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 #define _XOPEN_SOURCE 600
22 #include <sys/types.h>
24 #include <uuid/uuid.h>
29 #include "transaction.h"
30 #include "print-tree.h"
34 struct btrfs_device *dev;
38 static inline int nr_parity_stripes(struct map_lookup *map)
40 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
42 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
48 static inline int nr_data_stripes(struct map_lookup *map)
50 return map->num_stripes - nr_parity_stripes(map);
53 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
55 static LIST_HEAD(fs_uuids);
57 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
60 struct btrfs_device *dev;
61 struct list_head *cur;
63 list_for_each(cur, head) {
64 dev = list_entry(cur, struct btrfs_device, dev_list);
65 if (dev->devid == devid &&
66 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
73 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
75 struct list_head *cur;
76 struct btrfs_fs_devices *fs_devices;
78 list_for_each(cur, &fs_uuids) {
79 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
80 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
86 static int device_list_add(const char *path,
87 struct btrfs_super_block *disk_super,
88 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
90 struct btrfs_device *device;
91 struct btrfs_fs_devices *fs_devices;
92 u64 found_transid = btrfs_super_generation(disk_super);
94 fs_devices = find_fsid(disk_super->fsid);
96 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
99 INIT_LIST_HEAD(&fs_devices->devices);
100 list_add(&fs_devices->list, &fs_uuids);
101 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
102 fs_devices->latest_devid = devid;
103 fs_devices->latest_trans = found_transid;
104 fs_devices->lowest_devid = (u64)-1;
107 device = __find_device(&fs_devices->devices, devid,
108 disk_super->dev_item.uuid);
111 device = kzalloc(sizeof(*device), GFP_NOFS);
113 /* we can safely leave the fs_devices entry around */
117 device->devid = devid;
118 memcpy(device->uuid, disk_super->dev_item.uuid,
120 device->name = kstrdup(path, GFP_NOFS);
125 device->label = kstrdup(disk_super->label, GFP_NOFS);
126 device->total_devs = btrfs_super_num_devices(disk_super);
127 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
128 device->total_bytes =
129 btrfs_stack_device_total_bytes(&disk_super->dev_item);
131 btrfs_stack_device_bytes_used(&disk_super->dev_item);
132 list_add(&device->dev_list, &fs_devices->devices);
133 device->fs_devices = fs_devices;
134 } else if (!device->name || strcmp(device->name, path)) {
135 char *name = strdup(path);
143 if (found_transid > fs_devices->latest_trans) {
144 fs_devices->latest_devid = devid;
145 fs_devices->latest_trans = found_transid;
147 if (fs_devices->lowest_devid > devid) {
148 fs_devices->lowest_devid = devid;
150 *fs_devices_ret = fs_devices;
154 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
156 struct btrfs_fs_devices *seed_devices;
157 struct list_head *cur;
158 struct btrfs_device *device;
160 list_for_each(cur, &fs_devices->devices) {
161 device = list_entry(cur, struct btrfs_device, dev_list);
162 if (device->fd != -1) {
164 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
165 fprintf(stderr, "Warning, could not drop caches\n");
169 device->writeable = 0;
172 seed_devices = fs_devices->seed;
173 fs_devices->seed = NULL;
175 fs_devices = seed_devices;
182 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
185 struct list_head *head = &fs_devices->devices;
186 struct list_head *cur;
187 struct btrfs_device *device;
190 list_for_each(cur, head) {
191 device = list_entry(cur, struct btrfs_device, dev_list);
193 printk("no name for device %llu, skip it now\n", device->devid);
197 fd = open(device->name, flags);
203 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
204 fprintf(stderr, "Warning, could not drop caches\n");
206 if (device->devid == fs_devices->latest_devid)
207 fs_devices->latest_bdev = fd;
208 if (device->devid == fs_devices->lowest_devid)
209 fs_devices->lowest_bdev = fd;
212 device->writeable = 1;
216 btrfs_close_devices(fs_devices);
220 int btrfs_scan_one_device(int fd, const char *path,
221 struct btrfs_fs_devices **fs_devices_ret,
222 u64 *total_devs, u64 super_offset)
224 struct btrfs_super_block *disk_super;
235 disk_super = (struct btrfs_super_block *)buf;
236 ret = btrfs_read_dev_super(fd, disk_super, super_offset);
241 devid = le64_to_cpu(disk_super->dev_item.devid);
242 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
245 *total_devs = btrfs_super_num_devices(disk_super);
246 uuid_unparse(disk_super->fsid, uuidbuf);
248 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
257 * this uses a pretty simple search, the expectation is that it is
258 * called very infrequently and that a given device has a small number
261 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
262 struct btrfs_device *device,
263 struct btrfs_path *path,
264 u64 num_bytes, u64 *start)
266 struct btrfs_key key;
267 struct btrfs_root *root = device->dev_root;
268 struct btrfs_dev_extent *dev_extent = NULL;
271 u64 search_start = 0;
272 u64 search_end = device->total_bytes;
276 struct extent_buffer *l;
281 /* FIXME use last free of some kind */
283 /* we don't want to overwrite the superblock on the drive,
284 * so we make sure to start at an offset of at least 1MB
286 search_start = max((u64)1024 * 1024, search_start);
288 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
289 search_start = max(root->fs_info->alloc_start, search_start);
291 key.objectid = device->devid;
292 key.offset = search_start;
293 key.type = BTRFS_DEV_EXTENT_KEY;
294 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
297 ret = btrfs_previous_item(root, path, 0, key.type);
301 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
304 slot = path->slots[0];
305 if (slot >= btrfs_header_nritems(l)) {
306 ret = btrfs_next_leaf(root, path);
313 if (search_start >= search_end) {
317 *start = search_start;
321 *start = last_byte > search_start ?
322 last_byte : search_start;
323 if (search_end <= *start) {
329 btrfs_item_key_to_cpu(l, &key, slot);
331 if (key.objectid < device->devid)
334 if (key.objectid > device->devid)
337 if (key.offset >= search_start && key.offset > last_byte &&
339 if (last_byte < search_start)
340 last_byte = search_start;
341 hole_size = key.offset - last_byte;
342 if (key.offset > last_byte &&
343 hole_size >= num_bytes) {
348 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
353 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
354 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
360 /* we have to make sure we didn't find an extent that has already
361 * been allocated by the map tree or the original allocation
363 btrfs_release_path(root, path);
364 BUG_ON(*start < search_start);
366 if (*start + num_bytes > search_end) {
370 /* check for pending inserts here */
374 btrfs_release_path(root, path);
378 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
379 struct btrfs_device *device,
380 u64 chunk_tree, u64 chunk_objectid,
382 u64 num_bytes, u64 *start)
385 struct btrfs_path *path;
386 struct btrfs_root *root = device->dev_root;
387 struct btrfs_dev_extent *extent;
388 struct extent_buffer *leaf;
389 struct btrfs_key key;
391 path = btrfs_alloc_path();
395 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
400 key.objectid = device->devid;
402 key.type = BTRFS_DEV_EXTENT_KEY;
403 ret = btrfs_insert_empty_item(trans, root, path, &key,
407 leaf = path->nodes[0];
408 extent = btrfs_item_ptr(leaf, path->slots[0],
409 struct btrfs_dev_extent);
410 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
411 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
412 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
414 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
415 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
418 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
419 btrfs_mark_buffer_dirty(leaf);
421 btrfs_free_path(path);
425 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
427 struct btrfs_path *path;
429 struct btrfs_key key;
430 struct btrfs_chunk *chunk;
431 struct btrfs_key found_key;
433 path = btrfs_alloc_path();
436 key.objectid = objectid;
437 key.offset = (u64)-1;
438 key.type = BTRFS_CHUNK_ITEM_KEY;
440 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
446 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
450 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
452 if (found_key.objectid != objectid)
455 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
457 *offset = found_key.offset +
458 btrfs_chunk_length(path->nodes[0], chunk);
463 btrfs_free_path(path);
467 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
471 struct btrfs_key key;
472 struct btrfs_key found_key;
474 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
475 key.type = BTRFS_DEV_ITEM_KEY;
476 key.offset = (u64)-1;
478 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
484 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
489 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
491 *objectid = found_key.offset + 1;
495 btrfs_release_path(root, path);
500 * the device information is stored in the chunk root
501 * the btrfs_device struct should be fully filled in
503 int btrfs_add_device(struct btrfs_trans_handle *trans,
504 struct btrfs_root *root,
505 struct btrfs_device *device)
508 struct btrfs_path *path;
509 struct btrfs_dev_item *dev_item;
510 struct extent_buffer *leaf;
511 struct btrfs_key key;
515 root = root->fs_info->chunk_root;
517 path = btrfs_alloc_path();
521 ret = find_next_devid(root, path, &free_devid);
525 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
526 key.type = BTRFS_DEV_ITEM_KEY;
527 key.offset = free_devid;
529 ret = btrfs_insert_empty_item(trans, root, path, &key,
534 leaf = path->nodes[0];
535 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
537 device->devid = free_devid;
538 btrfs_set_device_id(leaf, dev_item, device->devid);
539 btrfs_set_device_generation(leaf, dev_item, 0);
540 btrfs_set_device_type(leaf, dev_item, device->type);
541 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
542 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
543 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
544 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
545 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
546 btrfs_set_device_group(leaf, dev_item, 0);
547 btrfs_set_device_seek_speed(leaf, dev_item, 0);
548 btrfs_set_device_bandwidth(leaf, dev_item, 0);
549 btrfs_set_device_start_offset(leaf, dev_item, 0);
551 ptr = (unsigned long)btrfs_device_uuid(dev_item);
552 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
553 ptr = (unsigned long)btrfs_device_fsid(dev_item);
554 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
555 btrfs_mark_buffer_dirty(leaf);
559 btrfs_free_path(path);
563 int btrfs_update_device(struct btrfs_trans_handle *trans,
564 struct btrfs_device *device)
567 struct btrfs_path *path;
568 struct btrfs_root *root;
569 struct btrfs_dev_item *dev_item;
570 struct extent_buffer *leaf;
571 struct btrfs_key key;
573 root = device->dev_root->fs_info->chunk_root;
575 path = btrfs_alloc_path();
579 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
580 key.type = BTRFS_DEV_ITEM_KEY;
581 key.offset = device->devid;
583 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
592 leaf = path->nodes[0];
593 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
595 btrfs_set_device_id(leaf, dev_item, device->devid);
596 btrfs_set_device_type(leaf, dev_item, device->type);
597 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
598 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
599 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
600 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
601 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
602 btrfs_mark_buffer_dirty(leaf);
605 btrfs_free_path(path);
609 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
610 struct btrfs_root *root,
611 struct btrfs_key *key,
612 struct btrfs_chunk *chunk, int item_size)
614 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
615 struct btrfs_disk_key disk_key;
619 array_size = btrfs_super_sys_array_size(super_copy);
620 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
623 ptr = super_copy->sys_chunk_array + array_size;
624 btrfs_cpu_key_to_disk(&disk_key, key);
625 memcpy(ptr, &disk_key, sizeof(disk_key));
626 ptr += sizeof(disk_key);
627 memcpy(ptr, chunk, item_size);
628 item_size += sizeof(disk_key);
629 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
633 static u64 div_factor(u64 num, int factor)
641 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
644 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
646 else if (type & BTRFS_BLOCK_GROUP_RAID10)
647 return calc_size * (num_stripes / sub_stripes);
648 else if (type & BTRFS_BLOCK_GROUP_RAID5)
649 return calc_size * (num_stripes - 1);
650 else if (type & BTRFS_BLOCK_GROUP_RAID6)
651 return calc_size * (num_stripes - 2);
653 return calc_size * num_stripes;
657 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
659 /* TODO, add a way to store the preferred stripe size */
663 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
664 struct btrfs_root *extent_root, u64 *start,
665 u64 *num_bytes, u64 type)
668 struct btrfs_fs_info *info = extent_root->fs_info;
669 struct btrfs_root *chunk_root = info->chunk_root;
670 struct btrfs_stripe *stripes;
671 struct btrfs_device *device = NULL;
672 struct btrfs_chunk *chunk;
673 struct list_head private_devs;
674 struct list_head *dev_list = &info->fs_devices->devices;
675 struct list_head *cur;
676 struct map_lookup *map;
677 int min_stripe_size = 1 * 1024 * 1024;
678 u64 calc_size = 8 * 1024 * 1024;
680 u64 max_chunk_size = 4 * calc_size;
690 int stripe_len = 64 * 1024;
691 struct btrfs_key key;
694 if (list_empty(dev_list)) {
698 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
699 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
700 BTRFS_BLOCK_GROUP_RAID10 |
701 BTRFS_BLOCK_GROUP_DUP)) {
702 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
703 calc_size = 8 * 1024 * 1024;
704 max_chunk_size = calc_size * 2;
705 min_stripe_size = 1 * 1024 * 1024;
706 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
707 calc_size = 1024 * 1024 * 1024;
708 max_chunk_size = 10 * calc_size;
709 min_stripe_size = 64 * 1024 * 1024;
710 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
711 calc_size = 1024 * 1024 * 1024;
712 max_chunk_size = 4 * calc_size;
713 min_stripe_size = 32 * 1024 * 1024;
716 if (type & BTRFS_BLOCK_GROUP_RAID1) {
717 num_stripes = min_t(u64, 2,
718 btrfs_super_num_devices(info->super_copy));
723 if (type & BTRFS_BLOCK_GROUP_DUP) {
727 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
728 num_stripes = btrfs_super_num_devices(info->super_copy);
731 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
732 num_stripes = btrfs_super_num_devices(info->super_copy);
735 num_stripes &= ~(u32)1;
739 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
740 num_stripes = btrfs_super_num_devices(info->super_copy);
744 stripe_len = find_raid56_stripe_len(num_stripes - 1,
745 btrfs_super_stripesize(info->super_copy));
747 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
748 num_stripes = btrfs_super_num_devices(info->super_copy);
752 stripe_len = find_raid56_stripe_len(num_stripes - 2,
753 btrfs_super_stripesize(info->super_copy));
756 /* we don't want a chunk larger than 10% of the FS */
757 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
758 max_chunk_size = min(percent_max, max_chunk_size);
761 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
763 calc_size = max_chunk_size;
764 calc_size /= num_stripes;
765 calc_size /= stripe_len;
766 calc_size *= stripe_len;
768 /* we don't want tiny stripes */
769 calc_size = max_t(u64, calc_size, min_stripe_size);
771 calc_size /= stripe_len;
772 calc_size *= stripe_len;
773 INIT_LIST_HEAD(&private_devs);
774 cur = dev_list->next;
777 if (type & BTRFS_BLOCK_GROUP_DUP)
778 min_free = calc_size * 2;
780 min_free = calc_size;
782 /* build a private list of devices we will allocate from */
783 while(index < num_stripes) {
784 device = list_entry(cur, struct btrfs_device, dev_list);
785 avail = device->total_bytes - device->bytes_used;
787 if (avail >= min_free) {
788 list_move_tail(&device->dev_list, &private_devs);
790 if (type & BTRFS_BLOCK_GROUP_DUP)
792 } else if (avail > max_avail)
797 if (index < num_stripes) {
798 list_splice(&private_devs, dev_list);
799 if (index >= min_stripes) {
801 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
802 num_stripes /= sub_stripes;
803 num_stripes *= sub_stripes;
808 if (!looped && max_avail > 0) {
810 calc_size = max_avail;
815 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
819 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
820 key.type = BTRFS_CHUNK_ITEM_KEY;
823 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
827 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
833 stripes = &chunk->stripe;
834 *num_bytes = chunk_bytes_by_type(type, calc_size,
835 num_stripes, sub_stripes);
837 while(index < num_stripes) {
838 struct btrfs_stripe *stripe;
839 BUG_ON(list_empty(&private_devs));
840 cur = private_devs.next;
841 device = list_entry(cur, struct btrfs_device, dev_list);
843 /* loop over this device again if we're doing a dup group */
844 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
845 (index == num_stripes - 1))
846 list_move_tail(&device->dev_list, dev_list);
848 ret = btrfs_alloc_dev_extent(trans, device,
849 info->chunk_root->root_key.objectid,
850 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
851 calc_size, &dev_offset);
854 device->bytes_used += calc_size;
855 ret = btrfs_update_device(trans, device);
858 map->stripes[index].dev = device;
859 map->stripes[index].physical = dev_offset;
860 stripe = stripes + index;
861 btrfs_set_stack_stripe_devid(stripe, device->devid);
862 btrfs_set_stack_stripe_offset(stripe, dev_offset);
863 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
866 BUG_ON(!list_empty(&private_devs));
868 /* key was set above */
869 btrfs_set_stack_chunk_length(chunk, *num_bytes);
870 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
871 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
872 btrfs_set_stack_chunk_type(chunk, type);
873 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
874 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
875 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
876 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
877 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
878 map->sector_size = extent_root->sectorsize;
879 map->stripe_len = stripe_len;
880 map->io_align = stripe_len;
881 map->io_width = stripe_len;
883 map->num_stripes = num_stripes;
884 map->sub_stripes = sub_stripes;
886 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
887 btrfs_chunk_item_size(num_stripes));
889 *start = key.offset;;
891 map->ce.start = key.offset;
892 map->ce.size = *num_bytes;
894 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
897 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
898 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
899 chunk, btrfs_chunk_item_size(num_stripes));
907 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
908 struct btrfs_root *extent_root, u64 *start,
909 u64 num_bytes, u64 type)
912 struct btrfs_fs_info *info = extent_root->fs_info;
913 struct btrfs_root *chunk_root = info->chunk_root;
914 struct btrfs_stripe *stripes;
915 struct btrfs_device *device = NULL;
916 struct btrfs_chunk *chunk;
917 struct list_head *dev_list = &info->fs_devices->devices;
918 struct list_head *cur;
919 struct map_lookup *map;
920 u64 calc_size = 8 * 1024 * 1024;
925 int stripe_len = 64 * 1024;
926 struct btrfs_key key;
928 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
929 key.type = BTRFS_CHUNK_ITEM_KEY;
930 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
935 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
939 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
945 stripes = &chunk->stripe;
946 calc_size = num_bytes;
949 cur = dev_list->next;
950 device = list_entry(cur, struct btrfs_device, dev_list);
952 while (index < num_stripes) {
953 struct btrfs_stripe *stripe;
955 ret = btrfs_alloc_dev_extent(trans, device,
956 info->chunk_root->root_key.objectid,
957 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
958 calc_size, &dev_offset);
961 device->bytes_used += calc_size;
962 ret = btrfs_update_device(trans, device);
965 map->stripes[index].dev = device;
966 map->stripes[index].physical = dev_offset;
967 stripe = stripes + index;
968 btrfs_set_stack_stripe_devid(stripe, device->devid);
969 btrfs_set_stack_stripe_offset(stripe, dev_offset);
970 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
974 /* key was set above */
975 btrfs_set_stack_chunk_length(chunk, num_bytes);
976 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
977 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
978 btrfs_set_stack_chunk_type(chunk, type);
979 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
980 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
981 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
982 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
983 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
984 map->sector_size = extent_root->sectorsize;
985 map->stripe_len = stripe_len;
986 map->io_align = stripe_len;
987 map->io_width = stripe_len;
989 map->num_stripes = num_stripes;
990 map->sub_stripes = sub_stripes;
992 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
993 btrfs_chunk_item_size(num_stripes));
997 map->ce.start = key.offset;
998 map->ce.size = num_bytes;
1000 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1007 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
1009 cache_tree_init(&tree->cache_tree);
1012 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1014 struct cache_extent *ce;
1015 struct map_lookup *map;
1018 ce = search_cache_extent(&map_tree->cache_tree, logical);
1020 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1021 map = container_of(ce, struct map_lookup, ce);
1023 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1024 ret = map->num_stripes;
1025 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1026 ret = map->sub_stripes;
1027 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1029 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1036 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1039 struct cache_extent *ce;
1040 struct map_lookup *map;
1042 ce = search_cache_extent(&map_tree->cache_tree, *logical);
1045 ce = next_cache_extent(ce);
1049 map = container_of(ce, struct map_lookup, ce);
1050 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1051 *logical = ce->start;
1060 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1061 u64 chunk_start, u64 physical, u64 devid,
1062 u64 **logical, int *naddrs, int *stripe_len)
1064 struct cache_extent *ce;
1065 struct map_lookup *map;
1073 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1075 map = container_of(ce, struct map_lookup, ce);
1078 rmap_len = map->stripe_len;
1079 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1080 length = ce->size / (map->num_stripes / map->sub_stripes);
1081 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1082 length = ce->size / map->num_stripes;
1083 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1084 BTRFS_BLOCK_GROUP_RAID6)) {
1085 length = ce->size / nr_data_stripes(map);
1086 rmap_len = map->stripe_len * nr_data_stripes(map);
1089 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1091 for (i = 0; i < map->num_stripes; i++) {
1092 if (devid && map->stripes[i].dev->devid != devid)
1094 if (map->stripes[i].physical > physical ||
1095 map->stripes[i].physical + length <= physical)
1098 stripe_nr = (physical - map->stripes[i].physical) /
1101 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1102 stripe_nr = (stripe_nr * map->num_stripes + i) /
1104 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1105 stripe_nr = stripe_nr * map->num_stripes + i;
1106 } /* else if RAID[56], multiply by nr_data_stripes().
1107 * Alternatively, just use rmap_len below instead of
1108 * map->stripe_len */
1110 bytenr = ce->start + stripe_nr * rmap_len;
1111 for (j = 0; j < nr; j++) {
1112 if (buf[j] == bytenr)
1121 *stripe_len = rmap_len;
1126 static inline int parity_smaller(u64 a, u64 b)
1131 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1132 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1134 struct btrfs_bio_stripe s;
1141 for (i = 0; i < bbio->num_stripes - 1; i++) {
1142 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1143 s = bbio->stripes[i];
1145 bbio->stripes[i] = bbio->stripes[i+1];
1146 raid_map[i] = raid_map[i+1];
1147 bbio->stripes[i+1] = s;
1155 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1156 u64 logical, u64 *length,
1157 struct btrfs_multi_bio **multi_ret, int mirror_num,
1160 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1161 multi_ret, mirror_num, raid_map_ret);
1164 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1165 u64 logical, u64 *length, u64 *type,
1166 struct btrfs_multi_bio **multi_ret, int mirror_num,
1169 struct cache_extent *ce;
1170 struct map_lookup *map;
1174 u64 *raid_map = NULL;
1175 int stripes_allocated = 8;
1176 int stripes_required = 1;
1179 struct btrfs_multi_bio *multi = NULL;
1181 if (multi_ret && rw == READ) {
1182 stripes_allocated = 1;
1185 ce = search_cache_extent(&map_tree->cache_tree, logical);
1191 if (ce->start > logical || ce->start + ce->size < logical) {
1198 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1203 map = container_of(ce, struct map_lookup, ce);
1204 offset = logical - ce->start;
1207 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1208 BTRFS_BLOCK_GROUP_DUP)) {
1209 stripes_required = map->num_stripes;
1210 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1211 stripes_required = map->sub_stripes;
1214 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1215 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1216 /* RAID[56] write or recovery. Return all stripes */
1217 stripes_required = map->num_stripes;
1219 /* Only allocate the map if we've already got a large enough multi_ret */
1220 if (stripes_allocated >= stripes_required) {
1221 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1229 /* if our multi bio struct is too small, back off and try again */
1230 if (multi_ret && stripes_allocated < stripes_required) {
1231 stripes_allocated = stripes_required;
1238 * stripe_nr counts the total number of stripes we have to stride
1239 * to get to this block
1241 stripe_nr = stripe_nr / map->stripe_len;
1243 stripe_offset = stripe_nr * map->stripe_len;
1244 BUG_ON(offset < stripe_offset);
1246 /* stripe_offset is the offset of this block in its stripe*/
1247 stripe_offset = offset - stripe_offset;
1249 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1250 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1251 BTRFS_BLOCK_GROUP_RAID10 |
1252 BTRFS_BLOCK_GROUP_DUP)) {
1253 /* we limit the length of each bio to what fits in a stripe */
1254 *length = min_t(u64, ce->size - offset,
1255 map->stripe_len - stripe_offset);
1257 *length = ce->size - offset;
1263 multi->num_stripes = 1;
1265 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1267 multi->num_stripes = map->num_stripes;
1268 else if (mirror_num)
1269 stripe_index = mirror_num - 1;
1271 stripe_index = stripe_nr % map->num_stripes;
1272 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1273 int factor = map->num_stripes / map->sub_stripes;
1275 stripe_index = stripe_nr % factor;
1276 stripe_index *= map->sub_stripes;
1279 multi->num_stripes = map->sub_stripes;
1280 else if (mirror_num)
1281 stripe_index += mirror_num - 1;
1283 stripe_nr = stripe_nr / factor;
1284 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1286 multi->num_stripes = map->num_stripes;
1287 else if (mirror_num)
1288 stripe_index = mirror_num - 1;
1289 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1290 BTRFS_BLOCK_GROUP_RAID6)) {
1295 u64 raid56_full_stripe_start;
1296 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1299 * align the start of our data stripe in the logical
1302 raid56_full_stripe_start = offset / full_stripe_len;
1303 raid56_full_stripe_start *= full_stripe_len;
1305 /* get the data stripe number */
1306 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1307 stripe_nr = stripe_nr / nr_data_stripes(map);
1309 /* Work out the disk rotation on this stripe-set */
1310 rot = stripe_nr % map->num_stripes;
1312 /* Fill in the logical address of each stripe */
1313 tmp = stripe_nr * nr_data_stripes(map);
1315 for (i = 0; i < nr_data_stripes(map); i++)
1316 raid_map[(i+rot) % map->num_stripes] =
1317 ce->start + (tmp + i) * map->stripe_len;
1319 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1320 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1321 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1323 *length = map->stripe_len;
1326 multi->num_stripes = map->num_stripes;
1328 stripe_index = stripe_nr % nr_data_stripes(map);
1329 stripe_nr = stripe_nr / nr_data_stripes(map);
1332 * Mirror #0 or #1 means the original data block.
1333 * Mirror #2 is RAID5 parity block.
1334 * Mirror #3 is RAID6 Q block.
1337 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1339 /* We distribute the parity blocks across stripes */
1340 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1344 * after this do_div call, stripe_nr is the number of stripes
1345 * on this device we have to walk to find the data, and
1346 * stripe_index is the number of our device in the stripe array
1348 stripe_index = stripe_nr % map->num_stripes;
1349 stripe_nr = stripe_nr / map->num_stripes;
1351 BUG_ON(stripe_index >= map->num_stripes);
1353 for (i = 0; i < multi->num_stripes; i++) {
1354 multi->stripes[i].physical =
1355 map->stripes[stripe_index].physical + stripe_offset +
1356 stripe_nr * map->stripe_len;
1357 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1366 sort_parity_stripes(multi, raid_map);
1367 *raid_map_ret = raid_map;
1373 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1376 struct btrfs_device *device;
1377 struct btrfs_fs_devices *cur_devices;
1379 cur_devices = root->fs_info->fs_devices;
1380 while (cur_devices) {
1382 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1383 device = __find_device(&cur_devices->devices,
1388 cur_devices = cur_devices->seed;
1393 struct btrfs_device *
1394 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1395 u64 devid, int instance)
1397 struct list_head *head = &fs_devices->devices;
1398 struct btrfs_device *dev;
1401 list_for_each_entry(dev, head, dev_list) {
1402 if (dev->devid == devid && num_found++ == instance)
1408 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1409 struct btrfs_fs_devices *fs_devices)
1411 struct map_lookup *map;
1412 u64 logical = BTRFS_SUPER_INFO_OFFSET;
1413 u64 length = BTRFS_SUPER_INFO_SIZE;
1414 int num_stripes = 0;
1415 int sub_stripes = 0;
1418 struct list_head *cur;
1420 list_for_each(cur, &fs_devices->devices) {
1423 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1427 map->ce.start = logical;
1428 map->ce.size = length;
1429 map->num_stripes = num_stripes;
1430 map->sub_stripes = sub_stripes;
1431 map->io_width = length;
1432 map->io_align = length;
1433 map->sector_size = length;
1434 map->stripe_len = length;
1435 map->type = BTRFS_BLOCK_GROUP_RAID1;
1438 list_for_each(cur, &fs_devices->devices) {
1439 struct btrfs_device *device = list_entry(cur,
1440 struct btrfs_device,
1442 map->stripes[i].physical = logical;
1443 map->stripes[i].dev = device;
1446 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1447 if (ret == -EEXIST) {
1448 struct cache_extent *old;
1449 struct map_lookup *old_map;
1450 old = lookup_cache_extent(&map_tree->cache_tree,
1452 old_map = container_of(old, struct map_lookup, ce);
1453 remove_cache_extent(&map_tree->cache_tree, old);
1455 ret = insert_cache_extent(&map_tree->cache_tree,
1462 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1464 struct cache_extent *ce;
1465 struct map_lookup *map;
1466 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1470 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1473 map = container_of(ce, struct map_lookup, ce);
1474 for (i = 0; i < map->num_stripes; i++) {
1475 if (!map->stripes[i].dev->writeable) {
1484 static struct btrfs_device *fill_missing_device(u64 devid)
1486 struct btrfs_device *device;
1488 device = kzalloc(sizeof(*device), GFP_NOFS);
1489 device->devid = devid;
1494 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1495 struct extent_buffer *leaf,
1496 struct btrfs_chunk *chunk)
1498 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1499 struct map_lookup *map;
1500 struct cache_extent *ce;
1504 u8 uuid[BTRFS_UUID_SIZE];
1509 logical = key->offset;
1510 length = btrfs_chunk_length(leaf, chunk);
1512 ce = search_cache_extent(&map_tree->cache_tree, logical);
1514 /* already mapped? */
1515 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1519 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1520 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1524 map->ce.start = logical;
1525 map->ce.size = length;
1526 map->num_stripes = num_stripes;
1527 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1528 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1529 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1530 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1531 map->type = btrfs_chunk_type(leaf, chunk);
1532 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1534 for (i = 0; i < num_stripes; i++) {
1535 map->stripes[i].physical =
1536 btrfs_stripe_offset_nr(leaf, chunk, i);
1537 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1538 read_extent_buffer(leaf, uuid, (unsigned long)
1539 btrfs_stripe_dev_uuid_nr(chunk, i),
1541 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1543 if (!map->stripes[i].dev) {
1544 map->stripes[i].dev = fill_missing_device(devid);
1545 printf("warning, device %llu is missing\n",
1546 (unsigned long long)devid);
1550 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1556 static int fill_device_from_item(struct extent_buffer *leaf,
1557 struct btrfs_dev_item *dev_item,
1558 struct btrfs_device *device)
1562 device->devid = btrfs_device_id(leaf, dev_item);
1563 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1564 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1565 device->type = btrfs_device_type(leaf, dev_item);
1566 device->io_align = btrfs_device_io_align(leaf, dev_item);
1567 device->io_width = btrfs_device_io_width(leaf, dev_item);
1568 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1570 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1571 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1576 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1578 struct btrfs_fs_devices *fs_devices;
1581 fs_devices = root->fs_info->fs_devices->seed;
1582 while (fs_devices) {
1583 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1587 fs_devices = fs_devices->seed;
1590 fs_devices = find_fsid(fsid);
1596 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1600 fs_devices->seed = root->fs_info->fs_devices->seed;
1601 root->fs_info->fs_devices->seed = fs_devices;
1606 static int read_one_dev(struct btrfs_root *root,
1607 struct extent_buffer *leaf,
1608 struct btrfs_dev_item *dev_item)
1610 struct btrfs_device *device;
1613 u8 fs_uuid[BTRFS_UUID_SIZE];
1614 u8 dev_uuid[BTRFS_UUID_SIZE];
1616 devid = btrfs_device_id(leaf, dev_item);
1617 read_extent_buffer(leaf, dev_uuid,
1618 (unsigned long)btrfs_device_uuid(dev_item),
1620 read_extent_buffer(leaf, fs_uuid,
1621 (unsigned long)btrfs_device_fsid(dev_item),
1624 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1625 ret = open_seed_devices(root, fs_uuid);
1630 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1632 printk("warning devid %llu not found already\n",
1633 (unsigned long long)devid);
1634 device = kmalloc(sizeof(*device), GFP_NOFS);
1637 device->total_ios = 0;
1638 list_add(&device->dev_list,
1639 &root->fs_info->fs_devices->devices);
1642 fill_device_from_item(leaf, dev_item, device);
1643 device->dev_root = root->fs_info->dev_root;
1647 int btrfs_read_sys_array(struct btrfs_root *root)
1649 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1650 struct extent_buffer *sb;
1651 struct btrfs_disk_key *disk_key;
1652 struct btrfs_chunk *chunk;
1653 struct btrfs_key key;
1658 unsigned long sb_ptr;
1662 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1663 BTRFS_SUPER_INFO_SIZE);
1666 btrfs_set_buffer_uptodate(sb);
1667 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1668 array_size = btrfs_super_sys_array_size(super_copy);
1671 * we do this loop twice, once for the device items and
1672 * once for all of the chunks. This way there are device
1673 * structs filled in for every chunk
1675 ptr = super_copy->sys_chunk_array;
1676 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1679 while (cur < array_size) {
1680 disk_key = (struct btrfs_disk_key *)ptr;
1681 btrfs_disk_key_to_cpu(&key, disk_key);
1683 len = sizeof(*disk_key);
1688 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1689 chunk = (struct btrfs_chunk *)sb_ptr;
1690 ret = read_one_chunk(root, &key, sb, chunk);
1693 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1694 len = btrfs_chunk_item_size(num_stripes);
1702 free_extent_buffer(sb);
1706 int btrfs_read_chunk_tree(struct btrfs_root *root)
1708 struct btrfs_path *path;
1709 struct extent_buffer *leaf;
1710 struct btrfs_key key;
1711 struct btrfs_key found_key;
1715 root = root->fs_info->chunk_root;
1717 path = btrfs_alloc_path();
1721 /* first we search for all of the device items, and then we
1722 * read in all of the chunk items. This way we can create chunk
1723 * mappings that reference all of the devices that are afound
1725 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1729 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1731 leaf = path->nodes[0];
1732 slot = path->slots[0];
1733 if (slot >= btrfs_header_nritems(leaf)) {
1734 ret = btrfs_next_leaf(root, path);
1741 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1742 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1743 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1745 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1746 struct btrfs_dev_item *dev_item;
1747 dev_item = btrfs_item_ptr(leaf, slot,
1748 struct btrfs_dev_item);
1749 ret = read_one_dev(root, leaf, dev_item);
1752 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1753 struct btrfs_chunk *chunk;
1754 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1755 ret = read_one_chunk(root, &found_key, leaf, chunk);
1760 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1762 btrfs_release_path(root, path);
1768 btrfs_free_path(path);
1772 struct list_head *btrfs_scanned_uuids(void)
1777 static int rmw_eb(struct btrfs_fs_info *info,
1778 struct extent_buffer *eb, struct extent_buffer *orig_eb)
1781 unsigned long orig_off = 0;
1782 unsigned long dest_off = 0;
1783 unsigned long copy_len = eb->len;
1785 ret = read_whole_eb(info, eb, 0);
1789 if (eb->start + eb->len <= orig_eb->start ||
1790 eb->start >= orig_eb->start + orig_eb->len)
1793 * | ----- orig_eb ------- |
1794 * | ----- stripe ------- |
1795 * | ----- orig_eb ------- |
1796 * | ----- orig_eb ------- |
1798 if (eb->start > orig_eb->start)
1799 orig_off = eb->start - orig_eb->start;
1800 if (orig_eb->start > eb->start)
1801 dest_off = orig_eb->start - eb->start;
1803 if (copy_len > orig_eb->len - orig_off)
1804 copy_len = orig_eb->len - orig_off;
1805 if (copy_len > eb->len - dest_off)
1806 copy_len = eb->len - dest_off;
1808 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1812 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1813 struct extent_buffer *orig_eb,
1814 struct extent_buffer **ebs,
1815 u64 stripe_len, u64 *raid_map,
1818 struct extent_buffer *eb;
1819 u64 start = orig_eb->start;
1824 for (i = 0; i < num_stripes; i++) {
1825 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1828 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1831 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1833 eb->start = raid_map[i];
1834 eb->len = stripe_len;
1838 eb->dev_bytenr = (u64)-1;
1840 this_eb_start = raid_map[i];
1842 if (start > this_eb_start ||
1843 start + orig_eb->len < this_eb_start + stripe_len) {
1844 ret = rmw_eb(info, eb, orig_eb);
1847 memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1853 int write_raid56_with_parity(struct btrfs_fs_info *info,
1854 struct extent_buffer *eb,
1855 struct btrfs_multi_bio *multi,
1856 u64 stripe_len, u64 *raid_map)
1858 struct extent_buffer *ebs[multi->num_stripes], *p_eb = NULL, *q_eb = NULL;
1862 int alloc_size = eb->len;
1864 if (stripe_len > alloc_size)
1865 alloc_size = stripe_len;
1867 split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1868 multi->num_stripes);
1870 for (i = 0; i < multi->num_stripes; i++) {
1871 struct extent_buffer *new_eb;
1872 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1873 ebs[i]->dev_bytenr = multi->stripes[i].physical;
1874 ebs[i]->fd = multi->stripes[i].dev->fd;
1875 multi->stripes[i].dev->total_ios++;
1876 BUG_ON(ebs[i]->start != raid_map[i]);
1879 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1881 new_eb->dev_bytenr = multi->stripes[i].physical;
1882 new_eb->fd = multi->stripes[i].dev->fd;
1883 multi->stripes[i].dev->total_ios++;
1884 new_eb->len = stripe_len;
1886 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1888 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1892 void *pointers[multi->num_stripes];
1893 ebs[multi->num_stripes - 2] = p_eb;
1894 ebs[multi->num_stripes - 1] = q_eb;
1896 for (i = 0; i < multi->num_stripes; i++)
1897 pointers[i] = ebs[i]->data;
1899 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1901 ebs[multi->num_stripes - 1] = p_eb;
1902 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1903 for (j = 1; j < multi->num_stripes - 1; j++) {
1904 for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1905 *(unsigned long *)(p_eb->data + i) ^=
1906 *(unsigned long *)(ebs[j]->data + i);
1911 for (i = 0; i < multi->num_stripes; i++) {
1912 ret = write_extent_to_disk(ebs[i]);