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 <asm/div64.h>
25 #include "extent_map.h"
27 #include "transaction.h"
28 #include "print-tree.h"
30 #include "async-thread.h"
40 struct btrfs_bio_stripe stripes[];
43 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
44 (sizeof(struct btrfs_bio_stripe) * (n)))
46 static DEFINE_MUTEX(uuid_mutex);
47 static LIST_HEAD(fs_uuids);
49 void btrfs_lock_volumes(void)
51 mutex_lock(&uuid_mutex);
54 void btrfs_unlock_volumes(void)
56 mutex_unlock(&uuid_mutex);
59 static void lock_chunks(struct btrfs_root *root)
61 mutex_lock(&root->fs_info->alloc_mutex);
62 mutex_lock(&root->fs_info->chunk_mutex);
65 static void unlock_chunks(struct btrfs_root *root)
67 mutex_unlock(&root->fs_info->chunk_mutex);
68 mutex_unlock(&root->fs_info->alloc_mutex);
71 int btrfs_cleanup_fs_uuids(void)
73 struct btrfs_fs_devices *fs_devices;
74 struct list_head *uuid_cur;
75 struct list_head *devices_cur;
76 struct btrfs_device *dev;
78 list_for_each(uuid_cur, &fs_uuids) {
79 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
81 while(!list_empty(&fs_devices->devices)) {
82 devices_cur = fs_devices->devices.next;
83 dev = list_entry(devices_cur, struct btrfs_device,
86 close_bdev_excl(dev->bdev);
87 fs_devices->open_devices--;
89 list_del(&dev->dev_list);
97 static noinline struct btrfs_device *__find_device(struct list_head *head,
100 struct btrfs_device *dev;
101 struct list_head *cur;
103 list_for_each(cur, head) {
104 dev = list_entry(cur, struct btrfs_device, dev_list);
105 if (dev->devid == devid &&
106 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
113 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 struct list_head *cur;
116 struct btrfs_fs_devices *fs_devices;
118 list_for_each(cur, &fs_uuids) {
119 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
120 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
127 * we try to collect pending bios for a device so we don't get a large
128 * number of procs sending bios down to the same device. This greatly
129 * improves the schedulers ability to collect and merge the bios.
131 * But, it also turns into a long list of bios to process and that is sure
132 * to eventually make the worker thread block. The solution here is to
133 * make some progress and then put this work struct back at the end of
134 * the list if the block device is congested. This way, multiple devices
135 * can make progress from a single worker thread.
137 static int noinline run_scheduled_bios(struct btrfs_device *device)
140 struct backing_dev_info *bdi;
141 struct btrfs_fs_info *fs_info;
145 unsigned long num_run = 0;
148 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
149 fs_info = device->dev_root->fs_info;
150 limit = btrfs_async_submit_limit(fs_info);
151 limit = limit * 2 / 3;
154 spin_lock(&device->io_lock);
156 /* take all the bios off the list at once and process them
157 * later on (without the lock held). But, remember the
158 * tail and other pointers so the bios can be properly reinserted
159 * into the list if we hit congestion
161 pending = device->pending_bios;
162 tail = device->pending_bio_tail;
163 WARN_ON(pending && !tail);
164 device->pending_bios = NULL;
165 device->pending_bio_tail = NULL;
168 * if pending was null this time around, no bios need processing
169 * at all and we can stop. Otherwise it'll loop back up again
170 * and do an additional check so no bios are missed.
172 * device->running_pending is used to synchronize with the
177 device->running_pending = 1;
180 device->running_pending = 0;
182 spin_unlock(&device->io_lock);
186 pending = pending->bi_next;
188 atomic_dec(&fs_info->nr_async_bios);
190 if (atomic_read(&fs_info->nr_async_bios) < limit &&
191 waitqueue_active(&fs_info->async_submit_wait))
192 wake_up(&fs_info->async_submit_wait);
194 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
196 submit_bio(cur->bi_rw, cur);
201 * we made progress, there is more work to do and the bdi
202 * is now congested. Back off and let other work structs
205 if (pending && bdi_write_congested(bdi)) {
206 struct bio *old_head;
208 spin_lock(&device->io_lock);
210 old_head = device->pending_bios;
211 device->pending_bios = pending;
212 if (device->pending_bio_tail)
213 tail->bi_next = old_head;
215 device->pending_bio_tail = tail;
217 spin_unlock(&device->io_lock);
218 btrfs_requeue_work(&device->work);
228 void pending_bios_fn(struct btrfs_work *work)
230 struct btrfs_device *device;
232 device = container_of(work, struct btrfs_device, work);
233 run_scheduled_bios(device);
236 static noinline int device_list_add(const char *path,
237 struct btrfs_super_block *disk_super,
238 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
240 struct btrfs_device *device;
241 struct btrfs_fs_devices *fs_devices;
242 u64 found_transid = btrfs_super_generation(disk_super);
244 fs_devices = find_fsid(disk_super->fsid);
246 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
249 INIT_LIST_HEAD(&fs_devices->devices);
250 INIT_LIST_HEAD(&fs_devices->alloc_list);
251 list_add(&fs_devices->list, &fs_uuids);
252 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
253 fs_devices->latest_devid = devid;
254 fs_devices->latest_trans = found_transid;
257 device = __find_device(&fs_devices->devices, devid,
258 disk_super->dev_item.uuid);
261 device = kzalloc(sizeof(*device), GFP_NOFS);
263 /* we can safely leave the fs_devices entry around */
266 device->devid = devid;
267 device->work.func = pending_bios_fn;
268 memcpy(device->uuid, disk_super->dev_item.uuid,
270 device->barriers = 1;
271 spin_lock_init(&device->io_lock);
272 device->name = kstrdup(path, GFP_NOFS);
277 list_add(&device->dev_list, &fs_devices->devices);
278 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
279 fs_devices->num_devices++;
282 if (found_transid > fs_devices->latest_trans) {
283 fs_devices->latest_devid = devid;
284 fs_devices->latest_trans = found_transid;
286 *fs_devices_ret = fs_devices;
290 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
292 struct list_head *head = &fs_devices->devices;
293 struct list_head *cur;
294 struct btrfs_device *device;
296 mutex_lock(&uuid_mutex);
298 list_for_each(cur, head) {
299 device = list_entry(cur, struct btrfs_device, dev_list);
300 if (!device->in_fs_metadata) {
301 struct block_device *bdev;
302 list_del(&device->dev_list);
303 list_del(&device->dev_alloc_list);
304 fs_devices->num_devices--;
307 fs_devices->open_devices--;
308 mutex_unlock(&uuid_mutex);
309 close_bdev_excl(bdev);
310 mutex_lock(&uuid_mutex);
317 mutex_unlock(&uuid_mutex);
321 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
323 struct list_head *head = &fs_devices->devices;
324 struct list_head *cur;
325 struct btrfs_device *device;
327 mutex_lock(&uuid_mutex);
328 list_for_each(cur, head) {
329 device = list_entry(cur, struct btrfs_device, dev_list);
331 close_bdev_excl(device->bdev);
332 fs_devices->open_devices--;
335 device->in_fs_metadata = 0;
337 fs_devices->mounted = 0;
338 mutex_unlock(&uuid_mutex);
342 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
343 int flags, void *holder)
345 struct block_device *bdev;
346 struct list_head *head = &fs_devices->devices;
347 struct list_head *cur;
348 struct btrfs_device *device;
349 struct block_device *latest_bdev = NULL;
350 struct buffer_head *bh;
351 struct btrfs_super_block *disk_super;
352 u64 latest_devid = 0;
353 u64 latest_transid = 0;
358 mutex_lock(&uuid_mutex);
359 if (fs_devices->mounted)
362 list_for_each(cur, head) {
363 device = list_entry(cur, struct btrfs_device, dev_list);
370 bdev = open_bdev_excl(device->name, flags, holder);
373 printk("open %s failed\n", device->name);
376 set_blocksize(bdev, 4096);
378 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
382 disk_super = (struct btrfs_super_block *)bh->b_data;
383 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
384 sizeof(disk_super->magic)))
387 devid = le64_to_cpu(disk_super->dev_item.devid);
388 if (devid != device->devid)
391 transid = btrfs_super_generation(disk_super);
392 if (!latest_transid || transid > latest_transid) {
393 latest_devid = devid;
394 latest_transid = transid;
399 device->in_fs_metadata = 0;
400 fs_devices->open_devices++;
406 close_bdev_excl(bdev);
410 if (fs_devices->open_devices == 0) {
414 fs_devices->mounted = 1;
415 fs_devices->latest_bdev = latest_bdev;
416 fs_devices->latest_devid = latest_devid;
417 fs_devices->latest_trans = latest_transid;
419 mutex_unlock(&uuid_mutex);
423 int btrfs_scan_one_device(const char *path, int flags, void *holder,
424 struct btrfs_fs_devices **fs_devices_ret)
426 struct btrfs_super_block *disk_super;
427 struct block_device *bdev;
428 struct buffer_head *bh;
433 mutex_lock(&uuid_mutex);
435 bdev = open_bdev_excl(path, flags, holder);
442 ret = set_blocksize(bdev, 4096);
445 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
450 disk_super = (struct btrfs_super_block *)bh->b_data;
451 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
452 sizeof(disk_super->magic))) {
456 devid = le64_to_cpu(disk_super->dev_item.devid);
457 transid = btrfs_super_generation(disk_super);
458 if (disk_super->label[0])
459 printk("device label %s ", disk_super->label);
461 /* FIXME, make a readl uuid parser */
462 printk("device fsid %llx-%llx ",
463 *(unsigned long long *)disk_super->fsid,
464 *(unsigned long long *)(disk_super->fsid + 8));
466 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
467 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
472 close_bdev_excl(bdev);
474 mutex_unlock(&uuid_mutex);
479 * this uses a pretty simple search, the expectation is that it is
480 * called very infrequently and that a given device has a small number
483 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
484 struct btrfs_device *device,
485 struct btrfs_path *path,
486 u64 num_bytes, u64 *start)
488 struct btrfs_key key;
489 struct btrfs_root *root = device->dev_root;
490 struct btrfs_dev_extent *dev_extent = NULL;
493 u64 search_start = 0;
494 u64 search_end = device->total_bytes;
498 struct extent_buffer *l;
503 /* FIXME use last free of some kind */
505 /* we don't want to overwrite the superblock on the drive,
506 * so we make sure to start at an offset of at least 1MB
508 search_start = max((u64)1024 * 1024, search_start);
510 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
511 search_start = max(root->fs_info->alloc_start, search_start);
513 key.objectid = device->devid;
514 key.offset = search_start;
515 key.type = BTRFS_DEV_EXTENT_KEY;
516 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
519 ret = btrfs_previous_item(root, path, 0, key.type);
523 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
526 slot = path->slots[0];
527 if (slot >= btrfs_header_nritems(l)) {
528 ret = btrfs_next_leaf(root, path);
535 if (search_start >= search_end) {
539 *start = search_start;
543 *start = last_byte > search_start ?
544 last_byte : search_start;
545 if (search_end <= *start) {
551 btrfs_item_key_to_cpu(l, &key, slot);
553 if (key.objectid < device->devid)
556 if (key.objectid > device->devid)
559 if (key.offset >= search_start && key.offset > last_byte &&
561 if (last_byte < search_start)
562 last_byte = search_start;
563 hole_size = key.offset - last_byte;
564 if (key.offset > last_byte &&
565 hole_size >= num_bytes) {
570 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
575 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
576 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
582 /* we have to make sure we didn't find an extent that has already
583 * been allocated by the map tree or the original allocation
585 btrfs_release_path(root, path);
586 BUG_ON(*start < search_start);
588 if (*start + num_bytes > search_end) {
592 /* check for pending inserts here */
596 btrfs_release_path(root, path);
600 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
601 struct btrfs_device *device,
605 struct btrfs_path *path;
606 struct btrfs_root *root = device->dev_root;
607 struct btrfs_key key;
608 struct btrfs_key found_key;
609 struct extent_buffer *leaf = NULL;
610 struct btrfs_dev_extent *extent = NULL;
612 path = btrfs_alloc_path();
616 key.objectid = device->devid;
618 key.type = BTRFS_DEV_EXTENT_KEY;
620 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
622 ret = btrfs_previous_item(root, path, key.objectid,
623 BTRFS_DEV_EXTENT_KEY);
625 leaf = path->nodes[0];
626 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
627 extent = btrfs_item_ptr(leaf, path->slots[0],
628 struct btrfs_dev_extent);
629 BUG_ON(found_key.offset > start || found_key.offset +
630 btrfs_dev_extent_length(leaf, extent) < start);
632 } else if (ret == 0) {
633 leaf = path->nodes[0];
634 extent = btrfs_item_ptr(leaf, path->slots[0],
635 struct btrfs_dev_extent);
639 if (device->bytes_used > 0)
640 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
641 ret = btrfs_del_item(trans, root, path);
644 btrfs_free_path(path);
648 int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
649 struct btrfs_device *device,
650 u64 chunk_tree, u64 chunk_objectid,
652 u64 num_bytes, u64 *start)
655 struct btrfs_path *path;
656 struct btrfs_root *root = device->dev_root;
657 struct btrfs_dev_extent *extent;
658 struct extent_buffer *leaf;
659 struct btrfs_key key;
661 WARN_ON(!device->in_fs_metadata);
662 path = btrfs_alloc_path();
666 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
671 key.objectid = device->devid;
673 key.type = BTRFS_DEV_EXTENT_KEY;
674 ret = btrfs_insert_empty_item(trans, root, path, &key,
678 leaf = path->nodes[0];
679 extent = btrfs_item_ptr(leaf, path->slots[0],
680 struct btrfs_dev_extent);
681 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
682 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
683 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
685 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
686 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
689 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
690 btrfs_mark_buffer_dirty(leaf);
692 btrfs_free_path(path);
696 static noinline int find_next_chunk(struct btrfs_root *root,
697 u64 objectid, u64 *offset)
699 struct btrfs_path *path;
701 struct btrfs_key key;
702 struct btrfs_chunk *chunk;
703 struct btrfs_key found_key;
705 path = btrfs_alloc_path();
708 key.objectid = objectid;
709 key.offset = (u64)-1;
710 key.type = BTRFS_CHUNK_ITEM_KEY;
712 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
718 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
722 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
724 if (found_key.objectid != objectid)
727 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
729 *offset = found_key.offset +
730 btrfs_chunk_length(path->nodes[0], chunk);
735 btrfs_free_path(path);
739 static noinline int find_next_devid(struct btrfs_root *root,
740 struct btrfs_path *path, u64 *objectid)
743 struct btrfs_key key;
744 struct btrfs_key found_key;
746 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
747 key.type = BTRFS_DEV_ITEM_KEY;
748 key.offset = (u64)-1;
750 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
756 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
761 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
763 *objectid = found_key.offset + 1;
767 btrfs_release_path(root, path);
772 * the device information is stored in the chunk root
773 * the btrfs_device struct should be fully filled in
775 int btrfs_add_device(struct btrfs_trans_handle *trans,
776 struct btrfs_root *root,
777 struct btrfs_device *device)
780 struct btrfs_path *path;
781 struct btrfs_dev_item *dev_item;
782 struct extent_buffer *leaf;
783 struct btrfs_key key;
787 root = root->fs_info->chunk_root;
789 path = btrfs_alloc_path();
793 ret = find_next_devid(root, path, &free_devid);
797 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
798 key.type = BTRFS_DEV_ITEM_KEY;
799 key.offset = free_devid;
801 ret = btrfs_insert_empty_item(trans, root, path, &key,
806 leaf = path->nodes[0];
807 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
809 device->devid = free_devid;
810 btrfs_set_device_id(leaf, dev_item, device->devid);
811 btrfs_set_device_type(leaf, dev_item, device->type);
812 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
813 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
814 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
815 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
816 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
817 btrfs_set_device_group(leaf, dev_item, 0);
818 btrfs_set_device_seek_speed(leaf, dev_item, 0);
819 btrfs_set_device_bandwidth(leaf, dev_item, 0);
821 ptr = (unsigned long)btrfs_device_uuid(dev_item);
822 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
823 btrfs_mark_buffer_dirty(leaf);
827 btrfs_free_path(path);
831 static int btrfs_rm_dev_item(struct btrfs_root *root,
832 struct btrfs_device *device)
835 struct btrfs_path *path;
836 struct block_device *bdev = device->bdev;
837 struct btrfs_device *next_dev;
838 struct btrfs_key key;
840 struct btrfs_fs_devices *fs_devices;
841 struct btrfs_trans_handle *trans;
843 root = root->fs_info->chunk_root;
845 path = btrfs_alloc_path();
849 trans = btrfs_start_transaction(root, 1);
850 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
851 key.type = BTRFS_DEV_ITEM_KEY;
852 key.offset = device->devid;
855 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
864 ret = btrfs_del_item(trans, root, path);
869 * at this point, the device is zero sized. We want to
870 * remove it from the devices list and zero out the old super
872 list_del_init(&device->dev_list);
873 list_del_init(&device->dev_alloc_list);
874 fs_devices = root->fs_info->fs_devices;
876 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
878 if (bdev == root->fs_info->sb->s_bdev)
879 root->fs_info->sb->s_bdev = next_dev->bdev;
880 if (bdev == fs_devices->latest_bdev)
881 fs_devices->latest_bdev = next_dev->bdev;
883 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
884 btrfs_set_super_num_devices(&root->fs_info->super_copy,
887 btrfs_free_path(path);
889 btrfs_commit_transaction(trans, root);
893 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
895 struct btrfs_device *device;
896 struct block_device *bdev;
897 struct buffer_head *bh = NULL;
898 struct btrfs_super_block *disk_super;
903 mutex_lock(&uuid_mutex);
904 mutex_lock(&root->fs_info->volume_mutex);
906 all_avail = root->fs_info->avail_data_alloc_bits |
907 root->fs_info->avail_system_alloc_bits |
908 root->fs_info->avail_metadata_alloc_bits;
910 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
911 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
912 printk("btrfs: unable to go below four devices on raid10\n");
917 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
918 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
919 printk("btrfs: unable to go below two devices on raid1\n");
924 if (strcmp(device_path, "missing") == 0) {
925 struct list_head *cur;
926 struct list_head *devices;
927 struct btrfs_device *tmp;
930 devices = &root->fs_info->fs_devices->devices;
931 list_for_each(cur, devices) {
932 tmp = list_entry(cur, struct btrfs_device, dev_list);
933 if (tmp->in_fs_metadata && !tmp->bdev) {
942 printk("btrfs: no missing devices found to remove\n");
947 bdev = open_bdev_excl(device_path, 0,
948 root->fs_info->bdev_holder);
954 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
959 disk_super = (struct btrfs_super_block *)bh->b_data;
960 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
961 sizeof(disk_super->magic))) {
965 if (memcmp(disk_super->fsid, root->fs_info->fsid,
970 devid = le64_to_cpu(disk_super->dev_item.devid);
971 device = btrfs_find_device(root, devid, NULL);
978 root->fs_info->fs_devices->num_devices--;
979 root->fs_info->fs_devices->open_devices--;
981 ret = btrfs_shrink_device(device, 0);
986 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
991 /* make sure this device isn't detected as part of
994 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
995 set_buffer_dirty(bh);
996 sync_dirty_buffer(bh);
1002 /* one close for the device struct or super_block */
1003 close_bdev_excl(device->bdev);
1006 /* one close for us */
1007 close_bdev_excl(bdev);
1009 kfree(device->name);
1018 close_bdev_excl(bdev);
1020 mutex_unlock(&root->fs_info->volume_mutex);
1021 mutex_unlock(&uuid_mutex);
1025 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1027 struct btrfs_trans_handle *trans;
1028 struct btrfs_device *device;
1029 struct block_device *bdev;
1030 struct list_head *cur;
1031 struct list_head *devices;
1036 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1041 mutex_lock(&root->fs_info->volume_mutex);
1043 trans = btrfs_start_transaction(root, 1);
1045 devices = &root->fs_info->fs_devices->devices;
1046 list_for_each(cur, devices) {
1047 device = list_entry(cur, struct btrfs_device, dev_list);
1048 if (device->bdev == bdev) {
1054 device = kzalloc(sizeof(*device), GFP_NOFS);
1056 /* we can safely leave the fs_devices entry around */
1058 goto out_close_bdev;
1061 device->barriers = 1;
1062 device->work.func = pending_bios_fn;
1063 generate_random_uuid(device->uuid);
1064 spin_lock_init(&device->io_lock);
1065 device->name = kstrdup(device_path, GFP_NOFS);
1066 if (!device->name) {
1068 goto out_close_bdev;
1070 device->io_width = root->sectorsize;
1071 device->io_align = root->sectorsize;
1072 device->sector_size = root->sectorsize;
1073 device->total_bytes = i_size_read(bdev->bd_inode);
1074 device->dev_root = root->fs_info->dev_root;
1075 device->bdev = bdev;
1076 device->in_fs_metadata = 1;
1078 ret = btrfs_add_device(trans, root, device);
1080 goto out_close_bdev;
1082 set_blocksize(device->bdev, 4096);
1084 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1085 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1086 total_bytes + device->total_bytes);
1088 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1089 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1092 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1093 list_add(&device->dev_alloc_list,
1094 &root->fs_info->fs_devices->alloc_list);
1095 root->fs_info->fs_devices->num_devices++;
1096 root->fs_info->fs_devices->open_devices++;
1098 unlock_chunks(root);
1099 btrfs_end_transaction(trans, root);
1100 mutex_unlock(&root->fs_info->volume_mutex);
1105 close_bdev_excl(bdev);
1109 int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1110 struct btrfs_device *device)
1113 struct btrfs_path *path;
1114 struct btrfs_root *root;
1115 struct btrfs_dev_item *dev_item;
1116 struct extent_buffer *leaf;
1117 struct btrfs_key key;
1119 root = device->dev_root->fs_info->chunk_root;
1121 path = btrfs_alloc_path();
1125 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1126 key.type = BTRFS_DEV_ITEM_KEY;
1127 key.offset = device->devid;
1129 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1138 leaf = path->nodes[0];
1139 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1141 btrfs_set_device_id(leaf, dev_item, device->devid);
1142 btrfs_set_device_type(leaf, dev_item, device->type);
1143 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1144 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1145 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1146 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1147 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1148 btrfs_mark_buffer_dirty(leaf);
1151 btrfs_free_path(path);
1155 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1156 struct btrfs_device *device, u64 new_size)
1158 struct btrfs_super_block *super_copy =
1159 &device->dev_root->fs_info->super_copy;
1160 u64 old_total = btrfs_super_total_bytes(super_copy);
1161 u64 diff = new_size - device->total_bytes;
1163 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1164 return btrfs_update_device(trans, device);
1167 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1168 struct btrfs_device *device, u64 new_size)
1171 lock_chunks(device->dev_root);
1172 ret = __btrfs_grow_device(trans, device, new_size);
1173 unlock_chunks(device->dev_root);
1177 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1178 struct btrfs_root *root,
1179 u64 chunk_tree, u64 chunk_objectid,
1183 struct btrfs_path *path;
1184 struct btrfs_key key;
1186 root = root->fs_info->chunk_root;
1187 path = btrfs_alloc_path();
1191 key.objectid = chunk_objectid;
1192 key.offset = chunk_offset;
1193 key.type = BTRFS_CHUNK_ITEM_KEY;
1195 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1198 ret = btrfs_del_item(trans, root, path);
1201 btrfs_free_path(path);
1205 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1208 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1209 struct btrfs_disk_key *disk_key;
1210 struct btrfs_chunk *chunk;
1217 struct btrfs_key key;
1219 array_size = btrfs_super_sys_array_size(super_copy);
1221 ptr = super_copy->sys_chunk_array;
1224 while (cur < array_size) {
1225 disk_key = (struct btrfs_disk_key *)ptr;
1226 btrfs_disk_key_to_cpu(&key, disk_key);
1228 len = sizeof(*disk_key);
1230 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1231 chunk = (struct btrfs_chunk *)(ptr + len);
1232 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1233 len += btrfs_chunk_item_size(num_stripes);
1238 if (key.objectid == chunk_objectid &&
1239 key.offset == chunk_offset) {
1240 memmove(ptr, ptr + len, array_size - (cur + len));
1242 btrfs_set_super_sys_array_size(super_copy, array_size);
1252 int btrfs_relocate_chunk(struct btrfs_root *root,
1253 u64 chunk_tree, u64 chunk_objectid,
1256 struct extent_map_tree *em_tree;
1257 struct btrfs_root *extent_root;
1258 struct btrfs_trans_handle *trans;
1259 struct extent_map *em;
1260 struct map_lookup *map;
1264 printk("btrfs relocating chunk %llu\n",
1265 (unsigned long long)chunk_offset);
1266 root = root->fs_info->chunk_root;
1267 extent_root = root->fs_info->extent_root;
1268 em_tree = &root->fs_info->mapping_tree.map_tree;
1270 /* step one, relocate all the extents inside this chunk */
1271 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1274 trans = btrfs_start_transaction(root, 1);
1280 * step two, delete the device extents and the
1281 * chunk tree entries
1283 spin_lock(&em_tree->lock);
1284 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1285 spin_unlock(&em_tree->lock);
1287 BUG_ON(em->start > chunk_offset ||
1288 em->start + em->len < chunk_offset);
1289 map = (struct map_lookup *)em->bdev;
1291 for (i = 0; i < map->num_stripes; i++) {
1292 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1293 map->stripes[i].physical);
1296 if (map->stripes[i].dev) {
1297 ret = btrfs_update_device(trans, map->stripes[i].dev);
1301 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1306 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1307 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1311 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1314 spin_lock(&em_tree->lock);
1315 remove_extent_mapping(em_tree, em);
1316 spin_unlock(&em_tree->lock);
1321 /* once for the tree */
1322 free_extent_map(em);
1324 free_extent_map(em);
1326 unlock_chunks(root);
1327 btrfs_end_transaction(trans, root);
1331 static u64 div_factor(u64 num, int factor)
1341 int btrfs_balance(struct btrfs_root *dev_root)
1344 struct list_head *cur;
1345 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1346 struct btrfs_device *device;
1349 struct btrfs_path *path;
1350 struct btrfs_key key;
1351 struct btrfs_chunk *chunk;
1352 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1353 struct btrfs_trans_handle *trans;
1354 struct btrfs_key found_key;
1357 mutex_lock(&dev_root->fs_info->volume_mutex);
1358 dev_root = dev_root->fs_info->dev_root;
1360 /* step one make some room on all the devices */
1361 list_for_each(cur, devices) {
1362 device = list_entry(cur, struct btrfs_device, dev_list);
1363 old_size = device->total_bytes;
1364 size_to_free = div_factor(old_size, 1);
1365 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1366 if (device->total_bytes - device->bytes_used > size_to_free)
1369 ret = btrfs_shrink_device(device, old_size - size_to_free);
1372 trans = btrfs_start_transaction(dev_root, 1);
1375 ret = btrfs_grow_device(trans, device, old_size);
1378 btrfs_end_transaction(trans, dev_root);
1381 /* step two, relocate all the chunks */
1382 path = btrfs_alloc_path();
1385 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1386 key.offset = (u64)-1;
1387 key.type = BTRFS_CHUNK_ITEM_KEY;
1390 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1395 * this shouldn't happen, it means the last relocate
1401 ret = btrfs_previous_item(chunk_root, path, 0,
1402 BTRFS_CHUNK_ITEM_KEY);
1406 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1408 if (found_key.objectid != key.objectid)
1411 chunk = btrfs_item_ptr(path->nodes[0],
1413 struct btrfs_chunk);
1414 key.offset = found_key.offset;
1415 /* chunk zero is special */
1416 if (key.offset == 0)
1419 btrfs_release_path(chunk_root, path);
1420 ret = btrfs_relocate_chunk(chunk_root,
1421 chunk_root->root_key.objectid,
1428 btrfs_free_path(path);
1429 mutex_unlock(&dev_root->fs_info->volume_mutex);
1434 * shrinking a device means finding all of the device extents past
1435 * the new size, and then following the back refs to the chunks.
1436 * The chunk relocation code actually frees the device extent
1438 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1440 struct btrfs_trans_handle *trans;
1441 struct btrfs_root *root = device->dev_root;
1442 struct btrfs_dev_extent *dev_extent = NULL;
1443 struct btrfs_path *path;
1450 struct extent_buffer *l;
1451 struct btrfs_key key;
1452 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1453 u64 old_total = btrfs_super_total_bytes(super_copy);
1454 u64 diff = device->total_bytes - new_size;
1457 path = btrfs_alloc_path();
1461 trans = btrfs_start_transaction(root, 1);
1471 device->total_bytes = new_size;
1472 ret = btrfs_update_device(trans, device);
1474 unlock_chunks(root);
1475 btrfs_end_transaction(trans, root);
1478 WARN_ON(diff > old_total);
1479 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1480 unlock_chunks(root);
1481 btrfs_end_transaction(trans, root);
1483 key.objectid = device->devid;
1484 key.offset = (u64)-1;
1485 key.type = BTRFS_DEV_EXTENT_KEY;
1488 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1492 ret = btrfs_previous_item(root, path, 0, key.type);
1501 slot = path->slots[0];
1502 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1504 if (key.objectid != device->devid)
1507 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1508 length = btrfs_dev_extent_length(l, dev_extent);
1510 if (key.offset + length <= new_size)
1513 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1514 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1515 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1516 btrfs_release_path(root, path);
1518 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1525 btrfs_free_path(path);
1529 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1530 struct btrfs_root *root,
1531 struct btrfs_key *key,
1532 struct btrfs_chunk *chunk, int item_size)
1534 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1535 struct btrfs_disk_key disk_key;
1539 array_size = btrfs_super_sys_array_size(super_copy);
1540 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1543 ptr = super_copy->sys_chunk_array + array_size;
1544 btrfs_cpu_key_to_disk(&disk_key, key);
1545 memcpy(ptr, &disk_key, sizeof(disk_key));
1546 ptr += sizeof(disk_key);
1547 memcpy(ptr, chunk, item_size);
1548 item_size += sizeof(disk_key);
1549 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1553 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1554 int num_stripes, int sub_stripes)
1556 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1558 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1559 return calc_size * (num_stripes / sub_stripes);
1561 return calc_size * num_stripes;
1565 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1566 struct btrfs_root *extent_root, u64 *start,
1567 u64 *num_bytes, u64 type)
1570 struct btrfs_fs_info *info = extent_root->fs_info;
1571 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1572 struct btrfs_path *path;
1573 struct btrfs_stripe *stripes;
1574 struct btrfs_device *device = NULL;
1575 struct btrfs_chunk *chunk;
1576 struct list_head private_devs;
1577 struct list_head *dev_list;
1578 struct list_head *cur;
1579 struct extent_map_tree *em_tree;
1580 struct map_lookup *map;
1581 struct extent_map *em;
1582 int min_stripe_size = 1 * 1024 * 1024;
1584 u64 calc_size = 1024 * 1024 * 1024;
1585 u64 max_chunk_size = calc_size;
1590 int num_stripes = 1;
1591 int min_stripes = 1;
1592 int sub_stripes = 0;
1596 int stripe_len = 64 * 1024;
1597 struct btrfs_key key;
1599 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1600 (type & BTRFS_BLOCK_GROUP_DUP)) {
1602 type &= ~BTRFS_BLOCK_GROUP_DUP;
1604 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1605 if (list_empty(dev_list))
1608 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1609 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1612 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1616 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1617 num_stripes = min_t(u64, 2,
1618 extent_root->fs_info->fs_devices->open_devices);
1619 if (num_stripes < 2)
1623 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1624 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1625 if (num_stripes < 4)
1627 num_stripes &= ~(u32)1;
1632 if (type & BTRFS_BLOCK_GROUP_DATA) {
1633 max_chunk_size = 10 * calc_size;
1634 min_stripe_size = 64 * 1024 * 1024;
1635 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1636 max_chunk_size = 4 * calc_size;
1637 min_stripe_size = 32 * 1024 * 1024;
1638 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1639 calc_size = 8 * 1024 * 1024;
1640 max_chunk_size = calc_size * 2;
1641 min_stripe_size = 1 * 1024 * 1024;
1644 path = btrfs_alloc_path();
1648 /* we don't want a chunk larger than 10% of the FS */
1649 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1650 max_chunk_size = min(percent_max, max_chunk_size);
1653 if (calc_size * num_stripes > max_chunk_size) {
1654 calc_size = max_chunk_size;
1655 do_div(calc_size, num_stripes);
1656 do_div(calc_size, stripe_len);
1657 calc_size *= stripe_len;
1659 /* we don't want tiny stripes */
1660 calc_size = max_t(u64, min_stripe_size, calc_size);
1662 do_div(calc_size, stripe_len);
1663 calc_size *= stripe_len;
1665 INIT_LIST_HEAD(&private_devs);
1666 cur = dev_list->next;
1669 if (type & BTRFS_BLOCK_GROUP_DUP)
1670 min_free = calc_size * 2;
1672 min_free = calc_size;
1675 * we add 1MB because we never use the first 1MB of the device, unless
1676 * we've looped, then we are likely allocating the maximum amount of
1677 * space left already
1680 min_free += 1024 * 1024;
1682 /* build a private list of devices we will allocate from */
1683 while(index < num_stripes) {
1684 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1686 if (device->total_bytes > device->bytes_used)
1687 avail = device->total_bytes - device->bytes_used;
1692 if (device->in_fs_metadata && avail >= min_free) {
1693 u64 ignored_start = 0;
1694 ret = find_free_dev_extent(trans, device, path,
1698 list_move_tail(&device->dev_alloc_list,
1701 if (type & BTRFS_BLOCK_GROUP_DUP)
1704 } else if (device->in_fs_metadata && avail > max_avail)
1706 if (cur == dev_list)
1709 if (index < num_stripes) {
1710 list_splice(&private_devs, dev_list);
1711 if (index >= min_stripes) {
1712 num_stripes = index;
1713 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1714 num_stripes /= sub_stripes;
1715 num_stripes *= sub_stripes;
1720 if (!looped && max_avail > 0) {
1722 calc_size = max_avail;
1725 btrfs_free_path(path);
1728 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1729 key.type = BTRFS_CHUNK_ITEM_KEY;
1730 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1733 btrfs_free_path(path);
1737 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1739 btrfs_free_path(path);
1743 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1746 btrfs_free_path(path);
1749 btrfs_free_path(path);
1752 stripes = &chunk->stripe;
1753 *num_bytes = chunk_bytes_by_type(type, calc_size,
1754 num_stripes, sub_stripes);
1757 while(index < num_stripes) {
1758 struct btrfs_stripe *stripe;
1759 BUG_ON(list_empty(&private_devs));
1760 cur = private_devs.next;
1761 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1763 /* loop over this device again if we're doing a dup group */
1764 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1765 (index == num_stripes - 1))
1766 list_move_tail(&device->dev_alloc_list, dev_list);
1768 ret = btrfs_alloc_dev_extent(trans, device,
1769 info->chunk_root->root_key.objectid,
1770 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1771 calc_size, &dev_offset);
1773 device->bytes_used += calc_size;
1774 ret = btrfs_update_device(trans, device);
1777 map->stripes[index].dev = device;
1778 map->stripes[index].physical = dev_offset;
1779 stripe = stripes + index;
1780 btrfs_set_stack_stripe_devid(stripe, device->devid);
1781 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1782 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1783 physical = dev_offset;
1786 BUG_ON(!list_empty(&private_devs));
1788 /* key was set above */
1789 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1790 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1791 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1792 btrfs_set_stack_chunk_type(chunk, type);
1793 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1794 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1795 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1796 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1797 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1798 map->sector_size = extent_root->sectorsize;
1799 map->stripe_len = stripe_len;
1800 map->io_align = stripe_len;
1801 map->io_width = stripe_len;
1803 map->num_stripes = num_stripes;
1804 map->sub_stripes = sub_stripes;
1806 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1807 btrfs_chunk_item_size(num_stripes));
1809 *start = key.offset;;
1811 em = alloc_extent_map(GFP_NOFS);
1814 em->bdev = (struct block_device *)map;
1815 em->start = key.offset;
1816 em->len = *num_bytes;
1817 em->block_start = 0;
1819 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1820 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1821 chunk, btrfs_chunk_item_size(num_stripes));
1826 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1827 spin_lock(&em_tree->lock);
1828 ret = add_extent_mapping(em_tree, em);
1829 spin_unlock(&em_tree->lock);
1831 free_extent_map(em);
1835 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1837 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1840 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1842 struct extent_map *em;
1845 spin_lock(&tree->map_tree.lock);
1846 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1848 remove_extent_mapping(&tree->map_tree, em);
1849 spin_unlock(&tree->map_tree.lock);
1854 free_extent_map(em);
1855 /* once for the tree */
1856 free_extent_map(em);
1860 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1862 struct extent_map *em;
1863 struct map_lookup *map;
1864 struct extent_map_tree *em_tree = &map_tree->map_tree;
1867 spin_lock(&em_tree->lock);
1868 em = lookup_extent_mapping(em_tree, logical, len);
1869 spin_unlock(&em_tree->lock);
1872 BUG_ON(em->start > logical || em->start + em->len < logical);
1873 map = (struct map_lookup *)em->bdev;
1874 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1875 ret = map->num_stripes;
1876 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1877 ret = map->sub_stripes;
1880 free_extent_map(em);
1884 static int find_live_mirror(struct map_lookup *map, int first, int num,
1888 if (map->stripes[optimal].dev->bdev)
1890 for (i = first; i < first + num; i++) {
1891 if (map->stripes[i].dev->bdev)
1894 /* we couldn't find one that doesn't fail. Just return something
1895 * and the io error handling code will clean up eventually
1900 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1901 u64 logical, u64 *length,
1902 struct btrfs_multi_bio **multi_ret,
1903 int mirror_num, struct page *unplug_page)
1905 struct extent_map *em;
1906 struct map_lookup *map;
1907 struct extent_map_tree *em_tree = &map_tree->map_tree;
1911 int stripes_allocated = 8;
1912 int stripes_required = 1;
1917 struct btrfs_multi_bio *multi = NULL;
1919 if (multi_ret && !(rw & (1 << BIO_RW))) {
1920 stripes_allocated = 1;
1924 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1929 atomic_set(&multi->error, 0);
1932 spin_lock(&em_tree->lock);
1933 em = lookup_extent_mapping(em_tree, logical, *length);
1934 spin_unlock(&em_tree->lock);
1936 if (!em && unplug_page)
1940 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1944 BUG_ON(em->start > logical || em->start + em->len < logical);
1945 map = (struct map_lookup *)em->bdev;
1946 offset = logical - em->start;
1948 if (mirror_num > map->num_stripes)
1951 /* if our multi bio struct is too small, back off and try again */
1952 if (rw & (1 << BIO_RW)) {
1953 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1954 BTRFS_BLOCK_GROUP_DUP)) {
1955 stripes_required = map->num_stripes;
1957 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1958 stripes_required = map->sub_stripes;
1962 if (multi_ret && rw == WRITE &&
1963 stripes_allocated < stripes_required) {
1964 stripes_allocated = map->num_stripes;
1965 free_extent_map(em);
1971 * stripe_nr counts the total number of stripes we have to stride
1972 * to get to this block
1974 do_div(stripe_nr, map->stripe_len);
1976 stripe_offset = stripe_nr * map->stripe_len;
1977 BUG_ON(offset < stripe_offset);
1979 /* stripe_offset is the offset of this block in its stripe*/
1980 stripe_offset = offset - stripe_offset;
1982 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1983 BTRFS_BLOCK_GROUP_RAID10 |
1984 BTRFS_BLOCK_GROUP_DUP)) {
1985 /* we limit the length of each bio to what fits in a stripe */
1986 *length = min_t(u64, em->len - offset,
1987 map->stripe_len - stripe_offset);
1989 *length = em->len - offset;
1992 if (!multi_ret && !unplug_page)
1997 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1998 if (unplug_page || (rw & (1 << BIO_RW)))
1999 num_stripes = map->num_stripes;
2000 else if (mirror_num)
2001 stripe_index = mirror_num - 1;
2003 stripe_index = find_live_mirror(map, 0,
2005 current->pid % map->num_stripes);
2008 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2009 if (rw & (1 << BIO_RW))
2010 num_stripes = map->num_stripes;
2011 else if (mirror_num)
2012 stripe_index = mirror_num - 1;
2014 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2015 int factor = map->num_stripes / map->sub_stripes;
2017 stripe_index = do_div(stripe_nr, factor);
2018 stripe_index *= map->sub_stripes;
2020 if (unplug_page || (rw & (1 << BIO_RW)))
2021 num_stripes = map->sub_stripes;
2022 else if (mirror_num)
2023 stripe_index += mirror_num - 1;
2025 stripe_index = find_live_mirror(map, stripe_index,
2026 map->sub_stripes, stripe_index +
2027 current->pid % map->sub_stripes);
2031 * after this do_div call, stripe_nr is the number of stripes
2032 * on this device we have to walk to find the data, and
2033 * stripe_index is the number of our device in the stripe array
2035 stripe_index = do_div(stripe_nr, map->num_stripes);
2037 BUG_ON(stripe_index >= map->num_stripes);
2039 for (i = 0; i < num_stripes; i++) {
2041 struct btrfs_device *device;
2042 struct backing_dev_info *bdi;
2044 device = map->stripes[stripe_index].dev;
2046 bdi = blk_get_backing_dev_info(device->bdev);
2047 if (bdi->unplug_io_fn) {
2048 bdi->unplug_io_fn(bdi, unplug_page);
2052 multi->stripes[i].physical =
2053 map->stripes[stripe_index].physical +
2054 stripe_offset + stripe_nr * map->stripe_len;
2055 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2061 multi->num_stripes = num_stripes;
2062 multi->max_errors = max_errors;
2065 free_extent_map(em);
2069 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2070 u64 logical, u64 *length,
2071 struct btrfs_multi_bio **multi_ret, int mirror_num)
2073 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2077 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2078 u64 logical, struct page *page)
2080 u64 length = PAGE_CACHE_SIZE;
2081 return __btrfs_map_block(map_tree, READ, logical, &length,
2086 static void end_bio_multi_stripe(struct bio *bio, int err)
2088 struct btrfs_multi_bio *multi = bio->bi_private;
2089 int is_orig_bio = 0;
2092 atomic_inc(&multi->error);
2094 if (bio == multi->orig_bio)
2097 if (atomic_dec_and_test(&multi->stripes_pending)) {
2100 bio = multi->orig_bio;
2102 bio->bi_private = multi->private;
2103 bio->bi_end_io = multi->end_io;
2104 /* only send an error to the higher layers if it is
2105 * beyond the tolerance of the multi-bio
2107 if (atomic_read(&multi->error) > multi->max_errors) {
2111 * this bio is actually up to date, we didn't
2112 * go over the max number of errors
2114 set_bit(BIO_UPTODATE, &bio->bi_flags);
2119 bio_endio(bio, err);
2120 } else if (!is_orig_bio) {
2125 struct async_sched {
2128 struct btrfs_fs_info *info;
2129 struct btrfs_work work;
2133 * see run_scheduled_bios for a description of why bios are collected for
2136 * This will add one bio to the pending list for a device and make sure
2137 * the work struct is scheduled.
2139 static int noinline schedule_bio(struct btrfs_root *root,
2140 struct btrfs_device *device,
2141 int rw, struct bio *bio)
2143 int should_queue = 1;
2145 /* don't bother with additional async steps for reads, right now */
2146 if (!(rw & (1 << BIO_RW))) {
2148 submit_bio(rw, bio);
2154 * nr_async_bios allows us to reliably return congestion to the
2155 * higher layers. Otherwise, the async bio makes it appear we have
2156 * made progress against dirty pages when we've really just put it
2157 * on a queue for later
2159 atomic_inc(&root->fs_info->nr_async_bios);
2160 WARN_ON(bio->bi_next);
2161 bio->bi_next = NULL;
2164 spin_lock(&device->io_lock);
2166 if (device->pending_bio_tail)
2167 device->pending_bio_tail->bi_next = bio;
2169 device->pending_bio_tail = bio;
2170 if (!device->pending_bios)
2171 device->pending_bios = bio;
2172 if (device->running_pending)
2175 spin_unlock(&device->io_lock);
2178 btrfs_queue_worker(&root->fs_info->submit_workers,
2183 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2184 int mirror_num, int async_submit)
2186 struct btrfs_mapping_tree *map_tree;
2187 struct btrfs_device *dev;
2188 struct bio *first_bio = bio;
2189 u64 logical = bio->bi_sector << 9;
2192 struct btrfs_multi_bio *multi = NULL;
2197 length = bio->bi_size;
2198 map_tree = &root->fs_info->mapping_tree;
2199 map_length = length;
2201 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2205 total_devs = multi->num_stripes;
2206 if (map_length < length) {
2207 printk("mapping failed logical %Lu bio len %Lu "
2208 "len %Lu\n", logical, length, map_length);
2211 multi->end_io = first_bio->bi_end_io;
2212 multi->private = first_bio->bi_private;
2213 multi->orig_bio = first_bio;
2214 atomic_set(&multi->stripes_pending, multi->num_stripes);
2216 while(dev_nr < total_devs) {
2217 if (total_devs > 1) {
2218 if (dev_nr < total_devs - 1) {
2219 bio = bio_clone(first_bio, GFP_NOFS);
2224 bio->bi_private = multi;
2225 bio->bi_end_io = end_bio_multi_stripe;
2227 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2228 dev = multi->stripes[dev_nr].dev;
2229 if (dev && dev->bdev) {
2230 bio->bi_bdev = dev->bdev;
2232 schedule_bio(root, dev, rw, bio);
2234 submit_bio(rw, bio);
2236 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2237 bio->bi_sector = logical >> 9;
2238 bio_endio(bio, -EIO);
2242 if (total_devs == 1)
2247 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2250 struct list_head *head = &root->fs_info->fs_devices->devices;
2252 return __find_device(head, devid, uuid);
2255 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2256 u64 devid, u8 *dev_uuid)
2258 struct btrfs_device *device;
2259 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2261 device = kzalloc(sizeof(*device), GFP_NOFS);
2262 list_add(&device->dev_list,
2263 &fs_devices->devices);
2264 list_add(&device->dev_alloc_list,
2265 &fs_devices->alloc_list);
2266 device->barriers = 1;
2267 device->dev_root = root->fs_info->dev_root;
2268 device->devid = devid;
2269 device->work.func = pending_bios_fn;
2270 fs_devices->num_devices++;
2271 spin_lock_init(&device->io_lock);
2272 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2277 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2278 struct extent_buffer *leaf,
2279 struct btrfs_chunk *chunk)
2281 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2282 struct map_lookup *map;
2283 struct extent_map *em;
2287 u8 uuid[BTRFS_UUID_SIZE];
2292 logical = key->offset;
2293 length = btrfs_chunk_length(leaf, chunk);
2295 spin_lock(&map_tree->map_tree.lock);
2296 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2297 spin_unlock(&map_tree->map_tree.lock);
2299 /* already mapped? */
2300 if (em && em->start <= logical && em->start + em->len > logical) {
2301 free_extent_map(em);
2304 free_extent_map(em);
2307 map = kzalloc(sizeof(*map), GFP_NOFS);
2311 em = alloc_extent_map(GFP_NOFS);
2314 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2315 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2317 free_extent_map(em);
2321 em->bdev = (struct block_device *)map;
2322 em->start = logical;
2324 em->block_start = 0;
2326 map->num_stripes = num_stripes;
2327 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2328 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2329 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2330 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2331 map->type = btrfs_chunk_type(leaf, chunk);
2332 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2333 for (i = 0; i < num_stripes; i++) {
2334 map->stripes[i].physical =
2335 btrfs_stripe_offset_nr(leaf, chunk, i);
2336 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2337 read_extent_buffer(leaf, uuid, (unsigned long)
2338 btrfs_stripe_dev_uuid_nr(chunk, i),
2340 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2342 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2344 free_extent_map(em);
2347 if (!map->stripes[i].dev) {
2348 map->stripes[i].dev =
2349 add_missing_dev(root, devid, uuid);
2350 if (!map->stripes[i].dev) {
2352 free_extent_map(em);
2356 map->stripes[i].dev->in_fs_metadata = 1;
2359 spin_lock(&map_tree->map_tree.lock);
2360 ret = add_extent_mapping(&map_tree->map_tree, em);
2361 spin_unlock(&map_tree->map_tree.lock);
2363 free_extent_map(em);
2368 static int fill_device_from_item(struct extent_buffer *leaf,
2369 struct btrfs_dev_item *dev_item,
2370 struct btrfs_device *device)
2374 device->devid = btrfs_device_id(leaf, dev_item);
2375 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2376 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2377 device->type = btrfs_device_type(leaf, dev_item);
2378 device->io_align = btrfs_device_io_align(leaf, dev_item);
2379 device->io_width = btrfs_device_io_width(leaf, dev_item);
2380 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2382 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2383 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2388 static int read_one_dev(struct btrfs_root *root,
2389 struct extent_buffer *leaf,
2390 struct btrfs_dev_item *dev_item)
2392 struct btrfs_device *device;
2395 u8 dev_uuid[BTRFS_UUID_SIZE];
2397 devid = btrfs_device_id(leaf, dev_item);
2398 read_extent_buffer(leaf, dev_uuid,
2399 (unsigned long)btrfs_device_uuid(dev_item),
2401 device = btrfs_find_device(root, devid, dev_uuid);
2403 printk("warning devid %Lu missing\n", devid);
2404 device = add_missing_dev(root, devid, dev_uuid);
2409 fill_device_from_item(leaf, dev_item, device);
2410 device->dev_root = root->fs_info->dev_root;
2411 device->in_fs_metadata = 1;
2414 ret = btrfs_open_device(device);
2422 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2424 struct btrfs_dev_item *dev_item;
2426 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2428 return read_one_dev(root, buf, dev_item);
2431 int btrfs_read_sys_array(struct btrfs_root *root)
2433 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2434 struct extent_buffer *sb;
2435 struct btrfs_disk_key *disk_key;
2436 struct btrfs_chunk *chunk;
2438 unsigned long sb_ptr;
2444 struct btrfs_key key;
2446 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2447 BTRFS_SUPER_INFO_SIZE);
2450 btrfs_set_buffer_uptodate(sb);
2451 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2452 array_size = btrfs_super_sys_array_size(super_copy);
2454 ptr = super_copy->sys_chunk_array;
2455 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2458 while (cur < array_size) {
2459 disk_key = (struct btrfs_disk_key *)ptr;
2460 btrfs_disk_key_to_cpu(&key, disk_key);
2462 len = sizeof(*disk_key); ptr += len;
2466 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2467 chunk = (struct btrfs_chunk *)sb_ptr;
2468 ret = read_one_chunk(root, &key, sb, chunk);
2471 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2472 len = btrfs_chunk_item_size(num_stripes);
2481 free_extent_buffer(sb);
2485 int btrfs_read_chunk_tree(struct btrfs_root *root)
2487 struct btrfs_path *path;
2488 struct extent_buffer *leaf;
2489 struct btrfs_key key;
2490 struct btrfs_key found_key;
2494 root = root->fs_info->chunk_root;
2496 path = btrfs_alloc_path();
2500 /* first we search for all of the device items, and then we
2501 * read in all of the chunk items. This way we can create chunk
2502 * mappings that reference all of the devices that are afound
2504 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2508 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2510 leaf = path->nodes[0];
2511 slot = path->slots[0];
2512 if (slot >= btrfs_header_nritems(leaf)) {
2513 ret = btrfs_next_leaf(root, path);
2520 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2521 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2522 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2524 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2525 struct btrfs_dev_item *dev_item;
2526 dev_item = btrfs_item_ptr(leaf, slot,
2527 struct btrfs_dev_item);
2528 ret = read_one_dev(root, leaf, dev_item);
2531 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2532 struct btrfs_chunk *chunk;
2533 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2534 ret = read_one_chunk(root, &found_key, leaf, chunk);
2538 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2540 btrfs_release_path(root, path);
2544 btrfs_free_path(path);