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"
35 struct btrfs_device *dev;
39 static inline int nr_parity_stripes(struct map_lookup *map)
41 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
43 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
49 static inline int nr_data_stripes(struct map_lookup *map)
51 return map->num_stripes - nr_parity_stripes(map);
54 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
56 static LIST_HEAD(fs_uuids);
58 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
61 struct btrfs_device *dev;
62 struct list_head *cur;
64 list_for_each(cur, head) {
65 dev = list_entry(cur, struct btrfs_device, dev_list);
66 if (dev->devid == devid &&
67 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
74 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
76 struct list_head *cur;
77 struct btrfs_fs_devices *fs_devices;
79 list_for_each(cur, &fs_uuids) {
80 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
81 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
87 static int device_list_add(const char *path,
88 struct btrfs_super_block *disk_super,
89 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
91 struct btrfs_device *device;
92 struct btrfs_fs_devices *fs_devices;
93 u64 found_transid = btrfs_super_generation(disk_super);
95 fs_devices = find_fsid(disk_super->fsid);
97 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
100 INIT_LIST_HEAD(&fs_devices->devices);
101 list_add(&fs_devices->list, &fs_uuids);
102 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
103 fs_devices->latest_devid = devid;
104 fs_devices->latest_trans = found_transid;
105 fs_devices->lowest_devid = (u64)-1;
108 device = __find_device(&fs_devices->devices, devid,
109 disk_super->dev_item.uuid);
112 device = kzalloc(sizeof(*device), GFP_NOFS);
114 /* we can safely leave the fs_devices entry around */
118 device->devid = devid;
119 memcpy(device->uuid, disk_super->dev_item.uuid,
121 device->name = kstrdup(path, GFP_NOFS);
126 device->label = kstrdup(disk_super->label, GFP_NOFS);
127 if (!device->label) {
132 device->total_devs = btrfs_super_num_devices(disk_super);
133 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
134 device->total_bytes =
135 btrfs_stack_device_total_bytes(&disk_super->dev_item);
137 btrfs_stack_device_bytes_used(&disk_super->dev_item);
138 list_add(&device->dev_list, &fs_devices->devices);
139 device->fs_devices = fs_devices;
140 } else if (!device->name || strcmp(device->name, path)) {
141 char *name = strdup(path);
149 if (found_transid > fs_devices->latest_trans) {
150 fs_devices->latest_devid = devid;
151 fs_devices->latest_trans = found_transid;
153 if (fs_devices->lowest_devid > devid) {
154 fs_devices->lowest_devid = devid;
156 *fs_devices_ret = fs_devices;
160 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
162 struct btrfs_fs_devices *seed_devices;
163 struct btrfs_device *device;
166 while (!list_empty(&fs_devices->devices)) {
167 device = list_entry(fs_devices->devices.next,
168 struct btrfs_device, dev_list);
169 if (device->fd != -1) {
171 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
172 fprintf(stderr, "Warning, could not drop caches\n");
176 device->writeable = 0;
177 list_del(&device->dev_list);
178 /* free the memory */
184 seed_devices = fs_devices->seed;
185 fs_devices->seed = NULL;
187 fs_devices = seed_devices;
191 list_del(&fs_devices->list);
196 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
199 struct list_head *head = &fs_devices->devices;
200 struct list_head *cur;
201 struct btrfs_device *device;
204 list_for_each(cur, head) {
205 device = list_entry(cur, struct btrfs_device, dev_list);
207 printk("no name for device %llu, skip it now\n", device->devid);
211 fd = open(device->name, flags);
217 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
218 fprintf(stderr, "Warning, could not drop caches\n");
220 if (device->devid == fs_devices->latest_devid)
221 fs_devices->latest_bdev = fd;
222 if (device->devid == fs_devices->lowest_devid)
223 fs_devices->lowest_bdev = fd;
226 device->writeable = 1;
230 btrfs_close_devices(fs_devices);
234 int btrfs_scan_one_device(int fd, const char *path,
235 struct btrfs_fs_devices **fs_devices_ret,
236 u64 *total_devs, u64 super_offset)
238 struct btrfs_super_block *disk_super;
248 disk_super = (struct btrfs_super_block *)buf;
249 ret = btrfs_read_dev_super(fd, disk_super, super_offset);
254 devid = btrfs_stack_device_id(&disk_super->dev_item);
255 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
258 *total_devs = btrfs_super_num_devices(disk_super);
260 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
269 * this uses a pretty simple search, the expectation is that it is
270 * called very infrequently and that a given device has a small number
273 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
274 struct btrfs_device *device,
275 struct btrfs_path *path,
276 u64 num_bytes, u64 *start)
278 struct btrfs_key key;
279 struct btrfs_root *root = device->dev_root;
280 struct btrfs_dev_extent *dev_extent = NULL;
283 u64 search_start = root->fs_info->alloc_start;
284 u64 search_end = device->total_bytes;
288 struct extent_buffer *l;
293 /* FIXME use last free of some kind */
295 /* we don't want to overwrite the superblock on the drive,
296 * so we make sure to start at an offset of at least 1MB
298 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
300 if (search_start >= search_end) {
305 key.objectid = device->devid;
306 key.offset = search_start;
307 key.type = BTRFS_DEV_EXTENT_KEY;
308 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
311 ret = btrfs_previous_item(root, path, 0, key.type);
315 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
318 slot = path->slots[0];
319 if (slot >= btrfs_header_nritems(l)) {
320 ret = btrfs_next_leaf(root, path);
327 if (search_start >= search_end) {
331 *start = search_start;
335 *start = last_byte > search_start ?
336 last_byte : search_start;
337 if (search_end <= *start) {
343 btrfs_item_key_to_cpu(l, &key, slot);
345 if (key.objectid < device->devid)
348 if (key.objectid > device->devid)
351 if (key.offset >= search_start && key.offset > last_byte &&
353 if (last_byte < search_start)
354 last_byte = search_start;
355 hole_size = key.offset - last_byte;
356 if (key.offset > last_byte &&
357 hole_size >= num_bytes) {
362 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
367 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
368 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
374 /* we have to make sure we didn't find an extent that has already
375 * been allocated by the map tree or the original allocation
377 btrfs_release_path(path);
378 BUG_ON(*start < search_start);
380 if (*start + num_bytes > search_end) {
384 /* check for pending inserts here */
388 btrfs_release_path(path);
392 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
393 struct btrfs_device *device,
394 u64 chunk_tree, u64 chunk_objectid,
396 u64 num_bytes, u64 *start)
399 struct btrfs_path *path;
400 struct btrfs_root *root = device->dev_root;
401 struct btrfs_dev_extent *extent;
402 struct extent_buffer *leaf;
403 struct btrfs_key key;
405 path = btrfs_alloc_path();
409 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
414 key.objectid = device->devid;
416 key.type = BTRFS_DEV_EXTENT_KEY;
417 ret = btrfs_insert_empty_item(trans, root, path, &key,
421 leaf = path->nodes[0];
422 extent = btrfs_item_ptr(leaf, path->slots[0],
423 struct btrfs_dev_extent);
424 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
425 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
426 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
428 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
429 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
432 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
433 btrfs_mark_buffer_dirty(leaf);
435 btrfs_free_path(path);
439 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
441 struct btrfs_path *path;
443 struct btrfs_key key;
444 struct btrfs_chunk *chunk;
445 struct btrfs_key found_key;
447 path = btrfs_alloc_path();
450 key.objectid = objectid;
451 key.offset = (u64)-1;
452 key.type = BTRFS_CHUNK_ITEM_KEY;
454 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
460 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
464 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
466 if (found_key.objectid != objectid)
469 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
471 *offset = found_key.offset +
472 btrfs_chunk_length(path->nodes[0], chunk);
477 btrfs_free_path(path);
481 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
485 struct btrfs_key key;
486 struct btrfs_key found_key;
488 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
489 key.type = BTRFS_DEV_ITEM_KEY;
490 key.offset = (u64)-1;
492 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
498 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
503 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
505 *objectid = found_key.offset + 1;
509 btrfs_release_path(path);
514 * the device information is stored in the chunk root
515 * the btrfs_device struct should be fully filled in
517 int btrfs_add_device(struct btrfs_trans_handle *trans,
518 struct btrfs_root *root,
519 struct btrfs_device *device)
522 struct btrfs_path *path;
523 struct btrfs_dev_item *dev_item;
524 struct extent_buffer *leaf;
525 struct btrfs_key key;
529 root = root->fs_info->chunk_root;
531 path = btrfs_alloc_path();
535 ret = find_next_devid(root, path, &free_devid);
539 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
540 key.type = BTRFS_DEV_ITEM_KEY;
541 key.offset = free_devid;
543 ret = btrfs_insert_empty_item(trans, root, path, &key,
548 leaf = path->nodes[0];
549 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
551 device->devid = free_devid;
552 btrfs_set_device_id(leaf, dev_item, device->devid);
553 btrfs_set_device_generation(leaf, dev_item, 0);
554 btrfs_set_device_type(leaf, dev_item, device->type);
555 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
556 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
557 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
558 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
559 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
560 btrfs_set_device_group(leaf, dev_item, 0);
561 btrfs_set_device_seek_speed(leaf, dev_item, 0);
562 btrfs_set_device_bandwidth(leaf, dev_item, 0);
563 btrfs_set_device_start_offset(leaf, dev_item, 0);
565 ptr = (unsigned long)btrfs_device_uuid(dev_item);
566 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
567 ptr = (unsigned long)btrfs_device_fsid(dev_item);
568 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
569 btrfs_mark_buffer_dirty(leaf);
573 btrfs_free_path(path);
577 int btrfs_update_device(struct btrfs_trans_handle *trans,
578 struct btrfs_device *device)
581 struct btrfs_path *path;
582 struct btrfs_root *root;
583 struct btrfs_dev_item *dev_item;
584 struct extent_buffer *leaf;
585 struct btrfs_key key;
587 root = device->dev_root->fs_info->chunk_root;
589 path = btrfs_alloc_path();
593 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
594 key.type = BTRFS_DEV_ITEM_KEY;
595 key.offset = device->devid;
597 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
606 leaf = path->nodes[0];
607 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
609 btrfs_set_device_id(leaf, dev_item, device->devid);
610 btrfs_set_device_type(leaf, dev_item, device->type);
611 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
612 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
613 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
614 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
615 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
616 btrfs_mark_buffer_dirty(leaf);
619 btrfs_free_path(path);
623 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root,
625 struct btrfs_key *key,
626 struct btrfs_chunk *chunk, int item_size)
628 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
629 struct btrfs_disk_key disk_key;
633 array_size = btrfs_super_sys_array_size(super_copy);
634 if (array_size + item_size + sizeof(disk_key)
635 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
638 ptr = super_copy->sys_chunk_array + array_size;
639 btrfs_cpu_key_to_disk(&disk_key, key);
640 memcpy(ptr, &disk_key, sizeof(disk_key));
641 ptr += sizeof(disk_key);
642 memcpy(ptr, chunk, item_size);
643 item_size += sizeof(disk_key);
644 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
648 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
651 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
653 else if (type & BTRFS_BLOCK_GROUP_RAID10)
654 return calc_size * (num_stripes / sub_stripes);
655 else if (type & BTRFS_BLOCK_GROUP_RAID5)
656 return calc_size * (num_stripes - 1);
657 else if (type & BTRFS_BLOCK_GROUP_RAID6)
658 return calc_size * (num_stripes - 2);
660 return calc_size * num_stripes;
664 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
666 /* TODO, add a way to store the preferred stripe size */
667 return BTRFS_STRIPE_LEN;
671 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
673 * It is not equal to "device->total_bytes - device->bytes_used".
674 * We do not allocate any chunk in 1M at beginning of device, and not
675 * allowed to allocate any chunk before alloc_start if it is specified.
676 * So search holes from max(1M, alloc_start) to device->total_bytes.
678 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
679 struct btrfs_device *device,
682 struct btrfs_path *path;
683 struct btrfs_root *root = device->dev_root;
684 struct btrfs_key key;
685 struct btrfs_dev_extent *dev_extent = NULL;
686 struct extent_buffer *l;
687 u64 search_start = root->fs_info->alloc_start;
688 u64 search_end = device->total_bytes;
694 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
696 path = btrfs_alloc_path();
700 key.objectid = device->devid;
701 key.offset = root->fs_info->alloc_start;
702 key.type = BTRFS_DEV_EXTENT_KEY;
705 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
708 ret = btrfs_previous_item(root, path, 0, key.type);
714 slot = path->slots[0];
715 if (slot >= btrfs_header_nritems(l)) {
716 ret = btrfs_next_leaf(root, path);
723 btrfs_item_key_to_cpu(l, &key, slot);
725 if (key.objectid < device->devid)
727 if (key.objectid > device->devid)
729 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
731 if (key.offset > search_end)
733 if (key.offset > search_start)
734 free_bytes += key.offset - search_start;
736 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
737 extent_end = key.offset + btrfs_dev_extent_length(l,
739 if (extent_end > search_start)
740 search_start = extent_end;
741 if (search_start > search_end)
748 if (search_start < search_end)
749 free_bytes += search_end - search_start;
751 *avail_bytes = free_bytes;
754 btrfs_free_path(path);
758 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
759 - sizeof(struct btrfs_item) \
760 - sizeof(struct btrfs_chunk)) \
761 / sizeof(struct btrfs_stripe) + 1)
763 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
764 - 2 * sizeof(struct btrfs_disk_key) \
765 - 2 * sizeof(struct btrfs_chunk)) \
766 / sizeof(struct btrfs_stripe) + 1)
768 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
769 struct btrfs_root *extent_root, u64 *start,
770 u64 *num_bytes, u64 type)
773 struct btrfs_fs_info *info = extent_root->fs_info;
774 struct btrfs_root *chunk_root = info->chunk_root;
775 struct btrfs_stripe *stripes;
776 struct btrfs_device *device = NULL;
777 struct btrfs_chunk *chunk;
778 struct list_head private_devs;
779 struct list_head *dev_list = &info->fs_devices->devices;
780 struct list_head *cur;
781 struct map_lookup *map;
782 int min_stripe_size = 1 * 1024 * 1024;
783 u64 calc_size = 8 * 1024 * 1024;
785 u64 max_chunk_size = 4 * calc_size;
796 int stripe_len = BTRFS_STRIPE_LEN;
797 struct btrfs_key key;
800 if (list_empty(dev_list)) {
804 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
805 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
806 BTRFS_BLOCK_GROUP_RAID10 |
807 BTRFS_BLOCK_GROUP_DUP)) {
808 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
809 calc_size = 8 * 1024 * 1024;
810 max_chunk_size = calc_size * 2;
811 min_stripe_size = 1 * 1024 * 1024;
812 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
813 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
814 calc_size = 1024 * 1024 * 1024;
815 max_chunk_size = 10 * calc_size;
816 min_stripe_size = 64 * 1024 * 1024;
817 max_stripes = BTRFS_MAX_DEVS(chunk_root);
818 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
819 calc_size = 1024 * 1024 * 1024;
820 max_chunk_size = 4 * calc_size;
821 min_stripe_size = 32 * 1024 * 1024;
822 max_stripes = BTRFS_MAX_DEVS(chunk_root);
825 if (type & BTRFS_BLOCK_GROUP_RAID1) {
826 num_stripes = min_t(u64, 2,
827 btrfs_super_num_devices(info->super_copy));
832 if (type & BTRFS_BLOCK_GROUP_DUP) {
836 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
837 num_stripes = btrfs_super_num_devices(info->super_copy);
838 if (num_stripes > max_stripes)
839 num_stripes = max_stripes;
842 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
843 num_stripes = btrfs_super_num_devices(info->super_copy);
844 if (num_stripes > max_stripes)
845 num_stripes = max_stripes;
848 num_stripes &= ~(u32)1;
852 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
853 num_stripes = btrfs_super_num_devices(info->super_copy);
854 if (num_stripes > max_stripes)
855 num_stripes = max_stripes;
859 stripe_len = find_raid56_stripe_len(num_stripes - 1,
860 btrfs_super_stripesize(info->super_copy));
862 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
863 num_stripes = btrfs_super_num_devices(info->super_copy);
864 if (num_stripes > max_stripes)
865 num_stripes = max_stripes;
869 stripe_len = find_raid56_stripe_len(num_stripes - 2,
870 btrfs_super_stripesize(info->super_copy));
873 /* we don't want a chunk larger than 10% of the FS */
874 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
875 max_chunk_size = min(percent_max, max_chunk_size);
878 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
880 calc_size = max_chunk_size;
881 calc_size /= num_stripes;
882 calc_size /= stripe_len;
883 calc_size *= stripe_len;
885 /* we don't want tiny stripes */
886 calc_size = max_t(u64, calc_size, min_stripe_size);
888 calc_size /= stripe_len;
889 calc_size *= stripe_len;
890 INIT_LIST_HEAD(&private_devs);
891 cur = dev_list->next;
894 if (type & BTRFS_BLOCK_GROUP_DUP)
895 min_free = calc_size * 2;
897 min_free = calc_size;
899 /* build a private list of devices we will allocate from */
900 while(index < num_stripes) {
901 device = list_entry(cur, struct btrfs_device, dev_list);
902 ret = btrfs_device_avail_bytes(trans, device, &avail);
906 if (avail >= min_free) {
907 list_move_tail(&device->dev_list, &private_devs);
909 if (type & BTRFS_BLOCK_GROUP_DUP)
911 } else if (avail > max_avail)
916 if (index < num_stripes) {
917 list_splice(&private_devs, dev_list);
918 if (index >= min_stripes) {
920 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
921 num_stripes /= sub_stripes;
922 num_stripes *= sub_stripes;
927 if (!looped && max_avail > 0) {
929 calc_size = max_avail;
934 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
938 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
939 key.type = BTRFS_CHUNK_ITEM_KEY;
942 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
946 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
952 stripes = &chunk->stripe;
953 *num_bytes = chunk_bytes_by_type(type, calc_size,
954 num_stripes, sub_stripes);
956 while(index < num_stripes) {
957 struct btrfs_stripe *stripe;
958 BUG_ON(list_empty(&private_devs));
959 cur = private_devs.next;
960 device = list_entry(cur, struct btrfs_device, dev_list);
962 /* loop over this device again if we're doing a dup group */
963 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
964 (index == num_stripes - 1))
965 list_move_tail(&device->dev_list, dev_list);
967 ret = btrfs_alloc_dev_extent(trans, device,
968 info->chunk_root->root_key.objectid,
969 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
970 calc_size, &dev_offset);
973 device->bytes_used += calc_size;
974 ret = btrfs_update_device(trans, device);
977 map->stripes[index].dev = device;
978 map->stripes[index].physical = dev_offset;
979 stripe = stripes + index;
980 btrfs_set_stack_stripe_devid(stripe, device->devid);
981 btrfs_set_stack_stripe_offset(stripe, dev_offset);
982 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
985 BUG_ON(!list_empty(&private_devs));
987 /* key was set above */
988 btrfs_set_stack_chunk_length(chunk, *num_bytes);
989 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
990 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
991 btrfs_set_stack_chunk_type(chunk, type);
992 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
993 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
994 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
995 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
996 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
997 map->sector_size = extent_root->sectorsize;
998 map->stripe_len = stripe_len;
999 map->io_align = stripe_len;
1000 map->io_width = stripe_len;
1002 map->num_stripes = num_stripes;
1003 map->sub_stripes = sub_stripes;
1005 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1006 btrfs_chunk_item_size(num_stripes));
1008 *start = key.offset;;
1010 map->ce.start = key.offset;
1011 map->ce.size = *num_bytes;
1013 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1016 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1017 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1018 chunk, btrfs_chunk_item_size(num_stripes));
1026 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1027 struct btrfs_root *extent_root, u64 *start,
1028 u64 num_bytes, u64 type)
1031 struct btrfs_fs_info *info = extent_root->fs_info;
1032 struct btrfs_root *chunk_root = info->chunk_root;
1033 struct btrfs_stripe *stripes;
1034 struct btrfs_device *device = NULL;
1035 struct btrfs_chunk *chunk;
1036 struct list_head *dev_list = &info->fs_devices->devices;
1037 struct list_head *cur;
1038 struct map_lookup *map;
1039 u64 calc_size = 8 * 1024 * 1024;
1040 int num_stripes = 1;
1041 int sub_stripes = 0;
1044 int stripe_len = BTRFS_STRIPE_LEN;
1045 struct btrfs_key key;
1047 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1048 key.type = BTRFS_CHUNK_ITEM_KEY;
1049 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1054 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1058 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1064 stripes = &chunk->stripe;
1065 calc_size = num_bytes;
1068 cur = dev_list->next;
1069 device = list_entry(cur, struct btrfs_device, dev_list);
1071 while (index < num_stripes) {
1072 struct btrfs_stripe *stripe;
1074 ret = btrfs_alloc_dev_extent(trans, device,
1075 info->chunk_root->root_key.objectid,
1076 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1077 calc_size, &dev_offset);
1080 device->bytes_used += calc_size;
1081 ret = btrfs_update_device(trans, device);
1084 map->stripes[index].dev = device;
1085 map->stripes[index].physical = dev_offset;
1086 stripe = stripes + index;
1087 btrfs_set_stack_stripe_devid(stripe, device->devid);
1088 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1089 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1093 /* key was set above */
1094 btrfs_set_stack_chunk_length(chunk, num_bytes);
1095 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1096 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1097 btrfs_set_stack_chunk_type(chunk, type);
1098 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1099 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1100 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1101 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1102 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1103 map->sector_size = extent_root->sectorsize;
1104 map->stripe_len = stripe_len;
1105 map->io_align = stripe_len;
1106 map->io_width = stripe_len;
1108 map->num_stripes = num_stripes;
1109 map->sub_stripes = sub_stripes;
1111 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1112 btrfs_chunk_item_size(num_stripes));
1114 *start = key.offset;
1116 map->ce.start = key.offset;
1117 map->ce.size = num_bytes;
1119 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1126 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1128 struct cache_extent *ce;
1129 struct map_lookup *map;
1132 ce = search_cache_extent(&map_tree->cache_tree, logical);
1134 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1135 map = container_of(ce, struct map_lookup, ce);
1137 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1138 ret = map->num_stripes;
1139 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1140 ret = map->sub_stripes;
1141 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1143 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1150 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1153 struct cache_extent *ce;
1154 struct map_lookup *map;
1156 ce = search_cache_extent(&map_tree->cache_tree, *logical);
1159 ce = next_cache_extent(ce);
1163 map = container_of(ce, struct map_lookup, ce);
1164 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1165 *logical = ce->start;
1174 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1175 u64 chunk_start, u64 physical, u64 devid,
1176 u64 **logical, int *naddrs, int *stripe_len)
1178 struct cache_extent *ce;
1179 struct map_lookup *map;
1187 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1189 map = container_of(ce, struct map_lookup, ce);
1192 rmap_len = map->stripe_len;
1193 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1194 length = ce->size / (map->num_stripes / map->sub_stripes);
1195 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1196 length = ce->size / map->num_stripes;
1197 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1198 BTRFS_BLOCK_GROUP_RAID6)) {
1199 length = ce->size / nr_data_stripes(map);
1200 rmap_len = map->stripe_len * nr_data_stripes(map);
1203 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1205 for (i = 0; i < map->num_stripes; i++) {
1206 if (devid && map->stripes[i].dev->devid != devid)
1208 if (map->stripes[i].physical > physical ||
1209 map->stripes[i].physical + length <= physical)
1212 stripe_nr = (physical - map->stripes[i].physical) /
1215 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1216 stripe_nr = (stripe_nr * map->num_stripes + i) /
1218 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1219 stripe_nr = stripe_nr * map->num_stripes + i;
1220 } /* else if RAID[56], multiply by nr_data_stripes().
1221 * Alternatively, just use rmap_len below instead of
1222 * map->stripe_len */
1224 bytenr = ce->start + stripe_nr * rmap_len;
1225 for (j = 0; j < nr; j++) {
1226 if (buf[j] == bytenr)
1235 *stripe_len = rmap_len;
1240 static inline int parity_smaller(u64 a, u64 b)
1245 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1246 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1248 struct btrfs_bio_stripe s;
1255 for (i = 0; i < bbio->num_stripes - 1; i++) {
1256 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1257 s = bbio->stripes[i];
1259 bbio->stripes[i] = bbio->stripes[i+1];
1260 raid_map[i] = raid_map[i+1];
1261 bbio->stripes[i+1] = s;
1269 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1270 u64 logical, u64 *length,
1271 struct btrfs_multi_bio **multi_ret, int mirror_num,
1274 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1275 multi_ret, mirror_num, raid_map_ret);
1278 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1279 u64 logical, u64 *length, u64 *type,
1280 struct btrfs_multi_bio **multi_ret, int mirror_num,
1283 struct cache_extent *ce;
1284 struct map_lookup *map;
1288 u64 *raid_map = NULL;
1289 int stripes_allocated = 8;
1290 int stripes_required = 1;
1293 struct btrfs_multi_bio *multi = NULL;
1295 if (multi_ret && rw == READ) {
1296 stripes_allocated = 1;
1299 ce = search_cache_extent(&map_tree->cache_tree, logical);
1304 if (ce->start > logical || ce->start + ce->size < logical) {
1310 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1315 map = container_of(ce, struct map_lookup, ce);
1316 offset = logical - ce->start;
1319 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1320 BTRFS_BLOCK_GROUP_DUP)) {
1321 stripes_required = map->num_stripes;
1322 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1323 stripes_required = map->sub_stripes;
1326 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1327 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1328 /* RAID[56] write or recovery. Return all stripes */
1329 stripes_required = map->num_stripes;
1331 /* Only allocate the map if we've already got a large enough multi_ret */
1332 if (stripes_allocated >= stripes_required) {
1333 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1341 /* if our multi bio struct is too small, back off and try again */
1342 if (multi_ret && stripes_allocated < stripes_required) {
1343 stripes_allocated = stripes_required;
1350 * stripe_nr counts the total number of stripes we have to stride
1351 * to get to this block
1353 stripe_nr = stripe_nr / map->stripe_len;
1355 stripe_offset = stripe_nr * map->stripe_len;
1356 BUG_ON(offset < stripe_offset);
1358 /* stripe_offset is the offset of this block in its stripe*/
1359 stripe_offset = offset - stripe_offset;
1361 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1362 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1363 BTRFS_BLOCK_GROUP_RAID10 |
1364 BTRFS_BLOCK_GROUP_DUP)) {
1365 /* we limit the length of each bio to what fits in a stripe */
1366 *length = min_t(u64, ce->size - offset,
1367 map->stripe_len - stripe_offset);
1369 *length = ce->size - offset;
1375 multi->num_stripes = 1;
1377 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1379 multi->num_stripes = map->num_stripes;
1380 else if (mirror_num)
1381 stripe_index = mirror_num - 1;
1383 stripe_index = stripe_nr % map->num_stripes;
1384 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1385 int factor = map->num_stripes / map->sub_stripes;
1387 stripe_index = stripe_nr % factor;
1388 stripe_index *= map->sub_stripes;
1391 multi->num_stripes = map->sub_stripes;
1392 else if (mirror_num)
1393 stripe_index += mirror_num - 1;
1395 stripe_nr = stripe_nr / factor;
1396 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1398 multi->num_stripes = map->num_stripes;
1399 else if (mirror_num)
1400 stripe_index = mirror_num - 1;
1401 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1402 BTRFS_BLOCK_GROUP_RAID6)) {
1407 u64 raid56_full_stripe_start;
1408 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1411 * align the start of our data stripe in the logical
1414 raid56_full_stripe_start = offset / full_stripe_len;
1415 raid56_full_stripe_start *= full_stripe_len;
1417 /* get the data stripe number */
1418 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1419 stripe_nr = stripe_nr / nr_data_stripes(map);
1421 /* Work out the disk rotation on this stripe-set */
1422 rot = stripe_nr % map->num_stripes;
1424 /* Fill in the logical address of each stripe */
1425 tmp = stripe_nr * nr_data_stripes(map);
1427 for (i = 0; i < nr_data_stripes(map); i++)
1428 raid_map[(i+rot) % map->num_stripes] =
1429 ce->start + (tmp + i) * map->stripe_len;
1431 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1432 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1433 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1435 *length = map->stripe_len;
1438 multi->num_stripes = map->num_stripes;
1440 stripe_index = stripe_nr % nr_data_stripes(map);
1441 stripe_nr = stripe_nr / nr_data_stripes(map);
1444 * Mirror #0 or #1 means the original data block.
1445 * Mirror #2 is RAID5 parity block.
1446 * Mirror #3 is RAID6 Q block.
1449 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1451 /* We distribute the parity blocks across stripes */
1452 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1456 * after this do_div call, stripe_nr is the number of stripes
1457 * on this device we have to walk to find the data, and
1458 * stripe_index is the number of our device in the stripe array
1460 stripe_index = stripe_nr % map->num_stripes;
1461 stripe_nr = stripe_nr / map->num_stripes;
1463 BUG_ON(stripe_index >= map->num_stripes);
1465 for (i = 0; i < multi->num_stripes; i++) {
1466 multi->stripes[i].physical =
1467 map->stripes[stripe_index].physical + stripe_offset +
1468 stripe_nr * map->stripe_len;
1469 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1478 sort_parity_stripes(multi, raid_map);
1479 *raid_map_ret = raid_map;
1485 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1488 struct btrfs_device *device;
1489 struct btrfs_fs_devices *cur_devices;
1491 cur_devices = root->fs_info->fs_devices;
1492 while (cur_devices) {
1494 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1495 device = __find_device(&cur_devices->devices,
1500 cur_devices = cur_devices->seed;
1505 struct btrfs_device *
1506 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1507 u64 devid, int instance)
1509 struct list_head *head = &fs_devices->devices;
1510 struct btrfs_device *dev;
1513 list_for_each_entry(dev, head, dev_list) {
1514 if (dev->devid == devid && num_found++ == instance)
1520 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1522 struct cache_extent *ce;
1523 struct map_lookup *map;
1524 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1529 * During chunk recovering, we may fail to find block group's
1530 * corresponding chunk, we will rebuild it later
1532 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1533 if (!root->fs_info->is_chunk_recover)
1538 map = container_of(ce, struct map_lookup, ce);
1539 for (i = 0; i < map->num_stripes; i++) {
1540 if (!map->stripes[i].dev->writeable) {
1549 static struct btrfs_device *fill_missing_device(u64 devid)
1551 struct btrfs_device *device;
1553 device = kzalloc(sizeof(*device), GFP_NOFS);
1554 device->devid = devid;
1559 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1560 struct extent_buffer *leaf,
1561 struct btrfs_chunk *chunk)
1563 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1564 struct map_lookup *map;
1565 struct cache_extent *ce;
1569 u8 uuid[BTRFS_UUID_SIZE];
1574 logical = key->offset;
1575 length = btrfs_chunk_length(leaf, chunk);
1577 ce = search_cache_extent(&map_tree->cache_tree, logical);
1579 /* already mapped? */
1580 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1584 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1585 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1589 map->ce.start = logical;
1590 map->ce.size = length;
1591 map->num_stripes = num_stripes;
1592 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1593 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1594 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1595 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1596 map->type = btrfs_chunk_type(leaf, chunk);
1597 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1599 for (i = 0; i < num_stripes; i++) {
1600 map->stripes[i].physical =
1601 btrfs_stripe_offset_nr(leaf, chunk, i);
1602 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1603 read_extent_buffer(leaf, uuid, (unsigned long)
1604 btrfs_stripe_dev_uuid_nr(chunk, i),
1606 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1608 if (!map->stripes[i].dev) {
1609 map->stripes[i].dev = fill_missing_device(devid);
1610 printf("warning, device %llu is missing\n",
1611 (unsigned long long)devid);
1615 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1621 static int fill_device_from_item(struct extent_buffer *leaf,
1622 struct btrfs_dev_item *dev_item,
1623 struct btrfs_device *device)
1627 device->devid = btrfs_device_id(leaf, dev_item);
1628 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1629 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1630 device->type = btrfs_device_type(leaf, dev_item);
1631 device->io_align = btrfs_device_io_align(leaf, dev_item);
1632 device->io_width = btrfs_device_io_width(leaf, dev_item);
1633 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1635 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1636 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1641 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1643 struct btrfs_fs_devices *fs_devices;
1646 fs_devices = root->fs_info->fs_devices->seed;
1647 while (fs_devices) {
1648 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1652 fs_devices = fs_devices->seed;
1655 fs_devices = find_fsid(fsid);
1661 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1665 fs_devices->seed = root->fs_info->fs_devices->seed;
1666 root->fs_info->fs_devices->seed = fs_devices;
1671 static int read_one_dev(struct btrfs_root *root,
1672 struct extent_buffer *leaf,
1673 struct btrfs_dev_item *dev_item)
1675 struct btrfs_device *device;
1678 u8 fs_uuid[BTRFS_UUID_SIZE];
1679 u8 dev_uuid[BTRFS_UUID_SIZE];
1681 devid = btrfs_device_id(leaf, dev_item);
1682 read_extent_buffer(leaf, dev_uuid,
1683 (unsigned long)btrfs_device_uuid(dev_item),
1685 read_extent_buffer(leaf, fs_uuid,
1686 (unsigned long)btrfs_device_fsid(dev_item),
1689 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1690 ret = open_seed_devices(root, fs_uuid);
1695 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1697 printk("warning devid %llu not found already\n",
1698 (unsigned long long)devid);
1699 device = kzalloc(sizeof(*device), GFP_NOFS);
1703 list_add(&device->dev_list,
1704 &root->fs_info->fs_devices->devices);
1707 fill_device_from_item(leaf, dev_item, device);
1708 device->dev_root = root->fs_info->dev_root;
1712 int btrfs_read_sys_array(struct btrfs_root *root)
1714 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1715 struct extent_buffer *sb;
1716 struct btrfs_disk_key *disk_key;
1717 struct btrfs_chunk *chunk;
1718 struct btrfs_key key;
1725 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1726 BTRFS_SUPER_INFO_SIZE);
1729 btrfs_set_buffer_uptodate(sb);
1730 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1731 array_end = ((u8 *)super_copy->sys_chunk_array) +
1732 btrfs_super_sys_array_size(super_copy);
1735 * we do this loop twice, once for the device items and
1736 * once for all of the chunks. This way there are device
1737 * structs filled in for every chunk
1739 ptr = super_copy->sys_chunk_array;
1741 while (ptr < array_end) {
1742 disk_key = (struct btrfs_disk_key *)ptr;
1743 btrfs_disk_key_to_cpu(&key, disk_key);
1745 len = sizeof(*disk_key);
1748 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1749 chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1750 ret = read_one_chunk(root, &key, sb, chunk);
1753 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1754 len = btrfs_chunk_item_size(num_stripes);
1760 free_extent_buffer(sb);
1764 int btrfs_read_chunk_tree(struct btrfs_root *root)
1766 struct btrfs_path *path;
1767 struct extent_buffer *leaf;
1768 struct btrfs_key key;
1769 struct btrfs_key found_key;
1773 root = root->fs_info->chunk_root;
1775 path = btrfs_alloc_path();
1780 * Read all device items, and then all the chunk items. All
1781 * device items are found before any chunk item (their object id
1782 * is smaller than the lowest possible object id for a chunk
1783 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1785 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1788 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1792 leaf = path->nodes[0];
1793 slot = path->slots[0];
1794 if (slot >= btrfs_header_nritems(leaf)) {
1795 ret = btrfs_next_leaf(root, path);
1802 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1803 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1804 struct btrfs_dev_item *dev_item;
1805 dev_item = btrfs_item_ptr(leaf, slot,
1806 struct btrfs_dev_item);
1807 ret = read_one_dev(root, leaf, dev_item);
1809 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1810 struct btrfs_chunk *chunk;
1811 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1812 ret = read_one_chunk(root, &found_key, leaf, chunk);
1820 btrfs_free_path(path);
1824 struct list_head *btrfs_scanned_uuids(void)
1829 static int rmw_eb(struct btrfs_fs_info *info,
1830 struct extent_buffer *eb, struct extent_buffer *orig_eb)
1833 unsigned long orig_off = 0;
1834 unsigned long dest_off = 0;
1835 unsigned long copy_len = eb->len;
1837 ret = read_whole_eb(info, eb, 0);
1841 if (eb->start + eb->len <= orig_eb->start ||
1842 eb->start >= orig_eb->start + orig_eb->len)
1845 * | ----- orig_eb ------- |
1846 * | ----- stripe ------- |
1847 * | ----- orig_eb ------- |
1848 * | ----- orig_eb ------- |
1850 if (eb->start > orig_eb->start)
1851 orig_off = eb->start - orig_eb->start;
1852 if (orig_eb->start > eb->start)
1853 dest_off = orig_eb->start - eb->start;
1855 if (copy_len > orig_eb->len - orig_off)
1856 copy_len = orig_eb->len - orig_off;
1857 if (copy_len > eb->len - dest_off)
1858 copy_len = eb->len - dest_off;
1860 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1864 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1865 struct extent_buffer *orig_eb,
1866 struct extent_buffer **ebs,
1867 u64 stripe_len, u64 *raid_map,
1870 struct extent_buffer *eb;
1871 u64 start = orig_eb->start;
1876 for (i = 0; i < num_stripes; i++) {
1877 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1880 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1883 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1885 eb->start = raid_map[i];
1886 eb->len = stripe_len;
1890 eb->dev_bytenr = (u64)-1;
1892 this_eb_start = raid_map[i];
1894 if (start > this_eb_start ||
1895 start + orig_eb->len < this_eb_start + stripe_len) {
1896 ret = rmw_eb(info, eb, orig_eb);
1899 memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1905 int write_raid56_with_parity(struct btrfs_fs_info *info,
1906 struct extent_buffer *eb,
1907 struct btrfs_multi_bio *multi,
1908 u64 stripe_len, u64 *raid_map)
1910 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1914 int alloc_size = eb->len;
1916 ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1919 if (stripe_len > alloc_size)
1920 alloc_size = stripe_len;
1922 split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1923 multi->num_stripes);
1925 for (i = 0; i < multi->num_stripes; i++) {
1926 struct extent_buffer *new_eb;
1927 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1928 ebs[i]->dev_bytenr = multi->stripes[i].physical;
1929 ebs[i]->fd = multi->stripes[i].dev->fd;
1930 multi->stripes[i].dev->total_ios++;
1931 BUG_ON(ebs[i]->start != raid_map[i]);
1934 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1936 new_eb->dev_bytenr = multi->stripes[i].physical;
1937 new_eb->fd = multi->stripes[i].dev->fd;
1938 multi->stripes[i].dev->total_ios++;
1939 new_eb->len = stripe_len;
1941 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1943 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1949 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
1953 ebs[multi->num_stripes - 2] = p_eb;
1954 ebs[multi->num_stripes - 1] = q_eb;
1956 for (i = 0; i < multi->num_stripes; i++)
1957 pointers[i] = ebs[i]->data;
1959 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1962 ebs[multi->num_stripes - 1] = p_eb;
1963 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1964 for (j = 1; j < multi->num_stripes - 1; j++) {
1965 for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1966 *(unsigned long *)(p_eb->data + i) ^=
1967 *(unsigned long *)(ebs[j]->data + i);
1972 for (i = 0; i < multi->num_stripes; i++) {
1973 ret = write_extent_to_disk(ebs[i]);