2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <linux/iocontext.h>
24 #include <asm/div64.h>
27 #include "extent_map.h"
29 #include "transaction.h"
30 #include "print-tree.h"
32 #include "async-thread.h"
42 struct btrfs_bio_stripe stripes[];
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 void btrfs_lock_volumes(void)
58 mutex_lock(&uuid_mutex);
61 void btrfs_unlock_volumes(void)
63 mutex_unlock(&uuid_mutex);
66 static void lock_chunks(struct btrfs_root *root)
68 mutex_lock(&root->fs_info->chunk_mutex);
71 static void unlock_chunks(struct btrfs_root *root)
73 mutex_unlock(&root->fs_info->chunk_mutex);
76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
90 int btrfs_cleanup_fs_uuids(void)
92 struct btrfs_fs_devices *fs_devices;
94 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
98 free_fs_devices(fs_devices);
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
106 struct btrfs_device *dev;
108 list_for_each_entry(dev, head, dev_list) {
109 if (dev->devid == devid &&
110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 struct btrfs_fs_devices *fs_devices;
121 list_for_each_entry(fs_devices, &fs_uuids, list) {
122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
129 struct bio *head, struct bio *tail)
132 struct bio *old_head;
134 old_head = pending_bios->head;
135 pending_bios->head = head;
136 if (pending_bios->tail)
137 tail->bi_next = old_head;
139 pending_bios->tail = tail;
143 * we try to collect pending bios for a device so we don't get a large
144 * number of procs sending bios down to the same device. This greatly
145 * improves the schedulers ability to collect and merge the bios.
147 * But, it also turns into a long list of bios to process and that is sure
148 * to eventually make the worker thread block. The solution here is to
149 * make some progress and then put this work struct back at the end of
150 * the list if the block device is congested. This way, multiple devices
151 * can make progress from a single worker thread.
153 static noinline int run_scheduled_bios(struct btrfs_device *device)
156 struct backing_dev_info *bdi;
157 struct btrfs_fs_info *fs_info;
158 struct btrfs_pending_bios *pending_bios;
162 unsigned long num_run;
163 unsigned long num_sync_run;
165 unsigned long last_waited = 0;
167 bdi = blk_get_backing_dev_info(device->bdev);
168 fs_info = device->dev_root->fs_info;
169 limit = btrfs_async_submit_limit(fs_info);
170 limit = limit * 2 / 3;
172 /* we want to make sure that every time we switch from the sync
173 * list to the normal list, we unplug
178 spin_lock(&device->io_lock);
183 /* take all the bios off the list at once and process them
184 * later on (without the lock held). But, remember the
185 * tail and other pointers so the bios can be properly reinserted
186 * into the list if we hit congestion
188 if (device->pending_sync_bios.head)
189 pending_bios = &device->pending_sync_bios;
191 pending_bios = &device->pending_bios;
193 pending = pending_bios->head;
194 tail = pending_bios->tail;
195 WARN_ON(pending && !tail);
198 * if pending was null this time around, no bios need processing
199 * at all and we can stop. Otherwise it'll loop back up again
200 * and do an additional check so no bios are missed.
202 * device->running_pending is used to synchronize with the
205 if (device->pending_sync_bios.head == NULL &&
206 device->pending_bios.head == NULL) {
208 device->running_pending = 0;
211 device->running_pending = 1;
214 pending_bios->head = NULL;
215 pending_bios->tail = NULL;
217 spin_unlock(&device->io_lock);
220 * if we're doing the regular priority list, make sure we unplug
221 * for any high prio bios we've sent down
223 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
225 blk_run_backing_dev(bdi, NULL);
231 if (pending_bios != &device->pending_sync_bios &&
232 device->pending_sync_bios.head &&
235 spin_lock(&device->io_lock);
236 requeue_list(pending_bios, pending, tail);
241 pending = pending->bi_next;
243 atomic_dec(&fs_info->nr_async_bios);
245 if (atomic_read(&fs_info->nr_async_bios) < limit &&
246 waitqueue_active(&fs_info->async_submit_wait))
247 wake_up(&fs_info->async_submit_wait);
249 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
250 submit_bio(cur->bi_rw, cur);
255 if (need_resched()) {
257 blk_run_backing_dev(bdi, NULL);
264 * we made progress, there is more work to do and the bdi
265 * is now congested. Back off and let other work structs
268 if (pending && bdi_write_congested(bdi) && num_run > 16 &&
269 fs_info->fs_devices->open_devices > 1) {
270 struct io_context *ioc;
272 ioc = current->io_context;
275 * the main goal here is that we don't want to
276 * block if we're going to be able to submit
277 * more requests without blocking.
279 * This code does two great things, it pokes into
280 * the elevator code from a filesystem _and_
281 * it makes assumptions about how batching works.
283 if (ioc && ioc->nr_batch_requests > 0 &&
284 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
286 ioc->last_waited == last_waited)) {
288 * we want to go through our batch of
289 * requests and stop. So, we copy out
290 * the ioc->last_waited time and test
291 * against it before looping
293 last_waited = ioc->last_waited;
294 if (need_resched()) {
296 blk_run_backing_dev(bdi, NULL);
303 spin_lock(&device->io_lock);
304 requeue_list(pending_bios, pending, tail);
305 device->running_pending = 1;
307 spin_unlock(&device->io_lock);
308 btrfs_requeue_work(&device->work);
315 blk_run_backing_dev(bdi, NULL);
322 spin_lock(&device->io_lock);
323 if (device->pending_bios.head || device->pending_sync_bios.head)
325 spin_unlock(&device->io_lock);
328 * IO has already been through a long path to get here. Checksumming,
329 * async helper threads, perhaps compression. We've done a pretty
330 * good job of collecting a batch of IO and should just unplug
331 * the device right away.
333 * This will help anyone who is waiting on the IO, they might have
334 * already unplugged, but managed to do so before the bio they
335 * cared about found its way down here.
337 blk_run_backing_dev(bdi, NULL);
342 static void pending_bios_fn(struct btrfs_work *work)
344 struct btrfs_device *device;
346 device = container_of(work, struct btrfs_device, work);
347 run_scheduled_bios(device);
350 static noinline int device_list_add(const char *path,
351 struct btrfs_super_block *disk_super,
352 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
354 struct btrfs_device *device;
355 struct btrfs_fs_devices *fs_devices;
356 u64 found_transid = btrfs_super_generation(disk_super);
358 fs_devices = find_fsid(disk_super->fsid);
360 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
363 INIT_LIST_HEAD(&fs_devices->devices);
364 INIT_LIST_HEAD(&fs_devices->alloc_list);
365 list_add(&fs_devices->list, &fs_uuids);
366 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
367 fs_devices->latest_devid = devid;
368 fs_devices->latest_trans = found_transid;
371 device = __find_device(&fs_devices->devices, devid,
372 disk_super->dev_item.uuid);
375 if (fs_devices->opened)
378 device = kzalloc(sizeof(*device), GFP_NOFS);
380 /* we can safely leave the fs_devices entry around */
383 device->devid = devid;
384 device->work.func = pending_bios_fn;
385 memcpy(device->uuid, disk_super->dev_item.uuid,
387 device->barriers = 1;
388 spin_lock_init(&device->io_lock);
389 device->name = kstrdup(path, GFP_NOFS);
394 INIT_LIST_HEAD(&device->dev_alloc_list);
395 list_add(&device->dev_list, &fs_devices->devices);
396 device->fs_devices = fs_devices;
397 fs_devices->num_devices++;
400 if (found_transid > fs_devices->latest_trans) {
401 fs_devices->latest_devid = devid;
402 fs_devices->latest_trans = found_transid;
404 *fs_devices_ret = fs_devices;
408 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
410 struct btrfs_fs_devices *fs_devices;
411 struct btrfs_device *device;
412 struct btrfs_device *orig_dev;
414 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
416 return ERR_PTR(-ENOMEM);
418 INIT_LIST_HEAD(&fs_devices->devices);
419 INIT_LIST_HEAD(&fs_devices->alloc_list);
420 INIT_LIST_HEAD(&fs_devices->list);
421 fs_devices->latest_devid = orig->latest_devid;
422 fs_devices->latest_trans = orig->latest_trans;
423 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
425 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
426 device = kzalloc(sizeof(*device), GFP_NOFS);
430 device->name = kstrdup(orig_dev->name, GFP_NOFS);
434 device->devid = orig_dev->devid;
435 device->work.func = pending_bios_fn;
436 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
437 device->barriers = 1;
438 spin_lock_init(&device->io_lock);
439 INIT_LIST_HEAD(&device->dev_list);
440 INIT_LIST_HEAD(&device->dev_alloc_list);
442 list_add(&device->dev_list, &fs_devices->devices);
443 device->fs_devices = fs_devices;
444 fs_devices->num_devices++;
448 free_fs_devices(fs_devices);
449 return ERR_PTR(-ENOMEM);
452 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
454 struct btrfs_device *device, *next;
456 mutex_lock(&uuid_mutex);
458 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
459 if (device->in_fs_metadata)
463 close_bdev_exclusive(device->bdev, device->mode);
465 fs_devices->open_devices--;
467 if (device->writeable) {
468 list_del_init(&device->dev_alloc_list);
469 device->writeable = 0;
470 fs_devices->rw_devices--;
472 list_del_init(&device->dev_list);
473 fs_devices->num_devices--;
478 if (fs_devices->seed) {
479 fs_devices = fs_devices->seed;
483 mutex_unlock(&uuid_mutex);
487 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
489 struct btrfs_device *device;
491 if (--fs_devices->opened > 0)
494 list_for_each_entry(device, &fs_devices->devices, dev_list) {
496 close_bdev_exclusive(device->bdev, device->mode);
497 fs_devices->open_devices--;
499 if (device->writeable) {
500 list_del_init(&device->dev_alloc_list);
501 fs_devices->rw_devices--;
505 device->writeable = 0;
506 device->in_fs_metadata = 0;
508 WARN_ON(fs_devices->open_devices);
509 WARN_ON(fs_devices->rw_devices);
510 fs_devices->opened = 0;
511 fs_devices->seeding = 0;
516 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
518 struct btrfs_fs_devices *seed_devices = NULL;
521 mutex_lock(&uuid_mutex);
522 ret = __btrfs_close_devices(fs_devices);
523 if (!fs_devices->opened) {
524 seed_devices = fs_devices->seed;
525 fs_devices->seed = NULL;
527 mutex_unlock(&uuid_mutex);
529 while (seed_devices) {
530 fs_devices = seed_devices;
531 seed_devices = fs_devices->seed;
532 __btrfs_close_devices(fs_devices);
533 free_fs_devices(fs_devices);
538 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
539 fmode_t flags, void *holder)
541 struct block_device *bdev;
542 struct list_head *head = &fs_devices->devices;
543 struct btrfs_device *device;
544 struct block_device *latest_bdev = NULL;
545 struct buffer_head *bh;
546 struct btrfs_super_block *disk_super;
547 u64 latest_devid = 0;
548 u64 latest_transid = 0;
553 list_for_each_entry(device, head, dev_list) {
559 bdev = open_bdev_exclusive(device->name, flags, holder);
561 printk(KERN_INFO "open %s failed\n", device->name);
564 set_blocksize(bdev, 4096);
566 bh = btrfs_read_dev_super(bdev);
570 disk_super = (struct btrfs_super_block *)bh->b_data;
571 devid = le64_to_cpu(disk_super->dev_item.devid);
572 if (devid != device->devid)
575 if (memcmp(device->uuid, disk_super->dev_item.uuid,
579 device->generation = btrfs_super_generation(disk_super);
580 if (!latest_transid || device->generation > latest_transid) {
581 latest_devid = devid;
582 latest_transid = device->generation;
586 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
587 device->writeable = 0;
589 device->writeable = !bdev_read_only(bdev);
594 device->in_fs_metadata = 0;
595 device->mode = flags;
597 fs_devices->open_devices++;
598 if (device->writeable) {
599 fs_devices->rw_devices++;
600 list_add(&device->dev_alloc_list,
601 &fs_devices->alloc_list);
608 close_bdev_exclusive(bdev, FMODE_READ);
612 if (fs_devices->open_devices == 0) {
616 fs_devices->seeding = seeding;
617 fs_devices->opened = 1;
618 fs_devices->latest_bdev = latest_bdev;
619 fs_devices->latest_devid = latest_devid;
620 fs_devices->latest_trans = latest_transid;
621 fs_devices->total_rw_bytes = 0;
626 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
627 fmode_t flags, void *holder)
631 mutex_lock(&uuid_mutex);
632 if (fs_devices->opened) {
633 fs_devices->opened++;
636 ret = __btrfs_open_devices(fs_devices, flags, holder);
638 mutex_unlock(&uuid_mutex);
642 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
643 struct btrfs_fs_devices **fs_devices_ret)
645 struct btrfs_super_block *disk_super;
646 struct block_device *bdev;
647 struct buffer_head *bh;
652 mutex_lock(&uuid_mutex);
654 bdev = open_bdev_exclusive(path, flags, holder);
661 ret = set_blocksize(bdev, 4096);
664 bh = btrfs_read_dev_super(bdev);
669 disk_super = (struct btrfs_super_block *)bh->b_data;
670 devid = le64_to_cpu(disk_super->dev_item.devid);
671 transid = btrfs_super_generation(disk_super);
672 if (disk_super->label[0])
673 printk(KERN_INFO "device label %s ", disk_super->label);
675 /* FIXME, make a readl uuid parser */
676 printk(KERN_INFO "device fsid %llx-%llx ",
677 *(unsigned long long *)disk_super->fsid,
678 *(unsigned long long *)(disk_super->fsid + 8));
680 printk(KERN_CONT "devid %llu transid %llu %s\n",
681 (unsigned long long)devid, (unsigned long long)transid, path);
682 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
686 close_bdev_exclusive(bdev, flags);
688 mutex_unlock(&uuid_mutex);
693 * this uses a pretty simple search, the expectation is that it is
694 * called very infrequently and that a given device has a small number
697 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
698 struct btrfs_device *device,
699 u64 num_bytes, u64 *start)
701 struct btrfs_key key;
702 struct btrfs_root *root = device->dev_root;
703 struct btrfs_dev_extent *dev_extent = NULL;
704 struct btrfs_path *path;
707 u64 search_start = 0;
708 u64 search_end = device->total_bytes;
712 struct extent_buffer *l;
714 path = btrfs_alloc_path();
720 /* FIXME use last free of some kind */
722 /* we don't want to overwrite the superblock on the drive,
723 * so we make sure to start at an offset of at least 1MB
725 search_start = max((u64)1024 * 1024, search_start);
727 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
728 search_start = max(root->fs_info->alloc_start, search_start);
730 key.objectid = device->devid;
731 key.offset = search_start;
732 key.type = BTRFS_DEV_EXTENT_KEY;
733 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
736 ret = btrfs_previous_item(root, path, 0, key.type);
740 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
743 slot = path->slots[0];
744 if (slot >= btrfs_header_nritems(l)) {
745 ret = btrfs_next_leaf(root, path);
752 if (search_start >= search_end) {
756 *start = search_start;
760 *start = last_byte > search_start ?
761 last_byte : search_start;
762 if (search_end <= *start) {
768 btrfs_item_key_to_cpu(l, &key, slot);
770 if (key.objectid < device->devid)
773 if (key.objectid > device->devid)
776 if (key.offset >= search_start && key.offset > last_byte &&
778 if (last_byte < search_start)
779 last_byte = search_start;
780 hole_size = key.offset - last_byte;
781 if (key.offset > last_byte &&
782 hole_size >= num_bytes) {
787 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
791 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
792 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
798 /* we have to make sure we didn't find an extent that has already
799 * been allocated by the map tree or the original allocation
801 BUG_ON(*start < search_start);
803 if (*start + num_bytes > search_end) {
807 /* check for pending inserts here */
811 btrfs_free_path(path);
815 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
816 struct btrfs_device *device,
820 struct btrfs_path *path;
821 struct btrfs_root *root = device->dev_root;
822 struct btrfs_key key;
823 struct btrfs_key found_key;
824 struct extent_buffer *leaf = NULL;
825 struct btrfs_dev_extent *extent = NULL;
827 path = btrfs_alloc_path();
831 key.objectid = device->devid;
833 key.type = BTRFS_DEV_EXTENT_KEY;
835 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
837 ret = btrfs_previous_item(root, path, key.objectid,
838 BTRFS_DEV_EXTENT_KEY);
840 leaf = path->nodes[0];
841 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
842 extent = btrfs_item_ptr(leaf, path->slots[0],
843 struct btrfs_dev_extent);
844 BUG_ON(found_key.offset > start || found_key.offset +
845 btrfs_dev_extent_length(leaf, extent) < start);
847 } else if (ret == 0) {
848 leaf = path->nodes[0];
849 extent = btrfs_item_ptr(leaf, path->slots[0],
850 struct btrfs_dev_extent);
854 if (device->bytes_used > 0)
855 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
856 ret = btrfs_del_item(trans, root, path);
859 btrfs_free_path(path);
863 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
864 struct btrfs_device *device,
865 u64 chunk_tree, u64 chunk_objectid,
866 u64 chunk_offset, u64 start, u64 num_bytes)
869 struct btrfs_path *path;
870 struct btrfs_root *root = device->dev_root;
871 struct btrfs_dev_extent *extent;
872 struct extent_buffer *leaf;
873 struct btrfs_key key;
875 WARN_ON(!device->in_fs_metadata);
876 path = btrfs_alloc_path();
880 key.objectid = device->devid;
882 key.type = BTRFS_DEV_EXTENT_KEY;
883 ret = btrfs_insert_empty_item(trans, root, path, &key,
887 leaf = path->nodes[0];
888 extent = btrfs_item_ptr(leaf, path->slots[0],
889 struct btrfs_dev_extent);
890 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
891 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
892 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
894 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
895 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
898 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
899 btrfs_mark_buffer_dirty(leaf);
900 btrfs_free_path(path);
904 static noinline int find_next_chunk(struct btrfs_root *root,
905 u64 objectid, u64 *offset)
907 struct btrfs_path *path;
909 struct btrfs_key key;
910 struct btrfs_chunk *chunk;
911 struct btrfs_key found_key;
913 path = btrfs_alloc_path();
916 key.objectid = objectid;
917 key.offset = (u64)-1;
918 key.type = BTRFS_CHUNK_ITEM_KEY;
920 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
926 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
930 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
932 if (found_key.objectid != objectid)
935 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
937 *offset = found_key.offset +
938 btrfs_chunk_length(path->nodes[0], chunk);
943 btrfs_free_path(path);
947 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
950 struct btrfs_key key;
951 struct btrfs_key found_key;
952 struct btrfs_path *path;
954 root = root->fs_info->chunk_root;
956 path = btrfs_alloc_path();
960 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
961 key.type = BTRFS_DEV_ITEM_KEY;
962 key.offset = (u64)-1;
964 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
970 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
975 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
977 *objectid = found_key.offset + 1;
981 btrfs_free_path(path);
986 * the device information is stored in the chunk root
987 * the btrfs_device struct should be fully filled in
989 int btrfs_add_device(struct btrfs_trans_handle *trans,
990 struct btrfs_root *root,
991 struct btrfs_device *device)
994 struct btrfs_path *path;
995 struct btrfs_dev_item *dev_item;
996 struct extent_buffer *leaf;
997 struct btrfs_key key;
1000 root = root->fs_info->chunk_root;
1002 path = btrfs_alloc_path();
1006 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1007 key.type = BTRFS_DEV_ITEM_KEY;
1008 key.offset = device->devid;
1010 ret = btrfs_insert_empty_item(trans, root, path, &key,
1015 leaf = path->nodes[0];
1016 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1018 btrfs_set_device_id(leaf, dev_item, device->devid);
1019 btrfs_set_device_generation(leaf, dev_item, 0);
1020 btrfs_set_device_type(leaf, dev_item, device->type);
1021 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1022 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1023 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1024 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1025 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1026 btrfs_set_device_group(leaf, dev_item, 0);
1027 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1028 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1029 btrfs_set_device_start_offset(leaf, dev_item, 0);
1031 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1032 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1033 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1034 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1035 btrfs_mark_buffer_dirty(leaf);
1039 btrfs_free_path(path);
1043 static int btrfs_rm_dev_item(struct btrfs_root *root,
1044 struct btrfs_device *device)
1047 struct btrfs_path *path;
1048 struct btrfs_key key;
1049 struct btrfs_trans_handle *trans;
1051 root = root->fs_info->chunk_root;
1053 path = btrfs_alloc_path();
1057 trans = btrfs_start_transaction(root, 1);
1058 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1059 key.type = BTRFS_DEV_ITEM_KEY;
1060 key.offset = device->devid;
1063 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1072 ret = btrfs_del_item(trans, root, path);
1076 btrfs_free_path(path);
1077 unlock_chunks(root);
1078 btrfs_commit_transaction(trans, root);
1082 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1084 struct btrfs_device *device;
1085 struct btrfs_device *next_device;
1086 struct block_device *bdev;
1087 struct buffer_head *bh = NULL;
1088 struct btrfs_super_block *disk_super;
1095 mutex_lock(&uuid_mutex);
1096 mutex_lock(&root->fs_info->volume_mutex);
1098 all_avail = root->fs_info->avail_data_alloc_bits |
1099 root->fs_info->avail_system_alloc_bits |
1100 root->fs_info->avail_metadata_alloc_bits;
1102 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1103 root->fs_info->fs_devices->rw_devices <= 4) {
1104 printk(KERN_ERR "btrfs: unable to go below four devices "
1110 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1111 root->fs_info->fs_devices->rw_devices <= 2) {
1112 printk(KERN_ERR "btrfs: unable to go below two "
1113 "devices on raid1\n");
1118 if (strcmp(device_path, "missing") == 0) {
1119 struct list_head *devices;
1120 struct btrfs_device *tmp;
1123 devices = &root->fs_info->fs_devices->devices;
1124 list_for_each_entry(tmp, devices, dev_list) {
1125 if (tmp->in_fs_metadata && !tmp->bdev) {
1134 printk(KERN_ERR "btrfs: no missing devices found to "
1139 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1140 root->fs_info->bdev_holder);
1142 ret = PTR_ERR(bdev);
1146 set_blocksize(bdev, 4096);
1147 bh = btrfs_read_dev_super(bdev);
1152 disk_super = (struct btrfs_super_block *)bh->b_data;
1153 devid = le64_to_cpu(disk_super->dev_item.devid);
1154 dev_uuid = disk_super->dev_item.uuid;
1155 device = btrfs_find_device(root, devid, dev_uuid,
1163 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1164 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1170 if (device->writeable) {
1171 list_del_init(&device->dev_alloc_list);
1172 root->fs_info->fs_devices->rw_devices--;
1175 ret = btrfs_shrink_device(device, 0);
1179 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1183 device->in_fs_metadata = 0;
1184 list_del_init(&device->dev_list);
1185 device->fs_devices->num_devices--;
1187 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1188 struct btrfs_device, dev_list);
1189 if (device->bdev == root->fs_info->sb->s_bdev)
1190 root->fs_info->sb->s_bdev = next_device->bdev;
1191 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1192 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1195 close_bdev_exclusive(device->bdev, device->mode);
1196 device->bdev = NULL;
1197 device->fs_devices->open_devices--;
1200 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1201 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1203 if (device->fs_devices->open_devices == 0) {
1204 struct btrfs_fs_devices *fs_devices;
1205 fs_devices = root->fs_info->fs_devices;
1206 while (fs_devices) {
1207 if (fs_devices->seed == device->fs_devices)
1209 fs_devices = fs_devices->seed;
1211 fs_devices->seed = device->fs_devices->seed;
1212 device->fs_devices->seed = NULL;
1213 __btrfs_close_devices(device->fs_devices);
1214 free_fs_devices(device->fs_devices);
1218 * at this point, the device is zero sized. We want to
1219 * remove it from the devices list and zero out the old super
1221 if (device->writeable) {
1222 /* make sure this device isn't detected as part of
1225 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1226 set_buffer_dirty(bh);
1227 sync_dirty_buffer(bh);
1230 kfree(device->name);
1238 close_bdev_exclusive(bdev, FMODE_READ);
1240 mutex_unlock(&root->fs_info->volume_mutex);
1241 mutex_unlock(&uuid_mutex);
1246 * does all the dirty work required for changing file system's UUID.
1248 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1249 struct btrfs_root *root)
1251 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1252 struct btrfs_fs_devices *old_devices;
1253 struct btrfs_fs_devices *seed_devices;
1254 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1255 struct btrfs_device *device;
1258 BUG_ON(!mutex_is_locked(&uuid_mutex));
1259 if (!fs_devices->seeding)
1262 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1266 old_devices = clone_fs_devices(fs_devices);
1267 if (IS_ERR(old_devices)) {
1268 kfree(seed_devices);
1269 return PTR_ERR(old_devices);
1272 list_add(&old_devices->list, &fs_uuids);
1274 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1275 seed_devices->opened = 1;
1276 INIT_LIST_HEAD(&seed_devices->devices);
1277 INIT_LIST_HEAD(&seed_devices->alloc_list);
1278 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1279 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1280 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1281 device->fs_devices = seed_devices;
1284 fs_devices->seeding = 0;
1285 fs_devices->num_devices = 0;
1286 fs_devices->open_devices = 0;
1287 fs_devices->seed = seed_devices;
1289 generate_random_uuid(fs_devices->fsid);
1290 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1291 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1292 super_flags = btrfs_super_flags(disk_super) &
1293 ~BTRFS_SUPER_FLAG_SEEDING;
1294 btrfs_set_super_flags(disk_super, super_flags);
1300 * strore the expected generation for seed devices in device items.
1302 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1303 struct btrfs_root *root)
1305 struct btrfs_path *path;
1306 struct extent_buffer *leaf;
1307 struct btrfs_dev_item *dev_item;
1308 struct btrfs_device *device;
1309 struct btrfs_key key;
1310 u8 fs_uuid[BTRFS_UUID_SIZE];
1311 u8 dev_uuid[BTRFS_UUID_SIZE];
1315 path = btrfs_alloc_path();
1319 root = root->fs_info->chunk_root;
1320 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1322 key.type = BTRFS_DEV_ITEM_KEY;
1325 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1329 leaf = path->nodes[0];
1331 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1332 ret = btrfs_next_leaf(root, path);
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 btrfs_release_path(root, path);
1343 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1344 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1345 key.type != BTRFS_DEV_ITEM_KEY)
1348 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1349 struct btrfs_dev_item);
1350 devid = btrfs_device_id(leaf, dev_item);
1351 read_extent_buffer(leaf, dev_uuid,
1352 (unsigned long)btrfs_device_uuid(dev_item),
1354 read_extent_buffer(leaf, fs_uuid,
1355 (unsigned long)btrfs_device_fsid(dev_item),
1357 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1360 if (device->fs_devices->seeding) {
1361 btrfs_set_device_generation(leaf, dev_item,
1362 device->generation);
1363 btrfs_mark_buffer_dirty(leaf);
1371 btrfs_free_path(path);
1375 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1377 struct btrfs_trans_handle *trans;
1378 struct btrfs_device *device;
1379 struct block_device *bdev;
1380 struct list_head *devices;
1381 struct super_block *sb = root->fs_info->sb;
1383 int seeding_dev = 0;
1386 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1389 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1393 if (root->fs_info->fs_devices->seeding) {
1395 down_write(&sb->s_umount);
1396 mutex_lock(&uuid_mutex);
1399 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1400 mutex_lock(&root->fs_info->volume_mutex);
1402 devices = &root->fs_info->fs_devices->devices;
1403 list_for_each_entry(device, devices, dev_list) {
1404 if (device->bdev == bdev) {
1410 device = kzalloc(sizeof(*device), GFP_NOFS);
1412 /* we can safely leave the fs_devices entry around */
1417 device->name = kstrdup(device_path, GFP_NOFS);
1418 if (!device->name) {
1424 ret = find_next_devid(root, &device->devid);
1430 trans = btrfs_start_transaction(root, 1);
1433 device->barriers = 1;
1434 device->writeable = 1;
1435 device->work.func = pending_bios_fn;
1436 generate_random_uuid(device->uuid);
1437 spin_lock_init(&device->io_lock);
1438 device->generation = trans->transid;
1439 device->io_width = root->sectorsize;
1440 device->io_align = root->sectorsize;
1441 device->sector_size = root->sectorsize;
1442 device->total_bytes = i_size_read(bdev->bd_inode);
1443 device->disk_total_bytes = device->total_bytes;
1444 device->dev_root = root->fs_info->dev_root;
1445 device->bdev = bdev;
1446 device->in_fs_metadata = 1;
1448 set_blocksize(device->bdev, 4096);
1451 sb->s_flags &= ~MS_RDONLY;
1452 ret = btrfs_prepare_sprout(trans, root);
1456 device->fs_devices = root->fs_info->fs_devices;
1457 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1458 list_add(&device->dev_alloc_list,
1459 &root->fs_info->fs_devices->alloc_list);
1460 root->fs_info->fs_devices->num_devices++;
1461 root->fs_info->fs_devices->open_devices++;
1462 root->fs_info->fs_devices->rw_devices++;
1463 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1465 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1466 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1467 total_bytes + device->total_bytes);
1469 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1470 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1474 ret = init_first_rw_device(trans, root, device);
1476 ret = btrfs_finish_sprout(trans, root);
1479 ret = btrfs_add_device(trans, root, device);
1483 * we've got more storage, clear any full flags on the space
1486 btrfs_clear_space_info_full(root->fs_info);
1488 unlock_chunks(root);
1489 btrfs_commit_transaction(trans, root);
1492 mutex_unlock(&uuid_mutex);
1493 up_write(&sb->s_umount);
1495 ret = btrfs_relocate_sys_chunks(root);
1499 mutex_unlock(&root->fs_info->volume_mutex);
1502 close_bdev_exclusive(bdev, 0);
1504 mutex_unlock(&uuid_mutex);
1505 up_write(&sb->s_umount);
1510 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1511 struct btrfs_device *device)
1514 struct btrfs_path *path;
1515 struct btrfs_root *root;
1516 struct btrfs_dev_item *dev_item;
1517 struct extent_buffer *leaf;
1518 struct btrfs_key key;
1520 root = device->dev_root->fs_info->chunk_root;
1522 path = btrfs_alloc_path();
1526 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1527 key.type = BTRFS_DEV_ITEM_KEY;
1528 key.offset = device->devid;
1530 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1539 leaf = path->nodes[0];
1540 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1542 btrfs_set_device_id(leaf, dev_item, device->devid);
1543 btrfs_set_device_type(leaf, dev_item, device->type);
1544 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1545 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1546 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1547 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1548 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1549 btrfs_mark_buffer_dirty(leaf);
1552 btrfs_free_path(path);
1556 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1557 struct btrfs_device *device, u64 new_size)
1559 struct btrfs_super_block *super_copy =
1560 &device->dev_root->fs_info->super_copy;
1561 u64 old_total = btrfs_super_total_bytes(super_copy);
1562 u64 diff = new_size - device->total_bytes;
1564 if (!device->writeable)
1566 if (new_size <= device->total_bytes)
1569 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1570 device->fs_devices->total_rw_bytes += diff;
1572 device->total_bytes = new_size;
1573 btrfs_clear_space_info_full(device->dev_root->fs_info);
1575 return btrfs_update_device(trans, device);
1578 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1579 struct btrfs_device *device, u64 new_size)
1582 lock_chunks(device->dev_root);
1583 ret = __btrfs_grow_device(trans, device, new_size);
1584 unlock_chunks(device->dev_root);
1588 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1589 struct btrfs_root *root,
1590 u64 chunk_tree, u64 chunk_objectid,
1594 struct btrfs_path *path;
1595 struct btrfs_key key;
1597 root = root->fs_info->chunk_root;
1598 path = btrfs_alloc_path();
1602 key.objectid = chunk_objectid;
1603 key.offset = chunk_offset;
1604 key.type = BTRFS_CHUNK_ITEM_KEY;
1606 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1609 ret = btrfs_del_item(trans, root, path);
1612 btrfs_free_path(path);
1616 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1619 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1620 struct btrfs_disk_key *disk_key;
1621 struct btrfs_chunk *chunk;
1628 struct btrfs_key key;
1630 array_size = btrfs_super_sys_array_size(super_copy);
1632 ptr = super_copy->sys_chunk_array;
1635 while (cur < array_size) {
1636 disk_key = (struct btrfs_disk_key *)ptr;
1637 btrfs_disk_key_to_cpu(&key, disk_key);
1639 len = sizeof(*disk_key);
1641 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1642 chunk = (struct btrfs_chunk *)(ptr + len);
1643 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1644 len += btrfs_chunk_item_size(num_stripes);
1649 if (key.objectid == chunk_objectid &&
1650 key.offset == chunk_offset) {
1651 memmove(ptr, ptr + len, array_size - (cur + len));
1653 btrfs_set_super_sys_array_size(super_copy, array_size);
1662 static int btrfs_relocate_chunk(struct btrfs_root *root,
1663 u64 chunk_tree, u64 chunk_objectid,
1666 struct extent_map_tree *em_tree;
1667 struct btrfs_root *extent_root;
1668 struct btrfs_trans_handle *trans;
1669 struct extent_map *em;
1670 struct map_lookup *map;
1674 root = root->fs_info->chunk_root;
1675 extent_root = root->fs_info->extent_root;
1676 em_tree = &root->fs_info->mapping_tree.map_tree;
1678 /* step one, relocate all the extents inside this chunk */
1679 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1682 trans = btrfs_start_transaction(root, 1);
1688 * step two, delete the device extents and the
1689 * chunk tree entries
1691 spin_lock(&em_tree->lock);
1692 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1693 spin_unlock(&em_tree->lock);
1695 BUG_ON(em->start > chunk_offset ||
1696 em->start + em->len < chunk_offset);
1697 map = (struct map_lookup *)em->bdev;
1699 for (i = 0; i < map->num_stripes; i++) {
1700 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1701 map->stripes[i].physical);
1704 if (map->stripes[i].dev) {
1705 ret = btrfs_update_device(trans, map->stripes[i].dev);
1709 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1714 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1715 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1719 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1722 spin_lock(&em_tree->lock);
1723 remove_extent_mapping(em_tree, em);
1724 spin_unlock(&em_tree->lock);
1729 /* once for the tree */
1730 free_extent_map(em);
1732 free_extent_map(em);
1734 unlock_chunks(root);
1735 btrfs_end_transaction(trans, root);
1739 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1741 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1742 struct btrfs_path *path;
1743 struct extent_buffer *leaf;
1744 struct btrfs_chunk *chunk;
1745 struct btrfs_key key;
1746 struct btrfs_key found_key;
1747 u64 chunk_tree = chunk_root->root_key.objectid;
1751 path = btrfs_alloc_path();
1755 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1756 key.offset = (u64)-1;
1757 key.type = BTRFS_CHUNK_ITEM_KEY;
1760 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1765 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1772 leaf = path->nodes[0];
1773 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1775 chunk = btrfs_item_ptr(leaf, path->slots[0],
1776 struct btrfs_chunk);
1777 chunk_type = btrfs_chunk_type(leaf, chunk);
1778 btrfs_release_path(chunk_root, path);
1780 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1781 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1787 if (found_key.offset == 0)
1789 key.offset = found_key.offset - 1;
1793 btrfs_free_path(path);
1797 static u64 div_factor(u64 num, int factor)
1806 int btrfs_balance(struct btrfs_root *dev_root)
1809 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1810 struct btrfs_device *device;
1813 struct btrfs_path *path;
1814 struct btrfs_key key;
1815 struct btrfs_chunk *chunk;
1816 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1817 struct btrfs_trans_handle *trans;
1818 struct btrfs_key found_key;
1820 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1823 mutex_lock(&dev_root->fs_info->volume_mutex);
1824 dev_root = dev_root->fs_info->dev_root;
1826 /* step one make some room on all the devices */
1827 list_for_each_entry(device, devices, dev_list) {
1828 old_size = device->total_bytes;
1829 size_to_free = div_factor(old_size, 1);
1830 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1831 if (!device->writeable ||
1832 device->total_bytes - device->bytes_used > size_to_free)
1835 ret = btrfs_shrink_device(device, old_size - size_to_free);
1838 trans = btrfs_start_transaction(dev_root, 1);
1841 ret = btrfs_grow_device(trans, device, old_size);
1844 btrfs_end_transaction(trans, dev_root);
1847 /* step two, relocate all the chunks */
1848 path = btrfs_alloc_path();
1851 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1852 key.offset = (u64)-1;
1853 key.type = BTRFS_CHUNK_ITEM_KEY;
1856 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1861 * this shouldn't happen, it means the last relocate
1867 ret = btrfs_previous_item(chunk_root, path, 0,
1868 BTRFS_CHUNK_ITEM_KEY);
1872 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1874 if (found_key.objectid != key.objectid)
1877 chunk = btrfs_item_ptr(path->nodes[0],
1879 struct btrfs_chunk);
1880 key.offset = found_key.offset;
1881 /* chunk zero is special */
1882 if (key.offset == 0)
1885 btrfs_release_path(chunk_root, path);
1886 ret = btrfs_relocate_chunk(chunk_root,
1887 chunk_root->root_key.objectid,
1894 btrfs_free_path(path);
1895 mutex_unlock(&dev_root->fs_info->volume_mutex);
1900 * shrinking a device means finding all of the device extents past
1901 * the new size, and then following the back refs to the chunks.
1902 * The chunk relocation code actually frees the device extent
1904 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1906 struct btrfs_trans_handle *trans;
1907 struct btrfs_root *root = device->dev_root;
1908 struct btrfs_dev_extent *dev_extent = NULL;
1909 struct btrfs_path *path;
1916 struct extent_buffer *l;
1917 struct btrfs_key key;
1918 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1919 u64 old_total = btrfs_super_total_bytes(super_copy);
1920 u64 diff = device->total_bytes - new_size;
1922 if (new_size >= device->total_bytes)
1925 path = btrfs_alloc_path();
1929 trans = btrfs_start_transaction(root, 1);
1939 device->total_bytes = new_size;
1940 if (device->writeable)
1941 device->fs_devices->total_rw_bytes -= diff;
1942 unlock_chunks(root);
1943 btrfs_end_transaction(trans, root);
1945 key.objectid = device->devid;
1946 key.offset = (u64)-1;
1947 key.type = BTRFS_DEV_EXTENT_KEY;
1950 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1954 ret = btrfs_previous_item(root, path, 0, key.type);
1963 slot = path->slots[0];
1964 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1966 if (key.objectid != device->devid)
1969 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1970 length = btrfs_dev_extent_length(l, dev_extent);
1972 if (key.offset + length <= new_size)
1975 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1976 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1977 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1978 btrfs_release_path(root, path);
1980 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1986 /* Shrinking succeeded, else we would be at "done". */
1987 trans = btrfs_start_transaction(root, 1);
1994 device->disk_total_bytes = new_size;
1995 /* Now btrfs_update_device() will change the on-disk size. */
1996 ret = btrfs_update_device(trans, device);
1998 unlock_chunks(root);
1999 btrfs_end_transaction(trans, root);
2002 WARN_ON(diff > old_total);
2003 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2004 unlock_chunks(root);
2005 btrfs_end_transaction(trans, root);
2007 btrfs_free_path(path);
2011 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2012 struct btrfs_root *root,
2013 struct btrfs_key *key,
2014 struct btrfs_chunk *chunk, int item_size)
2016 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2017 struct btrfs_disk_key disk_key;
2021 array_size = btrfs_super_sys_array_size(super_copy);
2022 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2025 ptr = super_copy->sys_chunk_array + array_size;
2026 btrfs_cpu_key_to_disk(&disk_key, key);
2027 memcpy(ptr, &disk_key, sizeof(disk_key));
2028 ptr += sizeof(disk_key);
2029 memcpy(ptr, chunk, item_size);
2030 item_size += sizeof(disk_key);
2031 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2035 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2036 int num_stripes, int sub_stripes)
2038 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2040 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2041 return calc_size * (num_stripes / sub_stripes);
2043 return calc_size * num_stripes;
2046 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2047 struct btrfs_root *extent_root,
2048 struct map_lookup **map_ret,
2049 u64 *num_bytes, u64 *stripe_size,
2050 u64 start, u64 type)
2052 struct btrfs_fs_info *info = extent_root->fs_info;
2053 struct btrfs_device *device = NULL;
2054 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2055 struct list_head *cur;
2056 struct map_lookup *map = NULL;
2057 struct extent_map_tree *em_tree;
2058 struct extent_map *em;
2059 struct list_head private_devs;
2060 int min_stripe_size = 1 * 1024 * 1024;
2061 u64 calc_size = 1024 * 1024 * 1024;
2062 u64 max_chunk_size = calc_size;
2067 int num_stripes = 1;
2068 int min_stripes = 1;
2069 int sub_stripes = 0;
2073 int stripe_len = 64 * 1024;
2075 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2076 (type & BTRFS_BLOCK_GROUP_DUP)) {
2078 type &= ~BTRFS_BLOCK_GROUP_DUP;
2080 if (list_empty(&fs_devices->alloc_list))
2083 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2084 num_stripes = fs_devices->rw_devices;
2087 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2091 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2092 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2093 if (num_stripes < 2)
2097 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2098 num_stripes = fs_devices->rw_devices;
2099 if (num_stripes < 4)
2101 num_stripes &= ~(u32)1;
2106 if (type & BTRFS_BLOCK_GROUP_DATA) {
2107 max_chunk_size = 10 * calc_size;
2108 min_stripe_size = 64 * 1024 * 1024;
2109 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2110 max_chunk_size = 4 * calc_size;
2111 min_stripe_size = 32 * 1024 * 1024;
2112 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2113 calc_size = 8 * 1024 * 1024;
2114 max_chunk_size = calc_size * 2;
2115 min_stripe_size = 1 * 1024 * 1024;
2118 /* we don't want a chunk larger than 10% of writeable space */
2119 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2123 if (!map || map->num_stripes != num_stripes) {
2125 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2128 map->num_stripes = num_stripes;
2131 if (calc_size * num_stripes > max_chunk_size) {
2132 calc_size = max_chunk_size;
2133 do_div(calc_size, num_stripes);
2134 do_div(calc_size, stripe_len);
2135 calc_size *= stripe_len;
2137 /* we don't want tiny stripes */
2138 calc_size = max_t(u64, min_stripe_size, calc_size);
2140 do_div(calc_size, stripe_len);
2141 calc_size *= stripe_len;
2143 cur = fs_devices->alloc_list.next;
2146 if (type & BTRFS_BLOCK_GROUP_DUP)
2147 min_free = calc_size * 2;
2149 min_free = calc_size;
2152 * we add 1MB because we never use the first 1MB of the device, unless
2153 * we've looped, then we are likely allocating the maximum amount of
2154 * space left already
2157 min_free += 1024 * 1024;
2159 INIT_LIST_HEAD(&private_devs);
2160 while (index < num_stripes) {
2161 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2162 BUG_ON(!device->writeable);
2163 if (device->total_bytes > device->bytes_used)
2164 avail = device->total_bytes - device->bytes_used;
2169 if (device->in_fs_metadata && avail >= min_free) {
2170 ret = find_free_dev_extent(trans, device,
2171 min_free, &dev_offset);
2173 list_move_tail(&device->dev_alloc_list,
2175 map->stripes[index].dev = device;
2176 map->stripes[index].physical = dev_offset;
2178 if (type & BTRFS_BLOCK_GROUP_DUP) {
2179 map->stripes[index].dev = device;
2180 map->stripes[index].physical =
2181 dev_offset + calc_size;
2185 } else if (device->in_fs_metadata && avail > max_avail)
2187 if (cur == &fs_devices->alloc_list)
2190 list_splice(&private_devs, &fs_devices->alloc_list);
2191 if (index < num_stripes) {
2192 if (index >= min_stripes) {
2193 num_stripes = index;
2194 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2195 num_stripes /= sub_stripes;
2196 num_stripes *= sub_stripes;
2201 if (!looped && max_avail > 0) {
2203 calc_size = max_avail;
2209 map->sector_size = extent_root->sectorsize;
2210 map->stripe_len = stripe_len;
2211 map->io_align = stripe_len;
2212 map->io_width = stripe_len;
2214 map->num_stripes = num_stripes;
2215 map->sub_stripes = sub_stripes;
2218 *stripe_size = calc_size;
2219 *num_bytes = chunk_bytes_by_type(type, calc_size,
2220 num_stripes, sub_stripes);
2222 em = alloc_extent_map(GFP_NOFS);
2227 em->bdev = (struct block_device *)map;
2229 em->len = *num_bytes;
2230 em->block_start = 0;
2231 em->block_len = em->len;
2233 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2234 spin_lock(&em_tree->lock);
2235 ret = add_extent_mapping(em_tree, em);
2236 spin_unlock(&em_tree->lock);
2238 free_extent_map(em);
2240 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2241 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2246 while (index < map->num_stripes) {
2247 device = map->stripes[index].dev;
2248 dev_offset = map->stripes[index].physical;
2250 ret = btrfs_alloc_dev_extent(trans, device,
2251 info->chunk_root->root_key.objectid,
2252 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2253 start, dev_offset, calc_size);
2261 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2262 struct btrfs_root *extent_root,
2263 struct map_lookup *map, u64 chunk_offset,
2264 u64 chunk_size, u64 stripe_size)
2267 struct btrfs_key key;
2268 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2269 struct btrfs_device *device;
2270 struct btrfs_chunk *chunk;
2271 struct btrfs_stripe *stripe;
2272 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2276 chunk = kzalloc(item_size, GFP_NOFS);
2281 while (index < map->num_stripes) {
2282 device = map->stripes[index].dev;
2283 device->bytes_used += stripe_size;
2284 ret = btrfs_update_device(trans, device);
2290 stripe = &chunk->stripe;
2291 while (index < map->num_stripes) {
2292 device = map->stripes[index].dev;
2293 dev_offset = map->stripes[index].physical;
2295 btrfs_set_stack_stripe_devid(stripe, device->devid);
2296 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2297 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2302 btrfs_set_stack_chunk_length(chunk, chunk_size);
2303 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2304 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2305 btrfs_set_stack_chunk_type(chunk, map->type);
2306 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2307 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2308 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2309 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2310 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2312 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2313 key.type = BTRFS_CHUNK_ITEM_KEY;
2314 key.offset = chunk_offset;
2316 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2319 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2320 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2329 * Chunk allocation falls into two parts. The first part does works
2330 * that make the new allocated chunk useable, but not do any operation
2331 * that modifies the chunk tree. The second part does the works that
2332 * require modifying the chunk tree. This division is important for the
2333 * bootstrap process of adding storage to a seed btrfs.
2335 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2336 struct btrfs_root *extent_root, u64 type)
2341 struct map_lookup *map;
2342 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2345 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2350 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2351 &stripe_size, chunk_offset, type);
2355 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2356 chunk_size, stripe_size);
2361 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2362 struct btrfs_root *root,
2363 struct btrfs_device *device)
2366 u64 sys_chunk_offset;
2370 u64 sys_stripe_size;
2372 struct map_lookup *map;
2373 struct map_lookup *sys_map;
2374 struct btrfs_fs_info *fs_info = root->fs_info;
2375 struct btrfs_root *extent_root = fs_info->extent_root;
2378 ret = find_next_chunk(fs_info->chunk_root,
2379 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2382 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2383 (fs_info->metadata_alloc_profile &
2384 fs_info->avail_metadata_alloc_bits);
2385 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2387 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2388 &stripe_size, chunk_offset, alloc_profile);
2391 sys_chunk_offset = chunk_offset + chunk_size;
2393 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2394 (fs_info->system_alloc_profile &
2395 fs_info->avail_system_alloc_bits);
2396 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2398 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2399 &sys_chunk_size, &sys_stripe_size,
2400 sys_chunk_offset, alloc_profile);
2403 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2407 * Modifying chunk tree needs allocating new blocks from both
2408 * system block group and metadata block group. So we only can
2409 * do operations require modifying the chunk tree after both
2410 * block groups were created.
2412 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2413 chunk_size, stripe_size);
2416 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2417 sys_chunk_offset, sys_chunk_size,
2423 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2425 struct extent_map *em;
2426 struct map_lookup *map;
2427 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2431 spin_lock(&map_tree->map_tree.lock);
2432 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2433 spin_unlock(&map_tree->map_tree.lock);
2437 map = (struct map_lookup *)em->bdev;
2438 for (i = 0; i < map->num_stripes; i++) {
2439 if (!map->stripes[i].dev->writeable) {
2444 free_extent_map(em);
2448 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2450 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2453 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2455 struct extent_map *em;
2458 spin_lock(&tree->map_tree.lock);
2459 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2461 remove_extent_mapping(&tree->map_tree, em);
2462 spin_unlock(&tree->map_tree.lock);
2467 free_extent_map(em);
2468 /* once for the tree */
2469 free_extent_map(em);
2473 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2475 struct extent_map *em;
2476 struct map_lookup *map;
2477 struct extent_map_tree *em_tree = &map_tree->map_tree;
2480 spin_lock(&em_tree->lock);
2481 em = lookup_extent_mapping(em_tree, logical, len);
2482 spin_unlock(&em_tree->lock);
2485 BUG_ON(em->start > logical || em->start + em->len < logical);
2486 map = (struct map_lookup *)em->bdev;
2487 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2488 ret = map->num_stripes;
2489 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2490 ret = map->sub_stripes;
2493 free_extent_map(em);
2497 static int find_live_mirror(struct map_lookup *map, int first, int num,
2501 if (map->stripes[optimal].dev->bdev)
2503 for (i = first; i < first + num; i++) {
2504 if (map->stripes[i].dev->bdev)
2507 /* we couldn't find one that doesn't fail. Just return something
2508 * and the io error handling code will clean up eventually
2513 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2514 u64 logical, u64 *length,
2515 struct btrfs_multi_bio **multi_ret,
2516 int mirror_num, struct page *unplug_page)
2518 struct extent_map *em;
2519 struct map_lookup *map;
2520 struct extent_map_tree *em_tree = &map_tree->map_tree;
2524 int stripes_allocated = 8;
2525 int stripes_required = 1;
2530 struct btrfs_multi_bio *multi = NULL;
2532 if (multi_ret && !(rw & (1 << BIO_RW)))
2533 stripes_allocated = 1;
2536 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2541 atomic_set(&multi->error, 0);
2544 spin_lock(&em_tree->lock);
2545 em = lookup_extent_mapping(em_tree, logical, *length);
2546 spin_unlock(&em_tree->lock);
2548 if (!em && unplug_page)
2552 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2553 (unsigned long long)logical,
2554 (unsigned long long)*length);
2558 BUG_ON(em->start > logical || em->start + em->len < logical);
2559 map = (struct map_lookup *)em->bdev;
2560 offset = logical - em->start;
2562 if (mirror_num > map->num_stripes)
2565 /* if our multi bio struct is too small, back off and try again */
2566 if (rw & (1 << BIO_RW)) {
2567 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2568 BTRFS_BLOCK_GROUP_DUP)) {
2569 stripes_required = map->num_stripes;
2571 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2572 stripes_required = map->sub_stripes;
2576 if (multi_ret && (rw & (1 << BIO_RW)) &&
2577 stripes_allocated < stripes_required) {
2578 stripes_allocated = map->num_stripes;
2579 free_extent_map(em);
2585 * stripe_nr counts the total number of stripes we have to stride
2586 * to get to this block
2588 do_div(stripe_nr, map->stripe_len);
2590 stripe_offset = stripe_nr * map->stripe_len;
2591 BUG_ON(offset < stripe_offset);
2593 /* stripe_offset is the offset of this block in its stripe*/
2594 stripe_offset = offset - stripe_offset;
2596 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2597 BTRFS_BLOCK_GROUP_RAID10 |
2598 BTRFS_BLOCK_GROUP_DUP)) {
2599 /* we limit the length of each bio to what fits in a stripe */
2600 *length = min_t(u64, em->len - offset,
2601 map->stripe_len - stripe_offset);
2603 *length = em->len - offset;
2606 if (!multi_ret && !unplug_page)
2611 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2612 if (unplug_page || (rw & (1 << BIO_RW)))
2613 num_stripes = map->num_stripes;
2614 else if (mirror_num)
2615 stripe_index = mirror_num - 1;
2617 stripe_index = find_live_mirror(map, 0,
2619 current->pid % map->num_stripes);
2622 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2623 if (rw & (1 << BIO_RW))
2624 num_stripes = map->num_stripes;
2625 else if (mirror_num)
2626 stripe_index = mirror_num - 1;
2628 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2629 int factor = map->num_stripes / map->sub_stripes;
2631 stripe_index = do_div(stripe_nr, factor);
2632 stripe_index *= map->sub_stripes;
2634 if (unplug_page || (rw & (1 << BIO_RW)))
2635 num_stripes = map->sub_stripes;
2636 else if (mirror_num)
2637 stripe_index += mirror_num - 1;
2639 stripe_index = find_live_mirror(map, stripe_index,
2640 map->sub_stripes, stripe_index +
2641 current->pid % map->sub_stripes);
2645 * after this do_div call, stripe_nr is the number of stripes
2646 * on this device we have to walk to find the data, and
2647 * stripe_index is the number of our device in the stripe array
2649 stripe_index = do_div(stripe_nr, map->num_stripes);
2651 BUG_ON(stripe_index >= map->num_stripes);
2653 for (i = 0; i < num_stripes; i++) {
2655 struct btrfs_device *device;
2656 struct backing_dev_info *bdi;
2658 device = map->stripes[stripe_index].dev;
2660 bdi = blk_get_backing_dev_info(device->bdev);
2661 if (bdi->unplug_io_fn)
2662 bdi->unplug_io_fn(bdi, unplug_page);
2665 multi->stripes[i].physical =
2666 map->stripes[stripe_index].physical +
2667 stripe_offset + stripe_nr * map->stripe_len;
2668 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2674 multi->num_stripes = num_stripes;
2675 multi->max_errors = max_errors;
2678 free_extent_map(em);
2682 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2683 u64 logical, u64 *length,
2684 struct btrfs_multi_bio **multi_ret, int mirror_num)
2686 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2690 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2691 u64 chunk_start, u64 physical, u64 devid,
2692 u64 **logical, int *naddrs, int *stripe_len)
2694 struct extent_map_tree *em_tree = &map_tree->map_tree;
2695 struct extent_map *em;
2696 struct map_lookup *map;
2703 spin_lock(&em_tree->lock);
2704 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2705 spin_unlock(&em_tree->lock);
2707 BUG_ON(!em || em->start != chunk_start);
2708 map = (struct map_lookup *)em->bdev;
2711 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2712 do_div(length, map->num_stripes / map->sub_stripes);
2713 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2714 do_div(length, map->num_stripes);
2716 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2719 for (i = 0; i < map->num_stripes; i++) {
2720 if (devid && map->stripes[i].dev->devid != devid)
2722 if (map->stripes[i].physical > physical ||
2723 map->stripes[i].physical + length <= physical)
2726 stripe_nr = physical - map->stripes[i].physical;
2727 do_div(stripe_nr, map->stripe_len);
2729 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2730 stripe_nr = stripe_nr * map->num_stripes + i;
2731 do_div(stripe_nr, map->sub_stripes);
2732 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2733 stripe_nr = stripe_nr * map->num_stripes + i;
2735 bytenr = chunk_start + stripe_nr * map->stripe_len;
2736 WARN_ON(nr >= map->num_stripes);
2737 for (j = 0; j < nr; j++) {
2738 if (buf[j] == bytenr)
2742 WARN_ON(nr >= map->num_stripes);
2747 for (i = 0; i > nr; i++) {
2748 struct btrfs_multi_bio *multi;
2749 struct btrfs_bio_stripe *stripe;
2753 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2754 &length, &multi, 0);
2757 stripe = multi->stripes;
2758 for (j = 0; j < multi->num_stripes; j++) {
2759 if (stripe->physical >= physical &&
2760 physical < stripe->physical + length)
2763 BUG_ON(j >= multi->num_stripes);
2769 *stripe_len = map->stripe_len;
2771 free_extent_map(em);
2775 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2776 u64 logical, struct page *page)
2778 u64 length = PAGE_CACHE_SIZE;
2779 return __btrfs_map_block(map_tree, READ, logical, &length,
2783 static void end_bio_multi_stripe(struct bio *bio, int err)
2785 struct btrfs_multi_bio *multi = bio->bi_private;
2786 int is_orig_bio = 0;
2789 atomic_inc(&multi->error);
2791 if (bio == multi->orig_bio)
2794 if (atomic_dec_and_test(&multi->stripes_pending)) {
2797 bio = multi->orig_bio;
2799 bio->bi_private = multi->private;
2800 bio->bi_end_io = multi->end_io;
2801 /* only send an error to the higher layers if it is
2802 * beyond the tolerance of the multi-bio
2804 if (atomic_read(&multi->error) > multi->max_errors) {
2808 * this bio is actually up to date, we didn't
2809 * go over the max number of errors
2811 set_bit(BIO_UPTODATE, &bio->bi_flags);
2816 bio_endio(bio, err);
2817 } else if (!is_orig_bio) {
2822 struct async_sched {
2825 struct btrfs_fs_info *info;
2826 struct btrfs_work work;
2830 * see run_scheduled_bios for a description of why bios are collected for
2833 * This will add one bio to the pending list for a device and make sure
2834 * the work struct is scheduled.
2836 static noinline int schedule_bio(struct btrfs_root *root,
2837 struct btrfs_device *device,
2838 int rw, struct bio *bio)
2840 int should_queue = 1;
2841 struct btrfs_pending_bios *pending_bios;
2843 /* don't bother with additional async steps for reads, right now */
2844 if (!(rw & (1 << BIO_RW))) {
2846 submit_bio(rw, bio);
2852 * nr_async_bios allows us to reliably return congestion to the
2853 * higher layers. Otherwise, the async bio makes it appear we have
2854 * made progress against dirty pages when we've really just put it
2855 * on a queue for later
2857 atomic_inc(&root->fs_info->nr_async_bios);
2858 WARN_ON(bio->bi_next);
2859 bio->bi_next = NULL;
2862 spin_lock(&device->io_lock);
2864 pending_bios = &device->pending_sync_bios;
2866 pending_bios = &device->pending_bios;
2868 if (pending_bios->tail)
2869 pending_bios->tail->bi_next = bio;
2871 pending_bios->tail = bio;
2872 if (!pending_bios->head)
2873 pending_bios->head = bio;
2874 if (device->running_pending)
2877 spin_unlock(&device->io_lock);
2880 btrfs_queue_worker(&root->fs_info->submit_workers,
2885 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2886 int mirror_num, int async_submit)
2888 struct btrfs_mapping_tree *map_tree;
2889 struct btrfs_device *dev;
2890 struct bio *first_bio = bio;
2891 u64 logical = (u64)bio->bi_sector << 9;
2894 struct btrfs_multi_bio *multi = NULL;
2899 length = bio->bi_size;
2900 map_tree = &root->fs_info->mapping_tree;
2901 map_length = length;
2903 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2907 total_devs = multi->num_stripes;
2908 if (map_length < length) {
2909 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2910 "len %llu\n", (unsigned long long)logical,
2911 (unsigned long long)length,
2912 (unsigned long long)map_length);
2915 multi->end_io = first_bio->bi_end_io;
2916 multi->private = first_bio->bi_private;
2917 multi->orig_bio = first_bio;
2918 atomic_set(&multi->stripes_pending, multi->num_stripes);
2920 while (dev_nr < total_devs) {
2921 if (total_devs > 1) {
2922 if (dev_nr < total_devs - 1) {
2923 bio = bio_clone(first_bio, GFP_NOFS);
2928 bio->bi_private = multi;
2929 bio->bi_end_io = end_bio_multi_stripe;
2931 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2932 dev = multi->stripes[dev_nr].dev;
2933 BUG_ON(rw == WRITE && !dev->writeable);
2934 if (dev && dev->bdev) {
2935 bio->bi_bdev = dev->bdev;
2937 schedule_bio(root, dev, rw, bio);
2939 submit_bio(rw, bio);
2941 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2942 bio->bi_sector = logical >> 9;
2943 bio_endio(bio, -EIO);
2947 if (total_devs == 1)
2952 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2955 struct btrfs_device *device;
2956 struct btrfs_fs_devices *cur_devices;
2958 cur_devices = root->fs_info->fs_devices;
2959 while (cur_devices) {
2961 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2962 device = __find_device(&cur_devices->devices,
2967 cur_devices = cur_devices->seed;
2972 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2973 u64 devid, u8 *dev_uuid)
2975 struct btrfs_device *device;
2976 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2978 device = kzalloc(sizeof(*device), GFP_NOFS);
2981 list_add(&device->dev_list,
2982 &fs_devices->devices);
2983 device->barriers = 1;
2984 device->dev_root = root->fs_info->dev_root;
2985 device->devid = devid;
2986 device->work.func = pending_bios_fn;
2987 device->fs_devices = fs_devices;
2988 fs_devices->num_devices++;
2989 spin_lock_init(&device->io_lock);
2990 INIT_LIST_HEAD(&device->dev_alloc_list);
2991 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2995 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2996 struct extent_buffer *leaf,
2997 struct btrfs_chunk *chunk)
2999 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3000 struct map_lookup *map;
3001 struct extent_map *em;
3005 u8 uuid[BTRFS_UUID_SIZE];
3010 logical = key->offset;
3011 length = btrfs_chunk_length(leaf, chunk);
3013 spin_lock(&map_tree->map_tree.lock);
3014 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3015 spin_unlock(&map_tree->map_tree.lock);
3017 /* already mapped? */
3018 if (em && em->start <= logical && em->start + em->len > logical) {
3019 free_extent_map(em);
3022 free_extent_map(em);
3025 em = alloc_extent_map(GFP_NOFS);
3028 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3029 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3031 free_extent_map(em);
3035 em->bdev = (struct block_device *)map;
3036 em->start = logical;
3038 em->block_start = 0;
3039 em->block_len = em->len;
3041 map->num_stripes = num_stripes;
3042 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3043 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3044 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3045 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3046 map->type = btrfs_chunk_type(leaf, chunk);
3047 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3048 for (i = 0; i < num_stripes; i++) {
3049 map->stripes[i].physical =
3050 btrfs_stripe_offset_nr(leaf, chunk, i);
3051 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3052 read_extent_buffer(leaf, uuid, (unsigned long)
3053 btrfs_stripe_dev_uuid_nr(chunk, i),
3055 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3057 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3059 free_extent_map(em);
3062 if (!map->stripes[i].dev) {
3063 map->stripes[i].dev =
3064 add_missing_dev(root, devid, uuid);
3065 if (!map->stripes[i].dev) {
3067 free_extent_map(em);
3071 map->stripes[i].dev->in_fs_metadata = 1;
3074 spin_lock(&map_tree->map_tree.lock);
3075 ret = add_extent_mapping(&map_tree->map_tree, em);
3076 spin_unlock(&map_tree->map_tree.lock);
3078 free_extent_map(em);
3083 static int fill_device_from_item(struct extent_buffer *leaf,
3084 struct btrfs_dev_item *dev_item,
3085 struct btrfs_device *device)
3089 device->devid = btrfs_device_id(leaf, dev_item);
3090 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3091 device->total_bytes = device->disk_total_bytes;
3092 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3093 device->type = btrfs_device_type(leaf, dev_item);
3094 device->io_align = btrfs_device_io_align(leaf, dev_item);
3095 device->io_width = btrfs_device_io_width(leaf, dev_item);
3096 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3098 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3099 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3104 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3106 struct btrfs_fs_devices *fs_devices;
3109 mutex_lock(&uuid_mutex);
3111 fs_devices = root->fs_info->fs_devices->seed;
3112 while (fs_devices) {
3113 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3117 fs_devices = fs_devices->seed;
3120 fs_devices = find_fsid(fsid);
3126 fs_devices = clone_fs_devices(fs_devices);
3127 if (IS_ERR(fs_devices)) {
3128 ret = PTR_ERR(fs_devices);
3132 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3133 root->fs_info->bdev_holder);
3137 if (!fs_devices->seeding) {
3138 __btrfs_close_devices(fs_devices);
3139 free_fs_devices(fs_devices);
3144 fs_devices->seed = root->fs_info->fs_devices->seed;
3145 root->fs_info->fs_devices->seed = fs_devices;
3147 mutex_unlock(&uuid_mutex);
3151 static int read_one_dev(struct btrfs_root *root,
3152 struct extent_buffer *leaf,
3153 struct btrfs_dev_item *dev_item)
3155 struct btrfs_device *device;
3158 u8 fs_uuid[BTRFS_UUID_SIZE];
3159 u8 dev_uuid[BTRFS_UUID_SIZE];
3161 devid = btrfs_device_id(leaf, dev_item);
3162 read_extent_buffer(leaf, dev_uuid,
3163 (unsigned long)btrfs_device_uuid(dev_item),
3165 read_extent_buffer(leaf, fs_uuid,
3166 (unsigned long)btrfs_device_fsid(dev_item),
3169 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3170 ret = open_seed_devices(root, fs_uuid);
3171 if (ret && !btrfs_test_opt(root, DEGRADED))
3175 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3176 if (!device || !device->bdev) {
3177 if (!btrfs_test_opt(root, DEGRADED))
3181 printk(KERN_WARNING "warning devid %llu missing\n",
3182 (unsigned long long)devid);
3183 device = add_missing_dev(root, devid, dev_uuid);
3189 if (device->fs_devices != root->fs_info->fs_devices) {
3190 BUG_ON(device->writeable);
3191 if (device->generation !=
3192 btrfs_device_generation(leaf, dev_item))
3196 fill_device_from_item(leaf, dev_item, device);
3197 device->dev_root = root->fs_info->dev_root;
3198 device->in_fs_metadata = 1;
3199 if (device->writeable)
3200 device->fs_devices->total_rw_bytes += device->total_bytes;
3205 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3207 struct btrfs_dev_item *dev_item;
3209 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3211 return read_one_dev(root, buf, dev_item);
3214 int btrfs_read_sys_array(struct btrfs_root *root)
3216 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3217 struct extent_buffer *sb;
3218 struct btrfs_disk_key *disk_key;
3219 struct btrfs_chunk *chunk;
3221 unsigned long sb_ptr;
3227 struct btrfs_key key;
3229 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3230 BTRFS_SUPER_INFO_SIZE);
3233 btrfs_set_buffer_uptodate(sb);
3234 btrfs_set_buffer_lockdep_class(sb, 0);
3236 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3237 array_size = btrfs_super_sys_array_size(super_copy);
3239 ptr = super_copy->sys_chunk_array;
3240 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3243 while (cur < array_size) {
3244 disk_key = (struct btrfs_disk_key *)ptr;
3245 btrfs_disk_key_to_cpu(&key, disk_key);
3247 len = sizeof(*disk_key); ptr += len;
3251 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3252 chunk = (struct btrfs_chunk *)sb_ptr;
3253 ret = read_one_chunk(root, &key, sb, chunk);
3256 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3257 len = btrfs_chunk_item_size(num_stripes);
3266 free_extent_buffer(sb);
3270 int btrfs_read_chunk_tree(struct btrfs_root *root)
3272 struct btrfs_path *path;
3273 struct extent_buffer *leaf;
3274 struct btrfs_key key;
3275 struct btrfs_key found_key;
3279 root = root->fs_info->chunk_root;
3281 path = btrfs_alloc_path();
3285 /* first we search for all of the device items, and then we
3286 * read in all of the chunk items. This way we can create chunk
3287 * mappings that reference all of the devices that are afound
3289 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3295 leaf = path->nodes[0];
3296 slot = path->slots[0];
3297 if (slot >= btrfs_header_nritems(leaf)) {
3298 ret = btrfs_next_leaf(root, path);
3305 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3306 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3307 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3309 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3310 struct btrfs_dev_item *dev_item;
3311 dev_item = btrfs_item_ptr(leaf, slot,
3312 struct btrfs_dev_item);
3313 ret = read_one_dev(root, leaf, dev_item);
3317 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3318 struct btrfs_chunk *chunk;
3319 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3320 ret = read_one_chunk(root, &found_key, leaf, chunk);
3326 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3328 btrfs_release_path(root, path);
3333 btrfs_free_path(path);