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.
20 #include <sys/types.h>
22 #include <uuid/uuid.h>
27 #include "transaction.h"
28 #include "print-tree.h"
33 struct btrfs_device *dev;
37 static inline int nr_parity_stripes(struct map_lookup *map)
39 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
41 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
47 static inline int nr_data_stripes(struct map_lookup *map)
49 return map->num_stripes - nr_parity_stripes(map);
52 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
54 static LIST_HEAD(fs_uuids);
56 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
59 struct btrfs_device *dev;
60 struct list_head *cur;
62 list_for_each(cur, head) {
63 dev = list_entry(cur, struct btrfs_device, dev_list);
64 if (dev->devid == devid &&
65 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
72 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
74 struct list_head *cur;
75 struct btrfs_fs_devices *fs_devices;
77 list_for_each(cur, &fs_uuids) {
78 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
79 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
85 static int device_list_add(const char *path,
86 struct btrfs_super_block *disk_super,
87 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
89 struct btrfs_device *device;
90 struct btrfs_fs_devices *fs_devices;
91 u64 found_transid = btrfs_super_generation(disk_super);
93 fs_devices = find_fsid(disk_super->fsid);
95 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
98 INIT_LIST_HEAD(&fs_devices->devices);
99 list_add(&fs_devices->list, &fs_uuids);
100 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
101 fs_devices->latest_devid = devid;
102 fs_devices->latest_trans = found_transid;
103 fs_devices->lowest_devid = (u64)-1;
106 device = __find_device(&fs_devices->devices, devid,
107 disk_super->dev_item.uuid);
110 device = kzalloc(sizeof(*device), GFP_NOFS);
112 /* we can safely leave the fs_devices entry around */
116 device->devid = devid;
117 device->generation = found_transid;
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 if (!device->label) {
131 device->total_devs = btrfs_super_num_devices(disk_super);
132 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
133 device->total_bytes =
134 btrfs_stack_device_total_bytes(&disk_super->dev_item);
136 btrfs_stack_device_bytes_used(&disk_super->dev_item);
137 list_add(&device->dev_list, &fs_devices->devices);
138 device->fs_devices = fs_devices;
139 } else if (!device->name || strcmp(device->name, path)) {
140 char *name = strdup(path);
148 if (found_transid > fs_devices->latest_trans) {
149 fs_devices->latest_devid = devid;
150 fs_devices->latest_trans = found_transid;
152 if (fs_devices->lowest_devid > devid) {
153 fs_devices->lowest_devid = devid;
155 *fs_devices_ret = fs_devices;
159 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
161 struct btrfs_fs_devices *seed_devices;
162 struct btrfs_device *device;
167 while (!list_empty(&fs_devices->devices)) {
168 device = list_entry(fs_devices->devices.next,
169 struct btrfs_device, dev_list);
170 if (device->fd != -1) {
172 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
173 fprintf(stderr, "Warning, could not drop caches\n");
177 device->writeable = 0;
178 list_del(&device->dev_list);
179 /* free the memory */
185 seed_devices = fs_devices->seed;
186 fs_devices->seed = NULL;
188 struct btrfs_fs_devices *orig;
191 fs_devices = seed_devices;
192 list_del(&orig->list);
196 list_del(&fs_devices->list);
203 void btrfs_close_all_devices(void)
205 struct btrfs_fs_devices *fs_devices;
207 while (!list_empty(&fs_uuids)) {
208 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
210 btrfs_close_devices(fs_devices);
214 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
217 struct list_head *head = &fs_devices->devices;
218 struct list_head *cur;
219 struct btrfs_device *device;
222 list_for_each(cur, head) {
223 device = list_entry(cur, struct btrfs_device, dev_list);
225 printk("no name for device %llu, skip it now\n", device->devid);
229 fd = open(device->name, flags);
232 error("cannot open device '%s': %s", device->name,
237 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
238 fprintf(stderr, "Warning, could not drop caches\n");
240 if (device->devid == fs_devices->latest_devid)
241 fs_devices->latest_bdev = fd;
242 if (device->devid == fs_devices->lowest_devid)
243 fs_devices->lowest_bdev = fd;
246 device->writeable = 1;
250 btrfs_close_devices(fs_devices);
254 int btrfs_scan_one_device(int fd, const char *path,
255 struct btrfs_fs_devices **fs_devices_ret,
256 u64 *total_devs, u64 super_offset, unsigned sbflags)
258 struct btrfs_super_block *disk_super;
259 char buf[BTRFS_SUPER_INFO_SIZE];
263 disk_super = (struct btrfs_super_block *)buf;
264 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
267 devid = btrfs_stack_device_id(&disk_super->dev_item);
268 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
271 *total_devs = btrfs_super_num_devices(disk_super);
273 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
279 * this uses a pretty simple search, the expectation is that it is
280 * called very infrequently and that a given device has a small number
283 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
284 struct btrfs_device *device,
285 struct btrfs_path *path,
286 u64 num_bytes, u64 *start)
288 struct btrfs_key key;
289 struct btrfs_root *root = device->dev_root;
290 struct btrfs_dev_extent *dev_extent = NULL;
293 u64 search_start = root->fs_info->alloc_start;
294 u64 search_end = device->total_bytes;
298 struct extent_buffer *l;
303 /* FIXME use last free of some kind */
305 /* we don't want to overwrite the superblock on the drive,
306 * so we make sure to start at an offset of at least 1MB
308 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
310 if (search_start >= search_end) {
315 key.objectid = device->devid;
316 key.offset = search_start;
317 key.type = BTRFS_DEV_EXTENT_KEY;
318 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
321 ret = btrfs_previous_item(root, path, 0, key.type);
325 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
328 slot = path->slots[0];
329 if (slot >= btrfs_header_nritems(l)) {
330 ret = btrfs_next_leaf(root, path);
337 if (search_start >= search_end) {
341 *start = search_start;
345 *start = last_byte > search_start ?
346 last_byte : search_start;
347 if (search_end <= *start) {
353 btrfs_item_key_to_cpu(l, &key, slot);
355 if (key.objectid < device->devid)
358 if (key.objectid > device->devid)
361 if (key.offset >= search_start && key.offset > last_byte &&
363 if (last_byte < search_start)
364 last_byte = search_start;
365 hole_size = key.offset - last_byte;
366 if (key.offset > last_byte &&
367 hole_size >= num_bytes) {
372 if (key.type != BTRFS_DEV_EXTENT_KEY) {
377 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
378 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
384 /* we have to make sure we didn't find an extent that has already
385 * been allocated by the map tree or the original allocation
387 btrfs_release_path(path);
388 BUG_ON(*start < search_start);
390 if (*start + num_bytes > search_end) {
394 /* check for pending inserts here */
398 btrfs_release_path(path);
402 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
403 struct btrfs_device *device,
404 u64 chunk_tree, u64 chunk_objectid,
406 u64 num_bytes, u64 *start, int convert)
409 struct btrfs_path *path;
410 struct btrfs_root *root = device->dev_root;
411 struct btrfs_dev_extent *extent;
412 struct extent_buffer *leaf;
413 struct btrfs_key key;
415 path = btrfs_alloc_path();
420 * For convert case, just skip search free dev_extent, as caller
421 * is responsible to make sure it's free.
424 ret = find_free_dev_extent(trans, device, path, num_bytes,
430 key.objectid = device->devid;
432 key.type = BTRFS_DEV_EXTENT_KEY;
433 ret = btrfs_insert_empty_item(trans, root, path, &key,
437 leaf = path->nodes[0];
438 extent = btrfs_item_ptr(leaf, path->slots[0],
439 struct btrfs_dev_extent);
440 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
441 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
442 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
444 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
445 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
448 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
449 btrfs_mark_buffer_dirty(leaf);
451 btrfs_free_path(path);
455 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
457 struct btrfs_path *path;
459 struct btrfs_key key;
460 struct btrfs_chunk *chunk;
461 struct btrfs_key found_key;
463 path = btrfs_alloc_path();
467 key.objectid = objectid;
468 key.offset = (u64)-1;
469 key.type = BTRFS_CHUNK_ITEM_KEY;
471 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
477 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
481 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
483 if (found_key.objectid != objectid)
486 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
488 *offset = found_key.offset +
489 btrfs_chunk_length(path->nodes[0], chunk);
494 btrfs_free_path(path);
498 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
502 struct btrfs_key key;
503 struct btrfs_key found_key;
505 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
506 key.type = BTRFS_DEV_ITEM_KEY;
507 key.offset = (u64)-1;
509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
515 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
520 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
522 *objectid = found_key.offset + 1;
526 btrfs_release_path(path);
531 * the device information is stored in the chunk root
532 * the btrfs_device struct should be fully filled in
534 int btrfs_add_device(struct btrfs_trans_handle *trans,
535 struct btrfs_root *root,
536 struct btrfs_device *device)
539 struct btrfs_path *path;
540 struct btrfs_dev_item *dev_item;
541 struct extent_buffer *leaf;
542 struct btrfs_key key;
546 root = root->fs_info->chunk_root;
548 path = btrfs_alloc_path();
552 ret = find_next_devid(root, path, &free_devid);
556 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
557 key.type = BTRFS_DEV_ITEM_KEY;
558 key.offset = free_devid;
560 ret = btrfs_insert_empty_item(trans, root, path, &key,
565 leaf = path->nodes[0];
566 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
568 device->devid = free_devid;
569 btrfs_set_device_id(leaf, dev_item, device->devid);
570 btrfs_set_device_generation(leaf, dev_item, 0);
571 btrfs_set_device_type(leaf, dev_item, device->type);
572 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
573 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
574 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
575 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
576 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
577 btrfs_set_device_group(leaf, dev_item, 0);
578 btrfs_set_device_seek_speed(leaf, dev_item, 0);
579 btrfs_set_device_bandwidth(leaf, dev_item, 0);
580 btrfs_set_device_start_offset(leaf, dev_item, 0);
582 ptr = (unsigned long)btrfs_device_uuid(dev_item);
583 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
584 ptr = (unsigned long)btrfs_device_fsid(dev_item);
585 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
586 btrfs_mark_buffer_dirty(leaf);
590 btrfs_free_path(path);
594 int btrfs_update_device(struct btrfs_trans_handle *trans,
595 struct btrfs_device *device)
598 struct btrfs_path *path;
599 struct btrfs_root *root;
600 struct btrfs_dev_item *dev_item;
601 struct extent_buffer *leaf;
602 struct btrfs_key key;
604 root = device->dev_root->fs_info->chunk_root;
606 path = btrfs_alloc_path();
610 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
611 key.type = BTRFS_DEV_ITEM_KEY;
612 key.offset = device->devid;
614 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
623 leaf = path->nodes[0];
624 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
626 btrfs_set_device_id(leaf, dev_item, device->devid);
627 btrfs_set_device_type(leaf, dev_item, device->type);
628 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
629 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
630 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
631 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
632 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
633 btrfs_mark_buffer_dirty(leaf);
636 btrfs_free_path(path);
640 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
641 struct btrfs_root *root,
642 struct btrfs_key *key,
643 struct btrfs_chunk *chunk, int item_size)
645 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
646 struct btrfs_disk_key disk_key;
650 array_size = btrfs_super_sys_array_size(super_copy);
651 if (array_size + item_size + sizeof(disk_key)
652 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
655 ptr = super_copy->sys_chunk_array + array_size;
656 btrfs_cpu_key_to_disk(&disk_key, key);
657 memcpy(ptr, &disk_key, sizeof(disk_key));
658 ptr += sizeof(disk_key);
659 memcpy(ptr, chunk, item_size);
660 item_size += sizeof(disk_key);
661 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
665 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
668 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
670 else if (type & BTRFS_BLOCK_GROUP_RAID10)
671 return calc_size * (num_stripes / sub_stripes);
672 else if (type & BTRFS_BLOCK_GROUP_RAID5)
673 return calc_size * (num_stripes - 1);
674 else if (type & BTRFS_BLOCK_GROUP_RAID6)
675 return calc_size * (num_stripes - 2);
677 return calc_size * num_stripes;
681 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
683 /* TODO, add a way to store the preferred stripe size */
684 return BTRFS_STRIPE_LEN;
688 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
690 * It is not equal to "device->total_bytes - device->bytes_used".
691 * We do not allocate any chunk in 1M at beginning of device, and not
692 * allowed to allocate any chunk before alloc_start if it is specified.
693 * So search holes from max(1M, alloc_start) to device->total_bytes.
695 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
696 struct btrfs_device *device,
699 struct btrfs_path *path;
700 struct btrfs_root *root = device->dev_root;
701 struct btrfs_key key;
702 struct btrfs_dev_extent *dev_extent = NULL;
703 struct extent_buffer *l;
704 u64 search_start = root->fs_info->alloc_start;
705 u64 search_end = device->total_bytes;
711 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
713 path = btrfs_alloc_path();
717 key.objectid = device->devid;
718 key.offset = root->fs_info->alloc_start;
719 key.type = BTRFS_DEV_EXTENT_KEY;
722 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
725 ret = btrfs_previous_item(root, path, 0, key.type);
731 slot = path->slots[0];
732 if (slot >= btrfs_header_nritems(l)) {
733 ret = btrfs_next_leaf(root, path);
740 btrfs_item_key_to_cpu(l, &key, slot);
742 if (key.objectid < device->devid)
744 if (key.objectid > device->devid)
746 if (key.type != BTRFS_DEV_EXTENT_KEY)
748 if (key.offset > search_end)
750 if (key.offset > search_start)
751 free_bytes += key.offset - search_start;
753 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
754 extent_end = key.offset + btrfs_dev_extent_length(l,
756 if (extent_end > search_start)
757 search_start = extent_end;
758 if (search_start > search_end)
765 if (search_start < search_end)
766 free_bytes += search_end - search_start;
768 *avail_bytes = free_bytes;
771 btrfs_free_path(path);
775 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
776 - sizeof(struct btrfs_item) \
777 - sizeof(struct btrfs_chunk)) \
778 / sizeof(struct btrfs_stripe) + 1)
780 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
781 - 2 * sizeof(struct btrfs_disk_key) \
782 - 2 * sizeof(struct btrfs_chunk)) \
783 / sizeof(struct btrfs_stripe) + 1)
785 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
786 struct btrfs_root *extent_root, u64 *start,
787 u64 *num_bytes, u64 type)
790 struct btrfs_fs_info *info = extent_root->fs_info;
791 struct btrfs_root *chunk_root = info->chunk_root;
792 struct btrfs_stripe *stripes;
793 struct btrfs_device *device = NULL;
794 struct btrfs_chunk *chunk;
795 struct list_head private_devs;
796 struct list_head *dev_list = &info->fs_devices->devices;
797 struct list_head *cur;
798 struct map_lookup *map;
799 int min_stripe_size = 1 * 1024 * 1024;
800 u64 calc_size = 8 * 1024 * 1024;
802 u64 max_chunk_size = 4 * calc_size;
813 int stripe_len = BTRFS_STRIPE_LEN;
814 struct btrfs_key key;
817 if (list_empty(dev_list)) {
821 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
822 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
823 BTRFS_BLOCK_GROUP_RAID10 |
824 BTRFS_BLOCK_GROUP_DUP)) {
825 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
826 calc_size = 8 * 1024 * 1024;
827 max_chunk_size = calc_size * 2;
828 min_stripe_size = 1 * 1024 * 1024;
829 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
830 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
831 calc_size = 1024 * 1024 * 1024;
832 max_chunk_size = 10 * calc_size;
833 min_stripe_size = 64 * 1024 * 1024;
834 max_stripes = BTRFS_MAX_DEVS(chunk_root);
835 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
836 calc_size = 1024 * 1024 * 1024;
837 max_chunk_size = 4 * calc_size;
838 min_stripe_size = 32 * 1024 * 1024;
839 max_stripes = BTRFS_MAX_DEVS(chunk_root);
842 if (type & BTRFS_BLOCK_GROUP_RAID1) {
843 num_stripes = min_t(u64, 2,
844 btrfs_super_num_devices(info->super_copy));
849 if (type & BTRFS_BLOCK_GROUP_DUP) {
853 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
854 num_stripes = btrfs_super_num_devices(info->super_copy);
855 if (num_stripes > max_stripes)
856 num_stripes = max_stripes;
859 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
860 num_stripes = btrfs_super_num_devices(info->super_copy);
861 if (num_stripes > max_stripes)
862 num_stripes = max_stripes;
865 num_stripes &= ~(u32)1;
869 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
870 num_stripes = btrfs_super_num_devices(info->super_copy);
871 if (num_stripes > max_stripes)
872 num_stripes = max_stripes;
876 stripe_len = find_raid56_stripe_len(num_stripes - 1,
877 btrfs_super_stripesize(info->super_copy));
879 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
880 num_stripes = btrfs_super_num_devices(info->super_copy);
881 if (num_stripes > max_stripes)
882 num_stripes = max_stripes;
886 stripe_len = find_raid56_stripe_len(num_stripes - 2,
887 btrfs_super_stripesize(info->super_copy));
890 /* we don't want a chunk larger than 10% of the FS */
891 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
892 max_chunk_size = min(percent_max, max_chunk_size);
895 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
897 calc_size = max_chunk_size;
898 calc_size /= num_stripes;
899 calc_size /= stripe_len;
900 calc_size *= stripe_len;
902 /* we don't want tiny stripes */
903 calc_size = max_t(u64, calc_size, min_stripe_size);
905 calc_size /= stripe_len;
906 calc_size *= stripe_len;
907 INIT_LIST_HEAD(&private_devs);
908 cur = dev_list->next;
911 if (type & BTRFS_BLOCK_GROUP_DUP)
912 min_free = calc_size * 2;
914 min_free = calc_size;
916 /* build a private list of devices we will allocate from */
917 while(index < num_stripes) {
918 device = list_entry(cur, struct btrfs_device, dev_list);
919 ret = btrfs_device_avail_bytes(trans, device, &avail);
923 if (avail >= min_free) {
924 list_move_tail(&device->dev_list, &private_devs);
926 if (type & BTRFS_BLOCK_GROUP_DUP)
928 } else if (avail > max_avail)
933 if (index < num_stripes) {
934 list_splice(&private_devs, dev_list);
935 if (index >= min_stripes) {
937 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
938 num_stripes /= sub_stripes;
939 num_stripes *= sub_stripes;
944 if (!looped && max_avail > 0) {
946 calc_size = max_avail;
951 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
955 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
956 key.type = BTRFS_CHUNK_ITEM_KEY;
959 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
963 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
969 stripes = &chunk->stripe;
970 *num_bytes = chunk_bytes_by_type(type, calc_size,
971 num_stripes, sub_stripes);
973 while(index < num_stripes) {
974 struct btrfs_stripe *stripe;
975 BUG_ON(list_empty(&private_devs));
976 cur = private_devs.next;
977 device = list_entry(cur, struct btrfs_device, dev_list);
979 /* loop over this device again if we're doing a dup group */
980 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
981 (index == num_stripes - 1))
982 list_move_tail(&device->dev_list, dev_list);
984 ret = btrfs_alloc_dev_extent(trans, device,
985 info->chunk_root->root_key.objectid,
986 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
987 calc_size, &dev_offset, 0);
990 device->bytes_used += calc_size;
991 ret = btrfs_update_device(trans, device);
994 map->stripes[index].dev = device;
995 map->stripes[index].physical = dev_offset;
996 stripe = stripes + index;
997 btrfs_set_stack_stripe_devid(stripe, device->devid);
998 btrfs_set_stack_stripe_offset(stripe, dev_offset);
999 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1002 BUG_ON(!list_empty(&private_devs));
1004 /* key was set above */
1005 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1006 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1007 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1008 btrfs_set_stack_chunk_type(chunk, type);
1009 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1010 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1011 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1012 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1013 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1014 map->sector_size = extent_root->sectorsize;
1015 map->stripe_len = stripe_len;
1016 map->io_align = stripe_len;
1017 map->io_width = stripe_len;
1019 map->num_stripes = num_stripes;
1020 map->sub_stripes = sub_stripes;
1022 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1023 btrfs_chunk_item_size(num_stripes));
1025 *start = key.offset;;
1027 map->ce.start = key.offset;
1028 map->ce.size = *num_bytes;
1030 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1033 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1034 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1035 chunk, btrfs_chunk_item_size(num_stripes));
1044 * Alloc a DATA chunk with SINGLE profile.
1046 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1047 * (btrfs logical bytenr == on-disk bytenr)
1048 * For that case, caller must make sure the chunk and dev_extent are not
1051 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1052 struct btrfs_root *extent_root, u64 *start,
1053 u64 num_bytes, u64 type, int convert)
1056 struct btrfs_fs_info *info = extent_root->fs_info;
1057 struct btrfs_root *chunk_root = info->chunk_root;
1058 struct btrfs_stripe *stripes;
1059 struct btrfs_device *device = NULL;
1060 struct btrfs_chunk *chunk;
1061 struct list_head *dev_list = &info->fs_devices->devices;
1062 struct list_head *cur;
1063 struct map_lookup *map;
1064 u64 calc_size = 8 * 1024 * 1024;
1065 int num_stripes = 1;
1066 int sub_stripes = 0;
1069 int stripe_len = BTRFS_STRIPE_LEN;
1070 struct btrfs_key key;
1072 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1073 key.type = BTRFS_CHUNK_ITEM_KEY;
1075 if (*start != round_down(*start, extent_root->sectorsize)) {
1076 error("DATA chunk start not sectorsize aligned: %llu",
1077 (unsigned long long)*start);
1080 key.offset = *start;
1081 dev_offset = *start;
1083 ret = find_next_chunk(chunk_root,
1084 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1090 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1094 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1100 stripes = &chunk->stripe;
1101 calc_size = num_bytes;
1104 cur = dev_list->next;
1105 device = list_entry(cur, struct btrfs_device, dev_list);
1107 while (index < num_stripes) {
1108 struct btrfs_stripe *stripe;
1110 ret = btrfs_alloc_dev_extent(trans, device,
1111 info->chunk_root->root_key.objectid,
1112 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1113 calc_size, &dev_offset, convert);
1116 device->bytes_used += calc_size;
1117 ret = btrfs_update_device(trans, device);
1120 map->stripes[index].dev = device;
1121 map->stripes[index].physical = dev_offset;
1122 stripe = stripes + index;
1123 btrfs_set_stack_stripe_devid(stripe, device->devid);
1124 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1125 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1129 /* key was set above */
1130 btrfs_set_stack_chunk_length(chunk, num_bytes);
1131 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1132 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1133 btrfs_set_stack_chunk_type(chunk, type);
1134 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1135 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1136 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1137 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1138 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1139 map->sector_size = extent_root->sectorsize;
1140 map->stripe_len = stripe_len;
1141 map->io_align = stripe_len;
1142 map->io_width = stripe_len;
1144 map->num_stripes = num_stripes;
1145 map->sub_stripes = sub_stripes;
1147 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1148 btrfs_chunk_item_size(num_stripes));
1151 *start = key.offset;
1153 map->ce.start = key.offset;
1154 map->ce.size = num_bytes;
1156 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1163 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1165 struct cache_extent *ce;
1166 struct map_lookup *map;
1169 ce = search_cache_extent(&map_tree->cache_tree, logical);
1171 fprintf(stderr, "No mapping for %llu-%llu\n",
1172 (unsigned long long)logical,
1173 (unsigned long long)logical+len);
1176 if (ce->start > logical || ce->start + ce->size < logical) {
1177 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1178 "%llu-%llu\n", (unsigned long long)logical,
1179 (unsigned long long)logical+len,
1180 (unsigned long long)ce->start,
1181 (unsigned long long)ce->start + ce->size);
1184 map = container_of(ce, struct map_lookup, ce);
1186 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1187 ret = map->num_stripes;
1188 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1189 ret = map->sub_stripes;
1190 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1192 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1199 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1200 u64 *size, u64 type)
1202 struct cache_extent *ce;
1203 struct map_lookup *map;
1206 ce = search_cache_extent(&map_tree->cache_tree, cur);
1210 * only jump to next bg if our cur is not 0
1211 * As the initial logical for btrfs_next_bg() is 0, and
1212 * if we jump to next bg, we skipped a valid bg.
1215 ce = next_cache_extent(ce);
1221 map = container_of(ce, struct map_lookup, ce);
1222 if (map->type & type) {
1223 *logical = ce->start;
1232 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1233 u64 chunk_start, u64 physical, u64 devid,
1234 u64 **logical, int *naddrs, int *stripe_len)
1236 struct cache_extent *ce;
1237 struct map_lookup *map;
1245 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1247 map = container_of(ce, struct map_lookup, ce);
1250 rmap_len = map->stripe_len;
1251 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1252 length = ce->size / (map->num_stripes / map->sub_stripes);
1253 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1254 length = ce->size / map->num_stripes;
1255 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1256 BTRFS_BLOCK_GROUP_RAID6)) {
1257 length = ce->size / nr_data_stripes(map);
1258 rmap_len = map->stripe_len * nr_data_stripes(map);
1261 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1263 for (i = 0; i < map->num_stripes; i++) {
1264 if (devid && map->stripes[i].dev->devid != devid)
1266 if (map->stripes[i].physical > physical ||
1267 map->stripes[i].physical + length <= physical)
1270 stripe_nr = (physical - map->stripes[i].physical) /
1273 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1274 stripe_nr = (stripe_nr * map->num_stripes + i) /
1276 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1277 stripe_nr = stripe_nr * map->num_stripes + i;
1278 } /* else if RAID[56], multiply by nr_data_stripes().
1279 * Alternatively, just use rmap_len below instead of
1280 * map->stripe_len */
1282 bytenr = ce->start + stripe_nr * rmap_len;
1283 for (j = 0; j < nr; j++) {
1284 if (buf[j] == bytenr)
1293 *stripe_len = rmap_len;
1298 static inline int parity_smaller(u64 a, u64 b)
1303 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1304 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1306 struct btrfs_bio_stripe s;
1313 for (i = 0; i < bbio->num_stripes - 1; i++) {
1314 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1315 s = bbio->stripes[i];
1317 bbio->stripes[i] = bbio->stripes[i+1];
1318 raid_map[i] = raid_map[i+1];
1319 bbio->stripes[i+1] = s;
1327 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1328 u64 logical, u64 *length,
1329 struct btrfs_multi_bio **multi_ret, int mirror_num,
1332 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1333 multi_ret, mirror_num, raid_map_ret);
1336 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1337 u64 logical, u64 *length, u64 *type,
1338 struct btrfs_multi_bio **multi_ret, int mirror_num,
1341 struct cache_extent *ce;
1342 struct map_lookup *map;
1346 u64 *raid_map = NULL;
1347 int stripes_allocated = 8;
1348 int stripes_required = 1;
1351 struct btrfs_multi_bio *multi = NULL;
1353 if (multi_ret && rw == READ) {
1354 stripes_allocated = 1;
1357 ce = search_cache_extent(&map_tree->cache_tree, logical);
1363 if (ce->start > logical) {
1365 *length = ce->start - logical;
1370 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1375 map = container_of(ce, struct map_lookup, ce);
1376 offset = logical - ce->start;
1379 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1380 BTRFS_BLOCK_GROUP_DUP)) {
1381 stripes_required = map->num_stripes;
1382 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1383 stripes_required = map->sub_stripes;
1386 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1387 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1388 /* RAID[56] write or recovery. Return all stripes */
1389 stripes_required = map->num_stripes;
1391 /* Only allocate the map if we've already got a large enough multi_ret */
1392 if (stripes_allocated >= stripes_required) {
1393 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1401 /* if our multi bio struct is too small, back off and try again */
1402 if (multi_ret && stripes_allocated < stripes_required) {
1403 stripes_allocated = stripes_required;
1410 * stripe_nr counts the total number of stripes we have to stride
1411 * to get to this block
1413 stripe_nr = stripe_nr / map->stripe_len;
1415 stripe_offset = stripe_nr * map->stripe_len;
1416 BUG_ON(offset < stripe_offset);
1418 /* stripe_offset is the offset of this block in its stripe*/
1419 stripe_offset = offset - stripe_offset;
1421 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1422 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1423 BTRFS_BLOCK_GROUP_RAID10 |
1424 BTRFS_BLOCK_GROUP_DUP)) {
1425 /* we limit the length of each bio to what fits in a stripe */
1426 *length = min_t(u64, ce->size - offset,
1427 map->stripe_len - stripe_offset);
1429 *length = ce->size - offset;
1435 multi->num_stripes = 1;
1437 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1439 multi->num_stripes = map->num_stripes;
1440 else if (mirror_num)
1441 stripe_index = mirror_num - 1;
1443 stripe_index = stripe_nr % map->num_stripes;
1444 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1445 int factor = map->num_stripes / map->sub_stripes;
1447 stripe_index = stripe_nr % factor;
1448 stripe_index *= map->sub_stripes;
1451 multi->num_stripes = map->sub_stripes;
1452 else if (mirror_num)
1453 stripe_index += mirror_num - 1;
1455 stripe_nr = stripe_nr / factor;
1456 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1458 multi->num_stripes = map->num_stripes;
1459 else if (mirror_num)
1460 stripe_index = mirror_num - 1;
1461 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1462 BTRFS_BLOCK_GROUP_RAID6)) {
1467 u64 raid56_full_stripe_start;
1468 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1471 * align the start of our data stripe in the logical
1474 raid56_full_stripe_start = offset / full_stripe_len;
1475 raid56_full_stripe_start *= full_stripe_len;
1477 /* get the data stripe number */
1478 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1479 stripe_nr = stripe_nr / nr_data_stripes(map);
1481 /* Work out the disk rotation on this stripe-set */
1482 rot = stripe_nr % map->num_stripes;
1484 /* Fill in the logical address of each stripe */
1485 tmp = stripe_nr * nr_data_stripes(map);
1487 for (i = 0; i < nr_data_stripes(map); i++)
1488 raid_map[(i+rot) % map->num_stripes] =
1489 ce->start + (tmp + i) * map->stripe_len;
1491 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1492 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1493 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1495 *length = map->stripe_len;
1498 multi->num_stripes = map->num_stripes;
1500 stripe_index = stripe_nr % nr_data_stripes(map);
1501 stripe_nr = stripe_nr / nr_data_stripes(map);
1504 * Mirror #0 or #1 means the original data block.
1505 * Mirror #2 is RAID5 parity block.
1506 * Mirror #3 is RAID6 Q block.
1509 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1511 /* We distribute the parity blocks across stripes */
1512 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1516 * after this do_div call, stripe_nr is the number of stripes
1517 * on this device we have to walk to find the data, and
1518 * stripe_index is the number of our device in the stripe array
1520 stripe_index = stripe_nr % map->num_stripes;
1521 stripe_nr = stripe_nr / map->num_stripes;
1523 BUG_ON(stripe_index >= map->num_stripes);
1525 for (i = 0; i < multi->num_stripes; i++) {
1526 multi->stripes[i].physical =
1527 map->stripes[stripe_index].physical + stripe_offset +
1528 stripe_nr * map->stripe_len;
1529 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1538 sort_parity_stripes(multi, raid_map);
1539 *raid_map_ret = raid_map;
1545 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1548 struct btrfs_device *device;
1549 struct btrfs_fs_devices *cur_devices;
1551 cur_devices = root->fs_info->fs_devices;
1552 while (cur_devices) {
1554 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1555 root->fs_info->ignore_fsid_mismatch)) {
1556 device = __find_device(&cur_devices->devices,
1561 cur_devices = cur_devices->seed;
1566 struct btrfs_device *
1567 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1568 u64 devid, int instance)
1570 struct list_head *head = &fs_devices->devices;
1571 struct btrfs_device *dev;
1574 list_for_each_entry(dev, head, dev_list) {
1575 if (dev->devid == devid && num_found++ == instance)
1581 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1583 struct cache_extent *ce;
1584 struct map_lookup *map;
1585 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1590 * During chunk recovering, we may fail to find block group's
1591 * corresponding chunk, we will rebuild it later
1593 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1594 if (!root->fs_info->is_chunk_recover)
1599 map = container_of(ce, struct map_lookup, ce);
1600 for (i = 0; i < map->num_stripes; i++) {
1601 if (!map->stripes[i].dev->writeable) {
1610 static struct btrfs_device *fill_missing_device(u64 devid)
1612 struct btrfs_device *device;
1614 device = kzalloc(sizeof(*device), GFP_NOFS);
1615 device->devid = devid;
1621 * slot == -1: SYSTEM chunk
1622 * return -EIO on error, otherwise return 0
1624 int btrfs_check_chunk_valid(struct btrfs_root *root,
1625 struct extent_buffer *leaf,
1626 struct btrfs_chunk *chunk,
1627 int slot, u64 logical)
1635 length = btrfs_chunk_length(leaf, chunk);
1636 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1637 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1638 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1639 type = btrfs_chunk_type(leaf, chunk);
1642 * These valid checks may be insufficient to cover every corner cases.
1644 if (!IS_ALIGNED(logical, root->sectorsize)) {
1645 error("invalid chunk logical %llu", logical);
1648 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1649 error("invalid chunk sectorsize %llu",
1650 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1653 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1654 error("invalid chunk length %llu", length);
1657 if (stripe_len != BTRFS_STRIPE_LEN) {
1658 error("invalid chunk stripe length: %llu", stripe_len);
1661 /* Check on chunk item type */
1662 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1663 error("invalid chunk type %llu", type);
1666 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1667 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1668 error("unrecognized chunk type: %llu",
1669 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1670 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1674 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1675 * it can't exceed the system chunk array size
1676 * For normal chunk, it should match its chunk item size.
1678 if (num_stripes < 1 ||
1679 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1680 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1681 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1682 btrfs_item_size_nr(leaf, slot))) {
1683 error("invalid num_stripes: %u", num_stripes);
1687 * Device number check against profile
1689 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1690 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1691 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1692 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1693 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1694 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1695 num_stripes != 1)) {
1696 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1697 num_stripes, sub_stripes,
1698 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1706 * Slot is used to verify the chunk item is valid
1708 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1710 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1711 struct extent_buffer *leaf,
1712 struct btrfs_chunk *chunk, int slot)
1714 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1715 struct map_lookup *map;
1716 struct cache_extent *ce;
1720 u8 uuid[BTRFS_UUID_SIZE];
1725 logical = key->offset;
1726 length = btrfs_chunk_length(leaf, chunk);
1727 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1728 /* Validation check */
1729 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1731 error("%s checksums match, but it has an invalid chunk, %s",
1732 (slot == -1) ? "Superblock" : "Metadata",
1733 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1737 ce = search_cache_extent(&map_tree->cache_tree, logical);
1739 /* already mapped? */
1740 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1744 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1748 map->ce.start = logical;
1749 map->ce.size = length;
1750 map->num_stripes = num_stripes;
1751 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1752 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1753 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1754 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1755 map->type = btrfs_chunk_type(leaf, chunk);
1756 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1758 for (i = 0; i < num_stripes; i++) {
1759 map->stripes[i].physical =
1760 btrfs_stripe_offset_nr(leaf, chunk, i);
1761 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1762 read_extent_buffer(leaf, uuid, (unsigned long)
1763 btrfs_stripe_dev_uuid_nr(chunk, i),
1765 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1767 if (!map->stripes[i].dev) {
1768 map->stripes[i].dev = fill_missing_device(devid);
1769 printf("warning, device %llu is missing\n",
1770 (unsigned long long)devid);
1771 list_add(&map->stripes[i].dev->dev_list,
1772 &root->fs_info->fs_devices->devices);
1776 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1782 static int fill_device_from_item(struct extent_buffer *leaf,
1783 struct btrfs_dev_item *dev_item,
1784 struct btrfs_device *device)
1788 device->devid = btrfs_device_id(leaf, dev_item);
1789 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1790 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1791 device->type = btrfs_device_type(leaf, dev_item);
1792 device->io_align = btrfs_device_io_align(leaf, dev_item);
1793 device->io_width = btrfs_device_io_width(leaf, dev_item);
1794 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1796 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1797 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1802 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1804 struct btrfs_fs_devices *fs_devices;
1807 fs_devices = root->fs_info->fs_devices->seed;
1808 while (fs_devices) {
1809 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1813 fs_devices = fs_devices->seed;
1816 fs_devices = find_fsid(fsid);
1818 /* missing all seed devices */
1819 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1824 INIT_LIST_HEAD(&fs_devices->devices);
1825 list_add(&fs_devices->list, &fs_uuids);
1826 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1829 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1833 fs_devices->seed = root->fs_info->fs_devices->seed;
1834 root->fs_info->fs_devices->seed = fs_devices;
1839 static int read_one_dev(struct btrfs_root *root,
1840 struct extent_buffer *leaf,
1841 struct btrfs_dev_item *dev_item)
1843 struct btrfs_device *device;
1846 u8 fs_uuid[BTRFS_UUID_SIZE];
1847 u8 dev_uuid[BTRFS_UUID_SIZE];
1849 devid = btrfs_device_id(leaf, dev_item);
1850 read_extent_buffer(leaf, dev_uuid,
1851 (unsigned long)btrfs_device_uuid(dev_item),
1853 read_extent_buffer(leaf, fs_uuid,
1854 (unsigned long)btrfs_device_fsid(dev_item),
1857 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1858 ret = open_seed_devices(root, fs_uuid);
1863 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1865 device = kzalloc(sizeof(*device), GFP_NOFS);
1869 list_add(&device->dev_list,
1870 &root->fs_info->fs_devices->devices);
1873 fill_device_from_item(leaf, dev_item, device);
1874 device->dev_root = root->fs_info->dev_root;
1878 int btrfs_read_sys_array(struct btrfs_root *root)
1880 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1881 struct extent_buffer *sb;
1882 struct btrfs_disk_key *disk_key;
1883 struct btrfs_chunk *chunk;
1885 unsigned long sb_array_offset;
1891 struct btrfs_key key;
1893 sb = btrfs_find_create_tree_block(root->fs_info,
1894 BTRFS_SUPER_INFO_OFFSET,
1895 BTRFS_SUPER_INFO_SIZE);
1898 btrfs_set_buffer_uptodate(sb);
1899 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1900 array_size = btrfs_super_sys_array_size(super_copy);
1902 array_ptr = super_copy->sys_chunk_array;
1903 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1906 while (cur_offset < array_size) {
1907 disk_key = (struct btrfs_disk_key *)array_ptr;
1908 len = sizeof(*disk_key);
1909 if (cur_offset + len > array_size)
1910 goto out_short_read;
1912 btrfs_disk_key_to_cpu(&key, disk_key);
1915 sb_array_offset += len;
1918 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1919 chunk = (struct btrfs_chunk *)sb_array_offset;
1921 * At least one btrfs_chunk with one stripe must be
1922 * present, exact stripe count check comes afterwards
1924 len = btrfs_chunk_item_size(1);
1925 if (cur_offset + len > array_size)
1926 goto out_short_read;
1928 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1931 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1932 num_stripes, cur_offset);
1937 len = btrfs_chunk_item_size(num_stripes);
1938 if (cur_offset + len > array_size)
1939 goto out_short_read;
1941 ret = read_one_chunk(root, &key, sb, chunk, -1);
1946 "ERROR: unexpected item type %u in sys_array at offset %u\n",
1947 (u32)key.type, cur_offset);
1952 sb_array_offset += len;
1955 free_extent_buffer(sb);
1959 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
1961 free_extent_buffer(sb);
1965 int btrfs_read_chunk_tree(struct btrfs_root *root)
1967 struct btrfs_path *path;
1968 struct extent_buffer *leaf;
1969 struct btrfs_key key;
1970 struct btrfs_key found_key;
1974 root = root->fs_info->chunk_root;
1976 path = btrfs_alloc_path();
1981 * Read all device items, and then all the chunk items. All
1982 * device items are found before any chunk item (their object id
1983 * is smaller than the lowest possible object id for a chunk
1984 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1986 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1989 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1993 leaf = path->nodes[0];
1994 slot = path->slots[0];
1995 if (slot >= btrfs_header_nritems(leaf)) {
1996 ret = btrfs_next_leaf(root, path);
2003 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2004 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2005 struct btrfs_dev_item *dev_item;
2006 dev_item = btrfs_item_ptr(leaf, slot,
2007 struct btrfs_dev_item);
2008 ret = read_one_dev(root, leaf, dev_item);
2010 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2011 struct btrfs_chunk *chunk;
2012 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2013 ret = read_one_chunk(root, &found_key, leaf, chunk,
2022 btrfs_free_path(path);
2026 struct list_head *btrfs_scanned_uuids(void)
2031 static int rmw_eb(struct btrfs_fs_info *info,
2032 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2035 unsigned long orig_off = 0;
2036 unsigned long dest_off = 0;
2037 unsigned long copy_len = eb->len;
2039 ret = read_whole_eb(info, eb, 0);
2043 if (eb->start + eb->len <= orig_eb->start ||
2044 eb->start >= orig_eb->start + orig_eb->len)
2047 * | ----- orig_eb ------- |
2048 * | ----- stripe ------- |
2049 * | ----- orig_eb ------- |
2050 * | ----- orig_eb ------- |
2052 if (eb->start > orig_eb->start)
2053 orig_off = eb->start - orig_eb->start;
2054 if (orig_eb->start > eb->start)
2055 dest_off = orig_eb->start - eb->start;
2057 if (copy_len > orig_eb->len - orig_off)
2058 copy_len = orig_eb->len - orig_off;
2059 if (copy_len > eb->len - dest_off)
2060 copy_len = eb->len - dest_off;
2062 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2066 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2067 struct extent_buffer *orig_eb,
2068 struct extent_buffer **ebs,
2069 u64 stripe_len, u64 *raid_map,
2072 struct extent_buffer **tmp_ebs;
2073 u64 start = orig_eb->start;
2078 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2082 /* Alloc memory in a row for data stripes */
2083 for (i = 0; i < num_stripes; i++) {
2084 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2087 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2094 for (i = 0; i < num_stripes; i++) {
2095 struct extent_buffer *eb = tmp_ebs[i];
2097 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2100 eb->start = raid_map[i];
2101 eb->len = stripe_len;
2105 eb->dev_bytenr = (u64)-1;
2107 this_eb_start = raid_map[i];
2109 if (start > this_eb_start ||
2110 start + orig_eb->len < this_eb_start + stripe_len) {
2111 ret = rmw_eb(info, eb, orig_eb);
2115 memcpy(eb->data, orig_eb->data + eb->start - start,
2123 for (i = 0; i < num_stripes; i++)
2129 int write_raid56_with_parity(struct btrfs_fs_info *info,
2130 struct extent_buffer *eb,
2131 struct btrfs_multi_bio *multi,
2132 u64 stripe_len, u64 *raid_map)
2134 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2138 int alloc_size = eb->len;
2141 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2142 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2143 if (!ebs || !pointers)
2146 if (stripe_len > alloc_size)
2147 alloc_size = stripe_len;
2149 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2150 multi->num_stripes);
2154 for (i = 0; i < multi->num_stripes; i++) {
2155 struct extent_buffer *new_eb;
2156 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2157 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2158 ebs[i]->fd = multi->stripes[i].dev->fd;
2159 multi->stripes[i].dev->total_ios++;
2160 if (ebs[i]->start != raid_map[i]) {
2162 goto out_free_split;
2166 new_eb = malloc(sizeof(*eb) + alloc_size);
2169 goto out_free_split;
2171 new_eb->dev_bytenr = multi->stripes[i].physical;
2172 new_eb->fd = multi->stripes[i].dev->fd;
2173 multi->stripes[i].dev->total_ios++;
2174 new_eb->len = stripe_len;
2176 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2178 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2182 ebs[multi->num_stripes - 2] = p_eb;
2183 ebs[multi->num_stripes - 1] = q_eb;
2185 for (i = 0; i < multi->num_stripes; i++)
2186 pointers[i] = ebs[i]->data;
2188 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2190 ebs[multi->num_stripes - 1] = p_eb;
2191 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2192 for (j = 1; j < multi->num_stripes - 1; j++) {
2193 for (i = 0; i < stripe_len; i += sizeof(u64)) {
2197 p_eb_data = get_unaligned_64(p_eb->data + i);
2198 ebs_data = get_unaligned_64(ebs[j]->data + i);
2199 p_eb_data ^= ebs_data;
2200 put_unaligned_64(p_eb_data, p_eb->data + i);
2205 for (i = 0; i < multi->num_stripes; i++) {
2206 ret = write_extent_to_disk(ebs[i]);
2208 goto out_free_split;
2212 for (i = 0; i < multi->num_stripes; i++) {