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 <asm/div64.h>
23 #include "extent_map.h"
25 #include "transaction.h"
26 #include "print-tree.h"
36 struct btrfs_bio_stripe stripes[];
39 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
40 (sizeof(struct btrfs_bio_stripe) * (n)))
42 static DEFINE_MUTEX(uuid_mutex);
43 static LIST_HEAD(fs_uuids);
45 int btrfs_cleanup_fs_uuids(void)
47 struct btrfs_fs_devices *fs_devices;
48 struct list_head *uuid_cur;
49 struct list_head *devices_cur;
50 struct btrfs_device *dev;
52 list_for_each(uuid_cur, &fs_uuids) {
53 fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices,
55 while(!list_empty(&fs_devices->devices)) {
56 devices_cur = fs_devices->devices.next;
57 dev = list_entry(devices_cur, struct btrfs_device,
59 printk("uuid cleanup finds %s\n", dev->name);
62 close_bdev_excl(dev->bdev);
64 list_del(&dev->dev_list);
71 static struct btrfs_device *__find_device(struct list_head *head, u64 devid)
73 struct btrfs_device *dev;
74 struct list_head *cur;
76 list_for_each(cur, head) {
77 dev = list_entry(cur, struct btrfs_device, dev_list);
78 if (dev->devid == devid)
84 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
86 struct list_head *cur;
87 struct btrfs_fs_devices *fs_devices;
89 list_for_each(cur, &fs_uuids) {
90 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
91 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
97 static int device_list_add(const char *path,
98 struct btrfs_super_block *disk_super,
99 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
101 struct btrfs_device *device;
102 struct btrfs_fs_devices *fs_devices;
103 u64 found_transid = btrfs_super_generation(disk_super);
105 fs_devices = find_fsid(disk_super->fsid);
107 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
110 INIT_LIST_HEAD(&fs_devices->devices);
111 list_add(&fs_devices->list, &fs_uuids);
112 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
113 fs_devices->latest_devid = devid;
114 fs_devices->latest_trans = found_transid;
115 fs_devices->lowest_devid = (u64)-1;
116 fs_devices->num_devices = 0;
119 device = __find_device(&fs_devices->devices, devid);
122 device = kzalloc(sizeof(*device), GFP_NOFS);
124 /* we can safely leave the fs_devices entry around */
127 device->devid = devid;
128 device->barriers = 1;
129 spin_lock_init(&device->io_lock);
130 device->name = kstrdup(path, GFP_NOFS);
135 list_add(&device->dev_list, &fs_devices->devices);
136 fs_devices->num_devices++;
139 if (found_transid > fs_devices->latest_trans) {
140 fs_devices->latest_devid = devid;
141 fs_devices->latest_trans = found_transid;
143 if (fs_devices->lowest_devid > devid) {
144 fs_devices->lowest_devid = devid;
145 printk("lowest devid now %Lu\n", devid);
147 *fs_devices_ret = fs_devices;
151 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
153 struct list_head *head = &fs_devices->devices;
154 struct list_head *cur;
155 struct btrfs_device *device;
157 mutex_lock(&uuid_mutex);
158 list_for_each(cur, head) {
159 device = list_entry(cur, struct btrfs_device, dev_list);
161 close_bdev_excl(device->bdev);
162 printk("close devices closes %s\n", device->name);
166 mutex_unlock(&uuid_mutex);
170 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
171 int flags, void *holder)
173 struct block_device *bdev;
174 struct list_head *head = &fs_devices->devices;
175 struct list_head *cur;
176 struct btrfs_device *device;
179 mutex_lock(&uuid_mutex);
180 list_for_each(cur, head) {
181 device = list_entry(cur, struct btrfs_device, dev_list);
182 bdev = open_bdev_excl(device->name, flags, holder);
183 printk("opening %s devid %Lu\n", device->name, device->devid);
185 printk("open %s failed\n", device->name);
189 if (device->devid == fs_devices->latest_devid)
190 fs_devices->latest_bdev = bdev;
191 if (device->devid == fs_devices->lowest_devid) {
192 fs_devices->lowest_bdev = bdev;
193 printk("lowest bdev %s\n", device->name);
197 mutex_unlock(&uuid_mutex);
200 mutex_unlock(&uuid_mutex);
201 btrfs_close_devices(fs_devices);
205 int btrfs_scan_one_device(const char *path, int flags, void *holder,
206 struct btrfs_fs_devices **fs_devices_ret)
208 struct btrfs_super_block *disk_super;
209 struct block_device *bdev;
210 struct buffer_head *bh;
215 mutex_lock(&uuid_mutex);
217 printk("scan one opens %s\n", path);
218 bdev = open_bdev_excl(path, flags, holder);
221 printk("open failed\n");
226 ret = set_blocksize(bdev, 4096);
229 bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
234 disk_super = (struct btrfs_super_block *)bh->b_data;
235 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
236 sizeof(disk_super->magic))) {
237 printk("no btrfs found on %s\n", path);
241 devid = le64_to_cpu(disk_super->dev_item.devid);
242 transid = btrfs_super_generation(disk_super);
243 printk("found device %Lu transid %Lu on %s\n", devid, transid, path);
244 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
249 close_bdev_excl(bdev);
251 mutex_unlock(&uuid_mutex);
256 * this uses a pretty simple search, the expectation is that it is
257 * called very infrequently and that a given device has a small number
260 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
261 struct btrfs_device *device,
262 struct btrfs_path *path,
263 u64 num_bytes, u64 *start)
265 struct btrfs_key key;
266 struct btrfs_root *root = device->dev_root;
267 struct btrfs_dev_extent *dev_extent = NULL;
270 u64 search_start = 0;
271 u64 search_end = device->total_bytes;
275 struct extent_buffer *l;
280 /* FIXME use last free of some kind */
282 /* we don't want to overwrite the superblock on the drive,
283 * so we make sure to start at an offset of at least 1MB
285 search_start = max((u64)1024 * 1024, search_start);
286 key.objectid = device->devid;
287 key.offset = search_start;
288 key.type = BTRFS_DEV_EXTENT_KEY;
289 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
292 ret = btrfs_previous_item(root, path, 0, key.type);
296 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
299 slot = path->slots[0];
300 if (slot >= btrfs_header_nritems(l)) {
301 ret = btrfs_next_leaf(root, path);
308 if (search_start >= search_end) {
312 *start = search_start;
316 *start = last_byte > search_start ?
317 last_byte : search_start;
318 if (search_end <= *start) {
324 btrfs_item_key_to_cpu(l, &key, slot);
326 if (key.objectid < device->devid)
329 if (key.objectid > device->devid)
332 if (key.offset >= search_start && key.offset > last_byte &&
334 if (last_byte < search_start)
335 last_byte = search_start;
336 hole_size = key.offset - last_byte;
337 if (key.offset > last_byte &&
338 hole_size >= num_bytes) {
343 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
348 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
349 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
355 /* we have to make sure we didn't find an extent that has already
356 * been allocated by the map tree or the original allocation
358 btrfs_release_path(root, path);
359 BUG_ON(*start < search_start);
361 if (*start + num_bytes > search_end) {
365 /* check for pending inserts here */
369 btrfs_release_path(root, path);
373 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
374 struct btrfs_device *device,
375 u64 owner, u64 num_bytes, u64 *start)
378 struct btrfs_path *path;
379 struct btrfs_root *root = device->dev_root;
380 struct btrfs_dev_extent *extent;
381 struct extent_buffer *leaf;
382 struct btrfs_key key;
384 path = btrfs_alloc_path();
388 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
393 key.objectid = device->devid;
395 key.type = BTRFS_DEV_EXTENT_KEY;
396 ret = btrfs_insert_empty_item(trans, root, path, &key,
400 leaf = path->nodes[0];
401 extent = btrfs_item_ptr(leaf, path->slots[0],
402 struct btrfs_dev_extent);
403 btrfs_set_dev_extent_owner(leaf, extent, owner);
404 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
405 btrfs_mark_buffer_dirty(leaf);
407 btrfs_free_path(path);
411 static int find_next_chunk(struct btrfs_root *root, u64 *objectid)
413 struct btrfs_path *path;
415 struct btrfs_key key;
416 struct btrfs_key found_key;
418 path = btrfs_alloc_path();
421 key.objectid = (u64)-1;
422 key.offset = (u64)-1;
423 key.type = BTRFS_CHUNK_ITEM_KEY;
425 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
431 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
435 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
437 *objectid = found_key.objectid + found_key.offset;
441 btrfs_free_path(path);
445 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
449 struct btrfs_key key;
450 struct btrfs_key found_key;
452 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
453 key.type = BTRFS_DEV_ITEM_KEY;
454 key.offset = (u64)-1;
456 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
462 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
467 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
469 *objectid = found_key.offset + 1;
473 btrfs_release_path(root, path);
478 * the device information is stored in the chunk root
479 * the btrfs_device struct should be fully filled in
481 int btrfs_add_device(struct btrfs_trans_handle *trans,
482 struct btrfs_root *root,
483 struct btrfs_device *device)
486 struct btrfs_path *path;
487 struct btrfs_dev_item *dev_item;
488 struct extent_buffer *leaf;
489 struct btrfs_key key;
493 root = root->fs_info->chunk_root;
495 path = btrfs_alloc_path();
499 ret = find_next_devid(root, path, &free_devid);
503 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
504 key.type = BTRFS_DEV_ITEM_KEY;
505 key.offset = free_devid;
507 ret = btrfs_insert_empty_item(trans, root, path, &key,
512 leaf = path->nodes[0];
513 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
515 device->devid = free_devid;
516 btrfs_set_device_id(leaf, dev_item, device->devid);
517 btrfs_set_device_type(leaf, dev_item, device->type);
518 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
519 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
520 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
521 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
522 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
524 ptr = (unsigned long)btrfs_device_uuid(dev_item);
525 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
526 btrfs_mark_buffer_dirty(leaf);
530 btrfs_free_path(path);
533 int btrfs_update_device(struct btrfs_trans_handle *trans,
534 struct btrfs_device *device)
537 struct btrfs_path *path;
538 struct btrfs_root *root;
539 struct btrfs_dev_item *dev_item;
540 struct extent_buffer *leaf;
541 struct btrfs_key key;
543 root = device->dev_root->fs_info->chunk_root;
545 path = btrfs_alloc_path();
549 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
550 key.type = BTRFS_DEV_ITEM_KEY;
551 key.offset = device->devid;
553 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
562 leaf = path->nodes[0];
563 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
565 btrfs_set_device_id(leaf, dev_item, device->devid);
566 btrfs_set_device_type(leaf, dev_item, device->type);
567 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
568 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
569 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
570 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
571 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
572 btrfs_mark_buffer_dirty(leaf);
575 btrfs_free_path(path);
579 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
580 struct btrfs_root *root,
581 struct btrfs_key *key,
582 struct btrfs_chunk *chunk, int item_size)
584 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
585 struct btrfs_disk_key disk_key;
589 array_size = btrfs_super_sys_array_size(super_copy);
590 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
593 ptr = super_copy->sys_chunk_array + array_size;
594 btrfs_cpu_key_to_disk(&disk_key, key);
595 memcpy(ptr, &disk_key, sizeof(disk_key));
596 ptr += sizeof(disk_key);
597 memcpy(ptr, chunk, item_size);
598 item_size += sizeof(disk_key);
599 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
603 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
604 struct btrfs_root *extent_root, u64 *start,
605 u64 *num_bytes, u64 type)
608 struct btrfs_fs_info *info = extent_root->fs_info;
609 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
610 struct btrfs_stripe *stripes;
611 struct btrfs_device *device = NULL;
612 struct btrfs_chunk *chunk;
613 struct list_head private_devs;
614 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
615 struct list_head *cur;
616 struct extent_map_tree *em_tree;
617 struct map_lookup *map;
618 struct extent_map *em;
620 u64 calc_size = 1024 * 1024 * 1024;
621 u64 min_free = calc_size;
628 int stripe_len = 64 * 1024;
629 struct btrfs_key key;
631 if (list_empty(dev_list))
634 if (type & (BTRFS_BLOCK_GROUP_RAID0))
635 num_stripes = btrfs_super_num_devices(&info->super_copy);
636 if (type & (BTRFS_BLOCK_GROUP_DUP))
638 if (type & (BTRFS_BLOCK_GROUP_RAID1)) {
639 num_stripes = min_t(u64, 2,
640 btrfs_super_num_devices(&info->super_copy));
643 INIT_LIST_HEAD(&private_devs);
644 cur = dev_list->next;
647 if (type & BTRFS_BLOCK_GROUP_DUP)
648 min_free = calc_size * 2;
650 /* build a private list of devices we will allocate from */
651 while(index < num_stripes) {
652 device = list_entry(cur, struct btrfs_device, dev_list);
654 avail = device->total_bytes - device->bytes_used;
656 if (avail > max_avail)
658 if (avail >= min_free) {
659 list_move_tail(&device->dev_list, &private_devs);
661 if (type & BTRFS_BLOCK_GROUP_DUP)
667 if (index < num_stripes) {
668 list_splice(&private_devs, dev_list);
669 if (!looped && max_avail > 0) {
671 calc_size = max_avail;
677 ret = find_next_chunk(chunk_root, &key.objectid);
681 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
685 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
691 stripes = &chunk->stripe;
693 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
694 *num_bytes = calc_size;
696 *num_bytes = calc_size * num_stripes;
699 printk("new chunk type %Lu start %Lu size %Lu\n", type, key.objectid, *num_bytes);
700 while(index < num_stripes) {
701 BUG_ON(list_empty(&private_devs));
702 cur = private_devs.next;
703 device = list_entry(cur, struct btrfs_device, dev_list);
705 /* loop over this device again if we're doing a dup group */
706 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
707 (index == num_stripes - 1))
708 list_move_tail(&device->dev_list, dev_list);
710 ret = btrfs_alloc_dev_extent(trans, device,
712 calc_size, &dev_offset);
714 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.objectid, calc_size, device->devid, type);
715 device->bytes_used += calc_size;
716 ret = btrfs_update_device(trans, device);
719 map->stripes[index].dev = device;
720 map->stripes[index].physical = dev_offset;
721 btrfs_set_stack_stripe_devid(stripes + index, device->devid);
722 btrfs_set_stack_stripe_offset(stripes + index, dev_offset);
723 physical = dev_offset;
726 BUG_ON(!list_empty(&private_devs));
728 /* key.objectid was set above */
729 key.offset = *num_bytes;
730 key.type = BTRFS_CHUNK_ITEM_KEY;
731 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
732 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
733 btrfs_set_stack_chunk_type(chunk, type);
734 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
735 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
736 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
737 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
738 map->sector_size = extent_root->sectorsize;
739 map->stripe_len = stripe_len;
740 map->io_align = stripe_len;
741 map->io_width = stripe_len;
743 map->num_stripes = num_stripes;
745 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
746 btrfs_chunk_item_size(num_stripes));
748 *start = key.objectid;
750 em = alloc_extent_map(GFP_NOFS);
753 em->bdev = (struct block_device *)map;
754 em->start = key.objectid;
755 em->len = key.offset;
760 em_tree = &extent_root->fs_info->mapping_tree.map_tree;
761 spin_lock(&em_tree->lock);
762 ret = add_extent_mapping(em_tree, em);
763 spin_unlock(&em_tree->lock);
769 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
771 extent_map_tree_init(&tree->map_tree, GFP_NOFS);
774 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
776 struct extent_map *em;
779 spin_lock(&tree->map_tree.lock);
780 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
782 remove_extent_mapping(&tree->map_tree, em);
783 spin_unlock(&tree->map_tree.lock);
789 /* once for the tree */
794 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
796 struct extent_map *em;
797 struct map_lookup *map;
798 struct extent_map_tree *em_tree = &map_tree->map_tree;
801 spin_lock(&em_tree->lock);
802 em = lookup_extent_mapping(em_tree, logical, len);
803 spin_unlock(&em_tree->lock);
806 BUG_ON(em->start > logical || em->start + em->len < logical);
807 map = (struct map_lookup *)em->bdev;
808 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
809 ret = map->num_stripes;
816 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
817 u64 logical, u64 *length,
818 struct btrfs_multi_bio **multi_ret, int mirror_num)
820 struct extent_map *em;
821 struct map_lookup *map;
822 struct extent_map_tree *em_tree = &map_tree->map_tree;
826 int stripes_allocated = 8;
829 struct btrfs_multi_bio *multi = NULL;
831 if (multi_ret && !(rw & (1 << BIO_RW))) {
832 stripes_allocated = 1;
836 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
842 spin_lock(&em_tree->lock);
843 em = lookup_extent_mapping(em_tree, logical, *length);
844 spin_unlock(&em_tree->lock);
847 BUG_ON(em->start > logical || em->start + em->len < logical);
848 map = (struct map_lookup *)em->bdev;
849 offset = logical - em->start;
851 if (mirror_num > map->num_stripes)
854 /* if our multi bio struct is too small, back off and try again */
855 if (multi_ret && (rw & (1 << BIO_RW)) &&
856 stripes_allocated < map->num_stripes &&
857 ((map->type & BTRFS_BLOCK_GROUP_RAID1) ||
858 (map->type & BTRFS_BLOCK_GROUP_DUP))) {
859 stripes_allocated = map->num_stripes;
866 * stripe_nr counts the total number of stripes we have to stride
867 * to get to this block
869 do_div(stripe_nr, map->stripe_len);
871 stripe_offset = stripe_nr * map->stripe_len;
872 BUG_ON(offset < stripe_offset);
874 /* stripe_offset is the offset of this block in its stripe*/
875 stripe_offset = offset - stripe_offset;
877 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
878 BTRFS_BLOCK_GROUP_DUP)) {
879 /* we limit the length of each bio to what fits in a stripe */
880 *length = min_t(u64, em->len - offset,
881 map->stripe_len - stripe_offset);
883 *length = em->len - offset;
888 multi->num_stripes = 1;
890 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
891 if (rw & (1 << BIO_RW))
892 multi->num_stripes = map->num_stripes;
893 else if (mirror_num) {
894 stripe_index = mirror_num - 1;
898 struct btrfs_device *cur;
900 for (i = 0; i < map->num_stripes; i++) {
901 cur = map->stripes[i].dev;
902 spin_lock(&cur->io_lock);
903 if (cur->total_ios < least) {
904 least = cur->total_ios;
907 spin_unlock(&cur->io_lock);
910 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
911 if (rw & (1 << BIO_RW))
912 multi->num_stripes = map->num_stripes;
914 stripe_index = mirror_num - 1;
917 * after this do_div call, stripe_nr is the number of stripes
918 * on this device we have to walk to find the data, and
919 * stripe_index is the number of our device in the stripe array
921 stripe_index = do_div(stripe_nr, map->num_stripes);
923 BUG_ON(stripe_index >= map->num_stripes);
924 BUG_ON(stripe_index != 0 && multi->num_stripes > 1);
926 for (i = 0; i < multi->num_stripes; i++) {
927 multi->stripes[i].physical =
928 map->stripes[stripe_index].physical + stripe_offset +
929 stripe_nr * map->stripe_len;
930 multi->stripes[i].dev = map->stripes[stripe_index].dev;
939 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
940 static void end_bio_multi_stripe(struct bio *bio, int err)
942 static int end_bio_multi_stripe(struct bio *bio,
943 unsigned int bytes_done, int err)
946 struct btrfs_multi_bio *multi = bio->bi_private;
948 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
955 if (atomic_dec_and_test(&multi->stripes_pending)) {
956 bio->bi_private = multi->private;
957 bio->bi_end_io = multi->end_io;
959 if (!err && multi->error)
963 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
964 bio_endio(bio, bio->bi_size, err);
971 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
976 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
979 struct btrfs_mapping_tree *map_tree;
980 struct btrfs_device *dev;
981 struct bio *first_bio = bio;
982 u64 logical = bio->bi_sector << 9;
985 struct bio_vec *bvec;
986 struct btrfs_multi_bio *multi = NULL;
992 bio_for_each_segment(bvec, bio, i) {
993 length += bvec->bv_len;
996 map_tree = &root->fs_info->mapping_tree;
999 ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi,
1003 total_devs = multi->num_stripes;
1004 if (map_length < length) {
1005 printk("mapping failed logical %Lu bio len %Lu "
1006 "len %Lu\n", logical, length, map_length);
1009 multi->end_io = first_bio->bi_end_io;
1010 multi->private = first_bio->bi_private;
1011 atomic_set(&multi->stripes_pending, multi->num_stripes);
1013 while(dev_nr < total_devs) {
1014 if (total_devs > 1) {
1015 if (dev_nr < total_devs - 1) {
1016 bio = bio_clone(first_bio, GFP_NOFS);
1021 bio->bi_private = multi;
1022 bio->bi_end_io = end_bio_multi_stripe;
1024 bio->bi_sector = multi->stripes[dev_nr].physical >> 9;
1025 dev = multi->stripes[dev_nr].dev;
1026 bio->bi_bdev = dev->bdev;
1027 spin_lock(&dev->io_lock);
1029 spin_unlock(&dev->io_lock);
1030 submit_bio(rw, bio);
1033 if (total_devs == 1)
1038 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid)
1040 struct list_head *head = &root->fs_info->fs_devices->devices;
1042 return __find_device(head, devid);
1045 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1046 struct extent_buffer *leaf,
1047 struct btrfs_chunk *chunk)
1049 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1050 struct map_lookup *map;
1051 struct extent_map *em;
1059 logical = key->objectid;
1060 length = key->offset;
1061 spin_lock(&map_tree->map_tree.lock);
1062 em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
1063 spin_unlock(&map_tree->map_tree.lock);
1065 /* already mapped? */
1066 if (em && em->start <= logical && em->start + em->len > logical) {
1067 free_extent_map(em);
1070 free_extent_map(em);
1073 map = kzalloc(sizeof(*map), GFP_NOFS);
1077 em = alloc_extent_map(GFP_NOFS);
1080 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1081 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1083 free_extent_map(em);
1087 em->bdev = (struct block_device *)map;
1088 em->start = logical;
1090 em->block_start = 0;
1092 map->num_stripes = num_stripes;
1093 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1094 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1095 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1096 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1097 map->type = btrfs_chunk_type(leaf, chunk);
1098 for (i = 0; i < num_stripes; i++) {
1099 map->stripes[i].physical =
1100 btrfs_stripe_offset_nr(leaf, chunk, i);
1101 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1102 map->stripes[i].dev = btrfs_find_device(root, devid);
1103 if (!map->stripes[i].dev) {
1105 free_extent_map(em);
1110 spin_lock(&map_tree->map_tree.lock);
1111 ret = add_extent_mapping(&map_tree->map_tree, em);
1112 spin_unlock(&map_tree->map_tree.lock);
1114 free_extent_map(em);
1119 static int fill_device_from_item(struct extent_buffer *leaf,
1120 struct btrfs_dev_item *dev_item,
1121 struct btrfs_device *device)
1125 device->devid = btrfs_device_id(leaf, dev_item);
1126 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1127 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1128 device->type = btrfs_device_type(leaf, dev_item);
1129 device->io_align = btrfs_device_io_align(leaf, dev_item);
1130 device->io_width = btrfs_device_io_width(leaf, dev_item);
1131 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1133 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1134 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_DEV_UUID_SIZE);
1139 static int read_one_dev(struct btrfs_root *root,
1140 struct extent_buffer *leaf,
1141 struct btrfs_dev_item *dev_item)
1143 struct btrfs_device *device;
1147 devid = btrfs_device_id(leaf, dev_item);
1148 device = btrfs_find_device(root, devid);
1150 printk("warning devid %Lu not found already\n", devid);
1151 device = kzalloc(sizeof(*device), GFP_NOFS);
1154 list_add(&device->dev_list,
1155 &root->fs_info->fs_devices->devices);
1156 device->barriers = 1;
1157 spin_lock_init(&device->io_lock);
1160 fill_device_from_item(leaf, dev_item, device);
1161 device->dev_root = root->fs_info->dev_root;
1164 ret = btrfs_open_device(device);
1172 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1174 struct btrfs_dev_item *dev_item;
1176 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1178 return read_one_dev(root, buf, dev_item);
1181 int btrfs_read_sys_array(struct btrfs_root *root)
1183 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1184 struct extent_buffer *sb = root->fs_info->sb_buffer;
1185 struct btrfs_disk_key *disk_key;
1186 struct btrfs_chunk *chunk;
1187 struct btrfs_key key;
1192 unsigned long sb_ptr;
1196 array_size = btrfs_super_sys_array_size(super_copy);
1199 * we do this loop twice, once for the device items and
1200 * once for all of the chunks. This way there are device
1201 * structs filled in for every chunk
1203 ptr = super_copy->sys_chunk_array;
1204 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1207 while (cur < array_size) {
1208 disk_key = (struct btrfs_disk_key *)ptr;
1209 btrfs_disk_key_to_cpu(&key, disk_key);
1211 len = sizeof(*disk_key);
1216 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1217 chunk = (struct btrfs_chunk *)sb_ptr;
1218 ret = read_one_chunk(root, &key, sb, chunk);
1220 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1221 len = btrfs_chunk_item_size(num_stripes);
1232 int btrfs_read_chunk_tree(struct btrfs_root *root)
1234 struct btrfs_path *path;
1235 struct extent_buffer *leaf;
1236 struct btrfs_key key;
1237 struct btrfs_key found_key;
1241 root = root->fs_info->chunk_root;
1243 path = btrfs_alloc_path();
1247 /* first we search for all of the device items, and then we
1248 * read in all of the chunk items. This way we can create chunk
1249 * mappings that reference all of the devices that are afound
1251 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1255 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1257 leaf = path->nodes[0];
1258 slot = path->slots[0];
1259 if (slot >= btrfs_header_nritems(leaf)) {
1260 ret = btrfs_next_leaf(root, path);
1267 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1268 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1269 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1271 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1272 struct btrfs_dev_item *dev_item;
1273 dev_item = btrfs_item_ptr(leaf, slot,
1274 struct btrfs_dev_item);
1275 ret = read_one_dev(root, leaf, dev_item);
1278 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1279 struct btrfs_chunk *chunk;
1280 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1281 ret = read_one_chunk(root, &found_key, leaf, chunk);
1285 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1287 btrfs_release_path(root, path);
1291 btrfs_free_path(path);