2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
54 #include "compression.h"
56 #include "free-space-cache.h"
57 #include "inode-map.h"
60 struct btrfs_iget_args
{
62 struct btrfs_root
*root
;
65 static const struct inode_operations btrfs_dir_inode_operations
;
66 static const struct inode_operations btrfs_symlink_inode_operations
;
67 static const struct inode_operations btrfs_dir_ro_inode_operations
;
68 static const struct inode_operations btrfs_special_inode_operations
;
69 static const struct inode_operations btrfs_file_inode_operations
;
70 static const struct address_space_operations btrfs_aops
;
71 static const struct address_space_operations btrfs_symlink_aops
;
72 static const struct file_operations btrfs_dir_file_operations
;
73 static struct extent_io_ops btrfs_extent_io_ops
;
75 static struct kmem_cache
*btrfs_inode_cachep
;
76 static struct kmem_cache
*btrfs_delalloc_work_cachep
;
77 struct kmem_cache
*btrfs_trans_handle_cachep
;
78 struct kmem_cache
*btrfs_transaction_cachep
;
79 struct kmem_cache
*btrfs_path_cachep
;
80 struct kmem_cache
*btrfs_free_space_cachep
;
83 static unsigned char btrfs_type_by_mode
[S_IFMT
>> S_SHIFT
] = {
84 [S_IFREG
>> S_SHIFT
] = BTRFS_FT_REG_FILE
,
85 [S_IFDIR
>> S_SHIFT
] = BTRFS_FT_DIR
,
86 [S_IFCHR
>> S_SHIFT
] = BTRFS_FT_CHRDEV
,
87 [S_IFBLK
>> S_SHIFT
] = BTRFS_FT_BLKDEV
,
88 [S_IFIFO
>> S_SHIFT
] = BTRFS_FT_FIFO
,
89 [S_IFSOCK
>> S_SHIFT
] = BTRFS_FT_SOCK
,
90 [S_IFLNK
>> S_SHIFT
] = BTRFS_FT_SYMLINK
,
93 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
);
94 static int btrfs_truncate(struct inode
*inode
);
95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
);
96 static noinline
int cow_file_range(struct inode
*inode
,
97 struct page
*locked_page
,
98 u64 start
, u64 end
, int *page_started
,
99 unsigned long *nr_written
, int unlock
);
100 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
101 u64 len
, u64 orig_start
,
102 u64 block_start
, u64 block_len
,
103 u64 orig_block_len
, u64 ram_bytes
,
106 static int btrfs_dirty_inode(struct inode
*inode
);
108 static int btrfs_init_inode_security(struct btrfs_trans_handle
*trans
,
109 struct inode
*inode
, struct inode
*dir
,
110 const struct qstr
*qstr
)
114 err
= btrfs_init_acl(trans
, inode
, dir
);
116 err
= btrfs_xattr_security_init(trans
, inode
, dir
, qstr
);
121 * this does all the hard work for inserting an inline extent into
122 * the btree. The caller should have done a btrfs_drop_extents so that
123 * no overlapping inline items exist in the btree
125 static noinline
int insert_inline_extent(struct btrfs_trans_handle
*trans
,
126 struct btrfs_root
*root
, struct inode
*inode
,
127 u64 start
, size_t size
, size_t compressed_size
,
129 struct page
**compressed_pages
)
131 struct btrfs_key key
;
132 struct btrfs_path
*path
;
133 struct extent_buffer
*leaf
;
134 struct page
*page
= NULL
;
137 struct btrfs_file_extent_item
*ei
;
140 size_t cur_size
= size
;
142 unsigned long offset
;
144 if (compressed_size
&& compressed_pages
)
145 cur_size
= compressed_size
;
147 path
= btrfs_alloc_path();
151 path
->leave_spinning
= 1;
153 key
.objectid
= btrfs_ino(inode
);
155 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
156 datasize
= btrfs_file_extent_calc_inline_size(cur_size
);
158 inode_add_bytes(inode
, size
);
159 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
165 leaf
= path
->nodes
[0];
166 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
167 struct btrfs_file_extent_item
);
168 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
169 btrfs_set_file_extent_type(leaf
, ei
, BTRFS_FILE_EXTENT_INLINE
);
170 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
171 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
172 btrfs_set_file_extent_ram_bytes(leaf
, ei
, size
);
173 ptr
= btrfs_file_extent_inline_start(ei
);
175 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
178 while (compressed_size
> 0) {
179 cpage
= compressed_pages
[i
];
180 cur_size
= min_t(unsigned long, compressed_size
,
183 kaddr
= kmap_atomic(cpage
);
184 write_extent_buffer(leaf
, kaddr
, ptr
, cur_size
);
185 kunmap_atomic(kaddr
);
189 compressed_size
-= cur_size
;
191 btrfs_set_file_extent_compression(leaf
, ei
,
194 page
= find_get_page(inode
->i_mapping
,
195 start
>> PAGE_CACHE_SHIFT
);
196 btrfs_set_file_extent_compression(leaf
, ei
, 0);
197 kaddr
= kmap_atomic(page
);
198 offset
= start
& (PAGE_CACHE_SIZE
- 1);
199 write_extent_buffer(leaf
, kaddr
+ offset
, ptr
, size
);
200 kunmap_atomic(kaddr
);
201 page_cache_release(page
);
203 btrfs_mark_buffer_dirty(leaf
);
204 btrfs_free_path(path
);
207 * we're an inline extent, so nobody can
208 * extend the file past i_size without locking
209 * a page we already have locked.
211 * We must do any isize and inode updates
212 * before we unlock the pages. Otherwise we
213 * could end up racing with unlink.
215 BTRFS_I(inode
)->disk_i_size
= inode
->i_size
;
216 ret
= btrfs_update_inode(trans
, root
, inode
);
220 btrfs_free_path(path
);
226 * conditionally insert an inline extent into the file. This
227 * does the checks required to make sure the data is small enough
228 * to fit as an inline extent.
230 static noinline
int cow_file_range_inline(struct btrfs_trans_handle
*trans
,
231 struct btrfs_root
*root
,
232 struct inode
*inode
, u64 start
, u64 end
,
233 size_t compressed_size
, int compress_type
,
234 struct page
**compressed_pages
)
236 u64 isize
= i_size_read(inode
);
237 u64 actual_end
= min(end
+ 1, isize
);
238 u64 inline_len
= actual_end
- start
;
239 u64 aligned_end
= ALIGN(end
, root
->sectorsize
);
240 u64 data_len
= inline_len
;
244 data_len
= compressed_size
;
247 actual_end
>= PAGE_CACHE_SIZE
||
248 data_len
>= BTRFS_MAX_INLINE_DATA_SIZE(root
) ||
250 (actual_end
& (root
->sectorsize
- 1)) == 0) ||
252 data_len
> root
->fs_info
->max_inline
) {
256 ret
= btrfs_drop_extents(trans
, root
, inode
, start
, aligned_end
, 1);
260 if (isize
> actual_end
)
261 inline_len
= min_t(u64
, isize
, actual_end
);
262 ret
= insert_inline_extent(trans
, root
, inode
, start
,
263 inline_len
, compressed_size
,
264 compress_type
, compressed_pages
);
265 if (ret
&& ret
!= -ENOSPC
) {
266 btrfs_abort_transaction(trans
, root
, ret
);
268 } else if (ret
== -ENOSPC
) {
272 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
273 btrfs_delalloc_release_metadata(inode
, end
+ 1 - start
);
274 btrfs_drop_extent_cache(inode
, start
, aligned_end
- 1, 0);
278 struct async_extent
{
283 unsigned long nr_pages
;
285 struct list_head list
;
290 struct btrfs_root
*root
;
291 struct page
*locked_page
;
294 struct list_head extents
;
295 struct btrfs_work work
;
298 static noinline
int add_async_extent(struct async_cow
*cow
,
299 u64 start
, u64 ram_size
,
302 unsigned long nr_pages
,
305 struct async_extent
*async_extent
;
307 async_extent
= kmalloc(sizeof(*async_extent
), GFP_NOFS
);
308 BUG_ON(!async_extent
); /* -ENOMEM */
309 async_extent
->start
= start
;
310 async_extent
->ram_size
= ram_size
;
311 async_extent
->compressed_size
= compressed_size
;
312 async_extent
->pages
= pages
;
313 async_extent
->nr_pages
= nr_pages
;
314 async_extent
->compress_type
= compress_type
;
315 list_add_tail(&async_extent
->list
, &cow
->extents
);
320 * we create compressed extents in two phases. The first
321 * phase compresses a range of pages that have already been
322 * locked (both pages and state bits are locked).
324 * This is done inside an ordered work queue, and the compression
325 * is spread across many cpus. The actual IO submission is step
326 * two, and the ordered work queue takes care of making sure that
327 * happens in the same order things were put onto the queue by
328 * writepages and friends.
330 * If this code finds it can't get good compression, it puts an
331 * entry onto the work queue to write the uncompressed bytes. This
332 * makes sure that both compressed inodes and uncompressed inodes
333 * are written in the same order that the flusher thread sent them
336 static noinline
int compress_file_range(struct inode
*inode
,
337 struct page
*locked_page
,
339 struct async_cow
*async_cow
,
342 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
343 struct btrfs_trans_handle
*trans
;
345 u64 blocksize
= root
->sectorsize
;
347 u64 isize
= i_size_read(inode
);
349 struct page
**pages
= NULL
;
350 unsigned long nr_pages
;
351 unsigned long nr_pages_ret
= 0;
352 unsigned long total_compressed
= 0;
353 unsigned long total_in
= 0;
354 unsigned long max_compressed
= 128 * 1024;
355 unsigned long max_uncompressed
= 128 * 1024;
358 int compress_type
= root
->fs_info
->compress_type
;
361 /* if this is a small write inside eof, kick off a defrag */
362 if ((end
- start
+ 1) < 16 * 1024 &&
363 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
364 btrfs_add_inode_defrag(NULL
, inode
);
366 actual_end
= min_t(u64
, isize
, end
+ 1);
369 nr_pages
= (end
>> PAGE_CACHE_SHIFT
) - (start
>> PAGE_CACHE_SHIFT
) + 1;
370 nr_pages
= min(nr_pages
, (128 * 1024UL) / PAGE_CACHE_SIZE
);
373 * we don't want to send crud past the end of i_size through
374 * compression, that's just a waste of CPU time. So, if the
375 * end of the file is before the start of our current
376 * requested range of bytes, we bail out to the uncompressed
377 * cleanup code that can deal with all of this.
379 * It isn't really the fastest way to fix things, but this is a
380 * very uncommon corner.
382 if (actual_end
<= start
)
383 goto cleanup_and_bail_uncompressed
;
385 total_compressed
= actual_end
- start
;
387 /* we want to make sure that amount of ram required to uncompress
388 * an extent is reasonable, so we limit the total size in ram
389 * of a compressed extent to 128k. This is a crucial number
390 * because it also controls how easily we can spread reads across
391 * cpus for decompression.
393 * We also want to make sure the amount of IO required to do
394 * a random read is reasonably small, so we limit the size of
395 * a compressed extent to 128k.
397 total_compressed
= min(total_compressed
, max_uncompressed
);
398 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
399 num_bytes
= max(blocksize
, num_bytes
);
404 * we do compression for mount -o compress and when the
405 * inode has not been flagged as nocompress. This flag can
406 * change at any time if we discover bad compression ratios.
408 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
) &&
409 (btrfs_test_opt(root
, COMPRESS
) ||
410 (BTRFS_I(inode
)->force_compress
) ||
411 (BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
))) {
413 pages
= kzalloc(sizeof(struct page
*) * nr_pages
, GFP_NOFS
);
415 /* just bail out to the uncompressed code */
419 if (BTRFS_I(inode
)->force_compress
)
420 compress_type
= BTRFS_I(inode
)->force_compress
;
423 * we need to call clear_page_dirty_for_io on each
424 * page in the range. Otherwise applications with the file
425 * mmap'd can wander in and change the page contents while
426 * we are compressing them.
428 * If the compression fails for any reason, we set the pages
429 * dirty again later on.
431 extent_range_clear_dirty_for_io(inode
, start
, end
);
433 ret
= btrfs_compress_pages(compress_type
,
434 inode
->i_mapping
, start
,
435 total_compressed
, pages
,
436 nr_pages
, &nr_pages_ret
,
442 unsigned long offset
= total_compressed
&
443 (PAGE_CACHE_SIZE
- 1);
444 struct page
*page
= pages
[nr_pages_ret
- 1];
447 /* zero the tail end of the last page, we might be
448 * sending it down to disk
451 kaddr
= kmap_atomic(page
);
452 memset(kaddr
+ offset
, 0,
453 PAGE_CACHE_SIZE
- offset
);
454 kunmap_atomic(kaddr
);
461 trans
= btrfs_join_transaction(root
);
463 ret
= PTR_ERR(trans
);
465 goto cleanup_and_out
;
467 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
469 /* lets try to make an inline extent */
470 if (ret
|| total_in
< (actual_end
- start
)) {
471 /* we didn't compress the entire range, try
472 * to make an uncompressed inline extent.
474 ret
= cow_file_range_inline(trans
, root
, inode
,
475 start
, end
, 0, 0, NULL
);
477 /* try making a compressed inline extent */
478 ret
= cow_file_range_inline(trans
, root
, inode
,
481 compress_type
, pages
);
485 * inline extent creation worked or returned error,
486 * we don't need to create any more async work items.
487 * Unlock and free up our temp pages.
489 extent_clear_unlock_delalloc(inode
,
490 &BTRFS_I(inode
)->io_tree
,
492 EXTENT_CLEAR_UNLOCK_PAGE
| EXTENT_CLEAR_DIRTY
|
493 EXTENT_CLEAR_DELALLOC
|
494 EXTENT_SET_WRITEBACK
| EXTENT_END_WRITEBACK
);
496 btrfs_end_transaction(trans
, root
);
499 btrfs_end_transaction(trans
, root
);
504 * we aren't doing an inline extent round the compressed size
505 * up to a block size boundary so the allocator does sane
508 total_compressed
= ALIGN(total_compressed
, blocksize
);
511 * one last check to make sure the compression is really a
512 * win, compare the page count read with the blocks on disk
514 total_in
= ALIGN(total_in
, PAGE_CACHE_SIZE
);
515 if (total_compressed
>= total_in
) {
518 num_bytes
= total_in
;
521 if (!will_compress
&& pages
) {
523 * the compression code ran but failed to make things smaller,
524 * free any pages it allocated and our page pointer array
526 for (i
= 0; i
< nr_pages_ret
; i
++) {
527 WARN_ON(pages
[i
]->mapping
);
528 page_cache_release(pages
[i
]);
532 total_compressed
= 0;
535 /* flag the file so we don't compress in the future */
536 if (!btrfs_test_opt(root
, FORCE_COMPRESS
) &&
537 !(BTRFS_I(inode
)->force_compress
)) {
538 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NOCOMPRESS
;
544 /* the async work queues will take care of doing actual
545 * allocation on disk for these compressed pages,
546 * and will submit them to the elevator.
548 add_async_extent(async_cow
, start
, num_bytes
,
549 total_compressed
, pages
, nr_pages_ret
,
552 if (start
+ num_bytes
< end
) {
559 cleanup_and_bail_uncompressed
:
561 * No compression, but we still need to write the pages in
562 * the file we've been given so far. redirty the locked
563 * page if it corresponds to our extent and set things up
564 * for the async work queue to run cow_file_range to do
565 * the normal delalloc dance
567 if (page_offset(locked_page
) >= start
&&
568 page_offset(locked_page
) <= end
) {
569 __set_page_dirty_nobuffers(locked_page
);
570 /* unlocked later on in the async handlers */
573 extent_range_redirty_for_io(inode
, start
, end
);
574 add_async_extent(async_cow
, start
, end
- start
+ 1,
575 0, NULL
, 0, BTRFS_COMPRESS_NONE
);
583 for (i
= 0; i
< nr_pages_ret
; i
++) {
584 WARN_ON(pages
[i
]->mapping
);
585 page_cache_release(pages
[i
]);
592 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
594 EXTENT_CLEAR_UNLOCK_PAGE
|
596 EXTENT_CLEAR_DELALLOC
|
597 EXTENT_SET_WRITEBACK
|
598 EXTENT_END_WRITEBACK
);
599 if (!trans
|| IS_ERR(trans
))
600 btrfs_error(root
->fs_info
, ret
, "Failed to join transaction");
602 btrfs_abort_transaction(trans
, root
, ret
);
607 * phase two of compressed writeback. This is the ordered portion
608 * of the code, which only gets called in the order the work was
609 * queued. We walk all the async extents created by compress_file_range
610 * and send them down to the disk.
612 static noinline
int submit_compressed_extents(struct inode
*inode
,
613 struct async_cow
*async_cow
)
615 struct async_extent
*async_extent
;
617 struct btrfs_trans_handle
*trans
;
618 struct btrfs_key ins
;
619 struct extent_map
*em
;
620 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
621 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
622 struct extent_io_tree
*io_tree
;
625 if (list_empty(&async_cow
->extents
))
629 while (!list_empty(&async_cow
->extents
)) {
630 async_extent
= list_entry(async_cow
->extents
.next
,
631 struct async_extent
, list
);
632 list_del(&async_extent
->list
);
634 io_tree
= &BTRFS_I(inode
)->io_tree
;
637 /* did the compression code fall back to uncompressed IO? */
638 if (!async_extent
->pages
) {
639 int page_started
= 0;
640 unsigned long nr_written
= 0;
642 lock_extent(io_tree
, async_extent
->start
,
643 async_extent
->start
+
644 async_extent
->ram_size
- 1);
646 /* allocate blocks */
647 ret
= cow_file_range(inode
, async_cow
->locked_page
,
649 async_extent
->start
+
650 async_extent
->ram_size
- 1,
651 &page_started
, &nr_written
, 0);
656 * if page_started, cow_file_range inserted an
657 * inline extent and took care of all the unlocking
658 * and IO for us. Otherwise, we need to submit
659 * all those pages down to the drive.
661 if (!page_started
&& !ret
)
662 extent_write_locked_range(io_tree
,
663 inode
, async_extent
->start
,
664 async_extent
->start
+
665 async_extent
->ram_size
- 1,
669 unlock_page(async_cow
->locked_page
);
675 lock_extent(io_tree
, async_extent
->start
,
676 async_extent
->start
+ async_extent
->ram_size
- 1);
678 trans
= btrfs_join_transaction(root
);
680 ret
= PTR_ERR(trans
);
682 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
683 ret
= btrfs_reserve_extent(trans
, root
,
684 async_extent
->compressed_size
,
685 async_extent
->compressed_size
,
686 0, alloc_hint
, &ins
, 1);
687 if (ret
&& ret
!= -ENOSPC
)
688 btrfs_abort_transaction(trans
, root
, ret
);
689 btrfs_end_transaction(trans
, root
);
695 for (i
= 0; i
< async_extent
->nr_pages
; i
++) {
696 WARN_ON(async_extent
->pages
[i
]->mapping
);
697 page_cache_release(async_extent
->pages
[i
]);
699 kfree(async_extent
->pages
);
700 async_extent
->nr_pages
= 0;
701 async_extent
->pages
= NULL
;
709 * here we're doing allocation and writeback of the
712 btrfs_drop_extent_cache(inode
, async_extent
->start
,
713 async_extent
->start
+
714 async_extent
->ram_size
- 1, 0);
716 em
= alloc_extent_map();
719 goto out_free_reserve
;
721 em
->start
= async_extent
->start
;
722 em
->len
= async_extent
->ram_size
;
723 em
->orig_start
= em
->start
;
724 em
->mod_start
= em
->start
;
725 em
->mod_len
= em
->len
;
727 em
->block_start
= ins
.objectid
;
728 em
->block_len
= ins
.offset
;
729 em
->orig_block_len
= ins
.offset
;
730 em
->ram_bytes
= async_extent
->ram_size
;
731 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
732 em
->compress_type
= async_extent
->compress_type
;
733 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
734 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
738 write_lock(&em_tree
->lock
);
739 ret
= add_extent_mapping(em_tree
, em
, 1);
740 write_unlock(&em_tree
->lock
);
741 if (ret
!= -EEXIST
) {
745 btrfs_drop_extent_cache(inode
, async_extent
->start
,
746 async_extent
->start
+
747 async_extent
->ram_size
- 1, 0);
751 goto out_free_reserve
;
753 ret
= btrfs_add_ordered_extent_compress(inode
,
756 async_extent
->ram_size
,
758 BTRFS_ORDERED_COMPRESSED
,
759 async_extent
->compress_type
);
761 goto out_free_reserve
;
764 * clear dirty, set writeback and unlock the pages.
766 extent_clear_unlock_delalloc(inode
,
767 &BTRFS_I(inode
)->io_tree
,
769 async_extent
->start
+
770 async_extent
->ram_size
- 1,
771 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
772 EXTENT_CLEAR_UNLOCK
|
773 EXTENT_CLEAR_DELALLOC
|
774 EXTENT_CLEAR_DIRTY
| EXTENT_SET_WRITEBACK
);
776 ret
= btrfs_submit_compressed_write(inode
,
778 async_extent
->ram_size
,
780 ins
.offset
, async_extent
->pages
,
781 async_extent
->nr_pages
);
782 alloc_hint
= ins
.objectid
+ ins
.offset
;
792 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
794 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
796 async_extent
->start
+
797 async_extent
->ram_size
- 1,
798 NULL
, EXTENT_CLEAR_UNLOCK_PAGE
|
799 EXTENT_CLEAR_UNLOCK
|
800 EXTENT_CLEAR_DELALLOC
|
802 EXTENT_SET_WRITEBACK
|
803 EXTENT_END_WRITEBACK
);
808 static u64
get_extent_allocation_hint(struct inode
*inode
, u64 start
,
811 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
812 struct extent_map
*em
;
815 read_lock(&em_tree
->lock
);
816 em
= search_extent_mapping(em_tree
, start
, num_bytes
);
819 * if block start isn't an actual block number then find the
820 * first block in this inode and use that as a hint. If that
821 * block is also bogus then just don't worry about it.
823 if (em
->block_start
>= EXTENT_MAP_LAST_BYTE
) {
825 em
= search_extent_mapping(em_tree
, 0, 0);
826 if (em
&& em
->block_start
< EXTENT_MAP_LAST_BYTE
)
827 alloc_hint
= em
->block_start
;
831 alloc_hint
= em
->block_start
;
835 read_unlock(&em_tree
->lock
);
841 * when extent_io.c finds a delayed allocation range in the file,
842 * the call backs end up in this code. The basic idea is to
843 * allocate extents on disk for the range, and create ordered data structs
844 * in ram to track those extents.
846 * locked_page is the page that writepage had locked already. We use
847 * it to make sure we don't do extra locks or unlocks.
849 * *page_started is set to one if we unlock locked_page and do everything
850 * required to start IO on it. It may be clean and already done with
853 static noinline
int __cow_file_range(struct btrfs_trans_handle
*trans
,
855 struct btrfs_root
*root
,
856 struct page
*locked_page
,
857 u64 start
, u64 end
, int *page_started
,
858 unsigned long *nr_written
,
863 unsigned long ram_size
;
866 u64 blocksize
= root
->sectorsize
;
867 struct btrfs_key ins
;
868 struct extent_map
*em
;
869 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
872 BUG_ON(btrfs_is_free_space_inode(inode
));
874 num_bytes
= ALIGN(end
- start
+ 1, blocksize
);
875 num_bytes
= max(blocksize
, num_bytes
);
876 disk_num_bytes
= num_bytes
;
878 /* if this is a small write inside eof, kick off defrag */
879 if (num_bytes
< 64 * 1024 &&
880 (start
> 0 || end
+ 1 < BTRFS_I(inode
)->disk_i_size
))
881 btrfs_add_inode_defrag(trans
, inode
);
884 /* lets try to make an inline extent */
885 ret
= cow_file_range_inline(trans
, root
, inode
,
886 start
, end
, 0, 0, NULL
);
888 extent_clear_unlock_delalloc(inode
,
889 &BTRFS_I(inode
)->io_tree
,
891 EXTENT_CLEAR_UNLOCK_PAGE
|
892 EXTENT_CLEAR_UNLOCK
|
893 EXTENT_CLEAR_DELALLOC
|
895 EXTENT_SET_WRITEBACK
|
896 EXTENT_END_WRITEBACK
);
898 *nr_written
= *nr_written
+
899 (end
- start
+ PAGE_CACHE_SIZE
) / PAGE_CACHE_SIZE
;
902 } else if (ret
< 0) {
903 btrfs_abort_transaction(trans
, root
, ret
);
908 BUG_ON(disk_num_bytes
>
909 btrfs_super_total_bytes(root
->fs_info
->super_copy
));
911 alloc_hint
= get_extent_allocation_hint(inode
, start
, num_bytes
);
912 btrfs_drop_extent_cache(inode
, start
, start
+ num_bytes
- 1, 0);
914 while (disk_num_bytes
> 0) {
917 cur_alloc_size
= disk_num_bytes
;
918 ret
= btrfs_reserve_extent(trans
, root
, cur_alloc_size
,
919 root
->sectorsize
, 0, alloc_hint
,
922 btrfs_abort_transaction(trans
, root
, ret
);
926 em
= alloc_extent_map();
932 em
->orig_start
= em
->start
;
933 ram_size
= ins
.offset
;
934 em
->len
= ins
.offset
;
935 em
->mod_start
= em
->start
;
936 em
->mod_len
= em
->len
;
938 em
->block_start
= ins
.objectid
;
939 em
->block_len
= ins
.offset
;
940 em
->orig_block_len
= ins
.offset
;
941 em
->ram_bytes
= ram_size
;
942 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
943 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
947 write_lock(&em_tree
->lock
);
948 ret
= add_extent_mapping(em_tree
, em
, 1);
949 write_unlock(&em_tree
->lock
);
950 if (ret
!= -EEXIST
) {
954 btrfs_drop_extent_cache(inode
, start
,
955 start
+ ram_size
- 1, 0);
960 cur_alloc_size
= ins
.offset
;
961 ret
= btrfs_add_ordered_extent(inode
, start
, ins
.objectid
,
962 ram_size
, cur_alloc_size
, 0);
966 if (root
->root_key
.objectid
==
967 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
968 ret
= btrfs_reloc_clone_csums(inode
, start
,
971 btrfs_abort_transaction(trans
, root
, ret
);
976 if (disk_num_bytes
< cur_alloc_size
)
979 /* we're not doing compressed IO, don't unlock the first
980 * page (which the caller expects to stay locked), don't
981 * clear any dirty bits and don't set any writeback bits
983 * Do set the Private2 bit so we know this page was properly
984 * setup for writepage
986 op
= unlock
? EXTENT_CLEAR_UNLOCK_PAGE
: 0;
987 op
|= EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
990 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
991 start
, start
+ ram_size
- 1,
993 disk_num_bytes
-= cur_alloc_size
;
994 num_bytes
-= cur_alloc_size
;
995 alloc_hint
= ins
.objectid
+ ins
.offset
;
996 start
+= cur_alloc_size
;
1002 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
1004 extent_clear_unlock_delalloc(inode
,
1005 &BTRFS_I(inode
)->io_tree
,
1006 start
, end
, locked_page
,
1007 EXTENT_CLEAR_UNLOCK_PAGE
|
1008 EXTENT_CLEAR_UNLOCK
|
1009 EXTENT_CLEAR_DELALLOC
|
1010 EXTENT_CLEAR_DIRTY
|
1011 EXTENT_SET_WRITEBACK
|
1012 EXTENT_END_WRITEBACK
);
1017 static noinline
int cow_file_range(struct inode
*inode
,
1018 struct page
*locked_page
,
1019 u64 start
, u64 end
, int *page_started
,
1020 unsigned long *nr_written
,
1023 struct btrfs_trans_handle
*trans
;
1024 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1027 trans
= btrfs_join_transaction(root
);
1028 if (IS_ERR(trans
)) {
1029 extent_clear_unlock_delalloc(inode
,
1030 &BTRFS_I(inode
)->io_tree
,
1031 start
, end
, locked_page
,
1032 EXTENT_CLEAR_UNLOCK_PAGE
|
1033 EXTENT_CLEAR_UNLOCK
|
1034 EXTENT_CLEAR_DELALLOC
|
1035 EXTENT_CLEAR_DIRTY
|
1036 EXTENT_SET_WRITEBACK
|
1037 EXTENT_END_WRITEBACK
);
1038 return PTR_ERR(trans
);
1040 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1042 ret
= __cow_file_range(trans
, inode
, root
, locked_page
, start
, end
,
1043 page_started
, nr_written
, unlock
);
1045 btrfs_end_transaction(trans
, root
);
1051 * work queue call back to started compression on a file and pages
1053 static noinline
void async_cow_start(struct btrfs_work
*work
)
1055 struct async_cow
*async_cow
;
1057 async_cow
= container_of(work
, struct async_cow
, work
);
1059 compress_file_range(async_cow
->inode
, async_cow
->locked_page
,
1060 async_cow
->start
, async_cow
->end
, async_cow
,
1062 if (num_added
== 0) {
1063 btrfs_add_delayed_iput(async_cow
->inode
);
1064 async_cow
->inode
= NULL
;
1069 * work queue call back to submit previously compressed pages
1071 static noinline
void async_cow_submit(struct btrfs_work
*work
)
1073 struct async_cow
*async_cow
;
1074 struct btrfs_root
*root
;
1075 unsigned long nr_pages
;
1077 async_cow
= container_of(work
, struct async_cow
, work
);
1079 root
= async_cow
->root
;
1080 nr_pages
= (async_cow
->end
- async_cow
->start
+ PAGE_CACHE_SIZE
) >>
1083 if (atomic_sub_return(nr_pages
, &root
->fs_info
->async_delalloc_pages
) <
1085 waitqueue_active(&root
->fs_info
->async_submit_wait
))
1086 wake_up(&root
->fs_info
->async_submit_wait
);
1088 if (async_cow
->inode
)
1089 submit_compressed_extents(async_cow
->inode
, async_cow
);
1092 static noinline
void async_cow_free(struct btrfs_work
*work
)
1094 struct async_cow
*async_cow
;
1095 async_cow
= container_of(work
, struct async_cow
, work
);
1096 if (async_cow
->inode
)
1097 btrfs_add_delayed_iput(async_cow
->inode
);
1101 static int cow_file_range_async(struct inode
*inode
, struct page
*locked_page
,
1102 u64 start
, u64 end
, int *page_started
,
1103 unsigned long *nr_written
)
1105 struct async_cow
*async_cow
;
1106 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1107 unsigned long nr_pages
;
1109 int limit
= 10 * 1024 * 1024;
1111 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, start
, end
, EXTENT_LOCKED
,
1112 1, 0, NULL
, GFP_NOFS
);
1113 while (start
< end
) {
1114 async_cow
= kmalloc(sizeof(*async_cow
), GFP_NOFS
);
1115 BUG_ON(!async_cow
); /* -ENOMEM */
1116 async_cow
->inode
= igrab(inode
);
1117 async_cow
->root
= root
;
1118 async_cow
->locked_page
= locked_page
;
1119 async_cow
->start
= start
;
1121 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NOCOMPRESS
)
1124 cur_end
= min(end
, start
+ 512 * 1024 - 1);
1126 async_cow
->end
= cur_end
;
1127 INIT_LIST_HEAD(&async_cow
->extents
);
1129 async_cow
->work
.func
= async_cow_start
;
1130 async_cow
->work
.ordered_func
= async_cow_submit
;
1131 async_cow
->work
.ordered_free
= async_cow_free
;
1132 async_cow
->work
.flags
= 0;
1134 nr_pages
= (cur_end
- start
+ PAGE_CACHE_SIZE
) >>
1136 atomic_add(nr_pages
, &root
->fs_info
->async_delalloc_pages
);
1138 btrfs_queue_worker(&root
->fs_info
->delalloc_workers
,
1141 if (atomic_read(&root
->fs_info
->async_delalloc_pages
) > limit
) {
1142 wait_event(root
->fs_info
->async_submit_wait
,
1143 (atomic_read(&root
->fs_info
->async_delalloc_pages
) <
1147 while (atomic_read(&root
->fs_info
->async_submit_draining
) &&
1148 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
1149 wait_event(root
->fs_info
->async_submit_wait
,
1150 (atomic_read(&root
->fs_info
->async_delalloc_pages
) ==
1154 *nr_written
+= nr_pages
;
1155 start
= cur_end
+ 1;
1161 static noinline
int csum_exist_in_range(struct btrfs_root
*root
,
1162 u64 bytenr
, u64 num_bytes
)
1165 struct btrfs_ordered_sum
*sums
;
1168 ret
= btrfs_lookup_csums_range(root
->fs_info
->csum_root
, bytenr
,
1169 bytenr
+ num_bytes
- 1, &list
, 0);
1170 if (ret
== 0 && list_empty(&list
))
1173 while (!list_empty(&list
)) {
1174 sums
= list_entry(list
.next
, struct btrfs_ordered_sum
, list
);
1175 list_del(&sums
->list
);
1182 * when nowcow writeback call back. This checks for snapshots or COW copies
1183 * of the extents that exist in the file, and COWs the file as required.
1185 * If no cow copies or snapshots exist, we write directly to the existing
1188 static noinline
int run_delalloc_nocow(struct inode
*inode
,
1189 struct page
*locked_page
,
1190 u64 start
, u64 end
, int *page_started
, int force
,
1191 unsigned long *nr_written
)
1193 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1194 struct btrfs_trans_handle
*trans
;
1195 struct extent_buffer
*leaf
;
1196 struct btrfs_path
*path
;
1197 struct btrfs_file_extent_item
*fi
;
1198 struct btrfs_key found_key
;
1213 u64 ino
= btrfs_ino(inode
);
1215 path
= btrfs_alloc_path();
1217 extent_clear_unlock_delalloc(inode
,
1218 &BTRFS_I(inode
)->io_tree
,
1219 start
, end
, locked_page
,
1220 EXTENT_CLEAR_UNLOCK_PAGE
|
1221 EXTENT_CLEAR_UNLOCK
|
1222 EXTENT_CLEAR_DELALLOC
|
1223 EXTENT_CLEAR_DIRTY
|
1224 EXTENT_SET_WRITEBACK
|
1225 EXTENT_END_WRITEBACK
);
1229 nolock
= btrfs_is_free_space_inode(inode
);
1232 trans
= btrfs_join_transaction_nolock(root
);
1234 trans
= btrfs_join_transaction(root
);
1236 if (IS_ERR(trans
)) {
1237 extent_clear_unlock_delalloc(inode
,
1238 &BTRFS_I(inode
)->io_tree
,
1239 start
, end
, locked_page
,
1240 EXTENT_CLEAR_UNLOCK_PAGE
|
1241 EXTENT_CLEAR_UNLOCK
|
1242 EXTENT_CLEAR_DELALLOC
|
1243 EXTENT_CLEAR_DIRTY
|
1244 EXTENT_SET_WRITEBACK
|
1245 EXTENT_END_WRITEBACK
);
1246 btrfs_free_path(path
);
1247 return PTR_ERR(trans
);
1250 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
1252 cow_start
= (u64
)-1;
1255 ret
= btrfs_lookup_file_extent(trans
, root
, path
, ino
,
1258 btrfs_abort_transaction(trans
, root
, ret
);
1261 if (ret
> 0 && path
->slots
[0] > 0 && check_prev
) {
1262 leaf
= path
->nodes
[0];
1263 btrfs_item_key_to_cpu(leaf
, &found_key
,
1264 path
->slots
[0] - 1);
1265 if (found_key
.objectid
== ino
&&
1266 found_key
.type
== BTRFS_EXTENT_DATA_KEY
)
1271 leaf
= path
->nodes
[0];
1272 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
1273 ret
= btrfs_next_leaf(root
, path
);
1275 btrfs_abort_transaction(trans
, root
, ret
);
1280 leaf
= path
->nodes
[0];
1286 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
1288 if (found_key
.objectid
> ino
||
1289 found_key
.type
> BTRFS_EXTENT_DATA_KEY
||
1290 found_key
.offset
> end
)
1293 if (found_key
.offset
> cur_offset
) {
1294 extent_end
= found_key
.offset
;
1299 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1300 struct btrfs_file_extent_item
);
1301 extent_type
= btrfs_file_extent_type(leaf
, fi
);
1303 ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
1304 if (extent_type
== BTRFS_FILE_EXTENT_REG
||
1305 extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1306 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
1307 extent_offset
= btrfs_file_extent_offset(leaf
, fi
);
1308 extent_end
= found_key
.offset
+
1309 btrfs_file_extent_num_bytes(leaf
, fi
);
1311 btrfs_file_extent_disk_num_bytes(leaf
, fi
);
1312 if (extent_end
<= start
) {
1316 if (disk_bytenr
== 0)
1318 if (btrfs_file_extent_compression(leaf
, fi
) ||
1319 btrfs_file_extent_encryption(leaf
, fi
) ||
1320 btrfs_file_extent_other_encoding(leaf
, fi
))
1322 if (extent_type
== BTRFS_FILE_EXTENT_REG
&& !force
)
1324 if (btrfs_extent_readonly(root
, disk_bytenr
))
1326 if (btrfs_cross_ref_exist(trans
, root
, ino
,
1328 extent_offset
, disk_bytenr
))
1330 disk_bytenr
+= extent_offset
;
1331 disk_bytenr
+= cur_offset
- found_key
.offset
;
1332 num_bytes
= min(end
+ 1, extent_end
) - cur_offset
;
1334 * force cow if csum exists in the range.
1335 * this ensure that csum for a given extent are
1336 * either valid or do not exist.
1338 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
1341 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
1342 extent_end
= found_key
.offset
+
1343 btrfs_file_extent_inline_len(leaf
, fi
);
1344 extent_end
= ALIGN(extent_end
, root
->sectorsize
);
1349 if (extent_end
<= start
) {
1354 if (cow_start
== (u64
)-1)
1355 cow_start
= cur_offset
;
1356 cur_offset
= extent_end
;
1357 if (cur_offset
> end
)
1363 btrfs_release_path(path
);
1364 if (cow_start
!= (u64
)-1) {
1365 ret
= __cow_file_range(trans
, inode
, root
, locked_page
,
1366 cow_start
, found_key
.offset
- 1,
1367 page_started
, nr_written
, 1);
1369 btrfs_abort_transaction(trans
, root
, ret
);
1372 cow_start
= (u64
)-1;
1375 if (extent_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
1376 struct extent_map
*em
;
1377 struct extent_map_tree
*em_tree
;
1378 em_tree
= &BTRFS_I(inode
)->extent_tree
;
1379 em
= alloc_extent_map();
1380 BUG_ON(!em
); /* -ENOMEM */
1381 em
->start
= cur_offset
;
1382 em
->orig_start
= found_key
.offset
- extent_offset
;
1383 em
->len
= num_bytes
;
1384 em
->block_len
= num_bytes
;
1385 em
->block_start
= disk_bytenr
;
1386 em
->orig_block_len
= disk_num_bytes
;
1387 em
->ram_bytes
= ram_bytes
;
1388 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
1389 em
->mod_start
= em
->start
;
1390 em
->mod_len
= em
->len
;
1391 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
1392 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
1393 em
->generation
= -1;
1395 write_lock(&em_tree
->lock
);
1396 ret
= add_extent_mapping(em_tree
, em
, 1);
1397 write_unlock(&em_tree
->lock
);
1398 if (ret
!= -EEXIST
) {
1399 free_extent_map(em
);
1402 btrfs_drop_extent_cache(inode
, em
->start
,
1403 em
->start
+ em
->len
- 1, 0);
1405 type
= BTRFS_ORDERED_PREALLOC
;
1407 type
= BTRFS_ORDERED_NOCOW
;
1410 ret
= btrfs_add_ordered_extent(inode
, cur_offset
, disk_bytenr
,
1411 num_bytes
, num_bytes
, type
);
1412 BUG_ON(ret
); /* -ENOMEM */
1414 if (root
->root_key
.objectid
==
1415 BTRFS_DATA_RELOC_TREE_OBJECTID
) {
1416 ret
= btrfs_reloc_clone_csums(inode
, cur_offset
,
1419 btrfs_abort_transaction(trans
, root
, ret
);
1424 extent_clear_unlock_delalloc(inode
, &BTRFS_I(inode
)->io_tree
,
1425 cur_offset
, cur_offset
+ num_bytes
- 1,
1426 locked_page
, EXTENT_CLEAR_UNLOCK_PAGE
|
1427 EXTENT_CLEAR_UNLOCK
| EXTENT_CLEAR_DELALLOC
|
1428 EXTENT_SET_PRIVATE2
);
1429 cur_offset
= extent_end
;
1430 if (cur_offset
> end
)
1433 btrfs_release_path(path
);
1435 if (cur_offset
<= end
&& cow_start
== (u64
)-1) {
1436 cow_start
= cur_offset
;
1440 if (cow_start
!= (u64
)-1) {
1441 ret
= __cow_file_range(trans
, inode
, root
, locked_page
,
1443 page_started
, nr_written
, 1);
1445 btrfs_abort_transaction(trans
, root
, ret
);
1451 err
= btrfs_end_transaction(trans
, root
);
1455 if (ret
&& cur_offset
< end
)
1456 extent_clear_unlock_delalloc(inode
,
1457 &BTRFS_I(inode
)->io_tree
,
1458 cur_offset
, end
, locked_page
,
1459 EXTENT_CLEAR_UNLOCK_PAGE
|
1460 EXTENT_CLEAR_UNLOCK
|
1461 EXTENT_CLEAR_DELALLOC
|
1462 EXTENT_CLEAR_DIRTY
|
1463 EXTENT_SET_WRITEBACK
|
1464 EXTENT_END_WRITEBACK
);
1466 btrfs_free_path(path
);
1471 * extent_io.c call back to do delayed allocation processing
1473 static int run_delalloc_range(struct inode
*inode
, struct page
*locked_page
,
1474 u64 start
, u64 end
, int *page_started
,
1475 unsigned long *nr_written
)
1478 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1480 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) {
1481 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1482 page_started
, 1, nr_written
);
1483 } else if (BTRFS_I(inode
)->flags
& BTRFS_INODE_PREALLOC
) {
1484 ret
= run_delalloc_nocow(inode
, locked_page
, start
, end
,
1485 page_started
, 0, nr_written
);
1486 } else if (!btrfs_test_opt(root
, COMPRESS
) &&
1487 !(BTRFS_I(inode
)->force_compress
) &&
1488 !(BTRFS_I(inode
)->flags
& BTRFS_INODE_COMPRESS
)) {
1489 ret
= cow_file_range(inode
, locked_page
, start
, end
,
1490 page_started
, nr_written
, 1);
1492 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT
,
1493 &BTRFS_I(inode
)->runtime_flags
);
1494 ret
= cow_file_range_async(inode
, locked_page
, start
, end
,
1495 page_started
, nr_written
);
1500 static void btrfs_split_extent_hook(struct inode
*inode
,
1501 struct extent_state
*orig
, u64 split
)
1503 /* not delalloc, ignore it */
1504 if (!(orig
->state
& EXTENT_DELALLOC
))
1507 spin_lock(&BTRFS_I(inode
)->lock
);
1508 BTRFS_I(inode
)->outstanding_extents
++;
1509 spin_unlock(&BTRFS_I(inode
)->lock
);
1513 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514 * extents so we can keep track of new extents that are just merged onto old
1515 * extents, such as when we are doing sequential writes, so we can properly
1516 * account for the metadata space we'll need.
1518 static void btrfs_merge_extent_hook(struct inode
*inode
,
1519 struct extent_state
*new,
1520 struct extent_state
*other
)
1522 /* not delalloc, ignore it */
1523 if (!(other
->state
& EXTENT_DELALLOC
))
1526 spin_lock(&BTRFS_I(inode
)->lock
);
1527 BTRFS_I(inode
)->outstanding_extents
--;
1528 spin_unlock(&BTRFS_I(inode
)->lock
);
1532 * extent_io.c set_bit_hook, used to track delayed allocation
1533 * bytes in this file, and to maintain the list of inodes that
1534 * have pending delalloc work to be done.
1536 static void btrfs_set_bit_hook(struct inode
*inode
,
1537 struct extent_state
*state
, unsigned long *bits
)
1541 * set_bit and clear bit hooks normally require _irqsave/restore
1542 * but in this case, we are only testing for the DELALLOC
1543 * bit, which is only set or cleared with irqs on
1545 if (!(state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1546 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1547 u64 len
= state
->end
+ 1 - state
->start
;
1548 bool do_list
= !btrfs_is_free_space_inode(inode
);
1550 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1551 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1553 spin_lock(&BTRFS_I(inode
)->lock
);
1554 BTRFS_I(inode
)->outstanding_extents
++;
1555 spin_unlock(&BTRFS_I(inode
)->lock
);
1558 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, len
,
1559 root
->fs_info
->delalloc_batch
);
1560 spin_lock(&BTRFS_I(inode
)->lock
);
1561 BTRFS_I(inode
)->delalloc_bytes
+= len
;
1562 if (do_list
&& !test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1563 &BTRFS_I(inode
)->runtime_flags
)) {
1564 spin_lock(&root
->fs_info
->delalloc_lock
);
1565 if (list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1566 list_add_tail(&BTRFS_I(inode
)->delalloc_inodes
,
1567 &root
->fs_info
->delalloc_inodes
);
1568 set_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1569 &BTRFS_I(inode
)->runtime_flags
);
1571 spin_unlock(&root
->fs_info
->delalloc_lock
);
1573 spin_unlock(&BTRFS_I(inode
)->lock
);
1578 * extent_io.c clear_bit_hook, see set_bit_hook for why
1580 static void btrfs_clear_bit_hook(struct inode
*inode
,
1581 struct extent_state
*state
,
1582 unsigned long *bits
)
1585 * set_bit and clear bit hooks normally require _irqsave/restore
1586 * but in this case, we are only testing for the DELALLOC
1587 * bit, which is only set or cleared with irqs on
1589 if ((state
->state
& EXTENT_DELALLOC
) && (*bits
& EXTENT_DELALLOC
)) {
1590 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1591 u64 len
= state
->end
+ 1 - state
->start
;
1592 bool do_list
= !btrfs_is_free_space_inode(inode
);
1594 if (*bits
& EXTENT_FIRST_DELALLOC
) {
1595 *bits
&= ~EXTENT_FIRST_DELALLOC
;
1596 } else if (!(*bits
& EXTENT_DO_ACCOUNTING
)) {
1597 spin_lock(&BTRFS_I(inode
)->lock
);
1598 BTRFS_I(inode
)->outstanding_extents
--;
1599 spin_unlock(&BTRFS_I(inode
)->lock
);
1602 if (*bits
& EXTENT_DO_ACCOUNTING
)
1603 btrfs_delalloc_release_metadata(inode
, len
);
1605 if (root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
1607 btrfs_free_reserved_data_space(inode
, len
);
1609 __percpu_counter_add(&root
->fs_info
->delalloc_bytes
, -len
,
1610 root
->fs_info
->delalloc_batch
);
1611 spin_lock(&BTRFS_I(inode
)->lock
);
1612 BTRFS_I(inode
)->delalloc_bytes
-= len
;
1613 if (do_list
&& BTRFS_I(inode
)->delalloc_bytes
== 0 &&
1614 test_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1615 &BTRFS_I(inode
)->runtime_flags
)) {
1616 spin_lock(&root
->fs_info
->delalloc_lock
);
1617 if (!list_empty(&BTRFS_I(inode
)->delalloc_inodes
)) {
1618 list_del_init(&BTRFS_I(inode
)->delalloc_inodes
);
1619 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
1620 &BTRFS_I(inode
)->runtime_flags
);
1622 spin_unlock(&root
->fs_info
->delalloc_lock
);
1624 spin_unlock(&BTRFS_I(inode
)->lock
);
1629 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1630 * we don't create bios that span stripes or chunks
1632 int btrfs_merge_bio_hook(int rw
, struct page
*page
, unsigned long offset
,
1633 size_t size
, struct bio
*bio
,
1634 unsigned long bio_flags
)
1636 struct btrfs_root
*root
= BTRFS_I(page
->mapping
->host
)->root
;
1637 u64 logical
= (u64
)bio
->bi_sector
<< 9;
1642 if (bio_flags
& EXTENT_BIO_COMPRESSED
)
1645 length
= bio
->bi_size
;
1646 map_length
= length
;
1647 ret
= btrfs_map_block(root
->fs_info
, rw
, logical
,
1648 &map_length
, NULL
, 0);
1649 /* Will always return 0 with map_multi == NULL */
1651 if (map_length
< length
+ size
)
1657 * in order to insert checksums into the metadata in large chunks,
1658 * we wait until bio submission time. All the pages in the bio are
1659 * checksummed and sums are attached onto the ordered extent record.
1661 * At IO completion time the cums attached on the ordered extent record
1662 * are inserted into the btree
1664 static int __btrfs_submit_bio_start(struct inode
*inode
, int rw
,
1665 struct bio
*bio
, int mirror_num
,
1666 unsigned long bio_flags
,
1669 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1672 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1673 BUG_ON(ret
); /* -ENOMEM */
1678 * in order to insert checksums into the metadata in large chunks,
1679 * we wait until bio submission time. All the pages in the bio are
1680 * checksummed and sums are attached onto the ordered extent record.
1682 * At IO completion time the cums attached on the ordered extent record
1683 * are inserted into the btree
1685 static int __btrfs_submit_bio_done(struct inode
*inode
, int rw
, struct bio
*bio
,
1686 int mirror_num
, unsigned long bio_flags
,
1689 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1692 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 1);
1694 bio_endio(bio
, ret
);
1699 * extent_io.c submission hook. This does the right thing for csum calculation
1700 * on write, or reading the csums from the tree before a read
1702 static int btrfs_submit_bio_hook(struct inode
*inode
, int rw
, struct bio
*bio
,
1703 int mirror_num
, unsigned long bio_flags
,
1706 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1710 int async
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
1712 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
1714 if (btrfs_is_free_space_inode(inode
))
1717 if (!(rw
& REQ_WRITE
)) {
1718 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, metadata
);
1722 if (bio_flags
& EXTENT_BIO_COMPRESSED
) {
1723 ret
= btrfs_submit_compressed_read(inode
, bio
,
1727 } else if (!skip_sum
) {
1728 ret
= btrfs_lookup_bio_sums(root
, inode
, bio
, NULL
);
1733 } else if (async
&& !skip_sum
) {
1734 /* csum items have already been cloned */
1735 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
1737 /* we're doing a write, do the async checksumming */
1738 ret
= btrfs_wq_submit_bio(BTRFS_I(inode
)->root
->fs_info
,
1739 inode
, rw
, bio
, mirror_num
,
1740 bio_flags
, bio_offset
,
1741 __btrfs_submit_bio_start
,
1742 __btrfs_submit_bio_done
);
1744 } else if (!skip_sum
) {
1745 ret
= btrfs_csum_one_bio(root
, inode
, bio
, 0, 0);
1751 ret
= btrfs_map_bio(root
, rw
, bio
, mirror_num
, 0);
1755 bio_endio(bio
, ret
);
1760 * given a list of ordered sums record them in the inode. This happens
1761 * at IO completion time based on sums calculated at bio submission time.
1763 static noinline
int add_pending_csums(struct btrfs_trans_handle
*trans
,
1764 struct inode
*inode
, u64 file_offset
,
1765 struct list_head
*list
)
1767 struct btrfs_ordered_sum
*sum
;
1769 list_for_each_entry(sum
, list
, list
) {
1770 trans
->adding_csums
= 1;
1771 btrfs_csum_file_blocks(trans
,
1772 BTRFS_I(inode
)->root
->fs_info
->csum_root
, sum
);
1773 trans
->adding_csums
= 0;
1778 int btrfs_set_extent_delalloc(struct inode
*inode
, u64 start
, u64 end
,
1779 struct extent_state
**cached_state
)
1781 WARN_ON((end
& (PAGE_CACHE_SIZE
- 1)) == 0);
1782 return set_extent_delalloc(&BTRFS_I(inode
)->io_tree
, start
, end
,
1783 cached_state
, GFP_NOFS
);
1786 /* see btrfs_writepage_start_hook for details on why this is required */
1787 struct btrfs_writepage_fixup
{
1789 struct btrfs_work work
;
1792 static void btrfs_writepage_fixup_worker(struct btrfs_work
*work
)
1794 struct btrfs_writepage_fixup
*fixup
;
1795 struct btrfs_ordered_extent
*ordered
;
1796 struct extent_state
*cached_state
= NULL
;
1798 struct inode
*inode
;
1803 fixup
= container_of(work
, struct btrfs_writepage_fixup
, work
);
1807 if (!page
->mapping
|| !PageDirty(page
) || !PageChecked(page
)) {
1808 ClearPageChecked(page
);
1812 inode
= page
->mapping
->host
;
1813 page_start
= page_offset(page
);
1814 page_end
= page_offset(page
) + PAGE_CACHE_SIZE
- 1;
1816 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
, 0,
1819 /* already ordered? We're done */
1820 if (PagePrivate2(page
))
1823 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
1825 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
,
1826 page_end
, &cached_state
, GFP_NOFS
);
1828 btrfs_start_ordered_extent(inode
, ordered
, 1);
1829 btrfs_put_ordered_extent(ordered
);
1833 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
1835 mapping_set_error(page
->mapping
, ret
);
1836 end_extent_writepage(page
, ret
, page_start
, page_end
);
1837 ClearPageChecked(page
);
1841 btrfs_set_extent_delalloc(inode
, page_start
, page_end
, &cached_state
);
1842 ClearPageChecked(page
);
1843 set_page_dirty(page
);
1845 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
1846 &cached_state
, GFP_NOFS
);
1849 page_cache_release(page
);
1854 * There are a few paths in the higher layers of the kernel that directly
1855 * set the page dirty bit without asking the filesystem if it is a
1856 * good idea. This causes problems because we want to make sure COW
1857 * properly happens and the data=ordered rules are followed.
1859 * In our case any range that doesn't have the ORDERED bit set
1860 * hasn't been properly setup for IO. We kick off an async process
1861 * to fix it up. The async helper will wait for ordered extents, set
1862 * the delalloc bit and make it safe to write the page.
1864 static int btrfs_writepage_start_hook(struct page
*page
, u64 start
, u64 end
)
1866 struct inode
*inode
= page
->mapping
->host
;
1867 struct btrfs_writepage_fixup
*fixup
;
1868 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1870 /* this page is properly in the ordered list */
1871 if (TestClearPagePrivate2(page
))
1874 if (PageChecked(page
))
1877 fixup
= kzalloc(sizeof(*fixup
), GFP_NOFS
);
1881 SetPageChecked(page
);
1882 page_cache_get(page
);
1883 fixup
->work
.func
= btrfs_writepage_fixup_worker
;
1885 btrfs_queue_worker(&root
->fs_info
->fixup_workers
, &fixup
->work
);
1889 static int insert_reserved_file_extent(struct btrfs_trans_handle
*trans
,
1890 struct inode
*inode
, u64 file_pos
,
1891 u64 disk_bytenr
, u64 disk_num_bytes
,
1892 u64 num_bytes
, u64 ram_bytes
,
1893 u8 compression
, u8 encryption
,
1894 u16 other_encoding
, int extent_type
)
1896 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
1897 struct btrfs_file_extent_item
*fi
;
1898 struct btrfs_path
*path
;
1899 struct extent_buffer
*leaf
;
1900 struct btrfs_key ins
;
1903 path
= btrfs_alloc_path();
1907 path
->leave_spinning
= 1;
1910 * we may be replacing one extent in the tree with another.
1911 * The new extent is pinned in the extent map, and we don't want
1912 * to drop it from the cache until it is completely in the btree.
1914 * So, tell btrfs_drop_extents to leave this extent in the cache.
1915 * the caller is expected to unpin it and allow it to be merged
1918 ret
= btrfs_drop_extents(trans
, root
, inode
, file_pos
,
1919 file_pos
+ num_bytes
, 0);
1923 ins
.objectid
= btrfs_ino(inode
);
1924 ins
.offset
= file_pos
;
1925 ins
.type
= BTRFS_EXTENT_DATA_KEY
;
1926 ret
= btrfs_insert_empty_item(trans
, root
, path
, &ins
, sizeof(*fi
));
1929 leaf
= path
->nodes
[0];
1930 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
1931 struct btrfs_file_extent_item
);
1932 btrfs_set_file_extent_generation(leaf
, fi
, trans
->transid
);
1933 btrfs_set_file_extent_type(leaf
, fi
, extent_type
);
1934 btrfs_set_file_extent_disk_bytenr(leaf
, fi
, disk_bytenr
);
1935 btrfs_set_file_extent_disk_num_bytes(leaf
, fi
, disk_num_bytes
);
1936 btrfs_set_file_extent_offset(leaf
, fi
, 0);
1937 btrfs_set_file_extent_num_bytes(leaf
, fi
, num_bytes
);
1938 btrfs_set_file_extent_ram_bytes(leaf
, fi
, ram_bytes
);
1939 btrfs_set_file_extent_compression(leaf
, fi
, compression
);
1940 btrfs_set_file_extent_encryption(leaf
, fi
, encryption
);
1941 btrfs_set_file_extent_other_encoding(leaf
, fi
, other_encoding
);
1943 btrfs_mark_buffer_dirty(leaf
);
1944 btrfs_release_path(path
);
1946 inode_add_bytes(inode
, num_bytes
);
1948 ins
.objectid
= disk_bytenr
;
1949 ins
.offset
= disk_num_bytes
;
1950 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1951 ret
= btrfs_alloc_reserved_file_extent(trans
, root
,
1952 root
->root_key
.objectid
,
1953 btrfs_ino(inode
), file_pos
, &ins
);
1955 btrfs_free_path(path
);
1960 /* snapshot-aware defrag */
1961 struct sa_defrag_extent_backref
{
1962 struct rb_node node
;
1963 struct old_sa_defrag_extent
*old
;
1972 struct old_sa_defrag_extent
{
1973 struct list_head list
;
1974 struct new_sa_defrag_extent
*new;
1983 struct new_sa_defrag_extent
{
1984 struct rb_root root
;
1985 struct list_head head
;
1986 struct btrfs_path
*path
;
1987 struct inode
*inode
;
1995 static int backref_comp(struct sa_defrag_extent_backref
*b1
,
1996 struct sa_defrag_extent_backref
*b2
)
1998 if (b1
->root_id
< b2
->root_id
)
2000 else if (b1
->root_id
> b2
->root_id
)
2003 if (b1
->inum
< b2
->inum
)
2005 else if (b1
->inum
> b2
->inum
)
2008 if (b1
->file_pos
< b2
->file_pos
)
2010 else if (b1
->file_pos
> b2
->file_pos
)
2014 * [------------------------------] ===> (a range of space)
2015 * |<--->| |<---->| =============> (fs/file tree A)
2016 * |<---------------------------->| ===> (fs/file tree B)
2018 * A range of space can refer to two file extents in one tree while
2019 * refer to only one file extent in another tree.
2021 * So we may process a disk offset more than one time(two extents in A)
2022 * and locate at the same extent(one extent in B), then insert two same
2023 * backrefs(both refer to the extent in B).
2028 static void backref_insert(struct rb_root
*root
,
2029 struct sa_defrag_extent_backref
*backref
)
2031 struct rb_node
**p
= &root
->rb_node
;
2032 struct rb_node
*parent
= NULL
;
2033 struct sa_defrag_extent_backref
*entry
;
2038 entry
= rb_entry(parent
, struct sa_defrag_extent_backref
, node
);
2040 ret
= backref_comp(backref
, entry
);
2044 p
= &(*p
)->rb_right
;
2047 rb_link_node(&backref
->node
, parent
, p
);
2048 rb_insert_color(&backref
->node
, root
);
2052 * Note the backref might has changed, and in this case we just return 0.
2054 static noinline
int record_one_backref(u64 inum
, u64 offset
, u64 root_id
,
2057 struct btrfs_file_extent_item
*extent
;
2058 struct btrfs_fs_info
*fs_info
;
2059 struct old_sa_defrag_extent
*old
= ctx
;
2060 struct new_sa_defrag_extent
*new = old
->new;
2061 struct btrfs_path
*path
= new->path
;
2062 struct btrfs_key key
;
2063 struct btrfs_root
*root
;
2064 struct sa_defrag_extent_backref
*backref
;
2065 struct extent_buffer
*leaf
;
2066 struct inode
*inode
= new->inode
;
2072 if (BTRFS_I(inode
)->root
->root_key
.objectid
== root_id
&&
2073 inum
== btrfs_ino(inode
))
2076 key
.objectid
= root_id
;
2077 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2078 key
.offset
= (u64
)-1;
2080 fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2081 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2083 if (PTR_ERR(root
) == -ENOENT
)
2086 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2087 inum
, offset
, root_id
);
2088 return PTR_ERR(root
);
2091 key
.objectid
= inum
;
2092 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2093 if (offset
> (u64
)-1 << 32)
2096 key
.offset
= offset
;
2098 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2107 leaf
= path
->nodes
[0];
2108 slot
= path
->slots
[0];
2110 if (slot
>= btrfs_header_nritems(leaf
)) {
2111 ret
= btrfs_next_leaf(root
, path
);
2114 } else if (ret
> 0) {
2123 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
2125 if (key
.objectid
> inum
)
2128 if (key
.objectid
< inum
|| key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2131 extent
= btrfs_item_ptr(leaf
, slot
,
2132 struct btrfs_file_extent_item
);
2134 if (btrfs_file_extent_disk_bytenr(leaf
, extent
) != old
->bytenr
)
2137 extent_offset
= btrfs_file_extent_offset(leaf
, extent
);
2138 if (key
.offset
- extent_offset
!= offset
)
2141 num_bytes
= btrfs_file_extent_num_bytes(leaf
, extent
);
2142 if (extent_offset
>= old
->extent_offset
+ old
->offset
+
2143 old
->len
|| extent_offset
+ num_bytes
<=
2144 old
->extent_offset
+ old
->offset
)
2150 backref
= kmalloc(sizeof(*backref
), GFP_NOFS
);
2156 backref
->root_id
= root_id
;
2157 backref
->inum
= inum
;
2158 backref
->file_pos
= offset
+ extent_offset
;
2159 backref
->num_bytes
= num_bytes
;
2160 backref
->extent_offset
= extent_offset
;
2161 backref
->generation
= btrfs_file_extent_generation(leaf
, extent
);
2163 backref_insert(&new->root
, backref
);
2166 btrfs_release_path(path
);
2171 static noinline
bool record_extent_backrefs(struct btrfs_path
*path
,
2172 struct new_sa_defrag_extent
*new)
2174 struct btrfs_fs_info
*fs_info
= BTRFS_I(new->inode
)->root
->fs_info
;
2175 struct old_sa_defrag_extent
*old
, *tmp
;
2180 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2181 ret
= iterate_inodes_from_logical(old
->bytenr
, fs_info
,
2182 path
, record_one_backref
,
2184 BUG_ON(ret
< 0 && ret
!= -ENOENT
);
2186 /* no backref to be processed for this extent */
2188 list_del(&old
->list
);
2193 if (list_empty(&new->head
))
2199 static int relink_is_mergable(struct extent_buffer
*leaf
,
2200 struct btrfs_file_extent_item
*fi
,
2203 if (btrfs_file_extent_disk_bytenr(leaf
, fi
) != disk_bytenr
)
2206 if (btrfs_file_extent_type(leaf
, fi
) != BTRFS_FILE_EXTENT_REG
)
2209 if (btrfs_file_extent_compression(leaf
, fi
) ||
2210 btrfs_file_extent_encryption(leaf
, fi
) ||
2211 btrfs_file_extent_other_encoding(leaf
, fi
))
2218 * Note the backref might has changed, and in this case we just return 0.
2220 static noinline
int relink_extent_backref(struct btrfs_path
*path
,
2221 struct sa_defrag_extent_backref
*prev
,
2222 struct sa_defrag_extent_backref
*backref
)
2224 struct btrfs_file_extent_item
*extent
;
2225 struct btrfs_file_extent_item
*item
;
2226 struct btrfs_ordered_extent
*ordered
;
2227 struct btrfs_trans_handle
*trans
;
2228 struct btrfs_fs_info
*fs_info
;
2229 struct btrfs_root
*root
;
2230 struct btrfs_key key
;
2231 struct extent_buffer
*leaf
;
2232 struct old_sa_defrag_extent
*old
= backref
->old
;
2233 struct new_sa_defrag_extent
*new = old
->new;
2234 struct inode
*src_inode
= new->inode
;
2235 struct inode
*inode
;
2236 struct extent_state
*cached
= NULL
;
2245 if (prev
&& prev
->root_id
== backref
->root_id
&&
2246 prev
->inum
== backref
->inum
&&
2247 prev
->file_pos
+ prev
->num_bytes
== backref
->file_pos
)
2250 /* step 1: get root */
2251 key
.objectid
= backref
->root_id
;
2252 key
.type
= BTRFS_ROOT_ITEM_KEY
;
2253 key
.offset
= (u64
)-1;
2255 fs_info
= BTRFS_I(src_inode
)->root
->fs_info
;
2256 index
= srcu_read_lock(&fs_info
->subvol_srcu
);
2258 root
= btrfs_read_fs_root_no_name(fs_info
, &key
);
2260 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2261 if (PTR_ERR(root
) == -ENOENT
)
2263 return PTR_ERR(root
);
2265 if (btrfs_root_refs(&root
->root_item
) == 0) {
2266 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2267 /* parse ENOENT to 0 */
2271 /* step 2: get inode */
2272 key
.objectid
= backref
->inum
;
2273 key
.type
= BTRFS_INODE_ITEM_KEY
;
2276 inode
= btrfs_iget(fs_info
->sb
, &key
, root
, NULL
);
2277 if (IS_ERR(inode
)) {
2278 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2282 srcu_read_unlock(&fs_info
->subvol_srcu
, index
);
2284 /* step 3: relink backref */
2285 lock_start
= backref
->file_pos
;
2286 lock_end
= backref
->file_pos
+ backref
->num_bytes
- 1;
2287 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2290 ordered
= btrfs_lookup_first_ordered_extent(inode
, lock_end
);
2292 btrfs_put_ordered_extent(ordered
);
2296 trans
= btrfs_join_transaction(root
);
2297 if (IS_ERR(trans
)) {
2298 ret
= PTR_ERR(trans
);
2302 key
.objectid
= backref
->inum
;
2303 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2304 key
.offset
= backref
->file_pos
;
2306 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2309 } else if (ret
> 0) {
2314 extent
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
2315 struct btrfs_file_extent_item
);
2317 if (btrfs_file_extent_generation(path
->nodes
[0], extent
) !=
2318 backref
->generation
)
2321 btrfs_release_path(path
);
2323 start
= backref
->file_pos
;
2324 if (backref
->extent_offset
< old
->extent_offset
+ old
->offset
)
2325 start
+= old
->extent_offset
+ old
->offset
-
2326 backref
->extent_offset
;
2328 len
= min(backref
->extent_offset
+ backref
->num_bytes
,
2329 old
->extent_offset
+ old
->offset
+ old
->len
);
2330 len
-= max(backref
->extent_offset
, old
->extent_offset
+ old
->offset
);
2332 ret
= btrfs_drop_extents(trans
, root
, inode
, start
,
2337 key
.objectid
= btrfs_ino(inode
);
2338 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2341 path
->leave_spinning
= 1;
2343 struct btrfs_file_extent_item
*fi
;
2345 struct btrfs_key found_key
;
2347 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 1, 1);
2352 leaf
= path
->nodes
[0];
2353 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2355 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
2356 struct btrfs_file_extent_item
);
2357 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
2359 if (relink_is_mergable(leaf
, fi
, new->bytenr
) &&
2360 extent_len
+ found_key
.offset
== start
) {
2361 btrfs_set_file_extent_num_bytes(leaf
, fi
,
2363 btrfs_mark_buffer_dirty(leaf
);
2364 inode_add_bytes(inode
, len
);
2370 btrfs_release_path(path
);
2375 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
2378 btrfs_abort_transaction(trans
, root
, ret
);
2382 leaf
= path
->nodes
[0];
2383 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
2384 struct btrfs_file_extent_item
);
2385 btrfs_set_file_extent_disk_bytenr(leaf
, item
, new->bytenr
);
2386 btrfs_set_file_extent_disk_num_bytes(leaf
, item
, new->disk_len
);
2387 btrfs_set_file_extent_offset(leaf
, item
, start
- new->file_pos
);
2388 btrfs_set_file_extent_num_bytes(leaf
, item
, len
);
2389 btrfs_set_file_extent_ram_bytes(leaf
, item
, new->len
);
2390 btrfs_set_file_extent_generation(leaf
, item
, trans
->transid
);
2391 btrfs_set_file_extent_type(leaf
, item
, BTRFS_FILE_EXTENT_REG
);
2392 btrfs_set_file_extent_compression(leaf
, item
, new->compress_type
);
2393 btrfs_set_file_extent_encryption(leaf
, item
, 0);
2394 btrfs_set_file_extent_other_encoding(leaf
, item
, 0);
2396 btrfs_mark_buffer_dirty(leaf
);
2397 inode_add_bytes(inode
, len
);
2398 btrfs_release_path(path
);
2400 ret
= btrfs_inc_extent_ref(trans
, root
, new->bytenr
,
2402 backref
->root_id
, backref
->inum
,
2403 new->file_pos
, 0); /* start - extent_offset */
2405 btrfs_abort_transaction(trans
, root
, ret
);
2411 btrfs_release_path(path
);
2412 path
->leave_spinning
= 0;
2413 btrfs_end_transaction(trans
, root
);
2415 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lock_start
, lock_end
,
2421 static void relink_file_extents(struct new_sa_defrag_extent
*new)
2423 struct btrfs_path
*path
;
2424 struct old_sa_defrag_extent
*old
, *tmp
;
2425 struct sa_defrag_extent_backref
*backref
;
2426 struct sa_defrag_extent_backref
*prev
= NULL
;
2427 struct inode
*inode
;
2428 struct btrfs_root
*root
;
2429 struct rb_node
*node
;
2433 root
= BTRFS_I(inode
)->root
;
2435 path
= btrfs_alloc_path();
2439 if (!record_extent_backrefs(path
, new)) {
2440 btrfs_free_path(path
);
2443 btrfs_release_path(path
);
2446 node
= rb_first(&new->root
);
2449 rb_erase(node
, &new->root
);
2451 backref
= rb_entry(node
, struct sa_defrag_extent_backref
, node
);
2453 ret
= relink_extent_backref(path
, prev
, backref
);
2466 btrfs_free_path(path
);
2468 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2469 list_del(&old
->list
);
2473 atomic_dec(&root
->fs_info
->defrag_running
);
2474 wake_up(&root
->fs_info
->transaction_wait
);
2479 static struct new_sa_defrag_extent
*
2480 record_old_file_extents(struct inode
*inode
,
2481 struct btrfs_ordered_extent
*ordered
)
2483 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2484 struct btrfs_path
*path
;
2485 struct btrfs_key key
;
2486 struct old_sa_defrag_extent
*old
, *tmp
;
2487 struct new_sa_defrag_extent
*new;
2490 new = kmalloc(sizeof(*new), GFP_NOFS
);
2495 new->file_pos
= ordered
->file_offset
;
2496 new->len
= ordered
->len
;
2497 new->bytenr
= ordered
->start
;
2498 new->disk_len
= ordered
->disk_len
;
2499 new->compress_type
= ordered
->compress_type
;
2500 new->root
= RB_ROOT
;
2501 INIT_LIST_HEAD(&new->head
);
2503 path
= btrfs_alloc_path();
2507 key
.objectid
= btrfs_ino(inode
);
2508 key
.type
= BTRFS_EXTENT_DATA_KEY
;
2509 key
.offset
= new->file_pos
;
2511 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2514 if (ret
> 0 && path
->slots
[0] > 0)
2517 /* find out all the old extents for the file range */
2519 struct btrfs_file_extent_item
*extent
;
2520 struct extent_buffer
*l
;
2529 slot
= path
->slots
[0];
2531 if (slot
>= btrfs_header_nritems(l
)) {
2532 ret
= btrfs_next_leaf(root
, path
);
2540 btrfs_item_key_to_cpu(l
, &key
, slot
);
2542 if (key
.objectid
!= btrfs_ino(inode
))
2544 if (key
.type
!= BTRFS_EXTENT_DATA_KEY
)
2546 if (key
.offset
>= new->file_pos
+ new->len
)
2549 extent
= btrfs_item_ptr(l
, slot
, struct btrfs_file_extent_item
);
2551 num_bytes
= btrfs_file_extent_num_bytes(l
, extent
);
2552 if (key
.offset
+ num_bytes
< new->file_pos
)
2555 disk_bytenr
= btrfs_file_extent_disk_bytenr(l
, extent
);
2559 extent_offset
= btrfs_file_extent_offset(l
, extent
);
2561 old
= kmalloc(sizeof(*old
), GFP_NOFS
);
2565 offset
= max(new->file_pos
, key
.offset
);
2566 end
= min(new->file_pos
+ new->len
, key
.offset
+ num_bytes
);
2568 old
->bytenr
= disk_bytenr
;
2569 old
->extent_offset
= extent_offset
;
2570 old
->offset
= offset
- key
.offset
;
2571 old
->len
= end
- offset
;
2574 list_add_tail(&old
->list
, &new->head
);
2580 btrfs_free_path(path
);
2581 atomic_inc(&root
->fs_info
->defrag_running
);
2586 list_for_each_entry_safe(old
, tmp
, &new->head
, list
) {
2587 list_del(&old
->list
);
2591 btrfs_free_path(path
);
2598 * helper function for btrfs_finish_ordered_io, this
2599 * just reads in some of the csum leaves to prime them into ram
2600 * before we start the transaction. It limits the amount of btree
2601 * reads required while inside the transaction.
2603 /* as ordered data IO finishes, this gets called so we can finish
2604 * an ordered extent if the range of bytes in the file it covers are
2607 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent
*ordered_extent
)
2609 struct inode
*inode
= ordered_extent
->inode
;
2610 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2611 struct btrfs_trans_handle
*trans
= NULL
;
2612 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
2613 struct extent_state
*cached_state
= NULL
;
2614 struct new_sa_defrag_extent
*new = NULL
;
2615 int compress_type
= 0;
2619 nolock
= btrfs_is_free_space_inode(inode
);
2621 if (test_bit(BTRFS_ORDERED_IOERR
, &ordered_extent
->flags
)) {
2626 if (test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
)) {
2627 BUG_ON(!list_empty(&ordered_extent
->list
)); /* Logic error */
2628 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2630 trans
= btrfs_join_transaction_nolock(root
);
2632 trans
= btrfs_join_transaction(root
);
2633 if (IS_ERR(trans
)) {
2634 ret
= PTR_ERR(trans
);
2638 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2639 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2640 if (ret
) /* -ENOMEM or corruption */
2641 btrfs_abort_transaction(trans
, root
, ret
);
2645 lock_extent_bits(io_tree
, ordered_extent
->file_offset
,
2646 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2649 ret
= test_range_bit(io_tree
, ordered_extent
->file_offset
,
2650 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2651 EXTENT_DEFRAG
, 1, cached_state
);
2653 u64 last_snapshot
= btrfs_root_last_snapshot(&root
->root_item
);
2654 if (last_snapshot
>= BTRFS_I(inode
)->generation
)
2655 /* the inode is shared */
2656 new = record_old_file_extents(inode
, ordered_extent
);
2658 clear_extent_bit(io_tree
, ordered_extent
->file_offset
,
2659 ordered_extent
->file_offset
+ ordered_extent
->len
- 1,
2660 EXTENT_DEFRAG
, 0, 0, &cached_state
, GFP_NOFS
);
2664 trans
= btrfs_join_transaction_nolock(root
);
2666 trans
= btrfs_join_transaction(root
);
2667 if (IS_ERR(trans
)) {
2668 ret
= PTR_ERR(trans
);
2672 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
2674 if (test_bit(BTRFS_ORDERED_COMPRESSED
, &ordered_extent
->flags
))
2675 compress_type
= ordered_extent
->compress_type
;
2676 if (test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
)) {
2677 BUG_ON(compress_type
);
2678 ret
= btrfs_mark_extent_written(trans
, inode
,
2679 ordered_extent
->file_offset
,
2680 ordered_extent
->file_offset
+
2681 ordered_extent
->len
);
2683 BUG_ON(root
== root
->fs_info
->tree_root
);
2684 ret
= insert_reserved_file_extent(trans
, inode
,
2685 ordered_extent
->file_offset
,
2686 ordered_extent
->start
,
2687 ordered_extent
->disk_len
,
2688 ordered_extent
->len
,
2689 ordered_extent
->len
,
2690 compress_type
, 0, 0,
2691 BTRFS_FILE_EXTENT_REG
);
2693 unpin_extent_cache(&BTRFS_I(inode
)->extent_tree
,
2694 ordered_extent
->file_offset
, ordered_extent
->len
,
2697 btrfs_abort_transaction(trans
, root
, ret
);
2701 add_pending_csums(trans
, inode
, ordered_extent
->file_offset
,
2702 &ordered_extent
->list
);
2704 btrfs_ordered_update_i_size(inode
, 0, ordered_extent
);
2705 ret
= btrfs_update_inode_fallback(trans
, root
, inode
);
2706 if (ret
) { /* -ENOMEM or corruption */
2707 btrfs_abort_transaction(trans
, root
, ret
);
2712 unlock_extent_cached(io_tree
, ordered_extent
->file_offset
,
2713 ordered_extent
->file_offset
+
2714 ordered_extent
->len
- 1, &cached_state
, GFP_NOFS
);
2716 if (root
!= root
->fs_info
->tree_root
)
2717 btrfs_delalloc_release_metadata(inode
, ordered_extent
->len
);
2719 btrfs_end_transaction(trans
, root
);
2722 clear_extent_uptodate(io_tree
, ordered_extent
->file_offset
,
2723 ordered_extent
->file_offset
+
2724 ordered_extent
->len
- 1, NULL
, GFP_NOFS
);
2727 * If the ordered extent had an IOERR or something else went
2728 * wrong we need to return the space for this ordered extent
2729 * back to the allocator.
2731 if (!test_bit(BTRFS_ORDERED_NOCOW
, &ordered_extent
->flags
) &&
2732 !test_bit(BTRFS_ORDERED_PREALLOC
, &ordered_extent
->flags
))
2733 btrfs_free_reserved_extent(root
, ordered_extent
->start
,
2734 ordered_extent
->disk_len
);
2739 * This needs to be done to make sure anybody waiting knows we are done
2740 * updating everything for this ordered extent.
2742 btrfs_remove_ordered_extent(inode
, ordered_extent
);
2744 /* for snapshot-aware defrag */
2746 relink_file_extents(new);
2749 btrfs_put_ordered_extent(ordered_extent
);
2750 /* once for the tree */
2751 btrfs_put_ordered_extent(ordered_extent
);
2756 static void finish_ordered_fn(struct btrfs_work
*work
)
2758 struct btrfs_ordered_extent
*ordered_extent
;
2759 ordered_extent
= container_of(work
, struct btrfs_ordered_extent
, work
);
2760 btrfs_finish_ordered_io(ordered_extent
);
2763 static int btrfs_writepage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2764 struct extent_state
*state
, int uptodate
)
2766 struct inode
*inode
= page
->mapping
->host
;
2767 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2768 struct btrfs_ordered_extent
*ordered_extent
= NULL
;
2769 struct btrfs_workers
*workers
;
2771 trace_btrfs_writepage_end_io_hook(page
, start
, end
, uptodate
);
2773 ClearPagePrivate2(page
);
2774 if (!btrfs_dec_test_ordered_pending(inode
, &ordered_extent
, start
,
2775 end
- start
+ 1, uptodate
))
2778 ordered_extent
->work
.func
= finish_ordered_fn
;
2779 ordered_extent
->work
.flags
= 0;
2781 if (btrfs_is_free_space_inode(inode
))
2782 workers
= &root
->fs_info
->endio_freespace_worker
;
2784 workers
= &root
->fs_info
->endio_write_workers
;
2785 btrfs_queue_worker(workers
, &ordered_extent
->work
);
2791 * when reads are done, we need to check csums to verify the data is correct
2792 * if there's a match, we allow the bio to finish. If not, the code in
2793 * extent_io.c will try to find good copies for us.
2795 static int btrfs_readpage_end_io_hook(struct page
*page
, u64 start
, u64 end
,
2796 struct extent_state
*state
, 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 u64
private = ~(u32
)0;
2804 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2806 static DEFINE_RATELIMIT_STATE(_rs
, DEFAULT_RATELIMIT_INTERVAL
,
2807 DEFAULT_RATELIMIT_BURST
);
2809 if (PageChecked(page
)) {
2810 ClearPageChecked(page
);
2814 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)
2817 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
&&
2818 test_range_bit(io_tree
, start
, end
, EXTENT_NODATASUM
, 1, NULL
)) {
2819 clear_extent_bits(io_tree
, start
, end
, EXTENT_NODATASUM
,
2824 if (state
&& state
->start
== start
) {
2825 private = state
->private;
2828 ret
= get_state_private(io_tree
, start
, &private);
2830 kaddr
= kmap_atomic(page
);
2834 csum
= btrfs_csum_data(kaddr
+ offset
, csum
, end
- start
+ 1);
2835 btrfs_csum_final(csum
, (char *)&csum
);
2836 if (csum
!= private)
2839 kunmap_atomic(kaddr
);
2844 if (__ratelimit(&_rs
))
2845 btrfs_info(root
->fs_info
, "csum failed ino %llu off %llu csum %u private %llu",
2846 (unsigned long long)btrfs_ino(page
->mapping
->host
),
2847 (unsigned long long)start
, csum
,
2848 (unsigned long long)private);
2849 memset(kaddr
+ offset
, 1, end
- start
+ 1);
2850 flush_dcache_page(page
);
2851 kunmap_atomic(kaddr
);
2857 struct delayed_iput
{
2858 struct list_head list
;
2859 struct inode
*inode
;
2862 /* JDM: If this is fs-wide, why can't we add a pointer to
2863 * btrfs_inode instead and avoid the allocation? */
2864 void btrfs_add_delayed_iput(struct inode
*inode
)
2866 struct btrfs_fs_info
*fs_info
= BTRFS_I(inode
)->root
->fs_info
;
2867 struct delayed_iput
*delayed
;
2869 if (atomic_add_unless(&inode
->i_count
, -1, 1))
2872 delayed
= kmalloc(sizeof(*delayed
), GFP_NOFS
| __GFP_NOFAIL
);
2873 delayed
->inode
= inode
;
2875 spin_lock(&fs_info
->delayed_iput_lock
);
2876 list_add_tail(&delayed
->list
, &fs_info
->delayed_iputs
);
2877 spin_unlock(&fs_info
->delayed_iput_lock
);
2880 void btrfs_run_delayed_iputs(struct btrfs_root
*root
)
2883 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2884 struct delayed_iput
*delayed
;
2887 spin_lock(&fs_info
->delayed_iput_lock
);
2888 empty
= list_empty(&fs_info
->delayed_iputs
);
2889 spin_unlock(&fs_info
->delayed_iput_lock
);
2893 spin_lock(&fs_info
->delayed_iput_lock
);
2894 list_splice_init(&fs_info
->delayed_iputs
, &list
);
2895 spin_unlock(&fs_info
->delayed_iput_lock
);
2897 while (!list_empty(&list
)) {
2898 delayed
= list_entry(list
.next
, struct delayed_iput
, list
);
2899 list_del(&delayed
->list
);
2900 iput(delayed
->inode
);
2906 * This is called in transaction commit time. If there are no orphan
2907 * files in the subvolume, it removes orphan item and frees block_rsv
2910 void btrfs_orphan_commit_root(struct btrfs_trans_handle
*trans
,
2911 struct btrfs_root
*root
)
2913 struct btrfs_block_rsv
*block_rsv
;
2916 if (atomic_read(&root
->orphan_inodes
) ||
2917 root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
)
2920 spin_lock(&root
->orphan_lock
);
2921 if (atomic_read(&root
->orphan_inodes
)) {
2922 spin_unlock(&root
->orphan_lock
);
2926 if (root
->orphan_cleanup_state
!= ORPHAN_CLEANUP_DONE
) {
2927 spin_unlock(&root
->orphan_lock
);
2931 block_rsv
= root
->orphan_block_rsv
;
2932 root
->orphan_block_rsv
= NULL
;
2933 spin_unlock(&root
->orphan_lock
);
2935 if (root
->orphan_item_inserted
&&
2936 btrfs_root_refs(&root
->root_item
) > 0) {
2937 ret
= btrfs_del_orphan_item(trans
, root
->fs_info
->tree_root
,
2938 root
->root_key
.objectid
);
2940 root
->orphan_item_inserted
= 0;
2944 WARN_ON(block_rsv
->size
> 0);
2945 btrfs_free_block_rsv(root
, block_rsv
);
2950 * This creates an orphan entry for the given inode in case something goes
2951 * wrong in the middle of an unlink/truncate.
2953 * NOTE: caller of this function should reserve 5 units of metadata for
2956 int btrfs_orphan_add(struct btrfs_trans_handle
*trans
, struct inode
*inode
)
2958 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
2959 struct btrfs_block_rsv
*block_rsv
= NULL
;
2964 if (!root
->orphan_block_rsv
) {
2965 block_rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
2970 spin_lock(&root
->orphan_lock
);
2971 if (!root
->orphan_block_rsv
) {
2972 root
->orphan_block_rsv
= block_rsv
;
2973 } else if (block_rsv
) {
2974 btrfs_free_block_rsv(root
, block_rsv
);
2978 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
2979 &BTRFS_I(inode
)->runtime_flags
)) {
2982 * For proper ENOSPC handling, we should do orphan
2983 * cleanup when mounting. But this introduces backward
2984 * compatibility issue.
2986 if (!xchg(&root
->orphan_item_inserted
, 1))
2992 atomic_inc(&root
->orphan_inodes
);
2995 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
2996 &BTRFS_I(inode
)->runtime_flags
))
2998 spin_unlock(&root
->orphan_lock
);
3000 /* grab metadata reservation from transaction handle */
3002 ret
= btrfs_orphan_reserve_metadata(trans
, inode
);
3003 BUG_ON(ret
); /* -ENOSPC in reservation; Logic error? JDM */
3006 /* insert an orphan item to track this unlinked/truncated file */
3008 ret
= btrfs_insert_orphan_item(trans
, root
, btrfs_ino(inode
));
3009 if (ret
&& ret
!= -EEXIST
) {
3010 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3011 &BTRFS_I(inode
)->runtime_flags
);
3012 btrfs_abort_transaction(trans
, root
, ret
);
3018 /* insert an orphan item to track subvolume contains orphan files */
3020 ret
= btrfs_insert_orphan_item(trans
, root
->fs_info
->tree_root
,
3021 root
->root_key
.objectid
);
3022 if (ret
&& ret
!= -EEXIST
) {
3023 btrfs_abort_transaction(trans
, root
, ret
);
3031 * We have done the truncate/delete so we can go ahead and remove the orphan
3032 * item for this particular inode.
3034 static int btrfs_orphan_del(struct btrfs_trans_handle
*trans
,
3035 struct inode
*inode
)
3037 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3038 int delete_item
= 0;
3039 int release_rsv
= 0;
3042 spin_lock(&root
->orphan_lock
);
3043 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3044 &BTRFS_I(inode
)->runtime_flags
))
3047 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED
,
3048 &BTRFS_I(inode
)->runtime_flags
))
3050 spin_unlock(&root
->orphan_lock
);
3052 if (trans
&& delete_item
) {
3053 ret
= btrfs_del_orphan_item(trans
, root
, btrfs_ino(inode
));
3054 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
3058 btrfs_orphan_release_metadata(inode
);
3059 atomic_dec(&root
->orphan_inodes
);
3066 * this cleans up any orphans that may be left on the list from the last use
3069 int btrfs_orphan_cleanup(struct btrfs_root
*root
)
3071 struct btrfs_path
*path
;
3072 struct extent_buffer
*leaf
;
3073 struct btrfs_key key
, found_key
;
3074 struct btrfs_trans_handle
*trans
;
3075 struct inode
*inode
;
3076 u64 last_objectid
= 0;
3077 int ret
= 0, nr_unlink
= 0, nr_truncate
= 0;
3079 if (cmpxchg(&root
->orphan_cleanup_state
, 0, ORPHAN_CLEANUP_STARTED
))
3082 path
= btrfs_alloc_path();
3089 key
.objectid
= BTRFS_ORPHAN_OBJECTID
;
3090 btrfs_set_key_type(&key
, BTRFS_ORPHAN_ITEM_KEY
);
3091 key
.offset
= (u64
)-1;
3094 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3099 * if ret == 0 means we found what we were searching for, which
3100 * is weird, but possible, so only screw with path if we didn't
3101 * find the key and see if we have stuff that matches
3105 if (path
->slots
[0] == 0)
3110 /* pull out the item */
3111 leaf
= path
->nodes
[0];
3112 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3114 /* make sure the item matches what we want */
3115 if (found_key
.objectid
!= BTRFS_ORPHAN_OBJECTID
)
3117 if (btrfs_key_type(&found_key
) != BTRFS_ORPHAN_ITEM_KEY
)
3120 /* release the path since we're done with it */
3121 btrfs_release_path(path
);
3124 * this is where we are basically btrfs_lookup, without the
3125 * crossing root thing. we store the inode number in the
3126 * offset of the orphan item.
3129 if (found_key
.offset
== last_objectid
) {
3130 btrfs_err(root
->fs_info
,
3131 "Error removing orphan entry, stopping orphan cleanup");
3136 last_objectid
= found_key
.offset
;
3138 found_key
.objectid
= found_key
.offset
;
3139 found_key
.type
= BTRFS_INODE_ITEM_KEY
;
3140 found_key
.offset
= 0;
3141 inode
= btrfs_iget(root
->fs_info
->sb
, &found_key
, root
, NULL
);
3142 ret
= PTR_RET(inode
);
3143 if (ret
&& ret
!= -ESTALE
)
3146 if (ret
== -ESTALE
&& root
== root
->fs_info
->tree_root
) {
3147 struct btrfs_root
*dead_root
;
3148 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3149 int is_dead_root
= 0;
3152 * this is an orphan in the tree root. Currently these
3153 * could come from 2 sources:
3154 * a) a snapshot deletion in progress
3155 * b) a free space cache inode
3156 * We need to distinguish those two, as the snapshot
3157 * orphan must not get deleted.
3158 * find_dead_roots already ran before us, so if this
3159 * is a snapshot deletion, we should find the root
3160 * in the dead_roots list
3162 spin_lock(&fs_info
->trans_lock
);
3163 list_for_each_entry(dead_root
, &fs_info
->dead_roots
,
3165 if (dead_root
->root_key
.objectid
==
3166 found_key
.objectid
) {
3171 spin_unlock(&fs_info
->trans_lock
);
3173 /* prevent this orphan from being found again */
3174 key
.offset
= found_key
.objectid
- 1;
3179 * Inode is already gone but the orphan item is still there,
3180 * kill the orphan item.
3182 if (ret
== -ESTALE
) {
3183 trans
= btrfs_start_transaction(root
, 1);
3184 if (IS_ERR(trans
)) {
3185 ret
= PTR_ERR(trans
);
3188 btrfs_debug(root
->fs_info
, "auto deleting %Lu",
3189 found_key
.objectid
);
3190 ret
= btrfs_del_orphan_item(trans
, root
,
3191 found_key
.objectid
);
3192 BUG_ON(ret
); /* -ENOMEM or corruption (JDM: Recheck) */
3193 btrfs_end_transaction(trans
, root
);
3198 * add this inode to the orphan list so btrfs_orphan_del does
3199 * the proper thing when we hit it
3201 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
3202 &BTRFS_I(inode
)->runtime_flags
);
3203 atomic_inc(&root
->orphan_inodes
);
3205 /* if we have links, this was a truncate, lets do that */
3206 if (inode
->i_nlink
) {
3207 if (!S_ISREG(inode
->i_mode
)) {
3214 /* 1 for the orphan item deletion. */
3215 trans
= btrfs_start_transaction(root
, 1);
3216 if (IS_ERR(trans
)) {
3217 ret
= PTR_ERR(trans
);
3220 ret
= btrfs_orphan_add(trans
, inode
);
3221 btrfs_end_transaction(trans
, root
);
3225 ret
= btrfs_truncate(inode
);
3227 btrfs_orphan_del(NULL
, inode
);
3232 /* this will do delete_inode and everything for us */
3237 /* release the path since we're done with it */
3238 btrfs_release_path(path
);
3240 root
->orphan_cleanup_state
= ORPHAN_CLEANUP_DONE
;
3242 if (root
->orphan_block_rsv
)
3243 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
,
3246 if (root
->orphan_block_rsv
|| root
->orphan_item_inserted
) {
3247 trans
= btrfs_join_transaction(root
);
3249 btrfs_end_transaction(trans
, root
);
3253 btrfs_debug(root
->fs_info
, "unlinked %d orphans", nr_unlink
);
3255 btrfs_debug(root
->fs_info
, "truncated %d orphans", nr_truncate
);
3259 btrfs_crit(root
->fs_info
,
3260 "could not do orphan cleanup %d", ret
);
3261 btrfs_free_path(path
);
3266 * very simple check to peek ahead in the leaf looking for xattrs. If we
3267 * don't find any xattrs, we know there can't be any acls.
3269 * slot is the slot the inode is in, objectid is the objectid of the inode
3271 static noinline
int acls_after_inode_item(struct extent_buffer
*leaf
,
3272 int slot
, u64 objectid
)
3274 u32 nritems
= btrfs_header_nritems(leaf
);
3275 struct btrfs_key found_key
;
3279 while (slot
< nritems
) {
3280 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3282 /* we found a different objectid, there must not be acls */
3283 if (found_key
.objectid
!= objectid
)
3286 /* we found an xattr, assume we've got an acl */
3287 if (found_key
.type
== BTRFS_XATTR_ITEM_KEY
)
3291 * we found a key greater than an xattr key, there can't
3292 * be any acls later on
3294 if (found_key
.type
> BTRFS_XATTR_ITEM_KEY
)
3301 * it goes inode, inode backrefs, xattrs, extents,
3302 * so if there are a ton of hard links to an inode there can
3303 * be a lot of backrefs. Don't waste time searching too hard,
3304 * this is just an optimization
3309 /* we hit the end of the leaf before we found an xattr or
3310 * something larger than an xattr. We have to assume the inode
3317 * read an inode from the btree into the in-memory inode
3319 static void btrfs_read_locked_inode(struct inode
*inode
)
3321 struct btrfs_path
*path
;
3322 struct extent_buffer
*leaf
;
3323 struct btrfs_inode_item
*inode_item
;
3324 struct btrfs_timespec
*tspec
;
3325 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3326 struct btrfs_key location
;
3330 bool filled
= false;
3332 ret
= btrfs_fill_inode(inode
, &rdev
);
3336 path
= btrfs_alloc_path();
3340 path
->leave_spinning
= 1;
3341 memcpy(&location
, &BTRFS_I(inode
)->location
, sizeof(location
));
3343 ret
= btrfs_lookup_inode(NULL
, root
, path
, &location
, 0);
3347 leaf
= path
->nodes
[0];
3352 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3353 struct btrfs_inode_item
);
3354 inode
->i_mode
= btrfs_inode_mode(leaf
, inode_item
);
3355 set_nlink(inode
, btrfs_inode_nlink(leaf
, inode_item
));
3356 i_uid_write(inode
, btrfs_inode_uid(leaf
, inode_item
));
3357 i_gid_write(inode
, btrfs_inode_gid(leaf
, inode_item
));
3358 btrfs_i_size_write(inode
, btrfs_inode_size(leaf
, inode_item
));
3360 tspec
= btrfs_inode_atime(inode_item
);
3361 inode
->i_atime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3362 inode
->i_atime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3364 tspec
= btrfs_inode_mtime(inode_item
);
3365 inode
->i_mtime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3366 inode
->i_mtime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3368 tspec
= btrfs_inode_ctime(inode_item
);
3369 inode
->i_ctime
.tv_sec
= btrfs_timespec_sec(leaf
, tspec
);
3370 inode
->i_ctime
.tv_nsec
= btrfs_timespec_nsec(leaf
, tspec
);
3372 inode_set_bytes(inode
, btrfs_inode_nbytes(leaf
, inode_item
));
3373 BTRFS_I(inode
)->generation
= btrfs_inode_generation(leaf
, inode_item
);
3374 BTRFS_I(inode
)->last_trans
= btrfs_inode_transid(leaf
, inode_item
);
3377 * If we were modified in the current generation and evicted from memory
3378 * and then re-read we need to do a full sync since we don't have any
3379 * idea about which extents were modified before we were evicted from
3382 if (BTRFS_I(inode
)->last_trans
== root
->fs_info
->generation
)
3383 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
3384 &BTRFS_I(inode
)->runtime_flags
);
3386 inode
->i_version
= btrfs_inode_sequence(leaf
, inode_item
);
3387 inode
->i_generation
= BTRFS_I(inode
)->generation
;
3389 rdev
= btrfs_inode_rdev(leaf
, inode_item
);
3391 BTRFS_I(inode
)->index_cnt
= (u64
)-1;
3392 BTRFS_I(inode
)->flags
= btrfs_inode_flags(leaf
, inode_item
);
3395 * try to precache a NULL acl entry for files that don't have
3396 * any xattrs or acls
3398 maybe_acls
= acls_after_inode_item(leaf
, path
->slots
[0],
3401 cache_no_acl(inode
);
3403 btrfs_free_path(path
);
3405 switch (inode
->i_mode
& S_IFMT
) {
3407 inode
->i_mapping
->a_ops
= &btrfs_aops
;
3408 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3409 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
3410 inode
->i_fop
= &btrfs_file_operations
;
3411 inode
->i_op
= &btrfs_file_inode_operations
;
3414 inode
->i_fop
= &btrfs_dir_file_operations
;
3415 if (root
== root
->fs_info
->tree_root
)
3416 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
3418 inode
->i_op
= &btrfs_dir_inode_operations
;
3421 inode
->i_op
= &btrfs_symlink_inode_operations
;
3422 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
3423 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
3426 inode
->i_op
= &btrfs_special_inode_operations
;
3427 init_special_inode(inode
, inode
->i_mode
, rdev
);
3431 btrfs_update_iflags(inode
);
3435 btrfs_free_path(path
);
3436 make_bad_inode(inode
);
3440 * given a leaf and an inode, copy the inode fields into the leaf
3442 static void fill_inode_item(struct btrfs_trans_handle
*trans
,
3443 struct extent_buffer
*leaf
,
3444 struct btrfs_inode_item
*item
,
3445 struct inode
*inode
)
3447 struct btrfs_map_token token
;
3449 btrfs_init_map_token(&token
);
3451 btrfs_set_token_inode_uid(leaf
, item
, i_uid_read(inode
), &token
);
3452 btrfs_set_token_inode_gid(leaf
, item
, i_gid_read(inode
), &token
);
3453 btrfs_set_token_inode_size(leaf
, item
, BTRFS_I(inode
)->disk_i_size
,
3455 btrfs_set_token_inode_mode(leaf
, item
, inode
->i_mode
, &token
);
3456 btrfs_set_token_inode_nlink(leaf
, item
, inode
->i_nlink
, &token
);
3458 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_atime(item
),
3459 inode
->i_atime
.tv_sec
, &token
);
3460 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_atime(item
),
3461 inode
->i_atime
.tv_nsec
, &token
);
3463 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_mtime(item
),
3464 inode
->i_mtime
.tv_sec
, &token
);
3465 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_mtime(item
),
3466 inode
->i_mtime
.tv_nsec
, &token
);
3468 btrfs_set_token_timespec_sec(leaf
, btrfs_inode_ctime(item
),
3469 inode
->i_ctime
.tv_sec
, &token
);
3470 btrfs_set_token_timespec_nsec(leaf
, btrfs_inode_ctime(item
),
3471 inode
->i_ctime
.tv_nsec
, &token
);
3473 btrfs_set_token_inode_nbytes(leaf
, item
, inode_get_bytes(inode
),
3475 btrfs_set_token_inode_generation(leaf
, item
, BTRFS_I(inode
)->generation
,
3477 btrfs_set_token_inode_sequence(leaf
, item
, inode
->i_version
, &token
);
3478 btrfs_set_token_inode_transid(leaf
, item
, trans
->transid
, &token
);
3479 btrfs_set_token_inode_rdev(leaf
, item
, inode
->i_rdev
, &token
);
3480 btrfs_set_token_inode_flags(leaf
, item
, BTRFS_I(inode
)->flags
, &token
);
3481 btrfs_set_token_inode_block_group(leaf
, item
, 0, &token
);
3485 * copy everything in the in-memory inode into the btree.
3487 static noinline
int btrfs_update_inode_item(struct btrfs_trans_handle
*trans
,
3488 struct btrfs_root
*root
, struct inode
*inode
)
3490 struct btrfs_inode_item
*inode_item
;
3491 struct btrfs_path
*path
;
3492 struct extent_buffer
*leaf
;
3495 path
= btrfs_alloc_path();
3499 path
->leave_spinning
= 1;
3500 ret
= btrfs_lookup_inode(trans
, root
, path
, &BTRFS_I(inode
)->location
,
3508 btrfs_unlock_up_safe(path
, 1);
3509 leaf
= path
->nodes
[0];
3510 inode_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
3511 struct btrfs_inode_item
);
3513 fill_inode_item(trans
, leaf
, inode_item
, inode
);
3514 btrfs_mark_buffer_dirty(leaf
);
3515 btrfs_set_inode_last_trans(trans
, inode
);
3518 btrfs_free_path(path
);
3523 * copy everything in the in-memory inode into the btree.
3525 noinline
int btrfs_update_inode(struct btrfs_trans_handle
*trans
,
3526 struct btrfs_root
*root
, struct inode
*inode
)
3531 * If the inode is a free space inode, we can deadlock during commit
3532 * if we put it into the delayed code.
3534 * The data relocation inode should also be directly updated
3537 if (!btrfs_is_free_space_inode(inode
)
3538 && root
->root_key
.objectid
!= BTRFS_DATA_RELOC_TREE_OBJECTID
) {
3539 btrfs_update_root_times(trans
, root
);
3541 ret
= btrfs_delayed_update_inode(trans
, root
, inode
);
3543 btrfs_set_inode_last_trans(trans
, inode
);
3547 return btrfs_update_inode_item(trans
, root
, inode
);
3550 noinline
int btrfs_update_inode_fallback(struct btrfs_trans_handle
*trans
,
3551 struct btrfs_root
*root
,
3552 struct inode
*inode
)
3556 ret
= btrfs_update_inode(trans
, root
, inode
);
3558 return btrfs_update_inode_item(trans
, root
, inode
);
3563 * unlink helper that gets used here in inode.c and in the tree logging
3564 * recovery code. It remove a link in a directory with a given name, and
3565 * also drops the back refs in the inode to the directory
3567 static int __btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3568 struct btrfs_root
*root
,
3569 struct inode
*dir
, struct inode
*inode
,
3570 const char *name
, int name_len
)
3572 struct btrfs_path
*path
;
3574 struct extent_buffer
*leaf
;
3575 struct btrfs_dir_item
*di
;
3576 struct btrfs_key key
;
3578 u64 ino
= btrfs_ino(inode
);
3579 u64 dir_ino
= btrfs_ino(dir
);
3581 path
= btrfs_alloc_path();
3587 path
->leave_spinning
= 1;
3588 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3589 name
, name_len
, -1);
3598 leaf
= path
->nodes
[0];
3599 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3600 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3603 btrfs_release_path(path
);
3605 ret
= btrfs_del_inode_ref(trans
, root
, name
, name_len
, ino
,
3608 btrfs_info(root
->fs_info
,
3609 "failed to delete reference to %.*s, inode %llu parent %llu",
3611 (unsigned long long)ino
, (unsigned long long)dir_ino
);
3612 btrfs_abort_transaction(trans
, root
, ret
);
3616 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3618 btrfs_abort_transaction(trans
, root
, ret
);
3622 ret
= btrfs_del_inode_ref_in_log(trans
, root
, name
, name_len
,
3624 if (ret
!= 0 && ret
!= -ENOENT
) {
3625 btrfs_abort_transaction(trans
, root
, ret
);
3629 ret
= btrfs_del_dir_entries_in_log(trans
, root
, name
, name_len
,
3634 btrfs_abort_transaction(trans
, root
, ret
);
3636 btrfs_free_path(path
);
3640 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3641 inode_inc_iversion(inode
);
3642 inode_inc_iversion(dir
);
3643 inode
->i_ctime
= dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3644 ret
= btrfs_update_inode(trans
, root
, dir
);
3649 int btrfs_unlink_inode(struct btrfs_trans_handle
*trans
,
3650 struct btrfs_root
*root
,
3651 struct inode
*dir
, struct inode
*inode
,
3652 const char *name
, int name_len
)
3655 ret
= __btrfs_unlink_inode(trans
, root
, dir
, inode
, name
, name_len
);
3657 btrfs_drop_nlink(inode
);
3658 ret
= btrfs_update_inode(trans
, root
, inode
);
3664 /* helper to check if there is any shared block in the path */
3665 static int check_path_shared(struct btrfs_root
*root
,
3666 struct btrfs_path
*path
)
3668 struct extent_buffer
*eb
;
3672 for (level
= 0; level
< BTRFS_MAX_LEVEL
; level
++) {
3675 if (!path
->nodes
[level
])
3677 eb
= path
->nodes
[level
];
3678 if (!btrfs_block_can_be_shared(root
, eb
))
3680 ret
= btrfs_lookup_extent_info(NULL
, root
, eb
->start
, level
, 1,
3689 * helper to start transaction for unlink and rmdir.
3691 * unlink and rmdir are special in btrfs, they do not always free space.
3692 * so in enospc case, we should make sure they will free space before
3693 * allowing them to use the global metadata reservation.
3695 static struct btrfs_trans_handle
*__unlink_start_trans(struct inode
*dir
,
3696 struct dentry
*dentry
)
3698 struct btrfs_trans_handle
*trans
;
3699 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3700 struct btrfs_path
*path
;
3701 struct btrfs_dir_item
*di
;
3702 struct inode
*inode
= dentry
->d_inode
;
3707 u64 ino
= btrfs_ino(inode
);
3708 u64 dir_ino
= btrfs_ino(dir
);
3711 * 1 for the possible orphan item
3712 * 1 for the dir item
3713 * 1 for the dir index
3714 * 1 for the inode ref
3717 trans
= btrfs_start_transaction(root
, 5);
3718 if (!IS_ERR(trans
) || PTR_ERR(trans
) != -ENOSPC
)
3721 if (ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
3722 return ERR_PTR(-ENOSPC
);
3724 /* check if there is someone else holds reference */
3725 if (S_ISDIR(inode
->i_mode
) && atomic_read(&inode
->i_count
) > 1)
3726 return ERR_PTR(-ENOSPC
);
3728 if (atomic_read(&inode
->i_count
) > 2)
3729 return ERR_PTR(-ENOSPC
);
3731 if (xchg(&root
->fs_info
->enospc_unlink
, 1))
3732 return ERR_PTR(-ENOSPC
);
3734 path
= btrfs_alloc_path();
3736 root
->fs_info
->enospc_unlink
= 0;
3737 return ERR_PTR(-ENOMEM
);
3740 /* 1 for the orphan item */
3741 trans
= btrfs_start_transaction(root
, 1);
3742 if (IS_ERR(trans
)) {
3743 btrfs_free_path(path
);
3744 root
->fs_info
->enospc_unlink
= 0;
3748 path
->skip_locking
= 1;
3749 path
->search_commit_root
= 1;
3751 ret
= btrfs_lookup_inode(trans
, root
, path
,
3752 &BTRFS_I(dir
)->location
, 0);
3758 if (check_path_shared(root
, path
))
3763 btrfs_release_path(path
);
3765 ret
= btrfs_lookup_inode(trans
, root
, path
,
3766 &BTRFS_I(inode
)->location
, 0);
3772 if (check_path_shared(root
, path
))
3777 btrfs_release_path(path
);
3779 if (ret
== 0 && S_ISREG(inode
->i_mode
)) {
3780 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
3786 BUG_ON(ret
== 0); /* Corruption */
3787 if (check_path_shared(root
, path
))
3789 btrfs_release_path(path
);
3797 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3798 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3804 if (check_path_shared(root
, path
))
3810 btrfs_release_path(path
);
3812 ret
= btrfs_get_inode_ref_index(trans
, root
, path
, dentry
->d_name
.name
,
3813 dentry
->d_name
.len
, ino
, dir_ino
, 0,
3820 if (check_path_shared(root
, path
))
3823 btrfs_release_path(path
);
3826 * This is a commit root search, if we can lookup inode item and other
3827 * relative items in the commit root, it means the transaction of
3828 * dir/file creation has been committed, and the dir index item that we
3829 * delay to insert has also been inserted into the commit root. So
3830 * we needn't worry about the delayed insertion of the dir index item
3833 di
= btrfs_lookup_dir_index_item(trans
, root
, path
, dir_ino
, index
,
3834 dentry
->d_name
.name
, dentry
->d_name
.len
, 0);
3839 BUG_ON(ret
== -ENOENT
);
3840 if (check_path_shared(root
, path
))
3845 btrfs_free_path(path
);
3846 /* Migrate the orphan reservation over */
3848 err
= btrfs_block_rsv_migrate(trans
->block_rsv
,
3849 &root
->fs_info
->global_block_rsv
,
3850 trans
->bytes_reserved
);
3853 btrfs_end_transaction(trans
, root
);
3854 root
->fs_info
->enospc_unlink
= 0;
3855 return ERR_PTR(err
);
3858 trans
->block_rsv
= &root
->fs_info
->global_block_rsv
;
3862 static void __unlink_end_trans(struct btrfs_trans_handle
*trans
,
3863 struct btrfs_root
*root
)
3865 if (trans
->block_rsv
->type
== BTRFS_BLOCK_RSV_GLOBAL
) {
3866 btrfs_block_rsv_release(root
, trans
->block_rsv
,
3867 trans
->bytes_reserved
);
3868 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
3869 BUG_ON(!root
->fs_info
->enospc_unlink
);
3870 root
->fs_info
->enospc_unlink
= 0;
3872 btrfs_end_transaction(trans
, root
);
3875 static int btrfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3877 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3878 struct btrfs_trans_handle
*trans
;
3879 struct inode
*inode
= dentry
->d_inode
;
3882 trans
= __unlink_start_trans(dir
, dentry
);
3884 return PTR_ERR(trans
);
3886 btrfs_record_unlink_dir(trans
, dir
, dentry
->d_inode
, 0);
3888 ret
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
3889 dentry
->d_name
.name
, dentry
->d_name
.len
);
3893 if (inode
->i_nlink
== 0) {
3894 ret
= btrfs_orphan_add(trans
, inode
);
3900 __unlink_end_trans(trans
, root
);
3901 btrfs_btree_balance_dirty(root
);
3905 int btrfs_unlink_subvol(struct btrfs_trans_handle
*trans
,
3906 struct btrfs_root
*root
,
3907 struct inode
*dir
, u64 objectid
,
3908 const char *name
, int name_len
)
3910 struct btrfs_path
*path
;
3911 struct extent_buffer
*leaf
;
3912 struct btrfs_dir_item
*di
;
3913 struct btrfs_key key
;
3916 u64 dir_ino
= btrfs_ino(dir
);
3918 path
= btrfs_alloc_path();
3922 di
= btrfs_lookup_dir_item(trans
, root
, path
, dir_ino
,
3923 name
, name_len
, -1);
3924 if (IS_ERR_OR_NULL(di
)) {
3932 leaf
= path
->nodes
[0];
3933 btrfs_dir_item_key_to_cpu(leaf
, di
, &key
);
3934 WARN_ON(key
.type
!= BTRFS_ROOT_ITEM_KEY
|| key
.objectid
!= objectid
);
3935 ret
= btrfs_delete_one_dir_name(trans
, root
, path
, di
);
3937 btrfs_abort_transaction(trans
, root
, ret
);
3940 btrfs_release_path(path
);
3942 ret
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
3943 objectid
, root
->root_key
.objectid
,
3944 dir_ino
, &index
, name
, name_len
);
3946 if (ret
!= -ENOENT
) {
3947 btrfs_abort_transaction(trans
, root
, ret
);
3950 di
= btrfs_search_dir_index_item(root
, path
, dir_ino
,
3952 if (IS_ERR_OR_NULL(di
)) {
3957 btrfs_abort_transaction(trans
, root
, ret
);
3961 leaf
= path
->nodes
[0];
3962 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3963 btrfs_release_path(path
);
3966 btrfs_release_path(path
);
3968 ret
= btrfs_delete_delayed_dir_index(trans
, root
, dir
, index
);
3970 btrfs_abort_transaction(trans
, root
, ret
);
3974 btrfs_i_size_write(dir
, dir
->i_size
- name_len
* 2);
3975 inode_inc_iversion(dir
);
3976 dir
->i_mtime
= dir
->i_ctime
= CURRENT_TIME
;
3977 ret
= btrfs_update_inode_fallback(trans
, root
, dir
);
3979 btrfs_abort_transaction(trans
, root
, ret
);
3981 btrfs_free_path(path
);
3985 static int btrfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3987 struct inode
*inode
= dentry
->d_inode
;
3989 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
3990 struct btrfs_trans_handle
*trans
;
3992 if (inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
3994 if (btrfs_ino(inode
) == BTRFS_FIRST_FREE_OBJECTID
)
3997 trans
= __unlink_start_trans(dir
, dentry
);
3999 return PTR_ERR(trans
);
4001 if (unlikely(btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
4002 err
= btrfs_unlink_subvol(trans
, root
, dir
,
4003 BTRFS_I(inode
)->location
.objectid
,
4004 dentry
->d_name
.name
,
4005 dentry
->d_name
.len
);
4009 err
= btrfs_orphan_add(trans
, inode
);
4013 /* now the directory is empty */
4014 err
= btrfs_unlink_inode(trans
, root
, dir
, dentry
->d_inode
,
4015 dentry
->d_name
.name
, dentry
->d_name
.len
);
4017 btrfs_i_size_write(inode
, 0);
4019 __unlink_end_trans(trans
, root
);
4020 btrfs_btree_balance_dirty(root
);
4026 * this can truncate away extent items, csum items and directory items.
4027 * It starts at a high offset and removes keys until it can't find
4028 * any higher than new_size
4030 * csum items that cross the new i_size are truncated to the new size
4033 * min_type is the minimum key type to truncate down to. If set to 0, this
4034 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4036 int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans
,
4037 struct btrfs_root
*root
,
4038 struct inode
*inode
,
4039 u64 new_size
, u32 min_type
)
4041 struct btrfs_path
*path
;
4042 struct extent_buffer
*leaf
;
4043 struct btrfs_file_extent_item
*fi
;
4044 struct btrfs_key key
;
4045 struct btrfs_key found_key
;
4046 u64 extent_start
= 0;
4047 u64 extent_num_bytes
= 0;
4048 u64 extent_offset
= 0;
4050 u32 found_type
= (u8
)-1;
4053 int pending_del_nr
= 0;
4054 int pending_del_slot
= 0;
4055 int extent_type
= -1;
4058 u64 ino
= btrfs_ino(inode
);
4060 BUG_ON(new_size
> 0 && min_type
!= BTRFS_EXTENT_DATA_KEY
);
4062 path
= btrfs_alloc_path();
4068 * We want to drop from the next block forward in case this new size is
4069 * not block aligned since we will be keeping the last block of the
4070 * extent just the way it is.
4072 if (root
->ref_cows
|| root
== root
->fs_info
->tree_root
)
4073 btrfs_drop_extent_cache(inode
, ALIGN(new_size
,
4074 root
->sectorsize
), (u64
)-1, 0);
4077 * This function is also used to drop the items in the log tree before
4078 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4079 * it is used to drop the loged items. So we shouldn't kill the delayed
4082 if (min_type
== 0 && root
== BTRFS_I(inode
)->root
)
4083 btrfs_kill_delayed_inode_items(inode
);
4086 key
.offset
= (u64
)-1;
4090 path
->leave_spinning
= 1;
4091 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
4098 /* there are no items in the tree for us to truncate, we're
4101 if (path
->slots
[0] == 0)
4108 leaf
= path
->nodes
[0];
4109 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4110 found_type
= btrfs_key_type(&found_key
);
4112 if (found_key
.objectid
!= ino
)
4115 if (found_type
< min_type
)
4118 item_end
= found_key
.offset
;
4119 if (found_type
== BTRFS_EXTENT_DATA_KEY
) {
4120 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4121 struct btrfs_file_extent_item
);
4122 extent_type
= btrfs_file_extent_type(leaf
, fi
);
4123 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4125 btrfs_file_extent_num_bytes(leaf
, fi
);
4126 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4127 item_end
+= btrfs_file_extent_inline_len(leaf
,
4132 if (found_type
> min_type
) {
4135 if (item_end
< new_size
)
4137 if (found_key
.offset
>= new_size
)
4143 /* FIXME, shrink the extent if the ref count is only 1 */
4144 if (found_type
!= BTRFS_EXTENT_DATA_KEY
)
4147 if (extent_type
!= BTRFS_FILE_EXTENT_INLINE
) {
4149 extent_start
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
4151 u64 orig_num_bytes
=
4152 btrfs_file_extent_num_bytes(leaf
, fi
);
4153 extent_num_bytes
= ALIGN(new_size
-
4156 btrfs_set_file_extent_num_bytes(leaf
, fi
,
4158 num_dec
= (orig_num_bytes
-
4160 if (root
->ref_cows
&& extent_start
!= 0)
4161 inode_sub_bytes(inode
, num_dec
);
4162 btrfs_mark_buffer_dirty(leaf
);
4165 btrfs_file_extent_disk_num_bytes(leaf
,
4167 extent_offset
= found_key
.offset
-
4168 btrfs_file_extent_offset(leaf
, fi
);
4170 /* FIXME blocksize != 4096 */
4171 num_dec
= btrfs_file_extent_num_bytes(leaf
, fi
);
4172 if (extent_start
!= 0) {
4175 inode_sub_bytes(inode
, num_dec
);
4178 } else if (extent_type
== BTRFS_FILE_EXTENT_INLINE
) {
4180 * we can't truncate inline items that have had
4184 btrfs_file_extent_compression(leaf
, fi
) == 0 &&
4185 btrfs_file_extent_encryption(leaf
, fi
) == 0 &&
4186 btrfs_file_extent_other_encoding(leaf
, fi
) == 0) {
4187 u32 size
= new_size
- found_key
.offset
;
4189 if (root
->ref_cows
) {
4190 inode_sub_bytes(inode
, item_end
+ 1 -
4194 btrfs_file_extent_calc_inline_size(size
);
4195 btrfs_truncate_item(root
, path
, size
, 1);
4196 } else if (root
->ref_cows
) {
4197 inode_sub_bytes(inode
, item_end
+ 1 -
4203 if (!pending_del_nr
) {
4204 /* no pending yet, add ourselves */
4205 pending_del_slot
= path
->slots
[0];
4207 } else if (pending_del_nr
&&
4208 path
->slots
[0] + 1 == pending_del_slot
) {
4209 /* hop on the pending chunk */
4211 pending_del_slot
= path
->slots
[0];
4218 if (found_extent
&& (root
->ref_cows
||
4219 root
== root
->fs_info
->tree_root
)) {
4220 btrfs_set_path_blocking(path
);
4221 ret
= btrfs_free_extent(trans
, root
, extent_start
,
4222 extent_num_bytes
, 0,
4223 btrfs_header_owner(leaf
),
4224 ino
, extent_offset
, 0);
4228 if (found_type
== BTRFS_INODE_ITEM_KEY
)
4231 if (path
->slots
[0] == 0 ||
4232 path
->slots
[0] != pending_del_slot
) {
4233 if (pending_del_nr
) {
4234 ret
= btrfs_del_items(trans
, root
, path
,
4238 btrfs_abort_transaction(trans
,
4244 btrfs_release_path(path
);
4251 if (pending_del_nr
) {
4252 ret
= btrfs_del_items(trans
, root
, path
, pending_del_slot
,
4255 btrfs_abort_transaction(trans
, root
, ret
);
4258 btrfs_free_path(path
);
4263 * btrfs_truncate_page - read, zero a chunk and write a page
4264 * @inode - inode that we're zeroing
4265 * @from - the offset to start zeroing
4266 * @len - the length to zero, 0 to zero the entire range respective to the
4268 * @front - zero up to the offset instead of from the offset on
4270 * This will find the page for the "from" offset and cow the page and zero the
4271 * part we want to zero. This is used with truncate and hole punching.
4273 int btrfs_truncate_page(struct inode
*inode
, loff_t from
, loff_t len
,
4276 struct address_space
*mapping
= inode
->i_mapping
;
4277 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4278 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4279 struct btrfs_ordered_extent
*ordered
;
4280 struct extent_state
*cached_state
= NULL
;
4282 u32 blocksize
= root
->sectorsize
;
4283 pgoff_t index
= from
>> PAGE_CACHE_SHIFT
;
4284 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
4286 gfp_t mask
= btrfs_alloc_write_mask(mapping
);
4291 if ((offset
& (blocksize
- 1)) == 0 &&
4292 (!len
|| ((len
& (blocksize
- 1)) == 0)))
4294 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
4299 page
= find_or_create_page(mapping
, index
, mask
);
4301 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4306 page_start
= page_offset(page
);
4307 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
4309 if (!PageUptodate(page
)) {
4310 ret
= btrfs_readpage(NULL
, page
);
4312 if (page
->mapping
!= mapping
) {
4314 page_cache_release(page
);
4317 if (!PageUptodate(page
)) {
4322 wait_on_page_writeback(page
);
4324 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
4325 set_page_extent_mapped(page
);
4327 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
4329 unlock_extent_cached(io_tree
, page_start
, page_end
,
4330 &cached_state
, GFP_NOFS
);
4332 page_cache_release(page
);
4333 btrfs_start_ordered_extent(inode
, ordered
, 1);
4334 btrfs_put_ordered_extent(ordered
);
4338 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
4339 EXTENT_DIRTY
| EXTENT_DELALLOC
|
4340 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
4341 0, 0, &cached_state
, GFP_NOFS
);
4343 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
4346 unlock_extent_cached(io_tree
, page_start
, page_end
,
4347 &cached_state
, GFP_NOFS
);
4351 if (offset
!= PAGE_CACHE_SIZE
) {
4353 len
= PAGE_CACHE_SIZE
- offset
;
4356 memset(kaddr
, 0, offset
);
4358 memset(kaddr
+ offset
, 0, len
);
4359 flush_dcache_page(page
);
4362 ClearPageChecked(page
);
4363 set_page_dirty(page
);
4364 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
,
4369 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
4371 page_cache_release(page
);
4377 * This function puts in dummy file extents for the area we're creating a hole
4378 * for. So if we are truncating this file to a larger size we need to insert
4379 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4380 * the range between oldsize and size
4382 int btrfs_cont_expand(struct inode
*inode
, loff_t oldsize
, loff_t size
)
4384 struct btrfs_trans_handle
*trans
;
4385 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4386 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
4387 struct extent_map
*em
= NULL
;
4388 struct extent_state
*cached_state
= NULL
;
4389 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
4390 u64 hole_start
= ALIGN(oldsize
, root
->sectorsize
);
4391 u64 block_end
= ALIGN(size
, root
->sectorsize
);
4397 if (size
<= hole_start
)
4401 struct btrfs_ordered_extent
*ordered
;
4402 btrfs_wait_ordered_range(inode
, hole_start
,
4403 block_end
- hole_start
);
4404 lock_extent_bits(io_tree
, hole_start
, block_end
- 1, 0,
4406 ordered
= btrfs_lookup_ordered_extent(inode
, hole_start
);
4409 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1,
4410 &cached_state
, GFP_NOFS
);
4411 btrfs_put_ordered_extent(ordered
);
4414 cur_offset
= hole_start
;
4416 em
= btrfs_get_extent(inode
, NULL
, 0, cur_offset
,
4417 block_end
- cur_offset
, 0);
4423 last_byte
= min(extent_map_end(em
), block_end
);
4424 last_byte
= ALIGN(last_byte
, root
->sectorsize
);
4425 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
)) {
4426 struct extent_map
*hole_em
;
4427 hole_size
= last_byte
- cur_offset
;
4429 trans
= btrfs_start_transaction(root
, 3);
4430 if (IS_ERR(trans
)) {
4431 err
= PTR_ERR(trans
);
4435 err
= btrfs_drop_extents(trans
, root
, inode
,
4437 cur_offset
+ hole_size
, 1);
4439 btrfs_abort_transaction(trans
, root
, err
);
4440 btrfs_end_transaction(trans
, root
);
4444 err
= btrfs_insert_file_extent(trans
, root
,
4445 btrfs_ino(inode
), cur_offset
, 0,
4446 0, hole_size
, 0, hole_size
,
4449 btrfs_abort_transaction(trans
, root
, err
);
4450 btrfs_end_transaction(trans
, root
);
4454 btrfs_drop_extent_cache(inode
, cur_offset
,
4455 cur_offset
+ hole_size
- 1, 0);
4456 hole_em
= alloc_extent_map();
4458 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
4459 &BTRFS_I(inode
)->runtime_flags
);
4462 hole_em
->start
= cur_offset
;
4463 hole_em
->len
= hole_size
;
4464 hole_em
->orig_start
= cur_offset
;
4466 hole_em
->block_start
= EXTENT_MAP_HOLE
;
4467 hole_em
->block_len
= 0;
4468 hole_em
->orig_block_len
= 0;
4469 hole_em
->ram_bytes
= hole_size
;
4470 hole_em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
4471 hole_em
->compress_type
= BTRFS_COMPRESS_NONE
;
4472 hole_em
->generation
= trans
->transid
;
4475 write_lock(&em_tree
->lock
);
4476 err
= add_extent_mapping(em_tree
, hole_em
, 1);
4477 write_unlock(&em_tree
->lock
);
4480 btrfs_drop_extent_cache(inode
, cur_offset
,
4484 free_extent_map(hole_em
);
4486 btrfs_update_inode(trans
, root
, inode
);
4487 btrfs_end_transaction(trans
, root
);
4489 free_extent_map(em
);
4491 cur_offset
= last_byte
;
4492 if (cur_offset
>= block_end
)
4496 free_extent_map(em
);
4497 unlock_extent_cached(io_tree
, hole_start
, block_end
- 1, &cached_state
,
4502 static int btrfs_setsize(struct inode
*inode
, struct iattr
*attr
)
4504 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4505 struct btrfs_trans_handle
*trans
;
4506 loff_t oldsize
= i_size_read(inode
);
4507 loff_t newsize
= attr
->ia_size
;
4508 int mask
= attr
->ia_valid
;
4511 if (newsize
== oldsize
)
4515 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4516 * special case where we need to update the times despite not having
4517 * these flags set. For all other operations the VFS set these flags
4518 * explicitly if it wants a timestamp update.
4520 if (newsize
!= oldsize
&& (!(mask
& (ATTR_CTIME
| ATTR_MTIME
))))
4521 inode
->i_ctime
= inode
->i_mtime
= current_fs_time(inode
->i_sb
);
4523 if (newsize
> oldsize
) {
4524 truncate_pagecache(inode
, oldsize
, newsize
);
4525 ret
= btrfs_cont_expand(inode
, oldsize
, newsize
);
4529 trans
= btrfs_start_transaction(root
, 1);
4531 return PTR_ERR(trans
);
4533 i_size_write(inode
, newsize
);
4534 btrfs_ordered_update_i_size(inode
, i_size_read(inode
), NULL
);
4535 ret
= btrfs_update_inode(trans
, root
, inode
);
4536 btrfs_end_transaction(trans
, root
);
4540 * We're truncating a file that used to have good data down to
4541 * zero. Make sure it gets into the ordered flush list so that
4542 * any new writes get down to disk quickly.
4545 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
4546 &BTRFS_I(inode
)->runtime_flags
);
4549 * 1 for the orphan item we're going to add
4550 * 1 for the orphan item deletion.
4552 trans
= btrfs_start_transaction(root
, 2);
4554 return PTR_ERR(trans
);
4557 * We need to do this in case we fail at _any_ point during the
4558 * actual truncate. Once we do the truncate_setsize we could
4559 * invalidate pages which forces any outstanding ordered io to
4560 * be instantly completed which will give us extents that need
4561 * to be truncated. If we fail to get an orphan inode down we
4562 * could have left over extents that were never meant to live,
4563 * so we need to garuntee from this point on that everything
4564 * will be consistent.
4566 ret
= btrfs_orphan_add(trans
, inode
);
4567 btrfs_end_transaction(trans
, root
);
4571 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4572 truncate_setsize(inode
, newsize
);
4574 /* Disable nonlocked read DIO to avoid the end less truncate */
4575 btrfs_inode_block_unlocked_dio(inode
);
4576 inode_dio_wait(inode
);
4577 btrfs_inode_resume_unlocked_dio(inode
);
4579 ret
= btrfs_truncate(inode
);
4580 if (ret
&& inode
->i_nlink
)
4581 btrfs_orphan_del(NULL
, inode
);
4587 static int btrfs_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4589 struct inode
*inode
= dentry
->d_inode
;
4590 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4593 if (btrfs_root_readonly(root
))
4596 err
= inode_change_ok(inode
, attr
);
4600 if (S_ISREG(inode
->i_mode
) && (attr
->ia_valid
& ATTR_SIZE
)) {
4601 err
= btrfs_setsize(inode
, attr
);
4606 if (attr
->ia_valid
) {
4607 setattr_copy(inode
, attr
);
4608 inode_inc_iversion(inode
);
4609 err
= btrfs_dirty_inode(inode
);
4611 if (!err
&& attr
->ia_valid
& ATTR_MODE
)
4612 err
= btrfs_acl_chmod(inode
);
4618 void btrfs_evict_inode(struct inode
*inode
)
4620 struct btrfs_trans_handle
*trans
;
4621 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4622 struct btrfs_block_rsv
*rsv
, *global_rsv
;
4623 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
4626 trace_btrfs_inode_evict(inode
);
4628 truncate_inode_pages(&inode
->i_data
, 0);
4629 if (inode
->i_nlink
&& (btrfs_root_refs(&root
->root_item
) != 0 ||
4630 btrfs_is_free_space_inode(inode
)))
4633 if (is_bad_inode(inode
)) {
4634 btrfs_orphan_del(NULL
, inode
);
4637 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4638 btrfs_wait_ordered_range(inode
, 0, (u64
)-1);
4640 if (root
->fs_info
->log_root_recovering
) {
4641 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
4642 &BTRFS_I(inode
)->runtime_flags
));
4646 if (inode
->i_nlink
> 0) {
4647 BUG_ON(btrfs_root_refs(&root
->root_item
) != 0);
4651 ret
= btrfs_commit_inode_delayed_inode(inode
);
4653 btrfs_orphan_del(NULL
, inode
);
4657 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
4659 btrfs_orphan_del(NULL
, inode
);
4662 rsv
->size
= min_size
;
4664 global_rsv
= &root
->fs_info
->global_block_rsv
;
4666 btrfs_i_size_write(inode
, 0);
4669 * This is a bit simpler than btrfs_truncate since we've already
4670 * reserved our space for our orphan item in the unlink, so we just
4671 * need to reserve some slack space in case we add bytes and update
4672 * inode item when doing the truncate.
4675 ret
= btrfs_block_rsv_refill(root
, rsv
, min_size
,
4676 BTRFS_RESERVE_FLUSH_LIMIT
);
4679 * Try and steal from the global reserve since we will
4680 * likely not use this space anyway, we want to try as
4681 * hard as possible to get this to work.
4684 ret
= btrfs_block_rsv_migrate(global_rsv
, rsv
, min_size
);
4687 btrfs_warn(root
->fs_info
,
4688 "Could not get space for a delete, will truncate on mount %d",
4690 btrfs_orphan_del(NULL
, inode
);
4691 btrfs_free_block_rsv(root
, rsv
);
4695 trans
= btrfs_join_transaction(root
);
4696 if (IS_ERR(trans
)) {
4697 btrfs_orphan_del(NULL
, inode
);
4698 btrfs_free_block_rsv(root
, rsv
);
4702 trans
->block_rsv
= rsv
;
4704 ret
= btrfs_truncate_inode_items(trans
, root
, inode
, 0, 0);
4708 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4709 btrfs_end_transaction(trans
, root
);
4711 btrfs_btree_balance_dirty(root
);
4714 btrfs_free_block_rsv(root
, rsv
);
4717 trans
->block_rsv
= root
->orphan_block_rsv
;
4718 ret
= btrfs_orphan_del(trans
, inode
);
4722 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
4723 if (!(root
== root
->fs_info
->tree_root
||
4724 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
))
4725 btrfs_return_ino(root
, btrfs_ino(inode
));
4727 btrfs_end_transaction(trans
, root
);
4728 btrfs_btree_balance_dirty(root
);
4730 btrfs_remove_delayed_node(inode
);
4736 * this returns the key found in the dir entry in the location pointer.
4737 * If no dir entries were found, location->objectid is 0.
4739 static int btrfs_inode_by_name(struct inode
*dir
, struct dentry
*dentry
,
4740 struct btrfs_key
*location
)
4742 const char *name
= dentry
->d_name
.name
;
4743 int namelen
= dentry
->d_name
.len
;
4744 struct btrfs_dir_item
*di
;
4745 struct btrfs_path
*path
;
4746 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
4749 path
= btrfs_alloc_path();
4753 di
= btrfs_lookup_dir_item(NULL
, root
, path
, btrfs_ino(dir
), name
,
4758 if (IS_ERR_OR_NULL(di
))
4761 btrfs_dir_item_key_to_cpu(path
->nodes
[0], di
, location
);
4763 btrfs_free_path(path
);
4766 location
->objectid
= 0;
4771 * when we hit a tree root in a directory, the btrfs part of the inode
4772 * needs to be changed to reflect the root directory of the tree root. This
4773 * is kind of like crossing a mount point.
4775 static int fixup_tree_root_location(struct btrfs_root
*root
,
4777 struct dentry
*dentry
,
4778 struct btrfs_key
*location
,
4779 struct btrfs_root
**sub_root
)
4781 struct btrfs_path
*path
;
4782 struct btrfs_root
*new_root
;
4783 struct btrfs_root_ref
*ref
;
4784 struct extent_buffer
*leaf
;
4788 path
= btrfs_alloc_path();
4795 ret
= btrfs_find_root_ref(root
->fs_info
->tree_root
, path
,
4796 BTRFS_I(dir
)->root
->root_key
.objectid
,
4797 location
->objectid
);
4804 leaf
= path
->nodes
[0];
4805 ref
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_root_ref
);
4806 if (btrfs_root_ref_dirid(leaf
, ref
) != btrfs_ino(dir
) ||
4807 btrfs_root_ref_name_len(leaf
, ref
) != dentry
->d_name
.len
)
4810 ret
= memcmp_extent_buffer(leaf
, dentry
->d_name
.name
,
4811 (unsigned long)(ref
+ 1),
4812 dentry
->d_name
.len
);
4816 btrfs_release_path(path
);
4818 new_root
= btrfs_read_fs_root_no_name(root
->fs_info
, location
);
4819 if (IS_ERR(new_root
)) {
4820 err
= PTR_ERR(new_root
);
4824 if (btrfs_root_refs(&new_root
->root_item
) == 0) {
4829 *sub_root
= new_root
;
4830 location
->objectid
= btrfs_root_dirid(&new_root
->root_item
);
4831 location
->type
= BTRFS_INODE_ITEM_KEY
;
4832 location
->offset
= 0;
4835 btrfs_free_path(path
);
4839 static void inode_tree_add(struct inode
*inode
)
4841 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4842 struct btrfs_inode
*entry
;
4844 struct rb_node
*parent
;
4845 u64 ino
= btrfs_ino(inode
);
4847 if (inode_unhashed(inode
))
4851 spin_lock(&root
->inode_lock
);
4852 p
= &root
->inode_tree
.rb_node
;
4855 entry
= rb_entry(parent
, struct btrfs_inode
, rb_node
);
4857 if (ino
< btrfs_ino(&entry
->vfs_inode
))
4858 p
= &parent
->rb_left
;
4859 else if (ino
> btrfs_ino(&entry
->vfs_inode
))
4860 p
= &parent
->rb_right
;
4862 WARN_ON(!(entry
->vfs_inode
.i_state
&
4863 (I_WILL_FREE
| I_FREEING
)));
4864 rb_erase(parent
, &root
->inode_tree
);
4865 RB_CLEAR_NODE(parent
);
4866 spin_unlock(&root
->inode_lock
);
4870 rb_link_node(&BTRFS_I(inode
)->rb_node
, parent
, p
);
4871 rb_insert_color(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4872 spin_unlock(&root
->inode_lock
);
4875 static void inode_tree_del(struct inode
*inode
)
4877 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4880 spin_lock(&root
->inode_lock
);
4881 if (!RB_EMPTY_NODE(&BTRFS_I(inode
)->rb_node
)) {
4882 rb_erase(&BTRFS_I(inode
)->rb_node
, &root
->inode_tree
);
4883 RB_CLEAR_NODE(&BTRFS_I(inode
)->rb_node
);
4884 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4886 spin_unlock(&root
->inode_lock
);
4889 * Free space cache has inodes in the tree root, but the tree root has a
4890 * root_refs of 0, so this could end up dropping the tree root as a
4891 * snapshot, so we need the extra !root->fs_info->tree_root check to
4892 * make sure we don't drop it.
4894 if (empty
&& btrfs_root_refs(&root
->root_item
) == 0 &&
4895 root
!= root
->fs_info
->tree_root
) {
4896 synchronize_srcu(&root
->fs_info
->subvol_srcu
);
4897 spin_lock(&root
->inode_lock
);
4898 empty
= RB_EMPTY_ROOT(&root
->inode_tree
);
4899 spin_unlock(&root
->inode_lock
);
4901 btrfs_add_dead_root(root
);
4905 void btrfs_invalidate_inodes(struct btrfs_root
*root
)
4907 struct rb_node
*node
;
4908 struct rb_node
*prev
;
4909 struct btrfs_inode
*entry
;
4910 struct inode
*inode
;
4913 WARN_ON(btrfs_root_refs(&root
->root_item
) != 0);
4915 spin_lock(&root
->inode_lock
);
4917 node
= root
->inode_tree
.rb_node
;
4921 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4923 if (objectid
< btrfs_ino(&entry
->vfs_inode
))
4924 node
= node
->rb_left
;
4925 else if (objectid
> btrfs_ino(&entry
->vfs_inode
))
4926 node
= node
->rb_right
;
4932 entry
= rb_entry(prev
, struct btrfs_inode
, rb_node
);
4933 if (objectid
<= btrfs_ino(&entry
->vfs_inode
)) {
4937 prev
= rb_next(prev
);
4941 entry
= rb_entry(node
, struct btrfs_inode
, rb_node
);
4942 objectid
= btrfs_ino(&entry
->vfs_inode
) + 1;
4943 inode
= igrab(&entry
->vfs_inode
);
4945 spin_unlock(&root
->inode_lock
);
4946 if (atomic_read(&inode
->i_count
) > 1)
4947 d_prune_aliases(inode
);
4949 * btrfs_drop_inode will have it removed from
4950 * the inode cache when its usage count
4955 spin_lock(&root
->inode_lock
);
4959 if (cond_resched_lock(&root
->inode_lock
))
4962 node
= rb_next(node
);
4964 spin_unlock(&root
->inode_lock
);
4967 static int btrfs_init_locked_inode(struct inode
*inode
, void *p
)
4969 struct btrfs_iget_args
*args
= p
;
4970 inode
->i_ino
= args
->ino
;
4971 BTRFS_I(inode
)->root
= args
->root
;
4975 static int btrfs_find_actor(struct inode
*inode
, void *opaque
)
4977 struct btrfs_iget_args
*args
= opaque
;
4978 return args
->ino
== btrfs_ino(inode
) &&
4979 args
->root
== BTRFS_I(inode
)->root
;
4982 static struct inode
*btrfs_iget_locked(struct super_block
*s
,
4984 struct btrfs_root
*root
)
4986 struct inode
*inode
;
4987 struct btrfs_iget_args args
;
4988 args
.ino
= objectid
;
4991 inode
= iget5_locked(s
, objectid
, btrfs_find_actor
,
4992 btrfs_init_locked_inode
,
4997 /* Get an inode object given its location and corresponding root.
4998 * Returns in *is_new if the inode was read from disk
5000 struct inode
*btrfs_iget(struct super_block
*s
, struct btrfs_key
*location
,
5001 struct btrfs_root
*root
, int *new)
5003 struct inode
*inode
;
5005 inode
= btrfs_iget_locked(s
, location
->objectid
, root
);
5007 return ERR_PTR(-ENOMEM
);
5009 if (inode
->i_state
& I_NEW
) {
5010 BTRFS_I(inode
)->root
= root
;
5011 memcpy(&BTRFS_I(inode
)->location
, location
, sizeof(*location
));
5012 btrfs_read_locked_inode(inode
);
5013 if (!is_bad_inode(inode
)) {
5014 inode_tree_add(inode
);
5015 unlock_new_inode(inode
);
5019 unlock_new_inode(inode
);
5021 inode
= ERR_PTR(-ESTALE
);
5028 static struct inode
*new_simple_dir(struct super_block
*s
,
5029 struct btrfs_key
*key
,
5030 struct btrfs_root
*root
)
5032 struct inode
*inode
= new_inode(s
);
5035 return ERR_PTR(-ENOMEM
);
5037 BTRFS_I(inode
)->root
= root
;
5038 memcpy(&BTRFS_I(inode
)->location
, key
, sizeof(*key
));
5039 set_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
);
5041 inode
->i_ino
= BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
;
5042 inode
->i_op
= &btrfs_dir_ro_inode_operations
;
5043 inode
->i_fop
= &simple_dir_operations
;
5044 inode
->i_mode
= S_IFDIR
| S_IRUGO
| S_IWUSR
| S_IXUGO
;
5045 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
5050 struct inode
*btrfs_lookup_dentry(struct inode
*dir
, struct dentry
*dentry
)
5052 struct inode
*inode
;
5053 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5054 struct btrfs_root
*sub_root
= root
;
5055 struct btrfs_key location
;
5059 if (dentry
->d_name
.len
> BTRFS_NAME_LEN
)
5060 return ERR_PTR(-ENAMETOOLONG
);
5062 ret
= btrfs_inode_by_name(dir
, dentry
, &location
);
5064 return ERR_PTR(ret
);
5066 if (location
.objectid
== 0)
5069 if (location
.type
== BTRFS_INODE_ITEM_KEY
) {
5070 inode
= btrfs_iget(dir
->i_sb
, &location
, root
, NULL
);
5074 BUG_ON(location
.type
!= BTRFS_ROOT_ITEM_KEY
);
5076 index
= srcu_read_lock(&root
->fs_info
->subvol_srcu
);
5077 ret
= fixup_tree_root_location(root
, dir
, dentry
,
5078 &location
, &sub_root
);
5081 inode
= ERR_PTR(ret
);
5083 inode
= new_simple_dir(dir
->i_sb
, &location
, sub_root
);
5085 inode
= btrfs_iget(dir
->i_sb
, &location
, sub_root
, NULL
);
5087 srcu_read_unlock(&root
->fs_info
->subvol_srcu
, index
);
5089 if (!IS_ERR(inode
) && root
!= sub_root
) {
5090 down_read(&root
->fs_info
->cleanup_work_sem
);
5091 if (!(inode
->i_sb
->s_flags
& MS_RDONLY
))
5092 ret
= btrfs_orphan_cleanup(sub_root
);
5093 up_read(&root
->fs_info
->cleanup_work_sem
);
5095 inode
= ERR_PTR(ret
);
5101 static int btrfs_dentry_delete(const struct dentry
*dentry
)
5103 struct btrfs_root
*root
;
5104 struct inode
*inode
= dentry
->d_inode
;
5106 if (!inode
&& !IS_ROOT(dentry
))
5107 inode
= dentry
->d_parent
->d_inode
;
5110 root
= BTRFS_I(inode
)->root
;
5111 if (btrfs_root_refs(&root
->root_item
) == 0)
5114 if (btrfs_ino(inode
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
5120 static void btrfs_dentry_release(struct dentry
*dentry
)
5122 if (dentry
->d_fsdata
)
5123 kfree(dentry
->d_fsdata
);
5126 static struct dentry
*btrfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
5131 ret
= d_splice_alias(btrfs_lookup_dentry(dir
, dentry
), dentry
);
5135 unsigned char btrfs_filetype_table
[] = {
5136 DT_UNKNOWN
, DT_REG
, DT_DIR
, DT_CHR
, DT_BLK
, DT_FIFO
, DT_SOCK
, DT_LNK
5139 static int btrfs_real_readdir(struct file
*filp
, void *dirent
,
5142 struct inode
*inode
= file_inode(filp
);
5143 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5144 struct btrfs_item
*item
;
5145 struct btrfs_dir_item
*di
;
5146 struct btrfs_key key
;
5147 struct btrfs_key found_key
;
5148 struct btrfs_path
*path
;
5149 struct list_head ins_list
;
5150 struct list_head del_list
;
5152 struct extent_buffer
*leaf
;
5154 unsigned char d_type
;
5159 int key_type
= BTRFS_DIR_INDEX_KEY
;
5163 int is_curr
= 0; /* filp->f_pos points to the current index? */
5165 /* FIXME, use a real flag for deciding about the key type */
5166 if (root
->fs_info
->tree_root
== root
)
5167 key_type
= BTRFS_DIR_ITEM_KEY
;
5169 /* special case for "." */
5170 if (filp
->f_pos
== 0) {
5171 over
= filldir(dirent
, ".", 1,
5172 filp
->f_pos
, btrfs_ino(inode
), DT_DIR
);
5177 /* special case for .., just use the back ref */
5178 if (filp
->f_pos
== 1) {
5179 u64 pino
= parent_ino(filp
->f_path
.dentry
);
5180 over
= filldir(dirent
, "..", 2,
5181 filp
->f_pos
, pino
, DT_DIR
);
5186 path
= btrfs_alloc_path();
5192 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5193 INIT_LIST_HEAD(&ins_list
);
5194 INIT_LIST_HEAD(&del_list
);
5195 btrfs_get_delayed_items(inode
, &ins_list
, &del_list
);
5198 btrfs_set_key_type(&key
, key_type
);
5199 key
.offset
= filp
->f_pos
;
5200 key
.objectid
= btrfs_ino(inode
);
5202 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5207 leaf
= path
->nodes
[0];
5208 slot
= path
->slots
[0];
5209 if (slot
>= btrfs_header_nritems(leaf
)) {
5210 ret
= btrfs_next_leaf(root
, path
);
5218 item
= btrfs_item_nr(leaf
, slot
);
5219 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
5221 if (found_key
.objectid
!= key
.objectid
)
5223 if (btrfs_key_type(&found_key
) != key_type
)
5225 if (found_key
.offset
< filp
->f_pos
)
5227 if (key_type
== BTRFS_DIR_INDEX_KEY
&&
5228 btrfs_should_delete_dir_index(&del_list
,
5232 filp
->f_pos
= found_key
.offset
;
5235 di
= btrfs_item_ptr(leaf
, slot
, struct btrfs_dir_item
);
5237 di_total
= btrfs_item_size(leaf
, item
);
5239 while (di_cur
< di_total
) {
5240 struct btrfs_key location
;
5242 if (verify_dir_item(root
, leaf
, di
))
5245 name_len
= btrfs_dir_name_len(leaf
, di
);
5246 if (name_len
<= sizeof(tmp_name
)) {
5247 name_ptr
= tmp_name
;
5249 name_ptr
= kmalloc(name_len
, GFP_NOFS
);
5255 read_extent_buffer(leaf
, name_ptr
,
5256 (unsigned long)(di
+ 1), name_len
);
5258 d_type
= btrfs_filetype_table
[btrfs_dir_type(leaf
, di
)];
5259 btrfs_dir_item_key_to_cpu(leaf
, di
, &location
);
5262 /* is this a reference to our own snapshot? If so
5265 * In contrast to old kernels, we insert the snapshot's
5266 * dir item and dir index after it has been created, so
5267 * we won't find a reference to our own snapshot. We
5268 * still keep the following code for backward
5271 if (location
.type
== BTRFS_ROOT_ITEM_KEY
&&
5272 location
.objectid
== root
->root_key
.objectid
) {
5276 over
= filldir(dirent
, name_ptr
, name_len
,
5277 found_key
.offset
, location
.objectid
,
5281 if (name_ptr
!= tmp_name
)
5286 di_len
= btrfs_dir_name_len(leaf
, di
) +
5287 btrfs_dir_data_len(leaf
, di
) + sizeof(*di
);
5289 di
= (struct btrfs_dir_item
*)((char *)di
+ di_len
);
5295 if (key_type
== BTRFS_DIR_INDEX_KEY
) {
5298 ret
= btrfs_readdir_delayed_dir_index(filp
, dirent
, filldir
,
5304 /* Reached end of directory/root. Bump pos past the last item. */
5305 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5307 * 32-bit glibc will use getdents64, but then strtol -
5308 * so the last number we can serve is this.
5310 filp
->f_pos
= 0x7fffffff;
5316 if (key_type
== BTRFS_DIR_INDEX_KEY
)
5317 btrfs_put_delayed_items(&ins_list
, &del_list
);
5318 btrfs_free_path(path
);
5322 int btrfs_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5324 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5325 struct btrfs_trans_handle
*trans
;
5327 bool nolock
= false;
5329 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5332 if (btrfs_fs_closing(root
->fs_info
) && btrfs_is_free_space_inode(inode
))
5335 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
5337 trans
= btrfs_join_transaction_nolock(root
);
5339 trans
= btrfs_join_transaction(root
);
5341 return PTR_ERR(trans
);
5342 ret
= btrfs_commit_transaction(trans
, root
);
5348 * This is somewhat expensive, updating the tree every time the
5349 * inode changes. But, it is most likely to find the inode in cache.
5350 * FIXME, needs more benchmarking...there are no reasons other than performance
5351 * to keep or drop this code.
5353 static int btrfs_dirty_inode(struct inode
*inode
)
5355 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5356 struct btrfs_trans_handle
*trans
;
5359 if (test_bit(BTRFS_INODE_DUMMY
, &BTRFS_I(inode
)->runtime_flags
))
5362 trans
= btrfs_join_transaction(root
);
5364 return PTR_ERR(trans
);
5366 ret
= btrfs_update_inode(trans
, root
, inode
);
5367 if (ret
&& ret
== -ENOSPC
) {
5368 /* whoops, lets try again with the full transaction */
5369 btrfs_end_transaction(trans
, root
);
5370 trans
= btrfs_start_transaction(root
, 1);
5372 return PTR_ERR(trans
);
5374 ret
= btrfs_update_inode(trans
, root
, inode
);
5376 btrfs_end_transaction(trans
, root
);
5377 if (BTRFS_I(inode
)->delayed_node
)
5378 btrfs_balance_delayed_items(root
);
5384 * This is a copy of file_update_time. We need this so we can return error on
5385 * ENOSPC for updating the inode in the case of file write and mmap writes.
5387 static int btrfs_update_time(struct inode
*inode
, struct timespec
*now
,
5390 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5392 if (btrfs_root_readonly(root
))
5395 if (flags
& S_VERSION
)
5396 inode_inc_iversion(inode
);
5397 if (flags
& S_CTIME
)
5398 inode
->i_ctime
= *now
;
5399 if (flags
& S_MTIME
)
5400 inode
->i_mtime
= *now
;
5401 if (flags
& S_ATIME
)
5402 inode
->i_atime
= *now
;
5403 return btrfs_dirty_inode(inode
);
5407 * find the highest existing sequence number in a directory
5408 * and then set the in-memory index_cnt variable to reflect
5409 * free sequence numbers
5411 static int btrfs_set_inode_index_count(struct inode
*inode
)
5413 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
5414 struct btrfs_key key
, found_key
;
5415 struct btrfs_path
*path
;
5416 struct extent_buffer
*leaf
;
5419 key
.objectid
= btrfs_ino(inode
);
5420 btrfs_set_key_type(&key
, BTRFS_DIR_INDEX_KEY
);
5421 key
.offset
= (u64
)-1;
5423 path
= btrfs_alloc_path();
5427 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5430 /* FIXME: we should be able to handle this */
5436 * MAGIC NUMBER EXPLANATION:
5437 * since we search a directory based on f_pos we have to start at 2
5438 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5439 * else has to start at 2
5441 if (path
->slots
[0] == 0) {
5442 BTRFS_I(inode
)->index_cnt
= 2;
5448 leaf
= path
->nodes
[0];
5449 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5451 if (found_key
.objectid
!= btrfs_ino(inode
) ||
5452 btrfs_key_type(&found_key
) != BTRFS_DIR_INDEX_KEY
) {
5453 BTRFS_I(inode
)->index_cnt
= 2;
5457 BTRFS_I(inode
)->index_cnt
= found_key
.offset
+ 1;
5459 btrfs_free_path(path
);
5464 * helper to find a free sequence number in a given directory. This current
5465 * code is very simple, later versions will do smarter things in the btree
5467 int btrfs_set_inode_index(struct inode
*dir
, u64
*index
)
5471 if (BTRFS_I(dir
)->index_cnt
== (u64
)-1) {
5472 ret
= btrfs_inode_delayed_dir_index_count(dir
);
5474 ret
= btrfs_set_inode_index_count(dir
);
5480 *index
= BTRFS_I(dir
)->index_cnt
;
5481 BTRFS_I(dir
)->index_cnt
++;
5486 static struct inode
*btrfs_new_inode(struct btrfs_trans_handle
*trans
,
5487 struct btrfs_root
*root
,
5489 const char *name
, int name_len
,
5490 u64 ref_objectid
, u64 objectid
,
5491 umode_t mode
, u64
*index
)
5493 struct inode
*inode
;
5494 struct btrfs_inode_item
*inode_item
;
5495 struct btrfs_key
*location
;
5496 struct btrfs_path
*path
;
5497 struct btrfs_inode_ref
*ref
;
5498 struct btrfs_key key
[2];
5504 path
= btrfs_alloc_path();
5506 return ERR_PTR(-ENOMEM
);
5508 inode
= new_inode(root
->fs_info
->sb
);
5510 btrfs_free_path(path
);
5511 return ERR_PTR(-ENOMEM
);
5515 * we have to initialize this early, so we can reclaim the inode
5516 * number if we fail afterwards in this function.
5518 inode
->i_ino
= objectid
;
5521 trace_btrfs_inode_request(dir
);
5523 ret
= btrfs_set_inode_index(dir
, index
);
5525 btrfs_free_path(path
);
5527 return ERR_PTR(ret
);
5531 * index_cnt is ignored for everything but a dir,
5532 * btrfs_get_inode_index_count has an explanation for the magic
5535 BTRFS_I(inode
)->index_cnt
= 2;
5536 BTRFS_I(inode
)->root
= root
;
5537 BTRFS_I(inode
)->generation
= trans
->transid
;
5538 inode
->i_generation
= BTRFS_I(inode
)->generation
;
5541 * We could have gotten an inode number from somebody who was fsynced
5542 * and then removed in this same transaction, so let's just set full
5543 * sync since it will be a full sync anyway and this will blow away the
5544 * old info in the log.
5546 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
5553 key
[0].objectid
= objectid
;
5554 btrfs_set_key_type(&key
[0], BTRFS_INODE_ITEM_KEY
);
5558 * Start new inodes with an inode_ref. This is slightly more
5559 * efficient for small numbers of hard links since they will
5560 * be packed into one item. Extended refs will kick in if we
5561 * add more hard links than can fit in the ref item.
5563 key
[1].objectid
= objectid
;
5564 btrfs_set_key_type(&key
[1], BTRFS_INODE_REF_KEY
);
5565 key
[1].offset
= ref_objectid
;
5567 sizes
[0] = sizeof(struct btrfs_inode_item
);
5568 sizes
[1] = name_len
+ sizeof(*ref
);
5570 path
->leave_spinning
= 1;
5571 ret
= btrfs_insert_empty_items(trans
, root
, path
, key
, sizes
, 2);
5575 inode_init_owner(inode
, dir
, mode
);
5576 inode_set_bytes(inode
, 0);
5577 inode
->i_mtime
= inode
->i_atime
= inode
->i_ctime
= CURRENT_TIME
;
5578 inode_item
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0],
5579 struct btrfs_inode_item
);
5580 memset_extent_buffer(path
->nodes
[0], 0, (unsigned long)inode_item
,
5581 sizeof(*inode_item
));
5582 fill_inode_item(trans
, path
->nodes
[0], inode_item
, inode
);
5584 ref
= btrfs_item_ptr(path
->nodes
[0], path
->slots
[0] + 1,
5585 struct btrfs_inode_ref
);
5586 btrfs_set_inode_ref_name_len(path
->nodes
[0], ref
, name_len
);
5587 btrfs_set_inode_ref_index(path
->nodes
[0], ref
, *index
);
5588 ptr
= (unsigned long)(ref
+ 1);
5589 write_extent_buffer(path
->nodes
[0], name
, ptr
, name_len
);
5591 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5592 btrfs_free_path(path
);
5594 location
= &BTRFS_I(inode
)->location
;
5595 location
->objectid
= objectid
;
5596 location
->offset
= 0;
5597 btrfs_set_key_type(location
, BTRFS_INODE_ITEM_KEY
);
5599 btrfs_inherit_iflags(inode
, dir
);
5601 if (S_ISREG(mode
)) {
5602 if (btrfs_test_opt(root
, NODATASUM
))
5603 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATASUM
;
5604 if (btrfs_test_opt(root
, NODATACOW
))
5605 BTRFS_I(inode
)->flags
|= BTRFS_INODE_NODATACOW
|
5606 BTRFS_INODE_NODATASUM
;
5609 insert_inode_hash(inode
);
5610 inode_tree_add(inode
);
5612 trace_btrfs_inode_new(inode
);
5613 btrfs_set_inode_last_trans(trans
, inode
);
5615 btrfs_update_root_times(trans
, root
);
5620 BTRFS_I(dir
)->index_cnt
--;
5621 btrfs_free_path(path
);
5623 return ERR_PTR(ret
);
5626 static inline u8
btrfs_inode_type(struct inode
*inode
)
5628 return btrfs_type_by_mode
[(inode
->i_mode
& S_IFMT
) >> S_SHIFT
];
5632 * utility function to add 'inode' into 'parent_inode' with
5633 * a give name and a given sequence number.
5634 * if 'add_backref' is true, also insert a backref from the
5635 * inode to the parent directory.
5637 int btrfs_add_link(struct btrfs_trans_handle
*trans
,
5638 struct inode
*parent_inode
, struct inode
*inode
,
5639 const char *name
, int name_len
, int add_backref
, u64 index
)
5642 struct btrfs_key key
;
5643 struct btrfs_root
*root
= BTRFS_I(parent_inode
)->root
;
5644 u64 ino
= btrfs_ino(inode
);
5645 u64 parent_ino
= btrfs_ino(parent_inode
);
5647 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5648 memcpy(&key
, &BTRFS_I(inode
)->root
->root_key
, sizeof(key
));
5651 btrfs_set_key_type(&key
, BTRFS_INODE_ITEM_KEY
);
5655 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5656 ret
= btrfs_add_root_ref(trans
, root
->fs_info
->tree_root
,
5657 key
.objectid
, root
->root_key
.objectid
,
5658 parent_ino
, index
, name
, name_len
);
5659 } else if (add_backref
) {
5660 ret
= btrfs_insert_inode_ref(trans
, root
, name
, name_len
, ino
,
5664 /* Nothing to clean up yet */
5668 ret
= btrfs_insert_dir_item(trans
, root
, name
, name_len
,
5670 btrfs_inode_type(inode
), index
);
5671 if (ret
== -EEXIST
|| ret
== -EOVERFLOW
)
5674 btrfs_abort_transaction(trans
, root
, ret
);
5678 btrfs_i_size_write(parent_inode
, parent_inode
->i_size
+
5680 inode_inc_iversion(parent_inode
);
5681 parent_inode
->i_mtime
= parent_inode
->i_ctime
= CURRENT_TIME
;
5682 ret
= btrfs_update_inode(trans
, root
, parent_inode
);
5684 btrfs_abort_transaction(trans
, root
, ret
);
5688 if (unlikely(ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
5691 err
= btrfs_del_root_ref(trans
, root
->fs_info
->tree_root
,
5692 key
.objectid
, root
->root_key
.objectid
,
5693 parent_ino
, &local_index
, name
, name_len
);
5695 } else if (add_backref
) {
5699 err
= btrfs_del_inode_ref(trans
, root
, name
, name_len
,
5700 ino
, parent_ino
, &local_index
);
5705 static int btrfs_add_nondir(struct btrfs_trans_handle
*trans
,
5706 struct inode
*dir
, struct dentry
*dentry
,
5707 struct inode
*inode
, int backref
, u64 index
)
5709 int err
= btrfs_add_link(trans
, dir
, inode
,
5710 dentry
->d_name
.name
, dentry
->d_name
.len
,
5717 static int btrfs_mknod(struct inode
*dir
, struct dentry
*dentry
,
5718 umode_t mode
, dev_t rdev
)
5720 struct btrfs_trans_handle
*trans
;
5721 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5722 struct inode
*inode
= NULL
;
5728 if (!new_valid_dev(rdev
))
5732 * 2 for inode item and ref
5734 * 1 for xattr if selinux is on
5736 trans
= btrfs_start_transaction(root
, 5);
5738 return PTR_ERR(trans
);
5740 err
= btrfs_find_free_ino(root
, &objectid
);
5744 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5745 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5747 if (IS_ERR(inode
)) {
5748 err
= PTR_ERR(inode
);
5752 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5759 * If the active LSM wants to access the inode during
5760 * d_instantiate it needs these. Smack checks to see
5761 * if the filesystem supports xattrs by looking at the
5765 inode
->i_op
= &btrfs_special_inode_operations
;
5766 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5770 init_special_inode(inode
, inode
->i_mode
, rdev
);
5771 btrfs_update_inode(trans
, root
, inode
);
5772 d_instantiate(dentry
, inode
);
5775 btrfs_end_transaction(trans
, root
);
5776 btrfs_btree_balance_dirty(root
);
5778 inode_dec_link_count(inode
);
5784 static int btrfs_create(struct inode
*dir
, struct dentry
*dentry
,
5785 umode_t mode
, bool excl
)
5787 struct btrfs_trans_handle
*trans
;
5788 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5789 struct inode
*inode
= NULL
;
5790 int drop_inode_on_err
= 0;
5796 * 2 for inode item and ref
5798 * 1 for xattr if selinux is on
5800 trans
= btrfs_start_transaction(root
, 5);
5802 return PTR_ERR(trans
);
5804 err
= btrfs_find_free_ino(root
, &objectid
);
5808 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5809 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5811 if (IS_ERR(inode
)) {
5812 err
= PTR_ERR(inode
);
5815 drop_inode_on_err
= 1;
5817 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5821 err
= btrfs_update_inode(trans
, root
, inode
);
5826 * If the active LSM wants to access the inode during
5827 * d_instantiate it needs these. Smack checks to see
5828 * if the filesystem supports xattrs by looking at the
5831 inode
->i_fop
= &btrfs_file_operations
;
5832 inode
->i_op
= &btrfs_file_inode_operations
;
5834 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
5838 inode
->i_mapping
->a_ops
= &btrfs_aops
;
5839 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
5840 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
5841 d_instantiate(dentry
, inode
);
5844 btrfs_end_transaction(trans
, root
);
5845 if (err
&& drop_inode_on_err
) {
5846 inode_dec_link_count(inode
);
5849 btrfs_btree_balance_dirty(root
);
5853 static int btrfs_link(struct dentry
*old_dentry
, struct inode
*dir
,
5854 struct dentry
*dentry
)
5856 struct btrfs_trans_handle
*trans
;
5857 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5858 struct inode
*inode
= old_dentry
->d_inode
;
5863 /* do not allow sys_link's with other subvols of the same device */
5864 if (root
->objectid
!= BTRFS_I(inode
)->root
->objectid
)
5867 if (inode
->i_nlink
>= BTRFS_LINK_MAX
)
5870 err
= btrfs_set_inode_index(dir
, &index
);
5875 * 2 items for inode and inode ref
5876 * 2 items for dir items
5877 * 1 item for parent inode
5879 trans
= btrfs_start_transaction(root
, 5);
5880 if (IS_ERR(trans
)) {
5881 err
= PTR_ERR(trans
);
5885 btrfs_inc_nlink(inode
);
5886 inode_inc_iversion(inode
);
5887 inode
->i_ctime
= CURRENT_TIME
;
5889 set_bit(BTRFS_INODE_COPY_EVERYTHING
, &BTRFS_I(inode
)->runtime_flags
);
5891 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 1, index
);
5896 struct dentry
*parent
= dentry
->d_parent
;
5897 err
= btrfs_update_inode(trans
, root
, inode
);
5900 d_instantiate(dentry
, inode
);
5901 btrfs_log_new_name(trans
, inode
, NULL
, parent
);
5904 btrfs_end_transaction(trans
, root
);
5907 inode_dec_link_count(inode
);
5910 btrfs_btree_balance_dirty(root
);
5914 static int btrfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
5916 struct inode
*inode
= NULL
;
5917 struct btrfs_trans_handle
*trans
;
5918 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
5920 int drop_on_err
= 0;
5925 * 2 items for inode and ref
5926 * 2 items for dir items
5927 * 1 for xattr if selinux is on
5929 trans
= btrfs_start_transaction(root
, 5);
5931 return PTR_ERR(trans
);
5933 err
= btrfs_find_free_ino(root
, &objectid
);
5937 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
5938 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
5939 S_IFDIR
| mode
, &index
);
5940 if (IS_ERR(inode
)) {
5941 err
= PTR_ERR(inode
);
5947 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
5951 inode
->i_op
= &btrfs_dir_inode_operations
;
5952 inode
->i_fop
= &btrfs_dir_file_operations
;
5954 btrfs_i_size_write(inode
, 0);
5955 err
= btrfs_update_inode(trans
, root
, inode
);
5959 err
= btrfs_add_link(trans
, dir
, inode
, dentry
->d_name
.name
,
5960 dentry
->d_name
.len
, 0, index
);
5964 d_instantiate(dentry
, inode
);
5968 btrfs_end_transaction(trans
, root
);
5971 btrfs_btree_balance_dirty(root
);
5975 /* helper for btfs_get_extent. Given an existing extent in the tree,
5976 * and an extent that you want to insert, deal with overlap and insert
5977 * the new extent into the tree.
5979 static int merge_extent_mapping(struct extent_map_tree
*em_tree
,
5980 struct extent_map
*existing
,
5981 struct extent_map
*em
,
5982 u64 map_start
, u64 map_len
)
5986 BUG_ON(map_start
< em
->start
|| map_start
>= extent_map_end(em
));
5987 start_diff
= map_start
- em
->start
;
5988 em
->start
= map_start
;
5990 if (em
->block_start
< EXTENT_MAP_LAST_BYTE
&&
5991 !test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
)) {
5992 em
->block_start
+= start_diff
;
5993 em
->block_len
-= start_diff
;
5995 return add_extent_mapping(em_tree
, em
, 0);
5998 static noinline
int uncompress_inline(struct btrfs_path
*path
,
5999 struct inode
*inode
, struct page
*page
,
6000 size_t pg_offset
, u64 extent_offset
,
6001 struct btrfs_file_extent_item
*item
)
6004 struct extent_buffer
*leaf
= path
->nodes
[0];
6007 unsigned long inline_size
;
6011 WARN_ON(pg_offset
!= 0);
6012 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6013 max_size
= btrfs_file_extent_ram_bytes(leaf
, item
);
6014 inline_size
= btrfs_file_extent_inline_item_len(leaf
,
6015 btrfs_item_nr(leaf
, path
->slots
[0]));
6016 tmp
= kmalloc(inline_size
, GFP_NOFS
);
6019 ptr
= btrfs_file_extent_inline_start(item
);
6021 read_extent_buffer(leaf
, tmp
, ptr
, inline_size
);
6023 max_size
= min_t(unsigned long, PAGE_CACHE_SIZE
, max_size
);
6024 ret
= btrfs_decompress(compress_type
, tmp
, page
,
6025 extent_offset
, inline_size
, max_size
);
6027 char *kaddr
= kmap_atomic(page
);
6028 unsigned long copy_size
= min_t(u64
,
6029 PAGE_CACHE_SIZE
- pg_offset
,
6030 max_size
- extent_offset
);
6031 memset(kaddr
+ pg_offset
, 0, copy_size
);
6032 kunmap_atomic(kaddr
);
6039 * a bit scary, this does extent mapping from logical file offset to the disk.
6040 * the ugly parts come from merging extents from the disk with the in-ram
6041 * representation. This gets more complex because of the data=ordered code,
6042 * where the in-ram extents might be locked pending data=ordered completion.
6044 * This also copies inline extents directly into the page.
6047 struct extent_map
*btrfs_get_extent(struct inode
*inode
, struct page
*page
,
6048 size_t pg_offset
, u64 start
, u64 len
,
6054 u64 extent_start
= 0;
6056 u64 objectid
= btrfs_ino(inode
);
6058 struct btrfs_path
*path
= NULL
;
6059 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6060 struct btrfs_file_extent_item
*item
;
6061 struct extent_buffer
*leaf
;
6062 struct btrfs_key found_key
;
6063 struct extent_map
*em
= NULL
;
6064 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
6065 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
6066 struct btrfs_trans_handle
*trans
= NULL
;
6070 read_lock(&em_tree
->lock
);
6071 em
= lookup_extent_mapping(em_tree
, start
, len
);
6073 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6074 read_unlock(&em_tree
->lock
);
6077 if (em
->start
> start
|| em
->start
+ em
->len
<= start
)
6078 free_extent_map(em
);
6079 else if (em
->block_start
== EXTENT_MAP_INLINE
&& page
)
6080 free_extent_map(em
);
6084 em
= alloc_extent_map();
6089 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6090 em
->start
= EXTENT_MAP_HOLE
;
6091 em
->orig_start
= EXTENT_MAP_HOLE
;
6093 em
->block_len
= (u64
)-1;
6096 path
= btrfs_alloc_path();
6102 * Chances are we'll be called again, so go ahead and do
6108 ret
= btrfs_lookup_file_extent(trans
, root
, path
,
6109 objectid
, start
, trans
!= NULL
);
6116 if (path
->slots
[0] == 0)
6121 leaf
= path
->nodes
[0];
6122 item
= btrfs_item_ptr(leaf
, path
->slots
[0],
6123 struct btrfs_file_extent_item
);
6124 /* are we inside the extent that was found? */
6125 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6126 found_type
= btrfs_key_type(&found_key
);
6127 if (found_key
.objectid
!= objectid
||
6128 found_type
!= BTRFS_EXTENT_DATA_KEY
) {
6132 found_type
= btrfs_file_extent_type(leaf
, item
);
6133 extent_start
= found_key
.offset
;
6134 compress_type
= btrfs_file_extent_compression(leaf
, item
);
6135 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6136 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6137 extent_end
= extent_start
+
6138 btrfs_file_extent_num_bytes(leaf
, item
);
6139 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6141 size
= btrfs_file_extent_inline_len(leaf
, item
);
6142 extent_end
= ALIGN(extent_start
+ size
, root
->sectorsize
);
6145 if (start
>= extent_end
) {
6147 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
6148 ret
= btrfs_next_leaf(root
, path
);
6155 leaf
= path
->nodes
[0];
6157 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
6158 if (found_key
.objectid
!= objectid
||
6159 found_key
.type
!= BTRFS_EXTENT_DATA_KEY
)
6161 if (start
+ len
<= found_key
.offset
)
6164 em
->orig_start
= start
;
6165 em
->len
= found_key
.offset
- start
;
6169 em
->ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, item
);
6170 if (found_type
== BTRFS_FILE_EXTENT_REG
||
6171 found_type
== BTRFS_FILE_EXTENT_PREALLOC
) {
6172 em
->start
= extent_start
;
6173 em
->len
= extent_end
- extent_start
;
6174 em
->orig_start
= extent_start
-
6175 btrfs_file_extent_offset(leaf
, item
);
6176 em
->orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
,
6178 bytenr
= btrfs_file_extent_disk_bytenr(leaf
, item
);
6180 em
->block_start
= EXTENT_MAP_HOLE
;
6183 if (compress_type
!= BTRFS_COMPRESS_NONE
) {
6184 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6185 em
->compress_type
= compress_type
;
6186 em
->block_start
= bytenr
;
6187 em
->block_len
= em
->orig_block_len
;
6189 bytenr
+= btrfs_file_extent_offset(leaf
, item
);
6190 em
->block_start
= bytenr
;
6191 em
->block_len
= em
->len
;
6192 if (found_type
== BTRFS_FILE_EXTENT_PREALLOC
)
6193 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6196 } else if (found_type
== BTRFS_FILE_EXTENT_INLINE
) {
6200 size_t extent_offset
;
6203 em
->block_start
= EXTENT_MAP_INLINE
;
6204 if (!page
|| create
) {
6205 em
->start
= extent_start
;
6206 em
->len
= extent_end
- extent_start
;
6210 size
= btrfs_file_extent_inline_len(leaf
, item
);
6211 extent_offset
= page_offset(page
) + pg_offset
- extent_start
;
6212 copy_size
= min_t(u64
, PAGE_CACHE_SIZE
- pg_offset
,
6213 size
- extent_offset
);
6214 em
->start
= extent_start
+ extent_offset
;
6215 em
->len
= ALIGN(copy_size
, root
->sectorsize
);
6216 em
->orig_block_len
= em
->len
;
6217 em
->orig_start
= em
->start
;
6218 if (compress_type
) {
6219 set_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
);
6220 em
->compress_type
= compress_type
;
6222 ptr
= btrfs_file_extent_inline_start(item
) + extent_offset
;
6223 if (create
== 0 && !PageUptodate(page
)) {
6224 if (btrfs_file_extent_compression(leaf
, item
) !=
6225 BTRFS_COMPRESS_NONE
) {
6226 ret
= uncompress_inline(path
, inode
, page
,
6228 extent_offset
, item
);
6229 BUG_ON(ret
); /* -ENOMEM */
6232 read_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6234 if (pg_offset
+ copy_size
< PAGE_CACHE_SIZE
) {
6235 memset(map
+ pg_offset
+ copy_size
, 0,
6236 PAGE_CACHE_SIZE
- pg_offset
-
6241 flush_dcache_page(page
);
6242 } else if (create
&& PageUptodate(page
)) {
6246 free_extent_map(em
);
6249 btrfs_release_path(path
);
6250 trans
= btrfs_join_transaction(root
);
6253 return ERR_CAST(trans
);
6257 write_extent_buffer(leaf
, map
+ pg_offset
, ptr
,
6260 btrfs_mark_buffer_dirty(leaf
);
6262 set_extent_uptodate(io_tree
, em
->start
,
6263 extent_map_end(em
) - 1, NULL
, GFP_NOFS
);
6266 WARN(1, KERN_ERR
"btrfs unknown found_type %d\n", found_type
);
6270 em
->orig_start
= start
;
6273 em
->block_start
= EXTENT_MAP_HOLE
;
6274 set_bit(EXTENT_FLAG_VACANCY
, &em
->flags
);
6276 btrfs_release_path(path
);
6277 if (em
->start
> start
|| extent_map_end(em
) <= start
) {
6278 btrfs_err(root
->fs_info
, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6279 (unsigned long long)em
->start
,
6280 (unsigned long long)em
->len
,
6281 (unsigned long long)start
,
6282 (unsigned long long)len
);
6288 write_lock(&em_tree
->lock
);
6289 ret
= add_extent_mapping(em_tree
, em
, 0);
6290 /* it is possible that someone inserted the extent into the tree
6291 * while we had the lock dropped. It is also possible that
6292 * an overlapping map exists in the tree
6294 if (ret
== -EEXIST
) {
6295 struct extent_map
*existing
;
6299 existing
= lookup_extent_mapping(em_tree
, start
, len
);
6300 if (existing
&& (existing
->start
> start
||
6301 existing
->start
+ existing
->len
<= start
)) {
6302 free_extent_map(existing
);
6306 existing
= lookup_extent_mapping(em_tree
, em
->start
,
6309 err
= merge_extent_mapping(em_tree
, existing
,
6312 free_extent_map(existing
);
6314 free_extent_map(em
);
6319 free_extent_map(em
);
6323 free_extent_map(em
);
6328 write_unlock(&em_tree
->lock
);
6332 trace_btrfs_get_extent(root
, em
);
6335 btrfs_free_path(path
);
6337 ret
= btrfs_end_transaction(trans
, root
);
6342 free_extent_map(em
);
6343 return ERR_PTR(err
);
6345 BUG_ON(!em
); /* Error is always set */
6349 struct extent_map
*btrfs_get_extent_fiemap(struct inode
*inode
, struct page
*page
,
6350 size_t pg_offset
, u64 start
, u64 len
,
6353 struct extent_map
*em
;
6354 struct extent_map
*hole_em
= NULL
;
6355 u64 range_start
= start
;
6361 em
= btrfs_get_extent(inode
, page
, pg_offset
, start
, len
, create
);
6368 * - a pre-alloc extent,
6369 * there might actually be delalloc bytes behind it.
6371 if (em
->block_start
!= EXTENT_MAP_HOLE
&&
6372 !test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6378 /* check to see if we've wrapped (len == -1 or similar) */
6387 /* ok, we didn't find anything, lets look for delalloc */
6388 found
= count_range_bits(&BTRFS_I(inode
)->io_tree
, &range_start
,
6389 end
, len
, EXTENT_DELALLOC
, 1);
6390 found_end
= range_start
+ found
;
6391 if (found_end
< range_start
)
6392 found_end
= (u64
)-1;
6395 * we didn't find anything useful, return
6396 * the original results from get_extent()
6398 if (range_start
> end
|| found_end
<= start
) {
6404 /* adjust the range_start to make sure it doesn't
6405 * go backwards from the start they passed in
6407 range_start
= max(start
,range_start
);
6408 found
= found_end
- range_start
;
6411 u64 hole_start
= start
;
6414 em
= alloc_extent_map();
6420 * when btrfs_get_extent can't find anything it
6421 * returns one huge hole
6423 * make sure what it found really fits our range, and
6424 * adjust to make sure it is based on the start from
6428 u64 calc_end
= extent_map_end(hole_em
);
6430 if (calc_end
<= start
|| (hole_em
->start
> end
)) {
6431 free_extent_map(hole_em
);
6434 hole_start
= max(hole_em
->start
, start
);
6435 hole_len
= calc_end
- hole_start
;
6439 if (hole_em
&& range_start
> hole_start
) {
6440 /* our hole starts before our delalloc, so we
6441 * have to return just the parts of the hole
6442 * that go until the delalloc starts
6444 em
->len
= min(hole_len
,
6445 range_start
- hole_start
);
6446 em
->start
= hole_start
;
6447 em
->orig_start
= hole_start
;
6449 * don't adjust block start at all,
6450 * it is fixed at EXTENT_MAP_HOLE
6452 em
->block_start
= hole_em
->block_start
;
6453 em
->block_len
= hole_len
;
6454 if (test_bit(EXTENT_FLAG_PREALLOC
, &hole_em
->flags
))
6455 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
6457 em
->start
= range_start
;
6459 em
->orig_start
= range_start
;
6460 em
->block_start
= EXTENT_MAP_DELALLOC
;
6461 em
->block_len
= found
;
6463 } else if (hole_em
) {
6468 free_extent_map(hole_em
);
6470 free_extent_map(em
);
6471 return ERR_PTR(err
);
6476 static struct extent_map
*btrfs_new_extent_direct(struct inode
*inode
,
6479 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6480 struct btrfs_trans_handle
*trans
;
6481 struct extent_map
*em
;
6482 struct btrfs_key ins
;
6486 trans
= btrfs_join_transaction(root
);
6488 return ERR_CAST(trans
);
6490 trans
->block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
6492 alloc_hint
= get_extent_allocation_hint(inode
, start
, len
);
6493 ret
= btrfs_reserve_extent(trans
, root
, len
, root
->sectorsize
, 0,
6494 alloc_hint
, &ins
, 1);
6500 em
= create_pinned_em(inode
, start
, ins
.offset
, start
, ins
.objectid
,
6501 ins
.offset
, ins
.offset
, ins
.offset
, 0);
6505 ret
= btrfs_add_ordered_extent_dio(inode
, start
, ins
.objectid
,
6506 ins
.offset
, ins
.offset
, 0);
6508 btrfs_free_reserved_extent(root
, ins
.objectid
, ins
.offset
);
6512 btrfs_end_transaction(trans
, root
);
6517 * returns 1 when the nocow is safe, < 1 on error, 0 if the
6518 * block must be cow'd
6520 static noinline
int can_nocow_odirect(struct btrfs_trans_handle
*trans
,
6521 struct inode
*inode
, u64 offset
, u64
*len
,
6522 u64
*orig_start
, u64
*orig_block_len
,
6525 struct btrfs_path
*path
;
6527 struct extent_buffer
*leaf
;
6528 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6529 struct btrfs_file_extent_item
*fi
;
6530 struct btrfs_key key
;
6538 path
= btrfs_alloc_path();
6542 ret
= btrfs_lookup_file_extent(trans
, root
, path
, btrfs_ino(inode
),
6547 slot
= path
->slots
[0];
6550 /* can't find the item, must cow */
6557 leaf
= path
->nodes
[0];
6558 btrfs_item_key_to_cpu(leaf
, &key
, slot
);
6559 if (key
.objectid
!= btrfs_ino(inode
) ||
6560 key
.type
!= BTRFS_EXTENT_DATA_KEY
) {
6561 /* not our file or wrong item type, must cow */
6565 if (key
.offset
> offset
) {
6566 /* Wrong offset, must cow */
6570 fi
= btrfs_item_ptr(leaf
, slot
, struct btrfs_file_extent_item
);
6571 found_type
= btrfs_file_extent_type(leaf
, fi
);
6572 if (found_type
!= BTRFS_FILE_EXTENT_REG
&&
6573 found_type
!= BTRFS_FILE_EXTENT_PREALLOC
) {
6574 /* not a regular extent, must cow */
6577 disk_bytenr
= btrfs_file_extent_disk_bytenr(leaf
, fi
);
6578 backref_offset
= btrfs_file_extent_offset(leaf
, fi
);
6580 *orig_start
= key
.offset
- backref_offset
;
6581 *orig_block_len
= btrfs_file_extent_disk_num_bytes(leaf
, fi
);
6582 *ram_bytes
= btrfs_file_extent_ram_bytes(leaf
, fi
);
6584 extent_end
= key
.offset
+ btrfs_file_extent_num_bytes(leaf
, fi
);
6585 if (extent_end
< offset
+ *len
) {
6586 /* extent doesn't include our full range, must cow */
6590 if (btrfs_extent_readonly(root
, disk_bytenr
))
6594 * look for other files referencing this extent, if we
6595 * find any we must cow
6597 if (btrfs_cross_ref_exist(trans
, root
, btrfs_ino(inode
),
6598 key
.offset
- backref_offset
, disk_bytenr
))
6602 * adjust disk_bytenr and num_bytes to cover just the bytes
6603 * in this extent we are about to write. If there
6604 * are any csums in that range we have to cow in order
6605 * to keep the csums correct
6607 disk_bytenr
+= backref_offset
;
6608 disk_bytenr
+= offset
- key
.offset
;
6609 num_bytes
= min(offset
+ *len
, extent_end
) - offset
;
6610 if (csum_exist_in_range(root
, disk_bytenr
, num_bytes
))
6613 * all of the above have passed, it is safe to overwrite this extent
6619 btrfs_free_path(path
);
6623 static int lock_extent_direct(struct inode
*inode
, u64 lockstart
, u64 lockend
,
6624 struct extent_state
**cached_state
, int writing
)
6626 struct btrfs_ordered_extent
*ordered
;
6630 lock_extent_bits(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6633 * We're concerned with the entire range that we're going to be
6634 * doing DIO to, so we need to make sure theres no ordered
6635 * extents in this range.
6637 ordered
= btrfs_lookup_ordered_range(inode
, lockstart
,
6638 lockend
- lockstart
+ 1);
6641 * We need to make sure there are no buffered pages in this
6642 * range either, we could have raced between the invalidate in
6643 * generic_file_direct_write and locking the extent. The
6644 * invalidate needs to happen so that reads after a write do not
6647 if (!ordered
&& (!writing
||
6648 !test_range_bit(&BTRFS_I(inode
)->io_tree
,
6649 lockstart
, lockend
, EXTENT_UPTODATE
, 0,
6653 unlock_extent_cached(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6654 cached_state
, GFP_NOFS
);
6657 btrfs_start_ordered_extent(inode
, ordered
, 1);
6658 btrfs_put_ordered_extent(ordered
);
6660 /* Screw you mmap */
6661 ret
= filemap_write_and_wait_range(inode
->i_mapping
,
6668 * If we found a page that couldn't be invalidated just
6669 * fall back to buffered.
6671 ret
= invalidate_inode_pages2_range(inode
->i_mapping
,
6672 lockstart
>> PAGE_CACHE_SHIFT
,
6673 lockend
>> PAGE_CACHE_SHIFT
);
6684 static struct extent_map
*create_pinned_em(struct inode
*inode
, u64 start
,
6685 u64 len
, u64 orig_start
,
6686 u64 block_start
, u64 block_len
,
6687 u64 orig_block_len
, u64 ram_bytes
,
6690 struct extent_map_tree
*em_tree
;
6691 struct extent_map
*em
;
6692 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6695 em_tree
= &BTRFS_I(inode
)->extent_tree
;
6696 em
= alloc_extent_map();
6698 return ERR_PTR(-ENOMEM
);
6701 em
->orig_start
= orig_start
;
6702 em
->mod_start
= start
;
6705 em
->block_len
= block_len
;
6706 em
->block_start
= block_start
;
6707 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
6708 em
->orig_block_len
= orig_block_len
;
6709 em
->ram_bytes
= ram_bytes
;
6710 em
->generation
= -1;
6711 set_bit(EXTENT_FLAG_PINNED
, &em
->flags
);
6712 if (type
== BTRFS_ORDERED_PREALLOC
)
6713 set_bit(EXTENT_FLAG_FILLING
, &em
->flags
);
6716 btrfs_drop_extent_cache(inode
, em
->start
,
6717 em
->start
+ em
->len
- 1, 0);
6718 write_lock(&em_tree
->lock
);
6719 ret
= add_extent_mapping(em_tree
, em
, 1);
6720 write_unlock(&em_tree
->lock
);
6721 } while (ret
== -EEXIST
);
6724 free_extent_map(em
);
6725 return ERR_PTR(ret
);
6732 static int btrfs_get_blocks_direct(struct inode
*inode
, sector_t iblock
,
6733 struct buffer_head
*bh_result
, int create
)
6735 struct extent_map
*em
;
6736 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6737 struct extent_state
*cached_state
= NULL
;
6738 u64 start
= iblock
<< inode
->i_blkbits
;
6739 u64 lockstart
, lockend
;
6740 u64 len
= bh_result
->b_size
;
6741 struct btrfs_trans_handle
*trans
;
6742 int unlock_bits
= EXTENT_LOCKED
;
6746 unlock_bits
|= EXTENT_DELALLOC
| EXTENT_DIRTY
;
6748 len
= min_t(u64
, len
, root
->sectorsize
);
6751 lockend
= start
+ len
- 1;
6754 * If this errors out it's because we couldn't invalidate pagecache for
6755 * this range and we need to fallback to buffered.
6757 if (lock_extent_direct(inode
, lockstart
, lockend
, &cached_state
, create
))
6760 em
= btrfs_get_extent(inode
, NULL
, 0, start
, len
, 0);
6767 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6768 * io. INLINE is special, and we could probably kludge it in here, but
6769 * it's still buffered so for safety lets just fall back to the generic
6772 * For COMPRESSED we _have_ to read the entire extent in so we can
6773 * decompress it, so there will be buffering required no matter what we
6774 * do, so go ahead and fallback to buffered.
6776 * We return -ENOTBLK because thats what makes DIO go ahead and go back
6777 * to buffered IO. Don't blame me, this is the price we pay for using
6780 if (test_bit(EXTENT_FLAG_COMPRESSED
, &em
->flags
) ||
6781 em
->block_start
== EXTENT_MAP_INLINE
) {
6782 free_extent_map(em
);
6787 /* Just a good old fashioned hole, return */
6788 if (!create
&& (em
->block_start
== EXTENT_MAP_HOLE
||
6789 test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))) {
6790 free_extent_map(em
);
6795 * We don't allocate a new extent in the following cases
6797 * 1) The inode is marked as NODATACOW. In this case we'll just use the
6799 * 2) The extent is marked as PREALLOC. We're good to go here and can
6800 * just use the extent.
6804 len
= min(len
, em
->len
- (start
- em
->start
));
6805 lockstart
= start
+ len
;
6809 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
) ||
6810 ((BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATACOW
) &&
6811 em
->block_start
!= EXTENT_MAP_HOLE
)) {
6814 u64 block_start
, orig_start
, orig_block_len
, ram_bytes
;
6816 if (test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6817 type
= BTRFS_ORDERED_PREALLOC
;
6819 type
= BTRFS_ORDERED_NOCOW
;
6820 len
= min(len
, em
->len
- (start
- em
->start
));
6821 block_start
= em
->block_start
+ (start
- em
->start
);
6824 * we're not going to log anything, but we do need
6825 * to make sure the current transaction stays open
6826 * while we look for nocow cross refs
6828 trans
= btrfs_join_transaction(root
);
6832 if (can_nocow_odirect(trans
, inode
, start
, &len
, &orig_start
,
6833 &orig_block_len
, &ram_bytes
) == 1) {
6834 if (type
== BTRFS_ORDERED_PREALLOC
) {
6835 free_extent_map(em
);
6836 em
= create_pinned_em(inode
, start
, len
,
6842 btrfs_end_transaction(trans
, root
);
6847 ret
= btrfs_add_ordered_extent_dio(inode
, start
,
6848 block_start
, len
, len
, type
);
6849 btrfs_end_transaction(trans
, root
);
6851 free_extent_map(em
);
6856 btrfs_end_transaction(trans
, root
);
6860 * this will cow the extent, reset the len in case we changed
6863 len
= bh_result
->b_size
;
6864 free_extent_map(em
);
6865 em
= btrfs_new_extent_direct(inode
, start
, len
);
6870 len
= min(len
, em
->len
- (start
- em
->start
));
6872 bh_result
->b_blocknr
= (em
->block_start
+ (start
- em
->start
)) >>
6874 bh_result
->b_size
= len
;
6875 bh_result
->b_bdev
= em
->bdev
;
6876 set_buffer_mapped(bh_result
);
6878 if (!test_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
))
6879 set_buffer_new(bh_result
);
6882 * Need to update the i_size under the extent lock so buffered
6883 * readers will get the updated i_size when we unlock.
6885 if (start
+ len
> i_size_read(inode
))
6886 i_size_write(inode
, start
+ len
);
6888 spin_lock(&BTRFS_I(inode
)->lock
);
6889 BTRFS_I(inode
)->outstanding_extents
++;
6890 spin_unlock(&BTRFS_I(inode
)->lock
);
6892 ret
= set_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6893 lockstart
+ len
- 1, EXTENT_DELALLOC
, NULL
,
6894 &cached_state
, GFP_NOFS
);
6899 * In the case of write we need to clear and unlock the entire range,
6900 * in the case of read we need to unlock only the end area that we
6901 * aren't using if there is any left over space.
6903 if (lockstart
< lockend
) {
6904 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
,
6905 lockend
, unlock_bits
, 1, 0,
6906 &cached_state
, GFP_NOFS
);
6908 free_extent_state(cached_state
);
6911 free_extent_map(em
);
6916 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, lockstart
, lockend
,
6917 unlock_bits
, 1, 0, &cached_state
, GFP_NOFS
);
6921 struct btrfs_dio_private
{
6922 struct inode
*inode
;
6928 /* number of bios pending for this dio */
6929 atomic_t pending_bios
;
6934 struct bio
*orig_bio
;
6937 static void btrfs_endio_direct_read(struct bio
*bio
, int err
)
6939 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6940 struct bio_vec
*bvec_end
= bio
->bi_io_vec
+ bio
->bi_vcnt
- 1;
6941 struct bio_vec
*bvec
= bio
->bi_io_vec
;
6942 struct inode
*inode
= dip
->inode
;
6943 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6946 start
= dip
->logical_offset
;
6948 if (!(BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
)) {
6949 struct page
*page
= bvec
->bv_page
;
6952 u64
private = ~(u32
)0;
6953 unsigned long flags
;
6955 if (get_state_private(&BTRFS_I(inode
)->io_tree
,
6958 local_irq_save(flags
);
6959 kaddr
= kmap_atomic(page
);
6960 csum
= btrfs_csum_data(kaddr
+ bvec
->bv_offset
,
6961 csum
, bvec
->bv_len
);
6962 btrfs_csum_final(csum
, (char *)&csum
);
6963 kunmap_atomic(kaddr
);
6964 local_irq_restore(flags
);
6966 flush_dcache_page(bvec
->bv_page
);
6967 if (csum
!= private) {
6969 btrfs_err(root
->fs_info
, "csum failed ino %llu off %llu csum %u private %u",
6970 (unsigned long long)btrfs_ino(inode
),
6971 (unsigned long long)start
,
6972 csum
, (unsigned)private);
6977 start
+= bvec
->bv_len
;
6979 } while (bvec
<= bvec_end
);
6981 unlock_extent(&BTRFS_I(inode
)->io_tree
, dip
->logical_offset
,
6982 dip
->logical_offset
+ dip
->bytes
- 1);
6983 bio
->bi_private
= dip
->private;
6987 /* If we had a csum failure make sure to clear the uptodate flag */
6989 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
6990 dio_end_io(bio
, err
);
6993 static void btrfs_endio_direct_write(struct bio
*bio
, int err
)
6995 struct btrfs_dio_private
*dip
= bio
->bi_private
;
6996 struct inode
*inode
= dip
->inode
;
6997 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
6998 struct btrfs_ordered_extent
*ordered
= NULL
;
6999 u64 ordered_offset
= dip
->logical_offset
;
7000 u64 ordered_bytes
= dip
->bytes
;
7006 ret
= btrfs_dec_test_first_ordered_pending(inode
, &ordered
,
7008 ordered_bytes
, !err
);
7012 ordered
->work
.func
= finish_ordered_fn
;
7013 ordered
->work
.flags
= 0;
7014 btrfs_queue_worker(&root
->fs_info
->endio_write_workers
,
7018 * our bio might span multiple ordered extents. If we haven't
7019 * completed the accounting for the whole dio, go back and try again
7021 if (ordered_offset
< dip
->logical_offset
+ dip
->bytes
) {
7022 ordered_bytes
= dip
->logical_offset
+ dip
->bytes
-
7028 bio
->bi_private
= dip
->private;
7032 /* If we had an error make sure to clear the uptodate flag */
7034 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
7035 dio_end_io(bio
, err
);
7038 static int __btrfs_submit_bio_start_direct_io(struct inode
*inode
, int rw
,
7039 struct bio
*bio
, int mirror_num
,
7040 unsigned long bio_flags
, u64 offset
)
7043 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7044 ret
= btrfs_csum_one_bio(root
, inode
, bio
, offset
, 1);
7045 BUG_ON(ret
); /* -ENOMEM */
7049 static void btrfs_end_dio_bio(struct bio
*bio
, int err
)
7051 struct btrfs_dio_private
*dip
= bio
->bi_private
;
7054 printk(KERN_ERR
"btrfs direct IO failed ino %llu rw %lu "
7055 "sector %#Lx len %u err no %d\n",
7056 (unsigned long long)btrfs_ino(dip
->inode
), bio
->bi_rw
,
7057 (unsigned long long)bio
->bi_sector
, bio
->bi_size
, err
);
7061 * before atomic variable goto zero, we must make sure
7062 * dip->errors is perceived to be set.
7064 smp_mb__before_atomic_dec();
7067 /* if there are more bios still pending for this dio, just exit */
7068 if (!atomic_dec_and_test(&dip
->pending_bios
))
7072 bio_io_error(dip
->orig_bio
);
7074 set_bit(BIO_UPTODATE
, &dip
->orig_bio
->bi_flags
);
7075 bio_endio(dip
->orig_bio
, 0);
7081 static struct bio
*btrfs_dio_bio_alloc(struct block_device
*bdev
,
7082 u64 first_sector
, gfp_t gfp_flags
)
7084 int nr_vecs
= bio_get_nr_vecs(bdev
);
7085 return btrfs_bio_alloc(bdev
, first_sector
, nr_vecs
, gfp_flags
);
7088 static inline int __btrfs_submit_dio_bio(struct bio
*bio
, struct inode
*inode
,
7089 int rw
, u64 file_offset
, int skip_sum
,
7092 int write
= rw
& REQ_WRITE
;
7093 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7097 async_submit
= !atomic_read(&BTRFS_I(inode
)->sync_writers
);
7102 ret
= btrfs_bio_wq_end_io(root
->fs_info
, bio
, 0);
7110 if (write
&& async_submit
) {
7111 ret
= btrfs_wq_submit_bio(root
->fs_info
,
7112 inode
, rw
, bio
, 0, 0,
7114 __btrfs_submit_bio_start_direct_io
,
7115 __btrfs_submit_bio_done
);
7119 * If we aren't doing async submit, calculate the csum of the
7122 ret
= btrfs_csum_one_bio(root
, inode
, bio
, file_offset
, 1);
7125 } else if (!skip_sum
) {
7126 ret
= btrfs_lookup_bio_sums_dio(root
, inode
, bio
, file_offset
);
7132 ret
= btrfs_map_bio(root
, rw
, bio
, 0, async_submit
);
7138 static int btrfs_submit_direct_hook(int rw
, struct btrfs_dio_private
*dip
,
7141 struct inode
*inode
= dip
->inode
;
7142 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7144 struct bio
*orig_bio
= dip
->orig_bio
;
7145 struct bio_vec
*bvec
= orig_bio
->bi_io_vec
;
7146 u64 start_sector
= orig_bio
->bi_sector
;
7147 u64 file_offset
= dip
->logical_offset
;
7152 int async_submit
= 0;
7154 map_length
= orig_bio
->bi_size
;
7155 ret
= btrfs_map_block(root
->fs_info
, rw
, start_sector
<< 9,
7156 &map_length
, NULL
, 0);
7161 if (map_length
>= orig_bio
->bi_size
) {
7166 /* async crcs make it difficult to collect full stripe writes. */
7167 if (btrfs_get_alloc_profile(root
, 1) &
7168 (BTRFS_BLOCK_GROUP_RAID5
| BTRFS_BLOCK_GROUP_RAID6
))
7173 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
, start_sector
, GFP_NOFS
);
7176 bio
->bi_private
= dip
;
7177 bio
->bi_end_io
= btrfs_end_dio_bio
;
7178 atomic_inc(&dip
->pending_bios
);
7180 while (bvec
<= (orig_bio
->bi_io_vec
+ orig_bio
->bi_vcnt
- 1)) {
7181 if (unlikely(map_length
< submit_len
+ bvec
->bv_len
||
7182 bio_add_page(bio
, bvec
->bv_page
, bvec
->bv_len
,
7183 bvec
->bv_offset
) < bvec
->bv_len
)) {
7185 * inc the count before we submit the bio so
7186 * we know the end IO handler won't happen before
7187 * we inc the count. Otherwise, the dip might get freed
7188 * before we're done setting it up
7190 atomic_inc(&dip
->pending_bios
);
7191 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
,
7192 file_offset
, skip_sum
,
7196 atomic_dec(&dip
->pending_bios
);
7200 start_sector
+= submit_len
>> 9;
7201 file_offset
+= submit_len
;
7206 bio
= btrfs_dio_bio_alloc(orig_bio
->bi_bdev
,
7207 start_sector
, GFP_NOFS
);
7210 bio
->bi_private
= dip
;
7211 bio
->bi_end_io
= btrfs_end_dio_bio
;
7213 map_length
= orig_bio
->bi_size
;
7214 ret
= btrfs_map_block(root
->fs_info
, rw
,
7216 &map_length
, NULL
, 0);
7222 submit_len
+= bvec
->bv_len
;
7229 ret
= __btrfs_submit_dio_bio(bio
, inode
, rw
, file_offset
, skip_sum
,
7238 * before atomic variable goto zero, we must
7239 * make sure dip->errors is perceived to be set.
7241 smp_mb__before_atomic_dec();
7242 if (atomic_dec_and_test(&dip
->pending_bios
))
7243 bio_io_error(dip
->orig_bio
);
7245 /* bio_end_io() will handle error, so we needn't return it */
7249 static void btrfs_submit_direct(int rw
, struct bio
*bio
, struct inode
*inode
,
7252 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7253 struct btrfs_dio_private
*dip
;
7254 struct bio_vec
*bvec
= bio
->bi_io_vec
;
7256 int write
= rw
& REQ_WRITE
;
7259 skip_sum
= BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
;
7261 dip
= kmalloc(sizeof(*dip
), GFP_NOFS
);
7267 dip
->private = bio
->bi_private
;
7269 dip
->logical_offset
= file_offset
;
7273 dip
->bytes
+= bvec
->bv_len
;
7275 } while (bvec
<= (bio
->bi_io_vec
+ bio
->bi_vcnt
- 1));
7277 dip
->disk_bytenr
= (u64
)bio
->bi_sector
<< 9;
7278 bio
->bi_private
= dip
;
7280 dip
->orig_bio
= bio
;
7281 atomic_set(&dip
->pending_bios
, 0);
7284 bio
->bi_end_io
= btrfs_endio_direct_write
;
7286 bio
->bi_end_io
= btrfs_endio_direct_read
;
7288 ret
= btrfs_submit_direct_hook(rw
, dip
, skip_sum
);
7293 * If this is a write, we need to clean up the reserved space and kill
7294 * the ordered extent.
7297 struct btrfs_ordered_extent
*ordered
;
7298 ordered
= btrfs_lookup_ordered_extent(inode
, file_offset
);
7299 if (!test_bit(BTRFS_ORDERED_PREALLOC
, &ordered
->flags
) &&
7300 !test_bit(BTRFS_ORDERED_NOCOW
, &ordered
->flags
))
7301 btrfs_free_reserved_extent(root
, ordered
->start
,
7303 btrfs_put_ordered_extent(ordered
);
7304 btrfs_put_ordered_extent(ordered
);
7306 bio_endio(bio
, ret
);
7309 static ssize_t
check_direct_IO(struct btrfs_root
*root
, int rw
, struct kiocb
*iocb
,
7310 const struct iovec
*iov
, loff_t offset
,
7311 unsigned long nr_segs
)
7317 unsigned blocksize_mask
= root
->sectorsize
- 1;
7318 ssize_t retval
= -EINVAL
;
7319 loff_t end
= offset
;
7321 if (offset
& blocksize_mask
)
7324 /* Check the memory alignment. Blocks cannot straddle pages */
7325 for (seg
= 0; seg
< nr_segs
; seg
++) {
7326 addr
= (unsigned long)iov
[seg
].iov_base
;
7327 size
= iov
[seg
].iov_len
;
7329 if ((addr
& blocksize_mask
) || (size
& blocksize_mask
))
7332 /* If this is a write we don't need to check anymore */
7337 * Check to make sure we don't have duplicate iov_base's in this
7338 * iovec, if so return EINVAL, otherwise we'll get csum errors
7339 * when reading back.
7341 for (i
= seg
+ 1; i
< nr_segs
; i
++) {
7342 if (iov
[seg
].iov_base
== iov
[i
].iov_base
)
7351 static ssize_t
btrfs_direct_IO(int rw
, struct kiocb
*iocb
,
7352 const struct iovec
*iov
, loff_t offset
,
7353 unsigned long nr_segs
)
7355 struct file
*file
= iocb
->ki_filp
;
7356 struct inode
*inode
= file
->f_mapping
->host
;
7360 bool relock
= false;
7363 if (check_direct_IO(BTRFS_I(inode
)->root
, rw
, iocb
, iov
,
7367 atomic_inc(&inode
->i_dio_count
);
7368 smp_mb__after_atomic_inc();
7371 count
= iov_length(iov
, nr_segs
);
7373 * If the write DIO is beyond the EOF, we need update
7374 * the isize, but it is protected by i_mutex. So we can
7375 * not unlock the i_mutex at this case.
7377 if (offset
+ count
<= inode
->i_size
) {
7378 mutex_unlock(&inode
->i_mutex
);
7381 ret
= btrfs_delalloc_reserve_space(inode
, count
);
7384 } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK
,
7385 &BTRFS_I(inode
)->runtime_flags
))) {
7386 inode_dio_done(inode
);
7387 flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
7391 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
7392 BTRFS_I(inode
)->root
->fs_info
->fs_devices
->latest_bdev
,
7393 iov
, offset
, nr_segs
, btrfs_get_blocks_direct
, NULL
,
7394 btrfs_submit_direct
, flags
);
7396 if (ret
< 0 && ret
!= -EIOCBQUEUED
)
7397 btrfs_delalloc_release_space(inode
, count
);
7398 else if (ret
>= 0 && (size_t)ret
< count
)
7399 btrfs_delalloc_release_space(inode
,
7400 count
- (size_t)ret
);
7402 btrfs_delalloc_release_metadata(inode
, 0);
7406 inode_dio_done(inode
);
7408 mutex_lock(&inode
->i_mutex
);
7413 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
7415 static int btrfs_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
7416 __u64 start
, __u64 len
)
7420 ret
= fiemap_check_flags(fieinfo
, BTRFS_FIEMAP_FLAGS
);
7424 return extent_fiemap(inode
, fieinfo
, start
, len
, btrfs_get_extent_fiemap
);
7427 int btrfs_readpage(struct file
*file
, struct page
*page
)
7429 struct extent_io_tree
*tree
;
7430 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7431 return extent_read_full_page(tree
, page
, btrfs_get_extent
, 0);
7434 static int btrfs_writepage(struct page
*page
, struct writeback_control
*wbc
)
7436 struct extent_io_tree
*tree
;
7439 if (current
->flags
& PF_MEMALLOC
) {
7440 redirty_page_for_writepage(wbc
, page
);
7444 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7445 return extent_write_full_page(tree
, page
, btrfs_get_extent
, wbc
);
7448 static int btrfs_writepages(struct address_space
*mapping
,
7449 struct writeback_control
*wbc
)
7451 struct extent_io_tree
*tree
;
7453 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7454 return extent_writepages(tree
, mapping
, btrfs_get_extent
, wbc
);
7458 btrfs_readpages(struct file
*file
, struct address_space
*mapping
,
7459 struct list_head
*pages
, unsigned nr_pages
)
7461 struct extent_io_tree
*tree
;
7462 tree
= &BTRFS_I(mapping
->host
)->io_tree
;
7463 return extent_readpages(tree
, mapping
, pages
, nr_pages
,
7466 static int __btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7468 struct extent_io_tree
*tree
;
7469 struct extent_map_tree
*map
;
7472 tree
= &BTRFS_I(page
->mapping
->host
)->io_tree
;
7473 map
= &BTRFS_I(page
->mapping
->host
)->extent_tree
;
7474 ret
= try_release_extent_mapping(map
, tree
, page
, gfp_flags
);
7476 ClearPagePrivate(page
);
7477 set_page_private(page
, 0);
7478 page_cache_release(page
);
7483 static int btrfs_releasepage(struct page
*page
, gfp_t gfp_flags
)
7485 if (PageWriteback(page
) || PageDirty(page
))
7487 return __btrfs_releasepage(page
, gfp_flags
& GFP_NOFS
);
7490 static void btrfs_invalidatepage(struct page
*page
, unsigned long offset
)
7492 struct inode
*inode
= page
->mapping
->host
;
7493 struct extent_io_tree
*tree
;
7494 struct btrfs_ordered_extent
*ordered
;
7495 struct extent_state
*cached_state
= NULL
;
7496 u64 page_start
= page_offset(page
);
7497 u64 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7500 * we have the page locked, so new writeback can't start,
7501 * and the dirty bit won't be cleared while we are here.
7503 * Wait for IO on this page so that we can safely clear
7504 * the PagePrivate2 bit and do ordered accounting
7506 wait_on_page_writeback(page
);
7508 tree
= &BTRFS_I(inode
)->io_tree
;
7510 btrfs_releasepage(page
, GFP_NOFS
);
7513 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
7514 ordered
= btrfs_lookup_ordered_extent(inode
, page_offset(page
));
7517 * IO on this page will never be started, so we need
7518 * to account for any ordered extents now
7520 clear_extent_bit(tree
, page_start
, page_end
,
7521 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7522 EXTENT_LOCKED
| EXTENT_DO_ACCOUNTING
|
7523 EXTENT_DEFRAG
, 1, 0, &cached_state
, GFP_NOFS
);
7525 * whoever cleared the private bit is responsible
7526 * for the finish_ordered_io
7528 if (TestClearPagePrivate2(page
) &&
7529 btrfs_dec_test_ordered_pending(inode
, &ordered
, page_start
,
7530 PAGE_CACHE_SIZE
, 1)) {
7531 btrfs_finish_ordered_io(ordered
);
7533 btrfs_put_ordered_extent(ordered
);
7534 cached_state
= NULL
;
7535 lock_extent_bits(tree
, page_start
, page_end
, 0, &cached_state
);
7537 clear_extent_bit(tree
, page_start
, page_end
,
7538 EXTENT_LOCKED
| EXTENT_DIRTY
| EXTENT_DELALLOC
|
7539 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
, 1, 1,
7540 &cached_state
, GFP_NOFS
);
7541 __btrfs_releasepage(page
, GFP_NOFS
);
7543 ClearPageChecked(page
);
7544 if (PagePrivate(page
)) {
7545 ClearPagePrivate(page
);
7546 set_page_private(page
, 0);
7547 page_cache_release(page
);
7552 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7553 * called from a page fault handler when a page is first dirtied. Hence we must
7554 * be careful to check for EOF conditions here. We set the page up correctly
7555 * for a written page which means we get ENOSPC checking when writing into
7556 * holes and correct delalloc and unwritten extent mapping on filesystems that
7557 * support these features.
7559 * We are not allowed to take the i_mutex here so we have to play games to
7560 * protect against truncate races as the page could now be beyond EOF. Because
7561 * vmtruncate() writes the inode size before removing pages, once we have the
7562 * page lock we can determine safely if the page is beyond EOF. If it is not
7563 * beyond EOF, then the page is guaranteed safe against truncation until we
7566 int btrfs_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
7568 struct page
*page
= vmf
->page
;
7569 struct inode
*inode
= file_inode(vma
->vm_file
);
7570 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7571 struct extent_io_tree
*io_tree
= &BTRFS_I(inode
)->io_tree
;
7572 struct btrfs_ordered_extent
*ordered
;
7573 struct extent_state
*cached_state
= NULL
;
7575 unsigned long zero_start
;
7582 sb_start_pagefault(inode
->i_sb
);
7583 ret
= btrfs_delalloc_reserve_space(inode
, PAGE_CACHE_SIZE
);
7585 ret
= file_update_time(vma
->vm_file
);
7591 else /* -ENOSPC, -EIO, etc */
7592 ret
= VM_FAULT_SIGBUS
;
7598 ret
= VM_FAULT_NOPAGE
; /* make the VM retry the fault */
7601 size
= i_size_read(inode
);
7602 page_start
= page_offset(page
);
7603 page_end
= page_start
+ PAGE_CACHE_SIZE
- 1;
7605 if ((page
->mapping
!= inode
->i_mapping
) ||
7606 (page_start
>= size
)) {
7607 /* page got truncated out from underneath us */
7610 wait_on_page_writeback(page
);
7612 lock_extent_bits(io_tree
, page_start
, page_end
, 0, &cached_state
);
7613 set_page_extent_mapped(page
);
7616 * we can't set the delalloc bits if there are pending ordered
7617 * extents. Drop our locks and wait for them to finish
7619 ordered
= btrfs_lookup_ordered_extent(inode
, page_start
);
7621 unlock_extent_cached(io_tree
, page_start
, page_end
,
7622 &cached_state
, GFP_NOFS
);
7624 btrfs_start_ordered_extent(inode
, ordered
, 1);
7625 btrfs_put_ordered_extent(ordered
);
7630 * XXX - page_mkwrite gets called every time the page is dirtied, even
7631 * if it was already dirty, so for space accounting reasons we need to
7632 * clear any delalloc bits for the range we are fixing to save. There
7633 * is probably a better way to do this, but for now keep consistent with
7634 * prepare_pages in the normal write path.
7636 clear_extent_bit(&BTRFS_I(inode
)->io_tree
, page_start
, page_end
,
7637 EXTENT_DIRTY
| EXTENT_DELALLOC
|
7638 EXTENT_DO_ACCOUNTING
| EXTENT_DEFRAG
,
7639 0, 0, &cached_state
, GFP_NOFS
);
7641 ret
= btrfs_set_extent_delalloc(inode
, page_start
, page_end
,
7644 unlock_extent_cached(io_tree
, page_start
, page_end
,
7645 &cached_state
, GFP_NOFS
);
7646 ret
= VM_FAULT_SIGBUS
;
7651 /* page is wholly or partially inside EOF */
7652 if (page_start
+ PAGE_CACHE_SIZE
> size
)
7653 zero_start
= size
& ~PAGE_CACHE_MASK
;
7655 zero_start
= PAGE_CACHE_SIZE
;
7657 if (zero_start
!= PAGE_CACHE_SIZE
) {
7659 memset(kaddr
+ zero_start
, 0, PAGE_CACHE_SIZE
- zero_start
);
7660 flush_dcache_page(page
);
7663 ClearPageChecked(page
);
7664 set_page_dirty(page
);
7665 SetPageUptodate(page
);
7667 BTRFS_I(inode
)->last_trans
= root
->fs_info
->generation
;
7668 BTRFS_I(inode
)->last_sub_trans
= BTRFS_I(inode
)->root
->log_transid
;
7669 BTRFS_I(inode
)->last_log_commit
= BTRFS_I(inode
)->root
->last_log_commit
;
7671 unlock_extent_cached(io_tree
, page_start
, page_end
, &cached_state
, GFP_NOFS
);
7675 sb_end_pagefault(inode
->i_sb
);
7676 return VM_FAULT_LOCKED
;
7680 btrfs_delalloc_release_space(inode
, PAGE_CACHE_SIZE
);
7682 sb_end_pagefault(inode
->i_sb
);
7686 static int btrfs_truncate(struct inode
*inode
)
7688 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7689 struct btrfs_block_rsv
*rsv
;
7692 struct btrfs_trans_handle
*trans
;
7693 u64 mask
= root
->sectorsize
- 1;
7694 u64 min_size
= btrfs_calc_trunc_metadata_size(root
, 1);
7696 ret
= btrfs_truncate_page(inode
, inode
->i_size
, 0, 0);
7700 btrfs_wait_ordered_range(inode
, inode
->i_size
& (~mask
), (u64
)-1);
7701 btrfs_ordered_update_i_size(inode
, inode
->i_size
, NULL
);
7704 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
7705 * 3 things going on here
7707 * 1) We need to reserve space for our orphan item and the space to
7708 * delete our orphan item. Lord knows we don't want to have a dangling
7709 * orphan item because we didn't reserve space to remove it.
7711 * 2) We need to reserve space to update our inode.
7713 * 3) We need to have something to cache all the space that is going to
7714 * be free'd up by the truncate operation, but also have some slack
7715 * space reserved in case it uses space during the truncate (thank you
7716 * very much snapshotting).
7718 * And we need these to all be seperate. The fact is we can use alot of
7719 * space doing the truncate, and we have no earthly idea how much space
7720 * we will use, so we need the truncate reservation to be seperate so it
7721 * doesn't end up using space reserved for updating the inode or
7722 * removing the orphan item. We also need to be able to stop the
7723 * transaction and start a new one, which means we need to be able to
7724 * update the inode several times, and we have no idea of knowing how
7725 * many times that will be, so we can't just reserve 1 item for the
7726 * entirety of the opration, so that has to be done seperately as well.
7727 * Then there is the orphan item, which does indeed need to be held on
7728 * to for the whole operation, and we need nobody to touch this reserved
7729 * space except the orphan code.
7731 * So that leaves us with
7733 * 1) root->orphan_block_rsv - for the orphan deletion.
7734 * 2) rsv - for the truncate reservation, which we will steal from the
7735 * transaction reservation.
7736 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7737 * updating the inode.
7739 rsv
= btrfs_alloc_block_rsv(root
, BTRFS_BLOCK_RSV_TEMP
);
7742 rsv
->size
= min_size
;
7746 * 1 for the truncate slack space
7747 * 1 for updating the inode.
7749 trans
= btrfs_start_transaction(root
, 2);
7750 if (IS_ERR(trans
)) {
7751 err
= PTR_ERR(trans
);
7755 /* Migrate the slack space for the truncate to our reserve */
7756 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
, rsv
,
7761 * setattr is responsible for setting the ordered_data_close flag,
7762 * but that is only tested during the last file release. That
7763 * could happen well after the next commit, leaving a great big
7764 * window where new writes may get lost if someone chooses to write
7765 * to this file after truncating to zero
7767 * The inode doesn't have any dirty data here, and so if we commit
7768 * this is a noop. If someone immediately starts writing to the inode
7769 * it is very likely we'll catch some of their writes in this
7770 * transaction, and the commit will find this file on the ordered
7771 * data list with good things to send down.
7773 * This is a best effort solution, there is still a window where
7774 * using truncate to replace the contents of the file will
7775 * end up with a zero length file after a crash.
7777 if (inode
->i_size
== 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE
,
7778 &BTRFS_I(inode
)->runtime_flags
))
7779 btrfs_add_ordered_operation(trans
, root
, inode
);
7782 * So if we truncate and then write and fsync we normally would just
7783 * write the extents that changed, which is a problem if we need to
7784 * first truncate that entire inode. So set this flag so we write out
7785 * all of the extents in the inode to the sync log so we're completely
7788 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
, &BTRFS_I(inode
)->runtime_flags
);
7789 trans
->block_rsv
= rsv
;
7792 ret
= btrfs_truncate_inode_items(trans
, root
, inode
,
7794 BTRFS_EXTENT_DATA_KEY
);
7795 if (ret
!= -ENOSPC
) {
7800 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7801 ret
= btrfs_update_inode(trans
, root
, inode
);
7807 btrfs_end_transaction(trans
, root
);
7808 btrfs_btree_balance_dirty(root
);
7810 trans
= btrfs_start_transaction(root
, 2);
7811 if (IS_ERR(trans
)) {
7812 ret
= err
= PTR_ERR(trans
);
7817 ret
= btrfs_block_rsv_migrate(&root
->fs_info
->trans_block_rsv
,
7819 BUG_ON(ret
); /* shouldn't happen */
7820 trans
->block_rsv
= rsv
;
7823 if (ret
== 0 && inode
->i_nlink
> 0) {
7824 trans
->block_rsv
= root
->orphan_block_rsv
;
7825 ret
= btrfs_orphan_del(trans
, inode
);
7831 trans
->block_rsv
= &root
->fs_info
->trans_block_rsv
;
7832 ret
= btrfs_update_inode(trans
, root
, inode
);
7836 ret
= btrfs_end_transaction(trans
, root
);
7837 btrfs_btree_balance_dirty(root
);
7841 btrfs_free_block_rsv(root
, rsv
);
7850 * create a new subvolume directory/inode (helper for the ioctl).
7852 int btrfs_create_subvol_root(struct btrfs_trans_handle
*trans
,
7853 struct btrfs_root
*new_root
, u64 new_dirid
)
7855 struct inode
*inode
;
7859 inode
= btrfs_new_inode(trans
, new_root
, NULL
, "..", 2,
7860 new_dirid
, new_dirid
,
7861 S_IFDIR
| (~current_umask() & S_IRWXUGO
),
7864 return PTR_ERR(inode
);
7865 inode
->i_op
= &btrfs_dir_inode_operations
;
7866 inode
->i_fop
= &btrfs_dir_file_operations
;
7868 set_nlink(inode
, 1);
7869 btrfs_i_size_write(inode
, 0);
7871 err
= btrfs_update_inode(trans
, new_root
, inode
);
7877 struct inode
*btrfs_alloc_inode(struct super_block
*sb
)
7879 struct btrfs_inode
*ei
;
7880 struct inode
*inode
;
7882 ei
= kmem_cache_alloc(btrfs_inode_cachep
, GFP_NOFS
);
7889 ei
->last_sub_trans
= 0;
7890 ei
->logged_trans
= 0;
7891 ei
->delalloc_bytes
= 0;
7892 ei
->disk_i_size
= 0;
7895 ei
->index_cnt
= (u64
)-1;
7896 ei
->last_unlink_trans
= 0;
7897 ei
->last_log_commit
= 0;
7899 spin_lock_init(&ei
->lock
);
7900 ei
->outstanding_extents
= 0;
7901 ei
->reserved_extents
= 0;
7903 ei
->runtime_flags
= 0;
7904 ei
->force_compress
= BTRFS_COMPRESS_NONE
;
7906 ei
->delayed_node
= NULL
;
7908 inode
= &ei
->vfs_inode
;
7909 extent_map_tree_init(&ei
->extent_tree
);
7910 extent_io_tree_init(&ei
->io_tree
, &inode
->i_data
);
7911 extent_io_tree_init(&ei
->io_failure_tree
, &inode
->i_data
);
7912 ei
->io_tree
.track_uptodate
= 1;
7913 ei
->io_failure_tree
.track_uptodate
= 1;
7914 atomic_set(&ei
->sync_writers
, 0);
7915 mutex_init(&ei
->log_mutex
);
7916 mutex_init(&ei
->delalloc_mutex
);
7917 btrfs_ordered_inode_tree_init(&ei
->ordered_tree
);
7918 INIT_LIST_HEAD(&ei
->delalloc_inodes
);
7919 INIT_LIST_HEAD(&ei
->ordered_operations
);
7920 RB_CLEAR_NODE(&ei
->rb_node
);
7925 static void btrfs_i_callback(struct rcu_head
*head
)
7927 struct inode
*inode
= container_of(head
, struct inode
, i_rcu
);
7928 kmem_cache_free(btrfs_inode_cachep
, BTRFS_I(inode
));
7931 void btrfs_destroy_inode(struct inode
*inode
)
7933 struct btrfs_ordered_extent
*ordered
;
7934 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7936 WARN_ON(!hlist_empty(&inode
->i_dentry
));
7937 WARN_ON(inode
->i_data
.nrpages
);
7938 WARN_ON(BTRFS_I(inode
)->outstanding_extents
);
7939 WARN_ON(BTRFS_I(inode
)->reserved_extents
);
7940 WARN_ON(BTRFS_I(inode
)->delalloc_bytes
);
7941 WARN_ON(BTRFS_I(inode
)->csum_bytes
);
7944 * This can happen where we create an inode, but somebody else also
7945 * created the same inode and we need to destroy the one we already
7952 * Make sure we're properly removed from the ordered operation
7956 if (!list_empty(&BTRFS_I(inode
)->ordered_operations
)) {
7957 spin_lock(&root
->fs_info
->ordered_extent_lock
);
7958 list_del_init(&BTRFS_I(inode
)->ordered_operations
);
7959 spin_unlock(&root
->fs_info
->ordered_extent_lock
);
7962 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM
,
7963 &BTRFS_I(inode
)->runtime_flags
)) {
7964 btrfs_info(root
->fs_info
, "inode %llu still on the orphan list",
7965 (unsigned long long)btrfs_ino(inode
));
7966 atomic_dec(&root
->orphan_inodes
);
7970 ordered
= btrfs_lookup_first_ordered_extent(inode
, (u64
)-1);
7974 btrfs_err(root
->fs_info
, "found ordered extent %llu %llu on inode cleanup",
7975 (unsigned long long)ordered
->file_offset
,
7976 (unsigned long long)ordered
->len
);
7977 btrfs_remove_ordered_extent(inode
, ordered
);
7978 btrfs_put_ordered_extent(ordered
);
7979 btrfs_put_ordered_extent(ordered
);
7982 inode_tree_del(inode
);
7983 btrfs_drop_extent_cache(inode
, 0, (u64
)-1, 0);
7985 call_rcu(&inode
->i_rcu
, btrfs_i_callback
);
7988 int btrfs_drop_inode(struct inode
*inode
)
7990 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
7992 /* the snap/subvol tree is on deleting */
7993 if (btrfs_root_refs(&root
->root_item
) == 0 &&
7994 root
!= root
->fs_info
->tree_root
)
7997 return generic_drop_inode(inode
);
8000 static void init_once(void *foo
)
8002 struct btrfs_inode
*ei
= (struct btrfs_inode
*) foo
;
8004 inode_init_once(&ei
->vfs_inode
);
8007 void btrfs_destroy_cachep(void)
8010 * Make sure all delayed rcu free inodes are flushed before we
8014 if (btrfs_inode_cachep
)
8015 kmem_cache_destroy(btrfs_inode_cachep
);
8016 if (btrfs_trans_handle_cachep
)
8017 kmem_cache_destroy(btrfs_trans_handle_cachep
);
8018 if (btrfs_transaction_cachep
)
8019 kmem_cache_destroy(btrfs_transaction_cachep
);
8020 if (btrfs_path_cachep
)
8021 kmem_cache_destroy(btrfs_path_cachep
);
8022 if (btrfs_free_space_cachep
)
8023 kmem_cache_destroy(btrfs_free_space_cachep
);
8024 if (btrfs_delalloc_work_cachep
)
8025 kmem_cache_destroy(btrfs_delalloc_work_cachep
);
8028 int btrfs_init_cachep(void)
8030 btrfs_inode_cachep
= kmem_cache_create("btrfs_inode",
8031 sizeof(struct btrfs_inode
), 0,
8032 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, init_once
);
8033 if (!btrfs_inode_cachep
)
8036 btrfs_trans_handle_cachep
= kmem_cache_create("btrfs_trans_handle",
8037 sizeof(struct btrfs_trans_handle
), 0,
8038 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8039 if (!btrfs_trans_handle_cachep
)
8042 btrfs_transaction_cachep
= kmem_cache_create("btrfs_transaction",
8043 sizeof(struct btrfs_transaction
), 0,
8044 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8045 if (!btrfs_transaction_cachep
)
8048 btrfs_path_cachep
= kmem_cache_create("btrfs_path",
8049 sizeof(struct btrfs_path
), 0,
8050 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8051 if (!btrfs_path_cachep
)
8054 btrfs_free_space_cachep
= kmem_cache_create("btrfs_free_space",
8055 sizeof(struct btrfs_free_space
), 0,
8056 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
, NULL
);
8057 if (!btrfs_free_space_cachep
)
8060 btrfs_delalloc_work_cachep
= kmem_cache_create("btrfs_delalloc_work",
8061 sizeof(struct btrfs_delalloc_work
), 0,
8062 SLAB_RECLAIM_ACCOUNT
| SLAB_MEM_SPREAD
,
8064 if (!btrfs_delalloc_work_cachep
)
8069 btrfs_destroy_cachep();
8073 static int btrfs_getattr(struct vfsmount
*mnt
,
8074 struct dentry
*dentry
, struct kstat
*stat
)
8077 struct inode
*inode
= dentry
->d_inode
;
8078 u32 blocksize
= inode
->i_sb
->s_blocksize
;
8080 generic_fillattr(inode
, stat
);
8081 stat
->dev
= BTRFS_I(inode
)->root
->anon_dev
;
8082 stat
->blksize
= PAGE_CACHE_SIZE
;
8084 spin_lock(&BTRFS_I(inode
)->lock
);
8085 delalloc_bytes
= BTRFS_I(inode
)->delalloc_bytes
;
8086 spin_unlock(&BTRFS_I(inode
)->lock
);
8087 stat
->blocks
= (ALIGN(inode_get_bytes(inode
), blocksize
) +
8088 ALIGN(delalloc_bytes
, blocksize
)) >> 9;
8092 static int btrfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
8093 struct inode
*new_dir
, struct dentry
*new_dentry
)
8095 struct btrfs_trans_handle
*trans
;
8096 struct btrfs_root
*root
= BTRFS_I(old_dir
)->root
;
8097 struct btrfs_root
*dest
= BTRFS_I(new_dir
)->root
;
8098 struct inode
*new_inode
= new_dentry
->d_inode
;
8099 struct inode
*old_inode
= old_dentry
->d_inode
;
8100 struct timespec ctime
= CURRENT_TIME
;
8104 u64 old_ino
= btrfs_ino(old_inode
);
8106 if (btrfs_ino(new_dir
) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)
8109 /* we only allow rename subvolume link between subvolumes */
8110 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
&& root
!= dest
)
8113 if (old_ino
== BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
||
8114 (new_inode
&& btrfs_ino(new_inode
) == BTRFS_FIRST_FREE_OBJECTID
))
8117 if (S_ISDIR(old_inode
->i_mode
) && new_inode
&&
8118 new_inode
->i_size
> BTRFS_EMPTY_DIR_SIZE
)
8122 /* check for collisions, even if the name isn't there */
8123 ret
= btrfs_check_dir_item_collision(root
, new_dir
->i_ino
,
8124 new_dentry
->d_name
.name
,
8125 new_dentry
->d_name
.len
);
8128 if (ret
== -EEXIST
) {
8130 * eexist without a new_inode */
8136 /* maybe -EOVERFLOW */
8143 * we're using rename to replace one file with another.
8144 * and the replacement file is large. Start IO on it now so
8145 * we don't add too much work to the end of the transaction
8147 if (new_inode
&& S_ISREG(old_inode
->i_mode
) && new_inode
->i_size
&&
8148 old_inode
->i_size
> BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT
)
8149 filemap_flush(old_inode
->i_mapping
);
8151 /* close the racy window with snapshot create/destroy ioctl */
8152 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8153 down_read(&root
->fs_info
->subvol_sem
);
8155 * We want to reserve the absolute worst case amount of items. So if
8156 * both inodes are subvols and we need to unlink them then that would
8157 * require 4 item modifications, but if they are both normal inodes it
8158 * would require 5 item modifications, so we'll assume their normal
8159 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8160 * should cover the worst case number of items we'll modify.
8162 trans
= btrfs_start_transaction(root
, 11);
8163 if (IS_ERR(trans
)) {
8164 ret
= PTR_ERR(trans
);
8169 btrfs_record_root_in_trans(trans
, dest
);
8171 ret
= btrfs_set_inode_index(new_dir
, &index
);
8175 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8176 /* force full log commit if subvolume involved. */
8177 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
8179 ret
= btrfs_insert_inode_ref(trans
, dest
,
8180 new_dentry
->d_name
.name
,
8181 new_dentry
->d_name
.len
,
8183 btrfs_ino(new_dir
), index
);
8187 * this is an ugly little race, but the rename is required
8188 * to make sure that if we crash, the inode is either at the
8189 * old name or the new one. pinning the log transaction lets
8190 * us make sure we don't allow a log commit to come in after
8191 * we unlink the name but before we add the new name back in.
8193 btrfs_pin_log_trans(root
);
8196 * make sure the inode gets flushed if it is replacing
8199 if (new_inode
&& new_inode
->i_size
&& S_ISREG(old_inode
->i_mode
))
8200 btrfs_add_ordered_operation(trans
, root
, old_inode
);
8202 inode_inc_iversion(old_dir
);
8203 inode_inc_iversion(new_dir
);
8204 inode_inc_iversion(old_inode
);
8205 old_dir
->i_ctime
= old_dir
->i_mtime
= ctime
;
8206 new_dir
->i_ctime
= new_dir
->i_mtime
= ctime
;
8207 old_inode
->i_ctime
= ctime
;
8209 if (old_dentry
->d_parent
!= new_dentry
->d_parent
)
8210 btrfs_record_unlink_dir(trans
, old_dir
, old_inode
, 1);
8212 if (unlikely(old_ino
== BTRFS_FIRST_FREE_OBJECTID
)) {
8213 root_objectid
= BTRFS_I(old_inode
)->root
->root_key
.objectid
;
8214 ret
= btrfs_unlink_subvol(trans
, root
, old_dir
, root_objectid
,
8215 old_dentry
->d_name
.name
,
8216 old_dentry
->d_name
.len
);
8218 ret
= __btrfs_unlink_inode(trans
, root
, old_dir
,
8219 old_dentry
->d_inode
,
8220 old_dentry
->d_name
.name
,
8221 old_dentry
->d_name
.len
);
8223 ret
= btrfs_update_inode(trans
, root
, old_inode
);
8226 btrfs_abort_transaction(trans
, root
, ret
);
8231 inode_inc_iversion(new_inode
);
8232 new_inode
->i_ctime
= CURRENT_TIME
;
8233 if (unlikely(btrfs_ino(new_inode
) ==
8234 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID
)) {
8235 root_objectid
= BTRFS_I(new_inode
)->location
.objectid
;
8236 ret
= btrfs_unlink_subvol(trans
, dest
, new_dir
,
8238 new_dentry
->d_name
.name
,
8239 new_dentry
->d_name
.len
);
8240 BUG_ON(new_inode
->i_nlink
== 0);
8242 ret
= btrfs_unlink_inode(trans
, dest
, new_dir
,
8243 new_dentry
->d_inode
,
8244 new_dentry
->d_name
.name
,
8245 new_dentry
->d_name
.len
);
8247 if (!ret
&& new_inode
->i_nlink
== 0) {
8248 ret
= btrfs_orphan_add(trans
, new_dentry
->d_inode
);
8252 btrfs_abort_transaction(trans
, root
, ret
);
8257 ret
= btrfs_add_link(trans
, new_dir
, old_inode
,
8258 new_dentry
->d_name
.name
,
8259 new_dentry
->d_name
.len
, 0, index
);
8261 btrfs_abort_transaction(trans
, root
, ret
);
8265 if (old_ino
!= BTRFS_FIRST_FREE_OBJECTID
) {
8266 struct dentry
*parent
= new_dentry
->d_parent
;
8267 btrfs_log_new_name(trans
, old_inode
, old_dir
, parent
);
8268 btrfs_end_log_trans(root
);
8271 btrfs_end_transaction(trans
, root
);
8273 if (old_ino
== BTRFS_FIRST_FREE_OBJECTID
)
8274 up_read(&root
->fs_info
->subvol_sem
);
8279 static void btrfs_run_delalloc_work(struct btrfs_work
*work
)
8281 struct btrfs_delalloc_work
*delalloc_work
;
8283 delalloc_work
= container_of(work
, struct btrfs_delalloc_work
,
8285 if (delalloc_work
->wait
)
8286 btrfs_wait_ordered_range(delalloc_work
->inode
, 0, (u64
)-1);
8288 filemap_flush(delalloc_work
->inode
->i_mapping
);
8290 if (delalloc_work
->delay_iput
)
8291 btrfs_add_delayed_iput(delalloc_work
->inode
);
8293 iput(delalloc_work
->inode
);
8294 complete(&delalloc_work
->completion
);
8297 struct btrfs_delalloc_work
*btrfs_alloc_delalloc_work(struct inode
*inode
,
8298 int wait
, int delay_iput
)
8300 struct btrfs_delalloc_work
*work
;
8302 work
= kmem_cache_zalloc(btrfs_delalloc_work_cachep
, GFP_NOFS
);
8306 init_completion(&work
->completion
);
8307 INIT_LIST_HEAD(&work
->list
);
8308 work
->inode
= inode
;
8310 work
->delay_iput
= delay_iput
;
8311 work
->work
.func
= btrfs_run_delalloc_work
;
8316 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work
*work
)
8318 wait_for_completion(&work
->completion
);
8319 kmem_cache_free(btrfs_delalloc_work_cachep
, work
);
8323 * some fairly slow code that needs optimization. This walks the list
8324 * of all the inodes with pending delalloc and forces them to disk.
8326 int btrfs_start_delalloc_inodes(struct btrfs_root
*root
, int delay_iput
)
8328 struct btrfs_inode
*binode
;
8329 struct inode
*inode
;
8330 struct btrfs_delalloc_work
*work
, *next
;
8331 struct list_head works
;
8332 struct list_head splice
;
8335 if (root
->fs_info
->sb
->s_flags
& MS_RDONLY
)
8338 INIT_LIST_HEAD(&works
);
8339 INIT_LIST_HEAD(&splice
);
8341 spin_lock(&root
->fs_info
->delalloc_lock
);
8342 list_splice_init(&root
->fs_info
->delalloc_inodes
, &splice
);
8343 while (!list_empty(&splice
)) {
8344 binode
= list_entry(splice
.next
, struct btrfs_inode
,
8347 list_del_init(&binode
->delalloc_inodes
);
8349 inode
= igrab(&binode
->vfs_inode
);
8351 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST
,
8352 &binode
->runtime_flags
);
8356 list_add_tail(&binode
->delalloc_inodes
,
8357 &root
->fs_info
->delalloc_inodes
);
8358 spin_unlock(&root
->fs_info
->delalloc_lock
);
8360 work
= btrfs_alloc_delalloc_work(inode
, 0, delay_iput
);
8361 if (unlikely(!work
)) {
8365 list_add_tail(&work
->list
, &works
);
8366 btrfs_queue_worker(&root
->fs_info
->flush_workers
,
8370 spin_lock(&root
->fs_info
->delalloc_lock
);
8372 spin_unlock(&root
->fs_info
->delalloc_lock
);
8374 list_for_each_entry_safe(work
, next
, &works
, list
) {
8375 list_del_init(&work
->list
);
8376 btrfs_wait_and_free_delalloc_work(work
);
8379 /* the filemap_flush will queue IO into the worker threads, but
8380 * we have to make sure the IO is actually started and that
8381 * ordered extents get created before we return
8383 atomic_inc(&root
->fs_info
->async_submit_draining
);
8384 while (atomic_read(&root
->fs_info
->nr_async_submits
) ||
8385 atomic_read(&root
->fs_info
->async_delalloc_pages
)) {
8386 wait_event(root
->fs_info
->async_submit_wait
,
8387 (atomic_read(&root
->fs_info
->nr_async_submits
) == 0 &&
8388 atomic_read(&root
->fs_info
->async_delalloc_pages
) == 0));
8390 atomic_dec(&root
->fs_info
->async_submit_draining
);
8393 list_for_each_entry_safe(work
, next
, &works
, list
) {
8394 list_del_init(&work
->list
);
8395 btrfs_wait_and_free_delalloc_work(work
);
8398 if (!list_empty_careful(&splice
)) {
8399 spin_lock(&root
->fs_info
->delalloc_lock
);
8400 list_splice_tail(&splice
, &root
->fs_info
->delalloc_inodes
);
8401 spin_unlock(&root
->fs_info
->delalloc_lock
);
8406 static int btrfs_symlink(struct inode
*dir
, struct dentry
*dentry
,
8407 const char *symname
)
8409 struct btrfs_trans_handle
*trans
;
8410 struct btrfs_root
*root
= BTRFS_I(dir
)->root
;
8411 struct btrfs_path
*path
;
8412 struct btrfs_key key
;
8413 struct inode
*inode
= NULL
;
8421 struct btrfs_file_extent_item
*ei
;
8422 struct extent_buffer
*leaf
;
8424 name_len
= strlen(symname
) + 1;
8425 if (name_len
> BTRFS_MAX_INLINE_DATA_SIZE(root
))
8426 return -ENAMETOOLONG
;
8429 * 2 items for inode item and ref
8430 * 2 items for dir items
8431 * 1 item for xattr if selinux is on
8433 trans
= btrfs_start_transaction(root
, 5);
8435 return PTR_ERR(trans
);
8437 err
= btrfs_find_free_ino(root
, &objectid
);
8441 inode
= btrfs_new_inode(trans
, root
, dir
, dentry
->d_name
.name
,
8442 dentry
->d_name
.len
, btrfs_ino(dir
), objectid
,
8443 S_IFLNK
|S_IRWXUGO
, &index
);
8444 if (IS_ERR(inode
)) {
8445 err
= PTR_ERR(inode
);
8449 err
= btrfs_init_inode_security(trans
, inode
, dir
, &dentry
->d_name
);
8456 * If the active LSM wants to access the inode during
8457 * d_instantiate it needs these. Smack checks to see
8458 * if the filesystem supports xattrs by looking at the
8461 inode
->i_fop
= &btrfs_file_operations
;
8462 inode
->i_op
= &btrfs_file_inode_operations
;
8464 err
= btrfs_add_nondir(trans
, dir
, dentry
, inode
, 0, index
);
8468 inode
->i_mapping
->a_ops
= &btrfs_aops
;
8469 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8470 BTRFS_I(inode
)->io_tree
.ops
= &btrfs_extent_io_ops
;
8475 path
= btrfs_alloc_path();
8481 key
.objectid
= btrfs_ino(inode
);
8483 btrfs_set_key_type(&key
, BTRFS_EXTENT_DATA_KEY
);
8484 datasize
= btrfs_file_extent_calc_inline_size(name_len
);
8485 err
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
8489 btrfs_free_path(path
);
8492 leaf
= path
->nodes
[0];
8493 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
8494 struct btrfs_file_extent_item
);
8495 btrfs_set_file_extent_generation(leaf
, ei
, trans
->transid
);
8496 btrfs_set_file_extent_type(leaf
, ei
,
8497 BTRFS_FILE_EXTENT_INLINE
);
8498 btrfs_set_file_extent_encryption(leaf
, ei
, 0);
8499 btrfs_set_file_extent_compression(leaf
, ei
, 0);
8500 btrfs_set_file_extent_other_encoding(leaf
, ei
, 0);
8501 btrfs_set_file_extent_ram_bytes(leaf
, ei
, name_len
);
8503 ptr
= btrfs_file_extent_inline_start(ei
);
8504 write_extent_buffer(leaf
, symname
, ptr
, name_len
);
8505 btrfs_mark_buffer_dirty(leaf
);
8506 btrfs_free_path(path
);
8508 inode
->i_op
= &btrfs_symlink_inode_operations
;
8509 inode
->i_mapping
->a_ops
= &btrfs_symlink_aops
;
8510 inode
->i_mapping
->backing_dev_info
= &root
->fs_info
->bdi
;
8511 inode_set_bytes(inode
, name_len
);
8512 btrfs_i_size_write(inode
, name_len
- 1);
8513 err
= btrfs_update_inode(trans
, root
, inode
);
8519 d_instantiate(dentry
, inode
);
8520 btrfs_end_transaction(trans
, root
);
8522 inode_dec_link_count(inode
);
8525 btrfs_btree_balance_dirty(root
);
8529 static int __btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8530 u64 start
, u64 num_bytes
, u64 min_size
,
8531 loff_t actual_len
, u64
*alloc_hint
,
8532 struct btrfs_trans_handle
*trans
)
8534 struct extent_map_tree
*em_tree
= &BTRFS_I(inode
)->extent_tree
;
8535 struct extent_map
*em
;
8536 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8537 struct btrfs_key ins
;
8538 u64 cur_offset
= start
;
8542 bool own_trans
= true;
8546 while (num_bytes
> 0) {
8548 trans
= btrfs_start_transaction(root
, 3);
8549 if (IS_ERR(trans
)) {
8550 ret
= PTR_ERR(trans
);
8555 cur_bytes
= min(num_bytes
, 256ULL * 1024 * 1024);
8556 cur_bytes
= max(cur_bytes
, min_size
);
8557 ret
= btrfs_reserve_extent(trans
, root
, cur_bytes
,
8558 min_size
, 0, *alloc_hint
, &ins
, 1);
8561 btrfs_end_transaction(trans
, root
);
8565 ret
= insert_reserved_file_extent(trans
, inode
,
8566 cur_offset
, ins
.objectid
,
8567 ins
.offset
, ins
.offset
,
8568 ins
.offset
, 0, 0, 0,
8569 BTRFS_FILE_EXTENT_PREALLOC
);
8571 btrfs_abort_transaction(trans
, root
, ret
);
8573 btrfs_end_transaction(trans
, root
);
8576 btrfs_drop_extent_cache(inode
, cur_offset
,
8577 cur_offset
+ ins
.offset
-1, 0);
8579 em
= alloc_extent_map();
8581 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC
,
8582 &BTRFS_I(inode
)->runtime_flags
);
8586 em
->start
= cur_offset
;
8587 em
->orig_start
= cur_offset
;
8588 em
->len
= ins
.offset
;
8589 em
->block_start
= ins
.objectid
;
8590 em
->block_len
= ins
.offset
;
8591 em
->orig_block_len
= ins
.offset
;
8592 em
->ram_bytes
= ins
.offset
;
8593 em
->bdev
= root
->fs_info
->fs_devices
->latest_bdev
;
8594 set_bit(EXTENT_FLAG_PREALLOC
, &em
->flags
);
8595 em
->generation
= trans
->transid
;
8598 write_lock(&em_tree
->lock
);
8599 ret
= add_extent_mapping(em_tree
, em
, 1);
8600 write_unlock(&em_tree
->lock
);
8603 btrfs_drop_extent_cache(inode
, cur_offset
,
8604 cur_offset
+ ins
.offset
- 1,
8607 free_extent_map(em
);
8609 num_bytes
-= ins
.offset
;
8610 cur_offset
+= ins
.offset
;
8611 *alloc_hint
= ins
.objectid
+ ins
.offset
;
8613 inode_inc_iversion(inode
);
8614 inode
->i_ctime
= CURRENT_TIME
;
8615 BTRFS_I(inode
)->flags
|= BTRFS_INODE_PREALLOC
;
8616 if (!(mode
& FALLOC_FL_KEEP_SIZE
) &&
8617 (actual_len
> inode
->i_size
) &&
8618 (cur_offset
> inode
->i_size
)) {
8619 if (cur_offset
> actual_len
)
8620 i_size
= actual_len
;
8622 i_size
= cur_offset
;
8623 i_size_write(inode
, i_size
);
8624 btrfs_ordered_update_i_size(inode
, i_size
, NULL
);
8627 ret
= btrfs_update_inode(trans
, root
, inode
);
8630 btrfs_abort_transaction(trans
, root
, ret
);
8632 btrfs_end_transaction(trans
, root
);
8637 btrfs_end_transaction(trans
, root
);
8642 int btrfs_prealloc_file_range(struct inode
*inode
, int mode
,
8643 u64 start
, u64 num_bytes
, u64 min_size
,
8644 loff_t actual_len
, u64
*alloc_hint
)
8646 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8647 min_size
, actual_len
, alloc_hint
,
8651 int btrfs_prealloc_file_range_trans(struct inode
*inode
,
8652 struct btrfs_trans_handle
*trans
, int mode
,
8653 u64 start
, u64 num_bytes
, u64 min_size
,
8654 loff_t actual_len
, u64
*alloc_hint
)
8656 return __btrfs_prealloc_file_range(inode
, mode
, start
, num_bytes
,
8657 min_size
, actual_len
, alloc_hint
, trans
);
8660 static int btrfs_set_page_dirty(struct page
*page
)
8662 return __set_page_dirty_nobuffers(page
);
8665 static int btrfs_permission(struct inode
*inode
, int mask
)
8667 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
8668 umode_t mode
= inode
->i_mode
;
8670 if (mask
& MAY_WRITE
&&
8671 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
))) {
8672 if (btrfs_root_readonly(root
))
8674 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_READONLY
)
8677 return generic_permission(inode
, mask
);
8680 static const struct inode_operations btrfs_dir_inode_operations
= {
8681 .getattr
= btrfs_getattr
,
8682 .lookup
= btrfs_lookup
,
8683 .create
= btrfs_create
,
8684 .unlink
= btrfs_unlink
,
8686 .mkdir
= btrfs_mkdir
,
8687 .rmdir
= btrfs_rmdir
,
8688 .rename
= btrfs_rename
,
8689 .symlink
= btrfs_symlink
,
8690 .setattr
= btrfs_setattr
,
8691 .mknod
= btrfs_mknod
,
8692 .setxattr
= btrfs_setxattr
,
8693 .getxattr
= btrfs_getxattr
,
8694 .listxattr
= btrfs_listxattr
,
8695 .removexattr
= btrfs_removexattr
,
8696 .permission
= btrfs_permission
,
8697 .get_acl
= btrfs_get_acl
,
8699 static const struct inode_operations btrfs_dir_ro_inode_operations
= {
8700 .lookup
= btrfs_lookup
,
8701 .permission
= btrfs_permission
,
8702 .get_acl
= btrfs_get_acl
,
8705 static const struct file_operations btrfs_dir_file_operations
= {
8706 .llseek
= generic_file_llseek
,
8707 .read
= generic_read_dir
,
8708 .readdir
= btrfs_real_readdir
,
8709 .unlocked_ioctl
= btrfs_ioctl
,
8710 #ifdef CONFIG_COMPAT
8711 .compat_ioctl
= btrfs_ioctl
,
8713 .release
= btrfs_release_file
,
8714 .fsync
= btrfs_sync_file
,
8717 static struct extent_io_ops btrfs_extent_io_ops
= {
8718 .fill_delalloc
= run_delalloc_range
,
8719 .submit_bio_hook
= btrfs_submit_bio_hook
,
8720 .merge_bio_hook
= btrfs_merge_bio_hook
,
8721 .readpage_end_io_hook
= btrfs_readpage_end_io_hook
,
8722 .writepage_end_io_hook
= btrfs_writepage_end_io_hook
,
8723 .writepage_start_hook
= btrfs_writepage_start_hook
,
8724 .set_bit_hook
= btrfs_set_bit_hook
,
8725 .clear_bit_hook
= btrfs_clear_bit_hook
,
8726 .merge_extent_hook
= btrfs_merge_extent_hook
,
8727 .split_extent_hook
= btrfs_split_extent_hook
,
8731 * btrfs doesn't support the bmap operation because swapfiles
8732 * use bmap to make a mapping of extents in the file. They assume
8733 * these extents won't change over the life of the file and they
8734 * use the bmap result to do IO directly to the drive.
8736 * the btrfs bmap call would return logical addresses that aren't
8737 * suitable for IO and they also will change frequently as COW
8738 * operations happen. So, swapfile + btrfs == corruption.
8740 * For now we're avoiding this by dropping bmap.
8742 static const struct address_space_operations btrfs_aops
= {
8743 .readpage
= btrfs_readpage
,
8744 .writepage
= btrfs_writepage
,
8745 .writepages
= btrfs_writepages
,
8746 .readpages
= btrfs_readpages
,
8747 .direct_IO
= btrfs_direct_IO
,
8748 .invalidatepage
= btrfs_invalidatepage
,
8749 .releasepage
= btrfs_releasepage
,
8750 .set_page_dirty
= btrfs_set_page_dirty
,
8751 .error_remove_page
= generic_error_remove_page
,
8754 static const struct address_space_operations btrfs_symlink_aops
= {
8755 .readpage
= btrfs_readpage
,
8756 .writepage
= btrfs_writepage
,
8757 .invalidatepage
= btrfs_invalidatepage
,
8758 .releasepage
= btrfs_releasepage
,
8761 static const struct inode_operations btrfs_file_inode_operations
= {
8762 .getattr
= btrfs_getattr
,
8763 .setattr
= btrfs_setattr
,
8764 .setxattr
= btrfs_setxattr
,
8765 .getxattr
= btrfs_getxattr
,
8766 .listxattr
= btrfs_listxattr
,
8767 .removexattr
= btrfs_removexattr
,
8768 .permission
= btrfs_permission
,
8769 .fiemap
= btrfs_fiemap
,
8770 .get_acl
= btrfs_get_acl
,
8771 .update_time
= btrfs_update_time
,
8773 static const struct inode_operations btrfs_special_inode_operations
= {
8774 .getattr
= btrfs_getattr
,
8775 .setattr
= btrfs_setattr
,
8776 .permission
= btrfs_permission
,
8777 .setxattr
= btrfs_setxattr
,
8778 .getxattr
= btrfs_getxattr
,
8779 .listxattr
= btrfs_listxattr
,
8780 .removexattr
= btrfs_removexattr
,
8781 .get_acl
= btrfs_get_acl
,
8782 .update_time
= btrfs_update_time
,
8784 static const struct inode_operations btrfs_symlink_inode_operations
= {
8785 .readlink
= generic_readlink
,
8786 .follow_link
= page_follow_link_light
,
8787 .put_link
= page_put_link
,
8788 .getattr
= btrfs_getattr
,
8789 .setattr
= btrfs_setattr
,
8790 .permission
= btrfs_permission
,
8791 .setxattr
= btrfs_setxattr
,
8792 .getxattr
= btrfs_getxattr
,
8793 .listxattr
= btrfs_listxattr
,
8794 .removexattr
= btrfs_removexattr
,
8795 .get_acl
= btrfs_get_acl
,
8796 .update_time
= btrfs_update_time
,
8799 const struct dentry_operations btrfs_dentry_operations
= {
8800 .d_delete
= btrfs_dentry_delete
,
8801 .d_release
= btrfs_dentry_release
,