2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <linux/random.h>
23 #include <asm/div64.h>
25 #include "extent_map.h"
27 #include "transaction.h"
28 #include "print-tree.h"
30 #include "async-thread.h"
40 struct btrfs_bio_stripe stripes[];
43 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
44 (sizeof(struct btrfs_bio_stripe) * (n)))
46 static DEFINE_MUTEX(uuid_mutex);
47 static LIST_HEAD(fs_uuids);
49 void btrfs_lock_volumes(void)
51 mutex_lock(&uuid_mutex);
54 void btrfs_unlock_volumes(void)
56 mutex_unlock(&uuid_mutex);
59 static void lock_chunks(struct btrfs_root *root)
61 mutex_lock(&root->fs_info->alloc_mutex);
62 mutex_lock(&root->fs_info->chunk_mutex);
65 static void unlock_chunks(struct btrfs_root *root)
67 mutex_unlock(&root->fs_info->alloc_mutex);
68 mutex_unlock(&root->fs_info->chunk_mutex);
71 int btrfs_cleanup_fs_uuids(void)
73 struct btrfs_fs_devices *fs_devices;
74 struct list_head *uuid_cur;
75 struct list_head *devices_cur;
76 struct btrfs_device *dev;
78 list_for_each(uuid_cur, &fs_uuids) {
79 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
81 while(!list_empty(&fs_devices->devices)) {
82 devices_cur = fs_devices->devices.next;
83 dev = list_entry(devices_cur, struct btrfs_device,
86 close_bdev_excl(dev->bdev);
87 fs_devices->open_devices--;
89 list_del(&dev->dev_list);
97 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
100 struct btrfs_device *dev;
101 struct list_head *cur;
103 list_for_each(cur, head) {
104 dev = list_entry(cur, struct btrfs_device, dev_list);
105 if (dev->devid == devid &&
106 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
113 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
115 struct list_head *cur;
116 struct btrfs_fs_devices *fs_devices;
118 list_for_each(cur, &fs_uuids) {
119 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
120 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
127 * we try to collect pending bios for a device so we don't get a large
128 * number of procs sending bios down to the same device. This greatly
129 * improves the schedulers ability to collect and merge the bios.
131 * But, it also turns into a long list of bios to process and that is sure
132 * to eventually make the worker thread block. The solution here is to
133 * make some progress and then put this work struct back at the end of
134 * the list if the block device is congested. This way, multiple devices
135 * can make progress from a single worker thread.
137 int run_scheduled_bios(struct btrfs_device *device)
140 struct backing_dev_info *bdi;
144 unsigned long num_run = 0;
146 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
148 spin_lock(&device->io_lock);
150 /* take all the bios off the list at once and process them
151 * later on (without the lock held). But, remember the
152 * tail and other pointers so the bios can be properly reinserted
153 * into the list if we hit congestion
155 pending = device->pending_bios;
156 tail = device->pending_bio_tail;
157 WARN_ON(pending && !tail);
158 device->pending_bios = NULL;
159 device->pending_bio_tail = NULL;
162 * if pending was null this time around, no bios need processing
163 * at all and we can stop. Otherwise it'll loop back up again
164 * and do an additional check so no bios are missed.
166 * device->running_pending is used to synchronize with the
171 device->running_pending = 1;
174 device->running_pending = 0;
176 spin_unlock(&device->io_lock);
180 pending = pending->bi_next;
182 atomic_dec(&device->dev_root->fs_info->nr_async_submits);
184 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
186 submit_bio(cur->bi_rw, cur);
191 * we made progress, there is more work to do and the bdi
192 * is now congested. Back off and let other work structs
195 if (pending && bdi_write_congested(bdi)) {
196 struct bio *old_head;
198 spin_lock(&device->io_lock);
200 old_head = device->pending_bios;
201 device->pending_bios = pending;
202 if (device->pending_bio_tail)
203 tail->bi_next = old_head;
205 device->pending_bio_tail = tail;
207 spin_unlock(&device->io_lock);
208 btrfs_requeue_work(&device->work);
218 void pending_bios_fn(struct btrfs_work *work)
220 struct btrfs_device *device;
222 device = container_of(work, struct btrfs_device, work);
223 run_scheduled_bios(device);
226 static int device_list_add(const char *path,
227 struct btrfs_super_block *disk_super,
228 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
230 struct btrfs_device *device;
231 struct btrfs_fs_devices *fs_devices;
232 u64 found_transid = btrfs_super_generation(disk_super);
234 fs_devices = find_fsid(disk_super->fsid);
236 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
239 INIT_LIST_HEAD(&fs_devices->devices);
240 INIT_LIST_HEAD(&fs_devices->alloc_list);
241 list_add(&fs_devices->list, &fs_uuids);
242 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
243 fs_devices->latest_devid = devid;
244 fs_devices->latest_trans = found_transid;
247 device = __find_device(&fs_devices->devices, devid,
248 disk_super->dev_item.uuid);
251 device = kzalloc(sizeof(*device), GFP_NOFS);
253 /* we can safely leave the fs_devices entry around */
256 device->devid = devid;
257 device->work.func = pending_bios_fn;
258 memcpy(device->uuid, disk_super->dev_item.uuid,
260 device->barriers = 1;
261 spin_lock_init(&device->io_lock);
262 device->name = kstrdup(path, GFP_NOFS);
267 list_add(&device->dev_list, &fs_devices->devices);
268 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
269 fs_devices->num_devices++;
272 if (found_transid > fs_devices->latest_trans) {
273 fs_devices->latest_devid = devid;
274 fs_devices->latest_trans = found_transid;
276 *fs_devices_ret = fs_devices;
280 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
282 struct list_head *head = &fs_devices->devices;
283 struct list_head *cur;
284 struct btrfs_device *device;
286 mutex_lock(&uuid_mutex);
288 list_for_each(cur, head) {
289 device = list_entry(cur, struct btrfs_device, dev_list);
290 if (!device->in_fs_metadata) {
291 struct block_device *bdev;
292 list_del(&device->dev_list);
293 list_del(&device->dev_alloc_list);
294 fs_devices->num_devices--;
297 fs_devices->open_devices--;
298 mutex_unlock(&uuid_mutex);
299 close_bdev_excl(bdev);
300 mutex_lock(&uuid_mutex);
307 mutex_unlock(&uuid_mutex);
311 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
313 struct list_head *head = &fs_devices->devices;
314 struct list_head *cur;
315 struct btrfs_device *device;
317 mutex_lock(&uuid_mutex);
318 list_for_each(cur, head) {
319 device = list_entry(cur, struct btrfs_device, dev_list);
321 close_bdev_excl(device->bdev);
322 fs_devices->open_devices--;
325 device->in_fs_metadata = 0;
327 fs_devices->mounted = 0;
328 mutex_unlock(&uuid_mutex);
332 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
333 int flags, void *holder)
335 struct block_device *bdev;
336 struct list_head *head = &fs_devices->devices;
337 struct list_head *cur;
338 struct btrfs_device *device;
339 struct block_device *latest_bdev = NULL;
340 struct buffer_head *bh;
341 struct btrfs_super_block *disk_super;
342 u64 latest_devid = 0;
343 u64 latest_transid = 0;
348 mutex_lock(&uuid_mutex);
349 if (fs_devices->mounted)
352 list_for_each(cur, head) {
353 device = list_entry(cur, struct btrfs_device, dev_list);
360 bdev = open_bdev_excl(device->name, flags, holder);
363 printk("open %s failed\n", device->name);
366 set_blocksize(bdev, 4096);
368 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
372 disk_super = (struct btrfs_super_block *)bh->b_data;
373 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
374 sizeof(disk_super->magic)))
377 devid = le64_to_cpu(disk_super->dev_item.devid);
378 if (devid != device->devid)
381 transid = btrfs_super_generation(disk_super);
382 if (!latest_transid || transid > latest_transid) {
383 latest_devid = devid;
384 latest_transid = transid;
389 device->in_fs_metadata = 0;
390 fs_devices->open_devices++;
396 close_bdev_excl(bdev);
400 if (fs_devices->open_devices == 0) {
404 fs_devices->mounted = 1;
405 fs_devices->latest_bdev = latest_bdev;
406 fs_devices->latest_devid = latest_devid;
407 fs_devices->latest_trans = latest_transid;
409 mutex_unlock(&uuid_mutex);
413 int btrfs_scan_one_device(const char *path, int flags, void *holder,
414 struct btrfs_fs_devices **fs_devices_ret)
416 struct btrfs_super_block *disk_super;
417 struct block_device *bdev;
418 struct buffer_head *bh;
423 mutex_lock(&uuid_mutex);
425 bdev = open_bdev_excl(path, flags, holder);
432 ret = set_blocksize(bdev, 4096);
435 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
440 disk_super = (struct btrfs_super_block *)bh->b_data;
441 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
442 sizeof(disk_super->magic))) {
446 devid = le64_to_cpu(disk_super->dev_item.devid);
447 transid = btrfs_super_generation(disk_super);
448 if (disk_super->label[0])
449 printk("device label %s ", disk_super->label);
451 /* FIXME, make a readl uuid parser */
452 printk("device fsid %llx-%llx ",
453 *(unsigned long long *)disk_super->fsid,
454 *(unsigned long long *)(disk_super->fsid + 8));
456 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
457 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
462 close_bdev_excl(bdev);
464 mutex_unlock(&uuid_mutex);
469 * this uses a pretty simple search, the expectation is that it is
470 * called very infrequently and that a given device has a small number
473 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
474 struct btrfs_device *device,
475 struct btrfs_path *path,
476 u64 num_bytes, u64 *start)
478 struct btrfs_key key;
479 struct btrfs_root *root = device->dev_root;
480 struct btrfs_dev_extent *dev_extent = NULL;
483 u64 search_start = 0;
484 u64 search_end = device->total_bytes;
488 struct extent_buffer *l;
493 /* FIXME use last free of some kind */
495 /* we don't want to overwrite the superblock on the drive,
496 * so we make sure to start at an offset of at least 1MB
498 search_start = max((u64)1024 * 1024, search_start);
500 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
501 search_start = max(root->fs_info->alloc_start, search_start);
503 key.objectid = device->devid;
504 key.offset = search_start;
505 key.type = BTRFS_DEV_EXTENT_KEY;
506 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
509 ret = btrfs_previous_item(root, path, 0, key.type);
513 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
516 slot = path->slots[0];
517 if (slot >= btrfs_header_nritems(l)) {
518 ret = btrfs_next_leaf(root, path);
525 if (search_start >= search_end) {
529 *start = search_start;
533 *start = last_byte > search_start ?
534 last_byte : search_start;
535 if (search_end <= *start) {
541 btrfs_item_key_to_cpu(l, &key, slot);
543 if (key.objectid < device->devid)
546 if (key.objectid > device->devid)
549 if (key.offset >= search_start && key.offset > last_byte &&
551 if (last_byte < search_start)
552 last_byte = search_start;
553 hole_size = key.offset - last_byte;
554 if (key.offset > last_byte &&
555 hole_size >= num_bytes) {
560 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
565 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
566 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
572 /* we have to make sure we didn't find an extent that has already
573 * been allocated by the map tree or the original allocation
575 btrfs_release_path(root, path);
576 BUG_ON(*start < search_start);
578 if (*start + num_bytes > search_end) {
582 /* check for pending inserts here */
586 btrfs_release_path(root, path);
590 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
591 struct btrfs_device *device,
595 struct btrfs_path *path;
596 struct btrfs_root *root = device->dev_root;
597 struct btrfs_key key;
598 struct btrfs_key found_key;
599 struct extent_buffer *leaf = NULL;
600 struct btrfs_dev_extent *extent = NULL;
602 path = btrfs_alloc_path();
606 key.objectid = device->devid;
608 key.type = BTRFS_DEV_EXTENT_KEY;
610 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
612 ret = btrfs_previous_item(root, path, key.objectid,
613 BTRFS_DEV_EXTENT_KEY);
615 leaf = path->nodes[0];
616 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
617 extent = btrfs_item_ptr(leaf, path->slots[0],
618 struct btrfs_dev_extent);
619 BUG_ON(found_key.offset > start || found_key.offset +
620 btrfs_dev_extent_length(leaf, extent) < start);
622 } else if (ret == 0) {
623 leaf = path->nodes[0];
624 extent = btrfs_item_ptr(leaf, path->slots[0],
625 struct btrfs_dev_extent);
629 if (device->bytes_used > 0)
630 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
631 ret = btrfs_del_item(trans, root, path);
634 btrfs_free_path(path);
638 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
639 struct btrfs_device *device,
640 u64 chunk_tree, u64 chunk_objectid,
642 u64 num_bytes, u64 *start)
645 struct btrfs_path *path;
646 struct btrfs_root *root = device->dev_root;
647 struct btrfs_dev_extent *extent;
648 struct extent_buffer *leaf;
649 struct btrfs_key key;
651 WARN_ON(!device->in_fs_metadata);
652 path = btrfs_alloc_path();
656 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
661 key.objectid = device->devid;
663 key.type = BTRFS_DEV_EXTENT_KEY;
664 ret = btrfs_insert_empty_item(trans, root, path, &key,
668 leaf = path->nodes[0];
669 extent = btrfs_item_ptr(leaf, path->slots[0],
670 struct btrfs_dev_extent);
671 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
672 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
673 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
675 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
676 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
679 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
680 btrfs_mark_buffer_dirty(leaf);
682 btrfs_free_path(path);
686 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
688 struct btrfs_path *path;
690 struct btrfs_key key;
691 struct btrfs_chunk *chunk;
692 struct btrfs_key found_key;
694 path = btrfs_alloc_path();
697 key.objectid = objectid;
698 key.offset = (u64)-1;
699 key.type = BTRFS_CHUNK_ITEM_KEY;
701 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
707 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
711 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
713 if (found_key.objectid != objectid)
716 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
718 *offset = found_key.offset +
719 btrfs_chunk_length(path->nodes[0], chunk);
724 btrfs_free_path(path);
728 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
732 struct btrfs_key key;
733 struct btrfs_key found_key;
735 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
736 key.type = BTRFS_DEV_ITEM_KEY;
737 key.offset = (u64)-1;
739 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
745 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
750 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
752 *objectid = found_key.offset + 1;
756 btrfs_release_path(root, path);
761 * the device information is stored in the chunk root
762 * the btrfs_device struct should be fully filled in
764 int btrfs_add_device(struct btrfs_trans_handle *trans,
765 struct btrfs_root *root,
766 struct btrfs_device *device)
769 struct btrfs_path *path;
770 struct btrfs_dev_item *dev_item;
771 struct extent_buffer *leaf;
772 struct btrfs_key key;
776 root = root->fs_info->chunk_root;
778 path = btrfs_alloc_path();
782 ret = find_next_devid(root, path, &free_devid);
786 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
787 key.type = BTRFS_DEV_ITEM_KEY;
788 key.offset = free_devid;
790 ret = btrfs_insert_empty_item(trans, root, path, &key,
795 leaf = path->nodes[0];
796 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
798 device->devid = free_devid;
799 btrfs_set_device_id(leaf, dev_item, device->devid);
800 btrfs_set_device_type(leaf, dev_item, device->type);
801 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
802 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
803 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
804 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
805 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
806 btrfs_set_device_group(leaf, dev_item, 0);
807 btrfs_set_device_seek_speed(leaf, dev_item, 0);
808 btrfs_set_device_bandwidth(leaf, dev_item, 0);
810 ptr = (unsigned long)btrfs_device_uuid(dev_item);
811 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
812 btrfs_mark_buffer_dirty(leaf);
816 btrfs_free_path(path);
820 static int btrfs_rm_dev_item(struct btrfs_root *root,
821 struct btrfs_device *device)
824 struct btrfs_path *path;
825 struct block_device *bdev = device->bdev;
826 struct btrfs_device *next_dev;
827 struct btrfs_key key;
829 struct btrfs_fs_devices *fs_devices;
830 struct btrfs_trans_handle *trans;
832 root = root->fs_info->chunk_root;
834 path = btrfs_alloc_path();
838 trans = btrfs_start_transaction(root, 1);
839 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
840 key.type = BTRFS_DEV_ITEM_KEY;
841 key.offset = device->devid;
844 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
853 ret = btrfs_del_item(trans, root, path);
858 * at this point, the device is zero sized. We want to
859 * remove it from the devices list and zero out the old super
861 list_del_init(&device->dev_list);
862 list_del_init(&device->dev_alloc_list);
863 fs_devices = root->fs_info->fs_devices;
865 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
867 if (bdev == root->fs_info->sb->s_bdev)
868 root->fs_info->sb->s_bdev = next_dev->bdev;
869 if (bdev == fs_devices->latest_bdev)
870 fs_devices->latest_bdev = next_dev->bdev;
872 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
873 btrfs_set_super_num_devices(&root->fs_info->super_copy,
876 btrfs_free_path(path);
878 btrfs_commit_transaction(trans, root);
882 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
884 struct btrfs_device *device;
885 struct block_device *bdev;
886 struct buffer_head *bh = NULL;
887 struct btrfs_super_block *disk_super;
892 mutex_lock(&uuid_mutex);
893 mutex_lock(&root->fs_info->volume_mutex);
895 all_avail = root->fs_info->avail_data_alloc_bits |
896 root->fs_info->avail_system_alloc_bits |
897 root->fs_info->avail_metadata_alloc_bits;
899 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
900 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
901 printk("btrfs: unable to go below four devices on raid10\n");
906 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
907 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
908 printk("btrfs: unable to go below two devices on raid1\n");
913 if (strcmp(device_path, "missing") == 0) {
914 struct list_head *cur;
915 struct list_head *devices;
916 struct btrfs_device *tmp;
919 devices = &root->fs_info->fs_devices->devices;
920 list_for_each(cur, devices) {
921 tmp = list_entry(cur, struct btrfs_device, dev_list);
922 if (tmp->in_fs_metadata && !tmp->bdev) {
931 printk("btrfs: no missing devices found to remove\n");
936 bdev = open_bdev_excl(device_path, 0,
937 root->fs_info->bdev_holder);
943 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
948 disk_super = (struct btrfs_super_block *)bh->b_data;
949 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
950 sizeof(disk_super->magic))) {
954 if (memcmp(disk_super->fsid, root->fs_info->fsid,
959 devid = le64_to_cpu(disk_super->dev_item.devid);
960 device = btrfs_find_device(root, devid, NULL);
967 root->fs_info->fs_devices->num_devices--;
968 root->fs_info->fs_devices->open_devices--;
970 ret = btrfs_shrink_device(device, 0);
975 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
980 /* make sure this device isn't detected as part of
983 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
984 set_buffer_dirty(bh);
985 sync_dirty_buffer(bh);
991 /* one close for the device struct or super_block */
992 close_bdev_excl(device->bdev);
995 /* one close for us */
996 close_bdev_excl(bdev);
1007 close_bdev_excl(bdev);
1009 mutex_unlock(&root->fs_info->volume_mutex);
1010 mutex_unlock(&uuid_mutex);
1014 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1016 struct btrfs_trans_handle *trans;
1017 struct btrfs_device *device;
1018 struct block_device *bdev;
1019 struct list_head *cur;
1020 struct list_head *devices;
1025 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1030 mutex_lock(&root->fs_info->volume_mutex);
1032 trans = btrfs_start_transaction(root, 1);
1034 devices = &root->fs_info->fs_devices->devices;
1035 list_for_each(cur, devices) {
1036 device = list_entry(cur, struct btrfs_device, dev_list);
1037 if (device->bdev == bdev) {
1043 device = kzalloc(sizeof(*device), GFP_NOFS);
1045 /* we can safely leave the fs_devices entry around */
1047 goto out_close_bdev;
1050 device->barriers = 1;
1051 device->work.func = pending_bios_fn;
1052 generate_random_uuid(device->uuid);
1053 spin_lock_init(&device->io_lock);
1054 device->name = kstrdup(device_path, GFP_NOFS);
1055 if (!device->name) {
1057 goto out_close_bdev;
1059 device->io_width = root->sectorsize;
1060 device->io_align = root->sectorsize;
1061 device->sector_size = root->sectorsize;
1062 device->total_bytes = i_size_read(bdev->bd_inode);
1063 device->dev_root = root->fs_info->dev_root;
1064 device->bdev = bdev;
1065 device->in_fs_metadata = 1;
1067 ret = btrfs_add_device(trans, root, device);
1069 goto out_close_bdev;
1071 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1072 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1073 total_bytes + device->total_bytes);
1075 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1076 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1079 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1080 list_add(&device->dev_alloc_list,
1081 &root->fs_info->fs_devices->alloc_list);
1082 root->fs_info->fs_devices->num_devices++;
1083 root->fs_info->fs_devices->open_devices++;
1085 unlock_chunks(root);
1086 btrfs_end_transaction(trans, root);
1087 mutex_unlock(&root->fs_info->volume_mutex);
1092 close_bdev_excl(bdev);
1096 int btrfs_update_device(struct btrfs_trans_handle *trans,
1097 struct btrfs_device *device)
1100 struct btrfs_path *path;
1101 struct btrfs_root *root;
1102 struct btrfs_dev_item *dev_item;
1103 struct extent_buffer *leaf;
1104 struct btrfs_key key;
1106 root = device->dev_root->fs_info->chunk_root;
1108 path = btrfs_alloc_path();
1112 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1113 key.type = BTRFS_DEV_ITEM_KEY;
1114 key.offset = device->devid;
1116 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1125 leaf = path->nodes[0];
1126 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1128 btrfs_set_device_id(leaf, dev_item, device->devid);
1129 btrfs_set_device_type(leaf, dev_item, device->type);
1130 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1131 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1132 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1133 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1134 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1135 btrfs_mark_buffer_dirty(leaf);
1138 btrfs_free_path(path);
1142 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1143 struct btrfs_device *device, u64 new_size)
1145 struct btrfs_super_block *super_copy =
1146 &device->dev_root->fs_info->super_copy;
1147 u64 old_total = btrfs_super_total_bytes(super_copy);
1148 u64 diff = new_size - device->total_bytes;
1150 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1151 return btrfs_update_device(trans, device);
1154 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1155 struct btrfs_device *device, u64 new_size)
1158 lock_chunks(device->dev_root);
1159 ret = __btrfs_grow_device(trans, device, new_size);
1160 unlock_chunks(device->dev_root);
1164 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1165 struct btrfs_root *root,
1166 u64 chunk_tree, u64 chunk_objectid,
1170 struct btrfs_path *path;
1171 struct btrfs_key key;
1173 root = root->fs_info->chunk_root;
1174 path = btrfs_alloc_path();
1178 key.objectid = chunk_objectid;
1179 key.offset = chunk_offset;
1180 key.type = BTRFS_CHUNK_ITEM_KEY;
1182 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1185 ret = btrfs_del_item(trans, root, path);
1188 btrfs_free_path(path);
1192 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1195 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1196 struct btrfs_disk_key *disk_key;
1197 struct btrfs_chunk *chunk;
1204 struct btrfs_key key;
1206 array_size = btrfs_super_sys_array_size(super_copy);
1208 ptr = super_copy->sys_chunk_array;
1211 while (cur < array_size) {
1212 disk_key = (struct btrfs_disk_key *)ptr;
1213 btrfs_disk_key_to_cpu(&key, disk_key);
1215 len = sizeof(*disk_key);
1217 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1218 chunk = (struct btrfs_chunk *)(ptr + len);
1219 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1220 len += btrfs_chunk_item_size(num_stripes);
1225 if (key.objectid == chunk_objectid &&
1226 key.offset == chunk_offset) {
1227 memmove(ptr, ptr + len, array_size - (cur + len));
1229 btrfs_set_super_sys_array_size(super_copy, array_size);
1239 int btrfs_relocate_chunk(struct btrfs_root *root,
1240 u64 chunk_tree, u64 chunk_objectid,
1243 struct extent_map_tree *em_tree;
1244 struct btrfs_root *extent_root;
1245 struct btrfs_trans_handle *trans;
1246 struct extent_map *em;
1247 struct map_lookup *map;
1251 printk("btrfs relocating chunk %llu\n",
1252 (unsigned long long)chunk_offset);
1253 root = root->fs_info->chunk_root;
1254 extent_root = root->fs_info->extent_root;
1255 em_tree = &root->fs_info->mapping_tree.map_tree;
1257 /* step one, relocate all the extents inside this chunk */
1258 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1261 trans = btrfs_start_transaction(root, 1);
1267 * step two, delete the device extents and the
1268 * chunk tree entries
1270 spin_lock(&em_tree->lock);
1271 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1272 spin_unlock(&em_tree->lock);
1274 BUG_ON(em->start > chunk_offset ||
1275 em->start + em->len < chunk_offset);
1276 map = (struct map_lookup *)em->bdev;
1278 for (i = 0; i < map->num_stripes; i++) {
1279 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1280 map->stripes[i].physical);
1283 if (map->stripes[i].dev) {
1284 ret = btrfs_update_device(trans, map->stripes[i].dev);
1288 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1293 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1294 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1298 spin_lock(&em_tree->lock);
1299 remove_extent_mapping(em_tree, em);
1303 /* once for the tree */
1304 free_extent_map(em);
1305 spin_unlock(&em_tree->lock);
1308 free_extent_map(em);
1310 unlock_chunks(root);
1311 btrfs_end_transaction(trans, root);
1315 static u64 div_factor(u64 num, int factor)
1325 int btrfs_balance(struct btrfs_root *dev_root)
1328 struct list_head *cur;
1329 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1330 struct btrfs_device *device;
1333 struct btrfs_path *path;
1334 struct btrfs_key key;
1335 struct btrfs_chunk *chunk;
1336 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1337 struct btrfs_trans_handle *trans;
1338 struct btrfs_key found_key;
1341 mutex_lock(&dev_root->fs_info->volume_mutex);
1342 dev_root = dev_root->fs_info->dev_root;
1344 /* step one make some room on all the devices */
1345 list_for_each(cur, devices) {
1346 device = list_entry(cur, struct btrfs_device, dev_list);
1347 old_size = device->total_bytes;
1348 size_to_free = div_factor(old_size, 1);
1349 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1350 if (device->total_bytes - device->bytes_used > size_to_free)
1353 ret = btrfs_shrink_device(device, old_size - size_to_free);
1356 trans = btrfs_start_transaction(dev_root, 1);
1359 ret = btrfs_grow_device(trans, device, old_size);
1362 btrfs_end_transaction(trans, dev_root);
1365 /* step two, relocate all the chunks */
1366 path = btrfs_alloc_path();
1369 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1370 key.offset = (u64)-1;
1371 key.type = BTRFS_CHUNK_ITEM_KEY;
1374 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1379 * this shouldn't happen, it means the last relocate
1385 ret = btrfs_previous_item(chunk_root, path, 0,
1386 BTRFS_CHUNK_ITEM_KEY);
1390 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1392 if (found_key.objectid != key.objectid)
1395 chunk = btrfs_item_ptr(path->nodes[0],
1397 struct btrfs_chunk);
1398 key.offset = found_key.offset;
1399 /* chunk zero is special */
1400 if (key.offset == 0)
1403 btrfs_release_path(chunk_root, path);
1404 ret = btrfs_relocate_chunk(chunk_root,
1405 chunk_root->root_key.objectid,
1412 btrfs_free_path(path);
1413 mutex_unlock(&dev_root->fs_info->volume_mutex);
1418 * shrinking a device means finding all of the device extents past
1419 * the new size, and then following the back refs to the chunks.
1420 * The chunk relocation code actually frees the device extent
1422 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1424 struct btrfs_trans_handle *trans;
1425 struct btrfs_root *root = device->dev_root;
1426 struct btrfs_dev_extent *dev_extent = NULL;
1427 struct btrfs_path *path;
1434 struct extent_buffer *l;
1435 struct btrfs_key key;
1436 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1437 u64 old_total = btrfs_super_total_bytes(super_copy);
1438 u64 diff = device->total_bytes - new_size;
1441 path = btrfs_alloc_path();
1445 trans = btrfs_start_transaction(root, 1);
1455 device->total_bytes = new_size;
1456 ret = btrfs_update_device(trans, device);
1458 unlock_chunks(root);
1459 btrfs_end_transaction(trans, root);
1462 WARN_ON(diff > old_total);
1463 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1464 unlock_chunks(root);
1465 btrfs_end_transaction(trans, root);
1467 key.objectid = device->devid;
1468 key.offset = (u64)-1;
1469 key.type = BTRFS_DEV_EXTENT_KEY;
1472 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1476 ret = btrfs_previous_item(root, path, 0, key.type);
1485 slot = path->slots[0];
1486 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1488 if (key.objectid != device->devid)
1491 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1492 length = btrfs_dev_extent_length(l, dev_extent);
1494 if (key.offset + length <= new_size)
1497 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1498 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1499 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1500 btrfs_release_path(root, path);
1502 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1509 btrfs_free_path(path);
1513 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1514 struct btrfs_root *root,
1515 struct btrfs_key *key,
1516 struct btrfs_chunk *chunk, int item_size)
1518 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1519 struct btrfs_disk_key disk_key;
1523 array_size = btrfs_super_sys_array_size(super_copy);
1524 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1527 ptr = super_copy->sys_chunk_array + array_size;
1528 btrfs_cpu_key_to_disk(&disk_key, key);
1529 memcpy(ptr, &disk_key, sizeof(disk_key));
1530 ptr += sizeof(disk_key);
1531 memcpy(ptr, chunk, item_size);
1532 item_size += sizeof(disk_key);
1533 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1537 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1540 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1542 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1543 return calc_size * (num_stripes / sub_stripes);
1545 return calc_size * num_stripes;
1549 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1550 struct btrfs_root *extent_root, u64 *start,
1551 u64 *num_bytes, u64 type)
1554 struct btrfs_fs_info *info = extent_root->fs_info;
1555 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1556 struct btrfs_path *path;
1557 struct btrfs_stripe *stripes;
1558 struct btrfs_device *device = NULL;
1559 struct btrfs_chunk *chunk;
1560 struct list_head private_devs;
1561 struct list_head *dev_list;
1562 struct list_head *cur;
1563 struct extent_map_tree *em_tree;
1564 struct map_lookup *map;
1565 struct extent_map *em;
1566 int min_stripe_size = 1 * 1024 * 1024;
1568 u64 calc_size = 1024 * 1024 * 1024;
1569 u64 max_chunk_size = calc_size;
1574 int num_stripes = 1;
1575 int min_stripes = 1;
1576 int sub_stripes = 0;
1580 int stripe_len = 64 * 1024;
1581 struct btrfs_key key;
1583 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1584 (type & BTRFS_BLOCK_GROUP_DUP)) {
1586 type &= ~BTRFS_BLOCK_GROUP_DUP;
1588 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1589 if (list_empty(dev_list))
1592 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1593 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1596 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1600 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1601 num_stripes = min_t(u64, 2,
1602 extent_root->fs_info->fs_devices->open_devices);
1603 if (num_stripes < 2)
1607 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1608 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1609 if (num_stripes < 4)
1611 num_stripes &= ~(u32)1;
1616 if (type & BTRFS_BLOCK_GROUP_DATA) {
1617 max_chunk_size = 10 * calc_size;
1618 min_stripe_size = 64 * 1024 * 1024;
1619 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1620 max_chunk_size = 4 * calc_size;
1621 min_stripe_size = 32 * 1024 * 1024;
1622 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1623 calc_size = 8 * 1024 * 1024;
1624 max_chunk_size = calc_size * 2;
1625 min_stripe_size = 1 * 1024 * 1024;
1628 path = btrfs_alloc_path();
1632 /* we don't want a chunk larger than 10% of the FS */
1633 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1634 max_chunk_size = min(percent_max, max_chunk_size);
1637 if (calc_size * num_stripes > max_chunk_size) {
1638 calc_size = max_chunk_size;
1639 do_div(calc_size, num_stripes);
1640 do_div(calc_size, stripe_len);
1641 calc_size *= stripe_len;
1643 /* we don't want tiny stripes */
1644 calc_size = max_t(u64, min_stripe_size, calc_size);
1646 do_div(calc_size, stripe_len);
1647 calc_size *= stripe_len;
1649 INIT_LIST_HEAD(&private_devs);
1650 cur = dev_list->next;
1653 if (type & BTRFS_BLOCK_GROUP_DUP)
1654 min_free = calc_size * 2;
1656 min_free = calc_size;
1658 /* we add 1MB because we never use the first 1MB of the device */
1659 min_free += 1024 * 1024;
1661 /* build a private list of devices we will allocate from */
1662 while(index < num_stripes) {
1663 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1665 if (device->total_bytes > device->bytes_used)
1666 avail = device->total_bytes - device->bytes_used;
1671 if (device->in_fs_metadata && avail >= min_free) {
1672 u64 ignored_start = 0;
1673 ret = find_free_dev_extent(trans, device, path,
1677 list_move_tail(&device->dev_alloc_list,
1680 if (type & BTRFS_BLOCK_GROUP_DUP)
1683 } else if (device->in_fs_metadata && avail > max_avail)
1685 if (cur == dev_list)
1688 if (index < num_stripes) {
1689 list_splice(&private_devs, dev_list);
1690 if (index >= min_stripes) {
1691 num_stripes = index;
1692 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1693 num_stripes /= sub_stripes;
1694 num_stripes *= sub_stripes;
1699 if (!looped && max_avail > 0) {
1701 calc_size = max_avail;
1704 btrfs_free_path(path);
1707 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1708 key.type = BTRFS_CHUNK_ITEM_KEY;
1709 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1712 btrfs_free_path(path);
1716 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1718 btrfs_free_path(path);
1722 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1725 btrfs_free_path(path);
1728 btrfs_free_path(path);
1731 stripes = &chunk->stripe;
1732 *num_bytes = chunk_bytes_by_type(type, calc_size,
1733 num_stripes, sub_stripes);
1736 while(index < num_stripes) {
1737 struct btrfs_stripe *stripe;
1738 BUG_ON(list_empty(&private_devs));
1739 cur = private_devs.next;
1740 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1742 /* loop over this device again if we're doing a dup group */
1743 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1744 (index == num_stripes - 1))
1745 list_move_tail(&device->dev_alloc_list, dev_list);
1747 ret = btrfs_alloc_dev_extent(trans, device,
1748 info->chunk_root->root_key.objectid,
1749 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1750 calc_size, &dev_offset);
1752 device->bytes_used += calc_size;
1753 ret = btrfs_update_device(trans, device);
1756 map->stripes[index].dev = device;
1757 map->stripes[index].physical = dev_offset;
1758 stripe = stripes + index;
1759 btrfs_set_stack_stripe_devid(stripe, device->devid);
1760 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1761 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1762 physical = dev_offset;
1765 BUG_ON(!list_empty(&private_devs));
1767 /* key was set above */
1768 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1769 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1770 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1771 btrfs_set_stack_chunk_type(chunk, type);
1772 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1773 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1774 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1775 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1776 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1777 map->sector_size = extent_root->sectorsize;
1778 map->stripe_len = stripe_len;
1779 map->io_align = stripe_len;
1780 map->io_width = stripe_len;
1782 map->num_stripes = num_stripes;
1783 map->sub_stripes = sub_stripes;
1785 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1786 btrfs_chunk_item_size(num_stripes));
1788 *start = key.offset;;
1790 em = alloc_extent_map(GFP_NOFS);
1793 em->bdev = (struct block_device *)map;
1794 em->start = key.offset;
1795 em->len = *num_bytes;
1796 em->block_start = 0;
1798 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1799 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1800 chunk, btrfs_chunk_item_size(num_stripes));
1805 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1806 spin_lock(&em_tree->lock);
1807 ret = add_extent_mapping(em_tree, em);
1808 spin_unlock(&em_tree->lock);
1810 free_extent_map(em);
1814 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1816 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1819 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1821 struct extent_map *em;
1824 spin_lock(&tree->map_tree.lock);
1825 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1827 remove_extent_mapping(&tree->map_tree, em);
1828 spin_unlock(&tree->map_tree.lock);
1833 free_extent_map(em);
1834 /* once for the tree */
1835 free_extent_map(em);
1839 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1841 struct extent_map *em;
1842 struct map_lookup *map;
1843 struct extent_map_tree *em_tree = &map_tree->map_tree;
1846 spin_lock(&em_tree->lock);
1847 em = lookup_extent_mapping(em_tree, logical, len);
1848 spin_unlock(&em_tree->lock);
1851 BUG_ON(em->start > logical || em->start + em->len < logical);
1852 map = (struct map_lookup *)em->bdev;
1853 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1854 ret = map->num_stripes;
1855 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1856 ret = map->sub_stripes;
1859 free_extent_map(em);
1863 static int find_live_mirror(struct map_lookup *map, int first, int num,
1867 if (map->stripes[optimal].dev->bdev)
1869 for (i = first; i < first + num; i++) {
1870 if (map->stripes[i].dev->bdev)
1873 /* we couldn't find one that doesn't fail. Just return something
1874 * and the io error handling code will clean up eventually
1879 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1880 u64 logical, u64 *length,
1881 struct btrfs_multi_bio **multi_ret,
1882 int mirror_num, struct page *unplug_page)
1884 struct extent_map *em;
1885 struct map_lookup *map;
1886 struct extent_map_tree *em_tree = &map_tree->map_tree;
1890 int stripes_allocated = 8;
1891 int stripes_required = 1;
1896 struct btrfs_multi_bio *multi = NULL;
1898 if (multi_ret && !(rw & (1 << BIO_RW))) {
1899 stripes_allocated = 1;
1903 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1908 atomic_set(&multi->error, 0);
1911 spin_lock(&em_tree->lock);
1912 em = lookup_extent_mapping(em_tree, logical, *length);
1913 spin_unlock(&em_tree->lock);
1915 if (!em && unplug_page)
1919 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1923 BUG_ON(em->start > logical || em->start + em->len < logical);
1924 map = (struct map_lookup *)em->bdev;
1925 offset = logical - em->start;
1927 if (mirror_num > map->num_stripes)
1930 /* if our multi bio struct is too small, back off and try again */
1931 if (rw & (1 << BIO_RW)) {
1932 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1933 BTRFS_BLOCK_GROUP_DUP)) {
1934 stripes_required = map->num_stripes;
1936 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1937 stripes_required = map->sub_stripes;
1941 if (multi_ret && rw == WRITE &&
1942 stripes_allocated < stripes_required) {
1943 stripes_allocated = map->num_stripes;
1944 free_extent_map(em);
1950 * stripe_nr counts the total number of stripes we have to stride
1951 * to get to this block
1953 do_div(stripe_nr, map->stripe_len);
1955 stripe_offset = stripe_nr * map->stripe_len;
1956 BUG_ON(offset < stripe_offset);
1958 /* stripe_offset is the offset of this block in its stripe*/
1959 stripe_offset = offset - stripe_offset;
1961 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1962 BTRFS_BLOCK_GROUP_RAID10 |
1963 BTRFS_BLOCK_GROUP_DUP)) {
1964 /* we limit the length of each bio to what fits in a stripe */
1965 *length = min_t(u64, em->len - offset,
1966 map->stripe_len - stripe_offset);
1968 *length = em->len - offset;
1971 if (!multi_ret && !unplug_page)
1976 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1977 if (unplug_page || (rw & (1 << BIO_RW)))
1978 num_stripes = map->num_stripes;
1979 else if (mirror_num)
1980 stripe_index = mirror_num - 1;
1982 stripe_index = find_live_mirror(map, 0,
1984 current->pid % map->num_stripes);
1987 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1988 if (rw & (1 << BIO_RW))
1989 num_stripes = map->num_stripes;
1990 else if (mirror_num)
1991 stripe_index = mirror_num - 1;
1993 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1994 int factor = map->num_stripes / map->sub_stripes;
1996 stripe_index = do_div(stripe_nr, factor);
1997 stripe_index *= map->sub_stripes;
1999 if (unplug_page || (rw & (1 << BIO_RW)))
2000 num_stripes = map->sub_stripes;
2001 else if (mirror_num)
2002 stripe_index += mirror_num - 1;
2004 stripe_index = find_live_mirror(map, stripe_index,
2005 map->sub_stripes, stripe_index +
2006 current->pid % map->sub_stripes);
2010 * after this do_div call, stripe_nr is the number of stripes
2011 * on this device we have to walk to find the data, and
2012 * stripe_index is the number of our device in the stripe array
2014 stripe_index = do_div(stripe_nr, map->num_stripes);
2016 BUG_ON(stripe_index >= map->num_stripes);
2018 for (i = 0; i < num_stripes; i++) {
2020 struct btrfs_device *device;
2021 struct backing_dev_info *bdi;
2023 device = map->stripes[stripe_index].dev;
2025 bdi = blk_get_backing_dev_info(device->bdev);
2026 if (bdi->unplug_io_fn) {
2027 bdi->unplug_io_fn(bdi, unplug_page);
2031 multi->stripes[i].physical =
2032 map->stripes[stripe_index].physical +
2033 stripe_offset + stripe_nr * map->stripe_len;
2034 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2040 multi->num_stripes = num_stripes;
2041 multi->max_errors = max_errors;
2044 free_extent_map(em);
2048 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2049 u64 logical, u64 *length,
2050 struct btrfs_multi_bio **multi_ret, int mirror_num)
2052 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2056 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2057 u64 logical, struct page *page)
2059 u64 length = PAGE_CACHE_SIZE;
2060 return __btrfs_map_block(map_tree, READ, logical, &length,
2065 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2066 static void end_bio_multi_stripe(struct bio *bio, int err)
2068 static int end_bio_multi_stripe(struct bio *bio,
2069 unsigned int bytes_done, int err)
2072 struct btrfs_multi_bio *multi = bio->bi_private;
2073 int is_orig_bio = 0;
2075 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2080 atomic_inc(&multi->error);
2082 if (bio == multi->orig_bio)
2085 if (atomic_dec_and_test(&multi->stripes_pending)) {
2088 bio = multi->orig_bio;
2090 bio->bi_private = multi->private;
2091 bio->bi_end_io = multi->end_io;
2092 /* only send an error to the higher layers if it is
2093 * beyond the tolerance of the multi-bio
2095 if (atomic_read(&multi->error) > multi->max_errors) {
2099 * this bio is actually up to date, we didn't
2100 * go over the max number of errors
2102 set_bit(BIO_UPTODATE, &bio->bi_flags);
2107 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2108 bio_endio(bio, bio->bi_size, err);
2110 bio_endio(bio, err);
2112 } else if (!is_orig_bio) {
2115 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2120 struct async_sched {
2123 struct btrfs_fs_info *info;
2124 struct btrfs_work work;
2128 * see run_scheduled_bios for a description of why bios are collected for
2131 * This will add one bio to the pending list for a device and make sure
2132 * the work struct is scheduled.
2134 int schedule_bio(struct btrfs_root *root, struct btrfs_device *device,
2135 int rw, struct bio *bio)
2137 int should_queue = 1;
2139 /* don't bother with additional async steps for reads, right now */
2140 if (!(rw & (1 << BIO_RW))) {
2142 submit_bio(rw, bio);
2148 * nr_async_sumbits allows us to reliably return congestion to the
2149 * higher layers. Otherwise, the async bio makes it appear we have
2150 * made progress against dirty pages when we've really just put it
2151 * on a queue for later
2153 atomic_inc(&root->fs_info->nr_async_submits);
2154 WARN_ON(bio->bi_next);
2155 bio->bi_next = NULL;
2158 spin_lock(&device->io_lock);
2160 if (device->pending_bio_tail)
2161 device->pending_bio_tail->bi_next = bio;
2163 device->pending_bio_tail = bio;
2164 if (!device->pending_bios)
2165 device->pending_bios = bio;
2166 if (device->running_pending)
2169 spin_unlock(&device->io_lock);
2172 btrfs_queue_worker(&root->fs_info->submit_workers,
2177 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2178 int mirror_num, int async_submit)
2180 struct btrfs_mapping_tree *map_tree;
2181 struct btrfs_device *dev;
2182 struct bio *first_bio = bio;
2183 u64 logical = bio->bi_sector << 9;
2186 struct btrfs_multi_bio *multi = NULL;
2191 length = bio->bi_size;
2192 map_tree = &root->fs_info->mapping_tree;
2193 map_length = length;
2195 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2199 total_devs = multi->num_stripes;
2200 if (map_length < length) {
2201 printk("mapping failed logical %Lu bio len %Lu "
2202 "len %Lu\n", logical, length, map_length);
2205 multi->end_io = first_bio->bi_end_io;
2206 multi->private = first_bio->bi_private;
2207 multi->orig_bio = first_bio;
2208 atomic_set(&multi->stripes_pending, multi->num_stripes);
2210 while(dev_nr < total_devs) {
2211 if (total_devs > 1) {
2212 if (dev_nr < total_devs - 1) {
2213 bio = bio_clone(first_bio, GFP_NOFS);
2218 bio->bi_private = multi;
2219 bio->bi_end_io = end_bio_multi_stripe;
2221 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2222 dev = multi->stripes[dev_nr].dev;
2223 if (dev && dev->bdev) {
2224 bio->bi_bdev = dev->bdev;
2226 schedule_bio(root, dev, rw, bio);
2228 submit_bio(rw, bio);
2230 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2231 bio->bi_sector = logical >> 9;
2232 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2233 bio_endio(bio, bio->bi_size, -EIO);
2235 bio_endio(bio, -EIO);
2240 if (total_devs == 1)
2245 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2248 struct list_head *head = &root->fs_info->fs_devices->devices;
2250 return __find_device(head, devid, uuid);
2253 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2254 u64 devid, u8 *dev_uuid)
2256 struct btrfs_device *device;
2257 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2259 device = kzalloc(sizeof(*device), GFP_NOFS);
2260 list_add(&device->dev_list,
2261 &fs_devices->devices);
2262 list_add(&device->dev_alloc_list,
2263 &fs_devices->alloc_list);
2264 device->barriers = 1;
2265 device->dev_root = root->fs_info->dev_root;
2266 device->devid = devid;
2267 device->work.func = pending_bios_fn;
2268 fs_devices->num_devices++;
2269 spin_lock_init(&device->io_lock);
2270 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2275 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2276 struct extent_buffer *leaf,
2277 struct btrfs_chunk *chunk)
2279 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2280 struct map_lookup *map;
2281 struct extent_map *em;
2285 u8 uuid[BTRFS_UUID_SIZE];
2290 logical = key->offset;
2291 length = btrfs_chunk_length(leaf, chunk);
2293 spin_lock(&map_tree->map_tree.lock);
2294 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2295 spin_unlock(&map_tree->map_tree.lock);
2297 /* already mapped? */
2298 if (em && em->start <= logical && em->start + em->len > logical) {
2299 free_extent_map(em);
2302 free_extent_map(em);
2305 map = kzalloc(sizeof(*map), GFP_NOFS);
2309 em = alloc_extent_map(GFP_NOFS);
2312 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2313 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2315 free_extent_map(em);
2319 em->bdev = (struct block_device *)map;
2320 em->start = logical;
2322 em->block_start = 0;
2324 map->num_stripes = num_stripes;
2325 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2326 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2327 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2328 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2329 map->type = btrfs_chunk_type(leaf, chunk);
2330 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2331 for (i = 0; i < num_stripes; i++) {
2332 map->stripes[i].physical =
2333 btrfs_stripe_offset_nr(leaf, chunk, i);
2334 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2335 read_extent_buffer(leaf, uuid, (unsigned long)
2336 btrfs_stripe_dev_uuid_nr(chunk, i),
2338 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2340 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2342 free_extent_map(em);
2345 if (!map->stripes[i].dev) {
2346 map->stripes[i].dev =
2347 add_missing_dev(root, devid, uuid);
2348 if (!map->stripes[i].dev) {
2350 free_extent_map(em);
2354 map->stripes[i].dev->in_fs_metadata = 1;
2357 spin_lock(&map_tree->map_tree.lock);
2358 ret = add_extent_mapping(&map_tree->map_tree, em);
2359 spin_unlock(&map_tree->map_tree.lock);
2361 free_extent_map(em);
2366 static int fill_device_from_item(struct extent_buffer *leaf,
2367 struct btrfs_dev_item *dev_item,
2368 struct btrfs_device *device)
2372 device->devid = btrfs_device_id(leaf, dev_item);
2373 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2374 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2375 device->type = btrfs_device_type(leaf, dev_item);
2376 device->io_align = btrfs_device_io_align(leaf, dev_item);
2377 device->io_width = btrfs_device_io_width(leaf, dev_item);
2378 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2380 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2381 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2386 static int read_one_dev(struct btrfs_root *root,
2387 struct extent_buffer *leaf,
2388 struct btrfs_dev_item *dev_item)
2390 struct btrfs_device *device;
2393 u8 dev_uuid[BTRFS_UUID_SIZE];
2395 devid = btrfs_device_id(leaf, dev_item);
2396 read_extent_buffer(leaf, dev_uuid,
2397 (unsigned long)btrfs_device_uuid(dev_item),
2399 device = btrfs_find_device(root, devid, dev_uuid);
2401 printk("warning devid %Lu missing\n", devid);
2402 device = add_missing_dev(root, devid, dev_uuid);
2407 fill_device_from_item(leaf, dev_item, device);
2408 device->dev_root = root->fs_info->dev_root;
2409 device->in_fs_metadata = 1;
2412 ret = btrfs_open_device(device);
2420 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2422 struct btrfs_dev_item *dev_item;
2424 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2426 return read_one_dev(root, buf, dev_item);
2429 int btrfs_read_sys_array(struct btrfs_root *root)
2431 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2432 struct extent_buffer *sb;
2433 struct btrfs_disk_key *disk_key;
2434 struct btrfs_chunk *chunk;
2436 unsigned long sb_ptr;
2442 struct btrfs_key key;
2444 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2445 BTRFS_SUPER_INFO_SIZE);
2448 btrfs_set_buffer_uptodate(sb);
2449 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2450 array_size = btrfs_super_sys_array_size(super_copy);
2452 ptr = super_copy->sys_chunk_array;
2453 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2456 while (cur < array_size) {
2457 disk_key = (struct btrfs_disk_key *)ptr;
2458 btrfs_disk_key_to_cpu(&key, disk_key);
2460 len = sizeof(*disk_key); ptr += len;
2464 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2465 chunk = (struct btrfs_chunk *)sb_ptr;
2466 ret = read_one_chunk(root, &key, sb, chunk);
2469 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2470 len = btrfs_chunk_item_size(num_stripes);
2479 free_extent_buffer(sb);
2483 int btrfs_read_chunk_tree(struct btrfs_root *root)
2485 struct btrfs_path *path;
2486 struct extent_buffer *leaf;
2487 struct btrfs_key key;
2488 struct btrfs_key found_key;
2492 root = root->fs_info->chunk_root;
2494 path = btrfs_alloc_path();
2498 /* first we search for all of the device items, and then we
2499 * read in all of the chunk items. This way we can create chunk
2500 * mappings that reference all of the devices that are afound
2502 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2508 leaf = path->nodes[0];
2509 slot = path->slots[0];
2510 if (slot >= btrfs_header_nritems(leaf)) {
2511 ret = btrfs_next_leaf(root, path);
2518 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2519 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2520 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2522 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2523 struct btrfs_dev_item *dev_item;
2524 dev_item = btrfs_item_ptr(leaf, slot,
2525 struct btrfs_dev_item);
2526 ret = read_one_dev(root, leaf, dev_item);
2529 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2530 struct btrfs_chunk *chunk;
2531 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2532 ret = read_one_chunk(root, &found_key, leaf, chunk);
2536 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2538 btrfs_release_path(root, path);
2542 btrfs_free_path(path);