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);
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 > 32 &&
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 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
723 struct btrfs_device *device,
724 u64 num_bytes, u64 *start,
727 struct btrfs_key key;
728 struct btrfs_root *root = device->dev_root;
729 struct btrfs_dev_extent *dev_extent = NULL;
730 struct btrfs_path *path;
733 u64 search_start = 0;
734 u64 search_end = device->total_bytes;
738 struct extent_buffer *l;
740 path = btrfs_alloc_path();
746 /* FIXME use last free of some kind */
748 /* we don't want to overwrite the superblock on the drive,
749 * so we make sure to start at an offset of at least 1MB
751 search_start = max((u64)1024 * 1024, search_start);
753 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
754 search_start = max(root->fs_info->alloc_start, search_start);
756 key.objectid = device->devid;
757 key.offset = search_start;
758 key.type = BTRFS_DEV_EXTENT_KEY;
759 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
763 ret = btrfs_previous_item(root, path, key.objectid, key.type);
770 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
773 slot = path->slots[0];
774 if (slot >= btrfs_header_nritems(l)) {
775 ret = btrfs_next_leaf(root, path);
782 if (search_start >= search_end) {
786 *start = search_start;
790 *start = last_byte > search_start ?
791 last_byte : search_start;
792 if (search_end <= *start) {
798 btrfs_item_key_to_cpu(l, &key, slot);
800 if (key.objectid < device->devid)
803 if (key.objectid > device->devid)
806 if (key.offset >= search_start && key.offset > last_byte &&
808 if (last_byte < search_start)
809 last_byte = search_start;
810 hole_size = key.offset - last_byte;
812 if (hole_size > *max_avail)
813 *max_avail = hole_size;
815 if (key.offset > last_byte &&
816 hole_size >= num_bytes) {
821 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
825 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
826 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
832 /* we have to make sure we didn't find an extent that has already
833 * been allocated by the map tree or the original allocation
835 BUG_ON(*start < search_start);
837 if (*start + num_bytes > search_end) {
841 /* check for pending inserts here */
845 btrfs_free_path(path);
849 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
850 struct btrfs_device *device,
854 struct btrfs_path *path;
855 struct btrfs_root *root = device->dev_root;
856 struct btrfs_key key;
857 struct btrfs_key found_key;
858 struct extent_buffer *leaf = NULL;
859 struct btrfs_dev_extent *extent = NULL;
861 path = btrfs_alloc_path();
865 key.objectid = device->devid;
867 key.type = BTRFS_DEV_EXTENT_KEY;
869 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
871 ret = btrfs_previous_item(root, path, key.objectid,
872 BTRFS_DEV_EXTENT_KEY);
874 leaf = path->nodes[0];
875 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
876 extent = btrfs_item_ptr(leaf, path->slots[0],
877 struct btrfs_dev_extent);
878 BUG_ON(found_key.offset > start || found_key.offset +
879 btrfs_dev_extent_length(leaf, extent) < start);
881 } else if (ret == 0) {
882 leaf = path->nodes[0];
883 extent = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_dev_extent);
888 if (device->bytes_used > 0)
889 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
890 ret = btrfs_del_item(trans, root, path);
893 btrfs_free_path(path);
897 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
898 struct btrfs_device *device,
899 u64 chunk_tree, u64 chunk_objectid,
900 u64 chunk_offset, u64 start, u64 num_bytes)
903 struct btrfs_path *path;
904 struct btrfs_root *root = device->dev_root;
905 struct btrfs_dev_extent *extent;
906 struct extent_buffer *leaf;
907 struct btrfs_key key;
909 WARN_ON(!device->in_fs_metadata);
910 path = btrfs_alloc_path();
914 key.objectid = device->devid;
916 key.type = BTRFS_DEV_EXTENT_KEY;
917 ret = btrfs_insert_empty_item(trans, root, path, &key,
921 leaf = path->nodes[0];
922 extent = btrfs_item_ptr(leaf, path->slots[0],
923 struct btrfs_dev_extent);
924 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
925 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
926 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
928 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
929 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
932 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
933 btrfs_mark_buffer_dirty(leaf);
934 btrfs_free_path(path);
938 static noinline int find_next_chunk(struct btrfs_root *root,
939 u64 objectid, u64 *offset)
941 struct btrfs_path *path;
943 struct btrfs_key key;
944 struct btrfs_chunk *chunk;
945 struct btrfs_key found_key;
947 path = btrfs_alloc_path();
950 key.objectid = objectid;
951 key.offset = (u64)-1;
952 key.type = BTRFS_CHUNK_ITEM_KEY;
954 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
960 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
964 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
966 if (found_key.objectid != objectid)
969 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
971 *offset = found_key.offset +
972 btrfs_chunk_length(path->nodes[0], chunk);
977 btrfs_free_path(path);
981 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
984 struct btrfs_key key;
985 struct btrfs_key found_key;
986 struct btrfs_path *path;
988 root = root->fs_info->chunk_root;
990 path = btrfs_alloc_path();
994 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
995 key.type = BTRFS_DEV_ITEM_KEY;
996 key.offset = (u64)-1;
998 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1004 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1005 BTRFS_DEV_ITEM_KEY);
1009 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1011 *objectid = found_key.offset + 1;
1015 btrfs_free_path(path);
1020 * the device information is stored in the chunk root
1021 * the btrfs_device struct should be fully filled in
1023 int btrfs_add_device(struct btrfs_trans_handle *trans,
1024 struct btrfs_root *root,
1025 struct btrfs_device *device)
1028 struct btrfs_path *path;
1029 struct btrfs_dev_item *dev_item;
1030 struct extent_buffer *leaf;
1031 struct btrfs_key key;
1034 root = root->fs_info->chunk_root;
1036 path = btrfs_alloc_path();
1040 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1041 key.type = BTRFS_DEV_ITEM_KEY;
1042 key.offset = device->devid;
1044 ret = btrfs_insert_empty_item(trans, root, path, &key,
1049 leaf = path->nodes[0];
1050 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1052 btrfs_set_device_id(leaf, dev_item, device->devid);
1053 btrfs_set_device_generation(leaf, dev_item, 0);
1054 btrfs_set_device_type(leaf, dev_item, device->type);
1055 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1056 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1057 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1058 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1059 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1060 btrfs_set_device_group(leaf, dev_item, 0);
1061 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1062 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1063 btrfs_set_device_start_offset(leaf, dev_item, 0);
1065 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1066 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1067 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1068 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1069 btrfs_mark_buffer_dirty(leaf);
1073 btrfs_free_path(path);
1077 static int btrfs_rm_dev_item(struct btrfs_root *root,
1078 struct btrfs_device *device)
1081 struct btrfs_path *path;
1082 struct btrfs_key key;
1083 struct btrfs_trans_handle *trans;
1085 root = root->fs_info->chunk_root;
1087 path = btrfs_alloc_path();
1091 trans = btrfs_start_transaction(root, 1);
1092 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1093 key.type = BTRFS_DEV_ITEM_KEY;
1094 key.offset = device->devid;
1097 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1106 ret = btrfs_del_item(trans, root, path);
1110 btrfs_free_path(path);
1111 unlock_chunks(root);
1112 btrfs_commit_transaction(trans, root);
1116 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1118 struct btrfs_device *device;
1119 struct btrfs_device *next_device;
1120 struct block_device *bdev;
1121 struct buffer_head *bh = NULL;
1122 struct btrfs_super_block *disk_super;
1129 mutex_lock(&uuid_mutex);
1130 mutex_lock(&root->fs_info->volume_mutex);
1132 all_avail = root->fs_info->avail_data_alloc_bits |
1133 root->fs_info->avail_system_alloc_bits |
1134 root->fs_info->avail_metadata_alloc_bits;
1136 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1137 root->fs_info->fs_devices->rw_devices <= 4) {
1138 printk(KERN_ERR "btrfs: unable to go below four devices "
1144 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1145 root->fs_info->fs_devices->rw_devices <= 2) {
1146 printk(KERN_ERR "btrfs: unable to go below two "
1147 "devices on raid1\n");
1152 if (strcmp(device_path, "missing") == 0) {
1153 struct list_head *devices;
1154 struct btrfs_device *tmp;
1157 devices = &root->fs_info->fs_devices->devices;
1158 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1159 list_for_each_entry(tmp, devices, dev_list) {
1160 if (tmp->in_fs_metadata && !tmp->bdev) {
1165 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1170 printk(KERN_ERR "btrfs: no missing devices found to "
1175 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1176 root->fs_info->bdev_holder);
1178 ret = PTR_ERR(bdev);
1182 set_blocksize(bdev, 4096);
1183 bh = btrfs_read_dev_super(bdev);
1188 disk_super = (struct btrfs_super_block *)bh->b_data;
1189 devid = le64_to_cpu(disk_super->dev_item.devid);
1190 dev_uuid = disk_super->dev_item.uuid;
1191 device = btrfs_find_device(root, devid, dev_uuid,
1199 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1200 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1206 if (device->writeable) {
1207 list_del_init(&device->dev_alloc_list);
1208 root->fs_info->fs_devices->rw_devices--;
1211 ret = btrfs_shrink_device(device, 0);
1215 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1219 device->in_fs_metadata = 0;
1222 * the device list mutex makes sure that we don't change
1223 * the device list while someone else is writing out all
1224 * the device supers.
1226 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1227 list_del_init(&device->dev_list);
1228 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1230 device->fs_devices->num_devices--;
1232 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1233 struct btrfs_device, dev_list);
1234 if (device->bdev == root->fs_info->sb->s_bdev)
1235 root->fs_info->sb->s_bdev = next_device->bdev;
1236 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1237 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1240 close_bdev_exclusive(device->bdev, device->mode);
1241 device->bdev = NULL;
1242 device->fs_devices->open_devices--;
1245 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1246 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1248 if (device->fs_devices->open_devices == 0) {
1249 struct btrfs_fs_devices *fs_devices;
1250 fs_devices = root->fs_info->fs_devices;
1251 while (fs_devices) {
1252 if (fs_devices->seed == device->fs_devices)
1254 fs_devices = fs_devices->seed;
1256 fs_devices->seed = device->fs_devices->seed;
1257 device->fs_devices->seed = NULL;
1258 __btrfs_close_devices(device->fs_devices);
1259 free_fs_devices(device->fs_devices);
1263 * at this point, the device is zero sized. We want to
1264 * remove it from the devices list and zero out the old super
1266 if (device->writeable) {
1267 /* make sure this device isn't detected as part of
1270 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1271 set_buffer_dirty(bh);
1272 sync_dirty_buffer(bh);
1275 kfree(device->name);
1283 close_bdev_exclusive(bdev, FMODE_READ);
1285 mutex_unlock(&root->fs_info->volume_mutex);
1286 mutex_unlock(&uuid_mutex);
1291 * does all the dirty work required for changing file system's UUID.
1293 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1294 struct btrfs_root *root)
1296 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1297 struct btrfs_fs_devices *old_devices;
1298 struct btrfs_fs_devices *seed_devices;
1299 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1300 struct btrfs_device *device;
1303 BUG_ON(!mutex_is_locked(&uuid_mutex));
1304 if (!fs_devices->seeding)
1307 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1311 old_devices = clone_fs_devices(fs_devices);
1312 if (IS_ERR(old_devices)) {
1313 kfree(seed_devices);
1314 return PTR_ERR(old_devices);
1317 list_add(&old_devices->list, &fs_uuids);
1319 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1320 seed_devices->opened = 1;
1321 INIT_LIST_HEAD(&seed_devices->devices);
1322 INIT_LIST_HEAD(&seed_devices->alloc_list);
1323 mutex_init(&seed_devices->device_list_mutex);
1324 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1325 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1326 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1327 device->fs_devices = seed_devices;
1330 fs_devices->seeding = 0;
1331 fs_devices->num_devices = 0;
1332 fs_devices->open_devices = 0;
1333 fs_devices->seed = seed_devices;
1335 generate_random_uuid(fs_devices->fsid);
1336 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1337 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1338 super_flags = btrfs_super_flags(disk_super) &
1339 ~BTRFS_SUPER_FLAG_SEEDING;
1340 btrfs_set_super_flags(disk_super, super_flags);
1346 * strore the expected generation for seed devices in device items.
1348 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1349 struct btrfs_root *root)
1351 struct btrfs_path *path;
1352 struct extent_buffer *leaf;
1353 struct btrfs_dev_item *dev_item;
1354 struct btrfs_device *device;
1355 struct btrfs_key key;
1356 u8 fs_uuid[BTRFS_UUID_SIZE];
1357 u8 dev_uuid[BTRFS_UUID_SIZE];
1361 path = btrfs_alloc_path();
1365 root = root->fs_info->chunk_root;
1366 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1368 key.type = BTRFS_DEV_ITEM_KEY;
1371 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1375 leaf = path->nodes[0];
1377 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1378 ret = btrfs_next_leaf(root, path);
1383 leaf = path->nodes[0];
1384 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1385 btrfs_release_path(root, path);
1389 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1390 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1391 key.type != BTRFS_DEV_ITEM_KEY)
1394 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1395 struct btrfs_dev_item);
1396 devid = btrfs_device_id(leaf, dev_item);
1397 read_extent_buffer(leaf, dev_uuid,
1398 (unsigned long)btrfs_device_uuid(dev_item),
1400 read_extent_buffer(leaf, fs_uuid,
1401 (unsigned long)btrfs_device_fsid(dev_item),
1403 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1406 if (device->fs_devices->seeding) {
1407 btrfs_set_device_generation(leaf, dev_item,
1408 device->generation);
1409 btrfs_mark_buffer_dirty(leaf);
1417 btrfs_free_path(path);
1421 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1423 struct btrfs_trans_handle *trans;
1424 struct btrfs_device *device;
1425 struct block_device *bdev;
1426 struct list_head *devices;
1427 struct super_block *sb = root->fs_info->sb;
1429 int seeding_dev = 0;
1432 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1435 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1439 if (root->fs_info->fs_devices->seeding) {
1441 down_write(&sb->s_umount);
1442 mutex_lock(&uuid_mutex);
1445 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1446 mutex_lock(&root->fs_info->volume_mutex);
1448 devices = &root->fs_info->fs_devices->devices;
1450 * we have the volume lock, so we don't need the extra
1451 * device list mutex while reading the list here.
1453 list_for_each_entry(device, devices, dev_list) {
1454 if (device->bdev == bdev) {
1460 device = kzalloc(sizeof(*device), GFP_NOFS);
1462 /* we can safely leave the fs_devices entry around */
1467 device->name = kstrdup(device_path, GFP_NOFS);
1468 if (!device->name) {
1474 ret = find_next_devid(root, &device->devid);
1480 trans = btrfs_start_transaction(root, 1);
1483 device->barriers = 1;
1484 device->writeable = 1;
1485 device->work.func = pending_bios_fn;
1486 generate_random_uuid(device->uuid);
1487 spin_lock_init(&device->io_lock);
1488 device->generation = trans->transid;
1489 device->io_width = root->sectorsize;
1490 device->io_align = root->sectorsize;
1491 device->sector_size = root->sectorsize;
1492 device->total_bytes = i_size_read(bdev->bd_inode);
1493 device->disk_total_bytes = device->total_bytes;
1494 device->dev_root = root->fs_info->dev_root;
1495 device->bdev = bdev;
1496 device->in_fs_metadata = 1;
1498 set_blocksize(device->bdev, 4096);
1501 sb->s_flags &= ~MS_RDONLY;
1502 ret = btrfs_prepare_sprout(trans, root);
1506 device->fs_devices = root->fs_info->fs_devices;
1509 * we don't want write_supers to jump in here with our device
1512 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1513 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1514 list_add(&device->dev_alloc_list,
1515 &root->fs_info->fs_devices->alloc_list);
1516 root->fs_info->fs_devices->num_devices++;
1517 root->fs_info->fs_devices->open_devices++;
1518 root->fs_info->fs_devices->rw_devices++;
1519 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1521 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1522 root->fs_info->fs_devices->rotating = 1;
1524 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1525 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1526 total_bytes + device->total_bytes);
1528 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1529 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1531 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1534 ret = init_first_rw_device(trans, root, device);
1536 ret = btrfs_finish_sprout(trans, root);
1539 ret = btrfs_add_device(trans, root, device);
1543 * we've got more storage, clear any full flags on the space
1546 btrfs_clear_space_info_full(root->fs_info);
1548 unlock_chunks(root);
1549 btrfs_commit_transaction(trans, root);
1552 mutex_unlock(&uuid_mutex);
1553 up_write(&sb->s_umount);
1555 ret = btrfs_relocate_sys_chunks(root);
1559 mutex_unlock(&root->fs_info->volume_mutex);
1562 close_bdev_exclusive(bdev, 0);
1564 mutex_unlock(&uuid_mutex);
1565 up_write(&sb->s_umount);
1570 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1571 struct btrfs_device *device)
1574 struct btrfs_path *path;
1575 struct btrfs_root *root;
1576 struct btrfs_dev_item *dev_item;
1577 struct extent_buffer *leaf;
1578 struct btrfs_key key;
1580 root = device->dev_root->fs_info->chunk_root;
1582 path = btrfs_alloc_path();
1586 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1587 key.type = BTRFS_DEV_ITEM_KEY;
1588 key.offset = device->devid;
1590 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1599 leaf = path->nodes[0];
1600 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1602 btrfs_set_device_id(leaf, dev_item, device->devid);
1603 btrfs_set_device_type(leaf, dev_item, device->type);
1604 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1605 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1606 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1607 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1608 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1609 btrfs_mark_buffer_dirty(leaf);
1612 btrfs_free_path(path);
1616 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1617 struct btrfs_device *device, u64 new_size)
1619 struct btrfs_super_block *super_copy =
1620 &device->dev_root->fs_info->super_copy;
1621 u64 old_total = btrfs_super_total_bytes(super_copy);
1622 u64 diff = new_size - device->total_bytes;
1624 if (!device->writeable)
1626 if (new_size <= device->total_bytes)
1629 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1630 device->fs_devices->total_rw_bytes += diff;
1632 device->total_bytes = new_size;
1633 device->disk_total_bytes = new_size;
1634 btrfs_clear_space_info_full(device->dev_root->fs_info);
1636 return btrfs_update_device(trans, device);
1639 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1640 struct btrfs_device *device, u64 new_size)
1643 lock_chunks(device->dev_root);
1644 ret = __btrfs_grow_device(trans, device, new_size);
1645 unlock_chunks(device->dev_root);
1649 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1650 struct btrfs_root *root,
1651 u64 chunk_tree, u64 chunk_objectid,
1655 struct btrfs_path *path;
1656 struct btrfs_key key;
1658 root = root->fs_info->chunk_root;
1659 path = btrfs_alloc_path();
1663 key.objectid = chunk_objectid;
1664 key.offset = chunk_offset;
1665 key.type = BTRFS_CHUNK_ITEM_KEY;
1667 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1670 ret = btrfs_del_item(trans, root, path);
1673 btrfs_free_path(path);
1677 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1680 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1681 struct btrfs_disk_key *disk_key;
1682 struct btrfs_chunk *chunk;
1689 struct btrfs_key key;
1691 array_size = btrfs_super_sys_array_size(super_copy);
1693 ptr = super_copy->sys_chunk_array;
1696 while (cur < array_size) {
1697 disk_key = (struct btrfs_disk_key *)ptr;
1698 btrfs_disk_key_to_cpu(&key, disk_key);
1700 len = sizeof(*disk_key);
1702 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1703 chunk = (struct btrfs_chunk *)(ptr + len);
1704 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1705 len += btrfs_chunk_item_size(num_stripes);
1710 if (key.objectid == chunk_objectid &&
1711 key.offset == chunk_offset) {
1712 memmove(ptr, ptr + len, array_size - (cur + len));
1714 btrfs_set_super_sys_array_size(super_copy, array_size);
1723 static int btrfs_relocate_chunk(struct btrfs_root *root,
1724 u64 chunk_tree, u64 chunk_objectid,
1727 struct extent_map_tree *em_tree;
1728 struct btrfs_root *extent_root;
1729 struct btrfs_trans_handle *trans;
1730 struct extent_map *em;
1731 struct map_lookup *map;
1735 root = root->fs_info->chunk_root;
1736 extent_root = root->fs_info->extent_root;
1737 em_tree = &root->fs_info->mapping_tree.map_tree;
1739 /* step one, relocate all the extents inside this chunk */
1740 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1743 trans = btrfs_start_transaction(root, 1);
1749 * step two, delete the device extents and the
1750 * chunk tree entries
1752 spin_lock(&em_tree->lock);
1753 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1754 spin_unlock(&em_tree->lock);
1756 BUG_ON(em->start > chunk_offset ||
1757 em->start + em->len < chunk_offset);
1758 map = (struct map_lookup *)em->bdev;
1760 for (i = 0; i < map->num_stripes; i++) {
1761 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1762 map->stripes[i].physical);
1765 if (map->stripes[i].dev) {
1766 ret = btrfs_update_device(trans, map->stripes[i].dev);
1770 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1775 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1776 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1780 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1783 spin_lock(&em_tree->lock);
1784 remove_extent_mapping(em_tree, em);
1785 spin_unlock(&em_tree->lock);
1790 /* once for the tree */
1791 free_extent_map(em);
1793 free_extent_map(em);
1795 unlock_chunks(root);
1796 btrfs_end_transaction(trans, root);
1800 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1802 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1803 struct btrfs_path *path;
1804 struct extent_buffer *leaf;
1805 struct btrfs_chunk *chunk;
1806 struct btrfs_key key;
1807 struct btrfs_key found_key;
1808 u64 chunk_tree = chunk_root->root_key.objectid;
1812 path = btrfs_alloc_path();
1816 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1817 key.offset = (u64)-1;
1818 key.type = BTRFS_CHUNK_ITEM_KEY;
1821 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1826 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1833 leaf = path->nodes[0];
1834 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1836 chunk = btrfs_item_ptr(leaf, path->slots[0],
1837 struct btrfs_chunk);
1838 chunk_type = btrfs_chunk_type(leaf, chunk);
1839 btrfs_release_path(chunk_root, path);
1841 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1842 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1848 if (found_key.offset == 0)
1850 key.offset = found_key.offset - 1;
1854 btrfs_free_path(path);
1858 static u64 div_factor(u64 num, int factor)
1867 int btrfs_balance(struct btrfs_root *dev_root)
1870 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1871 struct btrfs_device *device;
1874 struct btrfs_path *path;
1875 struct btrfs_key key;
1876 struct btrfs_chunk *chunk;
1877 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1878 struct btrfs_trans_handle *trans;
1879 struct btrfs_key found_key;
1881 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1884 mutex_lock(&dev_root->fs_info->volume_mutex);
1885 dev_root = dev_root->fs_info->dev_root;
1887 /* step one make some room on all the devices */
1888 list_for_each_entry(device, devices, dev_list) {
1889 old_size = device->total_bytes;
1890 size_to_free = div_factor(old_size, 1);
1891 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1892 if (!device->writeable ||
1893 device->total_bytes - device->bytes_used > size_to_free)
1896 ret = btrfs_shrink_device(device, old_size - size_to_free);
1899 trans = btrfs_start_transaction(dev_root, 1);
1902 ret = btrfs_grow_device(trans, device, old_size);
1905 btrfs_end_transaction(trans, dev_root);
1908 /* step two, relocate all the chunks */
1909 path = btrfs_alloc_path();
1912 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1913 key.offset = (u64)-1;
1914 key.type = BTRFS_CHUNK_ITEM_KEY;
1917 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1922 * this shouldn't happen, it means the last relocate
1928 ret = btrfs_previous_item(chunk_root, path, 0,
1929 BTRFS_CHUNK_ITEM_KEY);
1933 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1935 if (found_key.objectid != key.objectid)
1938 chunk = btrfs_item_ptr(path->nodes[0],
1940 struct btrfs_chunk);
1941 key.offset = found_key.offset;
1942 /* chunk zero is special */
1943 if (key.offset == 0)
1946 btrfs_release_path(chunk_root, path);
1947 ret = btrfs_relocate_chunk(chunk_root,
1948 chunk_root->root_key.objectid,
1955 btrfs_free_path(path);
1956 mutex_unlock(&dev_root->fs_info->volume_mutex);
1961 * shrinking a device means finding all of the device extents past
1962 * the new size, and then following the back refs to the chunks.
1963 * The chunk relocation code actually frees the device extent
1965 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1967 struct btrfs_trans_handle *trans;
1968 struct btrfs_root *root = device->dev_root;
1969 struct btrfs_dev_extent *dev_extent = NULL;
1970 struct btrfs_path *path;
1977 struct extent_buffer *l;
1978 struct btrfs_key key;
1979 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1980 u64 old_total = btrfs_super_total_bytes(super_copy);
1981 u64 diff = device->total_bytes - new_size;
1983 if (new_size >= device->total_bytes)
1986 path = btrfs_alloc_path();
1990 trans = btrfs_start_transaction(root, 1);
2000 device->total_bytes = new_size;
2001 if (device->writeable)
2002 device->fs_devices->total_rw_bytes -= diff;
2003 unlock_chunks(root);
2004 btrfs_end_transaction(trans, root);
2006 key.objectid = device->devid;
2007 key.offset = (u64)-1;
2008 key.type = BTRFS_DEV_EXTENT_KEY;
2011 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2015 ret = btrfs_previous_item(root, path, 0, key.type);
2024 slot = path->slots[0];
2025 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2027 if (key.objectid != device->devid)
2030 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2031 length = btrfs_dev_extent_length(l, dev_extent);
2033 if (key.offset + length <= new_size)
2036 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2037 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2038 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2039 btrfs_release_path(root, path);
2041 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2047 /* Shrinking succeeded, else we would be at "done". */
2048 trans = btrfs_start_transaction(root, 1);
2055 device->disk_total_bytes = new_size;
2056 /* Now btrfs_update_device() will change the on-disk size. */
2057 ret = btrfs_update_device(trans, device);
2059 unlock_chunks(root);
2060 btrfs_end_transaction(trans, root);
2063 WARN_ON(diff > old_total);
2064 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2065 unlock_chunks(root);
2066 btrfs_end_transaction(trans, root);
2068 btrfs_free_path(path);
2072 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2073 struct btrfs_root *root,
2074 struct btrfs_key *key,
2075 struct btrfs_chunk *chunk, int item_size)
2077 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2078 struct btrfs_disk_key disk_key;
2082 array_size = btrfs_super_sys_array_size(super_copy);
2083 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2086 ptr = super_copy->sys_chunk_array + array_size;
2087 btrfs_cpu_key_to_disk(&disk_key, key);
2088 memcpy(ptr, &disk_key, sizeof(disk_key));
2089 ptr += sizeof(disk_key);
2090 memcpy(ptr, chunk, item_size);
2091 item_size += sizeof(disk_key);
2092 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2096 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2097 int num_stripes, int sub_stripes)
2099 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2101 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2102 return calc_size * (num_stripes / sub_stripes);
2104 return calc_size * num_stripes;
2107 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2108 struct btrfs_root *extent_root,
2109 struct map_lookup **map_ret,
2110 u64 *num_bytes, u64 *stripe_size,
2111 u64 start, u64 type)
2113 struct btrfs_fs_info *info = extent_root->fs_info;
2114 struct btrfs_device *device = NULL;
2115 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2116 struct list_head *cur;
2117 struct map_lookup *map = NULL;
2118 struct extent_map_tree *em_tree;
2119 struct extent_map *em;
2120 struct list_head private_devs;
2121 int min_stripe_size = 1 * 1024 * 1024;
2122 u64 calc_size = 1024 * 1024 * 1024;
2123 u64 max_chunk_size = calc_size;
2128 int num_stripes = 1;
2129 int min_stripes = 1;
2130 int sub_stripes = 0;
2134 int stripe_len = 64 * 1024;
2136 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2137 (type & BTRFS_BLOCK_GROUP_DUP)) {
2139 type &= ~BTRFS_BLOCK_GROUP_DUP;
2141 if (list_empty(&fs_devices->alloc_list))
2144 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2145 num_stripes = fs_devices->rw_devices;
2148 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2152 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2153 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2154 if (num_stripes < 2)
2158 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2159 num_stripes = fs_devices->rw_devices;
2160 if (num_stripes < 4)
2162 num_stripes &= ~(u32)1;
2167 if (type & BTRFS_BLOCK_GROUP_DATA) {
2168 max_chunk_size = 10 * calc_size;
2169 min_stripe_size = 64 * 1024 * 1024;
2170 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2171 max_chunk_size = 4 * calc_size;
2172 min_stripe_size = 32 * 1024 * 1024;
2173 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2174 calc_size = 8 * 1024 * 1024;
2175 max_chunk_size = calc_size * 2;
2176 min_stripe_size = 1 * 1024 * 1024;
2179 /* we don't want a chunk larger than 10% of writeable space */
2180 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2185 if (!map || map->num_stripes != num_stripes) {
2187 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2190 map->num_stripes = num_stripes;
2193 if (calc_size * num_stripes > max_chunk_size) {
2194 calc_size = max_chunk_size;
2195 do_div(calc_size, num_stripes);
2196 do_div(calc_size, stripe_len);
2197 calc_size *= stripe_len;
2199 /* we don't want tiny stripes */
2200 calc_size = max_t(u64, min_stripe_size, calc_size);
2202 do_div(calc_size, stripe_len);
2203 calc_size *= stripe_len;
2205 cur = fs_devices->alloc_list.next;
2208 if (type & BTRFS_BLOCK_GROUP_DUP)
2209 min_free = calc_size * 2;
2211 min_free = calc_size;
2214 * we add 1MB because we never use the first 1MB of the device, unless
2215 * we've looped, then we are likely allocating the maximum amount of
2216 * space left already
2219 min_free += 1024 * 1024;
2221 INIT_LIST_HEAD(&private_devs);
2222 while (index < num_stripes) {
2223 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2224 BUG_ON(!device->writeable);
2225 if (device->total_bytes > device->bytes_used)
2226 avail = device->total_bytes - device->bytes_used;
2231 if (device->in_fs_metadata && avail >= min_free) {
2232 ret = find_free_dev_extent(trans, device,
2233 min_free, &dev_offset,
2236 list_move_tail(&device->dev_alloc_list,
2238 map->stripes[index].dev = device;
2239 map->stripes[index].physical = dev_offset;
2241 if (type & BTRFS_BLOCK_GROUP_DUP) {
2242 map->stripes[index].dev = device;
2243 map->stripes[index].physical =
2244 dev_offset + calc_size;
2248 } else if (device->in_fs_metadata && avail > max_avail)
2250 if (cur == &fs_devices->alloc_list)
2253 list_splice(&private_devs, &fs_devices->alloc_list);
2254 if (index < num_stripes) {
2255 if (index >= min_stripes) {
2256 num_stripes = index;
2257 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2258 num_stripes /= sub_stripes;
2259 num_stripes *= sub_stripes;
2264 if (!looped && max_avail > 0) {
2266 calc_size = max_avail;
2272 map->sector_size = extent_root->sectorsize;
2273 map->stripe_len = stripe_len;
2274 map->io_align = stripe_len;
2275 map->io_width = stripe_len;
2277 map->num_stripes = num_stripes;
2278 map->sub_stripes = sub_stripes;
2281 *stripe_size = calc_size;
2282 *num_bytes = chunk_bytes_by_type(type, calc_size,
2283 num_stripes, sub_stripes);
2285 em = alloc_extent_map(GFP_NOFS);
2290 em->bdev = (struct block_device *)map;
2292 em->len = *num_bytes;
2293 em->block_start = 0;
2294 em->block_len = em->len;
2296 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2297 spin_lock(&em_tree->lock);
2298 ret = add_extent_mapping(em_tree, em);
2299 spin_unlock(&em_tree->lock);
2301 free_extent_map(em);
2303 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2304 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2309 while (index < map->num_stripes) {
2310 device = map->stripes[index].dev;
2311 dev_offset = map->stripes[index].physical;
2313 ret = btrfs_alloc_dev_extent(trans, device,
2314 info->chunk_root->root_key.objectid,
2315 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2316 start, dev_offset, calc_size);
2324 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2325 struct btrfs_root *extent_root,
2326 struct map_lookup *map, u64 chunk_offset,
2327 u64 chunk_size, u64 stripe_size)
2330 struct btrfs_key key;
2331 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2332 struct btrfs_device *device;
2333 struct btrfs_chunk *chunk;
2334 struct btrfs_stripe *stripe;
2335 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2339 chunk = kzalloc(item_size, GFP_NOFS);
2344 while (index < map->num_stripes) {
2345 device = map->stripes[index].dev;
2346 device->bytes_used += stripe_size;
2347 ret = btrfs_update_device(trans, device);
2353 stripe = &chunk->stripe;
2354 while (index < map->num_stripes) {
2355 device = map->stripes[index].dev;
2356 dev_offset = map->stripes[index].physical;
2358 btrfs_set_stack_stripe_devid(stripe, device->devid);
2359 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2360 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2365 btrfs_set_stack_chunk_length(chunk, chunk_size);
2366 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2367 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2368 btrfs_set_stack_chunk_type(chunk, map->type);
2369 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2370 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2371 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2372 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2373 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2375 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2376 key.type = BTRFS_CHUNK_ITEM_KEY;
2377 key.offset = chunk_offset;
2379 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2382 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2383 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2392 * Chunk allocation falls into two parts. The first part does works
2393 * that make the new allocated chunk useable, but not do any operation
2394 * that modifies the chunk tree. The second part does the works that
2395 * require modifying the chunk tree. This division is important for the
2396 * bootstrap process of adding storage to a seed btrfs.
2398 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2399 struct btrfs_root *extent_root, u64 type)
2404 struct map_lookup *map;
2405 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2408 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2413 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2414 &stripe_size, chunk_offset, type);
2418 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2419 chunk_size, stripe_size);
2424 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2425 struct btrfs_root *root,
2426 struct btrfs_device *device)
2429 u64 sys_chunk_offset;
2433 u64 sys_stripe_size;
2435 struct map_lookup *map;
2436 struct map_lookup *sys_map;
2437 struct btrfs_fs_info *fs_info = root->fs_info;
2438 struct btrfs_root *extent_root = fs_info->extent_root;
2441 ret = find_next_chunk(fs_info->chunk_root,
2442 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2445 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2446 (fs_info->metadata_alloc_profile &
2447 fs_info->avail_metadata_alloc_bits);
2448 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2450 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2451 &stripe_size, chunk_offset, alloc_profile);
2454 sys_chunk_offset = chunk_offset + chunk_size;
2456 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2457 (fs_info->system_alloc_profile &
2458 fs_info->avail_system_alloc_bits);
2459 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2461 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2462 &sys_chunk_size, &sys_stripe_size,
2463 sys_chunk_offset, alloc_profile);
2466 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2470 * Modifying chunk tree needs allocating new blocks from both
2471 * system block group and metadata block group. So we only can
2472 * do operations require modifying the chunk tree after both
2473 * block groups were created.
2475 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2476 chunk_size, stripe_size);
2479 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2480 sys_chunk_offset, sys_chunk_size,
2486 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2488 struct extent_map *em;
2489 struct map_lookup *map;
2490 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2494 spin_lock(&map_tree->map_tree.lock);
2495 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2496 spin_unlock(&map_tree->map_tree.lock);
2500 map = (struct map_lookup *)em->bdev;
2501 for (i = 0; i < map->num_stripes; i++) {
2502 if (!map->stripes[i].dev->writeable) {
2507 free_extent_map(em);
2511 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2513 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2516 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2518 struct extent_map *em;
2521 spin_lock(&tree->map_tree.lock);
2522 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2524 remove_extent_mapping(&tree->map_tree, em);
2525 spin_unlock(&tree->map_tree.lock);
2530 free_extent_map(em);
2531 /* once for the tree */
2532 free_extent_map(em);
2536 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2538 struct extent_map *em;
2539 struct map_lookup *map;
2540 struct extent_map_tree *em_tree = &map_tree->map_tree;
2543 spin_lock(&em_tree->lock);
2544 em = lookup_extent_mapping(em_tree, logical, len);
2545 spin_unlock(&em_tree->lock);
2548 BUG_ON(em->start > logical || em->start + em->len < logical);
2549 map = (struct map_lookup *)em->bdev;
2550 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2551 ret = map->num_stripes;
2552 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2553 ret = map->sub_stripes;
2556 free_extent_map(em);
2560 static int find_live_mirror(struct map_lookup *map, int first, int num,
2564 if (map->stripes[optimal].dev->bdev)
2566 for (i = first; i < first + num; i++) {
2567 if (map->stripes[i].dev->bdev)
2570 /* we couldn't find one that doesn't fail. Just return something
2571 * and the io error handling code will clean up eventually
2576 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2577 u64 logical, u64 *length,
2578 struct btrfs_multi_bio **multi_ret,
2579 int mirror_num, struct page *unplug_page)
2581 struct extent_map *em;
2582 struct map_lookup *map;
2583 struct extent_map_tree *em_tree = &map_tree->map_tree;
2587 int stripes_allocated = 8;
2588 int stripes_required = 1;
2593 struct btrfs_multi_bio *multi = NULL;
2595 if (multi_ret && !(rw & (1 << BIO_RW)))
2596 stripes_allocated = 1;
2599 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2604 atomic_set(&multi->error, 0);
2607 spin_lock(&em_tree->lock);
2608 em = lookup_extent_mapping(em_tree, logical, *length);
2609 spin_unlock(&em_tree->lock);
2611 if (!em && unplug_page)
2615 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2616 (unsigned long long)logical,
2617 (unsigned long long)*length);
2621 BUG_ON(em->start > logical || em->start + em->len < logical);
2622 map = (struct map_lookup *)em->bdev;
2623 offset = logical - em->start;
2625 if (mirror_num > map->num_stripes)
2628 /* if our multi bio struct is too small, back off and try again */
2629 if (rw & (1 << BIO_RW)) {
2630 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2631 BTRFS_BLOCK_GROUP_DUP)) {
2632 stripes_required = map->num_stripes;
2634 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2635 stripes_required = map->sub_stripes;
2639 if (multi_ret && (rw & (1 << BIO_RW)) &&
2640 stripes_allocated < stripes_required) {
2641 stripes_allocated = map->num_stripes;
2642 free_extent_map(em);
2648 * stripe_nr counts the total number of stripes we have to stride
2649 * to get to this block
2651 do_div(stripe_nr, map->stripe_len);
2653 stripe_offset = stripe_nr * map->stripe_len;
2654 BUG_ON(offset < stripe_offset);
2656 /* stripe_offset is the offset of this block in its stripe*/
2657 stripe_offset = offset - stripe_offset;
2659 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2660 BTRFS_BLOCK_GROUP_RAID10 |
2661 BTRFS_BLOCK_GROUP_DUP)) {
2662 /* we limit the length of each bio to what fits in a stripe */
2663 *length = min_t(u64, em->len - offset,
2664 map->stripe_len - stripe_offset);
2666 *length = em->len - offset;
2669 if (!multi_ret && !unplug_page)
2674 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2675 if (unplug_page || (rw & (1 << BIO_RW)))
2676 num_stripes = map->num_stripes;
2677 else if (mirror_num)
2678 stripe_index = mirror_num - 1;
2680 stripe_index = find_live_mirror(map, 0,
2682 current->pid % map->num_stripes);
2685 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2686 if (rw & (1 << BIO_RW))
2687 num_stripes = map->num_stripes;
2688 else if (mirror_num)
2689 stripe_index = mirror_num - 1;
2691 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2692 int factor = map->num_stripes / map->sub_stripes;
2694 stripe_index = do_div(stripe_nr, factor);
2695 stripe_index *= map->sub_stripes;
2697 if (unplug_page || (rw & (1 << BIO_RW)))
2698 num_stripes = map->sub_stripes;
2699 else if (mirror_num)
2700 stripe_index += mirror_num - 1;
2702 stripe_index = find_live_mirror(map, stripe_index,
2703 map->sub_stripes, stripe_index +
2704 current->pid % map->sub_stripes);
2708 * after this do_div call, stripe_nr is the number of stripes
2709 * on this device we have to walk to find the data, and
2710 * stripe_index is the number of our device in the stripe array
2712 stripe_index = do_div(stripe_nr, map->num_stripes);
2714 BUG_ON(stripe_index >= map->num_stripes);
2716 for (i = 0; i < num_stripes; i++) {
2718 struct btrfs_device *device;
2719 struct backing_dev_info *bdi;
2721 device = map->stripes[stripe_index].dev;
2723 bdi = blk_get_backing_dev_info(device->bdev);
2724 if (bdi->unplug_io_fn)
2725 bdi->unplug_io_fn(bdi, unplug_page);
2728 multi->stripes[i].physical =
2729 map->stripes[stripe_index].physical +
2730 stripe_offset + stripe_nr * map->stripe_len;
2731 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2737 multi->num_stripes = num_stripes;
2738 multi->max_errors = max_errors;
2741 free_extent_map(em);
2745 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2746 u64 logical, u64 *length,
2747 struct btrfs_multi_bio **multi_ret, int mirror_num)
2749 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2753 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2754 u64 chunk_start, u64 physical, u64 devid,
2755 u64 **logical, int *naddrs, int *stripe_len)
2757 struct extent_map_tree *em_tree = &map_tree->map_tree;
2758 struct extent_map *em;
2759 struct map_lookup *map;
2766 spin_lock(&em_tree->lock);
2767 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2768 spin_unlock(&em_tree->lock);
2770 BUG_ON(!em || em->start != chunk_start);
2771 map = (struct map_lookup *)em->bdev;
2774 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2775 do_div(length, map->num_stripes / map->sub_stripes);
2776 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2777 do_div(length, map->num_stripes);
2779 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2782 for (i = 0; i < map->num_stripes; i++) {
2783 if (devid && map->stripes[i].dev->devid != devid)
2785 if (map->stripes[i].physical > physical ||
2786 map->stripes[i].physical + length <= physical)
2789 stripe_nr = physical - map->stripes[i].physical;
2790 do_div(stripe_nr, map->stripe_len);
2792 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2793 stripe_nr = stripe_nr * map->num_stripes + i;
2794 do_div(stripe_nr, map->sub_stripes);
2795 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2796 stripe_nr = stripe_nr * map->num_stripes + i;
2798 bytenr = chunk_start + stripe_nr * map->stripe_len;
2799 WARN_ON(nr >= map->num_stripes);
2800 for (j = 0; j < nr; j++) {
2801 if (buf[j] == bytenr)
2805 WARN_ON(nr >= map->num_stripes);
2812 *stripe_len = map->stripe_len;
2814 free_extent_map(em);
2818 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2819 u64 logical, struct page *page)
2821 u64 length = PAGE_CACHE_SIZE;
2822 return __btrfs_map_block(map_tree, READ, logical, &length,
2826 static void end_bio_multi_stripe(struct bio *bio, int err)
2828 struct btrfs_multi_bio *multi = bio->bi_private;
2829 int is_orig_bio = 0;
2832 atomic_inc(&multi->error);
2834 if (bio == multi->orig_bio)
2837 if (atomic_dec_and_test(&multi->stripes_pending)) {
2840 bio = multi->orig_bio;
2842 bio->bi_private = multi->private;
2843 bio->bi_end_io = multi->end_io;
2844 /* only send an error to the higher layers if it is
2845 * beyond the tolerance of the multi-bio
2847 if (atomic_read(&multi->error) > multi->max_errors) {
2851 * this bio is actually up to date, we didn't
2852 * go over the max number of errors
2854 set_bit(BIO_UPTODATE, &bio->bi_flags);
2859 bio_endio(bio, err);
2860 } else if (!is_orig_bio) {
2865 struct async_sched {
2868 struct btrfs_fs_info *info;
2869 struct btrfs_work work;
2873 * see run_scheduled_bios for a description of why bios are collected for
2876 * This will add one bio to the pending list for a device and make sure
2877 * the work struct is scheduled.
2879 static noinline int schedule_bio(struct btrfs_root *root,
2880 struct btrfs_device *device,
2881 int rw, struct bio *bio)
2883 int should_queue = 1;
2884 struct btrfs_pending_bios *pending_bios;
2886 /* don't bother with additional async steps for reads, right now */
2887 if (!(rw & (1 << BIO_RW))) {
2889 submit_bio(rw, bio);
2895 * nr_async_bios allows us to reliably return congestion to the
2896 * higher layers. Otherwise, the async bio makes it appear we have
2897 * made progress against dirty pages when we've really just put it
2898 * on a queue for later
2900 atomic_inc(&root->fs_info->nr_async_bios);
2901 WARN_ON(bio->bi_next);
2902 bio->bi_next = NULL;
2905 spin_lock(&device->io_lock);
2907 pending_bios = &device->pending_sync_bios;
2909 pending_bios = &device->pending_bios;
2911 if (pending_bios->tail)
2912 pending_bios->tail->bi_next = bio;
2914 pending_bios->tail = bio;
2915 if (!pending_bios->head)
2916 pending_bios->head = bio;
2917 if (device->running_pending)
2920 spin_unlock(&device->io_lock);
2923 btrfs_queue_worker(&root->fs_info->submit_workers,
2928 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2929 int mirror_num, int async_submit)
2931 struct btrfs_mapping_tree *map_tree;
2932 struct btrfs_device *dev;
2933 struct bio *first_bio = bio;
2934 u64 logical = (u64)bio->bi_sector << 9;
2937 struct btrfs_multi_bio *multi = NULL;
2942 length = bio->bi_size;
2943 map_tree = &root->fs_info->mapping_tree;
2944 map_length = length;
2946 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2950 total_devs = multi->num_stripes;
2951 if (map_length < length) {
2952 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2953 "len %llu\n", (unsigned long long)logical,
2954 (unsigned long long)length,
2955 (unsigned long long)map_length);
2958 multi->end_io = first_bio->bi_end_io;
2959 multi->private = first_bio->bi_private;
2960 multi->orig_bio = first_bio;
2961 atomic_set(&multi->stripes_pending, multi->num_stripes);
2963 while (dev_nr < total_devs) {
2964 if (total_devs > 1) {
2965 if (dev_nr < total_devs - 1) {
2966 bio = bio_clone(first_bio, GFP_NOFS);
2971 bio->bi_private = multi;
2972 bio->bi_end_io = end_bio_multi_stripe;
2974 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2975 dev = multi->stripes[dev_nr].dev;
2976 BUG_ON(rw == WRITE && !dev->writeable);
2977 if (dev && dev->bdev) {
2978 bio->bi_bdev = dev->bdev;
2980 schedule_bio(root, dev, rw, bio);
2982 submit_bio(rw, bio);
2984 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2985 bio->bi_sector = logical >> 9;
2986 bio_endio(bio, -EIO);
2990 if (total_devs == 1)
2995 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2998 struct btrfs_device *device;
2999 struct btrfs_fs_devices *cur_devices;
3001 cur_devices = root->fs_info->fs_devices;
3002 while (cur_devices) {
3004 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3005 device = __find_device(&cur_devices->devices,
3010 cur_devices = cur_devices->seed;
3015 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3016 u64 devid, u8 *dev_uuid)
3018 struct btrfs_device *device;
3019 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3021 device = kzalloc(sizeof(*device), GFP_NOFS);
3024 list_add(&device->dev_list,
3025 &fs_devices->devices);
3026 device->barriers = 1;
3027 device->dev_root = root->fs_info->dev_root;
3028 device->devid = devid;
3029 device->work.func = pending_bios_fn;
3030 device->fs_devices = fs_devices;
3031 fs_devices->num_devices++;
3032 spin_lock_init(&device->io_lock);
3033 INIT_LIST_HEAD(&device->dev_alloc_list);
3034 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3038 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3039 struct extent_buffer *leaf,
3040 struct btrfs_chunk *chunk)
3042 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3043 struct map_lookup *map;
3044 struct extent_map *em;
3048 u8 uuid[BTRFS_UUID_SIZE];
3053 logical = key->offset;
3054 length = btrfs_chunk_length(leaf, chunk);
3056 spin_lock(&map_tree->map_tree.lock);
3057 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3058 spin_unlock(&map_tree->map_tree.lock);
3060 /* already mapped? */
3061 if (em && em->start <= logical && em->start + em->len > logical) {
3062 free_extent_map(em);
3065 free_extent_map(em);
3068 em = alloc_extent_map(GFP_NOFS);
3071 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3072 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3074 free_extent_map(em);
3078 em->bdev = (struct block_device *)map;
3079 em->start = logical;
3081 em->block_start = 0;
3082 em->block_len = em->len;
3084 map->num_stripes = num_stripes;
3085 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3086 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3087 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3088 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3089 map->type = btrfs_chunk_type(leaf, chunk);
3090 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3091 for (i = 0; i < num_stripes; i++) {
3092 map->stripes[i].physical =
3093 btrfs_stripe_offset_nr(leaf, chunk, i);
3094 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3095 read_extent_buffer(leaf, uuid, (unsigned long)
3096 btrfs_stripe_dev_uuid_nr(chunk, i),
3098 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3100 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3102 free_extent_map(em);
3105 if (!map->stripes[i].dev) {
3106 map->stripes[i].dev =
3107 add_missing_dev(root, devid, uuid);
3108 if (!map->stripes[i].dev) {
3110 free_extent_map(em);
3114 map->stripes[i].dev->in_fs_metadata = 1;
3117 spin_lock(&map_tree->map_tree.lock);
3118 ret = add_extent_mapping(&map_tree->map_tree, em);
3119 spin_unlock(&map_tree->map_tree.lock);
3121 free_extent_map(em);
3126 static int fill_device_from_item(struct extent_buffer *leaf,
3127 struct btrfs_dev_item *dev_item,
3128 struct btrfs_device *device)
3132 device->devid = btrfs_device_id(leaf, dev_item);
3133 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3134 device->total_bytes = device->disk_total_bytes;
3135 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3136 device->type = btrfs_device_type(leaf, dev_item);
3137 device->io_align = btrfs_device_io_align(leaf, dev_item);
3138 device->io_width = btrfs_device_io_width(leaf, dev_item);
3139 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3141 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3142 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3147 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3149 struct btrfs_fs_devices *fs_devices;
3152 mutex_lock(&uuid_mutex);
3154 fs_devices = root->fs_info->fs_devices->seed;
3155 while (fs_devices) {
3156 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3160 fs_devices = fs_devices->seed;
3163 fs_devices = find_fsid(fsid);
3169 fs_devices = clone_fs_devices(fs_devices);
3170 if (IS_ERR(fs_devices)) {
3171 ret = PTR_ERR(fs_devices);
3175 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3176 root->fs_info->bdev_holder);
3180 if (!fs_devices->seeding) {
3181 __btrfs_close_devices(fs_devices);
3182 free_fs_devices(fs_devices);
3187 fs_devices->seed = root->fs_info->fs_devices->seed;
3188 root->fs_info->fs_devices->seed = fs_devices;
3190 mutex_unlock(&uuid_mutex);
3194 static int read_one_dev(struct btrfs_root *root,
3195 struct extent_buffer *leaf,
3196 struct btrfs_dev_item *dev_item)
3198 struct btrfs_device *device;
3201 u8 fs_uuid[BTRFS_UUID_SIZE];
3202 u8 dev_uuid[BTRFS_UUID_SIZE];
3204 devid = btrfs_device_id(leaf, dev_item);
3205 read_extent_buffer(leaf, dev_uuid,
3206 (unsigned long)btrfs_device_uuid(dev_item),
3208 read_extent_buffer(leaf, fs_uuid,
3209 (unsigned long)btrfs_device_fsid(dev_item),
3212 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3213 ret = open_seed_devices(root, fs_uuid);
3214 if (ret && !btrfs_test_opt(root, DEGRADED))
3218 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3219 if (!device || !device->bdev) {
3220 if (!btrfs_test_opt(root, DEGRADED))
3224 printk(KERN_WARNING "warning devid %llu missing\n",
3225 (unsigned long long)devid);
3226 device = add_missing_dev(root, devid, dev_uuid);
3232 if (device->fs_devices != root->fs_info->fs_devices) {
3233 BUG_ON(device->writeable);
3234 if (device->generation !=
3235 btrfs_device_generation(leaf, dev_item))
3239 fill_device_from_item(leaf, dev_item, device);
3240 device->dev_root = root->fs_info->dev_root;
3241 device->in_fs_metadata = 1;
3242 if (device->writeable)
3243 device->fs_devices->total_rw_bytes += device->total_bytes;
3248 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3250 struct btrfs_dev_item *dev_item;
3252 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3254 return read_one_dev(root, buf, dev_item);
3257 int btrfs_read_sys_array(struct btrfs_root *root)
3259 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3260 struct extent_buffer *sb;
3261 struct btrfs_disk_key *disk_key;
3262 struct btrfs_chunk *chunk;
3264 unsigned long sb_ptr;
3270 struct btrfs_key key;
3272 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3273 BTRFS_SUPER_INFO_SIZE);
3276 btrfs_set_buffer_uptodate(sb);
3277 btrfs_set_buffer_lockdep_class(sb, 0);
3279 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3280 array_size = btrfs_super_sys_array_size(super_copy);
3282 ptr = super_copy->sys_chunk_array;
3283 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3286 while (cur < array_size) {
3287 disk_key = (struct btrfs_disk_key *)ptr;
3288 btrfs_disk_key_to_cpu(&key, disk_key);
3290 len = sizeof(*disk_key); ptr += len;
3294 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3295 chunk = (struct btrfs_chunk *)sb_ptr;
3296 ret = read_one_chunk(root, &key, sb, chunk);
3299 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3300 len = btrfs_chunk_item_size(num_stripes);
3309 free_extent_buffer(sb);
3313 int btrfs_read_chunk_tree(struct btrfs_root *root)
3315 struct btrfs_path *path;
3316 struct extent_buffer *leaf;
3317 struct btrfs_key key;
3318 struct btrfs_key found_key;
3322 root = root->fs_info->chunk_root;
3324 path = btrfs_alloc_path();
3328 /* first we search for all of the device items, and then we
3329 * read in all of the chunk items. This way we can create chunk
3330 * mappings that reference all of the devices that are afound
3332 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3336 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3338 leaf = path->nodes[0];
3339 slot = path->slots[0];
3340 if (slot >= btrfs_header_nritems(leaf)) {
3341 ret = btrfs_next_leaf(root, path);
3348 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3349 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3350 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3352 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3353 struct btrfs_dev_item *dev_item;
3354 dev_item = btrfs_item_ptr(leaf, slot,
3355 struct btrfs_dev_item);
3356 ret = read_one_dev(root, leaf, dev_item);
3360 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3361 struct btrfs_chunk *chunk;
3362 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3363 ret = read_one_chunk(root, &found_key, leaf, chunk);
3369 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3371 btrfs_release_path(root, path);
3376 btrfs_free_path(path);