2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const 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_path_cachep;
74 struct kmem_cache *btrfs_free_space_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 int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 u64 start, size_t size, size_t compressed_size,
115 struct page **compressed_pages)
117 struct btrfs_key key;
118 struct btrfs_path *path;
119 struct extent_buffer *leaf;
120 struct page *page = NULL;
123 struct btrfs_file_extent_item *ei;
126 size_t cur_size = size;
128 unsigned long offset;
130 if (compressed_size && compressed_pages)
131 cur_size = compressed_size;
133 path = btrfs_alloc_path();
137 path->leave_spinning = 1;
138 btrfs_set_trans_block_group(trans, inode);
140 key.objectid = inode->i_ino;
142 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143 datasize = btrfs_file_extent_calc_inline_size(cur_size);
145 inode_add_bytes(inode, size);
146 ret = btrfs_insert_empty_item(trans, root, path, &key,
153 leaf = path->nodes[0];
154 ei = btrfs_item_ptr(leaf, path->slots[0],
155 struct btrfs_file_extent_item);
156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158 btrfs_set_file_extent_encryption(leaf, ei, 0);
159 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161 ptr = btrfs_file_extent_inline_start(ei);
163 if (compress_type != BTRFS_COMPRESS_NONE) {
166 while (compressed_size > 0) {
167 cpage = compressed_pages[i];
168 cur_size = min_t(unsigned long, compressed_size,
171 kaddr = kmap_atomic(cpage, KM_USER0);
172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
173 kunmap_atomic(kaddr, KM_USER0);
177 compressed_size -= cur_size;
179 btrfs_set_file_extent_compression(leaf, ei,
182 page = find_get_page(inode->i_mapping,
183 start >> PAGE_CACHE_SHIFT);
184 btrfs_set_file_extent_compression(leaf, ei, 0);
185 kaddr = kmap_atomic(page, KM_USER0);
186 offset = start & (PAGE_CACHE_SIZE - 1);
187 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188 kunmap_atomic(kaddr, KM_USER0);
189 page_cache_release(page);
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_free_path(path);
195 * we're an inline extent, so nobody can
196 * extend the file past i_size without locking
197 * a page we already have locked.
199 * We must do any isize and inode updates
200 * before we unlock the pages. Otherwise we
201 * could end up racing with unlink.
203 BTRFS_I(inode)->disk_i_size = inode->i_size;
204 btrfs_update_inode(trans, root, inode);
208 btrfs_free_path(path);
214 * conditionally insert an inline extent into the file. This
215 * does the checks required to make sure the data is small enough
216 * to fit as an inline extent.
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219 struct btrfs_root *root,
220 struct inode *inode, u64 start, u64 end,
221 size_t compressed_size, int compress_type,
222 struct page **compressed_pages)
224 u64 isize = i_size_read(inode);
225 u64 actual_end = min(end + 1, isize);
226 u64 inline_len = actual_end - start;
227 u64 aligned_end = (end + root->sectorsize - 1) &
228 ~((u64)root->sectorsize - 1);
230 u64 data_len = inline_len;
234 data_len = compressed_size;
237 actual_end >= PAGE_CACHE_SIZE ||
238 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240 (actual_end & (root->sectorsize - 1)) == 0) ||
242 data_len > root->fs_info->max_inline) {
246 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
250 if (isize > actual_end)
251 inline_len = min_t(u64, isize, actual_end);
252 ret = insert_inline_extent(trans, root, inode, start,
253 inline_len, compressed_size,
254 compress_type, compressed_pages);
256 btrfs_delalloc_release_metadata(inode, end + 1 - start);
257 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
261 struct async_extent {
266 unsigned long nr_pages;
268 struct list_head list;
273 struct btrfs_root *root;
274 struct page *locked_page;
277 struct list_head extents;
278 struct btrfs_work work;
281 static noinline int add_async_extent(struct async_cow *cow,
282 u64 start, u64 ram_size,
285 unsigned long nr_pages,
288 struct async_extent *async_extent;
290 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291 BUG_ON(!async_extent);
292 async_extent->start = start;
293 async_extent->ram_size = ram_size;
294 async_extent->compressed_size = compressed_size;
295 async_extent->pages = pages;
296 async_extent->nr_pages = nr_pages;
297 async_extent->compress_type = compress_type;
298 list_add_tail(&async_extent->list, &cow->extents);
303 * we create compressed extents in two phases. The first
304 * phase compresses a range of pages that have already been
305 * locked (both pages and state bits are locked).
307 * This is done inside an ordered work queue, and the compression
308 * is spread across many cpus. The actual IO submission is step
309 * two, and the ordered work queue takes care of making sure that
310 * happens in the same order things were put onto the queue by
311 * writepages and friends.
313 * If this code finds it can't get good compression, it puts an
314 * entry onto the work queue to write the uncompressed bytes. This
315 * makes sure that both compressed inodes and uncompressed inodes
316 * are written in the same order that pdflush sent them down.
318 static noinline int compress_file_range(struct inode *inode,
319 struct page *locked_page,
321 struct async_cow *async_cow,
324 struct btrfs_root *root = BTRFS_I(inode)->root;
325 struct btrfs_trans_handle *trans;
327 u64 blocksize = root->sectorsize;
329 u64 isize = i_size_read(inode);
331 struct page **pages = NULL;
332 unsigned long nr_pages;
333 unsigned long nr_pages_ret = 0;
334 unsigned long total_compressed = 0;
335 unsigned long total_in = 0;
336 unsigned long max_compressed = 128 * 1024;
337 unsigned long max_uncompressed = 128 * 1024;
340 int compress_type = root->fs_info->compress_type;
342 actual_end = min_t(u64, isize, end + 1);
345 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
346 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
349 * we don't want to send crud past the end of i_size through
350 * compression, that's just a waste of CPU time. So, if the
351 * end of the file is before the start of our current
352 * requested range of bytes, we bail out to the uncompressed
353 * cleanup code that can deal with all of this.
355 * It isn't really the fastest way to fix things, but this is a
356 * very uncommon corner.
358 if (actual_end <= start)
359 goto cleanup_and_bail_uncompressed;
361 total_compressed = actual_end - start;
363 /* we want to make sure that amount of ram required to uncompress
364 * an extent is reasonable, so we limit the total size in ram
365 * of a compressed extent to 128k. This is a crucial number
366 * because it also controls how easily we can spread reads across
367 * cpus for decompression.
369 * We also want to make sure the amount of IO required to do
370 * a random read is reasonably small, so we limit the size of
371 * a compressed extent to 128k.
373 total_compressed = min(total_compressed, max_uncompressed);
374 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
375 num_bytes = max(blocksize, num_bytes);
380 * we do compression for mount -o compress and when the
381 * inode has not been flagged as nocompress. This flag can
382 * change at any time if we discover bad compression ratios.
384 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
385 (btrfs_test_opt(root, COMPRESS) ||
386 (BTRFS_I(inode)->force_compress) ||
387 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
389 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
392 if (BTRFS_I(inode)->force_compress)
393 compress_type = BTRFS_I(inode)->force_compress;
395 ret = btrfs_compress_pages(compress_type,
396 inode->i_mapping, start,
397 total_compressed, pages,
398 nr_pages, &nr_pages_ret,
404 unsigned long offset = total_compressed &
405 (PAGE_CACHE_SIZE - 1);
406 struct page *page = pages[nr_pages_ret - 1];
409 /* zero the tail end of the last page, we might be
410 * sending it down to disk
413 kaddr = kmap_atomic(page, KM_USER0);
414 memset(kaddr + offset, 0,
415 PAGE_CACHE_SIZE - offset);
416 kunmap_atomic(kaddr, KM_USER0);
422 trans = btrfs_join_transaction(root, 1);
423 BUG_ON(IS_ERR(trans));
424 btrfs_set_trans_block_group(trans, inode);
425 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427 /* lets try to make an inline extent */
428 if (ret || total_in < (actual_end - start)) {
429 /* we didn't compress the entire range, try
430 * to make an uncompressed inline extent.
432 ret = cow_file_range_inline(trans, root, inode,
433 start, end, 0, 0, NULL);
435 /* try making a compressed inline extent */
436 ret = cow_file_range_inline(trans, root, inode,
439 compress_type, pages);
443 * inline extent creation worked, we don't need
444 * to create any more async work items. Unlock
445 * and free up our temp pages.
447 extent_clear_unlock_delalloc(inode,
448 &BTRFS_I(inode)->io_tree,
450 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
451 EXTENT_CLEAR_DELALLOC |
452 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
454 btrfs_end_transaction(trans, root);
457 btrfs_end_transaction(trans, root);
462 * we aren't doing an inline extent round the compressed size
463 * up to a block size boundary so the allocator does sane
466 total_compressed = (total_compressed + blocksize - 1) &
470 * one last check to make sure the compression is really a
471 * win, compare the page count read with the blocks on disk
473 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
474 ~(PAGE_CACHE_SIZE - 1);
475 if (total_compressed >= total_in) {
478 num_bytes = total_in;
481 if (!will_compress && pages) {
483 * the compression code ran but failed to make things smaller,
484 * free any pages it allocated and our page pointer array
486 for (i = 0; i < nr_pages_ret; i++) {
487 WARN_ON(pages[i]->mapping);
488 page_cache_release(pages[i]);
492 total_compressed = 0;
495 /* flag the file so we don't compress in the future */
496 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
497 !(BTRFS_I(inode)->force_compress)) {
498 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
504 /* the async work queues will take care of doing actual
505 * allocation on disk for these compressed pages,
506 * and will submit them to the elevator.
508 add_async_extent(async_cow, start, num_bytes,
509 total_compressed, pages, nr_pages_ret,
512 if (start + num_bytes < end) {
519 cleanup_and_bail_uncompressed:
521 * No compression, but we still need to write the pages in
522 * the file we've been given so far. redirty the locked
523 * page if it corresponds to our extent and set things up
524 * for the async work queue to run cow_file_range to do
525 * the normal delalloc dance
527 if (page_offset(locked_page) >= start &&
528 page_offset(locked_page) <= end) {
529 __set_page_dirty_nobuffers(locked_page);
530 /* unlocked later on in the async handlers */
532 add_async_extent(async_cow, start, end - start + 1,
533 0, NULL, 0, BTRFS_COMPRESS_NONE);
541 for (i = 0; i < nr_pages_ret; i++) {
542 WARN_ON(pages[i]->mapping);
543 page_cache_release(pages[i]);
551 * phase two of compressed writeback. This is the ordered portion
552 * of the code, which only gets called in the order the work was
553 * queued. We walk all the async extents created by compress_file_range
554 * and send them down to the disk.
556 static noinline int submit_compressed_extents(struct inode *inode,
557 struct async_cow *async_cow)
559 struct async_extent *async_extent;
561 struct btrfs_trans_handle *trans;
562 struct btrfs_key ins;
563 struct extent_map *em;
564 struct btrfs_root *root = BTRFS_I(inode)->root;
565 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
566 struct extent_io_tree *io_tree;
569 if (list_empty(&async_cow->extents))
573 while (!list_empty(&async_cow->extents)) {
574 async_extent = list_entry(async_cow->extents.next,
575 struct async_extent, list);
576 list_del(&async_extent->list);
578 io_tree = &BTRFS_I(inode)->io_tree;
581 /* did the compression code fall back to uncompressed IO? */
582 if (!async_extent->pages) {
583 int page_started = 0;
584 unsigned long nr_written = 0;
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start +
588 async_extent->ram_size - 1, GFP_NOFS);
590 /* allocate blocks */
591 ret = cow_file_range(inode, async_cow->locked_page,
593 async_extent->start +
594 async_extent->ram_size - 1,
595 &page_started, &nr_written, 0);
598 * if page_started, cow_file_range inserted an
599 * inline extent and took care of all the unlocking
600 * and IO for us. Otherwise, we need to submit
601 * all those pages down to the drive.
603 if (!page_started && !ret)
604 extent_write_locked_range(io_tree,
605 inode, async_extent->start,
606 async_extent->start +
607 async_extent->ram_size - 1,
615 lock_extent(io_tree, async_extent->start,
616 async_extent->start + async_extent->ram_size - 1,
619 trans = btrfs_join_transaction(root, 1);
620 BUG_ON(IS_ERR(trans));
621 ret = btrfs_reserve_extent(trans, root,
622 async_extent->compressed_size,
623 async_extent->compressed_size,
626 btrfs_end_transaction(trans, root);
630 for (i = 0; i < async_extent->nr_pages; i++) {
631 WARN_ON(async_extent->pages[i]->mapping);
632 page_cache_release(async_extent->pages[i]);
634 kfree(async_extent->pages);
635 async_extent->nr_pages = 0;
636 async_extent->pages = NULL;
637 unlock_extent(io_tree, async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1, GFP_NOFS);
644 * here we're doing allocation and writeback of the
647 btrfs_drop_extent_cache(inode, async_extent->start,
648 async_extent->start +
649 async_extent->ram_size - 1, 0);
651 em = alloc_extent_map(GFP_NOFS);
653 em->start = async_extent->start;
654 em->len = async_extent->ram_size;
655 em->orig_start = em->start;
657 em->block_start = ins.objectid;
658 em->block_len = ins.offset;
659 em->bdev = root->fs_info->fs_devices->latest_bdev;
660 em->compress_type = async_extent->compress_type;
661 set_bit(EXTENT_FLAG_PINNED, &em->flags);
662 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
665 write_lock(&em_tree->lock);
666 ret = add_extent_mapping(em_tree, em);
667 write_unlock(&em_tree->lock);
668 if (ret != -EEXIST) {
672 btrfs_drop_extent_cache(inode, async_extent->start,
673 async_extent->start +
674 async_extent->ram_size - 1, 0);
677 ret = btrfs_add_ordered_extent_compress(inode,
680 async_extent->ram_size,
682 BTRFS_ORDERED_COMPRESSED,
683 async_extent->compress_type);
687 * clear dirty, set writeback and unlock the pages.
689 extent_clear_unlock_delalloc(inode,
690 &BTRFS_I(inode)->io_tree,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
695 EXTENT_CLEAR_UNLOCK |
696 EXTENT_CLEAR_DELALLOC |
697 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
699 ret = btrfs_submit_compressed_write(inode,
701 async_extent->ram_size,
703 ins.offset, async_extent->pages,
704 async_extent->nr_pages);
707 alloc_hint = ins.objectid + ins.offset;
715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
718 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719 struct extent_map *em;
722 read_lock(&em_tree->lock);
723 em = search_extent_mapping(em_tree, start, num_bytes);
726 * if block start isn't an actual block number then find the
727 * first block in this inode and use that as a hint. If that
728 * block is also bogus then just don't worry about it.
730 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
732 em = search_extent_mapping(em_tree, 0, 0);
733 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
734 alloc_hint = em->block_start;
738 alloc_hint = em->block_start;
742 read_unlock(&em_tree->lock);
748 * when extent_io.c finds a delayed allocation range in the file,
749 * the call backs end up in this code. The basic idea is to
750 * allocate extents on disk for the range, and create ordered data structs
751 * in ram to track those extents.
753 * locked_page is the page that writepage had locked already. We use
754 * it to make sure we don't do extra locks or unlocks.
756 * *page_started is set to one if we unlock locked_page and do everything
757 * required to start IO on it. It may be clean and already done with
760 static noinline int cow_file_range(struct inode *inode,
761 struct page *locked_page,
762 u64 start, u64 end, int *page_started,
763 unsigned long *nr_written,
766 struct btrfs_root *root = BTRFS_I(inode)->root;
767 struct btrfs_trans_handle *trans;
770 unsigned long ram_size;
773 u64 blocksize = root->sectorsize;
774 struct btrfs_key ins;
775 struct extent_map *em;
776 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
779 BUG_ON(root == root->fs_info->tree_root);
780 trans = btrfs_join_transaction(root, 1);
781 BUG_ON(IS_ERR(trans));
782 btrfs_set_trans_block_group(trans, inode);
783 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
785 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
786 num_bytes = max(blocksize, num_bytes);
787 disk_num_bytes = num_bytes;
791 /* lets try to make an inline extent */
792 ret = cow_file_range_inline(trans, root, inode,
793 start, end, 0, 0, NULL);
795 extent_clear_unlock_delalloc(inode,
796 &BTRFS_I(inode)->io_tree,
798 EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
805 *nr_written = *nr_written +
806 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
813 BUG_ON(disk_num_bytes >
814 btrfs_super_total_bytes(&root->fs_info->super_copy));
816 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
817 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
819 while (disk_num_bytes > 0) {
822 cur_alloc_size = disk_num_bytes;
823 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
824 root->sectorsize, 0, alloc_hint,
828 em = alloc_extent_map(GFP_NOFS);
831 em->orig_start = em->start;
832 ram_size = ins.offset;
833 em->len = ins.offset;
835 em->block_start = ins.objectid;
836 em->block_len = ins.offset;
837 em->bdev = root->fs_info->fs_devices->latest_bdev;
838 set_bit(EXTENT_FLAG_PINNED, &em->flags);
841 write_lock(&em_tree->lock);
842 ret = add_extent_mapping(em_tree, em);
843 write_unlock(&em_tree->lock);
844 if (ret != -EEXIST) {
848 btrfs_drop_extent_cache(inode, start,
849 start + ram_size - 1, 0);
852 cur_alloc_size = ins.offset;
853 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
854 ram_size, cur_alloc_size, 0);
857 if (root->root_key.objectid ==
858 BTRFS_DATA_RELOC_TREE_OBJECTID) {
859 ret = btrfs_reloc_clone_csums(inode, start,
864 if (disk_num_bytes < cur_alloc_size)
867 /* we're not doing compressed IO, don't unlock the first
868 * page (which the caller expects to stay locked), don't
869 * clear any dirty bits and don't set any writeback bits
871 * Do set the Private2 bit so we know this page was properly
872 * setup for writepage
874 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
875 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
878 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
879 start, start + ram_size - 1,
881 disk_num_bytes -= cur_alloc_size;
882 num_bytes -= cur_alloc_size;
883 alloc_hint = ins.objectid + ins.offset;
884 start += cur_alloc_size;
888 btrfs_end_transaction(trans, root);
894 * work queue call back to started compression on a file and pages
896 static noinline void async_cow_start(struct btrfs_work *work)
898 struct async_cow *async_cow;
900 async_cow = container_of(work, struct async_cow, work);
902 compress_file_range(async_cow->inode, async_cow->locked_page,
903 async_cow->start, async_cow->end, async_cow,
906 async_cow->inode = NULL;
910 * work queue call back to submit previously compressed pages
912 static noinline void async_cow_submit(struct btrfs_work *work)
914 struct async_cow *async_cow;
915 struct btrfs_root *root;
916 unsigned long nr_pages;
918 async_cow = container_of(work, struct async_cow, work);
920 root = async_cow->root;
921 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
924 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
926 if (atomic_read(&root->fs_info->async_delalloc_pages) <
928 waitqueue_active(&root->fs_info->async_submit_wait))
929 wake_up(&root->fs_info->async_submit_wait);
931 if (async_cow->inode)
932 submit_compressed_extents(async_cow->inode, async_cow);
935 static noinline void async_cow_free(struct btrfs_work *work)
937 struct async_cow *async_cow;
938 async_cow = container_of(work, struct async_cow, work);
942 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
943 u64 start, u64 end, int *page_started,
944 unsigned long *nr_written)
946 struct async_cow *async_cow;
947 struct btrfs_root *root = BTRFS_I(inode)->root;
948 unsigned long nr_pages;
950 int limit = 10 * 1024 * 1042;
952 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
953 1, 0, NULL, GFP_NOFS);
954 while (start < end) {
955 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
956 async_cow->inode = inode;
957 async_cow->root = root;
958 async_cow->locked_page = locked_page;
959 async_cow->start = start;
961 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
964 cur_end = min(end, start + 512 * 1024 - 1);
966 async_cow->end = cur_end;
967 INIT_LIST_HEAD(&async_cow->extents);
969 async_cow->work.func = async_cow_start;
970 async_cow->work.ordered_func = async_cow_submit;
971 async_cow->work.ordered_free = async_cow_free;
972 async_cow->work.flags = 0;
974 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
976 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
978 btrfs_queue_worker(&root->fs_info->delalloc_workers,
981 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
982 wait_event(root->fs_info->async_submit_wait,
983 (atomic_read(&root->fs_info->async_delalloc_pages) <
987 while (atomic_read(&root->fs_info->async_submit_draining) &&
988 atomic_read(&root->fs_info->async_delalloc_pages)) {
989 wait_event(root->fs_info->async_submit_wait,
990 (atomic_read(&root->fs_info->async_delalloc_pages) ==
994 *nr_written += nr_pages;
1001 static noinline int csum_exist_in_range(struct btrfs_root *root,
1002 u64 bytenr, u64 num_bytes)
1005 struct btrfs_ordered_sum *sums;
1008 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1009 bytenr + num_bytes - 1, &list);
1010 if (ret == 0 && list_empty(&list))
1013 while (!list_empty(&list)) {
1014 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1015 list_del(&sums->list);
1022 * when nowcow writeback call back. This checks for snapshots or COW copies
1023 * of the extents that exist in the file, and COWs the file as required.
1025 * If no cow copies or snapshots exist, we write directly to the existing
1028 static noinline int run_delalloc_nocow(struct inode *inode,
1029 struct page *locked_page,
1030 u64 start, u64 end, int *page_started, int force,
1031 unsigned long *nr_written)
1033 struct btrfs_root *root = BTRFS_I(inode)->root;
1034 struct btrfs_trans_handle *trans;
1035 struct extent_buffer *leaf;
1036 struct btrfs_path *path;
1037 struct btrfs_file_extent_item *fi;
1038 struct btrfs_key found_key;
1050 bool nolock = false;
1052 path = btrfs_alloc_path();
1054 if (root == root->fs_info->tree_root) {
1056 trans = btrfs_join_transaction_nolock(root, 1);
1058 trans = btrfs_join_transaction(root, 1);
1060 BUG_ON(IS_ERR(trans));
1062 cow_start = (u64)-1;
1065 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1069 leaf = path->nodes[0];
1070 btrfs_item_key_to_cpu(leaf, &found_key,
1071 path->slots[0] - 1);
1072 if (found_key.objectid == inode->i_ino &&
1073 found_key.type == BTRFS_EXTENT_DATA_KEY)
1078 leaf = path->nodes[0];
1079 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1080 ret = btrfs_next_leaf(root, path);
1085 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093 if (found_key.objectid > inode->i_ino ||
1094 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1095 found_key.offset > end)
1098 if (found_key.offset > cur_offset) {
1099 extent_end = found_key.offset;
1104 fi = btrfs_item_ptr(leaf, path->slots[0],
1105 struct btrfs_file_extent_item);
1106 extent_type = btrfs_file_extent_type(leaf, fi);
1108 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1109 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1110 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1111 extent_offset = btrfs_file_extent_offset(leaf, fi);
1112 extent_end = found_key.offset +
1113 btrfs_file_extent_num_bytes(leaf, fi);
1114 if (extent_end <= start) {
1118 if (disk_bytenr == 0)
1120 if (btrfs_file_extent_compression(leaf, fi) ||
1121 btrfs_file_extent_encryption(leaf, fi) ||
1122 btrfs_file_extent_other_encoding(leaf, fi))
1124 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126 if (btrfs_extent_readonly(root, disk_bytenr))
1128 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130 extent_offset, disk_bytenr))
1132 disk_bytenr += extent_offset;
1133 disk_bytenr += cur_offset - found_key.offset;
1134 num_bytes = min(end + 1, extent_end) - cur_offset;
1136 * force cow if csum exists in the range.
1137 * this ensure that csum for a given extent are
1138 * either valid or do not exist.
1140 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1144 extent_end = found_key.offset +
1145 btrfs_file_extent_inline_len(leaf, fi);
1146 extent_end = ALIGN(extent_end, root->sectorsize);
1151 if (extent_end <= start) {
1156 if (cow_start == (u64)-1)
1157 cow_start = cur_offset;
1158 cur_offset = extent_end;
1159 if (cur_offset > end)
1165 btrfs_release_path(root, path);
1166 if (cow_start != (u64)-1) {
1167 ret = cow_file_range(inode, locked_page, cow_start,
1168 found_key.offset - 1, page_started,
1171 cow_start = (u64)-1;
1174 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1175 struct extent_map *em;
1176 struct extent_map_tree *em_tree;
1177 em_tree = &BTRFS_I(inode)->extent_tree;
1178 em = alloc_extent_map(GFP_NOFS);
1180 em->start = cur_offset;
1181 em->orig_start = em->start;
1182 em->len = num_bytes;
1183 em->block_len = num_bytes;
1184 em->block_start = disk_bytenr;
1185 em->bdev = root->fs_info->fs_devices->latest_bdev;
1186 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188 write_lock(&em_tree->lock);
1189 ret = add_extent_mapping(em_tree, em);
1190 write_unlock(&em_tree->lock);
1191 if (ret != -EEXIST) {
1192 free_extent_map(em);
1195 btrfs_drop_extent_cache(inode, em->start,
1196 em->start + em->len - 1, 0);
1198 type = BTRFS_ORDERED_PREALLOC;
1200 type = BTRFS_ORDERED_NOCOW;
1203 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1204 num_bytes, num_bytes, type);
1207 if (root->root_key.objectid ==
1208 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1209 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1214 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1215 cur_offset, cur_offset + num_bytes - 1,
1216 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1217 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1218 EXTENT_SET_PRIVATE2);
1219 cur_offset = extent_end;
1220 if (cur_offset > end)
1223 btrfs_release_path(root, path);
1225 if (cur_offset <= end && cow_start == (u64)-1)
1226 cow_start = cur_offset;
1227 if (cow_start != (u64)-1) {
1228 ret = cow_file_range(inode, locked_page, cow_start, end,
1229 page_started, nr_written, 1);
1234 ret = btrfs_end_transaction_nolock(trans, root);
1237 ret = btrfs_end_transaction(trans, root);
1240 btrfs_free_path(path);
1245 * extent_io.c call back to do delayed allocation processing
1247 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1248 u64 start, u64 end, int *page_started,
1249 unsigned long *nr_written)
1252 struct btrfs_root *root = BTRFS_I(inode)->root;
1254 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1255 ret = run_delalloc_nocow(inode, locked_page, start, end,
1256 page_started, 1, nr_written);
1257 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1258 ret = run_delalloc_nocow(inode, locked_page, start, end,
1259 page_started, 0, nr_written);
1260 else if (!btrfs_test_opt(root, COMPRESS) &&
1261 !(BTRFS_I(inode)->force_compress) &&
1262 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1263 ret = cow_file_range(inode, locked_page, start, end,
1264 page_started, nr_written, 1);
1266 ret = cow_file_range_async(inode, locked_page, start, end,
1267 page_started, nr_written);
1271 static int btrfs_split_extent_hook(struct inode *inode,
1272 struct extent_state *orig, u64 split)
1274 /* not delalloc, ignore it */
1275 if (!(orig->state & EXTENT_DELALLOC))
1278 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1283 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1284 * extents so we can keep track of new extents that are just merged onto old
1285 * extents, such as when we are doing sequential writes, so we can properly
1286 * account for the metadata space we'll need.
1288 static int btrfs_merge_extent_hook(struct inode *inode,
1289 struct extent_state *new,
1290 struct extent_state *other)
1292 /* not delalloc, ignore it */
1293 if (!(other->state & EXTENT_DELALLOC))
1296 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1301 * extent_io.c set_bit_hook, used to track delayed allocation
1302 * bytes in this file, and to maintain the list of inodes that
1303 * have pending delalloc work to be done.
1305 static int btrfs_set_bit_hook(struct inode *inode,
1306 struct extent_state *state, int *bits)
1310 * set_bit and clear bit hooks normally require _irqsave/restore
1311 * but in this case, we are only testeing for the DELALLOC
1312 * bit, which is only set or cleared with irqs on
1314 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1315 struct btrfs_root *root = BTRFS_I(inode)->root;
1316 u64 len = state->end + 1 - state->start;
1317 int do_list = (root->root_key.objectid !=
1318 BTRFS_ROOT_TREE_OBJECTID);
1320 if (*bits & EXTENT_FIRST_DELALLOC)
1321 *bits &= ~EXTENT_FIRST_DELALLOC;
1323 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325 spin_lock(&root->fs_info->delalloc_lock);
1326 BTRFS_I(inode)->delalloc_bytes += len;
1327 root->fs_info->delalloc_bytes += len;
1328 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1329 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1330 &root->fs_info->delalloc_inodes);
1332 spin_unlock(&root->fs_info->delalloc_lock);
1338 * extent_io.c clear_bit_hook, see set_bit_hook for why
1340 static int btrfs_clear_bit_hook(struct inode *inode,
1341 struct extent_state *state, int *bits)
1344 * set_bit and clear bit hooks normally require _irqsave/restore
1345 * but in this case, we are only testeing for the DELALLOC
1346 * bit, which is only set or cleared with irqs on
1348 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1349 struct btrfs_root *root = BTRFS_I(inode)->root;
1350 u64 len = state->end + 1 - state->start;
1351 int do_list = (root->root_key.objectid !=
1352 BTRFS_ROOT_TREE_OBJECTID);
1354 if (*bits & EXTENT_FIRST_DELALLOC)
1355 *bits &= ~EXTENT_FIRST_DELALLOC;
1356 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1357 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359 if (*bits & EXTENT_DO_ACCOUNTING)
1360 btrfs_delalloc_release_metadata(inode, len);
1362 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364 btrfs_free_reserved_data_space(inode, len);
1366 spin_lock(&root->fs_info->delalloc_lock);
1367 root->fs_info->delalloc_bytes -= len;
1368 BTRFS_I(inode)->delalloc_bytes -= len;
1370 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1371 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1372 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374 spin_unlock(&root->fs_info->delalloc_lock);
1380 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1381 * we don't create bios that span stripes or chunks
1383 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1384 size_t size, struct bio *bio,
1385 unsigned long bio_flags)
1387 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1388 struct btrfs_mapping_tree *map_tree;
1389 u64 logical = (u64)bio->bi_sector << 9;
1394 if (bio_flags & EXTENT_BIO_COMPRESSED)
1397 length = bio->bi_size;
1398 map_tree = &root->fs_info->mapping_tree;
1399 map_length = length;
1400 ret = btrfs_map_block(map_tree, READ, logical,
1401 &map_length, NULL, 0);
1403 if (map_length < length + size)
1409 * in order to insert checksums into the metadata in large chunks,
1410 * we wait until bio submission time. All the pages in the bio are
1411 * checksummed and sums are attached onto the ordered extent record.
1413 * At IO completion time the cums attached on the ordered extent record
1414 * are inserted into the btree
1416 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1417 struct bio *bio, int mirror_num,
1418 unsigned long bio_flags,
1421 struct btrfs_root *root = BTRFS_I(inode)->root;
1424 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1430 * in order to insert checksums into the metadata in large chunks,
1431 * we wait until bio submission time. All the pages in the bio are
1432 * checksummed and sums are attached onto the ordered extent record.
1434 * At IO completion time the cums attached on the ordered extent record
1435 * are inserted into the btree
1437 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1438 int mirror_num, unsigned long bio_flags,
1441 struct btrfs_root *root = BTRFS_I(inode)->root;
1442 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1446 * extent_io.c submission hook. This does the right thing for csum calculation
1447 * on write, or reading the csums from the tree before a read
1449 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1450 int mirror_num, unsigned long bio_flags,
1453 struct btrfs_root *root = BTRFS_I(inode)->root;
1457 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459 if (root == root->fs_info->tree_root)
1460 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465 if (!(rw & REQ_WRITE)) {
1466 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1467 return btrfs_submit_compressed_read(inode, bio,
1468 mirror_num, bio_flags);
1469 } else if (!skip_sum) {
1470 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1475 } else if (!skip_sum) {
1476 /* csum items have already been cloned */
1477 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479 /* we're doing a write, do the async checksumming */
1480 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1481 inode, rw, bio, mirror_num,
1482 bio_flags, bio_offset,
1483 __btrfs_submit_bio_start,
1484 __btrfs_submit_bio_done);
1488 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1492 * given a list of ordered sums record them in the inode. This happens
1493 * at IO completion time based on sums calculated at bio submission time.
1495 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1496 struct inode *inode, u64 file_offset,
1497 struct list_head *list)
1499 struct btrfs_ordered_sum *sum;
1501 btrfs_set_trans_block_group(trans, inode);
1503 list_for_each_entry(sum, list, list) {
1504 btrfs_csum_file_blocks(trans,
1505 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1510 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1511 struct extent_state **cached_state)
1513 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1516 cached_state, GFP_NOFS);
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup {
1522 struct btrfs_work work;
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 struct btrfs_writepage_fixup *fixup;
1528 struct btrfs_ordered_extent *ordered;
1529 struct extent_state *cached_state = NULL;
1531 struct inode *inode;
1535 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1539 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1540 ClearPageChecked(page);
1544 inode = page->mapping->host;
1545 page_start = page_offset(page);
1546 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1549 &cached_state, GFP_NOFS);
1551 /* already ordered? We're done */
1552 if (PagePrivate2(page))
1555 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1558 page_end, &cached_state, GFP_NOFS);
1560 btrfs_start_ordered_extent(inode, ordered, 1);
1565 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1566 ClearPageChecked(page);
1568 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1569 &cached_state, GFP_NOFS);
1572 page_cache_release(page);
1577 * There are a few paths in the higher layers of the kernel that directly
1578 * set the page dirty bit without asking the filesystem if it is a
1579 * good idea. This causes problems because we want to make sure COW
1580 * properly happens and the data=ordered rules are followed.
1582 * In our case any range that doesn't have the ORDERED bit set
1583 * hasn't been properly setup for IO. We kick off an async process
1584 * to fix it up. The async helper will wait for ordered extents, set
1585 * the delalloc bit and make it safe to write the page.
1587 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 struct inode *inode = page->mapping->host;
1590 struct btrfs_writepage_fixup *fixup;
1591 struct btrfs_root *root = BTRFS_I(inode)->root;
1593 /* this page is properly in the ordered list */
1594 if (TestClearPagePrivate2(page))
1597 if (PageChecked(page))
1600 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1604 SetPageChecked(page);
1605 page_cache_get(page);
1606 fixup->work.func = btrfs_writepage_fixup_worker;
1608 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1612 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1613 struct inode *inode, u64 file_pos,
1614 u64 disk_bytenr, u64 disk_num_bytes,
1615 u64 num_bytes, u64 ram_bytes,
1616 u8 compression, u8 encryption,
1617 u16 other_encoding, int extent_type)
1619 struct btrfs_root *root = BTRFS_I(inode)->root;
1620 struct btrfs_file_extent_item *fi;
1621 struct btrfs_path *path;
1622 struct extent_buffer *leaf;
1623 struct btrfs_key ins;
1627 path = btrfs_alloc_path();
1630 path->leave_spinning = 1;
1633 * we may be replacing one extent in the tree with another.
1634 * The new extent is pinned in the extent map, and we don't want
1635 * to drop it from the cache until it is completely in the btree.
1637 * So, tell btrfs_drop_extents to leave this extent in the cache.
1638 * the caller is expected to unpin it and allow it to be merged
1641 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1645 ins.objectid = inode->i_ino;
1646 ins.offset = file_pos;
1647 ins.type = BTRFS_EXTENT_DATA_KEY;
1648 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650 leaf = path->nodes[0];
1651 fi = btrfs_item_ptr(leaf, path->slots[0],
1652 struct btrfs_file_extent_item);
1653 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1654 btrfs_set_file_extent_type(leaf, fi, extent_type);
1655 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1656 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1657 btrfs_set_file_extent_offset(leaf, fi, 0);
1658 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1659 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1660 btrfs_set_file_extent_compression(leaf, fi, compression);
1661 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1662 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664 btrfs_unlock_up_safe(path, 1);
1665 btrfs_set_lock_blocking(leaf);
1667 btrfs_mark_buffer_dirty(leaf);
1669 inode_add_bytes(inode, num_bytes);
1671 ins.objectid = disk_bytenr;
1672 ins.offset = disk_num_bytes;
1673 ins.type = BTRFS_EXTENT_ITEM_KEY;
1674 ret = btrfs_alloc_reserved_file_extent(trans, root,
1675 root->root_key.objectid,
1676 inode->i_ino, file_pos, &ins);
1678 btrfs_free_path(path);
1684 * helper function for btrfs_finish_ordered_io, this
1685 * just reads in some of the csum leaves to prime them into ram
1686 * before we start the transaction. It limits the amount of btree
1687 * reads required while inside the transaction.
1689 /* as ordered data IO finishes, this gets called so we can finish
1690 * an ordered extent if the range of bytes in the file it covers are
1693 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 struct btrfs_root *root = BTRFS_I(inode)->root;
1696 struct btrfs_trans_handle *trans = NULL;
1697 struct btrfs_ordered_extent *ordered_extent = NULL;
1698 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1699 struct extent_state *cached_state = NULL;
1700 int compress_type = 0;
1702 bool nolock = false;
1704 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1708 BUG_ON(!ordered_extent);
1710 nolock = (root == root->fs_info->tree_root);
1712 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1713 BUG_ON(!list_empty(&ordered_extent->list));
1714 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717 trans = btrfs_join_transaction_nolock(root, 1);
1719 trans = btrfs_join_transaction(root, 1);
1720 BUG_ON(IS_ERR(trans));
1721 btrfs_set_trans_block_group(trans, inode);
1722 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1723 ret = btrfs_update_inode(trans, root, inode);
1729 lock_extent_bits(io_tree, ordered_extent->file_offset,
1730 ordered_extent->file_offset + ordered_extent->len - 1,
1731 0, &cached_state, GFP_NOFS);
1734 trans = btrfs_join_transaction_nolock(root, 1);
1736 trans = btrfs_join_transaction(root, 1);
1737 BUG_ON(IS_ERR(trans));
1738 btrfs_set_trans_block_group(trans, inode);
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1742 compress_type = ordered_extent->compress_type;
1743 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1744 BUG_ON(compress_type);
1745 ret = btrfs_mark_extent_written(trans, inode,
1746 ordered_extent->file_offset,
1747 ordered_extent->file_offset +
1748 ordered_extent->len);
1751 BUG_ON(root == root->fs_info->tree_root);
1752 ret = insert_reserved_file_extent(trans, inode,
1753 ordered_extent->file_offset,
1754 ordered_extent->start,
1755 ordered_extent->disk_len,
1756 ordered_extent->len,
1757 ordered_extent->len,
1758 compress_type, 0, 0,
1759 BTRFS_FILE_EXTENT_REG);
1760 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1761 ordered_extent->file_offset,
1762 ordered_extent->len);
1765 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1766 ordered_extent->file_offset +
1767 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769 add_pending_csums(trans, inode, ordered_extent->file_offset,
1770 &ordered_extent->list);
1772 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1773 ret = btrfs_update_inode(trans, root, inode);
1778 btrfs_end_transaction_nolock(trans, root);
1780 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1782 btrfs_end_transaction(trans, root);
1786 btrfs_put_ordered_extent(ordered_extent);
1787 /* once for the tree */
1788 btrfs_put_ordered_extent(ordered_extent);
1793 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1794 struct extent_state *state, int uptodate)
1796 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1798 ClearPagePrivate2(page);
1799 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1803 * When IO fails, either with EIO or csum verification fails, we
1804 * try other mirrors that might have a good copy of the data. This
1805 * io_failure_record is used to record state as we go through all the
1806 * mirrors. If another mirror has good data, the page is set up to date
1807 * and things continue. If a good mirror can't be found, the original
1808 * bio end_io callback is called to indicate things have failed.
1810 struct io_failure_record {
1815 unsigned long bio_flags;
1819 static int btrfs_io_failed_hook(struct bio *failed_bio,
1820 struct page *page, u64 start, u64 end,
1821 struct extent_state *state)
1823 struct io_failure_record *failrec = NULL;
1825 struct extent_map *em;
1826 struct inode *inode = page->mapping->host;
1827 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1828 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1835 ret = get_state_private(failure_tree, start, &private);
1837 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1840 failrec->start = start;
1841 failrec->len = end - start + 1;
1842 failrec->last_mirror = 0;
1843 failrec->bio_flags = 0;
1845 read_lock(&em_tree->lock);
1846 em = lookup_extent_mapping(em_tree, start, failrec->len);
1847 if (em->start > start || em->start + em->len < start) {
1848 free_extent_map(em);
1851 read_unlock(&em_tree->lock);
1853 if (!em || IS_ERR(em)) {
1857 logical = start - em->start;
1858 logical = em->block_start + logical;
1859 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1860 logical = em->block_start;
1861 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1862 extent_set_compress_type(&failrec->bio_flags,
1865 failrec->logical = logical;
1866 free_extent_map(em);
1867 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1868 EXTENT_DIRTY, GFP_NOFS);
1869 set_state_private(failure_tree, start,
1870 (u64)(unsigned long)failrec);
1872 failrec = (struct io_failure_record *)(unsigned long)private;
1874 num_copies = btrfs_num_copies(
1875 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1876 failrec->logical, failrec->len);
1877 failrec->last_mirror++;
1879 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1880 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1883 if (state && state->start != failrec->start)
1885 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1887 if (!state || failrec->last_mirror > num_copies) {
1888 set_state_private(failure_tree, failrec->start, 0);
1889 clear_extent_bits(failure_tree, failrec->start,
1890 failrec->start + failrec->len - 1,
1891 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1895 bio = bio_alloc(GFP_NOFS, 1);
1896 bio->bi_private = state;
1897 bio->bi_end_io = failed_bio->bi_end_io;
1898 bio->bi_sector = failrec->logical >> 9;
1899 bio->bi_bdev = failed_bio->bi_bdev;
1902 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1903 if (failed_bio->bi_rw & REQ_WRITE)
1908 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1909 failrec->last_mirror,
1910 failrec->bio_flags, 0);
1915 * each time an IO finishes, we do a fast check in the IO failure tree
1916 * to see if we need to process or clean up an io_failure_record
1918 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1921 u64 private_failure;
1922 struct io_failure_record *failure;
1926 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1927 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1928 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1929 start, &private_failure);
1931 failure = (struct io_failure_record *)(unsigned long)
1933 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1935 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1937 failure->start + failure->len - 1,
1938 EXTENT_DIRTY | EXTENT_LOCKED,
1947 * when reads are done, we need to check csums to verify the data is correct
1948 * if there's a match, we allow the bio to finish. If not, we go through
1949 * the io_failure_record routines to find good copies
1951 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1952 struct extent_state *state)
1954 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1955 struct inode *inode = page->mapping->host;
1956 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1958 u64 private = ~(u32)0;
1960 struct btrfs_root *root = BTRFS_I(inode)->root;
1963 if (PageChecked(page)) {
1964 ClearPageChecked(page);
1968 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1971 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1972 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1973 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1978 if (state && state->start == start) {
1979 private = state->private;
1982 ret = get_state_private(io_tree, start, &private);
1984 kaddr = kmap_atomic(page, KM_USER0);
1988 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1989 btrfs_csum_final(csum, (char *)&csum);
1990 if (csum != private)
1993 kunmap_atomic(kaddr, KM_USER0);
1995 /* if the io failure tree for this inode is non-empty,
1996 * check to see if we've recovered from a failed IO
1998 btrfs_clean_io_failures(inode, start);
2002 if (printk_ratelimit()) {
2003 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2004 "private %llu\n", page->mapping->host->i_ino,
2005 (unsigned long long)start, csum,
2006 (unsigned long long)private);
2008 memset(kaddr + offset, 1, end - start + 1);
2009 flush_dcache_page(page);
2010 kunmap_atomic(kaddr, KM_USER0);
2016 struct delayed_iput {
2017 struct list_head list;
2018 struct inode *inode;
2021 void btrfs_add_delayed_iput(struct inode *inode)
2023 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2024 struct delayed_iput *delayed;
2026 if (atomic_add_unless(&inode->i_count, -1, 1))
2029 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2030 delayed->inode = inode;
2032 spin_lock(&fs_info->delayed_iput_lock);
2033 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2034 spin_unlock(&fs_info->delayed_iput_lock);
2037 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2040 struct btrfs_fs_info *fs_info = root->fs_info;
2041 struct delayed_iput *delayed;
2044 spin_lock(&fs_info->delayed_iput_lock);
2045 empty = list_empty(&fs_info->delayed_iputs);
2046 spin_unlock(&fs_info->delayed_iput_lock);
2050 down_read(&root->fs_info->cleanup_work_sem);
2051 spin_lock(&fs_info->delayed_iput_lock);
2052 list_splice_init(&fs_info->delayed_iputs, &list);
2053 spin_unlock(&fs_info->delayed_iput_lock);
2055 while (!list_empty(&list)) {
2056 delayed = list_entry(list.next, struct delayed_iput, list);
2057 list_del(&delayed->list);
2058 iput(delayed->inode);
2061 up_read(&root->fs_info->cleanup_work_sem);
2065 * calculate extra metadata reservation when snapshotting a subvolume
2066 * contains orphan files.
2068 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2069 struct btrfs_pending_snapshot *pending,
2070 u64 *bytes_to_reserve)
2072 struct btrfs_root *root;
2073 struct btrfs_block_rsv *block_rsv;
2077 root = pending->root;
2078 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2081 block_rsv = root->orphan_block_rsv;
2083 /* orphan block reservation for the snapshot */
2084 num_bytes = block_rsv->size;
2087 * after the snapshot is created, COWing tree blocks may use more
2088 * space than it frees. So we should make sure there is enough
2091 index = trans->transid & 0x1;
2092 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2093 num_bytes += block_rsv->size -
2094 (block_rsv->reserved + block_rsv->freed[index]);
2097 *bytes_to_reserve += num_bytes;
2100 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2101 struct btrfs_pending_snapshot *pending)
2103 struct btrfs_root *root = pending->root;
2104 struct btrfs_root *snap = pending->snap;
2105 struct btrfs_block_rsv *block_rsv;
2110 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2113 /* refill source subvolume's orphan block reservation */
2114 block_rsv = root->orphan_block_rsv;
2115 index = trans->transid & 0x1;
2116 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2117 num_bytes = block_rsv->size -
2118 (block_rsv->reserved + block_rsv->freed[index]);
2119 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2120 root->orphan_block_rsv,
2125 /* setup orphan block reservation for the snapshot */
2126 block_rsv = btrfs_alloc_block_rsv(snap);
2129 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2130 snap->orphan_block_rsv = block_rsv;
2132 num_bytes = root->orphan_block_rsv->size;
2133 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2134 block_rsv, num_bytes);
2138 /* insert orphan item for the snapshot */
2139 WARN_ON(!root->orphan_item_inserted);
2140 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2141 snap->root_key.objectid);
2143 snap->orphan_item_inserted = 1;
2147 enum btrfs_orphan_cleanup_state {
2148 ORPHAN_CLEANUP_STARTED = 1,
2149 ORPHAN_CLEANUP_DONE = 2,
2153 * This is called in transaction commmit time. If there are no orphan
2154 * files in the subvolume, it removes orphan item and frees block_rsv
2157 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2158 struct btrfs_root *root)
2162 if (!list_empty(&root->orphan_list) ||
2163 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2166 if (root->orphan_item_inserted &&
2167 btrfs_root_refs(&root->root_item) > 0) {
2168 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2169 root->root_key.objectid);
2171 root->orphan_item_inserted = 0;
2174 if (root->orphan_block_rsv) {
2175 WARN_ON(root->orphan_block_rsv->size > 0);
2176 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2177 root->orphan_block_rsv = NULL;
2182 * This creates an orphan entry for the given inode in case something goes
2183 * wrong in the middle of an unlink/truncate.
2185 * NOTE: caller of this function should reserve 5 units of metadata for
2188 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2190 struct btrfs_root *root = BTRFS_I(inode)->root;
2191 struct btrfs_block_rsv *block_rsv = NULL;
2196 if (!root->orphan_block_rsv) {
2197 block_rsv = btrfs_alloc_block_rsv(root);
2201 spin_lock(&root->orphan_lock);
2202 if (!root->orphan_block_rsv) {
2203 root->orphan_block_rsv = block_rsv;
2204 } else if (block_rsv) {
2205 btrfs_free_block_rsv(root, block_rsv);
2209 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2210 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2213 * For proper ENOSPC handling, we should do orphan
2214 * cleanup when mounting. But this introduces backward
2215 * compatibility issue.
2217 if (!xchg(&root->orphan_item_inserted, 1))
2224 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2227 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2228 BTRFS_I(inode)->orphan_meta_reserved = 1;
2231 spin_unlock(&root->orphan_lock);
2234 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2236 /* grab metadata reservation from transaction handle */
2238 ret = btrfs_orphan_reserve_metadata(trans, inode);
2242 /* insert an orphan item to track this unlinked/truncated file */
2244 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2248 /* insert an orphan item to track subvolume contains orphan files */
2250 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2251 root->root_key.objectid);
2258 * We have done the truncate/delete so we can go ahead and remove the orphan
2259 * item for this particular inode.
2261 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2263 struct btrfs_root *root = BTRFS_I(inode)->root;
2264 int delete_item = 0;
2265 int release_rsv = 0;
2268 spin_lock(&root->orphan_lock);
2269 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2270 list_del_init(&BTRFS_I(inode)->i_orphan);
2274 if (BTRFS_I(inode)->orphan_meta_reserved) {
2275 BTRFS_I(inode)->orphan_meta_reserved = 0;
2278 spin_unlock(&root->orphan_lock);
2280 if (trans && delete_item) {
2281 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2286 btrfs_orphan_release_metadata(inode);
2292 * this cleans up any orphans that may be left on the list from the last use
2295 int btrfs_orphan_cleanup(struct btrfs_root *root)
2297 struct btrfs_path *path;
2298 struct extent_buffer *leaf;
2299 struct btrfs_key key, found_key;
2300 struct btrfs_trans_handle *trans;
2301 struct inode *inode;
2302 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2304 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2307 path = btrfs_alloc_path();
2314 key.objectid = BTRFS_ORPHAN_OBJECTID;
2315 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2316 key.offset = (u64)-1;
2319 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2324 * if ret == 0 means we found what we were searching for, which
2325 * is weird, but possible, so only screw with path if we didnt
2326 * find the key and see if we have stuff that matches
2330 if (path->slots[0] == 0)
2335 /* pull out the item */
2336 leaf = path->nodes[0];
2337 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2339 /* make sure the item matches what we want */
2340 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2342 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2345 /* release the path since we're done with it */
2346 btrfs_release_path(root, path);
2349 * this is where we are basically btrfs_lookup, without the
2350 * crossing root thing. we store the inode number in the
2351 * offset of the orphan item.
2353 found_key.objectid = found_key.offset;
2354 found_key.type = BTRFS_INODE_ITEM_KEY;
2355 found_key.offset = 0;
2356 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2357 if (IS_ERR(inode)) {
2358 ret = PTR_ERR(inode);
2363 * add this inode to the orphan list so btrfs_orphan_del does
2364 * the proper thing when we hit it
2366 spin_lock(&root->orphan_lock);
2367 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2368 spin_unlock(&root->orphan_lock);
2371 * if this is a bad inode, means we actually succeeded in
2372 * removing the inode, but not the orphan record, which means
2373 * we need to manually delete the orphan since iput will just
2374 * do a destroy_inode
2376 if (is_bad_inode(inode)) {
2377 trans = btrfs_start_transaction(root, 0);
2378 if (IS_ERR(trans)) {
2379 ret = PTR_ERR(trans);
2382 btrfs_orphan_del(trans, inode);
2383 btrfs_end_transaction(trans, root);
2388 /* if we have links, this was a truncate, lets do that */
2389 if (inode->i_nlink) {
2390 if (!S_ISREG(inode->i_mode)) {
2396 ret = btrfs_truncate(inode);
2401 /* this will do delete_inode and everything for us */
2406 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2408 if (root->orphan_block_rsv)
2409 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2412 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2413 trans = btrfs_join_transaction(root, 1);
2415 btrfs_end_transaction(trans, root);
2419 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2421 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2425 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2426 btrfs_free_path(path);
2431 * very simple check to peek ahead in the leaf looking for xattrs. If we
2432 * don't find any xattrs, we know there can't be any acls.
2434 * slot is the slot the inode is in, objectid is the objectid of the inode
2436 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2437 int slot, u64 objectid)
2439 u32 nritems = btrfs_header_nritems(leaf);
2440 struct btrfs_key found_key;
2444 while (slot < nritems) {
2445 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2447 /* we found a different objectid, there must not be acls */
2448 if (found_key.objectid != objectid)
2451 /* we found an xattr, assume we've got an acl */
2452 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2456 * we found a key greater than an xattr key, there can't
2457 * be any acls later on
2459 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2466 * it goes inode, inode backrefs, xattrs, extents,
2467 * so if there are a ton of hard links to an inode there can
2468 * be a lot of backrefs. Don't waste time searching too hard,
2469 * this is just an optimization
2474 /* we hit the end of the leaf before we found an xattr or
2475 * something larger than an xattr. We have to assume the inode
2482 * read an inode from the btree into the in-memory inode
2484 static void btrfs_read_locked_inode(struct inode *inode)
2486 struct btrfs_path *path;
2487 struct extent_buffer *leaf;
2488 struct btrfs_inode_item *inode_item;
2489 struct btrfs_timespec *tspec;
2490 struct btrfs_root *root = BTRFS_I(inode)->root;
2491 struct btrfs_key location;
2493 u64 alloc_group_block;
2497 path = btrfs_alloc_path();
2499 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2501 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2505 leaf = path->nodes[0];
2506 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2507 struct btrfs_inode_item);
2509 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2510 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2511 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2512 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2513 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2515 tspec = btrfs_inode_atime(inode_item);
2516 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2517 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2519 tspec = btrfs_inode_mtime(inode_item);
2520 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2521 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523 tspec = btrfs_inode_ctime(inode_item);
2524 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2525 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2527 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2528 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2529 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2530 inode->i_generation = BTRFS_I(inode)->generation;
2532 rdev = btrfs_inode_rdev(leaf, inode_item);
2534 BTRFS_I(inode)->index_cnt = (u64)-1;
2535 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2537 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2538 if (location.objectid == BTRFS_FREE_SPACE_OBJECTID)
2539 inode->i_mapping->flags &= ~__GFP_FS;
2542 * try to precache a NULL acl entry for files that don't have
2543 * any xattrs or acls
2545 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547 cache_no_acl(inode);
2549 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2550 alloc_group_block, 0);
2551 btrfs_free_path(path);
2554 switch (inode->i_mode & S_IFMT) {
2556 inode->i_mapping->a_ops = &btrfs_aops;
2557 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2558 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2559 inode->i_fop = &btrfs_file_operations;
2560 inode->i_op = &btrfs_file_inode_operations;
2563 inode->i_fop = &btrfs_dir_file_operations;
2564 if (root == root->fs_info->tree_root)
2565 inode->i_op = &btrfs_dir_ro_inode_operations;
2567 inode->i_op = &btrfs_dir_inode_operations;
2570 inode->i_op = &btrfs_symlink_inode_operations;
2571 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2572 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2575 inode->i_op = &btrfs_special_inode_operations;
2576 init_special_inode(inode, inode->i_mode, rdev);
2580 btrfs_update_iflags(inode);
2584 btrfs_free_path(path);
2585 make_bad_inode(inode);
2589 * given a leaf and an inode, copy the inode fields into the leaf
2591 static void fill_inode_item(struct btrfs_trans_handle *trans,
2592 struct extent_buffer *leaf,
2593 struct btrfs_inode_item *item,
2594 struct inode *inode)
2596 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2597 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2598 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2599 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2600 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2602 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2603 inode->i_atime.tv_sec);
2604 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2605 inode->i_atime.tv_nsec);
2607 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2608 inode->i_mtime.tv_sec);
2609 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2610 inode->i_mtime.tv_nsec);
2612 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2613 inode->i_ctime.tv_sec);
2614 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2615 inode->i_ctime.tv_nsec);
2617 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2618 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2619 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2620 btrfs_set_inode_transid(leaf, item, trans->transid);
2621 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2622 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2623 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2627 * copy everything in the in-memory inode into the btree.
2629 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2630 struct btrfs_root *root, struct inode *inode)
2632 struct btrfs_inode_item *inode_item;
2633 struct btrfs_path *path;
2634 struct extent_buffer *leaf;
2637 path = btrfs_alloc_path();
2639 path->leave_spinning = 1;
2640 ret = btrfs_lookup_inode(trans, root, path,
2641 &BTRFS_I(inode)->location, 1);
2648 btrfs_unlock_up_safe(path, 1);
2649 leaf = path->nodes[0];
2650 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2651 struct btrfs_inode_item);
2653 fill_inode_item(trans, leaf, inode_item, inode);
2654 btrfs_mark_buffer_dirty(leaf);
2655 btrfs_set_inode_last_trans(trans, inode);
2658 btrfs_free_path(path);
2664 * unlink helper that gets used here in inode.c and in the tree logging
2665 * recovery code. It remove a link in a directory with a given name, and
2666 * also drops the back refs in the inode to the directory
2668 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2669 struct btrfs_root *root,
2670 struct inode *dir, struct inode *inode,
2671 const char *name, int name_len)
2673 struct btrfs_path *path;
2675 struct extent_buffer *leaf;
2676 struct btrfs_dir_item *di;
2677 struct btrfs_key key;
2680 path = btrfs_alloc_path();
2686 path->leave_spinning = 1;
2687 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2688 name, name_len, -1);
2697 leaf = path->nodes[0];
2698 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2699 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2702 btrfs_release_path(root, path);
2704 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2706 dir->i_ino, &index);
2708 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2709 "inode %lu parent %lu\n", name_len, name,
2710 inode->i_ino, dir->i_ino);
2714 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2715 index, name, name_len, -1);
2724 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2725 btrfs_release_path(root, path);
2727 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2729 BUG_ON(ret != 0 && ret != -ENOENT);
2731 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2736 btrfs_free_path(path);
2740 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2741 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2742 btrfs_update_inode(trans, root, dir);
2747 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2748 struct btrfs_root *root,
2749 struct inode *dir, struct inode *inode,
2750 const char *name, int name_len)
2753 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2755 btrfs_drop_nlink(inode);
2756 ret = btrfs_update_inode(trans, root, inode);
2762 /* helper to check if there is any shared block in the path */
2763 static int check_path_shared(struct btrfs_root *root,
2764 struct btrfs_path *path)
2766 struct extent_buffer *eb;
2770 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2773 if (!path->nodes[level])
2775 eb = path->nodes[level];
2776 if (!btrfs_block_can_be_shared(root, eb))
2778 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2787 * helper to start transaction for unlink and rmdir.
2789 * unlink and rmdir are special in btrfs, they do not always free space.
2790 * so in enospc case, we should make sure they will free space before
2791 * allowing them to use the global metadata reservation.
2793 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2794 struct dentry *dentry)
2796 struct btrfs_trans_handle *trans;
2797 struct btrfs_root *root = BTRFS_I(dir)->root;
2798 struct btrfs_path *path;
2799 struct btrfs_inode_ref *ref;
2800 struct btrfs_dir_item *di;
2801 struct inode *inode = dentry->d_inode;
2807 trans = btrfs_start_transaction(root, 10);
2808 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2811 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2812 return ERR_PTR(-ENOSPC);
2814 /* check if there is someone else holds reference */
2815 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2816 return ERR_PTR(-ENOSPC);
2818 if (atomic_read(&inode->i_count) > 2)
2819 return ERR_PTR(-ENOSPC);
2821 if (xchg(&root->fs_info->enospc_unlink, 1))
2822 return ERR_PTR(-ENOSPC);
2824 path = btrfs_alloc_path();
2826 root->fs_info->enospc_unlink = 0;
2827 return ERR_PTR(-ENOMEM);
2830 trans = btrfs_start_transaction(root, 0);
2831 if (IS_ERR(trans)) {
2832 btrfs_free_path(path);
2833 root->fs_info->enospc_unlink = 0;
2837 path->skip_locking = 1;
2838 path->search_commit_root = 1;
2840 ret = btrfs_lookup_inode(trans, root, path,
2841 &BTRFS_I(dir)->location, 0);
2847 if (check_path_shared(root, path))
2852 btrfs_release_path(root, path);
2854 ret = btrfs_lookup_inode(trans, root, path,
2855 &BTRFS_I(inode)->location, 0);
2861 if (check_path_shared(root, path))
2866 btrfs_release_path(root, path);
2868 if (ret == 0 && S_ISREG(inode->i_mode)) {
2869 ret = btrfs_lookup_file_extent(trans, root, path,
2870 inode->i_ino, (u64)-1, 0);
2876 if (check_path_shared(root, path))
2878 btrfs_release_path(root, path);
2886 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2887 dentry->d_name.name, dentry->d_name.len, 0);
2893 if (check_path_shared(root, path))
2899 btrfs_release_path(root, path);
2901 ref = btrfs_lookup_inode_ref(trans, root, path,
2902 dentry->d_name.name, dentry->d_name.len,
2903 inode->i_ino, dir->i_ino, 0);
2909 if (check_path_shared(root, path))
2911 index = btrfs_inode_ref_index(path->nodes[0], ref);
2912 btrfs_release_path(root, path);
2914 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2915 dentry->d_name.name, dentry->d_name.len, 0);
2920 BUG_ON(ret == -ENOENT);
2921 if (check_path_shared(root, path))
2926 btrfs_free_path(path);
2928 btrfs_end_transaction(trans, root);
2929 root->fs_info->enospc_unlink = 0;
2930 return ERR_PTR(err);
2933 trans->block_rsv = &root->fs_info->global_block_rsv;
2937 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2938 struct btrfs_root *root)
2940 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2941 BUG_ON(!root->fs_info->enospc_unlink);
2942 root->fs_info->enospc_unlink = 0;
2944 btrfs_end_transaction_throttle(trans, root);
2947 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2949 struct btrfs_root *root = BTRFS_I(dir)->root;
2950 struct btrfs_trans_handle *trans;
2951 struct inode *inode = dentry->d_inode;
2953 unsigned long nr = 0;
2955 trans = __unlink_start_trans(dir, dentry);
2957 return PTR_ERR(trans);
2959 btrfs_set_trans_block_group(trans, dir);
2961 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2963 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2964 dentry->d_name.name, dentry->d_name.len);
2967 if (inode->i_nlink == 0) {
2968 ret = btrfs_orphan_add(trans, inode);
2972 nr = trans->blocks_used;
2973 __unlink_end_trans(trans, root);
2974 btrfs_btree_balance_dirty(root, nr);
2978 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2979 struct btrfs_root *root,
2980 struct inode *dir, u64 objectid,
2981 const char *name, int name_len)
2983 struct btrfs_path *path;
2984 struct extent_buffer *leaf;
2985 struct btrfs_dir_item *di;
2986 struct btrfs_key key;
2990 path = btrfs_alloc_path();
2994 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2995 name, name_len, -1);
2996 BUG_ON(!di || IS_ERR(di));
2998 leaf = path->nodes[0];
2999 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3000 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3001 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3003 btrfs_release_path(root, path);
3005 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3006 objectid, root->root_key.objectid,
3007 dir->i_ino, &index, name, name_len);
3009 BUG_ON(ret != -ENOENT);
3010 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3012 BUG_ON(!di || IS_ERR(di));
3014 leaf = path->nodes[0];
3015 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3016 btrfs_release_path(root, path);
3020 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3021 index, name, name_len, -1);
3022 BUG_ON(!di || IS_ERR(di));
3024 leaf = path->nodes[0];
3025 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3026 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3027 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3029 btrfs_release_path(root, path);
3031 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3032 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3033 ret = btrfs_update_inode(trans, root, dir);
3036 btrfs_free_path(path);
3040 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3042 struct inode *inode = dentry->d_inode;
3044 struct btrfs_root *root = BTRFS_I(dir)->root;
3045 struct btrfs_trans_handle *trans;
3046 unsigned long nr = 0;
3048 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3049 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3052 trans = __unlink_start_trans(dir, dentry);
3054 return PTR_ERR(trans);
3056 btrfs_set_trans_block_group(trans, dir);
3058 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3059 err = btrfs_unlink_subvol(trans, root, dir,
3060 BTRFS_I(inode)->location.objectid,
3061 dentry->d_name.name,
3062 dentry->d_name.len);
3066 err = btrfs_orphan_add(trans, inode);
3070 /* now the directory is empty */
3071 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3072 dentry->d_name.name, dentry->d_name.len);
3074 btrfs_i_size_write(inode, 0);
3076 nr = trans->blocks_used;
3077 __unlink_end_trans(trans, root);
3078 btrfs_btree_balance_dirty(root, nr);
3085 * when truncating bytes in a file, it is possible to avoid reading
3086 * the leaves that contain only checksum items. This can be the
3087 * majority of the IO required to delete a large file, but it must
3088 * be done carefully.
3090 * The keys in the level just above the leaves are checked to make sure
3091 * the lowest key in a given leaf is a csum key, and starts at an offset
3092 * after the new size.
3094 * Then the key for the next leaf is checked to make sure it also has
3095 * a checksum item for the same file. If it does, we know our target leaf
3096 * contains only checksum items, and it can be safely freed without reading
3099 * This is just an optimization targeted at large files. It may do
3100 * nothing. It will return 0 unless things went badly.
3102 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3103 struct btrfs_root *root,
3104 struct btrfs_path *path,
3105 struct inode *inode, u64 new_size)
3107 struct btrfs_key key;
3110 struct btrfs_key found_key;
3111 struct btrfs_key other_key;
3112 struct btrfs_leaf_ref *ref;
3116 path->lowest_level = 1;
3117 key.objectid = inode->i_ino;
3118 key.type = BTRFS_CSUM_ITEM_KEY;
3119 key.offset = new_size;
3121 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3125 if (path->nodes[1] == NULL) {
3130 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3131 nritems = btrfs_header_nritems(path->nodes[1]);
3136 if (path->slots[1] >= nritems)
3139 /* did we find a key greater than anything we want to delete? */
3140 if (found_key.objectid > inode->i_ino ||
3141 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3144 /* we check the next key in the node to make sure the leave contains
3145 * only checksum items. This comparison doesn't work if our
3146 * leaf is the last one in the node
3148 if (path->slots[1] + 1 >= nritems) {
3150 /* search forward from the last key in the node, this
3151 * will bring us into the next node in the tree
3153 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3155 /* unlikely, but we inc below, so check to be safe */
3156 if (found_key.offset == (u64)-1)
3159 /* search_forward needs a path with locks held, do the
3160 * search again for the original key. It is possible
3161 * this will race with a balance and return a path that
3162 * we could modify, but this drop is just an optimization
3163 * and is allowed to miss some leaves.
3165 btrfs_release_path(root, path);
3168 /* setup a max key for search_forward */
3169 other_key.offset = (u64)-1;
3170 other_key.type = key.type;
3171 other_key.objectid = key.objectid;
3173 path->keep_locks = 1;
3174 ret = btrfs_search_forward(root, &found_key, &other_key,
3176 path->keep_locks = 0;
3177 if (ret || found_key.objectid != key.objectid ||
3178 found_key.type != key.type) {
3183 key.offset = found_key.offset;
3184 btrfs_release_path(root, path);
3189 /* we know there's one more slot after us in the tree,
3190 * read that key so we can verify it is also a checksum item
3192 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3194 if (found_key.objectid < inode->i_ino)
3197 if (found_key.type != key.type || found_key.offset < new_size)
3201 * if the key for the next leaf isn't a csum key from this objectid,
3202 * we can't be sure there aren't good items inside this leaf.
3205 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3208 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3209 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3211 * it is safe to delete this leaf, it contains only
3212 * csum items from this inode at an offset >= new_size
3214 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3217 if (root->ref_cows && leaf_gen < trans->transid) {
3218 ref = btrfs_alloc_leaf_ref(root, 0);
3220 ref->root_gen = root->root_key.offset;
3221 ref->bytenr = leaf_start;
3223 ref->generation = leaf_gen;
3226 btrfs_sort_leaf_ref(ref);
3228 ret = btrfs_add_leaf_ref(root, ref, 0);
3230 btrfs_free_leaf_ref(root, ref);
3236 btrfs_release_path(root, path);
3238 if (other_key.objectid == inode->i_ino &&
3239 other_key.type == key.type && other_key.offset > key.offset) {
3240 key.offset = other_key.offset;
3246 /* fixup any changes we've made to the path */
3247 path->lowest_level = 0;
3248 path->keep_locks = 0;
3249 btrfs_release_path(root, path);
3256 * this can truncate away extent items, csum items and directory items.
3257 * It starts at a high offset and removes keys until it can't find
3258 * any higher than new_size
3260 * csum items that cross the new i_size are truncated to the new size
3263 * min_type is the minimum key type to truncate down to. If set to 0, this
3264 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3266 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3267 struct btrfs_root *root,
3268 struct inode *inode,
3269 u64 new_size, u32 min_type)
3271 struct btrfs_path *path;
3272 struct extent_buffer *leaf;
3273 struct btrfs_file_extent_item *fi;
3274 struct btrfs_key key;
3275 struct btrfs_key found_key;
3276 u64 extent_start = 0;
3277 u64 extent_num_bytes = 0;
3278 u64 extent_offset = 0;
3280 u64 mask = root->sectorsize - 1;
3281 u32 found_type = (u8)-1;
3284 int pending_del_nr = 0;
3285 int pending_del_slot = 0;
3286 int extent_type = -1;
3291 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3293 if (root->ref_cows || root == root->fs_info->tree_root)
3294 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3296 path = btrfs_alloc_path();
3300 key.objectid = inode->i_ino;
3301 key.offset = (u64)-1;
3305 path->leave_spinning = 1;
3306 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3313 /* there are no items in the tree for us to truncate, we're
3316 if (path->slots[0] == 0)
3323 leaf = path->nodes[0];
3324 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3325 found_type = btrfs_key_type(&found_key);
3328 if (found_key.objectid != inode->i_ino)
3331 if (found_type < min_type)
3334 item_end = found_key.offset;
3335 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3336 fi = btrfs_item_ptr(leaf, path->slots[0],
3337 struct btrfs_file_extent_item);
3338 extent_type = btrfs_file_extent_type(leaf, fi);
3339 encoding = btrfs_file_extent_compression(leaf, fi);
3340 encoding |= btrfs_file_extent_encryption(leaf, fi);
3341 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3343 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3345 btrfs_file_extent_num_bytes(leaf, fi);
3346 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3347 item_end += btrfs_file_extent_inline_len(leaf,
3352 if (found_type > min_type) {
3355 if (item_end < new_size)
3357 if (found_key.offset >= new_size)
3363 /* FIXME, shrink the extent if the ref count is only 1 */
3364 if (found_type != BTRFS_EXTENT_DATA_KEY)
3367 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3369 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3370 if (!del_item && !encoding) {
3371 u64 orig_num_bytes =
3372 btrfs_file_extent_num_bytes(leaf, fi);
3373 extent_num_bytes = new_size -
3374 found_key.offset + root->sectorsize - 1;
3375 extent_num_bytes = extent_num_bytes &
3376 ~((u64)root->sectorsize - 1);
3377 btrfs_set_file_extent_num_bytes(leaf, fi,
3379 num_dec = (orig_num_bytes -
3381 if (root->ref_cows && extent_start != 0)
3382 inode_sub_bytes(inode, num_dec);
3383 btrfs_mark_buffer_dirty(leaf);
3386 btrfs_file_extent_disk_num_bytes(leaf,
3388 extent_offset = found_key.offset -
3389 btrfs_file_extent_offset(leaf, fi);
3391 /* FIXME blocksize != 4096 */
3392 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3393 if (extent_start != 0) {
3396 inode_sub_bytes(inode, num_dec);
3399 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3401 * we can't truncate inline items that have had
3405 btrfs_file_extent_compression(leaf, fi) == 0 &&
3406 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3407 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3408 u32 size = new_size - found_key.offset;
3410 if (root->ref_cows) {
3411 inode_sub_bytes(inode, item_end + 1 -
3415 btrfs_file_extent_calc_inline_size(size);
3416 ret = btrfs_truncate_item(trans, root, path,
3419 } else if (root->ref_cows) {
3420 inode_sub_bytes(inode, item_end + 1 -
3426 if (!pending_del_nr) {
3427 /* no pending yet, add ourselves */
3428 pending_del_slot = path->slots[0];
3430 } else if (pending_del_nr &&
3431 path->slots[0] + 1 == pending_del_slot) {
3432 /* hop on the pending chunk */
3434 pending_del_slot = path->slots[0];
3441 if (found_extent && (root->ref_cows ||
3442 root == root->fs_info->tree_root)) {
3443 btrfs_set_path_blocking(path);
3444 ret = btrfs_free_extent(trans, root, extent_start,
3445 extent_num_bytes, 0,
3446 btrfs_header_owner(leaf),
3447 inode->i_ino, extent_offset);
3451 if (found_type == BTRFS_INODE_ITEM_KEY)
3454 if (path->slots[0] == 0 ||
3455 path->slots[0] != pending_del_slot) {
3456 if (root->ref_cows) {
3460 if (pending_del_nr) {
3461 ret = btrfs_del_items(trans, root, path,
3467 btrfs_release_path(root, path);
3474 if (pending_del_nr) {
3475 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3479 btrfs_free_path(path);
3484 * taken from block_truncate_page, but does cow as it zeros out
3485 * any bytes left in the last page in the file.
3487 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3489 struct inode *inode = mapping->host;
3490 struct btrfs_root *root = BTRFS_I(inode)->root;
3491 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3492 struct btrfs_ordered_extent *ordered;
3493 struct extent_state *cached_state = NULL;
3495 u32 blocksize = root->sectorsize;
3496 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3497 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3503 if ((offset & (blocksize - 1)) == 0)
3505 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3511 page = grab_cache_page(mapping, index);
3513 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3517 page_start = page_offset(page);
3518 page_end = page_start + PAGE_CACHE_SIZE - 1;
3520 if (!PageUptodate(page)) {
3521 ret = btrfs_readpage(NULL, page);
3523 if (page->mapping != mapping) {
3525 page_cache_release(page);
3528 if (!PageUptodate(page)) {
3533 wait_on_page_writeback(page);
3535 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3537 set_page_extent_mapped(page);
3539 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3541 unlock_extent_cached(io_tree, page_start, page_end,
3542 &cached_state, GFP_NOFS);
3544 page_cache_release(page);
3545 btrfs_start_ordered_extent(inode, ordered, 1);
3546 btrfs_put_ordered_extent(ordered);
3550 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3551 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3552 0, 0, &cached_state, GFP_NOFS);
3554 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3557 unlock_extent_cached(io_tree, page_start, page_end,
3558 &cached_state, GFP_NOFS);
3563 if (offset != PAGE_CACHE_SIZE) {
3565 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3566 flush_dcache_page(page);
3569 ClearPageChecked(page);
3570 set_page_dirty(page);
3571 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3576 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3578 page_cache_release(page);
3584 * This function puts in dummy file extents for the area we're creating a hole
3585 * for. So if we are truncating this file to a larger size we need to insert
3586 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3587 * the range between oldsize and size
3589 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3591 struct btrfs_trans_handle *trans;
3592 struct btrfs_root *root = BTRFS_I(inode)->root;
3593 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3594 struct extent_map *em = NULL;
3595 struct extent_state *cached_state = NULL;
3596 u64 mask = root->sectorsize - 1;
3597 u64 hole_start = (oldsize + mask) & ~mask;
3598 u64 block_end = (size + mask) & ~mask;
3604 if (size <= hole_start)
3608 struct btrfs_ordered_extent *ordered;
3609 btrfs_wait_ordered_range(inode, hole_start,
3610 block_end - hole_start);
3611 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3612 &cached_state, GFP_NOFS);
3613 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3616 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3617 &cached_state, GFP_NOFS);
3618 btrfs_put_ordered_extent(ordered);
3621 cur_offset = hole_start;
3623 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3624 block_end - cur_offset, 0);
3625 BUG_ON(IS_ERR(em) || !em);
3626 last_byte = min(extent_map_end(em), block_end);
3627 last_byte = (last_byte + mask) & ~mask;
3628 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3630 hole_size = last_byte - cur_offset;
3632 trans = btrfs_start_transaction(root, 2);
3633 if (IS_ERR(trans)) {
3634 err = PTR_ERR(trans);
3637 btrfs_set_trans_block_group(trans, inode);
3639 err = btrfs_drop_extents(trans, inode, cur_offset,
3640 cur_offset + hole_size,
3645 err = btrfs_insert_file_extent(trans, root,
3646 inode->i_ino, cur_offset, 0,
3647 0, hole_size, 0, hole_size,
3652 btrfs_drop_extent_cache(inode, hole_start,
3655 btrfs_end_transaction(trans, root);
3657 free_extent_map(em);
3659 cur_offset = last_byte;
3660 if (cur_offset >= block_end)
3664 free_extent_map(em);
3665 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3670 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3672 loff_t oldsize = i_size_read(inode);
3675 if (newsize == oldsize)
3678 if (newsize > oldsize) {
3679 i_size_write(inode, newsize);
3680 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3681 truncate_pagecache(inode, oldsize, newsize);
3682 ret = btrfs_cont_expand(inode, oldsize, newsize);
3684 btrfs_setsize(inode, oldsize);
3688 mark_inode_dirty(inode);
3692 * We're truncating a file that used to have good data down to
3693 * zero. Make sure it gets into the ordered flush list so that
3694 * any new writes get down to disk quickly.
3697 BTRFS_I(inode)->ordered_data_close = 1;
3699 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3700 truncate_setsize(inode, newsize);
3701 ret = btrfs_truncate(inode);
3707 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3709 struct inode *inode = dentry->d_inode;
3710 struct btrfs_root *root = BTRFS_I(inode)->root;
3713 if (btrfs_root_readonly(root))
3716 err = inode_change_ok(inode, attr);
3720 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3721 err = btrfs_setsize(inode, attr->ia_size);
3726 if (attr->ia_valid) {
3727 setattr_copy(inode, attr);
3728 mark_inode_dirty(inode);
3730 if (attr->ia_valid & ATTR_MODE)
3731 err = btrfs_acl_chmod(inode);
3737 void btrfs_evict_inode(struct inode *inode)
3739 struct btrfs_trans_handle *trans;
3740 struct btrfs_root *root = BTRFS_I(inode)->root;
3744 trace_btrfs_inode_evict(inode);
3746 truncate_inode_pages(&inode->i_data, 0);
3747 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3748 root == root->fs_info->tree_root))
3751 if (is_bad_inode(inode)) {
3752 btrfs_orphan_del(NULL, inode);
3755 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3756 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3758 if (root->fs_info->log_root_recovering) {
3759 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3763 if (inode->i_nlink > 0) {
3764 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3768 btrfs_i_size_write(inode, 0);
3771 trans = btrfs_start_transaction(root, 0);
3772 BUG_ON(IS_ERR(trans));
3773 btrfs_set_trans_block_group(trans, inode);
3774 trans->block_rsv = root->orphan_block_rsv;
3776 ret = btrfs_block_rsv_check(trans, root,
3777 root->orphan_block_rsv, 0, 5);
3779 BUG_ON(ret != -EAGAIN);
3780 ret = btrfs_commit_transaction(trans, root);
3785 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3789 nr = trans->blocks_used;
3790 btrfs_end_transaction(trans, root);
3792 btrfs_btree_balance_dirty(root, nr);
3797 ret = btrfs_orphan_del(trans, inode);
3801 nr = trans->blocks_used;
3802 btrfs_end_transaction(trans, root);
3803 btrfs_btree_balance_dirty(root, nr);
3805 end_writeback(inode);
3810 * this returns the key found in the dir entry in the location pointer.
3811 * If no dir entries were found, location->objectid is 0.
3813 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3814 struct btrfs_key *location)
3816 const char *name = dentry->d_name.name;
3817 int namelen = dentry->d_name.len;
3818 struct btrfs_dir_item *di;
3819 struct btrfs_path *path;
3820 struct btrfs_root *root = BTRFS_I(dir)->root;
3823 path = btrfs_alloc_path();
3826 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3831 if (!di || IS_ERR(di))
3834 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3836 btrfs_free_path(path);
3839 location->objectid = 0;
3844 * when we hit a tree root in a directory, the btrfs part of the inode
3845 * needs to be changed to reflect the root directory of the tree root. This
3846 * is kind of like crossing a mount point.
3848 static int fixup_tree_root_location(struct btrfs_root *root,
3850 struct dentry *dentry,
3851 struct btrfs_key *location,
3852 struct btrfs_root **sub_root)
3854 struct btrfs_path *path;
3855 struct btrfs_root *new_root;
3856 struct btrfs_root_ref *ref;
3857 struct extent_buffer *leaf;
3861 path = btrfs_alloc_path();
3868 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3869 BTRFS_I(dir)->root->root_key.objectid,
3870 location->objectid);
3877 leaf = path->nodes[0];
3878 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3879 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3880 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3883 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3884 (unsigned long)(ref + 1),
3885 dentry->d_name.len);
3889 btrfs_release_path(root->fs_info->tree_root, path);
3891 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3892 if (IS_ERR(new_root)) {
3893 err = PTR_ERR(new_root);
3897 if (btrfs_root_refs(&new_root->root_item) == 0) {
3902 *sub_root = new_root;
3903 location->objectid = btrfs_root_dirid(&new_root->root_item);
3904 location->type = BTRFS_INODE_ITEM_KEY;
3905 location->offset = 0;
3908 btrfs_free_path(path);
3912 static void inode_tree_add(struct inode *inode)
3914 struct btrfs_root *root = BTRFS_I(inode)->root;
3915 struct btrfs_inode *entry;
3917 struct rb_node *parent;
3919 p = &root->inode_tree.rb_node;
3922 if (inode_unhashed(inode))
3925 spin_lock(&root->inode_lock);
3928 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3930 if (inode->i_ino < entry->vfs_inode.i_ino)
3931 p = &parent->rb_left;
3932 else if (inode->i_ino > entry->vfs_inode.i_ino)
3933 p = &parent->rb_right;
3935 WARN_ON(!(entry->vfs_inode.i_state &
3936 (I_WILL_FREE | I_FREEING)));
3937 rb_erase(parent, &root->inode_tree);
3938 RB_CLEAR_NODE(parent);
3939 spin_unlock(&root->inode_lock);
3943 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3944 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3945 spin_unlock(&root->inode_lock);
3948 static void inode_tree_del(struct inode *inode)
3950 struct btrfs_root *root = BTRFS_I(inode)->root;
3953 spin_lock(&root->inode_lock);
3954 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3955 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3956 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3957 empty = RB_EMPTY_ROOT(&root->inode_tree);
3959 spin_unlock(&root->inode_lock);
3962 * Free space cache has inodes in the tree root, but the tree root has a
3963 * root_refs of 0, so this could end up dropping the tree root as a
3964 * snapshot, so we need the extra !root->fs_info->tree_root check to
3965 * make sure we don't drop it.
3967 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3968 root != root->fs_info->tree_root) {
3969 synchronize_srcu(&root->fs_info->subvol_srcu);
3970 spin_lock(&root->inode_lock);
3971 empty = RB_EMPTY_ROOT(&root->inode_tree);
3972 spin_unlock(&root->inode_lock);
3974 btrfs_add_dead_root(root);
3978 int btrfs_invalidate_inodes(struct btrfs_root *root)
3980 struct rb_node *node;
3981 struct rb_node *prev;
3982 struct btrfs_inode *entry;
3983 struct inode *inode;
3986 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3988 spin_lock(&root->inode_lock);
3990 node = root->inode_tree.rb_node;
3994 entry = rb_entry(node, struct btrfs_inode, rb_node);
3996 if (objectid < entry->vfs_inode.i_ino)
3997 node = node->rb_left;
3998 else if (objectid > entry->vfs_inode.i_ino)
3999 node = node->rb_right;
4005 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4006 if (objectid <= entry->vfs_inode.i_ino) {
4010 prev = rb_next(prev);
4014 entry = rb_entry(node, struct btrfs_inode, rb_node);
4015 objectid = entry->vfs_inode.i_ino + 1;
4016 inode = igrab(&entry->vfs_inode);
4018 spin_unlock(&root->inode_lock);
4019 if (atomic_read(&inode->i_count) > 1)
4020 d_prune_aliases(inode);
4022 * btrfs_drop_inode will have it removed from
4023 * the inode cache when its usage count
4028 spin_lock(&root->inode_lock);
4032 if (cond_resched_lock(&root->inode_lock))
4035 node = rb_next(node);
4037 spin_unlock(&root->inode_lock);
4041 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4043 struct btrfs_iget_args *args = p;
4044 inode->i_ino = args->ino;
4045 BTRFS_I(inode)->root = args->root;
4046 btrfs_set_inode_space_info(args->root, inode);
4050 static int btrfs_find_actor(struct inode *inode, void *opaque)
4052 struct btrfs_iget_args *args = opaque;
4053 return args->ino == inode->i_ino &&
4054 args->root == BTRFS_I(inode)->root;
4057 static struct inode *btrfs_iget_locked(struct super_block *s,
4059 struct btrfs_root *root)
4061 struct inode *inode;
4062 struct btrfs_iget_args args;
4063 args.ino = objectid;
4066 inode = iget5_locked(s, objectid, btrfs_find_actor,
4067 btrfs_init_locked_inode,
4072 /* Get an inode object given its location and corresponding root.
4073 * Returns in *is_new if the inode was read from disk
4075 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4076 struct btrfs_root *root, int *new)
4078 struct inode *inode;
4080 inode = btrfs_iget_locked(s, location->objectid, root);
4082 return ERR_PTR(-ENOMEM);
4084 if (inode->i_state & I_NEW) {
4085 BTRFS_I(inode)->root = root;
4086 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4087 btrfs_read_locked_inode(inode);
4088 inode_tree_add(inode);
4089 unlock_new_inode(inode);
4097 static struct inode *new_simple_dir(struct super_block *s,
4098 struct btrfs_key *key,
4099 struct btrfs_root *root)
4101 struct inode *inode = new_inode(s);
4104 return ERR_PTR(-ENOMEM);
4106 BTRFS_I(inode)->root = root;
4107 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4108 BTRFS_I(inode)->dummy_inode = 1;
4110 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4111 inode->i_op = &simple_dir_inode_operations;
4112 inode->i_fop = &simple_dir_operations;
4113 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4114 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4119 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4121 struct inode *inode;
4122 struct btrfs_root *root = BTRFS_I(dir)->root;
4123 struct btrfs_root *sub_root = root;
4124 struct btrfs_key location;
4128 if (dentry->d_name.len > BTRFS_NAME_LEN)
4129 return ERR_PTR(-ENAMETOOLONG);
4131 ret = btrfs_inode_by_name(dir, dentry, &location);
4134 return ERR_PTR(ret);
4136 if (location.objectid == 0)
4139 if (location.type == BTRFS_INODE_ITEM_KEY) {
4140 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4144 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4146 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4147 ret = fixup_tree_root_location(root, dir, dentry,
4148 &location, &sub_root);
4151 inode = ERR_PTR(ret);
4153 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4155 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4157 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4159 if (!IS_ERR(inode) && root != sub_root) {
4160 down_read(&root->fs_info->cleanup_work_sem);
4161 if (!(inode->i_sb->s_flags & MS_RDONLY))
4162 ret = btrfs_orphan_cleanup(sub_root);
4163 up_read(&root->fs_info->cleanup_work_sem);
4165 inode = ERR_PTR(ret);
4171 static int btrfs_dentry_delete(const struct dentry *dentry)
4173 struct btrfs_root *root;
4175 if (!dentry->d_inode && !IS_ROOT(dentry))
4176 dentry = dentry->d_parent;
4178 if (dentry->d_inode) {
4179 root = BTRFS_I(dentry->d_inode)->root;
4180 if (btrfs_root_refs(&root->root_item) == 0)
4186 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4187 struct nameidata *nd)
4189 struct inode *inode;
4191 inode = btrfs_lookup_dentry(dir, dentry);
4193 return ERR_CAST(inode);
4195 return d_splice_alias(inode, dentry);
4198 static unsigned char btrfs_filetype_table[] = {
4199 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4202 static int btrfs_real_readdir(struct file *filp, void *dirent,
4205 struct inode *inode = filp->f_dentry->d_inode;
4206 struct btrfs_root *root = BTRFS_I(inode)->root;
4207 struct btrfs_item *item;
4208 struct btrfs_dir_item *di;
4209 struct btrfs_key key;
4210 struct btrfs_key found_key;
4211 struct btrfs_path *path;
4214 struct extent_buffer *leaf;
4217 unsigned char d_type;
4222 int key_type = BTRFS_DIR_INDEX_KEY;
4227 /* FIXME, use a real flag for deciding about the key type */
4228 if (root->fs_info->tree_root == root)
4229 key_type = BTRFS_DIR_ITEM_KEY;
4231 /* special case for "." */
4232 if (filp->f_pos == 0) {
4233 over = filldir(dirent, ".", 1,
4240 /* special case for .., just use the back ref */
4241 if (filp->f_pos == 1) {
4242 u64 pino = parent_ino(filp->f_path.dentry);
4243 over = filldir(dirent, "..", 2,
4249 path = btrfs_alloc_path();
4252 btrfs_set_key_type(&key, key_type);
4253 key.offset = filp->f_pos;
4254 key.objectid = inode->i_ino;
4256 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4262 leaf = path->nodes[0];
4263 nritems = btrfs_header_nritems(leaf);
4264 slot = path->slots[0];
4265 if (advance || slot >= nritems) {
4266 if (slot >= nritems - 1) {
4267 ret = btrfs_next_leaf(root, path);
4270 leaf = path->nodes[0];
4271 nritems = btrfs_header_nritems(leaf);
4272 slot = path->slots[0];
4280 item = btrfs_item_nr(leaf, slot);
4281 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4283 if (found_key.objectid != key.objectid)
4285 if (btrfs_key_type(&found_key) != key_type)
4287 if (found_key.offset < filp->f_pos)
4290 filp->f_pos = found_key.offset;
4292 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4294 di_total = btrfs_item_size(leaf, item);
4296 while (di_cur < di_total) {
4297 struct btrfs_key location;
4299 if (verify_dir_item(root, leaf, di))
4302 name_len = btrfs_dir_name_len(leaf, di);
4303 if (name_len <= sizeof(tmp_name)) {
4304 name_ptr = tmp_name;
4306 name_ptr = kmalloc(name_len, GFP_NOFS);
4312 read_extent_buffer(leaf, name_ptr,
4313 (unsigned long)(di + 1), name_len);
4315 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4316 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4318 /* is this a reference to our own snapshot? If so
4321 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4322 location.objectid == root->root_key.objectid) {
4326 over = filldir(dirent, name_ptr, name_len,
4327 found_key.offset, location.objectid,
4331 if (name_ptr != tmp_name)
4336 di_len = btrfs_dir_name_len(leaf, di) +
4337 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4339 di = (struct btrfs_dir_item *)((char *)di + di_len);
4343 /* Reached end of directory/root. Bump pos past the last item. */
4344 if (key_type == BTRFS_DIR_INDEX_KEY)
4346 * 32-bit glibc will use getdents64, but then strtol -
4347 * so the last number we can serve is this.
4349 filp->f_pos = 0x7fffffff;
4355 btrfs_free_path(path);
4359 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4361 struct btrfs_root *root = BTRFS_I(inode)->root;
4362 struct btrfs_trans_handle *trans;
4364 bool nolock = false;
4366 if (BTRFS_I(inode)->dummy_inode)
4370 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4372 if (wbc->sync_mode == WB_SYNC_ALL) {
4374 trans = btrfs_join_transaction_nolock(root, 1);
4376 trans = btrfs_join_transaction(root, 1);
4378 return PTR_ERR(trans);
4379 btrfs_set_trans_block_group(trans, inode);
4381 ret = btrfs_end_transaction_nolock(trans, root);
4383 ret = btrfs_commit_transaction(trans, root);
4389 * This is somewhat expensive, updating the tree every time the
4390 * inode changes. But, it is most likely to find the inode in cache.
4391 * FIXME, needs more benchmarking...there are no reasons other than performance
4392 * to keep or drop this code.
4394 void btrfs_dirty_inode(struct inode *inode)
4396 struct btrfs_root *root = BTRFS_I(inode)->root;
4397 struct btrfs_trans_handle *trans;
4400 if (BTRFS_I(inode)->dummy_inode)
4403 trans = btrfs_join_transaction(root, 1);
4404 BUG_ON(IS_ERR(trans));
4405 btrfs_set_trans_block_group(trans, inode);
4407 ret = btrfs_update_inode(trans, root, inode);
4408 if (ret && ret == -ENOSPC) {
4409 /* whoops, lets try again with the full transaction */
4410 btrfs_end_transaction(trans, root);
4411 trans = btrfs_start_transaction(root, 1);
4412 if (IS_ERR(trans)) {
4413 if (printk_ratelimit()) {
4414 printk(KERN_ERR "btrfs: fail to "
4415 "dirty inode %lu error %ld\n",
4416 inode->i_ino, PTR_ERR(trans));
4420 btrfs_set_trans_block_group(trans, inode);
4422 ret = btrfs_update_inode(trans, root, inode);
4424 if (printk_ratelimit()) {
4425 printk(KERN_ERR "btrfs: fail to "
4426 "dirty inode %lu error %d\n",
4431 btrfs_end_transaction(trans, root);
4435 * find the highest existing sequence number in a directory
4436 * and then set the in-memory index_cnt variable to reflect
4437 * free sequence numbers
4439 static int btrfs_set_inode_index_count(struct inode *inode)
4441 struct btrfs_root *root = BTRFS_I(inode)->root;
4442 struct btrfs_key key, found_key;
4443 struct btrfs_path *path;
4444 struct extent_buffer *leaf;
4447 key.objectid = inode->i_ino;
4448 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4449 key.offset = (u64)-1;
4451 path = btrfs_alloc_path();
4455 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4458 /* FIXME: we should be able to handle this */
4464 * MAGIC NUMBER EXPLANATION:
4465 * since we search a directory based on f_pos we have to start at 2
4466 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4467 * else has to start at 2
4469 if (path->slots[0] == 0) {
4470 BTRFS_I(inode)->index_cnt = 2;
4476 leaf = path->nodes[0];
4477 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4479 if (found_key.objectid != inode->i_ino ||
4480 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4481 BTRFS_I(inode)->index_cnt = 2;
4485 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4487 btrfs_free_path(path);
4492 * helper to find a free sequence number in a given directory. This current
4493 * code is very simple, later versions will do smarter things in the btree
4495 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4499 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4500 ret = btrfs_set_inode_index_count(dir);
4505 *index = BTRFS_I(dir)->index_cnt;
4506 BTRFS_I(dir)->index_cnt++;
4511 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4512 struct btrfs_root *root,
4514 const char *name, int name_len,
4515 u64 ref_objectid, u64 objectid,
4516 u64 alloc_hint, int mode, u64 *index)
4518 struct inode *inode;
4519 struct btrfs_inode_item *inode_item;
4520 struct btrfs_key *location;
4521 struct btrfs_path *path;
4522 struct btrfs_inode_ref *ref;
4523 struct btrfs_key key[2];
4529 path = btrfs_alloc_path();
4532 inode = new_inode(root->fs_info->sb);
4534 return ERR_PTR(-ENOMEM);
4537 trace_btrfs_inode_request(dir);
4539 ret = btrfs_set_inode_index(dir, index);
4542 return ERR_PTR(ret);
4546 * index_cnt is ignored for everything but a dir,
4547 * btrfs_get_inode_index_count has an explanation for the magic
4550 BTRFS_I(inode)->index_cnt = 2;
4551 BTRFS_I(inode)->root = root;
4552 BTRFS_I(inode)->generation = trans->transid;
4553 inode->i_generation = BTRFS_I(inode)->generation;
4554 btrfs_set_inode_space_info(root, inode);
4560 BTRFS_I(inode)->block_group =
4561 btrfs_find_block_group(root, 0, alloc_hint, owner);
4563 key[0].objectid = objectid;
4564 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4567 key[1].objectid = objectid;
4568 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4569 key[1].offset = ref_objectid;
4571 sizes[0] = sizeof(struct btrfs_inode_item);
4572 sizes[1] = name_len + sizeof(*ref);
4574 path->leave_spinning = 1;
4575 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4579 inode_init_owner(inode, dir, mode);
4580 inode->i_ino = objectid;
4581 inode_set_bytes(inode, 0);
4582 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4583 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4584 struct btrfs_inode_item);
4585 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4587 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4588 struct btrfs_inode_ref);
4589 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4590 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4591 ptr = (unsigned long)(ref + 1);
4592 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4594 btrfs_mark_buffer_dirty(path->nodes[0]);
4595 btrfs_free_path(path);
4597 location = &BTRFS_I(inode)->location;
4598 location->objectid = objectid;
4599 location->offset = 0;
4600 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4602 btrfs_inherit_iflags(inode, dir);
4604 if ((mode & S_IFREG)) {
4605 if (btrfs_test_opt(root, NODATASUM))
4606 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4607 if (btrfs_test_opt(root, NODATACOW) ||
4608 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4609 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4612 insert_inode_hash(inode);
4613 inode_tree_add(inode);
4615 trace_btrfs_inode_new(inode);
4620 BTRFS_I(dir)->index_cnt--;
4621 btrfs_free_path(path);
4623 return ERR_PTR(ret);
4626 static inline u8 btrfs_inode_type(struct inode *inode)
4628 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4632 * utility function to add 'inode' into 'parent_inode' with
4633 * a give name and a given sequence number.
4634 * if 'add_backref' is true, also insert a backref from the
4635 * inode to the parent directory.
4637 int btrfs_add_link(struct btrfs_trans_handle *trans,
4638 struct inode *parent_inode, struct inode *inode,
4639 const char *name, int name_len, int add_backref, u64 index)
4642 struct btrfs_key key;
4643 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4645 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4646 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4648 key.objectid = inode->i_ino;
4649 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4653 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4654 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4655 key.objectid, root->root_key.objectid,
4656 parent_inode->i_ino,
4657 index, name, name_len);
4658 } else if (add_backref) {
4659 ret = btrfs_insert_inode_ref(trans, root,
4660 name, name_len, inode->i_ino,
4661 parent_inode->i_ino, index);
4665 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4666 parent_inode->i_ino, &key,
4667 btrfs_inode_type(inode), index);
4670 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4672 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4673 ret = btrfs_update_inode(trans, root, parent_inode);
4678 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4679 struct inode *dir, struct dentry *dentry,
4680 struct inode *inode, int backref, u64 index)
4682 int err = btrfs_add_link(trans, dir, inode,
4683 dentry->d_name.name, dentry->d_name.len,
4686 d_instantiate(dentry, inode);
4694 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4695 int mode, dev_t rdev)
4697 struct btrfs_trans_handle *trans;
4698 struct btrfs_root *root = BTRFS_I(dir)->root;
4699 struct inode *inode = NULL;
4703 unsigned long nr = 0;
4706 if (!new_valid_dev(rdev))
4709 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4714 * 2 for inode item and ref
4716 * 1 for xattr if selinux is on
4718 trans = btrfs_start_transaction(root, 5);
4720 return PTR_ERR(trans);
4722 btrfs_set_trans_block_group(trans, dir);
4724 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4725 dentry->d_name.len, dir->i_ino, objectid,
4726 BTRFS_I(dir)->block_group, mode, &index);
4727 err = PTR_ERR(inode);
4731 err = btrfs_init_inode_security(trans, inode, dir);
4737 btrfs_set_trans_block_group(trans, inode);
4738 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4742 inode->i_op = &btrfs_special_inode_operations;
4743 init_special_inode(inode, inode->i_mode, rdev);
4744 btrfs_update_inode(trans, root, inode);
4746 btrfs_update_inode_block_group(trans, inode);
4747 btrfs_update_inode_block_group(trans, dir);
4749 nr = trans->blocks_used;
4750 btrfs_end_transaction_throttle(trans, root);
4751 btrfs_btree_balance_dirty(root, nr);
4753 inode_dec_link_count(inode);
4759 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4760 int mode, struct nameidata *nd)
4762 struct btrfs_trans_handle *trans;
4763 struct btrfs_root *root = BTRFS_I(dir)->root;
4764 struct inode *inode = NULL;
4767 unsigned long nr = 0;
4771 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4775 * 2 for inode item and ref
4777 * 1 for xattr if selinux is on
4779 trans = btrfs_start_transaction(root, 5);
4781 return PTR_ERR(trans);
4783 btrfs_set_trans_block_group(trans, dir);
4785 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4786 dentry->d_name.len, dir->i_ino, objectid,
4787 BTRFS_I(dir)->block_group, mode, &index);
4788 err = PTR_ERR(inode);
4792 err = btrfs_init_inode_security(trans, inode, dir);
4798 btrfs_set_trans_block_group(trans, inode);
4799 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4803 inode->i_mapping->a_ops = &btrfs_aops;
4804 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4805 inode->i_fop = &btrfs_file_operations;
4806 inode->i_op = &btrfs_file_inode_operations;
4807 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4809 btrfs_update_inode_block_group(trans, inode);
4810 btrfs_update_inode_block_group(trans, dir);
4812 nr = trans->blocks_used;
4813 btrfs_end_transaction_throttle(trans, root);
4815 inode_dec_link_count(inode);
4818 btrfs_btree_balance_dirty(root, nr);
4822 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4823 struct dentry *dentry)
4825 struct btrfs_trans_handle *trans;
4826 struct btrfs_root *root = BTRFS_I(dir)->root;
4827 struct inode *inode = old_dentry->d_inode;
4829 unsigned long nr = 0;
4833 if (inode->i_nlink == 0)
4836 /* do not allow sys_link's with other subvols of the same device */
4837 if (root->objectid != BTRFS_I(inode)->root->objectid)
4840 if (inode->i_nlink == ~0U)
4843 btrfs_inc_nlink(inode);
4844 inode->i_ctime = CURRENT_TIME;
4846 err = btrfs_set_inode_index(dir, &index);
4851 * 2 items for inode and inode ref
4852 * 2 items for dir items
4853 * 1 item for parent inode
4855 trans = btrfs_start_transaction(root, 5);
4856 if (IS_ERR(trans)) {
4857 err = PTR_ERR(trans);
4861 btrfs_set_trans_block_group(trans, dir);
4864 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4869 struct dentry *parent = dget_parent(dentry);
4870 btrfs_update_inode_block_group(trans, dir);
4871 err = btrfs_update_inode(trans, root, inode);
4873 btrfs_log_new_name(trans, inode, NULL, parent);
4877 nr = trans->blocks_used;
4878 btrfs_end_transaction_throttle(trans, root);
4881 inode_dec_link_count(inode);
4884 btrfs_btree_balance_dirty(root, nr);
4888 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4890 struct inode *inode = NULL;
4891 struct btrfs_trans_handle *trans;
4892 struct btrfs_root *root = BTRFS_I(dir)->root;
4894 int drop_on_err = 0;
4897 unsigned long nr = 1;
4899 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4904 * 2 items for inode and ref
4905 * 2 items for dir items
4906 * 1 for xattr if selinux is on
4908 trans = btrfs_start_transaction(root, 5);
4910 return PTR_ERR(trans);
4911 btrfs_set_trans_block_group(trans, dir);
4913 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4914 dentry->d_name.len, dir->i_ino, objectid,
4915 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4917 if (IS_ERR(inode)) {
4918 err = PTR_ERR(inode);
4924 err = btrfs_init_inode_security(trans, inode, dir);
4928 inode->i_op = &btrfs_dir_inode_operations;
4929 inode->i_fop = &btrfs_dir_file_operations;
4930 btrfs_set_trans_block_group(trans, inode);
4932 btrfs_i_size_write(inode, 0);
4933 err = btrfs_update_inode(trans, root, inode);
4937 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4938 dentry->d_name.len, 0, index);
4942 d_instantiate(dentry, inode);
4944 btrfs_update_inode_block_group(trans, inode);
4945 btrfs_update_inode_block_group(trans, dir);
4948 nr = trans->blocks_used;
4949 btrfs_end_transaction_throttle(trans, root);
4952 btrfs_btree_balance_dirty(root, nr);
4956 /* helper for btfs_get_extent. Given an existing extent in the tree,
4957 * and an extent that you want to insert, deal with overlap and insert
4958 * the new extent into the tree.
4960 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4961 struct extent_map *existing,
4962 struct extent_map *em,
4963 u64 map_start, u64 map_len)
4967 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4968 start_diff = map_start - em->start;
4969 em->start = map_start;
4971 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4972 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4973 em->block_start += start_diff;
4974 em->block_len -= start_diff;
4976 return add_extent_mapping(em_tree, em);
4979 static noinline int uncompress_inline(struct btrfs_path *path,
4980 struct inode *inode, struct page *page,
4981 size_t pg_offset, u64 extent_offset,
4982 struct btrfs_file_extent_item *item)
4985 struct extent_buffer *leaf = path->nodes[0];
4988 unsigned long inline_size;
4992 WARN_ON(pg_offset != 0);
4993 compress_type = btrfs_file_extent_compression(leaf, item);
4994 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4995 inline_size = btrfs_file_extent_inline_item_len(leaf,
4996 btrfs_item_nr(leaf, path->slots[0]));
4997 tmp = kmalloc(inline_size, GFP_NOFS);
4998 ptr = btrfs_file_extent_inline_start(item);
5000 read_extent_buffer(leaf, tmp, ptr, inline_size);
5002 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5003 ret = btrfs_decompress(compress_type, tmp, page,
5004 extent_offset, inline_size, max_size);
5006 char *kaddr = kmap_atomic(page, KM_USER0);
5007 unsigned long copy_size = min_t(u64,
5008 PAGE_CACHE_SIZE - pg_offset,
5009 max_size - extent_offset);
5010 memset(kaddr + pg_offset, 0, copy_size);
5011 kunmap_atomic(kaddr, KM_USER0);
5018 * a bit scary, this does extent mapping from logical file offset to the disk.
5019 * the ugly parts come from merging extents from the disk with the in-ram
5020 * representation. This gets more complex because of the data=ordered code,
5021 * where the in-ram extents might be locked pending data=ordered completion.
5023 * This also copies inline extents directly into the page.
5026 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5027 size_t pg_offset, u64 start, u64 len,
5033 u64 extent_start = 0;
5035 u64 objectid = inode->i_ino;
5037 struct btrfs_path *path = NULL;
5038 struct btrfs_root *root = BTRFS_I(inode)->root;
5039 struct btrfs_file_extent_item *item;
5040 struct extent_buffer *leaf;
5041 struct btrfs_key found_key;
5042 struct extent_map *em = NULL;
5043 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5044 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5045 struct btrfs_trans_handle *trans = NULL;
5049 read_lock(&em_tree->lock);
5050 em = lookup_extent_mapping(em_tree, start, len);
5052 em->bdev = root->fs_info->fs_devices->latest_bdev;
5053 read_unlock(&em_tree->lock);
5056 if (em->start > start || em->start + em->len <= start)
5057 free_extent_map(em);
5058 else if (em->block_start == EXTENT_MAP_INLINE && page)
5059 free_extent_map(em);
5063 em = alloc_extent_map(GFP_NOFS);
5068 em->bdev = root->fs_info->fs_devices->latest_bdev;
5069 em->start = EXTENT_MAP_HOLE;
5070 em->orig_start = EXTENT_MAP_HOLE;
5072 em->block_len = (u64)-1;
5075 path = btrfs_alloc_path();
5079 ret = btrfs_lookup_file_extent(trans, root, path,
5080 objectid, start, trans != NULL);
5087 if (path->slots[0] == 0)
5092 leaf = path->nodes[0];
5093 item = btrfs_item_ptr(leaf, path->slots[0],
5094 struct btrfs_file_extent_item);
5095 /* are we inside the extent that was found? */
5096 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5097 found_type = btrfs_key_type(&found_key);
5098 if (found_key.objectid != objectid ||
5099 found_type != BTRFS_EXTENT_DATA_KEY) {
5103 found_type = btrfs_file_extent_type(leaf, item);
5104 extent_start = found_key.offset;
5105 compress_type = btrfs_file_extent_compression(leaf, item);
5106 if (found_type == BTRFS_FILE_EXTENT_REG ||
5107 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5108 extent_end = extent_start +
5109 btrfs_file_extent_num_bytes(leaf, item);
5110 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5112 size = btrfs_file_extent_inline_len(leaf, item);
5113 extent_end = (extent_start + size + root->sectorsize - 1) &
5114 ~((u64)root->sectorsize - 1);
5117 if (start >= extent_end) {
5119 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5120 ret = btrfs_next_leaf(root, path);
5127 leaf = path->nodes[0];
5129 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5130 if (found_key.objectid != objectid ||
5131 found_key.type != BTRFS_EXTENT_DATA_KEY)
5133 if (start + len <= found_key.offset)
5136 em->len = found_key.offset - start;
5140 if (found_type == BTRFS_FILE_EXTENT_REG ||
5141 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5142 em->start = extent_start;
5143 em->len = extent_end - extent_start;
5144 em->orig_start = extent_start -
5145 btrfs_file_extent_offset(leaf, item);
5146 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5148 em->block_start = EXTENT_MAP_HOLE;
5151 if (compress_type != BTRFS_COMPRESS_NONE) {
5152 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5153 em->compress_type = compress_type;
5154 em->block_start = bytenr;
5155 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5158 bytenr += btrfs_file_extent_offset(leaf, item);
5159 em->block_start = bytenr;
5160 em->block_len = em->len;
5161 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5162 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5165 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5169 size_t extent_offset;
5172 em->block_start = EXTENT_MAP_INLINE;
5173 if (!page || create) {
5174 em->start = extent_start;
5175 em->len = extent_end - extent_start;
5179 size = btrfs_file_extent_inline_len(leaf, item);
5180 extent_offset = page_offset(page) + pg_offset - extent_start;
5181 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5182 size - extent_offset);
5183 em->start = extent_start + extent_offset;
5184 em->len = (copy_size + root->sectorsize - 1) &
5185 ~((u64)root->sectorsize - 1);
5186 em->orig_start = EXTENT_MAP_INLINE;
5187 if (compress_type) {
5188 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5189 em->compress_type = compress_type;
5191 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5192 if (create == 0 && !PageUptodate(page)) {
5193 if (btrfs_file_extent_compression(leaf, item) !=
5194 BTRFS_COMPRESS_NONE) {
5195 ret = uncompress_inline(path, inode, page,
5197 extent_offset, item);
5201 read_extent_buffer(leaf, map + pg_offset, ptr,
5203 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5204 memset(map + pg_offset + copy_size, 0,
5205 PAGE_CACHE_SIZE - pg_offset -
5210 flush_dcache_page(page);
5211 } else if (create && PageUptodate(page)) {
5215 free_extent_map(em);
5217 btrfs_release_path(root, path);
5218 trans = btrfs_join_transaction(root, 1);
5220 return ERR_CAST(trans);
5224 write_extent_buffer(leaf, map + pg_offset, ptr,
5227 btrfs_mark_buffer_dirty(leaf);
5229 set_extent_uptodate(io_tree, em->start,
5230 extent_map_end(em) - 1, GFP_NOFS);
5233 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5240 em->block_start = EXTENT_MAP_HOLE;
5241 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5243 btrfs_release_path(root, path);
5244 if (em->start > start || extent_map_end(em) <= start) {
5245 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5246 "[%llu %llu]\n", (unsigned long long)em->start,
5247 (unsigned long long)em->len,
5248 (unsigned long long)start,
5249 (unsigned long long)len);
5255 write_lock(&em_tree->lock);
5256 ret = add_extent_mapping(em_tree, em);
5257 /* it is possible that someone inserted the extent into the tree
5258 * while we had the lock dropped. It is also possible that
5259 * an overlapping map exists in the tree
5261 if (ret == -EEXIST) {
5262 struct extent_map *existing;
5266 existing = lookup_extent_mapping(em_tree, start, len);
5267 if (existing && (existing->start > start ||
5268 existing->start + existing->len <= start)) {
5269 free_extent_map(existing);
5273 existing = lookup_extent_mapping(em_tree, em->start,
5276 err = merge_extent_mapping(em_tree, existing,
5279 free_extent_map(existing);
5281 free_extent_map(em);
5286 free_extent_map(em);
5290 free_extent_map(em);
5295 write_unlock(&em_tree->lock);
5298 trace_btrfs_get_extent(root, em);
5301 btrfs_free_path(path);
5303 ret = btrfs_end_transaction(trans, root);
5308 free_extent_map(em);
5309 return ERR_PTR(err);
5314 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5315 size_t pg_offset, u64 start, u64 len,
5318 struct extent_map *em;
5319 struct extent_map *hole_em = NULL;
5320 u64 range_start = start;
5326 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5331 * if our em maps to a hole, there might
5332 * actually be delalloc bytes behind it
5334 if (em->block_start != EXTENT_MAP_HOLE)
5340 /* check to see if we've wrapped (len == -1 or similar) */
5349 /* ok, we didn't find anything, lets look for delalloc */
5350 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5351 end, len, EXTENT_DELALLOC, 1);
5352 found_end = range_start + found;
5353 if (found_end < range_start)
5354 found_end = (u64)-1;
5357 * we didn't find anything useful, return
5358 * the original results from get_extent()
5360 if (range_start > end || found_end <= start) {
5366 /* adjust the range_start to make sure it doesn't
5367 * go backwards from the start they passed in
5369 range_start = max(start,range_start);
5370 found = found_end - range_start;
5373 u64 hole_start = start;
5376 em = alloc_extent_map(GFP_NOFS);
5382 * when btrfs_get_extent can't find anything it
5383 * returns one huge hole
5385 * make sure what it found really fits our range, and
5386 * adjust to make sure it is based on the start from
5390 u64 calc_end = extent_map_end(hole_em);
5392 if (calc_end <= start || (hole_em->start > end)) {
5393 free_extent_map(hole_em);
5396 hole_start = max(hole_em->start, start);
5397 hole_len = calc_end - hole_start;
5401 if (hole_em && range_start > hole_start) {
5402 /* our hole starts before our delalloc, so we
5403 * have to return just the parts of the hole
5404 * that go until the delalloc starts
5406 em->len = min(hole_len,
5407 range_start - hole_start);
5408 em->start = hole_start;
5409 em->orig_start = hole_start;
5411 * don't adjust block start at all,
5412 * it is fixed at EXTENT_MAP_HOLE
5414 em->block_start = hole_em->block_start;
5415 em->block_len = hole_len;
5417 em->start = range_start;
5419 em->orig_start = range_start;
5420 em->block_start = EXTENT_MAP_DELALLOC;
5421 em->block_len = found;
5423 } else if (hole_em) {
5428 free_extent_map(hole_em);
5430 free_extent_map(em);
5431 return ERR_PTR(err);
5436 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5439 struct btrfs_root *root = BTRFS_I(inode)->root;
5440 struct btrfs_trans_handle *trans;
5441 struct extent_map *em;
5442 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5443 struct btrfs_key ins;
5447 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5449 trans = btrfs_join_transaction(root, 0);
5451 return ERR_CAST(trans);
5453 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5455 alloc_hint = get_extent_allocation_hint(inode, start, len);
5456 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5457 alloc_hint, (u64)-1, &ins, 1);
5463 em = alloc_extent_map(GFP_NOFS);
5465 em = ERR_PTR(-ENOMEM);
5470 em->orig_start = em->start;
5471 em->len = ins.offset;
5473 em->block_start = ins.objectid;
5474 em->block_len = ins.offset;
5475 em->bdev = root->fs_info->fs_devices->latest_bdev;
5476 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5479 write_lock(&em_tree->lock);
5480 ret = add_extent_mapping(em_tree, em);
5481 write_unlock(&em_tree->lock);
5484 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5487 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5488 ins.offset, ins.offset, 0);
5490 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5494 btrfs_end_transaction(trans, root);
5499 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5500 * block must be cow'd
5502 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5503 struct inode *inode, u64 offset, u64 len)
5505 struct btrfs_path *path;
5507 struct extent_buffer *leaf;
5508 struct btrfs_root *root = BTRFS_I(inode)->root;
5509 struct btrfs_file_extent_item *fi;
5510 struct btrfs_key key;
5518 path = btrfs_alloc_path();
5522 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5527 slot = path->slots[0];
5530 /* can't find the item, must cow */
5537 leaf = path->nodes[0];
5538 btrfs_item_key_to_cpu(leaf, &key, slot);
5539 if (key.objectid != inode->i_ino ||
5540 key.type != BTRFS_EXTENT_DATA_KEY) {
5541 /* not our file or wrong item type, must cow */
5545 if (key.offset > offset) {
5546 /* Wrong offset, must cow */
5550 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5551 found_type = btrfs_file_extent_type(leaf, fi);
5552 if (found_type != BTRFS_FILE_EXTENT_REG &&
5553 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5554 /* not a regular extent, must cow */
5557 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5558 backref_offset = btrfs_file_extent_offset(leaf, fi);
5560 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5561 if (extent_end < offset + len) {
5562 /* extent doesn't include our full range, must cow */
5566 if (btrfs_extent_readonly(root, disk_bytenr))
5570 * look for other files referencing this extent, if we
5571 * find any we must cow
5573 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5574 key.offset - backref_offset, disk_bytenr))
5578 * adjust disk_bytenr and num_bytes to cover just the bytes
5579 * in this extent we are about to write. If there
5580 * are any csums in that range we have to cow in order
5581 * to keep the csums correct
5583 disk_bytenr += backref_offset;
5584 disk_bytenr += offset - key.offset;
5585 num_bytes = min(offset + len, extent_end) - offset;
5586 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5589 * all of the above have passed, it is safe to overwrite this extent
5594 btrfs_free_path(path);
5598 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5599 struct buffer_head *bh_result, int create)
5601 struct extent_map *em;
5602 struct btrfs_root *root = BTRFS_I(inode)->root;
5603 u64 start = iblock << inode->i_blkbits;
5604 u64 len = bh_result->b_size;
5605 struct btrfs_trans_handle *trans;
5607 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5612 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5613 * io. INLINE is special, and we could probably kludge it in here, but
5614 * it's still buffered so for safety lets just fall back to the generic
5617 * For COMPRESSED we _have_ to read the entire extent in so we can
5618 * decompress it, so there will be buffering required no matter what we
5619 * do, so go ahead and fallback to buffered.
5621 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5622 * to buffered IO. Don't blame me, this is the price we pay for using
5625 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5626 em->block_start == EXTENT_MAP_INLINE) {
5627 free_extent_map(em);
5631 /* Just a good old fashioned hole, return */
5632 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5633 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5634 free_extent_map(em);
5635 /* DIO will do one hole at a time, so just unlock a sector */
5636 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5637 start + root->sectorsize - 1, GFP_NOFS);
5642 * We don't allocate a new extent in the following cases
5644 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5646 * 2) The extent is marked as PREALLOC. We're good to go here and can
5647 * just use the extent.
5651 len = em->len - (start - em->start);
5655 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5656 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5657 em->block_start != EXTENT_MAP_HOLE)) {
5662 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5663 type = BTRFS_ORDERED_PREALLOC;
5665 type = BTRFS_ORDERED_NOCOW;
5666 len = min(len, em->len - (start - em->start));
5667 block_start = em->block_start + (start - em->start);
5670 * we're not going to log anything, but we do need
5671 * to make sure the current transaction stays open
5672 * while we look for nocow cross refs
5674 trans = btrfs_join_transaction(root, 0);
5678 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5679 ret = btrfs_add_ordered_extent_dio(inode, start,
5680 block_start, len, len, type);
5681 btrfs_end_transaction(trans, root);
5683 free_extent_map(em);
5688 btrfs_end_transaction(trans, root);
5692 * this will cow the extent, reset the len in case we changed
5695 len = bh_result->b_size;
5696 free_extent_map(em);
5697 em = btrfs_new_extent_direct(inode, start, len);
5700 len = min(len, em->len - (start - em->start));
5702 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5703 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5706 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5708 bh_result->b_size = len;
5709 bh_result->b_bdev = em->bdev;
5710 set_buffer_mapped(bh_result);
5711 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5712 set_buffer_new(bh_result);
5714 free_extent_map(em);
5719 struct btrfs_dio_private {
5720 struct inode *inode;
5727 /* number of bios pending for this dio */
5728 atomic_t pending_bios;
5733 struct bio *orig_bio;
5736 static void btrfs_endio_direct_read(struct bio *bio, int err)
5738 struct btrfs_dio_private *dip = bio->bi_private;
5739 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5740 struct bio_vec *bvec = bio->bi_io_vec;
5741 struct inode *inode = dip->inode;
5742 struct btrfs_root *root = BTRFS_I(inode)->root;
5744 u32 *private = dip->csums;
5746 start = dip->logical_offset;
5748 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5749 struct page *page = bvec->bv_page;
5752 unsigned long flags;
5754 local_irq_save(flags);
5755 kaddr = kmap_atomic(page, KM_IRQ0);
5756 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5757 csum, bvec->bv_len);
5758 btrfs_csum_final(csum, (char *)&csum);
5759 kunmap_atomic(kaddr, KM_IRQ0);
5760 local_irq_restore(flags);
5762 flush_dcache_page(bvec->bv_page);
5763 if (csum != *private) {
5764 printk(KERN_ERR "btrfs csum failed ino %lu off"
5765 " %llu csum %u private %u\n",
5766 inode->i_ino, (unsigned long long)start,
5772 start += bvec->bv_len;
5775 } while (bvec <= bvec_end);
5777 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5778 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5779 bio->bi_private = dip->private;
5784 /* If we had a csum failure make sure to clear the uptodate flag */
5786 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5787 dio_end_io(bio, err);
5790 static void btrfs_endio_direct_write(struct bio *bio, int err)
5792 struct btrfs_dio_private *dip = bio->bi_private;
5793 struct inode *inode = dip->inode;
5794 struct btrfs_root *root = BTRFS_I(inode)->root;
5795 struct btrfs_trans_handle *trans;
5796 struct btrfs_ordered_extent *ordered = NULL;
5797 struct extent_state *cached_state = NULL;
5798 u64 ordered_offset = dip->logical_offset;
5799 u64 ordered_bytes = dip->bytes;
5805 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5813 trans = btrfs_join_transaction(root, 1);
5814 if (IS_ERR(trans)) {
5818 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5820 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5821 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5823 ret = btrfs_update_inode(trans, root, inode);
5828 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5829 ordered->file_offset + ordered->len - 1, 0,
5830 &cached_state, GFP_NOFS);
5832 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5833 ret = btrfs_mark_extent_written(trans, inode,
5834 ordered->file_offset,
5835 ordered->file_offset +
5842 ret = insert_reserved_file_extent(trans, inode,
5843 ordered->file_offset,
5849 BTRFS_FILE_EXTENT_REG);
5850 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5851 ordered->file_offset, ordered->len);
5859 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5860 btrfs_ordered_update_i_size(inode, 0, ordered);
5861 btrfs_update_inode(trans, root, inode);
5863 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5864 ordered->file_offset + ordered->len - 1,
5865 &cached_state, GFP_NOFS);
5867 btrfs_delalloc_release_metadata(inode, ordered->len);
5868 btrfs_end_transaction(trans, root);
5869 ordered_offset = ordered->file_offset + ordered->len;
5870 btrfs_put_ordered_extent(ordered);
5871 btrfs_put_ordered_extent(ordered);
5875 * our bio might span multiple ordered extents. If we haven't
5876 * completed the accounting for the whole dio, go back and try again
5878 if (ordered_offset < dip->logical_offset + dip->bytes) {
5879 ordered_bytes = dip->logical_offset + dip->bytes -
5884 bio->bi_private = dip->private;
5889 /* If we had an error make sure to clear the uptodate flag */
5891 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5892 dio_end_io(bio, err);
5895 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5896 struct bio *bio, int mirror_num,
5897 unsigned long bio_flags, u64 offset)
5900 struct btrfs_root *root = BTRFS_I(inode)->root;
5901 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5906 static void btrfs_end_dio_bio(struct bio *bio, int err)
5908 struct btrfs_dio_private *dip = bio->bi_private;
5911 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5912 "sector %#Lx len %u err no %d\n",
5913 dip->inode->i_ino, bio->bi_rw,
5914 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5918 * before atomic variable goto zero, we must make sure
5919 * dip->errors is perceived to be set.
5921 smp_mb__before_atomic_dec();
5924 /* if there are more bios still pending for this dio, just exit */
5925 if (!atomic_dec_and_test(&dip->pending_bios))
5929 bio_io_error(dip->orig_bio);
5931 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5932 bio_endio(dip->orig_bio, 0);
5938 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5939 u64 first_sector, gfp_t gfp_flags)
5941 int nr_vecs = bio_get_nr_vecs(bdev);
5942 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5945 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5946 int rw, u64 file_offset, int skip_sum,
5949 int write = rw & REQ_WRITE;
5950 struct btrfs_root *root = BTRFS_I(inode)->root;
5954 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5958 if (write && !skip_sum) {
5959 ret = btrfs_wq_submit_bio(root->fs_info,
5960 inode, rw, bio, 0, 0,
5962 __btrfs_submit_bio_start_direct_io,
5963 __btrfs_submit_bio_done);
5965 } else if (!skip_sum) {
5966 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5967 file_offset, csums);
5972 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5978 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5981 struct inode *inode = dip->inode;
5982 struct btrfs_root *root = BTRFS_I(inode)->root;
5983 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5985 struct bio *orig_bio = dip->orig_bio;
5986 struct bio_vec *bvec = orig_bio->bi_io_vec;
5987 u64 start_sector = orig_bio->bi_sector;
5988 u64 file_offset = dip->logical_offset;
5992 u32 *csums = dip->csums;
5994 int write = rw & REQ_WRITE;
5996 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5999 bio->bi_private = dip;
6000 bio->bi_end_io = btrfs_end_dio_bio;
6001 atomic_inc(&dip->pending_bios);
6003 map_length = orig_bio->bi_size;
6004 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6005 &map_length, NULL, 0);
6011 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6012 if (unlikely(map_length < submit_len + bvec->bv_len ||
6013 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6014 bvec->bv_offset) < bvec->bv_len)) {
6016 * inc the count before we submit the bio so
6017 * we know the end IO handler won't happen before
6018 * we inc the count. Otherwise, the dip might get freed
6019 * before we're done setting it up
6021 atomic_inc(&dip->pending_bios);
6022 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6023 file_offset, skip_sum,
6027 atomic_dec(&dip->pending_bios);
6031 /* Write's use the ordered csums */
6032 if (!write && !skip_sum)
6033 csums = csums + nr_pages;
6034 start_sector += submit_len >> 9;
6035 file_offset += submit_len;
6040 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6041 start_sector, GFP_NOFS);
6044 bio->bi_private = dip;
6045 bio->bi_end_io = btrfs_end_dio_bio;
6047 map_length = orig_bio->bi_size;
6048 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6049 &map_length, NULL, 0);
6055 submit_len += bvec->bv_len;
6061 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6070 * before atomic variable goto zero, we must
6071 * make sure dip->errors is perceived to be set.
6073 smp_mb__before_atomic_dec();
6074 if (atomic_dec_and_test(&dip->pending_bios))
6075 bio_io_error(dip->orig_bio);
6077 /* bio_end_io() will handle error, so we needn't return it */
6081 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6084 struct btrfs_root *root = BTRFS_I(inode)->root;
6085 struct btrfs_dio_private *dip;
6086 struct bio_vec *bvec = bio->bi_io_vec;
6088 int write = rw & REQ_WRITE;
6091 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6093 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6100 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6101 if (!write && !skip_sum) {
6102 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6110 dip->private = bio->bi_private;
6112 dip->logical_offset = file_offset;
6116 dip->bytes += bvec->bv_len;
6118 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6120 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6121 bio->bi_private = dip;
6123 dip->orig_bio = bio;
6124 atomic_set(&dip->pending_bios, 0);
6127 bio->bi_end_io = btrfs_endio_direct_write;
6129 bio->bi_end_io = btrfs_endio_direct_read;
6131 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6136 * If this is a write, we need to clean up the reserved space and kill
6137 * the ordered extent.
6140 struct btrfs_ordered_extent *ordered;
6141 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6142 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6143 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6144 btrfs_free_reserved_extent(root, ordered->start,
6146 btrfs_put_ordered_extent(ordered);
6147 btrfs_put_ordered_extent(ordered);
6149 bio_endio(bio, ret);
6152 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6153 const struct iovec *iov, loff_t offset,
6154 unsigned long nr_segs)
6159 unsigned blocksize_mask = root->sectorsize - 1;
6160 ssize_t retval = -EINVAL;
6161 loff_t end = offset;
6163 if (offset & blocksize_mask)
6166 /* Check the memory alignment. Blocks cannot straddle pages */
6167 for (seg = 0; seg < nr_segs; seg++) {
6168 addr = (unsigned long)iov[seg].iov_base;
6169 size = iov[seg].iov_len;
6171 if ((addr & blocksize_mask) || (size & blocksize_mask))
6178 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6179 const struct iovec *iov, loff_t offset,
6180 unsigned long nr_segs)
6182 struct file *file = iocb->ki_filp;
6183 struct inode *inode = file->f_mapping->host;
6184 struct btrfs_ordered_extent *ordered;
6185 struct extent_state *cached_state = NULL;
6186 u64 lockstart, lockend;
6188 int writing = rw & WRITE;
6190 size_t count = iov_length(iov, nr_segs);
6192 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6198 lockend = offset + count - 1;
6201 ret = btrfs_delalloc_reserve_space(inode, count);
6207 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6208 0, &cached_state, GFP_NOFS);
6210 * We're concerned with the entire range that we're going to be
6211 * doing DIO to, so we need to make sure theres no ordered
6212 * extents in this range.
6214 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6215 lockend - lockstart + 1);
6218 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6219 &cached_state, GFP_NOFS);
6220 btrfs_start_ordered_extent(inode, ordered, 1);
6221 btrfs_put_ordered_extent(ordered);
6226 * we don't use btrfs_set_extent_delalloc because we don't want
6227 * the dirty or uptodate bits
6230 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6231 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6232 EXTENT_DELALLOC, 0, NULL, &cached_state,
6235 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6236 lockend, EXTENT_LOCKED | write_bits,
6237 1, 0, &cached_state, GFP_NOFS);
6242 free_extent_state(cached_state);
6243 cached_state = NULL;
6245 ret = __blockdev_direct_IO(rw, iocb, inode,
6246 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6247 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6248 btrfs_submit_direct, 0);
6250 if (ret < 0 && ret != -EIOCBQUEUED) {
6251 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6252 offset + iov_length(iov, nr_segs) - 1,
6253 EXTENT_LOCKED | write_bits, 1, 0,
6254 &cached_state, GFP_NOFS);
6255 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6257 * We're falling back to buffered, unlock the section we didn't
6260 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6261 offset + iov_length(iov, nr_segs) - 1,
6262 EXTENT_LOCKED | write_bits, 1, 0,
6263 &cached_state, GFP_NOFS);
6266 free_extent_state(cached_state);
6270 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6271 __u64 start, __u64 len)
6273 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6276 int btrfs_readpage(struct file *file, struct page *page)
6278 struct extent_io_tree *tree;
6279 tree = &BTRFS_I(page->mapping->host)->io_tree;
6280 return extent_read_full_page(tree, page, btrfs_get_extent);
6283 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6285 struct extent_io_tree *tree;
6288 if (current->flags & PF_MEMALLOC) {
6289 redirty_page_for_writepage(wbc, page);
6293 tree = &BTRFS_I(page->mapping->host)->io_tree;
6294 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6297 int btrfs_writepages(struct address_space *mapping,
6298 struct writeback_control *wbc)
6300 struct extent_io_tree *tree;
6302 tree = &BTRFS_I(mapping->host)->io_tree;
6303 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6307 btrfs_readpages(struct file *file, struct address_space *mapping,
6308 struct list_head *pages, unsigned nr_pages)
6310 struct extent_io_tree *tree;
6311 tree = &BTRFS_I(mapping->host)->io_tree;
6312 return extent_readpages(tree, mapping, pages, nr_pages,
6315 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6317 struct extent_io_tree *tree;
6318 struct extent_map_tree *map;
6321 tree = &BTRFS_I(page->mapping->host)->io_tree;
6322 map = &BTRFS_I(page->mapping->host)->extent_tree;
6323 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6325 ClearPagePrivate(page);
6326 set_page_private(page, 0);
6327 page_cache_release(page);
6332 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6334 if (PageWriteback(page) || PageDirty(page))
6336 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6339 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6341 struct extent_io_tree *tree;
6342 struct btrfs_ordered_extent *ordered;
6343 struct extent_state *cached_state = NULL;
6344 u64 page_start = page_offset(page);
6345 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6349 * we have the page locked, so new writeback can't start,
6350 * and the dirty bit won't be cleared while we are here.
6352 * Wait for IO on this page so that we can safely clear
6353 * the PagePrivate2 bit and do ordered accounting
6355 wait_on_page_writeback(page);
6357 tree = &BTRFS_I(page->mapping->host)->io_tree;
6359 btrfs_releasepage(page, GFP_NOFS);
6362 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6364 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6368 * IO on this page will never be started, so we need
6369 * to account for any ordered extents now
6371 clear_extent_bit(tree, page_start, page_end,
6372 EXTENT_DIRTY | EXTENT_DELALLOC |
6373 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6374 &cached_state, GFP_NOFS);
6376 * whoever cleared the private bit is responsible
6377 * for the finish_ordered_io
6379 if (TestClearPagePrivate2(page)) {
6380 btrfs_finish_ordered_io(page->mapping->host,
6381 page_start, page_end);
6383 btrfs_put_ordered_extent(ordered);
6384 cached_state = NULL;
6385 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6388 clear_extent_bit(tree, page_start, page_end,
6389 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6390 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6391 __btrfs_releasepage(page, GFP_NOFS);
6393 ClearPageChecked(page);
6394 if (PagePrivate(page)) {
6395 ClearPagePrivate(page);
6396 set_page_private(page, 0);
6397 page_cache_release(page);
6402 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6403 * called from a page fault handler when a page is first dirtied. Hence we must
6404 * be careful to check for EOF conditions here. We set the page up correctly
6405 * for a written page which means we get ENOSPC checking when writing into
6406 * holes and correct delalloc and unwritten extent mapping on filesystems that
6407 * support these features.
6409 * We are not allowed to take the i_mutex here so we have to play games to
6410 * protect against truncate races as the page could now be beyond EOF. Because
6411 * vmtruncate() writes the inode size before removing pages, once we have the
6412 * page lock we can determine safely if the page is beyond EOF. If it is not
6413 * beyond EOF, then the page is guaranteed safe against truncation until we
6416 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6418 struct page *page = vmf->page;
6419 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6420 struct btrfs_root *root = BTRFS_I(inode)->root;
6421 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6422 struct btrfs_ordered_extent *ordered;
6423 struct extent_state *cached_state = NULL;
6425 unsigned long zero_start;
6431 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6435 else /* -ENOSPC, -EIO, etc */
6436 ret = VM_FAULT_SIGBUS;
6440 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6443 size = i_size_read(inode);
6444 page_start = page_offset(page);
6445 page_end = page_start + PAGE_CACHE_SIZE - 1;
6447 if ((page->mapping != inode->i_mapping) ||
6448 (page_start >= size)) {
6449 /* page got truncated out from underneath us */
6452 wait_on_page_writeback(page);
6454 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6456 set_page_extent_mapped(page);
6459 * we can't set the delalloc bits if there are pending ordered
6460 * extents. Drop our locks and wait for them to finish
6462 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6464 unlock_extent_cached(io_tree, page_start, page_end,
6465 &cached_state, GFP_NOFS);
6467 btrfs_start_ordered_extent(inode, ordered, 1);
6468 btrfs_put_ordered_extent(ordered);
6473 * XXX - page_mkwrite gets called every time the page is dirtied, even
6474 * if it was already dirty, so for space accounting reasons we need to
6475 * clear any delalloc bits for the range we are fixing to save. There
6476 * is probably a better way to do this, but for now keep consistent with
6477 * prepare_pages in the normal write path.
6479 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6480 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6481 0, 0, &cached_state, GFP_NOFS);
6483 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6486 unlock_extent_cached(io_tree, page_start, page_end,
6487 &cached_state, GFP_NOFS);
6488 ret = VM_FAULT_SIGBUS;
6493 /* page is wholly or partially inside EOF */
6494 if (page_start + PAGE_CACHE_SIZE > size)
6495 zero_start = size & ~PAGE_CACHE_MASK;
6497 zero_start = PAGE_CACHE_SIZE;
6499 if (zero_start != PAGE_CACHE_SIZE) {
6501 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6502 flush_dcache_page(page);
6505 ClearPageChecked(page);
6506 set_page_dirty(page);
6507 SetPageUptodate(page);
6509 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6510 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6512 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6516 return VM_FAULT_LOCKED;
6518 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6523 static int btrfs_truncate(struct inode *inode)
6525 struct btrfs_root *root = BTRFS_I(inode)->root;
6528 struct btrfs_trans_handle *trans;
6530 u64 mask = root->sectorsize - 1;
6532 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6536 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6537 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6539 trans = btrfs_start_transaction(root, 5);
6541 return PTR_ERR(trans);
6543 btrfs_set_trans_block_group(trans, inode);
6545 ret = btrfs_orphan_add(trans, inode);
6547 btrfs_end_transaction(trans, root);
6551 nr = trans->blocks_used;
6552 btrfs_end_transaction(trans, root);
6553 btrfs_btree_balance_dirty(root, nr);
6555 /* Now start a transaction for the truncate */
6556 trans = btrfs_start_transaction(root, 0);
6558 return PTR_ERR(trans);
6559 btrfs_set_trans_block_group(trans, inode);
6560 trans->block_rsv = root->orphan_block_rsv;
6563 * setattr is responsible for setting the ordered_data_close flag,
6564 * but that is only tested during the last file release. That
6565 * could happen well after the next commit, leaving a great big
6566 * window where new writes may get lost if someone chooses to write
6567 * to this file after truncating to zero
6569 * The inode doesn't have any dirty data here, and so if we commit
6570 * this is a noop. If someone immediately starts writing to the inode
6571 * it is very likely we'll catch some of their writes in this
6572 * transaction, and the commit will find this file on the ordered
6573 * data list with good things to send down.
6575 * This is a best effort solution, there is still a window where
6576 * using truncate to replace the contents of the file will
6577 * end up with a zero length file after a crash.
6579 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6580 btrfs_add_ordered_operation(trans, root, inode);
6584 trans = btrfs_start_transaction(root, 0);
6586 return PTR_ERR(trans);
6587 btrfs_set_trans_block_group(trans, inode);
6588 trans->block_rsv = root->orphan_block_rsv;
6591 ret = btrfs_block_rsv_check(trans, root,
6592 root->orphan_block_rsv, 0, 5);
6593 if (ret == -EAGAIN) {
6594 ret = btrfs_commit_transaction(trans, root);
6604 ret = btrfs_truncate_inode_items(trans, root, inode,
6606 BTRFS_EXTENT_DATA_KEY);
6607 if (ret != -EAGAIN) {
6612 ret = btrfs_update_inode(trans, root, inode);
6618 nr = trans->blocks_used;
6619 btrfs_end_transaction(trans, root);
6621 btrfs_btree_balance_dirty(root, nr);
6624 if (ret == 0 && inode->i_nlink > 0) {
6625 ret = btrfs_orphan_del(trans, inode);
6628 } else if (ret && inode->i_nlink > 0) {
6630 * Failed to do the truncate, remove us from the in memory
6633 ret = btrfs_orphan_del(NULL, inode);
6636 ret = btrfs_update_inode(trans, root, inode);
6640 nr = trans->blocks_used;
6641 ret = btrfs_end_transaction_throttle(trans, root);
6644 btrfs_btree_balance_dirty(root, nr);
6650 * create a new subvolume directory/inode (helper for the ioctl).
6652 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6653 struct btrfs_root *new_root,
6654 u64 new_dirid, u64 alloc_hint)
6656 struct inode *inode;
6660 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6661 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6663 return PTR_ERR(inode);
6664 inode->i_op = &btrfs_dir_inode_operations;
6665 inode->i_fop = &btrfs_dir_file_operations;
6668 btrfs_i_size_write(inode, 0);
6670 err = btrfs_update_inode(trans, new_root, inode);
6677 /* helper function for file defrag and space balancing. This
6678 * forces readahead on a given range of bytes in an inode
6680 unsigned long btrfs_force_ra(struct address_space *mapping,
6681 struct file_ra_state *ra, struct file *file,
6682 pgoff_t offset, pgoff_t last_index)
6684 pgoff_t req_size = last_index - offset + 1;
6686 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6687 return offset + req_size;
6690 struct inode *btrfs_alloc_inode(struct super_block *sb)
6692 struct btrfs_inode *ei;
6693 struct inode *inode;
6695 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6700 ei->space_info = NULL;
6704 ei->last_sub_trans = 0;
6705 ei->logged_trans = 0;
6706 ei->delalloc_bytes = 0;
6707 ei->reserved_bytes = 0;
6708 ei->disk_i_size = 0;
6710 ei->index_cnt = (u64)-1;
6711 ei->last_unlink_trans = 0;
6713 atomic_set(&ei->outstanding_extents, 0);
6714 atomic_set(&ei->reserved_extents, 0);
6716 ei->ordered_data_close = 0;
6717 ei->orphan_meta_reserved = 0;
6718 ei->dummy_inode = 0;
6719 ei->force_compress = BTRFS_COMPRESS_NONE;
6721 inode = &ei->vfs_inode;
6722 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6723 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6724 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6725 mutex_init(&ei->log_mutex);
6726 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6727 INIT_LIST_HEAD(&ei->i_orphan);
6728 INIT_LIST_HEAD(&ei->delalloc_inodes);
6729 INIT_LIST_HEAD(&ei->ordered_operations);
6730 RB_CLEAR_NODE(&ei->rb_node);
6735 static void btrfs_i_callback(struct rcu_head *head)
6737 struct inode *inode = container_of(head, struct inode, i_rcu);
6738 INIT_LIST_HEAD(&inode->i_dentry);
6739 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6742 void btrfs_destroy_inode(struct inode *inode)
6744 struct btrfs_ordered_extent *ordered;
6745 struct btrfs_root *root = BTRFS_I(inode)->root;
6747 WARN_ON(!list_empty(&inode->i_dentry));
6748 WARN_ON(inode->i_data.nrpages);
6749 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6750 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6753 * This can happen where we create an inode, but somebody else also
6754 * created the same inode and we need to destroy the one we already
6761 * Make sure we're properly removed from the ordered operation
6765 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6766 spin_lock(&root->fs_info->ordered_extent_lock);
6767 list_del_init(&BTRFS_I(inode)->ordered_operations);
6768 spin_unlock(&root->fs_info->ordered_extent_lock);
6771 if (root == root->fs_info->tree_root) {
6772 struct btrfs_block_group_cache *block_group;
6774 block_group = btrfs_lookup_block_group(root->fs_info,
6775 BTRFS_I(inode)->block_group);
6776 if (block_group && block_group->inode == inode) {
6777 spin_lock(&block_group->lock);
6778 block_group->inode = NULL;
6779 spin_unlock(&block_group->lock);
6780 btrfs_put_block_group(block_group);
6781 } else if (block_group) {
6782 btrfs_put_block_group(block_group);
6786 spin_lock(&root->orphan_lock);
6787 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6788 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6790 list_del_init(&BTRFS_I(inode)->i_orphan);
6792 spin_unlock(&root->orphan_lock);
6795 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6799 printk(KERN_ERR "btrfs found ordered "
6800 "extent %llu %llu on inode cleanup\n",
6801 (unsigned long long)ordered->file_offset,
6802 (unsigned long long)ordered->len);
6803 btrfs_remove_ordered_extent(inode, ordered);
6804 btrfs_put_ordered_extent(ordered);
6805 btrfs_put_ordered_extent(ordered);
6808 inode_tree_del(inode);
6809 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6811 call_rcu(&inode->i_rcu, btrfs_i_callback);
6814 int btrfs_drop_inode(struct inode *inode)
6816 struct btrfs_root *root = BTRFS_I(inode)->root;
6818 if (btrfs_root_refs(&root->root_item) == 0 &&
6819 root != root->fs_info->tree_root)
6822 return generic_drop_inode(inode);
6825 static void init_once(void *foo)
6827 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6829 inode_init_once(&ei->vfs_inode);
6832 void btrfs_destroy_cachep(void)
6834 if (btrfs_inode_cachep)
6835 kmem_cache_destroy(btrfs_inode_cachep);
6836 if (btrfs_trans_handle_cachep)
6837 kmem_cache_destroy(btrfs_trans_handle_cachep);
6838 if (btrfs_transaction_cachep)
6839 kmem_cache_destroy(btrfs_transaction_cachep);
6840 if (btrfs_path_cachep)
6841 kmem_cache_destroy(btrfs_path_cachep);
6842 if (btrfs_free_space_cachep)
6843 kmem_cache_destroy(btrfs_free_space_cachep);
6846 int btrfs_init_cachep(void)
6848 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6849 sizeof(struct btrfs_inode), 0,
6850 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6851 if (!btrfs_inode_cachep)
6854 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6855 sizeof(struct btrfs_trans_handle), 0,
6856 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6857 if (!btrfs_trans_handle_cachep)
6860 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6861 sizeof(struct btrfs_transaction), 0,
6862 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6863 if (!btrfs_transaction_cachep)
6866 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6867 sizeof(struct btrfs_path), 0,
6868 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6869 if (!btrfs_path_cachep)
6872 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6873 sizeof(struct btrfs_free_space), 0,
6874 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6875 if (!btrfs_free_space_cachep)
6880 btrfs_destroy_cachep();
6884 static int btrfs_getattr(struct vfsmount *mnt,
6885 struct dentry *dentry, struct kstat *stat)
6887 struct inode *inode = dentry->d_inode;
6888 generic_fillattr(inode, stat);
6889 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6890 stat->blksize = PAGE_CACHE_SIZE;
6891 stat->blocks = (inode_get_bytes(inode) +
6892 BTRFS_I(inode)->delalloc_bytes) >> 9;
6897 * If a file is moved, it will inherit the cow and compression flags of the new
6900 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6902 struct btrfs_inode *b_dir = BTRFS_I(dir);
6903 struct btrfs_inode *b_inode = BTRFS_I(inode);
6905 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6906 b_inode->flags |= BTRFS_INODE_NODATACOW;
6908 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6910 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6911 b_inode->flags |= BTRFS_INODE_COMPRESS;
6913 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6916 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6917 struct inode *new_dir, struct dentry *new_dentry)
6919 struct btrfs_trans_handle *trans;
6920 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6921 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6922 struct inode *new_inode = new_dentry->d_inode;
6923 struct inode *old_inode = old_dentry->d_inode;
6924 struct timespec ctime = CURRENT_TIME;
6929 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6932 /* we only allow rename subvolume link between subvolumes */
6933 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6936 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6937 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6940 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6941 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6944 * we're using rename to replace one file with another.
6945 * and the replacement file is large. Start IO on it now so
6946 * we don't add too much work to the end of the transaction
6948 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6949 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6950 filemap_flush(old_inode->i_mapping);
6952 /* close the racy window with snapshot create/destroy ioctl */
6953 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6954 down_read(&root->fs_info->subvol_sem);
6956 * We want to reserve the absolute worst case amount of items. So if
6957 * both inodes are subvols and we need to unlink them then that would
6958 * require 4 item modifications, but if they are both normal inodes it
6959 * would require 5 item modifications, so we'll assume their normal
6960 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6961 * should cover the worst case number of items we'll modify.
6963 trans = btrfs_start_transaction(root, 20);
6965 return PTR_ERR(trans);
6967 btrfs_set_trans_block_group(trans, new_dir);
6970 btrfs_record_root_in_trans(trans, dest);
6972 ret = btrfs_set_inode_index(new_dir, &index);
6976 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6977 /* force full log commit if subvolume involved. */
6978 root->fs_info->last_trans_log_full_commit = trans->transid;
6980 ret = btrfs_insert_inode_ref(trans, dest,
6981 new_dentry->d_name.name,
6982 new_dentry->d_name.len,
6984 new_dir->i_ino, index);
6988 * this is an ugly little race, but the rename is required
6989 * to make sure that if we crash, the inode is either at the
6990 * old name or the new one. pinning the log transaction lets
6991 * us make sure we don't allow a log commit to come in after
6992 * we unlink the name but before we add the new name back in.
6994 btrfs_pin_log_trans(root);
6997 * make sure the inode gets flushed if it is replacing
7000 if (new_inode && new_inode->i_size &&
7001 old_inode && S_ISREG(old_inode->i_mode)) {
7002 btrfs_add_ordered_operation(trans, root, old_inode);
7005 old_dir->i_ctime = old_dir->i_mtime = ctime;
7006 new_dir->i_ctime = new_dir->i_mtime = ctime;
7007 old_inode->i_ctime = ctime;
7009 if (old_dentry->d_parent != new_dentry->d_parent)
7010 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7012 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7013 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7014 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7015 old_dentry->d_name.name,
7016 old_dentry->d_name.len);
7018 ret = __btrfs_unlink_inode(trans, root, old_dir,
7019 old_dentry->d_inode,
7020 old_dentry->d_name.name,
7021 old_dentry->d_name.len);
7023 ret = btrfs_update_inode(trans, root, old_inode);
7028 new_inode->i_ctime = CURRENT_TIME;
7029 if (unlikely(new_inode->i_ino ==
7030 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7031 root_objectid = BTRFS_I(new_inode)->location.objectid;
7032 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7034 new_dentry->d_name.name,
7035 new_dentry->d_name.len);
7036 BUG_ON(new_inode->i_nlink == 0);
7038 ret = btrfs_unlink_inode(trans, dest, new_dir,
7039 new_dentry->d_inode,
7040 new_dentry->d_name.name,
7041 new_dentry->d_name.len);
7044 if (new_inode->i_nlink == 0) {
7045 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7050 fixup_inode_flags(new_dir, old_inode);
7052 ret = btrfs_add_link(trans, new_dir, old_inode,
7053 new_dentry->d_name.name,
7054 new_dentry->d_name.len, 0, index);
7057 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7058 struct dentry *parent = dget_parent(new_dentry);
7059 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7061 btrfs_end_log_trans(root);
7064 btrfs_end_transaction_throttle(trans, root);
7066 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7067 up_read(&root->fs_info->subvol_sem);
7073 * some fairly slow code that needs optimization. This walks the list
7074 * of all the inodes with pending delalloc and forces them to disk.
7076 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7078 struct list_head *head = &root->fs_info->delalloc_inodes;
7079 struct btrfs_inode *binode;
7080 struct inode *inode;
7082 if (root->fs_info->sb->s_flags & MS_RDONLY)
7085 spin_lock(&root->fs_info->delalloc_lock);
7086 while (!list_empty(head)) {
7087 binode = list_entry(head->next, struct btrfs_inode,
7089 inode = igrab(&binode->vfs_inode);
7091 list_del_init(&binode->delalloc_inodes);
7092 spin_unlock(&root->fs_info->delalloc_lock);
7094 filemap_flush(inode->i_mapping);
7096 btrfs_add_delayed_iput(inode);
7101 spin_lock(&root->fs_info->delalloc_lock);
7103 spin_unlock(&root->fs_info->delalloc_lock);
7105 /* the filemap_flush will queue IO into the worker threads, but
7106 * we have to make sure the IO is actually started and that
7107 * ordered extents get created before we return
7109 atomic_inc(&root->fs_info->async_submit_draining);
7110 while (atomic_read(&root->fs_info->nr_async_submits) ||
7111 atomic_read(&root->fs_info->async_delalloc_pages)) {
7112 wait_event(root->fs_info->async_submit_wait,
7113 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7114 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7116 atomic_dec(&root->fs_info->async_submit_draining);
7120 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7123 struct btrfs_inode *binode;
7124 struct inode *inode = NULL;
7126 spin_lock(&root->fs_info->delalloc_lock);
7127 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7128 binode = list_entry(root->fs_info->delalloc_inodes.next,
7129 struct btrfs_inode, delalloc_inodes);
7130 inode = igrab(&binode->vfs_inode);
7132 list_move_tail(&binode->delalloc_inodes,
7133 &root->fs_info->delalloc_inodes);
7137 list_del_init(&binode->delalloc_inodes);
7138 cond_resched_lock(&root->fs_info->delalloc_lock);
7140 spin_unlock(&root->fs_info->delalloc_lock);
7144 filemap_write_and_wait(inode->i_mapping);
7146 * We have to do this because compression doesn't
7147 * actually set PG_writeback until it submits the pages
7148 * for IO, which happens in an async thread, so we could
7149 * race and not actually wait for any writeback pages
7150 * because they've not been submitted yet. Technically
7151 * this could still be the case for the ordered stuff
7152 * since the async thread may not have started to do its
7153 * work yet. If this becomes the case then we need to
7154 * figure out a way to make sure that in writepage we
7155 * wait for any async pages to be submitted before
7156 * returning so that fdatawait does what its supposed to
7159 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7161 filemap_flush(inode->i_mapping);
7164 btrfs_add_delayed_iput(inode);
7172 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7173 const char *symname)
7175 struct btrfs_trans_handle *trans;
7176 struct btrfs_root *root = BTRFS_I(dir)->root;
7177 struct btrfs_path *path;
7178 struct btrfs_key key;
7179 struct inode *inode = NULL;
7187 struct btrfs_file_extent_item *ei;
7188 struct extent_buffer *leaf;
7189 unsigned long nr = 0;
7191 name_len = strlen(symname) + 1;
7192 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7193 return -ENAMETOOLONG;
7195 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7199 * 2 items for inode item and ref
7200 * 2 items for dir items
7201 * 1 item for xattr if selinux is on
7203 trans = btrfs_start_transaction(root, 5);
7205 return PTR_ERR(trans);
7207 btrfs_set_trans_block_group(trans, dir);
7209 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7210 dentry->d_name.len, dir->i_ino, objectid,
7211 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7213 err = PTR_ERR(inode);
7217 err = btrfs_init_inode_security(trans, inode, dir);
7223 btrfs_set_trans_block_group(trans, inode);
7224 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7228 inode->i_mapping->a_ops = &btrfs_aops;
7229 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7230 inode->i_fop = &btrfs_file_operations;
7231 inode->i_op = &btrfs_file_inode_operations;
7232 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7234 btrfs_update_inode_block_group(trans, inode);
7235 btrfs_update_inode_block_group(trans, dir);
7239 path = btrfs_alloc_path();
7241 key.objectid = inode->i_ino;
7243 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7244 datasize = btrfs_file_extent_calc_inline_size(name_len);
7245 err = btrfs_insert_empty_item(trans, root, path, &key,
7251 leaf = path->nodes[0];
7252 ei = btrfs_item_ptr(leaf, path->slots[0],
7253 struct btrfs_file_extent_item);
7254 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7255 btrfs_set_file_extent_type(leaf, ei,
7256 BTRFS_FILE_EXTENT_INLINE);
7257 btrfs_set_file_extent_encryption(leaf, ei, 0);
7258 btrfs_set_file_extent_compression(leaf, ei, 0);
7259 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7260 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7262 ptr = btrfs_file_extent_inline_start(ei);
7263 write_extent_buffer(leaf, symname, ptr, name_len);
7264 btrfs_mark_buffer_dirty(leaf);
7265 btrfs_free_path(path);
7267 inode->i_op = &btrfs_symlink_inode_operations;
7268 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7269 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7270 inode_set_bytes(inode, name_len);
7271 btrfs_i_size_write(inode, name_len - 1);
7272 err = btrfs_update_inode(trans, root, inode);
7277 nr = trans->blocks_used;
7278 btrfs_end_transaction_throttle(trans, root);
7280 inode_dec_link_count(inode);
7283 btrfs_btree_balance_dirty(root, nr);
7287 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7288 u64 start, u64 num_bytes, u64 min_size,
7289 loff_t actual_len, u64 *alloc_hint,
7290 struct btrfs_trans_handle *trans)
7292 struct btrfs_root *root = BTRFS_I(inode)->root;
7293 struct btrfs_key ins;
7294 u64 cur_offset = start;
7297 bool own_trans = true;
7301 while (num_bytes > 0) {
7303 trans = btrfs_start_transaction(root, 3);
7304 if (IS_ERR(trans)) {
7305 ret = PTR_ERR(trans);
7310 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7311 0, *alloc_hint, (u64)-1, &ins, 1);
7314 btrfs_end_transaction(trans, root);
7318 ret = insert_reserved_file_extent(trans, inode,
7319 cur_offset, ins.objectid,
7320 ins.offset, ins.offset,
7321 ins.offset, 0, 0, 0,
7322 BTRFS_FILE_EXTENT_PREALLOC);
7324 btrfs_drop_extent_cache(inode, cur_offset,
7325 cur_offset + ins.offset -1, 0);
7327 num_bytes -= ins.offset;
7328 cur_offset += ins.offset;
7329 *alloc_hint = ins.objectid + ins.offset;
7331 inode->i_ctime = CURRENT_TIME;
7332 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7333 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7334 (actual_len > inode->i_size) &&
7335 (cur_offset > inode->i_size)) {
7336 if (cur_offset > actual_len)
7337 i_size = actual_len;
7339 i_size = cur_offset;
7340 i_size_write(inode, i_size);
7341 btrfs_ordered_update_i_size(inode, i_size, NULL);
7344 ret = btrfs_update_inode(trans, root, inode);
7348 btrfs_end_transaction(trans, root);
7353 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7354 u64 start, u64 num_bytes, u64 min_size,
7355 loff_t actual_len, u64 *alloc_hint)
7357 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7358 min_size, actual_len, alloc_hint,
7362 int btrfs_prealloc_file_range_trans(struct inode *inode,
7363 struct btrfs_trans_handle *trans, int mode,
7364 u64 start, u64 num_bytes, u64 min_size,
7365 loff_t actual_len, u64 *alloc_hint)
7367 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7368 min_size, actual_len, alloc_hint, trans);
7371 static int btrfs_set_page_dirty(struct page *page)
7373 return __set_page_dirty_nobuffers(page);
7376 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7378 struct btrfs_root *root = BTRFS_I(inode)->root;
7380 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7382 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7384 return generic_permission(inode, mask, flags, btrfs_check_acl);
7387 static const struct inode_operations btrfs_dir_inode_operations = {
7388 .getattr = btrfs_getattr,
7389 .lookup = btrfs_lookup,
7390 .create = btrfs_create,
7391 .unlink = btrfs_unlink,
7393 .mkdir = btrfs_mkdir,
7394 .rmdir = btrfs_rmdir,
7395 .rename = btrfs_rename,
7396 .symlink = btrfs_symlink,
7397 .setattr = btrfs_setattr,
7398 .mknod = btrfs_mknod,
7399 .setxattr = btrfs_setxattr,
7400 .getxattr = btrfs_getxattr,
7401 .listxattr = btrfs_listxattr,
7402 .removexattr = btrfs_removexattr,
7403 .permission = btrfs_permission,
7405 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7406 .lookup = btrfs_lookup,
7407 .permission = btrfs_permission,
7410 static const struct file_operations btrfs_dir_file_operations = {
7411 .llseek = generic_file_llseek,
7412 .read = generic_read_dir,
7413 .readdir = btrfs_real_readdir,
7414 .unlocked_ioctl = btrfs_ioctl,
7415 #ifdef CONFIG_COMPAT
7416 .compat_ioctl = btrfs_ioctl,
7418 .release = btrfs_release_file,
7419 .fsync = btrfs_sync_file,
7422 static struct extent_io_ops btrfs_extent_io_ops = {
7423 .fill_delalloc = run_delalloc_range,
7424 .submit_bio_hook = btrfs_submit_bio_hook,
7425 .merge_bio_hook = btrfs_merge_bio_hook,
7426 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7427 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7428 .writepage_start_hook = btrfs_writepage_start_hook,
7429 .readpage_io_failed_hook = btrfs_io_failed_hook,
7430 .set_bit_hook = btrfs_set_bit_hook,
7431 .clear_bit_hook = btrfs_clear_bit_hook,
7432 .merge_extent_hook = btrfs_merge_extent_hook,
7433 .split_extent_hook = btrfs_split_extent_hook,
7437 * btrfs doesn't support the bmap operation because swapfiles
7438 * use bmap to make a mapping of extents in the file. They assume
7439 * these extents won't change over the life of the file and they
7440 * use the bmap result to do IO directly to the drive.
7442 * the btrfs bmap call would return logical addresses that aren't
7443 * suitable for IO and they also will change frequently as COW
7444 * operations happen. So, swapfile + btrfs == corruption.
7446 * For now we're avoiding this by dropping bmap.
7448 static const struct address_space_operations btrfs_aops = {
7449 .readpage = btrfs_readpage,
7450 .writepage = btrfs_writepage,
7451 .writepages = btrfs_writepages,
7452 .readpages = btrfs_readpages,
7453 .sync_page = block_sync_page,
7454 .direct_IO = btrfs_direct_IO,
7455 .invalidatepage = btrfs_invalidatepage,
7456 .releasepage = btrfs_releasepage,
7457 .set_page_dirty = btrfs_set_page_dirty,
7458 .error_remove_page = generic_error_remove_page,
7461 static const struct address_space_operations btrfs_symlink_aops = {
7462 .readpage = btrfs_readpage,
7463 .writepage = btrfs_writepage,
7464 .invalidatepage = btrfs_invalidatepage,
7465 .releasepage = btrfs_releasepage,
7468 static const struct inode_operations btrfs_file_inode_operations = {
7469 .getattr = btrfs_getattr,
7470 .setattr = btrfs_setattr,
7471 .setxattr = btrfs_setxattr,
7472 .getxattr = btrfs_getxattr,
7473 .listxattr = btrfs_listxattr,
7474 .removexattr = btrfs_removexattr,
7475 .permission = btrfs_permission,
7476 .fiemap = btrfs_fiemap,
7478 static const struct inode_operations btrfs_special_inode_operations = {
7479 .getattr = btrfs_getattr,
7480 .setattr = btrfs_setattr,
7481 .permission = btrfs_permission,
7482 .setxattr = btrfs_setxattr,
7483 .getxattr = btrfs_getxattr,
7484 .listxattr = btrfs_listxattr,
7485 .removexattr = btrfs_removexattr,
7487 static const struct inode_operations btrfs_symlink_inode_operations = {
7488 .readlink = generic_readlink,
7489 .follow_link = page_follow_link_light,
7490 .put_link = page_put_link,
7491 .getattr = btrfs_getattr,
7492 .permission = btrfs_permission,
7493 .setxattr = btrfs_setxattr,
7494 .getxattr = btrfs_getxattr,
7495 .listxattr = btrfs_listxattr,
7496 .removexattr = btrfs_removexattr,
7499 const struct dentry_operations btrfs_dentry_operations = {
7500 .d_delete = btrfs_dentry_delete,