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);
183 submit_bio(cur->bi_rw, cur);
187 * we made progress, there is more work to do and the bdi
188 * is now congested. Back off and let other work structs
191 if (pending && num_run && bdi_write_congested(bdi)) {
192 struct bio *old_head;
194 spin_lock(&device->io_lock);
195 old_head = device->pending_bios;
196 device->pending_bios = pending;
197 if (device->pending_bio_tail)
198 tail->bi_next = old_head;
200 device->pending_bio_tail = tail;
202 spin_unlock(&device->io_lock);
203 btrfs_requeue_work(&device->work);
213 void pending_bios_fn(struct btrfs_work *work)
215 struct btrfs_device *device;
217 device = container_of(work, struct btrfs_device, work);
218 run_scheduled_bios(device);
221 static int device_list_add(const char *path,
222 struct btrfs_super_block *disk_super,
223 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
225 struct btrfs_device *device;
226 struct btrfs_fs_devices *fs_devices;
227 u64 found_transid = btrfs_super_generation(disk_super);
229 fs_devices = find_fsid(disk_super->fsid);
231 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
234 INIT_LIST_HEAD(&fs_devices->devices);
235 INIT_LIST_HEAD(&fs_devices->alloc_list);
236 list_add(&fs_devices->list, &fs_uuids);
237 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
238 fs_devices->latest_devid = devid;
239 fs_devices->latest_trans = found_transid;
242 device = __find_device(&fs_devices->devices, devid,
243 disk_super->dev_item.uuid);
246 device = kzalloc(sizeof(*device), GFP_NOFS);
248 /* we can safely leave the fs_devices entry around */
251 device->devid = devid;
252 device->work.func = pending_bios_fn;
253 memcpy(device->uuid, disk_super->dev_item.uuid,
255 device->barriers = 1;
256 spin_lock_init(&device->io_lock);
257 device->name = kstrdup(path, GFP_NOFS);
262 list_add(&device->dev_list, &fs_devices->devices);
263 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
264 fs_devices->num_devices++;
267 if (found_transid > fs_devices->latest_trans) {
268 fs_devices->latest_devid = devid;
269 fs_devices->latest_trans = found_transid;
271 *fs_devices_ret = fs_devices;
275 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
277 struct list_head *head = &fs_devices->devices;
278 struct list_head *cur;
279 struct btrfs_device *device;
281 mutex_lock(&uuid_mutex);
283 list_for_each(cur, head) {
284 device = list_entry(cur, struct btrfs_device, dev_list);
285 if (!device->in_fs_metadata) {
286 struct block_device *bdev;
287 list_del(&device->dev_list);
288 list_del(&device->dev_alloc_list);
289 fs_devices->num_devices--;
292 fs_devices->open_devices--;
293 mutex_unlock(&uuid_mutex);
294 close_bdev_excl(bdev);
295 mutex_lock(&uuid_mutex);
302 mutex_unlock(&uuid_mutex);
306 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
308 struct list_head *head = &fs_devices->devices;
309 struct list_head *cur;
310 struct btrfs_device *device;
312 mutex_lock(&uuid_mutex);
313 list_for_each(cur, head) {
314 device = list_entry(cur, struct btrfs_device, dev_list);
316 close_bdev_excl(device->bdev);
317 fs_devices->open_devices--;
320 device->in_fs_metadata = 0;
322 fs_devices->mounted = 0;
323 mutex_unlock(&uuid_mutex);
327 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
328 int flags, void *holder)
330 struct block_device *bdev;
331 struct list_head *head = &fs_devices->devices;
332 struct list_head *cur;
333 struct btrfs_device *device;
334 struct block_device *latest_bdev = NULL;
335 struct buffer_head *bh;
336 struct btrfs_super_block *disk_super;
337 u64 latest_devid = 0;
338 u64 latest_transid = 0;
343 mutex_lock(&uuid_mutex);
344 if (fs_devices->mounted)
347 list_for_each(cur, head) {
348 device = list_entry(cur, struct btrfs_device, dev_list);
355 bdev = open_bdev_excl(device->name, flags, holder);
358 printk("open %s failed\n", device->name);
361 set_blocksize(bdev, 4096);
363 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
367 disk_super = (struct btrfs_super_block *)bh->b_data;
368 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
369 sizeof(disk_super->magic)))
372 devid = le64_to_cpu(disk_super->dev_item.devid);
373 if (devid != device->devid)
376 transid = btrfs_super_generation(disk_super);
377 if (!latest_transid || transid > latest_transid) {
378 latest_devid = devid;
379 latest_transid = transid;
384 device->in_fs_metadata = 0;
385 fs_devices->open_devices++;
391 close_bdev_excl(bdev);
395 if (fs_devices->open_devices == 0) {
399 fs_devices->mounted = 1;
400 fs_devices->latest_bdev = latest_bdev;
401 fs_devices->latest_devid = latest_devid;
402 fs_devices->latest_trans = latest_transid;
404 mutex_unlock(&uuid_mutex);
408 int btrfs_scan_one_device(const char *path, int flags, void *holder,
409 struct btrfs_fs_devices **fs_devices_ret)
411 struct btrfs_super_block *disk_super;
412 struct block_device *bdev;
413 struct buffer_head *bh;
418 mutex_lock(&uuid_mutex);
420 bdev = open_bdev_excl(path, flags, holder);
427 ret = set_blocksize(bdev, 4096);
430 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
435 disk_super = (struct btrfs_super_block *)bh->b_data;
436 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
437 sizeof(disk_super->magic))) {
441 devid = le64_to_cpu(disk_super->dev_item.devid);
442 transid = btrfs_super_generation(disk_super);
443 if (disk_super->label[0])
444 printk("device label %s ", disk_super->label);
446 /* FIXME, make a readl uuid parser */
447 printk("device fsid %llx-%llx ",
448 *(unsigned long long *)disk_super->fsid,
449 *(unsigned long long *)(disk_super->fsid + 8));
451 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
452 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
457 close_bdev_excl(bdev);
459 mutex_unlock(&uuid_mutex);
464 * this uses a pretty simple search, the expectation is that it is
465 * called very infrequently and that a given device has a small number
468 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
469 struct btrfs_device *device,
470 struct btrfs_path *path,
471 u64 num_bytes, u64 *start)
473 struct btrfs_key key;
474 struct btrfs_root *root = device->dev_root;
475 struct btrfs_dev_extent *dev_extent = NULL;
478 u64 search_start = 0;
479 u64 search_end = device->total_bytes;
483 struct extent_buffer *l;
488 /* FIXME use last free of some kind */
490 /* we don't want to overwrite the superblock on the drive,
491 * so we make sure to start at an offset of at least 1MB
493 search_start = max((u64)1024 * 1024, search_start);
495 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
496 search_start = max(root->fs_info->alloc_start, search_start);
498 key.objectid = device->devid;
499 key.offset = search_start;
500 key.type = BTRFS_DEV_EXTENT_KEY;
501 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
504 ret = btrfs_previous_item(root, path, 0, key.type);
508 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
511 slot = path->slots[0];
512 if (slot >= btrfs_header_nritems(l)) {
513 ret = btrfs_next_leaf(root, path);
520 if (search_start >= search_end) {
524 *start = search_start;
528 *start = last_byte > search_start ?
529 last_byte : search_start;
530 if (search_end <= *start) {
536 btrfs_item_key_to_cpu(l, &key, slot);
538 if (key.objectid < device->devid)
541 if (key.objectid > device->devid)
544 if (key.offset >= search_start && key.offset > last_byte &&
546 if (last_byte < search_start)
547 last_byte = search_start;
548 hole_size = key.offset - last_byte;
549 if (key.offset > last_byte &&
550 hole_size >= num_bytes) {
555 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
560 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
561 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
567 /* we have to make sure we didn't find an extent that has already
568 * been allocated by the map tree or the original allocation
570 btrfs_release_path(root, path);
571 BUG_ON(*start < search_start);
573 if (*start + num_bytes > search_end) {
577 /* check for pending inserts here */
581 btrfs_release_path(root, path);
585 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
586 struct btrfs_device *device,
590 struct btrfs_path *path;
591 struct btrfs_root *root = device->dev_root;
592 struct btrfs_key key;
593 struct btrfs_key found_key;
594 struct extent_buffer *leaf = NULL;
595 struct btrfs_dev_extent *extent = NULL;
597 path = btrfs_alloc_path();
601 key.objectid = device->devid;
603 key.type = BTRFS_DEV_EXTENT_KEY;
605 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
607 ret = btrfs_previous_item(root, path, key.objectid,
608 BTRFS_DEV_EXTENT_KEY);
610 leaf = path->nodes[0];
611 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
612 extent = btrfs_item_ptr(leaf, path->slots[0],
613 struct btrfs_dev_extent);
614 BUG_ON(found_key.offset > start || found_key.offset +
615 btrfs_dev_extent_length(leaf, extent) < start);
617 } else if (ret == 0) {
618 leaf = path->nodes[0];
619 extent = btrfs_item_ptr(leaf, path->slots[0],
620 struct btrfs_dev_extent);
624 if (device->bytes_used > 0)
625 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
626 ret = btrfs_del_item(trans, root, path);
629 btrfs_free_path(path);
633 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
634 struct btrfs_device *device,
635 u64 chunk_tree, u64 chunk_objectid,
637 u64 num_bytes, u64 *start)
640 struct btrfs_path *path;
641 struct btrfs_root *root = device->dev_root;
642 struct btrfs_dev_extent *extent;
643 struct extent_buffer *leaf;
644 struct btrfs_key key;
646 WARN_ON(!device->in_fs_metadata);
647 path = btrfs_alloc_path();
651 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
656 key.objectid = device->devid;
658 key.type = BTRFS_DEV_EXTENT_KEY;
659 ret = btrfs_insert_empty_item(trans, root, path, &key,
663 leaf = path->nodes[0];
664 extent = btrfs_item_ptr(leaf, path->slots[0],
665 struct btrfs_dev_extent);
666 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
667 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
668 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
670 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
671 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
674 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
675 btrfs_mark_buffer_dirty(leaf);
677 btrfs_free_path(path);
681 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
683 struct btrfs_path *path;
685 struct btrfs_key key;
686 struct btrfs_chunk *chunk;
687 struct btrfs_key found_key;
689 path = btrfs_alloc_path();
692 key.objectid = objectid;
693 key.offset = (u64)-1;
694 key.type = BTRFS_CHUNK_ITEM_KEY;
696 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
702 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
706 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
708 if (found_key.objectid != objectid)
711 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
713 *offset = found_key.offset +
714 btrfs_chunk_length(path->nodes[0], chunk);
719 btrfs_free_path(path);
723 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
727 struct btrfs_key key;
728 struct btrfs_key found_key;
730 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
731 key.type = BTRFS_DEV_ITEM_KEY;
732 key.offset = (u64)-1;
734 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
740 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
745 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
747 *objectid = found_key.offset + 1;
751 btrfs_release_path(root, path);
756 * the device information is stored in the chunk root
757 * the btrfs_device struct should be fully filled in
759 int btrfs_add_device(struct btrfs_trans_handle *trans,
760 struct btrfs_root *root,
761 struct btrfs_device *device)
764 struct btrfs_path *path;
765 struct btrfs_dev_item *dev_item;
766 struct extent_buffer *leaf;
767 struct btrfs_key key;
771 root = root->fs_info->chunk_root;
773 path = btrfs_alloc_path();
777 ret = find_next_devid(root, path, &free_devid);
781 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
782 key.type = BTRFS_DEV_ITEM_KEY;
783 key.offset = free_devid;
785 ret = btrfs_insert_empty_item(trans, root, path, &key,
790 leaf = path->nodes[0];
791 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
793 device->devid = free_devid;
794 btrfs_set_device_id(leaf, dev_item, device->devid);
795 btrfs_set_device_type(leaf, dev_item, device->type);
796 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
797 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
798 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
799 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
800 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
801 btrfs_set_device_group(leaf, dev_item, 0);
802 btrfs_set_device_seek_speed(leaf, dev_item, 0);
803 btrfs_set_device_bandwidth(leaf, dev_item, 0);
805 ptr = (unsigned long)btrfs_device_uuid(dev_item);
806 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
807 btrfs_mark_buffer_dirty(leaf);
811 btrfs_free_path(path);
815 static int btrfs_rm_dev_item(struct btrfs_root *root,
816 struct btrfs_device *device)
819 struct btrfs_path *path;
820 struct block_device *bdev = device->bdev;
821 struct btrfs_device *next_dev;
822 struct btrfs_key key;
824 struct btrfs_fs_devices *fs_devices;
825 struct btrfs_trans_handle *trans;
827 root = root->fs_info->chunk_root;
829 path = btrfs_alloc_path();
833 trans = btrfs_start_transaction(root, 1);
834 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
835 key.type = BTRFS_DEV_ITEM_KEY;
836 key.offset = device->devid;
839 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
848 ret = btrfs_del_item(trans, root, path);
853 * at this point, the device is zero sized. We want to
854 * remove it from the devices list and zero out the old super
856 list_del_init(&device->dev_list);
857 list_del_init(&device->dev_alloc_list);
858 fs_devices = root->fs_info->fs_devices;
860 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
862 if (bdev == root->fs_info->sb->s_bdev)
863 root->fs_info->sb->s_bdev = next_dev->bdev;
864 if (bdev == fs_devices->latest_bdev)
865 fs_devices->latest_bdev = next_dev->bdev;
867 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
868 btrfs_set_super_num_devices(&root->fs_info->super_copy,
871 btrfs_free_path(path);
873 btrfs_commit_transaction(trans, root);
877 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
879 struct btrfs_device *device;
880 struct block_device *bdev;
881 struct buffer_head *bh = NULL;
882 struct btrfs_super_block *disk_super;
887 mutex_lock(&uuid_mutex);
888 mutex_lock(&root->fs_info->volume_mutex);
890 all_avail = root->fs_info->avail_data_alloc_bits |
891 root->fs_info->avail_system_alloc_bits |
892 root->fs_info->avail_metadata_alloc_bits;
894 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
895 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
896 printk("btrfs: unable to go below four devices on raid10\n");
901 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
902 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
903 printk("btrfs: unable to go below two devices on raid1\n");
908 if (strcmp(device_path, "missing") == 0) {
909 struct list_head *cur;
910 struct list_head *devices;
911 struct btrfs_device *tmp;
914 devices = &root->fs_info->fs_devices->devices;
915 list_for_each(cur, devices) {
916 tmp = list_entry(cur, struct btrfs_device, dev_list);
917 if (tmp->in_fs_metadata && !tmp->bdev) {
926 printk("btrfs: no missing devices found to remove\n");
931 bdev = open_bdev_excl(device_path, 0,
932 root->fs_info->bdev_holder);
938 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
943 disk_super = (struct btrfs_super_block *)bh->b_data;
944 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
945 sizeof(disk_super->magic))) {
949 if (memcmp(disk_super->fsid, root->fs_info->fsid,
954 devid = le64_to_cpu(disk_super->dev_item.devid);
955 device = btrfs_find_device(root, devid, NULL);
962 root->fs_info->fs_devices->num_devices--;
963 root->fs_info->fs_devices->open_devices--;
965 ret = btrfs_shrink_device(device, 0);
970 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
975 /* make sure this device isn't detected as part of
978 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
979 set_buffer_dirty(bh);
980 sync_dirty_buffer(bh);
986 /* one close for the device struct or super_block */
987 close_bdev_excl(device->bdev);
990 /* one close for us */
991 close_bdev_excl(bdev);
1002 close_bdev_excl(bdev);
1004 mutex_unlock(&root->fs_info->volume_mutex);
1005 mutex_unlock(&uuid_mutex);
1009 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1011 struct btrfs_trans_handle *trans;
1012 struct btrfs_device *device;
1013 struct block_device *bdev;
1014 struct list_head *cur;
1015 struct list_head *devices;
1020 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1025 mutex_lock(&root->fs_info->volume_mutex);
1027 trans = btrfs_start_transaction(root, 1);
1029 devices = &root->fs_info->fs_devices->devices;
1030 list_for_each(cur, devices) {
1031 device = list_entry(cur, struct btrfs_device, dev_list);
1032 if (device->bdev == bdev) {
1038 device = kzalloc(sizeof(*device), GFP_NOFS);
1040 /* we can safely leave the fs_devices entry around */
1042 goto out_close_bdev;
1045 device->barriers = 1;
1046 device->work.func = pending_bios_fn;
1047 generate_random_uuid(device->uuid);
1048 spin_lock_init(&device->io_lock);
1049 device->name = kstrdup(device_path, GFP_NOFS);
1050 if (!device->name) {
1052 goto out_close_bdev;
1054 device->io_width = root->sectorsize;
1055 device->io_align = root->sectorsize;
1056 device->sector_size = root->sectorsize;
1057 device->total_bytes = i_size_read(bdev->bd_inode);
1058 device->dev_root = root->fs_info->dev_root;
1059 device->bdev = bdev;
1060 device->in_fs_metadata = 1;
1062 ret = btrfs_add_device(trans, root, device);
1064 goto out_close_bdev;
1066 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1067 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1068 total_bytes + device->total_bytes);
1070 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1071 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1074 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1075 list_add(&device->dev_alloc_list,
1076 &root->fs_info->fs_devices->alloc_list);
1077 root->fs_info->fs_devices->num_devices++;
1078 root->fs_info->fs_devices->open_devices++;
1080 unlock_chunks(root);
1081 btrfs_end_transaction(trans, root);
1082 mutex_unlock(&root->fs_info->volume_mutex);
1087 close_bdev_excl(bdev);
1091 int btrfs_update_device(struct btrfs_trans_handle *trans,
1092 struct btrfs_device *device)
1095 struct btrfs_path *path;
1096 struct btrfs_root *root;
1097 struct btrfs_dev_item *dev_item;
1098 struct extent_buffer *leaf;
1099 struct btrfs_key key;
1101 root = device->dev_root->fs_info->chunk_root;
1103 path = btrfs_alloc_path();
1107 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1108 key.type = BTRFS_DEV_ITEM_KEY;
1109 key.offset = device->devid;
1111 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1120 leaf = path->nodes[0];
1121 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1123 btrfs_set_device_id(leaf, dev_item, device->devid);
1124 btrfs_set_device_type(leaf, dev_item, device->type);
1125 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1126 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1127 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1128 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1129 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1130 btrfs_mark_buffer_dirty(leaf);
1133 btrfs_free_path(path);
1137 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1138 struct btrfs_device *device, u64 new_size)
1140 struct btrfs_super_block *super_copy =
1141 &device->dev_root->fs_info->super_copy;
1142 u64 old_total = btrfs_super_total_bytes(super_copy);
1143 u64 diff = new_size - device->total_bytes;
1145 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1146 return btrfs_update_device(trans, device);
1149 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1150 struct btrfs_device *device, u64 new_size)
1153 lock_chunks(device->dev_root);
1154 ret = __btrfs_grow_device(trans, device, new_size);
1155 unlock_chunks(device->dev_root);
1159 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 u64 chunk_tree, u64 chunk_objectid,
1165 struct btrfs_path *path;
1166 struct btrfs_key key;
1168 root = root->fs_info->chunk_root;
1169 path = btrfs_alloc_path();
1173 key.objectid = chunk_objectid;
1174 key.offset = chunk_offset;
1175 key.type = BTRFS_CHUNK_ITEM_KEY;
1177 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1180 ret = btrfs_del_item(trans, root, path);
1183 btrfs_free_path(path);
1187 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1190 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1191 struct btrfs_disk_key *disk_key;
1192 struct btrfs_chunk *chunk;
1199 struct btrfs_key key;
1201 array_size = btrfs_super_sys_array_size(super_copy);
1203 ptr = super_copy->sys_chunk_array;
1206 while (cur < array_size) {
1207 disk_key = (struct btrfs_disk_key *)ptr;
1208 btrfs_disk_key_to_cpu(&key, disk_key);
1210 len = sizeof(*disk_key);
1212 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1213 chunk = (struct btrfs_chunk *)(ptr + len);
1214 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1215 len += btrfs_chunk_item_size(num_stripes);
1220 if (key.objectid == chunk_objectid &&
1221 key.offset == chunk_offset) {
1222 memmove(ptr, ptr + len, array_size - (cur + len));
1224 btrfs_set_super_sys_array_size(super_copy, array_size);
1234 int btrfs_relocate_chunk(struct btrfs_root *root,
1235 u64 chunk_tree, u64 chunk_objectid,
1238 struct extent_map_tree *em_tree;
1239 struct btrfs_root *extent_root;
1240 struct btrfs_trans_handle *trans;
1241 struct extent_map *em;
1242 struct map_lookup *map;
1246 printk("btrfs relocating chunk %llu\n",
1247 (unsigned long long)chunk_offset);
1248 root = root->fs_info->chunk_root;
1249 extent_root = root->fs_info->extent_root;
1250 em_tree = &root->fs_info->mapping_tree.map_tree;
1252 /* step one, relocate all the extents inside this chunk */
1253 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1256 trans = btrfs_start_transaction(root, 1);
1262 * step two, delete the device extents and the
1263 * chunk tree entries
1265 spin_lock(&em_tree->lock);
1266 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1267 spin_unlock(&em_tree->lock);
1269 BUG_ON(em->start > chunk_offset ||
1270 em->start + em->len < chunk_offset);
1271 map = (struct map_lookup *)em->bdev;
1273 for (i = 0; i < map->num_stripes; i++) {
1274 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1275 map->stripes[i].physical);
1278 if (map->stripes[i].dev) {
1279 ret = btrfs_update_device(trans, map->stripes[i].dev);
1283 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1288 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1289 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1293 spin_lock(&em_tree->lock);
1294 remove_extent_mapping(em_tree, em);
1298 /* once for the tree */
1299 free_extent_map(em);
1300 spin_unlock(&em_tree->lock);
1303 free_extent_map(em);
1305 unlock_chunks(root);
1306 btrfs_end_transaction(trans, root);
1310 static u64 div_factor(u64 num, int factor)
1320 int btrfs_balance(struct btrfs_root *dev_root)
1323 struct list_head *cur;
1324 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1325 struct btrfs_device *device;
1328 struct btrfs_path *path;
1329 struct btrfs_key key;
1330 struct btrfs_chunk *chunk;
1331 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1332 struct btrfs_trans_handle *trans;
1333 struct btrfs_key found_key;
1336 mutex_lock(&dev_root->fs_info->volume_mutex);
1337 dev_root = dev_root->fs_info->dev_root;
1339 /* step one make some room on all the devices */
1340 list_for_each(cur, devices) {
1341 device = list_entry(cur, struct btrfs_device, dev_list);
1342 old_size = device->total_bytes;
1343 size_to_free = div_factor(old_size, 1);
1344 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1345 if (device->total_bytes - device->bytes_used > size_to_free)
1348 ret = btrfs_shrink_device(device, old_size - size_to_free);
1351 trans = btrfs_start_transaction(dev_root, 1);
1354 ret = btrfs_grow_device(trans, device, old_size);
1357 btrfs_end_transaction(trans, dev_root);
1360 /* step two, relocate all the chunks */
1361 path = btrfs_alloc_path();
1364 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1365 key.offset = (u64)-1;
1366 key.type = BTRFS_CHUNK_ITEM_KEY;
1369 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1374 * this shouldn't happen, it means the last relocate
1380 ret = btrfs_previous_item(chunk_root, path, 0,
1381 BTRFS_CHUNK_ITEM_KEY);
1385 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1387 if (found_key.objectid != key.objectid)
1390 chunk = btrfs_item_ptr(path->nodes[0],
1392 struct btrfs_chunk);
1393 key.offset = found_key.offset;
1394 /* chunk zero is special */
1395 if (key.offset == 0)
1398 btrfs_release_path(chunk_root, path);
1399 ret = btrfs_relocate_chunk(chunk_root,
1400 chunk_root->root_key.objectid,
1407 btrfs_free_path(path);
1408 mutex_unlock(&dev_root->fs_info->volume_mutex);
1413 * shrinking a device means finding all of the device extents past
1414 * the new size, and then following the back refs to the chunks.
1415 * The chunk relocation code actually frees the device extent
1417 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1419 struct btrfs_trans_handle *trans;
1420 struct btrfs_root *root = device->dev_root;
1421 struct btrfs_dev_extent *dev_extent = NULL;
1422 struct btrfs_path *path;
1429 struct extent_buffer *l;
1430 struct btrfs_key key;
1431 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1432 u64 old_total = btrfs_super_total_bytes(super_copy);
1433 u64 diff = device->total_bytes - new_size;
1436 path = btrfs_alloc_path();
1440 trans = btrfs_start_transaction(root, 1);
1450 device->total_bytes = new_size;
1451 ret = btrfs_update_device(trans, device);
1453 unlock_chunks(root);
1454 btrfs_end_transaction(trans, root);
1457 WARN_ON(diff > old_total);
1458 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1459 unlock_chunks(root);
1460 btrfs_end_transaction(trans, root);
1462 key.objectid = device->devid;
1463 key.offset = (u64)-1;
1464 key.type = BTRFS_DEV_EXTENT_KEY;
1467 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1471 ret = btrfs_previous_item(root, path, 0, key.type);
1480 slot = path->slots[0];
1481 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1483 if (key.objectid != device->devid)
1486 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1487 length = btrfs_dev_extent_length(l, dev_extent);
1489 if (key.offset + length <= new_size)
1492 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1493 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1494 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1495 btrfs_release_path(root, path);
1497 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1504 btrfs_free_path(path);
1508 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1509 struct btrfs_root *root,
1510 struct btrfs_key *key,
1511 struct btrfs_chunk *chunk, int item_size)
1513 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1514 struct btrfs_disk_key disk_key;
1518 array_size = btrfs_super_sys_array_size(super_copy);
1519 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1522 ptr = super_copy->sys_chunk_array + array_size;
1523 btrfs_cpu_key_to_disk(&disk_key, key);
1524 memcpy(ptr, &disk_key, sizeof(disk_key));
1525 ptr += sizeof(disk_key);
1526 memcpy(ptr, chunk, item_size);
1527 item_size += sizeof(disk_key);
1528 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1532 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1535 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1537 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1538 return calc_size * (num_stripes / sub_stripes);
1540 return calc_size * num_stripes;
1544 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1545 struct btrfs_root *extent_root, u64 *start,
1546 u64 *num_bytes, u64 type)
1549 struct btrfs_fs_info *info = extent_root->fs_info;
1550 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1551 struct btrfs_path *path;
1552 struct btrfs_stripe *stripes;
1553 struct btrfs_device *device = NULL;
1554 struct btrfs_chunk *chunk;
1555 struct list_head private_devs;
1556 struct list_head *dev_list;
1557 struct list_head *cur;
1558 struct extent_map_tree *em_tree;
1559 struct map_lookup *map;
1560 struct extent_map *em;
1561 int min_stripe_size = 1 * 1024 * 1024;
1563 u64 calc_size = 1024 * 1024 * 1024;
1564 u64 max_chunk_size = calc_size;
1569 int num_stripes = 1;
1570 int min_stripes = 1;
1571 int sub_stripes = 0;
1575 int stripe_len = 64 * 1024;
1576 struct btrfs_key key;
1578 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1579 (type & BTRFS_BLOCK_GROUP_DUP)) {
1581 type &= ~BTRFS_BLOCK_GROUP_DUP;
1583 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1584 if (list_empty(dev_list))
1587 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1588 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1591 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1595 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1596 num_stripes = min_t(u64, 2,
1597 extent_root->fs_info->fs_devices->open_devices);
1598 if (num_stripes < 2)
1602 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1603 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1604 if (num_stripes < 4)
1606 num_stripes &= ~(u32)1;
1611 if (type & BTRFS_BLOCK_GROUP_DATA) {
1612 max_chunk_size = 10 * calc_size;
1613 min_stripe_size = 64 * 1024 * 1024;
1614 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1615 max_chunk_size = 4 * calc_size;
1616 min_stripe_size = 32 * 1024 * 1024;
1617 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1618 calc_size = 8 * 1024 * 1024;
1619 max_chunk_size = calc_size * 2;
1620 min_stripe_size = 1 * 1024 * 1024;
1623 path = btrfs_alloc_path();
1627 /* we don't want a chunk larger than 10% of the FS */
1628 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1629 max_chunk_size = min(percent_max, max_chunk_size);
1632 if (calc_size * num_stripes > max_chunk_size) {
1633 calc_size = max_chunk_size;
1634 do_div(calc_size, num_stripes);
1635 do_div(calc_size, stripe_len);
1636 calc_size *= stripe_len;
1638 /* we don't want tiny stripes */
1639 calc_size = max_t(u64, min_stripe_size, calc_size);
1641 do_div(calc_size, stripe_len);
1642 calc_size *= stripe_len;
1644 INIT_LIST_HEAD(&private_devs);
1645 cur = dev_list->next;
1648 if (type & BTRFS_BLOCK_GROUP_DUP)
1649 min_free = calc_size * 2;
1651 min_free = calc_size;
1653 /* we add 1MB because we never use the first 1MB of the device */
1654 min_free += 1024 * 1024;
1656 /* build a private list of devices we will allocate from */
1657 while(index < num_stripes) {
1658 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1660 if (device->total_bytes > device->bytes_used)
1661 avail = device->total_bytes - device->bytes_used;
1666 if (device->in_fs_metadata && avail >= min_free) {
1667 u64 ignored_start = 0;
1668 ret = find_free_dev_extent(trans, device, path,
1672 list_move_tail(&device->dev_alloc_list,
1675 if (type & BTRFS_BLOCK_GROUP_DUP)
1678 } else if (device->in_fs_metadata && avail > max_avail)
1680 if (cur == dev_list)
1683 if (index < num_stripes) {
1684 list_splice(&private_devs, dev_list);
1685 if (index >= min_stripes) {
1686 num_stripes = index;
1687 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1688 num_stripes /= sub_stripes;
1689 num_stripes *= sub_stripes;
1694 if (!looped && max_avail > 0) {
1696 calc_size = max_avail;
1699 btrfs_free_path(path);
1702 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1703 key.type = BTRFS_CHUNK_ITEM_KEY;
1704 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1707 btrfs_free_path(path);
1711 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1713 btrfs_free_path(path);
1717 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1720 btrfs_free_path(path);
1723 btrfs_free_path(path);
1726 stripes = &chunk->stripe;
1727 *num_bytes = chunk_bytes_by_type(type, calc_size,
1728 num_stripes, sub_stripes);
1731 while(index < num_stripes) {
1732 struct btrfs_stripe *stripe;
1733 BUG_ON(list_empty(&private_devs));
1734 cur = private_devs.next;
1735 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1737 /* loop over this device again if we're doing a dup group */
1738 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1739 (index == num_stripes - 1))
1740 list_move_tail(&device->dev_alloc_list, dev_list);
1742 ret = btrfs_alloc_dev_extent(trans, device,
1743 info->chunk_root->root_key.objectid,
1744 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1745 calc_size, &dev_offset);
1747 device->bytes_used += calc_size;
1748 ret = btrfs_update_device(trans, device);
1751 map->stripes[index].dev = device;
1752 map->stripes[index].physical = dev_offset;
1753 stripe = stripes + index;
1754 btrfs_set_stack_stripe_devid(stripe, device->devid);
1755 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1756 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1757 physical = dev_offset;
1760 BUG_ON(!list_empty(&private_devs));
1762 /* key was set above */
1763 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1764 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1765 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1766 btrfs_set_stack_chunk_type(chunk, type);
1767 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1768 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1769 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1770 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1771 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1772 map->sector_size = extent_root->sectorsize;
1773 map->stripe_len = stripe_len;
1774 map->io_align = stripe_len;
1775 map->io_width = stripe_len;
1777 map->num_stripes = num_stripes;
1778 map->sub_stripes = sub_stripes;
1780 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1781 btrfs_chunk_item_size(num_stripes));
1783 *start = key.offset;;
1785 em = alloc_extent_map(GFP_NOFS);
1788 em->bdev = (struct block_device *)map;
1789 em->start = key.offset;
1790 em->len = *num_bytes;
1791 em->block_start = 0;
1793 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1794 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1795 chunk, btrfs_chunk_item_size(num_stripes));
1800 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1801 spin_lock(&em_tree->lock);
1802 ret = add_extent_mapping(em_tree, em);
1803 spin_unlock(&em_tree->lock);
1805 free_extent_map(em);
1809 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1811 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1814 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1816 struct extent_map *em;
1819 spin_lock(&tree->map_tree.lock);
1820 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1822 remove_extent_mapping(&tree->map_tree, em);
1823 spin_unlock(&tree->map_tree.lock);
1828 free_extent_map(em);
1829 /* once for the tree */
1830 free_extent_map(em);
1834 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1836 struct extent_map *em;
1837 struct map_lookup *map;
1838 struct extent_map_tree *em_tree = &map_tree->map_tree;
1841 spin_lock(&em_tree->lock);
1842 em = lookup_extent_mapping(em_tree, logical, len);
1843 spin_unlock(&em_tree->lock);
1846 BUG_ON(em->start > logical || em->start + em->len < logical);
1847 map = (struct map_lookup *)em->bdev;
1848 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1849 ret = map->num_stripes;
1850 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1851 ret = map->sub_stripes;
1854 free_extent_map(em);
1858 static int find_live_mirror(struct map_lookup *map, int first, int num,
1862 if (map->stripes[optimal].dev->bdev)
1864 for (i = first; i < first + num; i++) {
1865 if (map->stripes[i].dev->bdev)
1868 /* we couldn't find one that doesn't fail. Just return something
1869 * and the io error handling code will clean up eventually
1874 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1875 u64 logical, u64 *length,
1876 struct btrfs_multi_bio **multi_ret,
1877 int mirror_num, struct page *unplug_page)
1879 struct extent_map *em;
1880 struct map_lookup *map;
1881 struct extent_map_tree *em_tree = &map_tree->map_tree;
1885 int stripes_allocated = 8;
1886 int stripes_required = 1;
1891 struct btrfs_multi_bio *multi = NULL;
1893 if (multi_ret && !(rw & (1 << BIO_RW))) {
1894 stripes_allocated = 1;
1898 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1903 atomic_set(&multi->error, 0);
1906 spin_lock(&em_tree->lock);
1907 em = lookup_extent_mapping(em_tree, logical, *length);
1908 spin_unlock(&em_tree->lock);
1910 if (!em && unplug_page)
1914 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1918 BUG_ON(em->start > logical || em->start + em->len < logical);
1919 map = (struct map_lookup *)em->bdev;
1920 offset = logical - em->start;
1922 if (mirror_num > map->num_stripes)
1925 /* if our multi bio struct is too small, back off and try again */
1926 if (rw & (1 << BIO_RW)) {
1927 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1928 BTRFS_BLOCK_GROUP_DUP)) {
1929 stripes_required = map->num_stripes;
1931 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1932 stripes_required = map->sub_stripes;
1936 if (multi_ret && rw == WRITE &&
1937 stripes_allocated < stripes_required) {
1938 stripes_allocated = map->num_stripes;
1939 free_extent_map(em);
1945 * stripe_nr counts the total number of stripes we have to stride
1946 * to get to this block
1948 do_div(stripe_nr, map->stripe_len);
1950 stripe_offset = stripe_nr * map->stripe_len;
1951 BUG_ON(offset < stripe_offset);
1953 /* stripe_offset is the offset of this block in its stripe*/
1954 stripe_offset = offset - stripe_offset;
1956 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1957 BTRFS_BLOCK_GROUP_RAID10 |
1958 BTRFS_BLOCK_GROUP_DUP)) {
1959 /* we limit the length of each bio to what fits in a stripe */
1960 *length = min_t(u64, em->len - offset,
1961 map->stripe_len - stripe_offset);
1963 *length = em->len - offset;
1966 if (!multi_ret && !unplug_page)
1971 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1972 if (unplug_page || (rw & (1 << BIO_RW)))
1973 num_stripes = map->num_stripes;
1974 else if (mirror_num)
1975 stripe_index = mirror_num - 1;
1977 stripe_index = find_live_mirror(map, 0,
1979 current->pid % map->num_stripes);
1982 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1983 if (rw & (1 << BIO_RW))
1984 num_stripes = map->num_stripes;
1985 else if (mirror_num)
1986 stripe_index = mirror_num - 1;
1988 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1989 int factor = map->num_stripes / map->sub_stripes;
1991 stripe_index = do_div(stripe_nr, factor);
1992 stripe_index *= map->sub_stripes;
1994 if (unplug_page || (rw & (1 << BIO_RW)))
1995 num_stripes = map->sub_stripes;
1996 else if (mirror_num)
1997 stripe_index += mirror_num - 1;
1999 stripe_index = find_live_mirror(map, stripe_index,
2000 map->sub_stripes, stripe_index +
2001 current->pid % map->sub_stripes);
2005 * after this do_div call, stripe_nr is the number of stripes
2006 * on this device we have to walk to find the data, and
2007 * stripe_index is the number of our device in the stripe array
2009 stripe_index = do_div(stripe_nr, map->num_stripes);
2011 BUG_ON(stripe_index >= map->num_stripes);
2013 for (i = 0; i < num_stripes; i++) {
2015 struct btrfs_device *device;
2016 struct backing_dev_info *bdi;
2018 device = map->stripes[stripe_index].dev;
2020 bdi = blk_get_backing_dev_info(device->bdev);
2021 if (bdi->unplug_io_fn) {
2022 bdi->unplug_io_fn(bdi, unplug_page);
2026 multi->stripes[i].physical =
2027 map->stripes[stripe_index].physical +
2028 stripe_offset + stripe_nr * map->stripe_len;
2029 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2035 multi->num_stripes = num_stripes;
2036 multi->max_errors = max_errors;
2039 free_extent_map(em);
2043 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2044 u64 logical, u64 *length,
2045 struct btrfs_multi_bio **multi_ret, int mirror_num)
2047 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2051 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2052 u64 logical, struct page *page)
2054 u64 length = PAGE_CACHE_SIZE;
2055 return __btrfs_map_block(map_tree, READ, logical, &length,
2060 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2061 static void end_bio_multi_stripe(struct bio *bio, int err)
2063 static int end_bio_multi_stripe(struct bio *bio,
2064 unsigned int bytes_done, int err)
2067 struct btrfs_multi_bio *multi = bio->bi_private;
2069 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2074 atomic_inc(&multi->error);
2076 if (atomic_dec_and_test(&multi->stripes_pending)) {
2077 bio->bi_private = multi->private;
2078 bio->bi_end_io = multi->end_io;
2079 /* only send an error to the higher layers if it is
2080 * beyond the tolerance of the multi-bio
2082 if (atomic_read(&multi->error) > multi->max_errors) {
2086 * this bio is actually up to date, we didn't
2087 * go over the max number of errors
2089 set_bit(BIO_UPTODATE, &bio->bi_flags);
2094 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2095 bio_endio(bio, bio->bi_size, err);
2097 bio_endio(bio, err);
2102 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2107 struct async_sched {
2110 struct btrfs_fs_info *info;
2111 struct btrfs_work work;
2115 * see run_scheduled_bios for a description of why bios are collected for
2118 * This will add one bio to the pending list for a device and make sure
2119 * the work struct is scheduled.
2121 int schedule_bio(struct btrfs_root *root, struct btrfs_device *device,
2122 int rw, struct bio *bio)
2124 int should_queue = 1;
2126 /* don't bother with additional async steps for reads, right now */
2127 if (!(rw & (1 << BIO_RW))) {
2128 submit_bio(rw, bio);
2133 * nr_async_sumbits allows us to reliably return congestion to the
2134 * higher layers. Otherwise, the async bio makes it appear we have
2135 * made progress against dirty pages when we've really just put it
2136 * on a queue for later
2138 atomic_inc(&root->fs_info->nr_async_submits);
2139 bio->bi_next = NULL;
2142 spin_lock(&device->io_lock);
2144 if (device->pending_bio_tail)
2145 device->pending_bio_tail->bi_next = bio;
2147 device->pending_bio_tail = bio;
2148 if (!device->pending_bios)
2149 device->pending_bios = bio;
2150 if (device->running_pending)
2153 spin_unlock(&device->io_lock);
2156 btrfs_queue_worker(&root->fs_info->submit_workers,
2161 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2162 int mirror_num, int async_submit)
2164 struct btrfs_mapping_tree *map_tree;
2165 struct btrfs_device *dev;
2166 struct bio *first_bio = bio;
2167 u64 logical = bio->bi_sector << 9;
2170 struct btrfs_multi_bio *multi = NULL;
2175 length = bio->bi_size;
2176 map_tree = &root->fs_info->mapping_tree;
2177 map_length = length;
2179 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2183 total_devs = multi->num_stripes;
2184 if (map_length < length) {
2185 printk("mapping failed logical %Lu bio len %Lu "
2186 "len %Lu\n", logical, length, map_length);
2189 multi->end_io = first_bio->bi_end_io;
2190 multi->private = first_bio->bi_private;
2191 atomic_set(&multi->stripes_pending, multi->num_stripes);
2193 while(dev_nr < total_devs) {
2194 if (total_devs > 1) {
2195 if (dev_nr < total_devs - 1) {
2196 bio = bio_clone(first_bio, GFP_NOFS);
2201 bio->bi_private = multi;
2202 bio->bi_end_io = end_bio_multi_stripe;
2204 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2205 dev = multi->stripes[dev_nr].dev;
2206 if (dev && dev->bdev) {
2207 bio->bi_bdev = dev->bdev;
2209 schedule_bio(root, dev, rw, bio);
2211 submit_bio(rw, bio);
2213 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2214 bio->bi_sector = logical >> 9;
2215 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2216 bio_endio(bio, bio->bi_size, -EIO);
2218 bio_endio(bio, -EIO);
2223 if (total_devs == 1)
2228 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2231 struct list_head *head = &root->fs_info->fs_devices->devices;
2233 return __find_device(head, devid, uuid);
2236 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2237 u64 devid, u8 *dev_uuid)
2239 struct btrfs_device *device;
2240 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2242 device = kzalloc(sizeof(*device), GFP_NOFS);
2243 list_add(&device->dev_list,
2244 &fs_devices->devices);
2245 list_add(&device->dev_alloc_list,
2246 &fs_devices->alloc_list);
2247 device->barriers = 1;
2248 device->dev_root = root->fs_info->dev_root;
2249 device->devid = devid;
2250 device->work.func = pending_bios_fn;
2251 fs_devices->num_devices++;
2252 spin_lock_init(&device->io_lock);
2253 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2258 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2259 struct extent_buffer *leaf,
2260 struct btrfs_chunk *chunk)
2262 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2263 struct map_lookup *map;
2264 struct extent_map *em;
2268 u8 uuid[BTRFS_UUID_SIZE];
2273 logical = key->offset;
2274 length = btrfs_chunk_length(leaf, chunk);
2276 spin_lock(&map_tree->map_tree.lock);
2277 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2278 spin_unlock(&map_tree->map_tree.lock);
2280 /* already mapped? */
2281 if (em && em->start <= logical && em->start + em->len > logical) {
2282 free_extent_map(em);
2285 free_extent_map(em);
2288 map = kzalloc(sizeof(*map), GFP_NOFS);
2292 em = alloc_extent_map(GFP_NOFS);
2295 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2296 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2298 free_extent_map(em);
2302 em->bdev = (struct block_device *)map;
2303 em->start = logical;
2305 em->block_start = 0;
2307 map->num_stripes = num_stripes;
2308 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2309 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2310 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2311 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2312 map->type = btrfs_chunk_type(leaf, chunk);
2313 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2314 for (i = 0; i < num_stripes; i++) {
2315 map->stripes[i].physical =
2316 btrfs_stripe_offset_nr(leaf, chunk, i);
2317 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2318 read_extent_buffer(leaf, uuid, (unsigned long)
2319 btrfs_stripe_dev_uuid_nr(chunk, i),
2321 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2323 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2325 free_extent_map(em);
2328 if (!map->stripes[i].dev) {
2329 map->stripes[i].dev =
2330 add_missing_dev(root, devid, uuid);
2331 if (!map->stripes[i].dev) {
2333 free_extent_map(em);
2337 map->stripes[i].dev->in_fs_metadata = 1;
2340 spin_lock(&map_tree->map_tree.lock);
2341 ret = add_extent_mapping(&map_tree->map_tree, em);
2342 spin_unlock(&map_tree->map_tree.lock);
2344 free_extent_map(em);
2349 static int fill_device_from_item(struct extent_buffer *leaf,
2350 struct btrfs_dev_item *dev_item,
2351 struct btrfs_device *device)
2355 device->devid = btrfs_device_id(leaf, dev_item);
2356 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2357 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2358 device->type = btrfs_device_type(leaf, dev_item);
2359 device->io_align = btrfs_device_io_align(leaf, dev_item);
2360 device->io_width = btrfs_device_io_width(leaf, dev_item);
2361 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2363 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2364 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2369 static int read_one_dev(struct btrfs_root *root,
2370 struct extent_buffer *leaf,
2371 struct btrfs_dev_item *dev_item)
2373 struct btrfs_device *device;
2376 u8 dev_uuid[BTRFS_UUID_SIZE];
2378 devid = btrfs_device_id(leaf, dev_item);
2379 read_extent_buffer(leaf, dev_uuid,
2380 (unsigned long)btrfs_device_uuid(dev_item),
2382 device = btrfs_find_device(root, devid, dev_uuid);
2384 printk("warning devid %Lu missing\n", devid);
2385 device = add_missing_dev(root, devid, dev_uuid);
2390 fill_device_from_item(leaf, dev_item, device);
2391 device->dev_root = root->fs_info->dev_root;
2392 device->in_fs_metadata = 1;
2395 ret = btrfs_open_device(device);
2403 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2405 struct btrfs_dev_item *dev_item;
2407 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2409 return read_one_dev(root, buf, dev_item);
2412 int btrfs_read_sys_array(struct btrfs_root *root)
2414 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2415 struct extent_buffer *sb;
2416 struct btrfs_disk_key *disk_key;
2417 struct btrfs_chunk *chunk;
2419 unsigned long sb_ptr;
2425 struct btrfs_key key;
2427 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2428 BTRFS_SUPER_INFO_SIZE);
2431 btrfs_set_buffer_uptodate(sb);
2432 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2433 array_size = btrfs_super_sys_array_size(super_copy);
2435 ptr = super_copy->sys_chunk_array;
2436 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2439 while (cur < array_size) {
2440 disk_key = (struct btrfs_disk_key *)ptr;
2441 btrfs_disk_key_to_cpu(&key, disk_key);
2443 len = sizeof(*disk_key); ptr += len;
2447 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2448 chunk = (struct btrfs_chunk *)sb_ptr;
2449 ret = read_one_chunk(root, &key, sb, chunk);
2452 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2453 len = btrfs_chunk_item_size(num_stripes);
2462 free_extent_buffer(sb);
2466 int btrfs_read_chunk_tree(struct btrfs_root *root)
2468 struct btrfs_path *path;
2469 struct extent_buffer *leaf;
2470 struct btrfs_key key;
2471 struct btrfs_key found_key;
2475 root = root->fs_info->chunk_root;
2477 path = btrfs_alloc_path();
2481 /* first we search for all of the device items, and then we
2482 * read in all of the chunk items. This way we can create chunk
2483 * mappings that reference all of the devices that are afound
2485 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2489 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2491 leaf = path->nodes[0];
2492 slot = path->slots[0];
2493 if (slot >= btrfs_header_nritems(leaf)) {
2494 ret = btrfs_next_leaf(root, path);
2501 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2502 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2503 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2505 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2506 struct btrfs_dev_item *dev_item;
2507 dev_item = btrfs_item_ptr(leaf, slot,
2508 struct btrfs_dev_item);
2509 ret = read_one_dev(root, leaf, dev_item);
2512 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2513 struct btrfs_chunk *chunk;
2514 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2515 ret = read_one_chunk(root, &found_key, leaf, chunk);
2519 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2521 btrfs_release_path(root, path);
2525 btrfs_free_path(path);