2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir,
97 const struct qstr *qstr)
101 err = btrfs_init_acl(trans, inode, dir);
103 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
108 * this does all the hard work for inserting an inline extent into
109 * the btree. The caller should have done a btrfs_drop_extents so that
110 * no overlapping inline items exist in the btree
112 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
113 struct btrfs_root *root, struct inode *inode,
114 u64 start, size_t size, size_t compressed_size,
116 struct page **compressed_pages)
118 struct btrfs_key key;
119 struct btrfs_path *path;
120 struct extent_buffer *leaf;
121 struct page *page = NULL;
124 struct btrfs_file_extent_item *ei;
127 size_t cur_size = size;
129 unsigned long offset;
131 if (compressed_size && compressed_pages)
132 cur_size = compressed_size;
134 path = btrfs_alloc_path();
138 path->leave_spinning = 1;
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);
423 BUG_ON(IS_ERR(trans));
424 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
426 /* lets try to make an inline extent */
427 if (ret || total_in < (actual_end - start)) {
428 /* we didn't compress the entire range, try
429 * to make an uncompressed inline extent.
431 ret = cow_file_range_inline(trans, root, inode,
432 start, end, 0, 0, NULL);
434 /* try making a compressed inline extent */
435 ret = cow_file_range_inline(trans, root, inode,
438 compress_type, pages);
442 * inline extent creation worked, we don't need
443 * to create any more async work items. Unlock
444 * and free up our temp pages.
446 extent_clear_unlock_delalloc(inode,
447 &BTRFS_I(inode)->io_tree,
449 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
450 EXTENT_CLEAR_DELALLOC |
451 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
453 btrfs_end_transaction(trans, root);
456 btrfs_end_transaction(trans, root);
461 * we aren't doing an inline extent round the compressed size
462 * up to a block size boundary so the allocator does sane
465 total_compressed = (total_compressed + blocksize - 1) &
469 * one last check to make sure the compression is really a
470 * win, compare the page count read with the blocks on disk
472 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
473 ~(PAGE_CACHE_SIZE - 1);
474 if (total_compressed >= total_in) {
477 num_bytes = total_in;
480 if (!will_compress && pages) {
482 * the compression code ran but failed to make things smaller,
483 * free any pages it allocated and our page pointer array
485 for (i = 0; i < nr_pages_ret; i++) {
486 WARN_ON(pages[i]->mapping);
487 page_cache_release(pages[i]);
491 total_compressed = 0;
494 /* flag the file so we don't compress in the future */
495 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
496 !(BTRFS_I(inode)->force_compress)) {
497 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
503 /* the async work queues will take care of doing actual
504 * allocation on disk for these compressed pages,
505 * and will submit them to the elevator.
507 add_async_extent(async_cow, start, num_bytes,
508 total_compressed, pages, nr_pages_ret,
511 if (start + num_bytes < end) {
518 cleanup_and_bail_uncompressed:
520 * No compression, but we still need to write the pages in
521 * the file we've been given so far. redirty the locked
522 * page if it corresponds to our extent and set things up
523 * for the async work queue to run cow_file_range to do
524 * the normal delalloc dance
526 if (page_offset(locked_page) >= start &&
527 page_offset(locked_page) <= end) {
528 __set_page_dirty_nobuffers(locked_page);
529 /* unlocked later on in the async handlers */
531 add_async_extent(async_cow, start, end - start + 1,
532 0, NULL, 0, BTRFS_COMPRESS_NONE);
540 for (i = 0; i < nr_pages_ret; i++) {
541 WARN_ON(pages[i]->mapping);
542 page_cache_release(pages[i]);
550 * phase two of compressed writeback. This is the ordered portion
551 * of the code, which only gets called in the order the work was
552 * queued. We walk all the async extents created by compress_file_range
553 * and send them down to the disk.
555 static noinline int submit_compressed_extents(struct inode *inode,
556 struct async_cow *async_cow)
558 struct async_extent *async_extent;
560 struct btrfs_trans_handle *trans;
561 struct btrfs_key ins;
562 struct extent_map *em;
563 struct btrfs_root *root = BTRFS_I(inode)->root;
564 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
565 struct extent_io_tree *io_tree;
568 if (list_empty(&async_cow->extents))
572 while (!list_empty(&async_cow->extents)) {
573 async_extent = list_entry(async_cow->extents.next,
574 struct async_extent, list);
575 list_del(&async_extent->list);
577 io_tree = &BTRFS_I(inode)->io_tree;
580 /* did the compression code fall back to uncompressed IO? */
581 if (!async_extent->pages) {
582 int page_started = 0;
583 unsigned long nr_written = 0;
585 lock_extent(io_tree, async_extent->start,
586 async_extent->start +
587 async_extent->ram_size - 1, GFP_NOFS);
589 /* allocate blocks */
590 ret = cow_file_range(inode, async_cow->locked_page,
592 async_extent->start +
593 async_extent->ram_size - 1,
594 &page_started, &nr_written, 0);
597 * if page_started, cow_file_range inserted an
598 * inline extent and took care of all the unlocking
599 * and IO for us. Otherwise, we need to submit
600 * all those pages down to the drive.
602 if (!page_started && !ret)
603 extent_write_locked_range(io_tree,
604 inode, async_extent->start,
605 async_extent->start +
606 async_extent->ram_size - 1,
614 lock_extent(io_tree, async_extent->start,
615 async_extent->start + async_extent->ram_size - 1,
618 trans = btrfs_join_transaction(root);
619 BUG_ON(IS_ERR(trans));
620 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
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);
781 BUG_ON(IS_ERR(trans));
782 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
784 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
785 num_bytes = max(blocksize, num_bytes);
786 disk_num_bytes = num_bytes;
790 /* lets try to make an inline extent */
791 ret = cow_file_range_inline(trans, root, inode,
792 start, end, 0, 0, NULL);
794 extent_clear_unlock_delalloc(inode,
795 &BTRFS_I(inode)->io_tree,
797 EXTENT_CLEAR_UNLOCK_PAGE |
798 EXTENT_CLEAR_UNLOCK |
799 EXTENT_CLEAR_DELALLOC |
801 EXTENT_SET_WRITEBACK |
802 EXTENT_END_WRITEBACK);
804 *nr_written = *nr_written +
805 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
812 BUG_ON(disk_num_bytes >
813 btrfs_super_total_bytes(&root->fs_info->super_copy));
815 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
816 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
818 while (disk_num_bytes > 0) {
821 cur_alloc_size = disk_num_bytes;
822 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
823 root->sectorsize, 0, alloc_hint,
827 em = alloc_extent_map(GFP_NOFS);
830 em->orig_start = em->start;
831 ram_size = ins.offset;
832 em->len = ins.offset;
834 em->block_start = ins.objectid;
835 em->block_len = ins.offset;
836 em->bdev = root->fs_info->fs_devices->latest_bdev;
837 set_bit(EXTENT_FLAG_PINNED, &em->flags);
840 write_lock(&em_tree->lock);
841 ret = add_extent_mapping(em_tree, em);
842 write_unlock(&em_tree->lock);
843 if (ret != -EEXIST) {
847 btrfs_drop_extent_cache(inode, start,
848 start + ram_size - 1, 0);
851 cur_alloc_size = ins.offset;
852 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
853 ram_size, cur_alloc_size, 0);
856 if (root->root_key.objectid ==
857 BTRFS_DATA_RELOC_TREE_OBJECTID) {
858 ret = btrfs_reloc_clone_csums(inode, start,
863 if (disk_num_bytes < cur_alloc_size)
866 /* we're not doing compressed IO, don't unlock the first
867 * page (which the caller expects to stay locked), don't
868 * clear any dirty bits and don't set any writeback bits
870 * Do set the Private2 bit so we know this page was properly
871 * setup for writepage
873 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
874 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
877 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
878 start, start + ram_size - 1,
880 disk_num_bytes -= cur_alloc_size;
881 num_bytes -= cur_alloc_size;
882 alloc_hint = ins.objectid + ins.offset;
883 start += cur_alloc_size;
887 btrfs_end_transaction(trans, root);
893 * work queue call back to started compression on a file and pages
895 static noinline void async_cow_start(struct btrfs_work *work)
897 struct async_cow *async_cow;
899 async_cow = container_of(work, struct async_cow, work);
901 compress_file_range(async_cow->inode, async_cow->locked_page,
902 async_cow->start, async_cow->end, async_cow,
905 async_cow->inode = NULL;
909 * work queue call back to submit previously compressed pages
911 static noinline void async_cow_submit(struct btrfs_work *work)
913 struct async_cow *async_cow;
914 struct btrfs_root *root;
915 unsigned long nr_pages;
917 async_cow = container_of(work, struct async_cow, work);
919 root = async_cow->root;
920 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
923 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
925 if (atomic_read(&root->fs_info->async_delalloc_pages) <
927 waitqueue_active(&root->fs_info->async_submit_wait))
928 wake_up(&root->fs_info->async_submit_wait);
930 if (async_cow->inode)
931 submit_compressed_extents(async_cow->inode, async_cow);
934 static noinline void async_cow_free(struct btrfs_work *work)
936 struct async_cow *async_cow;
937 async_cow = container_of(work, struct async_cow, work);
941 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
942 u64 start, u64 end, int *page_started,
943 unsigned long *nr_written)
945 struct async_cow *async_cow;
946 struct btrfs_root *root = BTRFS_I(inode)->root;
947 unsigned long nr_pages;
949 int limit = 10 * 1024 * 1042;
951 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
952 1, 0, NULL, GFP_NOFS);
953 while (start < end) {
954 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);
1058 trans = btrfs_join_transaction(root);
1060 BUG_ON(IS_ERR(trans));
1061 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1063 cow_start = (u64)-1;
1066 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1069 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070 leaf = path->nodes[0];
1071 btrfs_item_key_to_cpu(leaf, &found_key,
1072 path->slots[0] - 1);
1073 if (found_key.objectid == inode->i_ino &&
1074 found_key.type == BTRFS_EXTENT_DATA_KEY)
1079 leaf = path->nodes[0];
1080 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081 ret = btrfs_next_leaf(root, path);
1086 leaf = path->nodes[0];
1092 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1094 if (found_key.objectid > inode->i_ino ||
1095 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096 found_key.offset > end)
1099 if (found_key.offset > cur_offset) {
1100 extent_end = found_key.offset;
1105 fi = btrfs_item_ptr(leaf, path->slots[0],
1106 struct btrfs_file_extent_item);
1107 extent_type = btrfs_file_extent_type(leaf, fi);
1109 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112 extent_offset = btrfs_file_extent_offset(leaf, fi);
1113 extent_end = found_key.offset +
1114 btrfs_file_extent_num_bytes(leaf, fi);
1115 if (extent_end <= start) {
1119 if (disk_bytenr == 0)
1121 if (btrfs_file_extent_compression(leaf, fi) ||
1122 btrfs_file_extent_encryption(leaf, fi) ||
1123 btrfs_file_extent_other_encoding(leaf, fi))
1125 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1127 if (btrfs_extent_readonly(root, disk_bytenr))
1129 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1131 extent_offset, disk_bytenr))
1133 disk_bytenr += extent_offset;
1134 disk_bytenr += cur_offset - found_key.offset;
1135 num_bytes = min(end + 1, extent_end) - cur_offset;
1137 * force cow if csum exists in the range.
1138 * this ensure that csum for a given extent are
1139 * either valid or do not exist.
1141 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1144 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145 extent_end = found_key.offset +
1146 btrfs_file_extent_inline_len(leaf, fi);
1147 extent_end = ALIGN(extent_end, root->sectorsize);
1152 if (extent_end <= start) {
1157 if (cow_start == (u64)-1)
1158 cow_start = cur_offset;
1159 cur_offset = extent_end;
1160 if (cur_offset > end)
1166 btrfs_release_path(root, path);
1167 if (cow_start != (u64)-1) {
1168 ret = cow_file_range(inode, locked_page, cow_start,
1169 found_key.offset - 1, page_started,
1172 cow_start = (u64)-1;
1175 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176 struct extent_map *em;
1177 struct extent_map_tree *em_tree;
1178 em_tree = &BTRFS_I(inode)->extent_tree;
1179 em = alloc_extent_map(GFP_NOFS);
1181 em->start = cur_offset;
1182 em->orig_start = em->start;
1183 em->len = num_bytes;
1184 em->block_len = num_bytes;
1185 em->block_start = disk_bytenr;
1186 em->bdev = root->fs_info->fs_devices->latest_bdev;
1187 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1189 write_lock(&em_tree->lock);
1190 ret = add_extent_mapping(em_tree, em);
1191 write_unlock(&em_tree->lock);
1192 if (ret != -EEXIST) {
1193 free_extent_map(em);
1196 btrfs_drop_extent_cache(inode, em->start,
1197 em->start + em->len - 1, 0);
1199 type = BTRFS_ORDERED_PREALLOC;
1201 type = BTRFS_ORDERED_NOCOW;
1204 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205 num_bytes, num_bytes, type);
1208 if (root->root_key.objectid ==
1209 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1215 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216 cur_offset, cur_offset + num_bytes - 1,
1217 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219 EXTENT_SET_PRIVATE2);
1220 cur_offset = extent_end;
1221 if (cur_offset > end)
1224 btrfs_release_path(root, path);
1226 if (cur_offset <= end && cow_start == (u64)-1)
1227 cow_start = cur_offset;
1228 if (cow_start != (u64)-1) {
1229 ret = cow_file_range(inode, locked_page, cow_start, end,
1230 page_started, nr_written, 1);
1235 ret = btrfs_end_transaction_nolock(trans, root);
1238 ret = btrfs_end_transaction(trans, root);
1241 btrfs_free_path(path);
1246 * extent_io.c call back to do delayed allocation processing
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249 u64 start, u64 end, int *page_started,
1250 unsigned long *nr_written)
1253 struct btrfs_root *root = BTRFS_I(inode)->root;
1255 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257 page_started, 1, nr_written);
1258 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260 page_started, 0, nr_written);
1261 else if (!btrfs_test_opt(root, COMPRESS) &&
1262 !(BTRFS_I(inode)->force_compress) &&
1263 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264 ret = cow_file_range(inode, locked_page, start, end,
1265 page_started, nr_written, 1);
1267 ret = cow_file_range_async(inode, locked_page, start, end,
1268 page_started, nr_written);
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273 struct extent_state *orig, u64 split)
1275 /* not delalloc, ignore it */
1276 if (!(orig->state & EXTENT_DELALLOC))
1279 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1284 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285 * extents so we can keep track of new extents that are just merged onto old
1286 * extents, such as when we are doing sequential writes, so we can properly
1287 * account for the metadata space we'll need.
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290 struct extent_state *new,
1291 struct extent_state *other)
1293 /* not delalloc, ignore it */
1294 if (!(other->state & EXTENT_DELALLOC))
1297 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1302 * extent_io.c set_bit_hook, used to track delayed allocation
1303 * bytes in this file, and to maintain the list of inodes that
1304 * have pending delalloc work to be done.
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307 struct extent_state *state, int *bits)
1311 * set_bit and clear bit hooks normally require _irqsave/restore
1312 * but in this case, we are only testeing for the DELALLOC
1313 * bit, which is only set or cleared with irqs on
1315 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316 struct btrfs_root *root = BTRFS_I(inode)->root;
1317 u64 len = state->end + 1 - state->start;
1318 int do_list = (root->root_key.objectid !=
1319 BTRFS_ROOT_TREE_OBJECTID);
1321 if (*bits & EXTENT_FIRST_DELALLOC)
1322 *bits &= ~EXTENT_FIRST_DELALLOC;
1324 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1326 spin_lock(&root->fs_info->delalloc_lock);
1327 BTRFS_I(inode)->delalloc_bytes += len;
1328 root->fs_info->delalloc_bytes += len;
1329 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331 &root->fs_info->delalloc_inodes);
1333 spin_unlock(&root->fs_info->delalloc_lock);
1339 * extent_io.c clear_bit_hook, see set_bit_hook for why
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342 struct extent_state *state, int *bits)
1345 * set_bit and clear bit hooks normally require _irqsave/restore
1346 * but in this case, we are only testeing for the DELALLOC
1347 * bit, which is only set or cleared with irqs on
1349 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350 struct btrfs_root *root = BTRFS_I(inode)->root;
1351 u64 len = state->end + 1 - state->start;
1352 int do_list = (root->root_key.objectid !=
1353 BTRFS_ROOT_TREE_OBJECTID);
1355 if (*bits & EXTENT_FIRST_DELALLOC)
1356 *bits &= ~EXTENT_FIRST_DELALLOC;
1357 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1360 if (*bits & EXTENT_DO_ACCOUNTING)
1361 btrfs_delalloc_release_metadata(inode, len);
1363 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1365 btrfs_free_reserved_data_space(inode, len);
1367 spin_lock(&root->fs_info->delalloc_lock);
1368 root->fs_info->delalloc_bytes -= len;
1369 BTRFS_I(inode)->delalloc_bytes -= len;
1371 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1375 spin_unlock(&root->fs_info->delalloc_lock);
1381 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382 * we don't create bios that span stripes or chunks
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385 size_t size, struct bio *bio,
1386 unsigned long bio_flags)
1388 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389 struct btrfs_mapping_tree *map_tree;
1390 u64 logical = (u64)bio->bi_sector << 9;
1395 if (bio_flags & EXTENT_BIO_COMPRESSED)
1398 length = bio->bi_size;
1399 map_tree = &root->fs_info->mapping_tree;
1400 map_length = length;
1401 ret = btrfs_map_block(map_tree, READ, logical,
1402 &map_length, NULL, 0);
1404 if (map_length < length + size)
1410 * in order to insert checksums into the metadata in large chunks,
1411 * we wait until bio submission time. All the pages in the bio are
1412 * checksummed and sums are attached onto the ordered extent record.
1414 * At IO completion time the cums attached on the ordered extent record
1415 * are inserted into the btree
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418 struct bio *bio, int mirror_num,
1419 unsigned long bio_flags,
1422 struct btrfs_root *root = BTRFS_I(inode)->root;
1425 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1431 * in order to insert checksums into the metadata in large chunks,
1432 * we wait until bio submission time. All the pages in the bio are
1433 * checksummed and sums are attached onto the ordered extent record.
1435 * At IO completion time the cums attached on the ordered extent record
1436 * are inserted into the btree
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439 int mirror_num, unsigned long bio_flags,
1442 struct btrfs_root *root = BTRFS_I(inode)->root;
1443 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1447 * extent_io.c submission hook. This does the right thing for csum calculation
1448 * on write, or reading the csums from the tree before a read
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451 int mirror_num, unsigned long bio_flags,
1454 struct btrfs_root *root = BTRFS_I(inode)->root;
1458 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1460 if (root == root->fs_info->tree_root)
1461 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1463 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1466 if (!(rw & REQ_WRITE)) {
1467 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468 return btrfs_submit_compressed_read(inode, bio,
1469 mirror_num, bio_flags);
1470 } else if (!skip_sum) {
1471 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1476 } else if (!skip_sum) {
1477 /* csum items have already been cloned */
1478 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1480 /* we're doing a write, do the async checksumming */
1481 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482 inode, rw, bio, mirror_num,
1483 bio_flags, bio_offset,
1484 __btrfs_submit_bio_start,
1485 __btrfs_submit_bio_done);
1489 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1493 * given a list of ordered sums record them in the inode. This happens
1494 * at IO completion time based on sums calculated at bio submission time.
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497 struct inode *inode, u64 file_offset,
1498 struct list_head *list)
1500 struct btrfs_ordered_sum *sum;
1502 list_for_each_entry(sum, list, list) {
1503 btrfs_csum_file_blocks(trans,
1504 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1509 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1510 struct extent_state **cached_state)
1512 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1514 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1515 cached_state, GFP_NOFS);
1518 /* see btrfs_writepage_start_hook for details on why this is required */
1519 struct btrfs_writepage_fixup {
1521 struct btrfs_work work;
1524 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1526 struct btrfs_writepage_fixup *fixup;
1527 struct btrfs_ordered_extent *ordered;
1528 struct extent_state *cached_state = NULL;
1530 struct inode *inode;
1534 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1539 ClearPageChecked(page);
1543 inode = page->mapping->host;
1544 page_start = page_offset(page);
1545 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1547 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1548 &cached_state, GFP_NOFS);
1550 /* already ordered? We're done */
1551 if (PagePrivate2(page))
1554 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1556 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1557 page_end, &cached_state, GFP_NOFS);
1559 btrfs_start_ordered_extent(inode, ordered, 1);
1564 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1565 ClearPageChecked(page);
1567 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1568 &cached_state, GFP_NOFS);
1571 page_cache_release(page);
1576 * There are a few paths in the higher layers of the kernel that directly
1577 * set the page dirty bit without asking the filesystem if it is a
1578 * good idea. This causes problems because we want to make sure COW
1579 * properly happens and the data=ordered rules are followed.
1581 * In our case any range that doesn't have the ORDERED bit set
1582 * hasn't been properly setup for IO. We kick off an async process
1583 * to fix it up. The async helper will wait for ordered extents, set
1584 * the delalloc bit and make it safe to write the page.
1586 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1588 struct inode *inode = page->mapping->host;
1589 struct btrfs_writepage_fixup *fixup;
1590 struct btrfs_root *root = BTRFS_I(inode)->root;
1592 /* this page is properly in the ordered list */
1593 if (TestClearPagePrivate2(page))
1596 if (PageChecked(page))
1599 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1603 SetPageChecked(page);
1604 page_cache_get(page);
1605 fixup->work.func = btrfs_writepage_fixup_worker;
1607 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1611 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1612 struct inode *inode, u64 file_pos,
1613 u64 disk_bytenr, u64 disk_num_bytes,
1614 u64 num_bytes, u64 ram_bytes,
1615 u8 compression, u8 encryption,
1616 u16 other_encoding, int extent_type)
1618 struct btrfs_root *root = BTRFS_I(inode)->root;
1619 struct btrfs_file_extent_item *fi;
1620 struct btrfs_path *path;
1621 struct extent_buffer *leaf;
1622 struct btrfs_key ins;
1626 path = btrfs_alloc_path();
1629 path->leave_spinning = 1;
1632 * we may be replacing one extent in the tree with another.
1633 * The new extent is pinned in the extent map, and we don't want
1634 * to drop it from the cache until it is completely in the btree.
1636 * So, tell btrfs_drop_extents to leave this extent in the cache.
1637 * the caller is expected to unpin it and allow it to be merged
1640 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1644 ins.objectid = inode->i_ino;
1645 ins.offset = file_pos;
1646 ins.type = BTRFS_EXTENT_DATA_KEY;
1647 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1649 leaf = path->nodes[0];
1650 fi = btrfs_item_ptr(leaf, path->slots[0],
1651 struct btrfs_file_extent_item);
1652 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1653 btrfs_set_file_extent_type(leaf, fi, extent_type);
1654 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1655 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1656 btrfs_set_file_extent_offset(leaf, fi, 0);
1657 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1658 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1659 btrfs_set_file_extent_compression(leaf, fi, compression);
1660 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1661 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1663 btrfs_unlock_up_safe(path, 1);
1664 btrfs_set_lock_blocking(leaf);
1666 btrfs_mark_buffer_dirty(leaf);
1668 inode_add_bytes(inode, num_bytes);
1670 ins.objectid = disk_bytenr;
1671 ins.offset = disk_num_bytes;
1672 ins.type = BTRFS_EXTENT_ITEM_KEY;
1673 ret = btrfs_alloc_reserved_file_extent(trans, root,
1674 root->root_key.objectid,
1675 inode->i_ino, file_pos, &ins);
1677 btrfs_free_path(path);
1683 * helper function for btrfs_finish_ordered_io, this
1684 * just reads in some of the csum leaves to prime them into ram
1685 * before we start the transaction. It limits the amount of btree
1686 * reads required while inside the transaction.
1688 /* as ordered data IO finishes, this gets called so we can finish
1689 * an ordered extent if the range of bytes in the file it covers are
1692 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1694 struct btrfs_root *root = BTRFS_I(inode)->root;
1695 struct btrfs_trans_handle *trans = NULL;
1696 struct btrfs_ordered_extent *ordered_extent = NULL;
1697 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1698 struct extent_state *cached_state = NULL;
1699 int compress_type = 0;
1701 bool nolock = false;
1703 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1707 BUG_ON(!ordered_extent);
1709 nolock = (root == root->fs_info->tree_root);
1711 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1712 BUG_ON(!list_empty(&ordered_extent->list));
1713 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716 trans = btrfs_join_transaction_nolock(root);
1718 trans = btrfs_join_transaction(root);
1719 BUG_ON(IS_ERR(trans));
1720 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1721 ret = btrfs_update_inode(trans, root, inode);
1727 lock_extent_bits(io_tree, ordered_extent->file_offset,
1728 ordered_extent->file_offset + ordered_extent->len - 1,
1729 0, &cached_state, GFP_NOFS);
1732 trans = btrfs_join_transaction_nolock(root);
1734 trans = btrfs_join_transaction(root);
1735 BUG_ON(IS_ERR(trans));
1736 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1738 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1739 compress_type = ordered_extent->compress_type;
1740 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1741 BUG_ON(compress_type);
1742 ret = btrfs_mark_extent_written(trans, inode,
1743 ordered_extent->file_offset,
1744 ordered_extent->file_offset +
1745 ordered_extent->len);
1748 BUG_ON(root == root->fs_info->tree_root);
1749 ret = insert_reserved_file_extent(trans, inode,
1750 ordered_extent->file_offset,
1751 ordered_extent->start,
1752 ordered_extent->disk_len,
1753 ordered_extent->len,
1754 ordered_extent->len,
1755 compress_type, 0, 0,
1756 BTRFS_FILE_EXTENT_REG);
1757 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1758 ordered_extent->file_offset,
1759 ordered_extent->len);
1762 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1763 ordered_extent->file_offset +
1764 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1766 add_pending_csums(trans, inode, ordered_extent->file_offset,
1767 &ordered_extent->list);
1769 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1771 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 didn't
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);
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;
2494 path = btrfs_alloc_path();
2496 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2498 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2502 leaf = path->nodes[0];
2503 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2504 struct btrfs_inode_item);
2506 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2507 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2508 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2509 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2510 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2512 tspec = btrfs_inode_atime(inode_item);
2513 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2514 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2516 tspec = btrfs_inode_mtime(inode_item);
2517 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2518 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520 tspec = btrfs_inode_ctime(inode_item);
2521 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2525 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2526 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2527 inode->i_generation = BTRFS_I(inode)->generation;
2529 rdev = btrfs_inode_rdev(leaf, inode_item);
2531 BTRFS_I(inode)->index_cnt = (u64)-1;
2532 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2535 * try to precache a NULL acl entry for files that don't have
2536 * any xattrs or acls
2538 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2540 cache_no_acl(inode);
2542 btrfs_free_path(path);
2545 switch (inode->i_mode & S_IFMT) {
2547 inode->i_mapping->a_ops = &btrfs_aops;
2548 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2549 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2550 inode->i_fop = &btrfs_file_operations;
2551 inode->i_op = &btrfs_file_inode_operations;
2554 inode->i_fop = &btrfs_dir_file_operations;
2555 if (root == root->fs_info->tree_root)
2556 inode->i_op = &btrfs_dir_ro_inode_operations;
2558 inode->i_op = &btrfs_dir_inode_operations;
2561 inode->i_op = &btrfs_symlink_inode_operations;
2562 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2563 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2566 inode->i_op = &btrfs_special_inode_operations;
2567 init_special_inode(inode, inode->i_mode, rdev);
2571 btrfs_update_iflags(inode);
2575 btrfs_free_path(path);
2576 make_bad_inode(inode);
2580 * given a leaf and an inode, copy the inode fields into the leaf
2582 static void fill_inode_item(struct btrfs_trans_handle *trans,
2583 struct extent_buffer *leaf,
2584 struct btrfs_inode_item *item,
2585 struct inode *inode)
2587 if (!leaf->map_token)
2588 map_private_extent_buffer(leaf, (unsigned long)item,
2589 sizeof(struct btrfs_inode_item),
2590 &leaf->map_token, &leaf->kaddr,
2591 &leaf->map_start, &leaf->map_len,
2594 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2595 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2596 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2597 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2598 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2600 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2601 inode->i_atime.tv_sec);
2602 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2603 inode->i_atime.tv_nsec);
2605 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2606 inode->i_mtime.tv_sec);
2607 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2608 inode->i_mtime.tv_nsec);
2610 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2611 inode->i_ctime.tv_sec);
2612 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2613 inode->i_ctime.tv_nsec);
2615 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2616 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2617 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2618 btrfs_set_inode_transid(leaf, item, trans->transid);
2619 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2620 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2621 btrfs_set_inode_block_group(leaf, item, 0);
2623 if (leaf->map_token) {
2624 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2625 leaf->map_token = NULL;
2630 * copy everything in the in-memory inode into the btree.
2632 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2633 struct btrfs_root *root, struct inode *inode)
2635 struct btrfs_inode_item *inode_item;
2636 struct btrfs_path *path;
2637 struct extent_buffer *leaf;
2640 path = btrfs_alloc_path();
2642 path->leave_spinning = 1;
2643 ret = btrfs_lookup_inode(trans, root, path,
2644 &BTRFS_I(inode)->location, 1);
2651 btrfs_unlock_up_safe(path, 1);
2652 leaf = path->nodes[0];
2653 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2654 struct btrfs_inode_item);
2656 fill_inode_item(trans, leaf, inode_item, inode);
2657 btrfs_mark_buffer_dirty(leaf);
2658 btrfs_set_inode_last_trans(trans, inode);
2661 btrfs_free_path(path);
2667 * unlink helper that gets used here in inode.c and in the tree logging
2668 * recovery code. It remove a link in a directory with a given name, and
2669 * also drops the back refs in the inode to the directory
2671 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2672 struct btrfs_root *root,
2673 struct inode *dir, struct inode *inode,
2674 const char *name, int name_len)
2676 struct btrfs_path *path;
2678 struct extent_buffer *leaf;
2679 struct btrfs_dir_item *di;
2680 struct btrfs_key key;
2683 path = btrfs_alloc_path();
2689 path->leave_spinning = 1;
2690 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2691 name, name_len, -1);
2700 leaf = path->nodes[0];
2701 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2702 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2705 btrfs_release_path(root, path);
2707 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2709 dir->i_ino, &index);
2711 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2712 "inode %lu parent %lu\n", name_len, name,
2713 inode->i_ino, dir->i_ino);
2717 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2718 index, name, name_len, -1);
2727 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2728 btrfs_release_path(root, path);
2730 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2732 BUG_ON(ret != 0 && ret != -ENOENT);
2734 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2739 btrfs_free_path(path);
2743 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2744 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2745 btrfs_update_inode(trans, root, dir);
2750 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2751 struct btrfs_root *root,
2752 struct inode *dir, struct inode *inode,
2753 const char *name, int name_len)
2756 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2758 btrfs_drop_nlink(inode);
2759 ret = btrfs_update_inode(trans, root, inode);
2765 /* helper to check if there is any shared block in the path */
2766 static int check_path_shared(struct btrfs_root *root,
2767 struct btrfs_path *path)
2769 struct extent_buffer *eb;
2773 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2776 if (!path->nodes[level])
2778 eb = path->nodes[level];
2779 if (!btrfs_block_can_be_shared(root, eb))
2781 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2790 * helper to start transaction for unlink and rmdir.
2792 * unlink and rmdir are special in btrfs, they do not always free space.
2793 * so in enospc case, we should make sure they will free space before
2794 * allowing them to use the global metadata reservation.
2796 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2797 struct dentry *dentry)
2799 struct btrfs_trans_handle *trans;
2800 struct btrfs_root *root = BTRFS_I(dir)->root;
2801 struct btrfs_path *path;
2802 struct btrfs_inode_ref *ref;
2803 struct btrfs_dir_item *di;
2804 struct inode *inode = dentry->d_inode;
2810 trans = btrfs_start_transaction(root, 10);
2811 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2814 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2815 return ERR_PTR(-ENOSPC);
2817 /* check if there is someone else holds reference */
2818 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2819 return ERR_PTR(-ENOSPC);
2821 if (atomic_read(&inode->i_count) > 2)
2822 return ERR_PTR(-ENOSPC);
2824 if (xchg(&root->fs_info->enospc_unlink, 1))
2825 return ERR_PTR(-ENOSPC);
2827 path = btrfs_alloc_path();
2829 root->fs_info->enospc_unlink = 0;
2830 return ERR_PTR(-ENOMEM);
2833 trans = btrfs_start_transaction(root, 0);
2834 if (IS_ERR(trans)) {
2835 btrfs_free_path(path);
2836 root->fs_info->enospc_unlink = 0;
2840 path->skip_locking = 1;
2841 path->search_commit_root = 1;
2843 ret = btrfs_lookup_inode(trans, root, path,
2844 &BTRFS_I(dir)->location, 0);
2850 if (check_path_shared(root, path))
2855 btrfs_release_path(root, path);
2857 ret = btrfs_lookup_inode(trans, root, path,
2858 &BTRFS_I(inode)->location, 0);
2864 if (check_path_shared(root, path))
2869 btrfs_release_path(root, path);
2871 if (ret == 0 && S_ISREG(inode->i_mode)) {
2872 ret = btrfs_lookup_file_extent(trans, root, path,
2873 inode->i_ino, (u64)-1, 0);
2879 if (check_path_shared(root, path))
2881 btrfs_release_path(root, path);
2889 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2890 dentry->d_name.name, dentry->d_name.len, 0);
2896 if (check_path_shared(root, path))
2902 btrfs_release_path(root, path);
2904 ref = btrfs_lookup_inode_ref(trans, root, path,
2905 dentry->d_name.name, dentry->d_name.len,
2906 inode->i_ino, dir->i_ino, 0);
2912 if (check_path_shared(root, path))
2914 index = btrfs_inode_ref_index(path->nodes[0], ref);
2915 btrfs_release_path(root, path);
2917 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2918 dentry->d_name.name, dentry->d_name.len, 0);
2923 BUG_ON(ret == -ENOENT);
2924 if (check_path_shared(root, path))
2929 btrfs_free_path(path);
2931 btrfs_end_transaction(trans, root);
2932 root->fs_info->enospc_unlink = 0;
2933 return ERR_PTR(err);
2936 trans->block_rsv = &root->fs_info->global_block_rsv;
2940 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2941 struct btrfs_root *root)
2943 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2944 BUG_ON(!root->fs_info->enospc_unlink);
2945 root->fs_info->enospc_unlink = 0;
2947 btrfs_end_transaction_throttle(trans, root);
2950 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2952 struct btrfs_root *root = BTRFS_I(dir)->root;
2953 struct btrfs_trans_handle *trans;
2954 struct inode *inode = dentry->d_inode;
2956 unsigned long nr = 0;
2958 trans = __unlink_start_trans(dir, dentry);
2960 return PTR_ERR(trans);
2962 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2964 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2965 dentry->d_name.name, dentry->d_name.len);
2968 if (inode->i_nlink == 0) {
2969 ret = btrfs_orphan_add(trans, inode);
2973 nr = trans->blocks_used;
2974 __unlink_end_trans(trans, root);
2975 btrfs_btree_balance_dirty(root, nr);
2979 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2980 struct btrfs_root *root,
2981 struct inode *dir, u64 objectid,
2982 const char *name, int name_len)
2984 struct btrfs_path *path;
2985 struct extent_buffer *leaf;
2986 struct btrfs_dir_item *di;
2987 struct btrfs_key key;
2991 path = btrfs_alloc_path();
2995 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2996 name, name_len, -1);
2997 BUG_ON(!di || IS_ERR(di));
2999 leaf = path->nodes[0];
3000 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3001 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3002 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3004 btrfs_release_path(root, path);
3006 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3007 objectid, root->root_key.objectid,
3008 dir->i_ino, &index, name, name_len);
3010 BUG_ON(ret != -ENOENT);
3011 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3013 BUG_ON(!di || IS_ERR(di));
3015 leaf = path->nodes[0];
3016 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3017 btrfs_release_path(root, path);
3021 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3022 index, name, name_len, -1);
3023 BUG_ON(!di || IS_ERR(di));
3025 leaf = path->nodes[0];
3026 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3027 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3028 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3030 btrfs_release_path(root, path);
3032 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3033 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3034 ret = btrfs_update_inode(trans, root, dir);
3037 btrfs_free_path(path);
3041 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3043 struct inode *inode = dentry->d_inode;
3045 struct btrfs_root *root = BTRFS_I(dir)->root;
3046 struct btrfs_trans_handle *trans;
3047 unsigned long nr = 0;
3049 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3050 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3053 trans = __unlink_start_trans(dir, dentry);
3055 return PTR_ERR(trans);
3057 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3058 err = btrfs_unlink_subvol(trans, root, dir,
3059 BTRFS_I(inode)->location.objectid,
3060 dentry->d_name.name,
3061 dentry->d_name.len);
3065 err = btrfs_orphan_add(trans, inode);
3069 /* now the directory is empty */
3070 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3071 dentry->d_name.name, dentry->d_name.len);
3073 btrfs_i_size_write(inode, 0);
3075 nr = trans->blocks_used;
3076 __unlink_end_trans(trans, root);
3077 btrfs_btree_balance_dirty(root, nr);
3084 * when truncating bytes in a file, it is possible to avoid reading
3085 * the leaves that contain only checksum items. This can be the
3086 * majority of the IO required to delete a large file, but it must
3087 * be done carefully.
3089 * The keys in the level just above the leaves are checked to make sure
3090 * the lowest key in a given leaf is a csum key, and starts at an offset
3091 * after the new size.
3093 * Then the key for the next leaf is checked to make sure it also has
3094 * a checksum item for the same file. If it does, we know our target leaf
3095 * contains only checksum items, and it can be safely freed without reading
3098 * This is just an optimization targeted at large files. It may do
3099 * nothing. It will return 0 unless things went badly.
3101 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct btrfs_path *path,
3104 struct inode *inode, u64 new_size)
3106 struct btrfs_key key;
3109 struct btrfs_key found_key;
3110 struct btrfs_key other_key;
3111 struct btrfs_leaf_ref *ref;
3115 path->lowest_level = 1;
3116 key.objectid = inode->i_ino;
3117 key.type = BTRFS_CSUM_ITEM_KEY;
3118 key.offset = new_size;
3120 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3124 if (path->nodes[1] == NULL) {
3129 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3130 nritems = btrfs_header_nritems(path->nodes[1]);
3135 if (path->slots[1] >= nritems)
3138 /* did we find a key greater than anything we want to delete? */
3139 if (found_key.objectid > inode->i_ino ||
3140 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3143 /* we check the next key in the node to make sure the leave contains
3144 * only checksum items. This comparison doesn't work if our
3145 * leaf is the last one in the node
3147 if (path->slots[1] + 1 >= nritems) {
3149 /* search forward from the last key in the node, this
3150 * will bring us into the next node in the tree
3152 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3154 /* unlikely, but we inc below, so check to be safe */
3155 if (found_key.offset == (u64)-1)
3158 /* search_forward needs a path with locks held, do the
3159 * search again for the original key. It is possible
3160 * this will race with a balance and return a path that
3161 * we could modify, but this drop is just an optimization
3162 * and is allowed to miss some leaves.
3164 btrfs_release_path(root, path);
3167 /* setup a max key for search_forward */
3168 other_key.offset = (u64)-1;
3169 other_key.type = key.type;
3170 other_key.objectid = key.objectid;
3172 path->keep_locks = 1;
3173 ret = btrfs_search_forward(root, &found_key, &other_key,
3175 path->keep_locks = 0;
3176 if (ret || found_key.objectid != key.objectid ||
3177 found_key.type != key.type) {
3182 key.offset = found_key.offset;
3183 btrfs_release_path(root, path);
3188 /* we know there's one more slot after us in the tree,
3189 * read that key so we can verify it is also a checksum item
3191 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3193 if (found_key.objectid < inode->i_ino)
3196 if (found_key.type != key.type || found_key.offset < new_size)
3200 * if the key for the next leaf isn't a csum key from this objectid,
3201 * we can't be sure there aren't good items inside this leaf.
3204 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3207 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3208 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3210 * it is safe to delete this leaf, it contains only
3211 * csum items from this inode at an offset >= new_size
3213 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3216 if (root->ref_cows && leaf_gen < trans->transid) {
3217 ref = btrfs_alloc_leaf_ref(root, 0);
3219 ref->root_gen = root->root_key.offset;
3220 ref->bytenr = leaf_start;
3222 ref->generation = leaf_gen;
3225 btrfs_sort_leaf_ref(ref);
3227 ret = btrfs_add_leaf_ref(root, ref, 0);
3229 btrfs_free_leaf_ref(root, ref);
3235 btrfs_release_path(root, path);
3237 if (other_key.objectid == inode->i_ino &&
3238 other_key.type == key.type && other_key.offset > key.offset) {
3239 key.offset = other_key.offset;
3245 /* fixup any changes we've made to the path */
3246 path->lowest_level = 0;
3247 path->keep_locks = 0;
3248 btrfs_release_path(root, path);
3255 * this can truncate away extent items, csum items and directory items.
3256 * It starts at a high offset and removes keys until it can't find
3257 * any higher than new_size
3259 * csum items that cross the new i_size are truncated to the new size
3262 * min_type is the minimum key type to truncate down to. If set to 0, this
3263 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3265 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3266 struct btrfs_root *root,
3267 struct inode *inode,
3268 u64 new_size, u32 min_type)
3270 struct btrfs_path *path;
3271 struct extent_buffer *leaf;
3272 struct btrfs_file_extent_item *fi;
3273 struct btrfs_key key;
3274 struct btrfs_key found_key;
3275 u64 extent_start = 0;
3276 u64 extent_num_bytes = 0;
3277 u64 extent_offset = 0;
3279 u64 mask = root->sectorsize - 1;
3280 u32 found_type = (u8)-1;
3283 int pending_del_nr = 0;
3284 int pending_del_slot = 0;
3285 int extent_type = -1;
3290 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3292 if (root->ref_cows || root == root->fs_info->tree_root)
3293 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3295 path = btrfs_alloc_path();
3299 key.objectid = inode->i_ino;
3300 key.offset = (u64)-1;
3304 path->leave_spinning = 1;
3305 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3312 /* there are no items in the tree for us to truncate, we're
3315 if (path->slots[0] == 0)
3322 leaf = path->nodes[0];
3323 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3324 found_type = btrfs_key_type(&found_key);
3327 if (found_key.objectid != inode->i_ino)
3330 if (found_type < min_type)
3333 item_end = found_key.offset;
3334 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3335 fi = btrfs_item_ptr(leaf, path->slots[0],
3336 struct btrfs_file_extent_item);
3337 extent_type = btrfs_file_extent_type(leaf, fi);
3338 encoding = btrfs_file_extent_compression(leaf, fi);
3339 encoding |= btrfs_file_extent_encryption(leaf, fi);
3340 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3342 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3344 btrfs_file_extent_num_bytes(leaf, fi);
3345 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3346 item_end += btrfs_file_extent_inline_len(leaf,
3351 if (found_type > min_type) {
3354 if (item_end < new_size)
3356 if (found_key.offset >= new_size)
3362 /* FIXME, shrink the extent if the ref count is only 1 */
3363 if (found_type != BTRFS_EXTENT_DATA_KEY)
3366 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3368 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3369 if (!del_item && !encoding) {
3370 u64 orig_num_bytes =
3371 btrfs_file_extent_num_bytes(leaf, fi);
3372 extent_num_bytes = new_size -
3373 found_key.offset + root->sectorsize - 1;
3374 extent_num_bytes = extent_num_bytes &
3375 ~((u64)root->sectorsize - 1);
3376 btrfs_set_file_extent_num_bytes(leaf, fi,
3378 num_dec = (orig_num_bytes -
3380 if (root->ref_cows && extent_start != 0)
3381 inode_sub_bytes(inode, num_dec);
3382 btrfs_mark_buffer_dirty(leaf);
3385 btrfs_file_extent_disk_num_bytes(leaf,
3387 extent_offset = found_key.offset -
3388 btrfs_file_extent_offset(leaf, fi);
3390 /* FIXME blocksize != 4096 */
3391 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3392 if (extent_start != 0) {
3395 inode_sub_bytes(inode, num_dec);
3398 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3400 * we can't truncate inline items that have had
3404 btrfs_file_extent_compression(leaf, fi) == 0 &&
3405 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3406 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3407 u32 size = new_size - found_key.offset;
3409 if (root->ref_cows) {
3410 inode_sub_bytes(inode, item_end + 1 -
3414 btrfs_file_extent_calc_inline_size(size);
3415 ret = btrfs_truncate_item(trans, root, path,
3418 } else if (root->ref_cows) {
3419 inode_sub_bytes(inode, item_end + 1 -
3425 if (!pending_del_nr) {
3426 /* no pending yet, add ourselves */
3427 pending_del_slot = path->slots[0];
3429 } else if (pending_del_nr &&
3430 path->slots[0] + 1 == pending_del_slot) {
3431 /* hop on the pending chunk */
3433 pending_del_slot = path->slots[0];
3440 if (found_extent && (root->ref_cows ||
3441 root == root->fs_info->tree_root)) {
3442 btrfs_set_path_blocking(path);
3443 ret = btrfs_free_extent(trans, root, extent_start,
3444 extent_num_bytes, 0,
3445 btrfs_header_owner(leaf),
3446 inode->i_ino, extent_offset);
3450 if (found_type == BTRFS_INODE_ITEM_KEY)
3453 if (path->slots[0] == 0 ||
3454 path->slots[0] != pending_del_slot) {
3455 if (root->ref_cows) {
3459 if (pending_del_nr) {
3460 ret = btrfs_del_items(trans, root, path,
3466 btrfs_release_path(root, path);
3473 if (pending_del_nr) {
3474 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3478 btrfs_free_path(path);
3483 * taken from block_truncate_page, but does cow as it zeros out
3484 * any bytes left in the last page in the file.
3486 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3488 struct inode *inode = mapping->host;
3489 struct btrfs_root *root = BTRFS_I(inode)->root;
3490 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3491 struct btrfs_ordered_extent *ordered;
3492 struct extent_state *cached_state = NULL;
3494 u32 blocksize = root->sectorsize;
3495 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3496 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3502 if ((offset & (blocksize - 1)) == 0)
3504 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3510 page = grab_cache_page(mapping, index);
3512 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3516 page_start = page_offset(page);
3517 page_end = page_start + PAGE_CACHE_SIZE - 1;
3519 if (!PageUptodate(page)) {
3520 ret = btrfs_readpage(NULL, page);
3522 if (page->mapping != mapping) {
3524 page_cache_release(page);
3527 if (!PageUptodate(page)) {
3532 wait_on_page_writeback(page);
3534 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3536 set_page_extent_mapped(page);
3538 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3540 unlock_extent_cached(io_tree, page_start, page_end,
3541 &cached_state, GFP_NOFS);
3543 page_cache_release(page);
3544 btrfs_start_ordered_extent(inode, ordered, 1);
3545 btrfs_put_ordered_extent(ordered);
3549 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3550 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3551 0, 0, &cached_state, GFP_NOFS);
3553 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3556 unlock_extent_cached(io_tree, page_start, page_end,
3557 &cached_state, GFP_NOFS);
3562 if (offset != PAGE_CACHE_SIZE) {
3564 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3565 flush_dcache_page(page);
3568 ClearPageChecked(page);
3569 set_page_dirty(page);
3570 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3575 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3577 page_cache_release(page);
3583 * This function puts in dummy file extents for the area we're creating a hole
3584 * for. So if we are truncating this file to a larger size we need to insert
3585 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3586 * the range between oldsize and size
3588 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3590 struct btrfs_trans_handle *trans;
3591 struct btrfs_root *root = BTRFS_I(inode)->root;
3592 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3593 struct extent_map *em = NULL;
3594 struct extent_state *cached_state = NULL;
3595 u64 mask = root->sectorsize - 1;
3596 u64 hole_start = (oldsize + mask) & ~mask;
3597 u64 block_end = (size + mask) & ~mask;
3603 if (size <= hole_start)
3607 struct btrfs_ordered_extent *ordered;
3608 btrfs_wait_ordered_range(inode, hole_start,
3609 block_end - hole_start);
3610 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3611 &cached_state, GFP_NOFS);
3612 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3615 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3616 &cached_state, GFP_NOFS);
3617 btrfs_put_ordered_extent(ordered);
3620 cur_offset = hole_start;
3622 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3623 block_end - cur_offset, 0);
3624 BUG_ON(IS_ERR(em) || !em);
3625 last_byte = min(extent_map_end(em), block_end);
3626 last_byte = (last_byte + mask) & ~mask;
3627 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3629 hole_size = last_byte - cur_offset;
3631 trans = btrfs_start_transaction(root, 2);
3632 if (IS_ERR(trans)) {
3633 err = PTR_ERR(trans);
3637 err = btrfs_drop_extents(trans, inode, cur_offset,
3638 cur_offset + hole_size,
3643 err = btrfs_insert_file_extent(trans, root,
3644 inode->i_ino, cur_offset, 0,
3645 0, hole_size, 0, hole_size,
3650 btrfs_drop_extent_cache(inode, hole_start,
3653 btrfs_end_transaction(trans, root);
3655 free_extent_map(em);
3657 cur_offset = last_byte;
3658 if (cur_offset >= block_end)
3662 free_extent_map(em);
3663 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3668 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3670 loff_t oldsize = i_size_read(inode);
3673 if (newsize == oldsize)
3676 if (newsize > oldsize) {
3677 i_size_write(inode, newsize);
3678 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3679 truncate_pagecache(inode, oldsize, newsize);
3680 ret = btrfs_cont_expand(inode, oldsize, newsize);
3682 btrfs_setsize(inode, oldsize);
3686 mark_inode_dirty(inode);
3690 * We're truncating a file that used to have good data down to
3691 * zero. Make sure it gets into the ordered flush list so that
3692 * any new writes get down to disk quickly.
3695 BTRFS_I(inode)->ordered_data_close = 1;
3697 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3698 truncate_setsize(inode, newsize);
3699 ret = btrfs_truncate(inode);
3705 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3707 struct inode *inode = dentry->d_inode;
3708 struct btrfs_root *root = BTRFS_I(inode)->root;
3711 if (btrfs_root_readonly(root))
3714 err = inode_change_ok(inode, attr);
3718 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3719 err = btrfs_setsize(inode, attr->ia_size);
3724 if (attr->ia_valid) {
3725 setattr_copy(inode, attr);
3726 mark_inode_dirty(inode);
3728 if (attr->ia_valid & ATTR_MODE)
3729 err = btrfs_acl_chmod(inode);
3735 void btrfs_evict_inode(struct inode *inode)
3737 struct btrfs_trans_handle *trans;
3738 struct btrfs_root *root = BTRFS_I(inode)->root;
3742 trace_btrfs_inode_evict(inode);
3744 truncate_inode_pages(&inode->i_data, 0);
3745 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3746 root == root->fs_info->tree_root))
3749 if (is_bad_inode(inode)) {
3750 btrfs_orphan_del(NULL, inode);
3753 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3754 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3756 if (root->fs_info->log_root_recovering) {
3757 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3761 if (inode->i_nlink > 0) {
3762 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3766 btrfs_i_size_write(inode, 0);
3769 trans = btrfs_start_transaction(root, 0);
3770 BUG_ON(IS_ERR(trans));
3771 trans->block_rsv = root->orphan_block_rsv;
3773 ret = btrfs_block_rsv_check(trans, root,
3774 root->orphan_block_rsv, 0, 5);
3776 BUG_ON(ret != -EAGAIN);
3777 ret = btrfs_commit_transaction(trans, root);
3782 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3786 nr = trans->blocks_used;
3787 btrfs_end_transaction(trans, root);
3789 btrfs_btree_balance_dirty(root, nr);
3794 ret = btrfs_orphan_del(trans, inode);
3798 nr = trans->blocks_used;
3799 btrfs_end_transaction(trans, root);
3800 btrfs_btree_balance_dirty(root, nr);
3802 end_writeback(inode);
3807 * this returns the key found in the dir entry in the location pointer.
3808 * If no dir entries were found, location->objectid is 0.
3810 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3811 struct btrfs_key *location)
3813 const char *name = dentry->d_name.name;
3814 int namelen = dentry->d_name.len;
3815 struct btrfs_dir_item *di;
3816 struct btrfs_path *path;
3817 struct btrfs_root *root = BTRFS_I(dir)->root;
3820 path = btrfs_alloc_path();
3823 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3828 if (!di || IS_ERR(di))
3831 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3833 btrfs_free_path(path);
3836 location->objectid = 0;
3841 * when we hit a tree root in a directory, the btrfs part of the inode
3842 * needs to be changed to reflect the root directory of the tree root. This
3843 * is kind of like crossing a mount point.
3845 static int fixup_tree_root_location(struct btrfs_root *root,
3847 struct dentry *dentry,
3848 struct btrfs_key *location,
3849 struct btrfs_root **sub_root)
3851 struct btrfs_path *path;
3852 struct btrfs_root *new_root;
3853 struct btrfs_root_ref *ref;
3854 struct extent_buffer *leaf;
3858 path = btrfs_alloc_path();
3865 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3866 BTRFS_I(dir)->root->root_key.objectid,
3867 location->objectid);
3874 leaf = path->nodes[0];
3875 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3876 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3877 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3880 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3881 (unsigned long)(ref + 1),
3882 dentry->d_name.len);
3886 btrfs_release_path(root->fs_info->tree_root, path);
3888 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3889 if (IS_ERR(new_root)) {
3890 err = PTR_ERR(new_root);
3894 if (btrfs_root_refs(&new_root->root_item) == 0) {
3899 *sub_root = new_root;
3900 location->objectid = btrfs_root_dirid(&new_root->root_item);
3901 location->type = BTRFS_INODE_ITEM_KEY;
3902 location->offset = 0;
3905 btrfs_free_path(path);
3909 static void inode_tree_add(struct inode *inode)
3911 struct btrfs_root *root = BTRFS_I(inode)->root;
3912 struct btrfs_inode *entry;
3914 struct rb_node *parent;
3916 p = &root->inode_tree.rb_node;
3919 if (inode_unhashed(inode))
3922 spin_lock(&root->inode_lock);
3925 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3927 if (inode->i_ino < entry->vfs_inode.i_ino)
3928 p = &parent->rb_left;
3929 else if (inode->i_ino > entry->vfs_inode.i_ino)
3930 p = &parent->rb_right;
3932 WARN_ON(!(entry->vfs_inode.i_state &
3933 (I_WILL_FREE | I_FREEING)));
3934 rb_erase(parent, &root->inode_tree);
3935 RB_CLEAR_NODE(parent);
3936 spin_unlock(&root->inode_lock);
3940 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3941 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3942 spin_unlock(&root->inode_lock);
3945 static void inode_tree_del(struct inode *inode)
3947 struct btrfs_root *root = BTRFS_I(inode)->root;
3950 spin_lock(&root->inode_lock);
3951 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3952 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3953 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3954 empty = RB_EMPTY_ROOT(&root->inode_tree);
3956 spin_unlock(&root->inode_lock);
3959 * Free space cache has inodes in the tree root, but the tree root has a
3960 * root_refs of 0, so this could end up dropping the tree root as a
3961 * snapshot, so we need the extra !root->fs_info->tree_root check to
3962 * make sure we don't drop it.
3964 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3965 root != root->fs_info->tree_root) {
3966 synchronize_srcu(&root->fs_info->subvol_srcu);
3967 spin_lock(&root->inode_lock);
3968 empty = RB_EMPTY_ROOT(&root->inode_tree);
3969 spin_unlock(&root->inode_lock);
3971 btrfs_add_dead_root(root);
3975 int btrfs_invalidate_inodes(struct btrfs_root *root)
3977 struct rb_node *node;
3978 struct rb_node *prev;
3979 struct btrfs_inode *entry;
3980 struct inode *inode;
3983 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3985 spin_lock(&root->inode_lock);
3987 node = root->inode_tree.rb_node;
3991 entry = rb_entry(node, struct btrfs_inode, rb_node);
3993 if (objectid < entry->vfs_inode.i_ino)
3994 node = node->rb_left;
3995 else if (objectid > entry->vfs_inode.i_ino)
3996 node = node->rb_right;
4002 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4003 if (objectid <= entry->vfs_inode.i_ino) {
4007 prev = rb_next(prev);
4011 entry = rb_entry(node, struct btrfs_inode, rb_node);
4012 objectid = entry->vfs_inode.i_ino + 1;
4013 inode = igrab(&entry->vfs_inode);
4015 spin_unlock(&root->inode_lock);
4016 if (atomic_read(&inode->i_count) > 1)
4017 d_prune_aliases(inode);
4019 * btrfs_drop_inode will have it removed from
4020 * the inode cache when its usage count
4025 spin_lock(&root->inode_lock);
4029 if (cond_resched_lock(&root->inode_lock))
4032 node = rb_next(node);
4034 spin_unlock(&root->inode_lock);
4038 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4040 struct btrfs_iget_args *args = p;
4041 inode->i_ino = args->ino;
4042 BTRFS_I(inode)->root = args->root;
4043 btrfs_set_inode_space_info(args->root, inode);
4047 static int btrfs_find_actor(struct inode *inode, void *opaque)
4049 struct btrfs_iget_args *args = opaque;
4050 return args->ino == inode->i_ino &&
4051 args->root == BTRFS_I(inode)->root;
4054 static struct inode *btrfs_iget_locked(struct super_block *s,
4056 struct btrfs_root *root)
4058 struct inode *inode;
4059 struct btrfs_iget_args args;
4060 args.ino = objectid;
4063 inode = iget5_locked(s, objectid, btrfs_find_actor,
4064 btrfs_init_locked_inode,
4069 /* Get an inode object given its location and corresponding root.
4070 * Returns in *is_new if the inode was read from disk
4072 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4073 struct btrfs_root *root, int *new)
4075 struct inode *inode;
4077 inode = btrfs_iget_locked(s, location->objectid, root);
4079 return ERR_PTR(-ENOMEM);
4081 if (inode->i_state & I_NEW) {
4082 BTRFS_I(inode)->root = root;
4083 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4084 btrfs_read_locked_inode(inode);
4085 inode_tree_add(inode);
4086 unlock_new_inode(inode);
4094 static struct inode *new_simple_dir(struct super_block *s,
4095 struct btrfs_key *key,
4096 struct btrfs_root *root)
4098 struct inode *inode = new_inode(s);
4101 return ERR_PTR(-ENOMEM);
4103 BTRFS_I(inode)->root = root;
4104 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4105 BTRFS_I(inode)->dummy_inode = 1;
4107 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4108 inode->i_op = &simple_dir_inode_operations;
4109 inode->i_fop = &simple_dir_operations;
4110 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4111 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4116 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4118 struct inode *inode;
4119 struct btrfs_root *root = BTRFS_I(dir)->root;
4120 struct btrfs_root *sub_root = root;
4121 struct btrfs_key location;
4125 if (dentry->d_name.len > BTRFS_NAME_LEN)
4126 return ERR_PTR(-ENAMETOOLONG);
4128 ret = btrfs_inode_by_name(dir, dentry, &location);
4131 return ERR_PTR(ret);
4133 if (location.objectid == 0)
4136 if (location.type == BTRFS_INODE_ITEM_KEY) {
4137 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4141 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4143 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4144 ret = fixup_tree_root_location(root, dir, dentry,
4145 &location, &sub_root);
4148 inode = ERR_PTR(ret);
4150 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4152 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4154 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4156 if (!IS_ERR(inode) && root != sub_root) {
4157 down_read(&root->fs_info->cleanup_work_sem);
4158 if (!(inode->i_sb->s_flags & MS_RDONLY))
4159 ret = btrfs_orphan_cleanup(sub_root);
4160 up_read(&root->fs_info->cleanup_work_sem);
4162 inode = ERR_PTR(ret);
4168 static int btrfs_dentry_delete(const struct dentry *dentry)
4170 struct btrfs_root *root;
4172 if (!dentry->d_inode && !IS_ROOT(dentry))
4173 dentry = dentry->d_parent;
4175 if (dentry->d_inode) {
4176 root = BTRFS_I(dentry->d_inode)->root;
4177 if (btrfs_root_refs(&root->root_item) == 0)
4183 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4184 struct nameidata *nd)
4186 struct inode *inode;
4188 inode = btrfs_lookup_dentry(dir, dentry);
4190 return ERR_CAST(inode);
4192 return d_splice_alias(inode, dentry);
4195 static unsigned char btrfs_filetype_table[] = {
4196 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4199 static int btrfs_real_readdir(struct file *filp, void *dirent,
4202 struct inode *inode = filp->f_dentry->d_inode;
4203 struct btrfs_root *root = BTRFS_I(inode)->root;
4204 struct btrfs_item *item;
4205 struct btrfs_dir_item *di;
4206 struct btrfs_key key;
4207 struct btrfs_key found_key;
4208 struct btrfs_path *path;
4210 struct extent_buffer *leaf;
4212 unsigned char d_type;
4217 int key_type = BTRFS_DIR_INDEX_KEY;
4222 /* FIXME, use a real flag for deciding about the key type */
4223 if (root->fs_info->tree_root == root)
4224 key_type = BTRFS_DIR_ITEM_KEY;
4226 /* special case for "." */
4227 if (filp->f_pos == 0) {
4228 over = filldir(dirent, ".", 1,
4235 /* special case for .., just use the back ref */
4236 if (filp->f_pos == 1) {
4237 u64 pino = parent_ino(filp->f_path.dentry);
4238 over = filldir(dirent, "..", 2,
4244 path = btrfs_alloc_path();
4249 btrfs_set_key_type(&key, key_type);
4250 key.offset = filp->f_pos;
4251 key.objectid = inode->i_ino;
4253 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4258 leaf = path->nodes[0];
4259 slot = path->slots[0];
4260 if (slot >= btrfs_header_nritems(leaf)) {
4261 ret = btrfs_next_leaf(root, path);
4269 item = btrfs_item_nr(leaf, slot);
4270 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4272 if (found_key.objectid != key.objectid)
4274 if (btrfs_key_type(&found_key) != key_type)
4276 if (found_key.offset < filp->f_pos)
4279 filp->f_pos = found_key.offset;
4281 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4283 di_total = btrfs_item_size(leaf, item);
4285 while (di_cur < di_total) {
4286 struct btrfs_key location;
4288 if (verify_dir_item(root, leaf, di))
4291 name_len = btrfs_dir_name_len(leaf, di);
4292 if (name_len <= sizeof(tmp_name)) {
4293 name_ptr = tmp_name;
4295 name_ptr = kmalloc(name_len, GFP_NOFS);
4301 read_extent_buffer(leaf, name_ptr,
4302 (unsigned long)(di + 1), name_len);
4304 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4305 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4307 /* is this a reference to our own snapshot? If so
4310 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4311 location.objectid == root->root_key.objectid) {
4315 over = filldir(dirent, name_ptr, name_len,
4316 found_key.offset, location.objectid,
4320 if (name_ptr != tmp_name)
4325 di_len = btrfs_dir_name_len(leaf, di) +
4326 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4328 di = (struct btrfs_dir_item *)((char *)di + di_len);
4334 /* Reached end of directory/root. Bump pos past the last item. */
4335 if (key_type == BTRFS_DIR_INDEX_KEY)
4337 * 32-bit glibc will use getdents64, but then strtol -
4338 * so the last number we can serve is this.
4340 filp->f_pos = 0x7fffffff;
4346 btrfs_free_path(path);
4350 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4352 struct btrfs_root *root = BTRFS_I(inode)->root;
4353 struct btrfs_trans_handle *trans;
4355 bool nolock = false;
4357 if (BTRFS_I(inode)->dummy_inode)
4361 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4363 if (wbc->sync_mode == WB_SYNC_ALL) {
4365 trans = btrfs_join_transaction_nolock(root);
4367 trans = btrfs_join_transaction(root);
4369 return PTR_ERR(trans);
4371 ret = btrfs_end_transaction_nolock(trans, root);
4373 ret = btrfs_commit_transaction(trans, root);
4379 * This is somewhat expensive, updating the tree every time the
4380 * inode changes. But, it is most likely to find the inode in cache.
4381 * FIXME, needs more benchmarking...there are no reasons other than performance
4382 * to keep or drop this code.
4384 void btrfs_dirty_inode(struct inode *inode)
4386 struct btrfs_root *root = BTRFS_I(inode)->root;
4387 struct btrfs_trans_handle *trans;
4390 if (BTRFS_I(inode)->dummy_inode)
4393 trans = btrfs_join_transaction(root);
4394 BUG_ON(IS_ERR(trans));
4396 ret = btrfs_update_inode(trans, root, inode);
4397 if (ret && ret == -ENOSPC) {
4398 /* whoops, lets try again with the full transaction */
4399 btrfs_end_transaction(trans, root);
4400 trans = btrfs_start_transaction(root, 1);
4401 if (IS_ERR(trans)) {
4402 if (printk_ratelimit()) {
4403 printk(KERN_ERR "btrfs: fail to "
4404 "dirty inode %lu error %ld\n",
4405 inode->i_ino, PTR_ERR(trans));
4410 ret = btrfs_update_inode(trans, root, inode);
4412 if (printk_ratelimit()) {
4413 printk(KERN_ERR "btrfs: fail to "
4414 "dirty inode %lu error %d\n",
4419 btrfs_end_transaction(trans, root);
4423 * find the highest existing sequence number in a directory
4424 * and then set the in-memory index_cnt variable to reflect
4425 * free sequence numbers
4427 static int btrfs_set_inode_index_count(struct inode *inode)
4429 struct btrfs_root *root = BTRFS_I(inode)->root;
4430 struct btrfs_key key, found_key;
4431 struct btrfs_path *path;
4432 struct extent_buffer *leaf;
4435 key.objectid = inode->i_ino;
4436 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4437 key.offset = (u64)-1;
4439 path = btrfs_alloc_path();
4443 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4446 /* FIXME: we should be able to handle this */
4452 * MAGIC NUMBER EXPLANATION:
4453 * since we search a directory based on f_pos we have to start at 2
4454 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4455 * else has to start at 2
4457 if (path->slots[0] == 0) {
4458 BTRFS_I(inode)->index_cnt = 2;
4464 leaf = path->nodes[0];
4465 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4467 if (found_key.objectid != inode->i_ino ||
4468 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4469 BTRFS_I(inode)->index_cnt = 2;
4473 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4475 btrfs_free_path(path);
4480 * helper to find a free sequence number in a given directory. This current
4481 * code is very simple, later versions will do smarter things in the btree
4483 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4487 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4488 ret = btrfs_set_inode_index_count(dir);
4493 *index = BTRFS_I(dir)->index_cnt;
4494 BTRFS_I(dir)->index_cnt++;
4499 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4500 struct btrfs_root *root,
4502 const char *name, int name_len,
4503 u64 ref_objectid, u64 objectid, int mode,
4506 struct inode *inode;
4507 struct btrfs_inode_item *inode_item;
4508 struct btrfs_key *location;
4509 struct btrfs_path *path;
4510 struct btrfs_inode_ref *ref;
4511 struct btrfs_key key[2];
4517 path = btrfs_alloc_path();
4520 inode = new_inode(root->fs_info->sb);
4522 btrfs_free_path(path);
4523 return ERR_PTR(-ENOMEM);
4527 trace_btrfs_inode_request(dir);
4529 ret = btrfs_set_inode_index(dir, index);
4531 btrfs_free_path(path);
4533 return ERR_PTR(ret);
4537 * index_cnt is ignored for everything but a dir,
4538 * btrfs_get_inode_index_count has an explanation for the magic
4541 BTRFS_I(inode)->index_cnt = 2;
4542 BTRFS_I(inode)->root = root;
4543 BTRFS_I(inode)->generation = trans->transid;
4544 inode->i_generation = BTRFS_I(inode)->generation;
4545 btrfs_set_inode_space_info(root, inode);
4552 key[0].objectid = objectid;
4553 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4556 key[1].objectid = objectid;
4557 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4558 key[1].offset = ref_objectid;
4560 sizes[0] = sizeof(struct btrfs_inode_item);
4561 sizes[1] = name_len + sizeof(*ref);
4563 path->leave_spinning = 1;
4564 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4568 inode_init_owner(inode, dir, mode);
4569 inode->i_ino = objectid;
4570 inode_set_bytes(inode, 0);
4571 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4572 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4573 struct btrfs_inode_item);
4574 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4576 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4577 struct btrfs_inode_ref);
4578 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4579 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4580 ptr = (unsigned long)(ref + 1);
4581 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4583 btrfs_mark_buffer_dirty(path->nodes[0]);
4584 btrfs_free_path(path);
4586 location = &BTRFS_I(inode)->location;
4587 location->objectid = objectid;
4588 location->offset = 0;
4589 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4591 btrfs_inherit_iflags(inode, dir);
4593 if ((mode & S_IFREG)) {
4594 if (btrfs_test_opt(root, NODATASUM))
4595 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4596 if (btrfs_test_opt(root, NODATACOW) ||
4597 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4598 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4601 insert_inode_hash(inode);
4602 inode_tree_add(inode);
4604 trace_btrfs_inode_new(inode);
4609 BTRFS_I(dir)->index_cnt--;
4610 btrfs_free_path(path);
4612 return ERR_PTR(ret);
4615 static inline u8 btrfs_inode_type(struct inode *inode)
4617 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4621 * utility function to add 'inode' into 'parent_inode' with
4622 * a give name and a given sequence number.
4623 * if 'add_backref' is true, also insert a backref from the
4624 * inode to the parent directory.
4626 int btrfs_add_link(struct btrfs_trans_handle *trans,
4627 struct inode *parent_inode, struct inode *inode,
4628 const char *name, int name_len, int add_backref, u64 index)
4631 struct btrfs_key key;
4632 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4634 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4635 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4637 key.objectid = inode->i_ino;
4638 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4642 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4643 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4644 key.objectid, root->root_key.objectid,
4645 parent_inode->i_ino,
4646 index, name, name_len);
4647 } else if (add_backref) {
4648 ret = btrfs_insert_inode_ref(trans, root,
4649 name, name_len, inode->i_ino,
4650 parent_inode->i_ino, index);
4654 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4655 parent_inode->i_ino, &key,
4656 btrfs_inode_type(inode), index);
4659 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4661 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4662 ret = btrfs_update_inode(trans, root, parent_inode);
4667 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4668 struct inode *dir, struct dentry *dentry,
4669 struct inode *inode, int backref, u64 index)
4671 int err = btrfs_add_link(trans, dir, inode,
4672 dentry->d_name.name, dentry->d_name.len,
4675 d_instantiate(dentry, inode);
4683 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4684 int mode, dev_t rdev)
4686 struct btrfs_trans_handle *trans;
4687 struct btrfs_root *root = BTRFS_I(dir)->root;
4688 struct inode *inode = NULL;
4692 unsigned long nr = 0;
4695 if (!new_valid_dev(rdev))
4698 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4703 * 2 for inode item and ref
4705 * 1 for xattr if selinux is on
4707 trans = btrfs_start_transaction(root, 5);
4709 return PTR_ERR(trans);
4711 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4712 dentry->d_name.len, dir->i_ino, objectid,
4714 if (IS_ERR(inode)) {
4715 err = PTR_ERR(inode);
4719 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4725 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4729 inode->i_op = &btrfs_special_inode_operations;
4730 init_special_inode(inode, inode->i_mode, rdev);
4731 btrfs_update_inode(trans, root, inode);
4734 nr = trans->blocks_used;
4735 btrfs_end_transaction_throttle(trans, root);
4736 btrfs_btree_balance_dirty(root, nr);
4738 inode_dec_link_count(inode);
4744 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4745 int mode, struct nameidata *nd)
4747 struct btrfs_trans_handle *trans;
4748 struct btrfs_root *root = BTRFS_I(dir)->root;
4749 struct inode *inode = NULL;
4752 unsigned long nr = 0;
4756 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4760 * 2 for inode item and ref
4762 * 1 for xattr if selinux is on
4764 trans = btrfs_start_transaction(root, 5);
4766 return PTR_ERR(trans);
4768 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4769 dentry->d_name.len, dir->i_ino, objectid,
4771 if (IS_ERR(inode)) {
4772 err = PTR_ERR(inode);
4776 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4782 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4786 inode->i_mapping->a_ops = &btrfs_aops;
4787 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4788 inode->i_fop = &btrfs_file_operations;
4789 inode->i_op = &btrfs_file_inode_operations;
4790 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4793 nr = trans->blocks_used;
4794 btrfs_end_transaction_throttle(trans, root);
4796 inode_dec_link_count(inode);
4799 btrfs_btree_balance_dirty(root, nr);
4803 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4804 struct dentry *dentry)
4806 struct btrfs_trans_handle *trans;
4807 struct btrfs_root *root = BTRFS_I(dir)->root;
4808 struct inode *inode = old_dentry->d_inode;
4810 unsigned long nr = 0;
4814 /* do not allow sys_link's with other subvols of the same device */
4815 if (root->objectid != BTRFS_I(inode)->root->objectid)
4818 if (inode->i_nlink == ~0U)
4821 err = btrfs_set_inode_index(dir, &index);
4826 * 2 items for inode and inode ref
4827 * 2 items for dir items
4828 * 1 item for parent inode
4830 trans = btrfs_start_transaction(root, 5);
4831 if (IS_ERR(trans)) {
4832 err = PTR_ERR(trans);
4836 btrfs_inc_nlink(inode);
4837 inode->i_ctime = CURRENT_TIME;
4840 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4845 struct dentry *parent = dget_parent(dentry);
4846 err = btrfs_update_inode(trans, root, inode);
4848 btrfs_log_new_name(trans, inode, NULL, parent);
4852 nr = trans->blocks_used;
4853 btrfs_end_transaction_throttle(trans, root);
4856 inode_dec_link_count(inode);
4859 btrfs_btree_balance_dirty(root, nr);
4863 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4865 struct inode *inode = NULL;
4866 struct btrfs_trans_handle *trans;
4867 struct btrfs_root *root = BTRFS_I(dir)->root;
4869 int drop_on_err = 0;
4872 unsigned long nr = 1;
4874 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4879 * 2 items for inode and ref
4880 * 2 items for dir items
4881 * 1 for xattr if selinux is on
4883 trans = btrfs_start_transaction(root, 5);
4885 return PTR_ERR(trans);
4887 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4888 dentry->d_name.len, dir->i_ino, objectid,
4889 S_IFDIR | mode, &index);
4890 if (IS_ERR(inode)) {
4891 err = PTR_ERR(inode);
4897 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4901 inode->i_op = &btrfs_dir_inode_operations;
4902 inode->i_fop = &btrfs_dir_file_operations;
4904 btrfs_i_size_write(inode, 0);
4905 err = btrfs_update_inode(trans, root, inode);
4909 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4910 dentry->d_name.len, 0, index);
4914 d_instantiate(dentry, inode);
4918 nr = trans->blocks_used;
4919 btrfs_end_transaction_throttle(trans, root);
4922 btrfs_btree_balance_dirty(root, nr);
4926 /* helper for btfs_get_extent. Given an existing extent in the tree,
4927 * and an extent that you want to insert, deal with overlap and insert
4928 * the new extent into the tree.
4930 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4931 struct extent_map *existing,
4932 struct extent_map *em,
4933 u64 map_start, u64 map_len)
4937 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4938 start_diff = map_start - em->start;
4939 em->start = map_start;
4941 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4942 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4943 em->block_start += start_diff;
4944 em->block_len -= start_diff;
4946 return add_extent_mapping(em_tree, em);
4949 static noinline int uncompress_inline(struct btrfs_path *path,
4950 struct inode *inode, struct page *page,
4951 size_t pg_offset, u64 extent_offset,
4952 struct btrfs_file_extent_item *item)
4955 struct extent_buffer *leaf = path->nodes[0];
4958 unsigned long inline_size;
4962 WARN_ON(pg_offset != 0);
4963 compress_type = btrfs_file_extent_compression(leaf, item);
4964 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4965 inline_size = btrfs_file_extent_inline_item_len(leaf,
4966 btrfs_item_nr(leaf, path->slots[0]));
4967 tmp = kmalloc(inline_size, GFP_NOFS);
4970 ptr = btrfs_file_extent_inline_start(item);
4972 read_extent_buffer(leaf, tmp, ptr, inline_size);
4974 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4975 ret = btrfs_decompress(compress_type, tmp, page,
4976 extent_offset, inline_size, max_size);
4978 char *kaddr = kmap_atomic(page, KM_USER0);
4979 unsigned long copy_size = min_t(u64,
4980 PAGE_CACHE_SIZE - pg_offset,
4981 max_size - extent_offset);
4982 memset(kaddr + pg_offset, 0, copy_size);
4983 kunmap_atomic(kaddr, KM_USER0);
4990 * a bit scary, this does extent mapping from logical file offset to the disk.
4991 * the ugly parts come from merging extents from the disk with the in-ram
4992 * representation. This gets more complex because of the data=ordered code,
4993 * where the in-ram extents might be locked pending data=ordered completion.
4995 * This also copies inline extents directly into the page.
4998 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4999 size_t pg_offset, u64 start, u64 len,
5005 u64 extent_start = 0;
5007 u64 objectid = inode->i_ino;
5009 struct btrfs_path *path = NULL;
5010 struct btrfs_root *root = BTRFS_I(inode)->root;
5011 struct btrfs_file_extent_item *item;
5012 struct extent_buffer *leaf;
5013 struct btrfs_key found_key;
5014 struct extent_map *em = NULL;
5015 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5016 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5017 struct btrfs_trans_handle *trans = NULL;
5021 read_lock(&em_tree->lock);
5022 em = lookup_extent_mapping(em_tree, start, len);
5024 em->bdev = root->fs_info->fs_devices->latest_bdev;
5025 read_unlock(&em_tree->lock);
5028 if (em->start > start || em->start + em->len <= start)
5029 free_extent_map(em);
5030 else if (em->block_start == EXTENT_MAP_INLINE && page)
5031 free_extent_map(em);
5035 em = alloc_extent_map(GFP_NOFS);
5040 em->bdev = root->fs_info->fs_devices->latest_bdev;
5041 em->start = EXTENT_MAP_HOLE;
5042 em->orig_start = EXTENT_MAP_HOLE;
5044 em->block_len = (u64)-1;
5047 path = btrfs_alloc_path();
5053 * Chances are we'll be called again, so go ahead and do
5059 ret = btrfs_lookup_file_extent(trans, root, path,
5060 objectid, start, trans != NULL);
5067 if (path->slots[0] == 0)
5072 leaf = path->nodes[0];
5073 item = btrfs_item_ptr(leaf, path->slots[0],
5074 struct btrfs_file_extent_item);
5075 /* are we inside the extent that was found? */
5076 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5077 found_type = btrfs_key_type(&found_key);
5078 if (found_key.objectid != objectid ||
5079 found_type != BTRFS_EXTENT_DATA_KEY) {
5083 found_type = btrfs_file_extent_type(leaf, item);
5084 extent_start = found_key.offset;
5085 compress_type = btrfs_file_extent_compression(leaf, item);
5086 if (found_type == BTRFS_FILE_EXTENT_REG ||
5087 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5088 extent_end = extent_start +
5089 btrfs_file_extent_num_bytes(leaf, item);
5090 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5092 size = btrfs_file_extent_inline_len(leaf, item);
5093 extent_end = (extent_start + size + root->sectorsize - 1) &
5094 ~((u64)root->sectorsize - 1);
5097 if (start >= extent_end) {
5099 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5100 ret = btrfs_next_leaf(root, path);
5107 leaf = path->nodes[0];
5109 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5110 if (found_key.objectid != objectid ||
5111 found_key.type != BTRFS_EXTENT_DATA_KEY)
5113 if (start + len <= found_key.offset)
5116 em->len = found_key.offset - start;
5120 if (found_type == BTRFS_FILE_EXTENT_REG ||
5121 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5122 em->start = extent_start;
5123 em->len = extent_end - extent_start;
5124 em->orig_start = extent_start -
5125 btrfs_file_extent_offset(leaf, item);
5126 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5128 em->block_start = EXTENT_MAP_HOLE;
5131 if (compress_type != BTRFS_COMPRESS_NONE) {
5132 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5133 em->compress_type = compress_type;
5134 em->block_start = bytenr;
5135 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5138 bytenr += btrfs_file_extent_offset(leaf, item);
5139 em->block_start = bytenr;
5140 em->block_len = em->len;
5141 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5142 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5145 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5149 size_t extent_offset;
5152 em->block_start = EXTENT_MAP_INLINE;
5153 if (!page || create) {
5154 em->start = extent_start;
5155 em->len = extent_end - extent_start;
5159 size = btrfs_file_extent_inline_len(leaf, item);
5160 extent_offset = page_offset(page) + pg_offset - extent_start;
5161 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5162 size - extent_offset);
5163 em->start = extent_start + extent_offset;
5164 em->len = (copy_size + root->sectorsize - 1) &
5165 ~((u64)root->sectorsize - 1);
5166 em->orig_start = EXTENT_MAP_INLINE;
5167 if (compress_type) {
5168 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5169 em->compress_type = compress_type;
5171 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5172 if (create == 0 && !PageUptodate(page)) {
5173 if (btrfs_file_extent_compression(leaf, item) !=
5174 BTRFS_COMPRESS_NONE) {
5175 ret = uncompress_inline(path, inode, page,
5177 extent_offset, item);
5181 read_extent_buffer(leaf, map + pg_offset, ptr,
5183 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5184 memset(map + pg_offset + copy_size, 0,
5185 PAGE_CACHE_SIZE - pg_offset -
5190 flush_dcache_page(page);
5191 } else if (create && PageUptodate(page)) {
5195 free_extent_map(em);
5197 btrfs_release_path(root, path);
5198 trans = btrfs_join_transaction(root);
5200 return ERR_CAST(trans);
5204 write_extent_buffer(leaf, map + pg_offset, ptr,
5207 btrfs_mark_buffer_dirty(leaf);
5209 set_extent_uptodate(io_tree, em->start,
5210 extent_map_end(em) - 1, NULL, GFP_NOFS);
5213 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5220 em->block_start = EXTENT_MAP_HOLE;
5221 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5223 btrfs_release_path(root, path);
5224 if (em->start > start || extent_map_end(em) <= start) {
5225 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5226 "[%llu %llu]\n", (unsigned long long)em->start,
5227 (unsigned long long)em->len,
5228 (unsigned long long)start,
5229 (unsigned long long)len);
5235 write_lock(&em_tree->lock);
5236 ret = add_extent_mapping(em_tree, em);
5237 /* it is possible that someone inserted the extent into the tree
5238 * while we had the lock dropped. It is also possible that
5239 * an overlapping map exists in the tree
5241 if (ret == -EEXIST) {
5242 struct extent_map *existing;
5246 existing = lookup_extent_mapping(em_tree, start, len);
5247 if (existing && (existing->start > start ||
5248 existing->start + existing->len <= start)) {
5249 free_extent_map(existing);
5253 existing = lookup_extent_mapping(em_tree, em->start,
5256 err = merge_extent_mapping(em_tree, existing,
5259 free_extent_map(existing);
5261 free_extent_map(em);
5266 free_extent_map(em);
5270 free_extent_map(em);
5275 write_unlock(&em_tree->lock);
5278 trace_btrfs_get_extent(root, em);
5281 btrfs_free_path(path);
5283 ret = btrfs_end_transaction(trans, root);
5288 free_extent_map(em);
5289 return ERR_PTR(err);
5294 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5295 size_t pg_offset, u64 start, u64 len,
5298 struct extent_map *em;
5299 struct extent_map *hole_em = NULL;
5300 u64 range_start = start;
5306 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5311 * if our em maps to a hole, there might
5312 * actually be delalloc bytes behind it
5314 if (em->block_start != EXTENT_MAP_HOLE)
5320 /* check to see if we've wrapped (len == -1 or similar) */
5329 /* ok, we didn't find anything, lets look for delalloc */
5330 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5331 end, len, EXTENT_DELALLOC, 1);
5332 found_end = range_start + found;
5333 if (found_end < range_start)
5334 found_end = (u64)-1;
5337 * we didn't find anything useful, return
5338 * the original results from get_extent()
5340 if (range_start > end || found_end <= start) {
5346 /* adjust the range_start to make sure it doesn't
5347 * go backwards from the start they passed in
5349 range_start = max(start,range_start);
5350 found = found_end - range_start;
5353 u64 hole_start = start;
5356 em = alloc_extent_map(GFP_NOFS);
5362 * when btrfs_get_extent can't find anything it
5363 * returns one huge hole
5365 * make sure what it found really fits our range, and
5366 * adjust to make sure it is based on the start from
5370 u64 calc_end = extent_map_end(hole_em);
5372 if (calc_end <= start || (hole_em->start > end)) {
5373 free_extent_map(hole_em);
5376 hole_start = max(hole_em->start, start);
5377 hole_len = calc_end - hole_start;
5381 if (hole_em && range_start > hole_start) {
5382 /* our hole starts before our delalloc, so we
5383 * have to return just the parts of the hole
5384 * that go until the delalloc starts
5386 em->len = min(hole_len,
5387 range_start - hole_start);
5388 em->start = hole_start;
5389 em->orig_start = hole_start;
5391 * don't adjust block start at all,
5392 * it is fixed at EXTENT_MAP_HOLE
5394 em->block_start = hole_em->block_start;
5395 em->block_len = hole_len;
5397 em->start = range_start;
5399 em->orig_start = range_start;
5400 em->block_start = EXTENT_MAP_DELALLOC;
5401 em->block_len = found;
5403 } else if (hole_em) {
5408 free_extent_map(hole_em);
5410 free_extent_map(em);
5411 return ERR_PTR(err);
5416 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5417 struct extent_map *em,
5420 struct btrfs_root *root = BTRFS_I(inode)->root;
5421 struct btrfs_trans_handle *trans;
5422 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5423 struct btrfs_key ins;
5426 bool insert = false;
5429 * Ok if the extent map we looked up is a hole and is for the exact
5430 * range we want, there is no reason to allocate a new one, however if
5431 * it is not right then we need to free this one and drop the cache for
5434 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5436 free_extent_map(em);
5439 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5442 trans = btrfs_join_transaction(root);
5444 return ERR_CAST(trans);
5446 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5448 alloc_hint = get_extent_allocation_hint(inode, start, len);
5449 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5450 alloc_hint, (u64)-1, &ins, 1);
5457 em = alloc_extent_map(GFP_NOFS);
5459 em = ERR_PTR(-ENOMEM);
5465 em->orig_start = em->start;
5466 em->len = ins.offset;
5468 em->block_start = ins.objectid;
5469 em->block_len = ins.offset;
5470 em->bdev = root->fs_info->fs_devices->latest_bdev;
5473 * We need to do this because if we're using the original em we searched
5474 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5477 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5480 write_lock(&em_tree->lock);
5481 ret = add_extent_mapping(em_tree, em);
5482 write_unlock(&em_tree->lock);
5485 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5488 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5489 ins.offset, ins.offset, 0);
5491 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5495 btrfs_end_transaction(trans, root);
5500 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5501 * block must be cow'd
5503 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5504 struct inode *inode, u64 offset, u64 len)
5506 struct btrfs_path *path;
5508 struct extent_buffer *leaf;
5509 struct btrfs_root *root = BTRFS_I(inode)->root;
5510 struct btrfs_file_extent_item *fi;
5511 struct btrfs_key key;
5519 path = btrfs_alloc_path();
5523 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5528 slot = path->slots[0];
5531 /* can't find the item, must cow */
5538 leaf = path->nodes[0];
5539 btrfs_item_key_to_cpu(leaf, &key, slot);
5540 if (key.objectid != inode->i_ino ||
5541 key.type != BTRFS_EXTENT_DATA_KEY) {
5542 /* not our file or wrong item type, must cow */
5546 if (key.offset > offset) {
5547 /* Wrong offset, must cow */
5551 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5552 found_type = btrfs_file_extent_type(leaf, fi);
5553 if (found_type != BTRFS_FILE_EXTENT_REG &&
5554 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5555 /* not a regular extent, must cow */
5558 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5559 backref_offset = btrfs_file_extent_offset(leaf, fi);
5561 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5562 if (extent_end < offset + len) {
5563 /* extent doesn't include our full range, must cow */
5567 if (btrfs_extent_readonly(root, disk_bytenr))
5571 * look for other files referencing this extent, if we
5572 * find any we must cow
5574 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5575 key.offset - backref_offset, disk_bytenr))
5579 * adjust disk_bytenr and num_bytes to cover just the bytes
5580 * in this extent we are about to write. If there
5581 * are any csums in that range we have to cow in order
5582 * to keep the csums correct
5584 disk_bytenr += backref_offset;
5585 disk_bytenr += offset - key.offset;
5586 num_bytes = min(offset + len, extent_end) - offset;
5587 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5590 * all of the above have passed, it is safe to overwrite this extent
5595 btrfs_free_path(path);
5599 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5600 struct buffer_head *bh_result, int create)
5602 struct extent_map *em;
5603 struct btrfs_root *root = BTRFS_I(inode)->root;
5604 u64 start = iblock << inode->i_blkbits;
5605 u64 len = bh_result->b_size;
5606 struct btrfs_trans_handle *trans;
5608 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5613 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5614 * io. INLINE is special, and we could probably kludge it in here, but
5615 * it's still buffered so for safety lets just fall back to the generic
5618 * For COMPRESSED we _have_ to read the entire extent in so we can
5619 * decompress it, so there will be buffering required no matter what we
5620 * do, so go ahead and fallback to buffered.
5622 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5623 * to buffered IO. Don't blame me, this is the price we pay for using
5626 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5627 em->block_start == EXTENT_MAP_INLINE) {
5628 free_extent_map(em);
5632 /* Just a good old fashioned hole, return */
5633 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5634 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5635 free_extent_map(em);
5636 /* DIO will do one hole at a time, so just unlock a sector */
5637 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5638 start + root->sectorsize - 1, GFP_NOFS);
5643 * We don't allocate a new extent in the following cases
5645 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5647 * 2) The extent is marked as PREALLOC. We're good to go here and can
5648 * just use the extent.
5652 len = em->len - (start - em->start);
5656 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5657 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5658 em->block_start != EXTENT_MAP_HOLE)) {
5663 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5664 type = BTRFS_ORDERED_PREALLOC;
5666 type = BTRFS_ORDERED_NOCOW;
5667 len = min(len, em->len - (start - em->start));
5668 block_start = em->block_start + (start - em->start);
5671 * we're not going to log anything, but we do need
5672 * to make sure the current transaction stays open
5673 * while we look for nocow cross refs
5675 trans = btrfs_join_transaction(root);
5679 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5680 ret = btrfs_add_ordered_extent_dio(inode, start,
5681 block_start, len, len, type);
5682 btrfs_end_transaction(trans, root);
5684 free_extent_map(em);
5689 btrfs_end_transaction(trans, root);
5693 * this will cow the extent, reset the len in case we changed
5696 len = bh_result->b_size;
5697 em = btrfs_new_extent_direct(inode, em, start, len);
5700 len = min(len, em->len - (start - em->start));
5702 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5703 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5706 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5708 bh_result->b_size = len;
5709 bh_result->b_bdev = em->bdev;
5710 set_buffer_mapped(bh_result);
5711 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5712 set_buffer_new(bh_result);
5714 free_extent_map(em);
5719 struct btrfs_dio_private {
5720 struct inode *inode;
5727 /* number of bios pending for this dio */
5728 atomic_t pending_bios;
5733 struct bio *orig_bio;
5736 static void btrfs_endio_direct_read(struct bio *bio, int err)
5738 struct btrfs_dio_private *dip = bio->bi_private;
5739 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5740 struct bio_vec *bvec = bio->bi_io_vec;
5741 struct inode *inode = dip->inode;
5742 struct btrfs_root *root = BTRFS_I(inode)->root;
5744 u32 *private = dip->csums;
5746 start = dip->logical_offset;
5748 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5749 struct page *page = bvec->bv_page;
5752 unsigned long flags;
5754 local_irq_save(flags);
5755 kaddr = kmap_atomic(page, KM_IRQ0);
5756 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5757 csum, bvec->bv_len);
5758 btrfs_csum_final(csum, (char *)&csum);
5759 kunmap_atomic(kaddr, KM_IRQ0);
5760 local_irq_restore(flags);
5762 flush_dcache_page(bvec->bv_page);
5763 if (csum != *private) {
5764 printk(KERN_ERR "btrfs csum failed ino %lu off"
5765 " %llu csum %u private %u\n",
5766 inode->i_ino, (unsigned long long)start,
5772 start += bvec->bv_len;
5775 } while (bvec <= bvec_end);
5777 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5778 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5779 bio->bi_private = dip->private;
5784 /* If we had a csum failure make sure to clear the uptodate flag */
5786 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5787 dio_end_io(bio, err);
5790 static void btrfs_endio_direct_write(struct bio *bio, int err)
5792 struct btrfs_dio_private *dip = bio->bi_private;
5793 struct inode *inode = dip->inode;
5794 struct btrfs_root *root = BTRFS_I(inode)->root;
5795 struct btrfs_trans_handle *trans;
5796 struct btrfs_ordered_extent *ordered = NULL;
5797 struct extent_state *cached_state = NULL;
5798 u64 ordered_offset = dip->logical_offset;
5799 u64 ordered_bytes = dip->bytes;
5805 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5813 trans = btrfs_join_transaction(root);
5814 if (IS_ERR(trans)) {
5818 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5820 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5821 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5823 ret = btrfs_update_inode(trans, root, inode);
5828 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5829 ordered->file_offset + ordered->len - 1, 0,
5830 &cached_state, GFP_NOFS);
5832 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5833 ret = btrfs_mark_extent_written(trans, inode,
5834 ordered->file_offset,
5835 ordered->file_offset +
5842 ret = insert_reserved_file_extent(trans, inode,
5843 ordered->file_offset,
5849 BTRFS_FILE_EXTENT_REG);
5850 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5851 ordered->file_offset, ordered->len);
5859 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5860 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5862 btrfs_update_inode(trans, root, inode);
5865 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5866 ordered->file_offset + ordered->len - 1,
5867 &cached_state, GFP_NOFS);
5869 btrfs_delalloc_release_metadata(inode, ordered->len);
5870 btrfs_end_transaction(trans, root);
5871 ordered_offset = ordered->file_offset + ordered->len;
5872 btrfs_put_ordered_extent(ordered);
5873 btrfs_put_ordered_extent(ordered);
5877 * our bio might span multiple ordered extents. If we haven't
5878 * completed the accounting for the whole dio, go back and try again
5880 if (ordered_offset < dip->logical_offset + dip->bytes) {
5881 ordered_bytes = dip->logical_offset + dip->bytes -
5886 bio->bi_private = dip->private;
5891 /* If we had an error make sure to clear the uptodate flag */
5893 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5894 dio_end_io(bio, err);
5897 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5898 struct bio *bio, int mirror_num,
5899 unsigned long bio_flags, u64 offset)
5902 struct btrfs_root *root = BTRFS_I(inode)->root;
5903 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5908 static void btrfs_end_dio_bio(struct bio *bio, int err)
5910 struct btrfs_dio_private *dip = bio->bi_private;
5913 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5914 "sector %#Lx len %u err no %d\n",
5915 dip->inode->i_ino, bio->bi_rw,
5916 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5920 * before atomic variable goto zero, we must make sure
5921 * dip->errors is perceived to be set.
5923 smp_mb__before_atomic_dec();
5926 /* if there are more bios still pending for this dio, just exit */
5927 if (!atomic_dec_and_test(&dip->pending_bios))
5931 bio_io_error(dip->orig_bio);
5933 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5934 bio_endio(dip->orig_bio, 0);
5940 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5941 u64 first_sector, gfp_t gfp_flags)
5943 int nr_vecs = bio_get_nr_vecs(bdev);
5944 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5947 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5948 int rw, u64 file_offset, int skip_sum,
5949 u32 *csums, int async_submit)
5951 int write = rw & REQ_WRITE;
5952 struct btrfs_root *root = BTRFS_I(inode)->root;
5956 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5963 if (write && async_submit) {
5964 ret = btrfs_wq_submit_bio(root->fs_info,
5965 inode, rw, bio, 0, 0,
5967 __btrfs_submit_bio_start_direct_io,
5968 __btrfs_submit_bio_done);
5972 * If we aren't doing async submit, calculate the csum of the
5975 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5978 } else if (!skip_sum) {
5979 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5980 file_offset, csums);
5986 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5992 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5995 struct inode *inode = dip->inode;
5996 struct btrfs_root *root = BTRFS_I(inode)->root;
5997 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5999 struct bio *orig_bio = dip->orig_bio;
6000 struct bio_vec *bvec = orig_bio->bi_io_vec;
6001 u64 start_sector = orig_bio->bi_sector;
6002 u64 file_offset = dip->logical_offset;
6006 u32 *csums = dip->csums;
6008 int async_submit = 0;
6009 int write = rw & REQ_WRITE;
6011 map_length = orig_bio->bi_size;
6012 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6013 &map_length, NULL, 0);
6019 if (map_length >= orig_bio->bi_size) {
6025 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6028 bio->bi_private = dip;
6029 bio->bi_end_io = btrfs_end_dio_bio;
6030 atomic_inc(&dip->pending_bios);
6032 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6033 if (unlikely(map_length < submit_len + bvec->bv_len ||
6034 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6035 bvec->bv_offset) < bvec->bv_len)) {
6037 * inc the count before we submit the bio so
6038 * we know the end IO handler won't happen before
6039 * we inc the count. Otherwise, the dip might get freed
6040 * before we're done setting it up
6042 atomic_inc(&dip->pending_bios);
6043 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6044 file_offset, skip_sum,
6045 csums, async_submit);
6048 atomic_dec(&dip->pending_bios);
6052 /* Write's use the ordered csums */
6053 if (!write && !skip_sum)
6054 csums = csums + nr_pages;
6055 start_sector += submit_len >> 9;
6056 file_offset += submit_len;
6061 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6062 start_sector, GFP_NOFS);
6065 bio->bi_private = dip;
6066 bio->bi_end_io = btrfs_end_dio_bio;
6068 map_length = orig_bio->bi_size;
6069 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6070 &map_length, NULL, 0);
6076 submit_len += bvec->bv_len;
6083 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6084 csums, async_submit);
6092 * before atomic variable goto zero, we must
6093 * make sure dip->errors is perceived to be set.
6095 smp_mb__before_atomic_dec();
6096 if (atomic_dec_and_test(&dip->pending_bios))
6097 bio_io_error(dip->orig_bio);
6099 /* bio_end_io() will handle error, so we needn't return it */
6103 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6106 struct btrfs_root *root = BTRFS_I(inode)->root;
6107 struct btrfs_dio_private *dip;
6108 struct bio_vec *bvec = bio->bi_io_vec;
6110 int write = rw & REQ_WRITE;
6113 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6115 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6122 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6123 if (!write && !skip_sum) {
6124 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6132 dip->private = bio->bi_private;
6134 dip->logical_offset = file_offset;
6138 dip->bytes += bvec->bv_len;
6140 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6142 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6143 bio->bi_private = dip;
6145 dip->orig_bio = bio;
6146 atomic_set(&dip->pending_bios, 0);
6149 bio->bi_end_io = btrfs_endio_direct_write;
6151 bio->bi_end_io = btrfs_endio_direct_read;
6153 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6158 * If this is a write, we need to clean up the reserved space and kill
6159 * the ordered extent.
6162 struct btrfs_ordered_extent *ordered;
6163 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6164 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6165 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6166 btrfs_free_reserved_extent(root, ordered->start,
6168 btrfs_put_ordered_extent(ordered);
6169 btrfs_put_ordered_extent(ordered);
6171 bio_endio(bio, ret);
6174 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6175 const struct iovec *iov, loff_t offset,
6176 unsigned long nr_segs)
6182 unsigned blocksize_mask = root->sectorsize - 1;
6183 ssize_t retval = -EINVAL;
6184 loff_t end = offset;
6186 if (offset & blocksize_mask)
6189 /* Check the memory alignment. Blocks cannot straddle pages */
6190 for (seg = 0; seg < nr_segs; seg++) {
6191 addr = (unsigned long)iov[seg].iov_base;
6192 size = iov[seg].iov_len;
6194 if ((addr & blocksize_mask) || (size & blocksize_mask))
6197 /* If this is a write we don't need to check anymore */
6202 * Check to make sure we don't have duplicate iov_base's in this
6203 * iovec, if so return EINVAL, otherwise we'll get csum errors
6204 * when reading back.
6206 for (i = seg + 1; i < nr_segs; i++) {
6207 if (iov[seg].iov_base == iov[i].iov_base)
6215 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6216 const struct iovec *iov, loff_t offset,
6217 unsigned long nr_segs)
6219 struct file *file = iocb->ki_filp;
6220 struct inode *inode = file->f_mapping->host;
6221 struct btrfs_ordered_extent *ordered;
6222 struct extent_state *cached_state = NULL;
6223 u64 lockstart, lockend;
6225 int writing = rw & WRITE;
6227 size_t count = iov_length(iov, nr_segs);
6229 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6235 lockend = offset + count - 1;
6238 ret = btrfs_delalloc_reserve_space(inode, count);
6244 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6245 0, &cached_state, GFP_NOFS);
6247 * We're concerned with the entire range that we're going to be
6248 * doing DIO to, so we need to make sure theres no ordered
6249 * extents in this range.
6251 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6252 lockend - lockstart + 1);
6255 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6256 &cached_state, GFP_NOFS);
6257 btrfs_start_ordered_extent(inode, ordered, 1);
6258 btrfs_put_ordered_extent(ordered);
6263 * we don't use btrfs_set_extent_delalloc because we don't want
6264 * the dirty or uptodate bits
6267 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6268 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6269 EXTENT_DELALLOC, 0, NULL, &cached_state,
6272 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6273 lockend, EXTENT_LOCKED | write_bits,
6274 1, 0, &cached_state, GFP_NOFS);
6279 free_extent_state(cached_state);
6280 cached_state = NULL;
6282 ret = __blockdev_direct_IO(rw, iocb, inode,
6283 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6284 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6285 btrfs_submit_direct, 0);
6287 if (ret < 0 && ret != -EIOCBQUEUED) {
6288 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6289 offset + iov_length(iov, nr_segs) - 1,
6290 EXTENT_LOCKED | write_bits, 1, 0,
6291 &cached_state, GFP_NOFS);
6292 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6294 * We're falling back to buffered, unlock the section we didn't
6297 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6298 offset + iov_length(iov, nr_segs) - 1,
6299 EXTENT_LOCKED | write_bits, 1, 0,
6300 &cached_state, GFP_NOFS);
6303 free_extent_state(cached_state);
6307 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6308 __u64 start, __u64 len)
6310 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6313 int btrfs_readpage(struct file *file, struct page *page)
6315 struct extent_io_tree *tree;
6316 tree = &BTRFS_I(page->mapping->host)->io_tree;
6317 return extent_read_full_page(tree, page, btrfs_get_extent);
6320 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6322 struct extent_io_tree *tree;
6325 if (current->flags & PF_MEMALLOC) {
6326 redirty_page_for_writepage(wbc, page);
6330 tree = &BTRFS_I(page->mapping->host)->io_tree;
6331 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6334 int btrfs_writepages(struct address_space *mapping,
6335 struct writeback_control *wbc)
6337 struct extent_io_tree *tree;
6339 tree = &BTRFS_I(mapping->host)->io_tree;
6340 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6344 btrfs_readpages(struct file *file, struct address_space *mapping,
6345 struct list_head *pages, unsigned nr_pages)
6347 struct extent_io_tree *tree;
6348 tree = &BTRFS_I(mapping->host)->io_tree;
6349 return extent_readpages(tree, mapping, pages, nr_pages,
6352 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6354 struct extent_io_tree *tree;
6355 struct extent_map_tree *map;
6358 tree = &BTRFS_I(page->mapping->host)->io_tree;
6359 map = &BTRFS_I(page->mapping->host)->extent_tree;
6360 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6362 ClearPagePrivate(page);
6363 set_page_private(page, 0);
6364 page_cache_release(page);
6369 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6371 if (PageWriteback(page) || PageDirty(page))
6373 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6376 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6378 struct extent_io_tree *tree;
6379 struct btrfs_ordered_extent *ordered;
6380 struct extent_state *cached_state = NULL;
6381 u64 page_start = page_offset(page);
6382 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6386 * we have the page locked, so new writeback can't start,
6387 * and the dirty bit won't be cleared while we are here.
6389 * Wait for IO on this page so that we can safely clear
6390 * the PagePrivate2 bit and do ordered accounting
6392 wait_on_page_writeback(page);
6394 tree = &BTRFS_I(page->mapping->host)->io_tree;
6396 btrfs_releasepage(page, GFP_NOFS);
6399 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6401 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6405 * IO on this page will never be started, so we need
6406 * to account for any ordered extents now
6408 clear_extent_bit(tree, page_start, page_end,
6409 EXTENT_DIRTY | EXTENT_DELALLOC |
6410 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6411 &cached_state, GFP_NOFS);
6413 * whoever cleared the private bit is responsible
6414 * for the finish_ordered_io
6416 if (TestClearPagePrivate2(page)) {
6417 btrfs_finish_ordered_io(page->mapping->host,
6418 page_start, page_end);
6420 btrfs_put_ordered_extent(ordered);
6421 cached_state = NULL;
6422 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6425 clear_extent_bit(tree, page_start, page_end,
6426 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6427 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6428 __btrfs_releasepage(page, GFP_NOFS);
6430 ClearPageChecked(page);
6431 if (PagePrivate(page)) {
6432 ClearPagePrivate(page);
6433 set_page_private(page, 0);
6434 page_cache_release(page);
6439 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6440 * called from a page fault handler when a page is first dirtied. Hence we must
6441 * be careful to check for EOF conditions here. We set the page up correctly
6442 * for a written page which means we get ENOSPC checking when writing into
6443 * holes and correct delalloc and unwritten extent mapping on filesystems that
6444 * support these features.
6446 * We are not allowed to take the i_mutex here so we have to play games to
6447 * protect against truncate races as the page could now be beyond EOF. Because
6448 * vmtruncate() writes the inode size before removing pages, once we have the
6449 * page lock we can determine safely if the page is beyond EOF. If it is not
6450 * beyond EOF, then the page is guaranteed safe against truncation until we
6453 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6455 struct page *page = vmf->page;
6456 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6457 struct btrfs_root *root = BTRFS_I(inode)->root;
6458 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6459 struct btrfs_ordered_extent *ordered;
6460 struct extent_state *cached_state = NULL;
6462 unsigned long zero_start;
6468 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6472 else /* -ENOSPC, -EIO, etc */
6473 ret = VM_FAULT_SIGBUS;
6477 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6480 size = i_size_read(inode);
6481 page_start = page_offset(page);
6482 page_end = page_start + PAGE_CACHE_SIZE - 1;
6484 if ((page->mapping != inode->i_mapping) ||
6485 (page_start >= size)) {
6486 /* page got truncated out from underneath us */
6489 wait_on_page_writeback(page);
6491 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6493 set_page_extent_mapped(page);
6496 * we can't set the delalloc bits if there are pending ordered
6497 * extents. Drop our locks and wait for them to finish
6499 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6501 unlock_extent_cached(io_tree, page_start, page_end,
6502 &cached_state, GFP_NOFS);
6504 btrfs_start_ordered_extent(inode, ordered, 1);
6505 btrfs_put_ordered_extent(ordered);
6510 * XXX - page_mkwrite gets called every time the page is dirtied, even
6511 * if it was already dirty, so for space accounting reasons we need to
6512 * clear any delalloc bits for the range we are fixing to save. There
6513 * is probably a better way to do this, but for now keep consistent with
6514 * prepare_pages in the normal write path.
6516 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6517 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6518 0, 0, &cached_state, GFP_NOFS);
6520 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6523 unlock_extent_cached(io_tree, page_start, page_end,
6524 &cached_state, GFP_NOFS);
6525 ret = VM_FAULT_SIGBUS;
6530 /* page is wholly or partially inside EOF */
6531 if (page_start + PAGE_CACHE_SIZE > size)
6532 zero_start = size & ~PAGE_CACHE_MASK;
6534 zero_start = PAGE_CACHE_SIZE;
6536 if (zero_start != PAGE_CACHE_SIZE) {
6538 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6539 flush_dcache_page(page);
6542 ClearPageChecked(page);
6543 set_page_dirty(page);
6544 SetPageUptodate(page);
6546 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6547 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6549 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6553 return VM_FAULT_LOCKED;
6555 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6560 static int btrfs_truncate(struct inode *inode)
6562 struct btrfs_root *root = BTRFS_I(inode)->root;
6563 struct btrfs_block_rsv *rsv;
6566 struct btrfs_trans_handle *trans;
6568 u64 mask = root->sectorsize - 1;
6570 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6574 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6575 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6578 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6579 * 3 things going on here
6581 * 1) We need to reserve space for our orphan item and the space to
6582 * delete our orphan item. Lord knows we don't want to have a dangling
6583 * orphan item because we didn't reserve space to remove it.
6585 * 2) We need to reserve space to update our inode.
6587 * 3) We need to have something to cache all the space that is going to
6588 * be free'd up by the truncate operation, but also have some slack
6589 * space reserved in case it uses space during the truncate (thank you
6590 * very much snapshotting).
6592 * And we need these to all be seperate. The fact is we can use alot of
6593 * space doing the truncate, and we have no earthly idea how much space
6594 * we will use, so we need the truncate reservation to be seperate so it
6595 * doesn't end up using space reserved for updating the inode or
6596 * removing the orphan item. We also need to be able to stop the
6597 * transaction and start a new one, which means we need to be able to
6598 * update the inode several times, and we have no idea of knowing how
6599 * many times that will be, so we can't just reserve 1 item for the
6600 * entirety of the opration, so that has to be done seperately as well.
6601 * Then there is the orphan item, which does indeed need to be held on
6602 * to for the whole operation, and we need nobody to touch this reserved
6603 * space except the orphan code.
6605 * So that leaves us with
6607 * 1) root->orphan_block_rsv - for the orphan deletion.
6608 * 2) rsv - for the truncate reservation, which we will steal from the
6609 * transaction reservation.
6610 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6611 * updating the inode.
6613 rsv = btrfs_alloc_block_rsv(root);
6616 btrfs_add_durable_block_rsv(root->fs_info, rsv);
6618 trans = btrfs_start_transaction(root, 4);
6619 if (IS_ERR(trans)) {
6620 err = PTR_ERR(trans);
6625 * Reserve space for the truncate process. Truncate should be adding
6626 * space, but if there are snapshots it may end up using space.
6628 ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6631 ret = btrfs_orphan_add(trans, inode);
6633 btrfs_end_transaction(trans, root);
6637 nr = trans->blocks_used;
6638 btrfs_end_transaction(trans, root);
6639 btrfs_btree_balance_dirty(root, nr);
6642 * Ok so we've already migrated our bytes over for the truncate, so here
6643 * just reserve the one slot we need for updating the inode.
6645 trans = btrfs_start_transaction(root, 1);
6646 if (IS_ERR(trans)) {
6647 err = PTR_ERR(trans);
6650 trans->block_rsv = rsv;
6653 * setattr is responsible for setting the ordered_data_close flag,
6654 * but that is only tested during the last file release. That
6655 * could happen well after the next commit, leaving a great big
6656 * window where new writes may get lost if someone chooses to write
6657 * to this file after truncating to zero
6659 * The inode doesn't have any dirty data here, and so if we commit
6660 * this is a noop. If someone immediately starts writing to the inode
6661 * it is very likely we'll catch some of their writes in this
6662 * transaction, and the commit will find this file on the ordered
6663 * data list with good things to send down.
6665 * This is a best effort solution, there is still a window where
6666 * using truncate to replace the contents of the file will
6667 * end up with a zero length file after a crash.
6669 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6670 btrfs_add_ordered_operation(trans, root, inode);
6674 trans = btrfs_start_transaction(root, 3);
6675 if (IS_ERR(trans)) {
6676 err = PTR_ERR(trans);
6680 ret = btrfs_truncate_reserve_metadata(trans, root,
6684 trans->block_rsv = rsv;
6687 ret = btrfs_truncate_inode_items(trans, root, inode,
6689 BTRFS_EXTENT_DATA_KEY);
6690 if (ret != -EAGAIN) {
6695 trans->block_rsv = &root->fs_info->trans_block_rsv;
6696 ret = btrfs_update_inode(trans, root, inode);
6702 nr = trans->blocks_used;
6703 btrfs_end_transaction(trans, root);
6705 btrfs_btree_balance_dirty(root, nr);
6708 if (ret == 0 && inode->i_nlink > 0) {
6709 trans->block_rsv = root->orphan_block_rsv;
6710 ret = btrfs_orphan_del(trans, inode);
6713 } else if (ret && inode->i_nlink > 0) {
6715 * Failed to do the truncate, remove us from the in memory
6718 ret = btrfs_orphan_del(NULL, inode);
6721 trans->block_rsv = &root->fs_info->trans_block_rsv;
6722 ret = btrfs_update_inode(trans, root, inode);
6726 nr = trans->blocks_used;
6727 ret = btrfs_end_transaction_throttle(trans, root);
6728 btrfs_btree_balance_dirty(root, nr);
6731 btrfs_free_block_rsv(root, rsv);
6740 * create a new subvolume directory/inode (helper for the ioctl).
6742 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6743 struct btrfs_root *new_root, u64 new_dirid)
6745 struct inode *inode;
6749 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6750 new_dirid, S_IFDIR | 0700, &index);
6752 return PTR_ERR(inode);
6753 inode->i_op = &btrfs_dir_inode_operations;
6754 inode->i_fop = &btrfs_dir_file_operations;
6757 btrfs_i_size_write(inode, 0);
6759 err = btrfs_update_inode(trans, new_root, inode);
6766 /* helper function for file defrag and space balancing. This
6767 * forces readahead on a given range of bytes in an inode
6769 unsigned long btrfs_force_ra(struct address_space *mapping,
6770 struct file_ra_state *ra, struct file *file,
6771 pgoff_t offset, pgoff_t last_index)
6773 pgoff_t req_size = last_index - offset + 1;
6775 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6776 return offset + req_size;
6779 struct inode *btrfs_alloc_inode(struct super_block *sb)
6781 struct btrfs_inode *ei;
6782 struct inode *inode;
6784 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6789 ei->space_info = NULL;
6793 ei->last_sub_trans = 0;
6794 ei->logged_trans = 0;
6795 ei->delalloc_bytes = 0;
6796 ei->reserved_bytes = 0;
6797 ei->disk_i_size = 0;
6799 ei->index_cnt = (u64)-1;
6800 ei->last_unlink_trans = 0;
6802 atomic_set(&ei->outstanding_extents, 0);
6803 atomic_set(&ei->reserved_extents, 0);
6805 ei->ordered_data_close = 0;
6806 ei->orphan_meta_reserved = 0;
6807 ei->dummy_inode = 0;
6808 ei->force_compress = BTRFS_COMPRESS_NONE;
6810 inode = &ei->vfs_inode;
6811 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6812 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6813 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6814 mutex_init(&ei->log_mutex);
6815 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6816 INIT_LIST_HEAD(&ei->i_orphan);
6817 INIT_LIST_HEAD(&ei->delalloc_inodes);
6818 INIT_LIST_HEAD(&ei->ordered_operations);
6819 RB_CLEAR_NODE(&ei->rb_node);
6824 static void btrfs_i_callback(struct rcu_head *head)
6826 struct inode *inode = container_of(head, struct inode, i_rcu);
6827 INIT_LIST_HEAD(&inode->i_dentry);
6828 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6831 void btrfs_destroy_inode(struct inode *inode)
6833 struct btrfs_ordered_extent *ordered;
6834 struct btrfs_root *root = BTRFS_I(inode)->root;
6836 WARN_ON(!list_empty(&inode->i_dentry));
6837 WARN_ON(inode->i_data.nrpages);
6838 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6839 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6842 * This can happen where we create an inode, but somebody else also
6843 * created the same inode and we need to destroy the one we already
6850 * Make sure we're properly removed from the ordered operation
6854 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6855 spin_lock(&root->fs_info->ordered_extent_lock);
6856 list_del_init(&BTRFS_I(inode)->ordered_operations);
6857 spin_unlock(&root->fs_info->ordered_extent_lock);
6860 spin_lock(&root->orphan_lock);
6861 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6862 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6864 list_del_init(&BTRFS_I(inode)->i_orphan);
6866 spin_unlock(&root->orphan_lock);
6869 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6873 printk(KERN_ERR "btrfs found ordered "
6874 "extent %llu %llu on inode cleanup\n",
6875 (unsigned long long)ordered->file_offset,
6876 (unsigned long long)ordered->len);
6877 btrfs_remove_ordered_extent(inode, ordered);
6878 btrfs_put_ordered_extent(ordered);
6879 btrfs_put_ordered_extent(ordered);
6882 inode_tree_del(inode);
6883 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6885 call_rcu(&inode->i_rcu, btrfs_i_callback);
6888 int btrfs_drop_inode(struct inode *inode)
6890 struct btrfs_root *root = BTRFS_I(inode)->root;
6892 if (btrfs_root_refs(&root->root_item) == 0 &&
6893 root != root->fs_info->tree_root)
6896 return generic_drop_inode(inode);
6899 static void init_once(void *foo)
6901 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6903 inode_init_once(&ei->vfs_inode);
6906 void btrfs_destroy_cachep(void)
6908 if (btrfs_inode_cachep)
6909 kmem_cache_destroy(btrfs_inode_cachep);
6910 if (btrfs_trans_handle_cachep)
6911 kmem_cache_destroy(btrfs_trans_handle_cachep);
6912 if (btrfs_transaction_cachep)
6913 kmem_cache_destroy(btrfs_transaction_cachep);
6914 if (btrfs_path_cachep)
6915 kmem_cache_destroy(btrfs_path_cachep);
6916 if (btrfs_free_space_cachep)
6917 kmem_cache_destroy(btrfs_free_space_cachep);
6920 int btrfs_init_cachep(void)
6922 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6923 sizeof(struct btrfs_inode), 0,
6924 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6925 if (!btrfs_inode_cachep)
6928 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6929 sizeof(struct btrfs_trans_handle), 0,
6930 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6931 if (!btrfs_trans_handle_cachep)
6934 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6935 sizeof(struct btrfs_transaction), 0,
6936 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6937 if (!btrfs_transaction_cachep)
6940 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6941 sizeof(struct btrfs_path), 0,
6942 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6943 if (!btrfs_path_cachep)
6946 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6947 sizeof(struct btrfs_free_space), 0,
6948 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6949 if (!btrfs_free_space_cachep)
6954 btrfs_destroy_cachep();
6958 static int btrfs_getattr(struct vfsmount *mnt,
6959 struct dentry *dentry, struct kstat *stat)
6961 struct inode *inode = dentry->d_inode;
6962 generic_fillattr(inode, stat);
6963 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6964 stat->blksize = PAGE_CACHE_SIZE;
6965 stat->blocks = (inode_get_bytes(inode) +
6966 BTRFS_I(inode)->delalloc_bytes) >> 9;
6971 * If a file is moved, it will inherit the cow and compression flags of the new
6974 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6976 struct btrfs_inode *b_dir = BTRFS_I(dir);
6977 struct btrfs_inode *b_inode = BTRFS_I(inode);
6979 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6980 b_inode->flags |= BTRFS_INODE_NODATACOW;
6982 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6984 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6985 b_inode->flags |= BTRFS_INODE_COMPRESS;
6987 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6990 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6991 struct inode *new_dir, struct dentry *new_dentry)
6993 struct btrfs_trans_handle *trans;
6994 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6995 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6996 struct inode *new_inode = new_dentry->d_inode;
6997 struct inode *old_inode = old_dentry->d_inode;
6998 struct timespec ctime = CURRENT_TIME;
7003 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7006 /* we only allow rename subvolume link between subvolumes */
7007 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7010 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7011 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7014 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7015 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7018 * we're using rename to replace one file with another.
7019 * and the replacement file is large. Start IO on it now so
7020 * we don't add too much work to the end of the transaction
7022 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7023 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7024 filemap_flush(old_inode->i_mapping);
7026 /* close the racy window with snapshot create/destroy ioctl */
7027 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7028 down_read(&root->fs_info->subvol_sem);
7030 * We want to reserve the absolute worst case amount of items. So if
7031 * both inodes are subvols and we need to unlink them then that would
7032 * require 4 item modifications, but if they are both normal inodes it
7033 * would require 5 item modifications, so we'll assume their normal
7034 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7035 * should cover the worst case number of items we'll modify.
7037 trans = btrfs_start_transaction(root, 20);
7038 if (IS_ERR(trans)) {
7039 ret = PTR_ERR(trans);
7044 btrfs_record_root_in_trans(trans, dest);
7046 ret = btrfs_set_inode_index(new_dir, &index);
7050 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7051 /* force full log commit if subvolume involved. */
7052 root->fs_info->last_trans_log_full_commit = trans->transid;
7054 ret = btrfs_insert_inode_ref(trans, dest,
7055 new_dentry->d_name.name,
7056 new_dentry->d_name.len,
7058 new_dir->i_ino, index);
7062 * this is an ugly little race, but the rename is required
7063 * to make sure that if we crash, the inode is either at the
7064 * old name or the new one. pinning the log transaction lets
7065 * us make sure we don't allow a log commit to come in after
7066 * we unlink the name but before we add the new name back in.
7068 btrfs_pin_log_trans(root);
7071 * make sure the inode gets flushed if it is replacing
7074 if (new_inode && new_inode->i_size &&
7075 old_inode && S_ISREG(old_inode->i_mode)) {
7076 btrfs_add_ordered_operation(trans, root, old_inode);
7079 old_dir->i_ctime = old_dir->i_mtime = ctime;
7080 new_dir->i_ctime = new_dir->i_mtime = ctime;
7081 old_inode->i_ctime = ctime;
7083 if (old_dentry->d_parent != new_dentry->d_parent)
7084 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7086 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7087 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7088 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7089 old_dentry->d_name.name,
7090 old_dentry->d_name.len);
7092 ret = __btrfs_unlink_inode(trans, root, old_dir,
7093 old_dentry->d_inode,
7094 old_dentry->d_name.name,
7095 old_dentry->d_name.len);
7097 ret = btrfs_update_inode(trans, root, old_inode);
7102 new_inode->i_ctime = CURRENT_TIME;
7103 if (unlikely(new_inode->i_ino ==
7104 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7105 root_objectid = BTRFS_I(new_inode)->location.objectid;
7106 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7108 new_dentry->d_name.name,
7109 new_dentry->d_name.len);
7110 BUG_ON(new_inode->i_nlink == 0);
7112 ret = btrfs_unlink_inode(trans, dest, new_dir,
7113 new_dentry->d_inode,
7114 new_dentry->d_name.name,
7115 new_dentry->d_name.len);
7118 if (new_inode->i_nlink == 0) {
7119 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7124 fixup_inode_flags(new_dir, old_inode);
7126 ret = btrfs_add_link(trans, new_dir, old_inode,
7127 new_dentry->d_name.name,
7128 new_dentry->d_name.len, 0, index);
7131 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7132 struct dentry *parent = dget_parent(new_dentry);
7133 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7135 btrfs_end_log_trans(root);
7138 btrfs_end_transaction_throttle(trans, root);
7140 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7141 up_read(&root->fs_info->subvol_sem);
7147 * some fairly slow code that needs optimization. This walks the list
7148 * of all the inodes with pending delalloc and forces them to disk.
7150 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7152 struct list_head *head = &root->fs_info->delalloc_inodes;
7153 struct btrfs_inode *binode;
7154 struct inode *inode;
7156 if (root->fs_info->sb->s_flags & MS_RDONLY)
7159 spin_lock(&root->fs_info->delalloc_lock);
7160 while (!list_empty(head)) {
7161 binode = list_entry(head->next, struct btrfs_inode,
7163 inode = igrab(&binode->vfs_inode);
7165 list_del_init(&binode->delalloc_inodes);
7166 spin_unlock(&root->fs_info->delalloc_lock);
7168 filemap_flush(inode->i_mapping);
7170 btrfs_add_delayed_iput(inode);
7175 spin_lock(&root->fs_info->delalloc_lock);
7177 spin_unlock(&root->fs_info->delalloc_lock);
7179 /* the filemap_flush will queue IO into the worker threads, but
7180 * we have to make sure the IO is actually started and that
7181 * ordered extents get created before we return
7183 atomic_inc(&root->fs_info->async_submit_draining);
7184 while (atomic_read(&root->fs_info->nr_async_submits) ||
7185 atomic_read(&root->fs_info->async_delalloc_pages)) {
7186 wait_event(root->fs_info->async_submit_wait,
7187 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7188 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7190 atomic_dec(&root->fs_info->async_submit_draining);
7194 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7197 struct btrfs_inode *binode;
7198 struct inode *inode = NULL;
7200 spin_lock(&root->fs_info->delalloc_lock);
7201 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7202 binode = list_entry(root->fs_info->delalloc_inodes.next,
7203 struct btrfs_inode, delalloc_inodes);
7204 inode = igrab(&binode->vfs_inode);
7206 list_move_tail(&binode->delalloc_inodes,
7207 &root->fs_info->delalloc_inodes);
7211 list_del_init(&binode->delalloc_inodes);
7212 cond_resched_lock(&root->fs_info->delalloc_lock);
7214 spin_unlock(&root->fs_info->delalloc_lock);
7218 filemap_write_and_wait(inode->i_mapping);
7220 * We have to do this because compression doesn't
7221 * actually set PG_writeback until it submits the pages
7222 * for IO, which happens in an async thread, so we could
7223 * race and not actually wait for any writeback pages
7224 * because they've not been submitted yet. Technically
7225 * this could still be the case for the ordered stuff
7226 * since the async thread may not have started to do its
7227 * work yet. If this becomes the case then we need to
7228 * figure out a way to make sure that in writepage we
7229 * wait for any async pages to be submitted before
7230 * returning so that fdatawait does what its supposed to
7233 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7235 filemap_flush(inode->i_mapping);
7238 btrfs_add_delayed_iput(inode);
7246 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7247 const char *symname)
7249 struct btrfs_trans_handle *trans;
7250 struct btrfs_root *root = BTRFS_I(dir)->root;
7251 struct btrfs_path *path;
7252 struct btrfs_key key;
7253 struct inode *inode = NULL;
7261 struct btrfs_file_extent_item *ei;
7262 struct extent_buffer *leaf;
7263 unsigned long nr = 0;
7265 name_len = strlen(symname) + 1;
7266 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7267 return -ENAMETOOLONG;
7269 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7273 * 2 items for inode item and ref
7274 * 2 items for dir items
7275 * 1 item for xattr if selinux is on
7277 trans = btrfs_start_transaction(root, 5);
7279 return PTR_ERR(trans);
7281 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7282 dentry->d_name.len, dir->i_ino, objectid,
7283 S_IFLNK|S_IRWXUGO, &index);
7284 if (IS_ERR(inode)) {
7285 err = PTR_ERR(inode);
7289 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7295 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7299 inode->i_mapping->a_ops = &btrfs_aops;
7300 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7301 inode->i_fop = &btrfs_file_operations;
7302 inode->i_op = &btrfs_file_inode_operations;
7303 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7308 path = btrfs_alloc_path();
7310 key.objectid = inode->i_ino;
7312 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7313 datasize = btrfs_file_extent_calc_inline_size(name_len);
7314 err = btrfs_insert_empty_item(trans, root, path, &key,
7320 leaf = path->nodes[0];
7321 ei = btrfs_item_ptr(leaf, path->slots[0],
7322 struct btrfs_file_extent_item);
7323 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7324 btrfs_set_file_extent_type(leaf, ei,
7325 BTRFS_FILE_EXTENT_INLINE);
7326 btrfs_set_file_extent_encryption(leaf, ei, 0);
7327 btrfs_set_file_extent_compression(leaf, ei, 0);
7328 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7329 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7331 ptr = btrfs_file_extent_inline_start(ei);
7332 write_extent_buffer(leaf, symname, ptr, name_len);
7333 btrfs_mark_buffer_dirty(leaf);
7334 btrfs_free_path(path);
7336 inode->i_op = &btrfs_symlink_inode_operations;
7337 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7338 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7339 inode_set_bytes(inode, name_len);
7340 btrfs_i_size_write(inode, name_len - 1);
7341 err = btrfs_update_inode(trans, root, inode);
7346 nr = trans->blocks_used;
7347 btrfs_end_transaction_throttle(trans, root);
7349 inode_dec_link_count(inode);
7352 btrfs_btree_balance_dirty(root, nr);
7356 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7357 u64 start, u64 num_bytes, u64 min_size,
7358 loff_t actual_len, u64 *alloc_hint,
7359 struct btrfs_trans_handle *trans)
7361 struct btrfs_root *root = BTRFS_I(inode)->root;
7362 struct btrfs_key ins;
7363 u64 cur_offset = start;
7366 bool own_trans = true;
7370 while (num_bytes > 0) {
7372 trans = btrfs_start_transaction(root, 3);
7373 if (IS_ERR(trans)) {
7374 ret = PTR_ERR(trans);
7379 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7380 0, *alloc_hint, (u64)-1, &ins, 1);
7383 btrfs_end_transaction(trans, root);
7387 ret = insert_reserved_file_extent(trans, inode,
7388 cur_offset, ins.objectid,
7389 ins.offset, ins.offset,
7390 ins.offset, 0, 0, 0,
7391 BTRFS_FILE_EXTENT_PREALLOC);
7393 btrfs_drop_extent_cache(inode, cur_offset,
7394 cur_offset + ins.offset -1, 0);
7396 num_bytes -= ins.offset;
7397 cur_offset += ins.offset;
7398 *alloc_hint = ins.objectid + ins.offset;
7400 inode->i_ctime = CURRENT_TIME;
7401 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7402 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7403 (actual_len > inode->i_size) &&
7404 (cur_offset > inode->i_size)) {
7405 if (cur_offset > actual_len)
7406 i_size = actual_len;
7408 i_size = cur_offset;
7409 i_size_write(inode, i_size);
7410 btrfs_ordered_update_i_size(inode, i_size, NULL);
7413 ret = btrfs_update_inode(trans, root, inode);
7417 btrfs_end_transaction(trans, root);
7422 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7423 u64 start, u64 num_bytes, u64 min_size,
7424 loff_t actual_len, u64 *alloc_hint)
7426 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7427 min_size, actual_len, alloc_hint,
7431 int btrfs_prealloc_file_range_trans(struct inode *inode,
7432 struct btrfs_trans_handle *trans, int mode,
7433 u64 start, u64 num_bytes, u64 min_size,
7434 loff_t actual_len, u64 *alloc_hint)
7436 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7437 min_size, actual_len, alloc_hint, trans);
7440 static int btrfs_set_page_dirty(struct page *page)
7442 return __set_page_dirty_nobuffers(page);
7445 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7447 struct btrfs_root *root = BTRFS_I(inode)->root;
7449 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7451 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7453 return generic_permission(inode, mask, flags, btrfs_check_acl);
7456 static const struct inode_operations btrfs_dir_inode_operations = {
7457 .getattr = btrfs_getattr,
7458 .lookup = btrfs_lookup,
7459 .create = btrfs_create,
7460 .unlink = btrfs_unlink,
7462 .mkdir = btrfs_mkdir,
7463 .rmdir = btrfs_rmdir,
7464 .rename = btrfs_rename,
7465 .symlink = btrfs_symlink,
7466 .setattr = btrfs_setattr,
7467 .mknod = btrfs_mknod,
7468 .setxattr = btrfs_setxattr,
7469 .getxattr = btrfs_getxattr,
7470 .listxattr = btrfs_listxattr,
7471 .removexattr = btrfs_removexattr,
7472 .permission = btrfs_permission,
7474 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7475 .lookup = btrfs_lookup,
7476 .permission = btrfs_permission,
7479 static const struct file_operations btrfs_dir_file_operations = {
7480 .llseek = generic_file_llseek,
7481 .read = generic_read_dir,
7482 .readdir = btrfs_real_readdir,
7483 .unlocked_ioctl = btrfs_ioctl,
7484 #ifdef CONFIG_COMPAT
7485 .compat_ioctl = btrfs_ioctl,
7487 .release = btrfs_release_file,
7488 .fsync = btrfs_sync_file,
7491 static struct extent_io_ops btrfs_extent_io_ops = {
7492 .fill_delalloc = run_delalloc_range,
7493 .submit_bio_hook = btrfs_submit_bio_hook,
7494 .merge_bio_hook = btrfs_merge_bio_hook,
7495 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7496 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7497 .writepage_start_hook = btrfs_writepage_start_hook,
7498 .readpage_io_failed_hook = btrfs_io_failed_hook,
7499 .set_bit_hook = btrfs_set_bit_hook,
7500 .clear_bit_hook = btrfs_clear_bit_hook,
7501 .merge_extent_hook = btrfs_merge_extent_hook,
7502 .split_extent_hook = btrfs_split_extent_hook,
7506 * btrfs doesn't support the bmap operation because swapfiles
7507 * use bmap to make a mapping of extents in the file. They assume
7508 * these extents won't change over the life of the file and they
7509 * use the bmap result to do IO directly to the drive.
7511 * the btrfs bmap call would return logical addresses that aren't
7512 * suitable for IO and they also will change frequently as COW
7513 * operations happen. So, swapfile + btrfs == corruption.
7515 * For now we're avoiding this by dropping bmap.
7517 static const struct address_space_operations btrfs_aops = {
7518 .readpage = btrfs_readpage,
7519 .writepage = btrfs_writepage,
7520 .writepages = btrfs_writepages,
7521 .readpages = btrfs_readpages,
7522 .direct_IO = btrfs_direct_IO,
7523 .invalidatepage = btrfs_invalidatepage,
7524 .releasepage = btrfs_releasepage,
7525 .set_page_dirty = btrfs_set_page_dirty,
7526 .error_remove_page = generic_error_remove_page,
7529 static const struct address_space_operations btrfs_symlink_aops = {
7530 .readpage = btrfs_readpage,
7531 .writepage = btrfs_writepage,
7532 .invalidatepage = btrfs_invalidatepage,
7533 .releasepage = btrfs_releasepage,
7536 static const struct inode_operations btrfs_file_inode_operations = {
7537 .getattr = btrfs_getattr,
7538 .setattr = btrfs_setattr,
7539 .setxattr = btrfs_setxattr,
7540 .getxattr = btrfs_getxattr,
7541 .listxattr = btrfs_listxattr,
7542 .removexattr = btrfs_removexattr,
7543 .permission = btrfs_permission,
7544 .fiemap = btrfs_fiemap,
7546 static const struct inode_operations btrfs_special_inode_operations = {
7547 .getattr = btrfs_getattr,
7548 .setattr = btrfs_setattr,
7549 .permission = btrfs_permission,
7550 .setxattr = btrfs_setxattr,
7551 .getxattr = btrfs_getxattr,
7552 .listxattr = btrfs_listxattr,
7553 .removexattr = btrfs_removexattr,
7555 static const struct inode_operations btrfs_symlink_inode_operations = {
7556 .readlink = generic_readlink,
7557 .follow_link = page_follow_link_light,
7558 .put_link = page_put_link,
7559 .getattr = btrfs_getattr,
7560 .permission = btrfs_permission,
7561 .setxattr = btrfs_setxattr,
7562 .getxattr = btrfs_getxattr,
7563 .listxattr = btrfs_listxattr,
7564 .removexattr = btrfs_removexattr,
7567 const struct dentry_operations btrfs_dentry_operations = {
7568 .d_delete = btrfs_dentry_delete,