2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/slab.h>
21 #include <linux/buffer_head.h>
22 #include <linux/blkdev.h>
23 #include <linux/random.h>
24 #include <linux/iocontext.h>
25 #include <asm/div64.h>
28 #include "extent_map.h"
30 #include "transaction.h"
31 #include "print-tree.h"
33 #include "async-thread.h"
43 struct btrfs_bio_stripe stripes[];
46 static int init_first_rw_device(struct btrfs_trans_handle *trans,
47 struct btrfs_root *root,
48 struct btrfs_device *device);
49 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
51 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
52 (sizeof(struct btrfs_bio_stripe) * (n)))
54 static DEFINE_MUTEX(uuid_mutex);
55 static LIST_HEAD(fs_uuids);
57 void btrfs_lock_volumes(void)
59 mutex_lock(&uuid_mutex);
62 void btrfs_unlock_volumes(void)
64 mutex_unlock(&uuid_mutex);
67 static void lock_chunks(struct btrfs_root *root)
69 mutex_lock(&root->fs_info->chunk_mutex);
72 static void unlock_chunks(struct btrfs_root *root)
74 mutex_unlock(&root->fs_info->chunk_mutex);
77 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
79 struct btrfs_device *device;
80 WARN_ON(fs_devices->opened);
81 while (!list_empty(&fs_devices->devices)) {
82 device = list_entry(fs_devices->devices.next,
83 struct btrfs_device, dev_list);
84 list_del(&device->dev_list);
91 int btrfs_cleanup_fs_uuids(void)
93 struct btrfs_fs_devices *fs_devices;
95 while (!list_empty(&fs_uuids)) {
96 fs_devices = list_entry(fs_uuids.next,
97 struct btrfs_fs_devices, list);
98 list_del(&fs_devices->list);
99 free_fs_devices(fs_devices);
104 static noinline struct btrfs_device *__find_device(struct list_head *head,
107 struct btrfs_device *dev;
109 list_for_each_entry(dev, head, dev_list) {
110 if (dev->devid == devid &&
111 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
118 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
120 struct btrfs_fs_devices *fs_devices;
122 list_for_each_entry(fs_devices, &fs_uuids, list) {
123 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
129 static void requeue_list(struct btrfs_pending_bios *pending_bios,
130 struct bio *head, struct bio *tail)
133 struct bio *old_head;
135 old_head = pending_bios->head;
136 pending_bios->head = head;
137 if (pending_bios->tail)
138 tail->bi_next = old_head;
140 pending_bios->tail = tail;
144 * we try to collect pending bios for a device so we don't get a large
145 * number of procs sending bios down to the same device. This greatly
146 * improves the schedulers ability to collect and merge the bios.
148 * But, it also turns into a long list of bios to process and that is sure
149 * to eventually make the worker thread block. The solution here is to
150 * make some progress and then put this work struct back at the end of
151 * the list if the block device is congested. This way, multiple devices
152 * can make progress from a single worker thread.
154 static noinline int run_scheduled_bios(struct btrfs_device *device)
157 struct backing_dev_info *bdi;
158 struct btrfs_fs_info *fs_info;
159 struct btrfs_pending_bios *pending_bios;
163 unsigned long num_run;
164 unsigned long num_sync_run;
165 unsigned long batch_run = 0;
167 unsigned long last_waited = 0;
170 bdi = blk_get_backing_dev_info(device->bdev);
171 fs_info = device->dev_root->fs_info;
172 limit = btrfs_async_submit_limit(fs_info);
173 limit = limit * 2 / 3;
175 /* we want to make sure that every time we switch from the sync
176 * list to the normal list, we unplug
181 spin_lock(&device->io_lock);
186 /* take all the bios off the list at once and process them
187 * later on (without the lock held). But, remember the
188 * tail and other pointers so the bios can be properly reinserted
189 * into the list if we hit congestion
191 if (!force_reg && device->pending_sync_bios.head) {
192 pending_bios = &device->pending_sync_bios;
195 pending_bios = &device->pending_bios;
199 pending = pending_bios->head;
200 tail = pending_bios->tail;
201 WARN_ON(pending && !tail);
204 * if pending was null this time around, no bios need processing
205 * at all and we can stop. Otherwise it'll loop back up again
206 * and do an additional check so no bios are missed.
208 * device->running_pending is used to synchronize with the
211 if (device->pending_sync_bios.head == NULL &&
212 device->pending_bios.head == NULL) {
214 device->running_pending = 0;
217 device->running_pending = 1;
220 pending_bios->head = NULL;
221 pending_bios->tail = NULL;
223 spin_unlock(&device->io_lock);
226 * if we're doing the regular priority list, make sure we unplug
227 * for any high prio bios we've sent down
229 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
231 blk_run_backing_dev(bdi, NULL);
237 /* we want to work on both lists, but do more bios on the
238 * sync list than the regular list
241 pending_bios != &device->pending_sync_bios &&
242 device->pending_sync_bios.head) ||
243 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
244 device->pending_bios.head)) {
245 spin_lock(&device->io_lock);
246 requeue_list(pending_bios, pending, tail);
251 pending = pending->bi_next;
253 atomic_dec(&fs_info->nr_async_bios);
255 if (atomic_read(&fs_info->nr_async_bios) < limit &&
256 waitqueue_active(&fs_info->async_submit_wait))
257 wake_up(&fs_info->async_submit_wait);
259 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
261 if (cur->bi_rw & REQ_SYNC)
264 submit_bio(cur->bi_rw, cur);
267 if (need_resched()) {
269 blk_run_backing_dev(bdi, NULL);
276 * we made progress, there is more work to do and the bdi
277 * is now congested. Back off and let other work structs
280 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
281 fs_info->fs_devices->open_devices > 1) {
282 struct io_context *ioc;
284 ioc = current->io_context;
287 * the main goal here is that we don't want to
288 * block if we're going to be able to submit
289 * more requests without blocking.
291 * This code does two great things, it pokes into
292 * the elevator code from a filesystem _and_
293 * it makes assumptions about how batching works.
295 if (ioc && ioc->nr_batch_requests > 0 &&
296 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
298 ioc->last_waited == last_waited)) {
300 * we want to go through our batch of
301 * requests and stop. So, we copy out
302 * the ioc->last_waited time and test
303 * against it before looping
305 last_waited = ioc->last_waited;
306 if (need_resched()) {
308 blk_run_backing_dev(bdi, NULL);
315 spin_lock(&device->io_lock);
316 requeue_list(pending_bios, pending, tail);
317 device->running_pending = 1;
319 spin_unlock(&device->io_lock);
320 btrfs_requeue_work(&device->work);
327 blk_run_backing_dev(bdi, NULL);
330 * IO has already been through a long path to get here. Checksumming,
331 * async helper threads, perhaps compression. We've done a pretty
332 * good job of collecting a batch of IO and should just unplug
333 * the device right away.
335 * This will help anyone who is waiting on the IO, they might have
336 * already unplugged, but managed to do so before the bio they
337 * cared about found its way down here.
339 blk_run_backing_dev(bdi, NULL);
345 spin_lock(&device->io_lock);
346 if (device->pending_bios.head || device->pending_sync_bios.head)
348 spin_unlock(&device->io_lock);
354 static void pending_bios_fn(struct btrfs_work *work)
356 struct btrfs_device *device;
358 device = container_of(work, struct btrfs_device, work);
359 run_scheduled_bios(device);
362 static noinline int device_list_add(const char *path,
363 struct btrfs_super_block *disk_super,
364 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
366 struct btrfs_device *device;
367 struct btrfs_fs_devices *fs_devices;
368 u64 found_transid = btrfs_super_generation(disk_super);
371 fs_devices = find_fsid(disk_super->fsid);
373 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
376 INIT_LIST_HEAD(&fs_devices->devices);
377 INIT_LIST_HEAD(&fs_devices->alloc_list);
378 list_add(&fs_devices->list, &fs_uuids);
379 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
380 fs_devices->latest_devid = devid;
381 fs_devices->latest_trans = found_transid;
382 mutex_init(&fs_devices->device_list_mutex);
385 device = __find_device(&fs_devices->devices, devid,
386 disk_super->dev_item.uuid);
389 if (fs_devices->opened)
392 device = kzalloc(sizeof(*device), GFP_NOFS);
394 /* we can safely leave the fs_devices entry around */
397 device->devid = devid;
398 device->work.func = pending_bios_fn;
399 memcpy(device->uuid, disk_super->dev_item.uuid,
401 spin_lock_init(&device->io_lock);
402 device->name = kstrdup(path, GFP_NOFS);
407 INIT_LIST_HEAD(&device->dev_alloc_list);
409 mutex_lock(&fs_devices->device_list_mutex);
410 list_add(&device->dev_list, &fs_devices->devices);
411 mutex_unlock(&fs_devices->device_list_mutex);
413 device->fs_devices = fs_devices;
414 fs_devices->num_devices++;
415 } else if (strcmp(device->name, path)) {
416 name = kstrdup(path, GFP_NOFS);
423 if (found_transid > fs_devices->latest_trans) {
424 fs_devices->latest_devid = devid;
425 fs_devices->latest_trans = found_transid;
427 *fs_devices_ret = fs_devices;
431 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
433 struct btrfs_fs_devices *fs_devices;
434 struct btrfs_device *device;
435 struct btrfs_device *orig_dev;
437 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
439 return ERR_PTR(-ENOMEM);
441 INIT_LIST_HEAD(&fs_devices->devices);
442 INIT_LIST_HEAD(&fs_devices->alloc_list);
443 INIT_LIST_HEAD(&fs_devices->list);
444 mutex_init(&fs_devices->device_list_mutex);
445 fs_devices->latest_devid = orig->latest_devid;
446 fs_devices->latest_trans = orig->latest_trans;
447 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
449 mutex_lock(&orig->device_list_mutex);
450 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
451 device = kzalloc(sizeof(*device), GFP_NOFS);
455 device->name = kstrdup(orig_dev->name, GFP_NOFS);
461 device->devid = orig_dev->devid;
462 device->work.func = pending_bios_fn;
463 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
464 spin_lock_init(&device->io_lock);
465 INIT_LIST_HEAD(&device->dev_list);
466 INIT_LIST_HEAD(&device->dev_alloc_list);
468 list_add(&device->dev_list, &fs_devices->devices);
469 device->fs_devices = fs_devices;
470 fs_devices->num_devices++;
472 mutex_unlock(&orig->device_list_mutex);
475 mutex_unlock(&orig->device_list_mutex);
476 free_fs_devices(fs_devices);
477 return ERR_PTR(-ENOMEM);
480 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
482 struct btrfs_device *device, *next;
484 mutex_lock(&uuid_mutex);
486 mutex_lock(&fs_devices->device_list_mutex);
487 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
488 if (device->in_fs_metadata)
492 close_bdev_exclusive(device->bdev, device->mode);
494 fs_devices->open_devices--;
496 if (device->writeable) {
497 list_del_init(&device->dev_alloc_list);
498 device->writeable = 0;
499 fs_devices->rw_devices--;
501 list_del_init(&device->dev_list);
502 fs_devices->num_devices--;
506 mutex_unlock(&fs_devices->device_list_mutex);
508 if (fs_devices->seed) {
509 fs_devices = fs_devices->seed;
513 mutex_unlock(&uuid_mutex);
517 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
519 struct btrfs_device *device;
521 if (--fs_devices->opened > 0)
524 list_for_each_entry(device, &fs_devices->devices, dev_list) {
526 close_bdev_exclusive(device->bdev, device->mode);
527 fs_devices->open_devices--;
529 if (device->writeable) {
530 list_del_init(&device->dev_alloc_list);
531 fs_devices->rw_devices--;
535 device->writeable = 0;
536 device->in_fs_metadata = 0;
538 WARN_ON(fs_devices->open_devices);
539 WARN_ON(fs_devices->rw_devices);
540 fs_devices->opened = 0;
541 fs_devices->seeding = 0;
546 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
548 struct btrfs_fs_devices *seed_devices = NULL;
551 mutex_lock(&uuid_mutex);
552 ret = __btrfs_close_devices(fs_devices);
553 if (!fs_devices->opened) {
554 seed_devices = fs_devices->seed;
555 fs_devices->seed = NULL;
557 mutex_unlock(&uuid_mutex);
559 while (seed_devices) {
560 fs_devices = seed_devices;
561 seed_devices = fs_devices->seed;
562 __btrfs_close_devices(fs_devices);
563 free_fs_devices(fs_devices);
568 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
569 fmode_t flags, void *holder)
571 struct block_device *bdev;
572 struct list_head *head = &fs_devices->devices;
573 struct btrfs_device *device;
574 struct block_device *latest_bdev = NULL;
575 struct buffer_head *bh;
576 struct btrfs_super_block *disk_super;
577 u64 latest_devid = 0;
578 u64 latest_transid = 0;
583 list_for_each_entry(device, head, dev_list) {
589 bdev = open_bdev_exclusive(device->name, flags, holder);
591 printk(KERN_INFO "open %s failed\n", device->name);
594 set_blocksize(bdev, 4096);
596 bh = btrfs_read_dev_super(bdev);
600 disk_super = (struct btrfs_super_block *)bh->b_data;
601 devid = btrfs_stack_device_id(&disk_super->dev_item);
602 if (devid != device->devid)
605 if (memcmp(device->uuid, disk_super->dev_item.uuid,
609 device->generation = btrfs_super_generation(disk_super);
610 if (!latest_transid || device->generation > latest_transid) {
611 latest_devid = devid;
612 latest_transid = device->generation;
616 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
617 device->writeable = 0;
619 device->writeable = !bdev_read_only(bdev);
624 device->in_fs_metadata = 0;
625 device->mode = flags;
627 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
628 fs_devices->rotating = 1;
630 fs_devices->open_devices++;
631 if (device->writeable) {
632 fs_devices->rw_devices++;
633 list_add(&device->dev_alloc_list,
634 &fs_devices->alloc_list);
641 close_bdev_exclusive(bdev, FMODE_READ);
645 if (fs_devices->open_devices == 0) {
649 fs_devices->seeding = seeding;
650 fs_devices->opened = 1;
651 fs_devices->latest_bdev = latest_bdev;
652 fs_devices->latest_devid = latest_devid;
653 fs_devices->latest_trans = latest_transid;
654 fs_devices->total_rw_bytes = 0;
659 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
660 fmode_t flags, void *holder)
664 mutex_lock(&uuid_mutex);
665 if (fs_devices->opened) {
666 fs_devices->opened++;
669 ret = __btrfs_open_devices(fs_devices, flags, holder);
671 mutex_unlock(&uuid_mutex);
675 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
676 struct btrfs_fs_devices **fs_devices_ret)
678 struct btrfs_super_block *disk_super;
679 struct block_device *bdev;
680 struct buffer_head *bh;
685 mutex_lock(&uuid_mutex);
687 bdev = open_bdev_exclusive(path, flags, holder);
694 ret = set_blocksize(bdev, 4096);
697 bh = btrfs_read_dev_super(bdev);
702 disk_super = (struct btrfs_super_block *)bh->b_data;
703 devid = btrfs_stack_device_id(&disk_super->dev_item);
704 transid = btrfs_super_generation(disk_super);
705 if (disk_super->label[0])
706 printk(KERN_INFO "device label %s ", disk_super->label);
708 /* FIXME, make a readl uuid parser */
709 printk(KERN_INFO "device fsid %llx-%llx ",
710 *(unsigned long long *)disk_super->fsid,
711 *(unsigned long long *)(disk_super->fsid + 8));
713 printk(KERN_CONT "devid %llu transid %llu %s\n",
714 (unsigned long long)devid, (unsigned long long)transid, path);
715 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
719 close_bdev_exclusive(bdev, flags);
721 mutex_unlock(&uuid_mutex);
726 * this uses a pretty simple search, the expectation is that it is
727 * called very infrequently and that a given device has a small number
730 int find_free_dev_extent(struct btrfs_trans_handle *trans,
731 struct btrfs_device *device, u64 num_bytes,
732 u64 *start, u64 *max_avail)
734 struct btrfs_key key;
735 struct btrfs_root *root = device->dev_root;
736 struct btrfs_dev_extent *dev_extent = NULL;
737 struct btrfs_path *path;
740 u64 search_start = 0;
741 u64 search_end = device->total_bytes;
745 struct extent_buffer *l;
747 path = btrfs_alloc_path();
753 /* FIXME use last free of some kind */
755 /* we don't want to overwrite the superblock on the drive,
756 * so we make sure to start at an offset of at least 1MB
758 search_start = max((u64)1024 * 1024, search_start);
760 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
761 search_start = max(root->fs_info->alloc_start, search_start);
763 key.objectid = device->devid;
764 key.offset = search_start;
765 key.type = BTRFS_DEV_EXTENT_KEY;
766 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
770 ret = btrfs_previous_item(root, path, key.objectid, key.type);
777 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
780 slot = path->slots[0];
781 if (slot >= btrfs_header_nritems(l)) {
782 ret = btrfs_next_leaf(root, path);
789 if (search_start >= search_end) {
793 *start = search_start;
797 *start = last_byte > search_start ?
798 last_byte : search_start;
799 if (search_end <= *start) {
805 btrfs_item_key_to_cpu(l, &key, slot);
807 if (key.objectid < device->devid)
810 if (key.objectid > device->devid)
813 if (key.offset >= search_start && key.offset > last_byte &&
815 if (last_byte < search_start)
816 last_byte = search_start;
817 hole_size = key.offset - last_byte;
819 if (hole_size > *max_avail)
820 *max_avail = hole_size;
822 if (key.offset > last_byte &&
823 hole_size >= num_bytes) {
828 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
832 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
833 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
839 /* we have to make sure we didn't find an extent that has already
840 * been allocated by the map tree or the original allocation
842 BUG_ON(*start < search_start);
844 if (*start + num_bytes > search_end) {
848 /* check for pending inserts here */
852 btrfs_free_path(path);
856 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
857 struct btrfs_device *device,
861 struct btrfs_path *path;
862 struct btrfs_root *root = device->dev_root;
863 struct btrfs_key key;
864 struct btrfs_key found_key;
865 struct extent_buffer *leaf = NULL;
866 struct btrfs_dev_extent *extent = NULL;
868 path = btrfs_alloc_path();
872 key.objectid = device->devid;
874 key.type = BTRFS_DEV_EXTENT_KEY;
876 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
878 ret = btrfs_previous_item(root, path, key.objectid,
879 BTRFS_DEV_EXTENT_KEY);
881 leaf = path->nodes[0];
882 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
883 extent = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_dev_extent);
885 BUG_ON(found_key.offset > start || found_key.offset +
886 btrfs_dev_extent_length(leaf, extent) < start);
888 } else if (ret == 0) {
889 leaf = path->nodes[0];
890 extent = btrfs_item_ptr(leaf, path->slots[0],
891 struct btrfs_dev_extent);
895 if (device->bytes_used > 0)
896 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
897 ret = btrfs_del_item(trans, root, path);
900 btrfs_free_path(path);
904 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
905 struct btrfs_device *device,
906 u64 chunk_tree, u64 chunk_objectid,
907 u64 chunk_offset, u64 start, u64 num_bytes)
910 struct btrfs_path *path;
911 struct btrfs_root *root = device->dev_root;
912 struct btrfs_dev_extent *extent;
913 struct extent_buffer *leaf;
914 struct btrfs_key key;
916 WARN_ON(!device->in_fs_metadata);
917 path = btrfs_alloc_path();
921 key.objectid = device->devid;
923 key.type = BTRFS_DEV_EXTENT_KEY;
924 ret = btrfs_insert_empty_item(trans, root, path, &key,
928 leaf = path->nodes[0];
929 extent = btrfs_item_ptr(leaf, path->slots[0],
930 struct btrfs_dev_extent);
931 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
932 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
933 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
935 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
936 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
939 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
940 btrfs_mark_buffer_dirty(leaf);
941 btrfs_free_path(path);
945 static noinline int find_next_chunk(struct btrfs_root *root,
946 u64 objectid, u64 *offset)
948 struct btrfs_path *path;
950 struct btrfs_key key;
951 struct btrfs_chunk *chunk;
952 struct btrfs_key found_key;
954 path = btrfs_alloc_path();
957 key.objectid = objectid;
958 key.offset = (u64)-1;
959 key.type = BTRFS_CHUNK_ITEM_KEY;
961 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
967 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
971 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
973 if (found_key.objectid != objectid)
976 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
978 *offset = found_key.offset +
979 btrfs_chunk_length(path->nodes[0], chunk);
984 btrfs_free_path(path);
988 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
991 struct btrfs_key key;
992 struct btrfs_key found_key;
993 struct btrfs_path *path;
995 root = root->fs_info->chunk_root;
997 path = btrfs_alloc_path();
1001 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1002 key.type = BTRFS_DEV_ITEM_KEY;
1003 key.offset = (u64)-1;
1005 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1011 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1012 BTRFS_DEV_ITEM_KEY);
1016 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1018 *objectid = found_key.offset + 1;
1022 btrfs_free_path(path);
1027 * the device information is stored in the chunk root
1028 * the btrfs_device struct should be fully filled in
1030 int btrfs_add_device(struct btrfs_trans_handle *trans,
1031 struct btrfs_root *root,
1032 struct btrfs_device *device)
1035 struct btrfs_path *path;
1036 struct btrfs_dev_item *dev_item;
1037 struct extent_buffer *leaf;
1038 struct btrfs_key key;
1041 root = root->fs_info->chunk_root;
1043 path = btrfs_alloc_path();
1047 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1048 key.type = BTRFS_DEV_ITEM_KEY;
1049 key.offset = device->devid;
1051 ret = btrfs_insert_empty_item(trans, root, path, &key,
1056 leaf = path->nodes[0];
1057 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1059 btrfs_set_device_id(leaf, dev_item, device->devid);
1060 btrfs_set_device_generation(leaf, dev_item, 0);
1061 btrfs_set_device_type(leaf, dev_item, device->type);
1062 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1063 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1064 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1065 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1066 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1067 btrfs_set_device_group(leaf, dev_item, 0);
1068 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1069 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1070 btrfs_set_device_start_offset(leaf, dev_item, 0);
1072 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1073 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1074 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1075 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1076 btrfs_mark_buffer_dirty(leaf);
1080 btrfs_free_path(path);
1084 static int btrfs_rm_dev_item(struct btrfs_root *root,
1085 struct btrfs_device *device)
1088 struct btrfs_path *path;
1089 struct btrfs_key key;
1090 struct btrfs_trans_handle *trans;
1092 root = root->fs_info->chunk_root;
1094 path = btrfs_alloc_path();
1098 trans = btrfs_start_transaction(root, 0);
1099 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1100 key.type = BTRFS_DEV_ITEM_KEY;
1101 key.offset = device->devid;
1104 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1113 ret = btrfs_del_item(trans, root, path);
1117 btrfs_free_path(path);
1118 unlock_chunks(root);
1119 btrfs_commit_transaction(trans, root);
1123 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1125 struct btrfs_device *device;
1126 struct btrfs_device *next_device;
1127 struct block_device *bdev;
1128 struct buffer_head *bh = NULL;
1129 struct btrfs_super_block *disk_super;
1136 mutex_lock(&uuid_mutex);
1137 mutex_lock(&root->fs_info->volume_mutex);
1139 all_avail = root->fs_info->avail_data_alloc_bits |
1140 root->fs_info->avail_system_alloc_bits |
1141 root->fs_info->avail_metadata_alloc_bits;
1143 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1144 root->fs_info->fs_devices->num_devices <= 4) {
1145 printk(KERN_ERR "btrfs: unable to go below four devices "
1151 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1152 root->fs_info->fs_devices->num_devices <= 2) {
1153 printk(KERN_ERR "btrfs: unable to go below two "
1154 "devices on raid1\n");
1159 if (strcmp(device_path, "missing") == 0) {
1160 struct list_head *devices;
1161 struct btrfs_device *tmp;
1164 devices = &root->fs_info->fs_devices->devices;
1165 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1166 list_for_each_entry(tmp, devices, dev_list) {
1167 if (tmp->in_fs_metadata && !tmp->bdev) {
1172 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1177 printk(KERN_ERR "btrfs: no missing devices found to "
1182 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1183 root->fs_info->bdev_holder);
1185 ret = PTR_ERR(bdev);
1189 set_blocksize(bdev, 4096);
1190 bh = btrfs_read_dev_super(bdev);
1195 disk_super = (struct btrfs_super_block *)bh->b_data;
1196 devid = btrfs_stack_device_id(&disk_super->dev_item);
1197 dev_uuid = disk_super->dev_item.uuid;
1198 device = btrfs_find_device(root, devid, dev_uuid,
1206 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1207 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1213 if (device->writeable) {
1214 list_del_init(&device->dev_alloc_list);
1215 root->fs_info->fs_devices->rw_devices--;
1218 ret = btrfs_shrink_device(device, 0);
1222 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1226 device->in_fs_metadata = 0;
1229 * the device list mutex makes sure that we don't change
1230 * the device list while someone else is writing out all
1231 * the device supers.
1233 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1234 list_del_init(&device->dev_list);
1235 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1237 device->fs_devices->num_devices--;
1239 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1240 struct btrfs_device, dev_list);
1241 if (device->bdev == root->fs_info->sb->s_bdev)
1242 root->fs_info->sb->s_bdev = next_device->bdev;
1243 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1244 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1247 close_bdev_exclusive(device->bdev, device->mode);
1248 device->bdev = NULL;
1249 device->fs_devices->open_devices--;
1252 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1253 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1255 if (device->fs_devices->open_devices == 0) {
1256 struct btrfs_fs_devices *fs_devices;
1257 fs_devices = root->fs_info->fs_devices;
1258 while (fs_devices) {
1259 if (fs_devices->seed == device->fs_devices)
1261 fs_devices = fs_devices->seed;
1263 fs_devices->seed = device->fs_devices->seed;
1264 device->fs_devices->seed = NULL;
1265 __btrfs_close_devices(device->fs_devices);
1266 free_fs_devices(device->fs_devices);
1270 * at this point, the device is zero sized. We want to
1271 * remove it from the devices list and zero out the old super
1273 if (device->writeable) {
1274 /* make sure this device isn't detected as part of
1277 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1278 set_buffer_dirty(bh);
1279 sync_dirty_buffer(bh);
1282 kfree(device->name);
1290 close_bdev_exclusive(bdev, FMODE_READ);
1292 mutex_unlock(&root->fs_info->volume_mutex);
1293 mutex_unlock(&uuid_mutex);
1298 * does all the dirty work required for changing file system's UUID.
1300 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1301 struct btrfs_root *root)
1303 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1304 struct btrfs_fs_devices *old_devices;
1305 struct btrfs_fs_devices *seed_devices;
1306 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1307 struct btrfs_device *device;
1310 BUG_ON(!mutex_is_locked(&uuid_mutex));
1311 if (!fs_devices->seeding)
1314 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1318 old_devices = clone_fs_devices(fs_devices);
1319 if (IS_ERR(old_devices)) {
1320 kfree(seed_devices);
1321 return PTR_ERR(old_devices);
1324 list_add(&old_devices->list, &fs_uuids);
1326 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1327 seed_devices->opened = 1;
1328 INIT_LIST_HEAD(&seed_devices->devices);
1329 INIT_LIST_HEAD(&seed_devices->alloc_list);
1330 mutex_init(&seed_devices->device_list_mutex);
1331 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1332 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1333 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1334 device->fs_devices = seed_devices;
1337 fs_devices->seeding = 0;
1338 fs_devices->num_devices = 0;
1339 fs_devices->open_devices = 0;
1340 fs_devices->seed = seed_devices;
1342 generate_random_uuid(fs_devices->fsid);
1343 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1344 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1345 super_flags = btrfs_super_flags(disk_super) &
1346 ~BTRFS_SUPER_FLAG_SEEDING;
1347 btrfs_set_super_flags(disk_super, super_flags);
1353 * strore the expected generation for seed devices in device items.
1355 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1356 struct btrfs_root *root)
1358 struct btrfs_path *path;
1359 struct extent_buffer *leaf;
1360 struct btrfs_dev_item *dev_item;
1361 struct btrfs_device *device;
1362 struct btrfs_key key;
1363 u8 fs_uuid[BTRFS_UUID_SIZE];
1364 u8 dev_uuid[BTRFS_UUID_SIZE];
1368 path = btrfs_alloc_path();
1372 root = root->fs_info->chunk_root;
1373 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1375 key.type = BTRFS_DEV_ITEM_KEY;
1378 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1382 leaf = path->nodes[0];
1384 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1385 ret = btrfs_next_leaf(root, path);
1390 leaf = path->nodes[0];
1391 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1392 btrfs_release_path(root, path);
1396 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1397 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1398 key.type != BTRFS_DEV_ITEM_KEY)
1401 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1402 struct btrfs_dev_item);
1403 devid = btrfs_device_id(leaf, dev_item);
1404 read_extent_buffer(leaf, dev_uuid,
1405 (unsigned long)btrfs_device_uuid(dev_item),
1407 read_extent_buffer(leaf, fs_uuid,
1408 (unsigned long)btrfs_device_fsid(dev_item),
1410 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1413 if (device->fs_devices->seeding) {
1414 btrfs_set_device_generation(leaf, dev_item,
1415 device->generation);
1416 btrfs_mark_buffer_dirty(leaf);
1424 btrfs_free_path(path);
1428 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1430 struct btrfs_trans_handle *trans;
1431 struct btrfs_device *device;
1432 struct block_device *bdev;
1433 struct list_head *devices;
1434 struct super_block *sb = root->fs_info->sb;
1436 int seeding_dev = 0;
1439 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1442 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1444 return PTR_ERR(bdev);
1446 if (root->fs_info->fs_devices->seeding) {
1448 down_write(&sb->s_umount);
1449 mutex_lock(&uuid_mutex);
1452 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1453 mutex_lock(&root->fs_info->volume_mutex);
1455 devices = &root->fs_info->fs_devices->devices;
1457 * we have the volume lock, so we don't need the extra
1458 * device list mutex while reading the list here.
1460 list_for_each_entry(device, devices, dev_list) {
1461 if (device->bdev == bdev) {
1467 device = kzalloc(sizeof(*device), GFP_NOFS);
1469 /* we can safely leave the fs_devices entry around */
1474 device->name = kstrdup(device_path, GFP_NOFS);
1475 if (!device->name) {
1481 ret = find_next_devid(root, &device->devid);
1487 trans = btrfs_start_transaction(root, 0);
1490 device->writeable = 1;
1491 device->work.func = pending_bios_fn;
1492 generate_random_uuid(device->uuid);
1493 spin_lock_init(&device->io_lock);
1494 device->generation = trans->transid;
1495 device->io_width = root->sectorsize;
1496 device->io_align = root->sectorsize;
1497 device->sector_size = root->sectorsize;
1498 device->total_bytes = i_size_read(bdev->bd_inode);
1499 device->disk_total_bytes = device->total_bytes;
1500 device->dev_root = root->fs_info->dev_root;
1501 device->bdev = bdev;
1502 device->in_fs_metadata = 1;
1504 set_blocksize(device->bdev, 4096);
1507 sb->s_flags &= ~MS_RDONLY;
1508 ret = btrfs_prepare_sprout(trans, root);
1512 device->fs_devices = root->fs_info->fs_devices;
1515 * we don't want write_supers to jump in here with our device
1518 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1519 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1520 list_add(&device->dev_alloc_list,
1521 &root->fs_info->fs_devices->alloc_list);
1522 root->fs_info->fs_devices->num_devices++;
1523 root->fs_info->fs_devices->open_devices++;
1524 root->fs_info->fs_devices->rw_devices++;
1525 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1527 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1528 root->fs_info->fs_devices->rotating = 1;
1530 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1531 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1532 total_bytes + device->total_bytes);
1534 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1535 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1537 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1540 ret = init_first_rw_device(trans, root, device);
1542 ret = btrfs_finish_sprout(trans, root);
1545 ret = btrfs_add_device(trans, root, device);
1549 * we've got more storage, clear any full flags on the space
1552 btrfs_clear_space_info_full(root->fs_info);
1554 unlock_chunks(root);
1555 btrfs_commit_transaction(trans, root);
1558 mutex_unlock(&uuid_mutex);
1559 up_write(&sb->s_umount);
1561 ret = btrfs_relocate_sys_chunks(root);
1565 mutex_unlock(&root->fs_info->volume_mutex);
1568 close_bdev_exclusive(bdev, 0);
1570 mutex_unlock(&uuid_mutex);
1571 up_write(&sb->s_umount);
1576 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1577 struct btrfs_device *device)
1580 struct btrfs_path *path;
1581 struct btrfs_root *root;
1582 struct btrfs_dev_item *dev_item;
1583 struct extent_buffer *leaf;
1584 struct btrfs_key key;
1586 root = device->dev_root->fs_info->chunk_root;
1588 path = btrfs_alloc_path();
1592 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1593 key.type = BTRFS_DEV_ITEM_KEY;
1594 key.offset = device->devid;
1596 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1605 leaf = path->nodes[0];
1606 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1608 btrfs_set_device_id(leaf, dev_item, device->devid);
1609 btrfs_set_device_type(leaf, dev_item, device->type);
1610 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1611 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1612 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1613 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1614 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1615 btrfs_mark_buffer_dirty(leaf);
1618 btrfs_free_path(path);
1622 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1623 struct btrfs_device *device, u64 new_size)
1625 struct btrfs_super_block *super_copy =
1626 &device->dev_root->fs_info->super_copy;
1627 u64 old_total = btrfs_super_total_bytes(super_copy);
1628 u64 diff = new_size - device->total_bytes;
1630 if (!device->writeable)
1632 if (new_size <= device->total_bytes)
1635 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1636 device->fs_devices->total_rw_bytes += diff;
1638 device->total_bytes = new_size;
1639 device->disk_total_bytes = new_size;
1640 btrfs_clear_space_info_full(device->dev_root->fs_info);
1642 return btrfs_update_device(trans, device);
1645 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1646 struct btrfs_device *device, u64 new_size)
1649 lock_chunks(device->dev_root);
1650 ret = __btrfs_grow_device(trans, device, new_size);
1651 unlock_chunks(device->dev_root);
1655 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1656 struct btrfs_root *root,
1657 u64 chunk_tree, u64 chunk_objectid,
1661 struct btrfs_path *path;
1662 struct btrfs_key key;
1664 root = root->fs_info->chunk_root;
1665 path = btrfs_alloc_path();
1669 key.objectid = chunk_objectid;
1670 key.offset = chunk_offset;
1671 key.type = BTRFS_CHUNK_ITEM_KEY;
1673 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1676 ret = btrfs_del_item(trans, root, path);
1679 btrfs_free_path(path);
1683 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1686 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1687 struct btrfs_disk_key *disk_key;
1688 struct btrfs_chunk *chunk;
1695 struct btrfs_key key;
1697 array_size = btrfs_super_sys_array_size(super_copy);
1699 ptr = super_copy->sys_chunk_array;
1702 while (cur < array_size) {
1703 disk_key = (struct btrfs_disk_key *)ptr;
1704 btrfs_disk_key_to_cpu(&key, disk_key);
1706 len = sizeof(*disk_key);
1708 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1709 chunk = (struct btrfs_chunk *)(ptr + len);
1710 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1711 len += btrfs_chunk_item_size(num_stripes);
1716 if (key.objectid == chunk_objectid &&
1717 key.offset == chunk_offset) {
1718 memmove(ptr, ptr + len, array_size - (cur + len));
1720 btrfs_set_super_sys_array_size(super_copy, array_size);
1729 static int btrfs_relocate_chunk(struct btrfs_root *root,
1730 u64 chunk_tree, u64 chunk_objectid,
1733 struct extent_map_tree *em_tree;
1734 struct btrfs_root *extent_root;
1735 struct btrfs_trans_handle *trans;
1736 struct extent_map *em;
1737 struct map_lookup *map;
1741 root = root->fs_info->chunk_root;
1742 extent_root = root->fs_info->extent_root;
1743 em_tree = &root->fs_info->mapping_tree.map_tree;
1745 ret = btrfs_can_relocate(extent_root, chunk_offset);
1749 /* step one, relocate all the extents inside this chunk */
1750 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1754 trans = btrfs_start_transaction(root, 0);
1760 * step two, delete the device extents and the
1761 * chunk tree entries
1763 read_lock(&em_tree->lock);
1764 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1765 read_unlock(&em_tree->lock);
1767 BUG_ON(em->start > chunk_offset ||
1768 em->start + em->len < chunk_offset);
1769 map = (struct map_lookup *)em->bdev;
1771 for (i = 0; i < map->num_stripes; i++) {
1772 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1773 map->stripes[i].physical);
1776 if (map->stripes[i].dev) {
1777 ret = btrfs_update_device(trans, map->stripes[i].dev);
1781 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1786 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1787 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1791 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1794 write_lock(&em_tree->lock);
1795 remove_extent_mapping(em_tree, em);
1796 write_unlock(&em_tree->lock);
1801 /* once for the tree */
1802 free_extent_map(em);
1804 free_extent_map(em);
1806 unlock_chunks(root);
1807 btrfs_end_transaction(trans, root);
1811 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1813 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1814 struct btrfs_path *path;
1815 struct extent_buffer *leaf;
1816 struct btrfs_chunk *chunk;
1817 struct btrfs_key key;
1818 struct btrfs_key found_key;
1819 u64 chunk_tree = chunk_root->root_key.objectid;
1821 bool retried = false;
1825 path = btrfs_alloc_path();
1830 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1831 key.offset = (u64)-1;
1832 key.type = BTRFS_CHUNK_ITEM_KEY;
1835 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1840 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1847 leaf = path->nodes[0];
1848 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1850 chunk = btrfs_item_ptr(leaf, path->slots[0],
1851 struct btrfs_chunk);
1852 chunk_type = btrfs_chunk_type(leaf, chunk);
1853 btrfs_release_path(chunk_root, path);
1855 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1856 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1865 if (found_key.offset == 0)
1867 key.offset = found_key.offset - 1;
1870 if (failed && !retried) {
1874 } else if (failed && retried) {
1879 btrfs_free_path(path);
1883 static u64 div_factor(u64 num, int factor)
1892 int btrfs_balance(struct btrfs_root *dev_root)
1895 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1896 struct btrfs_device *device;
1899 struct btrfs_path *path;
1900 struct btrfs_key key;
1901 struct btrfs_chunk *chunk;
1902 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1903 struct btrfs_trans_handle *trans;
1904 struct btrfs_key found_key;
1906 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1909 mutex_lock(&dev_root->fs_info->volume_mutex);
1910 dev_root = dev_root->fs_info->dev_root;
1912 /* step one make some room on all the devices */
1913 list_for_each_entry(device, devices, dev_list) {
1914 old_size = device->total_bytes;
1915 size_to_free = div_factor(old_size, 1);
1916 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1917 if (!device->writeable ||
1918 device->total_bytes - device->bytes_used > size_to_free)
1921 ret = btrfs_shrink_device(device, old_size - size_to_free);
1926 trans = btrfs_start_transaction(dev_root, 0);
1929 ret = btrfs_grow_device(trans, device, old_size);
1932 btrfs_end_transaction(trans, dev_root);
1935 /* step two, relocate all the chunks */
1936 path = btrfs_alloc_path();
1939 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1940 key.offset = (u64)-1;
1941 key.type = BTRFS_CHUNK_ITEM_KEY;
1944 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1949 * this shouldn't happen, it means the last relocate
1955 ret = btrfs_previous_item(chunk_root, path, 0,
1956 BTRFS_CHUNK_ITEM_KEY);
1960 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1962 if (found_key.objectid != key.objectid)
1965 chunk = btrfs_item_ptr(path->nodes[0],
1967 struct btrfs_chunk);
1968 /* chunk zero is special */
1969 if (found_key.offset == 0)
1972 btrfs_release_path(chunk_root, path);
1973 ret = btrfs_relocate_chunk(chunk_root,
1974 chunk_root->root_key.objectid,
1977 BUG_ON(ret && ret != -ENOSPC);
1978 key.offset = found_key.offset - 1;
1982 btrfs_free_path(path);
1983 mutex_unlock(&dev_root->fs_info->volume_mutex);
1988 * shrinking a device means finding all of the device extents past
1989 * the new size, and then following the back refs to the chunks.
1990 * The chunk relocation code actually frees the device extent
1992 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1994 struct btrfs_trans_handle *trans;
1995 struct btrfs_root *root = device->dev_root;
1996 struct btrfs_dev_extent *dev_extent = NULL;
1997 struct btrfs_path *path;
2005 bool retried = false;
2006 struct extent_buffer *l;
2007 struct btrfs_key key;
2008 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2009 u64 old_total = btrfs_super_total_bytes(super_copy);
2010 u64 old_size = device->total_bytes;
2011 u64 diff = device->total_bytes - new_size;
2013 if (new_size >= device->total_bytes)
2016 path = btrfs_alloc_path();
2024 device->total_bytes = new_size;
2025 if (device->writeable)
2026 device->fs_devices->total_rw_bytes -= diff;
2027 unlock_chunks(root);
2030 key.objectid = device->devid;
2031 key.offset = (u64)-1;
2032 key.type = BTRFS_DEV_EXTENT_KEY;
2035 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2039 ret = btrfs_previous_item(root, path, 0, key.type);
2044 btrfs_release_path(root, path);
2049 slot = path->slots[0];
2050 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2052 if (key.objectid != device->devid) {
2053 btrfs_release_path(root, path);
2057 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2058 length = btrfs_dev_extent_length(l, dev_extent);
2060 if (key.offset + length <= new_size) {
2061 btrfs_release_path(root, path);
2065 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2066 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2067 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2068 btrfs_release_path(root, path);
2070 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2072 if (ret && ret != -ENOSPC)
2079 if (failed && !retried) {
2083 } else if (failed && retried) {
2087 device->total_bytes = old_size;
2088 if (device->writeable)
2089 device->fs_devices->total_rw_bytes += diff;
2090 unlock_chunks(root);
2094 /* Shrinking succeeded, else we would be at "done". */
2095 trans = btrfs_start_transaction(root, 0);
2098 device->disk_total_bytes = new_size;
2099 /* Now btrfs_update_device() will change the on-disk size. */
2100 ret = btrfs_update_device(trans, device);
2102 unlock_chunks(root);
2103 btrfs_end_transaction(trans, root);
2106 WARN_ON(diff > old_total);
2107 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2108 unlock_chunks(root);
2109 btrfs_end_transaction(trans, root);
2111 btrfs_free_path(path);
2115 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2116 struct btrfs_root *root,
2117 struct btrfs_key *key,
2118 struct btrfs_chunk *chunk, int item_size)
2120 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2121 struct btrfs_disk_key disk_key;
2125 array_size = btrfs_super_sys_array_size(super_copy);
2126 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2129 ptr = super_copy->sys_chunk_array + array_size;
2130 btrfs_cpu_key_to_disk(&disk_key, key);
2131 memcpy(ptr, &disk_key, sizeof(disk_key));
2132 ptr += sizeof(disk_key);
2133 memcpy(ptr, chunk, item_size);
2134 item_size += sizeof(disk_key);
2135 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2139 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2140 int num_stripes, int sub_stripes)
2142 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2144 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2145 return calc_size * (num_stripes / sub_stripes);
2147 return calc_size * num_stripes;
2150 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2151 struct btrfs_root *extent_root,
2152 struct map_lookup **map_ret,
2153 u64 *num_bytes, u64 *stripe_size,
2154 u64 start, u64 type)
2156 struct btrfs_fs_info *info = extent_root->fs_info;
2157 struct btrfs_device *device = NULL;
2158 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2159 struct list_head *cur;
2160 struct map_lookup *map = NULL;
2161 struct extent_map_tree *em_tree;
2162 struct extent_map *em;
2163 struct list_head private_devs;
2164 int min_stripe_size = 1 * 1024 * 1024;
2165 u64 calc_size = 1024 * 1024 * 1024;
2166 u64 max_chunk_size = calc_size;
2171 int num_stripes = 1;
2172 int min_stripes = 1;
2173 int sub_stripes = 0;
2177 int stripe_len = 64 * 1024;
2179 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2180 (type & BTRFS_BLOCK_GROUP_DUP)) {
2182 type &= ~BTRFS_BLOCK_GROUP_DUP;
2184 if (list_empty(&fs_devices->alloc_list))
2187 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2188 num_stripes = fs_devices->rw_devices;
2191 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2195 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2196 if (fs_devices->rw_devices < 2)
2201 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2202 num_stripes = fs_devices->rw_devices;
2203 if (num_stripes < 4)
2205 num_stripes &= ~(u32)1;
2210 if (type & BTRFS_BLOCK_GROUP_DATA) {
2211 max_chunk_size = 10 * calc_size;
2212 min_stripe_size = 64 * 1024 * 1024;
2213 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2214 max_chunk_size = 256 * 1024 * 1024;
2215 min_stripe_size = 32 * 1024 * 1024;
2216 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2217 calc_size = 8 * 1024 * 1024;
2218 max_chunk_size = calc_size * 2;
2219 min_stripe_size = 1 * 1024 * 1024;
2222 /* we don't want a chunk larger than 10% of writeable space */
2223 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2228 if (!map || map->num_stripes != num_stripes) {
2230 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2233 map->num_stripes = num_stripes;
2236 if (calc_size * num_stripes > max_chunk_size) {
2237 calc_size = max_chunk_size;
2238 do_div(calc_size, num_stripes);
2239 do_div(calc_size, stripe_len);
2240 calc_size *= stripe_len;
2243 /* we don't want tiny stripes */
2245 calc_size = max_t(u64, min_stripe_size, calc_size);
2248 * we're about to do_div by the stripe_len so lets make sure
2249 * we end up with something bigger than a stripe
2251 calc_size = max_t(u64, calc_size, stripe_len * 4);
2253 do_div(calc_size, stripe_len);
2254 calc_size *= stripe_len;
2256 cur = fs_devices->alloc_list.next;
2259 if (type & BTRFS_BLOCK_GROUP_DUP)
2260 min_free = calc_size * 2;
2262 min_free = calc_size;
2265 * we add 1MB because we never use the first 1MB of the device, unless
2266 * we've looped, then we are likely allocating the maximum amount of
2267 * space left already
2270 min_free += 1024 * 1024;
2272 INIT_LIST_HEAD(&private_devs);
2273 while (index < num_stripes) {
2274 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2275 BUG_ON(!device->writeable);
2276 if (device->total_bytes > device->bytes_used)
2277 avail = device->total_bytes - device->bytes_used;
2282 if (device->in_fs_metadata && avail >= min_free) {
2283 ret = find_free_dev_extent(trans, device,
2284 min_free, &dev_offset,
2287 list_move_tail(&device->dev_alloc_list,
2289 map->stripes[index].dev = device;
2290 map->stripes[index].physical = dev_offset;
2292 if (type & BTRFS_BLOCK_GROUP_DUP) {
2293 map->stripes[index].dev = device;
2294 map->stripes[index].physical =
2295 dev_offset + calc_size;
2299 } else if (device->in_fs_metadata && avail > max_avail)
2301 if (cur == &fs_devices->alloc_list)
2304 list_splice(&private_devs, &fs_devices->alloc_list);
2305 if (index < num_stripes) {
2306 if (index >= min_stripes) {
2307 num_stripes = index;
2308 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2309 num_stripes /= sub_stripes;
2310 num_stripes *= sub_stripes;
2315 if (!looped && max_avail > 0) {
2317 calc_size = max_avail;
2323 map->sector_size = extent_root->sectorsize;
2324 map->stripe_len = stripe_len;
2325 map->io_align = stripe_len;
2326 map->io_width = stripe_len;
2328 map->num_stripes = num_stripes;
2329 map->sub_stripes = sub_stripes;
2332 *stripe_size = calc_size;
2333 *num_bytes = chunk_bytes_by_type(type, calc_size,
2334 num_stripes, sub_stripes);
2336 em = alloc_extent_map(GFP_NOFS);
2341 em->bdev = (struct block_device *)map;
2343 em->len = *num_bytes;
2344 em->block_start = 0;
2345 em->block_len = em->len;
2347 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2348 write_lock(&em_tree->lock);
2349 ret = add_extent_mapping(em_tree, em);
2350 write_unlock(&em_tree->lock);
2352 free_extent_map(em);
2354 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2355 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2360 while (index < map->num_stripes) {
2361 device = map->stripes[index].dev;
2362 dev_offset = map->stripes[index].physical;
2364 ret = btrfs_alloc_dev_extent(trans, device,
2365 info->chunk_root->root_key.objectid,
2366 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2367 start, dev_offset, calc_size);
2375 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2376 struct btrfs_root *extent_root,
2377 struct map_lookup *map, u64 chunk_offset,
2378 u64 chunk_size, u64 stripe_size)
2381 struct btrfs_key key;
2382 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2383 struct btrfs_device *device;
2384 struct btrfs_chunk *chunk;
2385 struct btrfs_stripe *stripe;
2386 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2390 chunk = kzalloc(item_size, GFP_NOFS);
2395 while (index < map->num_stripes) {
2396 device = map->stripes[index].dev;
2397 device->bytes_used += stripe_size;
2398 ret = btrfs_update_device(trans, device);
2404 stripe = &chunk->stripe;
2405 while (index < map->num_stripes) {
2406 device = map->stripes[index].dev;
2407 dev_offset = map->stripes[index].physical;
2409 btrfs_set_stack_stripe_devid(stripe, device->devid);
2410 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2411 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2416 btrfs_set_stack_chunk_length(chunk, chunk_size);
2417 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2418 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2419 btrfs_set_stack_chunk_type(chunk, map->type);
2420 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2421 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2422 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2423 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2424 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2426 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2427 key.type = BTRFS_CHUNK_ITEM_KEY;
2428 key.offset = chunk_offset;
2430 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2433 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2434 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2443 * Chunk allocation falls into two parts. The first part does works
2444 * that make the new allocated chunk useable, but not do any operation
2445 * that modifies the chunk tree. The second part does the works that
2446 * require modifying the chunk tree. This division is important for the
2447 * bootstrap process of adding storage to a seed btrfs.
2449 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2450 struct btrfs_root *extent_root, u64 type)
2455 struct map_lookup *map;
2456 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2459 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2464 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2465 &stripe_size, chunk_offset, type);
2469 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2470 chunk_size, stripe_size);
2475 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2476 struct btrfs_root *root,
2477 struct btrfs_device *device)
2480 u64 sys_chunk_offset;
2484 u64 sys_stripe_size;
2486 struct map_lookup *map;
2487 struct map_lookup *sys_map;
2488 struct btrfs_fs_info *fs_info = root->fs_info;
2489 struct btrfs_root *extent_root = fs_info->extent_root;
2492 ret = find_next_chunk(fs_info->chunk_root,
2493 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2496 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2497 (fs_info->metadata_alloc_profile &
2498 fs_info->avail_metadata_alloc_bits);
2499 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2501 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2502 &stripe_size, chunk_offset, alloc_profile);
2505 sys_chunk_offset = chunk_offset + chunk_size;
2507 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2508 (fs_info->system_alloc_profile &
2509 fs_info->avail_system_alloc_bits);
2510 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2512 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2513 &sys_chunk_size, &sys_stripe_size,
2514 sys_chunk_offset, alloc_profile);
2517 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2521 * Modifying chunk tree needs allocating new blocks from both
2522 * system block group and metadata block group. So we only can
2523 * do operations require modifying the chunk tree after both
2524 * block groups were created.
2526 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2527 chunk_size, stripe_size);
2530 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2531 sys_chunk_offset, sys_chunk_size,
2537 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2539 struct extent_map *em;
2540 struct map_lookup *map;
2541 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2545 read_lock(&map_tree->map_tree.lock);
2546 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2547 read_unlock(&map_tree->map_tree.lock);
2551 if (btrfs_test_opt(root, DEGRADED)) {
2552 free_extent_map(em);
2556 map = (struct map_lookup *)em->bdev;
2557 for (i = 0; i < map->num_stripes; i++) {
2558 if (!map->stripes[i].dev->writeable) {
2563 free_extent_map(em);
2567 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2569 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2572 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2574 struct extent_map *em;
2577 write_lock(&tree->map_tree.lock);
2578 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2580 remove_extent_mapping(&tree->map_tree, em);
2581 write_unlock(&tree->map_tree.lock);
2586 free_extent_map(em);
2587 /* once for the tree */
2588 free_extent_map(em);
2592 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2594 struct extent_map *em;
2595 struct map_lookup *map;
2596 struct extent_map_tree *em_tree = &map_tree->map_tree;
2599 read_lock(&em_tree->lock);
2600 em = lookup_extent_mapping(em_tree, logical, len);
2601 read_unlock(&em_tree->lock);
2604 BUG_ON(em->start > logical || em->start + em->len < logical);
2605 map = (struct map_lookup *)em->bdev;
2606 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2607 ret = map->num_stripes;
2608 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2609 ret = map->sub_stripes;
2612 free_extent_map(em);
2616 static int find_live_mirror(struct map_lookup *map, int first, int num,
2620 if (map->stripes[optimal].dev->bdev)
2622 for (i = first; i < first + num; i++) {
2623 if (map->stripes[i].dev->bdev)
2626 /* we couldn't find one that doesn't fail. Just return something
2627 * and the io error handling code will clean up eventually
2632 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2633 u64 logical, u64 *length,
2634 struct btrfs_multi_bio **multi_ret,
2635 int mirror_num, struct page *unplug_page)
2637 struct extent_map *em;
2638 struct map_lookup *map;
2639 struct extent_map_tree *em_tree = &map_tree->map_tree;
2643 int stripes_allocated = 8;
2644 int stripes_required = 1;
2649 struct btrfs_multi_bio *multi = NULL;
2651 if (multi_ret && !(rw & REQ_WRITE))
2652 stripes_allocated = 1;
2655 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2660 atomic_set(&multi->error, 0);
2663 read_lock(&em_tree->lock);
2664 em = lookup_extent_mapping(em_tree, logical, *length);
2665 read_unlock(&em_tree->lock);
2667 if (!em && unplug_page) {
2673 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2674 (unsigned long long)logical,
2675 (unsigned long long)*length);
2679 BUG_ON(em->start > logical || em->start + em->len < logical);
2680 map = (struct map_lookup *)em->bdev;
2681 offset = logical - em->start;
2683 if (mirror_num > map->num_stripes)
2686 /* if our multi bio struct is too small, back off and try again */
2687 if (rw & REQ_WRITE) {
2688 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2689 BTRFS_BLOCK_GROUP_DUP)) {
2690 stripes_required = map->num_stripes;
2692 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2693 stripes_required = map->sub_stripes;
2697 if (multi_ret && (rw & REQ_WRITE) &&
2698 stripes_allocated < stripes_required) {
2699 stripes_allocated = map->num_stripes;
2700 free_extent_map(em);
2706 * stripe_nr counts the total number of stripes we have to stride
2707 * to get to this block
2709 do_div(stripe_nr, map->stripe_len);
2711 stripe_offset = stripe_nr * map->stripe_len;
2712 BUG_ON(offset < stripe_offset);
2714 /* stripe_offset is the offset of this block in its stripe*/
2715 stripe_offset = offset - stripe_offset;
2717 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2718 BTRFS_BLOCK_GROUP_RAID10 |
2719 BTRFS_BLOCK_GROUP_DUP)) {
2720 /* we limit the length of each bio to what fits in a stripe */
2721 *length = min_t(u64, em->len - offset,
2722 map->stripe_len - stripe_offset);
2724 *length = em->len - offset;
2727 if (!multi_ret && !unplug_page)
2732 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2733 if (unplug_page || (rw & REQ_WRITE))
2734 num_stripes = map->num_stripes;
2735 else if (mirror_num)
2736 stripe_index = mirror_num - 1;
2738 stripe_index = find_live_mirror(map, 0,
2740 current->pid % map->num_stripes);
2743 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2745 num_stripes = map->num_stripes;
2746 else if (mirror_num)
2747 stripe_index = mirror_num - 1;
2749 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2750 int factor = map->num_stripes / map->sub_stripes;
2752 stripe_index = do_div(stripe_nr, factor);
2753 stripe_index *= map->sub_stripes;
2755 if (unplug_page || (rw & REQ_WRITE))
2756 num_stripes = map->sub_stripes;
2757 else if (mirror_num)
2758 stripe_index += mirror_num - 1;
2760 stripe_index = find_live_mirror(map, stripe_index,
2761 map->sub_stripes, stripe_index +
2762 current->pid % map->sub_stripes);
2766 * after this do_div call, stripe_nr is the number of stripes
2767 * on this device we have to walk to find the data, and
2768 * stripe_index is the number of our device in the stripe array
2770 stripe_index = do_div(stripe_nr, map->num_stripes);
2772 BUG_ON(stripe_index >= map->num_stripes);
2774 for (i = 0; i < num_stripes; i++) {
2776 struct btrfs_device *device;
2777 struct backing_dev_info *bdi;
2779 device = map->stripes[stripe_index].dev;
2781 bdi = blk_get_backing_dev_info(device->bdev);
2782 if (bdi->unplug_io_fn)
2783 bdi->unplug_io_fn(bdi, unplug_page);
2786 multi->stripes[i].physical =
2787 map->stripes[stripe_index].physical +
2788 stripe_offset + stripe_nr * map->stripe_len;
2789 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2795 multi->num_stripes = num_stripes;
2796 multi->max_errors = max_errors;
2799 free_extent_map(em);
2803 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2804 u64 logical, u64 *length,
2805 struct btrfs_multi_bio **multi_ret, int mirror_num)
2807 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2811 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2812 u64 chunk_start, u64 physical, u64 devid,
2813 u64 **logical, int *naddrs, int *stripe_len)
2815 struct extent_map_tree *em_tree = &map_tree->map_tree;
2816 struct extent_map *em;
2817 struct map_lookup *map;
2824 read_lock(&em_tree->lock);
2825 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2826 read_unlock(&em_tree->lock);
2828 BUG_ON(!em || em->start != chunk_start);
2829 map = (struct map_lookup *)em->bdev;
2832 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2833 do_div(length, map->num_stripes / map->sub_stripes);
2834 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2835 do_div(length, map->num_stripes);
2837 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2840 for (i = 0; i < map->num_stripes; i++) {
2841 if (devid && map->stripes[i].dev->devid != devid)
2843 if (map->stripes[i].physical > physical ||
2844 map->stripes[i].physical + length <= physical)
2847 stripe_nr = physical - map->stripes[i].physical;
2848 do_div(stripe_nr, map->stripe_len);
2850 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2851 stripe_nr = stripe_nr * map->num_stripes + i;
2852 do_div(stripe_nr, map->sub_stripes);
2853 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2854 stripe_nr = stripe_nr * map->num_stripes + i;
2856 bytenr = chunk_start + stripe_nr * map->stripe_len;
2857 WARN_ON(nr >= map->num_stripes);
2858 for (j = 0; j < nr; j++) {
2859 if (buf[j] == bytenr)
2863 WARN_ON(nr >= map->num_stripes);
2870 *stripe_len = map->stripe_len;
2872 free_extent_map(em);
2876 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2877 u64 logical, struct page *page)
2879 u64 length = PAGE_CACHE_SIZE;
2880 return __btrfs_map_block(map_tree, READ, logical, &length,
2884 static void end_bio_multi_stripe(struct bio *bio, int err)
2886 struct btrfs_multi_bio *multi = bio->bi_private;
2887 int is_orig_bio = 0;
2890 atomic_inc(&multi->error);
2892 if (bio == multi->orig_bio)
2895 if (atomic_dec_and_test(&multi->stripes_pending)) {
2898 bio = multi->orig_bio;
2900 bio->bi_private = multi->private;
2901 bio->bi_end_io = multi->end_io;
2902 /* only send an error to the higher layers if it is
2903 * beyond the tolerance of the multi-bio
2905 if (atomic_read(&multi->error) > multi->max_errors) {
2909 * this bio is actually up to date, we didn't
2910 * go over the max number of errors
2912 set_bit(BIO_UPTODATE, &bio->bi_flags);
2917 bio_endio(bio, err);
2918 } else if (!is_orig_bio) {
2923 struct async_sched {
2926 struct btrfs_fs_info *info;
2927 struct btrfs_work work;
2931 * see run_scheduled_bios for a description of why bios are collected for
2934 * This will add one bio to the pending list for a device and make sure
2935 * the work struct is scheduled.
2937 static noinline int schedule_bio(struct btrfs_root *root,
2938 struct btrfs_device *device,
2939 int rw, struct bio *bio)
2941 int should_queue = 1;
2942 struct btrfs_pending_bios *pending_bios;
2944 /* don't bother with additional async steps for reads, right now */
2945 if (!(rw & REQ_WRITE)) {
2947 submit_bio(rw, bio);
2953 * nr_async_bios allows us to reliably return congestion to the
2954 * higher layers. Otherwise, the async bio makes it appear we have
2955 * made progress against dirty pages when we've really just put it
2956 * on a queue for later
2958 atomic_inc(&root->fs_info->nr_async_bios);
2959 WARN_ON(bio->bi_next);
2960 bio->bi_next = NULL;
2963 spin_lock(&device->io_lock);
2964 if (bio->bi_rw & REQ_SYNC)
2965 pending_bios = &device->pending_sync_bios;
2967 pending_bios = &device->pending_bios;
2969 if (pending_bios->tail)
2970 pending_bios->tail->bi_next = bio;
2972 pending_bios->tail = bio;
2973 if (!pending_bios->head)
2974 pending_bios->head = bio;
2975 if (device->running_pending)
2978 spin_unlock(&device->io_lock);
2981 btrfs_queue_worker(&root->fs_info->submit_workers,
2986 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2987 int mirror_num, int async_submit)
2989 struct btrfs_mapping_tree *map_tree;
2990 struct btrfs_device *dev;
2991 struct bio *first_bio = bio;
2992 u64 logical = (u64)bio->bi_sector << 9;
2995 struct btrfs_multi_bio *multi = NULL;
3000 length = bio->bi_size;
3001 map_tree = &root->fs_info->mapping_tree;
3002 map_length = length;
3004 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3008 total_devs = multi->num_stripes;
3009 if (map_length < length) {
3010 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3011 "len %llu\n", (unsigned long long)logical,
3012 (unsigned long long)length,
3013 (unsigned long long)map_length);
3016 multi->end_io = first_bio->bi_end_io;
3017 multi->private = first_bio->bi_private;
3018 multi->orig_bio = first_bio;
3019 atomic_set(&multi->stripes_pending, multi->num_stripes);
3021 while (dev_nr < total_devs) {
3022 if (total_devs > 1) {
3023 if (dev_nr < total_devs - 1) {
3024 bio = bio_clone(first_bio, GFP_NOFS);
3029 bio->bi_private = multi;
3030 bio->bi_end_io = end_bio_multi_stripe;
3032 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3033 dev = multi->stripes[dev_nr].dev;
3034 BUG_ON(rw == WRITE && !dev->writeable);
3035 if (dev && dev->bdev) {
3036 bio->bi_bdev = dev->bdev;
3038 schedule_bio(root, dev, rw, bio);
3040 submit_bio(rw, bio);
3042 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3043 bio->bi_sector = logical >> 9;
3044 bio_endio(bio, -EIO);
3048 if (total_devs == 1)
3053 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3056 struct btrfs_device *device;
3057 struct btrfs_fs_devices *cur_devices;
3059 cur_devices = root->fs_info->fs_devices;
3060 while (cur_devices) {
3062 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3063 device = __find_device(&cur_devices->devices,
3068 cur_devices = cur_devices->seed;
3073 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3074 u64 devid, u8 *dev_uuid)
3076 struct btrfs_device *device;
3077 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3079 device = kzalloc(sizeof(*device), GFP_NOFS);
3082 list_add(&device->dev_list,
3083 &fs_devices->devices);
3084 device->dev_root = root->fs_info->dev_root;
3085 device->devid = devid;
3086 device->work.func = pending_bios_fn;
3087 device->fs_devices = fs_devices;
3088 fs_devices->num_devices++;
3089 spin_lock_init(&device->io_lock);
3090 INIT_LIST_HEAD(&device->dev_alloc_list);
3091 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3095 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3096 struct extent_buffer *leaf,
3097 struct btrfs_chunk *chunk)
3099 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3100 struct map_lookup *map;
3101 struct extent_map *em;
3105 u8 uuid[BTRFS_UUID_SIZE];
3110 logical = key->offset;
3111 length = btrfs_chunk_length(leaf, chunk);
3113 read_lock(&map_tree->map_tree.lock);
3114 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3115 read_unlock(&map_tree->map_tree.lock);
3117 /* already mapped? */
3118 if (em && em->start <= logical && em->start + em->len > logical) {
3119 free_extent_map(em);
3122 free_extent_map(em);
3125 em = alloc_extent_map(GFP_NOFS);
3128 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3129 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3131 free_extent_map(em);
3135 em->bdev = (struct block_device *)map;
3136 em->start = logical;
3138 em->block_start = 0;
3139 em->block_len = em->len;
3141 map->num_stripes = num_stripes;
3142 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3143 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3144 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3145 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3146 map->type = btrfs_chunk_type(leaf, chunk);
3147 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3148 for (i = 0; i < num_stripes; i++) {
3149 map->stripes[i].physical =
3150 btrfs_stripe_offset_nr(leaf, chunk, i);
3151 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3152 read_extent_buffer(leaf, uuid, (unsigned long)
3153 btrfs_stripe_dev_uuid_nr(chunk, i),
3155 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3157 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3159 free_extent_map(em);
3162 if (!map->stripes[i].dev) {
3163 map->stripes[i].dev =
3164 add_missing_dev(root, devid, uuid);
3165 if (!map->stripes[i].dev) {
3167 free_extent_map(em);
3171 map->stripes[i].dev->in_fs_metadata = 1;
3174 write_lock(&map_tree->map_tree.lock);
3175 ret = add_extent_mapping(&map_tree->map_tree, em);
3176 write_unlock(&map_tree->map_tree.lock);
3178 free_extent_map(em);
3183 static int fill_device_from_item(struct extent_buffer *leaf,
3184 struct btrfs_dev_item *dev_item,
3185 struct btrfs_device *device)
3189 device->devid = btrfs_device_id(leaf, dev_item);
3190 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3191 device->total_bytes = device->disk_total_bytes;
3192 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3193 device->type = btrfs_device_type(leaf, dev_item);
3194 device->io_align = btrfs_device_io_align(leaf, dev_item);
3195 device->io_width = btrfs_device_io_width(leaf, dev_item);
3196 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3198 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3199 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3204 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3206 struct btrfs_fs_devices *fs_devices;
3209 mutex_lock(&uuid_mutex);
3211 fs_devices = root->fs_info->fs_devices->seed;
3212 while (fs_devices) {
3213 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3217 fs_devices = fs_devices->seed;
3220 fs_devices = find_fsid(fsid);
3226 fs_devices = clone_fs_devices(fs_devices);
3227 if (IS_ERR(fs_devices)) {
3228 ret = PTR_ERR(fs_devices);
3232 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3233 root->fs_info->bdev_holder);
3237 if (!fs_devices->seeding) {
3238 __btrfs_close_devices(fs_devices);
3239 free_fs_devices(fs_devices);
3244 fs_devices->seed = root->fs_info->fs_devices->seed;
3245 root->fs_info->fs_devices->seed = fs_devices;
3247 mutex_unlock(&uuid_mutex);
3251 static int read_one_dev(struct btrfs_root *root,
3252 struct extent_buffer *leaf,
3253 struct btrfs_dev_item *dev_item)
3255 struct btrfs_device *device;
3258 u8 fs_uuid[BTRFS_UUID_SIZE];
3259 u8 dev_uuid[BTRFS_UUID_SIZE];
3261 devid = btrfs_device_id(leaf, dev_item);
3262 read_extent_buffer(leaf, dev_uuid,
3263 (unsigned long)btrfs_device_uuid(dev_item),
3265 read_extent_buffer(leaf, fs_uuid,
3266 (unsigned long)btrfs_device_fsid(dev_item),
3269 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3270 ret = open_seed_devices(root, fs_uuid);
3271 if (ret && !btrfs_test_opt(root, DEGRADED))
3275 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3276 if (!device || !device->bdev) {
3277 if (!btrfs_test_opt(root, DEGRADED))
3281 printk(KERN_WARNING "warning devid %llu missing\n",
3282 (unsigned long long)devid);
3283 device = add_missing_dev(root, devid, dev_uuid);
3289 if (device->fs_devices != root->fs_info->fs_devices) {
3290 BUG_ON(device->writeable);
3291 if (device->generation !=
3292 btrfs_device_generation(leaf, dev_item))
3296 fill_device_from_item(leaf, dev_item, device);
3297 device->dev_root = root->fs_info->dev_root;
3298 device->in_fs_metadata = 1;
3299 if (device->writeable)
3300 device->fs_devices->total_rw_bytes += device->total_bytes;
3305 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3307 struct btrfs_dev_item *dev_item;
3309 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3311 return read_one_dev(root, buf, dev_item);
3314 int btrfs_read_sys_array(struct btrfs_root *root)
3316 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3317 struct extent_buffer *sb;
3318 struct btrfs_disk_key *disk_key;
3319 struct btrfs_chunk *chunk;
3321 unsigned long sb_ptr;
3327 struct btrfs_key key;
3329 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3330 BTRFS_SUPER_INFO_SIZE);
3333 btrfs_set_buffer_uptodate(sb);
3334 btrfs_set_buffer_lockdep_class(sb, 0);
3336 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3337 array_size = btrfs_super_sys_array_size(super_copy);
3339 ptr = super_copy->sys_chunk_array;
3340 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3343 while (cur < array_size) {
3344 disk_key = (struct btrfs_disk_key *)ptr;
3345 btrfs_disk_key_to_cpu(&key, disk_key);
3347 len = sizeof(*disk_key); ptr += len;
3351 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3352 chunk = (struct btrfs_chunk *)sb_ptr;
3353 ret = read_one_chunk(root, &key, sb, chunk);
3356 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3357 len = btrfs_chunk_item_size(num_stripes);
3366 free_extent_buffer(sb);
3370 int btrfs_read_chunk_tree(struct btrfs_root *root)
3372 struct btrfs_path *path;
3373 struct extent_buffer *leaf;
3374 struct btrfs_key key;
3375 struct btrfs_key found_key;
3379 root = root->fs_info->chunk_root;
3381 path = btrfs_alloc_path();
3385 /* first we search for all of the device items, and then we
3386 * read in all of the chunk items. This way we can create chunk
3387 * mappings that reference all of the devices that are afound
3389 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3393 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3397 leaf = path->nodes[0];
3398 slot = path->slots[0];
3399 if (slot >= btrfs_header_nritems(leaf)) {
3400 ret = btrfs_next_leaf(root, path);
3407 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3408 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3409 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3411 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3412 struct btrfs_dev_item *dev_item;
3413 dev_item = btrfs_item_ptr(leaf, slot,
3414 struct btrfs_dev_item);
3415 ret = read_one_dev(root, leaf, dev_item);
3419 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3420 struct btrfs_chunk *chunk;
3421 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3422 ret = read_one_chunk(root, &found_key, leaf, chunk);
3428 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3430 btrfs_release_path(root, path);
3435 btrfs_free_path(path);