripes / map->sub_stripes;
3028 stripe_index = do_div(stripe_nr, factor);
3029 stripe_index *= map->sub_stripes;
3032 num_stripes = map->sub_stripes;
3033 else if (rw & REQ_DISCARD)
3034 num_stripes = min_t(u64, map->sub_stripes *
3035 (stripe_nr_end - stripe_nr_orig),
3037 else if (mirror_num)
3038 stripe_index += mirror_num - 1;
3040 stripe_index = find_live_mirror(map, stripe_index,
3041 map->sub_stripes, stripe_index +
3042 current->pid % map->sub_stripes);
3043 mirror_num = stripe_index + 1;
3047 * after this do_div call, stripe_nr is the number of stripes
3048 * on this device we have to walk to find the data, and
3049 * stripe_index is the number of our device in the stripe array
3051 stripe_index = do_div(stripe_nr, map->num_stripes);
3052 mirror_num = stripe_index + 1;
3054 BUG_ON(stripe_index >= map->num_stripes);
3056 if (rw & REQ_DISCARD) {
3057 for (i = 0; i < num_stripes; i++) {
3058 bbio->stripes[i].physical =
3059 map->stripes[stripe_index].physical +
3060 stripe_offset + stripe_nr * map->stripe_len;
3061 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3063 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3065 u32 last_stripe = 0;
3068 div_u64_rem(stripe_nr_end - 1,
3072 for (j = 0; j < map->num_stripes; j++) {
3075 div_u64_rem(stripe_nr_end - 1 - j,
3076 map->num_stripes, &test);
3077 if (test == stripe_index)
3080 stripes = stripe_nr_end - 1 - j;
3081 do_div(stripes, map->num_stripes);
3082 bbio->stripes[i].length = map->stripe_len *
3083 (stripes - stripe_nr + 1);
3086 bbio->stripes[i].length -=
3090 if (stripe_index == last_stripe)
3091 bbio->stripes[i].length -=
3093 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3096 int factor = map->num_stripes /
3098 u32 last_stripe = 0;
3100 div_u64_rem(stripe_nr_end - 1,
3101 factor, &last_stripe);
3102 last_stripe *= map->sub_stripes;
3104 for (j = 0; j < factor; j++) {
3107 div_u64_rem(stripe_nr_end - 1 - j,
3111 stripe_index / map->sub_stripes)
3114 stripes = stripe_nr_end - 1 - j;
3115 do_div(stripes, factor);
3116 bbio->stripes[i].length = map->stripe_len *
3117 (stripes - stripe_nr + 1);
3119 if (i < map->sub_stripes) {
3120 bbio->stripes[i].length -=
3122 if (i == map->sub_stripes - 1)
3125 if (stripe_index >= last_stripe &&
3126 stripe_index <= (last_stripe +
3127 map->sub_stripes - 1)) {
3128 bbio->stripes[i].length -=
3132 bbio->stripes[i].length = *length;
3135 if (stripe_index == map->num_stripes) {
3136 /* This could only happen for RAID0/10 */
3142 for (i = 0; i < num_stripes; i++) {
3143 bbio->stripes[i].physical =
3144 map->stripes[stripe_index].physical +
3146 stripe_nr * map->stripe_len;
3147 bbio->stripes[i].dev =
3148 map->stripes[stripe_index].dev;
3154 bbio->num_stripes = num_stripes;
3155 bbio->max_errors = max_errors;
3156 bbio->mirror_num = mirror_num;
3159 free_extent_map(em);
3163 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3164 u64 logical, u64 *length,
3165 struct btrfs_bio **bbio_ret, int mirror_num)
3167 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3171 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3172 u64 chunk_start, u64 physical, u64 devid,
3173 u64 **logical, int *naddrs, int *stripe_len)
3175 struct extent_map_tree *em_tree = &map_tree->map_tree;
3176 struct extent_map *em;
3177 struct map_lookup *map;
3184 read_lock(&em_tree->lock);
3185 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3186 read_unlock(&em_tree->lock);
3188 BUG_ON(!em || em->start != chunk_start);
3189 map = (struct map_lookup *)em->bdev;
3192 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3193 do_div(length, map->num_stripes / map->sub_stripes);
3194 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3195 do_div(length, map->num_stripes);
3197 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3200 for (i = 0; i < map->num_stripes; i++) {
3201 if (devid && map->stripes[i].dev->devid != devid)
3203 if (map->stripes[i].physical > physical ||
3204 map->stripes[i].physical + length <= physical)
3207 stripe_nr = physical - map->stripes[i].physical;
3208 do_div(stripe_nr, map->stripe_len);
3210 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3211 stripe_nr = stripe_nr * map->num_stripes + i;
3212 do_div(stripe_nr, map->sub_stripes);
3213 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3214 stripe_nr = stripe_nr * map->num_stripes + i;
3216 bytenr = chunk_start + stripe_nr * map->stripe_len;
3217 WARN_ON(nr >= map->num_stripes);
3218 for (j = 0; j < nr; j++) {
3219 if (buf[j] == bytenr)
3223 WARN_ON(nr >= map->num_stripes);
3230 *stripe_len = map->stripe_len;
3232 free_extent_map(em);
3236 static void btrfs_end_bio(struct bio *bio, int err)
3238 struct btrfs_bio *bbio = bio->bi_private;
3239 int is_orig_bio = 0;
3242 atomic_inc(&bbio->error);
3244 if (bio == bbio->orig_bio)
3247 if (atomic_dec_and_test(&bbio->stripes_pending)) {
3250 bio = bbio->orig_bio;
3252 bio->bi_private = bbio->private;
3253 bio->bi_end_io = bbio->end_io;
3254 bio->bi_bdev = (struct block_device *)
3255 (unsigned long)bbio->mirror_num;
3256 /* only send an error to the higher layers if it is
3257 * beyond the tolerance of the multi-bio
3259 if (atomic_read(&bbio->error) > bbio->max_errors) {
3263 * this bio is actually up to date, we didn't
3264 * go over the max number of errors
3266 set_bit(BIO_UPTODATE, &bio->bi_flags);
3271 bio_endio(bio, err);
3272 } else if (!is_orig_bio) {
3277 struct async_sched {
3280 struct btrfs_fs_info *info;
3281 struct btrfs_work work;
3285 * see run_scheduled_bios for a description of why bios are collected for
3288 * This will add one bio to the pending list for a device and make sure
3289 * the work struct is scheduled.
3291 static noinline int schedule_bio(struct btrfs_root *root,
3292 struct btrfs_device *device,
3293 int rw, struct bio *bio)
3295 int should_queue = 1;
3296 struct btrfs_pending_bios *pending_bios;
3298 /* don't bother with additional async steps for reads, right now */
3299 if (!(rw & REQ_WRITE)) {
3301 submit_bio(rw, bio);
3307 * nr_async_bios allows us to reliably return congestion to the
3308 * higher layers. Otherwise, the async bio makes it appear we have
3309 * made progress against dirty pages when we've really just put it
3310 * on a queue for later
3312 atomic_inc(&root->fs_info->nr_async_bios);
3313 WARN_ON(bio->bi_next);
3314 bio->bi_next = NULL;
3317 spin_lock(&device->io_lock);
3318 if (bio->bi_rw & REQ_SYNC)
3319 pending_bios = &device->pending_sync_bios;
3321 pending_bios = &device->pending_bios;
3323 if (pending_bios->tail)
3324 pending_bios->tail->bi_next = bio;
3326 pending_bios->tail = bio;
3327 if (!pending_bios->head)
3328 pending_bios->head = bio;
3329 if (device->running_pending)
3332 spin_unlock(&device->io_lock);
3335 btrfs_queue_worker(&root->fs_info->submit_workers,
3340 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3341 int mirror_num, int async_submit)
3343 struct btrfs_mapping_tree *map_tree;
3344 struct btrfs_device *dev;
3345 struct bio *first_bio = bio;
3346 u64 logical = (u64)bio->bi_sector << 9;
3352 struct btrfs_bio *bbio = NULL;
3354 length = bio->bi_size;
3355 map_tree = &root->fs_info->mapping_tree;
3356 map_length = length;
3358 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
3362 total_devs = bbio->num_stripes;
3363 if (map_length < length) {
3364 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3365 "len %llu\n", (unsigned long long)logical,
3366 (unsigned long long)length,
3367 (unsigned long long)map_length);
3371 bbio->orig_bio = first_bio;
3372 bbio->private = first_bio->bi_private;
3373 bbio->end_io = first_bio->bi_end_io;
3374 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
3376 while (dev_nr < total_devs) {
3377 if (dev_nr < total_devs - 1) {
3378 bio = bio_clone(first_bio, GFP_NOFS);
3383 bio->bi_private = bbio;
3384 bio->bi_end_io = btrfs_end_bio;
3385 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
3386 dev = bbio->stripes[dev_nr].dev;
3387 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3388 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
3389 "(%s id %llu), size=%u\n", rw,
3390 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
3391 dev->name, dev->devid, bio->bi_size);
3392 bio->bi_bdev = dev->bdev;
3394 schedule_bio(root, dev, rw, bio);
3396 submit_bio(rw, bio);
3398 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3399 bio->bi_sector = logical >> 9;
3400 bio_endio(bio, -EIO);
3407 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3410 struct btrfs_device *device;
3411 struct btrfs_fs_devices *cur_devices;
3413 cur_devices = root->fs_info->fs_devices;
3414 while (cur_devices) {
3416 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3417 device = __find_device(&cur_devices->devices,
3422 cur_devices = cur_devices->seed;
3427 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3428 u64 devid, u8 *dev_uuid)
3430 struct btrfs_device *device;
3431 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3433 device = kzalloc(sizeof(*device), GFP_NOFS);
3436 list_add(&device->dev_list,
3437 &fs_devices->devices);
3438 device->dev_root = root->fs_info->dev_root;
3439 device->devid = devid;
3440 device->work.func = pending_bios_fn;
3441 device->fs_devices = fs_devices;
3442 device->missing = 1;
3443 fs_devices->num_devices++;
3444 fs_devices->missing_devices++;
3445 spin_lock_init(&device->io_lock);
3446 INIT_LIST_HEAD(&device->dev_alloc_list);
3447 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3451 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3452 struct extent_buffer *leaf,
3453 struct btrfs_chunk *chunk)
3455 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3456 struct map_lookup *map;
3457 struct extent_map *em;
3461 u8 uuid[BTRFS_UUID_SIZE];
3466 logical = key->offset;
3467 length = btrfs_chunk_length(leaf, chunk);
3469 read_lock(&map_tree->map_tree.lock);
3470 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3471 read_unlock(&map_tree->map_tree.lock);
3473 /* already mapped? */
3474 if (em && em->start <= logical && em->start + em->len > logical) {
3475 free_extent_map(em);
3478 free_extent_map(em);
3481 em = alloc_extent_map();
3484 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3485 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3487 free_extent_map(em);
3491 em->bdev = (struct block_device *)map;
3492 em->start = logical;
3494 em->block_start = 0;
3495 em->block_len = em->len;
3497 map->num_stripes = num_stripes;
3498 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3499 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3500 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3501 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3502 map->type = btrfs_chunk_type(leaf, chunk);
3503 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3504 for (i = 0; i < num_stripes; i++) {
3505 map->stripes[i].physical =
3506 btrfs_stripe_offset_nr(leaf, chunk, i);
3507 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3508 read_extent_buffer(leaf, uuid, (unsigned long)
3509 btrfs_stripe_dev_uuid_nr(chunk, i),
3511 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3513 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3515 free_extent_map(em);
3518 if (!map->stripes[i].dev) {
3519 map->stripes[i].dev =
3520 add_missing_dev(root, devid, uuid);
3521 if (!map->stripes[i].dev) {
3523 free_extent_map(em);
3527 map->stripes[i].dev->in_fs_metadata = 1;
3530 write_lock(&map_tree->map_tree.lock);
3531 ret = add_extent_mapping(&map_tree->map_tree, em);
3532 write_unlock(&map_tree->map_tree.lock);
3534 free_extent_map(em);
3539 static int fill_device_from_item(struct extent_buffer *leaf,
3540 struct btrfs_dev_item *dev_item,
3541 struct btrfs_device *device)
3545 device->devid = btrfs_device_id(leaf, dev_item);
3546 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3547 device->total_bytes = device->disk_total_bytes;
3548 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3549 device->type = btrfs_device_type(leaf, dev_item);
3550 device->io_align = btrfs_device_io_align(leaf, dev_item);
3551 device->io_width = btrfs_device_io_width(leaf, dev_item);
3552 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3554 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3555 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3560 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3562 struct btrfs_fs_devices *fs_devices;
3565 mutex_lock(&uuid_mutex);
3567 fs_devices = root->fs_info->fs_devices->seed;
3568 while (fs_devices) {
3569 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3573 fs_devices = fs_devices->seed;
3576 fs_devices = find_fsid(fsid);
3582 fs_devices = clone_fs_devices(fs_devices);
3583 if (IS_ERR(fs_devices)) {
3584 ret = PTR_ERR(fs_devices);
3588 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3589 root->fs_info->bdev_holder);
3593 if (!fs_devices->seeding) {
3594 __btrfs_close_devices(fs_devices);
3595 free_fs_devices(fs_devices);
3600 fs_devices->seed = root->fs_info->fs_devices->seed;
3601 root->fs_info->fs_devices->seed = fs_devices;
3603 mutex_unlock(&uuid_mutex);
3607 static int read_one_dev(struct btrfs_root *root,
3608 struct extent_buffer *leaf,
3609 struct btrfs_dev_item *dev_item)
3611 struct btrfs_device *device;
3614 u8 fs_uuid[BTRFS_UUID_SIZE];
3615 u8 dev_uuid[BTRFS_UUID_SIZE];
3617 devid = btrfs_device_id(leaf, dev_item);
3618 read_extent_buffer(leaf, dev_uuid,
3619 (unsigned long)btrfs_device_uuid(dev_item),
3621 read_extent_buffer(leaf, fs_uuid,
3622 (unsigned long)btrfs_device_fsid(dev_item),
3625 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3626 ret = open_seed_devices(root, fs_uuid);
3627 if (ret && !btrfs_test_opt(root, DEGRADED))
3631 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3632 if (!device || !device->bdev) {
3633 if (!btrfs_test_opt(root, DEGRADED))
3637 printk(KERN_WARNING "warning devid %llu missing\n",
3638 (unsigned long long)devid);
3639 device = add_missing_dev(root, devid, dev_uuid);
3642 } else if (!device->missing) {
3644 * this happens when a device that was properly setup
3645 * in the device info lists suddenly goes bad.
3646 * device->bdev is NULL, and so we have to set
3647 * device->missing to one here
3649 root->fs_info->fs_devices->missing_devices++;
3650 device->missing = 1;
3654 if (device->fs_devices != root->fs_info->fs_devices) {
3655 BUG_ON(device->writeable);
3656 if (device->generation !=
3657 btrfs_device_generation(leaf, dev_item))
3661 fill_device_from_item(leaf, dev_item, device);
3662 device->dev_root = root->fs_info->dev_root;
3663 device->in_fs_metadata = 1;
3664 if (device->writeable) {
3665 device->fs_devices->total_rw_bytes += device->total_bytes;
3666 spin_lock(&root->fs_info->free_chunk_lock);
3667 root->fs_info->free_chunk_space += device->total_bytes -
3669 spin_unlock(&root->fs_info->free_chunk_lock);
3675 int btrfs_read_sys_array(struct btrfs_root *root)
3677 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3678 struct extent_buffer *sb;
3679 struct btrfs_disk_key *disk_key;
3680 struct btrfs_chunk *chunk;
3682 unsigned long sb_ptr;
3688 struct btrfs_key key;
3690 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3691 BTRFS_SUPER_INFO_SIZE);
3694 btrfs_set_buffer_uptodate(sb);
3695 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3697 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3698 array_size = btrfs_super_sys_array_size(super_copy);
3700 ptr = super_copy->sys_chunk_array;
3701 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3704 while (cur < array_size) {
3705 disk_key = (struct btrfs_disk_key *)ptr;
3706 btrfs_disk_key_to_cpu(&key, disk_key);
3708 len = sizeof(*disk_key); ptr += len;
3712 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3713 chunk = (struct btrfs_chunk *)sb_ptr;
3714 ret = read_one_chunk(root, &key, sb, chunk);
3717 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3718 len = btrfs_chunk_item_size(num_stripes);
3727 free_extent_buffer(sb);
3731 int btrfs_read_chunk_tree(struct btrfs_root *root)
3733 struct btrfs_path *path;
3734 struct extent_buffer *leaf;
3735 struct btrfs_key key;
3736 struct btrfs_key found_key;
3740 root = root->fs_info->chunk_root;
3742 path = btrfs_alloc_path();
3746 /* first we search for all of the device items, and then we
3747 * read in all of the chunk items. This way we can create chunk
3748 * mappings that reference all of the devices that are afound
3750 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3754 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3758 leaf = path->nodes[0];
3759 slot = path->slots[0];
3760 if (slot >= btrfs_header_nritems(leaf)) {
3761 ret = btrfs_next_leaf(root, path);
3768 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3769 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3770 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3772 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3773 struct btrfs_dev_item *dev_item;
3774 dev_item = btrfs_item_ptr(leaf, slot,
3775 struct btrfs_dev_item);
3776 ret = read_one_dev(root, leaf, dev_item);
3780 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3781 struct btrfs_chunk *chunk;
3782 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3783 ret = read_one_chunk(root, &found_key, leaf, chunk);
3789 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3791 btrfs_release_path(path);
3796 btrfs_free_path(path);