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.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args
{
60 struct btrfs_root
*root
;
63 static const struct inode_operations btrfs_dir_inode_operations
;
64 static const struct inode_operations btrfs_symlink_inode_operations
;
65 static const struct inode_operations btrfs_dir_ro_inode_operations
;
66 static const struct inode_operations btrfs_special_inode_operations
;
67 static const struct inode_operations btrfs_file_inode_operations
;
68 static const struct address_space_operations btrfs_aops
;
69 static const struct address_space_operations btrfs_symlink_aops
;
70 static const struct file_operations btrfs_dir_file_operations
;
71 static struct extent_io_ops btrfs_extent_io_ops
;
73 static struct kmem_cache
*btrfs_inode_cachep
;
74 struct kmem_cache
*btrfs_trans_handle_cachep
;
75 struct kmem_cache
*btrfs_transaction_cachep
;
76 struct kmem_cache
*btrfs_path_cachep
;
77 struct kmem_cache
*btrfs_free_space_cachep
;
80 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
81 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
82 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
83 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
84 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
85 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
86 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
87 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
90 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
);
91 static int btrfs_truncate(struct inode
*inode
);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
93 static noinline
int cow_file_range(struct inode
*inode
,
94 struct page
*locked_page
,
95 u64 start
, u64 end
, int *page_started
,
96 unsigned long *nr_written
, int unlock
);
98 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
99 struct inode
*inode
, struct inode
*dir
,
100 const struct qstr
*qstr
)
104 err
= btrfs_init_acl(trans
, inode
, dir
);
106 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
111 * this does all the hard work for inserting an inline extent into
112 * the btree. The caller should have done a btrfs_drop_extents so that
113 * no overlapping inline items exist in the btree
115 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
116 struct btrfs_root
*root
, struct inode
*inode
,
117 u64 start
, size_t size
, size_t compressed_size
,
119 struct page
**compressed_pages
)
121 struct btrfs_key key
;
122 struct btrfs_path
*path
;
123 struct extent_buffer
*leaf
;
124 struct page
*page
= NULL
;
127 struct btrfs_file_extent_item
*ei
;
130 size_t cur_size
= size
;
132 unsigned long offset
;
134 if (compressed_size
&& compressed_pages
)
135 cur_size
= compressed_size
;
137 path
= btrfs_alloc_path();
141 path
->leave_spinning
= 1;
143 key
.objectid
= btrfs_ino(inode
);
145 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
146 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
148 inode_add_bytes(inode
, size
);
149 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
155 leaf
= path
->nodes
[0];
156 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
157 struct btrfs_file_extent_item
);
158 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
159 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
160 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
161 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
162 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
163 ptr
= btrfs_file_extent_inline_start(ei
);
165 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
168 while (compressed_size
> 0) {
169 cpage
= compressed_pages
[i
];
170 cur_size
= min_t(unsigned long, compressed_size
,
173 kaddr
= kmap_atomic(cpage
);
174 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
175 kunmap_atomic(kaddr
);
179 compressed_size
-= cur_size
;
181 btrfs_set_file_extent_compression(leaf
, ei
,
184 page
= find_get_page(inode
->i_mapping
,
185 start
>> PAGE_CACHE_SHIFT
);
186 btrfs_set_file_extent_compression(leaf
, ei
, 0);
187 kaddr
= kmap_atomic(page
);
188 offset
= start
& (PAGE_CACHE_SIZE
- 1);
189 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
190 kunmap_atomic(kaddr
);
191 page_cache_release(page
);
193 btrfs_mark_buffer_dirty(leaf
);
194 btrfs_free_path(path
);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
206 ret
= btrfs_update_inode(trans
, root
, inode
);
210 btrfs_free_path(path
);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
221 struct btrfs_root
*root
,
222 struct inode
*inode
, u64 start
, u64 end
,
223 size_t compressed_size
, int compress_type
,
224 struct page
**compressed_pages
)
226 u64 isize
= i_size_read(inode
);
227 u64 actual_end
= min(end
+ 1, isize
);
228 u64 inline_len
= actual_end
- start
;
229 u64 aligned_end
= (end
+ root
->sectorsize
- 1) &
230 ~((u64
)root
->sectorsize
- 1);
231 u64 data_len
= inline_len
;
235 data_len
= compressed_size
;
238 actual_end
>= PAGE_CACHE_SIZE
||
239 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
241 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
243 data_len
> root
->fs_info
->max_inline
) {
247 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, aligned_end
, 1);
251 if (isize
> actual_end
)
252 inline_len
= min_t(u64
, isize
, actual_end
);
253 ret
= insert_inline_extent(trans
, root
, inode
, start
,
254 inline_len
, compressed_size
,
255 compress_type
, compressed_pages
);
256 if (ret
&& ret
!= -ENOSPC
) {
257 btrfs_abort_transaction(trans
, root
, ret
);
259 } else if (ret
== -ENOSPC
) {
263 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
264 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
268 struct async_extent
{
273 unsigned long nr_pages
;
275 struct list_head list
;
280 struct btrfs_root
*root
;
281 struct page
*locked_page
;
284 struct list_head extents
;
285 struct btrfs_work work
;
288 static noinline
int add_async_extent(struct async_cow
*cow
,
289 u64 start
, u64 ram_size
,
292 unsigned long nr_pages
,
295 struct async_extent
*async_extent
;
297 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
298 BUG_ON(!async_extent
); /* -ENOMEM */
299 async_extent
->start
= start
;
300 async_extent
->ram_size
= ram_size
;
301 async_extent
->compressed_size
= compressed_size
;
302 async_extent
->pages
= pages
;
303 async_extent
->nr_pages
= nr_pages
;
304 async_extent
->compress_type
= compress_type
;
305 list_add_tail(&async_extent
->list
, &cow
->extents
);
310 * we create compressed extents in two phases. The first
311 * phase compresses a range of pages that have already been
312 * locked (both pages and state bits are locked).
314 * This is done inside an ordered work queue, and the compression
315 * is spread across many cpus. The actual IO submission is step
316 * two, and the ordered work queue takes care of making sure that
317 * happens in the same order things were put onto the queue by
318 * writepages and friends.
320 * If this code finds it can't get good compression, it puts an
321 * entry onto the work queue to write the uncompressed bytes. This
322 * makes sure that both compressed inodes and uncompressed inodes
323 * are written in the same order that the flusher thread sent them
326 static noinline
int compress_file_range(struct inode
*inode
,
327 struct page
*locked_page
,
329 struct async_cow
*async_cow
,
332 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
333 struct btrfs_trans_handle
*trans
;
335 u64 blocksize
= root
->sectorsize
;
337 u64 isize
= i_size_read(inode
);
339 struct page
**pages
= NULL
;
340 unsigned long nr_pages
;
341 unsigned long nr_pages_ret
= 0;
342 unsigned long total_compressed
= 0;
343 unsigned long total_in
= 0;
344 unsigned long max_compressed
= 128 * 1024;
345 unsigned long max_uncompressed
= 128 * 1024;
348 int compress_type
= root
->fs_info
->compress_type
;
350 /* if this is a small write inside eof, kick off a defrag */
351 if ((end
- start
+ 1) < 16 * 1024 &&
352 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
353 btrfs_add_inode_defrag(NULL
, inode
);
355 actual_end
= min_t(u64
, isize
, end
+ 1);
358 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
359 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
362 * we don't want to send crud past the end of i_size through
363 * compression, that's just a waste of CPU time. So, if the
364 * end of the file is before the start of our current
365 * requested range of bytes, we bail out to the uncompressed
366 * cleanup code that can deal with all of this.
368 * It isn't really the fastest way to fix things, but this is a
369 * very uncommon corner.
371 if (actual_end
<= start
)
372 goto cleanup_and_bail_uncompressed
;
374 total_compressed
= actual_end
- start
;
376 /* we want to make sure that amount of ram required to uncompress
377 * an extent is reasonable, so we limit the total size in ram
378 * of a compressed extent to 128k. This is a crucial number
379 * because it also controls how easily we can spread reads across
380 * cpus for decompression.
382 * We also want to make sure the amount of IO required to do
383 * a random read is reasonably small, so we limit the size of
384 * a compressed extent to 128k.
386 total_compressed
= min(total_compressed
, max_uncompressed
);
387 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
388 num_bytes
= max(blocksize
, num_bytes
);
393 * we do compression for mount -o compress and when the
394 * inode has not been flagged as nocompress. This flag can
395 * change at any time if we discover bad compression ratios.
397 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
398 (btrfs_test_opt(root
, COMPRESS
) ||
399 (BTRFS_I(inode
)->force_compress
) ||
400 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
402 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
404 /* just bail out to the uncompressed code */
408 if (BTRFS_I(inode
)->force_compress
)
409 compress_type
= BTRFS_I(inode
)->force_compress
;
411 ret
= btrfs_compress_pages(compress_type
,
412 inode
->i_mapping
, start
,
413 total_compressed
, pages
,
414 nr_pages
, &nr_pages_ret
,
420 unsigned long offset
= total_compressed
&
421 (PAGE_CACHE_SIZE
- 1);
422 struct page
*page
= pages
[nr_pages_ret
- 1];
425 /* zero the tail end of the last page, we might be
426 * sending it down to disk
429 kaddr
= kmap_atomic(page
);
430 memset(kaddr
+ offset
, 0,
431 PAGE_CACHE_SIZE
- offset
);
432 kunmap_atomic(kaddr
);
439 trans
= btrfs_join_transaction(root
);
441 ret
= PTR_ERR(trans
);
443 goto cleanup_and_out
;
445 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
447 /* lets try to make an inline extent */
448 if (ret
|| total_in
< (actual_end
- start
)) {
449 /* we didn't compress the entire range, try
450 * to make an uncompressed inline extent.
452 ret
= cow_file_range_inline(trans
, root
, inode
,
453 start
, end
, 0, 0, NULL
);
455 /* try making a compressed inline extent */
456 ret
= cow_file_range_inline(trans
, root
, inode
,
459 compress_type
, pages
);
463 * inline extent creation worked or returned error,
464 * we don't need to create any more async work items.
465 * Unlock and free up our temp pages.
467 extent_clear_unlock_delalloc(inode
,
468 &BTRFS_I(inode
)->io_tree
,
470 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
471 EXTENT_CLEAR_DELALLOC
|
472 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
474 btrfs_end_transaction(trans
, root
);
477 btrfs_end_transaction(trans
, root
);
482 * we aren't doing an inline extent round the compressed size
483 * up to a block size boundary so the allocator does sane
486 total_compressed
= (total_compressed
+ blocksize
- 1) &
490 * one last check to make sure the compression is really a
491 * win, compare the page count read with the blocks on disk
493 total_in
= (total_in
+ PAGE_CACHE_SIZE
- 1) &
494 ~(PAGE_CACHE_SIZE
- 1);
495 if (total_compressed
>= total_in
) {
498 num_bytes
= total_in
;
501 if (!will_compress
&& pages
) {
503 * the compression code ran but failed to make things smaller,
504 * free any pages it allocated and our page pointer array
506 for (i
= 0; i
< nr_pages_ret
; i
++) {
507 WARN_ON(pages
[i
]->mapping
);
508 page_cache_release(pages
[i
]);
512 total_compressed
= 0;
515 /* flag the file so we don't compress in the future */
516 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
517 !(BTRFS_I(inode
)->force_compress
)) {
518 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
524 /* the async work queues will take care of doing actual
525 * allocation on disk for these compressed pages,
526 * and will submit them to the elevator.
528 add_async_extent(async_cow
, start
, num_bytes
,
529 total_compressed
, pages
, nr_pages_ret
,
532 if (start
+ num_bytes
< end
) {
539 cleanup_and_bail_uncompressed
:
541 * No compression, but we still need to write the pages in
542 * the file we've been given so far. redirty the locked
543 * page if it corresponds to our extent and set things up
544 * for the async work queue to run cow_file_range to do
545 * the normal delalloc dance
547 if (page_offset(locked_page
) >= start
&&
548 page_offset(locked_page
) <= end
) {
549 __set_page_dirty_nobuffers(locked_page
);
550 /* unlocked later on in the async handlers */
552 add_async_extent(async_cow
, start
, end
- start
+ 1,
553 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
561 for (i
= 0; i
< nr_pages_ret
; i
++) {
562 WARN_ON(pages
[i
]->mapping
);
563 page_cache_release(pages
[i
]);
570 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
572 EXTENT_CLEAR_UNLOCK_PAGE
|
574 EXTENT_CLEAR_DELALLOC
|
575 EXTENT_SET_WRITEBACK
|
576 EXTENT_END_WRITEBACK
);
577 if (!trans
|| IS_ERR(trans
))
578 btrfs_error(root
->fs_info
, ret
, "Failed to join transaction");
580 btrfs_abort_transaction(trans
, root
, ret
);
585 * phase two of compressed writeback. This is the ordered portion
586 * of the code, which only gets called in the order the work was
587 * queued. We walk all the async extents created by compress_file_range
588 * and send them down to the disk.
590 static noinline
int submit_compressed_extents(struct inode
*inode
,
591 struct async_cow
*async_cow
)
593 struct async_extent
*async_extent
;
595 struct btrfs_trans_handle
*trans
;
596 struct btrfs_key ins
;
597 struct extent_map
*em
;
598 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
599 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
600 struct extent_io_tree
*io_tree
;
603 if (list_empty(&async_cow
->extents
))
607 while (!list_empty(&async_cow
->extents
)) {
608 async_extent
= list_entry(async_cow
->extents
.next
,
609 struct async_extent
, list
);
610 list_del(&async_extent
->list
);
612 io_tree
= &BTRFS_I(inode
)->io_tree
;
615 /* did the compression code fall back to uncompressed IO? */
616 if (!async_extent
->pages
) {
617 int page_started
= 0;
618 unsigned long nr_written
= 0;
620 lock_extent(io_tree
, async_extent
->start
,
621 async_extent
->start
+
622 async_extent
->ram_size
- 1);
624 /* allocate blocks */
625 ret
= cow_file_range(inode
, async_cow
->locked_page
,
627 async_extent
->start
+
628 async_extent
->ram_size
- 1,
629 &page_started
, &nr_written
, 0);
634 * if page_started, cow_file_range inserted an
635 * inline extent and took care of all the unlocking
636 * and IO for us. Otherwise, we need to submit
637 * all those pages down to the drive.
639 if (!page_started
&& !ret
)
640 extent_write_locked_range(io_tree
,
641 inode
, async_extent
->start
,
642 async_extent
->start
+
643 async_extent
->ram_size
- 1,
651 lock_extent(io_tree
, async_extent
->start
,
652 async_extent
->start
+ async_extent
->ram_size
- 1);
654 trans
= btrfs_join_transaction(root
);
656 ret
= PTR_ERR(trans
);
658 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
659 ret
= btrfs_reserve_extent(trans
, root
,
660 async_extent
->compressed_size
,
661 async_extent
->compressed_size
,
662 0, alloc_hint
, &ins
, 1);
663 if (ret
&& ret
!= -ENOSPC
)
664 btrfs_abort_transaction(trans
, root
, ret
);
665 btrfs_end_transaction(trans
, root
);
670 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
671 WARN_ON(async_extent
->pages
[i
]->mapping
);
672 page_cache_release(async_extent
->pages
[i
]);
674 kfree(async_extent
->pages
);
675 async_extent
->nr_pages
= 0;
676 async_extent
->pages
= NULL
;
677 unlock_extent(io_tree
, async_extent
->start
,
678 async_extent
->start
+
679 async_extent
->ram_size
- 1);
682 goto out_free
; /* JDM: Requeue? */
686 * here we're doing allocation and writeback of the
689 btrfs_drop_extent_cache(inode
, async_extent
->start
,
690 async_extent
->start
+
691 async_extent
->ram_size
- 1, 0);
693 em
= alloc_extent_map();
694 BUG_ON(!em
); /* -ENOMEM */
695 em
->start
= async_extent
->start
;
696 em
->len
= async_extent
->ram_size
;
697 em
->orig_start
= em
->start
;
699 em
->block_start
= ins
.objectid
;
700 em
->block_len
= ins
.offset
;
701 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
702 em
->compress_type
= async_extent
->compress_type
;
703 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
704 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
707 write_lock(&em_tree
->lock
);
708 ret
= add_extent_mapping(em_tree
, em
);
709 write_unlock(&em_tree
->lock
);
710 if (ret
!= -EEXIST
) {
714 btrfs_drop_extent_cache(inode
, async_extent
->start
,
715 async_extent
->start
+
716 async_extent
->ram_size
- 1, 0);
719 ret
= btrfs_add_ordered_extent_compress(inode
,
722 async_extent
->ram_size
,
724 BTRFS_ORDERED_COMPRESSED
,
725 async_extent
->compress_type
);
726 BUG_ON(ret
); /* -ENOMEM */
729 * clear dirty, set writeback and unlock the pages.
731 extent_clear_unlock_delalloc(inode
,
732 &BTRFS_I(inode
)->io_tree
,
734 async_extent
->start
+
735 async_extent
->ram_size
- 1,
736 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
737 EXTENT_CLEAR_UNLOCK
|
738 EXTENT_CLEAR_DELALLOC
|
739 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
741 ret
= btrfs_submit_compressed_write(inode
,
743 async_extent
->ram_size
,
745 ins
.offset
, async_extent
->pages
,
746 async_extent
->nr_pages
);
748 BUG_ON(ret
); /* -ENOMEM */
749 alloc_hint
= ins
.objectid
+ ins
.offset
;
761 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
764 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
765 struct extent_map
*em
;
768 read_lock(&em_tree
->lock
);
769 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
772 * if block start isn't an actual block number then find the
773 * first block in this inode and use that as a hint. If that
774 * block is also bogus then just don't worry about it.
776 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
778 em
= search_extent_mapping(em_tree
, 0, 0);
779 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
780 alloc_hint
= em
->block_start
;
784 alloc_hint
= em
->block_start
;
788 read_unlock(&em_tree
->lock
);
794 * when extent_io.c finds a delayed allocation range in the file,
795 * the call backs end up in this code. The basic idea is to
796 * allocate extents on disk for the range, and create ordered data structs
797 * in ram to track those extents.
799 * locked_page is the page that writepage had locked already. We use
800 * it to make sure we don't do extra locks or unlocks.
802 * *page_started is set to one if we unlock locked_page and do everything
803 * required to start IO on it. It may be clean and already done with
806 static noinline
int cow_file_range(struct inode
*inode
,
807 struct page
*locked_page
,
808 u64 start
, u64 end
, int *page_started
,
809 unsigned long *nr_written
,
812 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
813 struct btrfs_trans_handle
*trans
;
816 unsigned long ram_size
;
819 u64 blocksize
= root
->sectorsize
;
820 struct btrfs_key ins
;
821 struct extent_map
*em
;
822 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
825 BUG_ON(btrfs_is_free_space_inode(inode
));
826 trans
= btrfs_join_transaction(root
);
828 extent_clear_unlock_delalloc(inode
,
829 &BTRFS_I(inode
)->io_tree
,
830 start
, end
, locked_page
,
831 EXTENT_CLEAR_UNLOCK_PAGE
|
832 EXTENT_CLEAR_UNLOCK
|
833 EXTENT_CLEAR_DELALLOC
|
835 EXTENT_SET_WRITEBACK
|
836 EXTENT_END_WRITEBACK
);
837 return PTR_ERR(trans
);
839 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
841 num_bytes
= (end
- start
+ blocksize
) & ~(blocksize
- 1);
842 num_bytes
= max(blocksize
, num_bytes
);
843 disk_num_bytes
= num_bytes
;
846 /* if this is a small write inside eof, kick off defrag */
847 if (num_bytes
< 64 * 1024 &&
848 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
849 btrfs_add_inode_defrag(trans
, inode
);
852 /* lets try to make an inline extent */
853 ret
= cow_file_range_inline(trans
, root
, inode
,
854 start
, end
, 0, 0, NULL
);
856 extent_clear_unlock_delalloc(inode
,
857 &BTRFS_I(inode
)->io_tree
,
859 EXTENT_CLEAR_UNLOCK_PAGE
|
860 EXTENT_CLEAR_UNLOCK
|
861 EXTENT_CLEAR_DELALLOC
|
863 EXTENT_SET_WRITEBACK
|
864 EXTENT_END_WRITEBACK
);
866 *nr_written
= *nr_written
+
867 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
870 } else if (ret
< 0) {
871 btrfs_abort_transaction(trans
, root
, ret
);
876 BUG_ON(disk_num_bytes
>
877 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
879 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
880 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
882 while (disk_num_bytes
> 0) {
885 cur_alloc_size
= disk_num_bytes
;
886 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
887 root
->sectorsize
, 0, alloc_hint
,
890 btrfs_abort_transaction(trans
, root
, ret
);
894 em
= alloc_extent_map();
895 BUG_ON(!em
); /* -ENOMEM */
897 em
->orig_start
= em
->start
;
898 ram_size
= ins
.offset
;
899 em
->len
= ins
.offset
;
901 em
->block_start
= ins
.objectid
;
902 em
->block_len
= ins
.offset
;
903 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
904 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
907 write_lock(&em_tree
->lock
);
908 ret
= add_extent_mapping(em_tree
, em
);
909 write_unlock(&em_tree
->lock
);
910 if (ret
!= -EEXIST
) {
914 btrfs_drop_extent_cache(inode
, start
,
915 start
+ ram_size
- 1, 0);
918 cur_alloc_size
= ins
.offset
;
919 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
920 ram_size
, cur_alloc_size
, 0);
921 BUG_ON(ret
); /* -ENOMEM */
923 if (root
->root_key
.objectid
==
924 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
925 ret
= btrfs_reloc_clone_csums(inode
, start
,
928 btrfs_abort_transaction(trans
, root
, ret
);
933 if (disk_num_bytes
< cur_alloc_size
)
936 /* we're not doing compressed IO, don't unlock the first
937 * page (which the caller expects to stay locked), don't
938 * clear any dirty bits and don't set any writeback bits
940 * Do set the Private2 bit so we know this page was properly
941 * setup for writepage
943 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
944 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
947 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
948 start
, start
+ ram_size
- 1,
950 disk_num_bytes
-= cur_alloc_size
;
951 num_bytes
-= cur_alloc_size
;
952 alloc_hint
= ins
.objectid
+ ins
.offset
;
953 start
+= cur_alloc_size
;
957 btrfs_end_transaction(trans
, root
);
961 extent_clear_unlock_delalloc(inode
,
962 &BTRFS_I(inode
)->io_tree
,
963 start
, end
, locked_page
,
964 EXTENT_CLEAR_UNLOCK_PAGE
|
965 EXTENT_CLEAR_UNLOCK
|
966 EXTENT_CLEAR_DELALLOC
|
968 EXTENT_SET_WRITEBACK
|
969 EXTENT_END_WRITEBACK
);
975 * work queue call back to started compression on a file and pages
977 static noinline
void async_cow_start(struct btrfs_work
*work
)
979 struct async_cow
*async_cow
;
981 async_cow
= container_of(work
, struct async_cow
, work
);
983 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
984 async_cow
->start
, async_cow
->end
, async_cow
,
986 if (num_added
== 0) {
987 btrfs_add_delayed_iput(async_cow
->inode
);
988 async_cow
->inode
= NULL
;
993 * work queue call back to submit previously compressed pages
995 static noinline
void async_cow_submit(struct btrfs_work
*work
)
997 struct async_cow
*async_cow
;
998 struct btrfs_root
*root
;
999 unsigned long nr_pages
;
1001 async_cow
= container_of(work
, struct async_cow
, work
);
1003 root
= async_cow
->root
;
1004 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1007 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1009 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1010 wake_up(&root
->fs_info
->async_submit_wait
);
1012 if (async_cow
->inode
)
1013 submit_compressed_extents(async_cow
->inode
, async_cow
);
1016 static noinline
void async_cow_free(struct btrfs_work
*work
)
1018 struct async_cow
*async_cow
;
1019 async_cow
= container_of(work
, struct async_cow
, work
);
1020 if (async_cow
->inode
)
1021 btrfs_add_delayed_iput(async_cow
->inode
);
1025 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1026 u64 start
, u64 end
, int *page_started
,
1027 unsigned long *nr_written
)
1029 struct async_cow
*async_cow
;
1030 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1031 unsigned long nr_pages
;
1033 int limit
= 10 * 1024 * 1024;
1035 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1036 1, 0, NULL
, GFP_NOFS
);
1037 while (start
< end
) {
1038 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1039 BUG_ON(!async_cow
); /* -ENOMEM */
1040 async_cow
->inode
= igrab(inode
);
1041 async_cow
->root
= root
;
1042 async_cow
->locked_page
= locked_page
;
1043 async_cow
->start
= start
;
1045 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1048 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1050 async_cow
->end
= cur_end
;
1051 INIT_LIST_HEAD(&async_cow
->extents
);
1053 async_cow
->work
.func
= async_cow_start
;
1054 async_cow
->work
.ordered_func
= async_cow_submit
;
1055 async_cow
->work
.ordered_free
= async_cow_free
;
1056 async_cow
->work
.flags
= 0;
1058 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1060 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1062 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1065 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1066 wait_event(root
->fs_info
->async_submit_wait
,
1067 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1071 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1072 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1073 wait_event(root
->fs_info
->async_submit_wait
,
1074 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1078 *nr_written
+= nr_pages
;
1079 start
= cur_end
+ 1;
1085 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1086 u64 bytenr
, u64 num_bytes
)
1089 struct btrfs_ordered_sum
*sums
;
1092 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1093 bytenr
+ num_bytes
- 1, &list
, 0);
1094 if (ret
== 0 && list_empty(&list
))
1097 while (!list_empty(&list
)) {
1098 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1099 list_del(&sums
->list
);
1106 * when nowcow writeback call back. This checks for snapshots or COW copies
1107 * of the extents that exist in the file, and COWs the file as required.
1109 * If no cow copies or snapshots exist, we write directly to the existing
1112 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1113 struct page
*locked_page
,
1114 u64 start
, u64 end
, int *page_started
, int force
,
1115 unsigned long *nr_written
)
1117 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1118 struct btrfs_trans_handle
*trans
;
1119 struct extent_buffer
*leaf
;
1120 struct btrfs_path
*path
;
1121 struct btrfs_file_extent_item
*fi
;
1122 struct btrfs_key found_key
;
1135 u64 ino
= btrfs_ino(inode
);
1137 path
= btrfs_alloc_path();
1139 extent_clear_unlock_delalloc(inode
,
1140 &BTRFS_I(inode
)->io_tree
,
1141 start
, end
, locked_page
,
1142 EXTENT_CLEAR_UNLOCK_PAGE
|
1143 EXTENT_CLEAR_UNLOCK
|
1144 EXTENT_CLEAR_DELALLOC
|
1145 EXTENT_CLEAR_DIRTY
|
1146 EXTENT_SET_WRITEBACK
|
1147 EXTENT_END_WRITEBACK
);
1151 nolock
= btrfs_is_free_space_inode(inode
);
1154 trans
= btrfs_join_transaction_nolock(root
);
1156 trans
= btrfs_join_transaction(root
);
1158 if (IS_ERR(trans
)) {
1159 extent_clear_unlock_delalloc(inode
,
1160 &BTRFS_I(inode
)->io_tree
,
1161 start
, end
, locked_page
,
1162 EXTENT_CLEAR_UNLOCK_PAGE
|
1163 EXTENT_CLEAR_UNLOCK
|
1164 EXTENT_CLEAR_DELALLOC
|
1165 EXTENT_CLEAR_DIRTY
|
1166 EXTENT_SET_WRITEBACK
|
1167 EXTENT_END_WRITEBACK
);
1168 btrfs_free_path(path
);
1169 return PTR_ERR(trans
);
1172 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1174 cow_start
= (u64
)-1;
1177 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1180 btrfs_abort_transaction(trans
, root
, ret
);
1183 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1184 leaf
= path
->nodes
[0];
1185 btrfs_item_key_to_cpu(leaf
, &found_key
,
1186 path
->slots
[0] - 1);
1187 if (found_key
.objectid
== ino
&&
1188 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1193 leaf
= path
->nodes
[0];
1194 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1195 ret
= btrfs_next_leaf(root
, path
);
1197 btrfs_abort_transaction(trans
, root
, ret
);
1202 leaf
= path
->nodes
[0];
1208 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1210 if (found_key
.objectid
> ino
||
1211 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1212 found_key
.offset
> end
)
1215 if (found_key
.offset
> cur_offset
) {
1216 extent_end
= found_key
.offset
;
1221 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1222 struct btrfs_file_extent_item
);
1223 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1225 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1226 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1227 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1228 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1229 extent_end
= found_key
.offset
+
1230 btrfs_file_extent_num_bytes(leaf
, fi
);
1231 if (extent_end
<= start
) {
1235 if (disk_bytenr
== 0)
1237 if (btrfs_file_extent_compression(leaf
, fi
) ||
1238 btrfs_file_extent_encryption(leaf
, fi
) ||
1239 btrfs_file_extent_other_encoding(leaf
, fi
))
1241 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1243 if (btrfs_extent_readonly(root
, disk_bytenr
))
1245 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1247 extent_offset
, disk_bytenr
))
1249 disk_bytenr
+= extent_offset
;
1250 disk_bytenr
+= cur_offset
- found_key
.offset
;
1251 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1253 * force cow if csum exists in the range.
1254 * this ensure that csum for a given extent are
1255 * either valid or do not exist.
1257 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1260 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1261 extent_end
= found_key
.offset
+
1262 btrfs_file_extent_inline_len(leaf
, fi
);
1263 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1268 if (extent_end
<= start
) {
1273 if (cow_start
== (u64
)-1)
1274 cow_start
= cur_offset
;
1275 cur_offset
= extent_end
;
1276 if (cur_offset
> end
)
1282 btrfs_release_path(path
);
1283 if (cow_start
!= (u64
)-1) {
1284 ret
= cow_file_range(inode
, locked_page
, cow_start
,
1285 found_key
.offset
- 1, page_started
,
1288 btrfs_abort_transaction(trans
, root
, ret
);
1291 cow_start
= (u64
)-1;
1294 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1295 struct extent_map
*em
;
1296 struct extent_map_tree
*em_tree
;
1297 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1298 em
= alloc_extent_map();
1299 BUG_ON(!em
); /* -ENOMEM */
1300 em
->start
= cur_offset
;
1301 em
->orig_start
= em
->start
;
1302 em
->len
= num_bytes
;
1303 em
->block_len
= num_bytes
;
1304 em
->block_start
= disk_bytenr
;
1305 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1306 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1307 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
1309 write_lock(&em_tree
->lock
);
1310 ret
= add_extent_mapping(em_tree
, em
);
1311 write_unlock(&em_tree
->lock
);
1312 if (ret
!= -EEXIST
) {
1313 free_extent_map(em
);
1316 btrfs_drop_extent_cache(inode
, em
->start
,
1317 em
->start
+ em
->len
- 1, 0);
1319 type
= BTRFS_ORDERED_PREALLOC
;
1321 type
= BTRFS_ORDERED_NOCOW
;
1324 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1325 num_bytes
, num_bytes
, type
);
1326 BUG_ON(ret
); /* -ENOMEM */
1328 if (root
->root_key
.objectid
==
1329 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1330 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1333 btrfs_abort_transaction(trans
, root
, ret
);
1338 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1339 cur_offset
, cur_offset
+ num_bytes
- 1,
1340 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1341 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1342 EXTENT_SET_PRIVATE2
);
1343 cur_offset
= extent_end
;
1344 if (cur_offset
> end
)
1347 btrfs_release_path(path
);
1349 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1350 cow_start
= cur_offset
;
1354 if (cow_start
!= (u64
)-1) {
1355 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1356 page_started
, nr_written
, 1);
1358 btrfs_abort_transaction(trans
, root
, ret
);
1364 err
= btrfs_end_transaction(trans
, root
);
1368 if (ret
&& cur_offset
< end
)
1369 extent_clear_unlock_delalloc(inode
,
1370 &BTRFS_I(inode
)->io_tree
,
1371 cur_offset
, end
, locked_page
,
1372 EXTENT_CLEAR_UNLOCK_PAGE
|
1373 EXTENT_CLEAR_UNLOCK
|
1374 EXTENT_CLEAR_DELALLOC
|
1375 EXTENT_CLEAR_DIRTY
|
1376 EXTENT_SET_WRITEBACK
|
1377 EXTENT_END_WRITEBACK
);
1379 btrfs_free_path(path
);
1384 * extent_io.c call back to do delayed allocation processing
1386 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1387 u64 start
, u64 end
, int *page_started
,
1388 unsigned long *nr_written
)
1391 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1393 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1394 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1395 page_started
, 1, nr_written
);
1396 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1397 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1398 page_started
, 0, nr_written
);
1399 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1400 !(BTRFS_I(inode
)->force_compress
) &&
1401 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1402 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1403 page_started
, nr_written
, 1);
1405 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1406 &BTRFS_I(inode
)->runtime_flags
);
1407 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1408 page_started
, nr_written
);
1413 static void btrfs_split_extent_hook(struct inode
*inode
,
1414 struct extent_state
*orig
, u64 split
)
1416 /* not delalloc, ignore it */
1417 if (!(orig
->state
& EXTENT_DELALLOC
))
1420 spin_lock(&BTRFS_I(inode
)->lock
);
1421 BTRFS_I(inode
)->outstanding_extents
++;
1422 spin_unlock(&BTRFS_I(inode
)->lock
);
1426 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1427 * extents so we can keep track of new extents that are just merged onto old
1428 * extents, such as when we are doing sequential writes, so we can properly
1429 * account for the metadata space we'll need.
1431 static void btrfs_merge_extent_hook(struct inode
*inode
,
1432 struct extent_state
*new,
1433 struct extent_state
*other
)
1435 /* not delalloc, ignore it */
1436 if (!(other
->state
& EXTENT_DELALLOC
))
1439 spin_lock(&BTRFS_I(inode
)->lock
);
1440 BTRFS_I(inode
)->outstanding_extents
--;
1441 spin_unlock(&BTRFS_I(inode
)->lock
);
1445 * extent_io.c set_bit_hook, used to track delayed allocation
1446 * bytes in this file, and to maintain the list of inodes that
1447 * have pending delalloc work to be done.
1449 static void btrfs_set_bit_hook(struct inode
*inode
,
1450 struct extent_state
*state
, int *bits
)
1454 * set_bit and clear bit hooks normally require _irqsave/restore
1455 * but in this case, we are only testing for the DELALLOC
1456 * bit, which is only set or cleared with irqs on
1458 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1459 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1460 u64 len
= state
->end
+ 1 - state
->start
;
1461 bool do_list
= !btrfs_is_free_space_inode(inode
);
1463 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1464 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1466 spin_lock(&BTRFS_I(inode
)->lock
);
1467 BTRFS_I(inode
)->outstanding_extents
++;
1468 spin_unlock(&BTRFS_I(inode
)->lock
);
1471 spin_lock(&root
->fs_info
->delalloc_lock
);
1472 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1473 root
->fs_info
->delalloc_bytes
+= len
;
1474 if (do_list
&& list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1475 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1476 &root
->fs_info
->delalloc_inodes
);
1478 spin_unlock(&root
->fs_info
->delalloc_lock
);
1483 * extent_io.c clear_bit_hook, see set_bit_hook for why
1485 static void btrfs_clear_bit_hook(struct inode
*inode
,
1486 struct extent_state
*state
, int *bits
)
1489 * set_bit and clear bit hooks normally require _irqsave/restore
1490 * but in this case, we are only testing for the DELALLOC
1491 * bit, which is only set or cleared with irqs on
1493 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1494 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1495 u64 len
= state
->end
+ 1 - state
->start
;
1496 bool do_list
= !btrfs_is_free_space_inode(inode
);
1498 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1499 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1500 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1501 spin_lock(&BTRFS_I(inode
)->lock
);
1502 BTRFS_I(inode
)->outstanding_extents
--;
1503 spin_unlock(&BTRFS_I(inode
)->lock
);
1506 if (*bits
& EXTENT_DO_ACCOUNTING
)
1507 btrfs_delalloc_release_metadata(inode
, len
);
1509 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1511 btrfs_free_reserved_data_space(inode
, len
);
1513 spin_lock(&root
->fs_info
->delalloc_lock
);
1514 root
->fs_info
->delalloc_bytes
-= len
;
1515 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1517 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1518 !list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1519 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1521 spin_unlock(&root
->fs_info
->delalloc_lock
);
1526 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1527 * we don't create bios that span stripes or chunks
1529 int btrfs_merge_bio_hook(struct page
*page
, unsigned long offset
,
1530 size_t size
, struct bio
*bio
,
1531 unsigned long bio_flags
)
1533 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1534 struct btrfs_mapping_tree
*map_tree
;
1535 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1540 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1543 length
= bio
->bi_size
;
1544 map_tree
= &root
->fs_info
->mapping_tree
;
1545 map_length
= length
;
1546 ret
= btrfs_map_block(map_tree
, READ
, logical
,
1547 &map_length
, NULL
, 0);
1548 /* Will always return 0 or 1 with map_multi == NULL */
1550 if (map_length
< length
+ size
)
1556 * in order to insert checksums into the metadata in large chunks,
1557 * we wait until bio submission time. All the pages in the bio are
1558 * checksummed and sums are attached onto the ordered extent record.
1560 * At IO completion time the cums attached on the ordered extent record
1561 * are inserted into the btree
1563 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1564 struct bio
*bio
, int mirror_num
,
1565 unsigned long bio_flags
,
1568 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1571 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1572 BUG_ON(ret
); /* -ENOMEM */
1577 * in order to insert checksums into the metadata in large chunks,
1578 * we wait until bio submission time. All the pages in the bio are
1579 * checksummed and sums are attached onto the ordered extent record.
1581 * At IO completion time the cums attached on the ordered extent record
1582 * are inserted into the btree
1584 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1585 int mirror_num
, unsigned long bio_flags
,
1588 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1589 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1593 * extent_io.c submission hook. This does the right thing for csum calculation
1594 * on write, or reading the csums from the tree before a read
1596 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1597 int mirror_num
, unsigned long bio_flags
,
1600 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1605 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1607 if (btrfs_is_free_space_inode(inode
))
1610 if (!(rw
& REQ_WRITE
)) {
1611 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1615 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1616 return btrfs_submit_compressed_read(inode
, bio
,
1617 mirror_num
, bio_flags
);
1618 } else if (!skip_sum
) {
1619 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1624 } else if (!skip_sum
) {
1625 /* csum items have already been cloned */
1626 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1628 /* we're doing a write, do the async checksumming */
1629 return btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1630 inode
, rw
, bio
, mirror_num
,
1631 bio_flags
, bio_offset
,
1632 __btrfs_submit_bio_start
,
1633 __btrfs_submit_bio_done
);
1637 return btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1641 * given a list of ordered sums record them in the inode. This happens
1642 * at IO completion time based on sums calculated at bio submission time.
1644 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1645 struct inode
*inode
, u64 file_offset
,
1646 struct list_head
*list
)
1648 struct btrfs_ordered_sum
*sum
;
1650 list_for_each_entry(sum
, list
, list
) {
1651 btrfs_csum_file_blocks(trans
,
1652 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1657 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1658 struct extent_state
**cached_state
)
1660 if ((end
& (PAGE_CACHE_SIZE
- 1)) == 0)
1662 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1663 cached_state
, GFP_NOFS
);
1666 /* see btrfs_writepage_start_hook for details on why this is required */
1667 struct btrfs_writepage_fixup
{
1669 struct btrfs_work work
;
1672 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1674 struct btrfs_writepage_fixup
*fixup
;
1675 struct btrfs_ordered_extent
*ordered
;
1676 struct extent_state
*cached_state
= NULL
;
1678 struct inode
*inode
;
1683 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1687 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1688 ClearPageChecked(page
);
1692 inode
= page
->mapping
->host
;
1693 page_start
= page_offset(page
);
1694 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1696 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1699 /* already ordered? We're done */
1700 if (PagePrivate2(page
))
1703 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1705 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1706 page_end
, &cached_state
, GFP_NOFS
);
1708 btrfs_start_ordered_extent(inode
, ordered
, 1);
1709 btrfs_put_ordered_extent(ordered
);
1713 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1715 mapping_set_error(page
->mapping
, ret
);
1716 end_extent_writepage(page
, ret
, page_start
, page_end
);
1717 ClearPageChecked(page
);
1721 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1722 ClearPageChecked(page
);
1723 set_page_dirty(page
);
1725 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1726 &cached_state
, GFP_NOFS
);
1729 page_cache_release(page
);
1734 * There are a few paths in the higher layers of the kernel that directly
1735 * set the page dirty bit without asking the filesystem if it is a
1736 * good idea. This causes problems because we want to make sure COW
1737 * properly happens and the data=ordered rules are followed.
1739 * In our case any range that doesn't have the ORDERED bit set
1740 * hasn't been properly setup for IO. We kick off an async process
1741 * to fix it up. The async helper will wait for ordered extents, set
1742 * the delalloc bit and make it safe to write the page.
1744 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1746 struct inode
*inode
= page
->mapping
->host
;
1747 struct btrfs_writepage_fixup
*fixup
;
1748 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1750 /* this page is properly in the ordered list */
1751 if (TestClearPagePrivate2(page
))
1754 if (PageChecked(page
))
1757 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1761 SetPageChecked(page
);
1762 page_cache_get(page
);
1763 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1765 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1769 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1770 struct inode
*inode
, u64 file_pos
,
1771 u64 disk_bytenr
, u64 disk_num_bytes
,
1772 u64 num_bytes
, u64 ram_bytes
,
1773 u8 compression
, u8 encryption
,
1774 u16 other_encoding
, int extent_type
)
1776 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1777 struct btrfs_file_extent_item
*fi
;
1778 struct btrfs_path
*path
;
1779 struct extent_buffer
*leaf
;
1780 struct btrfs_key ins
;
1783 path
= btrfs_alloc_path();
1787 path
->leave_spinning
= 1;
1790 * we may be replacing one extent in the tree with another.
1791 * The new extent is pinned in the extent map, and we don't want
1792 * to drop it from the cache until it is completely in the btree.
1794 * So, tell btrfs_drop_extents to leave this extent in the cache.
1795 * the caller is expected to unpin it and allow it to be merged
1798 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1799 file_pos
+ num_bytes
, 0);
1803 ins
.objectid
= btrfs_ino(inode
);
1804 ins
.offset
= file_pos
;
1805 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1806 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1809 leaf
= path
->nodes
[0];
1810 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1811 struct btrfs_file_extent_item
);
1812 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1813 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1814 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1815 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1816 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1817 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1818 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1819 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1820 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1821 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1823 btrfs_mark_buffer_dirty(leaf
);
1824 btrfs_release_path(path
);
1826 inode_add_bytes(inode
, num_bytes
);
1828 ins
.objectid
= disk_bytenr
;
1829 ins
.offset
= disk_num_bytes
;
1830 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1831 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1832 root
->root_key
.objectid
,
1833 btrfs_ino(inode
), file_pos
, &ins
);
1835 btrfs_free_path(path
);
1841 * helper function for btrfs_finish_ordered_io, this
1842 * just reads in some of the csum leaves to prime them into ram
1843 * before we start the transaction. It limits the amount of btree
1844 * reads required while inside the transaction.
1846 /* as ordered data IO finishes, this gets called so we can finish
1847 * an ordered extent if the range of bytes in the file it covers are
1850 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
1852 struct inode
*inode
= ordered_extent
->inode
;
1853 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1854 struct btrfs_trans_handle
*trans
= NULL
;
1855 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
1856 struct extent_state
*cached_state
= NULL
;
1857 int compress_type
= 0;
1861 nolock
= btrfs_is_free_space_inode(inode
);
1863 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
1868 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
1869 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
1870 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1873 trans
= btrfs_join_transaction_nolock(root
);
1875 trans
= btrfs_join_transaction(root
);
1876 if (IS_ERR(trans
)) {
1877 ret
= PTR_ERR(trans
);
1881 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1882 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1883 if (ret
) /* -ENOMEM or corruption */
1884 btrfs_abort_transaction(trans
, root
, ret
);
1889 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
1890 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
1894 trans
= btrfs_join_transaction_nolock(root
);
1896 trans
= btrfs_join_transaction(root
);
1897 if (IS_ERR(trans
)) {
1898 ret
= PTR_ERR(trans
);
1902 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1904 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
1905 compress_type
= ordered_extent
->compress_type
;
1906 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1907 BUG_ON(compress_type
);
1908 ret
= btrfs_mark_extent_written(trans
, inode
,
1909 ordered_extent
->file_offset
,
1910 ordered_extent
->file_offset
+
1911 ordered_extent
->len
);
1913 BUG_ON(root
== root
->fs_info
->tree_root
);
1914 ret
= insert_reserved_file_extent(trans
, inode
,
1915 ordered_extent
->file_offset
,
1916 ordered_extent
->start
,
1917 ordered_extent
->disk_len
,
1918 ordered_extent
->len
,
1919 ordered_extent
->len
,
1920 compress_type
, 0, 0,
1921 BTRFS_FILE_EXTENT_REG
);
1923 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
1924 ordered_extent
->file_offset
, ordered_extent
->len
,
1927 btrfs_abort_transaction(trans
, root
, ret
);
1931 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
1932 &ordered_extent
->list
);
1934 ret
= btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
1935 if (!ret
|| !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
1936 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
1937 if (ret
) { /* -ENOMEM or corruption */
1938 btrfs_abort_transaction(trans
, root
, ret
);
1942 btrfs_set_inode_last_trans(trans
, inode
);
1946 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
1947 ordered_extent
->file_offset
+
1948 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
1950 if (root
!= root
->fs_info
->tree_root
)
1951 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
1953 btrfs_end_transaction(trans
, root
);
1956 clear_extent_uptodate(io_tree
, ordered_extent
->file_offset
,
1957 ordered_extent
->file_offset
+
1958 ordered_extent
->len
- 1, NULL
, GFP_NOFS
);
1961 * This needs to be done to make sure anybody waiting knows we are done
1962 * updating everything for this ordered extent.
1964 btrfs_remove_ordered_extent(inode
, ordered_extent
);
1967 btrfs_put_ordered_extent(ordered_extent
);
1968 /* once for the tree */
1969 btrfs_put_ordered_extent(ordered_extent
);
1974 static void finish_ordered_fn(struct btrfs_work
*work
)
1976 struct btrfs_ordered_extent
*ordered_extent
;
1977 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
1978 btrfs_finish_ordered_io(ordered_extent
);
1981 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
1982 struct extent_state
*state
, int uptodate
)
1984 struct inode
*inode
= page
->mapping
->host
;
1985 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1986 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
1987 struct btrfs_workers
*workers
;
1989 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
1991 ClearPagePrivate2(page
);
1992 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
1993 end
- start
+ 1, uptodate
))
1996 ordered_extent
->work
.func
= finish_ordered_fn
;
1997 ordered_extent
->work
.flags
= 0;
1999 if (btrfs_is_free_space_inode(inode
))
2000 workers
= &root
->fs_info
->endio_freespace_worker
;
2002 workers
= &root
->fs_info
->endio_write_workers
;
2003 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2009 * when reads are done, we need to check csums to verify the data is correct
2010 * if there's a match, we allow the bio to finish. If not, the code in
2011 * extent_io.c will try to find good copies for us.
2013 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2014 struct extent_state
*state
, int mirror
)
2016 size_t offset
= start
- ((u64
)page
->index
<< PAGE_CACHE_SHIFT
);
2017 struct inode
*inode
= page
->mapping
->host
;
2018 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2020 u64
private = ~(u32
)0;
2022 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2025 if (PageChecked(page
)) {
2026 ClearPageChecked(page
);
2030 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2033 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2034 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2035 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2040 if (state
&& state
->start
== start
) {
2041 private = state
->private;
2044 ret
= get_state_private(io_tree
, start
, &private);
2046 kaddr
= kmap_atomic(page
);
2050 csum
= btrfs_csum_data(root
, kaddr
+ offset
, csum
, end
- start
+ 1);
2051 btrfs_csum_final(csum
, (char *)&csum
);
2052 if (csum
!= private)
2055 kunmap_atomic(kaddr
);
2060 printk_ratelimited(KERN_INFO
"btrfs csum failed ino %llu off %llu csum %u "
2062 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2063 (unsigned long long)start
, csum
,
2064 (unsigned long long)private);
2065 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2066 flush_dcache_page(page
);
2067 kunmap_atomic(kaddr
);
2073 struct delayed_iput
{
2074 struct list_head list
;
2075 struct inode
*inode
;
2078 /* JDM: If this is fs-wide, why can't we add a pointer to
2079 * btrfs_inode instead and avoid the allocation? */
2080 void btrfs_add_delayed_iput(struct inode
*inode
)
2082 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2083 struct delayed_iput
*delayed
;
2085 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2088 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2089 delayed
->inode
= inode
;
2091 spin_lock(&fs_info
->delayed_iput_lock
);
2092 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2093 spin_unlock(&fs_info
->delayed_iput_lock
);
2096 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2099 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2100 struct delayed_iput
*delayed
;
2103 spin_lock(&fs_info
->delayed_iput_lock
);
2104 empty
= list_empty(&fs_info
->delayed_iputs
);
2105 spin_unlock(&fs_info
->delayed_iput_lock
);
2109 spin_lock(&fs_info
->delayed_iput_lock
);
2110 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2111 spin_unlock(&fs_info
->delayed_iput_lock
);
2113 while (!list_empty(&list
)) {
2114 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2115 list_del(&delayed
->list
);
2116 iput(delayed
->inode
);
2121 enum btrfs_orphan_cleanup_state
{
2122 ORPHAN_CLEANUP_STARTED
= 1,
2123 ORPHAN_CLEANUP_DONE
= 2,
2127 * This is called in transaction commit time. If there are no orphan
2128 * files in the subvolume, it removes orphan item and frees block_rsv
2131 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2132 struct btrfs_root
*root
)
2134 struct btrfs_block_rsv
*block_rsv
;
2137 if (atomic_read(&root
->orphan_inodes
) ||
2138 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2141 spin_lock(&root
->orphan_lock
);
2142 if (atomic_read(&root
->orphan_inodes
)) {
2143 spin_unlock(&root
->orphan_lock
);
2147 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2148 spin_unlock(&root
->orphan_lock
);
2152 block_rsv
= root
->orphan_block_rsv
;
2153 root
->orphan_block_rsv
= NULL
;
2154 spin_unlock(&root
->orphan_lock
);
2156 if (root
->orphan_item_inserted
&&
2157 btrfs_root_refs(&root
->root_item
) > 0) {
2158 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2159 root
->root_key
.objectid
);
2161 root
->orphan_item_inserted
= 0;
2165 WARN_ON(block_rsv
->size
> 0);
2166 btrfs_free_block_rsv(root
, block_rsv
);
2171 * This creates an orphan entry for the given inode in case something goes
2172 * wrong in the middle of an unlink/truncate.
2174 * NOTE: caller of this function should reserve 5 units of metadata for
2177 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2179 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2180 struct btrfs_block_rsv
*block_rsv
= NULL
;
2185 if (!root
->orphan_block_rsv
) {
2186 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2191 spin_lock(&root
->orphan_lock
);
2192 if (!root
->orphan_block_rsv
) {
2193 root
->orphan_block_rsv
= block_rsv
;
2194 } else if (block_rsv
) {
2195 btrfs_free_block_rsv(root
, block_rsv
);
2199 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2200 &BTRFS_I(inode
)->runtime_flags
)) {
2203 * For proper ENOSPC handling, we should do orphan
2204 * cleanup when mounting. But this introduces backward
2205 * compatibility issue.
2207 if (!xchg(&root
->orphan_item_inserted
, 1))
2213 atomic_inc(&root
->orphan_inodes
);
2216 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2217 &BTRFS_I(inode
)->runtime_flags
))
2219 spin_unlock(&root
->orphan_lock
);
2221 /* grab metadata reservation from transaction handle */
2223 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2224 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
2227 /* insert an orphan item to track this unlinked/truncated file */
2229 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
2230 if (ret
&& ret
!= -EEXIST
) {
2231 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2232 &BTRFS_I(inode
)->runtime_flags
);
2233 btrfs_abort_transaction(trans
, root
, ret
);
2239 /* insert an orphan item to track subvolume contains orphan files */
2241 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
2242 root
->root_key
.objectid
);
2243 if (ret
&& ret
!= -EEXIST
) {
2244 btrfs_abort_transaction(trans
, root
, ret
);
2252 * We have done the truncate/delete so we can go ahead and remove the orphan
2253 * item for this particular inode.
2255 int btrfs_orphan_del(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2257 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2258 int delete_item
= 0;
2259 int release_rsv
= 0;
2262 spin_lock(&root
->orphan_lock
);
2263 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2264 &BTRFS_I(inode
)->runtime_flags
))
2267 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2268 &BTRFS_I(inode
)->runtime_flags
))
2270 spin_unlock(&root
->orphan_lock
);
2272 if (trans
&& delete_item
) {
2273 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
2274 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2278 btrfs_orphan_release_metadata(inode
);
2279 atomic_dec(&root
->orphan_inodes
);
2286 * this cleans up any orphans that may be left on the list from the last use
2289 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
2291 struct btrfs_path
*path
;
2292 struct extent_buffer
*leaf
;
2293 struct btrfs_key key
, found_key
;
2294 struct btrfs_trans_handle
*trans
;
2295 struct inode
*inode
;
2296 u64 last_objectid
= 0;
2297 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
2299 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
2302 path
= btrfs_alloc_path();
2309 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
2310 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
2311 key
.offset
= (u64
)-1;
2314 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2319 * if ret == 0 means we found what we were searching for, which
2320 * is weird, but possible, so only screw with path if we didn't
2321 * find the key and see if we have stuff that matches
2325 if (path
->slots
[0] == 0)
2330 /* pull out the item */
2331 leaf
= path
->nodes
[0];
2332 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2334 /* make sure the item matches what we want */
2335 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
2337 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
2340 /* release the path since we're done with it */
2341 btrfs_release_path(path
);
2344 * this is where we are basically btrfs_lookup, without the
2345 * crossing root thing. we store the inode number in the
2346 * offset of the orphan item.
2349 if (found_key
.offset
== last_objectid
) {
2350 printk(KERN_ERR
"btrfs: Error removing orphan entry, "
2351 "stopping orphan cleanup\n");
2356 last_objectid
= found_key
.offset
;
2358 found_key
.objectid
= found_key
.offset
;
2359 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
2360 found_key
.offset
= 0;
2361 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
2362 ret
= PTR_RET(inode
);
2363 if (ret
&& ret
!= -ESTALE
)
2366 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
2367 struct btrfs_root
*dead_root
;
2368 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2369 int is_dead_root
= 0;
2372 * this is an orphan in the tree root. Currently these
2373 * could come from 2 sources:
2374 * a) a snapshot deletion in progress
2375 * b) a free space cache inode
2376 * We need to distinguish those two, as the snapshot
2377 * orphan must not get deleted.
2378 * find_dead_roots already ran before us, so if this
2379 * is a snapshot deletion, we should find the root
2380 * in the dead_roots list
2382 spin_lock(&fs_info
->trans_lock
);
2383 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
2385 if (dead_root
->root_key
.objectid
==
2386 found_key
.objectid
) {
2391 spin_unlock(&fs_info
->trans_lock
);
2393 /* prevent this orphan from being found again */
2394 key
.offset
= found_key
.objectid
- 1;
2399 * Inode is already gone but the orphan item is still there,
2400 * kill the orphan item.
2402 if (ret
== -ESTALE
) {
2403 trans
= btrfs_start_transaction(root
, 1);
2404 if (IS_ERR(trans
)) {
2405 ret
= PTR_ERR(trans
);
2408 printk(KERN_ERR
"auto deleting %Lu\n",
2409 found_key
.objectid
);
2410 ret
= btrfs_del_orphan_item(trans
, root
,
2411 found_key
.objectid
);
2412 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
2413 btrfs_end_transaction(trans
, root
);
2418 * add this inode to the orphan list so btrfs_orphan_del does
2419 * the proper thing when we hit it
2421 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2422 &BTRFS_I(inode
)->runtime_flags
);
2424 /* if we have links, this was a truncate, lets do that */
2425 if (inode
->i_nlink
) {
2426 if (!S_ISREG(inode
->i_mode
)) {
2432 ret
= btrfs_truncate(inode
);
2437 /* this will do delete_inode and everything for us */
2442 /* release the path since we're done with it */
2443 btrfs_release_path(path
);
2445 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
2447 if (root
->orphan_block_rsv
)
2448 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
2451 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
2452 trans
= btrfs_join_transaction(root
);
2454 btrfs_end_transaction(trans
, root
);
2458 printk(KERN_INFO
"btrfs: unlinked %d orphans\n", nr_unlink
);
2460 printk(KERN_INFO
"btrfs: truncated %d orphans\n", nr_truncate
);
2464 printk(KERN_CRIT
"btrfs: could not do orphan cleanup %d\n", ret
);
2465 btrfs_free_path(path
);
2470 * very simple check to peek ahead in the leaf looking for xattrs. If we
2471 * don't find any xattrs, we know there can't be any acls.
2473 * slot is the slot the inode is in, objectid is the objectid of the inode
2475 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
2476 int slot
, u64 objectid
)
2478 u32 nritems
= btrfs_header_nritems(leaf
);
2479 struct btrfs_key found_key
;
2483 while (slot
< nritems
) {
2484 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
2486 /* we found a different objectid, there must not be acls */
2487 if (found_key
.objectid
!= objectid
)
2490 /* we found an xattr, assume we've got an acl */
2491 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
2495 * we found a key greater than an xattr key, there can't
2496 * be any acls later on
2498 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
2505 * it goes inode, inode backrefs, xattrs, extents,
2506 * so if there are a ton of hard links to an inode there can
2507 * be a lot of backrefs. Don't waste time searching too hard,
2508 * this is just an optimization
2513 /* we hit the end of the leaf before we found an xattr or
2514 * something larger than an xattr. We have to assume the inode
2521 * read an inode from the btree into the in-memory inode
2523 static void btrfs_read_locked_inode(struct inode
*inode
)
2525 struct btrfs_path
*path
;
2526 struct extent_buffer
*leaf
;
2527 struct btrfs_inode_item
*inode_item
;
2528 struct btrfs_timespec
*tspec
;
2529 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2530 struct btrfs_key location
;
2534 bool filled
= false;
2536 ret
= btrfs_fill_inode(inode
, &rdev
);
2540 path
= btrfs_alloc_path();
2544 path
->leave_spinning
= 1;
2545 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
2547 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
2551 leaf
= path
->nodes
[0];
2556 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2557 struct btrfs_inode_item
);
2558 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
2559 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
2560 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
2561 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
2562 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
2564 tspec
= btrfs_inode_atime(inode_item
);
2565 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2566 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2568 tspec
= btrfs_inode_mtime(inode_item
);
2569 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2570 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2572 tspec
= btrfs_inode_ctime(inode_item
);
2573 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
2574 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
2576 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
2577 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
2578 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
2581 * If we were modified in the current generation and evicted from memory
2582 * and then re-read we need to do a full sync since we don't have any
2583 * idea about which extents were modified before we were evicted from
2586 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
2587 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
2588 &BTRFS_I(inode
)->runtime_flags
);
2590 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
2591 inode
->i_generation
= BTRFS_I(inode
)->generation
;
2593 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
2595 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
2596 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
2599 * try to precache a NULL acl entry for files that don't have
2600 * any xattrs or acls
2602 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
2605 cache_no_acl(inode
);
2607 btrfs_free_path(path
);
2609 switch (inode
->i_mode
& S_IFMT
) {
2611 inode
->i_mapping
->a_ops
= &btrfs_aops
;
2612 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2613 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
2614 inode
->i_fop
= &btrfs_file_operations
;
2615 inode
->i_op
= &btrfs_file_inode_operations
;
2618 inode
->i_fop
= &btrfs_dir_file_operations
;
2619 if (root
== root
->fs_info
->tree_root
)
2620 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
2622 inode
->i_op
= &btrfs_dir_inode_operations
;
2625 inode
->i_op
= &btrfs_symlink_inode_operations
;
2626 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
2627 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
2630 inode
->i_op
= &btrfs_special_inode_operations
;
2631 init_special_inode(inode
, inode
->i_mode
, rdev
);
2635 btrfs_update_iflags(inode
);
2639 btrfs_free_path(path
);
2640 make_bad_inode(inode
);
2644 * given a leaf and an inode, copy the inode fields into the leaf
2646 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
2647 struct extent_buffer
*leaf
,
2648 struct btrfs_inode_item
*item
,
2649 struct inode
*inode
)
2651 btrfs_set_inode_uid(leaf
, item
, i_uid_read(inode
));
2652 btrfs_set_inode_gid(leaf
, item
, i_gid_read(inode
));
2653 btrfs_set_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
);
2654 btrfs_set_inode_mode(leaf
, item
, inode
->i_mode
);
2655 btrfs_set_inode_nlink(leaf
, item
, inode
->i_nlink
);
2657 btrfs_set_timespec_sec(leaf
, btrfs_inode_atime(item
),
2658 inode
->i_atime
.tv_sec
);
2659 btrfs_set_timespec_nsec(leaf
, btrfs_inode_atime(item
),
2660 inode
->i_atime
.tv_nsec
);
2662 btrfs_set_timespec_sec(leaf
, btrfs_inode_mtime(item
),
2663 inode
->i_mtime
.tv_sec
);
2664 btrfs_set_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
2665 inode
->i_mtime
.tv_nsec
);
2667 btrfs_set_timespec_sec(leaf
, btrfs_inode_ctime(item
),
2668 inode
->i_ctime
.tv_sec
);
2669 btrfs_set_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
2670 inode
->i_ctime
.tv_nsec
);
2672 btrfs_set_inode_nbytes(leaf
, item
, inode_get_bytes(inode
));
2673 btrfs_set_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
);
2674 btrfs_set_inode_sequence(leaf
, item
, inode
->i_version
);
2675 btrfs_set_inode_transid(leaf
, item
, trans
->transid
);
2676 btrfs_set_inode_rdev(leaf
, item
, inode
->i_rdev
);
2677 btrfs_set_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
);
2678 btrfs_set_inode_block_group(leaf
, item
, 0);
2682 * copy everything in the in-memory inode into the btree.
2684 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
2685 struct btrfs_root
*root
, struct inode
*inode
)
2687 struct btrfs_inode_item
*inode_item
;
2688 struct btrfs_path
*path
;
2689 struct extent_buffer
*leaf
;
2692 path
= btrfs_alloc_path();
2696 path
->leave_spinning
= 1;
2697 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
2705 btrfs_unlock_up_safe(path
, 1);
2706 leaf
= path
->nodes
[0];
2707 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2708 struct btrfs_inode_item
);
2710 fill_inode_item(trans
, leaf
, inode_item
, inode
);
2711 btrfs_mark_buffer_dirty(leaf
);
2712 btrfs_set_inode_last_trans(trans
, inode
);
2715 btrfs_free_path(path
);
2720 * copy everything in the in-memory inode into the btree.
2722 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
2723 struct btrfs_root
*root
, struct inode
*inode
)
2728 * If the inode is a free space inode, we can deadlock during commit
2729 * if we put it into the delayed code.
2731 * The data relocation inode should also be directly updated
2734 if (!btrfs_is_free_space_inode(inode
)
2735 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
2736 btrfs_update_root_times(trans
, root
);
2738 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
2740 btrfs_set_inode_last_trans(trans
, inode
);
2744 return btrfs_update_inode_item(trans
, root
, inode
);
2747 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
2748 struct btrfs_root
*root
,
2749 struct inode
*inode
)
2753 ret
= btrfs_update_inode(trans
, root
, inode
);
2755 return btrfs_update_inode_item(trans
, root
, inode
);
2760 * unlink helper that gets used here in inode.c and in the tree logging
2761 * recovery code. It remove a link in a directory with a given name, and
2762 * also drops the back refs in the inode to the directory
2764 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2765 struct btrfs_root
*root
,
2766 struct inode
*dir
, struct inode
*inode
,
2767 const char *name
, int name_len
)
2769 struct btrfs_path
*path
;
2771 struct extent_buffer
*leaf
;
2772 struct btrfs_dir_item
*di
;
2773 struct btrfs_key key
;
2775 u64 ino
= btrfs_ino(inode
);
2776 u64 dir_ino
= btrfs_ino(dir
);
2778 path
= btrfs_alloc_path();
2784 path
->leave_spinning
= 1;
2785 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2786 name
, name_len
, -1);
2795 leaf
= path
->nodes
[0];
2796 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
2797 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
2800 btrfs_release_path(path
);
2802 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
2805 printk(KERN_INFO
"btrfs failed to delete reference to %.*s, "
2806 "inode %llu parent %llu\n", name_len
, name
,
2807 (unsigned long long)ino
, (unsigned long long)dir_ino
);
2808 btrfs_abort_transaction(trans
, root
, ret
);
2812 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
2814 btrfs_abort_transaction(trans
, root
, ret
);
2818 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
2820 if (ret
!= 0 && ret
!= -ENOENT
) {
2821 btrfs_abort_transaction(trans
, root
, ret
);
2825 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
2830 btrfs_free_path(path
);
2834 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
2835 inode_inc_iversion(inode
);
2836 inode_inc_iversion(dir
);
2837 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
2838 ret
= btrfs_update_inode(trans
, root
, dir
);
2843 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
2844 struct btrfs_root
*root
,
2845 struct inode
*dir
, struct inode
*inode
,
2846 const char *name
, int name_len
)
2849 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
2851 btrfs_drop_nlink(inode
);
2852 ret
= btrfs_update_inode(trans
, root
, inode
);
2858 /* helper to check if there is any shared block in the path */
2859 static int check_path_shared(struct btrfs_root
*root
,
2860 struct btrfs_path
*path
)
2862 struct extent_buffer
*eb
;
2866 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
2869 if (!path
->nodes
[level
])
2871 eb
= path
->nodes
[level
];
2872 if (!btrfs_block_can_be_shared(root
, eb
))
2874 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, eb
->len
,
2883 * helper to start transaction for unlink and rmdir.
2885 * unlink and rmdir are special in btrfs, they do not always free space.
2886 * so in enospc case, we should make sure they will free space before
2887 * allowing them to use the global metadata reservation.
2889 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
2890 struct dentry
*dentry
)
2892 struct btrfs_trans_handle
*trans
;
2893 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
2894 struct btrfs_path
*path
;
2895 struct btrfs_dir_item
*di
;
2896 struct inode
*inode
= dentry
->d_inode
;
2901 u64 ino
= btrfs_ino(inode
);
2902 u64 dir_ino
= btrfs_ino(dir
);
2905 * 1 for the possible orphan item
2906 * 1 for the dir item
2907 * 1 for the dir index
2908 * 1 for the inode ref
2909 * 1 for the inode ref in the tree log
2910 * 2 for the dir entries in the log
2913 trans
= btrfs_start_transaction(root
, 8);
2914 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
2917 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
2918 return ERR_PTR(-ENOSPC
);
2920 /* check if there is someone else holds reference */
2921 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
2922 return ERR_PTR(-ENOSPC
);
2924 if (atomic_read(&inode
->i_count
) > 2)
2925 return ERR_PTR(-ENOSPC
);
2927 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
2928 return ERR_PTR(-ENOSPC
);
2930 path
= btrfs_alloc_path();
2932 root
->fs_info
->enospc_unlink
= 0;
2933 return ERR_PTR(-ENOMEM
);
2936 /* 1 for the orphan item */
2937 trans
= btrfs_start_transaction(root
, 1);
2938 if (IS_ERR(trans
)) {
2939 btrfs_free_path(path
);
2940 root
->fs_info
->enospc_unlink
= 0;
2944 path
->skip_locking
= 1;
2945 path
->search_commit_root
= 1;
2947 ret
= btrfs_lookup_inode(trans
, root
, path
,
2948 &BTRFS_I(dir
)->location
, 0);
2954 if (check_path_shared(root
, path
))
2959 btrfs_release_path(path
);
2961 ret
= btrfs_lookup_inode(trans
, root
, path
,
2962 &BTRFS_I(inode
)->location
, 0);
2968 if (check_path_shared(root
, path
))
2973 btrfs_release_path(path
);
2975 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
2976 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
2982 BUG_ON(ret
== 0); /* Corruption */
2983 if (check_path_shared(root
, path
))
2985 btrfs_release_path(path
);
2993 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
2994 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3000 if (check_path_shared(root
, path
))
3006 btrfs_release_path(path
);
3008 ret
= btrfs_get_inode_ref_index(trans
, root
, path
, dentry
->d_name
.name
,
3009 dentry
->d_name
.len
, ino
, dir_ino
, 0,
3016 if (check_path_shared(root
, path
))
3019 btrfs_release_path(path
);
3022 * This is a commit root search, if we can lookup inode item and other
3023 * relative items in the commit root, it means the transaction of
3024 * dir/file creation has been committed, and the dir index item that we
3025 * delay to insert has also been inserted into the commit root. So
3026 * we needn't worry about the delayed insertion of the dir index item
3029 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
3030 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3035 BUG_ON(ret
== -ENOENT
);
3036 if (check_path_shared(root
, path
))
3041 btrfs_free_path(path
);
3042 /* Migrate the orphan reservation over */
3044 err
= btrfs_block_rsv_migrate(trans
->block_rsv
,
3045 &root
->fs_info
->global_block_rsv
,
3046 trans
->bytes_reserved
);
3049 btrfs_end_transaction(trans
, root
);
3050 root
->fs_info
->enospc_unlink
= 0;
3051 return ERR_PTR(err
);
3054 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
3058 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
3059 struct btrfs_root
*root
)
3061 if (trans
->block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
) {
3062 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3063 trans
->bytes_reserved
);
3064 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3065 BUG_ON(!root
->fs_info
->enospc_unlink
);
3066 root
->fs_info
->enospc_unlink
= 0;
3068 btrfs_end_transaction(trans
, root
);
3071 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3073 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3074 struct btrfs_trans_handle
*trans
;
3075 struct inode
*inode
= dentry
->d_inode
;
3077 unsigned long nr
= 0;
3079 trans
= __unlink_start_trans(dir
, dentry
);
3081 return PTR_ERR(trans
);
3083 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3085 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3086 dentry
->d_name
.name
, dentry
->d_name
.len
);
3090 if (inode
->i_nlink
== 0) {
3091 ret
= btrfs_orphan_add(trans
, inode
);
3097 nr
= trans
->blocks_used
;
3098 __unlink_end_trans(trans
, root
);
3099 btrfs_btree_balance_dirty(root
, nr
);
3103 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3104 struct btrfs_root
*root
,
3105 struct inode
*dir
, u64 objectid
,
3106 const char *name
, int name_len
)
3108 struct btrfs_path
*path
;
3109 struct extent_buffer
*leaf
;
3110 struct btrfs_dir_item
*di
;
3111 struct btrfs_key key
;
3114 u64 dir_ino
= btrfs_ino(dir
);
3116 path
= btrfs_alloc_path();
3120 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3121 name
, name_len
, -1);
3122 if (IS_ERR_OR_NULL(di
)) {
3130 leaf
= path
->nodes
[0];
3131 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3132 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3133 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3135 btrfs_abort_transaction(trans
, root
, ret
);
3138 btrfs_release_path(path
);
3140 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3141 objectid
, root
->root_key
.objectid
,
3142 dir_ino
, &index
, name
, name_len
);
3144 if (ret
!= -ENOENT
) {
3145 btrfs_abort_transaction(trans
, root
, ret
);
3148 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3150 if (IS_ERR_OR_NULL(di
)) {
3155 btrfs_abort_transaction(trans
, root
, ret
);
3159 leaf
= path
->nodes
[0];
3160 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3161 btrfs_release_path(path
);
3164 btrfs_release_path(path
);
3166 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3168 btrfs_abort_transaction(trans
, root
, ret
);
3172 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3173 inode_inc_iversion(dir
);
3174 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3175 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
3177 btrfs_abort_transaction(trans
, root
, ret
);
3179 btrfs_free_path(path
);
3183 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3185 struct inode
*inode
= dentry
->d_inode
;
3187 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3188 struct btrfs_trans_handle
*trans
;
3189 unsigned long nr
= 0;
3191 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
3193 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3196 trans
= __unlink_start_trans(dir
, dentry
);
3198 return PTR_ERR(trans
);
3200 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3201 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3202 BTRFS_I(inode
)->location
.objectid
,
3203 dentry
->d_name
.name
,
3204 dentry
->d_name
.len
);
3208 err
= btrfs_orphan_add(trans
, inode
);
3212 /* now the directory is empty */
3213 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3214 dentry
->d_name
.name
, dentry
->d_name
.len
);
3216 btrfs_i_size_write(inode
, 0);
3218 nr
= trans
->blocks_used
;
3219 __unlink_end_trans(trans
, root
);
3220 btrfs_btree_balance_dirty(root
, nr
);
3226 * this can truncate away extent items, csum items and directory items.
3227 * It starts at a high offset and removes keys until it can't find
3228 * any higher than new_size
3230 * csum items that cross the new i_size are truncated to the new size
3233 * min_type is the minimum key type to truncate down to. If set to 0, this
3234 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3236 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3237 struct btrfs_root
*root
,
3238 struct inode
*inode
,
3239 u64 new_size
, u32 min_type
)
3241 struct btrfs_path
*path
;
3242 struct extent_buffer
*leaf
;
3243 struct btrfs_file_extent_item
*fi
;
3244 struct btrfs_key key
;
3245 struct btrfs_key found_key
;
3246 u64 extent_start
= 0;
3247 u64 extent_num_bytes
= 0;
3248 u64 extent_offset
= 0;
3250 u64 mask
= root
->sectorsize
- 1;
3251 u32 found_type
= (u8
)-1;
3254 int pending_del_nr
= 0;
3255 int pending_del_slot
= 0;
3256 int extent_type
= -1;
3259 u64 ino
= btrfs_ino(inode
);
3261 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3263 path
= btrfs_alloc_path();
3269 * We want to drop from the next block forward in case this new size is
3270 * not block aligned since we will be keeping the last block of the
3271 * extent just the way it is.
3273 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3274 btrfs_drop_extent_cache(inode
, (new_size
+ mask
) & (~mask
), (u64
)-1, 0);
3277 * This function is also used to drop the items in the log tree before
3278 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3279 * it is used to drop the loged items. So we shouldn't kill the delayed
3282 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3283 btrfs_kill_delayed_inode_items(inode
);
3286 key
.offset
= (u64
)-1;
3290 path
->leave_spinning
= 1;
3291 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
3298 /* there are no items in the tree for us to truncate, we're
3301 if (path
->slots
[0] == 0)
3308 leaf
= path
->nodes
[0];
3309 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3310 found_type
= btrfs_key_type(&found_key
);
3312 if (found_key
.objectid
!= ino
)
3315 if (found_type
< min_type
)
3318 item_end
= found_key
.offset
;
3319 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
3320 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3321 struct btrfs_file_extent_item
);
3322 extent_type
= btrfs_file_extent_type(leaf
, fi
);
3323 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3325 btrfs_file_extent_num_bytes(leaf
, fi
);
3326 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3327 item_end
+= btrfs_file_extent_inline_len(leaf
,
3332 if (found_type
> min_type
) {
3335 if (item_end
< new_size
)
3337 if (found_key
.offset
>= new_size
)
3343 /* FIXME, shrink the extent if the ref count is only 1 */
3344 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
3347 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
3349 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
3351 u64 orig_num_bytes
=
3352 btrfs_file_extent_num_bytes(leaf
, fi
);
3353 extent_num_bytes
= new_size
-
3354 found_key
.offset
+ root
->sectorsize
- 1;
3355 extent_num_bytes
= extent_num_bytes
&
3356 ~((u64
)root
->sectorsize
- 1);
3357 btrfs_set_file_extent_num_bytes(leaf
, fi
,
3359 num_dec
= (orig_num_bytes
-
3361 if (root
->ref_cows
&& extent_start
!= 0)
3362 inode_sub_bytes(inode
, num_dec
);
3363 btrfs_mark_buffer_dirty(leaf
);
3366 btrfs_file_extent_disk_num_bytes(leaf
,
3368 extent_offset
= found_key
.offset
-
3369 btrfs_file_extent_offset(leaf
, fi
);
3371 /* FIXME blocksize != 4096 */
3372 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
3373 if (extent_start
!= 0) {
3376 inode_sub_bytes(inode
, num_dec
);
3379 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
3381 * we can't truncate inline items that have had
3385 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
3386 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
3387 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
3388 u32 size
= new_size
- found_key
.offset
;
3390 if (root
->ref_cows
) {
3391 inode_sub_bytes(inode
, item_end
+ 1 -
3395 btrfs_file_extent_calc_inline_size(size
);
3396 btrfs_truncate_item(trans
, root
, path
,
3398 } else if (root
->ref_cows
) {
3399 inode_sub_bytes(inode
, item_end
+ 1 -
3405 if (!pending_del_nr
) {
3406 /* no pending yet, add ourselves */
3407 pending_del_slot
= path
->slots
[0];
3409 } else if (pending_del_nr
&&
3410 path
->slots
[0] + 1 == pending_del_slot
) {
3411 /* hop on the pending chunk */
3413 pending_del_slot
= path
->slots
[0];
3420 if (found_extent
&& (root
->ref_cows
||
3421 root
== root
->fs_info
->tree_root
)) {
3422 btrfs_set_path_blocking(path
);
3423 ret
= btrfs_free_extent(trans
, root
, extent_start
,
3424 extent_num_bytes
, 0,
3425 btrfs_header_owner(leaf
),
3426 ino
, extent_offset
, 0);
3430 if (found_type
== BTRFS_INODE_ITEM_KEY
)
3433 if (path
->slots
[0] == 0 ||
3434 path
->slots
[0] != pending_del_slot
) {
3435 if (pending_del_nr
) {
3436 ret
= btrfs_del_items(trans
, root
, path
,
3440 btrfs_abort_transaction(trans
,
3446 btrfs_release_path(path
);
3453 if (pending_del_nr
) {
3454 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
3457 btrfs_abort_transaction(trans
, root
, ret
);
3460 btrfs_free_path(path
);
3465 * btrfs_truncate_page - read, zero a chunk and write a page
3466 * @inode - inode that we're zeroing
3467 * @from - the offset to start zeroing
3468 * @len - the length to zero, 0 to zero the entire range respective to the
3470 * @front - zero up to the offset instead of from the offset on
3472 * This will find the page for the "from" offset and cow the page and zero the
3473 * part we want to zero. This is used with truncate and hole punching.
3475 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
3478 struct address_space
*mapping
= inode
->i_mapping
;
3479 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3480 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3481 struct btrfs_ordered_extent
*ordered
;
3482 struct extent_state
*cached_state
= NULL
;
3484 u32 blocksize
= root
->sectorsize
;
3485 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
3486 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3488 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
3493 if ((offset
& (blocksize
- 1)) == 0 &&
3494 (!len
|| ((len
& (blocksize
- 1)) == 0)))
3496 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
3502 page
= find_or_create_page(mapping
, index
, mask
);
3504 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3508 page_start
= page_offset(page
);
3509 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
3511 if (!PageUptodate(page
)) {
3512 ret
= btrfs_readpage(NULL
, page
);
3514 if (page
->mapping
!= mapping
) {
3516 page_cache_release(page
);
3519 if (!PageUptodate(page
)) {
3524 wait_on_page_writeback(page
);
3526 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
3527 set_page_extent_mapped(page
);
3529 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
3531 unlock_extent_cached(io_tree
, page_start
, page_end
,
3532 &cached_state
, GFP_NOFS
);
3534 page_cache_release(page
);
3535 btrfs_start_ordered_extent(inode
, ordered
, 1);
3536 btrfs_put_ordered_extent(ordered
);
3540 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
3541 EXTENT_DIRTY
| EXTENT_DELALLOC
|
3542 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
3543 0, 0, &cached_state
, GFP_NOFS
);
3545 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
3548 unlock_extent_cached(io_tree
, page_start
, page_end
,
3549 &cached_state
, GFP_NOFS
);
3554 if (offset
!= PAGE_CACHE_SIZE
) {
3556 len
= PAGE_CACHE_SIZE
- offset
;
3559 memset(kaddr
, 0, offset
);
3561 memset(kaddr
+ offset
, 0, len
);
3562 flush_dcache_page(page
);
3565 ClearPageChecked(page
);
3566 set_page_dirty(page
);
3567 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
3572 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
3574 page_cache_release(page
);
3580 * This function puts in dummy file extents for the area we're creating a hole
3581 * for. So if we are truncating this file to a larger size we need to insert
3582 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3583 * the range between oldsize and size
3585 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
3587 struct btrfs_trans_handle
*trans
;
3588 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3589 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
3590 struct extent_map
*em
= NULL
;
3591 struct extent_state
*cached_state
= NULL
;
3592 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
3593 u64 mask
= root
->sectorsize
- 1;
3594 u64 hole_start
= (oldsize
+ mask
) & ~mask
;
3595 u64 block_end
= (size
+ mask
) & ~mask
;
3601 if (size
<= hole_start
)
3605 struct btrfs_ordered_extent
*ordered
;
3606 btrfs_wait_ordered_range(inode
, hole_start
,
3607 block_end
- hole_start
);
3608 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
3610 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
3613 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
3614 &cached_state
, GFP_NOFS
);
3615 btrfs_put_ordered_extent(ordered
);
3618 cur_offset
= hole_start
;
3620 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
3621 block_end
- cur_offset
, 0);
3626 last_byte
= min(extent_map_end(em
), block_end
);
3627 last_byte
= (last_byte
+ mask
) & ~mask
;
3628 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
3629 struct extent_map
*hole_em
;
3630 hole_size
= last_byte
- cur_offset
;
3632 trans
= btrfs_start_transaction(root
, 3);
3633 if (IS_ERR(trans
)) {
3634 err
= PTR_ERR(trans
);
3638 err
= btrfs_drop_extents(trans
, root
, inode
,
3640 cur_offset
+ hole_size
, 1);
3642 btrfs_abort_transaction(trans
, root
, err
);
3643 btrfs_end_transaction(trans
, root
);
3647 err
= btrfs_insert_file_extent(trans
, root
,
3648 btrfs_ino(inode
), cur_offset
, 0,
3649 0, hole_size
, 0, hole_size
,
3652 btrfs_abort_transaction(trans
, root
, err
);
3653 btrfs_end_transaction(trans
, root
);
3657 btrfs_drop_extent_cache(inode
, cur_offset
,
3658 cur_offset
+ hole_size
- 1, 0);
3659 hole_em
= alloc_extent_map();
3661 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3662 &BTRFS_I(inode
)->runtime_flags
);
3665 hole_em
->start
= cur_offset
;
3666 hole_em
->len
= hole_size
;
3667 hole_em
->orig_start
= cur_offset
;
3669 hole_em
->block_start
= EXTENT_MAP_HOLE
;
3670 hole_em
->block_len
= 0;
3671 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
3672 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
3673 hole_em
->generation
= trans
->transid
;
3676 write_lock(&em_tree
->lock
);
3677 err
= add_extent_mapping(em_tree
, hole_em
);
3679 list_move(&hole_em
->list
,
3680 &em_tree
->modified_extents
);
3681 write_unlock(&em_tree
->lock
);
3684 btrfs_drop_extent_cache(inode
, cur_offset
,
3688 free_extent_map(hole_em
);
3690 btrfs_update_inode(trans
, root
, inode
);
3691 btrfs_end_transaction(trans
, root
);
3693 free_extent_map(em
);
3695 cur_offset
= last_byte
;
3696 if (cur_offset
>= block_end
)
3700 free_extent_map(em
);
3701 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
3706 static int btrfs_setsize(struct inode
*inode
, loff_t newsize
)
3708 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3709 struct btrfs_trans_handle
*trans
;
3710 loff_t oldsize
= i_size_read(inode
);
3713 if (newsize
== oldsize
)
3716 if (newsize
> oldsize
) {
3717 truncate_pagecache(inode
, oldsize
, newsize
);
3718 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
3722 trans
= btrfs_start_transaction(root
, 1);
3724 return PTR_ERR(trans
);
3726 i_size_write(inode
, newsize
);
3727 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
3728 ret
= btrfs_update_inode(trans
, root
, inode
);
3729 btrfs_end_transaction(trans
, root
);
3733 * We're truncating a file that used to have good data down to
3734 * zero. Make sure it gets into the ordered flush list so that
3735 * any new writes get down to disk quickly.
3738 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
3739 &BTRFS_I(inode
)->runtime_flags
);
3741 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3742 truncate_setsize(inode
, newsize
);
3743 ret
= btrfs_truncate(inode
);
3749 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
3751 struct inode
*inode
= dentry
->d_inode
;
3752 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3755 if (btrfs_root_readonly(root
))
3758 err
= inode_change_ok(inode
, attr
);
3762 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
3763 err
= btrfs_setsize(inode
, attr
->ia_size
);
3768 if (attr
->ia_valid
) {
3769 setattr_copy(inode
, attr
);
3770 inode_inc_iversion(inode
);
3771 err
= btrfs_dirty_inode(inode
);
3773 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
3774 err
= btrfs_acl_chmod(inode
);
3780 void btrfs_evict_inode(struct inode
*inode
)
3782 struct btrfs_trans_handle
*trans
;
3783 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3784 struct btrfs_block_rsv
*rsv
, *global_rsv
;
3785 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
3789 trace_btrfs_inode_evict(inode
);
3791 truncate_inode_pages(&inode
->i_data
, 0);
3792 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
3793 btrfs_is_free_space_inode(inode
)))
3796 if (is_bad_inode(inode
)) {
3797 btrfs_orphan_del(NULL
, inode
);
3800 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3801 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
3803 if (root
->fs_info
->log_root_recovering
) {
3804 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3805 &BTRFS_I(inode
)->runtime_flags
));
3809 if (inode
->i_nlink
> 0) {
3810 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
3814 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
3816 btrfs_orphan_del(NULL
, inode
);
3819 rsv
->size
= min_size
;
3821 global_rsv
= &root
->fs_info
->global_block_rsv
;
3823 btrfs_i_size_write(inode
, 0);
3826 * This is a bit simpler than btrfs_truncate since we've already
3827 * reserved our space for our orphan item in the unlink, so we just
3828 * need to reserve some slack space in case we add bytes and update
3829 * inode item when doing the truncate.
3832 ret
= btrfs_block_rsv_refill_noflush(root
, rsv
, min_size
);
3835 * Try and steal from the global reserve since we will
3836 * likely not use this space anyway, we want to try as
3837 * hard as possible to get this to work.
3840 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
3843 printk(KERN_WARNING
"Could not get space for a "
3844 "delete, will truncate on mount %d\n", ret
);
3845 btrfs_orphan_del(NULL
, inode
);
3846 btrfs_free_block_rsv(root
, rsv
);
3850 trans
= btrfs_start_transaction_noflush(root
, 1);
3851 if (IS_ERR(trans
)) {
3852 btrfs_orphan_del(NULL
, inode
);
3853 btrfs_free_block_rsv(root
, rsv
);
3857 trans
->block_rsv
= rsv
;
3859 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
3863 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3864 ret
= btrfs_update_inode(trans
, root
, inode
);
3867 nr
= trans
->blocks_used
;
3868 btrfs_end_transaction(trans
, root
);
3870 btrfs_btree_balance_dirty(root
, nr
);
3873 btrfs_free_block_rsv(root
, rsv
);
3876 trans
->block_rsv
= root
->orphan_block_rsv
;
3877 ret
= btrfs_orphan_del(trans
, inode
);
3881 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3882 if (!(root
== root
->fs_info
->tree_root
||
3883 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
3884 btrfs_return_ino(root
, btrfs_ino(inode
));
3886 nr
= trans
->blocks_used
;
3887 btrfs_end_transaction(trans
, root
);
3888 btrfs_btree_balance_dirty(root
, nr
);
3895 * this returns the key found in the dir entry in the location pointer.
3896 * If no dir entries were found, location->objectid is 0.
3898 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
3899 struct btrfs_key
*location
)
3901 const char *name
= dentry
->d_name
.name
;
3902 int namelen
= dentry
->d_name
.len
;
3903 struct btrfs_dir_item
*di
;
3904 struct btrfs_path
*path
;
3905 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3908 path
= btrfs_alloc_path();
3912 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
3917 if (IS_ERR_OR_NULL(di
))
3920 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
3922 btrfs_free_path(path
);
3925 location
->objectid
= 0;
3930 * when we hit a tree root in a directory, the btrfs part of the inode
3931 * needs to be changed to reflect the root directory of the tree root. This
3932 * is kind of like crossing a mount point.
3934 static int fixup_tree_root_location(struct btrfs_root
*root
,
3936 struct dentry
*dentry
,
3937 struct btrfs_key
*location
,
3938 struct btrfs_root
**sub_root
)
3940 struct btrfs_path
*path
;
3941 struct btrfs_root
*new_root
;
3942 struct btrfs_root_ref
*ref
;
3943 struct extent_buffer
*leaf
;
3947 path
= btrfs_alloc_path();
3954 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
3955 BTRFS_I(dir
)->root
->root_key
.objectid
,
3956 location
->objectid
);
3963 leaf
= path
->nodes
[0];
3964 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
3965 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
3966 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
3969 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
3970 (unsigned long)(ref
+ 1),
3971 dentry
->d_name
.len
);
3975 btrfs_release_path(path
);
3977 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
3978 if (IS_ERR(new_root
)) {
3979 err
= PTR_ERR(new_root
);
3983 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
3988 *sub_root
= new_root
;
3989 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
3990 location
->type
= BTRFS_INODE_ITEM_KEY
;
3991 location
->offset
= 0;
3994 btrfs_free_path(path
);
3998 static void inode_tree_add(struct inode
*inode
)
4000 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4001 struct btrfs_inode
*entry
;
4003 struct rb_node
*parent
;
4004 u64 ino
= btrfs_ino(inode
);
4006 p
= &root
->inode_tree
.rb_node
;
4009 if (inode_unhashed(inode
))
4012 spin_lock(&root
->inode_lock
);
4015 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
4017 if (ino
< btrfs_ino(&entry
->vfs_inode
))
4018 p
= &parent
->rb_left
;
4019 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
4020 p
= &parent
->rb_right
;
4022 WARN_ON(!(entry
->vfs_inode
.i_state
&
4023 (I_WILL_FREE
| I_FREEING
)));
4024 rb_erase(parent
, &root
->inode_tree
);
4025 RB_CLEAR_NODE(parent
);
4026 spin_unlock(&root
->inode_lock
);
4030 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
4031 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4032 spin_unlock(&root
->inode_lock
);
4035 static void inode_tree_del(struct inode
*inode
)
4037 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4040 spin_lock(&root
->inode_lock
);
4041 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
4042 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4043 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
4044 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4046 spin_unlock(&root
->inode_lock
);
4049 * Free space cache has inodes in the tree root, but the tree root has a
4050 * root_refs of 0, so this could end up dropping the tree root as a
4051 * snapshot, so we need the extra !root->fs_info->tree_root check to
4052 * make sure we don't drop it.
4054 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
4055 root
!= root
->fs_info
->tree_root
) {
4056 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4057 spin_lock(&root
->inode_lock
);
4058 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4059 spin_unlock(&root
->inode_lock
);
4061 btrfs_add_dead_root(root
);
4065 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4067 struct rb_node
*node
;
4068 struct rb_node
*prev
;
4069 struct btrfs_inode
*entry
;
4070 struct inode
*inode
;
4073 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4075 spin_lock(&root
->inode_lock
);
4077 node
= root
->inode_tree
.rb_node
;
4081 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4083 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4084 node
= node
->rb_left
;
4085 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4086 node
= node
->rb_right
;
4092 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4093 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4097 prev
= rb_next(prev
);
4101 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4102 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4103 inode
= igrab(&entry
->vfs_inode
);
4105 spin_unlock(&root
->inode_lock
);
4106 if (atomic_read(&inode
->i_count
) > 1)
4107 d_prune_aliases(inode
);
4109 * btrfs_drop_inode will have it removed from
4110 * the inode cache when its usage count
4115 spin_lock(&root
->inode_lock
);
4119 if (cond_resched_lock(&root
->inode_lock
))
4122 node
= rb_next(node
);
4124 spin_unlock(&root
->inode_lock
);
4127 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4129 struct btrfs_iget_args
*args
= p
;
4130 inode
->i_ino
= args
->ino
;
4131 BTRFS_I(inode
)->root
= args
->root
;
4135 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4137 struct btrfs_iget_args
*args
= opaque
;
4138 return args
->ino
== btrfs_ino(inode
) &&
4139 args
->root
== BTRFS_I(inode
)->root
;
4142 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4144 struct btrfs_root
*root
)
4146 struct inode
*inode
;
4147 struct btrfs_iget_args args
;
4148 args
.ino
= objectid
;
4151 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4152 btrfs_init_locked_inode
,
4157 /* Get an inode object given its location and corresponding root.
4158 * Returns in *is_new if the inode was read from disk
4160 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
4161 struct btrfs_root
*root
, int *new)
4163 struct inode
*inode
;
4165 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
4167 return ERR_PTR(-ENOMEM
);
4169 if (inode
->i_state
& I_NEW
) {
4170 BTRFS_I(inode
)->root
= root
;
4171 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
4172 btrfs_read_locked_inode(inode
);
4173 if (!is_bad_inode(inode
)) {
4174 inode_tree_add(inode
);
4175 unlock_new_inode(inode
);
4179 unlock_new_inode(inode
);
4181 inode
= ERR_PTR(-ESTALE
);
4188 static struct inode
*new_simple_dir(struct super_block
*s
,
4189 struct btrfs_key
*key
,
4190 struct btrfs_root
*root
)
4192 struct inode
*inode
= new_inode(s
);
4195 return ERR_PTR(-ENOMEM
);
4197 BTRFS_I(inode
)->root
= root
;
4198 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
4199 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
4201 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
4202 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
4203 inode
->i_fop
= &simple_dir_operations
;
4204 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
4205 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4210 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
4212 struct inode
*inode
;
4213 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4214 struct btrfs_root
*sub_root
= root
;
4215 struct btrfs_key location
;
4219 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
4220 return ERR_PTR(-ENAMETOOLONG
);
4222 if (unlikely(d_need_lookup(dentry
))) {
4223 memcpy(&location
, dentry
->d_fsdata
, sizeof(struct btrfs_key
));
4224 kfree(dentry
->d_fsdata
);
4225 dentry
->d_fsdata
= NULL
;
4226 /* This thing is hashed, drop it for now */
4229 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
4233 return ERR_PTR(ret
);
4235 if (location
.objectid
== 0)
4238 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
4239 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
4243 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
4245 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
4246 ret
= fixup_tree_root_location(root
, dir
, dentry
,
4247 &location
, &sub_root
);
4250 inode
= ERR_PTR(ret
);
4252 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
4254 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
4256 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
4258 if (!IS_ERR(inode
) && root
!= sub_root
) {
4259 down_read(&root
->fs_info
->cleanup_work_sem
);
4260 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
4261 ret
= btrfs_orphan_cleanup(sub_root
);
4262 up_read(&root
->fs_info
->cleanup_work_sem
);
4264 inode
= ERR_PTR(ret
);
4270 static int btrfs_dentry_delete(const struct dentry
*dentry
)
4272 struct btrfs_root
*root
;
4273 struct inode
*inode
= dentry
->d_inode
;
4275 if (!inode
&& !IS_ROOT(dentry
))
4276 inode
= dentry
->d_parent
->d_inode
;
4279 root
= BTRFS_I(inode
)->root
;
4280 if (btrfs_root_refs(&root
->root_item
) == 0)
4283 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
4289 static void btrfs_dentry_release(struct dentry
*dentry
)
4291 if (dentry
->d_fsdata
)
4292 kfree(dentry
->d_fsdata
);
4295 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
4300 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
4301 if (unlikely(d_need_lookup(dentry
))) {
4302 spin_lock(&dentry
->d_lock
);
4303 dentry
->d_flags
&= ~DCACHE_NEED_LOOKUP
;
4304 spin_unlock(&dentry
->d_lock
);
4309 unsigned char btrfs_filetype_table
[] = {
4310 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
4313 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
4316 struct inode
*inode
= filp
->f_dentry
->d_inode
;
4317 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4318 struct btrfs_item
*item
;
4319 struct btrfs_dir_item
*di
;
4320 struct btrfs_key key
;
4321 struct btrfs_key found_key
;
4322 struct btrfs_path
*path
;
4323 struct list_head ins_list
;
4324 struct list_head del_list
;
4326 struct extent_buffer
*leaf
;
4328 unsigned char d_type
;
4333 int key_type
= BTRFS_DIR_INDEX_KEY
;
4337 int is_curr
= 0; /* filp->f_pos points to the current index? */
4339 /* FIXME, use a real flag for deciding about the key type */
4340 if (root
->fs_info
->tree_root
== root
)
4341 key_type
= BTRFS_DIR_ITEM_KEY
;
4343 /* special case for "." */
4344 if (filp
->f_pos
== 0) {
4345 over
= filldir(dirent
, ".", 1,
4346 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
4351 /* special case for .., just use the back ref */
4352 if (filp
->f_pos
== 1) {
4353 u64 pino
= parent_ino(filp
->f_path
.dentry
);
4354 over
= filldir(dirent
, "..", 2,
4355 filp
->f_pos
, pino
, DT_DIR
);
4360 path
= btrfs_alloc_path();
4366 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4367 INIT_LIST_HEAD(&ins_list
);
4368 INIT_LIST_HEAD(&del_list
);
4369 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
4372 btrfs_set_key_type(&key
, key_type
);
4373 key
.offset
= filp
->f_pos
;
4374 key
.objectid
= btrfs_ino(inode
);
4376 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4381 leaf
= path
->nodes
[0];
4382 slot
= path
->slots
[0];
4383 if (slot
>= btrfs_header_nritems(leaf
)) {
4384 ret
= btrfs_next_leaf(root
, path
);
4392 item
= btrfs_item_nr(leaf
, slot
);
4393 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4395 if (found_key
.objectid
!= key
.objectid
)
4397 if (btrfs_key_type(&found_key
) != key_type
)
4399 if (found_key
.offset
< filp
->f_pos
)
4401 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
4402 btrfs_should_delete_dir_index(&del_list
,
4406 filp
->f_pos
= found_key
.offset
;
4409 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
4411 di_total
= btrfs_item_size(leaf
, item
);
4413 while (di_cur
< di_total
) {
4414 struct btrfs_key location
;
4416 if (verify_dir_item(root
, leaf
, di
))
4419 name_len
= btrfs_dir_name_len(leaf
, di
);
4420 if (name_len
<= sizeof(tmp_name
)) {
4421 name_ptr
= tmp_name
;
4423 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
4429 read_extent_buffer(leaf
, name_ptr
,
4430 (unsigned long)(di
+ 1), name_len
);
4432 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
4433 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
4436 /* is this a reference to our own snapshot? If so
4439 * In contrast to old kernels, we insert the snapshot's
4440 * dir item and dir index after it has been created, so
4441 * we won't find a reference to our own snapshot. We
4442 * still keep the following code for backward
4445 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
4446 location
.objectid
== root
->root_key
.objectid
) {
4450 over
= filldir(dirent
, name_ptr
, name_len
,
4451 found_key
.offset
, location
.objectid
,
4455 if (name_ptr
!= tmp_name
)
4460 di_len
= btrfs_dir_name_len(leaf
, di
) +
4461 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
4463 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
4469 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
4472 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
4478 /* Reached end of directory/root. Bump pos past the last item. */
4479 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4481 * 32-bit glibc will use getdents64, but then strtol -
4482 * so the last number we can serve is this.
4484 filp
->f_pos
= 0x7fffffff;
4490 if (key_type
== BTRFS_DIR_INDEX_KEY
)
4491 btrfs_put_delayed_items(&ins_list
, &del_list
);
4492 btrfs_free_path(path
);
4496 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4498 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4499 struct btrfs_trans_handle
*trans
;
4501 bool nolock
= false;
4503 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4506 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
4509 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
4511 trans
= btrfs_join_transaction_nolock(root
);
4513 trans
= btrfs_join_transaction(root
);
4515 return PTR_ERR(trans
);
4516 ret
= btrfs_commit_transaction(trans
, root
);
4522 * This is somewhat expensive, updating the tree every time the
4523 * inode changes. But, it is most likely to find the inode in cache.
4524 * FIXME, needs more benchmarking...there are no reasons other than performance
4525 * to keep or drop this code.
4527 int btrfs_dirty_inode(struct inode
*inode
)
4529 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4530 struct btrfs_trans_handle
*trans
;
4533 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
4536 trans
= btrfs_join_transaction(root
);
4538 return PTR_ERR(trans
);
4540 ret
= btrfs_update_inode(trans
, root
, inode
);
4541 if (ret
&& ret
== -ENOSPC
) {
4542 /* whoops, lets try again with the full transaction */
4543 btrfs_end_transaction(trans
, root
);
4544 trans
= btrfs_start_transaction(root
, 1);
4546 return PTR_ERR(trans
);
4548 ret
= btrfs_update_inode(trans
, root
, inode
);
4550 btrfs_end_transaction(trans
, root
);
4551 if (BTRFS_I(inode
)->delayed_node
)
4552 btrfs_balance_delayed_items(root
);
4558 * This is a copy of file_update_time. We need this so we can return error on
4559 * ENOSPC for updating the inode in the case of file write and mmap writes.
4561 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
4564 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4566 if (btrfs_root_readonly(root
))
4569 if (flags
& S_VERSION
)
4570 inode_inc_iversion(inode
);
4571 if (flags
& S_CTIME
)
4572 inode
->i_ctime
= *now
;
4573 if (flags
& S_MTIME
)
4574 inode
->i_mtime
= *now
;
4575 if (flags
& S_ATIME
)
4576 inode
->i_atime
= *now
;
4577 return btrfs_dirty_inode(inode
);
4581 * find the highest existing sequence number in a directory
4582 * and then set the in-memory index_cnt variable to reflect
4583 * free sequence numbers
4585 static int btrfs_set_inode_index_count(struct inode
*inode
)
4587 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4588 struct btrfs_key key
, found_key
;
4589 struct btrfs_path
*path
;
4590 struct extent_buffer
*leaf
;
4593 key
.objectid
= btrfs_ino(inode
);
4594 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
4595 key
.offset
= (u64
)-1;
4597 path
= btrfs_alloc_path();
4601 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4604 /* FIXME: we should be able to handle this */
4610 * MAGIC NUMBER EXPLANATION:
4611 * since we search a directory based on f_pos we have to start at 2
4612 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4613 * else has to start at 2
4615 if (path
->slots
[0] == 0) {
4616 BTRFS_I(inode
)->index_cnt
= 2;
4622 leaf
= path
->nodes
[0];
4623 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4625 if (found_key
.objectid
!= btrfs_ino(inode
) ||
4626 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
4627 BTRFS_I(inode
)->index_cnt
= 2;
4631 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
4633 btrfs_free_path(path
);
4638 * helper to find a free sequence number in a given directory. This current
4639 * code is very simple, later versions will do smarter things in the btree
4641 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
4645 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
4646 ret
= btrfs_inode_delayed_dir_index_count(dir
);
4648 ret
= btrfs_set_inode_index_count(dir
);
4654 *index
= BTRFS_I(dir
)->index_cnt
;
4655 BTRFS_I(dir
)->index_cnt
++;
4660 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
4661 struct btrfs_root
*root
,
4663 const char *name
, int name_len
,
4664 u64 ref_objectid
, u64 objectid
,
4665 umode_t mode
, u64
*index
)
4667 struct inode
*inode
;
4668 struct btrfs_inode_item
*inode_item
;
4669 struct btrfs_key
*location
;
4670 struct btrfs_path
*path
;
4671 struct btrfs_inode_ref
*ref
;
4672 struct btrfs_key key
[2];
4678 path
= btrfs_alloc_path();
4680 return ERR_PTR(-ENOMEM
);
4682 inode
= new_inode(root
->fs_info
->sb
);
4684 btrfs_free_path(path
);
4685 return ERR_PTR(-ENOMEM
);
4689 * we have to initialize this early, so we can reclaim the inode
4690 * number if we fail afterwards in this function.
4692 inode
->i_ino
= objectid
;
4695 trace_btrfs_inode_request(dir
);
4697 ret
= btrfs_set_inode_index(dir
, index
);
4699 btrfs_free_path(path
);
4701 return ERR_PTR(ret
);
4705 * index_cnt is ignored for everything but a dir,
4706 * btrfs_get_inode_index_count has an explanation for the magic
4709 BTRFS_I(inode
)->index_cnt
= 2;
4710 BTRFS_I(inode
)->root
= root
;
4711 BTRFS_I(inode
)->generation
= trans
->transid
;
4712 inode
->i_generation
= BTRFS_I(inode
)->generation
;
4715 * We could have gotten an inode number from somebody who was fsynced
4716 * and then removed in this same transaction, so let's just set full
4717 * sync since it will be a full sync anyway and this will blow away the
4718 * old info in the log.
4720 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
4727 key
[0].objectid
= objectid
;
4728 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
4732 * Start new inodes with an inode_ref. This is slightly more
4733 * efficient for small numbers of hard links since they will
4734 * be packed into one item. Extended refs will kick in if we
4735 * add more hard links than can fit in the ref item.
4737 key
[1].objectid
= objectid
;
4738 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
4739 key
[1].offset
= ref_objectid
;
4741 sizes
[0] = sizeof(struct btrfs_inode_item
);
4742 sizes
[1] = name_len
+ sizeof(*ref
);
4744 path
->leave_spinning
= 1;
4745 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
4749 inode_init_owner(inode
, dir
, mode
);
4750 inode_set_bytes(inode
, 0);
4751 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
4752 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
4753 struct btrfs_inode_item
);
4754 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
4755 sizeof(*inode_item
));
4756 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
4758 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
4759 struct btrfs_inode_ref
);
4760 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
4761 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
4762 ptr
= (unsigned long)(ref
+ 1);
4763 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
4765 btrfs_mark_buffer_dirty(path
->nodes
[0]);
4766 btrfs_free_path(path
);
4768 location
= &BTRFS_I(inode
)->location
;
4769 location
->objectid
= objectid
;
4770 location
->offset
= 0;
4771 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
4773 btrfs_inherit_iflags(inode
, dir
);
4775 if (S_ISREG(mode
)) {
4776 if (btrfs_test_opt(root
, NODATASUM
))
4777 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
4778 if (btrfs_test_opt(root
, NODATACOW
) ||
4779 (BTRFS_I(dir
)->flags
& BTRFS_INODE_NODATACOW
))
4780 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
;
4783 insert_inode_hash(inode
);
4784 inode_tree_add(inode
);
4786 trace_btrfs_inode_new(inode
);
4787 btrfs_set_inode_last_trans(trans
, inode
);
4789 btrfs_update_root_times(trans
, root
);
4794 BTRFS_I(dir
)->index_cnt
--;
4795 btrfs_free_path(path
);
4797 return ERR_PTR(ret
);
4800 static inline u8
btrfs_inode_type(struct inode
*inode
)
4802 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
4806 * utility function to add 'inode' into 'parent_inode' with
4807 * a give name and a given sequence number.
4808 * if 'add_backref' is true, also insert a backref from the
4809 * inode to the parent directory.
4811 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
4812 struct inode
*parent_inode
, struct inode
*inode
,
4813 const char *name
, int name_len
, int add_backref
, u64 index
)
4816 struct btrfs_key key
;
4817 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
4818 u64 ino
= btrfs_ino(inode
);
4819 u64 parent_ino
= btrfs_ino(parent_inode
);
4821 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4822 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
4825 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
4829 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4830 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
4831 key
.objectid
, root
->root_key
.objectid
,
4832 parent_ino
, index
, name
, name_len
);
4833 } else if (add_backref
) {
4834 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
4838 /* Nothing to clean up yet */
4842 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
4844 btrfs_inode_type(inode
), index
);
4848 btrfs_abort_transaction(trans
, root
, ret
);
4852 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
4854 inode_inc_iversion(parent_inode
);
4855 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
4856 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
4858 btrfs_abort_transaction(trans
, root
, ret
);
4862 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
4865 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
4866 key
.objectid
, root
->root_key
.objectid
,
4867 parent_ino
, &local_index
, name
, name_len
);
4869 } else if (add_backref
) {
4873 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
4874 ino
, parent_ino
, &local_index
);
4879 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
4880 struct inode
*dir
, struct dentry
*dentry
,
4881 struct inode
*inode
, int backref
, u64 index
)
4883 int err
= btrfs_add_link(trans
, dir
, inode
,
4884 dentry
->d_name
.name
, dentry
->d_name
.len
,
4891 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
4892 umode_t mode
, dev_t rdev
)
4894 struct btrfs_trans_handle
*trans
;
4895 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4896 struct inode
*inode
= NULL
;
4900 unsigned long nr
= 0;
4903 if (!new_valid_dev(rdev
))
4907 * 2 for inode item and ref
4909 * 1 for xattr if selinux is on
4911 trans
= btrfs_start_transaction(root
, 5);
4913 return PTR_ERR(trans
);
4915 err
= btrfs_find_free_ino(root
, &objectid
);
4919 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4920 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4922 if (IS_ERR(inode
)) {
4923 err
= PTR_ERR(inode
);
4927 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
4934 * If the active LSM wants to access the inode during
4935 * d_instantiate it needs these. Smack checks to see
4936 * if the filesystem supports xattrs by looking at the
4940 inode
->i_op
= &btrfs_special_inode_operations
;
4941 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
4945 init_special_inode(inode
, inode
->i_mode
, rdev
);
4946 btrfs_update_inode(trans
, root
, inode
);
4947 d_instantiate(dentry
, inode
);
4950 nr
= trans
->blocks_used
;
4951 btrfs_end_transaction(trans
, root
);
4952 btrfs_btree_balance_dirty(root
, nr
);
4954 inode_dec_link_count(inode
);
4960 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
4961 umode_t mode
, bool excl
)
4963 struct btrfs_trans_handle
*trans
;
4964 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4965 struct inode
*inode
= NULL
;
4968 unsigned long nr
= 0;
4973 * 2 for inode item and ref
4975 * 1 for xattr if selinux is on
4977 trans
= btrfs_start_transaction(root
, 5);
4979 return PTR_ERR(trans
);
4981 err
= btrfs_find_free_ino(root
, &objectid
);
4985 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
4986 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
4988 if (IS_ERR(inode
)) {
4989 err
= PTR_ERR(inode
);
4993 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5000 * If the active LSM wants to access the inode during
5001 * d_instantiate it needs these. Smack checks to see
5002 * if the filesystem supports xattrs by looking at the
5005 inode
->i_fop
= &btrfs_file_operations
;
5006 inode
->i_op
= &btrfs_file_inode_operations
;
5008 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5012 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5013 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5014 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5015 d_instantiate(dentry
, inode
);
5018 nr
= trans
->blocks_used
;
5019 btrfs_end_transaction(trans
, root
);
5021 inode_dec_link_count(inode
);
5024 btrfs_btree_balance_dirty(root
, nr
);
5028 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
5029 struct dentry
*dentry
)
5031 struct btrfs_trans_handle
*trans
;
5032 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5033 struct inode
*inode
= old_dentry
->d_inode
;
5035 unsigned long nr
= 0;
5039 /* do not allow sys_link's with other subvols of the same device */
5040 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
5043 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
5046 err
= btrfs_set_inode_index(dir
, &index
);
5051 * 2 items for inode and inode ref
5052 * 2 items for dir items
5053 * 1 item for parent inode
5055 trans
= btrfs_start_transaction(root
, 5);
5056 if (IS_ERR(trans
)) {
5057 err
= PTR_ERR(trans
);
5061 btrfs_inc_nlink(inode
);
5062 inode_inc_iversion(inode
);
5063 inode
->i_ctime
= CURRENT_TIME
;
5066 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5071 struct dentry
*parent
= dentry
->d_parent
;
5072 err
= btrfs_update_inode(trans
, root
, inode
);
5075 d_instantiate(dentry
, inode
);
5076 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5079 nr
= trans
->blocks_used
;
5080 btrfs_end_transaction(trans
, root
);
5083 inode_dec_link_count(inode
);
5086 btrfs_btree_balance_dirty(root
, nr
);
5090 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5092 struct inode
*inode
= NULL
;
5093 struct btrfs_trans_handle
*trans
;
5094 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5096 int drop_on_err
= 0;
5099 unsigned long nr
= 1;
5102 * 2 items for inode and ref
5103 * 2 items for dir items
5104 * 1 for xattr if selinux is on
5106 trans
= btrfs_start_transaction(root
, 5);
5108 return PTR_ERR(trans
);
5110 err
= btrfs_find_free_ino(root
, &objectid
);
5114 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5115 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5116 S_IFDIR
| mode
, &index
);
5117 if (IS_ERR(inode
)) {
5118 err
= PTR_ERR(inode
);
5124 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5128 inode
->i_op
= &btrfs_dir_inode_operations
;
5129 inode
->i_fop
= &btrfs_dir_file_operations
;
5131 btrfs_i_size_write(inode
, 0);
5132 err
= btrfs_update_inode(trans
, root
, inode
);
5136 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5137 dentry
->d_name
.len
, 0, index
);
5141 d_instantiate(dentry
, inode
);
5145 nr
= trans
->blocks_used
;
5146 btrfs_end_transaction(trans
, root
);
5149 btrfs_btree_balance_dirty(root
, nr
);
5153 /* helper for btfs_get_extent. Given an existing extent in the tree,
5154 * and an extent that you want to insert, deal with overlap and insert
5155 * the new extent into the tree.
5157 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
5158 struct extent_map
*existing
,
5159 struct extent_map
*em
,
5160 u64 map_start
, u64 map_len
)
5164 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
5165 start_diff
= map_start
- em
->start
;
5166 em
->start
= map_start
;
5168 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
5169 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
5170 em
->block_start
+= start_diff
;
5171 em
->block_len
-= start_diff
;
5173 return add_extent_mapping(em_tree
, em
);
5176 static noinline
int uncompress_inline(struct btrfs_path
*path
,
5177 struct inode
*inode
, struct page
*page
,
5178 size_t pg_offset
, u64 extent_offset
,
5179 struct btrfs_file_extent_item
*item
)
5182 struct extent_buffer
*leaf
= path
->nodes
[0];
5185 unsigned long inline_size
;
5189 WARN_ON(pg_offset
!= 0);
5190 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5191 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
5192 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
5193 btrfs_item_nr(leaf
, path
->slots
[0]));
5194 tmp
= kmalloc(inline_size
, GFP_NOFS
);
5197 ptr
= btrfs_file_extent_inline_start(item
);
5199 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
5201 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
5202 ret
= btrfs_decompress(compress_type
, tmp
, page
,
5203 extent_offset
, inline_size
, max_size
);
5205 char *kaddr
= kmap_atomic(page
);
5206 unsigned long copy_size
= min_t(u64
,
5207 PAGE_CACHE_SIZE
- pg_offset
,
5208 max_size
- extent_offset
);
5209 memset(kaddr
+ pg_offset
, 0, copy_size
);
5210 kunmap_atomic(kaddr
);
5217 * a bit scary, this does extent mapping from logical file offset to the disk.
5218 * the ugly parts come from merging extents from the disk with the in-ram
5219 * representation. This gets more complex because of the data=ordered code,
5220 * where the in-ram extents might be locked pending data=ordered completion.
5222 * This also copies inline extents directly into the page.
5225 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
5226 size_t pg_offset
, u64 start
, u64 len
,
5232 u64 extent_start
= 0;
5234 u64 objectid
= btrfs_ino(inode
);
5236 struct btrfs_path
*path
= NULL
;
5237 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5238 struct btrfs_file_extent_item
*item
;
5239 struct extent_buffer
*leaf
;
5240 struct btrfs_key found_key
;
5241 struct extent_map
*em
= NULL
;
5242 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5243 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
5244 struct btrfs_trans_handle
*trans
= NULL
;
5248 read_lock(&em_tree
->lock
);
5249 em
= lookup_extent_mapping(em_tree
, start
, len
);
5251 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5252 read_unlock(&em_tree
->lock
);
5255 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
5256 free_extent_map(em
);
5257 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
5258 free_extent_map(em
);
5262 em
= alloc_extent_map();
5267 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5268 em
->start
= EXTENT_MAP_HOLE
;
5269 em
->orig_start
= EXTENT_MAP_HOLE
;
5271 em
->block_len
= (u64
)-1;
5274 path
= btrfs_alloc_path();
5280 * Chances are we'll be called again, so go ahead and do
5286 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
5287 objectid
, start
, trans
!= NULL
);
5294 if (path
->slots
[0] == 0)
5299 leaf
= path
->nodes
[0];
5300 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5301 struct btrfs_file_extent_item
);
5302 /* are we inside the extent that was found? */
5303 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5304 found_type
= btrfs_key_type(&found_key
);
5305 if (found_key
.objectid
!= objectid
||
5306 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
5310 found_type
= btrfs_file_extent_type(leaf
, item
);
5311 extent_start
= found_key
.offset
;
5312 compress_type
= btrfs_file_extent_compression(leaf
, item
);
5313 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5314 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5315 extent_end
= extent_start
+
5316 btrfs_file_extent_num_bytes(leaf
, item
);
5317 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5319 size
= btrfs_file_extent_inline_len(leaf
, item
);
5320 extent_end
= (extent_start
+ size
+ root
->sectorsize
- 1) &
5321 ~((u64
)root
->sectorsize
- 1);
5324 if (start
>= extent_end
) {
5326 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
5327 ret
= btrfs_next_leaf(root
, path
);
5334 leaf
= path
->nodes
[0];
5336 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5337 if (found_key
.objectid
!= objectid
||
5338 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
5340 if (start
+ len
<= found_key
.offset
)
5343 em
->len
= found_key
.offset
- start
;
5347 if (found_type
== BTRFS_FILE_EXTENT_REG
||
5348 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
5349 em
->start
= extent_start
;
5350 em
->len
= extent_end
- extent_start
;
5351 em
->orig_start
= extent_start
-
5352 btrfs_file_extent_offset(leaf
, item
);
5353 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
5355 em
->block_start
= EXTENT_MAP_HOLE
;
5358 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
5359 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5360 em
->compress_type
= compress_type
;
5361 em
->block_start
= bytenr
;
5362 em
->block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
5365 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
5366 em
->block_start
= bytenr
;
5367 em
->block_len
= em
->len
;
5368 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
5369 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5372 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
5376 size_t extent_offset
;
5379 em
->block_start
= EXTENT_MAP_INLINE
;
5380 if (!page
|| create
) {
5381 em
->start
= extent_start
;
5382 em
->len
= extent_end
- extent_start
;
5386 size
= btrfs_file_extent_inline_len(leaf
, item
);
5387 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
5388 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
5389 size
- extent_offset
);
5390 em
->start
= extent_start
+ extent_offset
;
5391 em
->len
= (copy_size
+ root
->sectorsize
- 1) &
5392 ~((u64
)root
->sectorsize
- 1);
5393 em
->orig_start
= EXTENT_MAP_INLINE
;
5394 if (compress_type
) {
5395 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
5396 em
->compress_type
= compress_type
;
5398 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
5399 if (create
== 0 && !PageUptodate(page
)) {
5400 if (btrfs_file_extent_compression(leaf
, item
) !=
5401 BTRFS_COMPRESS_NONE
) {
5402 ret
= uncompress_inline(path
, inode
, page
,
5404 extent_offset
, item
);
5405 BUG_ON(ret
); /* -ENOMEM */
5408 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5410 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
5411 memset(map
+ pg_offset
+ copy_size
, 0,
5412 PAGE_CACHE_SIZE
- pg_offset
-
5417 flush_dcache_page(page
);
5418 } else if (create
&& PageUptodate(page
)) {
5422 free_extent_map(em
);
5425 btrfs_release_path(path
);
5426 trans
= btrfs_join_transaction(root
);
5429 return ERR_CAST(trans
);
5433 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
5436 btrfs_mark_buffer_dirty(leaf
);
5438 set_extent_uptodate(io_tree
, em
->start
,
5439 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
5442 printk(KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
5449 em
->block_start
= EXTENT_MAP_HOLE
;
5450 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
5452 btrfs_release_path(path
);
5453 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
5454 printk(KERN_ERR
"Btrfs: bad extent! em: [%llu %llu] passed "
5455 "[%llu %llu]\n", (unsigned long long)em
->start
,
5456 (unsigned long long)em
->len
,
5457 (unsigned long long)start
,
5458 (unsigned long long)len
);
5464 write_lock(&em_tree
->lock
);
5465 ret
= add_extent_mapping(em_tree
, em
);
5466 /* it is possible that someone inserted the extent into the tree
5467 * while we had the lock dropped. It is also possible that
5468 * an overlapping map exists in the tree
5470 if (ret
== -EEXIST
) {
5471 struct extent_map
*existing
;
5475 existing
= lookup_extent_mapping(em_tree
, start
, len
);
5476 if (existing
&& (existing
->start
> start
||
5477 existing
->start
+ existing
->len
<= start
)) {
5478 free_extent_map(existing
);
5482 existing
= lookup_extent_mapping(em_tree
, em
->start
,
5485 err
= merge_extent_mapping(em_tree
, existing
,
5488 free_extent_map(existing
);
5490 free_extent_map(em
);
5495 free_extent_map(em
);
5499 free_extent_map(em
);
5504 write_unlock(&em_tree
->lock
);
5508 trace_btrfs_get_extent(root
, em
);
5511 btrfs_free_path(path
);
5513 ret
= btrfs_end_transaction(trans
, root
);
5518 free_extent_map(em
);
5519 return ERR_PTR(err
);
5521 BUG_ON(!em
); /* Error is always set */
5525 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
5526 size_t pg_offset
, u64 start
, u64 len
,
5529 struct extent_map
*em
;
5530 struct extent_map
*hole_em
= NULL
;
5531 u64 range_start
= start
;
5537 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
5542 * if our em maps to a hole, there might
5543 * actually be delalloc bytes behind it
5545 if (em
->block_start
!= EXTENT_MAP_HOLE
)
5551 /* check to see if we've wrapped (len == -1 or similar) */
5560 /* ok, we didn't find anything, lets look for delalloc */
5561 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
5562 end
, len
, EXTENT_DELALLOC
, 1);
5563 found_end
= range_start
+ found
;
5564 if (found_end
< range_start
)
5565 found_end
= (u64
)-1;
5568 * we didn't find anything useful, return
5569 * the original results from get_extent()
5571 if (range_start
> end
|| found_end
<= start
) {
5577 /* adjust the range_start to make sure it doesn't
5578 * go backwards from the start they passed in
5580 range_start
= max(start
,range_start
);
5581 found
= found_end
- range_start
;
5584 u64 hole_start
= start
;
5587 em
= alloc_extent_map();
5593 * when btrfs_get_extent can't find anything it
5594 * returns one huge hole
5596 * make sure what it found really fits our range, and
5597 * adjust to make sure it is based on the start from
5601 u64 calc_end
= extent_map_end(hole_em
);
5603 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
5604 free_extent_map(hole_em
);
5607 hole_start
= max(hole_em
->start
, start
);
5608 hole_len
= calc_end
- hole_start
;
5612 if (hole_em
&& range_start
> hole_start
) {
5613 /* our hole starts before our delalloc, so we
5614 * have to return just the parts of the hole
5615 * that go until the delalloc starts
5617 em
->len
= min(hole_len
,
5618 range_start
- hole_start
);
5619 em
->start
= hole_start
;
5620 em
->orig_start
= hole_start
;
5622 * don't adjust block start at all,
5623 * it is fixed at EXTENT_MAP_HOLE
5625 em
->block_start
= hole_em
->block_start
;
5626 em
->block_len
= hole_len
;
5628 em
->start
= range_start
;
5630 em
->orig_start
= range_start
;
5631 em
->block_start
= EXTENT_MAP_DELALLOC
;
5632 em
->block_len
= found
;
5634 } else if (hole_em
) {
5639 free_extent_map(hole_em
);
5641 free_extent_map(em
);
5642 return ERR_PTR(err
);
5647 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
5648 struct extent_map
*em
,
5651 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5652 struct btrfs_trans_handle
*trans
;
5653 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
5654 struct btrfs_key ins
;
5657 bool insert
= false;
5660 * Ok if the extent map we looked up is a hole and is for the exact
5661 * range we want, there is no reason to allocate a new one, however if
5662 * it is not right then we need to free this one and drop the cache for
5665 if (em
->block_start
!= EXTENT_MAP_HOLE
|| em
->start
!= start
||
5667 free_extent_map(em
);
5670 btrfs_drop_extent_cache(inode
, start
, start
+ len
- 1, 0);
5673 trans
= btrfs_join_transaction(root
);
5675 return ERR_CAST(trans
);
5677 if (start
<= BTRFS_I(inode
)->disk_i_size
&& len
< 64 * 1024)
5678 btrfs_add_inode_defrag(trans
, inode
);
5680 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
5682 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
5683 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
5684 alloc_hint
, &ins
, 1);
5691 em
= alloc_extent_map();
5693 em
= ERR_PTR(-ENOMEM
);
5699 em
->orig_start
= em
->start
;
5700 em
->len
= ins
.offset
;
5702 em
->block_start
= ins
.objectid
;
5703 em
->block_len
= ins
.offset
;
5704 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5707 * We need to do this because if we're using the original em we searched
5708 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5711 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5714 write_lock(&em_tree
->lock
);
5715 ret
= add_extent_mapping(em_tree
, em
);
5716 write_unlock(&em_tree
->lock
);
5719 btrfs_drop_extent_cache(inode
, start
, start
+ em
->len
- 1, 0);
5722 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
5723 ins
.offset
, ins
.offset
, 0);
5725 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
5729 btrfs_end_transaction(trans
, root
);
5734 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5735 * block must be cow'd
5737 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
5738 struct inode
*inode
, u64 offset
, u64 len
)
5740 struct btrfs_path
*path
;
5742 struct extent_buffer
*leaf
;
5743 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5744 struct btrfs_file_extent_item
*fi
;
5745 struct btrfs_key key
;
5753 path
= btrfs_alloc_path();
5757 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
5762 slot
= path
->slots
[0];
5765 /* can't find the item, must cow */
5772 leaf
= path
->nodes
[0];
5773 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
5774 if (key
.objectid
!= btrfs_ino(inode
) ||
5775 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
5776 /* not our file or wrong item type, must cow */
5780 if (key
.offset
> offset
) {
5781 /* Wrong offset, must cow */
5785 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
5786 found_type
= btrfs_file_extent_type(leaf
, fi
);
5787 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
5788 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
5789 /* not a regular extent, must cow */
5792 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
5793 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
5795 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
5796 if (extent_end
< offset
+ len
) {
5797 /* extent doesn't include our full range, must cow */
5801 if (btrfs_extent_readonly(root
, disk_bytenr
))
5805 * look for other files referencing this extent, if we
5806 * find any we must cow
5808 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
5809 key
.offset
- backref_offset
, disk_bytenr
))
5813 * adjust disk_bytenr and num_bytes to cover just the bytes
5814 * in this extent we are about to write. If there
5815 * are any csums in that range we have to cow in order
5816 * to keep the csums correct
5818 disk_bytenr
+= backref_offset
;
5819 disk_bytenr
+= offset
- key
.offset
;
5820 num_bytes
= min(offset
+ len
, extent_end
) - offset
;
5821 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
5824 * all of the above have passed, it is safe to overwrite this extent
5829 btrfs_free_path(path
);
5833 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
5834 struct extent_state
**cached_state
, int writing
)
5836 struct btrfs_ordered_extent
*ordered
;
5840 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
5843 * We're concerned with the entire range that we're going to be
5844 * doing DIO to, so we need to make sure theres no ordered
5845 * extents in this range.
5847 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
5848 lockend
- lockstart
+ 1);
5851 * We need to make sure there are no buffered pages in this
5852 * range either, we could have raced between the invalidate in
5853 * generic_file_direct_write and locking the extent. The
5854 * invalidate needs to happen so that reads after a write do not
5857 if (!ordered
&& (!writing
||
5858 !test_range_bit(&BTRFS_I(inode
)->io_tree
,
5859 lockstart
, lockend
, EXTENT_UPTODATE
, 0,
5863 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
5864 cached_state
, GFP_NOFS
);
5867 btrfs_start_ordered_extent(inode
, ordered
, 1);
5868 btrfs_put_ordered_extent(ordered
);
5870 /* Screw you mmap */
5871 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
5878 * If we found a page that couldn't be invalidated just
5879 * fall back to buffered.
5881 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
5882 lockstart
>> PAGE_CACHE_SHIFT
,
5883 lockend
>> PAGE_CACHE_SHIFT
);
5894 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
5895 u64 len
, u64 orig_start
,
5896 u64 block_start
, u64 block_len
,
5899 struct extent_map_tree
*em_tree
;
5900 struct extent_map
*em
;
5901 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5904 em_tree
= &BTRFS_I(inode
)->extent_tree
;
5905 em
= alloc_extent_map();
5907 return ERR_PTR(-ENOMEM
);
5910 em
->orig_start
= orig_start
;
5912 em
->block_len
= block_len
;
5913 em
->block_start
= block_start
;
5914 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
5915 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
5916 if (type
== BTRFS_ORDERED_PREALLOC
)
5917 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
5920 btrfs_drop_extent_cache(inode
, em
->start
,
5921 em
->start
+ em
->len
- 1, 0);
5922 write_lock(&em_tree
->lock
);
5923 ret
= add_extent_mapping(em_tree
, em
);
5924 write_unlock(&em_tree
->lock
);
5925 } while (ret
== -EEXIST
);
5928 free_extent_map(em
);
5929 return ERR_PTR(ret
);
5936 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
5937 struct buffer_head
*bh_result
, int create
)
5939 struct extent_map
*em
;
5940 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5941 struct extent_state
*cached_state
= NULL
;
5942 u64 start
= iblock
<< inode
->i_blkbits
;
5943 u64 lockstart
, lockend
;
5944 u64 len
= bh_result
->b_size
;
5945 struct btrfs_trans_handle
*trans
;
5946 int unlock_bits
= EXTENT_LOCKED
;
5950 ret
= btrfs_delalloc_reserve_space(inode
, len
);
5953 unlock_bits
|= EXTENT_DELALLOC
| EXTENT_DIRTY
;
5955 len
= min_t(u64
, len
, root
->sectorsize
);
5959 lockend
= start
+ len
- 1;
5962 * If this errors out it's because we couldn't invalidate pagecache for
5963 * this range and we need to fallback to buffered.
5965 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
5969 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
5970 lockend
, EXTENT_DELALLOC
, NULL
,
5971 &cached_state
, GFP_NOFS
);
5976 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
5983 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5984 * io. INLINE is special, and we could probably kludge it in here, but
5985 * it's still buffered so for safety lets just fall back to the generic
5988 * For COMPRESSED we _have_ to read the entire extent in so we can
5989 * decompress it, so there will be buffering required no matter what we
5990 * do, so go ahead and fallback to buffered.
5992 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5993 * to buffered IO. Don't blame me, this is the price we pay for using
5996 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
5997 em
->block_start
== EXTENT_MAP_INLINE
) {
5998 free_extent_map(em
);
6003 /* Just a good old fashioned hole, return */
6004 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
6005 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
6006 free_extent_map(em
);
6012 * We don't allocate a new extent in the following cases
6014 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6016 * 2) The extent is marked as PREALLOC. We're good to go here and can
6017 * just use the extent.
6021 len
= min(len
, em
->len
- (start
- em
->start
));
6022 lockstart
= start
+ len
;
6026 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
6027 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
6028 em
->block_start
!= EXTENT_MAP_HOLE
)) {
6033 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6034 type
= BTRFS_ORDERED_PREALLOC
;
6036 type
= BTRFS_ORDERED_NOCOW
;
6037 len
= min(len
, em
->len
- (start
- em
->start
));
6038 block_start
= em
->block_start
+ (start
- em
->start
);
6041 * we're not going to log anything, but we do need
6042 * to make sure the current transaction stays open
6043 * while we look for nocow cross refs
6045 trans
= btrfs_join_transaction(root
);
6049 if (can_nocow_odirect(trans
, inode
, start
, len
) == 1) {
6050 u64 orig_start
= em
->start
;
6052 if (type
== BTRFS_ORDERED_PREALLOC
) {
6053 free_extent_map(em
);
6054 em
= create_pinned_em(inode
, start
, len
,
6056 block_start
, len
, type
);
6058 btrfs_end_transaction(trans
, root
);
6063 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
6064 block_start
, len
, len
, type
);
6065 btrfs_end_transaction(trans
, root
);
6067 free_extent_map(em
);
6072 btrfs_end_transaction(trans
, root
);
6076 * this will cow the extent, reset the len in case we changed
6079 len
= bh_result
->b_size
;
6080 em
= btrfs_new_extent_direct(inode
, em
, start
, len
);
6085 len
= min(len
, em
->len
- (start
- em
->start
));
6087 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
6089 bh_result
->b_size
= len
;
6090 bh_result
->b_bdev
= em
->bdev
;
6091 set_buffer_mapped(bh_result
);
6093 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6094 set_buffer_new(bh_result
);
6097 * Need to update the i_size under the extent lock so buffered
6098 * readers will get the updated i_size when we unlock.
6100 if (start
+ len
> i_size_read(inode
))
6101 i_size_write(inode
, start
+ len
);
6105 * In the case of write we need to clear and unlock the entire range,
6106 * in the case of read we need to unlock only the end area that we
6107 * aren't using if there is any left over space.
6109 if (lockstart
< lockend
) {
6110 if (create
&& len
< lockend
- lockstart
) {
6111 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6112 lockstart
+ len
- 1,
6113 unlock_bits
| EXTENT_DEFRAG
, 1, 0,
6114 &cached_state
, GFP_NOFS
);
6116 * Beside unlock, we also need to cleanup reserved space
6117 * for the left range by attaching EXTENT_DO_ACCOUNTING.
6119 clear_extent_bit(&BTRFS_I(inode
)->io_tree
,
6120 lockstart
+ len
, lockend
,
6121 unlock_bits
| EXTENT_DO_ACCOUNTING
|
6122 EXTENT_DEFRAG
, 1, 0, NULL
, GFP_NOFS
);
6124 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6125 lockend
, unlock_bits
, 1, 0,
6126 &cached_state
, GFP_NOFS
);
6129 free_extent_state(cached_state
);
6132 free_extent_map(em
);
6138 unlock_bits
|= EXTENT_DO_ACCOUNTING
;
6140 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6141 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
6145 struct btrfs_dio_private
{
6146 struct inode
*inode
;
6152 /* number of bios pending for this dio */
6153 atomic_t pending_bios
;
6158 struct bio
*orig_bio
;
6161 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
6163 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6164 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
6165 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6166 struct inode
*inode
= dip
->inode
;
6167 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6170 start
= dip
->logical_offset
;
6172 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
6173 struct page
*page
= bvec
->bv_page
;
6176 u64
private = ~(u32
)0;
6177 unsigned long flags
;
6179 if (get_state_private(&BTRFS_I(inode
)->io_tree
,
6182 local_irq_save(flags
);
6183 kaddr
= kmap_atomic(page
);
6184 csum
= btrfs_csum_data(root
, kaddr
+ bvec
->bv_offset
,
6185 csum
, bvec
->bv_len
);
6186 btrfs_csum_final(csum
, (char *)&csum
);
6187 kunmap_atomic(kaddr
);
6188 local_irq_restore(flags
);
6190 flush_dcache_page(bvec
->bv_page
);
6191 if (csum
!= private) {
6193 printk(KERN_ERR
"btrfs csum failed ino %llu off"
6194 " %llu csum %u private %u\n",
6195 (unsigned long long)btrfs_ino(inode
),
6196 (unsigned long long)start
,
6197 csum
, (unsigned)private);
6202 start
+= bvec
->bv_len
;
6204 } while (bvec
<= bvec_end
);
6206 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
6207 dip
->logical_offset
+ dip
->bytes
- 1);
6208 bio
->bi_private
= dip
->private;
6212 /* If we had a csum failure make sure to clear the uptodate flag */
6214 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6215 dio_end_io(bio
, err
);
6218 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
6220 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6221 struct inode
*inode
= dip
->inode
;
6222 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6223 struct btrfs_ordered_extent
*ordered
= NULL
;
6224 u64 ordered_offset
= dip
->logical_offset
;
6225 u64 ordered_bytes
= dip
->bytes
;
6231 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
6233 ordered_bytes
, !err
);
6237 ordered
->work
.func
= finish_ordered_fn
;
6238 ordered
->work
.flags
= 0;
6239 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
6243 * our bio might span multiple ordered extents. If we haven't
6244 * completed the accounting for the whole dio, go back and try again
6246 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
6247 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
6253 bio
->bi_private
= dip
->private;
6257 /* If we had an error make sure to clear the uptodate flag */
6259 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6260 dio_end_io(bio
, err
);
6263 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
6264 struct bio
*bio
, int mirror_num
,
6265 unsigned long bio_flags
, u64 offset
)
6268 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6269 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
6270 BUG_ON(ret
); /* -ENOMEM */
6274 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
6276 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6279 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
6280 "sector %#Lx len %u err no %d\n",
6281 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
6282 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
6286 * before atomic variable goto zero, we must make sure
6287 * dip->errors is perceived to be set.
6289 smp_mb__before_atomic_dec();
6292 /* if there are more bios still pending for this dio, just exit */
6293 if (!atomic_dec_and_test(&dip
->pending_bios
))
6297 bio_io_error(dip
->orig_bio
);
6299 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
6300 bio_endio(dip
->orig_bio
, 0);
6306 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
6307 u64 first_sector
, gfp_t gfp_flags
)
6309 int nr_vecs
= bio_get_nr_vecs(bdev
);
6310 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
6313 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
6314 int rw
, u64 file_offset
, int skip_sum
,
6317 int write
= rw
& REQ_WRITE
;
6318 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6324 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
6332 if (write
&& async_submit
) {
6333 ret
= btrfs_wq_submit_bio(root
->fs_info
,
6334 inode
, rw
, bio
, 0, 0,
6336 __btrfs_submit_bio_start_direct_io
,
6337 __btrfs_submit_bio_done
);
6341 * If we aren't doing async submit, calculate the csum of the
6344 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
6347 } else if (!skip_sum
) {
6348 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
, file_offset
);
6354 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
6360 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
6363 struct inode
*inode
= dip
->inode
;
6364 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6365 struct btrfs_mapping_tree
*map_tree
= &root
->fs_info
->mapping_tree
;
6367 struct bio
*orig_bio
= dip
->orig_bio
;
6368 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
6369 u64 start_sector
= orig_bio
->bi_sector
;
6370 u64 file_offset
= dip
->logical_offset
;
6375 int async_submit
= 0;
6377 map_length
= orig_bio
->bi_size
;
6378 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6379 &map_length
, NULL
, 0);
6385 if (map_length
>= orig_bio
->bi_size
) {
6391 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
6394 bio
->bi_private
= dip
;
6395 bio
->bi_end_io
= btrfs_end_dio_bio
;
6396 atomic_inc(&dip
->pending_bios
);
6398 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
6399 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
6400 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
6401 bvec
->bv_offset
) < bvec
->bv_len
)) {
6403 * inc the count before we submit the bio so
6404 * we know the end IO handler won't happen before
6405 * we inc the count. Otherwise, the dip might get freed
6406 * before we're done setting it up
6408 atomic_inc(&dip
->pending_bios
);
6409 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
6410 file_offset
, skip_sum
,
6414 atomic_dec(&dip
->pending_bios
);
6418 start_sector
+= submit_len
>> 9;
6419 file_offset
+= submit_len
;
6424 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
6425 start_sector
, GFP_NOFS
);
6428 bio
->bi_private
= dip
;
6429 bio
->bi_end_io
= btrfs_end_dio_bio
;
6431 map_length
= orig_bio
->bi_size
;
6432 ret
= btrfs_map_block(map_tree
, READ
, start_sector
<< 9,
6433 &map_length
, NULL
, 0);
6439 submit_len
+= bvec
->bv_len
;
6446 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
6455 * before atomic variable goto zero, we must
6456 * make sure dip->errors is perceived to be set.
6458 smp_mb__before_atomic_dec();
6459 if (atomic_dec_and_test(&dip
->pending_bios
))
6460 bio_io_error(dip
->orig_bio
);
6462 /* bio_end_io() will handle error, so we needn't return it */
6466 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
6469 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6470 struct btrfs_dio_private
*dip
;
6471 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6473 int write
= rw
& REQ_WRITE
;
6476 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
6478 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
6484 dip
->private = bio
->bi_private
;
6486 dip
->logical_offset
= file_offset
;
6490 dip
->bytes
+= bvec
->bv_len
;
6492 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
6494 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
6495 bio
->bi_private
= dip
;
6497 dip
->orig_bio
= bio
;
6498 atomic_set(&dip
->pending_bios
, 0);
6501 bio
->bi_end_io
= btrfs_endio_direct_write
;
6503 bio
->bi_end_io
= btrfs_endio_direct_read
;
6505 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
6510 * If this is a write, we need to clean up the reserved space and kill
6511 * the ordered extent.
6514 struct btrfs_ordered_extent
*ordered
;
6515 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
6516 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
6517 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
6518 btrfs_free_reserved_extent(root
, ordered
->start
,
6520 btrfs_put_ordered_extent(ordered
);
6521 btrfs_put_ordered_extent(ordered
);
6523 bio_endio(bio
, ret
);
6526 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
6527 const struct iovec
*iov
, loff_t offset
,
6528 unsigned long nr_segs
)
6534 unsigned blocksize_mask
= root
->sectorsize
- 1;
6535 ssize_t retval
= -EINVAL
;
6536 loff_t end
= offset
;
6538 if (offset
& blocksize_mask
)
6541 /* Check the memory alignment. Blocks cannot straddle pages */
6542 for (seg
= 0; seg
< nr_segs
; seg
++) {
6543 addr
= (unsigned long)iov
[seg
].iov_base
;
6544 size
= iov
[seg
].iov_len
;
6546 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
6549 /* If this is a write we don't need to check anymore */
6554 * Check to make sure we don't have duplicate iov_base's in this
6555 * iovec, if so return EINVAL, otherwise we'll get csum errors
6556 * when reading back.
6558 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
6559 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
6568 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
6569 const struct iovec
*iov
, loff_t offset
,
6570 unsigned long nr_segs
)
6572 struct file
*file
= iocb
->ki_filp
;
6573 struct inode
*inode
= file
->f_mapping
->host
;
6575 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
6579 return __blockdev_direct_IO(rw
, iocb
, inode
,
6580 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
6581 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
6582 btrfs_submit_direct
, 0);
6585 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
6586 __u64 start
, __u64 len
)
6588 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
6591 int btrfs_readpage(struct file
*file
, struct page
*page
)
6593 struct extent_io_tree
*tree
;
6594 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6595 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
6598 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
6600 struct extent_io_tree
*tree
;
6603 if (current
->flags
& PF_MEMALLOC
) {
6604 redirty_page_for_writepage(wbc
, page
);
6608 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6609 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
6612 int btrfs_writepages(struct address_space
*mapping
,
6613 struct writeback_control
*wbc
)
6615 struct extent_io_tree
*tree
;
6617 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6618 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
6622 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
6623 struct list_head
*pages
, unsigned nr_pages
)
6625 struct extent_io_tree
*tree
;
6626 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
6627 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
6630 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6632 struct extent_io_tree
*tree
;
6633 struct extent_map_tree
*map
;
6636 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
6637 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
6638 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
6640 ClearPagePrivate(page
);
6641 set_page_private(page
, 0);
6642 page_cache_release(page
);
6647 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
6649 if (PageWriteback(page
) || PageDirty(page
))
6651 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
6654 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
6656 struct inode
*inode
= page
->mapping
->host
;
6657 struct extent_io_tree
*tree
;
6658 struct btrfs_ordered_extent
*ordered
;
6659 struct extent_state
*cached_state
= NULL
;
6660 u64 page_start
= page_offset(page
);
6661 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6664 * we have the page locked, so new writeback can't start,
6665 * and the dirty bit won't be cleared while we are here.
6667 * Wait for IO on this page so that we can safely clear
6668 * the PagePrivate2 bit and do ordered accounting
6670 wait_on_page_writeback(page
);
6672 tree
= &BTRFS_I(inode
)->io_tree
;
6674 btrfs_releasepage(page
, GFP_NOFS
);
6677 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6678 ordered
= btrfs_lookup_ordered_extent(inode
,
6682 * IO on this page will never be started, so we need
6683 * to account for any ordered extents now
6685 clear_extent_bit(tree
, page_start
, page_end
,
6686 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6687 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
6688 EXTENT_DEFRAG
, 1, 0, &cached_state
, GFP_NOFS
);
6690 * whoever cleared the private bit is responsible
6691 * for the finish_ordered_io
6693 if (TestClearPagePrivate2(page
) &&
6694 btrfs_dec_test_ordered_pending(inode
, &ordered
, page_start
,
6695 PAGE_CACHE_SIZE
, 1)) {
6696 btrfs_finish_ordered_io(ordered
);
6698 btrfs_put_ordered_extent(ordered
);
6699 cached_state
= NULL
;
6700 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
6702 clear_extent_bit(tree
, page_start
, page_end
,
6703 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
6704 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
6705 &cached_state
, GFP_NOFS
);
6706 __btrfs_releasepage(page
, GFP_NOFS
);
6708 ClearPageChecked(page
);
6709 if (PagePrivate(page
)) {
6710 ClearPagePrivate(page
);
6711 set_page_private(page
, 0);
6712 page_cache_release(page
);
6717 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6718 * called from a page fault handler when a page is first dirtied. Hence we must
6719 * be careful to check for EOF conditions here. We set the page up correctly
6720 * for a written page which means we get ENOSPC checking when writing into
6721 * holes and correct delalloc and unwritten extent mapping on filesystems that
6722 * support these features.
6724 * We are not allowed to take the i_mutex here so we have to play games to
6725 * protect against truncate races as the page could now be beyond EOF. Because
6726 * vmtruncate() writes the inode size before removing pages, once we have the
6727 * page lock we can determine safely if the page is beyond EOF. If it is not
6728 * beyond EOF, then the page is guaranteed safe against truncation until we
6731 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
6733 struct page
*page
= vmf
->page
;
6734 struct inode
*inode
= fdentry(vma
->vm_file
)->d_inode
;
6735 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6736 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6737 struct btrfs_ordered_extent
*ordered
;
6738 struct extent_state
*cached_state
= NULL
;
6740 unsigned long zero_start
;
6747 sb_start_pagefault(inode
->i_sb
);
6748 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
6750 ret
= file_update_time(vma
->vm_file
);
6756 else /* -ENOSPC, -EIO, etc */
6757 ret
= VM_FAULT_SIGBUS
;
6763 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
6766 size
= i_size_read(inode
);
6767 page_start
= page_offset(page
);
6768 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
6770 if ((page
->mapping
!= inode
->i_mapping
) ||
6771 (page_start
>= size
)) {
6772 /* page got truncated out from underneath us */
6775 wait_on_page_writeback(page
);
6777 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
6778 set_page_extent_mapped(page
);
6781 * we can't set the delalloc bits if there are pending ordered
6782 * extents. Drop our locks and wait for them to finish
6784 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
6786 unlock_extent_cached(io_tree
, page_start
, page_end
,
6787 &cached_state
, GFP_NOFS
);
6789 btrfs_start_ordered_extent(inode
, ordered
, 1);
6790 btrfs_put_ordered_extent(ordered
);
6795 * XXX - page_mkwrite gets called every time the page is dirtied, even
6796 * if it was already dirty, so for space accounting reasons we need to
6797 * clear any delalloc bits for the range we are fixing to save. There
6798 * is probably a better way to do this, but for now keep consistent with
6799 * prepare_pages in the normal write path.
6801 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
6802 EXTENT_DIRTY
| EXTENT_DELALLOC
|
6803 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
6804 0, 0, &cached_state
, GFP_NOFS
);
6806 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
6809 unlock_extent_cached(io_tree
, page_start
, page_end
,
6810 &cached_state
, GFP_NOFS
);
6811 ret
= VM_FAULT_SIGBUS
;
6816 /* page is wholly or partially inside EOF */
6817 if (page_start
+ PAGE_CACHE_SIZE
> size
)
6818 zero_start
= size
& ~PAGE_CACHE_MASK
;
6820 zero_start
= PAGE_CACHE_SIZE
;
6822 if (zero_start
!= PAGE_CACHE_SIZE
) {
6824 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
6825 flush_dcache_page(page
);
6828 ClearPageChecked(page
);
6829 set_page_dirty(page
);
6830 SetPageUptodate(page
);
6832 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
6833 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
6834 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
6836 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
6840 sb_end_pagefault(inode
->i_sb
);
6841 return VM_FAULT_LOCKED
;
6845 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
6847 sb_end_pagefault(inode
->i_sb
);
6851 static int btrfs_truncate(struct inode
*inode
)
6853 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6854 struct btrfs_block_rsv
*rsv
;
6857 struct btrfs_trans_handle
*trans
;
6859 u64 mask
= root
->sectorsize
- 1;
6860 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
6862 ret
= btrfs_truncate_page(inode
, inode
->i_size
, 0, 0);
6866 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
6867 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
6870 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6871 * 3 things going on here
6873 * 1) We need to reserve space for our orphan item and the space to
6874 * delete our orphan item. Lord knows we don't want to have a dangling
6875 * orphan item because we didn't reserve space to remove it.
6877 * 2) We need to reserve space to update our inode.
6879 * 3) We need to have something to cache all the space that is going to
6880 * be free'd up by the truncate operation, but also have some slack
6881 * space reserved in case it uses space during the truncate (thank you
6882 * very much snapshotting).
6884 * And we need these to all be seperate. The fact is we can use alot of
6885 * space doing the truncate, and we have no earthly idea how much space
6886 * we will use, so we need the truncate reservation to be seperate so it
6887 * doesn't end up using space reserved for updating the inode or
6888 * removing the orphan item. We also need to be able to stop the
6889 * transaction and start a new one, which means we need to be able to
6890 * update the inode several times, and we have no idea of knowing how
6891 * many times that will be, so we can't just reserve 1 item for the
6892 * entirety of the opration, so that has to be done seperately as well.
6893 * Then there is the orphan item, which does indeed need to be held on
6894 * to for the whole operation, and we need nobody to touch this reserved
6895 * space except the orphan code.
6897 * So that leaves us with
6899 * 1) root->orphan_block_rsv - for the orphan deletion.
6900 * 2) rsv - for the truncate reservation, which we will steal from the
6901 * transaction reservation.
6902 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6903 * updating the inode.
6905 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
6908 rsv
->size
= min_size
;
6912 * 1 for the truncate slack space
6913 * 1 for the orphan item we're going to add
6914 * 1 for the orphan item deletion
6915 * 1 for updating the inode.
6917 trans
= btrfs_start_transaction(root
, 4);
6918 if (IS_ERR(trans
)) {
6919 err
= PTR_ERR(trans
);
6923 /* Migrate the slack space for the truncate to our reserve */
6924 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
6928 ret
= btrfs_orphan_add(trans
, inode
);
6930 btrfs_end_transaction(trans
, root
);
6935 * setattr is responsible for setting the ordered_data_close flag,
6936 * but that is only tested during the last file release. That
6937 * could happen well after the next commit, leaving a great big
6938 * window where new writes may get lost if someone chooses to write
6939 * to this file after truncating to zero
6941 * The inode doesn't have any dirty data here, and so if we commit
6942 * this is a noop. If someone immediately starts writing to the inode
6943 * it is very likely we'll catch some of their writes in this
6944 * transaction, and the commit will find this file on the ordered
6945 * data list with good things to send down.
6947 * This is a best effort solution, there is still a window where
6948 * using truncate to replace the contents of the file will
6949 * end up with a zero length file after a crash.
6951 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
6952 &BTRFS_I(inode
)->runtime_flags
))
6953 btrfs_add_ordered_operation(trans
, root
, inode
);
6956 * So if we truncate and then write and fsync we normally would just
6957 * write the extents that changed, which is a problem if we need to
6958 * first truncate that entire inode. So set this flag so we write out
6959 * all of the extents in the inode to the sync log so we're completely
6962 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
6963 trans
->block_rsv
= rsv
;
6966 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
6968 BTRFS_EXTENT_DATA_KEY
);
6969 if (ret
!= -ENOSPC
) {
6974 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
6975 ret
= btrfs_update_inode(trans
, root
, inode
);
6981 nr
= trans
->blocks_used
;
6982 btrfs_end_transaction(trans
, root
);
6983 btrfs_btree_balance_dirty(root
, nr
);
6985 trans
= btrfs_start_transaction(root
, 2);
6986 if (IS_ERR(trans
)) {
6987 ret
= err
= PTR_ERR(trans
);
6992 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
6994 BUG_ON(ret
); /* shouldn't happen */
6995 trans
->block_rsv
= rsv
;
6998 if (ret
== 0 && inode
->i_nlink
> 0) {
6999 trans
->block_rsv
= root
->orphan_block_rsv
;
7000 ret
= btrfs_orphan_del(trans
, inode
);
7003 } else if (ret
&& inode
->i_nlink
> 0) {
7005 * Failed to do the truncate, remove us from the in memory
7008 ret
= btrfs_orphan_del(NULL
, inode
);
7012 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7013 ret
= btrfs_update_inode(trans
, root
, inode
);
7017 nr
= trans
->blocks_used
;
7018 ret
= btrfs_end_transaction(trans
, root
);
7019 btrfs_btree_balance_dirty(root
, nr
);
7023 btrfs_free_block_rsv(root
, rsv
);
7032 * create a new subvolume directory/inode (helper for the ioctl).
7034 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
7035 struct btrfs_root
*new_root
, u64 new_dirid
)
7037 struct inode
*inode
;
7041 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
7042 new_dirid
, new_dirid
,
7043 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
7046 return PTR_ERR(inode
);
7047 inode
->i_op
= &btrfs_dir_inode_operations
;
7048 inode
->i_fop
= &btrfs_dir_file_operations
;
7050 set_nlink(inode
, 1);
7051 btrfs_i_size_write(inode
, 0);
7053 err
= btrfs_update_inode(trans
, new_root
, inode
);
7059 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
7061 struct btrfs_inode
*ei
;
7062 struct inode
*inode
;
7064 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
7071 ei
->last_sub_trans
= 0;
7072 ei
->logged_trans
= 0;
7073 ei
->delalloc_bytes
= 0;
7074 ei
->disk_i_size
= 0;
7077 ei
->index_cnt
= (u64
)-1;
7078 ei
->last_unlink_trans
= 0;
7079 ei
->last_log_commit
= 0;
7081 spin_lock_init(&ei
->lock
);
7082 ei
->outstanding_extents
= 0;
7083 ei
->reserved_extents
= 0;
7085 ei
->runtime_flags
= 0;
7086 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
7088 ei
->delayed_node
= NULL
;
7090 inode
= &ei
->vfs_inode
;
7091 extent_map_tree_init(&ei
->extent_tree
);
7092 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
7093 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
7094 ei
->io_tree
.track_uptodate
= 1;
7095 ei
->io_failure_tree
.track_uptodate
= 1;
7096 mutex_init(&ei
->log_mutex
);
7097 mutex_init(&ei
->delalloc_mutex
);
7098 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
7099 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
7100 INIT_LIST_HEAD(&ei
->ordered_operations
);
7101 RB_CLEAR_NODE(&ei
->rb_node
);
7106 static void btrfs_i_callback(struct rcu_head
*head
)
7108 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
7109 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
7112 void btrfs_destroy_inode(struct inode
*inode
)
7114 struct btrfs_ordered_extent
*ordered
;
7115 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7117 WARN_ON(!hlist_empty(&inode
->i_dentry
));
7118 WARN_ON(inode
->i_data
.nrpages
);
7119 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
7120 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
7121 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
7122 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
7125 * This can happen where we create an inode, but somebody else also
7126 * created the same inode and we need to destroy the one we already
7133 * Make sure we're properly removed from the ordered operation
7137 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
7138 spin_lock(&root
->fs_info
->ordered_extent_lock
);
7139 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
7140 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
7143 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
7144 &BTRFS_I(inode
)->runtime_flags
)) {
7145 printk(KERN_INFO
"BTRFS: inode %llu still on the orphan list\n",
7146 (unsigned long long)btrfs_ino(inode
));
7147 atomic_dec(&root
->orphan_inodes
);
7151 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
7155 printk(KERN_ERR
"btrfs found ordered "
7156 "extent %llu %llu on inode cleanup\n",
7157 (unsigned long long)ordered
->file_offset
,
7158 (unsigned long long)ordered
->len
);
7159 btrfs_remove_ordered_extent(inode
, ordered
);
7160 btrfs_put_ordered_extent(ordered
);
7161 btrfs_put_ordered_extent(ordered
);
7164 inode_tree_del(inode
);
7165 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
7167 btrfs_remove_delayed_node(inode
);
7168 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
7171 int btrfs_drop_inode(struct inode
*inode
)
7173 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7175 if (btrfs_root_refs(&root
->root_item
) == 0 &&
7176 !btrfs_is_free_space_inode(inode
))
7179 return generic_drop_inode(inode
);
7182 static void init_once(void *foo
)
7184 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
7186 inode_init_once(&ei
->vfs_inode
);
7189 void btrfs_destroy_cachep(void)
7192 * Make sure all delayed rcu free inodes are flushed before we
7196 if (btrfs_inode_cachep
)
7197 kmem_cache_destroy(btrfs_inode_cachep
);
7198 if (btrfs_trans_handle_cachep
)
7199 kmem_cache_destroy(btrfs_trans_handle_cachep
);
7200 if (btrfs_transaction_cachep
)
7201 kmem_cache_destroy(btrfs_transaction_cachep
);
7202 if (btrfs_path_cachep
)
7203 kmem_cache_destroy(btrfs_path_cachep
);
7204 if (btrfs_free_space_cachep
)
7205 kmem_cache_destroy(btrfs_free_space_cachep
);
7208 int btrfs_init_cachep(void)
7210 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
7211 sizeof(struct btrfs_inode
), 0,
7212 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
7213 if (!btrfs_inode_cachep
)
7216 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
7217 sizeof(struct btrfs_trans_handle
), 0,
7218 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7219 if (!btrfs_trans_handle_cachep
)
7222 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
7223 sizeof(struct btrfs_transaction
), 0,
7224 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7225 if (!btrfs_transaction_cachep
)
7228 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
7229 sizeof(struct btrfs_path
), 0,
7230 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7231 if (!btrfs_path_cachep
)
7234 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
7235 sizeof(struct btrfs_free_space
), 0,
7236 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
7237 if (!btrfs_free_space_cachep
)
7242 btrfs_destroy_cachep();
7246 static int btrfs_getattr(struct vfsmount
*mnt
,
7247 struct dentry
*dentry
, struct kstat
*stat
)
7249 struct inode
*inode
= dentry
->d_inode
;
7250 u32 blocksize
= inode
->i_sb
->s_blocksize
;
7252 generic_fillattr(inode
, stat
);
7253 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
7254 stat
->blksize
= PAGE_CACHE_SIZE
;
7255 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
7256 ALIGN(BTRFS_I(inode
)->delalloc_bytes
, blocksize
)) >> 9;
7261 * If a file is moved, it will inherit the cow and compression flags of the new
7264 static void fixup_inode_flags(struct inode
*dir
, struct inode
*inode
)
7266 struct btrfs_inode
*b_dir
= BTRFS_I(dir
);
7267 struct btrfs_inode
*b_inode
= BTRFS_I(inode
);
7269 if (b_dir
->flags
& BTRFS_INODE_NODATACOW
)
7270 b_inode
->flags
|= BTRFS_INODE_NODATACOW
;
7272 b_inode
->flags
&= ~BTRFS_INODE_NODATACOW
;
7274 if (b_dir
->flags
& BTRFS_INODE_COMPRESS
) {
7275 b_inode
->flags
|= BTRFS_INODE_COMPRESS
;
7276 b_inode
->flags
&= ~BTRFS_INODE_NOCOMPRESS
;
7278 b_inode
->flags
&= ~(BTRFS_INODE_COMPRESS
|
7279 BTRFS_INODE_NOCOMPRESS
);
7283 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
7284 struct inode
*new_dir
, struct dentry
*new_dentry
)
7286 struct btrfs_trans_handle
*trans
;
7287 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
7288 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
7289 struct inode
*new_inode
= new_dentry
->d_inode
;
7290 struct inode
*old_inode
= old_dentry
->d_inode
;
7291 struct timespec ctime
= CURRENT_TIME
;
7295 u64 old_ino
= btrfs_ino(old_inode
);
7297 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
7300 /* we only allow rename subvolume link between subvolumes */
7301 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
7304 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
7305 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
7308 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
7309 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
7312 * we're using rename to replace one file with another.
7313 * and the replacement file is large. Start IO on it now so
7314 * we don't add too much work to the end of the transaction
7316 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
7317 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
7318 filemap_flush(old_inode
->i_mapping
);
7320 /* close the racy window with snapshot create/destroy ioctl */
7321 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7322 down_read(&root
->fs_info
->subvol_sem
);
7324 * We want to reserve the absolute worst case amount of items. So if
7325 * both inodes are subvols and we need to unlink them then that would
7326 * require 4 item modifications, but if they are both normal inodes it
7327 * would require 5 item modifications, so we'll assume their normal
7328 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7329 * should cover the worst case number of items we'll modify.
7331 trans
= btrfs_start_transaction(root
, 20);
7332 if (IS_ERR(trans
)) {
7333 ret
= PTR_ERR(trans
);
7338 btrfs_record_root_in_trans(trans
, dest
);
7340 ret
= btrfs_set_inode_index(new_dir
, &index
);
7344 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7345 /* force full log commit if subvolume involved. */
7346 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7348 ret
= btrfs_insert_inode_ref(trans
, dest
,
7349 new_dentry
->d_name
.name
,
7350 new_dentry
->d_name
.len
,
7352 btrfs_ino(new_dir
), index
);
7356 * this is an ugly little race, but the rename is required
7357 * to make sure that if we crash, the inode is either at the
7358 * old name or the new one. pinning the log transaction lets
7359 * us make sure we don't allow a log commit to come in after
7360 * we unlink the name but before we add the new name back in.
7362 btrfs_pin_log_trans(root
);
7365 * make sure the inode gets flushed if it is replacing
7368 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
7369 btrfs_add_ordered_operation(trans
, root
, old_inode
);
7371 inode_inc_iversion(old_dir
);
7372 inode_inc_iversion(new_dir
);
7373 inode_inc_iversion(old_inode
);
7374 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
7375 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
7376 old_inode
->i_ctime
= ctime
;
7378 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
7379 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
7381 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
7382 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
7383 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
7384 old_dentry
->d_name
.name
,
7385 old_dentry
->d_name
.len
);
7387 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
7388 old_dentry
->d_inode
,
7389 old_dentry
->d_name
.name
,
7390 old_dentry
->d_name
.len
);
7392 ret
= btrfs_update_inode(trans
, root
, old_inode
);
7395 btrfs_abort_transaction(trans
, root
, ret
);
7400 inode_inc_iversion(new_inode
);
7401 new_inode
->i_ctime
= CURRENT_TIME
;
7402 if (unlikely(btrfs_ino(new_inode
) ==
7403 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
7404 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
7405 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
7407 new_dentry
->d_name
.name
,
7408 new_dentry
->d_name
.len
);
7409 BUG_ON(new_inode
->i_nlink
== 0);
7411 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
7412 new_dentry
->d_inode
,
7413 new_dentry
->d_name
.name
,
7414 new_dentry
->d_name
.len
);
7416 if (!ret
&& new_inode
->i_nlink
== 0) {
7417 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
7421 btrfs_abort_transaction(trans
, root
, ret
);
7426 fixup_inode_flags(new_dir
, old_inode
);
7428 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
7429 new_dentry
->d_name
.name
,
7430 new_dentry
->d_name
.len
, 0, index
);
7432 btrfs_abort_transaction(trans
, root
, ret
);
7436 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
7437 struct dentry
*parent
= new_dentry
->d_parent
;
7438 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
7439 btrfs_end_log_trans(root
);
7442 btrfs_end_transaction(trans
, root
);
7444 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
7445 up_read(&root
->fs_info
->subvol_sem
);
7451 * some fairly slow code that needs optimization. This walks the list
7452 * of all the inodes with pending delalloc and forces them to disk.
7454 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
7456 struct list_head
*head
= &root
->fs_info
->delalloc_inodes
;
7457 struct btrfs_inode
*binode
;
7458 struct inode
*inode
;
7460 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
7463 spin_lock(&root
->fs_info
->delalloc_lock
);
7464 while (!list_empty(head
)) {
7465 binode
= list_entry(head
->next
, struct btrfs_inode
,
7467 inode
= igrab(&binode
->vfs_inode
);
7469 list_del_init(&binode
->delalloc_inodes
);
7470 spin_unlock(&root
->fs_info
->delalloc_lock
);
7472 filemap_flush(inode
->i_mapping
);
7474 btrfs_add_delayed_iput(inode
);
7479 spin_lock(&root
->fs_info
->delalloc_lock
);
7481 spin_unlock(&root
->fs_info
->delalloc_lock
);
7483 /* the filemap_flush will queue IO into the worker threads, but
7484 * we have to make sure the IO is actually started and that
7485 * ordered extents get created before we return
7487 atomic_inc(&root
->fs_info
->async_submit_draining
);
7488 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
7489 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
7490 wait_event(root
->fs_info
->async_submit_wait
,
7491 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
7492 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
7494 atomic_dec(&root
->fs_info
->async_submit_draining
);
7498 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
7499 const char *symname
)
7501 struct btrfs_trans_handle
*trans
;
7502 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
7503 struct btrfs_path
*path
;
7504 struct btrfs_key key
;
7505 struct inode
*inode
= NULL
;
7513 struct btrfs_file_extent_item
*ei
;
7514 struct extent_buffer
*leaf
;
7515 unsigned long nr
= 0;
7517 name_len
= strlen(symname
) + 1;
7518 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
7519 return -ENAMETOOLONG
;
7522 * 2 items for inode item and ref
7523 * 2 items for dir items
7524 * 1 item for xattr if selinux is on
7526 trans
= btrfs_start_transaction(root
, 5);
7528 return PTR_ERR(trans
);
7530 err
= btrfs_find_free_ino(root
, &objectid
);
7534 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
7535 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
7536 S_IFLNK
|S_IRWXUGO
, &index
);
7537 if (IS_ERR(inode
)) {
7538 err
= PTR_ERR(inode
);
7542 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
7549 * If the active LSM wants to access the inode during
7550 * d_instantiate it needs these. Smack checks to see
7551 * if the filesystem supports xattrs by looking at the
7554 inode
->i_fop
= &btrfs_file_operations
;
7555 inode
->i_op
= &btrfs_file_inode_operations
;
7557 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
7561 inode
->i_mapping
->a_ops
= &btrfs_aops
;
7562 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7563 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
7568 path
= btrfs_alloc_path();
7574 key
.objectid
= btrfs_ino(inode
);
7576 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
7577 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
7578 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
7582 btrfs_free_path(path
);
7585 leaf
= path
->nodes
[0];
7586 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
7587 struct btrfs_file_extent_item
);
7588 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
7589 btrfs_set_file_extent_type(leaf
, ei
,
7590 BTRFS_FILE_EXTENT_INLINE
);
7591 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
7592 btrfs_set_file_extent_compression(leaf
, ei
, 0);
7593 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
7594 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
7596 ptr
= btrfs_file_extent_inline_start(ei
);
7597 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
7598 btrfs_mark_buffer_dirty(leaf
);
7599 btrfs_free_path(path
);
7601 inode
->i_op
= &btrfs_symlink_inode_operations
;
7602 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
7603 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
7604 inode_set_bytes(inode
, name_len
);
7605 btrfs_i_size_write(inode
, name_len
- 1);
7606 err
= btrfs_update_inode(trans
, root
, inode
);
7612 d_instantiate(dentry
, inode
);
7613 nr
= trans
->blocks_used
;
7614 btrfs_end_transaction(trans
, root
);
7616 inode_dec_link_count(inode
);
7619 btrfs_btree_balance_dirty(root
, nr
);
7623 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7624 u64 start
, u64 num_bytes
, u64 min_size
,
7625 loff_t actual_len
, u64
*alloc_hint
,
7626 struct btrfs_trans_handle
*trans
)
7628 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
7629 struct extent_map
*em
;
7630 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7631 struct btrfs_key ins
;
7632 u64 cur_offset
= start
;
7635 bool own_trans
= true;
7639 while (num_bytes
> 0) {
7641 trans
= btrfs_start_transaction(root
, 3);
7642 if (IS_ERR(trans
)) {
7643 ret
= PTR_ERR(trans
);
7648 ret
= btrfs_reserve_extent(trans
, root
, num_bytes
, min_size
,
7649 0, *alloc_hint
, &ins
, 1);
7652 btrfs_end_transaction(trans
, root
);
7656 ret
= insert_reserved_file_extent(trans
, inode
,
7657 cur_offset
, ins
.objectid
,
7658 ins
.offset
, ins
.offset
,
7659 ins
.offset
, 0, 0, 0,
7660 BTRFS_FILE_EXTENT_PREALLOC
);
7662 btrfs_abort_transaction(trans
, root
, ret
);
7664 btrfs_end_transaction(trans
, root
);
7667 btrfs_drop_extent_cache(inode
, cur_offset
,
7668 cur_offset
+ ins
.offset
-1, 0);
7670 em
= alloc_extent_map();
7672 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
7673 &BTRFS_I(inode
)->runtime_flags
);
7677 em
->start
= cur_offset
;
7678 em
->orig_start
= cur_offset
;
7679 em
->len
= ins
.offset
;
7680 em
->block_start
= ins
.objectid
;
7681 em
->block_len
= ins
.offset
;
7682 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
7683 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
7684 em
->generation
= trans
->transid
;
7687 write_lock(&em_tree
->lock
);
7688 ret
= add_extent_mapping(em_tree
, em
);
7690 list_move(&em
->list
,
7691 &em_tree
->modified_extents
);
7692 write_unlock(&em_tree
->lock
);
7695 btrfs_drop_extent_cache(inode
, cur_offset
,
7696 cur_offset
+ ins
.offset
- 1,
7699 free_extent_map(em
);
7701 num_bytes
-= ins
.offset
;
7702 cur_offset
+= ins
.offset
;
7703 *alloc_hint
= ins
.objectid
+ ins
.offset
;
7705 inode_inc_iversion(inode
);
7706 inode
->i_ctime
= CURRENT_TIME
;
7707 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
7708 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
7709 (actual_len
> inode
->i_size
) &&
7710 (cur_offset
> inode
->i_size
)) {
7711 if (cur_offset
> actual_len
)
7712 i_size
= actual_len
;
7714 i_size
= cur_offset
;
7715 i_size_write(inode
, i_size
);
7716 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
7719 ret
= btrfs_update_inode(trans
, root
, inode
);
7722 btrfs_abort_transaction(trans
, root
, ret
);
7724 btrfs_end_transaction(trans
, root
);
7729 btrfs_end_transaction(trans
, root
);
7734 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
7735 u64 start
, u64 num_bytes
, u64 min_size
,
7736 loff_t actual_len
, u64
*alloc_hint
)
7738 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7739 min_size
, actual_len
, alloc_hint
,
7743 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
7744 struct btrfs_trans_handle
*trans
, int mode
,
7745 u64 start
, u64 num_bytes
, u64 min_size
,
7746 loff_t actual_len
, u64
*alloc_hint
)
7748 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
7749 min_size
, actual_len
, alloc_hint
, trans
);
7752 static int btrfs_set_page_dirty(struct page
*page
)
7754 return __set_page_dirty_nobuffers(page
);
7757 static int btrfs_permission(struct inode
*inode
, int mask
)
7759 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7760 umode_t mode
= inode
->i_mode
;
7762 if (mask
& MAY_WRITE
&&
7763 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
7764 if (btrfs_root_readonly(root
))
7766 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
7769 return generic_permission(inode
, mask
);
7772 static const struct inode_operations btrfs_dir_inode_operations
= {
7773 .getattr
= btrfs_getattr
,
7774 .lookup
= btrfs_lookup
,
7775 .create
= btrfs_create
,
7776 .unlink
= btrfs_unlink
,
7778 .mkdir
= btrfs_mkdir
,
7779 .rmdir
= btrfs_rmdir
,
7780 .rename
= btrfs_rename
,
7781 .symlink
= btrfs_symlink
,
7782 .setattr
= btrfs_setattr
,
7783 .mknod
= btrfs_mknod
,
7784 .setxattr
= btrfs_setxattr
,
7785 .getxattr
= btrfs_getxattr
,
7786 .listxattr
= btrfs_listxattr
,
7787 .removexattr
= btrfs_removexattr
,
7788 .permission
= btrfs_permission
,
7789 .get_acl
= btrfs_get_acl
,
7791 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
7792 .lookup
= btrfs_lookup
,
7793 .permission
= btrfs_permission
,
7794 .get_acl
= btrfs_get_acl
,
7797 static const struct file_operations btrfs_dir_file_operations
= {
7798 .llseek
= generic_file_llseek
,
7799 .read
= generic_read_dir
,
7800 .readdir
= btrfs_real_readdir
,
7801 .unlocked_ioctl
= btrfs_ioctl
,
7802 #ifdef CONFIG_COMPAT
7803 .compat_ioctl
= btrfs_ioctl
,
7805 .release
= btrfs_release_file
,
7806 .fsync
= btrfs_sync_file
,
7809 static struct extent_io_ops btrfs_extent_io_ops
= {
7810 .fill_delalloc
= run_delalloc_range
,
7811 .submit_bio_hook
= btrfs_submit_bio_hook
,
7812 .merge_bio_hook
= btrfs_merge_bio_hook
,
7813 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
7814 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
7815 .writepage_start_hook
= btrfs_writepage_start_hook
,
7816 .set_bit_hook
= btrfs_set_bit_hook
,
7817 .clear_bit_hook
= btrfs_clear_bit_hook
,
7818 .merge_extent_hook
= btrfs_merge_extent_hook
,
7819 .split_extent_hook
= btrfs_split_extent_hook
,
7823 * btrfs doesn't support the bmap operation because swapfiles
7824 * use bmap to make a mapping of extents in the file. They assume
7825 * these extents won't change over the life of the file and they
7826 * use the bmap result to do IO directly to the drive.
7828 * the btrfs bmap call would return logical addresses that aren't
7829 * suitable for IO and they also will change frequently as COW
7830 * operations happen. So, swapfile + btrfs == corruption.
7832 * For now we're avoiding this by dropping bmap.
7834 static const struct address_space_operations btrfs_aops
= {
7835 .readpage
= btrfs_readpage
,
7836 .writepage
= btrfs_writepage
,
7837 .writepages
= btrfs_writepages
,
7838 .readpages
= btrfs_readpages
,
7839 .direct_IO
= btrfs_direct_IO
,
7840 .invalidatepage
= btrfs_invalidatepage
,
7841 .releasepage
= btrfs_releasepage
,
7842 .set_page_dirty
= btrfs_set_page_dirty
,
7843 .error_remove_page
= generic_error_remove_page
,
7846 static const struct address_space_operations btrfs_symlink_aops
= {
7847 .readpage
= btrfs_readpage
,
7848 .writepage
= btrfs_writepage
,
7849 .invalidatepage
= btrfs_invalidatepage
,
7850 .releasepage
= btrfs_releasepage
,
7853 static const struct inode_operations btrfs_file_inode_operations
= {
7854 .getattr
= btrfs_getattr
,
7855 .setattr
= btrfs_setattr
,
7856 .setxattr
= btrfs_setxattr
,
7857 .getxattr
= btrfs_getxattr
,
7858 .listxattr
= btrfs_listxattr
,
7859 .removexattr
= btrfs_removexattr
,
7860 .permission
= btrfs_permission
,
7861 .fiemap
= btrfs_fiemap
,
7862 .get_acl
= btrfs_get_acl
,
7863 .update_time
= btrfs_update_time
,
7865 static const struct inode_operations btrfs_special_inode_operations
= {
7866 .getattr
= btrfs_getattr
,
7867 .setattr
= btrfs_setattr
,
7868 .permission
= btrfs_permission
,
7869 .setxattr
= btrfs_setxattr
,
7870 .getxattr
= btrfs_getxattr
,
7871 .listxattr
= btrfs_listxattr
,
7872 .removexattr
= btrfs_removexattr
,
7873 .get_acl
= btrfs_get_acl
,
7874 .update_time
= btrfs_update_time
,
7876 static const struct inode_operations btrfs_symlink_inode_operations
= {
7877 .readlink
= generic_readlink
,
7878 .follow_link
= page_follow_link_light
,
7879 .put_link
= page_put_link
,
7880 .getattr
= btrfs_getattr
,
7881 .setattr
= btrfs_setattr
,
7882 .permission
= btrfs_permission
,
7883 .setxattr
= btrfs_setxattr
,
7884 .getxattr
= btrfs_getxattr
,
7885 .listxattr
= btrfs_listxattr
,
7886 .removexattr
= btrfs_removexattr
,
7887 .get_acl
= btrfs_get_acl
,
7888 .update_time
= btrfs_update_time
,
7891 const struct dentry_operations btrfs_dentry_operations
= {
7892 .d_delete
= btrfs_dentry_delete
,
7893 .d_release
= btrfs_dentry_release
,