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 <linux/iocontext.h>
24 #include <asm/div64.h>
27 #include "extent_map.h"
29 #include "transaction.h"
30 #include "print-tree.h"
32 #include "async-thread.h"
42 struct btrfs_bio_stripe stripes[];
45 static int init_first_rw_device(struct btrfs_trans_handle *trans,
46 struct btrfs_root *root,
47 struct btrfs_device *device);
48 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
53 static DEFINE_MUTEX(uuid_mutex);
54 static LIST_HEAD(fs_uuids);
56 void btrfs_lock_volumes(void)
58 mutex_lock(&uuid_mutex);
61 void btrfs_unlock_volumes(void)
63 mutex_unlock(&uuid_mutex);
66 static void lock_chunks(struct btrfs_root *root)
68 mutex_lock(&root->fs_info->chunk_mutex);
71 static void unlock_chunks(struct btrfs_root *root)
73 mutex_unlock(&root->fs_info->chunk_mutex);
76 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
78 struct btrfs_device *device;
79 WARN_ON(fs_devices->opened);
80 while (!list_empty(&fs_devices->devices)) {
81 device = list_entry(fs_devices->devices.next,
82 struct btrfs_device, dev_list);
83 list_del(&device->dev_list);
90 int btrfs_cleanup_fs_uuids(void)
92 struct btrfs_fs_devices *fs_devices;
94 while (!list_empty(&fs_uuids)) {
95 fs_devices = list_entry(fs_uuids.next,
96 struct btrfs_fs_devices, list);
97 list_del(&fs_devices->list);
98 free_fs_devices(fs_devices);
103 static noinline struct btrfs_device *__find_device(struct list_head *head,
106 struct btrfs_device *dev;
108 list_for_each_entry(dev, head, dev_list) {
109 if (dev->devid == devid &&
110 (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
119 struct btrfs_fs_devices *fs_devices;
121 list_for_each_entry(fs_devices, &fs_uuids, list) {
122 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
128 static void requeue_list(struct btrfs_pending_bios *pending_bios,
129 struct bio *head, struct bio *tail)
132 struct bio *old_head;
134 old_head = pending_bios->head;
135 pending_bios->head = head;
136 if (pending_bios->tail)
137 tail->bi_next = old_head;
139 pending_bios->tail = tail;
143 * we try to collect pending bios for a device so we don't get a large
144 * number of procs sending bios down to the same device. This greatly
145 * improves the schedulers ability to collect and merge the bios.
147 * But, it also turns into a long list of bios to process and that is sure
148 * to eventually make the worker thread block. The solution here is to
149 * make some progress and then put this work struct back at the end of
150 * the list if the block device is congested. This way, multiple devices
151 * can make progress from a single worker thread.
153 static noinline int run_scheduled_bios(struct btrfs_device *device)
156 struct backing_dev_info *bdi;
157 struct btrfs_fs_info *fs_info;
158 struct btrfs_pending_bios *pending_bios;
162 unsigned long num_run;
163 unsigned long num_sync_run;
164 unsigned long batch_run = 0;
166 unsigned long last_waited = 0;
169 bdi = blk_get_backing_dev_info(device->bdev);
170 fs_info = device->dev_root->fs_info;
171 limit = btrfs_async_submit_limit(fs_info);
172 limit = limit * 2 / 3;
174 /* we want to make sure that every time we switch from the sync
175 * list to the normal list, we unplug
180 spin_lock(&device->io_lock);
185 /* take all the bios off the list at once and process them
186 * later on (without the lock held). But, remember the
187 * tail and other pointers so the bios can be properly reinserted
188 * into the list if we hit congestion
190 if (!force_reg && device->pending_sync_bios.head) {
191 pending_bios = &device->pending_sync_bios;
194 pending_bios = &device->pending_bios;
198 pending = pending_bios->head;
199 tail = pending_bios->tail;
200 WARN_ON(pending && !tail);
203 * if pending was null this time around, no bios need processing
204 * at all and we can stop. Otherwise it'll loop back up again
205 * and do an additional check so no bios are missed.
207 * device->running_pending is used to synchronize with the
210 if (device->pending_sync_bios.head == NULL &&
211 device->pending_bios.head == NULL) {
213 device->running_pending = 0;
216 device->running_pending = 1;
219 pending_bios->head = NULL;
220 pending_bios->tail = NULL;
222 spin_unlock(&device->io_lock);
225 * if we're doing the regular priority list, make sure we unplug
226 * for any high prio bios we've sent down
228 if (pending_bios == &device->pending_bios && num_sync_run > 0) {
230 blk_run_backing_dev(bdi, NULL);
236 /* we want to work on both lists, but do more bios on the
237 * sync list than the regular list
240 pending_bios != &device->pending_sync_bios &&
241 device->pending_sync_bios.head) ||
242 (num_run > 64 && pending_bios == &device->pending_sync_bios &&
243 device->pending_bios.head)) {
244 spin_lock(&device->io_lock);
245 requeue_list(pending_bios, pending, tail);
250 pending = pending->bi_next;
252 atomic_dec(&fs_info->nr_async_bios);
254 if (atomic_read(&fs_info->nr_async_bios) < limit &&
255 waitqueue_active(&fs_info->async_submit_wait))
256 wake_up(&fs_info->async_submit_wait);
258 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
259 submit_bio(cur->bi_rw, cur);
263 if (bio_rw_flagged(cur, BIO_RW_SYNCIO))
266 if (need_resched()) {
268 blk_run_backing_dev(bdi, NULL);
275 * we made progress, there is more work to do and the bdi
276 * is now congested. Back off and let other work structs
279 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
280 fs_info->fs_devices->open_devices > 1) {
281 struct io_context *ioc;
283 ioc = current->io_context;
286 * the main goal here is that we don't want to
287 * block if we're going to be able to submit
288 * more requests without blocking.
290 * This code does two great things, it pokes into
291 * the elevator code from a filesystem _and_
292 * it makes assumptions about how batching works.
294 if (ioc && ioc->nr_batch_requests > 0 &&
295 time_before(jiffies, ioc->last_waited + HZ/50UL) &&
297 ioc->last_waited == last_waited)) {
299 * we want to go through our batch of
300 * requests and stop. So, we copy out
301 * the ioc->last_waited time and test
302 * against it before looping
304 last_waited = ioc->last_waited;
305 if (need_resched()) {
307 blk_run_backing_dev(bdi, NULL);
314 spin_lock(&device->io_lock);
315 requeue_list(pending_bios, pending, tail);
316 device->running_pending = 1;
318 spin_unlock(&device->io_lock);
319 btrfs_requeue_work(&device->work);
326 blk_run_backing_dev(bdi, NULL);
333 spin_lock(&device->io_lock);
334 if (device->pending_bios.head || device->pending_sync_bios.head)
336 spin_unlock(&device->io_lock);
339 * IO has already been through a long path to get here. Checksumming,
340 * async helper threads, perhaps compression. We've done a pretty
341 * good job of collecting a batch of IO and should just unplug
342 * the device right away.
344 * This will help anyone who is waiting on the IO, they might have
345 * already unplugged, but managed to do so before the bio they
346 * cared about found its way down here.
348 blk_run_backing_dev(bdi, NULL);
353 static void pending_bios_fn(struct btrfs_work *work)
355 struct btrfs_device *device;
357 device = container_of(work, struct btrfs_device, work);
358 run_scheduled_bios(device);
361 static noinline int device_list_add(const char *path,
362 struct btrfs_super_block *disk_super,
363 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
365 struct btrfs_device *device;
366 struct btrfs_fs_devices *fs_devices;
367 u64 found_transid = btrfs_super_generation(disk_super);
369 fs_devices = find_fsid(disk_super->fsid);
371 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
374 INIT_LIST_HEAD(&fs_devices->devices);
375 INIT_LIST_HEAD(&fs_devices->alloc_list);
376 list_add(&fs_devices->list, &fs_uuids);
377 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
378 fs_devices->latest_devid = devid;
379 fs_devices->latest_trans = found_transid;
380 mutex_init(&fs_devices->device_list_mutex);
383 device = __find_device(&fs_devices->devices, devid,
384 disk_super->dev_item.uuid);
387 if (fs_devices->opened)
390 device = kzalloc(sizeof(*device), GFP_NOFS);
392 /* we can safely leave the fs_devices entry around */
395 device->devid = devid;
396 device->work.func = pending_bios_fn;
397 memcpy(device->uuid, disk_super->dev_item.uuid,
399 device->barriers = 1;
400 spin_lock_init(&device->io_lock);
401 device->name = kstrdup(path, GFP_NOFS);
406 INIT_LIST_HEAD(&device->dev_alloc_list);
408 mutex_lock(&fs_devices->device_list_mutex);
409 list_add(&device->dev_list, &fs_devices->devices);
410 mutex_unlock(&fs_devices->device_list_mutex);
412 device->fs_devices = fs_devices;
413 fs_devices->num_devices++;
416 if (found_transid > fs_devices->latest_trans) {
417 fs_devices->latest_devid = devid;
418 fs_devices->latest_trans = found_transid;
420 *fs_devices_ret = fs_devices;
424 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
426 struct btrfs_fs_devices *fs_devices;
427 struct btrfs_device *device;
428 struct btrfs_device *orig_dev;
430 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
432 return ERR_PTR(-ENOMEM);
434 INIT_LIST_HEAD(&fs_devices->devices);
435 INIT_LIST_HEAD(&fs_devices->alloc_list);
436 INIT_LIST_HEAD(&fs_devices->list);
437 mutex_init(&fs_devices->device_list_mutex);
438 fs_devices->latest_devid = orig->latest_devid;
439 fs_devices->latest_trans = orig->latest_trans;
440 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
442 mutex_lock(&orig->device_list_mutex);
443 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
444 device = kzalloc(sizeof(*device), GFP_NOFS);
448 device->name = kstrdup(orig_dev->name, GFP_NOFS);
454 device->devid = orig_dev->devid;
455 device->work.func = pending_bios_fn;
456 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
457 device->barriers = 1;
458 spin_lock_init(&device->io_lock);
459 INIT_LIST_HEAD(&device->dev_list);
460 INIT_LIST_HEAD(&device->dev_alloc_list);
462 list_add(&device->dev_list, &fs_devices->devices);
463 device->fs_devices = fs_devices;
464 fs_devices->num_devices++;
466 mutex_unlock(&orig->device_list_mutex);
469 mutex_unlock(&orig->device_list_mutex);
470 free_fs_devices(fs_devices);
471 return ERR_PTR(-ENOMEM);
474 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
476 struct btrfs_device *device, *next;
478 mutex_lock(&uuid_mutex);
480 mutex_lock(&fs_devices->device_list_mutex);
481 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
482 if (device->in_fs_metadata)
486 close_bdev_exclusive(device->bdev, device->mode);
488 fs_devices->open_devices--;
490 if (device->writeable) {
491 list_del_init(&device->dev_alloc_list);
492 device->writeable = 0;
493 fs_devices->rw_devices--;
495 list_del_init(&device->dev_list);
496 fs_devices->num_devices--;
500 mutex_unlock(&fs_devices->device_list_mutex);
502 if (fs_devices->seed) {
503 fs_devices = fs_devices->seed;
507 mutex_unlock(&uuid_mutex);
511 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
513 struct btrfs_device *device;
515 if (--fs_devices->opened > 0)
518 list_for_each_entry(device, &fs_devices->devices, dev_list) {
520 close_bdev_exclusive(device->bdev, device->mode);
521 fs_devices->open_devices--;
523 if (device->writeable) {
524 list_del_init(&device->dev_alloc_list);
525 fs_devices->rw_devices--;
529 device->writeable = 0;
530 device->in_fs_metadata = 0;
532 WARN_ON(fs_devices->open_devices);
533 WARN_ON(fs_devices->rw_devices);
534 fs_devices->opened = 0;
535 fs_devices->seeding = 0;
540 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
542 struct btrfs_fs_devices *seed_devices = NULL;
545 mutex_lock(&uuid_mutex);
546 ret = __btrfs_close_devices(fs_devices);
547 if (!fs_devices->opened) {
548 seed_devices = fs_devices->seed;
549 fs_devices->seed = NULL;
551 mutex_unlock(&uuid_mutex);
553 while (seed_devices) {
554 fs_devices = seed_devices;
555 seed_devices = fs_devices->seed;
556 __btrfs_close_devices(fs_devices);
557 free_fs_devices(fs_devices);
562 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
563 fmode_t flags, void *holder)
565 struct block_device *bdev;
566 struct list_head *head = &fs_devices->devices;
567 struct btrfs_device *device;
568 struct block_device *latest_bdev = NULL;
569 struct buffer_head *bh;
570 struct btrfs_super_block *disk_super;
571 u64 latest_devid = 0;
572 u64 latest_transid = 0;
577 list_for_each_entry(device, head, dev_list) {
583 bdev = open_bdev_exclusive(device->name, flags, holder);
585 printk(KERN_INFO "open %s failed\n", device->name);
588 set_blocksize(bdev, 4096);
590 bh = btrfs_read_dev_super(bdev);
594 disk_super = (struct btrfs_super_block *)bh->b_data;
595 devid = le64_to_cpu(disk_super->dev_item.devid);
596 if (devid != device->devid)
599 if (memcmp(device->uuid, disk_super->dev_item.uuid,
603 device->generation = btrfs_super_generation(disk_super);
604 if (!latest_transid || device->generation > latest_transid) {
605 latest_devid = devid;
606 latest_transid = device->generation;
610 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
611 device->writeable = 0;
613 device->writeable = !bdev_read_only(bdev);
618 device->in_fs_metadata = 0;
619 device->mode = flags;
621 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
622 fs_devices->rotating = 1;
624 fs_devices->open_devices++;
625 if (device->writeable) {
626 fs_devices->rw_devices++;
627 list_add(&device->dev_alloc_list,
628 &fs_devices->alloc_list);
635 close_bdev_exclusive(bdev, FMODE_READ);
639 if (fs_devices->open_devices == 0) {
643 fs_devices->seeding = seeding;
644 fs_devices->opened = 1;
645 fs_devices->latest_bdev = latest_bdev;
646 fs_devices->latest_devid = latest_devid;
647 fs_devices->latest_trans = latest_transid;
648 fs_devices->total_rw_bytes = 0;
653 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
654 fmode_t flags, void *holder)
658 mutex_lock(&uuid_mutex);
659 if (fs_devices->opened) {
660 fs_devices->opened++;
663 ret = __btrfs_open_devices(fs_devices, flags, holder);
665 mutex_unlock(&uuid_mutex);
669 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
670 struct btrfs_fs_devices **fs_devices_ret)
672 struct btrfs_super_block *disk_super;
673 struct block_device *bdev;
674 struct buffer_head *bh;
679 mutex_lock(&uuid_mutex);
681 bdev = open_bdev_exclusive(path, flags, holder);
688 ret = set_blocksize(bdev, 4096);
691 bh = btrfs_read_dev_super(bdev);
696 disk_super = (struct btrfs_super_block *)bh->b_data;
697 devid = le64_to_cpu(disk_super->dev_item.devid);
698 transid = btrfs_super_generation(disk_super);
699 if (disk_super->label[0])
700 printk(KERN_INFO "device label %s ", disk_super->label);
702 /* FIXME, make a readl uuid parser */
703 printk(KERN_INFO "device fsid %llx-%llx ",
704 *(unsigned long long *)disk_super->fsid,
705 *(unsigned long long *)(disk_super->fsid + 8));
707 printk(KERN_CONT "devid %llu transid %llu %s\n",
708 (unsigned long long)devid, (unsigned long long)transid, path);
709 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
713 close_bdev_exclusive(bdev, flags);
715 mutex_unlock(&uuid_mutex);
720 * this uses a pretty simple search, the expectation is that it is
721 * called very infrequently and that a given device has a small number
724 int find_free_dev_extent(struct btrfs_trans_handle *trans,
725 struct btrfs_device *device, u64 num_bytes,
726 u64 *start, u64 *max_avail)
728 struct btrfs_key key;
729 struct btrfs_root *root = device->dev_root;
730 struct btrfs_dev_extent *dev_extent = NULL;
731 struct btrfs_path *path;
734 u64 search_start = 0;
735 u64 search_end = device->total_bytes;
739 struct extent_buffer *l;
741 path = btrfs_alloc_path();
747 /* FIXME use last free of some kind */
749 /* we don't want to overwrite the superblock on the drive,
750 * so we make sure to start at an offset of at least 1MB
752 search_start = max((u64)1024 * 1024, search_start);
754 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
755 search_start = max(root->fs_info->alloc_start, search_start);
757 key.objectid = device->devid;
758 key.offset = search_start;
759 key.type = BTRFS_DEV_EXTENT_KEY;
760 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
764 ret = btrfs_previous_item(root, path, key.objectid, key.type);
771 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
774 slot = path->slots[0];
775 if (slot >= btrfs_header_nritems(l)) {
776 ret = btrfs_next_leaf(root, path);
783 if (search_start >= search_end) {
787 *start = search_start;
791 *start = last_byte > search_start ?
792 last_byte : search_start;
793 if (search_end <= *start) {
799 btrfs_item_key_to_cpu(l, &key, slot);
801 if (key.objectid < device->devid)
804 if (key.objectid > device->devid)
807 if (key.offset >= search_start && key.offset > last_byte &&
809 if (last_byte < search_start)
810 last_byte = search_start;
811 hole_size = key.offset - last_byte;
813 if (hole_size > *max_avail)
814 *max_avail = hole_size;
816 if (key.offset > last_byte &&
817 hole_size >= num_bytes) {
822 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
826 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
827 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
833 /* we have to make sure we didn't find an extent that has already
834 * been allocated by the map tree or the original allocation
836 BUG_ON(*start < search_start);
838 if (*start + num_bytes > search_end) {
842 /* check for pending inserts here */
846 btrfs_free_path(path);
850 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
851 struct btrfs_device *device,
855 struct btrfs_path *path;
856 struct btrfs_root *root = device->dev_root;
857 struct btrfs_key key;
858 struct btrfs_key found_key;
859 struct extent_buffer *leaf = NULL;
860 struct btrfs_dev_extent *extent = NULL;
862 path = btrfs_alloc_path();
866 key.objectid = device->devid;
868 key.type = BTRFS_DEV_EXTENT_KEY;
870 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
872 ret = btrfs_previous_item(root, path, key.objectid,
873 BTRFS_DEV_EXTENT_KEY);
875 leaf = path->nodes[0];
876 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
877 extent = btrfs_item_ptr(leaf, path->slots[0],
878 struct btrfs_dev_extent);
879 BUG_ON(found_key.offset > start || found_key.offset +
880 btrfs_dev_extent_length(leaf, extent) < start);
882 } else if (ret == 0) {
883 leaf = path->nodes[0];
884 extent = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_dev_extent);
889 if (device->bytes_used > 0)
890 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
891 ret = btrfs_del_item(trans, root, path);
894 btrfs_free_path(path);
898 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
899 struct btrfs_device *device,
900 u64 chunk_tree, u64 chunk_objectid,
901 u64 chunk_offset, u64 start, u64 num_bytes)
904 struct btrfs_path *path;
905 struct btrfs_root *root = device->dev_root;
906 struct btrfs_dev_extent *extent;
907 struct extent_buffer *leaf;
908 struct btrfs_key key;
910 WARN_ON(!device->in_fs_metadata);
911 path = btrfs_alloc_path();
915 key.objectid = device->devid;
917 key.type = BTRFS_DEV_EXTENT_KEY;
918 ret = btrfs_insert_empty_item(trans, root, path, &key,
922 leaf = path->nodes[0];
923 extent = btrfs_item_ptr(leaf, path->slots[0],
924 struct btrfs_dev_extent);
925 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
926 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
927 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
929 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
930 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
933 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
934 btrfs_mark_buffer_dirty(leaf);
935 btrfs_free_path(path);
939 static noinline int find_next_chunk(struct btrfs_root *root,
940 u64 objectid, u64 *offset)
942 struct btrfs_path *path;
944 struct btrfs_key key;
945 struct btrfs_chunk *chunk;
946 struct btrfs_key found_key;
948 path = btrfs_alloc_path();
951 key.objectid = objectid;
952 key.offset = (u64)-1;
953 key.type = BTRFS_CHUNK_ITEM_KEY;
955 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
961 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
965 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
967 if (found_key.objectid != objectid)
970 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
972 *offset = found_key.offset +
973 btrfs_chunk_length(path->nodes[0], chunk);
978 btrfs_free_path(path);
982 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
985 struct btrfs_key key;
986 struct btrfs_key found_key;
987 struct btrfs_path *path;
989 root = root->fs_info->chunk_root;
991 path = btrfs_alloc_path();
995 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
996 key.type = BTRFS_DEV_ITEM_KEY;
997 key.offset = (u64)-1;
999 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1005 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1006 BTRFS_DEV_ITEM_KEY);
1010 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1012 *objectid = found_key.offset + 1;
1016 btrfs_free_path(path);
1021 * the device information is stored in the chunk root
1022 * the btrfs_device struct should be fully filled in
1024 int btrfs_add_device(struct btrfs_trans_handle *trans,
1025 struct btrfs_root *root,
1026 struct btrfs_device *device)
1029 struct btrfs_path *path;
1030 struct btrfs_dev_item *dev_item;
1031 struct extent_buffer *leaf;
1032 struct btrfs_key key;
1035 root = root->fs_info->chunk_root;
1037 path = btrfs_alloc_path();
1041 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1042 key.type = BTRFS_DEV_ITEM_KEY;
1043 key.offset = device->devid;
1045 ret = btrfs_insert_empty_item(trans, root, path, &key,
1050 leaf = path->nodes[0];
1051 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1053 btrfs_set_device_id(leaf, dev_item, device->devid);
1054 btrfs_set_device_generation(leaf, dev_item, 0);
1055 btrfs_set_device_type(leaf, dev_item, device->type);
1056 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1057 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1058 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1059 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1060 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1061 btrfs_set_device_group(leaf, dev_item, 0);
1062 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1063 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1064 btrfs_set_device_start_offset(leaf, dev_item, 0);
1066 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1067 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1068 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1069 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1070 btrfs_mark_buffer_dirty(leaf);
1074 btrfs_free_path(path);
1078 static int btrfs_rm_dev_item(struct btrfs_root *root,
1079 struct btrfs_device *device)
1082 struct btrfs_path *path;
1083 struct btrfs_key key;
1084 struct btrfs_trans_handle *trans;
1086 root = root->fs_info->chunk_root;
1088 path = btrfs_alloc_path();
1092 trans = btrfs_start_transaction(root, 1);
1093 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1094 key.type = BTRFS_DEV_ITEM_KEY;
1095 key.offset = device->devid;
1098 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1107 ret = btrfs_del_item(trans, root, path);
1111 btrfs_free_path(path);
1112 unlock_chunks(root);
1113 btrfs_commit_transaction(trans, root);
1117 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1119 struct btrfs_device *device;
1120 struct btrfs_device *next_device;
1121 struct block_device *bdev;
1122 struct buffer_head *bh = NULL;
1123 struct btrfs_super_block *disk_super;
1130 mutex_lock(&uuid_mutex);
1131 mutex_lock(&root->fs_info->volume_mutex);
1133 all_avail = root->fs_info->avail_data_alloc_bits |
1134 root->fs_info->avail_system_alloc_bits |
1135 root->fs_info->avail_metadata_alloc_bits;
1137 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1138 root->fs_info->fs_devices->rw_devices <= 4) {
1139 printk(KERN_ERR "btrfs: unable to go below four devices "
1145 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1146 root->fs_info->fs_devices->rw_devices <= 2) {
1147 printk(KERN_ERR "btrfs: unable to go below two "
1148 "devices on raid1\n");
1153 if (strcmp(device_path, "missing") == 0) {
1154 struct list_head *devices;
1155 struct btrfs_device *tmp;
1158 devices = &root->fs_info->fs_devices->devices;
1159 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1160 list_for_each_entry(tmp, devices, dev_list) {
1161 if (tmp->in_fs_metadata && !tmp->bdev) {
1166 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1171 printk(KERN_ERR "btrfs: no missing devices found to "
1176 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1177 root->fs_info->bdev_holder);
1179 ret = PTR_ERR(bdev);
1183 set_blocksize(bdev, 4096);
1184 bh = btrfs_read_dev_super(bdev);
1189 disk_super = (struct btrfs_super_block *)bh->b_data;
1190 devid = le64_to_cpu(disk_super->dev_item.devid);
1191 dev_uuid = disk_super->dev_item.uuid;
1192 device = btrfs_find_device(root, devid, dev_uuid,
1200 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1201 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1207 if (device->writeable) {
1208 list_del_init(&device->dev_alloc_list);
1209 root->fs_info->fs_devices->rw_devices--;
1212 ret = btrfs_shrink_device(device, 0);
1216 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1220 device->in_fs_metadata = 0;
1223 * the device list mutex makes sure that we don't change
1224 * the device list while someone else is writing out all
1225 * the device supers.
1227 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1228 list_del_init(&device->dev_list);
1229 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1231 device->fs_devices->num_devices--;
1233 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1234 struct btrfs_device, dev_list);
1235 if (device->bdev == root->fs_info->sb->s_bdev)
1236 root->fs_info->sb->s_bdev = next_device->bdev;
1237 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1238 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1241 close_bdev_exclusive(device->bdev, device->mode);
1242 device->bdev = NULL;
1243 device->fs_devices->open_devices--;
1246 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1247 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1249 if (device->fs_devices->open_devices == 0) {
1250 struct btrfs_fs_devices *fs_devices;
1251 fs_devices = root->fs_info->fs_devices;
1252 while (fs_devices) {
1253 if (fs_devices->seed == device->fs_devices)
1255 fs_devices = fs_devices->seed;
1257 fs_devices->seed = device->fs_devices->seed;
1258 device->fs_devices->seed = NULL;
1259 __btrfs_close_devices(device->fs_devices);
1260 free_fs_devices(device->fs_devices);
1264 * at this point, the device is zero sized. We want to
1265 * remove it from the devices list and zero out the old super
1267 if (device->writeable) {
1268 /* make sure this device isn't detected as part of
1271 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1272 set_buffer_dirty(bh);
1273 sync_dirty_buffer(bh);
1276 kfree(device->name);
1284 close_bdev_exclusive(bdev, FMODE_READ);
1286 mutex_unlock(&root->fs_info->volume_mutex);
1287 mutex_unlock(&uuid_mutex);
1292 * does all the dirty work required for changing file system's UUID.
1294 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1295 struct btrfs_root *root)
1297 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1298 struct btrfs_fs_devices *old_devices;
1299 struct btrfs_fs_devices *seed_devices;
1300 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1301 struct btrfs_device *device;
1304 BUG_ON(!mutex_is_locked(&uuid_mutex));
1305 if (!fs_devices->seeding)
1308 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1312 old_devices = clone_fs_devices(fs_devices);
1313 if (IS_ERR(old_devices)) {
1314 kfree(seed_devices);
1315 return PTR_ERR(old_devices);
1318 list_add(&old_devices->list, &fs_uuids);
1320 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1321 seed_devices->opened = 1;
1322 INIT_LIST_HEAD(&seed_devices->devices);
1323 INIT_LIST_HEAD(&seed_devices->alloc_list);
1324 mutex_init(&seed_devices->device_list_mutex);
1325 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1326 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1327 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1328 device->fs_devices = seed_devices;
1331 fs_devices->seeding = 0;
1332 fs_devices->num_devices = 0;
1333 fs_devices->open_devices = 0;
1334 fs_devices->seed = seed_devices;
1336 generate_random_uuid(fs_devices->fsid);
1337 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1338 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1339 super_flags = btrfs_super_flags(disk_super) &
1340 ~BTRFS_SUPER_FLAG_SEEDING;
1341 btrfs_set_super_flags(disk_super, super_flags);
1347 * strore the expected generation for seed devices in device items.
1349 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1350 struct btrfs_root *root)
1352 struct btrfs_path *path;
1353 struct extent_buffer *leaf;
1354 struct btrfs_dev_item *dev_item;
1355 struct btrfs_device *device;
1356 struct btrfs_key key;
1357 u8 fs_uuid[BTRFS_UUID_SIZE];
1358 u8 dev_uuid[BTRFS_UUID_SIZE];
1362 path = btrfs_alloc_path();
1366 root = root->fs_info->chunk_root;
1367 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1369 key.type = BTRFS_DEV_ITEM_KEY;
1372 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1376 leaf = path->nodes[0];
1378 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1379 ret = btrfs_next_leaf(root, path);
1384 leaf = path->nodes[0];
1385 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1386 btrfs_release_path(root, path);
1390 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1391 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1392 key.type != BTRFS_DEV_ITEM_KEY)
1395 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1396 struct btrfs_dev_item);
1397 devid = btrfs_device_id(leaf, dev_item);
1398 read_extent_buffer(leaf, dev_uuid,
1399 (unsigned long)btrfs_device_uuid(dev_item),
1401 read_extent_buffer(leaf, fs_uuid,
1402 (unsigned long)btrfs_device_fsid(dev_item),
1404 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1407 if (device->fs_devices->seeding) {
1408 btrfs_set_device_generation(leaf, dev_item,
1409 device->generation);
1410 btrfs_mark_buffer_dirty(leaf);
1418 btrfs_free_path(path);
1422 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1424 struct btrfs_trans_handle *trans;
1425 struct btrfs_device *device;
1426 struct block_device *bdev;
1427 struct list_head *devices;
1428 struct super_block *sb = root->fs_info->sb;
1430 int seeding_dev = 0;
1433 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1436 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1440 if (root->fs_info->fs_devices->seeding) {
1442 down_write(&sb->s_umount);
1443 mutex_lock(&uuid_mutex);
1446 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1447 mutex_lock(&root->fs_info->volume_mutex);
1449 devices = &root->fs_info->fs_devices->devices;
1451 * we have the volume lock, so we don't need the extra
1452 * device list mutex while reading the list here.
1454 list_for_each_entry(device, devices, dev_list) {
1455 if (device->bdev == bdev) {
1461 device = kzalloc(sizeof(*device), GFP_NOFS);
1463 /* we can safely leave the fs_devices entry around */
1468 device->name = kstrdup(device_path, GFP_NOFS);
1469 if (!device->name) {
1475 ret = find_next_devid(root, &device->devid);
1481 trans = btrfs_start_transaction(root, 1);
1484 device->barriers = 1;
1485 device->writeable = 1;
1486 device->work.func = pending_bios_fn;
1487 generate_random_uuid(device->uuid);
1488 spin_lock_init(&device->io_lock);
1489 device->generation = trans->transid;
1490 device->io_width = root->sectorsize;
1491 device->io_align = root->sectorsize;
1492 device->sector_size = root->sectorsize;
1493 device->total_bytes = i_size_read(bdev->bd_inode);
1494 device->disk_total_bytes = device->total_bytes;
1495 device->dev_root = root->fs_info->dev_root;
1496 device->bdev = bdev;
1497 device->in_fs_metadata = 1;
1499 set_blocksize(device->bdev, 4096);
1502 sb->s_flags &= ~MS_RDONLY;
1503 ret = btrfs_prepare_sprout(trans, root);
1507 device->fs_devices = root->fs_info->fs_devices;
1510 * we don't want write_supers to jump in here with our device
1513 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1514 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1515 list_add(&device->dev_alloc_list,
1516 &root->fs_info->fs_devices->alloc_list);
1517 root->fs_info->fs_devices->num_devices++;
1518 root->fs_info->fs_devices->open_devices++;
1519 root->fs_info->fs_devices->rw_devices++;
1520 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1522 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1523 root->fs_info->fs_devices->rotating = 1;
1525 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1526 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1527 total_bytes + device->total_bytes);
1529 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1530 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1532 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1535 ret = init_first_rw_device(trans, root, device);
1537 ret = btrfs_finish_sprout(trans, root);
1540 ret = btrfs_add_device(trans, root, device);
1544 * we've got more storage, clear any full flags on the space
1547 btrfs_clear_space_info_full(root->fs_info);
1549 unlock_chunks(root);
1550 btrfs_commit_transaction(trans, root);
1553 mutex_unlock(&uuid_mutex);
1554 up_write(&sb->s_umount);
1556 ret = btrfs_relocate_sys_chunks(root);
1560 mutex_unlock(&root->fs_info->volume_mutex);
1563 close_bdev_exclusive(bdev, 0);
1565 mutex_unlock(&uuid_mutex);
1566 up_write(&sb->s_umount);
1571 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1572 struct btrfs_device *device)
1575 struct btrfs_path *path;
1576 struct btrfs_root *root;
1577 struct btrfs_dev_item *dev_item;
1578 struct extent_buffer *leaf;
1579 struct btrfs_key key;
1581 root = device->dev_root->fs_info->chunk_root;
1583 path = btrfs_alloc_path();
1587 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1588 key.type = BTRFS_DEV_ITEM_KEY;
1589 key.offset = device->devid;
1591 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1600 leaf = path->nodes[0];
1601 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1603 btrfs_set_device_id(leaf, dev_item, device->devid);
1604 btrfs_set_device_type(leaf, dev_item, device->type);
1605 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1606 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1607 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1608 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1609 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1610 btrfs_mark_buffer_dirty(leaf);
1613 btrfs_free_path(path);
1617 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1618 struct btrfs_device *device, u64 new_size)
1620 struct btrfs_super_block *super_copy =
1621 &device->dev_root->fs_info->super_copy;
1622 u64 old_total = btrfs_super_total_bytes(super_copy);
1623 u64 diff = new_size - device->total_bytes;
1625 if (!device->writeable)
1627 if (new_size <= device->total_bytes)
1630 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1631 device->fs_devices->total_rw_bytes += diff;
1633 device->total_bytes = new_size;
1634 device->disk_total_bytes = new_size;
1635 btrfs_clear_space_info_full(device->dev_root->fs_info);
1637 return btrfs_update_device(trans, device);
1640 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1641 struct btrfs_device *device, u64 new_size)
1644 lock_chunks(device->dev_root);
1645 ret = __btrfs_grow_device(trans, device, new_size);
1646 unlock_chunks(device->dev_root);
1650 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1651 struct btrfs_root *root,
1652 u64 chunk_tree, u64 chunk_objectid,
1656 struct btrfs_path *path;
1657 struct btrfs_key key;
1659 root = root->fs_info->chunk_root;
1660 path = btrfs_alloc_path();
1664 key.objectid = chunk_objectid;
1665 key.offset = chunk_offset;
1666 key.type = BTRFS_CHUNK_ITEM_KEY;
1668 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1671 ret = btrfs_del_item(trans, root, path);
1674 btrfs_free_path(path);
1678 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1681 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1682 struct btrfs_disk_key *disk_key;
1683 struct btrfs_chunk *chunk;
1690 struct btrfs_key key;
1692 array_size = btrfs_super_sys_array_size(super_copy);
1694 ptr = super_copy->sys_chunk_array;
1697 while (cur < array_size) {
1698 disk_key = (struct btrfs_disk_key *)ptr;
1699 btrfs_disk_key_to_cpu(&key, disk_key);
1701 len = sizeof(*disk_key);
1703 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1704 chunk = (struct btrfs_chunk *)(ptr + len);
1705 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1706 len += btrfs_chunk_item_size(num_stripes);
1711 if (key.objectid == chunk_objectid &&
1712 key.offset == chunk_offset) {
1713 memmove(ptr, ptr + len, array_size - (cur + len));
1715 btrfs_set_super_sys_array_size(super_copy, array_size);
1724 static int btrfs_relocate_chunk(struct btrfs_root *root,
1725 u64 chunk_tree, u64 chunk_objectid,
1728 struct extent_map_tree *em_tree;
1729 struct btrfs_root *extent_root;
1730 struct btrfs_trans_handle *trans;
1731 struct extent_map *em;
1732 struct map_lookup *map;
1736 root = root->fs_info->chunk_root;
1737 extent_root = root->fs_info->extent_root;
1738 em_tree = &root->fs_info->mapping_tree.map_tree;
1740 ret = btrfs_can_relocate(extent_root, chunk_offset);
1744 /* step one, relocate all the extents inside this chunk */
1745 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1748 trans = btrfs_start_transaction(root, 1);
1754 * step two, delete the device extents and the
1755 * chunk tree entries
1757 read_lock(&em_tree->lock);
1758 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1759 read_unlock(&em_tree->lock);
1761 BUG_ON(em->start > chunk_offset ||
1762 em->start + em->len < chunk_offset);
1763 map = (struct map_lookup *)em->bdev;
1765 for (i = 0; i < map->num_stripes; i++) {
1766 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1767 map->stripes[i].physical);
1770 if (map->stripes[i].dev) {
1771 ret = btrfs_update_device(trans, map->stripes[i].dev);
1775 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1780 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1781 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1785 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1788 write_lock(&em_tree->lock);
1789 remove_extent_mapping(em_tree, em);
1790 write_unlock(&em_tree->lock);
1795 /* once for the tree */
1796 free_extent_map(em);
1798 free_extent_map(em);
1800 unlock_chunks(root);
1801 btrfs_end_transaction(trans, root);
1805 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1807 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1808 struct btrfs_path *path;
1809 struct extent_buffer *leaf;
1810 struct btrfs_chunk *chunk;
1811 struct btrfs_key key;
1812 struct btrfs_key found_key;
1813 u64 chunk_tree = chunk_root->root_key.objectid;
1815 bool retried = false;
1819 path = btrfs_alloc_path();
1824 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1825 key.offset = (u64)-1;
1826 key.type = BTRFS_CHUNK_ITEM_KEY;
1829 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1834 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1841 leaf = path->nodes[0];
1842 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1844 chunk = btrfs_item_ptr(leaf, path->slots[0],
1845 struct btrfs_chunk);
1846 chunk_type = btrfs_chunk_type(leaf, chunk);
1847 btrfs_release_path(chunk_root, path);
1849 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1850 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1859 if (found_key.offset == 0)
1861 key.offset = found_key.offset - 1;
1864 if (failed && !retried) {
1868 } else if (failed && retried) {
1873 btrfs_free_path(path);
1877 static u64 div_factor(u64 num, int factor)
1886 int btrfs_balance(struct btrfs_root *dev_root)
1889 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1890 struct btrfs_device *device;
1893 struct btrfs_path *path;
1894 struct btrfs_key key;
1895 struct btrfs_chunk *chunk;
1896 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1897 struct btrfs_trans_handle *trans;
1898 struct btrfs_key found_key;
1900 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1903 mutex_lock(&dev_root->fs_info->volume_mutex);
1904 dev_root = dev_root->fs_info->dev_root;
1906 /* step one make some room on all the devices */
1907 list_for_each_entry(device, devices, dev_list) {
1908 old_size = device->total_bytes;
1909 size_to_free = div_factor(old_size, 1);
1910 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1911 if (!device->writeable ||
1912 device->total_bytes - device->bytes_used > size_to_free)
1915 ret = btrfs_shrink_device(device, old_size - size_to_free);
1920 trans = btrfs_start_transaction(dev_root, 1);
1923 ret = btrfs_grow_device(trans, device, old_size);
1926 btrfs_end_transaction(trans, dev_root);
1929 /* step two, relocate all the chunks */
1930 path = btrfs_alloc_path();
1933 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1934 key.offset = (u64)-1;
1935 key.type = BTRFS_CHUNK_ITEM_KEY;
1938 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1943 * this shouldn't happen, it means the last relocate
1949 ret = btrfs_previous_item(chunk_root, path, 0,
1950 BTRFS_CHUNK_ITEM_KEY);
1954 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1956 if (found_key.objectid != key.objectid)
1959 chunk = btrfs_item_ptr(path->nodes[0],
1961 struct btrfs_chunk);
1962 /* chunk zero is special */
1963 if (found_key.offset == 0)
1966 btrfs_release_path(chunk_root, path);
1967 ret = btrfs_relocate_chunk(chunk_root,
1968 chunk_root->root_key.objectid,
1971 BUG_ON(ret && ret != -ENOSPC);
1972 key.offset = found_key.offset - 1;
1976 btrfs_free_path(path);
1977 mutex_unlock(&dev_root->fs_info->volume_mutex);
1982 * shrinking a device means finding all of the device extents past
1983 * the new size, and then following the back refs to the chunks.
1984 * The chunk relocation code actually frees the device extent
1986 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1988 struct btrfs_trans_handle *trans;
1989 struct btrfs_root *root = device->dev_root;
1990 struct btrfs_dev_extent *dev_extent = NULL;
1991 struct btrfs_path *path;
1999 bool retried = false;
2000 struct extent_buffer *l;
2001 struct btrfs_key key;
2002 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2003 u64 old_total = btrfs_super_total_bytes(super_copy);
2004 u64 old_size = device->total_bytes;
2005 u64 diff = device->total_bytes - new_size;
2007 if (new_size >= device->total_bytes)
2010 path = btrfs_alloc_path();
2018 device->total_bytes = new_size;
2019 if (device->writeable)
2020 device->fs_devices->total_rw_bytes -= diff;
2021 unlock_chunks(root);
2024 key.objectid = device->devid;
2025 key.offset = (u64)-1;
2026 key.type = BTRFS_DEV_EXTENT_KEY;
2029 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2033 ret = btrfs_previous_item(root, path, 0, key.type);
2038 btrfs_release_path(root, path);
2043 slot = path->slots[0];
2044 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
2046 if (key.objectid != device->devid) {
2047 btrfs_release_path(root, path);
2051 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
2052 length = btrfs_dev_extent_length(l, dev_extent);
2054 if (key.offset + length <= new_size) {
2055 btrfs_release_path(root, path);
2059 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
2060 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
2061 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
2062 btrfs_release_path(root, path);
2064 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2066 if (ret && ret != -ENOSPC)
2073 if (failed && !retried) {
2077 } else if (failed && retried) {
2081 device->total_bytes = old_size;
2082 if (device->writeable)
2083 device->fs_devices->total_rw_bytes += diff;
2084 unlock_chunks(root);
2088 /* Shrinking succeeded, else we would be at "done". */
2089 trans = btrfs_start_transaction(root, 1);
2096 device->disk_total_bytes = new_size;
2097 /* Now btrfs_update_device() will change the on-disk size. */
2098 ret = btrfs_update_device(trans, device);
2100 unlock_chunks(root);
2101 btrfs_end_transaction(trans, root);
2104 WARN_ON(diff > old_total);
2105 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2106 unlock_chunks(root);
2107 btrfs_end_transaction(trans, root);
2109 btrfs_free_path(path);
2113 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
2115 struct btrfs_key *key,
2116 struct btrfs_chunk *chunk, int item_size)
2118 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2119 struct btrfs_disk_key disk_key;
2123 array_size = btrfs_super_sys_array_size(super_copy);
2124 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2127 ptr = super_copy->sys_chunk_array + array_size;
2128 btrfs_cpu_key_to_disk(&disk_key, key);
2129 memcpy(ptr, &disk_key, sizeof(disk_key));
2130 ptr += sizeof(disk_key);
2131 memcpy(ptr, chunk, item_size);
2132 item_size += sizeof(disk_key);
2133 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2137 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2138 int num_stripes, int sub_stripes)
2140 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2142 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2143 return calc_size * (num_stripes / sub_stripes);
2145 return calc_size * num_stripes;
2148 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *extent_root,
2150 struct map_lookup **map_ret,
2151 u64 *num_bytes, u64 *stripe_size,
2152 u64 start, u64 type)
2154 struct btrfs_fs_info *info = extent_root->fs_info;
2155 struct btrfs_device *device = NULL;
2156 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2157 struct list_head *cur;
2158 struct map_lookup *map = NULL;
2159 struct extent_map_tree *em_tree;
2160 struct extent_map *em;
2161 struct list_head private_devs;
2162 int min_stripe_size = 1 * 1024 * 1024;
2163 u64 calc_size = 1024 * 1024 * 1024;
2164 u64 max_chunk_size = calc_size;
2169 int num_stripes = 1;
2170 int min_stripes = 1;
2171 int sub_stripes = 0;
2175 int stripe_len = 64 * 1024;
2177 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2178 (type & BTRFS_BLOCK_GROUP_DUP)) {
2180 type &= ~BTRFS_BLOCK_GROUP_DUP;
2182 if (list_empty(&fs_devices->alloc_list))
2185 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2186 num_stripes = fs_devices->rw_devices;
2189 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2193 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2194 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2195 if (num_stripes < 2)
2199 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2200 num_stripes = fs_devices->rw_devices;
2201 if (num_stripes < 4)
2203 num_stripes &= ~(u32)1;
2208 if (type & BTRFS_BLOCK_GROUP_DATA) {
2209 max_chunk_size = 10 * calc_size;
2210 min_stripe_size = 64 * 1024 * 1024;
2211 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2212 max_chunk_size = 4 * calc_size;
2213 min_stripe_size = 32 * 1024 * 1024;
2214 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2215 calc_size = 8 * 1024 * 1024;
2216 max_chunk_size = calc_size * 2;
2217 min_stripe_size = 1 * 1024 * 1024;
2220 /* we don't want a chunk larger than 10% of writeable space */
2221 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2226 if (!map || map->num_stripes != num_stripes) {
2228 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2231 map->num_stripes = num_stripes;
2234 if (calc_size * num_stripes > max_chunk_size) {
2235 calc_size = max_chunk_size;
2236 do_div(calc_size, num_stripes);
2237 do_div(calc_size, stripe_len);
2238 calc_size *= stripe_len;
2240 /* we don't want tiny stripes */
2241 calc_size = max_t(u64, min_stripe_size, calc_size);
2243 do_div(calc_size, stripe_len);
2244 calc_size *= stripe_len;
2246 cur = fs_devices->alloc_list.next;
2249 if (type & BTRFS_BLOCK_GROUP_DUP)
2250 min_free = calc_size * 2;
2252 min_free = calc_size;
2255 * we add 1MB because we never use the first 1MB of the device, unless
2256 * we've looped, then we are likely allocating the maximum amount of
2257 * space left already
2260 min_free += 1024 * 1024;
2262 INIT_LIST_HEAD(&private_devs);
2263 while (index < num_stripes) {
2264 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2265 BUG_ON(!device->writeable);
2266 if (device->total_bytes > device->bytes_used)
2267 avail = device->total_bytes - device->bytes_used;
2272 if (device->in_fs_metadata && avail >= min_free) {
2273 ret = find_free_dev_extent(trans, device,
2274 min_free, &dev_offset,
2277 list_move_tail(&device->dev_alloc_list,
2279 map->stripes[index].dev = device;
2280 map->stripes[index].physical = dev_offset;
2282 if (type & BTRFS_BLOCK_GROUP_DUP) {
2283 map->stripes[index].dev = device;
2284 map->stripes[index].physical =
2285 dev_offset + calc_size;
2289 } else if (device->in_fs_metadata && avail > max_avail)
2291 if (cur == &fs_devices->alloc_list)
2294 list_splice(&private_devs, &fs_devices->alloc_list);
2295 if (index < num_stripes) {
2296 if (index >= min_stripes) {
2297 num_stripes = index;
2298 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2299 num_stripes /= sub_stripes;
2300 num_stripes *= sub_stripes;
2305 if (!looped && max_avail > 0) {
2307 calc_size = max_avail;
2313 map->sector_size = extent_root->sectorsize;
2314 map->stripe_len = stripe_len;
2315 map->io_align = stripe_len;
2316 map->io_width = stripe_len;
2318 map->num_stripes = num_stripes;
2319 map->sub_stripes = sub_stripes;
2322 *stripe_size = calc_size;
2323 *num_bytes = chunk_bytes_by_type(type, calc_size,
2324 num_stripes, sub_stripes);
2326 em = alloc_extent_map(GFP_NOFS);
2331 em->bdev = (struct block_device *)map;
2333 em->len = *num_bytes;
2334 em->block_start = 0;
2335 em->block_len = em->len;
2337 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2338 write_lock(&em_tree->lock);
2339 ret = add_extent_mapping(em_tree, em);
2340 write_unlock(&em_tree->lock);
2342 free_extent_map(em);
2344 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2345 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2350 while (index < map->num_stripes) {
2351 device = map->stripes[index].dev;
2352 dev_offset = map->stripes[index].physical;
2354 ret = btrfs_alloc_dev_extent(trans, device,
2355 info->chunk_root->root_key.objectid,
2356 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2357 start, dev_offset, calc_size);
2365 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2366 struct btrfs_root *extent_root,
2367 struct map_lookup *map, u64 chunk_offset,
2368 u64 chunk_size, u64 stripe_size)
2371 struct btrfs_key key;
2372 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2373 struct btrfs_device *device;
2374 struct btrfs_chunk *chunk;
2375 struct btrfs_stripe *stripe;
2376 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2380 chunk = kzalloc(item_size, GFP_NOFS);
2385 while (index < map->num_stripes) {
2386 device = map->stripes[index].dev;
2387 device->bytes_used += stripe_size;
2388 ret = btrfs_update_device(trans, device);
2394 stripe = &chunk->stripe;
2395 while (index < map->num_stripes) {
2396 device = map->stripes[index].dev;
2397 dev_offset = map->stripes[index].physical;
2399 btrfs_set_stack_stripe_devid(stripe, device->devid);
2400 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2401 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2406 btrfs_set_stack_chunk_length(chunk, chunk_size);
2407 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2408 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2409 btrfs_set_stack_chunk_type(chunk, map->type);
2410 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2411 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2412 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2413 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2414 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2416 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2417 key.type = BTRFS_CHUNK_ITEM_KEY;
2418 key.offset = chunk_offset;
2420 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2423 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2424 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2433 * Chunk allocation falls into two parts. The first part does works
2434 * that make the new allocated chunk useable, but not do any operation
2435 * that modifies the chunk tree. The second part does the works that
2436 * require modifying the chunk tree. This division is important for the
2437 * bootstrap process of adding storage to a seed btrfs.
2439 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2440 struct btrfs_root *extent_root, u64 type)
2445 struct map_lookup *map;
2446 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2449 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2454 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2455 &stripe_size, chunk_offset, type);
2459 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2460 chunk_size, stripe_size);
2465 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2466 struct btrfs_root *root,
2467 struct btrfs_device *device)
2470 u64 sys_chunk_offset;
2474 u64 sys_stripe_size;
2476 struct map_lookup *map;
2477 struct map_lookup *sys_map;
2478 struct btrfs_fs_info *fs_info = root->fs_info;
2479 struct btrfs_root *extent_root = fs_info->extent_root;
2482 ret = find_next_chunk(fs_info->chunk_root,
2483 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2486 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2487 (fs_info->metadata_alloc_profile &
2488 fs_info->avail_metadata_alloc_bits);
2489 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2491 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2492 &stripe_size, chunk_offset, alloc_profile);
2495 sys_chunk_offset = chunk_offset + chunk_size;
2497 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2498 (fs_info->system_alloc_profile &
2499 fs_info->avail_system_alloc_bits);
2500 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2502 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2503 &sys_chunk_size, &sys_stripe_size,
2504 sys_chunk_offset, alloc_profile);
2507 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2511 * Modifying chunk tree needs allocating new blocks from both
2512 * system block group and metadata block group. So we only can
2513 * do operations require modifying the chunk tree after both
2514 * block groups were created.
2516 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2517 chunk_size, stripe_size);
2520 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2521 sys_chunk_offset, sys_chunk_size,
2527 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2529 struct extent_map *em;
2530 struct map_lookup *map;
2531 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2535 read_lock(&map_tree->map_tree.lock);
2536 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2537 read_unlock(&map_tree->map_tree.lock);
2541 map = (struct map_lookup *)em->bdev;
2542 for (i = 0; i < map->num_stripes; i++) {
2543 if (!map->stripes[i].dev->writeable) {
2548 free_extent_map(em);
2552 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2554 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2557 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2559 struct extent_map *em;
2562 write_lock(&tree->map_tree.lock);
2563 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2565 remove_extent_mapping(&tree->map_tree, em);
2566 write_unlock(&tree->map_tree.lock);
2571 free_extent_map(em);
2572 /* once for the tree */
2573 free_extent_map(em);
2577 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2579 struct extent_map *em;
2580 struct map_lookup *map;
2581 struct extent_map_tree *em_tree = &map_tree->map_tree;
2584 read_lock(&em_tree->lock);
2585 em = lookup_extent_mapping(em_tree, logical, len);
2586 read_unlock(&em_tree->lock);
2589 BUG_ON(em->start > logical || em->start + em->len < logical);
2590 map = (struct map_lookup *)em->bdev;
2591 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2592 ret = map->num_stripes;
2593 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2594 ret = map->sub_stripes;
2597 free_extent_map(em);
2601 static int find_live_mirror(struct map_lookup *map, int first, int num,
2605 if (map->stripes[optimal].dev->bdev)
2607 for (i = first; i < first + num; i++) {
2608 if (map->stripes[i].dev->bdev)
2611 /* we couldn't find one that doesn't fail. Just return something
2612 * and the io error handling code will clean up eventually
2617 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2618 u64 logical, u64 *length,
2619 struct btrfs_multi_bio **multi_ret,
2620 int mirror_num, struct page *unplug_page)
2622 struct extent_map *em;
2623 struct map_lookup *map;
2624 struct extent_map_tree *em_tree = &map_tree->map_tree;
2628 int stripes_allocated = 8;
2629 int stripes_required = 1;
2634 struct btrfs_multi_bio *multi = NULL;
2636 if (multi_ret && !(rw & (1 << BIO_RW)))
2637 stripes_allocated = 1;
2640 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2645 atomic_set(&multi->error, 0);
2648 read_lock(&em_tree->lock);
2649 em = lookup_extent_mapping(em_tree, logical, *length);
2650 read_unlock(&em_tree->lock);
2652 if (!em && unplug_page)
2656 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2657 (unsigned long long)logical,
2658 (unsigned long long)*length);
2662 BUG_ON(em->start > logical || em->start + em->len < logical);
2663 map = (struct map_lookup *)em->bdev;
2664 offset = logical - em->start;
2666 if (mirror_num > map->num_stripes)
2669 /* if our multi bio struct is too small, back off and try again */
2670 if (rw & (1 << BIO_RW)) {
2671 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2672 BTRFS_BLOCK_GROUP_DUP)) {
2673 stripes_required = map->num_stripes;
2675 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2676 stripes_required = map->sub_stripes;
2680 if (multi_ret && (rw & (1 << BIO_RW)) &&
2681 stripes_allocated < stripes_required) {
2682 stripes_allocated = map->num_stripes;
2683 free_extent_map(em);
2689 * stripe_nr counts the total number of stripes we have to stride
2690 * to get to this block
2692 do_div(stripe_nr, map->stripe_len);
2694 stripe_offset = stripe_nr * map->stripe_len;
2695 BUG_ON(offset < stripe_offset);
2697 /* stripe_offset is the offset of this block in its stripe*/
2698 stripe_offset = offset - stripe_offset;
2700 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2701 BTRFS_BLOCK_GROUP_RAID10 |
2702 BTRFS_BLOCK_GROUP_DUP)) {
2703 /* we limit the length of each bio to what fits in a stripe */
2704 *length = min_t(u64, em->len - offset,
2705 map->stripe_len - stripe_offset);
2707 *length = em->len - offset;
2710 if (!multi_ret && !unplug_page)
2715 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2716 if (unplug_page || (rw & (1 << BIO_RW)))
2717 num_stripes = map->num_stripes;
2718 else if (mirror_num)
2719 stripe_index = mirror_num - 1;
2721 stripe_index = find_live_mirror(map, 0,
2723 current->pid % map->num_stripes);
2726 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2727 if (rw & (1 << BIO_RW))
2728 num_stripes = map->num_stripes;
2729 else if (mirror_num)
2730 stripe_index = mirror_num - 1;
2732 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2733 int factor = map->num_stripes / map->sub_stripes;
2735 stripe_index = do_div(stripe_nr, factor);
2736 stripe_index *= map->sub_stripes;
2738 if (unplug_page || (rw & (1 << BIO_RW)))
2739 num_stripes = map->sub_stripes;
2740 else if (mirror_num)
2741 stripe_index += mirror_num - 1;
2743 stripe_index = find_live_mirror(map, stripe_index,
2744 map->sub_stripes, stripe_index +
2745 current->pid % map->sub_stripes);
2749 * after this do_div call, stripe_nr is the number of stripes
2750 * on this device we have to walk to find the data, and
2751 * stripe_index is the number of our device in the stripe array
2753 stripe_index = do_div(stripe_nr, map->num_stripes);
2755 BUG_ON(stripe_index >= map->num_stripes);
2757 for (i = 0; i < num_stripes; i++) {
2759 struct btrfs_device *device;
2760 struct backing_dev_info *bdi;
2762 device = map->stripes[stripe_index].dev;
2764 bdi = blk_get_backing_dev_info(device->bdev);
2765 if (bdi->unplug_io_fn)
2766 bdi->unplug_io_fn(bdi, unplug_page);
2769 multi->stripes[i].physical =
2770 map->stripes[stripe_index].physical +
2771 stripe_offset + stripe_nr * map->stripe_len;
2772 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2778 multi->num_stripes = num_stripes;
2779 multi->max_errors = max_errors;
2782 free_extent_map(em);
2786 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2787 u64 logical, u64 *length,
2788 struct btrfs_multi_bio **multi_ret, int mirror_num)
2790 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2794 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2795 u64 chunk_start, u64 physical, u64 devid,
2796 u64 **logical, int *naddrs, int *stripe_len)
2798 struct extent_map_tree *em_tree = &map_tree->map_tree;
2799 struct extent_map *em;
2800 struct map_lookup *map;
2807 read_lock(&em_tree->lock);
2808 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2809 read_unlock(&em_tree->lock);
2811 BUG_ON(!em || em->start != chunk_start);
2812 map = (struct map_lookup *)em->bdev;
2815 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2816 do_div(length, map->num_stripes / map->sub_stripes);
2817 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2818 do_div(length, map->num_stripes);
2820 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2823 for (i = 0; i < map->num_stripes; i++) {
2824 if (devid && map->stripes[i].dev->devid != devid)
2826 if (map->stripes[i].physical > physical ||
2827 map->stripes[i].physical + length <= physical)
2830 stripe_nr = physical - map->stripes[i].physical;
2831 do_div(stripe_nr, map->stripe_len);
2833 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2834 stripe_nr = stripe_nr * map->num_stripes + i;
2835 do_div(stripe_nr, map->sub_stripes);
2836 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2837 stripe_nr = stripe_nr * map->num_stripes + i;
2839 bytenr = chunk_start + stripe_nr * map->stripe_len;
2840 WARN_ON(nr >= map->num_stripes);
2841 for (j = 0; j < nr; j++) {
2842 if (buf[j] == bytenr)
2846 WARN_ON(nr >= map->num_stripes);
2853 *stripe_len = map->stripe_len;
2855 free_extent_map(em);
2859 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2860 u64 logical, struct page *page)
2862 u64 length = PAGE_CACHE_SIZE;
2863 return __btrfs_map_block(map_tree, READ, logical, &length,
2867 static void end_bio_multi_stripe(struct bio *bio, int err)
2869 struct btrfs_multi_bio *multi = bio->bi_private;
2870 int is_orig_bio = 0;
2873 atomic_inc(&multi->error);
2875 if (bio == multi->orig_bio)
2878 if (atomic_dec_and_test(&multi->stripes_pending)) {
2881 bio = multi->orig_bio;
2883 bio->bi_private = multi->private;
2884 bio->bi_end_io = multi->end_io;
2885 /* only send an error to the higher layers if it is
2886 * beyond the tolerance of the multi-bio
2888 if (atomic_read(&multi->error) > multi->max_errors) {
2892 * this bio is actually up to date, we didn't
2893 * go over the max number of errors
2895 set_bit(BIO_UPTODATE, &bio->bi_flags);
2900 bio_endio(bio, err);
2901 } else if (!is_orig_bio) {
2906 struct async_sched {
2909 struct btrfs_fs_info *info;
2910 struct btrfs_work work;
2914 * see run_scheduled_bios for a description of why bios are collected for
2917 * This will add one bio to the pending list for a device and make sure
2918 * the work struct is scheduled.
2920 static noinline int schedule_bio(struct btrfs_root *root,
2921 struct btrfs_device *device,
2922 int rw, struct bio *bio)
2924 int should_queue = 1;
2925 struct btrfs_pending_bios *pending_bios;
2927 /* don't bother with additional async steps for reads, right now */
2928 if (!(rw & (1 << BIO_RW))) {
2930 submit_bio(rw, bio);
2936 * nr_async_bios allows us to reliably return congestion to the
2937 * higher layers. Otherwise, the async bio makes it appear we have
2938 * made progress against dirty pages when we've really just put it
2939 * on a queue for later
2941 atomic_inc(&root->fs_info->nr_async_bios);
2942 WARN_ON(bio->bi_next);
2943 bio->bi_next = NULL;
2946 spin_lock(&device->io_lock);
2947 if (bio_rw_flagged(bio, BIO_RW_SYNCIO))
2948 pending_bios = &device->pending_sync_bios;
2950 pending_bios = &device->pending_bios;
2952 if (pending_bios->tail)
2953 pending_bios->tail->bi_next = bio;
2955 pending_bios->tail = bio;
2956 if (!pending_bios->head)
2957 pending_bios->head = bio;
2958 if (device->running_pending)
2961 spin_unlock(&device->io_lock);
2964 btrfs_queue_worker(&root->fs_info->submit_workers,
2969 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2970 int mirror_num, int async_submit)
2972 struct btrfs_mapping_tree *map_tree;
2973 struct btrfs_device *dev;
2974 struct bio *first_bio = bio;
2975 u64 logical = (u64)bio->bi_sector << 9;
2978 struct btrfs_multi_bio *multi = NULL;
2983 length = bio->bi_size;
2984 map_tree = &root->fs_info->mapping_tree;
2985 map_length = length;
2987 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2991 total_devs = multi->num_stripes;
2992 if (map_length < length) {
2993 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2994 "len %llu\n", (unsigned long long)logical,
2995 (unsigned long long)length,
2996 (unsigned long long)map_length);
2999 multi->end_io = first_bio->bi_end_io;
3000 multi->private = first_bio->bi_private;
3001 multi->orig_bio = first_bio;
3002 atomic_set(&multi->stripes_pending, multi->num_stripes);
3004 while (dev_nr < total_devs) {
3005 if (total_devs > 1) {
3006 if (dev_nr < total_devs - 1) {
3007 bio = bio_clone(first_bio, GFP_NOFS);
3012 bio->bi_private = multi;
3013 bio->bi_end_io = end_bio_multi_stripe;
3015 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
3016 dev = multi->stripes[dev_nr].dev;
3017 BUG_ON(rw == WRITE && !dev->writeable);
3018 if (dev && dev->bdev) {
3019 bio->bi_bdev = dev->bdev;
3021 schedule_bio(root, dev, rw, bio);
3023 submit_bio(rw, bio);
3025 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
3026 bio->bi_sector = logical >> 9;
3027 bio_endio(bio, -EIO);
3031 if (total_devs == 1)
3036 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
3039 struct btrfs_device *device;
3040 struct btrfs_fs_devices *cur_devices;
3042 cur_devices = root->fs_info->fs_devices;
3043 while (cur_devices) {
3045 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3046 device = __find_device(&cur_devices->devices,
3051 cur_devices = cur_devices->seed;
3056 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
3057 u64 devid, u8 *dev_uuid)
3059 struct btrfs_device *device;
3060 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
3062 device = kzalloc(sizeof(*device), GFP_NOFS);
3065 list_add(&device->dev_list,
3066 &fs_devices->devices);
3067 device->barriers = 1;
3068 device->dev_root = root->fs_info->dev_root;
3069 device->devid = devid;
3070 device->work.func = pending_bios_fn;
3071 device->fs_devices = fs_devices;
3072 fs_devices->num_devices++;
3073 spin_lock_init(&device->io_lock);
3074 INIT_LIST_HEAD(&device->dev_alloc_list);
3075 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3079 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3080 struct extent_buffer *leaf,
3081 struct btrfs_chunk *chunk)
3083 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3084 struct map_lookup *map;
3085 struct extent_map *em;
3089 u8 uuid[BTRFS_UUID_SIZE];
3094 logical = key->offset;
3095 length = btrfs_chunk_length(leaf, chunk);
3097 read_lock(&map_tree->map_tree.lock);
3098 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3099 read_unlock(&map_tree->map_tree.lock);
3101 /* already mapped? */
3102 if (em && em->start <= logical && em->start + em->len > logical) {
3103 free_extent_map(em);
3106 free_extent_map(em);
3109 em = alloc_extent_map(GFP_NOFS);
3112 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3113 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3115 free_extent_map(em);
3119 em->bdev = (struct block_device *)map;
3120 em->start = logical;
3122 em->block_start = 0;
3123 em->block_len = em->len;
3125 map->num_stripes = num_stripes;
3126 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3127 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3128 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3129 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3130 map->type = btrfs_chunk_type(leaf, chunk);
3131 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3132 for (i = 0; i < num_stripes; i++) {
3133 map->stripes[i].physical =
3134 btrfs_stripe_offset_nr(leaf, chunk, i);
3135 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3136 read_extent_buffer(leaf, uuid, (unsigned long)
3137 btrfs_stripe_dev_uuid_nr(chunk, i),
3139 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3141 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3143 free_extent_map(em);
3146 if (!map->stripes[i].dev) {
3147 map->stripes[i].dev =
3148 add_missing_dev(root, devid, uuid);
3149 if (!map->stripes[i].dev) {
3151 free_extent_map(em);
3155 map->stripes[i].dev->in_fs_metadata = 1;
3158 write_lock(&map_tree->map_tree.lock);
3159 ret = add_extent_mapping(&map_tree->map_tree, em);
3160 write_unlock(&map_tree->map_tree.lock);
3162 free_extent_map(em);
3167 static int fill_device_from_item(struct extent_buffer *leaf,
3168 struct btrfs_dev_item *dev_item,
3169 struct btrfs_device *device)
3173 device->devid = btrfs_device_id(leaf, dev_item);
3174 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3175 device->total_bytes = device->disk_total_bytes;
3176 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3177 device->type = btrfs_device_type(leaf, dev_item);
3178 device->io_align = btrfs_device_io_align(leaf, dev_item);
3179 device->io_width = btrfs_device_io_width(leaf, dev_item);
3180 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3182 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3183 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3188 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3190 struct btrfs_fs_devices *fs_devices;
3193 mutex_lock(&uuid_mutex);
3195 fs_devices = root->fs_info->fs_devices->seed;
3196 while (fs_devices) {
3197 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3201 fs_devices = fs_devices->seed;
3204 fs_devices = find_fsid(fsid);
3210 fs_devices = clone_fs_devices(fs_devices);
3211 if (IS_ERR(fs_devices)) {
3212 ret = PTR_ERR(fs_devices);
3216 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3217 root->fs_info->bdev_holder);
3221 if (!fs_devices->seeding) {
3222 __btrfs_close_devices(fs_devices);
3223 free_fs_devices(fs_devices);
3228 fs_devices->seed = root->fs_info->fs_devices->seed;
3229 root->fs_info->fs_devices->seed = fs_devices;
3231 mutex_unlock(&uuid_mutex);
3235 static int read_one_dev(struct btrfs_root *root,
3236 struct extent_buffer *leaf,
3237 struct btrfs_dev_item *dev_item)
3239 struct btrfs_device *device;
3242 u8 fs_uuid[BTRFS_UUID_SIZE];
3243 u8 dev_uuid[BTRFS_UUID_SIZE];
3245 devid = btrfs_device_id(leaf, dev_item);
3246 read_extent_buffer(leaf, dev_uuid,
3247 (unsigned long)btrfs_device_uuid(dev_item),
3249 read_extent_buffer(leaf, fs_uuid,
3250 (unsigned long)btrfs_device_fsid(dev_item),
3253 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3254 ret = open_seed_devices(root, fs_uuid);
3255 if (ret && !btrfs_test_opt(root, DEGRADED))
3259 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3260 if (!device || !device->bdev) {
3261 if (!btrfs_test_opt(root, DEGRADED))
3265 printk(KERN_WARNING "warning devid %llu missing\n",
3266 (unsigned long long)devid);
3267 device = add_missing_dev(root, devid, dev_uuid);
3273 if (device->fs_devices != root->fs_info->fs_devices) {
3274 BUG_ON(device->writeable);
3275 if (device->generation !=
3276 btrfs_device_generation(leaf, dev_item))
3280 fill_device_from_item(leaf, dev_item, device);
3281 device->dev_root = root->fs_info->dev_root;
3282 device->in_fs_metadata = 1;
3283 if (device->writeable)
3284 device->fs_devices->total_rw_bytes += device->total_bytes;
3289 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3291 struct btrfs_dev_item *dev_item;
3293 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3295 return read_one_dev(root, buf, dev_item);
3298 int btrfs_read_sys_array(struct btrfs_root *root)
3300 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3301 struct extent_buffer *sb;
3302 struct btrfs_disk_key *disk_key;
3303 struct btrfs_chunk *chunk;
3305 unsigned long sb_ptr;
3311 struct btrfs_key key;
3313 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3314 BTRFS_SUPER_INFO_SIZE);
3317 btrfs_set_buffer_uptodate(sb);
3318 btrfs_set_buffer_lockdep_class(sb, 0);
3320 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3321 array_size = btrfs_super_sys_array_size(super_copy);
3323 ptr = super_copy->sys_chunk_array;
3324 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3327 while (cur < array_size) {
3328 disk_key = (struct btrfs_disk_key *)ptr;
3329 btrfs_disk_key_to_cpu(&key, disk_key);
3331 len = sizeof(*disk_key); ptr += len;
3335 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3336 chunk = (struct btrfs_chunk *)sb_ptr;
3337 ret = read_one_chunk(root, &key, sb, chunk);
3340 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3341 len = btrfs_chunk_item_size(num_stripes);
3350 free_extent_buffer(sb);
3354 int btrfs_read_chunk_tree(struct btrfs_root *root)
3356 struct btrfs_path *path;
3357 struct extent_buffer *leaf;
3358 struct btrfs_key key;
3359 struct btrfs_key found_key;
3363 root = root->fs_info->chunk_root;
3365 path = btrfs_alloc_path();
3369 /* first we search for all of the device items, and then we
3370 * read in all of the chunk items. This way we can create chunk
3371 * mappings that reference all of the devices that are afound
3373 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3377 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3379 leaf = path->nodes[0];
3380 slot = path->slots[0];
3381 if (slot >= btrfs_header_nritems(leaf)) {
3382 ret = btrfs_next_leaf(root, path);
3389 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3390 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3391 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3393 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3394 struct btrfs_dev_item *dev_item;
3395 dev_item = btrfs_item_ptr(leaf, slot,
3396 struct btrfs_dev_item);
3397 ret = read_one_dev(root, leaf, dev_item);
3401 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3402 struct btrfs_chunk *chunk;
3403 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3404 ret = read_one_chunk(root, &found_key, leaf, chunk);
3410 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3412 btrfs_release_path(root, path);
3417 btrfs_free_path(path);