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 list_head *cur;
164 struct btrfs_device *device;
166 list_for_each(cur, &fs_devices->devices) {
167 device = list_entry(cur, struct btrfs_device, dev_list);
168 if (device->fd != -1) {
170 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
171 fprintf(stderr, "Warning, could not drop caches\n");
175 device->writeable = 0;
178 seed_devices = fs_devices->seed;
179 fs_devices->seed = NULL;
181 fs_devices = seed_devices;
188 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
191 struct list_head *head = &fs_devices->devices;
192 struct list_head *cur;
193 struct btrfs_device *device;
196 list_for_each(cur, head) {
197 device = list_entry(cur, struct btrfs_device, dev_list);
199 printk("no name for device %llu, skip it now\n", device->devid);
203 fd = open(device->name, flags);
209 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
210 fprintf(stderr, "Warning, could not drop caches\n");
212 if (device->devid == fs_devices->latest_devid)
213 fs_devices->latest_bdev = fd;
214 if (device->devid == fs_devices->lowest_devid)
215 fs_devices->lowest_bdev = fd;
218 device->writeable = 1;
222 btrfs_close_devices(fs_devices);
226 int btrfs_scan_one_device(int fd, const char *path,
227 struct btrfs_fs_devices **fs_devices_ret,
228 u64 *total_devs, u64 super_offset)
230 struct btrfs_super_block *disk_super;
240 disk_super = (struct btrfs_super_block *)buf;
241 ret = btrfs_read_dev_super(fd, disk_super, super_offset);
246 devid = btrfs_stack_device_id(&disk_super->dev_item);
247 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
250 *total_devs = btrfs_super_num_devices(disk_super);
252 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
261 * this uses a pretty simple search, the expectation is that it is
262 * called very infrequently and that a given device has a small number
265 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
266 struct btrfs_device *device,
267 struct btrfs_path *path,
268 u64 num_bytes, u64 *start)
270 struct btrfs_key key;
271 struct btrfs_root *root = device->dev_root;
272 struct btrfs_dev_extent *dev_extent = NULL;
275 u64 search_start = 0;
276 u64 search_end = device->total_bytes;
280 struct extent_buffer *l;
285 /* FIXME use last free of some kind */
287 /* we don't want to overwrite the superblock on the drive,
288 * so we make sure to start at an offset of at least 1MB
290 search_start = max((u64)1024 * 1024, search_start);
292 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
293 search_start = max(root->fs_info->alloc_start, search_start);
295 key.objectid = device->devid;
296 key.offset = search_start;
297 key.type = BTRFS_DEV_EXTENT_KEY;
298 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
301 ret = btrfs_previous_item(root, path, 0, key.type);
305 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
308 slot = path->slots[0];
309 if (slot >= btrfs_header_nritems(l)) {
310 ret = btrfs_next_leaf(root, path);
317 if (search_start >= search_end) {
321 *start = search_start;
325 *start = last_byte > search_start ?
326 last_byte : search_start;
327 if (search_end <= *start) {
333 btrfs_item_key_to_cpu(l, &key, slot);
335 if (key.objectid < device->devid)
338 if (key.objectid > device->devid)
341 if (key.offset >= search_start && key.offset > last_byte &&
343 if (last_byte < search_start)
344 last_byte = search_start;
345 hole_size = key.offset - last_byte;
346 if (key.offset > last_byte &&
347 hole_size >= num_bytes) {
352 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
357 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
358 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
364 /* we have to make sure we didn't find an extent that has already
365 * been allocated by the map tree or the original allocation
367 btrfs_release_path(path);
368 BUG_ON(*start < search_start);
370 if (*start + num_bytes > search_end) {
374 /* check for pending inserts here */
378 btrfs_release_path(path);
382 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
383 struct btrfs_device *device,
384 u64 chunk_tree, u64 chunk_objectid,
386 u64 num_bytes, u64 *start)
389 struct btrfs_path *path;
390 struct btrfs_root *root = device->dev_root;
391 struct btrfs_dev_extent *extent;
392 struct extent_buffer *leaf;
393 struct btrfs_key key;
395 path = btrfs_alloc_path();
399 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
404 key.objectid = device->devid;
406 key.type = BTRFS_DEV_EXTENT_KEY;
407 ret = btrfs_insert_empty_item(trans, root, path, &key,
411 leaf = path->nodes[0];
412 extent = btrfs_item_ptr(leaf, path->slots[0],
413 struct btrfs_dev_extent);
414 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
415 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
416 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
418 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
419 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
422 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
423 btrfs_mark_buffer_dirty(leaf);
425 btrfs_free_path(path);
429 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
431 struct btrfs_path *path;
433 struct btrfs_key key;
434 struct btrfs_chunk *chunk;
435 struct btrfs_key found_key;
437 path = btrfs_alloc_path();
440 key.objectid = objectid;
441 key.offset = (u64)-1;
442 key.type = BTRFS_CHUNK_ITEM_KEY;
444 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
450 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
454 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
456 if (found_key.objectid != objectid)
459 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
461 *offset = found_key.offset +
462 btrfs_chunk_length(path->nodes[0], chunk);
467 btrfs_free_path(path);
471 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
475 struct btrfs_key key;
476 struct btrfs_key found_key;
478 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
479 key.type = BTRFS_DEV_ITEM_KEY;
480 key.offset = (u64)-1;
482 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
488 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
493 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
495 *objectid = found_key.offset + 1;
499 btrfs_release_path(path);
504 * the device information is stored in the chunk root
505 * the btrfs_device struct should be fully filled in
507 int btrfs_add_device(struct btrfs_trans_handle *trans,
508 struct btrfs_root *root,
509 struct btrfs_device *device)
512 struct btrfs_path *path;
513 struct btrfs_dev_item *dev_item;
514 struct extent_buffer *leaf;
515 struct btrfs_key key;
519 root = root->fs_info->chunk_root;
521 path = btrfs_alloc_path();
525 ret = find_next_devid(root, path, &free_devid);
529 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
530 key.type = BTRFS_DEV_ITEM_KEY;
531 key.offset = free_devid;
533 ret = btrfs_insert_empty_item(trans, root, path, &key,
538 leaf = path->nodes[0];
539 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
541 device->devid = free_devid;
542 btrfs_set_device_id(leaf, dev_item, device->devid);
543 btrfs_set_device_generation(leaf, dev_item, 0);
544 btrfs_set_device_type(leaf, dev_item, device->type);
545 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
546 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
547 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
548 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
549 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
550 btrfs_set_device_group(leaf, dev_item, 0);
551 btrfs_set_device_seek_speed(leaf, dev_item, 0);
552 btrfs_set_device_bandwidth(leaf, dev_item, 0);
553 btrfs_set_device_start_offset(leaf, dev_item, 0);
555 ptr = (unsigned long)btrfs_device_uuid(dev_item);
556 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
557 ptr = (unsigned long)btrfs_device_fsid(dev_item);
558 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
559 btrfs_mark_buffer_dirty(leaf);
563 btrfs_free_path(path);
567 int btrfs_update_device(struct btrfs_trans_handle *trans,
568 struct btrfs_device *device)
571 struct btrfs_path *path;
572 struct btrfs_root *root;
573 struct btrfs_dev_item *dev_item;
574 struct extent_buffer *leaf;
575 struct btrfs_key key;
577 root = device->dev_root->fs_info->chunk_root;
579 path = btrfs_alloc_path();
583 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
584 key.type = BTRFS_DEV_ITEM_KEY;
585 key.offset = device->devid;
587 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
596 leaf = path->nodes[0];
597 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
599 btrfs_set_device_id(leaf, dev_item, device->devid);
600 btrfs_set_device_type(leaf, dev_item, device->type);
601 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
602 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
603 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
604 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
605 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
606 btrfs_mark_buffer_dirty(leaf);
609 btrfs_free_path(path);
613 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
614 struct btrfs_root *root,
615 struct btrfs_key *key,
616 struct btrfs_chunk *chunk, int item_size)
618 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
619 struct btrfs_disk_key disk_key;
623 array_size = btrfs_super_sys_array_size(super_copy);
624 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
627 ptr = super_copy->sys_chunk_array + array_size;
628 btrfs_cpu_key_to_disk(&disk_key, key);
629 memcpy(ptr, &disk_key, sizeof(disk_key));
630 ptr += sizeof(disk_key);
631 memcpy(ptr, chunk, item_size);
632 item_size += sizeof(disk_key);
633 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
637 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
640 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
642 else if (type & BTRFS_BLOCK_GROUP_RAID10)
643 return calc_size * (num_stripes / sub_stripes);
644 else if (type & BTRFS_BLOCK_GROUP_RAID5)
645 return calc_size * (num_stripes - 1);
646 else if (type & BTRFS_BLOCK_GROUP_RAID6)
647 return calc_size * (num_stripes - 2);
649 return calc_size * num_stripes;
653 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
655 /* TODO, add a way to store the preferred stripe size */
659 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
660 struct btrfs_root *extent_root, u64 *start,
661 u64 *num_bytes, u64 type)
664 struct btrfs_fs_info *info = extent_root->fs_info;
665 struct btrfs_root *chunk_root = info->chunk_root;
666 struct btrfs_stripe *stripes;
667 struct btrfs_device *device = NULL;
668 struct btrfs_chunk *chunk;
669 struct list_head private_devs;
670 struct list_head *dev_list = &info->fs_devices->devices;
671 struct list_head *cur;
672 struct map_lookup *map;
673 int min_stripe_size = 1 * 1024 * 1024;
674 u64 calc_size = 8 * 1024 * 1024;
676 u64 max_chunk_size = 4 * calc_size;
686 int stripe_len = 64 * 1024;
687 struct btrfs_key key;
690 if (list_empty(dev_list)) {
694 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
695 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
696 BTRFS_BLOCK_GROUP_RAID10 |
697 BTRFS_BLOCK_GROUP_DUP)) {
698 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
699 calc_size = 8 * 1024 * 1024;
700 max_chunk_size = calc_size * 2;
701 min_stripe_size = 1 * 1024 * 1024;
702 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
703 calc_size = 1024 * 1024 * 1024;
704 max_chunk_size = 10 * calc_size;
705 min_stripe_size = 64 * 1024 * 1024;
706 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
707 calc_size = 1024 * 1024 * 1024;
708 max_chunk_size = 4 * calc_size;
709 min_stripe_size = 32 * 1024 * 1024;
712 if (type & BTRFS_BLOCK_GROUP_RAID1) {
713 num_stripes = min_t(u64, 2,
714 btrfs_super_num_devices(info->super_copy));
719 if (type & BTRFS_BLOCK_GROUP_DUP) {
723 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
724 num_stripes = btrfs_super_num_devices(info->super_copy);
727 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
728 num_stripes = btrfs_super_num_devices(info->super_copy);
731 num_stripes &= ~(u32)1;
735 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
736 num_stripes = btrfs_super_num_devices(info->super_copy);
740 stripe_len = find_raid56_stripe_len(num_stripes - 1,
741 btrfs_super_stripesize(info->super_copy));
743 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
744 num_stripes = btrfs_super_num_devices(info->super_copy);
748 stripe_len = find_raid56_stripe_len(num_stripes - 2,
749 btrfs_super_stripesize(info->super_copy));
752 /* we don't want a chunk larger than 10% of the FS */
753 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
754 max_chunk_size = min(percent_max, max_chunk_size);
757 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
759 calc_size = max_chunk_size;
760 calc_size /= num_stripes;
761 calc_size /= stripe_len;
762 calc_size *= stripe_len;
764 /* we don't want tiny stripes */
765 calc_size = max_t(u64, calc_size, min_stripe_size);
767 calc_size /= stripe_len;
768 calc_size *= stripe_len;
769 INIT_LIST_HEAD(&private_devs);
770 cur = dev_list->next;
773 if (type & BTRFS_BLOCK_GROUP_DUP)
774 min_free = calc_size * 2;
776 min_free = calc_size;
778 /* build a private list of devices we will allocate from */
779 while(index < num_stripes) {
780 device = list_entry(cur, struct btrfs_device, dev_list);
781 avail = device->total_bytes - device->bytes_used;
783 if (avail >= min_free) {
784 list_move_tail(&device->dev_list, &private_devs);
786 if (type & BTRFS_BLOCK_GROUP_DUP)
788 } else if (avail > max_avail)
793 if (index < num_stripes) {
794 list_splice(&private_devs, dev_list);
795 if (index >= min_stripes) {
797 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
798 num_stripes /= sub_stripes;
799 num_stripes *= sub_stripes;
804 if (!looped && max_avail > 0) {
806 calc_size = max_avail;
811 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
815 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
816 key.type = BTRFS_CHUNK_ITEM_KEY;
819 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
823 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
829 stripes = &chunk->stripe;
830 *num_bytes = chunk_bytes_by_type(type, calc_size,
831 num_stripes, sub_stripes);
833 while(index < num_stripes) {
834 struct btrfs_stripe *stripe;
835 BUG_ON(list_empty(&private_devs));
836 cur = private_devs.next;
837 device = list_entry(cur, struct btrfs_device, dev_list);
839 /* loop over this device again if we're doing a dup group */
840 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
841 (index == num_stripes - 1))
842 list_move_tail(&device->dev_list, dev_list);
844 ret = btrfs_alloc_dev_extent(trans, device,
845 info->chunk_root->root_key.objectid,
846 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
847 calc_size, &dev_offset);
850 device->bytes_used += calc_size;
851 ret = btrfs_update_device(trans, device);
854 map->stripes[index].dev = device;
855 map->stripes[index].physical = dev_offset;
856 stripe = stripes + index;
857 btrfs_set_stack_stripe_devid(stripe, device->devid);
858 btrfs_set_stack_stripe_offset(stripe, dev_offset);
859 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
862 BUG_ON(!list_empty(&private_devs));
864 /* key was set above */
865 btrfs_set_stack_chunk_length(chunk, *num_bytes);
866 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
867 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
868 btrfs_set_stack_chunk_type(chunk, type);
869 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
870 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
871 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
872 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
873 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
874 map->sector_size = extent_root->sectorsize;
875 map->stripe_len = stripe_len;
876 map->io_align = stripe_len;
877 map->io_width = stripe_len;
879 map->num_stripes = num_stripes;
880 map->sub_stripes = sub_stripes;
882 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
883 btrfs_chunk_item_size(num_stripes));
885 *start = key.offset;;
887 map->ce.start = key.offset;
888 map->ce.size = *num_bytes;
890 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
893 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
894 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
895 chunk, btrfs_chunk_item_size(num_stripes));
903 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
904 struct btrfs_root *extent_root, u64 *start,
905 u64 num_bytes, u64 type)
908 struct btrfs_fs_info *info = extent_root->fs_info;
909 struct btrfs_root *chunk_root = info->chunk_root;
910 struct btrfs_stripe *stripes;
911 struct btrfs_device *device = NULL;
912 struct btrfs_chunk *chunk;
913 struct list_head *dev_list = &info->fs_devices->devices;
914 struct list_head *cur;
915 struct map_lookup *map;
916 u64 calc_size = 8 * 1024 * 1024;
921 int stripe_len = 64 * 1024;
922 struct btrfs_key key;
924 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
925 key.type = BTRFS_CHUNK_ITEM_KEY;
926 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
931 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
935 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
941 stripes = &chunk->stripe;
942 calc_size = num_bytes;
945 cur = dev_list->next;
946 device = list_entry(cur, struct btrfs_device, dev_list);
948 while (index < num_stripes) {
949 struct btrfs_stripe *stripe;
951 ret = btrfs_alloc_dev_extent(trans, device,
952 info->chunk_root->root_key.objectid,
953 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
954 calc_size, &dev_offset);
957 device->bytes_used += calc_size;
958 ret = btrfs_update_device(trans, device);
961 map->stripes[index].dev = device;
962 map->stripes[index].physical = dev_offset;
963 stripe = stripes + index;
964 btrfs_set_stack_stripe_devid(stripe, device->devid);
965 btrfs_set_stack_stripe_offset(stripe, dev_offset);
966 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
970 /* key was set above */
971 btrfs_set_stack_chunk_length(chunk, num_bytes);
972 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
973 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
974 btrfs_set_stack_chunk_type(chunk, type);
975 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
976 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
977 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
978 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
979 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
980 map->sector_size = extent_root->sectorsize;
981 map->stripe_len = stripe_len;
982 map->io_align = stripe_len;
983 map->io_width = stripe_len;
985 map->num_stripes = num_stripes;
986 map->sub_stripes = sub_stripes;
988 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
989 btrfs_chunk_item_size(num_stripes));
993 map->ce.start = key.offset;
994 map->ce.size = num_bytes;
996 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1003 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1005 struct cache_extent *ce;
1006 struct map_lookup *map;
1009 ce = search_cache_extent(&map_tree->cache_tree, logical);
1011 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1012 map = container_of(ce, struct map_lookup, ce);
1014 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1015 ret = map->num_stripes;
1016 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1017 ret = map->sub_stripes;
1018 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1020 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1027 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1030 struct cache_extent *ce;
1031 struct map_lookup *map;
1033 ce = search_cache_extent(&map_tree->cache_tree, *logical);
1036 ce = next_cache_extent(ce);
1040 map = container_of(ce, struct map_lookup, ce);
1041 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1042 *logical = ce->start;
1051 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1052 u64 chunk_start, u64 physical, u64 devid,
1053 u64 **logical, int *naddrs, int *stripe_len)
1055 struct cache_extent *ce;
1056 struct map_lookup *map;
1064 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1066 map = container_of(ce, struct map_lookup, ce);
1069 rmap_len = map->stripe_len;
1070 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1071 length = ce->size / (map->num_stripes / map->sub_stripes);
1072 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1073 length = ce->size / map->num_stripes;
1074 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1075 BTRFS_BLOCK_GROUP_RAID6)) {
1076 length = ce->size / nr_data_stripes(map);
1077 rmap_len = map->stripe_len * nr_data_stripes(map);
1080 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1082 for (i = 0; i < map->num_stripes; i++) {
1083 if (devid && map->stripes[i].dev->devid != devid)
1085 if (map->stripes[i].physical > physical ||
1086 map->stripes[i].physical + length <= physical)
1089 stripe_nr = (physical - map->stripes[i].physical) /
1092 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1093 stripe_nr = (stripe_nr * map->num_stripes + i) /
1095 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1096 stripe_nr = stripe_nr * map->num_stripes + i;
1097 } /* else if RAID[56], multiply by nr_data_stripes().
1098 * Alternatively, just use rmap_len below instead of
1099 * map->stripe_len */
1101 bytenr = ce->start + stripe_nr * rmap_len;
1102 for (j = 0; j < nr; j++) {
1103 if (buf[j] == bytenr)
1112 *stripe_len = rmap_len;
1117 static inline int parity_smaller(u64 a, u64 b)
1122 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1123 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1125 struct btrfs_bio_stripe s;
1132 for (i = 0; i < bbio->num_stripes - 1; i++) {
1133 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1134 s = bbio->stripes[i];
1136 bbio->stripes[i] = bbio->stripes[i+1];
1137 raid_map[i] = raid_map[i+1];
1138 bbio->stripes[i+1] = s;
1146 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1147 u64 logical, u64 *length,
1148 struct btrfs_multi_bio **multi_ret, int mirror_num,
1151 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1152 multi_ret, mirror_num, raid_map_ret);
1155 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1156 u64 logical, u64 *length, u64 *type,
1157 struct btrfs_multi_bio **multi_ret, int mirror_num,
1160 struct cache_extent *ce;
1161 struct map_lookup *map;
1165 u64 *raid_map = NULL;
1166 int stripes_allocated = 8;
1167 int stripes_required = 1;
1170 struct btrfs_multi_bio *multi = NULL;
1172 if (multi_ret && rw == READ) {
1173 stripes_allocated = 1;
1176 ce = search_cache_extent(&map_tree->cache_tree, logical);
1182 if (ce->start > logical || ce->start + ce->size < logical) {
1189 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1194 map = container_of(ce, struct map_lookup, ce);
1195 offset = logical - ce->start;
1198 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1199 BTRFS_BLOCK_GROUP_DUP)) {
1200 stripes_required = map->num_stripes;
1201 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1202 stripes_required = map->sub_stripes;
1205 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1206 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1207 /* RAID[56] write or recovery. Return all stripes */
1208 stripes_required = map->num_stripes;
1210 /* Only allocate the map if we've already got a large enough multi_ret */
1211 if (stripes_allocated >= stripes_required) {
1212 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1220 /* if our multi bio struct is too small, back off and try again */
1221 if (multi_ret && stripes_allocated < stripes_required) {
1222 stripes_allocated = stripes_required;
1229 * stripe_nr counts the total number of stripes we have to stride
1230 * to get to this block
1232 stripe_nr = stripe_nr / map->stripe_len;
1234 stripe_offset = stripe_nr * map->stripe_len;
1235 BUG_ON(offset < stripe_offset);
1237 /* stripe_offset is the offset of this block in its stripe*/
1238 stripe_offset = offset - stripe_offset;
1240 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1241 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1242 BTRFS_BLOCK_GROUP_RAID10 |
1243 BTRFS_BLOCK_GROUP_DUP)) {
1244 /* we limit the length of each bio to what fits in a stripe */
1245 *length = min_t(u64, ce->size - offset,
1246 map->stripe_len - stripe_offset);
1248 *length = ce->size - offset;
1254 multi->num_stripes = 1;
1256 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1258 multi->num_stripes = map->num_stripes;
1259 else if (mirror_num)
1260 stripe_index = mirror_num - 1;
1262 stripe_index = stripe_nr % map->num_stripes;
1263 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1264 int factor = map->num_stripes / map->sub_stripes;
1266 stripe_index = stripe_nr % factor;
1267 stripe_index *= map->sub_stripes;
1270 multi->num_stripes = map->sub_stripes;
1271 else if (mirror_num)
1272 stripe_index += mirror_num - 1;
1274 stripe_nr = stripe_nr / factor;
1275 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1277 multi->num_stripes = map->num_stripes;
1278 else if (mirror_num)
1279 stripe_index = mirror_num - 1;
1280 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1281 BTRFS_BLOCK_GROUP_RAID6)) {
1286 u64 raid56_full_stripe_start;
1287 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1290 * align the start of our data stripe in the logical
1293 raid56_full_stripe_start = offset / full_stripe_len;
1294 raid56_full_stripe_start *= full_stripe_len;
1296 /* get the data stripe number */
1297 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1298 stripe_nr = stripe_nr / nr_data_stripes(map);
1300 /* Work out the disk rotation on this stripe-set */
1301 rot = stripe_nr % map->num_stripes;
1303 /* Fill in the logical address of each stripe */
1304 tmp = stripe_nr * nr_data_stripes(map);
1306 for (i = 0; i < nr_data_stripes(map); i++)
1307 raid_map[(i+rot) % map->num_stripes] =
1308 ce->start + (tmp + i) * map->stripe_len;
1310 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1311 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1312 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1314 *length = map->stripe_len;
1317 multi->num_stripes = map->num_stripes;
1319 stripe_index = stripe_nr % nr_data_stripes(map);
1320 stripe_nr = stripe_nr / nr_data_stripes(map);
1323 * Mirror #0 or #1 means the original data block.
1324 * Mirror #2 is RAID5 parity block.
1325 * Mirror #3 is RAID6 Q block.
1328 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1330 /* We distribute the parity blocks across stripes */
1331 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1335 * after this do_div call, stripe_nr is the number of stripes
1336 * on this device we have to walk to find the data, and
1337 * stripe_index is the number of our device in the stripe array
1339 stripe_index = stripe_nr % map->num_stripes;
1340 stripe_nr = stripe_nr / map->num_stripes;
1342 BUG_ON(stripe_index >= map->num_stripes);
1344 for (i = 0; i < multi->num_stripes; i++) {
1345 multi->stripes[i].physical =
1346 map->stripes[stripe_index].physical + stripe_offset +
1347 stripe_nr * map->stripe_len;
1348 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1357 sort_parity_stripes(multi, raid_map);
1358 *raid_map_ret = raid_map;
1364 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1367 struct btrfs_device *device;
1368 struct btrfs_fs_devices *cur_devices;
1370 cur_devices = root->fs_info->fs_devices;
1371 while (cur_devices) {
1373 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1374 device = __find_device(&cur_devices->devices,
1379 cur_devices = cur_devices->seed;
1384 struct btrfs_device *
1385 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1386 u64 devid, int instance)
1388 struct list_head *head = &fs_devices->devices;
1389 struct btrfs_device *dev;
1392 list_for_each_entry(dev, head, dev_list) {
1393 if (dev->devid == devid && num_found++ == instance)
1399 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1401 struct cache_extent *ce;
1402 struct map_lookup *map;
1403 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1407 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1410 map = container_of(ce, struct map_lookup, ce);
1411 for (i = 0; i < map->num_stripes; i++) {
1412 if (!map->stripes[i].dev->writeable) {
1421 static struct btrfs_device *fill_missing_device(u64 devid)
1423 struct btrfs_device *device;
1425 device = kzalloc(sizeof(*device), GFP_NOFS);
1426 device->devid = devid;
1431 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1432 struct extent_buffer *leaf,
1433 struct btrfs_chunk *chunk)
1435 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1436 struct map_lookup *map;
1437 struct cache_extent *ce;
1441 u8 uuid[BTRFS_UUID_SIZE];
1446 logical = key->offset;
1447 length = btrfs_chunk_length(leaf, chunk);
1449 ce = search_cache_extent(&map_tree->cache_tree, logical);
1451 /* already mapped? */
1452 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1456 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1457 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1461 map->ce.start = logical;
1462 map->ce.size = length;
1463 map->num_stripes = num_stripes;
1464 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1465 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1466 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1467 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1468 map->type = btrfs_chunk_type(leaf, chunk);
1469 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1471 for (i = 0; i < num_stripes; i++) {
1472 map->stripes[i].physical =
1473 btrfs_stripe_offset_nr(leaf, chunk, i);
1474 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1475 read_extent_buffer(leaf, uuid, (unsigned long)
1476 btrfs_stripe_dev_uuid_nr(chunk, i),
1478 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1480 if (!map->stripes[i].dev) {
1481 map->stripes[i].dev = fill_missing_device(devid);
1482 printf("warning, device %llu is missing\n",
1483 (unsigned long long)devid);
1487 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1493 static int fill_device_from_item(struct extent_buffer *leaf,
1494 struct btrfs_dev_item *dev_item,
1495 struct btrfs_device *device)
1499 device->devid = btrfs_device_id(leaf, dev_item);
1500 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1501 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1502 device->type = btrfs_device_type(leaf, dev_item);
1503 device->io_align = btrfs_device_io_align(leaf, dev_item);
1504 device->io_width = btrfs_device_io_width(leaf, dev_item);
1505 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1507 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1508 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1513 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1515 struct btrfs_fs_devices *fs_devices;
1518 fs_devices = root->fs_info->fs_devices->seed;
1519 while (fs_devices) {
1520 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1524 fs_devices = fs_devices->seed;
1527 fs_devices = find_fsid(fsid);
1533 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1537 fs_devices->seed = root->fs_info->fs_devices->seed;
1538 root->fs_info->fs_devices->seed = fs_devices;
1543 static int read_one_dev(struct btrfs_root *root,
1544 struct extent_buffer *leaf,
1545 struct btrfs_dev_item *dev_item)
1547 struct btrfs_device *device;
1550 u8 fs_uuid[BTRFS_UUID_SIZE];
1551 u8 dev_uuid[BTRFS_UUID_SIZE];
1553 devid = btrfs_device_id(leaf, dev_item);
1554 read_extent_buffer(leaf, dev_uuid,
1555 (unsigned long)btrfs_device_uuid(dev_item),
1557 read_extent_buffer(leaf, fs_uuid,
1558 (unsigned long)btrfs_device_fsid(dev_item),
1561 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1562 ret = open_seed_devices(root, fs_uuid);
1567 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1569 printk("warning devid %llu not found already\n",
1570 (unsigned long long)devid);
1571 device = kzalloc(sizeof(*device), GFP_NOFS);
1575 list_add(&device->dev_list,
1576 &root->fs_info->fs_devices->devices);
1579 fill_device_from_item(leaf, dev_item, device);
1580 device->dev_root = root->fs_info->dev_root;
1584 int btrfs_read_sys_array(struct btrfs_root *root)
1586 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1587 struct extent_buffer *sb;
1588 struct btrfs_disk_key *disk_key;
1589 struct btrfs_chunk *chunk;
1590 struct btrfs_key key;
1597 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1598 BTRFS_SUPER_INFO_SIZE);
1601 btrfs_set_buffer_uptodate(sb);
1602 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1603 array_end = ((u8 *)super_copy->sys_chunk_array) +
1604 btrfs_super_sys_array_size(super_copy);
1607 * we do this loop twice, once for the device items and
1608 * once for all of the chunks. This way there are device
1609 * structs filled in for every chunk
1611 ptr = super_copy->sys_chunk_array;
1613 while (ptr < array_end) {
1614 disk_key = (struct btrfs_disk_key *)ptr;
1615 btrfs_disk_key_to_cpu(&key, disk_key);
1617 len = sizeof(*disk_key);
1620 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1621 chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1622 ret = read_one_chunk(root, &key, sb, chunk);
1625 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1626 len = btrfs_chunk_item_size(num_stripes);
1632 free_extent_buffer(sb);
1636 int btrfs_read_chunk_tree(struct btrfs_root *root)
1638 struct btrfs_path *path;
1639 struct extent_buffer *leaf;
1640 struct btrfs_key key;
1641 struct btrfs_key found_key;
1645 root = root->fs_info->chunk_root;
1647 path = btrfs_alloc_path();
1652 * Read all device items, and then all the chunk items. All
1653 * device items are found before any chunk item (their object id
1654 * is smaller than the lowest possible object id for a chunk
1655 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1657 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1660 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1664 leaf = path->nodes[0];
1665 slot = path->slots[0];
1666 if (slot >= btrfs_header_nritems(leaf)) {
1667 ret = btrfs_next_leaf(root, path);
1674 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1675 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1676 struct btrfs_dev_item *dev_item;
1677 dev_item = btrfs_item_ptr(leaf, slot,
1678 struct btrfs_dev_item);
1679 ret = read_one_dev(root, leaf, dev_item);
1681 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1682 struct btrfs_chunk *chunk;
1683 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1684 ret = read_one_chunk(root, &found_key, leaf, chunk);
1692 btrfs_free_path(path);
1696 struct list_head *btrfs_scanned_uuids(void)
1701 static int rmw_eb(struct btrfs_fs_info *info,
1702 struct extent_buffer *eb, struct extent_buffer *orig_eb)
1705 unsigned long orig_off = 0;
1706 unsigned long dest_off = 0;
1707 unsigned long copy_len = eb->len;
1709 ret = read_whole_eb(info, eb, 0);
1713 if (eb->start + eb->len <= orig_eb->start ||
1714 eb->start >= orig_eb->start + orig_eb->len)
1717 * | ----- orig_eb ------- |
1718 * | ----- stripe ------- |
1719 * | ----- orig_eb ------- |
1720 * | ----- orig_eb ------- |
1722 if (eb->start > orig_eb->start)
1723 orig_off = eb->start - orig_eb->start;
1724 if (orig_eb->start > eb->start)
1725 dest_off = orig_eb->start - eb->start;
1727 if (copy_len > orig_eb->len - orig_off)
1728 copy_len = orig_eb->len - orig_off;
1729 if (copy_len > eb->len - dest_off)
1730 copy_len = eb->len - dest_off;
1732 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1736 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1737 struct extent_buffer *orig_eb,
1738 struct extent_buffer **ebs,
1739 u64 stripe_len, u64 *raid_map,
1742 struct extent_buffer *eb;
1743 u64 start = orig_eb->start;
1748 for (i = 0; i < num_stripes; i++) {
1749 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1752 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1755 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1757 eb->start = raid_map[i];
1758 eb->len = stripe_len;
1762 eb->dev_bytenr = (u64)-1;
1764 this_eb_start = raid_map[i];
1766 if (start > this_eb_start ||
1767 start + orig_eb->len < this_eb_start + stripe_len) {
1768 ret = rmw_eb(info, eb, orig_eb);
1771 memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1777 int write_raid56_with_parity(struct btrfs_fs_info *info,
1778 struct extent_buffer *eb,
1779 struct btrfs_multi_bio *multi,
1780 u64 stripe_len, u64 *raid_map)
1782 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1786 int alloc_size = eb->len;
1788 ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1791 if (stripe_len > alloc_size)
1792 alloc_size = stripe_len;
1794 split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1795 multi->num_stripes);
1797 for (i = 0; i < multi->num_stripes; i++) {
1798 struct extent_buffer *new_eb;
1799 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1800 ebs[i]->dev_bytenr = multi->stripes[i].physical;
1801 ebs[i]->fd = multi->stripes[i].dev->fd;
1802 multi->stripes[i].dev->total_ios++;
1803 BUG_ON(ebs[i]->start != raid_map[i]);
1806 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1808 new_eb->dev_bytenr = multi->stripes[i].physical;
1809 new_eb->fd = multi->stripes[i].dev->fd;
1810 multi->stripes[i].dev->total_ios++;
1811 new_eb->len = stripe_len;
1813 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1815 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1821 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
1825 ebs[multi->num_stripes - 2] = p_eb;
1826 ebs[multi->num_stripes - 1] = q_eb;
1828 for (i = 0; i < multi->num_stripes; i++)
1829 pointers[i] = ebs[i]->data;
1831 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1834 ebs[multi->num_stripes - 1] = p_eb;
1835 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1836 for (j = 1; j < multi->num_stripes - 1; j++) {
1837 for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1838 *(unsigned long *)(p_eb->data + i) ^=
1839 *(unsigned long *)(ebs[j]->data + i);
1844 for (i = 0; i < multi->num_stripes; i++) {
1845 ret = write_extent_to_disk(ebs[i]);