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->chunk_mutex);
64 static void unlock_chunks(struct btrfs_root *root)
66 mutex_unlock(&root->fs_info->chunk_mutex);
69 int btrfs_cleanup_fs_uuids(void)
71 struct btrfs_fs_devices *fs_devices;
72 struct list_head *uuid_cur;
73 struct list_head *devices_cur;
74 struct btrfs_device *dev;
76 list_for_each(uuid_cur, &fs_uuids) {
77 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
79 while(!list_empty(&fs_devices->devices)) {
80 devices_cur = fs_devices->devices.next;
81 dev = list_entry(devices_cur, struct btrfs_device,
84 close_bdev_excl(dev->bdev);
85 fs_devices->open_devices--;
87 list_del(&dev->dev_list);
95 static noinline struct btrfs_device *__find_device(struct list_head *head,
98 struct btrfs_device *dev;
99 struct list_head *cur;
101 list_for_each(cur, head) {
102 dev = list_entry(cur, struct btrfs_device, dev_list);
103 if (dev->devid == devid &&
104 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
111 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
113 struct list_head *cur;
114 struct btrfs_fs_devices *fs_devices;
116 list_for_each(cur, &fs_uuids) {
117 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
118 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
125 * we try to collect pending bios for a device so we don't get a large
126 * number of procs sending bios down to the same device. This greatly
127 * improves the schedulers ability to collect and merge the bios.
129 * But, it also turns into a long list of bios to process and that is sure
130 * to eventually make the worker thread block. The solution here is to
131 * make some progress and then put this work struct back at the end of
132 * the list if the block device is congested. This way, multiple devices
133 * can make progress from a single worker thread.
135 static int noinline run_scheduled_bios(struct btrfs_device *device)
138 struct backing_dev_info *bdi;
139 struct btrfs_fs_info *fs_info;
143 unsigned long num_run = 0;
146 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
147 fs_info = device->dev_root->fs_info;
148 limit = btrfs_async_submit_limit(fs_info);
149 limit = limit * 2 / 3;
152 spin_lock(&device->io_lock);
154 /* take all the bios off the list at once and process them
155 * later on (without the lock held). But, remember the
156 * tail and other pointers so the bios can be properly reinserted
157 * into the list if we hit congestion
159 pending = device->pending_bios;
160 tail = device->pending_bio_tail;
161 WARN_ON(pending && !tail);
162 device->pending_bios = NULL;
163 device->pending_bio_tail = NULL;
166 * if pending was null this time around, no bios need processing
167 * at all and we can stop. Otherwise it'll loop back up again
168 * and do an additional check so no bios are missed.
170 * device->running_pending is used to synchronize with the
175 device->running_pending = 1;
178 device->running_pending = 0;
180 spin_unlock(&device->io_lock);
184 pending = pending->bi_next;
186 atomic_dec(&fs_info->nr_async_bios);
188 if (atomic_read(&fs_info->nr_async_bios) < limit &&
189 waitqueue_active(&fs_info->async_submit_wait))
190 wake_up(&fs_info->async_submit_wait);
192 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
194 submit_bio(cur->bi_rw, cur);
199 * we made progress, there is more work to do and the bdi
200 * is now congested. Back off and let other work structs
203 if (pending && bdi_write_congested(bdi) &&
204 fs_info->fs_devices->open_devices > 1) {
205 struct bio *old_head;
207 spin_lock(&device->io_lock);
209 old_head = device->pending_bios;
210 device->pending_bios = pending;
211 if (device->pending_bio_tail)
212 tail->bi_next = old_head;
214 device->pending_bio_tail = tail;
216 spin_unlock(&device->io_lock);
217 btrfs_requeue_work(&device->work);
227 void pending_bios_fn(struct btrfs_work *work)
229 struct btrfs_device *device;
231 device = container_of(work, struct btrfs_device, work);
232 run_scheduled_bios(device);
235 static noinline int device_list_add(const char *path,
236 struct btrfs_super_block *disk_super,
237 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
239 struct btrfs_device *device;
240 struct btrfs_fs_devices *fs_devices;
241 u64 found_transid = btrfs_super_generation(disk_super);
243 fs_devices = find_fsid(disk_super->fsid);
245 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
248 INIT_LIST_HEAD(&fs_devices->devices);
249 INIT_LIST_HEAD(&fs_devices->alloc_list);
250 list_add(&fs_devices->list, &fs_uuids);
251 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
252 fs_devices->latest_devid = devid;
253 fs_devices->latest_trans = found_transid;
256 device = __find_device(&fs_devices->devices, devid,
257 disk_super->dev_item.uuid);
260 device = kzalloc(sizeof(*device), GFP_NOFS);
262 /* we can safely leave the fs_devices entry around */
265 device->devid = devid;
266 device->work.func = pending_bios_fn;
267 memcpy(device->uuid, disk_super->dev_item.uuid,
269 device->barriers = 1;
270 spin_lock_init(&device->io_lock);
271 device->name = kstrdup(path, GFP_NOFS);
276 list_add(&device->dev_list, &fs_devices->devices);
277 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
278 fs_devices->num_devices++;
281 if (found_transid > fs_devices->latest_trans) {
282 fs_devices->latest_devid = devid;
283 fs_devices->latest_trans = found_transid;
285 *fs_devices_ret = fs_devices;
289 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
291 struct list_head *head = &fs_devices->devices;
292 struct list_head *cur;
293 struct btrfs_device *device;
295 mutex_lock(&uuid_mutex);
297 list_for_each(cur, head) {
298 device = list_entry(cur, struct btrfs_device, dev_list);
299 if (!device->in_fs_metadata) {
300 struct block_device *bdev;
301 list_del(&device->dev_list);
302 list_del(&device->dev_alloc_list);
303 fs_devices->num_devices--;
306 fs_devices->open_devices--;
307 mutex_unlock(&uuid_mutex);
308 close_bdev_excl(bdev);
309 mutex_lock(&uuid_mutex);
316 mutex_unlock(&uuid_mutex);
320 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
322 struct list_head *head = &fs_devices->devices;
323 struct list_head *cur;
324 struct btrfs_device *device;
326 mutex_lock(&uuid_mutex);
327 list_for_each(cur, head) {
328 device = list_entry(cur, struct btrfs_device, dev_list);
330 close_bdev_excl(device->bdev);
331 fs_devices->open_devices--;
334 device->in_fs_metadata = 0;
336 fs_devices->mounted = 0;
337 mutex_unlock(&uuid_mutex);
341 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
342 int flags, void *holder)
344 struct block_device *bdev;
345 struct list_head *head = &fs_devices->devices;
346 struct list_head *cur;
347 struct btrfs_device *device;
348 struct block_device *latest_bdev = NULL;
349 struct buffer_head *bh;
350 struct btrfs_super_block *disk_super;
351 u64 latest_devid = 0;
352 u64 latest_transid = 0;
357 mutex_lock(&uuid_mutex);
358 if (fs_devices->mounted)
361 list_for_each(cur, head) {
362 device = list_entry(cur, struct btrfs_device, dev_list);
369 bdev = open_bdev_excl(device->name, flags, holder);
372 printk("open %s failed\n", device->name);
375 set_blocksize(bdev, 4096);
377 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
381 disk_super = (struct btrfs_super_block *)bh->b_data;
382 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
383 sizeof(disk_super->magic)))
386 devid = le64_to_cpu(disk_super->dev_item.devid);
387 if (devid != device->devid)
390 transid = btrfs_super_generation(disk_super);
391 if (!latest_transid || transid > latest_transid) {
392 latest_devid = devid;
393 latest_transid = transid;
398 device->in_fs_metadata = 0;
399 fs_devices->open_devices++;
405 close_bdev_excl(bdev);
409 if (fs_devices->open_devices == 0) {
413 fs_devices->mounted = 1;
414 fs_devices->latest_bdev = latest_bdev;
415 fs_devices->latest_devid = latest_devid;
416 fs_devices->latest_trans = latest_transid;
418 mutex_unlock(&uuid_mutex);
422 int btrfs_scan_one_device(const char *path, int flags, void *holder,
423 struct btrfs_fs_devices **fs_devices_ret)
425 struct btrfs_super_block *disk_super;
426 struct block_device *bdev;
427 struct buffer_head *bh;
432 mutex_lock(&uuid_mutex);
434 bdev = open_bdev_excl(path, flags, holder);
441 ret = set_blocksize(bdev, 4096);
444 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
449 disk_super = (struct btrfs_super_block *)bh->b_data;
450 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
451 sizeof(disk_super->magic))) {
455 devid = le64_to_cpu(disk_super->dev_item.devid);
456 transid = btrfs_super_generation(disk_super);
457 if (disk_super->label[0])
458 printk("device label %s ", disk_super->label);
460 /* FIXME, make a readl uuid parser */
461 printk("device fsid %llx-%llx ",
462 *(unsigned long long *)disk_super->fsid,
463 *(unsigned long long *)(disk_super->fsid + 8));
465 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
466 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
471 close_bdev_excl(bdev);
473 mutex_unlock(&uuid_mutex);
478 * this uses a pretty simple search, the expectation is that it is
479 * called very infrequently and that a given device has a small number
482 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
483 struct btrfs_device *device,
484 struct btrfs_path *path,
485 u64 num_bytes, u64 *start)
487 struct btrfs_key key;
488 struct btrfs_root *root = device->dev_root;
489 struct btrfs_dev_extent *dev_extent = NULL;
492 u64 search_start = 0;
493 u64 search_end = device->total_bytes;
497 struct extent_buffer *l;
502 /* FIXME use last free of some kind */
504 /* we don't want to overwrite the superblock on the drive,
505 * so we make sure to start at an offset of at least 1MB
507 search_start = max((u64)1024 * 1024, search_start);
509 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
510 search_start = max(root->fs_info->alloc_start, search_start);
512 key.objectid = device->devid;
513 key.offset = search_start;
514 key.type = BTRFS_DEV_EXTENT_KEY;
515 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
518 ret = btrfs_previous_item(root, path, 0, key.type);
522 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
525 slot = path->slots[0];
526 if (slot >= btrfs_header_nritems(l)) {
527 ret = btrfs_next_leaf(root, path);
534 if (search_start >= search_end) {
538 *start = search_start;
542 *start = last_byte > search_start ?
543 last_byte : search_start;
544 if (search_end <= *start) {
550 btrfs_item_key_to_cpu(l, &key, slot);
552 if (key.objectid < device->devid)
555 if (key.objectid > device->devid)
558 if (key.offset >= search_start && key.offset > last_byte &&
560 if (last_byte < search_start)
561 last_byte = search_start;
562 hole_size = key.offset - last_byte;
563 if (key.offset > last_byte &&
564 hole_size >= num_bytes) {
569 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
574 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
575 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
581 /* we have to make sure we didn't find an extent that has already
582 * been allocated by the map tree or the original allocation
584 btrfs_release_path(root, path);
585 BUG_ON(*start < search_start);
587 if (*start + num_bytes > search_end) {
591 /* check for pending inserts here */
595 btrfs_release_path(root, path);
599 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
600 struct btrfs_device *device,
604 struct btrfs_path *path;
605 struct btrfs_root *root = device->dev_root;
606 struct btrfs_key key;
607 struct btrfs_key found_key;
608 struct extent_buffer *leaf = NULL;
609 struct btrfs_dev_extent *extent = NULL;
611 path = btrfs_alloc_path();
615 key.objectid = device->devid;
617 key.type = BTRFS_DEV_EXTENT_KEY;
619 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
621 ret = btrfs_previous_item(root, path, key.objectid,
622 BTRFS_DEV_EXTENT_KEY);
624 leaf = path->nodes[0];
625 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
626 extent = btrfs_item_ptr(leaf, path->slots[0],
627 struct btrfs_dev_extent);
628 BUG_ON(found_key.offset > start || found_key.offset +
629 btrfs_dev_extent_length(leaf, extent) < start);
631 } else if (ret == 0) {
632 leaf = path->nodes[0];
633 extent = btrfs_item_ptr(leaf, path->slots[0],
634 struct btrfs_dev_extent);
638 if (device->bytes_used > 0)
639 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
640 ret = btrfs_del_item(trans, root, path);
643 btrfs_free_path(path);
647 int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
648 struct btrfs_device *device,
649 u64 chunk_tree, u64 chunk_objectid,
651 u64 num_bytes, u64 *start)
654 struct btrfs_path *path;
655 struct btrfs_root *root = device->dev_root;
656 struct btrfs_dev_extent *extent;
657 struct extent_buffer *leaf;
658 struct btrfs_key key;
660 WARN_ON(!device->in_fs_metadata);
661 path = btrfs_alloc_path();
665 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
670 key.objectid = device->devid;
672 key.type = BTRFS_DEV_EXTENT_KEY;
673 ret = btrfs_insert_empty_item(trans, root, path, &key,
677 leaf = path->nodes[0];
678 extent = btrfs_item_ptr(leaf, path->slots[0],
679 struct btrfs_dev_extent);
680 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
681 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
682 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
684 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
685 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
688 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
689 btrfs_mark_buffer_dirty(leaf);
691 btrfs_free_path(path);
695 static noinline int find_next_chunk(struct btrfs_root *root,
696 u64 objectid, u64 *offset)
698 struct btrfs_path *path;
700 struct btrfs_key key;
701 struct btrfs_chunk *chunk;
702 struct btrfs_key found_key;
704 path = btrfs_alloc_path();
707 key.objectid = objectid;
708 key.offset = (u64)-1;
709 key.type = BTRFS_CHUNK_ITEM_KEY;
711 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
717 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
721 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
723 if (found_key.objectid != objectid)
726 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
728 *offset = found_key.offset +
729 btrfs_chunk_length(path->nodes[0], chunk);
734 btrfs_free_path(path);
738 static noinline int find_next_devid(struct btrfs_root *root,
739 struct btrfs_path *path, u64 *objectid)
742 struct btrfs_key key;
743 struct btrfs_key found_key;
745 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
746 key.type = BTRFS_DEV_ITEM_KEY;
747 key.offset = (u64)-1;
749 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
755 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
760 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
762 *objectid = found_key.offset + 1;
766 btrfs_release_path(root, path);
771 * the device information is stored in the chunk root
772 * the btrfs_device struct should be fully filled in
774 int btrfs_add_device(struct btrfs_trans_handle *trans,
775 struct btrfs_root *root,
776 struct btrfs_device *device)
779 struct btrfs_path *path;
780 struct btrfs_dev_item *dev_item;
781 struct extent_buffer *leaf;
782 struct btrfs_key key;
786 root = root->fs_info->chunk_root;
788 path = btrfs_alloc_path();
792 ret = find_next_devid(root, path, &free_devid);
796 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
797 key.type = BTRFS_DEV_ITEM_KEY;
798 key.offset = free_devid;
800 ret = btrfs_insert_empty_item(trans, root, path, &key,
805 leaf = path->nodes[0];
806 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
808 device->devid = free_devid;
809 btrfs_set_device_id(leaf, dev_item, device->devid);
810 btrfs_set_device_type(leaf, dev_item, device->type);
811 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
812 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
813 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
814 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
815 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
816 btrfs_set_device_group(leaf, dev_item, 0);
817 btrfs_set_device_seek_speed(leaf, dev_item, 0);
818 btrfs_set_device_bandwidth(leaf, dev_item, 0);
820 ptr = (unsigned long)btrfs_device_uuid(dev_item);
821 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
822 btrfs_mark_buffer_dirty(leaf);
826 btrfs_free_path(path);
830 static int btrfs_rm_dev_item(struct btrfs_root *root,
831 struct btrfs_device *device)
834 struct btrfs_path *path;
835 struct block_device *bdev = device->bdev;
836 struct btrfs_device *next_dev;
837 struct btrfs_key key;
839 struct btrfs_fs_devices *fs_devices;
840 struct btrfs_trans_handle *trans;
842 root = root->fs_info->chunk_root;
844 path = btrfs_alloc_path();
848 trans = btrfs_start_transaction(root, 1);
849 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
850 key.type = BTRFS_DEV_ITEM_KEY;
851 key.offset = device->devid;
854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
863 ret = btrfs_del_item(trans, root, path);
868 * at this point, the device is zero sized. We want to
869 * remove it from the devices list and zero out the old super
871 list_del_init(&device->dev_list);
872 list_del_init(&device->dev_alloc_list);
873 fs_devices = root->fs_info->fs_devices;
875 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
877 if (bdev == root->fs_info->sb->s_bdev)
878 root->fs_info->sb->s_bdev = next_dev->bdev;
879 if (bdev == fs_devices->latest_bdev)
880 fs_devices->latest_bdev = next_dev->bdev;
882 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
883 btrfs_set_super_num_devices(&root->fs_info->super_copy,
886 btrfs_free_path(path);
888 btrfs_commit_transaction(trans, root);
892 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
894 struct btrfs_device *device;
895 struct block_device *bdev;
896 struct buffer_head *bh = NULL;
897 struct btrfs_super_block *disk_super;
902 mutex_lock(&uuid_mutex);
903 mutex_lock(&root->fs_info->volume_mutex);
905 all_avail = root->fs_info->avail_data_alloc_bits |
906 root->fs_info->avail_system_alloc_bits |
907 root->fs_info->avail_metadata_alloc_bits;
909 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
910 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
911 printk("btrfs: unable to go below four devices on raid10\n");
916 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
917 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
918 printk("btrfs: unable to go below two devices on raid1\n");
923 if (strcmp(device_path, "missing") == 0) {
924 struct list_head *cur;
925 struct list_head *devices;
926 struct btrfs_device *tmp;
929 devices = &root->fs_info->fs_devices->devices;
930 list_for_each(cur, devices) {
931 tmp = list_entry(cur, struct btrfs_device, dev_list);
932 if (tmp->in_fs_metadata && !tmp->bdev) {
941 printk("btrfs: no missing devices found to remove\n");
946 bdev = open_bdev_excl(device_path, 0,
947 root->fs_info->bdev_holder);
953 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
958 disk_super = (struct btrfs_super_block *)bh->b_data;
959 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
960 sizeof(disk_super->magic))) {
964 if (memcmp(disk_super->fsid, root->fs_info->fsid,
969 devid = le64_to_cpu(disk_super->dev_item.devid);
970 device = btrfs_find_device(root, devid, NULL);
977 root->fs_info->fs_devices->num_devices--;
978 root->fs_info->fs_devices->open_devices--;
980 ret = btrfs_shrink_device(device, 0);
985 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
990 /* make sure this device isn't detected as part of
993 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
994 set_buffer_dirty(bh);
995 sync_dirty_buffer(bh);
1001 /* one close for the device struct or super_block */
1002 close_bdev_excl(device->bdev);
1005 /* one close for us */
1006 close_bdev_excl(bdev);
1008 kfree(device->name);
1017 close_bdev_excl(bdev);
1019 mutex_unlock(&root->fs_info->volume_mutex);
1020 mutex_unlock(&uuid_mutex);
1024 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1026 struct btrfs_trans_handle *trans;
1027 struct btrfs_device *device;
1028 struct block_device *bdev;
1029 struct list_head *cur;
1030 struct list_head *devices;
1035 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1040 filemap_write_and_wait(bdev->bd_inode->i_mapping);
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;
1818 em->block_len = em->len;
1820 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1821 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1822 chunk, btrfs_chunk_item_size(num_stripes));
1827 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1828 spin_lock(&em_tree->lock);
1829 ret = add_extent_mapping(em_tree, em);
1830 spin_unlock(&em_tree->lock);
1832 free_extent_map(em);
1836 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1838 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1841 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1843 struct extent_map *em;
1846 spin_lock(&tree->map_tree.lock);
1847 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1849 remove_extent_mapping(&tree->map_tree, em);
1850 spin_unlock(&tree->map_tree.lock);
1855 free_extent_map(em);
1856 /* once for the tree */
1857 free_extent_map(em);
1861 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1863 struct extent_map *em;
1864 struct map_lookup *map;
1865 struct extent_map_tree *em_tree = &map_tree->map_tree;
1868 spin_lock(&em_tree->lock);
1869 em = lookup_extent_mapping(em_tree, logical, len);
1870 spin_unlock(&em_tree->lock);
1873 BUG_ON(em->start > logical || em->start + em->len < logical);
1874 map = (struct map_lookup *)em->bdev;
1875 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1876 ret = map->num_stripes;
1877 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1878 ret = map->sub_stripes;
1881 free_extent_map(em);
1885 static int find_live_mirror(struct map_lookup *map, int first, int num,
1889 if (map->stripes[optimal].dev->bdev)
1891 for (i = first; i < first + num; i++) {
1892 if (map->stripes[i].dev->bdev)
1895 /* we couldn't find one that doesn't fail. Just return something
1896 * and the io error handling code will clean up eventually
1901 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1902 u64 logical, u64 *length,
1903 struct btrfs_multi_bio **multi_ret,
1904 int mirror_num, struct page *unplug_page)
1906 struct extent_map *em;
1907 struct map_lookup *map;
1908 struct extent_map_tree *em_tree = &map_tree->map_tree;
1912 int stripes_allocated = 8;
1913 int stripes_required = 1;
1918 struct btrfs_multi_bio *multi = NULL;
1920 if (multi_ret && !(rw & (1 << BIO_RW))) {
1921 stripes_allocated = 1;
1925 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1930 atomic_set(&multi->error, 0);
1933 spin_lock(&em_tree->lock);
1934 em = lookup_extent_mapping(em_tree, logical, *length);
1935 spin_unlock(&em_tree->lock);
1937 if (!em && unplug_page)
1941 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1945 BUG_ON(em->start > logical || em->start + em->len < logical);
1946 map = (struct map_lookup *)em->bdev;
1947 offset = logical - em->start;
1949 if (mirror_num > map->num_stripes)
1952 /* if our multi bio struct is too small, back off and try again */
1953 if (rw & (1 << BIO_RW)) {
1954 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1955 BTRFS_BLOCK_GROUP_DUP)) {
1956 stripes_required = map->num_stripes;
1958 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1959 stripes_required = map->sub_stripes;
1963 if (multi_ret && rw == WRITE &&
1964 stripes_allocated < stripes_required) {
1965 stripes_allocated = map->num_stripes;
1966 free_extent_map(em);
1972 * stripe_nr counts the total number of stripes we have to stride
1973 * to get to this block
1975 do_div(stripe_nr, map->stripe_len);
1977 stripe_offset = stripe_nr * map->stripe_len;
1978 BUG_ON(offset < stripe_offset);
1980 /* stripe_offset is the offset of this block in its stripe*/
1981 stripe_offset = offset - stripe_offset;
1983 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1984 BTRFS_BLOCK_GROUP_RAID10 |
1985 BTRFS_BLOCK_GROUP_DUP)) {
1986 /* we limit the length of each bio to what fits in a stripe */
1987 *length = min_t(u64, em->len - offset,
1988 map->stripe_len - stripe_offset);
1990 *length = em->len - offset;
1993 if (!multi_ret && !unplug_page)
1998 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1999 if (unplug_page || (rw & (1 << BIO_RW)))
2000 num_stripes = map->num_stripes;
2001 else if (mirror_num)
2002 stripe_index = mirror_num - 1;
2004 stripe_index = find_live_mirror(map, 0,
2006 current->pid % map->num_stripes);
2009 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2010 if (rw & (1 << BIO_RW))
2011 num_stripes = map->num_stripes;
2012 else if (mirror_num)
2013 stripe_index = mirror_num - 1;
2015 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2016 int factor = map->num_stripes / map->sub_stripes;
2018 stripe_index = do_div(stripe_nr, factor);
2019 stripe_index *= map->sub_stripes;
2021 if (unplug_page || (rw & (1 << BIO_RW)))
2022 num_stripes = map->sub_stripes;
2023 else if (mirror_num)
2024 stripe_index += mirror_num - 1;
2026 stripe_index = find_live_mirror(map, stripe_index,
2027 map->sub_stripes, stripe_index +
2028 current->pid % map->sub_stripes);
2032 * after this do_div call, stripe_nr is the number of stripes
2033 * on this device we have to walk to find the data, and
2034 * stripe_index is the number of our device in the stripe array
2036 stripe_index = do_div(stripe_nr, map->num_stripes);
2038 BUG_ON(stripe_index >= map->num_stripes);
2040 for (i = 0; i < num_stripes; i++) {
2042 struct btrfs_device *device;
2043 struct backing_dev_info *bdi;
2045 device = map->stripes[stripe_index].dev;
2047 bdi = blk_get_backing_dev_info(device->bdev);
2048 if (bdi->unplug_io_fn) {
2049 bdi->unplug_io_fn(bdi, unplug_page);
2053 multi->stripes[i].physical =
2054 map->stripes[stripe_index].physical +
2055 stripe_offset + stripe_nr * map->stripe_len;
2056 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2062 multi->num_stripes = num_stripes;
2063 multi->max_errors = max_errors;
2066 free_extent_map(em);
2070 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2071 u64 logical, u64 *length,
2072 struct btrfs_multi_bio **multi_ret, int mirror_num)
2074 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2078 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2079 u64 logical, struct page *page)
2081 u64 length = PAGE_CACHE_SIZE;
2082 return __btrfs_map_block(map_tree, READ, logical, &length,
2087 static void end_bio_multi_stripe(struct bio *bio, int err)
2089 struct btrfs_multi_bio *multi = bio->bi_private;
2090 int is_orig_bio = 0;
2093 atomic_inc(&multi->error);
2095 if (bio == multi->orig_bio)
2098 if (atomic_dec_and_test(&multi->stripes_pending)) {
2101 bio = multi->orig_bio;
2103 bio->bi_private = multi->private;
2104 bio->bi_end_io = multi->end_io;
2105 /* only send an error to the higher layers if it is
2106 * beyond the tolerance of the multi-bio
2108 if (atomic_read(&multi->error) > multi->max_errors) {
2112 * this bio is actually up to date, we didn't
2113 * go over the max number of errors
2115 set_bit(BIO_UPTODATE, &bio->bi_flags);
2120 bio_endio(bio, err);
2121 } else if (!is_orig_bio) {
2126 struct async_sched {
2129 struct btrfs_fs_info *info;
2130 struct btrfs_work work;
2134 * see run_scheduled_bios for a description of why bios are collected for
2137 * This will add one bio to the pending list for a device and make sure
2138 * the work struct is scheduled.
2140 static int noinline schedule_bio(struct btrfs_root *root,
2141 struct btrfs_device *device,
2142 int rw, struct bio *bio)
2144 int should_queue = 1;
2146 /* don't bother with additional async steps for reads, right now */
2147 if (!(rw & (1 << BIO_RW))) {
2149 submit_bio(rw, bio);
2155 * nr_async_bios allows us to reliably return congestion to the
2156 * higher layers. Otherwise, the async bio makes it appear we have
2157 * made progress against dirty pages when we've really just put it
2158 * on a queue for later
2160 atomic_inc(&root->fs_info->nr_async_bios);
2161 WARN_ON(bio->bi_next);
2162 bio->bi_next = NULL;
2165 spin_lock(&device->io_lock);
2167 if (device->pending_bio_tail)
2168 device->pending_bio_tail->bi_next = bio;
2170 device->pending_bio_tail = bio;
2171 if (!device->pending_bios)
2172 device->pending_bios = bio;
2173 if (device->running_pending)
2176 spin_unlock(&device->io_lock);
2179 btrfs_queue_worker(&root->fs_info->submit_workers,
2184 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2185 int mirror_num, int async_submit)
2187 struct btrfs_mapping_tree *map_tree;
2188 struct btrfs_device *dev;
2189 struct bio *first_bio = bio;
2190 u64 logical = (u64)bio->bi_sector << 9;
2193 struct btrfs_multi_bio *multi = NULL;
2198 length = bio->bi_size;
2199 map_tree = &root->fs_info->mapping_tree;
2200 map_length = length;
2202 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2206 total_devs = multi->num_stripes;
2207 if (map_length < length) {
2208 printk("mapping failed logical %Lu bio len %Lu "
2209 "len %Lu\n", logical, length, map_length);
2212 multi->end_io = first_bio->bi_end_io;
2213 multi->private = first_bio->bi_private;
2214 multi->orig_bio = first_bio;
2215 atomic_set(&multi->stripes_pending, multi->num_stripes);
2217 while(dev_nr < total_devs) {
2218 if (total_devs > 1) {
2219 if (dev_nr < total_devs - 1) {
2220 bio = bio_clone(first_bio, GFP_NOFS);
2225 bio->bi_private = multi;
2226 bio->bi_end_io = end_bio_multi_stripe;
2228 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2229 dev = multi->stripes[dev_nr].dev;
2230 if (dev && dev->bdev) {
2231 bio->bi_bdev = dev->bdev;
2233 schedule_bio(root, dev, rw, bio);
2235 submit_bio(rw, bio);
2237 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2238 bio->bi_sector = logical >> 9;
2239 bio_endio(bio, -EIO);
2243 if (total_devs == 1)
2248 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2251 struct list_head *head = &root->fs_info->fs_devices->devices;
2253 return __find_device(head, devid, uuid);
2256 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2257 u64 devid, u8 *dev_uuid)
2259 struct btrfs_device *device;
2260 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2262 device = kzalloc(sizeof(*device), GFP_NOFS);
2263 list_add(&device->dev_list,
2264 &fs_devices->devices);
2265 list_add(&device->dev_alloc_list,
2266 &fs_devices->alloc_list);
2267 device->barriers = 1;
2268 device->dev_root = root->fs_info->dev_root;
2269 device->devid = devid;
2270 device->work.func = pending_bios_fn;
2271 fs_devices->num_devices++;
2272 spin_lock_init(&device->io_lock);
2273 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2278 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2279 struct extent_buffer *leaf,
2280 struct btrfs_chunk *chunk)
2282 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2283 struct map_lookup *map;
2284 struct extent_map *em;
2288 u8 uuid[BTRFS_UUID_SIZE];
2293 logical = key->offset;
2294 length = btrfs_chunk_length(leaf, chunk);
2296 spin_lock(&map_tree->map_tree.lock);
2297 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2298 spin_unlock(&map_tree->map_tree.lock);
2300 /* already mapped? */
2301 if (em && em->start <= logical && em->start + em->len > logical) {
2302 free_extent_map(em);
2305 free_extent_map(em);
2308 map = kzalloc(sizeof(*map), GFP_NOFS);
2312 em = alloc_extent_map(GFP_NOFS);
2315 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2316 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2318 free_extent_map(em);
2322 em->bdev = (struct block_device *)map;
2323 em->start = logical;
2325 em->block_start = 0;
2326 em->block_len = em->len;
2328 map->num_stripes = num_stripes;
2329 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2330 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2331 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2332 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2333 map->type = btrfs_chunk_type(leaf, chunk);
2334 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2335 for (i = 0; i < num_stripes; i++) {
2336 map->stripes[i].physical =
2337 btrfs_stripe_offset_nr(leaf, chunk, i);
2338 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2339 read_extent_buffer(leaf, uuid, (unsigned long)
2340 btrfs_stripe_dev_uuid_nr(chunk, i),
2342 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2344 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2346 free_extent_map(em);
2349 if (!map->stripes[i].dev) {
2350 map->stripes[i].dev =
2351 add_missing_dev(root, devid, uuid);
2352 if (!map->stripes[i].dev) {
2354 free_extent_map(em);
2358 map->stripes[i].dev->in_fs_metadata = 1;
2361 spin_lock(&map_tree->map_tree.lock);
2362 ret = add_extent_mapping(&map_tree->map_tree, em);
2363 spin_unlock(&map_tree->map_tree.lock);
2365 free_extent_map(em);
2370 static int fill_device_from_item(struct extent_buffer *leaf,
2371 struct btrfs_dev_item *dev_item,
2372 struct btrfs_device *device)
2376 device->devid = btrfs_device_id(leaf, dev_item);
2377 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2378 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2379 device->type = btrfs_device_type(leaf, dev_item);
2380 device->io_align = btrfs_device_io_align(leaf, dev_item);
2381 device->io_width = btrfs_device_io_width(leaf, dev_item);
2382 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2384 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2385 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2390 static int read_one_dev(struct btrfs_root *root,
2391 struct extent_buffer *leaf,
2392 struct btrfs_dev_item *dev_item)
2394 struct btrfs_device *device;
2397 u8 dev_uuid[BTRFS_UUID_SIZE];
2399 devid = btrfs_device_id(leaf, dev_item);
2400 read_extent_buffer(leaf, dev_uuid,
2401 (unsigned long)btrfs_device_uuid(dev_item),
2403 device = btrfs_find_device(root, devid, dev_uuid);
2405 printk("warning devid %Lu missing\n", devid);
2406 device = add_missing_dev(root, devid, dev_uuid);
2411 fill_device_from_item(leaf, dev_item, device);
2412 device->dev_root = root->fs_info->dev_root;
2413 device->in_fs_metadata = 1;
2416 ret = btrfs_open_device(device);
2424 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2426 struct btrfs_dev_item *dev_item;
2428 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2430 return read_one_dev(root, buf, dev_item);
2433 int btrfs_read_sys_array(struct btrfs_root *root)
2435 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2436 struct extent_buffer *sb;
2437 struct btrfs_disk_key *disk_key;
2438 struct btrfs_chunk *chunk;
2440 unsigned long sb_ptr;
2446 struct btrfs_key key;
2448 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2449 BTRFS_SUPER_INFO_SIZE);
2452 btrfs_set_buffer_uptodate(sb);
2453 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2454 array_size = btrfs_super_sys_array_size(super_copy);
2456 ptr = super_copy->sys_chunk_array;
2457 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2460 while (cur < array_size) {
2461 disk_key = (struct btrfs_disk_key *)ptr;
2462 btrfs_disk_key_to_cpu(&key, disk_key);
2464 len = sizeof(*disk_key); ptr += len;
2468 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2469 chunk = (struct btrfs_chunk *)sb_ptr;
2470 ret = read_one_chunk(root, &key, sb, chunk);
2473 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2474 len = btrfs_chunk_item_size(num_stripes);
2483 free_extent_buffer(sb);
2487 int btrfs_read_chunk_tree(struct btrfs_root *root)
2489 struct btrfs_path *path;
2490 struct extent_buffer *leaf;
2491 struct btrfs_key key;
2492 struct btrfs_key found_key;
2496 root = root->fs_info->chunk_root;
2498 path = btrfs_alloc_path();
2502 /* first we search for all of the device items, and then we
2503 * read in all of the chunk items. This way we can create chunk
2504 * mappings that reference all of the devices that are afound
2506 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512 leaf = path->nodes[0];
2513 slot = path->slots[0];
2514 if (slot >= btrfs_header_nritems(leaf)) {
2515 ret = btrfs_next_leaf(root, path);
2522 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2523 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2524 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2526 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2527 struct btrfs_dev_item *dev_item;
2528 dev_item = btrfs_item_ptr(leaf, slot,
2529 struct btrfs_dev_item);
2530 ret = read_one_dev(root, leaf, dev_item);
2533 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2534 struct btrfs_chunk *chunk;
2535 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2536 ret = read_one_chunk(root, &found_key, leaf, chunk);
2540 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2542 btrfs_release_path(root, path);
2546 btrfs_free_path(path);