2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <linux/capability.h>
26 #include <asm/div64.h>
29 #include "extent_map.h"
31 #include "transaction.h"
32 #include "print-tree.h"
34 #include "async-thread.h"
44 struct btrfs_bio_stripe stripes[];
47 static int init_first_rw_device(struct btrfs_trans_handle *trans,
48 struct btrfs_root *root,
49 struct btrfs_device *device);
50 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
52 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
53 (sizeof(struct btrfs_bio_stripe) * (n)))
55 static DEFINE_MUTEX(uuid_mutex);
56 static LIST_HEAD(fs_uuids);
58 void btrfs_lock_volumes(void)
60 mutex_lock(&uuid_mutex);
63 void btrfs_unlock_volumes(void)
65 mutex_unlock(&uuid_mutex);
68 static void lock_chunks(struct btrfs_root *root)
70 mutex_lock(&root->fs_info->chunk_mutex);
73 static void unlock_chunks(struct btrfs_root *root)
75 mutex_unlock(&root->fs_info->chunk_mutex);
78 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
80 struct btrfs_device *device;
81 WARN_ON(fs_devices->opened);
82 while (!list_empty(&fs_devices->devices)) {
83 device = list_entry(fs_devices->devices.next,
84 struct btrfs_device, dev_list);
85 list_del(&device->dev_list);
92 int btrfs_cleanup_fs_uuids(void)
94 struct btrfs_fs_devices *fs_devices;
96 while (!list_empty(&fs_uuids)) {
97 fs_devices = list_entry(fs_uuids.next,
98 struct btrfs_fs_devices, list);
99 list_del(&fs_devices->list);
100 free_fs_devices(fs_devices);
105 static noinline struct btrfs_device *__find_device(struct list_head *head,
108 struct btrfs_device *dev;
110 list_for_each_entry(dev, head, dev_list) {
111 if (dev->devid == devid &&
112 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
119 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
121 struct btrfs_fs_devices *fs_devices;
123 list_for_each_entry(fs_devices, &fs_uuids, list) {
124 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
130 static void requeue_list(struct btrfs_pending_bios *pending_bios,
131 struct bio *head, struct bio *tail)
134 struct bio *old_head;
136 old_head = pending_bios->head;
137 pending_bios->head = head;
138 if (pending_bios->tail)
139 tail->bi_next = old_head;
141 pending_bios->tail = tail;
145 * we try to collect pending bios for a device so we don't get a large
146 * number of procs sending bios down to the same device. This greatly
147 * improves the schedulers ability to collect and merge the bios.
149 * But, it also turns into a long list of bios to process and that is sure
150 * to eventually make the worker thread block. The solution here is to
151 * make some progress and then put this work struct back at the end of
152 * the list if the block device is congested. This way, multiple devices
153 * can make progress from a single worker thread.
155 static noinline int run_scheduled_bios(struct btrfs_device *device)
158 struct backing_dev_info *bdi;
159 struct btrfs_fs_info *fs_info;
160 struct btrfs_pending_bios *pending_bios;
164 unsigned long num_run;
165 unsigned long num_sync_run;
166 unsigned long batch_run = 0;
168 unsigned long last_waited = 0;
171 bdi = blk_get_backing_dev_info(device->bdev);
172 fs_info = device->dev_root->fs_info;
173 limit = btrfs_async_submit_limit(fs_info);
174 limit = limit * 2 / 3;
176 /* we want to make sure that every time we switch from the sync
177 * list to the normal list, we unplug
182 spin_lock(&device->io_lock);
187 /* take all the bios off the list at once and process them
188 * later on (without the lock held). But, remember the
189 * tail and other pointers so the bios can be properly reinserted
190 * into the list if we hit congestion
192 if (!force_reg && device->pending_sync_bios.head) {
193 pending_bios = &device->pending_sync_bios;
196 pending_bios = &device->pending_bios;
200 pending = pending_bios->head;
201 tail = pending_bios->tail;
202 WARN_ON(pending && !tail);
205 * if pending was null this time around, no bios need processing
206 * at all and we can stop. Otherwise it'll loop back up again
207 * and do an additional check so no bios are missed.
209 * device->running_pending is used to synchronize with the
212 if (device->pending_sync_bios.head == NULL &&
213 device->pending_bios.head == NULL) {
215 device->running_pending = 0;
218 device->running_pending = 1;
221 pending_bios->head = NULL;
222 pending_bios->tail = NULL;
224 spin_unlock(&device->io_lock);
227 * if we're doing the regular priority list, make sure we unplug
228 * for any high prio bios we've sent down
230 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
232 blk_run_backing_dev(bdi, NULL);
238 /* we want to work on both lists, but do more bios on the
239 * sync list than the regular list
242 pending_bios != &device->pending_sync_bios &&
243 device->pending_sync_bios.head) ||
244 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
245 device->pending_bios.head)) {
246 spin_lock(&device->io_lock);
247 requeue_list(pending_bios, pending, tail);
252 pending = pending->bi_next;
254 atomic_dec(&fs_info->nr_async_bios);
256 if (atomic_read(&fs_info->nr_async_bios) < limit &&
257 waitqueue_active(&fs_info->async_submit_wait))
258 wake_up(&fs_info->async_submit_wait);
260 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
262 if (cur->bi_rw & REQ_SYNC)
265 submit_bio(cur->bi_rw, cur);
268 if (need_resched()) {
270 blk_run_backing_dev(bdi, NULL);
277 * we made progress, there is more work to do and the bdi
278 * is now congested. Back off and let other work structs
281 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
282 fs_info->fs_devices->open_devices > 1) {
283 struct io_context *ioc;
285 ioc = current->io_context;
288 * the main goal here is that we don't want to
289 * block if we're going to be able to submit
290 * more requests without blocking.
292 * This code does two great things, it pokes into
293 * the elevator code from a filesystem _and_
294 * it makes assumptions about how batching works.
296 if (ioc && ioc->nr_batch_requests > 0 &&
297 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
299 ioc->last_waited == last_waited)) {
301 * we want to go through our batch of
302 * requests and stop. So, we copy out
303 * the ioc->last_waited time and test
304 * against it before looping
306 last_waited = ioc->last_waited;
307 if (need_resched()) {
309 blk_run_backing_dev(bdi, NULL);
316 spin_lock(&device->io_lock);
317 requeue_list(pending_bios, pending, tail);
318 device->running_pending = 1;
320 spin_unlock(&device->io_lock);
321 btrfs_requeue_work(&device->work);
328 blk_run_backing_dev(bdi, NULL);
331 * IO has already been through a long path to get here. Checksumming,
332 * async helper threads, perhaps compression. We've done a pretty
333 * good job of collecting a batch of IO and should just unplug
334 * the device right away.
336 * This will help anyone who is waiting on the IO, they might have
337 * already unplugged, but managed to do so before the bio they
338 * cared about found its way down here.
340 blk_run_backing_dev(bdi, NULL);
346 spin_lock(&device->io_lock);
347 if (device->pending_bios.head || device->pending_sync_bios.head)
349 spin_unlock(&device->io_lock);
355 static void pending_bios_fn(struct btrfs_work *work)
357 struct btrfs_device *device;
359 device = container_of(work, struct btrfs_device, work);
360 run_scheduled_bios(device);
363 static noinline int device_list_add(const char *path,
364 struct btrfs_super_block *disk_super,
365 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
367 struct btrfs_device *device;
368 struct btrfs_fs_devices *fs_devices;
369 u64 found_transid = btrfs_super_generation(disk_super);
372 fs_devices = find_fsid(disk_super->fsid);
374 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
377 INIT_LIST_HEAD(&fs_devices->devices);
378 INIT_LIST_HEAD(&fs_devices->alloc_list);
379 list_add(&fs_devices->list, &fs_uuids);
380 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
381 fs_devices->latest_devid = devid;
382 fs_devices->latest_trans = found_transid;
383 mutex_init(&fs_devices->device_list_mutex);
386 device = __find_device(&fs_devices->devices, devid,
387 disk_super->dev_item.uuid);
390 if (fs_devices->opened)
393 device = kzalloc(sizeof(*device), GFP_NOFS);
395 /* we can safely leave the fs_devices entry around */
398 device->devid = devid;
399 device->work.func = pending_bios_fn;
400 memcpy(device->uuid, disk_super->dev_item.uuid,
402 device->barriers = 1;
403 spin_lock_init(&device->io_lock);
404 device->name = kstrdup(path, GFP_NOFS);
409 INIT_LIST_HEAD(&device->dev_alloc_list);
411 mutex_lock(&fs_devices->device_list_mutex);
412 list_add(&device->dev_list, &fs_devices->devices);
413 mutex_unlock(&fs_devices->device_list_mutex);
415 device->fs_devices = fs_devices;
416 fs_devices->num_devices++;
417 } else if (!device->name || strcmp(device->name, path)) {
418 name = kstrdup(path, GFP_NOFS);
423 if (device->missing) {
424 fs_devices->missing_devices--;
429 if (found_transid > fs_devices->latest_trans) {
430 fs_devices->latest_devid = devid;
431 fs_devices->latest_trans = found_transid;
433 *fs_devices_ret = fs_devices;
437 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
439 struct btrfs_fs_devices *fs_devices;
440 struct btrfs_device *device;
441 struct btrfs_device *orig_dev;
443 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
445 return ERR_PTR(-ENOMEM);
447 INIT_LIST_HEAD(&fs_devices->devices);
448 INIT_LIST_HEAD(&fs_devices->alloc_list);
449 INIT_LIST_HEAD(&fs_devices->list);
450 mutex_init(&fs_devices->device_list_mutex);
451 fs_devices->latest_devid = orig->latest_devid;
452 fs_devices->latest_trans = orig->latest_trans;
453 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
455 mutex_lock(&orig->device_list_mutex);
456 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
457 device = kzalloc(sizeof(*device), GFP_NOFS);
461 device->name = kstrdup(orig_dev->name, GFP_NOFS);
467 device->devid = orig_dev->devid;
468 device->work.func = pending_bios_fn;
469 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
470 device->barriers = 1;
471 spin_lock_init(&device->io_lock);
472 INIT_LIST_HEAD(&device->dev_list);
473 INIT_LIST_HEAD(&device->dev_alloc_list);
475 list_add(&device->dev_list, &fs_devices->devices);
476 device->fs_devices = fs_devices;
477 fs_devices->num_devices++;
479 mutex_unlock(&orig->device_list_mutex);
482 mutex_unlock(&orig->device_list_mutex);
483 free_fs_devices(fs_devices);
484 return ERR_PTR(-ENOMEM);
487 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
489 struct btrfs_device *device, *next;
491 mutex_lock(&uuid_mutex);
493 mutex_lock(&fs_devices->device_list_mutex);
494 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
495 if (device->in_fs_metadata)
499 close_bdev_exclusive(device->bdev, device->mode);
501 fs_devices->open_devices--;
503 if (device->writeable) {
504 list_del_init(&device->dev_alloc_list);
505 device->writeable = 0;
506 fs_devices->rw_devices--;
508 list_del_init(&device->dev_list);
509 fs_devices->num_devices--;
513 mutex_unlock(&fs_devices->device_list_mutex);
515 if (fs_devices->seed) {
516 fs_devices = fs_devices->seed;
520 mutex_unlock(&uuid_mutex);
524 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
526 struct btrfs_device *device;
528 if (--fs_devices->opened > 0)
531 list_for_each_entry(device, &fs_devices->devices, dev_list) {
533 close_bdev_exclusive(device->bdev, device->mode);
534 fs_devices->open_devices--;
536 if (device->writeable) {
537 list_del_init(&device->dev_alloc_list);
538 fs_devices->rw_devices--;
542 device->writeable = 0;
543 device->in_fs_metadata = 0;
545 WARN_ON(fs_devices->open_devices);
546 WARN_ON(fs_devices->rw_devices);
547 fs_devices->opened = 0;
548 fs_devices->seeding = 0;
553 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
555 struct btrfs_fs_devices *seed_devices = NULL;
558 mutex_lock(&uuid_mutex);
559 ret = __btrfs_close_devices(fs_devices);
560 if (!fs_devices->opened) {
561 seed_devices = fs_devices->seed;
562 fs_devices->seed = NULL;
564 mutex_unlock(&uuid_mutex);
566 while (seed_devices) {
567 fs_devices = seed_devices;
568 seed_devices = fs_devices->seed;
569 __btrfs_close_devices(fs_devices);
570 free_fs_devices(fs_devices);
575 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
576 fmode_t flags, void *holder)
578 struct block_device *bdev;
579 struct list_head *head = &fs_devices->devices;
580 struct btrfs_device *device;
581 struct block_device *latest_bdev = NULL;
582 struct buffer_head *bh;
583 struct btrfs_super_block *disk_super;
584 u64 latest_devid = 0;
585 u64 latest_transid = 0;
590 list_for_each_entry(device, head, dev_list) {
596 bdev = open_bdev_exclusive(device->name, flags, holder);
598 printk(KERN_INFO "open %s failed\n", device->name);
601 set_blocksize(bdev, 4096);
603 bh = btrfs_read_dev_super(bdev);
609 disk_super = (struct btrfs_super_block *)bh->b_data;
610 devid = btrfs_stack_device_id(&disk_super->dev_item);
611 if (devid != device->devid)
614 if (memcmp(device->uuid, disk_super->dev_item.uuid,
618 device->generation = btrfs_super_generation(disk_super);
619 if (!latest_transid || device->generation > latest_transid) {
620 latest_devid = devid;
621 latest_transid = device->generation;
625 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
626 device->writeable = 0;
628 device->writeable = !bdev_read_only(bdev);
633 device->in_fs_metadata = 0;
634 device->mode = flags;
636 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
637 fs_devices->rotating = 1;
639 fs_devices->open_devices++;
640 if (device->writeable) {
641 fs_devices->rw_devices++;
642 list_add(&device->dev_alloc_list,
643 &fs_devices->alloc_list);
650 close_bdev_exclusive(bdev, FMODE_READ);
654 if (fs_devices->open_devices == 0) {
658 fs_devices->seeding = seeding;
659 fs_devices->opened = 1;
660 fs_devices->latest_bdev = latest_bdev;
661 fs_devices->latest_devid = latest_devid;
662 fs_devices->latest_trans = latest_transid;
663 fs_devices->total_rw_bytes = 0;
668 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
669 fmode_t flags, void *holder)
673 mutex_lock(&uuid_mutex);
674 if (fs_devices->opened) {
675 fs_devices->opened++;
678 ret = __btrfs_open_devices(fs_devices, flags, holder);
680 mutex_unlock(&uuid_mutex);
684 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
685 struct btrfs_fs_devices **fs_devices_ret)
687 struct btrfs_super_block *disk_super;
688 struct block_device *bdev;
689 struct buffer_head *bh;
694 mutex_lock(&uuid_mutex);
696 bdev = open_bdev_exclusive(path, flags, holder);
703 ret = set_blocksize(bdev, 4096);
706 bh = btrfs_read_dev_super(bdev);
711 disk_super = (struct btrfs_super_block *)bh->b_data;
712 devid = btrfs_stack_device_id(&disk_super->dev_item);
713 transid = btrfs_super_generation(disk_super);
714 if (disk_super->label[0])
715 printk(KERN_INFO "device label %s ", disk_super->label);
717 /* FIXME, make a readl uuid parser */
718 printk(KERN_INFO "device fsid %llx-%llx ",
719 *(unsigned long long *)disk_super->fsid,
720 *(unsigned long long *)(disk_super->fsid + 8));
722 printk(KERN_CONT "devid %llu transid %llu %s\n",
723 (unsigned long long)devid, (unsigned long long)transid, path);
724 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
728 close_bdev_exclusive(bdev, flags);
730 mutex_unlock(&uuid_mutex);
734 /* helper to account the used device space in the range */
735 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
736 u64 end, u64 *length)
738 struct btrfs_key key;
739 struct btrfs_root *root = device->dev_root;
740 struct btrfs_dev_extent *dev_extent;
741 struct btrfs_path *path;
745 struct extent_buffer *l;
749 if (start >= device->total_bytes)
752 path = btrfs_alloc_path();
757 key.objectid = device->devid;
759 key.type = BTRFS_DEV_EXTENT_KEY;
761 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
765 ret = btrfs_previous_item(root, path, key.objectid, key.type);
772 slot = path->slots[0];
773 if (slot >= btrfs_header_nritems(l)) {
774 ret = btrfs_next_leaf(root, path);
782 btrfs_item_key_to_cpu(l, &key, slot);
784 if (key.objectid < device->devid)
787 if (key.objectid > device->devid)
790 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
793 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
794 extent_end = key.offset + btrfs_dev_extent_length(l,
796 if (key.offset <= start && extent_end > end) {
797 *length = end - start + 1;
799 } else if (key.offset <= start && extent_end > start)
800 *length += extent_end - start;
801 else if (key.offset > start && extent_end <= end)
802 *length += extent_end - key.offset;
803 else if (key.offset > start && key.offset <= end) {
804 *length += end - key.offset + 1;
806 } else if (key.offset > end)
814 btrfs_free_path(path);
819 * find_free_dev_extent - find free space in the specified device
820 * @trans: transaction handler
821 * @device: the device which we search the free space in
822 * @num_bytes: the size of the free space that we need
823 * @start: store the start of the free space.
824 * @len: the size of the free space. that we find, or the size of the max
825 * free space if we don't find suitable free space
827 * this uses a pretty simple search, the expectation is that it is
828 * called very infrequently and that a given device has a small number
831 * @start is used to store the start of the free space if we find. But if we
832 * don't find suitable free space, it will be used to store the start position
833 * of the max free space.
835 * @len is used to store the size of the free space that we find.
836 * But if we don't find suitable free space, it is used to store the size of
837 * the max free space.
839 int find_free_dev_extent(struct btrfs_trans_handle *trans,
840 struct btrfs_device *device, u64 num_bytes,
841 u64 *start, u64 *len)
843 struct btrfs_key key;
844 struct btrfs_root *root = device->dev_root;
845 struct btrfs_dev_extent *dev_extent;
846 struct btrfs_path *path;
852 u64 search_end = device->total_bytes;
855 struct extent_buffer *l;
857 /* FIXME use last free of some kind */
859 /* we don't want to overwrite the superblock on the drive,
860 * so we make sure to start at an offset of at least 1MB
862 search_start = 1024 * 1024;
864 if (root->fs_info->alloc_start + num_bytes <= search_end)
865 search_start = max(root->fs_info->alloc_start, search_start);
867 max_hole_start = search_start;
870 if (search_start >= search_end) {
875 path = btrfs_alloc_path();
882 key.objectid = device->devid;
883 key.offset = search_start;
884 key.type = BTRFS_DEV_EXTENT_KEY;
886 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
890 ret = btrfs_previous_item(root, path, key.objectid, key.type);
897 slot = path->slots[0];
898 if (slot >= btrfs_header_nritems(l)) {
899 ret = btrfs_next_leaf(root, path);
907 btrfs_item_key_to_cpu(l, &key, slot);
909 if (key.objectid < device->devid)
912 if (key.objectid > device->devid)
915 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
918 if (key.offset > search_start) {
919 hole_size = key.offset - search_start;
921 if (hole_size > max_hole_size) {
922 max_hole_start = search_start;
923 max_hole_size = hole_size;
927 * If this free space is greater than which we need,
928 * it must be the max free space that we have found
929 * until now, so max_hole_start must point to the start
930 * of this free space and the length of this free space
931 * is stored in max_hole_size. Thus, we return
932 * max_hole_start and max_hole_size and go back to the
935 if (hole_size >= num_bytes) {
941 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
942 extent_end = key.offset + btrfs_dev_extent_length(l,
944 if (extent_end > search_start)
945 search_start = extent_end;
951 hole_size = search_end- search_start;
952 if (hole_size > max_hole_size) {
953 max_hole_start = search_start;
954 max_hole_size = hole_size;
958 if (hole_size < num_bytes)
964 btrfs_free_path(path);
966 *start = max_hole_start;
968 *len = max_hole_size;
972 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
973 struct btrfs_device *device,
977 struct btrfs_path *path;
978 struct btrfs_root *root = device->dev_root;
979 struct btrfs_key key;
980 struct btrfs_key found_key;
981 struct extent_buffer *leaf = NULL;
982 struct btrfs_dev_extent *extent = NULL;
984 path = btrfs_alloc_path();
988 key.objectid = device->devid;
990 key.type = BTRFS_DEV_EXTENT_KEY;
992 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
994 ret = btrfs_previous_item(root, path, key.objectid,
995 BTRFS_DEV_EXTENT_KEY);
997 leaf = path->nodes[0];
998 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
999 extent = btrfs_item_ptr(leaf, path->slots[0],
1000 struct btrfs_dev_extent);
1001 BUG_ON(found_key.offset > start || found_key.offset +
1002 btrfs_dev_extent_length(leaf, extent) < start);
1004 } else if (ret == 0) {
1005 leaf = path->nodes[0];
1006 extent = btrfs_item_ptr(leaf, path->slots[0],
1007 struct btrfs_dev_extent);
1011 if (device->bytes_used > 0)
1012 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1013 ret = btrfs_del_item(trans, root, path);
1016 btrfs_free_path(path);
1020 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1021 struct btrfs_device *device,
1022 u64 chunk_tree, u64 chunk_objectid,
1023 u64 chunk_offset, u64 start, u64 num_bytes)
1026 struct btrfs_path *path;
1027 struct btrfs_root *root = device->dev_root;
1028 struct btrfs_dev_extent *extent;
1029 struct extent_buffer *leaf;
1030 struct btrfs_key key;
1032 WARN_ON(!device->in_fs_metadata);
1033 path = btrfs_alloc_path();
1037 key.objectid = device->devid;
1039 key.type = BTRFS_DEV_EXTENT_KEY;
1040 ret = btrfs_insert_empty_item(trans, root, path, &key,
1044 leaf = path->nodes[0];
1045 extent = btrfs_item_ptr(leaf, path->slots[0],
1046 struct btrfs_dev_extent);
1047 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1048 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1049 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1051 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1052 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1055 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1056 btrfs_mark_buffer_dirty(leaf);
1057 btrfs_free_path(path);
1061 static noinline int find_next_chunk(struct btrfs_root *root,
1062 u64 objectid, u64 *offset)
1064 struct btrfs_path *path;
1066 struct btrfs_key key;
1067 struct btrfs_chunk *chunk;
1068 struct btrfs_key found_key;
1070 path = btrfs_alloc_path();
1073 key.objectid = objectid;
1074 key.offset = (u64)-1;
1075 key.type = BTRFS_CHUNK_ITEM_KEY;
1077 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1083 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1087 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1089 if (found_key.objectid != objectid)
1092 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1093 struct btrfs_chunk);
1094 *offset = found_key.offset +
1095 btrfs_chunk_length(path->nodes[0], chunk);
1100 btrfs_free_path(path);
1104 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1107 struct btrfs_key key;
1108 struct btrfs_key found_key;
1109 struct btrfs_path *path;
1111 root = root->fs_info->chunk_root;
1113 path = btrfs_alloc_path();
1117 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1118 key.type = BTRFS_DEV_ITEM_KEY;
1119 key.offset = (u64)-1;
1121 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1127 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1128 BTRFS_DEV_ITEM_KEY);
1132 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1134 *objectid = found_key.offset + 1;
1138 btrfs_free_path(path);
1143 * the device information is stored in the chunk root
1144 * the btrfs_device struct should be fully filled in
1146 int btrfs_add_device(struct btrfs_trans_handle *trans,
1147 struct btrfs_root *root,
1148 struct btrfs_device *device)
1151 struct btrfs_path *path;
1152 struct btrfs_dev_item *dev_item;
1153 struct extent_buffer *leaf;
1154 struct btrfs_key key;
1157 root = root->fs_info->chunk_root;
1159 path = btrfs_alloc_path();
1163 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1164 key.type = BTRFS_DEV_ITEM_KEY;
1165 key.offset = device->devid;
1167 ret = btrfs_insert_empty_item(trans, root, path, &key,
1172 leaf = path->nodes[0];
1173 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1175 btrfs_set_device_id(leaf, dev_item, device->devid);
1176 btrfs_set_device_generation(leaf, dev_item, 0);
1177 btrfs_set_device_type(leaf, dev_item, device->type);
1178 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1179 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1180 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1181 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1182 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1183 btrfs_set_device_group(leaf, dev_item, 0);
1184 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1185 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1186 btrfs_set_device_start_offset(leaf, dev_item, 0);
1188 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1189 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1190 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1191 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1192 btrfs_mark_buffer_dirty(leaf);
1196 btrfs_free_path(path);
1200 static int btrfs_rm_dev_item(struct btrfs_root *root,
1201 struct btrfs_device *device)
1204 struct btrfs_path *path;
1205 struct btrfs_key key;
1206 struct btrfs_trans_handle *trans;
1208 root = root->fs_info->chunk_root;
1210 path = btrfs_alloc_path();
1214 trans = btrfs_start_transaction(root, 0);
1215 if (IS_ERR(trans)) {
1216 btrfs_free_path(path);
1217 return PTR_ERR(trans);
1219 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1220 key.type = BTRFS_DEV_ITEM_KEY;
1221 key.offset = device->devid;
1224 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1233 ret = btrfs_del_item(trans, root, path);
1237 btrfs_free_path(path);
1238 unlock_chunks(root);
1239 btrfs_commit_transaction(trans, root);
1243 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1245 struct btrfs_device *device;
1246 struct btrfs_device *next_device;
1247 struct block_device *bdev;
1248 struct buffer_head *bh = NULL;
1249 struct btrfs_super_block *disk_super;
1256 mutex_lock(&uuid_mutex);
1257 mutex_lock(&root->fs_info->volume_mutex);
1259 all_avail = root->fs_info->avail_data_alloc_bits |
1260 root->fs_info->avail_system_alloc_bits |
1261 root->fs_info->avail_metadata_alloc_bits;
1263 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1264 root->fs_info->fs_devices->num_devices <= 4) {
1265 printk(KERN_ERR "btrfs: unable to go below four devices "
1271 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1272 root->fs_info->fs_devices->num_devices <= 2) {
1273 printk(KERN_ERR "btrfs: unable to go below two "
1274 "devices on raid1\n");
1279 if (strcmp(device_path, "missing") == 0) {
1280 struct list_head *devices;
1281 struct btrfs_device *tmp;
1284 devices = &root->fs_info->fs_devices->devices;
1285 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1286 list_for_each_entry(tmp, devices, dev_list) {
1287 if (tmp->in_fs_metadata && !tmp->bdev) {
1292 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1297 printk(KERN_ERR "btrfs: no missing devices found to "
1302 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1303 root->fs_info->bdev_holder);
1305 ret = PTR_ERR(bdev);
1309 set_blocksize(bdev, 4096);
1310 bh = btrfs_read_dev_super(bdev);
1315 disk_super = (struct btrfs_super_block *)bh->b_data;
1316 devid = btrfs_stack_device_id(&disk_super->dev_item);
1317 dev_uuid = disk_super->dev_item.uuid;
1318 device = btrfs_find_device(root, devid, dev_uuid,
1326 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1327 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1333 if (device->writeable) {
1334 list_del_init(&device->dev_alloc_list);
1335 root->fs_info->fs_devices->rw_devices--;
1338 ret = btrfs_shrink_device(device, 0);
1342 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1346 device->in_fs_metadata = 0;
1349 * the device list mutex makes sure that we don't change
1350 * the device list while someone else is writing out all
1351 * the device supers.
1353 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1354 list_del_init(&device->dev_list);
1355 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1357 device->fs_devices->num_devices--;
1359 if (device->missing)
1360 root->fs_info->fs_devices->missing_devices--;
1362 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1363 struct btrfs_device, dev_list);
1364 if (device->bdev == root->fs_info->sb->s_bdev)
1365 root->fs_info->sb->s_bdev = next_device->bdev;
1366 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1367 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1370 close_bdev_exclusive(device->bdev, device->mode);
1371 device->bdev = NULL;
1372 device->fs_devices->open_devices--;
1375 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1376 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1378 if (device->fs_devices->open_devices == 0) {
1379 struct btrfs_fs_devices *fs_devices;
1380 fs_devices = root->fs_info->fs_devices;
1381 while (fs_devices) {
1382 if (fs_devices->seed == device->fs_devices)
1384 fs_devices = fs_devices->seed;
1386 fs_devices->seed = device->fs_devices->seed;
1387 device->fs_devices->seed = NULL;
1388 __btrfs_close_devices(device->fs_devices);
1389 free_fs_devices(device->fs_devices);
1393 * at this point, the device is zero sized. We want to
1394 * remove it from the devices list and zero out the old super
1396 if (device->writeable) {
1397 /* make sure this device isn't detected as part of
1400 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1401 set_buffer_dirty(bh);
1402 sync_dirty_buffer(bh);
1405 kfree(device->name);
1413 close_bdev_exclusive(bdev, FMODE_READ);
1415 mutex_unlock(&root->fs_info->volume_mutex);
1416 mutex_unlock(&uuid_mutex);
1421 * does all the dirty work required for changing file system's UUID.
1423 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *root)
1426 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1427 struct btrfs_fs_devices *old_devices;
1428 struct btrfs_fs_devices *seed_devices;
1429 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1430 struct btrfs_device *device;
1433 BUG_ON(!mutex_is_locked(&uuid_mutex));
1434 if (!fs_devices->seeding)
1437 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1441 old_devices = clone_fs_devices(fs_devices);
1442 if (IS_ERR(old_devices)) {
1443 kfree(seed_devices);
1444 return PTR_ERR(old_devices);
1447 list_add(&old_devices->list, &fs_uuids);
1449 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1450 seed_devices->opened = 1;
1451 INIT_LIST_HEAD(&seed_devices->devices);
1452 INIT_LIST_HEAD(&seed_devices->alloc_list);
1453 mutex_init(&seed_devices->device_list_mutex);
1454 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1455 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1456 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1457 device->fs_devices = seed_devices;
1460 fs_devices->seeding = 0;
1461 fs_devices->num_devices = 0;
1462 fs_devices->open_devices = 0;
1463 fs_devices->seed = seed_devices;
1465 generate_random_uuid(fs_devices->fsid);
1466 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1467 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1468 super_flags = btrfs_super_flags(disk_super) &
1469 ~BTRFS_SUPER_FLAG_SEEDING;
1470 btrfs_set_super_flags(disk_super, super_flags);
1476 * strore the expected generation for seed devices in device items.
1478 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1479 struct btrfs_root *root)
1481 struct btrfs_path *path;
1482 struct extent_buffer *leaf;
1483 struct btrfs_dev_item *dev_item;
1484 struct btrfs_device *device;
1485 struct btrfs_key key;
1486 u8 fs_uuid[BTRFS_UUID_SIZE];
1487 u8 dev_uuid[BTRFS_UUID_SIZE];
1491 path = btrfs_alloc_path();
1495 root = root->fs_info->chunk_root;
1496 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1498 key.type = BTRFS_DEV_ITEM_KEY;
1501 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1505 leaf = path->nodes[0];
1507 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1508 ret = btrfs_next_leaf(root, path);
1513 leaf = path->nodes[0];
1514 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1515 btrfs_release_path(root, path);
1519 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1520 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1521 key.type != BTRFS_DEV_ITEM_KEY)
1524 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1525 struct btrfs_dev_item);
1526 devid = btrfs_device_id(leaf, dev_item);
1527 read_extent_buffer(leaf, dev_uuid,
1528 (unsigned long)btrfs_device_uuid(dev_item),
1530 read_extent_buffer(leaf, fs_uuid,
1531 (unsigned long)btrfs_device_fsid(dev_item),
1533 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1536 if (device->fs_devices->seeding) {
1537 btrfs_set_device_generation(leaf, dev_item,
1538 device->generation);
1539 btrfs_mark_buffer_dirty(leaf);
1547 btrfs_free_path(path);
1551 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1553 struct btrfs_trans_handle *trans;
1554 struct btrfs_device *device;
1555 struct block_device *bdev;
1556 struct list_head *devices;
1557 struct super_block *sb = root->fs_info->sb;
1559 int seeding_dev = 0;
1562 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1565 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1567 return PTR_ERR(bdev);
1569 if (root->fs_info->fs_devices->seeding) {
1571 down_write(&sb->s_umount);
1572 mutex_lock(&uuid_mutex);
1575 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1576 mutex_lock(&root->fs_info->volume_mutex);
1578 devices = &root->fs_info->fs_devices->devices;
1580 * we have the volume lock, so we don't need the extra
1581 * device list mutex while reading the list here.
1583 list_for_each_entry(device, devices, dev_list) {
1584 if (device->bdev == bdev) {
1590 device = kzalloc(sizeof(*device), GFP_NOFS);
1592 /* we can safely leave the fs_devices entry around */
1597 device->name = kstrdup(device_path, GFP_NOFS);
1598 if (!device->name) {
1604 ret = find_next_devid(root, &device->devid);
1606 kfree(device->name);
1611 trans = btrfs_start_transaction(root, 0);
1612 if (IS_ERR(trans)) {
1613 kfree(device->name);
1615 ret = PTR_ERR(trans);
1621 device->barriers = 1;
1622 device->writeable = 1;
1623 device->work.func = pending_bios_fn;
1624 generate_random_uuid(device->uuid);
1625 spin_lock_init(&device->io_lock);
1626 device->generation = trans->transid;
1627 device->io_width = root->sectorsize;
1628 device->io_align = root->sectorsize;
1629 device->sector_size = root->sectorsize;
1630 device->total_bytes = i_size_read(bdev->bd_inode);
1631 device->disk_total_bytes = device->total_bytes;
1632 device->dev_root = root->fs_info->dev_root;
1633 device->bdev = bdev;
1634 device->in_fs_metadata = 1;
1636 set_blocksize(device->bdev, 4096);
1639 sb->s_flags &= ~MS_RDONLY;
1640 ret = btrfs_prepare_sprout(trans, root);
1644 device->fs_devices = root->fs_info->fs_devices;
1647 * we don't want write_supers to jump in here with our device
1650 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1651 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1652 list_add(&device->dev_alloc_list,
1653 &root->fs_info->fs_devices->alloc_list);
1654 root->fs_info->fs_devices->num_devices++;
1655 root->fs_info->fs_devices->open_devices++;
1656 root->fs_info->fs_devices->rw_devices++;
1657 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1659 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1660 root->fs_info->fs_devices->rotating = 1;
1662 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1663 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1664 total_bytes + device->total_bytes);
1666 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1667 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1669 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1672 ret = init_first_rw_device(trans, root, device);
1674 ret = btrfs_finish_sprout(trans, root);
1677 ret = btrfs_add_device(trans, root, device);
1681 * we've got more storage, clear any full flags on the space
1684 btrfs_clear_space_info_full(root->fs_info);
1686 unlock_chunks(root);
1687 btrfs_commit_transaction(trans, root);
1690 mutex_unlock(&uuid_mutex);
1691 up_write(&sb->s_umount);
1693 ret = btrfs_relocate_sys_chunks(root);
1697 mutex_unlock(&root->fs_info->volume_mutex);
1700 close_bdev_exclusive(bdev, 0);
1702 mutex_unlock(&uuid_mutex);
1703 up_write(&sb->s_umount);
1708 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1709 struct btrfs_device *device)
1712 struct btrfs_path *path;
1713 struct btrfs_root *root;
1714 struct btrfs_dev_item *dev_item;
1715 struct extent_buffer *leaf;
1716 struct btrfs_key key;
1718 root = device->dev_root->fs_info->chunk_root;
1720 path = btrfs_alloc_path();
1724 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1725 key.type = BTRFS_DEV_ITEM_KEY;
1726 key.offset = device->devid;
1728 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1737 leaf = path->nodes[0];
1738 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1740 btrfs_set_device_id(leaf, dev_item, device->devid);
1741 btrfs_set_device_type(leaf, dev_item, device->type);
1742 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1743 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1744 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1745 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1746 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1747 btrfs_mark_buffer_dirty(leaf);
1750 btrfs_free_path(path);
1754 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1755 struct btrfs_device *device, u64 new_size)
1757 struct btrfs_super_block *super_copy =
1758 &device->dev_root->fs_info->super_copy;
1759 u64 old_total = btrfs_super_total_bytes(super_copy);
1760 u64 diff = new_size - device->total_bytes;
1762 if (!device->writeable)
1764 if (new_size <= device->total_bytes)
1767 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1768 device->fs_devices->total_rw_bytes += diff;
1770 device->total_bytes = new_size;
1771 device->disk_total_bytes = new_size;
1772 btrfs_clear_space_info_full(device->dev_root->fs_info);
1774 return btrfs_update_device(trans, device);
1777 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1778 struct btrfs_device *device, u64 new_size)
1781 lock_chunks(device->dev_root);
1782 ret = __btrfs_grow_device(trans, device, new_size);
1783 unlock_chunks(device->dev_root);
1787 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1788 struct btrfs_root *root,
1789 u64 chunk_tree, u64 chunk_objectid,
1793 struct btrfs_path *path;
1794 struct btrfs_key key;
1796 root = root->fs_info->chunk_root;
1797 path = btrfs_alloc_path();
1801 key.objectid = chunk_objectid;
1802 key.offset = chunk_offset;
1803 key.type = BTRFS_CHUNK_ITEM_KEY;
1805 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1808 ret = btrfs_del_item(trans, root, path);
1811 btrfs_free_path(path);
1815 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1818 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1819 struct btrfs_disk_key *disk_key;
1820 struct btrfs_chunk *chunk;
1827 struct btrfs_key key;
1829 array_size = btrfs_super_sys_array_size(super_copy);
1831 ptr = super_copy->sys_chunk_array;
1834 while (cur < array_size) {
1835 disk_key = (struct btrfs_disk_key *)ptr;
1836 btrfs_disk_key_to_cpu(&key, disk_key);
1838 len = sizeof(*disk_key);
1840 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1841 chunk = (struct btrfs_chunk *)(ptr + len);
1842 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1843 len += btrfs_chunk_item_size(num_stripes);
1848 if (key.objectid == chunk_objectid &&
1849 key.offset == chunk_offset) {
1850 memmove(ptr, ptr + len, array_size - (cur + len));
1852 btrfs_set_super_sys_array_size(super_copy, array_size);
1861 static int btrfs_relocate_chunk(struct btrfs_root *root,
1862 u64 chunk_tree, u64 chunk_objectid,
1865 struct extent_map_tree *em_tree;
1866 struct btrfs_root *extent_root;
1867 struct btrfs_trans_handle *trans;
1868 struct extent_map *em;
1869 struct map_lookup *map;
1873 root = root->fs_info->chunk_root;
1874 extent_root = root->fs_info->extent_root;
1875 em_tree = &root->fs_info->mapping_tree.map_tree;
1877 ret = btrfs_can_relocate(extent_root, chunk_offset);
1881 /* step one, relocate all the extents inside this chunk */
1882 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1886 trans = btrfs_start_transaction(root, 0);
1887 BUG_ON(IS_ERR(trans));
1892 * step two, delete the device extents and the
1893 * chunk tree entries
1895 read_lock(&em_tree->lock);
1896 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1897 read_unlock(&em_tree->lock);
1899 BUG_ON(em->start > chunk_offset ||
1900 em->start + em->len < chunk_offset);
1901 map = (struct map_lookup *)em->bdev;
1903 for (i = 0; i < map->num_stripes; i++) {
1904 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1905 map->stripes[i].physical);
1908 if (map->stripes[i].dev) {
1909 ret = btrfs_update_device(trans, map->stripes[i].dev);
1913 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1918 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1919 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1923 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1926 write_lock(&em_tree->lock);
1927 remove_extent_mapping(em_tree, em);
1928 write_unlock(&em_tree->lock);
1933 /* once for the tree */
1934 free_extent_map(em);
1936 free_extent_map(em);
1938 unlock_chunks(root);
1939 btrfs_end_transaction(trans, root);
1943 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1945 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1946 struct btrfs_path *path;
1947 struct extent_buffer *leaf;
1948 struct btrfs_chunk *chunk;
1949 struct btrfs_key key;
1950 struct btrfs_key found_key;
1951 u64 chunk_tree = chunk_root->root_key.objectid;
1953 bool retried = false;
1957 path = btrfs_alloc_path();
1962 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1963 key.offset = (u64)-1;
1964 key.type = BTRFS_CHUNK_ITEM_KEY;
1967 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1972 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1979 leaf = path->nodes[0];
1980 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1982 chunk = btrfs_item_ptr(leaf, path->slots[0],
1983 struct btrfs_chunk);
1984 chunk_type = btrfs_chunk_type(leaf, chunk);
1985 btrfs_release_path(chunk_root, path);
1987 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1988 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1997 if (found_key.offset == 0)
1999 key.offset = found_key.offset - 1;
2002 if (failed && !retried) {
2006 } else if (failed && retried) {
2011 btrfs_free_path(path);
2015 static u64 div_factor(u64 num, int factor)
2024 int btrfs_balance(struct btrfs_root *dev_root)
2027 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2028 struct btrfs_device *device;
2031 struct btrfs_path *path;
2032 struct btrfs_key key;
2033 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2034 struct btrfs_trans_handle *trans;
2035 struct btrfs_key found_key;
2037 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2040 if (!capable(CAP_SYS_ADMIN))
2043 mutex_lock(&dev_root->fs_info->volume_mutex);
2044 dev_root = dev_root->fs_info->dev_root;
2046 /* step one make some room on all the devices */
2047 list_for_each_entry(device, devices, dev_list) {
2048 old_size = device->total_bytes;
2049 size_to_free = div_factor(old_size, 1);
2050 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2051 if (!device->writeable ||
2052 device->total_bytes - device->bytes_used > size_to_free)
2055 ret = btrfs_shrink_device(device, old_size - size_to_free);
2060 trans = btrfs_start_transaction(dev_root, 0);
2061 BUG_ON(IS_ERR(trans));
2063 ret = btrfs_grow_device(trans, device, old_size);
2066 btrfs_end_transaction(trans, dev_root);
2069 /* step two, relocate all the chunks */
2070 path = btrfs_alloc_path();
2073 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2074 key.offset = (u64)-1;
2075 key.type = BTRFS_CHUNK_ITEM_KEY;
2078 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2083 * this shouldn't happen, it means the last relocate
2089 ret = btrfs_previous_item(chunk_root, path, 0,
2090 BTRFS_CHUNK_ITEM_KEY);
2094 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2096 if (found_key.objectid != key.objectid)
2099 /* chunk zero is special */
2100 if (found_key.offset == 0)
2103 btrfs_release_path(chunk_root, path);
2104 ret = btrfs_relocate_chunk(chunk_root,
2105 chunk_root->root_key.objectid,
2108 BUG_ON(ret && ret != -ENOSPC);
2109 key.offset = found_key.offset - 1;
2113 btrfs_free_path(path);
2114 mutex_unlock(&dev_root->fs_info->volume_mutex);
2119 * shrinking a device means finding all of the device extents past
2120 * the new size, and then following the back refs to the chunks.
2121 * The chunk relocation code actually frees the device extent
2123 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2125 struct btrfs_trans_handle *trans;
2126 struct btrfs_root *root = device->dev_root;
2127 struct btrfs_dev_extent *dev_extent = NULL;
2128 struct btrfs_path *path;
2136 bool retried = false;
2137 struct extent_buffer *l;
2138 struct btrfs_key key;
2139 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2140 u64 old_total = btrfs_super_total_bytes(super_copy);
2141 u64 old_size = device->total_bytes;
2142 u64 diff = device->total_bytes - new_size;
2144 if (new_size >= device->total_bytes)
2147 path = btrfs_alloc_path();
2155 device->total_bytes = new_size;
2156 if (device->writeable)
2157 device->fs_devices->total_rw_bytes -= diff;
2158 unlock_chunks(root);
2161 key.objectid = device->devid;
2162 key.offset = (u64)-1;
2163 key.type = BTRFS_DEV_EXTENT_KEY;
2166 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2170 ret = btrfs_previous_item(root, path, 0, key.type);
2175 btrfs_release_path(root, path);
2180 slot = path->slots[0];
2181 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2183 if (key.objectid != device->devid) {
2184 btrfs_release_path(root, path);
2188 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2189 length = btrfs_dev_extent_length(l, dev_extent);
2191 if (key.offset + length <= new_size) {
2192 btrfs_release_path(root, path);
2196 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2197 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2198 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2199 btrfs_release_path(root, path);
2201 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2203 if (ret && ret != -ENOSPC)
2210 if (failed && !retried) {
2214 } else if (failed && retried) {
2218 device->total_bytes = old_size;
2219 if (device->writeable)
2220 device->fs_devices->total_rw_bytes += diff;
2221 unlock_chunks(root);
2225 /* Shrinking succeeded, else we would be at "done". */
2226 trans = btrfs_start_transaction(root, 0);
2227 if (IS_ERR(trans)) {
2228 ret = PTR_ERR(trans);
2234 device->disk_total_bytes = new_size;
2235 /* Now btrfs_update_device() will change the on-disk size. */
2236 ret = btrfs_update_device(trans, device);
2238 unlock_chunks(root);
2239 btrfs_end_transaction(trans, root);
2242 WARN_ON(diff > old_total);
2243 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2244 unlock_chunks(root);
2245 btrfs_end_transaction(trans, root);
2247 btrfs_free_path(path);
2251 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2252 struct btrfs_root *root,
2253 struct btrfs_key *key,
2254 struct btrfs_chunk *chunk, int item_size)
2256 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2257 struct btrfs_disk_key disk_key;
2261 array_size = btrfs_super_sys_array_size(super_copy);
2262 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2265 ptr = super_copy->sys_chunk_array + array_size;
2266 btrfs_cpu_key_to_disk(&disk_key, key);
2267 memcpy(ptr, &disk_key, sizeof(disk_key));
2268 ptr += sizeof(disk_key);
2269 memcpy(ptr, chunk, item_size);
2270 item_size += sizeof(disk_key);
2271 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2275 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2276 int num_stripes, int sub_stripes)
2278 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2280 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2281 return calc_size * (num_stripes / sub_stripes);
2283 return calc_size * num_stripes;
2286 /* Used to sort the devices by max_avail(descending sort) */
2287 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2289 if (((struct btrfs_device_info *)dev_info1)->max_avail >
2290 ((struct btrfs_device_info *)dev_info2)->max_avail)
2292 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2293 ((struct btrfs_device_info *)dev_info2)->max_avail)
2299 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2300 int *num_stripes, int *min_stripes,
2307 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2308 *num_stripes = fs_devices->rw_devices;
2311 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2315 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2316 if (fs_devices->rw_devices < 2)
2321 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2322 *num_stripes = fs_devices->rw_devices;
2323 if (*num_stripes < 4)
2325 *num_stripes &= ~(u32)1;
2333 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2334 u64 proposed_size, u64 type,
2335 int num_stripes, int small_stripe)
2337 int min_stripe_size = 1 * 1024 * 1024;
2338 u64 calc_size = proposed_size;
2339 u64 max_chunk_size = calc_size;
2342 if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2343 BTRFS_BLOCK_GROUP_DUP |
2344 BTRFS_BLOCK_GROUP_RAID10))
2347 if (type & BTRFS_BLOCK_GROUP_DATA) {
2348 max_chunk_size = 10 * calc_size;
2349 min_stripe_size = 64 * 1024 * 1024;
2350 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2351 max_chunk_size = 256 * 1024 * 1024;
2352 min_stripe_size = 32 * 1024 * 1024;
2353 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2354 calc_size = 8 * 1024 * 1024;
2355 max_chunk_size = calc_size * 2;
2356 min_stripe_size = 1 * 1024 * 1024;
2359 /* we don't want a chunk larger than 10% of writeable space */
2360 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2363 if (calc_size * num_stripes > max_chunk_size * ncopies) {
2364 calc_size = max_chunk_size * ncopies;
2365 do_div(calc_size, num_stripes);
2366 do_div(calc_size, BTRFS_STRIPE_LEN);
2367 calc_size *= BTRFS_STRIPE_LEN;
2370 /* we don't want tiny stripes */
2372 calc_size = max_t(u64, min_stripe_size, calc_size);
2375 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2376 * we end up with something bigger than a stripe
2378 calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2380 do_div(calc_size, BTRFS_STRIPE_LEN);
2381 calc_size *= BTRFS_STRIPE_LEN;
2386 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2389 struct map_lookup *new;
2390 size_t len = map_lookup_size(num_stripes);
2392 BUG_ON(map->num_stripes < num_stripes);
2394 if (map->num_stripes == num_stripes)
2397 new = kmalloc(len, GFP_NOFS);
2399 /* just change map->num_stripes */
2400 map->num_stripes = num_stripes;
2404 memcpy(new, map, len);
2405 new->num_stripes = num_stripes;
2411 * helper to allocate device space from btrfs_device_info, in which we stored
2412 * max free space information of every device. It is used when we can not
2413 * allocate chunks by default size.
2415 * By this helper, we can allocate a new chunk as larger as possible.
2417 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2418 struct btrfs_fs_devices *fs_devices,
2419 struct btrfs_device_info *devices,
2420 int nr_device, u64 type,
2421 struct map_lookup **map_lookup,
2422 int min_stripes, u64 *stripe_size)
2424 int i, index, sort_again = 0;
2425 int min_devices = min_stripes;
2426 u64 max_avail, min_free;
2427 struct map_lookup *map = *map_lookup;
2430 if (nr_device < min_stripes)
2433 btrfs_descending_sort_devices(devices, nr_device);
2435 max_avail = devices[0].max_avail;
2439 for (i = 0; i < nr_device; i++) {
2441 * if dev_offset = 0, it means the free space of this device
2442 * is less than what we need, and we didn't search max avail
2443 * extent on this device, so do it now.
2445 if (!devices[i].dev_offset) {
2446 ret = find_free_dev_extent(trans, devices[i].dev,
2448 &devices[i].dev_offset,
2449 &devices[i].max_avail);
2450 if (ret != 0 && ret != -ENOSPC)
2456 /* we update the max avail free extent of each devices, sort again */
2458 btrfs_descending_sort_devices(devices, nr_device);
2460 if (type & BTRFS_BLOCK_GROUP_DUP)
2463 if (!devices[min_devices - 1].max_avail)
2466 max_avail = devices[min_devices - 1].max_avail;
2467 if (type & BTRFS_BLOCK_GROUP_DUP)
2468 do_div(max_avail, 2);
2470 max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2472 if (type & BTRFS_BLOCK_GROUP_DUP)
2473 min_free = max_avail * 2;
2475 min_free = max_avail;
2477 if (min_free > devices[min_devices - 1].max_avail)
2480 map = __shrink_map_lookup_stripes(map, min_stripes);
2481 *stripe_size = max_avail;
2484 for (i = 0; i < min_stripes; i++) {
2485 map->stripes[i].dev = devices[index].dev;
2486 map->stripes[i].physical = devices[index].dev_offset;
2487 if (type & BTRFS_BLOCK_GROUP_DUP) {
2489 map->stripes[i].dev = devices[index].dev;
2490 map->stripes[i].physical = devices[index].dev_offset +
2500 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2501 struct btrfs_root *extent_root,
2502 struct map_lookup **map_ret,
2503 u64 *num_bytes, u64 *stripe_size,
2504 u64 start, u64 type)
2506 struct btrfs_fs_info *info = extent_root->fs_info;
2507 struct btrfs_device *device = NULL;
2508 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2509 struct list_head *cur;
2510 struct map_lookup *map;
2511 struct extent_map_tree *em_tree;
2512 struct extent_map *em;
2513 struct btrfs_device_info *devices_info;
2514 struct list_head private_devs;
2515 u64 calc_size = 1024 * 1024 * 1024;
2522 int min_devices; /* the min number of devices we need */
2527 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2528 (type & BTRFS_BLOCK_GROUP_DUP)) {
2530 type &= ~BTRFS_BLOCK_GROUP_DUP;
2532 if (list_empty(&fs_devices->alloc_list))
2535 ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2536 &min_stripes, &sub_stripes);
2540 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2545 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2550 map->num_stripes = num_stripes;
2552 cur = fs_devices->alloc_list.next;
2556 calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2559 if (type & BTRFS_BLOCK_GROUP_DUP) {
2560 min_free = calc_size * 2;
2563 min_free = calc_size;
2564 min_devices = min_stripes;
2567 INIT_LIST_HEAD(&private_devs);
2568 while (index < num_stripes) {
2569 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2570 BUG_ON(!device->writeable);
2571 if (device->total_bytes > device->bytes_used)
2572 avail = device->total_bytes - device->bytes_used;
2577 if (device->in_fs_metadata && avail >= min_free) {
2578 ret = find_free_dev_extent(trans, device, min_free,
2579 &devices_info[i].dev_offset,
2580 &devices_info[i].max_avail);
2582 list_move_tail(&device->dev_alloc_list,
2584 map->stripes[index].dev = device;
2585 map->stripes[index].physical =
2586 devices_info[i].dev_offset;
2588 if (type & BTRFS_BLOCK_GROUP_DUP) {
2589 map->stripes[index].dev = device;
2590 map->stripes[index].physical =
2591 devices_info[i].dev_offset +
2595 } else if (ret != -ENOSPC)
2598 devices_info[i].dev = device;
2600 } else if (device->in_fs_metadata &&
2601 avail >= BTRFS_STRIPE_LEN) {
2602 devices_info[i].dev = device;
2603 devices_info[i].max_avail = avail;
2607 if (cur == &fs_devices->alloc_list)
2611 list_splice(&private_devs, &fs_devices->alloc_list);
2612 if (index < num_stripes) {
2613 if (index >= min_stripes) {
2614 num_stripes = index;
2615 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2616 num_stripes /= sub_stripes;
2617 num_stripes *= sub_stripes;
2620 map = __shrink_map_lookup_stripes(map, num_stripes);
2621 } else if (i >= min_devices) {
2622 ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2623 devices_info, i, type,
2633 map->sector_size = extent_root->sectorsize;
2634 map->stripe_len = BTRFS_STRIPE_LEN;
2635 map->io_align = BTRFS_STRIPE_LEN;
2636 map->io_width = BTRFS_STRIPE_LEN;
2638 map->sub_stripes = sub_stripes;
2641 *stripe_size = calc_size;
2642 *num_bytes = chunk_bytes_by_type(type, calc_size,
2643 map->num_stripes, sub_stripes);
2645 em = alloc_extent_map(GFP_NOFS);
2650 em->bdev = (struct block_device *)map;
2652 em->len = *num_bytes;
2653 em->block_start = 0;
2654 em->block_len = em->len;
2656 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2657 write_lock(&em_tree->lock);
2658 ret = add_extent_mapping(em_tree, em);
2659 write_unlock(&em_tree->lock);
2661 free_extent_map(em);
2663 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2664 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2669 while (index < map->num_stripes) {
2670 device = map->stripes[index].dev;
2671 dev_offset = map->stripes[index].physical;
2673 ret = btrfs_alloc_dev_extent(trans, device,
2674 info->chunk_root->root_key.objectid,
2675 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2676 start, dev_offset, calc_size);
2681 kfree(devices_info);
2686 kfree(devices_info);
2690 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2691 struct btrfs_root *extent_root,
2692 struct map_lookup *map, u64 chunk_offset,
2693 u64 chunk_size, u64 stripe_size)
2696 struct btrfs_key key;
2697 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2698 struct btrfs_device *device;
2699 struct btrfs_chunk *chunk;
2700 struct btrfs_stripe *stripe;
2701 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2705 chunk = kzalloc(item_size, GFP_NOFS);
2710 while (index < map->num_stripes) {
2711 device = map->stripes[index].dev;
2712 device->bytes_used += stripe_size;
2713 ret = btrfs_update_device(trans, device);
2719 stripe = &chunk->stripe;
2720 while (index < map->num_stripes) {
2721 device = map->stripes[index].dev;
2722 dev_offset = map->stripes[index].physical;
2724 btrfs_set_stack_stripe_devid(stripe, device->devid);
2725 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2726 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2731 btrfs_set_stack_chunk_length(chunk, chunk_size);
2732 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2733 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2734 btrfs_set_stack_chunk_type(chunk, map->type);
2735 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2736 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2737 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2738 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2739 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2741 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2742 key.type = BTRFS_CHUNK_ITEM_KEY;
2743 key.offset = chunk_offset;
2745 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2748 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2749 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2758 * Chunk allocation falls into two parts. The first part does works
2759 * that make the new allocated chunk useable, but not do any operation
2760 * that modifies the chunk tree. The second part does the works that
2761 * require modifying the chunk tree. This division is important for the
2762 * bootstrap process of adding storage to a seed btrfs.
2764 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2765 struct btrfs_root *extent_root, u64 type)
2770 struct map_lookup *map;
2771 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2774 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2779 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2780 &stripe_size, chunk_offset, type);
2784 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2785 chunk_size, stripe_size);
2790 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root,
2792 struct btrfs_device *device)
2795 u64 sys_chunk_offset;
2799 u64 sys_stripe_size;
2801 struct map_lookup *map;
2802 struct map_lookup *sys_map;
2803 struct btrfs_fs_info *fs_info = root->fs_info;
2804 struct btrfs_root *extent_root = fs_info->extent_root;
2807 ret = find_next_chunk(fs_info->chunk_root,
2808 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2811 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2812 (fs_info->metadata_alloc_profile &
2813 fs_info->avail_metadata_alloc_bits);
2814 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2816 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2817 &stripe_size, chunk_offset, alloc_profile);
2820 sys_chunk_offset = chunk_offset + chunk_size;
2822 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2823 (fs_info->system_alloc_profile &
2824 fs_info->avail_system_alloc_bits);
2825 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2827 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2828 &sys_chunk_size, &sys_stripe_size,
2829 sys_chunk_offset, alloc_profile);
2832 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2836 * Modifying chunk tree needs allocating new blocks from both
2837 * system block group and metadata block group. So we only can
2838 * do operations require modifying the chunk tree after both
2839 * block groups were created.
2841 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2842 chunk_size, stripe_size);
2845 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2846 sys_chunk_offset, sys_chunk_size,
2852 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2854 struct extent_map *em;
2855 struct map_lookup *map;
2856 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2860 read_lock(&map_tree->map_tree.lock);
2861 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2862 read_unlock(&map_tree->map_tree.lock);
2866 if (btrfs_test_opt(root, DEGRADED)) {
2867 free_extent_map(em);
2871 map = (struct map_lookup *)em->bdev;
2872 for (i = 0; i < map->num_stripes; i++) {
2873 if (!map->stripes[i].dev->writeable) {
2878 free_extent_map(em);
2882 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2884 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2887 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2889 struct extent_map *em;
2892 write_lock(&tree->map_tree.lock);
2893 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2895 remove_extent_mapping(&tree->map_tree, em);
2896 write_unlock(&tree->map_tree.lock);
2901 free_extent_map(em);
2902 /* once for the tree */
2903 free_extent_map(em);
2907 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2909 struct extent_map *em;
2910 struct map_lookup *map;
2911 struct extent_map_tree *em_tree = &map_tree->map_tree;
2914 read_lock(&em_tree->lock);
2915 em = lookup_extent_mapping(em_tree, logical, len);
2916 read_unlock(&em_tree->lock);
2919 BUG_ON(em->start > logical || em->start + em->len < logical);
2920 map = (struct map_lookup *)em->bdev;
2921 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2922 ret = map->num_stripes;
2923 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2924 ret = map->sub_stripes;
2927 free_extent_map(em);
2931 static int find_live_mirror(struct map_lookup *map, int first, int num,
2935 if (map->stripes[optimal].dev->bdev)
2937 for (i = first; i < first + num; i++) {
2938 if (map->stripes[i].dev->bdev)
2941 /* we couldn't find one that doesn't fail. Just return something
2942 * and the io error handling code will clean up eventually
2947 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2948 u64 logical, u64 *length,
2949 struct btrfs_multi_bio **multi_ret,
2950 int mirror_num, struct page *unplug_page)
2952 struct extent_map *em;
2953 struct map_lookup *map;
2954 struct extent_map_tree *em_tree = &map_tree->map_tree;
2958 int stripes_allocated = 8;
2959 int stripes_required = 1;
2964 struct btrfs_multi_bio *multi = NULL;
2966 if (multi_ret && !(rw & REQ_WRITE))
2967 stripes_allocated = 1;
2970 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2975 atomic_set(&multi->error, 0);
2978 read_lock(&em_tree->lock);
2979 em = lookup_extent_mapping(em_tree, logical, *length);
2980 read_unlock(&em_tree->lock);
2982 if (!em && unplug_page) {
2988 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2989 (unsigned long long)logical,
2990 (unsigned long long)*length);
2994 BUG_ON(em->start > logical || em->start + em->len < logical);
2995 map = (struct map_lookup *)em->bdev;
2996 offset = logical - em->start;
2998 if (mirror_num > map->num_stripes)
3001 /* if our multi bio struct is too small, back off and try again */
3002 if (rw & REQ_WRITE) {
3003 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3004 BTRFS_BLOCK_GROUP_DUP)) {
3005 stripes_required = map->num_stripes;
3007 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3008 stripes_required = map->sub_stripes;
3012 if (multi_ret && (rw & REQ_WRITE) &&
3013 stripes_allocated < stripes_required) {
3014 stripes_allocated = map->num_stripes;
3015 free_extent_map(em);
3021 * stripe_nr counts the total number of stripes we have to stride
3022 * to get to this block
3024 do_div(stripe_nr, map->stripe_len);
3026 stripe_offset = stripe_nr * map->stripe_len;
3027 BUG_ON(offset < stripe_offset);
3029 /* stripe_offset is the offset of this block in its stripe*/
3030 stripe_offset = offset - stripe_offset;
3032 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3033 BTRFS_BLOCK_GROUP_RAID10 |
3034 BTRFS_BLOCK_GROUP_DUP)) {
3035 /* we limit the length of each bio to what fits in a stripe */
3036 *length = min_t(u64, em->len - offset,
3037 map->stripe_len - stripe_offset);
3039 *length = em->len - offset;
3042 if (!multi_ret && !unplug_page)
3047 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3048 if (unplug_page || (rw & REQ_WRITE))
3049 num_stripes = map->num_stripes;
3050 else if (mirror_num)
3051 stripe_index = mirror_num - 1;
3053 stripe_index = find_live_mirror(map, 0,
3055 current->pid % map->num_stripes);
3058 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3060 num_stripes = map->num_stripes;
3061 else if (mirror_num)
3062 stripe_index = mirror_num - 1;
3064 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3065 int factor = map->num_stripes / map->sub_stripes;
3067 stripe_index = do_div(stripe_nr, factor);
3068 stripe_index *= map->sub_stripes;
3070 if (unplug_page || (rw & REQ_WRITE))
3071 num_stripes = map->sub_stripes;
3072 else if (mirror_num)
3073 stripe_index += mirror_num - 1;
3075 stripe_index = find_live_mirror(map, stripe_index,
3076 map->sub_stripes, stripe_index +
3077 current->pid % map->sub_stripes);
3081 * after this do_div call, stripe_nr is the number of stripes
3082 * on this device we have to walk to find the data, and
3083 * stripe_index is the number of our device in the stripe array
3085 stripe_index = do_div(stripe_nr, map->num_stripes);
3087 BUG_ON(stripe_index >= map->num_stripes);
3089 for (i = 0; i < num_stripes; i++) {
3091 struct btrfs_device *device;
3092 struct backing_dev_info *bdi;
3094 device = map->stripes[stripe_index].dev;
3096 bdi = blk_get_backing_dev_info(device->bdev);
3097 if (bdi->unplug_io_fn)
3098 bdi->unplug_io_fn(bdi, unplug_page);
3101 multi->stripes[i].physical =
3102 map->stripes[stripe_index].physical +
3103 stripe_offset + stripe_nr * map->stripe_len;
3104 multi->stripes[i].dev = map->stripes[stripe_index].dev;
3110 multi->num_stripes = num_stripes;
3111 multi->max_errors = max_errors;
3114 free_extent_map(em);
3118 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3119 u64 logical, u64 *length,
3120 struct btrfs_multi_bio **multi_ret, int mirror_num)
3122 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3126 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3127 u64 chunk_start, u64 physical, u64 devid,
3128 u64 **logical, int *naddrs, int *stripe_len)
3130 struct extent_map_tree *em_tree = &map_tree->map_tree;
3131 struct extent_map *em;
3132 struct map_lookup *map;
3139 read_lock(&em_tree->lock);
3140 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3141 read_unlock(&em_tree->lock);
3143 BUG_ON(!em || em->start != chunk_start);
3144 map = (struct map_lookup *)em->bdev;
3147 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3148 do_div(length, map->num_stripes / map->sub_stripes);
3149 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3150 do_div(length, map->num_stripes);
3152 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3155 for (i = 0; i < map->num_stripes; i++) {
3156 if (devid && map->stripes[i].dev->devid != devid)
3158 if (map->stripes[i].physical > physical ||
3159 map->stripes[i].physical + length <= physical)
3162 stripe_nr = physical - map->stripes[i].physical;
3163 do_div(stripe_nr, map->stripe_len);
3165 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3166 stripe_nr = stripe_nr * map->num_stripes + i;
3167 do_div(stripe_nr, map->sub_stripes);
3168 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3169 stripe_nr = stripe_nr * map->num_stripes + i;
3171 bytenr = chunk_start + stripe_nr * map->stripe_len;
3172 WARN_ON(nr >= map->num_stripes);
3173 for (j = 0; j < nr; j++) {
3174 if (buf[j] == bytenr)
3178 WARN_ON(nr >= map->num_stripes);
3185 *stripe_len = map->stripe_len;
3187 free_extent_map(em);
3191 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3192 u64 logical, struct page *page)
3194 u64 length = PAGE_CACHE_SIZE;
3195 return __btrfs_map_block(map_tree, READ, logical, &length,
3199 static void end_bio_multi_stripe(struct bio *bio, int err)
3201 struct btrfs_multi_bio *multi = bio->bi_private;
3202 int is_orig_bio = 0;
3205 atomic_inc(&multi->error);
3207 if (bio == multi->orig_bio)
3210 if (atomic_dec_and_test(&multi->stripes_pending)) {
3213 bio = multi->orig_bio;
3215 bio->bi_private = multi->private;
3216 bio->bi_end_io = multi->end_io;
3217 /* only send an error to the higher layers if it is
3218 * beyond the tolerance of the multi-bio
3220 if (atomic_read(&multi->error) > multi->max_errors) {
3224 * this bio is actually up to date, we didn't
3225 * go over the max number of errors
3227 set_bit(BIO_UPTODATE, &bio->bi_flags);
3232 bio_endio(bio, err);
3233 } else if (!is_orig_bio) {
3238 struct async_sched {
3241 struct btrfs_fs_info *info;
3242 struct btrfs_work work;
3246 * see run_scheduled_bios for a description of why bios are collected for
3249 * This will add one bio to the pending list for a device and make sure
3250 * the work struct is scheduled.
3252 static noinline int schedule_bio(struct btrfs_root *root,
3253 struct btrfs_device *device,
3254 int rw, struct bio *bio)
3256 int should_queue = 1;
3257 struct btrfs_pending_bios *pending_bios;
3259 /* don't bother with additional async steps for reads, right now */
3260 if (!(rw & REQ_WRITE)) {
3262 submit_bio(rw, bio);
3268 * nr_async_bios allows us to reliably return congestion to the
3269 * higher layers. Otherwise, the async bio makes it appear we have
3270 * made progress against dirty pages when we've really just put it
3271 * on a queue for later
3273 atomic_inc(&root->fs_info->nr_async_bios);
3274 WARN_ON(bio->bi_next);
3275 bio->bi_next = NULL;
3278 spin_lock(&device->io_lock);
3279 if (bio->bi_rw & REQ_SYNC)
3280 pending_bios = &device->pending_sync_bios;
3282 pending_bios = &device->pending_bios;
3284 if (pending_bios->tail)
3285 pending_bios->tail->bi_next = bio;
3287 pending_bios->tail = bio;
3288 if (!pending_bios->head)
3289 pending_bios->head = bio;
3290 if (device->running_pending)
3293 spin_unlock(&device->io_lock);
3296 btrfs_queue_worker(&root->fs_info->submit_workers,
3301 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3302 int mirror_num, int async_submit)
3304 struct btrfs_mapping_tree *map_tree;
3305 struct btrfs_device *dev;
3306 struct bio *first_bio = bio;
3307 u64 logical = (u64)bio->bi_sector << 9;
3310 struct btrfs_multi_bio *multi = NULL;
3315 length = bio->bi_size;
3316 map_tree = &root->fs_info->mapping_tree;
3317 map_length = length;
3319 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3323 total_devs = multi->num_stripes;
3324 if (map_length < length) {
3325 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3326 "len %llu\n", (unsigned long long)logical,
3327 (unsigned long long)length,
3328 (unsigned long long)map_length);
3331 multi->end_io = first_bio->bi_end_io;
3332 multi->private = first_bio->bi_private;
3333 multi->orig_bio = first_bio;
3334 atomic_set(&multi->stripes_pending, multi->num_stripes);
3336 while (dev_nr < total_devs) {
3337 if (total_devs > 1) {
3338 if (dev_nr < total_devs - 1) {
3339 bio = bio_clone(first_bio, GFP_NOFS);
3344 bio->bi_private = multi;
3345 bio->bi_end_io = end_bio_multi_stripe;
3347 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3348 dev = multi->stripes[dev_nr].dev;
3349 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3350 bio->bi_bdev = dev->bdev;
3352 schedule_bio(root, dev, rw, bio);
3354 submit_bio(rw, bio);
3356 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3357 bio->bi_sector = logical >> 9;
3358 bio_endio(bio, -EIO);
3362 if (total_devs == 1)
3367 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3370 struct btrfs_device *device;
3371 struct btrfs_fs_devices *cur_devices;
3373 cur_devices = root->fs_info->fs_devices;
3374 while (cur_devices) {
3376 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3377 device = __find_device(&cur_devices->devices,
3382 cur_devices = cur_devices->seed;
3387 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3388 u64 devid, u8 *dev_uuid)
3390 struct btrfs_device *device;
3391 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3393 device = kzalloc(sizeof(*device), GFP_NOFS);
3396 list_add(&device->dev_list,
3397 &fs_devices->devices);
3398 device->barriers = 1;
3399 device->dev_root = root->fs_info->dev_root;
3400 device->devid = devid;
3401 device->work.func = pending_bios_fn;
3402 device->fs_devices = fs_devices;
3403 device->missing = 1;
3404 fs_devices->num_devices++;
3405 fs_devices->missing_devices++;
3406 spin_lock_init(&device->io_lock);
3407 INIT_LIST_HEAD(&device->dev_alloc_list);
3408 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3412 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3413 struct extent_buffer *leaf,
3414 struct btrfs_chunk *chunk)
3416 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3417 struct map_lookup *map;
3418 struct extent_map *em;
3422 u8 uuid[BTRFS_UUID_SIZE];
3427 logical = key->offset;
3428 length = btrfs_chunk_length(leaf, chunk);
3430 read_lock(&map_tree->map_tree.lock);
3431 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3432 read_unlock(&map_tree->map_tree.lock);
3434 /* already mapped? */
3435 if (em && em->start <= logical && em->start + em->len > logical) {
3436 free_extent_map(em);
3439 free_extent_map(em);
3442 em = alloc_extent_map(GFP_NOFS);
3445 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3446 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3448 free_extent_map(em);
3452 em->bdev = (struct block_device *)map;
3453 em->start = logical;
3455 em->block_start = 0;
3456 em->block_len = em->len;
3458 map->num_stripes = num_stripes;
3459 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3460 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3461 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3462 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3463 map->type = btrfs_chunk_type(leaf, chunk);
3464 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3465 for (i = 0; i < num_stripes; i++) {
3466 map->stripes[i].physical =
3467 btrfs_stripe_offset_nr(leaf, chunk, i);
3468 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3469 read_extent_buffer(leaf, uuid, (unsigned long)
3470 btrfs_stripe_dev_uuid_nr(chunk, i),
3472 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3474 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3476 free_extent_map(em);
3479 if (!map->stripes[i].dev) {
3480 map->stripes[i].dev =
3481 add_missing_dev(root, devid, uuid);
3482 if (!map->stripes[i].dev) {
3484 free_extent_map(em);
3488 map->stripes[i].dev->in_fs_metadata = 1;
3491 write_lock(&map_tree->map_tree.lock);
3492 ret = add_extent_mapping(&map_tree->map_tree, em);
3493 write_unlock(&map_tree->map_tree.lock);
3495 free_extent_map(em);
3500 static int fill_device_from_item(struct extent_buffer *leaf,
3501 struct btrfs_dev_item *dev_item,
3502 struct btrfs_device *device)
3506 device->devid = btrfs_device_id(leaf, dev_item);
3507 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3508 device->total_bytes = device->disk_total_bytes;
3509 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3510 device->type = btrfs_device_type(leaf, dev_item);
3511 device->io_align = btrfs_device_io_align(leaf, dev_item);
3512 device->io_width = btrfs_device_io_width(leaf, dev_item);
3513 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3515 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3516 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3521 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3523 struct btrfs_fs_devices *fs_devices;
3526 mutex_lock(&uuid_mutex);
3528 fs_devices = root->fs_info->fs_devices->seed;
3529 while (fs_devices) {
3530 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3534 fs_devices = fs_devices->seed;
3537 fs_devices = find_fsid(fsid);
3543 fs_devices = clone_fs_devices(fs_devices);
3544 if (IS_ERR(fs_devices)) {
3545 ret = PTR_ERR(fs_devices);
3549 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3550 root->fs_info->bdev_holder);
3554 if (!fs_devices->seeding) {
3555 __btrfs_close_devices(fs_devices);
3556 free_fs_devices(fs_devices);
3561 fs_devices->seed = root->fs_info->fs_devices->seed;
3562 root->fs_info->fs_devices->seed = fs_devices;
3564 mutex_unlock(&uuid_mutex);
3568 static int read_one_dev(struct btrfs_root *root,
3569 struct extent_buffer *leaf,
3570 struct btrfs_dev_item *dev_item)
3572 struct btrfs_device *device;
3575 u8 fs_uuid[BTRFS_UUID_SIZE];
3576 u8 dev_uuid[BTRFS_UUID_SIZE];
3578 devid = btrfs_device_id(leaf, dev_item);
3579 read_extent_buffer(leaf, dev_uuid,
3580 (unsigned long)btrfs_device_uuid(dev_item),
3582 read_extent_buffer(leaf, fs_uuid,
3583 (unsigned long)btrfs_device_fsid(dev_item),
3586 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3587 ret = open_seed_devices(root, fs_uuid);
3588 if (ret && !btrfs_test_opt(root, DEGRADED))
3592 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3593 if (!device || !device->bdev) {
3594 if (!btrfs_test_opt(root, DEGRADED))
3598 printk(KERN_WARNING "warning devid %llu missing\n",
3599 (unsigned long long)devid);
3600 device = add_missing_dev(root, devid, dev_uuid);
3603 } else if (!device->missing) {
3605 * this happens when a device that was properly setup
3606 * in the device info lists suddenly goes bad.
3607 * device->bdev is NULL, and so we have to set
3608 * device->missing to one here
3610 root->fs_info->fs_devices->missing_devices++;
3611 device->missing = 1;
3615 if (device->fs_devices != root->fs_info->fs_devices) {
3616 BUG_ON(device->writeable);
3617 if (device->generation !=
3618 btrfs_device_generation(leaf, dev_item))
3622 fill_device_from_item(leaf, dev_item, device);
3623 device->dev_root = root->fs_info->dev_root;
3624 device->in_fs_metadata = 1;
3625 if (device->writeable)
3626 device->fs_devices->total_rw_bytes += device->total_bytes;
3631 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3633 struct btrfs_dev_item *dev_item;
3635 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3637 return read_one_dev(root, buf, dev_item);
3640 int btrfs_read_sys_array(struct btrfs_root *root)
3642 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3643 struct extent_buffer *sb;
3644 struct btrfs_disk_key *disk_key;
3645 struct btrfs_chunk *chunk;
3647 unsigned long sb_ptr;
3653 struct btrfs_key key;
3655 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3656 BTRFS_SUPER_INFO_SIZE);
3659 btrfs_set_buffer_uptodate(sb);
3660 btrfs_set_buffer_lockdep_class(sb, 0);
3662 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3663 array_size = btrfs_super_sys_array_size(super_copy);
3665 ptr = super_copy->sys_chunk_array;
3666 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3669 while (cur < array_size) {
3670 disk_key = (struct btrfs_disk_key *)ptr;
3671 btrfs_disk_key_to_cpu(&key, disk_key);
3673 len = sizeof(*disk_key); ptr += len;
3677 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3678 chunk = (struct btrfs_chunk *)sb_ptr;
3679 ret = read_one_chunk(root, &key, sb, chunk);
3682 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3683 len = btrfs_chunk_item_size(num_stripes);
3692 free_extent_buffer(sb);
3696 int btrfs_read_chunk_tree(struct btrfs_root *root)
3698 struct btrfs_path *path;
3699 struct extent_buffer *leaf;
3700 struct btrfs_key key;
3701 struct btrfs_key found_key;
3705 root = root->fs_info->chunk_root;
3707 path = btrfs_alloc_path();
3711 /* first we search for all of the device items, and then we
3712 * read in all of the chunk items. This way we can create chunk
3713 * mappings that reference all of the devices that are afound
3715 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3719 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3723 leaf = path->nodes[0];
3724 slot = path->slots[0];
3725 if (slot >= btrfs_header_nritems(leaf)) {
3726 ret = btrfs_next_leaf(root, path);
3733 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3734 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3735 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3737 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3738 struct btrfs_dev_item *dev_item;
3739 dev_item = btrfs_item_ptr(leaf, slot,
3740 struct btrfs_dev_item);
3741 ret = read_one_dev(root, leaf, dev_item);
3745 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3746 struct btrfs_chunk *chunk;
3747 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3748 ret = read_one_chunk(root, &found_key, leaf, chunk);
3754 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3756 btrfs_release_path(root, path);
3761 btrfs_free_path(path);