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,
97 const struct qstr *qstr)
101 err = btrfs_init_acl(trans, inode, dir);
103 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
108 * this does all the hard work for inserting an inline extent into
109 * the btree. The caller should have done a btrfs_drop_extents so that
110 * no overlapping inline items exist in the btree
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113 struct btrfs_root *root, struct inode *inode,
114 u64 start, size_t size, size_t compressed_size,
116 struct page **compressed_pages)
118 struct btrfs_key key;
119 struct btrfs_path *path;
120 struct extent_buffer *leaf;
121 struct page *page = NULL;
124 struct btrfs_file_extent_item *ei;
127 size_t cur_size = size;
129 unsigned long offset;
131 if (compressed_size && compressed_pages)
132 cur_size = compressed_size;
134 path = btrfs_alloc_path();
138 path->leave_spinning = 1;
139 btrfs_set_trans_block_group(trans, inode);
141 key.objectid = inode->i_ino;
143 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
144 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146 inode_add_bytes(inode, size);
147 ret = btrfs_insert_empty_item(trans, root, path, &key,
154 leaf = path->nodes[0];
155 ei = btrfs_item_ptr(leaf, path->slots[0],
156 struct btrfs_file_extent_item);
157 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
158 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
159 btrfs_set_file_extent_encryption(leaf, ei, 0);
160 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
161 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
162 ptr = btrfs_file_extent_inline_start(ei);
164 if (compress_type != BTRFS_COMPRESS_NONE) {
167 while (compressed_size > 0) {
168 cpage = compressed_pages[i];
169 cur_size = min_t(unsigned long, compressed_size,
172 kaddr = kmap_atomic(cpage, KM_USER0);
173 write_extent_buffer(leaf, kaddr, ptr, cur_size);
174 kunmap_atomic(kaddr, KM_USER0);
178 compressed_size -= cur_size;
180 btrfs_set_file_extent_compression(leaf, ei,
183 page = find_get_page(inode->i_mapping,
184 start >> PAGE_CACHE_SHIFT);
185 btrfs_set_file_extent_compression(leaf, ei, 0);
186 kaddr = kmap_atomic(page, KM_USER0);
187 offset = start & (PAGE_CACHE_SIZE - 1);
188 write_extent_buffer(leaf, kaddr + offset, ptr, size);
189 kunmap_atomic(kaddr, KM_USER0);
190 page_cache_release(page);
192 btrfs_mark_buffer_dirty(leaf);
193 btrfs_free_path(path);
196 * we're an inline extent, so nobody can
197 * extend the file past i_size without locking
198 * a page we already have locked.
200 * We must do any isize and inode updates
201 * before we unlock the pages. Otherwise we
202 * could end up racing with unlink.
204 BTRFS_I(inode)->disk_i_size = inode->i_size;
205 btrfs_update_inode(trans, root, inode);
209 btrfs_free_path(path);
215 * conditionally insert an inline extent into the file. This
216 * does the checks required to make sure the data is small enough
217 * to fit as an inline extent.
219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
220 struct btrfs_root *root,
221 struct inode *inode, u64 start, u64 end,
222 size_t compressed_size, int compress_type,
223 struct page **compressed_pages)
225 u64 isize = i_size_read(inode);
226 u64 actual_end = min(end + 1, isize);
227 u64 inline_len = actual_end - start;
228 u64 aligned_end = (end + root->sectorsize - 1) &
229 ~((u64)root->sectorsize - 1);
231 u64 data_len = inline_len;
235 data_len = compressed_size;
238 actual_end >= PAGE_CACHE_SIZE ||
239 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (actual_end & (root->sectorsize - 1)) == 0) ||
243 data_len > root->fs_info->max_inline) {
247 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
251 if (isize > actual_end)
252 inline_len = min_t(u64, isize, actual_end);
253 ret = insert_inline_extent(trans, root, inode, start,
254 inline_len, compressed_size,
255 compress_type, compressed_pages);
257 btrfs_delalloc_release_metadata(inode, end + 1 - start);
258 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
262 struct async_extent {
267 unsigned long nr_pages;
269 struct list_head list;
274 struct btrfs_root *root;
275 struct page *locked_page;
278 struct list_head extents;
279 struct btrfs_work work;
282 static noinline int add_async_extent(struct async_cow *cow,
283 u64 start, u64 ram_size,
286 unsigned long nr_pages,
289 struct async_extent *async_extent;
291 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
292 BUG_ON(!async_extent);
293 async_extent->start = start;
294 async_extent->ram_size = ram_size;
295 async_extent->compressed_size = compressed_size;
296 async_extent->pages = pages;
297 async_extent->nr_pages = nr_pages;
298 async_extent->compress_type = compress_type;
299 list_add_tail(&async_extent->list, &cow->extents);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline int compress_file_range(struct inode *inode,
320 struct page *locked_page,
322 struct async_cow *async_cow,
325 struct btrfs_root *root = BTRFS_I(inode)->root;
326 struct btrfs_trans_handle *trans;
328 u64 blocksize = root->sectorsize;
330 u64 isize = i_size_read(inode);
332 struct page **pages = NULL;
333 unsigned long nr_pages;
334 unsigned long nr_pages_ret = 0;
335 unsigned long total_compressed = 0;
336 unsigned long total_in = 0;
337 unsigned long max_compressed = 128 * 1024;
338 unsigned long max_uncompressed = 128 * 1024;
341 int compress_type = root->fs_info->compress_type;
343 actual_end = min_t(u64, isize, end + 1);
346 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end <= start)
360 goto cleanup_and_bail_uncompressed;
362 total_compressed = actual_end - start;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed = min(total_compressed, max_uncompressed);
375 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376 num_bytes = max(blocksize, num_bytes);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386 (btrfs_test_opt(root, COMPRESS) ||
387 (BTRFS_I(inode)->force_compress) ||
388 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
390 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
393 if (BTRFS_I(inode)->force_compress)
394 compress_type = BTRFS_I(inode)->force_compress;
396 ret = btrfs_compress_pages(compress_type,
397 inode->i_mapping, start,
398 total_compressed, pages,
399 nr_pages, &nr_pages_ret,
405 unsigned long offset = total_compressed &
406 (PAGE_CACHE_SIZE - 1);
407 struct page *page = pages[nr_pages_ret - 1];
410 /* zero the tail end of the last page, we might be
411 * sending it down to disk
414 kaddr = kmap_atomic(page, KM_USER0);
415 memset(kaddr + offset, 0,
416 PAGE_CACHE_SIZE - offset);
417 kunmap_atomic(kaddr, KM_USER0);
423 trans = btrfs_join_transaction(root, 1);
424 BUG_ON(IS_ERR(trans));
425 btrfs_set_trans_block_group(trans, inode);
426 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
428 /* lets try to make an inline extent */
429 if (ret || total_in < (actual_end - start)) {
430 /* we didn't compress the entire range, try
431 * to make an uncompressed inline extent.
433 ret = cow_file_range_inline(trans, root, inode,
434 start, end, 0, 0, NULL);
436 /* try making a compressed inline extent */
437 ret = cow_file_range_inline(trans, root, inode,
440 compress_type, pages);
444 * inline extent creation worked, we don't need
445 * to create any more async work items. Unlock
446 * and free up our temp pages.
448 extent_clear_unlock_delalloc(inode,
449 &BTRFS_I(inode)->io_tree,
451 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
452 EXTENT_CLEAR_DELALLOC |
453 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
455 btrfs_end_transaction(trans, root);
458 btrfs_end_transaction(trans, root);
463 * we aren't doing an inline extent round the compressed size
464 * up to a block size boundary so the allocator does sane
467 total_compressed = (total_compressed + blocksize - 1) &
471 * one last check to make sure the compression is really a
472 * win, compare the page count read with the blocks on disk
474 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
475 ~(PAGE_CACHE_SIZE - 1);
476 if (total_compressed >= total_in) {
479 num_bytes = total_in;
482 if (!will_compress && pages) {
484 * the compression code ran but failed to make things smaller,
485 * free any pages it allocated and our page pointer array
487 for (i = 0; i < nr_pages_ret; i++) {
488 WARN_ON(pages[i]->mapping);
489 page_cache_release(pages[i]);
493 total_compressed = 0;
496 /* flag the file so we don't compress in the future */
497 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
498 !(BTRFS_I(inode)->force_compress)) {
499 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
505 /* the async work queues will take care of doing actual
506 * allocation on disk for these compressed pages,
507 * and will submit them to the elevator.
509 add_async_extent(async_cow, start, num_bytes,
510 total_compressed, pages, nr_pages_ret,
513 if (start + num_bytes < end) {
520 cleanup_and_bail_uncompressed:
522 * No compression, but we still need to write the pages in
523 * the file we've been given so far. redirty the locked
524 * page if it corresponds to our extent and set things up
525 * for the async work queue to run cow_file_range to do
526 * the normal delalloc dance
528 if (page_offset(locked_page) >= start &&
529 page_offset(locked_page) <= end) {
530 __set_page_dirty_nobuffers(locked_page);
531 /* unlocked later on in the async handlers */
533 add_async_extent(async_cow, start, end - start + 1,
534 0, NULL, 0, BTRFS_COMPRESS_NONE);
542 for (i = 0; i < nr_pages_ret; i++) {
543 WARN_ON(pages[i]->mapping);
544 page_cache_release(pages[i]);
552 * phase two of compressed writeback. This is the ordered portion
553 * of the code, which only gets called in the order the work was
554 * queued. We walk all the async extents created by compress_file_range
555 * and send them down to the disk.
557 static noinline int submit_compressed_extents(struct inode *inode,
558 struct async_cow *async_cow)
560 struct async_extent *async_extent;
562 struct btrfs_trans_handle *trans;
563 struct btrfs_key ins;
564 struct extent_map *em;
565 struct btrfs_root *root = BTRFS_I(inode)->root;
566 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
567 struct extent_io_tree *io_tree;
570 if (list_empty(&async_cow->extents))
574 while (!list_empty(&async_cow->extents)) {
575 async_extent = list_entry(async_cow->extents.next,
576 struct async_extent, list);
577 list_del(&async_extent->list);
579 io_tree = &BTRFS_I(inode)->io_tree;
582 /* did the compression code fall back to uncompressed IO? */
583 if (!async_extent->pages) {
584 int page_started = 0;
585 unsigned long nr_written = 0;
587 lock_extent(io_tree, async_extent->start,
588 async_extent->start +
589 async_extent->ram_size - 1, GFP_NOFS);
591 /* allocate blocks */
592 ret = cow_file_range(inode, async_cow->locked_page,
594 async_extent->start +
595 async_extent->ram_size - 1,
596 &page_started, &nr_written, 0);
599 * if page_started, cow_file_range inserted an
600 * inline extent and took care of all the unlocking
601 * and IO for us. Otherwise, we need to submit
602 * all those pages down to the drive.
604 if (!page_started && !ret)
605 extent_write_locked_range(io_tree,
606 inode, async_extent->start,
607 async_extent->start +
608 async_extent->ram_size - 1,
616 lock_extent(io_tree, async_extent->start,
617 async_extent->start + async_extent->ram_size - 1,
620 trans = btrfs_join_transaction(root, 1);
621 BUG_ON(IS_ERR(trans));
622 ret = btrfs_reserve_extent(trans, root,
623 async_extent->compressed_size,
624 async_extent->compressed_size,
627 btrfs_end_transaction(trans, root);
631 for (i = 0; i < async_extent->nr_pages; i++) {
632 WARN_ON(async_extent->pages[i]->mapping);
633 page_cache_release(async_extent->pages[i]);
635 kfree(async_extent->pages);
636 async_extent->nr_pages = 0;
637 async_extent->pages = NULL;
638 unlock_extent(io_tree, async_extent->start,
639 async_extent->start +
640 async_extent->ram_size - 1, GFP_NOFS);
645 * here we're doing allocation and writeback of the
648 btrfs_drop_extent_cache(inode, async_extent->start,
649 async_extent->start +
650 async_extent->ram_size - 1, 0);
652 em = alloc_extent_map(GFP_NOFS);
654 em->start = async_extent->start;
655 em->len = async_extent->ram_size;
656 em->orig_start = em->start;
658 em->block_start = ins.objectid;
659 em->block_len = ins.offset;
660 em->bdev = root->fs_info->fs_devices->latest_bdev;
661 em->compress_type = async_extent->compress_type;
662 set_bit(EXTENT_FLAG_PINNED, &em->flags);
663 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
666 write_lock(&em_tree->lock);
667 ret = add_extent_mapping(em_tree, em);
668 write_unlock(&em_tree->lock);
669 if (ret != -EEXIST) {
673 btrfs_drop_extent_cache(inode, async_extent->start,
674 async_extent->start +
675 async_extent->ram_size - 1, 0);
678 ret = btrfs_add_ordered_extent_compress(inode,
681 async_extent->ram_size,
683 BTRFS_ORDERED_COMPRESSED,
684 async_extent->compress_type);
688 * clear dirty, set writeback and unlock the pages.
690 extent_clear_unlock_delalloc(inode,
691 &BTRFS_I(inode)->io_tree,
693 async_extent->start +
694 async_extent->ram_size - 1,
695 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
696 EXTENT_CLEAR_UNLOCK |
697 EXTENT_CLEAR_DELALLOC |
698 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
700 ret = btrfs_submit_compressed_write(inode,
702 async_extent->ram_size,
704 ins.offset, async_extent->pages,
705 async_extent->nr_pages);
708 alloc_hint = ins.objectid + ins.offset;
716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
719 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
720 struct extent_map *em;
723 read_lock(&em_tree->lock);
724 em = search_extent_mapping(em_tree, start, num_bytes);
727 * if block start isn't an actual block number then find the
728 * first block in this inode and use that as a hint. If that
729 * block is also bogus then just don't worry about it.
731 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
733 em = search_extent_mapping(em_tree, 0, 0);
734 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
735 alloc_hint = em->block_start;
739 alloc_hint = em->block_start;
743 read_unlock(&em_tree->lock);
749 * when extent_io.c finds a delayed allocation range in the file,
750 * the call backs end up in this code. The basic idea is to
751 * allocate extents on disk for the range, and create ordered data structs
752 * in ram to track those extents.
754 * locked_page is the page that writepage had locked already. We use
755 * it to make sure we don't do extra locks or unlocks.
757 * *page_started is set to one if we unlock locked_page and do everything
758 * required to start IO on it. It may be clean and already done with
761 static noinline int cow_file_range(struct inode *inode,
762 struct page *locked_page,
763 u64 start, u64 end, int *page_started,
764 unsigned long *nr_written,
767 struct btrfs_root *root = BTRFS_I(inode)->root;
768 struct btrfs_trans_handle *trans;
771 unsigned long ram_size;
774 u64 blocksize = root->sectorsize;
775 struct btrfs_key ins;
776 struct extent_map *em;
777 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
780 BUG_ON(root == root->fs_info->tree_root);
781 trans = btrfs_join_transaction(root, 1);
782 BUG_ON(IS_ERR(trans));
783 btrfs_set_trans_block_group(trans, inode);
784 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
786 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
787 num_bytes = max(blocksize, num_bytes);
788 disk_num_bytes = num_bytes;
792 /* lets try to make an inline extent */
793 ret = cow_file_range_inline(trans, root, inode,
794 start, end, 0, 0, NULL);
796 extent_clear_unlock_delalloc(inode,
797 &BTRFS_I(inode)->io_tree,
799 EXTENT_CLEAR_UNLOCK_PAGE |
800 EXTENT_CLEAR_UNLOCK |
801 EXTENT_CLEAR_DELALLOC |
803 EXTENT_SET_WRITEBACK |
804 EXTENT_END_WRITEBACK);
806 *nr_written = *nr_written +
807 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
814 BUG_ON(disk_num_bytes >
815 btrfs_super_total_bytes(&root->fs_info->super_copy));
817 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
818 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
820 while (disk_num_bytes > 0) {
823 cur_alloc_size = disk_num_bytes;
824 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
825 root->sectorsize, 0, alloc_hint,
829 em = alloc_extent_map(GFP_NOFS);
832 em->orig_start = em->start;
833 ram_size = ins.offset;
834 em->len = ins.offset;
836 em->block_start = ins.objectid;
837 em->block_len = ins.offset;
838 em->bdev = root->fs_info->fs_devices->latest_bdev;
839 set_bit(EXTENT_FLAG_PINNED, &em->flags);
842 write_lock(&em_tree->lock);
843 ret = add_extent_mapping(em_tree, em);
844 write_unlock(&em_tree->lock);
845 if (ret != -EEXIST) {
849 btrfs_drop_extent_cache(inode, start,
850 start + ram_size - 1, 0);
853 cur_alloc_size = ins.offset;
854 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
855 ram_size, cur_alloc_size, 0);
858 if (root->root_key.objectid ==
859 BTRFS_DATA_RELOC_TREE_OBJECTID) {
860 ret = btrfs_reloc_clone_csums(inode, start,
865 if (disk_num_bytes < cur_alloc_size)
868 /* we're not doing compressed IO, don't unlock the first
869 * page (which the caller expects to stay locked), don't
870 * clear any dirty bits and don't set any writeback bits
872 * Do set the Private2 bit so we know this page was properly
873 * setup for writepage
875 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
876 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
879 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
880 start, start + ram_size - 1,
882 disk_num_bytes -= cur_alloc_size;
883 num_bytes -= cur_alloc_size;
884 alloc_hint = ins.objectid + ins.offset;
885 start += cur_alloc_size;
889 btrfs_end_transaction(trans, root);
895 * work queue call back to started compression on a file and pages
897 static noinline void async_cow_start(struct btrfs_work *work)
899 struct async_cow *async_cow;
901 async_cow = container_of(work, struct async_cow, work);
903 compress_file_range(async_cow->inode, async_cow->locked_page,
904 async_cow->start, async_cow->end, async_cow,
907 async_cow->inode = NULL;
911 * work queue call back to submit previously compressed pages
913 static noinline void async_cow_submit(struct btrfs_work *work)
915 struct async_cow *async_cow;
916 struct btrfs_root *root;
917 unsigned long nr_pages;
919 async_cow = container_of(work, struct async_cow, work);
921 root = async_cow->root;
922 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
925 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
927 if (atomic_read(&root->fs_info->async_delalloc_pages) <
929 waitqueue_active(&root->fs_info->async_submit_wait))
930 wake_up(&root->fs_info->async_submit_wait);
932 if (async_cow->inode)
933 submit_compressed_extents(async_cow->inode, async_cow);
936 static noinline void async_cow_free(struct btrfs_work *work)
938 struct async_cow *async_cow;
939 async_cow = container_of(work, struct async_cow, work);
943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
944 u64 start, u64 end, int *page_started,
945 unsigned long *nr_written)
947 struct async_cow *async_cow;
948 struct btrfs_root *root = BTRFS_I(inode)->root;
949 unsigned long nr_pages;
951 int limit = 10 * 1024 * 1042;
953 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
954 1, 0, NULL, GFP_NOFS);
955 while (start < end) {
956 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
958 async_cow->inode = inode;
959 async_cow->root = root;
960 async_cow->locked_page = locked_page;
961 async_cow->start = start;
963 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
966 cur_end = min(end, start + 512 * 1024 - 1);
968 async_cow->end = cur_end;
969 INIT_LIST_HEAD(&async_cow->extents);
971 async_cow->work.func = async_cow_start;
972 async_cow->work.ordered_func = async_cow_submit;
973 async_cow->work.ordered_free = async_cow_free;
974 async_cow->work.flags = 0;
976 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
978 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
980 btrfs_queue_worker(&root->fs_info->delalloc_workers,
983 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
984 wait_event(root->fs_info->async_submit_wait,
985 (atomic_read(&root->fs_info->async_delalloc_pages) <
989 while (atomic_read(&root->fs_info->async_submit_draining) &&
990 atomic_read(&root->fs_info->async_delalloc_pages)) {
991 wait_event(root->fs_info->async_submit_wait,
992 (atomic_read(&root->fs_info->async_delalloc_pages) ==
996 *nr_written += nr_pages;
1003 static noinline int csum_exist_in_range(struct btrfs_root *root,
1004 u64 bytenr, u64 num_bytes)
1007 struct btrfs_ordered_sum *sums;
1010 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1011 bytenr + num_bytes - 1, &list);
1012 if (ret == 0 && list_empty(&list))
1015 while (!list_empty(&list)) {
1016 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1017 list_del(&sums->list);
1024 * when nowcow writeback call back. This checks for snapshots or COW copies
1025 * of the extents that exist in the file, and COWs the file as required.
1027 * If no cow copies or snapshots exist, we write directly to the existing
1030 static noinline int run_delalloc_nocow(struct inode *inode,
1031 struct page *locked_page,
1032 u64 start, u64 end, int *page_started, int force,
1033 unsigned long *nr_written)
1035 struct btrfs_root *root = BTRFS_I(inode)->root;
1036 struct btrfs_trans_handle *trans;
1037 struct extent_buffer *leaf;
1038 struct btrfs_path *path;
1039 struct btrfs_file_extent_item *fi;
1040 struct btrfs_key found_key;
1052 bool nolock = false;
1054 path = btrfs_alloc_path();
1056 if (root == root->fs_info->tree_root) {
1058 trans = btrfs_join_transaction_nolock(root, 1);
1060 trans = btrfs_join_transaction(root, 1);
1062 BUG_ON(IS_ERR(trans));
1064 cow_start = (u64)-1;
1067 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1070 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1071 leaf = path->nodes[0];
1072 btrfs_item_key_to_cpu(leaf, &found_key,
1073 path->slots[0] - 1);
1074 if (found_key.objectid == inode->i_ino &&
1075 found_key.type == BTRFS_EXTENT_DATA_KEY)
1080 leaf = path->nodes[0];
1081 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1082 ret = btrfs_next_leaf(root, path);
1087 leaf = path->nodes[0];
1093 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1095 if (found_key.objectid > inode->i_ino ||
1096 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1097 found_key.offset > end)
1100 if (found_key.offset > cur_offset) {
1101 extent_end = found_key.offset;
1106 fi = btrfs_item_ptr(leaf, path->slots[0],
1107 struct btrfs_file_extent_item);
1108 extent_type = btrfs_file_extent_type(leaf, fi);
1110 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1111 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1112 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1113 extent_offset = btrfs_file_extent_offset(leaf, fi);
1114 extent_end = found_key.offset +
1115 btrfs_file_extent_num_bytes(leaf, fi);
1116 if (extent_end <= start) {
1120 if (disk_bytenr == 0)
1122 if (btrfs_file_extent_compression(leaf, fi) ||
1123 btrfs_file_extent_encryption(leaf, fi) ||
1124 btrfs_file_extent_other_encoding(leaf, fi))
1126 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1128 if (btrfs_extent_readonly(root, disk_bytenr))
1130 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1132 extent_offset, disk_bytenr))
1134 disk_bytenr += extent_offset;
1135 disk_bytenr += cur_offset - found_key.offset;
1136 num_bytes = min(end + 1, extent_end) - cur_offset;
1138 * force cow if csum exists in the range.
1139 * this ensure that csum for a given extent are
1140 * either valid or do not exist.
1142 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1145 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1146 extent_end = found_key.offset +
1147 btrfs_file_extent_inline_len(leaf, fi);
1148 extent_end = ALIGN(extent_end, root->sectorsize);
1153 if (extent_end <= start) {
1158 if (cow_start == (u64)-1)
1159 cow_start = cur_offset;
1160 cur_offset = extent_end;
1161 if (cur_offset > end)
1167 btrfs_release_path(root, path);
1168 if (cow_start != (u64)-1) {
1169 ret = cow_file_range(inode, locked_page, cow_start,
1170 found_key.offset - 1, page_started,
1173 cow_start = (u64)-1;
1176 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1177 struct extent_map *em;
1178 struct extent_map_tree *em_tree;
1179 em_tree = &BTRFS_I(inode)->extent_tree;
1180 em = alloc_extent_map(GFP_NOFS);
1182 em->start = cur_offset;
1183 em->orig_start = em->start;
1184 em->len = num_bytes;
1185 em->block_len = num_bytes;
1186 em->block_start = disk_bytenr;
1187 em->bdev = root->fs_info->fs_devices->latest_bdev;
1188 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1190 write_lock(&em_tree->lock);
1191 ret = add_extent_mapping(em_tree, em);
1192 write_unlock(&em_tree->lock);
1193 if (ret != -EEXIST) {
1194 free_extent_map(em);
1197 btrfs_drop_extent_cache(inode, em->start,
1198 em->start + em->len - 1, 0);
1200 type = BTRFS_ORDERED_PREALLOC;
1202 type = BTRFS_ORDERED_NOCOW;
1205 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1206 num_bytes, num_bytes, type);
1209 if (root->root_key.objectid ==
1210 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1211 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1216 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1217 cur_offset, cur_offset + num_bytes - 1,
1218 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1219 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1220 EXTENT_SET_PRIVATE2);
1221 cur_offset = extent_end;
1222 if (cur_offset > end)
1225 btrfs_release_path(root, path);
1227 if (cur_offset <= end && cow_start == (u64)-1)
1228 cow_start = cur_offset;
1229 if (cow_start != (u64)-1) {
1230 ret = cow_file_range(inode, locked_page, cow_start, end,
1231 page_started, nr_written, 1);
1236 ret = btrfs_end_transaction_nolock(trans, root);
1239 ret = btrfs_end_transaction(trans, root);
1242 btrfs_free_path(path);
1247 * extent_io.c call back to do delayed allocation processing
1249 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1250 u64 start, u64 end, int *page_started,
1251 unsigned long *nr_written)
1254 struct btrfs_root *root = BTRFS_I(inode)->root;
1256 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1257 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258 page_started, 1, nr_written);
1259 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1260 ret = run_delalloc_nocow(inode, locked_page, start, end,
1261 page_started, 0, nr_written);
1262 else if (!btrfs_test_opt(root, COMPRESS) &&
1263 !(BTRFS_I(inode)->force_compress) &&
1264 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1265 ret = cow_file_range(inode, locked_page, start, end,
1266 page_started, nr_written, 1);
1268 ret = cow_file_range_async(inode, locked_page, start, end,
1269 page_started, nr_written);
1273 static int btrfs_split_extent_hook(struct inode *inode,
1274 struct extent_state *orig, u64 split)
1276 /* not delalloc, ignore it */
1277 if (!(orig->state & EXTENT_DELALLOC))
1280 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1285 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1286 * extents so we can keep track of new extents that are just merged onto old
1287 * extents, such as when we are doing sequential writes, so we can properly
1288 * account for the metadata space we'll need.
1290 static int btrfs_merge_extent_hook(struct inode *inode,
1291 struct extent_state *new,
1292 struct extent_state *other)
1294 /* not delalloc, ignore it */
1295 if (!(other->state & EXTENT_DELALLOC))
1298 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1303 * extent_io.c set_bit_hook, used to track delayed allocation
1304 * bytes in this file, and to maintain the list of inodes that
1305 * have pending delalloc work to be done.
1307 static int btrfs_set_bit_hook(struct inode *inode,
1308 struct extent_state *state, int *bits)
1312 * set_bit and clear bit hooks normally require _irqsave/restore
1313 * but in this case, we are only testeing for the DELALLOC
1314 * bit, which is only set or cleared with irqs on
1316 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1317 struct btrfs_root *root = BTRFS_I(inode)->root;
1318 u64 len = state->end + 1 - state->start;
1319 int do_list = (root->root_key.objectid !=
1320 BTRFS_ROOT_TREE_OBJECTID);
1322 if (*bits & EXTENT_FIRST_DELALLOC)
1323 *bits &= ~EXTENT_FIRST_DELALLOC;
1325 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1327 spin_lock(&root->fs_info->delalloc_lock);
1328 BTRFS_I(inode)->delalloc_bytes += len;
1329 root->fs_info->delalloc_bytes += len;
1330 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332 &root->fs_info->delalloc_inodes);
1334 spin_unlock(&root->fs_info->delalloc_lock);
1340 * extent_io.c clear_bit_hook, see set_bit_hook for why
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343 struct extent_state *state, int *bits)
1346 * set_bit and clear bit hooks normally require _irqsave/restore
1347 * but in this case, we are only testeing for the DELALLOC
1348 * bit, which is only set or cleared with irqs on
1350 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1351 struct btrfs_root *root = BTRFS_I(inode)->root;
1352 u64 len = state->end + 1 - state->start;
1353 int do_list = (root->root_key.objectid !=
1354 BTRFS_ROOT_TREE_OBJECTID);
1356 if (*bits & EXTENT_FIRST_DELALLOC)
1357 *bits &= ~EXTENT_FIRST_DELALLOC;
1358 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1359 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1361 if (*bits & EXTENT_DO_ACCOUNTING)
1362 btrfs_delalloc_release_metadata(inode, len);
1364 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1366 btrfs_free_reserved_data_space(inode, len);
1368 spin_lock(&root->fs_info->delalloc_lock);
1369 root->fs_info->delalloc_bytes -= len;
1370 BTRFS_I(inode)->delalloc_bytes -= len;
1372 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1373 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1374 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1376 spin_unlock(&root->fs_info->delalloc_lock);
1382 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1383 * we don't create bios that span stripes or chunks
1385 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1386 size_t size, struct bio *bio,
1387 unsigned long bio_flags)
1389 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1390 struct btrfs_mapping_tree *map_tree;
1391 u64 logical = (u64)bio->bi_sector << 9;
1396 if (bio_flags & EXTENT_BIO_COMPRESSED)
1399 length = bio->bi_size;
1400 map_tree = &root->fs_info->mapping_tree;
1401 map_length = length;
1402 ret = btrfs_map_block(map_tree, READ, logical,
1403 &map_length, NULL, 0);
1405 if (map_length < length + size)
1411 * in order to insert checksums into the metadata in large chunks,
1412 * we wait until bio submission time. All the pages in the bio are
1413 * checksummed and sums are attached onto the ordered extent record.
1415 * At IO completion time the cums attached on the ordered extent record
1416 * are inserted into the btree
1418 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1419 struct bio *bio, int mirror_num,
1420 unsigned long bio_flags,
1423 struct btrfs_root *root = BTRFS_I(inode)->root;
1426 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1432 * in order to insert checksums into the metadata in large chunks,
1433 * we wait until bio submission time. All the pages in the bio are
1434 * checksummed and sums are attached onto the ordered extent record.
1436 * At IO completion time the cums attached on the ordered extent record
1437 * are inserted into the btree
1439 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1440 int mirror_num, unsigned long bio_flags,
1443 struct btrfs_root *root = BTRFS_I(inode)->root;
1444 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1448 * extent_io.c submission hook. This does the right thing for csum calculation
1449 * on write, or reading the csums from the tree before a read
1451 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1452 int mirror_num, unsigned long bio_flags,
1455 struct btrfs_root *root = BTRFS_I(inode)->root;
1459 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1461 if (root == root->fs_info->tree_root)
1462 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1464 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1467 if (!(rw & REQ_WRITE)) {
1468 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1469 return btrfs_submit_compressed_read(inode, bio,
1470 mirror_num, bio_flags);
1471 } else if (!skip_sum) {
1472 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1477 } else if (!skip_sum) {
1478 /* csum items have already been cloned */
1479 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1481 /* we're doing a write, do the async checksumming */
1482 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483 inode, rw, bio, mirror_num,
1484 bio_flags, bio_offset,
1485 __btrfs_submit_bio_start,
1486 __btrfs_submit_bio_done);
1490 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1494 * given a list of ordered sums record them in the inode. This happens
1495 * at IO completion time based on sums calculated at bio submission time.
1497 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1498 struct inode *inode, u64 file_offset,
1499 struct list_head *list)
1501 struct btrfs_ordered_sum *sum;
1503 btrfs_set_trans_block_group(trans, inode);
1505 list_for_each_entry(sum, list, list) {
1506 btrfs_csum_file_blocks(trans,
1507 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1512 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1513 struct extent_state **cached_state)
1515 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1517 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1518 cached_state, GFP_NOFS);
1521 /* see btrfs_writepage_start_hook for details on why this is required */
1522 struct btrfs_writepage_fixup {
1524 struct btrfs_work work;
1527 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1529 struct btrfs_writepage_fixup *fixup;
1530 struct btrfs_ordered_extent *ordered;
1531 struct extent_state *cached_state = NULL;
1533 struct inode *inode;
1537 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1541 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1542 ClearPageChecked(page);
1546 inode = page->mapping->host;
1547 page_start = page_offset(page);
1548 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1550 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1551 &cached_state, GFP_NOFS);
1553 /* already ordered? We're done */
1554 if (PagePrivate2(page))
1557 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1559 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1560 page_end, &cached_state, GFP_NOFS);
1562 btrfs_start_ordered_extent(inode, ordered, 1);
1567 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1568 ClearPageChecked(page);
1570 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1571 &cached_state, GFP_NOFS);
1574 page_cache_release(page);
1579 * There are a few paths in the higher layers of the kernel that directly
1580 * set the page dirty bit without asking the filesystem if it is a
1581 * good idea. This causes problems because we want to make sure COW
1582 * properly happens and the data=ordered rules are followed.
1584 * In our case any range that doesn't have the ORDERED bit set
1585 * hasn't been properly setup for IO. We kick off an async process
1586 * to fix it up. The async helper will wait for ordered extents, set
1587 * the delalloc bit and make it safe to write the page.
1589 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1591 struct inode *inode = page->mapping->host;
1592 struct btrfs_writepage_fixup *fixup;
1593 struct btrfs_root *root = BTRFS_I(inode)->root;
1595 /* this page is properly in the ordered list */
1596 if (TestClearPagePrivate2(page))
1599 if (PageChecked(page))
1602 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1606 SetPageChecked(page);
1607 page_cache_get(page);
1608 fixup->work.func = btrfs_writepage_fixup_worker;
1610 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1614 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1615 struct inode *inode, u64 file_pos,
1616 u64 disk_bytenr, u64 disk_num_bytes,
1617 u64 num_bytes, u64 ram_bytes,
1618 u8 compression, u8 encryption,
1619 u16 other_encoding, int extent_type)
1621 struct btrfs_root *root = BTRFS_I(inode)->root;
1622 struct btrfs_file_extent_item *fi;
1623 struct btrfs_path *path;
1624 struct extent_buffer *leaf;
1625 struct btrfs_key ins;
1629 path = btrfs_alloc_path();
1632 path->leave_spinning = 1;
1635 * we may be replacing one extent in the tree with another.
1636 * The new extent is pinned in the extent map, and we don't want
1637 * to drop it from the cache until it is completely in the btree.
1639 * So, tell btrfs_drop_extents to leave this extent in the cache.
1640 * the caller is expected to unpin it and allow it to be merged
1643 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1647 ins.objectid = inode->i_ino;
1648 ins.offset = file_pos;
1649 ins.type = BTRFS_EXTENT_DATA_KEY;
1650 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1652 leaf = path->nodes[0];
1653 fi = btrfs_item_ptr(leaf, path->slots[0],
1654 struct btrfs_file_extent_item);
1655 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1656 btrfs_set_file_extent_type(leaf, fi, extent_type);
1657 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1658 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1659 btrfs_set_file_extent_offset(leaf, fi, 0);
1660 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1661 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1662 btrfs_set_file_extent_compression(leaf, fi, compression);
1663 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1664 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1666 btrfs_unlock_up_safe(path, 1);
1667 btrfs_set_lock_blocking(leaf);
1669 btrfs_mark_buffer_dirty(leaf);
1671 inode_add_bytes(inode, num_bytes);
1673 ins.objectid = disk_bytenr;
1674 ins.offset = disk_num_bytes;
1675 ins.type = BTRFS_EXTENT_ITEM_KEY;
1676 ret = btrfs_alloc_reserved_file_extent(trans, root,
1677 root->root_key.objectid,
1678 inode->i_ino, file_pos, &ins);
1680 btrfs_free_path(path);
1686 * helper function for btrfs_finish_ordered_io, this
1687 * just reads in some of the csum leaves to prime them into ram
1688 * before we start the transaction. It limits the amount of btree
1689 * reads required while inside the transaction.
1691 /* as ordered data IO finishes, this gets called so we can finish
1692 * an ordered extent if the range of bytes in the file it covers are
1695 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1697 struct btrfs_root *root = BTRFS_I(inode)->root;
1698 struct btrfs_trans_handle *trans = NULL;
1699 struct btrfs_ordered_extent *ordered_extent = NULL;
1700 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1701 struct extent_state *cached_state = NULL;
1702 int compress_type = 0;
1704 bool nolock = false;
1706 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1710 BUG_ON(!ordered_extent);
1712 nolock = (root == root->fs_info->tree_root);
1714 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1715 BUG_ON(!list_empty(&ordered_extent->list));
1716 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1719 trans = btrfs_join_transaction_nolock(root, 1);
1721 trans = btrfs_join_transaction(root, 1);
1722 BUG_ON(IS_ERR(trans));
1723 btrfs_set_trans_block_group(trans, inode);
1724 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1725 ret = btrfs_update_inode(trans, root, inode);
1731 lock_extent_bits(io_tree, ordered_extent->file_offset,
1732 ordered_extent->file_offset + ordered_extent->len - 1,
1733 0, &cached_state, GFP_NOFS);
1736 trans = btrfs_join_transaction_nolock(root, 1);
1738 trans = btrfs_join_transaction(root, 1);
1739 BUG_ON(IS_ERR(trans));
1740 btrfs_set_trans_block_group(trans, inode);
1741 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1743 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1744 compress_type = ordered_extent->compress_type;
1745 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1746 BUG_ON(compress_type);
1747 ret = btrfs_mark_extent_written(trans, inode,
1748 ordered_extent->file_offset,
1749 ordered_extent->file_offset +
1750 ordered_extent->len);
1753 BUG_ON(root == root->fs_info->tree_root);
1754 ret = insert_reserved_file_extent(trans, inode,
1755 ordered_extent->file_offset,
1756 ordered_extent->start,
1757 ordered_extent->disk_len,
1758 ordered_extent->len,
1759 ordered_extent->len,
1760 compress_type, 0, 0,
1761 BTRFS_FILE_EXTENT_REG);
1762 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1763 ordered_extent->file_offset,
1764 ordered_extent->len);
1767 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1768 ordered_extent->file_offset +
1769 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1771 add_pending_csums(trans, inode, ordered_extent->file_offset,
1772 &ordered_extent->list);
1774 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1776 ret = btrfs_update_inode(trans, root, inode);
1783 btrfs_end_transaction_nolock(trans, root);
1785 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1787 btrfs_end_transaction(trans, root);
1791 btrfs_put_ordered_extent(ordered_extent);
1792 /* once for the tree */
1793 btrfs_put_ordered_extent(ordered_extent);
1798 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1799 struct extent_state *state, int uptodate)
1801 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1803 ClearPagePrivate2(page);
1804 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1808 * When IO fails, either with EIO or csum verification fails, we
1809 * try other mirrors that might have a good copy of the data. This
1810 * io_failure_record is used to record state as we go through all the
1811 * mirrors. If another mirror has good data, the page is set up to date
1812 * and things continue. If a good mirror can't be found, the original
1813 * bio end_io callback is called to indicate things have failed.
1815 struct io_failure_record {
1820 unsigned long bio_flags;
1824 static int btrfs_io_failed_hook(struct bio *failed_bio,
1825 struct page *page, u64 start, u64 end,
1826 struct extent_state *state)
1828 struct io_failure_record *failrec = NULL;
1830 struct extent_map *em;
1831 struct inode *inode = page->mapping->host;
1832 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1833 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1840 ret = get_state_private(failure_tree, start, &private);
1842 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1845 failrec->start = start;
1846 failrec->len = end - start + 1;
1847 failrec->last_mirror = 0;
1848 failrec->bio_flags = 0;
1850 read_lock(&em_tree->lock);
1851 em = lookup_extent_mapping(em_tree, start, failrec->len);
1852 if (em->start > start || em->start + em->len < start) {
1853 free_extent_map(em);
1856 read_unlock(&em_tree->lock);
1858 if (!em || IS_ERR(em)) {
1862 logical = start - em->start;
1863 logical = em->block_start + logical;
1864 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1865 logical = em->block_start;
1866 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1867 extent_set_compress_type(&failrec->bio_flags,
1870 failrec->logical = logical;
1871 free_extent_map(em);
1872 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1873 EXTENT_DIRTY, GFP_NOFS);
1874 set_state_private(failure_tree, start,
1875 (u64)(unsigned long)failrec);
1877 failrec = (struct io_failure_record *)(unsigned long)private;
1879 num_copies = btrfs_num_copies(
1880 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1881 failrec->logical, failrec->len);
1882 failrec->last_mirror++;
1884 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1885 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1888 if (state && state->start != failrec->start)
1890 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1892 if (!state || failrec->last_mirror > num_copies) {
1893 set_state_private(failure_tree, failrec->start, 0);
1894 clear_extent_bits(failure_tree, failrec->start,
1895 failrec->start + failrec->len - 1,
1896 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1900 bio = bio_alloc(GFP_NOFS, 1);
1901 bio->bi_private = state;
1902 bio->bi_end_io = failed_bio->bi_end_io;
1903 bio->bi_sector = failrec->logical >> 9;
1904 bio->bi_bdev = failed_bio->bi_bdev;
1907 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1908 if (failed_bio->bi_rw & REQ_WRITE)
1913 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1914 failrec->last_mirror,
1915 failrec->bio_flags, 0);
1920 * each time an IO finishes, we do a fast check in the IO failure tree
1921 * to see if we need to process or clean up an io_failure_record
1923 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1926 u64 private_failure;
1927 struct io_failure_record *failure;
1931 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1932 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1933 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1934 start, &private_failure);
1936 failure = (struct io_failure_record *)(unsigned long)
1938 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1940 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1942 failure->start + failure->len - 1,
1943 EXTENT_DIRTY | EXTENT_LOCKED,
1952 * when reads are done, we need to check csums to verify the data is correct
1953 * if there's a match, we allow the bio to finish. If not, we go through
1954 * the io_failure_record routines to find good copies
1956 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1957 struct extent_state *state)
1959 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1960 struct inode *inode = page->mapping->host;
1961 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1963 u64 private = ~(u32)0;
1965 struct btrfs_root *root = BTRFS_I(inode)->root;
1968 if (PageChecked(page)) {
1969 ClearPageChecked(page);
1973 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1976 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1977 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1978 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1983 if (state && state->start == start) {
1984 private = state->private;
1987 ret = get_state_private(io_tree, start, &private);
1989 kaddr = kmap_atomic(page, KM_USER0);
1993 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1994 btrfs_csum_final(csum, (char *)&csum);
1995 if (csum != private)
1998 kunmap_atomic(kaddr, KM_USER0);
2000 /* if the io failure tree for this inode is non-empty,
2001 * check to see if we've recovered from a failed IO
2003 btrfs_clean_io_failures(inode, start);
2007 if (printk_ratelimit()) {
2008 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2009 "private %llu\n", page->mapping->host->i_ino,
2010 (unsigned long long)start, csum,
2011 (unsigned long long)private);
2013 memset(kaddr + offset, 1, end - start + 1);
2014 flush_dcache_page(page);
2015 kunmap_atomic(kaddr, KM_USER0);
2021 struct delayed_iput {
2022 struct list_head list;
2023 struct inode *inode;
2026 void btrfs_add_delayed_iput(struct inode *inode)
2028 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2029 struct delayed_iput *delayed;
2031 if (atomic_add_unless(&inode->i_count, -1, 1))
2034 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2035 delayed->inode = inode;
2037 spin_lock(&fs_info->delayed_iput_lock);
2038 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2039 spin_unlock(&fs_info->delayed_iput_lock);
2042 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2045 struct btrfs_fs_info *fs_info = root->fs_info;
2046 struct delayed_iput *delayed;
2049 spin_lock(&fs_info->delayed_iput_lock);
2050 empty = list_empty(&fs_info->delayed_iputs);
2051 spin_unlock(&fs_info->delayed_iput_lock);
2055 down_read(&root->fs_info->cleanup_work_sem);
2056 spin_lock(&fs_info->delayed_iput_lock);
2057 list_splice_init(&fs_info->delayed_iputs, &list);
2058 spin_unlock(&fs_info->delayed_iput_lock);
2060 while (!list_empty(&list)) {
2061 delayed = list_entry(list.next, struct delayed_iput, list);
2062 list_del(&delayed->list);
2063 iput(delayed->inode);
2066 up_read(&root->fs_info->cleanup_work_sem);
2070 * calculate extra metadata reservation when snapshotting a subvolume
2071 * contains orphan files.
2073 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2074 struct btrfs_pending_snapshot *pending,
2075 u64 *bytes_to_reserve)
2077 struct btrfs_root *root;
2078 struct btrfs_block_rsv *block_rsv;
2082 root = pending->root;
2083 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2086 block_rsv = root->orphan_block_rsv;
2088 /* orphan block reservation for the snapshot */
2089 num_bytes = block_rsv->size;
2092 * after the snapshot is created, COWing tree blocks may use more
2093 * space than it frees. So we should make sure there is enough
2096 index = trans->transid & 0x1;
2097 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2098 num_bytes += block_rsv->size -
2099 (block_rsv->reserved + block_rsv->freed[index]);
2102 *bytes_to_reserve += num_bytes;
2105 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2106 struct btrfs_pending_snapshot *pending)
2108 struct btrfs_root *root = pending->root;
2109 struct btrfs_root *snap = pending->snap;
2110 struct btrfs_block_rsv *block_rsv;
2115 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2118 /* refill source subvolume's orphan block reservation */
2119 block_rsv = root->orphan_block_rsv;
2120 index = trans->transid & 0x1;
2121 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122 num_bytes = block_rsv->size -
2123 (block_rsv->reserved + block_rsv->freed[index]);
2124 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2125 root->orphan_block_rsv,
2130 /* setup orphan block reservation for the snapshot */
2131 block_rsv = btrfs_alloc_block_rsv(snap);
2134 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2135 snap->orphan_block_rsv = block_rsv;
2137 num_bytes = root->orphan_block_rsv->size;
2138 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139 block_rsv, num_bytes);
2143 /* insert orphan item for the snapshot */
2144 WARN_ON(!root->orphan_item_inserted);
2145 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2146 snap->root_key.objectid);
2148 snap->orphan_item_inserted = 1;
2152 enum btrfs_orphan_cleanup_state {
2153 ORPHAN_CLEANUP_STARTED = 1,
2154 ORPHAN_CLEANUP_DONE = 2,
2158 * This is called in transaction commmit time. If there are no orphan
2159 * files in the subvolume, it removes orphan item and frees block_rsv
2162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2163 struct btrfs_root *root)
2167 if (!list_empty(&root->orphan_list) ||
2168 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2171 if (root->orphan_item_inserted &&
2172 btrfs_root_refs(&root->root_item) > 0) {
2173 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2174 root->root_key.objectid);
2176 root->orphan_item_inserted = 0;
2179 if (root->orphan_block_rsv) {
2180 WARN_ON(root->orphan_block_rsv->size > 0);
2181 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2182 root->orphan_block_rsv = NULL;
2187 * This creates an orphan entry for the given inode in case something goes
2188 * wrong in the middle of an unlink/truncate.
2190 * NOTE: caller of this function should reserve 5 units of metadata for
2193 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2195 struct btrfs_root *root = BTRFS_I(inode)->root;
2196 struct btrfs_block_rsv *block_rsv = NULL;
2201 if (!root->orphan_block_rsv) {
2202 block_rsv = btrfs_alloc_block_rsv(root);
2206 spin_lock(&root->orphan_lock);
2207 if (!root->orphan_block_rsv) {
2208 root->orphan_block_rsv = block_rsv;
2209 } else if (block_rsv) {
2210 btrfs_free_block_rsv(root, block_rsv);
2214 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2215 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2218 * For proper ENOSPC handling, we should do orphan
2219 * cleanup when mounting. But this introduces backward
2220 * compatibility issue.
2222 if (!xchg(&root->orphan_item_inserted, 1))
2230 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2231 BTRFS_I(inode)->orphan_meta_reserved = 1;
2234 spin_unlock(&root->orphan_lock);
2237 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2239 /* grab metadata reservation from transaction handle */
2241 ret = btrfs_orphan_reserve_metadata(trans, inode);
2245 /* insert an orphan item to track this unlinked/truncated file */
2247 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2251 /* insert an orphan item to track subvolume contains orphan files */
2253 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2254 root->root_key.objectid);
2261 * We have done the truncate/delete so we can go ahead and remove the orphan
2262 * item for this particular inode.
2264 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2266 struct btrfs_root *root = BTRFS_I(inode)->root;
2267 int delete_item = 0;
2268 int release_rsv = 0;
2271 spin_lock(&root->orphan_lock);
2272 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2273 list_del_init(&BTRFS_I(inode)->i_orphan);
2277 if (BTRFS_I(inode)->orphan_meta_reserved) {
2278 BTRFS_I(inode)->orphan_meta_reserved = 0;
2281 spin_unlock(&root->orphan_lock);
2283 if (trans && delete_item) {
2284 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2289 btrfs_orphan_release_metadata(inode);
2295 * this cleans up any orphans that may be left on the list from the last use
2298 int btrfs_orphan_cleanup(struct btrfs_root *root)
2300 struct btrfs_path *path;
2301 struct extent_buffer *leaf;
2302 struct btrfs_key key, found_key;
2303 struct btrfs_trans_handle *trans;
2304 struct inode *inode;
2305 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2307 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2310 path = btrfs_alloc_path();
2317 key.objectid = BTRFS_ORPHAN_OBJECTID;
2318 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2319 key.offset = (u64)-1;
2322 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2327 * if ret == 0 means we found what we were searching for, which
2328 * is weird, but possible, so only screw with path if we didn't
2329 * find the key and see if we have stuff that matches
2333 if (path->slots[0] == 0)
2338 /* pull out the item */
2339 leaf = path->nodes[0];
2340 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2342 /* make sure the item matches what we want */
2343 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2345 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2348 /* release the path since we're done with it */
2349 btrfs_release_path(root, path);
2352 * this is where we are basically btrfs_lookup, without the
2353 * crossing root thing. we store the inode number in the
2354 * offset of the orphan item.
2356 found_key.objectid = found_key.offset;
2357 found_key.type = BTRFS_INODE_ITEM_KEY;
2358 found_key.offset = 0;
2359 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2360 if (IS_ERR(inode)) {
2361 ret = PTR_ERR(inode);
2366 * add this inode to the orphan list so btrfs_orphan_del does
2367 * the proper thing when we hit it
2369 spin_lock(&root->orphan_lock);
2370 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2371 spin_unlock(&root->orphan_lock);
2374 * if this is a bad inode, means we actually succeeded in
2375 * removing the inode, but not the orphan record, which means
2376 * we need to manually delete the orphan since iput will just
2377 * do a destroy_inode
2379 if (is_bad_inode(inode)) {
2380 trans = btrfs_start_transaction(root, 0);
2381 if (IS_ERR(trans)) {
2382 ret = PTR_ERR(trans);
2385 btrfs_orphan_del(trans, inode);
2386 btrfs_end_transaction(trans, root);
2391 /* if we have links, this was a truncate, lets do that */
2392 if (inode->i_nlink) {
2393 if (!S_ISREG(inode->i_mode)) {
2399 ret = btrfs_truncate(inode);
2404 /* this will do delete_inode and everything for us */
2409 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2411 if (root->orphan_block_rsv)
2412 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2415 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2416 trans = btrfs_join_transaction(root, 1);
2418 btrfs_end_transaction(trans, root);
2422 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2424 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2428 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2429 btrfs_free_path(path);
2434 * very simple check to peek ahead in the leaf looking for xattrs. If we
2435 * don't find any xattrs, we know there can't be any acls.
2437 * slot is the slot the inode is in, objectid is the objectid of the inode
2439 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2440 int slot, u64 objectid)
2442 u32 nritems = btrfs_header_nritems(leaf);
2443 struct btrfs_key found_key;
2447 while (slot < nritems) {
2448 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2450 /* we found a different objectid, there must not be acls */
2451 if (found_key.objectid != objectid)
2454 /* we found an xattr, assume we've got an acl */
2455 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2459 * we found a key greater than an xattr key, there can't
2460 * be any acls later on
2462 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2469 * it goes inode, inode backrefs, xattrs, extents,
2470 * so if there are a ton of hard links to an inode there can
2471 * be a lot of backrefs. Don't waste time searching too hard,
2472 * this is just an optimization
2477 /* we hit the end of the leaf before we found an xattr or
2478 * something larger than an xattr. We have to assume the inode
2485 * read an inode from the btree into the in-memory inode
2487 static void btrfs_read_locked_inode(struct inode *inode)
2489 struct btrfs_path *path;
2490 struct extent_buffer *leaf;
2491 struct btrfs_inode_item *inode_item;
2492 struct btrfs_timespec *tspec;
2493 struct btrfs_root *root = BTRFS_I(inode)->root;
2494 struct btrfs_key location;
2496 u64 alloc_group_block;
2500 path = btrfs_alloc_path();
2502 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2504 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2508 leaf = path->nodes[0];
2509 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2510 struct btrfs_inode_item);
2512 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2513 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2514 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2515 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2516 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2518 tspec = btrfs_inode_atime(inode_item);
2519 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2520 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2522 tspec = btrfs_inode_mtime(inode_item);
2523 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2524 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2526 tspec = btrfs_inode_ctime(inode_item);
2527 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2528 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2530 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2531 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2532 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2533 inode->i_generation = BTRFS_I(inode)->generation;
2535 rdev = btrfs_inode_rdev(leaf, inode_item);
2537 BTRFS_I(inode)->index_cnt = (u64)-1;
2538 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2540 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2543 * try to precache a NULL acl entry for files that don't have
2544 * any xattrs or acls
2546 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2548 cache_no_acl(inode);
2550 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2551 alloc_group_block, 0);
2552 btrfs_free_path(path);
2555 switch (inode->i_mode & S_IFMT) {
2557 inode->i_mapping->a_ops = &btrfs_aops;
2558 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2559 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2560 inode->i_fop = &btrfs_file_operations;
2561 inode->i_op = &btrfs_file_inode_operations;
2564 inode->i_fop = &btrfs_dir_file_operations;
2565 if (root == root->fs_info->tree_root)
2566 inode->i_op = &btrfs_dir_ro_inode_operations;
2568 inode->i_op = &btrfs_dir_inode_operations;
2571 inode->i_op = &btrfs_symlink_inode_operations;
2572 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2573 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2576 inode->i_op = &btrfs_special_inode_operations;
2577 init_special_inode(inode, inode->i_mode, rdev);
2581 btrfs_update_iflags(inode);
2585 btrfs_free_path(path);
2586 make_bad_inode(inode);
2590 * given a leaf and an inode, copy the inode fields into the leaf
2592 static void fill_inode_item(struct btrfs_trans_handle *trans,
2593 struct extent_buffer *leaf,
2594 struct btrfs_inode_item *item,
2595 struct inode *inode)
2597 if (!leaf->map_token)
2598 map_private_extent_buffer(leaf, (unsigned long)item,
2599 sizeof(struct btrfs_inode_item),
2600 &leaf->map_token, &leaf->kaddr,
2601 &leaf->map_start, &leaf->map_len,
2604 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2605 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2606 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2607 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2608 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2610 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2611 inode->i_atime.tv_sec);
2612 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2613 inode->i_atime.tv_nsec);
2615 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2616 inode->i_mtime.tv_sec);
2617 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2618 inode->i_mtime.tv_nsec);
2620 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2621 inode->i_ctime.tv_sec);
2622 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2623 inode->i_ctime.tv_nsec);
2625 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2626 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2627 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2628 btrfs_set_inode_transid(leaf, item, trans->transid);
2629 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2630 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2631 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2633 if (leaf->map_token) {
2634 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2635 leaf->map_token = NULL;
2640 * copy everything in the in-memory inode into the btree.
2642 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2643 struct btrfs_root *root, struct inode *inode)
2645 struct btrfs_inode_item *inode_item;
2646 struct btrfs_path *path;
2647 struct extent_buffer *leaf;
2650 path = btrfs_alloc_path();
2652 path->leave_spinning = 1;
2653 ret = btrfs_lookup_inode(trans, root, path,
2654 &BTRFS_I(inode)->location, 1);
2661 btrfs_unlock_up_safe(path, 1);
2662 leaf = path->nodes[0];
2663 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2664 struct btrfs_inode_item);
2666 fill_inode_item(trans, leaf, inode_item, inode);
2667 btrfs_mark_buffer_dirty(leaf);
2668 btrfs_set_inode_last_trans(trans, inode);
2671 btrfs_free_path(path);
2677 * unlink helper that gets used here in inode.c and in the tree logging
2678 * recovery code. It remove a link in a directory with a given name, and
2679 * also drops the back refs in the inode to the directory
2681 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2682 struct btrfs_root *root,
2683 struct inode *dir, struct inode *inode,
2684 const char *name, int name_len)
2686 struct btrfs_path *path;
2688 struct extent_buffer *leaf;
2689 struct btrfs_dir_item *di;
2690 struct btrfs_key key;
2693 path = btrfs_alloc_path();
2699 path->leave_spinning = 1;
2700 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2701 name, name_len, -1);
2710 leaf = path->nodes[0];
2711 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2712 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2715 btrfs_release_path(root, path);
2717 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2719 dir->i_ino, &index);
2721 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2722 "inode %lu parent %lu\n", name_len, name,
2723 inode->i_ino, dir->i_ino);
2727 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2728 index, name, name_len, -1);
2737 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2738 btrfs_release_path(root, path);
2740 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2742 BUG_ON(ret != 0 && ret != -ENOENT);
2744 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2749 btrfs_free_path(path);
2753 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2754 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2755 btrfs_update_inode(trans, root, dir);
2760 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2761 struct btrfs_root *root,
2762 struct inode *dir, struct inode *inode,
2763 const char *name, int name_len)
2766 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2768 btrfs_drop_nlink(inode);
2769 ret = btrfs_update_inode(trans, root, inode);
2775 /* helper to check if there is any shared block in the path */
2776 static int check_path_shared(struct btrfs_root *root,
2777 struct btrfs_path *path)
2779 struct extent_buffer *eb;
2783 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2786 if (!path->nodes[level])
2788 eb = path->nodes[level];
2789 if (!btrfs_block_can_be_shared(root, eb))
2791 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2800 * helper to start transaction for unlink and rmdir.
2802 * unlink and rmdir are special in btrfs, they do not always free space.
2803 * so in enospc case, we should make sure they will free space before
2804 * allowing them to use the global metadata reservation.
2806 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2807 struct dentry *dentry)
2809 struct btrfs_trans_handle *trans;
2810 struct btrfs_root *root = BTRFS_I(dir)->root;
2811 struct btrfs_path *path;
2812 struct btrfs_inode_ref *ref;
2813 struct btrfs_dir_item *di;
2814 struct inode *inode = dentry->d_inode;
2820 trans = btrfs_start_transaction(root, 10);
2821 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2824 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2825 return ERR_PTR(-ENOSPC);
2827 /* check if there is someone else holds reference */
2828 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2829 return ERR_PTR(-ENOSPC);
2831 if (atomic_read(&inode->i_count) > 2)
2832 return ERR_PTR(-ENOSPC);
2834 if (xchg(&root->fs_info->enospc_unlink, 1))
2835 return ERR_PTR(-ENOSPC);
2837 path = btrfs_alloc_path();
2839 root->fs_info->enospc_unlink = 0;
2840 return ERR_PTR(-ENOMEM);
2843 trans = btrfs_start_transaction(root, 0);
2844 if (IS_ERR(trans)) {
2845 btrfs_free_path(path);
2846 root->fs_info->enospc_unlink = 0;
2850 path->skip_locking = 1;
2851 path->search_commit_root = 1;
2853 ret = btrfs_lookup_inode(trans, root, path,
2854 &BTRFS_I(dir)->location, 0);
2860 if (check_path_shared(root, path))
2865 btrfs_release_path(root, path);
2867 ret = btrfs_lookup_inode(trans, root, path,
2868 &BTRFS_I(inode)->location, 0);
2874 if (check_path_shared(root, path))
2879 btrfs_release_path(root, path);
2881 if (ret == 0 && S_ISREG(inode->i_mode)) {
2882 ret = btrfs_lookup_file_extent(trans, root, path,
2883 inode->i_ino, (u64)-1, 0);
2889 if (check_path_shared(root, path))
2891 btrfs_release_path(root, path);
2899 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2900 dentry->d_name.name, dentry->d_name.len, 0);
2906 if (check_path_shared(root, path))
2912 btrfs_release_path(root, path);
2914 ref = btrfs_lookup_inode_ref(trans, root, path,
2915 dentry->d_name.name, dentry->d_name.len,
2916 inode->i_ino, dir->i_ino, 0);
2922 if (check_path_shared(root, path))
2924 index = btrfs_inode_ref_index(path->nodes[0], ref);
2925 btrfs_release_path(root, path);
2927 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2928 dentry->d_name.name, dentry->d_name.len, 0);
2933 BUG_ON(ret == -ENOENT);
2934 if (check_path_shared(root, path))
2939 btrfs_free_path(path);
2941 btrfs_end_transaction(trans, root);
2942 root->fs_info->enospc_unlink = 0;
2943 return ERR_PTR(err);
2946 trans->block_rsv = &root->fs_info->global_block_rsv;
2950 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2951 struct btrfs_root *root)
2953 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2954 BUG_ON(!root->fs_info->enospc_unlink);
2955 root->fs_info->enospc_unlink = 0;
2957 btrfs_end_transaction_throttle(trans, root);
2960 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2962 struct btrfs_root *root = BTRFS_I(dir)->root;
2963 struct btrfs_trans_handle *trans;
2964 struct inode *inode = dentry->d_inode;
2966 unsigned long nr = 0;
2968 trans = __unlink_start_trans(dir, dentry);
2970 return PTR_ERR(trans);
2972 btrfs_set_trans_block_group(trans, dir);
2974 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2976 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2977 dentry->d_name.name, dentry->d_name.len);
2980 if (inode->i_nlink == 0) {
2981 ret = btrfs_orphan_add(trans, inode);
2985 nr = trans->blocks_used;
2986 __unlink_end_trans(trans, root);
2987 btrfs_btree_balance_dirty(root, nr);
2991 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2992 struct btrfs_root *root,
2993 struct inode *dir, u64 objectid,
2994 const char *name, int name_len)
2996 struct btrfs_path *path;
2997 struct extent_buffer *leaf;
2998 struct btrfs_dir_item *di;
2999 struct btrfs_key key;
3003 path = btrfs_alloc_path();
3007 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3008 name, name_len, -1);
3009 BUG_ON(!di || IS_ERR(di));
3011 leaf = path->nodes[0];
3012 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3013 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3014 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3016 btrfs_release_path(root, path);
3018 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3019 objectid, root->root_key.objectid,
3020 dir->i_ino, &index, name, name_len);
3022 BUG_ON(ret != -ENOENT);
3023 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3025 BUG_ON(!di || IS_ERR(di));
3027 leaf = path->nodes[0];
3028 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3029 btrfs_release_path(root, path);
3033 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3034 index, name, name_len, -1);
3035 BUG_ON(!di || IS_ERR(di));
3037 leaf = path->nodes[0];
3038 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3039 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3040 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3042 btrfs_release_path(root, path);
3044 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3045 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3046 ret = btrfs_update_inode(trans, root, dir);
3049 btrfs_free_path(path);
3053 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3055 struct inode *inode = dentry->d_inode;
3057 struct btrfs_root *root = BTRFS_I(dir)->root;
3058 struct btrfs_trans_handle *trans;
3059 unsigned long nr = 0;
3061 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3062 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3065 trans = __unlink_start_trans(dir, dentry);
3067 return PTR_ERR(trans);
3069 btrfs_set_trans_block_group(trans, dir);
3071 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3072 err = btrfs_unlink_subvol(trans, root, dir,
3073 BTRFS_I(inode)->location.objectid,
3074 dentry->d_name.name,
3075 dentry->d_name.len);
3079 err = btrfs_orphan_add(trans, inode);
3083 /* now the directory is empty */
3084 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3085 dentry->d_name.name, dentry->d_name.len);
3087 btrfs_i_size_write(inode, 0);
3089 nr = trans->blocks_used;
3090 __unlink_end_trans(trans, root);
3091 btrfs_btree_balance_dirty(root, nr);
3098 * when truncating bytes in a file, it is possible to avoid reading
3099 * the leaves that contain only checksum items. This can be the
3100 * majority of the IO required to delete a large file, but it must
3101 * be done carefully.
3103 * The keys in the level just above the leaves are checked to make sure
3104 * the lowest key in a given leaf is a csum key, and starts at an offset
3105 * after the new size.
3107 * Then the key for the next leaf is checked to make sure it also has
3108 * a checksum item for the same file. If it does, we know our target leaf
3109 * contains only checksum items, and it can be safely freed without reading
3112 * This is just an optimization targeted at large files. It may do
3113 * nothing. It will return 0 unless things went badly.
3115 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3116 struct btrfs_root *root,
3117 struct btrfs_path *path,
3118 struct inode *inode, u64 new_size)
3120 struct btrfs_key key;
3123 struct btrfs_key found_key;
3124 struct btrfs_key other_key;
3125 struct btrfs_leaf_ref *ref;
3129 path->lowest_level = 1;
3130 key.objectid = inode->i_ino;
3131 key.type = BTRFS_CSUM_ITEM_KEY;
3132 key.offset = new_size;
3134 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3138 if (path->nodes[1] == NULL) {
3143 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3144 nritems = btrfs_header_nritems(path->nodes[1]);
3149 if (path->slots[1] >= nritems)
3152 /* did we find a key greater than anything we want to delete? */
3153 if (found_key.objectid > inode->i_ino ||
3154 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3157 /* we check the next key in the node to make sure the leave contains
3158 * only checksum items. This comparison doesn't work if our
3159 * leaf is the last one in the node
3161 if (path->slots[1] + 1 >= nritems) {
3163 /* search forward from the last key in the node, this
3164 * will bring us into the next node in the tree
3166 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3168 /* unlikely, but we inc below, so check to be safe */
3169 if (found_key.offset == (u64)-1)
3172 /* search_forward needs a path with locks held, do the
3173 * search again for the original key. It is possible
3174 * this will race with a balance and return a path that
3175 * we could modify, but this drop is just an optimization
3176 * and is allowed to miss some leaves.
3178 btrfs_release_path(root, path);
3181 /* setup a max key for search_forward */
3182 other_key.offset = (u64)-1;
3183 other_key.type = key.type;
3184 other_key.objectid = key.objectid;
3186 path->keep_locks = 1;
3187 ret = btrfs_search_forward(root, &found_key, &other_key,
3189 path->keep_locks = 0;
3190 if (ret || found_key.objectid != key.objectid ||
3191 found_key.type != key.type) {
3196 key.offset = found_key.offset;
3197 btrfs_release_path(root, path);
3202 /* we know there's one more slot after us in the tree,
3203 * read that key so we can verify it is also a checksum item
3205 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3207 if (found_key.objectid < inode->i_ino)
3210 if (found_key.type != key.type || found_key.offset < new_size)
3214 * if the key for the next leaf isn't a csum key from this objectid,
3215 * we can't be sure there aren't good items inside this leaf.
3218 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3221 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3222 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3224 * it is safe to delete this leaf, it contains only
3225 * csum items from this inode at an offset >= new_size
3227 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3230 if (root->ref_cows && leaf_gen < trans->transid) {
3231 ref = btrfs_alloc_leaf_ref(root, 0);
3233 ref->root_gen = root->root_key.offset;
3234 ref->bytenr = leaf_start;
3236 ref->generation = leaf_gen;
3239 btrfs_sort_leaf_ref(ref);
3241 ret = btrfs_add_leaf_ref(root, ref, 0);
3243 btrfs_free_leaf_ref(root, ref);
3249 btrfs_release_path(root, path);
3251 if (other_key.objectid == inode->i_ino &&
3252 other_key.type == key.type && other_key.offset > key.offset) {
3253 key.offset = other_key.offset;
3259 /* fixup any changes we've made to the path */
3260 path->lowest_level = 0;
3261 path->keep_locks = 0;
3262 btrfs_release_path(root, path);
3269 * this can truncate away extent items, csum items and directory items.
3270 * It starts at a high offset and removes keys until it can't find
3271 * any higher than new_size
3273 * csum items that cross the new i_size are truncated to the new size
3276 * min_type is the minimum key type to truncate down to. If set to 0, this
3277 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3279 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3280 struct btrfs_root *root,
3281 struct inode *inode,
3282 u64 new_size, u32 min_type)
3284 struct btrfs_path *path;
3285 struct extent_buffer *leaf;
3286 struct btrfs_file_extent_item *fi;
3287 struct btrfs_key key;
3288 struct btrfs_key found_key;
3289 u64 extent_start = 0;
3290 u64 extent_num_bytes = 0;
3291 u64 extent_offset = 0;
3293 u64 mask = root->sectorsize - 1;
3294 u32 found_type = (u8)-1;
3297 int pending_del_nr = 0;
3298 int pending_del_slot = 0;
3299 int extent_type = -1;
3304 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3306 if (root->ref_cows || root == root->fs_info->tree_root)
3307 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3309 path = btrfs_alloc_path();
3313 key.objectid = inode->i_ino;
3314 key.offset = (u64)-1;
3318 path->leave_spinning = 1;
3319 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3326 /* there are no items in the tree for us to truncate, we're
3329 if (path->slots[0] == 0)
3336 leaf = path->nodes[0];
3337 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3338 found_type = btrfs_key_type(&found_key);
3341 if (found_key.objectid != inode->i_ino)
3344 if (found_type < min_type)
3347 item_end = found_key.offset;
3348 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3349 fi = btrfs_item_ptr(leaf, path->slots[0],
3350 struct btrfs_file_extent_item);
3351 extent_type = btrfs_file_extent_type(leaf, fi);
3352 encoding = btrfs_file_extent_compression(leaf, fi);
3353 encoding |= btrfs_file_extent_encryption(leaf, fi);
3354 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3356 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3358 btrfs_file_extent_num_bytes(leaf, fi);
3359 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3360 item_end += btrfs_file_extent_inline_len(leaf,
3365 if (found_type > min_type) {
3368 if (item_end < new_size)
3370 if (found_key.offset >= new_size)
3376 /* FIXME, shrink the extent if the ref count is only 1 */
3377 if (found_type != BTRFS_EXTENT_DATA_KEY)
3380 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3382 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3383 if (!del_item && !encoding) {
3384 u64 orig_num_bytes =
3385 btrfs_file_extent_num_bytes(leaf, fi);
3386 extent_num_bytes = new_size -
3387 found_key.offset + root->sectorsize - 1;
3388 extent_num_bytes = extent_num_bytes &
3389 ~((u64)root->sectorsize - 1);
3390 btrfs_set_file_extent_num_bytes(leaf, fi,
3392 num_dec = (orig_num_bytes -
3394 if (root->ref_cows && extent_start != 0)
3395 inode_sub_bytes(inode, num_dec);
3396 btrfs_mark_buffer_dirty(leaf);
3399 btrfs_file_extent_disk_num_bytes(leaf,
3401 extent_offset = found_key.offset -
3402 btrfs_file_extent_offset(leaf, fi);
3404 /* FIXME blocksize != 4096 */
3405 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3406 if (extent_start != 0) {
3409 inode_sub_bytes(inode, num_dec);
3412 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3414 * we can't truncate inline items that have had
3418 btrfs_file_extent_compression(leaf, fi) == 0 &&
3419 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3420 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3421 u32 size = new_size - found_key.offset;
3423 if (root->ref_cows) {
3424 inode_sub_bytes(inode, item_end + 1 -
3428 btrfs_file_extent_calc_inline_size(size);
3429 ret = btrfs_truncate_item(trans, root, path,
3432 } else if (root->ref_cows) {
3433 inode_sub_bytes(inode, item_end + 1 -
3439 if (!pending_del_nr) {
3440 /* no pending yet, add ourselves */
3441 pending_del_slot = path->slots[0];
3443 } else if (pending_del_nr &&
3444 path->slots[0] + 1 == pending_del_slot) {
3445 /* hop on the pending chunk */
3447 pending_del_slot = path->slots[0];
3454 if (found_extent && (root->ref_cows ||
3455 root == root->fs_info->tree_root)) {
3456 btrfs_set_path_blocking(path);
3457 ret = btrfs_free_extent(trans, root, extent_start,
3458 extent_num_bytes, 0,
3459 btrfs_header_owner(leaf),
3460 inode->i_ino, extent_offset);
3464 if (found_type == BTRFS_INODE_ITEM_KEY)
3467 if (path->slots[0] == 0 ||
3468 path->slots[0] != pending_del_slot) {
3469 if (root->ref_cows) {
3473 if (pending_del_nr) {
3474 ret = btrfs_del_items(trans, root, path,
3480 btrfs_release_path(root, path);
3487 if (pending_del_nr) {
3488 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3492 btrfs_free_path(path);
3497 * taken from block_truncate_page, but does cow as it zeros out
3498 * any bytes left in the last page in the file.
3500 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3502 struct inode *inode = mapping->host;
3503 struct btrfs_root *root = BTRFS_I(inode)->root;
3504 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3505 struct btrfs_ordered_extent *ordered;
3506 struct extent_state *cached_state = NULL;
3508 u32 blocksize = root->sectorsize;
3509 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3510 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3516 if ((offset & (blocksize - 1)) == 0)
3518 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3524 page = grab_cache_page(mapping, index);
3526 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3530 page_start = page_offset(page);
3531 page_end = page_start + PAGE_CACHE_SIZE - 1;
3533 if (!PageUptodate(page)) {
3534 ret = btrfs_readpage(NULL, page);
3536 if (page->mapping != mapping) {
3538 page_cache_release(page);
3541 if (!PageUptodate(page)) {
3546 wait_on_page_writeback(page);
3548 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3550 set_page_extent_mapped(page);
3552 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3554 unlock_extent_cached(io_tree, page_start, page_end,
3555 &cached_state, GFP_NOFS);
3557 page_cache_release(page);
3558 btrfs_start_ordered_extent(inode, ordered, 1);
3559 btrfs_put_ordered_extent(ordered);
3563 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3564 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3565 0, 0, &cached_state, GFP_NOFS);
3567 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3570 unlock_extent_cached(io_tree, page_start, page_end,
3571 &cached_state, GFP_NOFS);
3576 if (offset != PAGE_CACHE_SIZE) {
3578 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3579 flush_dcache_page(page);
3582 ClearPageChecked(page);
3583 set_page_dirty(page);
3584 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3589 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3591 page_cache_release(page);
3597 * This function puts in dummy file extents for the area we're creating a hole
3598 * for. So if we are truncating this file to a larger size we need to insert
3599 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3600 * the range between oldsize and size
3602 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3604 struct btrfs_trans_handle *trans;
3605 struct btrfs_root *root = BTRFS_I(inode)->root;
3606 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3607 struct extent_map *em = NULL;
3608 struct extent_state *cached_state = NULL;
3609 u64 mask = root->sectorsize - 1;
3610 u64 hole_start = (oldsize + mask) & ~mask;
3611 u64 block_end = (size + mask) & ~mask;
3617 if (size <= hole_start)
3621 struct btrfs_ordered_extent *ordered;
3622 btrfs_wait_ordered_range(inode, hole_start,
3623 block_end - hole_start);
3624 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3625 &cached_state, GFP_NOFS);
3626 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3629 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3630 &cached_state, GFP_NOFS);
3631 btrfs_put_ordered_extent(ordered);
3634 cur_offset = hole_start;
3636 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3637 block_end - cur_offset, 0);
3638 BUG_ON(IS_ERR(em) || !em);
3639 last_byte = min(extent_map_end(em), block_end);
3640 last_byte = (last_byte + mask) & ~mask;
3641 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3643 hole_size = last_byte - cur_offset;
3645 trans = btrfs_start_transaction(root, 2);
3646 if (IS_ERR(trans)) {
3647 err = PTR_ERR(trans);
3650 btrfs_set_trans_block_group(trans, inode);
3652 err = btrfs_drop_extents(trans, inode, cur_offset,
3653 cur_offset + hole_size,
3658 err = btrfs_insert_file_extent(trans, root,
3659 inode->i_ino, cur_offset, 0,
3660 0, hole_size, 0, hole_size,
3665 btrfs_drop_extent_cache(inode, hole_start,
3668 btrfs_end_transaction(trans, root);
3670 free_extent_map(em);
3672 cur_offset = last_byte;
3673 if (cur_offset >= block_end)
3677 free_extent_map(em);
3678 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3683 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3685 loff_t oldsize = i_size_read(inode);
3688 if (newsize == oldsize)
3691 if (newsize > oldsize) {
3692 i_size_write(inode, newsize);
3693 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3694 truncate_pagecache(inode, oldsize, newsize);
3695 ret = btrfs_cont_expand(inode, oldsize, newsize);
3697 btrfs_setsize(inode, oldsize);
3701 mark_inode_dirty(inode);
3705 * We're truncating a file that used to have good data down to
3706 * zero. Make sure it gets into the ordered flush list so that
3707 * any new writes get down to disk quickly.
3710 BTRFS_I(inode)->ordered_data_close = 1;
3712 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3713 truncate_setsize(inode, newsize);
3714 ret = btrfs_truncate(inode);
3720 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3722 struct inode *inode = dentry->d_inode;
3723 struct btrfs_root *root = BTRFS_I(inode)->root;
3726 if (btrfs_root_readonly(root))
3729 err = inode_change_ok(inode, attr);
3733 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3734 err = btrfs_setsize(inode, attr->ia_size);
3739 if (attr->ia_valid) {
3740 setattr_copy(inode, attr);
3741 mark_inode_dirty(inode);
3743 if (attr->ia_valid & ATTR_MODE)
3744 err = btrfs_acl_chmod(inode);
3750 void btrfs_evict_inode(struct inode *inode)
3752 struct btrfs_trans_handle *trans;
3753 struct btrfs_root *root = BTRFS_I(inode)->root;
3757 trace_btrfs_inode_evict(inode);
3759 truncate_inode_pages(&inode->i_data, 0);
3760 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3761 root == root->fs_info->tree_root))
3764 if (is_bad_inode(inode)) {
3765 btrfs_orphan_del(NULL, inode);
3768 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3769 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3771 if (root->fs_info->log_root_recovering) {
3772 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3776 if (inode->i_nlink > 0) {
3777 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3781 btrfs_i_size_write(inode, 0);
3784 trans = btrfs_start_transaction(root, 0);
3785 BUG_ON(IS_ERR(trans));
3786 btrfs_set_trans_block_group(trans, inode);
3787 trans->block_rsv = root->orphan_block_rsv;
3789 ret = btrfs_block_rsv_check(trans, root,
3790 root->orphan_block_rsv, 0, 5);
3792 BUG_ON(ret != -EAGAIN);
3793 ret = btrfs_commit_transaction(trans, root);
3798 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3802 nr = trans->blocks_used;
3803 btrfs_end_transaction(trans, root);
3805 btrfs_btree_balance_dirty(root, nr);
3810 ret = btrfs_orphan_del(trans, inode);
3814 nr = trans->blocks_used;
3815 btrfs_end_transaction(trans, root);
3816 btrfs_btree_balance_dirty(root, nr);
3818 end_writeback(inode);
3823 * this returns the key found in the dir entry in the location pointer.
3824 * If no dir entries were found, location->objectid is 0.
3826 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3827 struct btrfs_key *location)
3829 const char *name = dentry->d_name.name;
3830 int namelen = dentry->d_name.len;
3831 struct btrfs_dir_item *di;
3832 struct btrfs_path *path;
3833 struct btrfs_root *root = BTRFS_I(dir)->root;
3836 path = btrfs_alloc_path();
3839 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3844 if (!di || IS_ERR(di))
3847 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3849 btrfs_free_path(path);
3852 location->objectid = 0;
3857 * when we hit a tree root in a directory, the btrfs part of the inode
3858 * needs to be changed to reflect the root directory of the tree root. This
3859 * is kind of like crossing a mount point.
3861 static int fixup_tree_root_location(struct btrfs_root *root,
3863 struct dentry *dentry,
3864 struct btrfs_key *location,
3865 struct btrfs_root **sub_root)
3867 struct btrfs_path *path;
3868 struct btrfs_root *new_root;
3869 struct btrfs_root_ref *ref;
3870 struct extent_buffer *leaf;
3874 path = btrfs_alloc_path();
3881 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3882 BTRFS_I(dir)->root->root_key.objectid,
3883 location->objectid);
3890 leaf = path->nodes[0];
3891 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3892 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3893 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3896 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3897 (unsigned long)(ref + 1),
3898 dentry->d_name.len);
3902 btrfs_release_path(root->fs_info->tree_root, path);
3904 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3905 if (IS_ERR(new_root)) {
3906 err = PTR_ERR(new_root);
3910 if (btrfs_root_refs(&new_root->root_item) == 0) {
3915 *sub_root = new_root;
3916 location->objectid = btrfs_root_dirid(&new_root->root_item);
3917 location->type = BTRFS_INODE_ITEM_KEY;
3918 location->offset = 0;
3921 btrfs_free_path(path);
3925 static void inode_tree_add(struct inode *inode)
3927 struct btrfs_root *root = BTRFS_I(inode)->root;
3928 struct btrfs_inode *entry;
3930 struct rb_node *parent;
3932 p = &root->inode_tree.rb_node;
3935 if (inode_unhashed(inode))
3938 spin_lock(&root->inode_lock);
3941 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3943 if (inode->i_ino < entry->vfs_inode.i_ino)
3944 p = &parent->rb_left;
3945 else if (inode->i_ino > entry->vfs_inode.i_ino)
3946 p = &parent->rb_right;
3948 WARN_ON(!(entry->vfs_inode.i_state &
3949 (I_WILL_FREE | I_FREEING)));
3950 rb_erase(parent, &root->inode_tree);
3951 RB_CLEAR_NODE(parent);
3952 spin_unlock(&root->inode_lock);
3956 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3957 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3958 spin_unlock(&root->inode_lock);
3961 static void inode_tree_del(struct inode *inode)
3963 struct btrfs_root *root = BTRFS_I(inode)->root;
3966 spin_lock(&root->inode_lock);
3967 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3968 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3969 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3970 empty = RB_EMPTY_ROOT(&root->inode_tree);
3972 spin_unlock(&root->inode_lock);
3975 * Free space cache has inodes in the tree root, but the tree root has a
3976 * root_refs of 0, so this could end up dropping the tree root as a
3977 * snapshot, so we need the extra !root->fs_info->tree_root check to
3978 * make sure we don't drop it.
3980 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3981 root != root->fs_info->tree_root) {
3982 synchronize_srcu(&root->fs_info->subvol_srcu);
3983 spin_lock(&root->inode_lock);
3984 empty = RB_EMPTY_ROOT(&root->inode_tree);
3985 spin_unlock(&root->inode_lock);
3987 btrfs_add_dead_root(root);
3991 int btrfs_invalidate_inodes(struct btrfs_root *root)
3993 struct rb_node *node;
3994 struct rb_node *prev;
3995 struct btrfs_inode *entry;
3996 struct inode *inode;
3999 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4001 spin_lock(&root->inode_lock);
4003 node = root->inode_tree.rb_node;
4007 entry = rb_entry(node, struct btrfs_inode, rb_node);
4009 if (objectid < entry->vfs_inode.i_ino)
4010 node = node->rb_left;
4011 else if (objectid > entry->vfs_inode.i_ino)
4012 node = node->rb_right;
4018 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4019 if (objectid <= entry->vfs_inode.i_ino) {
4023 prev = rb_next(prev);
4027 entry = rb_entry(node, struct btrfs_inode, rb_node);
4028 objectid = entry->vfs_inode.i_ino + 1;
4029 inode = igrab(&entry->vfs_inode);
4031 spin_unlock(&root->inode_lock);
4032 if (atomic_read(&inode->i_count) > 1)
4033 d_prune_aliases(inode);
4035 * btrfs_drop_inode will have it removed from
4036 * the inode cache when its usage count
4041 spin_lock(&root->inode_lock);
4045 if (cond_resched_lock(&root->inode_lock))
4048 node = rb_next(node);
4050 spin_unlock(&root->inode_lock);
4054 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4056 struct btrfs_iget_args *args = p;
4057 inode->i_ino = args->ino;
4058 BTRFS_I(inode)->root = args->root;
4059 btrfs_set_inode_space_info(args->root, inode);
4063 static int btrfs_find_actor(struct inode *inode, void *opaque)
4065 struct btrfs_iget_args *args = opaque;
4066 return args->ino == inode->i_ino &&
4067 args->root == BTRFS_I(inode)->root;
4070 static struct inode *btrfs_iget_locked(struct super_block *s,
4072 struct btrfs_root *root)
4074 struct inode *inode;
4075 struct btrfs_iget_args args;
4076 args.ino = objectid;
4079 inode = iget5_locked(s, objectid, btrfs_find_actor,
4080 btrfs_init_locked_inode,
4085 /* Get an inode object given its location and corresponding root.
4086 * Returns in *is_new if the inode was read from disk
4088 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4089 struct btrfs_root *root, int *new)
4091 struct inode *inode;
4093 inode = btrfs_iget_locked(s, location->objectid, root);
4095 return ERR_PTR(-ENOMEM);
4097 if (inode->i_state & I_NEW) {
4098 BTRFS_I(inode)->root = root;
4099 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4100 btrfs_read_locked_inode(inode);
4101 inode_tree_add(inode);
4102 unlock_new_inode(inode);
4110 static struct inode *new_simple_dir(struct super_block *s,
4111 struct btrfs_key *key,
4112 struct btrfs_root *root)
4114 struct inode *inode = new_inode(s);
4117 return ERR_PTR(-ENOMEM);
4119 BTRFS_I(inode)->root = root;
4120 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4121 BTRFS_I(inode)->dummy_inode = 1;
4123 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4124 inode->i_op = &simple_dir_inode_operations;
4125 inode->i_fop = &simple_dir_operations;
4126 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4127 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4132 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4134 struct inode *inode;
4135 struct btrfs_root *root = BTRFS_I(dir)->root;
4136 struct btrfs_root *sub_root = root;
4137 struct btrfs_key location;
4141 if (dentry->d_name.len > BTRFS_NAME_LEN)
4142 return ERR_PTR(-ENAMETOOLONG);
4144 ret = btrfs_inode_by_name(dir, dentry, &location);
4147 return ERR_PTR(ret);
4149 if (location.objectid == 0)
4152 if (location.type == BTRFS_INODE_ITEM_KEY) {
4153 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4157 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4159 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4160 ret = fixup_tree_root_location(root, dir, dentry,
4161 &location, &sub_root);
4164 inode = ERR_PTR(ret);
4166 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4168 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4170 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4172 if (!IS_ERR(inode) && root != sub_root) {
4173 down_read(&root->fs_info->cleanup_work_sem);
4174 if (!(inode->i_sb->s_flags & MS_RDONLY))
4175 ret = btrfs_orphan_cleanup(sub_root);
4176 up_read(&root->fs_info->cleanup_work_sem);
4178 inode = ERR_PTR(ret);
4184 static int btrfs_dentry_delete(const struct dentry *dentry)
4186 struct btrfs_root *root;
4188 if (!dentry->d_inode && !IS_ROOT(dentry))
4189 dentry = dentry->d_parent;
4191 if (dentry->d_inode) {
4192 root = BTRFS_I(dentry->d_inode)->root;
4193 if (btrfs_root_refs(&root->root_item) == 0)
4199 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4200 struct nameidata *nd)
4202 struct inode *inode;
4204 inode = btrfs_lookup_dentry(dir, dentry);
4206 return ERR_CAST(inode);
4208 return d_splice_alias(inode, dentry);
4211 static unsigned char btrfs_filetype_table[] = {
4212 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4215 static int btrfs_real_readdir(struct file *filp, void *dirent,
4218 struct inode *inode = filp->f_dentry->d_inode;
4219 struct btrfs_root *root = BTRFS_I(inode)->root;
4220 struct btrfs_item *item;
4221 struct btrfs_dir_item *di;
4222 struct btrfs_key key;
4223 struct btrfs_key found_key;
4224 struct btrfs_path *path;
4226 struct extent_buffer *leaf;
4228 unsigned char d_type;
4233 int key_type = BTRFS_DIR_INDEX_KEY;
4238 /* FIXME, use a real flag for deciding about the key type */
4239 if (root->fs_info->tree_root == root)
4240 key_type = BTRFS_DIR_ITEM_KEY;
4242 /* special case for "." */
4243 if (filp->f_pos == 0) {
4244 over = filldir(dirent, ".", 1,
4251 /* special case for .., just use the back ref */
4252 if (filp->f_pos == 1) {
4253 u64 pino = parent_ino(filp->f_path.dentry);
4254 over = filldir(dirent, "..", 2,
4260 path = btrfs_alloc_path();
4263 btrfs_set_key_type(&key, key_type);
4264 key.offset = filp->f_pos;
4265 key.objectid = inode->i_ino;
4267 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4272 leaf = path->nodes[0];
4273 slot = path->slots[0];
4274 if (slot >= btrfs_header_nritems(leaf)) {
4275 ret = btrfs_next_leaf(root, path);
4283 item = btrfs_item_nr(leaf, slot);
4284 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4286 if (found_key.objectid != key.objectid)
4288 if (btrfs_key_type(&found_key) != key_type)
4290 if (found_key.offset < filp->f_pos)
4293 filp->f_pos = found_key.offset;
4295 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4297 di_total = btrfs_item_size(leaf, item);
4299 while (di_cur < di_total) {
4300 struct btrfs_key location;
4302 if (verify_dir_item(root, leaf, di))
4305 name_len = btrfs_dir_name_len(leaf, di);
4306 if (name_len <= sizeof(tmp_name)) {
4307 name_ptr = tmp_name;
4309 name_ptr = kmalloc(name_len, GFP_NOFS);
4315 read_extent_buffer(leaf, name_ptr,
4316 (unsigned long)(di + 1), name_len);
4318 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4319 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4321 /* is this a reference to our own snapshot? If so
4324 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4325 location.objectid == root->root_key.objectid) {
4329 over = filldir(dirent, name_ptr, name_len,
4330 found_key.offset, location.objectid,
4334 if (name_ptr != tmp_name)
4339 di_len = btrfs_dir_name_len(leaf, di) +
4340 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4342 di = (struct btrfs_dir_item *)((char *)di + di_len);
4348 /* Reached end of directory/root. Bump pos past the last item. */
4349 if (key_type == BTRFS_DIR_INDEX_KEY)
4351 * 32-bit glibc will use getdents64, but then strtol -
4352 * so the last number we can serve is this.
4354 filp->f_pos = 0x7fffffff;
4360 btrfs_free_path(path);
4364 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4366 struct btrfs_root *root = BTRFS_I(inode)->root;
4367 struct btrfs_trans_handle *trans;
4369 bool nolock = false;
4371 if (BTRFS_I(inode)->dummy_inode)
4375 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4377 if (wbc->sync_mode == WB_SYNC_ALL) {
4379 trans = btrfs_join_transaction_nolock(root, 1);
4381 trans = btrfs_join_transaction(root, 1);
4383 return PTR_ERR(trans);
4384 btrfs_set_trans_block_group(trans, inode);
4386 ret = btrfs_end_transaction_nolock(trans, root);
4388 ret = btrfs_commit_transaction(trans, root);
4394 * This is somewhat expensive, updating the tree every time the
4395 * inode changes. But, it is most likely to find the inode in cache.
4396 * FIXME, needs more benchmarking...there are no reasons other than performance
4397 * to keep or drop this code.
4399 void btrfs_dirty_inode(struct inode *inode)
4401 struct btrfs_root *root = BTRFS_I(inode)->root;
4402 struct btrfs_trans_handle *trans;
4405 if (BTRFS_I(inode)->dummy_inode)
4408 trans = btrfs_join_transaction(root, 1);
4409 BUG_ON(IS_ERR(trans));
4410 btrfs_set_trans_block_group(trans, inode);
4412 ret = btrfs_update_inode(trans, root, inode);
4413 if (ret && ret == -ENOSPC) {
4414 /* whoops, lets try again with the full transaction */
4415 btrfs_end_transaction(trans, root);
4416 trans = btrfs_start_transaction(root, 1);
4417 if (IS_ERR(trans)) {
4418 if (printk_ratelimit()) {
4419 printk(KERN_ERR "btrfs: fail to "
4420 "dirty inode %lu error %ld\n",
4421 inode->i_ino, PTR_ERR(trans));
4425 btrfs_set_trans_block_group(trans, inode);
4427 ret = btrfs_update_inode(trans, root, inode);
4429 if (printk_ratelimit()) {
4430 printk(KERN_ERR "btrfs: fail to "
4431 "dirty inode %lu error %d\n",
4436 btrfs_end_transaction(trans, root);
4440 * find the highest existing sequence number in a directory
4441 * and then set the in-memory index_cnt variable to reflect
4442 * free sequence numbers
4444 static int btrfs_set_inode_index_count(struct inode *inode)
4446 struct btrfs_root *root = BTRFS_I(inode)->root;
4447 struct btrfs_key key, found_key;
4448 struct btrfs_path *path;
4449 struct extent_buffer *leaf;
4452 key.objectid = inode->i_ino;
4453 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4454 key.offset = (u64)-1;
4456 path = btrfs_alloc_path();
4460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4463 /* FIXME: we should be able to handle this */
4469 * MAGIC NUMBER EXPLANATION:
4470 * since we search a directory based on f_pos we have to start at 2
4471 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4472 * else has to start at 2
4474 if (path->slots[0] == 0) {
4475 BTRFS_I(inode)->index_cnt = 2;
4481 leaf = path->nodes[0];
4482 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4484 if (found_key.objectid != inode->i_ino ||
4485 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4486 BTRFS_I(inode)->index_cnt = 2;
4490 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4492 btrfs_free_path(path);
4497 * helper to find a free sequence number in a given directory. This current
4498 * code is very simple, later versions will do smarter things in the btree
4500 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4504 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4505 ret = btrfs_set_inode_index_count(dir);
4510 *index = BTRFS_I(dir)->index_cnt;
4511 BTRFS_I(dir)->index_cnt++;
4516 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4517 struct btrfs_root *root,
4519 const char *name, int name_len,
4520 u64 ref_objectid, u64 objectid,
4521 u64 alloc_hint, int mode, u64 *index)
4523 struct inode *inode;
4524 struct btrfs_inode_item *inode_item;
4525 struct btrfs_key *location;
4526 struct btrfs_path *path;
4527 struct btrfs_inode_ref *ref;
4528 struct btrfs_key key[2];
4534 path = btrfs_alloc_path();
4537 inode = new_inode(root->fs_info->sb);
4539 btrfs_free_path(path);
4540 return ERR_PTR(-ENOMEM);
4544 trace_btrfs_inode_request(dir);
4546 ret = btrfs_set_inode_index(dir, index);
4548 btrfs_free_path(path);
4550 return ERR_PTR(ret);
4554 * index_cnt is ignored for everything but a dir,
4555 * btrfs_get_inode_index_count has an explanation for the magic
4558 BTRFS_I(inode)->index_cnt = 2;
4559 BTRFS_I(inode)->root = root;
4560 BTRFS_I(inode)->generation = trans->transid;
4561 inode->i_generation = BTRFS_I(inode)->generation;
4562 btrfs_set_inode_space_info(root, inode);
4568 BTRFS_I(inode)->block_group =
4569 btrfs_find_block_group(root, 0, alloc_hint, owner);
4571 key[0].objectid = objectid;
4572 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4575 key[1].objectid = objectid;
4576 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4577 key[1].offset = ref_objectid;
4579 sizes[0] = sizeof(struct btrfs_inode_item);
4580 sizes[1] = name_len + sizeof(*ref);
4582 path->leave_spinning = 1;
4583 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4587 inode_init_owner(inode, dir, mode);
4588 inode->i_ino = objectid;
4589 inode_set_bytes(inode, 0);
4590 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4591 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4592 struct btrfs_inode_item);
4593 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4595 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4596 struct btrfs_inode_ref);
4597 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4598 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4599 ptr = (unsigned long)(ref + 1);
4600 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4602 btrfs_mark_buffer_dirty(path->nodes[0]);
4603 btrfs_free_path(path);
4605 location = &BTRFS_I(inode)->location;
4606 location->objectid = objectid;
4607 location->offset = 0;
4608 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4610 btrfs_inherit_iflags(inode, dir);
4612 if ((mode & S_IFREG)) {
4613 if (btrfs_test_opt(root, NODATASUM))
4614 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4615 if (btrfs_test_opt(root, NODATACOW) ||
4616 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4617 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4620 insert_inode_hash(inode);
4621 inode_tree_add(inode);
4623 trace_btrfs_inode_new(inode);
4628 BTRFS_I(dir)->index_cnt--;
4629 btrfs_free_path(path);
4631 return ERR_PTR(ret);
4634 static inline u8 btrfs_inode_type(struct inode *inode)
4636 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4640 * utility function to add 'inode' into 'parent_inode' with
4641 * a give name and a given sequence number.
4642 * if 'add_backref' is true, also insert a backref from the
4643 * inode to the parent directory.
4645 int btrfs_add_link(struct btrfs_trans_handle *trans,
4646 struct inode *parent_inode, struct inode *inode,
4647 const char *name, int name_len, int add_backref, u64 index)
4650 struct btrfs_key key;
4651 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4653 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4654 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4656 key.objectid = inode->i_ino;
4657 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4661 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4662 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4663 key.objectid, root->root_key.objectid,
4664 parent_inode->i_ino,
4665 index, name, name_len);
4666 } else if (add_backref) {
4667 ret = btrfs_insert_inode_ref(trans, root,
4668 name, name_len, inode->i_ino,
4669 parent_inode->i_ino, index);
4673 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4674 parent_inode->i_ino, &key,
4675 btrfs_inode_type(inode), index);
4678 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4680 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4681 ret = btrfs_update_inode(trans, root, parent_inode);
4686 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4687 struct inode *dir, struct dentry *dentry,
4688 struct inode *inode, int backref, u64 index)
4690 int err = btrfs_add_link(trans, dir, inode,
4691 dentry->d_name.name, dentry->d_name.len,
4694 d_instantiate(dentry, inode);
4702 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4703 int mode, dev_t rdev)
4705 struct btrfs_trans_handle *trans;
4706 struct btrfs_root *root = BTRFS_I(dir)->root;
4707 struct inode *inode = NULL;
4711 unsigned long nr = 0;
4714 if (!new_valid_dev(rdev))
4717 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4722 * 2 for inode item and ref
4724 * 1 for xattr if selinux is on
4726 trans = btrfs_start_transaction(root, 5);
4728 return PTR_ERR(trans);
4730 btrfs_set_trans_block_group(trans, dir);
4732 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4733 dentry->d_name.len, dir->i_ino, objectid,
4734 BTRFS_I(dir)->block_group, mode, &index);
4735 if (IS_ERR(inode)) {
4736 err = PTR_ERR(inode);
4740 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4746 btrfs_set_trans_block_group(trans, inode);
4747 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4751 inode->i_op = &btrfs_special_inode_operations;
4752 init_special_inode(inode, inode->i_mode, rdev);
4753 btrfs_update_inode(trans, root, inode);
4755 btrfs_update_inode_block_group(trans, inode);
4756 btrfs_update_inode_block_group(trans, dir);
4758 nr = trans->blocks_used;
4759 btrfs_end_transaction_throttle(trans, root);
4760 btrfs_btree_balance_dirty(root, nr);
4762 inode_dec_link_count(inode);
4768 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4769 int mode, struct nameidata *nd)
4771 struct btrfs_trans_handle *trans;
4772 struct btrfs_root *root = BTRFS_I(dir)->root;
4773 struct inode *inode = NULL;
4776 unsigned long nr = 0;
4780 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4784 * 2 for inode item and ref
4786 * 1 for xattr if selinux is on
4788 trans = btrfs_start_transaction(root, 5);
4790 return PTR_ERR(trans);
4792 btrfs_set_trans_block_group(trans, dir);
4794 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4795 dentry->d_name.len, dir->i_ino, objectid,
4796 BTRFS_I(dir)->block_group, mode, &index);
4797 if (IS_ERR(inode)) {
4798 err = PTR_ERR(inode);
4802 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4808 btrfs_set_trans_block_group(trans, inode);
4809 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4813 inode->i_mapping->a_ops = &btrfs_aops;
4814 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4815 inode->i_fop = &btrfs_file_operations;
4816 inode->i_op = &btrfs_file_inode_operations;
4817 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4819 btrfs_update_inode_block_group(trans, inode);
4820 btrfs_update_inode_block_group(trans, dir);
4822 nr = trans->blocks_used;
4823 btrfs_end_transaction_throttle(trans, root);
4825 inode_dec_link_count(inode);
4828 btrfs_btree_balance_dirty(root, nr);
4832 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4833 struct dentry *dentry)
4835 struct btrfs_trans_handle *trans;
4836 struct btrfs_root *root = BTRFS_I(dir)->root;
4837 struct inode *inode = old_dentry->d_inode;
4839 unsigned long nr = 0;
4843 /* do not allow sys_link's with other subvols of the same device */
4844 if (root->objectid != BTRFS_I(inode)->root->objectid)
4847 if (inode->i_nlink == ~0U)
4850 err = btrfs_set_inode_index(dir, &index);
4855 * 2 items for inode and inode ref
4856 * 2 items for dir items
4857 * 1 item for parent inode
4859 trans = btrfs_start_transaction(root, 5);
4860 if (IS_ERR(trans)) {
4861 err = PTR_ERR(trans);
4865 btrfs_inc_nlink(inode);
4866 inode->i_ctime = CURRENT_TIME;
4868 btrfs_set_trans_block_group(trans, dir);
4871 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4876 struct dentry *parent = dget_parent(dentry);
4877 btrfs_update_inode_block_group(trans, dir);
4878 err = btrfs_update_inode(trans, root, inode);
4880 btrfs_log_new_name(trans, inode, NULL, parent);
4884 nr = trans->blocks_used;
4885 btrfs_end_transaction_throttle(trans, root);
4888 inode_dec_link_count(inode);
4891 btrfs_btree_balance_dirty(root, nr);
4895 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4897 struct inode *inode = NULL;
4898 struct btrfs_trans_handle *trans;
4899 struct btrfs_root *root = BTRFS_I(dir)->root;
4901 int drop_on_err = 0;
4904 unsigned long nr = 1;
4906 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4911 * 2 items for inode and ref
4912 * 2 items for dir items
4913 * 1 for xattr if selinux is on
4915 trans = btrfs_start_transaction(root, 5);
4917 return PTR_ERR(trans);
4918 btrfs_set_trans_block_group(trans, dir);
4920 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4921 dentry->d_name.len, dir->i_ino, objectid,
4922 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4924 if (IS_ERR(inode)) {
4925 err = PTR_ERR(inode);
4931 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4935 inode->i_op = &btrfs_dir_inode_operations;
4936 inode->i_fop = &btrfs_dir_file_operations;
4937 btrfs_set_trans_block_group(trans, inode);
4939 btrfs_i_size_write(inode, 0);
4940 err = btrfs_update_inode(trans, root, inode);
4944 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4945 dentry->d_name.len, 0, index);
4949 d_instantiate(dentry, inode);
4951 btrfs_update_inode_block_group(trans, inode);
4952 btrfs_update_inode_block_group(trans, dir);
4955 nr = trans->blocks_used;
4956 btrfs_end_transaction_throttle(trans, root);
4959 btrfs_btree_balance_dirty(root, nr);
4963 /* helper for btfs_get_extent. Given an existing extent in the tree,
4964 * and an extent that you want to insert, deal with overlap and insert
4965 * the new extent into the tree.
4967 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4968 struct extent_map *existing,
4969 struct extent_map *em,
4970 u64 map_start, u64 map_len)
4974 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4975 start_diff = map_start - em->start;
4976 em->start = map_start;
4978 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4979 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4980 em->block_start += start_diff;
4981 em->block_len -= start_diff;
4983 return add_extent_mapping(em_tree, em);
4986 static noinline int uncompress_inline(struct btrfs_path *path,
4987 struct inode *inode, struct page *page,
4988 size_t pg_offset, u64 extent_offset,
4989 struct btrfs_file_extent_item *item)
4992 struct extent_buffer *leaf = path->nodes[0];
4995 unsigned long inline_size;
4999 WARN_ON(pg_offset != 0);
5000 compress_type = btrfs_file_extent_compression(leaf, item);
5001 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5002 inline_size = btrfs_file_extent_inline_item_len(leaf,
5003 btrfs_item_nr(leaf, path->slots[0]));
5004 tmp = kmalloc(inline_size, GFP_NOFS);
5007 ptr = btrfs_file_extent_inline_start(item);
5009 read_extent_buffer(leaf, tmp, ptr, inline_size);
5011 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5012 ret = btrfs_decompress(compress_type, tmp, page,
5013 extent_offset, inline_size, max_size);
5015 char *kaddr = kmap_atomic(page, KM_USER0);
5016 unsigned long copy_size = min_t(u64,
5017 PAGE_CACHE_SIZE - pg_offset,
5018 max_size - extent_offset);
5019 memset(kaddr + pg_offset, 0, copy_size);
5020 kunmap_atomic(kaddr, KM_USER0);
5027 * a bit scary, this does extent mapping from logical file offset to the disk.
5028 * the ugly parts come from merging extents from the disk with the in-ram
5029 * representation. This gets more complex because of the data=ordered code,
5030 * where the in-ram extents might be locked pending data=ordered completion.
5032 * This also copies inline extents directly into the page.
5035 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5036 size_t pg_offset, u64 start, u64 len,
5042 u64 extent_start = 0;
5044 u64 objectid = inode->i_ino;
5046 struct btrfs_path *path = NULL;
5047 struct btrfs_root *root = BTRFS_I(inode)->root;
5048 struct btrfs_file_extent_item *item;
5049 struct extent_buffer *leaf;
5050 struct btrfs_key found_key;
5051 struct extent_map *em = NULL;
5052 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5053 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5054 struct btrfs_trans_handle *trans = NULL;
5058 read_lock(&em_tree->lock);
5059 em = lookup_extent_mapping(em_tree, start, len);
5061 em->bdev = root->fs_info->fs_devices->latest_bdev;
5062 read_unlock(&em_tree->lock);
5065 if (em->start > start || em->start + em->len <= start)
5066 free_extent_map(em);
5067 else if (em->block_start == EXTENT_MAP_INLINE && page)
5068 free_extent_map(em);
5072 em = alloc_extent_map(GFP_NOFS);
5077 em->bdev = root->fs_info->fs_devices->latest_bdev;
5078 em->start = EXTENT_MAP_HOLE;
5079 em->orig_start = EXTENT_MAP_HOLE;
5081 em->block_len = (u64)-1;
5084 path = btrfs_alloc_path();
5088 ret = btrfs_lookup_file_extent(trans, root, path,
5089 objectid, start, trans != NULL);
5096 if (path->slots[0] == 0)
5101 leaf = path->nodes[0];
5102 item = btrfs_item_ptr(leaf, path->slots[0],
5103 struct btrfs_file_extent_item);
5104 /* are we inside the extent that was found? */
5105 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5106 found_type = btrfs_key_type(&found_key);
5107 if (found_key.objectid != objectid ||
5108 found_type != BTRFS_EXTENT_DATA_KEY) {
5112 found_type = btrfs_file_extent_type(leaf, item);
5113 extent_start = found_key.offset;
5114 compress_type = btrfs_file_extent_compression(leaf, item);
5115 if (found_type == BTRFS_FILE_EXTENT_REG ||
5116 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5117 extent_end = extent_start +
5118 btrfs_file_extent_num_bytes(leaf, item);
5119 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5121 size = btrfs_file_extent_inline_len(leaf, item);
5122 extent_end = (extent_start + size + root->sectorsize - 1) &
5123 ~((u64)root->sectorsize - 1);
5126 if (start >= extent_end) {
5128 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5129 ret = btrfs_next_leaf(root, path);
5136 leaf = path->nodes[0];
5138 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5139 if (found_key.objectid != objectid ||
5140 found_key.type != BTRFS_EXTENT_DATA_KEY)
5142 if (start + len <= found_key.offset)
5145 em->len = found_key.offset - start;
5149 if (found_type == BTRFS_FILE_EXTENT_REG ||
5150 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5151 em->start = extent_start;
5152 em->len = extent_end - extent_start;
5153 em->orig_start = extent_start -
5154 btrfs_file_extent_offset(leaf, item);
5155 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5157 em->block_start = EXTENT_MAP_HOLE;
5160 if (compress_type != BTRFS_COMPRESS_NONE) {
5161 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5162 em->compress_type = compress_type;
5163 em->block_start = bytenr;
5164 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5167 bytenr += btrfs_file_extent_offset(leaf, item);
5168 em->block_start = bytenr;
5169 em->block_len = em->len;
5170 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5171 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5174 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5178 size_t extent_offset;
5181 em->block_start = EXTENT_MAP_INLINE;
5182 if (!page || create) {
5183 em->start = extent_start;
5184 em->len = extent_end - extent_start;
5188 size = btrfs_file_extent_inline_len(leaf, item);
5189 extent_offset = page_offset(page) + pg_offset - extent_start;
5190 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5191 size - extent_offset);
5192 em->start = extent_start + extent_offset;
5193 em->len = (copy_size + root->sectorsize - 1) &
5194 ~((u64)root->sectorsize - 1);
5195 em->orig_start = EXTENT_MAP_INLINE;
5196 if (compress_type) {
5197 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5198 em->compress_type = compress_type;
5200 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5201 if (create == 0 && !PageUptodate(page)) {
5202 if (btrfs_file_extent_compression(leaf, item) !=
5203 BTRFS_COMPRESS_NONE) {
5204 ret = uncompress_inline(path, inode, page,
5206 extent_offset, item);
5210 read_extent_buffer(leaf, map + pg_offset, ptr,
5212 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5213 memset(map + pg_offset + copy_size, 0,
5214 PAGE_CACHE_SIZE - pg_offset -
5219 flush_dcache_page(page);
5220 } else if (create && PageUptodate(page)) {
5224 free_extent_map(em);
5226 btrfs_release_path(root, path);
5227 trans = btrfs_join_transaction(root, 1);
5229 return ERR_CAST(trans);
5233 write_extent_buffer(leaf, map + pg_offset, ptr,
5236 btrfs_mark_buffer_dirty(leaf);
5238 set_extent_uptodate(io_tree, em->start,
5239 extent_map_end(em) - 1, NULL, GFP_NOFS);
5242 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5249 em->block_start = EXTENT_MAP_HOLE;
5250 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5252 btrfs_release_path(root, path);
5253 if (em->start > start || extent_map_end(em) <= start) {
5254 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5255 "[%llu %llu]\n", (unsigned long long)em->start,
5256 (unsigned long long)em->len,
5257 (unsigned long long)start,
5258 (unsigned long long)len);
5264 write_lock(&em_tree->lock);
5265 ret = add_extent_mapping(em_tree, em);
5266 /* it is possible that someone inserted the extent into the tree
5267 * while we had the lock dropped. It is also possible that
5268 * an overlapping map exists in the tree
5270 if (ret == -EEXIST) {
5271 struct extent_map *existing;
5275 existing = lookup_extent_mapping(em_tree, start, len);
5276 if (existing && (existing->start > start ||
5277 existing->start + existing->len <= start)) {
5278 free_extent_map(existing);
5282 existing = lookup_extent_mapping(em_tree, em->start,
5285 err = merge_extent_mapping(em_tree, existing,
5288 free_extent_map(existing);
5290 free_extent_map(em);
5295 free_extent_map(em);
5299 free_extent_map(em);
5304 write_unlock(&em_tree->lock);
5307 trace_btrfs_get_extent(root, em);
5310 btrfs_free_path(path);
5312 ret = btrfs_end_transaction(trans, root);
5317 free_extent_map(em);
5318 return ERR_PTR(err);
5323 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5324 size_t pg_offset, u64 start, u64 len,
5327 struct extent_map *em;
5328 struct extent_map *hole_em = NULL;
5329 u64 range_start = start;
5335 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5340 * if our em maps to a hole, there might
5341 * actually be delalloc bytes behind it
5343 if (em->block_start != EXTENT_MAP_HOLE)
5349 /* check to see if we've wrapped (len == -1 or similar) */
5358 /* ok, we didn't find anything, lets look for delalloc */
5359 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5360 end, len, EXTENT_DELALLOC, 1);
5361 found_end = range_start + found;
5362 if (found_end < range_start)
5363 found_end = (u64)-1;
5366 * we didn't find anything useful, return
5367 * the original results from get_extent()
5369 if (range_start > end || found_end <= start) {
5375 /* adjust the range_start to make sure it doesn't
5376 * go backwards from the start they passed in
5378 range_start = max(start,range_start);
5379 found = found_end - range_start;
5382 u64 hole_start = start;
5385 em = alloc_extent_map(GFP_NOFS);
5391 * when btrfs_get_extent can't find anything it
5392 * returns one huge hole
5394 * make sure what it found really fits our range, and
5395 * adjust to make sure it is based on the start from
5399 u64 calc_end = extent_map_end(hole_em);
5401 if (calc_end <= start || (hole_em->start > end)) {
5402 free_extent_map(hole_em);
5405 hole_start = max(hole_em->start, start);
5406 hole_len = calc_end - hole_start;
5410 if (hole_em && range_start > hole_start) {
5411 /* our hole starts before our delalloc, so we
5412 * have to return just the parts of the hole
5413 * that go until the delalloc starts
5415 em->len = min(hole_len,
5416 range_start - hole_start);
5417 em->start = hole_start;
5418 em->orig_start = hole_start;
5420 * don't adjust block start at all,
5421 * it is fixed at EXTENT_MAP_HOLE
5423 em->block_start = hole_em->block_start;
5424 em->block_len = hole_len;
5426 em->start = range_start;
5428 em->orig_start = range_start;
5429 em->block_start = EXTENT_MAP_DELALLOC;
5430 em->block_len = found;
5432 } else if (hole_em) {
5437 free_extent_map(hole_em);
5439 free_extent_map(em);
5440 return ERR_PTR(err);
5445 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5446 struct extent_map *em,
5449 struct btrfs_root *root = BTRFS_I(inode)->root;
5450 struct btrfs_trans_handle *trans;
5451 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5452 struct btrfs_key ins;
5455 bool insert = false;
5458 * Ok if the extent map we looked up is a hole and is for the exact
5459 * range we want, there is no reason to allocate a new one, however if
5460 * it is not right then we need to free this one and drop the cache for
5463 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5465 free_extent_map(em);
5468 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5471 trans = btrfs_join_transaction(root, 0);
5473 return ERR_CAST(trans);
5475 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5477 alloc_hint = get_extent_allocation_hint(inode, start, len);
5478 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5479 alloc_hint, (u64)-1, &ins, 1);
5486 em = alloc_extent_map(GFP_NOFS);
5488 em = ERR_PTR(-ENOMEM);
5494 em->orig_start = em->start;
5495 em->len = ins.offset;
5497 em->block_start = ins.objectid;
5498 em->block_len = ins.offset;
5499 em->bdev = root->fs_info->fs_devices->latest_bdev;
5502 * We need to do this because if we're using the original em we searched
5503 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5506 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5509 write_lock(&em_tree->lock);
5510 ret = add_extent_mapping(em_tree, em);
5511 write_unlock(&em_tree->lock);
5514 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5517 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5518 ins.offset, ins.offset, 0);
5520 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5524 btrfs_end_transaction(trans, root);
5529 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5530 * block must be cow'd
5532 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5533 struct inode *inode, u64 offset, u64 len)
5535 struct btrfs_path *path;
5537 struct extent_buffer *leaf;
5538 struct btrfs_root *root = BTRFS_I(inode)->root;
5539 struct btrfs_file_extent_item *fi;
5540 struct btrfs_key key;
5548 path = btrfs_alloc_path();
5552 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5557 slot = path->slots[0];
5560 /* can't find the item, must cow */
5567 leaf = path->nodes[0];
5568 btrfs_item_key_to_cpu(leaf, &key, slot);
5569 if (key.objectid != inode->i_ino ||
5570 key.type != BTRFS_EXTENT_DATA_KEY) {
5571 /* not our file or wrong item type, must cow */
5575 if (key.offset > offset) {
5576 /* Wrong offset, must cow */
5580 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5581 found_type = btrfs_file_extent_type(leaf, fi);
5582 if (found_type != BTRFS_FILE_EXTENT_REG &&
5583 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5584 /* not a regular extent, must cow */
5587 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5588 backref_offset = btrfs_file_extent_offset(leaf, fi);
5590 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5591 if (extent_end < offset + len) {
5592 /* extent doesn't include our full range, must cow */
5596 if (btrfs_extent_readonly(root, disk_bytenr))
5600 * look for other files referencing this extent, if we
5601 * find any we must cow
5603 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5604 key.offset - backref_offset, disk_bytenr))
5608 * adjust disk_bytenr and num_bytes to cover just the bytes
5609 * in this extent we are about to write. If there
5610 * are any csums in that range we have to cow in order
5611 * to keep the csums correct
5613 disk_bytenr += backref_offset;
5614 disk_bytenr += offset - key.offset;
5615 num_bytes = min(offset + len, extent_end) - offset;
5616 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5619 * all of the above have passed, it is safe to overwrite this extent
5624 btrfs_free_path(path);
5628 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5629 struct buffer_head *bh_result, int create)
5631 struct extent_map *em;
5632 struct btrfs_root *root = BTRFS_I(inode)->root;
5633 u64 start = iblock << inode->i_blkbits;
5634 u64 len = bh_result->b_size;
5635 struct btrfs_trans_handle *trans;
5637 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5642 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5643 * io. INLINE is special, and we could probably kludge it in here, but
5644 * it's still buffered so for safety lets just fall back to the generic
5647 * For COMPRESSED we _have_ to read the entire extent in so we can
5648 * decompress it, so there will be buffering required no matter what we
5649 * do, so go ahead and fallback to buffered.
5651 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5652 * to buffered IO. Don't blame me, this is the price we pay for using
5655 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5656 em->block_start == EXTENT_MAP_INLINE) {
5657 free_extent_map(em);
5661 /* Just a good old fashioned hole, return */
5662 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5663 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5664 free_extent_map(em);
5665 /* DIO will do one hole at a time, so just unlock a sector */
5666 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5667 start + root->sectorsize - 1, GFP_NOFS);
5672 * We don't allocate a new extent in the following cases
5674 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5676 * 2) The extent is marked as PREALLOC. We're good to go here and can
5677 * just use the extent.
5681 len = em->len - (start - em->start);
5685 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5686 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5687 em->block_start != EXTENT_MAP_HOLE)) {
5692 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5693 type = BTRFS_ORDERED_PREALLOC;
5695 type = BTRFS_ORDERED_NOCOW;
5696 len = min(len, em->len - (start - em->start));
5697 block_start = em->block_start + (start - em->start);
5700 * we're not going to log anything, but we do need
5701 * to make sure the current transaction stays open
5702 * while we look for nocow cross refs
5704 trans = btrfs_join_transaction(root, 0);
5708 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5709 ret = btrfs_add_ordered_extent_dio(inode, start,
5710 block_start, len, len, type);
5711 btrfs_end_transaction(trans, root);
5713 free_extent_map(em);
5718 btrfs_end_transaction(trans, root);
5722 * this will cow the extent, reset the len in case we changed
5725 len = bh_result->b_size;
5726 em = btrfs_new_extent_direct(inode, em, start, len);
5729 len = min(len, em->len - (start - em->start));
5731 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5732 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5735 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5737 bh_result->b_size = len;
5738 bh_result->b_bdev = em->bdev;
5739 set_buffer_mapped(bh_result);
5740 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5741 set_buffer_new(bh_result);
5743 free_extent_map(em);
5748 struct btrfs_dio_private {
5749 struct inode *inode;
5756 /* number of bios pending for this dio */
5757 atomic_t pending_bios;
5762 struct bio *orig_bio;
5765 static void btrfs_endio_direct_read(struct bio *bio, int err)
5767 struct btrfs_dio_private *dip = bio->bi_private;
5768 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5769 struct bio_vec *bvec = bio->bi_io_vec;
5770 struct inode *inode = dip->inode;
5771 struct btrfs_root *root = BTRFS_I(inode)->root;
5773 u32 *private = dip->csums;
5775 start = dip->logical_offset;
5777 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5778 struct page *page = bvec->bv_page;
5781 unsigned long flags;
5783 local_irq_save(flags);
5784 kaddr = kmap_atomic(page, KM_IRQ0);
5785 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5786 csum, bvec->bv_len);
5787 btrfs_csum_final(csum, (char *)&csum);
5788 kunmap_atomic(kaddr, KM_IRQ0);
5789 local_irq_restore(flags);
5791 flush_dcache_page(bvec->bv_page);
5792 if (csum != *private) {
5793 printk(KERN_ERR "btrfs csum failed ino %lu off"
5794 " %llu csum %u private %u\n",
5795 inode->i_ino, (unsigned long long)start,
5801 start += bvec->bv_len;
5804 } while (bvec <= bvec_end);
5806 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5807 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5808 bio->bi_private = dip->private;
5813 /* If we had a csum failure make sure to clear the uptodate flag */
5815 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5816 dio_end_io(bio, err);
5819 static void btrfs_endio_direct_write(struct bio *bio, int err)
5821 struct btrfs_dio_private *dip = bio->bi_private;
5822 struct inode *inode = dip->inode;
5823 struct btrfs_root *root = BTRFS_I(inode)->root;
5824 struct btrfs_trans_handle *trans;
5825 struct btrfs_ordered_extent *ordered = NULL;
5826 struct extent_state *cached_state = NULL;
5827 u64 ordered_offset = dip->logical_offset;
5828 u64 ordered_bytes = dip->bytes;
5834 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5842 trans = btrfs_join_transaction(root, 1);
5843 if (IS_ERR(trans)) {
5847 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5849 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5850 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5852 ret = btrfs_update_inode(trans, root, inode);
5857 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5858 ordered->file_offset + ordered->len - 1, 0,
5859 &cached_state, GFP_NOFS);
5861 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5862 ret = btrfs_mark_extent_written(trans, inode,
5863 ordered->file_offset,
5864 ordered->file_offset +
5871 ret = insert_reserved_file_extent(trans, inode,
5872 ordered->file_offset,
5878 BTRFS_FILE_EXTENT_REG);
5879 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5880 ordered->file_offset, ordered->len);
5888 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5889 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5891 btrfs_update_inode(trans, root, inode);
5894 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5895 ordered->file_offset + ordered->len - 1,
5896 &cached_state, GFP_NOFS);
5898 btrfs_delalloc_release_metadata(inode, ordered->len);
5899 btrfs_end_transaction(trans, root);
5900 ordered_offset = ordered->file_offset + ordered->len;
5901 btrfs_put_ordered_extent(ordered);
5902 btrfs_put_ordered_extent(ordered);
5906 * our bio might span multiple ordered extents. If we haven't
5907 * completed the accounting for the whole dio, go back and try again
5909 if (ordered_offset < dip->logical_offset + dip->bytes) {
5910 ordered_bytes = dip->logical_offset + dip->bytes -
5915 bio->bi_private = dip->private;
5920 /* If we had an error make sure to clear the uptodate flag */
5922 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5923 dio_end_io(bio, err);
5926 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5927 struct bio *bio, int mirror_num,
5928 unsigned long bio_flags, u64 offset)
5931 struct btrfs_root *root = BTRFS_I(inode)->root;
5932 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5937 static void btrfs_end_dio_bio(struct bio *bio, int err)
5939 struct btrfs_dio_private *dip = bio->bi_private;
5942 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5943 "sector %#Lx len %u err no %d\n",
5944 dip->inode->i_ino, bio->bi_rw,
5945 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5949 * before atomic variable goto zero, we must make sure
5950 * dip->errors is perceived to be set.
5952 smp_mb__before_atomic_dec();
5955 /* if there are more bios still pending for this dio, just exit */
5956 if (!atomic_dec_and_test(&dip->pending_bios))
5960 bio_io_error(dip->orig_bio);
5962 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5963 bio_endio(dip->orig_bio, 0);
5969 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5970 u64 first_sector, gfp_t gfp_flags)
5972 int nr_vecs = bio_get_nr_vecs(bdev);
5973 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5976 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5977 int rw, u64 file_offset, int skip_sum,
5978 u32 *csums, int async_submit)
5980 int write = rw & REQ_WRITE;
5981 struct btrfs_root *root = BTRFS_I(inode)->root;
5985 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5992 if (write && async_submit) {
5993 ret = btrfs_wq_submit_bio(root->fs_info,
5994 inode, rw, bio, 0, 0,
5996 __btrfs_submit_bio_start_direct_io,
5997 __btrfs_submit_bio_done);
6001 * If we aren't doing async submit, calculate the csum of the
6004 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6007 } else if (!skip_sum) {
6008 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6009 file_offset, csums);
6015 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6021 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6024 struct inode *inode = dip->inode;
6025 struct btrfs_root *root = BTRFS_I(inode)->root;
6026 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6028 struct bio *orig_bio = dip->orig_bio;
6029 struct bio_vec *bvec = orig_bio->bi_io_vec;
6030 u64 start_sector = orig_bio->bi_sector;
6031 u64 file_offset = dip->logical_offset;
6035 u32 *csums = dip->csums;
6037 int async_submit = 0;
6038 int write = rw & REQ_WRITE;
6040 map_length = orig_bio->bi_size;
6041 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6042 &map_length, NULL, 0);
6048 if (map_length >= orig_bio->bi_size) {
6054 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6057 bio->bi_private = dip;
6058 bio->bi_end_io = btrfs_end_dio_bio;
6059 atomic_inc(&dip->pending_bios);
6061 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6062 if (unlikely(map_length < submit_len + bvec->bv_len ||
6063 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6064 bvec->bv_offset) < bvec->bv_len)) {
6066 * inc the count before we submit the bio so
6067 * we know the end IO handler won't happen before
6068 * we inc the count. Otherwise, the dip might get freed
6069 * before we're done setting it up
6071 atomic_inc(&dip->pending_bios);
6072 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6073 file_offset, skip_sum,
6074 csums, async_submit);
6077 atomic_dec(&dip->pending_bios);
6081 /* Write's use the ordered csums */
6082 if (!write && !skip_sum)
6083 csums = csums + nr_pages;
6084 start_sector += submit_len >> 9;
6085 file_offset += submit_len;
6090 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6091 start_sector, GFP_NOFS);
6094 bio->bi_private = dip;
6095 bio->bi_end_io = btrfs_end_dio_bio;
6097 map_length = orig_bio->bi_size;
6098 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6099 &map_length, NULL, 0);
6105 submit_len += bvec->bv_len;
6112 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6113 csums, async_submit);
6121 * before atomic variable goto zero, we must
6122 * make sure dip->errors is perceived to be set.
6124 smp_mb__before_atomic_dec();
6125 if (atomic_dec_and_test(&dip->pending_bios))
6126 bio_io_error(dip->orig_bio);
6128 /* bio_end_io() will handle error, so we needn't return it */
6132 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6135 struct btrfs_root *root = BTRFS_I(inode)->root;
6136 struct btrfs_dio_private *dip;
6137 struct bio_vec *bvec = bio->bi_io_vec;
6139 int write = rw & REQ_WRITE;
6142 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6144 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6151 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6152 if (!write && !skip_sum) {
6153 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6161 dip->private = bio->bi_private;
6163 dip->logical_offset = file_offset;
6167 dip->bytes += bvec->bv_len;
6169 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6171 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6172 bio->bi_private = dip;
6174 dip->orig_bio = bio;
6175 atomic_set(&dip->pending_bios, 0);
6178 bio->bi_end_io = btrfs_endio_direct_write;
6180 bio->bi_end_io = btrfs_endio_direct_read;
6182 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6187 * If this is a write, we need to clean up the reserved space and kill
6188 * the ordered extent.
6191 struct btrfs_ordered_extent *ordered;
6192 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6193 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6194 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6195 btrfs_free_reserved_extent(root, ordered->start,
6197 btrfs_put_ordered_extent(ordered);
6198 btrfs_put_ordered_extent(ordered);
6200 bio_endio(bio, ret);
6203 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6204 const struct iovec *iov, loff_t offset,
6205 unsigned long nr_segs)
6211 unsigned blocksize_mask = root->sectorsize - 1;
6212 ssize_t retval = -EINVAL;
6213 loff_t end = offset;
6215 if (offset & blocksize_mask)
6218 /* Check the memory alignment. Blocks cannot straddle pages */
6219 for (seg = 0; seg < nr_segs; seg++) {
6220 addr = (unsigned long)iov[seg].iov_base;
6221 size = iov[seg].iov_len;
6223 if ((addr & blocksize_mask) || (size & blocksize_mask))
6226 /* If this is a write we don't need to check anymore */
6231 * Check to make sure we don't have duplicate iov_base's in this
6232 * iovec, if so return EINVAL, otherwise we'll get csum errors
6233 * when reading back.
6235 for (i = seg + 1; i < nr_segs; i++) {
6236 if (iov[seg].iov_base == iov[i].iov_base)
6244 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6245 const struct iovec *iov, loff_t offset,
6246 unsigned long nr_segs)
6248 struct file *file = iocb->ki_filp;
6249 struct inode *inode = file->f_mapping->host;
6250 struct btrfs_ordered_extent *ordered;
6251 struct extent_state *cached_state = NULL;
6252 u64 lockstart, lockend;
6254 int writing = rw & WRITE;
6256 size_t count = iov_length(iov, nr_segs);
6258 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6264 lockend = offset + count - 1;
6267 ret = btrfs_delalloc_reserve_space(inode, count);
6273 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6274 0, &cached_state, GFP_NOFS);
6276 * We're concerned with the entire range that we're going to be
6277 * doing DIO to, so we need to make sure theres no ordered
6278 * extents in this range.
6280 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6281 lockend - lockstart + 1);
6284 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6285 &cached_state, GFP_NOFS);
6286 btrfs_start_ordered_extent(inode, ordered, 1);
6287 btrfs_put_ordered_extent(ordered);
6292 * we don't use btrfs_set_extent_delalloc because we don't want
6293 * the dirty or uptodate bits
6296 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6297 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6298 EXTENT_DELALLOC, 0, NULL, &cached_state,
6301 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6302 lockend, EXTENT_LOCKED | write_bits,
6303 1, 0, &cached_state, GFP_NOFS);
6308 free_extent_state(cached_state);
6309 cached_state = NULL;
6311 ret = __blockdev_direct_IO(rw, iocb, inode,
6312 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6313 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6314 btrfs_submit_direct, 0);
6316 if (ret < 0 && ret != -EIOCBQUEUED) {
6317 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6318 offset + iov_length(iov, nr_segs) - 1,
6319 EXTENT_LOCKED | write_bits, 1, 0,
6320 &cached_state, GFP_NOFS);
6321 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6323 * We're falling back to buffered, unlock the section we didn't
6326 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6327 offset + iov_length(iov, nr_segs) - 1,
6328 EXTENT_LOCKED | write_bits, 1, 0,
6329 &cached_state, GFP_NOFS);
6332 free_extent_state(cached_state);
6336 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6337 __u64 start, __u64 len)
6339 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6342 int btrfs_readpage(struct file *file, struct page *page)
6344 struct extent_io_tree *tree;
6345 tree = &BTRFS_I(page->mapping->host)->io_tree;
6346 return extent_read_full_page(tree, page, btrfs_get_extent);
6349 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6351 struct extent_io_tree *tree;
6354 if (current->flags & PF_MEMALLOC) {
6355 redirty_page_for_writepage(wbc, page);
6359 tree = &BTRFS_I(page->mapping->host)->io_tree;
6360 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6363 int btrfs_writepages(struct address_space *mapping,
6364 struct writeback_control *wbc)
6366 struct extent_io_tree *tree;
6368 tree = &BTRFS_I(mapping->host)->io_tree;
6369 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6373 btrfs_readpages(struct file *file, struct address_space *mapping,
6374 struct list_head *pages, unsigned nr_pages)
6376 struct extent_io_tree *tree;
6377 tree = &BTRFS_I(mapping->host)->io_tree;
6378 return extent_readpages(tree, mapping, pages, nr_pages,
6381 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6383 struct extent_io_tree *tree;
6384 struct extent_map_tree *map;
6387 tree = &BTRFS_I(page->mapping->host)->io_tree;
6388 map = &BTRFS_I(page->mapping->host)->extent_tree;
6389 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6391 ClearPagePrivate(page);
6392 set_page_private(page, 0);
6393 page_cache_release(page);
6398 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6400 if (PageWriteback(page) || PageDirty(page))
6402 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6405 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6407 struct extent_io_tree *tree;
6408 struct btrfs_ordered_extent *ordered;
6409 struct extent_state *cached_state = NULL;
6410 u64 page_start = page_offset(page);
6411 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6415 * we have the page locked, so new writeback can't start,
6416 * and the dirty bit won't be cleared while we are here.
6418 * Wait for IO on this page so that we can safely clear
6419 * the PagePrivate2 bit and do ordered accounting
6421 wait_on_page_writeback(page);
6423 tree = &BTRFS_I(page->mapping->host)->io_tree;
6425 btrfs_releasepage(page, GFP_NOFS);
6428 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6430 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6434 * IO on this page will never be started, so we need
6435 * to account for any ordered extents now
6437 clear_extent_bit(tree, page_start, page_end,
6438 EXTENT_DIRTY | EXTENT_DELALLOC |
6439 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6440 &cached_state, GFP_NOFS);
6442 * whoever cleared the private bit is responsible
6443 * for the finish_ordered_io
6445 if (TestClearPagePrivate2(page)) {
6446 btrfs_finish_ordered_io(page->mapping->host,
6447 page_start, page_end);
6449 btrfs_put_ordered_extent(ordered);
6450 cached_state = NULL;
6451 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6454 clear_extent_bit(tree, page_start, page_end,
6455 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6456 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6457 __btrfs_releasepage(page, GFP_NOFS);
6459 ClearPageChecked(page);
6460 if (PagePrivate(page)) {
6461 ClearPagePrivate(page);
6462 set_page_private(page, 0);
6463 page_cache_release(page);
6468 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6469 * called from a page fault handler when a page is first dirtied. Hence we must
6470 * be careful to check for EOF conditions here. We set the page up correctly
6471 * for a written page which means we get ENOSPC checking when writing into
6472 * holes and correct delalloc and unwritten extent mapping on filesystems that
6473 * support these features.
6475 * We are not allowed to take the i_mutex here so we have to play games to
6476 * protect against truncate races as the page could now be beyond EOF. Because
6477 * vmtruncate() writes the inode size before removing pages, once we have the
6478 * page lock we can determine safely if the page is beyond EOF. If it is not
6479 * beyond EOF, then the page is guaranteed safe against truncation until we
6482 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6484 struct page *page = vmf->page;
6485 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6486 struct btrfs_root *root = BTRFS_I(inode)->root;
6487 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6488 struct btrfs_ordered_extent *ordered;
6489 struct extent_state *cached_state = NULL;
6491 unsigned long zero_start;
6497 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6501 else /* -ENOSPC, -EIO, etc */
6502 ret = VM_FAULT_SIGBUS;
6506 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6509 size = i_size_read(inode);
6510 page_start = page_offset(page);
6511 page_end = page_start + PAGE_CACHE_SIZE - 1;
6513 if ((page->mapping != inode->i_mapping) ||
6514 (page_start >= size)) {
6515 /* page got truncated out from underneath us */
6518 wait_on_page_writeback(page);
6520 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6522 set_page_extent_mapped(page);
6525 * we can't set the delalloc bits if there are pending ordered
6526 * extents. Drop our locks and wait for them to finish
6528 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6530 unlock_extent_cached(io_tree, page_start, page_end,
6531 &cached_state, GFP_NOFS);
6533 btrfs_start_ordered_extent(inode, ordered, 1);
6534 btrfs_put_ordered_extent(ordered);
6539 * XXX - page_mkwrite gets called every time the page is dirtied, even
6540 * if it was already dirty, so for space accounting reasons we need to
6541 * clear any delalloc bits for the range we are fixing to save. There
6542 * is probably a better way to do this, but for now keep consistent with
6543 * prepare_pages in the normal write path.
6545 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6546 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6547 0, 0, &cached_state, GFP_NOFS);
6549 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6552 unlock_extent_cached(io_tree, page_start, page_end,
6553 &cached_state, GFP_NOFS);
6554 ret = VM_FAULT_SIGBUS;
6559 /* page is wholly or partially inside EOF */
6560 if (page_start + PAGE_CACHE_SIZE > size)
6561 zero_start = size & ~PAGE_CACHE_MASK;
6563 zero_start = PAGE_CACHE_SIZE;
6565 if (zero_start != PAGE_CACHE_SIZE) {
6567 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6568 flush_dcache_page(page);
6571 ClearPageChecked(page);
6572 set_page_dirty(page);
6573 SetPageUptodate(page);
6575 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6576 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6578 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6582 return VM_FAULT_LOCKED;
6584 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6589 static int btrfs_truncate(struct inode *inode)
6591 struct btrfs_root *root = BTRFS_I(inode)->root;
6594 struct btrfs_trans_handle *trans;
6596 u64 mask = root->sectorsize - 1;
6598 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6602 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6603 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6605 trans = btrfs_start_transaction(root, 5);
6607 return PTR_ERR(trans);
6609 btrfs_set_trans_block_group(trans, inode);
6611 ret = btrfs_orphan_add(trans, inode);
6613 btrfs_end_transaction(trans, root);
6617 nr = trans->blocks_used;
6618 btrfs_end_transaction(trans, root);
6619 btrfs_btree_balance_dirty(root, nr);
6621 /* Now start a transaction for the truncate */
6622 trans = btrfs_start_transaction(root, 0);
6624 return PTR_ERR(trans);
6625 btrfs_set_trans_block_group(trans, inode);
6626 trans->block_rsv = root->orphan_block_rsv;
6629 * setattr is responsible for setting the ordered_data_close flag,
6630 * but that is only tested during the last file release. That
6631 * could happen well after the next commit, leaving a great big
6632 * window where new writes may get lost if someone chooses to write
6633 * to this file after truncating to zero
6635 * The inode doesn't have any dirty data here, and so if we commit
6636 * this is a noop. If someone immediately starts writing to the inode
6637 * it is very likely we'll catch some of their writes in this
6638 * transaction, and the commit will find this file on the ordered
6639 * data list with good things to send down.
6641 * This is a best effort solution, there is still a window where
6642 * using truncate to replace the contents of the file will
6643 * end up with a zero length file after a crash.
6645 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6646 btrfs_add_ordered_operation(trans, root, inode);
6650 trans = btrfs_start_transaction(root, 0);
6652 return PTR_ERR(trans);
6653 btrfs_set_trans_block_group(trans, inode);
6654 trans->block_rsv = root->orphan_block_rsv;
6657 ret = btrfs_block_rsv_check(trans, root,
6658 root->orphan_block_rsv, 0, 5);
6659 if (ret == -EAGAIN) {
6660 ret = btrfs_commit_transaction(trans, root);
6670 ret = btrfs_truncate_inode_items(trans, root, inode,
6672 BTRFS_EXTENT_DATA_KEY);
6673 if (ret != -EAGAIN) {
6678 ret = btrfs_update_inode(trans, root, inode);
6684 nr = trans->blocks_used;
6685 btrfs_end_transaction(trans, root);
6687 btrfs_btree_balance_dirty(root, nr);
6690 if (ret == 0 && inode->i_nlink > 0) {
6691 ret = btrfs_orphan_del(trans, inode);
6694 } else if (ret && inode->i_nlink > 0) {
6696 * Failed to do the truncate, remove us from the in memory
6699 ret = btrfs_orphan_del(NULL, inode);
6702 ret = btrfs_update_inode(trans, root, inode);
6706 nr = trans->blocks_used;
6707 ret = btrfs_end_transaction_throttle(trans, root);
6710 btrfs_btree_balance_dirty(root, nr);
6716 * create a new subvolume directory/inode (helper for the ioctl).
6718 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6719 struct btrfs_root *new_root,
6720 u64 new_dirid, u64 alloc_hint)
6722 struct inode *inode;
6726 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6727 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6729 return PTR_ERR(inode);
6730 inode->i_op = &btrfs_dir_inode_operations;
6731 inode->i_fop = &btrfs_dir_file_operations;
6734 btrfs_i_size_write(inode, 0);
6736 err = btrfs_update_inode(trans, new_root, inode);
6743 /* helper function for file defrag and space balancing. This
6744 * forces readahead on a given range of bytes in an inode
6746 unsigned long btrfs_force_ra(struct address_space *mapping,
6747 struct file_ra_state *ra, struct file *file,
6748 pgoff_t offset, pgoff_t last_index)
6750 pgoff_t req_size = last_index - offset + 1;
6752 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6753 return offset + req_size;
6756 struct inode *btrfs_alloc_inode(struct super_block *sb)
6758 struct btrfs_inode *ei;
6759 struct inode *inode;
6761 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6766 ei->space_info = NULL;
6770 ei->last_sub_trans = 0;
6771 ei->logged_trans = 0;
6772 ei->delalloc_bytes = 0;
6773 ei->reserved_bytes = 0;
6774 ei->disk_i_size = 0;
6776 ei->index_cnt = (u64)-1;
6777 ei->last_unlink_trans = 0;
6779 atomic_set(&ei->outstanding_extents, 0);
6780 atomic_set(&ei->reserved_extents, 0);
6782 ei->ordered_data_close = 0;
6783 ei->orphan_meta_reserved = 0;
6784 ei->dummy_inode = 0;
6785 ei->force_compress = BTRFS_COMPRESS_NONE;
6787 inode = &ei->vfs_inode;
6788 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6789 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6790 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6791 mutex_init(&ei->log_mutex);
6792 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6793 INIT_LIST_HEAD(&ei->i_orphan);
6794 INIT_LIST_HEAD(&ei->delalloc_inodes);
6795 INIT_LIST_HEAD(&ei->ordered_operations);
6796 RB_CLEAR_NODE(&ei->rb_node);
6801 static void btrfs_i_callback(struct rcu_head *head)
6803 struct inode *inode = container_of(head, struct inode, i_rcu);
6804 INIT_LIST_HEAD(&inode->i_dentry);
6805 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6808 void btrfs_destroy_inode(struct inode *inode)
6810 struct btrfs_ordered_extent *ordered;
6811 struct btrfs_root *root = BTRFS_I(inode)->root;
6813 WARN_ON(!list_empty(&inode->i_dentry));
6814 WARN_ON(inode->i_data.nrpages);
6815 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6816 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6819 * This can happen where we create an inode, but somebody else also
6820 * created the same inode and we need to destroy the one we already
6827 * Make sure we're properly removed from the ordered operation
6831 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6832 spin_lock(&root->fs_info->ordered_extent_lock);
6833 list_del_init(&BTRFS_I(inode)->ordered_operations);
6834 spin_unlock(&root->fs_info->ordered_extent_lock);
6837 if (root == root->fs_info->tree_root) {
6838 struct btrfs_block_group_cache *block_group;
6840 block_group = btrfs_lookup_block_group(root->fs_info,
6841 BTRFS_I(inode)->block_group);
6842 if (block_group && block_group->inode == inode) {
6843 spin_lock(&block_group->lock);
6844 block_group->inode = NULL;
6845 spin_unlock(&block_group->lock);
6846 btrfs_put_block_group(block_group);
6847 } else if (block_group) {
6848 btrfs_put_block_group(block_group);
6852 spin_lock(&root->orphan_lock);
6853 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6854 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6856 list_del_init(&BTRFS_I(inode)->i_orphan);
6858 spin_unlock(&root->orphan_lock);
6861 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6865 printk(KERN_ERR "btrfs found ordered "
6866 "extent %llu %llu on inode cleanup\n",
6867 (unsigned long long)ordered->file_offset,
6868 (unsigned long long)ordered->len);
6869 btrfs_remove_ordered_extent(inode, ordered);
6870 btrfs_put_ordered_extent(ordered);
6871 btrfs_put_ordered_extent(ordered);
6874 inode_tree_del(inode);
6875 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6877 call_rcu(&inode->i_rcu, btrfs_i_callback);
6880 int btrfs_drop_inode(struct inode *inode)
6882 struct btrfs_root *root = BTRFS_I(inode)->root;
6884 if (btrfs_root_refs(&root->root_item) == 0 &&
6885 root != root->fs_info->tree_root)
6888 return generic_drop_inode(inode);
6891 static void init_once(void *foo)
6893 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6895 inode_init_once(&ei->vfs_inode);
6898 void btrfs_destroy_cachep(void)
6900 if (btrfs_inode_cachep)
6901 kmem_cache_destroy(btrfs_inode_cachep);
6902 if (btrfs_trans_handle_cachep)
6903 kmem_cache_destroy(btrfs_trans_handle_cachep);
6904 if (btrfs_transaction_cachep)
6905 kmem_cache_destroy(btrfs_transaction_cachep);
6906 if (btrfs_path_cachep)
6907 kmem_cache_destroy(btrfs_path_cachep);
6908 if (btrfs_free_space_cachep)
6909 kmem_cache_destroy(btrfs_free_space_cachep);
6912 int btrfs_init_cachep(void)
6914 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6915 sizeof(struct btrfs_inode), 0,
6916 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6917 if (!btrfs_inode_cachep)
6920 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6921 sizeof(struct btrfs_trans_handle), 0,
6922 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6923 if (!btrfs_trans_handle_cachep)
6926 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6927 sizeof(struct btrfs_transaction), 0,
6928 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6929 if (!btrfs_transaction_cachep)
6932 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6933 sizeof(struct btrfs_path), 0,
6934 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6935 if (!btrfs_path_cachep)
6938 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6939 sizeof(struct btrfs_free_space), 0,
6940 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6941 if (!btrfs_free_space_cachep)
6946 btrfs_destroy_cachep();
6950 static int btrfs_getattr(struct vfsmount *mnt,
6951 struct dentry *dentry, struct kstat *stat)
6953 struct inode *inode = dentry->d_inode;
6954 generic_fillattr(inode, stat);
6955 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6956 stat->blksize = PAGE_CACHE_SIZE;
6957 stat->blocks = (inode_get_bytes(inode) +
6958 BTRFS_I(inode)->delalloc_bytes) >> 9;
6963 * If a file is moved, it will inherit the cow and compression flags of the new
6966 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6968 struct btrfs_inode *b_dir = BTRFS_I(dir);
6969 struct btrfs_inode *b_inode = BTRFS_I(inode);
6971 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6972 b_inode->flags |= BTRFS_INODE_NODATACOW;
6974 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6976 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6977 b_inode->flags |= BTRFS_INODE_COMPRESS;
6979 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6982 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6983 struct inode *new_dir, struct dentry *new_dentry)
6985 struct btrfs_trans_handle *trans;
6986 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6987 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6988 struct inode *new_inode = new_dentry->d_inode;
6989 struct inode *old_inode = old_dentry->d_inode;
6990 struct timespec ctime = CURRENT_TIME;
6995 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6998 /* we only allow rename subvolume link between subvolumes */
6999 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7002 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7003 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7006 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7007 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7010 * we're using rename to replace one file with another.
7011 * and the replacement file is large. Start IO on it now so
7012 * we don't add too much work to the end of the transaction
7014 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7015 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7016 filemap_flush(old_inode->i_mapping);
7018 /* close the racy window with snapshot create/destroy ioctl */
7019 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7020 down_read(&root->fs_info->subvol_sem);
7022 * We want to reserve the absolute worst case amount of items. So if
7023 * both inodes are subvols and we need to unlink them then that would
7024 * require 4 item modifications, but if they are both normal inodes it
7025 * would require 5 item modifications, so we'll assume their normal
7026 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7027 * should cover the worst case number of items we'll modify.
7029 trans = btrfs_start_transaction(root, 20);
7030 if (IS_ERR(trans)) {
7031 ret = PTR_ERR(trans);
7035 btrfs_set_trans_block_group(trans, new_dir);
7038 btrfs_record_root_in_trans(trans, dest);
7040 ret = btrfs_set_inode_index(new_dir, &index);
7044 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7045 /* force full log commit if subvolume involved. */
7046 root->fs_info->last_trans_log_full_commit = trans->transid;
7048 ret = btrfs_insert_inode_ref(trans, dest,
7049 new_dentry->d_name.name,
7050 new_dentry->d_name.len,
7052 new_dir->i_ino, index);
7056 * this is an ugly little race, but the rename is required
7057 * to make sure that if we crash, the inode is either at the
7058 * old name or the new one. pinning the log transaction lets
7059 * us make sure we don't allow a log commit to come in after
7060 * we unlink the name but before we add the new name back in.
7062 btrfs_pin_log_trans(root);
7065 * make sure the inode gets flushed if it is replacing
7068 if (new_inode && new_inode->i_size &&
7069 old_inode && S_ISREG(old_inode->i_mode)) {
7070 btrfs_add_ordered_operation(trans, root, old_inode);
7073 old_dir->i_ctime = old_dir->i_mtime = ctime;
7074 new_dir->i_ctime = new_dir->i_mtime = ctime;
7075 old_inode->i_ctime = ctime;
7077 if (old_dentry->d_parent != new_dentry->d_parent)
7078 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7080 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7081 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7082 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7083 old_dentry->d_name.name,
7084 old_dentry->d_name.len);
7086 ret = __btrfs_unlink_inode(trans, root, old_dir,
7087 old_dentry->d_inode,
7088 old_dentry->d_name.name,
7089 old_dentry->d_name.len);
7091 ret = btrfs_update_inode(trans, root, old_inode);
7096 new_inode->i_ctime = CURRENT_TIME;
7097 if (unlikely(new_inode->i_ino ==
7098 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7099 root_objectid = BTRFS_I(new_inode)->location.objectid;
7100 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7102 new_dentry->d_name.name,
7103 new_dentry->d_name.len);
7104 BUG_ON(new_inode->i_nlink == 0);
7106 ret = btrfs_unlink_inode(trans, dest, new_dir,
7107 new_dentry->d_inode,
7108 new_dentry->d_name.name,
7109 new_dentry->d_name.len);
7112 if (new_inode->i_nlink == 0) {
7113 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7118 fixup_inode_flags(new_dir, old_inode);
7120 ret = btrfs_add_link(trans, new_dir, old_inode,
7121 new_dentry->d_name.name,
7122 new_dentry->d_name.len, 0, index);
7125 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7126 struct dentry *parent = dget_parent(new_dentry);
7127 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7129 btrfs_end_log_trans(root);
7132 btrfs_end_transaction_throttle(trans, root);
7134 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7135 up_read(&root->fs_info->subvol_sem);
7141 * some fairly slow code that needs optimization. This walks the list
7142 * of all the inodes with pending delalloc and forces them to disk.
7144 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7146 struct list_head *head = &root->fs_info->delalloc_inodes;
7147 struct btrfs_inode *binode;
7148 struct inode *inode;
7150 if (root->fs_info->sb->s_flags & MS_RDONLY)
7153 spin_lock(&root->fs_info->delalloc_lock);
7154 while (!list_empty(head)) {
7155 binode = list_entry(head->next, struct btrfs_inode,
7157 inode = igrab(&binode->vfs_inode);
7159 list_del_init(&binode->delalloc_inodes);
7160 spin_unlock(&root->fs_info->delalloc_lock);
7162 filemap_flush(inode->i_mapping);
7164 btrfs_add_delayed_iput(inode);
7169 spin_lock(&root->fs_info->delalloc_lock);
7171 spin_unlock(&root->fs_info->delalloc_lock);
7173 /* the filemap_flush will queue IO into the worker threads, but
7174 * we have to make sure the IO is actually started and that
7175 * ordered extents get created before we return
7177 atomic_inc(&root->fs_info->async_submit_draining);
7178 while (atomic_read(&root->fs_info->nr_async_submits) ||
7179 atomic_read(&root->fs_info->async_delalloc_pages)) {
7180 wait_event(root->fs_info->async_submit_wait,
7181 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7182 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7184 atomic_dec(&root->fs_info->async_submit_draining);
7188 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7191 struct btrfs_inode *binode;
7192 struct inode *inode = NULL;
7194 spin_lock(&root->fs_info->delalloc_lock);
7195 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7196 binode = list_entry(root->fs_info->delalloc_inodes.next,
7197 struct btrfs_inode, delalloc_inodes);
7198 inode = igrab(&binode->vfs_inode);
7200 list_move_tail(&binode->delalloc_inodes,
7201 &root->fs_info->delalloc_inodes);
7205 list_del_init(&binode->delalloc_inodes);
7206 cond_resched_lock(&root->fs_info->delalloc_lock);
7208 spin_unlock(&root->fs_info->delalloc_lock);
7212 filemap_write_and_wait(inode->i_mapping);
7214 * We have to do this because compression doesn't
7215 * actually set PG_writeback until it submits the pages
7216 * for IO, which happens in an async thread, so we could
7217 * race and not actually wait for any writeback pages
7218 * because they've not been submitted yet. Technically
7219 * this could still be the case for the ordered stuff
7220 * since the async thread may not have started to do its
7221 * work yet. If this becomes the case then we need to
7222 * figure out a way to make sure that in writepage we
7223 * wait for any async pages to be submitted before
7224 * returning so that fdatawait does what its supposed to
7227 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7229 filemap_flush(inode->i_mapping);
7232 btrfs_add_delayed_iput(inode);
7240 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7241 const char *symname)
7243 struct btrfs_trans_handle *trans;
7244 struct btrfs_root *root = BTRFS_I(dir)->root;
7245 struct btrfs_path *path;
7246 struct btrfs_key key;
7247 struct inode *inode = NULL;
7255 struct btrfs_file_extent_item *ei;
7256 struct extent_buffer *leaf;
7257 unsigned long nr = 0;
7259 name_len = strlen(symname) + 1;
7260 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7261 return -ENAMETOOLONG;
7263 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7267 * 2 items for inode item and ref
7268 * 2 items for dir items
7269 * 1 item for xattr if selinux is on
7271 trans = btrfs_start_transaction(root, 5);
7273 return PTR_ERR(trans);
7275 btrfs_set_trans_block_group(trans, dir);
7277 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7278 dentry->d_name.len, dir->i_ino, objectid,
7279 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7281 if (IS_ERR(inode)) {
7282 err = PTR_ERR(inode);
7286 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7292 btrfs_set_trans_block_group(trans, inode);
7293 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7297 inode->i_mapping->a_ops = &btrfs_aops;
7298 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7299 inode->i_fop = &btrfs_file_operations;
7300 inode->i_op = &btrfs_file_inode_operations;
7301 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7303 btrfs_update_inode_block_group(trans, inode);
7304 btrfs_update_inode_block_group(trans, dir);
7308 path = btrfs_alloc_path();
7310 key.objectid = inode->i_ino;
7312 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7313 datasize = btrfs_file_extent_calc_inline_size(name_len);
7314 err = btrfs_insert_empty_item(trans, root, path, &key,
7320 leaf = path->nodes[0];
7321 ei = btrfs_item_ptr(leaf, path->slots[0],
7322 struct btrfs_file_extent_item);
7323 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7324 btrfs_set_file_extent_type(leaf, ei,
7325 BTRFS_FILE_EXTENT_INLINE);
7326 btrfs_set_file_extent_encryption(leaf, ei, 0);
7327 btrfs_set_file_extent_compression(leaf, ei, 0);
7328 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7329 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7331 ptr = btrfs_file_extent_inline_start(ei);
7332 write_extent_buffer(leaf, symname, ptr, name_len);
7333 btrfs_mark_buffer_dirty(leaf);
7334 btrfs_free_path(path);
7336 inode->i_op = &btrfs_symlink_inode_operations;
7337 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7338 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7339 inode_set_bytes(inode, name_len);
7340 btrfs_i_size_write(inode, name_len - 1);
7341 err = btrfs_update_inode(trans, root, inode);
7346 nr = trans->blocks_used;
7347 btrfs_end_transaction_throttle(trans, root);
7349 inode_dec_link_count(inode);
7352 btrfs_btree_balance_dirty(root, nr);
7356 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7357 u64 start, u64 num_bytes, u64 min_size,
7358 loff_t actual_len, u64 *alloc_hint,
7359 struct btrfs_trans_handle *trans)
7361 struct btrfs_root *root = BTRFS_I(inode)->root;
7362 struct btrfs_key ins;
7363 u64 cur_offset = start;
7366 bool own_trans = true;
7370 while (num_bytes > 0) {
7372 trans = btrfs_start_transaction(root, 3);
7373 if (IS_ERR(trans)) {
7374 ret = PTR_ERR(trans);
7379 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7380 0, *alloc_hint, (u64)-1, &ins, 1);
7383 btrfs_end_transaction(trans, root);
7387 ret = insert_reserved_file_extent(trans, inode,
7388 cur_offset, ins.objectid,
7389 ins.offset, ins.offset,
7390 ins.offset, 0, 0, 0,
7391 BTRFS_FILE_EXTENT_PREALLOC);
7393 btrfs_drop_extent_cache(inode, cur_offset,
7394 cur_offset + ins.offset -1, 0);
7396 num_bytes -= ins.offset;
7397 cur_offset += ins.offset;
7398 *alloc_hint = ins.objectid + ins.offset;
7400 inode->i_ctime = CURRENT_TIME;
7401 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7402 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7403 (actual_len > inode->i_size) &&
7404 (cur_offset > inode->i_size)) {
7405 if (cur_offset > actual_len)
7406 i_size = actual_len;
7408 i_size = cur_offset;
7409 i_size_write(inode, i_size);
7410 btrfs_ordered_update_i_size(inode, i_size, NULL);
7413 ret = btrfs_update_inode(trans, root, inode);
7417 btrfs_end_transaction(trans, root);
7422 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7423 u64 start, u64 num_bytes, u64 min_size,
7424 loff_t actual_len, u64 *alloc_hint)
7426 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7427 min_size, actual_len, alloc_hint,
7431 int btrfs_prealloc_file_range_trans(struct inode *inode,
7432 struct btrfs_trans_handle *trans, int mode,
7433 u64 start, u64 num_bytes, u64 min_size,
7434 loff_t actual_len, u64 *alloc_hint)
7436 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7437 min_size, actual_len, alloc_hint, trans);
7440 static int btrfs_set_page_dirty(struct page *page)
7442 return __set_page_dirty_nobuffers(page);
7445 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7447 struct btrfs_root *root = BTRFS_I(inode)->root;
7449 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7451 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7453 return generic_permission(inode, mask, flags, btrfs_check_acl);
7456 static const struct inode_operations btrfs_dir_inode_operations = {
7457 .getattr = btrfs_getattr,
7458 .lookup = btrfs_lookup,
7459 .create = btrfs_create,
7460 .unlink = btrfs_unlink,
7462 .mkdir = btrfs_mkdir,
7463 .rmdir = btrfs_rmdir,
7464 .rename = btrfs_rename,
7465 .symlink = btrfs_symlink,
7466 .setattr = btrfs_setattr,
7467 .mknod = btrfs_mknod,
7468 .setxattr = btrfs_setxattr,
7469 .getxattr = btrfs_getxattr,
7470 .listxattr = btrfs_listxattr,
7471 .removexattr = btrfs_removexattr,
7472 .permission = btrfs_permission,
7474 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7475 .lookup = btrfs_lookup,
7476 .permission = btrfs_permission,
7479 static const struct file_operations btrfs_dir_file_operations = {
7480 .llseek = generic_file_llseek,
7481 .read = generic_read_dir,
7482 .readdir = btrfs_real_readdir,
7483 .unlocked_ioctl = btrfs_ioctl,
7484 #ifdef CONFIG_COMPAT
7485 .compat_ioctl = btrfs_ioctl,
7487 .release = btrfs_release_file,
7488 .fsync = btrfs_sync_file,
7491 static struct extent_io_ops btrfs_extent_io_ops = {
7492 .fill_delalloc = run_delalloc_range,
7493 .submit_bio_hook = btrfs_submit_bio_hook,
7494 .merge_bio_hook = btrfs_merge_bio_hook,
7495 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7496 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7497 .writepage_start_hook = btrfs_writepage_start_hook,
7498 .readpage_io_failed_hook = btrfs_io_failed_hook,
7499 .set_bit_hook = btrfs_set_bit_hook,
7500 .clear_bit_hook = btrfs_clear_bit_hook,
7501 .merge_extent_hook = btrfs_merge_extent_hook,
7502 .split_extent_hook = btrfs_split_extent_hook,
7506 * btrfs doesn't support the bmap operation because swapfiles
7507 * use bmap to make a mapping of extents in the file. They assume
7508 * these extents won't change over the life of the file and they
7509 * use the bmap result to do IO directly to the drive.
7511 * the btrfs bmap call would return logical addresses that aren't
7512 * suitable for IO and they also will change frequently as COW
7513 * operations happen. So, swapfile + btrfs == corruption.
7515 * For now we're avoiding this by dropping bmap.
7517 static const struct address_space_operations btrfs_aops = {
7518 .readpage = btrfs_readpage,
7519 .writepage = btrfs_writepage,
7520 .writepages = btrfs_writepages,
7521 .readpages = btrfs_readpages,
7522 .direct_IO = btrfs_direct_IO,
7523 .invalidatepage = btrfs_invalidatepage,
7524 .releasepage = btrfs_releasepage,
7525 .set_page_dirty = btrfs_set_page_dirty,
7526 .error_remove_page = generic_error_remove_page,
7529 static const struct address_space_operations btrfs_symlink_aops = {
7530 .readpage = btrfs_readpage,
7531 .writepage = btrfs_writepage,
7532 .invalidatepage = btrfs_invalidatepage,
7533 .releasepage = btrfs_releasepage,
7536 static const struct inode_operations btrfs_file_inode_operations = {
7537 .getattr = btrfs_getattr,
7538 .setattr = btrfs_setattr,
7539 .setxattr = btrfs_setxattr,
7540 .getxattr = btrfs_getxattr,
7541 .listxattr = btrfs_listxattr,
7542 .removexattr = btrfs_removexattr,
7543 .permission = btrfs_permission,
7544 .fiemap = btrfs_fiemap,
7546 static const struct inode_operations btrfs_special_inode_operations = {
7547 .getattr = btrfs_getattr,
7548 .setattr = btrfs_setattr,
7549 .permission = btrfs_permission,
7550 .setxattr = btrfs_setxattr,
7551 .getxattr = btrfs_getxattr,
7552 .listxattr = btrfs_listxattr,
7553 .removexattr = btrfs_removexattr,
7555 static const struct inode_operations btrfs_symlink_inode_operations = {
7556 .readlink = generic_readlink,
7557 .follow_link = page_follow_link_light,
7558 .put_link = page_put_link,
7559 .getattr = btrfs_getattr,
7560 .permission = btrfs_permission,
7561 .setxattr = btrfs_setxattr,
7562 .getxattr = btrfs_getxattr,
7563 .listxattr = btrfs_listxattr,
7564 .removexattr = btrfs_removexattr,
7567 const struct dentry_operations btrfs_dentry_operations = {
7568 .d_delete = btrfs_dentry_delete,