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/smp_lock.h>
30 #include <linux/backing-dev.h>
31 #include <linux/mpage.h>
32 #include <linux/swap.h>
33 #include <linux/writeback.h>
34 #include <linux/statfs.h>
35 #include <linux/compat.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/version.h>
38 #include <linux/xattr.h>
39 #include <linux/posix_acl.h>
40 #include <linux/falloc.h>
44 #include "transaction.h"
45 #include "btrfs_inode.h"
47 #include "print-tree.h"
49 #include "ordered-data.h"
52 #include "ref-cache.h"
53 #include "compression.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static struct inode_operations btrfs_dir_inode_operations;
61 static struct inode_operations btrfs_symlink_inode_operations;
62 static struct inode_operations btrfs_dir_ro_inode_operations;
63 static struct inode_operations btrfs_special_inode_operations;
64 static struct inode_operations btrfs_file_inode_operations;
65 static struct address_space_operations btrfs_aops;
66 static struct address_space_operations btrfs_symlink_aops;
67 static struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_bit_radix_cachep;
74 struct kmem_cache *btrfs_path_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
95 * a very lame attempt at stopping writes when the FS is 85% full. There
96 * are countless ways this is incorrect, but it is better than nothing.
98 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
106 spin_lock(&root->fs_info->delalloc_lock);
107 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
108 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
116 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
118 spin_unlock(&root->fs_info->delalloc_lock);
123 * this does all the hard work for inserting an inline extent into
124 * the btree. The caller should have done a btrfs_drop_extents so that
125 * no overlapping inline items exist in the btree
127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
128 struct btrfs_root *root, struct inode *inode,
129 u64 start, size_t size, size_t compressed_size,
130 struct page **compressed_pages)
132 struct btrfs_key key;
133 struct btrfs_path *path;
134 struct extent_buffer *leaf;
135 struct page *page = NULL;
138 struct btrfs_file_extent_item *ei;
141 size_t cur_size = size;
143 unsigned long offset;
144 int use_compress = 0;
146 if (compressed_size && compressed_pages) {
148 cur_size = compressed_size;
151 path = btrfs_alloc_path();
155 btrfs_set_trans_block_group(trans, inode);
157 key.objectid = inode->i_ino;
159 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
160 datasize = btrfs_file_extent_calc_inline_size(cur_size);
162 inode_add_bytes(inode, size);
163 ret = btrfs_insert_empty_item(trans, root, path, &key,
170 leaf = path->nodes[0];
171 ei = btrfs_item_ptr(leaf, path->slots[0],
172 struct btrfs_file_extent_item);
173 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
174 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
175 btrfs_set_file_extent_encryption(leaf, ei, 0);
176 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
177 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
178 ptr = btrfs_file_extent_inline_start(ei);
183 while (compressed_size > 0) {
184 cpage = compressed_pages[i];
185 cur_size = min_t(unsigned long, compressed_size,
189 write_extent_buffer(leaf, kaddr, ptr, cur_size);
194 compressed_size -= cur_size;
196 btrfs_set_file_extent_compression(leaf, ei,
197 BTRFS_COMPRESS_ZLIB);
199 page = find_get_page(inode->i_mapping,
200 start >> PAGE_CACHE_SHIFT);
201 btrfs_set_file_extent_compression(leaf, ei, 0);
202 kaddr = kmap_atomic(page, KM_USER0);
203 offset = start & (PAGE_CACHE_SIZE - 1);
204 write_extent_buffer(leaf, kaddr + offset, ptr, size);
205 kunmap_atomic(kaddr, KM_USER0);
206 page_cache_release(page);
208 btrfs_mark_buffer_dirty(leaf);
209 btrfs_free_path(path);
211 BTRFS_I(inode)->disk_i_size = inode->i_size;
212 btrfs_update_inode(trans, root, inode);
215 btrfs_free_path(path);
221 * conditionally insert an inline extent into the file. This
222 * does the checks required to make sure the data is small enough
223 * to fit as an inline extent.
225 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
226 struct btrfs_root *root,
227 struct inode *inode, u64 start, u64 end,
228 size_t compressed_size,
229 struct page **compressed_pages)
231 u64 isize = i_size_read(inode);
232 u64 actual_end = min(end + 1, isize);
233 u64 inline_len = actual_end - start;
234 u64 aligned_end = (end + root->sectorsize - 1) &
235 ~((u64)root->sectorsize - 1);
237 u64 data_len = inline_len;
241 data_len = compressed_size;
244 actual_end >= PAGE_CACHE_SIZE ||
245 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
247 (actual_end & (root->sectorsize - 1)) == 0) ||
249 data_len > root->fs_info->max_inline) {
253 ret = btrfs_drop_extents(trans, root, inode, start,
254 aligned_end, start, &hint_byte);
257 if (isize > actual_end)
258 inline_len = min_t(u64, isize, actual_end);
259 ret = insert_inline_extent(trans, root, inode, start,
260 inline_len, compressed_size,
263 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
267 struct async_extent {
272 unsigned long nr_pages;
273 struct list_head list;
278 struct btrfs_root *root;
279 struct page *locked_page;
282 struct list_head extents;
283 struct btrfs_work work;
286 static noinline int add_async_extent(struct async_cow *cow,
287 u64 start, u64 ram_size,
290 unsigned long nr_pages)
292 struct async_extent *async_extent;
294 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
295 async_extent->start = start;
296 async_extent->ram_size = ram_size;
297 async_extent->compressed_size = compressed_size;
298 async_extent->pages = pages;
299 async_extent->nr_pages = nr_pages;
300 list_add_tail(&async_extent->list, &cow->extents);
305 * we create compressed extents in two phases. The first
306 * phase compresses a range of pages that have already been
307 * locked (both pages and state bits are locked).
309 * This is done inside an ordered work queue, and the compression
310 * is spread across many cpus. The actual IO submission is step
311 * two, and the ordered work queue takes care of making sure that
312 * happens in the same order things were put onto the queue by
313 * writepages and friends.
315 * If this code finds it can't get good compression, it puts an
316 * entry onto the work queue to write the uncompressed bytes. This
317 * makes sure that both compressed inodes and uncompressed inodes
318 * are written in the same order that pdflush sent them down.
320 static noinline int compress_file_range(struct inode *inode,
321 struct page *locked_page,
323 struct async_cow *async_cow,
326 struct btrfs_root *root = BTRFS_I(inode)->root;
327 struct btrfs_trans_handle *trans;
331 u64 blocksize = root->sectorsize;
333 u64 isize = i_size_read(inode);
335 struct page **pages = NULL;
336 unsigned long nr_pages;
337 unsigned long nr_pages_ret = 0;
338 unsigned long total_compressed = 0;
339 unsigned long total_in = 0;
340 unsigned long max_compressed = 128 * 1024;
341 unsigned long max_uncompressed = 128 * 1024;
347 actual_end = min_t(u64, isize, end + 1);
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353 total_compressed = actual_end - start;
355 /* we want to make sure that amount of ram required to uncompress
356 * an extent is reasonable, so we limit the total size in ram
357 * of a compressed extent to 128k. This is a crucial number
358 * because it also controls how easily we can spread reads across
359 * cpus for decompression.
361 * We also want to make sure the amount of IO required to do
362 * a random read is reasonably small, so we limit the size of
363 * a compressed extent to 128k.
365 total_compressed = min(total_compressed, max_uncompressed);
366 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
367 num_bytes = max(blocksize, num_bytes);
368 disk_num_bytes = num_bytes;
373 * we do compression for mount -o compress and when the
374 * inode has not been flagged as nocompress. This flag can
375 * change at any time if we discover bad compression ratios.
377 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
378 btrfs_test_opt(root, COMPRESS)) {
380 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
382 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
383 total_compressed, pages,
384 nr_pages, &nr_pages_ret,
390 unsigned long offset = total_compressed &
391 (PAGE_CACHE_SIZE - 1);
392 struct page *page = pages[nr_pages_ret - 1];
395 /* zero the tail end of the last page, we might be
396 * sending it down to disk
399 kaddr = kmap_atomic(page, KM_USER0);
400 memset(kaddr + offset, 0,
401 PAGE_CACHE_SIZE - offset);
402 kunmap_atomic(kaddr, KM_USER0);
408 trans = btrfs_join_transaction(root, 1);
410 btrfs_set_trans_block_group(trans, inode);
412 /* lets try to make an inline extent */
413 if (ret || total_in < (actual_end - start)) {
414 /* we didn't compress the entire range, try
415 * to make an uncompressed inline extent.
417 ret = cow_file_range_inline(trans, root, inode,
418 start, end, 0, NULL);
420 /* try making a compressed inline extent */
421 ret = cow_file_range_inline(trans, root, inode,
423 total_compressed, pages);
425 btrfs_end_transaction(trans, root);
428 * inline extent creation worked, we don't need
429 * to create any more async work items. Unlock
430 * and free up our temp pages.
432 extent_clear_unlock_delalloc(inode,
433 &BTRFS_I(inode)->io_tree,
434 start, end, NULL, 1, 0,
443 * we aren't doing an inline extent round the compressed size
444 * up to a block size boundary so the allocator does sane
447 total_compressed = (total_compressed + blocksize - 1) &
451 * one last check to make sure the compression is really a
452 * win, compare the page count read with the blocks on disk
454 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
455 ~(PAGE_CACHE_SIZE - 1);
456 if (total_compressed >= total_in) {
459 disk_num_bytes = total_compressed;
460 num_bytes = total_in;
463 if (!will_compress && pages) {
465 * the compression code ran but failed to make things smaller,
466 * free any pages it allocated and our page pointer array
468 for (i = 0; i < nr_pages_ret; i++) {
469 WARN_ON(pages[i]->mapping);
470 page_cache_release(pages[i]);
474 total_compressed = 0;
477 /* flag the file so we don't compress in the future */
478 btrfs_set_flag(inode, NOCOMPRESS);
483 /* the async work queues will take care of doing actual
484 * allocation on disk for these compressed pages,
485 * and will submit them to the elevator.
487 add_async_extent(async_cow, start, num_bytes,
488 total_compressed, pages, nr_pages_ret);
490 if (start + num_bytes < end && start + num_bytes < actual_end) {
498 * No compression, but we still need to write the pages in
499 * the file we've been given so far. redirty the locked
500 * page if it corresponds to our extent and set things up
501 * for the async work queue to run cow_file_range to do
502 * the normal delalloc dance
504 if (page_offset(locked_page) >= start &&
505 page_offset(locked_page) <= end) {
506 __set_page_dirty_nobuffers(locked_page);
507 /* unlocked later on in the async handlers */
509 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
517 for (i = 0; i < nr_pages_ret; i++) {
518 WARN_ON(pages[i]->mapping);
519 page_cache_release(pages[i]);
527 * phase two of compressed writeback. This is the ordered portion
528 * of the code, which only gets called in the order the work was
529 * queued. We walk all the async extents created by compress_file_range
530 * and send them down to the disk.
532 static noinline int submit_compressed_extents(struct inode *inode,
533 struct async_cow *async_cow)
535 struct async_extent *async_extent;
537 struct btrfs_trans_handle *trans;
538 struct btrfs_key ins;
539 struct extent_map *em;
540 struct btrfs_root *root = BTRFS_I(inode)->root;
541 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
542 struct extent_io_tree *io_tree;
545 if (list_empty(&async_cow->extents))
548 trans = btrfs_join_transaction(root, 1);
550 while (!list_empty(&async_cow->extents)) {
551 async_extent = list_entry(async_cow->extents.next,
552 struct async_extent, list);
553 list_del(&async_extent->list);
555 io_tree = &BTRFS_I(inode)->io_tree;
557 /* did the compression code fall back to uncompressed IO? */
558 if (!async_extent->pages) {
559 int page_started = 0;
560 unsigned long nr_written = 0;
562 lock_extent(io_tree, async_extent->start,
563 async_extent->start +
564 async_extent->ram_size - 1, GFP_NOFS);
566 /* allocate blocks */
567 cow_file_range(inode, async_cow->locked_page,
569 async_extent->start +
570 async_extent->ram_size - 1,
571 &page_started, &nr_written, 0);
574 * if page_started, cow_file_range inserted an
575 * inline extent and took care of all the unlocking
576 * and IO for us. Otherwise, we need to submit
577 * all those pages down to the drive.
580 extent_write_locked_range(io_tree,
581 inode, async_extent->start,
582 async_extent->start +
583 async_extent->ram_size - 1,
591 lock_extent(io_tree, async_extent->start,
592 async_extent->start + async_extent->ram_size - 1,
595 * here we're doing allocation and writeback of the
598 btrfs_drop_extent_cache(inode, async_extent->start,
599 async_extent->start +
600 async_extent->ram_size - 1, 0);
602 ret = btrfs_reserve_extent(trans, root,
603 async_extent->compressed_size,
604 async_extent->compressed_size,
608 em = alloc_extent_map(GFP_NOFS);
609 em->start = async_extent->start;
610 em->len = async_extent->ram_size;
611 em->orig_start = em->start;
613 em->block_start = ins.objectid;
614 em->block_len = ins.offset;
615 em->bdev = root->fs_info->fs_devices->latest_bdev;
616 set_bit(EXTENT_FLAG_PINNED, &em->flags);
617 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
620 spin_lock(&em_tree->lock);
621 ret = add_extent_mapping(em_tree, em);
622 spin_unlock(&em_tree->lock);
623 if (ret != -EEXIST) {
627 btrfs_drop_extent_cache(inode, async_extent->start,
628 async_extent->start +
629 async_extent->ram_size - 1, 0);
632 ret = btrfs_add_ordered_extent(inode, async_extent->start,
634 async_extent->ram_size,
636 BTRFS_ORDERED_COMPRESSED);
639 btrfs_end_transaction(trans, root);
642 * clear dirty, set writeback and unlock the pages.
644 extent_clear_unlock_delalloc(inode,
645 &BTRFS_I(inode)->io_tree,
647 async_extent->start +
648 async_extent->ram_size - 1,
649 NULL, 1, 1, 0, 1, 1, 0);
651 ret = btrfs_submit_compressed_write(inode,
653 async_extent->ram_size,
655 ins.offset, async_extent->pages,
656 async_extent->nr_pages);
659 trans = btrfs_join_transaction(root, 1);
660 alloc_hint = ins.objectid + ins.offset;
665 btrfs_end_transaction(trans, root);
670 * when extent_io.c finds a delayed allocation range in the file,
671 * the call backs end up in this code. The basic idea is to
672 * allocate extents on disk for the range, and create ordered data structs
673 * in ram to track those extents.
675 * locked_page is the page that writepage had locked already. We use
676 * it to make sure we don't do extra locks or unlocks.
678 * *page_started is set to one if we unlock locked_page and do everything
679 * required to start IO on it. It may be clean and already done with
682 static noinline int cow_file_range(struct inode *inode,
683 struct page *locked_page,
684 u64 start, u64 end, int *page_started,
685 unsigned long *nr_written,
688 struct btrfs_root *root = BTRFS_I(inode)->root;
689 struct btrfs_trans_handle *trans;
692 unsigned long ram_size;
695 u64 blocksize = root->sectorsize;
697 u64 isize = i_size_read(inode);
698 struct btrfs_key ins;
699 struct extent_map *em;
700 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
703 trans = btrfs_join_transaction(root, 1);
705 btrfs_set_trans_block_group(trans, inode);
707 actual_end = min_t(u64, isize, end + 1);
709 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
710 num_bytes = max(blocksize, num_bytes);
711 disk_num_bytes = num_bytes;
715 /* lets try to make an inline extent */
716 ret = cow_file_range_inline(trans, root, inode,
717 start, end, 0, NULL);
719 extent_clear_unlock_delalloc(inode,
720 &BTRFS_I(inode)->io_tree,
721 start, end, NULL, 1, 1,
723 *nr_written = *nr_written +
724 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
731 BUG_ON(disk_num_bytes >
732 btrfs_super_total_bytes(&root->fs_info->super_copy));
734 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
736 while (disk_num_bytes > 0) {
737 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
738 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
739 root->sectorsize, 0, alloc_hint,
743 em = alloc_extent_map(GFP_NOFS);
745 em->orig_start = em->start;
747 ram_size = ins.offset;
748 em->len = ins.offset;
750 em->block_start = ins.objectid;
751 em->block_len = ins.offset;
752 em->bdev = root->fs_info->fs_devices->latest_bdev;
753 set_bit(EXTENT_FLAG_PINNED, &em->flags);
756 spin_lock(&em_tree->lock);
757 ret = add_extent_mapping(em_tree, em);
758 spin_unlock(&em_tree->lock);
759 if (ret != -EEXIST) {
763 btrfs_drop_extent_cache(inode, start,
764 start + ram_size - 1, 0);
767 cur_alloc_size = ins.offset;
768 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
769 ram_size, cur_alloc_size, 0);
772 if (root->root_key.objectid ==
773 BTRFS_DATA_RELOC_TREE_OBJECTID) {
774 ret = btrfs_reloc_clone_csums(inode, start,
779 if (disk_num_bytes < cur_alloc_size)
782 /* we're not doing compressed IO, don't unlock the first
783 * page (which the caller expects to stay locked), don't
784 * clear any dirty bits and don't set any writeback bits
786 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
787 start, start + ram_size - 1,
788 locked_page, unlock, 1,
790 disk_num_bytes -= cur_alloc_size;
791 num_bytes -= cur_alloc_size;
792 alloc_hint = ins.objectid + ins.offset;
793 start += cur_alloc_size;
797 btrfs_end_transaction(trans, root);
803 * work queue call back to started compression on a file and pages
805 static noinline void async_cow_start(struct btrfs_work *work)
807 struct async_cow *async_cow;
809 async_cow = container_of(work, struct async_cow, work);
811 compress_file_range(async_cow->inode, async_cow->locked_page,
812 async_cow->start, async_cow->end, async_cow,
815 async_cow->inode = NULL;
819 * work queue call back to submit previously compressed pages
821 static noinline void async_cow_submit(struct btrfs_work *work)
823 struct async_cow *async_cow;
824 struct btrfs_root *root;
825 unsigned long nr_pages;
827 async_cow = container_of(work, struct async_cow, work);
829 root = async_cow->root;
830 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
833 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
835 if (atomic_read(&root->fs_info->async_delalloc_pages) <
837 waitqueue_active(&root->fs_info->async_submit_wait))
838 wake_up(&root->fs_info->async_submit_wait);
840 if (async_cow->inode)
841 submit_compressed_extents(async_cow->inode, async_cow);
844 static noinline void async_cow_free(struct btrfs_work *work)
846 struct async_cow *async_cow;
847 async_cow = container_of(work, struct async_cow, work);
851 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
852 u64 start, u64 end, int *page_started,
853 unsigned long *nr_written)
855 struct async_cow *async_cow;
856 struct btrfs_root *root = BTRFS_I(inode)->root;
857 unsigned long nr_pages;
859 int limit = 10 * 1024 * 1042;
861 if (!btrfs_test_opt(root, COMPRESS)) {
862 return cow_file_range(inode, locked_page, start, end,
863 page_started, nr_written, 1);
866 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
867 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
868 while (start < end) {
869 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
870 async_cow->inode = inode;
871 async_cow->root = root;
872 async_cow->locked_page = locked_page;
873 async_cow->start = start;
875 if (btrfs_test_flag(inode, NOCOMPRESS))
878 cur_end = min(end, start + 512 * 1024 - 1);
880 async_cow->end = cur_end;
881 INIT_LIST_HEAD(&async_cow->extents);
883 async_cow->work.func = async_cow_start;
884 async_cow->work.ordered_func = async_cow_submit;
885 async_cow->work.ordered_free = async_cow_free;
886 async_cow->work.flags = 0;
888 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
890 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
892 btrfs_queue_worker(&root->fs_info->delalloc_workers,
895 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
896 wait_event(root->fs_info->async_submit_wait,
897 (atomic_read(&root->fs_info->async_delalloc_pages) <
901 while (atomic_read(&root->fs_info->async_submit_draining) &&
902 atomic_read(&root->fs_info->async_delalloc_pages)) {
903 wait_event(root->fs_info->async_submit_wait,
904 (atomic_read(&root->fs_info->async_delalloc_pages) ==
908 *nr_written += nr_pages;
915 static noinline int csum_exist_in_range(struct btrfs_root *root,
916 u64 bytenr, u64 num_bytes)
919 struct btrfs_ordered_sum *sums;
922 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
923 bytenr + num_bytes - 1, &list);
924 if (ret == 0 && list_empty(&list))
927 while (!list_empty(&list)) {
928 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
929 list_del(&sums->list);
936 * when nowcow writeback call back. This checks for snapshots or COW copies
937 * of the extents that exist in the file, and COWs the file as required.
939 * If no cow copies or snapshots exist, we write directly to the existing
942 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
943 u64 start, u64 end, int *page_started, int force,
944 unsigned long *nr_written)
946 struct btrfs_root *root = BTRFS_I(inode)->root;
947 struct btrfs_trans_handle *trans;
948 struct extent_buffer *leaf;
949 struct btrfs_path *path;
950 struct btrfs_file_extent_item *fi;
951 struct btrfs_key found_key;
963 path = btrfs_alloc_path();
965 trans = btrfs_join_transaction(root, 1);
971 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
974 if (ret > 0 && path->slots[0] > 0 && check_prev) {
975 leaf = path->nodes[0];
976 btrfs_item_key_to_cpu(leaf, &found_key,
978 if (found_key.objectid == inode->i_ino &&
979 found_key.type == BTRFS_EXTENT_DATA_KEY)
984 leaf = path->nodes[0];
985 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
986 ret = btrfs_next_leaf(root, path);
991 leaf = path->nodes[0];
997 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
999 if (found_key.objectid > inode->i_ino ||
1000 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1001 found_key.offset > end)
1004 if (found_key.offset > cur_offset) {
1005 extent_end = found_key.offset;
1009 fi = btrfs_item_ptr(leaf, path->slots[0],
1010 struct btrfs_file_extent_item);
1011 extent_type = btrfs_file_extent_type(leaf, fi);
1013 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1014 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1015 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1016 extent_end = found_key.offset +
1017 btrfs_file_extent_num_bytes(leaf, fi);
1018 if (extent_end <= start) {
1022 if (disk_bytenr == 0)
1024 if (btrfs_file_extent_compression(leaf, fi) ||
1025 btrfs_file_extent_encryption(leaf, fi) ||
1026 btrfs_file_extent_other_encoding(leaf, fi))
1028 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1030 if (btrfs_extent_readonly(root, disk_bytenr))
1032 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1035 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1036 disk_bytenr += cur_offset - found_key.offset;
1037 num_bytes = min(end + 1, extent_end) - cur_offset;
1039 * force cow if csum exists in the range.
1040 * this ensure that csum for a given extent are
1041 * either valid or do not exist.
1043 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1046 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1047 extent_end = found_key.offset +
1048 btrfs_file_extent_inline_len(leaf, fi);
1049 extent_end = ALIGN(extent_end, root->sectorsize);
1054 if (extent_end <= start) {
1059 if (cow_start == (u64)-1)
1060 cow_start = cur_offset;
1061 cur_offset = extent_end;
1062 if (cur_offset > end)
1068 btrfs_release_path(root, path);
1069 if (cow_start != (u64)-1) {
1070 ret = cow_file_range(inode, locked_page, cow_start,
1071 found_key.offset - 1, page_started,
1074 cow_start = (u64)-1;
1077 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1078 struct extent_map *em;
1079 struct extent_map_tree *em_tree;
1080 em_tree = &BTRFS_I(inode)->extent_tree;
1081 em = alloc_extent_map(GFP_NOFS);
1082 em->start = cur_offset;
1083 em->orig_start = em->start;
1084 em->len = num_bytes;
1085 em->block_len = num_bytes;
1086 em->block_start = disk_bytenr;
1087 em->bdev = root->fs_info->fs_devices->latest_bdev;
1088 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1090 spin_lock(&em_tree->lock);
1091 ret = add_extent_mapping(em_tree, em);
1092 spin_unlock(&em_tree->lock);
1093 if (ret != -EEXIST) {
1094 free_extent_map(em);
1097 btrfs_drop_extent_cache(inode, em->start,
1098 em->start + em->len - 1, 0);
1100 type = BTRFS_ORDERED_PREALLOC;
1102 type = BTRFS_ORDERED_NOCOW;
1105 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1106 num_bytes, num_bytes, type);
1109 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1110 cur_offset, cur_offset + num_bytes - 1,
1111 locked_page, 1, 1, 1, 0, 0, 0);
1112 cur_offset = extent_end;
1113 if (cur_offset > end)
1116 btrfs_release_path(root, path);
1118 if (cur_offset <= end && cow_start == (u64)-1)
1119 cow_start = cur_offset;
1120 if (cow_start != (u64)-1) {
1121 ret = cow_file_range(inode, locked_page, cow_start, end,
1122 page_started, nr_written, 1);
1126 ret = btrfs_end_transaction(trans, root);
1128 btrfs_free_path(path);
1133 * extent_io.c call back to do delayed allocation processing
1135 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1136 u64 start, u64 end, int *page_started,
1137 unsigned long *nr_written)
1141 if (btrfs_test_flag(inode, NODATACOW))
1142 ret = run_delalloc_nocow(inode, locked_page, start, end,
1143 page_started, 1, nr_written);
1144 else if (btrfs_test_flag(inode, PREALLOC))
1145 ret = run_delalloc_nocow(inode, locked_page, start, end,
1146 page_started, 0, nr_written);
1148 ret = cow_file_range_async(inode, locked_page, start, end,
1149 page_started, nr_written);
1155 * extent_io.c set_bit_hook, used to track delayed allocation
1156 * bytes in this file, and to maintain the list of inodes that
1157 * have pending delalloc work to be done.
1159 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1160 unsigned long old, unsigned long bits)
1163 * set_bit and clear bit hooks normally require _irqsave/restore
1164 * but in this case, we are only testeing for the DELALLOC
1165 * bit, which is only set or cleared with irqs on
1167 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1168 struct btrfs_root *root = BTRFS_I(inode)->root;
1169 spin_lock(&root->fs_info->delalloc_lock);
1170 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1171 root->fs_info->delalloc_bytes += end - start + 1;
1172 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1173 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1174 &root->fs_info->delalloc_inodes);
1176 spin_unlock(&root->fs_info->delalloc_lock);
1182 * extent_io.c clear_bit_hook, see set_bit_hook for why
1184 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1185 unsigned long old, unsigned long bits)
1188 * set_bit and clear bit hooks normally require _irqsave/restore
1189 * but in this case, we are only testeing for the DELALLOC
1190 * bit, which is only set or cleared with irqs on
1192 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1193 struct btrfs_root *root = BTRFS_I(inode)->root;
1195 spin_lock(&root->fs_info->delalloc_lock);
1196 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1197 printk(KERN_INFO "btrfs warning: delalloc account "
1199 (unsigned long long)end - start + 1,
1200 (unsigned long long)
1201 root->fs_info->delalloc_bytes);
1202 root->fs_info->delalloc_bytes = 0;
1203 BTRFS_I(inode)->delalloc_bytes = 0;
1205 root->fs_info->delalloc_bytes -= end - start + 1;
1206 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1208 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1209 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1210 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1212 spin_unlock(&root->fs_info->delalloc_lock);
1218 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1219 * we don't create bios that span stripes or chunks
1221 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1222 size_t size, struct bio *bio,
1223 unsigned long bio_flags)
1225 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1226 struct btrfs_mapping_tree *map_tree;
1227 u64 logical = (u64)bio->bi_sector << 9;
1232 if (bio_flags & EXTENT_BIO_COMPRESSED)
1235 length = bio->bi_size;
1236 map_tree = &root->fs_info->mapping_tree;
1237 map_length = length;
1238 ret = btrfs_map_block(map_tree, READ, logical,
1239 &map_length, NULL, 0);
1241 if (map_length < length + size)
1247 * in order to insert checksums into the metadata in large chunks,
1248 * we wait until bio submission time. All the pages in the bio are
1249 * checksummed and sums are attached onto the ordered extent record.
1251 * At IO completion time the cums attached on the ordered extent record
1252 * are inserted into the btree
1254 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1255 struct bio *bio, int mirror_num,
1256 unsigned long bio_flags)
1258 struct btrfs_root *root = BTRFS_I(inode)->root;
1261 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1267 * in order to insert checksums into the metadata in large chunks,
1268 * we wait until bio submission time. All the pages in the bio are
1269 * checksummed and sums are attached onto the ordered extent record.
1271 * At IO completion time the cums attached on the ordered extent record
1272 * are inserted into the btree
1274 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1275 int mirror_num, unsigned long bio_flags)
1277 struct btrfs_root *root = BTRFS_I(inode)->root;
1278 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1282 * extent_io.c submission hook. This does the right thing for csum calculation
1283 * on write, or reading the csums from the tree before a read
1285 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1286 int mirror_num, unsigned long bio_flags)
1288 struct btrfs_root *root = BTRFS_I(inode)->root;
1292 skip_sum = btrfs_test_flag(inode, NODATASUM);
1294 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1297 if (!(rw & (1 << BIO_RW))) {
1298 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1299 return btrfs_submit_compressed_read(inode, bio,
1300 mirror_num, bio_flags);
1301 } else if (!skip_sum)
1302 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1304 } else if (!skip_sum) {
1305 /* csum items have already been cloned */
1306 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1308 /* we're doing a write, do the async checksumming */
1309 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1310 inode, rw, bio, mirror_num,
1311 bio_flags, __btrfs_submit_bio_start,
1312 __btrfs_submit_bio_done);
1316 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1320 * given a list of ordered sums record them in the inode. This happens
1321 * at IO completion time based on sums calculated at bio submission time.
1323 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1324 struct inode *inode, u64 file_offset,
1325 struct list_head *list)
1327 struct list_head *cur;
1328 struct btrfs_ordered_sum *sum;
1330 btrfs_set_trans_block_group(trans, inode);
1331 list_for_each(cur, list) {
1332 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1333 btrfs_csum_file_blocks(trans,
1334 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1339 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1341 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1343 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1347 /* see btrfs_writepage_start_hook for details on why this is required */
1348 struct btrfs_writepage_fixup {
1350 struct btrfs_work work;
1353 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1355 struct btrfs_writepage_fixup *fixup;
1356 struct btrfs_ordered_extent *ordered;
1358 struct inode *inode;
1362 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1366 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1367 ClearPageChecked(page);
1371 inode = page->mapping->host;
1372 page_start = page_offset(page);
1373 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1375 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1377 /* already ordered? We're done */
1378 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1379 EXTENT_ORDERED, 0)) {
1383 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1385 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1386 page_end, GFP_NOFS);
1388 btrfs_start_ordered_extent(inode, ordered, 1);
1392 btrfs_set_extent_delalloc(inode, page_start, page_end);
1393 ClearPageChecked(page);
1395 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1398 page_cache_release(page);
1402 * There are a few paths in the higher layers of the kernel that directly
1403 * set the page dirty bit without asking the filesystem if it is a
1404 * good idea. This causes problems because we want to make sure COW
1405 * properly happens and the data=ordered rules are followed.
1407 * In our case any range that doesn't have the ORDERED bit set
1408 * hasn't been properly setup for IO. We kick off an async process
1409 * to fix it up. The async helper will wait for ordered extents, set
1410 * the delalloc bit and make it safe to write the page.
1412 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1414 struct inode *inode = page->mapping->host;
1415 struct btrfs_writepage_fixup *fixup;
1416 struct btrfs_root *root = BTRFS_I(inode)->root;
1419 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1424 if (PageChecked(page))
1427 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1431 SetPageChecked(page);
1432 page_cache_get(page);
1433 fixup->work.func = btrfs_writepage_fixup_worker;
1435 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1439 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1440 struct inode *inode, u64 file_pos,
1441 u64 disk_bytenr, u64 disk_num_bytes,
1442 u64 num_bytes, u64 ram_bytes,
1443 u8 compression, u8 encryption,
1444 u16 other_encoding, int extent_type)
1446 struct btrfs_root *root = BTRFS_I(inode)->root;
1447 struct btrfs_file_extent_item *fi;
1448 struct btrfs_path *path;
1449 struct extent_buffer *leaf;
1450 struct btrfs_key ins;
1454 path = btrfs_alloc_path();
1457 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1458 file_pos + num_bytes, file_pos, &hint);
1461 ins.objectid = inode->i_ino;
1462 ins.offset = file_pos;
1463 ins.type = BTRFS_EXTENT_DATA_KEY;
1464 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1466 leaf = path->nodes[0];
1467 fi = btrfs_item_ptr(leaf, path->slots[0],
1468 struct btrfs_file_extent_item);
1469 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1470 btrfs_set_file_extent_type(leaf, fi, extent_type);
1471 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1472 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1473 btrfs_set_file_extent_offset(leaf, fi, 0);
1474 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1475 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1476 btrfs_set_file_extent_compression(leaf, fi, compression);
1477 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1478 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1479 btrfs_mark_buffer_dirty(leaf);
1481 inode_add_bytes(inode, num_bytes);
1482 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1484 ins.objectid = disk_bytenr;
1485 ins.offset = disk_num_bytes;
1486 ins.type = BTRFS_EXTENT_ITEM_KEY;
1487 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1488 root->root_key.objectid,
1489 trans->transid, inode->i_ino, &ins);
1492 btrfs_free_path(path);
1496 /* as ordered data IO finishes, this gets called so we can finish
1497 * an ordered extent if the range of bytes in the file it covers are
1500 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1502 struct btrfs_root *root = BTRFS_I(inode)->root;
1503 struct btrfs_trans_handle *trans;
1504 struct btrfs_ordered_extent *ordered_extent;
1505 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1509 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1513 trans = btrfs_join_transaction(root, 1);
1515 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1516 BUG_ON(!ordered_extent);
1517 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1520 lock_extent(io_tree, ordered_extent->file_offset,
1521 ordered_extent->file_offset + ordered_extent->len - 1,
1524 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1526 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1528 ret = btrfs_mark_extent_written(trans, root, inode,
1529 ordered_extent->file_offset,
1530 ordered_extent->file_offset +
1531 ordered_extent->len);
1534 ret = insert_reserved_file_extent(trans, inode,
1535 ordered_extent->file_offset,
1536 ordered_extent->start,
1537 ordered_extent->disk_len,
1538 ordered_extent->len,
1539 ordered_extent->len,
1541 BTRFS_FILE_EXTENT_REG);
1544 unlock_extent(io_tree, ordered_extent->file_offset,
1545 ordered_extent->file_offset + ordered_extent->len - 1,
1548 add_pending_csums(trans, inode, ordered_extent->file_offset,
1549 &ordered_extent->list);
1551 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1552 btrfs_ordered_update_i_size(inode, ordered_extent);
1553 btrfs_update_inode(trans, root, inode);
1554 btrfs_remove_ordered_extent(inode, ordered_extent);
1555 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1558 btrfs_put_ordered_extent(ordered_extent);
1559 /* once for the tree */
1560 btrfs_put_ordered_extent(ordered_extent);
1562 btrfs_end_transaction(trans, root);
1566 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1567 struct extent_state *state, int uptodate)
1569 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1573 * When IO fails, either with EIO or csum verification fails, we
1574 * try other mirrors that might have a good copy of the data. This
1575 * io_failure_record is used to record state as we go through all the
1576 * mirrors. If another mirror has good data, the page is set up to date
1577 * and things continue. If a good mirror can't be found, the original
1578 * bio end_io callback is called to indicate things have failed.
1580 struct io_failure_record {
1585 unsigned long bio_flags;
1589 static int btrfs_io_failed_hook(struct bio *failed_bio,
1590 struct page *page, u64 start, u64 end,
1591 struct extent_state *state)
1593 struct io_failure_record *failrec = NULL;
1595 struct extent_map *em;
1596 struct inode *inode = page->mapping->host;
1597 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1598 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1605 ret = get_state_private(failure_tree, start, &private);
1607 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1610 failrec->start = start;
1611 failrec->len = end - start + 1;
1612 failrec->last_mirror = 0;
1613 failrec->bio_flags = 0;
1615 spin_lock(&em_tree->lock);
1616 em = lookup_extent_mapping(em_tree, start, failrec->len);
1617 if (em->start > start || em->start + em->len < start) {
1618 free_extent_map(em);
1621 spin_unlock(&em_tree->lock);
1623 if (!em || IS_ERR(em)) {
1627 logical = start - em->start;
1628 logical = em->block_start + logical;
1629 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1630 logical = em->block_start;
1631 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1633 failrec->logical = logical;
1634 free_extent_map(em);
1635 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1636 EXTENT_DIRTY, GFP_NOFS);
1637 set_state_private(failure_tree, start,
1638 (u64)(unsigned long)failrec);
1640 failrec = (struct io_failure_record *)(unsigned long)private;
1642 num_copies = btrfs_num_copies(
1643 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1644 failrec->logical, failrec->len);
1645 failrec->last_mirror++;
1647 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1648 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1651 if (state && state->start != failrec->start)
1653 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1655 if (!state || failrec->last_mirror > num_copies) {
1656 set_state_private(failure_tree, failrec->start, 0);
1657 clear_extent_bits(failure_tree, failrec->start,
1658 failrec->start + failrec->len - 1,
1659 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1663 bio = bio_alloc(GFP_NOFS, 1);
1664 bio->bi_private = state;
1665 bio->bi_end_io = failed_bio->bi_end_io;
1666 bio->bi_sector = failrec->logical >> 9;
1667 bio->bi_bdev = failed_bio->bi_bdev;
1670 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1671 if (failed_bio->bi_rw & (1 << BIO_RW))
1676 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1677 failrec->last_mirror,
1678 failrec->bio_flags);
1683 * each time an IO finishes, we do a fast check in the IO failure tree
1684 * to see if we need to process or clean up an io_failure_record
1686 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1689 u64 private_failure;
1690 struct io_failure_record *failure;
1694 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1695 (u64)-1, 1, EXTENT_DIRTY)) {
1696 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1697 start, &private_failure);
1699 failure = (struct io_failure_record *)(unsigned long)
1701 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1703 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1705 failure->start + failure->len - 1,
1706 EXTENT_DIRTY | EXTENT_LOCKED,
1715 * when reads are done, we need to check csums to verify the data is correct
1716 * if there's a match, we allow the bio to finish. If not, we go through
1717 * the io_failure_record routines to find good copies
1719 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1720 struct extent_state *state)
1722 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1723 struct inode *inode = page->mapping->host;
1724 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1726 u64 private = ~(u32)0;
1728 struct btrfs_root *root = BTRFS_I(inode)->root;
1730 unsigned long flags;
1732 if (PageChecked(page)) {
1733 ClearPageChecked(page);
1736 if (btrfs_test_flag(inode, NODATASUM))
1739 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1740 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1)) {
1741 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1746 if (state && state->start == start) {
1747 private = state->private;
1750 ret = get_state_private(io_tree, start, &private);
1752 local_irq_save(flags);
1753 kaddr = kmap_atomic(page, KM_IRQ0);
1757 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1758 btrfs_csum_final(csum, (char *)&csum);
1759 if (csum != private)
1762 kunmap_atomic(kaddr, KM_IRQ0);
1763 local_irq_restore(flags);
1765 /* if the io failure tree for this inode is non-empty,
1766 * check to see if we've recovered from a failed IO
1768 btrfs_clean_io_failures(inode, start);
1772 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1773 "private %llu\n", page->mapping->host->i_ino,
1774 (unsigned long long)start, csum,
1775 (unsigned long long)private);
1776 memset(kaddr + offset, 1, end - start + 1);
1777 flush_dcache_page(page);
1778 kunmap_atomic(kaddr, KM_IRQ0);
1779 local_irq_restore(flags);
1786 * This creates an orphan entry for the given inode in case something goes
1787 * wrong in the middle of an unlink/truncate.
1789 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1791 struct btrfs_root *root = BTRFS_I(inode)->root;
1794 spin_lock(&root->list_lock);
1796 /* already on the orphan list, we're good */
1797 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1798 spin_unlock(&root->list_lock);
1802 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1804 spin_unlock(&root->list_lock);
1807 * insert an orphan item to track this unlinked/truncated file
1809 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1815 * We have done the truncate/delete so we can go ahead and remove the orphan
1816 * item for this particular inode.
1818 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1820 struct btrfs_root *root = BTRFS_I(inode)->root;
1823 spin_lock(&root->list_lock);
1825 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1826 spin_unlock(&root->list_lock);
1830 list_del_init(&BTRFS_I(inode)->i_orphan);
1832 spin_unlock(&root->list_lock);
1836 spin_unlock(&root->list_lock);
1838 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1844 * this cleans up any orphans that may be left on the list from the last use
1847 void btrfs_orphan_cleanup(struct btrfs_root *root)
1849 struct btrfs_path *path;
1850 struct extent_buffer *leaf;
1851 struct btrfs_item *item;
1852 struct btrfs_key key, found_key;
1853 struct btrfs_trans_handle *trans;
1854 struct inode *inode;
1855 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1857 path = btrfs_alloc_path();
1862 key.objectid = BTRFS_ORPHAN_OBJECTID;
1863 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1864 key.offset = (u64)-1;
1868 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1870 printk(KERN_ERR "Error searching slot for orphan: %d"
1876 * if ret == 0 means we found what we were searching for, which
1877 * is weird, but possible, so only screw with path if we didnt
1878 * find the key and see if we have stuff that matches
1881 if (path->slots[0] == 0)
1886 /* pull out the item */
1887 leaf = path->nodes[0];
1888 item = btrfs_item_nr(leaf, path->slots[0]);
1889 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1891 /* make sure the item matches what we want */
1892 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1894 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1897 /* release the path since we're done with it */
1898 btrfs_release_path(root, path);
1901 * this is where we are basically btrfs_lookup, without the
1902 * crossing root thing. we store the inode number in the
1903 * offset of the orphan item.
1905 inode = btrfs_iget_locked(root->fs_info->sb,
1906 found_key.offset, root);
1910 if (inode->i_state & I_NEW) {
1911 BTRFS_I(inode)->root = root;
1913 /* have to set the location manually */
1914 BTRFS_I(inode)->location.objectid = inode->i_ino;
1915 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1916 BTRFS_I(inode)->location.offset = 0;
1918 btrfs_read_locked_inode(inode);
1919 unlock_new_inode(inode);
1923 * add this inode to the orphan list so btrfs_orphan_del does
1924 * the proper thing when we hit it
1926 spin_lock(&root->list_lock);
1927 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1928 spin_unlock(&root->list_lock);
1931 * if this is a bad inode, means we actually succeeded in
1932 * removing the inode, but not the orphan record, which means
1933 * we need to manually delete the orphan since iput will just
1934 * do a destroy_inode
1936 if (is_bad_inode(inode)) {
1937 trans = btrfs_start_transaction(root, 1);
1938 btrfs_orphan_del(trans, inode);
1939 btrfs_end_transaction(trans, root);
1944 /* if we have links, this was a truncate, lets do that */
1945 if (inode->i_nlink) {
1947 btrfs_truncate(inode);
1952 /* this will do delete_inode and everything for us */
1957 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1959 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1961 btrfs_free_path(path);
1965 * read an inode from the btree into the in-memory inode
1967 void btrfs_read_locked_inode(struct inode *inode)
1969 struct btrfs_path *path;
1970 struct extent_buffer *leaf;
1971 struct btrfs_inode_item *inode_item;
1972 struct btrfs_timespec *tspec;
1973 struct btrfs_root *root = BTRFS_I(inode)->root;
1974 struct btrfs_key location;
1975 u64 alloc_group_block;
1979 path = btrfs_alloc_path();
1981 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1983 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1987 leaf = path->nodes[0];
1988 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1989 struct btrfs_inode_item);
1991 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1992 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1993 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1994 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1995 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1997 tspec = btrfs_inode_atime(inode_item);
1998 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1999 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2001 tspec = btrfs_inode_mtime(inode_item);
2002 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2003 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2005 tspec = btrfs_inode_ctime(inode_item);
2006 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2007 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2009 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2010 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2011 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2012 inode->i_generation = BTRFS_I(inode)->generation;
2014 rdev = btrfs_inode_rdev(leaf, inode_item);
2016 BTRFS_I(inode)->index_cnt = (u64)-1;
2017 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2019 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2020 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2021 alloc_group_block, 0);
2022 btrfs_free_path(path);
2025 switch (inode->i_mode & S_IFMT) {
2027 inode->i_mapping->a_ops = &btrfs_aops;
2028 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2029 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2030 inode->i_fop = &btrfs_file_operations;
2031 inode->i_op = &btrfs_file_inode_operations;
2034 inode->i_fop = &btrfs_dir_file_operations;
2035 if (root == root->fs_info->tree_root)
2036 inode->i_op = &btrfs_dir_ro_inode_operations;
2038 inode->i_op = &btrfs_dir_inode_operations;
2041 inode->i_op = &btrfs_symlink_inode_operations;
2042 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2043 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2046 init_special_inode(inode, inode->i_mode, rdev);
2052 btrfs_free_path(path);
2053 make_bad_inode(inode);
2057 * given a leaf and an inode, copy the inode fields into the leaf
2059 static void fill_inode_item(struct btrfs_trans_handle *trans,
2060 struct extent_buffer *leaf,
2061 struct btrfs_inode_item *item,
2062 struct inode *inode)
2064 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2065 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2066 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2067 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2068 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2070 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2071 inode->i_atime.tv_sec);
2072 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2073 inode->i_atime.tv_nsec);
2075 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2076 inode->i_mtime.tv_sec);
2077 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2078 inode->i_mtime.tv_nsec);
2080 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2081 inode->i_ctime.tv_sec);
2082 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2083 inode->i_ctime.tv_nsec);
2085 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2086 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2087 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2088 btrfs_set_inode_transid(leaf, item, trans->transid);
2089 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2090 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2091 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2095 * copy everything in the in-memory inode into the btree.
2097 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2098 struct btrfs_root *root, struct inode *inode)
2100 struct btrfs_inode_item *inode_item;
2101 struct btrfs_path *path;
2102 struct extent_buffer *leaf;
2105 path = btrfs_alloc_path();
2107 ret = btrfs_lookup_inode(trans, root, path,
2108 &BTRFS_I(inode)->location, 1);
2115 leaf = path->nodes[0];
2116 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2117 struct btrfs_inode_item);
2119 fill_inode_item(trans, leaf, inode_item, inode);
2120 btrfs_mark_buffer_dirty(leaf);
2121 btrfs_set_inode_last_trans(trans, inode);
2124 btrfs_free_path(path);
2130 * unlink helper that gets used here in inode.c and in the tree logging
2131 * recovery code. It remove a link in a directory with a given name, and
2132 * also drops the back refs in the inode to the directory
2134 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2135 struct btrfs_root *root,
2136 struct inode *dir, struct inode *inode,
2137 const char *name, int name_len)
2139 struct btrfs_path *path;
2141 struct extent_buffer *leaf;
2142 struct btrfs_dir_item *di;
2143 struct btrfs_key key;
2146 path = btrfs_alloc_path();
2152 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2153 name, name_len, -1);
2162 leaf = path->nodes[0];
2163 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2164 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2167 btrfs_release_path(root, path);
2169 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2171 dir->i_ino, &index);
2173 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2174 "inode %lu parent %lu\n", name_len, name,
2175 inode->i_ino, dir->i_ino);
2179 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2180 index, name, name_len, -1);
2189 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2190 btrfs_release_path(root, path);
2192 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2194 BUG_ON(ret != 0 && ret != -ENOENT);
2196 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2198 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2202 btrfs_free_path(path);
2206 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2207 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2208 btrfs_update_inode(trans, root, dir);
2209 btrfs_drop_nlink(inode);
2210 ret = btrfs_update_inode(trans, root, inode);
2211 dir->i_sb->s_dirt = 1;
2216 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2218 struct btrfs_root *root;
2219 struct btrfs_trans_handle *trans;
2220 struct inode *inode = dentry->d_inode;
2222 unsigned long nr = 0;
2224 root = BTRFS_I(dir)->root;
2226 ret = btrfs_check_free_space(root, 1, 1);
2230 trans = btrfs_start_transaction(root, 1);
2232 btrfs_set_trans_block_group(trans, dir);
2233 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2234 dentry->d_name.name, dentry->d_name.len);
2236 if (inode->i_nlink == 0)
2237 ret = btrfs_orphan_add(trans, inode);
2239 nr = trans->blocks_used;
2241 btrfs_end_transaction_throttle(trans, root);
2243 btrfs_btree_balance_dirty(root, nr);
2247 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2249 struct inode *inode = dentry->d_inode;
2252 struct btrfs_root *root = BTRFS_I(dir)->root;
2253 struct btrfs_trans_handle *trans;
2254 unsigned long nr = 0;
2257 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2258 * the root of a subvolume or snapshot
2260 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2261 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2265 ret = btrfs_check_free_space(root, 1, 1);
2269 trans = btrfs_start_transaction(root, 1);
2270 btrfs_set_trans_block_group(trans, dir);
2272 err = btrfs_orphan_add(trans, inode);
2276 /* now the directory is empty */
2277 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2278 dentry->d_name.name, dentry->d_name.len);
2280 btrfs_i_size_write(inode, 0);
2283 nr = trans->blocks_used;
2284 ret = btrfs_end_transaction_throttle(trans, root);
2286 btrfs_btree_balance_dirty(root, nr);
2295 * when truncating bytes in a file, it is possible to avoid reading
2296 * the leaves that contain only checksum items. This can be the
2297 * majority of the IO required to delete a large file, but it must
2298 * be done carefully.
2300 * The keys in the level just above the leaves are checked to make sure
2301 * the lowest key in a given leaf is a csum key, and starts at an offset
2302 * after the new size.
2304 * Then the key for the next leaf is checked to make sure it also has
2305 * a checksum item for the same file. If it does, we know our target leaf
2306 * contains only checksum items, and it can be safely freed without reading
2309 * This is just an optimization targeted at large files. It may do
2310 * nothing. It will return 0 unless things went badly.
2312 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2313 struct btrfs_root *root,
2314 struct btrfs_path *path,
2315 struct inode *inode, u64 new_size)
2317 struct btrfs_key key;
2320 struct btrfs_key found_key;
2321 struct btrfs_key other_key;
2322 struct btrfs_leaf_ref *ref;
2326 path->lowest_level = 1;
2327 key.objectid = inode->i_ino;
2328 key.type = BTRFS_CSUM_ITEM_KEY;
2329 key.offset = new_size;
2331 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2335 if (path->nodes[1] == NULL) {
2340 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2341 nritems = btrfs_header_nritems(path->nodes[1]);
2346 if (path->slots[1] >= nritems)
2349 /* did we find a key greater than anything we want to delete? */
2350 if (found_key.objectid > inode->i_ino ||
2351 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2354 /* we check the next key in the node to make sure the leave contains
2355 * only checksum items. This comparison doesn't work if our
2356 * leaf is the last one in the node
2358 if (path->slots[1] + 1 >= nritems) {
2360 /* search forward from the last key in the node, this
2361 * will bring us into the next node in the tree
2363 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2365 /* unlikely, but we inc below, so check to be safe */
2366 if (found_key.offset == (u64)-1)
2369 /* search_forward needs a path with locks held, do the
2370 * search again for the original key. It is possible
2371 * this will race with a balance and return a path that
2372 * we could modify, but this drop is just an optimization
2373 * and is allowed to miss some leaves.
2375 btrfs_release_path(root, path);
2378 /* setup a max key for search_forward */
2379 other_key.offset = (u64)-1;
2380 other_key.type = key.type;
2381 other_key.objectid = key.objectid;
2383 path->keep_locks = 1;
2384 ret = btrfs_search_forward(root, &found_key, &other_key,
2386 path->keep_locks = 0;
2387 if (ret || found_key.objectid != key.objectid ||
2388 found_key.type != key.type) {
2393 key.offset = found_key.offset;
2394 btrfs_release_path(root, path);
2399 /* we know there's one more slot after us in the tree,
2400 * read that key so we can verify it is also a checksum item
2402 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2404 if (found_key.objectid < inode->i_ino)
2407 if (found_key.type != key.type || found_key.offset < new_size)
2411 * if the key for the next leaf isn't a csum key from this objectid,
2412 * we can't be sure there aren't good items inside this leaf.
2415 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2418 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2419 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2421 * it is safe to delete this leaf, it contains only
2422 * csum items from this inode at an offset >= new_size
2424 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2427 if (root->ref_cows && leaf_gen < trans->transid) {
2428 ref = btrfs_alloc_leaf_ref(root, 0);
2430 ref->root_gen = root->root_key.offset;
2431 ref->bytenr = leaf_start;
2433 ref->generation = leaf_gen;
2436 ret = btrfs_add_leaf_ref(root, ref, 0);
2438 btrfs_free_leaf_ref(root, ref);
2444 btrfs_release_path(root, path);
2446 if (other_key.objectid == inode->i_ino &&
2447 other_key.type == key.type && other_key.offset > key.offset) {
2448 key.offset = other_key.offset;
2454 /* fixup any changes we've made to the path */
2455 path->lowest_level = 0;
2456 path->keep_locks = 0;
2457 btrfs_release_path(root, path);
2464 * this can truncate away extent items, csum items and directory items.
2465 * It starts at a high offset and removes keys until it can't find
2466 * any higher than new_size
2468 * csum items that cross the new i_size are truncated to the new size
2471 * min_type is the minimum key type to truncate down to. If set to 0, this
2472 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2474 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2475 struct btrfs_root *root,
2476 struct inode *inode,
2477 u64 new_size, u32 min_type)
2480 struct btrfs_path *path;
2481 struct btrfs_key key;
2482 struct btrfs_key found_key;
2484 struct extent_buffer *leaf;
2485 struct btrfs_file_extent_item *fi;
2486 u64 extent_start = 0;
2487 u64 extent_num_bytes = 0;
2493 int pending_del_nr = 0;
2494 int pending_del_slot = 0;
2495 int extent_type = -1;
2497 u64 mask = root->sectorsize - 1;
2500 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2501 path = btrfs_alloc_path();
2505 /* FIXME, add redo link to tree so we don't leak on crash */
2506 key.objectid = inode->i_ino;
2507 key.offset = (u64)-1;
2510 btrfs_init_path(path);
2513 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2518 /* there are no items in the tree for us to truncate, we're
2521 if (path->slots[0] == 0) {
2530 leaf = path->nodes[0];
2531 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2532 found_type = btrfs_key_type(&found_key);
2535 if (found_key.objectid != inode->i_ino)
2538 if (found_type < min_type)
2541 item_end = found_key.offset;
2542 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2543 fi = btrfs_item_ptr(leaf, path->slots[0],
2544 struct btrfs_file_extent_item);
2545 extent_type = btrfs_file_extent_type(leaf, fi);
2546 encoding = btrfs_file_extent_compression(leaf, fi);
2547 encoding |= btrfs_file_extent_encryption(leaf, fi);
2548 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2550 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2552 btrfs_file_extent_num_bytes(leaf, fi);
2553 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2554 item_end += btrfs_file_extent_inline_len(leaf,
2559 if (item_end < new_size) {
2560 if (found_type == BTRFS_DIR_ITEM_KEY)
2561 found_type = BTRFS_INODE_ITEM_KEY;
2562 else if (found_type == BTRFS_EXTENT_ITEM_KEY)
2563 found_type = BTRFS_EXTENT_DATA_KEY;
2564 else if (found_type == BTRFS_EXTENT_DATA_KEY)
2565 found_type = BTRFS_XATTR_ITEM_KEY;
2566 else if (found_type == BTRFS_XATTR_ITEM_KEY)
2567 found_type = BTRFS_INODE_REF_KEY;
2568 else if (found_type)
2572 btrfs_set_key_type(&key, found_type);
2575 if (found_key.offset >= new_size)
2581 /* FIXME, shrink the extent if the ref count is only 1 */
2582 if (found_type != BTRFS_EXTENT_DATA_KEY)
2585 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2587 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2588 if (!del_item && !encoding) {
2589 u64 orig_num_bytes =
2590 btrfs_file_extent_num_bytes(leaf, fi);
2591 extent_num_bytes = new_size -
2592 found_key.offset + root->sectorsize - 1;
2593 extent_num_bytes = extent_num_bytes &
2594 ~((u64)root->sectorsize - 1);
2595 btrfs_set_file_extent_num_bytes(leaf, fi,
2597 num_dec = (orig_num_bytes -
2599 if (root->ref_cows && extent_start != 0)
2600 inode_sub_bytes(inode, num_dec);
2601 btrfs_mark_buffer_dirty(leaf);
2604 btrfs_file_extent_disk_num_bytes(leaf,
2606 /* FIXME blocksize != 4096 */
2607 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2608 if (extent_start != 0) {
2611 inode_sub_bytes(inode, num_dec);
2613 root_gen = btrfs_header_generation(leaf);
2614 root_owner = btrfs_header_owner(leaf);
2616 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2618 * we can't truncate inline items that have had
2622 btrfs_file_extent_compression(leaf, fi) == 0 &&
2623 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2624 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2625 u32 size = new_size - found_key.offset;
2627 if (root->ref_cows) {
2628 inode_sub_bytes(inode, item_end + 1 -
2632 btrfs_file_extent_calc_inline_size(size);
2633 ret = btrfs_truncate_item(trans, root, path,
2636 } else if (root->ref_cows) {
2637 inode_sub_bytes(inode, item_end + 1 -
2643 if (!pending_del_nr) {
2644 /* no pending yet, add ourselves */
2645 pending_del_slot = path->slots[0];
2647 } else if (pending_del_nr &&
2648 path->slots[0] + 1 == pending_del_slot) {
2649 /* hop on the pending chunk */
2651 pending_del_slot = path->slots[0];
2659 ret = btrfs_free_extent(trans, root, extent_start,
2661 leaf->start, root_owner,
2662 root_gen, inode->i_ino, 0);
2666 if (path->slots[0] == 0) {
2669 btrfs_release_path(root, path);
2674 if (pending_del_nr &&
2675 path->slots[0] + 1 != pending_del_slot) {
2676 struct btrfs_key debug;
2678 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2680 ret = btrfs_del_items(trans, root, path,
2685 btrfs_release_path(root, path);
2691 if (pending_del_nr) {
2692 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2695 btrfs_free_path(path);
2696 inode->i_sb->s_dirt = 1;
2701 * taken from block_truncate_page, but does cow as it zeros out
2702 * any bytes left in the last page in the file.
2704 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2706 struct inode *inode = mapping->host;
2707 struct btrfs_root *root = BTRFS_I(inode)->root;
2708 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2709 struct btrfs_ordered_extent *ordered;
2711 u32 blocksize = root->sectorsize;
2712 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2713 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2719 if ((offset & (blocksize - 1)) == 0)
2724 page = grab_cache_page(mapping, index);
2728 page_start = page_offset(page);
2729 page_end = page_start + PAGE_CACHE_SIZE - 1;
2731 if (!PageUptodate(page)) {
2732 ret = btrfs_readpage(NULL, page);
2734 if (page->mapping != mapping) {
2736 page_cache_release(page);
2739 if (!PageUptodate(page)) {
2744 wait_on_page_writeback(page);
2746 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2747 set_page_extent_mapped(page);
2749 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2751 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2753 page_cache_release(page);
2754 btrfs_start_ordered_extent(inode, ordered, 1);
2755 btrfs_put_ordered_extent(ordered);
2759 btrfs_set_extent_delalloc(inode, page_start, page_end);
2761 if (offset != PAGE_CACHE_SIZE) {
2763 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2764 flush_dcache_page(page);
2767 ClearPageChecked(page);
2768 set_page_dirty(page);
2769 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2773 page_cache_release(page);
2778 int btrfs_cont_expand(struct inode *inode, loff_t size)
2780 struct btrfs_trans_handle *trans;
2781 struct btrfs_root *root = BTRFS_I(inode)->root;
2782 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2783 struct extent_map *em;
2784 u64 mask = root->sectorsize - 1;
2785 u64 hole_start = (inode->i_size + mask) & ~mask;
2786 u64 block_end = (size + mask) & ~mask;
2792 if (size <= hole_start)
2795 err = btrfs_check_free_space(root, 1, 0);
2799 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2802 struct btrfs_ordered_extent *ordered;
2803 btrfs_wait_ordered_range(inode, hole_start,
2804 block_end - hole_start);
2805 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2806 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2809 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2810 btrfs_put_ordered_extent(ordered);
2813 trans = btrfs_start_transaction(root, 1);
2814 btrfs_set_trans_block_group(trans, inode);
2816 cur_offset = hole_start;
2818 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2819 block_end - cur_offset, 0);
2820 BUG_ON(IS_ERR(em) || !em);
2821 last_byte = min(extent_map_end(em), block_end);
2822 last_byte = (last_byte + mask) & ~mask;
2823 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2825 hole_size = last_byte - cur_offset;
2826 err = btrfs_drop_extents(trans, root, inode,
2828 cur_offset + hole_size,
2829 cur_offset, &hint_byte);
2832 err = btrfs_insert_file_extent(trans, root,
2833 inode->i_ino, cur_offset, 0,
2834 0, hole_size, 0, hole_size,
2836 btrfs_drop_extent_cache(inode, hole_start,
2839 free_extent_map(em);
2840 cur_offset = last_byte;
2841 if (err || cur_offset >= block_end)
2845 btrfs_end_transaction(trans, root);
2846 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2850 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2852 struct inode *inode = dentry->d_inode;
2855 err = inode_change_ok(inode, attr);
2859 if (S_ISREG(inode->i_mode) &&
2860 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2861 err = btrfs_cont_expand(inode, attr->ia_size);
2866 err = inode_setattr(inode, attr);
2868 if (!err && ((attr->ia_valid & ATTR_MODE)))
2869 err = btrfs_acl_chmod(inode);
2873 void btrfs_delete_inode(struct inode *inode)
2875 struct btrfs_trans_handle *trans;
2876 struct btrfs_root *root = BTRFS_I(inode)->root;
2880 truncate_inode_pages(&inode->i_data, 0);
2881 if (is_bad_inode(inode)) {
2882 btrfs_orphan_del(NULL, inode);
2885 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2887 btrfs_i_size_write(inode, 0);
2888 trans = btrfs_join_transaction(root, 1);
2890 btrfs_set_trans_block_group(trans, inode);
2891 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2893 btrfs_orphan_del(NULL, inode);
2894 goto no_delete_lock;
2897 btrfs_orphan_del(trans, inode);
2899 nr = trans->blocks_used;
2902 btrfs_end_transaction(trans, root);
2903 btrfs_btree_balance_dirty(root, nr);
2907 nr = trans->blocks_used;
2908 btrfs_end_transaction(trans, root);
2909 btrfs_btree_balance_dirty(root, nr);
2915 * this returns the key found in the dir entry in the location pointer.
2916 * If no dir entries were found, location->objectid is 0.
2918 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2919 struct btrfs_key *location)
2921 const char *name = dentry->d_name.name;
2922 int namelen = dentry->d_name.len;
2923 struct btrfs_dir_item *di;
2924 struct btrfs_path *path;
2925 struct btrfs_root *root = BTRFS_I(dir)->root;
2928 path = btrfs_alloc_path();
2931 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2936 if (!di || IS_ERR(di))
2939 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2941 btrfs_free_path(path);
2944 location->objectid = 0;
2949 * when we hit a tree root in a directory, the btrfs part of the inode
2950 * needs to be changed to reflect the root directory of the tree root. This
2951 * is kind of like crossing a mount point.
2953 static int fixup_tree_root_location(struct btrfs_root *root,
2954 struct btrfs_key *location,
2955 struct btrfs_root **sub_root,
2956 struct dentry *dentry)
2958 struct btrfs_root_item *ri;
2960 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2962 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2965 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2966 dentry->d_name.name,
2967 dentry->d_name.len);
2968 if (IS_ERR(*sub_root))
2969 return PTR_ERR(*sub_root);
2971 ri = &(*sub_root)->root_item;
2972 location->objectid = btrfs_root_dirid(ri);
2973 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2974 location->offset = 0;
2979 static noinline void init_btrfs_i(struct inode *inode)
2981 struct btrfs_inode *bi = BTRFS_I(inode);
2984 bi->i_default_acl = NULL;
2989 bi->logged_trans = 0;
2990 bi->delalloc_bytes = 0;
2991 bi->disk_i_size = 0;
2993 bi->index_cnt = (u64)-1;
2994 bi->log_dirty_trans = 0;
2995 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2996 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2997 inode->i_mapping, GFP_NOFS);
2998 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2999 inode->i_mapping, GFP_NOFS);
3000 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3001 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3002 mutex_init(&BTRFS_I(inode)->extent_mutex);
3003 mutex_init(&BTRFS_I(inode)->log_mutex);
3006 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3008 struct btrfs_iget_args *args = p;
3009 inode->i_ino = args->ino;
3010 init_btrfs_i(inode);
3011 BTRFS_I(inode)->root = args->root;
3015 static int btrfs_find_actor(struct inode *inode, void *opaque)
3017 struct btrfs_iget_args *args = opaque;
3018 return args->ino == inode->i_ino &&
3019 args->root == BTRFS_I(inode)->root;
3022 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
3023 struct btrfs_root *root, int wait)
3025 struct inode *inode;
3026 struct btrfs_iget_args args;
3027 args.ino = objectid;
3031 inode = ilookup5(s, objectid, btrfs_find_actor,
3034 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3040 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3041 struct btrfs_root *root)
3043 struct inode *inode;
3044 struct btrfs_iget_args args;
3045 args.ino = objectid;
3048 inode = iget5_locked(s, objectid, btrfs_find_actor,
3049 btrfs_init_locked_inode,
3054 /* Get an inode object given its location and corresponding root.
3055 * Returns in *is_new if the inode was read from disk
3057 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3058 struct btrfs_root *root, int *is_new)
3060 struct inode *inode;
3062 inode = btrfs_iget_locked(s, location->objectid, root);
3064 return ERR_PTR(-EACCES);
3066 if (inode->i_state & I_NEW) {
3067 BTRFS_I(inode)->root = root;
3068 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3069 btrfs_read_locked_inode(inode);
3070 unlock_new_inode(inode);
3081 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3083 struct inode *inode;
3084 struct btrfs_inode *bi = BTRFS_I(dir);
3085 struct btrfs_root *root = bi->root;
3086 struct btrfs_root *sub_root = root;
3087 struct btrfs_key location;
3090 if (dentry->d_name.len > BTRFS_NAME_LEN)
3091 return ERR_PTR(-ENAMETOOLONG);
3093 ret = btrfs_inode_by_name(dir, dentry, &location);
3096 return ERR_PTR(ret);
3099 if (location.objectid) {
3100 ret = fixup_tree_root_location(root, &location, &sub_root,
3103 return ERR_PTR(ret);
3105 return ERR_PTR(-ENOENT);
3106 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3108 return ERR_CAST(inode);
3113 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3114 struct nameidata *nd)
3116 struct inode *inode;
3118 if (dentry->d_name.len > BTRFS_NAME_LEN)
3119 return ERR_PTR(-ENAMETOOLONG);
3121 inode = btrfs_lookup_dentry(dir, dentry);
3123 return ERR_CAST(inode);
3125 return d_splice_alias(inode, dentry);
3128 static unsigned char btrfs_filetype_table[] = {
3129 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3132 static int btrfs_real_readdir(struct file *filp, void *dirent,
3135 struct inode *inode = filp->f_dentry->d_inode;
3136 struct btrfs_root *root = BTRFS_I(inode)->root;
3137 struct btrfs_item *item;
3138 struct btrfs_dir_item *di;
3139 struct btrfs_key key;
3140 struct btrfs_key found_key;
3141 struct btrfs_path *path;
3144 struct extent_buffer *leaf;
3147 unsigned char d_type;
3152 int key_type = BTRFS_DIR_INDEX_KEY;
3157 /* FIXME, use a real flag for deciding about the key type */
3158 if (root->fs_info->tree_root == root)
3159 key_type = BTRFS_DIR_ITEM_KEY;
3161 /* special case for "." */
3162 if (filp->f_pos == 0) {
3163 over = filldir(dirent, ".", 1,
3170 /* special case for .., just use the back ref */
3171 if (filp->f_pos == 1) {
3172 u64 pino = parent_ino(filp->f_path.dentry);
3173 over = filldir(dirent, "..", 2,
3179 path = btrfs_alloc_path();
3182 btrfs_set_key_type(&key, key_type);
3183 key.offset = filp->f_pos;
3184 key.objectid = inode->i_ino;
3186 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3192 leaf = path->nodes[0];
3193 nritems = btrfs_header_nritems(leaf);
3194 slot = path->slots[0];
3195 if (advance || slot >= nritems) {
3196 if (slot >= nritems - 1) {
3197 ret = btrfs_next_leaf(root, path);
3200 leaf = path->nodes[0];
3201 nritems = btrfs_header_nritems(leaf);
3202 slot = path->slots[0];
3210 item = btrfs_item_nr(leaf, slot);
3211 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3213 if (found_key.objectid != key.objectid)
3215 if (btrfs_key_type(&found_key) != key_type)
3217 if (found_key.offset < filp->f_pos)
3220 filp->f_pos = found_key.offset;
3222 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3224 di_total = btrfs_item_size(leaf, item);
3226 while (di_cur < di_total) {
3227 struct btrfs_key location;
3229 name_len = btrfs_dir_name_len(leaf, di);
3230 if (name_len <= sizeof(tmp_name)) {
3231 name_ptr = tmp_name;
3233 name_ptr = kmalloc(name_len, GFP_NOFS);
3239 read_extent_buffer(leaf, name_ptr,
3240 (unsigned long)(di + 1), name_len);
3242 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3243 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3245 /* is this a reference to our own snapshot? If so
3248 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3249 location.objectid == root->root_key.objectid) {
3253 over = filldir(dirent, name_ptr, name_len,
3254 found_key.offset, location.objectid,
3258 if (name_ptr != tmp_name)
3263 di_len = btrfs_dir_name_len(leaf, di) +
3264 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3266 di = (struct btrfs_dir_item *)((char *)di + di_len);
3270 /* Reached end of directory/root. Bump pos past the last item. */
3271 if (key_type == BTRFS_DIR_INDEX_KEY)
3272 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3278 btrfs_free_path(path);
3282 int btrfs_write_inode(struct inode *inode, int wait)
3284 struct btrfs_root *root = BTRFS_I(inode)->root;
3285 struct btrfs_trans_handle *trans;
3288 if (root->fs_info->btree_inode == inode)
3292 trans = btrfs_join_transaction(root, 1);
3293 btrfs_set_trans_block_group(trans, inode);
3294 ret = btrfs_commit_transaction(trans, root);
3300 * This is somewhat expensive, updating the tree every time the
3301 * inode changes. But, it is most likely to find the inode in cache.
3302 * FIXME, needs more benchmarking...there are no reasons other than performance
3303 * to keep or drop this code.
3305 void btrfs_dirty_inode(struct inode *inode)
3307 struct btrfs_root *root = BTRFS_I(inode)->root;
3308 struct btrfs_trans_handle *trans;
3310 trans = btrfs_join_transaction(root, 1);
3311 btrfs_set_trans_block_group(trans, inode);
3312 btrfs_update_inode(trans, root, inode);
3313 btrfs_end_transaction(trans, root);
3317 * find the highest existing sequence number in a directory
3318 * and then set the in-memory index_cnt variable to reflect
3319 * free sequence numbers
3321 static int btrfs_set_inode_index_count(struct inode *inode)
3323 struct btrfs_root *root = BTRFS_I(inode)->root;
3324 struct btrfs_key key, found_key;
3325 struct btrfs_path *path;
3326 struct extent_buffer *leaf;
3329 key.objectid = inode->i_ino;
3330 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3331 key.offset = (u64)-1;
3333 path = btrfs_alloc_path();
3337 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3340 /* FIXME: we should be able to handle this */
3346 * MAGIC NUMBER EXPLANATION:
3347 * since we search a directory based on f_pos we have to start at 2
3348 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3349 * else has to start at 2
3351 if (path->slots[0] == 0) {
3352 BTRFS_I(inode)->index_cnt = 2;
3358 leaf = path->nodes[0];
3359 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3361 if (found_key.objectid != inode->i_ino ||
3362 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3363 BTRFS_I(inode)->index_cnt = 2;
3367 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3369 btrfs_free_path(path);
3374 * helper to find a free sequence number in a given directory. This current
3375 * code is very simple, later versions will do smarter things in the btree
3377 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3381 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3382 ret = btrfs_set_inode_index_count(dir);
3387 *index = BTRFS_I(dir)->index_cnt;
3388 BTRFS_I(dir)->index_cnt++;
3393 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3394 struct btrfs_root *root,
3396 const char *name, int name_len,
3397 u64 ref_objectid, u64 objectid,
3398 u64 alloc_hint, int mode, u64 *index)
3400 struct inode *inode;
3401 struct btrfs_inode_item *inode_item;
3402 struct btrfs_key *location;
3403 struct btrfs_path *path;
3404 struct btrfs_inode_ref *ref;
3405 struct btrfs_key key[2];
3411 path = btrfs_alloc_path();
3414 inode = new_inode(root->fs_info->sb);
3416 return ERR_PTR(-ENOMEM);
3419 ret = btrfs_set_inode_index(dir, index);
3421 return ERR_PTR(ret);
3424 * index_cnt is ignored for everything but a dir,
3425 * btrfs_get_inode_index_count has an explanation for the magic
3428 init_btrfs_i(inode);
3429 BTRFS_I(inode)->index_cnt = 2;
3430 BTRFS_I(inode)->root = root;
3431 BTRFS_I(inode)->generation = trans->transid;
3437 BTRFS_I(inode)->block_group =
3438 btrfs_find_block_group(root, 0, alloc_hint, owner);
3439 if ((mode & S_IFREG)) {
3440 if (btrfs_test_opt(root, NODATASUM))
3441 btrfs_set_flag(inode, NODATASUM);
3442 if (btrfs_test_opt(root, NODATACOW))
3443 btrfs_set_flag(inode, NODATACOW);
3446 key[0].objectid = objectid;
3447 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3450 key[1].objectid = objectid;
3451 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3452 key[1].offset = ref_objectid;
3454 sizes[0] = sizeof(struct btrfs_inode_item);
3455 sizes[1] = name_len + sizeof(*ref);
3457 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3461 if (objectid > root->highest_inode)
3462 root->highest_inode = objectid;
3464 inode->i_uid = current_fsuid();
3465 inode->i_gid = current_fsgid();
3466 inode->i_mode = mode;
3467 inode->i_ino = objectid;
3468 inode_set_bytes(inode, 0);
3469 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3470 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3471 struct btrfs_inode_item);
3472 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3474 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3475 struct btrfs_inode_ref);
3476 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3477 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3478 ptr = (unsigned long)(ref + 1);
3479 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3481 btrfs_mark_buffer_dirty(path->nodes[0]);
3482 btrfs_free_path(path);
3484 location = &BTRFS_I(inode)->location;
3485 location->objectid = objectid;
3486 location->offset = 0;
3487 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3489 insert_inode_hash(inode);
3493 BTRFS_I(dir)->index_cnt--;
3494 btrfs_free_path(path);
3495 return ERR_PTR(ret);
3498 static inline u8 btrfs_inode_type(struct inode *inode)
3500 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3504 * utility function to add 'inode' into 'parent_inode' with
3505 * a give name and a given sequence number.
3506 * if 'add_backref' is true, also insert a backref from the
3507 * inode to the parent directory.
3509 int btrfs_add_link(struct btrfs_trans_handle *trans,
3510 struct inode *parent_inode, struct inode *inode,
3511 const char *name, int name_len, int add_backref, u64 index)
3514 struct btrfs_key key;
3515 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3517 key.objectid = inode->i_ino;
3518 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3521 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3522 parent_inode->i_ino,
3523 &key, btrfs_inode_type(inode),
3527 ret = btrfs_insert_inode_ref(trans, root,
3530 parent_inode->i_ino,
3533 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3535 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3536 ret = btrfs_update_inode(trans, root, parent_inode);
3541 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3542 struct dentry *dentry, struct inode *inode,
3543 int backref, u64 index)
3545 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3546 inode, dentry->d_name.name,
3547 dentry->d_name.len, backref, index);
3549 d_instantiate(dentry, inode);
3557 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3558 int mode, dev_t rdev)
3560 struct btrfs_trans_handle *trans;
3561 struct btrfs_root *root = BTRFS_I(dir)->root;
3562 struct inode *inode = NULL;
3566 unsigned long nr = 0;
3569 if (!new_valid_dev(rdev))
3572 err = btrfs_check_free_space(root, 1, 0);
3576 trans = btrfs_start_transaction(root, 1);
3577 btrfs_set_trans_block_group(trans, dir);
3579 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3585 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3587 dentry->d_parent->d_inode->i_ino, objectid,
3588 BTRFS_I(dir)->block_group, mode, &index);
3589 err = PTR_ERR(inode);
3593 err = btrfs_init_acl(inode, dir);
3599 btrfs_set_trans_block_group(trans, inode);
3600 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3604 inode->i_op = &btrfs_special_inode_operations;
3605 init_special_inode(inode, inode->i_mode, rdev);
3606 btrfs_update_inode(trans, root, inode);
3608 dir->i_sb->s_dirt = 1;
3609 btrfs_update_inode_block_group(trans, inode);
3610 btrfs_update_inode_block_group(trans, dir);
3612 nr = trans->blocks_used;
3613 btrfs_end_transaction_throttle(trans, root);
3616 inode_dec_link_count(inode);
3619 btrfs_btree_balance_dirty(root, nr);
3623 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3624 int mode, struct nameidata *nd)
3626 struct btrfs_trans_handle *trans;
3627 struct btrfs_root *root = BTRFS_I(dir)->root;
3628 struct inode *inode = NULL;
3631 unsigned long nr = 0;
3635 err = btrfs_check_free_space(root, 1, 0);
3638 trans = btrfs_start_transaction(root, 1);
3639 btrfs_set_trans_block_group(trans, dir);
3641 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3647 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3649 dentry->d_parent->d_inode->i_ino,
3650 objectid, BTRFS_I(dir)->block_group, mode,
3652 err = PTR_ERR(inode);
3656 err = btrfs_init_acl(inode, dir);
3662 btrfs_set_trans_block_group(trans, inode);
3663 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3667 inode->i_mapping->a_ops = &btrfs_aops;
3668 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3669 inode->i_fop = &btrfs_file_operations;
3670 inode->i_op = &btrfs_file_inode_operations;
3671 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3673 dir->i_sb->s_dirt = 1;
3674 btrfs_update_inode_block_group(trans, inode);
3675 btrfs_update_inode_block_group(trans, dir);
3677 nr = trans->blocks_used;
3678 btrfs_end_transaction_throttle(trans, root);
3681 inode_dec_link_count(inode);
3684 btrfs_btree_balance_dirty(root, nr);
3688 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3689 struct dentry *dentry)
3691 struct btrfs_trans_handle *trans;
3692 struct btrfs_root *root = BTRFS_I(dir)->root;
3693 struct inode *inode = old_dentry->d_inode;
3695 unsigned long nr = 0;
3699 if (inode->i_nlink == 0)
3702 btrfs_inc_nlink(inode);
3703 err = btrfs_check_free_space(root, 1, 0);
3706 err = btrfs_set_inode_index(dir, &index);
3710 trans = btrfs_start_transaction(root, 1);
3712 btrfs_set_trans_block_group(trans, dir);
3713 atomic_inc(&inode->i_count);
3715 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3720 dir->i_sb->s_dirt = 1;
3721 btrfs_update_inode_block_group(trans, dir);
3722 err = btrfs_update_inode(trans, root, inode);
3727 nr = trans->blocks_used;
3728 btrfs_end_transaction_throttle(trans, root);
3731 inode_dec_link_count(inode);
3734 btrfs_btree_balance_dirty(root, nr);
3738 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3740 struct inode *inode = NULL;
3741 struct btrfs_trans_handle *trans;
3742 struct btrfs_root *root = BTRFS_I(dir)->root;
3744 int drop_on_err = 0;
3747 unsigned long nr = 1;
3749 err = btrfs_check_free_space(root, 1, 0);
3753 trans = btrfs_start_transaction(root, 1);
3754 btrfs_set_trans_block_group(trans, dir);
3756 if (IS_ERR(trans)) {
3757 err = PTR_ERR(trans);
3761 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3767 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3769 dentry->d_parent->d_inode->i_ino, objectid,
3770 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3772 if (IS_ERR(inode)) {
3773 err = PTR_ERR(inode);
3779 err = btrfs_init_acl(inode, dir);
3783 inode->i_op = &btrfs_dir_inode_operations;
3784 inode->i_fop = &btrfs_dir_file_operations;
3785 btrfs_set_trans_block_group(trans, inode);
3787 btrfs_i_size_write(inode, 0);
3788 err = btrfs_update_inode(trans, root, inode);
3792 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3793 inode, dentry->d_name.name,
3794 dentry->d_name.len, 0, index);
3798 d_instantiate(dentry, inode);
3800 dir->i_sb->s_dirt = 1;
3801 btrfs_update_inode_block_group(trans, inode);
3802 btrfs_update_inode_block_group(trans, dir);
3805 nr = trans->blocks_used;
3806 btrfs_end_transaction_throttle(trans, root);
3811 btrfs_btree_balance_dirty(root, nr);
3815 /* helper for btfs_get_extent. Given an existing extent in the tree,
3816 * and an extent that you want to insert, deal with overlap and insert
3817 * the new extent into the tree.
3819 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3820 struct extent_map *existing,
3821 struct extent_map *em,
3822 u64 map_start, u64 map_len)
3826 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3827 start_diff = map_start - em->start;
3828 em->start = map_start;
3830 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3831 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3832 em->block_start += start_diff;
3833 em->block_len -= start_diff;
3835 return add_extent_mapping(em_tree, em);
3838 static noinline int uncompress_inline(struct btrfs_path *path,
3839 struct inode *inode, struct page *page,
3840 size_t pg_offset, u64 extent_offset,
3841 struct btrfs_file_extent_item *item)
3844 struct extent_buffer *leaf = path->nodes[0];
3847 unsigned long inline_size;
3850 WARN_ON(pg_offset != 0);
3851 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3852 inline_size = btrfs_file_extent_inline_item_len(leaf,
3853 btrfs_item_nr(leaf, path->slots[0]));
3854 tmp = kmalloc(inline_size, GFP_NOFS);
3855 ptr = btrfs_file_extent_inline_start(item);
3857 read_extent_buffer(leaf, tmp, ptr, inline_size);
3859 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3860 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3861 inline_size, max_size);
3863 char *kaddr = kmap_atomic(page, KM_USER0);
3864 unsigned long copy_size = min_t(u64,
3865 PAGE_CACHE_SIZE - pg_offset,
3866 max_size - extent_offset);
3867 memset(kaddr + pg_offset, 0, copy_size);
3868 kunmap_atomic(kaddr, KM_USER0);
3875 * a bit scary, this does extent mapping from logical file offset to the disk.
3876 * the ugly parts come from merging extents from the disk with the in-ram
3877 * representation. This gets more complex because of the data=ordered code,
3878 * where the in-ram extents might be locked pending data=ordered completion.
3880 * This also copies inline extents directly into the page.
3883 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3884 size_t pg_offset, u64 start, u64 len,
3890 u64 extent_start = 0;
3892 u64 objectid = inode->i_ino;
3894 struct btrfs_path *path = NULL;
3895 struct btrfs_root *root = BTRFS_I(inode)->root;
3896 struct btrfs_file_extent_item *item;
3897 struct extent_buffer *leaf;
3898 struct btrfs_key found_key;
3899 struct extent_map *em = NULL;
3900 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3901 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3902 struct btrfs_trans_handle *trans = NULL;
3906 spin_lock(&em_tree->lock);
3907 em = lookup_extent_mapping(em_tree, start, len);
3909 em->bdev = root->fs_info->fs_devices->latest_bdev;
3910 spin_unlock(&em_tree->lock);
3913 if (em->start > start || em->start + em->len <= start)
3914 free_extent_map(em);
3915 else if (em->block_start == EXTENT_MAP_INLINE && page)
3916 free_extent_map(em);
3920 em = alloc_extent_map(GFP_NOFS);
3925 em->bdev = root->fs_info->fs_devices->latest_bdev;
3926 em->start = EXTENT_MAP_HOLE;
3927 em->orig_start = EXTENT_MAP_HOLE;
3929 em->block_len = (u64)-1;
3932 path = btrfs_alloc_path();
3936 ret = btrfs_lookup_file_extent(trans, root, path,
3937 objectid, start, trans != NULL);
3944 if (path->slots[0] == 0)
3949 leaf = path->nodes[0];
3950 item = btrfs_item_ptr(leaf, path->slots[0],
3951 struct btrfs_file_extent_item);
3952 /* are we inside the extent that was found? */
3953 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3954 found_type = btrfs_key_type(&found_key);
3955 if (found_key.objectid != objectid ||
3956 found_type != BTRFS_EXTENT_DATA_KEY) {
3960 found_type = btrfs_file_extent_type(leaf, item);
3961 extent_start = found_key.offset;
3962 compressed = btrfs_file_extent_compression(leaf, item);
3963 if (found_type == BTRFS_FILE_EXTENT_REG ||
3964 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3965 extent_end = extent_start +
3966 btrfs_file_extent_num_bytes(leaf, item);
3967 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3969 size = btrfs_file_extent_inline_len(leaf, item);
3970 extent_end = (extent_start + size + root->sectorsize - 1) &
3971 ~((u64)root->sectorsize - 1);
3974 if (start >= extent_end) {
3976 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3977 ret = btrfs_next_leaf(root, path);
3984 leaf = path->nodes[0];
3986 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3987 if (found_key.objectid != objectid ||
3988 found_key.type != BTRFS_EXTENT_DATA_KEY)
3990 if (start + len <= found_key.offset)
3993 em->len = found_key.offset - start;
3997 if (found_type == BTRFS_FILE_EXTENT_REG ||
3998 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3999 em->start = extent_start;
4000 em->len = extent_end - extent_start;
4001 em->orig_start = extent_start -
4002 btrfs_file_extent_offset(leaf, item);
4003 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4005 em->block_start = EXTENT_MAP_HOLE;
4009 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4010 em->block_start = bytenr;
4011 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4014 bytenr += btrfs_file_extent_offset(leaf, item);
4015 em->block_start = bytenr;
4016 em->block_len = em->len;
4017 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4018 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4021 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4025 size_t extent_offset;
4028 em->block_start = EXTENT_MAP_INLINE;
4029 if (!page || create) {
4030 em->start = extent_start;
4031 em->len = extent_end - extent_start;
4035 size = btrfs_file_extent_inline_len(leaf, item);
4036 extent_offset = page_offset(page) + pg_offset - extent_start;
4037 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4038 size - extent_offset);
4039 em->start = extent_start + extent_offset;
4040 em->len = (copy_size + root->sectorsize - 1) &
4041 ~((u64)root->sectorsize - 1);
4042 em->orig_start = EXTENT_MAP_INLINE;
4044 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4045 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4046 if (create == 0 && !PageUptodate(page)) {
4047 if (btrfs_file_extent_compression(leaf, item) ==
4048 BTRFS_COMPRESS_ZLIB) {
4049 ret = uncompress_inline(path, inode, page,
4051 extent_offset, item);
4055 read_extent_buffer(leaf, map + pg_offset, ptr,
4059 flush_dcache_page(page);
4060 } else if (create && PageUptodate(page)) {
4063 free_extent_map(em);
4065 btrfs_release_path(root, path);
4066 trans = btrfs_join_transaction(root, 1);
4070 write_extent_buffer(leaf, map + pg_offset, ptr,
4073 btrfs_mark_buffer_dirty(leaf);
4075 set_extent_uptodate(io_tree, em->start,
4076 extent_map_end(em) - 1, GFP_NOFS);
4079 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4086 em->block_start = EXTENT_MAP_HOLE;
4087 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4089 btrfs_release_path(root, path);
4090 if (em->start > start || extent_map_end(em) <= start) {
4091 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4092 "[%llu %llu]\n", (unsigned long long)em->start,
4093 (unsigned long long)em->len,
4094 (unsigned long long)start,
4095 (unsigned long long)len);
4101 spin_lock(&em_tree->lock);
4102 ret = add_extent_mapping(em_tree, em);
4103 /* it is possible that someone inserted the extent into the tree
4104 * while we had the lock dropped. It is also possible that
4105 * an overlapping map exists in the tree
4107 if (ret == -EEXIST) {
4108 struct extent_map *existing;
4112 existing = lookup_extent_mapping(em_tree, start, len);
4113 if (existing && (existing->start > start ||
4114 existing->start + existing->len <= start)) {
4115 free_extent_map(existing);
4119 existing = lookup_extent_mapping(em_tree, em->start,
4122 err = merge_extent_mapping(em_tree, existing,
4125 free_extent_map(existing);
4127 free_extent_map(em);
4132 free_extent_map(em);
4136 free_extent_map(em);
4141 spin_unlock(&em_tree->lock);
4144 btrfs_free_path(path);
4146 ret = btrfs_end_transaction(trans, root);
4151 free_extent_map(em);
4153 return ERR_PTR(err);
4158 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4159 const struct iovec *iov, loff_t offset,
4160 unsigned long nr_segs)
4165 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4167 return extent_bmap(mapping, iblock, btrfs_get_extent);
4170 int btrfs_readpage(struct file *file, struct page *page)
4172 struct extent_io_tree *tree;
4173 tree = &BTRFS_I(page->mapping->host)->io_tree;
4174 return extent_read_full_page(tree, page, btrfs_get_extent);
4177 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4179 struct extent_io_tree *tree;
4182 if (current->flags & PF_MEMALLOC) {
4183 redirty_page_for_writepage(wbc, page);
4187 tree = &BTRFS_I(page->mapping->host)->io_tree;
4188 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4191 int btrfs_writepages(struct address_space *mapping,
4192 struct writeback_control *wbc)
4194 struct extent_io_tree *tree;
4196 tree = &BTRFS_I(mapping->host)->io_tree;
4197 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4201 btrfs_readpages(struct file *file, struct address_space *mapping,
4202 struct list_head *pages, unsigned nr_pages)
4204 struct extent_io_tree *tree;
4205 tree = &BTRFS_I(mapping->host)->io_tree;
4206 return extent_readpages(tree, mapping, pages, nr_pages,
4209 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4211 struct extent_io_tree *tree;
4212 struct extent_map_tree *map;
4215 tree = &BTRFS_I(page->mapping->host)->io_tree;
4216 map = &BTRFS_I(page->mapping->host)->extent_tree;
4217 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4219 ClearPagePrivate(page);
4220 set_page_private(page, 0);
4221 page_cache_release(page);
4226 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4228 if (PageWriteback(page) || PageDirty(page))
4230 return __btrfs_releasepage(page, gfp_flags);
4233 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4235 struct extent_io_tree *tree;
4236 struct btrfs_ordered_extent *ordered;
4237 u64 page_start = page_offset(page);
4238 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4240 wait_on_page_writeback(page);
4241 tree = &BTRFS_I(page->mapping->host)->io_tree;
4243 btrfs_releasepage(page, GFP_NOFS);
4247 lock_extent(tree, page_start, page_end, GFP_NOFS);
4248 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4252 * IO on this page will never be started, so we need
4253 * to account for any ordered extents now
4255 clear_extent_bit(tree, page_start, page_end,
4256 EXTENT_DIRTY | EXTENT_DELALLOC |
4257 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4258 btrfs_finish_ordered_io(page->mapping->host,
4259 page_start, page_end);
4260 btrfs_put_ordered_extent(ordered);
4261 lock_extent(tree, page_start, page_end, GFP_NOFS);
4263 clear_extent_bit(tree, page_start, page_end,
4264 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4267 __btrfs_releasepage(page, GFP_NOFS);
4269 ClearPageChecked(page);
4270 if (PagePrivate(page)) {
4271 ClearPagePrivate(page);
4272 set_page_private(page, 0);
4273 page_cache_release(page);
4278 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4279 * called from a page fault handler when a page is first dirtied. Hence we must
4280 * be careful to check for EOF conditions here. We set the page up correctly
4281 * for a written page which means we get ENOSPC checking when writing into
4282 * holes and correct delalloc and unwritten extent mapping on filesystems that
4283 * support these features.
4285 * We are not allowed to take the i_mutex here so we have to play games to
4286 * protect against truncate races as the page could now be beyond EOF. Because
4287 * vmtruncate() writes the inode size before removing pages, once we have the
4288 * page lock we can determine safely if the page is beyond EOF. If it is not
4289 * beyond EOF, then the page is guaranteed safe against truncation until we
4292 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4294 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4295 struct btrfs_root *root = BTRFS_I(inode)->root;
4296 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4297 struct btrfs_ordered_extent *ordered;
4299 unsigned long zero_start;
4305 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4312 size = i_size_read(inode);
4313 page_start = page_offset(page);
4314 page_end = page_start + PAGE_CACHE_SIZE - 1;
4316 if ((page->mapping != inode->i_mapping) ||
4317 (page_start >= size)) {
4318 /* page got truncated out from underneath us */
4321 wait_on_page_writeback(page);
4323 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4324 set_page_extent_mapped(page);
4327 * we can't set the delalloc bits if there are pending ordered
4328 * extents. Drop our locks and wait for them to finish
4330 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4332 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4334 btrfs_start_ordered_extent(inode, ordered, 1);
4335 btrfs_put_ordered_extent(ordered);
4339 btrfs_set_extent_delalloc(inode, page_start, page_end);
4342 /* page is wholly or partially inside EOF */
4343 if (page_start + PAGE_CACHE_SIZE > size)
4344 zero_start = size & ~PAGE_CACHE_MASK;
4346 zero_start = PAGE_CACHE_SIZE;
4348 if (zero_start != PAGE_CACHE_SIZE) {
4350 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4351 flush_dcache_page(page);
4354 ClearPageChecked(page);
4355 set_page_dirty(page);
4356 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4364 static void btrfs_truncate(struct inode *inode)
4366 struct btrfs_root *root = BTRFS_I(inode)->root;
4368 struct btrfs_trans_handle *trans;
4370 u64 mask = root->sectorsize - 1;
4372 if (!S_ISREG(inode->i_mode))
4374 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4377 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4378 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4380 trans = btrfs_start_transaction(root, 1);
4381 btrfs_set_trans_block_group(trans, inode);
4382 btrfs_i_size_write(inode, inode->i_size);
4384 ret = btrfs_orphan_add(trans, inode);
4387 /* FIXME, add redo link to tree so we don't leak on crash */
4388 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4389 BTRFS_EXTENT_DATA_KEY);
4390 btrfs_update_inode(trans, root, inode);
4392 ret = btrfs_orphan_del(trans, inode);
4396 nr = trans->blocks_used;
4397 ret = btrfs_end_transaction_throttle(trans, root);
4399 btrfs_btree_balance_dirty(root, nr);
4403 * create a new subvolume directory/inode (helper for the ioctl).
4405 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
4406 struct btrfs_root *new_root, struct dentry *dentry,
4407 u64 new_dirid, u64 alloc_hint)
4409 struct inode *inode;
4413 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4414 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
4416 return PTR_ERR(inode);
4417 inode->i_op = &btrfs_dir_inode_operations;
4418 inode->i_fop = &btrfs_dir_file_operations;
4421 btrfs_i_size_write(inode, 0);
4423 error = btrfs_update_inode(trans, new_root, inode);
4427 d_instantiate(dentry, inode);
4431 /* helper function for file defrag and space balancing. This
4432 * forces readahead on a given range of bytes in an inode
4434 unsigned long btrfs_force_ra(struct address_space *mapping,
4435 struct file_ra_state *ra, struct file *file,
4436 pgoff_t offset, pgoff_t last_index)
4438 pgoff_t req_size = last_index - offset + 1;
4440 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4441 return offset + req_size;
4444 struct inode *btrfs_alloc_inode(struct super_block *sb)
4446 struct btrfs_inode *ei;
4448 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4452 ei->logged_trans = 0;
4453 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4454 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4455 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4456 INIT_LIST_HEAD(&ei->i_orphan);
4457 return &ei->vfs_inode;
4460 void btrfs_destroy_inode(struct inode *inode)
4462 struct btrfs_ordered_extent *ordered;
4463 WARN_ON(!list_empty(&inode->i_dentry));
4464 WARN_ON(inode->i_data.nrpages);
4466 if (BTRFS_I(inode)->i_acl &&
4467 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4468 posix_acl_release(BTRFS_I(inode)->i_acl);
4469 if (BTRFS_I(inode)->i_default_acl &&
4470 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4471 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4473 spin_lock(&BTRFS_I(inode)->root->list_lock);
4474 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4475 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4476 " list\n", inode->i_ino);
4479 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4482 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4486 printk(KERN_ERR "btrfs found ordered "
4487 "extent %llu %llu on inode cleanup\n",
4488 (unsigned long long)ordered->file_offset,
4489 (unsigned long long)ordered->len);
4490 btrfs_remove_ordered_extent(inode, ordered);
4491 btrfs_put_ordered_extent(ordered);
4492 btrfs_put_ordered_extent(ordered);
4495 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4496 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4499 static void init_once(void *foo)
4501 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4503 inode_init_once(&ei->vfs_inode);
4506 void btrfs_destroy_cachep(void)
4508 if (btrfs_inode_cachep)
4509 kmem_cache_destroy(btrfs_inode_cachep);
4510 if (btrfs_trans_handle_cachep)
4511 kmem_cache_destroy(btrfs_trans_handle_cachep);
4512 if (btrfs_transaction_cachep)
4513 kmem_cache_destroy(btrfs_transaction_cachep);
4514 if (btrfs_bit_radix_cachep)
4515 kmem_cache_destroy(btrfs_bit_radix_cachep);
4516 if (btrfs_path_cachep)
4517 kmem_cache_destroy(btrfs_path_cachep);
4520 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4521 unsigned long extra_flags,
4522 void (*ctor)(void *))
4524 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4525 SLAB_MEM_SPREAD | extra_flags), ctor);
4528 int btrfs_init_cachep(void)
4530 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4531 sizeof(struct btrfs_inode),
4533 if (!btrfs_inode_cachep)
4535 btrfs_trans_handle_cachep =
4536 btrfs_cache_create("btrfs_trans_handle_cache",
4537 sizeof(struct btrfs_trans_handle),
4539 if (!btrfs_trans_handle_cachep)
4541 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4542 sizeof(struct btrfs_transaction),
4544 if (!btrfs_transaction_cachep)
4546 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4547 sizeof(struct btrfs_path),
4549 if (!btrfs_path_cachep)
4551 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4552 SLAB_DESTROY_BY_RCU, NULL);
4553 if (!btrfs_bit_radix_cachep)
4557 btrfs_destroy_cachep();
4561 static int btrfs_getattr(struct vfsmount *mnt,
4562 struct dentry *dentry, struct kstat *stat)
4564 struct inode *inode = dentry->d_inode;
4565 generic_fillattr(inode, stat);
4566 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4567 stat->blksize = PAGE_CACHE_SIZE;
4568 stat->blocks = (inode_get_bytes(inode) +
4569 BTRFS_I(inode)->delalloc_bytes) >> 9;
4573 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4574 struct inode *new_dir, struct dentry *new_dentry)
4576 struct btrfs_trans_handle *trans;
4577 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4578 struct inode *new_inode = new_dentry->d_inode;
4579 struct inode *old_inode = old_dentry->d_inode;
4580 struct timespec ctime = CURRENT_TIME;
4584 /* we're not allowed to rename between subvolumes */
4585 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4586 BTRFS_I(new_dir)->root->root_key.objectid)
4589 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4590 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4594 /* to rename a snapshot or subvolume, we need to juggle the
4595 * backrefs. This isn't coded yet
4597 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4600 ret = btrfs_check_free_space(root, 1, 0);
4604 trans = btrfs_start_transaction(root, 1);
4606 btrfs_set_trans_block_group(trans, new_dir);
4608 btrfs_inc_nlink(old_dentry->d_inode);
4609 old_dir->i_ctime = old_dir->i_mtime = ctime;
4610 new_dir->i_ctime = new_dir->i_mtime = ctime;
4611 old_inode->i_ctime = ctime;
4613 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4614 old_dentry->d_name.name,
4615 old_dentry->d_name.len);
4620 new_inode->i_ctime = CURRENT_TIME;
4621 ret = btrfs_unlink_inode(trans, root, new_dir,
4622 new_dentry->d_inode,
4623 new_dentry->d_name.name,
4624 new_dentry->d_name.len);
4627 if (new_inode->i_nlink == 0) {
4628 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4634 ret = btrfs_set_inode_index(new_dir, &index);
4638 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4639 old_inode, new_dentry->d_name.name,
4640 new_dentry->d_name.len, 1, index);
4645 btrfs_end_transaction_throttle(trans, root);
4651 * some fairly slow code that needs optimization. This walks the list
4652 * of all the inodes with pending delalloc and forces them to disk.
4654 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4656 struct list_head *head = &root->fs_info->delalloc_inodes;
4657 struct btrfs_inode *binode;
4658 struct inode *inode;
4660 if (root->fs_info->sb->s_flags & MS_RDONLY)
4663 spin_lock(&root->fs_info->delalloc_lock);
4664 while (!list_empty(head)) {
4665 binode = list_entry(head->next, struct btrfs_inode,
4667 inode = igrab(&binode->vfs_inode);
4669 list_del_init(&binode->delalloc_inodes);
4670 spin_unlock(&root->fs_info->delalloc_lock);
4672 filemap_flush(inode->i_mapping);
4676 spin_lock(&root->fs_info->delalloc_lock);
4678 spin_unlock(&root->fs_info->delalloc_lock);
4680 /* the filemap_flush will queue IO into the worker threads, but
4681 * we have to make sure the IO is actually started and that
4682 * ordered extents get created before we return
4684 atomic_inc(&root->fs_info->async_submit_draining);
4685 while (atomic_read(&root->fs_info->nr_async_submits) ||
4686 atomic_read(&root->fs_info->async_delalloc_pages)) {
4687 wait_event(root->fs_info->async_submit_wait,
4688 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4689 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4691 atomic_dec(&root->fs_info->async_submit_draining);
4695 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4696 const char *symname)
4698 struct btrfs_trans_handle *trans;
4699 struct btrfs_root *root = BTRFS_I(dir)->root;
4700 struct btrfs_path *path;
4701 struct btrfs_key key;
4702 struct inode *inode = NULL;
4710 struct btrfs_file_extent_item *ei;
4711 struct extent_buffer *leaf;
4712 unsigned long nr = 0;
4714 name_len = strlen(symname) + 1;
4715 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4716 return -ENAMETOOLONG;
4718 err = btrfs_check_free_space(root, 1, 0);
4722 trans = btrfs_start_transaction(root, 1);
4723 btrfs_set_trans_block_group(trans, dir);
4725 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4731 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4733 dentry->d_parent->d_inode->i_ino, objectid,
4734 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4736 err = PTR_ERR(inode);
4740 err = btrfs_init_acl(inode, dir);
4746 btrfs_set_trans_block_group(trans, inode);
4747 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4751 inode->i_mapping->a_ops = &btrfs_aops;
4752 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4753 inode->i_fop = &btrfs_file_operations;
4754 inode->i_op = &btrfs_file_inode_operations;
4755 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4757 dir->i_sb->s_dirt = 1;
4758 btrfs_update_inode_block_group(trans, inode);
4759 btrfs_update_inode_block_group(trans, dir);
4763 path = btrfs_alloc_path();
4765 key.objectid = inode->i_ino;
4767 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4768 datasize = btrfs_file_extent_calc_inline_size(name_len);
4769 err = btrfs_insert_empty_item(trans, root, path, &key,
4775 leaf = path->nodes[0];
4776 ei = btrfs_item_ptr(leaf, path->slots[0],
4777 struct btrfs_file_extent_item);
4778 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4779 btrfs_set_file_extent_type(leaf, ei,
4780 BTRFS_FILE_EXTENT_INLINE);
4781 btrfs_set_file_extent_encryption(leaf, ei, 0);
4782 btrfs_set_file_extent_compression(leaf, ei, 0);
4783 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4784 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4786 ptr = btrfs_file_extent_inline_start(ei);
4787 write_extent_buffer(leaf, symname, ptr, name_len);
4788 btrfs_mark_buffer_dirty(leaf);
4789 btrfs_free_path(path);
4791 inode->i_op = &btrfs_symlink_inode_operations;
4792 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4793 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4794 inode_set_bytes(inode, name_len);
4795 btrfs_i_size_write(inode, name_len - 1);
4796 err = btrfs_update_inode(trans, root, inode);
4801 nr = trans->blocks_used;
4802 btrfs_end_transaction_throttle(trans, root);
4805 inode_dec_link_count(inode);
4808 btrfs_btree_balance_dirty(root, nr);
4812 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4813 u64 alloc_hint, int mode)
4815 struct btrfs_trans_handle *trans;
4816 struct btrfs_root *root = BTRFS_I(inode)->root;
4817 struct btrfs_key ins;
4819 u64 cur_offset = start;
4820 u64 num_bytes = end - start;
4823 trans = btrfs_join_transaction(root, 1);
4825 btrfs_set_trans_block_group(trans, inode);
4827 while (num_bytes > 0) {
4828 alloc_size = min(num_bytes, root->fs_info->max_extent);
4829 ret = btrfs_reserve_extent(trans, root, alloc_size,
4830 root->sectorsize, 0, alloc_hint,
4836 ret = insert_reserved_file_extent(trans, inode,
4837 cur_offset, ins.objectid,
4838 ins.offset, ins.offset,
4839 ins.offset, 0, 0, 0,
4840 BTRFS_FILE_EXTENT_PREALLOC);
4842 num_bytes -= ins.offset;
4843 cur_offset += ins.offset;
4844 alloc_hint = ins.objectid + ins.offset;
4847 if (cur_offset > start) {
4848 inode->i_ctime = CURRENT_TIME;
4849 btrfs_set_flag(inode, PREALLOC);
4850 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4851 cur_offset > i_size_read(inode))
4852 btrfs_i_size_write(inode, cur_offset);
4853 ret = btrfs_update_inode(trans, root, inode);
4857 btrfs_end_transaction(trans, root);
4861 static long btrfs_fallocate(struct inode *inode, int mode,
4862 loff_t offset, loff_t len)
4869 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4870 struct extent_map *em;
4873 alloc_start = offset & ~mask;
4874 alloc_end = (offset + len + mask) & ~mask;
4876 mutex_lock(&inode->i_mutex);
4877 if (alloc_start > inode->i_size) {
4878 ret = btrfs_cont_expand(inode, alloc_start);
4884 struct btrfs_ordered_extent *ordered;
4885 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4886 alloc_end - 1, GFP_NOFS);
4887 ordered = btrfs_lookup_first_ordered_extent(inode,
4890 ordered->file_offset + ordered->len > alloc_start &&
4891 ordered->file_offset < alloc_end) {
4892 btrfs_put_ordered_extent(ordered);
4893 unlock_extent(&BTRFS_I(inode)->io_tree,
4894 alloc_start, alloc_end - 1, GFP_NOFS);
4895 btrfs_wait_ordered_range(inode, alloc_start,
4896 alloc_end - alloc_start);
4899 btrfs_put_ordered_extent(ordered);
4904 cur_offset = alloc_start;
4906 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4907 alloc_end - cur_offset, 0);
4908 BUG_ON(IS_ERR(em) || !em);
4909 last_byte = min(extent_map_end(em), alloc_end);
4910 last_byte = (last_byte + mask) & ~mask;
4911 if (em->block_start == EXTENT_MAP_HOLE) {
4912 ret = prealloc_file_range(inode, cur_offset,
4913 last_byte, alloc_hint, mode);
4915 free_extent_map(em);
4919 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4920 alloc_hint = em->block_start;
4921 free_extent_map(em);
4923 cur_offset = last_byte;
4924 if (cur_offset >= alloc_end) {
4929 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4932 mutex_unlock(&inode->i_mutex);
4936 static int btrfs_set_page_dirty(struct page *page)
4938 return __set_page_dirty_nobuffers(page);
4941 static int btrfs_permission(struct inode *inode, int mask)
4943 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4945 return generic_permission(inode, mask, btrfs_check_acl);
4948 static struct inode_operations btrfs_dir_inode_operations = {
4949 .getattr = btrfs_getattr,
4950 .lookup = btrfs_lookup,
4951 .create = btrfs_create,
4952 .unlink = btrfs_unlink,
4954 .mkdir = btrfs_mkdir,
4955 .rmdir = btrfs_rmdir,
4956 .rename = btrfs_rename,
4957 .symlink = btrfs_symlink,
4958 .setattr = btrfs_setattr,
4959 .mknod = btrfs_mknod,
4960 .setxattr = btrfs_setxattr,
4961 .getxattr = btrfs_getxattr,
4962 .listxattr = btrfs_listxattr,
4963 .removexattr = btrfs_removexattr,
4964 .permission = btrfs_permission,
4966 static struct inode_operations btrfs_dir_ro_inode_operations = {
4967 .lookup = btrfs_lookup,
4968 .permission = btrfs_permission,
4970 static struct file_operations btrfs_dir_file_operations = {
4971 .llseek = generic_file_llseek,
4972 .read = generic_read_dir,
4973 .readdir = btrfs_real_readdir,
4974 .unlocked_ioctl = btrfs_ioctl,
4975 #ifdef CONFIG_COMPAT
4976 .compat_ioctl = btrfs_ioctl,
4978 .release = btrfs_release_file,
4979 .fsync = btrfs_sync_file,
4982 static struct extent_io_ops btrfs_extent_io_ops = {
4983 .fill_delalloc = run_delalloc_range,
4984 .submit_bio_hook = btrfs_submit_bio_hook,
4985 .merge_bio_hook = btrfs_merge_bio_hook,
4986 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4987 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4988 .writepage_start_hook = btrfs_writepage_start_hook,
4989 .readpage_io_failed_hook = btrfs_io_failed_hook,
4990 .set_bit_hook = btrfs_set_bit_hook,
4991 .clear_bit_hook = btrfs_clear_bit_hook,
4994 static struct address_space_operations btrfs_aops = {
4995 .readpage = btrfs_readpage,
4996 .writepage = btrfs_writepage,
4997 .writepages = btrfs_writepages,
4998 .readpages = btrfs_readpages,
4999 .sync_page = block_sync_page,
5001 .direct_IO = btrfs_direct_IO,
5002 .invalidatepage = btrfs_invalidatepage,
5003 .releasepage = btrfs_releasepage,
5004 .set_page_dirty = btrfs_set_page_dirty,
5007 static struct address_space_operations btrfs_symlink_aops = {
5008 .readpage = btrfs_readpage,
5009 .writepage = btrfs_writepage,
5010 .invalidatepage = btrfs_invalidatepage,
5011 .releasepage = btrfs_releasepage,
5014 static struct inode_operations btrfs_file_inode_operations = {
5015 .truncate = btrfs_truncate,
5016 .getattr = btrfs_getattr,
5017 .setattr = btrfs_setattr,
5018 .setxattr = btrfs_setxattr,
5019 .getxattr = btrfs_getxattr,
5020 .listxattr = btrfs_listxattr,
5021 .removexattr = btrfs_removexattr,
5022 .permission = btrfs_permission,
5023 .fallocate = btrfs_fallocate,
5025 static struct inode_operations btrfs_special_inode_operations = {
5026 .getattr = btrfs_getattr,
5027 .setattr = btrfs_setattr,
5028 .permission = btrfs_permission,
5029 .setxattr = btrfs_setxattr,
5030 .getxattr = btrfs_getxattr,
5031 .listxattr = btrfs_listxattr,
5032 .removexattr = btrfs_removexattr,
5034 static struct inode_operations btrfs_symlink_inode_operations = {
5035 .readlink = generic_readlink,
5036 .follow_link = page_follow_link_light,
5037 .put_link = page_put_link,
5038 .permission = btrfs_permission,