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 void btrfs_truncate(struct inode *inode);
88 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, int *page_started,
92 unsigned long *nr_written, int unlock);
94 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
95 struct inode *inode, struct inode *dir)
99 err = btrfs_init_acl(trans, inode, dir);
101 err = btrfs_xattr_security_init(trans, inode, dir);
106 * this does all the hard work for inserting an inline extent into
107 * the btree. The caller should have done a btrfs_drop_extents so that
108 * no overlapping inline items exist in the btree
110 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
111 struct btrfs_root *root, struct inode *inode,
112 u64 start, size_t size, size_t compressed_size,
113 struct page **compressed_pages)
115 struct btrfs_key key;
116 struct btrfs_path *path;
117 struct extent_buffer *leaf;
118 struct page *page = NULL;
121 struct btrfs_file_extent_item *ei;
124 size_t cur_size = size;
126 unsigned long offset;
127 int compress_type = BTRFS_COMPRESS_NONE;
129 if (compressed_size && compressed_pages) {
130 compress_type = root->fs_info->compress_type;
131 cur_size = compressed_size;
134 path = btrfs_alloc_path();
138 path->leave_spinning = 1;
139 btrfs_set_trans_block_group(trans, inode);
141 key.objectid = inode->i_ino;
143 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
144 datasize = btrfs_file_extent_calc_inline_size(cur_size);
146 inode_add_bytes(inode, size);
147 ret = btrfs_insert_empty_item(trans, root, path, &key,
154 leaf = path->nodes[0];
155 ei = btrfs_item_ptr(leaf, path->slots[0],
156 struct btrfs_file_extent_item);
157 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
158 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
159 btrfs_set_file_extent_encryption(leaf, ei, 0);
160 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
161 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
162 ptr = btrfs_file_extent_inline_start(ei);
164 if (compress_type != BTRFS_COMPRESS_NONE) {
167 while (compressed_size > 0) {
168 cpage = compressed_pages[i];
169 cur_size = min_t(unsigned long, compressed_size,
172 kaddr = kmap_atomic(cpage, KM_USER0);
173 write_extent_buffer(leaf, kaddr, ptr, cur_size);
174 kunmap_atomic(kaddr, KM_USER0);
178 compressed_size -= cur_size;
180 btrfs_set_file_extent_compression(leaf, ei,
183 page = find_get_page(inode->i_mapping,
184 start >> PAGE_CACHE_SHIFT);
185 btrfs_set_file_extent_compression(leaf, ei, 0);
186 kaddr = kmap_atomic(page, KM_USER0);
187 offset = start & (PAGE_CACHE_SIZE - 1);
188 write_extent_buffer(leaf, kaddr + offset, ptr, size);
189 kunmap_atomic(kaddr, KM_USER0);
190 page_cache_release(page);
192 btrfs_mark_buffer_dirty(leaf);
193 btrfs_free_path(path);
196 * we're an inline extent, so nobody can
197 * extend the file past i_size without locking
198 * a page we already have locked.
200 * We must do any isize and inode updates
201 * before we unlock the pages. Otherwise we
202 * could end up racing with unlink.
204 BTRFS_I(inode)->disk_i_size = inode->i_size;
205 btrfs_update_inode(trans, root, inode);
209 btrfs_free_path(path);
215 * conditionally insert an inline extent into the file. This
216 * does the checks required to make sure the data is small enough
217 * to fit as an inline extent.
219 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
220 struct btrfs_root *root,
221 struct inode *inode, u64 start, u64 end,
222 size_t compressed_size,
223 struct page **compressed_pages)
225 u64 isize = i_size_read(inode);
226 u64 actual_end = min(end + 1, isize);
227 u64 inline_len = actual_end - start;
228 u64 aligned_end = (end + root->sectorsize - 1) &
229 ~((u64)root->sectorsize - 1);
231 u64 data_len = inline_len;
235 data_len = compressed_size;
238 actual_end >= PAGE_CACHE_SIZE ||
239 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
241 (actual_end & (root->sectorsize - 1)) == 0) ||
243 data_len > root->fs_info->max_inline) {
247 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
251 if (isize > actual_end)
252 inline_len = min_t(u64, isize, actual_end);
253 ret = insert_inline_extent(trans, root, inode, start,
254 inline_len, compressed_size,
257 btrfs_delalloc_release_metadata(inode, end + 1 - start);
258 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
262 struct async_extent {
267 unsigned long nr_pages;
269 struct list_head list;
274 struct btrfs_root *root;
275 struct page *locked_page;
278 struct list_head extents;
279 struct btrfs_work work;
282 static noinline int add_async_extent(struct async_cow *cow,
283 u64 start, u64 ram_size,
286 unsigned long nr_pages,
289 struct async_extent *async_extent;
291 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
292 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))) {
388 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
390 if (BTRFS_I(inode)->force_compress)
391 compress_type = BTRFS_I(inode)->force_compress;
393 ret = btrfs_compress_pages(compress_type,
394 inode->i_mapping, start,
395 total_compressed, pages,
396 nr_pages, &nr_pages_ret,
402 unsigned long offset = total_compressed &
403 (PAGE_CACHE_SIZE - 1);
404 struct page *page = pages[nr_pages_ret - 1];
407 /* zero the tail end of the last page, we might be
408 * sending it down to disk
411 kaddr = kmap_atomic(page, KM_USER0);
412 memset(kaddr + offset, 0,
413 PAGE_CACHE_SIZE - offset);
414 kunmap_atomic(kaddr, KM_USER0);
420 trans = btrfs_join_transaction(root, 1);
421 BUG_ON(IS_ERR(trans));
422 btrfs_set_trans_block_group(trans, inode);
423 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
425 /* lets try to make an inline extent */
426 if (ret || total_in < (actual_end - start)) {
427 /* we didn't compress the entire range, try
428 * to make an uncompressed inline extent.
430 ret = cow_file_range_inline(trans, root, inode,
431 start, end, 0, NULL);
433 /* try making a compressed inline extent */
434 ret = cow_file_range_inline(trans, root, inode,
436 total_compressed, pages);
440 * inline extent creation worked, we don't need
441 * to create any more async work items. Unlock
442 * and free up our temp pages.
444 extent_clear_unlock_delalloc(inode,
445 &BTRFS_I(inode)->io_tree,
447 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
448 EXTENT_CLEAR_DELALLOC |
449 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
451 btrfs_end_transaction(trans, root);
454 btrfs_end_transaction(trans, root);
459 * we aren't doing an inline extent round the compressed size
460 * up to a block size boundary so the allocator does sane
463 total_compressed = (total_compressed + blocksize - 1) &
467 * one last check to make sure the compression is really a
468 * win, compare the page count read with the blocks on disk
470 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
471 ~(PAGE_CACHE_SIZE - 1);
472 if (total_compressed >= total_in) {
475 num_bytes = total_in;
478 if (!will_compress && pages) {
480 * the compression code ran but failed to make things smaller,
481 * free any pages it allocated and our page pointer array
483 for (i = 0; i < nr_pages_ret; i++) {
484 WARN_ON(pages[i]->mapping);
485 page_cache_release(pages[i]);
489 total_compressed = 0;
492 /* flag the file so we don't compress in the future */
493 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
494 !(BTRFS_I(inode)->force_compress)) {
495 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
501 /* the async work queues will take care of doing actual
502 * allocation on disk for these compressed pages,
503 * and will submit them to the elevator.
505 add_async_extent(async_cow, start, num_bytes,
506 total_compressed, pages, nr_pages_ret,
509 if (start + num_bytes < end) {
516 cleanup_and_bail_uncompressed:
518 * No compression, but we still need to write the pages in
519 * the file we've been given so far. redirty the locked
520 * page if it corresponds to our extent and set things up
521 * for the async work queue to run cow_file_range to do
522 * the normal delalloc dance
524 if (page_offset(locked_page) >= start &&
525 page_offset(locked_page) <= end) {
526 __set_page_dirty_nobuffers(locked_page);
527 /* unlocked later on in the async handlers */
529 add_async_extent(async_cow, start, end - start + 1,
530 0, NULL, 0, BTRFS_COMPRESS_NONE);
538 for (i = 0; i < nr_pages_ret; i++) {
539 WARN_ON(pages[i]->mapping);
540 page_cache_release(pages[i]);
548 * phase two of compressed writeback. This is the ordered portion
549 * of the code, which only gets called in the order the work was
550 * queued. We walk all the async extents created by compress_file_range
551 * and send them down to the disk.
553 static noinline int submit_compressed_extents(struct inode *inode,
554 struct async_cow *async_cow)
556 struct async_extent *async_extent;
558 struct btrfs_trans_handle *trans;
559 struct btrfs_key ins;
560 struct extent_map *em;
561 struct btrfs_root *root = BTRFS_I(inode)->root;
562 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
563 struct extent_io_tree *io_tree;
566 if (list_empty(&async_cow->extents))
570 while (!list_empty(&async_cow->extents)) {
571 async_extent = list_entry(async_cow->extents.next,
572 struct async_extent, list);
573 list_del(&async_extent->list);
575 io_tree = &BTRFS_I(inode)->io_tree;
578 /* did the compression code fall back to uncompressed IO? */
579 if (!async_extent->pages) {
580 int page_started = 0;
581 unsigned long nr_written = 0;
583 lock_extent(io_tree, async_extent->start,
584 async_extent->start +
585 async_extent->ram_size - 1, GFP_NOFS);
587 /* allocate blocks */
588 ret = cow_file_range(inode, async_cow->locked_page,
590 async_extent->start +
591 async_extent->ram_size - 1,
592 &page_started, &nr_written, 0);
595 * if page_started, cow_file_range inserted an
596 * inline extent and took care of all the unlocking
597 * and IO for us. Otherwise, we need to submit
598 * all those pages down to the drive.
600 if (!page_started && !ret)
601 extent_write_locked_range(io_tree,
602 inode, async_extent->start,
603 async_extent->start +
604 async_extent->ram_size - 1,
612 lock_extent(io_tree, async_extent->start,
613 async_extent->start + async_extent->ram_size - 1,
616 trans = btrfs_join_transaction(root, 1);
617 BUG_ON(IS_ERR(trans));
618 ret = btrfs_reserve_extent(trans, root,
619 async_extent->compressed_size,
620 async_extent->compressed_size,
623 btrfs_end_transaction(trans, root);
627 for (i = 0; i < async_extent->nr_pages; i++) {
628 WARN_ON(async_extent->pages[i]->mapping);
629 page_cache_release(async_extent->pages[i]);
631 kfree(async_extent->pages);
632 async_extent->nr_pages = 0;
633 async_extent->pages = NULL;
634 unlock_extent(io_tree, async_extent->start,
635 async_extent->start +
636 async_extent->ram_size - 1, GFP_NOFS);
641 * here we're doing allocation and writeback of the
644 btrfs_drop_extent_cache(inode, async_extent->start,
645 async_extent->start +
646 async_extent->ram_size - 1, 0);
648 em = alloc_extent_map(GFP_NOFS);
650 em->start = async_extent->start;
651 em->len = async_extent->ram_size;
652 em->orig_start = em->start;
654 em->block_start = ins.objectid;
655 em->block_len = ins.offset;
656 em->bdev = root->fs_info->fs_devices->latest_bdev;
657 em->compress_type = async_extent->compress_type;
658 set_bit(EXTENT_FLAG_PINNED, &em->flags);
659 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
662 write_lock(&em_tree->lock);
663 ret = add_extent_mapping(em_tree, em);
664 write_unlock(&em_tree->lock);
665 if (ret != -EEXIST) {
669 btrfs_drop_extent_cache(inode, async_extent->start,
670 async_extent->start +
671 async_extent->ram_size - 1, 0);
674 ret = btrfs_add_ordered_extent_compress(inode,
677 async_extent->ram_size,
679 BTRFS_ORDERED_COMPRESSED,
680 async_extent->compress_type);
684 * clear dirty, set writeback and unlock the pages.
686 extent_clear_unlock_delalloc(inode,
687 &BTRFS_I(inode)->io_tree,
689 async_extent->start +
690 async_extent->ram_size - 1,
691 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
692 EXTENT_CLEAR_UNLOCK |
693 EXTENT_CLEAR_DELALLOC |
694 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
696 ret = btrfs_submit_compressed_write(inode,
698 async_extent->ram_size,
700 ins.offset, async_extent->pages,
701 async_extent->nr_pages);
704 alloc_hint = ins.objectid + ins.offset;
712 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
715 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
716 struct extent_map *em;
719 read_lock(&em_tree->lock);
720 em = search_extent_mapping(em_tree, start, num_bytes);
723 * if block start isn't an actual block number then find the
724 * first block in this inode and use that as a hint. If that
725 * block is also bogus then just don't worry about it.
727 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
729 em = search_extent_mapping(em_tree, 0, 0);
730 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
731 alloc_hint = em->block_start;
735 alloc_hint = em->block_start;
739 read_unlock(&em_tree->lock);
745 * when extent_io.c finds a delayed allocation range in the file,
746 * the call backs end up in this code. The basic idea is to
747 * allocate extents on disk for the range, and create ordered data structs
748 * in ram to track those extents.
750 * locked_page is the page that writepage had locked already. We use
751 * it to make sure we don't do extra locks or unlocks.
753 * *page_started is set to one if we unlock locked_page and do everything
754 * required to start IO on it. It may be clean and already done with
757 static noinline int cow_file_range(struct inode *inode,
758 struct page *locked_page,
759 u64 start, u64 end, int *page_started,
760 unsigned long *nr_written,
763 struct btrfs_root *root = BTRFS_I(inode)->root;
764 struct btrfs_trans_handle *trans;
767 unsigned long ram_size;
770 u64 blocksize = root->sectorsize;
771 struct btrfs_key ins;
772 struct extent_map *em;
773 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
776 BUG_ON(root == root->fs_info->tree_root);
777 trans = btrfs_join_transaction(root, 1);
778 BUG_ON(IS_ERR(trans));
779 btrfs_set_trans_block_group(trans, inode);
780 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
782 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
783 num_bytes = max(blocksize, num_bytes);
784 disk_num_bytes = num_bytes;
788 /* lets try to make an inline extent */
789 ret = cow_file_range_inline(trans, root, inode,
790 start, end, 0, NULL);
792 extent_clear_unlock_delalloc(inode,
793 &BTRFS_I(inode)->io_tree,
795 EXTENT_CLEAR_UNLOCK_PAGE |
796 EXTENT_CLEAR_UNLOCK |
797 EXTENT_CLEAR_DELALLOC |
799 EXTENT_SET_WRITEBACK |
800 EXTENT_END_WRITEBACK);
802 *nr_written = *nr_written +
803 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
810 BUG_ON(disk_num_bytes >
811 btrfs_super_total_bytes(&root->fs_info->super_copy));
813 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
814 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
816 while (disk_num_bytes > 0) {
819 cur_alloc_size = disk_num_bytes;
820 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
821 root->sectorsize, 0, alloc_hint,
825 em = alloc_extent_map(GFP_NOFS);
828 em->orig_start = em->start;
829 ram_size = ins.offset;
830 em->len = ins.offset;
832 em->block_start = ins.objectid;
833 em->block_len = ins.offset;
834 em->bdev = root->fs_info->fs_devices->latest_bdev;
835 set_bit(EXTENT_FLAG_PINNED, &em->flags);
838 write_lock(&em_tree->lock);
839 ret = add_extent_mapping(em_tree, em);
840 write_unlock(&em_tree->lock);
841 if (ret != -EEXIST) {
845 btrfs_drop_extent_cache(inode, start,
846 start + ram_size - 1, 0);
849 cur_alloc_size = ins.offset;
850 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
851 ram_size, cur_alloc_size, 0);
854 if (root->root_key.objectid ==
855 BTRFS_DATA_RELOC_TREE_OBJECTID) {
856 ret = btrfs_reloc_clone_csums(inode, start,
861 if (disk_num_bytes < cur_alloc_size)
864 /* we're not doing compressed IO, don't unlock the first
865 * page (which the caller expects to stay locked), don't
866 * clear any dirty bits and don't set any writeback bits
868 * Do set the Private2 bit so we know this page was properly
869 * setup for writepage
871 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
872 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
875 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
876 start, start + ram_size - 1,
878 disk_num_bytes -= cur_alloc_size;
879 num_bytes -= cur_alloc_size;
880 alloc_hint = ins.objectid + ins.offset;
881 start += cur_alloc_size;
885 btrfs_end_transaction(trans, root);
891 * work queue call back to started compression on a file and pages
893 static noinline void async_cow_start(struct btrfs_work *work)
895 struct async_cow *async_cow;
897 async_cow = container_of(work, struct async_cow, work);
899 compress_file_range(async_cow->inode, async_cow->locked_page,
900 async_cow->start, async_cow->end, async_cow,
903 async_cow->inode = NULL;
907 * work queue call back to submit previously compressed pages
909 static noinline void async_cow_submit(struct btrfs_work *work)
911 struct async_cow *async_cow;
912 struct btrfs_root *root;
913 unsigned long nr_pages;
915 async_cow = container_of(work, struct async_cow, work);
917 root = async_cow->root;
918 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
921 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
923 if (atomic_read(&root->fs_info->async_delalloc_pages) <
925 waitqueue_active(&root->fs_info->async_submit_wait))
926 wake_up(&root->fs_info->async_submit_wait);
928 if (async_cow->inode)
929 submit_compressed_extents(async_cow->inode, async_cow);
932 static noinline void async_cow_free(struct btrfs_work *work)
934 struct async_cow *async_cow;
935 async_cow = container_of(work, struct async_cow, work);
939 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
940 u64 start, u64 end, int *page_started,
941 unsigned long *nr_written)
943 struct async_cow *async_cow;
944 struct btrfs_root *root = BTRFS_I(inode)->root;
945 unsigned long nr_pages;
947 int limit = 10 * 1024 * 1042;
949 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
950 1, 0, NULL, GFP_NOFS);
951 while (start < end) {
952 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
953 async_cow->inode = inode;
954 async_cow->root = root;
955 async_cow->locked_page = locked_page;
956 async_cow->start = start;
958 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
961 cur_end = min(end, start + 512 * 1024 - 1);
963 async_cow->end = cur_end;
964 INIT_LIST_HEAD(&async_cow->extents);
966 async_cow->work.func = async_cow_start;
967 async_cow->work.ordered_func = async_cow_submit;
968 async_cow->work.ordered_free = async_cow_free;
969 async_cow->work.flags = 0;
971 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
973 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
975 btrfs_queue_worker(&root->fs_info->delalloc_workers,
978 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
979 wait_event(root->fs_info->async_submit_wait,
980 (atomic_read(&root->fs_info->async_delalloc_pages) <
984 while (atomic_read(&root->fs_info->async_submit_draining) &&
985 atomic_read(&root->fs_info->async_delalloc_pages)) {
986 wait_event(root->fs_info->async_submit_wait,
987 (atomic_read(&root->fs_info->async_delalloc_pages) ==
991 *nr_written += nr_pages;
998 static noinline int csum_exist_in_range(struct btrfs_root *root,
999 u64 bytenr, u64 num_bytes)
1002 struct btrfs_ordered_sum *sums;
1005 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1006 bytenr + num_bytes - 1, &list);
1007 if (ret == 0 && list_empty(&list))
1010 while (!list_empty(&list)) {
1011 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1012 list_del(&sums->list);
1019 * when nowcow writeback call back. This checks for snapshots or COW copies
1020 * of the extents that exist in the file, and COWs the file as required.
1022 * If no cow copies or snapshots exist, we write directly to the existing
1025 static noinline int run_delalloc_nocow(struct inode *inode,
1026 struct page *locked_page,
1027 u64 start, u64 end, int *page_started, int force,
1028 unsigned long *nr_written)
1030 struct btrfs_root *root = BTRFS_I(inode)->root;
1031 struct btrfs_trans_handle *trans;
1032 struct extent_buffer *leaf;
1033 struct btrfs_path *path;
1034 struct btrfs_file_extent_item *fi;
1035 struct btrfs_key found_key;
1047 bool nolock = false;
1049 path = btrfs_alloc_path();
1051 if (root == root->fs_info->tree_root) {
1053 trans = btrfs_join_transaction_nolock(root, 1);
1055 trans = btrfs_join_transaction(root, 1);
1057 BUG_ON(IS_ERR(trans));
1059 cow_start = (u64)-1;
1062 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1065 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1066 leaf = path->nodes[0];
1067 btrfs_item_key_to_cpu(leaf, &found_key,
1068 path->slots[0] - 1);
1069 if (found_key.objectid == inode->i_ino &&
1070 found_key.type == BTRFS_EXTENT_DATA_KEY)
1075 leaf = path->nodes[0];
1076 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1077 ret = btrfs_next_leaf(root, path);
1082 leaf = path->nodes[0];
1088 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1090 if (found_key.objectid > inode->i_ino ||
1091 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1092 found_key.offset > end)
1095 if (found_key.offset > cur_offset) {
1096 extent_end = found_key.offset;
1101 fi = btrfs_item_ptr(leaf, path->slots[0],
1102 struct btrfs_file_extent_item);
1103 extent_type = btrfs_file_extent_type(leaf, fi);
1105 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1106 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1107 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1108 extent_offset = btrfs_file_extent_offset(leaf, fi);
1109 extent_end = found_key.offset +
1110 btrfs_file_extent_num_bytes(leaf, fi);
1111 if (extent_end <= start) {
1115 if (disk_bytenr == 0)
1117 if (btrfs_file_extent_compression(leaf, fi) ||
1118 btrfs_file_extent_encryption(leaf, fi) ||
1119 btrfs_file_extent_other_encoding(leaf, fi))
1121 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1123 if (btrfs_extent_readonly(root, disk_bytenr))
1125 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1127 extent_offset, disk_bytenr))
1129 disk_bytenr += extent_offset;
1130 disk_bytenr += cur_offset - found_key.offset;
1131 num_bytes = min(end + 1, extent_end) - cur_offset;
1133 * force cow if csum exists in the range.
1134 * this ensure that csum for a given extent are
1135 * either valid or do not exist.
1137 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1140 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1141 extent_end = found_key.offset +
1142 btrfs_file_extent_inline_len(leaf, fi);
1143 extent_end = ALIGN(extent_end, root->sectorsize);
1148 if (extent_end <= start) {
1153 if (cow_start == (u64)-1)
1154 cow_start = cur_offset;
1155 cur_offset = extent_end;
1156 if (cur_offset > end)
1162 btrfs_release_path(root, path);
1163 if (cow_start != (u64)-1) {
1164 ret = cow_file_range(inode, locked_page, cow_start,
1165 found_key.offset - 1, page_started,
1168 cow_start = (u64)-1;
1171 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1172 struct extent_map *em;
1173 struct extent_map_tree *em_tree;
1174 em_tree = &BTRFS_I(inode)->extent_tree;
1175 em = alloc_extent_map(GFP_NOFS);
1177 em->start = cur_offset;
1178 em->orig_start = em->start;
1179 em->len = num_bytes;
1180 em->block_len = num_bytes;
1181 em->block_start = disk_bytenr;
1182 em->bdev = root->fs_info->fs_devices->latest_bdev;
1183 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1185 write_lock(&em_tree->lock);
1186 ret = add_extent_mapping(em_tree, em);
1187 write_unlock(&em_tree->lock);
1188 if (ret != -EEXIST) {
1189 free_extent_map(em);
1192 btrfs_drop_extent_cache(inode, em->start,
1193 em->start + em->len - 1, 0);
1195 type = BTRFS_ORDERED_PREALLOC;
1197 type = BTRFS_ORDERED_NOCOW;
1200 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1201 num_bytes, num_bytes, type);
1204 if (root->root_key.objectid ==
1205 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1206 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1211 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1212 cur_offset, cur_offset + num_bytes - 1,
1213 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1214 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1215 EXTENT_SET_PRIVATE2);
1216 cur_offset = extent_end;
1217 if (cur_offset > end)
1220 btrfs_release_path(root, path);
1222 if (cur_offset <= end && cow_start == (u64)-1)
1223 cow_start = cur_offset;
1224 if (cow_start != (u64)-1) {
1225 ret = cow_file_range(inode, locked_page, cow_start, end,
1226 page_started, nr_written, 1);
1231 ret = btrfs_end_transaction_nolock(trans, root);
1234 ret = btrfs_end_transaction(trans, root);
1237 btrfs_free_path(path);
1242 * extent_io.c call back to do delayed allocation processing
1244 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1245 u64 start, u64 end, int *page_started,
1246 unsigned long *nr_written)
1249 struct btrfs_root *root = BTRFS_I(inode)->root;
1251 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1252 ret = run_delalloc_nocow(inode, locked_page, start, end,
1253 page_started, 1, nr_written);
1254 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1255 ret = run_delalloc_nocow(inode, locked_page, start, end,
1256 page_started, 0, nr_written);
1257 else if (!btrfs_test_opt(root, COMPRESS) &&
1258 !(BTRFS_I(inode)->force_compress))
1259 ret = cow_file_range(inode, locked_page, start, end,
1260 page_started, nr_written, 1);
1262 ret = cow_file_range_async(inode, locked_page, start, end,
1263 page_started, nr_written);
1267 static int btrfs_split_extent_hook(struct inode *inode,
1268 struct extent_state *orig, u64 split)
1270 /* not delalloc, ignore it */
1271 if (!(orig->state & EXTENT_DELALLOC))
1274 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1279 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1280 * extents so we can keep track of new extents that are just merged onto old
1281 * extents, such as when we are doing sequential writes, so we can properly
1282 * account for the metadata space we'll need.
1284 static int btrfs_merge_extent_hook(struct inode *inode,
1285 struct extent_state *new,
1286 struct extent_state *other)
1288 /* not delalloc, ignore it */
1289 if (!(other->state & EXTENT_DELALLOC))
1292 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1297 * extent_io.c set_bit_hook, used to track delayed allocation
1298 * bytes in this file, and to maintain the list of inodes that
1299 * have pending delalloc work to be done.
1301 static int btrfs_set_bit_hook(struct inode *inode,
1302 struct extent_state *state, int *bits)
1306 * set_bit and clear bit hooks normally require _irqsave/restore
1307 * but in this case, we are only testeing for the DELALLOC
1308 * bit, which is only set or cleared with irqs on
1310 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1311 struct btrfs_root *root = BTRFS_I(inode)->root;
1312 u64 len = state->end + 1 - state->start;
1313 int do_list = (root->root_key.objectid !=
1314 BTRFS_ROOT_TREE_OBJECTID);
1316 if (*bits & EXTENT_FIRST_DELALLOC)
1317 *bits &= ~EXTENT_FIRST_DELALLOC;
1319 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1321 spin_lock(&root->fs_info->delalloc_lock);
1322 BTRFS_I(inode)->delalloc_bytes += len;
1323 root->fs_info->delalloc_bytes += len;
1324 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1325 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1326 &root->fs_info->delalloc_inodes);
1328 spin_unlock(&root->fs_info->delalloc_lock);
1334 * extent_io.c clear_bit_hook, see set_bit_hook for why
1336 static int btrfs_clear_bit_hook(struct inode *inode,
1337 struct extent_state *state, int *bits)
1340 * set_bit and clear bit hooks normally require _irqsave/restore
1341 * but in this case, we are only testeing for the DELALLOC
1342 * bit, which is only set or cleared with irqs on
1344 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1345 struct btrfs_root *root = BTRFS_I(inode)->root;
1346 u64 len = state->end + 1 - state->start;
1347 int do_list = (root->root_key.objectid !=
1348 BTRFS_ROOT_TREE_OBJECTID);
1350 if (*bits & EXTENT_FIRST_DELALLOC)
1351 *bits &= ~EXTENT_FIRST_DELALLOC;
1352 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1353 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1355 if (*bits & EXTENT_DO_ACCOUNTING)
1356 btrfs_delalloc_release_metadata(inode, len);
1358 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1360 btrfs_free_reserved_data_space(inode, len);
1362 spin_lock(&root->fs_info->delalloc_lock);
1363 root->fs_info->delalloc_bytes -= len;
1364 BTRFS_I(inode)->delalloc_bytes -= len;
1366 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1367 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1368 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1370 spin_unlock(&root->fs_info->delalloc_lock);
1376 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1377 * we don't create bios that span stripes or chunks
1379 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1380 size_t size, struct bio *bio,
1381 unsigned long bio_flags)
1383 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1384 struct btrfs_mapping_tree *map_tree;
1385 u64 logical = (u64)bio->bi_sector << 9;
1390 if (bio_flags & EXTENT_BIO_COMPRESSED)
1393 length = bio->bi_size;
1394 map_tree = &root->fs_info->mapping_tree;
1395 map_length = length;
1396 ret = btrfs_map_block(map_tree, READ, logical,
1397 &map_length, NULL, 0);
1399 if (map_length < length + size)
1405 * in order to insert checksums into the metadata in large chunks,
1406 * we wait until bio submission time. All the pages in the bio are
1407 * checksummed and sums are attached onto the ordered extent record.
1409 * At IO completion time the cums attached on the ordered extent record
1410 * are inserted into the btree
1412 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1413 struct bio *bio, int mirror_num,
1414 unsigned long bio_flags,
1417 struct btrfs_root *root = BTRFS_I(inode)->root;
1420 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1426 * in order to insert checksums into the metadata in large chunks,
1427 * we wait until bio submission time. All the pages in the bio are
1428 * checksummed and sums are attached onto the ordered extent record.
1430 * At IO completion time the cums attached on the ordered extent record
1431 * are inserted into the btree
1433 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1434 int mirror_num, unsigned long bio_flags,
1437 struct btrfs_root *root = BTRFS_I(inode)->root;
1438 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1442 * extent_io.c submission hook. This does the right thing for csum calculation
1443 * on write, or reading the csums from the tree before a read
1445 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1446 int mirror_num, unsigned long bio_flags,
1449 struct btrfs_root *root = BTRFS_I(inode)->root;
1453 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1455 if (root == root->fs_info->tree_root)
1456 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1458 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1461 if (!(rw & REQ_WRITE)) {
1462 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1463 return btrfs_submit_compressed_read(inode, bio,
1464 mirror_num, bio_flags);
1465 } else if (!skip_sum)
1466 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1468 } else if (!skip_sum) {
1469 /* csum items have already been cloned */
1470 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1472 /* we're doing a write, do the async checksumming */
1473 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1474 inode, rw, bio, mirror_num,
1475 bio_flags, bio_offset,
1476 __btrfs_submit_bio_start,
1477 __btrfs_submit_bio_done);
1481 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1485 * given a list of ordered sums record them in the inode. This happens
1486 * at IO completion time based on sums calculated at bio submission time.
1488 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1489 struct inode *inode, u64 file_offset,
1490 struct list_head *list)
1492 struct btrfs_ordered_sum *sum;
1494 btrfs_set_trans_block_group(trans, inode);
1496 list_for_each_entry(sum, list, list) {
1497 btrfs_csum_file_blocks(trans,
1498 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1503 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1504 struct extent_state **cached_state)
1506 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1508 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1509 cached_state, GFP_NOFS);
1512 /* see btrfs_writepage_start_hook for details on why this is required */
1513 struct btrfs_writepage_fixup {
1515 struct btrfs_work work;
1518 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1520 struct btrfs_writepage_fixup *fixup;
1521 struct btrfs_ordered_extent *ordered;
1522 struct extent_state *cached_state = NULL;
1524 struct inode *inode;
1528 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1532 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1533 ClearPageChecked(page);
1537 inode = page->mapping->host;
1538 page_start = page_offset(page);
1539 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1541 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1542 &cached_state, GFP_NOFS);
1544 /* already ordered? We're done */
1545 if (PagePrivate2(page))
1548 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1550 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1551 page_end, &cached_state, GFP_NOFS);
1553 btrfs_start_ordered_extent(inode, ordered, 1);
1558 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1559 ClearPageChecked(page);
1561 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1562 &cached_state, GFP_NOFS);
1565 page_cache_release(page);
1570 * There are a few paths in the higher layers of the kernel that directly
1571 * set the page dirty bit without asking the filesystem if it is a
1572 * good idea. This causes problems because we want to make sure COW
1573 * properly happens and the data=ordered rules are followed.
1575 * In our case any range that doesn't have the ORDERED bit set
1576 * hasn't been properly setup for IO. We kick off an async process
1577 * to fix it up. The async helper will wait for ordered extents, set
1578 * the delalloc bit and make it safe to write the page.
1580 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1582 struct inode *inode = page->mapping->host;
1583 struct btrfs_writepage_fixup *fixup;
1584 struct btrfs_root *root = BTRFS_I(inode)->root;
1586 /* this page is properly in the ordered list */
1587 if (TestClearPagePrivate2(page))
1590 if (PageChecked(page))
1593 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1597 SetPageChecked(page);
1598 page_cache_get(page);
1599 fixup->work.func = btrfs_writepage_fixup_worker;
1601 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1605 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1606 struct inode *inode, u64 file_pos,
1607 u64 disk_bytenr, u64 disk_num_bytes,
1608 u64 num_bytes, u64 ram_bytes,
1609 u8 compression, u8 encryption,
1610 u16 other_encoding, int extent_type)
1612 struct btrfs_root *root = BTRFS_I(inode)->root;
1613 struct btrfs_file_extent_item *fi;
1614 struct btrfs_path *path;
1615 struct extent_buffer *leaf;
1616 struct btrfs_key ins;
1620 path = btrfs_alloc_path();
1623 path->leave_spinning = 1;
1626 * we may be replacing one extent in the tree with another.
1627 * The new extent is pinned in the extent map, and we don't want
1628 * to drop it from the cache until it is completely in the btree.
1630 * So, tell btrfs_drop_extents to leave this extent in the cache.
1631 * the caller is expected to unpin it and allow it to be merged
1634 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1638 ins.objectid = inode->i_ino;
1639 ins.offset = file_pos;
1640 ins.type = BTRFS_EXTENT_DATA_KEY;
1641 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1643 leaf = path->nodes[0];
1644 fi = btrfs_item_ptr(leaf, path->slots[0],
1645 struct btrfs_file_extent_item);
1646 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1647 btrfs_set_file_extent_type(leaf, fi, extent_type);
1648 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1649 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1650 btrfs_set_file_extent_offset(leaf, fi, 0);
1651 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1652 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1653 btrfs_set_file_extent_compression(leaf, fi, compression);
1654 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1655 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1657 btrfs_unlock_up_safe(path, 1);
1658 btrfs_set_lock_blocking(leaf);
1660 btrfs_mark_buffer_dirty(leaf);
1662 inode_add_bytes(inode, num_bytes);
1664 ins.objectid = disk_bytenr;
1665 ins.offset = disk_num_bytes;
1666 ins.type = BTRFS_EXTENT_ITEM_KEY;
1667 ret = btrfs_alloc_reserved_file_extent(trans, root,
1668 root->root_key.objectid,
1669 inode->i_ino, file_pos, &ins);
1671 btrfs_free_path(path);
1677 * helper function for btrfs_finish_ordered_io, this
1678 * just reads in some of the csum leaves to prime them into ram
1679 * before we start the transaction. It limits the amount of btree
1680 * reads required while inside the transaction.
1682 /* as ordered data IO finishes, this gets called so we can finish
1683 * an ordered extent if the range of bytes in the file it covers are
1686 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1688 struct btrfs_root *root = BTRFS_I(inode)->root;
1689 struct btrfs_trans_handle *trans = NULL;
1690 struct btrfs_ordered_extent *ordered_extent = NULL;
1691 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1692 struct extent_state *cached_state = NULL;
1693 int compress_type = 0;
1695 bool nolock = false;
1697 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1701 BUG_ON(!ordered_extent);
1703 nolock = (root == root->fs_info->tree_root);
1705 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1706 BUG_ON(!list_empty(&ordered_extent->list));
1707 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1710 trans = btrfs_join_transaction_nolock(root, 1);
1712 trans = btrfs_join_transaction(root, 1);
1713 BUG_ON(IS_ERR(trans));
1714 btrfs_set_trans_block_group(trans, inode);
1715 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1716 ret = btrfs_update_inode(trans, root, inode);
1722 lock_extent_bits(io_tree, ordered_extent->file_offset,
1723 ordered_extent->file_offset + ordered_extent->len - 1,
1724 0, &cached_state, GFP_NOFS);
1727 trans = btrfs_join_transaction_nolock(root, 1);
1729 trans = btrfs_join_transaction(root, 1);
1730 BUG_ON(IS_ERR(trans));
1731 btrfs_set_trans_block_group(trans, inode);
1732 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1734 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1735 compress_type = ordered_extent->compress_type;
1736 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1737 BUG_ON(compress_type);
1738 ret = btrfs_mark_extent_written(trans, inode,
1739 ordered_extent->file_offset,
1740 ordered_extent->file_offset +
1741 ordered_extent->len);
1744 BUG_ON(root == root->fs_info->tree_root);
1745 ret = insert_reserved_file_extent(trans, inode,
1746 ordered_extent->file_offset,
1747 ordered_extent->start,
1748 ordered_extent->disk_len,
1749 ordered_extent->len,
1750 ordered_extent->len,
1751 compress_type, 0, 0,
1752 BTRFS_FILE_EXTENT_REG);
1753 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1754 ordered_extent->file_offset,
1755 ordered_extent->len);
1758 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1759 ordered_extent->file_offset +
1760 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1762 add_pending_csums(trans, inode, ordered_extent->file_offset,
1763 &ordered_extent->list);
1765 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1766 ret = btrfs_update_inode(trans, root, inode);
1771 btrfs_end_transaction_nolock(trans, root);
1773 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1775 btrfs_end_transaction(trans, root);
1779 btrfs_put_ordered_extent(ordered_extent);
1780 /* once for the tree */
1781 btrfs_put_ordered_extent(ordered_extent);
1786 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1787 struct extent_state *state, int uptodate)
1789 ClearPagePrivate2(page);
1790 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1794 * When IO fails, either with EIO or csum verification fails, we
1795 * try other mirrors that might have a good copy of the data. This
1796 * io_failure_record is used to record state as we go through all the
1797 * mirrors. If another mirror has good data, the page is set up to date
1798 * and things continue. If a good mirror can't be found, the original
1799 * bio end_io callback is called to indicate things have failed.
1801 struct io_failure_record {
1806 unsigned long bio_flags;
1810 static int btrfs_io_failed_hook(struct bio *failed_bio,
1811 struct page *page, u64 start, u64 end,
1812 struct extent_state *state)
1814 struct io_failure_record *failrec = NULL;
1816 struct extent_map *em;
1817 struct inode *inode = page->mapping->host;
1818 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1819 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1826 ret = get_state_private(failure_tree, start, &private);
1828 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1831 failrec->start = start;
1832 failrec->len = end - start + 1;
1833 failrec->last_mirror = 0;
1834 failrec->bio_flags = 0;
1836 read_lock(&em_tree->lock);
1837 em = lookup_extent_mapping(em_tree, start, failrec->len);
1838 if (em->start > start || em->start + em->len < start) {
1839 free_extent_map(em);
1842 read_unlock(&em_tree->lock);
1844 if (!em || IS_ERR(em)) {
1848 logical = start - em->start;
1849 logical = em->block_start + logical;
1850 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1851 logical = em->block_start;
1852 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1853 extent_set_compress_type(&failrec->bio_flags,
1856 failrec->logical = logical;
1857 free_extent_map(em);
1858 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1859 EXTENT_DIRTY, GFP_NOFS);
1860 set_state_private(failure_tree, start,
1861 (u64)(unsigned long)failrec);
1863 failrec = (struct io_failure_record *)(unsigned long)private;
1865 num_copies = btrfs_num_copies(
1866 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1867 failrec->logical, failrec->len);
1868 failrec->last_mirror++;
1870 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1871 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1874 if (state && state->start != failrec->start)
1876 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1878 if (!state || failrec->last_mirror > num_copies) {
1879 set_state_private(failure_tree, failrec->start, 0);
1880 clear_extent_bits(failure_tree, failrec->start,
1881 failrec->start + failrec->len - 1,
1882 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1886 bio = bio_alloc(GFP_NOFS, 1);
1887 bio->bi_private = state;
1888 bio->bi_end_io = failed_bio->bi_end_io;
1889 bio->bi_sector = failrec->logical >> 9;
1890 bio->bi_bdev = failed_bio->bi_bdev;
1893 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1894 if (failed_bio->bi_rw & REQ_WRITE)
1899 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1900 failrec->last_mirror,
1901 failrec->bio_flags, 0);
1906 * each time an IO finishes, we do a fast check in the IO failure tree
1907 * to see if we need to process or clean up an io_failure_record
1909 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1912 u64 private_failure;
1913 struct io_failure_record *failure;
1917 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1918 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1919 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1920 start, &private_failure);
1922 failure = (struct io_failure_record *)(unsigned long)
1924 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1926 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1928 failure->start + failure->len - 1,
1929 EXTENT_DIRTY | EXTENT_LOCKED,
1938 * when reads are done, we need to check csums to verify the data is correct
1939 * if there's a match, we allow the bio to finish. If not, we go through
1940 * the io_failure_record routines to find good copies
1942 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1943 struct extent_state *state)
1945 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1946 struct inode *inode = page->mapping->host;
1947 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1949 u64 private = ~(u32)0;
1951 struct btrfs_root *root = BTRFS_I(inode)->root;
1954 if (PageChecked(page)) {
1955 ClearPageChecked(page);
1959 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1962 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1963 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1964 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1969 if (state && state->start == start) {
1970 private = state->private;
1973 ret = get_state_private(io_tree, start, &private);
1975 kaddr = kmap_atomic(page, KM_USER0);
1979 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1980 btrfs_csum_final(csum, (char *)&csum);
1981 if (csum != private)
1984 kunmap_atomic(kaddr, KM_USER0);
1986 /* if the io failure tree for this inode is non-empty,
1987 * check to see if we've recovered from a failed IO
1989 btrfs_clean_io_failures(inode, start);
1993 if (printk_ratelimit()) {
1994 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1995 "private %llu\n", page->mapping->host->i_ino,
1996 (unsigned long long)start, csum,
1997 (unsigned long long)private);
1999 memset(kaddr + offset, 1, end - start + 1);
2000 flush_dcache_page(page);
2001 kunmap_atomic(kaddr, KM_USER0);
2007 struct delayed_iput {
2008 struct list_head list;
2009 struct inode *inode;
2012 void btrfs_add_delayed_iput(struct inode *inode)
2014 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2015 struct delayed_iput *delayed;
2017 if (atomic_add_unless(&inode->i_count, -1, 1))
2020 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2021 delayed->inode = inode;
2023 spin_lock(&fs_info->delayed_iput_lock);
2024 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2025 spin_unlock(&fs_info->delayed_iput_lock);
2028 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2031 struct btrfs_fs_info *fs_info = root->fs_info;
2032 struct delayed_iput *delayed;
2035 spin_lock(&fs_info->delayed_iput_lock);
2036 empty = list_empty(&fs_info->delayed_iputs);
2037 spin_unlock(&fs_info->delayed_iput_lock);
2041 down_read(&root->fs_info->cleanup_work_sem);
2042 spin_lock(&fs_info->delayed_iput_lock);
2043 list_splice_init(&fs_info->delayed_iputs, &list);
2044 spin_unlock(&fs_info->delayed_iput_lock);
2046 while (!list_empty(&list)) {
2047 delayed = list_entry(list.next, struct delayed_iput, list);
2048 list_del(&delayed->list);
2049 iput(delayed->inode);
2052 up_read(&root->fs_info->cleanup_work_sem);
2056 * calculate extra metadata reservation when snapshotting a subvolume
2057 * contains orphan files.
2059 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2060 struct btrfs_pending_snapshot *pending,
2061 u64 *bytes_to_reserve)
2063 struct btrfs_root *root;
2064 struct btrfs_block_rsv *block_rsv;
2068 root = pending->root;
2069 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2072 block_rsv = root->orphan_block_rsv;
2074 /* orphan block reservation for the snapshot */
2075 num_bytes = block_rsv->size;
2078 * after the snapshot is created, COWing tree blocks may use more
2079 * space than it frees. So we should make sure there is enough
2082 index = trans->transid & 0x1;
2083 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2084 num_bytes += block_rsv->size -
2085 (block_rsv->reserved + block_rsv->freed[index]);
2088 *bytes_to_reserve += num_bytes;
2091 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2092 struct btrfs_pending_snapshot *pending)
2094 struct btrfs_root *root = pending->root;
2095 struct btrfs_root *snap = pending->snap;
2096 struct btrfs_block_rsv *block_rsv;
2101 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2104 /* refill source subvolume's orphan block reservation */
2105 block_rsv = root->orphan_block_rsv;
2106 index = trans->transid & 0x1;
2107 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2108 num_bytes = block_rsv->size -
2109 (block_rsv->reserved + block_rsv->freed[index]);
2110 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2111 root->orphan_block_rsv,
2116 /* setup orphan block reservation for the snapshot */
2117 block_rsv = btrfs_alloc_block_rsv(snap);
2120 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2121 snap->orphan_block_rsv = block_rsv;
2123 num_bytes = root->orphan_block_rsv->size;
2124 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2125 block_rsv, num_bytes);
2129 /* insert orphan item for the snapshot */
2130 WARN_ON(!root->orphan_item_inserted);
2131 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2132 snap->root_key.objectid);
2134 snap->orphan_item_inserted = 1;
2138 enum btrfs_orphan_cleanup_state {
2139 ORPHAN_CLEANUP_STARTED = 1,
2140 ORPHAN_CLEANUP_DONE = 2,
2144 * This is called in transaction commmit time. If there are no orphan
2145 * files in the subvolume, it removes orphan item and frees block_rsv
2148 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2149 struct btrfs_root *root)
2153 if (!list_empty(&root->orphan_list) ||
2154 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2157 if (root->orphan_item_inserted &&
2158 btrfs_root_refs(&root->root_item) > 0) {
2159 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2160 root->root_key.objectid);
2162 root->orphan_item_inserted = 0;
2165 if (root->orphan_block_rsv) {
2166 WARN_ON(root->orphan_block_rsv->size > 0);
2167 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2168 root->orphan_block_rsv = NULL;
2173 * This creates an orphan entry for the given inode in case something goes
2174 * wrong in the middle of an unlink/truncate.
2176 * NOTE: caller of this function should reserve 5 units of metadata for
2179 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2181 struct btrfs_root *root = BTRFS_I(inode)->root;
2182 struct btrfs_block_rsv *block_rsv = NULL;
2187 if (!root->orphan_block_rsv) {
2188 block_rsv = btrfs_alloc_block_rsv(root);
2192 spin_lock(&root->orphan_lock);
2193 if (!root->orphan_block_rsv) {
2194 root->orphan_block_rsv = block_rsv;
2195 } else if (block_rsv) {
2196 btrfs_free_block_rsv(root, block_rsv);
2200 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2201 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2204 * For proper ENOSPC handling, we should do orphan
2205 * cleanup when mounting. But this introduces backward
2206 * compatibility issue.
2208 if (!xchg(&root->orphan_item_inserted, 1))
2215 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2218 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2219 BTRFS_I(inode)->orphan_meta_reserved = 1;
2222 spin_unlock(&root->orphan_lock);
2225 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2227 /* grab metadata reservation from transaction handle */
2229 ret = btrfs_orphan_reserve_metadata(trans, inode);
2233 /* insert an orphan item to track this unlinked/truncated file */
2235 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2239 /* insert an orphan item to track subvolume contains orphan files */
2241 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2242 root->root_key.objectid);
2249 * We have done the truncate/delete so we can go ahead and remove the orphan
2250 * item for this particular inode.
2252 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2254 struct btrfs_root *root = BTRFS_I(inode)->root;
2255 int delete_item = 0;
2256 int release_rsv = 0;
2259 spin_lock(&root->orphan_lock);
2260 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2261 list_del_init(&BTRFS_I(inode)->i_orphan);
2265 if (BTRFS_I(inode)->orphan_meta_reserved) {
2266 BTRFS_I(inode)->orphan_meta_reserved = 0;
2269 spin_unlock(&root->orphan_lock);
2271 if (trans && delete_item) {
2272 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2277 btrfs_orphan_release_metadata(inode);
2283 * this cleans up any orphans that may be left on the list from the last use
2286 void btrfs_orphan_cleanup(struct btrfs_root *root)
2288 struct btrfs_path *path;
2289 struct extent_buffer *leaf;
2290 struct btrfs_key key, found_key;
2291 struct btrfs_trans_handle *trans;
2292 struct inode *inode;
2293 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2295 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2298 path = btrfs_alloc_path();
2302 key.objectid = BTRFS_ORPHAN_OBJECTID;
2303 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2304 key.offset = (u64)-1;
2307 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2309 printk(KERN_ERR "Error searching slot for orphan: %d"
2315 * if ret == 0 means we found what we were searching for, which
2316 * is weird, but possible, so only screw with path if we didnt
2317 * find the key and see if we have stuff that matches
2320 if (path->slots[0] == 0)
2325 /* pull out the item */
2326 leaf = path->nodes[0];
2327 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2329 /* make sure the item matches what we want */
2330 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2332 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2335 /* release the path since we're done with it */
2336 btrfs_release_path(root, path);
2339 * this is where we are basically btrfs_lookup, without the
2340 * crossing root thing. we store the inode number in the
2341 * offset of the orphan item.
2343 found_key.objectid = found_key.offset;
2344 found_key.type = BTRFS_INODE_ITEM_KEY;
2345 found_key.offset = 0;
2346 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2347 BUG_ON(IS_ERR(inode));
2350 * add this inode to the orphan list so btrfs_orphan_del does
2351 * the proper thing when we hit it
2353 spin_lock(&root->orphan_lock);
2354 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2355 spin_unlock(&root->orphan_lock);
2358 * if this is a bad inode, means we actually succeeded in
2359 * removing the inode, but not the orphan record, which means
2360 * we need to manually delete the orphan since iput will just
2361 * do a destroy_inode
2363 if (is_bad_inode(inode)) {
2364 trans = btrfs_start_transaction(root, 0);
2365 BUG_ON(IS_ERR(trans));
2366 btrfs_orphan_del(trans, inode);
2367 btrfs_end_transaction(trans, root);
2372 /* if we have links, this was a truncate, lets do that */
2373 if (inode->i_nlink) {
2375 btrfs_truncate(inode);
2380 /* this will do delete_inode and everything for us */
2383 btrfs_free_path(path);
2385 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2387 if (root->orphan_block_rsv)
2388 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2391 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2392 trans = btrfs_join_transaction(root, 1);
2393 BUG_ON(IS_ERR(trans));
2394 btrfs_end_transaction(trans, root);
2398 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2400 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2404 * very simple check to peek ahead in the leaf looking for xattrs. If we
2405 * don't find any xattrs, we know there can't be any acls.
2407 * slot is the slot the inode is in, objectid is the objectid of the inode
2409 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2410 int slot, u64 objectid)
2412 u32 nritems = btrfs_header_nritems(leaf);
2413 struct btrfs_key found_key;
2417 while (slot < nritems) {
2418 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2420 /* we found a different objectid, there must not be acls */
2421 if (found_key.objectid != objectid)
2424 /* we found an xattr, assume we've got an acl */
2425 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2429 * we found a key greater than an xattr key, there can't
2430 * be any acls later on
2432 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2439 * it goes inode, inode backrefs, xattrs, extents,
2440 * so if there are a ton of hard links to an inode there can
2441 * be a lot of backrefs. Don't waste time searching too hard,
2442 * this is just an optimization
2447 /* we hit the end of the leaf before we found an xattr or
2448 * something larger than an xattr. We have to assume the inode
2455 * read an inode from the btree into the in-memory inode
2457 static void btrfs_read_locked_inode(struct inode *inode)
2459 struct btrfs_path *path;
2460 struct extent_buffer *leaf;
2461 struct btrfs_inode_item *inode_item;
2462 struct btrfs_timespec *tspec;
2463 struct btrfs_root *root = BTRFS_I(inode)->root;
2464 struct btrfs_key location;
2466 u64 alloc_group_block;
2470 path = btrfs_alloc_path();
2472 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2474 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2478 leaf = path->nodes[0];
2479 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2480 struct btrfs_inode_item);
2482 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2483 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2484 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2485 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2486 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2488 tspec = btrfs_inode_atime(inode_item);
2489 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2490 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2492 tspec = btrfs_inode_mtime(inode_item);
2493 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2494 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2496 tspec = btrfs_inode_ctime(inode_item);
2497 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2498 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2500 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2501 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2502 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2503 inode->i_generation = BTRFS_I(inode)->generation;
2505 rdev = btrfs_inode_rdev(leaf, inode_item);
2507 BTRFS_I(inode)->index_cnt = (u64)-1;
2508 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2510 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2513 * try to precache a NULL acl entry for files that don't have
2514 * any xattrs or acls
2516 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2518 cache_no_acl(inode);
2520 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2521 alloc_group_block, 0);
2522 btrfs_free_path(path);
2525 switch (inode->i_mode & S_IFMT) {
2527 inode->i_mapping->a_ops = &btrfs_aops;
2528 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2529 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2530 inode->i_fop = &btrfs_file_operations;
2531 inode->i_op = &btrfs_file_inode_operations;
2534 inode->i_fop = &btrfs_dir_file_operations;
2535 if (root == root->fs_info->tree_root)
2536 inode->i_op = &btrfs_dir_ro_inode_operations;
2538 inode->i_op = &btrfs_dir_inode_operations;
2541 inode->i_op = &btrfs_symlink_inode_operations;
2542 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2543 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2546 inode->i_op = &btrfs_special_inode_operations;
2547 init_special_inode(inode, inode->i_mode, rdev);
2551 btrfs_update_iflags(inode);
2555 btrfs_free_path(path);
2556 make_bad_inode(inode);
2560 * given a leaf and an inode, copy the inode fields into the leaf
2562 static void fill_inode_item(struct btrfs_trans_handle *trans,
2563 struct extent_buffer *leaf,
2564 struct btrfs_inode_item *item,
2565 struct inode *inode)
2567 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2568 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2569 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2570 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2571 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2573 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2574 inode->i_atime.tv_sec);
2575 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2576 inode->i_atime.tv_nsec);
2578 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2579 inode->i_mtime.tv_sec);
2580 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2581 inode->i_mtime.tv_nsec);
2583 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2584 inode->i_ctime.tv_sec);
2585 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2586 inode->i_ctime.tv_nsec);
2588 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2589 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2590 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2591 btrfs_set_inode_transid(leaf, item, trans->transid);
2592 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2593 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2594 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2598 * copy everything in the in-memory inode into the btree.
2600 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2601 struct btrfs_root *root, struct inode *inode)
2603 struct btrfs_inode_item *inode_item;
2604 struct btrfs_path *path;
2605 struct extent_buffer *leaf;
2608 path = btrfs_alloc_path();
2610 path->leave_spinning = 1;
2611 ret = btrfs_lookup_inode(trans, root, path,
2612 &BTRFS_I(inode)->location, 1);
2619 btrfs_unlock_up_safe(path, 1);
2620 leaf = path->nodes[0];
2621 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2622 struct btrfs_inode_item);
2624 fill_inode_item(trans, leaf, inode_item, inode);
2625 btrfs_mark_buffer_dirty(leaf);
2626 btrfs_set_inode_last_trans(trans, inode);
2629 btrfs_free_path(path);
2635 * unlink helper that gets used here in inode.c and in the tree logging
2636 * recovery code. It remove a link in a directory with a given name, and
2637 * also drops the back refs in the inode to the directory
2639 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2640 struct btrfs_root *root,
2641 struct inode *dir, struct inode *inode,
2642 const char *name, int name_len)
2644 struct btrfs_path *path;
2646 struct extent_buffer *leaf;
2647 struct btrfs_dir_item *di;
2648 struct btrfs_key key;
2651 path = btrfs_alloc_path();
2657 path->leave_spinning = 1;
2658 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2659 name, name_len, -1);
2668 leaf = path->nodes[0];
2669 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2670 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2673 btrfs_release_path(root, path);
2675 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2677 dir->i_ino, &index);
2679 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2680 "inode %lu parent %lu\n", name_len, name,
2681 inode->i_ino, dir->i_ino);
2685 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2686 index, name, name_len, -1);
2695 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2696 btrfs_release_path(root, path);
2698 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2700 BUG_ON(ret != 0 && ret != -ENOENT);
2702 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2707 btrfs_free_path(path);
2711 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2712 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2713 btrfs_update_inode(trans, root, dir);
2714 btrfs_drop_nlink(inode);
2715 ret = btrfs_update_inode(trans, root, inode);
2720 /* helper to check if there is any shared block in the path */
2721 static int check_path_shared(struct btrfs_root *root,
2722 struct btrfs_path *path)
2724 struct extent_buffer *eb;
2728 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2731 if (!path->nodes[level])
2733 eb = path->nodes[level];
2734 if (!btrfs_block_can_be_shared(root, eb))
2736 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2745 * helper to start transaction for unlink and rmdir.
2747 * unlink and rmdir are special in btrfs, they do not always free space.
2748 * so in enospc case, we should make sure they will free space before
2749 * allowing them to use the global metadata reservation.
2751 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2752 struct dentry *dentry)
2754 struct btrfs_trans_handle *trans;
2755 struct btrfs_root *root = BTRFS_I(dir)->root;
2756 struct btrfs_path *path;
2757 struct btrfs_inode_ref *ref;
2758 struct btrfs_dir_item *di;
2759 struct inode *inode = dentry->d_inode;
2765 trans = btrfs_start_transaction(root, 10);
2766 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2769 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2770 return ERR_PTR(-ENOSPC);
2772 /* check if there is someone else holds reference */
2773 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2774 return ERR_PTR(-ENOSPC);
2776 if (atomic_read(&inode->i_count) > 2)
2777 return ERR_PTR(-ENOSPC);
2779 if (xchg(&root->fs_info->enospc_unlink, 1))
2780 return ERR_PTR(-ENOSPC);
2782 path = btrfs_alloc_path();
2784 root->fs_info->enospc_unlink = 0;
2785 return ERR_PTR(-ENOMEM);
2788 trans = btrfs_start_transaction(root, 0);
2789 if (IS_ERR(trans)) {
2790 btrfs_free_path(path);
2791 root->fs_info->enospc_unlink = 0;
2795 path->skip_locking = 1;
2796 path->search_commit_root = 1;
2798 ret = btrfs_lookup_inode(trans, root, path,
2799 &BTRFS_I(dir)->location, 0);
2805 if (check_path_shared(root, path))
2810 btrfs_release_path(root, path);
2812 ret = btrfs_lookup_inode(trans, root, path,
2813 &BTRFS_I(inode)->location, 0);
2819 if (check_path_shared(root, path))
2824 btrfs_release_path(root, path);
2826 if (ret == 0 && S_ISREG(inode->i_mode)) {
2827 ret = btrfs_lookup_file_extent(trans, root, path,
2828 inode->i_ino, (u64)-1, 0);
2834 if (check_path_shared(root, path))
2836 btrfs_release_path(root, path);
2844 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2845 dentry->d_name.name, dentry->d_name.len, 0);
2851 if (check_path_shared(root, path))
2857 btrfs_release_path(root, path);
2859 ref = btrfs_lookup_inode_ref(trans, root, path,
2860 dentry->d_name.name, dentry->d_name.len,
2861 inode->i_ino, dir->i_ino, 0);
2867 if (check_path_shared(root, path))
2869 index = btrfs_inode_ref_index(path->nodes[0], ref);
2870 btrfs_release_path(root, path);
2872 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2873 dentry->d_name.name, dentry->d_name.len, 0);
2878 BUG_ON(ret == -ENOENT);
2879 if (check_path_shared(root, path))
2884 btrfs_free_path(path);
2886 btrfs_end_transaction(trans, root);
2887 root->fs_info->enospc_unlink = 0;
2888 return ERR_PTR(err);
2891 trans->block_rsv = &root->fs_info->global_block_rsv;
2895 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2896 struct btrfs_root *root)
2898 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2899 BUG_ON(!root->fs_info->enospc_unlink);
2900 root->fs_info->enospc_unlink = 0;
2902 btrfs_end_transaction_throttle(trans, root);
2905 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2907 struct btrfs_root *root = BTRFS_I(dir)->root;
2908 struct btrfs_trans_handle *trans;
2909 struct inode *inode = dentry->d_inode;
2911 unsigned long nr = 0;
2913 trans = __unlink_start_trans(dir, dentry);
2915 return PTR_ERR(trans);
2917 btrfs_set_trans_block_group(trans, dir);
2919 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2921 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2922 dentry->d_name.name, dentry->d_name.len);
2925 if (inode->i_nlink == 0) {
2926 ret = btrfs_orphan_add(trans, inode);
2930 nr = trans->blocks_used;
2931 __unlink_end_trans(trans, root);
2932 btrfs_btree_balance_dirty(root, nr);
2936 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2937 struct btrfs_root *root,
2938 struct inode *dir, u64 objectid,
2939 const char *name, int name_len)
2941 struct btrfs_path *path;
2942 struct extent_buffer *leaf;
2943 struct btrfs_dir_item *di;
2944 struct btrfs_key key;
2948 path = btrfs_alloc_path();
2952 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2953 name, name_len, -1);
2954 BUG_ON(!di || IS_ERR(di));
2956 leaf = path->nodes[0];
2957 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2958 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2959 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2961 btrfs_release_path(root, path);
2963 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2964 objectid, root->root_key.objectid,
2965 dir->i_ino, &index, name, name_len);
2967 BUG_ON(ret != -ENOENT);
2968 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2970 BUG_ON(!di || IS_ERR(di));
2972 leaf = path->nodes[0];
2973 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2974 btrfs_release_path(root, path);
2978 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2979 index, name, name_len, -1);
2980 BUG_ON(!di || IS_ERR(di));
2982 leaf = path->nodes[0];
2983 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2984 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2985 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2987 btrfs_release_path(root, path);
2989 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2990 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2991 ret = btrfs_update_inode(trans, root, dir);
2994 btrfs_free_path(path);
2998 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3000 struct inode *inode = dentry->d_inode;
3002 struct btrfs_root *root = BTRFS_I(dir)->root;
3003 struct btrfs_trans_handle *trans;
3004 unsigned long nr = 0;
3006 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3007 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3010 trans = __unlink_start_trans(dir, dentry);
3012 return PTR_ERR(trans);
3014 btrfs_set_trans_block_group(trans, dir);
3016 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3017 err = btrfs_unlink_subvol(trans, root, dir,
3018 BTRFS_I(inode)->location.objectid,
3019 dentry->d_name.name,
3020 dentry->d_name.len);
3024 err = btrfs_orphan_add(trans, inode);
3028 /* now the directory is empty */
3029 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3030 dentry->d_name.name, dentry->d_name.len);
3032 btrfs_i_size_write(inode, 0);
3034 nr = trans->blocks_used;
3035 __unlink_end_trans(trans, root);
3036 btrfs_btree_balance_dirty(root, nr);
3043 * when truncating bytes in a file, it is possible to avoid reading
3044 * the leaves that contain only checksum items. This can be the
3045 * majority of the IO required to delete a large file, but it must
3046 * be done carefully.
3048 * The keys in the level just above the leaves are checked to make sure
3049 * the lowest key in a given leaf is a csum key, and starts at an offset
3050 * after the new size.
3052 * Then the key for the next leaf is checked to make sure it also has
3053 * a checksum item for the same file. If it does, we know our target leaf
3054 * contains only checksum items, and it can be safely freed without reading
3057 * This is just an optimization targeted at large files. It may do
3058 * nothing. It will return 0 unless things went badly.
3060 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3061 struct btrfs_root *root,
3062 struct btrfs_path *path,
3063 struct inode *inode, u64 new_size)
3065 struct btrfs_key key;
3068 struct btrfs_key found_key;
3069 struct btrfs_key other_key;
3070 struct btrfs_leaf_ref *ref;
3074 path->lowest_level = 1;
3075 key.objectid = inode->i_ino;
3076 key.type = BTRFS_CSUM_ITEM_KEY;
3077 key.offset = new_size;
3079 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3083 if (path->nodes[1] == NULL) {
3088 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3089 nritems = btrfs_header_nritems(path->nodes[1]);
3094 if (path->slots[1] >= nritems)
3097 /* did we find a key greater than anything we want to delete? */
3098 if (found_key.objectid > inode->i_ino ||
3099 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3102 /* we check the next key in the node to make sure the leave contains
3103 * only checksum items. This comparison doesn't work if our
3104 * leaf is the last one in the node
3106 if (path->slots[1] + 1 >= nritems) {
3108 /* search forward from the last key in the node, this
3109 * will bring us into the next node in the tree
3111 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3113 /* unlikely, but we inc below, so check to be safe */
3114 if (found_key.offset == (u64)-1)
3117 /* search_forward needs a path with locks held, do the
3118 * search again for the original key. It is possible
3119 * this will race with a balance and return a path that
3120 * we could modify, but this drop is just an optimization
3121 * and is allowed to miss some leaves.
3123 btrfs_release_path(root, path);
3126 /* setup a max key for search_forward */
3127 other_key.offset = (u64)-1;
3128 other_key.type = key.type;
3129 other_key.objectid = key.objectid;
3131 path->keep_locks = 1;
3132 ret = btrfs_search_forward(root, &found_key, &other_key,
3134 path->keep_locks = 0;
3135 if (ret || found_key.objectid != key.objectid ||
3136 found_key.type != key.type) {
3141 key.offset = found_key.offset;
3142 btrfs_release_path(root, path);
3147 /* we know there's one more slot after us in the tree,
3148 * read that key so we can verify it is also a checksum item
3150 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3152 if (found_key.objectid < inode->i_ino)
3155 if (found_key.type != key.type || found_key.offset < new_size)
3159 * if the key for the next leaf isn't a csum key from this objectid,
3160 * we can't be sure there aren't good items inside this leaf.
3163 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3166 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3167 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3169 * it is safe to delete this leaf, it contains only
3170 * csum items from this inode at an offset >= new_size
3172 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3175 if (root->ref_cows && leaf_gen < trans->transid) {
3176 ref = btrfs_alloc_leaf_ref(root, 0);
3178 ref->root_gen = root->root_key.offset;
3179 ref->bytenr = leaf_start;
3181 ref->generation = leaf_gen;
3184 btrfs_sort_leaf_ref(ref);
3186 ret = btrfs_add_leaf_ref(root, ref, 0);
3188 btrfs_free_leaf_ref(root, ref);
3194 btrfs_release_path(root, path);
3196 if (other_key.objectid == inode->i_ino &&
3197 other_key.type == key.type && other_key.offset > key.offset) {
3198 key.offset = other_key.offset;
3204 /* fixup any changes we've made to the path */
3205 path->lowest_level = 0;
3206 path->keep_locks = 0;
3207 btrfs_release_path(root, path);
3214 * this can truncate away extent items, csum items and directory items.
3215 * It starts at a high offset and removes keys until it can't find
3216 * any higher than new_size
3218 * csum items that cross the new i_size are truncated to the new size
3221 * min_type is the minimum key type to truncate down to. If set to 0, this
3222 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3224 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3225 struct btrfs_root *root,
3226 struct inode *inode,
3227 u64 new_size, u32 min_type)
3229 struct btrfs_path *path;
3230 struct extent_buffer *leaf;
3231 struct btrfs_file_extent_item *fi;
3232 struct btrfs_key key;
3233 struct btrfs_key found_key;
3234 u64 extent_start = 0;
3235 u64 extent_num_bytes = 0;
3236 u64 extent_offset = 0;
3238 u64 mask = root->sectorsize - 1;
3239 u32 found_type = (u8)-1;
3242 int pending_del_nr = 0;
3243 int pending_del_slot = 0;
3244 int extent_type = -1;
3249 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3251 if (root->ref_cows || root == root->fs_info->tree_root)
3252 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3254 path = btrfs_alloc_path();
3258 key.objectid = inode->i_ino;
3259 key.offset = (u64)-1;
3263 path->leave_spinning = 1;
3264 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3271 /* there are no items in the tree for us to truncate, we're
3274 if (path->slots[0] == 0)
3281 leaf = path->nodes[0];
3282 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3283 found_type = btrfs_key_type(&found_key);
3286 if (found_key.objectid != inode->i_ino)
3289 if (found_type < min_type)
3292 item_end = found_key.offset;
3293 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3294 fi = btrfs_item_ptr(leaf, path->slots[0],
3295 struct btrfs_file_extent_item);
3296 extent_type = btrfs_file_extent_type(leaf, fi);
3297 encoding = btrfs_file_extent_compression(leaf, fi);
3298 encoding |= btrfs_file_extent_encryption(leaf, fi);
3299 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3301 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3303 btrfs_file_extent_num_bytes(leaf, fi);
3304 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3305 item_end += btrfs_file_extent_inline_len(leaf,
3310 if (found_type > min_type) {
3313 if (item_end < new_size)
3315 if (found_key.offset >= new_size)
3321 /* FIXME, shrink the extent if the ref count is only 1 */
3322 if (found_type != BTRFS_EXTENT_DATA_KEY)
3325 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3327 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3328 if (!del_item && !encoding) {
3329 u64 orig_num_bytes =
3330 btrfs_file_extent_num_bytes(leaf, fi);
3331 extent_num_bytes = new_size -
3332 found_key.offset + root->sectorsize - 1;
3333 extent_num_bytes = extent_num_bytes &
3334 ~((u64)root->sectorsize - 1);
3335 btrfs_set_file_extent_num_bytes(leaf, fi,
3337 num_dec = (orig_num_bytes -
3339 if (root->ref_cows && extent_start != 0)
3340 inode_sub_bytes(inode, num_dec);
3341 btrfs_mark_buffer_dirty(leaf);
3344 btrfs_file_extent_disk_num_bytes(leaf,
3346 extent_offset = found_key.offset -
3347 btrfs_file_extent_offset(leaf, fi);
3349 /* FIXME blocksize != 4096 */
3350 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3351 if (extent_start != 0) {
3354 inode_sub_bytes(inode, num_dec);
3357 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3359 * we can't truncate inline items that have had
3363 btrfs_file_extent_compression(leaf, fi) == 0 &&
3364 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3365 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3366 u32 size = new_size - found_key.offset;
3368 if (root->ref_cows) {
3369 inode_sub_bytes(inode, item_end + 1 -
3373 btrfs_file_extent_calc_inline_size(size);
3374 ret = btrfs_truncate_item(trans, root, path,
3377 } else if (root->ref_cows) {
3378 inode_sub_bytes(inode, item_end + 1 -
3384 if (!pending_del_nr) {
3385 /* no pending yet, add ourselves */
3386 pending_del_slot = path->slots[0];
3388 } else if (pending_del_nr &&
3389 path->slots[0] + 1 == pending_del_slot) {
3390 /* hop on the pending chunk */
3392 pending_del_slot = path->slots[0];
3399 if (found_extent && (root->ref_cows ||
3400 root == root->fs_info->tree_root)) {
3401 btrfs_set_path_blocking(path);
3402 ret = btrfs_free_extent(trans, root, extent_start,
3403 extent_num_bytes, 0,
3404 btrfs_header_owner(leaf),
3405 inode->i_ino, extent_offset);
3409 if (found_type == BTRFS_INODE_ITEM_KEY)
3412 if (path->slots[0] == 0 ||
3413 path->slots[0] != pending_del_slot) {
3414 if (root->ref_cows) {
3418 if (pending_del_nr) {
3419 ret = btrfs_del_items(trans, root, path,
3425 btrfs_release_path(root, path);
3432 if (pending_del_nr) {
3433 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3437 btrfs_free_path(path);
3442 * taken from block_truncate_page, but does cow as it zeros out
3443 * any bytes left in the last page in the file.
3445 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3447 struct inode *inode = mapping->host;
3448 struct btrfs_root *root = BTRFS_I(inode)->root;
3449 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3450 struct btrfs_ordered_extent *ordered;
3451 struct extent_state *cached_state = NULL;
3453 u32 blocksize = root->sectorsize;
3454 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3455 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3461 if ((offset & (blocksize - 1)) == 0)
3463 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3469 page = grab_cache_page(mapping, index);
3471 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3475 page_start = page_offset(page);
3476 page_end = page_start + PAGE_CACHE_SIZE - 1;
3478 if (!PageUptodate(page)) {
3479 ret = btrfs_readpage(NULL, page);
3481 if (page->mapping != mapping) {
3483 page_cache_release(page);
3486 if (!PageUptodate(page)) {
3491 wait_on_page_writeback(page);
3493 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3495 set_page_extent_mapped(page);
3497 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3499 unlock_extent_cached(io_tree, page_start, page_end,
3500 &cached_state, GFP_NOFS);
3502 page_cache_release(page);
3503 btrfs_start_ordered_extent(inode, ordered, 1);
3504 btrfs_put_ordered_extent(ordered);
3508 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3509 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3510 0, 0, &cached_state, GFP_NOFS);
3512 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3515 unlock_extent_cached(io_tree, page_start, page_end,
3516 &cached_state, GFP_NOFS);
3521 if (offset != PAGE_CACHE_SIZE) {
3523 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3524 flush_dcache_page(page);
3527 ClearPageChecked(page);
3528 set_page_dirty(page);
3529 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3534 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3536 page_cache_release(page);
3541 int btrfs_cont_expand(struct inode *inode, loff_t size)
3543 struct btrfs_trans_handle *trans;
3544 struct btrfs_root *root = BTRFS_I(inode)->root;
3545 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3546 struct extent_map *em = NULL;
3547 struct extent_state *cached_state = NULL;
3548 u64 mask = root->sectorsize - 1;
3549 u64 hole_start = (inode->i_size + mask) & ~mask;
3550 u64 block_end = (size + mask) & ~mask;
3556 if (size <= hole_start)
3560 struct btrfs_ordered_extent *ordered;
3561 btrfs_wait_ordered_range(inode, hole_start,
3562 block_end - hole_start);
3563 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3564 &cached_state, GFP_NOFS);
3565 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3568 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3569 &cached_state, GFP_NOFS);
3570 btrfs_put_ordered_extent(ordered);
3573 cur_offset = hole_start;
3575 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3576 block_end - cur_offset, 0);
3577 BUG_ON(IS_ERR(em) || !em);
3578 last_byte = min(extent_map_end(em), block_end);
3579 last_byte = (last_byte + mask) & ~mask;
3580 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3582 hole_size = last_byte - cur_offset;
3584 trans = btrfs_start_transaction(root, 2);
3585 if (IS_ERR(trans)) {
3586 err = PTR_ERR(trans);
3589 btrfs_set_trans_block_group(trans, inode);
3591 err = btrfs_drop_extents(trans, inode, cur_offset,
3592 cur_offset + hole_size,
3596 err = btrfs_insert_file_extent(trans, root,
3597 inode->i_ino, cur_offset, 0,
3598 0, hole_size, 0, hole_size,
3602 btrfs_drop_extent_cache(inode, hole_start,
3605 btrfs_end_transaction(trans, root);
3607 free_extent_map(em);
3609 cur_offset = last_byte;
3610 if (cur_offset >= block_end)
3614 free_extent_map(em);
3615 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3620 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3622 struct btrfs_root *root = BTRFS_I(inode)->root;
3623 struct btrfs_trans_handle *trans;
3627 if (attr->ia_size == inode->i_size)
3630 if (attr->ia_size > inode->i_size) {
3631 unsigned long limit;
3632 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3633 if (attr->ia_size > inode->i_sb->s_maxbytes)
3635 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3636 send_sig(SIGXFSZ, current, 0);
3641 trans = btrfs_start_transaction(root, 5);
3643 return PTR_ERR(trans);
3645 btrfs_set_trans_block_group(trans, inode);
3647 ret = btrfs_orphan_add(trans, inode);
3650 nr = trans->blocks_used;
3651 btrfs_end_transaction(trans, root);
3652 btrfs_btree_balance_dirty(root, nr);
3654 if (attr->ia_size > inode->i_size) {
3655 ret = btrfs_cont_expand(inode, attr->ia_size);
3657 btrfs_truncate(inode);
3661 i_size_write(inode, attr->ia_size);
3662 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3664 trans = btrfs_start_transaction(root, 0);
3665 BUG_ON(IS_ERR(trans));
3666 btrfs_set_trans_block_group(trans, inode);
3667 trans->block_rsv = root->orphan_block_rsv;
3668 BUG_ON(!trans->block_rsv);
3670 ret = btrfs_update_inode(trans, root, inode);
3672 if (inode->i_nlink > 0) {
3673 ret = btrfs_orphan_del(trans, inode);
3676 nr = trans->blocks_used;
3677 btrfs_end_transaction(trans, root);
3678 btrfs_btree_balance_dirty(root, nr);
3683 * We're truncating a file that used to have good data down to
3684 * zero. Make sure it gets into the ordered flush list so that
3685 * any new writes get down to disk quickly.
3687 if (attr->ia_size == 0)
3688 BTRFS_I(inode)->ordered_data_close = 1;
3690 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3691 ret = vmtruncate(inode, attr->ia_size);
3697 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3699 struct inode *inode = dentry->d_inode;
3700 struct btrfs_root *root = BTRFS_I(inode)->root;
3703 if (btrfs_root_readonly(root))
3706 err = inode_change_ok(inode, attr);
3710 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3711 err = btrfs_setattr_size(inode, attr);
3716 if (attr->ia_valid) {
3717 setattr_copy(inode, attr);
3718 mark_inode_dirty(inode);
3720 if (attr->ia_valid & ATTR_MODE)
3721 err = btrfs_acl_chmod(inode);
3727 void btrfs_evict_inode(struct inode *inode)
3729 struct btrfs_trans_handle *trans;
3730 struct btrfs_root *root = BTRFS_I(inode)->root;
3734 truncate_inode_pages(&inode->i_data, 0);
3735 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3736 root == root->fs_info->tree_root))
3739 if (is_bad_inode(inode)) {
3740 btrfs_orphan_del(NULL, inode);
3743 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3744 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3746 if (root->fs_info->log_root_recovering) {
3747 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3751 if (inode->i_nlink > 0) {
3752 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3756 btrfs_i_size_write(inode, 0);
3759 trans = btrfs_start_transaction(root, 0);
3760 BUG_ON(IS_ERR(trans));
3761 btrfs_set_trans_block_group(trans, inode);
3762 trans->block_rsv = root->orphan_block_rsv;
3764 ret = btrfs_block_rsv_check(trans, root,
3765 root->orphan_block_rsv, 0, 5);
3767 BUG_ON(ret != -EAGAIN);
3768 ret = btrfs_commit_transaction(trans, root);
3773 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3777 nr = trans->blocks_used;
3778 btrfs_end_transaction(trans, root);
3780 btrfs_btree_balance_dirty(root, nr);
3785 ret = btrfs_orphan_del(trans, inode);
3789 nr = trans->blocks_used;
3790 btrfs_end_transaction(trans, root);
3791 btrfs_btree_balance_dirty(root, nr);
3793 end_writeback(inode);
3798 * this returns the key found in the dir entry in the location pointer.
3799 * If no dir entries were found, location->objectid is 0.
3801 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3802 struct btrfs_key *location)
3804 const char *name = dentry->d_name.name;
3805 int namelen = dentry->d_name.len;
3806 struct btrfs_dir_item *di;
3807 struct btrfs_path *path;
3808 struct btrfs_root *root = BTRFS_I(dir)->root;
3811 path = btrfs_alloc_path();
3814 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3819 if (!di || IS_ERR(di))
3822 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3824 btrfs_free_path(path);
3827 location->objectid = 0;
3832 * when we hit a tree root in a directory, the btrfs part of the inode
3833 * needs to be changed to reflect the root directory of the tree root. This
3834 * is kind of like crossing a mount point.
3836 static int fixup_tree_root_location(struct btrfs_root *root,
3838 struct dentry *dentry,
3839 struct btrfs_key *location,
3840 struct btrfs_root **sub_root)
3842 struct btrfs_path *path;
3843 struct btrfs_root *new_root;
3844 struct btrfs_root_ref *ref;
3845 struct extent_buffer *leaf;
3849 path = btrfs_alloc_path();
3856 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3857 BTRFS_I(dir)->root->root_key.objectid,
3858 location->objectid);
3865 leaf = path->nodes[0];
3866 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3867 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3868 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3871 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3872 (unsigned long)(ref + 1),
3873 dentry->d_name.len);
3877 btrfs_release_path(root->fs_info->tree_root, path);
3879 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3880 if (IS_ERR(new_root)) {
3881 err = PTR_ERR(new_root);
3885 if (btrfs_root_refs(&new_root->root_item) == 0) {
3890 *sub_root = new_root;
3891 location->objectid = btrfs_root_dirid(&new_root->root_item);
3892 location->type = BTRFS_INODE_ITEM_KEY;
3893 location->offset = 0;
3896 btrfs_free_path(path);
3900 static void inode_tree_add(struct inode *inode)
3902 struct btrfs_root *root = BTRFS_I(inode)->root;
3903 struct btrfs_inode *entry;
3905 struct rb_node *parent;
3907 p = &root->inode_tree.rb_node;
3910 if (inode_unhashed(inode))
3913 spin_lock(&root->inode_lock);
3916 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3918 if (inode->i_ino < entry->vfs_inode.i_ino)
3919 p = &parent->rb_left;
3920 else if (inode->i_ino > entry->vfs_inode.i_ino)
3921 p = &parent->rb_right;
3923 WARN_ON(!(entry->vfs_inode.i_state &
3924 (I_WILL_FREE | I_FREEING)));
3925 rb_erase(parent, &root->inode_tree);
3926 RB_CLEAR_NODE(parent);
3927 spin_unlock(&root->inode_lock);
3931 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3932 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3933 spin_unlock(&root->inode_lock);
3936 static void inode_tree_del(struct inode *inode)
3938 struct btrfs_root *root = BTRFS_I(inode)->root;
3941 spin_lock(&root->inode_lock);
3942 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3943 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3944 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3945 empty = RB_EMPTY_ROOT(&root->inode_tree);
3947 spin_unlock(&root->inode_lock);
3950 * Free space cache has inodes in the tree root, but the tree root has a
3951 * root_refs of 0, so this could end up dropping the tree root as a
3952 * snapshot, so we need the extra !root->fs_info->tree_root check to
3953 * make sure we don't drop it.
3955 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3956 root != root->fs_info->tree_root) {
3957 synchronize_srcu(&root->fs_info->subvol_srcu);
3958 spin_lock(&root->inode_lock);
3959 empty = RB_EMPTY_ROOT(&root->inode_tree);
3960 spin_unlock(&root->inode_lock);
3962 btrfs_add_dead_root(root);
3966 int btrfs_invalidate_inodes(struct btrfs_root *root)
3968 struct rb_node *node;
3969 struct rb_node *prev;
3970 struct btrfs_inode *entry;
3971 struct inode *inode;
3974 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3976 spin_lock(&root->inode_lock);
3978 node = root->inode_tree.rb_node;
3982 entry = rb_entry(node, struct btrfs_inode, rb_node);
3984 if (objectid < entry->vfs_inode.i_ino)
3985 node = node->rb_left;
3986 else if (objectid > entry->vfs_inode.i_ino)
3987 node = node->rb_right;
3993 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3994 if (objectid <= entry->vfs_inode.i_ino) {
3998 prev = rb_next(prev);
4002 entry = rb_entry(node, struct btrfs_inode, rb_node);
4003 objectid = entry->vfs_inode.i_ino + 1;
4004 inode = igrab(&entry->vfs_inode);
4006 spin_unlock(&root->inode_lock);
4007 if (atomic_read(&inode->i_count) > 1)
4008 d_prune_aliases(inode);
4010 * btrfs_drop_inode will have it removed from
4011 * the inode cache when its usage count
4016 spin_lock(&root->inode_lock);
4020 if (cond_resched_lock(&root->inode_lock))
4023 node = rb_next(node);
4025 spin_unlock(&root->inode_lock);
4029 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4031 struct btrfs_iget_args *args = p;
4032 inode->i_ino = args->ino;
4033 BTRFS_I(inode)->root = args->root;
4034 btrfs_set_inode_space_info(args->root, inode);
4038 static int btrfs_find_actor(struct inode *inode, void *opaque)
4040 struct btrfs_iget_args *args = opaque;
4041 return args->ino == inode->i_ino &&
4042 args->root == BTRFS_I(inode)->root;
4045 static struct inode *btrfs_iget_locked(struct super_block *s,
4047 struct btrfs_root *root)
4049 struct inode *inode;
4050 struct btrfs_iget_args args;
4051 args.ino = objectid;
4054 inode = iget5_locked(s, objectid, btrfs_find_actor,
4055 btrfs_init_locked_inode,
4060 /* Get an inode object given its location and corresponding root.
4061 * Returns in *is_new if the inode was read from disk
4063 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4064 struct btrfs_root *root, int *new)
4066 struct inode *inode;
4068 inode = btrfs_iget_locked(s, location->objectid, root);
4070 return ERR_PTR(-ENOMEM);
4072 if (inode->i_state & I_NEW) {
4073 BTRFS_I(inode)->root = root;
4074 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4075 btrfs_read_locked_inode(inode);
4077 inode_tree_add(inode);
4078 unlock_new_inode(inode);
4086 static struct inode *new_simple_dir(struct super_block *s,
4087 struct btrfs_key *key,
4088 struct btrfs_root *root)
4090 struct inode *inode = new_inode(s);
4093 return ERR_PTR(-ENOMEM);
4095 BTRFS_I(inode)->root = root;
4096 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4097 BTRFS_I(inode)->dummy_inode = 1;
4099 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4100 inode->i_op = &simple_dir_inode_operations;
4101 inode->i_fop = &simple_dir_operations;
4102 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4103 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4108 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4110 struct inode *inode;
4111 struct btrfs_root *root = BTRFS_I(dir)->root;
4112 struct btrfs_root *sub_root = root;
4113 struct btrfs_key location;
4117 if (dentry->d_name.len > BTRFS_NAME_LEN)
4118 return ERR_PTR(-ENAMETOOLONG);
4120 ret = btrfs_inode_by_name(dir, dentry, &location);
4123 return ERR_PTR(ret);
4125 if (location.objectid == 0)
4128 if (location.type == BTRFS_INODE_ITEM_KEY) {
4129 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4133 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4135 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4136 ret = fixup_tree_root_location(root, dir, dentry,
4137 &location, &sub_root);
4140 inode = ERR_PTR(ret);
4142 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4144 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4146 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4148 if (!IS_ERR(inode) && root != sub_root) {
4149 down_read(&root->fs_info->cleanup_work_sem);
4150 if (!(inode->i_sb->s_flags & MS_RDONLY))
4151 btrfs_orphan_cleanup(sub_root);
4152 up_read(&root->fs_info->cleanup_work_sem);
4158 static int btrfs_dentry_delete(const struct dentry *dentry)
4160 struct btrfs_root *root;
4162 if (!dentry->d_inode && !IS_ROOT(dentry))
4163 dentry = dentry->d_parent;
4165 if (dentry->d_inode) {
4166 root = BTRFS_I(dentry->d_inode)->root;
4167 if (btrfs_root_refs(&root->root_item) == 0)
4173 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4174 struct nameidata *nd)
4176 struct inode *inode;
4178 inode = btrfs_lookup_dentry(dir, dentry);
4180 return ERR_CAST(inode);
4182 return d_splice_alias(inode, dentry);
4185 static unsigned char btrfs_filetype_table[] = {
4186 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4189 static int btrfs_real_readdir(struct file *filp, void *dirent,
4192 struct inode *inode = filp->f_dentry->d_inode;
4193 struct btrfs_root *root = BTRFS_I(inode)->root;
4194 struct btrfs_item *item;
4195 struct btrfs_dir_item *di;
4196 struct btrfs_key key;
4197 struct btrfs_key found_key;
4198 struct btrfs_path *path;
4201 struct extent_buffer *leaf;
4204 unsigned char d_type;
4209 int key_type = BTRFS_DIR_INDEX_KEY;
4214 /* FIXME, use a real flag for deciding about the key type */
4215 if (root->fs_info->tree_root == root)
4216 key_type = BTRFS_DIR_ITEM_KEY;
4218 /* special case for "." */
4219 if (filp->f_pos == 0) {
4220 over = filldir(dirent, ".", 1,
4227 /* special case for .., just use the back ref */
4228 if (filp->f_pos == 1) {
4229 u64 pino = parent_ino(filp->f_path.dentry);
4230 over = filldir(dirent, "..", 2,
4236 path = btrfs_alloc_path();
4239 btrfs_set_key_type(&key, key_type);
4240 key.offset = filp->f_pos;
4241 key.objectid = inode->i_ino;
4243 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4249 leaf = path->nodes[0];
4250 nritems = btrfs_header_nritems(leaf);
4251 slot = path->slots[0];
4252 if (advance || slot >= nritems) {
4253 if (slot >= nritems - 1) {
4254 ret = btrfs_next_leaf(root, path);
4257 leaf = path->nodes[0];
4258 nritems = btrfs_header_nritems(leaf);
4259 slot = path->slots[0];
4267 item = btrfs_item_nr(leaf, slot);
4268 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4270 if (found_key.objectid != key.objectid)
4272 if (btrfs_key_type(&found_key) != key_type)
4274 if (found_key.offset < filp->f_pos)
4277 filp->f_pos = found_key.offset;
4279 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4281 di_total = btrfs_item_size(leaf, item);
4283 while (di_cur < di_total) {
4284 struct btrfs_key location;
4286 name_len = btrfs_dir_name_len(leaf, di);
4287 if (name_len <= sizeof(tmp_name)) {
4288 name_ptr = tmp_name;
4290 name_ptr = kmalloc(name_len, GFP_NOFS);
4296 read_extent_buffer(leaf, name_ptr,
4297 (unsigned long)(di + 1), name_len);
4299 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4300 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4302 /* is this a reference to our own snapshot? If so
4305 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4306 location.objectid == root->root_key.objectid) {
4310 over = filldir(dirent, name_ptr, name_len,
4311 found_key.offset, location.objectid,
4315 if (name_ptr != tmp_name)
4320 di_len = btrfs_dir_name_len(leaf, di) +
4321 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4323 di = (struct btrfs_dir_item *)((char *)di + di_len);
4327 /* Reached end of directory/root. Bump pos past the last item. */
4328 if (key_type == BTRFS_DIR_INDEX_KEY)
4330 * 32-bit glibc will use getdents64, but then strtol -
4331 * so the last number we can serve is this.
4333 filp->f_pos = 0x7fffffff;
4339 btrfs_free_path(path);
4343 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4346 struct btrfs_trans_handle *trans;
4348 bool nolock = false;
4350 if (BTRFS_I(inode)->dummy_inode)
4354 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4356 if (wbc->sync_mode == WB_SYNC_ALL) {
4358 trans = btrfs_join_transaction_nolock(root, 1);
4360 trans = btrfs_join_transaction(root, 1);
4362 return PTR_ERR(trans);
4363 btrfs_set_trans_block_group(trans, inode);
4365 ret = btrfs_end_transaction_nolock(trans, root);
4367 ret = btrfs_commit_transaction(trans, root);
4373 * This is somewhat expensive, updating the tree every time the
4374 * inode changes. But, it is most likely to find the inode in cache.
4375 * FIXME, needs more benchmarking...there are no reasons other than performance
4376 * to keep or drop this code.
4378 void btrfs_dirty_inode(struct inode *inode)
4380 struct btrfs_root *root = BTRFS_I(inode)->root;
4381 struct btrfs_trans_handle *trans;
4384 if (BTRFS_I(inode)->dummy_inode)
4387 trans = btrfs_join_transaction(root, 1);
4388 BUG_ON(IS_ERR(trans));
4389 btrfs_set_trans_block_group(trans, inode);
4391 ret = btrfs_update_inode(trans, root, inode);
4392 if (ret && ret == -ENOSPC) {
4393 /* whoops, lets try again with the full transaction */
4394 btrfs_end_transaction(trans, root);
4395 trans = btrfs_start_transaction(root, 1);
4396 if (IS_ERR(trans)) {
4397 if (printk_ratelimit()) {
4398 printk(KERN_ERR "btrfs: fail to "
4399 "dirty inode %lu error %ld\n",
4400 inode->i_ino, PTR_ERR(trans));
4404 btrfs_set_trans_block_group(trans, inode);
4406 ret = btrfs_update_inode(trans, root, inode);
4408 if (printk_ratelimit()) {
4409 printk(KERN_ERR "btrfs: fail to "
4410 "dirty inode %lu error %d\n",
4415 btrfs_end_transaction(trans, root);
4419 * find the highest existing sequence number in a directory
4420 * and then set the in-memory index_cnt variable to reflect
4421 * free sequence numbers
4423 static int btrfs_set_inode_index_count(struct inode *inode)
4425 struct btrfs_root *root = BTRFS_I(inode)->root;
4426 struct btrfs_key key, found_key;
4427 struct btrfs_path *path;
4428 struct extent_buffer *leaf;
4431 key.objectid = inode->i_ino;
4432 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4433 key.offset = (u64)-1;
4435 path = btrfs_alloc_path();
4439 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4442 /* FIXME: we should be able to handle this */
4448 * MAGIC NUMBER EXPLANATION:
4449 * since we search a directory based on f_pos we have to start at 2
4450 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4451 * else has to start at 2
4453 if (path->slots[0] == 0) {
4454 BTRFS_I(inode)->index_cnt = 2;
4460 leaf = path->nodes[0];
4461 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4463 if (found_key.objectid != inode->i_ino ||
4464 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4465 BTRFS_I(inode)->index_cnt = 2;
4469 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4471 btrfs_free_path(path);
4476 * helper to find a free sequence number in a given directory. This current
4477 * code is very simple, later versions will do smarter things in the btree
4479 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4483 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4484 ret = btrfs_set_inode_index_count(dir);
4489 *index = BTRFS_I(dir)->index_cnt;
4490 BTRFS_I(dir)->index_cnt++;
4495 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4496 struct btrfs_root *root,
4498 const char *name, int name_len,
4499 u64 ref_objectid, u64 objectid,
4500 u64 alloc_hint, int mode, u64 *index)
4502 struct inode *inode;
4503 struct btrfs_inode_item *inode_item;
4504 struct btrfs_key *location;
4505 struct btrfs_path *path;
4506 struct btrfs_inode_ref *ref;
4507 struct btrfs_key key[2];
4513 path = btrfs_alloc_path();
4516 inode = new_inode(root->fs_info->sb);
4518 return ERR_PTR(-ENOMEM);
4521 ret = btrfs_set_inode_index(dir, index);
4524 return ERR_PTR(ret);
4528 * index_cnt is ignored for everything but a dir,
4529 * btrfs_get_inode_index_count has an explanation for the magic
4532 BTRFS_I(inode)->index_cnt = 2;
4533 BTRFS_I(inode)->root = root;
4534 BTRFS_I(inode)->generation = trans->transid;
4535 inode->i_generation = BTRFS_I(inode)->generation;
4536 btrfs_set_inode_space_info(root, inode);
4542 BTRFS_I(inode)->block_group =
4543 btrfs_find_block_group(root, 0, alloc_hint, owner);
4545 key[0].objectid = objectid;
4546 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4549 key[1].objectid = objectid;
4550 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4551 key[1].offset = ref_objectid;
4553 sizes[0] = sizeof(struct btrfs_inode_item);
4554 sizes[1] = name_len + sizeof(*ref);
4556 path->leave_spinning = 1;
4557 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4561 inode_init_owner(inode, dir, mode);
4562 inode->i_ino = objectid;
4563 inode_set_bytes(inode, 0);
4564 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4565 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4566 struct btrfs_inode_item);
4567 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4569 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4570 struct btrfs_inode_ref);
4571 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4572 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4573 ptr = (unsigned long)(ref + 1);
4574 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4576 btrfs_mark_buffer_dirty(path->nodes[0]);
4577 btrfs_free_path(path);
4579 location = &BTRFS_I(inode)->location;
4580 location->objectid = objectid;
4581 location->offset = 0;
4582 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4584 btrfs_inherit_iflags(inode, dir);
4586 if ((mode & S_IFREG)) {
4587 if (btrfs_test_opt(root, NODATASUM))
4588 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4589 if (btrfs_test_opt(root, NODATACOW))
4590 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4593 insert_inode_hash(inode);
4594 inode_tree_add(inode);
4598 BTRFS_I(dir)->index_cnt--;
4599 btrfs_free_path(path);
4601 return ERR_PTR(ret);
4604 static inline u8 btrfs_inode_type(struct inode *inode)
4606 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4610 * utility function to add 'inode' into 'parent_inode' with
4611 * a give name and a given sequence number.
4612 * if 'add_backref' is true, also insert a backref from the
4613 * inode to the parent directory.
4615 int btrfs_add_link(struct btrfs_trans_handle *trans,
4616 struct inode *parent_inode, struct inode *inode,
4617 const char *name, int name_len, int add_backref, u64 index)
4620 struct btrfs_key key;
4621 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4623 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4624 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4626 key.objectid = inode->i_ino;
4627 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4631 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4632 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4633 key.objectid, root->root_key.objectid,
4634 parent_inode->i_ino,
4635 index, name, name_len);
4636 } else if (add_backref) {
4637 ret = btrfs_insert_inode_ref(trans, root,
4638 name, name_len, inode->i_ino,
4639 parent_inode->i_ino, index);
4643 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4644 parent_inode->i_ino, &key,
4645 btrfs_inode_type(inode), index);
4648 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4650 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4651 ret = btrfs_update_inode(trans, root, parent_inode);
4656 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4657 struct inode *dir, struct dentry *dentry,
4658 struct inode *inode, int backref, u64 index)
4660 int err = btrfs_add_link(trans, dir, inode,
4661 dentry->d_name.name, dentry->d_name.len,
4664 d_instantiate(dentry, inode);
4672 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4673 int mode, dev_t rdev)
4675 struct btrfs_trans_handle *trans;
4676 struct btrfs_root *root = BTRFS_I(dir)->root;
4677 struct inode *inode = NULL;
4681 unsigned long nr = 0;
4684 if (!new_valid_dev(rdev))
4687 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4692 * 2 for inode item and ref
4694 * 1 for xattr if selinux is on
4696 trans = btrfs_start_transaction(root, 5);
4698 return PTR_ERR(trans);
4700 btrfs_set_trans_block_group(trans, dir);
4702 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4703 dentry->d_name.len, dir->i_ino, objectid,
4704 BTRFS_I(dir)->block_group, mode, &index);
4705 err = PTR_ERR(inode);
4709 err = btrfs_init_inode_security(trans, inode, dir);
4715 btrfs_set_trans_block_group(trans, inode);
4716 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4720 inode->i_op = &btrfs_special_inode_operations;
4721 init_special_inode(inode, inode->i_mode, rdev);
4722 btrfs_update_inode(trans, root, inode);
4724 btrfs_update_inode_block_group(trans, inode);
4725 btrfs_update_inode_block_group(trans, dir);
4727 nr = trans->blocks_used;
4728 btrfs_end_transaction_throttle(trans, root);
4729 btrfs_btree_balance_dirty(root, nr);
4731 inode_dec_link_count(inode);
4737 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4738 int mode, struct nameidata *nd)
4740 struct btrfs_trans_handle *trans;
4741 struct btrfs_root *root = BTRFS_I(dir)->root;
4742 struct inode *inode = NULL;
4745 unsigned long nr = 0;
4749 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4753 * 2 for inode item and ref
4755 * 1 for xattr if selinux is on
4757 trans = btrfs_start_transaction(root, 5);
4759 return PTR_ERR(trans);
4761 btrfs_set_trans_block_group(trans, dir);
4763 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4764 dentry->d_name.len, dir->i_ino, objectid,
4765 BTRFS_I(dir)->block_group, mode, &index);
4766 err = PTR_ERR(inode);
4770 err = btrfs_init_inode_security(trans, inode, dir);
4776 btrfs_set_trans_block_group(trans, inode);
4777 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4781 inode->i_mapping->a_ops = &btrfs_aops;
4782 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4783 inode->i_fop = &btrfs_file_operations;
4784 inode->i_op = &btrfs_file_inode_operations;
4785 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4787 btrfs_update_inode_block_group(trans, inode);
4788 btrfs_update_inode_block_group(trans, dir);
4790 nr = trans->blocks_used;
4791 btrfs_end_transaction_throttle(trans, root);
4793 inode_dec_link_count(inode);
4796 btrfs_btree_balance_dirty(root, nr);
4800 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4801 struct dentry *dentry)
4803 struct btrfs_trans_handle *trans;
4804 struct btrfs_root *root = BTRFS_I(dir)->root;
4805 struct inode *inode = old_dentry->d_inode;
4807 unsigned long nr = 0;
4811 if (inode->i_nlink == 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 btrfs_inc_nlink(inode);
4819 inode->i_ctime = CURRENT_TIME;
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_set_trans_block_group(trans, dir);
4839 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4844 struct dentry *parent = dget_parent(dentry);
4845 btrfs_update_inode_block_group(trans, dir);
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);
4886 btrfs_set_trans_block_group(trans, dir);
4888 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4889 dentry->d_name.len, dir->i_ino, objectid,
4890 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4892 if (IS_ERR(inode)) {
4893 err = PTR_ERR(inode);
4899 err = btrfs_init_inode_security(trans, inode, dir);
4903 inode->i_op = &btrfs_dir_inode_operations;
4904 inode->i_fop = &btrfs_dir_file_operations;
4905 btrfs_set_trans_block_group(trans, inode);
4907 btrfs_i_size_write(inode, 0);
4908 err = btrfs_update_inode(trans, root, inode);
4912 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4913 dentry->d_name.len, 0, index);
4917 d_instantiate(dentry, inode);
4919 btrfs_update_inode_block_group(trans, inode);
4920 btrfs_update_inode_block_group(trans, dir);
4923 nr = trans->blocks_used;
4924 btrfs_end_transaction_throttle(trans, root);
4927 btrfs_btree_balance_dirty(root, nr);
4931 /* helper for btfs_get_extent. Given an existing extent in the tree,
4932 * and an extent that you want to insert, deal with overlap and insert
4933 * the new extent into the tree.
4935 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4936 struct extent_map *existing,
4937 struct extent_map *em,
4938 u64 map_start, u64 map_len)
4942 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4943 start_diff = map_start - em->start;
4944 em->start = map_start;
4946 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4947 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4948 em->block_start += start_diff;
4949 em->block_len -= start_diff;
4951 return add_extent_mapping(em_tree, em);
4954 static noinline int uncompress_inline(struct btrfs_path *path,
4955 struct inode *inode, struct page *page,
4956 size_t pg_offset, u64 extent_offset,
4957 struct btrfs_file_extent_item *item)
4960 struct extent_buffer *leaf = path->nodes[0];
4963 unsigned long inline_size;
4967 WARN_ON(pg_offset != 0);
4968 compress_type = btrfs_file_extent_compression(leaf, item);
4969 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4970 inline_size = btrfs_file_extent_inline_item_len(leaf,
4971 btrfs_item_nr(leaf, path->slots[0]));
4972 tmp = kmalloc(inline_size, GFP_NOFS);
4973 ptr = btrfs_file_extent_inline_start(item);
4975 read_extent_buffer(leaf, tmp, ptr, inline_size);
4977 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4978 ret = btrfs_decompress(compress_type, tmp, page,
4979 extent_offset, inline_size, max_size);
4981 char *kaddr = kmap_atomic(page, KM_USER0);
4982 unsigned long copy_size = min_t(u64,
4983 PAGE_CACHE_SIZE - pg_offset,
4984 max_size - extent_offset);
4985 memset(kaddr + pg_offset, 0, copy_size);
4986 kunmap_atomic(kaddr, KM_USER0);
4993 * a bit scary, this does extent mapping from logical file offset to the disk.
4994 * the ugly parts come from merging extents from the disk with the in-ram
4995 * representation. This gets more complex because of the data=ordered code,
4996 * where the in-ram extents might be locked pending data=ordered completion.
4998 * This also copies inline extents directly into the page.
5001 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5002 size_t pg_offset, u64 start, u64 len,
5008 u64 extent_start = 0;
5010 u64 objectid = inode->i_ino;
5012 struct btrfs_path *path = NULL;
5013 struct btrfs_root *root = BTRFS_I(inode)->root;
5014 struct btrfs_file_extent_item *item;
5015 struct extent_buffer *leaf;
5016 struct btrfs_key found_key;
5017 struct extent_map *em = NULL;
5018 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5019 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5020 struct btrfs_trans_handle *trans = NULL;
5024 read_lock(&em_tree->lock);
5025 em = lookup_extent_mapping(em_tree, start, len);
5027 em->bdev = root->fs_info->fs_devices->latest_bdev;
5028 read_unlock(&em_tree->lock);
5031 if (em->start > start || em->start + em->len <= start)
5032 free_extent_map(em);
5033 else if (em->block_start == EXTENT_MAP_INLINE && page)
5034 free_extent_map(em);
5038 em = alloc_extent_map(GFP_NOFS);
5043 em->bdev = root->fs_info->fs_devices->latest_bdev;
5044 em->start = EXTENT_MAP_HOLE;
5045 em->orig_start = EXTENT_MAP_HOLE;
5047 em->block_len = (u64)-1;
5050 path = btrfs_alloc_path();
5054 ret = btrfs_lookup_file_extent(trans, root, path,
5055 objectid, start, trans != NULL);
5062 if (path->slots[0] == 0)
5067 leaf = path->nodes[0];
5068 item = btrfs_item_ptr(leaf, path->slots[0],
5069 struct btrfs_file_extent_item);
5070 /* are we inside the extent that was found? */
5071 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5072 found_type = btrfs_key_type(&found_key);
5073 if (found_key.objectid != objectid ||
5074 found_type != BTRFS_EXTENT_DATA_KEY) {
5078 found_type = btrfs_file_extent_type(leaf, item);
5079 extent_start = found_key.offset;
5080 compress_type = btrfs_file_extent_compression(leaf, item);
5081 if (found_type == BTRFS_FILE_EXTENT_REG ||
5082 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5083 extent_end = extent_start +
5084 btrfs_file_extent_num_bytes(leaf, item);
5085 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5087 size = btrfs_file_extent_inline_len(leaf, item);
5088 extent_end = (extent_start + size + root->sectorsize - 1) &
5089 ~((u64)root->sectorsize - 1);
5092 if (start >= extent_end) {
5094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5095 ret = btrfs_next_leaf(root, path);
5102 leaf = path->nodes[0];
5104 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5105 if (found_key.objectid != objectid ||
5106 found_key.type != BTRFS_EXTENT_DATA_KEY)
5108 if (start + len <= found_key.offset)
5111 em->len = found_key.offset - start;
5115 if (found_type == BTRFS_FILE_EXTENT_REG ||
5116 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5117 em->start = extent_start;
5118 em->len = extent_end - extent_start;
5119 em->orig_start = extent_start -
5120 btrfs_file_extent_offset(leaf, item);
5121 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5123 em->block_start = EXTENT_MAP_HOLE;
5126 if (compress_type != BTRFS_COMPRESS_NONE) {
5127 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5128 em->compress_type = compress_type;
5129 em->block_start = bytenr;
5130 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5133 bytenr += btrfs_file_extent_offset(leaf, item);
5134 em->block_start = bytenr;
5135 em->block_len = em->len;
5136 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5137 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5140 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5144 size_t extent_offset;
5147 em->block_start = EXTENT_MAP_INLINE;
5148 if (!page || create) {
5149 em->start = extent_start;
5150 em->len = extent_end - extent_start;
5154 size = btrfs_file_extent_inline_len(leaf, item);
5155 extent_offset = page_offset(page) + pg_offset - extent_start;
5156 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5157 size - extent_offset);
5158 em->start = extent_start + extent_offset;
5159 em->len = (copy_size + root->sectorsize - 1) &
5160 ~((u64)root->sectorsize - 1);
5161 em->orig_start = EXTENT_MAP_INLINE;
5162 if (compress_type) {
5163 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5164 em->compress_type = compress_type;
5166 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5167 if (create == 0 && !PageUptodate(page)) {
5168 if (btrfs_file_extent_compression(leaf, item) !=
5169 BTRFS_COMPRESS_NONE) {
5170 ret = uncompress_inline(path, inode, page,
5172 extent_offset, item);
5176 read_extent_buffer(leaf, map + pg_offset, ptr,
5178 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5179 memset(map + pg_offset + copy_size, 0,
5180 PAGE_CACHE_SIZE - pg_offset -
5185 flush_dcache_page(page);
5186 } else if (create && PageUptodate(page)) {
5190 free_extent_map(em);
5192 btrfs_release_path(root, path);
5193 trans = btrfs_join_transaction(root, 1);
5195 return ERR_CAST(trans);
5199 write_extent_buffer(leaf, map + pg_offset, ptr,
5202 btrfs_mark_buffer_dirty(leaf);
5204 set_extent_uptodate(io_tree, em->start,
5205 extent_map_end(em) - 1, GFP_NOFS);
5208 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5215 em->block_start = EXTENT_MAP_HOLE;
5216 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5218 btrfs_release_path(root, path);
5219 if (em->start > start || extent_map_end(em) <= start) {
5220 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5221 "[%llu %llu]\n", (unsigned long long)em->start,
5222 (unsigned long long)em->len,
5223 (unsigned long long)start,
5224 (unsigned long long)len);
5230 write_lock(&em_tree->lock);
5231 ret = add_extent_mapping(em_tree, em);
5232 /* it is possible that someone inserted the extent into the tree
5233 * while we had the lock dropped. It is also possible that
5234 * an overlapping map exists in the tree
5236 if (ret == -EEXIST) {
5237 struct extent_map *existing;
5241 existing = lookup_extent_mapping(em_tree, start, len);
5242 if (existing && (existing->start > start ||
5243 existing->start + existing->len <= start)) {
5244 free_extent_map(existing);
5248 existing = lookup_extent_mapping(em_tree, em->start,
5251 err = merge_extent_mapping(em_tree, existing,
5254 free_extent_map(existing);
5256 free_extent_map(em);
5261 free_extent_map(em);
5265 free_extent_map(em);
5270 write_unlock(&em_tree->lock);
5273 btrfs_free_path(path);
5275 ret = btrfs_end_transaction(trans, root);
5280 free_extent_map(em);
5281 return ERR_PTR(err);
5286 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5287 size_t pg_offset, u64 start, u64 len,
5290 struct extent_map *em;
5291 struct extent_map *hole_em = NULL;
5292 u64 range_start = start;
5298 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5303 * if our em maps to a hole, there might
5304 * actually be delalloc bytes behind it
5306 if (em->block_start != EXTENT_MAP_HOLE)
5312 /* check to see if we've wrapped (len == -1 or similar) */
5321 /* ok, we didn't find anything, lets look for delalloc */
5322 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5323 end, len, EXTENT_DELALLOC, 1);
5324 found_end = range_start + found;
5325 if (found_end < range_start)
5326 found_end = (u64)-1;
5329 * we didn't find anything useful, return
5330 * the original results from get_extent()
5332 if (range_start > end || found_end <= start) {
5338 /* adjust the range_start to make sure it doesn't
5339 * go backwards from the start they passed in
5341 range_start = max(start,range_start);
5342 found = found_end - range_start;
5345 u64 hole_start = start;
5348 em = alloc_extent_map(GFP_NOFS);
5354 * when btrfs_get_extent can't find anything it
5355 * returns one huge hole
5357 * make sure what it found really fits our range, and
5358 * adjust to make sure it is based on the start from
5362 u64 calc_end = extent_map_end(hole_em);
5364 if (calc_end <= start || (hole_em->start > end)) {
5365 free_extent_map(hole_em);
5368 hole_start = max(hole_em->start, start);
5369 hole_len = calc_end - hole_start;
5373 if (hole_em && range_start > hole_start) {
5374 /* our hole starts before our delalloc, so we
5375 * have to return just the parts of the hole
5376 * that go until the delalloc starts
5378 em->len = min(hole_len,
5379 range_start - hole_start);
5380 em->start = hole_start;
5381 em->orig_start = hole_start;
5383 * don't adjust block start at all,
5384 * it is fixed at EXTENT_MAP_HOLE
5386 em->block_start = hole_em->block_start;
5387 em->block_len = hole_len;
5389 em->start = range_start;
5391 em->orig_start = range_start;
5392 em->block_start = EXTENT_MAP_DELALLOC;
5393 em->block_len = found;
5395 } else if (hole_em) {
5400 free_extent_map(hole_em);
5402 free_extent_map(em);
5403 return ERR_PTR(err);
5408 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5411 struct btrfs_root *root = BTRFS_I(inode)->root;
5412 struct btrfs_trans_handle *trans;
5413 struct extent_map *em;
5414 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5415 struct btrfs_key ins;
5419 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5421 trans = btrfs_join_transaction(root, 0);
5423 return ERR_CAST(trans);
5425 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5427 alloc_hint = get_extent_allocation_hint(inode, start, len);
5428 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5429 alloc_hint, (u64)-1, &ins, 1);
5435 em = alloc_extent_map(GFP_NOFS);
5437 em = ERR_PTR(-ENOMEM);
5442 em->orig_start = em->start;
5443 em->len = ins.offset;
5445 em->block_start = ins.objectid;
5446 em->block_len = ins.offset;
5447 em->bdev = root->fs_info->fs_devices->latest_bdev;
5448 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5451 write_lock(&em_tree->lock);
5452 ret = add_extent_mapping(em_tree, em);
5453 write_unlock(&em_tree->lock);
5456 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5459 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5460 ins.offset, ins.offset, 0);
5462 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5466 btrfs_end_transaction(trans, root);
5471 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5472 * block must be cow'd
5474 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5475 struct inode *inode, u64 offset, u64 len)
5477 struct btrfs_path *path;
5479 struct extent_buffer *leaf;
5480 struct btrfs_root *root = BTRFS_I(inode)->root;
5481 struct btrfs_file_extent_item *fi;
5482 struct btrfs_key key;
5490 path = btrfs_alloc_path();
5494 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5499 slot = path->slots[0];
5502 /* can't find the item, must cow */
5509 leaf = path->nodes[0];
5510 btrfs_item_key_to_cpu(leaf, &key, slot);
5511 if (key.objectid != inode->i_ino ||
5512 key.type != BTRFS_EXTENT_DATA_KEY) {
5513 /* not our file or wrong item type, must cow */
5517 if (key.offset > offset) {
5518 /* Wrong offset, must cow */
5522 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5523 found_type = btrfs_file_extent_type(leaf, fi);
5524 if (found_type != BTRFS_FILE_EXTENT_REG &&
5525 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5526 /* not a regular extent, must cow */
5529 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5530 backref_offset = btrfs_file_extent_offset(leaf, fi);
5532 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5533 if (extent_end < offset + len) {
5534 /* extent doesn't include our full range, must cow */
5538 if (btrfs_extent_readonly(root, disk_bytenr))
5542 * look for other files referencing this extent, if we
5543 * find any we must cow
5545 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5546 key.offset - backref_offset, disk_bytenr))
5550 * adjust disk_bytenr and num_bytes to cover just the bytes
5551 * in this extent we are about to write. If there
5552 * are any csums in that range we have to cow in order
5553 * to keep the csums correct
5555 disk_bytenr += backref_offset;
5556 disk_bytenr += offset - key.offset;
5557 num_bytes = min(offset + len, extent_end) - offset;
5558 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5561 * all of the above have passed, it is safe to overwrite this extent
5566 btrfs_free_path(path);
5570 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5571 struct buffer_head *bh_result, int create)
5573 struct extent_map *em;
5574 struct btrfs_root *root = BTRFS_I(inode)->root;
5575 u64 start = iblock << inode->i_blkbits;
5576 u64 len = bh_result->b_size;
5577 struct btrfs_trans_handle *trans;
5579 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5584 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5585 * io. INLINE is special, and we could probably kludge it in here, but
5586 * it's still buffered so for safety lets just fall back to the generic
5589 * For COMPRESSED we _have_ to read the entire extent in so we can
5590 * decompress it, so there will be buffering required no matter what we
5591 * do, so go ahead and fallback to buffered.
5593 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5594 * to buffered IO. Don't blame me, this is the price we pay for using
5597 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5598 em->block_start == EXTENT_MAP_INLINE) {
5599 free_extent_map(em);
5603 /* Just a good old fashioned hole, return */
5604 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5605 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5606 free_extent_map(em);
5607 /* DIO will do one hole at a time, so just unlock a sector */
5608 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5609 start + root->sectorsize - 1, GFP_NOFS);
5614 * We don't allocate a new extent in the following cases
5616 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5618 * 2) The extent is marked as PREALLOC. We're good to go here and can
5619 * just use the extent.
5623 len = em->len - (start - em->start);
5627 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5628 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5629 em->block_start != EXTENT_MAP_HOLE)) {
5634 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5635 type = BTRFS_ORDERED_PREALLOC;
5637 type = BTRFS_ORDERED_NOCOW;
5638 len = min(len, em->len - (start - em->start));
5639 block_start = em->block_start + (start - em->start);
5642 * we're not going to log anything, but we do need
5643 * to make sure the current transaction stays open
5644 * while we look for nocow cross refs
5646 trans = btrfs_join_transaction(root, 0);
5650 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5651 ret = btrfs_add_ordered_extent_dio(inode, start,
5652 block_start, len, len, type);
5653 btrfs_end_transaction(trans, root);
5655 free_extent_map(em);
5660 btrfs_end_transaction(trans, root);
5664 * this will cow the extent, reset the len in case we changed
5667 len = bh_result->b_size;
5668 free_extent_map(em);
5669 em = btrfs_new_extent_direct(inode, start, len);
5672 len = min(len, em->len - (start - em->start));
5674 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5675 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5678 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5680 bh_result->b_size = len;
5681 bh_result->b_bdev = em->bdev;
5682 set_buffer_mapped(bh_result);
5683 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5684 set_buffer_new(bh_result);
5686 free_extent_map(em);
5691 struct btrfs_dio_private {
5692 struct inode *inode;
5699 /* number of bios pending for this dio */
5700 atomic_t pending_bios;
5705 struct bio *orig_bio;
5708 static void btrfs_endio_direct_read(struct bio *bio, int err)
5710 struct btrfs_dio_private *dip = bio->bi_private;
5711 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5712 struct bio_vec *bvec = bio->bi_io_vec;
5713 struct inode *inode = dip->inode;
5714 struct btrfs_root *root = BTRFS_I(inode)->root;
5716 u32 *private = dip->csums;
5718 start = dip->logical_offset;
5720 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5721 struct page *page = bvec->bv_page;
5724 unsigned long flags;
5726 local_irq_save(flags);
5727 kaddr = kmap_atomic(page, KM_IRQ0);
5728 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5729 csum, bvec->bv_len);
5730 btrfs_csum_final(csum, (char *)&csum);
5731 kunmap_atomic(kaddr, KM_IRQ0);
5732 local_irq_restore(flags);
5734 flush_dcache_page(bvec->bv_page);
5735 if (csum != *private) {
5736 printk(KERN_ERR "btrfs csum failed ino %lu off"
5737 " %llu csum %u private %u\n",
5738 inode->i_ino, (unsigned long long)start,
5744 start += bvec->bv_len;
5747 } while (bvec <= bvec_end);
5749 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5750 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5751 bio->bi_private = dip->private;
5755 dio_end_io(bio, err);
5758 static void btrfs_endio_direct_write(struct bio *bio, int err)
5760 struct btrfs_dio_private *dip = bio->bi_private;
5761 struct inode *inode = dip->inode;
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5763 struct btrfs_trans_handle *trans;
5764 struct btrfs_ordered_extent *ordered = NULL;
5765 struct extent_state *cached_state = NULL;
5766 u64 ordered_offset = dip->logical_offset;
5767 u64 ordered_bytes = dip->bytes;
5773 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5781 trans = btrfs_join_transaction(root, 1);
5782 if (IS_ERR(trans)) {
5786 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5788 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5789 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5791 ret = btrfs_update_inode(trans, root, inode);
5796 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5797 ordered->file_offset + ordered->len - 1, 0,
5798 &cached_state, GFP_NOFS);
5800 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5801 ret = btrfs_mark_extent_written(trans, inode,
5802 ordered->file_offset,
5803 ordered->file_offset +
5810 ret = insert_reserved_file_extent(trans, inode,
5811 ordered->file_offset,
5817 BTRFS_FILE_EXTENT_REG);
5818 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5819 ordered->file_offset, ordered->len);
5827 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5828 btrfs_ordered_update_i_size(inode, 0, ordered);
5829 btrfs_update_inode(trans, root, inode);
5831 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5832 ordered->file_offset + ordered->len - 1,
5833 &cached_state, GFP_NOFS);
5835 btrfs_delalloc_release_metadata(inode, ordered->len);
5836 btrfs_end_transaction(trans, root);
5837 ordered_offset = ordered->file_offset + ordered->len;
5838 btrfs_put_ordered_extent(ordered);
5839 btrfs_put_ordered_extent(ordered);
5843 * our bio might span multiple ordered extents. If we haven't
5844 * completed the accounting for the whole dio, go back and try again
5846 if (ordered_offset < dip->logical_offset + dip->bytes) {
5847 ordered_bytes = dip->logical_offset + dip->bytes -
5852 bio->bi_private = dip->private;
5856 dio_end_io(bio, err);
5859 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5860 struct bio *bio, int mirror_num,
5861 unsigned long bio_flags, u64 offset)
5864 struct btrfs_root *root = BTRFS_I(inode)->root;
5865 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5870 static void btrfs_end_dio_bio(struct bio *bio, int err)
5872 struct btrfs_dio_private *dip = bio->bi_private;
5875 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5876 "sector %#Lx len %u err no %d\n",
5877 dip->inode->i_ino, bio->bi_rw,
5878 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5882 * before atomic variable goto zero, we must make sure
5883 * dip->errors is perceived to be set.
5885 smp_mb__before_atomic_dec();
5888 /* if there are more bios still pending for this dio, just exit */
5889 if (!atomic_dec_and_test(&dip->pending_bios))
5893 bio_io_error(dip->orig_bio);
5895 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5896 bio_endio(dip->orig_bio, 0);
5902 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5903 u64 first_sector, gfp_t gfp_flags)
5905 int nr_vecs = bio_get_nr_vecs(bdev);
5906 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5909 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5910 int rw, u64 file_offset, int skip_sum,
5913 int write = rw & REQ_WRITE;
5914 struct btrfs_root *root = BTRFS_I(inode)->root;
5918 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5922 if (write && !skip_sum) {
5923 ret = btrfs_wq_submit_bio(root->fs_info,
5924 inode, rw, bio, 0, 0,
5926 __btrfs_submit_bio_start_direct_io,
5927 __btrfs_submit_bio_done);
5929 } else if (!skip_sum)
5930 btrfs_lookup_bio_sums_dio(root, inode, bio,
5931 file_offset, csums);
5933 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5939 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5942 struct inode *inode = dip->inode;
5943 struct btrfs_root *root = BTRFS_I(inode)->root;
5944 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5946 struct bio *orig_bio = dip->orig_bio;
5947 struct bio_vec *bvec = orig_bio->bi_io_vec;
5948 u64 start_sector = orig_bio->bi_sector;
5949 u64 file_offset = dip->logical_offset;
5953 u32 *csums = dip->csums;
5956 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5959 bio->bi_private = dip;
5960 bio->bi_end_io = btrfs_end_dio_bio;
5961 atomic_inc(&dip->pending_bios);
5963 map_length = orig_bio->bi_size;
5964 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5965 &map_length, NULL, 0);
5971 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5972 if (unlikely(map_length < submit_len + bvec->bv_len ||
5973 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5974 bvec->bv_offset) < bvec->bv_len)) {
5976 * inc the count before we submit the bio so
5977 * we know the end IO handler won't happen before
5978 * we inc the count. Otherwise, the dip might get freed
5979 * before we're done setting it up
5981 atomic_inc(&dip->pending_bios);
5982 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5983 file_offset, skip_sum,
5987 atomic_dec(&dip->pending_bios);
5992 csums = csums + nr_pages;
5993 start_sector += submit_len >> 9;
5994 file_offset += submit_len;
5999 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6000 start_sector, GFP_NOFS);
6003 bio->bi_private = dip;
6004 bio->bi_end_io = btrfs_end_dio_bio;
6006 map_length = orig_bio->bi_size;
6007 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6008 &map_length, NULL, 0);
6014 submit_len += bvec->bv_len;
6020 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6029 * before atomic variable goto zero, we must
6030 * make sure dip->errors is perceived to be set.
6032 smp_mb__before_atomic_dec();
6033 if (atomic_dec_and_test(&dip->pending_bios))
6034 bio_io_error(dip->orig_bio);
6036 /* bio_end_io() will handle error, so we needn't return it */
6040 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6043 struct btrfs_root *root = BTRFS_I(inode)->root;
6044 struct btrfs_dio_private *dip;
6045 struct bio_vec *bvec = bio->bi_io_vec;
6047 int write = rw & REQ_WRITE;
6050 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6052 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6060 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6068 dip->private = bio->bi_private;
6070 dip->logical_offset = file_offset;
6074 dip->bytes += bvec->bv_len;
6076 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6078 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6079 bio->bi_private = dip;
6081 dip->orig_bio = bio;
6082 atomic_set(&dip->pending_bios, 0);
6085 bio->bi_end_io = btrfs_endio_direct_write;
6087 bio->bi_end_io = btrfs_endio_direct_read;
6089 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6094 * If this is a write, we need to clean up the reserved space and kill
6095 * the ordered extent.
6098 struct btrfs_ordered_extent *ordered;
6099 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6100 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6101 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6102 btrfs_free_reserved_extent(root, ordered->start,
6104 btrfs_put_ordered_extent(ordered);
6105 btrfs_put_ordered_extent(ordered);
6107 bio_endio(bio, ret);
6110 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6111 const struct iovec *iov, loff_t offset,
6112 unsigned long nr_segs)
6117 unsigned blocksize_mask = root->sectorsize - 1;
6118 ssize_t retval = -EINVAL;
6119 loff_t end = offset;
6121 if (offset & blocksize_mask)
6124 /* Check the memory alignment. Blocks cannot straddle pages */
6125 for (seg = 0; seg < nr_segs; seg++) {
6126 addr = (unsigned long)iov[seg].iov_base;
6127 size = iov[seg].iov_len;
6129 if ((addr & blocksize_mask) || (size & blocksize_mask))
6136 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6137 const struct iovec *iov, loff_t offset,
6138 unsigned long nr_segs)
6140 struct file *file = iocb->ki_filp;
6141 struct inode *inode = file->f_mapping->host;
6142 struct btrfs_ordered_extent *ordered;
6143 struct extent_state *cached_state = NULL;
6144 u64 lockstart, lockend;
6146 int writing = rw & WRITE;
6148 size_t count = iov_length(iov, nr_segs);
6150 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6156 lockend = offset + count - 1;
6159 ret = btrfs_delalloc_reserve_space(inode, count);
6165 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6166 0, &cached_state, GFP_NOFS);
6168 * We're concerned with the entire range that we're going to be
6169 * doing DIO to, so we need to make sure theres no ordered
6170 * extents in this range.
6172 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6173 lockend - lockstart + 1);
6176 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6177 &cached_state, GFP_NOFS);
6178 btrfs_start_ordered_extent(inode, ordered, 1);
6179 btrfs_put_ordered_extent(ordered);
6184 * we don't use btrfs_set_extent_delalloc because we don't want
6185 * the dirty or uptodate bits
6188 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6189 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6190 EXTENT_DELALLOC, 0, NULL, &cached_state,
6193 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6194 lockend, EXTENT_LOCKED | write_bits,
6195 1, 0, &cached_state, GFP_NOFS);
6200 free_extent_state(cached_state);
6201 cached_state = NULL;
6203 ret = __blockdev_direct_IO(rw, iocb, inode,
6204 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6205 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6206 btrfs_submit_direct, 0);
6208 if (ret < 0 && ret != -EIOCBQUEUED) {
6209 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6210 offset + iov_length(iov, nr_segs) - 1,
6211 EXTENT_LOCKED | write_bits, 1, 0,
6212 &cached_state, GFP_NOFS);
6213 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6215 * We're falling back to buffered, unlock the section we didn't
6218 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6219 offset + iov_length(iov, nr_segs) - 1,
6220 EXTENT_LOCKED | write_bits, 1, 0,
6221 &cached_state, GFP_NOFS);
6224 free_extent_state(cached_state);
6228 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6229 __u64 start, __u64 len)
6231 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6234 int btrfs_readpage(struct file *file, struct page *page)
6236 struct extent_io_tree *tree;
6237 tree = &BTRFS_I(page->mapping->host)->io_tree;
6238 return extent_read_full_page(tree, page, btrfs_get_extent);
6241 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6243 struct extent_io_tree *tree;
6246 if (current->flags & PF_MEMALLOC) {
6247 redirty_page_for_writepage(wbc, page);
6251 tree = &BTRFS_I(page->mapping->host)->io_tree;
6252 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6255 int btrfs_writepages(struct address_space *mapping,
6256 struct writeback_control *wbc)
6258 struct extent_io_tree *tree;
6260 tree = &BTRFS_I(mapping->host)->io_tree;
6261 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6265 btrfs_readpages(struct file *file, struct address_space *mapping,
6266 struct list_head *pages, unsigned nr_pages)
6268 struct extent_io_tree *tree;
6269 tree = &BTRFS_I(mapping->host)->io_tree;
6270 return extent_readpages(tree, mapping, pages, nr_pages,
6273 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6275 struct extent_io_tree *tree;
6276 struct extent_map_tree *map;
6279 tree = &BTRFS_I(page->mapping->host)->io_tree;
6280 map = &BTRFS_I(page->mapping->host)->extent_tree;
6281 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6283 ClearPagePrivate(page);
6284 set_page_private(page, 0);
6285 page_cache_release(page);
6290 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6292 if (PageWriteback(page) || PageDirty(page))
6294 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6297 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6299 struct extent_io_tree *tree;
6300 struct btrfs_ordered_extent *ordered;
6301 struct extent_state *cached_state = NULL;
6302 u64 page_start = page_offset(page);
6303 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6307 * we have the page locked, so new writeback can't start,
6308 * and the dirty bit won't be cleared while we are here.
6310 * Wait for IO on this page so that we can safely clear
6311 * the PagePrivate2 bit and do ordered accounting
6313 wait_on_page_writeback(page);
6315 tree = &BTRFS_I(page->mapping->host)->io_tree;
6317 btrfs_releasepage(page, GFP_NOFS);
6320 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6322 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6326 * IO on this page will never be started, so we need
6327 * to account for any ordered extents now
6329 clear_extent_bit(tree, page_start, page_end,
6330 EXTENT_DIRTY | EXTENT_DELALLOC |
6331 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6332 &cached_state, GFP_NOFS);
6334 * whoever cleared the private bit is responsible
6335 * for the finish_ordered_io
6337 if (TestClearPagePrivate2(page)) {
6338 btrfs_finish_ordered_io(page->mapping->host,
6339 page_start, page_end);
6341 btrfs_put_ordered_extent(ordered);
6342 cached_state = NULL;
6343 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6346 clear_extent_bit(tree, page_start, page_end,
6347 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6348 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6349 __btrfs_releasepage(page, GFP_NOFS);
6351 ClearPageChecked(page);
6352 if (PagePrivate(page)) {
6353 ClearPagePrivate(page);
6354 set_page_private(page, 0);
6355 page_cache_release(page);
6360 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6361 * called from a page fault handler when a page is first dirtied. Hence we must
6362 * be careful to check for EOF conditions here. We set the page up correctly
6363 * for a written page which means we get ENOSPC checking when writing into
6364 * holes and correct delalloc and unwritten extent mapping on filesystems that
6365 * support these features.
6367 * We are not allowed to take the i_mutex here so we have to play games to
6368 * protect against truncate races as the page could now be beyond EOF. Because
6369 * vmtruncate() writes the inode size before removing pages, once we have the
6370 * page lock we can determine safely if the page is beyond EOF. If it is not
6371 * beyond EOF, then the page is guaranteed safe against truncation until we
6374 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6376 struct page *page = vmf->page;
6377 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6378 struct btrfs_root *root = BTRFS_I(inode)->root;
6379 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6380 struct btrfs_ordered_extent *ordered;
6381 struct extent_state *cached_state = NULL;
6383 unsigned long zero_start;
6389 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6393 else /* -ENOSPC, -EIO, etc */
6394 ret = VM_FAULT_SIGBUS;
6398 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6401 size = i_size_read(inode);
6402 page_start = page_offset(page);
6403 page_end = page_start + PAGE_CACHE_SIZE - 1;
6405 if ((page->mapping != inode->i_mapping) ||
6406 (page_start >= size)) {
6407 /* page got truncated out from underneath us */
6410 wait_on_page_writeback(page);
6412 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6414 set_page_extent_mapped(page);
6417 * we can't set the delalloc bits if there are pending ordered
6418 * extents. Drop our locks and wait for them to finish
6420 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6422 unlock_extent_cached(io_tree, page_start, page_end,
6423 &cached_state, GFP_NOFS);
6425 btrfs_start_ordered_extent(inode, ordered, 1);
6426 btrfs_put_ordered_extent(ordered);
6431 * XXX - page_mkwrite gets called every time the page is dirtied, even
6432 * if it was already dirty, so for space accounting reasons we need to
6433 * clear any delalloc bits for the range we are fixing to save. There
6434 * is probably a better way to do this, but for now keep consistent with
6435 * prepare_pages in the normal write path.
6437 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6438 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6439 0, 0, &cached_state, GFP_NOFS);
6441 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6444 unlock_extent_cached(io_tree, page_start, page_end,
6445 &cached_state, GFP_NOFS);
6446 ret = VM_FAULT_SIGBUS;
6451 /* page is wholly or partially inside EOF */
6452 if (page_start + PAGE_CACHE_SIZE > size)
6453 zero_start = size & ~PAGE_CACHE_MASK;
6455 zero_start = PAGE_CACHE_SIZE;
6457 if (zero_start != PAGE_CACHE_SIZE) {
6459 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6460 flush_dcache_page(page);
6463 ClearPageChecked(page);
6464 set_page_dirty(page);
6465 SetPageUptodate(page);
6467 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6468 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6470 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6474 return VM_FAULT_LOCKED;
6476 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6481 static void btrfs_truncate(struct inode *inode)
6483 struct btrfs_root *root = BTRFS_I(inode)->root;
6485 struct btrfs_trans_handle *trans;
6487 u64 mask = root->sectorsize - 1;
6489 if (!S_ISREG(inode->i_mode)) {
6494 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6498 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6499 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6501 trans = btrfs_start_transaction(root, 0);
6502 BUG_ON(IS_ERR(trans));
6503 btrfs_set_trans_block_group(trans, inode);
6504 trans->block_rsv = root->orphan_block_rsv;
6507 * setattr is responsible for setting the ordered_data_close flag,
6508 * but that is only tested during the last file release. That
6509 * could happen well after the next commit, leaving a great big
6510 * window where new writes may get lost if someone chooses to write
6511 * to this file after truncating to zero
6513 * The inode doesn't have any dirty data here, and so if we commit
6514 * this is a noop. If someone immediately starts writing to the inode
6515 * it is very likely we'll catch some of their writes in this
6516 * transaction, and the commit will find this file on the ordered
6517 * data list with good things to send down.
6519 * This is a best effort solution, there is still a window where
6520 * using truncate to replace the contents of the file will
6521 * end up with a zero length file after a crash.
6523 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6524 btrfs_add_ordered_operation(trans, root, inode);
6528 trans = btrfs_start_transaction(root, 0);
6529 BUG_ON(IS_ERR(trans));
6530 btrfs_set_trans_block_group(trans, inode);
6531 trans->block_rsv = root->orphan_block_rsv;
6534 ret = btrfs_block_rsv_check(trans, root,
6535 root->orphan_block_rsv, 0, 5);
6537 BUG_ON(ret != -EAGAIN);
6538 ret = btrfs_commit_transaction(trans, root);
6544 ret = btrfs_truncate_inode_items(trans, root, inode,
6546 BTRFS_EXTENT_DATA_KEY);
6550 ret = btrfs_update_inode(trans, root, inode);
6553 nr = trans->blocks_used;
6554 btrfs_end_transaction(trans, root);
6556 btrfs_btree_balance_dirty(root, nr);
6559 if (ret == 0 && inode->i_nlink > 0) {
6560 ret = btrfs_orphan_del(trans, inode);
6564 ret = btrfs_update_inode(trans, root, inode);
6567 nr = trans->blocks_used;
6568 ret = btrfs_end_transaction_throttle(trans, root);
6570 btrfs_btree_balance_dirty(root, nr);
6574 * create a new subvolume directory/inode (helper for the ioctl).
6576 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6577 struct btrfs_root *new_root,
6578 u64 new_dirid, u64 alloc_hint)
6580 struct inode *inode;
6584 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6585 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6587 return PTR_ERR(inode);
6588 inode->i_op = &btrfs_dir_inode_operations;
6589 inode->i_fop = &btrfs_dir_file_operations;
6592 btrfs_i_size_write(inode, 0);
6594 err = btrfs_update_inode(trans, new_root, inode);
6601 /* helper function for file defrag and space balancing. This
6602 * forces readahead on a given range of bytes in an inode
6604 unsigned long btrfs_force_ra(struct address_space *mapping,
6605 struct file_ra_state *ra, struct file *file,
6606 pgoff_t offset, pgoff_t last_index)
6608 pgoff_t req_size = last_index - offset + 1;
6610 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6611 return offset + req_size;
6614 struct inode *btrfs_alloc_inode(struct super_block *sb)
6616 struct btrfs_inode *ei;
6617 struct inode *inode;
6619 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6624 ei->space_info = NULL;
6628 ei->last_sub_trans = 0;
6629 ei->logged_trans = 0;
6630 ei->delalloc_bytes = 0;
6631 ei->reserved_bytes = 0;
6632 ei->disk_i_size = 0;
6634 ei->index_cnt = (u64)-1;
6635 ei->last_unlink_trans = 0;
6637 atomic_set(&ei->outstanding_extents, 0);
6638 atomic_set(&ei->reserved_extents, 0);
6640 ei->ordered_data_close = 0;
6641 ei->orphan_meta_reserved = 0;
6642 ei->dummy_inode = 0;
6643 ei->force_compress = BTRFS_COMPRESS_NONE;
6645 inode = &ei->vfs_inode;
6646 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6647 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6648 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6649 mutex_init(&ei->log_mutex);
6650 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6651 INIT_LIST_HEAD(&ei->i_orphan);
6652 INIT_LIST_HEAD(&ei->delalloc_inodes);
6653 INIT_LIST_HEAD(&ei->ordered_operations);
6654 RB_CLEAR_NODE(&ei->rb_node);
6659 static void btrfs_i_callback(struct rcu_head *head)
6661 struct inode *inode = container_of(head, struct inode, i_rcu);
6662 INIT_LIST_HEAD(&inode->i_dentry);
6663 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6666 void btrfs_destroy_inode(struct inode *inode)
6668 struct btrfs_ordered_extent *ordered;
6669 struct btrfs_root *root = BTRFS_I(inode)->root;
6671 WARN_ON(!list_empty(&inode->i_dentry));
6672 WARN_ON(inode->i_data.nrpages);
6673 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6674 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6677 * This can happen where we create an inode, but somebody else also
6678 * created the same inode and we need to destroy the one we already
6685 * Make sure we're properly removed from the ordered operation
6689 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6690 spin_lock(&root->fs_info->ordered_extent_lock);
6691 list_del_init(&BTRFS_I(inode)->ordered_operations);
6692 spin_unlock(&root->fs_info->ordered_extent_lock);
6695 if (root == root->fs_info->tree_root) {
6696 struct btrfs_block_group_cache *block_group;
6698 block_group = btrfs_lookup_block_group(root->fs_info,
6699 BTRFS_I(inode)->block_group);
6700 if (block_group && block_group->inode == inode) {
6701 spin_lock(&block_group->lock);
6702 block_group->inode = NULL;
6703 spin_unlock(&block_group->lock);
6704 btrfs_put_block_group(block_group);
6705 } else if (block_group) {
6706 btrfs_put_block_group(block_group);
6710 spin_lock(&root->orphan_lock);
6711 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6712 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6714 list_del_init(&BTRFS_I(inode)->i_orphan);
6716 spin_unlock(&root->orphan_lock);
6719 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6723 printk(KERN_ERR "btrfs found ordered "
6724 "extent %llu %llu on inode cleanup\n",
6725 (unsigned long long)ordered->file_offset,
6726 (unsigned long long)ordered->len);
6727 btrfs_remove_ordered_extent(inode, ordered);
6728 btrfs_put_ordered_extent(ordered);
6729 btrfs_put_ordered_extent(ordered);
6732 inode_tree_del(inode);
6733 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6735 call_rcu(&inode->i_rcu, btrfs_i_callback);
6738 int btrfs_drop_inode(struct inode *inode)
6740 struct btrfs_root *root = BTRFS_I(inode)->root;
6742 if (btrfs_root_refs(&root->root_item) == 0 &&
6743 root != root->fs_info->tree_root)
6746 return generic_drop_inode(inode);
6749 static void init_once(void *foo)
6751 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6753 inode_init_once(&ei->vfs_inode);
6756 void btrfs_destroy_cachep(void)
6758 if (btrfs_inode_cachep)
6759 kmem_cache_destroy(btrfs_inode_cachep);
6760 if (btrfs_trans_handle_cachep)
6761 kmem_cache_destroy(btrfs_trans_handle_cachep);
6762 if (btrfs_transaction_cachep)
6763 kmem_cache_destroy(btrfs_transaction_cachep);
6764 if (btrfs_path_cachep)
6765 kmem_cache_destroy(btrfs_path_cachep);
6766 if (btrfs_free_space_cachep)
6767 kmem_cache_destroy(btrfs_free_space_cachep);
6770 int btrfs_init_cachep(void)
6772 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6773 sizeof(struct btrfs_inode), 0,
6774 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6775 if (!btrfs_inode_cachep)
6778 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6779 sizeof(struct btrfs_trans_handle), 0,
6780 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6781 if (!btrfs_trans_handle_cachep)
6784 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6785 sizeof(struct btrfs_transaction), 0,
6786 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6787 if (!btrfs_transaction_cachep)
6790 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6791 sizeof(struct btrfs_path), 0,
6792 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6793 if (!btrfs_path_cachep)
6796 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6797 sizeof(struct btrfs_free_space), 0,
6798 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6799 if (!btrfs_free_space_cachep)
6804 btrfs_destroy_cachep();
6808 static int btrfs_getattr(struct vfsmount *mnt,
6809 struct dentry *dentry, struct kstat *stat)
6811 struct inode *inode = dentry->d_inode;
6812 generic_fillattr(inode, stat);
6813 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6814 stat->blksize = PAGE_CACHE_SIZE;
6815 stat->blocks = (inode_get_bytes(inode) +
6816 BTRFS_I(inode)->delalloc_bytes) >> 9;
6820 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6821 struct inode *new_dir, struct dentry *new_dentry)
6823 struct btrfs_trans_handle *trans;
6824 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6825 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6826 struct inode *new_inode = new_dentry->d_inode;
6827 struct inode *old_inode = old_dentry->d_inode;
6828 struct timespec ctime = CURRENT_TIME;
6833 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6836 /* we only allow rename subvolume link between subvolumes */
6837 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6840 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6841 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6844 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6845 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6848 * we're using rename to replace one file with another.
6849 * and the replacement file is large. Start IO on it now so
6850 * we don't add too much work to the end of the transaction
6852 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6853 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6854 filemap_flush(old_inode->i_mapping);
6856 /* close the racy window with snapshot create/destroy ioctl */
6857 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6858 down_read(&root->fs_info->subvol_sem);
6860 * We want to reserve the absolute worst case amount of items. So if
6861 * both inodes are subvols and we need to unlink them then that would
6862 * require 4 item modifications, but if they are both normal inodes it
6863 * would require 5 item modifications, so we'll assume their normal
6864 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6865 * should cover the worst case number of items we'll modify.
6867 trans = btrfs_start_transaction(root, 20);
6869 return PTR_ERR(trans);
6871 btrfs_set_trans_block_group(trans, new_dir);
6874 btrfs_record_root_in_trans(trans, dest);
6876 ret = btrfs_set_inode_index(new_dir, &index);
6880 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6881 /* force full log commit if subvolume involved. */
6882 root->fs_info->last_trans_log_full_commit = trans->transid;
6884 ret = btrfs_insert_inode_ref(trans, dest,
6885 new_dentry->d_name.name,
6886 new_dentry->d_name.len,
6888 new_dir->i_ino, index);
6892 * this is an ugly little race, but the rename is required
6893 * to make sure that if we crash, the inode is either at the
6894 * old name or the new one. pinning the log transaction lets
6895 * us make sure we don't allow a log commit to come in after
6896 * we unlink the name but before we add the new name back in.
6898 btrfs_pin_log_trans(root);
6901 * make sure the inode gets flushed if it is replacing
6904 if (new_inode && new_inode->i_size &&
6905 old_inode && S_ISREG(old_inode->i_mode)) {
6906 btrfs_add_ordered_operation(trans, root, old_inode);
6909 old_dir->i_ctime = old_dir->i_mtime = ctime;
6910 new_dir->i_ctime = new_dir->i_mtime = ctime;
6911 old_inode->i_ctime = ctime;
6913 if (old_dentry->d_parent != new_dentry->d_parent)
6914 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6916 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6917 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6918 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6919 old_dentry->d_name.name,
6920 old_dentry->d_name.len);
6922 btrfs_inc_nlink(old_dentry->d_inode);
6923 ret = btrfs_unlink_inode(trans, root, old_dir,
6924 old_dentry->d_inode,
6925 old_dentry->d_name.name,
6926 old_dentry->d_name.len);
6931 new_inode->i_ctime = CURRENT_TIME;
6932 if (unlikely(new_inode->i_ino ==
6933 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6934 root_objectid = BTRFS_I(new_inode)->location.objectid;
6935 ret = btrfs_unlink_subvol(trans, dest, new_dir,
6937 new_dentry->d_name.name,
6938 new_dentry->d_name.len);
6939 BUG_ON(new_inode->i_nlink == 0);
6941 ret = btrfs_unlink_inode(trans, dest, new_dir,
6942 new_dentry->d_inode,
6943 new_dentry->d_name.name,
6944 new_dentry->d_name.len);
6947 if (new_inode->i_nlink == 0) {
6948 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6953 ret = btrfs_add_link(trans, new_dir, old_inode,
6954 new_dentry->d_name.name,
6955 new_dentry->d_name.len, 0, index);
6958 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6959 struct dentry *parent = dget_parent(new_dentry);
6960 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6962 btrfs_end_log_trans(root);
6965 btrfs_end_transaction_throttle(trans, root);
6967 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6968 up_read(&root->fs_info->subvol_sem);
6974 * some fairly slow code that needs optimization. This walks the list
6975 * of all the inodes with pending delalloc and forces them to disk.
6977 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6979 struct list_head *head = &root->fs_info->delalloc_inodes;
6980 struct btrfs_inode *binode;
6981 struct inode *inode;
6983 if (root->fs_info->sb->s_flags & MS_RDONLY)
6986 spin_lock(&root->fs_info->delalloc_lock);
6987 while (!list_empty(head)) {
6988 binode = list_entry(head->next, struct btrfs_inode,
6990 inode = igrab(&binode->vfs_inode);
6992 list_del_init(&binode->delalloc_inodes);
6993 spin_unlock(&root->fs_info->delalloc_lock);
6995 filemap_flush(inode->i_mapping);
6997 btrfs_add_delayed_iput(inode);
7002 spin_lock(&root->fs_info->delalloc_lock);
7004 spin_unlock(&root->fs_info->delalloc_lock);
7006 /* the filemap_flush will queue IO into the worker threads, but
7007 * we have to make sure the IO is actually started and that
7008 * ordered extents get created before we return
7010 atomic_inc(&root->fs_info->async_submit_draining);
7011 while (atomic_read(&root->fs_info->nr_async_submits) ||
7012 atomic_read(&root->fs_info->async_delalloc_pages)) {
7013 wait_event(root->fs_info->async_submit_wait,
7014 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7015 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7017 atomic_dec(&root->fs_info->async_submit_draining);
7021 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7024 struct btrfs_inode *binode;
7025 struct inode *inode = NULL;
7027 spin_lock(&root->fs_info->delalloc_lock);
7028 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7029 binode = list_entry(root->fs_info->delalloc_inodes.next,
7030 struct btrfs_inode, delalloc_inodes);
7031 inode = igrab(&binode->vfs_inode);
7033 list_move_tail(&binode->delalloc_inodes,
7034 &root->fs_info->delalloc_inodes);
7038 list_del_init(&binode->delalloc_inodes);
7039 cond_resched_lock(&root->fs_info->delalloc_lock);
7041 spin_unlock(&root->fs_info->delalloc_lock);
7045 filemap_write_and_wait(inode->i_mapping);
7047 * We have to do this because compression doesn't
7048 * actually set PG_writeback until it submits the pages
7049 * for IO, which happens in an async thread, so we could
7050 * race and not actually wait for any writeback pages
7051 * because they've not been submitted yet. Technically
7052 * this could still be the case for the ordered stuff
7053 * since the async thread may not have started to do its
7054 * work yet. If this becomes the case then we need to
7055 * figure out a way to make sure that in writepage we
7056 * wait for any async pages to be submitted before
7057 * returning so that fdatawait does what its supposed to
7060 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7062 filemap_flush(inode->i_mapping);
7065 btrfs_add_delayed_iput(inode);
7073 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7074 const char *symname)
7076 struct btrfs_trans_handle *trans;
7077 struct btrfs_root *root = BTRFS_I(dir)->root;
7078 struct btrfs_path *path;
7079 struct btrfs_key key;
7080 struct inode *inode = NULL;
7088 struct btrfs_file_extent_item *ei;
7089 struct extent_buffer *leaf;
7090 unsigned long nr = 0;
7092 name_len = strlen(symname) + 1;
7093 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7094 return -ENAMETOOLONG;
7096 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7100 * 2 items for inode item and ref
7101 * 2 items for dir items
7102 * 1 item for xattr if selinux is on
7104 trans = btrfs_start_transaction(root, 5);
7106 return PTR_ERR(trans);
7108 btrfs_set_trans_block_group(trans, dir);
7110 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7111 dentry->d_name.len, dir->i_ino, objectid,
7112 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7114 err = PTR_ERR(inode);
7118 err = btrfs_init_inode_security(trans, inode, dir);
7124 btrfs_set_trans_block_group(trans, inode);
7125 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7129 inode->i_mapping->a_ops = &btrfs_aops;
7130 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7131 inode->i_fop = &btrfs_file_operations;
7132 inode->i_op = &btrfs_file_inode_operations;
7133 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7135 btrfs_update_inode_block_group(trans, inode);
7136 btrfs_update_inode_block_group(trans, dir);
7140 path = btrfs_alloc_path();
7142 key.objectid = inode->i_ino;
7144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7145 datasize = btrfs_file_extent_calc_inline_size(name_len);
7146 err = btrfs_insert_empty_item(trans, root, path, &key,
7152 leaf = path->nodes[0];
7153 ei = btrfs_item_ptr(leaf, path->slots[0],
7154 struct btrfs_file_extent_item);
7155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7156 btrfs_set_file_extent_type(leaf, ei,
7157 BTRFS_FILE_EXTENT_INLINE);
7158 btrfs_set_file_extent_encryption(leaf, ei, 0);
7159 btrfs_set_file_extent_compression(leaf, ei, 0);
7160 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7161 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7163 ptr = btrfs_file_extent_inline_start(ei);
7164 write_extent_buffer(leaf, symname, ptr, name_len);
7165 btrfs_mark_buffer_dirty(leaf);
7166 btrfs_free_path(path);
7168 inode->i_op = &btrfs_symlink_inode_operations;
7169 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7170 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7171 inode_set_bytes(inode, name_len);
7172 btrfs_i_size_write(inode, name_len - 1);
7173 err = btrfs_update_inode(trans, root, inode);
7178 nr = trans->blocks_used;
7179 btrfs_end_transaction_throttle(trans, root);
7181 inode_dec_link_count(inode);
7184 btrfs_btree_balance_dirty(root, nr);
7188 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7189 u64 start, u64 num_bytes, u64 min_size,
7190 loff_t actual_len, u64 *alloc_hint,
7191 struct btrfs_trans_handle *trans)
7193 struct btrfs_root *root = BTRFS_I(inode)->root;
7194 struct btrfs_key ins;
7195 u64 cur_offset = start;
7198 bool own_trans = true;
7202 while (num_bytes > 0) {
7204 trans = btrfs_start_transaction(root, 3);
7205 if (IS_ERR(trans)) {
7206 ret = PTR_ERR(trans);
7211 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7212 0, *alloc_hint, (u64)-1, &ins, 1);
7215 btrfs_end_transaction(trans, root);
7219 ret = insert_reserved_file_extent(trans, inode,
7220 cur_offset, ins.objectid,
7221 ins.offset, ins.offset,
7222 ins.offset, 0, 0, 0,
7223 BTRFS_FILE_EXTENT_PREALLOC);
7225 btrfs_drop_extent_cache(inode, cur_offset,
7226 cur_offset + ins.offset -1, 0);
7228 num_bytes -= ins.offset;
7229 cur_offset += ins.offset;
7230 *alloc_hint = ins.objectid + ins.offset;
7232 inode->i_ctime = CURRENT_TIME;
7233 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7234 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7235 (actual_len > inode->i_size) &&
7236 (cur_offset > inode->i_size)) {
7237 if (cur_offset > actual_len)
7238 i_size = actual_len;
7240 i_size = cur_offset;
7241 i_size_write(inode, i_size);
7242 btrfs_ordered_update_i_size(inode, i_size, NULL);
7245 ret = btrfs_update_inode(trans, root, inode);
7249 btrfs_end_transaction(trans, root);
7254 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7255 u64 start, u64 num_bytes, u64 min_size,
7256 loff_t actual_len, u64 *alloc_hint)
7258 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7259 min_size, actual_len, alloc_hint,
7263 int btrfs_prealloc_file_range_trans(struct inode *inode,
7264 struct btrfs_trans_handle *trans, int mode,
7265 u64 start, u64 num_bytes, u64 min_size,
7266 loff_t actual_len, u64 *alloc_hint)
7268 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7269 min_size, actual_len, alloc_hint, trans);
7272 static int btrfs_set_page_dirty(struct page *page)
7274 return __set_page_dirty_nobuffers(page);
7277 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7279 struct btrfs_root *root = BTRFS_I(inode)->root;
7281 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7283 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7285 return generic_permission(inode, mask, flags, btrfs_check_acl);
7288 static const struct inode_operations btrfs_dir_inode_operations = {
7289 .getattr = btrfs_getattr,
7290 .lookup = btrfs_lookup,
7291 .create = btrfs_create,
7292 .unlink = btrfs_unlink,
7294 .mkdir = btrfs_mkdir,
7295 .rmdir = btrfs_rmdir,
7296 .rename = btrfs_rename,
7297 .symlink = btrfs_symlink,
7298 .setattr = btrfs_setattr,
7299 .mknod = btrfs_mknod,
7300 .setxattr = btrfs_setxattr,
7301 .getxattr = btrfs_getxattr,
7302 .listxattr = btrfs_listxattr,
7303 .removexattr = btrfs_removexattr,
7304 .permission = btrfs_permission,
7306 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7307 .lookup = btrfs_lookup,
7308 .permission = btrfs_permission,
7311 static const struct file_operations btrfs_dir_file_operations = {
7312 .llseek = generic_file_llseek,
7313 .read = generic_read_dir,
7314 .readdir = btrfs_real_readdir,
7315 .unlocked_ioctl = btrfs_ioctl,
7316 #ifdef CONFIG_COMPAT
7317 .compat_ioctl = btrfs_ioctl,
7319 .release = btrfs_release_file,
7320 .fsync = btrfs_sync_file,
7323 static struct extent_io_ops btrfs_extent_io_ops = {
7324 .fill_delalloc = run_delalloc_range,
7325 .submit_bio_hook = btrfs_submit_bio_hook,
7326 .merge_bio_hook = btrfs_merge_bio_hook,
7327 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7328 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7329 .writepage_start_hook = btrfs_writepage_start_hook,
7330 .readpage_io_failed_hook = btrfs_io_failed_hook,
7331 .set_bit_hook = btrfs_set_bit_hook,
7332 .clear_bit_hook = btrfs_clear_bit_hook,
7333 .merge_extent_hook = btrfs_merge_extent_hook,
7334 .split_extent_hook = btrfs_split_extent_hook,
7338 * btrfs doesn't support the bmap operation because swapfiles
7339 * use bmap to make a mapping of extents in the file. They assume
7340 * these extents won't change over the life of the file and they
7341 * use the bmap result to do IO directly to the drive.
7343 * the btrfs bmap call would return logical addresses that aren't
7344 * suitable for IO and they also will change frequently as COW
7345 * operations happen. So, swapfile + btrfs == corruption.
7347 * For now we're avoiding this by dropping bmap.
7349 static const struct address_space_operations btrfs_aops = {
7350 .readpage = btrfs_readpage,
7351 .writepage = btrfs_writepage,
7352 .writepages = btrfs_writepages,
7353 .readpages = btrfs_readpages,
7354 .sync_page = block_sync_page,
7355 .direct_IO = btrfs_direct_IO,
7356 .invalidatepage = btrfs_invalidatepage,
7357 .releasepage = btrfs_releasepage,
7358 .set_page_dirty = btrfs_set_page_dirty,
7359 .error_remove_page = generic_error_remove_page,
7362 static const struct address_space_operations btrfs_symlink_aops = {
7363 .readpage = btrfs_readpage,
7364 .writepage = btrfs_writepage,
7365 .invalidatepage = btrfs_invalidatepage,
7366 .releasepage = btrfs_releasepage,
7369 static const struct inode_operations btrfs_file_inode_operations = {
7370 .truncate = btrfs_truncate,
7371 .getattr = btrfs_getattr,
7372 .setattr = btrfs_setattr,
7373 .setxattr = btrfs_setxattr,
7374 .getxattr = btrfs_getxattr,
7375 .listxattr = btrfs_listxattr,
7376 .removexattr = btrfs_removexattr,
7377 .permission = btrfs_permission,
7378 .fiemap = btrfs_fiemap,
7380 static const struct inode_operations btrfs_special_inode_operations = {
7381 .getattr = btrfs_getattr,
7382 .setattr = btrfs_setattr,
7383 .permission = btrfs_permission,
7384 .setxattr = btrfs_setxattr,
7385 .getxattr = btrfs_getxattr,
7386 .listxattr = btrfs_listxattr,
7387 .removexattr = btrfs_removexattr,
7389 static const struct inode_operations btrfs_symlink_inode_operations = {
7390 .readlink = generic_readlink,
7391 .follow_link = page_follow_link_light,
7392 .put_link = page_put_link,
7393 .getattr = btrfs_getattr,
7394 .permission = btrfs_permission,
7395 .setxattr = btrfs_setxattr,
7396 .getxattr = btrfs_getxattr,
7397 .listxattr = btrfs_listxattr,
7398 .removexattr = btrfs_removexattr,
7401 const struct dentry_operations btrfs_dentry_operations = {
7402 .d_delete = btrfs_dentry_delete,