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 struct bio *old_head;
206 spin_lock(&device->io_lock);
208 old_head = device->pending_bios;
209 device->pending_bios = pending;
210 if (device->pending_bio_tail)
211 tail->bi_next = old_head;
213 device->pending_bio_tail = tail;
215 spin_unlock(&device->io_lock);
216 btrfs_requeue_work(&device->work);
226 void pending_bios_fn(struct btrfs_work *work)
228 struct btrfs_device *device;
230 device = container_of(work, struct btrfs_device, work);
231 run_scheduled_bios(device);
234 static noinline int device_list_add(const char *path,
235 struct btrfs_super_block *disk_super,
236 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
238 struct btrfs_device *device;
239 struct btrfs_fs_devices *fs_devices;
240 u64 found_transid = btrfs_super_generation(disk_super);
242 fs_devices = find_fsid(disk_super->fsid);
244 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
247 INIT_LIST_HEAD(&fs_devices->devices);
248 INIT_LIST_HEAD(&fs_devices->alloc_list);
249 list_add(&fs_devices->list, &fs_uuids);
250 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
251 fs_devices->latest_devid = devid;
252 fs_devices->latest_trans = found_transid;
255 device = __find_device(&fs_devices->devices, devid,
256 disk_super->dev_item.uuid);
259 device = kzalloc(sizeof(*device), GFP_NOFS);
261 /* we can safely leave the fs_devices entry around */
264 device->devid = devid;
265 device->work.func = pending_bios_fn;
266 memcpy(device->uuid, disk_super->dev_item.uuid,
268 device->barriers = 1;
269 spin_lock_init(&device->io_lock);
270 device->name = kstrdup(path, GFP_NOFS);
275 list_add(&device->dev_list, &fs_devices->devices);
276 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
277 fs_devices->num_devices++;
280 if (found_transid > fs_devices->latest_trans) {
281 fs_devices->latest_devid = devid;
282 fs_devices->latest_trans = found_transid;
284 *fs_devices_ret = fs_devices;
288 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
290 struct list_head *head = &fs_devices->devices;
291 struct list_head *cur;
292 struct btrfs_device *device;
294 mutex_lock(&uuid_mutex);
296 list_for_each(cur, head) {
297 device = list_entry(cur, struct btrfs_device, dev_list);
298 if (!device->in_fs_metadata) {
299 struct block_device *bdev;
300 list_del(&device->dev_list);
301 list_del(&device->dev_alloc_list);
302 fs_devices->num_devices--;
305 fs_devices->open_devices--;
306 mutex_unlock(&uuid_mutex);
307 close_bdev_excl(bdev);
308 mutex_lock(&uuid_mutex);
315 mutex_unlock(&uuid_mutex);
319 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
321 struct list_head *head = &fs_devices->devices;
322 struct list_head *cur;
323 struct btrfs_device *device;
325 mutex_lock(&uuid_mutex);
326 list_for_each(cur, head) {
327 device = list_entry(cur, struct btrfs_device, dev_list);
329 close_bdev_excl(device->bdev);
330 fs_devices->open_devices--;
333 device->in_fs_metadata = 0;
335 fs_devices->mounted = 0;
336 mutex_unlock(&uuid_mutex);
340 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
341 int flags, void *holder)
343 struct block_device *bdev;
344 struct list_head *head = &fs_devices->devices;
345 struct list_head *cur;
346 struct btrfs_device *device;
347 struct block_device *latest_bdev = NULL;
348 struct buffer_head *bh;
349 struct btrfs_super_block *disk_super;
350 u64 latest_devid = 0;
351 u64 latest_transid = 0;
356 mutex_lock(&uuid_mutex);
357 if (fs_devices->mounted)
360 list_for_each(cur, head) {
361 device = list_entry(cur, struct btrfs_device, dev_list);
368 bdev = open_bdev_excl(device->name, flags, holder);
371 printk("open %s failed\n", device->name);
374 set_blocksize(bdev, 4096);
376 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
380 disk_super = (struct btrfs_super_block *)bh->b_data;
381 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
382 sizeof(disk_super->magic)))
385 devid = le64_to_cpu(disk_super->dev_item.devid);
386 if (devid != device->devid)
389 transid = btrfs_super_generation(disk_super);
390 if (!latest_transid || transid > latest_transid) {
391 latest_devid = devid;
392 latest_transid = transid;
397 device->in_fs_metadata = 0;
398 fs_devices->open_devices++;
404 close_bdev_excl(bdev);
408 if (fs_devices->open_devices == 0) {
412 fs_devices->mounted = 1;
413 fs_devices->latest_bdev = latest_bdev;
414 fs_devices->latest_devid = latest_devid;
415 fs_devices->latest_trans = latest_transid;
417 mutex_unlock(&uuid_mutex);
421 int btrfs_scan_one_device(const char *path, int flags, void *holder,
422 struct btrfs_fs_devices **fs_devices_ret)
424 struct btrfs_super_block *disk_super;
425 struct block_device *bdev;
426 struct buffer_head *bh;
431 mutex_lock(&uuid_mutex);
433 bdev = open_bdev_excl(path, flags, holder);
440 ret = set_blocksize(bdev, 4096);
443 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
448 disk_super = (struct btrfs_super_block *)bh->b_data;
449 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
450 sizeof(disk_super->magic))) {
454 devid = le64_to_cpu(disk_super->dev_item.devid);
455 transid = btrfs_super_generation(disk_super);
456 if (disk_super->label[0])
457 printk("device label %s ", disk_super->label);
459 /* FIXME, make a readl uuid parser */
460 printk("device fsid %llx-%llx ",
461 *(unsigned long long *)disk_super->fsid,
462 *(unsigned long long *)(disk_super->fsid + 8));
464 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
465 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
470 close_bdev_excl(bdev);
472 mutex_unlock(&uuid_mutex);
477 * this uses a pretty simple search, the expectation is that it is
478 * called very infrequently and that a given device has a small number
481 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
482 struct btrfs_device *device,
483 struct btrfs_path *path,
484 u64 num_bytes, u64 *start)
486 struct btrfs_key key;
487 struct btrfs_root *root = device->dev_root;
488 struct btrfs_dev_extent *dev_extent = NULL;
491 u64 search_start = 0;
492 u64 search_end = device->total_bytes;
496 struct extent_buffer *l;
501 /* FIXME use last free of some kind */
503 /* we don't want to overwrite the superblock on the drive,
504 * so we make sure to start at an offset of at least 1MB
506 search_start = max((u64)1024 * 1024, search_start);
508 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
509 search_start = max(root->fs_info->alloc_start, search_start);
511 key.objectid = device->devid;
512 key.offset = search_start;
513 key.type = BTRFS_DEV_EXTENT_KEY;
514 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
517 ret = btrfs_previous_item(root, path, 0, key.type);
521 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
524 slot = path->slots[0];
525 if (slot >= btrfs_header_nritems(l)) {
526 ret = btrfs_next_leaf(root, path);
533 if (search_start >= search_end) {
537 *start = search_start;
541 *start = last_byte > search_start ?
542 last_byte : search_start;
543 if (search_end <= *start) {
549 btrfs_item_key_to_cpu(l, &key, slot);
551 if (key.objectid < device->devid)
554 if (key.objectid > device->devid)
557 if (key.offset >= search_start && key.offset > last_byte &&
559 if (last_byte < search_start)
560 last_byte = search_start;
561 hole_size = key.offset - last_byte;
562 if (key.offset > last_byte &&
563 hole_size >= num_bytes) {
568 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
573 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
574 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
580 /* we have to make sure we didn't find an extent that has already
581 * been allocated by the map tree or the original allocation
583 btrfs_release_path(root, path);
584 BUG_ON(*start < search_start);
586 if (*start + num_bytes > search_end) {
590 /* check for pending inserts here */
594 btrfs_release_path(root, path);
598 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
599 struct btrfs_device *device,
603 struct btrfs_path *path;
604 struct btrfs_root *root = device->dev_root;
605 struct btrfs_key key;
606 struct btrfs_key found_key;
607 struct extent_buffer *leaf = NULL;
608 struct btrfs_dev_extent *extent = NULL;
610 path = btrfs_alloc_path();
614 key.objectid = device->devid;
616 key.type = BTRFS_DEV_EXTENT_KEY;
618 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
620 ret = btrfs_previous_item(root, path, key.objectid,
621 BTRFS_DEV_EXTENT_KEY);
623 leaf = path->nodes[0];
624 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
625 extent = btrfs_item_ptr(leaf, path->slots[0],
626 struct btrfs_dev_extent);
627 BUG_ON(found_key.offset > start || found_key.offset +
628 btrfs_dev_extent_length(leaf, extent) < start);
630 } else if (ret == 0) {
631 leaf = path->nodes[0];
632 extent = btrfs_item_ptr(leaf, path->slots[0],
633 struct btrfs_dev_extent);
637 if (device->bytes_used > 0)
638 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
639 ret = btrfs_del_item(trans, root, path);
642 btrfs_free_path(path);
646 int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
647 struct btrfs_device *device,
648 u64 chunk_tree, u64 chunk_objectid,
650 u64 num_bytes, u64 *start)
653 struct btrfs_path *path;
654 struct btrfs_root *root = device->dev_root;
655 struct btrfs_dev_extent *extent;
656 struct extent_buffer *leaf;
657 struct btrfs_key key;
659 WARN_ON(!device->in_fs_metadata);
660 path = btrfs_alloc_path();
664 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
669 key.objectid = device->devid;
671 key.type = BTRFS_DEV_EXTENT_KEY;
672 ret = btrfs_insert_empty_item(trans, root, path, &key,
676 leaf = path->nodes[0];
677 extent = btrfs_item_ptr(leaf, path->slots[0],
678 struct btrfs_dev_extent);
679 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
680 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
681 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
683 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
684 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
687 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
688 btrfs_mark_buffer_dirty(leaf);
690 btrfs_free_path(path);
694 static noinline int find_next_chunk(struct btrfs_root *root,
695 u64 objectid, u64 *offset)
697 struct btrfs_path *path;
699 struct btrfs_key key;
700 struct btrfs_chunk *chunk;
701 struct btrfs_key found_key;
703 path = btrfs_alloc_path();
706 key.objectid = objectid;
707 key.offset = (u64)-1;
708 key.type = BTRFS_CHUNK_ITEM_KEY;
710 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
716 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
720 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
722 if (found_key.objectid != objectid)
725 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
727 *offset = found_key.offset +
728 btrfs_chunk_length(path->nodes[0], chunk);
733 btrfs_free_path(path);
737 static noinline int find_next_devid(struct btrfs_root *root,
738 struct btrfs_path *path, u64 *objectid)
741 struct btrfs_key key;
742 struct btrfs_key found_key;
744 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
745 key.type = BTRFS_DEV_ITEM_KEY;
746 key.offset = (u64)-1;
748 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
754 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
759 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
761 *objectid = found_key.offset + 1;
765 btrfs_release_path(root, path);
770 * the device information is stored in the chunk root
771 * the btrfs_device struct should be fully filled in
773 int btrfs_add_device(struct btrfs_trans_handle *trans,
774 struct btrfs_root *root,
775 struct btrfs_device *device)
778 struct btrfs_path *path;
779 struct btrfs_dev_item *dev_item;
780 struct extent_buffer *leaf;
781 struct btrfs_key key;
785 root = root->fs_info->chunk_root;
787 path = btrfs_alloc_path();
791 ret = find_next_devid(root, path, &free_devid);
795 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
796 key.type = BTRFS_DEV_ITEM_KEY;
797 key.offset = free_devid;
799 ret = btrfs_insert_empty_item(trans, root, path, &key,
804 leaf = path->nodes[0];
805 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
807 device->devid = free_devid;
808 btrfs_set_device_id(leaf, dev_item, device->devid);
809 btrfs_set_device_type(leaf, dev_item, device->type);
810 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
811 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
812 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
813 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
814 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
815 btrfs_set_device_group(leaf, dev_item, 0);
816 btrfs_set_device_seek_speed(leaf, dev_item, 0);
817 btrfs_set_device_bandwidth(leaf, dev_item, 0);
819 ptr = (unsigned long)btrfs_device_uuid(dev_item);
820 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
821 btrfs_mark_buffer_dirty(leaf);
825 btrfs_free_path(path);
829 static int btrfs_rm_dev_item(struct btrfs_root *root,
830 struct btrfs_device *device)
833 struct btrfs_path *path;
834 struct block_device *bdev = device->bdev;
835 struct btrfs_device *next_dev;
836 struct btrfs_key key;
838 struct btrfs_fs_devices *fs_devices;
839 struct btrfs_trans_handle *trans;
841 root = root->fs_info->chunk_root;
843 path = btrfs_alloc_path();
847 trans = btrfs_start_transaction(root, 1);
848 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
849 key.type = BTRFS_DEV_ITEM_KEY;
850 key.offset = device->devid;
853 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
862 ret = btrfs_del_item(trans, root, path);
867 * at this point, the device is zero sized. We want to
868 * remove it from the devices list and zero out the old super
870 list_del_init(&device->dev_list);
871 list_del_init(&device->dev_alloc_list);
872 fs_devices = root->fs_info->fs_devices;
874 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
876 if (bdev == root->fs_info->sb->s_bdev)
877 root->fs_info->sb->s_bdev = next_dev->bdev;
878 if (bdev == fs_devices->latest_bdev)
879 fs_devices->latest_bdev = next_dev->bdev;
881 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
882 btrfs_set_super_num_devices(&root->fs_info->super_copy,
885 btrfs_free_path(path);
887 btrfs_commit_transaction(trans, root);
891 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
893 struct btrfs_device *device;
894 struct block_device *bdev;
895 struct buffer_head *bh = NULL;
896 struct btrfs_super_block *disk_super;
901 mutex_lock(&uuid_mutex);
902 mutex_lock(&root->fs_info->volume_mutex);
904 all_avail = root->fs_info->avail_data_alloc_bits |
905 root->fs_info->avail_system_alloc_bits |
906 root->fs_info->avail_metadata_alloc_bits;
908 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
909 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
910 printk("btrfs: unable to go below four devices on raid10\n");
915 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
916 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
917 printk("btrfs: unable to go below two devices on raid1\n");
922 if (strcmp(device_path, "missing") == 0) {
923 struct list_head *cur;
924 struct list_head *devices;
925 struct btrfs_device *tmp;
928 devices = &root->fs_info->fs_devices->devices;
929 list_for_each(cur, devices) {
930 tmp = list_entry(cur, struct btrfs_device, dev_list);
931 if (tmp->in_fs_metadata && !tmp->bdev) {
940 printk("btrfs: no missing devices found to remove\n");
945 bdev = open_bdev_excl(device_path, 0,
946 root->fs_info->bdev_holder);
952 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
957 disk_super = (struct btrfs_super_block *)bh->b_data;
958 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
959 sizeof(disk_super->magic))) {
963 if (memcmp(disk_super->fsid, root->fs_info->fsid,
968 devid = le64_to_cpu(disk_super->dev_item.devid);
969 device = btrfs_find_device(root, devid, NULL);
976 root->fs_info->fs_devices->num_devices--;
977 root->fs_info->fs_devices->open_devices--;
979 ret = btrfs_shrink_device(device, 0);
984 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
989 /* make sure this device isn't detected as part of
992 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
993 set_buffer_dirty(bh);
994 sync_dirty_buffer(bh);
1000 /* one close for the device struct or super_block */
1001 close_bdev_excl(device->bdev);
1004 /* one close for us */
1005 close_bdev_excl(bdev);
1007 kfree(device->name);
1016 close_bdev_excl(bdev);
1018 mutex_unlock(&root->fs_info->volume_mutex);
1019 mutex_unlock(&uuid_mutex);
1023 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1025 struct btrfs_trans_handle *trans;
1026 struct btrfs_device *device;
1027 struct block_device *bdev;
1028 struct list_head *cur;
1029 struct list_head *devices;
1034 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1039 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1040 mutex_lock(&root->fs_info->volume_mutex);
1042 trans = btrfs_start_transaction(root, 1);
1044 devices = &root->fs_info->fs_devices->devices;
1045 list_for_each(cur, devices) {
1046 device = list_entry(cur, struct btrfs_device, dev_list);
1047 if (device->bdev == bdev) {
1053 device = kzalloc(sizeof(*device), GFP_NOFS);
1055 /* we can safely leave the fs_devices entry around */
1057 goto out_close_bdev;
1060 device->barriers = 1;
1061 device->work.func = pending_bios_fn;
1062 generate_random_uuid(device->uuid);
1063 spin_lock_init(&device->io_lock);
1064 device->name = kstrdup(device_path, GFP_NOFS);
1065 if (!device->name) {
1067 goto out_close_bdev;
1069 device->io_width = root->sectorsize;
1070 device->io_align = root->sectorsize;
1071 device->sector_size = root->sectorsize;
1072 device->total_bytes = i_size_read(bdev->bd_inode);
1073 device->dev_root = root->fs_info->dev_root;
1074 device->bdev = bdev;
1075 device->in_fs_metadata = 1;
1077 ret = btrfs_add_device(trans, root, device);
1079 goto out_close_bdev;
1081 set_blocksize(device->bdev, 4096);
1083 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1084 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1085 total_bytes + device->total_bytes);
1087 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1088 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1091 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1092 list_add(&device->dev_alloc_list,
1093 &root->fs_info->fs_devices->alloc_list);
1094 root->fs_info->fs_devices->num_devices++;
1095 root->fs_info->fs_devices->open_devices++;
1097 unlock_chunks(root);
1098 btrfs_end_transaction(trans, root);
1099 mutex_unlock(&root->fs_info->volume_mutex);
1104 close_bdev_excl(bdev);
1108 int noinline btrfs_update_device(struct btrfs_trans_handle *trans,
1109 struct btrfs_device *device)
1112 struct btrfs_path *path;
1113 struct btrfs_root *root;
1114 struct btrfs_dev_item *dev_item;
1115 struct extent_buffer *leaf;
1116 struct btrfs_key key;
1118 root = device->dev_root->fs_info->chunk_root;
1120 path = btrfs_alloc_path();
1124 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1125 key.type = BTRFS_DEV_ITEM_KEY;
1126 key.offset = device->devid;
1128 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1137 leaf = path->nodes[0];
1138 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1140 btrfs_set_device_id(leaf, dev_item, device->devid);
1141 btrfs_set_device_type(leaf, dev_item, device->type);
1142 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1143 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1144 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1145 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1146 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1147 btrfs_mark_buffer_dirty(leaf);
1150 btrfs_free_path(path);
1154 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1155 struct btrfs_device *device, u64 new_size)
1157 struct btrfs_super_block *super_copy =
1158 &device->dev_root->fs_info->super_copy;
1159 u64 old_total = btrfs_super_total_bytes(super_copy);
1160 u64 diff = new_size - device->total_bytes;
1162 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1163 return btrfs_update_device(trans, device);
1166 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1167 struct btrfs_device *device, u64 new_size)
1170 lock_chunks(device->dev_root);
1171 ret = __btrfs_grow_device(trans, device, new_size);
1172 unlock_chunks(device->dev_root);
1176 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1177 struct btrfs_root *root,
1178 u64 chunk_tree, u64 chunk_objectid,
1182 struct btrfs_path *path;
1183 struct btrfs_key key;
1185 root = root->fs_info->chunk_root;
1186 path = btrfs_alloc_path();
1190 key.objectid = chunk_objectid;
1191 key.offset = chunk_offset;
1192 key.type = BTRFS_CHUNK_ITEM_KEY;
1194 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1197 ret = btrfs_del_item(trans, root, path);
1200 btrfs_free_path(path);
1204 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1207 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1208 struct btrfs_disk_key *disk_key;
1209 struct btrfs_chunk *chunk;
1216 struct btrfs_key key;
1218 array_size = btrfs_super_sys_array_size(super_copy);
1220 ptr = super_copy->sys_chunk_array;
1223 while (cur < array_size) {
1224 disk_key = (struct btrfs_disk_key *)ptr;
1225 btrfs_disk_key_to_cpu(&key, disk_key);
1227 len = sizeof(*disk_key);
1229 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1230 chunk = (struct btrfs_chunk *)(ptr + len);
1231 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1232 len += btrfs_chunk_item_size(num_stripes);
1237 if (key.objectid == chunk_objectid &&
1238 key.offset == chunk_offset) {
1239 memmove(ptr, ptr + len, array_size - (cur + len));
1241 btrfs_set_super_sys_array_size(super_copy, array_size);
1251 int btrfs_relocate_chunk(struct btrfs_root *root,
1252 u64 chunk_tree, u64 chunk_objectid,
1255 struct extent_map_tree *em_tree;
1256 struct btrfs_root *extent_root;
1257 struct btrfs_trans_handle *trans;
1258 struct extent_map *em;
1259 struct map_lookup *map;
1263 printk("btrfs relocating chunk %llu\n",
1264 (unsigned long long)chunk_offset);
1265 root = root->fs_info->chunk_root;
1266 extent_root = root->fs_info->extent_root;
1267 em_tree = &root->fs_info->mapping_tree.map_tree;
1269 /* step one, relocate all the extents inside this chunk */
1270 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1273 trans = btrfs_start_transaction(root, 1);
1279 * step two, delete the device extents and the
1280 * chunk tree entries
1282 spin_lock(&em_tree->lock);
1283 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1284 spin_unlock(&em_tree->lock);
1286 BUG_ON(em->start > chunk_offset ||
1287 em->start + em->len < chunk_offset);
1288 map = (struct map_lookup *)em->bdev;
1290 for (i = 0; i < map->num_stripes; i++) {
1291 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1292 map->stripes[i].physical);
1295 if (map->stripes[i].dev) {
1296 ret = btrfs_update_device(trans, map->stripes[i].dev);
1300 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1305 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1306 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1310 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1313 spin_lock(&em_tree->lock);
1314 remove_extent_mapping(em_tree, em);
1315 spin_unlock(&em_tree->lock);
1320 /* once for the tree */
1321 free_extent_map(em);
1323 free_extent_map(em);
1325 unlock_chunks(root);
1326 btrfs_end_transaction(trans, root);
1330 static u64 div_factor(u64 num, int factor)
1340 int btrfs_balance(struct btrfs_root *dev_root)
1343 struct list_head *cur;
1344 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1345 struct btrfs_device *device;
1348 struct btrfs_path *path;
1349 struct btrfs_key key;
1350 struct btrfs_chunk *chunk;
1351 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1352 struct btrfs_trans_handle *trans;
1353 struct btrfs_key found_key;
1356 mutex_lock(&dev_root->fs_info->volume_mutex);
1357 dev_root = dev_root->fs_info->dev_root;
1359 /* step one make some room on all the devices */
1360 list_for_each(cur, devices) {
1361 device = list_entry(cur, struct btrfs_device, dev_list);
1362 old_size = device->total_bytes;
1363 size_to_free = div_factor(old_size, 1);
1364 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1365 if (device->total_bytes - device->bytes_used > size_to_free)
1368 ret = btrfs_shrink_device(device, old_size - size_to_free);
1371 trans = btrfs_start_transaction(dev_root, 1);
1374 ret = btrfs_grow_device(trans, device, old_size);
1377 btrfs_end_transaction(trans, dev_root);
1380 /* step two, relocate all the chunks */
1381 path = btrfs_alloc_path();
1384 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1385 key.offset = (u64)-1;
1386 key.type = BTRFS_CHUNK_ITEM_KEY;
1389 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1394 * this shouldn't happen, it means the last relocate
1400 ret = btrfs_previous_item(chunk_root, path, 0,
1401 BTRFS_CHUNK_ITEM_KEY);
1405 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1407 if (found_key.objectid != key.objectid)
1410 chunk = btrfs_item_ptr(path->nodes[0],
1412 struct btrfs_chunk);
1413 key.offset = found_key.offset;
1414 /* chunk zero is special */
1415 if (key.offset == 0)
1418 btrfs_release_path(chunk_root, path);
1419 ret = btrfs_relocate_chunk(chunk_root,
1420 chunk_root->root_key.objectid,
1427 btrfs_free_path(path);
1428 mutex_unlock(&dev_root->fs_info->volume_mutex);
1433 * shrinking a device means finding all of the device extents past
1434 * the new size, and then following the back refs to the chunks.
1435 * The chunk relocation code actually frees the device extent
1437 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1439 struct btrfs_trans_handle *trans;
1440 struct btrfs_root *root = device->dev_root;
1441 struct btrfs_dev_extent *dev_extent = NULL;
1442 struct btrfs_path *path;
1449 struct extent_buffer *l;
1450 struct btrfs_key key;
1451 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1452 u64 old_total = btrfs_super_total_bytes(super_copy);
1453 u64 diff = device->total_bytes - new_size;
1456 path = btrfs_alloc_path();
1460 trans = btrfs_start_transaction(root, 1);
1470 device->total_bytes = new_size;
1471 ret = btrfs_update_device(trans, device);
1473 unlock_chunks(root);
1474 btrfs_end_transaction(trans, root);
1477 WARN_ON(diff > old_total);
1478 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1479 unlock_chunks(root);
1480 btrfs_end_transaction(trans, root);
1482 key.objectid = device->devid;
1483 key.offset = (u64)-1;
1484 key.type = BTRFS_DEV_EXTENT_KEY;
1487 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1491 ret = btrfs_previous_item(root, path, 0, key.type);
1500 slot = path->slots[0];
1501 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1503 if (key.objectid != device->devid)
1506 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1507 length = btrfs_dev_extent_length(l, dev_extent);
1509 if (key.offset + length <= new_size)
1512 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1513 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1514 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1515 btrfs_release_path(root, path);
1517 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1524 btrfs_free_path(path);
1528 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1529 struct btrfs_root *root,
1530 struct btrfs_key *key,
1531 struct btrfs_chunk *chunk, int item_size)
1533 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1534 struct btrfs_disk_key disk_key;
1538 array_size = btrfs_super_sys_array_size(super_copy);
1539 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1542 ptr = super_copy->sys_chunk_array + array_size;
1543 btrfs_cpu_key_to_disk(&disk_key, key);
1544 memcpy(ptr, &disk_key, sizeof(disk_key));
1545 ptr += sizeof(disk_key);
1546 memcpy(ptr, chunk, item_size);
1547 item_size += sizeof(disk_key);
1548 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1552 static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size,
1553 int num_stripes, int sub_stripes)
1555 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1557 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1558 return calc_size * (num_stripes / sub_stripes);
1560 return calc_size * num_stripes;
1564 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1565 struct btrfs_root *extent_root, u64 *start,
1566 u64 *num_bytes, u64 type)
1569 struct btrfs_fs_info *info = extent_root->fs_info;
1570 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1571 struct btrfs_path *path;
1572 struct btrfs_stripe *stripes;
1573 struct btrfs_device *device = NULL;
1574 struct btrfs_chunk *chunk;
1575 struct list_head private_devs;
1576 struct list_head *dev_list;
1577 struct list_head *cur;
1578 struct extent_map_tree *em_tree;
1579 struct map_lookup *map;
1580 struct extent_map *em;
1581 int min_stripe_size = 1 * 1024 * 1024;
1583 u64 calc_size = 1024 * 1024 * 1024;
1584 u64 max_chunk_size = calc_size;
1589 int num_stripes = 1;
1590 int min_stripes = 1;
1591 int sub_stripes = 0;
1595 int stripe_len = 64 * 1024;
1596 struct btrfs_key key;
1598 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1599 (type & BTRFS_BLOCK_GROUP_DUP)) {
1601 type &= ~BTRFS_BLOCK_GROUP_DUP;
1603 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1604 if (list_empty(dev_list))
1607 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1608 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1611 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1615 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1616 num_stripes = min_t(u64, 2,
1617 extent_root->fs_info->fs_devices->open_devices);
1618 if (num_stripes < 2)
1622 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1623 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1624 if (num_stripes < 4)
1626 num_stripes &= ~(u32)1;
1631 if (type & BTRFS_BLOCK_GROUP_DATA) {
1632 max_chunk_size = 10 * calc_size;
1633 min_stripe_size = 64 * 1024 * 1024;
1634 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1635 max_chunk_size = 4 * calc_size;
1636 min_stripe_size = 32 * 1024 * 1024;
1637 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1638 calc_size = 8 * 1024 * 1024;
1639 max_chunk_size = calc_size * 2;
1640 min_stripe_size = 1 * 1024 * 1024;
1643 path = btrfs_alloc_path();
1647 /* we don't want a chunk larger than 10% of the FS */
1648 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1649 max_chunk_size = min(percent_max, max_chunk_size);
1652 if (calc_size * num_stripes > max_chunk_size) {
1653 calc_size = max_chunk_size;
1654 do_div(calc_size, num_stripes);
1655 do_div(calc_size, stripe_len);
1656 calc_size *= stripe_len;
1658 /* we don't want tiny stripes */
1659 calc_size = max_t(u64, min_stripe_size, calc_size);
1661 do_div(calc_size, stripe_len);
1662 calc_size *= stripe_len;
1664 INIT_LIST_HEAD(&private_devs);
1665 cur = dev_list->next;
1668 if (type & BTRFS_BLOCK_GROUP_DUP)
1669 min_free = calc_size * 2;
1671 min_free = calc_size;
1674 * we add 1MB because we never use the first 1MB of the device, unless
1675 * we've looped, then we are likely allocating the maximum amount of
1676 * space left already
1679 min_free += 1024 * 1024;
1681 /* build a private list of devices we will allocate from */
1682 while(index < num_stripes) {
1683 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1685 if (device->total_bytes > device->bytes_used)
1686 avail = device->total_bytes - device->bytes_used;
1691 if (device->in_fs_metadata && avail >= min_free) {
1692 u64 ignored_start = 0;
1693 ret = find_free_dev_extent(trans, device, path,
1697 list_move_tail(&device->dev_alloc_list,
1700 if (type & BTRFS_BLOCK_GROUP_DUP)
1703 } else if (device->in_fs_metadata && avail > max_avail)
1705 if (cur == dev_list)
1708 if (index < num_stripes) {
1709 list_splice(&private_devs, dev_list);
1710 if (index >= min_stripes) {
1711 num_stripes = index;
1712 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1713 num_stripes /= sub_stripes;
1714 num_stripes *= sub_stripes;
1719 if (!looped && max_avail > 0) {
1721 calc_size = max_avail;
1724 btrfs_free_path(path);
1727 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1728 key.type = BTRFS_CHUNK_ITEM_KEY;
1729 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1732 btrfs_free_path(path);
1736 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1738 btrfs_free_path(path);
1742 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1745 btrfs_free_path(path);
1748 btrfs_free_path(path);
1751 stripes = &chunk->stripe;
1752 *num_bytes = chunk_bytes_by_type(type, calc_size,
1753 num_stripes, sub_stripes);
1756 while(index < num_stripes) {
1757 struct btrfs_stripe *stripe;
1758 BUG_ON(list_empty(&private_devs));
1759 cur = private_devs.next;
1760 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1762 /* loop over this device again if we're doing a dup group */
1763 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1764 (index == num_stripes - 1))
1765 list_move_tail(&device->dev_alloc_list, dev_list);
1767 ret = btrfs_alloc_dev_extent(trans, device,
1768 info->chunk_root->root_key.objectid,
1769 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1770 calc_size, &dev_offset);
1772 device->bytes_used += calc_size;
1773 ret = btrfs_update_device(trans, device);
1776 map->stripes[index].dev = device;
1777 map->stripes[index].physical = dev_offset;
1778 stripe = stripes + index;
1779 btrfs_set_stack_stripe_devid(stripe, device->devid);
1780 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1781 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1782 physical = dev_offset;
1785 BUG_ON(!list_empty(&private_devs));
1787 /* key was set above */
1788 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1789 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1790 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1791 btrfs_set_stack_chunk_type(chunk, type);
1792 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1793 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1794 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1795 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1796 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1797 map->sector_size = extent_root->sectorsize;
1798 map->stripe_len = stripe_len;
1799 map->io_align = stripe_len;
1800 map->io_width = stripe_len;
1802 map->num_stripes = num_stripes;
1803 map->sub_stripes = sub_stripes;
1805 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1806 btrfs_chunk_item_size(num_stripes));
1808 *start = key.offset;;
1810 em = alloc_extent_map(GFP_NOFS);
1813 em->bdev = (struct block_device *)map;
1814 em->start = key.offset;
1815 em->len = *num_bytes;
1816 em->block_start = 0;
1817 em->block_len = em->len;
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 = (u64)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;
2325 em->block_len = em->len;
2327 map->num_stripes = num_stripes;
2328 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2329 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2330 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2331 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2332 map->type = btrfs_chunk_type(leaf, chunk);
2333 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2334 for (i = 0; i < num_stripes; i++) {
2335 map->stripes[i].physical =
2336 btrfs_stripe_offset_nr(leaf, chunk, i);
2337 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2338 read_extent_buffer(leaf, uuid, (unsigned long)
2339 btrfs_stripe_dev_uuid_nr(chunk, i),
2341 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2343 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2345 free_extent_map(em);
2348 if (!map->stripes[i].dev) {
2349 map->stripes[i].dev =
2350 add_missing_dev(root, devid, uuid);
2351 if (!map->stripes[i].dev) {
2353 free_extent_map(em);
2357 map->stripes[i].dev->in_fs_metadata = 1;
2360 spin_lock(&map_tree->map_tree.lock);
2361 ret = add_extent_mapping(&map_tree->map_tree, em);
2362 spin_unlock(&map_tree->map_tree.lock);
2364 free_extent_map(em);
2369 static int fill_device_from_item(struct extent_buffer *leaf,
2370 struct btrfs_dev_item *dev_item,
2371 struct btrfs_device *device)
2375 device->devid = btrfs_device_id(leaf, dev_item);
2376 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2377 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2378 device->type = btrfs_device_type(leaf, dev_item);
2379 device->io_align = btrfs_device_io_align(leaf, dev_item);
2380 device->io_width = btrfs_device_io_width(leaf, dev_item);
2381 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2383 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2384 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2389 static int read_one_dev(struct btrfs_root *root,
2390 struct extent_buffer *leaf,
2391 struct btrfs_dev_item *dev_item)
2393 struct btrfs_device *device;
2396 u8 dev_uuid[BTRFS_UUID_SIZE];
2398 devid = btrfs_device_id(leaf, dev_item);
2399 read_extent_buffer(leaf, dev_uuid,
2400 (unsigned long)btrfs_device_uuid(dev_item),
2402 device = btrfs_find_device(root, devid, dev_uuid);
2404 printk("warning devid %Lu missing\n", devid);
2405 device = add_missing_dev(root, devid, dev_uuid);
2410 fill_device_from_item(leaf, dev_item, device);
2411 device->dev_root = root->fs_info->dev_root;
2412 device->in_fs_metadata = 1;
2415 ret = btrfs_open_device(device);
2423 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2425 struct btrfs_dev_item *dev_item;
2427 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2429 return read_one_dev(root, buf, dev_item);
2432 int btrfs_read_sys_array(struct btrfs_root *root)
2434 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2435 struct extent_buffer *sb;
2436 struct btrfs_disk_key *disk_key;
2437 struct btrfs_chunk *chunk;
2439 unsigned long sb_ptr;
2445 struct btrfs_key key;
2447 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2448 BTRFS_SUPER_INFO_SIZE);
2451 btrfs_set_buffer_uptodate(sb);
2452 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2453 array_size = btrfs_super_sys_array_size(super_copy);
2455 ptr = super_copy->sys_chunk_array;
2456 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2459 while (cur < array_size) {
2460 disk_key = (struct btrfs_disk_key *)ptr;
2461 btrfs_disk_key_to_cpu(&key, disk_key);
2463 len = sizeof(*disk_key); ptr += len;
2467 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2468 chunk = (struct btrfs_chunk *)sb_ptr;
2469 ret = read_one_chunk(root, &key, sb, chunk);
2472 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2473 len = btrfs_chunk_item_size(num_stripes);
2482 free_extent_buffer(sb);
2486 int btrfs_read_chunk_tree(struct btrfs_root *root)
2488 struct btrfs_path *path;
2489 struct extent_buffer *leaf;
2490 struct btrfs_key key;
2491 struct btrfs_key found_key;
2495 root = root->fs_info->chunk_root;
2497 path = btrfs_alloc_path();
2501 /* first we search for all of the device items, and then we
2502 * read in all of the chunk items. This way we can create chunk
2503 * mappings that reference all of the devices that are afound
2505 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2511 leaf = path->nodes[0];
2512 slot = path->slots[0];
2513 if (slot >= btrfs_header_nritems(leaf)) {
2514 ret = btrfs_next_leaf(root, path);
2521 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2522 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2523 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2525 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2526 struct btrfs_dev_item *dev_item;
2527 dev_item = btrfs_item_ptr(leaf, slot,
2528 struct btrfs_dev_item);
2529 ret = read_one_dev(root, leaf, dev_item);
2532 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2533 struct btrfs_chunk *chunk;
2534 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2535 ret = read_one_chunk(root, &found_key, leaf, chunk);
2539 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2541 btrfs_release_path(root, path);
2545 btrfs_free_path(path);