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/version.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;
107 struct list_head *cur;
109 list_for_each(cur, head) {
110 dev = list_entry(cur, struct btrfs_device, dev_list);
111 if (dev->devid == devid &&
112 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
119 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
121 struct list_head *cur;
122 struct btrfs_fs_devices *fs_devices;
124 list_for_each(cur, &fs_uuids) {
125 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
126 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
133 * we try to collect pending bios for a device so we don't get a large
134 * number of procs sending bios down to the same device. This greatly
135 * improves the schedulers ability to collect and merge the bios.
137 * But, it also turns into a long list of bios to process and that is sure
138 * to eventually make the worker thread block. The solution here is to
139 * make some progress and then put this work struct back at the end of
140 * the list if the block device is congested. This way, multiple devices
141 * can make progress from a single worker thread.
143 static int noinline run_scheduled_bios(struct btrfs_device *device)
146 struct backing_dev_info *bdi;
147 struct btrfs_fs_info *fs_info;
151 unsigned long num_run = 0;
154 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
155 fs_info = device->dev_root->fs_info;
156 limit = btrfs_async_submit_limit(fs_info);
157 limit = limit * 2 / 3;
160 spin_lock(&device->io_lock);
162 /* take all the bios off the list at once and process them
163 * later on (without the lock held). But, remember the
164 * tail and other pointers so the bios can be properly reinserted
165 * into the list if we hit congestion
167 pending = device->pending_bios;
168 tail = device->pending_bio_tail;
169 WARN_ON(pending && !tail);
170 device->pending_bios = NULL;
171 device->pending_bio_tail = NULL;
174 * if pending was null this time around, no bios need processing
175 * at all and we can stop. Otherwise it'll loop back up again
176 * and do an additional check so no bios are missed.
178 * device->running_pending is used to synchronize with the
183 device->running_pending = 1;
186 device->running_pending = 0;
188 spin_unlock(&device->io_lock);
192 pending = pending->bi_next;
194 atomic_dec(&fs_info->nr_async_bios);
196 if (atomic_read(&fs_info->nr_async_bios) < limit &&
197 waitqueue_active(&fs_info->async_submit_wait))
198 wake_up(&fs_info->async_submit_wait);
200 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
202 submit_bio(cur->bi_rw, cur);
207 * we made progress, there is more work to do and the bdi
208 * is now congested. Back off and let other work structs
211 if (pending && bdi_write_congested(bdi) &&
212 fs_info->fs_devices->open_devices > 1) {
213 struct bio *old_head;
215 spin_lock(&device->io_lock);
217 old_head = device->pending_bios;
218 device->pending_bios = pending;
219 if (device->pending_bio_tail)
220 tail->bi_next = old_head;
222 device->pending_bio_tail = tail;
224 spin_unlock(&device->io_lock);
225 btrfs_requeue_work(&device->work);
235 static void pending_bios_fn(struct btrfs_work *work)
237 struct btrfs_device *device;
239 device = container_of(work, struct btrfs_device, work);
240 run_scheduled_bios(device);
243 static noinline int device_list_add(const char *path,
244 struct btrfs_super_block *disk_super,
245 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
247 struct btrfs_device *device;
248 struct btrfs_fs_devices *fs_devices;
249 u64 found_transid = btrfs_super_generation(disk_super);
251 fs_devices = find_fsid(disk_super->fsid);
253 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
256 INIT_LIST_HEAD(&fs_devices->devices);
257 INIT_LIST_HEAD(&fs_devices->alloc_list);
258 list_add(&fs_devices->list, &fs_uuids);
259 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
260 fs_devices->latest_devid = devid;
261 fs_devices->latest_trans = found_transid;
264 device = __find_device(&fs_devices->devices, devid,
265 disk_super->dev_item.uuid);
268 if (fs_devices->opened)
271 device = kzalloc(sizeof(*device), GFP_NOFS);
273 /* we can safely leave the fs_devices entry around */
276 device->devid = devid;
277 device->work.func = pending_bios_fn;
278 memcpy(device->uuid, disk_super->dev_item.uuid,
280 device->barriers = 1;
281 spin_lock_init(&device->io_lock);
282 device->name = kstrdup(path, GFP_NOFS);
287 INIT_LIST_HEAD(&device->dev_alloc_list);
288 list_add(&device->dev_list, &fs_devices->devices);
289 device->fs_devices = fs_devices;
290 fs_devices->num_devices++;
293 if (found_transid > fs_devices->latest_trans) {
294 fs_devices->latest_devid = devid;
295 fs_devices->latest_trans = found_transid;
297 *fs_devices_ret = fs_devices;
301 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
303 struct btrfs_fs_devices *fs_devices;
304 struct btrfs_device *device;
305 struct btrfs_device *orig_dev;
307 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
309 return ERR_PTR(-ENOMEM);
311 INIT_LIST_HEAD(&fs_devices->devices);
312 INIT_LIST_HEAD(&fs_devices->alloc_list);
313 INIT_LIST_HEAD(&fs_devices->list);
314 fs_devices->latest_devid = orig->latest_devid;
315 fs_devices->latest_trans = orig->latest_trans;
316 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
318 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
319 device = kzalloc(sizeof(*device), GFP_NOFS);
323 device->name = kstrdup(orig_dev->name, GFP_NOFS);
327 device->devid = orig_dev->devid;
328 device->work.func = pending_bios_fn;
329 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
330 device->barriers = 1;
331 spin_lock_init(&device->io_lock);
332 INIT_LIST_HEAD(&device->dev_list);
333 INIT_LIST_HEAD(&device->dev_alloc_list);
335 list_add(&device->dev_list, &fs_devices->devices);
336 device->fs_devices = fs_devices;
337 fs_devices->num_devices++;
341 free_fs_devices(fs_devices);
342 return ERR_PTR(-ENOMEM);
345 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
347 struct list_head *tmp;
348 struct list_head *cur;
349 struct btrfs_device *device;
351 mutex_lock(&uuid_mutex);
353 list_for_each_safe(cur, tmp, &fs_devices->devices) {
354 device = list_entry(cur, struct btrfs_device, dev_list);
355 if (device->in_fs_metadata)
359 close_bdev_exclusive(device->bdev, device->mode);
361 fs_devices->open_devices--;
363 if (device->writeable) {
364 list_del_init(&device->dev_alloc_list);
365 device->writeable = 0;
366 fs_devices->rw_devices--;
368 list_del_init(&device->dev_list);
369 fs_devices->num_devices--;
374 if (fs_devices->seed) {
375 fs_devices = fs_devices->seed;
379 mutex_unlock(&uuid_mutex);
383 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
385 struct list_head *cur;
386 struct btrfs_device *device;
388 if (--fs_devices->opened > 0)
391 list_for_each(cur, &fs_devices->devices) {
392 device = list_entry(cur, struct btrfs_device, dev_list);
394 close_bdev_exclusive(device->bdev, device->mode);
395 fs_devices->open_devices--;
397 if (device->writeable) {
398 list_del_init(&device->dev_alloc_list);
399 fs_devices->rw_devices--;
403 device->writeable = 0;
404 device->in_fs_metadata = 0;
406 WARN_ON(fs_devices->open_devices);
407 WARN_ON(fs_devices->rw_devices);
408 fs_devices->opened = 0;
409 fs_devices->seeding = 0;
414 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
416 struct btrfs_fs_devices *seed_devices = NULL;
419 mutex_lock(&uuid_mutex);
420 ret = __btrfs_close_devices(fs_devices);
421 if (!fs_devices->opened) {
422 seed_devices = fs_devices->seed;
423 fs_devices->seed = NULL;
425 mutex_unlock(&uuid_mutex);
427 while (seed_devices) {
428 fs_devices = seed_devices;
429 seed_devices = fs_devices->seed;
430 __btrfs_close_devices(fs_devices);
431 free_fs_devices(fs_devices);
436 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
437 fmode_t flags, void *holder)
439 struct block_device *bdev;
440 struct list_head *head = &fs_devices->devices;
441 struct list_head *cur;
442 struct btrfs_device *device;
443 struct block_device *latest_bdev = NULL;
444 struct buffer_head *bh;
445 struct btrfs_super_block *disk_super;
446 u64 latest_devid = 0;
447 u64 latest_transid = 0;
452 list_for_each(cur, head) {
453 device = list_entry(cur, struct btrfs_device, dev_list);
459 bdev = open_bdev_exclusive(device->name, flags, holder);
461 printk("open %s failed\n", device->name);
464 set_blocksize(bdev, 4096);
466 bh = btrfs_read_dev_super(bdev);
470 disk_super = (struct btrfs_super_block *)bh->b_data;
471 devid = le64_to_cpu(disk_super->dev_item.devid);
472 if (devid != device->devid)
475 if (memcmp(device->uuid, disk_super->dev_item.uuid,
479 device->generation = btrfs_super_generation(disk_super);
480 if (!latest_transid || device->generation > latest_transid) {
481 latest_devid = devid;
482 latest_transid = device->generation;
486 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
487 device->writeable = 0;
489 device->writeable = !bdev_read_only(bdev);
494 device->in_fs_metadata = 0;
495 device->mode = flags;
497 fs_devices->open_devices++;
498 if (device->writeable) {
499 fs_devices->rw_devices++;
500 list_add(&device->dev_alloc_list,
501 &fs_devices->alloc_list);
508 close_bdev_exclusive(bdev, FMODE_READ);
512 if (fs_devices->open_devices == 0) {
516 fs_devices->seeding = seeding;
517 fs_devices->opened = 1;
518 fs_devices->latest_bdev = latest_bdev;
519 fs_devices->latest_devid = latest_devid;
520 fs_devices->latest_trans = latest_transid;
521 fs_devices->total_rw_bytes = 0;
526 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
527 fmode_t flags, void *holder)
531 mutex_lock(&uuid_mutex);
532 if (fs_devices->opened) {
533 fs_devices->opened++;
536 ret = __btrfs_open_devices(fs_devices, flags, holder);
538 mutex_unlock(&uuid_mutex);
542 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
543 struct btrfs_fs_devices **fs_devices_ret)
545 struct btrfs_super_block *disk_super;
546 struct block_device *bdev;
547 struct buffer_head *bh;
552 mutex_lock(&uuid_mutex);
554 bdev = open_bdev_exclusive(path, flags, holder);
561 ret = set_blocksize(bdev, 4096);
564 bh = btrfs_read_dev_super(bdev);
569 disk_super = (struct btrfs_super_block *)bh->b_data;
570 devid = le64_to_cpu(disk_super->dev_item.devid);
571 transid = btrfs_super_generation(disk_super);
572 if (disk_super->label[0])
573 printk("device label %s ", disk_super->label);
575 /* FIXME, make a readl uuid parser */
576 printk("device fsid %llx-%llx ",
577 *(unsigned long long *)disk_super->fsid,
578 *(unsigned long long *)(disk_super->fsid + 8));
580 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
581 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
585 close_bdev_exclusive(bdev, flags);
587 mutex_unlock(&uuid_mutex);
592 * this uses a pretty simple search, the expectation is that it is
593 * called very infrequently and that a given device has a small number
596 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
597 struct btrfs_device *device,
598 u64 num_bytes, u64 *start)
600 struct btrfs_key key;
601 struct btrfs_root *root = device->dev_root;
602 struct btrfs_dev_extent *dev_extent = NULL;
603 struct btrfs_path *path;
606 u64 search_start = 0;
607 u64 search_end = device->total_bytes;
611 struct extent_buffer *l;
613 path = btrfs_alloc_path();
619 /* FIXME use last free of some kind */
621 /* we don't want to overwrite the superblock on the drive,
622 * so we make sure to start at an offset of at least 1MB
624 search_start = max((u64)1024 * 1024, search_start);
626 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
627 search_start = max(root->fs_info->alloc_start, search_start);
629 key.objectid = device->devid;
630 key.offset = search_start;
631 key.type = BTRFS_DEV_EXTENT_KEY;
632 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
635 ret = btrfs_previous_item(root, path, 0, key.type);
639 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
642 slot = path->slots[0];
643 if (slot >= btrfs_header_nritems(l)) {
644 ret = btrfs_next_leaf(root, path);
651 if (search_start >= search_end) {
655 *start = search_start;
659 *start = last_byte > search_start ?
660 last_byte : search_start;
661 if (search_end <= *start) {
667 btrfs_item_key_to_cpu(l, &key, slot);
669 if (key.objectid < device->devid)
672 if (key.objectid > device->devid)
675 if (key.offset >= search_start && key.offset > last_byte &&
677 if (last_byte < search_start)
678 last_byte = search_start;
679 hole_size = key.offset - last_byte;
680 if (key.offset > last_byte &&
681 hole_size >= num_bytes) {
686 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
691 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
692 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
698 /* we have to make sure we didn't find an extent that has already
699 * been allocated by the map tree or the original allocation
701 BUG_ON(*start < search_start);
703 if (*start + num_bytes > search_end) {
707 /* check for pending inserts here */
711 btrfs_free_path(path);
715 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
716 struct btrfs_device *device,
720 struct btrfs_path *path;
721 struct btrfs_root *root = device->dev_root;
722 struct btrfs_key key;
723 struct btrfs_key found_key;
724 struct extent_buffer *leaf = NULL;
725 struct btrfs_dev_extent *extent = NULL;
727 path = btrfs_alloc_path();
731 key.objectid = device->devid;
733 key.type = BTRFS_DEV_EXTENT_KEY;
735 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
737 ret = btrfs_previous_item(root, path, key.objectid,
738 BTRFS_DEV_EXTENT_KEY);
740 leaf = path->nodes[0];
741 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
742 extent = btrfs_item_ptr(leaf, path->slots[0],
743 struct btrfs_dev_extent);
744 BUG_ON(found_key.offset > start || found_key.offset +
745 btrfs_dev_extent_length(leaf, extent) < start);
747 } else if (ret == 0) {
748 leaf = path->nodes[0];
749 extent = btrfs_item_ptr(leaf, path->slots[0],
750 struct btrfs_dev_extent);
754 if (device->bytes_used > 0)
755 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
756 ret = btrfs_del_item(trans, root, path);
759 btrfs_free_path(path);
763 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
764 struct btrfs_device *device,
765 u64 chunk_tree, u64 chunk_objectid,
766 u64 chunk_offset, u64 start, u64 num_bytes)
769 struct btrfs_path *path;
770 struct btrfs_root *root = device->dev_root;
771 struct btrfs_dev_extent *extent;
772 struct extent_buffer *leaf;
773 struct btrfs_key key;
775 WARN_ON(!device->in_fs_metadata);
776 path = btrfs_alloc_path();
780 key.objectid = device->devid;
782 key.type = BTRFS_DEV_EXTENT_KEY;
783 ret = btrfs_insert_empty_item(trans, root, path, &key,
787 leaf = path->nodes[0];
788 extent = btrfs_item_ptr(leaf, path->slots[0],
789 struct btrfs_dev_extent);
790 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
791 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
792 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
794 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
795 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
798 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
799 btrfs_mark_buffer_dirty(leaf);
800 btrfs_free_path(path);
804 static noinline int find_next_chunk(struct btrfs_root *root,
805 u64 objectid, u64 *offset)
807 struct btrfs_path *path;
809 struct btrfs_key key;
810 struct btrfs_chunk *chunk;
811 struct btrfs_key found_key;
813 path = btrfs_alloc_path();
816 key.objectid = objectid;
817 key.offset = (u64)-1;
818 key.type = BTRFS_CHUNK_ITEM_KEY;
820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
826 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
830 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
832 if (found_key.objectid != objectid)
835 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
837 *offset = found_key.offset +
838 btrfs_chunk_length(path->nodes[0], chunk);
843 btrfs_free_path(path);
847 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
850 struct btrfs_key key;
851 struct btrfs_key found_key;
852 struct btrfs_path *path;
854 root = root->fs_info->chunk_root;
856 path = btrfs_alloc_path();
860 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
861 key.type = BTRFS_DEV_ITEM_KEY;
862 key.offset = (u64)-1;
864 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
870 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
875 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
877 *objectid = found_key.offset + 1;
881 btrfs_free_path(path);
886 * the device information is stored in the chunk root
887 * the btrfs_device struct should be fully filled in
889 int btrfs_add_device(struct btrfs_trans_handle *trans,
890 struct btrfs_root *root,
891 struct btrfs_device *device)
894 struct btrfs_path *path;
895 struct btrfs_dev_item *dev_item;
896 struct extent_buffer *leaf;
897 struct btrfs_key key;
900 root = root->fs_info->chunk_root;
902 path = btrfs_alloc_path();
906 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
907 key.type = BTRFS_DEV_ITEM_KEY;
908 key.offset = device->devid;
910 ret = btrfs_insert_empty_item(trans, root, path, &key,
915 leaf = path->nodes[0];
916 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
918 btrfs_set_device_id(leaf, dev_item, device->devid);
919 btrfs_set_device_generation(leaf, dev_item, 0);
920 btrfs_set_device_type(leaf, dev_item, device->type);
921 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
922 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
923 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
924 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
925 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
926 btrfs_set_device_group(leaf, dev_item, 0);
927 btrfs_set_device_seek_speed(leaf, dev_item, 0);
928 btrfs_set_device_bandwidth(leaf, dev_item, 0);
929 btrfs_set_device_start_offset(leaf, dev_item, 0);
931 ptr = (unsigned long)btrfs_device_uuid(dev_item);
932 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
933 ptr = (unsigned long)btrfs_device_fsid(dev_item);
934 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
935 btrfs_mark_buffer_dirty(leaf);
939 btrfs_free_path(path);
943 static int btrfs_rm_dev_item(struct btrfs_root *root,
944 struct btrfs_device *device)
947 struct btrfs_path *path;
948 struct btrfs_key key;
949 struct btrfs_trans_handle *trans;
951 root = root->fs_info->chunk_root;
953 path = btrfs_alloc_path();
957 trans = btrfs_start_transaction(root, 1);
958 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
959 key.type = BTRFS_DEV_ITEM_KEY;
960 key.offset = device->devid;
963 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
972 ret = btrfs_del_item(trans, root, path);
976 btrfs_free_path(path);
978 btrfs_commit_transaction(trans, root);
982 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
984 struct btrfs_device *device;
985 struct btrfs_device *next_device;
986 struct block_device *bdev;
987 struct buffer_head *bh = NULL;
988 struct btrfs_super_block *disk_super;
995 mutex_lock(&uuid_mutex);
996 mutex_lock(&root->fs_info->volume_mutex);
998 all_avail = root->fs_info->avail_data_alloc_bits |
999 root->fs_info->avail_system_alloc_bits |
1000 root->fs_info->avail_metadata_alloc_bits;
1002 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1003 root->fs_info->fs_devices->rw_devices <= 4) {
1004 printk("btrfs: unable to go below four devices on raid10\n");
1009 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1010 root->fs_info->fs_devices->rw_devices <= 2) {
1011 printk("btrfs: unable to go below two devices on raid1\n");
1016 if (strcmp(device_path, "missing") == 0) {
1017 struct list_head *cur;
1018 struct list_head *devices;
1019 struct btrfs_device *tmp;
1022 devices = &root->fs_info->fs_devices->devices;
1023 list_for_each(cur, devices) {
1024 tmp = list_entry(cur, struct btrfs_device, dev_list);
1025 if (tmp->in_fs_metadata && !tmp->bdev) {
1034 printk("btrfs: no missing devices found to remove\n");
1038 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1039 root->fs_info->bdev_holder);
1041 ret = PTR_ERR(bdev);
1045 set_blocksize(bdev, 4096);
1046 bh = btrfs_read_dev_super(bdev);
1051 disk_super = (struct btrfs_super_block *)bh->b_data;
1052 devid = le64_to_cpu(disk_super->dev_item.devid);
1053 dev_uuid = disk_super->dev_item.uuid;
1054 device = btrfs_find_device(root, devid, dev_uuid,
1062 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1063 printk("btrfs: unable to remove the only writeable device\n");
1068 if (device->writeable) {
1069 list_del_init(&device->dev_alloc_list);
1070 root->fs_info->fs_devices->rw_devices--;
1073 ret = btrfs_shrink_device(device, 0);
1077 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1081 device->in_fs_metadata = 0;
1082 list_del_init(&device->dev_list);
1083 device->fs_devices->num_devices--;
1085 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1086 struct btrfs_device, dev_list);
1087 if (device->bdev == root->fs_info->sb->s_bdev)
1088 root->fs_info->sb->s_bdev = next_device->bdev;
1089 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1090 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1093 close_bdev_exclusive(device->bdev, device->mode);
1094 device->bdev = NULL;
1095 device->fs_devices->open_devices--;
1098 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1099 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1101 if (device->fs_devices->open_devices == 0) {
1102 struct btrfs_fs_devices *fs_devices;
1103 fs_devices = root->fs_info->fs_devices;
1104 while (fs_devices) {
1105 if (fs_devices->seed == device->fs_devices)
1107 fs_devices = fs_devices->seed;
1109 fs_devices->seed = device->fs_devices->seed;
1110 device->fs_devices->seed = NULL;
1111 __btrfs_close_devices(device->fs_devices);
1112 free_fs_devices(device->fs_devices);
1116 * at this point, the device is zero sized. We want to
1117 * remove it from the devices list and zero out the old super
1119 if (device->writeable) {
1120 /* make sure this device isn't detected as part of
1123 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1124 set_buffer_dirty(bh);
1125 sync_dirty_buffer(bh);
1128 kfree(device->name);
1136 close_bdev_exclusive(bdev, FMODE_READ);
1138 mutex_unlock(&root->fs_info->volume_mutex);
1139 mutex_unlock(&uuid_mutex);
1144 * does all the dirty work required for changing file system's UUID.
1146 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1147 struct btrfs_root *root)
1149 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1150 struct btrfs_fs_devices *old_devices;
1151 struct btrfs_fs_devices *seed_devices;
1152 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1153 struct btrfs_device *device;
1156 BUG_ON(!mutex_is_locked(&uuid_mutex));
1157 if (!fs_devices->seeding)
1160 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1164 old_devices = clone_fs_devices(fs_devices);
1165 if (IS_ERR(old_devices)) {
1166 kfree(seed_devices);
1167 return PTR_ERR(old_devices);
1170 list_add(&old_devices->list, &fs_uuids);
1172 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1173 seed_devices->opened = 1;
1174 INIT_LIST_HEAD(&seed_devices->devices);
1175 INIT_LIST_HEAD(&seed_devices->alloc_list);
1176 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1177 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1178 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1179 device->fs_devices = seed_devices;
1182 fs_devices->seeding = 0;
1183 fs_devices->num_devices = 0;
1184 fs_devices->open_devices = 0;
1185 fs_devices->seed = seed_devices;
1187 generate_random_uuid(fs_devices->fsid);
1188 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1189 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1190 super_flags = btrfs_super_flags(disk_super) &
1191 ~BTRFS_SUPER_FLAG_SEEDING;
1192 btrfs_set_super_flags(disk_super, super_flags);
1198 * strore the expected generation for seed devices in device items.
1200 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1201 struct btrfs_root *root)
1203 struct btrfs_path *path;
1204 struct extent_buffer *leaf;
1205 struct btrfs_dev_item *dev_item;
1206 struct btrfs_device *device;
1207 struct btrfs_key key;
1208 u8 fs_uuid[BTRFS_UUID_SIZE];
1209 u8 dev_uuid[BTRFS_UUID_SIZE];
1213 path = btrfs_alloc_path();
1217 root = root->fs_info->chunk_root;
1218 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1220 key.type = BTRFS_DEV_ITEM_KEY;
1223 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1227 leaf = path->nodes[0];
1229 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1230 ret = btrfs_next_leaf(root, path);
1235 leaf = path->nodes[0];
1236 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1237 btrfs_release_path(root, path);
1241 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1242 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1243 key.type != BTRFS_DEV_ITEM_KEY)
1246 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1247 struct btrfs_dev_item);
1248 devid = btrfs_device_id(leaf, dev_item);
1249 read_extent_buffer(leaf, dev_uuid,
1250 (unsigned long)btrfs_device_uuid(dev_item),
1252 read_extent_buffer(leaf, fs_uuid,
1253 (unsigned long)btrfs_device_fsid(dev_item),
1255 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1258 if (device->fs_devices->seeding) {
1259 btrfs_set_device_generation(leaf, dev_item,
1260 device->generation);
1261 btrfs_mark_buffer_dirty(leaf);
1269 btrfs_free_path(path);
1273 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1275 struct btrfs_trans_handle *trans;
1276 struct btrfs_device *device;
1277 struct block_device *bdev;
1278 struct list_head *cur;
1279 struct list_head *devices;
1280 struct super_block *sb = root->fs_info->sb;
1282 int seeding_dev = 0;
1285 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1288 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1293 if (root->fs_info->fs_devices->seeding) {
1295 down_write(&sb->s_umount);
1296 mutex_lock(&uuid_mutex);
1299 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1300 mutex_lock(&root->fs_info->volume_mutex);
1302 devices = &root->fs_info->fs_devices->devices;
1303 list_for_each(cur, devices) {
1304 device = list_entry(cur, struct btrfs_device, dev_list);
1305 if (device->bdev == bdev) {
1311 device = kzalloc(sizeof(*device), GFP_NOFS);
1313 /* we can safely leave the fs_devices entry around */
1318 device->name = kstrdup(device_path, GFP_NOFS);
1319 if (!device->name) {
1325 ret = find_next_devid(root, &device->devid);
1331 trans = btrfs_start_transaction(root, 1);
1334 device->barriers = 1;
1335 device->writeable = 1;
1336 device->work.func = pending_bios_fn;
1337 generate_random_uuid(device->uuid);
1338 spin_lock_init(&device->io_lock);
1339 device->generation = trans->transid;
1340 device->io_width = root->sectorsize;
1341 device->io_align = root->sectorsize;
1342 device->sector_size = root->sectorsize;
1343 device->total_bytes = i_size_read(bdev->bd_inode);
1344 device->dev_root = root->fs_info->dev_root;
1345 device->bdev = bdev;
1346 device->in_fs_metadata = 1;
1348 set_blocksize(device->bdev, 4096);
1351 sb->s_flags &= ~MS_RDONLY;
1352 ret = btrfs_prepare_sprout(trans, root);
1356 device->fs_devices = root->fs_info->fs_devices;
1357 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1358 list_add(&device->dev_alloc_list,
1359 &root->fs_info->fs_devices->alloc_list);
1360 root->fs_info->fs_devices->num_devices++;
1361 root->fs_info->fs_devices->open_devices++;
1362 root->fs_info->fs_devices->rw_devices++;
1363 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1365 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1366 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1367 total_bytes + device->total_bytes);
1369 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1370 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1374 ret = init_first_rw_device(trans, root, device);
1376 ret = btrfs_finish_sprout(trans, root);
1379 ret = btrfs_add_device(trans, root, device);
1382 unlock_chunks(root);
1383 btrfs_commit_transaction(trans, root);
1386 mutex_unlock(&uuid_mutex);
1387 up_write(&sb->s_umount);
1389 ret = btrfs_relocate_sys_chunks(root);
1393 mutex_unlock(&root->fs_info->volume_mutex);
1396 close_bdev_exclusive(bdev, 0);
1398 mutex_unlock(&uuid_mutex);
1399 up_write(&sb->s_umount);
1404 static int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1405 struct btrfs_device *device)
1408 struct btrfs_path *path;
1409 struct btrfs_root *root;
1410 struct btrfs_dev_item *dev_item;
1411 struct extent_buffer *leaf;
1412 struct btrfs_key key;
1414 root = device->dev_root->fs_info->chunk_root;
1416 path = btrfs_alloc_path();
1420 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1421 key.type = BTRFS_DEV_ITEM_KEY;
1422 key.offset = device->devid;
1424 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1433 leaf = path->nodes[0];
1434 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1436 btrfs_set_device_id(leaf, dev_item, device->devid);
1437 btrfs_set_device_type(leaf, dev_item, device->type);
1438 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1439 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1440 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1441 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1442 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1443 btrfs_mark_buffer_dirty(leaf);
1446 btrfs_free_path(path);
1450 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1451 struct btrfs_device *device, u64 new_size)
1453 struct btrfs_super_block *super_copy =
1454 &device->dev_root->fs_info->super_copy;
1455 u64 old_total = btrfs_super_total_bytes(super_copy);
1456 u64 diff = new_size - device->total_bytes;
1458 if (!device->writeable)
1460 if (new_size <= device->total_bytes)
1463 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1464 device->fs_devices->total_rw_bytes += diff;
1466 device->total_bytes = new_size;
1467 return btrfs_update_device(trans, device);
1470 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1471 struct btrfs_device *device, u64 new_size)
1474 lock_chunks(device->dev_root);
1475 ret = __btrfs_grow_device(trans, device, new_size);
1476 unlock_chunks(device->dev_root);
1480 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1481 struct btrfs_root *root,
1482 u64 chunk_tree, u64 chunk_objectid,
1486 struct btrfs_path *path;
1487 struct btrfs_key key;
1489 root = root->fs_info->chunk_root;
1490 path = btrfs_alloc_path();
1494 key.objectid = chunk_objectid;
1495 key.offset = chunk_offset;
1496 key.type = BTRFS_CHUNK_ITEM_KEY;
1498 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1501 ret = btrfs_del_item(trans, root, path);
1504 btrfs_free_path(path);
1508 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1511 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1512 struct btrfs_disk_key *disk_key;
1513 struct btrfs_chunk *chunk;
1520 struct btrfs_key key;
1522 array_size = btrfs_super_sys_array_size(super_copy);
1524 ptr = super_copy->sys_chunk_array;
1527 while (cur < array_size) {
1528 disk_key = (struct btrfs_disk_key *)ptr;
1529 btrfs_disk_key_to_cpu(&key, disk_key);
1531 len = sizeof(*disk_key);
1533 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1534 chunk = (struct btrfs_chunk *)(ptr + len);
1535 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1536 len += btrfs_chunk_item_size(num_stripes);
1541 if (key.objectid == chunk_objectid &&
1542 key.offset == chunk_offset) {
1543 memmove(ptr, ptr + len, array_size - (cur + len));
1545 btrfs_set_super_sys_array_size(super_copy, array_size);
1554 static int btrfs_relocate_chunk(struct btrfs_root *root,
1555 u64 chunk_tree, u64 chunk_objectid,
1558 struct extent_map_tree *em_tree;
1559 struct btrfs_root *extent_root;
1560 struct btrfs_trans_handle *trans;
1561 struct extent_map *em;
1562 struct map_lookup *map;
1566 printk("btrfs relocating chunk %llu\n",
1567 (unsigned long long)chunk_offset);
1568 root = root->fs_info->chunk_root;
1569 extent_root = root->fs_info->extent_root;
1570 em_tree = &root->fs_info->mapping_tree.map_tree;
1572 /* step one, relocate all the extents inside this chunk */
1573 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1576 trans = btrfs_start_transaction(root, 1);
1582 * step two, delete the device extents and the
1583 * chunk tree entries
1585 spin_lock(&em_tree->lock);
1586 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1587 spin_unlock(&em_tree->lock);
1589 BUG_ON(em->start > chunk_offset ||
1590 em->start + em->len < chunk_offset);
1591 map = (struct map_lookup *)em->bdev;
1593 for (i = 0; i < map->num_stripes; i++) {
1594 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1595 map->stripes[i].physical);
1598 if (map->stripes[i].dev) {
1599 ret = btrfs_update_device(trans, map->stripes[i].dev);
1603 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1608 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1609 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1613 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1616 spin_lock(&em_tree->lock);
1617 remove_extent_mapping(em_tree, em);
1618 spin_unlock(&em_tree->lock);
1623 /* once for the tree */
1624 free_extent_map(em);
1626 free_extent_map(em);
1628 unlock_chunks(root);
1629 btrfs_end_transaction(trans, root);
1633 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1635 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1636 struct btrfs_path *path;
1637 struct extent_buffer *leaf;
1638 struct btrfs_chunk *chunk;
1639 struct btrfs_key key;
1640 struct btrfs_key found_key;
1641 u64 chunk_tree = chunk_root->root_key.objectid;
1645 path = btrfs_alloc_path();
1649 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1650 key.offset = (u64)-1;
1651 key.type = BTRFS_CHUNK_ITEM_KEY;
1654 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1659 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1666 leaf = path->nodes[0];
1667 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1669 chunk = btrfs_item_ptr(leaf, path->slots[0],
1670 struct btrfs_chunk);
1671 chunk_type = btrfs_chunk_type(leaf, chunk);
1672 btrfs_release_path(chunk_root, path);
1674 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1675 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1681 if (found_key.offset == 0)
1683 key.offset = found_key.offset - 1;
1687 btrfs_free_path(path);
1691 static u64 div_factor(u64 num, int factor)
1700 int btrfs_balance(struct btrfs_root *dev_root)
1703 struct list_head *cur;
1704 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1705 struct btrfs_device *device;
1708 struct btrfs_path *path;
1709 struct btrfs_key key;
1710 struct btrfs_chunk *chunk;
1711 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1712 struct btrfs_trans_handle *trans;
1713 struct btrfs_key found_key;
1715 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1718 mutex_lock(&dev_root->fs_info->volume_mutex);
1719 dev_root = dev_root->fs_info->dev_root;
1721 /* step one make some room on all the devices */
1722 list_for_each(cur, devices) {
1723 device = list_entry(cur, struct btrfs_device, dev_list);
1724 old_size = device->total_bytes;
1725 size_to_free = div_factor(old_size, 1);
1726 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1727 if (!device->writeable ||
1728 device->total_bytes - device->bytes_used > size_to_free)
1731 ret = btrfs_shrink_device(device, old_size - size_to_free);
1734 trans = btrfs_start_transaction(dev_root, 1);
1737 ret = btrfs_grow_device(trans, device, old_size);
1740 btrfs_end_transaction(trans, dev_root);
1743 /* step two, relocate all the chunks */
1744 path = btrfs_alloc_path();
1747 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1748 key.offset = (u64)-1;
1749 key.type = BTRFS_CHUNK_ITEM_KEY;
1752 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1757 * this shouldn't happen, it means the last relocate
1763 ret = btrfs_previous_item(chunk_root, path, 0,
1764 BTRFS_CHUNK_ITEM_KEY);
1768 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1770 if (found_key.objectid != key.objectid)
1773 chunk = btrfs_item_ptr(path->nodes[0],
1775 struct btrfs_chunk);
1776 key.offset = found_key.offset;
1777 /* chunk zero is special */
1778 if (key.offset == 0)
1781 btrfs_release_path(chunk_root, path);
1782 ret = btrfs_relocate_chunk(chunk_root,
1783 chunk_root->root_key.objectid,
1790 btrfs_free_path(path);
1791 mutex_unlock(&dev_root->fs_info->volume_mutex);
1796 * shrinking a device means finding all of the device extents past
1797 * the new size, and then following the back refs to the chunks.
1798 * The chunk relocation code actually frees the device extent
1800 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1802 struct btrfs_trans_handle *trans;
1803 struct btrfs_root *root = device->dev_root;
1804 struct btrfs_dev_extent *dev_extent = NULL;
1805 struct btrfs_path *path;
1812 struct extent_buffer *l;
1813 struct btrfs_key key;
1814 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1815 u64 old_total = btrfs_super_total_bytes(super_copy);
1816 u64 diff = device->total_bytes - new_size;
1818 if (new_size >= device->total_bytes)
1821 path = btrfs_alloc_path();
1825 trans = btrfs_start_transaction(root, 1);
1835 device->total_bytes = new_size;
1836 if (device->writeable)
1837 device->fs_devices->total_rw_bytes -= diff;
1838 ret = btrfs_update_device(trans, device);
1840 unlock_chunks(root);
1841 btrfs_end_transaction(trans, root);
1844 WARN_ON(diff > old_total);
1845 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1846 unlock_chunks(root);
1847 btrfs_end_transaction(trans, root);
1849 key.objectid = device->devid;
1850 key.offset = (u64)-1;
1851 key.type = BTRFS_DEV_EXTENT_KEY;
1854 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1858 ret = btrfs_previous_item(root, path, 0, key.type);
1867 slot = path->slots[0];
1868 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1870 if (key.objectid != device->devid)
1873 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1874 length = btrfs_dev_extent_length(l, dev_extent);
1876 if (key.offset + length <= new_size)
1879 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1880 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1881 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1882 btrfs_release_path(root, path);
1884 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1891 btrfs_free_path(path);
1895 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1896 struct btrfs_root *root,
1897 struct btrfs_key *key,
1898 struct btrfs_chunk *chunk, int item_size)
1900 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1901 struct btrfs_disk_key disk_key;
1905 array_size = btrfs_super_sys_array_size(super_copy);
1906 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1909 ptr = super_copy->sys_chunk_array + array_size;
1910 btrfs_cpu_key_to_disk(&disk_key, key);
1911 memcpy(ptr, &disk_key, sizeof(disk_key));
1912 ptr += sizeof(disk_key);
1913 memcpy(ptr, chunk, item_size);
1914 item_size += sizeof(disk_key);
1915 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1919 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1920 int num_stripes, int sub_stripes)
1922 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1924 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1925 return calc_size * (num_stripes / sub_stripes);
1927 return calc_size * num_stripes;
1930 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1931 struct btrfs_root *extent_root,
1932 struct map_lookup **map_ret,
1933 u64 *num_bytes, u64 *stripe_size,
1934 u64 start, u64 type)
1936 struct btrfs_fs_info *info = extent_root->fs_info;
1937 struct btrfs_device *device = NULL;
1938 struct btrfs_fs_devices *fs_devices = info->fs_devices;
1939 struct list_head *cur;
1940 struct map_lookup *map = NULL;
1941 struct extent_map_tree *em_tree;
1942 struct extent_map *em;
1943 struct list_head private_devs;
1944 int min_stripe_size = 1 * 1024 * 1024;
1945 u64 calc_size = 1024 * 1024 * 1024;
1946 u64 max_chunk_size = calc_size;
1951 int num_stripes = 1;
1952 int min_stripes = 1;
1953 int sub_stripes = 0;
1957 int stripe_len = 64 * 1024;
1959 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1960 (type & BTRFS_BLOCK_GROUP_DUP)) {
1962 type &= ~BTRFS_BLOCK_GROUP_DUP;
1964 if (list_empty(&fs_devices->alloc_list))
1967 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1968 num_stripes = fs_devices->rw_devices;
1971 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1975 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1976 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
1977 if (num_stripes < 2)
1981 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1982 num_stripes = fs_devices->rw_devices;
1983 if (num_stripes < 4)
1985 num_stripes &= ~(u32)1;
1990 if (type & BTRFS_BLOCK_GROUP_DATA) {
1991 max_chunk_size = 10 * calc_size;
1992 min_stripe_size = 64 * 1024 * 1024;
1993 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1994 max_chunk_size = 4 * calc_size;
1995 min_stripe_size = 32 * 1024 * 1024;
1996 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1997 calc_size = 8 * 1024 * 1024;
1998 max_chunk_size = calc_size * 2;
1999 min_stripe_size = 1 * 1024 * 1024;
2002 /* we don't want a chunk larger than 10% of writeable space */
2003 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2007 if (!map || map->num_stripes != num_stripes) {
2009 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2012 map->num_stripes = num_stripes;
2015 if (calc_size * num_stripes > max_chunk_size) {
2016 calc_size = max_chunk_size;
2017 do_div(calc_size, num_stripes);
2018 do_div(calc_size, stripe_len);
2019 calc_size *= stripe_len;
2021 /* we don't want tiny stripes */
2022 calc_size = max_t(u64, min_stripe_size, calc_size);
2024 do_div(calc_size, stripe_len);
2025 calc_size *= stripe_len;
2027 cur = fs_devices->alloc_list.next;
2030 if (type & BTRFS_BLOCK_GROUP_DUP)
2031 min_free = calc_size * 2;
2033 min_free = calc_size;
2036 * we add 1MB because we never use the first 1MB of the device, unless
2037 * we've looped, then we are likely allocating the maximum amount of
2038 * space left already
2041 min_free += 1024 * 1024;
2043 INIT_LIST_HEAD(&private_devs);
2044 while(index < num_stripes) {
2045 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2046 BUG_ON(!device->writeable);
2047 if (device->total_bytes > device->bytes_used)
2048 avail = device->total_bytes - device->bytes_used;
2053 if (device->in_fs_metadata && avail >= min_free) {
2054 ret = find_free_dev_extent(trans, device,
2055 min_free, &dev_offset);
2057 list_move_tail(&device->dev_alloc_list,
2059 map->stripes[index].dev = device;
2060 map->stripes[index].physical = dev_offset;
2062 if (type & BTRFS_BLOCK_GROUP_DUP) {
2063 map->stripes[index].dev = device;
2064 map->stripes[index].physical =
2065 dev_offset + calc_size;
2069 } else if (device->in_fs_metadata && avail > max_avail)
2071 if (cur == &fs_devices->alloc_list)
2074 list_splice(&private_devs, &fs_devices->alloc_list);
2075 if (index < num_stripes) {
2076 if (index >= min_stripes) {
2077 num_stripes = index;
2078 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2079 num_stripes /= sub_stripes;
2080 num_stripes *= sub_stripes;
2085 if (!looped && max_avail > 0) {
2087 calc_size = max_avail;
2093 map->sector_size = extent_root->sectorsize;
2094 map->stripe_len = stripe_len;
2095 map->io_align = stripe_len;
2096 map->io_width = stripe_len;
2098 map->num_stripes = num_stripes;
2099 map->sub_stripes = sub_stripes;
2102 *stripe_size = calc_size;
2103 *num_bytes = chunk_bytes_by_type(type, calc_size,
2104 num_stripes, sub_stripes);
2106 em = alloc_extent_map(GFP_NOFS);
2111 em->bdev = (struct block_device *)map;
2113 em->len = *num_bytes;
2114 em->block_start = 0;
2115 em->block_len = em->len;
2117 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2118 spin_lock(&em_tree->lock);
2119 ret = add_extent_mapping(em_tree, em);
2120 spin_unlock(&em_tree->lock);
2122 free_extent_map(em);
2124 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2125 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2130 while (index < map->num_stripes) {
2131 device = map->stripes[index].dev;
2132 dev_offset = map->stripes[index].physical;
2134 ret = btrfs_alloc_dev_extent(trans, device,
2135 info->chunk_root->root_key.objectid,
2136 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2137 start, dev_offset, calc_size);
2145 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2146 struct btrfs_root *extent_root,
2147 struct map_lookup *map, u64 chunk_offset,
2148 u64 chunk_size, u64 stripe_size)
2151 struct btrfs_key key;
2152 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2153 struct btrfs_device *device;
2154 struct btrfs_chunk *chunk;
2155 struct btrfs_stripe *stripe;
2156 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2160 chunk = kzalloc(item_size, GFP_NOFS);
2165 while (index < map->num_stripes) {
2166 device = map->stripes[index].dev;
2167 device->bytes_used += stripe_size;
2168 ret = btrfs_update_device(trans, device);
2174 stripe = &chunk->stripe;
2175 while (index < map->num_stripes) {
2176 device = map->stripes[index].dev;
2177 dev_offset = map->stripes[index].physical;
2179 btrfs_set_stack_stripe_devid(stripe, device->devid);
2180 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2181 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2186 btrfs_set_stack_chunk_length(chunk, chunk_size);
2187 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2188 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2189 btrfs_set_stack_chunk_type(chunk, map->type);
2190 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2191 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2192 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2193 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2194 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2196 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2197 key.type = BTRFS_CHUNK_ITEM_KEY;
2198 key.offset = chunk_offset;
2200 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2203 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2204 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2213 * Chunk allocation falls into two parts. The first part does works
2214 * that make the new allocated chunk useable, but not do any operation
2215 * that modifies the chunk tree. The second part does the works that
2216 * require modifying the chunk tree. This division is important for the
2217 * bootstrap process of adding storage to a seed btrfs.
2219 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2220 struct btrfs_root *extent_root, u64 type)
2225 struct map_lookup *map;
2226 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2229 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2234 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2235 &stripe_size, chunk_offset, type);
2239 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2240 chunk_size, stripe_size);
2245 static int noinline init_first_rw_device(struct btrfs_trans_handle *trans,
2246 struct btrfs_root *root,
2247 struct btrfs_device *device)
2250 u64 sys_chunk_offset;
2254 u64 sys_stripe_size;
2256 struct map_lookup *map;
2257 struct map_lookup *sys_map;
2258 struct btrfs_fs_info *fs_info = root->fs_info;
2259 struct btrfs_root *extent_root = fs_info->extent_root;
2262 ret = find_next_chunk(fs_info->chunk_root,
2263 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2266 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2267 (fs_info->metadata_alloc_profile &
2268 fs_info->avail_metadata_alloc_bits);
2269 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2271 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2272 &stripe_size, chunk_offset, alloc_profile);
2275 sys_chunk_offset = chunk_offset + chunk_size;
2277 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2278 (fs_info->system_alloc_profile &
2279 fs_info->avail_system_alloc_bits);
2280 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2282 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2283 &sys_chunk_size, &sys_stripe_size,
2284 sys_chunk_offset, alloc_profile);
2287 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2291 * Modifying chunk tree needs allocating new blocks from both
2292 * system block group and metadata block group. So we only can
2293 * do operations require modifying the chunk tree after both
2294 * block groups were created.
2296 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2297 chunk_size, stripe_size);
2300 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2301 sys_chunk_offset, sys_chunk_size,
2307 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2309 struct extent_map *em;
2310 struct map_lookup *map;
2311 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2315 spin_lock(&map_tree->map_tree.lock);
2316 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2317 spin_unlock(&map_tree->map_tree.lock);
2321 map = (struct map_lookup *)em->bdev;
2322 for (i = 0; i < map->num_stripes; i++) {
2323 if (!map->stripes[i].dev->writeable) {
2328 free_extent_map(em);
2332 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2334 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2337 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2339 struct extent_map *em;
2342 spin_lock(&tree->map_tree.lock);
2343 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2345 remove_extent_mapping(&tree->map_tree, em);
2346 spin_unlock(&tree->map_tree.lock);
2351 free_extent_map(em);
2352 /* once for the tree */
2353 free_extent_map(em);
2357 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2359 struct extent_map *em;
2360 struct map_lookup *map;
2361 struct extent_map_tree *em_tree = &map_tree->map_tree;
2364 spin_lock(&em_tree->lock);
2365 em = lookup_extent_mapping(em_tree, logical, len);
2366 spin_unlock(&em_tree->lock);
2369 BUG_ON(em->start > logical || em->start + em->len < logical);
2370 map = (struct map_lookup *)em->bdev;
2371 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2372 ret = map->num_stripes;
2373 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2374 ret = map->sub_stripes;
2377 free_extent_map(em);
2381 static int find_live_mirror(struct map_lookup *map, int first, int num,
2385 if (map->stripes[optimal].dev->bdev)
2387 for (i = first; i < first + num; i++) {
2388 if (map->stripes[i].dev->bdev)
2391 /* we couldn't find one that doesn't fail. Just return something
2392 * and the io error handling code will clean up eventually
2397 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2398 u64 logical, u64 *length,
2399 struct btrfs_multi_bio **multi_ret,
2400 int mirror_num, struct page *unplug_page)
2402 struct extent_map *em;
2403 struct map_lookup *map;
2404 struct extent_map_tree *em_tree = &map_tree->map_tree;
2408 int stripes_allocated = 8;
2409 int stripes_required = 1;
2414 struct btrfs_multi_bio *multi = NULL;
2416 if (multi_ret && !(rw & (1 << BIO_RW))) {
2417 stripes_allocated = 1;
2421 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2426 atomic_set(&multi->error, 0);
2429 spin_lock(&em_tree->lock);
2430 em = lookup_extent_mapping(em_tree, logical, *length);
2431 spin_unlock(&em_tree->lock);
2433 if (!em && unplug_page)
2437 printk("unable to find logical %Lu len %Lu\n", logical, *length);
2441 BUG_ON(em->start > logical || em->start + em->len < logical);
2442 map = (struct map_lookup *)em->bdev;
2443 offset = logical - em->start;
2445 if (mirror_num > map->num_stripes)
2448 /* if our multi bio struct is too small, back off and try again */
2449 if (rw & (1 << BIO_RW)) {
2450 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2451 BTRFS_BLOCK_GROUP_DUP)) {
2452 stripes_required = map->num_stripes;
2454 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2455 stripes_required = map->sub_stripes;
2459 if (multi_ret && rw == WRITE &&
2460 stripes_allocated < stripes_required) {
2461 stripes_allocated = map->num_stripes;
2462 free_extent_map(em);
2468 * stripe_nr counts the total number of stripes we have to stride
2469 * to get to this block
2471 do_div(stripe_nr, map->stripe_len);
2473 stripe_offset = stripe_nr * map->stripe_len;
2474 BUG_ON(offset < stripe_offset);
2476 /* stripe_offset is the offset of this block in its stripe*/
2477 stripe_offset = offset - stripe_offset;
2479 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2480 BTRFS_BLOCK_GROUP_RAID10 |
2481 BTRFS_BLOCK_GROUP_DUP)) {
2482 /* we limit the length of each bio to what fits in a stripe */
2483 *length = min_t(u64, em->len - offset,
2484 map->stripe_len - stripe_offset);
2486 *length = em->len - offset;
2489 if (!multi_ret && !unplug_page)
2494 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2495 if (unplug_page || (rw & (1 << BIO_RW)))
2496 num_stripes = map->num_stripes;
2497 else if (mirror_num)
2498 stripe_index = mirror_num - 1;
2500 stripe_index = find_live_mirror(map, 0,
2502 current->pid % map->num_stripes);
2505 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2506 if (rw & (1 << BIO_RW))
2507 num_stripes = map->num_stripes;
2508 else if (mirror_num)
2509 stripe_index = mirror_num - 1;
2511 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2512 int factor = map->num_stripes / map->sub_stripes;
2514 stripe_index = do_div(stripe_nr, factor);
2515 stripe_index *= map->sub_stripes;
2517 if (unplug_page || (rw & (1 << BIO_RW)))
2518 num_stripes = map->sub_stripes;
2519 else if (mirror_num)
2520 stripe_index += mirror_num - 1;
2522 stripe_index = find_live_mirror(map, stripe_index,
2523 map->sub_stripes, stripe_index +
2524 current->pid % map->sub_stripes);
2528 * after this do_div call, stripe_nr is the number of stripes
2529 * on this device we have to walk to find the data, and
2530 * stripe_index is the number of our device in the stripe array
2532 stripe_index = do_div(stripe_nr, map->num_stripes);
2534 BUG_ON(stripe_index >= map->num_stripes);
2536 for (i = 0; i < num_stripes; i++) {
2538 struct btrfs_device *device;
2539 struct backing_dev_info *bdi;
2541 device = map->stripes[stripe_index].dev;
2543 bdi = blk_get_backing_dev_info(device->bdev);
2544 if (bdi->unplug_io_fn) {
2545 bdi->unplug_io_fn(bdi, unplug_page);
2549 multi->stripes[i].physical =
2550 map->stripes[stripe_index].physical +
2551 stripe_offset + stripe_nr * map->stripe_len;
2552 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2558 multi->num_stripes = num_stripes;
2559 multi->max_errors = max_errors;
2562 free_extent_map(em);
2566 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2567 u64 logical, u64 *length,
2568 struct btrfs_multi_bio **multi_ret, int mirror_num)
2570 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2574 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2575 u64 chunk_start, u64 physical, u64 devid,
2576 u64 **logical, int *naddrs, int *stripe_len)
2578 struct extent_map_tree *em_tree = &map_tree->map_tree;
2579 struct extent_map *em;
2580 struct map_lookup *map;
2587 spin_lock(&em_tree->lock);
2588 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2589 spin_unlock(&em_tree->lock);
2591 BUG_ON(!em || em->start != chunk_start);
2592 map = (struct map_lookup *)em->bdev;
2595 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2596 do_div(length, map->num_stripes / map->sub_stripes);
2597 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2598 do_div(length, map->num_stripes);
2600 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2603 for (i = 0; i < map->num_stripes; i++) {
2604 if (devid && map->stripes[i].dev->devid != devid)
2606 if (map->stripes[i].physical > physical ||
2607 map->stripes[i].physical + length <= physical)
2610 stripe_nr = physical - map->stripes[i].physical;
2611 do_div(stripe_nr, map->stripe_len);
2613 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2614 stripe_nr = stripe_nr * map->num_stripes + i;
2615 do_div(stripe_nr, map->sub_stripes);
2616 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2617 stripe_nr = stripe_nr * map->num_stripes + i;
2619 bytenr = chunk_start + stripe_nr * map->stripe_len;
2620 WARN_ON(nr >= map->num_stripes);
2621 for (j = 0; j < nr; j++) {
2622 if (buf[j] == bytenr)
2626 WARN_ON(nr >= map->num_stripes);
2631 for (i = 0; i > nr; i++) {
2632 struct btrfs_multi_bio *multi;
2633 struct btrfs_bio_stripe *stripe;
2637 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2638 &length, &multi, 0);
2641 stripe = multi->stripes;
2642 for (j = 0; j < multi->num_stripes; j++) {
2643 if (stripe->physical >= physical &&
2644 physical < stripe->physical + length)
2647 BUG_ON(j >= multi->num_stripes);
2653 *stripe_len = map->stripe_len;
2655 free_extent_map(em);
2659 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2660 u64 logical, struct page *page)
2662 u64 length = PAGE_CACHE_SIZE;
2663 return __btrfs_map_block(map_tree, READ, logical, &length,
2667 static void end_bio_multi_stripe(struct bio *bio, int err)
2669 struct btrfs_multi_bio *multi = bio->bi_private;
2670 int is_orig_bio = 0;
2673 atomic_inc(&multi->error);
2675 if (bio == multi->orig_bio)
2678 if (atomic_dec_and_test(&multi->stripes_pending)) {
2681 bio = multi->orig_bio;
2683 bio->bi_private = multi->private;
2684 bio->bi_end_io = multi->end_io;
2685 /* only send an error to the higher layers if it is
2686 * beyond the tolerance of the multi-bio
2688 if (atomic_read(&multi->error) > multi->max_errors) {
2692 * this bio is actually up to date, we didn't
2693 * go over the max number of errors
2695 set_bit(BIO_UPTODATE, &bio->bi_flags);
2700 bio_endio(bio, err);
2701 } else if (!is_orig_bio) {
2706 struct async_sched {
2709 struct btrfs_fs_info *info;
2710 struct btrfs_work work;
2714 * see run_scheduled_bios for a description of why bios are collected for
2717 * This will add one bio to the pending list for a device and make sure
2718 * the work struct is scheduled.
2720 static int noinline schedule_bio(struct btrfs_root *root,
2721 struct btrfs_device *device,
2722 int rw, struct bio *bio)
2724 int should_queue = 1;
2726 /* don't bother with additional async steps for reads, right now */
2727 if (!(rw & (1 << BIO_RW))) {
2729 submit_bio(rw, bio);
2735 * nr_async_bios allows us to reliably return congestion to the
2736 * higher layers. Otherwise, the async bio makes it appear we have
2737 * made progress against dirty pages when we've really just put it
2738 * on a queue for later
2740 atomic_inc(&root->fs_info->nr_async_bios);
2741 WARN_ON(bio->bi_next);
2742 bio->bi_next = NULL;
2745 spin_lock(&device->io_lock);
2747 if (device->pending_bio_tail)
2748 device->pending_bio_tail->bi_next = bio;
2750 device->pending_bio_tail = bio;
2751 if (!device->pending_bios)
2752 device->pending_bios = bio;
2753 if (device->running_pending)
2756 spin_unlock(&device->io_lock);
2759 btrfs_queue_worker(&root->fs_info->submit_workers,
2764 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2765 int mirror_num, int async_submit)
2767 struct btrfs_mapping_tree *map_tree;
2768 struct btrfs_device *dev;
2769 struct bio *first_bio = bio;
2770 u64 logical = (u64)bio->bi_sector << 9;
2773 struct btrfs_multi_bio *multi = NULL;
2778 length = bio->bi_size;
2779 map_tree = &root->fs_info->mapping_tree;
2780 map_length = length;
2782 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2786 total_devs = multi->num_stripes;
2787 if (map_length < length) {
2788 printk("mapping failed logical %Lu bio len %Lu "
2789 "len %Lu\n", logical, length, map_length);
2792 multi->end_io = first_bio->bi_end_io;
2793 multi->private = first_bio->bi_private;
2794 multi->orig_bio = first_bio;
2795 atomic_set(&multi->stripes_pending, multi->num_stripes);
2797 while(dev_nr < total_devs) {
2798 if (total_devs > 1) {
2799 if (dev_nr < total_devs - 1) {
2800 bio = bio_clone(first_bio, GFP_NOFS);
2805 bio->bi_private = multi;
2806 bio->bi_end_io = end_bio_multi_stripe;
2808 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2809 dev = multi->stripes[dev_nr].dev;
2810 BUG_ON(rw == WRITE && !dev->writeable);
2811 if (dev && dev->bdev) {
2812 bio->bi_bdev = dev->bdev;
2814 schedule_bio(root, dev, rw, bio);
2816 submit_bio(rw, bio);
2818 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2819 bio->bi_sector = logical >> 9;
2820 bio_endio(bio, -EIO);
2824 if (total_devs == 1)
2829 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2832 struct btrfs_device *device;
2833 struct btrfs_fs_devices *cur_devices;
2835 cur_devices = root->fs_info->fs_devices;
2836 while (cur_devices) {
2838 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2839 device = __find_device(&cur_devices->devices,
2844 cur_devices = cur_devices->seed;
2849 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2850 u64 devid, u8 *dev_uuid)
2852 struct btrfs_device *device;
2853 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2855 device = kzalloc(sizeof(*device), GFP_NOFS);
2858 list_add(&device->dev_list,
2859 &fs_devices->devices);
2860 device->barriers = 1;
2861 device->dev_root = root->fs_info->dev_root;
2862 device->devid = devid;
2863 device->work.func = pending_bios_fn;
2864 device->fs_devices = fs_devices;
2865 fs_devices->num_devices++;
2866 spin_lock_init(&device->io_lock);
2867 INIT_LIST_HEAD(&device->dev_alloc_list);
2868 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2872 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2873 struct extent_buffer *leaf,
2874 struct btrfs_chunk *chunk)
2876 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2877 struct map_lookup *map;
2878 struct extent_map *em;
2882 u8 uuid[BTRFS_UUID_SIZE];
2887 logical = key->offset;
2888 length = btrfs_chunk_length(leaf, chunk);
2890 spin_lock(&map_tree->map_tree.lock);
2891 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2892 spin_unlock(&map_tree->map_tree.lock);
2894 /* already mapped? */
2895 if (em && em->start <= logical && em->start + em->len > logical) {
2896 free_extent_map(em);
2899 free_extent_map(em);
2902 map = kzalloc(sizeof(*map), GFP_NOFS);
2906 em = alloc_extent_map(GFP_NOFS);
2909 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2910 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2912 free_extent_map(em);
2916 em->bdev = (struct block_device *)map;
2917 em->start = logical;
2919 em->block_start = 0;
2920 em->block_len = em->len;
2922 map->num_stripes = num_stripes;
2923 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2924 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2925 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2926 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2927 map->type = btrfs_chunk_type(leaf, chunk);
2928 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2929 for (i = 0; i < num_stripes; i++) {
2930 map->stripes[i].physical =
2931 btrfs_stripe_offset_nr(leaf, chunk, i);
2932 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2933 read_extent_buffer(leaf, uuid, (unsigned long)
2934 btrfs_stripe_dev_uuid_nr(chunk, i),
2936 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
2938 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2940 free_extent_map(em);
2943 if (!map->stripes[i].dev) {
2944 map->stripes[i].dev =
2945 add_missing_dev(root, devid, uuid);
2946 if (!map->stripes[i].dev) {
2948 free_extent_map(em);
2952 map->stripes[i].dev->in_fs_metadata = 1;
2955 spin_lock(&map_tree->map_tree.lock);
2956 ret = add_extent_mapping(&map_tree->map_tree, em);
2957 spin_unlock(&map_tree->map_tree.lock);
2959 free_extent_map(em);
2964 static int fill_device_from_item(struct extent_buffer *leaf,
2965 struct btrfs_dev_item *dev_item,
2966 struct btrfs_device *device)
2970 device->devid = btrfs_device_id(leaf, dev_item);
2971 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2972 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2973 device->type = btrfs_device_type(leaf, dev_item);
2974 device->io_align = btrfs_device_io_align(leaf, dev_item);
2975 device->io_width = btrfs_device_io_width(leaf, dev_item);
2976 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2978 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2979 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2984 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
2986 struct btrfs_fs_devices *fs_devices;
2989 mutex_lock(&uuid_mutex);
2991 fs_devices = root->fs_info->fs_devices->seed;
2992 while (fs_devices) {
2993 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2997 fs_devices = fs_devices->seed;
3000 fs_devices = find_fsid(fsid);
3006 fs_devices = clone_fs_devices(fs_devices);
3007 if (IS_ERR(fs_devices)) {
3008 ret = PTR_ERR(fs_devices);
3012 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3013 root->fs_info->bdev_holder);
3017 if (!fs_devices->seeding) {
3018 __btrfs_close_devices(fs_devices);
3019 free_fs_devices(fs_devices);
3024 fs_devices->seed = root->fs_info->fs_devices->seed;
3025 root->fs_info->fs_devices->seed = fs_devices;
3027 mutex_unlock(&uuid_mutex);
3031 static int read_one_dev(struct btrfs_root *root,
3032 struct extent_buffer *leaf,
3033 struct btrfs_dev_item *dev_item)
3035 struct btrfs_device *device;
3038 u8 fs_uuid[BTRFS_UUID_SIZE];
3039 u8 dev_uuid[BTRFS_UUID_SIZE];
3041 devid = btrfs_device_id(leaf, dev_item);
3042 read_extent_buffer(leaf, dev_uuid,
3043 (unsigned long)btrfs_device_uuid(dev_item),
3045 read_extent_buffer(leaf, fs_uuid,
3046 (unsigned long)btrfs_device_fsid(dev_item),
3049 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3050 ret = open_seed_devices(root, fs_uuid);
3051 if (ret && !btrfs_test_opt(root, DEGRADED))
3055 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3056 if (!device || !device->bdev) {
3057 if (!btrfs_test_opt(root, DEGRADED))
3061 printk("warning devid %Lu missing\n", devid);
3062 device = add_missing_dev(root, devid, dev_uuid);
3068 if (device->fs_devices != root->fs_info->fs_devices) {
3069 BUG_ON(device->writeable);
3070 if (device->generation !=
3071 btrfs_device_generation(leaf, dev_item))
3075 fill_device_from_item(leaf, dev_item, device);
3076 device->dev_root = root->fs_info->dev_root;
3077 device->in_fs_metadata = 1;
3078 if (device->writeable)
3079 device->fs_devices->total_rw_bytes += device->total_bytes;
3082 ret = btrfs_open_device(device);
3090 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3092 struct btrfs_dev_item *dev_item;
3094 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3096 return read_one_dev(root, buf, dev_item);
3099 int btrfs_read_sys_array(struct btrfs_root *root)
3101 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3102 struct extent_buffer *sb;
3103 struct btrfs_disk_key *disk_key;
3104 struct btrfs_chunk *chunk;
3106 unsigned long sb_ptr;
3112 struct btrfs_key key;
3114 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3115 BTRFS_SUPER_INFO_SIZE);
3118 btrfs_set_buffer_uptodate(sb);
3119 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3120 array_size = btrfs_super_sys_array_size(super_copy);
3122 ptr = super_copy->sys_chunk_array;
3123 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3126 while (cur < array_size) {
3127 disk_key = (struct btrfs_disk_key *)ptr;
3128 btrfs_disk_key_to_cpu(&key, disk_key);
3130 len = sizeof(*disk_key); ptr += len;
3134 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3135 chunk = (struct btrfs_chunk *)sb_ptr;
3136 ret = read_one_chunk(root, &key, sb, chunk);
3139 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3140 len = btrfs_chunk_item_size(num_stripes);
3149 free_extent_buffer(sb);
3153 int btrfs_read_chunk_tree(struct btrfs_root *root)
3155 struct btrfs_path *path;
3156 struct extent_buffer *leaf;
3157 struct btrfs_key key;
3158 struct btrfs_key found_key;
3162 root = root->fs_info->chunk_root;
3164 path = btrfs_alloc_path();
3168 /* first we search for all of the device items, and then we
3169 * read in all of the chunk items. This way we can create chunk
3170 * mappings that reference all of the devices that are afound
3172 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3176 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3178 leaf = path->nodes[0];
3179 slot = path->slots[0];
3180 if (slot >= btrfs_header_nritems(leaf)) {
3181 ret = btrfs_next_leaf(root, path);
3188 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3189 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3190 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3192 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3193 struct btrfs_dev_item *dev_item;
3194 dev_item = btrfs_item_ptr(leaf, slot,
3195 struct btrfs_dev_item);
3196 ret = read_one_dev(root, leaf, dev_item);
3200 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3201 struct btrfs_chunk *chunk;
3202 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3203 ret = read_one_chunk(root, &found_key, leaf, chunk);
3209 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3211 btrfs_release_path(root, path);
3216 btrfs_free_path(path);