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 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1216 key.type = BTRFS_DEV_ITEM_KEY;
1217 key.offset = device->devid;
1220 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1229 ret = btrfs_del_item(trans, root, path);
1233 btrfs_free_path(path);
1234 unlock_chunks(root);
1235 btrfs_commit_transaction(trans, root);
1239 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1241 struct btrfs_device *device;
1242 struct btrfs_device *next_device;
1243 struct block_device *bdev;
1244 struct buffer_head *bh = NULL;
1245 struct btrfs_super_block *disk_super;
1252 mutex_lock(&uuid_mutex);
1253 mutex_lock(&root->fs_info->volume_mutex);
1255 all_avail = root->fs_info->avail_data_alloc_bits |
1256 root->fs_info->avail_system_alloc_bits |
1257 root->fs_info->avail_metadata_alloc_bits;
1259 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1260 root->fs_info->fs_devices->num_devices <= 4) {
1261 printk(KERN_ERR "btrfs: unable to go below four devices "
1267 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1268 root->fs_info->fs_devices->num_devices <= 2) {
1269 printk(KERN_ERR "btrfs: unable to go below two "
1270 "devices on raid1\n");
1275 if (strcmp(device_path, "missing") == 0) {
1276 struct list_head *devices;
1277 struct btrfs_device *tmp;
1280 devices = &root->fs_info->fs_devices->devices;
1281 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1282 list_for_each_entry(tmp, devices, dev_list) {
1283 if (tmp->in_fs_metadata && !tmp->bdev) {
1288 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1293 printk(KERN_ERR "btrfs: no missing devices found to "
1298 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1299 root->fs_info->bdev_holder);
1301 ret = PTR_ERR(bdev);
1305 set_blocksize(bdev, 4096);
1306 bh = btrfs_read_dev_super(bdev);
1311 disk_super = (struct btrfs_super_block *)bh->b_data;
1312 devid = btrfs_stack_device_id(&disk_super->dev_item);
1313 dev_uuid = disk_super->dev_item.uuid;
1314 device = btrfs_find_device(root, devid, dev_uuid,
1322 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1323 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1329 if (device->writeable) {
1330 list_del_init(&device->dev_alloc_list);
1331 root->fs_info->fs_devices->rw_devices--;
1334 ret = btrfs_shrink_device(device, 0);
1338 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1342 device->in_fs_metadata = 0;
1345 * the device list mutex makes sure that we don't change
1346 * the device list while someone else is writing out all
1347 * the device supers.
1349 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1350 list_del_init(&device->dev_list);
1351 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1353 device->fs_devices->num_devices--;
1355 if (device->missing)
1356 root->fs_info->fs_devices->missing_devices--;
1358 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1359 struct btrfs_device, dev_list);
1360 if (device->bdev == root->fs_info->sb->s_bdev)
1361 root->fs_info->sb->s_bdev = next_device->bdev;
1362 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1363 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1366 close_bdev_exclusive(device->bdev, device->mode);
1367 device->bdev = NULL;
1368 device->fs_devices->open_devices--;
1371 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1372 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1374 if (device->fs_devices->open_devices == 0) {
1375 struct btrfs_fs_devices *fs_devices;
1376 fs_devices = root->fs_info->fs_devices;
1377 while (fs_devices) {
1378 if (fs_devices->seed == device->fs_devices)
1380 fs_devices = fs_devices->seed;
1382 fs_devices->seed = device->fs_devices->seed;
1383 device->fs_devices->seed = NULL;
1384 __btrfs_close_devices(device->fs_devices);
1385 free_fs_devices(device->fs_devices);
1389 * at this point, the device is zero sized. We want to
1390 * remove it from the devices list and zero out the old super
1392 if (device->writeable) {
1393 /* make sure this device isn't detected as part of
1396 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1397 set_buffer_dirty(bh);
1398 sync_dirty_buffer(bh);
1401 kfree(device->name);
1409 close_bdev_exclusive(bdev, FMODE_READ);
1411 mutex_unlock(&root->fs_info->volume_mutex);
1412 mutex_unlock(&uuid_mutex);
1417 * does all the dirty work required for changing file system's UUID.
1419 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1420 struct btrfs_root *root)
1422 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1423 struct btrfs_fs_devices *old_devices;
1424 struct btrfs_fs_devices *seed_devices;
1425 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1426 struct btrfs_device *device;
1429 BUG_ON(!mutex_is_locked(&uuid_mutex));
1430 if (!fs_devices->seeding)
1433 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1437 old_devices = clone_fs_devices(fs_devices);
1438 if (IS_ERR(old_devices)) {
1439 kfree(seed_devices);
1440 return PTR_ERR(old_devices);
1443 list_add(&old_devices->list, &fs_uuids);
1445 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1446 seed_devices->opened = 1;
1447 INIT_LIST_HEAD(&seed_devices->devices);
1448 INIT_LIST_HEAD(&seed_devices->alloc_list);
1449 mutex_init(&seed_devices->device_list_mutex);
1450 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1451 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1452 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1453 device->fs_devices = seed_devices;
1456 fs_devices->seeding = 0;
1457 fs_devices->num_devices = 0;
1458 fs_devices->open_devices = 0;
1459 fs_devices->seed = seed_devices;
1461 generate_random_uuid(fs_devices->fsid);
1462 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1463 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1464 super_flags = btrfs_super_flags(disk_super) &
1465 ~BTRFS_SUPER_FLAG_SEEDING;
1466 btrfs_set_super_flags(disk_super, super_flags);
1472 * strore the expected generation for seed devices in device items.
1474 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1475 struct btrfs_root *root)
1477 struct btrfs_path *path;
1478 struct extent_buffer *leaf;
1479 struct btrfs_dev_item *dev_item;
1480 struct btrfs_device *device;
1481 struct btrfs_key key;
1482 u8 fs_uuid[BTRFS_UUID_SIZE];
1483 u8 dev_uuid[BTRFS_UUID_SIZE];
1487 path = btrfs_alloc_path();
1491 root = root->fs_info->chunk_root;
1492 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1494 key.type = BTRFS_DEV_ITEM_KEY;
1497 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1501 leaf = path->nodes[0];
1503 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1504 ret = btrfs_next_leaf(root, path);
1509 leaf = path->nodes[0];
1510 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1511 btrfs_release_path(root, path);
1515 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1516 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1517 key.type != BTRFS_DEV_ITEM_KEY)
1520 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1521 struct btrfs_dev_item);
1522 devid = btrfs_device_id(leaf, dev_item);
1523 read_extent_buffer(leaf, dev_uuid,
1524 (unsigned long)btrfs_device_uuid(dev_item),
1526 read_extent_buffer(leaf, fs_uuid,
1527 (unsigned long)btrfs_device_fsid(dev_item),
1529 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1532 if (device->fs_devices->seeding) {
1533 btrfs_set_device_generation(leaf, dev_item,
1534 device->generation);
1535 btrfs_mark_buffer_dirty(leaf);
1543 btrfs_free_path(path);
1547 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1549 struct btrfs_trans_handle *trans;
1550 struct btrfs_device *device;
1551 struct block_device *bdev;
1552 struct list_head *devices;
1553 struct super_block *sb = root->fs_info->sb;
1555 int seeding_dev = 0;
1558 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1561 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1563 return PTR_ERR(bdev);
1565 if (root->fs_info->fs_devices->seeding) {
1567 down_write(&sb->s_umount);
1568 mutex_lock(&uuid_mutex);
1571 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1572 mutex_lock(&root->fs_info->volume_mutex);
1574 devices = &root->fs_info->fs_devices->devices;
1576 * we have the volume lock, so we don't need the extra
1577 * device list mutex while reading the list here.
1579 list_for_each_entry(device, devices, dev_list) {
1580 if (device->bdev == bdev) {
1586 device = kzalloc(sizeof(*device), GFP_NOFS);
1588 /* we can safely leave the fs_devices entry around */
1593 device->name = kstrdup(device_path, GFP_NOFS);
1594 if (!device->name) {
1600 ret = find_next_devid(root, &device->devid);
1606 trans = btrfs_start_transaction(root, 0);
1609 device->barriers = 1;
1610 device->writeable = 1;
1611 device->work.func = pending_bios_fn;
1612 generate_random_uuid(device->uuid);
1613 spin_lock_init(&device->io_lock);
1614 device->generation = trans->transid;
1615 device->io_width = root->sectorsize;
1616 device->io_align = root->sectorsize;
1617 device->sector_size = root->sectorsize;
1618 device->total_bytes = i_size_read(bdev->bd_inode);
1619 device->disk_total_bytes = device->total_bytes;
1620 device->dev_root = root->fs_info->dev_root;
1621 device->bdev = bdev;
1622 device->in_fs_metadata = 1;
1624 set_blocksize(device->bdev, 4096);
1627 sb->s_flags &= ~MS_RDONLY;
1628 ret = btrfs_prepare_sprout(trans, root);
1632 device->fs_devices = root->fs_info->fs_devices;
1635 * we don't want write_supers to jump in here with our device
1638 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1639 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1640 list_add(&device->dev_alloc_list,
1641 &root->fs_info->fs_devices->alloc_list);
1642 root->fs_info->fs_devices->num_devices++;
1643 root->fs_info->fs_devices->open_devices++;
1644 root->fs_info->fs_devices->rw_devices++;
1645 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1647 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1648 root->fs_info->fs_devices->rotating = 1;
1650 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1651 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1652 total_bytes + device->total_bytes);
1654 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1655 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1657 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1660 ret = init_first_rw_device(trans, root, device);
1662 ret = btrfs_finish_sprout(trans, root);
1665 ret = btrfs_add_device(trans, root, device);
1669 * we've got more storage, clear any full flags on the space
1672 btrfs_clear_space_info_full(root->fs_info);
1674 unlock_chunks(root);
1675 btrfs_commit_transaction(trans, root);
1678 mutex_unlock(&uuid_mutex);
1679 up_write(&sb->s_umount);
1681 ret = btrfs_relocate_sys_chunks(root);
1685 mutex_unlock(&root->fs_info->volume_mutex);
1688 close_bdev_exclusive(bdev, 0);
1690 mutex_unlock(&uuid_mutex);
1691 up_write(&sb->s_umount);
1696 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1697 struct btrfs_device *device)
1700 struct btrfs_path *path;
1701 struct btrfs_root *root;
1702 struct btrfs_dev_item *dev_item;
1703 struct extent_buffer *leaf;
1704 struct btrfs_key key;
1706 root = device->dev_root->fs_info->chunk_root;
1708 path = btrfs_alloc_path();
1712 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1713 key.type = BTRFS_DEV_ITEM_KEY;
1714 key.offset = device->devid;
1716 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1725 leaf = path->nodes[0];
1726 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1728 btrfs_set_device_id(leaf, dev_item, device->devid);
1729 btrfs_set_device_type(leaf, dev_item, device->type);
1730 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1731 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1732 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1733 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1734 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1735 btrfs_mark_buffer_dirty(leaf);
1738 btrfs_free_path(path);
1742 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1743 struct btrfs_device *device, u64 new_size)
1745 struct btrfs_super_block *super_copy =
1746 &device->dev_root->fs_info->super_copy;
1747 u64 old_total = btrfs_super_total_bytes(super_copy);
1748 u64 diff = new_size - device->total_bytes;
1750 if (!device->writeable)
1752 if (new_size <= device->total_bytes)
1755 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1756 device->fs_devices->total_rw_bytes += diff;
1758 device->total_bytes = new_size;
1759 device->disk_total_bytes = new_size;
1760 btrfs_clear_space_info_full(device->dev_root->fs_info);
1762 return btrfs_update_device(trans, device);
1765 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1766 struct btrfs_device *device, u64 new_size)
1769 lock_chunks(device->dev_root);
1770 ret = __btrfs_grow_device(trans, device, new_size);
1771 unlock_chunks(device->dev_root);
1775 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1776 struct btrfs_root *root,
1777 u64 chunk_tree, u64 chunk_objectid,
1781 struct btrfs_path *path;
1782 struct btrfs_key key;
1784 root = root->fs_info->chunk_root;
1785 path = btrfs_alloc_path();
1789 key.objectid = chunk_objectid;
1790 key.offset = chunk_offset;
1791 key.type = BTRFS_CHUNK_ITEM_KEY;
1793 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1796 ret = btrfs_del_item(trans, root, path);
1799 btrfs_free_path(path);
1803 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1806 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1807 struct btrfs_disk_key *disk_key;
1808 struct btrfs_chunk *chunk;
1815 struct btrfs_key key;
1817 array_size = btrfs_super_sys_array_size(super_copy);
1819 ptr = super_copy->sys_chunk_array;
1822 while (cur < array_size) {
1823 disk_key = (struct btrfs_disk_key *)ptr;
1824 btrfs_disk_key_to_cpu(&key, disk_key);
1826 len = sizeof(*disk_key);
1828 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1829 chunk = (struct btrfs_chunk *)(ptr + len);
1830 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1831 len += btrfs_chunk_item_size(num_stripes);
1836 if (key.objectid == chunk_objectid &&
1837 key.offset == chunk_offset) {
1838 memmove(ptr, ptr + len, array_size - (cur + len));
1840 btrfs_set_super_sys_array_size(super_copy, array_size);
1849 static int btrfs_relocate_chunk(struct btrfs_root *root,
1850 u64 chunk_tree, u64 chunk_objectid,
1853 struct extent_map_tree *em_tree;
1854 struct btrfs_root *extent_root;
1855 struct btrfs_trans_handle *trans;
1856 struct extent_map *em;
1857 struct map_lookup *map;
1861 root = root->fs_info->chunk_root;
1862 extent_root = root->fs_info->extent_root;
1863 em_tree = &root->fs_info->mapping_tree.map_tree;
1865 ret = btrfs_can_relocate(extent_root, chunk_offset);
1869 /* step one, relocate all the extents inside this chunk */
1870 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1874 trans = btrfs_start_transaction(root, 0);
1880 * step two, delete the device extents and the
1881 * chunk tree entries
1883 read_lock(&em_tree->lock);
1884 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1885 read_unlock(&em_tree->lock);
1887 BUG_ON(em->start > chunk_offset ||
1888 em->start + em->len < chunk_offset);
1889 map = (struct map_lookup *)em->bdev;
1891 for (i = 0; i < map->num_stripes; i++) {
1892 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1893 map->stripes[i].physical);
1896 if (map->stripes[i].dev) {
1897 ret = btrfs_update_device(trans, map->stripes[i].dev);
1901 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1906 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1907 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1911 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1914 write_lock(&em_tree->lock);
1915 remove_extent_mapping(em_tree, em);
1916 write_unlock(&em_tree->lock);
1921 /* once for the tree */
1922 free_extent_map(em);
1924 free_extent_map(em);
1926 unlock_chunks(root);
1927 btrfs_end_transaction(trans, root);
1931 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1933 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1934 struct btrfs_path *path;
1935 struct extent_buffer *leaf;
1936 struct btrfs_chunk *chunk;
1937 struct btrfs_key key;
1938 struct btrfs_key found_key;
1939 u64 chunk_tree = chunk_root->root_key.objectid;
1941 bool retried = false;
1945 path = btrfs_alloc_path();
1950 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1951 key.offset = (u64)-1;
1952 key.type = BTRFS_CHUNK_ITEM_KEY;
1955 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1960 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1967 leaf = path->nodes[0];
1968 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1970 chunk = btrfs_item_ptr(leaf, path->slots[0],
1971 struct btrfs_chunk);
1972 chunk_type = btrfs_chunk_type(leaf, chunk);
1973 btrfs_release_path(chunk_root, path);
1975 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1976 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1985 if (found_key.offset == 0)
1987 key.offset = found_key.offset - 1;
1990 if (failed && !retried) {
1994 } else if (failed && retried) {
1999 btrfs_free_path(path);
2003 static u64 div_factor(u64 num, int factor)
2012 int btrfs_balance(struct btrfs_root *dev_root)
2015 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2016 struct btrfs_device *device;
2019 struct btrfs_path *path;
2020 struct btrfs_key key;
2021 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2022 struct btrfs_trans_handle *trans;
2023 struct btrfs_key found_key;
2025 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2028 if (!capable(CAP_SYS_ADMIN))
2031 mutex_lock(&dev_root->fs_info->volume_mutex);
2032 dev_root = dev_root->fs_info->dev_root;
2034 /* step one make some room on all the devices */
2035 list_for_each_entry(device, devices, dev_list) {
2036 old_size = device->total_bytes;
2037 size_to_free = div_factor(old_size, 1);
2038 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2039 if (!device->writeable ||
2040 device->total_bytes - device->bytes_used > size_to_free)
2043 ret = btrfs_shrink_device(device, old_size - size_to_free);
2048 trans = btrfs_start_transaction(dev_root, 0);
2051 ret = btrfs_grow_device(trans, device, old_size);
2054 btrfs_end_transaction(trans, dev_root);
2057 /* step two, relocate all the chunks */
2058 path = btrfs_alloc_path();
2061 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2062 key.offset = (u64)-1;
2063 key.type = BTRFS_CHUNK_ITEM_KEY;
2066 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2071 * this shouldn't happen, it means the last relocate
2077 ret = btrfs_previous_item(chunk_root, path, 0,
2078 BTRFS_CHUNK_ITEM_KEY);
2082 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2084 if (found_key.objectid != key.objectid)
2087 /* chunk zero is special */
2088 if (found_key.offset == 0)
2091 btrfs_release_path(chunk_root, path);
2092 ret = btrfs_relocate_chunk(chunk_root,
2093 chunk_root->root_key.objectid,
2096 BUG_ON(ret && ret != -ENOSPC);
2097 key.offset = found_key.offset - 1;
2101 btrfs_free_path(path);
2102 mutex_unlock(&dev_root->fs_info->volume_mutex);
2107 * shrinking a device means finding all of the device extents past
2108 * the new size, and then following the back refs to the chunks.
2109 * The chunk relocation code actually frees the device extent
2111 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2113 struct btrfs_trans_handle *trans;
2114 struct btrfs_root *root = device->dev_root;
2115 struct btrfs_dev_extent *dev_extent = NULL;
2116 struct btrfs_path *path;
2124 bool retried = false;
2125 struct extent_buffer *l;
2126 struct btrfs_key key;
2127 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2128 u64 old_total = btrfs_super_total_bytes(super_copy);
2129 u64 old_size = device->total_bytes;
2130 u64 diff = device->total_bytes - new_size;
2132 if (new_size >= device->total_bytes)
2135 path = btrfs_alloc_path();
2143 device->total_bytes = new_size;
2144 if (device->writeable)
2145 device->fs_devices->total_rw_bytes -= diff;
2146 unlock_chunks(root);
2149 key.objectid = device->devid;
2150 key.offset = (u64)-1;
2151 key.type = BTRFS_DEV_EXTENT_KEY;
2154 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2158 ret = btrfs_previous_item(root, path, 0, key.type);
2163 btrfs_release_path(root, path);
2168 slot = path->slots[0];
2169 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2171 if (key.objectid != device->devid) {
2172 btrfs_release_path(root, path);
2176 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2177 length = btrfs_dev_extent_length(l, dev_extent);
2179 if (key.offset + length <= new_size) {
2180 btrfs_release_path(root, path);
2184 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2185 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2186 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2187 btrfs_release_path(root, path);
2189 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2191 if (ret && ret != -ENOSPC)
2198 if (failed && !retried) {
2202 } else if (failed && retried) {
2206 device->total_bytes = old_size;
2207 if (device->writeable)
2208 device->fs_devices->total_rw_bytes += diff;
2209 unlock_chunks(root);
2213 /* Shrinking succeeded, else we would be at "done". */
2214 trans = btrfs_start_transaction(root, 0);
2217 device->disk_total_bytes = new_size;
2218 /* Now btrfs_update_device() will change the on-disk size. */
2219 ret = btrfs_update_device(trans, device);
2221 unlock_chunks(root);
2222 btrfs_end_transaction(trans, root);
2225 WARN_ON(diff > old_total);
2226 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2227 unlock_chunks(root);
2228 btrfs_end_transaction(trans, root);
2230 btrfs_free_path(path);
2234 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2235 struct btrfs_root *root,
2236 struct btrfs_key *key,
2237 struct btrfs_chunk *chunk, int item_size)
2239 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2240 struct btrfs_disk_key disk_key;
2244 array_size = btrfs_super_sys_array_size(super_copy);
2245 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2248 ptr = super_copy->sys_chunk_array + array_size;
2249 btrfs_cpu_key_to_disk(&disk_key, key);
2250 memcpy(ptr, &disk_key, sizeof(disk_key));
2251 ptr += sizeof(disk_key);
2252 memcpy(ptr, chunk, item_size);
2253 item_size += sizeof(disk_key);
2254 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2258 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2259 int num_stripes, int sub_stripes)
2261 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2263 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2264 return calc_size * (num_stripes / sub_stripes);
2266 return calc_size * num_stripes;
2269 /* Used to sort the devices by max_avail(descending sort) */
2270 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2272 if (((struct btrfs_device_info *)dev_info1)->max_avail >
2273 ((struct btrfs_device_info *)dev_info2)->max_avail)
2275 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2276 ((struct btrfs_device_info *)dev_info2)->max_avail)
2282 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2283 int *num_stripes, int *min_stripes,
2290 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2291 *num_stripes = fs_devices->rw_devices;
2294 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2298 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2299 if (fs_devices->rw_devices < 2)
2304 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2305 *num_stripes = fs_devices->rw_devices;
2306 if (*num_stripes < 4)
2308 *num_stripes &= ~(u32)1;
2316 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2317 u64 proposed_size, u64 type,
2318 int num_stripes, int small_stripe)
2320 int min_stripe_size = 1 * 1024 * 1024;
2321 u64 calc_size = proposed_size;
2322 u64 max_chunk_size = calc_size;
2325 if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2326 BTRFS_BLOCK_GROUP_DUP |
2327 BTRFS_BLOCK_GROUP_RAID10))
2330 if (type & BTRFS_BLOCK_GROUP_DATA) {
2331 max_chunk_size = 10 * calc_size;
2332 min_stripe_size = 64 * 1024 * 1024;
2333 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2334 max_chunk_size = 256 * 1024 * 1024;
2335 min_stripe_size = 32 * 1024 * 1024;
2336 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2337 calc_size = 8 * 1024 * 1024;
2338 max_chunk_size = calc_size * 2;
2339 min_stripe_size = 1 * 1024 * 1024;
2342 /* we don't want a chunk larger than 10% of writeable space */
2343 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2346 if (calc_size * num_stripes > max_chunk_size * ncopies) {
2347 calc_size = max_chunk_size * ncopies;
2348 do_div(calc_size, num_stripes);
2349 do_div(calc_size, BTRFS_STRIPE_LEN);
2350 calc_size *= BTRFS_STRIPE_LEN;
2353 /* we don't want tiny stripes */
2355 calc_size = max_t(u64, min_stripe_size, calc_size);
2358 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2359 * we end up with something bigger than a stripe
2361 calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2363 do_div(calc_size, BTRFS_STRIPE_LEN);
2364 calc_size *= BTRFS_STRIPE_LEN;
2369 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2372 struct map_lookup *new;
2373 size_t len = map_lookup_size(num_stripes);
2375 BUG_ON(map->num_stripes < num_stripes);
2377 if (map->num_stripes == num_stripes)
2380 new = kmalloc(len, GFP_NOFS);
2382 /* just change map->num_stripes */
2383 map->num_stripes = num_stripes;
2387 memcpy(new, map, len);
2388 new->num_stripes = num_stripes;
2394 * helper to allocate device space from btrfs_device_info, in which we stored
2395 * max free space information of every device. It is used when we can not
2396 * allocate chunks by default size.
2398 * By this helper, we can allocate a new chunk as larger as possible.
2400 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2401 struct btrfs_fs_devices *fs_devices,
2402 struct btrfs_device_info *devices,
2403 int nr_device, u64 type,
2404 struct map_lookup **map_lookup,
2405 int min_stripes, u64 *stripe_size)
2407 int i, index, sort_again = 0;
2408 int min_devices = min_stripes;
2409 u64 max_avail, min_free;
2410 struct map_lookup *map = *map_lookup;
2413 if (nr_device < min_stripes)
2416 btrfs_descending_sort_devices(devices, nr_device);
2418 max_avail = devices[0].max_avail;
2422 for (i = 0; i < nr_device; i++) {
2424 * if dev_offset = 0, it means the free space of this device
2425 * is less than what we need, and we didn't search max avail
2426 * extent on this device, so do it now.
2428 if (!devices[i].dev_offset) {
2429 ret = find_free_dev_extent(trans, devices[i].dev,
2431 &devices[i].dev_offset,
2432 &devices[i].max_avail);
2433 if (ret != 0 && ret != -ENOSPC)
2439 /* we update the max avail free extent of each devices, sort again */
2441 btrfs_descending_sort_devices(devices, nr_device);
2443 if (type & BTRFS_BLOCK_GROUP_DUP)
2446 if (!devices[min_devices - 1].max_avail)
2449 max_avail = devices[min_devices - 1].max_avail;
2450 if (type & BTRFS_BLOCK_GROUP_DUP)
2451 do_div(max_avail, 2);
2453 max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2455 if (type & BTRFS_BLOCK_GROUP_DUP)
2456 min_free = max_avail * 2;
2458 min_free = max_avail;
2460 if (min_free > devices[min_devices - 1].max_avail)
2463 map = __shrink_map_lookup_stripes(map, min_stripes);
2464 *stripe_size = max_avail;
2467 for (i = 0; i < min_stripes; i++) {
2468 map->stripes[i].dev = devices[index].dev;
2469 map->stripes[i].physical = devices[index].dev_offset;
2470 if (type & BTRFS_BLOCK_GROUP_DUP) {
2472 map->stripes[i].dev = devices[index].dev;
2473 map->stripes[i].physical = devices[index].dev_offset +
2483 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2484 struct btrfs_root *extent_root,
2485 struct map_lookup **map_ret,
2486 u64 *num_bytes, u64 *stripe_size,
2487 u64 start, u64 type)
2489 struct btrfs_fs_info *info = extent_root->fs_info;
2490 struct btrfs_device *device = NULL;
2491 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2492 struct list_head *cur;
2493 struct map_lookup *map;
2494 struct extent_map_tree *em_tree;
2495 struct extent_map *em;
2496 struct btrfs_device_info *devices_info;
2497 struct list_head private_devs;
2498 u64 calc_size = 1024 * 1024 * 1024;
2505 int min_devices; /* the min number of devices we need */
2510 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2511 (type & BTRFS_BLOCK_GROUP_DUP)) {
2513 type &= ~BTRFS_BLOCK_GROUP_DUP;
2515 if (list_empty(&fs_devices->alloc_list))
2518 ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2519 &min_stripes, &sub_stripes);
2523 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2528 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2533 map->num_stripes = num_stripes;
2535 cur = fs_devices->alloc_list.next;
2539 calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2542 if (type & BTRFS_BLOCK_GROUP_DUP) {
2543 min_free = calc_size * 2;
2546 min_free = calc_size;
2547 min_devices = min_stripes;
2550 INIT_LIST_HEAD(&private_devs);
2551 while (index < num_stripes) {
2552 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2553 BUG_ON(!device->writeable);
2554 if (device->total_bytes > device->bytes_used)
2555 avail = device->total_bytes - device->bytes_used;
2560 if (device->in_fs_metadata && avail >= min_free) {
2561 ret = find_free_dev_extent(trans, device, min_free,
2562 &devices_info[i].dev_offset,
2563 &devices_info[i].max_avail);
2565 list_move_tail(&device->dev_alloc_list,
2567 map->stripes[index].dev = device;
2568 map->stripes[index].physical =
2569 devices_info[i].dev_offset;
2571 if (type & BTRFS_BLOCK_GROUP_DUP) {
2572 map->stripes[index].dev = device;
2573 map->stripes[index].physical =
2574 devices_info[i].dev_offset +
2578 } else if (ret != -ENOSPC)
2581 devices_info[i].dev = device;
2583 } else if (device->in_fs_metadata &&
2584 avail >= BTRFS_STRIPE_LEN) {
2585 devices_info[i].dev = device;
2586 devices_info[i].max_avail = avail;
2590 if (cur == &fs_devices->alloc_list)
2594 list_splice(&private_devs, &fs_devices->alloc_list);
2595 if (index < num_stripes) {
2596 if (index >= min_stripes) {
2597 num_stripes = index;
2598 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2599 num_stripes /= sub_stripes;
2600 num_stripes *= sub_stripes;
2603 map = __shrink_map_lookup_stripes(map, num_stripes);
2604 } else if (i >= min_devices) {
2605 ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2606 devices_info, i, type,
2616 map->sector_size = extent_root->sectorsize;
2617 map->stripe_len = BTRFS_STRIPE_LEN;
2618 map->io_align = BTRFS_STRIPE_LEN;
2619 map->io_width = BTRFS_STRIPE_LEN;
2621 map->sub_stripes = sub_stripes;
2624 *stripe_size = calc_size;
2625 *num_bytes = chunk_bytes_by_type(type, calc_size,
2626 map->num_stripes, sub_stripes);
2628 em = alloc_extent_map(GFP_NOFS);
2633 em->bdev = (struct block_device *)map;
2635 em->len = *num_bytes;
2636 em->block_start = 0;
2637 em->block_len = em->len;
2639 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2640 write_lock(&em_tree->lock);
2641 ret = add_extent_mapping(em_tree, em);
2642 write_unlock(&em_tree->lock);
2644 free_extent_map(em);
2646 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2647 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2652 while (index < map->num_stripes) {
2653 device = map->stripes[index].dev;
2654 dev_offset = map->stripes[index].physical;
2656 ret = btrfs_alloc_dev_extent(trans, device,
2657 info->chunk_root->root_key.objectid,
2658 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2659 start, dev_offset, calc_size);
2664 kfree(devices_info);
2669 kfree(devices_info);
2673 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2674 struct btrfs_root *extent_root,
2675 struct map_lookup *map, u64 chunk_offset,
2676 u64 chunk_size, u64 stripe_size)
2679 struct btrfs_key key;
2680 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2681 struct btrfs_device *device;
2682 struct btrfs_chunk *chunk;
2683 struct btrfs_stripe *stripe;
2684 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2688 chunk = kzalloc(item_size, GFP_NOFS);
2693 while (index < map->num_stripes) {
2694 device = map->stripes[index].dev;
2695 device->bytes_used += stripe_size;
2696 ret = btrfs_update_device(trans, device);
2702 stripe = &chunk->stripe;
2703 while (index < map->num_stripes) {
2704 device = map->stripes[index].dev;
2705 dev_offset = map->stripes[index].physical;
2707 btrfs_set_stack_stripe_devid(stripe, device->devid);
2708 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2709 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2714 btrfs_set_stack_chunk_length(chunk, chunk_size);
2715 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2716 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2717 btrfs_set_stack_chunk_type(chunk, map->type);
2718 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2719 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2720 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2721 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2722 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2724 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2725 key.type = BTRFS_CHUNK_ITEM_KEY;
2726 key.offset = chunk_offset;
2728 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2731 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2732 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2741 * Chunk allocation falls into two parts. The first part does works
2742 * that make the new allocated chunk useable, but not do any operation
2743 * that modifies the chunk tree. The second part does the works that
2744 * require modifying the chunk tree. This division is important for the
2745 * bootstrap process of adding storage to a seed btrfs.
2747 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *extent_root, u64 type)
2753 struct map_lookup *map;
2754 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2757 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2762 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2763 &stripe_size, chunk_offset, type);
2767 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2768 chunk_size, stripe_size);
2773 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2774 struct btrfs_root *root,
2775 struct btrfs_device *device)
2778 u64 sys_chunk_offset;
2782 u64 sys_stripe_size;
2784 struct map_lookup *map;
2785 struct map_lookup *sys_map;
2786 struct btrfs_fs_info *fs_info = root->fs_info;
2787 struct btrfs_root *extent_root = fs_info->extent_root;
2790 ret = find_next_chunk(fs_info->chunk_root,
2791 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2794 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2795 (fs_info->metadata_alloc_profile &
2796 fs_info->avail_metadata_alloc_bits);
2797 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2799 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2800 &stripe_size, chunk_offset, alloc_profile);
2803 sys_chunk_offset = chunk_offset + chunk_size;
2805 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2806 (fs_info->system_alloc_profile &
2807 fs_info->avail_system_alloc_bits);
2808 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2810 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2811 &sys_chunk_size, &sys_stripe_size,
2812 sys_chunk_offset, alloc_profile);
2815 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2819 * Modifying chunk tree needs allocating new blocks from both
2820 * system block group and metadata block group. So we only can
2821 * do operations require modifying the chunk tree after both
2822 * block groups were created.
2824 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2825 chunk_size, stripe_size);
2828 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2829 sys_chunk_offset, sys_chunk_size,
2835 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2837 struct extent_map *em;
2838 struct map_lookup *map;
2839 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2843 read_lock(&map_tree->map_tree.lock);
2844 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2845 read_unlock(&map_tree->map_tree.lock);
2849 if (btrfs_test_opt(root, DEGRADED)) {
2850 free_extent_map(em);
2854 map = (struct map_lookup *)em->bdev;
2855 for (i = 0; i < map->num_stripes; i++) {
2856 if (!map->stripes[i].dev->writeable) {
2861 free_extent_map(em);
2865 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2867 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2870 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2872 struct extent_map *em;
2875 write_lock(&tree->map_tree.lock);
2876 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2878 remove_extent_mapping(&tree->map_tree, em);
2879 write_unlock(&tree->map_tree.lock);
2884 free_extent_map(em);
2885 /* once for the tree */
2886 free_extent_map(em);
2890 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2892 struct extent_map *em;
2893 struct map_lookup *map;
2894 struct extent_map_tree *em_tree = &map_tree->map_tree;
2897 read_lock(&em_tree->lock);
2898 em = lookup_extent_mapping(em_tree, logical, len);
2899 read_unlock(&em_tree->lock);
2902 BUG_ON(em->start > logical || em->start + em->len < logical);
2903 map = (struct map_lookup *)em->bdev;
2904 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2905 ret = map->num_stripes;
2906 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2907 ret = map->sub_stripes;
2910 free_extent_map(em);
2914 static int find_live_mirror(struct map_lookup *map, int first, int num,
2918 if (map->stripes[optimal].dev->bdev)
2920 for (i = first; i < first + num; i++) {
2921 if (map->stripes[i].dev->bdev)
2924 /* we couldn't find one that doesn't fail. Just return something
2925 * and the io error handling code will clean up eventually
2930 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2931 u64 logical, u64 *length,
2932 struct btrfs_multi_bio **multi_ret,
2933 int mirror_num, struct page *unplug_page)
2935 struct extent_map *em;
2936 struct map_lookup *map;
2937 struct extent_map_tree *em_tree = &map_tree->map_tree;
2941 int stripes_allocated = 8;
2942 int stripes_required = 1;
2947 struct btrfs_multi_bio *multi = NULL;
2949 if (multi_ret && !(rw & REQ_WRITE))
2950 stripes_allocated = 1;
2953 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2958 atomic_set(&multi->error, 0);
2961 read_lock(&em_tree->lock);
2962 em = lookup_extent_mapping(em_tree, logical, *length);
2963 read_unlock(&em_tree->lock);
2965 if (!em && unplug_page) {
2971 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2972 (unsigned long long)logical,
2973 (unsigned long long)*length);
2977 BUG_ON(em->start > logical || em->start + em->len < logical);
2978 map = (struct map_lookup *)em->bdev;
2979 offset = logical - em->start;
2981 if (mirror_num > map->num_stripes)
2984 /* if our multi bio struct is too small, back off and try again */
2985 if (rw & REQ_WRITE) {
2986 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2987 BTRFS_BLOCK_GROUP_DUP)) {
2988 stripes_required = map->num_stripes;
2990 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2991 stripes_required = map->sub_stripes;
2995 if (multi_ret && (rw & REQ_WRITE) &&
2996 stripes_allocated < stripes_required) {
2997 stripes_allocated = map->num_stripes;
2998 free_extent_map(em);
3004 * stripe_nr counts the total number of stripes we have to stride
3005 * to get to this block
3007 do_div(stripe_nr, map->stripe_len);
3009 stripe_offset = stripe_nr * map->stripe_len;
3010 BUG_ON(offset < stripe_offset);
3012 /* stripe_offset is the offset of this block in its stripe*/
3013 stripe_offset = offset - stripe_offset;
3015 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3016 BTRFS_BLOCK_GROUP_RAID10 |
3017 BTRFS_BLOCK_GROUP_DUP)) {
3018 /* we limit the length of each bio to what fits in a stripe */
3019 *length = min_t(u64, em->len - offset,
3020 map->stripe_len - stripe_offset);
3022 *length = em->len - offset;
3025 if (!multi_ret && !unplug_page)
3030 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3031 if (unplug_page || (rw & REQ_WRITE))
3032 num_stripes = map->num_stripes;
3033 else if (mirror_num)
3034 stripe_index = mirror_num - 1;
3036 stripe_index = find_live_mirror(map, 0,
3038 current->pid % map->num_stripes);
3041 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3043 num_stripes = map->num_stripes;
3044 else if (mirror_num)
3045 stripe_index = mirror_num - 1;
3047 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3048 int factor = map->num_stripes / map->sub_stripes;
3050 stripe_index = do_div(stripe_nr, factor);
3051 stripe_index *= map->sub_stripes;
3053 if (unplug_page || (rw & REQ_WRITE))
3054 num_stripes = map->sub_stripes;
3055 else if (mirror_num)
3056 stripe_index += mirror_num - 1;
3058 stripe_index = find_live_mirror(map, stripe_index,
3059 map->sub_stripes, stripe_index +
3060 current->pid % map->sub_stripes);
3064 * after this do_div call, stripe_nr is the number of stripes
3065 * on this device we have to walk to find the data, and
3066 * stripe_index is the number of our device in the stripe array
3068 stripe_index = do_div(stripe_nr, map->num_stripes);
3070 BUG_ON(stripe_index >= map->num_stripes);
3072 for (i = 0; i < num_stripes; i++) {
3074 struct btrfs_device *device;
3075 struct backing_dev_info *bdi;
3077 device = map->stripes[stripe_index].dev;
3079 bdi = blk_get_backing_dev_info(device->bdev);
3080 if (bdi->unplug_io_fn)
3081 bdi->unplug_io_fn(bdi, unplug_page);
3084 multi->stripes[i].physical =
3085 map->stripes[stripe_index].physical +
3086 stripe_offset + stripe_nr * map->stripe_len;
3087 multi->stripes[i].dev = map->stripes[stripe_index].dev;
3093 multi->num_stripes = num_stripes;
3094 multi->max_errors = max_errors;
3097 free_extent_map(em);
3101 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3102 u64 logical, u64 *length,
3103 struct btrfs_multi_bio **multi_ret, int mirror_num)
3105 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3109 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3110 u64 chunk_start, u64 physical, u64 devid,
3111 u64 **logical, int *naddrs, int *stripe_len)
3113 struct extent_map_tree *em_tree = &map_tree->map_tree;
3114 struct extent_map *em;
3115 struct map_lookup *map;
3122 read_lock(&em_tree->lock);
3123 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3124 read_unlock(&em_tree->lock);
3126 BUG_ON(!em || em->start != chunk_start);
3127 map = (struct map_lookup *)em->bdev;
3130 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3131 do_div(length, map->num_stripes / map->sub_stripes);
3132 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3133 do_div(length, map->num_stripes);
3135 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3138 for (i = 0; i < map->num_stripes; i++) {
3139 if (devid && map->stripes[i].dev->devid != devid)
3141 if (map->stripes[i].physical > physical ||
3142 map->stripes[i].physical + length <= physical)
3145 stripe_nr = physical - map->stripes[i].physical;
3146 do_div(stripe_nr, map->stripe_len);
3148 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3149 stripe_nr = stripe_nr * map->num_stripes + i;
3150 do_div(stripe_nr, map->sub_stripes);
3151 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3152 stripe_nr = stripe_nr * map->num_stripes + i;
3154 bytenr = chunk_start + stripe_nr * map->stripe_len;
3155 WARN_ON(nr >= map->num_stripes);
3156 for (j = 0; j < nr; j++) {
3157 if (buf[j] == bytenr)
3161 WARN_ON(nr >= map->num_stripes);
3168 *stripe_len = map->stripe_len;
3170 free_extent_map(em);
3174 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3175 u64 logical, struct page *page)
3177 u64 length = PAGE_CACHE_SIZE;
3178 return __btrfs_map_block(map_tree, READ, logical, &length,
3182 static void end_bio_multi_stripe(struct bio *bio, int err)
3184 struct btrfs_multi_bio *multi = bio->bi_private;
3185 int is_orig_bio = 0;
3188 atomic_inc(&multi->error);
3190 if (bio == multi->orig_bio)
3193 if (atomic_dec_and_test(&multi->stripes_pending)) {
3196 bio = multi->orig_bio;
3198 bio->bi_private = multi->private;
3199 bio->bi_end_io = multi->end_io;
3200 /* only send an error to the higher layers if it is
3201 * beyond the tolerance of the multi-bio
3203 if (atomic_read(&multi->error) > multi->max_errors) {
3207 * this bio is actually up to date, we didn't
3208 * go over the max number of errors
3210 set_bit(BIO_UPTODATE, &bio->bi_flags);
3215 bio_endio(bio, err);
3216 } else if (!is_orig_bio) {
3221 struct async_sched {
3224 struct btrfs_fs_info *info;
3225 struct btrfs_work work;
3229 * see run_scheduled_bios for a description of why bios are collected for
3232 * This will add one bio to the pending list for a device and make sure
3233 * the work struct is scheduled.
3235 static noinline int schedule_bio(struct btrfs_root *root,
3236 struct btrfs_device *device,
3237 int rw, struct bio *bio)
3239 int should_queue = 1;
3240 struct btrfs_pending_bios *pending_bios;
3242 /* don't bother with additional async steps for reads, right now */
3243 if (!(rw & REQ_WRITE)) {
3245 submit_bio(rw, bio);
3251 * nr_async_bios allows us to reliably return congestion to the
3252 * higher layers. Otherwise, the async bio makes it appear we have
3253 * made progress against dirty pages when we've really just put it
3254 * on a queue for later
3256 atomic_inc(&root->fs_info->nr_async_bios);
3257 WARN_ON(bio->bi_next);
3258 bio->bi_next = NULL;
3261 spin_lock(&device->io_lock);
3262 if (bio->bi_rw & REQ_SYNC)
3263 pending_bios = &device->pending_sync_bios;
3265 pending_bios = &device->pending_bios;
3267 if (pending_bios->tail)
3268 pending_bios->tail->bi_next = bio;
3270 pending_bios->tail = bio;
3271 if (!pending_bios->head)
3272 pending_bios->head = bio;
3273 if (device->running_pending)
3276 spin_unlock(&device->io_lock);
3279 btrfs_queue_worker(&root->fs_info->submit_workers,
3284 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3285 int mirror_num, int async_submit)
3287 struct btrfs_mapping_tree *map_tree;
3288 struct btrfs_device *dev;
3289 struct bio *first_bio = bio;
3290 u64 logical = (u64)bio->bi_sector << 9;
3293 struct btrfs_multi_bio *multi = NULL;
3298 length = bio->bi_size;
3299 map_tree = &root->fs_info->mapping_tree;
3300 map_length = length;
3302 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3306 total_devs = multi->num_stripes;
3307 if (map_length < length) {
3308 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3309 "len %llu\n", (unsigned long long)logical,
3310 (unsigned long long)length,
3311 (unsigned long long)map_length);
3314 multi->end_io = first_bio->bi_end_io;
3315 multi->private = first_bio->bi_private;
3316 multi->orig_bio = first_bio;
3317 atomic_set(&multi->stripes_pending, multi->num_stripes);
3319 while (dev_nr < total_devs) {
3320 if (total_devs > 1) {
3321 if (dev_nr < total_devs - 1) {
3322 bio = bio_clone(first_bio, GFP_NOFS);
3327 bio->bi_private = multi;
3328 bio->bi_end_io = end_bio_multi_stripe;
3330 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3331 dev = multi->stripes[dev_nr].dev;
3332 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3333 bio->bi_bdev = dev->bdev;
3335 schedule_bio(root, dev, rw, bio);
3337 submit_bio(rw, bio);
3339 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3340 bio->bi_sector = logical >> 9;
3341 bio_endio(bio, -EIO);
3345 if (total_devs == 1)
3350 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3353 struct btrfs_device *device;
3354 struct btrfs_fs_devices *cur_devices;
3356 cur_devices = root->fs_info->fs_devices;
3357 while (cur_devices) {
3359 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3360 device = __find_device(&cur_devices->devices,
3365 cur_devices = cur_devices->seed;
3370 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3371 u64 devid, u8 *dev_uuid)
3373 struct btrfs_device *device;
3374 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3376 device = kzalloc(sizeof(*device), GFP_NOFS);
3379 list_add(&device->dev_list,
3380 &fs_devices->devices);
3381 device->barriers = 1;
3382 device->dev_root = root->fs_info->dev_root;
3383 device->devid = devid;
3384 device->work.func = pending_bios_fn;
3385 device->fs_devices = fs_devices;
3386 device->missing = 1;
3387 fs_devices->num_devices++;
3388 fs_devices->missing_devices++;
3389 spin_lock_init(&device->io_lock);
3390 INIT_LIST_HEAD(&device->dev_alloc_list);
3391 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3395 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3396 struct extent_buffer *leaf,
3397 struct btrfs_chunk *chunk)
3399 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3400 struct map_lookup *map;
3401 struct extent_map *em;
3405 u8 uuid[BTRFS_UUID_SIZE];
3410 logical = key->offset;
3411 length = btrfs_chunk_length(leaf, chunk);
3413 read_lock(&map_tree->map_tree.lock);
3414 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3415 read_unlock(&map_tree->map_tree.lock);
3417 /* already mapped? */
3418 if (em && em->start <= logical && em->start + em->len > logical) {
3419 free_extent_map(em);
3422 free_extent_map(em);
3425 em = alloc_extent_map(GFP_NOFS);
3428 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3429 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3431 free_extent_map(em);
3435 em->bdev = (struct block_device *)map;
3436 em->start = logical;
3438 em->block_start = 0;
3439 em->block_len = em->len;
3441 map->num_stripes = num_stripes;
3442 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3443 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3444 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3445 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3446 map->type = btrfs_chunk_type(leaf, chunk);
3447 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3448 for (i = 0; i < num_stripes; i++) {
3449 map->stripes[i].physical =
3450 btrfs_stripe_offset_nr(leaf, chunk, i);
3451 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3452 read_extent_buffer(leaf, uuid, (unsigned long)
3453 btrfs_stripe_dev_uuid_nr(chunk, i),
3455 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3457 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3459 free_extent_map(em);
3462 if (!map->stripes[i].dev) {
3463 map->stripes[i].dev =
3464 add_missing_dev(root, devid, uuid);
3465 if (!map->stripes[i].dev) {
3467 free_extent_map(em);
3471 map->stripes[i].dev->in_fs_metadata = 1;
3474 write_lock(&map_tree->map_tree.lock);
3475 ret = add_extent_mapping(&map_tree->map_tree, em);
3476 write_unlock(&map_tree->map_tree.lock);
3478 free_extent_map(em);
3483 static int fill_device_from_item(struct extent_buffer *leaf,
3484 struct btrfs_dev_item *dev_item,
3485 struct btrfs_device *device)
3489 device->devid = btrfs_device_id(leaf, dev_item);
3490 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3491 device->total_bytes = device->disk_total_bytes;
3492 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3493 device->type = btrfs_device_type(leaf, dev_item);
3494 device->io_align = btrfs_device_io_align(leaf, dev_item);
3495 device->io_width = btrfs_device_io_width(leaf, dev_item);
3496 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3498 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3499 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3504 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3506 struct btrfs_fs_devices *fs_devices;
3509 mutex_lock(&uuid_mutex);
3511 fs_devices = root->fs_info->fs_devices->seed;
3512 while (fs_devices) {
3513 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3517 fs_devices = fs_devices->seed;
3520 fs_devices = find_fsid(fsid);
3526 fs_devices = clone_fs_devices(fs_devices);
3527 if (IS_ERR(fs_devices)) {
3528 ret = PTR_ERR(fs_devices);
3532 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3533 root->fs_info->bdev_holder);
3537 if (!fs_devices->seeding) {
3538 __btrfs_close_devices(fs_devices);
3539 free_fs_devices(fs_devices);
3544 fs_devices->seed = root->fs_info->fs_devices->seed;
3545 root->fs_info->fs_devices->seed = fs_devices;
3547 mutex_unlock(&uuid_mutex);
3551 static int read_one_dev(struct btrfs_root *root,
3552 struct extent_buffer *leaf,
3553 struct btrfs_dev_item *dev_item)
3555 struct btrfs_device *device;
3558 u8 fs_uuid[BTRFS_UUID_SIZE];
3559 u8 dev_uuid[BTRFS_UUID_SIZE];
3561 devid = btrfs_device_id(leaf, dev_item);
3562 read_extent_buffer(leaf, dev_uuid,
3563 (unsigned long)btrfs_device_uuid(dev_item),
3565 read_extent_buffer(leaf, fs_uuid,
3566 (unsigned long)btrfs_device_fsid(dev_item),
3569 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3570 ret = open_seed_devices(root, fs_uuid);
3571 if (ret && !btrfs_test_opt(root, DEGRADED))
3575 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3576 if (!device || !device->bdev) {
3577 if (!btrfs_test_opt(root, DEGRADED))
3581 printk(KERN_WARNING "warning devid %llu missing\n",
3582 (unsigned long long)devid);
3583 device = add_missing_dev(root, devid, dev_uuid);
3586 } else if (!device->missing) {
3588 * this happens when a device that was properly setup
3589 * in the device info lists suddenly goes bad.
3590 * device->bdev is NULL, and so we have to set
3591 * device->missing to one here
3593 root->fs_info->fs_devices->missing_devices++;
3594 device->missing = 1;
3598 if (device->fs_devices != root->fs_info->fs_devices) {
3599 BUG_ON(device->writeable);
3600 if (device->generation !=
3601 btrfs_device_generation(leaf, dev_item))
3605 fill_device_from_item(leaf, dev_item, device);
3606 device->dev_root = root->fs_info->dev_root;
3607 device->in_fs_metadata = 1;
3608 if (device->writeable)
3609 device->fs_devices->total_rw_bytes += device->total_bytes;
3614 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3616 struct btrfs_dev_item *dev_item;
3618 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3620 return read_one_dev(root, buf, dev_item);
3623 int btrfs_read_sys_array(struct btrfs_root *root)
3625 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3626 struct extent_buffer *sb;
3627 struct btrfs_disk_key *disk_key;
3628 struct btrfs_chunk *chunk;
3630 unsigned long sb_ptr;
3636 struct btrfs_key key;
3638 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3639 BTRFS_SUPER_INFO_SIZE);
3642 btrfs_set_buffer_uptodate(sb);
3643 btrfs_set_buffer_lockdep_class(sb, 0);
3645 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3646 array_size = btrfs_super_sys_array_size(super_copy);
3648 ptr = super_copy->sys_chunk_array;
3649 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3652 while (cur < array_size) {
3653 disk_key = (struct btrfs_disk_key *)ptr;
3654 btrfs_disk_key_to_cpu(&key, disk_key);
3656 len = sizeof(*disk_key); ptr += len;
3660 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3661 chunk = (struct btrfs_chunk *)sb_ptr;
3662 ret = read_one_chunk(root, &key, sb, chunk);
3665 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3666 len = btrfs_chunk_item_size(num_stripes);
3675 free_extent_buffer(sb);
3679 int btrfs_read_chunk_tree(struct btrfs_root *root)
3681 struct btrfs_path *path;
3682 struct extent_buffer *leaf;
3683 struct btrfs_key key;
3684 struct btrfs_key found_key;
3688 root = root->fs_info->chunk_root;
3690 path = btrfs_alloc_path();
3694 /* first we search for all of the device items, and then we
3695 * read in all of the chunk items. This way we can create chunk
3696 * mappings that reference all of the devices that are afound
3698 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3702 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3706 leaf = path->nodes[0];
3707 slot = path->slots[0];
3708 if (slot >= btrfs_header_nritems(leaf)) {
3709 ret = btrfs_next_leaf(root, path);
3716 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3717 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3718 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3720 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3721 struct btrfs_dev_item *dev_item;
3722 dev_item = btrfs_item_ptr(leaf, slot,
3723 struct btrfs_dev_item);
3724 ret = read_one_dev(root, leaf, dev_item);
3728 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3729 struct btrfs_chunk *chunk;
3730 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3731 ret = read_one_chunk(root, &found_key, leaf, chunk);
3737 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3739 btrfs_release_path(root, path);
3744 btrfs_free_path(path);