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/aio.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/xattr.h>
38 #include <linux/posix_acl.h>
39 #include <linux/falloc.h>
40 #include <linux/slab.h>
41 #include <linux/ratelimit.h>
42 #include <linux/mount.h>
43 #include <linux/btrfs.h>
44 #include <linux/blkdev.h>
45 #include <linux/posix_acl_xattr.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
63 struct btrfs_iget_args
{
64 struct btrfs_key
*location
;
65 struct btrfs_root
*root
;
68 static const struct inode_operations btrfs_dir_inode_operations
;
69 static const struct inode_operations btrfs_symlink_inode_operations
;
70 static const struct inode_operations btrfs_dir_ro_inode_operations
;
71 static const struct inode_operations btrfs_special_inode_operations
;
72 static const struct inode_operations btrfs_file_inode_operations
;
73 static const struct address_space_operations btrfs_aops
;
74 static const struct address_space_operations btrfs_symlink_aops
;
75 static const struct file_operations btrfs_dir_file_operations
;
76 static struct extent_io_ops btrfs_extent_io_ops
;
78 static struct kmem_cache
*btrfs_inode_cachep
;
79 static struct kmem_cache
*btrfs_delalloc_work_cachep
;
80 struct kmem_cache
*btrfs_trans_handle_cachep
;
81 struct kmem_cache
*btrfs_transaction_cachep
;
82 struct kmem_cache
*btrfs_path_cachep
;
83 struct kmem_cache
*btrfs_free_space_cachep
;
86 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
87 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
88 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
89 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
90 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
91 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
92 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
93 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
96 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
97 static int btrfs_truncate(struct inode
*inode
);
98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
99 static noinline
int cow_file_range(struct inode
*inode
,
100 struct page
*locked_page
,
101 u64 start
, u64 end
, int *page_started
,
102 unsigned long *nr_written
, int unlock
);
103 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
104 u64 len
, u64 orig_start
,
105 u64 block_start
, u64 block_len
,
106 u64 orig_block_len
, u64 ram_bytes
,
109 static int btrfs_dirty_inode(struct inode
*inode
);
111 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
112 struct inode
*inode
, struct inode
*dir
,
113 const struct qstr
*qstr
)
117 err
= btrfs_init_acl(trans
, inode
, dir
);
119 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
129 struct btrfs_path
*path
, int extent_inserted
,
130 struct btrfs_root
*root
, struct inode
*inode
,
131 u64 start
, size_t size
, size_t compressed_size
,
133 struct page
**compressed_pages
)
135 struct extent_buffer
*leaf
;
136 struct page
*page
= NULL
;
139 struct btrfs_file_extent_item
*ei
;
142 size_t cur_size
= size
;
143 unsigned long offset
;
145 if (compressed_size
&& compressed_pages
)
146 cur_size
= compressed_size
;
148 inode_add_bytes(inode
, size
);
150 if (!extent_inserted
) {
151 struct btrfs_key key
;
154 key
.objectid
= btrfs_ino(inode
);
156 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
158 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
159 path
->leave_spinning
= 1;
160 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
167 leaf
= path
->nodes
[0];
168 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
169 struct btrfs_file_extent_item
);
170 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
171 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
172 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
173 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
174 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
175 ptr
= btrfs_file_extent_inline_start(ei
);
177 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
180 while (compressed_size
> 0) {
181 cpage
= compressed_pages
[i
];
182 cur_size
= min_t(unsigned long, compressed_size
,
185 kaddr
= kmap_atomic(cpage
);
186 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
187 kunmap_atomic(kaddr
);
191 compressed_size
-= cur_size
;
193 btrfs_set_file_extent_compression(leaf
, ei
,
196 page
= find_get_page(inode
->i_mapping
,
197 start
>> PAGE_CACHE_SHIFT
);
198 btrfs_set_file_extent_compression(leaf
, ei
, 0);
199 kaddr
= kmap_atomic(page
);
200 offset
= start
& (PAGE_CACHE_SIZE
- 1);
201 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
202 kunmap_atomic(kaddr
);
203 page_cache_release(page
);
205 btrfs_mark_buffer_dirty(leaf
);
206 btrfs_release_path(path
);
209 * we're an inline extent, so nobody can
210 * extend the file past i_size without locking
211 * a page we already have locked.
213 * We must do any isize and inode updates
214 * before we unlock the pages. Otherwise we
215 * could end up racing with unlink.
217 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
218 ret
= btrfs_update_inode(trans
, root
, inode
);
227 * conditionally insert an inline extent into the file. This
228 * does the checks required to make sure the data is small enough
229 * to fit as an inline extent.
231 static noinline
int cow_file_range_inline(struct btrfs_root
*root
,
232 struct inode
*inode
, u64 start
,
233 u64 end
, size_t compressed_size
,
235 struct page
**compressed_pages
)
237 struct btrfs_trans_handle
*trans
;
238 u64 isize
= i_size_read(inode
);
239 u64 actual_end
= min(end
+ 1, isize
);
240 u64 inline_len
= actual_end
- start
;
241 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
242 u64 data_len
= inline_len
;
244 struct btrfs_path
*path
;
245 int extent_inserted
= 0;
246 u32 extent_item_size
;
249 data_len
= compressed_size
;
252 actual_end
>= PAGE_CACHE_SIZE
||
253 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
255 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
257 data_len
> root
->fs_info
->max_inline
) {
261 path
= btrfs_alloc_path();
265 trans
= btrfs_join_transaction(root
);
267 btrfs_free_path(path
);
268 return PTR_ERR(trans
);
270 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
272 if (compressed_size
&& compressed_pages
)
273 extent_item_size
= btrfs_file_extent_calc_inline_size(
276 extent_item_size
= btrfs_file_extent_calc_inline_size(
279 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
,
280 start
, aligned_end
, NULL
,
281 1, 1, extent_item_size
, &extent_inserted
);
283 btrfs_abort_transaction(trans
, root
, ret
);
287 if (isize
> actual_end
)
288 inline_len
= min_t(u64
, isize
, actual_end
);
289 ret
= insert_inline_extent(trans
, path
, extent_inserted
,
291 inline_len
, compressed_size
,
292 compress_type
, compressed_pages
);
293 if (ret
&& ret
!= -ENOSPC
) {
294 btrfs_abort_transaction(trans
, root
, ret
);
296 } else if (ret
== -ENOSPC
) {
301 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
302 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
303 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
305 btrfs_free_path(path
);
306 btrfs_end_transaction(trans
, root
);
310 struct async_extent
{
315 unsigned long nr_pages
;
317 struct list_head list
;
322 struct btrfs_root
*root
;
323 struct page
*locked_page
;
326 struct list_head extents
;
327 struct btrfs_work work
;
330 static noinline
int add_async_extent(struct async_cow
*cow
,
331 u64 start
, u64 ram_size
,
334 unsigned long nr_pages
,
337 struct async_extent
*async_extent
;
339 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
340 BUG_ON(!async_extent
); /* -ENOMEM */
341 async_extent
->start
= start
;
342 async_extent
->ram_size
= ram_size
;
343 async_extent
->compressed_size
= compressed_size
;
344 async_extent
->pages
= pages
;
345 async_extent
->nr_pages
= nr_pages
;
346 async_extent
->compress_type
= compress_type
;
347 list_add_tail(&async_extent
->list
, &cow
->extents
);
352 * we create compressed extents in two phases. The first
353 * phase compresses a range of pages that have already been
354 * locked (both pages and state bits are locked).
356 * This is done inside an ordered work queue, and the compression
357 * is spread across many cpus. The actual IO submission is step
358 * two, and the ordered work queue takes care of making sure that
359 * happens in the same order things were put onto the queue by
360 * writepages and friends.
362 * If this code finds it can't get good compression, it puts an
363 * entry onto the work queue to write the uncompressed bytes. This
364 * makes sure that both compressed inodes and uncompressed inodes
365 * are written in the same order that the flusher thread sent them
368 static noinline
int compress_file_range(struct inode
*inode
,
369 struct page
*locked_page
,
371 struct async_cow
*async_cow
,
374 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
376 u64 blocksize
= root
->sectorsize
;
378 u64 isize
= i_size_read(inode
);
380 struct page
**pages
= NULL
;
381 unsigned long nr_pages
;
382 unsigned long nr_pages_ret
= 0;
383 unsigned long total_compressed
= 0;
384 unsigned long total_in
= 0;
385 unsigned long max_compressed
= 128 * 1024;
386 unsigned long max_uncompressed
= 128 * 1024;
389 int compress_type
= root
->fs_info
->compress_type
;
392 /* if this is a small write inside eof, kick off a defrag */
393 if ((end
- start
+ 1) < 16 * 1024 &&
394 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
395 btrfs_add_inode_defrag(NULL
, inode
);
397 actual_end
= min_t(u64
, isize
, end
+ 1);
400 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
401 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
404 * we don't want to send crud past the end of i_size through
405 * compression, that's just a waste of CPU time. So, if the
406 * end of the file is before the start of our current
407 * requested range of bytes, we bail out to the uncompressed
408 * cleanup code that can deal with all of this.
410 * It isn't really the fastest way to fix things, but this is a
411 * very uncommon corner.
413 if (actual_end
<= start
)
414 goto cleanup_and_bail_uncompressed
;
416 total_compressed
= actual_end
- start
;
418 /* we want to make sure that amount of ram required to uncompress
419 * an extent is reasonable, so we limit the total size in ram
420 * of a compressed extent to 128k. This is a crucial number
421 * because it also controls how easily we can spread reads across
422 * cpus for decompression.
424 * We also want to make sure the amount of IO required to do
425 * a random read is reasonably small, so we limit the size of
426 * a compressed extent to 128k.
428 total_compressed
= min(total_compressed
, max_uncompressed
);
429 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
430 num_bytes
= max(blocksize
, num_bytes
);
435 * we do compression for mount -o compress and when the
436 * inode has not been flagged as nocompress. This flag can
437 * change at any time if we discover bad compression ratios.
439 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
440 (btrfs_test_opt(root
, COMPRESS
) ||
441 (BTRFS_I(inode
)->force_compress
) ||
442 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
444 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
446 /* just bail out to the uncompressed code */
450 if (BTRFS_I(inode
)->force_compress
)
451 compress_type
= BTRFS_I(inode
)->force_compress
;
454 * we need to call clear_page_dirty_for_io on each
455 * page in the range. Otherwise applications with the file
456 * mmap'd can wander in and change the page contents while
457 * we are compressing them.
459 * If the compression fails for any reason, we set the pages
460 * dirty again later on.
462 extent_range_clear_dirty_for_io(inode
, start
, end
);
464 ret
= btrfs_compress_pages(compress_type
,
465 inode
->i_mapping
, start
,
466 total_compressed
, pages
,
467 nr_pages
, &nr_pages_ret
,
473 unsigned long offset
= total_compressed
&
474 (PAGE_CACHE_SIZE
- 1);
475 struct page
*page
= pages
[nr_pages_ret
- 1];
478 /* zero the tail end of the last page, we might be
479 * sending it down to disk
482 kaddr
= kmap_atomic(page
);
483 memset(kaddr
+ offset
, 0,
484 PAGE_CACHE_SIZE
- offset
);
485 kunmap_atomic(kaddr
);
492 /* lets try to make an inline extent */
493 if (ret
|| total_in
< (actual_end
- start
)) {
494 /* we didn't compress the entire range, try
495 * to make an uncompressed inline extent.
497 ret
= cow_file_range_inline(root
, inode
, start
, end
,
500 /* try making a compressed inline extent */
501 ret
= cow_file_range_inline(root
, inode
, start
, end
,
503 compress_type
, pages
);
506 unsigned long clear_flags
= EXTENT_DELALLOC
|
508 clear_flags
|= (ret
< 0) ? EXTENT_DO_ACCOUNTING
: 0;
511 * inline extent creation worked or returned error,
512 * we don't need to create any more async work items.
513 * Unlock and free up our temp pages.
515 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
516 clear_flags
, PAGE_UNLOCK
|
526 * we aren't doing an inline extent round the compressed size
527 * up to a block size boundary so the allocator does sane
530 total_compressed
= ALIGN(total_compressed
, blocksize
);
533 * one last check to make sure the compression is really a
534 * win, compare the page count read with the blocks on disk
536 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
537 if (total_compressed
>= total_in
) {
540 num_bytes
= total_in
;
543 if (!will_compress
&& pages
) {
545 * the compression code ran but failed to make things smaller,
546 * free any pages it allocated and our page pointer array
548 for (i
= 0; i
< nr_pages_ret
; i
++) {
549 WARN_ON(pages
[i
]->mapping
);
550 page_cache_release(pages
[i
]);
554 total_compressed
= 0;
557 /* flag the file so we don't compress in the future */
558 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
559 !(BTRFS_I(inode
)->force_compress
)) {
560 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
566 /* the async work queues will take care of doing actual
567 * allocation on disk for these compressed pages,
568 * and will submit them to the elevator.
570 add_async_extent(async_cow
, start
, num_bytes
,
571 total_compressed
, pages
, nr_pages_ret
,
574 if (start
+ num_bytes
< end
) {
581 cleanup_and_bail_uncompressed
:
583 * No compression, but we still need to write the pages in
584 * the file we've been given so far. redirty the locked
585 * page if it corresponds to our extent and set things up
586 * for the async work queue to run cow_file_range to do
587 * the normal delalloc dance
589 if (page_offset(locked_page
) >= start
&&
590 page_offset(locked_page
) <= end
) {
591 __set_page_dirty_nobuffers(locked_page
);
592 /* unlocked later on in the async handlers */
595 extent_range_redirty_for_io(inode
, start
, end
);
596 add_async_extent(async_cow
, start
, end
- start
+ 1,
597 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
605 for (i
= 0; i
< nr_pages_ret
; i
++) {
606 WARN_ON(pages
[i
]->mapping
);
607 page_cache_release(pages
[i
]);
615 * phase two of compressed writeback. This is the ordered portion
616 * of the code, which only gets called in the order the work was
617 * queued. We walk all the async extents created by compress_file_range
618 * and send them down to the disk.
620 static noinline
int submit_compressed_extents(struct inode
*inode
,
621 struct async_cow
*async_cow
)
623 struct async_extent
*async_extent
;
625 struct btrfs_key ins
;
626 struct extent_map
*em
;
627 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
628 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
629 struct extent_io_tree
*io_tree
;
632 if (list_empty(&async_cow
->extents
))
636 while (!list_empty(&async_cow
->extents
)) {
637 async_extent
= list_entry(async_cow
->extents
.next
,
638 struct async_extent
, list
);
639 list_del(&async_extent
->list
);
641 io_tree
= &BTRFS_I(inode
)->io_tree
;
644 /* did the compression code fall back to uncompressed IO? */
645 if (!async_extent
->pages
) {
646 int page_started
= 0;
647 unsigned long nr_written
= 0;
649 lock_extent(io_tree
, async_extent
->start
,
650 async_extent
->start
+
651 async_extent
->ram_size
- 1);
653 /* allocate blocks */
654 ret
= cow_file_range(inode
, async_cow
->locked_page
,
656 async_extent
->start
+
657 async_extent
->ram_size
- 1,
658 &page_started
, &nr_written
, 0);
663 * if page_started, cow_file_range inserted an
664 * inline extent and took care of all the unlocking
665 * and IO for us. Otherwise, we need to submit
666 * all those pages down to the drive.
668 if (!page_started
&& !ret
)
669 extent_write_locked_range(io_tree
,
670 inode
, async_extent
->start
,
671 async_extent
->start
+
672 async_extent
->ram_size
- 1,
676 unlock_page(async_cow
->locked_page
);
682 lock_extent(io_tree
, async_extent
->start
,
683 async_extent
->start
+ async_extent
->ram_size
- 1);
685 ret
= btrfs_reserve_extent(root
,
686 async_extent
->compressed_size
,
687 async_extent
->compressed_size
,
688 0, alloc_hint
, &ins
, 1);
692 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
693 WARN_ON(async_extent
->pages
[i
]->mapping
);
694 page_cache_release(async_extent
->pages
[i
]);
696 kfree(async_extent
->pages
);
697 async_extent
->nr_pages
= 0;
698 async_extent
->pages
= NULL
;
700 if (ret
== -ENOSPC
) {
701 unlock_extent(io_tree
, async_extent
->start
,
702 async_extent
->start
+
703 async_extent
->ram_size
- 1);
710 * here we're doing allocation and writeback of the
713 btrfs_drop_extent_cache(inode
, async_extent
->start
,
714 async_extent
->start
+
715 async_extent
->ram_size
- 1, 0);
717 em
= alloc_extent_map();
720 goto out_free_reserve
;
722 em
->start
= async_extent
->start
;
723 em
->len
= async_extent
->ram_size
;
724 em
->orig_start
= em
->start
;
725 em
->mod_start
= em
->start
;
726 em
->mod_len
= em
->len
;
728 em
->block_start
= ins
.objectid
;
729 em
->block_len
= ins
.offset
;
730 em
->orig_block_len
= ins
.offset
;
731 em
->ram_bytes
= async_extent
->ram_size
;
732 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
733 em
->compress_type
= async_extent
->compress_type
;
734 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
735 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
739 write_lock(&em_tree
->lock
);
740 ret
= add_extent_mapping(em_tree
, em
, 1);
741 write_unlock(&em_tree
->lock
);
742 if (ret
!= -EEXIST
) {
746 btrfs_drop_extent_cache(inode
, async_extent
->start
,
747 async_extent
->start
+
748 async_extent
->ram_size
- 1, 0);
752 goto out_free_reserve
;
754 ret
= btrfs_add_ordered_extent_compress(inode
,
757 async_extent
->ram_size
,
759 BTRFS_ORDERED_COMPRESSED
,
760 async_extent
->compress_type
);
762 goto out_free_reserve
;
765 * clear dirty, set writeback and unlock the pages.
767 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
768 async_extent
->start
+
769 async_extent
->ram_size
- 1,
770 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
,
771 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
773 ret
= btrfs_submit_compressed_write(inode
,
775 async_extent
->ram_size
,
777 ins
.offset
, async_extent
->pages
,
778 async_extent
->nr_pages
);
779 alloc_hint
= ins
.objectid
+ ins
.offset
;
789 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
791 extent_clear_unlock_delalloc(inode
, async_extent
->start
,
792 async_extent
->start
+
793 async_extent
->ram_size
- 1,
794 NULL
, EXTENT_LOCKED
| EXTENT_DELALLOC
|
795 EXTENT_DEFRAG
| EXTENT_DO_ACCOUNTING
,
796 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
797 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
802 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
805 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
806 struct extent_map
*em
;
809 read_lock(&em_tree
->lock
);
810 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
813 * if block start isn't an actual block number then find the
814 * first block in this inode and use that as a hint. If that
815 * block is also bogus then just don't worry about it.
817 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
819 em
= search_extent_mapping(em_tree
, 0, 0);
820 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
821 alloc_hint
= em
->block_start
;
825 alloc_hint
= em
->block_start
;
829 read_unlock(&em_tree
->lock
);
835 * when extent_io.c finds a delayed allocation range in the file,
836 * the call backs end up in this code. The basic idea is to
837 * allocate extents on disk for the range, and create ordered data structs
838 * in ram to track those extents.
840 * locked_page is the page that writepage had locked already. We use
841 * it to make sure we don't do extra locks or unlocks.
843 * *page_started is set to one if we unlock locked_page and do everything
844 * required to start IO on it. It may be clean and already done with
847 static noinline
int cow_file_range(struct inode
*inode
,
848 struct page
*locked_page
,
849 u64 start
, u64 end
, int *page_started
,
850 unsigned long *nr_written
,
853 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
856 unsigned long ram_size
;
859 u64 blocksize
= root
->sectorsize
;
860 struct btrfs_key ins
;
861 struct extent_map
*em
;
862 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
865 if (btrfs_is_free_space_inode(inode
)) {
871 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
872 num_bytes
= max(blocksize
, num_bytes
);
873 disk_num_bytes
= num_bytes
;
875 /* if this is a small write inside eof, kick off defrag */
876 if (num_bytes
< 64 * 1024 &&
877 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
878 btrfs_add_inode_defrag(NULL
, inode
);
881 /* lets try to make an inline extent */
882 ret
= cow_file_range_inline(root
, inode
, start
, end
, 0, 0,
885 extent_clear_unlock_delalloc(inode
, start
, end
, NULL
,
886 EXTENT_LOCKED
| EXTENT_DELALLOC
|
887 EXTENT_DEFRAG
, PAGE_UNLOCK
|
888 PAGE_CLEAR_DIRTY
| PAGE_SET_WRITEBACK
|
891 *nr_written
= *nr_written
+
892 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
895 } else if (ret
< 0) {
900 BUG_ON(disk_num_bytes
>
901 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
903 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
904 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
906 while (disk_num_bytes
> 0) {
909 cur_alloc_size
= disk_num_bytes
;
910 ret
= btrfs_reserve_extent(root
, cur_alloc_size
,
911 root
->sectorsize
, 0, alloc_hint
,
916 em
= alloc_extent_map();
922 em
->orig_start
= em
->start
;
923 ram_size
= ins
.offset
;
924 em
->len
= ins
.offset
;
925 em
->mod_start
= em
->start
;
926 em
->mod_len
= em
->len
;
928 em
->block_start
= ins
.objectid
;
929 em
->block_len
= ins
.offset
;
930 em
->orig_block_len
= ins
.offset
;
931 em
->ram_bytes
= ram_size
;
932 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
933 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
937 write_lock(&em_tree
->lock
);
938 ret
= add_extent_mapping(em_tree
, em
, 1);
939 write_unlock(&em_tree
->lock
);
940 if (ret
!= -EEXIST
) {
944 btrfs_drop_extent_cache(inode
, start
,
945 start
+ ram_size
- 1, 0);
950 cur_alloc_size
= ins
.offset
;
951 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
952 ram_size
, cur_alloc_size
, 0);
956 if (root
->root_key
.objectid
==
957 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
958 ret
= btrfs_reloc_clone_csums(inode
, start
,
964 if (disk_num_bytes
< cur_alloc_size
)
967 /* we're not doing compressed IO, don't unlock the first
968 * page (which the caller expects to stay locked), don't
969 * clear any dirty bits and don't set any writeback bits
971 * Do set the Private2 bit so we know this page was properly
972 * setup for writepage
974 op
= unlock
? PAGE_UNLOCK
: 0;
975 op
|= PAGE_SET_PRIVATE2
;
977 extent_clear_unlock_delalloc(inode
, start
,
978 start
+ ram_size
- 1, locked_page
,
979 EXTENT_LOCKED
| EXTENT_DELALLOC
,
981 disk_num_bytes
-= cur_alloc_size
;
982 num_bytes
-= cur_alloc_size
;
983 alloc_hint
= ins
.objectid
+ ins
.offset
;
984 start
+= cur_alloc_size
;
990 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
992 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
993 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
994 EXTENT_DELALLOC
| EXTENT_DEFRAG
,
995 PAGE_UNLOCK
| PAGE_CLEAR_DIRTY
|
996 PAGE_SET_WRITEBACK
| PAGE_END_WRITEBACK
);
1001 * work queue call back to started compression on a file and pages
1003 static noinline
void async_cow_start(struct btrfs_work
*work
)
1005 struct async_cow
*async_cow
;
1007 async_cow
= container_of(work
, struct async_cow
, work
);
1009 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1010 async_cow
->start
, async_cow
->end
, async_cow
,
1012 if (num_added
== 0) {
1013 btrfs_add_delayed_iput(async_cow
->inode
);
1014 async_cow
->inode
= NULL
;
1019 * work queue call back to submit previously compressed pages
1021 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1023 struct async_cow
*async_cow
;
1024 struct btrfs_root
*root
;
1025 unsigned long nr_pages
;
1027 async_cow
= container_of(work
, struct async_cow
, work
);
1029 root
= async_cow
->root
;
1030 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1033 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1035 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1036 wake_up(&root
->fs_info
->async_submit_wait
);
1038 if (async_cow
->inode
)
1039 submit_compressed_extents(async_cow
->inode
, async_cow
);
1042 static noinline
void async_cow_free(struct btrfs_work
*work
)
1044 struct async_cow
*async_cow
;
1045 async_cow
= container_of(work
, struct async_cow
, work
);
1046 if (async_cow
->inode
)
1047 btrfs_add_delayed_iput(async_cow
->inode
);
1051 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1052 u64 start
, u64 end
, int *page_started
,
1053 unsigned long *nr_written
)
1055 struct async_cow
*async_cow
;
1056 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1057 unsigned long nr_pages
;
1059 int limit
= 10 * 1024 * 1024;
1061 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1062 1, 0, NULL
, GFP_NOFS
);
1063 while (start
< end
) {
1064 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1065 BUG_ON(!async_cow
); /* -ENOMEM */
1066 async_cow
->inode
= igrab(inode
);
1067 async_cow
->root
= root
;
1068 async_cow
->locked_page
= locked_page
;
1069 async_cow
->start
= start
;
1071 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1074 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1076 async_cow
->end
= cur_end
;
1077 INIT_LIST_HEAD(&async_cow
->extents
);
1079 async_cow
->work
.func
= async_cow_start
;
1080 async_cow
->work
.ordered_func
= async_cow_submit
;
1081 async_cow
->work
.ordered_free
= async_cow_free
;
1082 async_cow
->work
.flags
= 0;
1084 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1086 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1088 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1091 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1092 wait_event(root
->fs_info
->async_submit_wait
,
1093 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1097 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1098 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1099 wait_event(root
->fs_info
->async_submit_wait
,
1100 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1104 *nr_written
+= nr_pages
;
1105 start
= cur_end
+ 1;
1111 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1112 u64 bytenr
, u64 num_bytes
)
1115 struct btrfs_ordered_sum
*sums
;
1118 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1119 bytenr
+ num_bytes
- 1, &list
, 0);
1120 if (ret
== 0 && list_empty(&list
))
1123 while (!list_empty(&list
)) {
1124 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1125 list_del(&sums
->list
);
1132 * when nowcow writeback call back. This checks for snapshots or COW copies
1133 * of the extents that exist in the file, and COWs the file as required.
1135 * If no cow copies or snapshots exist, we write directly to the existing
1138 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1139 struct page
*locked_page
,
1140 u64 start
, u64 end
, int *page_started
, int force
,
1141 unsigned long *nr_written
)
1143 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1144 struct btrfs_trans_handle
*trans
;
1145 struct extent_buffer
*leaf
;
1146 struct btrfs_path
*path
;
1147 struct btrfs_file_extent_item
*fi
;
1148 struct btrfs_key found_key
;
1163 u64 ino
= btrfs_ino(inode
);
1165 path
= btrfs_alloc_path();
1167 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1168 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1169 EXTENT_DO_ACCOUNTING
|
1170 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1172 PAGE_SET_WRITEBACK
|
1173 PAGE_END_WRITEBACK
);
1177 nolock
= btrfs_is_free_space_inode(inode
);
1180 trans
= btrfs_join_transaction_nolock(root
);
1182 trans
= btrfs_join_transaction(root
);
1184 if (IS_ERR(trans
)) {
1185 extent_clear_unlock_delalloc(inode
, start
, end
, locked_page
,
1186 EXTENT_LOCKED
| EXTENT_DELALLOC
|
1187 EXTENT_DO_ACCOUNTING
|
1188 EXTENT_DEFRAG
, PAGE_UNLOCK
|
1190 PAGE_SET_WRITEBACK
|
1191 PAGE_END_WRITEBACK
);
1192 btrfs_free_path(path
);
1193 return PTR_ERR(trans
);
1196 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1198 cow_start
= (u64
)-1;
1201 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1205 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1206 leaf
= path
->nodes
[0];
1207 btrfs_item_key_to_cpu(leaf
, &found_key
,
1208 path
->slots
[0] - 1);
1209 if (found_key
.objectid
== ino
&&
1210 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1215 leaf
= path
->nodes
[0];
1216 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1217 ret
= btrfs_next_leaf(root
, path
);
1222 leaf
= path
->nodes
[0];
1228 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1230 if (found_key
.objectid
> ino
||
1231 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1232 found_key
.offset
> end
)
1235 if (found_key
.offset
> cur_offset
) {
1236 extent_end
= found_key
.offset
;
1241 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1242 struct btrfs_file_extent_item
);
1243 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1245 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1246 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1247 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1248 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1249 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1250 extent_end
= found_key
.offset
+
1251 btrfs_file_extent_num_bytes(leaf
, fi
);
1253 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1254 if (extent_end
<= start
) {
1258 if (disk_bytenr
== 0)
1260 if (btrfs_file_extent_compression(leaf
, fi
) ||
1261 btrfs_file_extent_encryption(leaf
, fi
) ||
1262 btrfs_file_extent_other_encoding(leaf
, fi
))
1264 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1266 if (btrfs_extent_readonly(root
, disk_bytenr
))
1268 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1270 extent_offset
, disk_bytenr
))
1272 disk_bytenr
+= extent_offset
;
1273 disk_bytenr
+= cur_offset
- found_key
.offset
;
1274 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1276 * force cow if csum exists in the range.
1277 * this ensure that csum for a given extent are
1278 * either valid or do not exist.
1280 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1283 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1284 extent_end
= found_key
.offset
+
1285 btrfs_file_extent_inline_len(leaf
,
1286 path
->slots
[0], fi
);
1287 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1292 if (extent_end
<= start
) {
1297 if (cow_start
== (u64
)-1)
1298 cow_start
= cur_offset
;
1299 cur_offset
= extent_end
;
1300 if (cur_offset
> end
)
1306 btrfs_release_path(path
);
1307 if (cow_start
!= (u64
)-1) {
1308 ret
= cow_file_range(inode
, locked_page
,
1309 cow_start
, found_key
.offset
- 1,
1310 page_started
, nr_written
, 1);
1313 cow_start
= (u64
)-1;
1316 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1317 struct extent_map
*em
;
1318 struct extent_map_tree
*em_tree
;
1319 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1320 em
= alloc_extent_map();
1321 BUG_ON(!em
); /* -ENOMEM */
1322 em
->start
= cur_offset
;
1323 em
->orig_start
= found_key
.offset
- extent_offset
;
1324 em
->len
= num_bytes
;
1325 em
->block_len
= num_bytes
;
1326 em
->block_start
= disk_bytenr
;
1327 em
->orig_block_len
= disk_num_bytes
;
1328 em
->ram_bytes
= ram_bytes
;
1329 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1330 em
->mod_start
= em
->start
;
1331 em
->mod_len
= em
->len
;
1332 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1333 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1334 em
->generation
= -1;
1336 write_lock(&em_tree
->lock
);
1337 ret
= add_extent_mapping(em_tree
, em
, 1);
1338 write_unlock(&em_tree
->lock
);
1339 if (ret
!= -EEXIST
) {
1340 free_extent_map(em
);
1343 btrfs_drop_extent_cache(inode
, em
->start
,
1344 em
->start
+ em
->len
- 1, 0);
1346 type
= BTRFS_ORDERED_PREALLOC
;
1348 type
= BTRFS_ORDERED_NOCOW
;
1351 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1352 num_bytes
, num_bytes
, type
);
1353 BUG_ON(ret
); /* -ENOMEM */
1355 if (root
->root_key
.objectid
==
1356 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1357 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1363 extent_clear_unlock_delalloc(inode
, cur_offset
,
1364 cur_offset
+ num_bytes
- 1,
1365 locked_page
, EXTENT_LOCKED
|
1366 EXTENT_DELALLOC
, PAGE_UNLOCK
|
1368 cur_offset
= extent_end
;
1369 if (cur_offset
> end
)
1372 btrfs_release_path(path
);
1374 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1375 cow_start
= cur_offset
;
1379 if (cow_start
!= (u64
)-1) {
1380 ret
= cow_file_range(inode
, locked_page
, cow_start
, end
,
1381 page_started
, nr_written
, 1);
1387 err
= btrfs_end_transaction(trans
, root
);
1391 if (ret
&& cur_offset
< end
)
1392 extent_clear_unlock_delalloc(inode
, cur_offset
, end
,
1393 locked_page
, EXTENT_LOCKED
|
1394 EXTENT_DELALLOC
| EXTENT_DEFRAG
|
1395 EXTENT_DO_ACCOUNTING
, PAGE_UNLOCK
|
1397 PAGE_SET_WRITEBACK
|
1398 PAGE_END_WRITEBACK
);
1399 btrfs_free_path(path
);
1404 * extent_io.c call back to do delayed allocation processing
1406 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1407 u64 start
, u64 end
, int *page_started
,
1408 unsigned long *nr_written
)
1411 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1413 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1414 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1415 page_started
, 1, nr_written
);
1416 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1417 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1418 page_started
, 0, nr_written
);
1419 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1420 !(BTRFS_I(inode
)->force_compress
) &&
1421 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1422 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1423 page_started
, nr_written
, 1);
1425 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1426 &BTRFS_I(inode
)->runtime_flags
);
1427 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1428 page_started
, nr_written
);
1433 static void btrfs_split_extent_hook(struct inode
*inode
,
1434 struct extent_state
*orig
, u64 split
)
1436 /* not delalloc, ignore it */
1437 if (!(orig
->state
& EXTENT_DELALLOC
))
1440 spin_lock(&BTRFS_I(inode
)->lock
);
1441 BTRFS_I(inode
)->outstanding_extents
++;
1442 spin_unlock(&BTRFS_I(inode
)->lock
);
1446 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1447 * extents so we can keep track of new extents that are just merged onto old
1448 * extents, such as when we are doing sequential writes, so we can properly
1449 * account for the metadata space we'll need.
1451 static void btrfs_merge_extent_hook(struct inode
*inode
,
1452 struct extent_state
*new,
1453 struct extent_state
*other
)
1455 /* not delalloc, ignore it */
1456 if (!(other
->state
& EXTENT_DELALLOC
))
1459 spin_lock(&BTRFS_I(inode
)->lock
);
1460 BTRFS_I(inode
)->outstanding_extents
--;
1461 spin_unlock(&BTRFS_I(inode
)->lock
);
1464 static void btrfs_add_delalloc_inodes(struct btrfs_root
*root
,
1465 struct inode
*inode
)
1467 spin_lock(&root
->delalloc_lock
);
1468 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1469 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1470 &root
->delalloc_inodes
);
1471 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1472 &BTRFS_I(inode
)->runtime_flags
);
1473 root
->nr_delalloc_inodes
++;
1474 if (root
->nr_delalloc_inodes
== 1) {
1475 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1476 BUG_ON(!list_empty(&root
->delalloc_root
));
1477 list_add_tail(&root
->delalloc_root
,
1478 &root
->fs_info
->delalloc_roots
);
1479 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1482 spin_unlock(&root
->delalloc_lock
);
1485 static void btrfs_del_delalloc_inode(struct btrfs_root
*root
,
1486 struct inode
*inode
)
1488 spin_lock(&root
->delalloc_lock
);
1489 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1490 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1491 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1492 &BTRFS_I(inode
)->runtime_flags
);
1493 root
->nr_delalloc_inodes
--;
1494 if (!root
->nr_delalloc_inodes
) {
1495 spin_lock(&root
->fs_info
->delalloc_root_lock
);
1496 BUG_ON(list_empty(&root
->delalloc_root
));
1497 list_del_init(&root
->delalloc_root
);
1498 spin_unlock(&root
->fs_info
->delalloc_root_lock
);
1501 spin_unlock(&root
->delalloc_lock
);
1505 * extent_io.c set_bit_hook, used to track delayed allocation
1506 * bytes in this file, and to maintain the list of inodes that
1507 * have pending delalloc work to be done.
1509 static void btrfs_set_bit_hook(struct inode
*inode
,
1510 struct extent_state
*state
, unsigned long *bits
)
1514 * set_bit and clear bit hooks normally require _irqsave/restore
1515 * but in this case, we are only testing for the DELALLOC
1516 * bit, which is only set or cleared with irqs on
1518 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1519 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1520 u64 len
= state
->end
+ 1 - state
->start
;
1521 bool do_list
= !btrfs_is_free_space_inode(inode
);
1523 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1524 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1526 spin_lock(&BTRFS_I(inode
)->lock
);
1527 BTRFS_I(inode
)->outstanding_extents
++;
1528 spin_unlock(&BTRFS_I(inode
)->lock
);
1531 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1532 root
->fs_info
->delalloc_batch
);
1533 spin_lock(&BTRFS_I(inode
)->lock
);
1534 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1535 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1536 &BTRFS_I(inode
)->runtime_flags
))
1537 btrfs_add_delalloc_inodes(root
, inode
);
1538 spin_unlock(&BTRFS_I(inode
)->lock
);
1543 * extent_io.c clear_bit_hook, see set_bit_hook for why
1545 static void btrfs_clear_bit_hook(struct inode
*inode
,
1546 struct extent_state
*state
,
1547 unsigned long *bits
)
1550 * set_bit and clear bit hooks normally require _irqsave/restore
1551 * but in this case, we are only testing for the DELALLOC
1552 * bit, which is only set or cleared with irqs on
1554 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1555 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1556 u64 len
= state
->end
+ 1 - state
->start
;
1557 bool do_list
= !btrfs_is_free_space_inode(inode
);
1559 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1560 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1561 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1562 spin_lock(&BTRFS_I(inode
)->lock
);
1563 BTRFS_I(inode
)->outstanding_extents
--;
1564 spin_unlock(&BTRFS_I(inode
)->lock
);
1568 * We don't reserve metadata space for space cache inodes so we
1569 * don't need to call dellalloc_release_metadata if there is an
1572 if (*bits
& EXTENT_DO_ACCOUNTING
&&
1573 root
!= root
->fs_info
->tree_root
)
1574 btrfs_delalloc_release_metadata(inode
, len
);
1576 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1577 && do_list
&& !(state
->state
& EXTENT_NORESERVE
))
1578 btrfs_free_reserved_data_space(inode
, len
);
1580 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1581 root
->fs_info
->delalloc_batch
);
1582 spin_lock(&BTRFS_I(inode
)->lock
);
1583 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1584 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1585 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1586 &BTRFS_I(inode
)->runtime_flags
))
1587 btrfs_del_delalloc_inode(root
, inode
);
1588 spin_unlock(&BTRFS_I(inode
)->lock
);
1593 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1594 * we don't create bios that span stripes or chunks
1596 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1597 size_t size
, struct bio
*bio
,
1598 unsigned long bio_flags
)
1600 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1601 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1606 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1609 length
= bio
->bi_size
;
1610 map_length
= length
;
1611 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1612 &map_length
, NULL
, 0);
1613 /* Will always return 0 with map_multi == NULL */
1615 if (map_length
< length
+ size
)
1621 * in order to insert checksums into the metadata in large chunks,
1622 * we wait until bio submission time. All the pages in the bio are
1623 * checksummed and sums are attached onto the ordered extent record.
1625 * At IO completion time the cums attached on the ordered extent record
1626 * are inserted into the btree
1628 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1629 struct bio
*bio
, int mirror_num
,
1630 unsigned long bio_flags
,
1633 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1636 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1637 BUG_ON(ret
); /* -ENOMEM */
1642 * in order to insert checksums into the metadata in large chunks,
1643 * we wait until bio submission time. All the pages in the bio are
1644 * checksummed and sums are attached onto the ordered extent record.
1646 * At IO completion time the cums attached on the ordered extent record
1647 * are inserted into the btree
1649 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1650 int mirror_num
, unsigned long bio_flags
,
1653 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1656 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1658 bio_endio(bio
, ret
);
1663 * extent_io.c submission hook. This does the right thing for csum calculation
1664 * on write, or reading the csums from the tree before a read
1666 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1667 int mirror_num
, unsigned long bio_flags
,
1670 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1674 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1676 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1678 if (btrfs_is_free_space_inode(inode
))
1681 if (!(rw
& REQ_WRITE
)) {
1682 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1686 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1687 ret
= btrfs_submit_compressed_read(inode
, bio
,
1691 } else if (!skip_sum
) {
1692 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1697 } else if (async
&& !skip_sum
) {
1698 /* csum items have already been cloned */
1699 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1701 /* we're doing a write, do the async checksumming */
1702 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1703 inode
, rw
, bio
, mirror_num
,
1704 bio_flags
, bio_offset
,
1705 __btrfs_submit_bio_start
,
1706 __btrfs_submit_bio_done
);
1708 } else if (!skip_sum
) {
1709 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1715 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1719 bio_endio(bio
, ret
);
1724 * given a list of ordered sums record them in the inode. This happens
1725 * at IO completion time based on sums calculated at bio submission time.
1727 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1728 struct inode
*inode
, u64 file_offset
,
1729 struct list_head
*list
)
1731 struct btrfs_ordered_sum
*sum
;
1733 list_for_each_entry(sum
, list
, list
) {
1734 trans
->adding_csums
= 1;
1735 btrfs_csum_file_blocks(trans
,
1736 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1737 trans
->adding_csums
= 0;
1742 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1743 struct extent_state
**cached_state
)
1745 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1746 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1747 cached_state
, GFP_NOFS
);
1750 /* see btrfs_writepage_start_hook for details on why this is required */
1751 struct btrfs_writepage_fixup
{
1753 struct btrfs_work work
;
1756 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1758 struct btrfs_writepage_fixup
*fixup
;
1759 struct btrfs_ordered_extent
*ordered
;
1760 struct extent_state
*cached_state
= NULL
;
1762 struct inode
*inode
;
1767 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1771 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1772 ClearPageChecked(page
);
1776 inode
= page
->mapping
->host
;
1777 page_start
= page_offset(page
);
1778 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1780 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1783 /* already ordered? We're done */
1784 if (PagePrivate2(page
))
1787 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1789 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1790 page_end
, &cached_state
, GFP_NOFS
);
1792 btrfs_start_ordered_extent(inode
, ordered
, 1);
1793 btrfs_put_ordered_extent(ordered
);
1797 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1799 mapping_set_error(page
->mapping
, ret
);
1800 end_extent_writepage(page
, ret
, page_start
, page_end
);
1801 ClearPageChecked(page
);
1805 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1806 ClearPageChecked(page
);
1807 set_page_dirty(page
);
1809 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1810 &cached_state
, GFP_NOFS
);
1813 page_cache_release(page
);
1818 * There are a few paths in the higher layers of the kernel that directly
1819 * set the page dirty bit without asking the filesystem if it is a
1820 * good idea. This causes problems because we want to make sure COW
1821 * properly happens and the data=ordered rules are followed.
1823 * In our case any range that doesn't have the ORDERED bit set
1824 * hasn't been properly setup for IO. We kick off an async process
1825 * to fix it up. The async helper will wait for ordered extents, set
1826 * the delalloc bit and make it safe to write the page.
1828 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1830 struct inode
*inode
= page
->mapping
->host
;
1831 struct btrfs_writepage_fixup
*fixup
;
1832 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1834 /* this page is properly in the ordered list */
1835 if (TestClearPagePrivate2(page
))
1838 if (PageChecked(page
))
1841 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1845 SetPageChecked(page
);
1846 page_cache_get(page
);
1847 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1849 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1853 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1854 struct inode
*inode
, u64 file_pos
,
1855 u64 disk_bytenr
, u64 disk_num_bytes
,
1856 u64 num_bytes
, u64 ram_bytes
,
1857 u8 compression
, u8 encryption
,
1858 u16 other_encoding
, int extent_type
)
1860 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1861 struct btrfs_file_extent_item
*fi
;
1862 struct btrfs_path
*path
;
1863 struct extent_buffer
*leaf
;
1864 struct btrfs_key ins
;
1865 int extent_inserted
= 0;
1868 path
= btrfs_alloc_path();
1873 * we may be replacing one extent in the tree with another.
1874 * The new extent is pinned in the extent map, and we don't want
1875 * to drop it from the cache until it is completely in the btree.
1877 * So, tell btrfs_drop_extents to leave this extent in the cache.
1878 * the caller is expected to unpin it and allow it to be merged
1881 ret
= __btrfs_drop_extents(trans
, root
, inode
, path
, file_pos
,
1882 file_pos
+ num_bytes
, NULL
, 0,
1883 1, sizeof(*fi
), &extent_inserted
);
1887 if (!extent_inserted
) {
1888 ins
.objectid
= btrfs_ino(inode
);
1889 ins
.offset
= file_pos
;
1890 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1892 path
->leave_spinning
= 1;
1893 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
,
1898 leaf
= path
->nodes
[0];
1899 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1900 struct btrfs_file_extent_item
);
1901 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1902 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1903 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1904 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1905 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1906 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1907 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1908 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1909 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1910 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1912 btrfs_mark_buffer_dirty(leaf
);
1913 btrfs_release_path(path
);
1915 inode_add_bytes(inode
, num_bytes
);
1917 ins
.objectid
= disk_bytenr
;
1918 ins
.offset
= disk_num_bytes
;
1919 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1920 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1921 root
->root_key
.objectid
,
1922 btrfs_ino(inode
), file_pos
, &ins
);
1924 btrfs_free_path(path
);
1929 /* snapshot-aware defrag */
1930 struct sa_defrag_extent_backref
{
1931 struct rb_node node
;
1932 struct old_sa_defrag_extent
*old
;
1941 struct old_sa_defrag_extent
{
1942 struct list_head list
;
1943 struct new_sa_defrag_extent
*new;
1952 struct new_sa_defrag_extent
{
1953 struct rb_root root
;
1954 struct list_head head
;
1955 struct btrfs_path
*path
;
1956 struct inode
*inode
;
1964 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
1965 struct sa_defrag_extent_backref
*b2
)
1967 if (b1
->root_id
< b2
->root_id
)
1969 else if (b1
->root_id
> b2
->root_id
)
1972 if (b1
->inum
< b2
->inum
)
1974 else if (b1
->inum
> b2
->inum
)
1977 if (b1
->file_pos
< b2
->file_pos
)
1979 else if (b1
->file_pos
> b2
->file_pos
)
1983 * [------------------------------] ===> (a range of space)
1984 * |<--->| |<---->| =============> (fs/file tree A)
1985 * |<---------------------------->| ===> (fs/file tree B)
1987 * A range of space can refer to two file extents in one tree while
1988 * refer to only one file extent in another tree.
1990 * So we may process a disk offset more than one time(two extents in A)
1991 * and locate at the same extent(one extent in B), then insert two same
1992 * backrefs(both refer to the extent in B).
1997 static void backref_insert(struct rb_root
*root
,
1998 struct sa_defrag_extent_backref
*backref
)
2000 struct rb_node
**p
= &root
->rb_node
;
2001 struct rb_node
*parent
= NULL
;
2002 struct sa_defrag_extent_backref
*entry
;
2007 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2009 ret
= backref_comp(backref
, entry
);
2013 p
= &(*p
)->rb_right
;
2016 rb_link_node(&backref
->node
, parent
, p
);
2017 rb_insert_color(&backref
->node
, root
);
2021 * Note the backref might has changed, and in this case we just return 0.
2023 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2026 struct btrfs_file_extent_item
*extent
;
2027 struct btrfs_fs_info
*fs_info
;
2028 struct old_sa_defrag_extent
*old
= ctx
;
2029 struct new_sa_defrag_extent
*new = old
->new;
2030 struct btrfs_path
*path
= new->path
;
2031 struct btrfs_key key
;
2032 struct btrfs_root
*root
;
2033 struct sa_defrag_extent_backref
*backref
;
2034 struct extent_buffer
*leaf
;
2035 struct inode
*inode
= new->inode
;
2041 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2042 inum
== btrfs_ino(inode
))
2045 key
.objectid
= root_id
;
2046 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2047 key
.offset
= (u64
)-1;
2049 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2050 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2052 if (PTR_ERR(root
) == -ENOENT
)
2055 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2056 inum
, offset
, root_id
);
2057 return PTR_ERR(root
);
2060 key
.objectid
= inum
;
2061 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2062 if (offset
> (u64
)-1 << 32)
2065 key
.offset
= offset
;
2067 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2068 if (WARN_ON(ret
< 0))
2075 leaf
= path
->nodes
[0];
2076 slot
= path
->slots
[0];
2078 if (slot
>= btrfs_header_nritems(leaf
)) {
2079 ret
= btrfs_next_leaf(root
, path
);
2082 } else if (ret
> 0) {
2091 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2093 if (key
.objectid
> inum
)
2096 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2099 extent
= btrfs_item_ptr(leaf
, slot
,
2100 struct btrfs_file_extent_item
);
2102 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2106 * 'offset' refers to the exact key.offset,
2107 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2108 * (key.offset - extent_offset).
2110 if (key
.offset
!= offset
)
2113 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2114 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2116 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2117 old
->len
|| extent_offset
+ num_bytes
<=
2118 old
->extent_offset
+ old
->offset
)
2123 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2129 backref
->root_id
= root_id
;
2130 backref
->inum
= inum
;
2131 backref
->file_pos
= offset
;
2132 backref
->num_bytes
= num_bytes
;
2133 backref
->extent_offset
= extent_offset
;
2134 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2136 backref_insert(&new->root
, backref
);
2139 btrfs_release_path(path
);
2144 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2145 struct new_sa_defrag_extent
*new)
2147 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2148 struct old_sa_defrag_extent
*old
, *tmp
;
2153 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2154 ret
= iterate_inodes_from_logical(old
->bytenr
+
2155 old
->extent_offset
, fs_info
,
2156 path
, record_one_backref
,
2158 if (ret
< 0 && ret
!= -ENOENT
)
2161 /* no backref to be processed for this extent */
2163 list_del(&old
->list
);
2168 if (list_empty(&new->head
))
2174 static int relink_is_mergable(struct extent_buffer
*leaf
,
2175 struct btrfs_file_extent_item
*fi
,
2176 struct new_sa_defrag_extent
*new)
2178 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != new->bytenr
)
2181 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2184 if (btrfs_file_extent_compression(leaf
, fi
) != new->compress_type
)
2187 if (btrfs_file_extent_encryption(leaf
, fi
) ||
2188 btrfs_file_extent_other_encoding(leaf
, fi
))
2195 * Note the backref might has changed, and in this case we just return 0.
2197 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2198 struct sa_defrag_extent_backref
*prev
,
2199 struct sa_defrag_extent_backref
*backref
)
2201 struct btrfs_file_extent_item
*extent
;
2202 struct btrfs_file_extent_item
*item
;
2203 struct btrfs_ordered_extent
*ordered
;
2204 struct btrfs_trans_handle
*trans
;
2205 struct btrfs_fs_info
*fs_info
;
2206 struct btrfs_root
*root
;
2207 struct btrfs_key key
;
2208 struct extent_buffer
*leaf
;
2209 struct old_sa_defrag_extent
*old
= backref
->old
;
2210 struct new_sa_defrag_extent
*new = old
->new;
2211 struct inode
*src_inode
= new->inode
;
2212 struct inode
*inode
;
2213 struct extent_state
*cached
= NULL
;
2222 if (prev
&& prev
->root_id
== backref
->root_id
&&
2223 prev
->inum
== backref
->inum
&&
2224 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2227 /* step 1: get root */
2228 key
.objectid
= backref
->root_id
;
2229 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2230 key
.offset
= (u64
)-1;
2232 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2233 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2235 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2237 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2238 if (PTR_ERR(root
) == -ENOENT
)
2240 return PTR_ERR(root
);
2243 if (btrfs_root_readonly(root
)) {
2244 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2248 /* step 2: get inode */
2249 key
.objectid
= backref
->inum
;
2250 key
.type
= BTRFS_INODE_ITEM_KEY
;
2253 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2254 if (IS_ERR(inode
)) {
2255 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2259 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2261 /* step 3: relink backref */
2262 lock_start
= backref
->file_pos
;
2263 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2264 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2267 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2269 btrfs_put_ordered_extent(ordered
);
2273 trans
= btrfs_join_transaction(root
);
2274 if (IS_ERR(trans
)) {
2275 ret
= PTR_ERR(trans
);
2279 key
.objectid
= backref
->inum
;
2280 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2281 key
.offset
= backref
->file_pos
;
2283 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2286 } else if (ret
> 0) {
2291 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2292 struct btrfs_file_extent_item
);
2294 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2295 backref
->generation
)
2298 btrfs_release_path(path
);
2300 start
= backref
->file_pos
;
2301 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2302 start
+= old
->extent_offset
+ old
->offset
-
2303 backref
->extent_offset
;
2305 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2306 old
->extent_offset
+ old
->offset
+ old
->len
);
2307 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2309 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2314 key
.objectid
= btrfs_ino(inode
);
2315 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2318 path
->leave_spinning
= 1;
2320 struct btrfs_file_extent_item
*fi
;
2322 struct btrfs_key found_key
;
2324 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
2329 leaf
= path
->nodes
[0];
2330 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2332 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2333 struct btrfs_file_extent_item
);
2334 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2336 if (extent_len
+ found_key
.offset
== start
&&
2337 relink_is_mergable(leaf
, fi
, new)) {
2338 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2340 btrfs_mark_buffer_dirty(leaf
);
2341 inode_add_bytes(inode
, len
);
2347 btrfs_release_path(path
);
2352 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2355 btrfs_abort_transaction(trans
, root
, ret
);
2359 leaf
= path
->nodes
[0];
2360 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2361 struct btrfs_file_extent_item
);
2362 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2363 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2364 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2365 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2366 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2367 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2368 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2369 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2370 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2371 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2373 btrfs_mark_buffer_dirty(leaf
);
2374 inode_add_bytes(inode
, len
);
2375 btrfs_release_path(path
);
2377 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2379 backref
->root_id
, backref
->inum
,
2380 new->file_pos
, 0); /* start - extent_offset */
2382 btrfs_abort_transaction(trans
, root
, ret
);
2388 btrfs_release_path(path
);
2389 path
->leave_spinning
= 0;
2390 btrfs_end_transaction(trans
, root
);
2392 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2398 static void free_sa_defrag_extent(struct new_sa_defrag_extent
*new)
2400 struct old_sa_defrag_extent
*old
, *tmp
;
2405 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2406 list_del(&old
->list
);
2412 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2414 struct btrfs_path
*path
;
2415 struct sa_defrag_extent_backref
*backref
;
2416 struct sa_defrag_extent_backref
*prev
= NULL
;
2417 struct inode
*inode
;
2418 struct btrfs_root
*root
;
2419 struct rb_node
*node
;
2423 root
= BTRFS_I(inode
)->root
;
2425 path
= btrfs_alloc_path();
2429 if (!record_extent_backrefs(path
, new)) {
2430 btrfs_free_path(path
);
2433 btrfs_release_path(path
);
2436 node
= rb_first(&new->root
);
2439 rb_erase(node
, &new->root
);
2441 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2443 ret
= relink_extent_backref(path
, prev
, backref
);
2456 btrfs_free_path(path
);
2458 free_sa_defrag_extent(new);
2460 atomic_dec(&root
->fs_info
->defrag_running
);
2461 wake_up(&root
->fs_info
->transaction_wait
);
2464 static struct new_sa_defrag_extent
*
2465 record_old_file_extents(struct inode
*inode
,
2466 struct btrfs_ordered_extent
*ordered
)
2468 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2469 struct btrfs_path
*path
;
2470 struct btrfs_key key
;
2471 struct old_sa_defrag_extent
*old
;
2472 struct new_sa_defrag_extent
*new;
2475 new = kmalloc(sizeof(*new), GFP_NOFS
);
2480 new->file_pos
= ordered
->file_offset
;
2481 new->len
= ordered
->len
;
2482 new->bytenr
= ordered
->start
;
2483 new->disk_len
= ordered
->disk_len
;
2484 new->compress_type
= ordered
->compress_type
;
2485 new->root
= RB_ROOT
;
2486 INIT_LIST_HEAD(&new->head
);
2488 path
= btrfs_alloc_path();
2492 key
.objectid
= btrfs_ino(inode
);
2493 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2494 key
.offset
= new->file_pos
;
2496 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2499 if (ret
> 0 && path
->slots
[0] > 0)
2502 /* find out all the old extents for the file range */
2504 struct btrfs_file_extent_item
*extent
;
2505 struct extent_buffer
*l
;
2514 slot
= path
->slots
[0];
2516 if (slot
>= btrfs_header_nritems(l
)) {
2517 ret
= btrfs_next_leaf(root
, path
);
2525 btrfs_item_key_to_cpu(l
, &key
, slot
);
2527 if (key
.objectid
!= btrfs_ino(inode
))
2529 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2531 if (key
.offset
>= new->file_pos
+ new->len
)
2534 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2536 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2537 if (key
.offset
+ num_bytes
< new->file_pos
)
2540 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2544 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2546 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2550 offset
= max(new->file_pos
, key
.offset
);
2551 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2553 old
->bytenr
= disk_bytenr
;
2554 old
->extent_offset
= extent_offset
;
2555 old
->offset
= offset
- key
.offset
;
2556 old
->len
= end
- offset
;
2559 list_add_tail(&old
->list
, &new->head
);
2565 btrfs_free_path(path
);
2566 atomic_inc(&root
->fs_info
->defrag_running
);
2571 btrfs_free_path(path
);
2573 free_sa_defrag_extent(new);
2577 /* as ordered data IO finishes, this gets called so we can finish
2578 * an ordered extent if the range of bytes in the file it covers are
2581 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2583 struct inode
*inode
= ordered_extent
->inode
;
2584 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2585 struct btrfs_trans_handle
*trans
= NULL
;
2586 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2587 struct extent_state
*cached_state
= NULL
;
2588 struct new_sa_defrag_extent
*new = NULL
;
2589 int compress_type
= 0;
2591 u64 logical_len
= ordered_extent
->len
;
2593 bool truncated
= false;
2595 nolock
= btrfs_is_free_space_inode(inode
);
2597 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2602 if (test_bit(BTRFS_ORDERED_TRUNCATED
, &ordered_extent
->flags
)) {
2604 logical_len
= ordered_extent
->truncated_len
;
2605 /* Truncated the entire extent, don't bother adding */
2610 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2611 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2612 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2614 trans
= btrfs_join_transaction_nolock(root
);
2616 trans
= btrfs_join_transaction(root
);
2617 if (IS_ERR(trans
)) {
2618 ret
= PTR_ERR(trans
);
2622 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2623 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2624 if (ret
) /* -ENOMEM or corruption */
2625 btrfs_abort_transaction(trans
, root
, ret
);
2629 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2630 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2633 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2634 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2635 EXTENT_DEFRAG
, 1, cached_state
);
2637 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2638 if (0 && last_snapshot
>= BTRFS_I(inode
)->generation
)
2639 /* the inode is shared */
2640 new = record_old_file_extents(inode
, ordered_extent
);
2642 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2643 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2644 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2648 trans
= btrfs_join_transaction_nolock(root
);
2650 trans
= btrfs_join_transaction(root
);
2651 if (IS_ERR(trans
)) {
2652 ret
= PTR_ERR(trans
);
2656 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2658 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2659 compress_type
= ordered_extent
->compress_type
;
2660 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2661 BUG_ON(compress_type
);
2662 ret
= btrfs_mark_extent_written(trans
, inode
,
2663 ordered_extent
->file_offset
,
2664 ordered_extent
->file_offset
+
2667 BUG_ON(root
== root
->fs_info
->tree_root
);
2668 ret
= insert_reserved_file_extent(trans
, inode
,
2669 ordered_extent
->file_offset
,
2670 ordered_extent
->start
,
2671 ordered_extent
->disk_len
,
2672 logical_len
, logical_len
,
2673 compress_type
, 0, 0,
2674 BTRFS_FILE_EXTENT_REG
);
2676 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2677 ordered_extent
->file_offset
, ordered_extent
->len
,
2680 btrfs_abort_transaction(trans
, root
, ret
);
2684 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2685 &ordered_extent
->list
);
2687 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2688 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2689 if (ret
) { /* -ENOMEM or corruption */
2690 btrfs_abort_transaction(trans
, root
, ret
);
2695 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2696 ordered_extent
->file_offset
+
2697 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2699 if (root
!= root
->fs_info
->tree_root
)
2700 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2702 btrfs_end_transaction(trans
, root
);
2704 if (ret
|| truncated
) {
2708 start
= ordered_extent
->file_offset
+ logical_len
;
2710 start
= ordered_extent
->file_offset
;
2711 end
= ordered_extent
->file_offset
+ ordered_extent
->len
- 1;
2712 clear_extent_uptodate(io_tree
, start
, end
, NULL
, GFP_NOFS
);
2714 /* Drop the cache for the part of the extent we didn't write. */
2715 btrfs_drop_extent_cache(inode
, start
, end
, 0);
2718 * If the ordered extent had an IOERR or something else went
2719 * wrong we need to return the space for this ordered extent
2720 * back to the allocator. We only free the extent in the
2721 * truncated case if we didn't write out the extent at all.
2723 if ((ret
|| !logical_len
) &&
2724 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2725 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2726 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2727 ordered_extent
->disk_len
);
2732 * This needs to be done to make sure anybody waiting knows we are done
2733 * updating everything for this ordered extent.
2735 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2737 /* for snapshot-aware defrag */
2740 free_sa_defrag_extent(new);
2741 atomic_dec(&root
->fs_info
->defrag_running
);
2743 relink_file_extents(new);
2748 btrfs_put_ordered_extent(ordered_extent
);
2749 /* once for the tree */
2750 btrfs_put_ordered_extent(ordered_extent
);
2755 static void finish_ordered_fn(struct btrfs_work
*work
)
2757 struct btrfs_ordered_extent
*ordered_extent
;
2758 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
2759 btrfs_finish_ordered_io(ordered_extent
);
2762 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2763 struct extent_state
*state
, int uptodate
)
2765 struct inode
*inode
= page
->mapping
->host
;
2766 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2767 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
2768 struct btrfs_workers
*workers
;
2770 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2772 ClearPagePrivate2(page
);
2773 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2774 end
- start
+ 1, uptodate
))
2777 ordered_extent
->work
.func
= finish_ordered_fn
;
2778 ordered_extent
->work
.flags
= 0;
2780 if (btrfs_is_free_space_inode(inode
))
2781 workers
= &root
->fs_info
->endio_freespace_worker
;
2783 workers
= &root
->fs_info
->endio_write_workers
;
2784 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2790 * when reads are done, we need to check csums to verify the data is correct
2791 * if there's a match, we allow the bio to finish. If not, the code in
2792 * extent_io.c will try to find good copies for us.
2794 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio
*io_bio
,
2795 u64 phy_offset
, struct page
*page
,
2796 u64 start
, u64 end
, int mirror
)
2798 size_t offset
= start
- page_offset(page
);
2799 struct inode
*inode
= page
->mapping
->host
;
2800 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2802 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2805 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2806 DEFAULT_RATELIMIT_BURST
);
2808 if (PageChecked(page
)) {
2809 ClearPageChecked(page
);
2813 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2816 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2817 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2818 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2823 phy_offset
>>= inode
->i_sb
->s_blocksize_bits
;
2824 csum_expected
= *(((u32
*)io_bio
->csum
) + phy_offset
);
2826 kaddr
= kmap_atomic(page
);
2827 csum
= btrfs_csum_data(kaddr
+ offset
, csum
, end
- start
+ 1);
2828 btrfs_csum_final(csum
, (char *)&csum
);
2829 if (csum
!= csum_expected
)
2832 kunmap_atomic(kaddr
);
2837 if (__ratelimit(&_rs
))
2838 btrfs_info(root
->fs_info
, "csum failed ino %llu off %llu csum %u expected csum %u",
2839 btrfs_ino(page
->mapping
->host
), start
, csum
, csum_expected
);
2840 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2841 flush_dcache_page(page
);
2842 kunmap_atomic(kaddr
);
2843 if (csum_expected
== 0)
2848 struct delayed_iput
{
2849 struct list_head list
;
2850 struct inode
*inode
;
2853 /* JDM: If this is fs-wide, why can't we add a pointer to
2854 * btrfs_inode instead and avoid the allocation? */
2855 void btrfs_add_delayed_iput(struct inode
*inode
)
2857 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2858 struct delayed_iput
*delayed
;
2860 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2863 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2864 delayed
->inode
= inode
;
2866 spin_lock(&fs_info
->delayed_iput_lock
);
2867 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2868 spin_unlock(&fs_info
->delayed_iput_lock
);
2871 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2874 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2875 struct delayed_iput
*delayed
;
2878 spin_lock(&fs_info
->delayed_iput_lock
);
2879 empty
= list_empty(&fs_info
->delayed_iputs
);
2880 spin_unlock(&fs_info
->delayed_iput_lock
);
2884 spin_lock(&fs_info
->delayed_iput_lock
);
2885 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2886 spin_unlock(&fs_info
->delayed_iput_lock
);
2888 while (!list_empty(&list
)) {
2889 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2890 list_del(&delayed
->list
);
2891 iput(delayed
->inode
);
2897 * This is called in transaction commit time. If there are no orphan
2898 * files in the subvolume, it removes orphan item and frees block_rsv
2901 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2902 struct btrfs_root
*root
)
2904 struct btrfs_block_rsv
*block_rsv
;
2907 if (atomic_read(&root
->orphan_inodes
) ||
2908 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2911 spin_lock(&root
->orphan_lock
);
2912 if (atomic_read(&root
->orphan_inodes
)) {
2913 spin_unlock(&root
->orphan_lock
);
2917 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2918 spin_unlock(&root
->orphan_lock
);
2922 block_rsv
= root
->orphan_block_rsv
;
2923 root
->orphan_block_rsv
= NULL
;
2924 spin_unlock(&root
->orphan_lock
);
2926 if (root
->orphan_item_inserted
&&
2927 btrfs_root_refs(&root
->root_item
) > 0) {
2928 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2929 root
->root_key
.objectid
);
2931 btrfs_abort_transaction(trans
, root
, ret
);
2933 root
->orphan_item_inserted
= 0;
2937 WARN_ON(block_rsv
->size
> 0);
2938 btrfs_free_block_rsv(root
, block_rsv
);
2943 * This creates an orphan entry for the given inode in case something goes
2944 * wrong in the middle of an unlink/truncate.
2946 * NOTE: caller of this function should reserve 5 units of metadata for
2949 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2951 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2952 struct btrfs_block_rsv
*block_rsv
= NULL
;
2957 if (!root
->orphan_block_rsv
) {
2958 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2963 spin_lock(&root
->orphan_lock
);
2964 if (!root
->orphan_block_rsv
) {
2965 root
->orphan_block_rsv
= block_rsv
;
2966 } else if (block_rsv
) {
2967 btrfs_free_block_rsv(root
, block_rsv
);
2971 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2972 &BTRFS_I(inode
)->runtime_flags
)) {
2975 * For proper ENOSPC handling, we should do orphan
2976 * cleanup when mounting. But this introduces backward
2977 * compatibility issue.
2979 if (!xchg(&root
->orphan_item_inserted
, 1))
2985 atomic_inc(&root
->orphan_inodes
);
2988 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2989 &BTRFS_I(inode
)->runtime_flags
))
2991 spin_unlock(&root
->orphan_lock
);
2993 /* grab metadata reservation from transaction handle */
2995 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
2996 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
2999 /* insert an orphan item to track this unlinked/truncated file */
3001 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3003 atomic_dec(&root
->orphan_inodes
);
3005 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3006 &BTRFS_I(inode
)->runtime_flags
);
3007 btrfs_orphan_release_metadata(inode
);
3009 if (ret
!= -EEXIST
) {
3010 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3011 &BTRFS_I(inode
)->runtime_flags
);
3012 btrfs_abort_transaction(trans
, root
, ret
);
3019 /* insert an orphan item to track subvolume contains orphan files */
3021 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3022 root
->root_key
.objectid
);
3023 if (ret
&& ret
!= -EEXIST
) {
3024 btrfs_abort_transaction(trans
, root
, ret
);
3032 * We have done the truncate/delete so we can go ahead and remove the orphan
3033 * item for this particular inode.
3035 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3036 struct inode
*inode
)
3038 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3039 int delete_item
= 0;
3040 int release_rsv
= 0;
3043 spin_lock(&root
->orphan_lock
);
3044 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3045 &BTRFS_I(inode
)->runtime_flags
))
3048 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3049 &BTRFS_I(inode
)->runtime_flags
))
3051 spin_unlock(&root
->orphan_lock
);
3054 atomic_dec(&root
->orphan_inodes
);
3056 ret
= btrfs_del_orphan_item(trans
, root
,
3061 btrfs_orphan_release_metadata(inode
);
3067 * this cleans up any orphans that may be left on the list from the last use
3070 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3072 struct btrfs_path
*path
;
3073 struct extent_buffer
*leaf
;
3074 struct btrfs_key key
, found_key
;
3075 struct btrfs_trans_handle
*trans
;
3076 struct inode
*inode
;
3077 u64 last_objectid
= 0;
3078 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3080 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3083 path
= btrfs_alloc_path();
3090 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3091 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
3092 key
.offset
= (u64
)-1;
3095 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3100 * if ret == 0 means we found what we were searching for, which
3101 * is weird, but possible, so only screw with path if we didn't
3102 * find the key and see if we have stuff that matches
3106 if (path
->slots
[0] == 0)
3111 /* pull out the item */
3112 leaf
= path
->nodes
[0];
3113 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3115 /* make sure the item matches what we want */
3116 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3118 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
3121 /* release the path since we're done with it */
3122 btrfs_release_path(path
);
3125 * this is where we are basically btrfs_lookup, without the
3126 * crossing root thing. we store the inode number in the
3127 * offset of the orphan item.
3130 if (found_key
.offset
== last_objectid
) {
3131 btrfs_err(root
->fs_info
,
3132 "Error removing orphan entry, stopping orphan cleanup");
3137 last_objectid
= found_key
.offset
;
3139 found_key
.objectid
= found_key
.offset
;
3140 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3141 found_key
.offset
= 0;
3142 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3143 ret
= PTR_ERR_OR_ZERO(inode
);
3144 if (ret
&& ret
!= -ESTALE
)
3147 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3148 struct btrfs_root
*dead_root
;
3149 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3150 int is_dead_root
= 0;
3153 * this is an orphan in the tree root. Currently these
3154 * could come from 2 sources:
3155 * a) a snapshot deletion in progress
3156 * b) a free space cache inode
3157 * We need to distinguish those two, as the snapshot
3158 * orphan must not get deleted.
3159 * find_dead_roots already ran before us, so if this
3160 * is a snapshot deletion, we should find the root
3161 * in the dead_roots list
3163 spin_lock(&fs_info
->trans_lock
);
3164 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3166 if (dead_root
->root_key
.objectid
==
3167 found_key
.objectid
) {
3172 spin_unlock(&fs_info
->trans_lock
);
3174 /* prevent this orphan from being found again */
3175 key
.offset
= found_key
.objectid
- 1;
3180 * Inode is already gone but the orphan item is still there,
3181 * kill the orphan item.
3183 if (ret
== -ESTALE
) {
3184 trans
= btrfs_start_transaction(root
, 1);
3185 if (IS_ERR(trans
)) {
3186 ret
= PTR_ERR(trans
);
3189 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3190 found_key
.objectid
);
3191 ret
= btrfs_del_orphan_item(trans
, root
,
3192 found_key
.objectid
);
3193 btrfs_end_transaction(trans
, root
);
3200 * add this inode to the orphan list so btrfs_orphan_del does
3201 * the proper thing when we hit it
3203 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3204 &BTRFS_I(inode
)->runtime_flags
);
3205 atomic_inc(&root
->orphan_inodes
);
3207 /* if we have links, this was a truncate, lets do that */
3208 if (inode
->i_nlink
) {
3209 if (WARN_ON(!S_ISREG(inode
->i_mode
))) {
3215 /* 1 for the orphan item deletion. */
3216 trans
= btrfs_start_transaction(root
, 1);
3217 if (IS_ERR(trans
)) {
3219 ret
= PTR_ERR(trans
);
3222 ret
= btrfs_orphan_add(trans
, inode
);
3223 btrfs_end_transaction(trans
, root
);
3229 ret
= btrfs_truncate(inode
);
3231 btrfs_orphan_del(NULL
, inode
);
3236 /* this will do delete_inode and everything for us */
3241 /* release the path since we're done with it */
3242 btrfs_release_path(path
);
3244 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3246 if (root
->orphan_block_rsv
)
3247 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3250 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
3251 trans
= btrfs_join_transaction(root
);
3253 btrfs_end_transaction(trans
, root
);
3257 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3259 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3263 btrfs_crit(root
->fs_info
,
3264 "could not do orphan cleanup %d", ret
);
3265 btrfs_free_path(path
);
3270 * very simple check to peek ahead in the leaf looking for xattrs. If we
3271 * don't find any xattrs, we know there can't be any acls.
3273 * slot is the slot the inode is in, objectid is the objectid of the inode
3275 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3276 int slot
, u64 objectid
,
3277 int *first_xattr_slot
)
3279 u32 nritems
= btrfs_header_nritems(leaf
);
3280 struct btrfs_key found_key
;
3281 static u64 xattr_access
= 0;
3282 static u64 xattr_default
= 0;
3285 if (!xattr_access
) {
3286 xattr_access
= btrfs_name_hash(POSIX_ACL_XATTR_ACCESS
,
3287 strlen(POSIX_ACL_XATTR_ACCESS
));
3288 xattr_default
= btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT
,
3289 strlen(POSIX_ACL_XATTR_DEFAULT
));
3293 *first_xattr_slot
= -1;
3294 while (slot
< nritems
) {
3295 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3297 /* we found a different objectid, there must not be acls */
3298 if (found_key
.objectid
!= objectid
)
3301 /* we found an xattr, assume we've got an acl */
3302 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
) {
3303 if (*first_xattr_slot
== -1)
3304 *first_xattr_slot
= slot
;
3305 if (found_key
.offset
== xattr_access
||
3306 found_key
.offset
== xattr_default
)
3311 * we found a key greater than an xattr key, there can't
3312 * be any acls later on
3314 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3321 * it goes inode, inode backrefs, xattrs, extents,
3322 * so if there are a ton of hard links to an inode there can
3323 * be a lot of backrefs. Don't waste time searching too hard,
3324 * this is just an optimization
3329 /* we hit the end of the leaf before we found an xattr or
3330 * something larger than an xattr. We have to assume the inode
3333 if (*first_xattr_slot
== -1)
3334 *first_xattr_slot
= slot
;
3339 * read an inode from the btree into the in-memory inode
3341 static void btrfs_read_locked_inode(struct inode
*inode
)
3343 struct btrfs_path
*path
;
3344 struct extent_buffer
*leaf
;
3345 struct btrfs_inode_item
*inode_item
;
3346 struct btrfs_timespec
*tspec
;
3347 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3348 struct btrfs_key location
;
3353 bool filled
= false;
3354 int first_xattr_slot
;
3356 ret
= btrfs_fill_inode(inode
, &rdev
);
3360 path
= btrfs_alloc_path();
3364 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3366 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3370 leaf
= path
->nodes
[0];
3375 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3376 struct btrfs_inode_item
);
3377 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3378 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3379 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3380 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3381 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3383 tspec
= btrfs_inode_atime(inode_item
);
3384 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3385 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3387 tspec
= btrfs_inode_mtime(inode_item
);
3388 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3389 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3391 tspec
= btrfs_inode_ctime(inode_item
);
3392 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3393 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3395 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3396 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3397 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3400 * If we were modified in the current generation and evicted from memory
3401 * and then re-read we need to do a full sync since we don't have any
3402 * idea about which extents were modified before we were evicted from
3405 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3406 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3407 &BTRFS_I(inode
)->runtime_flags
);
3409 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3410 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3412 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3414 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3415 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3419 if (inode
->i_nlink
!= 1 ||
3420 path
->slots
[0] >= btrfs_header_nritems(leaf
))
3423 btrfs_item_key_to_cpu(leaf
, &location
, path
->slots
[0]);
3424 if (location
.objectid
!= btrfs_ino(inode
))
3427 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3428 if (location
.type
== BTRFS_INODE_REF_KEY
) {
3429 struct btrfs_inode_ref
*ref
;
3431 ref
= (struct btrfs_inode_ref
*)ptr
;
3432 BTRFS_I(inode
)->dir_index
= btrfs_inode_ref_index(leaf
, ref
);
3433 } else if (location
.type
== BTRFS_INODE_EXTREF_KEY
) {
3434 struct btrfs_inode_extref
*extref
;
3436 extref
= (struct btrfs_inode_extref
*)ptr
;
3437 BTRFS_I(inode
)->dir_index
= btrfs_inode_extref_index(leaf
,
3442 * try to precache a NULL acl entry for files that don't have
3443 * any xattrs or acls
3445 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3446 btrfs_ino(inode
), &first_xattr_slot
);
3447 if (first_xattr_slot
!= -1) {
3448 path
->slots
[0] = first_xattr_slot
;
3449 ret
= btrfs_load_inode_props(inode
, path
);
3451 btrfs_err(root
->fs_info
,
3452 "error loading props for ino %llu (root %llu): %d\n",
3454 root
->root_key
.objectid
, ret
);
3456 btrfs_free_path(path
);
3459 cache_no_acl(inode
);
3461 switch (inode
->i_mode
& S_IFMT
) {
3463 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3464 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3465 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3466 inode
->i_fop
= &btrfs_file_operations
;
3467 inode
->i_op
= &btrfs_file_inode_operations
;
3470 inode
->i_fop
= &btrfs_dir_file_operations
;
3471 if (root
== root
->fs_info
->tree_root
)
3472 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3474 inode
->i_op
= &btrfs_dir_inode_operations
;
3477 inode
->i_op
= &btrfs_symlink_inode_operations
;
3478 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3479 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3482 inode
->i_op
= &btrfs_special_inode_operations
;
3483 init_special_inode(inode
, inode
->i_mode
, rdev
);
3487 btrfs_update_iflags(inode
);
3491 btrfs_free_path(path
);
3492 make_bad_inode(inode
);
3496 * given a leaf and an inode, copy the inode fields into the leaf
3498 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3499 struct extent_buffer
*leaf
,
3500 struct btrfs_inode_item
*item
,
3501 struct inode
*inode
)
3503 struct btrfs_map_token token
;
3505 btrfs_init_map_token(&token
);
3507 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3508 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3509 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3511 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3512 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3514 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_atime(item
),
3515 inode
->i_atime
.tv_sec
, &token
);
3516 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_atime(item
),
3517 inode
->i_atime
.tv_nsec
, &token
);
3519 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_mtime(item
),
3520 inode
->i_mtime
.tv_sec
, &token
);
3521 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
3522 inode
->i_mtime
.tv_nsec
, &token
);
3524 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_ctime(item
),
3525 inode
->i_ctime
.tv_sec
, &token
);
3526 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
3527 inode
->i_ctime
.tv_nsec
, &token
);
3529 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3531 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3533 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3534 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3535 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3536 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3537 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3541 * copy everything in the in-memory inode into the btree.
3543 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3544 struct btrfs_root
*root
, struct inode
*inode
)
3546 struct btrfs_inode_item
*inode_item
;
3547 struct btrfs_path
*path
;
3548 struct extent_buffer
*leaf
;
3551 path
= btrfs_alloc_path();
3555 path
->leave_spinning
= 1;
3556 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3564 leaf
= path
->nodes
[0];
3565 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3566 struct btrfs_inode_item
);
3568 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3569 btrfs_mark_buffer_dirty(leaf
);
3570 btrfs_set_inode_last_trans(trans
, inode
);
3573 btrfs_free_path(path
);
3578 * copy everything in the in-memory inode into the btree.
3580 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3581 struct btrfs_root
*root
, struct inode
*inode
)
3586 * If the inode is a free space inode, we can deadlock during commit
3587 * if we put it into the delayed code.
3589 * The data relocation inode should also be directly updated
3592 if (!btrfs_is_free_space_inode(inode
)
3593 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
3594 btrfs_update_root_times(trans
, root
);
3596 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3598 btrfs_set_inode_last_trans(trans
, inode
);
3602 return btrfs_update_inode_item(trans
, root
, inode
);
3605 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3606 struct btrfs_root
*root
,
3607 struct inode
*inode
)
3611 ret
= btrfs_update_inode(trans
, root
, inode
);
3613 return btrfs_update_inode_item(trans
, root
, inode
);
3618 * unlink helper that gets used here in inode.c and in the tree logging
3619 * recovery code. It remove a link in a directory with a given name, and
3620 * also drops the back refs in the inode to the directory
3622 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3623 struct btrfs_root
*root
,
3624 struct inode
*dir
, struct inode
*inode
,
3625 const char *name
, int name_len
)
3627 struct btrfs_path
*path
;
3629 struct extent_buffer
*leaf
;
3630 struct btrfs_dir_item
*di
;
3631 struct btrfs_key key
;
3633 u64 ino
= btrfs_ino(inode
);
3634 u64 dir_ino
= btrfs_ino(dir
);
3636 path
= btrfs_alloc_path();
3642 path
->leave_spinning
= 1;
3643 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3644 name
, name_len
, -1);
3653 leaf
= path
->nodes
[0];
3654 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3655 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3658 btrfs_release_path(path
);
3661 * If we don't have dir index, we have to get it by looking up
3662 * the inode ref, since we get the inode ref, remove it directly,
3663 * it is unnecessary to do delayed deletion.
3665 * But if we have dir index, needn't search inode ref to get it.
3666 * Since the inode ref is close to the inode item, it is better
3667 * that we delay to delete it, and just do this deletion when
3668 * we update the inode item.
3670 if (BTRFS_I(inode
)->dir_index
) {
3671 ret
= btrfs_delayed_delete_inode_ref(inode
);
3673 index
= BTRFS_I(inode
)->dir_index
;
3678 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3681 btrfs_info(root
->fs_info
,
3682 "failed to delete reference to %.*s, inode %llu parent %llu",
3683 name_len
, name
, ino
, dir_ino
);
3684 btrfs_abort_transaction(trans
, root
, ret
);
3688 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3690 btrfs_abort_transaction(trans
, root
, ret
);
3694 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3696 if (ret
!= 0 && ret
!= -ENOENT
) {
3697 btrfs_abort_transaction(trans
, root
, ret
);
3701 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
3706 btrfs_abort_transaction(trans
, root
, ret
);
3708 btrfs_free_path(path
);
3712 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3713 inode_inc_iversion(inode
);
3714 inode_inc_iversion(dir
);
3715 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3716 ret
= btrfs_update_inode(trans
, root
, dir
);
3721 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3722 struct btrfs_root
*root
,
3723 struct inode
*dir
, struct inode
*inode
,
3724 const char *name
, int name_len
)
3727 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
3730 ret
= btrfs_update_inode(trans
, root
, inode
);
3736 * helper to start transaction for unlink and rmdir.
3738 * unlink and rmdir are special in btrfs, they do not always free space, so
3739 * if we cannot make our reservations the normal way try and see if there is
3740 * plenty of slack room in the global reserve to migrate, otherwise we cannot
3741 * allow the unlink to occur.
3743 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
)
3745 struct btrfs_trans_handle
*trans
;
3746 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3750 * 1 for the possible orphan item
3751 * 1 for the dir item
3752 * 1 for the dir index
3753 * 1 for the inode ref
3756 trans
= btrfs_start_transaction(root
, 5);
3757 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
3760 if (PTR_ERR(trans
) == -ENOSPC
) {
3761 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3763 trans
= btrfs_start_transaction(root
, 0);
3766 ret
= btrfs_cond_migrate_bytes(root
->fs_info
,
3767 &root
->fs_info
->trans_block_rsv
,
3770 btrfs_end_transaction(trans
, root
);
3771 return ERR_PTR(ret
);
3773 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3774 trans
->bytes_reserved
= num_bytes
;
3779 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3781 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3782 struct btrfs_trans_handle
*trans
;
3783 struct inode
*inode
= dentry
->d_inode
;
3786 trans
= __unlink_start_trans(dir
);
3788 return PTR_ERR(trans
);
3790 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3792 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3793 dentry
->d_name
.name
, dentry
->d_name
.len
);
3797 if (inode
->i_nlink
== 0) {
3798 ret
= btrfs_orphan_add(trans
, inode
);
3804 btrfs_end_transaction(trans
, root
);
3805 btrfs_btree_balance_dirty(root
);
3809 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3810 struct btrfs_root
*root
,
3811 struct inode
*dir
, u64 objectid
,
3812 const char *name
, int name_len
)
3814 struct btrfs_path
*path
;
3815 struct extent_buffer
*leaf
;
3816 struct btrfs_dir_item
*di
;
3817 struct btrfs_key key
;
3820 u64 dir_ino
= btrfs_ino(dir
);
3822 path
= btrfs_alloc_path();
3826 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3827 name
, name_len
, -1);
3828 if (IS_ERR_OR_NULL(di
)) {
3836 leaf
= path
->nodes
[0];
3837 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3838 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3839 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3841 btrfs_abort_transaction(trans
, root
, ret
);
3844 btrfs_release_path(path
);
3846 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3847 objectid
, root
->root_key
.objectid
,
3848 dir_ino
, &index
, name
, name_len
);
3850 if (ret
!= -ENOENT
) {
3851 btrfs_abort_transaction(trans
, root
, ret
);
3854 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3856 if (IS_ERR_OR_NULL(di
)) {
3861 btrfs_abort_transaction(trans
, root
, ret
);
3865 leaf
= path
->nodes
[0];
3866 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3867 btrfs_release_path(path
);
3870 btrfs_release_path(path
);
3872 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3874 btrfs_abort_transaction(trans
, root
, ret
);
3878 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3879 inode_inc_iversion(dir
);
3880 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3881 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
3883 btrfs_abort_transaction(trans
, root
, ret
);
3885 btrfs_free_path(path
);
3889 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3891 struct inode
*inode
= dentry
->d_inode
;
3893 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3894 struct btrfs_trans_handle
*trans
;
3896 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
3898 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3901 trans
= __unlink_start_trans(dir
);
3903 return PTR_ERR(trans
);
3905 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
3906 err
= btrfs_unlink_subvol(trans
, root
, dir
,
3907 BTRFS_I(inode
)->location
.objectid
,
3908 dentry
->d_name
.name
,
3909 dentry
->d_name
.len
);
3913 err
= btrfs_orphan_add(trans
, inode
);
3917 /* now the directory is empty */
3918 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3919 dentry
->d_name
.name
, dentry
->d_name
.len
);
3921 btrfs_i_size_write(inode
, 0);
3923 btrfs_end_transaction(trans
, root
);
3924 btrfs_btree_balance_dirty(root
);
3930 * this can truncate away extent items, csum items and directory items.
3931 * It starts at a high offset and removes keys until it can't find
3932 * any higher than new_size
3934 * csum items that cross the new i_size are truncated to the new size
3937 * min_type is the minimum key type to truncate down to. If set to 0, this
3938 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3940 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
3941 struct btrfs_root
*root
,
3942 struct inode
*inode
,
3943 u64 new_size
, u32 min_type
)
3945 struct btrfs_path
*path
;
3946 struct extent_buffer
*leaf
;
3947 struct btrfs_file_extent_item
*fi
;
3948 struct btrfs_key key
;
3949 struct btrfs_key found_key
;
3950 u64 extent_start
= 0;
3951 u64 extent_num_bytes
= 0;
3952 u64 extent_offset
= 0;
3954 u64 last_size
= (u64
)-1;
3955 u32 found_type
= (u8
)-1;
3958 int pending_del_nr
= 0;
3959 int pending_del_slot
= 0;
3960 int extent_type
= -1;
3963 u64 ino
= btrfs_ino(inode
);
3965 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
3967 path
= btrfs_alloc_path();
3973 * We want to drop from the next block forward in case this new size is
3974 * not block aligned since we will be keeping the last block of the
3975 * extent just the way it is.
3977 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
3978 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
3979 root
->sectorsize
), (u64
)-1, 0);
3982 * This function is also used to drop the items in the log tree before
3983 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3984 * it is used to drop the loged items. So we shouldn't kill the delayed
3987 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
3988 btrfs_kill_delayed_inode_items(inode
);
3991 key
.offset
= (u64
)-1;
3995 path
->leave_spinning
= 1;
3996 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4003 /* there are no items in the tree for us to truncate, we're
4006 if (path
->slots
[0] == 0)
4013 leaf
= path
->nodes
[0];
4014 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4015 found_type
= btrfs_key_type(&found_key
);
4017 if (found_key
.objectid
!= ino
)
4020 if (found_type
< min_type
)
4023 item_end
= found_key
.offset
;
4024 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4025 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4026 struct btrfs_file_extent_item
);
4027 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4028 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4030 btrfs_file_extent_num_bytes(leaf
, fi
);
4031 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4032 item_end
+= btrfs_file_extent_inline_len(leaf
,
4033 path
->slots
[0], fi
);
4037 if (found_type
> min_type
) {
4040 if (item_end
< new_size
)
4042 if (found_key
.offset
>= new_size
)
4048 /* FIXME, shrink the extent if the ref count is only 1 */
4049 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4053 last_size
= found_key
.offset
;
4055 last_size
= new_size
;
4057 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4059 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4061 u64 orig_num_bytes
=
4062 btrfs_file_extent_num_bytes(leaf
, fi
);
4063 extent_num_bytes
= ALIGN(new_size
-
4066 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4068 num_dec
= (orig_num_bytes
-
4070 if (root
->ref_cows
&& extent_start
!= 0)
4071 inode_sub_bytes(inode
, num_dec
);
4072 btrfs_mark_buffer_dirty(leaf
);
4075 btrfs_file_extent_disk_num_bytes(leaf
,
4077 extent_offset
= found_key
.offset
-
4078 btrfs_file_extent_offset(leaf
, fi
);
4080 /* FIXME blocksize != 4096 */
4081 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4082 if (extent_start
!= 0) {
4085 inode_sub_bytes(inode
, num_dec
);
4088 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4090 * we can't truncate inline items that have had
4094 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
4095 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4096 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4097 u32 size
= new_size
- found_key
.offset
;
4099 if (root
->ref_cows
) {
4100 inode_sub_bytes(inode
, item_end
+ 1 -
4105 * update the ram bytes to properly reflect
4106 * the new size of our item
4108 btrfs_set_file_extent_ram_bytes(leaf
, fi
, size
);
4110 btrfs_file_extent_calc_inline_size(size
);
4111 btrfs_truncate_item(root
, path
, size
, 1);
4112 } else if (root
->ref_cows
) {
4113 inode_sub_bytes(inode
, item_end
+ 1 -
4119 if (!pending_del_nr
) {
4120 /* no pending yet, add ourselves */
4121 pending_del_slot
= path
->slots
[0];
4123 } else if (pending_del_nr
&&
4124 path
->slots
[0] + 1 == pending_del_slot
) {
4125 /* hop on the pending chunk */
4127 pending_del_slot
= path
->slots
[0];
4134 if (found_extent
&& (root
->ref_cows
||
4135 root
== root
->fs_info
->tree_root
)) {
4136 btrfs_set_path_blocking(path
);
4137 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4138 extent_num_bytes
, 0,
4139 btrfs_header_owner(leaf
),
4140 ino
, extent_offset
, 0);
4144 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4147 if (path
->slots
[0] == 0 ||
4148 path
->slots
[0] != pending_del_slot
) {
4149 if (pending_del_nr
) {
4150 ret
= btrfs_del_items(trans
, root
, path
,
4154 btrfs_abort_transaction(trans
,
4160 btrfs_release_path(path
);
4167 if (pending_del_nr
) {
4168 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4171 btrfs_abort_transaction(trans
, root
, ret
);
4174 if (last_size
!= (u64
)-1)
4175 btrfs_ordered_update_i_size(inode
, last_size
, NULL
);
4176 btrfs_free_path(path
);
4181 * btrfs_truncate_page - read, zero a chunk and write a page
4182 * @inode - inode that we're zeroing
4183 * @from - the offset to start zeroing
4184 * @len - the length to zero, 0 to zero the entire range respective to the
4186 * @front - zero up to the offset instead of from the offset on
4188 * This will find the page for the "from" offset and cow the page and zero the
4189 * part we want to zero. This is used with truncate and hole punching.
4191 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4194 struct address_space
*mapping
= inode
->i_mapping
;
4195 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4196 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4197 struct btrfs_ordered_extent
*ordered
;
4198 struct extent_state
*cached_state
= NULL
;
4200 u32 blocksize
= root
->sectorsize
;
4201 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4202 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4204 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4209 if ((offset
& (blocksize
- 1)) == 0 &&
4210 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4212 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
4217 page
= find_or_create_page(mapping
, index
, mask
);
4219 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4224 page_start
= page_offset(page
);
4225 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4227 if (!PageUptodate(page
)) {
4228 ret
= btrfs_readpage(NULL
, page
);
4230 if (page
->mapping
!= mapping
) {
4232 page_cache_release(page
);
4235 if (!PageUptodate(page
)) {
4240 wait_on_page_writeback(page
);
4242 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4243 set_page_extent_mapped(page
);
4245 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4247 unlock_extent_cached(io_tree
, page_start
, page_end
,
4248 &cached_state
, GFP_NOFS
);
4250 page_cache_release(page
);
4251 btrfs_start_ordered_extent(inode
, ordered
, 1);
4252 btrfs_put_ordered_extent(ordered
);
4256 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4257 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4258 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4259 0, 0, &cached_state
, GFP_NOFS
);
4261 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4264 unlock_extent_cached(io_tree
, page_start
, page_end
,
4265 &cached_state
, GFP_NOFS
);
4269 if (offset
!= PAGE_CACHE_SIZE
) {
4271 len
= PAGE_CACHE_SIZE
- offset
;
4274 memset(kaddr
, 0, offset
);
4276 memset(kaddr
+ offset
, 0, len
);
4277 flush_dcache_page(page
);
4280 ClearPageChecked(page
);
4281 set_page_dirty(page
);
4282 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4287 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4289 page_cache_release(page
);
4294 static int maybe_insert_hole(struct btrfs_root
*root
, struct inode
*inode
,
4295 u64 offset
, u64 len
)
4297 struct btrfs_trans_handle
*trans
;
4301 * Still need to make sure the inode looks like it's been updated so
4302 * that any holes get logged if we fsync.
4304 if (btrfs_fs_incompat(root
->fs_info
, NO_HOLES
)) {
4305 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
4306 BTRFS_I(inode
)->last_sub_trans
= root
->log_transid
;
4307 BTRFS_I(inode
)->last_log_commit
= root
->last_log_commit
;
4312 * 1 - for the one we're dropping
4313 * 1 - for the one we're adding
4314 * 1 - for updating the inode.
4316 trans
= btrfs_start_transaction(root
, 3);
4318 return PTR_ERR(trans
);
4320 ret
= btrfs_drop_extents(trans
, root
, inode
, offset
, offset
+ len
, 1);
4322 btrfs_abort_transaction(trans
, root
, ret
);
4323 btrfs_end_transaction(trans
, root
);
4327 ret
= btrfs_insert_file_extent(trans
, root
, btrfs_ino(inode
), offset
,
4328 0, 0, len
, 0, len
, 0, 0, 0);
4330 btrfs_abort_transaction(trans
, root
, ret
);
4332 btrfs_update_inode(trans
, root
, inode
);
4333 btrfs_end_transaction(trans
, root
);
4338 * This function puts in dummy file extents for the area we're creating a hole
4339 * for. So if we are truncating this file to a larger size we need to insert
4340 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4341 * the range between oldsize and size
4343 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4345 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4346 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4347 struct extent_map
*em
= NULL
;
4348 struct extent_state
*cached_state
= NULL
;
4349 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4350 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4351 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4358 * If our size started in the middle of a page we need to zero out the
4359 * rest of the page before we expand the i_size, otherwise we could
4360 * expose stale data.
4362 err
= btrfs_truncate_page(inode
, oldsize
, 0, 0);
4366 if (size
<= hole_start
)
4370 struct btrfs_ordered_extent
*ordered
;
4372 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4374 ordered
= btrfs_lookup_ordered_range(inode
, hole_start
,
4375 block_end
- hole_start
);
4378 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4379 &cached_state
, GFP_NOFS
);
4380 btrfs_start_ordered_extent(inode
, ordered
, 1);
4381 btrfs_put_ordered_extent(ordered
);
4384 cur_offset
= hole_start
;
4386 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4387 block_end
- cur_offset
, 0);
4393 last_byte
= min(extent_map_end(em
), block_end
);
4394 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4395 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4396 struct extent_map
*hole_em
;
4397 hole_size
= last_byte
- cur_offset
;
4399 err
= maybe_insert_hole(root
, inode
, cur_offset
,
4403 btrfs_drop_extent_cache(inode
, cur_offset
,
4404 cur_offset
+ hole_size
- 1, 0);
4405 hole_em
= alloc_extent_map();
4407 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4408 &BTRFS_I(inode
)->runtime_flags
);
4411 hole_em
->start
= cur_offset
;
4412 hole_em
->len
= hole_size
;
4413 hole_em
->orig_start
= cur_offset
;
4415 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4416 hole_em
->block_len
= 0;
4417 hole_em
->orig_block_len
= 0;
4418 hole_em
->ram_bytes
= hole_size
;
4419 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4420 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4421 hole_em
->generation
= root
->fs_info
->generation
;
4424 write_lock(&em_tree
->lock
);
4425 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4426 write_unlock(&em_tree
->lock
);
4429 btrfs_drop_extent_cache(inode
, cur_offset
,
4433 free_extent_map(hole_em
);
4436 free_extent_map(em
);
4438 cur_offset
= last_byte
;
4439 if (cur_offset
>= block_end
)
4442 free_extent_map(em
);
4443 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4448 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4450 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4451 struct btrfs_trans_handle
*trans
;
4452 loff_t oldsize
= i_size_read(inode
);
4453 loff_t newsize
= attr
->ia_size
;
4454 int mask
= attr
->ia_valid
;
4458 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4459 * special case where we need to update the times despite not having
4460 * these flags set. For all other operations the VFS set these flags
4461 * explicitly if it wants a timestamp update.
4463 if (newsize
!= oldsize
&& (!(mask
& (ATTR_CTIME
| ATTR_MTIME
))))
4464 inode
->i_ctime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
4466 if (newsize
> oldsize
) {
4467 truncate_pagecache(inode
, newsize
);
4468 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4472 trans
= btrfs_start_transaction(root
, 1);
4474 return PTR_ERR(trans
);
4476 i_size_write(inode
, newsize
);
4477 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4478 ret
= btrfs_update_inode(trans
, root
, inode
);
4479 btrfs_end_transaction(trans
, root
);
4483 * We're truncating a file that used to have good data down to
4484 * zero. Make sure it gets into the ordered flush list so that
4485 * any new writes get down to disk quickly.
4488 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4489 &BTRFS_I(inode
)->runtime_flags
);
4492 * 1 for the orphan item we're going to add
4493 * 1 for the orphan item deletion.
4495 trans
= btrfs_start_transaction(root
, 2);
4497 return PTR_ERR(trans
);
4500 * We need to do this in case we fail at _any_ point during the
4501 * actual truncate. Once we do the truncate_setsize we could
4502 * invalidate pages which forces any outstanding ordered io to
4503 * be instantly completed which will give us extents that need
4504 * to be truncated. If we fail to get an orphan inode down we
4505 * could have left over extents that were never meant to live,
4506 * so we need to garuntee from this point on that everything
4507 * will be consistent.
4509 ret
= btrfs_orphan_add(trans
, inode
);
4510 btrfs_end_transaction(trans
, root
);
4514 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4515 truncate_setsize(inode
, newsize
);
4517 /* Disable nonlocked read DIO to avoid the end less truncate */
4518 btrfs_inode_block_unlocked_dio(inode
);
4519 inode_dio_wait(inode
);
4520 btrfs_inode_resume_unlocked_dio(inode
);
4522 ret
= btrfs_truncate(inode
);
4523 if (ret
&& inode
->i_nlink
) {
4527 * failed to truncate, disk_i_size is only adjusted down
4528 * as we remove extents, so it should represent the true
4529 * size of the inode, so reset the in memory size and
4530 * delete our orphan entry.
4532 trans
= btrfs_join_transaction(root
);
4533 if (IS_ERR(trans
)) {
4534 btrfs_orphan_del(NULL
, inode
);
4537 i_size_write(inode
, BTRFS_I(inode
)->disk_i_size
);
4538 err
= btrfs_orphan_del(trans
, inode
);
4540 btrfs_abort_transaction(trans
, root
, err
);
4541 btrfs_end_transaction(trans
, root
);
4548 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4550 struct inode
*inode
= dentry
->d_inode
;
4551 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4554 if (btrfs_root_readonly(root
))
4557 err
= inode_change_ok(inode
, attr
);
4561 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
4562 err
= btrfs_setsize(inode
, attr
);
4567 if (attr
->ia_valid
) {
4568 setattr_copy(inode
, attr
);
4569 inode_inc_iversion(inode
);
4570 err
= btrfs_dirty_inode(inode
);
4572 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
4573 err
= btrfs_acl_chmod(inode
);
4580 * While truncating the inode pages during eviction, we get the VFS calling
4581 * btrfs_invalidatepage() against each page of the inode. This is slow because
4582 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4583 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4584 * extent_state structures over and over, wasting lots of time.
4586 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4587 * those expensive operations on a per page basis and do only the ordered io
4588 * finishing, while we release here the extent_map and extent_state structures,
4589 * without the excessive merging and splitting.
4591 static void evict_inode_truncate_pages(struct inode
*inode
)
4593 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4594 struct extent_map_tree
*map_tree
= &BTRFS_I(inode
)->extent_tree
;
4595 struct rb_node
*node
;
4597 ASSERT(inode
->i_state
& I_FREEING
);
4598 truncate_inode_pages(&inode
->i_data
, 0);
4600 write_lock(&map_tree
->lock
);
4601 while (!RB_EMPTY_ROOT(&map_tree
->map
)) {
4602 struct extent_map
*em
;
4604 node
= rb_first(&map_tree
->map
);
4605 em
= rb_entry(node
, struct extent_map
, rb_node
);
4606 clear_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
4607 clear_bit(EXTENT_FLAG_LOGGING
, &em
->flags
);
4608 remove_extent_mapping(map_tree
, em
);
4609 free_extent_map(em
);
4611 write_unlock(&map_tree
->lock
);
4613 spin_lock(&io_tree
->lock
);
4614 while (!RB_EMPTY_ROOT(&io_tree
->state
)) {
4615 struct extent_state
*state
;
4616 struct extent_state
*cached_state
= NULL
;
4618 node
= rb_first(&io_tree
->state
);
4619 state
= rb_entry(node
, struct extent_state
, rb_node
);
4620 atomic_inc(&state
->refs
);
4621 spin_unlock(&io_tree
->lock
);
4623 lock_extent_bits(io_tree
, state
->start
, state
->end
,
4625 clear_extent_bit(io_tree
, state
->start
, state
->end
,
4626 EXTENT_LOCKED
| EXTENT_DIRTY
|
4627 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
4628 EXTENT_DEFRAG
, 1, 1,
4629 &cached_state
, GFP_NOFS
);
4630 free_extent_state(state
);
4632 spin_lock(&io_tree
->lock
);
4634 spin_unlock(&io_tree
->lock
);
4637 void btrfs_evict_inode(struct inode
*inode
)
4639 struct btrfs_trans_handle
*trans
;
4640 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4641 struct btrfs_block_rsv
*rsv
, *global_rsv
;
4642 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
4645 trace_btrfs_inode_evict(inode
);
4647 evict_inode_truncate_pages(inode
);
4649 if (inode
->i_nlink
&&
4650 ((btrfs_root_refs(&root
->root_item
) != 0 &&
4651 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
) ||
4652 btrfs_is_free_space_inode(inode
)))
4655 if (is_bad_inode(inode
)) {
4656 btrfs_orphan_del(NULL
, inode
);
4659 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4660 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
4662 if (root
->fs_info
->log_root_recovering
) {
4663 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
4664 &BTRFS_I(inode
)->runtime_flags
));
4668 if (inode
->i_nlink
> 0) {
4669 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0 &&
4670 root
->root_key
.objectid
!= BTRFS_ROOT_TREE_OBJECTID
);
4674 ret
= btrfs_commit_inode_delayed_inode(inode
);
4676 btrfs_orphan_del(NULL
, inode
);
4680 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
4682 btrfs_orphan_del(NULL
, inode
);
4685 rsv
->size
= min_size
;
4687 global_rsv
= &root
->fs_info
->global_block_rsv
;
4689 btrfs_i_size_write(inode
, 0);
4692 * This is a bit simpler than btrfs_truncate since we've already
4693 * reserved our space for our orphan item in the unlink, so we just
4694 * need to reserve some slack space in case we add bytes and update
4695 * inode item when doing the truncate.
4698 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
4699 BTRFS_RESERVE_FLUSH_LIMIT
);
4702 * Try and steal from the global reserve since we will
4703 * likely not use this space anyway, we want to try as
4704 * hard as possible to get this to work.
4707 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
4710 btrfs_warn(root
->fs_info
,
4711 "Could not get space for a delete, will truncate on mount %d",
4713 btrfs_orphan_del(NULL
, inode
);
4714 btrfs_free_block_rsv(root
, rsv
);
4718 trans
= btrfs_join_transaction(root
);
4719 if (IS_ERR(trans
)) {
4720 btrfs_orphan_del(NULL
, inode
);
4721 btrfs_free_block_rsv(root
, rsv
);
4725 trans
->block_rsv
= rsv
;
4727 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
4731 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4732 btrfs_end_transaction(trans
, root
);
4734 btrfs_btree_balance_dirty(root
);
4737 btrfs_free_block_rsv(root
, rsv
);
4740 * Errors here aren't a big deal, it just means we leave orphan items
4741 * in the tree. They will be cleaned up on the next mount.
4744 trans
->block_rsv
= root
->orphan_block_rsv
;
4745 btrfs_orphan_del(trans
, inode
);
4747 btrfs_orphan_del(NULL
, inode
);
4750 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4751 if (!(root
== root
->fs_info
->tree_root
||
4752 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
4753 btrfs_return_ino(root
, btrfs_ino(inode
));
4755 btrfs_end_transaction(trans
, root
);
4756 btrfs_btree_balance_dirty(root
);
4758 btrfs_remove_delayed_node(inode
);
4764 * this returns the key found in the dir entry in the location pointer.
4765 * If no dir entries were found, location->objectid is 0.
4767 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
4768 struct btrfs_key
*location
)
4770 const char *name
= dentry
->d_name
.name
;
4771 int namelen
= dentry
->d_name
.len
;
4772 struct btrfs_dir_item
*di
;
4773 struct btrfs_path
*path
;
4774 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4777 path
= btrfs_alloc_path();
4781 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
4786 if (IS_ERR_OR_NULL(di
))
4789 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
4791 btrfs_free_path(path
);
4794 location
->objectid
= 0;
4799 * when we hit a tree root in a directory, the btrfs part of the inode
4800 * needs to be changed to reflect the root directory of the tree root. This
4801 * is kind of like crossing a mount point.
4803 static int fixup_tree_root_location(struct btrfs_root
*root
,
4805 struct dentry
*dentry
,
4806 struct btrfs_key
*location
,
4807 struct btrfs_root
**sub_root
)
4809 struct btrfs_path
*path
;
4810 struct btrfs_root
*new_root
;
4811 struct btrfs_root_ref
*ref
;
4812 struct extent_buffer
*leaf
;
4816 path
= btrfs_alloc_path();
4823 ret
= btrfs_find_item(root
->fs_info
->tree_root
, path
,
4824 BTRFS_I(dir
)->root
->root_key
.objectid
,
4825 location
->objectid
, BTRFS_ROOT_REF_KEY
, NULL
);
4832 leaf
= path
->nodes
[0];
4833 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
4834 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
4835 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
4838 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
4839 (unsigned long)(ref
+ 1),
4840 dentry
->d_name
.len
);
4844 btrfs_release_path(path
);
4846 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
4847 if (IS_ERR(new_root
)) {
4848 err
= PTR_ERR(new_root
);
4852 *sub_root
= new_root
;
4853 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
4854 location
->type
= BTRFS_INODE_ITEM_KEY
;
4855 location
->offset
= 0;
4858 btrfs_free_path(path
);
4862 static void inode_tree_add(struct inode
*inode
)
4864 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4865 struct btrfs_inode
*entry
;
4867 struct rb_node
*parent
;
4868 struct rb_node
*new = &BTRFS_I(inode
)->rb_node
;
4869 u64 ino
= btrfs_ino(inode
);
4871 if (inode_unhashed(inode
))
4874 spin_lock(&root
->inode_lock
);
4875 p
= &root
->inode_tree
.rb_node
;
4878 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
4880 if (ino
< btrfs_ino(&entry
->vfs_inode
))
4881 p
= &parent
->rb_left
;
4882 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
4883 p
= &parent
->rb_right
;
4885 WARN_ON(!(entry
->vfs_inode
.i_state
&
4886 (I_WILL_FREE
| I_FREEING
)));
4887 rb_replace_node(parent
, new, &root
->inode_tree
);
4888 RB_CLEAR_NODE(parent
);
4889 spin_unlock(&root
->inode_lock
);
4893 rb_link_node(new, parent
, p
);
4894 rb_insert_color(new, &root
->inode_tree
);
4895 spin_unlock(&root
->inode_lock
);
4898 static void inode_tree_del(struct inode
*inode
)
4900 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4903 spin_lock(&root
->inode_lock
);
4904 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
4905 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4906 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
4907 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4909 spin_unlock(&root
->inode_lock
);
4911 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0) {
4912 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4913 spin_lock(&root
->inode_lock
);
4914 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4915 spin_unlock(&root
->inode_lock
);
4917 btrfs_add_dead_root(root
);
4921 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4923 struct rb_node
*node
;
4924 struct rb_node
*prev
;
4925 struct btrfs_inode
*entry
;
4926 struct inode
*inode
;
4929 if (!test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
4930 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4932 spin_lock(&root
->inode_lock
);
4934 node
= root
->inode_tree
.rb_node
;
4938 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4940 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4941 node
= node
->rb_left
;
4942 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4943 node
= node
->rb_right
;
4949 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4950 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4954 prev
= rb_next(prev
);
4958 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4959 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4960 inode
= igrab(&entry
->vfs_inode
);
4962 spin_unlock(&root
->inode_lock
);
4963 if (atomic_read(&inode
->i_count
) > 1)
4964 d_prune_aliases(inode
);
4966 * btrfs_drop_inode will have it removed from
4967 * the inode cache when its usage count
4972 spin_lock(&root
->inode_lock
);
4976 if (cond_resched_lock(&root
->inode_lock
))
4979 node
= rb_next(node
);
4981 spin_unlock(&root
->inode_lock
);
4984 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4986 struct btrfs_iget_args
*args
= p
;
4987 inode
->i_ino
= args
->location
->objectid
;
4988 memcpy(&BTRFS_I(inode
)->location
, args
->location
,
4989 sizeof(*args
->location
));
4990 BTRFS_I(inode
)->root
= args
->root
;
4994 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4996 struct btrfs_iget_args
*args
= opaque
;
4997 return args
->location
->objectid
== BTRFS_I(inode
)->location
.objectid
&&
4998 args
->root
== BTRFS_I(inode
)->root
;
5001 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
5002 struct btrfs_key
*location
,
5003 struct btrfs_root
*root
)
5005 struct inode
*inode
;
5006 struct btrfs_iget_args args
;
5007 unsigned long hashval
= btrfs_inode_hash(location
->objectid
, root
);
5009 args
.location
= location
;
5012 inode
= iget5_locked(s
, hashval
, btrfs_find_actor
,
5013 btrfs_init_locked_inode
,
5018 /* Get an inode object given its location and corresponding root.
5019 * Returns in *is_new if the inode was read from disk
5021 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5022 struct btrfs_root
*root
, int *new)
5024 struct inode
*inode
;
5026 inode
= btrfs_iget_locked(s
, location
, root
);
5028 return ERR_PTR(-ENOMEM
);
5030 if (inode
->i_state
& I_NEW
) {
5031 btrfs_read_locked_inode(inode
);
5032 if (!is_bad_inode(inode
)) {
5033 inode_tree_add(inode
);
5034 unlock_new_inode(inode
);
5038 unlock_new_inode(inode
);
5040 inode
= ERR_PTR(-ESTALE
);
5047 static struct inode
*new_simple_dir(struct super_block
*s
,
5048 struct btrfs_key
*key
,
5049 struct btrfs_root
*root
)
5051 struct inode
*inode
= new_inode(s
);
5054 return ERR_PTR(-ENOMEM
);
5056 BTRFS_I(inode
)->root
= root
;
5057 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5058 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5060 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5061 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5062 inode
->i_fop
= &simple_dir_operations
;
5063 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5064 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
5069 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5071 struct inode
*inode
;
5072 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5073 struct btrfs_root
*sub_root
= root
;
5074 struct btrfs_key location
;
5078 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5079 return ERR_PTR(-ENAMETOOLONG
);
5081 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5083 return ERR_PTR(ret
);
5085 if (location
.objectid
== 0)
5086 return ERR_PTR(-ENOENT
);
5088 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5089 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5093 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5095 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5096 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5097 &location
, &sub_root
);
5100 inode
= ERR_PTR(ret
);
5102 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5104 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5106 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5108 if (!IS_ERR(inode
) && root
!= sub_root
) {
5109 down_read(&root
->fs_info
->cleanup_work_sem
);
5110 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5111 ret
= btrfs_orphan_cleanup(sub_root
);
5112 up_read(&root
->fs_info
->cleanup_work_sem
);
5115 inode
= ERR_PTR(ret
);
5122 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5124 struct btrfs_root
*root
;
5125 struct inode
*inode
= dentry
->d_inode
;
5127 if (!inode
&& !IS_ROOT(dentry
))
5128 inode
= dentry
->d_parent
->d_inode
;
5131 root
= BTRFS_I(inode
)->root
;
5132 if (btrfs_root_refs(&root
->root_item
) == 0)
5135 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5141 static void btrfs_dentry_release(struct dentry
*dentry
)
5143 if (dentry
->d_fsdata
)
5144 kfree(dentry
->d_fsdata
);
5147 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5150 struct inode
*inode
;
5152 inode
= btrfs_lookup_dentry(dir
, dentry
);
5153 if (IS_ERR(inode
)) {
5154 if (PTR_ERR(inode
) == -ENOENT
)
5157 return ERR_CAST(inode
);
5160 return d_materialise_unique(dentry
, inode
);
5163 unsigned char btrfs_filetype_table
[] = {
5164 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5167 static int btrfs_real_readdir(struct file
*file
, struct dir_context
*ctx
)
5169 struct inode
*inode
= file_inode(file
);
5170 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5171 struct btrfs_item
*item
;
5172 struct btrfs_dir_item
*di
;
5173 struct btrfs_key key
;
5174 struct btrfs_key found_key
;
5175 struct btrfs_path
*path
;
5176 struct list_head ins_list
;
5177 struct list_head del_list
;
5179 struct extent_buffer
*leaf
;
5181 unsigned char d_type
;
5186 int key_type
= BTRFS_DIR_INDEX_KEY
;
5190 int is_curr
= 0; /* ctx->pos points to the current index? */
5192 /* FIXME, use a real flag for deciding about the key type */
5193 if (root
->fs_info
->tree_root
== root
)
5194 key_type
= BTRFS_DIR_ITEM_KEY
;
5196 if (!dir_emit_dots(file
, ctx
))
5199 path
= btrfs_alloc_path();
5205 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5206 INIT_LIST_HEAD(&ins_list
);
5207 INIT_LIST_HEAD(&del_list
);
5208 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5211 btrfs_set_key_type(&key
, key_type
);
5212 key
.offset
= ctx
->pos
;
5213 key
.objectid
= btrfs_ino(inode
);
5215 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5220 leaf
= path
->nodes
[0];
5221 slot
= path
->slots
[0];
5222 if (slot
>= btrfs_header_nritems(leaf
)) {
5223 ret
= btrfs_next_leaf(root
, path
);
5231 item
= btrfs_item_nr(slot
);
5232 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5234 if (found_key
.objectid
!= key
.objectid
)
5236 if (btrfs_key_type(&found_key
) != key_type
)
5238 if (found_key
.offset
< ctx
->pos
)
5240 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5241 btrfs_should_delete_dir_index(&del_list
,
5245 ctx
->pos
= found_key
.offset
;
5248 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5250 di_total
= btrfs_item_size(leaf
, item
);
5252 while (di_cur
< di_total
) {
5253 struct btrfs_key location
;
5255 if (verify_dir_item(root
, leaf
, di
))
5258 name_len
= btrfs_dir_name_len(leaf
, di
);
5259 if (name_len
<= sizeof(tmp_name
)) {
5260 name_ptr
= tmp_name
;
5262 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5268 read_extent_buffer(leaf
, name_ptr
,
5269 (unsigned long)(di
+ 1), name_len
);
5271 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5272 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5275 /* is this a reference to our own snapshot? If so
5278 * In contrast to old kernels, we insert the snapshot's
5279 * dir item and dir index after it has been created, so
5280 * we won't find a reference to our own snapshot. We
5281 * still keep the following code for backward
5284 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5285 location
.objectid
== root
->root_key
.objectid
) {
5289 over
= !dir_emit(ctx
, name_ptr
, name_len
,
5290 location
.objectid
, d_type
);
5293 if (name_ptr
!= tmp_name
)
5298 di_len
= btrfs_dir_name_len(leaf
, di
) +
5299 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5301 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5307 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5310 ret
= btrfs_readdir_delayed_dir_index(ctx
, &ins_list
);
5315 /* Reached end of directory/root. Bump pos past the last item. */
5319 * Stop new entries from being returned after we return the last
5322 * New directory entries are assigned a strictly increasing
5323 * offset. This means that new entries created during readdir
5324 * are *guaranteed* to be seen in the future by that readdir.
5325 * This has broken buggy programs which operate on names as
5326 * they're returned by readdir. Until we re-use freed offsets
5327 * we have this hack to stop new entries from being returned
5328 * under the assumption that they'll never reach this huge
5331 * This is being careful not to overflow 32bit loff_t unless the
5332 * last entry requires it because doing so has broken 32bit apps
5335 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5336 if (ctx
->pos
>= INT_MAX
)
5337 ctx
->pos
= LLONG_MAX
;
5344 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5345 btrfs_put_delayed_items(&ins_list
, &del_list
);
5346 btrfs_free_path(path
);
5350 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5352 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5353 struct btrfs_trans_handle
*trans
;
5355 bool nolock
= false;
5357 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5360 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5363 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5365 trans
= btrfs_join_transaction_nolock(root
);
5367 trans
= btrfs_join_transaction(root
);
5369 return PTR_ERR(trans
);
5370 ret
= btrfs_commit_transaction(trans
, root
);
5376 * This is somewhat expensive, updating the tree every time the
5377 * inode changes. But, it is most likely to find the inode in cache.
5378 * FIXME, needs more benchmarking...there are no reasons other than performance
5379 * to keep or drop this code.
5381 static int btrfs_dirty_inode(struct inode
*inode
)
5383 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5384 struct btrfs_trans_handle
*trans
;
5387 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5390 trans
= btrfs_join_transaction(root
);
5392 return PTR_ERR(trans
);
5394 ret
= btrfs_update_inode(trans
, root
, inode
);
5395 if (ret
&& ret
== -ENOSPC
) {
5396 /* whoops, lets try again with the full transaction */
5397 btrfs_end_transaction(trans
, root
);
5398 trans
= btrfs_start_transaction(root
, 1);
5400 return PTR_ERR(trans
);
5402 ret
= btrfs_update_inode(trans
, root
, inode
);
5404 btrfs_end_transaction(trans
, root
);
5405 if (BTRFS_I(inode
)->delayed_node
)
5406 btrfs_balance_delayed_items(root
);
5412 * This is a copy of file_update_time. We need this so we can return error on
5413 * ENOSPC for updating the inode in the case of file write and mmap writes.
5415 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5418 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5420 if (btrfs_root_readonly(root
))
5423 if (flags
& S_VERSION
)
5424 inode_inc_iversion(inode
);
5425 if (flags
& S_CTIME
)
5426 inode
->i_ctime
= *now
;
5427 if (flags
& S_MTIME
)
5428 inode
->i_mtime
= *now
;
5429 if (flags
& S_ATIME
)
5430 inode
->i_atime
= *now
;
5431 return btrfs_dirty_inode(inode
);
5435 * find the highest existing sequence number in a directory
5436 * and then set the in-memory index_cnt variable to reflect
5437 * free sequence numbers
5439 static int btrfs_set_inode_index_count(struct inode
*inode
)
5441 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5442 struct btrfs_key key
, found_key
;
5443 struct btrfs_path
*path
;
5444 struct extent_buffer
*leaf
;
5447 key
.objectid
= btrfs_ino(inode
);
5448 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
5449 key
.offset
= (u64
)-1;
5451 path
= btrfs_alloc_path();
5455 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5458 /* FIXME: we should be able to handle this */
5464 * MAGIC NUMBER EXPLANATION:
5465 * since we search a directory based on f_pos we have to start at 2
5466 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5467 * else has to start at 2
5469 if (path
->slots
[0] == 0) {
5470 BTRFS_I(inode
)->index_cnt
= 2;
5476 leaf
= path
->nodes
[0];
5477 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5479 if (found_key
.objectid
!= btrfs_ino(inode
) ||
5480 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
5481 BTRFS_I(inode
)->index_cnt
= 2;
5485 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
5487 btrfs_free_path(path
);
5492 * helper to find a free sequence number in a given directory. This current
5493 * code is very simple, later versions will do smarter things in the btree
5495 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
5499 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
5500 ret
= btrfs_inode_delayed_dir_index_count(dir
);
5502 ret
= btrfs_set_inode_index_count(dir
);
5508 *index
= BTRFS_I(dir
)->index_cnt
;
5509 BTRFS_I(dir
)->index_cnt
++;
5514 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
5515 struct btrfs_root
*root
,
5517 const char *name
, int name_len
,
5518 u64 ref_objectid
, u64 objectid
,
5519 umode_t mode
, u64
*index
)
5521 struct inode
*inode
;
5522 struct btrfs_inode_item
*inode_item
;
5523 struct btrfs_key
*location
;
5524 struct btrfs_path
*path
;
5525 struct btrfs_inode_ref
*ref
;
5526 struct btrfs_key key
[2];
5531 path
= btrfs_alloc_path();
5533 return ERR_PTR(-ENOMEM
);
5535 inode
= new_inode(root
->fs_info
->sb
);
5537 btrfs_free_path(path
);
5538 return ERR_PTR(-ENOMEM
);
5542 * we have to initialize this early, so we can reclaim the inode
5543 * number if we fail afterwards in this function.
5545 inode
->i_ino
= objectid
;
5548 trace_btrfs_inode_request(dir
);
5550 ret
= btrfs_set_inode_index(dir
, index
);
5552 btrfs_free_path(path
);
5554 return ERR_PTR(ret
);
5558 * index_cnt is ignored for everything but a dir,
5559 * btrfs_get_inode_index_count has an explanation for the magic
5562 BTRFS_I(inode
)->index_cnt
= 2;
5563 BTRFS_I(inode
)->dir_index
= *index
;
5564 BTRFS_I(inode
)->root
= root
;
5565 BTRFS_I(inode
)->generation
= trans
->transid
;
5566 inode
->i_generation
= BTRFS_I(inode
)->generation
;
5569 * We could have gotten an inode number from somebody who was fsynced
5570 * and then removed in this same transaction, so let's just set full
5571 * sync since it will be a full sync anyway and this will blow away the
5572 * old info in the log.
5574 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
5576 key
[0].objectid
= objectid
;
5577 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
5581 * Start new inodes with an inode_ref. This is slightly more
5582 * efficient for small numbers of hard links since they will
5583 * be packed into one item. Extended refs will kick in if we
5584 * add more hard links than can fit in the ref item.
5586 key
[1].objectid
= objectid
;
5587 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
5588 key
[1].offset
= ref_objectid
;
5590 sizes
[0] = sizeof(struct btrfs_inode_item
);
5591 sizes
[1] = name_len
+ sizeof(*ref
);
5593 path
->leave_spinning
= 1;
5594 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
5598 inode_init_owner(inode
, dir
, mode
);
5599 inode_set_bytes(inode
, 0);
5600 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
5601 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5602 struct btrfs_inode_item
);
5603 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
5604 sizeof(*inode_item
));
5605 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
5607 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
5608 struct btrfs_inode_ref
);
5609 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
5610 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
5611 ptr
= (unsigned long)(ref
+ 1);
5612 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
5614 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5615 btrfs_free_path(path
);
5617 location
= &BTRFS_I(inode
)->location
;
5618 location
->objectid
= objectid
;
5619 location
->offset
= 0;
5620 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
5622 btrfs_inherit_iflags(inode
, dir
);
5624 if (S_ISREG(mode
)) {
5625 if (btrfs_test_opt(root
, NODATASUM
))
5626 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
5627 if (btrfs_test_opt(root
, NODATACOW
))
5628 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
5629 BTRFS_INODE_NODATASUM
;
5632 btrfs_insert_inode_hash(inode
);
5633 inode_tree_add(inode
);
5635 trace_btrfs_inode_new(inode
);
5636 btrfs_set_inode_last_trans(trans
, inode
);
5638 btrfs_update_root_times(trans
, root
);
5640 ret
= btrfs_inode_inherit_props(trans
, inode
, dir
);
5642 btrfs_err(root
->fs_info
,
5643 "error inheriting props for ino %llu (root %llu): %d",
5644 btrfs_ino(inode
), root
->root_key
.objectid
, ret
);
5649 BTRFS_I(dir
)->index_cnt
--;
5650 btrfs_free_path(path
);
5652 return ERR_PTR(ret
);
5655 static inline u8
btrfs_inode_type(struct inode
*inode
)
5657 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
5661 * utility function to add 'inode' into 'parent_inode' with
5662 * a give name and a given sequence number.
5663 * if 'add_backref' is true, also insert a backref from the
5664 * inode to the parent directory.
5666 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
5667 struct inode
*parent_inode
, struct inode
*inode
,
5668 const char *name
, int name_len
, int add_backref
, u64 index
)
5671 struct btrfs_key key
;
5672 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
5673 u64 ino
= btrfs_ino(inode
);
5674 u64 parent_ino
= btrfs_ino(parent_inode
);
5676 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5677 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
5680 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
5684 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5685 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
5686 key
.objectid
, root
->root_key
.objectid
,
5687 parent_ino
, index
, name
, name_len
);
5688 } else if (add_backref
) {
5689 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
5693 /* Nothing to clean up yet */
5697 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
5699 btrfs_inode_type(inode
), index
);
5700 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
5703 btrfs_abort_transaction(trans
, root
, ret
);
5707 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
5709 inode_inc_iversion(parent_inode
);
5710 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
5711 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
5713 btrfs_abort_transaction(trans
, root
, ret
);
5717 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5720 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
5721 key
.objectid
, root
->root_key
.objectid
,
5722 parent_ino
, &local_index
, name
, name_len
);
5724 } else if (add_backref
) {
5728 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
5729 ino
, parent_ino
, &local_index
);
5734 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
5735 struct inode
*dir
, struct dentry
*dentry
,
5736 struct inode
*inode
, int backref
, u64 index
)
5738 int err
= btrfs_add_link(trans
, dir
, inode
,
5739 dentry
->d_name
.name
, dentry
->d_name
.len
,
5746 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
5747 umode_t mode
, dev_t rdev
)
5749 struct btrfs_trans_handle
*trans
;
5750 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5751 struct inode
*inode
= NULL
;
5757 if (!new_valid_dev(rdev
))
5761 * 2 for inode item and ref
5763 * 1 for xattr if selinux is on
5765 trans
= btrfs_start_transaction(root
, 5);
5767 return PTR_ERR(trans
);
5769 err
= btrfs_find_free_ino(root
, &objectid
);
5773 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5774 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5776 if (IS_ERR(inode
)) {
5777 err
= PTR_ERR(inode
);
5781 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5788 * If the active LSM wants to access the inode during
5789 * d_instantiate it needs these. Smack checks to see
5790 * if the filesystem supports xattrs by looking at the
5794 inode
->i_op
= &btrfs_special_inode_operations
;
5795 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5799 init_special_inode(inode
, inode
->i_mode
, rdev
);
5800 btrfs_update_inode(trans
, root
, inode
);
5801 d_instantiate(dentry
, inode
);
5804 btrfs_end_transaction(trans
, root
);
5805 btrfs_balance_delayed_items(root
);
5806 btrfs_btree_balance_dirty(root
);
5808 inode_dec_link_count(inode
);
5814 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
5815 umode_t mode
, bool excl
)
5817 struct btrfs_trans_handle
*trans
;
5818 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5819 struct inode
*inode
= NULL
;
5820 int drop_inode_on_err
= 0;
5826 * 2 for inode item and ref
5828 * 1 for xattr if selinux is on
5830 trans
= btrfs_start_transaction(root
, 5);
5832 return PTR_ERR(trans
);
5834 err
= btrfs_find_free_ino(root
, &objectid
);
5838 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5839 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5841 if (IS_ERR(inode
)) {
5842 err
= PTR_ERR(inode
);
5845 drop_inode_on_err
= 1;
5847 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5851 err
= btrfs_update_inode(trans
, root
, inode
);
5856 * If the active LSM wants to access the inode during
5857 * d_instantiate it needs these. Smack checks to see
5858 * if the filesystem supports xattrs by looking at the
5861 inode
->i_fop
= &btrfs_file_operations
;
5862 inode
->i_op
= &btrfs_file_inode_operations
;
5864 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5868 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5869 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5870 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5871 d_instantiate(dentry
, inode
);
5874 btrfs_end_transaction(trans
, root
);
5875 if (err
&& drop_inode_on_err
) {
5876 inode_dec_link_count(inode
);
5879 btrfs_balance_delayed_items(root
);
5880 btrfs_btree_balance_dirty(root
);
5884 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
5885 struct dentry
*dentry
)
5887 struct btrfs_trans_handle
*trans
;
5888 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5889 struct inode
*inode
= old_dentry
->d_inode
;
5894 /* do not allow sys_link's with other subvols of the same device */
5895 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
5898 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
5901 err
= btrfs_set_inode_index(dir
, &index
);
5906 * 2 items for inode and inode ref
5907 * 2 items for dir items
5908 * 1 item for parent inode
5910 trans
= btrfs_start_transaction(root
, 5);
5911 if (IS_ERR(trans
)) {
5912 err
= PTR_ERR(trans
);
5916 /* There are several dir indexes for this inode, clear the cache. */
5917 BTRFS_I(inode
)->dir_index
= 0ULL;
5919 inode_inc_iversion(inode
);
5920 inode
->i_ctime
= CURRENT_TIME
;
5922 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
5924 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5929 struct dentry
*parent
= dentry
->d_parent
;
5930 err
= btrfs_update_inode(trans
, root
, inode
);
5933 d_instantiate(dentry
, inode
);
5934 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5937 btrfs_end_transaction(trans
, root
);
5938 btrfs_balance_delayed_items(root
);
5941 inode_dec_link_count(inode
);
5944 btrfs_btree_balance_dirty(root
);
5948 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5950 struct inode
*inode
= NULL
;
5951 struct btrfs_trans_handle
*trans
;
5952 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5954 int drop_on_err
= 0;
5959 * 2 items for inode and ref
5960 * 2 items for dir items
5961 * 1 for xattr if selinux is on
5963 trans
= btrfs_start_transaction(root
, 5);
5965 return PTR_ERR(trans
);
5967 err
= btrfs_find_free_ino(root
, &objectid
);
5971 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5972 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5973 S_IFDIR
| mode
, &index
);
5974 if (IS_ERR(inode
)) {
5975 err
= PTR_ERR(inode
);
5981 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5985 inode
->i_op
= &btrfs_dir_inode_operations
;
5986 inode
->i_fop
= &btrfs_dir_file_operations
;
5988 btrfs_i_size_write(inode
, 0);
5989 err
= btrfs_update_inode(trans
, root
, inode
);
5993 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5994 dentry
->d_name
.len
, 0, index
);
5998 d_instantiate(dentry
, inode
);
6002 btrfs_end_transaction(trans
, root
);
6005 btrfs_balance_delayed_items(root
);
6006 btrfs_btree_balance_dirty(root
);
6010 /* helper for btfs_get_extent. Given an existing extent in the tree,
6011 * and an extent that you want to insert, deal with overlap and insert
6012 * the new extent into the tree.
6014 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
6015 struct extent_map
*existing
,
6016 struct extent_map
*em
,
6017 u64 map_start
, u64 map_len
)
6021 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
6022 start_diff
= map_start
- em
->start
;
6023 em
->start
= map_start
;
6025 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
6026 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
6027 em
->block_start
+= start_diff
;
6028 em
->block_len
-= start_diff
;
6030 return add_extent_mapping(em_tree
, em
, 0);
6033 static noinline
int uncompress_inline(struct btrfs_path
*path
,
6034 struct inode
*inode
, struct page
*page
,
6035 size_t pg_offset
, u64 extent_offset
,
6036 struct btrfs_file_extent_item
*item
)
6039 struct extent_buffer
*leaf
= path
->nodes
[0];
6042 unsigned long inline_size
;
6046 WARN_ON(pg_offset
!= 0);
6047 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6048 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6049 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6050 btrfs_item_nr(path
->slots
[0]));
6051 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6054 ptr
= btrfs_file_extent_inline_start(item
);
6056 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6058 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6059 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6060 extent_offset
, inline_size
, max_size
);
6062 char *kaddr
= kmap_atomic(page
);
6063 unsigned long copy_size
= min_t(u64
,
6064 PAGE_CACHE_SIZE
- pg_offset
,
6065 max_size
- extent_offset
);
6066 memset(kaddr
+ pg_offset
, 0, copy_size
);
6067 kunmap_atomic(kaddr
);
6074 * a bit scary, this does extent mapping from logical file offset to the disk.
6075 * the ugly parts come from merging extents from the disk with the in-ram
6076 * representation. This gets more complex because of the data=ordered code,
6077 * where the in-ram extents might be locked pending data=ordered completion.
6079 * This also copies inline extents directly into the page.
6082 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6083 size_t pg_offset
, u64 start
, u64 len
,
6089 u64 extent_start
= 0;
6091 u64 objectid
= btrfs_ino(inode
);
6093 struct btrfs_path
*path
= NULL
;
6094 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6095 struct btrfs_file_extent_item
*item
;
6096 struct extent_buffer
*leaf
;
6097 struct btrfs_key found_key
;
6098 struct extent_map
*em
= NULL
;
6099 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6100 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6101 struct btrfs_trans_handle
*trans
= NULL
;
6105 read_lock(&em_tree
->lock
);
6106 em
= lookup_extent_mapping(em_tree
, start
, len
);
6108 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6109 read_unlock(&em_tree
->lock
);
6112 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6113 free_extent_map(em
);
6114 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6115 free_extent_map(em
);
6119 em
= alloc_extent_map();
6124 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6125 em
->start
= EXTENT_MAP_HOLE
;
6126 em
->orig_start
= EXTENT_MAP_HOLE
;
6128 em
->block_len
= (u64
)-1;
6131 path
= btrfs_alloc_path();
6137 * Chances are we'll be called again, so go ahead and do
6143 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6144 objectid
, start
, trans
!= NULL
);
6151 if (path
->slots
[0] == 0)
6156 leaf
= path
->nodes
[0];
6157 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6158 struct btrfs_file_extent_item
);
6159 /* are we inside the extent that was found? */
6160 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6161 found_type
= btrfs_key_type(&found_key
);
6162 if (found_key
.objectid
!= objectid
||
6163 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6165 * If we backup past the first extent we want to move forward
6166 * and see if there is an extent in front of us, otherwise we'll
6167 * say there is a hole for our whole search range which can
6174 found_type
= btrfs_file_extent_type(leaf
, item
);
6175 extent_start
= found_key
.offset
;
6176 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6177 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6178 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6179 extent_end
= extent_start
+
6180 btrfs_file_extent_num_bytes(leaf
, item
);
6181 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6183 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6184 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6187 if (start
>= extent_end
) {
6189 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6190 ret
= btrfs_next_leaf(root
, path
);
6197 leaf
= path
->nodes
[0];
6199 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6200 if (found_key
.objectid
!= objectid
||
6201 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6203 if (start
+ len
<= found_key
.offset
)
6206 em
->orig_start
= start
;
6207 em
->len
= found_key
.offset
- start
;
6211 em
->ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, item
);
6212 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6213 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6214 em
->start
= extent_start
;
6215 em
->len
= extent_end
- extent_start
;
6216 em
->orig_start
= extent_start
-
6217 btrfs_file_extent_offset(leaf
, item
);
6218 em
->orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
6220 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
6222 em
->block_start
= EXTENT_MAP_HOLE
;
6225 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
6226 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6227 em
->compress_type
= compress_type
;
6228 em
->block_start
= bytenr
;
6229 em
->block_len
= em
->orig_block_len
;
6231 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
6232 em
->block_start
= bytenr
;
6233 em
->block_len
= em
->len
;
6234 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
6235 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6238 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6242 size_t extent_offset
;
6245 em
->block_start
= EXTENT_MAP_INLINE
;
6246 if (!page
|| create
) {
6247 em
->start
= extent_start
;
6248 em
->len
= extent_end
- extent_start
;
6252 size
= btrfs_file_extent_inline_len(leaf
, path
->slots
[0], item
);
6253 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6254 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6255 size
- extent_offset
);
6256 em
->start
= extent_start
+ extent_offset
;
6257 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6258 em
->orig_block_len
= em
->len
;
6259 em
->orig_start
= em
->start
;
6260 if (compress_type
) {
6261 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6262 em
->compress_type
= compress_type
;
6264 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6265 if (create
== 0 && !PageUptodate(page
)) {
6266 if (btrfs_file_extent_compression(leaf
, item
) !=
6267 BTRFS_COMPRESS_NONE
) {
6268 ret
= uncompress_inline(path
, inode
, page
,
6270 extent_offset
, item
);
6271 BUG_ON(ret
); /* -ENOMEM */
6274 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6276 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6277 memset(map
+ pg_offset
+ copy_size
, 0,
6278 PAGE_CACHE_SIZE
- pg_offset
-
6283 flush_dcache_page(page
);
6284 } else if (create
&& PageUptodate(page
)) {
6288 free_extent_map(em
);
6291 btrfs_release_path(path
);
6292 trans
= btrfs_join_transaction(root
);
6295 return ERR_CAST(trans
);
6299 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6302 btrfs_mark_buffer_dirty(leaf
);
6304 set_extent_uptodate(io_tree
, em
->start
,
6305 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6308 WARN(1, KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
6312 em
->orig_start
= start
;
6315 em
->block_start
= EXTENT_MAP_HOLE
;
6316 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6318 btrfs_release_path(path
);
6319 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6320 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6321 em
->start
, em
->len
, start
, len
);
6327 write_lock(&em_tree
->lock
);
6328 ret
= add_extent_mapping(em_tree
, em
, 0);
6329 /* it is possible that someone inserted the extent into the tree
6330 * while we had the lock dropped. It is also possible that
6331 * an overlapping map exists in the tree
6333 if (ret
== -EEXIST
) {
6334 struct extent_map
*existing
;
6338 existing
= lookup_extent_mapping(em_tree
, start
, len
);
6339 if (existing
&& (existing
->start
> start
||
6340 existing
->start
+ existing
->len
<= start
)) {
6341 free_extent_map(existing
);
6345 existing
= lookup_extent_mapping(em_tree
, em
->start
,
6348 err
= merge_extent_mapping(em_tree
, existing
,
6351 free_extent_map(existing
);
6353 free_extent_map(em
);
6358 free_extent_map(em
);
6362 free_extent_map(em
);
6367 write_unlock(&em_tree
->lock
);
6370 trace_btrfs_get_extent(root
, em
);
6373 btrfs_free_path(path
);
6375 ret
= btrfs_end_transaction(trans
, root
);
6380 free_extent_map(em
);
6381 return ERR_PTR(err
);
6383 BUG_ON(!em
); /* Error is always set */
6387 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6388 size_t pg_offset
, u64 start
, u64 len
,
6391 struct extent_map
*em
;
6392 struct extent_map
*hole_em
= NULL
;
6393 u64 range_start
= start
;
6399 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6406 * - a pre-alloc extent,
6407 * there might actually be delalloc bytes behind it.
6409 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
6410 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6416 /* check to see if we've wrapped (len == -1 or similar) */
6425 /* ok, we didn't find anything, lets look for delalloc */
6426 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
6427 end
, len
, EXTENT_DELALLOC
, 1);
6428 found_end
= range_start
+ found
;
6429 if (found_end
< range_start
)
6430 found_end
= (u64
)-1;
6433 * we didn't find anything useful, return
6434 * the original results from get_extent()
6436 if (range_start
> end
|| found_end
<= start
) {
6442 /* adjust the range_start to make sure it doesn't
6443 * go backwards from the start they passed in
6445 range_start
= max(start
, range_start
);
6446 found
= found_end
- range_start
;
6449 u64 hole_start
= start
;
6452 em
= alloc_extent_map();
6458 * when btrfs_get_extent can't find anything it
6459 * returns one huge hole
6461 * make sure what it found really fits our range, and
6462 * adjust to make sure it is based on the start from
6466 u64 calc_end
= extent_map_end(hole_em
);
6468 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
6469 free_extent_map(hole_em
);
6472 hole_start
= max(hole_em
->start
, start
);
6473 hole_len
= calc_end
- hole_start
;
6477 if (hole_em
&& range_start
> hole_start
) {
6478 /* our hole starts before our delalloc, so we
6479 * have to return just the parts of the hole
6480 * that go until the delalloc starts
6482 em
->len
= min(hole_len
,
6483 range_start
- hole_start
);
6484 em
->start
= hole_start
;
6485 em
->orig_start
= hole_start
;
6487 * don't adjust block start at all,
6488 * it is fixed at EXTENT_MAP_HOLE
6490 em
->block_start
= hole_em
->block_start
;
6491 em
->block_len
= hole_len
;
6492 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
6493 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6495 em
->start
= range_start
;
6497 em
->orig_start
= range_start
;
6498 em
->block_start
= EXTENT_MAP_DELALLOC
;
6499 em
->block_len
= found
;
6501 } else if (hole_em
) {
6506 free_extent_map(hole_em
);
6508 free_extent_map(em
);
6509 return ERR_PTR(err
);
6514 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
6517 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6518 struct extent_map
*em
;
6519 struct btrfs_key ins
;
6523 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
6524 ret
= btrfs_reserve_extent(root
, len
, root
->sectorsize
, 0,
6525 alloc_hint
, &ins
, 1);
6527 return ERR_PTR(ret
);
6529 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
6530 ins
.offset
, ins
.offset
, ins
.offset
, 0);
6532 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
6536 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
6537 ins
.offset
, ins
.offset
, 0);
6539 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
6540 free_extent_map(em
);
6541 return ERR_PTR(ret
);
6548 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6549 * block must be cow'd
6551 noinline
int can_nocow_extent(struct inode
*inode
, u64 offset
, u64
*len
,
6552 u64
*orig_start
, u64
*orig_block_len
,
6555 struct btrfs_trans_handle
*trans
;
6556 struct btrfs_path
*path
;
6558 struct extent_buffer
*leaf
;
6559 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6560 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6561 struct btrfs_file_extent_item
*fi
;
6562 struct btrfs_key key
;
6569 bool nocow
= (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
);
6571 path
= btrfs_alloc_path();
6575 ret
= btrfs_lookup_file_extent(NULL
, root
, path
, btrfs_ino(inode
),
6580 slot
= path
->slots
[0];
6583 /* can't find the item, must cow */
6590 leaf
= path
->nodes
[0];
6591 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
6592 if (key
.objectid
!= btrfs_ino(inode
) ||
6593 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6594 /* not our file or wrong item type, must cow */
6598 if (key
.offset
> offset
) {
6599 /* Wrong offset, must cow */
6603 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
6604 found_type
= btrfs_file_extent_type(leaf
, fi
);
6605 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
6606 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
6607 /* not a regular extent, must cow */
6611 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_REG
)
6614 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
6615 if (extent_end
<= offset
)
6618 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6619 if (disk_bytenr
== 0)
6622 if (btrfs_file_extent_compression(leaf
, fi
) ||
6623 btrfs_file_extent_encryption(leaf
, fi
) ||
6624 btrfs_file_extent_other_encoding(leaf
, fi
))
6627 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
6630 *orig_start
= key
.offset
- backref_offset
;
6631 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6632 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6635 if (btrfs_extent_readonly(root
, disk_bytenr
))
6638 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
6639 if (!nocow
&& found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6642 range_end
= round_up(offset
+ num_bytes
, root
->sectorsize
) - 1;
6643 ret
= test_range_bit(io_tree
, offset
, range_end
,
6644 EXTENT_DELALLOC
, 0, NULL
);
6651 btrfs_release_path(path
);
6654 * look for other files referencing this extent, if we
6655 * find any we must cow
6657 trans
= btrfs_join_transaction(root
);
6658 if (IS_ERR(trans
)) {
6663 ret
= btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
6664 key
.offset
- backref_offset
, disk_bytenr
);
6665 btrfs_end_transaction(trans
, root
);
6672 * adjust disk_bytenr and num_bytes to cover just the bytes
6673 * in this extent we are about to write. If there
6674 * are any csums in that range we have to cow in order
6675 * to keep the csums correct
6677 disk_bytenr
+= backref_offset
;
6678 disk_bytenr
+= offset
- key
.offset
;
6679 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
6682 * all of the above have passed, it is safe to overwrite this extent
6688 btrfs_free_path(path
);
6692 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
6693 struct extent_state
**cached_state
, int writing
)
6695 struct btrfs_ordered_extent
*ordered
;
6699 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6702 * We're concerned with the entire range that we're going to be
6703 * doing DIO to, so we need to make sure theres no ordered
6704 * extents in this range.
6706 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6707 lockend
- lockstart
+ 1);
6710 * We need to make sure there are no buffered pages in this
6711 * range either, we could have raced between the invalidate in
6712 * generic_file_direct_write and locking the extent. The
6713 * invalidate needs to happen so that reads after a write do not
6716 if (!ordered
&& (!writing
||
6717 !test_range_bit(&BTRFS_I(inode
)->io_tree
,
6718 lockstart
, lockend
, EXTENT_UPTODATE
, 0,
6722 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6723 cached_state
, GFP_NOFS
);
6726 btrfs_start_ordered_extent(inode
, ordered
, 1);
6727 btrfs_put_ordered_extent(ordered
);
6729 /* Screw you mmap */
6730 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
6737 * If we found a page that couldn't be invalidated just
6738 * fall back to buffered.
6740 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6741 lockstart
>> PAGE_CACHE_SHIFT
,
6742 lockend
>> PAGE_CACHE_SHIFT
);
6753 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
6754 u64 len
, u64 orig_start
,
6755 u64 block_start
, u64 block_len
,
6756 u64 orig_block_len
, u64 ram_bytes
,
6759 struct extent_map_tree
*em_tree
;
6760 struct extent_map
*em
;
6761 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6764 em_tree
= &BTRFS_I(inode
)->extent_tree
;
6765 em
= alloc_extent_map();
6767 return ERR_PTR(-ENOMEM
);
6770 em
->orig_start
= orig_start
;
6771 em
->mod_start
= start
;
6774 em
->block_len
= block_len
;
6775 em
->block_start
= block_start
;
6776 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6777 em
->orig_block_len
= orig_block_len
;
6778 em
->ram_bytes
= ram_bytes
;
6779 em
->generation
= -1;
6780 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6781 if (type
== BTRFS_ORDERED_PREALLOC
)
6782 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
6785 btrfs_drop_extent_cache(inode
, em
->start
,
6786 em
->start
+ em
->len
- 1, 0);
6787 write_lock(&em_tree
->lock
);
6788 ret
= add_extent_mapping(em_tree
, em
, 1);
6789 write_unlock(&em_tree
->lock
);
6790 } while (ret
== -EEXIST
);
6793 free_extent_map(em
);
6794 return ERR_PTR(ret
);
6801 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
6802 struct buffer_head
*bh_result
, int create
)
6804 struct extent_map
*em
;
6805 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6806 struct extent_state
*cached_state
= NULL
;
6807 u64 start
= iblock
<< inode
->i_blkbits
;
6808 u64 lockstart
, lockend
;
6809 u64 len
= bh_result
->b_size
;
6810 int unlock_bits
= EXTENT_LOCKED
;
6814 unlock_bits
|= EXTENT_DELALLOC
| EXTENT_DIRTY
;
6816 len
= min_t(u64
, len
, root
->sectorsize
);
6819 lockend
= start
+ len
- 1;
6822 * If this errors out it's because we couldn't invalidate pagecache for
6823 * this range and we need to fallback to buffered.
6825 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
6828 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
6835 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6836 * io. INLINE is special, and we could probably kludge it in here, but
6837 * it's still buffered so for safety lets just fall back to the generic
6840 * For COMPRESSED we _have_ to read the entire extent in so we can
6841 * decompress it, so there will be buffering required no matter what we
6842 * do, so go ahead and fallback to buffered.
6844 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6845 * to buffered IO. Don't blame me, this is the price we pay for using
6848 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
6849 em
->block_start
== EXTENT_MAP_INLINE
) {
6850 free_extent_map(em
);
6855 /* Just a good old fashioned hole, return */
6856 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
6857 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
6858 free_extent_map(em
);
6863 * We don't allocate a new extent in the following cases
6865 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6867 * 2) The extent is marked as PREALLOC. We're good to go here and can
6868 * just use the extent.
6872 len
= min(len
, em
->len
- (start
- em
->start
));
6873 lockstart
= start
+ len
;
6877 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
6878 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
6879 em
->block_start
!= EXTENT_MAP_HOLE
)) {
6882 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
6884 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6885 type
= BTRFS_ORDERED_PREALLOC
;
6887 type
= BTRFS_ORDERED_NOCOW
;
6888 len
= min(len
, em
->len
- (start
- em
->start
));
6889 block_start
= em
->block_start
+ (start
- em
->start
);
6891 if (can_nocow_extent(inode
, start
, &len
, &orig_start
,
6892 &orig_block_len
, &ram_bytes
) == 1) {
6893 if (type
== BTRFS_ORDERED_PREALLOC
) {
6894 free_extent_map(em
);
6895 em
= create_pinned_em(inode
, start
, len
,
6904 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
6905 block_start
, len
, len
, type
);
6907 free_extent_map(em
);
6915 * this will cow the extent, reset the len in case we changed
6918 len
= bh_result
->b_size
;
6919 free_extent_map(em
);
6920 em
= btrfs_new_extent_direct(inode
, start
, len
);
6925 len
= min(len
, em
->len
- (start
- em
->start
));
6927 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
6929 bh_result
->b_size
= len
;
6930 bh_result
->b_bdev
= em
->bdev
;
6931 set_buffer_mapped(bh_result
);
6933 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6934 set_buffer_new(bh_result
);
6937 * Need to update the i_size under the extent lock so buffered
6938 * readers will get the updated i_size when we unlock.
6940 if (start
+ len
> i_size_read(inode
))
6941 i_size_write(inode
, start
+ len
);
6943 spin_lock(&BTRFS_I(inode
)->lock
);
6944 BTRFS_I(inode
)->outstanding_extents
++;
6945 spin_unlock(&BTRFS_I(inode
)->lock
);
6947 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6948 lockstart
+ len
- 1, EXTENT_DELALLOC
, NULL
,
6949 &cached_state
, GFP_NOFS
);
6954 * In the case of write we need to clear and unlock the entire range,
6955 * in the case of read we need to unlock only the end area that we
6956 * aren't using if there is any left over space.
6958 if (lockstart
< lockend
) {
6959 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6960 lockend
, unlock_bits
, 1, 0,
6961 &cached_state
, GFP_NOFS
);
6963 free_extent_state(cached_state
);
6966 free_extent_map(em
);
6971 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6972 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
6976 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
6978 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6979 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
6980 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6981 struct inode
*inode
= dip
->inode
;
6982 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6983 struct bio
*dio_bio
;
6984 u32
*csums
= (u32
*)dip
->csum
;
6988 start
= dip
->logical_offset
;
6990 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
6991 struct page
*page
= bvec
->bv_page
;
6994 unsigned long flags
;
6996 local_irq_save(flags
);
6997 kaddr
= kmap_atomic(page
);
6998 csum
= btrfs_csum_data(kaddr
+ bvec
->bv_offset
,
6999 csum
, bvec
->bv_len
);
7000 btrfs_csum_final(csum
, (char *)&csum
);
7001 kunmap_atomic(kaddr
);
7002 local_irq_restore(flags
);
7004 flush_dcache_page(bvec
->bv_page
);
7005 if (csum
!= csums
[index
]) {
7006 btrfs_err(root
->fs_info
, "csum failed ino %llu off %llu csum %u expected csum %u",
7007 btrfs_ino(inode
), start
, csum
,
7013 start
+= bvec
->bv_len
;
7016 } while (bvec
<= bvec_end
);
7018 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
7019 dip
->logical_offset
+ dip
->bytes
- 1);
7020 dio_bio
= dip
->dio_bio
;
7024 /* If we had a csum failure make sure to clear the uptodate flag */
7026 clear_bit(BIO_UPTODATE
, &dio_bio
->bi_flags
);
7027 dio_end_io(dio_bio
, err
);
7031 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
7033 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7034 struct inode
*inode
= dip
->inode
;
7035 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7036 struct btrfs_ordered_extent
*ordered
= NULL
;
7037 u64 ordered_offset
= dip
->logical_offset
;
7038 u64 ordered_bytes
= dip
->bytes
;
7039 struct bio
*dio_bio
;
7045 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
7047 ordered_bytes
, !err
);
7051 ordered
->work
.func
= finish_ordered_fn
;
7052 ordered
->work
.flags
= 0;
7053 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
7057 * our bio might span multiple ordered extents. If we haven't
7058 * completed the accounting for the whole dio, go back and try again
7060 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
7061 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
7067 dio_bio
= dip
->dio_bio
;
7071 /* If we had an error make sure to clear the uptodate flag */
7073 clear_bit(BIO_UPTODATE
, &dio_bio
->bi_flags
);
7074 dio_end_io(dio_bio
, err
);
7078 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
7079 struct bio
*bio
, int mirror_num
,
7080 unsigned long bio_flags
, u64 offset
)
7083 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7084 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
7085 BUG_ON(ret
); /* -ENOMEM */
7089 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
7091 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7094 btrfs_err(BTRFS_I(dip
->inode
)->root
->fs_info
,
7095 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7096 btrfs_ino(dip
->inode
), bio
->bi_rw
,
7097 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
7101 * before atomic variable goto zero, we must make sure
7102 * dip->errors is perceived to be set.
7104 smp_mb__before_atomic_dec();
7107 /* if there are more bios still pending for this dio, just exit */
7108 if (!atomic_dec_and_test(&dip
->pending_bios
))
7112 bio_io_error(dip
->orig_bio
);
7114 set_bit(BIO_UPTODATE
, &dip
->dio_bio
->bi_flags
);
7115 bio_endio(dip
->orig_bio
, 0);
7121 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
7122 u64 first_sector
, gfp_t gfp_flags
)
7124 int nr_vecs
= bio_get_nr_vecs(bdev
);
7125 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
7128 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
7129 int rw
, u64 file_offset
, int skip_sum
,
7132 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7133 int write
= rw
& REQ_WRITE
;
7134 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7138 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
7143 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
7151 if (write
&& async_submit
) {
7152 ret
= btrfs_wq_submit_bio(root
->fs_info
,
7153 inode
, rw
, bio
, 0, 0,
7155 __btrfs_submit_bio_start_direct_io
,
7156 __btrfs_submit_bio_done
);
7160 * If we aren't doing async submit, calculate the csum of the
7163 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
7166 } else if (!skip_sum
) {
7167 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, dip
, bio
,
7174 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
7180 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
7183 struct inode
*inode
= dip
->inode
;
7184 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7186 struct bio
*orig_bio
= dip
->orig_bio
;
7187 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
7188 u64 start_sector
= orig_bio
->bi_sector
;
7189 u64 file_offset
= dip
->logical_offset
;
7194 int async_submit
= 0;
7196 map_length
= orig_bio
->bi_size
;
7197 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
7198 &map_length
, NULL
, 0);
7204 if (map_length
>= orig_bio
->bi_size
) {
7209 /* async crcs make it difficult to collect full stripe writes. */
7210 if (btrfs_get_alloc_profile(root
, 1) &
7211 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
))
7216 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
7219 bio
->bi_private
= dip
;
7220 bio
->bi_end_io
= btrfs_end_dio_bio
;
7221 atomic_inc(&dip
->pending_bios
);
7223 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
7224 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
7225 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
7226 bvec
->bv_offset
) < bvec
->bv_len
)) {
7228 * inc the count before we submit the bio so
7229 * we know the end IO handler won't happen before
7230 * we inc the count. Otherwise, the dip might get freed
7231 * before we're done setting it up
7233 atomic_inc(&dip
->pending_bios
);
7234 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
7235 file_offset
, skip_sum
,
7239 atomic_dec(&dip
->pending_bios
);
7243 start_sector
+= submit_len
>> 9;
7244 file_offset
+= submit_len
;
7249 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
7250 start_sector
, GFP_NOFS
);
7253 bio
->bi_private
= dip
;
7254 bio
->bi_end_io
= btrfs_end_dio_bio
;
7256 map_length
= orig_bio
->bi_size
;
7257 ret
= btrfs_map_block(root
->fs_info
, rw
,
7259 &map_length
, NULL
, 0);
7265 submit_len
+= bvec
->bv_len
;
7272 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
7281 * before atomic variable goto zero, we must
7282 * make sure dip->errors is perceived to be set.
7284 smp_mb__before_atomic_dec();
7285 if (atomic_dec_and_test(&dip
->pending_bios
))
7286 bio_io_error(dip
->orig_bio
);
7288 /* bio_end_io() will handle error, so we needn't return it */
7292 static void btrfs_submit_direct(int rw
, struct bio
*dio_bio
,
7293 struct inode
*inode
, loff_t file_offset
)
7295 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7296 struct btrfs_dio_private
*dip
;
7300 int write
= rw
& REQ_WRITE
;
7304 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7306 io_bio
= btrfs_bio_clone(dio_bio
, GFP_NOFS
);
7312 if (!skip_sum
&& !write
) {
7313 csum_size
= btrfs_super_csum_size(root
->fs_info
->super_copy
);
7314 sum_len
= dio_bio
->bi_size
>> inode
->i_sb
->s_blocksize_bits
;
7315 sum_len
*= csum_size
;
7320 dip
= kmalloc(sizeof(*dip
) + sum_len
, GFP_NOFS
);
7326 dip
->private = dio_bio
->bi_private
;
7328 dip
->logical_offset
= file_offset
;
7329 dip
->bytes
= dio_bio
->bi_size
;
7330 dip
->disk_bytenr
= (u64
)dio_bio
->bi_sector
<< 9;
7331 io_bio
->bi_private
= dip
;
7333 dip
->orig_bio
= io_bio
;
7334 dip
->dio_bio
= dio_bio
;
7335 atomic_set(&dip
->pending_bios
, 0);
7338 io_bio
->bi_end_io
= btrfs_endio_direct_write
;
7340 io_bio
->bi_end_io
= btrfs_endio_direct_read
;
7342 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
7351 * If this is a write, we need to clean up the reserved space and kill
7352 * the ordered extent.
7355 struct btrfs_ordered_extent
*ordered
;
7356 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
7357 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
7358 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
7359 btrfs_free_reserved_extent(root
, ordered
->start
,
7361 btrfs_put_ordered_extent(ordered
);
7362 btrfs_put_ordered_extent(ordered
);
7364 bio_endio(dio_bio
, ret
);
7367 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
7368 const struct iovec
*iov
, loff_t offset
,
7369 unsigned long nr_segs
)
7375 unsigned blocksize_mask
= root
->sectorsize
- 1;
7376 ssize_t retval
= -EINVAL
;
7377 loff_t end
= offset
;
7379 if (offset
& blocksize_mask
)
7382 /* Check the memory alignment. Blocks cannot straddle pages */
7383 for (seg
= 0; seg
< nr_segs
; seg
++) {
7384 addr
= (unsigned long)iov
[seg
].iov_base
;
7385 size
= iov
[seg
].iov_len
;
7387 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
7390 /* If this is a write we don't need to check anymore */
7395 * Check to make sure we don't have duplicate iov_base's in this
7396 * iovec, if so return EINVAL, otherwise we'll get csum errors
7397 * when reading back.
7399 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
7400 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
7409 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
7410 const struct iovec
*iov
, loff_t offset
,
7411 unsigned long nr_segs
)
7413 struct file
*file
= iocb
->ki_filp
;
7414 struct inode
*inode
= file
->f_mapping
->host
;
7418 bool relock
= false;
7421 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
7425 atomic_inc(&inode
->i_dio_count
);
7426 smp_mb__after_atomic_inc();
7429 * The generic stuff only does filemap_write_and_wait_range, which isn't
7430 * enough if we've written compressed pages to this area, so we need to
7431 * call btrfs_wait_ordered_range to make absolutely sure that any
7432 * outstanding dirty pages are on disk.
7434 count
= iov_length(iov
, nr_segs
);
7435 ret
= btrfs_wait_ordered_range(inode
, offset
, count
);
7441 * If the write DIO is beyond the EOF, we need update
7442 * the isize, but it is protected by i_mutex. So we can
7443 * not unlock the i_mutex at this case.
7445 if (offset
+ count
<= inode
->i_size
) {
7446 mutex_unlock(&inode
->i_mutex
);
7449 ret
= btrfs_delalloc_reserve_space(inode
, count
);
7452 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
7453 &BTRFS_I(inode
)->runtime_flags
))) {
7454 inode_dio_done(inode
);
7455 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
7459 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
7460 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
7461 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
7462 btrfs_submit_direct
, flags
);
7464 if (ret
< 0 && ret
!= -EIOCBQUEUED
)
7465 btrfs_delalloc_release_space(inode
, count
);
7466 else if (ret
>= 0 && (size_t)ret
< count
)
7467 btrfs_delalloc_release_space(inode
,
7468 count
- (size_t)ret
);
7470 btrfs_delalloc_release_metadata(inode
, 0);
7474 inode_dio_done(inode
);
7476 mutex_lock(&inode
->i_mutex
);
7481 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7483 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
7484 __u64 start
, __u64 len
)
7488 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
7492 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
7495 int btrfs_readpage(struct file
*file
, struct page
*page
)
7497 struct extent_io_tree
*tree
;
7498 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7499 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
7502 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
7504 struct extent_io_tree
*tree
;
7507 if (current
->flags
& PF_MEMALLOC
) {
7508 redirty_page_for_writepage(wbc
, page
);
7512 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7513 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
7516 static int btrfs_writepages(struct address_space
*mapping
,
7517 struct writeback_control
*wbc
)
7519 struct extent_io_tree
*tree
;
7521 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7522 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
7526 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
7527 struct list_head
*pages
, unsigned nr_pages
)
7529 struct extent_io_tree
*tree
;
7530 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7531 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
7534 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7536 struct extent_io_tree
*tree
;
7537 struct extent_map_tree
*map
;
7540 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7541 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
7542 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
7544 ClearPagePrivate(page
);
7545 set_page_private(page
, 0);
7546 page_cache_release(page
);
7551 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7553 if (PageWriteback(page
) || PageDirty(page
))
7555 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
7558 static void btrfs_invalidatepage(struct page
*page
, unsigned int offset
,
7559 unsigned int length
)
7561 struct inode
*inode
= page
->mapping
->host
;
7562 struct extent_io_tree
*tree
;
7563 struct btrfs_ordered_extent
*ordered
;
7564 struct extent_state
*cached_state
= NULL
;
7565 u64 page_start
= page_offset(page
);
7566 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7567 int inode_evicting
= inode
->i_state
& I_FREEING
;
7570 * we have the page locked, so new writeback can't start,
7571 * and the dirty bit won't be cleared while we are here.
7573 * Wait for IO on this page so that we can safely clear
7574 * the PagePrivate2 bit and do ordered accounting
7576 wait_on_page_writeback(page
);
7578 tree
= &BTRFS_I(inode
)->io_tree
;
7580 btrfs_releasepage(page
, GFP_NOFS
);
7584 if (!inode_evicting
)
7585 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
7586 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
7589 * IO on this page will never be started, so we need
7590 * to account for any ordered extents now
7592 if (!inode_evicting
)
7593 clear_extent_bit(tree
, page_start
, page_end
,
7594 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7595 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
7596 EXTENT_DEFRAG
, 1, 0, &cached_state
,
7599 * whoever cleared the private bit is responsible
7600 * for the finish_ordered_io
7602 if (TestClearPagePrivate2(page
)) {
7603 struct btrfs_ordered_inode_tree
*tree
;
7606 tree
= &BTRFS_I(inode
)->ordered_tree
;
7608 spin_lock_irq(&tree
->lock
);
7609 set_bit(BTRFS_ORDERED_TRUNCATED
, &ordered
->flags
);
7610 new_len
= page_start
- ordered
->file_offset
;
7611 if (new_len
< ordered
->truncated_len
)
7612 ordered
->truncated_len
= new_len
;
7613 spin_unlock_irq(&tree
->lock
);
7615 if (btrfs_dec_test_ordered_pending(inode
, &ordered
,
7617 PAGE_CACHE_SIZE
, 1))
7618 btrfs_finish_ordered_io(ordered
);
7620 btrfs_put_ordered_extent(ordered
);
7621 if (!inode_evicting
) {
7622 cached_state
= NULL
;
7623 lock_extent_bits(tree
, page_start
, page_end
, 0,
7628 if (!inode_evicting
) {
7629 clear_extent_bit(tree
, page_start
, page_end
,
7630 EXTENT_LOCKED
| EXTENT_DIRTY
|
7631 EXTENT_DELALLOC
| EXTENT_DO_ACCOUNTING
|
7632 EXTENT_DEFRAG
, 1, 1,
7633 &cached_state
, GFP_NOFS
);
7635 __btrfs_releasepage(page
, GFP_NOFS
);
7638 ClearPageChecked(page
);
7639 if (PagePrivate(page
)) {
7640 ClearPagePrivate(page
);
7641 set_page_private(page
, 0);
7642 page_cache_release(page
);
7647 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7648 * called from a page fault handler when a page is first dirtied. Hence we must
7649 * be careful to check for EOF conditions here. We set the page up correctly
7650 * for a written page which means we get ENOSPC checking when writing into
7651 * holes and correct delalloc and unwritten extent mapping on filesystems that
7652 * support these features.
7654 * We are not allowed to take the i_mutex here so we have to play games to
7655 * protect against truncate races as the page could now be beyond EOF. Because
7656 * vmtruncate() writes the inode size before removing pages, once we have the
7657 * page lock we can determine safely if the page is beyond EOF. If it is not
7658 * beyond EOF, then the page is guaranteed safe against truncation until we
7661 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
7663 struct page
*page
= vmf
->page
;
7664 struct inode
*inode
= file_inode(vma
->vm_file
);
7665 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7666 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7667 struct btrfs_ordered_extent
*ordered
;
7668 struct extent_state
*cached_state
= NULL
;
7670 unsigned long zero_start
;
7677 sb_start_pagefault(inode
->i_sb
);
7678 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
7680 ret
= file_update_time(vma
->vm_file
);
7686 else /* -ENOSPC, -EIO, etc */
7687 ret
= VM_FAULT_SIGBUS
;
7693 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
7696 size
= i_size_read(inode
);
7697 page_start
= page_offset(page
);
7698 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7700 if ((page
->mapping
!= inode
->i_mapping
) ||
7701 (page_start
>= size
)) {
7702 /* page got truncated out from underneath us */
7705 wait_on_page_writeback(page
);
7707 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
7708 set_page_extent_mapped(page
);
7711 * we can't set the delalloc bits if there are pending ordered
7712 * extents. Drop our locks and wait for them to finish
7714 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
7716 unlock_extent_cached(io_tree
, page_start
, page_end
,
7717 &cached_state
, GFP_NOFS
);
7719 btrfs_start_ordered_extent(inode
, ordered
, 1);
7720 btrfs_put_ordered_extent(ordered
);
7725 * XXX - page_mkwrite gets called every time the page is dirtied, even
7726 * if it was already dirty, so for space accounting reasons we need to
7727 * clear any delalloc bits for the range we are fixing to save. There
7728 * is probably a better way to do this, but for now keep consistent with
7729 * prepare_pages in the normal write path.
7731 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
7732 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7733 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
7734 0, 0, &cached_state
, GFP_NOFS
);
7736 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
7739 unlock_extent_cached(io_tree
, page_start
, page_end
,
7740 &cached_state
, GFP_NOFS
);
7741 ret
= VM_FAULT_SIGBUS
;
7746 /* page is wholly or partially inside EOF */
7747 if (page_start
+ PAGE_CACHE_SIZE
> size
)
7748 zero_start
= size
& ~PAGE_CACHE_MASK
;
7750 zero_start
= PAGE_CACHE_SIZE
;
7752 if (zero_start
!= PAGE_CACHE_SIZE
) {
7754 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
7755 flush_dcache_page(page
);
7758 ClearPageChecked(page
);
7759 set_page_dirty(page
);
7760 SetPageUptodate(page
);
7762 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
7763 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
7764 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
7766 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
7770 sb_end_pagefault(inode
->i_sb
);
7771 return VM_FAULT_LOCKED
;
7775 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
7777 sb_end_pagefault(inode
->i_sb
);
7781 static int btrfs_truncate(struct inode
*inode
)
7783 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7784 struct btrfs_block_rsv
*rsv
;
7787 struct btrfs_trans_handle
*trans
;
7788 u64 mask
= root
->sectorsize
- 1;
7789 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
7791 ret
= btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
),
7797 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7798 * 3 things going on here
7800 * 1) We need to reserve space for our orphan item and the space to
7801 * delete our orphan item. Lord knows we don't want to have a dangling
7802 * orphan item because we didn't reserve space to remove it.
7804 * 2) We need to reserve space to update our inode.
7806 * 3) We need to have something to cache all the space that is going to
7807 * be free'd up by the truncate operation, but also have some slack
7808 * space reserved in case it uses space during the truncate (thank you
7809 * very much snapshotting).
7811 * And we need these to all be seperate. The fact is we can use alot of
7812 * space doing the truncate, and we have no earthly idea how much space
7813 * we will use, so we need the truncate reservation to be seperate so it
7814 * doesn't end up using space reserved for updating the inode or
7815 * removing the orphan item. We also need to be able to stop the
7816 * transaction and start a new one, which means we need to be able to
7817 * update the inode several times, and we have no idea of knowing how
7818 * many times that will be, so we can't just reserve 1 item for the
7819 * entirety of the opration, so that has to be done seperately as well.
7820 * Then there is the orphan item, which does indeed need to be held on
7821 * to for the whole operation, and we need nobody to touch this reserved
7822 * space except the orphan code.
7824 * So that leaves us with
7826 * 1) root->orphan_block_rsv - for the orphan deletion.
7827 * 2) rsv - for the truncate reservation, which we will steal from the
7828 * transaction reservation.
7829 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7830 * updating the inode.
7832 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
7835 rsv
->size
= min_size
;
7839 * 1 for the truncate slack space
7840 * 1 for updating the inode.
7842 trans
= btrfs_start_transaction(root
, 2);
7843 if (IS_ERR(trans
)) {
7844 err
= PTR_ERR(trans
);
7848 /* Migrate the slack space for the truncate to our reserve */
7849 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
7854 * setattr is responsible for setting the ordered_data_close flag,
7855 * but that is only tested during the last file release. That
7856 * could happen well after the next commit, leaving a great big
7857 * window where new writes may get lost if someone chooses to write
7858 * to this file after truncating to zero
7860 * The inode doesn't have any dirty data here, and so if we commit
7861 * this is a noop. If someone immediately starts writing to the inode
7862 * it is very likely we'll catch some of their writes in this
7863 * transaction, and the commit will find this file on the ordered
7864 * data list with good things to send down.
7866 * This is a best effort solution, there is still a window where
7867 * using truncate to replace the contents of the file will
7868 * end up with a zero length file after a crash.
7870 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
7871 &BTRFS_I(inode
)->runtime_flags
))
7872 btrfs_add_ordered_operation(trans
, root
, inode
);
7875 * So if we truncate and then write and fsync we normally would just
7876 * write the extents that changed, which is a problem if we need to
7877 * first truncate that entire inode. So set this flag so we write out
7878 * all of the extents in the inode to the sync log so we're completely
7881 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
7882 trans
->block_rsv
= rsv
;
7885 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
7887 BTRFS_EXTENT_DATA_KEY
);
7888 if (ret
!= -ENOSPC
) {
7893 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7894 ret
= btrfs_update_inode(trans
, root
, inode
);
7900 btrfs_end_transaction(trans
, root
);
7901 btrfs_btree_balance_dirty(root
);
7903 trans
= btrfs_start_transaction(root
, 2);
7904 if (IS_ERR(trans
)) {
7905 ret
= err
= PTR_ERR(trans
);
7910 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
7912 BUG_ON(ret
); /* shouldn't happen */
7913 trans
->block_rsv
= rsv
;
7916 if (ret
== 0 && inode
->i_nlink
> 0) {
7917 trans
->block_rsv
= root
->orphan_block_rsv
;
7918 ret
= btrfs_orphan_del(trans
, inode
);
7924 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7925 ret
= btrfs_update_inode(trans
, root
, inode
);
7929 ret
= btrfs_end_transaction(trans
, root
);
7930 btrfs_btree_balance_dirty(root
);
7934 btrfs_free_block_rsv(root
, rsv
);
7943 * create a new subvolume directory/inode (helper for the ioctl).
7945 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
7946 struct btrfs_root
*new_root
,
7947 struct btrfs_root
*parent_root
,
7950 struct inode
*inode
;
7954 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
7955 new_dirid
, new_dirid
,
7956 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
7959 return PTR_ERR(inode
);
7960 inode
->i_op
= &btrfs_dir_inode_operations
;
7961 inode
->i_fop
= &btrfs_dir_file_operations
;
7963 set_nlink(inode
, 1);
7964 btrfs_i_size_write(inode
, 0);
7966 err
= btrfs_subvol_inherit_props(trans
, new_root
, parent_root
);
7968 btrfs_err(new_root
->fs_info
,
7969 "error inheriting subvolume %llu properties: %d\n",
7970 new_root
->root_key
.objectid
, err
);
7972 err
= btrfs_update_inode(trans
, new_root
, inode
);
7978 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
7980 struct btrfs_inode
*ei
;
7981 struct inode
*inode
;
7983 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
7990 ei
->last_sub_trans
= 0;
7991 ei
->logged_trans
= 0;
7992 ei
->delalloc_bytes
= 0;
7993 ei
->disk_i_size
= 0;
7996 ei
->index_cnt
= (u64
)-1;
7998 ei
->last_unlink_trans
= 0;
7999 ei
->last_log_commit
= 0;
8001 spin_lock_init(&ei
->lock
);
8002 ei
->outstanding_extents
= 0;
8003 ei
->reserved_extents
= 0;
8005 ei
->runtime_flags
= 0;
8006 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
8008 ei
->delayed_node
= NULL
;
8010 inode
= &ei
->vfs_inode
;
8011 extent_map_tree_init(&ei
->extent_tree
);
8012 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
8013 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
8014 ei
->io_tree
.track_uptodate
= 1;
8015 ei
->io_failure_tree
.track_uptodate
= 1;
8016 atomic_set(&ei
->sync_writers
, 0);
8017 mutex_init(&ei
->log_mutex
);
8018 mutex_init(&ei
->delalloc_mutex
);
8019 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
8020 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
8021 INIT_LIST_HEAD(&ei
->ordered_operations
);
8022 RB_CLEAR_NODE(&ei
->rb_node
);
8027 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8028 void btrfs_test_destroy_inode(struct inode
*inode
)
8030 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
8031 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8035 static void btrfs_i_callback(struct rcu_head
*head
)
8037 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
8038 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
8041 void btrfs_destroy_inode(struct inode
*inode
)
8043 struct btrfs_ordered_extent
*ordered
;
8044 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8046 WARN_ON(!hlist_empty(&inode
->i_dentry
));
8047 WARN_ON(inode
->i_data
.nrpages
);
8048 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
8049 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
8050 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
8051 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
8054 * This can happen where we create an inode, but somebody else also
8055 * created the same inode and we need to destroy the one we already
8062 * Make sure we're properly removed from the ordered operation
8066 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
8067 spin_lock(&root
->fs_info
->ordered_root_lock
);
8068 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
8069 spin_unlock(&root
->fs_info
->ordered_root_lock
);
8072 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
8073 &BTRFS_I(inode
)->runtime_flags
)) {
8074 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
8076 atomic_dec(&root
->orphan_inodes
);
8080 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
8084 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
8085 ordered
->file_offset
, ordered
->len
);
8086 btrfs_remove_ordered_extent(inode
, ordered
);
8087 btrfs_put_ordered_extent(ordered
);
8088 btrfs_put_ordered_extent(ordered
);
8091 inode_tree_del(inode
);
8092 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
8094 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
8097 int btrfs_drop_inode(struct inode
*inode
)
8099 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8104 /* the snap/subvol tree is on deleting */
8105 if (btrfs_root_refs(&root
->root_item
) == 0)
8108 return generic_drop_inode(inode
);
8111 static void init_once(void *foo
)
8113 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
8115 inode_init_once(&ei
->vfs_inode
);
8118 void btrfs_destroy_cachep(void)
8121 * Make sure all delayed rcu free inodes are flushed before we
8125 if (btrfs_inode_cachep
)
8126 kmem_cache_destroy(btrfs_inode_cachep
);
8127 if (btrfs_trans_handle_cachep
)
8128 kmem_cache_destroy(btrfs_trans_handle_cachep
);
8129 if (btrfs_transaction_cachep
)
8130 kmem_cache_destroy(btrfs_transaction_cachep
);
8131 if (btrfs_path_cachep
)
8132 kmem_cache_destroy(btrfs_path_cachep
);
8133 if (btrfs_free_space_cachep
)
8134 kmem_cache_destroy(btrfs_free_space_cachep
);
8135 if (btrfs_delalloc_work_cachep
)
8136 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
8139 int btrfs_init_cachep(void)
8141 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
8142 sizeof(struct btrfs_inode
), 0,
8143 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
8144 if (!btrfs_inode_cachep
)
8147 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
8148 sizeof(struct btrfs_trans_handle
), 0,
8149 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8150 if (!btrfs_trans_handle_cachep
)
8153 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
8154 sizeof(struct btrfs_transaction
), 0,
8155 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8156 if (!btrfs_transaction_cachep
)
8159 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
8160 sizeof(struct btrfs_path
), 0,
8161 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8162 if (!btrfs_path_cachep
)
8165 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
8166 sizeof(struct btrfs_free_space
), 0,
8167 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8168 if (!btrfs_free_space_cachep
)
8171 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
8172 sizeof(struct btrfs_delalloc_work
), 0,
8173 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
8175 if (!btrfs_delalloc_work_cachep
)
8180 btrfs_destroy_cachep();
8184 static int btrfs_getattr(struct vfsmount
*mnt
,
8185 struct dentry
*dentry
, struct kstat
*stat
)
8188 struct inode
*inode
= dentry
->d_inode
;
8189 u32 blocksize
= inode
->i_sb
->s_blocksize
;
8191 generic_fillattr(inode
, stat
);
8192 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
8193 stat
->blksize
= PAGE_CACHE_SIZE
;
8195 spin_lock(&BTRFS_I(inode
)->lock
);
8196 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
8197 spin_unlock(&BTRFS_I(inode
)->lock
);
8198 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
8199 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
8203 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
8204 struct inode
*new_dir
, struct dentry
*new_dentry
)
8206 struct btrfs_trans_handle
*trans
;
8207 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
8208 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
8209 struct inode
*new_inode
= new_dentry
->d_inode
;
8210 struct inode
*old_inode
= old_dentry
->d_inode
;
8211 struct timespec ctime
= CURRENT_TIME
;
8215 u64 old_ino
= btrfs_ino(old_inode
);
8217 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
8220 /* we only allow rename subvolume link between subvolumes */
8221 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
8224 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
8225 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
8228 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
8229 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
8233 /* check for collisions, even if the name isn't there */
8234 ret
= btrfs_check_dir_item_collision(dest
, new_dir
->i_ino
,
8235 new_dentry
->d_name
.name
,
8236 new_dentry
->d_name
.len
);
8239 if (ret
== -EEXIST
) {
8241 * eexist without a new_inode */
8242 if (WARN_ON(!new_inode
)) {
8246 /* maybe -EOVERFLOW */
8253 * we're using rename to replace one file with another.
8254 * and the replacement file is large. Start IO on it now so
8255 * we don't add too much work to the end of the transaction
8257 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
8258 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
8259 filemap_flush(old_inode
->i_mapping
);
8261 /* close the racy window with snapshot create/destroy ioctl */
8262 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8263 down_read(&root
->fs_info
->subvol_sem
);
8265 * We want to reserve the absolute worst case amount of items. So if
8266 * both inodes are subvols and we need to unlink them then that would
8267 * require 4 item modifications, but if they are both normal inodes it
8268 * would require 5 item modifications, so we'll assume their normal
8269 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8270 * should cover the worst case number of items we'll modify.
8272 trans
= btrfs_start_transaction(root
, 11);
8273 if (IS_ERR(trans
)) {
8274 ret
= PTR_ERR(trans
);
8279 btrfs_record_root_in_trans(trans
, dest
);
8281 ret
= btrfs_set_inode_index(new_dir
, &index
);
8285 BTRFS_I(old_inode
)->dir_index
= 0ULL;
8286 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8287 /* force full log commit if subvolume involved. */
8288 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8290 ret
= btrfs_insert_inode_ref(trans
, dest
,
8291 new_dentry
->d_name
.name
,
8292 new_dentry
->d_name
.len
,
8294 btrfs_ino(new_dir
), index
);
8298 * this is an ugly little race, but the rename is required
8299 * to make sure that if we crash, the inode is either at the
8300 * old name or the new one. pinning the log transaction lets
8301 * us make sure we don't allow a log commit to come in after
8302 * we unlink the name but before we add the new name back in.
8304 btrfs_pin_log_trans(root
);
8307 * make sure the inode gets flushed if it is replacing
8310 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
8311 btrfs_add_ordered_operation(trans
, root
, old_inode
);
8313 inode_inc_iversion(old_dir
);
8314 inode_inc_iversion(new_dir
);
8315 inode_inc_iversion(old_inode
);
8316 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
8317 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
8318 old_inode
->i_ctime
= ctime
;
8320 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
8321 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
8323 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8324 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
8325 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
8326 old_dentry
->d_name
.name
,
8327 old_dentry
->d_name
.len
);
8329 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
8330 old_dentry
->d_inode
,
8331 old_dentry
->d_name
.name
,
8332 old_dentry
->d_name
.len
);
8334 ret
= btrfs_update_inode(trans
, root
, old_inode
);
8337 btrfs_abort_transaction(trans
, root
, ret
);
8342 inode_inc_iversion(new_inode
);
8343 new_inode
->i_ctime
= CURRENT_TIME
;
8344 if (unlikely(btrfs_ino(new_inode
) ==
8345 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
8346 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
8347 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
8349 new_dentry
->d_name
.name
,
8350 new_dentry
->d_name
.len
);
8351 BUG_ON(new_inode
->i_nlink
== 0);
8353 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
8354 new_dentry
->d_inode
,
8355 new_dentry
->d_name
.name
,
8356 new_dentry
->d_name
.len
);
8358 if (!ret
&& new_inode
->i_nlink
== 0)
8359 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
8361 btrfs_abort_transaction(trans
, root
, ret
);
8366 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
8367 new_dentry
->d_name
.name
,
8368 new_dentry
->d_name
.len
, 0, index
);
8370 btrfs_abort_transaction(trans
, root
, ret
);
8374 if (old_inode
->i_nlink
== 1)
8375 BTRFS_I(old_inode
)->dir_index
= index
;
8377 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
8378 struct dentry
*parent
= new_dentry
->d_parent
;
8379 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
8380 btrfs_end_log_trans(root
);
8383 btrfs_end_transaction(trans
, root
);
8385 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8386 up_read(&root
->fs_info
->subvol_sem
);
8391 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
8393 struct btrfs_delalloc_work
*delalloc_work
;
8394 struct inode
*inode
;
8396 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
8398 inode
= delalloc_work
->inode
;
8399 if (delalloc_work
->wait
) {
8400 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
8402 filemap_flush(inode
->i_mapping
);
8403 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
8404 &BTRFS_I(inode
)->runtime_flags
))
8405 filemap_flush(inode
->i_mapping
);
8408 if (delalloc_work
->delay_iput
)
8409 btrfs_add_delayed_iput(inode
);
8412 complete(&delalloc_work
->completion
);
8415 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
8416 int wait
, int delay_iput
)
8418 struct btrfs_delalloc_work
*work
;
8420 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
8424 init_completion(&work
->completion
);
8425 INIT_LIST_HEAD(&work
->list
);
8426 work
->inode
= inode
;
8428 work
->delay_iput
= delay_iput
;
8429 work
->work
.func
= btrfs_run_delalloc_work
;
8434 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
8436 wait_for_completion(&work
->completion
);
8437 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
8441 * some fairly slow code that needs optimization. This walks the list
8442 * of all the inodes with pending delalloc and forces them to disk.
8444 static int __start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
8446 struct btrfs_inode
*binode
;
8447 struct inode
*inode
;
8448 struct btrfs_delalloc_work
*work
, *next
;
8449 struct list_head works
;
8450 struct list_head splice
;
8453 INIT_LIST_HEAD(&works
);
8454 INIT_LIST_HEAD(&splice
);
8456 spin_lock(&root
->delalloc_lock
);
8457 list_splice_init(&root
->delalloc_inodes
, &splice
);
8458 while (!list_empty(&splice
)) {
8459 binode
= list_entry(splice
.next
, struct btrfs_inode
,
8462 list_move_tail(&binode
->delalloc_inodes
,
8463 &root
->delalloc_inodes
);
8464 inode
= igrab(&binode
->vfs_inode
);
8466 cond_resched_lock(&root
->delalloc_lock
);
8469 spin_unlock(&root
->delalloc_lock
);
8471 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
8472 if (unlikely(!work
)) {
8474 btrfs_add_delayed_iput(inode
);
8480 list_add_tail(&work
->list
, &works
);
8481 btrfs_queue_worker(&root
->fs_info
->flush_workers
,
8485 spin_lock(&root
->delalloc_lock
);
8487 spin_unlock(&root
->delalloc_lock
);
8489 list_for_each_entry_safe(work
, next
, &works
, list
) {
8490 list_del_init(&work
->list
);
8491 btrfs_wait_and_free_delalloc_work(work
);
8495 list_for_each_entry_safe(work
, next
, &works
, list
) {
8496 list_del_init(&work
->list
);
8497 btrfs_wait_and_free_delalloc_work(work
);
8500 if (!list_empty_careful(&splice
)) {
8501 spin_lock(&root
->delalloc_lock
);
8502 list_splice_tail(&splice
, &root
->delalloc_inodes
);
8503 spin_unlock(&root
->delalloc_lock
);
8508 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
8512 if (test_bit(BTRFS_FS_STATE_ERROR
, &root
->fs_info
->fs_state
))
8515 ret
= __start_delalloc_inodes(root
, delay_iput
);
8517 * the filemap_flush will queue IO into the worker threads, but
8518 * we have to make sure the IO is actually started and that
8519 * ordered extents get created before we return
8521 atomic_inc(&root
->fs_info
->async_submit_draining
);
8522 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
8523 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
8524 wait_event(root
->fs_info
->async_submit_wait
,
8525 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
8526 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
8528 atomic_dec(&root
->fs_info
->async_submit_draining
);
8532 int btrfs_start_delalloc_roots(struct btrfs_fs_info
*fs_info
, int delay_iput
)
8534 struct btrfs_root
*root
;
8535 struct list_head splice
;
8538 if (test_bit(BTRFS_FS_STATE_ERROR
, &fs_info
->fs_state
))
8541 INIT_LIST_HEAD(&splice
);
8543 spin_lock(&fs_info
->delalloc_root_lock
);
8544 list_splice_init(&fs_info
->delalloc_roots
, &splice
);
8545 while (!list_empty(&splice
)) {
8546 root
= list_first_entry(&splice
, struct btrfs_root
,
8548 root
= btrfs_grab_fs_root(root
);
8550 list_move_tail(&root
->delalloc_root
,
8551 &fs_info
->delalloc_roots
);
8552 spin_unlock(&fs_info
->delalloc_root_lock
);
8554 ret
= __start_delalloc_inodes(root
, delay_iput
);
8555 btrfs_put_fs_root(root
);
8559 spin_lock(&fs_info
->delalloc_root_lock
);
8561 spin_unlock(&fs_info
->delalloc_root_lock
);
8563 atomic_inc(&fs_info
->async_submit_draining
);
8564 while (atomic_read(&fs_info
->nr_async_submits
) ||
8565 atomic_read(&fs_info
->async_delalloc_pages
)) {
8566 wait_event(fs_info
->async_submit_wait
,
8567 (atomic_read(&fs_info
->nr_async_submits
) == 0 &&
8568 atomic_read(&fs_info
->async_delalloc_pages
) == 0));
8570 atomic_dec(&fs_info
->async_submit_draining
);
8573 if (!list_empty_careful(&splice
)) {
8574 spin_lock(&fs_info
->delalloc_root_lock
);
8575 list_splice_tail(&splice
, &fs_info
->delalloc_roots
);
8576 spin_unlock(&fs_info
->delalloc_root_lock
);
8581 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
8582 const char *symname
)
8584 struct btrfs_trans_handle
*trans
;
8585 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
8586 struct btrfs_path
*path
;
8587 struct btrfs_key key
;
8588 struct inode
*inode
= NULL
;
8596 struct btrfs_file_extent_item
*ei
;
8597 struct extent_buffer
*leaf
;
8599 name_len
= strlen(symname
);
8600 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
8601 return -ENAMETOOLONG
;
8604 * 2 items for inode item and ref
8605 * 2 items for dir items
8606 * 1 item for xattr if selinux is on
8608 trans
= btrfs_start_transaction(root
, 5);
8610 return PTR_ERR(trans
);
8612 err
= btrfs_find_free_ino(root
, &objectid
);
8616 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
8617 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
8618 S_IFLNK
|S_IRWXUGO
, &index
);
8619 if (IS_ERR(inode
)) {
8620 err
= PTR_ERR(inode
);
8624 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
8631 * If the active LSM wants to access the inode during
8632 * d_instantiate it needs these. Smack checks to see
8633 * if the filesystem supports xattrs by looking at the
8636 inode
->i_fop
= &btrfs_file_operations
;
8637 inode
->i_op
= &btrfs_file_inode_operations
;
8639 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
8643 inode
->i_mapping
->a_ops
= &btrfs_aops
;
8644 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8645 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
8650 path
= btrfs_alloc_path();
8656 key
.objectid
= btrfs_ino(inode
);
8658 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
8659 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
8660 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
8664 btrfs_free_path(path
);
8667 leaf
= path
->nodes
[0];
8668 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
8669 struct btrfs_file_extent_item
);
8670 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
8671 btrfs_set_file_extent_type(leaf
, ei
,
8672 BTRFS_FILE_EXTENT_INLINE
);
8673 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
8674 btrfs_set_file_extent_compression(leaf
, ei
, 0);
8675 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
8676 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
8678 ptr
= btrfs_file_extent_inline_start(ei
);
8679 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
8680 btrfs_mark_buffer_dirty(leaf
);
8681 btrfs_free_path(path
);
8683 inode
->i_op
= &btrfs_symlink_inode_operations
;
8684 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
8685 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8686 inode_set_bytes(inode
, name_len
);
8687 btrfs_i_size_write(inode
, name_len
);
8688 err
= btrfs_update_inode(trans
, root
, inode
);
8694 d_instantiate(dentry
, inode
);
8695 btrfs_end_transaction(trans
, root
);
8697 inode_dec_link_count(inode
);
8700 btrfs_btree_balance_dirty(root
);
8704 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8705 u64 start
, u64 num_bytes
, u64 min_size
,
8706 loff_t actual_len
, u64
*alloc_hint
,
8707 struct btrfs_trans_handle
*trans
)
8709 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
8710 struct extent_map
*em
;
8711 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8712 struct btrfs_key ins
;
8713 u64 cur_offset
= start
;
8717 bool own_trans
= true;
8721 while (num_bytes
> 0) {
8723 trans
= btrfs_start_transaction(root
, 3);
8724 if (IS_ERR(trans
)) {
8725 ret
= PTR_ERR(trans
);
8730 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
8731 cur_bytes
= max(cur_bytes
, min_size
);
8732 ret
= btrfs_reserve_extent(root
, cur_bytes
, min_size
, 0,
8733 *alloc_hint
, &ins
, 1);
8736 btrfs_end_transaction(trans
, root
);
8740 ret
= insert_reserved_file_extent(trans
, inode
,
8741 cur_offset
, ins
.objectid
,
8742 ins
.offset
, ins
.offset
,
8743 ins
.offset
, 0, 0, 0,
8744 BTRFS_FILE_EXTENT_PREALLOC
);
8746 btrfs_free_reserved_extent(root
, ins
.objectid
,
8748 btrfs_abort_transaction(trans
, root
, ret
);
8750 btrfs_end_transaction(trans
, root
);
8753 btrfs_drop_extent_cache(inode
, cur_offset
,
8754 cur_offset
+ ins
.offset
-1, 0);
8756 em
= alloc_extent_map();
8758 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
8759 &BTRFS_I(inode
)->runtime_flags
);
8763 em
->start
= cur_offset
;
8764 em
->orig_start
= cur_offset
;
8765 em
->len
= ins
.offset
;
8766 em
->block_start
= ins
.objectid
;
8767 em
->block_len
= ins
.offset
;
8768 em
->orig_block_len
= ins
.offset
;
8769 em
->ram_bytes
= ins
.offset
;
8770 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
8771 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
8772 em
->generation
= trans
->transid
;
8775 write_lock(&em_tree
->lock
);
8776 ret
= add_extent_mapping(em_tree
, em
, 1);
8777 write_unlock(&em_tree
->lock
);
8780 btrfs_drop_extent_cache(inode
, cur_offset
,
8781 cur_offset
+ ins
.offset
- 1,
8784 free_extent_map(em
);
8786 num_bytes
-= ins
.offset
;
8787 cur_offset
+= ins
.offset
;
8788 *alloc_hint
= ins
.objectid
+ ins
.offset
;
8790 inode_inc_iversion(inode
);
8791 inode
->i_ctime
= CURRENT_TIME
;
8792 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
8793 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
8794 (actual_len
> inode
->i_size
) &&
8795 (cur_offset
> inode
->i_size
)) {
8796 if (cur_offset
> actual_len
)
8797 i_size
= actual_len
;
8799 i_size
= cur_offset
;
8800 i_size_write(inode
, i_size
);
8801 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
8804 ret
= btrfs_update_inode(trans
, root
, inode
);
8807 btrfs_abort_transaction(trans
, root
, ret
);
8809 btrfs_end_transaction(trans
, root
);
8814 btrfs_end_transaction(trans
, root
);
8819 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8820 u64 start
, u64 num_bytes
, u64 min_size
,
8821 loff_t actual_len
, u64
*alloc_hint
)
8823 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8824 min_size
, actual_len
, alloc_hint
,
8828 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
8829 struct btrfs_trans_handle
*trans
, int mode
,
8830 u64 start
, u64 num_bytes
, u64 min_size
,
8831 loff_t actual_len
, u64
*alloc_hint
)
8833 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8834 min_size
, actual_len
, alloc_hint
, trans
);
8837 static int btrfs_set_page_dirty(struct page
*page
)
8839 return __set_page_dirty_nobuffers(page
);
8842 static int btrfs_permission(struct inode
*inode
, int mask
)
8844 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8845 umode_t mode
= inode
->i_mode
;
8847 if (mask
& MAY_WRITE
&&
8848 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
8849 if (btrfs_root_readonly(root
))
8851 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
8854 return generic_permission(inode
, mask
);
8857 static const struct inode_operations btrfs_dir_inode_operations
= {
8858 .getattr
= btrfs_getattr
,
8859 .lookup
= btrfs_lookup
,
8860 .create
= btrfs_create
,
8861 .unlink
= btrfs_unlink
,
8863 .mkdir
= btrfs_mkdir
,
8864 .rmdir
= btrfs_rmdir
,
8865 .rename
= btrfs_rename
,
8866 .symlink
= btrfs_symlink
,
8867 .setattr
= btrfs_setattr
,
8868 .mknod
= btrfs_mknod
,
8869 .setxattr
= btrfs_setxattr
,
8870 .getxattr
= btrfs_getxattr
,
8871 .listxattr
= btrfs_listxattr
,
8872 .removexattr
= btrfs_removexattr
,
8873 .permission
= btrfs_permission
,
8874 .get_acl
= btrfs_get_acl
,
8875 .update_time
= btrfs_update_time
,
8877 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
8878 .lookup
= btrfs_lookup
,
8879 .permission
= btrfs_permission
,
8880 .get_acl
= btrfs_get_acl
,
8881 .update_time
= btrfs_update_time
,
8884 static const struct file_operations btrfs_dir_file_operations
= {
8885 .llseek
= generic_file_llseek
,
8886 .read
= generic_read_dir
,
8887 .iterate
= btrfs_real_readdir
,
8888 .unlocked_ioctl
= btrfs_ioctl
,
8889 #ifdef CONFIG_COMPAT
8890 .compat_ioctl
= btrfs_ioctl
,
8892 .release
= btrfs_release_file
,
8893 .fsync
= btrfs_sync_file
,
8896 static struct extent_io_ops btrfs_extent_io_ops
= {
8897 .fill_delalloc
= run_delalloc_range
,
8898 .submit_bio_hook
= btrfs_submit_bio_hook
,
8899 .merge_bio_hook
= btrfs_merge_bio_hook
,
8900 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
8901 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
8902 .writepage_start_hook
= btrfs_writepage_start_hook
,
8903 .set_bit_hook
= btrfs_set_bit_hook
,
8904 .clear_bit_hook
= btrfs_clear_bit_hook
,
8905 .merge_extent_hook
= btrfs_merge_extent_hook
,
8906 .split_extent_hook
= btrfs_split_extent_hook
,
8910 * btrfs doesn't support the bmap operation because swapfiles
8911 * use bmap to make a mapping of extents in the file. They assume
8912 * these extents won't change over the life of the file and they
8913 * use the bmap result to do IO directly to the drive.
8915 * the btrfs bmap call would return logical addresses that aren't
8916 * suitable for IO and they also will change frequently as COW
8917 * operations happen. So, swapfile + btrfs == corruption.
8919 * For now we're avoiding this by dropping bmap.
8921 static const struct address_space_operations btrfs_aops
= {
8922 .readpage
= btrfs_readpage
,
8923 .writepage
= btrfs_writepage
,
8924 .writepages
= btrfs_writepages
,
8925 .readpages
= btrfs_readpages
,
8926 .direct_IO
= btrfs_direct_IO
,
8927 .invalidatepage
= btrfs_invalidatepage
,
8928 .releasepage
= btrfs_releasepage
,
8929 .set_page_dirty
= btrfs_set_page_dirty
,
8930 .error_remove_page
= generic_error_remove_page
,
8933 static const struct address_space_operations btrfs_symlink_aops
= {
8934 .readpage
= btrfs_readpage
,
8935 .writepage
= btrfs_writepage
,
8936 .invalidatepage
= btrfs_invalidatepage
,
8937 .releasepage
= btrfs_releasepage
,
8940 static const struct inode_operations btrfs_file_inode_operations
= {
8941 .getattr
= btrfs_getattr
,
8942 .setattr
= btrfs_setattr
,
8943 .setxattr
= btrfs_setxattr
,
8944 .getxattr
= btrfs_getxattr
,
8945 .listxattr
= btrfs_listxattr
,
8946 .removexattr
= btrfs_removexattr
,
8947 .permission
= btrfs_permission
,
8948 .fiemap
= btrfs_fiemap
,
8949 .get_acl
= btrfs_get_acl
,
8950 .update_time
= btrfs_update_time
,
8952 static const struct inode_operations btrfs_special_inode_operations
= {
8953 .getattr
= btrfs_getattr
,
8954 .setattr
= btrfs_setattr
,
8955 .permission
= btrfs_permission
,
8956 .setxattr
= btrfs_setxattr
,
8957 .getxattr
= btrfs_getxattr
,
8958 .listxattr
= btrfs_listxattr
,
8959 .removexattr
= btrfs_removexattr
,
8960 .get_acl
= btrfs_get_acl
,
8961 .update_time
= btrfs_update_time
,
8963 static const struct inode_operations btrfs_symlink_inode_operations
= {
8964 .readlink
= generic_readlink
,
8965 .follow_link
= page_follow_link_light
,
8966 .put_link
= page_put_link
,
8967 .getattr
= btrfs_getattr
,
8968 .setattr
= btrfs_setattr
,
8969 .permission
= btrfs_permission
,
8970 .setxattr
= btrfs_setxattr
,
8971 .getxattr
= btrfs_getxattr
,
8972 .listxattr
= btrfs_listxattr
,
8973 .removexattr
= btrfs_removexattr
,
8974 .get_acl
= btrfs_get_acl
,
8975 .update_time
= btrfs_update_time
,
8978 const struct dentry_operations btrfs_dentry_operations
= {
8979 .d_delete
= btrfs_dentry_delete
,
8980 .d_release
= btrfs_dentry_release
,