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 device->barriers = 1;
402 spin_lock_init(&device->io_lock);
403 device->name = kstrdup(path, GFP_NOFS);
408 INIT_LIST_HEAD(&device->dev_alloc_list);
410 mutex_lock(&fs_devices->device_list_mutex);
411 list_add(&device->dev_list, &fs_devices->devices);
412 mutex_unlock(&fs_devices->device_list_mutex);
414 device->fs_devices = fs_devices;
415 fs_devices->num_devices++;
416 } else if (!device->name || strcmp(device->name, path)) {
417 name = kstrdup(path, GFP_NOFS);
422 if (device->missing) {
423 fs_devices->missing_devices--;
428 if (found_transid > fs_devices->latest_trans) {
429 fs_devices->latest_devid = devid;
430 fs_devices->latest_trans = found_transid;
432 *fs_devices_ret = fs_devices;
436 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
438 struct btrfs_fs_devices *fs_devices;
439 struct btrfs_device *device;
440 struct btrfs_device *orig_dev;
442 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
444 return ERR_PTR(-ENOMEM);
446 INIT_LIST_HEAD(&fs_devices->devices);
447 INIT_LIST_HEAD(&fs_devices->alloc_list);
448 INIT_LIST_HEAD(&fs_devices->list);
449 mutex_init(&fs_devices->device_list_mutex);
450 fs_devices->latest_devid = orig->latest_devid;
451 fs_devices->latest_trans = orig->latest_trans;
452 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
454 mutex_lock(&orig->device_list_mutex);
455 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
456 device = kzalloc(sizeof(*device), GFP_NOFS);
460 device->name = kstrdup(orig_dev->name, GFP_NOFS);
466 device->devid = orig_dev->devid;
467 device->work.func = pending_bios_fn;
468 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
469 device->barriers = 1;
470 spin_lock_init(&device->io_lock);
471 INIT_LIST_HEAD(&device->dev_list);
472 INIT_LIST_HEAD(&device->dev_alloc_list);
474 list_add(&device->dev_list, &fs_devices->devices);
475 device->fs_devices = fs_devices;
476 fs_devices->num_devices++;
478 mutex_unlock(&orig->device_list_mutex);
481 mutex_unlock(&orig->device_list_mutex);
482 free_fs_devices(fs_devices);
483 return ERR_PTR(-ENOMEM);
486 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
488 struct btrfs_device *device, *next;
490 mutex_lock(&uuid_mutex);
492 mutex_lock(&fs_devices->device_list_mutex);
493 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
494 if (device->in_fs_metadata)
498 close_bdev_exclusive(device->bdev, device->mode);
500 fs_devices->open_devices--;
502 if (device->writeable) {
503 list_del_init(&device->dev_alloc_list);
504 device->writeable = 0;
505 fs_devices->rw_devices--;
507 list_del_init(&device->dev_list);
508 fs_devices->num_devices--;
512 mutex_unlock(&fs_devices->device_list_mutex);
514 if (fs_devices->seed) {
515 fs_devices = fs_devices->seed;
519 mutex_unlock(&uuid_mutex);
523 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
525 struct btrfs_device *device;
527 if (--fs_devices->opened > 0)
530 list_for_each_entry(device, &fs_devices->devices, dev_list) {
532 close_bdev_exclusive(device->bdev, device->mode);
533 fs_devices->open_devices--;
535 if (device->writeable) {
536 list_del_init(&device->dev_alloc_list);
537 fs_devices->rw_devices--;
541 device->writeable = 0;
542 device->in_fs_metadata = 0;
544 WARN_ON(fs_devices->open_devices);
545 WARN_ON(fs_devices->rw_devices);
546 fs_devices->opened = 0;
547 fs_devices->seeding = 0;
552 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
554 struct btrfs_fs_devices *seed_devices = NULL;
557 mutex_lock(&uuid_mutex);
558 ret = __btrfs_close_devices(fs_devices);
559 if (!fs_devices->opened) {
560 seed_devices = fs_devices->seed;
561 fs_devices->seed = NULL;
563 mutex_unlock(&uuid_mutex);
565 while (seed_devices) {
566 fs_devices = seed_devices;
567 seed_devices = fs_devices->seed;
568 __btrfs_close_devices(fs_devices);
569 free_fs_devices(fs_devices);
574 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
575 fmode_t flags, void *holder)
577 struct block_device *bdev;
578 struct list_head *head = &fs_devices->devices;
579 struct btrfs_device *device;
580 struct block_device *latest_bdev = NULL;
581 struct buffer_head *bh;
582 struct btrfs_super_block *disk_super;
583 u64 latest_devid = 0;
584 u64 latest_transid = 0;
589 list_for_each_entry(device, head, dev_list) {
595 bdev = open_bdev_exclusive(device->name, flags, holder);
597 printk(KERN_INFO "open %s failed\n", device->name);
600 set_blocksize(bdev, 4096);
602 bh = btrfs_read_dev_super(bdev);
606 disk_super = (struct btrfs_super_block *)bh->b_data;
607 devid = btrfs_stack_device_id(&disk_super->dev_item);
608 if (devid != device->devid)
611 if (memcmp(device->uuid, disk_super->dev_item.uuid,
615 device->generation = btrfs_super_generation(disk_super);
616 if (!latest_transid || device->generation > latest_transid) {
617 latest_devid = devid;
618 latest_transid = device->generation;
622 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
623 device->writeable = 0;
625 device->writeable = !bdev_read_only(bdev);
630 device->in_fs_metadata = 0;
631 device->mode = flags;
633 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
634 fs_devices->rotating = 1;
636 fs_devices->open_devices++;
637 if (device->writeable) {
638 fs_devices->rw_devices++;
639 list_add(&device->dev_alloc_list,
640 &fs_devices->alloc_list);
647 close_bdev_exclusive(bdev, FMODE_READ);
651 if (fs_devices->open_devices == 0) {
655 fs_devices->seeding = seeding;
656 fs_devices->opened = 1;
657 fs_devices->latest_bdev = latest_bdev;
658 fs_devices->latest_devid = latest_devid;
659 fs_devices->latest_trans = latest_transid;
660 fs_devices->total_rw_bytes = 0;
665 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
666 fmode_t flags, void *holder)
670 mutex_lock(&uuid_mutex);
671 if (fs_devices->opened) {
672 fs_devices->opened++;
675 ret = __btrfs_open_devices(fs_devices, flags, holder);
677 mutex_unlock(&uuid_mutex);
681 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
682 struct btrfs_fs_devices **fs_devices_ret)
684 struct btrfs_super_block *disk_super;
685 struct block_device *bdev;
686 struct buffer_head *bh;
691 mutex_lock(&uuid_mutex);
693 bdev = open_bdev_exclusive(path, flags, holder);
700 ret = set_blocksize(bdev, 4096);
703 bh = btrfs_read_dev_super(bdev);
708 disk_super = (struct btrfs_super_block *)bh->b_data;
709 devid = btrfs_stack_device_id(&disk_super->dev_item);
710 transid = btrfs_super_generation(disk_super);
711 if (disk_super->label[0])
712 printk(KERN_INFO "device label %s ", disk_super->label);
714 /* FIXME, make a readl uuid parser */
715 printk(KERN_INFO "device fsid %llx-%llx ",
716 *(unsigned long long *)disk_super->fsid,
717 *(unsigned long long *)(disk_super->fsid + 8));
719 printk(KERN_CONT "devid %llu transid %llu %s\n",
720 (unsigned long long)devid, (unsigned long long)transid, path);
721 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
725 close_bdev_exclusive(bdev, flags);
727 mutex_unlock(&uuid_mutex);
732 * this uses a pretty simple search, the expectation is that it is
733 * called very infrequently and that a given device has a small number
736 int find_free_dev_extent(struct btrfs_trans_handle *trans,
737 struct btrfs_device *device, u64 num_bytes,
738 u64 *start, u64 *max_avail)
740 struct btrfs_key key;
741 struct btrfs_root *root = device->dev_root;
742 struct btrfs_dev_extent *dev_extent = NULL;
743 struct btrfs_path *path;
746 u64 search_start = 0;
747 u64 search_end = device->total_bytes;
751 struct extent_buffer *l;
753 path = btrfs_alloc_path();
759 /* FIXME use last free of some kind */
761 /* we don't want to overwrite the superblock on the drive,
762 * so we make sure to start at an offset of at least 1MB
764 search_start = max((u64)1024 * 1024, search_start);
766 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
767 search_start = max(root->fs_info->alloc_start, search_start);
769 key.objectid = device->devid;
770 key.offset = search_start;
771 key.type = BTRFS_DEV_EXTENT_KEY;
772 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
776 ret = btrfs_previous_item(root, path, key.objectid, key.type);
783 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
786 slot = path->slots[0];
787 if (slot >= btrfs_header_nritems(l)) {
788 ret = btrfs_next_leaf(root, path);
795 if (search_start >= search_end) {
799 *start = search_start;
803 *start = last_byte > search_start ?
804 last_byte : search_start;
805 if (search_end <= *start) {
811 btrfs_item_key_to_cpu(l, &key, slot);
813 if (key.objectid < device->devid)
816 if (key.objectid > device->devid)
819 if (key.offset >= search_start && key.offset > last_byte &&
821 if (last_byte < search_start)
822 last_byte = search_start;
823 hole_size = key.offset - last_byte;
825 if (hole_size > *max_avail)
826 *max_avail = hole_size;
828 if (key.offset > last_byte &&
829 hole_size >= num_bytes) {
834 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
838 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
839 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
845 /* we have to make sure we didn't find an extent that has already
846 * been allocated by the map tree or the original allocation
848 BUG_ON(*start < search_start);
850 if (*start + num_bytes > search_end) {
854 /* check for pending inserts here */
858 btrfs_free_path(path);
862 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
863 struct btrfs_device *device,
867 struct btrfs_path *path;
868 struct btrfs_root *root = device->dev_root;
869 struct btrfs_key key;
870 struct btrfs_key found_key;
871 struct extent_buffer *leaf = NULL;
872 struct btrfs_dev_extent *extent = NULL;
874 path = btrfs_alloc_path();
878 key.objectid = device->devid;
880 key.type = BTRFS_DEV_EXTENT_KEY;
882 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
884 ret = btrfs_previous_item(root, path, key.objectid,
885 BTRFS_DEV_EXTENT_KEY);
887 leaf = path->nodes[0];
888 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
889 extent = btrfs_item_ptr(leaf, path->slots[0],
890 struct btrfs_dev_extent);
891 BUG_ON(found_key.offset > start || found_key.offset +
892 btrfs_dev_extent_length(leaf, extent) < start);
894 } else if (ret == 0) {
895 leaf = path->nodes[0];
896 extent = btrfs_item_ptr(leaf, path->slots[0],
897 struct btrfs_dev_extent);
901 if (device->bytes_used > 0)
902 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
903 ret = btrfs_del_item(trans, root, path);
906 btrfs_free_path(path);
910 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
911 struct btrfs_device *device,
912 u64 chunk_tree, u64 chunk_objectid,
913 u64 chunk_offset, u64 start, u64 num_bytes)
916 struct btrfs_path *path;
917 struct btrfs_root *root = device->dev_root;
918 struct btrfs_dev_extent *extent;
919 struct extent_buffer *leaf;
920 struct btrfs_key key;
922 WARN_ON(!device->in_fs_metadata);
923 path = btrfs_alloc_path();
927 key.objectid = device->devid;
929 key.type = BTRFS_DEV_EXTENT_KEY;
930 ret = btrfs_insert_empty_item(trans, root, path, &key,
934 leaf = path->nodes[0];
935 extent = btrfs_item_ptr(leaf, path->slots[0],
936 struct btrfs_dev_extent);
937 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
938 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
939 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
941 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
942 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
945 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
946 btrfs_mark_buffer_dirty(leaf);
947 btrfs_free_path(path);
951 static noinline int find_next_chunk(struct btrfs_root *root,
952 u64 objectid, u64 *offset)
954 struct btrfs_path *path;
956 struct btrfs_key key;
957 struct btrfs_chunk *chunk;
958 struct btrfs_key found_key;
960 path = btrfs_alloc_path();
963 key.objectid = objectid;
964 key.offset = (u64)-1;
965 key.type = BTRFS_CHUNK_ITEM_KEY;
967 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
973 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
977 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
979 if (found_key.objectid != objectid)
982 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
984 *offset = found_key.offset +
985 btrfs_chunk_length(path->nodes[0], chunk);
990 btrfs_free_path(path);
994 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
997 struct btrfs_key key;
998 struct btrfs_key found_key;
999 struct btrfs_path *path;
1001 root = root->fs_info->chunk_root;
1003 path = btrfs_alloc_path();
1007 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1008 key.type = BTRFS_DEV_ITEM_KEY;
1009 key.offset = (u64)-1;
1011 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1017 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1018 BTRFS_DEV_ITEM_KEY);
1022 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1024 *objectid = found_key.offset + 1;
1028 btrfs_free_path(path);
1033 * the device information is stored in the chunk root
1034 * the btrfs_device struct should be fully filled in
1036 int btrfs_add_device(struct btrfs_trans_handle *trans,
1037 struct btrfs_root *root,
1038 struct btrfs_device *device)
1041 struct btrfs_path *path;
1042 struct btrfs_dev_item *dev_item;
1043 struct extent_buffer *leaf;
1044 struct btrfs_key key;
1047 root = root->fs_info->chunk_root;
1049 path = btrfs_alloc_path();
1053 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1054 key.type = BTRFS_DEV_ITEM_KEY;
1055 key.offset = device->devid;
1057 ret = btrfs_insert_empty_item(trans, root, path, &key,
1062 leaf = path->nodes[0];
1063 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1065 btrfs_set_device_id(leaf, dev_item, device->devid);
1066 btrfs_set_device_generation(leaf, dev_item, 0);
1067 btrfs_set_device_type(leaf, dev_item, device->type);
1068 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1069 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1070 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1071 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1072 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1073 btrfs_set_device_group(leaf, dev_item, 0);
1074 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1075 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1076 btrfs_set_device_start_offset(leaf, dev_item, 0);
1078 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1079 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1080 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1081 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1082 btrfs_mark_buffer_dirty(leaf);
1086 btrfs_free_path(path);
1090 static int btrfs_rm_dev_item(struct btrfs_root *root,
1091 struct btrfs_device *device)
1094 struct btrfs_path *path;
1095 struct btrfs_key key;
1096 struct btrfs_trans_handle *trans;
1098 root = root->fs_info->chunk_root;
1100 path = btrfs_alloc_path();
1104 trans = btrfs_start_transaction(root, 0);
1105 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1106 key.type = BTRFS_DEV_ITEM_KEY;
1107 key.offset = device->devid;
1110 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1119 ret = btrfs_del_item(trans, root, path);
1123 btrfs_free_path(path);
1124 unlock_chunks(root);
1125 btrfs_commit_transaction(trans, root);
1129 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1131 struct btrfs_device *device;
1132 struct btrfs_device *next_device;
1133 struct block_device *bdev;
1134 struct buffer_head *bh = NULL;
1135 struct btrfs_super_block *disk_super;
1142 mutex_lock(&uuid_mutex);
1143 mutex_lock(&root->fs_info->volume_mutex);
1145 all_avail = root->fs_info->avail_data_alloc_bits |
1146 root->fs_info->avail_system_alloc_bits |
1147 root->fs_info->avail_metadata_alloc_bits;
1149 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1150 root->fs_info->fs_devices->num_devices <= 4) {
1151 printk(KERN_ERR "btrfs: unable to go below four devices "
1157 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1158 root->fs_info->fs_devices->num_devices <= 2) {
1159 printk(KERN_ERR "btrfs: unable to go below two "
1160 "devices on raid1\n");
1165 if (strcmp(device_path, "missing") == 0) {
1166 struct list_head *devices;
1167 struct btrfs_device *tmp;
1170 devices = &root->fs_info->fs_devices->devices;
1171 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1172 list_for_each_entry(tmp, devices, dev_list) {
1173 if (tmp->in_fs_metadata && !tmp->bdev) {
1178 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1183 printk(KERN_ERR "btrfs: no missing devices found to "
1188 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1189 root->fs_info->bdev_holder);
1191 ret = PTR_ERR(bdev);
1195 set_blocksize(bdev, 4096);
1196 bh = btrfs_read_dev_super(bdev);
1201 disk_super = (struct btrfs_super_block *)bh->b_data;
1202 devid = btrfs_stack_device_id(&disk_super->dev_item);
1203 dev_uuid = disk_super->dev_item.uuid;
1204 device = btrfs_find_device(root, devid, dev_uuid,
1212 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1213 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1219 if (device->writeable) {
1220 list_del_init(&device->dev_alloc_list);
1221 root->fs_info->fs_devices->rw_devices--;
1224 ret = btrfs_shrink_device(device, 0);
1228 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1232 device->in_fs_metadata = 0;
1235 * the device list mutex makes sure that we don't change
1236 * the device list while someone else is writing out all
1237 * the device supers.
1239 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1240 list_del_init(&device->dev_list);
1241 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1243 device->fs_devices->num_devices--;
1245 if (device->missing)
1246 root->fs_info->fs_devices->missing_devices--;
1248 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1249 struct btrfs_device, dev_list);
1250 if (device->bdev == root->fs_info->sb->s_bdev)
1251 root->fs_info->sb->s_bdev = next_device->bdev;
1252 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1253 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1256 close_bdev_exclusive(device->bdev, device->mode);
1257 device->bdev = NULL;
1258 device->fs_devices->open_devices--;
1261 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1262 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1264 if (device->fs_devices->open_devices == 0) {
1265 struct btrfs_fs_devices *fs_devices;
1266 fs_devices = root->fs_info->fs_devices;
1267 while (fs_devices) {
1268 if (fs_devices->seed == device->fs_devices)
1270 fs_devices = fs_devices->seed;
1272 fs_devices->seed = device->fs_devices->seed;
1273 device->fs_devices->seed = NULL;
1274 __btrfs_close_devices(device->fs_devices);
1275 free_fs_devices(device->fs_devices);
1279 * at this point, the device is zero sized. We want to
1280 * remove it from the devices list and zero out the old super
1282 if (device->writeable) {
1283 /* make sure this device isn't detected as part of
1286 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1287 set_buffer_dirty(bh);
1288 sync_dirty_buffer(bh);
1291 kfree(device->name);
1299 close_bdev_exclusive(bdev, FMODE_READ);
1301 mutex_unlock(&root->fs_info->volume_mutex);
1302 mutex_unlock(&uuid_mutex);
1307 * does all the dirty work required for changing file system's UUID.
1309 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1310 struct btrfs_root *root)
1312 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1313 struct btrfs_fs_devices *old_devices;
1314 struct btrfs_fs_devices *seed_devices;
1315 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1316 struct btrfs_device *device;
1319 BUG_ON(!mutex_is_locked(&uuid_mutex));
1320 if (!fs_devices->seeding)
1323 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1327 old_devices = clone_fs_devices(fs_devices);
1328 if (IS_ERR(old_devices)) {
1329 kfree(seed_devices);
1330 return PTR_ERR(old_devices);
1333 list_add(&old_devices->list, &fs_uuids);
1335 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1336 seed_devices->opened = 1;
1337 INIT_LIST_HEAD(&seed_devices->devices);
1338 INIT_LIST_HEAD(&seed_devices->alloc_list);
1339 mutex_init(&seed_devices->device_list_mutex);
1340 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1341 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1342 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1343 device->fs_devices = seed_devices;
1346 fs_devices->seeding = 0;
1347 fs_devices->num_devices = 0;
1348 fs_devices->open_devices = 0;
1349 fs_devices->seed = seed_devices;
1351 generate_random_uuid(fs_devices->fsid);
1352 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1353 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1354 super_flags = btrfs_super_flags(disk_super) &
1355 ~BTRFS_SUPER_FLAG_SEEDING;
1356 btrfs_set_super_flags(disk_super, super_flags);
1362 * strore the expected generation for seed devices in device items.
1364 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1365 struct btrfs_root *root)
1367 struct btrfs_path *path;
1368 struct extent_buffer *leaf;
1369 struct btrfs_dev_item *dev_item;
1370 struct btrfs_device *device;
1371 struct btrfs_key key;
1372 u8 fs_uuid[BTRFS_UUID_SIZE];
1373 u8 dev_uuid[BTRFS_UUID_SIZE];
1377 path = btrfs_alloc_path();
1381 root = root->fs_info->chunk_root;
1382 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1384 key.type = BTRFS_DEV_ITEM_KEY;
1387 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1391 leaf = path->nodes[0];
1393 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1394 ret = btrfs_next_leaf(root, path);
1399 leaf = path->nodes[0];
1400 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1401 btrfs_release_path(root, path);
1405 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1406 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1407 key.type != BTRFS_DEV_ITEM_KEY)
1410 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1411 struct btrfs_dev_item);
1412 devid = btrfs_device_id(leaf, dev_item);
1413 read_extent_buffer(leaf, dev_uuid,
1414 (unsigned long)btrfs_device_uuid(dev_item),
1416 read_extent_buffer(leaf, fs_uuid,
1417 (unsigned long)btrfs_device_fsid(dev_item),
1419 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1422 if (device->fs_devices->seeding) {
1423 btrfs_set_device_generation(leaf, dev_item,
1424 device->generation);
1425 btrfs_mark_buffer_dirty(leaf);
1433 btrfs_free_path(path);
1437 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1439 struct btrfs_trans_handle *trans;
1440 struct btrfs_device *device;
1441 struct block_device *bdev;
1442 struct list_head *devices;
1443 struct super_block *sb = root->fs_info->sb;
1445 int seeding_dev = 0;
1448 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1451 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1453 return PTR_ERR(bdev);
1455 if (root->fs_info->fs_devices->seeding) {
1457 down_write(&sb->s_umount);
1458 mutex_lock(&uuid_mutex);
1461 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1462 mutex_lock(&root->fs_info->volume_mutex);
1464 devices = &root->fs_info->fs_devices->devices;
1466 * we have the volume lock, so we don't need the extra
1467 * device list mutex while reading the list here.
1469 list_for_each_entry(device, devices, dev_list) {
1470 if (device->bdev == bdev) {
1476 device = kzalloc(sizeof(*device), GFP_NOFS);
1478 /* we can safely leave the fs_devices entry around */
1483 device->name = kstrdup(device_path, GFP_NOFS);
1484 if (!device->name) {
1490 ret = find_next_devid(root, &device->devid);
1496 trans = btrfs_start_transaction(root, 0);
1499 device->barriers = 1;
1500 device->writeable = 1;
1501 device->work.func = pending_bios_fn;
1502 generate_random_uuid(device->uuid);
1503 spin_lock_init(&device->io_lock);
1504 device->generation = trans->transid;
1505 device->io_width = root->sectorsize;
1506 device->io_align = root->sectorsize;
1507 device->sector_size = root->sectorsize;
1508 device->total_bytes = i_size_read(bdev->bd_inode);
1509 device->disk_total_bytes = device->total_bytes;
1510 device->dev_root = root->fs_info->dev_root;
1511 device->bdev = bdev;
1512 device->in_fs_metadata = 1;
1514 set_blocksize(device->bdev, 4096);
1517 sb->s_flags &= ~MS_RDONLY;
1518 ret = btrfs_prepare_sprout(trans, root);
1522 device->fs_devices = root->fs_info->fs_devices;
1525 * we don't want write_supers to jump in here with our device
1528 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1529 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1530 list_add(&device->dev_alloc_list,
1531 &root->fs_info->fs_devices->alloc_list);
1532 root->fs_info->fs_devices->num_devices++;
1533 root->fs_info->fs_devices->open_devices++;
1534 root->fs_info->fs_devices->rw_devices++;
1535 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1537 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1538 root->fs_info->fs_devices->rotating = 1;
1540 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1541 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1542 total_bytes + device->total_bytes);
1544 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1545 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1547 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1550 ret = init_first_rw_device(trans, root, device);
1552 ret = btrfs_finish_sprout(trans, root);
1555 ret = btrfs_add_device(trans, root, device);
1559 * we've got more storage, clear any full flags on the space
1562 btrfs_clear_space_info_full(root->fs_info);
1564 unlock_chunks(root);
1565 btrfs_commit_transaction(trans, root);
1568 mutex_unlock(&uuid_mutex);
1569 up_write(&sb->s_umount);
1571 ret = btrfs_relocate_sys_chunks(root);
1575 mutex_unlock(&root->fs_info->volume_mutex);
1578 close_bdev_exclusive(bdev, 0);
1580 mutex_unlock(&uuid_mutex);
1581 up_write(&sb->s_umount);
1586 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1587 struct btrfs_device *device)
1590 struct btrfs_path *path;
1591 struct btrfs_root *root;
1592 struct btrfs_dev_item *dev_item;
1593 struct extent_buffer *leaf;
1594 struct btrfs_key key;
1596 root = device->dev_root->fs_info->chunk_root;
1598 path = btrfs_alloc_path();
1602 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1603 key.type = BTRFS_DEV_ITEM_KEY;
1604 key.offset = device->devid;
1606 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1615 leaf = path->nodes[0];
1616 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1618 btrfs_set_device_id(leaf, dev_item, device->devid);
1619 btrfs_set_device_type(leaf, dev_item, device->type);
1620 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1621 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1622 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1623 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1624 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1625 btrfs_mark_buffer_dirty(leaf);
1628 btrfs_free_path(path);
1632 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1633 struct btrfs_device *device, u64 new_size)
1635 struct btrfs_super_block *super_copy =
1636 &device->dev_root->fs_info->super_copy;
1637 u64 old_total = btrfs_super_total_bytes(super_copy);
1638 u64 diff = new_size - device->total_bytes;
1640 if (!device->writeable)
1642 if (new_size <= device->total_bytes)
1645 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1646 device->fs_devices->total_rw_bytes += diff;
1648 device->total_bytes = new_size;
1649 device->disk_total_bytes = new_size;
1650 btrfs_clear_space_info_full(device->dev_root->fs_info);
1652 return btrfs_update_device(trans, device);
1655 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1656 struct btrfs_device *device, u64 new_size)
1659 lock_chunks(device->dev_root);
1660 ret = __btrfs_grow_device(trans, device, new_size);
1661 unlock_chunks(device->dev_root);
1665 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1666 struct btrfs_root *root,
1667 u64 chunk_tree, u64 chunk_objectid,
1671 struct btrfs_path *path;
1672 struct btrfs_key key;
1674 root = root->fs_info->chunk_root;
1675 path = btrfs_alloc_path();
1679 key.objectid = chunk_objectid;
1680 key.offset = chunk_offset;
1681 key.type = BTRFS_CHUNK_ITEM_KEY;
1683 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1686 ret = btrfs_del_item(trans, root, path);
1689 btrfs_free_path(path);
1693 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1696 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1697 struct btrfs_disk_key *disk_key;
1698 struct btrfs_chunk *chunk;
1705 struct btrfs_key key;
1707 array_size = btrfs_super_sys_array_size(super_copy);
1709 ptr = super_copy->sys_chunk_array;
1712 while (cur < array_size) {
1713 disk_key = (struct btrfs_disk_key *)ptr;
1714 btrfs_disk_key_to_cpu(&key, disk_key);
1716 len = sizeof(*disk_key);
1718 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1719 chunk = (struct btrfs_chunk *)(ptr + len);
1720 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1721 len += btrfs_chunk_item_size(num_stripes);
1726 if (key.objectid == chunk_objectid &&
1727 key.offset == chunk_offset) {
1728 memmove(ptr, ptr + len, array_size - (cur + len));
1730 btrfs_set_super_sys_array_size(super_copy, array_size);
1739 static int btrfs_relocate_chunk(struct btrfs_root *root,
1740 u64 chunk_tree, u64 chunk_objectid,
1743 struct extent_map_tree *em_tree;
1744 struct btrfs_root *extent_root;
1745 struct btrfs_trans_handle *trans;
1746 struct extent_map *em;
1747 struct map_lookup *map;
1751 root = root->fs_info->chunk_root;
1752 extent_root = root->fs_info->extent_root;
1753 em_tree = &root->fs_info->mapping_tree.map_tree;
1755 ret = btrfs_can_relocate(extent_root, chunk_offset);
1759 /* step one, relocate all the extents inside this chunk */
1760 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1764 trans = btrfs_start_transaction(root, 0);
1770 * step two, delete the device extents and the
1771 * chunk tree entries
1773 read_lock(&em_tree->lock);
1774 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1775 read_unlock(&em_tree->lock);
1777 BUG_ON(em->start > chunk_offset ||
1778 em->start + em->len < chunk_offset);
1779 map = (struct map_lookup *)em->bdev;
1781 for (i = 0; i < map->num_stripes; i++) {
1782 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1783 map->stripes[i].physical);
1786 if (map->stripes[i].dev) {
1787 ret = btrfs_update_device(trans, map->stripes[i].dev);
1791 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1796 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1797 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1801 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1804 write_lock(&em_tree->lock);
1805 remove_extent_mapping(em_tree, em);
1806 write_unlock(&em_tree->lock);
1811 /* once for the tree */
1812 free_extent_map(em);
1814 free_extent_map(em);
1816 unlock_chunks(root);
1817 btrfs_end_transaction(trans, root);
1821 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1823 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1824 struct btrfs_path *path;
1825 struct extent_buffer *leaf;
1826 struct btrfs_chunk *chunk;
1827 struct btrfs_key key;
1828 struct btrfs_key found_key;
1829 u64 chunk_tree = chunk_root->root_key.objectid;
1831 bool retried = false;
1835 path = btrfs_alloc_path();
1840 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1841 key.offset = (u64)-1;
1842 key.type = BTRFS_CHUNK_ITEM_KEY;
1845 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1850 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1857 leaf = path->nodes[0];
1858 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1860 chunk = btrfs_item_ptr(leaf, path->slots[0],
1861 struct btrfs_chunk);
1862 chunk_type = btrfs_chunk_type(leaf, chunk);
1863 btrfs_release_path(chunk_root, path);
1865 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1866 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1875 if (found_key.offset == 0)
1877 key.offset = found_key.offset - 1;
1880 if (failed && !retried) {
1884 } else if (failed && retried) {
1889 btrfs_free_path(path);
1893 static u64 div_factor(u64 num, int factor)
1902 int btrfs_balance(struct btrfs_root *dev_root)
1905 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1906 struct btrfs_device *device;
1909 struct btrfs_path *path;
1910 struct btrfs_key key;
1911 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1912 struct btrfs_trans_handle *trans;
1913 struct btrfs_key found_key;
1915 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1918 mutex_lock(&dev_root->fs_info->volume_mutex);
1919 dev_root = dev_root->fs_info->dev_root;
1921 /* step one make some room on all the devices */
1922 list_for_each_entry(device, devices, dev_list) {
1923 old_size = device->total_bytes;
1924 size_to_free = div_factor(old_size, 1);
1925 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1926 if (!device->writeable ||
1927 device->total_bytes - device->bytes_used > size_to_free)
1930 ret = btrfs_shrink_device(device, old_size - size_to_free);
1935 trans = btrfs_start_transaction(dev_root, 0);
1938 ret = btrfs_grow_device(trans, device, old_size);
1941 btrfs_end_transaction(trans, dev_root);
1944 /* step two, relocate all the chunks */
1945 path = btrfs_alloc_path();
1948 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1949 key.offset = (u64)-1;
1950 key.type = BTRFS_CHUNK_ITEM_KEY;
1953 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1958 * this shouldn't happen, it means the last relocate
1964 ret = btrfs_previous_item(chunk_root, path, 0,
1965 BTRFS_CHUNK_ITEM_KEY);
1969 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1971 if (found_key.objectid != key.objectid)
1974 /* chunk zero is special */
1975 if (found_key.offset == 0)
1978 btrfs_release_path(chunk_root, path);
1979 ret = btrfs_relocate_chunk(chunk_root,
1980 chunk_root->root_key.objectid,
1983 BUG_ON(ret && ret != -ENOSPC);
1984 key.offset = found_key.offset - 1;
1988 btrfs_free_path(path);
1989 mutex_unlock(&dev_root->fs_info->volume_mutex);
1994 * shrinking a device means finding all of the device extents past
1995 * the new size, and then following the back refs to the chunks.
1996 * The chunk relocation code actually frees the device extent
1998 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
2000 struct btrfs_trans_handle *trans;
2001 struct btrfs_root *root = device->dev_root;
2002 struct btrfs_dev_extent *dev_extent = NULL;
2003 struct btrfs_path *path;
2011 bool retried = false;
2012 struct extent_buffer *l;
2013 struct btrfs_key key;
2014 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2015 u64 old_total = btrfs_super_total_bytes(super_copy);
2016 u64 old_size = device->total_bytes;
2017 u64 diff = device->total_bytes - new_size;
2019 if (new_size >= device->total_bytes)
2022 path = btrfs_alloc_path();
2030 device->total_bytes = new_size;
2031 if (device->writeable)
2032 device->fs_devices->total_rw_bytes -= diff;
2033 unlock_chunks(root);
2036 key.objectid = device->devid;
2037 key.offset = (u64)-1;
2038 key.type = BTRFS_DEV_EXTENT_KEY;
2041 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2045 ret = btrfs_previous_item(root, path, 0, key.type);
2050 btrfs_release_path(root, path);
2055 slot = path->slots[0];
2056 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2058 if (key.objectid != device->devid) {
2059 btrfs_release_path(root, path);
2063 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2064 length = btrfs_dev_extent_length(l, dev_extent);
2066 if (key.offset + length <= new_size) {
2067 btrfs_release_path(root, path);
2071 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2072 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2073 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2074 btrfs_release_path(root, path);
2076 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2078 if (ret && ret != -ENOSPC)
2085 if (failed && !retried) {
2089 } else if (failed && retried) {
2093 device->total_bytes = old_size;
2094 if (device->writeable)
2095 device->fs_devices->total_rw_bytes += diff;
2096 unlock_chunks(root);
2100 /* Shrinking succeeded, else we would be at "done". */
2101 trans = btrfs_start_transaction(root, 0);
2104 device->disk_total_bytes = new_size;
2105 /* Now btrfs_update_device() will change the on-disk size. */
2106 ret = btrfs_update_device(trans, device);
2108 unlock_chunks(root);
2109 btrfs_end_transaction(trans, root);
2112 WARN_ON(diff > old_total);
2113 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2114 unlock_chunks(root);
2115 btrfs_end_transaction(trans, root);
2117 btrfs_free_path(path);
2121 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2122 struct btrfs_root *root,
2123 struct btrfs_key *key,
2124 struct btrfs_chunk *chunk, int item_size)
2126 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2127 struct btrfs_disk_key disk_key;
2131 array_size = btrfs_super_sys_array_size(super_copy);
2132 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2135 ptr = super_copy->sys_chunk_array + array_size;
2136 btrfs_cpu_key_to_disk(&disk_key, key);
2137 memcpy(ptr, &disk_key, sizeof(disk_key));
2138 ptr += sizeof(disk_key);
2139 memcpy(ptr, chunk, item_size);
2140 item_size += sizeof(disk_key);
2141 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2145 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2146 int num_stripes, int sub_stripes)
2148 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2150 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2151 return calc_size * (num_stripes / sub_stripes);
2153 return calc_size * num_stripes;
2156 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2157 struct btrfs_root *extent_root,
2158 struct map_lookup **map_ret,
2159 u64 *num_bytes, u64 *stripe_size,
2160 u64 start, u64 type)
2162 struct btrfs_fs_info *info = extent_root->fs_info;
2163 struct btrfs_device *device = NULL;
2164 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2165 struct list_head *cur;
2166 struct map_lookup *map = NULL;
2167 struct extent_map_tree *em_tree;
2168 struct extent_map *em;
2169 struct list_head private_devs;
2170 int min_stripe_size = 1 * 1024 * 1024;
2171 u64 calc_size = 1024 * 1024 * 1024;
2172 u64 max_chunk_size = calc_size;
2177 int num_stripes = 1;
2178 int min_stripes = 1;
2179 int sub_stripes = 0;
2184 int stripe_len = 64 * 1024;
2186 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2187 (type & BTRFS_BLOCK_GROUP_DUP)) {
2189 type &= ~BTRFS_BLOCK_GROUP_DUP;
2191 if (list_empty(&fs_devices->alloc_list))
2194 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2195 num_stripes = fs_devices->rw_devices;
2198 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2203 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2204 if (fs_devices->rw_devices < 2)
2210 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2211 num_stripes = fs_devices->rw_devices;
2212 if (num_stripes < 4)
2214 num_stripes &= ~(u32)1;
2220 if (type & BTRFS_BLOCK_GROUP_DATA) {
2221 max_chunk_size = 10 * calc_size;
2222 min_stripe_size = 64 * 1024 * 1024;
2223 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2224 max_chunk_size = 256 * 1024 * 1024;
2225 min_stripe_size = 32 * 1024 * 1024;
2226 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2227 calc_size = 8 * 1024 * 1024;
2228 max_chunk_size = calc_size * 2;
2229 min_stripe_size = 1 * 1024 * 1024;
2232 /* we don't want a chunk larger than 10% of writeable space */
2233 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2238 if (!map || map->num_stripes != num_stripes) {
2240 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2243 map->num_stripes = num_stripes;
2246 if (calc_size * num_stripes > max_chunk_size * ncopies) {
2247 calc_size = max_chunk_size * ncopies;
2248 do_div(calc_size, num_stripes);
2249 do_div(calc_size, stripe_len);
2250 calc_size *= stripe_len;
2253 /* we don't want tiny stripes */
2255 calc_size = max_t(u64, min_stripe_size, calc_size);
2258 * we're about to do_div by the stripe_len so lets make sure
2259 * we end up with something bigger than a stripe
2261 calc_size = max_t(u64, calc_size, stripe_len * 4);
2263 do_div(calc_size, stripe_len);
2264 calc_size *= stripe_len;
2266 cur = fs_devices->alloc_list.next;
2269 if (type & BTRFS_BLOCK_GROUP_DUP)
2270 min_free = calc_size * 2;
2272 min_free = calc_size;
2275 * we add 1MB because we never use the first 1MB of the device, unless
2276 * we've looped, then we are likely allocating the maximum amount of
2277 * space left already
2280 min_free += 1024 * 1024;
2282 INIT_LIST_HEAD(&private_devs);
2283 while (index < num_stripes) {
2284 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2285 BUG_ON(!device->writeable);
2286 if (device->total_bytes > device->bytes_used)
2287 avail = device->total_bytes - device->bytes_used;
2292 if (device->in_fs_metadata && avail >= min_free) {
2293 ret = find_free_dev_extent(trans, device,
2294 min_free, &dev_offset,
2297 list_move_tail(&device->dev_alloc_list,
2299 map->stripes[index].dev = device;
2300 map->stripes[index].physical = dev_offset;
2302 if (type & BTRFS_BLOCK_GROUP_DUP) {
2303 map->stripes[index].dev = device;
2304 map->stripes[index].physical =
2305 dev_offset + calc_size;
2309 } else if (device->in_fs_metadata && avail > max_avail)
2311 if (cur == &fs_devices->alloc_list)
2314 list_splice(&private_devs, &fs_devices->alloc_list);
2315 if (index < num_stripes) {
2316 if (index >= min_stripes) {
2317 num_stripes = index;
2318 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2319 num_stripes /= sub_stripes;
2320 num_stripes *= sub_stripes;
2325 if (!looped && max_avail > 0) {
2327 calc_size = max_avail;
2328 if (type & BTRFS_BLOCK_GROUP_DUP)
2329 do_div(calc_size, 2);
2335 map->sector_size = extent_root->sectorsize;
2336 map->stripe_len = stripe_len;
2337 map->io_align = stripe_len;
2338 map->io_width = stripe_len;
2340 map->num_stripes = num_stripes;
2341 map->sub_stripes = sub_stripes;
2344 *stripe_size = calc_size;
2345 *num_bytes = chunk_bytes_by_type(type, calc_size,
2346 num_stripes, sub_stripes);
2348 em = alloc_extent_map(GFP_NOFS);
2353 em->bdev = (struct block_device *)map;
2355 em->len = *num_bytes;
2356 em->block_start = 0;
2357 em->block_len = em->len;
2359 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2360 write_lock(&em_tree->lock);
2361 ret = add_extent_mapping(em_tree, em);
2362 write_unlock(&em_tree->lock);
2364 free_extent_map(em);
2366 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2367 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2372 while (index < map->num_stripes) {
2373 device = map->stripes[index].dev;
2374 dev_offset = map->stripes[index].physical;
2376 ret = btrfs_alloc_dev_extent(trans, device,
2377 info->chunk_root->root_key.objectid,
2378 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2379 start, dev_offset, calc_size);
2387 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2388 struct btrfs_root *extent_root,
2389 struct map_lookup *map, u64 chunk_offset,
2390 u64 chunk_size, u64 stripe_size)
2393 struct btrfs_key key;
2394 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2395 struct btrfs_device *device;
2396 struct btrfs_chunk *chunk;
2397 struct btrfs_stripe *stripe;
2398 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2402 chunk = kzalloc(item_size, GFP_NOFS);
2407 while (index < map->num_stripes) {
2408 device = map->stripes[index].dev;
2409 device->bytes_used += stripe_size;
2410 ret = btrfs_update_device(trans, device);
2416 stripe = &chunk->stripe;
2417 while (index < map->num_stripes) {
2418 device = map->stripes[index].dev;
2419 dev_offset = map->stripes[index].physical;
2421 btrfs_set_stack_stripe_devid(stripe, device->devid);
2422 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2423 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2428 btrfs_set_stack_chunk_length(chunk, chunk_size);
2429 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2430 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2431 btrfs_set_stack_chunk_type(chunk, map->type);
2432 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2433 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2434 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2435 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2436 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2438 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2439 key.type = BTRFS_CHUNK_ITEM_KEY;
2440 key.offset = chunk_offset;
2442 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2445 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2446 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2455 * Chunk allocation falls into two parts. The first part does works
2456 * that make the new allocated chunk useable, but not do any operation
2457 * that modifies the chunk tree. The second part does the works that
2458 * require modifying the chunk tree. This division is important for the
2459 * bootstrap process of adding storage to a seed btrfs.
2461 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2462 struct btrfs_root *extent_root, u64 type)
2467 struct map_lookup *map;
2468 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2471 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2476 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2477 &stripe_size, chunk_offset, type);
2481 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2482 chunk_size, stripe_size);
2487 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2488 struct btrfs_root *root,
2489 struct btrfs_device *device)
2492 u64 sys_chunk_offset;
2496 u64 sys_stripe_size;
2498 struct map_lookup *map;
2499 struct map_lookup *sys_map;
2500 struct btrfs_fs_info *fs_info = root->fs_info;
2501 struct btrfs_root *extent_root = fs_info->extent_root;
2504 ret = find_next_chunk(fs_info->chunk_root,
2505 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2508 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2509 (fs_info->metadata_alloc_profile &
2510 fs_info->avail_metadata_alloc_bits);
2511 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2513 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2514 &stripe_size, chunk_offset, alloc_profile);
2517 sys_chunk_offset = chunk_offset + chunk_size;
2519 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2520 (fs_info->system_alloc_profile &
2521 fs_info->avail_system_alloc_bits);
2522 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2524 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2525 &sys_chunk_size, &sys_stripe_size,
2526 sys_chunk_offset, alloc_profile);
2529 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2533 * Modifying chunk tree needs allocating new blocks from both
2534 * system block group and metadata block group. So we only can
2535 * do operations require modifying the chunk tree after both
2536 * block groups were created.
2538 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2539 chunk_size, stripe_size);
2542 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2543 sys_chunk_offset, sys_chunk_size,
2549 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2551 struct extent_map *em;
2552 struct map_lookup *map;
2553 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2557 read_lock(&map_tree->map_tree.lock);
2558 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2559 read_unlock(&map_tree->map_tree.lock);
2563 if (btrfs_test_opt(root, DEGRADED)) {
2564 free_extent_map(em);
2568 map = (struct map_lookup *)em->bdev;
2569 for (i = 0; i < map->num_stripes; i++) {
2570 if (!map->stripes[i].dev->writeable) {
2575 free_extent_map(em);
2579 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2581 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2584 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2586 struct extent_map *em;
2589 write_lock(&tree->map_tree.lock);
2590 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2592 remove_extent_mapping(&tree->map_tree, em);
2593 write_unlock(&tree->map_tree.lock);
2598 free_extent_map(em);
2599 /* once for the tree */
2600 free_extent_map(em);
2604 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2606 struct extent_map *em;
2607 struct map_lookup *map;
2608 struct extent_map_tree *em_tree = &map_tree->map_tree;
2611 read_lock(&em_tree->lock);
2612 em = lookup_extent_mapping(em_tree, logical, len);
2613 read_unlock(&em_tree->lock);
2616 BUG_ON(em->start > logical || em->start + em->len < logical);
2617 map = (struct map_lookup *)em->bdev;
2618 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2619 ret = map->num_stripes;
2620 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2621 ret = map->sub_stripes;
2624 free_extent_map(em);
2628 static int find_live_mirror(struct map_lookup *map, int first, int num,
2632 if (map->stripes[optimal].dev->bdev)
2634 for (i = first; i < first + num; i++) {
2635 if (map->stripes[i].dev->bdev)
2638 /* we couldn't find one that doesn't fail. Just return something
2639 * and the io error handling code will clean up eventually
2644 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2645 u64 logical, u64 *length,
2646 struct btrfs_multi_bio **multi_ret,
2647 int mirror_num, struct page *unplug_page)
2649 struct extent_map *em;
2650 struct map_lookup *map;
2651 struct extent_map_tree *em_tree = &map_tree->map_tree;
2655 int stripes_allocated = 8;
2656 int stripes_required = 1;
2661 struct btrfs_multi_bio *multi = NULL;
2663 if (multi_ret && !(rw & REQ_WRITE))
2664 stripes_allocated = 1;
2667 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2672 atomic_set(&multi->error, 0);
2675 read_lock(&em_tree->lock);
2676 em = lookup_extent_mapping(em_tree, logical, *length);
2677 read_unlock(&em_tree->lock);
2679 if (!em && unplug_page) {
2685 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2686 (unsigned long long)logical,
2687 (unsigned long long)*length);
2691 BUG_ON(em->start > logical || em->start + em->len < logical);
2692 map = (struct map_lookup *)em->bdev;
2693 offset = logical - em->start;
2695 if (mirror_num > map->num_stripes)
2698 /* if our multi bio struct is too small, back off and try again */
2699 if (rw & REQ_WRITE) {
2700 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2701 BTRFS_BLOCK_GROUP_DUP)) {
2702 stripes_required = map->num_stripes;
2704 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2705 stripes_required = map->sub_stripes;
2709 if (multi_ret && (rw & REQ_WRITE) &&
2710 stripes_allocated < stripes_required) {
2711 stripes_allocated = map->num_stripes;
2712 free_extent_map(em);
2718 * stripe_nr counts the total number of stripes we have to stride
2719 * to get to this block
2721 do_div(stripe_nr, map->stripe_len);
2723 stripe_offset = stripe_nr * map->stripe_len;
2724 BUG_ON(offset < stripe_offset);
2726 /* stripe_offset is the offset of this block in its stripe*/
2727 stripe_offset = offset - stripe_offset;
2729 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2730 BTRFS_BLOCK_GROUP_RAID10 |
2731 BTRFS_BLOCK_GROUP_DUP)) {
2732 /* we limit the length of each bio to what fits in a stripe */
2733 *length = min_t(u64, em->len - offset,
2734 map->stripe_len - stripe_offset);
2736 *length = em->len - offset;
2739 if (!multi_ret && !unplug_page)
2744 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2745 if (unplug_page || (rw & REQ_WRITE))
2746 num_stripes = map->num_stripes;
2747 else if (mirror_num)
2748 stripe_index = mirror_num - 1;
2750 stripe_index = find_live_mirror(map, 0,
2752 current->pid % map->num_stripes);
2755 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2757 num_stripes = map->num_stripes;
2758 else if (mirror_num)
2759 stripe_index = mirror_num - 1;
2761 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2762 int factor = map->num_stripes / map->sub_stripes;
2764 stripe_index = do_div(stripe_nr, factor);
2765 stripe_index *= map->sub_stripes;
2767 if (unplug_page || (rw & REQ_WRITE))
2768 num_stripes = map->sub_stripes;
2769 else if (mirror_num)
2770 stripe_index += mirror_num - 1;
2772 stripe_index = find_live_mirror(map, stripe_index,
2773 map->sub_stripes, stripe_index +
2774 current->pid % map->sub_stripes);
2778 * after this do_div call, stripe_nr is the number of stripes
2779 * on this device we have to walk to find the data, and
2780 * stripe_index is the number of our device in the stripe array
2782 stripe_index = do_div(stripe_nr, map->num_stripes);
2784 BUG_ON(stripe_index >= map->num_stripes);
2786 for (i = 0; i < num_stripes; i++) {
2788 struct btrfs_device *device;
2789 struct backing_dev_info *bdi;
2791 device = map->stripes[stripe_index].dev;
2793 bdi = blk_get_backing_dev_info(device->bdev);
2794 if (bdi->unplug_io_fn)
2795 bdi->unplug_io_fn(bdi, unplug_page);
2798 multi->stripes[i].physical =
2799 map->stripes[stripe_index].physical +
2800 stripe_offset + stripe_nr * map->stripe_len;
2801 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2807 multi->num_stripes = num_stripes;
2808 multi->max_errors = max_errors;
2811 free_extent_map(em);
2815 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2816 u64 logical, u64 *length,
2817 struct btrfs_multi_bio **multi_ret, int mirror_num)
2819 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2823 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2824 u64 chunk_start, u64 physical, u64 devid,
2825 u64 **logical, int *naddrs, int *stripe_len)
2827 struct extent_map_tree *em_tree = &map_tree->map_tree;
2828 struct extent_map *em;
2829 struct map_lookup *map;
2836 read_lock(&em_tree->lock);
2837 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2838 read_unlock(&em_tree->lock);
2840 BUG_ON(!em || em->start != chunk_start);
2841 map = (struct map_lookup *)em->bdev;
2844 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2845 do_div(length, map->num_stripes / map->sub_stripes);
2846 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2847 do_div(length, map->num_stripes);
2849 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2852 for (i = 0; i < map->num_stripes; i++) {
2853 if (devid && map->stripes[i].dev->devid != devid)
2855 if (map->stripes[i].physical > physical ||
2856 map->stripes[i].physical + length <= physical)
2859 stripe_nr = physical - map->stripes[i].physical;
2860 do_div(stripe_nr, map->stripe_len);
2862 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2863 stripe_nr = stripe_nr * map->num_stripes + i;
2864 do_div(stripe_nr, map->sub_stripes);
2865 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2866 stripe_nr = stripe_nr * map->num_stripes + i;
2868 bytenr = chunk_start + stripe_nr * map->stripe_len;
2869 WARN_ON(nr >= map->num_stripes);
2870 for (j = 0; j < nr; j++) {
2871 if (buf[j] == bytenr)
2875 WARN_ON(nr >= map->num_stripes);
2882 *stripe_len = map->stripe_len;
2884 free_extent_map(em);
2888 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2889 u64 logical, struct page *page)
2891 u64 length = PAGE_CACHE_SIZE;
2892 return __btrfs_map_block(map_tree, READ, logical, &length,
2896 static void end_bio_multi_stripe(struct bio *bio, int err)
2898 struct btrfs_multi_bio *multi = bio->bi_private;
2899 int is_orig_bio = 0;
2902 atomic_inc(&multi->error);
2904 if (bio == multi->orig_bio)
2907 if (atomic_dec_and_test(&multi->stripes_pending)) {
2910 bio = multi->orig_bio;
2912 bio->bi_private = multi->private;
2913 bio->bi_end_io = multi->end_io;
2914 /* only send an error to the higher layers if it is
2915 * beyond the tolerance of the multi-bio
2917 if (atomic_read(&multi->error) > multi->max_errors) {
2921 * this bio is actually up to date, we didn't
2922 * go over the max number of errors
2924 set_bit(BIO_UPTODATE, &bio->bi_flags);
2929 bio_endio(bio, err);
2930 } else if (!is_orig_bio) {
2935 struct async_sched {
2938 struct btrfs_fs_info *info;
2939 struct btrfs_work work;
2943 * see run_scheduled_bios for a description of why bios are collected for
2946 * This will add one bio to the pending list for a device and make sure
2947 * the work struct is scheduled.
2949 static noinline int schedule_bio(struct btrfs_root *root,
2950 struct btrfs_device *device,
2951 int rw, struct bio *bio)
2953 int should_queue = 1;
2954 struct btrfs_pending_bios *pending_bios;
2956 /* don't bother with additional async steps for reads, right now */
2957 if (!(rw & REQ_WRITE)) {
2959 submit_bio(rw, bio);
2965 * nr_async_bios allows us to reliably return congestion to the
2966 * higher layers. Otherwise, the async bio makes it appear we have
2967 * made progress against dirty pages when we've really just put it
2968 * on a queue for later
2970 atomic_inc(&root->fs_info->nr_async_bios);
2971 WARN_ON(bio->bi_next);
2972 bio->bi_next = NULL;
2975 spin_lock(&device->io_lock);
2976 if (bio->bi_rw & REQ_SYNC)
2977 pending_bios = &device->pending_sync_bios;
2979 pending_bios = &device->pending_bios;
2981 if (pending_bios->tail)
2982 pending_bios->tail->bi_next = bio;
2984 pending_bios->tail = bio;
2985 if (!pending_bios->head)
2986 pending_bios->head = bio;
2987 if (device->running_pending)
2990 spin_unlock(&device->io_lock);
2993 btrfs_queue_worker(&root->fs_info->submit_workers,
2998 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2999 int mirror_num, int async_submit)
3001 struct btrfs_mapping_tree *map_tree;
3002 struct btrfs_device *dev;
3003 struct bio *first_bio = bio;
3004 u64 logical = (u64)bio->bi_sector << 9;
3007 struct btrfs_multi_bio *multi = NULL;
3012 length = bio->bi_size;
3013 map_tree = &root->fs_info->mapping_tree;
3014 map_length = length;
3016 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
3020 total_devs = multi->num_stripes;
3021 if (map_length < length) {
3022 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
3023 "len %llu\n", (unsigned long long)logical,
3024 (unsigned long long)length,
3025 (unsigned long long)map_length);
3028 multi->end_io = first_bio->bi_end_io;
3029 multi->private = first_bio->bi_private;
3030 multi->orig_bio = first_bio;
3031 atomic_set(&multi->stripes_pending, multi->num_stripes);
3033 while (dev_nr < total_devs) {
3034 if (total_devs > 1) {
3035 if (dev_nr < total_devs - 1) {
3036 bio = bio_clone(first_bio, GFP_NOFS);
3041 bio->bi_private = multi;
3042 bio->bi_end_io = end_bio_multi_stripe;
3044 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3045 dev = multi->stripes[dev_nr].dev;
3046 if (dev && dev->bdev && (rw != WRITE || dev->writeable)) {
3047 bio->bi_bdev = dev->bdev;
3049 schedule_bio(root, dev, rw, bio);
3051 submit_bio(rw, bio);
3053 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3054 bio->bi_sector = logical >> 9;
3055 bio_endio(bio, -EIO);
3059 if (total_devs == 1)
3064 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3067 struct btrfs_device *device;
3068 struct btrfs_fs_devices *cur_devices;
3070 cur_devices = root->fs_info->fs_devices;
3071 while (cur_devices) {
3073 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3074 device = __find_device(&cur_devices->devices,
3079 cur_devices = cur_devices->seed;
3084 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3085 u64 devid, u8 *dev_uuid)
3087 struct btrfs_device *device;
3088 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3090 device = kzalloc(sizeof(*device), GFP_NOFS);
3093 list_add(&device->dev_list,
3094 &fs_devices->devices);
3095 device->barriers = 1;
3096 device->dev_root = root->fs_info->dev_root;
3097 device->devid = devid;
3098 device->work.func = pending_bios_fn;
3099 device->fs_devices = fs_devices;
3100 device->missing = 1;
3101 fs_devices->num_devices++;
3102 fs_devices->missing_devices++;
3103 spin_lock_init(&device->io_lock);
3104 INIT_LIST_HEAD(&device->dev_alloc_list);
3105 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3109 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3110 struct extent_buffer *leaf,
3111 struct btrfs_chunk *chunk)
3113 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3114 struct map_lookup *map;
3115 struct extent_map *em;
3119 u8 uuid[BTRFS_UUID_SIZE];
3124 logical = key->offset;
3125 length = btrfs_chunk_length(leaf, chunk);
3127 read_lock(&map_tree->map_tree.lock);
3128 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3129 read_unlock(&map_tree->map_tree.lock);
3131 /* already mapped? */
3132 if (em && em->start <= logical && em->start + em->len > logical) {
3133 free_extent_map(em);
3136 free_extent_map(em);
3139 em = alloc_extent_map(GFP_NOFS);
3142 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3143 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3145 free_extent_map(em);
3149 em->bdev = (struct block_device *)map;
3150 em->start = logical;
3152 em->block_start = 0;
3153 em->block_len = em->len;
3155 map->num_stripes = num_stripes;
3156 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3157 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3158 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3159 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3160 map->type = btrfs_chunk_type(leaf, chunk);
3161 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3162 for (i = 0; i < num_stripes; i++) {
3163 map->stripes[i].physical =
3164 btrfs_stripe_offset_nr(leaf, chunk, i);
3165 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3166 read_extent_buffer(leaf, uuid, (unsigned long)
3167 btrfs_stripe_dev_uuid_nr(chunk, i),
3169 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3171 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3173 free_extent_map(em);
3176 if (!map->stripes[i].dev) {
3177 map->stripes[i].dev =
3178 add_missing_dev(root, devid, uuid);
3179 if (!map->stripes[i].dev) {
3181 free_extent_map(em);
3185 map->stripes[i].dev->in_fs_metadata = 1;
3188 write_lock(&map_tree->map_tree.lock);
3189 ret = add_extent_mapping(&map_tree->map_tree, em);
3190 write_unlock(&map_tree->map_tree.lock);
3192 free_extent_map(em);
3197 static int fill_device_from_item(struct extent_buffer *leaf,
3198 struct btrfs_dev_item *dev_item,
3199 struct btrfs_device *device)
3203 device->devid = btrfs_device_id(leaf, dev_item);
3204 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3205 device->total_bytes = device->disk_total_bytes;
3206 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3207 device->type = btrfs_device_type(leaf, dev_item);
3208 device->io_align = btrfs_device_io_align(leaf, dev_item);
3209 device->io_width = btrfs_device_io_width(leaf, dev_item);
3210 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3212 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3213 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3218 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3220 struct btrfs_fs_devices *fs_devices;
3223 mutex_lock(&uuid_mutex);
3225 fs_devices = root->fs_info->fs_devices->seed;
3226 while (fs_devices) {
3227 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3231 fs_devices = fs_devices->seed;
3234 fs_devices = find_fsid(fsid);
3240 fs_devices = clone_fs_devices(fs_devices);
3241 if (IS_ERR(fs_devices)) {
3242 ret = PTR_ERR(fs_devices);
3246 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3247 root->fs_info->bdev_holder);
3251 if (!fs_devices->seeding) {
3252 __btrfs_close_devices(fs_devices);
3253 free_fs_devices(fs_devices);
3258 fs_devices->seed = root->fs_info->fs_devices->seed;
3259 root->fs_info->fs_devices->seed = fs_devices;
3261 mutex_unlock(&uuid_mutex);
3265 static int read_one_dev(struct btrfs_root *root,
3266 struct extent_buffer *leaf,
3267 struct btrfs_dev_item *dev_item)
3269 struct btrfs_device *device;
3272 u8 fs_uuid[BTRFS_UUID_SIZE];
3273 u8 dev_uuid[BTRFS_UUID_SIZE];
3275 devid = btrfs_device_id(leaf, dev_item);
3276 read_extent_buffer(leaf, dev_uuid,
3277 (unsigned long)btrfs_device_uuid(dev_item),
3279 read_extent_buffer(leaf, fs_uuid,
3280 (unsigned long)btrfs_device_fsid(dev_item),
3283 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3284 ret = open_seed_devices(root, fs_uuid);
3285 if (ret && !btrfs_test_opt(root, DEGRADED))
3289 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3290 if (!device || !device->bdev) {
3291 if (!btrfs_test_opt(root, DEGRADED))
3295 printk(KERN_WARNING "warning devid %llu missing\n",
3296 (unsigned long long)devid);
3297 device = add_missing_dev(root, devid, dev_uuid);
3300 } else if (!device->missing) {
3302 * this happens when a device that was properly setup
3303 * in the device info lists suddenly goes bad.
3304 * device->bdev is NULL, and so we have to set
3305 * device->missing to one here
3307 root->fs_info->fs_devices->missing_devices++;
3308 device->missing = 1;
3312 if (device->fs_devices != root->fs_info->fs_devices) {
3313 BUG_ON(device->writeable);
3314 if (device->generation !=
3315 btrfs_device_generation(leaf, dev_item))
3319 fill_device_from_item(leaf, dev_item, device);
3320 device->dev_root = root->fs_info->dev_root;
3321 device->in_fs_metadata = 1;
3322 if (device->writeable)
3323 device->fs_devices->total_rw_bytes += device->total_bytes;
3328 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3330 struct btrfs_dev_item *dev_item;
3332 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3334 return read_one_dev(root, buf, dev_item);
3337 int btrfs_read_sys_array(struct btrfs_root *root)
3339 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3340 struct extent_buffer *sb;
3341 struct btrfs_disk_key *disk_key;
3342 struct btrfs_chunk *chunk;
3344 unsigned long sb_ptr;
3350 struct btrfs_key key;
3352 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3353 BTRFS_SUPER_INFO_SIZE);
3356 btrfs_set_buffer_uptodate(sb);
3357 btrfs_set_buffer_lockdep_class(sb, 0);
3359 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3360 array_size = btrfs_super_sys_array_size(super_copy);
3362 ptr = super_copy->sys_chunk_array;
3363 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3366 while (cur < array_size) {
3367 disk_key = (struct btrfs_disk_key *)ptr;
3368 btrfs_disk_key_to_cpu(&key, disk_key);
3370 len = sizeof(*disk_key); ptr += len;
3374 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3375 chunk = (struct btrfs_chunk *)sb_ptr;
3376 ret = read_one_chunk(root, &key, sb, chunk);
3379 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3380 len = btrfs_chunk_item_size(num_stripes);
3389 free_extent_buffer(sb);
3393 int btrfs_read_chunk_tree(struct btrfs_root *root)
3395 struct btrfs_path *path;
3396 struct extent_buffer *leaf;
3397 struct btrfs_key key;
3398 struct btrfs_key found_key;
3402 root = root->fs_info->chunk_root;
3404 path = btrfs_alloc_path();
3408 /* first we search for all of the device items, and then we
3409 * read in all of the chunk items. This way we can create chunk
3410 * mappings that reference all of the devices that are afound
3412 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3416 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3420 leaf = path->nodes[0];
3421 slot = path->slots[0];
3422 if (slot >= btrfs_header_nritems(leaf)) {
3423 ret = btrfs_next_leaf(root, path);
3430 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3431 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3432 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3434 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3435 struct btrfs_dev_item *dev_item;
3436 dev_item = btrfs_item_ptr(leaf, slot,
3437 struct btrfs_dev_item);
3438 ret = read_one_dev(root, leaf, dev_item);
3442 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3443 struct btrfs_chunk *chunk;
3444 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3445 ret = read_one_chunk(root, &found_key, leaf, chunk);
3451 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3453 btrfs_release_path(root, path);
3458 btrfs_free_path(path);