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);
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 > 32 &&
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;
382 device = __find_device(&fs_devices->devices, devid,
383 disk_super->dev_item.uuid);
386 if (fs_devices->opened)
389 device = kzalloc(sizeof(*device), GFP_NOFS);
391 /* we can safely leave the fs_devices entry around */
394 device->devid = devid;
395 device->work.func = pending_bios_fn;
396 memcpy(device->uuid, disk_super->dev_item.uuid,
398 device->barriers = 1;
399 spin_lock_init(&device->io_lock);
400 device->name = kstrdup(path, GFP_NOFS);
405 INIT_LIST_HEAD(&device->dev_alloc_list);
406 list_add(&device->dev_list, &fs_devices->devices);
407 device->fs_devices = fs_devices;
408 fs_devices->num_devices++;
411 if (found_transid > fs_devices->latest_trans) {
412 fs_devices->latest_devid = devid;
413 fs_devices->latest_trans = found_transid;
415 *fs_devices_ret = fs_devices;
419 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
421 struct btrfs_fs_devices *fs_devices;
422 struct btrfs_device *device;
423 struct btrfs_device *orig_dev;
425 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
427 return ERR_PTR(-ENOMEM);
429 INIT_LIST_HEAD(&fs_devices->devices);
430 INIT_LIST_HEAD(&fs_devices->alloc_list);
431 INIT_LIST_HEAD(&fs_devices->list);
432 fs_devices->latest_devid = orig->latest_devid;
433 fs_devices->latest_trans = orig->latest_trans;
434 memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
436 list_for_each_entry(orig_dev, &orig->devices, dev_list) {
437 device = kzalloc(sizeof(*device), GFP_NOFS);
441 device->name = kstrdup(orig_dev->name, GFP_NOFS);
445 device->devid = orig_dev->devid;
446 device->work.func = pending_bios_fn;
447 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
448 device->barriers = 1;
449 spin_lock_init(&device->io_lock);
450 INIT_LIST_HEAD(&device->dev_list);
451 INIT_LIST_HEAD(&device->dev_alloc_list);
453 list_add(&device->dev_list, &fs_devices->devices);
454 device->fs_devices = fs_devices;
455 fs_devices->num_devices++;
459 free_fs_devices(fs_devices);
460 return ERR_PTR(-ENOMEM);
463 int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices)
465 struct btrfs_device *device, *next;
467 mutex_lock(&uuid_mutex);
469 list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
470 if (device->in_fs_metadata)
474 close_bdev_exclusive(device->bdev, device->mode);
476 fs_devices->open_devices--;
478 if (device->writeable) {
479 list_del_init(&device->dev_alloc_list);
480 device->writeable = 0;
481 fs_devices->rw_devices--;
483 list_del_init(&device->dev_list);
484 fs_devices->num_devices--;
489 if (fs_devices->seed) {
490 fs_devices = fs_devices->seed;
494 mutex_unlock(&uuid_mutex);
498 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
500 struct btrfs_device *device;
502 if (--fs_devices->opened > 0)
505 list_for_each_entry(device, &fs_devices->devices, dev_list) {
507 close_bdev_exclusive(device->bdev, device->mode);
508 fs_devices->open_devices--;
510 if (device->writeable) {
511 list_del_init(&device->dev_alloc_list);
512 fs_devices->rw_devices--;
516 device->writeable = 0;
517 device->in_fs_metadata = 0;
519 WARN_ON(fs_devices->open_devices);
520 WARN_ON(fs_devices->rw_devices);
521 fs_devices->opened = 0;
522 fs_devices->seeding = 0;
527 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
529 struct btrfs_fs_devices *seed_devices = NULL;
532 mutex_lock(&uuid_mutex);
533 ret = __btrfs_close_devices(fs_devices);
534 if (!fs_devices->opened) {
535 seed_devices = fs_devices->seed;
536 fs_devices->seed = NULL;
538 mutex_unlock(&uuid_mutex);
540 while (seed_devices) {
541 fs_devices = seed_devices;
542 seed_devices = fs_devices->seed;
543 __btrfs_close_devices(fs_devices);
544 free_fs_devices(fs_devices);
549 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
550 fmode_t flags, void *holder)
552 struct block_device *bdev;
553 struct list_head *head = &fs_devices->devices;
554 struct btrfs_device *device;
555 struct block_device *latest_bdev = NULL;
556 struct buffer_head *bh;
557 struct btrfs_super_block *disk_super;
558 u64 latest_devid = 0;
559 u64 latest_transid = 0;
564 list_for_each_entry(device, head, dev_list) {
570 bdev = open_bdev_exclusive(device->name, flags, holder);
572 printk(KERN_INFO "open %s failed\n", device->name);
575 set_blocksize(bdev, 4096);
577 bh = btrfs_read_dev_super(bdev);
581 disk_super = (struct btrfs_super_block *)bh->b_data;
582 devid = le64_to_cpu(disk_super->dev_item.devid);
583 if (devid != device->devid)
586 if (memcmp(device->uuid, disk_super->dev_item.uuid,
590 device->generation = btrfs_super_generation(disk_super);
591 if (!latest_transid || device->generation > latest_transid) {
592 latest_devid = devid;
593 latest_transid = device->generation;
597 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
598 device->writeable = 0;
600 device->writeable = !bdev_read_only(bdev);
605 device->in_fs_metadata = 0;
606 device->mode = flags;
608 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
609 fs_devices->rotating = 1;
611 fs_devices->open_devices++;
612 if (device->writeable) {
613 fs_devices->rw_devices++;
614 list_add(&device->dev_alloc_list,
615 &fs_devices->alloc_list);
622 close_bdev_exclusive(bdev, FMODE_READ);
626 if (fs_devices->open_devices == 0) {
630 fs_devices->seeding = seeding;
631 fs_devices->opened = 1;
632 fs_devices->latest_bdev = latest_bdev;
633 fs_devices->latest_devid = latest_devid;
634 fs_devices->latest_trans = latest_transid;
635 fs_devices->total_rw_bytes = 0;
640 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
641 fmode_t flags, void *holder)
645 mutex_lock(&uuid_mutex);
646 if (fs_devices->opened) {
647 fs_devices->opened++;
650 ret = __btrfs_open_devices(fs_devices, flags, holder);
652 mutex_unlock(&uuid_mutex);
656 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
657 struct btrfs_fs_devices **fs_devices_ret)
659 struct btrfs_super_block *disk_super;
660 struct block_device *bdev;
661 struct buffer_head *bh;
666 mutex_lock(&uuid_mutex);
668 bdev = open_bdev_exclusive(path, flags, holder);
675 ret = set_blocksize(bdev, 4096);
678 bh = btrfs_read_dev_super(bdev);
683 disk_super = (struct btrfs_super_block *)bh->b_data;
684 devid = le64_to_cpu(disk_super->dev_item.devid);
685 transid = btrfs_super_generation(disk_super);
686 if (disk_super->label[0])
687 printk(KERN_INFO "device label %s ", disk_super->label);
689 /* FIXME, make a readl uuid parser */
690 printk(KERN_INFO "device fsid %llx-%llx ",
691 *(unsigned long long *)disk_super->fsid,
692 *(unsigned long long *)(disk_super->fsid + 8));
694 printk(KERN_CONT "devid %llu transid %llu %s\n",
695 (unsigned long long)devid, (unsigned long long)transid, path);
696 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
700 close_bdev_exclusive(bdev, flags);
702 mutex_unlock(&uuid_mutex);
707 * this uses a pretty simple search, the expectation is that it is
708 * called very infrequently and that a given device has a small number
711 static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
712 struct btrfs_device *device,
713 u64 num_bytes, u64 *start)
715 struct btrfs_key key;
716 struct btrfs_root *root = device->dev_root;
717 struct btrfs_dev_extent *dev_extent = NULL;
718 struct btrfs_path *path;
721 u64 search_start = 0;
722 u64 search_end = device->total_bytes;
726 struct extent_buffer *l;
728 path = btrfs_alloc_path();
734 /* FIXME use last free of some kind */
736 /* we don't want to overwrite the superblock on the drive,
737 * so we make sure to start at an offset of at least 1MB
739 search_start = max((u64)1024 * 1024, search_start);
741 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
742 search_start = max(root->fs_info->alloc_start, search_start);
744 key.objectid = device->devid;
745 key.offset = search_start;
746 key.type = BTRFS_DEV_EXTENT_KEY;
747 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
750 ret = btrfs_previous_item(root, path, 0, key.type);
754 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
757 slot = path->slots[0];
758 if (slot >= btrfs_header_nritems(l)) {
759 ret = btrfs_next_leaf(root, path);
766 if (search_start >= search_end) {
770 *start = search_start;
774 *start = last_byte > search_start ?
775 last_byte : search_start;
776 if (search_end <= *start) {
782 btrfs_item_key_to_cpu(l, &key, slot);
784 if (key.objectid < device->devid)
787 if (key.objectid > device->devid)
790 if (key.offset >= search_start && key.offset > last_byte &&
792 if (last_byte < search_start)
793 last_byte = search_start;
794 hole_size = key.offset - last_byte;
795 if (key.offset > last_byte &&
796 hole_size >= num_bytes) {
801 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
805 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
806 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
812 /* we have to make sure we didn't find an extent that has already
813 * been allocated by the map tree or the original allocation
815 BUG_ON(*start < search_start);
817 if (*start + num_bytes > search_end) {
821 /* check for pending inserts here */
825 btrfs_free_path(path);
829 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
830 struct btrfs_device *device,
834 struct btrfs_path *path;
835 struct btrfs_root *root = device->dev_root;
836 struct btrfs_key key;
837 struct btrfs_key found_key;
838 struct extent_buffer *leaf = NULL;
839 struct btrfs_dev_extent *extent = NULL;
841 path = btrfs_alloc_path();
845 key.objectid = device->devid;
847 key.type = BTRFS_DEV_EXTENT_KEY;
849 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
851 ret = btrfs_previous_item(root, path, key.objectid,
852 BTRFS_DEV_EXTENT_KEY);
854 leaf = path->nodes[0];
855 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
856 extent = btrfs_item_ptr(leaf, path->slots[0],
857 struct btrfs_dev_extent);
858 BUG_ON(found_key.offset > start || found_key.offset +
859 btrfs_dev_extent_length(leaf, extent) < start);
861 } else if (ret == 0) {
862 leaf = path->nodes[0];
863 extent = btrfs_item_ptr(leaf, path->slots[0],
864 struct btrfs_dev_extent);
868 if (device->bytes_used > 0)
869 device->bytes_used -= btrfs_dev_extent_length(leaf, extent);
870 ret = btrfs_del_item(trans, root, path);
873 btrfs_free_path(path);
877 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
878 struct btrfs_device *device,
879 u64 chunk_tree, u64 chunk_objectid,
880 u64 chunk_offset, u64 start, u64 num_bytes)
883 struct btrfs_path *path;
884 struct btrfs_root *root = device->dev_root;
885 struct btrfs_dev_extent *extent;
886 struct extent_buffer *leaf;
887 struct btrfs_key key;
889 WARN_ON(!device->in_fs_metadata);
890 path = btrfs_alloc_path();
894 key.objectid = device->devid;
896 key.type = BTRFS_DEV_EXTENT_KEY;
897 ret = btrfs_insert_empty_item(trans, root, path, &key,
901 leaf = path->nodes[0];
902 extent = btrfs_item_ptr(leaf, path->slots[0],
903 struct btrfs_dev_extent);
904 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
905 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
906 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
908 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
909 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
912 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
913 btrfs_mark_buffer_dirty(leaf);
914 btrfs_free_path(path);
918 static noinline int find_next_chunk(struct btrfs_root *root,
919 u64 objectid, u64 *offset)
921 struct btrfs_path *path;
923 struct btrfs_key key;
924 struct btrfs_chunk *chunk;
925 struct btrfs_key found_key;
927 path = btrfs_alloc_path();
930 key.objectid = objectid;
931 key.offset = (u64)-1;
932 key.type = BTRFS_CHUNK_ITEM_KEY;
934 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
940 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
944 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
946 if (found_key.objectid != objectid)
949 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
951 *offset = found_key.offset +
952 btrfs_chunk_length(path->nodes[0], chunk);
957 btrfs_free_path(path);
961 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
964 struct btrfs_key key;
965 struct btrfs_key found_key;
966 struct btrfs_path *path;
968 root = root->fs_info->chunk_root;
970 path = btrfs_alloc_path();
974 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
975 key.type = BTRFS_DEV_ITEM_KEY;
976 key.offset = (u64)-1;
978 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
984 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
989 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
991 *objectid = found_key.offset + 1;
995 btrfs_free_path(path);
1000 * the device information is stored in the chunk root
1001 * the btrfs_device struct should be fully filled in
1003 int btrfs_add_device(struct btrfs_trans_handle *trans,
1004 struct btrfs_root *root,
1005 struct btrfs_device *device)
1008 struct btrfs_path *path;
1009 struct btrfs_dev_item *dev_item;
1010 struct extent_buffer *leaf;
1011 struct btrfs_key key;
1014 root = root->fs_info->chunk_root;
1016 path = btrfs_alloc_path();
1020 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1021 key.type = BTRFS_DEV_ITEM_KEY;
1022 key.offset = device->devid;
1024 ret = btrfs_insert_empty_item(trans, root, path, &key,
1029 leaf = path->nodes[0];
1030 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1032 btrfs_set_device_id(leaf, dev_item, device->devid);
1033 btrfs_set_device_generation(leaf, dev_item, 0);
1034 btrfs_set_device_type(leaf, dev_item, device->type);
1035 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1036 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1037 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1038 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1039 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1040 btrfs_set_device_group(leaf, dev_item, 0);
1041 btrfs_set_device_seek_speed(leaf, dev_item, 0);
1042 btrfs_set_device_bandwidth(leaf, dev_item, 0);
1043 btrfs_set_device_start_offset(leaf, dev_item, 0);
1045 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1046 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1047 ptr = (unsigned long)btrfs_device_fsid(dev_item);
1048 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1049 btrfs_mark_buffer_dirty(leaf);
1053 btrfs_free_path(path);
1057 static int btrfs_rm_dev_item(struct btrfs_root *root,
1058 struct btrfs_device *device)
1061 struct btrfs_path *path;
1062 struct btrfs_key key;
1063 struct btrfs_trans_handle *trans;
1065 root = root->fs_info->chunk_root;
1067 path = btrfs_alloc_path();
1071 trans = btrfs_start_transaction(root, 1);
1072 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1073 key.type = BTRFS_DEV_ITEM_KEY;
1074 key.offset = device->devid;
1077 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1086 ret = btrfs_del_item(trans, root, path);
1090 btrfs_free_path(path);
1091 unlock_chunks(root);
1092 btrfs_commit_transaction(trans, root);
1096 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1098 struct btrfs_device *device;
1099 struct btrfs_device *next_device;
1100 struct block_device *bdev;
1101 struct buffer_head *bh = NULL;
1102 struct btrfs_super_block *disk_super;
1109 mutex_lock(&uuid_mutex);
1110 mutex_lock(&root->fs_info->volume_mutex);
1112 all_avail = root->fs_info->avail_data_alloc_bits |
1113 root->fs_info->avail_system_alloc_bits |
1114 root->fs_info->avail_metadata_alloc_bits;
1116 if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) &&
1117 root->fs_info->fs_devices->rw_devices <= 4) {
1118 printk(KERN_ERR "btrfs: unable to go below four devices "
1124 if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) &&
1125 root->fs_info->fs_devices->rw_devices <= 2) {
1126 printk(KERN_ERR "btrfs: unable to go below two "
1127 "devices on raid1\n");
1132 if (strcmp(device_path, "missing") == 0) {
1133 struct list_head *devices;
1134 struct btrfs_device *tmp;
1137 devices = &root->fs_info->fs_devices->devices;
1138 list_for_each_entry(tmp, devices, dev_list) {
1139 if (tmp->in_fs_metadata && !tmp->bdev) {
1148 printk(KERN_ERR "btrfs: no missing devices found to "
1153 bdev = open_bdev_exclusive(device_path, FMODE_READ,
1154 root->fs_info->bdev_holder);
1156 ret = PTR_ERR(bdev);
1160 set_blocksize(bdev, 4096);
1161 bh = btrfs_read_dev_super(bdev);
1166 disk_super = (struct btrfs_super_block *)bh->b_data;
1167 devid = le64_to_cpu(disk_super->dev_item.devid);
1168 dev_uuid = disk_super->dev_item.uuid;
1169 device = btrfs_find_device(root, devid, dev_uuid,
1177 if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1178 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1184 if (device->writeable) {
1185 list_del_init(&device->dev_alloc_list);
1186 root->fs_info->fs_devices->rw_devices--;
1189 ret = btrfs_shrink_device(device, 0);
1193 ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1197 device->in_fs_metadata = 0;
1198 list_del_init(&device->dev_list);
1199 device->fs_devices->num_devices--;
1201 next_device = list_entry(root->fs_info->fs_devices->devices.next,
1202 struct btrfs_device, dev_list);
1203 if (device->bdev == root->fs_info->sb->s_bdev)
1204 root->fs_info->sb->s_bdev = next_device->bdev;
1205 if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1206 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1209 close_bdev_exclusive(device->bdev, device->mode);
1210 device->bdev = NULL;
1211 device->fs_devices->open_devices--;
1214 num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1215 btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices);
1217 if (device->fs_devices->open_devices == 0) {
1218 struct btrfs_fs_devices *fs_devices;
1219 fs_devices = root->fs_info->fs_devices;
1220 while (fs_devices) {
1221 if (fs_devices->seed == device->fs_devices)
1223 fs_devices = fs_devices->seed;
1225 fs_devices->seed = device->fs_devices->seed;
1226 device->fs_devices->seed = NULL;
1227 __btrfs_close_devices(device->fs_devices);
1228 free_fs_devices(device->fs_devices);
1232 * at this point, the device is zero sized. We want to
1233 * remove it from the devices list and zero out the old super
1235 if (device->writeable) {
1236 /* make sure this device isn't detected as part of
1239 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1240 set_buffer_dirty(bh);
1241 sync_dirty_buffer(bh);
1244 kfree(device->name);
1252 close_bdev_exclusive(bdev, FMODE_READ);
1254 mutex_unlock(&root->fs_info->volume_mutex);
1255 mutex_unlock(&uuid_mutex);
1260 * does all the dirty work required for changing file system's UUID.
1262 static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans,
1263 struct btrfs_root *root)
1265 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1266 struct btrfs_fs_devices *old_devices;
1267 struct btrfs_fs_devices *seed_devices;
1268 struct btrfs_super_block *disk_super = &root->fs_info->super_copy;
1269 struct btrfs_device *device;
1272 BUG_ON(!mutex_is_locked(&uuid_mutex));
1273 if (!fs_devices->seeding)
1276 seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1280 old_devices = clone_fs_devices(fs_devices);
1281 if (IS_ERR(old_devices)) {
1282 kfree(seed_devices);
1283 return PTR_ERR(old_devices);
1286 list_add(&old_devices->list, &fs_uuids);
1288 memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1289 seed_devices->opened = 1;
1290 INIT_LIST_HEAD(&seed_devices->devices);
1291 INIT_LIST_HEAD(&seed_devices->alloc_list);
1292 list_splice_init(&fs_devices->devices, &seed_devices->devices);
1293 list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1294 list_for_each_entry(device, &seed_devices->devices, dev_list) {
1295 device->fs_devices = seed_devices;
1298 fs_devices->seeding = 0;
1299 fs_devices->num_devices = 0;
1300 fs_devices->open_devices = 0;
1301 fs_devices->seed = seed_devices;
1303 generate_random_uuid(fs_devices->fsid);
1304 memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1305 memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1306 super_flags = btrfs_super_flags(disk_super) &
1307 ~BTRFS_SUPER_FLAG_SEEDING;
1308 btrfs_set_super_flags(disk_super, super_flags);
1314 * strore the expected generation for seed devices in device items.
1316 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1317 struct btrfs_root *root)
1319 struct btrfs_path *path;
1320 struct extent_buffer *leaf;
1321 struct btrfs_dev_item *dev_item;
1322 struct btrfs_device *device;
1323 struct btrfs_key key;
1324 u8 fs_uuid[BTRFS_UUID_SIZE];
1325 u8 dev_uuid[BTRFS_UUID_SIZE];
1329 path = btrfs_alloc_path();
1333 root = root->fs_info->chunk_root;
1334 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1336 key.type = BTRFS_DEV_ITEM_KEY;
1339 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1343 leaf = path->nodes[0];
1345 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1346 ret = btrfs_next_leaf(root, path);
1351 leaf = path->nodes[0];
1352 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1353 btrfs_release_path(root, path);
1357 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1358 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1359 key.type != BTRFS_DEV_ITEM_KEY)
1362 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1363 struct btrfs_dev_item);
1364 devid = btrfs_device_id(leaf, dev_item);
1365 read_extent_buffer(leaf, dev_uuid,
1366 (unsigned long)btrfs_device_uuid(dev_item),
1368 read_extent_buffer(leaf, fs_uuid,
1369 (unsigned long)btrfs_device_fsid(dev_item),
1371 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1374 if (device->fs_devices->seeding) {
1375 btrfs_set_device_generation(leaf, dev_item,
1376 device->generation);
1377 btrfs_mark_buffer_dirty(leaf);
1385 btrfs_free_path(path);
1389 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1391 struct btrfs_trans_handle *trans;
1392 struct btrfs_device *device;
1393 struct block_device *bdev;
1394 struct list_head *devices;
1395 struct super_block *sb = root->fs_info->sb;
1397 int seeding_dev = 0;
1400 if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1403 bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder);
1407 if (root->fs_info->fs_devices->seeding) {
1409 down_write(&sb->s_umount);
1410 mutex_lock(&uuid_mutex);
1413 filemap_write_and_wait(bdev->bd_inode->i_mapping);
1414 mutex_lock(&root->fs_info->volume_mutex);
1416 devices = &root->fs_info->fs_devices->devices;
1417 list_for_each_entry(device, devices, dev_list) {
1418 if (device->bdev == bdev) {
1424 device = kzalloc(sizeof(*device), GFP_NOFS);
1426 /* we can safely leave the fs_devices entry around */
1431 device->name = kstrdup(device_path, GFP_NOFS);
1432 if (!device->name) {
1438 ret = find_next_devid(root, &device->devid);
1444 trans = btrfs_start_transaction(root, 1);
1447 device->barriers = 1;
1448 device->writeable = 1;
1449 device->work.func = pending_bios_fn;
1450 generate_random_uuid(device->uuid);
1451 spin_lock_init(&device->io_lock);
1452 device->generation = trans->transid;
1453 device->io_width = root->sectorsize;
1454 device->io_align = root->sectorsize;
1455 device->sector_size = root->sectorsize;
1456 device->total_bytes = i_size_read(bdev->bd_inode);
1457 device->disk_total_bytes = device->total_bytes;
1458 device->dev_root = root->fs_info->dev_root;
1459 device->bdev = bdev;
1460 device->in_fs_metadata = 1;
1462 set_blocksize(device->bdev, 4096);
1465 sb->s_flags &= ~MS_RDONLY;
1466 ret = btrfs_prepare_sprout(trans, root);
1470 device->fs_devices = root->fs_info->fs_devices;
1471 list_add(&device->dev_list, &root->fs_info->fs_devices->devices);
1472 list_add(&device->dev_alloc_list,
1473 &root->fs_info->fs_devices->alloc_list);
1474 root->fs_info->fs_devices->num_devices++;
1475 root->fs_info->fs_devices->open_devices++;
1476 root->fs_info->fs_devices->rw_devices++;
1477 root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
1479 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
1480 root->fs_info->fs_devices->rotating = 1;
1482 total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
1483 btrfs_set_super_total_bytes(&root->fs_info->super_copy,
1484 total_bytes + device->total_bytes);
1486 total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy);
1487 btrfs_set_super_num_devices(&root->fs_info->super_copy,
1491 ret = init_first_rw_device(trans, root, device);
1493 ret = btrfs_finish_sprout(trans, root);
1496 ret = btrfs_add_device(trans, root, device);
1500 * we've got more storage, clear any full flags on the space
1503 btrfs_clear_space_info_full(root->fs_info);
1505 unlock_chunks(root);
1506 btrfs_commit_transaction(trans, root);
1509 mutex_unlock(&uuid_mutex);
1510 up_write(&sb->s_umount);
1512 ret = btrfs_relocate_sys_chunks(root);
1516 mutex_unlock(&root->fs_info->volume_mutex);
1519 close_bdev_exclusive(bdev, 0);
1521 mutex_unlock(&uuid_mutex);
1522 up_write(&sb->s_umount);
1527 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
1528 struct btrfs_device *device)
1531 struct btrfs_path *path;
1532 struct btrfs_root *root;
1533 struct btrfs_dev_item *dev_item;
1534 struct extent_buffer *leaf;
1535 struct btrfs_key key;
1537 root = device->dev_root->fs_info->chunk_root;
1539 path = btrfs_alloc_path();
1543 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1544 key.type = BTRFS_DEV_ITEM_KEY;
1545 key.offset = device->devid;
1547 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1556 leaf = path->nodes[0];
1557 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1559 btrfs_set_device_id(leaf, dev_item, device->devid);
1560 btrfs_set_device_type(leaf, dev_item, device->type);
1561 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1562 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1563 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1564 btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
1565 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1566 btrfs_mark_buffer_dirty(leaf);
1569 btrfs_free_path(path);
1573 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
1574 struct btrfs_device *device, u64 new_size)
1576 struct btrfs_super_block *super_copy =
1577 &device->dev_root->fs_info->super_copy;
1578 u64 old_total = btrfs_super_total_bytes(super_copy);
1579 u64 diff = new_size - device->total_bytes;
1581 if (!device->writeable)
1583 if (new_size <= device->total_bytes)
1586 btrfs_set_super_total_bytes(super_copy, old_total + diff);
1587 device->fs_devices->total_rw_bytes += diff;
1589 device->total_bytes = new_size;
1590 btrfs_clear_space_info_full(device->dev_root->fs_info);
1592 return btrfs_update_device(trans, device);
1595 int btrfs_grow_device(struct btrfs_trans_handle *trans,
1596 struct btrfs_device *device, u64 new_size)
1599 lock_chunks(device->dev_root);
1600 ret = __btrfs_grow_device(trans, device, new_size);
1601 unlock_chunks(device->dev_root);
1605 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
1606 struct btrfs_root *root,
1607 u64 chunk_tree, u64 chunk_objectid,
1611 struct btrfs_path *path;
1612 struct btrfs_key key;
1614 root = root->fs_info->chunk_root;
1615 path = btrfs_alloc_path();
1619 key.objectid = chunk_objectid;
1620 key.offset = chunk_offset;
1621 key.type = BTRFS_CHUNK_ITEM_KEY;
1623 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1626 ret = btrfs_del_item(trans, root, path);
1629 btrfs_free_path(path);
1633 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
1636 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1637 struct btrfs_disk_key *disk_key;
1638 struct btrfs_chunk *chunk;
1645 struct btrfs_key key;
1647 array_size = btrfs_super_sys_array_size(super_copy);
1649 ptr = super_copy->sys_chunk_array;
1652 while (cur < array_size) {
1653 disk_key = (struct btrfs_disk_key *)ptr;
1654 btrfs_disk_key_to_cpu(&key, disk_key);
1656 len = sizeof(*disk_key);
1658 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1659 chunk = (struct btrfs_chunk *)(ptr + len);
1660 num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1661 len += btrfs_chunk_item_size(num_stripes);
1666 if (key.objectid == chunk_objectid &&
1667 key.offset == chunk_offset) {
1668 memmove(ptr, ptr + len, array_size - (cur + len));
1670 btrfs_set_super_sys_array_size(super_copy, array_size);
1679 static int btrfs_relocate_chunk(struct btrfs_root *root,
1680 u64 chunk_tree, u64 chunk_objectid,
1683 struct extent_map_tree *em_tree;
1684 struct btrfs_root *extent_root;
1685 struct btrfs_trans_handle *trans;
1686 struct extent_map *em;
1687 struct map_lookup *map;
1691 root = root->fs_info->chunk_root;
1692 extent_root = root->fs_info->extent_root;
1693 em_tree = &root->fs_info->mapping_tree.map_tree;
1695 /* step one, relocate all the extents inside this chunk */
1696 ret = btrfs_relocate_block_group(extent_root, chunk_offset);
1699 trans = btrfs_start_transaction(root, 1);
1705 * step two, delete the device extents and the
1706 * chunk tree entries
1708 spin_lock(&em_tree->lock);
1709 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
1710 spin_unlock(&em_tree->lock);
1712 BUG_ON(em->start > chunk_offset ||
1713 em->start + em->len < chunk_offset);
1714 map = (struct map_lookup *)em->bdev;
1716 for (i = 0; i < map->num_stripes; i++) {
1717 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
1718 map->stripes[i].physical);
1721 if (map->stripes[i].dev) {
1722 ret = btrfs_update_device(trans, map->stripes[i].dev);
1726 ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
1731 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
1732 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
1736 ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
1739 spin_lock(&em_tree->lock);
1740 remove_extent_mapping(em_tree, em);
1741 spin_unlock(&em_tree->lock);
1746 /* once for the tree */
1747 free_extent_map(em);
1749 free_extent_map(em);
1751 unlock_chunks(root);
1752 btrfs_end_transaction(trans, root);
1756 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
1758 struct btrfs_root *chunk_root = root->fs_info->chunk_root;
1759 struct btrfs_path *path;
1760 struct extent_buffer *leaf;
1761 struct btrfs_chunk *chunk;
1762 struct btrfs_key key;
1763 struct btrfs_key found_key;
1764 u64 chunk_tree = chunk_root->root_key.objectid;
1768 path = btrfs_alloc_path();
1772 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1773 key.offset = (u64)-1;
1774 key.type = BTRFS_CHUNK_ITEM_KEY;
1777 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1782 ret = btrfs_previous_item(chunk_root, path, key.objectid,
1789 leaf = path->nodes[0];
1790 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1792 chunk = btrfs_item_ptr(leaf, path->slots[0],
1793 struct btrfs_chunk);
1794 chunk_type = btrfs_chunk_type(leaf, chunk);
1795 btrfs_release_path(chunk_root, path);
1797 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
1798 ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
1804 if (found_key.offset == 0)
1806 key.offset = found_key.offset - 1;
1810 btrfs_free_path(path);
1814 static u64 div_factor(u64 num, int factor)
1823 int btrfs_balance(struct btrfs_root *dev_root)
1826 struct list_head *devices = &dev_root->fs_info->fs_devices->devices;
1827 struct btrfs_device *device;
1830 struct btrfs_path *path;
1831 struct btrfs_key key;
1832 struct btrfs_chunk *chunk;
1833 struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root;
1834 struct btrfs_trans_handle *trans;
1835 struct btrfs_key found_key;
1837 if (dev_root->fs_info->sb->s_flags & MS_RDONLY)
1840 mutex_lock(&dev_root->fs_info->volume_mutex);
1841 dev_root = dev_root->fs_info->dev_root;
1843 /* step one make some room on all the devices */
1844 list_for_each_entry(device, devices, dev_list) {
1845 old_size = device->total_bytes;
1846 size_to_free = div_factor(old_size, 1);
1847 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
1848 if (!device->writeable ||
1849 device->total_bytes - device->bytes_used > size_to_free)
1852 ret = btrfs_shrink_device(device, old_size - size_to_free);
1855 trans = btrfs_start_transaction(dev_root, 1);
1858 ret = btrfs_grow_device(trans, device, old_size);
1861 btrfs_end_transaction(trans, dev_root);
1864 /* step two, relocate all the chunks */
1865 path = btrfs_alloc_path();
1868 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1869 key.offset = (u64)-1;
1870 key.type = BTRFS_CHUNK_ITEM_KEY;
1873 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
1878 * this shouldn't happen, it means the last relocate
1884 ret = btrfs_previous_item(chunk_root, path, 0,
1885 BTRFS_CHUNK_ITEM_KEY);
1889 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1891 if (found_key.objectid != key.objectid)
1894 chunk = btrfs_item_ptr(path->nodes[0],
1896 struct btrfs_chunk);
1897 key.offset = found_key.offset;
1898 /* chunk zero is special */
1899 if (key.offset == 0)
1902 btrfs_release_path(chunk_root, path);
1903 ret = btrfs_relocate_chunk(chunk_root,
1904 chunk_root->root_key.objectid,
1911 btrfs_free_path(path);
1912 mutex_unlock(&dev_root->fs_info->volume_mutex);
1917 * shrinking a device means finding all of the device extents past
1918 * the new size, and then following the back refs to the chunks.
1919 * The chunk relocation code actually frees the device extent
1921 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
1923 struct btrfs_trans_handle *trans;
1924 struct btrfs_root *root = device->dev_root;
1925 struct btrfs_dev_extent *dev_extent = NULL;
1926 struct btrfs_path *path;
1933 struct extent_buffer *l;
1934 struct btrfs_key key;
1935 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1936 u64 old_total = btrfs_super_total_bytes(super_copy);
1937 u64 diff = device->total_bytes - new_size;
1939 if (new_size >= device->total_bytes)
1942 path = btrfs_alloc_path();
1946 trans = btrfs_start_transaction(root, 1);
1956 device->total_bytes = new_size;
1957 if (device->writeable)
1958 device->fs_devices->total_rw_bytes -= diff;
1959 unlock_chunks(root);
1960 btrfs_end_transaction(trans, root);
1962 key.objectid = device->devid;
1963 key.offset = (u64)-1;
1964 key.type = BTRFS_DEV_EXTENT_KEY;
1967 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1971 ret = btrfs_previous_item(root, path, 0, key.type);
1980 slot = path->slots[0];
1981 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
1983 if (key.objectid != device->devid)
1986 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1987 length = btrfs_dev_extent_length(l, dev_extent);
1989 if (key.offset + length <= new_size)
1992 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
1993 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
1994 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
1995 btrfs_release_path(root, path);
1997 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
2003 /* Shrinking succeeded, else we would be at "done". */
2004 trans = btrfs_start_transaction(root, 1);
2011 device->disk_total_bytes = new_size;
2012 /* Now btrfs_update_device() will change the on-disk size. */
2013 ret = btrfs_update_device(trans, device);
2015 unlock_chunks(root);
2016 btrfs_end_transaction(trans, root);
2019 WARN_ON(diff > old_total);
2020 btrfs_set_super_total_bytes(super_copy, old_total - diff);
2021 unlock_chunks(root);
2022 btrfs_end_transaction(trans, root);
2024 btrfs_free_path(path);
2028 static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
2029 struct btrfs_root *root,
2030 struct btrfs_key *key,
2031 struct btrfs_chunk *chunk, int item_size)
2033 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
2034 struct btrfs_disk_key disk_key;
2038 array_size = btrfs_super_sys_array_size(super_copy);
2039 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
2042 ptr = super_copy->sys_chunk_array + array_size;
2043 btrfs_cpu_key_to_disk(&disk_key, key);
2044 memcpy(ptr, &disk_key, sizeof(disk_key));
2045 ptr += sizeof(disk_key);
2046 memcpy(ptr, chunk, item_size);
2047 item_size += sizeof(disk_key);
2048 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
2052 static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size,
2053 int num_stripes, int sub_stripes)
2055 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
2057 else if (type & BTRFS_BLOCK_GROUP_RAID10)
2058 return calc_size * (num_stripes / sub_stripes);
2060 return calc_size * num_stripes;
2063 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2064 struct btrfs_root *extent_root,
2065 struct map_lookup **map_ret,
2066 u64 *num_bytes, u64 *stripe_size,
2067 u64 start, u64 type)
2069 struct btrfs_fs_info *info = extent_root->fs_info;
2070 struct btrfs_device *device = NULL;
2071 struct btrfs_fs_devices *fs_devices = info->fs_devices;
2072 struct list_head *cur;
2073 struct map_lookup *map = NULL;
2074 struct extent_map_tree *em_tree;
2075 struct extent_map *em;
2076 struct list_head private_devs;
2077 int min_stripe_size = 1 * 1024 * 1024;
2078 u64 calc_size = 1024 * 1024 * 1024;
2079 u64 max_chunk_size = calc_size;
2084 int num_stripes = 1;
2085 int min_stripes = 1;
2086 int sub_stripes = 0;
2090 int stripe_len = 64 * 1024;
2092 if ((type & BTRFS_BLOCK_GROUP_RAID1) &&
2093 (type & BTRFS_BLOCK_GROUP_DUP)) {
2095 type &= ~BTRFS_BLOCK_GROUP_DUP;
2097 if (list_empty(&fs_devices->alloc_list))
2100 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
2101 num_stripes = fs_devices->rw_devices;
2104 if (type & (BTRFS_BLOCK_GROUP_DUP)) {
2108 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
2109 num_stripes = min_t(u64, 2, fs_devices->rw_devices);
2110 if (num_stripes < 2)
2114 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2115 num_stripes = fs_devices->rw_devices;
2116 if (num_stripes < 4)
2118 num_stripes &= ~(u32)1;
2123 if (type & BTRFS_BLOCK_GROUP_DATA) {
2124 max_chunk_size = 10 * calc_size;
2125 min_stripe_size = 64 * 1024 * 1024;
2126 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
2127 max_chunk_size = 4 * calc_size;
2128 min_stripe_size = 32 * 1024 * 1024;
2129 } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
2130 calc_size = 8 * 1024 * 1024;
2131 max_chunk_size = calc_size * 2;
2132 min_stripe_size = 1 * 1024 * 1024;
2135 /* we don't want a chunk larger than 10% of writeable space */
2136 max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
2140 if (!map || map->num_stripes != num_stripes) {
2142 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
2145 map->num_stripes = num_stripes;
2148 if (calc_size * num_stripes > max_chunk_size) {
2149 calc_size = max_chunk_size;
2150 do_div(calc_size, num_stripes);
2151 do_div(calc_size, stripe_len);
2152 calc_size *= stripe_len;
2154 /* we don't want tiny stripes */
2155 calc_size = max_t(u64, min_stripe_size, calc_size);
2157 do_div(calc_size, stripe_len);
2158 calc_size *= stripe_len;
2160 cur = fs_devices->alloc_list.next;
2163 if (type & BTRFS_BLOCK_GROUP_DUP)
2164 min_free = calc_size * 2;
2166 min_free = calc_size;
2169 * we add 1MB because we never use the first 1MB of the device, unless
2170 * we've looped, then we are likely allocating the maximum amount of
2171 * space left already
2174 min_free += 1024 * 1024;
2176 INIT_LIST_HEAD(&private_devs);
2177 while (index < num_stripes) {
2178 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
2179 BUG_ON(!device->writeable);
2180 if (device->total_bytes > device->bytes_used)
2181 avail = device->total_bytes - device->bytes_used;
2186 if (device->in_fs_metadata && avail >= min_free) {
2187 ret = find_free_dev_extent(trans, device,
2188 min_free, &dev_offset);
2190 list_move_tail(&device->dev_alloc_list,
2192 map->stripes[index].dev = device;
2193 map->stripes[index].physical = dev_offset;
2195 if (type & BTRFS_BLOCK_GROUP_DUP) {
2196 map->stripes[index].dev = device;
2197 map->stripes[index].physical =
2198 dev_offset + calc_size;
2202 } else if (device->in_fs_metadata && avail > max_avail)
2204 if (cur == &fs_devices->alloc_list)
2207 list_splice(&private_devs, &fs_devices->alloc_list);
2208 if (index < num_stripes) {
2209 if (index >= min_stripes) {
2210 num_stripes = index;
2211 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
2212 num_stripes /= sub_stripes;
2213 num_stripes *= sub_stripes;
2218 if (!looped && max_avail > 0) {
2220 calc_size = max_avail;
2226 map->sector_size = extent_root->sectorsize;
2227 map->stripe_len = stripe_len;
2228 map->io_align = stripe_len;
2229 map->io_width = stripe_len;
2231 map->num_stripes = num_stripes;
2232 map->sub_stripes = sub_stripes;
2235 *stripe_size = calc_size;
2236 *num_bytes = chunk_bytes_by_type(type, calc_size,
2237 num_stripes, sub_stripes);
2239 em = alloc_extent_map(GFP_NOFS);
2244 em->bdev = (struct block_device *)map;
2246 em->len = *num_bytes;
2247 em->block_start = 0;
2248 em->block_len = em->len;
2250 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
2251 spin_lock(&em_tree->lock);
2252 ret = add_extent_mapping(em_tree, em);
2253 spin_unlock(&em_tree->lock);
2255 free_extent_map(em);
2257 ret = btrfs_make_block_group(trans, extent_root, 0, type,
2258 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2263 while (index < map->num_stripes) {
2264 device = map->stripes[index].dev;
2265 dev_offset = map->stripes[index].physical;
2267 ret = btrfs_alloc_dev_extent(trans, device,
2268 info->chunk_root->root_key.objectid,
2269 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2270 start, dev_offset, calc_size);
2278 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
2279 struct btrfs_root *extent_root,
2280 struct map_lookup *map, u64 chunk_offset,
2281 u64 chunk_size, u64 stripe_size)
2284 struct btrfs_key key;
2285 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2286 struct btrfs_device *device;
2287 struct btrfs_chunk *chunk;
2288 struct btrfs_stripe *stripe;
2289 size_t item_size = btrfs_chunk_item_size(map->num_stripes);
2293 chunk = kzalloc(item_size, GFP_NOFS);
2298 while (index < map->num_stripes) {
2299 device = map->stripes[index].dev;
2300 device->bytes_used += stripe_size;
2301 ret = btrfs_update_device(trans, device);
2307 stripe = &chunk->stripe;
2308 while (index < map->num_stripes) {
2309 device = map->stripes[index].dev;
2310 dev_offset = map->stripes[index].physical;
2312 btrfs_set_stack_stripe_devid(stripe, device->devid);
2313 btrfs_set_stack_stripe_offset(stripe, dev_offset);
2314 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
2319 btrfs_set_stack_chunk_length(chunk, chunk_size);
2320 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
2321 btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
2322 btrfs_set_stack_chunk_type(chunk, map->type);
2323 btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
2324 btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
2325 btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
2326 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
2327 btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
2329 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2330 key.type = BTRFS_CHUNK_ITEM_KEY;
2331 key.offset = chunk_offset;
2333 ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
2336 if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2337 ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk,
2346 * Chunk allocation falls into two parts. The first part does works
2347 * that make the new allocated chunk useable, but not do any operation
2348 * that modifies the chunk tree. The second part does the works that
2349 * require modifying the chunk tree. This division is important for the
2350 * bootstrap process of adding storage to a seed btrfs.
2352 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
2353 struct btrfs_root *extent_root, u64 type)
2358 struct map_lookup *map;
2359 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
2362 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2367 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2368 &stripe_size, chunk_offset, type);
2372 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2373 chunk_size, stripe_size);
2378 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
2379 struct btrfs_root *root,
2380 struct btrfs_device *device)
2383 u64 sys_chunk_offset;
2387 u64 sys_stripe_size;
2389 struct map_lookup *map;
2390 struct map_lookup *sys_map;
2391 struct btrfs_fs_info *fs_info = root->fs_info;
2392 struct btrfs_root *extent_root = fs_info->extent_root;
2395 ret = find_next_chunk(fs_info->chunk_root,
2396 BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
2399 alloc_profile = BTRFS_BLOCK_GROUP_METADATA |
2400 (fs_info->metadata_alloc_profile &
2401 fs_info->avail_metadata_alloc_bits);
2402 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2404 ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
2405 &stripe_size, chunk_offset, alloc_profile);
2408 sys_chunk_offset = chunk_offset + chunk_size;
2410 alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM |
2411 (fs_info->system_alloc_profile &
2412 fs_info->avail_system_alloc_bits);
2413 alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile);
2415 ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
2416 &sys_chunk_size, &sys_stripe_size,
2417 sys_chunk_offset, alloc_profile);
2420 ret = btrfs_add_device(trans, fs_info->chunk_root, device);
2424 * Modifying chunk tree needs allocating new blocks from both
2425 * system block group and metadata block group. So we only can
2426 * do operations require modifying the chunk tree after both
2427 * block groups were created.
2429 ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
2430 chunk_size, stripe_size);
2433 ret = __finish_chunk_alloc(trans, extent_root, sys_map,
2434 sys_chunk_offset, sys_chunk_size,
2440 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
2442 struct extent_map *em;
2443 struct map_lookup *map;
2444 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
2448 spin_lock(&map_tree->map_tree.lock);
2449 em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
2450 spin_unlock(&map_tree->map_tree.lock);
2454 map = (struct map_lookup *)em->bdev;
2455 for (i = 0; i < map->num_stripes; i++) {
2456 if (!map->stripes[i].dev->writeable) {
2461 free_extent_map(em);
2465 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
2467 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
2470 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
2472 struct extent_map *em;
2475 spin_lock(&tree->map_tree.lock);
2476 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
2478 remove_extent_mapping(&tree->map_tree, em);
2479 spin_unlock(&tree->map_tree.lock);
2484 free_extent_map(em);
2485 /* once for the tree */
2486 free_extent_map(em);
2490 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
2492 struct extent_map *em;
2493 struct map_lookup *map;
2494 struct extent_map_tree *em_tree = &map_tree->map_tree;
2497 spin_lock(&em_tree->lock);
2498 em = lookup_extent_mapping(em_tree, logical, len);
2499 spin_unlock(&em_tree->lock);
2502 BUG_ON(em->start > logical || em->start + em->len < logical);
2503 map = (struct map_lookup *)em->bdev;
2504 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
2505 ret = map->num_stripes;
2506 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2507 ret = map->sub_stripes;
2510 free_extent_map(em);
2514 static int find_live_mirror(struct map_lookup *map, int first, int num,
2518 if (map->stripes[optimal].dev->bdev)
2520 for (i = first; i < first + num; i++) {
2521 if (map->stripes[i].dev->bdev)
2524 /* we couldn't find one that doesn't fail. Just return something
2525 * and the io error handling code will clean up eventually
2530 static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2531 u64 logical, u64 *length,
2532 struct btrfs_multi_bio **multi_ret,
2533 int mirror_num, struct page *unplug_page)
2535 struct extent_map *em;
2536 struct map_lookup *map;
2537 struct extent_map_tree *em_tree = &map_tree->map_tree;
2541 int stripes_allocated = 8;
2542 int stripes_required = 1;
2547 struct btrfs_multi_bio *multi = NULL;
2549 if (multi_ret && !(rw & (1 << BIO_RW)))
2550 stripes_allocated = 1;
2553 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
2558 atomic_set(&multi->error, 0);
2561 spin_lock(&em_tree->lock);
2562 em = lookup_extent_mapping(em_tree, logical, *length);
2563 spin_unlock(&em_tree->lock);
2565 if (!em && unplug_page)
2569 printk(KERN_CRIT "unable to find logical %llu len %llu\n",
2570 (unsigned long long)logical,
2571 (unsigned long long)*length);
2575 BUG_ON(em->start > logical || em->start + em->len < logical);
2576 map = (struct map_lookup *)em->bdev;
2577 offset = logical - em->start;
2579 if (mirror_num > map->num_stripes)
2582 /* if our multi bio struct is too small, back off and try again */
2583 if (rw & (1 << BIO_RW)) {
2584 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
2585 BTRFS_BLOCK_GROUP_DUP)) {
2586 stripes_required = map->num_stripes;
2588 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2589 stripes_required = map->sub_stripes;
2593 if (multi_ret && (rw & (1 << BIO_RW)) &&
2594 stripes_allocated < stripes_required) {
2595 stripes_allocated = map->num_stripes;
2596 free_extent_map(em);
2602 * stripe_nr counts the total number of stripes we have to stride
2603 * to get to this block
2605 do_div(stripe_nr, map->stripe_len);
2607 stripe_offset = stripe_nr * map->stripe_len;
2608 BUG_ON(offset < stripe_offset);
2610 /* stripe_offset is the offset of this block in its stripe*/
2611 stripe_offset = offset - stripe_offset;
2613 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
2614 BTRFS_BLOCK_GROUP_RAID10 |
2615 BTRFS_BLOCK_GROUP_DUP)) {
2616 /* we limit the length of each bio to what fits in a stripe */
2617 *length = min_t(u64, em->len - offset,
2618 map->stripe_len - stripe_offset);
2620 *length = em->len - offset;
2623 if (!multi_ret && !unplug_page)
2628 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
2629 if (unplug_page || (rw & (1 << BIO_RW)))
2630 num_stripes = map->num_stripes;
2631 else if (mirror_num)
2632 stripe_index = mirror_num - 1;
2634 stripe_index = find_live_mirror(map, 0,
2636 current->pid % map->num_stripes);
2639 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
2640 if (rw & (1 << BIO_RW))
2641 num_stripes = map->num_stripes;
2642 else if (mirror_num)
2643 stripe_index = mirror_num - 1;
2645 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2646 int factor = map->num_stripes / map->sub_stripes;
2648 stripe_index = do_div(stripe_nr, factor);
2649 stripe_index *= map->sub_stripes;
2651 if (unplug_page || (rw & (1 << BIO_RW)))
2652 num_stripes = map->sub_stripes;
2653 else if (mirror_num)
2654 stripe_index += mirror_num - 1;
2656 stripe_index = find_live_mirror(map, stripe_index,
2657 map->sub_stripes, stripe_index +
2658 current->pid % map->sub_stripes);
2662 * after this do_div call, stripe_nr is the number of stripes
2663 * on this device we have to walk to find the data, and
2664 * stripe_index is the number of our device in the stripe array
2666 stripe_index = do_div(stripe_nr, map->num_stripes);
2668 BUG_ON(stripe_index >= map->num_stripes);
2670 for (i = 0; i < num_stripes; i++) {
2672 struct btrfs_device *device;
2673 struct backing_dev_info *bdi;
2675 device = map->stripes[stripe_index].dev;
2677 bdi = blk_get_backing_dev_info(device->bdev);
2678 if (bdi->unplug_io_fn)
2679 bdi->unplug_io_fn(bdi, unplug_page);
2682 multi->stripes[i].physical =
2683 map->stripes[stripe_index].physical +
2684 stripe_offset + stripe_nr * map->stripe_len;
2685 multi->stripes[i].dev = map->stripes[stripe_index].dev;
2691 multi->num_stripes = num_stripes;
2692 multi->max_errors = max_errors;
2695 free_extent_map(em);
2699 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
2700 u64 logical, u64 *length,
2701 struct btrfs_multi_bio **multi_ret, int mirror_num)
2703 return __btrfs_map_block(map_tree, rw, logical, length, multi_ret,
2707 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
2708 u64 chunk_start, u64 physical, u64 devid,
2709 u64 **logical, int *naddrs, int *stripe_len)
2711 struct extent_map_tree *em_tree = &map_tree->map_tree;
2712 struct extent_map *em;
2713 struct map_lookup *map;
2720 spin_lock(&em_tree->lock);
2721 em = lookup_extent_mapping(em_tree, chunk_start, 1);
2722 spin_unlock(&em_tree->lock);
2724 BUG_ON(!em || em->start != chunk_start);
2725 map = (struct map_lookup *)em->bdev;
2728 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
2729 do_div(length, map->num_stripes / map->sub_stripes);
2730 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
2731 do_div(length, map->num_stripes);
2733 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
2736 for (i = 0; i < map->num_stripes; i++) {
2737 if (devid && map->stripes[i].dev->devid != devid)
2739 if (map->stripes[i].physical > physical ||
2740 map->stripes[i].physical + length <= physical)
2743 stripe_nr = physical - map->stripes[i].physical;
2744 do_div(stripe_nr, map->stripe_len);
2746 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
2747 stripe_nr = stripe_nr * map->num_stripes + i;
2748 do_div(stripe_nr, map->sub_stripes);
2749 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
2750 stripe_nr = stripe_nr * map->num_stripes + i;
2752 bytenr = chunk_start + stripe_nr * map->stripe_len;
2753 WARN_ON(nr >= map->num_stripes);
2754 for (j = 0; j < nr; j++) {
2755 if (buf[j] == bytenr)
2759 WARN_ON(nr >= map->num_stripes);
2764 for (i = 0; i > nr; i++) {
2765 struct btrfs_multi_bio *multi;
2766 struct btrfs_bio_stripe *stripe;
2770 ret = btrfs_map_block(map_tree, WRITE, buf[i],
2771 &length, &multi, 0);
2774 stripe = multi->stripes;
2775 for (j = 0; j < multi->num_stripes; j++) {
2776 if (stripe->physical >= physical &&
2777 physical < stripe->physical + length)
2780 BUG_ON(j >= multi->num_stripes);
2786 *stripe_len = map->stripe_len;
2788 free_extent_map(em);
2792 int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree,
2793 u64 logical, struct page *page)
2795 u64 length = PAGE_CACHE_SIZE;
2796 return __btrfs_map_block(map_tree, READ, logical, &length,
2800 static void end_bio_multi_stripe(struct bio *bio, int err)
2802 struct btrfs_multi_bio *multi = bio->bi_private;
2803 int is_orig_bio = 0;
2806 atomic_inc(&multi->error);
2808 if (bio == multi->orig_bio)
2811 if (atomic_dec_and_test(&multi->stripes_pending)) {
2814 bio = multi->orig_bio;
2816 bio->bi_private = multi->private;
2817 bio->bi_end_io = multi->end_io;
2818 /* only send an error to the higher layers if it is
2819 * beyond the tolerance of the multi-bio
2821 if (atomic_read(&multi->error) > multi->max_errors) {
2825 * this bio is actually up to date, we didn't
2826 * go over the max number of errors
2828 set_bit(BIO_UPTODATE, &bio->bi_flags);
2833 bio_endio(bio, err);
2834 } else if (!is_orig_bio) {
2839 struct async_sched {
2842 struct btrfs_fs_info *info;
2843 struct btrfs_work work;
2847 * see run_scheduled_bios for a description of why bios are collected for
2850 * This will add one bio to the pending list for a device and make sure
2851 * the work struct is scheduled.
2853 static noinline int schedule_bio(struct btrfs_root *root,
2854 struct btrfs_device *device,
2855 int rw, struct bio *bio)
2857 int should_queue = 1;
2858 struct btrfs_pending_bios *pending_bios;
2860 /* don't bother with additional async steps for reads, right now */
2861 if (!(rw & (1 << BIO_RW))) {
2863 submit_bio(rw, bio);
2869 * nr_async_bios allows us to reliably return congestion to the
2870 * higher layers. Otherwise, the async bio makes it appear we have
2871 * made progress against dirty pages when we've really just put it
2872 * on a queue for later
2874 atomic_inc(&root->fs_info->nr_async_bios);
2875 WARN_ON(bio->bi_next);
2876 bio->bi_next = NULL;
2879 spin_lock(&device->io_lock);
2881 pending_bios = &device->pending_sync_bios;
2883 pending_bios = &device->pending_bios;
2885 if (pending_bios->tail)
2886 pending_bios->tail->bi_next = bio;
2888 pending_bios->tail = bio;
2889 if (!pending_bios->head)
2890 pending_bios->head = bio;
2891 if (device->running_pending)
2894 spin_unlock(&device->io_lock);
2897 btrfs_queue_worker(&root->fs_info->submit_workers,
2902 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
2903 int mirror_num, int async_submit)
2905 struct btrfs_mapping_tree *map_tree;
2906 struct btrfs_device *dev;
2907 struct bio *first_bio = bio;
2908 u64 logical = (u64)bio->bi_sector << 9;
2911 struct btrfs_multi_bio *multi = NULL;
2916 length = bio->bi_size;
2917 map_tree = &root->fs_info->mapping_tree;
2918 map_length = length;
2920 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
2924 total_devs = multi->num_stripes;
2925 if (map_length < length) {
2926 printk(KERN_CRIT "mapping failed logical %llu bio len %llu "
2927 "len %llu\n", (unsigned long long)logical,
2928 (unsigned long long)length,
2929 (unsigned long long)map_length);
2932 multi->end_io = first_bio->bi_end_io;
2933 multi->private = first_bio->bi_private;
2934 multi->orig_bio = first_bio;
2935 atomic_set(&multi->stripes_pending, multi->num_stripes);
2937 while (dev_nr < total_devs) {
2938 if (total_devs > 1) {
2939 if (dev_nr < total_devs - 1) {
2940 bio = bio_clone(first_bio, GFP_NOFS);
2945 bio->bi_private = multi;
2946 bio->bi_end_io = end_bio_multi_stripe;
2948 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
2949 dev = multi->stripes[dev_nr].dev;
2950 BUG_ON(rw == WRITE && !dev->writeable);
2951 if (dev && dev->bdev) {
2952 bio->bi_bdev = dev->bdev;
2954 schedule_bio(root, dev, rw, bio);
2956 submit_bio(rw, bio);
2958 bio->bi_bdev = root->fs_info->fs_devices->latest_bdev;
2959 bio->bi_sector = logical >> 9;
2960 bio_endio(bio, -EIO);
2964 if (total_devs == 1)
2969 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
2972 struct btrfs_device *device;
2973 struct btrfs_fs_devices *cur_devices;
2975 cur_devices = root->fs_info->fs_devices;
2976 while (cur_devices) {
2978 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
2979 device = __find_device(&cur_devices->devices,
2984 cur_devices = cur_devices->seed;
2989 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
2990 u64 devid, u8 *dev_uuid)
2992 struct btrfs_device *device;
2993 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
2995 device = kzalloc(sizeof(*device), GFP_NOFS);
2998 list_add(&device->dev_list,
2999 &fs_devices->devices);
3000 device->barriers = 1;
3001 device->dev_root = root->fs_info->dev_root;
3002 device->devid = devid;
3003 device->work.func = pending_bios_fn;
3004 device->fs_devices = fs_devices;
3005 fs_devices->num_devices++;
3006 spin_lock_init(&device->io_lock);
3007 INIT_LIST_HEAD(&device->dev_alloc_list);
3008 memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
3012 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
3013 struct extent_buffer *leaf,
3014 struct btrfs_chunk *chunk)
3016 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
3017 struct map_lookup *map;
3018 struct extent_map *em;
3022 u8 uuid[BTRFS_UUID_SIZE];
3027 logical = key->offset;
3028 length = btrfs_chunk_length(leaf, chunk);
3030 spin_lock(&map_tree->map_tree.lock);
3031 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
3032 spin_unlock(&map_tree->map_tree.lock);
3034 /* already mapped? */
3035 if (em && em->start <= logical && em->start + em->len > logical) {
3036 free_extent_map(em);
3039 free_extent_map(em);
3042 em = alloc_extent_map(GFP_NOFS);
3045 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
3046 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3048 free_extent_map(em);
3052 em->bdev = (struct block_device *)map;
3053 em->start = logical;
3055 em->block_start = 0;
3056 em->block_len = em->len;
3058 map->num_stripes = num_stripes;
3059 map->io_width = btrfs_chunk_io_width(leaf, chunk);
3060 map->io_align = btrfs_chunk_io_align(leaf, chunk);
3061 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
3062 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
3063 map->type = btrfs_chunk_type(leaf, chunk);
3064 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
3065 for (i = 0; i < num_stripes; i++) {
3066 map->stripes[i].physical =
3067 btrfs_stripe_offset_nr(leaf, chunk, i);
3068 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
3069 read_extent_buffer(leaf, uuid, (unsigned long)
3070 btrfs_stripe_dev_uuid_nr(chunk, i),
3072 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
3074 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
3076 free_extent_map(em);
3079 if (!map->stripes[i].dev) {
3080 map->stripes[i].dev =
3081 add_missing_dev(root, devid, uuid);
3082 if (!map->stripes[i].dev) {
3084 free_extent_map(em);
3088 map->stripes[i].dev->in_fs_metadata = 1;
3091 spin_lock(&map_tree->map_tree.lock);
3092 ret = add_extent_mapping(&map_tree->map_tree, em);
3093 spin_unlock(&map_tree->map_tree.lock);
3095 free_extent_map(em);
3100 static int fill_device_from_item(struct extent_buffer *leaf,
3101 struct btrfs_dev_item *dev_item,
3102 struct btrfs_device *device)
3106 device->devid = btrfs_device_id(leaf, dev_item);
3107 device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
3108 device->total_bytes = device->disk_total_bytes;
3109 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
3110 device->type = btrfs_device_type(leaf, dev_item);
3111 device->io_align = btrfs_device_io_align(leaf, dev_item);
3112 device->io_width = btrfs_device_io_width(leaf, dev_item);
3113 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
3115 ptr = (unsigned long)btrfs_device_uuid(dev_item);
3116 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
3121 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
3123 struct btrfs_fs_devices *fs_devices;
3126 mutex_lock(&uuid_mutex);
3128 fs_devices = root->fs_info->fs_devices->seed;
3129 while (fs_devices) {
3130 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
3134 fs_devices = fs_devices->seed;
3137 fs_devices = find_fsid(fsid);
3143 fs_devices = clone_fs_devices(fs_devices);
3144 if (IS_ERR(fs_devices)) {
3145 ret = PTR_ERR(fs_devices);
3149 ret = __btrfs_open_devices(fs_devices, FMODE_READ,
3150 root->fs_info->bdev_holder);
3154 if (!fs_devices->seeding) {
3155 __btrfs_close_devices(fs_devices);
3156 free_fs_devices(fs_devices);
3161 fs_devices->seed = root->fs_info->fs_devices->seed;
3162 root->fs_info->fs_devices->seed = fs_devices;
3164 mutex_unlock(&uuid_mutex);
3168 static int read_one_dev(struct btrfs_root *root,
3169 struct extent_buffer *leaf,
3170 struct btrfs_dev_item *dev_item)
3172 struct btrfs_device *device;
3175 u8 fs_uuid[BTRFS_UUID_SIZE];
3176 u8 dev_uuid[BTRFS_UUID_SIZE];
3178 devid = btrfs_device_id(leaf, dev_item);
3179 read_extent_buffer(leaf, dev_uuid,
3180 (unsigned long)btrfs_device_uuid(dev_item),
3182 read_extent_buffer(leaf, fs_uuid,
3183 (unsigned long)btrfs_device_fsid(dev_item),
3186 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
3187 ret = open_seed_devices(root, fs_uuid);
3188 if (ret && !btrfs_test_opt(root, DEGRADED))
3192 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
3193 if (!device || !device->bdev) {
3194 if (!btrfs_test_opt(root, DEGRADED))
3198 printk(KERN_WARNING "warning devid %llu missing\n",
3199 (unsigned long long)devid);
3200 device = add_missing_dev(root, devid, dev_uuid);
3206 if (device->fs_devices != root->fs_info->fs_devices) {
3207 BUG_ON(device->writeable);
3208 if (device->generation !=
3209 btrfs_device_generation(leaf, dev_item))
3213 fill_device_from_item(leaf, dev_item, device);
3214 device->dev_root = root->fs_info->dev_root;
3215 device->in_fs_metadata = 1;
3216 if (device->writeable)
3217 device->fs_devices->total_rw_bytes += device->total_bytes;
3222 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
3224 struct btrfs_dev_item *dev_item;
3226 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
3228 return read_one_dev(root, buf, dev_item);
3231 int btrfs_read_sys_array(struct btrfs_root *root)
3233 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
3234 struct extent_buffer *sb;
3235 struct btrfs_disk_key *disk_key;
3236 struct btrfs_chunk *chunk;
3238 unsigned long sb_ptr;
3244 struct btrfs_key key;
3246 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
3247 BTRFS_SUPER_INFO_SIZE);
3250 btrfs_set_buffer_uptodate(sb);
3251 btrfs_set_buffer_lockdep_class(sb, 0);
3253 write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
3254 array_size = btrfs_super_sys_array_size(super_copy);
3256 ptr = super_copy->sys_chunk_array;
3257 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
3260 while (cur < array_size) {
3261 disk_key = (struct btrfs_disk_key *)ptr;
3262 btrfs_disk_key_to_cpu(&key, disk_key);
3264 len = sizeof(*disk_key); ptr += len;
3268 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
3269 chunk = (struct btrfs_chunk *)sb_ptr;
3270 ret = read_one_chunk(root, &key, sb, chunk);
3273 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
3274 len = btrfs_chunk_item_size(num_stripes);
3283 free_extent_buffer(sb);
3287 int btrfs_read_chunk_tree(struct btrfs_root *root)
3289 struct btrfs_path *path;
3290 struct extent_buffer *leaf;
3291 struct btrfs_key key;
3292 struct btrfs_key found_key;
3296 root = root->fs_info->chunk_root;
3298 path = btrfs_alloc_path();
3302 /* first we search for all of the device items, and then we
3303 * read in all of the chunk items. This way we can create chunk
3304 * mappings that reference all of the devices that are afound
3306 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
3310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3312 leaf = path->nodes[0];
3313 slot = path->slots[0];
3314 if (slot >= btrfs_header_nritems(leaf)) {
3315 ret = btrfs_next_leaf(root, path);
3322 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3323 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3324 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
3326 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
3327 struct btrfs_dev_item *dev_item;
3328 dev_item = btrfs_item_ptr(leaf, slot,
3329 struct btrfs_dev_item);
3330 ret = read_one_dev(root, leaf, dev_item);
3334 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
3335 struct btrfs_chunk *chunk;
3336 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
3337 ret = read_one_chunk(root, &found_key, leaf, chunk);
3343 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
3345 btrfs_release_path(root, path);
3350 btrfs_free_path(path);