2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <linux/ratelimit.h>
27 #include <linux/kthread.h>
28 #include <asm/div64.h>
31 #include "extent_map.h"
33 #include "transaction.h"
34 #include "print-tree.h"
36 #include "async-thread.h"
37 #include "check-integrity.h"
38 #include "rcu-string.h"
40 static int init_first_rw_device(struct btrfs_trans_handle *trans,
41 struct btrfs_root *root,
42 struct btrfs_device *device);
43 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
44 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
45 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
47 static DEFINE_MUTEX(uuid_mutex);
48 static LIST_HEAD(fs_uuids);
50 static void lock_chunks(struct btrfs_root *root)
52 mutex_lock(&root->fs_info->chunk_mutex);
55 static void unlock_chunks(struct btrfs_root *root)
57 mutex_unlock(&root->fs_info->chunk_mutex);
60 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
62 struct btrfs_device *device;
63 WARN_ON(fs_devices->opened);
64 while (!list_empty(&fs_devices->devices)) {
65 device = list_entry(fs_devices->devices.next,
66 struct btrfs_device, dev_list);
67 list_del(&device->dev_list);
68 rcu_string_free(device->name);
74 void btrfs_cleanup_fs_uuids(void)
76 struct btrfs_fs_devices *fs_devices;
78 while (!list_empty(&fs_uuids)) {
79 fs_devices = list_entry(fs_uuids.next,
80 struct btrfs_fs_devices, list);
81 list_del(&fs_devices->list);
82 free_fs_devices(fs_devices);
86 static noinline struct btrfs_device *__find_device(struct list_head *head,
89 struct btrfs_device *dev;
91 list_for_each_entry(dev, head, dev_list) {
92 if (dev->devid == devid &&
93 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
100 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
102 struct btrfs_fs_devices *fs_devices;
104 list_for_each_entry(fs_devices, &fs_uuids, list) {
105 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
111 static void requeue_list(struct btrfs_pending_bios *pending_bios,
112 struct bio *head, struct bio *tail)
115 struct bio *old_head;
117 old_head = pending_bios->head;
118 pending_bios->head = head;
119 if (pending_bios->tail)
120 tail->bi_next = old_head;
122 pending_bios->tail = tail;
126 * we try to collect pending bios for a device so we don't get a large
127 * number of procs sending bios down to the same device. This greatly
128 * improves the schedulers ability to collect and merge the bios.
130 * But, it also turns into a long list of bios to process and that is sure
131 * to eventually make the worker thread block. The solution here is to
132 * make some progress and then put this work struct back at the end of
133 * the list if the block device is congested. This way, multiple devices
134 * can make progress from a single worker thread.
136 static noinline void run_scheduled_bios(struct btrfs_device *device)
139 struct backing_dev_info *bdi;
140 struct btrfs_fs_info *fs_info;
141 struct btrfs_pending_bios *pending_bios;
145 unsigned long num_run;
146 unsigned long batch_run = 0;
148 unsigned long last_waited = 0;
150 int sync_pending = 0;
151 struct blk_plug plug;
154 * this function runs all the bios we've collected for
155 * a particular device. We don't want to wander off to
156 * another device without first sending all of these down.
157 * So, setup a plug here and finish it off before we return
159 blk_start_plug(&plug);
161 bdi = blk_get_backing_dev_info(device->bdev);
162 fs_info = device->dev_root->fs_info;
163 limit = btrfs_async_submit_limit(fs_info);
164 limit = limit * 2 / 3;
167 spin_lock(&device->io_lock);
172 /* take all the bios off the list at once and process them
173 * later on (without the lock held). But, remember the
174 * tail and other pointers so the bios can be properly reinserted
175 * into the list if we hit congestion
177 if (!force_reg && device->pending_sync_bios.head) {
178 pending_bios = &device->pending_sync_bios;
181 pending_bios = &device->pending_bios;
185 pending = pending_bios->head;
186 tail = pending_bios->tail;
187 WARN_ON(pending && !tail);
190 * if pending was null this time around, no bios need processing
191 * at all and we can stop. Otherwise it'll loop back up again
192 * and do an additional check so no bios are missed.
194 * device->running_pending is used to synchronize with the
197 if (device->pending_sync_bios.head == NULL &&
198 device->pending_bios.head == NULL) {
200 device->running_pending = 0;
203 device->running_pending = 1;
206 pending_bios->head = NULL;
207 pending_bios->tail = NULL;
209 spin_unlock(&device->io_lock);
214 /* we want to work on both lists, but do more bios on the
215 * sync list than the regular list
218 pending_bios != &device->pending_sync_bios &&
219 device->pending_sync_bios.head) ||
220 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
221 device->pending_bios.head)) {
222 spin_lock(&device->io_lock);
223 requeue_list(pending_bios, pending, tail);
228 pending = pending->bi_next;
231 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
232 waitqueue_active(&fs_info->async_submit_wait))
233 wake_up(&fs_info->async_submit_wait);
235 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
238 * if we're doing the sync list, record that our
239 * plug has some sync requests on it
241 * If we're doing the regular list and there are
242 * sync requests sitting around, unplug before
245 if (pending_bios == &device->pending_sync_bios) {
247 } else if (sync_pending) {
248 blk_finish_plug(&plug);
249 blk_start_plug(&plug);
253 btrfsic_submit_bio(cur->bi_rw, cur);
260 * we made progress, there is more work to do and the bdi
261 * is now congested. Back off and let other work structs
264 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
265 fs_info->fs_devices->open_devices > 1) {
266 struct io_context *ioc;
268 ioc = current->io_context;
271 * the main goal here is that we don't want to
272 * block if we're going to be able to submit
273 * more requests without blocking.
275 * This code does two great things, it pokes into
276 * the elevator code from a filesystem _and_
277 * it makes assumptions about how batching works.
279 if (ioc && ioc->nr_batch_requests > 0 &&
280 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
282 ioc->last_waited == last_waited)) {
284 * we want to go through our batch of
285 * requests and stop. So, we copy out
286 * the ioc->last_waited time and test
287 * against it before looping
289 last_waited = ioc->last_waited;
294 spin_lock(&device->io_lock);
295 requeue_list(pending_bios, pending, tail);
296 device->running_pending = 1;
298 spin_unlock(&device->io_lock);
299 btrfs_requeue_work(&device->work);
302 /* unplug every 64 requests just for good measure */
303 if (batch_run % 64 == 0) {
304 blk_finish_plug(&plug);
305 blk_start_plug(&plug);
314 spin_lock(&device->io_lock);
315 if (device->pending_bios.head || device->pending_sync_bios.head)
317 spin_unlock(&device->io_lock);
320 blk_finish_plug(&plug);
323 static void pending_bios_fn(struct btrfs_work *work)
325 struct btrfs_device *device;
327 device = container_of(work, struct btrfs_device, work);
328 run_scheduled_bios(device);
331 static noinline int device_list_add(const char *path,
332 struct btrfs_super_block *disk_super,
333 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
335 struct btrfs_device *device;
336 struct btrfs_fs_devices *fs_devices;
337 struct rcu_string *name;
338 u64 found_transid = btrfs_super_generation(disk_super);
340 fs_devices = find_fsid(disk_super->fsid);
342 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
345 INIT_LIST_HEAD(&fs_devices->devices);
346 INIT_LIST_HEAD(&fs_devices->alloc_list);
347 list_add(&fs_devices->list, &fs_uuids);
348 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
349 fs_devices->latest_devid = devid;
350 fs_devices->latest_trans = found_transid;
351 mutex_init(&fs_devices->device_list_mutex);
354 device = __find_device(&fs_devices->devices, devid,
355 disk_super->dev_item.uuid);
358 if (fs_devices->opened)
361 device = kzalloc(sizeof(*device), GFP_NOFS);
363 /* we can safely leave the fs_devices entry around */
366 device->devid = devid;
367 device->dev_stats_valid = 0;
368 device->work.func = pending_bios_fn;
369 memcpy(device->uuid, disk_super->dev_item.uuid,
371 spin_lock_init(&device->io_lock);
373 name = rcu_string_strdup(path, GFP_NOFS);
378 rcu_assign_pointer(device->name, name);
379 INIT_LIST_HEAD(&device->dev_alloc_list);
381 /* init readahead state */
382 spin_lock_init(&device->reada_lock);
383 device->reada_curr_zone = NULL;
384 atomic_set(&device->reada_in_flight, 0);
385 device->reada_next = 0;
386 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
387 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
389 mutex_lock(&fs_devices->device_list_mutex);
390 list_add_rcu(&device->dev_list, &fs_devices->devices);
391 mutex_unlock(&fs_devices->device_list_mutex);
393 device->fs_devices = fs_devices;
394 fs_devices->num_devices++;
395 } else if (!device->name || strcmp(device->name->str, path)) {
396 name = rcu_string_strdup(path, GFP_NOFS);
399 rcu_string_free(device->name);
400 rcu_assign_pointer(device->name, name);
401 if (device->missing) {
402 fs_devices->missing_devices--;
407 if (found_transid > fs_devices->latest_trans) {
408 fs_devices->latest_devid = devid;
409 fs_devices->latest_trans = found_transid;
411 *fs_devices_ret = fs_devices;
415 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
417 struct btrfs_fs_devices *fs_devices;
418 struct btrfs_device *device;
419 struct btrfs_device *orig_dev;
421 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
423 return ERR_PTR(-ENOMEM);
425 INIT_LIST_HEAD(&fs_devices->devices);
426 INIT_LIST_HEAD(&fs_devices->alloc_list);
427 INIT_LIST_HEAD(&fs_devices->list);
428 mutex_init(&fs_devices->device_list_mutex);
429 fs_devices->latest_devid = orig->latest_devid;
430 fs_devices->latest_trans = orig->latest_trans;
431 fs_devices->total_devices = orig->total_devices;
432 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
434 /* We have held the volume lock, it is safe to get the devices. */
435 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
436 struct rcu_string *name;
438 device = kzalloc(sizeof(*device), GFP_NOFS);
443 * This is ok to do without rcu read locked because we hold the
444 * uuid mutex so nothing we touch in here is going to disappear.
446 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
451 rcu_assign_pointer(device->name, name);
453 device->devid = orig_dev->devid;
454 device->work.func = pending_bios_fn;
455 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
456 spin_lock_init(&device->io_lock);
457 INIT_LIST_HEAD(&device->dev_list);
458 INIT_LIST_HEAD(&device->dev_alloc_list);
460 list_add(&device->dev_list, &fs_devices->devices);
461 device->fs_devices = fs_devices;
462 fs_devices->num_devices++;
466 free_fs_devices(fs_devices);
467 return ERR_PTR(-ENOMEM);
470 void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
472 struct btrfs_device *device, *next;
474 struct block_device *latest_bdev = NULL;
475 u64 latest_devid = 0;
476 u64 latest_transid = 0;
478 mutex_lock(&uuid_mutex);
480 /* This is the initialized path, it is safe to release the devices. */
481 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
482 if (device->in_fs_metadata) {
483 if (!latest_transid ||
484 device->generation > latest_transid) {
485 latest_devid = device->devid;
486 latest_transid = device->generation;
487 latest_bdev = device->bdev;
493 blkdev_put(device->bdev, device->mode);
495 fs_devices->open_devices--;
497 if (device->writeable) {
498 list_del_init(&device->dev_alloc_list);
499 device->writeable = 0;
500 fs_devices->rw_devices--;
502 list_del_init(&device->dev_list);
503 fs_devices->num_devices--;
504 rcu_string_free(device->name);
508 if (fs_devices->seed) {
509 fs_devices = fs_devices->seed;
513 fs_devices->latest_bdev = latest_bdev;
514 fs_devices->latest_devid = latest_devid;
515 fs_devices->latest_trans = latest_transid;
517 mutex_unlock(&uuid_mutex);
520 static void __free_device(struct work_struct *work)
522 struct btrfs_device *device;
524 device = container_of(work, struct btrfs_device, rcu_work);
527 blkdev_put(device->bdev, device->mode);
529 rcu_string_free(device->name);
533 static void free_device(struct rcu_head *head)
535 struct btrfs_device *device;
537 device = container_of(head, struct btrfs_device, rcu);
539 INIT_WORK(&device->rcu_work, __free_device);
540 schedule_work(&device->rcu_work);
543 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
545 struct btrfs_device *device;
547 if (--fs_devices->opened > 0)
550 mutex_lock(&fs_devices->device_list_mutex);
551 list_for_each_entry(device, &fs_devices->devices, dev_list) {
552 struct btrfs_device *new_device;
553 struct rcu_string *name;
556 fs_devices->open_devices--;
558 if (device->writeable) {
559 list_del_init(&device->dev_alloc_list);
560 fs_devices->rw_devices--;
563 if (device->can_discard)
564 fs_devices->num_can_discard--;
566 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
567 BUG_ON(!new_device); /* -ENOMEM */
568 memcpy(new_device, device, sizeof(*new_device));
570 /* Safe because we are under uuid_mutex */
572 name = rcu_string_strdup(device->name->str, GFP_NOFS);
573 BUG_ON(device->name && !name); /* -ENOMEM */
574 rcu_assign_pointer(new_device->name, name);
576 new_device->bdev = NULL;
577 new_device->writeable = 0;
578 new_device->in_fs_metadata = 0;
579 new_device->can_discard = 0;
580 list_replace_rcu(&device->dev_list, &new_device->dev_list);
582 call_rcu(&device->rcu, free_device);
584 mutex_unlock(&fs_devices->device_list_mutex);
586 WARN_ON(fs_devices->open_devices);
587 WARN_ON(fs_devices->rw_devices);
588 fs_devices->opened = 0;
589 fs_devices->seeding = 0;
594 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
596 struct btrfs_fs_devices *seed_devices = NULL;
599 mutex_lock(&uuid_mutex);
600 ret = __btrfs_close_devices(fs_devices);
601 if (!fs_devices->opened) {
602 seed_devices = fs_devices->seed;
603 fs_devices->seed = NULL;
605 mutex_unlock(&uuid_mutex);
607 while (seed_devices) {
608 fs_devices = seed_devices;
609 seed_devices = fs_devices->seed;
610 __btrfs_close_devices(fs_devices);
611 free_fs_devices(fs_devices);
616 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
617 fmode_t flags, void *holder)
619 struct request_queue *q;
620 struct block_device *bdev;
621 struct list_head *head = &fs_devices->devices;
622 struct btrfs_device *device;
623 struct block_device *latest_bdev = NULL;
624 struct buffer_head *bh;
625 struct btrfs_super_block *disk_super;
626 u64 latest_devid = 0;
627 u64 latest_transid = 0;
634 list_for_each_entry(device, head, dev_list) {
640 bdev = blkdev_get_by_path(device->name->str, flags, holder);
642 printk(KERN_INFO "btrfs: open %s failed\n", device->name->str);
645 filemap_write_and_wait(bdev->bd_inode->i_mapping);
646 invalidate_bdev(bdev);
647 set_blocksize(bdev, 4096);
649 bh = btrfs_read_dev_super(bdev);
653 disk_super = (struct btrfs_super_block *)bh->b_data;
654 devid = btrfs_stack_device_id(&disk_super->dev_item);
655 if (devid != device->devid)
658 if (memcmp(device->uuid, disk_super->dev_item.uuid,
662 device->generation = btrfs_super_generation(disk_super);
663 if (!latest_transid || device->generation > latest_transid) {
664 latest_devid = devid;
665 latest_transid = device->generation;
669 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
670 device->writeable = 0;
672 device->writeable = !bdev_read_only(bdev);
676 q = bdev_get_queue(bdev);
677 if (blk_queue_discard(q)) {
678 device->can_discard = 1;
679 fs_devices->num_can_discard++;
683 device->in_fs_metadata = 0;
684 device->mode = flags;
686 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
687 fs_devices->rotating = 1;
689 fs_devices->open_devices++;
690 if (device->writeable) {
691 fs_devices->rw_devices++;
692 list_add(&device->dev_alloc_list,
693 &fs_devices->alloc_list);
701 blkdev_put(bdev, flags);
705 if (fs_devices->open_devices == 0) {
709 fs_devices->seeding = seeding;
710 fs_devices->opened = 1;
711 fs_devices->latest_bdev = latest_bdev;
712 fs_devices->latest_devid = latest_devid;
713 fs_devices->latest_trans = latest_transid;
714 fs_devices->total_rw_bytes = 0;
719 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
720 fmode_t flags, void *holder)
724 mutex_lock(&uuid_mutex);
725 if (fs_devices->opened) {
726 fs_devices->opened++;
729 ret = __btrfs_open_devices(fs_devices, flags, holder);
731 mutex_unlock(&uuid_mutex);
735 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
736 struct btrfs_fs_devices **fs_devices_ret)
738 struct btrfs_super_block *disk_super;
739 struct block_device *bdev;
740 struct buffer_head *bh;
747 bdev = blkdev_get_by_path(path, flags, holder);
754 mutex_lock(&uuid_mutex);
755 ret = set_blocksize(bdev, 4096);
758 bh = btrfs_read_dev_super(bdev);
763 disk_super = (struct btrfs_super_block *)bh->b_data;
764 devid = btrfs_stack_device_id(&disk_super->dev_item);
765 transid = btrfs_super_generation(disk_super);
766 total_devices = btrfs_super_num_devices(disk_super);
767 if (disk_super->label[0])
768 printk(KERN_INFO "device label %s ", disk_super->label);
770 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
771 printk(KERN_CONT "devid %llu transid %llu %s\n",
772 (unsigned long long)devid, (unsigned long long)transid, path);
773 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
774 if (!ret && fs_devices_ret)
775 (*fs_devices_ret)->total_devices = total_devices;
778 mutex_unlock(&uuid_mutex);
779 blkdev_put(bdev, flags);
784 /* helper to account the used device space in the range */
785 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
786 u64 end, u64 *length)
788 struct btrfs_key key;
789 struct btrfs_root *root = device->dev_root;
790 struct btrfs_dev_extent *dev_extent;
791 struct btrfs_path *path;
795 struct extent_buffer *l;
799 if (start >= device->total_bytes)
802 path = btrfs_alloc_path();
807 key.objectid = device->devid;
809 key.type = BTRFS_DEV_EXTENT_KEY;
811 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
815 ret = btrfs_previous_item(root, path, key.objectid, key.type);
822 slot = path->slots[0];
823 if (slot >= btrfs_header_nritems(l)) {
824 ret = btrfs_next_leaf(root, path);
832 btrfs_item_key_to_cpu(l, &key, slot);
834 if (key.objectid < device->devid)
837 if (key.objectid > device->devid)
840 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
843 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
844 extent_end = key.offset + btrfs_dev_extent_length(l,
846 if (key.offset <= start && extent_end > end) {
847 *length = end - start + 1;
849 } else if (key.offset <= start && extent_end > start)
850 *length += extent_end - start;
851 else if (key.offset > start && extent_end <= end)
852 *length += extent_end - key.offset;
853 else if (key.offset > start && key.offset <= end) {
854 *length += end - key.offset + 1;
856 } else if (key.offset > end)
864 btrfs_free_path(path);
869 * find_free_dev_extent - find free space in the specified device
870 * @device: the device which we search the free space in
871 * @num_bytes: the size of the free space that we need
872 * @start: store the start of the free space.
873 * @len: the size of the free space. that we find, or the size of the max
874 * free space if we don't find suitable free space
876 * this uses a pretty simple search, the expectation is that it is
877 * called very infrequently and that a given device has a small number
880 * @start is used to store the start of the free space if we find. But if we
881 * don't find suitable free space, it will be used to store the start position
882 * of the max free space.
884 * @len is used to store the size of the free space that we find.
885 * But if we don't find suitable free space, it is used to store the size of
886 * the max free space.
888 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
889 u64 *start, u64 *len)
891 struct btrfs_key key;
892 struct btrfs_root *root = device->dev_root;
893 struct btrfs_dev_extent *dev_extent;
894 struct btrfs_path *path;
900 u64 search_end = device->total_bytes;
903 struct extent_buffer *l;
905 /* FIXME use last free of some kind */
907 /* we don't want to overwrite the superblock on the drive,
908 * so we make sure to start at an offset of at least 1MB
910 search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
912 max_hole_start = search_start;
916 if (search_start >= search_end) {
921 path = btrfs_alloc_path();
928 key.objectid = device->devid;
929 key.offset = search_start;
930 key.type = BTRFS_DEV_EXTENT_KEY;
932 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
936 ret = btrfs_previous_item(root, path, key.objectid, key.type);
943 slot = path->slots[0];
944 if (slot >= btrfs_header_nritems(l)) {
945 ret = btrfs_next_leaf(root, path);
953 btrfs_item_key_to_cpu(l, &key, slot);
955 if (key.objectid < device->devid)
958 if (key.objectid > device->devid)
961 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
964 if (key.offset > search_start) {
965 hole_size = key.offset - search_start;
967 if (hole_size > max_hole_size) {
968 max_hole_start = search_start;
969 max_hole_size = hole_size;
973 * If this free space is greater than which we need,
974 * it must be the max free space that we have found
975 * until now, so max_hole_start must point to the start
976 * of this free space and the length of this free space
977 * is stored in max_hole_size. Thus, we return
978 * max_hole_start and max_hole_size and go back to the
981 if (hole_size >= num_bytes) {
987 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
988 extent_end = key.offset + btrfs_dev_extent_length(l,
990 if (extent_end > search_start)
991 search_start = extent_end;
998 * At this point, search_start should be the end of
999 * allocated dev extents, and when shrinking the device,
1000 * search_end may be smaller than search_start.
1002 if (search_end > search_start)
1003 hole_size = search_end - search_start;
1005 if (hole_size > max_hole_size) {
1006 max_hole_start = search_start;
1007 max_hole_size = hole_size;
1011 if (hole_size < num_bytes)
1017 btrfs_free_path(path);
1019 *start = max_hole_start;
1021 *len = max_hole_size;
1025 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1026 struct btrfs_device *device,
1030 struct btrfs_path *path;
1031 struct btrfs_root *root = device->dev_root;
1032 struct btrfs_key key;
1033 struct btrfs_key found_key;
1034 struct extent_buffer *leaf = NULL;
1035 struct btrfs_dev_extent *extent = NULL;
1037 path = btrfs_alloc_path();
1041 key.objectid = device->devid;
1043 key.type = BTRFS_DEV_EXTENT_KEY;
1045 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1047 ret = btrfs_previous_item(root, path, key.objectid,
1048 BTRFS_DEV_EXTENT_KEY);
1051 leaf = path->nodes[0];
1052 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1053 extent = btrfs_item_ptr(leaf, path->slots[0],
1054 struct btrfs_dev_extent);
1055 BUG_ON(found_key.offset > start || found_key.offset +
1056 btrfs_dev_extent_length(leaf, extent) < start);
1058 btrfs_release_path(path);
1060 } else if (ret == 0) {
1061 leaf = path->nodes[0];
1062 extent = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_dev_extent);
1065 btrfs_error(root->fs_info, ret, "Slot search failed");
1069 if (device->bytes_used > 0) {
1070 u64 len = btrfs_dev_extent_length(leaf, extent);
1071 device->bytes_used -= len;
1072 spin_lock(&root->fs_info->free_chunk_lock);
1073 root->fs_info->free_chunk_space += len;
1074 spin_unlock(&root->fs_info->free_chunk_lock);
1076 ret = btrfs_del_item(trans, root, path);
1078 btrfs_error(root->fs_info, ret,
1079 "Failed to remove dev extent item");
1082 btrfs_free_path(path);
1086 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1087 struct btrfs_device *device,
1088 u64 chunk_tree, u64 chunk_objectid,
1089 u64 chunk_offset, u64 start, u64 num_bytes)
1092 struct btrfs_path *path;
1093 struct btrfs_root *root = device->dev_root;
1094 struct btrfs_dev_extent *extent;
1095 struct extent_buffer *leaf;
1096 struct btrfs_key key;
1098 WARN_ON(!device->in_fs_metadata);
1099 path = btrfs_alloc_path();
1103 key.objectid = device->devid;
1105 key.type = BTRFS_DEV_EXTENT_KEY;
1106 ret = btrfs_insert_empty_item(trans, root, path, &key,
1111 leaf = path->nodes[0];
1112 extent = btrfs_item_ptr(leaf, path->slots[0],
1113 struct btrfs_dev_extent);
1114 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1115 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1116 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1118 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1119 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1122 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1123 btrfs_mark_buffer_dirty(leaf);
1125 btrfs_free_path(path);
1129 static noinline int find_next_chunk(struct btrfs_root *root,
1130 u64 objectid, u64 *offset)
1132 struct btrfs_path *path;
1134 struct btrfs_key key;
1135 struct btrfs_chunk *chunk;
1136 struct btrfs_key found_key;
1138 path = btrfs_alloc_path();
1142 key.objectid = objectid;
1143 key.offset = (u64)-1;
1144 key.type = BTRFS_CHUNK_ITEM_KEY;
1146 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1150 BUG_ON(ret == 0); /* Corruption */
1152 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1156 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1158 if (found_key.objectid != objectid)
1161 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1162 struct btrfs_chunk);
1163 *offset = found_key.offset +
1164 btrfs_chunk_length(path->nodes[0], chunk);
1169 btrfs_free_path(path);
1173 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1176 struct btrfs_key key;
1177 struct btrfs_key found_key;
1178 struct btrfs_path *path;
1180 root = root->fs_info->chunk_root;
1182 path = btrfs_alloc_path();
1186 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1187 key.type = BTRFS_DEV_ITEM_KEY;
1188 key.offset = (u64)-1;
1190 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1194 BUG_ON(ret == 0); /* Corruption */
1196 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1197 BTRFS_DEV_ITEM_KEY);
1201 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1203 *objectid = found_key.offset + 1;
1207 btrfs_free_path(path);
1212 * the device information is stored in the chunk root
1213 * the btrfs_device struct should be fully filled in
1215 int btrfs_add_device(struct btrfs_trans_handle *trans,
1216 struct btrfs_root *root,
1217 struct btrfs_device *device)
1220 struct btrfs_path *path;
1221 struct btrfs_dev_item *dev_item;
1222 struct extent_buffer *leaf;
1223 struct btrfs_key key;
1226 root = root->fs_info->chunk_root;
1228 path = btrfs_alloc_path();
1232 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1233 key.type = BTRFS_DEV_ITEM_KEY;
1234 key.offset = device->devid;
1236 ret = btrfs_insert_empty_item(trans, root, path, &key,
1241 leaf = path->nodes[0];
1242 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1244 btrfs_set_device_id(leaf, dev_item, device->devid);
1245 btrfs_set_device_generation(leaf, dev_item, 0);
1246 btrfs_set_device_type(leaf, dev_item, device->type);
1247 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1248 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1249 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1250 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1251 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1252 btrfs_set_device_group(leaf, dev_item, 0);
1253 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1254 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1255 btrfs_set_device_start_offset(leaf, dev_item, 0);
1257 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1258 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1259 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1260 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1261 btrfs_mark_buffer_dirty(leaf);
1265 btrfs_free_path(path);
1269 static int btrfs_rm_dev_item(struct btrfs_root *root,
1270 struct btrfs_device *device)
1273 struct btrfs_path *path;
1274 struct btrfs_key key;
1275 struct btrfs_trans_handle *trans;
1277 root = root->fs_info->chunk_root;
1279 path = btrfs_alloc_path();
1283 trans = btrfs_start_transaction(root, 0);
1284 if (IS_ERR(trans)) {
1285 btrfs_free_path(path);
1286 return PTR_ERR(trans);
1288 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1289 key.type = BTRFS_DEV_ITEM_KEY;
1290 key.offset = device->devid;
1293 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1302 ret = btrfs_del_item(trans, root, path);
1306 btrfs_free_path(path);
1307 unlock_chunks(root);
1308 btrfs_commit_transaction(trans, root);
1312 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1314 struct btrfs_device *device;
1315 struct btrfs_device *next_device;
1316 struct block_device *bdev;
1317 struct buffer_head *bh = NULL;
1318 struct btrfs_super_block *disk_super;
1319 struct btrfs_fs_devices *cur_devices;
1325 bool clear_super = false;
1327 mutex_lock(&uuid_mutex);
1329 all_avail = root->fs_info->avail_data_alloc_bits |
1330 root->fs_info->avail_system_alloc_bits |
1331 root->fs_info->avail_metadata_alloc_bits;
1333 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1334 root->fs_info->fs_devices->num_devices <= 4) {
1335 printk(KERN_ERR "btrfs: unable to go below four devices "
1341 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1342 root->fs_info->fs_devices->num_devices <= 2) {
1343 printk(KERN_ERR "btrfs: unable to go below two "
1344 "devices on raid1\n");
1349 if (strcmp(device_path, "missing") == 0) {
1350 struct list_head *devices;
1351 struct btrfs_device *tmp;
1354 devices = &root->fs_info->fs_devices->devices;
1356 * It is safe to read the devices since the volume_mutex
1359 list_for_each_entry(tmp, devices, dev_list) {
1360 if (tmp->in_fs_metadata && !tmp->bdev) {
1369 printk(KERN_ERR "btrfs: no missing devices found to "
1374 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1375 root->fs_info->bdev_holder);
1377 ret = PTR_ERR(bdev);
1381 set_blocksize(bdev, 4096);
1382 invalidate_bdev(bdev);
1383 bh = btrfs_read_dev_super(bdev);
1388 disk_super = (struct btrfs_super_block *)bh->b_data;
1389 devid = btrfs_stack_device_id(&disk_super->dev_item);
1390 dev_uuid = disk_super->dev_item.uuid;
1391 device = btrfs_find_device(root, devid, dev_uuid,
1399 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1400 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1406 if (device->writeable) {
1408 list_del_init(&device->dev_alloc_list);
1409 unlock_chunks(root);
1410 root->fs_info->fs_devices->rw_devices--;
1414 ret = btrfs_shrink_device(device, 0);
1418 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1422 spin_lock(&root->fs_info->free_chunk_lock);
1423 root->fs_info->free_chunk_space = device->total_bytes -
1425 spin_unlock(&root->fs_info->free_chunk_lock);
1427 device->in_fs_metadata = 0;
1428 btrfs_scrub_cancel_dev(root, device);
1431 * the device list mutex makes sure that we don't change
1432 * the device list while someone else is writing out all
1433 * the device supers.
1436 cur_devices = device->fs_devices;
1437 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1438 list_del_rcu(&device->dev_list);
1440 device->fs_devices->num_devices--;
1441 device->fs_devices->total_devices--;
1443 if (device->missing)
1444 root->fs_info->fs_devices->missing_devices--;
1446 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1447 struct btrfs_device, dev_list);
1448 if (device->bdev == root->fs_info->sb->s_bdev)
1449 root->fs_info->sb->s_bdev = next_device->bdev;
1450 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1451 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1454 device->fs_devices->open_devices--;
1456 call_rcu(&device->rcu, free_device);
1457 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1459 num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1460 btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1462 if (cur_devices->open_devices == 0) {
1463 struct btrfs_fs_devices *fs_devices;
1464 fs_devices = root->fs_info->fs_devices;
1465 while (fs_devices) {
1466 if (fs_devices->seed == cur_devices)
1468 fs_devices = fs_devices->seed;
1470 fs_devices->seed = cur_devices->seed;
1471 cur_devices->seed = NULL;
1473 __btrfs_close_devices(cur_devices);
1474 unlock_chunks(root);
1475 free_fs_devices(cur_devices);
1478 root->fs_info->num_tolerated_disk_barrier_failures =
1479 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1482 * at this point, the device is zero sized. We want to
1483 * remove it from the devices list and zero out the old super
1486 /* make sure this device isn't detected as part of
1489 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1490 set_buffer_dirty(bh);
1491 sync_dirty_buffer(bh);
1500 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1502 mutex_unlock(&uuid_mutex);
1505 if (device->writeable) {
1507 list_add(&device->dev_alloc_list,
1508 &root->fs_info->fs_devices->alloc_list);
1509 unlock_chunks(root);
1510 root->fs_info->fs_devices->rw_devices++;
1516 * does all the dirty work required for changing file system's UUID.
1518 static int btrfs_prepare_sprout(struct btrfs_root *root)
1520 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1521 struct btrfs_fs_devices *old_devices;
1522 struct btrfs_fs_devices *seed_devices;
1523 struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1524 struct btrfs_device *device;
1527 BUG_ON(!mutex_is_locked(&uuid_mutex));
1528 if (!fs_devices->seeding)
1531 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1535 old_devices = clone_fs_devices(fs_devices);
1536 if (IS_ERR(old_devices)) {
1537 kfree(seed_devices);
1538 return PTR_ERR(old_devices);
1541 list_add(&old_devices->list, &fs_uuids);
1543 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1544 seed_devices->opened = 1;
1545 INIT_LIST_HEAD(&seed_devices->devices);
1546 INIT_LIST_HEAD(&seed_devices->alloc_list);
1547 mutex_init(&seed_devices->device_list_mutex);
1549 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1550 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1552 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1554 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1555 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1556 device->fs_devices = seed_devices;
1559 fs_devices->seeding = 0;
1560 fs_devices->num_devices = 0;
1561 fs_devices->open_devices = 0;
1562 fs_devices->total_devices = 0;
1563 fs_devices->seed = seed_devices;
1565 generate_random_uuid(fs_devices->fsid);
1566 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1567 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1568 super_flags = btrfs_super_flags(disk_super) &
1569 ~BTRFS_SUPER_FLAG_SEEDING;
1570 btrfs_set_super_flags(disk_super, super_flags);
1576 * strore the expected generation for seed devices in device items.
1578 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1579 struct btrfs_root *root)
1581 struct btrfs_path *path;
1582 struct extent_buffer *leaf;
1583 struct btrfs_dev_item *dev_item;
1584 struct btrfs_device *device;
1585 struct btrfs_key key;
1586 u8 fs_uuid[BTRFS_UUID_SIZE];
1587 u8 dev_uuid[BTRFS_UUID_SIZE];
1591 path = btrfs_alloc_path();
1595 root = root->fs_info->chunk_root;
1596 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1598 key.type = BTRFS_DEV_ITEM_KEY;
1601 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1605 leaf = path->nodes[0];
1607 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1608 ret = btrfs_next_leaf(root, path);
1613 leaf = path->nodes[0];
1614 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1615 btrfs_release_path(path);
1619 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1620 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1621 key.type != BTRFS_DEV_ITEM_KEY)
1624 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1625 struct btrfs_dev_item);
1626 devid = btrfs_device_id(leaf, dev_item);
1627 read_extent_buffer(leaf, dev_uuid,
1628 (unsigned long)btrfs_device_uuid(dev_item),
1630 read_extent_buffer(leaf, fs_uuid,
1631 (unsigned long)btrfs_device_fsid(dev_item),
1633 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1634 BUG_ON(!device); /* Logic error */
1636 if (device->fs_devices->seeding) {
1637 btrfs_set_device_generation(leaf, dev_item,
1638 device->generation);
1639 btrfs_mark_buffer_dirty(leaf);
1647 btrfs_free_path(path);
1651 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1653 struct request_queue *q;
1654 struct btrfs_trans_handle *trans;
1655 struct btrfs_device *device;
1656 struct block_device *bdev;
1657 struct list_head *devices;
1658 struct super_block *sb = root->fs_info->sb;
1659 struct rcu_string *name;
1661 int seeding_dev = 0;
1664 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1667 bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1668 root->fs_info->bdev_holder);
1670 return PTR_ERR(bdev);
1672 if (root->fs_info->fs_devices->seeding) {
1674 down_write(&sb->s_umount);
1675 mutex_lock(&uuid_mutex);
1678 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1680 devices = &root->fs_info->fs_devices->devices;
1682 * we have the volume lock, so we don't need the extra
1683 * device list mutex while reading the list here.
1685 list_for_each_entry(device, devices, dev_list) {
1686 if (device->bdev == bdev) {
1692 device = kzalloc(sizeof(*device), GFP_NOFS);
1694 /* we can safely leave the fs_devices entry around */
1699 name = rcu_string_strdup(device_path, GFP_NOFS);
1705 rcu_assign_pointer(device->name, name);
1707 ret = find_next_devid(root, &device->devid);
1709 rcu_string_free(device->name);
1714 trans = btrfs_start_transaction(root, 0);
1715 if (IS_ERR(trans)) {
1716 rcu_string_free(device->name);
1718 ret = PTR_ERR(trans);
1724 q = bdev_get_queue(bdev);
1725 if (blk_queue_discard(q))
1726 device->can_discard = 1;
1727 device->writeable = 1;
1728 device->work.func = pending_bios_fn;
1729 generate_random_uuid(device->uuid);
1730 spin_lock_init(&device->io_lock);
1731 device->generation = trans->transid;
1732 device->io_width = root->sectorsize;
1733 device->io_align = root->sectorsize;
1734 device->sector_size = root->sectorsize;
1735 device->total_bytes = i_size_read(bdev->bd_inode);
1736 device->disk_total_bytes = device->total_bytes;
1737 device->dev_root = root->fs_info->dev_root;
1738 device->bdev = bdev;
1739 device->in_fs_metadata = 1;
1740 device->mode = FMODE_EXCL;
1741 set_blocksize(device->bdev, 4096);
1744 sb->s_flags &= ~MS_RDONLY;
1745 ret = btrfs_prepare_sprout(root);
1746 BUG_ON(ret); /* -ENOMEM */
1749 device->fs_devices = root->fs_info->fs_devices;
1751 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1752 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1753 list_add(&device->dev_alloc_list,
1754 &root->fs_info->fs_devices->alloc_list);
1755 root->fs_info->fs_devices->num_devices++;
1756 root->fs_info->fs_devices->open_devices++;
1757 root->fs_info->fs_devices->rw_devices++;
1758 root->fs_info->fs_devices->total_devices++;
1759 if (device->can_discard)
1760 root->fs_info->fs_devices->num_can_discard++;
1761 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1763 spin_lock(&root->fs_info->free_chunk_lock);
1764 root->fs_info->free_chunk_space += device->total_bytes;
1765 spin_unlock(&root->fs_info->free_chunk_lock);
1767 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1768 root->fs_info->fs_devices->rotating = 1;
1770 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
1771 btrfs_set_super_total_bytes(root->fs_info->super_copy,
1772 total_bytes + device->total_bytes);
1774 total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
1775 btrfs_set_super_num_devices(root->fs_info->super_copy,
1777 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1780 ret = init_first_rw_device(trans, root, device);
1782 btrfs_abort_transaction(trans, root, ret);
1785 ret = btrfs_finish_sprout(trans, root);
1787 btrfs_abort_transaction(trans, root, ret);
1791 ret = btrfs_add_device(trans, root, device);
1793 btrfs_abort_transaction(trans, root, ret);
1799 * we've got more storage, clear any full flags on the space
1802 btrfs_clear_space_info_full(root->fs_info);
1804 unlock_chunks(root);
1805 root->fs_info->num_tolerated_disk_barrier_failures =
1806 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1807 ret = btrfs_commit_transaction(trans, root);
1810 mutex_unlock(&uuid_mutex);
1811 up_write(&sb->s_umount);
1813 if (ret) /* transaction commit */
1816 ret = btrfs_relocate_sys_chunks(root);
1818 btrfs_error(root->fs_info, ret,
1819 "Failed to relocate sys chunks after "
1820 "device initialization. This can be fixed "
1821 "using the \"btrfs balance\" command.");
1827 unlock_chunks(root);
1828 btrfs_end_transaction(trans, root);
1829 rcu_string_free(device->name);
1832 blkdev_put(bdev, FMODE_EXCL);
1834 mutex_unlock(&uuid_mutex);
1835 up_write(&sb->s_umount);
1840 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1841 struct btrfs_device *device)
1844 struct btrfs_path *path;
1845 struct btrfs_root *root;
1846 struct btrfs_dev_item *dev_item;
1847 struct extent_buffer *leaf;
1848 struct btrfs_key key;
1850 root = device->dev_root->fs_info->chunk_root;
1852 path = btrfs_alloc_path();
1856 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1857 key.type = BTRFS_DEV_ITEM_KEY;
1858 key.offset = device->devid;
1860 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1869 leaf = path->nodes[0];
1870 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1872 btrfs_set_device_id(leaf, dev_item, device->devid);
1873 btrfs_set_device_type(leaf, dev_item, device->type);
1874 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1875 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1876 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1877 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1878 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1879 btrfs_mark_buffer_dirty(leaf);
1882 btrfs_free_path(path);
1886 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1887 struct btrfs_device *device, u64 new_size)
1889 struct btrfs_super_block *super_copy =
1890 device->dev_root->fs_info->super_copy;
1891 u64 old_total = btrfs_super_total_bytes(super_copy);
1892 u64 diff = new_size - device->total_bytes;
1894 if (!device->writeable)
1896 if (new_size <= device->total_bytes)
1899 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1900 device->fs_devices->total_rw_bytes += diff;
1902 device->total_bytes = new_size;
1903 device->disk_total_bytes = new_size;
1904 btrfs_clear_space_info_full(device->dev_root->fs_info);
1906 return btrfs_update_device(trans, device);
1909 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1910 struct btrfs_device *device, u64 new_size)
1913 lock_chunks(device->dev_root);
1914 ret = __btrfs_grow_device(trans, device, new_size);
1915 unlock_chunks(device->dev_root);
1919 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1920 struct btrfs_root *root,
1921 u64 chunk_tree, u64 chunk_objectid,
1925 struct btrfs_path *path;
1926 struct btrfs_key key;
1928 root = root->fs_info->chunk_root;
1929 path = btrfs_alloc_path();
1933 key.objectid = chunk_objectid;
1934 key.offset = chunk_offset;
1935 key.type = BTRFS_CHUNK_ITEM_KEY;
1937 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1940 else if (ret > 0) { /* Logic error or corruption */
1941 btrfs_error(root->fs_info, -ENOENT,
1942 "Failed lookup while freeing chunk.");
1947 ret = btrfs_del_item(trans, root, path);
1949 btrfs_error(root->fs_info, ret,
1950 "Failed to delete chunk item.");
1952 btrfs_free_path(path);
1956 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1959 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1960 struct btrfs_disk_key *disk_key;
1961 struct btrfs_chunk *chunk;
1968 struct btrfs_key key;
1970 array_size = btrfs_super_sys_array_size(super_copy);
1972 ptr = super_copy->sys_chunk_array;
1975 while (cur < array_size) {
1976 disk_key = (struct btrfs_disk_key *)ptr;
1977 btrfs_disk_key_to_cpu(&key, disk_key);
1979 len = sizeof(*disk_key);
1981 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1982 chunk = (struct btrfs_chunk *)(ptr + len);
1983 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1984 len += btrfs_chunk_item_size(num_stripes);
1989 if (key.objectid == chunk_objectid &&
1990 key.offset == chunk_offset) {
1991 memmove(ptr, ptr + len, array_size - (cur + len));
1993 btrfs_set_super_sys_array_size(super_copy, array_size);
2002 static int btrfs_relocate_chunk(struct btrfs_root *root,
2003 u64 chunk_tree, u64 chunk_objectid,
2006 struct extent_map_tree *em_tree;
2007 struct btrfs_root *extent_root;
2008 struct btrfs_trans_handle *trans;
2009 struct extent_map *em;
2010 struct map_lookup *map;
2014 root = root->fs_info->chunk_root;
2015 extent_root = root->fs_info->extent_root;
2016 em_tree = &root->fs_info->mapping_tree.map_tree;
2018 ret = btrfs_can_relocate(extent_root, chunk_offset);
2022 /* step one, relocate all the extents inside this chunk */
2023 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2027 trans = btrfs_start_transaction(root, 0);
2028 BUG_ON(IS_ERR(trans));
2033 * step two, delete the device extents and the
2034 * chunk tree entries
2036 read_lock(&em_tree->lock);
2037 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2038 read_unlock(&em_tree->lock);
2040 BUG_ON(!em || em->start > chunk_offset ||
2041 em->start + em->len < chunk_offset);
2042 map = (struct map_lookup *)em->bdev;
2044 for (i = 0; i < map->num_stripes; i++) {
2045 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2046 map->stripes[i].physical);
2049 if (map->stripes[i].dev) {
2050 ret = btrfs_update_device(trans, map->stripes[i].dev);
2054 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2059 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2061 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2062 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2066 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2069 write_lock(&em_tree->lock);
2070 remove_extent_mapping(em_tree, em);
2071 write_unlock(&em_tree->lock);
2076 /* once for the tree */
2077 free_extent_map(em);
2079 free_extent_map(em);
2081 unlock_chunks(root);
2082 btrfs_end_transaction(trans, root);
2086 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2088 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2089 struct btrfs_path *path;
2090 struct extent_buffer *leaf;
2091 struct btrfs_chunk *chunk;
2092 struct btrfs_key key;
2093 struct btrfs_key found_key;
2094 u64 chunk_tree = chunk_root->root_key.objectid;
2096 bool retried = false;
2100 path = btrfs_alloc_path();
2105 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2106 key.offset = (u64)-1;
2107 key.type = BTRFS_CHUNK_ITEM_KEY;
2110 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2113 BUG_ON(ret == 0); /* Corruption */
2115 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2122 leaf = path->nodes[0];
2123 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2125 chunk = btrfs_item_ptr(leaf, path->slots[0],
2126 struct btrfs_chunk);
2127 chunk_type = btrfs_chunk_type(leaf, chunk);
2128 btrfs_release_path(path);
2130 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2131 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2140 if (found_key.offset == 0)
2142 key.offset = found_key.offset - 1;
2145 if (failed && !retried) {
2149 } else if (failed && retried) {
2154 btrfs_free_path(path);
2158 static int insert_balance_item(struct btrfs_root *root,
2159 struct btrfs_balance_control *bctl)
2161 struct btrfs_trans_handle *trans;
2162 struct btrfs_balance_item *item;
2163 struct btrfs_disk_balance_args disk_bargs;
2164 struct btrfs_path *path;
2165 struct extent_buffer *leaf;
2166 struct btrfs_key key;
2169 path = btrfs_alloc_path();
2173 trans = btrfs_start_transaction(root, 0);
2174 if (IS_ERR(trans)) {
2175 btrfs_free_path(path);
2176 return PTR_ERR(trans);
2179 key.objectid = BTRFS_BALANCE_OBJECTID;
2180 key.type = BTRFS_BALANCE_ITEM_KEY;
2183 ret = btrfs_insert_empty_item(trans, root, path, &key,
2188 leaf = path->nodes[0];
2189 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2191 memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2193 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2194 btrfs_set_balance_data(leaf, item, &disk_bargs);
2195 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2196 btrfs_set_balance_meta(leaf, item, &disk_bargs);
2197 btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2198 btrfs_set_balance_sys(leaf, item, &disk_bargs);
2200 btrfs_set_balance_flags(leaf, item, bctl->flags);
2202 btrfs_mark_buffer_dirty(leaf);
2204 btrfs_free_path(path);
2205 err = btrfs_commit_transaction(trans, root);
2211 static int del_balance_item(struct btrfs_root *root)
2213 struct btrfs_trans_handle *trans;
2214 struct btrfs_path *path;
2215 struct btrfs_key key;
2218 path = btrfs_alloc_path();
2222 trans = btrfs_start_transaction(root, 0);
2223 if (IS_ERR(trans)) {
2224 btrfs_free_path(path);
2225 return PTR_ERR(trans);
2228 key.objectid = BTRFS_BALANCE_OBJECTID;
2229 key.type = BTRFS_BALANCE_ITEM_KEY;
2232 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2240 ret = btrfs_del_item(trans, root, path);
2242 btrfs_free_path(path);
2243 err = btrfs_commit_transaction(trans, root);
2250 * This is a heuristic used to reduce the number of chunks balanced on
2251 * resume after balance was interrupted.
2253 static void update_balance_args(struct btrfs_balance_control *bctl)
2256 * Turn on soft mode for chunk types that were being converted.
2258 if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2259 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2260 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2261 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2262 if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2263 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2266 * Turn on usage filter if is not already used. The idea is
2267 * that chunks that we have already balanced should be
2268 * reasonably full. Don't do it for chunks that are being
2269 * converted - that will keep us from relocating unconverted
2270 * (albeit full) chunks.
2272 if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2273 !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2274 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2275 bctl->data.usage = 90;
2277 if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2278 !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2279 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2280 bctl->sys.usage = 90;
2282 if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2283 !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2284 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2285 bctl->meta.usage = 90;
2290 * Should be called with both balance and volume mutexes held to
2291 * serialize other volume operations (add_dev/rm_dev/resize) with
2292 * restriper. Same goes for unset_balance_control.
2294 static void set_balance_control(struct btrfs_balance_control *bctl)
2296 struct btrfs_fs_info *fs_info = bctl->fs_info;
2298 BUG_ON(fs_info->balance_ctl);
2300 spin_lock(&fs_info->balance_lock);
2301 fs_info->balance_ctl = bctl;
2302 spin_unlock(&fs_info->balance_lock);
2305 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2307 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2309 BUG_ON(!fs_info->balance_ctl);
2311 spin_lock(&fs_info->balance_lock);
2312 fs_info->balance_ctl = NULL;
2313 spin_unlock(&fs_info->balance_lock);
2319 * Balance filters. Return 1 if chunk should be filtered out
2320 * (should not be balanced).
2322 static int chunk_profiles_filter(u64 chunk_type,
2323 struct btrfs_balance_args *bargs)
2325 chunk_type = chunk_to_extended(chunk_type) &
2326 BTRFS_EXTENDED_PROFILE_MASK;
2328 if (bargs->profiles & chunk_type)
2334 static u64 div_factor_fine(u64 num, int factor)
2346 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2347 struct btrfs_balance_args *bargs)
2349 struct btrfs_block_group_cache *cache;
2350 u64 chunk_used, user_thresh;
2353 cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2354 chunk_used = btrfs_block_group_used(&cache->item);
2356 user_thresh = div_factor_fine(cache->key.offset, bargs->usage);
2357 if (chunk_used < user_thresh)
2360 btrfs_put_block_group(cache);
2364 static int chunk_devid_filter(struct extent_buffer *leaf,
2365 struct btrfs_chunk *chunk,
2366 struct btrfs_balance_args *bargs)
2368 struct btrfs_stripe *stripe;
2369 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2372 for (i = 0; i < num_stripes; i++) {
2373 stripe = btrfs_stripe_nr(chunk, i);
2374 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2381 /* [pstart, pend) */
2382 static int chunk_drange_filter(struct extent_buffer *leaf,
2383 struct btrfs_chunk *chunk,
2385 struct btrfs_balance_args *bargs)
2387 struct btrfs_stripe *stripe;
2388 int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2394 if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2397 if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2398 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))
2402 factor = num_stripes / factor;
2404 for (i = 0; i < num_stripes; i++) {
2405 stripe = btrfs_stripe_nr(chunk, i);
2406 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2409 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2410 stripe_length = btrfs_chunk_length(leaf, chunk);
2411 do_div(stripe_length, factor);
2413 if (stripe_offset < bargs->pend &&
2414 stripe_offset + stripe_length > bargs->pstart)
2421 /* [vstart, vend) */
2422 static int chunk_vrange_filter(struct extent_buffer *leaf,
2423 struct btrfs_chunk *chunk,
2425 struct btrfs_balance_args *bargs)
2427 if (chunk_offset < bargs->vend &&
2428 chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2429 /* at least part of the chunk is inside this vrange */
2435 static int chunk_soft_convert_filter(u64 chunk_type,
2436 struct btrfs_balance_args *bargs)
2438 if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2441 chunk_type = chunk_to_extended(chunk_type) &
2442 BTRFS_EXTENDED_PROFILE_MASK;
2444 if (bargs->target == chunk_type)
2450 static int should_balance_chunk(struct btrfs_root *root,
2451 struct extent_buffer *leaf,
2452 struct btrfs_chunk *chunk, u64 chunk_offset)
2454 struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2455 struct btrfs_balance_args *bargs = NULL;
2456 u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2459 if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2460 (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2464 if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2465 bargs = &bctl->data;
2466 else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2468 else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2469 bargs = &bctl->meta;
2471 /* profiles filter */
2472 if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2473 chunk_profiles_filter(chunk_type, bargs)) {
2478 if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2479 chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2484 if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2485 chunk_devid_filter(leaf, chunk, bargs)) {
2489 /* drange filter, makes sense only with devid filter */
2490 if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2491 chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2496 if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2497 chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2501 /* soft profile changing mode */
2502 if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2503 chunk_soft_convert_filter(chunk_type, bargs)) {
2510 static u64 div_factor(u64 num, int factor)
2519 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2521 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2522 struct btrfs_root *chunk_root = fs_info->chunk_root;
2523 struct btrfs_root *dev_root = fs_info->dev_root;
2524 struct list_head *devices;
2525 struct btrfs_device *device;
2528 struct btrfs_chunk *chunk;
2529 struct btrfs_path *path;
2530 struct btrfs_key key;
2531 struct btrfs_key found_key;
2532 struct btrfs_trans_handle *trans;
2533 struct extent_buffer *leaf;
2536 int enospc_errors = 0;
2537 bool counting = true;
2539 /* step one make some room on all the devices */
2540 devices = &fs_info->fs_devices->devices;
2541 list_for_each_entry(device, devices, dev_list) {
2542 old_size = device->total_bytes;
2543 size_to_free = div_factor(old_size, 1);
2544 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2545 if (!device->writeable ||
2546 device->total_bytes - device->bytes_used > size_to_free)
2549 ret = btrfs_shrink_device(device, old_size - size_to_free);
2554 trans = btrfs_start_transaction(dev_root, 0);
2555 BUG_ON(IS_ERR(trans));
2557 ret = btrfs_grow_device(trans, device, old_size);
2560 btrfs_end_transaction(trans, dev_root);
2563 /* step two, relocate all the chunks */
2564 path = btrfs_alloc_path();
2570 /* zero out stat counters */
2571 spin_lock(&fs_info->balance_lock);
2572 memset(&bctl->stat, 0, sizeof(bctl->stat));
2573 spin_unlock(&fs_info->balance_lock);
2575 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2576 key.offset = (u64)-1;
2577 key.type = BTRFS_CHUNK_ITEM_KEY;
2580 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2581 atomic_read(&fs_info->balance_cancel_req)) {
2586 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2591 * this shouldn't happen, it means the last relocate
2595 BUG(); /* FIXME break ? */
2597 ret = btrfs_previous_item(chunk_root, path, 0,
2598 BTRFS_CHUNK_ITEM_KEY);
2604 leaf = path->nodes[0];
2605 slot = path->slots[0];
2606 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2608 if (found_key.objectid != key.objectid)
2611 /* chunk zero is special */
2612 if (found_key.offset == 0)
2615 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2618 spin_lock(&fs_info->balance_lock);
2619 bctl->stat.considered++;
2620 spin_unlock(&fs_info->balance_lock);
2623 ret = should_balance_chunk(chunk_root, leaf, chunk,
2625 btrfs_release_path(path);
2630 spin_lock(&fs_info->balance_lock);
2631 bctl->stat.expected++;
2632 spin_unlock(&fs_info->balance_lock);
2636 ret = btrfs_relocate_chunk(chunk_root,
2637 chunk_root->root_key.objectid,
2640 if (ret && ret != -ENOSPC)
2642 if (ret == -ENOSPC) {
2645 spin_lock(&fs_info->balance_lock);
2646 bctl->stat.completed++;
2647 spin_unlock(&fs_info->balance_lock);
2650 key.offset = found_key.offset - 1;
2654 btrfs_release_path(path);
2659 btrfs_free_path(path);
2660 if (enospc_errors) {
2661 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
2671 * alloc_profile_is_valid - see if a given profile is valid and reduced
2672 * @flags: profile to validate
2673 * @extended: if true @flags is treated as an extended profile
2675 static int alloc_profile_is_valid(u64 flags, int extended)
2677 u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
2678 BTRFS_BLOCK_GROUP_PROFILE_MASK);
2680 flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
2682 /* 1) check that all other bits are zeroed */
2686 /* 2) see if profile is reduced */
2688 return !extended; /* "0" is valid for usual profiles */
2690 /* true if exactly one bit set */
2691 return (flags & (flags - 1)) == 0;
2694 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
2696 /* cancel requested || normal exit path */
2697 return atomic_read(&fs_info->balance_cancel_req) ||
2698 (atomic_read(&fs_info->balance_pause_req) == 0 &&
2699 atomic_read(&fs_info->balance_cancel_req) == 0);
2702 static void __cancel_balance(struct btrfs_fs_info *fs_info)
2706 unset_balance_control(fs_info);
2707 ret = del_balance_item(fs_info->tree_root);
2711 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
2712 struct btrfs_ioctl_balance_args *bargs);
2715 * Should be called with both balance and volume mutexes held
2717 int btrfs_balance(struct btrfs_balance_control *bctl,
2718 struct btrfs_ioctl_balance_args *bargs)
2720 struct btrfs_fs_info *fs_info = bctl->fs_info;
2725 if (btrfs_fs_closing(fs_info) ||
2726 atomic_read(&fs_info->balance_pause_req) ||
2727 atomic_read(&fs_info->balance_cancel_req)) {
2732 allowed = btrfs_super_incompat_flags(fs_info->super_copy);
2733 if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
2737 * In case of mixed groups both data and meta should be picked,
2738 * and identical options should be given for both of them.
2740 allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
2741 if (mixed && (bctl->flags & allowed)) {
2742 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
2743 !(bctl->flags & BTRFS_BALANCE_METADATA) ||
2744 memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
2745 printk(KERN_ERR "btrfs: with mixed groups data and "
2746 "metadata balance options must be the same\n");
2752 allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
2753 if (fs_info->fs_devices->num_devices == 1)
2754 allowed |= BTRFS_BLOCK_GROUP_DUP;
2755 else if (fs_info->fs_devices->num_devices < 4)
2756 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
2758 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2759 BTRFS_BLOCK_GROUP_RAID10);
2761 if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2762 (!alloc_profile_is_valid(bctl->data.target, 1) ||
2763 (bctl->data.target & ~allowed))) {
2764 printk(KERN_ERR "btrfs: unable to start balance with target "
2765 "data profile %llu\n",
2766 (unsigned long long)bctl->data.target);
2770 if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2771 (!alloc_profile_is_valid(bctl->meta.target, 1) ||
2772 (bctl->meta.target & ~allowed))) {
2773 printk(KERN_ERR "btrfs: unable to start balance with target "
2774 "metadata profile %llu\n",
2775 (unsigned long long)bctl->meta.target);
2779 if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2780 (!alloc_profile_is_valid(bctl->sys.target, 1) ||
2781 (bctl->sys.target & ~allowed))) {
2782 printk(KERN_ERR "btrfs: unable to start balance with target "
2783 "system profile %llu\n",
2784 (unsigned long long)bctl->sys.target);
2789 /* allow dup'ed data chunks only in mixed mode */
2790 if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2791 (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
2792 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
2797 /* allow to reduce meta or sys integrity only if force set */
2798 allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
2799 BTRFS_BLOCK_GROUP_RAID10;
2800 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2801 (fs_info->avail_system_alloc_bits & allowed) &&
2802 !(bctl->sys.target & allowed)) ||
2803 ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
2804 (fs_info->avail_metadata_alloc_bits & allowed) &&
2805 !(bctl->meta.target & allowed))) {
2806 if (bctl->flags & BTRFS_BALANCE_FORCE) {
2807 printk(KERN_INFO "btrfs: force reducing metadata "
2810 printk(KERN_ERR "btrfs: balance will reduce metadata "
2811 "integrity, use force if you want this\n");
2817 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2818 int num_tolerated_disk_barrier_failures;
2819 u64 target = bctl->sys.target;
2821 num_tolerated_disk_barrier_failures =
2822 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2823 if (num_tolerated_disk_barrier_failures > 0 &&
2825 (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
2826 BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
2827 num_tolerated_disk_barrier_failures = 0;
2828 else if (num_tolerated_disk_barrier_failures > 1 &&
2830 (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
2831 num_tolerated_disk_barrier_failures = 1;
2833 fs_info->num_tolerated_disk_barrier_failures =
2834 num_tolerated_disk_barrier_failures;
2837 ret = insert_balance_item(fs_info->tree_root, bctl);
2838 if (ret && ret != -EEXIST)
2841 if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
2842 BUG_ON(ret == -EEXIST);
2843 set_balance_control(bctl);
2845 BUG_ON(ret != -EEXIST);
2846 spin_lock(&fs_info->balance_lock);
2847 update_balance_args(bctl);
2848 spin_unlock(&fs_info->balance_lock);
2851 atomic_inc(&fs_info->balance_running);
2852 mutex_unlock(&fs_info->balance_mutex);
2854 ret = __btrfs_balance(fs_info);
2856 mutex_lock(&fs_info->balance_mutex);
2857 atomic_dec(&fs_info->balance_running);
2860 memset(bargs, 0, sizeof(*bargs));
2861 update_ioctl_balance_args(fs_info, 0, bargs);
2864 if ((ret && ret != -ECANCELED && ret != -ENOSPC) ||
2865 balance_need_close(fs_info)) {
2866 __cancel_balance(fs_info);
2869 if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
2870 fs_info->num_tolerated_disk_barrier_failures =
2871 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2874 wake_up(&fs_info->balance_wait_q);
2878 if (bctl->flags & BTRFS_BALANCE_RESUME)
2879 __cancel_balance(fs_info);
2885 static int balance_kthread(void *data)
2887 struct btrfs_fs_info *fs_info = data;
2890 mutex_lock(&fs_info->volume_mutex);
2891 mutex_lock(&fs_info->balance_mutex);
2893 if (fs_info->balance_ctl) {
2894 printk(KERN_INFO "btrfs: continuing balance\n");
2895 ret = btrfs_balance(fs_info->balance_ctl, NULL);
2898 mutex_unlock(&fs_info->balance_mutex);
2899 mutex_unlock(&fs_info->volume_mutex);
2904 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
2906 struct task_struct *tsk;
2908 spin_lock(&fs_info->balance_lock);
2909 if (!fs_info->balance_ctl) {
2910 spin_unlock(&fs_info->balance_lock);
2913 spin_unlock(&fs_info->balance_lock);
2915 if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
2916 printk(KERN_INFO "btrfs: force skipping balance\n");
2920 tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
2922 return PTR_ERR(tsk);
2927 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
2929 struct btrfs_balance_control *bctl;
2930 struct btrfs_balance_item *item;
2931 struct btrfs_disk_balance_args disk_bargs;
2932 struct btrfs_path *path;
2933 struct extent_buffer *leaf;
2934 struct btrfs_key key;
2937 path = btrfs_alloc_path();
2941 key.objectid = BTRFS_BALANCE_OBJECTID;
2942 key.type = BTRFS_BALANCE_ITEM_KEY;
2945 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
2948 if (ret > 0) { /* ret = -ENOENT; */
2953 bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
2959 leaf = path->nodes[0];
2960 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2962 bctl->fs_info = fs_info;
2963 bctl->flags = btrfs_balance_flags(leaf, item);
2964 bctl->flags |= BTRFS_BALANCE_RESUME;
2966 btrfs_balance_data(leaf, item, &disk_bargs);
2967 btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
2968 btrfs_balance_meta(leaf, item, &disk_bargs);
2969 btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
2970 btrfs_balance_sys(leaf, item, &disk_bargs);
2971 btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
2973 mutex_lock(&fs_info->volume_mutex);
2974 mutex_lock(&fs_info->balance_mutex);
2976 set_balance_control(bctl);
2978 mutex_unlock(&fs_info->balance_mutex);
2979 mutex_unlock(&fs_info->volume_mutex);
2981 btrfs_free_path(path);
2985 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
2989 mutex_lock(&fs_info->balance_mutex);
2990 if (!fs_info->balance_ctl) {
2991 mutex_unlock(&fs_info->balance_mutex);
2995 if (atomic_read(&fs_info->balance_running)) {
2996 atomic_inc(&fs_info->balance_pause_req);
2997 mutex_unlock(&fs_info->balance_mutex);
2999 wait_event(fs_info->balance_wait_q,
3000 atomic_read(&fs_info->balance_running) == 0);
3002 mutex_lock(&fs_info->balance_mutex);
3003 /* we are good with balance_ctl ripped off from under us */
3004 BUG_ON(atomic_read(&fs_info->balance_running));
3005 atomic_dec(&fs_info->balance_pause_req);
3010 mutex_unlock(&fs_info->balance_mutex);
3014 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3016 mutex_lock(&fs_info->balance_mutex);
3017 if (!fs_info->balance_ctl) {
3018 mutex_unlock(&fs_info->balance_mutex);
3022 atomic_inc(&fs_info->balance_cancel_req);
3024 * if we are running just wait and return, balance item is
3025 * deleted in btrfs_balance in this case
3027 if (atomic_read(&fs_info->balance_running)) {
3028 mutex_unlock(&fs_info->balance_mutex);
3029 wait_event(fs_info->balance_wait_q,
3030 atomic_read(&fs_info->balance_running) == 0);
3031 mutex_lock(&fs_info->balance_mutex);
3033 /* __cancel_balance needs volume_mutex */
3034 mutex_unlock(&fs_info->balance_mutex);
3035 mutex_lock(&fs_info->volume_mutex);
3036 mutex_lock(&fs_info->balance_mutex);
3038 if (fs_info->balance_ctl)
3039 __cancel_balance(fs_info);
3041 mutex_unlock(&fs_info->volume_mutex);
3044 BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3045 atomic_dec(&fs_info->balance_cancel_req);
3046 mutex_unlock(&fs_info->balance_mutex);
3051 * shrinking a device means finding all of the device extents past
3052 * the new size, and then following the back refs to the chunks.
3053 * The chunk relocation code actually frees the device extent
3055 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3057 struct btrfs_trans_handle *trans;
3058 struct btrfs_root *root = device->dev_root;
3059 struct btrfs_dev_extent *dev_extent = NULL;
3060 struct btrfs_path *path;
3068 bool retried = false;
3069 struct extent_buffer *l;
3070 struct btrfs_key key;
3071 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3072 u64 old_total = btrfs_super_total_bytes(super_copy);
3073 u64 old_size = device->total_bytes;
3074 u64 diff = device->total_bytes - new_size;
3076 if (new_size >= device->total_bytes)
3079 path = btrfs_alloc_path();
3087 device->total_bytes = new_size;
3088 if (device->writeable) {
3089 device->fs_devices->total_rw_bytes -= diff;
3090 spin_lock(&root->fs_info->free_chunk_lock);
3091 root->fs_info->free_chunk_space -= diff;
3092 spin_unlock(&root->fs_info->free_chunk_lock);
3094 unlock_chunks(root);
3097 key.objectid = device->devid;
3098 key.offset = (u64)-1;
3099 key.type = BTRFS_DEV_EXTENT_KEY;
3102 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3106 ret = btrfs_previous_item(root, path, 0, key.type);
3111 btrfs_release_path(path);
3116 slot = path->slots[0];
3117 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3119 if (key.objectid != device->devid) {
3120 btrfs_release_path(path);
3124 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3125 length = btrfs_dev_extent_length(l, dev_extent);
3127 if (key.offset + length <= new_size) {
3128 btrfs_release_path(path);
3132 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3133 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3134 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3135 btrfs_release_path(path);
3137 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3139 if (ret && ret != -ENOSPC)
3143 } while (key.offset-- > 0);
3145 if (failed && !retried) {
3149 } else if (failed && retried) {
3153 device->total_bytes = old_size;
3154 if (device->writeable)
3155 device->fs_devices->total_rw_bytes += diff;
3156 spin_lock(&root->fs_info->free_chunk_lock);
3157 root->fs_info->free_chunk_space += diff;
3158 spin_unlock(&root->fs_info->free_chunk_lock);
3159 unlock_chunks(root);
3163 /* Shrinking succeeded, else we would be at "done". */
3164 trans = btrfs_start_transaction(root, 0);
3165 if (IS_ERR(trans)) {
3166 ret = PTR_ERR(trans);
3172 device->disk_total_bytes = new_size;
3173 /* Now btrfs_update_device() will change the on-disk size. */
3174 ret = btrfs_update_device(trans, device);
3176 unlock_chunks(root);
3177 btrfs_end_transaction(trans, root);
3180 WARN_ON(diff > old_total);
3181 btrfs_set_super_total_bytes(super_copy, old_total - diff);
3182 unlock_chunks(root);
3183 btrfs_end_transaction(trans, root);
3185 btrfs_free_path(path);
3189 static int btrfs_add_system_chunk(struct btrfs_root *root,
3190 struct btrfs_key *key,
3191 struct btrfs_chunk *chunk, int item_size)
3193 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3194 struct btrfs_disk_key disk_key;
3198 array_size = btrfs_super_sys_array_size(super_copy);
3199 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3202 ptr = super_copy->sys_chunk_array + array_size;
3203 btrfs_cpu_key_to_disk(&disk_key, key);
3204 memcpy(ptr, &disk_key, sizeof(disk_key));
3205 ptr += sizeof(disk_key);
3206 memcpy(ptr, chunk, item_size);
3207 item_size += sizeof(disk_key);
3208 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3213 * sort the devices in descending order by max_avail, total_avail
3215 static int btrfs_cmp_device_info(const void *a, const void *b)
3217 const struct btrfs_device_info *di_a = a;
3218 const struct btrfs_device_info *di_b = b;
3220 if (di_a->max_avail > di_b->max_avail)
3222 if (di_a->max_avail < di_b->max_avail)
3224 if (di_a->total_avail > di_b->total_avail)
3226 if (di_a->total_avail < di_b->total_avail)
3231 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3232 struct btrfs_root *extent_root,
3233 struct map_lookup **map_ret,
3234 u64 *num_bytes_out, u64 *stripe_size_out,
3235 u64 start, u64 type)
3237 struct btrfs_fs_info *info = extent_root->fs_info;
3238 struct btrfs_fs_devices *fs_devices = info->fs_devices;
3239 struct list_head *cur;
3240 struct map_lookup *map = NULL;
3241 struct extent_map_tree *em_tree;
3242 struct extent_map *em;
3243 struct btrfs_device_info *devices_info = NULL;
3245 int num_stripes; /* total number of stripes to allocate */
3246 int sub_stripes; /* sub_stripes info for map */
3247 int dev_stripes; /* stripes per dev */
3248 int devs_max; /* max devs to use */
3249 int devs_min; /* min devs needed */
3250 int devs_increment; /* ndevs has to be a multiple of this */
3251 int ncopies; /* how many copies to data has */
3253 u64 max_stripe_size;
3261 BUG_ON(!alloc_profile_is_valid(type, 0));
3263 if (list_empty(&fs_devices->alloc_list))
3270 devs_max = 0; /* 0 == as many as possible */
3274 * define the properties of each RAID type.
3275 * FIXME: move this to a global table and use it in all RAID
3278 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
3282 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
3284 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
3289 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
3298 if (type & BTRFS_BLOCK_GROUP_DATA) {
3299 max_stripe_size = 1024 * 1024 * 1024;
3300 max_chunk_size = 10 * max_stripe_size;
3301 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3302 /* for larger filesystems, use larger metadata chunks */
3303 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3304 max_stripe_size = 1024 * 1024 * 1024;
3306 max_stripe_size = 256 * 1024 * 1024;
3307 max_chunk_size = max_stripe_size;
3308 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3309 max_stripe_size = 32 * 1024 * 1024;
3310 max_chunk_size = 2 * max_stripe_size;
3312 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3317 /* we don't want a chunk larger than 10% of writeable space */
3318 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3321 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3326 cur = fs_devices->alloc_list.next;
3329 * in the first pass through the devices list, we gather information
3330 * about the available holes on each device.
3333 while (cur != &fs_devices->alloc_list) {
3334 struct btrfs_device *device;
3338 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3342 if (!device->writeable) {
3344 "btrfs: read-only device in alloc_list\n");
3349 if (!device->in_fs_metadata)
3352 if (device->total_bytes > device->bytes_used)
3353 total_avail = device->total_bytes - device->bytes_used;
3357 /* If there is no space on this device, skip it. */
3358 if (total_avail == 0)
3361 ret = find_free_dev_extent(device,
3362 max_stripe_size * dev_stripes,
3363 &dev_offset, &max_avail);
3364 if (ret && ret != -ENOSPC)
3368 max_avail = max_stripe_size * dev_stripes;
3370 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3373 devices_info[ndevs].dev_offset = dev_offset;
3374 devices_info[ndevs].max_avail = max_avail;
3375 devices_info[ndevs].total_avail = total_avail;
3376 devices_info[ndevs].dev = device;
3381 * now sort the devices by hole size / available space
3383 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3384 btrfs_cmp_device_info, NULL);
3386 /* round down to number of usable stripes */
3387 ndevs -= ndevs % devs_increment;
3389 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3394 if (devs_max && ndevs > devs_max)
3397 * the primary goal is to maximize the number of stripes, so use as many
3398 * devices as possible, even if the stripes are not maximum sized.
3400 stripe_size = devices_info[ndevs-1].max_avail;
3401 num_stripes = ndevs * dev_stripes;
3403 if (stripe_size * ndevs > max_chunk_size * ncopies) {
3404 stripe_size = max_chunk_size * ncopies;
3405 do_div(stripe_size, ndevs);
3408 do_div(stripe_size, dev_stripes);
3410 /* align to BTRFS_STRIPE_LEN */
3411 do_div(stripe_size, BTRFS_STRIPE_LEN);
3412 stripe_size *= BTRFS_STRIPE_LEN;
3414 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3419 map->num_stripes = num_stripes;
3421 for (i = 0; i < ndevs; ++i) {
3422 for (j = 0; j < dev_stripes; ++j) {
3423 int s = i * dev_stripes + j;
3424 map->stripes[s].dev = devices_info[i].dev;
3425 map->stripes[s].physical = devices_info[i].dev_offset +
3429 map->sector_size = extent_root->sectorsize;
3430 map->stripe_len = BTRFS_STRIPE_LEN;
3431 map->io_align = BTRFS_STRIPE_LEN;
3432 map->io_width = BTRFS_STRIPE_LEN;
3434 map->sub_stripes = sub_stripes;
3437 num_bytes = stripe_size * (num_stripes / ncopies);
3439 *stripe_size_out = stripe_size;
3440 *num_bytes_out = num_bytes;
3442 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3444 em = alloc_extent_map();
3449 em->bdev = (struct block_device *)map;
3451 em->len = num_bytes;
3452 em->block_start = 0;
3453 em->block_len = em->len;
3455 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3456 write_lock(&em_tree->lock);
3457 ret = add_extent_mapping(em_tree, em);
3458 write_unlock(&em_tree->lock);
3459 free_extent_map(em);
3463 ret = btrfs_make_block_group(trans, extent_root, 0, type,
3464 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3469 for (i = 0; i < map->num_stripes; ++i) {
3470 struct btrfs_device *device;
3473 device = map->stripes[i].dev;
3474 dev_offset = map->stripes[i].physical;
3476 ret = btrfs_alloc_dev_extent(trans, device,
3477 info->chunk_root->root_key.objectid,
3478 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3479 start, dev_offset, stripe_size);
3481 btrfs_abort_transaction(trans, extent_root, ret);
3486 kfree(devices_info);
3491 kfree(devices_info);
3495 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3496 struct btrfs_root *extent_root,
3497 struct map_lookup *map, u64 chunk_offset,
3498 u64 chunk_size, u64 stripe_size)
3501 struct btrfs_key key;
3502 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3503 struct btrfs_device *device;
3504 struct btrfs_chunk *chunk;
3505 struct btrfs_stripe *stripe;
3506 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3510 chunk = kzalloc(item_size, GFP_NOFS);
3515 while (index < map->num_stripes) {
3516 device = map->stripes[index].dev;
3517 device->bytes_used += stripe_size;
3518 ret = btrfs_update_device(trans, device);
3524 spin_lock(&extent_root->fs_info->free_chunk_lock);
3525 extent_root->fs_info->free_chunk_space -= (stripe_size *
3527 spin_unlock(&extent_root->fs_info->free_chunk_lock);
3530 stripe = &chunk->stripe;
3531 while (index < map->num_stripes) {
3532 device = map->stripes[index].dev;
3533 dev_offset = map->stripes[index].physical;
3535 btrfs_set_stack_stripe_devid(stripe, device->devid);
3536 btrfs_set_stack_stripe_offset(stripe, dev_offset);
3537 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
3542 btrfs_set_stack_chunk_length(chunk, chunk_size);
3543 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
3544 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
3545 btrfs_set_stack_chunk_type(chunk, map->type);
3546 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
3547 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
3548 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
3549 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
3550 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
3552 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
3553 key.type = BTRFS_CHUNK_ITEM_KEY;
3554 key.offset = chunk_offset;
3556 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
3558 if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
3560 * TODO: Cleanup of inserted chunk root in case of
3563 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
3573 * Chunk allocation falls into two parts. The first part does works
3574 * that make the new allocated chunk useable, but not do any operation
3575 * that modifies the chunk tree. The second part does the works that
3576 * require modifying the chunk tree. This division is important for the
3577 * bootstrap process of adding storage to a seed btrfs.
3579 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3580 struct btrfs_root *extent_root, u64 type)
3585 struct map_lookup *map;
3586 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3589 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3594 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3595 &stripe_size, chunk_offset, type);
3599 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3600 chunk_size, stripe_size);
3606 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
3607 struct btrfs_root *root,
3608 struct btrfs_device *device)
3611 u64 sys_chunk_offset;
3615 u64 sys_stripe_size;
3617 struct map_lookup *map;
3618 struct map_lookup *sys_map;
3619 struct btrfs_fs_info *fs_info = root->fs_info;
3620 struct btrfs_root *extent_root = fs_info->extent_root;
3623 ret = find_next_chunk(fs_info->chunk_root,
3624 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
3628 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
3629 fs_info->avail_metadata_alloc_bits;
3630 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3632 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
3633 &stripe_size, chunk_offset, alloc_profile);
3637 sys_chunk_offset = chunk_offset + chunk_size;
3639 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
3640 fs_info->avail_system_alloc_bits;
3641 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
3643 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
3644 &sys_chunk_size, &sys_stripe_size,
3645 sys_chunk_offset, alloc_profile);
3647 btrfs_abort_transaction(trans, root, ret);
3651 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
3653 btrfs_abort_transaction(trans, root, ret);
3658 * Modifying chunk tree needs allocating new blocks from both
3659 * system block group and metadata block group. So we only can
3660 * do operations require modifying the chunk tree after both
3661 * block groups were created.
3663 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
3664 chunk_size, stripe_size);
3666 btrfs_abort_transaction(trans, root, ret);
3670 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
3671 sys_chunk_offset, sys_chunk_size,
3674 btrfs_abort_transaction(trans, root, ret);
3681 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
3683 struct extent_map *em;
3684 struct map_lookup *map;
3685 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3689 read_lock(&map_tree->map_tree.lock);
3690 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
3691 read_unlock(&map_tree->map_tree.lock);
3695 if (btrfs_test_opt(root, DEGRADED)) {
3696 free_extent_map(em);
3700 map = (struct map_lookup *)em->bdev;
3701 for (i = 0; i < map->num_stripes; i++) {
3702 if (!map->stripes[i].dev->writeable) {
3707 free_extent_map(em);
3711 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
3713 extent_map_tree_init(&tree->map_tree);
3716 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
3718 struct extent_map *em;
3721 write_lock(&tree->map_tree.lock);
3722 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
3724 remove_extent_mapping(&tree->map_tree, em);
3725 write_unlock(&tree->map_tree.lock);
3730 free_extent_map(em);
3731 /* once for the tree */
3732 free_extent_map(em);
3736 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
3738 struct extent_map *em;
3739 struct map_lookup *map;
3740 struct extent_map_tree *em_tree = &map_tree->map_tree;
3743 read_lock(&em_tree->lock);
3744 em = lookup_extent_mapping(em_tree, logical, len);
3745 read_unlock(&em_tree->lock);
3748 BUG_ON(em->start > logical || em->start + em->len < logical);
3749 map = (struct map_lookup *)em->bdev;
3750 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
3751 ret = map->num_stripes;
3752 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3753 ret = map->sub_stripes;
3756 free_extent_map(em);
3760 static int find_live_mirror(struct map_lookup *map, int first, int num,
3764 if (map->stripes[optimal].dev->bdev)
3766 for (i = first; i < first + num; i++) {
3767 if (map->stripes[i].dev->bdev)
3770 /* we couldn't find one that doesn't fail. Just return something
3771 * and the io error handling code will clean up eventually
3776 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3777 u64 logical, u64 *length,
3778 struct btrfs_bio **bbio_ret,
3781 struct extent_map *em;
3782 struct map_lookup *map;
3783 struct extent_map_tree *em_tree = &map_tree->map_tree;
3786 u64 stripe_end_offset;
3795 struct btrfs_bio *bbio = NULL;
3797 read_lock(&em_tree->lock);
3798 em = lookup_extent_mapping(em_tree, logical, *length);
3799 read_unlock(&em_tree->lock);
3802 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
3803 (unsigned long long)logical,
3804 (unsigned long long)*length);
3808 BUG_ON(em->start > logical || em->start + em->len < logical);
3809 map = (struct map_lookup *)em->bdev;
3810 offset = logical - em->start;
3812 if (mirror_num > map->num_stripes)
3817 * stripe_nr counts the total number of stripes we have to stride
3818 * to get to this block
3820 do_div(stripe_nr, map->stripe_len);
3822 stripe_offset = stripe_nr * map->stripe_len;
3823 BUG_ON(offset < stripe_offset);
3825 /* stripe_offset is the offset of this block in its stripe*/
3826 stripe_offset = offset - stripe_offset;
3828 if (rw & REQ_DISCARD)
3829 *length = min_t(u64, em->len - offset, *length);
3830 else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
3831 /* we limit the length of each bio to what fits in a stripe */
3832 *length = min_t(u64, em->len - offset,
3833 map->stripe_len - stripe_offset);
3835 *length = em->len - offset;
3843 stripe_nr_orig = stripe_nr;
3844 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
3845 (~(map->stripe_len - 1));
3846 do_div(stripe_nr_end, map->stripe_len);
3847 stripe_end_offset = stripe_nr_end * map->stripe_len -
3849 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3850 if (rw & REQ_DISCARD)
3851 num_stripes = min_t(u64, map->num_stripes,
3852 stripe_nr_end - stripe_nr_orig);
3853 stripe_index = do_div(stripe_nr, map->num_stripes);
3854 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3855 if (rw & (REQ_WRITE | REQ_DISCARD))
3856 num_stripes = map->num_stripes;
3857 else if (mirror_num)
3858 stripe_index = mirror_num - 1;
3860 stripe_index = find_live_mirror(map, 0,
3862 current->pid % map->num_stripes);
3863 mirror_num = stripe_index + 1;
3866 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3867 if (rw & (REQ_WRITE | REQ_DISCARD)) {
3868 num_stripes = map->num_stripes;
3869 } else if (mirror_num) {
3870 stripe_index = mirror_num - 1;
3875 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3876 int factor = map->num_stripes / map->sub_stripes;
3878 stripe_index = do_div(stripe_nr, factor);
3879 stripe_index *= map->sub_stripes;
3882 num_stripes = map->sub_stripes;
3883 else if (rw & REQ_DISCARD)
3884 num_stripes = min_t(u64, map->sub_stripes *
3885 (stripe_nr_end - stripe_nr_orig),
3887 else if (mirror_num)
3888 stripe_index += mirror_num - 1;
3890 int old_stripe_index = stripe_index;
3891 stripe_index = find_live_mirror(map, stripe_index,
3892 map->sub_stripes, stripe_index +
3893 current->pid % map->sub_stripes);
3894 mirror_num = stripe_index - old_stripe_index + 1;
3898 * after this do_div call, stripe_nr is the number of stripes
3899 * on this device we have to walk to find the data, and
3900 * stripe_index is the number of our device in the stripe array
3902 stripe_index = do_div(stripe_nr, map->num_stripes);
3903 mirror_num = stripe_index + 1;
3905 BUG_ON(stripe_index >= map->num_stripes);
3907 bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS);
3912 atomic_set(&bbio->error, 0);
3914 if (rw & REQ_DISCARD) {
3916 int sub_stripes = 0;
3917 u64 stripes_per_dev = 0;
3918 u32 remaining_stripes = 0;
3919 u32 last_stripe = 0;
3922 (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
3923 if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3926 sub_stripes = map->sub_stripes;
3928 factor = map->num_stripes / sub_stripes;
3929 stripes_per_dev = div_u64_rem(stripe_nr_end -
3932 &remaining_stripes);
3933 div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
3934 last_stripe *= sub_stripes;
3937 for (i = 0; i < num_stripes; i++) {
3938 bbio->stripes[i].physical =
3939 map->stripes[stripe_index].physical +
3940 stripe_offset + stripe_nr * map->stripe_len;
3941 bbio->stripes[i].dev = map->stripes[stripe_index].dev;
3943 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
3944 BTRFS_BLOCK_GROUP_RAID10)) {
3945 bbio->stripes[i].length = stripes_per_dev *
3948 if (i / sub_stripes < remaining_stripes)
3949 bbio->stripes[i].length +=
3953 * Special for the first stripe and
3956 * |-------|...|-------|
3960 if (i < sub_stripes)
3961 bbio->stripes[i].length -=
3964 if (stripe_index >= last_stripe &&
3965 stripe_index <= (last_stripe +
3967 bbio->stripes[i].length -=
3970 if (i == sub_stripes - 1)
3973 bbio->stripes[i].length = *length;
3976 if (stripe_index == map->num_stripes) {
3977 /* This could only happen for RAID0/10 */
3983 for (i = 0; i < num_stripes; i++) {
3984 bbio->stripes[i].physical =
3985 map->stripes[stripe_index].physical +
3987 stripe_nr * map->stripe_len;
3988 bbio->stripes[i].dev =
3989 map->stripes[stripe_index].dev;
3994 if (rw & REQ_WRITE) {
3995 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3996 BTRFS_BLOCK_GROUP_RAID10 |
3997 BTRFS_BLOCK_GROUP_DUP)) {
4003 bbio->num_stripes = num_stripes;
4004 bbio->max_errors = max_errors;
4005 bbio->mirror_num = mirror_num;
4007 free_extent_map(em);
4011 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
4012 u64 logical, u64 *length,
4013 struct btrfs_bio **bbio_ret, int mirror_num)
4015 return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret,
4019 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4020 u64 chunk_start, u64 physical, u64 devid,
4021 u64 **logical, int *naddrs, int *stripe_len)
4023 struct extent_map_tree *em_tree = &map_tree->map_tree;
4024 struct extent_map *em;
4025 struct map_lookup *map;
4032 read_lock(&em_tree->lock);
4033 em = lookup_extent_mapping(em_tree, chunk_start, 1);
4034 read_unlock(&em_tree->lock);
4036 BUG_ON(!em || em->start != chunk_start);
4037 map = (struct map_lookup *)em->bdev;
4040 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4041 do_div(length, map->num_stripes / map->sub_stripes);
4042 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4043 do_div(length, map->num_stripes);
4045 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4046 BUG_ON(!buf); /* -ENOMEM */
4048 for (i = 0; i < map->num_stripes; i++) {
4049 if (devid && map->stripes[i].dev->devid != devid)
4051 if (map->stripes[i].physical > physical ||
4052 map->stripes[i].physical + length <= physical)
4055 stripe_nr = physical - map->stripes[i].physical;
4056 do_div(stripe_nr, map->stripe_len);
4058 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4059 stripe_nr = stripe_nr * map->num_stripes + i;
4060 do_div(stripe_nr, map->sub_stripes);
4061 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4062 stripe_nr = stripe_nr * map->num_stripes + i;
4064 bytenr = chunk_start + stripe_nr * map->stripe_len;
4065 WARN_ON(nr >= map->num_stripes);
4066 for (j = 0; j < nr; j++) {
4067 if (buf[j] == bytenr)
4071 WARN_ON(nr >= map->num_stripes);
4078 *stripe_len = map->stripe_len;
4080 free_extent_map(em);
4084 static void *merge_stripe_index_into_bio_private(void *bi_private,
4085 unsigned int stripe_index)
4088 * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4090 * The alternative solution (instead of stealing bits from the
4091 * pointer) would be to allocate an intermediate structure
4092 * that contains the old private pointer plus the stripe_index.
4094 BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4095 BUG_ON(stripe_index > 3);
4096 return (void *)(((uintptr_t)bi_private) | stripe_index);
4099 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4101 return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4104 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4106 return (unsigned int)((uintptr_t)bi_private) & 3;
4109 static void btrfs_end_bio(struct bio *bio, int err)
4111 struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4112 int is_orig_bio = 0;
4115 atomic_inc(&bbio->error);
4116 if (err == -EIO || err == -EREMOTEIO) {
4117 unsigned int stripe_index =
4118 extract_stripe_index_from_bio_private(
4120 struct btrfs_device *dev;
4122 BUG_ON(stripe_index >= bbio->num_stripes);
4123 dev = bbio->stripes[stripe_index].dev;
4125 if (bio->bi_rw & WRITE)
4126 btrfs_dev_stat_inc(dev,
4127 BTRFS_DEV_STAT_WRITE_ERRS);
4129 btrfs_dev_stat_inc(dev,
4130 BTRFS_DEV_STAT_READ_ERRS);
4131 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
4132 btrfs_dev_stat_inc(dev,
4133 BTRFS_DEV_STAT_FLUSH_ERRS);
4134 btrfs_dev_stat_print_on_error(dev);
4139 if (bio == bbio->orig_bio)
4142 if (atomic_dec_and_test(&bbio->stripes_pending)) {
4145 bio = bbio->orig_bio;
4147 bio->bi_private = bbio->private;
4148 bio->bi_end_io = bbio->end_io;
4149 bio->bi_bdev = (struct block_device *)
4150 (unsigned long)bbio->mirror_num;
4151 /* only send an error to the higher layers if it is
4152 * beyond the tolerance of the multi-bio
4154 if (atomic_read(&bbio->error) > bbio->max_errors) {
4158 * this bio is actually up to date, we didn't
4159 * go over the max number of errors
4161 set_bit(BIO_UPTODATE, &bio->bi_flags);
4166 bio_endio(bio, err);
4167 } else if (!is_orig_bio) {
4172 struct async_sched {
4175 struct btrfs_fs_info *info;
4176 struct btrfs_work work;
4180 * see run_scheduled_bios for a description of why bios are collected for
4183 * This will add one bio to the pending list for a device and make sure
4184 * the work struct is scheduled.
4186 static noinline void schedule_bio(struct btrfs_root *root,
4187 struct btrfs_device *device,
4188 int rw, struct bio *bio)
4190 int should_queue = 1;
4191 struct btrfs_pending_bios *pending_bios;
4193 /* don't bother with additional async steps for reads, right now */
4194 if (!(rw & REQ_WRITE)) {
4196 btrfsic_submit_bio(rw, bio);
4202 * nr_async_bios allows us to reliably return congestion to the
4203 * higher layers. Otherwise, the async bio makes it appear we have
4204 * made progress against dirty pages when we've really just put it
4205 * on a queue for later
4207 atomic_inc(&root->fs_info->nr_async_bios);
4208 WARN_ON(bio->bi_next);
4209 bio->bi_next = NULL;
4212 spin_lock(&device->io_lock);
4213 if (bio->bi_rw & REQ_SYNC)
4214 pending_bios = &device->pending_sync_bios;
4216 pending_bios = &device->pending_bios;
4218 if (pending_bios->tail)
4219 pending_bios->tail->bi_next = bio;
4221 pending_bios->tail = bio;
4222 if (!pending_bios->head)
4223 pending_bios->head = bio;
4224 if (device->running_pending)
4227 spin_unlock(&device->io_lock);
4230 btrfs_queue_worker(&root->fs_info->submit_workers,
4234 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
4235 int mirror_num, int async_submit)
4237 struct btrfs_mapping_tree *map_tree;
4238 struct btrfs_device *dev;
4239 struct bio *first_bio = bio;
4240 u64 logical = (u64)bio->bi_sector << 9;
4246 struct btrfs_bio *bbio = NULL;
4248 length = bio->bi_size;
4249 map_tree = &root->fs_info->mapping_tree;
4250 map_length = length;
4252 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio,
4254 if (ret) /* -ENOMEM */
4257 total_devs = bbio->num_stripes;
4258 if (map_length < length) {
4259 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
4260 "len %llu\n", (unsigned long long)logical,
4261 (unsigned long long)length,
4262 (unsigned long long)map_length);
4266 bbio->orig_bio = first_bio;
4267 bbio->private = first_bio->bi_private;
4268 bbio->end_io = first_bio->bi_end_io;
4269 atomic_set(&bbio->stripes_pending, bbio->num_stripes);
4271 while (dev_nr < total_devs) {
4272 if (dev_nr < total_devs - 1) {
4273 bio = bio_clone(first_bio, GFP_NOFS);
4274 BUG_ON(!bio); /* -ENOMEM */
4278 bio->bi_private = bbio;
4279 bio->bi_private = merge_stripe_index_into_bio_private(
4280 bio->bi_private, (unsigned int)dev_nr);
4281 bio->bi_end_io = btrfs_end_bio;
4282 bio->bi_sector = bbio->stripes[dev_nr].physical >> 9;
4283 dev = bbio->stripes[dev_nr].dev;
4284 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
4286 struct rcu_string *name;
4289 name = rcu_dereference(dev->name);
4290 pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu "
4291 "(%s id %llu), size=%u\n", rw,
4292 (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
4293 name->str, dev->devid, bio->bi_size);
4296 bio->bi_bdev = dev->bdev;
4298 schedule_bio(root, dev, rw, bio);
4300 btrfsic_submit_bio(rw, bio);
4302 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
4303 bio->bi_sector = logical >> 9;
4304 bio_endio(bio, -EIO);
4311 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
4314 struct btrfs_device *device;
4315 struct btrfs_fs_devices *cur_devices;
4317 cur_devices = root->fs_info->fs_devices;
4318 while (cur_devices) {
4320 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4321 device = __find_device(&cur_devices->devices,
4326 cur_devices = cur_devices->seed;
4331 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
4332 u64 devid, u8 *dev_uuid)
4334 struct btrfs_device *device;
4335 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4337 device = kzalloc(sizeof(*device), GFP_NOFS);
4340 list_add(&device->dev_list,
4341 &fs_devices->devices);
4342 device->dev_root = root->fs_info->dev_root;
4343 device->devid = devid;
4344 device->work.func = pending_bios_fn;
4345 device->fs_devices = fs_devices;
4346 device->missing = 1;
4347 fs_devices->num_devices++;
4348 fs_devices->missing_devices++;
4349 spin_lock_init(&device->io_lock);
4350 INIT_LIST_HEAD(&device->dev_alloc_list);
4351 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
4355 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
4356 struct extent_buffer *leaf,
4357 struct btrfs_chunk *chunk)
4359 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4360 struct map_lookup *map;
4361 struct extent_map *em;
4365 u8 uuid[BTRFS_UUID_SIZE];
4370 logical = key->offset;
4371 length = btrfs_chunk_length(leaf, chunk);
4373 read_lock(&map_tree->map_tree.lock);
4374 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
4375 read_unlock(&map_tree->map_tree.lock);
4377 /* already mapped? */
4378 if (em && em->start <= logical && em->start + em->len > logical) {
4379 free_extent_map(em);
4382 free_extent_map(em);
4385 em = alloc_extent_map();
4388 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
4389 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
4391 free_extent_map(em);
4395 em->bdev = (struct block_device *)map;
4396 em->start = logical;
4398 em->block_start = 0;
4399 em->block_len = em->len;
4401 map->num_stripes = num_stripes;
4402 map->io_width = btrfs_chunk_io_width(leaf, chunk);
4403 map->io_align = btrfs_chunk_io_align(leaf, chunk);
4404 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
4405 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
4406 map->type = btrfs_chunk_type(leaf, chunk);
4407 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
4408 for (i = 0; i < num_stripes; i++) {
4409 map->stripes[i].physical =
4410 btrfs_stripe_offset_nr(leaf, chunk, i);
4411 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
4412 read_extent_buffer(leaf, uuid, (unsigned long)
4413 btrfs_stripe_dev_uuid_nr(chunk, i),
4415 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
4417 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
4419 free_extent_map(em);
4422 if (!map->stripes[i].dev) {
4423 map->stripes[i].dev =
4424 add_missing_dev(root, devid, uuid);
4425 if (!map->stripes[i].dev) {
4427 free_extent_map(em);
4431 map->stripes[i].dev->in_fs_metadata = 1;
4434 write_lock(&map_tree->map_tree.lock);
4435 ret = add_extent_mapping(&map_tree->map_tree, em);
4436 write_unlock(&map_tree->map_tree.lock);
4437 BUG_ON(ret); /* Tree corruption */
4438 free_extent_map(em);
4443 static void fill_device_from_item(struct extent_buffer *leaf,
4444 struct btrfs_dev_item *dev_item,
4445 struct btrfs_device *device)
4449 device->devid = btrfs_device_id(leaf, dev_item);
4450 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
4451 device->total_bytes = device->disk_total_bytes;
4452 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
4453 device->type = btrfs_device_type(leaf, dev_item);
4454 device->io_align = btrfs_device_io_align(leaf, dev_item);
4455 device->io_width = btrfs_device_io_width(leaf, dev_item);
4456 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
4458 ptr = (unsigned long)btrfs_device_uuid(dev_item);
4459 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
4462 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
4464 struct btrfs_fs_devices *fs_devices;
4467 BUG_ON(!mutex_is_locked(&uuid_mutex));
4469 fs_devices = root->fs_info->fs_devices->seed;
4470 while (fs_devices) {
4471 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
4475 fs_devices = fs_devices->seed;
4478 fs_devices = find_fsid(fsid);
4484 fs_devices = clone_fs_devices(fs_devices);
4485 if (IS_ERR(fs_devices)) {
4486 ret = PTR_ERR(fs_devices);
4490 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
4491 root->fs_info->bdev_holder);
4493 free_fs_devices(fs_devices);
4497 if (!fs_devices->seeding) {
4498 __btrfs_close_devices(fs_devices);
4499 free_fs_devices(fs_devices);
4504 fs_devices->seed = root->fs_info->fs_devices->seed;
4505 root->fs_info->fs_devices->seed = fs_devices;
4510 static int read_one_dev(struct btrfs_root *root,
4511 struct extent_buffer *leaf,
4512 struct btrfs_dev_item *dev_item)
4514 struct btrfs_device *device;
4517 u8 fs_uuid[BTRFS_UUID_SIZE];
4518 u8 dev_uuid[BTRFS_UUID_SIZE];
4520 devid = btrfs_device_id(leaf, dev_item);
4521 read_extent_buffer(leaf, dev_uuid,
4522 (unsigned long)btrfs_device_uuid(dev_item),
4524 read_extent_buffer(leaf, fs_uuid,
4525 (unsigned long)btrfs_device_fsid(dev_item),
4528 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
4529 ret = open_seed_devices(root, fs_uuid);
4530 if (ret && !btrfs_test_opt(root, DEGRADED))
4534 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
4535 if (!device || !device->bdev) {
4536 if (!btrfs_test_opt(root, DEGRADED))
4540 printk(KERN_WARNING "warning devid %llu missing\n",
4541 (unsigned long long)devid);
4542 device = add_missing_dev(root, devid, dev_uuid);
4545 } else if (!device->missing) {
4547 * this happens when a device that was properly setup
4548 * in the device info lists suddenly goes bad.
4549 * device->bdev is NULL, and so we have to set
4550 * device->missing to one here
4552 root->fs_info->fs_devices->missing_devices++;
4553 device->missing = 1;
4557 if (device->fs_devices != root->fs_info->fs_devices) {
4558 BUG_ON(device->writeable);
4559 if (device->generation !=
4560 btrfs_device_generation(leaf, dev_item))
4564 fill_device_from_item(leaf, dev_item, device);
4565 device->dev_root = root->fs_info->dev_root;
4566 device->in_fs_metadata = 1;
4567 if (device->writeable) {
4568 device->fs_devices->total_rw_bytes += device->total_bytes;
4569 spin_lock(&root->fs_info->free_chunk_lock);
4570 root->fs_info->free_chunk_space += device->total_bytes -
4572 spin_unlock(&root->fs_info->free_chunk_lock);
4578 int btrfs_read_sys_array(struct btrfs_root *root)
4580 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
4581 struct extent_buffer *sb;
4582 struct btrfs_disk_key *disk_key;
4583 struct btrfs_chunk *chunk;
4585 unsigned long sb_ptr;
4591 struct btrfs_key key;
4593 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
4594 BTRFS_SUPER_INFO_SIZE);
4597 btrfs_set_buffer_uptodate(sb);
4598 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
4600 * The sb extent buffer is artifical and just used to read the system array.
4601 * btrfs_set_buffer_uptodate() call does not properly mark all it's
4602 * pages up-to-date when the page is larger: extent does not cover the
4603 * whole page and consequently check_page_uptodate does not find all
4604 * the page's extents up-to-date (the hole beyond sb),
4605 * write_extent_buffer then triggers a WARN_ON.
4607 * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
4608 * but sb spans only this function. Add an explicit SetPageUptodate call
4609 * to silence the warning eg. on PowerPC 64.
4611 if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
4612 SetPageUptodate(sb->pages[0]);
4614 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
4615 array_size = btrfs_super_sys_array_size(super_copy);
4617 ptr = super_copy->sys_chunk_array;
4618 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
4621 while (cur < array_size) {
4622 disk_key = (struct btrfs_disk_key *)ptr;
4623 btrfs_disk_key_to_cpu(&key, disk_key);
4625 len = sizeof(*disk_key); ptr += len;
4629 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
4630 chunk = (struct btrfs_chunk *)sb_ptr;
4631 ret = read_one_chunk(root, &key, sb, chunk);
4634 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
4635 len = btrfs_chunk_item_size(num_stripes);
4644 free_extent_buffer(sb);
4648 int btrfs_read_chunk_tree(struct btrfs_root *root)
4650 struct btrfs_path *path;
4651 struct extent_buffer *leaf;
4652 struct btrfs_key key;
4653 struct btrfs_key found_key;
4657 root = root->fs_info->chunk_root;
4659 path = btrfs_alloc_path();
4663 mutex_lock(&uuid_mutex);
4666 /* first we search for all of the device items, and then we
4667 * read in all of the chunk items. This way we can create chunk
4668 * mappings that reference all of the devices that are afound
4670 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
4674 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4678 leaf = path->nodes[0];
4679 slot = path->slots[0];
4680 if (slot >= btrfs_header_nritems(leaf)) {
4681 ret = btrfs_next_leaf(root, path);
4688 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4689 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4690 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
4692 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
4693 struct btrfs_dev_item *dev_item;
4694 dev_item = btrfs_item_ptr(leaf, slot,
4695 struct btrfs_dev_item);
4696 ret = read_one_dev(root, leaf, dev_item);
4700 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
4701 struct btrfs_chunk *chunk;
4702 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
4703 ret = read_one_chunk(root, &found_key, leaf, chunk);
4709 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
4711 btrfs_release_path(path);
4716 unlock_chunks(root);
4717 mutex_unlock(&uuid_mutex);
4719 btrfs_free_path(path);
4723 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
4727 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4728 btrfs_dev_stat_reset(dev, i);
4731 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
4733 struct btrfs_key key;
4734 struct btrfs_key found_key;
4735 struct btrfs_root *dev_root = fs_info->dev_root;
4736 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4737 struct extent_buffer *eb;
4740 struct btrfs_device *device;
4741 struct btrfs_path *path = NULL;
4744 path = btrfs_alloc_path();
4750 mutex_lock(&fs_devices->device_list_mutex);
4751 list_for_each_entry(device, &fs_devices->devices, dev_list) {
4753 struct btrfs_dev_stats_item *ptr;
4756 key.type = BTRFS_DEV_STATS_KEY;
4757 key.offset = device->devid;
4758 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
4760 __btrfs_reset_dev_stats(device);
4761 device->dev_stats_valid = 1;
4762 btrfs_release_path(path);
4765 slot = path->slots[0];
4766 eb = path->nodes[0];
4767 btrfs_item_key_to_cpu(eb, &found_key, slot);
4768 item_size = btrfs_item_size_nr(eb, slot);
4770 ptr = btrfs_item_ptr(eb, slot,
4771 struct btrfs_dev_stats_item);
4773 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4774 if (item_size >= (1 + i) * sizeof(__le64))
4775 btrfs_dev_stat_set(device, i,
4776 btrfs_dev_stats_value(eb, ptr, i));
4778 btrfs_dev_stat_reset(device, i);
4781 device->dev_stats_valid = 1;
4782 btrfs_dev_stat_print_on_load(device);
4783 btrfs_release_path(path);
4785 mutex_unlock(&fs_devices->device_list_mutex);
4788 btrfs_free_path(path);
4789 return ret < 0 ? ret : 0;
4792 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
4793 struct btrfs_root *dev_root,
4794 struct btrfs_device *device)
4796 struct btrfs_path *path;
4797 struct btrfs_key key;
4798 struct extent_buffer *eb;
4799 struct btrfs_dev_stats_item *ptr;
4804 key.type = BTRFS_DEV_STATS_KEY;
4805 key.offset = device->devid;
4807 path = btrfs_alloc_path();
4809 ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
4811 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
4812 ret, rcu_str_deref(device->name));
4817 btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
4818 /* need to delete old one and insert a new one */
4819 ret = btrfs_del_item(trans, dev_root, path);
4821 printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
4822 rcu_str_deref(device->name), ret);
4829 /* need to insert a new item */
4830 btrfs_release_path(path);
4831 ret = btrfs_insert_empty_item(trans, dev_root, path,
4832 &key, sizeof(*ptr));
4834 printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
4835 rcu_str_deref(device->name), ret);
4840 eb = path->nodes[0];
4841 ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
4842 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4843 btrfs_set_dev_stats_value(eb, ptr, i,
4844 btrfs_dev_stat_read(device, i));
4845 btrfs_mark_buffer_dirty(eb);
4848 btrfs_free_path(path);
4853 * called from commit_transaction. Writes all changed device stats to disk.
4855 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
4856 struct btrfs_fs_info *fs_info)
4858 struct btrfs_root *dev_root = fs_info->dev_root;
4859 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
4860 struct btrfs_device *device;
4863 mutex_lock(&fs_devices->device_list_mutex);
4864 list_for_each_entry(device, &fs_devices->devices, dev_list) {
4865 if (!device->dev_stats_valid || !device->dev_stats_dirty)
4868 ret = update_dev_stat_item(trans, dev_root, device);
4870 device->dev_stats_dirty = 0;
4872 mutex_unlock(&fs_devices->device_list_mutex);
4877 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
4879 btrfs_dev_stat_inc(dev, index);
4880 btrfs_dev_stat_print_on_error(dev);
4883 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
4885 if (!dev->dev_stats_valid)
4887 printk_ratelimited_in_rcu(KERN_ERR
4888 "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4889 rcu_str_deref(dev->name),
4890 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4891 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4892 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4893 btrfs_dev_stat_read(dev,
4894 BTRFS_DEV_STAT_CORRUPTION_ERRS),
4895 btrfs_dev_stat_read(dev,
4896 BTRFS_DEV_STAT_GENERATION_ERRS));
4899 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
4903 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4904 if (btrfs_dev_stat_read(dev, i) != 0)
4906 if (i == BTRFS_DEV_STAT_VALUES_MAX)
4907 return; /* all values == 0, suppress message */
4909 printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
4910 rcu_str_deref(dev->name),
4911 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
4912 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
4913 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
4914 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
4915 btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
4918 int btrfs_get_dev_stats(struct btrfs_root *root,
4919 struct btrfs_ioctl_get_dev_stats *stats)
4921 struct btrfs_device *dev;
4922 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
4925 mutex_lock(&fs_devices->device_list_mutex);
4926 dev = btrfs_find_device(root, stats->devid, NULL, NULL);
4927 mutex_unlock(&fs_devices->device_list_mutex);
4931 "btrfs: get dev_stats failed, device not found\n");
4933 } else if (!dev->dev_stats_valid) {
4935 "btrfs: get dev_stats failed, not yet valid\n");
4937 } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
4938 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
4939 if (stats->nr_items > i)
4941 btrfs_dev_stat_read_and_reset(dev, i);
4943 btrfs_dev_stat_reset(dev, i);
4946 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
4947 if (stats->nr_items > i)
4948 stats->values[i] = btrfs_dev_stat_read(dev, i);
4950 if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
4951 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;