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,
107 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
108 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
109 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
117 if (used + root->fs_info->delalloc_bytes + num_required > thresh)
119 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
124 * this does all the hard work for inserting an inline extent into
125 * the btree. The caller should have done a btrfs_drop_extents so that
126 * no overlapping inline items exist in the btree
128 static int noinline insert_inline_extent(struct btrfs_trans_handle *trans,
129 struct btrfs_root *root, struct inode *inode,
130 u64 start, size_t size, size_t compressed_size,
131 struct page **compressed_pages)
133 struct btrfs_key key;
134 struct btrfs_path *path;
135 struct extent_buffer *leaf;
136 struct page *page = NULL;
139 struct btrfs_file_extent_item *ei;
142 size_t cur_size = size;
144 unsigned long offset;
145 int use_compress = 0;
147 if (compressed_size && compressed_pages) {
149 cur_size = compressed_size;
152 path = btrfs_alloc_path(); if (!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 inode_add_bytes(inode, size);
161 datasize = btrfs_file_extent_calc_inline_size(cur_size);
163 inode_add_bytes(inode, size);
164 ret = btrfs_insert_empty_item(trans, root, path, &key,
169 printk("got bad ret %d\n", ret);
172 leaf = path->nodes[0];
173 ei = btrfs_item_ptr(leaf, path->slots[0],
174 struct btrfs_file_extent_item);
175 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
176 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
177 btrfs_set_file_extent_encryption(leaf, ei, 0);
178 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
179 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
180 ptr = btrfs_file_extent_inline_start(ei);
185 while(compressed_size > 0) {
186 cpage = compressed_pages[i];
187 cur_size = min_t(unsigned long, compressed_size,
191 write_extent_buffer(leaf, kaddr, ptr, cur_size);
196 compressed_size -= cur_size;
198 btrfs_set_file_extent_compression(leaf, ei,
199 BTRFS_COMPRESS_ZLIB);
201 page = find_get_page(inode->i_mapping,
202 start >> PAGE_CACHE_SHIFT);
203 btrfs_set_file_extent_compression(leaf, ei, 0);
204 kaddr = kmap_atomic(page, KM_USER0);
205 offset = start & (PAGE_CACHE_SIZE - 1);
206 write_extent_buffer(leaf, kaddr + offset, ptr, size);
207 kunmap_atomic(kaddr, KM_USER0);
208 page_cache_release(page);
210 btrfs_mark_buffer_dirty(leaf);
211 btrfs_free_path(path);
213 BTRFS_I(inode)->disk_i_size = inode->i_size;
214 btrfs_update_inode(trans, root, inode);
217 btrfs_free_path(path);
223 * conditionally insert an inline extent into the file. This
224 * does the checks required to make sure the data is small enough
225 * to fit as an inline extent.
227 static int cow_file_range_inline(struct btrfs_trans_handle *trans,
228 struct btrfs_root *root,
229 struct inode *inode, u64 start, u64 end,
230 size_t compressed_size,
231 struct page **compressed_pages)
233 u64 isize = i_size_read(inode);
234 u64 actual_end = min(end + 1, isize);
235 u64 inline_len = actual_end - start;
236 u64 aligned_end = (end + root->sectorsize - 1) &
237 ~((u64)root->sectorsize - 1);
239 u64 data_len = inline_len;
243 data_len = compressed_size;
246 actual_end >= PAGE_CACHE_SIZE ||
247 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
249 (actual_end & (root->sectorsize - 1)) == 0) ||
251 data_len > root->fs_info->max_inline) {
255 ret = btrfs_drop_extents(trans, root, inode, start,
256 aligned_end, start, &hint_byte);
259 if (isize > actual_end)
260 inline_len = min_t(u64, isize, actual_end);
261 ret = insert_inline_extent(trans, root, inode, start,
262 inline_len, compressed_size,
265 btrfs_drop_extent_cache(inode, start, aligned_end, 0);
269 struct async_extent {
274 unsigned long nr_pages;
275 struct list_head list;
280 struct btrfs_root *root;
281 struct page *locked_page;
284 struct list_head extents;
285 struct btrfs_work work;
288 static noinline int add_async_extent(struct async_cow *cow,
289 u64 start, u64 ram_size,
292 unsigned long nr_pages)
294 struct async_extent *async_extent;
296 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
297 async_extent->start = start;
298 async_extent->ram_size = ram_size;
299 async_extent->compressed_size = compressed_size;
300 async_extent->pages = pages;
301 async_extent->nr_pages = nr_pages;
302 list_add_tail(&async_extent->list, &cow->extents);
307 * we create compressed extents in two phases. The first
308 * phase compresses a range of pages that have already been
309 * locked (both pages and state bits are locked).
311 * This is done inside an ordered work queue, and the compression
312 * is spread across many cpus. The actual IO submission is step
313 * two, and the ordered work queue takes care of making sure that
314 * happens in the same order things were put onto the queue by
315 * writepages and friends.
317 * If this code finds it can't get good compression, it puts an
318 * entry onto the work queue to write the uncompressed bytes. This
319 * makes sure that both compressed inodes and uncompressed inodes
320 * are written in the same order that pdflush sent them down.
322 static noinline int compress_file_range(struct inode *inode,
323 struct page *locked_page,
325 struct async_cow *async_cow,
328 struct btrfs_root *root = BTRFS_I(inode)->root;
329 struct btrfs_trans_handle *trans;
333 u64 blocksize = root->sectorsize;
336 struct page **pages = NULL;
337 unsigned long nr_pages;
338 unsigned long nr_pages_ret = 0;
339 unsigned long total_compressed = 0;
340 unsigned long total_in = 0;
341 unsigned long max_compressed = 128 * 1024;
342 unsigned long max_uncompressed = 128 * 1024;
350 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
351 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
353 actual_end = min_t(u64, i_size_read(inode), end + 1);
354 total_compressed = actual_end - start;
356 /* we want to make sure that amount of ram required to uncompress
357 * an extent is reasonable, so we limit the total size in ram
358 * of a compressed extent to 128k. This is a crucial number
359 * because it also controls how easily we can spread reads across
360 * cpus for decompression.
362 * We also want to make sure the amount of IO required to do
363 * a random read is reasonably small, so we limit the size of
364 * a compressed extent to 128k.
366 total_compressed = min(total_compressed, max_uncompressed);
367 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
368 num_bytes = max(blocksize, num_bytes);
369 disk_num_bytes = num_bytes;
374 * we do compression for mount -o compress and when the
375 * inode has not been flagged as nocompress. This flag can
376 * change at any time if we discover bad compression ratios.
378 if (!btrfs_test_flag(inode, NOCOMPRESS) &&
379 btrfs_test_opt(root, COMPRESS)) {
381 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
383 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
384 total_compressed, pages,
385 nr_pages, &nr_pages_ret,
391 unsigned long offset = total_compressed &
392 (PAGE_CACHE_SIZE - 1);
393 struct page *page = pages[nr_pages_ret - 1];
396 /* zero the tail end of the last page, we might be
397 * sending it down to disk
400 kaddr = kmap_atomic(page, KM_USER0);
401 memset(kaddr + offset, 0,
402 PAGE_CACHE_SIZE - offset);
403 kunmap_atomic(kaddr, KM_USER0);
409 trans = btrfs_join_transaction(root, 1);
411 btrfs_set_trans_block_group(trans, inode);
413 /* lets try to make an inline extent */
414 if (ret || total_in < (actual_end - start)) {
415 /* we didn't compress the entire range, try
416 * to make an uncompressed inline extent.
418 ret = cow_file_range_inline(trans, root, inode,
419 start, end, 0, NULL);
421 /* try making a compressed inline extent */
422 ret = cow_file_range_inline(trans, root, inode,
424 total_compressed, pages);
426 btrfs_end_transaction(trans, root);
429 * inline extent creation worked, we don't need
430 * to create any more async work items. Unlock
431 * and free up our temp pages.
433 extent_clear_unlock_delalloc(inode,
434 &BTRFS_I(inode)->io_tree,
435 start, end, NULL, 1, 0,
444 * we aren't doing an inline extent round the compressed size
445 * up to a block size boundary so the allocator does sane
448 total_compressed = (total_compressed + blocksize - 1) &
452 * one last check to make sure the compression is really a
453 * win, compare the page count read with the blocks on disk
455 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
456 ~(PAGE_CACHE_SIZE - 1);
457 if (total_compressed >= total_in) {
460 disk_num_bytes = total_compressed;
461 num_bytes = total_in;
464 if (!will_compress && pages) {
466 * the compression code ran but failed to make things smaller,
467 * free any pages it allocated and our page pointer array
469 for (i = 0; i < nr_pages_ret; i++) {
470 WARN_ON(pages[i]->mapping);
471 page_cache_release(pages[i]);
475 total_compressed = 0;
478 /* flag the file so we don't compress in the future */
479 btrfs_set_flag(inode, NOCOMPRESS);
484 /* the async work queues will take care of doing actual
485 * allocation on disk for these compressed pages,
486 * and will submit them to the elevator.
488 add_async_extent(async_cow, start, num_bytes,
489 total_compressed, pages, nr_pages_ret);
491 if (start + num_bytes < end) {
499 * No compression, but we still need to write the pages in
500 * the file we've been given so far. redirty the locked
501 * page if it corresponds to our extent and set things up
502 * for the async work queue to run cow_file_range to do
503 * the normal delalloc dance
505 if (page_offset(locked_page) >= start &&
506 page_offset(locked_page) <= end) {
507 __set_page_dirty_nobuffers(locked_page);
508 /* unlocked later on in the async handlers */
510 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
518 for (i = 0; i < nr_pages_ret; i++) {
519 WARN_ON(pages[i]->mapping);
520 page_cache_release(pages[i]);
529 * phase two of compressed writeback. This is the ordered portion
530 * of the code, which only gets called in the order the work was
531 * queued. We walk all the async extents created by compress_file_range
532 * and send them down to the disk.
534 static noinline int submit_compressed_extents(struct inode *inode,
535 struct async_cow *async_cow)
537 struct async_extent *async_extent;
539 struct btrfs_trans_handle *trans;
540 struct btrfs_key ins;
541 struct extent_map *em;
542 struct btrfs_root *root = BTRFS_I(inode)->root;
543 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
544 struct extent_io_tree *io_tree;
547 if (list_empty(&async_cow->extents))
550 trans = btrfs_join_transaction(root, 1);
552 while(!list_empty(&async_cow->extents)) {
553 async_extent = list_entry(async_cow->extents.next,
554 struct async_extent, list);
555 list_del(&async_extent->list);
557 io_tree = &BTRFS_I(inode)->io_tree;
559 /* did the compression code fall back to uncompressed IO? */
560 if (!async_extent->pages) {
561 int page_started = 0;
562 unsigned long nr_written = 0;
564 lock_extent(io_tree, async_extent->start,
565 async_extent->start + async_extent->ram_size - 1,
568 /* allocate blocks */
569 cow_file_range(inode, async_cow->locked_page,
571 async_extent->start +
572 async_extent->ram_size - 1,
573 &page_started, &nr_written, 0);
576 * if page_started, cow_file_range inserted an
577 * inline extent and took care of all the unlocking
578 * and IO for us. Otherwise, we need to submit
579 * all those pages down to the drive.
582 extent_write_locked_range(io_tree,
583 inode, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1,
593 lock_extent(io_tree, async_extent->start,
594 async_extent->start + async_extent->ram_size - 1,
597 * here we're doing allocation and writeback of the
600 btrfs_drop_extent_cache(inode, async_extent->start,
601 async_extent->start +
602 async_extent->ram_size - 1, 0);
604 ret = btrfs_reserve_extent(trans, root,
605 async_extent->compressed_size,
606 async_extent->compressed_size,
610 em = alloc_extent_map(GFP_NOFS);
611 em->start = async_extent->start;
612 em->len = async_extent->ram_size;
613 em->orig_start = em->start;
615 em->block_start = ins.objectid;
616 em->block_len = ins.offset;
617 em->bdev = root->fs_info->fs_devices->latest_bdev;
618 set_bit(EXTENT_FLAG_PINNED, &em->flags);
619 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
622 spin_lock(&em_tree->lock);
623 ret = add_extent_mapping(em_tree, em);
624 spin_unlock(&em_tree->lock);
625 if (ret != -EEXIST) {
629 btrfs_drop_extent_cache(inode, async_extent->start,
630 async_extent->start +
631 async_extent->ram_size - 1, 0);
634 ret = btrfs_add_ordered_extent(inode, async_extent->start,
636 async_extent->ram_size,
638 BTRFS_ORDERED_COMPRESSED);
641 btrfs_end_transaction(trans, root);
644 * clear dirty, set writeback and unlock the pages.
646 extent_clear_unlock_delalloc(inode,
647 &BTRFS_I(inode)->io_tree,
649 async_extent->start +
650 async_extent->ram_size - 1,
651 NULL, 1, 1, 0, 1, 1, 0);
653 ret = btrfs_submit_compressed_write(inode,
655 async_extent->ram_size,
657 ins.offset, async_extent->pages,
658 async_extent->nr_pages);
661 trans = btrfs_join_transaction(root, 1);
662 alloc_hint = ins.objectid + ins.offset;
667 btrfs_end_transaction(trans, root);
672 * when extent_io.c finds a delayed allocation range in the file,
673 * the call backs end up in this code. The basic idea is to
674 * allocate extents on disk for the range, and create ordered data structs
675 * in ram to track those extents.
677 * locked_page is the page that writepage had locked already. We use
678 * it to make sure we don't do extra locks or unlocks.
680 * *page_started is set to one if we unlock locked_page and do everything
681 * required to start IO on it. It may be clean and already done with
684 static noinline int cow_file_range(struct inode *inode,
685 struct page *locked_page,
686 u64 start, u64 end, int *page_started,
687 unsigned long *nr_written,
690 struct btrfs_root *root = BTRFS_I(inode)->root;
691 struct btrfs_trans_handle *trans;
694 unsigned long ram_size;
697 u64 blocksize = root->sectorsize;
699 struct btrfs_key ins;
700 struct extent_map *em;
701 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
704 trans = btrfs_join_transaction(root, 1);
706 btrfs_set_trans_block_group(trans, inode);
708 actual_end = min_t(u64, i_size_read(inode), end + 1);
710 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
711 num_bytes = max(blocksize, num_bytes);
712 disk_num_bytes = num_bytes;
716 /* lets try to make an inline extent */
717 ret = cow_file_range_inline(trans, root, inode,
718 start, end, 0, NULL);
720 extent_clear_unlock_delalloc(inode,
721 &BTRFS_I(inode)->io_tree,
722 start, end, NULL, 1, 1,
724 *nr_written = *nr_written +
725 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
732 BUG_ON(disk_num_bytes >
733 btrfs_super_total_bytes(&root->fs_info->super_copy));
735 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
737 while(disk_num_bytes > 0) {
738 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
739 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
740 root->sectorsize, 0, alloc_hint,
745 em = alloc_extent_map(GFP_NOFS);
747 em->orig_start = em->start;
749 ram_size = ins.offset;
750 em->len = ins.offset;
752 em->block_start = ins.objectid;
753 em->block_len = ins.offset;
754 em->bdev = root->fs_info->fs_devices->latest_bdev;
755 set_bit(EXTENT_FLAG_PINNED, &em->flags);
758 spin_lock(&em_tree->lock);
759 ret = add_extent_mapping(em_tree, em);
760 spin_unlock(&em_tree->lock);
761 if (ret != -EEXIST) {
765 btrfs_drop_extent_cache(inode, start,
766 start + ram_size - 1, 0);
769 cur_alloc_size = ins.offset;
770 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
771 ram_size, cur_alloc_size, 0);
774 if (disk_num_bytes < cur_alloc_size) {
775 printk("num_bytes %Lu cur_alloc %Lu\n", disk_num_bytes,
779 /* we're not doing compressed IO, don't unlock the first
780 * page (which the caller expects to stay locked), don't
781 * clear any dirty bits and don't set any writeback bits
783 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
784 start, start + ram_size - 1,
785 locked_page, unlock, 1,
787 disk_num_bytes -= cur_alloc_size;
788 num_bytes -= cur_alloc_size;
789 alloc_hint = ins.objectid + ins.offset;
790 start += cur_alloc_size;
794 btrfs_end_transaction(trans, root);
800 * work queue call back to started compression on a file and pages
802 static noinline void async_cow_start(struct btrfs_work *work)
804 struct async_cow *async_cow;
806 async_cow = container_of(work, struct async_cow, work);
808 compress_file_range(async_cow->inode, async_cow->locked_page,
809 async_cow->start, async_cow->end, async_cow,
812 async_cow->inode = NULL;
816 * work queue call back to submit previously compressed pages
818 static noinline void async_cow_submit(struct btrfs_work *work)
820 struct async_cow *async_cow;
821 struct btrfs_root *root;
822 unsigned long nr_pages;
824 async_cow = container_of(work, struct async_cow, work);
826 root = async_cow->root;
827 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
830 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
832 if (atomic_read(&root->fs_info->async_delalloc_pages) <
834 waitqueue_active(&root->fs_info->async_submit_wait))
835 wake_up(&root->fs_info->async_submit_wait);
837 if (async_cow->inode) {
838 submit_compressed_extents(async_cow->inode, async_cow);
842 static noinline void async_cow_free(struct btrfs_work *work)
844 struct async_cow *async_cow;
845 async_cow = container_of(work, struct async_cow, work);
849 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
850 u64 start, u64 end, int *page_started,
851 unsigned long *nr_written)
853 struct async_cow *async_cow;
854 struct btrfs_root *root = BTRFS_I(inode)->root;
855 unsigned long nr_pages;
857 int limit = 10 * 1024 * 1042;
859 if (!btrfs_test_opt(root, COMPRESS)) {
860 return cow_file_range(inode, locked_page, start, end,
861 page_started, nr_written, 1);
864 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED |
865 EXTENT_DELALLOC, 1, 0, GFP_NOFS);
867 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
868 async_cow->inode = inode;
869 async_cow->root = root;
870 async_cow->locked_page = locked_page;
871 async_cow->start = start;
873 if (btrfs_test_flag(inode, NOCOMPRESS))
876 cur_end = min(end, start + 512 * 1024 - 1);
878 async_cow->end = cur_end;
879 INIT_LIST_HEAD(&async_cow->extents);
881 async_cow->work.func = async_cow_start;
882 async_cow->work.ordered_func = async_cow_submit;
883 async_cow->work.ordered_free = async_cow_free;
884 async_cow->work.flags = 0;
886 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
888 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
890 btrfs_queue_worker(&root->fs_info->delalloc_workers,
893 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
894 wait_event(root->fs_info->async_submit_wait,
895 (atomic_read(&root->fs_info->async_delalloc_pages) <
899 while(atomic_read(&root->fs_info->async_submit_draining) &&
900 atomic_read(&root->fs_info->async_delalloc_pages)) {
901 wait_event(root->fs_info->async_submit_wait,
902 (atomic_read(&root->fs_info->async_delalloc_pages) ==
906 *nr_written += nr_pages;
914 * when nowcow writeback call back. This checks for snapshots or COW copies
915 * of the extents that exist in the file, and COWs the file as required.
917 * If no cow copies or snapshots exist, we write directly to the existing
920 static int run_delalloc_nocow(struct inode *inode, struct page *locked_page,
921 u64 start, u64 end, int *page_started, int force,
922 unsigned long *nr_written)
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 struct btrfs_trans_handle *trans;
926 struct extent_buffer *leaf;
927 struct btrfs_path *path;
928 struct btrfs_file_extent_item *fi;
929 struct btrfs_key found_key;
941 path = btrfs_alloc_path();
943 trans = btrfs_join_transaction(root, 1);
949 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
952 if (ret > 0 && path->slots[0] > 0 && check_prev) {
953 leaf = path->nodes[0];
954 btrfs_item_key_to_cpu(leaf, &found_key,
956 if (found_key.objectid == inode->i_ino &&
957 found_key.type == BTRFS_EXTENT_DATA_KEY)
962 leaf = path->nodes[0];
963 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
964 ret = btrfs_next_leaf(root, path);
969 leaf = path->nodes[0];
974 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
976 if (found_key.objectid > inode->i_ino ||
977 found_key.type > BTRFS_EXTENT_DATA_KEY ||
978 found_key.offset > end)
981 if (found_key.offset > cur_offset) {
982 extent_end = found_key.offset;
986 fi = btrfs_item_ptr(leaf, path->slots[0],
987 struct btrfs_file_extent_item);
988 extent_type = btrfs_file_extent_type(leaf, fi);
990 if (extent_type == BTRFS_FILE_EXTENT_REG ||
991 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
992 struct btrfs_block_group_cache *block_group;
993 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
994 extent_end = found_key.offset +
995 btrfs_file_extent_num_bytes(leaf, fi);
996 if (extent_end <= start) {
1000 if (btrfs_file_extent_compression(leaf, fi) ||
1001 btrfs_file_extent_encryption(leaf, fi) ||
1002 btrfs_file_extent_other_encoding(leaf, fi))
1004 if (disk_bytenr == 0)
1006 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1008 if (btrfs_cross_ref_exist(trans, root, disk_bytenr))
1010 block_group = btrfs_lookup_block_group(root->fs_info,
1012 if (!block_group || block_group->ro)
1014 disk_bytenr += btrfs_file_extent_offset(leaf, fi);
1016 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1017 extent_end = found_key.offset +
1018 btrfs_file_extent_inline_len(leaf, fi);
1019 extent_end = ALIGN(extent_end, root->sectorsize);
1024 if (extent_end <= start) {
1029 if (cow_start == (u64)-1)
1030 cow_start = cur_offset;
1031 cur_offset = extent_end;
1032 if (cur_offset > end)
1038 btrfs_release_path(root, path);
1039 if (cow_start != (u64)-1) {
1040 ret = cow_file_range(inode, locked_page, cow_start,
1041 found_key.offset - 1, page_started,
1044 cow_start = (u64)-1;
1047 disk_bytenr += cur_offset - found_key.offset;
1048 num_bytes = min(end + 1, extent_end) - cur_offset;
1049 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1050 struct extent_map *em;
1051 struct extent_map_tree *em_tree;
1052 em_tree = &BTRFS_I(inode)->extent_tree;
1053 em = alloc_extent_map(GFP_NOFS);
1054 em->start = cur_offset;
1055 em->orig_start = em->start;
1056 em->len = num_bytes;
1057 em->block_len = num_bytes;
1058 em->block_start = disk_bytenr;
1059 em->bdev = root->fs_info->fs_devices->latest_bdev;
1060 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1062 spin_lock(&em_tree->lock);
1063 ret = add_extent_mapping(em_tree, em);
1064 spin_unlock(&em_tree->lock);
1065 if (ret != -EEXIST) {
1066 free_extent_map(em);
1069 btrfs_drop_extent_cache(inode, em->start,
1070 em->start + em->len - 1, 0);
1072 type = BTRFS_ORDERED_PREALLOC;
1074 type = BTRFS_ORDERED_NOCOW;
1077 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1078 num_bytes, num_bytes, type);
1081 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1082 cur_offset, cur_offset + num_bytes - 1,
1083 locked_page, 1, 1, 1, 0, 0, 0);
1084 cur_offset = extent_end;
1085 if (cur_offset > end)
1088 btrfs_release_path(root, path);
1090 if (cur_offset <= end && cow_start == (u64)-1)
1091 cow_start = cur_offset;
1092 if (cow_start != (u64)-1) {
1093 ret = cow_file_range(inode, locked_page, cow_start, end,
1094 page_started, nr_written, 1);
1098 ret = btrfs_end_transaction(trans, root);
1100 btrfs_free_path(path);
1105 * extent_io.c call back to do delayed allocation processing
1107 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1108 u64 start, u64 end, int *page_started,
1109 unsigned long *nr_written)
1111 struct btrfs_root *root = BTRFS_I(inode)->root;
1114 if (btrfs_test_opt(root, NODATACOW) ||
1115 btrfs_test_flag(inode, NODATACOW))
1116 ret = run_delalloc_nocow(inode, locked_page, start, end,
1117 page_started, 1, nr_written);
1118 else if (btrfs_test_flag(inode, PREALLOC))
1119 ret = run_delalloc_nocow(inode, locked_page, start, end,
1120 page_started, 0, nr_written);
1122 ret = cow_file_range_async(inode, locked_page, start, end,
1123 page_started, nr_written);
1129 * extent_io.c set_bit_hook, used to track delayed allocation
1130 * bytes in this file, and to maintain the list of inodes that
1131 * have pending delalloc work to be done.
1133 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1134 unsigned long old, unsigned long bits)
1136 unsigned long flags;
1137 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1140 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1141 root->fs_info->delalloc_bytes += end - start + 1;
1142 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1143 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1144 &root->fs_info->delalloc_inodes);
1146 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1152 * extent_io.c clear_bit_hook, see set_bit_hook for why
1154 static int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
1155 unsigned long old, unsigned long bits)
1157 if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1158 struct btrfs_root *root = BTRFS_I(inode)->root;
1159 unsigned long flags;
1161 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
1162 if (end - start + 1 > root->fs_info->delalloc_bytes) {
1163 printk("warning: delalloc account %Lu %Lu\n",
1164 end - start + 1, root->fs_info->delalloc_bytes);
1165 root->fs_info->delalloc_bytes = 0;
1166 BTRFS_I(inode)->delalloc_bytes = 0;
1168 root->fs_info->delalloc_bytes -= end - start + 1;
1169 BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
1171 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1172 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1173 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1175 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
1181 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1182 * we don't create bios that span stripes or chunks
1184 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1185 size_t size, struct bio *bio,
1186 unsigned long bio_flags)
1188 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1189 struct btrfs_mapping_tree *map_tree;
1190 u64 logical = (u64)bio->bi_sector << 9;
1195 if (bio_flags & EXTENT_BIO_COMPRESSED)
1198 length = bio->bi_size;
1199 map_tree = &root->fs_info->mapping_tree;
1200 map_length = length;
1201 ret = btrfs_map_block(map_tree, READ, logical,
1202 &map_length, NULL, 0);
1204 if (map_length < length + size) {
1211 * in order to insert checksums into the metadata in large chunks,
1212 * we wait until bio submission time. All the pages in the bio are
1213 * checksummed and sums are attached onto the ordered extent record.
1215 * At IO completion time the cums attached on the ordered extent record
1216 * are inserted into the btree
1218 static int __btrfs_submit_bio_start(struct inode *inode, int rw, struct bio *bio,
1219 int mirror_num, unsigned long bio_flags)
1221 struct btrfs_root *root = BTRFS_I(inode)->root;
1224 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1230 * in order to insert checksums into the metadata in large chunks,
1231 * we wait until bio submission time. All the pages in the bio are
1232 * checksummed and sums are attached onto the ordered extent record.
1234 * At IO completion time the cums attached on the ordered extent record
1235 * are inserted into the btree
1237 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1238 int mirror_num, unsigned long bio_flags)
1240 struct btrfs_root *root = BTRFS_I(inode)->root;
1241 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1245 * extent_io.c submission hook. This does the right thing for csum calculation on write,
1246 * or reading the csums from the tree before a read
1248 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1249 int mirror_num, unsigned long bio_flags)
1251 struct btrfs_root *root = BTRFS_I(inode)->root;
1255 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1258 skip_sum = btrfs_test_opt(root, NODATASUM) ||
1259 btrfs_test_flag(inode, NODATASUM);
1261 if (!(rw & (1 << BIO_RW))) {
1262 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1263 return btrfs_submit_compressed_read(inode, bio,
1264 mirror_num, bio_flags);
1265 } else if (!skip_sum)
1266 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1268 } else if (!skip_sum) {
1269 /* we're doing a write, do the async checksumming */
1270 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1271 inode, rw, bio, mirror_num,
1272 bio_flags, __btrfs_submit_bio_start,
1273 __btrfs_submit_bio_done);
1277 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1281 * given a list of ordered sums record them in the inode. This happens
1282 * at IO completion time based on sums calculated at bio submission time.
1284 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1285 struct inode *inode, u64 file_offset,
1286 struct list_head *list)
1288 struct list_head *cur;
1289 struct btrfs_ordered_sum *sum;
1291 btrfs_set_trans_block_group(trans, inode);
1292 list_for_each(cur, list) {
1293 sum = list_entry(cur, struct btrfs_ordered_sum, list);
1294 btrfs_csum_file_blocks(trans,
1295 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1300 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1302 if ((end & (PAGE_CACHE_SIZE - 1)) == 0) {
1305 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1309 /* see btrfs_writepage_start_hook for details on why this is required */
1310 struct btrfs_writepage_fixup {
1312 struct btrfs_work work;
1315 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1317 struct btrfs_writepage_fixup *fixup;
1318 struct btrfs_ordered_extent *ordered;
1320 struct inode *inode;
1324 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1328 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1329 ClearPageChecked(page);
1333 inode = page->mapping->host;
1334 page_start = page_offset(page);
1335 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1337 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1339 /* already ordered? We're done */
1340 if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
1341 EXTENT_ORDERED, 0)) {
1345 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1347 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1348 page_end, GFP_NOFS);
1350 btrfs_start_ordered_extent(inode, ordered, 1);
1354 btrfs_set_extent_delalloc(inode, page_start, page_end);
1355 ClearPageChecked(page);
1357 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1360 page_cache_release(page);
1364 * There are a few paths in the higher layers of the kernel that directly
1365 * set the page dirty bit without asking the filesystem if it is a
1366 * good idea. This causes problems because we want to make sure COW
1367 * properly happens and the data=ordered rules are followed.
1369 * In our case any range that doesn't have the ORDERED bit set
1370 * hasn't been properly setup for IO. We kick off an async process
1371 * to fix it up. The async helper will wait for ordered extents, set
1372 * the delalloc bit and make it safe to write the page.
1374 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1376 struct inode *inode = page->mapping->host;
1377 struct btrfs_writepage_fixup *fixup;
1378 struct btrfs_root *root = BTRFS_I(inode)->root;
1381 ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1386 if (PageChecked(page))
1389 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1393 SetPageChecked(page);
1394 page_cache_get(page);
1395 fixup->work.func = btrfs_writepage_fixup_worker;
1397 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1401 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1402 struct inode *inode, u64 file_pos,
1403 u64 disk_bytenr, u64 disk_num_bytes,
1404 u64 num_bytes, u64 ram_bytes,
1405 u8 compression, u8 encryption,
1406 u16 other_encoding, int extent_type)
1408 struct btrfs_root *root = BTRFS_I(inode)->root;
1409 struct btrfs_file_extent_item *fi;
1410 struct btrfs_path *path;
1411 struct extent_buffer *leaf;
1412 struct btrfs_key ins;
1416 path = btrfs_alloc_path();
1419 ret = btrfs_drop_extents(trans, root, inode, file_pos,
1420 file_pos + num_bytes, file_pos, &hint);
1423 ins.objectid = inode->i_ino;
1424 ins.offset = file_pos;
1425 ins.type = BTRFS_EXTENT_DATA_KEY;
1426 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1428 leaf = path->nodes[0];
1429 fi = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_file_extent_item);
1431 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1432 btrfs_set_file_extent_type(leaf, fi, extent_type);
1433 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1434 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1435 btrfs_set_file_extent_offset(leaf, fi, 0);
1436 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1437 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1438 btrfs_set_file_extent_compression(leaf, fi, compression);
1439 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1440 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1441 btrfs_mark_buffer_dirty(leaf);
1443 inode_add_bytes(inode, num_bytes);
1444 btrfs_drop_extent_cache(inode, file_pos, file_pos + num_bytes - 1, 0);
1446 ins.objectid = disk_bytenr;
1447 ins.offset = disk_num_bytes;
1448 ins.type = BTRFS_EXTENT_ITEM_KEY;
1449 ret = btrfs_alloc_reserved_extent(trans, root, leaf->start,
1450 root->root_key.objectid,
1451 trans->transid, inode->i_ino, &ins);
1454 btrfs_free_path(path);
1458 /* as ordered data IO finishes, this gets called so we can finish
1459 * an ordered extent if the range of bytes in the file it covers are
1462 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1464 struct btrfs_root *root = BTRFS_I(inode)->root;
1465 struct btrfs_trans_handle *trans;
1466 struct btrfs_ordered_extent *ordered_extent;
1467 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1471 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1475 trans = btrfs_join_transaction(root, 1);
1477 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1478 BUG_ON(!ordered_extent);
1479 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
1482 lock_extent(io_tree, ordered_extent->file_offset,
1483 ordered_extent->file_offset + ordered_extent->len - 1,
1486 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1488 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1490 ret = btrfs_mark_extent_written(trans, root, inode,
1491 ordered_extent->file_offset,
1492 ordered_extent->file_offset +
1493 ordered_extent->len);
1496 ret = insert_reserved_file_extent(trans, inode,
1497 ordered_extent->file_offset,
1498 ordered_extent->start,
1499 ordered_extent->disk_len,
1500 ordered_extent->len,
1501 ordered_extent->len,
1503 BTRFS_FILE_EXTENT_REG);
1506 unlock_extent(io_tree, ordered_extent->file_offset,
1507 ordered_extent->file_offset + ordered_extent->len - 1,
1510 add_pending_csums(trans, inode, ordered_extent->file_offset,
1511 &ordered_extent->list);
1513 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1514 btrfs_ordered_update_i_size(inode, ordered_extent);
1515 btrfs_update_inode(trans, root, inode);
1516 btrfs_remove_ordered_extent(inode, ordered_extent);
1517 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1520 btrfs_put_ordered_extent(ordered_extent);
1521 /* once for the tree */
1522 btrfs_put_ordered_extent(ordered_extent);
1524 btrfs_end_transaction(trans, root);
1528 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1529 struct extent_state *state, int uptodate)
1531 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1535 * When IO fails, either with EIO or csum verification fails, we
1536 * try other mirrors that might have a good copy of the data. This
1537 * io_failure_record is used to record state as we go through all the
1538 * mirrors. If another mirror has good data, the page is set up to date
1539 * and things continue. If a good mirror can't be found, the original
1540 * bio end_io callback is called to indicate things have failed.
1542 struct io_failure_record {
1547 unsigned long bio_flags;
1551 static int btrfs_io_failed_hook(struct bio *failed_bio,
1552 struct page *page, u64 start, u64 end,
1553 struct extent_state *state)
1555 struct io_failure_record *failrec = NULL;
1557 struct extent_map *em;
1558 struct inode *inode = page->mapping->host;
1559 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1560 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1567 ret = get_state_private(failure_tree, start, &private);
1569 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1572 failrec->start = start;
1573 failrec->len = end - start + 1;
1574 failrec->last_mirror = 0;
1575 failrec->bio_flags = 0;
1577 spin_lock(&em_tree->lock);
1578 em = lookup_extent_mapping(em_tree, start, failrec->len);
1579 if (em->start > start || em->start + em->len < start) {
1580 free_extent_map(em);
1583 spin_unlock(&em_tree->lock);
1585 if (!em || IS_ERR(em)) {
1589 logical = start - em->start;
1590 logical = em->block_start + logical;
1591 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1592 logical = em->block_start;
1593 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1595 failrec->logical = logical;
1596 free_extent_map(em);
1597 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1598 EXTENT_DIRTY, GFP_NOFS);
1599 set_state_private(failure_tree, start,
1600 (u64)(unsigned long)failrec);
1602 failrec = (struct io_failure_record *)(unsigned long)private;
1604 num_copies = btrfs_num_copies(
1605 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1606 failrec->logical, failrec->len);
1607 failrec->last_mirror++;
1609 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
1610 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1613 if (state && state->start != failrec->start)
1615 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
1617 if (!state || failrec->last_mirror > num_copies) {
1618 set_state_private(failure_tree, failrec->start, 0);
1619 clear_extent_bits(failure_tree, failrec->start,
1620 failrec->start + failrec->len - 1,
1621 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1625 bio = bio_alloc(GFP_NOFS, 1);
1626 bio->bi_private = state;
1627 bio->bi_end_io = failed_bio->bi_end_io;
1628 bio->bi_sector = failrec->logical >> 9;
1629 bio->bi_bdev = failed_bio->bi_bdev;
1632 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1633 if (failed_bio->bi_rw & (1 << BIO_RW))
1638 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1639 failrec->last_mirror,
1640 failrec->bio_flags);
1645 * each time an IO finishes, we do a fast check in the IO failure tree
1646 * to see if we need to process or clean up an io_failure_record
1648 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1651 u64 private_failure;
1652 struct io_failure_record *failure;
1656 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1657 (u64)-1, 1, EXTENT_DIRTY)) {
1658 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1659 start, &private_failure);
1661 failure = (struct io_failure_record *)(unsigned long)
1663 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1665 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1667 failure->start + failure->len - 1,
1668 EXTENT_DIRTY | EXTENT_LOCKED,
1677 * when reads are done, we need to check csums to verify the data is correct
1678 * if there's a match, we allow the bio to finish. If not, we go through
1679 * the io_failure_record routines to find good copies
1681 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1682 struct extent_state *state)
1684 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1685 struct inode *inode = page->mapping->host;
1686 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1688 u64 private = ~(u32)0;
1690 struct btrfs_root *root = BTRFS_I(inode)->root;
1692 unsigned long flags;
1694 if (PageChecked(page)) {
1695 ClearPageChecked(page);
1698 if (btrfs_test_opt(root, NODATASUM) ||
1699 btrfs_test_flag(inode, NODATASUM))
1702 if (state && state->start == start) {
1703 private = state->private;
1706 ret = get_state_private(io_tree, start, &private);
1708 local_irq_save(flags);
1709 kaddr = kmap_atomic(page, KM_IRQ0);
1713 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1714 btrfs_csum_final(csum, (char *)&csum);
1715 if (csum != private) {
1718 kunmap_atomic(kaddr, KM_IRQ0);
1719 local_irq_restore(flags);
1721 /* if the io failure tree for this inode is non-empty,
1722 * check to see if we've recovered from a failed IO
1724 btrfs_clean_io_failures(inode, start);
1728 printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
1729 page->mapping->host->i_ino, (unsigned long long)start, csum,
1731 memset(kaddr + offset, 1, end - start + 1);
1732 flush_dcache_page(page);
1733 kunmap_atomic(kaddr, KM_IRQ0);
1734 local_irq_restore(flags);
1741 * This creates an orphan entry for the given inode in case something goes
1742 * wrong in the middle of an unlink/truncate.
1744 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1746 struct btrfs_root *root = BTRFS_I(inode)->root;
1749 spin_lock(&root->list_lock);
1751 /* already on the orphan list, we're good */
1752 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
1753 spin_unlock(&root->list_lock);
1757 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1759 spin_unlock(&root->list_lock);
1762 * insert an orphan item to track this unlinked/truncated file
1764 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
1770 * We have done the truncate/delete so we can go ahead and remove the orphan
1771 * item for this particular inode.
1773 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
1775 struct btrfs_root *root = BTRFS_I(inode)->root;
1778 spin_lock(&root->list_lock);
1780 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
1781 spin_unlock(&root->list_lock);
1785 list_del_init(&BTRFS_I(inode)->i_orphan);
1787 spin_unlock(&root->list_lock);
1791 spin_unlock(&root->list_lock);
1793 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
1799 * this cleans up any orphans that may be left on the list from the last use
1802 void btrfs_orphan_cleanup(struct btrfs_root *root)
1804 struct btrfs_path *path;
1805 struct extent_buffer *leaf;
1806 struct btrfs_item *item;
1807 struct btrfs_key key, found_key;
1808 struct btrfs_trans_handle *trans;
1809 struct inode *inode;
1810 int ret = 0, nr_unlink = 0, nr_truncate = 0;
1812 path = btrfs_alloc_path();
1817 key.objectid = BTRFS_ORPHAN_OBJECTID;
1818 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
1819 key.offset = (u64)-1;
1823 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1825 printk(KERN_ERR "Error searching slot for orphan: %d"
1831 * if ret == 0 means we found what we were searching for, which
1832 * is weird, but possible, so only screw with path if we didnt
1833 * find the key and see if we have stuff that matches
1836 if (path->slots[0] == 0)
1841 /* pull out the item */
1842 leaf = path->nodes[0];
1843 item = btrfs_item_nr(leaf, path->slots[0]);
1844 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1846 /* make sure the item matches what we want */
1847 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
1849 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
1852 /* release the path since we're done with it */
1853 btrfs_release_path(root, path);
1856 * this is where we are basically btrfs_lookup, without the
1857 * crossing root thing. we store the inode number in the
1858 * offset of the orphan item.
1860 inode = btrfs_iget_locked(root->fs_info->sb,
1861 found_key.offset, root);
1865 if (inode->i_state & I_NEW) {
1866 BTRFS_I(inode)->root = root;
1868 /* have to set the location manually */
1869 BTRFS_I(inode)->location.objectid = inode->i_ino;
1870 BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
1871 BTRFS_I(inode)->location.offset = 0;
1873 btrfs_read_locked_inode(inode);
1874 unlock_new_inode(inode);
1878 * add this inode to the orphan list so btrfs_orphan_del does
1879 * the proper thing when we hit it
1881 spin_lock(&root->list_lock);
1882 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
1883 spin_unlock(&root->list_lock);
1886 * if this is a bad inode, means we actually succeeded in
1887 * removing the inode, but not the orphan record, which means
1888 * we need to manually delete the orphan since iput will just
1889 * do a destroy_inode
1891 if (is_bad_inode(inode)) {
1892 trans = btrfs_start_transaction(root, 1);
1893 btrfs_orphan_del(trans, inode);
1894 btrfs_end_transaction(trans, root);
1899 /* if we have links, this was a truncate, lets do that */
1900 if (inode->i_nlink) {
1902 btrfs_truncate(inode);
1907 /* this will do delete_inode and everything for us */
1912 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
1914 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
1916 btrfs_free_path(path);
1920 * read an inode from the btree into the in-memory inode
1922 void btrfs_read_locked_inode(struct inode *inode)
1924 struct btrfs_path *path;
1925 struct extent_buffer *leaf;
1926 struct btrfs_inode_item *inode_item;
1927 struct btrfs_timespec *tspec;
1928 struct btrfs_root *root = BTRFS_I(inode)->root;
1929 struct btrfs_key location;
1930 u64 alloc_group_block;
1934 path = btrfs_alloc_path();
1936 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1938 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1942 leaf = path->nodes[0];
1943 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1944 struct btrfs_inode_item);
1946 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1947 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1948 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1949 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1950 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1952 tspec = btrfs_inode_atime(inode_item);
1953 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1954 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1956 tspec = btrfs_inode_mtime(inode_item);
1957 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1958 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1960 tspec = btrfs_inode_ctime(inode_item);
1961 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1962 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1964 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
1965 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1966 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
1967 inode->i_generation = BTRFS_I(inode)->generation;
1969 rdev = btrfs_inode_rdev(leaf, inode_item);
1971 BTRFS_I(inode)->index_cnt = (u64)-1;
1973 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1974 BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1976 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1977 if (!BTRFS_I(inode)->block_group) {
1978 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1980 BTRFS_BLOCK_GROUP_METADATA, 0);
1982 btrfs_free_path(path);
1985 switch (inode->i_mode & S_IFMT) {
1987 inode->i_mapping->a_ops = &btrfs_aops;
1988 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1989 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1990 inode->i_fop = &btrfs_file_operations;
1991 inode->i_op = &btrfs_file_inode_operations;
1994 inode->i_fop = &btrfs_dir_file_operations;
1995 if (root == root->fs_info->tree_root)
1996 inode->i_op = &btrfs_dir_ro_inode_operations;
1998 inode->i_op = &btrfs_dir_inode_operations;
2001 inode->i_op = &btrfs_symlink_inode_operations;
2002 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2003 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2006 init_special_inode(inode, inode->i_mode, rdev);
2012 btrfs_free_path(path);
2013 make_bad_inode(inode);
2017 * given a leaf and an inode, copy the inode fields into the leaf
2019 static void fill_inode_item(struct btrfs_trans_handle *trans,
2020 struct extent_buffer *leaf,
2021 struct btrfs_inode_item *item,
2022 struct inode *inode)
2024 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2025 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2026 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2027 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2028 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2030 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2031 inode->i_atime.tv_sec);
2032 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2033 inode->i_atime.tv_nsec);
2035 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2036 inode->i_mtime.tv_sec);
2037 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2038 inode->i_mtime.tv_nsec);
2040 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2041 inode->i_ctime.tv_sec);
2042 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2043 inode->i_ctime.tv_nsec);
2045 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2046 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2047 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2048 btrfs_set_inode_transid(leaf, item, trans->transid);
2049 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2050 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2051 btrfs_set_inode_block_group(leaf, item,
2052 BTRFS_I(inode)->block_group->key.objectid);
2056 * copy everything in the in-memory inode into the btree.
2058 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
2059 struct btrfs_root *root,
2060 struct inode *inode)
2062 struct btrfs_inode_item *inode_item;
2063 struct btrfs_path *path;
2064 struct extent_buffer *leaf;
2067 path = btrfs_alloc_path();
2069 ret = btrfs_lookup_inode(trans, root, path,
2070 &BTRFS_I(inode)->location, 1);
2077 leaf = path->nodes[0];
2078 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2079 struct btrfs_inode_item);
2081 fill_inode_item(trans, leaf, inode_item, inode);
2082 btrfs_mark_buffer_dirty(leaf);
2083 btrfs_set_inode_last_trans(trans, inode);
2086 btrfs_free_path(path);
2092 * unlink helper that gets used here in inode.c and in the tree logging
2093 * recovery code. It remove a link in a directory with a given name, and
2094 * also drops the back refs in the inode to the directory
2096 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2097 struct btrfs_root *root,
2098 struct inode *dir, struct inode *inode,
2099 const char *name, int name_len)
2101 struct btrfs_path *path;
2103 struct extent_buffer *leaf;
2104 struct btrfs_dir_item *di;
2105 struct btrfs_key key;
2108 path = btrfs_alloc_path();
2114 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2115 name, name_len, -1);
2124 leaf = path->nodes[0];
2125 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2126 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2129 btrfs_release_path(root, path);
2131 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2133 dir->i_ino, &index);
2135 printk("failed to delete reference to %.*s, "
2136 "inode %lu parent %lu\n", name_len, name,
2137 inode->i_ino, dir->i_ino);
2141 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2142 index, name, name_len, -1);
2151 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2152 btrfs_release_path(root, path);
2154 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2156 BUG_ON(ret != 0 && ret != -ENOENT);
2158 BTRFS_I(dir)->log_dirty_trans = trans->transid;
2160 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2164 btrfs_free_path(path);
2168 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2169 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2170 btrfs_update_inode(trans, root, dir);
2171 btrfs_drop_nlink(inode);
2172 ret = btrfs_update_inode(trans, root, inode);
2173 dir->i_sb->s_dirt = 1;
2178 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2180 struct btrfs_root *root;
2181 struct btrfs_trans_handle *trans;
2182 struct inode *inode = dentry->d_inode;
2184 unsigned long nr = 0;
2186 root = BTRFS_I(dir)->root;
2188 ret = btrfs_check_free_space(root, 1, 1);
2192 trans = btrfs_start_transaction(root, 1);
2194 btrfs_set_trans_block_group(trans, dir);
2195 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2196 dentry->d_name.name, dentry->d_name.len);
2198 if (inode->i_nlink == 0)
2199 ret = btrfs_orphan_add(trans, inode);
2201 nr = trans->blocks_used;
2203 btrfs_end_transaction_throttle(trans, root);
2205 btrfs_btree_balance_dirty(root, nr);
2209 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2211 struct inode *inode = dentry->d_inode;
2214 struct btrfs_root *root = BTRFS_I(dir)->root;
2215 struct btrfs_trans_handle *trans;
2216 unsigned long nr = 0;
2219 * the FIRST_FREE_OBJECTID check makes sure we don't try to rmdir
2220 * the root of a subvolume or snapshot
2222 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2223 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID) {
2227 ret = btrfs_check_free_space(root, 1, 1);
2231 trans = btrfs_start_transaction(root, 1);
2232 btrfs_set_trans_block_group(trans, dir);
2234 err = btrfs_orphan_add(trans, inode);
2238 /* now the directory is empty */
2239 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2240 dentry->d_name.name, dentry->d_name.len);
2242 btrfs_i_size_write(inode, 0);
2246 nr = trans->blocks_used;
2247 ret = btrfs_end_transaction_throttle(trans, root);
2249 btrfs_btree_balance_dirty(root, nr);
2258 * when truncating bytes in a file, it is possible to avoid reading
2259 * the leaves that contain only checksum items. This can be the
2260 * majority of the IO required to delete a large file, but it must
2261 * be done carefully.
2263 * The keys in the level just above the leaves are checked to make sure
2264 * the lowest key in a given leaf is a csum key, and starts at an offset
2265 * after the new size.
2267 * Then the key for the next leaf is checked to make sure it also has
2268 * a checksum item for the same file. If it does, we know our target leaf
2269 * contains only checksum items, and it can be safely freed without reading
2272 * This is just an optimization targeted at large files. It may do
2273 * nothing. It will return 0 unless things went badly.
2275 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2276 struct btrfs_root *root,
2277 struct btrfs_path *path,
2278 struct inode *inode, u64 new_size)
2280 struct btrfs_key key;
2283 struct btrfs_key found_key;
2284 struct btrfs_key other_key;
2285 struct btrfs_leaf_ref *ref;
2289 path->lowest_level = 1;
2290 key.objectid = inode->i_ino;
2291 key.type = BTRFS_CSUM_ITEM_KEY;
2292 key.offset = new_size;
2294 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2298 if (path->nodes[1] == NULL) {
2303 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2304 nritems = btrfs_header_nritems(path->nodes[1]);
2309 if (path->slots[1] >= nritems)
2312 /* did we find a key greater than anything we want to delete? */
2313 if (found_key.objectid > inode->i_ino ||
2314 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2317 /* we check the next key in the node to make sure the leave contains
2318 * only checksum items. This comparison doesn't work if our
2319 * leaf is the last one in the node
2321 if (path->slots[1] + 1 >= nritems) {
2323 /* search forward from the last key in the node, this
2324 * will bring us into the next node in the tree
2326 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2328 /* unlikely, but we inc below, so check to be safe */
2329 if (found_key.offset == (u64)-1)
2332 /* search_forward needs a path with locks held, do the
2333 * search again for the original key. It is possible
2334 * this will race with a balance and return a path that
2335 * we could modify, but this drop is just an optimization
2336 * and is allowed to miss some leaves.
2338 btrfs_release_path(root, path);
2341 /* setup a max key for search_forward */
2342 other_key.offset = (u64)-1;
2343 other_key.type = key.type;
2344 other_key.objectid = key.objectid;
2346 path->keep_locks = 1;
2347 ret = btrfs_search_forward(root, &found_key, &other_key,
2349 path->keep_locks = 0;
2350 if (ret || found_key.objectid != key.objectid ||
2351 found_key.type != key.type) {
2356 key.offset = found_key.offset;
2357 btrfs_release_path(root, path);
2362 /* we know there's one more slot after us in the tree,
2363 * read that key so we can verify it is also a checksum item
2365 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2367 if (found_key.objectid < inode->i_ino)
2370 if (found_key.type != key.type || found_key.offset < new_size)
2374 * if the key for the next leaf isn't a csum key from this objectid,
2375 * we can't be sure there aren't good items inside this leaf.
2378 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2381 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2382 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2384 * it is safe to delete this leaf, it contains only
2385 * csum items from this inode at an offset >= new_size
2387 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2390 if (root->ref_cows && leaf_gen < trans->transid) {
2391 ref = btrfs_alloc_leaf_ref(root, 0);
2393 ref->root_gen = root->root_key.offset;
2394 ref->bytenr = leaf_start;
2396 ref->generation = leaf_gen;
2399 ret = btrfs_add_leaf_ref(root, ref, 0);
2401 btrfs_free_leaf_ref(root, ref);
2407 btrfs_release_path(root, path);
2409 if (other_key.objectid == inode->i_ino &&
2410 other_key.type == key.type && other_key.offset > key.offset) {
2411 key.offset = other_key.offset;
2417 /* fixup any changes we've made to the path */
2418 path->lowest_level = 0;
2419 path->keep_locks = 0;
2420 btrfs_release_path(root, path);
2427 * this can truncate away extent items, csum items and directory items.
2428 * It starts at a high offset and removes keys until it can't find
2429 * any higher than new_size
2431 * csum items that cross the new i_size are truncated to the new size
2434 * min_type is the minimum key type to truncate down to. If set to 0, this
2435 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2437 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2438 struct btrfs_root *root,
2439 struct inode *inode,
2440 u64 new_size, u32 min_type)
2443 struct btrfs_path *path;
2444 struct btrfs_key key;
2445 struct btrfs_key found_key;
2447 struct extent_buffer *leaf;
2448 struct btrfs_file_extent_item *fi;
2449 u64 extent_start = 0;
2450 u64 extent_num_bytes = 0;
2456 int pending_del_nr = 0;
2457 int pending_del_slot = 0;
2458 int extent_type = -1;
2460 u64 mask = root->sectorsize - 1;
2463 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2464 path = btrfs_alloc_path();
2468 /* FIXME, add redo link to tree so we don't leak on crash */
2469 key.objectid = inode->i_ino;
2470 key.offset = (u64)-1;
2473 btrfs_init_path(path);
2476 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2481 /* there are no items in the tree for us to truncate, we're
2484 if (path->slots[0] == 0) {
2493 leaf = path->nodes[0];
2494 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2495 found_type = btrfs_key_type(&found_key);
2498 if (found_key.objectid != inode->i_ino)
2501 if (found_type < min_type)
2504 item_end = found_key.offset;
2505 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2506 fi = btrfs_item_ptr(leaf, path->slots[0],
2507 struct btrfs_file_extent_item);
2508 extent_type = btrfs_file_extent_type(leaf, fi);
2509 encoding = btrfs_file_extent_compression(leaf, fi);
2510 encoding |= btrfs_file_extent_encryption(leaf, fi);
2511 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2513 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2515 btrfs_file_extent_num_bytes(leaf, fi);
2516 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2517 item_end += btrfs_file_extent_inline_len(leaf,
2522 if (item_end < new_size) {
2523 if (found_type == BTRFS_DIR_ITEM_KEY) {
2524 found_type = BTRFS_INODE_ITEM_KEY;
2525 } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
2526 found_type = BTRFS_EXTENT_DATA_KEY;
2527 } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
2528 found_type = BTRFS_XATTR_ITEM_KEY;
2529 } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
2530 found_type = BTRFS_INODE_REF_KEY;
2531 } else if (found_type) {
2536 btrfs_set_key_type(&key, found_type);
2539 if (found_key.offset >= new_size)
2545 /* FIXME, shrink the extent if the ref count is only 1 */
2546 if (found_type != BTRFS_EXTENT_DATA_KEY)
2549 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2551 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2552 if (!del_item && !encoding) {
2553 u64 orig_num_bytes =
2554 btrfs_file_extent_num_bytes(leaf, fi);
2555 extent_num_bytes = new_size -
2556 found_key.offset + root->sectorsize - 1;
2557 extent_num_bytes = extent_num_bytes &
2558 ~((u64)root->sectorsize - 1);
2559 btrfs_set_file_extent_num_bytes(leaf, fi,
2561 num_dec = (orig_num_bytes -
2563 if (root->ref_cows && extent_start != 0)
2564 inode_sub_bytes(inode, num_dec);
2565 btrfs_mark_buffer_dirty(leaf);
2568 btrfs_file_extent_disk_num_bytes(leaf,
2570 /* FIXME blocksize != 4096 */
2571 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2572 if (extent_start != 0) {
2575 inode_sub_bytes(inode, num_dec);
2577 root_gen = btrfs_header_generation(leaf);
2578 root_owner = btrfs_header_owner(leaf);
2580 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2582 * we can't truncate inline items that have had
2586 btrfs_file_extent_compression(leaf, fi) == 0 &&
2587 btrfs_file_extent_encryption(leaf, fi) == 0 &&
2588 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
2589 u32 size = new_size - found_key.offset;
2591 if (root->ref_cows) {
2592 inode_sub_bytes(inode, item_end + 1 -
2596 btrfs_file_extent_calc_inline_size(size);
2597 ret = btrfs_truncate_item(trans, root, path,
2600 } else if (root->ref_cows) {
2601 inode_sub_bytes(inode, item_end + 1 -
2607 if (!pending_del_nr) {
2608 /* no pending yet, add ourselves */
2609 pending_del_slot = path->slots[0];
2611 } else if (pending_del_nr &&
2612 path->slots[0] + 1 == pending_del_slot) {
2613 /* hop on the pending chunk */
2615 pending_del_slot = path->slots[0];
2617 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
2623 ret = btrfs_free_extent(trans, root, extent_start,
2625 leaf->start, root_owner,
2626 root_gen, inode->i_ino, 0);
2630 if (path->slots[0] == 0) {
2633 btrfs_release_path(root, path);
2638 if (pending_del_nr &&
2639 path->slots[0] + 1 != pending_del_slot) {
2640 struct btrfs_key debug;
2642 btrfs_item_key_to_cpu(path->nodes[0], &debug,
2644 ret = btrfs_del_items(trans, root, path,
2649 btrfs_release_path(root, path);
2655 if (pending_del_nr) {
2656 ret = btrfs_del_items(trans, root, path, pending_del_slot,
2659 btrfs_free_path(path);
2660 inode->i_sb->s_dirt = 1;
2665 * taken from block_truncate_page, but does cow as it zeros out
2666 * any bytes left in the last page in the file.
2668 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
2670 struct inode *inode = mapping->host;
2671 struct btrfs_root *root = BTRFS_I(inode)->root;
2672 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2673 struct btrfs_ordered_extent *ordered;
2675 u32 blocksize = root->sectorsize;
2676 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2677 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2683 if ((offset & (blocksize - 1)) == 0)
2688 page = grab_cache_page(mapping, index);
2692 page_start = page_offset(page);
2693 page_end = page_start + PAGE_CACHE_SIZE - 1;
2695 if (!PageUptodate(page)) {
2696 ret = btrfs_readpage(NULL, page);
2698 if (page->mapping != mapping) {
2700 page_cache_release(page);
2703 if (!PageUptodate(page)) {
2708 wait_on_page_writeback(page);
2710 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2711 set_page_extent_mapped(page);
2713 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2715 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2717 page_cache_release(page);
2718 btrfs_start_ordered_extent(inode, ordered, 1);
2719 btrfs_put_ordered_extent(ordered);
2723 btrfs_set_extent_delalloc(inode, page_start, page_end);
2725 if (offset != PAGE_CACHE_SIZE) {
2727 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
2728 flush_dcache_page(page);
2731 ClearPageChecked(page);
2732 set_page_dirty(page);
2733 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2737 page_cache_release(page);
2742 int btrfs_cont_expand(struct inode *inode, loff_t size)
2744 struct btrfs_trans_handle *trans;
2745 struct btrfs_root *root = BTRFS_I(inode)->root;
2746 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2747 struct extent_map *em;
2748 u64 mask = root->sectorsize - 1;
2749 u64 hole_start = (inode->i_size + mask) & ~mask;
2750 u64 block_end = (size + mask) & ~mask;
2756 if (size <= hole_start)
2759 err = btrfs_check_free_space(root, 1, 0);
2763 btrfs_truncate_page(inode->i_mapping, inode->i_size);
2766 struct btrfs_ordered_extent *ordered;
2767 btrfs_wait_ordered_range(inode, hole_start,
2768 block_end - hole_start);
2769 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2770 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
2773 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2774 btrfs_put_ordered_extent(ordered);
2777 trans = btrfs_start_transaction(root, 1);
2778 btrfs_set_trans_block_group(trans, inode);
2780 cur_offset = hole_start;
2782 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
2783 block_end - cur_offset, 0);
2784 BUG_ON(IS_ERR(em) || !em);
2785 last_byte = min(extent_map_end(em), block_end);
2786 last_byte = (last_byte + mask) & ~mask;
2787 if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
2789 hole_size = last_byte - cur_offset;
2790 err = btrfs_drop_extents(trans, root, inode,
2792 cur_offset + hole_size,
2793 cur_offset, &hint_byte);
2796 err = btrfs_insert_file_extent(trans, root,
2797 inode->i_ino, cur_offset, 0,
2798 0, hole_size, 0, hole_size,
2800 btrfs_drop_extent_cache(inode, hole_start,
2803 free_extent_map(em);
2804 cur_offset = last_byte;
2805 if (err || cur_offset >= block_end)
2809 btrfs_end_transaction(trans, root);
2810 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
2814 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
2816 struct inode *inode = dentry->d_inode;
2819 err = inode_change_ok(inode, attr);
2823 if (S_ISREG(inode->i_mode) &&
2824 attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
2825 err = btrfs_cont_expand(inode, attr->ia_size);
2830 err = inode_setattr(inode, attr);
2832 if (!err && ((attr->ia_valid & ATTR_MODE)))
2833 err = btrfs_acl_chmod(inode);
2837 void btrfs_delete_inode(struct inode *inode)
2839 struct btrfs_trans_handle *trans;
2840 struct btrfs_root *root = BTRFS_I(inode)->root;
2844 truncate_inode_pages(&inode->i_data, 0);
2845 if (is_bad_inode(inode)) {
2846 btrfs_orphan_del(NULL, inode);
2849 btrfs_wait_ordered_range(inode, 0, (u64)-1);
2851 btrfs_i_size_write(inode, 0);
2852 trans = btrfs_start_transaction(root, 1);
2854 btrfs_set_trans_block_group(trans, inode);
2855 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
2857 btrfs_orphan_del(NULL, inode);
2858 goto no_delete_lock;
2861 btrfs_orphan_del(trans, inode);
2863 nr = trans->blocks_used;
2866 btrfs_end_transaction(trans, root);
2867 btrfs_btree_balance_dirty(root, nr);
2871 nr = trans->blocks_used;
2872 btrfs_end_transaction(trans, root);
2873 btrfs_btree_balance_dirty(root, nr);
2879 * this returns the key found in the dir entry in the location pointer.
2880 * If no dir entries were found, location->objectid is 0.
2882 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
2883 struct btrfs_key *location)
2885 const char *name = dentry->d_name.name;
2886 int namelen = dentry->d_name.len;
2887 struct btrfs_dir_item *di;
2888 struct btrfs_path *path;
2889 struct btrfs_root *root = BTRFS_I(dir)->root;
2892 path = btrfs_alloc_path();
2895 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
2899 if (!di || IS_ERR(di)) {
2902 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
2904 btrfs_free_path(path);
2907 location->objectid = 0;
2912 * when we hit a tree root in a directory, the btrfs part of the inode
2913 * needs to be changed to reflect the root directory of the tree root. This
2914 * is kind of like crossing a mount point.
2916 static int fixup_tree_root_location(struct btrfs_root *root,
2917 struct btrfs_key *location,
2918 struct btrfs_root **sub_root,
2919 struct dentry *dentry)
2921 struct btrfs_root_item *ri;
2923 if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
2925 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
2928 *sub_root = btrfs_read_fs_root(root->fs_info, location,
2929 dentry->d_name.name,
2930 dentry->d_name.len);
2931 if (IS_ERR(*sub_root))
2932 return PTR_ERR(*sub_root);
2934 ri = &(*sub_root)->root_item;
2935 location->objectid = btrfs_root_dirid(ri);
2936 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2937 location->offset = 0;
2942 static noinline void init_btrfs_i(struct inode *inode)
2944 struct btrfs_inode *bi = BTRFS_I(inode);
2947 bi->i_default_acl = NULL;
2952 bi->logged_trans = 0;
2953 bi->delalloc_bytes = 0;
2954 bi->disk_i_size = 0;
2956 bi->index_cnt = (u64)-1;
2957 bi->log_dirty_trans = 0;
2958 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2959 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2960 inode->i_mapping, GFP_NOFS);
2961 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2962 inode->i_mapping, GFP_NOFS);
2963 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2964 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2965 mutex_init(&BTRFS_I(inode)->extent_mutex);
2966 mutex_init(&BTRFS_I(inode)->log_mutex);
2969 static int btrfs_init_locked_inode(struct inode *inode, void *p)
2971 struct btrfs_iget_args *args = p;
2972 inode->i_ino = args->ino;
2973 init_btrfs_i(inode);
2974 BTRFS_I(inode)->root = args->root;
2978 static int btrfs_find_actor(struct inode *inode, void *opaque)
2980 struct btrfs_iget_args *args = opaque;
2981 return (args->ino == inode->i_ino &&
2982 args->root == BTRFS_I(inode)->root);
2985 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
2986 struct btrfs_root *root, int wait)
2988 struct inode *inode;
2989 struct btrfs_iget_args args;
2990 args.ino = objectid;
2994 inode = ilookup5(s, objectid, btrfs_find_actor,
2997 inode = ilookup5_nowait(s, objectid, btrfs_find_actor,
3003 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
3004 struct btrfs_root *root)
3006 struct inode *inode;
3007 struct btrfs_iget_args args;
3008 args.ino = objectid;
3011 inode = iget5_locked(s, objectid, btrfs_find_actor,
3012 btrfs_init_locked_inode,
3017 /* Get an inode object given its location and corresponding root.
3018 * Returns in *is_new if the inode was read from disk
3020 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3021 struct btrfs_root *root, int *is_new)
3023 struct inode *inode;
3025 inode = btrfs_iget_locked(s, location->objectid, root);
3027 return ERR_PTR(-EACCES);
3029 if (inode->i_state & I_NEW) {
3030 BTRFS_I(inode)->root = root;
3031 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3032 btrfs_read_locked_inode(inode);
3033 unlock_new_inode(inode);
3044 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3046 struct inode * inode;
3047 struct btrfs_inode *bi = BTRFS_I(dir);
3048 struct btrfs_root *root = bi->root;
3049 struct btrfs_root *sub_root = root;
3050 struct btrfs_key location;
3053 if (dentry->d_name.len > BTRFS_NAME_LEN)
3054 return ERR_PTR(-ENAMETOOLONG);
3056 ret = btrfs_inode_by_name(dir, dentry, &location);
3059 return ERR_PTR(ret);
3062 if (location.objectid) {
3063 ret = fixup_tree_root_location(root, &location, &sub_root,
3066 return ERR_PTR(ret);
3068 return ERR_PTR(-ENOENT);
3069 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
3071 return ERR_CAST(inode);
3076 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3077 struct nameidata *nd)
3079 struct inode *inode;
3081 if (dentry->d_name.len > BTRFS_NAME_LEN)
3082 return ERR_PTR(-ENAMETOOLONG);
3084 inode = btrfs_lookup_dentry(dir, dentry);
3086 return ERR_CAST(inode);
3088 return d_splice_alias(inode, dentry);
3091 static unsigned char btrfs_filetype_table[] = {
3092 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3095 static int btrfs_real_readdir(struct file *filp, void *dirent,
3098 struct inode *inode = filp->f_dentry->d_inode;
3099 struct btrfs_root *root = BTRFS_I(inode)->root;
3100 struct btrfs_item *item;
3101 struct btrfs_dir_item *di;
3102 struct btrfs_key key;
3103 struct btrfs_key found_key;
3104 struct btrfs_path *path;
3107 struct extent_buffer *leaf;
3110 unsigned char d_type;
3115 int key_type = BTRFS_DIR_INDEX_KEY;
3120 /* FIXME, use a real flag for deciding about the key type */
3121 if (root->fs_info->tree_root == root)
3122 key_type = BTRFS_DIR_ITEM_KEY;
3124 /* special case for "." */
3125 if (filp->f_pos == 0) {
3126 over = filldir(dirent, ".", 1,
3133 /* special case for .., just use the back ref */
3134 if (filp->f_pos == 1) {
3135 u64 pino = parent_ino(filp->f_path.dentry);
3136 over = filldir(dirent, "..", 2,
3142 path = btrfs_alloc_path();
3145 btrfs_set_key_type(&key, key_type);
3146 key.offset = filp->f_pos;
3147 key.objectid = inode->i_ino;
3149 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3155 leaf = path->nodes[0];
3156 nritems = btrfs_header_nritems(leaf);
3157 slot = path->slots[0];
3158 if (advance || slot >= nritems) {
3159 if (slot >= nritems - 1) {
3160 ret = btrfs_next_leaf(root, path);
3163 leaf = path->nodes[0];
3164 nritems = btrfs_header_nritems(leaf);
3165 slot = path->slots[0];
3173 item = btrfs_item_nr(leaf, slot);
3174 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3176 if (found_key.objectid != key.objectid)
3178 if (btrfs_key_type(&found_key) != key_type)
3180 if (found_key.offset < filp->f_pos)
3183 filp->f_pos = found_key.offset;
3185 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3187 di_total = btrfs_item_size(leaf, item);
3189 while (di_cur < di_total) {
3190 struct btrfs_key location;
3192 name_len = btrfs_dir_name_len(leaf, di);
3193 if (name_len <= sizeof(tmp_name)) {
3194 name_ptr = tmp_name;
3196 name_ptr = kmalloc(name_len, GFP_NOFS);
3202 read_extent_buffer(leaf, name_ptr,
3203 (unsigned long)(di + 1), name_len);
3205 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3206 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3208 /* is this a reference to our own snapshot? If so
3211 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3212 location.objectid == root->root_key.objectid) {
3216 over = filldir(dirent, name_ptr, name_len,
3217 found_key.offset, location.objectid,
3221 if (name_ptr != tmp_name)
3226 di_len = btrfs_dir_name_len(leaf, di) +
3227 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3229 di = (struct btrfs_dir_item *)((char *)di + di_len);
3233 /* Reached end of directory/root. Bump pos past the last item. */
3234 if (key_type == BTRFS_DIR_INDEX_KEY)
3235 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
3241 btrfs_free_path(path);
3245 int btrfs_write_inode(struct inode *inode, int wait)
3247 struct btrfs_root *root = BTRFS_I(inode)->root;
3248 struct btrfs_trans_handle *trans;
3251 if (root->fs_info->btree_inode == inode)
3255 trans = btrfs_join_transaction(root, 1);
3256 btrfs_set_trans_block_group(trans, inode);
3257 ret = btrfs_commit_transaction(trans, root);
3263 * This is somewhat expensive, updating the tree every time the
3264 * inode changes. But, it is most likely to find the inode in cache.
3265 * FIXME, needs more benchmarking...there are no reasons other than performance
3266 * to keep or drop this code.
3268 void btrfs_dirty_inode(struct inode *inode)
3270 struct btrfs_root *root = BTRFS_I(inode)->root;
3271 struct btrfs_trans_handle *trans;
3273 trans = btrfs_join_transaction(root, 1);
3274 btrfs_set_trans_block_group(trans, inode);
3275 btrfs_update_inode(trans, root, inode);
3276 btrfs_end_transaction(trans, root);
3280 * find the highest existing sequence number in a directory
3281 * and then set the in-memory index_cnt variable to reflect
3282 * free sequence numbers
3284 static int btrfs_set_inode_index_count(struct inode *inode)
3286 struct btrfs_root *root = BTRFS_I(inode)->root;
3287 struct btrfs_key key, found_key;
3288 struct btrfs_path *path;
3289 struct extent_buffer *leaf;
3292 key.objectid = inode->i_ino;
3293 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
3294 key.offset = (u64)-1;
3296 path = btrfs_alloc_path();
3300 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3303 /* FIXME: we should be able to handle this */
3309 * MAGIC NUMBER EXPLANATION:
3310 * since we search a directory based on f_pos we have to start at 2
3311 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
3312 * else has to start at 2
3314 if (path->slots[0] == 0) {
3315 BTRFS_I(inode)->index_cnt = 2;
3321 leaf = path->nodes[0];
3322 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3324 if (found_key.objectid != inode->i_ino ||
3325 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
3326 BTRFS_I(inode)->index_cnt = 2;
3330 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
3332 btrfs_free_path(path);
3337 * helper to find a free sequence number in a given directory. This current
3338 * code is very simple, later versions will do smarter things in the btree
3340 int btrfs_set_inode_index(struct inode *dir, u64 *index)
3344 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
3345 ret = btrfs_set_inode_index_count(dir);
3351 *index = BTRFS_I(dir)->index_cnt;
3352 BTRFS_I(dir)->index_cnt++;
3357 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
3358 struct btrfs_root *root,
3360 const char *name, int name_len,
3363 struct btrfs_block_group_cache *group,
3364 int mode, u64 *index)
3366 struct inode *inode;
3367 struct btrfs_inode_item *inode_item;
3368 struct btrfs_block_group_cache *new_inode_group;
3369 struct btrfs_key *location;
3370 struct btrfs_path *path;
3371 struct btrfs_inode_ref *ref;
3372 struct btrfs_key key[2];
3378 path = btrfs_alloc_path();
3381 inode = new_inode(root->fs_info->sb);
3383 return ERR_PTR(-ENOMEM);
3386 ret = btrfs_set_inode_index(dir, index);
3388 return ERR_PTR(ret);
3391 * index_cnt is ignored for everything but a dir,
3392 * btrfs_get_inode_index_count has an explanation for the magic
3395 init_btrfs_i(inode);
3396 BTRFS_I(inode)->index_cnt = 2;
3397 BTRFS_I(inode)->root = root;
3398 BTRFS_I(inode)->generation = trans->transid;
3404 new_inode_group = btrfs_find_block_group(root, group, 0,
3405 BTRFS_BLOCK_GROUP_METADATA, owner);
3406 if (!new_inode_group) {
3407 printk("find_block group failed\n");
3408 new_inode_group = group;
3410 BTRFS_I(inode)->block_group = new_inode_group;
3412 key[0].objectid = objectid;
3413 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
3416 key[1].objectid = objectid;
3417 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
3418 key[1].offset = ref_objectid;
3420 sizes[0] = sizeof(struct btrfs_inode_item);
3421 sizes[1] = name_len + sizeof(*ref);
3423 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
3427 if (objectid > root->highest_inode)
3428 root->highest_inode = objectid;
3430 inode->i_uid = current_fsuid();
3431 inode->i_gid = current_fsgid();
3432 inode->i_mode = mode;
3433 inode->i_ino = objectid;
3434 inode_set_bytes(inode, 0);
3435 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3436 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
3437 struct btrfs_inode_item);
3438 fill_inode_item(trans, path->nodes[0], inode_item, inode);
3440 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
3441 struct btrfs_inode_ref);
3442 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
3443 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
3444 ptr = (unsigned long)(ref + 1);
3445 write_extent_buffer(path->nodes[0], name, ptr, name_len);
3447 btrfs_mark_buffer_dirty(path->nodes[0]);
3448 btrfs_free_path(path);
3450 location = &BTRFS_I(inode)->location;
3451 location->objectid = objectid;
3452 location->offset = 0;
3453 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
3455 insert_inode_hash(inode);
3459 BTRFS_I(dir)->index_cnt--;
3460 btrfs_free_path(path);
3461 return ERR_PTR(ret);
3464 static inline u8 btrfs_inode_type(struct inode *inode)
3466 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
3470 * utility function to add 'inode' into 'parent_inode' with
3471 * a give name and a given sequence number.
3472 * if 'add_backref' is true, also insert a backref from the
3473 * inode to the parent directory.
3475 int btrfs_add_link(struct btrfs_trans_handle *trans,
3476 struct inode *parent_inode, struct inode *inode,
3477 const char *name, int name_len, int add_backref, u64 index)
3480 struct btrfs_key key;
3481 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
3483 key.objectid = inode->i_ino;
3484 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
3487 ret = btrfs_insert_dir_item(trans, root, name, name_len,
3488 parent_inode->i_ino,
3489 &key, btrfs_inode_type(inode),
3493 ret = btrfs_insert_inode_ref(trans, root,
3496 parent_inode->i_ino,
3499 btrfs_i_size_write(parent_inode, parent_inode->i_size +
3501 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
3502 ret = btrfs_update_inode(trans, root, parent_inode);
3507 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
3508 struct dentry *dentry, struct inode *inode,
3509 int backref, u64 index)
3511 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3512 inode, dentry->d_name.name,
3513 dentry->d_name.len, backref, index);
3515 d_instantiate(dentry, inode);
3523 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
3524 int mode, dev_t rdev)
3526 struct btrfs_trans_handle *trans;
3527 struct btrfs_root *root = BTRFS_I(dir)->root;
3528 struct inode *inode = NULL;
3532 unsigned long nr = 0;
3535 if (!new_valid_dev(rdev))
3538 err = btrfs_check_free_space(root, 1, 0);
3542 trans = btrfs_start_transaction(root, 1);
3543 btrfs_set_trans_block_group(trans, dir);
3545 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3551 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3553 dentry->d_parent->d_inode->i_ino, objectid,
3554 BTRFS_I(dir)->block_group, mode, &index);
3555 err = PTR_ERR(inode);
3559 err = btrfs_init_acl(inode, dir);
3565 btrfs_set_trans_block_group(trans, inode);
3566 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3570 inode->i_op = &btrfs_special_inode_operations;
3571 init_special_inode(inode, inode->i_mode, rdev);
3572 btrfs_update_inode(trans, root, inode);
3574 dir->i_sb->s_dirt = 1;
3575 btrfs_update_inode_block_group(trans, inode);
3576 btrfs_update_inode_block_group(trans, dir);
3578 nr = trans->blocks_used;
3579 btrfs_end_transaction_throttle(trans, root);
3582 inode_dec_link_count(inode);
3585 btrfs_btree_balance_dirty(root, nr);
3589 static int btrfs_create(struct inode *dir, struct dentry *dentry,
3590 int mode, struct nameidata *nd)
3592 struct btrfs_trans_handle *trans;
3593 struct btrfs_root *root = BTRFS_I(dir)->root;
3594 struct inode *inode = NULL;
3597 unsigned long nr = 0;
3601 err = btrfs_check_free_space(root, 1, 0);
3604 trans = btrfs_start_transaction(root, 1);
3605 btrfs_set_trans_block_group(trans, dir);
3607 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3613 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3615 dentry->d_parent->d_inode->i_ino,
3616 objectid, BTRFS_I(dir)->block_group, mode,
3618 err = PTR_ERR(inode);
3622 err = btrfs_init_acl(inode, dir);
3628 btrfs_set_trans_block_group(trans, inode);
3629 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3633 inode->i_mapping->a_ops = &btrfs_aops;
3634 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3635 inode->i_fop = &btrfs_file_operations;
3636 inode->i_op = &btrfs_file_inode_operations;
3637 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3639 dir->i_sb->s_dirt = 1;
3640 btrfs_update_inode_block_group(trans, inode);
3641 btrfs_update_inode_block_group(trans, dir);
3643 nr = trans->blocks_used;
3644 btrfs_end_transaction_throttle(trans, root);
3647 inode_dec_link_count(inode);
3650 btrfs_btree_balance_dirty(root, nr);
3654 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
3655 struct dentry *dentry)
3657 struct btrfs_trans_handle *trans;
3658 struct btrfs_root *root = BTRFS_I(dir)->root;
3659 struct inode *inode = old_dentry->d_inode;
3661 unsigned long nr = 0;
3665 if (inode->i_nlink == 0)
3668 btrfs_inc_nlink(inode);
3669 err = btrfs_check_free_space(root, 1, 0);
3672 err = btrfs_set_inode_index(dir, &index);
3676 trans = btrfs_start_transaction(root, 1);
3678 btrfs_set_trans_block_group(trans, dir);
3679 atomic_inc(&inode->i_count);
3681 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
3686 dir->i_sb->s_dirt = 1;
3687 btrfs_update_inode_block_group(trans, dir);
3688 err = btrfs_update_inode(trans, root, inode);
3693 nr = trans->blocks_used;
3694 btrfs_end_transaction_throttle(trans, root);
3697 inode_dec_link_count(inode);
3700 btrfs_btree_balance_dirty(root, nr);
3704 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
3706 struct inode *inode = NULL;
3707 struct btrfs_trans_handle *trans;
3708 struct btrfs_root *root = BTRFS_I(dir)->root;
3710 int drop_on_err = 0;
3713 unsigned long nr = 1;
3715 err = btrfs_check_free_space(root, 1, 0);
3719 trans = btrfs_start_transaction(root, 1);
3720 btrfs_set_trans_block_group(trans, dir);
3722 if (IS_ERR(trans)) {
3723 err = PTR_ERR(trans);
3727 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3733 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3735 dentry->d_parent->d_inode->i_ino, objectid,
3736 BTRFS_I(dir)->block_group, S_IFDIR | mode,
3738 if (IS_ERR(inode)) {
3739 err = PTR_ERR(inode);
3745 err = btrfs_init_acl(inode, dir);
3749 inode->i_op = &btrfs_dir_inode_operations;
3750 inode->i_fop = &btrfs_dir_file_operations;
3751 btrfs_set_trans_block_group(trans, inode);
3753 btrfs_i_size_write(inode, 0);
3754 err = btrfs_update_inode(trans, root, inode);
3758 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
3759 inode, dentry->d_name.name,
3760 dentry->d_name.len, 0, index);
3764 d_instantiate(dentry, inode);
3766 dir->i_sb->s_dirt = 1;
3767 btrfs_update_inode_block_group(trans, inode);
3768 btrfs_update_inode_block_group(trans, dir);
3771 nr = trans->blocks_used;
3772 btrfs_end_transaction_throttle(trans, root);
3777 btrfs_btree_balance_dirty(root, nr);
3781 /* helper for btfs_get_extent. Given an existing extent in the tree,
3782 * and an extent that you want to insert, deal with overlap and insert
3783 * the new extent into the tree.
3785 static int merge_extent_mapping(struct extent_map_tree *em_tree,
3786 struct extent_map *existing,
3787 struct extent_map *em,
3788 u64 map_start, u64 map_len)
3792 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
3793 start_diff = map_start - em->start;
3794 em->start = map_start;
3796 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
3797 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
3798 em->block_start += start_diff;
3799 em->block_len -= start_diff;
3801 return add_extent_mapping(em_tree, em);
3804 static noinline int uncompress_inline(struct btrfs_path *path,
3805 struct inode *inode, struct page *page,
3806 size_t pg_offset, u64 extent_offset,
3807 struct btrfs_file_extent_item *item)
3810 struct extent_buffer *leaf = path->nodes[0];
3813 unsigned long inline_size;
3816 WARN_ON(pg_offset != 0);
3817 max_size = btrfs_file_extent_ram_bytes(leaf, item);
3818 inline_size = btrfs_file_extent_inline_item_len(leaf,
3819 btrfs_item_nr(leaf, path->slots[0]));
3820 tmp = kmalloc(inline_size, GFP_NOFS);
3821 ptr = btrfs_file_extent_inline_start(item);
3823 read_extent_buffer(leaf, tmp, ptr, inline_size);
3825 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
3826 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
3827 inline_size, max_size);
3829 char *kaddr = kmap_atomic(page, KM_USER0);
3830 unsigned long copy_size = min_t(u64,
3831 PAGE_CACHE_SIZE - pg_offset,
3832 max_size - extent_offset);
3833 memset(kaddr + pg_offset, 0, copy_size);
3834 kunmap_atomic(kaddr, KM_USER0);
3841 * a bit scary, this does extent mapping from logical file offset to the disk.
3842 * the ugly parts come from merging extents from the disk with the
3843 * in-ram representation. This gets more complex because of the data=ordered code,
3844 * where the in-ram extents might be locked pending data=ordered completion.
3846 * This also copies inline extents directly into the page.
3848 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
3849 size_t pg_offset, u64 start, u64 len,
3855 u64 extent_start = 0;
3857 u64 objectid = inode->i_ino;
3859 struct btrfs_path *path = NULL;
3860 struct btrfs_root *root = BTRFS_I(inode)->root;
3861 struct btrfs_file_extent_item *item;
3862 struct extent_buffer *leaf;
3863 struct btrfs_key found_key;
3864 struct extent_map *em = NULL;
3865 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3866 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3867 struct btrfs_trans_handle *trans = NULL;
3871 spin_lock(&em_tree->lock);
3872 em = lookup_extent_mapping(em_tree, start, len);
3874 em->bdev = root->fs_info->fs_devices->latest_bdev;
3875 spin_unlock(&em_tree->lock);
3878 if (em->start > start || em->start + em->len <= start)
3879 free_extent_map(em);
3880 else if (em->block_start == EXTENT_MAP_INLINE && page)
3881 free_extent_map(em);
3885 em = alloc_extent_map(GFP_NOFS);
3890 em->bdev = root->fs_info->fs_devices->latest_bdev;
3891 em->start = EXTENT_MAP_HOLE;
3892 em->orig_start = EXTENT_MAP_HOLE;
3894 em->block_len = (u64)-1;
3897 path = btrfs_alloc_path();
3901 ret = btrfs_lookup_file_extent(trans, root, path,
3902 objectid, start, trans != NULL);
3909 if (path->slots[0] == 0)
3914 leaf = path->nodes[0];
3915 item = btrfs_item_ptr(leaf, path->slots[0],
3916 struct btrfs_file_extent_item);
3917 /* are we inside the extent that was found? */
3918 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3919 found_type = btrfs_key_type(&found_key);
3920 if (found_key.objectid != objectid ||
3921 found_type != BTRFS_EXTENT_DATA_KEY) {
3925 found_type = btrfs_file_extent_type(leaf, item);
3926 extent_start = found_key.offset;
3927 compressed = btrfs_file_extent_compression(leaf, item);
3928 if (found_type == BTRFS_FILE_EXTENT_REG ||
3929 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3930 extent_end = extent_start +
3931 btrfs_file_extent_num_bytes(leaf, item);
3932 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3934 size = btrfs_file_extent_inline_len(leaf, item);
3935 extent_end = (extent_start + size + root->sectorsize - 1) &
3936 ~((u64)root->sectorsize - 1);
3939 if (start >= extent_end) {
3941 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3942 ret = btrfs_next_leaf(root, path);
3949 leaf = path->nodes[0];
3951 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3952 if (found_key.objectid != objectid ||
3953 found_key.type != BTRFS_EXTENT_DATA_KEY)
3955 if (start + len <= found_key.offset)
3958 em->len = found_key.offset - start;
3962 if (found_type == BTRFS_FILE_EXTENT_REG ||
3963 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
3964 em->start = extent_start;
3965 em->len = extent_end - extent_start;
3966 em->orig_start = extent_start -
3967 btrfs_file_extent_offset(leaf, item);
3968 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
3970 em->block_start = EXTENT_MAP_HOLE;
3974 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
3975 em->block_start = bytenr;
3976 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
3979 bytenr += btrfs_file_extent_offset(leaf, item);
3980 em->block_start = bytenr;
3981 em->block_len = em->len;
3982 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
3983 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
3986 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3990 size_t extent_offset;
3993 em->block_start = EXTENT_MAP_INLINE;
3994 if (!page || create) {
3995 em->start = extent_start;
3996 em->len = extent_end - extent_start;
4000 size = btrfs_file_extent_inline_len(leaf, item);
4001 extent_offset = page_offset(page) + pg_offset - extent_start;
4002 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4003 size - extent_offset);
4004 em->start = extent_start + extent_offset;
4005 em->len = (copy_size + root->sectorsize - 1) &
4006 ~((u64)root->sectorsize - 1);
4007 em->orig_start = EXTENT_MAP_INLINE;
4009 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4010 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4011 if (create == 0 && !PageUptodate(page)) {
4012 if (btrfs_file_extent_compression(leaf, item) ==
4013 BTRFS_COMPRESS_ZLIB) {
4014 ret = uncompress_inline(path, inode, page,
4016 extent_offset, item);
4020 read_extent_buffer(leaf, map + pg_offset, ptr,
4024 flush_dcache_page(page);
4025 } else if (create && PageUptodate(page)) {
4028 free_extent_map(em);
4030 btrfs_release_path(root, path);
4031 trans = btrfs_join_transaction(root, 1);
4035 write_extent_buffer(leaf, map + pg_offset, ptr,
4038 btrfs_mark_buffer_dirty(leaf);
4040 set_extent_uptodate(io_tree, em->start,
4041 extent_map_end(em) - 1, GFP_NOFS);
4044 printk("unkknown found_type %d\n", found_type);
4051 em->block_start = EXTENT_MAP_HOLE;
4052 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4054 btrfs_release_path(root, path);
4055 if (em->start > start || extent_map_end(em) <= start) {
4056 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
4062 spin_lock(&em_tree->lock);
4063 ret = add_extent_mapping(em_tree, em);
4064 /* it is possible that someone inserted the extent into the tree
4065 * while we had the lock dropped. It is also possible that
4066 * an overlapping map exists in the tree
4068 if (ret == -EEXIST) {
4069 struct extent_map *existing;
4073 existing = lookup_extent_mapping(em_tree, start, len);
4074 if (existing && (existing->start > start ||
4075 existing->start + existing->len <= start)) {
4076 free_extent_map(existing);
4080 existing = lookup_extent_mapping(em_tree, em->start,
4083 err = merge_extent_mapping(em_tree, existing,
4086 free_extent_map(existing);
4088 free_extent_map(em);
4093 printk("failing to insert %Lu %Lu\n",
4095 free_extent_map(em);
4099 free_extent_map(em);
4104 spin_unlock(&em_tree->lock);
4107 btrfs_free_path(path);
4109 ret = btrfs_end_transaction(trans, root);
4115 free_extent_map(em);
4117 return ERR_PTR(err);
4122 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4123 const struct iovec *iov, loff_t offset,
4124 unsigned long nr_segs)
4129 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
4131 return extent_bmap(mapping, iblock, btrfs_get_extent);
4134 int btrfs_readpage(struct file *file, struct page *page)
4136 struct extent_io_tree *tree;
4137 tree = &BTRFS_I(page->mapping->host)->io_tree;
4138 return extent_read_full_page(tree, page, btrfs_get_extent);
4141 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4143 struct extent_io_tree *tree;
4146 if (current->flags & PF_MEMALLOC) {
4147 redirty_page_for_writepage(wbc, page);
4151 tree = &BTRFS_I(page->mapping->host)->io_tree;
4152 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4155 int btrfs_writepages(struct address_space *mapping,
4156 struct writeback_control *wbc)
4158 struct extent_io_tree *tree;
4160 tree = &BTRFS_I(mapping->host)->io_tree;
4161 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4165 btrfs_readpages(struct file *file, struct address_space *mapping,
4166 struct list_head *pages, unsigned nr_pages)
4168 struct extent_io_tree *tree;
4169 tree = &BTRFS_I(mapping->host)->io_tree;
4170 return extent_readpages(tree, mapping, pages, nr_pages,
4173 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4175 struct extent_io_tree *tree;
4176 struct extent_map_tree *map;
4179 tree = &BTRFS_I(page->mapping->host)->io_tree;
4180 map = &BTRFS_I(page->mapping->host)->extent_tree;
4181 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4183 ClearPagePrivate(page);
4184 set_page_private(page, 0);
4185 page_cache_release(page);
4190 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4192 if (PageWriteback(page) || PageDirty(page))
4194 return __btrfs_releasepage(page, gfp_flags);
4197 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
4199 struct extent_io_tree *tree;
4200 struct btrfs_ordered_extent *ordered;
4201 u64 page_start = page_offset(page);
4202 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
4204 wait_on_page_writeback(page);
4205 tree = &BTRFS_I(page->mapping->host)->io_tree;
4207 btrfs_releasepage(page, GFP_NOFS);
4211 lock_extent(tree, page_start, page_end, GFP_NOFS);
4212 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
4216 * IO on this page will never be started, so we need
4217 * to account for any ordered extents now
4219 clear_extent_bit(tree, page_start, page_end,
4220 EXTENT_DIRTY | EXTENT_DELALLOC |
4221 EXTENT_LOCKED, 1, 0, GFP_NOFS);
4222 btrfs_finish_ordered_io(page->mapping->host,
4223 page_start, page_end);
4224 btrfs_put_ordered_extent(ordered);
4225 lock_extent(tree, page_start, page_end, GFP_NOFS);
4227 clear_extent_bit(tree, page_start, page_end,
4228 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
4231 __btrfs_releasepage(page, GFP_NOFS);
4233 ClearPageChecked(page);
4234 if (PagePrivate(page)) {
4235 ClearPagePrivate(page);
4236 set_page_private(page, 0);
4237 page_cache_release(page);
4242 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
4243 * called from a page fault handler when a page is first dirtied. Hence we must
4244 * be careful to check for EOF conditions here. We set the page up correctly
4245 * for a written page which means we get ENOSPC checking when writing into
4246 * holes and correct delalloc and unwritten extent mapping on filesystems that
4247 * support these features.
4249 * We are not allowed to take the i_mutex here so we have to play games to
4250 * protect against truncate races as the page could now be beyond EOF. Because
4251 * vmtruncate() writes the inode size before removing pages, once we have the
4252 * page lock we can determine safely if the page is beyond EOF. If it is not
4253 * beyond EOF, then the page is guaranteed safe against truncation until we
4256 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
4258 struct inode *inode = fdentry(vma->vm_file)->d_inode;
4259 struct btrfs_root *root = BTRFS_I(inode)->root;
4260 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4261 struct btrfs_ordered_extent *ordered;
4263 unsigned long zero_start;
4269 ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
4276 size = i_size_read(inode);
4277 page_start = page_offset(page);
4278 page_end = page_start + PAGE_CACHE_SIZE - 1;
4280 if ((page->mapping != inode->i_mapping) ||
4281 (page_start >= size)) {
4282 /* page got truncated out from underneath us */
4285 wait_on_page_writeback(page);
4287 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
4288 set_page_extent_mapped(page);
4291 * we can't set the delalloc bits if there are pending ordered
4292 * extents. Drop our locks and wait for them to finish
4294 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4296 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4298 btrfs_start_ordered_extent(inode, ordered, 1);
4299 btrfs_put_ordered_extent(ordered);
4303 btrfs_set_extent_delalloc(inode, page_start, page_end);
4306 /* page is wholly or partially inside EOF */
4307 if (page_start + PAGE_CACHE_SIZE > size)
4308 zero_start = size & ~PAGE_CACHE_MASK;
4310 zero_start = PAGE_CACHE_SIZE;
4312 if (zero_start != PAGE_CACHE_SIZE) {
4314 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
4315 flush_dcache_page(page);
4318 ClearPageChecked(page);
4319 set_page_dirty(page);
4320 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
4328 static void btrfs_truncate(struct inode *inode)
4330 struct btrfs_root *root = BTRFS_I(inode)->root;
4332 struct btrfs_trans_handle *trans;
4334 u64 mask = root->sectorsize - 1;
4336 if (!S_ISREG(inode->i_mode))
4338 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4341 btrfs_truncate_page(inode->i_mapping, inode->i_size);
4342 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
4344 trans = btrfs_start_transaction(root, 1);
4345 btrfs_set_trans_block_group(trans, inode);
4346 btrfs_i_size_write(inode, inode->i_size);
4348 ret = btrfs_orphan_add(trans, inode);
4351 /* FIXME, add redo link to tree so we don't leak on crash */
4352 ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
4353 BTRFS_EXTENT_DATA_KEY);
4354 btrfs_update_inode(trans, root, inode);
4356 ret = btrfs_orphan_del(trans, inode);
4360 nr = trans->blocks_used;
4361 ret = btrfs_end_transaction_throttle(trans, root);
4363 btrfs_btree_balance_dirty(root, nr);
4367 * create a new subvolume directory/inode (helper for the ioctl).
4369 int btrfs_create_subvol_root(struct btrfs_root *new_root, struct dentry *dentry,
4370 struct btrfs_trans_handle *trans, u64 new_dirid,
4371 struct btrfs_block_group_cache *block_group)
4373 struct inode *inode;
4377 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
4378 new_dirid, block_group, S_IFDIR | 0700, &index);
4380 return PTR_ERR(inode);
4381 inode->i_op = &btrfs_dir_inode_operations;
4382 inode->i_fop = &btrfs_dir_file_operations;
4385 btrfs_i_size_write(inode, 0);
4387 error = btrfs_update_inode(trans, new_root, inode);
4391 d_instantiate(dentry, inode);
4395 /* helper function for file defrag and space balancing. This
4396 * forces readahead on a given range of bytes in an inode
4398 unsigned long btrfs_force_ra(struct address_space *mapping,
4399 struct file_ra_state *ra, struct file *file,
4400 pgoff_t offset, pgoff_t last_index)
4402 pgoff_t req_size = last_index - offset + 1;
4404 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
4405 return offset + req_size;
4408 struct inode *btrfs_alloc_inode(struct super_block *sb)
4410 struct btrfs_inode *ei;
4412 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
4416 ei->logged_trans = 0;
4417 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
4418 ei->i_acl = BTRFS_ACL_NOT_CACHED;
4419 ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
4420 INIT_LIST_HEAD(&ei->i_orphan);
4421 return &ei->vfs_inode;
4424 void btrfs_destroy_inode(struct inode *inode)
4426 struct btrfs_ordered_extent *ordered;
4427 WARN_ON(!list_empty(&inode->i_dentry));
4428 WARN_ON(inode->i_data.nrpages);
4430 if (BTRFS_I(inode)->i_acl &&
4431 BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
4432 posix_acl_release(BTRFS_I(inode)->i_acl);
4433 if (BTRFS_I(inode)->i_default_acl &&
4434 BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
4435 posix_acl_release(BTRFS_I(inode)->i_default_acl);
4437 spin_lock(&BTRFS_I(inode)->root->list_lock);
4438 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
4439 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
4440 " list\n", inode->i_ino);
4443 spin_unlock(&BTRFS_I(inode)->root->list_lock);
4446 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
4450 printk("found ordered extent %Lu %Lu\n",
4451 ordered->file_offset, ordered->len);
4452 btrfs_remove_ordered_extent(inode, ordered);
4453 btrfs_put_ordered_extent(ordered);
4454 btrfs_put_ordered_extent(ordered);
4457 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
4458 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
4461 static void init_once(void *foo)
4463 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
4465 inode_init_once(&ei->vfs_inode);
4468 void btrfs_destroy_cachep(void)
4470 if (btrfs_inode_cachep)
4471 kmem_cache_destroy(btrfs_inode_cachep);
4472 if (btrfs_trans_handle_cachep)
4473 kmem_cache_destroy(btrfs_trans_handle_cachep);
4474 if (btrfs_transaction_cachep)
4475 kmem_cache_destroy(btrfs_transaction_cachep);
4476 if (btrfs_bit_radix_cachep)
4477 kmem_cache_destroy(btrfs_bit_radix_cachep);
4478 if (btrfs_path_cachep)
4479 kmem_cache_destroy(btrfs_path_cachep);
4482 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
4483 unsigned long extra_flags,
4484 void (*ctor)(void *))
4486 return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
4487 SLAB_MEM_SPREAD | extra_flags), ctor);
4490 int btrfs_init_cachep(void)
4492 btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
4493 sizeof(struct btrfs_inode),
4495 if (!btrfs_inode_cachep)
4497 btrfs_trans_handle_cachep =
4498 btrfs_cache_create("btrfs_trans_handle_cache",
4499 sizeof(struct btrfs_trans_handle),
4501 if (!btrfs_trans_handle_cachep)
4503 btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
4504 sizeof(struct btrfs_transaction),
4506 if (!btrfs_transaction_cachep)
4508 btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
4509 sizeof(struct btrfs_path),
4511 if (!btrfs_path_cachep)
4513 btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
4514 SLAB_DESTROY_BY_RCU, NULL);
4515 if (!btrfs_bit_radix_cachep)
4519 btrfs_destroy_cachep();
4523 static int btrfs_getattr(struct vfsmount *mnt,
4524 struct dentry *dentry, struct kstat *stat)
4526 struct inode *inode = dentry->d_inode;
4527 generic_fillattr(inode, stat);
4528 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
4529 stat->blksize = PAGE_CACHE_SIZE;
4530 stat->blocks = (inode_get_bytes(inode) +
4531 BTRFS_I(inode)->delalloc_bytes) >> 9;
4535 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
4536 struct inode * new_dir,struct dentry *new_dentry)
4538 struct btrfs_trans_handle *trans;
4539 struct btrfs_root *root = BTRFS_I(old_dir)->root;
4540 struct inode *new_inode = new_dentry->d_inode;
4541 struct inode *old_inode = old_dentry->d_inode;
4542 struct timespec ctime = CURRENT_TIME;
4546 /* we're not allowed to rename between subvolumes */
4547 if (BTRFS_I(old_inode)->root->root_key.objectid !=
4548 BTRFS_I(new_dir)->root->root_key.objectid)
4551 if (S_ISDIR(old_inode->i_mode) && new_inode &&
4552 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
4556 /* to rename a snapshot or subvolume, we need to juggle the
4557 * backrefs. This isn't coded yet
4559 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
4562 ret = btrfs_check_free_space(root, 1, 0);
4566 trans = btrfs_start_transaction(root, 1);
4568 btrfs_set_trans_block_group(trans, new_dir);
4570 btrfs_inc_nlink(old_dentry->d_inode);
4571 old_dir->i_ctime = old_dir->i_mtime = ctime;
4572 new_dir->i_ctime = new_dir->i_mtime = ctime;
4573 old_inode->i_ctime = ctime;
4575 ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
4576 old_dentry->d_name.name,
4577 old_dentry->d_name.len);
4582 new_inode->i_ctime = CURRENT_TIME;
4583 ret = btrfs_unlink_inode(trans, root, new_dir,
4584 new_dentry->d_inode,
4585 new_dentry->d_name.name,
4586 new_dentry->d_name.len);
4589 if (new_inode->i_nlink == 0) {
4590 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
4596 ret = btrfs_set_inode_index(new_dir, &index);
4600 ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
4601 old_inode, new_dentry->d_name.name,
4602 new_dentry->d_name.len, 1, index);
4607 btrfs_end_transaction_throttle(trans, root);
4613 * some fairly slow code that needs optimization. This walks the list
4614 * of all the inodes with pending delalloc and forces them to disk.
4616 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
4618 struct list_head *head = &root->fs_info->delalloc_inodes;
4619 struct btrfs_inode *binode;
4620 struct inode *inode;
4621 unsigned long flags;
4623 if (root->fs_info->sb->s_flags & MS_RDONLY)
4626 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4627 while(!list_empty(head)) {
4628 binode = list_entry(head->next, struct btrfs_inode,
4630 inode = igrab(&binode->vfs_inode);
4632 list_del_init(&binode->delalloc_inodes);
4633 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4635 filemap_flush(inode->i_mapping);
4639 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
4641 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
4643 /* the filemap_flush will queue IO into the worker threads, but
4644 * we have to make sure the IO is actually started and that
4645 * ordered extents get created before we return
4647 atomic_inc(&root->fs_info->async_submit_draining);
4648 while(atomic_read(&root->fs_info->nr_async_submits) ||
4649 atomic_read(&root->fs_info->async_delalloc_pages)) {
4650 wait_event(root->fs_info->async_submit_wait,
4651 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
4652 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
4654 atomic_dec(&root->fs_info->async_submit_draining);
4658 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
4659 const char *symname)
4661 struct btrfs_trans_handle *trans;
4662 struct btrfs_root *root = BTRFS_I(dir)->root;
4663 struct btrfs_path *path;
4664 struct btrfs_key key;
4665 struct inode *inode = NULL;
4673 struct btrfs_file_extent_item *ei;
4674 struct extent_buffer *leaf;
4675 unsigned long nr = 0;
4677 name_len = strlen(symname) + 1;
4678 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
4679 return -ENAMETOOLONG;
4681 err = btrfs_check_free_space(root, 1, 0);
4685 trans = btrfs_start_transaction(root, 1);
4686 btrfs_set_trans_block_group(trans, dir);
4688 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4694 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4696 dentry->d_parent->d_inode->i_ino, objectid,
4697 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
4699 err = PTR_ERR(inode);
4703 err = btrfs_init_acl(inode, dir);
4709 btrfs_set_trans_block_group(trans, inode);
4710 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4714 inode->i_mapping->a_ops = &btrfs_aops;
4715 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4716 inode->i_fop = &btrfs_file_operations;
4717 inode->i_op = &btrfs_file_inode_operations;
4718 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4720 dir->i_sb->s_dirt = 1;
4721 btrfs_update_inode_block_group(trans, inode);
4722 btrfs_update_inode_block_group(trans, dir);
4726 path = btrfs_alloc_path();
4728 key.objectid = inode->i_ino;
4730 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
4731 datasize = btrfs_file_extent_calc_inline_size(name_len);
4732 err = btrfs_insert_empty_item(trans, root, path, &key,
4738 leaf = path->nodes[0];
4739 ei = btrfs_item_ptr(leaf, path->slots[0],
4740 struct btrfs_file_extent_item);
4741 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
4742 btrfs_set_file_extent_type(leaf, ei,
4743 BTRFS_FILE_EXTENT_INLINE);
4744 btrfs_set_file_extent_encryption(leaf, ei, 0);
4745 btrfs_set_file_extent_compression(leaf, ei, 0);
4746 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
4747 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
4749 ptr = btrfs_file_extent_inline_start(ei);
4750 write_extent_buffer(leaf, symname, ptr, name_len);
4751 btrfs_mark_buffer_dirty(leaf);
4752 btrfs_free_path(path);
4754 inode->i_op = &btrfs_symlink_inode_operations;
4755 inode->i_mapping->a_ops = &btrfs_symlink_aops;
4756 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4757 inode_set_bytes(inode, name_len);
4758 btrfs_i_size_write(inode, name_len - 1);
4759 err = btrfs_update_inode(trans, root, inode);
4764 nr = trans->blocks_used;
4765 btrfs_end_transaction_throttle(trans, root);
4768 inode_dec_link_count(inode);
4771 btrfs_btree_balance_dirty(root, nr);
4775 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
4776 u64 alloc_hint, int mode)
4778 struct btrfs_trans_handle *trans;
4779 struct btrfs_root *root = BTRFS_I(inode)->root;
4780 struct btrfs_key ins;
4782 u64 cur_offset = start;
4783 u64 num_bytes = end - start;
4786 trans = btrfs_join_transaction(root, 1);
4788 btrfs_set_trans_block_group(trans, inode);
4790 while (num_bytes > 0) {
4791 alloc_size = min(num_bytes, root->fs_info->max_extent);
4792 ret = btrfs_reserve_extent(trans, root, alloc_size,
4793 root->sectorsize, 0, alloc_hint,
4799 ret = insert_reserved_file_extent(trans, inode,
4800 cur_offset, ins.objectid,
4801 ins.offset, ins.offset,
4802 ins.offset, 0, 0, 0,
4803 BTRFS_FILE_EXTENT_PREALLOC);
4805 num_bytes -= ins.offset;
4806 cur_offset += ins.offset;
4807 alloc_hint = ins.objectid + ins.offset;
4810 if (cur_offset > start) {
4811 inode->i_ctime = CURRENT_TIME;
4812 btrfs_set_flag(inode, PREALLOC);
4813 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
4814 cur_offset > i_size_read(inode))
4815 btrfs_i_size_write(inode, cur_offset);
4816 ret = btrfs_update_inode(trans, root, inode);
4820 btrfs_end_transaction(trans, root);
4824 static long btrfs_fallocate(struct inode *inode, int mode,
4825 loff_t offset, loff_t len)
4832 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
4833 struct extent_map *em;
4836 alloc_start = offset & ~mask;
4837 alloc_end = (offset + len + mask) & ~mask;
4839 mutex_lock(&inode->i_mutex);
4840 if (alloc_start > inode->i_size) {
4841 ret = btrfs_cont_expand(inode, alloc_start);
4847 struct btrfs_ordered_extent *ordered;
4848 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start,
4849 alloc_end - 1, GFP_NOFS);
4850 ordered = btrfs_lookup_first_ordered_extent(inode,
4853 ordered->file_offset + ordered->len > alloc_start &&
4854 ordered->file_offset < alloc_end) {
4855 btrfs_put_ordered_extent(ordered);
4856 unlock_extent(&BTRFS_I(inode)->io_tree,
4857 alloc_start, alloc_end - 1, GFP_NOFS);
4858 btrfs_wait_ordered_range(inode, alloc_start,
4859 alloc_end - alloc_start);
4862 btrfs_put_ordered_extent(ordered);
4867 cur_offset = alloc_start;
4869 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4870 alloc_end - cur_offset, 0);
4871 BUG_ON(IS_ERR(em) || !em);
4872 last_byte = min(extent_map_end(em), alloc_end);
4873 last_byte = (last_byte + mask) & ~mask;
4874 if (em->block_start == EXTENT_MAP_HOLE) {
4875 ret = prealloc_file_range(inode, cur_offset,
4876 last_byte, alloc_hint, mode);
4878 free_extent_map(em);
4882 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
4883 alloc_hint = em->block_start;
4884 free_extent_map(em);
4886 cur_offset = last_byte;
4887 if (cur_offset >= alloc_end) {
4892 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, alloc_end - 1,
4895 mutex_unlock(&inode->i_mutex);
4899 static int btrfs_set_page_dirty(struct page *page)
4901 return __set_page_dirty_nobuffers(page);
4904 static int btrfs_permission(struct inode *inode, int mask)
4906 if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
4908 return generic_permission(inode, mask, btrfs_check_acl);
4911 static struct inode_operations btrfs_dir_inode_operations = {
4912 .getattr = btrfs_getattr,
4913 .lookup = btrfs_lookup,
4914 .create = btrfs_create,
4915 .unlink = btrfs_unlink,
4917 .mkdir = btrfs_mkdir,
4918 .rmdir = btrfs_rmdir,
4919 .rename = btrfs_rename,
4920 .symlink = btrfs_symlink,
4921 .setattr = btrfs_setattr,
4922 .mknod = btrfs_mknod,
4923 .setxattr = btrfs_setxattr,
4924 .getxattr = btrfs_getxattr,
4925 .listxattr = btrfs_listxattr,
4926 .removexattr = btrfs_removexattr,
4927 .permission = btrfs_permission,
4929 static struct inode_operations btrfs_dir_ro_inode_operations = {
4930 .lookup = btrfs_lookup,
4931 .permission = btrfs_permission,
4933 static struct file_operations btrfs_dir_file_operations = {
4934 .llseek = generic_file_llseek,
4935 .read = generic_read_dir,
4936 .readdir = btrfs_real_readdir,
4937 .unlocked_ioctl = btrfs_ioctl,
4938 #ifdef CONFIG_COMPAT
4939 .compat_ioctl = btrfs_ioctl,
4941 .release = btrfs_release_file,
4942 .fsync = btrfs_sync_file,
4945 static struct extent_io_ops btrfs_extent_io_ops = {
4946 .fill_delalloc = run_delalloc_range,
4947 .submit_bio_hook = btrfs_submit_bio_hook,
4948 .merge_bio_hook = btrfs_merge_bio_hook,
4949 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
4950 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
4951 .writepage_start_hook = btrfs_writepage_start_hook,
4952 .readpage_io_failed_hook = btrfs_io_failed_hook,
4953 .set_bit_hook = btrfs_set_bit_hook,
4954 .clear_bit_hook = btrfs_clear_bit_hook,
4957 static struct address_space_operations btrfs_aops = {
4958 .readpage = btrfs_readpage,
4959 .writepage = btrfs_writepage,
4960 .writepages = btrfs_writepages,
4961 .readpages = btrfs_readpages,
4962 .sync_page = block_sync_page,
4964 .direct_IO = btrfs_direct_IO,
4965 .invalidatepage = btrfs_invalidatepage,
4966 .releasepage = btrfs_releasepage,
4967 .set_page_dirty = btrfs_set_page_dirty,
4970 static struct address_space_operations btrfs_symlink_aops = {
4971 .readpage = btrfs_readpage,
4972 .writepage = btrfs_writepage,
4973 .invalidatepage = btrfs_invalidatepage,
4974 .releasepage = btrfs_releasepage,
4977 static struct inode_operations btrfs_file_inode_operations = {
4978 .truncate = btrfs_truncate,
4979 .getattr = btrfs_getattr,
4980 .setattr = btrfs_setattr,
4981 .setxattr = btrfs_setxattr,
4982 .getxattr = btrfs_getxattr,
4983 .listxattr = btrfs_listxattr,
4984 .removexattr = btrfs_removexattr,
4985 .permission = btrfs_permission,
4986 .fallocate = btrfs_fallocate,
4988 static struct inode_operations btrfs_special_inode_operations = {
4989 .getattr = btrfs_getattr,
4990 .setattr = btrfs_setattr,
4991 .permission = btrfs_permission,
4992 .setxattr = btrfs_setxattr,
4993 .getxattr = btrfs_getxattr,
4994 .listxattr = btrfs_listxattr,
4995 .removexattr = btrfs_removexattr,
4997 static struct inode_operations btrfs_symlink_inode_operations = {
4998 .readlink = generic_readlink,
4999 .follow_link = page_follow_link_light,
5000 .put_link = page_put_link,
5001 .permission = btrfs_permission,