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 int btrfs_cleanup_fs_uuids(void)
61 struct btrfs_fs_devices *fs_devices;
62 struct list_head *uuid_cur;
63 struct list_head *devices_cur;
64 struct btrfs_device *dev;
66 list_for_each(uuid_cur, &fs_uuids) {
67 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
69 while(!list_empty(&fs_devices->devices)) {
70 devices_cur = fs_devices->devices.next;
71 dev = list_entry(devices_cur, struct btrfs_device,
74 close_bdev_excl(dev->bdev);
75 fs_devices->open_devices--;
77 list_del(&dev->dev_list);
85 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
88 struct btrfs_device *dev;
89 struct list_head *cur;
91 list_for_each(cur, head) {
92 dev = list_entry(cur, struct btrfs_device, dev_list);
93 if (dev->devid == devid &&
94 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
101 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
103 struct list_head *cur;
104 struct btrfs_fs_devices *fs_devices;
106 list_for_each(cur, &fs_uuids) {
107 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
108 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
115 * we try to collect pending bios for a device so we don't get a large
116 * number of procs sending bios down to the same device. This greatly
117 * improves the schedulers ability to collect and merge the bios.
119 * But, it also turns into a long list of bios to process and that is sure
120 * to eventually make the worker thread block. The solution here is to
121 * make some progress and then put this work struct back at the end of
122 * the list if the block device is congested. This way, multiple devices
123 * can make progress from a single worker thread.
125 int run_scheduled_bios(struct btrfs_device *device)
128 struct backing_dev_info *bdi;
132 unsigned long num_run = 0;
134 bdi = device->bdev->bd_inode->i_mapping->backing_dev_info;
136 spin_lock(&device->io_lock);
138 /* take all the bios off the list at once and process them
139 * later on (without the lock held). But, remember the
140 * tail and other pointers so the bios can be properly reinserted
141 * into the list if we hit congestion
143 pending = device->pending_bios;
144 tail = device->pending_bio_tail;
145 WARN_ON(pending && !tail);
146 device->pending_bios = NULL;
147 device->pending_bio_tail = NULL;
150 * if pending was null this time around, no bios need processing
151 * at all and we can stop. Otherwise it'll loop back up again
152 * and do an additional check so no bios are missed.
154 * device->running_pending is used to synchronize with the
159 device->running_pending = 1;
162 device->running_pending = 0;
164 spin_unlock(&device->io_lock);
168 pending = pending->bi_next;
170 atomic_dec(&device->dev_root->fs_info->nr_async_submits);
171 submit_bio(cur->bi_rw, cur);
175 * we made progress, there is more work to do and the bdi
176 * is now congested. Back off and let other work structs
179 if (pending && num_run && bdi_write_congested(bdi)) {
180 struct bio *old_head;
182 spin_lock(&device->io_lock);
183 old_head = device->pending_bios;
184 device->pending_bios = pending;
185 if (device->pending_bio_tail)
186 tail->bi_next = old_head;
188 device->pending_bio_tail = tail;
190 spin_unlock(&device->io_lock);
191 btrfs_requeue_work(&device->work);
201 void pending_bios_fn(struct btrfs_work *work)
203 struct btrfs_device *device;
205 device = container_of(work, struct btrfs_device, work);
206 run_scheduled_bios(device);
209 static int device_list_add(const char *path,
210 struct btrfs_super_block *disk_super,
211 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
213 struct btrfs_device *device;
214 struct btrfs_fs_devices *fs_devices;
215 u64 found_transid = btrfs_super_generation(disk_super);
217 fs_devices = find_fsid(disk_super->fsid);
219 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
222 INIT_LIST_HEAD(&fs_devices->devices);
223 INIT_LIST_HEAD(&fs_devices->alloc_list);
224 list_add(&fs_devices->list, &fs_uuids);
225 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
226 fs_devices->latest_devid = devid;
227 fs_devices->latest_trans = found_transid;
230 device = __find_device(&fs_devices->devices, devid,
231 disk_super->dev_item.uuid);
234 device = kzalloc(sizeof(*device), GFP_NOFS);
236 /* we can safely leave the fs_devices entry around */
239 device->devid = devid;
240 device->work.func = pending_bios_fn;
241 memcpy(device->uuid, disk_super->dev_item.uuid,
243 device->barriers = 1;
244 spin_lock_init(&device->io_lock);
245 device->name = kstrdup(path, GFP_NOFS);
250 list_add(&device->dev_list, &fs_devices->devices);
251 list_add(&device->dev_alloc_list, &fs_devices->alloc_list);
252 fs_devices->num_devices++;
255 if (found_transid > fs_devices->latest_trans) {
256 fs_devices->latest_devid = devid;
257 fs_devices->latest_trans = found_transid;
259 *fs_devices_ret = fs_devices;
263 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
265 struct list_head *head = &fs_devices->devices;
266 struct list_head *cur;
267 struct btrfs_device *device;
269 mutex_lock(&uuid_mutex);
271 list_for_each(cur, head) {
272 device = list_entry(cur, struct btrfs_device, dev_list);
273 if (!device->in_fs_metadata) {
275 close_bdev_excl(device->bdev);
276 fs_devices->open_devices--;
278 list_del(&device->dev_list);
279 list_del(&device->dev_alloc_list);
280 fs_devices->num_devices--;
286 mutex_unlock(&uuid_mutex);
290 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
292 struct list_head *head = &fs_devices->devices;
293 struct list_head *cur;
294 struct btrfs_device *device;
296 mutex_lock(&uuid_mutex);
297 list_for_each(cur, head) {
298 device = list_entry(cur, struct btrfs_device, dev_list);
300 close_bdev_excl(device->bdev);
301 fs_devices->open_devices--;
304 device->in_fs_metadata = 0;
306 fs_devices->mounted = 0;
307 mutex_unlock(&uuid_mutex);
311 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
312 int flags, void *holder)
314 struct block_device *bdev;
315 struct list_head *head = &fs_devices->devices;
316 struct list_head *cur;
317 struct btrfs_device *device;
318 struct block_device *latest_bdev = NULL;
319 struct buffer_head *bh;
320 struct btrfs_super_block *disk_super;
321 u64 latest_devid = 0;
322 u64 latest_transid = 0;
327 mutex_lock(&uuid_mutex);
328 if (fs_devices->mounted)
331 list_for_each(cur, head) {
332 device = list_entry(cur, struct btrfs_device, dev_list);
339 bdev = open_bdev_excl(device->name, flags, holder);
342 printk("open %s failed\n", device->name);
345 set_blocksize(bdev, 4096);
347 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
351 disk_super = (struct btrfs_super_block *)bh->b_data;
352 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
353 sizeof(disk_super->magic)))
356 devid = le64_to_cpu(disk_super->dev_item.devid);
357 if (devid != device->devid)
360 transid = btrfs_super_generation(disk_super);
361 if (!latest_transid || transid > latest_transid) {
362 latest_devid = devid;
363 latest_transid = transid;
368 device->in_fs_metadata = 0;
369 fs_devices->open_devices++;
375 close_bdev_excl(bdev);
379 if (fs_devices->open_devices == 0) {
383 fs_devices->mounted = 1;
384 fs_devices->latest_bdev = latest_bdev;
385 fs_devices->latest_devid = latest_devid;
386 fs_devices->latest_trans = latest_transid;
388 mutex_unlock(&uuid_mutex);
392 int btrfs_scan_one_device(const char *path, int flags, void *holder,
393 struct btrfs_fs_devices **fs_devices_ret)
395 struct btrfs_super_block *disk_super;
396 struct block_device *bdev;
397 struct buffer_head *bh;
402 mutex_lock(&uuid_mutex);
404 bdev = open_bdev_excl(path, flags, holder);
411 ret = set_blocksize(bdev, 4096);
414 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
419 disk_super = (struct btrfs_super_block *)bh->b_data;
420 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
421 sizeof(disk_super->magic))) {
425 devid = le64_to_cpu(disk_super->dev_item.devid);
426 transid = btrfs_super_generation(disk_super);
427 if (disk_super->label[0])
428 printk("device label %s ", disk_super->label);
430 /* FIXME, make a readl uuid parser */
431 printk("device fsid %llx-%llx ",
432 *(unsigned long long *)disk_super->fsid,
433 *(unsigned long long *)(disk_super->fsid + 8));
435 printk("devid %Lu transid %Lu %s\n", devid, transid, path);
436 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
441 close_bdev_excl(bdev);
443 mutex_unlock(&uuid_mutex);
448 * this uses a pretty simple search, the expectation is that it is
449 * called very infrequently and that a given device has a small number
452 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
453 struct btrfs_device *device,
454 struct btrfs_path *path,
455 u64 num_bytes, u64 *start)
457 struct btrfs_key key;
458 struct btrfs_root *root = device->dev_root;
459 struct btrfs_dev_extent *dev_extent = NULL;
462 u64 search_start = 0;
463 u64 search_end = device->total_bytes;
467 struct extent_buffer *l;
472 /* FIXME use last free of some kind */
474 /* we don't want to overwrite the superblock on the drive,
475 * so we make sure to start at an offset of at least 1MB
477 search_start = max((u64)1024 * 1024, search_start);
479 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
480 search_start = max(root->fs_info->alloc_start, search_start);
482 key.objectid = device->devid;
483 key.offset = search_start;
484 key.type = BTRFS_DEV_EXTENT_KEY;
485 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
488 ret = btrfs_previous_item(root, path, 0, key.type);
492 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
495 slot = path->slots[0];
496 if (slot >= btrfs_header_nritems(l)) {
497 ret = btrfs_next_leaf(root, path);
504 if (search_start >= search_end) {
508 *start = search_start;
512 *start = last_byte > search_start ?
513 last_byte : search_start;
514 if (search_end <= *start) {
520 btrfs_item_key_to_cpu(l, &key, slot);
522 if (key.objectid < device->devid)
525 if (key.objectid > device->devid)
528 if (key.offset >= search_start && key.offset > last_byte &&
530 if (last_byte < search_start)
531 last_byte = search_start;
532 hole_size = key.offset - last_byte;
533 if (key.offset > last_byte &&
534 hole_size >= num_bytes) {
539 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
544 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
545 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
551 /* we have to make sure we didn't find an extent that has already
552 * been allocated by the map tree or the original allocation
554 btrfs_release_path(root, path);
555 BUG_ON(*start < search_start);
557 if (*start + num_bytes > search_end) {
561 /* check for pending inserts here */
565 btrfs_release_path(root, path);
569 int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
570 struct btrfs_device *device,
574 struct btrfs_path *path;
575 struct btrfs_root *root = device->dev_root;
576 struct btrfs_key key;
577 struct btrfs_key found_key;
578 struct extent_buffer *leaf = NULL;
579 struct btrfs_dev_extent *extent = NULL;
581 path = btrfs_alloc_path();
585 key.objectid = device->devid;
587 key.type = BTRFS_DEV_EXTENT_KEY;
589 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
591 ret = btrfs_previous_item(root, path, key.objectid,
592 BTRFS_DEV_EXTENT_KEY);
594 leaf = path->nodes[0];
595 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
596 extent = btrfs_item_ptr(leaf, path->slots[0],
597 struct btrfs_dev_extent);
598 BUG_ON(found_key.offset > start || found_key.offset +
599 btrfs_dev_extent_length(leaf, extent) < start);
601 } else if (ret == 0) {
602 leaf = path->nodes[0];
603 extent = btrfs_item_ptr(leaf, path->slots[0],
604 struct btrfs_dev_extent);
608 if (device->bytes_used > 0)
609 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
610 ret = btrfs_del_item(trans, root, path);
613 btrfs_free_path(path);
617 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
618 struct btrfs_device *device,
619 u64 chunk_tree, u64 chunk_objectid,
621 u64 num_bytes, u64 *start)
624 struct btrfs_path *path;
625 struct btrfs_root *root = device->dev_root;
626 struct btrfs_dev_extent *extent;
627 struct extent_buffer *leaf;
628 struct btrfs_key key;
630 WARN_ON(!device->in_fs_metadata);
631 path = btrfs_alloc_path();
635 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
640 key.objectid = device->devid;
642 key.type = BTRFS_DEV_EXTENT_KEY;
643 ret = btrfs_insert_empty_item(trans, root, path, &key,
647 leaf = path->nodes[0];
648 extent = btrfs_item_ptr(leaf, path->slots[0],
649 struct btrfs_dev_extent);
650 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
651 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
652 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
654 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
655 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
658 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
659 btrfs_mark_buffer_dirty(leaf);
661 btrfs_free_path(path);
665 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
667 struct btrfs_path *path;
669 struct btrfs_key key;
670 struct btrfs_chunk *chunk;
671 struct btrfs_key found_key;
673 path = btrfs_alloc_path();
676 key.objectid = objectid;
677 key.offset = (u64)-1;
678 key.type = BTRFS_CHUNK_ITEM_KEY;
680 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
686 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
690 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
692 if (found_key.objectid != objectid)
695 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
697 *offset = found_key.offset +
698 btrfs_chunk_length(path->nodes[0], chunk);
703 btrfs_free_path(path);
707 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
711 struct btrfs_key key;
712 struct btrfs_key found_key;
714 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
715 key.type = BTRFS_DEV_ITEM_KEY;
716 key.offset = (u64)-1;
718 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
724 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
729 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
731 *objectid = found_key.offset + 1;
735 btrfs_release_path(root, path);
740 * the device information is stored in the chunk root
741 * the btrfs_device struct should be fully filled in
743 int btrfs_add_device(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root,
745 struct btrfs_device *device)
748 struct btrfs_path *path;
749 struct btrfs_dev_item *dev_item;
750 struct extent_buffer *leaf;
751 struct btrfs_key key;
755 root = root->fs_info->chunk_root;
757 path = btrfs_alloc_path();
761 ret = find_next_devid(root, path, &free_devid);
765 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
766 key.type = BTRFS_DEV_ITEM_KEY;
767 key.offset = free_devid;
769 ret = btrfs_insert_empty_item(trans, root, path, &key,
774 leaf = path->nodes[0];
775 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
777 device->devid = free_devid;
778 btrfs_set_device_id(leaf, dev_item, device->devid);
779 btrfs_set_device_type(leaf, dev_item, device->type);
780 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
781 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
782 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
783 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
784 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
785 btrfs_set_device_group(leaf, dev_item, 0);
786 btrfs_set_device_seek_speed(leaf, dev_item, 0);
787 btrfs_set_device_bandwidth(leaf, dev_item, 0);
789 ptr = (unsigned long)btrfs_device_uuid(dev_item);
790 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
791 btrfs_mark_buffer_dirty(leaf);
795 btrfs_free_path(path);
799 static int btrfs_rm_dev_item(struct btrfs_root *root,
800 struct btrfs_device *device)
803 struct btrfs_path *path;
804 struct block_device *bdev = device->bdev;
805 struct btrfs_device *next_dev;
806 struct btrfs_key key;
808 struct btrfs_fs_devices *fs_devices;
809 struct btrfs_trans_handle *trans;
811 root = root->fs_info->chunk_root;
813 path = btrfs_alloc_path();
817 trans = btrfs_start_transaction(root, 1);
818 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
819 key.type = BTRFS_DEV_ITEM_KEY;
820 key.offset = device->devid;
822 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
831 ret = btrfs_del_item(trans, root, path);
836 * at this point, the device is zero sized. We want to
837 * remove it from the devices list and zero out the old super
839 list_del_init(&device->dev_list);
840 list_del_init(&device->dev_alloc_list);
841 fs_devices = root->fs_info->fs_devices;
843 next_dev = list_entry(fs_devices->devices.next, struct btrfs_device,
845 if (bdev == root->fs_info->sb->s_bdev)
846 root->fs_info->sb->s_bdev = next_dev->bdev;
847 if (bdev == fs_devices->latest_bdev)
848 fs_devices->latest_bdev = next_dev->bdev;
850 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
851 btrfs_set_super_num_devices(&root->fs_info->super_copy,
854 btrfs_free_path(path);
855 btrfs_commit_transaction(trans, root);
859 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
861 struct btrfs_device *device;
862 struct block_device *bdev;
863 struct buffer_head *bh = NULL;
864 struct btrfs_super_block *disk_super;
869 mutex_lock(&root->fs_info->fs_mutex);
870 mutex_lock(&uuid_mutex);
872 all_avail = root->fs_info->avail_data_alloc_bits |
873 root->fs_info->avail_system_alloc_bits |
874 root->fs_info->avail_metadata_alloc_bits;
876 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
877 btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) {
878 printk("btrfs: unable to go below four devices on raid10\n");
883 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
884 btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) {
885 printk("btrfs: unable to go below two devices on raid1\n");
890 if (strcmp(device_path, "missing") == 0) {
891 struct list_head *cur;
892 struct list_head *devices;
893 struct btrfs_device *tmp;
896 devices = &root->fs_info->fs_devices->devices;
897 list_for_each(cur, devices) {
898 tmp = list_entry(cur, struct btrfs_device, dev_list);
899 if (tmp->in_fs_metadata && !tmp->bdev) {
908 printk("btrfs: no missing devices found to remove\n");
913 bdev = open_bdev_excl(device_path, 0,
914 root->fs_info->bdev_holder);
920 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
925 disk_super = (struct btrfs_super_block *)bh->b_data;
926 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
927 sizeof(disk_super->magic))) {
931 if (memcmp(disk_super->fsid, root->fs_info->fsid,
936 devid = le64_to_cpu(disk_super->dev_item.devid);
937 device = btrfs_find_device(root, devid, NULL);
944 root->fs_info->fs_devices->num_devices--;
945 root->fs_info->fs_devices->open_devices--;
947 ret = btrfs_shrink_device(device, 0);
952 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
957 /* make sure this device isn't detected as part of
960 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
961 set_buffer_dirty(bh);
962 sync_dirty_buffer(bh);
968 /* one close for the device struct or super_block */
969 close_bdev_excl(device->bdev);
972 /* one close for us */
973 close_bdev_excl(bdev);
984 close_bdev_excl(bdev);
986 mutex_unlock(&uuid_mutex);
987 mutex_unlock(&root->fs_info->fs_mutex);
991 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
993 struct btrfs_trans_handle *trans;
994 struct btrfs_device *device;
995 struct block_device *bdev;
996 struct list_head *cur;
997 struct list_head *devices;
1002 bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder);
1006 mutex_lock(&root->fs_info->fs_mutex);
1007 trans = btrfs_start_transaction(root, 1);
1008 devices = &root->fs_info->fs_devices->devices;
1009 list_for_each(cur, devices) {
1010 device = list_entry(cur, struct btrfs_device, dev_list);
1011 if (device->bdev == bdev) {
1017 device = kzalloc(sizeof(*device), GFP_NOFS);
1019 /* we can safely leave the fs_devices entry around */
1021 goto out_close_bdev;
1024 device->barriers = 1;
1025 device->work.func = pending_bios_fn;
1026 generate_random_uuid(device->uuid);
1027 spin_lock_init(&device->io_lock);
1028 device->name = kstrdup(device_path, GFP_NOFS);
1029 if (!device->name) {
1031 goto out_close_bdev;
1033 device->io_width = root->sectorsize;
1034 device->io_align = root->sectorsize;
1035 device->sector_size = root->sectorsize;
1036 device->total_bytes = i_size_read(bdev->bd_inode);
1037 device->dev_root = root->fs_info->dev_root;
1038 device->bdev = bdev;
1039 device->in_fs_metadata = 1;
1041 ret = btrfs_add_device(trans, root, device);
1043 goto out_close_bdev;
1045 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1046 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1047 total_bytes + device->total_bytes);
1049 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1050 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1053 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1054 list_add(&device->dev_alloc_list,
1055 &root->fs_info->fs_devices->alloc_list);
1056 root->fs_info->fs_devices->num_devices++;
1057 root->fs_info->fs_devices->open_devices++;
1059 btrfs_end_transaction(trans, root);
1060 mutex_unlock(&root->fs_info->fs_mutex);
1064 close_bdev_excl(bdev);
1068 int btrfs_update_device(struct btrfs_trans_handle *trans,
1069 struct btrfs_device *device)
1072 struct btrfs_path *path;
1073 struct btrfs_root *root;
1074 struct btrfs_dev_item *dev_item;
1075 struct extent_buffer *leaf;
1076 struct btrfs_key key;
1078 root = device->dev_root->fs_info->chunk_root;
1080 path = btrfs_alloc_path();
1084 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1085 key.type = BTRFS_DEV_ITEM_KEY;
1086 key.offset = device->devid;
1088 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1097 leaf = path->nodes[0];
1098 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1100 btrfs_set_device_id(leaf, dev_item, device->devid);
1101 btrfs_set_device_type(leaf, dev_item, device->type);
1102 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1103 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1104 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1105 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1106 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1107 btrfs_mark_buffer_dirty(leaf);
1110 btrfs_free_path(path);
1114 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1115 struct btrfs_device *device, u64 new_size)
1117 struct btrfs_super_block *super_copy =
1118 &device->dev_root->fs_info->super_copy;
1119 u64 old_total = btrfs_super_total_bytes(super_copy);
1120 u64 diff = new_size - device->total_bytes;
1122 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1123 return btrfs_update_device(trans, device);
1126 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1127 struct btrfs_root *root,
1128 u64 chunk_tree, u64 chunk_objectid,
1132 struct btrfs_path *path;
1133 struct btrfs_key key;
1135 root = root->fs_info->chunk_root;
1136 path = btrfs_alloc_path();
1140 key.objectid = chunk_objectid;
1141 key.offset = chunk_offset;
1142 key.type = BTRFS_CHUNK_ITEM_KEY;
1144 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1147 ret = btrfs_del_item(trans, root, path);
1150 btrfs_free_path(path);
1154 int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1157 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1158 struct btrfs_disk_key *disk_key;
1159 struct btrfs_chunk *chunk;
1166 struct btrfs_key key;
1168 array_size = btrfs_super_sys_array_size(super_copy);
1170 ptr = super_copy->sys_chunk_array;
1173 while (cur < array_size) {
1174 disk_key = (struct btrfs_disk_key *)ptr;
1175 btrfs_disk_key_to_cpu(&key, disk_key);
1177 len = sizeof(*disk_key);
1179 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1180 chunk = (struct btrfs_chunk *)(ptr + len);
1181 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1182 len += btrfs_chunk_item_size(num_stripes);
1187 if (key.objectid == chunk_objectid &&
1188 key.offset == chunk_offset) {
1189 memmove(ptr, ptr + len, array_size - (cur + len));
1191 btrfs_set_super_sys_array_size(super_copy, array_size);
1201 int btrfs_relocate_chunk(struct btrfs_root *root,
1202 u64 chunk_tree, u64 chunk_objectid,
1205 struct extent_map_tree *em_tree;
1206 struct btrfs_root *extent_root;
1207 struct btrfs_trans_handle *trans;
1208 struct extent_map *em;
1209 struct map_lookup *map;
1213 printk("btrfs relocating chunk %llu\n",
1214 (unsigned long long)chunk_offset);
1215 root = root->fs_info->chunk_root;
1216 extent_root = root->fs_info->extent_root;
1217 em_tree = &root->fs_info->mapping_tree.map_tree;
1219 /* step one, relocate all the extents inside this chunk */
1220 ret = btrfs_shrink_extent_tree(extent_root, chunk_offset);
1223 trans = btrfs_start_transaction(root, 1);
1227 * step two, delete the device extents and the
1228 * chunk tree entries
1230 spin_lock(&em_tree->lock);
1231 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1232 spin_unlock(&em_tree->lock);
1234 BUG_ON(em->start > chunk_offset ||
1235 em->start + em->len < chunk_offset);
1236 map = (struct map_lookup *)em->bdev;
1238 for (i = 0; i < map->num_stripes; i++) {
1239 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1240 map->stripes[i].physical);
1243 if (map->stripes[i].dev) {
1244 ret = btrfs_update_device(trans, map->stripes[i].dev);
1248 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1253 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1254 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1258 spin_lock(&em_tree->lock);
1259 remove_extent_mapping(em_tree, em);
1263 /* once for the tree */
1264 free_extent_map(em);
1265 spin_unlock(&em_tree->lock);
1268 free_extent_map(em);
1270 btrfs_end_transaction(trans, root);
1274 static u64 div_factor(u64 num, int factor)
1284 int btrfs_balance(struct btrfs_root *dev_root)
1287 struct list_head *cur;
1288 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1289 struct btrfs_device *device;
1292 struct btrfs_path *path;
1293 struct btrfs_key key;
1294 struct btrfs_chunk *chunk;
1295 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1296 struct btrfs_trans_handle *trans;
1297 struct btrfs_key found_key;
1300 dev_root = dev_root->fs_info->dev_root;
1302 mutex_lock(&dev_root->fs_info->fs_mutex);
1303 /* step one make some room on all the devices */
1304 list_for_each(cur, devices) {
1305 device = list_entry(cur, struct btrfs_device, dev_list);
1306 old_size = device->total_bytes;
1307 size_to_free = div_factor(old_size, 1);
1308 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1309 if (device->total_bytes - device->bytes_used > size_to_free)
1312 ret = btrfs_shrink_device(device, old_size - size_to_free);
1315 trans = btrfs_start_transaction(dev_root, 1);
1318 ret = btrfs_grow_device(trans, device, old_size);
1321 btrfs_end_transaction(trans, dev_root);
1324 /* step two, relocate all the chunks */
1325 path = btrfs_alloc_path();
1328 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1329 key.offset = (u64)-1;
1330 key.type = BTRFS_CHUNK_ITEM_KEY;
1333 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1338 * this shouldn't happen, it means the last relocate
1344 ret = btrfs_previous_item(chunk_root, path, 0,
1345 BTRFS_CHUNK_ITEM_KEY);
1349 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1351 if (found_key.objectid != key.objectid)
1353 chunk = btrfs_item_ptr(path->nodes[0],
1355 struct btrfs_chunk);
1356 key.offset = found_key.offset;
1357 /* chunk zero is special */
1358 if (key.offset == 0)
1361 ret = btrfs_relocate_chunk(chunk_root,
1362 chunk_root->root_key.objectid,
1366 btrfs_release_path(chunk_root, path);
1370 btrfs_free_path(path);
1371 mutex_unlock(&dev_root->fs_info->fs_mutex);
1376 * shrinking a device means finding all of the device extents past
1377 * the new size, and then following the back refs to the chunks.
1378 * The chunk relocation code actually frees the device extent
1380 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1382 struct btrfs_trans_handle *trans;
1383 struct btrfs_root *root = device->dev_root;
1384 struct btrfs_dev_extent *dev_extent = NULL;
1385 struct btrfs_path *path;
1392 struct extent_buffer *l;
1393 struct btrfs_key key;
1394 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1395 u64 old_total = btrfs_super_total_bytes(super_copy);
1396 u64 diff = device->total_bytes - new_size;
1399 path = btrfs_alloc_path();
1403 trans = btrfs_start_transaction(root, 1);
1411 device->total_bytes = new_size;
1412 ret = btrfs_update_device(trans, device);
1414 btrfs_end_transaction(trans, root);
1417 WARN_ON(diff > old_total);
1418 btrfs_set_super_total_bytes(super_copy, old_total - diff);
1419 btrfs_end_transaction(trans, root);
1421 key.objectid = device->devid;
1422 key.offset = (u64)-1;
1423 key.type = BTRFS_DEV_EXTENT_KEY;
1426 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1430 ret = btrfs_previous_item(root, path, 0, key.type);
1439 slot = path->slots[0];
1440 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1442 if (key.objectid != device->devid)
1445 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1446 length = btrfs_dev_extent_length(l, dev_extent);
1448 if (key.offset + length <= new_size)
1451 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1452 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1453 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1454 btrfs_release_path(root, path);
1456 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
1463 btrfs_free_path(path);
1467 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
1468 struct btrfs_root *root,
1469 struct btrfs_key *key,
1470 struct btrfs_chunk *chunk, int item_size)
1472 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1473 struct btrfs_disk_key disk_key;
1477 array_size = btrfs_super_sys_array_size(super_copy);
1478 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
1481 ptr = super_copy->sys_chunk_array + array_size;
1482 btrfs_cpu_key_to_disk(&disk_key, key);
1483 memcpy(ptr, &disk_key, sizeof(disk_key));
1484 ptr += sizeof(disk_key);
1485 memcpy(ptr, chunk, item_size);
1486 item_size += sizeof(disk_key);
1487 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
1491 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
1494 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
1496 else if (type & BTRFS_BLOCK_GROUP_RAID10)
1497 return calc_size * (num_stripes / sub_stripes);
1499 return calc_size * num_stripes;
1503 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
1504 struct btrfs_root *extent_root, u64 *start,
1505 u64 *num_bytes, u64 type)
1508 struct btrfs_fs_info *info = extent_root->fs_info;
1509 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
1510 struct btrfs_path *path;
1511 struct btrfs_stripe *stripes;
1512 struct btrfs_device *device = NULL;
1513 struct btrfs_chunk *chunk;
1514 struct list_head private_devs;
1515 struct list_head *dev_list;
1516 struct list_head *cur;
1517 struct extent_map_tree *em_tree;
1518 struct map_lookup *map;
1519 struct extent_map *em;
1520 int min_stripe_size = 1 * 1024 * 1024;
1522 u64 calc_size = 1024 * 1024 * 1024;
1523 u64 max_chunk_size = calc_size;
1528 int num_stripes = 1;
1529 int min_stripes = 1;
1530 int sub_stripes = 0;
1534 int stripe_len = 64 * 1024;
1535 struct btrfs_key key;
1537 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
1538 (type & BTRFS_BLOCK_GROUP_DUP)) {
1540 type &= ~BTRFS_BLOCK_GROUP_DUP;
1542 dev_list = &extent_root->fs_info->fs_devices->alloc_list;
1543 if (list_empty(dev_list))
1546 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
1547 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1550 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
1554 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
1555 num_stripes = min_t(u64, 2,
1556 extent_root->fs_info->fs_devices->open_devices);
1557 if (num_stripes < 2)
1561 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1562 num_stripes = extent_root->fs_info->fs_devices->open_devices;
1563 if (num_stripes < 4)
1565 num_stripes &= ~(u32)1;
1570 if (type & BTRFS_BLOCK_GROUP_DATA) {
1571 max_chunk_size = 10 * calc_size;
1572 min_stripe_size = 64 * 1024 * 1024;
1573 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
1574 max_chunk_size = 4 * calc_size;
1575 min_stripe_size = 32 * 1024 * 1024;
1576 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1577 calc_size = 8 * 1024 * 1024;
1578 max_chunk_size = calc_size * 2;
1579 min_stripe_size = 1 * 1024 * 1024;
1582 path = btrfs_alloc_path();
1586 /* we don't want a chunk larger than 10% of the FS */
1587 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
1588 max_chunk_size = min(percent_max, max_chunk_size);
1591 if (calc_size * num_stripes > max_chunk_size) {
1592 calc_size = max_chunk_size;
1593 do_div(calc_size, num_stripes);
1594 do_div(calc_size, stripe_len);
1595 calc_size *= stripe_len;
1597 /* we don't want tiny stripes */
1598 calc_size = max_t(u64, min_stripe_size, calc_size);
1600 do_div(calc_size, stripe_len);
1601 calc_size *= stripe_len;
1603 INIT_LIST_HEAD(&private_devs);
1604 cur = dev_list->next;
1607 if (type & BTRFS_BLOCK_GROUP_DUP)
1608 min_free = calc_size * 2;
1610 min_free = calc_size;
1612 /* we add 1MB because we never use the first 1MB of the device */
1613 min_free += 1024 * 1024;
1615 /* build a private list of devices we will allocate from */
1616 while(index < num_stripes) {
1617 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1619 if (device->total_bytes > device->bytes_used)
1620 avail = device->total_bytes - device->bytes_used;
1625 if (device->in_fs_metadata && avail >= min_free) {
1626 u64 ignored_start = 0;
1627 ret = find_free_dev_extent(trans, device, path,
1631 list_move_tail(&device->dev_alloc_list,
1634 if (type & BTRFS_BLOCK_GROUP_DUP)
1637 } else if (device->in_fs_metadata && avail > max_avail)
1639 if (cur == dev_list)
1642 if (index < num_stripes) {
1643 list_splice(&private_devs, dev_list);
1644 if (index >= min_stripes) {
1645 num_stripes = index;
1646 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
1647 num_stripes /= sub_stripes;
1648 num_stripes *= sub_stripes;
1653 if (!looped && max_avail > 0) {
1655 calc_size = max_avail;
1658 btrfs_free_path(path);
1661 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1662 key.type = BTRFS_CHUNK_ITEM_KEY;
1663 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1666 btrfs_free_path(path);
1670 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1672 btrfs_free_path(path);
1676 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1679 btrfs_free_path(path);
1682 btrfs_free_path(path);
1685 stripes = &chunk->stripe;
1686 *num_bytes = chunk_bytes_by_type(type, calc_size,
1687 num_stripes, sub_stripes);
1690 while(index < num_stripes) {
1691 struct btrfs_stripe *stripe;
1692 BUG_ON(list_empty(&private_devs));
1693 cur = private_devs.next;
1694 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
1696 /* loop over this device again if we're doing a dup group */
1697 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1698 (index == num_stripes - 1))
1699 list_move_tail(&device->dev_alloc_list, dev_list);
1701 ret = btrfs_alloc_dev_extent(trans, device,
1702 info->chunk_root->root_key.objectid,
1703 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1704 calc_size, &dev_offset);
1706 device->bytes_used += calc_size;
1707 ret = btrfs_update_device(trans, device);
1710 map->stripes[index].dev = device;
1711 map->stripes[index].physical = dev_offset;
1712 stripe = stripes + index;
1713 btrfs_set_stack_stripe_devid(stripe, device->devid);
1714 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1715 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1716 physical = dev_offset;
1719 BUG_ON(!list_empty(&private_devs));
1721 /* key was set above */
1722 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1723 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1724 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1725 btrfs_set_stack_chunk_type(chunk, type);
1726 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1727 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1728 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1729 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1730 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1731 map->sector_size = extent_root->sectorsize;
1732 map->stripe_len = stripe_len;
1733 map->io_align = stripe_len;
1734 map->io_width = stripe_len;
1736 map->num_stripes = num_stripes;
1737 map->sub_stripes = sub_stripes;
1739 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1740 btrfs_chunk_item_size(num_stripes));
1742 *start = key.offset;;
1744 em = alloc_extent_map(GFP_NOFS);
1747 em->bdev = (struct block_device *)map;
1748 em->start = key.offset;
1749 em->len = *num_bytes;
1750 em->block_start = 0;
1752 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1753 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1754 chunk, btrfs_chunk_item_size(num_stripes));
1759 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
1760 spin_lock(&em_tree->lock);
1761 ret = add_extent_mapping(em_tree, em);
1762 spin_unlock(&em_tree->lock);
1764 free_extent_map(em);
1768 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1770 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
1773 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
1775 struct extent_map *em;
1778 spin_lock(&tree->map_tree.lock);
1779 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
1781 remove_extent_mapping(&tree->map_tree, em);
1782 spin_unlock(&tree->map_tree.lock);
1787 free_extent_map(em);
1788 /* once for the tree */
1789 free_extent_map(em);
1793 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1795 struct extent_map *em;
1796 struct map_lookup *map;
1797 struct extent_map_tree *em_tree = &map_tree->map_tree;
1800 spin_lock(&em_tree->lock);
1801 em = lookup_extent_mapping(em_tree, logical, len);
1802 spin_unlock(&em_tree->lock);
1805 BUG_ON(em->start > logical || em->start + em->len < logical);
1806 map = (struct map_lookup *)em->bdev;
1807 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1808 ret = map->num_stripes;
1809 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1810 ret = map->sub_stripes;
1813 free_extent_map(em);
1817 static int find_live_mirror(struct map_lookup *map, int first, int num,
1821 if (map->stripes[optimal].dev->bdev)
1823 for (i = first; i < first + num; i++) {
1824 if (map->stripes[i].dev->bdev)
1827 /* we couldn't find one that doesn't fail. Just return something
1828 * and the io error handling code will clean up eventually
1833 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1834 u64 logical, u64 *length,
1835 struct btrfs_multi_bio **multi_ret,
1836 int mirror_num, struct page *unplug_page)
1838 struct extent_map *em;
1839 struct map_lookup *map;
1840 struct extent_map_tree *em_tree = &map_tree->map_tree;
1844 int stripes_allocated = 8;
1845 int stripes_required = 1;
1850 struct btrfs_multi_bio *multi = NULL;
1852 if (multi_ret && !(rw & (1 << BIO_RW))) {
1853 stripes_allocated = 1;
1857 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1862 atomic_set(&multi->error, 0);
1865 spin_lock(&em_tree->lock);
1866 em = lookup_extent_mapping(em_tree, logical, *length);
1867 spin_unlock(&em_tree->lock);
1869 if (!em && unplug_page)
1873 printk("unable to find logical %Lu len %Lu\n", logical, *length);
1877 BUG_ON(em->start > logical || em->start + em->len < logical);
1878 map = (struct map_lookup *)em->bdev;
1879 offset = logical - em->start;
1881 if (mirror_num > map->num_stripes)
1884 /* if our multi bio struct is too small, back off and try again */
1885 if (rw & (1 << BIO_RW)) {
1886 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1887 BTRFS_BLOCK_GROUP_DUP)) {
1888 stripes_required = map->num_stripes;
1890 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1891 stripes_required = map->sub_stripes;
1895 if (multi_ret && rw == WRITE &&
1896 stripes_allocated < stripes_required) {
1897 stripes_allocated = map->num_stripes;
1898 free_extent_map(em);
1904 * stripe_nr counts the total number of stripes we have to stride
1905 * to get to this block
1907 do_div(stripe_nr, map->stripe_len);
1909 stripe_offset = stripe_nr * map->stripe_len;
1910 BUG_ON(offset < stripe_offset);
1912 /* stripe_offset is the offset of this block in its stripe*/
1913 stripe_offset = offset - stripe_offset;
1915 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1916 BTRFS_BLOCK_GROUP_RAID10 |
1917 BTRFS_BLOCK_GROUP_DUP)) {
1918 /* we limit the length of each bio to what fits in a stripe */
1919 *length = min_t(u64, em->len - offset,
1920 map->stripe_len - stripe_offset);
1922 *length = em->len - offset;
1925 if (!multi_ret && !unplug_page)
1930 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1931 if (unplug_page || (rw & (1 << BIO_RW)))
1932 num_stripes = map->num_stripes;
1933 else if (mirror_num)
1934 stripe_index = mirror_num - 1;
1936 stripe_index = find_live_mirror(map, 0,
1938 current->pid % map->num_stripes);
1941 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1942 if (rw & (1 << BIO_RW))
1943 num_stripes = map->num_stripes;
1944 else if (mirror_num)
1945 stripe_index = mirror_num - 1;
1947 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1948 int factor = map->num_stripes / map->sub_stripes;
1950 stripe_index = do_div(stripe_nr, factor);
1951 stripe_index *= map->sub_stripes;
1953 if (unplug_page || (rw & (1 << BIO_RW)))
1954 num_stripes = map->sub_stripes;
1955 else if (mirror_num)
1956 stripe_index += mirror_num - 1;
1958 stripe_index = find_live_mirror(map, stripe_index,
1959 map->sub_stripes, stripe_index +
1960 current->pid % map->sub_stripes);
1964 * after this do_div call, stripe_nr is the number of stripes
1965 * on this device we have to walk to find the data, and
1966 * stripe_index is the number of our device in the stripe array
1968 stripe_index = do_div(stripe_nr, map->num_stripes);
1970 BUG_ON(stripe_index >= map->num_stripes);
1972 for (i = 0; i < num_stripes; i++) {
1974 struct btrfs_device *device;
1975 struct backing_dev_info *bdi;
1977 device = map->stripes[stripe_index].dev;
1979 bdi = blk_get_backing_dev_info(device->bdev);
1980 if (bdi->unplug_io_fn) {
1981 bdi->unplug_io_fn(bdi, unplug_page);
1985 multi->stripes[i].physical =
1986 map->stripes[stripe_index].physical +
1987 stripe_offset + stripe_nr * map->stripe_len;
1988 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1994 multi->num_stripes = num_stripes;
1995 multi->max_errors = max_errors;
1998 free_extent_map(em);
2002 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2003 u64 logical, u64 *length,
2004 struct btrfs_multi_bio **multi_ret, int mirror_num)
2006 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2010 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2011 u64 logical, struct page *page)
2013 u64 length = PAGE_CACHE_SIZE;
2014 return __btrfs_map_block(map_tree, READ, logical, &length,
2019 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2020 static void end_bio_multi_stripe(struct bio *bio, int err)
2022 static int end_bio_multi_stripe(struct bio *bio,
2023 unsigned int bytes_done, int err)
2026 struct btrfs_multi_bio *multi = bio->bi_private;
2028 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2033 atomic_inc(&multi->error);
2035 if (atomic_dec_and_test(&multi->stripes_pending)) {
2036 bio->bi_private = multi->private;
2037 bio->bi_end_io = multi->end_io;
2038 /* only send an error to the higher layers if it is
2039 * beyond the tolerance of the multi-bio
2041 if (atomic_read(&multi->error) > multi->max_errors) {
2045 * this bio is actually up to date, we didn't
2046 * go over the max number of errors
2048 set_bit(BIO_UPTODATE, &bio->bi_flags);
2053 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2054 bio_endio(bio, bio->bi_size, err);
2056 bio_endio(bio, err);
2061 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2066 struct async_sched {
2069 struct btrfs_fs_info *info;
2070 struct btrfs_work work;
2074 * see run_scheduled_bios for a description of why bios are collected for
2077 * This will add one bio to the pending list for a device and make sure
2078 * the work struct is scheduled.
2080 int schedule_bio(struct btrfs_root *root, struct btrfs_device *device,
2081 int rw, struct bio *bio)
2083 int should_queue = 1;
2085 /* don't bother with additional async steps for reads, right now */
2086 if (!(rw & (1 << BIO_RW))) {
2087 submit_bio(rw, bio);
2092 * nr_async_sumbits allows us to reliably return congestion to the
2093 * higher layers. Otherwise, the async bio makes it appear we have
2094 * made progress against dirty pages when we've really just put it
2095 * on a queue for later
2097 atomic_inc(&root->fs_info->nr_async_submits);
2098 bio->bi_next = NULL;
2101 spin_lock(&device->io_lock);
2103 if (device->pending_bio_tail)
2104 device->pending_bio_tail->bi_next = bio;
2106 device->pending_bio_tail = bio;
2107 if (!device->pending_bios)
2108 device->pending_bios = bio;
2109 if (device->running_pending)
2112 spin_unlock(&device->io_lock);
2115 btrfs_queue_worker(&root->fs_info->workers, &device->work);
2119 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2120 int mirror_num, int async_submit)
2122 struct btrfs_mapping_tree *map_tree;
2123 struct btrfs_device *dev;
2124 struct bio *first_bio = bio;
2125 u64 logical = bio->bi_sector << 9;
2128 struct btrfs_multi_bio *multi = NULL;
2133 length = bio->bi_size;
2134 map_tree = &root->fs_info->mapping_tree;
2135 map_length = length;
2137 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2141 total_devs = multi->num_stripes;
2142 if (map_length < length) {
2143 printk("mapping failed logical %Lu bio len %Lu "
2144 "len %Lu\n", logical, length, map_length);
2147 multi->end_io = first_bio->bi_end_io;
2148 multi->private = first_bio->bi_private;
2149 atomic_set(&multi->stripes_pending, multi->num_stripes);
2151 while(dev_nr < total_devs) {
2152 if (total_devs > 1) {
2153 if (dev_nr < total_devs - 1) {
2154 bio = bio_clone(first_bio, GFP_NOFS);
2159 bio->bi_private = multi;
2160 bio->bi_end_io = end_bio_multi_stripe;
2162 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2163 dev = multi->stripes[dev_nr].dev;
2164 if (dev && dev->bdev) {
2165 bio->bi_bdev = dev->bdev;
2167 schedule_bio(root, dev, rw, bio);
2169 submit_bio(rw, bio);
2171 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2172 bio->bi_sector = logical >> 9;
2173 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
2174 bio_endio(bio, bio->bi_size, -EIO);
2176 bio_endio(bio, -EIO);
2181 if (total_devs == 1)
2186 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2189 struct list_head *head = &root->fs_info->fs_devices->devices;
2191 return __find_device(head, devid, uuid);
2194 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2195 u64 devid, u8 *dev_uuid)
2197 struct btrfs_device *device;
2198 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2200 device = kzalloc(sizeof(*device), GFP_NOFS);
2201 list_add(&device->dev_list,
2202 &fs_devices->devices);
2203 list_add(&device->dev_alloc_list,
2204 &fs_devices->alloc_list);
2205 device->barriers = 1;
2206 device->dev_root = root->fs_info->dev_root;
2207 device->devid = devid;
2208 device->work.func = pending_bios_fn;
2209 fs_devices->num_devices++;
2210 spin_lock_init(&device->io_lock);
2211 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
2216 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
2217 struct extent_buffer *leaf,
2218 struct btrfs_chunk *chunk)
2220 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2221 struct map_lookup *map;
2222 struct extent_map *em;
2226 u8 uuid[BTRFS_UUID_SIZE];
2231 logical = key->offset;
2232 length = btrfs_chunk_length(leaf, chunk);
2234 spin_lock(&map_tree->map_tree.lock);
2235 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
2236 spin_unlock(&map_tree->map_tree.lock);
2238 /* already mapped? */
2239 if (em && em->start <= logical && em->start + em->len > logical) {
2240 free_extent_map(em);
2243 free_extent_map(em);
2246 map = kzalloc(sizeof(*map), GFP_NOFS);
2250 em = alloc_extent_map(GFP_NOFS);
2253 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2254 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2256 free_extent_map(em);
2260 em->bdev = (struct block_device *)map;
2261 em->start = logical;
2263 em->block_start = 0;
2265 map->num_stripes = num_stripes;
2266 map->io_width = btrfs_chunk_io_width(leaf, chunk);
2267 map->io_align = btrfs_chunk_io_align(leaf, chunk);
2268 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
2269 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
2270 map->type = btrfs_chunk_type(leaf, chunk);
2271 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
2272 for (i = 0; i < num_stripes; i++) {
2273 map->stripes[i].physical =
2274 btrfs_stripe_offset_nr(leaf, chunk, i);
2275 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
2276 read_extent_buffer(leaf, uuid, (unsigned long)
2277 btrfs_stripe_dev_uuid_nr(chunk, i),
2279 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
2281 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
2283 free_extent_map(em);
2286 if (!map->stripes[i].dev) {
2287 map->stripes[i].dev =
2288 add_missing_dev(root, devid, uuid);
2289 if (!map->stripes[i].dev) {
2291 free_extent_map(em);
2295 map->stripes[i].dev->in_fs_metadata = 1;
2298 spin_lock(&map_tree->map_tree.lock);
2299 ret = add_extent_mapping(&map_tree->map_tree, em);
2300 spin_unlock(&map_tree->map_tree.lock);
2302 free_extent_map(em);
2307 static int fill_device_from_item(struct extent_buffer *leaf,
2308 struct btrfs_dev_item *dev_item,
2309 struct btrfs_device *device)
2313 device->devid = btrfs_device_id(leaf, dev_item);
2314 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2315 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2316 device->type = btrfs_device_type(leaf, dev_item);
2317 device->io_align = btrfs_device_io_align(leaf, dev_item);
2318 device->io_width = btrfs_device_io_width(leaf, dev_item);
2319 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
2321 ptr = (unsigned long)btrfs_device_uuid(dev_item);
2322 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
2327 static int read_one_dev(struct btrfs_root *root,
2328 struct extent_buffer *leaf,
2329 struct btrfs_dev_item *dev_item)
2331 struct btrfs_device *device;
2334 u8 dev_uuid[BTRFS_UUID_SIZE];
2336 devid = btrfs_device_id(leaf, dev_item);
2337 read_extent_buffer(leaf, dev_uuid,
2338 (unsigned long)btrfs_device_uuid(dev_item),
2340 device = btrfs_find_device(root, devid, dev_uuid);
2342 printk("warning devid %Lu missing\n", devid);
2343 device = add_missing_dev(root, devid, dev_uuid);
2348 fill_device_from_item(leaf, dev_item, device);
2349 device->dev_root = root->fs_info->dev_root;
2350 device->in_fs_metadata = 1;
2353 ret = btrfs_open_device(device);
2361 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
2363 struct btrfs_dev_item *dev_item;
2365 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
2367 return read_one_dev(root, buf, dev_item);
2370 int btrfs_read_sys_array(struct btrfs_root *root)
2372 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2373 struct extent_buffer *sb;
2374 struct btrfs_disk_key *disk_key;
2375 struct btrfs_chunk *chunk;
2377 unsigned long sb_ptr;
2383 struct btrfs_key key;
2385 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
2386 BTRFS_SUPER_INFO_SIZE);
2389 btrfs_set_buffer_uptodate(sb);
2390 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
2391 array_size = btrfs_super_sys_array_size(super_copy);
2393 ptr = super_copy->sys_chunk_array;
2394 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
2397 while (cur < array_size) {
2398 disk_key = (struct btrfs_disk_key *)ptr;
2399 btrfs_disk_key_to_cpu(&key, disk_key);
2401 len = sizeof(*disk_key); ptr += len;
2405 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2406 chunk = (struct btrfs_chunk *)sb_ptr;
2407 ret = read_one_chunk(root, &key, sb, chunk);
2410 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
2411 len = btrfs_chunk_item_size(num_stripes);
2420 free_extent_buffer(sb);
2424 int btrfs_read_chunk_tree(struct btrfs_root *root)
2426 struct btrfs_path *path;
2427 struct extent_buffer *leaf;
2428 struct btrfs_key key;
2429 struct btrfs_key found_key;
2433 root = root->fs_info->chunk_root;
2435 path = btrfs_alloc_path();
2439 /* first we search for all of the device items, and then we
2440 * read in all of the chunk items. This way we can create chunk
2441 * mappings that reference all of the devices that are afound
2443 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2447 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2449 leaf = path->nodes[0];
2450 slot = path->slots[0];
2451 if (slot >= btrfs_header_nritems(leaf)) {
2452 ret = btrfs_next_leaf(root, path);
2459 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2460 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2461 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
2463 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2464 struct btrfs_dev_item *dev_item;
2465 dev_item = btrfs_item_ptr(leaf, slot,
2466 struct btrfs_dev_item);
2467 ret = read_one_dev(root, leaf, dev_item);
2470 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2471 struct btrfs_chunk *chunk;
2472 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2473 ret = read_one_chunk(root, &found_key, leaf, chunk);
2477 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
2479 btrfs_release_path(root, path);
2483 btrfs_free_path(path);