2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/bio.h>
20 #include <linux/buffer_head.h>
21 #include <linux/blkdev.h>
22 #include <asm/div64.h>
24 #include "extent_map.h"
26 #include "transaction.h"
27 #include "print-tree.h"
38 struct btrfs_bio_stripe stripes
[];
41 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
42 (sizeof(struct btrfs_bio_stripe) * (n)))
44 static DEFINE_MUTEX(uuid_mutex
);
45 static LIST_HEAD(fs_uuids
);
47 int btrfs_cleanup_fs_uuids(void)
49 struct btrfs_fs_devices
*fs_devices
;
50 struct list_head
*uuid_cur
;
51 struct list_head
*devices_cur
;
52 struct btrfs_device
*dev
;
54 list_for_each(uuid_cur
, &fs_uuids
) {
55 fs_devices
= list_entry(uuid_cur
, struct btrfs_fs_devices
,
57 while(!list_empty(&fs_devices
->devices
)) {
58 devices_cur
= fs_devices
->devices
.next
;
59 dev
= list_entry(devices_cur
, struct btrfs_device
,
62 close_bdev_excl(dev
->bdev
);
64 list_del(&dev
->dev_list
);
71 static struct btrfs_device
*__find_device(struct list_head
*head
, u64 devid
,
74 struct btrfs_device
*dev
;
75 struct list_head
*cur
;
77 list_for_each(cur
, head
) {
78 dev
= list_entry(cur
, struct btrfs_device
, dev_list
);
79 if (dev
->devid
== devid
&&
80 !memcmp(dev
->uuid
, uuid
, BTRFS_UUID_SIZE
)) {
87 static struct btrfs_fs_devices
*find_fsid(u8
*fsid
)
89 struct list_head
*cur
;
90 struct btrfs_fs_devices
*fs_devices
;
92 list_for_each(cur
, &fs_uuids
) {
93 fs_devices
= list_entry(cur
, struct btrfs_fs_devices
, list
);
94 if (memcmp(fsid
, fs_devices
->fsid
, BTRFS_FSID_SIZE
) == 0)
100 static int device_list_add(const char *path
,
101 struct btrfs_super_block
*disk_super
,
102 u64 devid
, struct btrfs_fs_devices
**fs_devices_ret
)
104 struct btrfs_device
*device
;
105 struct btrfs_fs_devices
*fs_devices
;
106 u64 found_transid
= btrfs_super_generation(disk_super
);
108 fs_devices
= find_fsid(disk_super
->fsid
);
110 fs_devices
= kmalloc(sizeof(*fs_devices
), GFP_NOFS
);
113 INIT_LIST_HEAD(&fs_devices
->devices
);
114 INIT_LIST_HEAD(&fs_devices
->alloc_list
);
115 list_add(&fs_devices
->list
, &fs_uuids
);
116 memcpy(fs_devices
->fsid
, disk_super
->fsid
, BTRFS_FSID_SIZE
);
117 fs_devices
->latest_devid
= devid
;
118 fs_devices
->latest_trans
= found_transid
;
119 fs_devices
->lowest_devid
= (u64
)-1;
120 fs_devices
->num_devices
= 0;
123 device
= __find_device(&fs_devices
->devices
, devid
,
124 disk_super
->dev_item
.uuid
);
127 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
129 /* we can safely leave the fs_devices entry around */
132 device
->devid
= devid
;
133 memcpy(device
->uuid
, disk_super
->dev_item
.uuid
,
135 device
->barriers
= 1;
136 spin_lock_init(&device
->io_lock
);
137 device
->name
= kstrdup(path
, GFP_NOFS
);
142 list_add(&device
->dev_list
, &fs_devices
->devices
);
143 list_add(&device
->dev_alloc_list
, &fs_devices
->alloc_list
);
144 fs_devices
->num_devices
++;
147 if (found_transid
> fs_devices
->latest_trans
) {
148 fs_devices
->latest_devid
= devid
;
149 fs_devices
->latest_trans
= found_transid
;
151 if (fs_devices
->lowest_devid
> devid
) {
152 fs_devices
->lowest_devid
= devid
;
154 *fs_devices_ret
= fs_devices
;
158 int btrfs_close_devices(struct btrfs_fs_devices
*fs_devices
)
160 struct list_head
*head
= &fs_devices
->devices
;
161 struct list_head
*cur
;
162 struct btrfs_device
*device
;
164 mutex_lock(&uuid_mutex
);
165 list_for_each(cur
, head
) {
166 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
168 close_bdev_excl(device
->bdev
);
172 mutex_unlock(&uuid_mutex
);
176 int btrfs_open_devices(struct btrfs_fs_devices
*fs_devices
,
177 int flags
, void *holder
)
179 struct block_device
*bdev
;
180 struct list_head
*head
= &fs_devices
->devices
;
181 struct list_head
*cur
;
182 struct btrfs_device
*device
;
185 mutex_lock(&uuid_mutex
);
186 list_for_each(cur
, head
) {
187 device
= list_entry(cur
, struct btrfs_device
, dev_list
);
188 bdev
= open_bdev_excl(device
->name
, flags
, holder
);
191 printk("open %s failed\n", device
->name
);
195 if (device
->devid
== fs_devices
->latest_devid
)
196 fs_devices
->latest_bdev
= bdev
;
197 if (device
->devid
== fs_devices
->lowest_devid
) {
198 fs_devices
->lowest_bdev
= bdev
;
202 mutex_unlock(&uuid_mutex
);
205 mutex_unlock(&uuid_mutex
);
206 btrfs_close_devices(fs_devices
);
210 int btrfs_scan_one_device(const char *path
, int flags
, void *holder
,
211 struct btrfs_fs_devices
**fs_devices_ret
)
213 struct btrfs_super_block
*disk_super
;
214 struct block_device
*bdev
;
215 struct buffer_head
*bh
;
220 mutex_lock(&uuid_mutex
);
222 bdev
= open_bdev_excl(path
, flags
, holder
);
229 ret
= set_blocksize(bdev
, 4096);
232 bh
= __bread(bdev
, BTRFS_SUPER_INFO_OFFSET
/ 4096, 4096);
237 disk_super
= (struct btrfs_super_block
*)bh
->b_data
;
238 if (strncmp((char *)(&disk_super
->magic
), BTRFS_MAGIC
,
239 sizeof(disk_super
->magic
))) {
243 devid
= le64_to_cpu(disk_super
->dev_item
.devid
);
244 transid
= btrfs_super_generation(disk_super
);
245 if (disk_super
->label
[0])
246 printk("device label %s ", disk_super
->label
);
248 /* FIXME, make a readl uuid parser */
249 printk("device fsid %llx-%llx ",
250 *(unsigned long long *)disk_super
->fsid
,
251 *(unsigned long long *)(disk_super
->fsid
+ 8));
253 printk("devid %Lu transid %Lu %s\n", devid
, transid
, path
);
254 ret
= device_list_add(path
, disk_super
, devid
, fs_devices_ret
);
259 close_bdev_excl(bdev
);
261 mutex_unlock(&uuid_mutex
);
266 * this uses a pretty simple search, the expectation is that it is
267 * called very infrequently and that a given device has a small number
270 static int find_free_dev_extent(struct btrfs_trans_handle
*trans
,
271 struct btrfs_device
*device
,
272 struct btrfs_path
*path
,
273 u64 num_bytes
, u64
*start
)
275 struct btrfs_key key
;
276 struct btrfs_root
*root
= device
->dev_root
;
277 struct btrfs_dev_extent
*dev_extent
= NULL
;
280 u64 search_start
= 0;
281 u64 search_end
= device
->total_bytes
;
285 struct extent_buffer
*l
;
290 /* FIXME use last free of some kind */
292 /* we don't want to overwrite the superblock on the drive,
293 * so we make sure to start at an offset of at least 1MB
295 search_start
= max((u64
)1024 * 1024, search_start
);
296 key
.objectid
= device
->devid
;
297 key
.offset
= search_start
;
298 key
.type
= BTRFS_DEV_EXTENT_KEY
;
299 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 0);
302 ret
= btrfs_previous_item(root
, path
, 0, key
.type
);
306 btrfs_item_key_to_cpu(l
, &key
, path
->slots
[0]);
309 slot
= path
->slots
[0];
310 if (slot
>= btrfs_header_nritems(l
)) {
311 ret
= btrfs_next_leaf(root
, path
);
318 if (search_start
>= search_end
) {
322 *start
= search_start
;
326 *start
= last_byte
> search_start
?
327 last_byte
: search_start
;
328 if (search_end
<= *start
) {
334 btrfs_item_key_to_cpu(l
, &key
, slot
);
336 if (key
.objectid
< device
->devid
)
339 if (key
.objectid
> device
->devid
)
342 if (key
.offset
>= search_start
&& key
.offset
> last_byte
&&
344 if (last_byte
< search_start
)
345 last_byte
= search_start
;
346 hole_size
= key
.offset
- last_byte
;
347 if (key
.offset
> last_byte
&&
348 hole_size
>= num_bytes
) {
353 if (btrfs_key_type(&key
) != BTRFS_DEV_EXTENT_KEY
) {
358 dev_extent
= btrfs_item_ptr(l
, slot
, struct btrfs_dev_extent
);
359 last_byte
= key
.offset
+ btrfs_dev_extent_length(l
, dev_extent
);
365 /* we have to make sure we didn't find an extent that has already
366 * been allocated by the map tree or the original allocation
368 btrfs_release_path(root
, path
);
369 BUG_ON(*start
< search_start
);
371 if (*start
+ num_bytes
> search_end
) {
375 /* check for pending inserts here */
379 btrfs_release_path(root
, path
);
383 int btrfs_alloc_dev_extent(struct btrfs_trans_handle
*trans
,
384 struct btrfs_device
*device
,
385 u64 chunk_tree
, u64 chunk_objectid
,
387 u64 num_bytes
, u64
*start
)
390 struct btrfs_path
*path
;
391 struct btrfs_root
*root
= device
->dev_root
;
392 struct btrfs_dev_extent
*extent
;
393 struct extent_buffer
*leaf
;
394 struct btrfs_key key
;
396 path
= btrfs_alloc_path();
400 ret
= find_free_dev_extent(trans
, device
, path
, num_bytes
, start
);
405 key
.objectid
= device
->devid
;
407 key
.type
= BTRFS_DEV_EXTENT_KEY
;
408 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
412 leaf
= path
->nodes
[0];
413 extent
= btrfs_item_ptr(leaf
, path
->slots
[0],
414 struct btrfs_dev_extent
);
415 btrfs_set_dev_extent_chunk_tree(leaf
, extent
, chunk_tree
);
416 btrfs_set_dev_extent_chunk_objectid(leaf
, extent
, chunk_objectid
);
417 btrfs_set_dev_extent_chunk_offset(leaf
, extent
, chunk_offset
);
419 write_extent_buffer(leaf
, root
->fs_info
->chunk_tree_uuid
,
420 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent
),
423 btrfs_set_dev_extent_length(leaf
, extent
, num_bytes
);
424 btrfs_mark_buffer_dirty(leaf
);
426 btrfs_free_path(path
);
430 static int find_next_chunk(struct btrfs_root
*root
, u64 objectid
, u64
*offset
)
432 struct btrfs_path
*path
;
434 struct btrfs_key key
;
435 struct btrfs_chunk
*chunk
;
436 struct btrfs_key found_key
;
438 path
= btrfs_alloc_path();
441 key
.objectid
= objectid
;
442 key
.offset
= (u64
)-1;
443 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
445 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
451 ret
= btrfs_previous_item(root
, path
, 0, BTRFS_CHUNK_ITEM_KEY
);
455 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
457 if (found_key
.objectid
!= objectid
)
460 chunk
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
462 *offset
= found_key
.offset
+
463 btrfs_chunk_length(path
->nodes
[0], chunk
);
468 btrfs_free_path(path
);
472 static int find_next_devid(struct btrfs_root
*root
, struct btrfs_path
*path
,
476 struct btrfs_key key
;
477 struct btrfs_key found_key
;
479 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
480 key
.type
= BTRFS_DEV_ITEM_KEY
;
481 key
.offset
= (u64
)-1;
483 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
489 ret
= btrfs_previous_item(root
, path
, BTRFS_DEV_ITEMS_OBJECTID
,
494 btrfs_item_key_to_cpu(path
->nodes
[0], &found_key
,
496 *objectid
= found_key
.offset
+ 1;
500 btrfs_release_path(root
, path
);
505 * the device information is stored in the chunk root
506 * the btrfs_device struct should be fully filled in
508 int btrfs_add_device(struct btrfs_trans_handle
*trans
,
509 struct btrfs_root
*root
,
510 struct btrfs_device
*device
)
513 struct btrfs_path
*path
;
514 struct btrfs_dev_item
*dev_item
;
515 struct extent_buffer
*leaf
;
516 struct btrfs_key key
;
520 root
= root
->fs_info
->chunk_root
;
522 path
= btrfs_alloc_path();
526 ret
= find_next_devid(root
, path
, &free_devid
);
530 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
531 key
.type
= BTRFS_DEV_ITEM_KEY
;
532 key
.offset
= free_devid
;
534 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
539 leaf
= path
->nodes
[0];
540 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
542 device
->devid
= free_devid
;
543 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
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);
554 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
555 write_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
556 btrfs_mark_buffer_dirty(leaf
);
560 btrfs_free_path(path
);
563 int btrfs_update_device(struct btrfs_trans_handle
*trans
,
564 struct btrfs_device
*device
)
567 struct btrfs_path
*path
;
568 struct btrfs_root
*root
;
569 struct btrfs_dev_item
*dev_item
;
570 struct extent_buffer
*leaf
;
571 struct btrfs_key key
;
573 root
= device
->dev_root
->fs_info
->chunk_root
;
575 path
= btrfs_alloc_path();
579 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
580 key
.type
= BTRFS_DEV_ITEM_KEY
;
581 key
.offset
= device
->devid
;
583 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
592 leaf
= path
->nodes
[0];
593 dev_item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_dev_item
);
595 btrfs_set_device_id(leaf
, dev_item
, device
->devid
);
596 btrfs_set_device_type(leaf
, dev_item
, device
->type
);
597 btrfs_set_device_io_align(leaf
, dev_item
, device
->io_align
);
598 btrfs_set_device_io_width(leaf
, dev_item
, device
->io_width
);
599 btrfs_set_device_sector_size(leaf
, dev_item
, device
->sector_size
);
600 btrfs_set_device_total_bytes(leaf
, dev_item
, device
->total_bytes
);
601 btrfs_set_device_bytes_used(leaf
, dev_item
, device
->bytes_used
);
602 btrfs_mark_buffer_dirty(leaf
);
605 btrfs_free_path(path
);
609 int btrfs_add_system_chunk(struct btrfs_trans_handle
*trans
,
610 struct btrfs_root
*root
,
611 struct btrfs_key
*key
,
612 struct btrfs_chunk
*chunk
, int item_size
)
614 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
615 struct btrfs_disk_key disk_key
;
619 array_size
= btrfs_super_sys_array_size(super_copy
);
620 if (array_size
+ item_size
> BTRFS_SYSTEM_CHUNK_ARRAY_SIZE
)
623 ptr
= super_copy
->sys_chunk_array
+ array_size
;
624 btrfs_cpu_key_to_disk(&disk_key
, key
);
625 memcpy(ptr
, &disk_key
, sizeof(disk_key
));
626 ptr
+= sizeof(disk_key
);
627 memcpy(ptr
, chunk
, item_size
);
628 item_size
+= sizeof(disk_key
);
629 btrfs_set_super_sys_array_size(super_copy
, array_size
+ item_size
);
633 static u64
div_factor(u64 num
, int factor
)
642 static u64
chunk_bytes_by_type(u64 type
, u64 calc_size
, int num_stripes
,
645 if (type
& (BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_DUP
))
647 else if (type
& BTRFS_BLOCK_GROUP_RAID10
)
648 return calc_size
* (num_stripes
/ sub_stripes
);
650 return calc_size
* num_stripes
;
654 int btrfs_alloc_chunk(struct btrfs_trans_handle
*trans
,
655 struct btrfs_root
*extent_root
, u64
*start
,
656 u64
*num_bytes
, u64 type
)
659 struct btrfs_fs_info
*info
= extent_root
->fs_info
;
660 struct btrfs_root
*chunk_root
= extent_root
->fs_info
->chunk_root
;
661 struct btrfs_stripe
*stripes
;
662 struct btrfs_device
*device
= NULL
;
663 struct btrfs_chunk
*chunk
;
664 struct list_head private_devs
;
665 struct list_head
*dev_list
;
666 struct list_head
*cur
;
667 struct extent_map_tree
*em_tree
;
668 struct map_lookup
*map
;
669 struct extent_map
*em
;
670 int min_stripe_size
= 1 * 1024 * 1024;
672 u64 calc_size
= 1024 * 1024 * 1024;
673 u64 max_chunk_size
= calc_size
;
684 int stripe_len
= 64 * 1024;
685 struct btrfs_key key
;
687 dev_list
= &extent_root
->fs_info
->fs_devices
->alloc_list
;
688 if (list_empty(dev_list
))
691 if (type
& (BTRFS_BLOCK_GROUP_RAID0
)) {
692 num_stripes
= btrfs_super_num_devices(&info
->super_copy
);
695 if (type
& (BTRFS_BLOCK_GROUP_DUP
)) {
699 if (type
& (BTRFS_BLOCK_GROUP_RAID1
)) {
700 num_stripes
= min_t(u64
, 2,
701 btrfs_super_num_devices(&info
->super_copy
));
706 if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
707 num_stripes
= btrfs_super_num_devices(&info
->super_copy
);
710 num_stripes
&= ~(u32
)1;
715 if (type
& BTRFS_BLOCK_GROUP_DATA
) {
716 max_chunk_size
= 10 * calc_size
;
717 min_stripe_size
= 64 * 1024 * 1024;
718 } else if (type
& BTRFS_BLOCK_GROUP_METADATA
) {
719 max_chunk_size
= 4 * calc_size
;
720 min_stripe_size
= 32 * 1024 * 1024;
721 } else if (type
& BTRFS_BLOCK_GROUP_SYSTEM
) {
722 calc_size
= 8 * 1024 * 1024;
723 max_chunk_size
= calc_size
* 2;
724 min_stripe_size
= 1 * 1024 * 1024;
727 /* we don't want a chunk larger than 10% of the FS */
728 percent_max
= div_factor(btrfs_super_total_bytes(&info
->super_copy
), 1);
729 max_chunk_size
= min(percent_max
, max_chunk_size
);
732 if (calc_size
* num_stripes
> max_chunk_size
) {
733 calc_size
= max_chunk_size
;
734 do_div(calc_size
, num_stripes
);
735 do_div(calc_size
, stripe_len
);
736 calc_size
*= stripe_len
;
738 /* we don't want tiny stripes */
739 calc_size
= max_t(u64
, min_stripe_size
, calc_size
);
741 do_div(calc_size
, stripe_len
);
742 calc_size
*= stripe_len
;
744 INIT_LIST_HEAD(&private_devs
);
745 cur
= dev_list
->next
;
748 if (type
& BTRFS_BLOCK_GROUP_DUP
)
749 min_free
= calc_size
* 2;
751 min_free
= calc_size
;
753 /* we add 1MB because we never use the first 1MB of the device */
754 min_free
+= 1024 * 1024;
756 /* build a private list of devices we will allocate from */
757 while(index
< num_stripes
) {
758 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
760 avail
= device
->total_bytes
- device
->bytes_used
;
762 if (avail
>= min_free
) {
763 list_move_tail(&device
->dev_alloc_list
, &private_devs
);
765 if (type
& BTRFS_BLOCK_GROUP_DUP
)
767 } else if (avail
> max_avail
)
772 if (index
< num_stripes
) {
773 list_splice(&private_devs
, dev_list
);
774 if (index
>= min_stripes
) {
776 if (type
& (BTRFS_BLOCK_GROUP_RAID10
)) {
777 num_stripes
/= sub_stripes
;
778 num_stripes
*= sub_stripes
;
783 if (!looped
&& max_avail
> 0) {
785 calc_size
= max_avail
;
790 key
.objectid
= BTRFS_FIRST_CHUNK_TREE_OBJECTID
;
791 key
.type
= BTRFS_CHUNK_ITEM_KEY
;
792 ret
= find_next_chunk(chunk_root
, BTRFS_FIRST_CHUNK_TREE_OBJECTID
,
797 chunk
= kmalloc(btrfs_chunk_item_size(num_stripes
), GFP_NOFS
);
801 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
807 stripes
= &chunk
->stripe
;
808 *num_bytes
= chunk_bytes_by_type(type
, calc_size
,
809 num_stripes
, sub_stripes
);
813 printk("new chunk type %Lu start %Lu size %Lu\n", type
, key
.offset
, *num_bytes
);
814 while(index
< num_stripes
) {
815 struct btrfs_stripe
*stripe
;
816 BUG_ON(list_empty(&private_devs
));
817 cur
= private_devs
.next
;
818 device
= list_entry(cur
, struct btrfs_device
, dev_alloc_list
);
820 /* loop over this device again if we're doing a dup group */
821 if (!(type
& BTRFS_BLOCK_GROUP_DUP
) ||
822 (index
== num_stripes
- 1))
823 list_move_tail(&device
->dev_alloc_list
, dev_list
);
825 ret
= btrfs_alloc_dev_extent(trans
, device
,
826 info
->chunk_root
->root_key
.objectid
,
827 BTRFS_FIRST_CHUNK_TREE_OBJECTID
, key
.offset
,
828 calc_size
, &dev_offset
);
830 printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key
.offset
, calc_size
, device
->devid
, type
);
831 device
->bytes_used
+= calc_size
;
832 ret
= btrfs_update_device(trans
, device
);
835 map
->stripes
[index
].dev
= device
;
836 map
->stripes
[index
].physical
= dev_offset
;
837 stripe
= stripes
+ index
;
838 btrfs_set_stack_stripe_devid(stripe
, device
->devid
);
839 btrfs_set_stack_stripe_offset(stripe
, dev_offset
);
840 memcpy(stripe
->dev_uuid
, device
->uuid
, BTRFS_UUID_SIZE
);
841 physical
= dev_offset
;
844 BUG_ON(!list_empty(&private_devs
));
846 /* key was set above */
847 btrfs_set_stack_chunk_length(chunk
, *num_bytes
);
848 btrfs_set_stack_chunk_owner(chunk
, extent_root
->root_key
.objectid
);
849 btrfs_set_stack_chunk_stripe_len(chunk
, stripe_len
);
850 btrfs_set_stack_chunk_type(chunk
, type
);
851 btrfs_set_stack_chunk_num_stripes(chunk
, num_stripes
);
852 btrfs_set_stack_chunk_io_align(chunk
, stripe_len
);
853 btrfs_set_stack_chunk_io_width(chunk
, stripe_len
);
854 btrfs_set_stack_chunk_sector_size(chunk
, extent_root
->sectorsize
);
855 btrfs_set_stack_chunk_sub_stripes(chunk
, sub_stripes
);
856 map
->sector_size
= extent_root
->sectorsize
;
857 map
->stripe_len
= stripe_len
;
858 map
->io_align
= stripe_len
;
859 map
->io_width
= stripe_len
;
861 map
->num_stripes
= num_stripes
;
862 map
->sub_stripes
= sub_stripes
;
864 ret
= btrfs_insert_item(trans
, chunk_root
, &key
, chunk
,
865 btrfs_chunk_item_size(num_stripes
));
867 *start
= key
.offset
;;
869 em
= alloc_extent_map(GFP_NOFS
);
872 em
->bdev
= (struct block_device
*)map
;
873 em
->start
= key
.offset
;
874 em
->len
= *num_bytes
;
879 em_tree
= &extent_root
->fs_info
->mapping_tree
.map_tree
;
880 spin_lock(&em_tree
->lock
);
881 ret
= add_extent_mapping(em_tree
, em
);
882 spin_unlock(&em_tree
->lock
);
888 void btrfs_mapping_init(struct btrfs_mapping_tree
*tree
)
890 extent_map_tree_init(&tree
->map_tree
, GFP_NOFS
);
893 void btrfs_mapping_tree_free(struct btrfs_mapping_tree
*tree
)
895 struct extent_map
*em
;
898 spin_lock(&tree
->map_tree
.lock
);
899 em
= lookup_extent_mapping(&tree
->map_tree
, 0, (u64
)-1);
901 remove_extent_mapping(&tree
->map_tree
, em
);
902 spin_unlock(&tree
->map_tree
.lock
);
908 /* once for the tree */
913 int btrfs_num_copies(struct btrfs_mapping_tree
*map_tree
, u64 logical
, u64 len
)
915 struct extent_map
*em
;
916 struct map_lookup
*map
;
917 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
920 spin_lock(&em_tree
->lock
);
921 em
= lookup_extent_mapping(em_tree
, logical
, len
);
922 spin_unlock(&em_tree
->lock
);
925 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
926 map
= (struct map_lookup
*)em
->bdev
;
927 if (map
->type
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
))
928 ret
= map
->num_stripes
;
929 else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
)
930 ret
= map
->sub_stripes
;
937 static int __btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
938 u64 logical
, u64
*length
,
939 struct btrfs_multi_bio
**multi_ret
,
940 int mirror_num
, struct page
*unplug_page
)
942 struct extent_map
*em
;
943 struct map_lookup
*map
;
944 struct extent_map_tree
*em_tree
= &map_tree
->map_tree
;
948 int stripes_allocated
= 8;
949 int stripes_required
= 1;
953 struct btrfs_multi_bio
*multi
= NULL
;
955 if (multi_ret
&& !(rw
& (1 << BIO_RW
))) {
956 stripes_allocated
= 1;
960 multi
= kzalloc(btrfs_multi_bio_size(stripes_allocated
),
966 spin_lock(&em_tree
->lock
);
967 em
= lookup_extent_mapping(em_tree
, logical
, *length
);
968 spin_unlock(&em_tree
->lock
);
970 if (!em
&& unplug_page
)
974 printk("unable to find logical %Lu\n", logical
);
978 BUG_ON(em
->start
> logical
|| em
->start
+ em
->len
< logical
);
979 map
= (struct map_lookup
*)em
->bdev
;
980 offset
= logical
- em
->start
;
982 if (mirror_num
> map
->num_stripes
)
985 /* if our multi bio struct is too small, back off and try again */
986 if (rw
& (1 << BIO_RW
)) {
987 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID1
|
988 BTRFS_BLOCK_GROUP_DUP
)) {
989 stripes_required
= map
->num_stripes
;
990 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
991 stripes_required
= map
->sub_stripes
;
994 if (multi_ret
&& rw
== WRITE
&&
995 stripes_allocated
< stripes_required
) {
996 stripes_allocated
= map
->num_stripes
;
1003 * stripe_nr counts the total number of stripes we have to stride
1004 * to get to this block
1006 do_div(stripe_nr
, map
->stripe_len
);
1008 stripe_offset
= stripe_nr
* map
->stripe_len
;
1009 BUG_ON(offset
< stripe_offset
);
1011 /* stripe_offset is the offset of this block in its stripe*/
1012 stripe_offset
= offset
- stripe_offset
;
1014 if (map
->type
& (BTRFS_BLOCK_GROUP_RAID0
| BTRFS_BLOCK_GROUP_RAID1
|
1015 BTRFS_BLOCK_GROUP_RAID10
|
1016 BTRFS_BLOCK_GROUP_DUP
)) {
1017 /* we limit the length of each bio to what fits in a stripe */
1018 *length
= min_t(u64
, em
->len
- offset
,
1019 map
->stripe_len
- stripe_offset
);
1021 *length
= em
->len
- offset
;
1024 if (!multi_ret
&& !unplug_page
)
1029 if (map
->type
& BTRFS_BLOCK_GROUP_RAID1
) {
1030 if (unplug_page
|| (rw
& (1 << BIO_RW
)))
1031 num_stripes
= map
->num_stripes
;
1032 else if (mirror_num
) {
1033 stripe_index
= mirror_num
- 1;
1035 u64 orig_stripe_nr
= stripe_nr
;
1036 stripe_index
= do_div(orig_stripe_nr
, num_stripes
);
1038 } else if (map
->type
& BTRFS_BLOCK_GROUP_DUP
) {
1039 if (rw
& (1 << BIO_RW
))
1040 num_stripes
= map
->num_stripes
;
1041 else if (mirror_num
)
1042 stripe_index
= mirror_num
- 1;
1043 } else if (map
->type
& BTRFS_BLOCK_GROUP_RAID10
) {
1044 int factor
= map
->num_stripes
/ map
->sub_stripes
;
1046 stripe_index
= do_div(stripe_nr
, factor
);
1047 stripe_index
*= map
->sub_stripes
;
1049 if (unplug_page
|| (rw
& (1 << BIO_RW
)))
1050 num_stripes
= map
->sub_stripes
;
1051 else if (mirror_num
)
1052 stripe_index
+= mirror_num
- 1;
1054 u64 orig_stripe_nr
= stripe_nr
;
1055 stripe_index
+= do_div(orig_stripe_nr
,
1060 * after this do_div call, stripe_nr is the number of stripes
1061 * on this device we have to walk to find the data, and
1062 * stripe_index is the number of our device in the stripe array
1064 stripe_index
= do_div(stripe_nr
, map
->num_stripes
);
1066 BUG_ON(stripe_index
>= map
->num_stripes
);
1068 for (i
= 0; i
< num_stripes
; i
++) {
1070 struct btrfs_device
*device
;
1071 struct backing_dev_info
*bdi
;
1073 device
= map
->stripes
[stripe_index
].dev
;
1074 bdi
= blk_get_backing_dev_info(device
->bdev
);
1075 if (bdi
->unplug_io_fn
) {
1076 bdi
->unplug_io_fn(bdi
, unplug_page
);
1079 multi
->stripes
[i
].physical
=
1080 map
->stripes
[stripe_index
].physical
+
1081 stripe_offset
+ stripe_nr
* map
->stripe_len
;
1082 multi
->stripes
[i
].dev
= map
->stripes
[stripe_index
].dev
;
1088 multi
->num_stripes
= num_stripes
;
1091 free_extent_map(em
);
1095 int btrfs_map_block(struct btrfs_mapping_tree
*map_tree
, int rw
,
1096 u64 logical
, u64
*length
,
1097 struct btrfs_multi_bio
**multi_ret
, int mirror_num
)
1099 return __btrfs_map_block(map_tree
, rw
, logical
, length
, multi_ret
,
1103 int btrfs_unplug_page(struct btrfs_mapping_tree
*map_tree
,
1104 u64 logical
, struct page
*page
)
1106 u64 length
= PAGE_CACHE_SIZE
;
1107 return __btrfs_map_block(map_tree
, READ
, logical
, &length
,
1112 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1113 static void end_bio_multi_stripe(struct bio
*bio
, int err
)
1115 static int end_bio_multi_stripe(struct bio
*bio
,
1116 unsigned int bytes_done
, int err
)
1119 struct btrfs_multi_bio
*multi
= bio
->bi_private
;
1121 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1128 if (atomic_dec_and_test(&multi
->stripes_pending
)) {
1129 bio
->bi_private
= multi
->private;
1130 bio
->bi_end_io
= multi
->end_io
;
1132 if (!err
&& multi
->error
)
1136 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1137 bio_endio(bio
, bio
->bi_size
, err
);
1139 bio_endio(bio
, err
);
1144 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1149 int btrfs_map_bio(struct btrfs_root
*root
, int rw
, struct bio
*bio
,
1152 struct btrfs_mapping_tree
*map_tree
;
1153 struct btrfs_device
*dev
;
1154 struct bio
*first_bio
= bio
;
1155 u64 logical
= bio
->bi_sector
<< 9;
1158 struct btrfs_multi_bio
*multi
= NULL
;
1163 length
= bio
->bi_size
;
1164 map_tree
= &root
->fs_info
->mapping_tree
;
1165 map_length
= length
;
1167 ret
= btrfs_map_block(map_tree
, rw
, logical
, &map_length
, &multi
,
1171 total_devs
= multi
->num_stripes
;
1172 if (map_length
< length
) {
1173 printk("mapping failed logical %Lu bio len %Lu "
1174 "len %Lu\n", logical
, length
, map_length
);
1177 multi
->end_io
= first_bio
->bi_end_io
;
1178 multi
->private = first_bio
->bi_private
;
1179 atomic_set(&multi
->stripes_pending
, multi
->num_stripes
);
1181 while(dev_nr
< total_devs
) {
1182 if (total_devs
> 1) {
1183 if (dev_nr
< total_devs
- 1) {
1184 bio
= bio_clone(first_bio
, GFP_NOFS
);
1189 bio
->bi_private
= multi
;
1190 bio
->bi_end_io
= end_bio_multi_stripe
;
1192 bio
->bi_sector
= multi
->stripes
[dev_nr
].physical
>> 9;
1193 dev
= multi
->stripes
[dev_nr
].dev
;
1195 bio
->bi_bdev
= dev
->bdev
;
1196 spin_lock(&dev
->io_lock
);
1198 spin_unlock(&dev
->io_lock
);
1199 submit_bio(rw
, bio
);
1202 if (total_devs
== 1)
1207 struct btrfs_device
*btrfs_find_device(struct btrfs_root
*root
, u64 devid
,
1210 struct list_head
*head
= &root
->fs_info
->fs_devices
->devices
;
1212 return __find_device(head
, devid
, uuid
);
1215 static int read_one_chunk(struct btrfs_root
*root
, struct btrfs_key
*key
,
1216 struct extent_buffer
*leaf
,
1217 struct btrfs_chunk
*chunk
)
1219 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
1220 struct map_lookup
*map
;
1221 struct extent_map
*em
;
1225 u8 uuid
[BTRFS_UUID_SIZE
];
1230 logical
= key
->offset
;
1231 length
= btrfs_chunk_length(leaf
, chunk
);
1232 spin_lock(&map_tree
->map_tree
.lock
);
1233 em
= lookup_extent_mapping(&map_tree
->map_tree
, logical
, 1);
1234 spin_unlock(&map_tree
->map_tree
.lock
);
1236 /* already mapped? */
1237 if (em
&& em
->start
<= logical
&& em
->start
+ em
->len
> logical
) {
1238 free_extent_map(em
);
1241 free_extent_map(em
);
1244 map
= kzalloc(sizeof(*map
), GFP_NOFS
);
1248 em
= alloc_extent_map(GFP_NOFS
);
1251 num_stripes
= btrfs_chunk_num_stripes(leaf
, chunk
);
1252 map
= kmalloc(map_lookup_size(num_stripes
), GFP_NOFS
);
1254 free_extent_map(em
);
1258 em
->bdev
= (struct block_device
*)map
;
1259 em
->start
= logical
;
1261 em
->block_start
= 0;
1263 map
->num_stripes
= num_stripes
;
1264 map
->io_width
= btrfs_chunk_io_width(leaf
, chunk
);
1265 map
->io_align
= btrfs_chunk_io_align(leaf
, chunk
);
1266 map
->sector_size
= btrfs_chunk_sector_size(leaf
, chunk
);
1267 map
->stripe_len
= btrfs_chunk_stripe_len(leaf
, chunk
);
1268 map
->type
= btrfs_chunk_type(leaf
, chunk
);
1269 map
->sub_stripes
= btrfs_chunk_sub_stripes(leaf
, chunk
);
1270 for (i
= 0; i
< num_stripes
; i
++) {
1271 map
->stripes
[i
].physical
=
1272 btrfs_stripe_offset_nr(leaf
, chunk
, i
);
1273 devid
= btrfs_stripe_devid_nr(leaf
, chunk
, i
);
1274 read_extent_buffer(leaf
, uuid
, (unsigned long)
1275 btrfs_stripe_dev_uuid_nr(chunk
, i
),
1277 map
->stripes
[i
].dev
= btrfs_find_device(root
, devid
, uuid
);
1278 if (!map
->stripes
[i
].dev
) {
1280 free_extent_map(em
);
1285 spin_lock(&map_tree
->map_tree
.lock
);
1286 ret
= add_extent_mapping(&map_tree
->map_tree
, em
);
1287 spin_unlock(&map_tree
->map_tree
.lock
);
1289 free_extent_map(em
);
1294 static int fill_device_from_item(struct extent_buffer
*leaf
,
1295 struct btrfs_dev_item
*dev_item
,
1296 struct btrfs_device
*device
)
1300 device
->devid
= btrfs_device_id(leaf
, dev_item
);
1301 device
->total_bytes
= btrfs_device_total_bytes(leaf
, dev_item
);
1302 device
->bytes_used
= btrfs_device_bytes_used(leaf
, dev_item
);
1303 device
->type
= btrfs_device_type(leaf
, dev_item
);
1304 device
->io_align
= btrfs_device_io_align(leaf
, dev_item
);
1305 device
->io_width
= btrfs_device_io_width(leaf
, dev_item
);
1306 device
->sector_size
= btrfs_device_sector_size(leaf
, dev_item
);
1308 ptr
= (unsigned long)btrfs_device_uuid(dev_item
);
1309 read_extent_buffer(leaf
, device
->uuid
, ptr
, BTRFS_UUID_SIZE
);
1314 static int read_one_dev(struct btrfs_root
*root
,
1315 struct extent_buffer
*leaf
,
1316 struct btrfs_dev_item
*dev_item
)
1318 struct btrfs_device
*device
;
1321 u8 dev_uuid
[BTRFS_UUID_SIZE
];
1323 devid
= btrfs_device_id(leaf
, dev_item
);
1324 read_extent_buffer(leaf
, dev_uuid
,
1325 (unsigned long)btrfs_device_uuid(dev_item
),
1327 device
= btrfs_find_device(root
, devid
, dev_uuid
);
1329 printk("warning devid %Lu not found already\n", devid
);
1330 device
= kzalloc(sizeof(*device
), GFP_NOFS
);
1333 list_add(&device
->dev_list
,
1334 &root
->fs_info
->fs_devices
->devices
);
1335 list_add(&device
->dev_alloc_list
,
1336 &root
->fs_info
->fs_devices
->alloc_list
);
1337 device
->barriers
= 1;
1338 spin_lock_init(&device
->io_lock
);
1341 fill_device_from_item(leaf
, dev_item
, device
);
1342 device
->dev_root
= root
->fs_info
->dev_root
;
1345 ret
= btrfs_open_device(device
);
1353 int btrfs_read_super_device(struct btrfs_root
*root
, struct extent_buffer
*buf
)
1355 struct btrfs_dev_item
*dev_item
;
1357 dev_item
= (struct btrfs_dev_item
*)offsetof(struct btrfs_super_block
,
1359 return read_one_dev(root
, buf
, dev_item
);
1362 int btrfs_read_sys_array(struct btrfs_root
*root
)
1364 struct btrfs_super_block
*super_copy
= &root
->fs_info
->super_copy
;
1365 struct extent_buffer
*sb
= root
->fs_info
->sb_buffer
;
1366 struct btrfs_disk_key
*disk_key
;
1367 struct btrfs_chunk
*chunk
;
1368 struct btrfs_key key
;
1373 unsigned long sb_ptr
;
1377 array_size
= btrfs_super_sys_array_size(super_copy
);
1380 * we do this loop twice, once for the device items and
1381 * once for all of the chunks. This way there are device
1382 * structs filled in for every chunk
1384 ptr
= super_copy
->sys_chunk_array
;
1385 sb_ptr
= offsetof(struct btrfs_super_block
, sys_chunk_array
);
1388 while (cur
< array_size
) {
1389 disk_key
= (struct btrfs_disk_key
*)ptr
;
1390 btrfs_disk_key_to_cpu(&key
, disk_key
);
1392 len
= sizeof(*disk_key
);
1397 if (key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
1398 chunk
= (struct btrfs_chunk
*)sb_ptr
;
1399 ret
= read_one_chunk(root
, &key
, sb
, chunk
);
1401 num_stripes
= btrfs_chunk_num_stripes(sb
, chunk
);
1402 len
= btrfs_chunk_item_size(num_stripes
);
1413 int btrfs_read_chunk_tree(struct btrfs_root
*root
)
1415 struct btrfs_path
*path
;
1416 struct extent_buffer
*leaf
;
1417 struct btrfs_key key
;
1418 struct btrfs_key found_key
;
1422 root
= root
->fs_info
->chunk_root
;
1424 path
= btrfs_alloc_path();
1428 /* first we search for all of the device items, and then we
1429 * read in all of the chunk items. This way we can create chunk
1430 * mappings that reference all of the devices that are afound
1432 key
.objectid
= BTRFS_DEV_ITEMS_OBJECTID
;
1436 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
1438 leaf
= path
->nodes
[0];
1439 slot
= path
->slots
[0];
1440 if (slot
>= btrfs_header_nritems(leaf
)) {
1441 ret
= btrfs_next_leaf(root
, path
);
1448 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
1449 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
1450 if (found_key
.objectid
!= BTRFS_DEV_ITEMS_OBJECTID
)
1452 if (found_key
.type
== BTRFS_DEV_ITEM_KEY
) {
1453 struct btrfs_dev_item
*dev_item
;
1454 dev_item
= btrfs_item_ptr(leaf
, slot
,
1455 struct btrfs_dev_item
);
1456 ret
= read_one_dev(root
, leaf
, dev_item
);
1459 } else if (found_key
.type
== BTRFS_CHUNK_ITEM_KEY
) {
1460 struct btrfs_chunk
*chunk
;
1461 chunk
= btrfs_item_ptr(leaf
, slot
, struct btrfs_chunk
);
1462 ret
= read_one_chunk(root
, &found_key
, leaf
, chunk
);
1466 if (key
.objectid
== BTRFS_DEV_ITEMS_OBJECTID
) {
1468 btrfs_release_path(root
, path
);
1472 btrfs_free_path(path
);