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/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/iocontext.h>
24 #include <asm/div64.h>
27 #include "extent_map.h"
29 #include "transaction.h"
30 #include "print-tree.h"
32 #include "async-thread.h"
42 struct btrfs_bio_stripe stripes[];
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 void btrfs_lock_volumes(void)
58 mutex_lock(&uuid_mutex);
61 void btrfs_unlock_volumes(void)
63 mutex_unlock(&uuid_mutex);
66 static void lock_chunks(struct btrfs_root *root)
68 mutex_lock(&root->fs_info->chunk_mutex);
71 static void unlock_chunks(struct btrfs_root *root)
73 mutex_unlock(&root->fs_info->chunk_mutex);
76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
90 int btrfs_cleanup_fs_uuids(void)
92 struct btrfs_fs_devices *fs_devices;
94 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
98 free_fs_devices(fs_devices);
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
106 struct btrfs_device *dev;
108 list_for_each_entry(dev, head, dev_list) {
109 if (dev->devid == devid &&
110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 struct btrfs_fs_devices *fs_devices;
121 list_for_each_entry(fs_devices, &fs_uuids, list) {
122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
129 struct bio *head, struct bio *tail)
132 struct bio *old_head;
134 old_head = pending_bios->head;
135 pending_bios->head = head;
136 if (pending_bios->tail)
137 tail->bi_next = old_head;
139 pending_bios->tail = tail;
143 * we try to collect pending bios for a device so we don't get a large
144 * number of procs sending bios down to the same device. This greatly
145 * improves the schedulers ability to collect and merge the bios.
147 * But, it also turns into a long list of bios to process and that is sure
148 * to eventually make the worker thread block. The solution here is to
149 * make some progress and then put this work struct back at the end of
150 * the list if the block device is congested. This way, multiple devices
151 * can make progress from a single worker thread.
153 static noinline int run_scheduled_bios(struct btrfs_device *device)
156 struct backing_dev_info *bdi;
157 struct btrfs_fs_info *fs_info;
158 struct btrfs_pending_bios *pending_bios;
162 unsigned long num_run;
163 unsigned long num_sync_run;
164 unsigned long batch_run = 0;
166 unsigned long last_waited = 0;
169 bdi = blk_get_backing_dev_info(device->bdev);
170 fs_info = device->dev_root->fs_info;
171 limit = btrfs_async_submit_limit(fs_info);
172 limit = limit * 2 / 3;
174 /* we want to make sure that every time we switch from the sync
175 * list to the normal list, we unplug
180 spin_lock(&device->io_lock);
185 /* take all the bios off the list at once and process them
186 * later on (without the lock held). But, remember the
187 * tail and other pointers so the bios can be properly reinserted
188 * into the list if we hit congestion
190 if (!force_reg && device->pending_sync_bios.head) {
191 pending_bios = &device->pending_sync_bios;
194 pending_bios = &device->pending_bios;
198 pending = pending_bios->head;
199 tail = pending_bios->tail;
200 WARN_ON(pending && !tail);
203 * if pending was null this time around, no bios need processing
204 * at all and we can stop. Otherwise it'll loop back up again
205 * and do an additional check so no bios are missed.
207 * device->running_pending is used to synchronize with the
210 if (device->pending_sync_bios.head == NULL &&
211 device->pending_bios.head == NULL) {
213 device->running_pending = 0;
216 device->running_pending = 1;
219 pending_bios->head = NULL;
220 pending_bios->tail = NULL;
222 spin_unlock(&device->io_lock);
225 * if we're doing the regular priority list, make sure we unplug
226 * for any high prio bios we've sent down
228 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
230 blk_run_backing_dev(bdi, NULL);
236 /* we want to work on both lists, but do more bios on the
237 * sync list than the regular list
240 pending_bios != &device->pending_sync_bios &&
241 device->pending_sync_bios.head) ||
242 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
243 device->pending_bios.head)) {
244 spin_lock(&device->io_lock);
245 requeue_list(pending_bios, pending, tail);
250 pending = pending->bi_next;
252 atomic_dec(&fs_info->nr_async_bios);
254 if (atomic_read(&fs_info->nr_async_bios) < limit &&
255 waitqueue_active(&fs_info->async_submit_wait))
256 wake_up(&fs_info->async_submit_wait);
258 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
259 submit_bio(cur->bi_rw, cur);
263 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
266 if (need_resched()) {
268 blk_run_backing_dev(bdi, NULL);
275 * we made progress, there is more work to do and the bdi
276 * is now congested. Back off and let other work structs
279 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
280 fs_info->fs_devices->open_devices > 1) {
281 struct io_context *ioc;
283 ioc = current->io_context;
286 * the main goal here is that we don't want to
287 * block if we're going to be able to submit
288 * more requests without blocking.
290 * This code does two great things, it pokes into
291 * the elevator code from a filesystem _and_
292 * it makes assumptions about how batching works.
294 if (ioc && ioc->nr_batch_requests > 0 &&
295 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
297 ioc->last_waited == last_waited)) {
299 * we want to go through our batch of
300 * requests and stop. So, we copy out
301 * the ioc->last_waited time and test
302 * against it before looping
304 last_waited = ioc->last_waited;
305 if (need_resched()) {
307 blk_run_backing_dev(bdi, NULL);
314 spin_lock(&device->io_lock);
315 requeue_list(pending_bios, pending, tail);
316 device->running_pending = 1;
318 spin_unlock(&device->io_lock);
319 btrfs_requeue_work(&device->work);
326 blk_run_backing_dev(bdi, NULL);
333 spin_lock(&device->io_lock);
334 if (device->pending_bios.head || device->pending_sync_bios.head)
336 spin_unlock(&device->io_lock);
339 * IO has already been through a long path to get here. Checksumming,
340 * async helper threads, perhaps compression. We've done a pretty
341 * good job of collecting a batch of IO and should just unplug
342 * the device right away.
344 * This will help anyone who is waiting on the IO, they might have
345 * already unplugged, but managed to do so before the bio they
346 * cared about found its way down here.
348 blk_run_backing_dev(bdi, NULL);
353 static void pending_bios_fn(struct btrfs_work *work)
355 struct btrfs_device *device;
357 device = container_of(work, struct btrfs_device, work);
358 run_scheduled_bios(device);
361 static noinline int device_list_add(const char *path,
362 struct btrfs_super_block *disk_super,
363 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
365 struct btrfs_device *device;
366 struct btrfs_fs_devices *fs_devices;
367 u64 found_transid = btrfs_super_generation(disk_super);
369 fs_devices = find_fsid(disk_super->fsid);
371 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
374 INIT_LIST_HEAD(&fs_devices->devices);
375 INIT_LIST_HEAD(&fs_devices->alloc_list);
376 list_add(&fs_devices->list, &fs_uuids);
377 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
378 fs_devices->latest_devid = devid;
379 fs_devices->latest_trans = found_transid;
380 mutex_init(&fs_devices->device_list_mutex);
383 device = __find_device(&fs_devices->devices, devid,
384 disk_super->dev_item.uuid);
387 if (fs_devices->opened)
390 device = kzalloc(sizeof(*device), GFP_NOFS);
392 /* we can safely leave the fs_devices entry around */
395 device->devid = devid;
396 device->work.func = pending_bios_fn;
397 memcpy(device->uuid, disk_super->dev_item.uuid,
399 device->barriers = 1;
400 spin_lock_init(&device->io_lock);
401 device->name = kstrdup(path, GFP_NOFS);
406 INIT_LIST_HEAD(&device->dev_alloc_list);
408 mutex_lock(&fs_devices->device_list_mutex);
409 list_add(&device->dev_list, &fs_devices->devices);
410 mutex_unlock(&fs_devices->device_list_mutex);
412 device->fs_devices = fs_devices;
413 fs_devices->num_devices++;
416 if (found_transid > fs_devices->latest_trans) {
417 fs_devices->latest_devid = devid;
418 fs_devices->latest_trans = found_transid;
420 *fs_devices_ret = fs_devices;
424 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
426 struct btrfs_fs_devices *fs_devices;
427 struct btrfs_device *device;
428 struct btrfs_device *orig_dev;
430 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
432 return ERR_PTR(-ENOMEM);
434 INIT_LIST_HEAD(&fs_devices->devices);
435 INIT_LIST_HEAD(&fs_devices->alloc_list);
436 INIT_LIST_HEAD(&fs_devices->list);
437 mutex_init(&fs_devices->device_list_mutex);
438 fs_devices->latest_devid = orig->latest_devid;
439 fs_devices->latest_trans = orig->latest_trans;
440 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
442 mutex_lock(&orig->device_list_mutex);
443 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
444 device = kzalloc(sizeof(*device), GFP_NOFS);
448 device->name = kstrdup(orig_dev->name, GFP_NOFS);
452 device->devid = orig_dev->devid;
453 device->work.func = pending_bios_fn;
454 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
455 device->barriers = 1;
456 spin_lock_init(&device->io_lock);
457 INIT_LIST_HEAD(&device->dev_list);
458 INIT_LIST_HEAD(&device->dev_alloc_list);
460 list_add(&device->dev_list, &fs_devices->devices);
461 device->fs_devices = fs_devices;
462 fs_devices->num_devices++;
464 mutex_unlock(&orig->device_list_mutex);
467 mutex_unlock(&orig->device_list_mutex);
468 free_fs_devices(fs_devices);
469 return ERR_PTR(-ENOMEM);
472 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
474 struct btrfs_device *device, *next;
476 mutex_lock(&uuid_mutex);
478 mutex_lock(&fs_devices->device_list_mutex);
479 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
480 if (device->in_fs_metadata)
484 close_bdev_exclusive(device->bdev, device->mode);
486 fs_devices->open_devices--;
488 if (device->writeable) {
489 list_del_init(&device->dev_alloc_list);
490 device->writeable = 0;
491 fs_devices->rw_devices--;
493 list_del_init(&device->dev_list);
494 fs_devices->num_devices--;
498 mutex_unlock(&fs_devices->device_list_mutex);
500 if (fs_devices->seed) {
501 fs_devices = fs_devices->seed;
505 mutex_unlock(&uuid_mutex);
509 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
511 struct btrfs_device *device;
513 if (--fs_devices->opened > 0)
516 list_for_each_entry(device, &fs_devices->devices, dev_list) {
518 close_bdev_exclusive(device->bdev, device->mode);
519 fs_devices->open_devices--;
521 if (device->writeable) {
522 list_del_init(&device->dev_alloc_list);
523 fs_devices->rw_devices--;
527 device->writeable = 0;
528 device->in_fs_metadata = 0;
530 WARN_ON(fs_devices->open_devices);
531 WARN_ON(fs_devices->rw_devices);
532 fs_devices->opened = 0;
533 fs_devices->seeding = 0;
538 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
540 struct btrfs_fs_devices *seed_devices = NULL;
543 mutex_lock(&uuid_mutex);
544 ret = __btrfs_close_devices(fs_devices);
545 if (!fs_devices->opened) {
546 seed_devices = fs_devices->seed;
547 fs_devices->seed = NULL;
549 mutex_unlock(&uuid_mutex);
551 while (seed_devices) {
552 fs_devices = seed_devices;
553 seed_devices = fs_devices->seed;
554 __btrfs_close_devices(fs_devices);
555 free_fs_devices(fs_devices);
560 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
561 fmode_t flags, void *holder)
563 struct block_device *bdev;
564 struct list_head *head = &fs_devices->devices;
565 struct btrfs_device *device;
566 struct block_device *latest_bdev = NULL;
567 struct buffer_head *bh;
568 struct btrfs_super_block *disk_super;
569 u64 latest_devid = 0;
570 u64 latest_transid = 0;
575 list_for_each_entry(device, head, dev_list) {
581 bdev = open_bdev_exclusive(device->name, flags, holder);
583 printk(KERN_INFO "open %s failed\n", device->name);
586 set_blocksize(bdev, 4096);
588 bh = btrfs_read_dev_super(bdev);
592 disk_super = (struct btrfs_super_block *)bh->b_data;
593 devid = le64_to_cpu(disk_super->dev_item.devid);
594 if (devid != device->devid)
597 if (memcmp(device->uuid, disk_super->dev_item.uuid,
601 device->generation = btrfs_super_generation(disk_super);
602 if (!latest_transid || device->generation > latest_transid) {
603 latest_devid = devid;
604 latest_transid = device->generation;
608 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
609 device->writeable = 0;
611 device->writeable = !bdev_read_only(bdev);
616 device->in_fs_metadata = 0;
617 device->mode = flags;
619 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
620 fs_devices->rotating = 1;
622 fs_devices->open_devices++;
623 if (device->writeable) {
624 fs_devices->rw_devices++;
625 list_add(&device->dev_alloc_list,
626 &fs_devices->alloc_list);
633 close_bdev_exclusive(bdev, FMODE_READ);
637 if (fs_devices->open_devices == 0) {
641 fs_devices->seeding = seeding;
642 fs_devices->opened = 1;
643 fs_devices->latest_bdev = latest_bdev;
644 fs_devices->latest_devid = latest_devid;
645 fs_devices->latest_trans = latest_transid;
646 fs_devices->total_rw_bytes = 0;
651 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
652 fmode_t flags, void *holder)
656 mutex_lock(&uuid_mutex);
657 if (fs_devices->opened) {
658 fs_devices->opened++;
661 ret = __btrfs_open_devices(fs_devices, flags, holder);
663 mutex_unlock(&uuid_mutex);
667 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
668 struct btrfs_fs_devices **fs_devices_ret)
670 struct btrfs_super_block *disk_super;
671 struct block_device *bdev;
672 struct buffer_head *bh;
677 mutex_lock(&uuid_mutex);
679 bdev = open_bdev_exclusive(path, flags, holder);
686 ret = set_blocksize(bdev, 4096);
689 bh = btrfs_read_dev_super(bdev);
694 disk_super = (struct btrfs_super_block *)bh->b_data;
695 devid = le64_to_cpu(disk_super->dev_item.devid);
696 transid = btrfs_super_generation(disk_super);
697 if (disk_super->label[0])
698 printk(KERN_INFO "device label %s ", disk_super->label);
700 /* FIXME, make a readl uuid parser */
701 printk(KERN_INFO "device fsid %llx-%llx ",
702 *(unsigned long long *)disk_super->fsid,
703 *(unsigned long long *)(disk_super->fsid + 8));
705 printk(KERN_CONT "devid %llu transid %llu %s\n",
706 (unsigned long long)devid, (unsigned long long)transid, path);
707 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
711 close_bdev_exclusive(bdev, flags);
713 mutex_unlock(&uuid_mutex);
718 * this uses a pretty simple search, the expectation is that it is
719 * called very infrequently and that a given device has a small number
722 int find_free_dev_extent(struct btrfs_trans_handle *trans,
723 struct btrfs_device *device, u64 num_bytes,
724 u64 *start, u64 *max_avail)
726 struct btrfs_key key;
727 struct btrfs_root *root = device->dev_root;
728 struct btrfs_dev_extent *dev_extent = NULL;
729 struct btrfs_path *path;
732 u64 search_start = 0;
733 u64 search_end = device->total_bytes;
737 struct extent_buffer *l;
739 path = btrfs_alloc_path();
745 /* FIXME use last free of some kind */
747 /* we don't want to overwrite the superblock on the drive,
748 * so we make sure to start at an offset of at least 1MB
750 search_start = max((u64)1024 * 1024, search_start);
752 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
753 search_start = max(root->fs_info->alloc_start, search_start);
755 key.objectid = device->devid;
756 key.offset = search_start;
757 key.type = BTRFS_DEV_EXTENT_KEY;
758 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
762 ret = btrfs_previous_item(root, path, key.objectid, key.type);
769 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
772 slot = path->slots[0];
773 if (slot >= btrfs_header_nritems(l)) {
774 ret = btrfs_next_leaf(root, path);
781 if (search_start >= search_end) {
785 *start = search_start;
789 *start = last_byte > search_start ?
790 last_byte : search_start;
791 if (search_end <= *start) {
797 btrfs_item_key_to_cpu(l, &key, slot);
799 if (key.objectid < device->devid)
802 if (key.objectid > device->devid)
805 if (key.offset >= search_start && key.offset > last_byte &&
807 if (last_byte < search_start)
808 last_byte = search_start;
809 hole_size = key.offset - last_byte;
811 if (hole_size > *max_avail)
812 *max_avail = hole_size;
814 if (key.offset > last_byte &&
815 hole_size >= num_bytes) {
820 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
824 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
825 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
831 /* we have to make sure we didn't find an extent that has already
832 * been allocated by the map tree or the original allocation
834 BUG_ON(*start < search_start);
836 if (*start + num_bytes > search_end) {
840 /* check for pending inserts here */
844 btrfs_free_path(path);
848 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
849 struct btrfs_device *device,
853 struct btrfs_path *path;
854 struct btrfs_root *root = device->dev_root;
855 struct btrfs_key key;
856 struct btrfs_key found_key;
857 struct extent_buffer *leaf = NULL;
858 struct btrfs_dev_extent *extent = NULL;
860 path = btrfs_alloc_path();
864 key.objectid = device->devid;
866 key.type = BTRFS_DEV_EXTENT_KEY;
868 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
870 ret = btrfs_previous_item(root, path, key.objectid,
871 BTRFS_DEV_EXTENT_KEY);
873 leaf = path->nodes[0];
874 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
875 extent = btrfs_item_ptr(leaf, path->slots[0],
876 struct btrfs_dev_extent);
877 BUG_ON(found_key.offset > start || found_key.offset +
878 btrfs_dev_extent_length(leaf, extent) < start);
880 } else if (ret == 0) {
881 leaf = path->nodes[0];
882 extent = btrfs_item_ptr(leaf, path->slots[0],
883 struct btrfs_dev_extent);
887 if (device->bytes_used > 0)
888 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
889 ret = btrfs_del_item(trans, root, path);
892 btrfs_free_path(path);
896 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
897 struct btrfs_device *device,
898 u64 chunk_tree, u64 chunk_objectid,
899 u64 chunk_offset, u64 start, u64 num_bytes)
902 struct btrfs_path *path;
903 struct btrfs_root *root = device->dev_root;
904 struct btrfs_dev_extent *extent;
905 struct extent_buffer *leaf;
906 struct btrfs_key key;
908 WARN_ON(!device->in_fs_metadata);
909 path = btrfs_alloc_path();
913 key.objectid = device->devid;
915 key.type = BTRFS_DEV_EXTENT_KEY;
916 ret = btrfs_insert_empty_item(trans, root, path, &key,
920 leaf = path->nodes[0];
921 extent = btrfs_item_ptr(leaf, path->slots[0],
922 struct btrfs_dev_extent);
923 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
924 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
925 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
927 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
928 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
931 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
932 btrfs_mark_buffer_dirty(leaf);
933 btrfs_free_path(path);
937 static noinline int find_next_chunk(struct btrfs_root *root,
938 u64 objectid, u64 *offset)
940 struct btrfs_path *path;
942 struct btrfs_key key;
943 struct btrfs_chunk *chunk;
944 struct btrfs_key found_key;
946 path = btrfs_alloc_path();
949 key.objectid = objectid;
950 key.offset = (u64)-1;
951 key.type = BTRFS_CHUNK_ITEM_KEY;
953 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
959 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
963 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
965 if (found_key.objectid != objectid)
968 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
970 *offset = found_key.offset +
971 btrfs_chunk_length(path->nodes[0], chunk);
976 btrfs_free_path(path);
980 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
983 struct btrfs_key key;
984 struct btrfs_key found_key;
985 struct btrfs_path *path;
987 root = root->fs_info->chunk_root;
989 path = btrfs_alloc_path();
993 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
994 key.type = BTRFS_DEV_ITEM_KEY;
995 key.offset = (u64)-1;
997 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1003 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1004 BTRFS_DEV_ITEM_KEY);
1008 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1010 *objectid = found_key.offset + 1;
1014 btrfs_free_path(path);
1019 * the device information is stored in the chunk root
1020 * the btrfs_device struct should be fully filled in
1022 int btrfs_add_device(struct btrfs_trans_handle *trans,
1023 struct btrfs_root *root,
1024 struct btrfs_device *device)
1027 struct btrfs_path *path;
1028 struct btrfs_dev_item *dev_item;
1029 struct extent_buffer *leaf;
1030 struct btrfs_key key;
1033 root = root->fs_info->chunk_root;
1035 path = btrfs_alloc_path();
1039 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1040 key.type = BTRFS_DEV_ITEM_KEY;
1041 key.offset = device->devid;
1043 ret = btrfs_insert_empty_item(trans, root, path, &key,
1048 leaf = path->nodes[0];
1049 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1051 btrfs_set_device_id(leaf, dev_item, device->devid);
1052 btrfs_set_device_generation(leaf, dev_item, 0);
1053 btrfs_set_device_type(leaf, dev_item, device->type);
1054 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1055 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1056 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1057 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1058 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1059 btrfs_set_device_group(leaf, dev_item, 0);
1060 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1061 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1062 btrfs_set_device_start_offset(leaf, dev_item, 0);
1064 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1065 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1066 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1067 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1068 btrfs_mark_buffer_dirty(leaf);
1072 btrfs_free_path(path);
1076 static int btrfs_rm_dev_item(struct btrfs_root *root,
1077 struct btrfs_device *device)
1080 struct btrfs_path *path;
1081 struct btrfs_key key;
1082 struct btrfs_trans_handle *trans;
1084 root = root->fs_info->chunk_root;
1086 path = btrfs_alloc_path();
1090 trans = btrfs_start_transaction(root, 1);
1091 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1092 key.type = BTRFS_DEV_ITEM_KEY;
1093 key.offset = device->devid;
1096 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1105 ret = btrfs_del_item(trans, root, path);
1109 btrfs_free_path(path);
1110 unlock_chunks(root);
1111 btrfs_commit_transaction(trans, root);
1115 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1117 struct btrfs_device *device;
1118 struct btrfs_device *next_device;
1119 struct block_device *bdev;
1120 struct buffer_head *bh = NULL;
1121 struct btrfs_super_block *disk_super;
1128 mutex_lock(&uuid_mutex);
1129 mutex_lock(&root->fs_info->volume_mutex);
1131 all_avail = root->fs_info->avail_data_alloc_bits |
1132 root->fs_info->avail_system_alloc_bits |
1133 root->fs_info->avail_metadata_alloc_bits;
1135 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1136 root->fs_info->fs_devices->rw_devices <= 4) {
1137 printk(KERN_ERR "btrfs: unable to go below four devices "
1143 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1144 root->fs_info->fs_devices->rw_devices <= 2) {
1145 printk(KERN_ERR "btrfs: unable to go below two "
1146 "devices on raid1\n");
1151 if (strcmp(device_path, "missing") == 0) {
1152 struct list_head *devices;
1153 struct btrfs_device *tmp;
1156 devices = &root->fs_info->fs_devices->devices;
1157 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1158 list_for_each_entry(tmp, devices, dev_list) {
1159 if (tmp->in_fs_metadata && !tmp->bdev) {
1164 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1169 printk(KERN_ERR "btrfs: no missing devices found to "
1174 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1175 root->fs_info->bdev_holder);
1177 ret = PTR_ERR(bdev);
1181 set_blocksize(bdev, 4096);
1182 bh = btrfs_read_dev_super(bdev);
1187 disk_super = (struct btrfs_super_block *)bh->b_data;
1188 devid = le64_to_cpu(disk_super->dev_item.devid);
1189 dev_uuid = disk_super->dev_item.uuid;
1190 device = btrfs_find_device(root, devid, dev_uuid,
1198 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1199 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1205 if (device->writeable) {
1206 list_del_init(&device->dev_alloc_list);
1207 root->fs_info->fs_devices->rw_devices--;
1210 ret = btrfs_shrink_device(device, 0);
1214 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1218 device->in_fs_metadata = 0;
1221 * the device list mutex makes sure that we don't change
1222 * the device list while someone else is writing out all
1223 * the device supers.
1225 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1226 list_del_init(&device->dev_list);
1227 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1229 device->fs_devices->num_devices--;
1231 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1232 struct btrfs_device, dev_list);
1233 if (device->bdev == root->fs_info->sb->s_bdev)
1234 root->fs_info->sb->s_bdev = next_device->bdev;
1235 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1236 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1239 close_bdev_exclusive(device->bdev, device->mode);
1240 device->bdev = NULL;
1241 device->fs_devices->open_devices--;
1244 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1245 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1247 if (device->fs_devices->open_devices == 0) {
1248 struct btrfs_fs_devices *fs_devices;
1249 fs_devices = root->fs_info->fs_devices;
1250 while (fs_devices) {
1251 if (fs_devices->seed == device->fs_devices)
1253 fs_devices = fs_devices->seed;
1255 fs_devices->seed = device->fs_devices->seed;
1256 device->fs_devices->seed = NULL;
1257 __btrfs_close_devices(device->fs_devices);
1258 free_fs_devices(device->fs_devices);
1262 * at this point, the device is zero sized. We want to
1263 * remove it from the devices list and zero out the old super
1265 if (device->writeable) {
1266 /* make sure this device isn't detected as part of
1269 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1270 set_buffer_dirty(bh);
1271 sync_dirty_buffer(bh);
1274 kfree(device->name);
1282 close_bdev_exclusive(bdev, FMODE_READ);
1284 mutex_unlock(&root->fs_info->volume_mutex);
1285 mutex_unlock(&uuid_mutex);
1290 * does all the dirty work required for changing file system's UUID.
1292 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1293 struct btrfs_root *root)
1295 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1296 struct btrfs_fs_devices *old_devices;
1297 struct btrfs_fs_devices *seed_devices;
1298 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1299 struct btrfs_device *device;
1302 BUG_ON(!mutex_is_locked(&uuid_mutex));
1303 if (!fs_devices->seeding)
1306 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1310 old_devices = clone_fs_devices(fs_devices);
1311 if (IS_ERR(old_devices)) {
1312 kfree(seed_devices);
1313 return PTR_ERR(old_devices);
1316 list_add(&old_devices->list, &fs_uuids);
1318 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1319 seed_devices->opened = 1;
1320 INIT_LIST_HEAD(&seed_devices->devices);
1321 INIT_LIST_HEAD(&seed_devices->alloc_list);
1322 mutex_init(&seed_devices->device_list_mutex);
1323 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1324 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1325 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1326 device->fs_devices = seed_devices;
1329 fs_devices->seeding = 0;
1330 fs_devices->num_devices = 0;
1331 fs_devices->open_devices = 0;
1332 fs_devices->seed = seed_devices;
1334 generate_random_uuid(fs_devices->fsid);
1335 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1336 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1337 super_flags = btrfs_super_flags(disk_super) &
1338 ~BTRFS_SUPER_FLAG_SEEDING;
1339 btrfs_set_super_flags(disk_super, super_flags);
1345 * strore the expected generation for seed devices in device items.
1347 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1348 struct btrfs_root *root)
1350 struct btrfs_path *path;
1351 struct extent_buffer *leaf;
1352 struct btrfs_dev_item *dev_item;
1353 struct btrfs_device *device;
1354 struct btrfs_key key;
1355 u8 fs_uuid[BTRFS_UUID_SIZE];
1356 u8 dev_uuid[BTRFS_UUID_SIZE];
1360 path = btrfs_alloc_path();
1364 root = root->fs_info->chunk_root;
1365 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1367 key.type = BTRFS_DEV_ITEM_KEY;
1370 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1374 leaf = path->nodes[0];
1376 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1377 ret = btrfs_next_leaf(root, path);
1382 leaf = path->nodes[0];
1383 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1384 btrfs_release_path(root, path);
1388 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1389 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1390 key.type != BTRFS_DEV_ITEM_KEY)
1393 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1394 struct btrfs_dev_item);
1395 devid = btrfs_device_id(leaf, dev_item);
1396 read_extent_buffer(leaf, dev_uuid,
1397 (unsigned long)btrfs_device_uuid(dev_item),
1399 read_extent_buffer(leaf, fs_uuid,
1400 (unsigned long)btrfs_device_fsid(dev_item),
1402 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1405 if (device->fs_devices->seeding) {
1406 btrfs_set_device_generation(leaf, dev_item,
1407 device->generation);
1408 btrfs_mark_buffer_dirty(leaf);
1416 btrfs_free_path(path);
1420 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1422 struct btrfs_trans_handle *trans;
1423 struct btrfs_device *device;
1424 struct block_device *bdev;
1425 struct list_head *devices;
1426 struct super_block *sb = root->fs_info->sb;
1428 int seeding_dev = 0;
1431 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1434 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1438 if (root->fs_info->fs_devices->seeding) {
1440 down_write(&sb->s_umount);
1441 mutex_lock(&uuid_mutex);
1444 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1445 mutex_lock(&root->fs_info->volume_mutex);
1447 devices = &root->fs_info->fs_devices->devices;
1449 * we have the volume lock, so we don't need the extra
1450 * device list mutex while reading the list here.
1452 list_for_each_entry(device, devices, dev_list) {
1453 if (device->bdev == bdev) {
1459 device = kzalloc(sizeof(*device), GFP_NOFS);
1461 /* we can safely leave the fs_devices entry around */
1466 device->name = kstrdup(device_path, GFP_NOFS);
1467 if (!device->name) {
1473 ret = find_next_devid(root, &device->devid);
1479 trans = btrfs_start_transaction(root, 1);
1482 device->barriers = 1;
1483 device->writeable = 1;
1484 device->work.func = pending_bios_fn;
1485 generate_random_uuid(device->uuid);
1486 spin_lock_init(&device->io_lock);
1487 device->generation = trans->transid;
1488 device->io_width = root->sectorsize;
1489 device->io_align = root->sectorsize;
1490 device->sector_size = root->sectorsize;
1491 device->total_bytes = i_size_read(bdev->bd_inode);
1492 device->disk_total_bytes = device->total_bytes;
1493 device->dev_root = root->fs_info->dev_root;
1494 device->bdev = bdev;
1495 device->in_fs_metadata = 1;
1497 set_blocksize(device->bdev, 4096);
1500 sb->s_flags &= ~MS_RDONLY;
1501 ret = btrfs_prepare_sprout(trans, root);
1505 device->fs_devices = root->fs_info->fs_devices;
1508 * we don't want write_supers to jump in here with our device
1511 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1512 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1513 list_add(&device->dev_alloc_list,
1514 &root->fs_info->fs_devices->alloc_list);
1515 root->fs_info->fs_devices->num_devices++;
1516 root->fs_info->fs_devices->open_devices++;
1517 root->fs_info->fs_devices->rw_devices++;
1518 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1520 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1521 root->fs_info->fs_devices->rotating = 1;
1523 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1524 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1525 total_bytes + device->total_bytes);
1527 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1528 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1530 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1533 ret = init_first_rw_device(trans, root, device);
1535 ret = btrfs_finish_sprout(trans, root);
1538 ret = btrfs_add_device(trans, root, device);
1542 * we've got more storage, clear any full flags on the space
1545 btrfs_clear_space_info_full(root->fs_info);
1547 unlock_chunks(root);
1548 btrfs_commit_transaction(trans, root);
1551 mutex_unlock(&uuid_mutex);
1552 up_write(&sb->s_umount);
1554 ret = btrfs_relocate_sys_chunks(root);
1558 mutex_unlock(&root->fs_info->volume_mutex);
1561 close_bdev_exclusive(bdev, 0);
1563 mutex_unlock(&uuid_mutex);
1564 up_write(&sb->s_umount);
1569 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1570 struct btrfs_device *device)
1573 struct btrfs_path *path;
1574 struct btrfs_root *root;
1575 struct btrfs_dev_item *dev_item;
1576 struct extent_buffer *leaf;
1577 struct btrfs_key key;
1579 root = device->dev_root->fs_info->chunk_root;
1581 path = btrfs_alloc_path();
1585 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1586 key.type = BTRFS_DEV_ITEM_KEY;
1587 key.offset = device->devid;
1589 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1598 leaf = path->nodes[0];
1599 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1601 btrfs_set_device_id(leaf, dev_item, device->devid);
1602 btrfs_set_device_type(leaf, dev_item, device->type);
1603 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1604 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1605 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1606 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1607 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1608 btrfs_mark_buffer_dirty(leaf);
1611 btrfs_free_path(path);
1615 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1616 struct btrfs_device *device, u64 new_size)
1618 struct btrfs_super_block *super_copy =
1619 &device->dev_root->fs_info->super_copy;
1620 u64 old_total = btrfs_super_total_bytes(super_copy);
1621 u64 diff = new_size - device->total_bytes;
1623 if (!device->writeable)
1625 if (new_size <= device->total_bytes)
1628 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1629 device->fs_devices->total_rw_bytes += diff;
1631 device->total_bytes = new_size;
1632 device->disk_total_bytes = new_size;
1633 btrfs_clear_space_info_full(device->dev_root->fs_info);
1635 return btrfs_update_device(trans, device);
1638 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1639 struct btrfs_device *device, u64 new_size)
1642 lock_chunks(device->dev_root);
1643 ret = __btrfs_grow_device(trans, device, new_size);
1644 unlock_chunks(device->dev_root);
1648 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1649 struct btrfs_root *root,
1650 u64 chunk_tree, u64 chunk_objectid,
1654 struct btrfs_path *path;
1655 struct btrfs_key key;
1657 root = root->fs_info->chunk_root;
1658 path = btrfs_alloc_path();
1662 key.objectid = chunk_objectid;
1663 key.offset = chunk_offset;
1664 key.type = BTRFS_CHUNK_ITEM_KEY;
1666 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1669 ret = btrfs_del_item(trans, root, path);
1672 btrfs_free_path(path);
1676 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1679 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1680 struct btrfs_disk_key *disk_key;
1681 struct btrfs_chunk *chunk;
1688 struct btrfs_key key;
1690 array_size = btrfs_super_sys_array_size(super_copy);
1692 ptr = super_copy->sys_chunk_array;
1695 while (cur < array_size) {
1696 disk_key = (struct btrfs_disk_key *)ptr;
1697 btrfs_disk_key_to_cpu(&key, disk_key);
1699 len = sizeof(*disk_key);
1701 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1702 chunk = (struct btrfs_chunk *)(ptr + len);
1703 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1704 len += btrfs_chunk_item_size(num_stripes);
1709 if (key.objectid == chunk_objectid &&
1710 key.offset == chunk_offset) {
1711 memmove(ptr, ptr + len, array_size - (cur + len));
1713 btrfs_set_super_sys_array_size(super_copy, array_size);
1722 static int btrfs_relocate_chunk(struct btrfs_root *root,
1723 u64 chunk_tree, u64 chunk_objectid,
1726 struct extent_map_tree *em_tree;
1727 struct btrfs_root *extent_root;
1728 struct btrfs_trans_handle *trans;
1729 struct extent_map *em;
1730 struct map_lookup *map;
1734 root = root->fs_info->chunk_root;
1735 extent_root = root->fs_info->extent_root;
1736 em_tree = &root->fs_info->mapping_tree.map_tree;
1738 ret = btrfs_can_relocate(extent_root, chunk_offset);
1742 /* step one, relocate all the extents inside this chunk */
1743 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1746 trans = btrfs_start_transaction(root, 1);
1752 * step two, delete the device extents and the
1753 * chunk tree entries
1755 read_lock(&em_tree->lock);
1756 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1757 read_unlock(&em_tree->lock);
1759 BUG_ON(em->start > chunk_offset ||
1760 em->start + em->len < chunk_offset);
1761 map = (struct map_lookup *)em->bdev;
1763 for (i = 0; i < map->num_stripes; i++) {
1764 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1765 map->stripes[i].physical);
1768 if (map->stripes[i].dev) {
1769 ret = btrfs_update_device(trans, map->stripes[i].dev);
1773 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1778 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1779 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1783 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1786 write_lock(&em_tree->lock);
1787 remove_extent_mapping(em_tree, em);
1788 write_unlock(&em_tree->lock);
1793 /* once for the tree */
1794 free_extent_map(em);
1796 free_extent_map(em);
1798 unlock_chunks(root);
1799 btrfs_end_transaction(trans, root);
1803 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1805 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1806 struct btrfs_path *path;
1807 struct extent_buffer *leaf;
1808 struct btrfs_chunk *chunk;
1809 struct btrfs_key key;
1810 struct btrfs_key found_key;
1811 u64 chunk_tree = chunk_root->root_key.objectid;
1813 bool retried = false;
1817 path = btrfs_alloc_path();
1822 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1823 key.offset = (u64)-1;
1824 key.type = BTRFS_CHUNK_ITEM_KEY;
1827 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1832 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1839 leaf = path->nodes[0];
1840 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1842 chunk = btrfs_item_ptr(leaf, path->slots[0],
1843 struct btrfs_chunk);
1844 chunk_type = btrfs_chunk_type(leaf, chunk);
1845 btrfs_release_path(chunk_root, path);
1847 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1848 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1857 if (found_key.offset == 0)
1859 key.offset = found_key.offset - 1;
1862 if (failed && !retried) {
1866 } else if (failed && retried) {
1871 btrfs_free_path(path);
1875 static u64 div_factor(u64 num, int factor)
1884 int btrfs_balance(struct btrfs_root *dev_root)
1887 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1888 struct btrfs_device *device;
1891 struct btrfs_path *path;
1892 struct btrfs_key key;
1893 struct btrfs_chunk *chunk;
1894 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1895 struct btrfs_trans_handle *trans;
1896 struct btrfs_key found_key;
1898 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1901 mutex_lock(&dev_root->fs_info->volume_mutex);
1902 dev_root = dev_root->fs_info->dev_root;
1904 /* step one make some room on all the devices */
1905 list_for_each_entry(device, devices, dev_list) {
1906 old_size = device->total_bytes;
1907 size_to_free = div_factor(old_size, 1);
1908 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1909 if (!device->writeable ||
1910 device->total_bytes - device->bytes_used > size_to_free)
1913 ret = btrfs_shrink_device(device, old_size - size_to_free);
1918 trans = btrfs_start_transaction(dev_root, 1);
1921 ret = btrfs_grow_device(trans, device, old_size);
1924 btrfs_end_transaction(trans, dev_root);
1927 /* step two, relocate all the chunks */
1928 path = btrfs_alloc_path();
1931 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1932 key.offset = (u64)-1;
1933 key.type = BTRFS_CHUNK_ITEM_KEY;
1936 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1941 * this shouldn't happen, it means the last relocate
1947 ret = btrfs_previous_item(chunk_root, path, 0,
1948 BTRFS_CHUNK_ITEM_KEY);
1952 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1954 if (found_key.objectid != key.objectid)
1957 chunk = btrfs_item_ptr(path->nodes[0],
1959 struct btrfs_chunk);
1960 /* chunk zero is special */
1961 if (found_key.offset == 0)
1964 btrfs_release_path(chunk_root, path);
1965 ret = btrfs_relocate_chunk(chunk_root,
1966 chunk_root->root_key.objectid,
1969 BUG_ON(ret && ret != -ENOSPC);
1970 key.offset = found_key.offset - 1;
1974 btrfs_free_path(path);
1975 mutex_unlock(&dev_root->fs_info->volume_mutex);
1980 * shrinking a device means finding all of the device extents past
1981 * the new size, and then following the back refs to the chunks.
1982 * The chunk relocation code actually frees the device extent
1984 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1986 struct btrfs_trans_handle *trans;
1987 struct btrfs_root *root = device->dev_root;
1988 struct btrfs_dev_extent *dev_extent = NULL;
1989 struct btrfs_path *path;
1997 bool retried = false;
1998 struct extent_buffer *l;
1999 struct btrfs_key key;
2000 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2001 u64 old_total = btrfs_super_total_bytes(super_copy);
2002 u64 old_size = device->total_bytes;
2003 u64 diff = device->total_bytes - new_size;
2005 if (new_size >= device->total_bytes)
2008 path = btrfs_alloc_path();
2016 device->total_bytes = new_size;
2017 if (device->writeable)
2018 device->fs_devices->total_rw_bytes -= diff;
2019 unlock_chunks(root);
2022 key.objectid = device->devid;
2023 key.offset = (u64)-1;
2024 key.type = BTRFS_DEV_EXTENT_KEY;
2027 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2031 ret = btrfs_previous_item(root, path, 0, key.type);
2036 btrfs_release_path(root, path);
2041 slot = path->slots[0];
2042 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2044 if (key.objectid != device->devid) {
2045 btrfs_release_path(root, path);
2049 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2050 length = btrfs_dev_extent_length(l, dev_extent);
2052 if (key.offset + length <= new_size) {
2053 btrfs_release_path(root, path);
2057 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2058 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2059 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2060 btrfs_release_path(root, path);
2062 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2064 if (ret && ret != -ENOSPC)
2071 if (failed && !retried) {
2075 } else if (failed && retried) {
2079 device->total_bytes = old_size;
2080 if (device->writeable)
2081 device->fs_devices->total_rw_bytes += diff;
2082 unlock_chunks(root);
2086 /* Shrinking succeeded, else we would be at "done". */
2087 trans = btrfs_start_transaction(root, 1);
2094 device->disk_total_bytes = new_size;
2095 /* Now btrfs_update_device() will change the on-disk size. */
2096 ret = btrfs_update_device(trans, device);
2098 unlock_chunks(root);
2099 btrfs_end_transaction(trans, root);
2102 WARN_ON(diff > old_total);
2103 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2104 unlock_chunks(root);
2105 btrfs_end_transaction(trans, root);
2107 btrfs_free_path(path);
2111 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2112 struct btrfs_root *root,
2113 struct btrfs_key *key,
2114 struct btrfs_chunk *chunk, int item_size)
2116 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2117 struct btrfs_disk_key disk_key;
2121 array_size = btrfs_super_sys_array_size(super_copy);
2122 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2125 ptr = super_copy->sys_chunk_array + array_size;
2126 btrfs_cpu_key_to_disk(&disk_key, key);
2127 memcpy(ptr, &disk_key, sizeof(disk_key));
2128 ptr += sizeof(disk_key);
2129 memcpy(ptr, chunk, item_size);
2130 item_size += sizeof(disk_key);
2131 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2135 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2136 int num_stripes, int sub_stripes)
2138 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2140 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2141 return calc_size * (num_stripes / sub_stripes);
2143 return calc_size * num_stripes;
2146 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2147 struct btrfs_root *extent_root,
2148 struct map_lookup **map_ret,
2149 u64 *num_bytes, u64 *stripe_size,
2150 u64 start, u64 type)
2152 struct btrfs_fs_info *info = extent_root->fs_info;
2153 struct btrfs_device *device = NULL;
2154 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2155 struct list_head *cur;
2156 struct map_lookup *map = NULL;
2157 struct extent_map_tree *em_tree;
2158 struct extent_map *em;
2159 struct list_head private_devs;
2160 int min_stripe_size = 1 * 1024 * 1024;
2161 u64 calc_size = 1024 * 1024 * 1024;
2162 u64 max_chunk_size = calc_size;
2167 int num_stripes = 1;
2168 int min_stripes = 1;
2169 int sub_stripes = 0;
2173 int stripe_len = 64 * 1024;
2175 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2176 (type & BTRFS_BLOCK_GROUP_DUP)) {
2178 type &= ~BTRFS_BLOCK_GROUP_DUP;
2180 if (list_empty(&fs_devices->alloc_list))
2183 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2184 num_stripes = fs_devices->rw_devices;
2187 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2191 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2192 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2193 if (num_stripes < 2)
2197 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2198 num_stripes = fs_devices->rw_devices;
2199 if (num_stripes < 4)
2201 num_stripes &= ~(u32)1;
2206 if (type & BTRFS_BLOCK_GROUP_DATA) {
2207 max_chunk_size = 10 * calc_size;
2208 min_stripe_size = 64 * 1024 * 1024;
2209 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2210 max_chunk_size = 4 * calc_size;
2211 min_stripe_size = 32 * 1024 * 1024;
2212 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2213 calc_size = 8 * 1024 * 1024;
2214 max_chunk_size = calc_size * 2;
2215 min_stripe_size = 1 * 1024 * 1024;
2218 /* we don't want a chunk larger than 10% of writeable space */
2219 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2224 if (!map || map->num_stripes != num_stripes) {
2226 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2229 map->num_stripes = num_stripes;
2232 if (calc_size * num_stripes > max_chunk_size) {
2233 calc_size = max_chunk_size;
2234 do_div(calc_size, num_stripes);
2235 do_div(calc_size, stripe_len);
2236 calc_size *= stripe_len;
2238 /* we don't want tiny stripes */
2239 calc_size = max_t(u64, min_stripe_size, calc_size);
2241 do_div(calc_size, stripe_len);
2242 calc_size *= stripe_len;
2244 cur = fs_devices->alloc_list.next;
2247 if (type & BTRFS_BLOCK_GROUP_DUP)
2248 min_free = calc_size * 2;
2250 min_free = calc_size;
2253 * we add 1MB because we never use the first 1MB of the device, unless
2254 * we've looped, then we are likely allocating the maximum amount of
2255 * space left already
2258 min_free += 1024 * 1024;
2260 INIT_LIST_HEAD(&private_devs);
2261 while (index < num_stripes) {
2262 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2263 BUG_ON(!device->writeable);
2264 if (device->total_bytes > device->bytes_used)
2265 avail = device->total_bytes - device->bytes_used;
2270 if (device->in_fs_metadata && avail >= min_free) {
2271 ret = find_free_dev_extent(trans, device,
2272 min_free, &dev_offset,
2275 list_move_tail(&device->dev_alloc_list,
2277 map->stripes[index].dev = device;
2278 map->stripes[index].physical = dev_offset;
2280 if (type & BTRFS_BLOCK_GROUP_DUP) {
2281 map->stripes[index].dev = device;
2282 map->stripes[index].physical =
2283 dev_offset + calc_size;
2287 } else if (device->in_fs_metadata && avail > max_avail)
2289 if (cur == &fs_devices->alloc_list)
2292 list_splice(&private_devs, &fs_devices->alloc_list);
2293 if (index < num_stripes) {
2294 if (index >= min_stripes) {
2295 num_stripes = index;
2296 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2297 num_stripes /= sub_stripes;
2298 num_stripes *= sub_stripes;
2303 if (!looped && max_avail > 0) {
2305 calc_size = max_avail;
2311 map->sector_size = extent_root->sectorsize;
2312 map->stripe_len = stripe_len;
2313 map->io_align = stripe_len;
2314 map->io_width = stripe_len;
2316 map->num_stripes = num_stripes;
2317 map->sub_stripes = sub_stripes;
2320 *stripe_size = calc_size;
2321 *num_bytes = chunk_bytes_by_type(type, calc_size,
2322 num_stripes, sub_stripes);
2324 em = alloc_extent_map(GFP_NOFS);
2329 em->bdev = (struct block_device *)map;
2331 em->len = *num_bytes;
2332 em->block_start = 0;
2333 em->block_len = em->len;
2335 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2336 write_lock(&em_tree->lock);
2337 ret = add_extent_mapping(em_tree, em);
2338 write_unlock(&em_tree->lock);
2340 free_extent_map(em);
2342 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2343 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2348 while (index < map->num_stripes) {
2349 device = map->stripes[index].dev;
2350 dev_offset = map->stripes[index].physical;
2352 ret = btrfs_alloc_dev_extent(trans, device,
2353 info->chunk_root->root_key.objectid,
2354 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2355 start, dev_offset, calc_size);
2363 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2364 struct btrfs_root *extent_root,
2365 struct map_lookup *map, u64 chunk_offset,
2366 u64 chunk_size, u64 stripe_size)
2369 struct btrfs_key key;
2370 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2371 struct btrfs_device *device;
2372 struct btrfs_chunk *chunk;
2373 struct btrfs_stripe *stripe;
2374 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2378 chunk = kzalloc(item_size, GFP_NOFS);
2383 while (index < map->num_stripes) {
2384 device = map->stripes[index].dev;
2385 device->bytes_used += stripe_size;
2386 ret = btrfs_update_device(trans, device);
2392 stripe = &chunk->stripe;
2393 while (index < map->num_stripes) {
2394 device = map->stripes[index].dev;
2395 dev_offset = map->stripes[index].physical;
2397 btrfs_set_stack_stripe_devid(stripe, device->devid);
2398 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2399 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2404 btrfs_set_stack_chunk_length(chunk, chunk_size);
2405 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2406 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2407 btrfs_set_stack_chunk_type(chunk, map->type);
2408 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2409 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2410 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2411 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2412 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2414 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2415 key.type = BTRFS_CHUNK_ITEM_KEY;
2416 key.offset = chunk_offset;
2418 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2421 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2422 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2431 * Chunk allocation falls into two parts. The first part does works
2432 * that make the new allocated chunk useable, but not do any operation
2433 * that modifies the chunk tree. The second part does the works that
2434 * require modifying the chunk tree. This division is important for the
2435 * bootstrap process of adding storage to a seed btrfs.
2437 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2438 struct btrfs_root *extent_root, u64 type)
2443 struct map_lookup *map;
2444 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2447 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2452 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2453 &stripe_size, chunk_offset, type);
2457 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2458 chunk_size, stripe_size);
2463 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2464 struct btrfs_root *root,
2465 struct btrfs_device *device)
2468 u64 sys_chunk_offset;
2472 u64 sys_stripe_size;
2474 struct map_lookup *map;
2475 struct map_lookup *sys_map;
2476 struct btrfs_fs_info *fs_info = root->fs_info;
2477 struct btrfs_root *extent_root = fs_info->extent_root;
2480 ret = find_next_chunk(fs_info->chunk_root,
2481 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2484 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2485 (fs_info->metadata_alloc_profile &
2486 fs_info->avail_metadata_alloc_bits);
2487 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2489 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2490 &stripe_size, chunk_offset, alloc_profile);
2493 sys_chunk_offset = chunk_offset + chunk_size;
2495 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2496 (fs_info->system_alloc_profile &
2497 fs_info->avail_system_alloc_bits);
2498 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2500 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2501 &sys_chunk_size, &sys_stripe_size,
2502 sys_chunk_offset, alloc_profile);
2505 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2509 * Modifying chunk tree needs allocating new blocks from both
2510 * system block group and metadata block group. So we only can
2511 * do operations require modifying the chunk tree after both
2512 * block groups were created.
2514 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2515 chunk_size, stripe_size);
2518 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2519 sys_chunk_offset, sys_chunk_size,
2525 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2527 struct extent_map *em;
2528 struct map_lookup *map;
2529 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2533 read_lock(&map_tree->map_tree.lock);
2534 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2535 read_unlock(&map_tree->map_tree.lock);
2539 map = (struct map_lookup *)em->bdev;
2540 for (i = 0; i < map->num_stripes; i++) {
2541 if (!map->stripes[i].dev->writeable) {
2546 free_extent_map(em);
2550 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2552 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2555 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2557 struct extent_map *em;
2560 write_lock(&tree->map_tree.lock);
2561 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2563 remove_extent_mapping(&tree->map_tree, em);
2564 write_unlock(&tree->map_tree.lock);
2569 free_extent_map(em);
2570 /* once for the tree */
2571 free_extent_map(em);
2575 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2577 struct extent_map *em;
2578 struct map_lookup *map;
2579 struct extent_map_tree *em_tree = &map_tree->map_tree;
2582 read_lock(&em_tree->lock);
2583 em = lookup_extent_mapping(em_tree, logical, len);
2584 read_unlock(&em_tree->lock);
2587 BUG_ON(em->start > logical || em->start + em->len < logical);
2588 map = (struct map_lookup *)em->bdev;
2589 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2590 ret = map->num_stripes;
2591 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2592 ret = map->sub_stripes;
2595 free_extent_map(em);
2599 static int find_live_mirror(struct map_lookup *map, int first, int num,
2603 if (map->stripes[optimal].dev->bdev)
2605 for (i = first; i < first + num; i++) {
2606 if (map->stripes[i].dev->bdev)
2609 /* we couldn't find one that doesn't fail. Just return something
2610 * and the io error handling code will clean up eventually
2615 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2616 u64 logical, u64 *length,
2617 struct btrfs_multi_bio **multi_ret,
2618 int mirror_num, struct page *unplug_page)
2620 struct extent_map *em;
2621 struct map_lookup *map;
2622 struct extent_map_tree *em_tree = &map_tree->map_tree;
2626 int stripes_allocated = 8;
2627 int stripes_required = 1;
2632 struct btrfs_multi_bio *multi = NULL;
2634 if (multi_ret && !(rw & (1 << BIO_RW)))
2635 stripes_allocated = 1;
2638 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2643 atomic_set(&multi->error, 0);
2646 read_lock(&em_tree->lock);
2647 em = lookup_extent_mapping(em_tree, logical, *length);
2648 read_unlock(&em_tree->lock);
2650 if (!em && unplug_page)
2654 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2655 (unsigned long long)logical,
2656 (unsigned long long)*length);
2660 BUG_ON(em->start > logical || em->start + em->len < logical);
2661 map = (struct map_lookup *)em->bdev;
2662 offset = logical - em->start;
2664 if (mirror_num > map->num_stripes)
2667 /* if our multi bio struct is too small, back off and try again */
2668 if (rw & (1 << BIO_RW)) {
2669 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2670 BTRFS_BLOCK_GROUP_DUP)) {
2671 stripes_required = map->num_stripes;
2673 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2674 stripes_required = map->sub_stripes;
2678 if (multi_ret && (rw & (1 << BIO_RW)) &&
2679 stripes_allocated < stripes_required) {
2680 stripes_allocated = map->num_stripes;
2681 free_extent_map(em);
2687 * stripe_nr counts the total number of stripes we have to stride
2688 * to get to this block
2690 do_div(stripe_nr, map->stripe_len);
2692 stripe_offset = stripe_nr * map->stripe_len;
2693 BUG_ON(offset < stripe_offset);
2695 /* stripe_offset is the offset of this block in its stripe*/
2696 stripe_offset = offset - stripe_offset;
2698 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2699 BTRFS_BLOCK_GROUP_RAID10 |
2700 BTRFS_BLOCK_GROUP_DUP)) {
2701 /* we limit the length of each bio to what fits in a stripe */
2702 *length = min_t(u64, em->len - offset,
2703 map->stripe_len - stripe_offset);
2705 *length = em->len - offset;
2708 if (!multi_ret && !unplug_page)
2713 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2714 if (unplug_page || (rw & (1 << BIO_RW)))
2715 num_stripes = map->num_stripes;
2716 else if (mirror_num)
2717 stripe_index = mirror_num - 1;
2719 stripe_index = find_live_mirror(map, 0,
2721 current->pid % map->num_stripes);
2724 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2725 if (rw & (1 << BIO_RW))
2726 num_stripes = map->num_stripes;
2727 else if (mirror_num)
2728 stripe_index = mirror_num - 1;
2730 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2731 int factor = map->num_stripes / map->sub_stripes;
2733 stripe_index = do_div(stripe_nr, factor);
2734 stripe_index *= map->sub_stripes;
2736 if (unplug_page || (rw & (1 << BIO_RW)))
2737 num_stripes = map->sub_stripes;
2738 else if (mirror_num)
2739 stripe_index += mirror_num - 1;
2741 stripe_index = find_live_mirror(map, stripe_index,
2742 map->sub_stripes, stripe_index +
2743 current->pid % map->sub_stripes);
2747 * after this do_div call, stripe_nr is the number of stripes
2748 * on this device we have to walk to find the data, and
2749 * stripe_index is the number of our device in the stripe array
2751 stripe_index = do_div(stripe_nr, map->num_stripes);
2753 BUG_ON(stripe_index >= map->num_stripes);
2755 for (i = 0; i < num_stripes; i++) {
2757 struct btrfs_device *device;
2758 struct backing_dev_info *bdi;
2760 device = map->stripes[stripe_index].dev;
2762 bdi = blk_get_backing_dev_info(device->bdev);
2763 if (bdi->unplug_io_fn)
2764 bdi->unplug_io_fn(bdi, unplug_page);
2767 multi->stripes[i].physical =
2768 map->stripes[stripe_index].physical +
2769 stripe_offset + stripe_nr * map->stripe_len;
2770 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2776 multi->num_stripes = num_stripes;
2777 multi->max_errors = max_errors;
2780 free_extent_map(em);
2784 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2785 u64 logical, u64 *length,
2786 struct btrfs_multi_bio **multi_ret, int mirror_num)
2788 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2792 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2793 u64 chunk_start, u64 physical, u64 devid,
2794 u64 **logical, int *naddrs, int *stripe_len)
2796 struct extent_map_tree *em_tree = &map_tree->map_tree;
2797 struct extent_map *em;
2798 struct map_lookup *map;
2805 read_lock(&em_tree->lock);
2806 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2807 read_unlock(&em_tree->lock);
2809 BUG_ON(!em || em->start != chunk_start);
2810 map = (struct map_lookup *)em->bdev;
2813 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2814 do_div(length, map->num_stripes / map->sub_stripes);
2815 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2816 do_div(length, map->num_stripes);
2818 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2821 for (i = 0; i < map->num_stripes; i++) {
2822 if (devid && map->stripes[i].dev->devid != devid)
2824 if (map->stripes[i].physical > physical ||
2825 map->stripes[i].physical + length <= physical)
2828 stripe_nr = physical - map->stripes[i].physical;
2829 do_div(stripe_nr, map->stripe_len);
2831 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2832 stripe_nr = stripe_nr * map->num_stripes + i;
2833 do_div(stripe_nr, map->sub_stripes);
2834 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2835 stripe_nr = stripe_nr * map->num_stripes + i;
2837 bytenr = chunk_start + stripe_nr * map->stripe_len;
2838 WARN_ON(nr >= map->num_stripes);
2839 for (j = 0; j < nr; j++) {
2840 if (buf[j] == bytenr)
2844 WARN_ON(nr >= map->num_stripes);
2851 *stripe_len = map->stripe_len;
2853 free_extent_map(em);
2857 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2858 u64 logical, struct page *page)
2860 u64 length = PAGE_CACHE_SIZE;
2861 return __btrfs_map_block(map_tree, READ, logical, &length,
2865 static void end_bio_multi_stripe(struct bio *bio, int err)
2867 struct btrfs_multi_bio *multi = bio->bi_private;
2868 int is_orig_bio = 0;
2871 atomic_inc(&multi->error);
2873 if (bio == multi->orig_bio)
2876 if (atomic_dec_and_test(&multi->stripes_pending)) {
2879 bio = multi->orig_bio;
2881 bio->bi_private = multi->private;
2882 bio->bi_end_io = multi->end_io;
2883 /* only send an error to the higher layers if it is
2884 * beyond the tolerance of the multi-bio
2886 if (atomic_read(&multi->error) > multi->max_errors) {
2890 * this bio is actually up to date, we didn't
2891 * go over the max number of errors
2893 set_bit(BIO_UPTODATE, &bio->bi_flags);
2898 bio_endio(bio, err);
2899 } else if (!is_orig_bio) {
2904 struct async_sched {
2907 struct btrfs_fs_info *info;
2908 struct btrfs_work work;
2912 * see run_scheduled_bios for a description of why bios are collected for
2915 * This will add one bio to the pending list for a device and make sure
2916 * the work struct is scheduled.
2918 static noinline int schedule_bio(struct btrfs_root *root,
2919 struct btrfs_device *device,
2920 int rw, struct bio *bio)
2922 int should_queue = 1;
2923 struct btrfs_pending_bios *pending_bios;
2925 /* don't bother with additional async steps for reads, right now */
2926 if (!(rw & (1 << BIO_RW))) {
2928 submit_bio(rw, bio);
2934 * nr_async_bios allows us to reliably return congestion to the
2935 * higher layers. Otherwise, the async bio makes it appear we have
2936 * made progress against dirty pages when we've really just put it
2937 * on a queue for later
2939 atomic_inc(&root->fs_info->nr_async_bios);
2940 WARN_ON(bio->bi_next);
2941 bio->bi_next = NULL;
2944 spin_lock(&device->io_lock);
2945 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2946 pending_bios = &device->pending_sync_bios;
2948 pending_bios = &device->pending_bios;
2950 if (pending_bios->tail)
2951 pending_bios->tail->bi_next = bio;
2953 pending_bios->tail = bio;
2954 if (!pending_bios->head)
2955 pending_bios->head = bio;
2956 if (device->running_pending)
2959 spin_unlock(&device->io_lock);
2962 btrfs_queue_worker(&root->fs_info->submit_workers,
2967 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2968 int mirror_num, int async_submit)
2970 struct btrfs_mapping_tree *map_tree;
2971 struct btrfs_device *dev;
2972 struct bio *first_bio = bio;
2973 u64 logical = (u64)bio->bi_sector << 9;
2976 struct btrfs_multi_bio *multi = NULL;
2981 length = bio->bi_size;
2982 map_tree = &root->fs_info->mapping_tree;
2983 map_length = length;
2985 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2989 total_devs = multi->num_stripes;
2990 if (map_length < length) {
2991 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2992 "len %llu\n", (unsigned long long)logical,
2993 (unsigned long long)length,
2994 (unsigned long long)map_length);
2997 multi->end_io = first_bio->bi_end_io;
2998 multi->private = first_bio->bi_private;
2999 multi->orig_bio = first_bio;
3000 atomic_set(&multi->stripes_pending, multi->num_stripes);
3002 while (dev_nr < total_devs) {
3003 if (total_devs > 1) {
3004 if (dev_nr < total_devs - 1) {
3005 bio = bio_clone(first_bio, GFP_NOFS);
3010 bio->bi_private = multi;
3011 bio->bi_end_io = end_bio_multi_stripe;
3013 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3014 dev = multi->stripes[dev_nr].dev;
3015 BUG_ON(rw == WRITE && !dev->writeable);
3016 if (dev && dev->bdev) {
3017 bio->bi_bdev = dev->bdev;
3019 schedule_bio(root, dev, rw, bio);
3021 submit_bio(rw, bio);
3023 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3024 bio->bi_sector = logical >> 9;
3025 bio_endio(bio, -EIO);
3029 if (total_devs == 1)
3034 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3037 struct btrfs_device *device;
3038 struct btrfs_fs_devices *cur_devices;
3040 cur_devices = root->fs_info->fs_devices;
3041 while (cur_devices) {
3043 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3044 device = __find_device(&cur_devices->devices,
3049 cur_devices = cur_devices->seed;
3054 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3055 u64 devid, u8 *dev_uuid)
3057 struct btrfs_device *device;
3058 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3060 device = kzalloc(sizeof(*device), GFP_NOFS);
3063 list_add(&device->dev_list,
3064 &fs_devices->devices);
3065 device->barriers = 1;
3066 device->dev_root = root->fs_info->dev_root;
3067 device->devid = devid;
3068 device->work.func = pending_bios_fn;
3069 device->fs_devices = fs_devices;
3070 fs_devices->num_devices++;
3071 spin_lock_init(&device->io_lock);
3072 INIT_LIST_HEAD(&device->dev_alloc_list);
3073 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3077 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3078 struct extent_buffer *leaf,
3079 struct btrfs_chunk *chunk)
3081 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3082 struct map_lookup *map;
3083 struct extent_map *em;
3087 u8 uuid[BTRFS_UUID_SIZE];
3092 logical = key->offset;
3093 length = btrfs_chunk_length(leaf, chunk);
3095 read_lock(&map_tree->map_tree.lock);
3096 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3097 read_unlock(&map_tree->map_tree.lock);
3099 /* already mapped? */
3100 if (em && em->start <= logical && em->start + em->len > logical) {
3101 free_extent_map(em);
3104 free_extent_map(em);
3107 em = alloc_extent_map(GFP_NOFS);
3110 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3111 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3113 free_extent_map(em);
3117 em->bdev = (struct block_device *)map;
3118 em->start = logical;
3120 em->block_start = 0;
3121 em->block_len = em->len;
3123 map->num_stripes = num_stripes;
3124 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3125 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3126 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3127 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3128 map->type = btrfs_chunk_type(leaf, chunk);
3129 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3130 for (i = 0; i < num_stripes; i++) {
3131 map->stripes[i].physical =
3132 btrfs_stripe_offset_nr(leaf, chunk, i);
3133 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3134 read_extent_buffer(leaf, uuid, (unsigned long)
3135 btrfs_stripe_dev_uuid_nr(chunk, i),
3137 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3139 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3141 free_extent_map(em);
3144 if (!map->stripes[i].dev) {
3145 map->stripes[i].dev =
3146 add_missing_dev(root, devid, uuid);
3147 if (!map->stripes[i].dev) {
3149 free_extent_map(em);
3153 map->stripes[i].dev->in_fs_metadata = 1;
3156 write_lock(&map_tree->map_tree.lock);
3157 ret = add_extent_mapping(&map_tree->map_tree, em);
3158 write_unlock(&map_tree->map_tree.lock);
3160 free_extent_map(em);
3165 static int fill_device_from_item(struct extent_buffer *leaf,
3166 struct btrfs_dev_item *dev_item,
3167 struct btrfs_device *device)
3171 device->devid = btrfs_device_id(leaf, dev_item);
3172 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3173 device->total_bytes = device->disk_total_bytes;
3174 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3175 device->type = btrfs_device_type(leaf, dev_item);
3176 device->io_align = btrfs_device_io_align(leaf, dev_item);
3177 device->io_width = btrfs_device_io_width(leaf, dev_item);
3178 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3180 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3181 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3186 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3188 struct btrfs_fs_devices *fs_devices;
3191 mutex_lock(&uuid_mutex);
3193 fs_devices = root->fs_info->fs_devices->seed;
3194 while (fs_devices) {
3195 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3199 fs_devices = fs_devices->seed;
3202 fs_devices = find_fsid(fsid);
3208 fs_devices = clone_fs_devices(fs_devices);
3209 if (IS_ERR(fs_devices)) {
3210 ret = PTR_ERR(fs_devices);
3214 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3215 root->fs_info->bdev_holder);
3219 if (!fs_devices->seeding) {
3220 __btrfs_close_devices(fs_devices);
3221 free_fs_devices(fs_devices);
3226 fs_devices->seed = root->fs_info->fs_devices->seed;
3227 root->fs_info->fs_devices->seed = fs_devices;
3229 mutex_unlock(&uuid_mutex);
3233 static int read_one_dev(struct btrfs_root *root,
3234 struct extent_buffer *leaf,
3235 struct btrfs_dev_item *dev_item)
3237 struct btrfs_device *device;
3240 u8 fs_uuid[BTRFS_UUID_SIZE];
3241 u8 dev_uuid[BTRFS_UUID_SIZE];
3243 devid = btrfs_device_id(leaf, dev_item);
3244 read_extent_buffer(leaf, dev_uuid,
3245 (unsigned long)btrfs_device_uuid(dev_item),
3247 read_extent_buffer(leaf, fs_uuid,
3248 (unsigned long)btrfs_device_fsid(dev_item),
3251 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3252 ret = open_seed_devices(root, fs_uuid);
3253 if (ret && !btrfs_test_opt(root, DEGRADED))
3257 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3258 if (!device || !device->bdev) {
3259 if (!btrfs_test_opt(root, DEGRADED))
3263 printk(KERN_WARNING "warning devid %llu missing\n",
3264 (unsigned long long)devid);
3265 device = add_missing_dev(root, devid, dev_uuid);
3271 if (device->fs_devices != root->fs_info->fs_devices) {
3272 BUG_ON(device->writeable);
3273 if (device->generation !=
3274 btrfs_device_generation(leaf, dev_item))
3278 fill_device_from_item(leaf, dev_item, device);
3279 device->dev_root = root->fs_info->dev_root;
3280 device->in_fs_metadata = 1;
3281 if (device->writeable)
3282 device->fs_devices->total_rw_bytes += device->total_bytes;
3287 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3289 struct btrfs_dev_item *dev_item;
3291 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3293 return read_one_dev(root, buf, dev_item);
3296 int btrfs_read_sys_array(struct btrfs_root *root)
3298 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3299 struct extent_buffer *sb;
3300 struct btrfs_disk_key *disk_key;
3301 struct btrfs_chunk *chunk;
3303 unsigned long sb_ptr;
3309 struct btrfs_key key;
3311 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3312 BTRFS_SUPER_INFO_SIZE);
3315 btrfs_set_buffer_uptodate(sb);
3316 btrfs_set_buffer_lockdep_class(sb, 0);
3318 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3319 array_size = btrfs_super_sys_array_size(super_copy);
3321 ptr = super_copy->sys_chunk_array;
3322 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3325 while (cur < array_size) {
3326 disk_key = (struct btrfs_disk_key *)ptr;
3327 btrfs_disk_key_to_cpu(&key, disk_key);
3329 len = sizeof(*disk_key); ptr += len;
3333 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3334 chunk = (struct btrfs_chunk *)sb_ptr;
3335 ret = read_one_chunk(root, &key, sb, chunk);
3338 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3339 len = btrfs_chunk_item_size(num_stripes);
3348 free_extent_buffer(sb);
3352 int btrfs_read_chunk_tree(struct btrfs_root *root)
3354 struct btrfs_path *path;
3355 struct extent_buffer *leaf;
3356 struct btrfs_key key;
3357 struct btrfs_key found_key;
3361 root = root->fs_info->chunk_root;
3363 path = btrfs_alloc_path();
3367 /* first we search for all of the device items, and then we
3368 * read in all of the chunk items. This way we can create chunk
3369 * mappings that reference all of the devices that are afound
3371 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3375 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3377 leaf = path->nodes[0];
3378 slot = path->slots[0];
3379 if (slot >= btrfs_header_nritems(leaf)) {
3380 ret = btrfs_next_leaf(root, path);
3387 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3388 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3389 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3391 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3392 struct btrfs_dev_item *dev_item;
3393 dev_item = btrfs_item_ptr(leaf, slot,
3394 struct btrfs_dev_item);
3395 ret = read_one_dev(root, leaf, dev_item);
3399 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3400 struct btrfs_chunk *chunk;
3401 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3402 ret = read_one_chunk(root, &found_key, leaf, chunk);
3408 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3410 btrfs_release_path(root, path);
3415 btrfs_free_path(path);