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);
1419 if (device->writeable) {
1420 list_add(&device->dev_alloc_list,
1421 &root->fs_info->fs_devices->alloc_list);
1422 root->fs_info->fs_devices->rw_devices++;
1428 * does all the dirty work required for changing file system's UUID.
1430 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1431 struct btrfs_root *root)
1433 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1434 struct btrfs_fs_devices *old_devices;
1435 struct btrfs_fs_devices *seed_devices;
1436 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1437 struct btrfs_device *device;
1440 BUG_ON(!mutex_is_locked(&uuid_mutex));
1441 if (!fs_devices->seeding)
1444 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1448 old_devices = clone_fs_devices(fs_devices);
1449 if (IS_ERR(old_devices)) {
1450 kfree(seed_devices);
1451 return PTR_ERR(old_devices);
1454 list_add(&old_devices->list, &fs_uuids);
1456 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1457 seed_devices->opened = 1;
1458 INIT_LIST_HEAD(&seed_devices->devices);
1459 INIT_LIST_HEAD(&seed_devices->alloc_list);
1460 mutex_init(&seed_devices->device_list_mutex);
1461 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1462 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1463 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1464 device->fs_devices = seed_devices;
1467 fs_devices->seeding = 0;
1468 fs_devices->num_devices = 0;
1469 fs_devices->open_devices = 0;
1470 fs_devices->seed = seed_devices;
1472 generate_random_uuid(fs_devices->fsid);
1473 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1474 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1475 super_flags = btrfs_super_flags(disk_super) &
1476 ~BTRFS_SUPER_FLAG_SEEDING;
1477 btrfs_set_super_flags(disk_super, super_flags);
1483 * strore the expected generation for seed devices in device items.
1485 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1486 struct btrfs_root *root)
1488 struct btrfs_path *path;
1489 struct extent_buffer *leaf;
1490 struct btrfs_dev_item *dev_item;
1491 struct btrfs_device *device;
1492 struct btrfs_key key;
1493 u8 fs_uuid[BTRFS_UUID_SIZE];
1494 u8 dev_uuid[BTRFS_UUID_SIZE];
1498 path = btrfs_alloc_path();
1502 root = root->fs_info->chunk_root;
1503 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1505 key.type = BTRFS_DEV_ITEM_KEY;
1508 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1512 leaf = path->nodes[0];
1514 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1515 ret = btrfs_next_leaf(root, path);
1520 leaf = path->nodes[0];
1521 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1522 btrfs_release_path(root, path);
1526 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1527 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1528 key.type != BTRFS_DEV_ITEM_KEY)
1531 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1532 struct btrfs_dev_item);
1533 devid = btrfs_device_id(leaf, dev_item);
1534 read_extent_buffer(leaf, dev_uuid,
1535 (unsigned long)btrfs_device_uuid(dev_item),
1537 read_extent_buffer(leaf, fs_uuid,
1538 (unsigned long)btrfs_device_fsid(dev_item),
1540 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1543 if (device->fs_devices->seeding) {
1544 btrfs_set_device_generation(leaf, dev_item,
1545 device->generation);
1546 btrfs_mark_buffer_dirty(leaf);
1554 btrfs_free_path(path);
1558 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1560 struct btrfs_trans_handle *trans;
1561 struct btrfs_device *device;
1562 struct block_device *bdev;
1563 struct list_head *devices;
1564 struct super_block *sb = root->fs_info->sb;
1566 int seeding_dev = 0;
1569 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1572 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1574 return PTR_ERR(bdev);
1576 if (root->fs_info->fs_devices->seeding) {
1578 down_write(&sb->s_umount);
1579 mutex_lock(&uuid_mutex);
1582 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1583 mutex_lock(&root->fs_info->volume_mutex);
1585 devices = &root->fs_info->fs_devices->devices;
1587 * we have the volume lock, so we don't need the extra
1588 * device list mutex while reading the list here.
1590 list_for_each_entry(device, devices, dev_list) {
1591 if (device->bdev == bdev) {
1597 device = kzalloc(sizeof(*device), GFP_NOFS);
1599 /* we can safely leave the fs_devices entry around */
1604 device->name = kstrdup(device_path, GFP_NOFS);
1605 if (!device->name) {
1611 ret = find_next_devid(root, &device->devid);
1613 kfree(device->name);
1618 trans = btrfs_start_transaction(root, 0);
1619 if (IS_ERR(trans)) {
1620 kfree(device->name);
1622 ret = PTR_ERR(trans);
1628 device->barriers = 1;
1629 device->writeable = 1;
1630 device->work.func = pending_bios_fn;
1631 generate_random_uuid(device->uuid);
1632 spin_lock_init(&device->io_lock);
1633 device->generation = trans->transid;
1634 device->io_width = root->sectorsize;
1635 device->io_align = root->sectorsize;
1636 device->sector_size = root->sectorsize;
1637 device->total_bytes = i_size_read(bdev->bd_inode);
1638 device->disk_total_bytes = device->total_bytes;
1639 device->dev_root = root->fs_info->dev_root;
1640 device->bdev = bdev;
1641 device->in_fs_metadata = 1;
1642 device->mode = FMODE_EXCL;
1643 set_blocksize(device->bdev, 4096);
1646 sb->s_flags &= ~MS_RDONLY;
1647 ret = btrfs_prepare_sprout(trans, root);
1651 device->fs_devices = root->fs_info->fs_devices;
1654 * we don't want write_supers to jump in here with our device
1657 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1658 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1659 list_add(&device->dev_alloc_list,
1660 &root->fs_info->fs_devices->alloc_list);
1661 root->fs_info->fs_devices->num_devices++;
1662 root->fs_info->fs_devices->open_devices++;
1663 root->fs_info->fs_devices->rw_devices++;
1664 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1666 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1667 root->fs_info->fs_devices->rotating = 1;
1669 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1670 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1671 total_bytes + device->total_bytes);
1673 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1674 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1676 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1679 ret = init_first_rw_device(trans, root, device);
1681 ret = btrfs_finish_sprout(trans, root);
1684 ret = btrfs_add_device(trans, root, device);
1688 * we've got more storage, clear any full flags on the space
1691 btrfs_clear_space_info_full(root->fs_info);
1693 unlock_chunks(root);
1694 btrfs_commit_transaction(trans, root);
1697 mutex_unlock(&uuid_mutex);
1698 up_write(&sb->s_umount);
1700 ret = btrfs_relocate_sys_chunks(root);
1704 mutex_unlock(&root->fs_info->volume_mutex);
1707 close_bdev_exclusive(bdev, 0);
1709 mutex_unlock(&uuid_mutex);
1710 up_write(&sb->s_umount);
1715 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1716 struct btrfs_device *device)
1719 struct btrfs_path *path;
1720 struct btrfs_root *root;
1721 struct btrfs_dev_item *dev_item;
1722 struct extent_buffer *leaf;
1723 struct btrfs_key key;
1725 root = device->dev_root->fs_info->chunk_root;
1727 path = btrfs_alloc_path();
1731 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1732 key.type = BTRFS_DEV_ITEM_KEY;
1733 key.offset = device->devid;
1735 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1744 leaf = path->nodes[0];
1745 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1747 btrfs_set_device_id(leaf, dev_item, device->devid);
1748 btrfs_set_device_type(leaf, dev_item, device->type);
1749 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1750 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1751 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1752 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1753 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1754 btrfs_mark_buffer_dirty(leaf);
1757 btrfs_free_path(path);
1761 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1762 struct btrfs_device *device, u64 new_size)
1764 struct btrfs_super_block *super_copy =
1765 &device->dev_root->fs_info->super_copy;
1766 u64 old_total = btrfs_super_total_bytes(super_copy);
1767 u64 diff = new_size - device->total_bytes;
1769 if (!device->writeable)
1771 if (new_size <= device->total_bytes)
1774 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1775 device->fs_devices->total_rw_bytes += diff;
1777 device->total_bytes = new_size;
1778 device->disk_total_bytes = new_size;
1779 btrfs_clear_space_info_full(device->dev_root->fs_info);
1781 return btrfs_update_device(trans, device);
1784 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1785 struct btrfs_device *device, u64 new_size)
1788 lock_chunks(device->dev_root);
1789 ret = __btrfs_grow_device(trans, device, new_size);
1790 unlock_chunks(device->dev_root);
1794 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1795 struct btrfs_root *root,
1796 u64 chunk_tree, u64 chunk_objectid,
1800 struct btrfs_path *path;
1801 struct btrfs_key key;
1803 root = root->fs_info->chunk_root;
1804 path = btrfs_alloc_path();
1808 key.objectid = chunk_objectid;
1809 key.offset = chunk_offset;
1810 key.type = BTRFS_CHUNK_ITEM_KEY;
1812 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1815 ret = btrfs_del_item(trans, root, path);
1818 btrfs_free_path(path);
1822 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1825 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1826 struct btrfs_disk_key *disk_key;
1827 struct btrfs_chunk *chunk;
1834 struct btrfs_key key;
1836 array_size = btrfs_super_sys_array_size(super_copy);
1838 ptr = super_copy->sys_chunk_array;
1841 while (cur < array_size) {
1842 disk_key = (struct btrfs_disk_key *)ptr;
1843 btrfs_disk_key_to_cpu(&key, disk_key);
1845 len = sizeof(*disk_key);
1847 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1848 chunk = (struct btrfs_chunk *)(ptr + len);
1849 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1850 len += btrfs_chunk_item_size(num_stripes);
1855 if (key.objectid == chunk_objectid &&
1856 key.offset == chunk_offset) {
1857 memmove(ptr, ptr + len, array_size - (cur + len));
1859 btrfs_set_super_sys_array_size(super_copy, array_size);
1868 static int btrfs_relocate_chunk(struct btrfs_root *root,
1869 u64 chunk_tree, u64 chunk_objectid,
1872 struct extent_map_tree *em_tree;
1873 struct btrfs_root *extent_root;
1874 struct btrfs_trans_handle *trans;
1875 struct extent_map *em;
1876 struct map_lookup *map;
1880 root = root->fs_info->chunk_root;
1881 extent_root = root->fs_info->extent_root;
1882 em_tree = &root->fs_info->mapping_tree.map_tree;
1884 ret = btrfs_can_relocate(extent_root, chunk_offset);
1888 /* step one, relocate all the extents inside this chunk */
1889 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1893 trans = btrfs_start_transaction(root, 0);
1894 BUG_ON(IS_ERR(trans));
1899 * step two, delete the device extents and the
1900 * chunk tree entries
1902 read_lock(&em_tree->lock);
1903 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1904 read_unlock(&em_tree->lock);
1906 BUG_ON(em->start > chunk_offset ||
1907 em->start + em->len < chunk_offset);
1908 map = (struct map_lookup *)em->bdev;
1910 for (i = 0; i < map->num_stripes; i++) {
1911 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1912 map->stripes[i].physical);
1915 if (map->stripes[i].dev) {
1916 ret = btrfs_update_device(trans, map->stripes[i].dev);
1920 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1925 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1926 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1930 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1933 write_lock(&em_tree->lock);
1934 remove_extent_mapping(em_tree, em);
1935 write_unlock(&em_tree->lock);
1940 /* once for the tree */
1941 free_extent_map(em);
1943 free_extent_map(em);
1945 unlock_chunks(root);
1946 btrfs_end_transaction(trans, root);
1950 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1952 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1953 struct btrfs_path *path;
1954 struct extent_buffer *leaf;
1955 struct btrfs_chunk *chunk;
1956 struct btrfs_key key;
1957 struct btrfs_key found_key;
1958 u64 chunk_tree = chunk_root->root_key.objectid;
1960 bool retried = false;
1964 path = btrfs_alloc_path();
1969 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1970 key.offset = (u64)-1;
1971 key.type = BTRFS_CHUNK_ITEM_KEY;
1974 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1979 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1986 leaf = path->nodes[0];
1987 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1989 chunk = btrfs_item_ptr(leaf, path->slots[0],
1990 struct btrfs_chunk);
1991 chunk_type = btrfs_chunk_type(leaf, chunk);
1992 btrfs_release_path(chunk_root, path);
1994 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1995 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2004 if (found_key.offset == 0)
2006 key.offset = found_key.offset - 1;
2009 if (failed && !retried) {
2013 } else if (failed && retried) {
2018 btrfs_free_path(path);
2022 static u64 div_factor(u64 num, int factor)
2031 int btrfs_balance(struct btrfs_root *dev_root)
2034 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2035 struct btrfs_device *device;
2038 struct btrfs_path *path;
2039 struct btrfs_key key;
2040 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2041 struct btrfs_trans_handle *trans;
2042 struct btrfs_key found_key;
2044 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2047 if (!capable(CAP_SYS_ADMIN))
2050 mutex_lock(&dev_root->fs_info->volume_mutex);
2051 dev_root = dev_root->fs_info->dev_root;
2053 /* step one make some room on all the devices */
2054 list_for_each_entry(device, devices, dev_list) {
2055 old_size = device->total_bytes;
2056 size_to_free = div_factor(old_size, 1);
2057 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2058 if (!device->writeable ||
2059 device->total_bytes - device->bytes_used > size_to_free)
2062 ret = btrfs_shrink_device(device, old_size - size_to_free);
2067 trans = btrfs_start_transaction(dev_root, 0);
2068 BUG_ON(IS_ERR(trans));
2070 ret = btrfs_grow_device(trans, device, old_size);
2073 btrfs_end_transaction(trans, dev_root);
2076 /* step two, relocate all the chunks */
2077 path = btrfs_alloc_path();
2080 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2081 key.offset = (u64)-1;
2082 key.type = BTRFS_CHUNK_ITEM_KEY;
2085 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2090 * this shouldn't happen, it means the last relocate
2096 ret = btrfs_previous_item(chunk_root, path, 0,
2097 BTRFS_CHUNK_ITEM_KEY);
2101 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2103 if (found_key.objectid != key.objectid)
2106 /* chunk zero is special */
2107 if (found_key.offset == 0)
2110 btrfs_release_path(chunk_root, path);
2111 ret = btrfs_relocate_chunk(chunk_root,
2112 chunk_root->root_key.objectid,
2115 BUG_ON(ret && ret != -ENOSPC);
2116 key.offset = found_key.offset - 1;
2120 btrfs_free_path(path);
2121 mutex_unlock(&dev_root->fs_info->volume_mutex);
2126 * shrinking a device means finding all of the device extents past
2127 * the new size, and then following the back refs to the chunks.
2128 * The chunk relocation code actually frees the device extent
2130 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2132 struct btrfs_trans_handle *trans;
2133 struct btrfs_root *root = device->dev_root;
2134 struct btrfs_dev_extent *dev_extent = NULL;
2135 struct btrfs_path *path;
2143 bool retried = false;
2144 struct extent_buffer *l;
2145 struct btrfs_key key;
2146 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2147 u64 old_total = btrfs_super_total_bytes(super_copy);
2148 u64 old_size = device->total_bytes;
2149 u64 diff = device->total_bytes - new_size;
2151 if (new_size >= device->total_bytes)
2154 path = btrfs_alloc_path();
2162 device->total_bytes = new_size;
2163 if (device->writeable)
2164 device->fs_devices->total_rw_bytes -= diff;
2165 unlock_chunks(root);
2168 key.objectid = device->devid;
2169 key.offset = (u64)-1;
2170 key.type = BTRFS_DEV_EXTENT_KEY;
2173 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2177 ret = btrfs_previous_item(root, path, 0, key.type);
2182 btrfs_release_path(root, path);
2187 slot = path->slots[0];
2188 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2190 if (key.objectid != device->devid) {
2191 btrfs_release_path(root, path);
2195 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2196 length = btrfs_dev_extent_length(l, dev_extent);
2198 if (key.offset + length <= new_size) {
2199 btrfs_release_path(root, path);
2203 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2204 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2205 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2206 btrfs_release_path(root, path);
2208 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2210 if (ret && ret != -ENOSPC)
2217 if (failed && !retried) {
2221 } else if (failed && retried) {
2225 device->total_bytes = old_size;
2226 if (device->writeable)
2227 device->fs_devices->total_rw_bytes += diff;
2228 unlock_chunks(root);
2232 /* Shrinking succeeded, else we would be at "done". */
2233 trans = btrfs_start_transaction(root, 0);
2234 if (IS_ERR(trans)) {
2235 ret = PTR_ERR(trans);
2241 device->disk_total_bytes = new_size;
2242 /* Now btrfs_update_device() will change the on-disk size. */
2243 ret = btrfs_update_device(trans, device);
2245 unlock_chunks(root);
2246 btrfs_end_transaction(trans, root);
2249 WARN_ON(diff > old_total);
2250 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2251 unlock_chunks(root);
2252 btrfs_end_transaction(trans, root);
2254 btrfs_free_path(path);
2258 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2259 struct btrfs_root *root,
2260 struct btrfs_key *key,
2261 struct btrfs_chunk *chunk, int item_size)
2263 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2264 struct btrfs_disk_key disk_key;
2268 array_size = btrfs_super_sys_array_size(super_copy);
2269 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2272 ptr = super_copy->sys_chunk_array + array_size;
2273 btrfs_cpu_key_to_disk(&disk_key, key);
2274 memcpy(ptr, &disk_key, sizeof(disk_key));
2275 ptr += sizeof(disk_key);
2276 memcpy(ptr, chunk, item_size);
2277 item_size += sizeof(disk_key);
2278 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2282 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2283 int num_stripes, int sub_stripes)
2285 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2287 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2288 return calc_size * (num_stripes / sub_stripes);
2290 return calc_size * num_stripes;
2293 /* Used to sort the devices by max_avail(descending sort) */
2294 int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2)
2296 if (((struct btrfs_device_info *)dev_info1)->max_avail >
2297 ((struct btrfs_device_info *)dev_info2)->max_avail)
2299 else if (((struct btrfs_device_info *)dev_info1)->max_avail <
2300 ((struct btrfs_device_info *)dev_info2)->max_avail)
2306 static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type,
2307 int *num_stripes, int *min_stripes,
2314 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2315 *num_stripes = fs_devices->rw_devices;
2318 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2322 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2323 if (fs_devices->rw_devices < 2)
2328 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2329 *num_stripes = fs_devices->rw_devices;
2330 if (*num_stripes < 4)
2332 *num_stripes &= ~(u32)1;
2340 static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices,
2341 u64 proposed_size, u64 type,
2342 int num_stripes, int small_stripe)
2344 int min_stripe_size = 1 * 1024 * 1024;
2345 u64 calc_size = proposed_size;
2346 u64 max_chunk_size = calc_size;
2349 if (type & (BTRFS_BLOCK_GROUP_RAID1 |
2350 BTRFS_BLOCK_GROUP_DUP |
2351 BTRFS_BLOCK_GROUP_RAID10))
2354 if (type & BTRFS_BLOCK_GROUP_DATA) {
2355 max_chunk_size = 10 * calc_size;
2356 min_stripe_size = 64 * 1024 * 1024;
2357 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2358 max_chunk_size = 256 * 1024 * 1024;
2359 min_stripe_size = 32 * 1024 * 1024;
2360 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2361 calc_size = 8 * 1024 * 1024;
2362 max_chunk_size = calc_size * 2;
2363 min_stripe_size = 1 * 1024 * 1024;
2366 /* we don't want a chunk larger than 10% of writeable space */
2367 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2370 if (calc_size * num_stripes > max_chunk_size * ncopies) {
2371 calc_size = max_chunk_size * ncopies;
2372 do_div(calc_size, num_stripes);
2373 do_div(calc_size, BTRFS_STRIPE_LEN);
2374 calc_size *= BTRFS_STRIPE_LEN;
2377 /* we don't want tiny stripes */
2379 calc_size = max_t(u64, min_stripe_size, calc_size);
2382 * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure
2383 * we end up with something bigger than a stripe
2385 calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN);
2387 do_div(calc_size, BTRFS_STRIPE_LEN);
2388 calc_size *= BTRFS_STRIPE_LEN;
2393 static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map,
2396 struct map_lookup *new;
2397 size_t len = map_lookup_size(num_stripes);
2399 BUG_ON(map->num_stripes < num_stripes);
2401 if (map->num_stripes == num_stripes)
2404 new = kmalloc(len, GFP_NOFS);
2406 /* just change map->num_stripes */
2407 map->num_stripes = num_stripes;
2411 memcpy(new, map, len);
2412 new->num_stripes = num_stripes;
2418 * helper to allocate device space from btrfs_device_info, in which we stored
2419 * max free space information of every device. It is used when we can not
2420 * allocate chunks by default size.
2422 * By this helper, we can allocate a new chunk as larger as possible.
2424 static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans,
2425 struct btrfs_fs_devices *fs_devices,
2426 struct btrfs_device_info *devices,
2427 int nr_device, u64 type,
2428 struct map_lookup **map_lookup,
2429 int min_stripes, u64 *stripe_size)
2431 int i, index, sort_again = 0;
2432 int min_devices = min_stripes;
2433 u64 max_avail, min_free;
2434 struct map_lookup *map = *map_lookup;
2437 if (nr_device < min_stripes)
2440 btrfs_descending_sort_devices(devices, nr_device);
2442 max_avail = devices[0].max_avail;
2446 for (i = 0; i < nr_device; i++) {
2448 * if dev_offset = 0, it means the free space of this device
2449 * is less than what we need, and we didn't search max avail
2450 * extent on this device, so do it now.
2452 if (!devices[i].dev_offset) {
2453 ret = find_free_dev_extent(trans, devices[i].dev,
2455 &devices[i].dev_offset,
2456 &devices[i].max_avail);
2457 if (ret != 0 && ret != -ENOSPC)
2463 /* we update the max avail free extent of each devices, sort again */
2465 btrfs_descending_sort_devices(devices, nr_device);
2467 if (type & BTRFS_BLOCK_GROUP_DUP)
2470 if (!devices[min_devices - 1].max_avail)
2473 max_avail = devices[min_devices - 1].max_avail;
2474 if (type & BTRFS_BLOCK_GROUP_DUP)
2475 do_div(max_avail, 2);
2477 max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type,
2479 if (type & BTRFS_BLOCK_GROUP_DUP)
2480 min_free = max_avail * 2;
2482 min_free = max_avail;
2484 if (min_free > devices[min_devices - 1].max_avail)
2487 map = __shrink_map_lookup_stripes(map, min_stripes);
2488 *stripe_size = max_avail;
2491 for (i = 0; i < min_stripes; i++) {
2492 map->stripes[i].dev = devices[index].dev;
2493 map->stripes[i].physical = devices[index].dev_offset;
2494 if (type & BTRFS_BLOCK_GROUP_DUP) {
2496 map->stripes[i].dev = devices[index].dev;
2497 map->stripes[i].physical = devices[index].dev_offset +
2507 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2508 struct btrfs_root *extent_root,
2509 struct map_lookup **map_ret,
2510 u64 *num_bytes, u64 *stripe_size,
2511 u64 start, u64 type)
2513 struct btrfs_fs_info *info = extent_root->fs_info;
2514 struct btrfs_device *device = NULL;
2515 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2516 struct list_head *cur;
2517 struct map_lookup *map;
2518 struct extent_map_tree *em_tree;
2519 struct extent_map *em;
2520 struct btrfs_device_info *devices_info;
2521 struct list_head private_devs;
2522 u64 calc_size = 1024 * 1024 * 1024;
2529 int min_devices; /* the min number of devices we need */
2534 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2535 (type & BTRFS_BLOCK_GROUP_DUP)) {
2537 type &= ~BTRFS_BLOCK_GROUP_DUP;
2539 if (list_empty(&fs_devices->alloc_list))
2542 ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes,
2543 &min_stripes, &sub_stripes);
2547 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2552 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2557 map->num_stripes = num_stripes;
2559 cur = fs_devices->alloc_list.next;
2563 calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type,
2566 if (type & BTRFS_BLOCK_GROUP_DUP) {
2567 min_free = calc_size * 2;
2570 min_free = calc_size;
2571 min_devices = min_stripes;
2574 INIT_LIST_HEAD(&private_devs);
2575 while (index < num_stripes) {
2576 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2577 BUG_ON(!device->writeable);
2578 if (device->total_bytes > device->bytes_used)
2579 avail = device->total_bytes - device->bytes_used;
2584 if (device->in_fs_metadata && avail >= min_free) {
2585 ret = find_free_dev_extent(trans, device, min_free,
2586 &devices_info[i].dev_offset,
2587 &devices_info[i].max_avail);
2589 list_move_tail(&device->dev_alloc_list,
2591 map->stripes[index].dev = device;
2592 map->stripes[index].physical =
2593 devices_info[i].dev_offset;
2595 if (type & BTRFS_BLOCK_GROUP_DUP) {
2596 map->stripes[index].dev = device;
2597 map->stripes[index].physical =
2598 devices_info[i].dev_offset +
2602 } else if (ret != -ENOSPC)
2605 devices_info[i].dev = device;
2607 } else if (device->in_fs_metadata &&
2608 avail >= BTRFS_STRIPE_LEN) {
2609 devices_info[i].dev = device;
2610 devices_info[i].max_avail = avail;
2614 if (cur == &fs_devices->alloc_list)
2618 list_splice(&private_devs, &fs_devices->alloc_list);
2619 if (index < num_stripes) {
2620 if (index >= min_stripes) {
2621 num_stripes = index;
2622 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2623 num_stripes /= sub_stripes;
2624 num_stripes *= sub_stripes;
2627 map = __shrink_map_lookup_stripes(map, num_stripes);
2628 } else if (i >= min_devices) {
2629 ret = __btrfs_alloc_tiny_space(trans, fs_devices,
2630 devices_info, i, type,
2640 map->sector_size = extent_root->sectorsize;
2641 map->stripe_len = BTRFS_STRIPE_LEN;
2642 map->io_align = BTRFS_STRIPE_LEN;
2643 map->io_width = BTRFS_STRIPE_LEN;
2645 map->sub_stripes = sub_stripes;
2648 *stripe_size = calc_size;
2649 *num_bytes = chunk_bytes_by_type(type, calc_size,
2650 map->num_stripes, sub_stripes);
2652 em = alloc_extent_map(GFP_NOFS);
2657 em->bdev = (struct block_device *)map;
2659 em->len = *num_bytes;
2660 em->block_start = 0;
2661 em->block_len = em->len;
2663 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2664 write_lock(&em_tree->lock);
2665 ret = add_extent_mapping(em_tree, em);
2666 write_unlock(&em_tree->lock);
2668 free_extent_map(em);
2670 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2671 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2676 while (index < map->num_stripes) {
2677 device = map->stripes[index].dev;
2678 dev_offset = map->stripes[index].physical;
2680 ret = btrfs_alloc_dev_extent(trans, device,
2681 info->chunk_root->root_key.objectid,
2682 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2683 start, dev_offset, calc_size);
2688 kfree(devices_info);
2693 kfree(devices_info);
2697 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2698 struct btrfs_root *extent_root,
2699 struct map_lookup *map, u64 chunk_offset,
2700 u64 chunk_size, u64 stripe_size)
2703 struct btrfs_key key;
2704 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2705 struct btrfs_device *device;
2706 struct btrfs_chunk *chunk;
2707 struct btrfs_stripe *stripe;
2708 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2712 chunk = kzalloc(item_size, GFP_NOFS);
2717 while (index < map->num_stripes) {
2718 device = map->stripes[index].dev;
2719 device->bytes_used += stripe_size;
2720 ret = btrfs_update_device(trans, device);
2726 stripe = &chunk->stripe;
2727 while (index < map->num_stripes) {
2728 device = map->stripes[index].dev;
2729 dev_offset = map->stripes[index].physical;
2731 btrfs_set_stack_stripe_devid(stripe, device->devid);
2732 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2733 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2738 btrfs_set_stack_chunk_length(chunk, chunk_size);
2739 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2740 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2741 btrfs_set_stack_chunk_type(chunk, map->type);
2742 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2743 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2744 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2745 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2746 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2748 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2749 key.type = BTRFS_CHUNK_ITEM_KEY;
2750 key.offset = chunk_offset;
2752 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2755 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2756 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2765 * Chunk allocation falls into two parts. The first part does works
2766 * that make the new allocated chunk useable, but not do any operation
2767 * that modifies the chunk tree. The second part does the works that
2768 * require modifying the chunk tree. This division is important for the
2769 * bootstrap process of adding storage to a seed btrfs.
2771 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2772 struct btrfs_root *extent_root, u64 type)
2777 struct map_lookup *map;
2778 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2781 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2786 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2787 &stripe_size, chunk_offset, type);
2791 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2792 chunk_size, stripe_size);
2797 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2798 struct btrfs_root *root,
2799 struct btrfs_device *device)
2802 u64 sys_chunk_offset;
2806 u64 sys_stripe_size;
2808 struct map_lookup *map;
2809 struct map_lookup *sys_map;
2810 struct btrfs_fs_info *fs_info = root->fs_info;
2811 struct btrfs_root *extent_root = fs_info->extent_root;
2814 ret = find_next_chunk(fs_info->chunk_root,
2815 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2818 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2819 (fs_info->metadata_alloc_profile &
2820 fs_info->avail_metadata_alloc_bits);
2821 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2823 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2824 &stripe_size, chunk_offset, alloc_profile);
2827 sys_chunk_offset = chunk_offset + chunk_size;
2829 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2830 (fs_info->system_alloc_profile &
2831 fs_info->avail_system_alloc_bits);
2832 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2834 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2835 &sys_chunk_size, &sys_stripe_size,
2836 sys_chunk_offset, alloc_profile);
2839 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2843 * Modifying chunk tree needs allocating new blocks from both
2844 * system block group and metadata block group. So we only can
2845 * do operations require modifying the chunk tree after both
2846 * block groups were created.
2848 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2849 chunk_size, stripe_size);
2852 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2853 sys_chunk_offset, sys_chunk_size,
2859 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2861 struct extent_map *em;
2862 struct map_lookup *map;
2863 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2867 read_lock(&map_tree->map_tree.lock);
2868 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2869 read_unlock(&map_tree->map_tree.lock);
2873 if (btrfs_test_opt(root, DEGRADED)) {
2874 free_extent_map(em);
2878 map = (struct map_lookup *)em->bdev;
2879 for (i = 0; i < map->num_stripes; i++) {
2880 if (!map->stripes[i].dev->writeable) {
2885 free_extent_map(em);
2889 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2891 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2894 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2896 struct extent_map *em;
2899 write_lock(&tree->map_tree.lock);
2900 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2902 remove_extent_mapping(&tree->map_tree, em);
2903 write_unlock(&tree->map_tree.lock);
2908 free_extent_map(em);
2909 /* once for the tree */
2910 free_extent_map(em);
2914 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2916 struct extent_map *em;
2917 struct map_lookup *map;
2918 struct extent_map_tree *em_tree = &map_tree->map_tree;
2921 read_lock(&em_tree->lock);
2922 em = lookup_extent_mapping(em_tree, logical, len);
2923 read_unlock(&em_tree->lock);
2926 BUG_ON(em->start > logical || em->start + em->len < logical);
2927 map = (struct map_lookup *)em->bdev;
2928 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2929 ret = map->num_stripes;
2930 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2931 ret = map->sub_stripes;
2934 free_extent_map(em);
2938 static int find_live_mirror(struct map_lookup *map, int first, int num,
2942 if (map->stripes[optimal].dev->bdev)
2944 for (i = first; i < first + num; i++) {
2945 if (map->stripes[i].dev->bdev)
2948 /* we couldn't find one that doesn't fail. Just return something
2949 * and the io error handling code will clean up eventually
2954 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2955 u64 logical, u64 *length,
2956 struct btrfs_multi_bio **multi_ret,
2957 int mirror_num, struct page *unplug_page)
2959 struct extent_map *em;
2960 struct map_lookup *map;
2961 struct extent_map_tree *em_tree = &map_tree->map_tree;
2965 int stripes_allocated = 8;
2966 int stripes_required = 1;
2971 struct btrfs_multi_bio *multi = NULL;
2973 if (multi_ret && !(rw & REQ_WRITE))
2974 stripes_allocated = 1;
2977 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2982 atomic_set(&multi->error, 0);
2985 read_lock(&em_tree->lock);
2986 em = lookup_extent_mapping(em_tree, logical, *length);
2987 read_unlock(&em_tree->lock);
2989 if (!em && unplug_page) {
2995 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2996 (unsigned long long)logical,
2997 (unsigned long long)*length);
3001 BUG_ON(em->start > logical || em->start + em->len < logical);
3002 map = (struct map_lookup *)em->bdev;
3003 offset = logical - em->start;
3005 if (mirror_num > map->num_stripes)
3008 /* if our multi bio struct is too small, back off and try again */
3009 if (rw & REQ_WRITE) {
3010 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
3011 BTRFS_BLOCK_GROUP_DUP)) {
3012 stripes_required = map->num_stripes;
3014 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3015 stripes_required = map->sub_stripes;
3019 if (multi_ret && (rw & REQ_WRITE) &&
3020 stripes_allocated < stripes_required) {
3021 stripes_allocated = map->num_stripes;
3022 free_extent_map(em);
3028 * stripe_nr counts the total number of stripes we have to stride
3029 * to get to this block
3031 do_div(stripe_nr, map->stripe_len);
3033 stripe_offset = stripe_nr * map->stripe_len;
3034 BUG_ON(offset < stripe_offset);
3036 /* stripe_offset is the offset of this block in its stripe*/
3037 stripe_offset = offset - stripe_offset;
3039 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3040 BTRFS_BLOCK_GROUP_RAID10 |
3041 BTRFS_BLOCK_GROUP_DUP)) {
3042 /* we limit the length of each bio to what fits in a stripe */
3043 *length = min_t(u64, em->len - offset,
3044 map->stripe_len - stripe_offset);
3046 *length = em->len - offset;
3049 if (!multi_ret && !unplug_page)
3054 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
3055 if (unplug_page || (rw & REQ_WRITE))
3056 num_stripes = map->num_stripes;
3057 else if (mirror_num)
3058 stripe_index = mirror_num - 1;
3060 stripe_index = find_live_mirror(map, 0,
3062 current->pid % map->num_stripes);
3065 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
3067 num_stripes = map->num_stripes;
3068 else if (mirror_num)
3069 stripe_index = mirror_num - 1;
3071 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3072 int factor = map->num_stripes / map->sub_stripes;
3074 stripe_index = do_div(stripe_nr, factor);
3075 stripe_index *= map->sub_stripes;
3077 if (unplug_page || (rw & REQ_WRITE))
3078 num_stripes = map->sub_stripes;
3079 else if (mirror_num)
3080 stripe_index += mirror_num - 1;
3082 stripe_index = find_live_mirror(map, stripe_index,
3083 map->sub_stripes, stripe_index +
3084 current->pid % map->sub_stripes);
3088 * after this do_div call, stripe_nr is the number of stripes
3089 * on this device we have to walk to find the data, and
3090 * stripe_index is the number of our device in the stripe array
3092 stripe_index = do_div(stripe_nr, map->num_stripes);
3094 BUG_ON(stripe_index >= map->num_stripes);
3096 for (i = 0; i < num_stripes; i++) {
3098 struct btrfs_device *device;
3099 struct backing_dev_info *bdi;
3101 device = map->stripes[stripe_index].dev;
3103 bdi = blk_get_backing_dev_info(device->bdev);
3104 if (bdi->unplug_io_fn)
3105 bdi->unplug_io_fn(bdi, unplug_page);
3108 multi->stripes[i].physical =
3109 map->stripes[stripe_index].physical +
3110 stripe_offset + stripe_nr * map->stripe_len;
3111 multi->stripes[i].dev = map->stripes[stripe_index].dev;
3117 multi->num_stripes = num_stripes;
3118 multi->max_errors = max_errors;
3121 free_extent_map(em);
3125 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3126 u64 logical, u64 *length,
3127 struct btrfs_multi_bio **multi_ret, int mirror_num)
3129 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3133 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3134 u64 chunk_start, u64 physical, u64 devid,
3135 u64 **logical, int *naddrs, int *stripe_len)
3137 struct extent_map_tree *em_tree = &map_tree->map_tree;
3138 struct extent_map *em;
3139 struct map_lookup *map;
3146 read_lock(&em_tree->lock);
3147 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3148 read_unlock(&em_tree->lock);
3150 BUG_ON(!em || em->start != chunk_start);
3151 map = (struct map_lookup *)em->bdev;
3154 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3155 do_div(length, map->num_stripes / map->sub_stripes);
3156 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3157 do_div(length, map->num_stripes);
3159 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3162 for (i = 0; i < map->num_stripes; i++) {
3163 if (devid && map->stripes[i].dev->devid != devid)
3165 if (map->stripes[i].physical > physical ||
3166 map->stripes[i].physical + length <= physical)
3169 stripe_nr = physical - map->stripes[i].physical;
3170 do_div(stripe_nr, map->stripe_len);
3172 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3173 stripe_nr = stripe_nr * map->num_stripes + i;
3174 do_div(stripe_nr, map->sub_stripes);
3175 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3176 stripe_nr = stripe_nr * map->num_stripes + i;
3178 bytenr = chunk_start + stripe_nr * map->stripe_len;
3179 WARN_ON(nr >= map->num_stripes);
3180 for (j = 0; j < nr; j++) {
3181 if (buf[j] == bytenr)
3185 WARN_ON(nr >= map->num_stripes);
3192 *stripe_len = map->stripe_len;
3194 free_extent_map(em);
3198 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
3199 u64 logical, struct page *page)
3201 u64 length = PAGE_CACHE_SIZE;
3202 return __btrfs_map_block(map_tree, READ, logical, &length,
3206 static void end_bio_multi_stripe(struct bio *bio, int err)
3208 struct btrfs_multi_bio *multi = bio->bi_private;
3209 int is_orig_bio = 0;
3212 atomic_inc(&multi->error);
3214 if (bio == multi->orig_bio)
3217 if (atomic_dec_and_test(&multi->stripes_pending)) {
3220 bio = multi->orig_bio;
3222 bio->bi_private = multi->private;
3223 bio->bi_end_io = multi->end_io;
3224 /* only send an error to the higher layers if it is
3225 * beyond the tolerance of the multi-bio
3227 if (atomic_read(&multi->error) > multi->max_errors) {
3231 * this bio is actually up to date, we didn't
3232 * go over the max number of errors
3234 set_bit(BIO_UPTODATE, &bio->bi_flags);
3239 bio_endio(bio, err);
3240 } else if (!is_orig_bio) {
3245 struct async_sched {
3248 struct btrfs_fs_info *info;
3249 struct btrfs_work work;
3253 * see run_scheduled_bios for a description of why bios are collected for
3256 * This will add one bio to the pending list for a device and make sure
3257 * the work struct is scheduled.
3259 static noinline int schedule_bio(struct btrfs_root *root,
3260 struct btrfs_device *device,
3261 int rw, struct bio *bio)
3263 int should_queue = 1;
3264 struct btrfs_pending_bios *pending_bios;
3266 /* don't bother with additional async steps for reads, right now */
3267 if (!(rw & REQ_WRITE)) {
3269 submit_bio(rw, bio);
3275 * nr_async_bios allows us to reliably return congestion to the
3276 * higher layers. Otherwise, the async bio makes it appear we have
3277 * made progress against dirty pages when we've really just put it
3278 * on a queue for later
3280 atomic_inc(&root->fs_info->nr_async_bios);
3281 WARN_ON(bio->bi_next);
3282 bio->bi_next = NULL;
3285 spin_lock(&device->io_lock);
3286 if (bio->bi_rw & REQ_SYNC)
3287 pending_bios = &device->pending_sync_bios;
3289 pending_bios = &device->pending_bios;
3291 if (pending_bios->tail)
3292 pending_bios->tail->bi_next = bio;
3294 pending_bios->tail = bio;
3295 if (!pending_bios->head)
3296 pending_bios->head = bio;
3297 if (device->running_pending)
3300 spin_unlock(&device->io_lock);
3303 btrfs_queue_worker(&root->fs_info->submit_workers,
3308 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3309 int mirror_num, int async_submit)
3311 struct btrfs_mapping_tree *map_tree;
3312 struct btrfs_device *dev;
3313 struct bio *first_bio = bio;
3314 u64 logical = (u64)bio->bi_sector << 9;
3317 struct btrfs_multi_bio *multi = NULL;
3322 length = bio->bi_size;
3323 map_tree = &root->fs_info->mapping_tree;
3324 map_length = length;
3326 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3330 total_devs = multi->num_stripes;
3331 if (map_length < length) {
3332 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3333 "len %llu\n", (unsigned long long)logical,
3334 (unsigned long long)length,
3335 (unsigned long long)map_length);
3338 multi->end_io = first_bio->bi_end_io;
3339 multi->private = first_bio->bi_private;
3340 multi->orig_bio = first_bio;
3341 atomic_set(&multi->stripes_pending, multi->num_stripes);
3343 while (dev_nr < total_devs) {
3344 if (total_devs > 1) {
3345 if (dev_nr < total_devs - 1) {
3346 bio = bio_clone(first_bio, GFP_NOFS);
3351 bio->bi_private = multi;
3352 bio->bi_end_io = end_bio_multi_stripe;
3354 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3355 dev = multi->stripes[dev_nr].dev;
3356 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3357 bio->bi_bdev = dev->bdev;
3359 schedule_bio(root, dev, rw, bio);
3361 submit_bio(rw, bio);
3363 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3364 bio->bi_sector = logical >> 9;
3365 bio_endio(bio, -EIO);
3369 if (total_devs == 1)
3374 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3377 struct btrfs_device *device;
3378 struct btrfs_fs_devices *cur_devices;
3380 cur_devices = root->fs_info->fs_devices;
3381 while (cur_devices) {
3383 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3384 device = __find_device(&cur_devices->devices,
3389 cur_devices = cur_devices->seed;
3394 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3395 u64 devid, u8 *dev_uuid)
3397 struct btrfs_device *device;
3398 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3400 device = kzalloc(sizeof(*device), GFP_NOFS);
3403 list_add(&device->dev_list,
3404 &fs_devices->devices);
3405 device->barriers = 1;
3406 device->dev_root = root->fs_info->dev_root;
3407 device->devid = devid;
3408 device->work.func = pending_bios_fn;
3409 device->fs_devices = fs_devices;
3410 device->missing = 1;
3411 fs_devices->num_devices++;
3412 fs_devices->missing_devices++;
3413 spin_lock_init(&device->io_lock);
3414 INIT_LIST_HEAD(&device->dev_alloc_list);
3415 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3419 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3420 struct extent_buffer *leaf,
3421 struct btrfs_chunk *chunk)
3423 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3424 struct map_lookup *map;
3425 struct extent_map *em;
3429 u8 uuid[BTRFS_UUID_SIZE];
3434 logical = key->offset;
3435 length = btrfs_chunk_length(leaf, chunk);
3437 read_lock(&map_tree->map_tree.lock);
3438 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3439 read_unlock(&map_tree->map_tree.lock);
3441 /* already mapped? */
3442 if (em && em->start <= logical && em->start + em->len > logical) {
3443 free_extent_map(em);
3446 free_extent_map(em);
3449 em = alloc_extent_map(GFP_NOFS);
3452 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3453 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3455 free_extent_map(em);
3459 em->bdev = (struct block_device *)map;
3460 em->start = logical;
3462 em->block_start = 0;
3463 em->block_len = em->len;
3465 map->num_stripes = num_stripes;
3466 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3467 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3468 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3469 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3470 map->type = btrfs_chunk_type(leaf, chunk);
3471 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3472 for (i = 0; i < num_stripes; i++) {
3473 map->stripes[i].physical =
3474 btrfs_stripe_offset_nr(leaf, chunk, i);
3475 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3476 read_extent_buffer(leaf, uuid, (unsigned long)
3477 btrfs_stripe_dev_uuid_nr(chunk, i),
3479 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3481 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3483 free_extent_map(em);
3486 if (!map->stripes[i].dev) {
3487 map->stripes[i].dev =
3488 add_missing_dev(root, devid, uuid);
3489 if (!map->stripes[i].dev) {
3491 free_extent_map(em);
3495 map->stripes[i].dev->in_fs_metadata = 1;
3498 write_lock(&map_tree->map_tree.lock);
3499 ret = add_extent_mapping(&map_tree->map_tree, em);
3500 write_unlock(&map_tree->map_tree.lock);
3502 free_extent_map(em);
3507 static int fill_device_from_item(struct extent_buffer *leaf,
3508 struct btrfs_dev_item *dev_item,
3509 struct btrfs_device *device)
3513 device->devid = btrfs_device_id(leaf, dev_item);
3514 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3515 device->total_bytes = device->disk_total_bytes;
3516 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3517 device->type = btrfs_device_type(leaf, dev_item);
3518 device->io_align = btrfs_device_io_align(leaf, dev_item);
3519 device->io_width = btrfs_device_io_width(leaf, dev_item);
3520 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3522 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3523 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3528 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3530 struct btrfs_fs_devices *fs_devices;
3533 mutex_lock(&uuid_mutex);
3535 fs_devices = root->fs_info->fs_devices->seed;
3536 while (fs_devices) {
3537 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3541 fs_devices = fs_devices->seed;
3544 fs_devices = find_fsid(fsid);
3550 fs_devices = clone_fs_devices(fs_devices);
3551 if (IS_ERR(fs_devices)) {
3552 ret = PTR_ERR(fs_devices);
3556 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3557 root->fs_info->bdev_holder);
3561 if (!fs_devices->seeding) {
3562 __btrfs_close_devices(fs_devices);
3563 free_fs_devices(fs_devices);
3568 fs_devices->seed = root->fs_info->fs_devices->seed;
3569 root->fs_info->fs_devices->seed = fs_devices;
3571 mutex_unlock(&uuid_mutex);
3575 static int read_one_dev(struct btrfs_root *root,
3576 struct extent_buffer *leaf,
3577 struct btrfs_dev_item *dev_item)
3579 struct btrfs_device *device;
3582 u8 fs_uuid[BTRFS_UUID_SIZE];
3583 u8 dev_uuid[BTRFS_UUID_SIZE];
3585 devid = btrfs_device_id(leaf, dev_item);
3586 read_extent_buffer(leaf, dev_uuid,
3587 (unsigned long)btrfs_device_uuid(dev_item),
3589 read_extent_buffer(leaf, fs_uuid,
3590 (unsigned long)btrfs_device_fsid(dev_item),
3593 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3594 ret = open_seed_devices(root, fs_uuid);
3595 if (ret && !btrfs_test_opt(root, DEGRADED))
3599 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3600 if (!device || !device->bdev) {
3601 if (!btrfs_test_opt(root, DEGRADED))
3605 printk(KERN_WARNING "warning devid %llu missing\n",
3606 (unsigned long long)devid);
3607 device = add_missing_dev(root, devid, dev_uuid);
3610 } else if (!device->missing) {
3612 * this happens when a device that was properly setup
3613 * in the device info lists suddenly goes bad.
3614 * device->bdev is NULL, and so we have to set
3615 * device->missing to one here
3617 root->fs_info->fs_devices->missing_devices++;
3618 device->missing = 1;
3622 if (device->fs_devices != root->fs_info->fs_devices) {
3623 BUG_ON(device->writeable);
3624 if (device->generation !=
3625 btrfs_device_generation(leaf, dev_item))
3629 fill_device_from_item(leaf, dev_item, device);
3630 device->dev_root = root->fs_info->dev_root;
3631 device->in_fs_metadata = 1;
3632 if (device->writeable)
3633 device->fs_devices->total_rw_bytes += device->total_bytes;
3638 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3640 struct btrfs_dev_item *dev_item;
3642 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3644 return read_one_dev(root, buf, dev_item);
3647 int btrfs_read_sys_array(struct btrfs_root *root)
3649 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3650 struct extent_buffer *sb;
3651 struct btrfs_disk_key *disk_key;
3652 struct btrfs_chunk *chunk;
3654 unsigned long sb_ptr;
3660 struct btrfs_key key;
3662 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3663 BTRFS_SUPER_INFO_SIZE);
3666 btrfs_set_buffer_uptodate(sb);
3667 btrfs_set_buffer_lockdep_class(sb, 0);
3669 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3670 array_size = btrfs_super_sys_array_size(super_copy);
3672 ptr = super_copy->sys_chunk_array;
3673 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3676 while (cur < array_size) {
3677 disk_key = (struct btrfs_disk_key *)ptr;
3678 btrfs_disk_key_to_cpu(&key, disk_key);
3680 len = sizeof(*disk_key); ptr += len;
3684 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3685 chunk = (struct btrfs_chunk *)sb_ptr;
3686 ret = read_one_chunk(root, &key, sb, chunk);
3689 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3690 len = btrfs_chunk_item_size(num_stripes);
3699 free_extent_buffer(sb);
3703 int btrfs_read_chunk_tree(struct btrfs_root *root)
3705 struct btrfs_path *path;
3706 struct extent_buffer *leaf;
3707 struct btrfs_key key;
3708 struct btrfs_key found_key;
3712 root = root->fs_info->chunk_root;
3714 path = btrfs_alloc_path();
3718 /* first we search for all of the device items, and then we
3719 * read in all of the chunk items. This way we can create chunk
3720 * mappings that reference all of the devices that are afound
3722 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3726 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3730 leaf = path->nodes[0];
3731 slot = path->slots[0];
3732 if (slot >= btrfs_header_nritems(leaf)) {
3733 ret = btrfs_next_leaf(root, path);
3740 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3741 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3742 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3744 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3745 struct btrfs_dev_item *dev_item;
3746 dev_item = btrfs_item_ptr(leaf, slot,
3747 struct btrfs_dev_item);
3748 ret = read_one_dev(root, leaf, dev_item);
3752 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3753 struct btrfs_chunk *chunk;
3754 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3755 ret = read_one_chunk(root, &found_key, leaf, chunk);
3761 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3763 btrfs_release_path(root, path);
3768 btrfs_free_path(path);