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"
36 static int init_first_rw_device(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct btrfs_device *device);
39 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
41 static DEFINE_MUTEX(uuid_mutex);
42 static LIST_HEAD(fs_uuids);
44 static void lock_chunks(struct btrfs_root *root)
46 mutex_lock(&root->fs_info->chunk_mutex);
49 static void unlock_chunks(struct btrfs_root *root)
51 mutex_unlock(&root->fs_info->chunk_mutex);
54 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
56 struct btrfs_device *device;
57 WARN_ON(fs_devices->opened);
58 while (!list_empty(&fs_devices->devices)) {
59 device = list_entry(fs_devices->devices.next,
60 struct btrfs_device, dev_list);
61 list_del(&device->dev_list);
68 int btrfs_cleanup_fs_uuids(void)
70 struct btrfs_fs_devices *fs_devices;
72 while (!list_empty(&fs_uuids)) {
73 fs_devices = list_entry(fs_uuids.next,
74 struct btrfs_fs_devices, list);
75 list_del(&fs_devices->list);
76 free_fs_devices(fs_devices);
81 static noinline struct btrfs_device *__find_device(struct list_head *head,
84 struct btrfs_device *dev;
86 list_for_each_entry(dev, head, dev_list) {
87 if (dev->devid == devid &&
88 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
95 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
97 struct btrfs_fs_devices *fs_devices;
99 list_for_each_entry(fs_devices, &fs_uuids, list) {
100 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
106 static void requeue_list(struct btrfs_pending_bios *pending_bios,
107 struct bio *head, struct bio *tail)
110 struct bio *old_head;
112 old_head = pending_bios->head;
113 pending_bios->head = head;
114 if (pending_bios->tail)
115 tail->bi_next = old_head;
117 pending_bios->tail = tail;
121 * we try to collect pending bios for a device so we don't get a large
122 * number of procs sending bios down to the same device. This greatly
123 * improves the schedulers ability to collect and merge the bios.
125 * But, it also turns into a long list of bios to process and that is sure
126 * to eventually make the worker thread block. The solution here is to
127 * make some progress and then put this work struct back at the end of
128 * the list if the block device is congested. This way, multiple devices
129 * can make progress from a single worker thread.
131 static noinline int run_scheduled_bios(struct btrfs_device *device)
134 struct backing_dev_info *bdi;
135 struct btrfs_fs_info *fs_info;
136 struct btrfs_pending_bios *pending_bios;
140 unsigned long num_run;
141 unsigned long batch_run = 0;
143 unsigned long last_waited = 0;
145 int sync_pending = 0;
146 struct blk_plug plug;
149 * this function runs all the bios we've collected for
150 * a particular device. We don't want to wander off to
151 * another device without first sending all of these down.
152 * So, setup a plug here and finish it off before we return
154 blk_start_plug(&plug);
156 bdi = blk_get_backing_dev_info(device->bdev);
157 fs_info = device->dev_root->fs_info;
158 limit = btrfs_async_submit_limit(fs_info);
159 limit = limit * 2 / 3;
162 spin_lock(&device->io_lock);
167 /* take all the bios off the list at once and process them
168 * later on (without the lock held). But, remember the
169 * tail and other pointers so the bios can be properly reinserted
170 * into the list if we hit congestion
172 if (!force_reg && device->pending_sync_bios.head) {
173 pending_bios = &device->pending_sync_bios;
176 pending_bios = &device->pending_bios;
180 pending = pending_bios->head;
181 tail = pending_bios->tail;
182 WARN_ON(pending && !tail);
185 * if pending was null this time around, no bios need processing
186 * at all and we can stop. Otherwise it'll loop back up again
187 * and do an additional check so no bios are missed.
189 * device->running_pending is used to synchronize with the
192 if (device->pending_sync_bios.head == NULL &&
193 device->pending_bios.head == NULL) {
195 device->running_pending = 0;
198 device->running_pending = 1;
201 pending_bios->head = NULL;
202 pending_bios->tail = NULL;
204 spin_unlock(&device->io_lock);
209 /* we want to work on both lists, but do more bios on the
210 * sync list than the regular list
213 pending_bios != &device->pending_sync_bios &&
214 device->pending_sync_bios.head) ||
215 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
216 device->pending_bios.head)) {
217 spin_lock(&device->io_lock);
218 requeue_list(pending_bios, pending, tail);
223 pending = pending->bi_next;
225 atomic_dec(&fs_info->nr_async_bios);
227 if (atomic_read(&fs_info->nr_async_bios) < limit &&
228 waitqueue_active(&fs_info->async_submit_wait))
229 wake_up(&fs_info->async_submit_wait);
231 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
234 * if we're doing the sync list, record that our
235 * plug has some sync requests on it
237 * If we're doing the regular list and there are
238 * sync requests sitting around, unplug before
241 if (pending_bios == &device->pending_sync_bios) {
243 } else if (sync_pending) {
244 blk_finish_plug(&plug);
245 blk_start_plug(&plug);
249 submit_bio(cur->bi_rw, cur);
256 * we made progress, there is more work to do and the bdi
257 * is now congested. Back off and let other work structs
260 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
261 fs_info->fs_devices->open_devices > 1) {
262 struct io_context *ioc;
264 ioc = current->io_context;
267 * the main goal here is that we don't want to
268 * block if we're going to be able to submit
269 * more requests without blocking.
271 * This code does two great things, it pokes into
272 * the elevator code from a filesystem _and_
273 * it makes assumptions about how batching works.
275 if (ioc && ioc->nr_batch_requests > 0 &&
276 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
278 ioc->last_waited == last_waited)) {
280 * we want to go through our batch of
281 * requests and stop. So, we copy out
282 * the ioc->last_waited time and test
283 * against it before looping
285 last_waited = ioc->last_waited;
290 spin_lock(&device->io_lock);
291 requeue_list(pending_bios, pending, tail);
292 device->running_pending = 1;
294 spin_unlock(&device->io_lock);
295 btrfs_requeue_work(&device->work);
304 spin_lock(&device->io_lock);
305 if (device->pending_bios.head || device->pending_sync_bios.head)
307 spin_unlock(&device->io_lock);
310 blk_finish_plug(&plug);
314 static void pending_bios_fn(struct btrfs_work *work)
316 struct btrfs_device *device;
318 device = container_of(work, struct btrfs_device, work);
319 run_scheduled_bios(device);
322 static noinline int device_list_add(const char *path,
323 struct btrfs_super_block *disk_super,
324 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
326 struct btrfs_device *device;
327 struct btrfs_fs_devices *fs_devices;
328 u64 found_transid = btrfs_super_generation(disk_super);
331 fs_devices = find_fsid(disk_super->fsid);
333 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
336 INIT_LIST_HEAD(&fs_devices->devices);
337 INIT_LIST_HEAD(&fs_devices->alloc_list);
338 list_add(&fs_devices->list, &fs_uuids);
339 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
340 fs_devices->latest_devid = devid;
341 fs_devices->latest_trans = found_transid;
342 mutex_init(&fs_devices->device_list_mutex);
345 device = __find_device(&fs_devices->devices, devid,
346 disk_super->dev_item.uuid);
349 if (fs_devices->opened)
352 device = kzalloc(sizeof(*device), GFP_NOFS);
354 /* we can safely leave the fs_devices entry around */
357 device->devid = devid;
358 device->work.func = pending_bios_fn;
359 memcpy(device->uuid, disk_super->dev_item.uuid,
361 spin_lock_init(&device->io_lock);
362 device->name = kstrdup(path, GFP_NOFS);
367 INIT_LIST_HEAD(&device->dev_alloc_list);
369 mutex_lock(&fs_devices->device_list_mutex);
370 list_add_rcu(&device->dev_list, &fs_devices->devices);
371 mutex_unlock(&fs_devices->device_list_mutex);
373 device->fs_devices = fs_devices;
374 fs_devices->num_devices++;
375 } else if (!device->name || strcmp(device->name, path)) {
376 name = kstrdup(path, GFP_NOFS);
381 if (device->missing) {
382 fs_devices->missing_devices--;
387 if (found_transid > fs_devices->latest_trans) {
388 fs_devices->latest_devid = devid;
389 fs_devices->latest_trans = found_transid;
391 *fs_devices_ret = fs_devices;
395 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
397 struct btrfs_fs_devices *fs_devices;
398 struct btrfs_device *device;
399 struct btrfs_device *orig_dev;
401 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
403 return ERR_PTR(-ENOMEM);
405 INIT_LIST_HEAD(&fs_devices->devices);
406 INIT_LIST_HEAD(&fs_devices->alloc_list);
407 INIT_LIST_HEAD(&fs_devices->list);
408 mutex_init(&fs_devices->device_list_mutex);
409 fs_devices->latest_devid = orig->latest_devid;
410 fs_devices->latest_trans = orig->latest_trans;
411 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
413 /* We have held the volume lock, it is safe to get the devices. */
414 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
415 device = kzalloc(sizeof(*device), GFP_NOFS);
419 device->name = kstrdup(orig_dev->name, GFP_NOFS);
425 device->devid = orig_dev->devid;
426 device->work.func = pending_bios_fn;
427 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
428 spin_lock_init(&device->io_lock);
429 INIT_LIST_HEAD(&device->dev_list);
430 INIT_LIST_HEAD(&device->dev_alloc_list);
432 list_add(&device->dev_list, &fs_devices->devices);
433 device->fs_devices = fs_devices;
434 fs_devices->num_devices++;
438 free_fs_devices(fs_devices);
439 return ERR_PTR(-ENOMEM);
442 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
444 struct btrfs_device *device, *next;
446 mutex_lock(&uuid_mutex);
448 /* This is the initialized path, it is safe to release the devices. */
449 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
450 if (device->in_fs_metadata)
454 blkdev_put(device->bdev, device->mode);
456 fs_devices->open_devices--;
458 if (device->writeable) {
459 list_del_init(&device->dev_alloc_list);
460 device->writeable = 0;
461 fs_devices->rw_devices--;
463 list_del_init(&device->dev_list);
464 fs_devices->num_devices--;
469 if (fs_devices->seed) {
470 fs_devices = fs_devices->seed;
474 mutex_unlock(&uuid_mutex);
478 static void __free_device(struct work_struct *work)
480 struct btrfs_device *device;
482 device = container_of(work, struct btrfs_device, rcu_work);
485 blkdev_put(device->bdev, device->mode);
491 static void free_device(struct rcu_head *head)
493 struct btrfs_device *device;
495 device = container_of(head, struct btrfs_device, rcu);
497 INIT_WORK(&device->rcu_work, __free_device);
498 schedule_work(&device->rcu_work);
501 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
503 struct btrfs_device *device;
505 if (--fs_devices->opened > 0)
508 mutex_lock(&fs_devices->device_list_mutex);
509 list_for_each_entry(device, &fs_devices->devices, dev_list) {
510 struct btrfs_device *new_device;
513 fs_devices->open_devices--;
515 if (device->writeable) {
516 list_del_init(&device->dev_alloc_list);
517 fs_devices->rw_devices--;
520 if (device->can_discard)
521 fs_devices->num_can_discard--;
523 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
525 memcpy(new_device, device, sizeof(*new_device));
526 new_device->name = kstrdup(device->name, GFP_NOFS);
527 BUG_ON(device->name && !new_device->name);
528 new_device->bdev = NULL;
529 new_device->writeable = 0;
530 new_device->in_fs_metadata = 0;
531 new_device->can_discard = 0;
532 list_replace_rcu(&device->dev_list, &new_device->dev_list);
534 call_rcu(&device->rcu, free_device);
536 mutex_unlock(&fs_devices->device_list_mutex);
538 WARN_ON(fs_devices->open_devices);
539 WARN_ON(fs_devices->rw_devices);
540 fs_devices->opened = 0;
541 fs_devices->seeding = 0;
546 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
548 struct btrfs_fs_devices *seed_devices = NULL;
551 mutex_lock(&uuid_mutex);
552 ret = __btrfs_close_devices(fs_devices);
553 if (!fs_devices->opened) {
554 seed_devices = fs_devices->seed;
555 fs_devices->seed = NULL;
557 mutex_unlock(&uuid_mutex);
559 while (seed_devices) {
560 fs_devices = seed_devices;
561 seed_devices = fs_devices->seed;
562 __btrfs_close_devices(fs_devices);
563 free_fs_devices(fs_devices);
568 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
569 fmode_t flags, void *holder)
571 struct request_queue *q;
572 struct block_device *bdev;
573 struct list_head *head = &fs_devices->devices;
574 struct btrfs_device *device;
575 struct block_device *latest_bdev = NULL;
576 struct buffer_head *bh;
577 struct btrfs_super_block *disk_super;
578 u64 latest_devid = 0;
579 u64 latest_transid = 0;
586 list_for_each_entry(device, head, dev_list) {
592 bdev = blkdev_get_by_path(device->name, flags, holder);
594 printk(KERN_INFO "open %s failed\n", device->name);
597 set_blocksize(bdev, 4096);
599 bh = btrfs_read_dev_super(bdev);
605 disk_super = (struct btrfs_super_block *)bh->b_data;
606 devid = btrfs_stack_device_id(&disk_super->dev_item);
607 if (devid != device->devid)
610 if (memcmp(device->uuid, disk_super->dev_item.uuid,
614 device->generation = btrfs_super_generation(disk_super);
615 if (!latest_transid || device->generation > latest_transid) {
616 latest_devid = devid;
617 latest_transid = device->generation;
621 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
622 device->writeable = 0;
624 device->writeable = !bdev_read_only(bdev);
628 q = bdev_get_queue(bdev);
629 if (blk_queue_discard(q)) {
630 device->can_discard = 1;
631 fs_devices->num_can_discard++;
635 device->in_fs_metadata = 0;
636 device->mode = flags;
638 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
639 fs_devices->rotating = 1;
641 fs_devices->open_devices++;
642 if (device->writeable) {
643 fs_devices->rw_devices++;
644 list_add(&device->dev_alloc_list,
645 &fs_devices->alloc_list);
653 blkdev_put(bdev, flags);
657 if (fs_devices->open_devices == 0) {
661 fs_devices->seeding = seeding;
662 fs_devices->opened = 1;
663 fs_devices->latest_bdev = latest_bdev;
664 fs_devices->latest_devid = latest_devid;
665 fs_devices->latest_trans = latest_transid;
666 fs_devices->total_rw_bytes = 0;
671 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
672 fmode_t flags, void *holder)
676 mutex_lock(&uuid_mutex);
677 if (fs_devices->opened) {
678 fs_devices->opened++;
681 ret = __btrfs_open_devices(fs_devices, flags, holder);
683 mutex_unlock(&uuid_mutex);
687 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
688 struct btrfs_fs_devices **fs_devices_ret)
690 struct btrfs_super_block *disk_super;
691 struct block_device *bdev;
692 struct buffer_head *bh;
697 mutex_lock(&uuid_mutex);
700 bdev = blkdev_get_by_path(path, flags, holder);
707 ret = set_blocksize(bdev, 4096);
710 bh = btrfs_read_dev_super(bdev);
715 disk_super = (struct btrfs_super_block *)bh->b_data;
716 devid = btrfs_stack_device_id(&disk_super->dev_item);
717 transid = btrfs_super_generation(disk_super);
718 if (disk_super->label[0])
719 printk(KERN_INFO "device label %s ", disk_super->label);
721 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
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 blkdev_put(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 = max(root->fs_info->alloc_start, 1024ull * 1024);
864 max_hole_start = search_start;
868 if (search_start >= search_end) {
873 path = btrfs_alloc_path();
880 key.objectid = device->devid;
881 key.offset = search_start;
882 key.type = BTRFS_DEV_EXTENT_KEY;
884 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
888 ret = btrfs_previous_item(root, path, key.objectid, key.type);
895 slot = path->slots[0];
896 if (slot >= btrfs_header_nritems(l)) {
897 ret = btrfs_next_leaf(root, path);
905 btrfs_item_key_to_cpu(l, &key, slot);
907 if (key.objectid < device->devid)
910 if (key.objectid > device->devid)
913 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
916 if (key.offset > search_start) {
917 hole_size = key.offset - search_start;
919 if (hole_size > max_hole_size) {
920 max_hole_start = search_start;
921 max_hole_size = hole_size;
925 * If this free space is greater than which we need,
926 * it must be the max free space that we have found
927 * until now, so max_hole_start must point to the start
928 * of this free space and the length of this free space
929 * is stored in max_hole_size. Thus, we return
930 * max_hole_start and max_hole_size and go back to the
933 if (hole_size >= num_bytes) {
939 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
940 extent_end = key.offset + btrfs_dev_extent_length(l,
942 if (extent_end > search_start)
943 search_start = extent_end;
950 * At this point, search_start should be the end of
951 * allocated dev extents, and when shrinking the device,
952 * search_end may be smaller than search_start.
954 if (search_end > search_start)
955 hole_size = search_end - search_start;
957 if (hole_size > max_hole_size) {
958 max_hole_start = search_start;
959 max_hole_size = hole_size;
963 if (hole_size < num_bytes)
969 btrfs_free_path(path);
971 *start = max_hole_start;
973 *len = max_hole_size;
977 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
978 struct btrfs_device *device,
982 struct btrfs_path *path;
983 struct btrfs_root *root = device->dev_root;
984 struct btrfs_key key;
985 struct btrfs_key found_key;
986 struct extent_buffer *leaf = NULL;
987 struct btrfs_dev_extent *extent = NULL;
989 path = btrfs_alloc_path();
993 key.objectid = device->devid;
995 key.type = BTRFS_DEV_EXTENT_KEY;
997 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
999 ret = btrfs_previous_item(root, path, key.objectid,
1000 BTRFS_DEV_EXTENT_KEY);
1003 leaf = path->nodes[0];
1004 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1005 extent = btrfs_item_ptr(leaf, path->slots[0],
1006 struct btrfs_dev_extent);
1007 BUG_ON(found_key.offset > start || found_key.offset +
1008 btrfs_dev_extent_length(leaf, extent) < start);
1009 } else if (ret == 0) {
1010 leaf = path->nodes[0];
1011 extent = btrfs_item_ptr(leaf, path->slots[0],
1012 struct btrfs_dev_extent);
1016 if (device->bytes_used > 0)
1017 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
1018 ret = btrfs_del_item(trans, root, path);
1021 btrfs_free_path(path);
1025 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1026 struct btrfs_device *device,
1027 u64 chunk_tree, u64 chunk_objectid,
1028 u64 chunk_offset, u64 start, u64 num_bytes)
1031 struct btrfs_path *path;
1032 struct btrfs_root *root = device->dev_root;
1033 struct btrfs_dev_extent *extent;
1034 struct extent_buffer *leaf;
1035 struct btrfs_key key;
1037 WARN_ON(!device->in_fs_metadata);
1038 path = btrfs_alloc_path();
1042 key.objectid = device->devid;
1044 key.type = BTRFS_DEV_EXTENT_KEY;
1045 ret = btrfs_insert_empty_item(trans, root, path, &key,
1049 leaf = path->nodes[0];
1050 extent = btrfs_item_ptr(leaf, path->slots[0],
1051 struct btrfs_dev_extent);
1052 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1053 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1054 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1056 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1057 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1060 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1061 btrfs_mark_buffer_dirty(leaf);
1062 btrfs_free_path(path);
1066 static noinline int find_next_chunk(struct btrfs_root *root,
1067 u64 objectid, u64 *offset)
1069 struct btrfs_path *path;
1071 struct btrfs_key key;
1072 struct btrfs_chunk *chunk;
1073 struct btrfs_key found_key;
1075 path = btrfs_alloc_path();
1079 key.objectid = objectid;
1080 key.offset = (u64)-1;
1081 key.type = BTRFS_CHUNK_ITEM_KEY;
1083 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1089 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1093 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1095 if (found_key.objectid != objectid)
1098 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1099 struct btrfs_chunk);
1100 *offset = found_key.offset +
1101 btrfs_chunk_length(path->nodes[0], chunk);
1106 btrfs_free_path(path);
1110 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1113 struct btrfs_key key;
1114 struct btrfs_key found_key;
1115 struct btrfs_path *path;
1117 root = root->fs_info->chunk_root;
1119 path = btrfs_alloc_path();
1123 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1124 key.type = BTRFS_DEV_ITEM_KEY;
1125 key.offset = (u64)-1;
1127 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1133 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1134 BTRFS_DEV_ITEM_KEY);
1138 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1140 *objectid = found_key.offset + 1;
1144 btrfs_free_path(path);
1149 * the device information is stored in the chunk root
1150 * the btrfs_device struct should be fully filled in
1152 int btrfs_add_device(struct btrfs_trans_handle *trans,
1153 struct btrfs_root *root,
1154 struct btrfs_device *device)
1157 struct btrfs_path *path;
1158 struct btrfs_dev_item *dev_item;
1159 struct extent_buffer *leaf;
1160 struct btrfs_key key;
1163 root = root->fs_info->chunk_root;
1165 path = btrfs_alloc_path();
1169 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1170 key.type = BTRFS_DEV_ITEM_KEY;
1171 key.offset = device->devid;
1173 ret = btrfs_insert_empty_item(trans, root, path, &key,
1178 leaf = path->nodes[0];
1179 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1181 btrfs_set_device_id(leaf, dev_item, device->devid);
1182 btrfs_set_device_generation(leaf, dev_item, 0);
1183 btrfs_set_device_type(leaf, dev_item, device->type);
1184 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1185 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1186 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1187 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1188 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1189 btrfs_set_device_group(leaf, dev_item, 0);
1190 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1191 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1192 btrfs_set_device_start_offset(leaf, dev_item, 0);
1194 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1195 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1196 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1197 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1198 btrfs_mark_buffer_dirty(leaf);
1202 btrfs_free_path(path);
1206 static int btrfs_rm_dev_item(struct btrfs_root *root,
1207 struct btrfs_device *device)
1210 struct btrfs_path *path;
1211 struct btrfs_key key;
1212 struct btrfs_trans_handle *trans;
1214 root = root->fs_info->chunk_root;
1216 path = btrfs_alloc_path();
1220 trans = btrfs_start_transaction(root, 0);
1221 if (IS_ERR(trans)) {
1222 btrfs_free_path(path);
1223 return PTR_ERR(trans);
1225 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1226 key.type = BTRFS_DEV_ITEM_KEY;
1227 key.offset = device->devid;
1230 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1239 ret = btrfs_del_item(trans, root, path);
1243 btrfs_free_path(path);
1244 unlock_chunks(root);
1245 btrfs_commit_transaction(trans, root);
1249 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1251 struct btrfs_device *device;
1252 struct btrfs_device *next_device;
1253 struct block_device *bdev;
1254 struct buffer_head *bh = NULL;
1255 struct btrfs_super_block *disk_super;
1256 struct btrfs_fs_devices *cur_devices;
1262 bool clear_super = false;
1264 mutex_lock(&uuid_mutex);
1265 mutex_lock(&root->fs_info->volume_mutex);
1267 all_avail = root->fs_info->avail_data_alloc_bits |
1268 root->fs_info->avail_system_alloc_bits |
1269 root->fs_info->avail_metadata_alloc_bits;
1271 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1272 root->fs_info->fs_devices->num_devices <= 4) {
1273 printk(KERN_ERR "btrfs: unable to go below four devices "
1279 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1280 root->fs_info->fs_devices->num_devices <= 2) {
1281 printk(KERN_ERR "btrfs: unable to go below two "
1282 "devices on raid1\n");
1287 if (strcmp(device_path, "missing") == 0) {
1288 struct list_head *devices;
1289 struct btrfs_device *tmp;
1292 devices = &root->fs_info->fs_devices->devices;
1294 * It is safe to read the devices since the volume_mutex
1297 list_for_each_entry(tmp, devices, dev_list) {
1298 if (tmp->in_fs_metadata && !tmp->bdev) {
1307 printk(KERN_ERR "btrfs: no missing devices found to "
1312 bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL,
1313 root->fs_info->bdev_holder);
1315 ret = PTR_ERR(bdev);
1319 set_blocksize(bdev, 4096);
1320 bh = btrfs_read_dev_super(bdev);
1325 disk_super = (struct btrfs_super_block *)bh->b_data;
1326 devid = btrfs_stack_device_id(&disk_super->dev_item);
1327 dev_uuid = disk_super->dev_item.uuid;
1328 device = btrfs_find_device(root, devid, dev_uuid,
1336 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1337 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1343 if (device->writeable) {
1345 list_del_init(&device->dev_alloc_list);
1346 unlock_chunks(root);
1347 root->fs_info->fs_devices->rw_devices--;
1351 ret = btrfs_shrink_device(device, 0);
1355 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1359 device->in_fs_metadata = 0;
1360 btrfs_scrub_cancel_dev(root, device);
1363 * the device list mutex makes sure that we don't change
1364 * the device list while someone else is writing out all
1365 * the device supers.
1368 cur_devices = device->fs_devices;
1369 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1370 list_del_rcu(&device->dev_list);
1372 device->fs_devices->num_devices--;
1374 if (device->missing)
1375 root->fs_info->fs_devices->missing_devices--;
1377 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1378 struct btrfs_device, dev_list);
1379 if (device->bdev == root->fs_info->sb->s_bdev)
1380 root->fs_info->sb->s_bdev = next_device->bdev;
1381 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1382 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1385 device->fs_devices->open_devices--;
1387 call_rcu(&device->rcu, free_device);
1388 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1390 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1391 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1393 if (cur_devices->open_devices == 0) {
1394 struct btrfs_fs_devices *fs_devices;
1395 fs_devices = root->fs_info->fs_devices;
1396 while (fs_devices) {
1397 if (fs_devices->seed == cur_devices)
1399 fs_devices = fs_devices->seed;
1401 fs_devices->seed = cur_devices->seed;
1402 cur_devices->seed = NULL;
1404 __btrfs_close_devices(cur_devices);
1405 unlock_chunks(root);
1406 free_fs_devices(cur_devices);
1410 * at this point, the device is zero sized. We want to
1411 * remove it from the devices list and zero out the old super
1414 /* make sure this device isn't detected as part of
1417 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1418 set_buffer_dirty(bh);
1419 sync_dirty_buffer(bh);
1428 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1430 mutex_unlock(&root->fs_info->volume_mutex);
1431 mutex_unlock(&uuid_mutex);
1434 if (device->writeable) {
1436 list_add(&device->dev_alloc_list,
1437 &root->fs_info->fs_devices->alloc_list);
1438 unlock_chunks(root);
1439 root->fs_info->fs_devices->rw_devices++;
1445 * does all the dirty work required for changing file system's UUID.
1447 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1448 struct btrfs_root *root)
1450 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1451 struct btrfs_fs_devices *old_devices;
1452 struct btrfs_fs_devices *seed_devices;
1453 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1454 struct btrfs_device *device;
1457 BUG_ON(!mutex_is_locked(&uuid_mutex));
1458 if (!fs_devices->seeding)
1461 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1465 old_devices = clone_fs_devices(fs_devices);
1466 if (IS_ERR(old_devices)) {
1467 kfree(seed_devices);
1468 return PTR_ERR(old_devices);
1471 list_add(&old_devices->list, &fs_uuids);
1473 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1474 seed_devices->opened = 1;
1475 INIT_LIST_HEAD(&seed_devices->devices);
1476 INIT_LIST_HEAD(&seed_devices->alloc_list);
1477 mutex_init(&seed_devices->device_list_mutex);
1479 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1480 list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1482 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1484 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1485 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1486 device->fs_devices = seed_devices;
1489 fs_devices->seeding = 0;
1490 fs_devices->num_devices = 0;
1491 fs_devices->open_devices = 0;
1492 fs_devices->seed = seed_devices;
1494 generate_random_uuid(fs_devices->fsid);
1495 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1496 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1497 super_flags = btrfs_super_flags(disk_super) &
1498 ~BTRFS_SUPER_FLAG_SEEDING;
1499 btrfs_set_super_flags(disk_super, super_flags);
1505 * strore the expected generation for seed devices in device items.
1507 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1508 struct btrfs_root *root)
1510 struct btrfs_path *path;
1511 struct extent_buffer *leaf;
1512 struct btrfs_dev_item *dev_item;
1513 struct btrfs_device *device;
1514 struct btrfs_key key;
1515 u8 fs_uuid[BTRFS_UUID_SIZE];
1516 u8 dev_uuid[BTRFS_UUID_SIZE];
1520 path = btrfs_alloc_path();
1524 root = root->fs_info->chunk_root;
1525 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1527 key.type = BTRFS_DEV_ITEM_KEY;
1530 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1534 leaf = path->nodes[0];
1536 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1537 ret = btrfs_next_leaf(root, path);
1542 leaf = path->nodes[0];
1543 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1544 btrfs_release_path(path);
1548 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1549 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1550 key.type != BTRFS_DEV_ITEM_KEY)
1553 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1554 struct btrfs_dev_item);
1555 devid = btrfs_device_id(leaf, dev_item);
1556 read_extent_buffer(leaf, dev_uuid,
1557 (unsigned long)btrfs_device_uuid(dev_item),
1559 read_extent_buffer(leaf, fs_uuid,
1560 (unsigned long)btrfs_device_fsid(dev_item),
1562 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1565 if (device->fs_devices->seeding) {
1566 btrfs_set_device_generation(leaf, dev_item,
1567 device->generation);
1568 btrfs_mark_buffer_dirty(leaf);
1576 btrfs_free_path(path);
1580 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1582 struct request_queue *q;
1583 struct btrfs_trans_handle *trans;
1584 struct btrfs_device *device;
1585 struct block_device *bdev;
1586 struct list_head *devices;
1587 struct super_block *sb = root->fs_info->sb;
1589 int seeding_dev = 0;
1592 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1595 bdev = blkdev_get_by_path(device_path, FMODE_EXCL,
1596 root->fs_info->bdev_holder);
1598 return PTR_ERR(bdev);
1600 if (root->fs_info->fs_devices->seeding) {
1602 down_write(&sb->s_umount);
1603 mutex_lock(&uuid_mutex);
1606 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1607 mutex_lock(&root->fs_info->volume_mutex);
1609 devices = &root->fs_info->fs_devices->devices;
1611 * we have the volume lock, so we don't need the extra
1612 * device list mutex while reading the list here.
1614 list_for_each_entry(device, devices, dev_list) {
1615 if (device->bdev == bdev) {
1621 device = kzalloc(sizeof(*device), GFP_NOFS);
1623 /* we can safely leave the fs_devices entry around */
1628 device->name = kstrdup(device_path, GFP_NOFS);
1629 if (!device->name) {
1635 ret = find_next_devid(root, &device->devid);
1637 kfree(device->name);
1642 trans = btrfs_start_transaction(root, 0);
1643 if (IS_ERR(trans)) {
1644 kfree(device->name);
1646 ret = PTR_ERR(trans);
1652 q = bdev_get_queue(bdev);
1653 if (blk_queue_discard(q))
1654 device->can_discard = 1;
1655 device->writeable = 1;
1656 device->work.func = pending_bios_fn;
1657 generate_random_uuid(device->uuid);
1658 spin_lock_init(&device->io_lock);
1659 device->generation = trans->transid;
1660 device->io_width = root->sectorsize;
1661 device->io_align = root->sectorsize;
1662 device->sector_size = root->sectorsize;
1663 device->total_bytes = i_size_read(bdev->bd_inode);
1664 device->disk_total_bytes = device->total_bytes;
1665 device->dev_root = root->fs_info->dev_root;
1666 device->bdev = bdev;
1667 device->in_fs_metadata = 1;
1668 device->mode = FMODE_EXCL;
1669 set_blocksize(device->bdev, 4096);
1672 sb->s_flags &= ~MS_RDONLY;
1673 ret = btrfs_prepare_sprout(trans, root);
1677 device->fs_devices = root->fs_info->fs_devices;
1680 * we don't want write_supers to jump in here with our device
1683 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1684 list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
1685 list_add(&device->dev_alloc_list,
1686 &root->fs_info->fs_devices->alloc_list);
1687 root->fs_info->fs_devices->num_devices++;
1688 root->fs_info->fs_devices->open_devices++;
1689 root->fs_info->fs_devices->rw_devices++;
1690 if (device->can_discard)
1691 root->fs_info->fs_devices->num_can_discard++;
1692 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1694 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1695 root->fs_info->fs_devices->rotating = 1;
1697 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1698 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1699 total_bytes + device->total_bytes);
1701 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1702 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1704 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1707 ret = init_first_rw_device(trans, root, device);
1709 ret = btrfs_finish_sprout(trans, root);
1712 ret = btrfs_add_device(trans, root, device);
1716 * we've got more storage, clear any full flags on the space
1719 btrfs_clear_space_info_full(root->fs_info);
1721 unlock_chunks(root);
1722 btrfs_commit_transaction(trans, root);
1725 mutex_unlock(&uuid_mutex);
1726 up_write(&sb->s_umount);
1728 ret = btrfs_relocate_sys_chunks(root);
1732 mutex_unlock(&root->fs_info->volume_mutex);
1735 blkdev_put(bdev, FMODE_EXCL);
1737 mutex_unlock(&uuid_mutex);
1738 up_write(&sb->s_umount);
1743 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1744 struct btrfs_device *device)
1747 struct btrfs_path *path;
1748 struct btrfs_root *root;
1749 struct btrfs_dev_item *dev_item;
1750 struct extent_buffer *leaf;
1751 struct btrfs_key key;
1753 root = device->dev_root->fs_info->chunk_root;
1755 path = btrfs_alloc_path();
1759 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1760 key.type = BTRFS_DEV_ITEM_KEY;
1761 key.offset = device->devid;
1763 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1772 leaf = path->nodes[0];
1773 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1775 btrfs_set_device_id(leaf, dev_item, device->devid);
1776 btrfs_set_device_type(leaf, dev_item, device->type);
1777 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1778 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1779 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1780 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1781 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1782 btrfs_mark_buffer_dirty(leaf);
1785 btrfs_free_path(path);
1789 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1790 struct btrfs_device *device, u64 new_size)
1792 struct btrfs_super_block *super_copy =
1793 &device->dev_root->fs_info->super_copy;
1794 u64 old_total = btrfs_super_total_bytes(super_copy);
1795 u64 diff = new_size - device->total_bytes;
1797 if (!device->writeable)
1799 if (new_size <= device->total_bytes)
1802 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1803 device->fs_devices->total_rw_bytes += diff;
1805 device->total_bytes = new_size;
1806 device->disk_total_bytes = new_size;
1807 btrfs_clear_space_info_full(device->dev_root->fs_info);
1809 return btrfs_update_device(trans, device);
1812 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1813 struct btrfs_device *device, u64 new_size)
1816 lock_chunks(device->dev_root);
1817 ret = __btrfs_grow_device(trans, device, new_size);
1818 unlock_chunks(device->dev_root);
1822 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1823 struct btrfs_root *root,
1824 u64 chunk_tree, u64 chunk_objectid,
1828 struct btrfs_path *path;
1829 struct btrfs_key key;
1831 root = root->fs_info->chunk_root;
1832 path = btrfs_alloc_path();
1836 key.objectid = chunk_objectid;
1837 key.offset = chunk_offset;
1838 key.type = BTRFS_CHUNK_ITEM_KEY;
1840 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1843 ret = btrfs_del_item(trans, root, path);
1845 btrfs_free_path(path);
1849 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1852 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1853 struct btrfs_disk_key *disk_key;
1854 struct btrfs_chunk *chunk;
1861 struct btrfs_key key;
1863 array_size = btrfs_super_sys_array_size(super_copy);
1865 ptr = super_copy->sys_chunk_array;
1868 while (cur < array_size) {
1869 disk_key = (struct btrfs_disk_key *)ptr;
1870 btrfs_disk_key_to_cpu(&key, disk_key);
1872 len = sizeof(*disk_key);
1874 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1875 chunk = (struct btrfs_chunk *)(ptr + len);
1876 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1877 len += btrfs_chunk_item_size(num_stripes);
1882 if (key.objectid == chunk_objectid &&
1883 key.offset == chunk_offset) {
1884 memmove(ptr, ptr + len, array_size - (cur + len));
1886 btrfs_set_super_sys_array_size(super_copy, array_size);
1895 static int btrfs_relocate_chunk(struct btrfs_root *root,
1896 u64 chunk_tree, u64 chunk_objectid,
1899 struct extent_map_tree *em_tree;
1900 struct btrfs_root *extent_root;
1901 struct btrfs_trans_handle *trans;
1902 struct extent_map *em;
1903 struct map_lookup *map;
1907 root = root->fs_info->chunk_root;
1908 extent_root = root->fs_info->extent_root;
1909 em_tree = &root->fs_info->mapping_tree.map_tree;
1911 ret = btrfs_can_relocate(extent_root, chunk_offset);
1915 /* step one, relocate all the extents inside this chunk */
1916 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1920 trans = btrfs_start_transaction(root, 0);
1921 BUG_ON(IS_ERR(trans));
1926 * step two, delete the device extents and the
1927 * chunk tree entries
1929 read_lock(&em_tree->lock);
1930 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1931 read_unlock(&em_tree->lock);
1933 BUG_ON(em->start > chunk_offset ||
1934 em->start + em->len < chunk_offset);
1935 map = (struct map_lookup *)em->bdev;
1937 for (i = 0; i < map->num_stripes; i++) {
1938 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1939 map->stripes[i].physical);
1942 if (map->stripes[i].dev) {
1943 ret = btrfs_update_device(trans, map->stripes[i].dev);
1947 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1952 trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
1954 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1955 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1959 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1962 write_lock(&em_tree->lock);
1963 remove_extent_mapping(em_tree, em);
1964 write_unlock(&em_tree->lock);
1969 /* once for the tree */
1970 free_extent_map(em);
1972 free_extent_map(em);
1974 unlock_chunks(root);
1975 btrfs_end_transaction(trans, root);
1979 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1981 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1982 struct btrfs_path *path;
1983 struct extent_buffer *leaf;
1984 struct btrfs_chunk *chunk;
1985 struct btrfs_key key;
1986 struct btrfs_key found_key;
1987 u64 chunk_tree = chunk_root->root_key.objectid;
1989 bool retried = false;
1993 path = btrfs_alloc_path();
1998 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1999 key.offset = (u64)-1;
2000 key.type = BTRFS_CHUNK_ITEM_KEY;
2003 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2008 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2015 leaf = path->nodes[0];
2016 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2018 chunk = btrfs_item_ptr(leaf, path->slots[0],
2019 struct btrfs_chunk);
2020 chunk_type = btrfs_chunk_type(leaf, chunk);
2021 btrfs_release_path(path);
2023 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2024 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2033 if (found_key.offset == 0)
2035 key.offset = found_key.offset - 1;
2038 if (failed && !retried) {
2042 } else if (failed && retried) {
2047 btrfs_free_path(path);
2051 static u64 div_factor(u64 num, int factor)
2060 int btrfs_balance(struct btrfs_root *dev_root)
2063 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
2064 struct btrfs_device *device;
2067 struct btrfs_path *path;
2068 struct btrfs_key key;
2069 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
2070 struct btrfs_trans_handle *trans;
2071 struct btrfs_key found_key;
2073 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
2076 if (!capable(CAP_SYS_ADMIN))
2079 mutex_lock(&dev_root->fs_info->volume_mutex);
2080 dev_root = dev_root->fs_info->dev_root;
2082 /* step one make some room on all the devices */
2083 list_for_each_entry(device, devices, dev_list) {
2084 old_size = device->total_bytes;
2085 size_to_free = div_factor(old_size, 1);
2086 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2087 if (!device->writeable ||
2088 device->total_bytes - device->bytes_used > size_to_free)
2091 ret = btrfs_shrink_device(device, old_size - size_to_free);
2096 trans = btrfs_start_transaction(dev_root, 0);
2097 BUG_ON(IS_ERR(trans));
2099 ret = btrfs_grow_device(trans, device, old_size);
2102 btrfs_end_transaction(trans, dev_root);
2105 /* step two, relocate all the chunks */
2106 path = btrfs_alloc_path();
2111 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2112 key.offset = (u64)-1;
2113 key.type = BTRFS_CHUNK_ITEM_KEY;
2116 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2121 * this shouldn't happen, it means the last relocate
2127 ret = btrfs_previous_item(chunk_root, path, 0,
2128 BTRFS_CHUNK_ITEM_KEY);
2132 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
2134 if (found_key.objectid != key.objectid)
2137 /* chunk zero is special */
2138 if (found_key.offset == 0)
2141 btrfs_release_path(path);
2142 ret = btrfs_relocate_chunk(chunk_root,
2143 chunk_root->root_key.objectid,
2146 if (ret && ret != -ENOSPC)
2148 key.offset = found_key.offset - 1;
2152 btrfs_free_path(path);
2153 mutex_unlock(&dev_root->fs_info->volume_mutex);
2158 * shrinking a device means finding all of the device extents past
2159 * the new size, and then following the back refs to the chunks.
2160 * The chunk relocation code actually frees the device extent
2162 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2164 struct btrfs_trans_handle *trans;
2165 struct btrfs_root *root = device->dev_root;
2166 struct btrfs_dev_extent *dev_extent = NULL;
2167 struct btrfs_path *path;
2175 bool retried = false;
2176 struct extent_buffer *l;
2177 struct btrfs_key key;
2178 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2179 u64 old_total = btrfs_super_total_bytes(super_copy);
2180 u64 old_size = device->total_bytes;
2181 u64 diff = device->total_bytes - new_size;
2183 if (new_size >= device->total_bytes)
2186 path = btrfs_alloc_path();
2194 device->total_bytes = new_size;
2195 if (device->writeable)
2196 device->fs_devices->total_rw_bytes -= diff;
2197 unlock_chunks(root);
2200 key.objectid = device->devid;
2201 key.offset = (u64)-1;
2202 key.type = BTRFS_DEV_EXTENT_KEY;
2205 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2209 ret = btrfs_previous_item(root, path, 0, key.type);
2214 btrfs_release_path(path);
2219 slot = path->slots[0];
2220 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2222 if (key.objectid != device->devid) {
2223 btrfs_release_path(path);
2227 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2228 length = btrfs_dev_extent_length(l, dev_extent);
2230 if (key.offset + length <= new_size) {
2231 btrfs_release_path(path);
2235 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2236 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2237 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2238 btrfs_release_path(path);
2240 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2242 if (ret && ret != -ENOSPC)
2249 if (failed && !retried) {
2253 } else if (failed && retried) {
2257 device->total_bytes = old_size;
2258 if (device->writeable)
2259 device->fs_devices->total_rw_bytes += diff;
2260 unlock_chunks(root);
2264 /* Shrinking succeeded, else we would be at "done". */
2265 trans = btrfs_start_transaction(root, 0);
2266 if (IS_ERR(trans)) {
2267 ret = PTR_ERR(trans);
2273 device->disk_total_bytes = new_size;
2274 /* Now btrfs_update_device() will change the on-disk size. */
2275 ret = btrfs_update_device(trans, device);
2277 unlock_chunks(root);
2278 btrfs_end_transaction(trans, root);
2281 WARN_ON(diff > old_total);
2282 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2283 unlock_chunks(root);
2284 btrfs_end_transaction(trans, root);
2286 btrfs_free_path(path);
2290 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2291 struct btrfs_root *root,
2292 struct btrfs_key *key,
2293 struct btrfs_chunk *chunk, int item_size)
2295 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2296 struct btrfs_disk_key disk_key;
2300 array_size = btrfs_super_sys_array_size(super_copy);
2301 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2304 ptr = super_copy->sys_chunk_array + array_size;
2305 btrfs_cpu_key_to_disk(&disk_key, key);
2306 memcpy(ptr, &disk_key, sizeof(disk_key));
2307 ptr += sizeof(disk_key);
2308 memcpy(ptr, chunk, item_size);
2309 item_size += sizeof(disk_key);
2310 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2315 * sort the devices in descending order by max_avail, total_avail
2317 static int btrfs_cmp_device_info(const void *a, const void *b)
2319 const struct btrfs_device_info *di_a = a;
2320 const struct btrfs_device_info *di_b = b;
2322 if (di_a->max_avail > di_b->max_avail)
2324 if (di_a->max_avail < di_b->max_avail)
2326 if (di_a->total_avail > di_b->total_avail)
2328 if (di_a->total_avail < di_b->total_avail)
2333 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2334 struct btrfs_root *extent_root,
2335 struct map_lookup **map_ret,
2336 u64 *num_bytes_out, u64 *stripe_size_out,
2337 u64 start, u64 type)
2339 struct btrfs_fs_info *info = extent_root->fs_info;
2340 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2341 struct list_head *cur;
2342 struct map_lookup *map = NULL;
2343 struct extent_map_tree *em_tree;
2344 struct extent_map *em;
2345 struct btrfs_device_info *devices_info = NULL;
2347 int num_stripes; /* total number of stripes to allocate */
2348 int sub_stripes; /* sub_stripes info for map */
2349 int dev_stripes; /* stripes per dev */
2350 int devs_max; /* max devs to use */
2351 int devs_min; /* min devs needed */
2352 int devs_increment; /* ndevs has to be a multiple of this */
2353 int ncopies; /* how many copies to data has */
2355 u64 max_stripe_size;
2363 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2364 (type & BTRFS_BLOCK_GROUP_DUP)) {
2366 type &= ~BTRFS_BLOCK_GROUP_DUP;
2369 if (list_empty(&fs_devices->alloc_list))
2376 devs_max = 0; /* 0 == as many as possible */
2380 * define the properties of each RAID type.
2381 * FIXME: move this to a global table and use it in all RAID
2384 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2388 } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2390 } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2395 } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2404 if (type & BTRFS_BLOCK_GROUP_DATA) {
2405 max_stripe_size = 1024 * 1024 * 1024;
2406 max_chunk_size = 10 * max_stripe_size;
2407 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2408 max_stripe_size = 256 * 1024 * 1024;
2409 max_chunk_size = max_stripe_size;
2410 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2411 max_stripe_size = 8 * 1024 * 1024;
2412 max_chunk_size = 2 * max_stripe_size;
2414 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
2419 /* we don't want a chunk larger than 10% of writeable space */
2420 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2423 devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
2428 cur = fs_devices->alloc_list.next;
2431 * in the first pass through the devices list, we gather information
2432 * about the available holes on each device.
2435 while (cur != &fs_devices->alloc_list) {
2436 struct btrfs_device *device;
2440 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2444 if (!device->writeable) {
2446 "btrfs: read-only device in alloc_list\n");
2451 if (!device->in_fs_metadata)
2454 if (device->total_bytes > device->bytes_used)
2455 total_avail = device->total_bytes - device->bytes_used;
2459 /* If there is no space on this device, skip it. */
2460 if (total_avail == 0)
2463 ret = find_free_dev_extent(trans, device,
2464 max_stripe_size * dev_stripes,
2465 &dev_offset, &max_avail);
2466 if (ret && ret != -ENOSPC)
2470 max_avail = max_stripe_size * dev_stripes;
2472 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
2475 devices_info[ndevs].dev_offset = dev_offset;
2476 devices_info[ndevs].max_avail = max_avail;
2477 devices_info[ndevs].total_avail = total_avail;
2478 devices_info[ndevs].dev = device;
2483 * now sort the devices by hole size / available space
2485 sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
2486 btrfs_cmp_device_info, NULL);
2488 /* round down to number of usable stripes */
2489 ndevs -= ndevs % devs_increment;
2491 if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
2496 if (devs_max && ndevs > devs_max)
2499 * the primary goal is to maximize the number of stripes, so use as many
2500 * devices as possible, even if the stripes are not maximum sized.
2502 stripe_size = devices_info[ndevs-1].max_avail;
2503 num_stripes = ndevs * dev_stripes;
2505 if (stripe_size * num_stripes > max_chunk_size * ncopies) {
2506 stripe_size = max_chunk_size * ncopies;
2507 do_div(stripe_size, num_stripes);
2510 do_div(stripe_size, dev_stripes);
2511 do_div(stripe_size, BTRFS_STRIPE_LEN);
2512 stripe_size *= BTRFS_STRIPE_LEN;
2514 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2519 map->num_stripes = num_stripes;
2521 for (i = 0; i < ndevs; ++i) {
2522 for (j = 0; j < dev_stripes; ++j) {
2523 int s = i * dev_stripes + j;
2524 map->stripes[s].dev = devices_info[i].dev;
2525 map->stripes[s].physical = devices_info[i].dev_offset +
2529 map->sector_size = extent_root->sectorsize;
2530 map->stripe_len = BTRFS_STRIPE_LEN;
2531 map->io_align = BTRFS_STRIPE_LEN;
2532 map->io_width = BTRFS_STRIPE_LEN;
2534 map->sub_stripes = sub_stripes;
2537 num_bytes = stripe_size * (num_stripes / ncopies);
2539 *stripe_size_out = stripe_size;
2540 *num_bytes_out = num_bytes;
2542 trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
2544 em = alloc_extent_map();
2549 em->bdev = (struct block_device *)map;
2551 em->len = num_bytes;
2552 em->block_start = 0;
2553 em->block_len = em->len;
2555 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2556 write_lock(&em_tree->lock);
2557 ret = add_extent_mapping(em_tree, em);
2558 write_unlock(&em_tree->lock);
2560 free_extent_map(em);
2562 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2563 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2567 for (i = 0; i < map->num_stripes; ++i) {
2568 struct btrfs_device *device;
2571 device = map->stripes[i].dev;
2572 dev_offset = map->stripes[i].physical;
2574 ret = btrfs_alloc_dev_extent(trans, device,
2575 info->chunk_root->root_key.objectid,
2576 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2577 start, dev_offset, stripe_size);
2581 kfree(devices_info);
2586 kfree(devices_info);
2590 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2591 struct btrfs_root *extent_root,
2592 struct map_lookup *map, u64 chunk_offset,
2593 u64 chunk_size, u64 stripe_size)
2596 struct btrfs_key key;
2597 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2598 struct btrfs_device *device;
2599 struct btrfs_chunk *chunk;
2600 struct btrfs_stripe *stripe;
2601 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2605 chunk = kzalloc(item_size, GFP_NOFS);
2610 while (index < map->num_stripes) {
2611 device = map->stripes[index].dev;
2612 device->bytes_used += stripe_size;
2613 ret = btrfs_update_device(trans, device);
2619 stripe = &chunk->stripe;
2620 while (index < map->num_stripes) {
2621 device = map->stripes[index].dev;
2622 dev_offset = map->stripes[index].physical;
2624 btrfs_set_stack_stripe_devid(stripe, device->devid);
2625 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2626 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2631 btrfs_set_stack_chunk_length(chunk, chunk_size);
2632 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2633 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2634 btrfs_set_stack_chunk_type(chunk, map->type);
2635 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2636 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2637 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2638 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2639 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2641 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2642 key.type = BTRFS_CHUNK_ITEM_KEY;
2643 key.offset = chunk_offset;
2645 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2648 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2649 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2659 * Chunk allocation falls into two parts. The first part does works
2660 * that make the new allocated chunk useable, but not do any operation
2661 * that modifies the chunk tree. The second part does the works that
2662 * require modifying the chunk tree. This division is important for the
2663 * bootstrap process of adding storage to a seed btrfs.
2665 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2666 struct btrfs_root *extent_root, u64 type)
2671 struct map_lookup *map;
2672 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2675 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2680 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2681 &stripe_size, chunk_offset, type);
2685 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2686 chunk_size, stripe_size);
2691 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2692 struct btrfs_root *root,
2693 struct btrfs_device *device)
2696 u64 sys_chunk_offset;
2700 u64 sys_stripe_size;
2702 struct map_lookup *map;
2703 struct map_lookup *sys_map;
2704 struct btrfs_fs_info *fs_info = root->fs_info;
2705 struct btrfs_root *extent_root = fs_info->extent_root;
2708 ret = find_next_chunk(fs_info->chunk_root,
2709 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2713 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2714 (fs_info->metadata_alloc_profile &
2715 fs_info->avail_metadata_alloc_bits);
2716 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2718 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2719 &stripe_size, chunk_offset, alloc_profile);
2722 sys_chunk_offset = chunk_offset + chunk_size;
2724 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2725 (fs_info->system_alloc_profile &
2726 fs_info->avail_system_alloc_bits);
2727 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2729 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2730 &sys_chunk_size, &sys_stripe_size,
2731 sys_chunk_offset, alloc_profile);
2734 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2738 * Modifying chunk tree needs allocating new blocks from both
2739 * system block group and metadata block group. So we only can
2740 * do operations require modifying the chunk tree after both
2741 * block groups were created.
2743 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2744 chunk_size, stripe_size);
2747 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2748 sys_chunk_offset, sys_chunk_size,
2754 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2756 struct extent_map *em;
2757 struct map_lookup *map;
2758 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2762 read_lock(&map_tree->map_tree.lock);
2763 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2764 read_unlock(&map_tree->map_tree.lock);
2768 if (btrfs_test_opt(root, DEGRADED)) {
2769 free_extent_map(em);
2773 map = (struct map_lookup *)em->bdev;
2774 for (i = 0; i < map->num_stripes; i++) {
2775 if (!map->stripes[i].dev->writeable) {
2780 free_extent_map(em);
2784 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2786 extent_map_tree_init(&tree->map_tree);
2789 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2791 struct extent_map *em;
2794 write_lock(&tree->map_tree.lock);
2795 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2797 remove_extent_mapping(&tree->map_tree, em);
2798 write_unlock(&tree->map_tree.lock);
2803 free_extent_map(em);
2804 /* once for the tree */
2805 free_extent_map(em);
2809 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2811 struct extent_map *em;
2812 struct map_lookup *map;
2813 struct extent_map_tree *em_tree = &map_tree->map_tree;
2816 read_lock(&em_tree->lock);
2817 em = lookup_extent_mapping(em_tree, logical, len);
2818 read_unlock(&em_tree->lock);
2821 BUG_ON(em->start > logical || em->start + em->len < logical);
2822 map = (struct map_lookup *)em->bdev;
2823 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2824 ret = map->num_stripes;
2825 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2826 ret = map->sub_stripes;
2829 free_extent_map(em);
2833 static int find_live_mirror(struct map_lookup *map, int first, int num,
2837 if (map->stripes[optimal].dev->bdev)
2839 for (i = first; i < first + num; i++) {
2840 if (map->stripes[i].dev->bdev)
2843 /* we couldn't find one that doesn't fail. Just return something
2844 * and the io error handling code will clean up eventually
2849 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2850 u64 logical, u64 *length,
2851 struct btrfs_multi_bio **multi_ret,
2854 struct extent_map *em;
2855 struct map_lookup *map;
2856 struct extent_map_tree *em_tree = &map_tree->map_tree;
2859 u64 stripe_end_offset;
2863 int stripes_allocated = 8;
2864 int stripes_required = 1;
2869 struct btrfs_multi_bio *multi = NULL;
2871 if (multi_ret && !(rw & (REQ_WRITE | REQ_DISCARD)))
2872 stripes_allocated = 1;
2875 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2880 atomic_set(&multi->error, 0);
2883 read_lock(&em_tree->lock);
2884 em = lookup_extent_mapping(em_tree, logical, *length);
2885 read_unlock(&em_tree->lock);
2888 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2889 (unsigned long long)logical,
2890 (unsigned long long)*length);
2894 BUG_ON(em->start > logical || em->start + em->len < logical);
2895 map = (struct map_lookup *)em->bdev;
2896 offset = logical - em->start;
2898 if (mirror_num > map->num_stripes)
2901 /* if our multi bio struct is too small, back off and try again */
2902 if (rw & REQ_WRITE) {
2903 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2904 BTRFS_BLOCK_GROUP_DUP)) {
2905 stripes_required = map->num_stripes;
2907 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2908 stripes_required = map->sub_stripes;
2912 if (rw & REQ_DISCARD) {
2913 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2914 BTRFS_BLOCK_GROUP_RAID1 |
2915 BTRFS_BLOCK_GROUP_DUP |
2916 BTRFS_BLOCK_GROUP_RAID10)) {
2917 stripes_required = map->num_stripes;
2920 if (multi_ret && (rw & (REQ_WRITE | REQ_DISCARD)) &&
2921 stripes_allocated < stripes_required) {
2922 stripes_allocated = map->num_stripes;
2923 free_extent_map(em);
2929 * stripe_nr counts the total number of stripes we have to stride
2930 * to get to this block
2932 do_div(stripe_nr, map->stripe_len);
2934 stripe_offset = stripe_nr * map->stripe_len;
2935 BUG_ON(offset < stripe_offset);
2937 /* stripe_offset is the offset of this block in its stripe*/
2938 stripe_offset = offset - stripe_offset;
2940 if (rw & REQ_DISCARD)
2941 *length = min_t(u64, em->len - offset, *length);
2942 else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
2943 BTRFS_BLOCK_GROUP_RAID1 |
2944 BTRFS_BLOCK_GROUP_RAID10 |
2945 BTRFS_BLOCK_GROUP_DUP)) {
2946 /* we limit the length of each bio to what fits in a stripe */
2947 *length = min_t(u64, em->len - offset,
2948 map->stripe_len - stripe_offset);
2950 *length = em->len - offset;
2958 stripe_nr_orig = stripe_nr;
2959 stripe_nr_end = (offset + *length + map->stripe_len - 1) &
2960 (~(map->stripe_len - 1));
2961 do_div(stripe_nr_end, map->stripe_len);
2962 stripe_end_offset = stripe_nr_end * map->stripe_len -
2964 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2965 if (rw & REQ_DISCARD)
2966 num_stripes = min_t(u64, map->num_stripes,
2967 stripe_nr_end - stripe_nr_orig);
2968 stripe_index = do_div(stripe_nr, map->num_stripes);
2969 } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2970 if (rw & (REQ_WRITE | REQ_DISCARD))
2971 num_stripes = map->num_stripes;
2972 else if (mirror_num)
2973 stripe_index = mirror_num - 1;
2975 stripe_index = find_live_mirror(map, 0,
2977 current->pid % map->num_stripes);
2980 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2981 if (rw & (REQ_WRITE | REQ_DISCARD))
2982 num_stripes = map->num_stripes;
2983 else if (mirror_num)
2984 stripe_index = mirror_num - 1;
2986 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2987 int factor = map->num_stripes / map->sub_stripes;
2989 stripe_index = do_div(stripe_nr, factor);
2990 stripe_index *= map->sub_stripes;
2993 num_stripes = map->sub_stripes;
2994 else if (rw & REQ_DISCARD)
2995 num_stripes = min_t(u64, map->sub_stripes *
2996 (stripe_nr_end - stripe_nr_orig),
2998 else if (mirror_num)
2999 stripe_index += mirror_num - 1;
3001 stripe_index = find_live_mirror(map, stripe_index,
3002 map->sub_stripes, stripe_index +
3003 current->pid % map->sub_stripes);
3007 * after this do_div call, stripe_nr is the number of stripes
3008 * on this device we have to walk to find the data, and
3009 * stripe_index is the number of our device in the stripe array
3011 stripe_index = do_div(stripe_nr, map->num_stripes);
3013 BUG_ON(stripe_index >= map->num_stripes);
3015 if (rw & REQ_DISCARD) {
3016 for (i = 0; i < num_stripes; i++) {
3017 multi->stripes[i].physical =
3018 map->stripes[stripe_index].physical +
3019 stripe_offset + stripe_nr * map->stripe_len;
3020 multi->stripes[i].dev = map->stripes[stripe_index].dev;
3022 if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3024 u32 last_stripe = 0;
3027 div_u64_rem(stripe_nr_end - 1,
3031 for (j = 0; j < map->num_stripes; j++) {
3034 div_u64_rem(stripe_nr_end - 1 - j,
3035 map->num_stripes, &test);
3036 if (test == stripe_index)
3039 stripes = stripe_nr_end - 1 - j;
3040 do_div(stripes, map->num_stripes);
3041 multi->stripes[i].length = map->stripe_len *
3042 (stripes - stripe_nr + 1);
3045 multi->stripes[i].length -=
3049 if (stripe_index == last_stripe)
3050 multi->stripes[i].length -=
3052 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3055 int factor = map->num_stripes /
3057 u32 last_stripe = 0;
3059 div_u64_rem(stripe_nr_end - 1,
3060 factor, &last_stripe);
3061 last_stripe *= map->sub_stripes;
3063 for (j = 0; j < factor; j++) {
3066 div_u64_rem(stripe_nr_end - 1 - j,
3070 stripe_index / map->sub_stripes)
3073 stripes = stripe_nr_end - 1 - j;
3074 do_div(stripes, factor);
3075 multi->stripes[i].length = map->stripe_len *
3076 (stripes - stripe_nr + 1);
3078 if (i < map->sub_stripes) {
3079 multi->stripes[i].length -=
3081 if (i == map->sub_stripes - 1)
3084 if (stripe_index >= last_stripe &&
3085 stripe_index <= (last_stripe +
3086 map->sub_stripes - 1)) {
3087 multi->stripes[i].length -=
3091 multi->stripes[i].length = *length;
3094 if (stripe_index == map->num_stripes) {
3095 /* This could only happen for RAID0/10 */
3101 for (i = 0; i < num_stripes; i++) {
3102 multi->stripes[i].physical =
3103 map->stripes[stripe_index].physical +
3105 stripe_nr * map->stripe_len;
3106 multi->stripes[i].dev =
3107 map->stripes[stripe_index].dev;
3113 multi->num_stripes = num_stripes;
3114 multi->max_errors = max_errors;
3117 free_extent_map(em);
3121 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
3122 u64 logical, u64 *length,
3123 struct btrfs_multi_bio **multi_ret, int mirror_num)
3125 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
3129 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
3130 u64 chunk_start, u64 physical, u64 devid,
3131 u64 **logical, int *naddrs, int *stripe_len)
3133 struct extent_map_tree *em_tree = &map_tree->map_tree;
3134 struct extent_map *em;
3135 struct map_lookup *map;
3142 read_lock(&em_tree->lock);
3143 em = lookup_extent_mapping(em_tree, chunk_start, 1);
3144 read_unlock(&em_tree->lock);
3146 BUG_ON(!em || em->start != chunk_start);
3147 map = (struct map_lookup *)em->bdev;
3150 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
3151 do_div(length, map->num_stripes / map->sub_stripes);
3152 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
3153 do_div(length, map->num_stripes);
3155 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
3158 for (i = 0; i < map->num_stripes; i++) {
3159 if (devid && map->stripes[i].dev->devid != devid)
3161 if (map->stripes[i].physical > physical ||
3162 map->stripes[i].physical + length <= physical)
3165 stripe_nr = physical - map->stripes[i].physical;
3166 do_div(stripe_nr, map->stripe_len);
3168 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
3169 stripe_nr = stripe_nr * map->num_stripes + i;
3170 do_div(stripe_nr, map->sub_stripes);
3171 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
3172 stripe_nr = stripe_nr * map->num_stripes + i;
3174 bytenr = chunk_start + stripe_nr * map->stripe_len;
3175 WARN_ON(nr >= map->num_stripes);
3176 for (j = 0; j < nr; j++) {
3177 if (buf[j] == bytenr)
3181 WARN_ON(nr >= map->num_stripes);
3188 *stripe_len = map->stripe_len;
3190 free_extent_map(em);
3194 static void end_bio_multi_stripe(struct bio *bio, int err)
3196 struct btrfs_multi_bio *multi = bio->bi_private;
3197 int is_orig_bio = 0;
3200 atomic_inc(&multi->error);
3202 if (bio == multi->orig_bio)
3205 if (atomic_dec_and_test(&multi->stripes_pending)) {
3208 bio = multi->orig_bio;
3210 bio->bi_private = multi->private;
3211 bio->bi_end_io = multi->end_io;
3212 /* only send an error to the higher layers if it is
3213 * beyond the tolerance of the multi-bio
3215 if (atomic_read(&multi->error) > multi->max_errors) {
3219 * this bio is actually up to date, we didn't
3220 * go over the max number of errors
3222 set_bit(BIO_UPTODATE, &bio->bi_flags);
3227 bio_endio(bio, err);
3228 } else if (!is_orig_bio) {
3233 struct async_sched {
3236 struct btrfs_fs_info *info;
3237 struct btrfs_work work;
3241 * see run_scheduled_bios for a description of why bios are collected for
3244 * This will add one bio to the pending list for a device and make sure
3245 * the work struct is scheduled.
3247 static noinline int schedule_bio(struct btrfs_root *root,
3248 struct btrfs_device *device,
3249 int rw, struct bio *bio)
3251 int should_queue = 1;
3252 struct btrfs_pending_bios *pending_bios;
3254 /* don't bother with additional async steps for reads, right now */
3255 if (!(rw & REQ_WRITE)) {
3257 submit_bio(rw, bio);
3263 * nr_async_bios allows us to reliably return congestion to the
3264 * higher layers. Otherwise, the async bio makes it appear we have
3265 * made progress against dirty pages when we've really just put it
3266 * on a queue for later
3268 atomic_inc(&root->fs_info->nr_async_bios);
3269 WARN_ON(bio->bi_next);
3270 bio->bi_next = NULL;
3273 spin_lock(&device->io_lock);
3274 if (bio->bi_rw & REQ_SYNC)
3275 pending_bios = &device->pending_sync_bios;
3277 pending_bios = &device->pending_bios;
3279 if (pending_bios->tail)
3280 pending_bios->tail->bi_next = bio;
3282 pending_bios->tail = bio;
3283 if (!pending_bios->head)
3284 pending_bios->head = bio;
3285 if (device->running_pending)
3288 spin_unlock(&device->io_lock);
3291 btrfs_queue_worker(&root->fs_info->submit_workers,
3296 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
3297 int mirror_num, int async_submit)
3299 struct btrfs_mapping_tree *map_tree;
3300 struct btrfs_device *dev;
3301 struct bio *first_bio = bio;
3302 u64 logical = (u64)bio->bi_sector << 9;
3305 struct btrfs_multi_bio *multi = NULL;
3310 length = bio->bi_size;
3311 map_tree = &root->fs_info->mapping_tree;
3312 map_length = length;
3314 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3318 total_devs = multi->num_stripes;
3319 if (map_length < length) {
3320 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3321 "len %llu\n", (unsigned long long)logical,
3322 (unsigned long long)length,
3323 (unsigned long long)map_length);
3326 multi->end_io = first_bio->bi_end_io;
3327 multi->private = first_bio->bi_private;
3328 multi->orig_bio = first_bio;
3329 atomic_set(&multi->stripes_pending, multi->num_stripes);
3331 while (dev_nr < total_devs) {
3332 if (total_devs > 1) {
3333 if (dev_nr < total_devs - 1) {
3334 bio = bio_clone(first_bio, GFP_NOFS);
3339 bio->bi_private = multi;
3340 bio->bi_end_io = end_bio_multi_stripe;
3342 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3343 dev = multi->stripes[dev_nr].dev;
3344 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3345 bio->bi_bdev = dev->bdev;
3347 schedule_bio(root, dev, rw, bio);
3349 submit_bio(rw, bio);
3351 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3352 bio->bi_sector = logical >> 9;
3353 bio_endio(bio, -EIO);
3357 if (total_devs == 1)
3362 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3365 struct btrfs_device *device;
3366 struct btrfs_fs_devices *cur_devices;
3368 cur_devices = root->fs_info->fs_devices;
3369 while (cur_devices) {
3371 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3372 device = __find_device(&cur_devices->devices,
3377 cur_devices = cur_devices->seed;
3382 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3383 u64 devid, u8 *dev_uuid)
3385 struct btrfs_device *device;
3386 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3388 device = kzalloc(sizeof(*device), GFP_NOFS);
3391 list_add(&device->dev_list,
3392 &fs_devices->devices);
3393 device->dev_root = root->fs_info->dev_root;
3394 device->devid = devid;
3395 device->work.func = pending_bios_fn;
3396 device->fs_devices = fs_devices;
3397 device->missing = 1;
3398 fs_devices->num_devices++;
3399 fs_devices->missing_devices++;
3400 spin_lock_init(&device->io_lock);
3401 INIT_LIST_HEAD(&device->dev_alloc_list);
3402 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3406 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3407 struct extent_buffer *leaf,
3408 struct btrfs_chunk *chunk)
3410 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3411 struct map_lookup *map;
3412 struct extent_map *em;
3416 u8 uuid[BTRFS_UUID_SIZE];
3421 logical = key->offset;
3422 length = btrfs_chunk_length(leaf, chunk);
3424 read_lock(&map_tree->map_tree.lock);
3425 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3426 read_unlock(&map_tree->map_tree.lock);
3428 /* already mapped? */
3429 if (em && em->start <= logical && em->start + em->len > logical) {
3430 free_extent_map(em);
3433 free_extent_map(em);
3436 em = alloc_extent_map();
3439 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3440 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3442 free_extent_map(em);
3446 em->bdev = (struct block_device *)map;
3447 em->start = logical;
3449 em->block_start = 0;
3450 em->block_len = em->len;
3452 map->num_stripes = num_stripes;
3453 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3454 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3455 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3456 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3457 map->type = btrfs_chunk_type(leaf, chunk);
3458 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3459 for (i = 0; i < num_stripes; i++) {
3460 map->stripes[i].physical =
3461 btrfs_stripe_offset_nr(leaf, chunk, i);
3462 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3463 read_extent_buffer(leaf, uuid, (unsigned long)
3464 btrfs_stripe_dev_uuid_nr(chunk, i),
3466 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3468 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3470 free_extent_map(em);
3473 if (!map->stripes[i].dev) {
3474 map->stripes[i].dev =
3475 add_missing_dev(root, devid, uuid);
3476 if (!map->stripes[i].dev) {
3478 free_extent_map(em);
3482 map->stripes[i].dev->in_fs_metadata = 1;
3485 write_lock(&map_tree->map_tree.lock);
3486 ret = add_extent_mapping(&map_tree->map_tree, em);
3487 write_unlock(&map_tree->map_tree.lock);
3489 free_extent_map(em);
3494 static int fill_device_from_item(struct extent_buffer *leaf,
3495 struct btrfs_dev_item *dev_item,
3496 struct btrfs_device *device)
3500 device->devid = btrfs_device_id(leaf, dev_item);
3501 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3502 device->total_bytes = device->disk_total_bytes;
3503 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3504 device->type = btrfs_device_type(leaf, dev_item);
3505 device->io_align = btrfs_device_io_align(leaf, dev_item);
3506 device->io_width = btrfs_device_io_width(leaf, dev_item);
3507 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3509 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3510 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3515 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3517 struct btrfs_fs_devices *fs_devices;
3520 mutex_lock(&uuid_mutex);
3522 fs_devices = root->fs_info->fs_devices->seed;
3523 while (fs_devices) {
3524 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3528 fs_devices = fs_devices->seed;
3531 fs_devices = find_fsid(fsid);
3537 fs_devices = clone_fs_devices(fs_devices);
3538 if (IS_ERR(fs_devices)) {
3539 ret = PTR_ERR(fs_devices);
3543 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3544 root->fs_info->bdev_holder);
3548 if (!fs_devices->seeding) {
3549 __btrfs_close_devices(fs_devices);
3550 free_fs_devices(fs_devices);
3555 fs_devices->seed = root->fs_info->fs_devices->seed;
3556 root->fs_info->fs_devices->seed = fs_devices;
3558 mutex_unlock(&uuid_mutex);
3562 static int read_one_dev(struct btrfs_root *root,
3563 struct extent_buffer *leaf,
3564 struct btrfs_dev_item *dev_item)
3566 struct btrfs_device *device;
3569 u8 fs_uuid[BTRFS_UUID_SIZE];
3570 u8 dev_uuid[BTRFS_UUID_SIZE];
3572 devid = btrfs_device_id(leaf, dev_item);
3573 read_extent_buffer(leaf, dev_uuid,
3574 (unsigned long)btrfs_device_uuid(dev_item),
3576 read_extent_buffer(leaf, fs_uuid,
3577 (unsigned long)btrfs_device_fsid(dev_item),
3580 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3581 ret = open_seed_devices(root, fs_uuid);
3582 if (ret && !btrfs_test_opt(root, DEGRADED))
3586 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3587 if (!device || !device->bdev) {
3588 if (!btrfs_test_opt(root, DEGRADED))
3592 printk(KERN_WARNING "warning devid %llu missing\n",
3593 (unsigned long long)devid);
3594 device = add_missing_dev(root, devid, dev_uuid);
3597 } else if (!device->missing) {
3599 * this happens when a device that was properly setup
3600 * in the device info lists suddenly goes bad.
3601 * device->bdev is NULL, and so we have to set
3602 * device->missing to one here
3604 root->fs_info->fs_devices->missing_devices++;
3605 device->missing = 1;
3609 if (device->fs_devices != root->fs_info->fs_devices) {
3610 BUG_ON(device->writeable);
3611 if (device->generation !=
3612 btrfs_device_generation(leaf, dev_item))
3616 fill_device_from_item(leaf, dev_item, device);
3617 device->dev_root = root->fs_info->dev_root;
3618 device->in_fs_metadata = 1;
3619 if (device->writeable)
3620 device->fs_devices->total_rw_bytes += device->total_bytes;
3625 int btrfs_read_sys_array(struct btrfs_root *root)
3627 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3628 struct extent_buffer *sb;
3629 struct btrfs_disk_key *disk_key;
3630 struct btrfs_chunk *chunk;
3632 unsigned long sb_ptr;
3638 struct btrfs_key key;
3640 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3641 BTRFS_SUPER_INFO_SIZE);
3644 btrfs_set_buffer_uptodate(sb);
3645 btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
3647 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3648 array_size = btrfs_super_sys_array_size(super_copy);
3650 ptr = super_copy->sys_chunk_array;
3651 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3654 while (cur < array_size) {
3655 disk_key = (struct btrfs_disk_key *)ptr;
3656 btrfs_disk_key_to_cpu(&key, disk_key);
3658 len = sizeof(*disk_key); ptr += len;
3662 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3663 chunk = (struct btrfs_chunk *)sb_ptr;
3664 ret = read_one_chunk(root, &key, sb, chunk);
3667 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3668 len = btrfs_chunk_item_size(num_stripes);
3677 free_extent_buffer(sb);
3681 int btrfs_read_chunk_tree(struct btrfs_root *root)
3683 struct btrfs_path *path;
3684 struct extent_buffer *leaf;
3685 struct btrfs_key key;
3686 struct btrfs_key found_key;
3690 root = root->fs_info->chunk_root;
3692 path = btrfs_alloc_path();
3696 /* first we search for all of the device items, and then we
3697 * read in all of the chunk items. This way we can create chunk
3698 * mappings that reference all of the devices that are afound
3700 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3704 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3708 leaf = path->nodes[0];
3709 slot = path->slots[0];
3710 if (slot >= btrfs_header_nritems(leaf)) {
3711 ret = btrfs_next_leaf(root, path);
3718 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3719 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3720 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3722 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3723 struct btrfs_dev_item *dev_item;
3724 dev_item = btrfs_item_ptr(leaf, slot,
3725 struct btrfs_dev_item);
3726 ret = read_one_dev(root, leaf, dev_item);
3730 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3731 struct btrfs_chunk *chunk;
3732 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3733 ret = read_one_chunk(root, &found_key, leaf, chunk);
3739 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3741 btrfs_release_path(path);
3746 btrfs_free_path(path);