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 int 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);
83 static noinline struct btrfs_device *__find_device(struct list_head *head,
86 struct btrfs_device *dev;
88 list_for_each_entry(dev, head, dev_list) {
89 if (dev->devid == devid &&
90 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
97 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
99 struct btrfs_fs_devices *fs_devices;
101 list_for_each_entry(fs_devices, &fs_uuids, list) {
102 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
108 static void requeue_list(struct btrfs_pending_bios *pending_bios,
109 struct bio *head, struct bio *tail)
112 struct bio *old_head;
114 old_head = pending_bios->head;
115 pending_bios->head = head;
116 if (pending_bios->tail)
117 tail->bi_next = old_head;
119 pending_bios->tail = tail;
123 * we try to collect pending bios for a device so we don't get a large
124 * number of procs sending bios down to the same device. This greatly
125 * improves the schedulers ability to collect and merge the bios.
127 * But, it also turns into a long list of bios to process and that is sure
128 * to eventually make the worker thread block. The solution here is to
129 * make some progress and then put this work struct back at the end of
130 * the list if the block device is congested. This way, multiple devices
131 * can make progress from a single worker thread.
133 static noinline int run_scheduled_bios(struct btrfs_device *device)
136 struct backing_dev_info *bdi;
137 struct btrfs_fs_info *fs_info;
138 struct btrfs_pending_bios *pending_bios;
142 unsigned long num_run;
143 unsigned long batch_run = 0;
145 unsigned long last_waited = 0;
147 int sync_pending = 0;
148 struct blk_plug plug;
151 * this function runs all the bios we've collected for
152 * a particular device. We don't want to wander off to
153 * another device without first sending all of these down.
154 * So, setup a plug here and finish it off before we return
156 blk_start_plug(&plug);
158 bdi = blk_get_backing_dev_info(device->bdev);
159 fs_info = device->dev_root->fs_info;
160 limit = btrfs_async_submit_limit(fs_info);
161 limit = limit * 2 / 3;
164 spin_lock(&device->io_lock);
169 /* take all the bios off the list at once and process them
170 * later on (without the lock held). But, remember the
171 * tail and other pointers so the bios can be properly reinserted
172 * into the list if we hit congestion
174 if (!force_reg && device->pending_sync_bios.head) {
175 pending_bios = &device->pending_sync_bios;
178 pending_bios = &device->pending_bios;
182 pending = pending_bios->head;
183 tail = pending_bios->tail;
184 WARN_ON(pending && !tail);
187 * if pending was null this time around, no bios need processing
188 * at all and we can stop. Otherwise it'll loop back up again
189 * and do an additional check so no bios are missed.
191 * device->running_pending is used to synchronize with the
194 if (device->pending_sync_bios.head == NULL &&
195 device->pending_bios.head == NULL) {
197 device->running_pending = 0;
200 device->running_pending = 1;
203 pending_bios->head = NULL;
204 pending_bios->tail = NULL;
206 spin_unlock(&device->io_lock);
211 /* we want to work on both lists, but do more bios on the
212 * sync list than the regular list
215 pending_bios != &device->pending_sync_bios &&
216 device->pending_sync_bios.head) ||
217 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
218 device->pending_bios.head)) {
219 spin_lock(&device->io_lock);
220 requeue_list(pending_bios, pending, tail);
225 pending = pending->bi_next;
227 atomic_dec(&fs_info->nr_async_bios);
229 if (atomic_read(&fs_info->nr_async_bios) < limit &&
230 waitqueue_active(&fs_info->async_submit_wait))
231 wake_up(&fs_info->async_submit_wait);
233 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
236 * if we're doing the sync list, record that our
237 * plug has some sync requests on it
239 * If we're doing the regular list and there are
240 * sync requests sitting around, unplug before
243 if (pending_bios == &device->pending_sync_bios) {
245 } else if (sync_pending) {
246 blk_finish_plug(&plug);
247 blk_start_plug(&plug);
251 btrfsic_submit_bio(cur->bi_rw, cur);
258 * we made progress, there is more work to do and the bdi
259 * is now congested. Back off and let other work structs
262 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
263 fs_info->fs_devices->open_devices > 1) {
264 struct io_context *ioc;
266 ioc = current->io_context;
269 * the main goal here is that we don't want to
270 * block if we're going to be able to submit
271 * more requests without blocking.
273 * This code does two great things, it pokes into
274 * the elevator code from a filesystem _and_
275 * it makes assumptions about how batching works.
277 if (ioc && ioc->nr_batch_requests > 0 &&
278 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
280 ioc->last_waited == last_waited)) {
282 * we want to go through our batch of
283 * requests and stop. So, we copy out
284 * the ioc->last_waited time and test
285 * against it before looping
287 last_waited = ioc->last_waited;
292 spin_lock(&device->io_lock);
293 requeue_list(pending_bios, pending, tail);
294 device->running_pending = 1;
296 spin_unlock(&device->io_lock);
297 btrfs_requeue_work(&device->work);
300 /* unplug every 64 requests just for good measure */
301 if (batch_run % 64 == 0) {
302 blk_finish_plug(&plug);
303 blk_start_plug(&plug);
312 spin_lock(&device->io_lock);
313 if (device->pending_bios.head || device->pending_sync_bios.head)
315 spin_unlock(&device->io_lock);
318 blk_finish_plug(&plug);
322 static void pending_bios_fn(struct btrfs_work *work)
324 struct btrfs_device *device;
326 device = container_of(work, struct btrfs_device, work);
327 run_scheduled_bios(device);
330 static noinline int device_list_add(const char *path,
331 struct btrfs_super_block *disk_super,
332 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
334 struct btrfs_device *device;
335 struct btrfs_fs_devices *fs_devices;
336 u64 found_transid = btrfs_super_generation(disk_super);
339 fs_devices = find_fsid(disk_super->fsid);
341 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
344 INIT_LIST_HEAD(&fs_devices->devices);
345 INIT_LIST_HEAD(&fs_devices->alloc_list);
346 list_add(&fs_devices->list, &fs_uuids);
347 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
348 fs_devices->latest_devid = devid;
349 fs_devices->latest_trans = found_transid;
350 mutex_init(&fs_devices->device_list_mutex);
353 device = __find_device(&fs_devices->devices, devid,
354 disk_super->dev_item.uuid);
357 if (fs_devices->opened)
360 device = kzalloc(sizeof(*device), GFP_NOFS);
362 /* we can safely leave the fs_devices entry around */
365 device->devid = devid;
366 device->work.func = pending_bios_fn;
367 memcpy(device->uuid, disk_super->dev_item.uuid,
369 spin_lock_init(&device->io_lock);
370 device->name = kstrdup(path, GFP_NOFS);
375 INIT_LIST_HEAD(&device->dev_alloc_list);
377 /* init readahead state */
378 spin_lock_init(&device->reada_lock);
379 device->reada_curr_zone = NULL;
380 atomic_set(&device->reada_in_flight, 0);
381 device->reada_next = 0;
382 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
383 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
385 mutex_lock(&fs_devices->device_list_mutex);
386 list_add_rcu(&device->dev_list, &fs_devices->devices);
387 mutex_unlock(&fs_devices->device_list_mutex);
389 device->fs_devices = fs_devices;
390 fs_devices->num_devices++;
391 } else if (!device->name || strcmp(device->name, path)) {
392 name = kstrdup(path, GFP_NOFS);
397 if (device->missing) {
398 fs_devices->missing_devices--;
403 if (found_transid > fs_devices->latest_trans) {
404 fs_devices->latest_devid = devid;
405 fs_devices->latest_trans = found_transid;
407 *fs_devices_ret = fs_devices;
411 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
413 struct btrfs_fs_devices *fs_devices;
414 struct btrfs_device *device;
415 struct btrfs_device *orig_dev;
417 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
419 return ERR_PTR(-ENOMEM);
421 INIT_LIST_HEAD(&fs_devices->devices);
422 INIT_LIST_HEAD(&fs_devices->alloc_list);
423 INIT_LIST_HEAD(&fs_devices->list);
424 mutex_init(&fs_devices->device_list_mutex);
425 fs_devices->latest_devid = orig->latest_devid;
426 fs_devices->latest_trans = orig->latest_trans;
427 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
429 /* We have held the volume lock, it is safe to get the devices. */
430 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
431 device = kzalloc(sizeof(*device), GFP_NOFS);
435 device->name = kstrdup(orig_dev->name, GFP_NOFS);
441 device->devid = orig_dev->devid;
442 device->work.func = pending_bios_fn;
443 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
444 spin_lock_init(&device->io_lock);
445 INIT_LIST_HEAD(&device->dev_list);
446 INIT_LIST_HEAD(&device->dev_alloc_list);
448 list_add(&device->dev_list, &fs_devices->devices);
449 device->fs_devices = fs_devices;
450 fs_devices->num_devices++;
454 free_fs_devices(fs_devices);
455 return ERR_PTR(-ENOMEM);
458 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
460 struct btrfs_device *device, *next;
462 struct block_device *latest_bdev = NULL;
463 u64 latest_devid = 0;
464 u64 latest_transid = 0;
466 mutex_lock(&uuid_mutex);
468 /* This is the initialized path, it is safe to release the devices. */
469 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
470 if (device->in_fs_metadata) {
471 if (!latest_transid ||
472 device->generation > latest_transid) {
473 latest_devid = device->devid;
474 latest_transid = device->generation;
475 latest_bdev = device->bdev;
481 blkdev_put(device->bdev, device->mode);
483 fs_devices->open_devices--;
485 if (device->writeable) {
486 list_del_init(&device->dev_alloc_list);
487 device->writeable = 0;
488 fs_devices->rw_devices--;
490 list_del_init(&device->dev_list);
491 fs_devices->num_devices--;
496 if (fs_devices->seed) {
497 fs_devices = fs_devices->seed;
501 fs_devices->latest_bdev = latest_bdev;
502 fs_devices->latest_devid = latest_devid;
503 fs_devices->latest_trans = latest_transid;
505 mutex_unlock(&uuid_mutex);
509 static void __free_device(struct work_struct *work)
511 struct btrfs_device *device;
513 device = container_of(work, struct btrfs_device, rcu_work);
516 blkdev_put(device->bdev, device->mode);
522 static void free_device(struct rcu_head *head)
524 struct btrfs_device *device;
526 device = container_of(head, struct btrfs_device, rcu);
528 INIT_WORK(&device->rcu_work, __free_device);
529 schedule_work(&device->rcu_work);
532 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
534 struct btrfs_device *device;
536 if (--fs_devices->opened > 0)
539 mutex_lock(&fs_devices->device_list_mutex);
540 list_for_each_entry(device, &fs_devices->devices, dev_list) {
541 struct btrfs_device *new_device;
544 fs_devices->open_devices--;
546 if (device->writeable) {
547 list_del_init(&device->dev_alloc_list);
548 fs_devices->rw_devices--;
551 if (device->can_discard)
552 fs_devices->num_can_discard--;
554 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
556 memcpy(new_device, device, sizeof(*new_device));
557 new_device->name = kstrdup(device->name, GFP_NOFS);
558 BUG_ON(device->name && !new_device->name);
559 new_device->bdev = NULL;
560 new_device->writeable = 0;
561 new_device->in_fs_metadata = 0;
562 new_device->can_discard = 0;
563 list_replace_rcu(&device->dev_list, &new_device->dev_list);
565 call_rcu(&device->rcu, free_device);
567 mutex_unlock(&fs_devices->device_list_mutex);
569 WARN_ON(fs_devices->open_devices);
570 WARN_ON(fs_devices->rw_devices);
571 fs_devices->opened = 0;
572 fs_devices->seeding = 0;
577 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
579 struct btrfs_fs_devices *seed_devices = NULL;
582 mutex_lock(&uuid_mutex);
583 ret = __btrfs_close_devices(fs_devices);
584 if (!fs_devices->opened) {
585 seed_devices = fs_devices->seed;
586 fs_devices->seed = NULL;
588 mutex_unlock(&uuid_mutex);
590 while (seed_devices) {
591 fs_devices = seed_devices;
592 seed_devices = fs_devices->seed;
593 __btrfs_close_devices(fs_devices);
594 free_fs_devices(fs_devices);
599 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
600 fmode_t flags, void *holder)
602 struct request_queue *q;
603 struct block_device *bdev;
604 struct list_head *head = &fs_devices->devices;
605 struct btrfs_device *device;
606 struct block_device *latest_bdev = NULL;
607 struct buffer_head *bh;
608 struct btrfs_super_block *disk_super;
609 u64 latest_devid = 0;
610 u64 latest_transid = 0;
617 list_for_each_entry(device, head, dev_list) {
623 bdev = blkdev_get_by_path(device->name, flags, holder);
625 printk(KERN_INFO "open %s failed\n", device->name);
628 set_blocksize(bdev, 4096);
630 bh = btrfs_read_dev_super(bdev);
634 disk_super = (struct btrfs_super_block *)bh->b_data;
635 devid = btrfs_stack_device_id(&disk_super->dev_item);
636 if (devid != device->devid)
639 if (memcmp(device->uuid, disk_super->dev_item.uuid,
643 device->generation = btrfs_super_generation(disk_super);
644 if (!latest_transid || device->generation > latest_transid) {
645 latest_devid = devid;
646 latest_transid = device->generation;
650 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
651 device->writeable = 0;
653 device->writeable = !bdev_read_only(bdev);
657 q = bdev_get_queue(bdev);
658 if (blk_queue_discard(q)) {
659 device->can_discard = 1;
660 fs_devices->num_can_discard++;
664 device->in_fs_metadata = 0;
665 device->mode = flags;
667 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
668 fs_devices->rotating = 1;
670 fs_devices->open_devices++;
671 if (device->writeable) {
672 fs_devices->rw_devices++;
673 list_add(&device->dev_alloc_list,
674 &fs_devices->alloc_list);
682 blkdev_put(bdev, flags);
686 if (fs_devices->open_devices == 0) {
690 fs_devices->seeding = seeding;
691 fs_devices->opened = 1;
692 fs_devices->latest_bdev = latest_bdev;
693 fs_devices->latest_devid = latest_devid;
694 fs_devices->latest_trans = latest_transid;
695 fs_devices->total_rw_bytes = 0;
700 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
701 fmode_t flags, void *holder)
705 mutex_lock(&uuid_mutex);
706 if (fs_devices->opened) {
707 fs_devices->opened++;
710 ret = __btrfs_open_devices(fs_devices, flags, holder);
712 mutex_unlock(&uuid_mutex);
716 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
717 struct btrfs_fs_devices **fs_devices_ret)
719 struct btrfs_super_block *disk_super;
720 struct block_device *bdev;
721 struct buffer_head *bh;
727 bdev = blkdev_get_by_path(path, flags, holder);
734 mutex_lock(&uuid_mutex);
735 ret = set_blocksize(bdev, 4096);
738 bh = btrfs_read_dev_super(bdev);
743 disk_super = (struct btrfs_super_block *)bh->b_data;
744 devid = btrfs_stack_device_id(&disk_super->dev_item);
745 transid = btrfs_super_generation(disk_super);
746 if (disk_super->label[0])
747 printk(KERN_INFO "device label %s ", disk_super->label);
749 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
750 printk(KERN_CONT "devid %llu transid %llu %s\n",
751 (unsigned long long)devid, (unsigned long long)transid, path);
752 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
756 mutex_unlock(&uuid_mutex);
757 blkdev_put(bdev, flags);
762 /* helper to account the used device space in the range */
763 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
764 u64 end, u64 *length)
766 struct btrfs_key key;
767 struct btrfs_root *root = device->dev_root;
768 struct btrfs_dev_extent *dev_extent;
769 struct btrfs_path *path;
773 struct extent_buffer *l;
777 if (start >= device->total_bytes)
780 path = btrfs_alloc_path();
785 key.objectid = device->devid;
787 key.type = BTRFS_DEV_EXTENT_KEY;
789 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
793 ret = btrfs_previous_item(root, path, key.objectid, key.type);
800 slot = path->slots[0];
801 if (slot >= btrfs_header_nritems(l)) {
802 ret = btrfs_next_leaf(root, path);
810 btrfs_item_key_to_cpu(l, &key, slot);
812 if (key.objectid < device->devid)
815 if (key.objectid > device->devid)
818 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
821 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
822 extent_end = key.offset + btrfs_dev_extent_length(l,
824 if (key.offset <= start && extent_end > end) {
825 *length = end - start + 1;
827 } else if (key.offset <= start && extent_end > start)
828 *length += extent_end - start;
829 else if (key.offset > start && extent_end <= end)
830 *length += extent_end - key.offset;
831 else if (key.offset > start && key.offset <= end) {
832 *length += end - key.offset + 1;
834 } else if (key.offset > end)
842 btrfs_free_path(path);
847 * find_free_dev_extent - find free space in the specified device
848 * @device: the device which we search the free space in
849 * @num_bytes: the size of the free space that we need
850 * @start: store the start of the free space.
851 * @len: the size of the free space. that we find, or the size of the max
852 * free space if we don't find suitable free space
854 * this uses a pretty simple search, the expectation is that it is
855 * called very infrequently and that a given device has a small number
858 * @start is used to store the start of the free space if we find. But if we
859 * don't find suitable free space, it will be used to store the start position
860 * of the max free space.
862 * @len is used to store the size of the free space that we find.
863 * But if we don't find suitable free space, it is used to store the size of
864 * the max free space.
866 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
867 u64 *start, u64 *len)
869 struct btrfs_key key;
870 struct btrfs_root *root = device->dev_root;
871 struct btrfs_dev_extent *dev_extent;
872 struct btrfs_path *path;
878 u64 search_end = device->total_bytes;
881 struct extent_buffer *l;
883 /* FIXME use last free of some kind */
885 /* we don't want to overwrite the superblock on the drive,
886 * so we make sure to start at an offset of at least 1MB
888 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
890 max_hole_start = search_start;
894 if (search_start >= search_end) {
899 path = btrfs_alloc_path();
906 key.objectid = device->devid;
907 key.offset = search_start;
908 key.type = BTRFS_DEV_EXTENT_KEY;
910 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
914 ret = btrfs_previous_item(root, path, key.objectid, key.type);
921 slot = path->slots[0];
922 if (slot >= btrfs_header_nritems(l)) {
923 ret = btrfs_next_leaf(root, path);
931 btrfs_item_key_to_cpu(l, &key, slot);
933 if (key.objectid < device->devid)
936 if (key.objectid > device->devid)
939 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
942 if (key.offset > search_start) {
943 hole_size = key.offset - search_start;
945 if (hole_size > max_hole_size) {
946 max_hole_start = search_start;
947 max_hole_size = hole_size;
951 * If this free space is greater than which we need,
952 * it must be the max free space that we have found
953 * until now, so max_hole_start must point to the start
954 * of this free space and the length of this free space
955 * is stored in max_hole_size. Thus, we return
956 * max_hole_start and max_hole_size and go back to the
959 if (hole_size >= num_bytes) {
965 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
966 extent_end = key.offset + btrfs_dev_extent_length(l,
968 if (extent_end > search_start)
969 search_start = extent_end;
976 * At this point, search_start should be the end of
977 * allocated dev extents, and when shrinking the device,
978 * search_end may be smaller than search_start.
980 if (search_end > search_start)
981 hole_size = search_end - search_start;
983 if (hole_size > max_hole_size) {
984 max_hole_start = search_start;
985 max_hole_size = hole_size;
989 if (hole_size < num_bytes)
995 btrfs_free_path(path);
997 *start = max_hole_start;
999 *len = max_hole_size;
1003 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1004 struct btrfs_device *device,
1008 struct btrfs_path *path;
1009 struct btrfs_root *root = device->dev_root;
1010 struct btrfs_key key;
1011 struct btrfs_key found_key;
1012 struct extent_buffer *leaf = NULL;
1013 struct btrfs_dev_extent *extent = NULL;
1015 path = btrfs_alloc_path();
1019 key.objectid = device->devid;
1021 key.type = BTRFS_DEV_EXTENT_KEY;
1023 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1025 ret = btrfs_previous_item(root, path, key.objectid,
1026 BTRFS_DEV_EXTENT_KEY);
1029 leaf = path->nodes[0];
1030 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1031 extent = btrfs_item_ptr(leaf, path->slots[0],
1032 struct btrfs_dev_extent);
1033 BUG_ON(found_key.offset > start || found_key.offset +
1034 btrfs_dev_extent_length(leaf, extent) < start);
1036 btrfs_release_path(path);
1038 } else if (ret == 0) {
1039 leaf = path->nodes[0];
1040 extent = btrfs_item_ptr(leaf, path->slots[0],
1041 struct btrfs_dev_extent);
1045 if (device->bytes_used > 0) {
1046 u64 len = btrfs_dev_extent_length(leaf, extent);
1047 device->bytes_used -= len;
1048 spin_lock(&root->fs_info->free_chunk_lock);
1049 root->fs_info->free_chunk_space += len;
1050 spin_unlock(&root->fs_info->free_chunk_lock);
1052 ret = btrfs_del_item(trans, root, path);
1055 btrfs_free_path(path);
1059 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1060 struct btrfs_device *device,
1061 u64 chunk_tree, u64 chunk_objectid,
1062 u64 chunk_offset, u64 start, u64 num_bytes)
1065 struct btrfs_path *path;
1066 struct btrfs_root *root = device->dev_root;
1067 struct btrfs_dev_extent *extent;
1068 struct extent_buffer *leaf;
1069 struct btrfs_key key;
1071 WARN_ON(!device->in_fs_metadata);
1072 path = btrfs_alloc_path();
1076 key.objectid = device->devid;
1078 key.type = BTRFS_DEV_EXTENT_KEY;
1079 ret = btrfs_insert_empty_item(trans, root, path, &key,
1083 leaf = path->nodes[0];
1084 extent = btrfs_item_ptr(leaf, path->slots[0],
1085 struct btrfs_dev_extent);
1086 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1087 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1088 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1090 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1091 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1094 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1095 btrfs_mark_buffer_dirty(leaf);
1096 btrfs_free_path(path);
1100 static noinline int find_next_chunk(struct btrfs_root *root,
1101 u64 objectid, u64 *offset)
1103 struct btrfs_path *path;
1105 struct btrfs_key key;
1106 struct btrfs_chunk *chunk;
1107 struct btrfs_key found_key;
1109 path = btrfs_alloc_path();
1113 key.objectid = objectid;
1114 key.offset = (u64)-1;
1115 key.type = BTRFS_CHUNK_ITEM_KEY;
1117 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1123 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1127 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1129 if (found_key.objectid != objectid)
1132 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1133 struct btrfs_chunk);
1134 *offset = found_key.offset +
1135 btrfs_chunk_length(path->nodes[0], chunk);
1140 btrfs_free_path(path);
1144 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1147 struct btrfs_key key;
1148 struct btrfs_key found_key;
1149 struct btrfs_path *path;
1151 root = root->fs_info->chunk_root;
1153 path = btrfs_alloc_path();
1157 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1158 key.type = BTRFS_DEV_ITEM_KEY;
1159 key.offset = (u64)-1;
1161 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1167 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1168 BTRFS_DEV_ITEM_KEY);
1172 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1174 *objectid = found_key.offset + 1;
1178 btrfs_free_path(path);
1183 * the device information is stored in the chunk root
1184 * the btrfs_device struct should be fully filled in
1186 int btrfs_add_device(struct btrfs_trans_handle *trans,
1187 struct btrfs_root *root,
1188 struct btrfs_device *device)
1191 struct btrfs_path *path;
1192 struct btrfs_dev_item *dev_item;
1193 struct extent_buffer *leaf;
1194 struct btrfs_key key;
1197 root = root->fs_info->chunk_root;
1199 path = btrfs_alloc_path();
1203 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1204 key.type = BTRFS_DEV_ITEM_KEY;
1205 key.offset = device->devid;
1207 ret = btrfs_insert_empty_item(trans, root, path, &key,
1212 leaf = path->nodes[0];
1213 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1215 btrfs_set_device_id(leaf, dev_item, device->devid);
1216 btrfs_set_device_generation(leaf, dev_item, 0);
1217 btrfs_set_device_type(leaf, dev_item, device->type);
1218 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1219 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1220 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1221 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1222 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1223 btrfs_set_device_group(leaf, dev_item, 0);
1224 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1225 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1226 btrfs_set_device_start_offset(leaf, dev_item, 0);
1228 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1229 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1230 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1231 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1232 btrfs_mark_buffer_dirty(leaf);
1236 btrfs_free_path(path);
1240 static int btrfs_rm_dev_item(struct btrfs_root *root,
1241 struct btrfs_device *device)
1244 struct btrfs_path *path;
1245 struct btrfs_key key;
1246 struct btrfs_trans_handle *trans;
1248 root = root->fs_info->chunk_root;
1250 path = btrfs_alloc_path();
1254 trans = btrfs_start_transaction(root, 0);
1255 if (IS_ERR(trans)) {
1256 btrfs_free_path(path);
1257 return PTR_ERR(trans);
1259 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1260 key.type = BTRFS_DEV_ITEM_KEY;
1261 key.offset = device->devid;
1264 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1273 ret = btrfs_del_item(trans, root, path);
1277 btrfs_free_path(path);
1278 unlock_chunks(root);
1279 btrfs_commit_transaction(trans, root);
1283 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1285 struct btrfs_device *device;
1286 struct btrfs_device *next_device;
1287 struct block_device *bdev;
1288 struct buffer_head *bh = NULL;
1289 struct btrfs_super_block *disk_super;
1290 struct btrfs_fs_devices *cur_devices;
1296 bool clear_super = false;
1298 mutex_lock(&uuid_mutex);
1300 all_avail = root->fs_info->avail_data_alloc_bits |
1301 root->fs_info->avail_system_alloc_bits |
1302 root->fs_info->avail_metadata_alloc_bits;
1304 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1305 root->fs_info->fs_devices->num_devices <= 4) {
1306 printk(KERN_ERR "btrfs: unable to go below four devices "
1312 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1313 root->fs_info->fs_devices->num_devices <= 2) {
1314 printk(KERN_ERR "btrfs: unable to go below two "
1315 "devices on raid1\n");
1320 if (strcmp(device_path, "missing") == 0) {
1321 struct list_head *devices;
1322 struct btrfs_device *tmp;
1325 devices = &root->fs_info->fs_devices->devices;
1327 * It is safe to read the devices since the volume_mutex
1330 list_for_each_entry(tmp, devices, dev_list) {
1331 if (tmp->in_fs_metadata && !tmp->bdev) {
1340 printk(KERN_ERR "btrfs: no missing devices found to "
1345 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1346 root->fs_info->bdev_holder);
1348 ret = PTR_ERR(bdev);
1352 set_blocksize(bdev, 4096);
1353 bh = btrfs_read_dev_super(bdev);
1358 disk_super = (struct btrfs_super_block *)bh->b_data;
1359 devid = btrfs_stack_device_id(&disk_super->dev_item);
1360 dev_uuid = disk_super->dev_item.uuid;
1361 device = btrfs_find_device(root, devid, dev_uuid,
1369 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1370 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1376 if (device->writeable) {
1378 list_del_init(&device->dev_alloc_list);
1379 unlock_chunks(root);
1380 root->fs_info->fs_devices->rw_devices--;
1384 ret = btrfs_shrink_device(device, 0);
1388 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1392 spin_lock(&root->fs_info->free_chunk_lock);
1393 root->fs_info->free_chunk_space = device->total_bytes -
1395 spin_unlock(&root->fs_info->free_chunk_lock);
1397 device->in_fs_metadata = 0;
1398 btrfs_scrub_cancel_dev(root, device);
1401 * the device list mutex makes sure that we don't change
1402 * the device list while someone else is writing out all
1403 * the device supers.
1406 cur_devices = device->fs_devices;
1407 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1408 list_del_rcu(&device->dev_list);
1410 device->fs_devices->num_devices--;
1412 if (device->missing)
1413 root->fs_info->fs_devices->missing_devices--;
1415 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1416 struct btrfs_device, dev_list);
1417 if (device->bdev == root->fs_info->sb->s_bdev)
1418 root->fs_info->sb->s_bdev = next_device->bdev;
1419 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1420 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1423 device->fs_devices->open_devices--;
1425 call_rcu(&device->rcu, free_device);
1426 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1428 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1429 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1431 if (cur_devices->open_devices == 0) {
1432 struct btrfs_fs_devices *fs_devices;
1433 fs_devices = root->fs_info->fs_devices;
1434 while (fs_devices) {
1435 if (fs_devices->seed == cur_devices)
1437 fs_devices = fs_devices->seed;
1439 fs_devices->seed = cur_devices->seed;
1440 cur_devices->seed = NULL;
1442 __btrfs_close_devices(cur_devices);
1443 unlock_chunks(root);
1444 free_fs_devices(cur_devices);
1448 * at this point, the device is zero sized. We want to
1449 * remove it from the devices list and zero out the old super
1452 /* make sure this device isn't detected as part of
1455 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1456 set_buffer_dirty(bh);
1457 sync_dirty_buffer(bh);
1466 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1468 mutex_unlock(&uuid_mutex);
1471 if (device->writeable) {
1473 list_add(&device->dev_alloc_list,
1474 &root->fs_info->fs_devices->alloc_list);
1475 unlock_chunks(root);
1476 root->fs_info->fs_devices->rw_devices++;
1482 * does all the dirty work required for changing file system's UUID.
1484 static int btrfs_prepare_sprout(struct btrfs_root *root)
1486 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1487 struct btrfs_fs_devices *old_devices;
1488 struct btrfs_fs_devices *seed_devices;
1489 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1490 struct btrfs_device *device;
1493 BUG_ON(!mutex_is_locked(&uuid_mutex));
1494 if (!fs_devices->seeding)
1497 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1501 old_devices = clone_fs_devices(fs_devices);
1502 if (IS_ERR(old_devices)) {
1503 kfree(seed_devices);
1504 return PTR_ERR(old_devices);
1507 list_add(&old_devices->list, &fs_uuids);
1509 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1510 seed_devices->opened = 1;
1511 INIT_LIST_HEAD(&seed_devices->devices);
1512 INIT_LIST_HEAD(&seed_devices->alloc_list);
1513 mutex_init(&seed_devices->device_list_mutex);
1515 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1516 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1518 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1520 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1521 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1522 device->fs_devices = seed_devices;
1525 fs_devices->seeding = 0;
1526 fs_devices->num_devices = 0;
1527 fs_devices->open_devices = 0;
1528 fs_devices->seed = seed_devices;
1530 generate_random_uuid(fs_devices->fsid);
1531 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1532 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1533 super_flags = btrfs_super_flags(disk_super) &
1534 ~BTRFS_SUPER_FLAG_SEEDING;
1535 btrfs_set_super_flags(disk_super, super_flags);
1541 * strore the expected generation for seed devices in device items.
1543 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1544 struct btrfs_root *root)
1546 struct btrfs_path *path;
1547 struct extent_buffer *leaf;
1548 struct btrfs_dev_item *dev_item;
1549 struct btrfs_device *device;
1550 struct btrfs_key key;
1551 u8 fs_uuid[BTRFS_UUID_SIZE];
1552 u8 dev_uuid[BTRFS_UUID_SIZE];
1556 path = btrfs_alloc_path();
1560 root = root->fs_info->chunk_root;
1561 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1563 key.type = BTRFS_DEV_ITEM_KEY;
1566 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1570 leaf = path->nodes[0];
1572 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1573 ret = btrfs_next_leaf(root, path);
1578 leaf = path->nodes[0];
1579 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1580 btrfs_release_path(path);
1584 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1585 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1586 key.type != BTRFS_DEV_ITEM_KEY)
1589 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1590 struct btrfs_dev_item);
1591 devid = btrfs_device_id(leaf, dev_item);
1592 read_extent_buffer(leaf, dev_uuid,
1593 (unsigned long)btrfs_device_uuid(dev_item),
1595 read_extent_buffer(leaf, fs_uuid,
1596 (unsigned long)btrfs_device_fsid(dev_item),
1598 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1601 if (device->fs_devices->seeding) {
1602 btrfs_set_device_generation(leaf, dev_item,
1603 device->generation);
1604 btrfs_mark_buffer_dirty(leaf);
1612 btrfs_free_path(path);
1616 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1618 struct request_queue *q;
1619 struct btrfs_trans_handle *trans;
1620 struct btrfs_device *device;
1621 struct block_device *bdev;
1622 struct list_head *devices;
1623 struct super_block *sb = root->fs_info->sb;
1625 int seeding_dev = 0;
1628 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1631 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1632 root->fs_info->bdev_holder);
1634 return PTR_ERR(bdev);
1636 if (root->fs_info->fs_devices->seeding) {
1638 down_write(&sb->s_umount);
1639 mutex_lock(&uuid_mutex);
1642 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1644 devices = &root->fs_info->fs_devices->devices;
1646 * we have the volume lock, so we don't need the extra
1647 * device list mutex while reading the list here.
1649 list_for_each_entry(device, devices, dev_list) {
1650 if (device->bdev == bdev) {
1656 device = kzalloc(sizeof(*device), GFP_NOFS);
1658 /* we can safely leave the fs_devices entry around */
1663 device->name = kstrdup(device_path, GFP_NOFS);
1664 if (!device->name) {
1670 ret = find_next_devid(root, &device->devid);
1672 kfree(device->name);
1677 trans = btrfs_start_transaction(root, 0);
1678 if (IS_ERR(trans)) {
1679 kfree(device->name);
1681 ret = PTR_ERR(trans);
1687 q = bdev_get_queue(bdev);
1688 if (blk_queue_discard(q))
1689 device->can_discard = 1;
1690 device->writeable = 1;
1691 device->work.func = pending_bios_fn;
1692 generate_random_uuid(device->uuid);
1693 spin_lock_init(&device->io_lock);
1694 device->generation = trans->transid;
1695 device->io_width = root->sectorsize;
1696 device->io_align = root->sectorsize;
1697 device->sector_size = root->sectorsize;
1698 device->total_bytes = i_size_read(bdev->bd_inode);
1699 device->disk_total_bytes = device->total_bytes;
1700 device->dev_root = root->fs_info->dev_root;
1701 device->bdev = bdev;
1702 device->in_fs_metadata = 1;
1703 device->mode = FMODE_EXCL;
1704 set_blocksize(device->bdev, 4096);
1707 sb->s_flags &= ~MS_RDONLY;
1708 ret = btrfs_prepare_sprout(root);
1712 device->fs_devices = root->fs_info->fs_devices;
1715 * we don't want write_supers to jump in here with our device
1718 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1719 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1720 list_add(&device->dev_alloc_list,
1721 &root->fs_info->fs_devices->alloc_list);
1722 root->fs_info->fs_devices->num_devices++;
1723 root->fs_info->fs_devices->open_devices++;
1724 root->fs_info->fs_devices->rw_devices++;
1725 if (device->can_discard)
1726 root->fs_info->fs_devices->num_can_discard++;
1727 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1729 spin_lock(&root->fs_info->free_chunk_lock);
1730 root->fs_info->free_chunk_space += device->total_bytes;
1731 spin_unlock(&root->fs_info->free_chunk_lock);
1733 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1734 root->fs_info->fs_devices->rotating = 1;
1736 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1737 btrfs_set_super_total_bytes(root->fs_info->super_copy,
1738 total_bytes + device->total_bytes);
1740 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1741 btrfs_set_super_num_devices(root->fs_info->super_copy,
1743 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1746 ret = init_first_rw_device(trans, root, device);
1748 ret = btrfs_finish_sprout(trans, root);
1751 ret = btrfs_add_device(trans, root, device);
1755 * we've got more storage, clear any full flags on the space
1758 btrfs_clear_space_info_full(root->fs_info);
1760 unlock_chunks(root);
1761 btrfs_commit_transaction(trans, root);
1764 mutex_unlock(&uuid_mutex);
1765 up_write(&sb->s_umount);
1767 ret = btrfs_relocate_sys_chunks(root);
1773 blkdev_put(bdev, FMODE_EXCL);
1775 mutex_unlock(&uuid_mutex);
1776 up_write(&sb->s_umount);
1781 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1782 struct btrfs_device *device)
1785 struct btrfs_path *path;
1786 struct btrfs_root *root;
1787 struct btrfs_dev_item *dev_item;
1788 struct extent_buffer *leaf;
1789 struct btrfs_key key;
1791 root = device->dev_root->fs_info->chunk_root;
1793 path = btrfs_alloc_path();
1797 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1798 key.type = BTRFS_DEV_ITEM_KEY;
1799 key.offset = device->devid;
1801 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1810 leaf = path->nodes[0];
1811 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1813 btrfs_set_device_id(leaf, dev_item, device->devid);
1814 btrfs_set_device_type(leaf, dev_item, device->type);
1815 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1816 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1817 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1818 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1819 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1820 btrfs_mark_buffer_dirty(leaf);
1823 btrfs_free_path(path);
1827 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1828 struct btrfs_device *device, u64 new_size)
1830 struct btrfs_super_block *super_copy =
1831 device->dev_root->fs_info->super_copy;
1832 u64 old_total = btrfs_super_total_bytes(super_copy);
1833 u64 diff = new_size - device->total_bytes;
1835 if (!device->writeable)
1837 if (new_size <= device->total_bytes)
1840 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1841 device->fs_devices->total_rw_bytes += diff;
1843 device->total_bytes = new_size;
1844 device->disk_total_bytes = new_size;
1845 btrfs_clear_space_info_full(device->dev_root->fs_info);
1847 return btrfs_update_device(trans, device);
1850 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1851 struct btrfs_device *device, u64 new_size)
1854 lock_chunks(device->dev_root);
1855 ret = __btrfs_grow_device(trans, device, new_size);
1856 unlock_chunks(device->dev_root);
1860 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1861 struct btrfs_root *root,
1862 u64 chunk_tree, u64 chunk_objectid,
1866 struct btrfs_path *path;
1867 struct btrfs_key key;
1869 root = root->fs_info->chunk_root;
1870 path = btrfs_alloc_path();
1874 key.objectid = chunk_objectid;
1875 key.offset = chunk_offset;
1876 key.type = BTRFS_CHUNK_ITEM_KEY;
1878 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1881 ret = btrfs_del_item(trans, root, path);
1883 btrfs_free_path(path);
1887 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1890 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1891 struct btrfs_disk_key *disk_key;
1892 struct btrfs_chunk *chunk;
1899 struct btrfs_key key;
1901 array_size = btrfs_super_sys_array_size(super_copy);
1903 ptr = super_copy->sys_chunk_array;
1906 while (cur < array_size) {
1907 disk_key = (struct btrfs_disk_key *)ptr;
1908 btrfs_disk_key_to_cpu(&key, disk_key);
1910 len = sizeof(*disk_key);
1912 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1913 chunk = (struct btrfs_chunk *)(ptr + len);
1914 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1915 len += btrfs_chunk_item_size(num_stripes);
1920 if (key.objectid == chunk_objectid &&
1921 key.offset == chunk_offset) {
1922 memmove(ptr, ptr + len, array_size - (cur + len));
1924 btrfs_set_super_sys_array_size(super_copy, array_size);
1933 static int btrfs_relocate_chunk(struct btrfs_root *root,
1934 u64 chunk_tree, u64 chunk_objectid,
1937 struct extent_map_tree *em_tree;
1938 struct btrfs_root *extent_root;
1939 struct btrfs_trans_handle *trans;
1940 struct extent_map *em;
1941 struct map_lookup *map;
1945 root = root->fs_info->chunk_root;
1946 extent_root = root->fs_info->extent_root;
1947 em_tree = &root->fs_info->mapping_tree.map_tree;
1949 ret = btrfs_can_relocate(extent_root, chunk_offset);
1953 /* step one, relocate all the extents inside this chunk */
1954 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1958 trans = btrfs_start_transaction(root, 0);
1959 BUG_ON(IS_ERR(trans));
1964 * step two, delete the device extents and the
1965 * chunk tree entries
1967 read_lock(&em_tree->lock);
1968 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1969 read_unlock(&em_tree->lock);
1971 BUG_ON(!em || em->start > chunk_offset ||
1972 em->start + em->len < chunk_offset);
1973 map = (struct map_lookup *)em->bdev;
1975 for (i = 0; i < map->num_stripes; i++) {
1976 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1977 map->stripes[i].physical);
1980 if (map->stripes[i].dev) {
1981 ret = btrfs_update_device(trans, map->stripes[i].dev);
1985 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1990 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1992 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1993 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1997 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2000 write_lock(&em_tree->lock);
2001 remove_extent_mapping(em_tree, em);
2002 write_unlock(&em_tree->lock);
2007 /* once for the tree */
2008 free_extent_map(em);
2010 free_extent_map(em);
2012 unlock_chunks(root);
2013 btrfs_end_transaction(trans, root);
2017 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2019 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2020 struct btrfs_path *path;
2021 struct extent_buffer *leaf;
2022 struct btrfs_chunk *chunk;
2023 struct btrfs_key key;
2024 struct btrfs_key found_key;
2025 u64 chunk_tree = chunk_root->root_key.objectid;
2027 bool retried = false;
2031 path = btrfs_alloc_path();
2036 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2037 key.offset = (u64)-1;
2038 key.type = BTRFS_CHUNK_ITEM_KEY;
2041 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2046 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2053 leaf = path->nodes[0];
2054 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2056 chunk = btrfs_item_ptr(leaf, path->slots[0],
2057 struct btrfs_chunk);
2058 chunk_type = btrfs_chunk_type(leaf, chunk);
2059 btrfs_release_path(path);
2061 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2062 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2071 if (found_key.offset == 0)
2073 key.offset = found_key.offset - 1;
2076 if (failed && !retried) {
2080 } else if (failed && retried) {
2085 btrfs_free_path(path);
2089 static int insert_balance_item(struct btrfs_root *root,
2090 struct btrfs_balance_control *bctl)
2092 struct btrfs_trans_handle *trans;
2093 struct btrfs_balance_item *item;
2094 struct btrfs_disk_balance_args disk_bargs;
2095 struct btrfs_path *path;
2096 struct extent_buffer *leaf;
2097 struct btrfs_key key;
2100 path = btrfs_alloc_path();
2104 trans = btrfs_start_transaction(root, 0);
2105 if (IS_ERR(trans)) {
2106 btrfs_free_path(path);
2107 return PTR_ERR(trans);
2110 key.objectid = BTRFS_BALANCE_OBJECTID;
2111 key.type = BTRFS_BALANCE_ITEM_KEY;
2114 ret = btrfs_insert_empty_item(trans, root, path, &key,
2119 leaf = path->nodes[0];
2120 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2122 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2124 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2125 btrfs_set_balance_data(leaf, item, &disk_bargs);
2126 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2127 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2128 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2129 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2131 btrfs_set_balance_flags(leaf, item, bctl->flags);
2133 btrfs_mark_buffer_dirty(leaf);
2135 btrfs_free_path(path);
2136 err = btrfs_commit_transaction(trans, root);
2142 static int del_balance_item(struct btrfs_root *root)
2144 struct btrfs_trans_handle *trans;
2145 struct btrfs_path *path;
2146 struct btrfs_key key;
2149 path = btrfs_alloc_path();
2153 trans = btrfs_start_transaction(root, 0);
2154 if (IS_ERR(trans)) {
2155 btrfs_free_path(path);
2156 return PTR_ERR(trans);
2159 key.objectid = BTRFS_BALANCE_OBJECTID;
2160 key.type = BTRFS_BALANCE_ITEM_KEY;
2163 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2171 ret = btrfs_del_item(trans, root, path);
2173 btrfs_free_path(path);
2174 err = btrfs_commit_transaction(trans, root);
2181 * This is a heuristic used to reduce the number of chunks balanced on
2182 * resume after balance was interrupted.
2184 static void update_balance_args(struct btrfs_balance_control *bctl)
2187 * Turn on soft mode for chunk types that were being converted.
2189 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2190 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2191 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2192 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2193 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2194 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2197 * Turn on usage filter if is not already used. The idea is
2198 * that chunks that we have already balanced should be
2199 * reasonably full. Don't do it for chunks that are being
2200 * converted - that will keep us from relocating unconverted
2201 * (albeit full) chunks.
2203 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2204 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2205 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2206 bctl->data.usage = 90;
2208 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2209 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2210 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2211 bctl->sys.usage = 90;
2213 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2214 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2215 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2216 bctl->meta.usage = 90;
2221 * Should be called with both balance and volume mutexes held to
2222 * serialize other volume operations (add_dev/rm_dev/resize) with
2223 * restriper. Same goes for unset_balance_control.
2225 static void set_balance_control(struct btrfs_balance_control *bctl)
2227 struct btrfs_fs_info *fs_info = bctl->fs_info;
2229 BUG_ON(fs_info->balance_ctl);
2231 spin_lock(&fs_info->balance_lock);
2232 fs_info->balance_ctl = bctl;
2233 spin_unlock(&fs_info->balance_lock);
2236 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2238 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2240 BUG_ON(!fs_info->balance_ctl);
2242 spin_lock(&fs_info->balance_lock);
2243 fs_info->balance_ctl = NULL;
2244 spin_unlock(&fs_info->balance_lock);
2250 * Balance filters. Return 1 if chunk should be filtered out
2251 * (should not be balanced).
2253 static int chunk_profiles_filter(u64 chunk_type,
2254 struct btrfs_balance_args *bargs)
2256 chunk_type = chunk_to_extended(chunk_type) &
2257 BTRFS_EXTENDED_PROFILE_MASK;
2259 if (bargs->profiles & chunk_type)
2265 static u64 div_factor_fine(u64 num, int factor)
2277 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2278 struct btrfs_balance_args *bargs)
2280 struct btrfs_block_group_cache *cache;
2281 u64 chunk_used, user_thresh;
2284 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2285 chunk_used = btrfs_block_group_used(&cache->item);
2287 user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2288 if (chunk_used < user_thresh)
2291 btrfs_put_block_group(cache);
2295 static int chunk_devid_filter(struct extent_buffer *leaf,
2296 struct btrfs_chunk *chunk,
2297 struct btrfs_balance_args *bargs)
2299 struct btrfs_stripe *stripe;
2300 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2303 for (i = 0; i < num_stripes; i++) {
2304 stripe = btrfs_stripe_nr(chunk, i);
2305 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2312 /* [pstart, pend) */
2313 static int chunk_drange_filter(struct extent_buffer *leaf,
2314 struct btrfs_chunk *chunk,
2316 struct btrfs_balance_args *bargs)
2318 struct btrfs_stripe *stripe;
2319 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2325 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2328 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2329 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2333 factor = num_stripes / factor;
2335 for (i = 0; i < num_stripes; i++) {
2336 stripe = btrfs_stripe_nr(chunk, i);
2337 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2340 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2341 stripe_length = btrfs_chunk_length(leaf, chunk);
2342 do_div(stripe_length, factor);
2344 if (stripe_offset < bargs->pend &&
2345 stripe_offset + stripe_length > bargs->pstart)
2352 /* [vstart, vend) */
2353 static int chunk_vrange_filter(struct extent_buffer *leaf,
2354 struct btrfs_chunk *chunk,
2356 struct btrfs_balance_args *bargs)
2358 if (chunk_offset < bargs->vend &&
2359 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2360 /* at least part of the chunk is inside this vrange */
2366 static int chunk_soft_convert_filter(u64 chunk_type,
2367 struct btrfs_balance_args *bargs)
2369 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2372 chunk_type = chunk_to_extended(chunk_type) &
2373 BTRFS_EXTENDED_PROFILE_MASK;
2375 if (bargs->target == chunk_type)
2381 static int should_balance_chunk(struct btrfs_root *root,
2382 struct extent_buffer *leaf,
2383 struct btrfs_chunk *chunk, u64 chunk_offset)
2385 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2386 struct btrfs_balance_args *bargs = NULL;
2387 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2390 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2391 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2395 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2396 bargs = &bctl->data;
2397 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2399 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2400 bargs = &bctl->meta;
2402 /* profiles filter */
2403 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2404 chunk_profiles_filter(chunk_type, bargs)) {
2409 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2410 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2415 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2416 chunk_devid_filter(leaf, chunk, bargs)) {
2420 /* drange filter, makes sense only with devid filter */
2421 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2422 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2427 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2428 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2432 /* soft profile changing mode */
2433 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2434 chunk_soft_convert_filter(chunk_type, bargs)) {
2441 static u64 div_factor(u64 num, int factor)
2450 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2452 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2453 struct btrfs_root *chunk_root = fs_info->chunk_root;
2454 struct btrfs_root *dev_root = fs_info->dev_root;
2455 struct list_head *devices;
2456 struct btrfs_device *device;
2459 struct btrfs_chunk *chunk;
2460 struct btrfs_path *path;
2461 struct btrfs_key key;
2462 struct btrfs_key found_key;
2463 struct btrfs_trans_handle *trans;
2464 struct extent_buffer *leaf;
2467 int enospc_errors = 0;
2468 bool counting = true;
2470 /* step one make some room on all the devices */
2471 devices = &fs_info->fs_devices->devices;
2472 list_for_each_entry(device, devices, dev_list) {
2473 old_size = device->total_bytes;
2474 size_to_free = div_factor(old_size, 1);
2475 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2476 if (!device->writeable ||
2477 device->total_bytes - device->bytes_used > size_to_free)
2480 ret = btrfs_shrink_device(device, old_size - size_to_free);
2485 trans = btrfs_start_transaction(dev_root, 0);
2486 BUG_ON(IS_ERR(trans));
2488 ret = btrfs_grow_device(trans, device, old_size);
2491 btrfs_end_transaction(trans, dev_root);
2494 /* step two, relocate all the chunks */
2495 path = btrfs_alloc_path();
2501 /* zero out stat counters */
2502 spin_lock(&fs_info->balance_lock);
2503 memset(&bctl->stat, 0, sizeof(bctl->stat));
2504 spin_unlock(&fs_info->balance_lock);
2506 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2507 key.offset = (u64)-1;
2508 key.type = BTRFS_CHUNK_ITEM_KEY;
2511 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2512 atomic_read(&fs_info->balance_cancel_req)) {
2517 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2522 * this shouldn't happen, it means the last relocate
2526 BUG(); /* FIXME break ? */
2528 ret = btrfs_previous_item(chunk_root, path, 0,
2529 BTRFS_CHUNK_ITEM_KEY);
2535 leaf = path->nodes[0];
2536 slot = path->slots[0];
2537 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2539 if (found_key.objectid != key.objectid)
2542 /* chunk zero is special */
2543 if (found_key.offset == 0)
2546 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2549 spin_lock(&fs_info->balance_lock);
2550 bctl->stat.considered++;
2551 spin_unlock(&fs_info->balance_lock);
2554 ret = should_balance_chunk(chunk_root, leaf, chunk,
2556 btrfs_release_path(path);
2561 spin_lock(&fs_info->balance_lock);
2562 bctl->stat.expected++;
2563 spin_unlock(&fs_info->balance_lock);
2567 ret = btrfs_relocate_chunk(chunk_root,
2568 chunk_root->root_key.objectid,
2571 if (ret && ret != -ENOSPC)
2573 if (ret == -ENOSPC) {
2576 spin_lock(&fs_info->balance_lock);
2577 bctl->stat.completed++;
2578 spin_unlock(&fs_info->balance_lock);
2581 key.offset = found_key.offset - 1;
2585 btrfs_release_path(path);
2590 btrfs_free_path(path);
2591 if (enospc_errors) {
2592 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2602 * alloc_profile_is_valid - see if a given profile is valid and reduced
2603 * @flags: profile to validate
2604 * @extended: if true @flags is treated as an extended profile
2606 static int alloc_profile_is_valid(u64 flags, int extended)
2608 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2609 BTRFS_BLOCK_GROUP_PROFILE_MASK);
2611 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2613 /* 1) check that all other bits are zeroed */
2617 /* 2) see if profile is reduced */
2619 return !extended; /* "0" is valid for usual profiles */
2621 /* true if exactly one bit set */
2622 return (flags & (flags - 1)) == 0;
2625 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2627 /* cancel requested || normal exit path */
2628 return atomic_read(&fs_info->balance_cancel_req) ||
2629 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2630 atomic_read(&fs_info->balance_cancel_req) == 0);
2633 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2637 unset_balance_control(fs_info);
2638 ret = del_balance_item(fs_info->tree_root);
2642 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2643 struct btrfs_ioctl_balance_args *bargs);
2646 * Should be called with both balance and volume mutexes held
2648 int btrfs_balance(struct btrfs_balance_control *bctl,
2649 struct btrfs_ioctl_balance_args *bargs)
2651 struct btrfs_fs_info *fs_info = bctl->fs_info;
2656 if (btrfs_fs_closing(fs_info) ||
2657 atomic_read(&fs_info->balance_pause_req) ||
2658 atomic_read(&fs_info->balance_cancel_req)) {
2663 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2664 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2668 * In case of mixed groups both data and meta should be picked,
2669 * and identical options should be given for both of them.
2671 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2672 if (mixed && (bctl->flags & allowed)) {
2673 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2674 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2675 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2676 printk(KERN_ERR "btrfs: with mixed groups data and "
2677 "metadata balance options must be the same\n");
2683 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2684 if (fs_info->fs_devices->num_devices == 1)
2685 allowed |= BTRFS_BLOCK_GROUP_DUP;
2686 else if (fs_info->fs_devices->num_devices < 4)
2687 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2689 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2690 BTRFS_BLOCK_GROUP_RAID10);
2692 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2693 (!alloc_profile_is_valid(bctl->data.target, 1) ||
2694 (bctl->data.target & ~allowed))) {
2695 printk(KERN_ERR "btrfs: unable to start balance with target "
2696 "data profile %llu\n",
2697 (unsigned long long)bctl->data.target);
2701 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2702 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2703 (bctl->meta.target & ~allowed))) {
2704 printk(KERN_ERR "btrfs: unable to start balance with target "
2705 "metadata profile %llu\n",
2706 (unsigned long long)bctl->meta.target);
2710 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2711 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2712 (bctl->sys.target & ~allowed))) {
2713 printk(KERN_ERR "btrfs: unable to start balance with target "
2714 "system profile %llu\n",
2715 (unsigned long long)bctl->sys.target);
2720 /* allow dup'ed data chunks only in mixed mode */
2721 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2722 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2723 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2728 /* allow to reduce meta or sys integrity only if force set */
2729 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2730 BTRFS_BLOCK_GROUP_RAID10;
2731 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2732 (fs_info->avail_system_alloc_bits & allowed) &&
2733 !(bctl->sys.target & allowed)) ||
2734 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2735 (fs_info->avail_metadata_alloc_bits & allowed) &&
2736 !(bctl->meta.target & allowed))) {
2737 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2738 printk(KERN_INFO "btrfs: force reducing metadata "
2741 printk(KERN_ERR "btrfs: balance will reduce metadata "
2742 "integrity, use force if you want this\n");
2748 ret = insert_balance_item(fs_info->tree_root, bctl);
2749 if (ret && ret != -EEXIST)
2752 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2753 BUG_ON(ret == -EEXIST);
2754 set_balance_control(bctl);
2756 BUG_ON(ret != -EEXIST);
2757 spin_lock(&fs_info->balance_lock);
2758 update_balance_args(bctl);
2759 spin_unlock(&fs_info->balance_lock);
2762 atomic_inc(&fs_info->balance_running);
2763 mutex_unlock(&fs_info->balance_mutex);
2765 ret = __btrfs_balance(fs_info);
2767 mutex_lock(&fs_info->balance_mutex);
2768 atomic_dec(&fs_info->balance_running);
2771 memset(bargs, 0, sizeof(*bargs));
2772 update_ioctl_balance_args(fs_info, 0, bargs);
2775 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2776 balance_need_close(fs_info)) {
2777 __cancel_balance(fs_info);
2780 wake_up(&fs_info->balance_wait_q);
2784 if (bctl->flags & BTRFS_BALANCE_RESUME)
2785 __cancel_balance(fs_info);
2791 static int balance_kthread(void *data)
2793 struct btrfs_balance_control *bctl =
2794 (struct btrfs_balance_control *)data;
2795 struct btrfs_fs_info *fs_info = bctl->fs_info;
2798 mutex_lock(&fs_info->volume_mutex);
2799 mutex_lock(&fs_info->balance_mutex);
2801 set_balance_control(bctl);
2803 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2804 printk(KERN_INFO "btrfs: force skipping balance\n");
2806 printk(KERN_INFO "btrfs: continuing balance\n");
2807 ret = btrfs_balance(bctl, NULL);
2810 mutex_unlock(&fs_info->balance_mutex);
2811 mutex_unlock(&fs_info->volume_mutex);
2815 int btrfs_recover_balance(struct btrfs_root *tree_root)
2817 struct task_struct *tsk;
2818 struct btrfs_balance_control *bctl;
2819 struct btrfs_balance_item *item;
2820 struct btrfs_disk_balance_args disk_bargs;
2821 struct btrfs_path *path;
2822 struct extent_buffer *leaf;
2823 struct btrfs_key key;
2826 path = btrfs_alloc_path();
2830 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2836 key.objectid = BTRFS_BALANCE_OBJECTID;
2837 key.type = BTRFS_BALANCE_ITEM_KEY;
2840 ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0);
2843 if (ret > 0) { /* ret = -ENOENT; */
2848 leaf = path->nodes[0];
2849 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2851 bctl->fs_info = tree_root->fs_info;
2852 bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME;
2854 btrfs_balance_data(leaf, item, &disk_bargs);
2855 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2856 btrfs_balance_meta(leaf, item, &disk_bargs);
2857 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2858 btrfs_balance_sys(leaf, item, &disk_bargs);
2859 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2861 tsk = kthread_run(balance_kthread, bctl, "btrfs-balance");
2870 btrfs_free_path(path);
2874 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2878 mutex_lock(&fs_info->balance_mutex);
2879 if (!fs_info->balance_ctl) {
2880 mutex_unlock(&fs_info->balance_mutex);
2884 if (atomic_read(&fs_info->balance_running)) {
2885 atomic_inc(&fs_info->balance_pause_req);
2886 mutex_unlock(&fs_info->balance_mutex);
2888 wait_event(fs_info->balance_wait_q,
2889 atomic_read(&fs_info->balance_running) == 0);
2891 mutex_lock(&fs_info->balance_mutex);
2892 /* we are good with balance_ctl ripped off from under us */
2893 BUG_ON(atomic_read(&fs_info->balance_running));
2894 atomic_dec(&fs_info->balance_pause_req);
2899 mutex_unlock(&fs_info->balance_mutex);
2903 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
2905 mutex_lock(&fs_info->balance_mutex);
2906 if (!fs_info->balance_ctl) {
2907 mutex_unlock(&fs_info->balance_mutex);
2911 atomic_inc(&fs_info->balance_cancel_req);
2913 * if we are running just wait and return, balance item is
2914 * deleted in btrfs_balance in this case
2916 if (atomic_read(&fs_info->balance_running)) {
2917 mutex_unlock(&fs_info->balance_mutex);
2918 wait_event(fs_info->balance_wait_q,
2919 atomic_read(&fs_info->balance_running) == 0);
2920 mutex_lock(&fs_info->balance_mutex);
2922 /* __cancel_balance needs volume_mutex */
2923 mutex_unlock(&fs_info->balance_mutex);
2924 mutex_lock(&fs_info->volume_mutex);
2925 mutex_lock(&fs_info->balance_mutex);
2927 if (fs_info->balance_ctl)
2928 __cancel_balance(fs_info);
2930 mutex_unlock(&fs_info->volume_mutex);
2933 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
2934 atomic_dec(&fs_info->balance_cancel_req);
2935 mutex_unlock(&fs_info->balance_mutex);
2940 * shrinking a device means finding all of the device extents past
2941 * the new size, and then following the back refs to the chunks.
2942 * The chunk relocation code actually frees the device extent
2944 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2946 struct btrfs_trans_handle *trans;
2947 struct btrfs_root *root = device->dev_root;
2948 struct btrfs_dev_extent *dev_extent = NULL;
2949 struct btrfs_path *path;
2957 bool retried = false;
2958 struct extent_buffer *l;
2959 struct btrfs_key key;
2960 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2961 u64 old_total = btrfs_super_total_bytes(super_copy);
2962 u64 old_size = device->total_bytes;
2963 u64 diff = device->total_bytes - new_size;
2965 if (new_size >= device->total_bytes)
2968 path = btrfs_alloc_path();
2976 device->total_bytes = new_size;
2977 if (device->writeable) {
2978 device->fs_devices->total_rw_bytes -= diff;
2979 spin_lock(&root->fs_info->free_chunk_lock);
2980 root->fs_info->free_chunk_space -= diff;
2981 spin_unlock(&root->fs_info->free_chunk_lock);
2983 unlock_chunks(root);
2986 key.objectid = device->devid;
2987 key.offset = (u64)-1;
2988 key.type = BTRFS_DEV_EXTENT_KEY;
2991 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2995 ret = btrfs_previous_item(root, path, 0, key.type);
3000 btrfs_release_path(path);
3005 slot = path->slots[0];
3006 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3008 if (key.objectid != device->devid) {
3009 btrfs_release_path(path);
3013 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3014 length = btrfs_dev_extent_length(l, dev_extent);
3016 if (key.offset + length <= new_size) {
3017 btrfs_release_path(path);
3021 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3022 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3023 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3024 btrfs_release_path(path);
3026 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3028 if (ret && ret != -ENOSPC)
3035 if (failed && !retried) {
3039 } else if (failed && retried) {
3043 device->total_bytes = old_size;
3044 if (device->writeable)
3045 device->fs_devices->total_rw_bytes += diff;
3046 spin_lock(&root->fs_info->free_chunk_lock);
3047 root->fs_info->free_chunk_space += diff;
3048 spin_unlock(&root->fs_info->free_chunk_lock);
3049 unlock_chunks(root);
3053 /* Shrinking succeeded, else we would be at "done". */
3054 trans = btrfs_start_transaction(root, 0);
3055 if (IS_ERR(trans)) {
3056 ret = PTR_ERR(trans);
3062 device->disk_total_bytes = new_size;
3063 /* Now btrfs_update_device() will change the on-disk size. */
3064 ret = btrfs_update_device(trans, device);
3066 unlock_chunks(root);
3067 btrfs_end_transaction(trans, root);
3070 WARN_ON(diff > old_total);
3071 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3072 unlock_chunks(root);
3073 btrfs_end_transaction(trans, root);
3075 btrfs_free_path(path);
3079 static int btrfs_add_system_chunk(struct btrfs_root *root,
3080 struct btrfs_key *key,
3081 struct btrfs_chunk *chunk, int item_size)
3083 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3084 struct btrfs_disk_key disk_key;
3088 array_size = btrfs_super_sys_array_size(super_copy);
3089 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3092 ptr = super_copy->sys_chunk_array + array_size;
3093 btrfs_cpu_key_to_disk(&disk_key, key);
3094 memcpy(ptr, &disk_key, sizeof(disk_key));
3095 ptr += sizeof(disk_key);
3096 memcpy(ptr, chunk, item_size);
3097 item_size += sizeof(disk_key);
3098 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3103 * sort the devices in descending order by max_avail, total_avail
3105 static int btrfs_cmp_device_info(const void *a, const void *b)
3107 const struct btrfs_device_info *di_a = a;
3108 const struct btrfs_device_info *di_b = b;
3110 if (di_a->max_avail > di_b->max_avail)
3112 if (di_a->max_avail < di_b->max_avail)
3114 if (di_a->total_avail > di_b->total_avail)
3116 if (di_a->total_avail < di_b->total_avail)
3121 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3122 struct btrfs_root *extent_root,
3123 struct map_lookup **map_ret,
3124 u64 *num_bytes_out, u64 *stripe_size_out,
3125 u64 start, u64 type)
3127 struct btrfs_fs_info *info = extent_root->fs_info;
3128 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3129 struct list_head *cur;
3130 struct map_lookup *map = NULL;
3131 struct extent_map_tree *em_tree;
3132 struct extent_map *em;
3133 struct btrfs_device_info *devices_info = NULL;
3135 int num_stripes; /* total number of stripes to allocate */
3136 int sub_stripes; /* sub_stripes info for map */
3137 int dev_stripes; /* stripes per dev */
3138 int devs_max; /* max devs to use */
3139 int devs_min; /* min devs needed */
3140 int devs_increment; /* ndevs has to be a multiple of this */
3141 int ncopies; /* how many copies to data has */
3143 u64 max_stripe_size;
3151 BUG_ON(!alloc_profile_is_valid(type, 0));
3153 if (list_empty(&fs_devices->alloc_list))
3160 devs_max = 0; /* 0 == as many as possible */
3164 * define the properties of each RAID type.
3165 * FIXME: move this to a global table and use it in all RAID
3168 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3172 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3174 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3179 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3188 if (type & BTRFS_BLOCK_GROUP_DATA) {
3189 max_stripe_size = 1024 * 1024 * 1024;
3190 max_chunk_size = 10 * max_stripe_size;
3191 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3192 /* for larger filesystems, use larger metadata chunks */
3193 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3194 max_stripe_size = 1024 * 1024 * 1024;
3196 max_stripe_size = 256 * 1024 * 1024;
3197 max_chunk_size = max_stripe_size;
3198 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3199 max_stripe_size = 32 * 1024 * 1024;
3200 max_chunk_size = 2 * max_stripe_size;
3202 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3207 /* we don't want a chunk larger than 10% of writeable space */
3208 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3211 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3216 cur = fs_devices->alloc_list.next;
3219 * in the first pass through the devices list, we gather information
3220 * about the available holes on each device.
3223 while (cur != &fs_devices->alloc_list) {
3224 struct btrfs_device *device;
3228 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3232 if (!device->writeable) {
3234 "btrfs: read-only device in alloc_list\n");
3239 if (!device->in_fs_metadata)
3242 if (device->total_bytes > device->bytes_used)
3243 total_avail = device->total_bytes - device->bytes_used;
3247 /* If there is no space on this device, skip it. */
3248 if (total_avail == 0)
3251 ret = find_free_dev_extent(device,
3252 max_stripe_size * dev_stripes,
3253 &dev_offset, &max_avail);
3254 if (ret && ret != -ENOSPC)
3258 max_avail = max_stripe_size * dev_stripes;
3260 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3263 devices_info[ndevs].dev_offset = dev_offset;
3264 devices_info[ndevs].max_avail = max_avail;
3265 devices_info[ndevs].total_avail = total_avail;
3266 devices_info[ndevs].dev = device;
3271 * now sort the devices by hole size / available space
3273 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3274 btrfs_cmp_device_info, NULL);
3276 /* round down to number of usable stripes */
3277 ndevs -= ndevs % devs_increment;
3279 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3284 if (devs_max && ndevs > devs_max)
3287 * the primary goal is to maximize the number of stripes, so use as many
3288 * devices as possible, even if the stripes are not maximum sized.
3290 stripe_size = devices_info[ndevs-1].max_avail;
3291 num_stripes = ndevs * dev_stripes;
3293 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
3294 stripe_size = max_chunk_size * ncopies;
3295 do_div(stripe_size, num_stripes);
3298 do_div(stripe_size, dev_stripes);
3299 do_div(stripe_size, BTRFS_STRIPE_LEN);
3300 stripe_size *= BTRFS_STRIPE_LEN;
3302 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3307 map->num_stripes = num_stripes;
3309 for (i = 0; i < ndevs; ++i) {
3310 for (j = 0; j < dev_stripes; ++j) {
3311 int s = i * dev_stripes + j;
3312 map->stripes[s].dev = devices_info[i].dev;
3313 map->stripes[s].physical = devices_info[i].dev_offset +
3317 map->sector_size = extent_root->sectorsize;
3318 map->stripe_len = BTRFS_STRIPE_LEN;
3319 map->io_align = BTRFS_STRIPE_LEN;
3320 map->io_width = BTRFS_STRIPE_LEN;
3322 map->sub_stripes = sub_stripes;
3325 num_bytes = stripe_size * (num_stripes / ncopies);
3327 *stripe_size_out = stripe_size;
3328 *num_bytes_out = num_bytes;
3330 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3332 em = alloc_extent_map();
3337 em->bdev = (struct block_device *)map;
3339 em->len = num_bytes;
3340 em->block_start = 0;
3341 em->block_len = em->len;
3343 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3344 write_lock(&em_tree->lock);
3345 ret = add_extent_mapping(em_tree, em);
3346 write_unlock(&em_tree->lock);
3348 free_extent_map(em);
3350 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3351 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3355 for (i = 0; i < map->num_stripes; ++i) {
3356 struct btrfs_device *device;
3359 device = map->stripes[i].dev;
3360 dev_offset = map->stripes[i].physical;
3362 ret = btrfs_alloc_dev_extent(trans, device,
3363 info->chunk_root->root_key.objectid,
3364 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3365 start, dev_offset, stripe_size);
3369 kfree(devices_info);
3374 kfree(devices_info);
3378 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *extent_root,
3380 struct map_lookup *map, u64 chunk_offset,
3381 u64 chunk_size, u64 stripe_size)
3384 struct btrfs_key key;
3385 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3386 struct btrfs_device *device;
3387 struct btrfs_chunk *chunk;
3388 struct btrfs_stripe *stripe;
3389 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3393 chunk = kzalloc(item_size, GFP_NOFS);
3398 while (index < map->num_stripes) {
3399 device = map->stripes[index].dev;
3400 device->bytes_used += stripe_size;
3401 ret = btrfs_update_device(trans, device);
3406 spin_lock(&extent_root->fs_info->free_chunk_lock);
3407 extent_root->fs_info->free_chunk_space -= (stripe_size *
3409 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3412 stripe = &chunk->stripe;
3413 while (index < map->num_stripes) {
3414 device = map->stripes[index].dev;
3415 dev_offset = map->stripes[index].physical;
3417 btrfs_set_stack_stripe_devid(stripe, device->devid);
3418 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3419 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3424 btrfs_set_stack_chunk_length(chunk, chunk_size);
3425 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3426 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3427 btrfs_set_stack_chunk_type(chunk, map->type);
3428 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3429 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3430 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3431 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3432 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3434 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3435 key.type = BTRFS_CHUNK_ITEM_KEY;
3436 key.offset = chunk_offset;
3438 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3441 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3442 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3452 * Chunk allocation falls into two parts. The first part does works
3453 * that make the new allocated chunk useable, but not do any operation
3454 * that modifies the chunk tree. The second part does the works that
3455 * require modifying the chunk tree. This division is important for the
3456 * bootstrap process of adding storage to a seed btrfs.
3458 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3459 struct btrfs_root *extent_root, u64 type)
3464 struct map_lookup *map;
3465 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3468 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3473 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3474 &stripe_size, chunk_offset, type);
3478 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3479 chunk_size, stripe_size);
3484 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3485 struct btrfs_root *root,
3486 struct btrfs_device *device)
3489 u64 sys_chunk_offset;
3493 u64 sys_stripe_size;
3495 struct map_lookup *map;
3496 struct map_lookup *sys_map;
3497 struct btrfs_fs_info *fs_info = root->fs_info;
3498 struct btrfs_root *extent_root = fs_info->extent_root;
3501 ret = find_next_chunk(fs_info->chunk_root,
3502 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3506 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3507 fs_info->avail_metadata_alloc_bits;
3508 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3510 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3511 &stripe_size, chunk_offset, alloc_profile);
3514 sys_chunk_offset = chunk_offset + chunk_size;
3516 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3517 fs_info->avail_system_alloc_bits;
3518 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3520 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3521 &sys_chunk_size, &sys_stripe_size,
3522 sys_chunk_offset, alloc_profile);
3525 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3529 * Modifying chunk tree needs allocating new blocks from both
3530 * system block group and metadata block group. So we only can
3531 * do operations require modifying the chunk tree after both
3532 * block groups were created.
3534 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3535 chunk_size, stripe_size);
3538 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3539 sys_chunk_offset, sys_chunk_size,
3545 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3547 struct extent_map *em;
3548 struct map_lookup *map;
3549 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3553 read_lock(&map_tree->map_tree.lock);
3554 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3555 read_unlock(&map_tree->map_tree.lock);
3559 if (btrfs_test_opt(root, DEGRADED)) {
3560 free_extent_map(em);
3564 map = (struct map_lookup *)em->bdev;
3565 for (i = 0; i < map->num_stripes; i++) {
3566 if (!map->stripes[i].dev->writeable) {
3571 free_extent_map(em);
3575 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3577 extent_map_tree_init(&tree->map_tree);
3580 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3582 struct extent_map *em;
3585 write_lock(&tree->map_tree.lock);
3586 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3588 remove_extent_mapping(&tree->map_tree, em);
3589 write_unlock(&tree->map_tree.lock);
3594 free_extent_map(em);
3595 /* once for the tree */
3596 free_extent_map(em);
3600 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3602 struct extent_map *em;
3603 struct map_lookup *map;
3604 struct extent_map_tree *em_tree = &map_tree->map_tree;
3607 read_lock(&em_tree->lock);
3608 em = lookup_extent_mapping(em_tree, logical, len);
3609 read_unlock(&em_tree->lock);
3612 BUG_ON(em->start > logical || em->start + em->len < logical);
3613 map = (struct map_lookup *)em->bdev;
3614 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3615 ret = map->num_stripes;
3616 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3617 ret = map->sub_stripes;
3620 free_extent_map(em);
3624 static int find_live_mirror(struct map_lookup *map, int first, int num,
3628 if (map->stripes[optimal].dev->bdev)
3630 for (i = first; i < first + num; i++) {
3631 if (map->stripes[i].dev->bdev)
3634 /* we couldn't find one that doesn't fail. Just return something
3635 * and the io error handling code will clean up eventually
3640 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3641 u64 logical, u64 *length,
3642 struct btrfs_bio **bbio_ret,
3645 struct extent_map *em;
3646 struct map_lookup *map;
3647 struct extent_map_tree *em_tree = &map_tree->map_tree;
3650 u64 stripe_end_offset;
3659 struct btrfs_bio *bbio = NULL;
3661 read_lock(&em_tree->lock);
3662 em = lookup_extent_mapping(em_tree, logical, *length);
3663 read_unlock(&em_tree->lock);
3666 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
3667 (unsigned long long)logical,
3668 (unsigned long long)*length);
3672 BUG_ON(em->start > logical || em->start + em->len < logical);
3673 map = (struct map_lookup *)em->bdev;
3674 offset = logical - em->start;
3676 if (mirror_num > map->num_stripes)
3681 * stripe_nr counts the total number of stripes we have to stride
3682 * to get to this block
3684 do_div(stripe_nr, map->stripe_len);
3686 stripe_offset = stripe_nr * map->stripe_len;
3687 BUG_ON(offset < stripe_offset);
3689 /* stripe_offset is the offset of this block in its stripe*/
3690 stripe_offset = offset - stripe_offset;
3692 if (rw & REQ_DISCARD)
3693 *length = min_t(u64, em->len - offset, *length);
3694 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3695 /* we limit the length of each bio to what fits in a stripe */
3696 *length = min_t(u64, em->len - offset,
3697 map->stripe_len - stripe_offset);
3699 *length = em->len - offset;
3707 stripe_nr_orig = stripe_nr;
3708 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3709 (~(map->stripe_len - 1));
3710 do_div(stripe_nr_end, map->stripe_len);
3711 stripe_end_offset = stripe_nr_end * map->stripe_len -
3713 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3714 if (rw & REQ_DISCARD)
3715 num_stripes = min_t(u64, map->num_stripes,
3716 stripe_nr_end - stripe_nr_orig);
3717 stripe_index = do_div(stripe_nr, map->num_stripes);
3718 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3719 if (rw & (REQ_WRITE | REQ_DISCARD))
3720 num_stripes = map->num_stripes;
3721 else if (mirror_num)
3722 stripe_index = mirror_num - 1;
3724 stripe_index = find_live_mirror(map, 0,
3726 current->pid % map->num_stripes);
3727 mirror_num = stripe_index + 1;
3730 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3731 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3732 num_stripes = map->num_stripes;
3733 } else if (mirror_num) {
3734 stripe_index = mirror_num - 1;
3739 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3740 int factor = map->num_stripes / map->sub_stripes;
3742 stripe_index = do_div(stripe_nr, factor);
3743 stripe_index *= map->sub_stripes;
3746 num_stripes = map->sub_stripes;
3747 else if (rw & REQ_DISCARD)
3748 num_stripes = min_t(u64, map->sub_stripes *
3749 (stripe_nr_end - stripe_nr_orig),
3751 else if (mirror_num)
3752 stripe_index += mirror_num - 1;
3754 stripe_index = find_live_mirror(map, stripe_index,
3755 map->sub_stripes, stripe_index +
3756 current->pid % map->sub_stripes);
3757 mirror_num = stripe_index + 1;
3761 * after this do_div call, stripe_nr is the number of stripes
3762 * on this device we have to walk to find the data, and
3763 * stripe_index is the number of our device in the stripe array
3765 stripe_index = do_div(stripe_nr, map->num_stripes);
3766 mirror_num = stripe_index + 1;
3768 BUG_ON(stripe_index >= map->num_stripes);
3770 bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3775 atomic_set(&bbio->error, 0);
3777 if (rw & REQ_DISCARD) {
3779 int sub_stripes = 0;
3780 u64 stripes_per_dev = 0;
3781 u32 remaining_stripes = 0;
3784 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3785 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3788 sub_stripes = map->sub_stripes;
3790 factor = map->num_stripes / sub_stripes;
3791 stripes_per_dev = div_u64_rem(stripe_nr_end -
3794 &remaining_stripes);
3797 for (i = 0; i < num_stripes; i++) {
3798 bbio->stripes[i].physical =
3799 map->stripes[stripe_index].physical +
3800 stripe_offset + stripe_nr * map->stripe_len;
3801 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3803 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3804 BTRFS_BLOCK_GROUP_RAID10)) {
3805 bbio->stripes[i].length = stripes_per_dev *
3807 if (i / sub_stripes < remaining_stripes)
3808 bbio->stripes[i].length +=
3810 if (i < sub_stripes)
3811 bbio->stripes[i].length -=
3813 if ((i / sub_stripes + 1) %
3814 sub_stripes == remaining_stripes)
3815 bbio->stripes[i].length -=
3817 if (i == sub_stripes - 1)
3820 bbio->stripes[i].length = *length;
3823 if (stripe_index == map->num_stripes) {
3824 /* This could only happen for RAID0/10 */
3830 for (i = 0; i < num_stripes; i++) {
3831 bbio->stripes[i].physical =
3832 map->stripes[stripe_index].physical +
3834 stripe_nr * map->stripe_len;
3835 bbio->stripes[i].dev =
3836 map->stripes[stripe_index].dev;
3841 if (rw & REQ_WRITE) {
3842 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3843 BTRFS_BLOCK_GROUP_RAID10 |
3844 BTRFS_BLOCK_GROUP_DUP)) {
3850 bbio->num_stripes = num_stripes;
3851 bbio->max_errors = max_errors;
3852 bbio->mirror_num = mirror_num;
3854 free_extent_map(em);
3858 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3859 u64 logical, u64 *length,
3860 struct btrfs_bio **bbio_ret, int mirror_num)
3862 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
3866 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3867 u64 chunk_start, u64 physical, u64 devid,
3868 u64 **logical, int *naddrs, int *stripe_len)
3870 struct extent_map_tree *em_tree = &map_tree->map_tree;
3871 struct extent_map *em;
3872 struct map_lookup *map;
3879 read_lock(&em_tree->lock);
3880 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3881 read_unlock(&em_tree->lock);
3883 BUG_ON(!em || em->start != chunk_start);
3884 map = (struct map_lookup *)em->bdev;
3887 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3888 do_div(length, map->num_stripes / map->sub_stripes);
3889 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3890 do_div(length, map->num_stripes);
3892 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3895 for (i = 0; i < map->num_stripes; i++) {
3896 if (devid && map->stripes[i].dev->devid != devid)
3898 if (map->stripes[i].physical > physical ||
3899 map->stripes[i].physical + length <= physical)
3902 stripe_nr = physical - map->stripes[i].physical;
3903 do_div(stripe_nr, map->stripe_len);
3905 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3906 stripe_nr = stripe_nr * map->num_stripes + i;
3907 do_div(stripe_nr, map->sub_stripes);
3908 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3909 stripe_nr = stripe_nr * map->num_stripes + i;
3911 bytenr = chunk_start + stripe_nr * map->stripe_len;
3912 WARN_ON(nr >= map->num_stripes);
3913 for (j = 0; j < nr; j++) {
3914 if (buf[j] == bytenr)
3918 WARN_ON(nr >= map->num_stripes);
3925 *stripe_len = map->stripe_len;
3927 free_extent_map(em);
3931 static void btrfs_end_bio(struct bio *bio, int err)
3933 struct btrfs_bio *bbio = bio->bi_private;
3934 int is_orig_bio = 0;
3937 atomic_inc(&bbio->error);
3939 if (bio == bbio->orig_bio)
3942 if (atomic_dec_and_test(&bbio->stripes_pending)) {
3945 bio = bbio->orig_bio;
3947 bio->bi_private = bbio->private;
3948 bio->bi_end_io = bbio->end_io;
3949 bio->bi_bdev = (struct block_device *)
3950 (unsigned long)bbio->mirror_num;
3951 /* only send an error to the higher layers if it is
3952 * beyond the tolerance of the multi-bio
3954 if (atomic_read(&bbio->error) > bbio->max_errors) {
3958 * this bio is actually up to date, we didn't
3959 * go over the max number of errors
3961 set_bit(BIO_UPTODATE, &bio->bi_flags);
3966 bio_endio(bio, err);
3967 } else if (!is_orig_bio) {
3972 struct async_sched {
3975 struct btrfs_fs_info *info;
3976 struct btrfs_work work;
3980 * see run_scheduled_bios for a description of why bios are collected for
3983 * This will add one bio to the pending list for a device and make sure
3984 * the work struct is scheduled.
3986 static noinline int schedule_bio(struct btrfs_root *root,
3987 struct btrfs_device *device,
3988 int rw, struct bio *bio)
3990 int should_queue = 1;
3991 struct btrfs_pending_bios *pending_bios;
3993 /* don't bother with additional async steps for reads, right now */
3994 if (!(rw & REQ_WRITE)) {
3996 btrfsic_submit_bio(rw, bio);
4002 * nr_async_bios allows us to reliably return congestion to the
4003 * higher layers. Otherwise, the async bio makes it appear we have
4004 * made progress against dirty pages when we've really just put it
4005 * on a queue for later
4007 atomic_inc(&root->fs_info->nr_async_bios);
4008 WARN_ON(bio->bi_next);
4009 bio->bi_next = NULL;
4012 spin_lock(&device->io_lock);
4013 if (bio->bi_rw & REQ_SYNC)
4014 pending_bios = &device->pending_sync_bios;
4016 pending_bios = &device->pending_bios;
4018 if (pending_bios->tail)
4019 pending_bios->tail->bi_next = bio;
4021 pending_bios->tail = bio;
4022 if (!pending_bios->head)
4023 pending_bios->head = bio;
4024 if (device->running_pending)
4027 spin_unlock(&device->io_lock);
4030 btrfs_queue_worker(&root->fs_info->submit_workers,
4035 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4036 int mirror_num, int async_submit)
4038 struct btrfs_mapping_tree *map_tree;
4039 struct btrfs_device *dev;
4040 struct bio *first_bio = bio;
4041 u64 logical = (u64)bio->bi_sector << 9;
4047 struct btrfs_bio *bbio = NULL;
4049 length = bio->bi_size;
4050 map_tree = &root->fs_info->mapping_tree;
4051 map_length = length;
4053 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4057 total_devs = bbio->num_stripes;
4058 if (map_length < length) {
4059 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
4060 "len %llu\n", (unsigned long long)logical,
4061 (unsigned long long)length,
4062 (unsigned long long)map_length);
4066 bbio->orig_bio = first_bio;
4067 bbio->private = first_bio->bi_private;
4068 bbio->end_io = first_bio->bi_end_io;
4069 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4071 while (dev_nr < total_devs) {
4072 if (dev_nr < total_devs - 1) {
4073 bio = bio_clone(first_bio, GFP_NOFS);
4078 bio->bi_private = bbio;
4079 bio->bi_end_io = btrfs_end_bio;
4080 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4081 dev = bbio->stripes[dev_nr].dev;
4082 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4083 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4084 "(%s id %llu), size=%u\n", rw,
4085 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4086 dev->name, dev->devid, bio->bi_size);
4087 bio->bi_bdev = dev->bdev;
4089 schedule_bio(root, dev, rw, bio);
4091 btrfsic_submit_bio(rw, bio);
4093 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4094 bio->bi_sector = logical >> 9;
4095 bio_endio(bio, -EIO);
4102 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4105 struct btrfs_device *device;
4106 struct btrfs_fs_devices *cur_devices;
4108 cur_devices = root->fs_info->fs_devices;
4109 while (cur_devices) {
4111 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4112 device = __find_device(&cur_devices->devices,
4117 cur_devices = cur_devices->seed;
4122 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4123 u64 devid, u8 *dev_uuid)
4125 struct btrfs_device *device;
4126 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4128 device = kzalloc(sizeof(*device), GFP_NOFS);
4131 list_add(&device->dev_list,
4132 &fs_devices->devices);
4133 device->dev_root = root->fs_info->dev_root;
4134 device->devid = devid;
4135 device->work.func = pending_bios_fn;
4136 device->fs_devices = fs_devices;
4137 device->missing = 1;
4138 fs_devices->num_devices++;
4139 fs_devices->missing_devices++;
4140 spin_lock_init(&device->io_lock);
4141 INIT_LIST_HEAD(&device->dev_alloc_list);
4142 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4146 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4147 struct extent_buffer *leaf,
4148 struct btrfs_chunk *chunk)
4150 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4151 struct map_lookup *map;
4152 struct extent_map *em;
4156 u8 uuid[BTRFS_UUID_SIZE];
4161 logical = key->offset;
4162 length = btrfs_chunk_length(leaf, chunk);
4164 read_lock(&map_tree->map_tree.lock);
4165 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4166 read_unlock(&map_tree->map_tree.lock);
4168 /* already mapped? */
4169 if (em && em->start <= logical && em->start + em->len > logical) {
4170 free_extent_map(em);
4173 free_extent_map(em);
4176 em = alloc_extent_map();
4179 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4180 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4182 free_extent_map(em);
4186 em->bdev = (struct block_device *)map;
4187 em->start = logical;
4189 em->block_start = 0;
4190 em->block_len = em->len;
4192 map->num_stripes = num_stripes;
4193 map->io_width = btrfs_chunk_io_width(leaf, chunk);
4194 map->io_align = btrfs_chunk_io_align(leaf, chunk);
4195 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4196 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4197 map->type = btrfs_chunk_type(leaf, chunk);
4198 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4199 for (i = 0; i < num_stripes; i++) {
4200 map->stripes[i].physical =
4201 btrfs_stripe_offset_nr(leaf, chunk, i);
4202 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4203 read_extent_buffer(leaf, uuid, (unsigned long)
4204 btrfs_stripe_dev_uuid_nr(chunk, i),
4206 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4208 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4210 free_extent_map(em);
4213 if (!map->stripes[i].dev) {
4214 map->stripes[i].dev =
4215 add_missing_dev(root, devid, uuid);
4216 if (!map->stripes[i].dev) {
4218 free_extent_map(em);
4222 map->stripes[i].dev->in_fs_metadata = 1;
4225 write_lock(&map_tree->map_tree.lock);
4226 ret = add_extent_mapping(&map_tree->map_tree, em);
4227 write_unlock(&map_tree->map_tree.lock);
4229 free_extent_map(em);
4234 static int fill_device_from_item(struct extent_buffer *leaf,
4235 struct btrfs_dev_item *dev_item,
4236 struct btrfs_device *device)
4240 device->devid = btrfs_device_id(leaf, dev_item);
4241 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4242 device->total_bytes = device->disk_total_bytes;
4243 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4244 device->type = btrfs_device_type(leaf, dev_item);
4245 device->io_align = btrfs_device_io_align(leaf, dev_item);
4246 device->io_width = btrfs_device_io_width(leaf, dev_item);
4247 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4249 ptr = (unsigned long)btrfs_device_uuid(dev_item);
4250 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4255 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4257 struct btrfs_fs_devices *fs_devices;
4260 BUG_ON(!mutex_is_locked(&uuid_mutex));
4262 fs_devices = root->fs_info->fs_devices->seed;
4263 while (fs_devices) {
4264 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4268 fs_devices = fs_devices->seed;
4271 fs_devices = find_fsid(fsid);
4277 fs_devices = clone_fs_devices(fs_devices);
4278 if (IS_ERR(fs_devices)) {
4279 ret = PTR_ERR(fs_devices);
4283 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4284 root->fs_info->bdev_holder);
4288 if (!fs_devices->seeding) {
4289 __btrfs_close_devices(fs_devices);
4290 free_fs_devices(fs_devices);
4295 fs_devices->seed = root->fs_info->fs_devices->seed;
4296 root->fs_info->fs_devices->seed = fs_devices;
4301 static int read_one_dev(struct btrfs_root *root,
4302 struct extent_buffer *leaf,
4303 struct btrfs_dev_item *dev_item)
4305 struct btrfs_device *device;
4308 u8 fs_uuid[BTRFS_UUID_SIZE];
4309 u8 dev_uuid[BTRFS_UUID_SIZE];
4311 devid = btrfs_device_id(leaf, dev_item);
4312 read_extent_buffer(leaf, dev_uuid,
4313 (unsigned long)btrfs_device_uuid(dev_item),
4315 read_extent_buffer(leaf, fs_uuid,
4316 (unsigned long)btrfs_device_fsid(dev_item),
4319 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4320 ret = open_seed_devices(root, fs_uuid);
4321 if (ret && !btrfs_test_opt(root, DEGRADED))
4325 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4326 if (!device || !device->bdev) {
4327 if (!btrfs_test_opt(root, DEGRADED))
4331 printk(KERN_WARNING "warning devid %llu missing\n",
4332 (unsigned long long)devid);
4333 device = add_missing_dev(root, devid, dev_uuid);
4336 } else if (!device->missing) {
4338 * this happens when a device that was properly setup
4339 * in the device info lists suddenly goes bad.
4340 * device->bdev is NULL, and so we have to set
4341 * device->missing to one here
4343 root->fs_info->fs_devices->missing_devices++;
4344 device->missing = 1;
4348 if (device->fs_devices != root->fs_info->fs_devices) {
4349 BUG_ON(device->writeable);
4350 if (device->generation !=
4351 btrfs_device_generation(leaf, dev_item))
4355 fill_device_from_item(leaf, dev_item, device);
4356 device->dev_root = root->fs_info->dev_root;
4357 device->in_fs_metadata = 1;
4358 if (device->writeable) {
4359 device->fs_devices->total_rw_bytes += device->total_bytes;
4360 spin_lock(&root->fs_info->free_chunk_lock);
4361 root->fs_info->free_chunk_space += device->total_bytes -
4363 spin_unlock(&root->fs_info->free_chunk_lock);
4369 int btrfs_read_sys_array(struct btrfs_root *root)
4371 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4372 struct extent_buffer *sb;
4373 struct btrfs_disk_key *disk_key;
4374 struct btrfs_chunk *chunk;
4376 unsigned long sb_ptr;
4382 struct btrfs_key key;
4384 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4385 BTRFS_SUPER_INFO_SIZE);
4388 btrfs_set_buffer_uptodate(sb);
4389 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4391 * The sb extent buffer is artifical and just used to read the system array.
4392 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4393 * pages up-to-date when the page is larger: extent does not cover the
4394 * whole page and consequently check_page_uptodate does not find all
4395 * the page's extents up-to-date (the hole beyond sb),
4396 * write_extent_buffer then triggers a WARN_ON.
4398 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4399 * but sb spans only this function. Add an explicit SetPageUptodate call
4400 * to silence the warning eg. on PowerPC 64.
4402 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4403 SetPageUptodate(sb->pages[0]);
4405 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4406 array_size = btrfs_super_sys_array_size(super_copy);
4408 ptr = super_copy->sys_chunk_array;
4409 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4412 while (cur < array_size) {
4413 disk_key = (struct btrfs_disk_key *)ptr;
4414 btrfs_disk_key_to_cpu(&key, disk_key);
4416 len = sizeof(*disk_key); ptr += len;
4420 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4421 chunk = (struct btrfs_chunk *)sb_ptr;
4422 ret = read_one_chunk(root, &key, sb, chunk);
4425 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4426 len = btrfs_chunk_item_size(num_stripes);
4435 free_extent_buffer(sb);
4439 int btrfs_read_chunk_tree(struct btrfs_root *root)
4441 struct btrfs_path *path;
4442 struct extent_buffer *leaf;
4443 struct btrfs_key key;
4444 struct btrfs_key found_key;
4448 root = root->fs_info->chunk_root;
4450 path = btrfs_alloc_path();
4454 mutex_lock(&uuid_mutex);
4457 /* first we search for all of the device items, and then we
4458 * read in all of the chunk items. This way we can create chunk
4459 * mappings that reference all of the devices that are afound
4461 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4465 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4469 leaf = path->nodes[0];
4470 slot = path->slots[0];
4471 if (slot >= btrfs_header_nritems(leaf)) {
4472 ret = btrfs_next_leaf(root, path);
4479 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4480 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4481 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4483 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4484 struct btrfs_dev_item *dev_item;
4485 dev_item = btrfs_item_ptr(leaf, slot,
4486 struct btrfs_dev_item);
4487 ret = read_one_dev(root, leaf, dev_item);
4491 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4492 struct btrfs_chunk *chunk;
4493 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4494 ret = read_one_chunk(root, &found_key, leaf, chunk);
4500 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4502 btrfs_release_path(path);
4507 unlock_chunks(root);
4508 mutex_unlock(&uuid_mutex);
4510 btrfs_free_path(path);