2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 u64 start, size_t size, size_t compressed_size,
115 struct page **compressed_pages)
117 struct btrfs_key key;
118 struct btrfs_path *path;
119 struct extent_buffer *leaf;
120 struct page *page = NULL;
123 struct btrfs_file_extent_item *ei;
126 size_t cur_size = size;
128 unsigned long offset;
130 if (compressed_size && compressed_pages)
131 cur_size = compressed_size;
133 path = btrfs_alloc_path();
137 path->leave_spinning = 1;
138 btrfs_set_trans_block_group(trans, inode);
140 key.objectid = inode->i_ino;
142 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
143 datasize = btrfs_file_extent_calc_inline_size(cur_size);
145 inode_add_bytes(inode, size);
146 ret = btrfs_insert_empty_item(trans, root, path, &key,
153 leaf = path->nodes[0];
154 ei = btrfs_item_ptr(leaf, path->slots[0],
155 struct btrfs_file_extent_item);
156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
157 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
158 btrfs_set_file_extent_encryption(leaf, ei, 0);
159 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
160 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
161 ptr = btrfs_file_extent_inline_start(ei);
163 if (compress_type != BTRFS_COMPRESS_NONE) {
166 while (compressed_size > 0) {
167 cpage = compressed_pages[i];
168 cur_size = min_t(unsigned long, compressed_size,
171 kaddr = kmap_atomic(cpage, KM_USER0);
172 write_extent_buffer(leaf, kaddr, ptr, cur_size);
173 kunmap_atomic(kaddr, KM_USER0);
177 compressed_size -= cur_size;
179 btrfs_set_file_extent_compression(leaf, ei,
182 page = find_get_page(inode->i_mapping,
183 start >> PAGE_CACHE_SHIFT);
184 btrfs_set_file_extent_compression(leaf, ei, 0);
185 kaddr = kmap_atomic(page, KM_USER0);
186 offset = start & (PAGE_CACHE_SIZE - 1);
187 write_extent_buffer(leaf, kaddr + offset, ptr, size);
188 kunmap_atomic(kaddr, KM_USER0);
189 page_cache_release(page);
191 btrfs_mark_buffer_dirty(leaf);
192 btrfs_free_path(path);
195 * we're an inline extent, so nobody can
196 * extend the file past i_size without locking
197 * a page we already have locked.
199 * We must do any isize and inode updates
200 * before we unlock the pages. Otherwise we
201 * could end up racing with unlink.
203 BTRFS_I(inode)->disk_i_size = inode->i_size;
204 btrfs_update_inode(trans, root, inode);
208 btrfs_free_path(path);
214 * conditionally insert an inline extent into the file. This
215 * does the checks required to make sure the data is small enough
216 * to fit as an inline extent.
218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
219 struct btrfs_root *root,
220 struct inode *inode, u64 start, u64 end,
221 size_t compressed_size, int compress_type,
222 struct page **compressed_pages)
224 u64 isize = i_size_read(inode);
225 u64 actual_end = min(end + 1, isize);
226 u64 inline_len = actual_end - start;
227 u64 aligned_end = (end + root->sectorsize - 1) &
228 ~((u64)root->sectorsize - 1);
230 u64 data_len = inline_len;
234 data_len = compressed_size;
237 actual_end >= PAGE_CACHE_SIZE ||
238 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
240 (actual_end & (root->sectorsize - 1)) == 0) ||
242 data_len > root->fs_info->max_inline) {
246 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
250 if (isize > actual_end)
251 inline_len = min_t(u64, isize, actual_end);
252 ret = insert_inline_extent(trans, root, inode, start,
253 inline_len, compressed_size,
254 compress_type, compressed_pages);
256 btrfs_delalloc_release_metadata(inode, end + 1 - start);
257 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
261 struct async_extent {
266 unsigned long nr_pages;
268 struct list_head list;
273 struct btrfs_root *root;
274 struct page *locked_page;
277 struct list_head extents;
278 struct btrfs_work work;
281 static noinline int add_async_extent(struct async_cow *cow,
282 u64 start, u64 ram_size,
285 unsigned long nr_pages,
288 struct async_extent *async_extent;
290 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
291 BUG_ON(!async_extent);
292 async_extent->start = start;
293 async_extent->ram_size = ram_size;
294 async_extent->compressed_size = compressed_size;
295 async_extent->pages = pages;
296 async_extent->nr_pages = nr_pages;
297 async_extent->compress_type = compress_type;
298 list_add_tail(&async_extent->list, &cow->extents);
303 * we create compressed extents in two phases. The first
304 * phase compresses a range of pages that have already been
305 * locked (both pages and state bits are locked).
307 * This is done inside an ordered work queue, and the compression
308 * is spread across many cpus. The actual IO submission is step
309 * two, and the ordered work queue takes care of making sure that
310 * happens in the same order things were put onto the queue by
311 * writepages and friends.
313 * If this code finds it can't get good compression, it puts an
314 * entry onto the work queue to write the uncompressed bytes. This
315 * makes sure that both compressed inodes and uncompressed inodes
316 * are written in the same order that pdflush sent them down.
318 static noinline int compress_file_range(struct inode *inode,
319 struct page *locked_page,
321 struct async_cow *async_cow,
324 struct btrfs_root *root = BTRFS_I(inode)->root;
325 struct btrfs_trans_handle *trans;
327 u64 blocksize = root->sectorsize;
329 u64 isize = i_size_read(inode);
331 struct page **pages = NULL;
332 unsigned long nr_pages;
333 unsigned long nr_pages_ret = 0;
334 unsigned long total_compressed = 0;
335 unsigned long total_in = 0;
336 unsigned long max_compressed = 128 * 1024;
337 unsigned long max_uncompressed = 128 * 1024;
340 int compress_type = root->fs_info->compress_type;
342 actual_end = min_t(u64, isize, end + 1);
345 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
346 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
349 * we don't want to send crud past the end of i_size through
350 * compression, that's just a waste of CPU time. So, if the
351 * end of the file is before the start of our current
352 * requested range of bytes, we bail out to the uncompressed
353 * cleanup code that can deal with all of this.
355 * It isn't really the fastest way to fix things, but this is a
356 * very uncommon corner.
358 if (actual_end <= start)
359 goto cleanup_and_bail_uncompressed;
361 total_compressed = actual_end - start;
363 /* we want to make sure that amount of ram required to uncompress
364 * an extent is reasonable, so we limit the total size in ram
365 * of a compressed extent to 128k. This is a crucial number
366 * because it also controls how easily we can spread reads across
367 * cpus for decompression.
369 * We also want to make sure the amount of IO required to do
370 * a random read is reasonably small, so we limit the size of
371 * a compressed extent to 128k.
373 total_compressed = min(total_compressed, max_uncompressed);
374 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
375 num_bytes = max(blocksize, num_bytes);
380 * we do compression for mount -o compress and when the
381 * inode has not been flagged as nocompress. This flag can
382 * change at any time if we discover bad compression ratios.
384 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
385 (btrfs_test_opt(root, COMPRESS) ||
386 (BTRFS_I(inode)->force_compress) ||
387 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
389 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
392 if (BTRFS_I(inode)->force_compress)
393 compress_type = BTRFS_I(inode)->force_compress;
395 ret = btrfs_compress_pages(compress_type,
396 inode->i_mapping, start,
397 total_compressed, pages,
398 nr_pages, &nr_pages_ret,
404 unsigned long offset = total_compressed &
405 (PAGE_CACHE_SIZE - 1);
406 struct page *page = pages[nr_pages_ret - 1];
409 /* zero the tail end of the last page, we might be
410 * sending it down to disk
413 kaddr = kmap_atomic(page, KM_USER0);
414 memset(kaddr + offset, 0,
415 PAGE_CACHE_SIZE - offset);
416 kunmap_atomic(kaddr, KM_USER0);
422 trans = btrfs_join_transaction(root, 1);
423 BUG_ON(IS_ERR(trans));
424 btrfs_set_trans_block_group(trans, inode);
425 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427 /* lets try to make an inline extent */
428 if (ret || total_in < (actual_end - start)) {
429 /* we didn't compress the entire range, try
430 * to make an uncompressed inline extent.
432 ret = cow_file_range_inline(trans, root, inode,
433 start, end, 0, 0, NULL);
435 /* try making a compressed inline extent */
436 ret = cow_file_range_inline(trans, root, inode,
439 compress_type, pages);
443 * inline extent creation worked, we don't need
444 * to create any more async work items. Unlock
445 * and free up our temp pages.
447 extent_clear_unlock_delalloc(inode,
448 &BTRFS_I(inode)->io_tree,
450 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
451 EXTENT_CLEAR_DELALLOC |
452 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
454 btrfs_end_transaction(trans, root);
457 btrfs_end_transaction(trans, root);
462 * we aren't doing an inline extent round the compressed size
463 * up to a block size boundary so the allocator does sane
466 total_compressed = (total_compressed + blocksize - 1) &
470 * one last check to make sure the compression is really a
471 * win, compare the page count read with the blocks on disk
473 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
474 ~(PAGE_CACHE_SIZE - 1);
475 if (total_compressed >= total_in) {
478 num_bytes = total_in;
481 if (!will_compress && pages) {
483 * the compression code ran but failed to make things smaller,
484 * free any pages it allocated and our page pointer array
486 for (i = 0; i < nr_pages_ret; i++) {
487 WARN_ON(pages[i]->mapping);
488 page_cache_release(pages[i]);
492 total_compressed = 0;
495 /* flag the file so we don't compress in the future */
496 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
497 !(BTRFS_I(inode)->force_compress)) {
498 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
504 /* the async work queues will take care of doing actual
505 * allocation on disk for these compressed pages,
506 * and will submit them to the elevator.
508 add_async_extent(async_cow, start, num_bytes,
509 total_compressed, pages, nr_pages_ret,
512 if (start + num_bytes < end) {
519 cleanup_and_bail_uncompressed:
521 * No compression, but we still need to write the pages in
522 * the file we've been given so far. redirty the locked
523 * page if it corresponds to our extent and set things up
524 * for the async work queue to run cow_file_range to do
525 * the normal delalloc dance
527 if (page_offset(locked_page) >= start &&
528 page_offset(locked_page) <= end) {
529 __set_page_dirty_nobuffers(locked_page);
530 /* unlocked later on in the async handlers */
532 add_async_extent(async_cow, start, end - start + 1,
533 0, NULL, 0, BTRFS_COMPRESS_NONE);
541 for (i = 0; i < nr_pages_ret; i++) {
542 WARN_ON(pages[i]->mapping);
543 page_cache_release(pages[i]);
551 * phase two of compressed writeback. This is the ordered portion
552 * of the code, which only gets called in the order the work was
553 * queued. We walk all the async extents created by compress_file_range
554 * and send them down to the disk.
556 static noinline int submit_compressed_extents(struct inode *inode,
557 struct async_cow *async_cow)
559 struct async_extent *async_extent;
561 struct btrfs_trans_handle *trans;
562 struct btrfs_key ins;
563 struct extent_map *em;
564 struct btrfs_root *root = BTRFS_I(inode)->root;
565 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
566 struct extent_io_tree *io_tree;
569 if (list_empty(&async_cow->extents))
573 while (!list_empty(&async_cow->extents)) {
574 async_extent = list_entry(async_cow->extents.next,
575 struct async_extent, list);
576 list_del(&async_extent->list);
578 io_tree = &BTRFS_I(inode)->io_tree;
581 /* did the compression code fall back to uncompressed IO? */
582 if (!async_extent->pages) {
583 int page_started = 0;
584 unsigned long nr_written = 0;
586 lock_extent(io_tree, async_extent->start,
587 async_extent->start +
588 async_extent->ram_size - 1, GFP_NOFS);
590 /* allocate blocks */
591 ret = cow_file_range(inode, async_cow->locked_page,
593 async_extent->start +
594 async_extent->ram_size - 1,
595 &page_started, &nr_written, 0);
598 * if page_started, cow_file_range inserted an
599 * inline extent and took care of all the unlocking
600 * and IO for us. Otherwise, we need to submit
601 * all those pages down to the drive.
603 if (!page_started && !ret)
604 extent_write_locked_range(io_tree,
605 inode, async_extent->start,
606 async_extent->start +
607 async_extent->ram_size - 1,
615 lock_extent(io_tree, async_extent->start,
616 async_extent->start + async_extent->ram_size - 1,
619 trans = btrfs_join_transaction(root, 1);
620 BUG_ON(IS_ERR(trans));
621 ret = btrfs_reserve_extent(trans, root,
622 async_extent->compressed_size,
623 async_extent->compressed_size,
626 btrfs_end_transaction(trans, root);
630 for (i = 0; i < async_extent->nr_pages; i++) {
631 WARN_ON(async_extent->pages[i]->mapping);
632 page_cache_release(async_extent->pages[i]);
634 kfree(async_extent->pages);
635 async_extent->nr_pages = 0;
636 async_extent->pages = NULL;
637 unlock_extent(io_tree, async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1, GFP_NOFS);
644 * here we're doing allocation and writeback of the
647 btrfs_drop_extent_cache(inode, async_extent->start,
648 async_extent->start +
649 async_extent->ram_size - 1, 0);
651 em = alloc_extent_map(GFP_NOFS);
653 em->start = async_extent->start;
654 em->len = async_extent->ram_size;
655 em->orig_start = em->start;
657 em->block_start = ins.objectid;
658 em->block_len = ins.offset;
659 em->bdev = root->fs_info->fs_devices->latest_bdev;
660 em->compress_type = async_extent->compress_type;
661 set_bit(EXTENT_FLAG_PINNED, &em->flags);
662 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
665 write_lock(&em_tree->lock);
666 ret = add_extent_mapping(em_tree, em);
667 write_unlock(&em_tree->lock);
668 if (ret != -EEXIST) {
672 btrfs_drop_extent_cache(inode, async_extent->start,
673 async_extent->start +
674 async_extent->ram_size - 1, 0);
677 ret = btrfs_add_ordered_extent_compress(inode,
680 async_extent->ram_size,
682 BTRFS_ORDERED_COMPRESSED,
683 async_extent->compress_type);
687 * clear dirty, set writeback and unlock the pages.
689 extent_clear_unlock_delalloc(inode,
690 &BTRFS_I(inode)->io_tree,
692 async_extent->start +
693 async_extent->ram_size - 1,
694 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
695 EXTENT_CLEAR_UNLOCK |
696 EXTENT_CLEAR_DELALLOC |
697 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
699 ret = btrfs_submit_compressed_write(inode,
701 async_extent->ram_size,
703 ins.offset, async_extent->pages,
704 async_extent->nr_pages);
707 alloc_hint = ins.objectid + ins.offset;
715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
718 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
719 struct extent_map *em;
722 read_lock(&em_tree->lock);
723 em = search_extent_mapping(em_tree, start, num_bytes);
726 * if block start isn't an actual block number then find the
727 * first block in this inode and use that as a hint. If that
728 * block is also bogus then just don't worry about it.
730 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
732 em = search_extent_mapping(em_tree, 0, 0);
733 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
734 alloc_hint = em->block_start;
738 alloc_hint = em->block_start;
742 read_unlock(&em_tree->lock);
748 * when extent_io.c finds a delayed allocation range in the file,
749 * the call backs end up in this code. The basic idea is to
750 * allocate extents on disk for the range, and create ordered data structs
751 * in ram to track those extents.
753 * locked_page is the page that writepage had locked already. We use
754 * it to make sure we don't do extra locks or unlocks.
756 * *page_started is set to one if we unlock locked_page and do everything
757 * required to start IO on it. It may be clean and already done with
760 static noinline int cow_file_range(struct inode *inode,
761 struct page *locked_page,
762 u64 start, u64 end, int *page_started,
763 unsigned long *nr_written,
766 struct btrfs_root *root = BTRFS_I(inode)->root;
767 struct btrfs_trans_handle *trans;
770 unsigned long ram_size;
773 u64 blocksize = root->sectorsize;
774 struct btrfs_key ins;
775 struct extent_map *em;
776 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
779 BUG_ON(root == root->fs_info->tree_root);
780 trans = btrfs_join_transaction(root, 1);
781 BUG_ON(IS_ERR(trans));
782 btrfs_set_trans_block_group(trans, inode);
783 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
785 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
786 num_bytes = max(blocksize, num_bytes);
787 disk_num_bytes = num_bytes;
791 /* lets try to make an inline extent */
792 ret = cow_file_range_inline(trans, root, inode,
793 start, end, 0, 0, NULL);
795 extent_clear_unlock_delalloc(inode,
796 &BTRFS_I(inode)->io_tree,
798 EXTENT_CLEAR_UNLOCK_PAGE |
799 EXTENT_CLEAR_UNLOCK |
800 EXTENT_CLEAR_DELALLOC |
802 EXTENT_SET_WRITEBACK |
803 EXTENT_END_WRITEBACK);
805 *nr_written = *nr_written +
806 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
813 BUG_ON(disk_num_bytes >
814 btrfs_super_total_bytes(&root->fs_info->super_copy));
816 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
817 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
819 while (disk_num_bytes > 0) {
822 cur_alloc_size = disk_num_bytes;
823 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
824 root->sectorsize, 0, alloc_hint,
828 em = alloc_extent_map(GFP_NOFS);
831 em->orig_start = em->start;
832 ram_size = ins.offset;
833 em->len = ins.offset;
835 em->block_start = ins.objectid;
836 em->block_len = ins.offset;
837 em->bdev = root->fs_info->fs_devices->latest_bdev;
838 set_bit(EXTENT_FLAG_PINNED, &em->flags);
841 write_lock(&em_tree->lock);
842 ret = add_extent_mapping(em_tree, em);
843 write_unlock(&em_tree->lock);
844 if (ret != -EEXIST) {
848 btrfs_drop_extent_cache(inode, start,
849 start + ram_size - 1, 0);
852 cur_alloc_size = ins.offset;
853 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
854 ram_size, cur_alloc_size, 0);
857 if (root->root_key.objectid ==
858 BTRFS_DATA_RELOC_TREE_OBJECTID) {
859 ret = btrfs_reloc_clone_csums(inode, start,
864 if (disk_num_bytes < cur_alloc_size)
867 /* we're not doing compressed IO, don't unlock the first
868 * page (which the caller expects to stay locked), don't
869 * clear any dirty bits and don't set any writeback bits
871 * Do set the Private2 bit so we know this page was properly
872 * setup for writepage
874 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
875 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
878 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
879 start, start + ram_size - 1,
881 disk_num_bytes -= cur_alloc_size;
882 num_bytes -= cur_alloc_size;
883 alloc_hint = ins.objectid + ins.offset;
884 start += cur_alloc_size;
888 btrfs_end_transaction(trans, root);
894 * work queue call back to started compression on a file and pages
896 static noinline void async_cow_start(struct btrfs_work *work)
898 struct async_cow *async_cow;
900 async_cow = container_of(work, struct async_cow, work);
902 compress_file_range(async_cow->inode, async_cow->locked_page,
903 async_cow->start, async_cow->end, async_cow,
906 async_cow->inode = NULL;
910 * work queue call back to submit previously compressed pages
912 static noinline void async_cow_submit(struct btrfs_work *work)
914 struct async_cow *async_cow;
915 struct btrfs_root *root;
916 unsigned long nr_pages;
918 async_cow = container_of(work, struct async_cow, work);
920 root = async_cow->root;
921 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
924 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
926 if (atomic_read(&root->fs_info->async_delalloc_pages) <
928 waitqueue_active(&root->fs_info->async_submit_wait))
929 wake_up(&root->fs_info->async_submit_wait);
931 if (async_cow->inode)
932 submit_compressed_extents(async_cow->inode, async_cow);
935 static noinline void async_cow_free(struct btrfs_work *work)
937 struct async_cow *async_cow;
938 async_cow = container_of(work, struct async_cow, work);
942 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
943 u64 start, u64 end, int *page_started,
944 unsigned long *nr_written)
946 struct async_cow *async_cow;
947 struct btrfs_root *root = BTRFS_I(inode)->root;
948 unsigned long nr_pages;
950 int limit = 10 * 1024 * 1042;
952 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
953 1, 0, NULL, GFP_NOFS);
954 while (start < end) {
955 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
956 async_cow->inode = inode;
957 async_cow->root = root;
958 async_cow->locked_page = locked_page;
959 async_cow->start = start;
961 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
964 cur_end = min(end, start + 512 * 1024 - 1);
966 async_cow->end = cur_end;
967 INIT_LIST_HEAD(&async_cow->extents);
969 async_cow->work.func = async_cow_start;
970 async_cow->work.ordered_func = async_cow_submit;
971 async_cow->work.ordered_free = async_cow_free;
972 async_cow->work.flags = 0;
974 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
976 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
978 btrfs_queue_worker(&root->fs_info->delalloc_workers,
981 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
982 wait_event(root->fs_info->async_submit_wait,
983 (atomic_read(&root->fs_info->async_delalloc_pages) <
987 while (atomic_read(&root->fs_info->async_submit_draining) &&
988 atomic_read(&root->fs_info->async_delalloc_pages)) {
989 wait_event(root->fs_info->async_submit_wait,
990 (atomic_read(&root->fs_info->async_delalloc_pages) ==
994 *nr_written += nr_pages;
1001 static noinline int csum_exist_in_range(struct btrfs_root *root,
1002 u64 bytenr, u64 num_bytes)
1005 struct btrfs_ordered_sum *sums;
1008 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1009 bytenr + num_bytes - 1, &list);
1010 if (ret == 0 && list_empty(&list))
1013 while (!list_empty(&list)) {
1014 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1015 list_del(&sums->list);
1022 * when nowcow writeback call back. This checks for snapshots or COW copies
1023 * of the extents that exist in the file, and COWs the file as required.
1025 * If no cow copies or snapshots exist, we write directly to the existing
1028 static noinline int run_delalloc_nocow(struct inode *inode,
1029 struct page *locked_page,
1030 u64 start, u64 end, int *page_started, int force,
1031 unsigned long *nr_written)
1033 struct btrfs_root *root = BTRFS_I(inode)->root;
1034 struct btrfs_trans_handle *trans;
1035 struct extent_buffer *leaf;
1036 struct btrfs_path *path;
1037 struct btrfs_file_extent_item *fi;
1038 struct btrfs_key found_key;
1050 bool nolock = false;
1052 path = btrfs_alloc_path();
1054 if (root == root->fs_info->tree_root) {
1056 trans = btrfs_join_transaction_nolock(root, 1);
1058 trans = btrfs_join_transaction(root, 1);
1060 BUG_ON(IS_ERR(trans));
1062 cow_start = (u64)-1;
1065 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1068 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1069 leaf = path->nodes[0];
1070 btrfs_item_key_to_cpu(leaf, &found_key,
1071 path->slots[0] - 1);
1072 if (found_key.objectid == inode->i_ino &&
1073 found_key.type == BTRFS_EXTENT_DATA_KEY)
1078 leaf = path->nodes[0];
1079 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1080 ret = btrfs_next_leaf(root, path);
1085 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093 if (found_key.objectid > inode->i_ino ||
1094 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1095 found_key.offset > end)
1098 if (found_key.offset > cur_offset) {
1099 extent_end = found_key.offset;
1104 fi = btrfs_item_ptr(leaf, path->slots[0],
1105 struct btrfs_file_extent_item);
1106 extent_type = btrfs_file_extent_type(leaf, fi);
1108 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1109 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1110 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1111 extent_offset = btrfs_file_extent_offset(leaf, fi);
1112 extent_end = found_key.offset +
1113 btrfs_file_extent_num_bytes(leaf, fi);
1114 if (extent_end <= start) {
1118 if (disk_bytenr == 0)
1120 if (btrfs_file_extent_compression(leaf, fi) ||
1121 btrfs_file_extent_encryption(leaf, fi) ||
1122 btrfs_file_extent_other_encoding(leaf, fi))
1124 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126 if (btrfs_extent_readonly(root, disk_bytenr))
1128 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130 extent_offset, disk_bytenr))
1132 disk_bytenr += extent_offset;
1133 disk_bytenr += cur_offset - found_key.offset;
1134 num_bytes = min(end + 1, extent_end) - cur_offset;
1136 * force cow if csum exists in the range.
1137 * this ensure that csum for a given extent are
1138 * either valid or do not exist.
1140 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1143 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1144 extent_end = found_key.offset +
1145 btrfs_file_extent_inline_len(leaf, fi);
1146 extent_end = ALIGN(extent_end, root->sectorsize);
1151 if (extent_end <= start) {
1156 if (cow_start == (u64)-1)
1157 cow_start = cur_offset;
1158 cur_offset = extent_end;
1159 if (cur_offset > end)
1165 btrfs_release_path(root, path);
1166 if (cow_start != (u64)-1) {
1167 ret = cow_file_range(inode, locked_page, cow_start,
1168 found_key.offset - 1, page_started,
1171 cow_start = (u64)-1;
1174 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1175 struct extent_map *em;
1176 struct extent_map_tree *em_tree;
1177 em_tree = &BTRFS_I(inode)->extent_tree;
1178 em = alloc_extent_map(GFP_NOFS);
1180 em->start = cur_offset;
1181 em->orig_start = em->start;
1182 em->len = num_bytes;
1183 em->block_len = num_bytes;
1184 em->block_start = disk_bytenr;
1185 em->bdev = root->fs_info->fs_devices->latest_bdev;
1186 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188 write_lock(&em_tree->lock);
1189 ret = add_extent_mapping(em_tree, em);
1190 write_unlock(&em_tree->lock);
1191 if (ret != -EEXIST) {
1192 free_extent_map(em);
1195 btrfs_drop_extent_cache(inode, em->start,
1196 em->start + em->len - 1, 0);
1198 type = BTRFS_ORDERED_PREALLOC;
1200 type = BTRFS_ORDERED_NOCOW;
1203 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1204 num_bytes, num_bytes, type);
1207 if (root->root_key.objectid ==
1208 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1209 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1214 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1215 cur_offset, cur_offset + num_bytes - 1,
1216 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1217 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1218 EXTENT_SET_PRIVATE2);
1219 cur_offset = extent_end;
1220 if (cur_offset > end)
1223 btrfs_release_path(root, path);
1225 if (cur_offset <= end && cow_start == (u64)-1)
1226 cow_start = cur_offset;
1227 if (cow_start != (u64)-1) {
1228 ret = cow_file_range(inode, locked_page, cow_start, end,
1229 page_started, nr_written, 1);
1234 ret = btrfs_end_transaction_nolock(trans, root);
1237 ret = btrfs_end_transaction(trans, root);
1240 btrfs_free_path(path);
1245 * extent_io.c call back to do delayed allocation processing
1247 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1248 u64 start, u64 end, int *page_started,
1249 unsigned long *nr_written)
1252 struct btrfs_root *root = BTRFS_I(inode)->root;
1254 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1255 ret = run_delalloc_nocow(inode, locked_page, start, end,
1256 page_started, 1, nr_written);
1257 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1258 ret = run_delalloc_nocow(inode, locked_page, start, end,
1259 page_started, 0, nr_written);
1260 else if (!btrfs_test_opt(root, COMPRESS) &&
1261 !(BTRFS_I(inode)->force_compress) &&
1262 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1263 ret = cow_file_range(inode, locked_page, start, end,
1264 page_started, nr_written, 1);
1266 ret = cow_file_range_async(inode, locked_page, start, end,
1267 page_started, nr_written);
1271 static int btrfs_split_extent_hook(struct inode *inode,
1272 struct extent_state *orig, u64 split)
1274 /* not delalloc, ignore it */
1275 if (!(orig->state & EXTENT_DELALLOC))
1278 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1283 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1284 * extents so we can keep track of new extents that are just merged onto old
1285 * extents, such as when we are doing sequential writes, so we can properly
1286 * account for the metadata space we'll need.
1288 static int btrfs_merge_extent_hook(struct inode *inode,
1289 struct extent_state *new,
1290 struct extent_state *other)
1292 /* not delalloc, ignore it */
1293 if (!(other->state & EXTENT_DELALLOC))
1296 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1301 * extent_io.c set_bit_hook, used to track delayed allocation
1302 * bytes in this file, and to maintain the list of inodes that
1303 * have pending delalloc work to be done.
1305 static int btrfs_set_bit_hook(struct inode *inode,
1306 struct extent_state *state, int *bits)
1310 * set_bit and clear bit hooks normally require _irqsave/restore
1311 * but in this case, we are only testeing for the DELALLOC
1312 * bit, which is only set or cleared with irqs on
1314 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1315 struct btrfs_root *root = BTRFS_I(inode)->root;
1316 u64 len = state->end + 1 - state->start;
1317 int do_list = (root->root_key.objectid !=
1318 BTRFS_ROOT_TREE_OBJECTID);
1320 if (*bits & EXTENT_FIRST_DELALLOC)
1321 *bits &= ~EXTENT_FIRST_DELALLOC;
1323 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325 spin_lock(&root->fs_info->delalloc_lock);
1326 BTRFS_I(inode)->delalloc_bytes += len;
1327 root->fs_info->delalloc_bytes += len;
1328 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1329 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1330 &root->fs_info->delalloc_inodes);
1332 spin_unlock(&root->fs_info->delalloc_lock);
1338 * extent_io.c clear_bit_hook, see set_bit_hook for why
1340 static int btrfs_clear_bit_hook(struct inode *inode,
1341 struct extent_state *state, int *bits)
1344 * set_bit and clear bit hooks normally require _irqsave/restore
1345 * but in this case, we are only testeing for the DELALLOC
1346 * bit, which is only set or cleared with irqs on
1348 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1349 struct btrfs_root *root = BTRFS_I(inode)->root;
1350 u64 len = state->end + 1 - state->start;
1351 int do_list = (root->root_key.objectid !=
1352 BTRFS_ROOT_TREE_OBJECTID);
1354 if (*bits & EXTENT_FIRST_DELALLOC)
1355 *bits &= ~EXTENT_FIRST_DELALLOC;
1356 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1357 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359 if (*bits & EXTENT_DO_ACCOUNTING)
1360 btrfs_delalloc_release_metadata(inode, len);
1362 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364 btrfs_free_reserved_data_space(inode, len);
1366 spin_lock(&root->fs_info->delalloc_lock);
1367 root->fs_info->delalloc_bytes -= len;
1368 BTRFS_I(inode)->delalloc_bytes -= len;
1370 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1371 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1372 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374 spin_unlock(&root->fs_info->delalloc_lock);
1380 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1381 * we don't create bios that span stripes or chunks
1383 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1384 size_t size, struct bio *bio,
1385 unsigned long bio_flags)
1387 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1388 struct btrfs_mapping_tree *map_tree;
1389 u64 logical = (u64)bio->bi_sector << 9;
1394 if (bio_flags & EXTENT_BIO_COMPRESSED)
1397 length = bio->bi_size;
1398 map_tree = &root->fs_info->mapping_tree;
1399 map_length = length;
1400 ret = btrfs_map_block(map_tree, READ, logical,
1401 &map_length, NULL, 0);
1403 if (map_length < length + size)
1409 * in order to insert checksums into the metadata in large chunks,
1410 * we wait until bio submission time. All the pages in the bio are
1411 * checksummed and sums are attached onto the ordered extent record.
1413 * At IO completion time the cums attached on the ordered extent record
1414 * are inserted into the btree
1416 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1417 struct bio *bio, int mirror_num,
1418 unsigned long bio_flags,
1421 struct btrfs_root *root = BTRFS_I(inode)->root;
1424 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1430 * in order to insert checksums into the metadata in large chunks,
1431 * we wait until bio submission time. All the pages in the bio are
1432 * checksummed and sums are attached onto the ordered extent record.
1434 * At IO completion time the cums attached on the ordered extent record
1435 * are inserted into the btree
1437 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1438 int mirror_num, unsigned long bio_flags,
1441 struct btrfs_root *root = BTRFS_I(inode)->root;
1442 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1446 * extent_io.c submission hook. This does the right thing for csum calculation
1447 * on write, or reading the csums from the tree before a read
1449 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1450 int mirror_num, unsigned long bio_flags,
1453 struct btrfs_root *root = BTRFS_I(inode)->root;
1457 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459 if (root == root->fs_info->tree_root)
1460 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1465 if (!(rw & REQ_WRITE)) {
1466 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1467 return btrfs_submit_compressed_read(inode, bio,
1468 mirror_num, bio_flags);
1469 } else if (!skip_sum) {
1470 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1475 } else if (!skip_sum) {
1476 /* csum items have already been cloned */
1477 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479 /* we're doing a write, do the async checksumming */
1480 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1481 inode, rw, bio, mirror_num,
1482 bio_flags, bio_offset,
1483 __btrfs_submit_bio_start,
1484 __btrfs_submit_bio_done);
1488 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1492 * given a list of ordered sums record them in the inode. This happens
1493 * at IO completion time based on sums calculated at bio submission time.
1495 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1496 struct inode *inode, u64 file_offset,
1497 struct list_head *list)
1499 struct btrfs_ordered_sum *sum;
1501 btrfs_set_trans_block_group(trans, inode);
1503 list_for_each_entry(sum, list, list) {
1504 btrfs_csum_file_blocks(trans,
1505 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1510 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1511 struct extent_state **cached_state)
1513 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1516 cached_state, GFP_NOFS);
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup {
1522 struct btrfs_work work;
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 struct btrfs_writepage_fixup *fixup;
1528 struct btrfs_ordered_extent *ordered;
1529 struct extent_state *cached_state = NULL;
1531 struct inode *inode;
1535 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1539 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1540 ClearPageChecked(page);
1544 inode = page->mapping->host;
1545 page_start = page_offset(page);
1546 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1549 &cached_state, GFP_NOFS);
1551 /* already ordered? We're done */
1552 if (PagePrivate2(page))
1555 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1558 page_end, &cached_state, GFP_NOFS);
1560 btrfs_start_ordered_extent(inode, ordered, 1);
1565 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1566 ClearPageChecked(page);
1568 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1569 &cached_state, GFP_NOFS);
1572 page_cache_release(page);
1577 * There are a few paths in the higher layers of the kernel that directly
1578 * set the page dirty bit without asking the filesystem if it is a
1579 * good idea. This causes problems because we want to make sure COW
1580 * properly happens and the data=ordered rules are followed.
1582 * In our case any range that doesn't have the ORDERED bit set
1583 * hasn't been properly setup for IO. We kick off an async process
1584 * to fix it up. The async helper will wait for ordered extents, set
1585 * the delalloc bit and make it safe to write the page.
1587 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 struct inode *inode = page->mapping->host;
1590 struct btrfs_writepage_fixup *fixup;
1591 struct btrfs_root *root = BTRFS_I(inode)->root;
1593 /* this page is properly in the ordered list */
1594 if (TestClearPagePrivate2(page))
1597 if (PageChecked(page))
1600 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1604 SetPageChecked(page);
1605 page_cache_get(page);
1606 fixup->work.func = btrfs_writepage_fixup_worker;
1608 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1612 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1613 struct inode *inode, u64 file_pos,
1614 u64 disk_bytenr, u64 disk_num_bytes,
1615 u64 num_bytes, u64 ram_bytes,
1616 u8 compression, u8 encryption,
1617 u16 other_encoding, int extent_type)
1619 struct btrfs_root *root = BTRFS_I(inode)->root;
1620 struct btrfs_file_extent_item *fi;
1621 struct btrfs_path *path;
1622 struct extent_buffer *leaf;
1623 struct btrfs_key ins;
1627 path = btrfs_alloc_path();
1630 path->leave_spinning = 1;
1633 * we may be replacing one extent in the tree with another.
1634 * The new extent is pinned in the extent map, and we don't want
1635 * to drop it from the cache until it is completely in the btree.
1637 * So, tell btrfs_drop_extents to leave this extent in the cache.
1638 * the caller is expected to unpin it and allow it to be merged
1641 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1645 ins.objectid = inode->i_ino;
1646 ins.offset = file_pos;
1647 ins.type = BTRFS_EXTENT_DATA_KEY;
1648 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650 leaf = path->nodes[0];
1651 fi = btrfs_item_ptr(leaf, path->slots[0],
1652 struct btrfs_file_extent_item);
1653 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1654 btrfs_set_file_extent_type(leaf, fi, extent_type);
1655 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1656 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1657 btrfs_set_file_extent_offset(leaf, fi, 0);
1658 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1659 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1660 btrfs_set_file_extent_compression(leaf, fi, compression);
1661 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1662 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664 btrfs_unlock_up_safe(path, 1);
1665 btrfs_set_lock_blocking(leaf);
1667 btrfs_mark_buffer_dirty(leaf);
1669 inode_add_bytes(inode, num_bytes);
1671 ins.objectid = disk_bytenr;
1672 ins.offset = disk_num_bytes;
1673 ins.type = BTRFS_EXTENT_ITEM_KEY;
1674 ret = btrfs_alloc_reserved_file_extent(trans, root,
1675 root->root_key.objectid,
1676 inode->i_ino, file_pos, &ins);
1678 btrfs_free_path(path);
1684 * helper function for btrfs_finish_ordered_io, this
1685 * just reads in some of the csum leaves to prime them into ram
1686 * before we start the transaction. It limits the amount of btree
1687 * reads required while inside the transaction.
1689 /* as ordered data IO finishes, this gets called so we can finish
1690 * an ordered extent if the range of bytes in the file it covers are
1693 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 struct btrfs_root *root = BTRFS_I(inode)->root;
1696 struct btrfs_trans_handle *trans = NULL;
1697 struct btrfs_ordered_extent *ordered_extent = NULL;
1698 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1699 struct extent_state *cached_state = NULL;
1700 int compress_type = 0;
1702 bool nolock = false;
1704 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1708 BUG_ON(!ordered_extent);
1710 nolock = (root == root->fs_info->tree_root);
1712 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1713 BUG_ON(!list_empty(&ordered_extent->list));
1714 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717 trans = btrfs_join_transaction_nolock(root, 1);
1719 trans = btrfs_join_transaction(root, 1);
1720 BUG_ON(IS_ERR(trans));
1721 btrfs_set_trans_block_group(trans, inode);
1722 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1723 ret = btrfs_update_inode(trans, root, inode);
1729 lock_extent_bits(io_tree, ordered_extent->file_offset,
1730 ordered_extent->file_offset + ordered_extent->len - 1,
1731 0, &cached_state, GFP_NOFS);
1734 trans = btrfs_join_transaction_nolock(root, 1);
1736 trans = btrfs_join_transaction(root, 1);
1737 BUG_ON(IS_ERR(trans));
1738 btrfs_set_trans_block_group(trans, inode);
1739 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1742 compress_type = ordered_extent->compress_type;
1743 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1744 BUG_ON(compress_type);
1745 ret = btrfs_mark_extent_written(trans, inode,
1746 ordered_extent->file_offset,
1747 ordered_extent->file_offset +
1748 ordered_extent->len);
1751 BUG_ON(root == root->fs_info->tree_root);
1752 ret = insert_reserved_file_extent(trans, inode,
1753 ordered_extent->file_offset,
1754 ordered_extent->start,
1755 ordered_extent->disk_len,
1756 ordered_extent->len,
1757 ordered_extent->len,
1758 compress_type, 0, 0,
1759 BTRFS_FILE_EXTENT_REG);
1760 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1761 ordered_extent->file_offset,
1762 ordered_extent->len);
1765 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1766 ordered_extent->file_offset +
1767 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769 add_pending_csums(trans, inode, ordered_extent->file_offset,
1770 &ordered_extent->list);
1772 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1773 ret = btrfs_update_inode(trans, root, inode);
1778 btrfs_end_transaction_nolock(trans, root);
1780 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1782 btrfs_end_transaction(trans, root);
1786 btrfs_put_ordered_extent(ordered_extent);
1787 /* once for the tree */
1788 btrfs_put_ordered_extent(ordered_extent);
1793 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1794 struct extent_state *state, int uptodate)
1796 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1798 ClearPagePrivate2(page);
1799 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1803 * When IO fails, either with EIO or csum verification fails, we
1804 * try other mirrors that might have a good copy of the data. This
1805 * io_failure_record is used to record state as we go through all the
1806 * mirrors. If another mirror has good data, the page is set up to date
1807 * and things continue. If a good mirror can't be found, the original
1808 * bio end_io callback is called to indicate things have failed.
1810 struct io_failure_record {
1815 unsigned long bio_flags;
1819 static int btrfs_io_failed_hook(struct bio *failed_bio,
1820 struct page *page, u64 start, u64 end,
1821 struct extent_state *state)
1823 struct io_failure_record *failrec = NULL;
1825 struct extent_map *em;
1826 struct inode *inode = page->mapping->host;
1827 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1828 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1835 ret = get_state_private(failure_tree, start, &private);
1837 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1840 failrec->start = start;
1841 failrec->len = end - start + 1;
1842 failrec->last_mirror = 0;
1843 failrec->bio_flags = 0;
1845 read_lock(&em_tree->lock);
1846 em = lookup_extent_mapping(em_tree, start, failrec->len);
1847 if (em->start > start || em->start + em->len < start) {
1848 free_extent_map(em);
1851 read_unlock(&em_tree->lock);
1853 if (!em || IS_ERR(em)) {
1857 logical = start - em->start;
1858 logical = em->block_start + logical;
1859 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1860 logical = em->block_start;
1861 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1862 extent_set_compress_type(&failrec->bio_flags,
1865 failrec->logical = logical;
1866 free_extent_map(em);
1867 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1868 EXTENT_DIRTY, GFP_NOFS);
1869 set_state_private(failure_tree, start,
1870 (u64)(unsigned long)failrec);
1872 failrec = (struct io_failure_record *)(unsigned long)private;
1874 num_copies = btrfs_num_copies(
1875 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1876 failrec->logical, failrec->len);
1877 failrec->last_mirror++;
1879 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1880 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1883 if (state && state->start != failrec->start)
1885 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1887 if (!state || failrec->last_mirror > num_copies) {
1888 set_state_private(failure_tree, failrec->start, 0);
1889 clear_extent_bits(failure_tree, failrec->start,
1890 failrec->start + failrec->len - 1,
1891 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1895 bio = bio_alloc(GFP_NOFS, 1);
1896 bio->bi_private = state;
1897 bio->bi_end_io = failed_bio->bi_end_io;
1898 bio->bi_sector = failrec->logical >> 9;
1899 bio->bi_bdev = failed_bio->bi_bdev;
1902 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1903 if (failed_bio->bi_rw & REQ_WRITE)
1908 ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1909 failrec->last_mirror,
1910 failrec->bio_flags, 0);
1915 * each time an IO finishes, we do a fast check in the IO failure tree
1916 * to see if we need to process or clean up an io_failure_record
1918 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1921 u64 private_failure;
1922 struct io_failure_record *failure;
1926 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1927 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1928 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1929 start, &private_failure);
1931 failure = (struct io_failure_record *)(unsigned long)
1933 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1935 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1937 failure->start + failure->len - 1,
1938 EXTENT_DIRTY | EXTENT_LOCKED,
1947 * when reads are done, we need to check csums to verify the data is correct
1948 * if there's a match, we allow the bio to finish. If not, we go through
1949 * the io_failure_record routines to find good copies
1951 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1952 struct extent_state *state)
1954 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1955 struct inode *inode = page->mapping->host;
1956 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1958 u64 private = ~(u32)0;
1960 struct btrfs_root *root = BTRFS_I(inode)->root;
1963 if (PageChecked(page)) {
1964 ClearPageChecked(page);
1968 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1971 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1972 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1973 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1978 if (state && state->start == start) {
1979 private = state->private;
1982 ret = get_state_private(io_tree, start, &private);
1984 kaddr = kmap_atomic(page, KM_USER0);
1988 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1989 btrfs_csum_final(csum, (char *)&csum);
1990 if (csum != private)
1993 kunmap_atomic(kaddr, KM_USER0);
1995 /* if the io failure tree for this inode is non-empty,
1996 * check to see if we've recovered from a failed IO
1998 btrfs_clean_io_failures(inode, start);
2002 if (printk_ratelimit()) {
2003 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2004 "private %llu\n", page->mapping->host->i_ino,
2005 (unsigned long long)start, csum,
2006 (unsigned long long)private);
2008 memset(kaddr + offset, 1, end - start + 1);
2009 flush_dcache_page(page);
2010 kunmap_atomic(kaddr, KM_USER0);
2016 struct delayed_iput {
2017 struct list_head list;
2018 struct inode *inode;
2021 void btrfs_add_delayed_iput(struct inode *inode)
2023 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2024 struct delayed_iput *delayed;
2026 if (atomic_add_unless(&inode->i_count, -1, 1))
2029 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2030 delayed->inode = inode;
2032 spin_lock(&fs_info->delayed_iput_lock);
2033 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2034 spin_unlock(&fs_info->delayed_iput_lock);
2037 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2040 struct btrfs_fs_info *fs_info = root->fs_info;
2041 struct delayed_iput *delayed;
2044 spin_lock(&fs_info->delayed_iput_lock);
2045 empty = list_empty(&fs_info->delayed_iputs);
2046 spin_unlock(&fs_info->delayed_iput_lock);
2050 down_read(&root->fs_info->cleanup_work_sem);
2051 spin_lock(&fs_info->delayed_iput_lock);
2052 list_splice_init(&fs_info->delayed_iputs, &list);
2053 spin_unlock(&fs_info->delayed_iput_lock);
2055 while (!list_empty(&list)) {
2056 delayed = list_entry(list.next, struct delayed_iput, list);
2057 list_del(&delayed->list);
2058 iput(delayed->inode);
2061 up_read(&root->fs_info->cleanup_work_sem);
2065 * calculate extra metadata reservation when snapshotting a subvolume
2066 * contains orphan files.
2068 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2069 struct btrfs_pending_snapshot *pending,
2070 u64 *bytes_to_reserve)
2072 struct btrfs_root *root;
2073 struct btrfs_block_rsv *block_rsv;
2077 root = pending->root;
2078 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2081 block_rsv = root->orphan_block_rsv;
2083 /* orphan block reservation for the snapshot */
2084 num_bytes = block_rsv->size;
2087 * after the snapshot is created, COWing tree blocks may use more
2088 * space than it frees. So we should make sure there is enough
2091 index = trans->transid & 0x1;
2092 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2093 num_bytes += block_rsv->size -
2094 (block_rsv->reserved + block_rsv->freed[index]);
2097 *bytes_to_reserve += num_bytes;
2100 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2101 struct btrfs_pending_snapshot *pending)
2103 struct btrfs_root *root = pending->root;
2104 struct btrfs_root *snap = pending->snap;
2105 struct btrfs_block_rsv *block_rsv;
2110 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2113 /* refill source subvolume's orphan block reservation */
2114 block_rsv = root->orphan_block_rsv;
2115 index = trans->transid & 0x1;
2116 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2117 num_bytes = block_rsv->size -
2118 (block_rsv->reserved + block_rsv->freed[index]);
2119 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2120 root->orphan_block_rsv,
2125 /* setup orphan block reservation for the snapshot */
2126 block_rsv = btrfs_alloc_block_rsv(snap);
2129 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2130 snap->orphan_block_rsv = block_rsv;
2132 num_bytes = root->orphan_block_rsv->size;
2133 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2134 block_rsv, num_bytes);
2138 /* insert orphan item for the snapshot */
2139 WARN_ON(!root->orphan_item_inserted);
2140 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2141 snap->root_key.objectid);
2143 snap->orphan_item_inserted = 1;
2147 enum btrfs_orphan_cleanup_state {
2148 ORPHAN_CLEANUP_STARTED = 1,
2149 ORPHAN_CLEANUP_DONE = 2,
2153 * This is called in transaction commmit time. If there are no orphan
2154 * files in the subvolume, it removes orphan item and frees block_rsv
2157 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2158 struct btrfs_root *root)
2162 if (!list_empty(&root->orphan_list) ||
2163 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2166 if (root->orphan_item_inserted &&
2167 btrfs_root_refs(&root->root_item) > 0) {
2168 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2169 root->root_key.objectid);
2171 root->orphan_item_inserted = 0;
2174 if (root->orphan_block_rsv) {
2175 WARN_ON(root->orphan_block_rsv->size > 0);
2176 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2177 root->orphan_block_rsv = NULL;
2182 * This creates an orphan entry for the given inode in case something goes
2183 * wrong in the middle of an unlink/truncate.
2185 * NOTE: caller of this function should reserve 5 units of metadata for
2188 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2190 struct btrfs_root *root = BTRFS_I(inode)->root;
2191 struct btrfs_block_rsv *block_rsv = NULL;
2196 if (!root->orphan_block_rsv) {
2197 block_rsv = btrfs_alloc_block_rsv(root);
2201 spin_lock(&root->orphan_lock);
2202 if (!root->orphan_block_rsv) {
2203 root->orphan_block_rsv = block_rsv;
2204 } else if (block_rsv) {
2205 btrfs_free_block_rsv(root, block_rsv);
2209 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2210 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2213 * For proper ENOSPC handling, we should do orphan
2214 * cleanup when mounting. But this introduces backward
2215 * compatibility issue.
2217 if (!xchg(&root->orphan_item_inserted, 1))
2225 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2226 BTRFS_I(inode)->orphan_meta_reserved = 1;
2229 spin_unlock(&root->orphan_lock);
2232 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2234 /* grab metadata reservation from transaction handle */
2236 ret = btrfs_orphan_reserve_metadata(trans, inode);
2240 /* insert an orphan item to track this unlinked/truncated file */
2242 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2246 /* insert an orphan item to track subvolume contains orphan files */
2248 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2249 root->root_key.objectid);
2256 * We have done the truncate/delete so we can go ahead and remove the orphan
2257 * item for this particular inode.
2259 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2261 struct btrfs_root *root = BTRFS_I(inode)->root;
2262 int delete_item = 0;
2263 int release_rsv = 0;
2266 spin_lock(&root->orphan_lock);
2267 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2268 list_del_init(&BTRFS_I(inode)->i_orphan);
2272 if (BTRFS_I(inode)->orphan_meta_reserved) {
2273 BTRFS_I(inode)->orphan_meta_reserved = 0;
2276 spin_unlock(&root->orphan_lock);
2278 if (trans && delete_item) {
2279 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2284 btrfs_orphan_release_metadata(inode);
2290 * this cleans up any orphans that may be left on the list from the last use
2293 int btrfs_orphan_cleanup(struct btrfs_root *root)
2295 struct btrfs_path *path;
2296 struct extent_buffer *leaf;
2297 struct btrfs_key key, found_key;
2298 struct btrfs_trans_handle *trans;
2299 struct inode *inode;
2300 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2302 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2305 path = btrfs_alloc_path();
2312 key.objectid = BTRFS_ORPHAN_OBJECTID;
2313 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2314 key.offset = (u64)-1;
2317 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2322 * if ret == 0 means we found what we were searching for, which
2323 * is weird, but possible, so only screw with path if we didnt
2324 * find the key and see if we have stuff that matches
2328 if (path->slots[0] == 0)
2333 /* pull out the item */
2334 leaf = path->nodes[0];
2335 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2337 /* make sure the item matches what we want */
2338 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2340 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2343 /* release the path since we're done with it */
2344 btrfs_release_path(root, path);
2347 * this is where we are basically btrfs_lookup, without the
2348 * crossing root thing. we store the inode number in the
2349 * offset of the orphan item.
2351 found_key.objectid = found_key.offset;
2352 found_key.type = BTRFS_INODE_ITEM_KEY;
2353 found_key.offset = 0;
2354 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2355 if (IS_ERR(inode)) {
2356 ret = PTR_ERR(inode);
2361 * add this inode to the orphan list so btrfs_orphan_del does
2362 * the proper thing when we hit it
2364 spin_lock(&root->orphan_lock);
2365 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2366 spin_unlock(&root->orphan_lock);
2369 * if this is a bad inode, means we actually succeeded in
2370 * removing the inode, but not the orphan record, which means
2371 * we need to manually delete the orphan since iput will just
2372 * do a destroy_inode
2374 if (is_bad_inode(inode)) {
2375 trans = btrfs_start_transaction(root, 0);
2376 if (IS_ERR(trans)) {
2377 ret = PTR_ERR(trans);
2380 btrfs_orphan_del(trans, inode);
2381 btrfs_end_transaction(trans, root);
2386 /* if we have links, this was a truncate, lets do that */
2387 if (inode->i_nlink) {
2388 if (!S_ISREG(inode->i_mode)) {
2394 ret = btrfs_truncate(inode);
2399 /* this will do delete_inode and everything for us */
2404 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2406 if (root->orphan_block_rsv)
2407 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2410 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2411 trans = btrfs_join_transaction(root, 1);
2413 btrfs_end_transaction(trans, root);
2417 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2419 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2423 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2424 btrfs_free_path(path);
2429 * very simple check to peek ahead in the leaf looking for xattrs. If we
2430 * don't find any xattrs, we know there can't be any acls.
2432 * slot is the slot the inode is in, objectid is the objectid of the inode
2434 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2435 int slot, u64 objectid)
2437 u32 nritems = btrfs_header_nritems(leaf);
2438 struct btrfs_key found_key;
2442 while (slot < nritems) {
2443 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2445 /* we found a different objectid, there must not be acls */
2446 if (found_key.objectid != objectid)
2449 /* we found an xattr, assume we've got an acl */
2450 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2454 * we found a key greater than an xattr key, there can't
2455 * be any acls later on
2457 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2464 * it goes inode, inode backrefs, xattrs, extents,
2465 * so if there are a ton of hard links to an inode there can
2466 * be a lot of backrefs. Don't waste time searching too hard,
2467 * this is just an optimization
2472 /* we hit the end of the leaf before we found an xattr or
2473 * something larger than an xattr. We have to assume the inode
2480 * read an inode from the btree into the in-memory inode
2482 static void btrfs_read_locked_inode(struct inode *inode)
2484 struct btrfs_path *path;
2485 struct extent_buffer *leaf;
2486 struct btrfs_inode_item *inode_item;
2487 struct btrfs_timespec *tspec;
2488 struct btrfs_root *root = BTRFS_I(inode)->root;
2489 struct btrfs_key location;
2491 u64 alloc_group_block;
2495 path = btrfs_alloc_path();
2497 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2499 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2503 leaf = path->nodes[0];
2504 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2505 struct btrfs_inode_item);
2507 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2508 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2509 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2510 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2511 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2513 tspec = btrfs_inode_atime(inode_item);
2514 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2515 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2517 tspec = btrfs_inode_mtime(inode_item);
2518 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2521 tspec = btrfs_inode_ctime(inode_item);
2522 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2525 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2526 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2527 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2528 inode->i_generation = BTRFS_I(inode)->generation;
2530 rdev = btrfs_inode_rdev(leaf, inode_item);
2532 BTRFS_I(inode)->index_cnt = (u64)-1;
2533 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2535 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2538 * try to precache a NULL acl entry for files that don't have
2539 * any xattrs or acls
2541 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2543 cache_no_acl(inode);
2545 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2546 alloc_group_block, 0);
2547 btrfs_free_path(path);
2550 switch (inode->i_mode & S_IFMT) {
2552 inode->i_mapping->a_ops = &btrfs_aops;
2553 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2554 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2555 inode->i_fop = &btrfs_file_operations;
2556 inode->i_op = &btrfs_file_inode_operations;
2559 inode->i_fop = &btrfs_dir_file_operations;
2560 if (root == root->fs_info->tree_root)
2561 inode->i_op = &btrfs_dir_ro_inode_operations;
2563 inode->i_op = &btrfs_dir_inode_operations;
2566 inode->i_op = &btrfs_symlink_inode_operations;
2567 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2568 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2571 inode->i_op = &btrfs_special_inode_operations;
2572 init_special_inode(inode, inode->i_mode, rdev);
2576 btrfs_update_iflags(inode);
2580 btrfs_free_path(path);
2581 make_bad_inode(inode);
2585 * given a leaf and an inode, copy the inode fields into the leaf
2587 static void fill_inode_item(struct btrfs_trans_handle *trans,
2588 struct extent_buffer *leaf,
2589 struct btrfs_inode_item *item,
2590 struct inode *inode)
2592 if (!leaf->map_token)
2593 map_private_extent_buffer(leaf, (unsigned long)item,
2594 sizeof(struct btrfs_inode_item),
2595 &leaf->map_token, &leaf->kaddr,
2596 &leaf->map_start, &leaf->map_len,
2599 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2600 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2601 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2602 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2603 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2605 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2606 inode->i_atime.tv_sec);
2607 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2608 inode->i_atime.tv_nsec);
2610 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2611 inode->i_mtime.tv_sec);
2612 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2613 inode->i_mtime.tv_nsec);
2615 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2616 inode->i_ctime.tv_sec);
2617 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2618 inode->i_ctime.tv_nsec);
2620 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2621 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2622 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2623 btrfs_set_inode_transid(leaf, item, trans->transid);
2624 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2625 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2626 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2628 if (leaf->map_token) {
2629 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2630 leaf->map_token = NULL;
2635 * copy everything in the in-memory inode into the btree.
2637 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2638 struct btrfs_root *root, struct inode *inode)
2640 struct btrfs_inode_item *inode_item;
2641 struct btrfs_path *path;
2642 struct extent_buffer *leaf;
2645 path = btrfs_alloc_path();
2647 path->leave_spinning = 1;
2648 ret = btrfs_lookup_inode(trans, root, path,
2649 &BTRFS_I(inode)->location, 1);
2656 btrfs_unlock_up_safe(path, 1);
2657 leaf = path->nodes[0];
2658 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2659 struct btrfs_inode_item);
2661 fill_inode_item(trans, leaf, inode_item, inode);
2662 btrfs_mark_buffer_dirty(leaf);
2663 btrfs_set_inode_last_trans(trans, inode);
2666 btrfs_free_path(path);
2672 * unlink helper that gets used here in inode.c and in the tree logging
2673 * recovery code. It remove a link in a directory with a given name, and
2674 * also drops the back refs in the inode to the directory
2676 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2677 struct btrfs_root *root,
2678 struct inode *dir, struct inode *inode,
2679 const char *name, int name_len)
2681 struct btrfs_path *path;
2683 struct extent_buffer *leaf;
2684 struct btrfs_dir_item *di;
2685 struct btrfs_key key;
2688 path = btrfs_alloc_path();
2694 path->leave_spinning = 1;
2695 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2696 name, name_len, -1);
2705 leaf = path->nodes[0];
2706 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2707 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2710 btrfs_release_path(root, path);
2712 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2714 dir->i_ino, &index);
2716 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2717 "inode %lu parent %lu\n", name_len, name,
2718 inode->i_ino, dir->i_ino);
2722 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2723 index, name, name_len, -1);
2732 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2733 btrfs_release_path(root, path);
2735 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2737 BUG_ON(ret != 0 && ret != -ENOENT);
2739 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2744 btrfs_free_path(path);
2748 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2749 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2750 btrfs_update_inode(trans, root, dir);
2755 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2756 struct btrfs_root *root,
2757 struct inode *dir, struct inode *inode,
2758 const char *name, int name_len)
2761 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2763 btrfs_drop_nlink(inode);
2764 ret = btrfs_update_inode(trans, root, inode);
2770 /* helper to check if there is any shared block in the path */
2771 static int check_path_shared(struct btrfs_root *root,
2772 struct btrfs_path *path)
2774 struct extent_buffer *eb;
2778 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2781 if (!path->nodes[level])
2783 eb = path->nodes[level];
2784 if (!btrfs_block_can_be_shared(root, eb))
2786 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2795 * helper to start transaction for unlink and rmdir.
2797 * unlink and rmdir are special in btrfs, they do not always free space.
2798 * so in enospc case, we should make sure they will free space before
2799 * allowing them to use the global metadata reservation.
2801 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2802 struct dentry *dentry)
2804 struct btrfs_trans_handle *trans;
2805 struct btrfs_root *root = BTRFS_I(dir)->root;
2806 struct btrfs_path *path;
2807 struct btrfs_inode_ref *ref;
2808 struct btrfs_dir_item *di;
2809 struct inode *inode = dentry->d_inode;
2815 trans = btrfs_start_transaction(root, 10);
2816 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2819 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2820 return ERR_PTR(-ENOSPC);
2822 /* check if there is someone else holds reference */
2823 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2824 return ERR_PTR(-ENOSPC);
2826 if (atomic_read(&inode->i_count) > 2)
2827 return ERR_PTR(-ENOSPC);
2829 if (xchg(&root->fs_info->enospc_unlink, 1))
2830 return ERR_PTR(-ENOSPC);
2832 path = btrfs_alloc_path();
2834 root->fs_info->enospc_unlink = 0;
2835 return ERR_PTR(-ENOMEM);
2838 trans = btrfs_start_transaction(root, 0);
2839 if (IS_ERR(trans)) {
2840 btrfs_free_path(path);
2841 root->fs_info->enospc_unlink = 0;
2845 path->skip_locking = 1;
2846 path->search_commit_root = 1;
2848 ret = btrfs_lookup_inode(trans, root, path,
2849 &BTRFS_I(dir)->location, 0);
2855 if (check_path_shared(root, path))
2860 btrfs_release_path(root, path);
2862 ret = btrfs_lookup_inode(trans, root, path,
2863 &BTRFS_I(inode)->location, 0);
2869 if (check_path_shared(root, path))
2874 btrfs_release_path(root, path);
2876 if (ret == 0 && S_ISREG(inode->i_mode)) {
2877 ret = btrfs_lookup_file_extent(trans, root, path,
2878 inode->i_ino, (u64)-1, 0);
2884 if (check_path_shared(root, path))
2886 btrfs_release_path(root, path);
2894 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2895 dentry->d_name.name, dentry->d_name.len, 0);
2901 if (check_path_shared(root, path))
2907 btrfs_release_path(root, path);
2909 ref = btrfs_lookup_inode_ref(trans, root, path,
2910 dentry->d_name.name, dentry->d_name.len,
2911 inode->i_ino, dir->i_ino, 0);
2917 if (check_path_shared(root, path))
2919 index = btrfs_inode_ref_index(path->nodes[0], ref);
2920 btrfs_release_path(root, path);
2922 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2923 dentry->d_name.name, dentry->d_name.len, 0);
2928 BUG_ON(ret == -ENOENT);
2929 if (check_path_shared(root, path))
2934 btrfs_free_path(path);
2936 btrfs_end_transaction(trans, root);
2937 root->fs_info->enospc_unlink = 0;
2938 return ERR_PTR(err);
2941 trans->block_rsv = &root->fs_info->global_block_rsv;
2945 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2946 struct btrfs_root *root)
2948 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2949 BUG_ON(!root->fs_info->enospc_unlink);
2950 root->fs_info->enospc_unlink = 0;
2952 btrfs_end_transaction_throttle(trans, root);
2955 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2957 struct btrfs_root *root = BTRFS_I(dir)->root;
2958 struct btrfs_trans_handle *trans;
2959 struct inode *inode = dentry->d_inode;
2961 unsigned long nr = 0;
2963 trans = __unlink_start_trans(dir, dentry);
2965 return PTR_ERR(trans);
2967 btrfs_set_trans_block_group(trans, dir);
2969 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2971 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2972 dentry->d_name.name, dentry->d_name.len);
2975 if (inode->i_nlink == 0) {
2976 ret = btrfs_orphan_add(trans, inode);
2980 nr = trans->blocks_used;
2981 __unlink_end_trans(trans, root);
2982 btrfs_btree_balance_dirty(root, nr);
2986 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2987 struct btrfs_root *root,
2988 struct inode *dir, u64 objectid,
2989 const char *name, int name_len)
2991 struct btrfs_path *path;
2992 struct extent_buffer *leaf;
2993 struct btrfs_dir_item *di;
2994 struct btrfs_key key;
2998 path = btrfs_alloc_path();
3002 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3003 name, name_len, -1);
3004 BUG_ON(!di || IS_ERR(di));
3006 leaf = path->nodes[0];
3007 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3008 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3009 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3011 btrfs_release_path(root, path);
3013 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3014 objectid, root->root_key.objectid,
3015 dir->i_ino, &index, name, name_len);
3017 BUG_ON(ret != -ENOENT);
3018 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3020 BUG_ON(!di || IS_ERR(di));
3022 leaf = path->nodes[0];
3023 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3024 btrfs_release_path(root, path);
3028 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3029 index, name, name_len, -1);
3030 BUG_ON(!di || IS_ERR(di));
3032 leaf = path->nodes[0];
3033 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3034 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3035 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3037 btrfs_release_path(root, path);
3039 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3040 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3041 ret = btrfs_update_inode(trans, root, dir);
3044 btrfs_free_path(path);
3048 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3050 struct inode *inode = dentry->d_inode;
3052 struct btrfs_root *root = BTRFS_I(dir)->root;
3053 struct btrfs_trans_handle *trans;
3054 unsigned long nr = 0;
3056 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3057 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3060 trans = __unlink_start_trans(dir, dentry);
3062 return PTR_ERR(trans);
3064 btrfs_set_trans_block_group(trans, dir);
3066 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3067 err = btrfs_unlink_subvol(trans, root, dir,
3068 BTRFS_I(inode)->location.objectid,
3069 dentry->d_name.name,
3070 dentry->d_name.len);
3074 err = btrfs_orphan_add(trans, inode);
3078 /* now the directory is empty */
3079 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3080 dentry->d_name.name, dentry->d_name.len);
3082 btrfs_i_size_write(inode, 0);
3084 nr = trans->blocks_used;
3085 __unlink_end_trans(trans, root);
3086 btrfs_btree_balance_dirty(root, nr);
3093 * when truncating bytes in a file, it is possible to avoid reading
3094 * the leaves that contain only checksum items. This can be the
3095 * majority of the IO required to delete a large file, but it must
3096 * be done carefully.
3098 * The keys in the level just above the leaves are checked to make sure
3099 * the lowest key in a given leaf is a csum key, and starts at an offset
3100 * after the new size.
3102 * Then the key for the next leaf is checked to make sure it also has
3103 * a checksum item for the same file. If it does, we know our target leaf
3104 * contains only checksum items, and it can be safely freed without reading
3107 * This is just an optimization targeted at large files. It may do
3108 * nothing. It will return 0 unless things went badly.
3110 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3111 struct btrfs_root *root,
3112 struct btrfs_path *path,
3113 struct inode *inode, u64 new_size)
3115 struct btrfs_key key;
3118 struct btrfs_key found_key;
3119 struct btrfs_key other_key;
3120 struct btrfs_leaf_ref *ref;
3124 path->lowest_level = 1;
3125 key.objectid = inode->i_ino;
3126 key.type = BTRFS_CSUM_ITEM_KEY;
3127 key.offset = new_size;
3129 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3133 if (path->nodes[1] == NULL) {
3138 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3139 nritems = btrfs_header_nritems(path->nodes[1]);
3144 if (path->slots[1] >= nritems)
3147 /* did we find a key greater than anything we want to delete? */
3148 if (found_key.objectid > inode->i_ino ||
3149 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3152 /* we check the next key in the node to make sure the leave contains
3153 * only checksum items. This comparison doesn't work if our
3154 * leaf is the last one in the node
3156 if (path->slots[1] + 1 >= nritems) {
3158 /* search forward from the last key in the node, this
3159 * will bring us into the next node in the tree
3161 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3163 /* unlikely, but we inc below, so check to be safe */
3164 if (found_key.offset == (u64)-1)
3167 /* search_forward needs a path with locks held, do the
3168 * search again for the original key. It is possible
3169 * this will race with a balance and return a path that
3170 * we could modify, but this drop is just an optimization
3171 * and is allowed to miss some leaves.
3173 btrfs_release_path(root, path);
3176 /* setup a max key for search_forward */
3177 other_key.offset = (u64)-1;
3178 other_key.type = key.type;
3179 other_key.objectid = key.objectid;
3181 path->keep_locks = 1;
3182 ret = btrfs_search_forward(root, &found_key, &other_key,
3184 path->keep_locks = 0;
3185 if (ret || found_key.objectid != key.objectid ||
3186 found_key.type != key.type) {
3191 key.offset = found_key.offset;
3192 btrfs_release_path(root, path);
3197 /* we know there's one more slot after us in the tree,
3198 * read that key so we can verify it is also a checksum item
3200 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3202 if (found_key.objectid < inode->i_ino)
3205 if (found_key.type != key.type || found_key.offset < new_size)
3209 * if the key for the next leaf isn't a csum key from this objectid,
3210 * we can't be sure there aren't good items inside this leaf.
3213 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3216 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3217 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3219 * it is safe to delete this leaf, it contains only
3220 * csum items from this inode at an offset >= new_size
3222 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3225 if (root->ref_cows && leaf_gen < trans->transid) {
3226 ref = btrfs_alloc_leaf_ref(root, 0);
3228 ref->root_gen = root->root_key.offset;
3229 ref->bytenr = leaf_start;
3231 ref->generation = leaf_gen;
3234 btrfs_sort_leaf_ref(ref);
3236 ret = btrfs_add_leaf_ref(root, ref, 0);
3238 btrfs_free_leaf_ref(root, ref);
3244 btrfs_release_path(root, path);
3246 if (other_key.objectid == inode->i_ino &&
3247 other_key.type == key.type && other_key.offset > key.offset) {
3248 key.offset = other_key.offset;
3254 /* fixup any changes we've made to the path */
3255 path->lowest_level = 0;
3256 path->keep_locks = 0;
3257 btrfs_release_path(root, path);
3264 * this can truncate away extent items, csum items and directory items.
3265 * It starts at a high offset and removes keys until it can't find
3266 * any higher than new_size
3268 * csum items that cross the new i_size are truncated to the new size
3271 * min_type is the minimum key type to truncate down to. If set to 0, this
3272 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3274 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3275 struct btrfs_root *root,
3276 struct inode *inode,
3277 u64 new_size, u32 min_type)
3279 struct btrfs_path *path;
3280 struct extent_buffer *leaf;
3281 struct btrfs_file_extent_item *fi;
3282 struct btrfs_key key;
3283 struct btrfs_key found_key;
3284 u64 extent_start = 0;
3285 u64 extent_num_bytes = 0;
3286 u64 extent_offset = 0;
3288 u64 mask = root->sectorsize - 1;
3289 u32 found_type = (u8)-1;
3292 int pending_del_nr = 0;
3293 int pending_del_slot = 0;
3294 int extent_type = -1;
3299 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3301 if (root->ref_cows || root == root->fs_info->tree_root)
3302 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3304 path = btrfs_alloc_path();
3308 key.objectid = inode->i_ino;
3309 key.offset = (u64)-1;
3313 path->leave_spinning = 1;
3314 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3321 /* there are no items in the tree for us to truncate, we're
3324 if (path->slots[0] == 0)
3331 leaf = path->nodes[0];
3332 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3333 found_type = btrfs_key_type(&found_key);
3336 if (found_key.objectid != inode->i_ino)
3339 if (found_type < min_type)
3342 item_end = found_key.offset;
3343 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3344 fi = btrfs_item_ptr(leaf, path->slots[0],
3345 struct btrfs_file_extent_item);
3346 extent_type = btrfs_file_extent_type(leaf, fi);
3347 encoding = btrfs_file_extent_compression(leaf, fi);
3348 encoding |= btrfs_file_extent_encryption(leaf, fi);
3349 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3351 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3353 btrfs_file_extent_num_bytes(leaf, fi);
3354 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3355 item_end += btrfs_file_extent_inline_len(leaf,
3360 if (found_type > min_type) {
3363 if (item_end < new_size)
3365 if (found_key.offset >= new_size)
3371 /* FIXME, shrink the extent if the ref count is only 1 */
3372 if (found_type != BTRFS_EXTENT_DATA_KEY)
3375 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3377 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3378 if (!del_item && !encoding) {
3379 u64 orig_num_bytes =
3380 btrfs_file_extent_num_bytes(leaf, fi);
3381 extent_num_bytes = new_size -
3382 found_key.offset + root->sectorsize - 1;
3383 extent_num_bytes = extent_num_bytes &
3384 ~((u64)root->sectorsize - 1);
3385 btrfs_set_file_extent_num_bytes(leaf, fi,
3387 num_dec = (orig_num_bytes -
3389 if (root->ref_cows && extent_start != 0)
3390 inode_sub_bytes(inode, num_dec);
3391 btrfs_mark_buffer_dirty(leaf);
3394 btrfs_file_extent_disk_num_bytes(leaf,
3396 extent_offset = found_key.offset -
3397 btrfs_file_extent_offset(leaf, fi);
3399 /* FIXME blocksize != 4096 */
3400 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3401 if (extent_start != 0) {
3404 inode_sub_bytes(inode, num_dec);
3407 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3409 * we can't truncate inline items that have had
3413 btrfs_file_extent_compression(leaf, fi) == 0 &&
3414 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3415 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3416 u32 size = new_size - found_key.offset;
3418 if (root->ref_cows) {
3419 inode_sub_bytes(inode, item_end + 1 -
3423 btrfs_file_extent_calc_inline_size(size);
3424 ret = btrfs_truncate_item(trans, root, path,
3427 } else if (root->ref_cows) {
3428 inode_sub_bytes(inode, item_end + 1 -
3434 if (!pending_del_nr) {
3435 /* no pending yet, add ourselves */
3436 pending_del_slot = path->slots[0];
3438 } else if (pending_del_nr &&
3439 path->slots[0] + 1 == pending_del_slot) {
3440 /* hop on the pending chunk */
3442 pending_del_slot = path->slots[0];
3449 if (found_extent && (root->ref_cows ||
3450 root == root->fs_info->tree_root)) {
3451 btrfs_set_path_blocking(path);
3452 ret = btrfs_free_extent(trans, root, extent_start,
3453 extent_num_bytes, 0,
3454 btrfs_header_owner(leaf),
3455 inode->i_ino, extent_offset);
3459 if (found_type == BTRFS_INODE_ITEM_KEY)
3462 if (path->slots[0] == 0 ||
3463 path->slots[0] != pending_del_slot) {
3464 if (root->ref_cows) {
3468 if (pending_del_nr) {
3469 ret = btrfs_del_items(trans, root, path,
3475 btrfs_release_path(root, path);
3482 if (pending_del_nr) {
3483 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3487 btrfs_free_path(path);
3492 * taken from block_truncate_page, but does cow as it zeros out
3493 * any bytes left in the last page in the file.
3495 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3497 struct inode *inode = mapping->host;
3498 struct btrfs_root *root = BTRFS_I(inode)->root;
3499 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3500 struct btrfs_ordered_extent *ordered;
3501 struct extent_state *cached_state = NULL;
3503 u32 blocksize = root->sectorsize;
3504 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3505 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3511 if ((offset & (blocksize - 1)) == 0)
3513 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3519 page = grab_cache_page(mapping, index);
3521 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3525 page_start = page_offset(page);
3526 page_end = page_start + PAGE_CACHE_SIZE - 1;
3528 if (!PageUptodate(page)) {
3529 ret = btrfs_readpage(NULL, page);
3531 if (page->mapping != mapping) {
3533 page_cache_release(page);
3536 if (!PageUptodate(page)) {
3541 wait_on_page_writeback(page);
3543 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3545 set_page_extent_mapped(page);
3547 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3549 unlock_extent_cached(io_tree, page_start, page_end,
3550 &cached_state, GFP_NOFS);
3552 page_cache_release(page);
3553 btrfs_start_ordered_extent(inode, ordered, 1);
3554 btrfs_put_ordered_extent(ordered);
3558 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3559 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3560 0, 0, &cached_state, GFP_NOFS);
3562 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3565 unlock_extent_cached(io_tree, page_start, page_end,
3566 &cached_state, GFP_NOFS);
3571 if (offset != PAGE_CACHE_SIZE) {
3573 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3574 flush_dcache_page(page);
3577 ClearPageChecked(page);
3578 set_page_dirty(page);
3579 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3584 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3586 page_cache_release(page);
3592 * This function puts in dummy file extents for the area we're creating a hole
3593 * for. So if we are truncating this file to a larger size we need to insert
3594 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3595 * the range between oldsize and size
3597 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3599 struct btrfs_trans_handle *trans;
3600 struct btrfs_root *root = BTRFS_I(inode)->root;
3601 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3602 struct extent_map *em = NULL;
3603 struct extent_state *cached_state = NULL;
3604 u64 mask = root->sectorsize - 1;
3605 u64 hole_start = (oldsize + mask) & ~mask;
3606 u64 block_end = (size + mask) & ~mask;
3612 if (size <= hole_start)
3616 struct btrfs_ordered_extent *ordered;
3617 btrfs_wait_ordered_range(inode, hole_start,
3618 block_end - hole_start);
3619 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3620 &cached_state, GFP_NOFS);
3621 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3624 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3625 &cached_state, GFP_NOFS);
3626 btrfs_put_ordered_extent(ordered);
3629 cur_offset = hole_start;
3631 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3632 block_end - cur_offset, 0);
3633 BUG_ON(IS_ERR(em) || !em);
3634 last_byte = min(extent_map_end(em), block_end);
3635 last_byte = (last_byte + mask) & ~mask;
3636 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3638 hole_size = last_byte - cur_offset;
3640 trans = btrfs_start_transaction(root, 2);
3641 if (IS_ERR(trans)) {
3642 err = PTR_ERR(trans);
3645 btrfs_set_trans_block_group(trans, inode);
3647 err = btrfs_drop_extents(trans, inode, cur_offset,
3648 cur_offset + hole_size,
3653 err = btrfs_insert_file_extent(trans, root,
3654 inode->i_ino, cur_offset, 0,
3655 0, hole_size, 0, hole_size,
3660 btrfs_drop_extent_cache(inode, hole_start,
3663 btrfs_end_transaction(trans, root);
3665 free_extent_map(em);
3667 cur_offset = last_byte;
3668 if (cur_offset >= block_end)
3672 free_extent_map(em);
3673 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3678 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3680 loff_t oldsize = i_size_read(inode);
3683 if (newsize == oldsize)
3686 if (newsize > oldsize) {
3687 i_size_write(inode, newsize);
3688 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3689 truncate_pagecache(inode, oldsize, newsize);
3690 ret = btrfs_cont_expand(inode, oldsize, newsize);
3692 btrfs_setsize(inode, oldsize);
3696 mark_inode_dirty(inode);
3700 * We're truncating a file that used to have good data down to
3701 * zero. Make sure it gets into the ordered flush list so that
3702 * any new writes get down to disk quickly.
3705 BTRFS_I(inode)->ordered_data_close = 1;
3707 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3708 truncate_setsize(inode, newsize);
3709 ret = btrfs_truncate(inode);
3715 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3717 struct inode *inode = dentry->d_inode;
3718 struct btrfs_root *root = BTRFS_I(inode)->root;
3721 if (btrfs_root_readonly(root))
3724 err = inode_change_ok(inode, attr);
3728 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3729 err = btrfs_setsize(inode, attr->ia_size);
3734 if (attr->ia_valid) {
3735 setattr_copy(inode, attr);
3736 mark_inode_dirty(inode);
3738 if (attr->ia_valid & ATTR_MODE)
3739 err = btrfs_acl_chmod(inode);
3745 void btrfs_evict_inode(struct inode *inode)
3747 struct btrfs_trans_handle *trans;
3748 struct btrfs_root *root = BTRFS_I(inode)->root;
3752 trace_btrfs_inode_evict(inode);
3754 truncate_inode_pages(&inode->i_data, 0);
3755 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3756 root == root->fs_info->tree_root))
3759 if (is_bad_inode(inode)) {
3760 btrfs_orphan_del(NULL, inode);
3763 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3764 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3766 if (root->fs_info->log_root_recovering) {
3767 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3771 if (inode->i_nlink > 0) {
3772 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3776 btrfs_i_size_write(inode, 0);
3779 trans = btrfs_start_transaction(root, 0);
3780 BUG_ON(IS_ERR(trans));
3781 btrfs_set_trans_block_group(trans, inode);
3782 trans->block_rsv = root->orphan_block_rsv;
3784 ret = btrfs_block_rsv_check(trans, root,
3785 root->orphan_block_rsv, 0, 5);
3787 BUG_ON(ret != -EAGAIN);
3788 ret = btrfs_commit_transaction(trans, root);
3793 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3797 nr = trans->blocks_used;
3798 btrfs_end_transaction(trans, root);
3800 btrfs_btree_balance_dirty(root, nr);
3805 ret = btrfs_orphan_del(trans, inode);
3809 nr = trans->blocks_used;
3810 btrfs_end_transaction(trans, root);
3811 btrfs_btree_balance_dirty(root, nr);
3813 end_writeback(inode);
3818 * this returns the key found in the dir entry in the location pointer.
3819 * If no dir entries were found, location->objectid is 0.
3821 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3822 struct btrfs_key *location)
3824 const char *name = dentry->d_name.name;
3825 int namelen = dentry->d_name.len;
3826 struct btrfs_dir_item *di;
3827 struct btrfs_path *path;
3828 struct btrfs_root *root = BTRFS_I(dir)->root;
3831 path = btrfs_alloc_path();
3834 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3839 if (!di || IS_ERR(di))
3842 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3844 btrfs_free_path(path);
3847 location->objectid = 0;
3852 * when we hit a tree root in a directory, the btrfs part of the inode
3853 * needs to be changed to reflect the root directory of the tree root. This
3854 * is kind of like crossing a mount point.
3856 static int fixup_tree_root_location(struct btrfs_root *root,
3858 struct dentry *dentry,
3859 struct btrfs_key *location,
3860 struct btrfs_root **sub_root)
3862 struct btrfs_path *path;
3863 struct btrfs_root *new_root;
3864 struct btrfs_root_ref *ref;
3865 struct extent_buffer *leaf;
3869 path = btrfs_alloc_path();
3876 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3877 BTRFS_I(dir)->root->root_key.objectid,
3878 location->objectid);
3885 leaf = path->nodes[0];
3886 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3887 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3888 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3891 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3892 (unsigned long)(ref + 1),
3893 dentry->d_name.len);
3897 btrfs_release_path(root->fs_info->tree_root, path);
3899 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3900 if (IS_ERR(new_root)) {
3901 err = PTR_ERR(new_root);
3905 if (btrfs_root_refs(&new_root->root_item) == 0) {
3910 *sub_root = new_root;
3911 location->objectid = btrfs_root_dirid(&new_root->root_item);
3912 location->type = BTRFS_INODE_ITEM_KEY;
3913 location->offset = 0;
3916 btrfs_free_path(path);
3920 static void inode_tree_add(struct inode *inode)
3922 struct btrfs_root *root = BTRFS_I(inode)->root;
3923 struct btrfs_inode *entry;
3925 struct rb_node *parent;
3927 p = &root->inode_tree.rb_node;
3930 if (inode_unhashed(inode))
3933 spin_lock(&root->inode_lock);
3936 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3938 if (inode->i_ino < entry->vfs_inode.i_ino)
3939 p = &parent->rb_left;
3940 else if (inode->i_ino > entry->vfs_inode.i_ino)
3941 p = &parent->rb_right;
3943 WARN_ON(!(entry->vfs_inode.i_state &
3944 (I_WILL_FREE | I_FREEING)));
3945 rb_erase(parent, &root->inode_tree);
3946 RB_CLEAR_NODE(parent);
3947 spin_unlock(&root->inode_lock);
3951 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3952 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3953 spin_unlock(&root->inode_lock);
3956 static void inode_tree_del(struct inode *inode)
3958 struct btrfs_root *root = BTRFS_I(inode)->root;
3961 spin_lock(&root->inode_lock);
3962 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3963 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3964 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3965 empty = RB_EMPTY_ROOT(&root->inode_tree);
3967 spin_unlock(&root->inode_lock);
3970 * Free space cache has inodes in the tree root, but the tree root has a
3971 * root_refs of 0, so this could end up dropping the tree root as a
3972 * snapshot, so we need the extra !root->fs_info->tree_root check to
3973 * make sure we don't drop it.
3975 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3976 root != root->fs_info->tree_root) {
3977 synchronize_srcu(&root->fs_info->subvol_srcu);
3978 spin_lock(&root->inode_lock);
3979 empty = RB_EMPTY_ROOT(&root->inode_tree);
3980 spin_unlock(&root->inode_lock);
3982 btrfs_add_dead_root(root);
3986 int btrfs_invalidate_inodes(struct btrfs_root *root)
3988 struct rb_node *node;
3989 struct rb_node *prev;
3990 struct btrfs_inode *entry;
3991 struct inode *inode;
3994 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3996 spin_lock(&root->inode_lock);
3998 node = root->inode_tree.rb_node;
4002 entry = rb_entry(node, struct btrfs_inode, rb_node);
4004 if (objectid < entry->vfs_inode.i_ino)
4005 node = node->rb_left;
4006 else if (objectid > entry->vfs_inode.i_ino)
4007 node = node->rb_right;
4013 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4014 if (objectid <= entry->vfs_inode.i_ino) {
4018 prev = rb_next(prev);
4022 entry = rb_entry(node, struct btrfs_inode, rb_node);
4023 objectid = entry->vfs_inode.i_ino + 1;
4024 inode = igrab(&entry->vfs_inode);
4026 spin_unlock(&root->inode_lock);
4027 if (atomic_read(&inode->i_count) > 1)
4028 d_prune_aliases(inode);
4030 * btrfs_drop_inode will have it removed from
4031 * the inode cache when its usage count
4036 spin_lock(&root->inode_lock);
4040 if (cond_resched_lock(&root->inode_lock))
4043 node = rb_next(node);
4045 spin_unlock(&root->inode_lock);
4049 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4051 struct btrfs_iget_args *args = p;
4052 inode->i_ino = args->ino;
4053 BTRFS_I(inode)->root = args->root;
4054 btrfs_set_inode_space_info(args->root, inode);
4058 static int btrfs_find_actor(struct inode *inode, void *opaque)
4060 struct btrfs_iget_args *args = opaque;
4061 return args->ino == inode->i_ino &&
4062 args->root == BTRFS_I(inode)->root;
4065 static struct inode *btrfs_iget_locked(struct super_block *s,
4067 struct btrfs_root *root)
4069 struct inode *inode;
4070 struct btrfs_iget_args args;
4071 args.ino = objectid;
4074 inode = iget5_locked(s, objectid, btrfs_find_actor,
4075 btrfs_init_locked_inode,
4080 /* Get an inode object given its location and corresponding root.
4081 * Returns in *is_new if the inode was read from disk
4083 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4084 struct btrfs_root *root, int *new)
4086 struct inode *inode;
4088 inode = btrfs_iget_locked(s, location->objectid, root);
4090 return ERR_PTR(-ENOMEM);
4092 if (inode->i_state & I_NEW) {
4093 BTRFS_I(inode)->root = root;
4094 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4095 btrfs_read_locked_inode(inode);
4096 inode_tree_add(inode);
4097 unlock_new_inode(inode);
4105 static struct inode *new_simple_dir(struct super_block *s,
4106 struct btrfs_key *key,
4107 struct btrfs_root *root)
4109 struct inode *inode = new_inode(s);
4112 return ERR_PTR(-ENOMEM);
4114 BTRFS_I(inode)->root = root;
4115 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4116 BTRFS_I(inode)->dummy_inode = 1;
4118 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4119 inode->i_op = &simple_dir_inode_operations;
4120 inode->i_fop = &simple_dir_operations;
4121 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4122 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4127 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4129 struct inode *inode;
4130 struct btrfs_root *root = BTRFS_I(dir)->root;
4131 struct btrfs_root *sub_root = root;
4132 struct btrfs_key location;
4136 if (dentry->d_name.len > BTRFS_NAME_LEN)
4137 return ERR_PTR(-ENAMETOOLONG);
4139 ret = btrfs_inode_by_name(dir, dentry, &location);
4142 return ERR_PTR(ret);
4144 if (location.objectid == 0)
4147 if (location.type == BTRFS_INODE_ITEM_KEY) {
4148 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4152 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4154 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4155 ret = fixup_tree_root_location(root, dir, dentry,
4156 &location, &sub_root);
4159 inode = ERR_PTR(ret);
4161 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4163 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4165 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4167 if (!IS_ERR(inode) && root != sub_root) {
4168 down_read(&root->fs_info->cleanup_work_sem);
4169 if (!(inode->i_sb->s_flags & MS_RDONLY))
4170 ret = btrfs_orphan_cleanup(sub_root);
4171 up_read(&root->fs_info->cleanup_work_sem);
4173 inode = ERR_PTR(ret);
4179 static int btrfs_dentry_delete(const struct dentry *dentry)
4181 struct btrfs_root *root;
4183 if (!dentry->d_inode && !IS_ROOT(dentry))
4184 dentry = dentry->d_parent;
4186 if (dentry->d_inode) {
4187 root = BTRFS_I(dentry->d_inode)->root;
4188 if (btrfs_root_refs(&root->root_item) == 0)
4194 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4195 struct nameidata *nd)
4197 struct inode *inode;
4199 inode = btrfs_lookup_dentry(dir, dentry);
4201 return ERR_CAST(inode);
4203 return d_splice_alias(inode, dentry);
4206 static unsigned char btrfs_filetype_table[] = {
4207 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4210 static int btrfs_real_readdir(struct file *filp, void *dirent,
4213 struct inode *inode = filp->f_dentry->d_inode;
4214 struct btrfs_root *root = BTRFS_I(inode)->root;
4215 struct btrfs_item *item;
4216 struct btrfs_dir_item *di;
4217 struct btrfs_key key;
4218 struct btrfs_key found_key;
4219 struct btrfs_path *path;
4222 struct extent_buffer *leaf;
4225 unsigned char d_type;
4230 int key_type = BTRFS_DIR_INDEX_KEY;
4235 /* FIXME, use a real flag for deciding about the key type */
4236 if (root->fs_info->tree_root == root)
4237 key_type = BTRFS_DIR_ITEM_KEY;
4239 /* special case for "." */
4240 if (filp->f_pos == 0) {
4241 over = filldir(dirent, ".", 1,
4248 /* special case for .., just use the back ref */
4249 if (filp->f_pos == 1) {
4250 u64 pino = parent_ino(filp->f_path.dentry);
4251 over = filldir(dirent, "..", 2,
4257 path = btrfs_alloc_path();
4260 btrfs_set_key_type(&key, key_type);
4261 key.offset = filp->f_pos;
4262 key.objectid = inode->i_ino;
4264 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4270 leaf = path->nodes[0];
4271 nritems = btrfs_header_nritems(leaf);
4272 slot = path->slots[0];
4273 if (advance || slot >= nritems) {
4274 if (slot >= nritems - 1) {
4275 ret = btrfs_next_leaf(root, path);
4278 leaf = path->nodes[0];
4279 nritems = btrfs_header_nritems(leaf);
4280 slot = path->slots[0];
4288 item = btrfs_item_nr(leaf, slot);
4289 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4291 if (found_key.objectid != key.objectid)
4293 if (btrfs_key_type(&found_key) != key_type)
4295 if (found_key.offset < filp->f_pos)
4298 filp->f_pos = found_key.offset;
4300 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4302 di_total = btrfs_item_size(leaf, item);
4304 while (di_cur < di_total) {
4305 struct btrfs_key location;
4307 if (verify_dir_item(root, leaf, di))
4310 name_len = btrfs_dir_name_len(leaf, di);
4311 if (name_len <= sizeof(tmp_name)) {
4312 name_ptr = tmp_name;
4314 name_ptr = kmalloc(name_len, GFP_NOFS);
4320 read_extent_buffer(leaf, name_ptr,
4321 (unsigned long)(di + 1), name_len);
4323 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4324 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4326 /* is this a reference to our own snapshot? If so
4329 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4330 location.objectid == root->root_key.objectid) {
4334 over = filldir(dirent, name_ptr, name_len,
4335 found_key.offset, location.objectid,
4339 if (name_ptr != tmp_name)
4344 di_len = btrfs_dir_name_len(leaf, di) +
4345 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4347 di = (struct btrfs_dir_item *)((char *)di + di_len);
4351 /* Reached end of directory/root. Bump pos past the last item. */
4352 if (key_type == BTRFS_DIR_INDEX_KEY)
4354 * 32-bit glibc will use getdents64, but then strtol -
4355 * so the last number we can serve is this.
4357 filp->f_pos = 0x7fffffff;
4363 btrfs_free_path(path);
4367 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4369 struct btrfs_root *root = BTRFS_I(inode)->root;
4370 struct btrfs_trans_handle *trans;
4372 bool nolock = false;
4374 if (BTRFS_I(inode)->dummy_inode)
4378 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4380 if (wbc->sync_mode == WB_SYNC_ALL) {
4382 trans = btrfs_join_transaction_nolock(root, 1);
4384 trans = btrfs_join_transaction(root, 1);
4386 return PTR_ERR(trans);
4387 btrfs_set_trans_block_group(trans, inode);
4389 ret = btrfs_end_transaction_nolock(trans, root);
4391 ret = btrfs_commit_transaction(trans, root);
4397 * This is somewhat expensive, updating the tree every time the
4398 * inode changes. But, it is most likely to find the inode in cache.
4399 * FIXME, needs more benchmarking...there are no reasons other than performance
4400 * to keep or drop this code.
4402 void btrfs_dirty_inode(struct inode *inode)
4404 struct btrfs_root *root = BTRFS_I(inode)->root;
4405 struct btrfs_trans_handle *trans;
4408 if (BTRFS_I(inode)->dummy_inode)
4411 trans = btrfs_join_transaction(root, 1);
4412 BUG_ON(IS_ERR(trans));
4413 btrfs_set_trans_block_group(trans, inode);
4415 ret = btrfs_update_inode(trans, root, inode);
4416 if (ret && ret == -ENOSPC) {
4417 /* whoops, lets try again with the full transaction */
4418 btrfs_end_transaction(trans, root);
4419 trans = btrfs_start_transaction(root, 1);
4420 if (IS_ERR(trans)) {
4421 if (printk_ratelimit()) {
4422 printk(KERN_ERR "btrfs: fail to "
4423 "dirty inode %lu error %ld\n",
4424 inode->i_ino, PTR_ERR(trans));
4428 btrfs_set_trans_block_group(trans, inode);
4430 ret = btrfs_update_inode(trans, root, inode);
4432 if (printk_ratelimit()) {
4433 printk(KERN_ERR "btrfs: fail to "
4434 "dirty inode %lu error %d\n",
4439 btrfs_end_transaction(trans, root);
4443 * find the highest existing sequence number in a directory
4444 * and then set the in-memory index_cnt variable to reflect
4445 * free sequence numbers
4447 static int btrfs_set_inode_index_count(struct inode *inode)
4449 struct btrfs_root *root = BTRFS_I(inode)->root;
4450 struct btrfs_key key, found_key;
4451 struct btrfs_path *path;
4452 struct extent_buffer *leaf;
4455 key.objectid = inode->i_ino;
4456 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4457 key.offset = (u64)-1;
4459 path = btrfs_alloc_path();
4463 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4466 /* FIXME: we should be able to handle this */
4472 * MAGIC NUMBER EXPLANATION:
4473 * since we search a directory based on f_pos we have to start at 2
4474 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4475 * else has to start at 2
4477 if (path->slots[0] == 0) {
4478 BTRFS_I(inode)->index_cnt = 2;
4484 leaf = path->nodes[0];
4485 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4487 if (found_key.objectid != inode->i_ino ||
4488 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4489 BTRFS_I(inode)->index_cnt = 2;
4493 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4495 btrfs_free_path(path);
4500 * helper to find a free sequence number in a given directory. This current
4501 * code is very simple, later versions will do smarter things in the btree
4503 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4507 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4508 ret = btrfs_set_inode_index_count(dir);
4513 *index = BTRFS_I(dir)->index_cnt;
4514 BTRFS_I(dir)->index_cnt++;
4519 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4520 struct btrfs_root *root,
4522 const char *name, int name_len,
4523 u64 ref_objectid, u64 objectid,
4524 u64 alloc_hint, int mode, u64 *index)
4526 struct inode *inode;
4527 struct btrfs_inode_item *inode_item;
4528 struct btrfs_key *location;
4529 struct btrfs_path *path;
4530 struct btrfs_inode_ref *ref;
4531 struct btrfs_key key[2];
4537 path = btrfs_alloc_path();
4540 inode = new_inode(root->fs_info->sb);
4542 return ERR_PTR(-ENOMEM);
4545 trace_btrfs_inode_request(dir);
4547 ret = btrfs_set_inode_index(dir, index);
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 err = PTR_ERR(inode);
4739 err = btrfs_init_inode_security(trans, inode, dir);
4745 btrfs_set_trans_block_group(trans, inode);
4746 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4750 inode->i_op = &btrfs_special_inode_operations;
4751 init_special_inode(inode, inode->i_mode, rdev);
4752 btrfs_update_inode(trans, root, inode);
4754 btrfs_update_inode_block_group(trans, inode);
4755 btrfs_update_inode_block_group(trans, dir);
4757 nr = trans->blocks_used;
4758 btrfs_end_transaction_throttle(trans, root);
4759 btrfs_btree_balance_dirty(root, nr);
4761 inode_dec_link_count(inode);
4767 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4768 int mode, struct nameidata *nd)
4770 struct btrfs_trans_handle *trans;
4771 struct btrfs_root *root = BTRFS_I(dir)->root;
4772 struct inode *inode = NULL;
4775 unsigned long nr = 0;
4779 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4783 * 2 for inode item and ref
4785 * 1 for xattr if selinux is on
4787 trans = btrfs_start_transaction(root, 5);
4789 return PTR_ERR(trans);
4791 btrfs_set_trans_block_group(trans, dir);
4793 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4794 dentry->d_name.len, dir->i_ino, objectid,
4795 BTRFS_I(dir)->block_group, mode, &index);
4796 err = PTR_ERR(inode);
4800 err = btrfs_init_inode_security(trans, inode, dir);
4806 btrfs_set_trans_block_group(trans, inode);
4807 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4811 inode->i_mapping->a_ops = &btrfs_aops;
4812 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4813 inode->i_fop = &btrfs_file_operations;
4814 inode->i_op = &btrfs_file_inode_operations;
4815 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4817 btrfs_update_inode_block_group(trans, inode);
4818 btrfs_update_inode_block_group(trans, dir);
4820 nr = trans->blocks_used;
4821 btrfs_end_transaction_throttle(trans, root);
4823 inode_dec_link_count(inode);
4826 btrfs_btree_balance_dirty(root, nr);
4830 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4831 struct dentry *dentry)
4833 struct btrfs_trans_handle *trans;
4834 struct btrfs_root *root = BTRFS_I(dir)->root;
4835 struct inode *inode = old_dentry->d_inode;
4837 unsigned long nr = 0;
4841 if (inode->i_nlink == 0)
4844 /* do not allow sys_link's with other subvols of the same device */
4845 if (root->objectid != BTRFS_I(inode)->root->objectid)
4848 if (inode->i_nlink == ~0U)
4851 btrfs_inc_nlink(inode);
4852 inode->i_ctime = CURRENT_TIME;
4854 err = btrfs_set_inode_index(dir, &index);
4859 * 2 items for inode and inode ref
4860 * 2 items for dir items
4861 * 1 item for parent inode
4863 trans = btrfs_start_transaction(root, 5);
4864 if (IS_ERR(trans)) {
4865 err = PTR_ERR(trans);
4869 btrfs_set_trans_block_group(trans, dir);
4872 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4877 struct dentry *parent = dget_parent(dentry);
4878 btrfs_update_inode_block_group(trans, dir);
4879 err = btrfs_update_inode(trans, root, inode);
4881 btrfs_log_new_name(trans, inode, NULL, parent);
4885 nr = trans->blocks_used;
4886 btrfs_end_transaction_throttle(trans, root);
4889 inode_dec_link_count(inode);
4892 btrfs_btree_balance_dirty(root, nr);
4896 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4898 struct inode *inode = NULL;
4899 struct btrfs_trans_handle *trans;
4900 struct btrfs_root *root = BTRFS_I(dir)->root;
4902 int drop_on_err = 0;
4905 unsigned long nr = 1;
4907 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4912 * 2 items for inode and ref
4913 * 2 items for dir items
4914 * 1 for xattr if selinux is on
4916 trans = btrfs_start_transaction(root, 5);
4918 return PTR_ERR(trans);
4919 btrfs_set_trans_block_group(trans, dir);
4921 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4922 dentry->d_name.len, dir->i_ino, objectid,
4923 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4925 if (IS_ERR(inode)) {
4926 err = PTR_ERR(inode);
4932 err = btrfs_init_inode_security(trans, inode, dir);
4936 inode->i_op = &btrfs_dir_inode_operations;
4937 inode->i_fop = &btrfs_dir_file_operations;
4938 btrfs_set_trans_block_group(trans, inode);
4940 btrfs_i_size_write(inode, 0);
4941 err = btrfs_update_inode(trans, root, inode);
4945 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4946 dentry->d_name.len, 0, index);
4950 d_instantiate(dentry, inode);
4952 btrfs_update_inode_block_group(trans, inode);
4953 btrfs_update_inode_block_group(trans, dir);
4956 nr = trans->blocks_used;
4957 btrfs_end_transaction_throttle(trans, root);
4960 btrfs_btree_balance_dirty(root, nr);
4964 /* helper for btfs_get_extent. Given an existing extent in the tree,
4965 * and an extent that you want to insert, deal with overlap and insert
4966 * the new extent into the tree.
4968 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4969 struct extent_map *existing,
4970 struct extent_map *em,
4971 u64 map_start, u64 map_len)
4975 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4976 start_diff = map_start - em->start;
4977 em->start = map_start;
4979 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4980 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4981 em->block_start += start_diff;
4982 em->block_len -= start_diff;
4984 return add_extent_mapping(em_tree, em);
4987 static noinline int uncompress_inline(struct btrfs_path *path,
4988 struct inode *inode, struct page *page,
4989 size_t pg_offset, u64 extent_offset,
4990 struct btrfs_file_extent_item *item)
4993 struct extent_buffer *leaf = path->nodes[0];
4996 unsigned long inline_size;
5000 WARN_ON(pg_offset != 0);
5001 compress_type = btrfs_file_extent_compression(leaf, item);
5002 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5003 inline_size = btrfs_file_extent_inline_item_len(leaf,
5004 btrfs_item_nr(leaf, path->slots[0]));
5005 tmp = kmalloc(inline_size, GFP_NOFS);
5006 ptr = btrfs_file_extent_inline_start(item);
5008 read_extent_buffer(leaf, tmp, ptr, inline_size);
5010 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5011 ret = btrfs_decompress(compress_type, tmp, page,
5012 extent_offset, inline_size, max_size);
5014 char *kaddr = kmap_atomic(page, KM_USER0);
5015 unsigned long copy_size = min_t(u64,
5016 PAGE_CACHE_SIZE - pg_offset,
5017 max_size - extent_offset);
5018 memset(kaddr + pg_offset, 0, copy_size);
5019 kunmap_atomic(kaddr, KM_USER0);
5026 * a bit scary, this does extent mapping from logical file offset to the disk.
5027 * the ugly parts come from merging extents from the disk with the in-ram
5028 * representation. This gets more complex because of the data=ordered code,
5029 * where the in-ram extents might be locked pending data=ordered completion.
5031 * This also copies inline extents directly into the page.
5034 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5035 size_t pg_offset, u64 start, u64 len,
5041 u64 extent_start = 0;
5043 u64 objectid = inode->i_ino;
5045 struct btrfs_path *path = NULL;
5046 struct btrfs_root *root = BTRFS_I(inode)->root;
5047 struct btrfs_file_extent_item *item;
5048 struct extent_buffer *leaf;
5049 struct btrfs_key found_key;
5050 struct extent_map *em = NULL;
5051 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5052 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5053 struct btrfs_trans_handle *trans = NULL;
5057 read_lock(&em_tree->lock);
5058 em = lookup_extent_mapping(em_tree, start, len);
5060 em->bdev = root->fs_info->fs_devices->latest_bdev;
5061 read_unlock(&em_tree->lock);
5064 if (em->start > start || em->start + em->len <= start)
5065 free_extent_map(em);
5066 else if (em->block_start == EXTENT_MAP_INLINE && page)
5067 free_extent_map(em);
5071 em = alloc_extent_map(GFP_NOFS);
5076 em->bdev = root->fs_info->fs_devices->latest_bdev;
5077 em->start = EXTENT_MAP_HOLE;
5078 em->orig_start = EXTENT_MAP_HOLE;
5080 em->block_len = (u64)-1;
5083 path = btrfs_alloc_path();
5087 ret = btrfs_lookup_file_extent(trans, root, path,
5088 objectid, start, trans != NULL);
5095 if (path->slots[0] == 0)
5100 leaf = path->nodes[0];
5101 item = btrfs_item_ptr(leaf, path->slots[0],
5102 struct btrfs_file_extent_item);
5103 /* are we inside the extent that was found? */
5104 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5105 found_type = btrfs_key_type(&found_key);
5106 if (found_key.objectid != objectid ||
5107 found_type != BTRFS_EXTENT_DATA_KEY) {
5111 found_type = btrfs_file_extent_type(leaf, item);
5112 extent_start = found_key.offset;
5113 compress_type = btrfs_file_extent_compression(leaf, item);
5114 if (found_type == BTRFS_FILE_EXTENT_REG ||
5115 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5116 extent_end = extent_start +
5117 btrfs_file_extent_num_bytes(leaf, item);
5118 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5120 size = btrfs_file_extent_inline_len(leaf, item);
5121 extent_end = (extent_start + size + root->sectorsize - 1) &
5122 ~((u64)root->sectorsize - 1);
5125 if (start >= extent_end) {
5127 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5128 ret = btrfs_next_leaf(root, path);
5135 leaf = path->nodes[0];
5137 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5138 if (found_key.objectid != objectid ||
5139 found_key.type != BTRFS_EXTENT_DATA_KEY)
5141 if (start + len <= found_key.offset)
5144 em->len = found_key.offset - start;
5148 if (found_type == BTRFS_FILE_EXTENT_REG ||
5149 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5150 em->start = extent_start;
5151 em->len = extent_end - extent_start;
5152 em->orig_start = extent_start -
5153 btrfs_file_extent_offset(leaf, item);
5154 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5156 em->block_start = EXTENT_MAP_HOLE;
5159 if (compress_type != BTRFS_COMPRESS_NONE) {
5160 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5161 em->compress_type = compress_type;
5162 em->block_start = bytenr;
5163 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5166 bytenr += btrfs_file_extent_offset(leaf, item);
5167 em->block_start = bytenr;
5168 em->block_len = em->len;
5169 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5170 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5173 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5177 size_t extent_offset;
5180 em->block_start = EXTENT_MAP_INLINE;
5181 if (!page || create) {
5182 em->start = extent_start;
5183 em->len = extent_end - extent_start;
5187 size = btrfs_file_extent_inline_len(leaf, item);
5188 extent_offset = page_offset(page) + pg_offset - extent_start;
5189 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5190 size - extent_offset);
5191 em->start = extent_start + extent_offset;
5192 em->len = (copy_size + root->sectorsize - 1) &
5193 ~((u64)root->sectorsize - 1);
5194 em->orig_start = EXTENT_MAP_INLINE;
5195 if (compress_type) {
5196 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5197 em->compress_type = compress_type;
5199 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5200 if (create == 0 && !PageUptodate(page)) {
5201 if (btrfs_file_extent_compression(leaf, item) !=
5202 BTRFS_COMPRESS_NONE) {
5203 ret = uncompress_inline(path, inode, page,
5205 extent_offset, item);
5209 read_extent_buffer(leaf, map + pg_offset, ptr,
5211 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5212 memset(map + pg_offset + copy_size, 0,
5213 PAGE_CACHE_SIZE - pg_offset -
5218 flush_dcache_page(page);
5219 } else if (create && PageUptodate(page)) {
5223 free_extent_map(em);
5225 btrfs_release_path(root, path);
5226 trans = btrfs_join_transaction(root, 1);
5228 return ERR_CAST(trans);
5232 write_extent_buffer(leaf, map + pg_offset, ptr,
5235 btrfs_mark_buffer_dirty(leaf);
5237 set_extent_uptodate(io_tree, em->start,
5238 extent_map_end(em) - 1, GFP_NOFS);
5241 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5248 em->block_start = EXTENT_MAP_HOLE;
5249 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5251 btrfs_release_path(root, path);
5252 if (em->start > start || extent_map_end(em) <= start) {
5253 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5254 "[%llu %llu]\n", (unsigned long long)em->start,
5255 (unsigned long long)em->len,
5256 (unsigned long long)start,
5257 (unsigned long long)len);
5263 write_lock(&em_tree->lock);
5264 ret = add_extent_mapping(em_tree, em);
5265 /* it is possible that someone inserted the extent into the tree
5266 * while we had the lock dropped. It is also possible that
5267 * an overlapping map exists in the tree
5269 if (ret == -EEXIST) {
5270 struct extent_map *existing;
5274 existing = lookup_extent_mapping(em_tree, start, len);
5275 if (existing && (existing->start > start ||
5276 existing->start + existing->len <= start)) {
5277 free_extent_map(existing);
5281 existing = lookup_extent_mapping(em_tree, em->start,
5284 err = merge_extent_mapping(em_tree, existing,
5287 free_extent_map(existing);
5289 free_extent_map(em);
5294 free_extent_map(em);
5298 free_extent_map(em);
5303 write_unlock(&em_tree->lock);
5306 trace_btrfs_get_extent(root, em);
5309 btrfs_free_path(path);
5311 ret = btrfs_end_transaction(trans, root);
5316 free_extent_map(em);
5317 return ERR_PTR(err);
5322 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5323 size_t pg_offset, u64 start, u64 len,
5326 struct extent_map *em;
5327 struct extent_map *hole_em = NULL;
5328 u64 range_start = start;
5334 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5339 * if our em maps to a hole, there might
5340 * actually be delalloc bytes behind it
5342 if (em->block_start != EXTENT_MAP_HOLE)
5348 /* check to see if we've wrapped (len == -1 or similar) */
5357 /* ok, we didn't find anything, lets look for delalloc */
5358 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5359 end, len, EXTENT_DELALLOC, 1);
5360 found_end = range_start + found;
5361 if (found_end < range_start)
5362 found_end = (u64)-1;
5365 * we didn't find anything useful, return
5366 * the original results from get_extent()
5368 if (range_start > end || found_end <= start) {
5374 /* adjust the range_start to make sure it doesn't
5375 * go backwards from the start they passed in
5377 range_start = max(start,range_start);
5378 found = found_end - range_start;
5381 u64 hole_start = start;
5384 em = alloc_extent_map(GFP_NOFS);
5390 * when btrfs_get_extent can't find anything it
5391 * returns one huge hole
5393 * make sure what it found really fits our range, and
5394 * adjust to make sure it is based on the start from
5398 u64 calc_end = extent_map_end(hole_em);
5400 if (calc_end <= start || (hole_em->start > end)) {
5401 free_extent_map(hole_em);
5404 hole_start = max(hole_em->start, start);
5405 hole_len = calc_end - hole_start;
5409 if (hole_em && range_start > hole_start) {
5410 /* our hole starts before our delalloc, so we
5411 * have to return just the parts of the hole
5412 * that go until the delalloc starts
5414 em->len = min(hole_len,
5415 range_start - hole_start);
5416 em->start = hole_start;
5417 em->orig_start = hole_start;
5419 * don't adjust block start at all,
5420 * it is fixed at EXTENT_MAP_HOLE
5422 em->block_start = hole_em->block_start;
5423 em->block_len = hole_len;
5425 em->start = range_start;
5427 em->orig_start = range_start;
5428 em->block_start = EXTENT_MAP_DELALLOC;
5429 em->block_len = found;
5431 } else if (hole_em) {
5436 free_extent_map(hole_em);
5438 free_extent_map(em);
5439 return ERR_PTR(err);
5444 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5447 struct btrfs_root *root = BTRFS_I(inode)->root;
5448 struct btrfs_trans_handle *trans;
5449 struct extent_map *em;
5450 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5451 struct btrfs_key ins;
5455 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5457 trans = btrfs_join_transaction(root, 0);
5459 return ERR_CAST(trans);
5461 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5463 alloc_hint = get_extent_allocation_hint(inode, start, len);
5464 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5465 alloc_hint, (u64)-1, &ins, 1);
5471 em = alloc_extent_map(GFP_NOFS);
5473 em = ERR_PTR(-ENOMEM);
5478 em->orig_start = em->start;
5479 em->len = ins.offset;
5481 em->block_start = ins.objectid;
5482 em->block_len = ins.offset;
5483 em->bdev = root->fs_info->fs_devices->latest_bdev;
5484 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5487 write_lock(&em_tree->lock);
5488 ret = add_extent_mapping(em_tree, em);
5489 write_unlock(&em_tree->lock);
5492 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5495 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5496 ins.offset, ins.offset, 0);
5498 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5502 btrfs_end_transaction(trans, root);
5507 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5508 * block must be cow'd
5510 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5511 struct inode *inode, u64 offset, u64 len)
5513 struct btrfs_path *path;
5515 struct extent_buffer *leaf;
5516 struct btrfs_root *root = BTRFS_I(inode)->root;
5517 struct btrfs_file_extent_item *fi;
5518 struct btrfs_key key;
5526 path = btrfs_alloc_path();
5530 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5535 slot = path->slots[0];
5538 /* can't find the item, must cow */
5545 leaf = path->nodes[0];
5546 btrfs_item_key_to_cpu(leaf, &key, slot);
5547 if (key.objectid != inode->i_ino ||
5548 key.type != BTRFS_EXTENT_DATA_KEY) {
5549 /* not our file or wrong item type, must cow */
5553 if (key.offset > offset) {
5554 /* Wrong offset, must cow */
5558 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5559 found_type = btrfs_file_extent_type(leaf, fi);
5560 if (found_type != BTRFS_FILE_EXTENT_REG &&
5561 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5562 /* not a regular extent, must cow */
5565 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5566 backref_offset = btrfs_file_extent_offset(leaf, fi);
5568 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5569 if (extent_end < offset + len) {
5570 /* extent doesn't include our full range, must cow */
5574 if (btrfs_extent_readonly(root, disk_bytenr))
5578 * look for other files referencing this extent, if we
5579 * find any we must cow
5581 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5582 key.offset - backref_offset, disk_bytenr))
5586 * adjust disk_bytenr and num_bytes to cover just the bytes
5587 * in this extent we are about to write. If there
5588 * are any csums in that range we have to cow in order
5589 * to keep the csums correct
5591 disk_bytenr += backref_offset;
5592 disk_bytenr += offset - key.offset;
5593 num_bytes = min(offset + len, extent_end) - offset;
5594 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5597 * all of the above have passed, it is safe to overwrite this extent
5602 btrfs_free_path(path);
5606 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5607 struct buffer_head *bh_result, int create)
5609 struct extent_map *em;
5610 struct btrfs_root *root = BTRFS_I(inode)->root;
5611 u64 start = iblock << inode->i_blkbits;
5612 u64 len = bh_result->b_size;
5613 struct btrfs_trans_handle *trans;
5615 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5620 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5621 * io. INLINE is special, and we could probably kludge it in here, but
5622 * it's still buffered so for safety lets just fall back to the generic
5625 * For COMPRESSED we _have_ to read the entire extent in so we can
5626 * decompress it, so there will be buffering required no matter what we
5627 * do, so go ahead and fallback to buffered.
5629 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5630 * to buffered IO. Don't blame me, this is the price we pay for using
5633 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5634 em->block_start == EXTENT_MAP_INLINE) {
5635 free_extent_map(em);
5639 /* Just a good old fashioned hole, return */
5640 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5641 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5642 free_extent_map(em);
5643 /* DIO will do one hole at a time, so just unlock a sector */
5644 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5645 start + root->sectorsize - 1, GFP_NOFS);
5650 * We don't allocate a new extent in the following cases
5652 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5654 * 2) The extent is marked as PREALLOC. We're good to go here and can
5655 * just use the extent.
5659 len = em->len - (start - em->start);
5663 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5664 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5665 em->block_start != EXTENT_MAP_HOLE)) {
5670 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5671 type = BTRFS_ORDERED_PREALLOC;
5673 type = BTRFS_ORDERED_NOCOW;
5674 len = min(len, em->len - (start - em->start));
5675 block_start = em->block_start + (start - em->start);
5678 * we're not going to log anything, but we do need
5679 * to make sure the current transaction stays open
5680 * while we look for nocow cross refs
5682 trans = btrfs_join_transaction(root, 0);
5686 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5687 ret = btrfs_add_ordered_extent_dio(inode, start,
5688 block_start, len, len, type);
5689 btrfs_end_transaction(trans, root);
5691 free_extent_map(em);
5696 btrfs_end_transaction(trans, root);
5700 * this will cow the extent, reset the len in case we changed
5703 len = bh_result->b_size;
5704 free_extent_map(em);
5705 em = btrfs_new_extent_direct(inode, start, len);
5708 len = min(len, em->len - (start - em->start));
5710 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5711 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5714 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5716 bh_result->b_size = len;
5717 bh_result->b_bdev = em->bdev;
5718 set_buffer_mapped(bh_result);
5719 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5720 set_buffer_new(bh_result);
5722 free_extent_map(em);
5727 struct btrfs_dio_private {
5728 struct inode *inode;
5735 /* number of bios pending for this dio */
5736 atomic_t pending_bios;
5741 struct bio *orig_bio;
5744 static void btrfs_endio_direct_read(struct bio *bio, int err)
5746 struct btrfs_dio_private *dip = bio->bi_private;
5747 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5748 struct bio_vec *bvec = bio->bi_io_vec;
5749 struct inode *inode = dip->inode;
5750 struct btrfs_root *root = BTRFS_I(inode)->root;
5752 u32 *private = dip->csums;
5754 start = dip->logical_offset;
5756 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5757 struct page *page = bvec->bv_page;
5760 unsigned long flags;
5762 local_irq_save(flags);
5763 kaddr = kmap_atomic(page, KM_IRQ0);
5764 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5765 csum, bvec->bv_len);
5766 btrfs_csum_final(csum, (char *)&csum);
5767 kunmap_atomic(kaddr, KM_IRQ0);
5768 local_irq_restore(flags);
5770 flush_dcache_page(bvec->bv_page);
5771 if (csum != *private) {
5772 printk(KERN_ERR "btrfs csum failed ino %lu off"
5773 " %llu csum %u private %u\n",
5774 inode->i_ino, (unsigned long long)start,
5780 start += bvec->bv_len;
5783 } while (bvec <= bvec_end);
5785 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5786 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5787 bio->bi_private = dip->private;
5792 /* If we had a csum failure make sure to clear the uptodate flag */
5794 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5795 dio_end_io(bio, err);
5798 static void btrfs_endio_direct_write(struct bio *bio, int err)
5800 struct btrfs_dio_private *dip = bio->bi_private;
5801 struct inode *inode = dip->inode;
5802 struct btrfs_root *root = BTRFS_I(inode)->root;
5803 struct btrfs_trans_handle *trans;
5804 struct btrfs_ordered_extent *ordered = NULL;
5805 struct extent_state *cached_state = NULL;
5806 u64 ordered_offset = dip->logical_offset;
5807 u64 ordered_bytes = dip->bytes;
5813 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5821 trans = btrfs_join_transaction(root, 1);
5822 if (IS_ERR(trans)) {
5826 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5828 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5829 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5831 ret = btrfs_update_inode(trans, root, inode);
5836 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5837 ordered->file_offset + ordered->len - 1, 0,
5838 &cached_state, GFP_NOFS);
5840 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5841 ret = btrfs_mark_extent_written(trans, inode,
5842 ordered->file_offset,
5843 ordered->file_offset +
5850 ret = insert_reserved_file_extent(trans, inode,
5851 ordered->file_offset,
5857 BTRFS_FILE_EXTENT_REG);
5858 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5859 ordered->file_offset, ordered->len);
5867 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5868 btrfs_ordered_update_i_size(inode, 0, ordered);
5869 btrfs_update_inode(trans, root, inode);
5871 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5872 ordered->file_offset + ordered->len - 1,
5873 &cached_state, GFP_NOFS);
5875 btrfs_delalloc_release_metadata(inode, ordered->len);
5876 btrfs_end_transaction(trans, root);
5877 ordered_offset = ordered->file_offset + ordered->len;
5878 btrfs_put_ordered_extent(ordered);
5879 btrfs_put_ordered_extent(ordered);
5883 * our bio might span multiple ordered extents. If we haven't
5884 * completed the accounting for the whole dio, go back and try again
5886 if (ordered_offset < dip->logical_offset + dip->bytes) {
5887 ordered_bytes = dip->logical_offset + dip->bytes -
5892 bio->bi_private = dip->private;
5897 /* If we had an error make sure to clear the uptodate flag */
5899 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5900 dio_end_io(bio, err);
5903 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5904 struct bio *bio, int mirror_num,
5905 unsigned long bio_flags, u64 offset)
5908 struct btrfs_root *root = BTRFS_I(inode)->root;
5909 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5914 static void btrfs_end_dio_bio(struct bio *bio, int err)
5916 struct btrfs_dio_private *dip = bio->bi_private;
5919 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5920 "sector %#Lx len %u err no %d\n",
5921 dip->inode->i_ino, bio->bi_rw,
5922 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5926 * before atomic variable goto zero, we must make sure
5927 * dip->errors is perceived to be set.
5929 smp_mb__before_atomic_dec();
5932 /* if there are more bios still pending for this dio, just exit */
5933 if (!atomic_dec_and_test(&dip->pending_bios))
5937 bio_io_error(dip->orig_bio);
5939 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5940 bio_endio(dip->orig_bio, 0);
5946 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5947 u64 first_sector, gfp_t gfp_flags)
5949 int nr_vecs = bio_get_nr_vecs(bdev);
5950 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5953 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5954 int rw, u64 file_offset, int skip_sum,
5957 int write = rw & REQ_WRITE;
5958 struct btrfs_root *root = BTRFS_I(inode)->root;
5962 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5966 if (write && !skip_sum) {
5967 ret = btrfs_wq_submit_bio(root->fs_info,
5968 inode, rw, bio, 0, 0,
5970 __btrfs_submit_bio_start_direct_io,
5971 __btrfs_submit_bio_done);
5973 } else if (!skip_sum) {
5974 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5975 file_offset, csums);
5980 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5986 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5989 struct inode *inode = dip->inode;
5990 struct btrfs_root *root = BTRFS_I(inode)->root;
5991 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5993 struct bio *orig_bio = dip->orig_bio;
5994 struct bio_vec *bvec = orig_bio->bi_io_vec;
5995 u64 start_sector = orig_bio->bi_sector;
5996 u64 file_offset = dip->logical_offset;
6000 u32 *csums = dip->csums;
6002 int write = rw & REQ_WRITE;
6004 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6007 bio->bi_private = dip;
6008 bio->bi_end_io = btrfs_end_dio_bio;
6009 atomic_inc(&dip->pending_bios);
6011 map_length = orig_bio->bi_size;
6012 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6013 &map_length, NULL, 0);
6019 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6020 if (unlikely(map_length < submit_len + bvec->bv_len ||
6021 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6022 bvec->bv_offset) < bvec->bv_len)) {
6024 * inc the count before we submit the bio so
6025 * we know the end IO handler won't happen before
6026 * we inc the count. Otherwise, the dip might get freed
6027 * before we're done setting it up
6029 atomic_inc(&dip->pending_bios);
6030 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6031 file_offset, skip_sum,
6035 atomic_dec(&dip->pending_bios);
6039 /* Write's use the ordered csums */
6040 if (!write && !skip_sum)
6041 csums = csums + nr_pages;
6042 start_sector += submit_len >> 9;
6043 file_offset += submit_len;
6048 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6049 start_sector, GFP_NOFS);
6052 bio->bi_private = dip;
6053 bio->bi_end_io = btrfs_end_dio_bio;
6055 map_length = orig_bio->bi_size;
6056 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6057 &map_length, NULL, 0);
6063 submit_len += bvec->bv_len;
6069 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6078 * before atomic variable goto zero, we must
6079 * make sure dip->errors is perceived to be set.
6081 smp_mb__before_atomic_dec();
6082 if (atomic_dec_and_test(&dip->pending_bios))
6083 bio_io_error(dip->orig_bio);
6085 /* bio_end_io() will handle error, so we needn't return it */
6089 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6092 struct btrfs_root *root = BTRFS_I(inode)->root;
6093 struct btrfs_dio_private *dip;
6094 struct bio_vec *bvec = bio->bi_io_vec;
6096 int write = rw & REQ_WRITE;
6099 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6101 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6108 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6109 if (!write && !skip_sum) {
6110 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6118 dip->private = bio->bi_private;
6120 dip->logical_offset = file_offset;
6124 dip->bytes += bvec->bv_len;
6126 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6128 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6129 bio->bi_private = dip;
6131 dip->orig_bio = bio;
6132 atomic_set(&dip->pending_bios, 0);
6135 bio->bi_end_io = btrfs_endio_direct_write;
6137 bio->bi_end_io = btrfs_endio_direct_read;
6139 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6144 * If this is a write, we need to clean up the reserved space and kill
6145 * the ordered extent.
6148 struct btrfs_ordered_extent *ordered;
6149 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6150 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6151 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6152 btrfs_free_reserved_extent(root, ordered->start,
6154 btrfs_put_ordered_extent(ordered);
6155 btrfs_put_ordered_extent(ordered);
6157 bio_endio(bio, ret);
6160 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6161 const struct iovec *iov, loff_t offset,
6162 unsigned long nr_segs)
6167 unsigned blocksize_mask = root->sectorsize - 1;
6168 ssize_t retval = -EINVAL;
6169 loff_t end = offset;
6171 if (offset & blocksize_mask)
6174 /* Check the memory alignment. Blocks cannot straddle pages */
6175 for (seg = 0; seg < nr_segs; seg++) {
6176 addr = (unsigned long)iov[seg].iov_base;
6177 size = iov[seg].iov_len;
6179 if ((addr & blocksize_mask) || (size & blocksize_mask))
6186 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6187 const struct iovec *iov, loff_t offset,
6188 unsigned long nr_segs)
6190 struct file *file = iocb->ki_filp;
6191 struct inode *inode = file->f_mapping->host;
6192 struct btrfs_ordered_extent *ordered;
6193 struct extent_state *cached_state = NULL;
6194 u64 lockstart, lockend;
6196 int writing = rw & WRITE;
6198 size_t count = iov_length(iov, nr_segs);
6200 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6206 lockend = offset + count - 1;
6209 ret = btrfs_delalloc_reserve_space(inode, count);
6215 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6216 0, &cached_state, GFP_NOFS);
6218 * We're concerned with the entire range that we're going to be
6219 * doing DIO to, so we need to make sure theres no ordered
6220 * extents in this range.
6222 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6223 lockend - lockstart + 1);
6226 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6227 &cached_state, GFP_NOFS);
6228 btrfs_start_ordered_extent(inode, ordered, 1);
6229 btrfs_put_ordered_extent(ordered);
6234 * we don't use btrfs_set_extent_delalloc because we don't want
6235 * the dirty or uptodate bits
6238 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6239 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6240 EXTENT_DELALLOC, 0, NULL, &cached_state,
6243 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6244 lockend, EXTENT_LOCKED | write_bits,
6245 1, 0, &cached_state, GFP_NOFS);
6250 free_extent_state(cached_state);
6251 cached_state = NULL;
6253 ret = __blockdev_direct_IO(rw, iocb, inode,
6254 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6255 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6256 btrfs_submit_direct, 0);
6258 if (ret < 0 && ret != -EIOCBQUEUED) {
6259 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6260 offset + iov_length(iov, nr_segs) - 1,
6261 EXTENT_LOCKED | write_bits, 1, 0,
6262 &cached_state, GFP_NOFS);
6263 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6265 * We're falling back to buffered, unlock the section we didn't
6268 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6269 offset + iov_length(iov, nr_segs) - 1,
6270 EXTENT_LOCKED | write_bits, 1, 0,
6271 &cached_state, GFP_NOFS);
6274 free_extent_state(cached_state);
6278 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6279 __u64 start, __u64 len)
6281 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6284 int btrfs_readpage(struct file *file, struct page *page)
6286 struct extent_io_tree *tree;
6287 tree = &BTRFS_I(page->mapping->host)->io_tree;
6288 return extent_read_full_page(tree, page, btrfs_get_extent);
6291 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6293 struct extent_io_tree *tree;
6296 if (current->flags & PF_MEMALLOC) {
6297 redirty_page_for_writepage(wbc, page);
6301 tree = &BTRFS_I(page->mapping->host)->io_tree;
6302 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6305 int btrfs_writepages(struct address_space *mapping,
6306 struct writeback_control *wbc)
6308 struct extent_io_tree *tree;
6310 tree = &BTRFS_I(mapping->host)->io_tree;
6311 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6315 btrfs_readpages(struct file *file, struct address_space *mapping,
6316 struct list_head *pages, unsigned nr_pages)
6318 struct extent_io_tree *tree;
6319 tree = &BTRFS_I(mapping->host)->io_tree;
6320 return extent_readpages(tree, mapping, pages, nr_pages,
6323 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6325 struct extent_io_tree *tree;
6326 struct extent_map_tree *map;
6329 tree = &BTRFS_I(page->mapping->host)->io_tree;
6330 map = &BTRFS_I(page->mapping->host)->extent_tree;
6331 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6333 ClearPagePrivate(page);
6334 set_page_private(page, 0);
6335 page_cache_release(page);
6340 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6342 if (PageWriteback(page) || PageDirty(page))
6344 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6347 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6349 struct extent_io_tree *tree;
6350 struct btrfs_ordered_extent *ordered;
6351 struct extent_state *cached_state = NULL;
6352 u64 page_start = page_offset(page);
6353 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6357 * we have the page locked, so new writeback can't start,
6358 * and the dirty bit won't be cleared while we are here.
6360 * Wait for IO on this page so that we can safely clear
6361 * the PagePrivate2 bit and do ordered accounting
6363 wait_on_page_writeback(page);
6365 tree = &BTRFS_I(page->mapping->host)->io_tree;
6367 btrfs_releasepage(page, GFP_NOFS);
6370 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6372 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6376 * IO on this page will never be started, so we need
6377 * to account for any ordered extents now
6379 clear_extent_bit(tree, page_start, page_end,
6380 EXTENT_DIRTY | EXTENT_DELALLOC |
6381 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6382 &cached_state, GFP_NOFS);
6384 * whoever cleared the private bit is responsible
6385 * for the finish_ordered_io
6387 if (TestClearPagePrivate2(page)) {
6388 btrfs_finish_ordered_io(page->mapping->host,
6389 page_start, page_end);
6391 btrfs_put_ordered_extent(ordered);
6392 cached_state = NULL;
6393 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6396 clear_extent_bit(tree, page_start, page_end,
6397 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6398 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6399 __btrfs_releasepage(page, GFP_NOFS);
6401 ClearPageChecked(page);
6402 if (PagePrivate(page)) {
6403 ClearPagePrivate(page);
6404 set_page_private(page, 0);
6405 page_cache_release(page);
6410 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6411 * called from a page fault handler when a page is first dirtied. Hence we must
6412 * be careful to check for EOF conditions here. We set the page up correctly
6413 * for a written page which means we get ENOSPC checking when writing into
6414 * holes and correct delalloc and unwritten extent mapping on filesystems that
6415 * support these features.
6417 * We are not allowed to take the i_mutex here so we have to play games to
6418 * protect against truncate races as the page could now be beyond EOF. Because
6419 * vmtruncate() writes the inode size before removing pages, once we have the
6420 * page lock we can determine safely if the page is beyond EOF. If it is not
6421 * beyond EOF, then the page is guaranteed safe against truncation until we
6424 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6426 struct page *page = vmf->page;
6427 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6428 struct btrfs_root *root = BTRFS_I(inode)->root;
6429 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6430 struct btrfs_ordered_extent *ordered;
6431 struct extent_state *cached_state = NULL;
6433 unsigned long zero_start;
6439 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6443 else /* -ENOSPC, -EIO, etc */
6444 ret = VM_FAULT_SIGBUS;
6448 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6451 size = i_size_read(inode);
6452 page_start = page_offset(page);
6453 page_end = page_start + PAGE_CACHE_SIZE - 1;
6455 if ((page->mapping != inode->i_mapping) ||
6456 (page_start >= size)) {
6457 /* page got truncated out from underneath us */
6460 wait_on_page_writeback(page);
6462 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6464 set_page_extent_mapped(page);
6467 * we can't set the delalloc bits if there are pending ordered
6468 * extents. Drop our locks and wait for them to finish
6470 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6472 unlock_extent_cached(io_tree, page_start, page_end,
6473 &cached_state, GFP_NOFS);
6475 btrfs_start_ordered_extent(inode, ordered, 1);
6476 btrfs_put_ordered_extent(ordered);
6481 * XXX - page_mkwrite gets called every time the page is dirtied, even
6482 * if it was already dirty, so for space accounting reasons we need to
6483 * clear any delalloc bits for the range we are fixing to save. There
6484 * is probably a better way to do this, but for now keep consistent with
6485 * prepare_pages in the normal write path.
6487 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6488 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6489 0, 0, &cached_state, GFP_NOFS);
6491 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6494 unlock_extent_cached(io_tree, page_start, page_end,
6495 &cached_state, GFP_NOFS);
6496 ret = VM_FAULT_SIGBUS;
6501 /* page is wholly or partially inside EOF */
6502 if (page_start + PAGE_CACHE_SIZE > size)
6503 zero_start = size & ~PAGE_CACHE_MASK;
6505 zero_start = PAGE_CACHE_SIZE;
6507 if (zero_start != PAGE_CACHE_SIZE) {
6509 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6510 flush_dcache_page(page);
6513 ClearPageChecked(page);
6514 set_page_dirty(page);
6515 SetPageUptodate(page);
6517 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6518 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6520 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6524 return VM_FAULT_LOCKED;
6526 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6531 static int btrfs_truncate(struct inode *inode)
6533 struct btrfs_root *root = BTRFS_I(inode)->root;
6536 struct btrfs_trans_handle *trans;
6538 u64 mask = root->sectorsize - 1;
6540 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6544 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6545 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6547 trans = btrfs_start_transaction(root, 5);
6549 return PTR_ERR(trans);
6551 btrfs_set_trans_block_group(trans, inode);
6553 ret = btrfs_orphan_add(trans, inode);
6555 btrfs_end_transaction(trans, root);
6559 nr = trans->blocks_used;
6560 btrfs_end_transaction(trans, root);
6561 btrfs_btree_balance_dirty(root, nr);
6563 /* Now start a transaction for the truncate */
6564 trans = btrfs_start_transaction(root, 0);
6566 return PTR_ERR(trans);
6567 btrfs_set_trans_block_group(trans, inode);
6568 trans->block_rsv = root->orphan_block_rsv;
6571 * setattr is responsible for setting the ordered_data_close flag,
6572 * but that is only tested during the last file release. That
6573 * could happen well after the next commit, leaving a great big
6574 * window where new writes may get lost if someone chooses to write
6575 * to this file after truncating to zero
6577 * The inode doesn't have any dirty data here, and so if we commit
6578 * this is a noop. If someone immediately starts writing to the inode
6579 * it is very likely we'll catch some of their writes in this
6580 * transaction, and the commit will find this file on the ordered
6581 * data list with good things to send down.
6583 * This is a best effort solution, there is still a window where
6584 * using truncate to replace the contents of the file will
6585 * end up with a zero length file after a crash.
6587 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6588 btrfs_add_ordered_operation(trans, root, inode);
6592 trans = btrfs_start_transaction(root, 0);
6594 return PTR_ERR(trans);
6595 btrfs_set_trans_block_group(trans, inode);
6596 trans->block_rsv = root->orphan_block_rsv;
6599 ret = btrfs_block_rsv_check(trans, root,
6600 root->orphan_block_rsv, 0, 5);
6601 if (ret == -EAGAIN) {
6602 ret = btrfs_commit_transaction(trans, root);
6612 ret = btrfs_truncate_inode_items(trans, root, inode,
6614 BTRFS_EXTENT_DATA_KEY);
6615 if (ret != -EAGAIN) {
6620 ret = btrfs_update_inode(trans, root, inode);
6626 nr = trans->blocks_used;
6627 btrfs_end_transaction(trans, root);
6629 btrfs_btree_balance_dirty(root, nr);
6632 if (ret == 0 && inode->i_nlink > 0) {
6633 ret = btrfs_orphan_del(trans, inode);
6636 } else if (ret && inode->i_nlink > 0) {
6638 * Failed to do the truncate, remove us from the in memory
6641 ret = btrfs_orphan_del(NULL, inode);
6644 ret = btrfs_update_inode(trans, root, inode);
6648 nr = trans->blocks_used;
6649 ret = btrfs_end_transaction_throttle(trans, root);
6652 btrfs_btree_balance_dirty(root, nr);
6658 * create a new subvolume directory/inode (helper for the ioctl).
6660 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6661 struct btrfs_root *new_root,
6662 u64 new_dirid, u64 alloc_hint)
6664 struct inode *inode;
6668 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6669 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6671 return PTR_ERR(inode);
6672 inode->i_op = &btrfs_dir_inode_operations;
6673 inode->i_fop = &btrfs_dir_file_operations;
6676 btrfs_i_size_write(inode, 0);
6678 err = btrfs_update_inode(trans, new_root, inode);
6685 /* helper function for file defrag and space balancing. This
6686 * forces readahead on a given range of bytes in an inode
6688 unsigned long btrfs_force_ra(struct address_space *mapping,
6689 struct file_ra_state *ra, struct file *file,
6690 pgoff_t offset, pgoff_t last_index)
6692 pgoff_t req_size = last_index - offset + 1;
6694 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6695 return offset + req_size;
6698 struct inode *btrfs_alloc_inode(struct super_block *sb)
6700 struct btrfs_inode *ei;
6701 struct inode *inode;
6703 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6708 ei->space_info = NULL;
6712 ei->last_sub_trans = 0;
6713 ei->logged_trans = 0;
6714 ei->delalloc_bytes = 0;
6715 ei->reserved_bytes = 0;
6716 ei->disk_i_size = 0;
6718 ei->index_cnt = (u64)-1;
6719 ei->last_unlink_trans = 0;
6721 atomic_set(&ei->outstanding_extents, 0);
6722 atomic_set(&ei->reserved_extents, 0);
6724 ei->ordered_data_close = 0;
6725 ei->orphan_meta_reserved = 0;
6726 ei->dummy_inode = 0;
6727 ei->force_compress = BTRFS_COMPRESS_NONE;
6729 inode = &ei->vfs_inode;
6730 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6731 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6732 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6733 mutex_init(&ei->log_mutex);
6734 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6735 INIT_LIST_HEAD(&ei->i_orphan);
6736 INIT_LIST_HEAD(&ei->delalloc_inodes);
6737 INIT_LIST_HEAD(&ei->ordered_operations);
6738 RB_CLEAR_NODE(&ei->rb_node);
6743 static void btrfs_i_callback(struct rcu_head *head)
6745 struct inode *inode = container_of(head, struct inode, i_rcu);
6746 INIT_LIST_HEAD(&inode->i_dentry);
6747 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6750 void btrfs_destroy_inode(struct inode *inode)
6752 struct btrfs_ordered_extent *ordered;
6753 struct btrfs_root *root = BTRFS_I(inode)->root;
6755 WARN_ON(!list_empty(&inode->i_dentry));
6756 WARN_ON(inode->i_data.nrpages);
6757 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6758 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6761 * This can happen where we create an inode, but somebody else also
6762 * created the same inode and we need to destroy the one we already
6769 * Make sure we're properly removed from the ordered operation
6773 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6774 spin_lock(&root->fs_info->ordered_extent_lock);
6775 list_del_init(&BTRFS_I(inode)->ordered_operations);
6776 spin_unlock(&root->fs_info->ordered_extent_lock);
6779 if (root == root->fs_info->tree_root) {
6780 struct btrfs_block_group_cache *block_group;
6782 block_group = btrfs_lookup_block_group(root->fs_info,
6783 BTRFS_I(inode)->block_group);
6784 if (block_group && block_group->inode == inode) {
6785 spin_lock(&block_group->lock);
6786 block_group->inode = NULL;
6787 spin_unlock(&block_group->lock);
6788 btrfs_put_block_group(block_group);
6789 } else if (block_group) {
6790 btrfs_put_block_group(block_group);
6794 spin_lock(&root->orphan_lock);
6795 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6796 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6798 list_del_init(&BTRFS_I(inode)->i_orphan);
6800 spin_unlock(&root->orphan_lock);
6803 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6807 printk(KERN_ERR "btrfs found ordered "
6808 "extent %llu %llu on inode cleanup\n",
6809 (unsigned long long)ordered->file_offset,
6810 (unsigned long long)ordered->len);
6811 btrfs_remove_ordered_extent(inode, ordered);
6812 btrfs_put_ordered_extent(ordered);
6813 btrfs_put_ordered_extent(ordered);
6816 inode_tree_del(inode);
6817 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6819 call_rcu(&inode->i_rcu, btrfs_i_callback);
6822 int btrfs_drop_inode(struct inode *inode)
6824 struct btrfs_root *root = BTRFS_I(inode)->root;
6826 if (btrfs_root_refs(&root->root_item) == 0 &&
6827 root != root->fs_info->tree_root)
6830 return generic_drop_inode(inode);
6833 static void init_once(void *foo)
6835 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6837 inode_init_once(&ei->vfs_inode);
6840 void btrfs_destroy_cachep(void)
6842 if (btrfs_inode_cachep)
6843 kmem_cache_destroy(btrfs_inode_cachep);
6844 if (btrfs_trans_handle_cachep)
6845 kmem_cache_destroy(btrfs_trans_handle_cachep);
6846 if (btrfs_transaction_cachep)
6847 kmem_cache_destroy(btrfs_transaction_cachep);
6848 if (btrfs_path_cachep)
6849 kmem_cache_destroy(btrfs_path_cachep);
6850 if (btrfs_free_space_cachep)
6851 kmem_cache_destroy(btrfs_free_space_cachep);
6854 int btrfs_init_cachep(void)
6856 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6857 sizeof(struct btrfs_inode), 0,
6858 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6859 if (!btrfs_inode_cachep)
6862 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6863 sizeof(struct btrfs_trans_handle), 0,
6864 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6865 if (!btrfs_trans_handle_cachep)
6868 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6869 sizeof(struct btrfs_transaction), 0,
6870 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6871 if (!btrfs_transaction_cachep)
6874 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6875 sizeof(struct btrfs_path), 0,
6876 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6877 if (!btrfs_path_cachep)
6880 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6881 sizeof(struct btrfs_free_space), 0,
6882 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6883 if (!btrfs_free_space_cachep)
6888 btrfs_destroy_cachep();
6892 static int btrfs_getattr(struct vfsmount *mnt,
6893 struct dentry *dentry, struct kstat *stat)
6895 struct inode *inode = dentry->d_inode;
6896 generic_fillattr(inode, stat);
6897 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6898 stat->blksize = PAGE_CACHE_SIZE;
6899 stat->blocks = (inode_get_bytes(inode) +
6900 BTRFS_I(inode)->delalloc_bytes) >> 9;
6905 * If a file is moved, it will inherit the cow and compression flags of the new
6908 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6910 struct btrfs_inode *b_dir = BTRFS_I(dir);
6911 struct btrfs_inode *b_inode = BTRFS_I(inode);
6913 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6914 b_inode->flags |= BTRFS_INODE_NODATACOW;
6916 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6918 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6919 b_inode->flags |= BTRFS_INODE_COMPRESS;
6921 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6924 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6925 struct inode *new_dir, struct dentry *new_dentry)
6927 struct btrfs_trans_handle *trans;
6928 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6929 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6930 struct inode *new_inode = new_dentry->d_inode;
6931 struct inode *old_inode = old_dentry->d_inode;
6932 struct timespec ctime = CURRENT_TIME;
6937 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6940 /* we only allow rename subvolume link between subvolumes */
6941 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6944 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6945 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6948 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6949 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6952 * we're using rename to replace one file with another.
6953 * and the replacement file is large. Start IO on it now so
6954 * we don't add too much work to the end of the transaction
6956 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6957 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6958 filemap_flush(old_inode->i_mapping);
6960 /* close the racy window with snapshot create/destroy ioctl */
6961 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6962 down_read(&root->fs_info->subvol_sem);
6964 * We want to reserve the absolute worst case amount of items. So if
6965 * both inodes are subvols and we need to unlink them then that would
6966 * require 4 item modifications, but if they are both normal inodes it
6967 * would require 5 item modifications, so we'll assume their normal
6968 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6969 * should cover the worst case number of items we'll modify.
6971 trans = btrfs_start_transaction(root, 20);
6972 if (IS_ERR(trans)) {
6973 ret = PTR_ERR(trans);
6977 btrfs_set_trans_block_group(trans, new_dir);
6980 btrfs_record_root_in_trans(trans, dest);
6982 ret = btrfs_set_inode_index(new_dir, &index);
6986 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6987 /* force full log commit if subvolume involved. */
6988 root->fs_info->last_trans_log_full_commit = trans->transid;
6990 ret = btrfs_insert_inode_ref(trans, dest,
6991 new_dentry->d_name.name,
6992 new_dentry->d_name.len,
6994 new_dir->i_ino, index);
6998 * this is an ugly little race, but the rename is required
6999 * to make sure that if we crash, the inode is either at the
7000 * old name or the new one. pinning the log transaction lets
7001 * us make sure we don't allow a log commit to come in after
7002 * we unlink the name but before we add the new name back in.
7004 btrfs_pin_log_trans(root);
7007 * make sure the inode gets flushed if it is replacing
7010 if (new_inode && new_inode->i_size &&
7011 old_inode && S_ISREG(old_inode->i_mode)) {
7012 btrfs_add_ordered_operation(trans, root, old_inode);
7015 old_dir->i_ctime = old_dir->i_mtime = ctime;
7016 new_dir->i_ctime = new_dir->i_mtime = ctime;
7017 old_inode->i_ctime = ctime;
7019 if (old_dentry->d_parent != new_dentry->d_parent)
7020 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7022 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7023 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7024 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7025 old_dentry->d_name.name,
7026 old_dentry->d_name.len);
7028 ret = __btrfs_unlink_inode(trans, root, old_dir,
7029 old_dentry->d_inode,
7030 old_dentry->d_name.name,
7031 old_dentry->d_name.len);
7033 ret = btrfs_update_inode(trans, root, old_inode);
7038 new_inode->i_ctime = CURRENT_TIME;
7039 if (unlikely(new_inode->i_ino ==
7040 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7041 root_objectid = BTRFS_I(new_inode)->location.objectid;
7042 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7044 new_dentry->d_name.name,
7045 new_dentry->d_name.len);
7046 BUG_ON(new_inode->i_nlink == 0);
7048 ret = btrfs_unlink_inode(trans, dest, new_dir,
7049 new_dentry->d_inode,
7050 new_dentry->d_name.name,
7051 new_dentry->d_name.len);
7054 if (new_inode->i_nlink == 0) {
7055 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7060 fixup_inode_flags(new_dir, old_inode);
7062 ret = btrfs_add_link(trans, new_dir, old_inode,
7063 new_dentry->d_name.name,
7064 new_dentry->d_name.len, 0, index);
7067 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7068 struct dentry *parent = dget_parent(new_dentry);
7069 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7071 btrfs_end_log_trans(root);
7074 btrfs_end_transaction_throttle(trans, root);
7076 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7077 up_read(&root->fs_info->subvol_sem);
7083 * some fairly slow code that needs optimization. This walks the list
7084 * of all the inodes with pending delalloc and forces them to disk.
7086 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7088 struct list_head *head = &root->fs_info->delalloc_inodes;
7089 struct btrfs_inode *binode;
7090 struct inode *inode;
7092 if (root->fs_info->sb->s_flags & MS_RDONLY)
7095 spin_lock(&root->fs_info->delalloc_lock);
7096 while (!list_empty(head)) {
7097 binode = list_entry(head->next, struct btrfs_inode,
7099 inode = igrab(&binode->vfs_inode);
7101 list_del_init(&binode->delalloc_inodes);
7102 spin_unlock(&root->fs_info->delalloc_lock);
7104 filemap_flush(inode->i_mapping);
7106 btrfs_add_delayed_iput(inode);
7111 spin_lock(&root->fs_info->delalloc_lock);
7113 spin_unlock(&root->fs_info->delalloc_lock);
7115 /* the filemap_flush will queue IO into the worker threads, but
7116 * we have to make sure the IO is actually started and that
7117 * ordered extents get created before we return
7119 atomic_inc(&root->fs_info->async_submit_draining);
7120 while (atomic_read(&root->fs_info->nr_async_submits) ||
7121 atomic_read(&root->fs_info->async_delalloc_pages)) {
7122 wait_event(root->fs_info->async_submit_wait,
7123 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7124 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7126 atomic_dec(&root->fs_info->async_submit_draining);
7130 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7133 struct btrfs_inode *binode;
7134 struct inode *inode = NULL;
7136 spin_lock(&root->fs_info->delalloc_lock);
7137 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7138 binode = list_entry(root->fs_info->delalloc_inodes.next,
7139 struct btrfs_inode, delalloc_inodes);
7140 inode = igrab(&binode->vfs_inode);
7142 list_move_tail(&binode->delalloc_inodes,
7143 &root->fs_info->delalloc_inodes);
7147 list_del_init(&binode->delalloc_inodes);
7148 cond_resched_lock(&root->fs_info->delalloc_lock);
7150 spin_unlock(&root->fs_info->delalloc_lock);
7154 filemap_write_and_wait(inode->i_mapping);
7156 * We have to do this because compression doesn't
7157 * actually set PG_writeback until it submits the pages
7158 * for IO, which happens in an async thread, so we could
7159 * race and not actually wait for any writeback pages
7160 * because they've not been submitted yet. Technically
7161 * this could still be the case for the ordered stuff
7162 * since the async thread may not have started to do its
7163 * work yet. If this becomes the case then we need to
7164 * figure out a way to make sure that in writepage we
7165 * wait for any async pages to be submitted before
7166 * returning so that fdatawait does what its supposed to
7169 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7171 filemap_flush(inode->i_mapping);
7174 btrfs_add_delayed_iput(inode);
7182 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7183 const char *symname)
7185 struct btrfs_trans_handle *trans;
7186 struct btrfs_root *root = BTRFS_I(dir)->root;
7187 struct btrfs_path *path;
7188 struct btrfs_key key;
7189 struct inode *inode = NULL;
7197 struct btrfs_file_extent_item *ei;
7198 struct extent_buffer *leaf;
7199 unsigned long nr = 0;
7201 name_len = strlen(symname) + 1;
7202 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7203 return -ENAMETOOLONG;
7205 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7209 * 2 items for inode item and ref
7210 * 2 items for dir items
7211 * 1 item for xattr if selinux is on
7213 trans = btrfs_start_transaction(root, 5);
7215 return PTR_ERR(trans);
7217 btrfs_set_trans_block_group(trans, dir);
7219 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7220 dentry->d_name.len, dir->i_ino, objectid,
7221 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7223 err = PTR_ERR(inode);
7227 err = btrfs_init_inode_security(trans, inode, dir);
7233 btrfs_set_trans_block_group(trans, inode);
7234 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7238 inode->i_mapping->a_ops = &btrfs_aops;
7239 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7240 inode->i_fop = &btrfs_file_operations;
7241 inode->i_op = &btrfs_file_inode_operations;
7242 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7244 btrfs_update_inode_block_group(trans, inode);
7245 btrfs_update_inode_block_group(trans, dir);
7249 path = btrfs_alloc_path();
7251 key.objectid = inode->i_ino;
7253 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7254 datasize = btrfs_file_extent_calc_inline_size(name_len);
7255 err = btrfs_insert_empty_item(trans, root, path, &key,
7261 leaf = path->nodes[0];
7262 ei = btrfs_item_ptr(leaf, path->slots[0],
7263 struct btrfs_file_extent_item);
7264 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7265 btrfs_set_file_extent_type(leaf, ei,
7266 BTRFS_FILE_EXTENT_INLINE);
7267 btrfs_set_file_extent_encryption(leaf, ei, 0);
7268 btrfs_set_file_extent_compression(leaf, ei, 0);
7269 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7270 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7272 ptr = btrfs_file_extent_inline_start(ei);
7273 write_extent_buffer(leaf, symname, ptr, name_len);
7274 btrfs_mark_buffer_dirty(leaf);
7275 btrfs_free_path(path);
7277 inode->i_op = &btrfs_symlink_inode_operations;
7278 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7279 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7280 inode_set_bytes(inode, name_len);
7281 btrfs_i_size_write(inode, name_len - 1);
7282 err = btrfs_update_inode(trans, root, inode);
7287 nr = trans->blocks_used;
7288 btrfs_end_transaction_throttle(trans, root);
7290 inode_dec_link_count(inode);
7293 btrfs_btree_balance_dirty(root, nr);
7297 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7298 u64 start, u64 num_bytes, u64 min_size,
7299 loff_t actual_len, u64 *alloc_hint,
7300 struct btrfs_trans_handle *trans)
7302 struct btrfs_root *root = BTRFS_I(inode)->root;
7303 struct btrfs_key ins;
7304 u64 cur_offset = start;
7307 bool own_trans = true;
7311 while (num_bytes > 0) {
7313 trans = btrfs_start_transaction(root, 3);
7314 if (IS_ERR(trans)) {
7315 ret = PTR_ERR(trans);
7320 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7321 0, *alloc_hint, (u64)-1, &ins, 1);
7324 btrfs_end_transaction(trans, root);
7328 ret = insert_reserved_file_extent(trans, inode,
7329 cur_offset, ins.objectid,
7330 ins.offset, ins.offset,
7331 ins.offset, 0, 0, 0,
7332 BTRFS_FILE_EXTENT_PREALLOC);
7334 btrfs_drop_extent_cache(inode, cur_offset,
7335 cur_offset + ins.offset -1, 0);
7337 num_bytes -= ins.offset;
7338 cur_offset += ins.offset;
7339 *alloc_hint = ins.objectid + ins.offset;
7341 inode->i_ctime = CURRENT_TIME;
7342 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7343 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7344 (actual_len > inode->i_size) &&
7345 (cur_offset > inode->i_size)) {
7346 if (cur_offset > actual_len)
7347 i_size = actual_len;
7349 i_size = cur_offset;
7350 i_size_write(inode, i_size);
7351 btrfs_ordered_update_i_size(inode, i_size, NULL);
7354 ret = btrfs_update_inode(trans, root, inode);
7358 btrfs_end_transaction(trans, root);
7363 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7364 u64 start, u64 num_bytes, u64 min_size,
7365 loff_t actual_len, u64 *alloc_hint)
7367 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7368 min_size, actual_len, alloc_hint,
7372 int btrfs_prealloc_file_range_trans(struct inode *inode,
7373 struct btrfs_trans_handle *trans, int mode,
7374 u64 start, u64 num_bytes, u64 min_size,
7375 loff_t actual_len, u64 *alloc_hint)
7377 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7378 min_size, actual_len, alloc_hint, trans);
7381 static int btrfs_set_page_dirty(struct page *page)
7383 return __set_page_dirty_nobuffers(page);
7386 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7388 struct btrfs_root *root = BTRFS_I(inode)->root;
7390 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7392 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7394 return generic_permission(inode, mask, flags, btrfs_check_acl);
7397 static const struct inode_operations btrfs_dir_inode_operations = {
7398 .getattr = btrfs_getattr,
7399 .lookup = btrfs_lookup,
7400 .create = btrfs_create,
7401 .unlink = btrfs_unlink,
7403 .mkdir = btrfs_mkdir,
7404 .rmdir = btrfs_rmdir,
7405 .rename = btrfs_rename,
7406 .symlink = btrfs_symlink,
7407 .setattr = btrfs_setattr,
7408 .mknod = btrfs_mknod,
7409 .setxattr = btrfs_setxattr,
7410 .getxattr = btrfs_getxattr,
7411 .listxattr = btrfs_listxattr,
7412 .removexattr = btrfs_removexattr,
7413 .permission = btrfs_permission,
7415 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7416 .lookup = btrfs_lookup,
7417 .permission = btrfs_permission,
7420 static const struct file_operations btrfs_dir_file_operations = {
7421 .llseek = generic_file_llseek,
7422 .read = generic_read_dir,
7423 .readdir = btrfs_real_readdir,
7424 .unlocked_ioctl = btrfs_ioctl,
7425 #ifdef CONFIG_COMPAT
7426 .compat_ioctl = btrfs_ioctl,
7428 .release = btrfs_release_file,
7429 .fsync = btrfs_sync_file,
7432 static struct extent_io_ops btrfs_extent_io_ops = {
7433 .fill_delalloc = run_delalloc_range,
7434 .submit_bio_hook = btrfs_submit_bio_hook,
7435 .merge_bio_hook = btrfs_merge_bio_hook,
7436 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7437 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7438 .writepage_start_hook = btrfs_writepage_start_hook,
7439 .readpage_io_failed_hook = btrfs_io_failed_hook,
7440 .set_bit_hook = btrfs_set_bit_hook,
7441 .clear_bit_hook = btrfs_clear_bit_hook,
7442 .merge_extent_hook = btrfs_merge_extent_hook,
7443 .split_extent_hook = btrfs_split_extent_hook,
7447 * btrfs doesn't support the bmap operation because swapfiles
7448 * use bmap to make a mapping of extents in the file. They assume
7449 * these extents won't change over the life of the file and they
7450 * use the bmap result to do IO directly to the drive.
7452 * the btrfs bmap call would return logical addresses that aren't
7453 * suitable for IO and they also will change frequently as COW
7454 * operations happen. So, swapfile + btrfs == corruption.
7456 * For now we're avoiding this by dropping bmap.
7458 static const struct address_space_operations btrfs_aops = {
7459 .readpage = btrfs_readpage,
7460 .writepage = btrfs_writepage,
7461 .writepages = btrfs_writepages,
7462 .readpages = btrfs_readpages,
7463 .sync_page = block_sync_page,
7464 .direct_IO = btrfs_direct_IO,
7465 .invalidatepage = btrfs_invalidatepage,
7466 .releasepage = btrfs_releasepage,
7467 .set_page_dirty = btrfs_set_page_dirty,
7468 .error_remove_page = generic_error_remove_page,
7471 static const struct address_space_operations btrfs_symlink_aops = {
7472 .readpage = btrfs_readpage,
7473 .writepage = btrfs_writepage,
7474 .invalidatepage = btrfs_invalidatepage,
7475 .releasepage = btrfs_releasepage,
7478 static const struct inode_operations btrfs_file_inode_operations = {
7479 .getattr = btrfs_getattr,
7480 .setattr = btrfs_setattr,
7481 .setxattr = btrfs_setxattr,
7482 .getxattr = btrfs_getxattr,
7483 .listxattr = btrfs_listxattr,
7484 .removexattr = btrfs_removexattr,
7485 .permission = btrfs_permission,
7486 .fiemap = btrfs_fiemap,
7488 static const struct inode_operations btrfs_special_inode_operations = {
7489 .getattr = btrfs_getattr,
7490 .setattr = btrfs_setattr,
7491 .permission = btrfs_permission,
7492 .setxattr = btrfs_setxattr,
7493 .getxattr = btrfs_getxattr,
7494 .listxattr = btrfs_listxattr,
7495 .removexattr = btrfs_removexattr,
7497 static const struct inode_operations btrfs_symlink_inode_operations = {
7498 .readlink = generic_readlink,
7499 .follow_link = page_follow_link_light,
7500 .put_link = page_put_link,
7501 .getattr = btrfs_getattr,
7502 .permission = btrfs_permission,
7503 .setxattr = btrfs_setxattr,
7504 .getxattr = btrfs_getxattr,
7505 .listxattr = btrfs_listxattr,
7506 .removexattr = btrfs_removexattr,
7509 const struct dentry_operations btrfs_dentry_operations = {
7510 .d_delete = btrfs_dentry_delete,