2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
53 #include "free-space-cache.h"
55 struct btrfs_iget_args {
57 struct btrfs_root *root;
60 static const struct inode_operations btrfs_dir_inode_operations;
61 static const struct inode_operations btrfs_symlink_inode_operations;
62 static const struct inode_operations btrfs_dir_ro_inode_operations;
63 static const struct inode_operations btrfs_special_inode_operations;
64 static const struct inode_operations btrfs_file_inode_operations;
65 static const struct address_space_operations btrfs_aops;
66 static const struct address_space_operations btrfs_symlink_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_transaction_cachep;
73 struct kmem_cache *btrfs_path_cachep;
74 struct kmem_cache *btrfs_free_space_cachep;
77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
78 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
79 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
80 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
81 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
82 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
83 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
84 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
88 static int btrfs_truncate(struct inode *inode);
89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
90 static noinline int cow_file_range(struct inode *inode,
91 struct page *locked_page,
92 u64 start, u64 end, int *page_started,
93 unsigned long *nr_written, int unlock);
95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
96 struct inode *inode, struct inode *dir)
100 err = btrfs_init_acl(trans, inode, dir);
102 err = btrfs_xattr_security_init(trans, inode, dir);
107 * this does all the hard work for inserting an inline extent into
108 * the btree. The caller should have done a btrfs_drop_extents so that
109 * no overlapping inline items exist in the btree
111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
112 struct btrfs_root *root, struct inode *inode,
113 u64 start, size_t size, size_t compressed_size,
114 struct page **compressed_pages)
116 struct btrfs_key key;
117 struct btrfs_path *path;
118 struct extent_buffer *leaf;
119 struct page *page = NULL;
122 struct btrfs_file_extent_item *ei;
125 size_t cur_size = size;
127 unsigned long offset;
128 int compress_type = BTRFS_COMPRESS_NONE;
130 if (compressed_size && compressed_pages) {
131 compress_type = root->fs_info->compress_type;
132 cur_size = compressed_size;
135 path = btrfs_alloc_path();
139 path->leave_spinning = 1;
140 btrfs_set_trans_block_group(trans, inode);
142 key.objectid = inode->i_ino;
144 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
145 datasize = btrfs_file_extent_calc_inline_size(cur_size);
147 inode_add_bytes(inode, size);
148 ret = btrfs_insert_empty_item(trans, root, path, &key,
155 leaf = path->nodes[0];
156 ei = btrfs_item_ptr(leaf, path->slots[0],
157 struct btrfs_file_extent_item);
158 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
159 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
160 btrfs_set_file_extent_encryption(leaf, ei, 0);
161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
162 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
163 ptr = btrfs_file_extent_inline_start(ei);
165 if (compress_type != BTRFS_COMPRESS_NONE) {
168 while (compressed_size > 0) {
169 cpage = compressed_pages[i];
170 cur_size = min_t(unsigned long, compressed_size,
173 kaddr = kmap_atomic(cpage, KM_USER0);
174 write_extent_buffer(leaf, kaddr, ptr, cur_size);
175 kunmap_atomic(kaddr, KM_USER0);
179 compressed_size -= cur_size;
181 btrfs_set_file_extent_compression(leaf, ei,
184 page = find_get_page(inode->i_mapping,
185 start >> PAGE_CACHE_SHIFT);
186 btrfs_set_file_extent_compression(leaf, ei, 0);
187 kaddr = kmap_atomic(page, KM_USER0);
188 offset = start & (PAGE_CACHE_SIZE - 1);
189 write_extent_buffer(leaf, kaddr + offset, ptr, size);
190 kunmap_atomic(kaddr, KM_USER0);
191 page_cache_release(page);
193 btrfs_mark_buffer_dirty(leaf);
194 btrfs_free_path(path);
197 * we're an inline extent, so nobody can
198 * extend the file past i_size without locking
199 * a page we already have locked.
201 * We must do any isize and inode updates
202 * before we unlock the pages. Otherwise we
203 * could end up racing with unlink.
205 BTRFS_I(inode)->disk_i_size = inode->i_size;
206 btrfs_update_inode(trans, root, inode);
210 btrfs_free_path(path);
216 * conditionally insert an inline extent into the file. This
217 * does the checks required to make sure the data is small enough
218 * to fit as an inline extent.
220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
221 struct btrfs_root *root,
222 struct inode *inode, u64 start, u64 end,
223 size_t compressed_size,
224 struct page **compressed_pages)
226 u64 isize = i_size_read(inode);
227 u64 actual_end = min(end + 1, isize);
228 u64 inline_len = actual_end - start;
229 u64 aligned_end = (end + root->sectorsize - 1) &
230 ~((u64)root->sectorsize - 1);
232 u64 data_len = inline_len;
236 data_len = compressed_size;
239 actual_end >= PAGE_CACHE_SIZE ||
240 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
242 (actual_end & (root->sectorsize - 1)) == 0) ||
244 data_len > root->fs_info->max_inline) {
248 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
252 if (isize > actual_end)
253 inline_len = min_t(u64, isize, actual_end);
254 ret = insert_inline_extent(trans, root, inode, start,
255 inline_len, compressed_size,
258 btrfs_delalloc_release_metadata(inode, end + 1 - start);
259 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
263 struct async_extent {
268 unsigned long nr_pages;
270 struct list_head list;
275 struct btrfs_root *root;
276 struct page *locked_page;
279 struct list_head extents;
280 struct btrfs_work work;
283 static noinline int add_async_extent(struct async_cow *cow,
284 u64 start, u64 ram_size,
287 unsigned long nr_pages,
290 struct async_extent *async_extent;
292 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
293 async_extent->start = start;
294 async_extent->ram_size = ram_size;
295 async_extent->compressed_size = compressed_size;
296 async_extent->pages = pages;
297 async_extent->nr_pages = nr_pages;
298 async_extent->compress_type = compress_type;
299 list_add_tail(&async_extent->list, &cow->extents);
304 * we create compressed extents in two phases. The first
305 * phase compresses a range of pages that have already been
306 * locked (both pages and state bits are locked).
308 * This is done inside an ordered work queue, and the compression
309 * is spread across many cpus. The actual IO submission is step
310 * two, and the ordered work queue takes care of making sure that
311 * happens in the same order things were put onto the queue by
312 * writepages and friends.
314 * If this code finds it can't get good compression, it puts an
315 * entry onto the work queue to write the uncompressed bytes. This
316 * makes sure that both compressed inodes and uncompressed inodes
317 * are written in the same order that pdflush sent them down.
319 static noinline int compress_file_range(struct inode *inode,
320 struct page *locked_page,
322 struct async_cow *async_cow,
325 struct btrfs_root *root = BTRFS_I(inode)->root;
326 struct btrfs_trans_handle *trans;
328 u64 blocksize = root->sectorsize;
330 u64 isize = i_size_read(inode);
332 struct page **pages = NULL;
333 unsigned long nr_pages;
334 unsigned long nr_pages_ret = 0;
335 unsigned long total_compressed = 0;
336 unsigned long total_in = 0;
337 unsigned long max_compressed = 128 * 1024;
338 unsigned long max_uncompressed = 128 * 1024;
341 int compress_type = root->fs_info->compress_type;
343 actual_end = min_t(u64, isize, end + 1);
346 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
347 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
350 * we don't want to send crud past the end of i_size through
351 * compression, that's just a waste of CPU time. So, if the
352 * end of the file is before the start of our current
353 * requested range of bytes, we bail out to the uncompressed
354 * cleanup code that can deal with all of this.
356 * It isn't really the fastest way to fix things, but this is a
357 * very uncommon corner.
359 if (actual_end <= start)
360 goto cleanup_and_bail_uncompressed;
362 total_compressed = actual_end - start;
364 /* we want to make sure that amount of ram required to uncompress
365 * an extent is reasonable, so we limit the total size in ram
366 * of a compressed extent to 128k. This is a crucial number
367 * because it also controls how easily we can spread reads across
368 * cpus for decompression.
370 * We also want to make sure the amount of IO required to do
371 * a random read is reasonably small, so we limit the size of
372 * a compressed extent to 128k.
374 total_compressed = min(total_compressed, max_uncompressed);
375 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
376 num_bytes = max(blocksize, num_bytes);
381 * we do compression for mount -o compress and when the
382 * inode has not been flagged as nocompress. This flag can
383 * change at any time if we discover bad compression ratios.
385 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
386 (btrfs_test_opt(root, COMPRESS) ||
387 (BTRFS_I(inode)->force_compress) ||
388 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
390 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
392 if (BTRFS_I(inode)->force_compress)
393 compress_type = BTRFS_I(inode)->force_compress;
395 ret = btrfs_compress_pages(compress_type,
396 inode->i_mapping, start,
397 total_compressed, pages,
398 nr_pages, &nr_pages_ret,
404 unsigned long offset = total_compressed &
405 (PAGE_CACHE_SIZE - 1);
406 struct page *page = pages[nr_pages_ret - 1];
409 /* zero the tail end of the last page, we might be
410 * sending it down to disk
413 kaddr = kmap_atomic(page, KM_USER0);
414 memset(kaddr + offset, 0,
415 PAGE_CACHE_SIZE - offset);
416 kunmap_atomic(kaddr, KM_USER0);
422 trans = btrfs_join_transaction(root, 1);
423 BUG_ON(IS_ERR(trans));
424 btrfs_set_trans_block_group(trans, inode);
425 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
427 /* lets try to make an inline extent */
428 if (ret || total_in < (actual_end - start)) {
429 /* we didn't compress the entire range, try
430 * to make an uncompressed inline extent.
432 ret = cow_file_range_inline(trans, root, inode,
433 start, end, 0, NULL);
435 /* try making a compressed inline extent */
436 ret = cow_file_range_inline(trans, root, inode,
438 total_compressed, pages);
442 * inline extent creation worked, we don't need
443 * to create any more async work items. Unlock
444 * and free up our temp pages.
446 extent_clear_unlock_delalloc(inode,
447 &BTRFS_I(inode)->io_tree,
449 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
450 EXTENT_CLEAR_DELALLOC |
451 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
453 btrfs_end_transaction(trans, root);
456 btrfs_end_transaction(trans, root);
461 * we aren't doing an inline extent round the compressed size
462 * up to a block size boundary so the allocator does sane
465 total_compressed = (total_compressed + blocksize - 1) &
469 * one last check to make sure the compression is really a
470 * win, compare the page count read with the blocks on disk
472 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
473 ~(PAGE_CACHE_SIZE - 1);
474 if (total_compressed >= total_in) {
477 num_bytes = total_in;
480 if (!will_compress && pages) {
482 * the compression code ran but failed to make things smaller,
483 * free any pages it allocated and our page pointer array
485 for (i = 0; i < nr_pages_ret; i++) {
486 WARN_ON(pages[i]->mapping);
487 page_cache_release(pages[i]);
491 total_compressed = 0;
494 /* flag the file so we don't compress in the future */
495 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
496 !(BTRFS_I(inode)->force_compress)) {
497 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
503 /* the async work queues will take care of doing actual
504 * allocation on disk for these compressed pages,
505 * and will submit them to the elevator.
507 add_async_extent(async_cow, start, num_bytes,
508 total_compressed, pages, nr_pages_ret,
511 if (start + num_bytes < end) {
518 cleanup_and_bail_uncompressed:
520 * No compression, but we still need to write the pages in
521 * the file we've been given so far. redirty the locked
522 * page if it corresponds to our extent and set things up
523 * for the async work queue to run cow_file_range to do
524 * the normal delalloc dance
526 if (page_offset(locked_page) >= start &&
527 page_offset(locked_page) <= end) {
528 __set_page_dirty_nobuffers(locked_page);
529 /* unlocked later on in the async handlers */
531 add_async_extent(async_cow, start, end - start + 1,
532 0, NULL, 0, BTRFS_COMPRESS_NONE);
540 for (i = 0; i < nr_pages_ret; i++) {
541 WARN_ON(pages[i]->mapping);
542 page_cache_release(pages[i]);
550 * phase two of compressed writeback. This is the ordered portion
551 * of the code, which only gets called in the order the work was
552 * queued. We walk all the async extents created by compress_file_range
553 * and send them down to the disk.
555 static noinline int submit_compressed_extents(struct inode *inode,
556 struct async_cow *async_cow)
558 struct async_extent *async_extent;
560 struct btrfs_trans_handle *trans;
561 struct btrfs_key ins;
562 struct extent_map *em;
563 struct btrfs_root *root = BTRFS_I(inode)->root;
564 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
565 struct extent_io_tree *io_tree;
568 if (list_empty(&async_cow->extents))
572 while (!list_empty(&async_cow->extents)) {
573 async_extent = list_entry(async_cow->extents.next,
574 struct async_extent, list);
575 list_del(&async_extent->list);
577 io_tree = &BTRFS_I(inode)->io_tree;
580 /* did the compression code fall back to uncompressed IO? */
581 if (!async_extent->pages) {
582 int page_started = 0;
583 unsigned long nr_written = 0;
585 lock_extent(io_tree, async_extent->start,
586 async_extent->start +
587 async_extent->ram_size - 1, GFP_NOFS);
589 /* allocate blocks */
590 ret = cow_file_range(inode, async_cow->locked_page,
592 async_extent->start +
593 async_extent->ram_size - 1,
594 &page_started, &nr_written, 0);
597 * if page_started, cow_file_range inserted an
598 * inline extent and took care of all the unlocking
599 * and IO for us. Otherwise, we need to submit
600 * all those pages down to the drive.
602 if (!page_started && !ret)
603 extent_write_locked_range(io_tree,
604 inode, async_extent->start,
605 async_extent->start +
606 async_extent->ram_size - 1,
614 lock_extent(io_tree, async_extent->start,
615 async_extent->start + async_extent->ram_size - 1,
618 trans = btrfs_join_transaction(root, 1);
619 BUG_ON(IS_ERR(trans));
620 ret = btrfs_reserve_extent(trans, root,
621 async_extent->compressed_size,
622 async_extent->compressed_size,
625 btrfs_end_transaction(trans, root);
629 for (i = 0; i < async_extent->nr_pages; i++) {
630 WARN_ON(async_extent->pages[i]->mapping);
631 page_cache_release(async_extent->pages[i]);
633 kfree(async_extent->pages);
634 async_extent->nr_pages = 0;
635 async_extent->pages = NULL;
636 unlock_extent(io_tree, async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1, GFP_NOFS);
643 * here we're doing allocation and writeback of the
646 btrfs_drop_extent_cache(inode, async_extent->start,
647 async_extent->start +
648 async_extent->ram_size - 1, 0);
650 em = alloc_extent_map(GFP_NOFS);
652 em->start = async_extent->start;
653 em->len = async_extent->ram_size;
654 em->orig_start = em->start;
656 em->block_start = ins.objectid;
657 em->block_len = ins.offset;
658 em->bdev = root->fs_info->fs_devices->latest_bdev;
659 em->compress_type = async_extent->compress_type;
660 set_bit(EXTENT_FLAG_PINNED, &em->flags);
661 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
664 write_lock(&em_tree->lock);
665 ret = add_extent_mapping(em_tree, em);
666 write_unlock(&em_tree->lock);
667 if (ret != -EEXIST) {
671 btrfs_drop_extent_cache(inode, async_extent->start,
672 async_extent->start +
673 async_extent->ram_size - 1, 0);
676 ret = btrfs_add_ordered_extent_compress(inode,
679 async_extent->ram_size,
681 BTRFS_ORDERED_COMPRESSED,
682 async_extent->compress_type);
686 * clear dirty, set writeback and unlock the pages.
688 extent_clear_unlock_delalloc(inode,
689 &BTRFS_I(inode)->io_tree,
691 async_extent->start +
692 async_extent->ram_size - 1,
693 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
694 EXTENT_CLEAR_UNLOCK |
695 EXTENT_CLEAR_DELALLOC |
696 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
698 ret = btrfs_submit_compressed_write(inode,
700 async_extent->ram_size,
702 ins.offset, async_extent->pages,
703 async_extent->nr_pages);
706 alloc_hint = ins.objectid + ins.offset;
714 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
717 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
718 struct extent_map *em;
721 read_lock(&em_tree->lock);
722 em = search_extent_mapping(em_tree, start, num_bytes);
725 * if block start isn't an actual block number then find the
726 * first block in this inode and use that as a hint. If that
727 * block is also bogus then just don't worry about it.
729 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
731 em = search_extent_mapping(em_tree, 0, 0);
732 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
733 alloc_hint = em->block_start;
737 alloc_hint = em->block_start;
741 read_unlock(&em_tree->lock);
747 * when extent_io.c finds a delayed allocation range in the file,
748 * the call backs end up in this code. The basic idea is to
749 * allocate extents on disk for the range, and create ordered data structs
750 * in ram to track those extents.
752 * locked_page is the page that writepage had locked already. We use
753 * it to make sure we don't do extra locks or unlocks.
755 * *page_started is set to one if we unlock locked_page and do everything
756 * required to start IO on it. It may be clean and already done with
759 static noinline int cow_file_range(struct inode *inode,
760 struct page *locked_page,
761 u64 start, u64 end, int *page_started,
762 unsigned long *nr_written,
765 struct btrfs_root *root = BTRFS_I(inode)->root;
766 struct btrfs_trans_handle *trans;
769 unsigned long ram_size;
772 u64 blocksize = root->sectorsize;
773 struct btrfs_key ins;
774 struct extent_map *em;
775 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
778 BUG_ON(root == root->fs_info->tree_root);
779 trans = btrfs_join_transaction(root, 1);
780 BUG_ON(IS_ERR(trans));
781 btrfs_set_trans_block_group(trans, inode);
782 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
784 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
785 num_bytes = max(blocksize, num_bytes);
786 disk_num_bytes = num_bytes;
790 /* lets try to make an inline extent */
791 ret = cow_file_range_inline(trans, root, inode,
792 start, end, 0, NULL);
794 extent_clear_unlock_delalloc(inode,
795 &BTRFS_I(inode)->io_tree,
797 EXTENT_CLEAR_UNLOCK_PAGE |
798 EXTENT_CLEAR_UNLOCK |
799 EXTENT_CLEAR_DELALLOC |
801 EXTENT_SET_WRITEBACK |
802 EXTENT_END_WRITEBACK);
804 *nr_written = *nr_written +
805 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
812 BUG_ON(disk_num_bytes >
813 btrfs_super_total_bytes(&root->fs_info->super_copy));
815 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
816 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
818 while (disk_num_bytes > 0) {
821 cur_alloc_size = disk_num_bytes;
822 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
823 root->sectorsize, 0, alloc_hint,
827 em = alloc_extent_map(GFP_NOFS);
830 em->orig_start = em->start;
831 ram_size = ins.offset;
832 em->len = ins.offset;
834 em->block_start = ins.objectid;
835 em->block_len = ins.offset;
836 em->bdev = root->fs_info->fs_devices->latest_bdev;
837 set_bit(EXTENT_FLAG_PINNED, &em->flags);
840 write_lock(&em_tree->lock);
841 ret = add_extent_mapping(em_tree, em);
842 write_unlock(&em_tree->lock);
843 if (ret != -EEXIST) {
847 btrfs_drop_extent_cache(inode, start,
848 start + ram_size - 1, 0);
851 cur_alloc_size = ins.offset;
852 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
853 ram_size, cur_alloc_size, 0);
856 if (root->root_key.objectid ==
857 BTRFS_DATA_RELOC_TREE_OBJECTID) {
858 ret = btrfs_reloc_clone_csums(inode, start,
863 if (disk_num_bytes < cur_alloc_size)
866 /* we're not doing compressed IO, don't unlock the first
867 * page (which the caller expects to stay locked), don't
868 * clear any dirty bits and don't set any writeback bits
870 * Do set the Private2 bit so we know this page was properly
871 * setup for writepage
873 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
874 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
877 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
878 start, start + ram_size - 1,
880 disk_num_bytes -= cur_alloc_size;
881 num_bytes -= cur_alloc_size;
882 alloc_hint = ins.objectid + ins.offset;
883 start += cur_alloc_size;
887 btrfs_end_transaction(trans, root);
893 * work queue call back to started compression on a file and pages
895 static noinline void async_cow_start(struct btrfs_work *work)
897 struct async_cow *async_cow;
899 async_cow = container_of(work, struct async_cow, work);
901 compress_file_range(async_cow->inode, async_cow->locked_page,
902 async_cow->start, async_cow->end, async_cow,
905 async_cow->inode = NULL;
909 * work queue call back to submit previously compressed pages
911 static noinline void async_cow_submit(struct btrfs_work *work)
913 struct async_cow *async_cow;
914 struct btrfs_root *root;
915 unsigned long nr_pages;
917 async_cow = container_of(work, struct async_cow, work);
919 root = async_cow->root;
920 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
923 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
925 if (atomic_read(&root->fs_info->async_delalloc_pages) <
927 waitqueue_active(&root->fs_info->async_submit_wait))
928 wake_up(&root->fs_info->async_submit_wait);
930 if (async_cow->inode)
931 submit_compressed_extents(async_cow->inode, async_cow);
934 static noinline void async_cow_free(struct btrfs_work *work)
936 struct async_cow *async_cow;
937 async_cow = container_of(work, struct async_cow, work);
941 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
942 u64 start, u64 end, int *page_started,
943 unsigned long *nr_written)
945 struct async_cow *async_cow;
946 struct btrfs_root *root = BTRFS_I(inode)->root;
947 unsigned long nr_pages;
949 int limit = 10 * 1024 * 1042;
951 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
952 1, 0, NULL, GFP_NOFS);
953 while (start < end) {
954 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
955 async_cow->inode = inode;
956 async_cow->root = root;
957 async_cow->locked_page = locked_page;
958 async_cow->start = start;
960 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
963 cur_end = min(end, start + 512 * 1024 - 1);
965 async_cow->end = cur_end;
966 INIT_LIST_HEAD(&async_cow->extents);
968 async_cow->work.func = async_cow_start;
969 async_cow->work.ordered_func = async_cow_submit;
970 async_cow->work.ordered_free = async_cow_free;
971 async_cow->work.flags = 0;
973 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
975 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
977 btrfs_queue_worker(&root->fs_info->delalloc_workers,
980 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
981 wait_event(root->fs_info->async_submit_wait,
982 (atomic_read(&root->fs_info->async_delalloc_pages) <
986 while (atomic_read(&root->fs_info->async_submit_draining) &&
987 atomic_read(&root->fs_info->async_delalloc_pages)) {
988 wait_event(root->fs_info->async_submit_wait,
989 (atomic_read(&root->fs_info->async_delalloc_pages) ==
993 *nr_written += nr_pages;
1000 static noinline int csum_exist_in_range(struct btrfs_root *root,
1001 u64 bytenr, u64 num_bytes)
1004 struct btrfs_ordered_sum *sums;
1007 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1008 bytenr + num_bytes - 1, &list);
1009 if (ret == 0 && list_empty(&list))
1012 while (!list_empty(&list)) {
1013 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1014 list_del(&sums->list);
1021 * when nowcow writeback call back. This checks for snapshots or COW copies
1022 * of the extents that exist in the file, and COWs the file as required.
1024 * If no cow copies or snapshots exist, we write directly to the existing
1027 static noinline int run_delalloc_nocow(struct inode *inode,
1028 struct page *locked_page,
1029 u64 start, u64 end, int *page_started, int force,
1030 unsigned long *nr_written)
1032 struct btrfs_root *root = BTRFS_I(inode)->root;
1033 struct btrfs_trans_handle *trans;
1034 struct extent_buffer *leaf;
1035 struct btrfs_path *path;
1036 struct btrfs_file_extent_item *fi;
1037 struct btrfs_key found_key;
1049 bool nolock = false;
1051 path = btrfs_alloc_path();
1053 if (root == root->fs_info->tree_root) {
1055 trans = btrfs_join_transaction_nolock(root, 1);
1057 trans = btrfs_join_transaction(root, 1);
1059 BUG_ON(IS_ERR(trans));
1061 cow_start = (u64)-1;
1064 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1067 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1068 leaf = path->nodes[0];
1069 btrfs_item_key_to_cpu(leaf, &found_key,
1070 path->slots[0] - 1);
1071 if (found_key.objectid == inode->i_ino &&
1072 found_key.type == BTRFS_EXTENT_DATA_KEY)
1077 leaf = path->nodes[0];
1078 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1079 ret = btrfs_next_leaf(root, path);
1084 leaf = path->nodes[0];
1090 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1092 if (found_key.objectid > inode->i_ino ||
1093 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1094 found_key.offset > end)
1097 if (found_key.offset > cur_offset) {
1098 extent_end = found_key.offset;
1103 fi = btrfs_item_ptr(leaf, path->slots[0],
1104 struct btrfs_file_extent_item);
1105 extent_type = btrfs_file_extent_type(leaf, fi);
1107 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1108 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1109 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1110 extent_offset = btrfs_file_extent_offset(leaf, fi);
1111 extent_end = found_key.offset +
1112 btrfs_file_extent_num_bytes(leaf, fi);
1113 if (extent_end <= start) {
1117 if (disk_bytenr == 0)
1119 if (btrfs_file_extent_compression(leaf, fi) ||
1120 btrfs_file_extent_encryption(leaf, fi) ||
1121 btrfs_file_extent_other_encoding(leaf, fi))
1123 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1125 if (btrfs_extent_readonly(root, disk_bytenr))
1127 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1129 extent_offset, disk_bytenr))
1131 disk_bytenr += extent_offset;
1132 disk_bytenr += cur_offset - found_key.offset;
1133 num_bytes = min(end + 1, extent_end) - cur_offset;
1135 * force cow if csum exists in the range.
1136 * this ensure that csum for a given extent are
1137 * either valid or do not exist.
1139 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1142 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1143 extent_end = found_key.offset +
1144 btrfs_file_extent_inline_len(leaf, fi);
1145 extent_end = ALIGN(extent_end, root->sectorsize);
1150 if (extent_end <= start) {
1155 if (cow_start == (u64)-1)
1156 cow_start = cur_offset;
1157 cur_offset = extent_end;
1158 if (cur_offset > end)
1164 btrfs_release_path(root, path);
1165 if (cow_start != (u64)-1) {
1166 ret = cow_file_range(inode, locked_page, cow_start,
1167 found_key.offset - 1, page_started,
1170 cow_start = (u64)-1;
1173 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1174 struct extent_map *em;
1175 struct extent_map_tree *em_tree;
1176 em_tree = &BTRFS_I(inode)->extent_tree;
1177 em = alloc_extent_map(GFP_NOFS);
1179 em->start = cur_offset;
1180 em->orig_start = em->start;
1181 em->len = num_bytes;
1182 em->block_len = num_bytes;
1183 em->block_start = disk_bytenr;
1184 em->bdev = root->fs_info->fs_devices->latest_bdev;
1185 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1187 write_lock(&em_tree->lock);
1188 ret = add_extent_mapping(em_tree, em);
1189 write_unlock(&em_tree->lock);
1190 if (ret != -EEXIST) {
1191 free_extent_map(em);
1194 btrfs_drop_extent_cache(inode, em->start,
1195 em->start + em->len - 1, 0);
1197 type = BTRFS_ORDERED_PREALLOC;
1199 type = BTRFS_ORDERED_NOCOW;
1202 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1203 num_bytes, num_bytes, type);
1206 if (root->root_key.objectid ==
1207 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1208 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1213 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1214 cur_offset, cur_offset + num_bytes - 1,
1215 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1216 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1217 EXTENT_SET_PRIVATE2);
1218 cur_offset = extent_end;
1219 if (cur_offset > end)
1222 btrfs_release_path(root, path);
1224 if (cur_offset <= end && cow_start == (u64)-1)
1225 cow_start = cur_offset;
1226 if (cow_start != (u64)-1) {
1227 ret = cow_file_range(inode, locked_page, cow_start, end,
1228 page_started, nr_written, 1);
1233 ret = btrfs_end_transaction_nolock(trans, root);
1236 ret = btrfs_end_transaction(trans, root);
1239 btrfs_free_path(path);
1244 * extent_io.c call back to do delayed allocation processing
1246 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1247 u64 start, u64 end, int *page_started,
1248 unsigned long *nr_written)
1251 struct btrfs_root *root = BTRFS_I(inode)->root;
1253 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1254 ret = run_delalloc_nocow(inode, locked_page, start, end,
1255 page_started, 1, nr_written);
1256 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1257 ret = run_delalloc_nocow(inode, locked_page, start, end,
1258 page_started, 0, nr_written);
1259 else if (!btrfs_test_opt(root, COMPRESS) &&
1260 !(BTRFS_I(inode)->force_compress) &&
1261 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1262 ret = cow_file_range(inode, locked_page, start, end,
1263 page_started, nr_written, 1);
1265 ret = cow_file_range_async(inode, locked_page, start, end,
1266 page_started, nr_written);
1270 static int btrfs_split_extent_hook(struct inode *inode,
1271 struct extent_state *orig, u64 split)
1273 /* not delalloc, ignore it */
1274 if (!(orig->state & EXTENT_DELALLOC))
1277 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1282 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1283 * extents so we can keep track of new extents that are just merged onto old
1284 * extents, such as when we are doing sequential writes, so we can properly
1285 * account for the metadata space we'll need.
1287 static int btrfs_merge_extent_hook(struct inode *inode,
1288 struct extent_state *new,
1289 struct extent_state *other)
1291 /* not delalloc, ignore it */
1292 if (!(other->state & EXTENT_DELALLOC))
1295 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1300 * extent_io.c set_bit_hook, used to track delayed allocation
1301 * bytes in this file, and to maintain the list of inodes that
1302 * have pending delalloc work to be done.
1304 static int btrfs_set_bit_hook(struct inode *inode,
1305 struct extent_state *state, int *bits)
1309 * set_bit and clear bit hooks normally require _irqsave/restore
1310 * but in this case, we are only testeing for the DELALLOC
1311 * bit, which is only set or cleared with irqs on
1313 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1314 struct btrfs_root *root = BTRFS_I(inode)->root;
1315 u64 len = state->end + 1 - state->start;
1316 int do_list = (root->root_key.objectid !=
1317 BTRFS_ROOT_TREE_OBJECTID);
1319 if (*bits & EXTENT_FIRST_DELALLOC)
1320 *bits &= ~EXTENT_FIRST_DELALLOC;
1322 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1324 spin_lock(&root->fs_info->delalloc_lock);
1325 BTRFS_I(inode)->delalloc_bytes += len;
1326 root->fs_info->delalloc_bytes += len;
1327 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1328 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1329 &root->fs_info->delalloc_inodes);
1331 spin_unlock(&root->fs_info->delalloc_lock);
1337 * extent_io.c clear_bit_hook, see set_bit_hook for why
1339 static int btrfs_clear_bit_hook(struct inode *inode,
1340 struct extent_state *state, int *bits)
1343 * set_bit and clear bit hooks normally require _irqsave/restore
1344 * but in this case, we are only testeing for the DELALLOC
1345 * bit, which is only set or cleared with irqs on
1347 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1348 struct btrfs_root *root = BTRFS_I(inode)->root;
1349 u64 len = state->end + 1 - state->start;
1350 int do_list = (root->root_key.objectid !=
1351 BTRFS_ROOT_TREE_OBJECTID);
1353 if (*bits & EXTENT_FIRST_DELALLOC)
1354 *bits &= ~EXTENT_FIRST_DELALLOC;
1355 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1356 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1358 if (*bits & EXTENT_DO_ACCOUNTING)
1359 btrfs_delalloc_release_metadata(inode, len);
1361 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1363 btrfs_free_reserved_data_space(inode, len);
1365 spin_lock(&root->fs_info->delalloc_lock);
1366 root->fs_info->delalloc_bytes -= len;
1367 BTRFS_I(inode)->delalloc_bytes -= len;
1369 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1370 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1371 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1373 spin_unlock(&root->fs_info->delalloc_lock);
1379 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1380 * we don't create bios that span stripes or chunks
1382 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1383 size_t size, struct bio *bio,
1384 unsigned long bio_flags)
1386 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1387 struct btrfs_mapping_tree *map_tree;
1388 u64 logical = (u64)bio->bi_sector << 9;
1393 if (bio_flags & EXTENT_BIO_COMPRESSED)
1396 length = bio->bi_size;
1397 map_tree = &root->fs_info->mapping_tree;
1398 map_length = length;
1399 ret = btrfs_map_block(map_tree, READ, logical,
1400 &map_length, NULL, 0);
1402 if (map_length < length + size)
1408 * in order to insert checksums into the metadata in large chunks,
1409 * we wait until bio submission time. All the pages in the bio are
1410 * checksummed and sums are attached onto the ordered extent record.
1412 * At IO completion time the cums attached on the ordered extent record
1413 * are inserted into the btree
1415 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1416 struct bio *bio, int mirror_num,
1417 unsigned long bio_flags,
1420 struct btrfs_root *root = BTRFS_I(inode)->root;
1423 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1429 * in order to insert checksums into the metadata in large chunks,
1430 * we wait until bio submission time. All the pages in the bio are
1431 * checksummed and sums are attached onto the ordered extent record.
1433 * At IO completion time the cums attached on the ordered extent record
1434 * are inserted into the btree
1436 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1437 int mirror_num, unsigned long bio_flags,
1440 struct btrfs_root *root = BTRFS_I(inode)->root;
1441 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1445 * extent_io.c submission hook. This does the right thing for csum calculation
1446 * on write, or reading the csums from the tree before a read
1448 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1449 int mirror_num, unsigned long bio_flags,
1452 struct btrfs_root *root = BTRFS_I(inode)->root;
1456 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1458 if (root == root->fs_info->tree_root)
1459 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1461 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1464 if (!(rw & REQ_WRITE)) {
1465 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1466 return btrfs_submit_compressed_read(inode, bio,
1467 mirror_num, bio_flags);
1468 } else if (!skip_sum)
1469 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1471 } else if (!skip_sum) {
1472 /* csum items have already been cloned */
1473 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1475 /* we're doing a write, do the async checksumming */
1476 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1477 inode, rw, bio, mirror_num,
1478 bio_flags, bio_offset,
1479 __btrfs_submit_bio_start,
1480 __btrfs_submit_bio_done);
1484 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1488 * given a list of ordered sums record them in the inode. This happens
1489 * at IO completion time based on sums calculated at bio submission time.
1491 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1492 struct inode *inode, u64 file_offset,
1493 struct list_head *list)
1495 struct btrfs_ordered_sum *sum;
1497 btrfs_set_trans_block_group(trans, inode);
1499 list_for_each_entry(sum, list, list) {
1500 btrfs_csum_file_blocks(trans,
1501 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1506 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1507 struct extent_state **cached_state)
1509 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1511 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1512 cached_state, GFP_NOFS);
1515 /* see btrfs_writepage_start_hook for details on why this is required */
1516 struct btrfs_writepage_fixup {
1518 struct btrfs_work work;
1521 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1523 struct btrfs_writepage_fixup *fixup;
1524 struct btrfs_ordered_extent *ordered;
1525 struct extent_state *cached_state = NULL;
1527 struct inode *inode;
1531 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1535 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1536 ClearPageChecked(page);
1540 inode = page->mapping->host;
1541 page_start = page_offset(page);
1542 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1544 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1545 &cached_state, GFP_NOFS);
1547 /* already ordered? We're done */
1548 if (PagePrivate2(page))
1551 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1553 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1554 page_end, &cached_state, GFP_NOFS);
1556 btrfs_start_ordered_extent(inode, ordered, 1);
1561 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1562 ClearPageChecked(page);
1564 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1565 &cached_state, GFP_NOFS);
1568 page_cache_release(page);
1573 * There are a few paths in the higher layers of the kernel that directly
1574 * set the page dirty bit without asking the filesystem if it is a
1575 * good idea. This causes problems because we want to make sure COW
1576 * properly happens and the data=ordered rules are followed.
1578 * In our case any range that doesn't have the ORDERED bit set
1579 * hasn't been properly setup for IO. We kick off an async process
1580 * to fix it up. The async helper will wait for ordered extents, set
1581 * the delalloc bit and make it safe to write the page.
1583 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1585 struct inode *inode = page->mapping->host;
1586 struct btrfs_writepage_fixup *fixup;
1587 struct btrfs_root *root = BTRFS_I(inode)->root;
1589 /* this page is properly in the ordered list */
1590 if (TestClearPagePrivate2(page))
1593 if (PageChecked(page))
1596 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1600 SetPageChecked(page);
1601 page_cache_get(page);
1602 fixup->work.func = btrfs_writepage_fixup_worker;
1604 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1608 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1609 struct inode *inode, u64 file_pos,
1610 u64 disk_bytenr, u64 disk_num_bytes,
1611 u64 num_bytes, u64 ram_bytes,
1612 u8 compression, u8 encryption,
1613 u16 other_encoding, int extent_type)
1615 struct btrfs_root *root = BTRFS_I(inode)->root;
1616 struct btrfs_file_extent_item *fi;
1617 struct btrfs_path *path;
1618 struct extent_buffer *leaf;
1619 struct btrfs_key ins;
1623 path = btrfs_alloc_path();
1626 path->leave_spinning = 1;
1629 * we may be replacing one extent in the tree with another.
1630 * The new extent is pinned in the extent map, and we don't want
1631 * to drop it from the cache until it is completely in the btree.
1633 * So, tell btrfs_drop_extents to leave this extent in the cache.
1634 * the caller is expected to unpin it and allow it to be merged
1637 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1641 ins.objectid = inode->i_ino;
1642 ins.offset = file_pos;
1643 ins.type = BTRFS_EXTENT_DATA_KEY;
1644 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1646 leaf = path->nodes[0];
1647 fi = btrfs_item_ptr(leaf, path->slots[0],
1648 struct btrfs_file_extent_item);
1649 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1650 btrfs_set_file_extent_type(leaf, fi, extent_type);
1651 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1652 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1653 btrfs_set_file_extent_offset(leaf, fi, 0);
1654 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1655 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1656 btrfs_set_file_extent_compression(leaf, fi, compression);
1657 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1658 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1660 btrfs_unlock_up_safe(path, 1);
1661 btrfs_set_lock_blocking(leaf);
1663 btrfs_mark_buffer_dirty(leaf);
1665 inode_add_bytes(inode, num_bytes);
1667 ins.objectid = disk_bytenr;
1668 ins.offset = disk_num_bytes;
1669 ins.type = BTRFS_EXTENT_ITEM_KEY;
1670 ret = btrfs_alloc_reserved_file_extent(trans, root,
1671 root->root_key.objectid,
1672 inode->i_ino, file_pos, &ins);
1674 btrfs_free_path(path);
1680 * helper function for btrfs_finish_ordered_io, this
1681 * just reads in some of the csum leaves to prime them into ram
1682 * before we start the transaction. It limits the amount of btree
1683 * reads required while inside the transaction.
1685 /* as ordered data IO finishes, this gets called so we can finish
1686 * an ordered extent if the range of bytes in the file it covers are
1689 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1691 struct btrfs_root *root = BTRFS_I(inode)->root;
1692 struct btrfs_trans_handle *trans = NULL;
1693 struct btrfs_ordered_extent *ordered_extent = NULL;
1694 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1695 struct extent_state *cached_state = NULL;
1696 int compress_type = 0;
1698 bool nolock = false;
1700 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1704 BUG_ON(!ordered_extent);
1706 nolock = (root == root->fs_info->tree_root);
1708 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1709 BUG_ON(!list_empty(&ordered_extent->list));
1710 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1713 trans = btrfs_join_transaction_nolock(root, 1);
1715 trans = btrfs_join_transaction(root, 1);
1716 BUG_ON(IS_ERR(trans));
1717 btrfs_set_trans_block_group(trans, inode);
1718 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1719 ret = btrfs_update_inode(trans, root, inode);
1725 lock_extent_bits(io_tree, ordered_extent->file_offset,
1726 ordered_extent->file_offset + ordered_extent->len - 1,
1727 0, &cached_state, GFP_NOFS);
1730 trans = btrfs_join_transaction_nolock(root, 1);
1732 trans = btrfs_join_transaction(root, 1);
1733 BUG_ON(IS_ERR(trans));
1734 btrfs_set_trans_block_group(trans, inode);
1735 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1737 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1738 compress_type = ordered_extent->compress_type;
1739 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1740 BUG_ON(compress_type);
1741 ret = btrfs_mark_extent_written(trans, inode,
1742 ordered_extent->file_offset,
1743 ordered_extent->file_offset +
1744 ordered_extent->len);
1747 BUG_ON(root == root->fs_info->tree_root);
1748 ret = insert_reserved_file_extent(trans, inode,
1749 ordered_extent->file_offset,
1750 ordered_extent->start,
1751 ordered_extent->disk_len,
1752 ordered_extent->len,
1753 ordered_extent->len,
1754 compress_type, 0, 0,
1755 BTRFS_FILE_EXTENT_REG);
1756 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1757 ordered_extent->file_offset,
1758 ordered_extent->len);
1761 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1762 ordered_extent->file_offset +
1763 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1765 add_pending_csums(trans, inode, ordered_extent->file_offset,
1766 &ordered_extent->list);
1768 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1769 ret = btrfs_update_inode(trans, root, inode);
1774 btrfs_end_transaction_nolock(trans, root);
1776 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1778 btrfs_end_transaction(trans, root);
1782 btrfs_put_ordered_extent(ordered_extent);
1783 /* once for the tree */
1784 btrfs_put_ordered_extent(ordered_extent);
1789 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1790 struct extent_state *state, int uptodate)
1792 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1794 ClearPagePrivate2(page);
1795 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1799 * When IO fails, either with EIO or csum verification fails, we
1800 * try other mirrors that might have a good copy of the data. This
1801 * io_failure_record is used to record state as we go through all the
1802 * mirrors. If another mirror has good data, the page is set up to date
1803 * and things continue. If a good mirror can't be found, the original
1804 * bio end_io callback is called to indicate things have failed.
1806 struct io_failure_record {
1811 unsigned long bio_flags;
1815 static int btrfs_io_failed_hook(struct bio *failed_bio,
1816 struct page *page, u64 start, u64 end,
1817 struct extent_state *state)
1819 struct io_failure_record *failrec = NULL;
1821 struct extent_map *em;
1822 struct inode *inode = page->mapping->host;
1823 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1824 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1831 ret = get_state_private(failure_tree, start, &private);
1833 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1836 failrec->start = start;
1837 failrec->len = end - start + 1;
1838 failrec->last_mirror = 0;
1839 failrec->bio_flags = 0;
1841 read_lock(&em_tree->lock);
1842 em = lookup_extent_mapping(em_tree, start, failrec->len);
1843 if (em->start > start || em->start + em->len < start) {
1844 free_extent_map(em);
1847 read_unlock(&em_tree->lock);
1849 if (!em || IS_ERR(em)) {
1853 logical = start - em->start;
1854 logical = em->block_start + logical;
1855 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1856 logical = em->block_start;
1857 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1858 extent_set_compress_type(&failrec->bio_flags,
1861 failrec->logical = logical;
1862 free_extent_map(em);
1863 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1864 EXTENT_DIRTY, GFP_NOFS);
1865 set_state_private(failure_tree, start,
1866 (u64)(unsigned long)failrec);
1868 failrec = (struct io_failure_record *)(unsigned long)private;
1870 num_copies = btrfs_num_copies(
1871 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1872 failrec->logical, failrec->len);
1873 failrec->last_mirror++;
1875 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1876 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1879 if (state && state->start != failrec->start)
1881 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1883 if (!state || failrec->last_mirror > num_copies) {
1884 set_state_private(failure_tree, failrec->start, 0);
1885 clear_extent_bits(failure_tree, failrec->start,
1886 failrec->start + failrec->len - 1,
1887 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1891 bio = bio_alloc(GFP_NOFS, 1);
1892 bio->bi_private = state;
1893 bio->bi_end_io = failed_bio->bi_end_io;
1894 bio->bi_sector = failrec->logical >> 9;
1895 bio->bi_bdev = failed_bio->bi_bdev;
1898 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1899 if (failed_bio->bi_rw & REQ_WRITE)
1904 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1905 failrec->last_mirror,
1906 failrec->bio_flags, 0);
1911 * each time an IO finishes, we do a fast check in the IO failure tree
1912 * to see if we need to process or clean up an io_failure_record
1914 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1917 u64 private_failure;
1918 struct io_failure_record *failure;
1922 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1923 (u64)-1, 1, EXTENT_DIRTY, 0)) {
1924 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1925 start, &private_failure);
1927 failure = (struct io_failure_record *)(unsigned long)
1929 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1931 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1933 failure->start + failure->len - 1,
1934 EXTENT_DIRTY | EXTENT_LOCKED,
1943 * when reads are done, we need to check csums to verify the data is correct
1944 * if there's a match, we allow the bio to finish. If not, we go through
1945 * the io_failure_record routines to find good copies
1947 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1948 struct extent_state *state)
1950 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1951 struct inode *inode = page->mapping->host;
1952 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1954 u64 private = ~(u32)0;
1956 struct btrfs_root *root = BTRFS_I(inode)->root;
1959 if (PageChecked(page)) {
1960 ClearPageChecked(page);
1964 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1967 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1968 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1969 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1974 if (state && state->start == start) {
1975 private = state->private;
1978 ret = get_state_private(io_tree, start, &private);
1980 kaddr = kmap_atomic(page, KM_USER0);
1984 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1985 btrfs_csum_final(csum, (char *)&csum);
1986 if (csum != private)
1989 kunmap_atomic(kaddr, KM_USER0);
1991 /* if the io failure tree for this inode is non-empty,
1992 * check to see if we've recovered from a failed IO
1994 btrfs_clean_io_failures(inode, start);
1998 if (printk_ratelimit()) {
1999 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2000 "private %llu\n", page->mapping->host->i_ino,
2001 (unsigned long long)start, csum,
2002 (unsigned long long)private);
2004 memset(kaddr + offset, 1, end - start + 1);
2005 flush_dcache_page(page);
2006 kunmap_atomic(kaddr, KM_USER0);
2012 struct delayed_iput {
2013 struct list_head list;
2014 struct inode *inode;
2017 void btrfs_add_delayed_iput(struct inode *inode)
2019 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2020 struct delayed_iput *delayed;
2022 if (atomic_add_unless(&inode->i_count, -1, 1))
2025 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2026 delayed->inode = inode;
2028 spin_lock(&fs_info->delayed_iput_lock);
2029 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2030 spin_unlock(&fs_info->delayed_iput_lock);
2033 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2036 struct btrfs_fs_info *fs_info = root->fs_info;
2037 struct delayed_iput *delayed;
2040 spin_lock(&fs_info->delayed_iput_lock);
2041 empty = list_empty(&fs_info->delayed_iputs);
2042 spin_unlock(&fs_info->delayed_iput_lock);
2046 down_read(&root->fs_info->cleanup_work_sem);
2047 spin_lock(&fs_info->delayed_iput_lock);
2048 list_splice_init(&fs_info->delayed_iputs, &list);
2049 spin_unlock(&fs_info->delayed_iput_lock);
2051 while (!list_empty(&list)) {
2052 delayed = list_entry(list.next, struct delayed_iput, list);
2053 list_del(&delayed->list);
2054 iput(delayed->inode);
2057 up_read(&root->fs_info->cleanup_work_sem);
2061 * calculate extra metadata reservation when snapshotting a subvolume
2062 * contains orphan files.
2064 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2065 struct btrfs_pending_snapshot *pending,
2066 u64 *bytes_to_reserve)
2068 struct btrfs_root *root;
2069 struct btrfs_block_rsv *block_rsv;
2073 root = pending->root;
2074 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2077 block_rsv = root->orphan_block_rsv;
2079 /* orphan block reservation for the snapshot */
2080 num_bytes = block_rsv->size;
2083 * after the snapshot is created, COWing tree blocks may use more
2084 * space than it frees. So we should make sure there is enough
2087 index = trans->transid & 0x1;
2088 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2089 num_bytes += block_rsv->size -
2090 (block_rsv->reserved + block_rsv->freed[index]);
2093 *bytes_to_reserve += num_bytes;
2096 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2097 struct btrfs_pending_snapshot *pending)
2099 struct btrfs_root *root = pending->root;
2100 struct btrfs_root *snap = pending->snap;
2101 struct btrfs_block_rsv *block_rsv;
2106 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2109 /* refill source subvolume's orphan block reservation */
2110 block_rsv = root->orphan_block_rsv;
2111 index = trans->transid & 0x1;
2112 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2113 num_bytes = block_rsv->size -
2114 (block_rsv->reserved + block_rsv->freed[index]);
2115 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2116 root->orphan_block_rsv,
2121 /* setup orphan block reservation for the snapshot */
2122 block_rsv = btrfs_alloc_block_rsv(snap);
2125 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2126 snap->orphan_block_rsv = block_rsv;
2128 num_bytes = root->orphan_block_rsv->size;
2129 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2130 block_rsv, num_bytes);
2134 /* insert orphan item for the snapshot */
2135 WARN_ON(!root->orphan_item_inserted);
2136 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2137 snap->root_key.objectid);
2139 snap->orphan_item_inserted = 1;
2143 enum btrfs_orphan_cleanup_state {
2144 ORPHAN_CLEANUP_STARTED = 1,
2145 ORPHAN_CLEANUP_DONE = 2,
2149 * This is called in transaction commmit time. If there are no orphan
2150 * files in the subvolume, it removes orphan item and frees block_rsv
2153 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2154 struct btrfs_root *root)
2158 if (!list_empty(&root->orphan_list) ||
2159 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2162 if (root->orphan_item_inserted &&
2163 btrfs_root_refs(&root->root_item) > 0) {
2164 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2165 root->root_key.objectid);
2167 root->orphan_item_inserted = 0;
2170 if (root->orphan_block_rsv) {
2171 WARN_ON(root->orphan_block_rsv->size > 0);
2172 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2173 root->orphan_block_rsv = NULL;
2178 * This creates an orphan entry for the given inode in case something goes
2179 * wrong in the middle of an unlink/truncate.
2181 * NOTE: caller of this function should reserve 5 units of metadata for
2184 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2186 struct btrfs_root *root = BTRFS_I(inode)->root;
2187 struct btrfs_block_rsv *block_rsv = NULL;
2192 if (!root->orphan_block_rsv) {
2193 block_rsv = btrfs_alloc_block_rsv(root);
2197 spin_lock(&root->orphan_lock);
2198 if (!root->orphan_block_rsv) {
2199 root->orphan_block_rsv = block_rsv;
2200 } else if (block_rsv) {
2201 btrfs_free_block_rsv(root, block_rsv);
2205 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2206 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2209 * For proper ENOSPC handling, we should do orphan
2210 * cleanup when mounting. But this introduces backward
2211 * compatibility issue.
2213 if (!xchg(&root->orphan_item_inserted, 1))
2220 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2223 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2224 BTRFS_I(inode)->orphan_meta_reserved = 1;
2227 spin_unlock(&root->orphan_lock);
2230 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2232 /* grab metadata reservation from transaction handle */
2234 ret = btrfs_orphan_reserve_metadata(trans, inode);
2238 /* insert an orphan item to track this unlinked/truncated file */
2240 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2244 /* insert an orphan item to track subvolume contains orphan files */
2246 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2247 root->root_key.objectid);
2254 * We have done the truncate/delete so we can go ahead and remove the orphan
2255 * item for this particular inode.
2257 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2259 struct btrfs_root *root = BTRFS_I(inode)->root;
2260 int delete_item = 0;
2261 int release_rsv = 0;
2264 spin_lock(&root->orphan_lock);
2265 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2266 list_del_init(&BTRFS_I(inode)->i_orphan);
2270 if (BTRFS_I(inode)->orphan_meta_reserved) {
2271 BTRFS_I(inode)->orphan_meta_reserved = 0;
2274 spin_unlock(&root->orphan_lock);
2276 if (trans && delete_item) {
2277 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2282 btrfs_orphan_release_metadata(inode);
2288 * this cleans up any orphans that may be left on the list from the last use
2291 int btrfs_orphan_cleanup(struct btrfs_root *root)
2293 struct btrfs_path *path;
2294 struct extent_buffer *leaf;
2295 struct btrfs_key key, found_key;
2296 struct btrfs_trans_handle *trans;
2297 struct inode *inode;
2298 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2300 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2303 path = btrfs_alloc_path();
2310 key.objectid = BTRFS_ORPHAN_OBJECTID;
2311 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2312 key.offset = (u64)-1;
2315 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2320 * if ret == 0 means we found what we were searching for, which
2321 * is weird, but possible, so only screw with path if we didnt
2322 * find the key and see if we have stuff that matches
2326 if (path->slots[0] == 0)
2331 /* pull out the item */
2332 leaf = path->nodes[0];
2333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2335 /* make sure the item matches what we want */
2336 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2338 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2341 /* release the path since we're done with it */
2342 btrfs_release_path(root, path);
2345 * this is where we are basically btrfs_lookup, without the
2346 * crossing root thing. we store the inode number in the
2347 * offset of the orphan item.
2349 found_key.objectid = found_key.offset;
2350 found_key.type = BTRFS_INODE_ITEM_KEY;
2351 found_key.offset = 0;
2352 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2353 if (IS_ERR(inode)) {
2354 ret = PTR_ERR(inode);
2359 * add this inode to the orphan list so btrfs_orphan_del does
2360 * the proper thing when we hit it
2362 spin_lock(&root->orphan_lock);
2363 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2364 spin_unlock(&root->orphan_lock);
2367 * if this is a bad inode, means we actually succeeded in
2368 * removing the inode, but not the orphan record, which means
2369 * we need to manually delete the orphan since iput will just
2370 * do a destroy_inode
2372 if (is_bad_inode(inode)) {
2373 trans = btrfs_start_transaction(root, 0);
2374 if (IS_ERR(trans)) {
2375 ret = PTR_ERR(trans);
2378 btrfs_orphan_del(trans, inode);
2379 btrfs_end_transaction(trans, root);
2384 /* if we have links, this was a truncate, lets do that */
2385 if (inode->i_nlink) {
2386 if (!S_ISREG(inode->i_mode)) {
2392 ret = btrfs_truncate(inode);
2397 /* this will do delete_inode and everything for us */
2402 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2404 if (root->orphan_block_rsv)
2405 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2408 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2409 trans = btrfs_join_transaction(root, 1);
2411 btrfs_end_transaction(trans, root);
2415 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2417 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2421 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2422 btrfs_free_path(path);
2427 * very simple check to peek ahead in the leaf looking for xattrs. If we
2428 * don't find any xattrs, we know there can't be any acls.
2430 * slot is the slot the inode is in, objectid is the objectid of the inode
2432 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2433 int slot, u64 objectid)
2435 u32 nritems = btrfs_header_nritems(leaf);
2436 struct btrfs_key found_key;
2440 while (slot < nritems) {
2441 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2443 /* we found a different objectid, there must not be acls */
2444 if (found_key.objectid != objectid)
2447 /* we found an xattr, assume we've got an acl */
2448 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2452 * we found a key greater than an xattr key, there can't
2453 * be any acls later on
2455 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2462 * it goes inode, inode backrefs, xattrs, extents,
2463 * so if there are a ton of hard links to an inode there can
2464 * be a lot of backrefs. Don't waste time searching too hard,
2465 * this is just an optimization
2470 /* we hit the end of the leaf before we found an xattr or
2471 * something larger than an xattr. We have to assume the inode
2478 * read an inode from the btree into the in-memory inode
2480 static void btrfs_read_locked_inode(struct inode *inode)
2482 struct btrfs_path *path;
2483 struct extent_buffer *leaf;
2484 struct btrfs_inode_item *inode_item;
2485 struct btrfs_timespec *tspec;
2486 struct btrfs_root *root = BTRFS_I(inode)->root;
2487 struct btrfs_key location;
2489 u64 alloc_group_block;
2493 path = btrfs_alloc_path();
2495 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2497 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2501 leaf = path->nodes[0];
2502 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2503 struct btrfs_inode_item);
2505 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2506 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2507 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2508 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2509 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2511 tspec = btrfs_inode_atime(inode_item);
2512 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2513 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2515 tspec = btrfs_inode_mtime(inode_item);
2516 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2517 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2519 tspec = btrfs_inode_ctime(inode_item);
2520 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2521 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2524 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2525 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2526 inode->i_generation = BTRFS_I(inode)->generation;
2528 rdev = btrfs_inode_rdev(leaf, inode_item);
2530 BTRFS_I(inode)->index_cnt = (u64)-1;
2531 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2533 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2536 * try to precache a NULL acl entry for files that don't have
2537 * any xattrs or acls
2539 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2541 cache_no_acl(inode);
2543 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2544 alloc_group_block, 0);
2545 btrfs_free_path(path);
2548 switch (inode->i_mode & S_IFMT) {
2550 inode->i_mapping->a_ops = &btrfs_aops;
2551 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2552 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2553 inode->i_fop = &btrfs_file_operations;
2554 inode->i_op = &btrfs_file_inode_operations;
2557 inode->i_fop = &btrfs_dir_file_operations;
2558 if (root == root->fs_info->tree_root)
2559 inode->i_op = &btrfs_dir_ro_inode_operations;
2561 inode->i_op = &btrfs_dir_inode_operations;
2564 inode->i_op = &btrfs_symlink_inode_operations;
2565 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2566 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2569 inode->i_op = &btrfs_special_inode_operations;
2570 init_special_inode(inode, inode->i_mode, rdev);
2574 btrfs_update_iflags(inode);
2578 btrfs_free_path(path);
2579 make_bad_inode(inode);
2583 * given a leaf and an inode, copy the inode fields into the leaf
2585 static void fill_inode_item(struct btrfs_trans_handle *trans,
2586 struct extent_buffer *leaf,
2587 struct btrfs_inode_item *item,
2588 struct inode *inode)
2590 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2591 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2592 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2593 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2594 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2596 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2597 inode->i_atime.tv_sec);
2598 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2599 inode->i_atime.tv_nsec);
2601 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2602 inode->i_mtime.tv_sec);
2603 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2604 inode->i_mtime.tv_nsec);
2606 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2607 inode->i_ctime.tv_sec);
2608 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2609 inode->i_ctime.tv_nsec);
2611 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2612 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2613 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2614 btrfs_set_inode_transid(leaf, item, trans->transid);
2615 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2616 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2617 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2621 * copy everything in the in-memory inode into the btree.
2623 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2624 struct btrfs_root *root, struct inode *inode)
2626 struct btrfs_inode_item *inode_item;
2627 struct btrfs_path *path;
2628 struct extent_buffer *leaf;
2631 path = btrfs_alloc_path();
2633 path->leave_spinning = 1;
2634 ret = btrfs_lookup_inode(trans, root, path,
2635 &BTRFS_I(inode)->location, 1);
2642 btrfs_unlock_up_safe(path, 1);
2643 leaf = path->nodes[0];
2644 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2645 struct btrfs_inode_item);
2647 fill_inode_item(trans, leaf, inode_item, inode);
2648 btrfs_mark_buffer_dirty(leaf);
2649 btrfs_set_inode_last_trans(trans, inode);
2652 btrfs_free_path(path);
2658 * unlink helper that gets used here in inode.c and in the tree logging
2659 * recovery code. It remove a link in a directory with a given name, and
2660 * also drops the back refs in the inode to the directory
2662 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2663 struct btrfs_root *root,
2664 struct inode *dir, struct inode *inode,
2665 const char *name, int name_len)
2667 struct btrfs_path *path;
2669 struct extent_buffer *leaf;
2670 struct btrfs_dir_item *di;
2671 struct btrfs_key key;
2674 path = btrfs_alloc_path();
2680 path->leave_spinning = 1;
2681 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2682 name, name_len, -1);
2691 leaf = path->nodes[0];
2692 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2693 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2696 btrfs_release_path(root, path);
2698 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2700 dir->i_ino, &index);
2702 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2703 "inode %lu parent %lu\n", name_len, name,
2704 inode->i_ino, dir->i_ino);
2708 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2709 index, name, name_len, -1);
2718 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2719 btrfs_release_path(root, path);
2721 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2723 BUG_ON(ret != 0 && ret != -ENOENT);
2725 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2730 btrfs_free_path(path);
2734 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2735 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2736 btrfs_update_inode(trans, root, dir);
2737 btrfs_drop_nlink(inode);
2738 ret = btrfs_update_inode(trans, root, inode);
2743 /* helper to check if there is any shared block in the path */
2744 static int check_path_shared(struct btrfs_root *root,
2745 struct btrfs_path *path)
2747 struct extent_buffer *eb;
2751 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2754 if (!path->nodes[level])
2756 eb = path->nodes[level];
2757 if (!btrfs_block_can_be_shared(root, eb))
2759 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2768 * helper to start transaction for unlink and rmdir.
2770 * unlink and rmdir are special in btrfs, they do not always free space.
2771 * so in enospc case, we should make sure they will free space before
2772 * allowing them to use the global metadata reservation.
2774 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2775 struct dentry *dentry)
2777 struct btrfs_trans_handle *trans;
2778 struct btrfs_root *root = BTRFS_I(dir)->root;
2779 struct btrfs_path *path;
2780 struct btrfs_inode_ref *ref;
2781 struct btrfs_dir_item *di;
2782 struct inode *inode = dentry->d_inode;
2788 trans = btrfs_start_transaction(root, 10);
2789 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2792 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2793 return ERR_PTR(-ENOSPC);
2795 /* check if there is someone else holds reference */
2796 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2797 return ERR_PTR(-ENOSPC);
2799 if (atomic_read(&inode->i_count) > 2)
2800 return ERR_PTR(-ENOSPC);
2802 if (xchg(&root->fs_info->enospc_unlink, 1))
2803 return ERR_PTR(-ENOSPC);
2805 path = btrfs_alloc_path();
2807 root->fs_info->enospc_unlink = 0;
2808 return ERR_PTR(-ENOMEM);
2811 trans = btrfs_start_transaction(root, 0);
2812 if (IS_ERR(trans)) {
2813 btrfs_free_path(path);
2814 root->fs_info->enospc_unlink = 0;
2818 path->skip_locking = 1;
2819 path->search_commit_root = 1;
2821 ret = btrfs_lookup_inode(trans, root, path,
2822 &BTRFS_I(dir)->location, 0);
2828 if (check_path_shared(root, path))
2833 btrfs_release_path(root, path);
2835 ret = btrfs_lookup_inode(trans, root, path,
2836 &BTRFS_I(inode)->location, 0);
2842 if (check_path_shared(root, path))
2847 btrfs_release_path(root, path);
2849 if (ret == 0 && S_ISREG(inode->i_mode)) {
2850 ret = btrfs_lookup_file_extent(trans, root, path,
2851 inode->i_ino, (u64)-1, 0);
2857 if (check_path_shared(root, path))
2859 btrfs_release_path(root, path);
2867 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2868 dentry->d_name.name, dentry->d_name.len, 0);
2874 if (check_path_shared(root, path))
2880 btrfs_release_path(root, path);
2882 ref = btrfs_lookup_inode_ref(trans, root, path,
2883 dentry->d_name.name, dentry->d_name.len,
2884 inode->i_ino, dir->i_ino, 0);
2890 if (check_path_shared(root, path))
2892 index = btrfs_inode_ref_index(path->nodes[0], ref);
2893 btrfs_release_path(root, path);
2895 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2896 dentry->d_name.name, dentry->d_name.len, 0);
2901 BUG_ON(ret == -ENOENT);
2902 if (check_path_shared(root, path))
2907 btrfs_free_path(path);
2909 btrfs_end_transaction(trans, root);
2910 root->fs_info->enospc_unlink = 0;
2911 return ERR_PTR(err);
2914 trans->block_rsv = &root->fs_info->global_block_rsv;
2918 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2919 struct btrfs_root *root)
2921 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2922 BUG_ON(!root->fs_info->enospc_unlink);
2923 root->fs_info->enospc_unlink = 0;
2925 btrfs_end_transaction_throttle(trans, root);
2928 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2930 struct btrfs_root *root = BTRFS_I(dir)->root;
2931 struct btrfs_trans_handle *trans;
2932 struct inode *inode = dentry->d_inode;
2934 unsigned long nr = 0;
2936 trans = __unlink_start_trans(dir, dentry);
2938 return PTR_ERR(trans);
2940 btrfs_set_trans_block_group(trans, dir);
2942 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2944 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2945 dentry->d_name.name, dentry->d_name.len);
2948 if (inode->i_nlink == 0) {
2949 ret = btrfs_orphan_add(trans, inode);
2953 nr = trans->blocks_used;
2954 __unlink_end_trans(trans, root);
2955 btrfs_btree_balance_dirty(root, nr);
2959 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2960 struct btrfs_root *root,
2961 struct inode *dir, u64 objectid,
2962 const char *name, int name_len)
2964 struct btrfs_path *path;
2965 struct extent_buffer *leaf;
2966 struct btrfs_dir_item *di;
2967 struct btrfs_key key;
2971 path = btrfs_alloc_path();
2975 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2976 name, name_len, -1);
2977 BUG_ON(!di || IS_ERR(di));
2979 leaf = path->nodes[0];
2980 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2981 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2982 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2984 btrfs_release_path(root, path);
2986 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2987 objectid, root->root_key.objectid,
2988 dir->i_ino, &index, name, name_len);
2990 BUG_ON(ret != -ENOENT);
2991 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2993 BUG_ON(!di || IS_ERR(di));
2995 leaf = path->nodes[0];
2996 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2997 btrfs_release_path(root, path);
3001 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3002 index, name, name_len, -1);
3003 BUG_ON(!di || IS_ERR(di));
3005 leaf = path->nodes[0];
3006 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3007 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3008 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3010 btrfs_release_path(root, path);
3012 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3013 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3014 ret = btrfs_update_inode(trans, root, dir);
3017 btrfs_free_path(path);
3021 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3023 struct inode *inode = dentry->d_inode;
3025 struct btrfs_root *root = BTRFS_I(dir)->root;
3026 struct btrfs_trans_handle *trans;
3027 unsigned long nr = 0;
3029 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3030 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3033 trans = __unlink_start_trans(dir, dentry);
3035 return PTR_ERR(trans);
3037 btrfs_set_trans_block_group(trans, dir);
3039 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3040 err = btrfs_unlink_subvol(trans, root, dir,
3041 BTRFS_I(inode)->location.objectid,
3042 dentry->d_name.name,
3043 dentry->d_name.len);
3047 err = btrfs_orphan_add(trans, inode);
3051 /* now the directory is empty */
3052 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3053 dentry->d_name.name, dentry->d_name.len);
3055 btrfs_i_size_write(inode, 0);
3057 nr = trans->blocks_used;
3058 __unlink_end_trans(trans, root);
3059 btrfs_btree_balance_dirty(root, nr);
3066 * when truncating bytes in a file, it is possible to avoid reading
3067 * the leaves that contain only checksum items. This can be the
3068 * majority of the IO required to delete a large file, but it must
3069 * be done carefully.
3071 * The keys in the level just above the leaves are checked to make sure
3072 * the lowest key in a given leaf is a csum key, and starts at an offset
3073 * after the new size.
3075 * Then the key for the next leaf is checked to make sure it also has
3076 * a checksum item for the same file. If it does, we know our target leaf
3077 * contains only checksum items, and it can be safely freed without reading
3080 * This is just an optimization targeted at large files. It may do
3081 * nothing. It will return 0 unless things went badly.
3083 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3084 struct btrfs_root *root,
3085 struct btrfs_path *path,
3086 struct inode *inode, u64 new_size)
3088 struct btrfs_key key;
3091 struct btrfs_key found_key;
3092 struct btrfs_key other_key;
3093 struct btrfs_leaf_ref *ref;
3097 path->lowest_level = 1;
3098 key.objectid = inode->i_ino;
3099 key.type = BTRFS_CSUM_ITEM_KEY;
3100 key.offset = new_size;
3102 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3106 if (path->nodes[1] == NULL) {
3111 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3112 nritems = btrfs_header_nritems(path->nodes[1]);
3117 if (path->slots[1] >= nritems)
3120 /* did we find a key greater than anything we want to delete? */
3121 if (found_key.objectid > inode->i_ino ||
3122 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3125 /* we check the next key in the node to make sure the leave contains
3126 * only checksum items. This comparison doesn't work if our
3127 * leaf is the last one in the node
3129 if (path->slots[1] + 1 >= nritems) {
3131 /* search forward from the last key in the node, this
3132 * will bring us into the next node in the tree
3134 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3136 /* unlikely, but we inc below, so check to be safe */
3137 if (found_key.offset == (u64)-1)
3140 /* search_forward needs a path with locks held, do the
3141 * search again for the original key. It is possible
3142 * this will race with a balance and return a path that
3143 * we could modify, but this drop is just an optimization
3144 * and is allowed to miss some leaves.
3146 btrfs_release_path(root, path);
3149 /* setup a max key for search_forward */
3150 other_key.offset = (u64)-1;
3151 other_key.type = key.type;
3152 other_key.objectid = key.objectid;
3154 path->keep_locks = 1;
3155 ret = btrfs_search_forward(root, &found_key, &other_key,
3157 path->keep_locks = 0;
3158 if (ret || found_key.objectid != key.objectid ||
3159 found_key.type != key.type) {
3164 key.offset = found_key.offset;
3165 btrfs_release_path(root, path);
3170 /* we know there's one more slot after us in the tree,
3171 * read that key so we can verify it is also a checksum item
3173 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3175 if (found_key.objectid < inode->i_ino)
3178 if (found_key.type != key.type || found_key.offset < new_size)
3182 * if the key for the next leaf isn't a csum key from this objectid,
3183 * we can't be sure there aren't good items inside this leaf.
3186 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3189 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3190 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3192 * it is safe to delete this leaf, it contains only
3193 * csum items from this inode at an offset >= new_size
3195 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3198 if (root->ref_cows && leaf_gen < trans->transid) {
3199 ref = btrfs_alloc_leaf_ref(root, 0);
3201 ref->root_gen = root->root_key.offset;
3202 ref->bytenr = leaf_start;
3204 ref->generation = leaf_gen;
3207 btrfs_sort_leaf_ref(ref);
3209 ret = btrfs_add_leaf_ref(root, ref, 0);
3211 btrfs_free_leaf_ref(root, ref);
3217 btrfs_release_path(root, path);
3219 if (other_key.objectid == inode->i_ino &&
3220 other_key.type == key.type && other_key.offset > key.offset) {
3221 key.offset = other_key.offset;
3227 /* fixup any changes we've made to the path */
3228 path->lowest_level = 0;
3229 path->keep_locks = 0;
3230 btrfs_release_path(root, path);
3237 * this can truncate away extent items, csum items and directory items.
3238 * It starts at a high offset and removes keys until it can't find
3239 * any higher than new_size
3241 * csum items that cross the new i_size are truncated to the new size
3244 * min_type is the minimum key type to truncate down to. If set to 0, this
3245 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3247 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3248 struct btrfs_root *root,
3249 struct inode *inode,
3250 u64 new_size, u32 min_type)
3252 struct btrfs_path *path;
3253 struct extent_buffer *leaf;
3254 struct btrfs_file_extent_item *fi;
3255 struct btrfs_key key;
3256 struct btrfs_key found_key;
3257 u64 extent_start = 0;
3258 u64 extent_num_bytes = 0;
3259 u64 extent_offset = 0;
3261 u64 mask = root->sectorsize - 1;
3262 u32 found_type = (u8)-1;
3265 int pending_del_nr = 0;
3266 int pending_del_slot = 0;
3267 int extent_type = -1;
3272 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3274 if (root->ref_cows || root == root->fs_info->tree_root)
3275 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3277 path = btrfs_alloc_path();
3281 key.objectid = inode->i_ino;
3282 key.offset = (u64)-1;
3286 path->leave_spinning = 1;
3287 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3294 /* there are no items in the tree for us to truncate, we're
3297 if (path->slots[0] == 0)
3304 leaf = path->nodes[0];
3305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3306 found_type = btrfs_key_type(&found_key);
3309 if (found_key.objectid != inode->i_ino)
3312 if (found_type < min_type)
3315 item_end = found_key.offset;
3316 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3317 fi = btrfs_item_ptr(leaf, path->slots[0],
3318 struct btrfs_file_extent_item);
3319 extent_type = btrfs_file_extent_type(leaf, fi);
3320 encoding = btrfs_file_extent_compression(leaf, fi);
3321 encoding |= btrfs_file_extent_encryption(leaf, fi);
3322 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3324 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3326 btrfs_file_extent_num_bytes(leaf, fi);
3327 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3328 item_end += btrfs_file_extent_inline_len(leaf,
3333 if (found_type > min_type) {
3336 if (item_end < new_size)
3338 if (found_key.offset >= new_size)
3344 /* FIXME, shrink the extent if the ref count is only 1 */
3345 if (found_type != BTRFS_EXTENT_DATA_KEY)
3348 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3350 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3351 if (!del_item && !encoding) {
3352 u64 orig_num_bytes =
3353 btrfs_file_extent_num_bytes(leaf, fi);
3354 extent_num_bytes = new_size -
3355 found_key.offset + root->sectorsize - 1;
3356 extent_num_bytes = extent_num_bytes &
3357 ~((u64)root->sectorsize - 1);
3358 btrfs_set_file_extent_num_bytes(leaf, fi,
3360 num_dec = (orig_num_bytes -
3362 if (root->ref_cows && extent_start != 0)
3363 inode_sub_bytes(inode, num_dec);
3364 btrfs_mark_buffer_dirty(leaf);
3367 btrfs_file_extent_disk_num_bytes(leaf,
3369 extent_offset = found_key.offset -
3370 btrfs_file_extent_offset(leaf, fi);
3372 /* FIXME blocksize != 4096 */
3373 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3374 if (extent_start != 0) {
3377 inode_sub_bytes(inode, num_dec);
3380 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3382 * we can't truncate inline items that have had
3386 btrfs_file_extent_compression(leaf, fi) == 0 &&
3387 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3388 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3389 u32 size = new_size - found_key.offset;
3391 if (root->ref_cows) {
3392 inode_sub_bytes(inode, item_end + 1 -
3396 btrfs_file_extent_calc_inline_size(size);
3397 ret = btrfs_truncate_item(trans, root, path,
3400 } else if (root->ref_cows) {
3401 inode_sub_bytes(inode, item_end + 1 -
3407 if (!pending_del_nr) {
3408 /* no pending yet, add ourselves */
3409 pending_del_slot = path->slots[0];
3411 } else if (pending_del_nr &&
3412 path->slots[0] + 1 == pending_del_slot) {
3413 /* hop on the pending chunk */
3415 pending_del_slot = path->slots[0];
3422 if (found_extent && (root->ref_cows ||
3423 root == root->fs_info->tree_root)) {
3424 btrfs_set_path_blocking(path);
3425 ret = btrfs_free_extent(trans, root, extent_start,
3426 extent_num_bytes, 0,
3427 btrfs_header_owner(leaf),
3428 inode->i_ino, extent_offset);
3432 if (found_type == BTRFS_INODE_ITEM_KEY)
3435 if (path->slots[0] == 0 ||
3436 path->slots[0] != pending_del_slot) {
3437 if (root->ref_cows) {
3441 if (pending_del_nr) {
3442 ret = btrfs_del_items(trans, root, path,
3448 btrfs_release_path(root, path);
3455 if (pending_del_nr) {
3456 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3460 btrfs_free_path(path);
3465 * taken from block_truncate_page, but does cow as it zeros out
3466 * any bytes left in the last page in the file.
3468 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3470 struct inode *inode = mapping->host;
3471 struct btrfs_root *root = BTRFS_I(inode)->root;
3472 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3473 struct btrfs_ordered_extent *ordered;
3474 struct extent_state *cached_state = NULL;
3476 u32 blocksize = root->sectorsize;
3477 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3478 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3484 if ((offset & (blocksize - 1)) == 0)
3486 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3492 page = grab_cache_page(mapping, index);
3494 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3498 page_start = page_offset(page);
3499 page_end = page_start + PAGE_CACHE_SIZE - 1;
3501 if (!PageUptodate(page)) {
3502 ret = btrfs_readpage(NULL, page);
3504 if (page->mapping != mapping) {
3506 page_cache_release(page);
3509 if (!PageUptodate(page)) {
3514 wait_on_page_writeback(page);
3516 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3518 set_page_extent_mapped(page);
3520 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3522 unlock_extent_cached(io_tree, page_start, page_end,
3523 &cached_state, GFP_NOFS);
3525 page_cache_release(page);
3526 btrfs_start_ordered_extent(inode, ordered, 1);
3527 btrfs_put_ordered_extent(ordered);
3531 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3532 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3533 0, 0, &cached_state, GFP_NOFS);
3535 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3538 unlock_extent_cached(io_tree, page_start, page_end,
3539 &cached_state, GFP_NOFS);
3544 if (offset != PAGE_CACHE_SIZE) {
3546 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3547 flush_dcache_page(page);
3550 ClearPageChecked(page);
3551 set_page_dirty(page);
3552 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3557 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3559 page_cache_release(page);
3565 * This function puts in dummy file extents for the area we're creating a hole
3566 * for. So if we are truncating this file to a larger size we need to insert
3567 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3568 * the range between oldsize and size
3570 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3572 struct btrfs_trans_handle *trans;
3573 struct btrfs_root *root = BTRFS_I(inode)->root;
3574 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3575 struct extent_map *em = NULL;
3576 struct extent_state *cached_state = NULL;
3577 u64 mask = root->sectorsize - 1;
3578 u64 hole_start = (oldsize + mask) & ~mask;
3579 u64 block_end = (size + mask) & ~mask;
3585 if (size <= hole_start)
3589 struct btrfs_ordered_extent *ordered;
3590 btrfs_wait_ordered_range(inode, hole_start,
3591 block_end - hole_start);
3592 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3593 &cached_state, GFP_NOFS);
3594 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3597 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3598 &cached_state, GFP_NOFS);
3599 btrfs_put_ordered_extent(ordered);
3602 cur_offset = hole_start;
3604 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3605 block_end - cur_offset, 0);
3606 BUG_ON(IS_ERR(em) || !em);
3607 last_byte = min(extent_map_end(em), block_end);
3608 last_byte = (last_byte + mask) & ~mask;
3609 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3611 hole_size = last_byte - cur_offset;
3613 trans = btrfs_start_transaction(root, 2);
3614 if (IS_ERR(trans)) {
3615 err = PTR_ERR(trans);
3618 btrfs_set_trans_block_group(trans, inode);
3620 err = btrfs_drop_extents(trans, inode, cur_offset,
3621 cur_offset + hole_size,
3626 err = btrfs_insert_file_extent(trans, root,
3627 inode->i_ino, cur_offset, 0,
3628 0, hole_size, 0, hole_size,
3633 btrfs_drop_extent_cache(inode, hole_start,
3636 btrfs_end_transaction(trans, root);
3638 free_extent_map(em);
3640 cur_offset = last_byte;
3641 if (cur_offset >= block_end)
3645 free_extent_map(em);
3646 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3651 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3653 loff_t oldsize = i_size_read(inode);
3656 if (newsize == oldsize)
3659 if (newsize > oldsize) {
3660 i_size_write(inode, newsize);
3661 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3662 truncate_pagecache(inode, oldsize, newsize);
3663 ret = btrfs_cont_expand(inode, oldsize, newsize);
3665 btrfs_setsize(inode, oldsize);
3669 mark_inode_dirty(inode);
3673 * We're truncating a file that used to have good data down to
3674 * zero. Make sure it gets into the ordered flush list so that
3675 * any new writes get down to disk quickly.
3678 BTRFS_I(inode)->ordered_data_close = 1;
3680 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3681 truncate_setsize(inode, newsize);
3682 ret = btrfs_truncate(inode);
3688 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3690 struct inode *inode = dentry->d_inode;
3691 struct btrfs_root *root = BTRFS_I(inode)->root;
3694 if (btrfs_root_readonly(root))
3697 err = inode_change_ok(inode, attr);
3701 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3702 err = btrfs_setsize(inode, attr->ia_size);
3707 if (attr->ia_valid) {
3708 setattr_copy(inode, attr);
3709 mark_inode_dirty(inode);
3711 if (attr->ia_valid & ATTR_MODE)
3712 err = btrfs_acl_chmod(inode);
3718 void btrfs_evict_inode(struct inode *inode)
3720 struct btrfs_trans_handle *trans;
3721 struct btrfs_root *root = BTRFS_I(inode)->root;
3725 trace_btrfs_inode_evict(inode);
3727 truncate_inode_pages(&inode->i_data, 0);
3728 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3729 root == root->fs_info->tree_root))
3732 if (is_bad_inode(inode)) {
3733 btrfs_orphan_del(NULL, inode);
3736 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3737 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3739 if (root->fs_info->log_root_recovering) {
3740 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3744 if (inode->i_nlink > 0) {
3745 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3749 btrfs_i_size_write(inode, 0);
3752 trans = btrfs_start_transaction(root, 0);
3753 BUG_ON(IS_ERR(trans));
3754 btrfs_set_trans_block_group(trans, inode);
3755 trans->block_rsv = root->orphan_block_rsv;
3757 ret = btrfs_block_rsv_check(trans, root,
3758 root->orphan_block_rsv, 0, 5);
3760 BUG_ON(ret != -EAGAIN);
3761 ret = btrfs_commit_transaction(trans, root);
3766 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3770 nr = trans->blocks_used;
3771 btrfs_end_transaction(trans, root);
3773 btrfs_btree_balance_dirty(root, nr);
3778 ret = btrfs_orphan_del(trans, inode);
3782 nr = trans->blocks_used;
3783 btrfs_end_transaction(trans, root);
3784 btrfs_btree_balance_dirty(root, nr);
3786 end_writeback(inode);
3791 * this returns the key found in the dir entry in the location pointer.
3792 * If no dir entries were found, location->objectid is 0.
3794 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3795 struct btrfs_key *location)
3797 const char *name = dentry->d_name.name;
3798 int namelen = dentry->d_name.len;
3799 struct btrfs_dir_item *di;
3800 struct btrfs_path *path;
3801 struct btrfs_root *root = BTRFS_I(dir)->root;
3804 path = btrfs_alloc_path();
3807 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3812 if (!di || IS_ERR(di))
3815 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3817 btrfs_free_path(path);
3820 location->objectid = 0;
3825 * when we hit a tree root in a directory, the btrfs part of the inode
3826 * needs to be changed to reflect the root directory of the tree root. This
3827 * is kind of like crossing a mount point.
3829 static int fixup_tree_root_location(struct btrfs_root *root,
3831 struct dentry *dentry,
3832 struct btrfs_key *location,
3833 struct btrfs_root **sub_root)
3835 struct btrfs_path *path;
3836 struct btrfs_root *new_root;
3837 struct btrfs_root_ref *ref;
3838 struct extent_buffer *leaf;
3842 path = btrfs_alloc_path();
3849 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3850 BTRFS_I(dir)->root->root_key.objectid,
3851 location->objectid);
3858 leaf = path->nodes[0];
3859 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3860 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3861 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3864 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3865 (unsigned long)(ref + 1),
3866 dentry->d_name.len);
3870 btrfs_release_path(root->fs_info->tree_root, path);
3872 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3873 if (IS_ERR(new_root)) {
3874 err = PTR_ERR(new_root);
3878 if (btrfs_root_refs(&new_root->root_item) == 0) {
3883 *sub_root = new_root;
3884 location->objectid = btrfs_root_dirid(&new_root->root_item);
3885 location->type = BTRFS_INODE_ITEM_KEY;
3886 location->offset = 0;
3889 btrfs_free_path(path);
3893 static void inode_tree_add(struct inode *inode)
3895 struct btrfs_root *root = BTRFS_I(inode)->root;
3896 struct btrfs_inode *entry;
3898 struct rb_node *parent;
3900 p = &root->inode_tree.rb_node;
3903 if (inode_unhashed(inode))
3906 spin_lock(&root->inode_lock);
3909 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3911 if (inode->i_ino < entry->vfs_inode.i_ino)
3912 p = &parent->rb_left;
3913 else if (inode->i_ino > entry->vfs_inode.i_ino)
3914 p = &parent->rb_right;
3916 WARN_ON(!(entry->vfs_inode.i_state &
3917 (I_WILL_FREE | I_FREEING)));
3918 rb_erase(parent, &root->inode_tree);
3919 RB_CLEAR_NODE(parent);
3920 spin_unlock(&root->inode_lock);
3924 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3925 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3926 spin_unlock(&root->inode_lock);
3929 static void inode_tree_del(struct inode *inode)
3931 struct btrfs_root *root = BTRFS_I(inode)->root;
3934 spin_lock(&root->inode_lock);
3935 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3936 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3937 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3938 empty = RB_EMPTY_ROOT(&root->inode_tree);
3940 spin_unlock(&root->inode_lock);
3943 * Free space cache has inodes in the tree root, but the tree root has a
3944 * root_refs of 0, so this could end up dropping the tree root as a
3945 * snapshot, so we need the extra !root->fs_info->tree_root check to
3946 * make sure we don't drop it.
3948 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3949 root != root->fs_info->tree_root) {
3950 synchronize_srcu(&root->fs_info->subvol_srcu);
3951 spin_lock(&root->inode_lock);
3952 empty = RB_EMPTY_ROOT(&root->inode_tree);
3953 spin_unlock(&root->inode_lock);
3955 btrfs_add_dead_root(root);
3959 int btrfs_invalidate_inodes(struct btrfs_root *root)
3961 struct rb_node *node;
3962 struct rb_node *prev;
3963 struct btrfs_inode *entry;
3964 struct inode *inode;
3967 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3969 spin_lock(&root->inode_lock);
3971 node = root->inode_tree.rb_node;
3975 entry = rb_entry(node, struct btrfs_inode, rb_node);
3977 if (objectid < entry->vfs_inode.i_ino)
3978 node = node->rb_left;
3979 else if (objectid > entry->vfs_inode.i_ino)
3980 node = node->rb_right;
3986 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3987 if (objectid <= entry->vfs_inode.i_ino) {
3991 prev = rb_next(prev);
3995 entry = rb_entry(node, struct btrfs_inode, rb_node);
3996 objectid = entry->vfs_inode.i_ino + 1;
3997 inode = igrab(&entry->vfs_inode);
3999 spin_unlock(&root->inode_lock);
4000 if (atomic_read(&inode->i_count) > 1)
4001 d_prune_aliases(inode);
4003 * btrfs_drop_inode will have it removed from
4004 * the inode cache when its usage count
4009 spin_lock(&root->inode_lock);
4013 if (cond_resched_lock(&root->inode_lock))
4016 node = rb_next(node);
4018 spin_unlock(&root->inode_lock);
4022 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4024 struct btrfs_iget_args *args = p;
4025 inode->i_ino = args->ino;
4026 BTRFS_I(inode)->root = args->root;
4027 btrfs_set_inode_space_info(args->root, inode);
4031 static int btrfs_find_actor(struct inode *inode, void *opaque)
4033 struct btrfs_iget_args *args = opaque;
4034 return args->ino == inode->i_ino &&
4035 args->root == BTRFS_I(inode)->root;
4038 static struct inode *btrfs_iget_locked(struct super_block *s,
4040 struct btrfs_root *root)
4042 struct inode *inode;
4043 struct btrfs_iget_args args;
4044 args.ino = objectid;
4047 inode = iget5_locked(s, objectid, btrfs_find_actor,
4048 btrfs_init_locked_inode,
4053 /* Get an inode object given its location and corresponding root.
4054 * Returns in *is_new if the inode was read from disk
4056 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4057 struct btrfs_root *root, int *new)
4059 struct inode *inode;
4061 inode = btrfs_iget_locked(s, location->objectid, root);
4063 return ERR_PTR(-ENOMEM);
4065 if (inode->i_state & I_NEW) {
4066 BTRFS_I(inode)->root = root;
4067 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4068 btrfs_read_locked_inode(inode);
4070 inode_tree_add(inode);
4071 unlock_new_inode(inode);
4079 static struct inode *new_simple_dir(struct super_block *s,
4080 struct btrfs_key *key,
4081 struct btrfs_root *root)
4083 struct inode *inode = new_inode(s);
4086 return ERR_PTR(-ENOMEM);
4088 BTRFS_I(inode)->root = root;
4089 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4090 BTRFS_I(inode)->dummy_inode = 1;
4092 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4093 inode->i_op = &simple_dir_inode_operations;
4094 inode->i_fop = &simple_dir_operations;
4095 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4096 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4101 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4103 struct inode *inode;
4104 struct btrfs_root *root = BTRFS_I(dir)->root;
4105 struct btrfs_root *sub_root = root;
4106 struct btrfs_key location;
4110 if (dentry->d_name.len > BTRFS_NAME_LEN)
4111 return ERR_PTR(-ENAMETOOLONG);
4113 ret = btrfs_inode_by_name(dir, dentry, &location);
4116 return ERR_PTR(ret);
4118 if (location.objectid == 0)
4121 if (location.type == BTRFS_INODE_ITEM_KEY) {
4122 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4126 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4128 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4129 ret = fixup_tree_root_location(root, dir, dentry,
4130 &location, &sub_root);
4133 inode = ERR_PTR(ret);
4135 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4137 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4139 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4141 if (!IS_ERR(inode) && root != sub_root) {
4142 down_read(&root->fs_info->cleanup_work_sem);
4143 if (!(inode->i_sb->s_flags & MS_RDONLY))
4144 ret = btrfs_orphan_cleanup(sub_root);
4145 up_read(&root->fs_info->cleanup_work_sem);
4147 inode = ERR_PTR(ret);
4153 static int btrfs_dentry_delete(const struct dentry *dentry)
4155 struct btrfs_root *root;
4157 if (!dentry->d_inode && !IS_ROOT(dentry))
4158 dentry = dentry->d_parent;
4160 if (dentry->d_inode) {
4161 root = BTRFS_I(dentry->d_inode)->root;
4162 if (btrfs_root_refs(&root->root_item) == 0)
4168 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4169 struct nameidata *nd)
4171 struct inode *inode;
4173 inode = btrfs_lookup_dentry(dir, dentry);
4175 return ERR_CAST(inode);
4177 return d_splice_alias(inode, dentry);
4180 static unsigned char btrfs_filetype_table[] = {
4181 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4184 static int btrfs_real_readdir(struct file *filp, void *dirent,
4187 struct inode *inode = filp->f_dentry->d_inode;
4188 struct btrfs_root *root = BTRFS_I(inode)->root;
4189 struct btrfs_item *item;
4190 struct btrfs_dir_item *di;
4191 struct btrfs_key key;
4192 struct btrfs_key found_key;
4193 struct btrfs_path *path;
4196 struct extent_buffer *leaf;
4199 unsigned char d_type;
4204 int key_type = BTRFS_DIR_INDEX_KEY;
4209 /* FIXME, use a real flag for deciding about the key type */
4210 if (root->fs_info->tree_root == root)
4211 key_type = BTRFS_DIR_ITEM_KEY;
4213 /* special case for "." */
4214 if (filp->f_pos == 0) {
4215 over = filldir(dirent, ".", 1,
4222 /* special case for .., just use the back ref */
4223 if (filp->f_pos == 1) {
4224 u64 pino = parent_ino(filp->f_path.dentry);
4225 over = filldir(dirent, "..", 2,
4231 path = btrfs_alloc_path();
4234 btrfs_set_key_type(&key, key_type);
4235 key.offset = filp->f_pos;
4236 key.objectid = inode->i_ino;
4238 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4244 leaf = path->nodes[0];
4245 nritems = btrfs_header_nritems(leaf);
4246 slot = path->slots[0];
4247 if (advance || slot >= nritems) {
4248 if (slot >= nritems - 1) {
4249 ret = btrfs_next_leaf(root, path);
4252 leaf = path->nodes[0];
4253 nritems = btrfs_header_nritems(leaf);
4254 slot = path->slots[0];
4262 item = btrfs_item_nr(leaf, slot);
4263 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4265 if (found_key.objectid != key.objectid)
4267 if (btrfs_key_type(&found_key) != key_type)
4269 if (found_key.offset < filp->f_pos)
4272 filp->f_pos = found_key.offset;
4274 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4276 di_total = btrfs_item_size(leaf, item);
4278 while (di_cur < di_total) {
4279 struct btrfs_key location;
4281 if (verify_dir_item(root, leaf, di))
4284 name_len = btrfs_dir_name_len(leaf, di);
4285 if (name_len <= sizeof(tmp_name)) {
4286 name_ptr = tmp_name;
4288 name_ptr = kmalloc(name_len, GFP_NOFS);
4294 read_extent_buffer(leaf, name_ptr,
4295 (unsigned long)(di + 1), name_len);
4297 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4298 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4300 /* is this a reference to our own snapshot? If so
4303 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4304 location.objectid == root->root_key.objectid) {
4308 over = filldir(dirent, name_ptr, name_len,
4309 found_key.offset, location.objectid,
4313 if (name_ptr != tmp_name)
4318 di_len = btrfs_dir_name_len(leaf, di) +
4319 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4321 di = (struct btrfs_dir_item *)((char *)di + di_len);
4325 /* Reached end of directory/root. Bump pos past the last item. */
4326 if (key_type == BTRFS_DIR_INDEX_KEY)
4328 * 32-bit glibc will use getdents64, but then strtol -
4329 * so the last number we can serve is this.
4331 filp->f_pos = 0x7fffffff;
4337 btrfs_free_path(path);
4341 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4343 struct btrfs_root *root = BTRFS_I(inode)->root;
4344 struct btrfs_trans_handle *trans;
4346 bool nolock = false;
4348 if (BTRFS_I(inode)->dummy_inode)
4352 nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4354 if (wbc->sync_mode == WB_SYNC_ALL) {
4356 trans = btrfs_join_transaction_nolock(root, 1);
4358 trans = btrfs_join_transaction(root, 1);
4360 return PTR_ERR(trans);
4361 btrfs_set_trans_block_group(trans, inode);
4363 ret = btrfs_end_transaction_nolock(trans, root);
4365 ret = btrfs_commit_transaction(trans, root);
4371 * This is somewhat expensive, updating the tree every time the
4372 * inode changes. But, it is most likely to find the inode in cache.
4373 * FIXME, needs more benchmarking...there are no reasons other than performance
4374 * to keep or drop this code.
4376 void btrfs_dirty_inode(struct inode *inode)
4378 struct btrfs_root *root = BTRFS_I(inode)->root;
4379 struct btrfs_trans_handle *trans;
4382 if (BTRFS_I(inode)->dummy_inode)
4385 trans = btrfs_join_transaction(root, 1);
4386 BUG_ON(IS_ERR(trans));
4387 btrfs_set_trans_block_group(trans, inode);
4389 ret = btrfs_update_inode(trans, root, inode);
4390 if (ret && ret == -ENOSPC) {
4391 /* whoops, lets try again with the full transaction */
4392 btrfs_end_transaction(trans, root);
4393 trans = btrfs_start_transaction(root, 1);
4394 if (IS_ERR(trans)) {
4395 if (printk_ratelimit()) {
4396 printk(KERN_ERR "btrfs: fail to "
4397 "dirty inode %lu error %ld\n",
4398 inode->i_ino, PTR_ERR(trans));
4402 btrfs_set_trans_block_group(trans, inode);
4404 ret = btrfs_update_inode(trans, root, inode);
4406 if (printk_ratelimit()) {
4407 printk(KERN_ERR "btrfs: fail to "
4408 "dirty inode %lu error %d\n",
4413 btrfs_end_transaction(trans, root);
4417 * find the highest existing sequence number in a directory
4418 * and then set the in-memory index_cnt variable to reflect
4419 * free sequence numbers
4421 static int btrfs_set_inode_index_count(struct inode *inode)
4423 struct btrfs_root *root = BTRFS_I(inode)->root;
4424 struct btrfs_key key, found_key;
4425 struct btrfs_path *path;
4426 struct extent_buffer *leaf;
4429 key.objectid = inode->i_ino;
4430 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4431 key.offset = (u64)-1;
4433 path = btrfs_alloc_path();
4437 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4440 /* FIXME: we should be able to handle this */
4446 * MAGIC NUMBER EXPLANATION:
4447 * since we search a directory based on f_pos we have to start at 2
4448 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4449 * else has to start at 2
4451 if (path->slots[0] == 0) {
4452 BTRFS_I(inode)->index_cnt = 2;
4458 leaf = path->nodes[0];
4459 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4461 if (found_key.objectid != inode->i_ino ||
4462 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4463 BTRFS_I(inode)->index_cnt = 2;
4467 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4469 btrfs_free_path(path);
4474 * helper to find a free sequence number in a given directory. This current
4475 * code is very simple, later versions will do smarter things in the btree
4477 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4481 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4482 ret = btrfs_set_inode_index_count(dir);
4487 *index = BTRFS_I(dir)->index_cnt;
4488 BTRFS_I(dir)->index_cnt++;
4493 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4494 struct btrfs_root *root,
4496 const char *name, int name_len,
4497 u64 ref_objectid, u64 objectid,
4498 u64 alloc_hint, int mode, u64 *index)
4500 struct inode *inode;
4501 struct btrfs_inode_item *inode_item;
4502 struct btrfs_key *location;
4503 struct btrfs_path *path;
4504 struct btrfs_inode_ref *ref;
4505 struct btrfs_key key[2];
4511 path = btrfs_alloc_path();
4514 inode = new_inode(root->fs_info->sb);
4516 return ERR_PTR(-ENOMEM);
4519 trace_btrfs_inode_request(dir);
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(dir)->flags & BTRFS_INODE_NODATACOW))
4591 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4594 insert_inode_hash(inode);
4595 inode_tree_add(inode);
4597 trace_btrfs_inode_new(inode);
4602 BTRFS_I(dir)->index_cnt--;
4603 btrfs_free_path(path);
4605 return ERR_PTR(ret);
4608 static inline u8 btrfs_inode_type(struct inode *inode)
4610 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4614 * utility function to add 'inode' into 'parent_inode' with
4615 * a give name and a given sequence number.
4616 * if 'add_backref' is true, also insert a backref from the
4617 * inode to the parent directory.
4619 int btrfs_add_link(struct btrfs_trans_handle *trans,
4620 struct inode *parent_inode, struct inode *inode,
4621 const char *name, int name_len, int add_backref, u64 index)
4624 struct btrfs_key key;
4625 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4627 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4628 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4630 key.objectid = inode->i_ino;
4631 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4635 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4636 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4637 key.objectid, root->root_key.objectid,
4638 parent_inode->i_ino,
4639 index, name, name_len);
4640 } else if (add_backref) {
4641 ret = btrfs_insert_inode_ref(trans, root,
4642 name, name_len, inode->i_ino,
4643 parent_inode->i_ino, index);
4647 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4648 parent_inode->i_ino, &key,
4649 btrfs_inode_type(inode), index);
4652 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4654 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4655 ret = btrfs_update_inode(trans, root, parent_inode);
4660 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4661 struct inode *dir, struct dentry *dentry,
4662 struct inode *inode, int backref, u64 index)
4664 int err = btrfs_add_link(trans, dir, inode,
4665 dentry->d_name.name, dentry->d_name.len,
4668 d_instantiate(dentry, inode);
4676 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4677 int mode, dev_t rdev)
4679 struct btrfs_trans_handle *trans;
4680 struct btrfs_root *root = BTRFS_I(dir)->root;
4681 struct inode *inode = NULL;
4685 unsigned long nr = 0;
4688 if (!new_valid_dev(rdev))
4691 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4696 * 2 for inode item and ref
4698 * 1 for xattr if selinux is on
4700 trans = btrfs_start_transaction(root, 5);
4702 return PTR_ERR(trans);
4704 btrfs_set_trans_block_group(trans, dir);
4706 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4707 dentry->d_name.len, dir->i_ino, objectid,
4708 BTRFS_I(dir)->block_group, mode, &index);
4709 err = PTR_ERR(inode);
4713 err = btrfs_init_inode_security(trans, inode, dir);
4719 btrfs_set_trans_block_group(trans, inode);
4720 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4724 inode->i_op = &btrfs_special_inode_operations;
4725 init_special_inode(inode, inode->i_mode, rdev);
4726 btrfs_update_inode(trans, root, inode);
4728 btrfs_update_inode_block_group(trans, inode);
4729 btrfs_update_inode_block_group(trans, dir);
4731 nr = trans->blocks_used;
4732 btrfs_end_transaction_throttle(trans, root);
4733 btrfs_btree_balance_dirty(root, nr);
4735 inode_dec_link_count(inode);
4741 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4742 int mode, struct nameidata *nd)
4744 struct btrfs_trans_handle *trans;
4745 struct btrfs_root *root = BTRFS_I(dir)->root;
4746 struct inode *inode = NULL;
4749 unsigned long nr = 0;
4753 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4757 * 2 for inode item and ref
4759 * 1 for xattr if selinux is on
4761 trans = btrfs_start_transaction(root, 5);
4763 return PTR_ERR(trans);
4765 btrfs_set_trans_block_group(trans, dir);
4767 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4768 dentry->d_name.len, dir->i_ino, objectid,
4769 BTRFS_I(dir)->block_group, mode, &index);
4770 err = PTR_ERR(inode);
4774 err = btrfs_init_inode_security(trans, inode, dir);
4780 btrfs_set_trans_block_group(trans, inode);
4781 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4785 inode->i_mapping->a_ops = &btrfs_aops;
4786 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4787 inode->i_fop = &btrfs_file_operations;
4788 inode->i_op = &btrfs_file_inode_operations;
4789 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4791 btrfs_update_inode_block_group(trans, inode);
4792 btrfs_update_inode_block_group(trans, dir);
4794 nr = trans->blocks_used;
4795 btrfs_end_transaction_throttle(trans, root);
4797 inode_dec_link_count(inode);
4800 btrfs_btree_balance_dirty(root, nr);
4804 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4805 struct dentry *dentry)
4807 struct btrfs_trans_handle *trans;
4808 struct btrfs_root *root = BTRFS_I(dir)->root;
4809 struct inode *inode = old_dentry->d_inode;
4811 unsigned long nr = 0;
4815 if (inode->i_nlink == 0)
4818 /* do not allow sys_link's with other subvols of the same device */
4819 if (root->objectid != BTRFS_I(inode)->root->objectid)
4822 btrfs_inc_nlink(inode);
4823 inode->i_ctime = CURRENT_TIME;
4825 err = btrfs_set_inode_index(dir, &index);
4830 * 2 items for inode and inode ref
4831 * 2 items for dir items
4832 * 1 item for parent inode
4834 trans = btrfs_start_transaction(root, 5);
4835 if (IS_ERR(trans)) {
4836 err = PTR_ERR(trans);
4840 btrfs_set_trans_block_group(trans, dir);
4843 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4848 struct dentry *parent = dget_parent(dentry);
4849 btrfs_update_inode_block_group(trans, dir);
4850 err = btrfs_update_inode(trans, root, inode);
4852 btrfs_log_new_name(trans, inode, NULL, parent);
4856 nr = trans->blocks_used;
4857 btrfs_end_transaction_throttle(trans, root);
4860 inode_dec_link_count(inode);
4863 btrfs_btree_balance_dirty(root, nr);
4867 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4869 struct inode *inode = NULL;
4870 struct btrfs_trans_handle *trans;
4871 struct btrfs_root *root = BTRFS_I(dir)->root;
4873 int drop_on_err = 0;
4876 unsigned long nr = 1;
4878 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4883 * 2 items for inode and ref
4884 * 2 items for dir items
4885 * 1 for xattr if selinux is on
4887 trans = btrfs_start_transaction(root, 5);
4889 return PTR_ERR(trans);
4890 btrfs_set_trans_block_group(trans, dir);
4892 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4893 dentry->d_name.len, dir->i_ino, objectid,
4894 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4896 if (IS_ERR(inode)) {
4897 err = PTR_ERR(inode);
4903 err = btrfs_init_inode_security(trans, inode, dir);
4907 inode->i_op = &btrfs_dir_inode_operations;
4908 inode->i_fop = &btrfs_dir_file_operations;
4909 btrfs_set_trans_block_group(trans, inode);
4911 btrfs_i_size_write(inode, 0);
4912 err = btrfs_update_inode(trans, root, inode);
4916 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4917 dentry->d_name.len, 0, index);
4921 d_instantiate(dentry, inode);
4923 btrfs_update_inode_block_group(trans, inode);
4924 btrfs_update_inode_block_group(trans, dir);
4927 nr = trans->blocks_used;
4928 btrfs_end_transaction_throttle(trans, root);
4931 btrfs_btree_balance_dirty(root, nr);
4935 /* helper for btfs_get_extent. Given an existing extent in the tree,
4936 * and an extent that you want to insert, deal with overlap and insert
4937 * the new extent into the tree.
4939 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4940 struct extent_map *existing,
4941 struct extent_map *em,
4942 u64 map_start, u64 map_len)
4946 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4947 start_diff = map_start - em->start;
4948 em->start = map_start;
4950 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4951 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4952 em->block_start += start_diff;
4953 em->block_len -= start_diff;
4955 return add_extent_mapping(em_tree, em);
4958 static noinline int uncompress_inline(struct btrfs_path *path,
4959 struct inode *inode, struct page *page,
4960 size_t pg_offset, u64 extent_offset,
4961 struct btrfs_file_extent_item *item)
4964 struct extent_buffer *leaf = path->nodes[0];
4967 unsigned long inline_size;
4971 WARN_ON(pg_offset != 0);
4972 compress_type = btrfs_file_extent_compression(leaf, item);
4973 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4974 inline_size = btrfs_file_extent_inline_item_len(leaf,
4975 btrfs_item_nr(leaf, path->slots[0]));
4976 tmp = kmalloc(inline_size, GFP_NOFS);
4977 ptr = btrfs_file_extent_inline_start(item);
4979 read_extent_buffer(leaf, tmp, ptr, inline_size);
4981 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4982 ret = btrfs_decompress(compress_type, tmp, page,
4983 extent_offset, inline_size, max_size);
4985 char *kaddr = kmap_atomic(page, KM_USER0);
4986 unsigned long copy_size = min_t(u64,
4987 PAGE_CACHE_SIZE - pg_offset,
4988 max_size - extent_offset);
4989 memset(kaddr + pg_offset, 0, copy_size);
4990 kunmap_atomic(kaddr, KM_USER0);
4997 * a bit scary, this does extent mapping from logical file offset to the disk.
4998 * the ugly parts come from merging extents from the disk with the in-ram
4999 * representation. This gets more complex because of the data=ordered code,
5000 * where the in-ram extents might be locked pending data=ordered completion.
5002 * This also copies inline extents directly into the page.
5005 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5006 size_t pg_offset, u64 start, u64 len,
5012 u64 extent_start = 0;
5014 u64 objectid = inode->i_ino;
5016 struct btrfs_path *path = NULL;
5017 struct btrfs_root *root = BTRFS_I(inode)->root;
5018 struct btrfs_file_extent_item *item;
5019 struct extent_buffer *leaf;
5020 struct btrfs_key found_key;
5021 struct extent_map *em = NULL;
5022 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5023 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5024 struct btrfs_trans_handle *trans = NULL;
5028 read_lock(&em_tree->lock);
5029 em = lookup_extent_mapping(em_tree, start, len);
5031 em->bdev = root->fs_info->fs_devices->latest_bdev;
5032 read_unlock(&em_tree->lock);
5035 if (em->start > start || em->start + em->len <= start)
5036 free_extent_map(em);
5037 else if (em->block_start == EXTENT_MAP_INLINE && page)
5038 free_extent_map(em);
5042 em = alloc_extent_map(GFP_NOFS);
5047 em->bdev = root->fs_info->fs_devices->latest_bdev;
5048 em->start = EXTENT_MAP_HOLE;
5049 em->orig_start = EXTENT_MAP_HOLE;
5051 em->block_len = (u64)-1;
5054 path = btrfs_alloc_path();
5058 ret = btrfs_lookup_file_extent(trans, root, path,
5059 objectid, start, trans != NULL);
5066 if (path->slots[0] == 0)
5071 leaf = path->nodes[0];
5072 item = btrfs_item_ptr(leaf, path->slots[0],
5073 struct btrfs_file_extent_item);
5074 /* are we inside the extent that was found? */
5075 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5076 found_type = btrfs_key_type(&found_key);
5077 if (found_key.objectid != objectid ||
5078 found_type != BTRFS_EXTENT_DATA_KEY) {
5082 found_type = btrfs_file_extent_type(leaf, item);
5083 extent_start = found_key.offset;
5084 compress_type = btrfs_file_extent_compression(leaf, item);
5085 if (found_type == BTRFS_FILE_EXTENT_REG ||
5086 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5087 extent_end = extent_start +
5088 btrfs_file_extent_num_bytes(leaf, item);
5089 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5091 size = btrfs_file_extent_inline_len(leaf, item);
5092 extent_end = (extent_start + size + root->sectorsize - 1) &
5093 ~((u64)root->sectorsize - 1);
5096 if (start >= extent_end) {
5098 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5099 ret = btrfs_next_leaf(root, path);
5106 leaf = path->nodes[0];
5108 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5109 if (found_key.objectid != objectid ||
5110 found_key.type != BTRFS_EXTENT_DATA_KEY)
5112 if (start + len <= found_key.offset)
5115 em->len = found_key.offset - start;
5119 if (found_type == BTRFS_FILE_EXTENT_REG ||
5120 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5121 em->start = extent_start;
5122 em->len = extent_end - extent_start;
5123 em->orig_start = extent_start -
5124 btrfs_file_extent_offset(leaf, item);
5125 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5127 em->block_start = EXTENT_MAP_HOLE;
5130 if (compress_type != BTRFS_COMPRESS_NONE) {
5131 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5132 em->compress_type = compress_type;
5133 em->block_start = bytenr;
5134 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5137 bytenr += btrfs_file_extent_offset(leaf, item);
5138 em->block_start = bytenr;
5139 em->block_len = em->len;
5140 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5141 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5144 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5148 size_t extent_offset;
5151 em->block_start = EXTENT_MAP_INLINE;
5152 if (!page || create) {
5153 em->start = extent_start;
5154 em->len = extent_end - extent_start;
5158 size = btrfs_file_extent_inline_len(leaf, item);
5159 extent_offset = page_offset(page) + pg_offset - extent_start;
5160 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5161 size - extent_offset);
5162 em->start = extent_start + extent_offset;
5163 em->len = (copy_size + root->sectorsize - 1) &
5164 ~((u64)root->sectorsize - 1);
5165 em->orig_start = EXTENT_MAP_INLINE;
5166 if (compress_type) {
5167 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5168 em->compress_type = compress_type;
5170 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5171 if (create == 0 && !PageUptodate(page)) {
5172 if (btrfs_file_extent_compression(leaf, item) !=
5173 BTRFS_COMPRESS_NONE) {
5174 ret = uncompress_inline(path, inode, page,
5176 extent_offset, item);
5180 read_extent_buffer(leaf, map + pg_offset, ptr,
5182 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5183 memset(map + pg_offset + copy_size, 0,
5184 PAGE_CACHE_SIZE - pg_offset -
5189 flush_dcache_page(page);
5190 } else if (create && PageUptodate(page)) {
5194 free_extent_map(em);
5196 btrfs_release_path(root, path);
5197 trans = btrfs_join_transaction(root, 1);
5199 return ERR_CAST(trans);
5203 write_extent_buffer(leaf, map + pg_offset, ptr,
5206 btrfs_mark_buffer_dirty(leaf);
5208 set_extent_uptodate(io_tree, em->start,
5209 extent_map_end(em) - 1, GFP_NOFS);
5212 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5219 em->block_start = EXTENT_MAP_HOLE;
5220 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5222 btrfs_release_path(root, path);
5223 if (em->start > start || extent_map_end(em) <= start) {
5224 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5225 "[%llu %llu]\n", (unsigned long long)em->start,
5226 (unsigned long long)em->len,
5227 (unsigned long long)start,
5228 (unsigned long long)len);
5234 write_lock(&em_tree->lock);
5235 ret = add_extent_mapping(em_tree, em);
5236 /* it is possible that someone inserted the extent into the tree
5237 * while we had the lock dropped. It is also possible that
5238 * an overlapping map exists in the tree
5240 if (ret == -EEXIST) {
5241 struct extent_map *existing;
5245 existing = lookup_extent_mapping(em_tree, start, len);
5246 if (existing && (existing->start > start ||
5247 existing->start + existing->len <= start)) {
5248 free_extent_map(existing);
5252 existing = lookup_extent_mapping(em_tree, em->start,
5255 err = merge_extent_mapping(em_tree, existing,
5258 free_extent_map(existing);
5260 free_extent_map(em);
5265 free_extent_map(em);
5269 free_extent_map(em);
5274 write_unlock(&em_tree->lock);
5277 trace_btrfs_get_extent(root, em);
5280 btrfs_free_path(path);
5282 ret = btrfs_end_transaction(trans, root);
5287 free_extent_map(em);
5288 return ERR_PTR(err);
5293 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5294 size_t pg_offset, u64 start, u64 len,
5297 struct extent_map *em;
5298 struct extent_map *hole_em = NULL;
5299 u64 range_start = start;
5305 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5310 * if our em maps to a hole, there might
5311 * actually be delalloc bytes behind it
5313 if (em->block_start != EXTENT_MAP_HOLE)
5319 /* check to see if we've wrapped (len == -1 or similar) */
5328 /* ok, we didn't find anything, lets look for delalloc */
5329 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5330 end, len, EXTENT_DELALLOC, 1);
5331 found_end = range_start + found;
5332 if (found_end < range_start)
5333 found_end = (u64)-1;
5336 * we didn't find anything useful, return
5337 * the original results from get_extent()
5339 if (range_start > end || found_end <= start) {
5345 /* adjust the range_start to make sure it doesn't
5346 * go backwards from the start they passed in
5348 range_start = max(start,range_start);
5349 found = found_end - range_start;
5352 u64 hole_start = start;
5355 em = alloc_extent_map(GFP_NOFS);
5361 * when btrfs_get_extent can't find anything it
5362 * returns one huge hole
5364 * make sure what it found really fits our range, and
5365 * adjust to make sure it is based on the start from
5369 u64 calc_end = extent_map_end(hole_em);
5371 if (calc_end <= start || (hole_em->start > end)) {
5372 free_extent_map(hole_em);
5375 hole_start = max(hole_em->start, start);
5376 hole_len = calc_end - hole_start;
5380 if (hole_em && range_start > hole_start) {
5381 /* our hole starts before our delalloc, so we
5382 * have to return just the parts of the hole
5383 * that go until the delalloc starts
5385 em->len = min(hole_len,
5386 range_start - hole_start);
5387 em->start = hole_start;
5388 em->orig_start = hole_start;
5390 * don't adjust block start at all,
5391 * it is fixed at EXTENT_MAP_HOLE
5393 em->block_start = hole_em->block_start;
5394 em->block_len = hole_len;
5396 em->start = range_start;
5398 em->orig_start = range_start;
5399 em->block_start = EXTENT_MAP_DELALLOC;
5400 em->block_len = found;
5402 } else if (hole_em) {
5407 free_extent_map(hole_em);
5409 free_extent_map(em);
5410 return ERR_PTR(err);
5415 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5418 struct btrfs_root *root = BTRFS_I(inode)->root;
5419 struct btrfs_trans_handle *trans;
5420 struct extent_map *em;
5421 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5422 struct btrfs_key ins;
5426 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5428 trans = btrfs_join_transaction(root, 0);
5430 return ERR_CAST(trans);
5432 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5434 alloc_hint = get_extent_allocation_hint(inode, start, len);
5435 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5436 alloc_hint, (u64)-1, &ins, 1);
5442 em = alloc_extent_map(GFP_NOFS);
5444 em = ERR_PTR(-ENOMEM);
5449 em->orig_start = em->start;
5450 em->len = ins.offset;
5452 em->block_start = ins.objectid;
5453 em->block_len = ins.offset;
5454 em->bdev = root->fs_info->fs_devices->latest_bdev;
5455 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5458 write_lock(&em_tree->lock);
5459 ret = add_extent_mapping(em_tree, em);
5460 write_unlock(&em_tree->lock);
5463 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5466 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5467 ins.offset, ins.offset, 0);
5469 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5473 btrfs_end_transaction(trans, root);
5478 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5479 * block must be cow'd
5481 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5482 struct inode *inode, u64 offset, u64 len)
5484 struct btrfs_path *path;
5486 struct extent_buffer *leaf;
5487 struct btrfs_root *root = BTRFS_I(inode)->root;
5488 struct btrfs_file_extent_item *fi;
5489 struct btrfs_key key;
5497 path = btrfs_alloc_path();
5501 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5506 slot = path->slots[0];
5509 /* can't find the item, must cow */
5516 leaf = path->nodes[0];
5517 btrfs_item_key_to_cpu(leaf, &key, slot);
5518 if (key.objectid != inode->i_ino ||
5519 key.type != BTRFS_EXTENT_DATA_KEY) {
5520 /* not our file or wrong item type, must cow */
5524 if (key.offset > offset) {
5525 /* Wrong offset, must cow */
5529 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5530 found_type = btrfs_file_extent_type(leaf, fi);
5531 if (found_type != BTRFS_FILE_EXTENT_REG &&
5532 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5533 /* not a regular extent, must cow */
5536 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5537 backref_offset = btrfs_file_extent_offset(leaf, fi);
5539 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5540 if (extent_end < offset + len) {
5541 /* extent doesn't include our full range, must cow */
5545 if (btrfs_extent_readonly(root, disk_bytenr))
5549 * look for other files referencing this extent, if we
5550 * find any we must cow
5552 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5553 key.offset - backref_offset, disk_bytenr))
5557 * adjust disk_bytenr and num_bytes to cover just the bytes
5558 * in this extent we are about to write. If there
5559 * are any csums in that range we have to cow in order
5560 * to keep the csums correct
5562 disk_bytenr += backref_offset;
5563 disk_bytenr += offset - key.offset;
5564 num_bytes = min(offset + len, extent_end) - offset;
5565 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5568 * all of the above have passed, it is safe to overwrite this extent
5573 btrfs_free_path(path);
5577 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5578 struct buffer_head *bh_result, int create)
5580 struct extent_map *em;
5581 struct btrfs_root *root = BTRFS_I(inode)->root;
5582 u64 start = iblock << inode->i_blkbits;
5583 u64 len = bh_result->b_size;
5584 struct btrfs_trans_handle *trans;
5586 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5591 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5592 * io. INLINE is special, and we could probably kludge it in here, but
5593 * it's still buffered so for safety lets just fall back to the generic
5596 * For COMPRESSED we _have_ to read the entire extent in so we can
5597 * decompress it, so there will be buffering required no matter what we
5598 * do, so go ahead and fallback to buffered.
5600 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5601 * to buffered IO. Don't blame me, this is the price we pay for using
5604 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5605 em->block_start == EXTENT_MAP_INLINE) {
5606 free_extent_map(em);
5610 /* Just a good old fashioned hole, return */
5611 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5612 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5613 free_extent_map(em);
5614 /* DIO will do one hole at a time, so just unlock a sector */
5615 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5616 start + root->sectorsize - 1, GFP_NOFS);
5621 * We don't allocate a new extent in the following cases
5623 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5625 * 2) The extent is marked as PREALLOC. We're good to go here and can
5626 * just use the extent.
5630 len = em->len - (start - em->start);
5634 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5635 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5636 em->block_start != EXTENT_MAP_HOLE)) {
5641 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5642 type = BTRFS_ORDERED_PREALLOC;
5644 type = BTRFS_ORDERED_NOCOW;
5645 len = min(len, em->len - (start - em->start));
5646 block_start = em->block_start + (start - em->start);
5649 * we're not going to log anything, but we do need
5650 * to make sure the current transaction stays open
5651 * while we look for nocow cross refs
5653 trans = btrfs_join_transaction(root, 0);
5657 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5658 ret = btrfs_add_ordered_extent_dio(inode, start,
5659 block_start, len, len, type);
5660 btrfs_end_transaction(trans, root);
5662 free_extent_map(em);
5667 btrfs_end_transaction(trans, root);
5671 * this will cow the extent, reset the len in case we changed
5674 len = bh_result->b_size;
5675 free_extent_map(em);
5676 em = btrfs_new_extent_direct(inode, start, len);
5679 len = min(len, em->len - (start - em->start));
5681 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5682 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5685 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5687 bh_result->b_size = len;
5688 bh_result->b_bdev = em->bdev;
5689 set_buffer_mapped(bh_result);
5690 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5691 set_buffer_new(bh_result);
5693 free_extent_map(em);
5698 struct btrfs_dio_private {
5699 struct inode *inode;
5706 /* number of bios pending for this dio */
5707 atomic_t pending_bios;
5712 struct bio *orig_bio;
5715 static void btrfs_endio_direct_read(struct bio *bio, int err)
5717 struct btrfs_dio_private *dip = bio->bi_private;
5718 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5719 struct bio_vec *bvec = bio->bi_io_vec;
5720 struct inode *inode = dip->inode;
5721 struct btrfs_root *root = BTRFS_I(inode)->root;
5723 u32 *private = dip->csums;
5725 start = dip->logical_offset;
5727 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5728 struct page *page = bvec->bv_page;
5731 unsigned long flags;
5733 local_irq_save(flags);
5734 kaddr = kmap_atomic(page, KM_IRQ0);
5735 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5736 csum, bvec->bv_len);
5737 btrfs_csum_final(csum, (char *)&csum);
5738 kunmap_atomic(kaddr, KM_IRQ0);
5739 local_irq_restore(flags);
5741 flush_dcache_page(bvec->bv_page);
5742 if (csum != *private) {
5743 printk(KERN_ERR "btrfs csum failed ino %lu off"
5744 " %llu csum %u private %u\n",
5745 inode->i_ino, (unsigned long long)start,
5751 start += bvec->bv_len;
5754 } while (bvec <= bvec_end);
5756 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5757 dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5758 bio->bi_private = dip->private;
5763 /* If we had a csum failure make sure to clear the uptodate flag */
5765 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5766 dio_end_io(bio, err);
5769 static void btrfs_endio_direct_write(struct bio *bio, int err)
5771 struct btrfs_dio_private *dip = bio->bi_private;
5772 struct inode *inode = dip->inode;
5773 struct btrfs_root *root = BTRFS_I(inode)->root;
5774 struct btrfs_trans_handle *trans;
5775 struct btrfs_ordered_extent *ordered = NULL;
5776 struct extent_state *cached_state = NULL;
5777 u64 ordered_offset = dip->logical_offset;
5778 u64 ordered_bytes = dip->bytes;
5784 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5792 trans = btrfs_join_transaction(root, 1);
5793 if (IS_ERR(trans)) {
5797 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5799 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5800 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5802 ret = btrfs_update_inode(trans, root, inode);
5807 lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5808 ordered->file_offset + ordered->len - 1, 0,
5809 &cached_state, GFP_NOFS);
5811 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5812 ret = btrfs_mark_extent_written(trans, inode,
5813 ordered->file_offset,
5814 ordered->file_offset +
5821 ret = insert_reserved_file_extent(trans, inode,
5822 ordered->file_offset,
5828 BTRFS_FILE_EXTENT_REG);
5829 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5830 ordered->file_offset, ordered->len);
5838 add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5839 btrfs_ordered_update_i_size(inode, 0, ordered);
5840 btrfs_update_inode(trans, root, inode);
5842 unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5843 ordered->file_offset + ordered->len - 1,
5844 &cached_state, GFP_NOFS);
5846 btrfs_delalloc_release_metadata(inode, ordered->len);
5847 btrfs_end_transaction(trans, root);
5848 ordered_offset = ordered->file_offset + ordered->len;
5849 btrfs_put_ordered_extent(ordered);
5850 btrfs_put_ordered_extent(ordered);
5854 * our bio might span multiple ordered extents. If we haven't
5855 * completed the accounting for the whole dio, go back and try again
5857 if (ordered_offset < dip->logical_offset + dip->bytes) {
5858 ordered_bytes = dip->logical_offset + dip->bytes -
5863 bio->bi_private = dip->private;
5868 /* If we had an error make sure to clear the uptodate flag */
5870 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5871 dio_end_io(bio, err);
5874 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5875 struct bio *bio, int mirror_num,
5876 unsigned long bio_flags, u64 offset)
5879 struct btrfs_root *root = BTRFS_I(inode)->root;
5880 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5885 static void btrfs_end_dio_bio(struct bio *bio, int err)
5887 struct btrfs_dio_private *dip = bio->bi_private;
5890 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5891 "sector %#Lx len %u err no %d\n",
5892 dip->inode->i_ino, bio->bi_rw,
5893 (unsigned long long)bio->bi_sector, bio->bi_size, err);
5897 * before atomic variable goto zero, we must make sure
5898 * dip->errors is perceived to be set.
5900 smp_mb__before_atomic_dec();
5903 /* if there are more bios still pending for this dio, just exit */
5904 if (!atomic_dec_and_test(&dip->pending_bios))
5908 bio_io_error(dip->orig_bio);
5910 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5911 bio_endio(dip->orig_bio, 0);
5917 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5918 u64 first_sector, gfp_t gfp_flags)
5920 int nr_vecs = bio_get_nr_vecs(bdev);
5921 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5924 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5925 int rw, u64 file_offset, int skip_sum,
5928 int write = rw & REQ_WRITE;
5929 struct btrfs_root *root = BTRFS_I(inode)->root;
5933 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5937 if (write && !skip_sum) {
5938 ret = btrfs_wq_submit_bio(root->fs_info,
5939 inode, rw, bio, 0, 0,
5941 __btrfs_submit_bio_start_direct_io,
5942 __btrfs_submit_bio_done);
5944 } else if (!skip_sum)
5945 btrfs_lookup_bio_sums_dio(root, inode, bio,
5946 file_offset, csums);
5948 ret = btrfs_map_bio(root, rw, bio, 0, 1);
5954 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5957 struct inode *inode = dip->inode;
5958 struct btrfs_root *root = BTRFS_I(inode)->root;
5959 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5961 struct bio *orig_bio = dip->orig_bio;
5962 struct bio_vec *bvec = orig_bio->bi_io_vec;
5963 u64 start_sector = orig_bio->bi_sector;
5964 u64 file_offset = dip->logical_offset;
5968 u32 *csums = dip->csums;
5970 int write = rw & REQ_WRITE;
5972 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5975 bio->bi_private = dip;
5976 bio->bi_end_io = btrfs_end_dio_bio;
5977 atomic_inc(&dip->pending_bios);
5979 map_length = orig_bio->bi_size;
5980 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5981 &map_length, NULL, 0);
5987 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5988 if (unlikely(map_length < submit_len + bvec->bv_len ||
5989 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5990 bvec->bv_offset) < bvec->bv_len)) {
5992 * inc the count before we submit the bio so
5993 * we know the end IO handler won't happen before
5994 * we inc the count. Otherwise, the dip might get freed
5995 * before we're done setting it up
5997 atomic_inc(&dip->pending_bios);
5998 ret = __btrfs_submit_dio_bio(bio, inode, rw,
5999 file_offset, skip_sum,
6003 atomic_dec(&dip->pending_bios);
6007 /* Write's use the ordered csums */
6008 if (!write && !skip_sum)
6009 csums = csums + nr_pages;
6010 start_sector += submit_len >> 9;
6011 file_offset += submit_len;
6016 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6017 start_sector, GFP_NOFS);
6020 bio->bi_private = dip;
6021 bio->bi_end_io = btrfs_end_dio_bio;
6023 map_length = orig_bio->bi_size;
6024 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6025 &map_length, NULL, 0);
6031 submit_len += bvec->bv_len;
6037 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6046 * before atomic variable goto zero, we must
6047 * make sure dip->errors is perceived to be set.
6049 smp_mb__before_atomic_dec();
6050 if (atomic_dec_and_test(&dip->pending_bios))
6051 bio_io_error(dip->orig_bio);
6053 /* bio_end_io() will handle error, so we needn't return it */
6057 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6060 struct btrfs_root *root = BTRFS_I(inode)->root;
6061 struct btrfs_dio_private *dip;
6062 struct bio_vec *bvec = bio->bi_io_vec;
6064 int write = rw & REQ_WRITE;
6067 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6069 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6076 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6077 if (!write && !skip_sum) {
6078 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6086 dip->private = bio->bi_private;
6088 dip->logical_offset = file_offset;
6092 dip->bytes += bvec->bv_len;
6094 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6096 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6097 bio->bi_private = dip;
6099 dip->orig_bio = bio;
6100 atomic_set(&dip->pending_bios, 0);
6103 bio->bi_end_io = btrfs_endio_direct_write;
6105 bio->bi_end_io = btrfs_endio_direct_read;
6107 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6112 * If this is a write, we need to clean up the reserved space and kill
6113 * the ordered extent.
6116 struct btrfs_ordered_extent *ordered;
6117 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6118 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6119 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6120 btrfs_free_reserved_extent(root, ordered->start,
6122 btrfs_put_ordered_extent(ordered);
6123 btrfs_put_ordered_extent(ordered);
6125 bio_endio(bio, ret);
6128 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6129 const struct iovec *iov, loff_t offset,
6130 unsigned long nr_segs)
6135 unsigned blocksize_mask = root->sectorsize - 1;
6136 ssize_t retval = -EINVAL;
6137 loff_t end = offset;
6139 if (offset & blocksize_mask)
6142 /* Check the memory alignment. Blocks cannot straddle pages */
6143 for (seg = 0; seg < nr_segs; seg++) {
6144 addr = (unsigned long)iov[seg].iov_base;
6145 size = iov[seg].iov_len;
6147 if ((addr & blocksize_mask) || (size & blocksize_mask))
6154 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6155 const struct iovec *iov, loff_t offset,
6156 unsigned long nr_segs)
6158 struct file *file = iocb->ki_filp;
6159 struct inode *inode = file->f_mapping->host;
6160 struct btrfs_ordered_extent *ordered;
6161 struct extent_state *cached_state = NULL;
6162 u64 lockstart, lockend;
6164 int writing = rw & WRITE;
6166 size_t count = iov_length(iov, nr_segs);
6168 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6174 lockend = offset + count - 1;
6177 ret = btrfs_delalloc_reserve_space(inode, count);
6183 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6184 0, &cached_state, GFP_NOFS);
6186 * We're concerned with the entire range that we're going to be
6187 * doing DIO to, so we need to make sure theres no ordered
6188 * extents in this range.
6190 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6191 lockend - lockstart + 1);
6194 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6195 &cached_state, GFP_NOFS);
6196 btrfs_start_ordered_extent(inode, ordered, 1);
6197 btrfs_put_ordered_extent(ordered);
6202 * we don't use btrfs_set_extent_delalloc because we don't want
6203 * the dirty or uptodate bits
6206 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6207 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6208 EXTENT_DELALLOC, 0, NULL, &cached_state,
6211 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6212 lockend, EXTENT_LOCKED | write_bits,
6213 1, 0, &cached_state, GFP_NOFS);
6218 free_extent_state(cached_state);
6219 cached_state = NULL;
6221 ret = __blockdev_direct_IO(rw, iocb, inode,
6222 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6223 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6224 btrfs_submit_direct, 0);
6226 if (ret < 0 && ret != -EIOCBQUEUED) {
6227 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6228 offset + iov_length(iov, nr_segs) - 1,
6229 EXTENT_LOCKED | write_bits, 1, 0,
6230 &cached_state, GFP_NOFS);
6231 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6233 * We're falling back to buffered, unlock the section we didn't
6236 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6237 offset + iov_length(iov, nr_segs) - 1,
6238 EXTENT_LOCKED | write_bits, 1, 0,
6239 &cached_state, GFP_NOFS);
6242 free_extent_state(cached_state);
6246 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6247 __u64 start, __u64 len)
6249 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6252 int btrfs_readpage(struct file *file, struct page *page)
6254 struct extent_io_tree *tree;
6255 tree = &BTRFS_I(page->mapping->host)->io_tree;
6256 return extent_read_full_page(tree, page, btrfs_get_extent);
6259 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6261 struct extent_io_tree *tree;
6264 if (current->flags & PF_MEMALLOC) {
6265 redirty_page_for_writepage(wbc, page);
6269 tree = &BTRFS_I(page->mapping->host)->io_tree;
6270 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6273 int btrfs_writepages(struct address_space *mapping,
6274 struct writeback_control *wbc)
6276 struct extent_io_tree *tree;
6278 tree = &BTRFS_I(mapping->host)->io_tree;
6279 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6283 btrfs_readpages(struct file *file, struct address_space *mapping,
6284 struct list_head *pages, unsigned nr_pages)
6286 struct extent_io_tree *tree;
6287 tree = &BTRFS_I(mapping->host)->io_tree;
6288 return extent_readpages(tree, mapping, pages, nr_pages,
6291 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6293 struct extent_io_tree *tree;
6294 struct extent_map_tree *map;
6297 tree = &BTRFS_I(page->mapping->host)->io_tree;
6298 map = &BTRFS_I(page->mapping->host)->extent_tree;
6299 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6301 ClearPagePrivate(page);
6302 set_page_private(page, 0);
6303 page_cache_release(page);
6308 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6310 if (PageWriteback(page) || PageDirty(page))
6312 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6315 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6317 struct extent_io_tree *tree;
6318 struct btrfs_ordered_extent *ordered;
6319 struct extent_state *cached_state = NULL;
6320 u64 page_start = page_offset(page);
6321 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6325 * we have the page locked, so new writeback can't start,
6326 * and the dirty bit won't be cleared while we are here.
6328 * Wait for IO on this page so that we can safely clear
6329 * the PagePrivate2 bit and do ordered accounting
6331 wait_on_page_writeback(page);
6333 tree = &BTRFS_I(page->mapping->host)->io_tree;
6335 btrfs_releasepage(page, GFP_NOFS);
6338 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6340 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6344 * IO on this page will never be started, so we need
6345 * to account for any ordered extents now
6347 clear_extent_bit(tree, page_start, page_end,
6348 EXTENT_DIRTY | EXTENT_DELALLOC |
6349 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6350 &cached_state, GFP_NOFS);
6352 * whoever cleared the private bit is responsible
6353 * for the finish_ordered_io
6355 if (TestClearPagePrivate2(page)) {
6356 btrfs_finish_ordered_io(page->mapping->host,
6357 page_start, page_end);
6359 btrfs_put_ordered_extent(ordered);
6360 cached_state = NULL;
6361 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6364 clear_extent_bit(tree, page_start, page_end,
6365 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6366 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6367 __btrfs_releasepage(page, GFP_NOFS);
6369 ClearPageChecked(page);
6370 if (PagePrivate(page)) {
6371 ClearPagePrivate(page);
6372 set_page_private(page, 0);
6373 page_cache_release(page);
6378 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6379 * called from a page fault handler when a page is first dirtied. Hence we must
6380 * be careful to check for EOF conditions here. We set the page up correctly
6381 * for a written page which means we get ENOSPC checking when writing into
6382 * holes and correct delalloc and unwritten extent mapping on filesystems that
6383 * support these features.
6385 * We are not allowed to take the i_mutex here so we have to play games to
6386 * protect against truncate races as the page could now be beyond EOF. Because
6387 * vmtruncate() writes the inode size before removing pages, once we have the
6388 * page lock we can determine safely if the page is beyond EOF. If it is not
6389 * beyond EOF, then the page is guaranteed safe against truncation until we
6392 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6394 struct page *page = vmf->page;
6395 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6396 struct btrfs_root *root = BTRFS_I(inode)->root;
6397 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6398 struct btrfs_ordered_extent *ordered;
6399 struct extent_state *cached_state = NULL;
6401 unsigned long zero_start;
6407 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6411 else /* -ENOSPC, -EIO, etc */
6412 ret = VM_FAULT_SIGBUS;
6416 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6419 size = i_size_read(inode);
6420 page_start = page_offset(page);
6421 page_end = page_start + PAGE_CACHE_SIZE - 1;
6423 if ((page->mapping != inode->i_mapping) ||
6424 (page_start >= size)) {
6425 /* page got truncated out from underneath us */
6428 wait_on_page_writeback(page);
6430 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6432 set_page_extent_mapped(page);
6435 * we can't set the delalloc bits if there are pending ordered
6436 * extents. Drop our locks and wait for them to finish
6438 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6440 unlock_extent_cached(io_tree, page_start, page_end,
6441 &cached_state, GFP_NOFS);
6443 btrfs_start_ordered_extent(inode, ordered, 1);
6444 btrfs_put_ordered_extent(ordered);
6449 * XXX - page_mkwrite gets called every time the page is dirtied, even
6450 * if it was already dirty, so for space accounting reasons we need to
6451 * clear any delalloc bits for the range we are fixing to save. There
6452 * is probably a better way to do this, but for now keep consistent with
6453 * prepare_pages in the normal write path.
6455 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6456 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6457 0, 0, &cached_state, GFP_NOFS);
6459 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6462 unlock_extent_cached(io_tree, page_start, page_end,
6463 &cached_state, GFP_NOFS);
6464 ret = VM_FAULT_SIGBUS;
6469 /* page is wholly or partially inside EOF */
6470 if (page_start + PAGE_CACHE_SIZE > size)
6471 zero_start = size & ~PAGE_CACHE_MASK;
6473 zero_start = PAGE_CACHE_SIZE;
6475 if (zero_start != PAGE_CACHE_SIZE) {
6477 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6478 flush_dcache_page(page);
6481 ClearPageChecked(page);
6482 set_page_dirty(page);
6483 SetPageUptodate(page);
6485 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6486 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6488 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6492 return VM_FAULT_LOCKED;
6494 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6499 static int btrfs_truncate(struct inode *inode)
6501 struct btrfs_root *root = BTRFS_I(inode)->root;
6504 struct btrfs_trans_handle *trans;
6506 u64 mask = root->sectorsize - 1;
6508 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6512 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6513 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6515 trans = btrfs_start_transaction(root, 5);
6517 return PTR_ERR(trans);
6519 btrfs_set_trans_block_group(trans, inode);
6521 ret = btrfs_orphan_add(trans, inode);
6523 btrfs_end_transaction(trans, root);
6527 nr = trans->blocks_used;
6528 btrfs_end_transaction(trans, root);
6529 btrfs_btree_balance_dirty(root, nr);
6531 /* Now start a transaction for the truncate */
6532 trans = btrfs_start_transaction(root, 0);
6534 return PTR_ERR(trans);
6535 btrfs_set_trans_block_group(trans, inode);
6536 trans->block_rsv = root->orphan_block_rsv;
6539 * setattr is responsible for setting the ordered_data_close flag,
6540 * but that is only tested during the last file release. That
6541 * could happen well after the next commit, leaving a great big
6542 * window where new writes may get lost if someone chooses to write
6543 * to this file after truncating to zero
6545 * The inode doesn't have any dirty data here, and so if we commit
6546 * this is a noop. If someone immediately starts writing to the inode
6547 * it is very likely we'll catch some of their writes in this
6548 * transaction, and the commit will find this file on the ordered
6549 * data list with good things to send down.
6551 * This is a best effort solution, there is still a window where
6552 * using truncate to replace the contents of the file will
6553 * end up with a zero length file after a crash.
6555 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6556 btrfs_add_ordered_operation(trans, root, inode);
6560 trans = btrfs_start_transaction(root, 0);
6562 return PTR_ERR(trans);
6563 btrfs_set_trans_block_group(trans, inode);
6564 trans->block_rsv = root->orphan_block_rsv;
6567 ret = btrfs_block_rsv_check(trans, root,
6568 root->orphan_block_rsv, 0, 5);
6569 if (ret == -EAGAIN) {
6570 ret = btrfs_commit_transaction(trans, root);
6580 ret = btrfs_truncate_inode_items(trans, root, inode,
6582 BTRFS_EXTENT_DATA_KEY);
6583 if (ret != -EAGAIN) {
6588 ret = btrfs_update_inode(trans, root, inode);
6594 nr = trans->blocks_used;
6595 btrfs_end_transaction(trans, root);
6597 btrfs_btree_balance_dirty(root, nr);
6600 if (ret == 0 && inode->i_nlink > 0) {
6601 ret = btrfs_orphan_del(trans, inode);
6604 } else if (ret && inode->i_nlink > 0) {
6606 * Failed to do the truncate, remove us from the in memory
6609 ret = btrfs_orphan_del(NULL, inode);
6612 ret = btrfs_update_inode(trans, root, inode);
6616 nr = trans->blocks_used;
6617 ret = btrfs_end_transaction_throttle(trans, root);
6620 btrfs_btree_balance_dirty(root, nr);
6626 * create a new subvolume directory/inode (helper for the ioctl).
6628 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6629 struct btrfs_root *new_root,
6630 u64 new_dirid, u64 alloc_hint)
6632 struct inode *inode;
6636 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6637 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6639 return PTR_ERR(inode);
6640 inode->i_op = &btrfs_dir_inode_operations;
6641 inode->i_fop = &btrfs_dir_file_operations;
6644 btrfs_i_size_write(inode, 0);
6646 err = btrfs_update_inode(trans, new_root, inode);
6653 /* helper function for file defrag and space balancing. This
6654 * forces readahead on a given range of bytes in an inode
6656 unsigned long btrfs_force_ra(struct address_space *mapping,
6657 struct file_ra_state *ra, struct file *file,
6658 pgoff_t offset, pgoff_t last_index)
6660 pgoff_t req_size = last_index - offset + 1;
6662 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6663 return offset + req_size;
6666 struct inode *btrfs_alloc_inode(struct super_block *sb)
6668 struct btrfs_inode *ei;
6669 struct inode *inode;
6671 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6676 ei->space_info = NULL;
6680 ei->last_sub_trans = 0;
6681 ei->logged_trans = 0;
6682 ei->delalloc_bytes = 0;
6683 ei->reserved_bytes = 0;
6684 ei->disk_i_size = 0;
6686 ei->index_cnt = (u64)-1;
6687 ei->last_unlink_trans = 0;
6689 atomic_set(&ei->outstanding_extents, 0);
6690 atomic_set(&ei->reserved_extents, 0);
6692 ei->ordered_data_close = 0;
6693 ei->orphan_meta_reserved = 0;
6694 ei->dummy_inode = 0;
6695 ei->force_compress = BTRFS_COMPRESS_NONE;
6697 inode = &ei->vfs_inode;
6698 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6699 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6700 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6701 mutex_init(&ei->log_mutex);
6702 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6703 INIT_LIST_HEAD(&ei->i_orphan);
6704 INIT_LIST_HEAD(&ei->delalloc_inodes);
6705 INIT_LIST_HEAD(&ei->ordered_operations);
6706 RB_CLEAR_NODE(&ei->rb_node);
6711 static void btrfs_i_callback(struct rcu_head *head)
6713 struct inode *inode = container_of(head, struct inode, i_rcu);
6714 INIT_LIST_HEAD(&inode->i_dentry);
6715 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6718 void btrfs_destroy_inode(struct inode *inode)
6720 struct btrfs_ordered_extent *ordered;
6721 struct btrfs_root *root = BTRFS_I(inode)->root;
6723 WARN_ON(!list_empty(&inode->i_dentry));
6724 WARN_ON(inode->i_data.nrpages);
6725 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6726 WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6729 * This can happen where we create an inode, but somebody else also
6730 * created the same inode and we need to destroy the one we already
6737 * Make sure we're properly removed from the ordered operation
6741 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6742 spin_lock(&root->fs_info->ordered_extent_lock);
6743 list_del_init(&BTRFS_I(inode)->ordered_operations);
6744 spin_unlock(&root->fs_info->ordered_extent_lock);
6747 if (root == root->fs_info->tree_root) {
6748 struct btrfs_block_group_cache *block_group;
6750 block_group = btrfs_lookup_block_group(root->fs_info,
6751 BTRFS_I(inode)->block_group);
6752 if (block_group && block_group->inode == inode) {
6753 spin_lock(&block_group->lock);
6754 block_group->inode = NULL;
6755 spin_unlock(&block_group->lock);
6756 btrfs_put_block_group(block_group);
6757 } else if (block_group) {
6758 btrfs_put_block_group(block_group);
6762 spin_lock(&root->orphan_lock);
6763 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6764 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6766 list_del_init(&BTRFS_I(inode)->i_orphan);
6768 spin_unlock(&root->orphan_lock);
6771 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6775 printk(KERN_ERR "btrfs found ordered "
6776 "extent %llu %llu on inode cleanup\n",
6777 (unsigned long long)ordered->file_offset,
6778 (unsigned long long)ordered->len);
6779 btrfs_remove_ordered_extent(inode, ordered);
6780 btrfs_put_ordered_extent(ordered);
6781 btrfs_put_ordered_extent(ordered);
6784 inode_tree_del(inode);
6785 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6787 call_rcu(&inode->i_rcu, btrfs_i_callback);
6790 int btrfs_drop_inode(struct inode *inode)
6792 struct btrfs_root *root = BTRFS_I(inode)->root;
6794 if (btrfs_root_refs(&root->root_item) == 0 &&
6795 root != root->fs_info->tree_root)
6798 return generic_drop_inode(inode);
6801 static void init_once(void *foo)
6803 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6805 inode_init_once(&ei->vfs_inode);
6808 void btrfs_destroy_cachep(void)
6810 if (btrfs_inode_cachep)
6811 kmem_cache_destroy(btrfs_inode_cachep);
6812 if (btrfs_trans_handle_cachep)
6813 kmem_cache_destroy(btrfs_trans_handle_cachep);
6814 if (btrfs_transaction_cachep)
6815 kmem_cache_destroy(btrfs_transaction_cachep);
6816 if (btrfs_path_cachep)
6817 kmem_cache_destroy(btrfs_path_cachep);
6818 if (btrfs_free_space_cachep)
6819 kmem_cache_destroy(btrfs_free_space_cachep);
6822 int btrfs_init_cachep(void)
6824 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6825 sizeof(struct btrfs_inode), 0,
6826 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6827 if (!btrfs_inode_cachep)
6830 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6831 sizeof(struct btrfs_trans_handle), 0,
6832 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6833 if (!btrfs_trans_handle_cachep)
6836 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6837 sizeof(struct btrfs_transaction), 0,
6838 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6839 if (!btrfs_transaction_cachep)
6842 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6843 sizeof(struct btrfs_path), 0,
6844 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6845 if (!btrfs_path_cachep)
6848 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6849 sizeof(struct btrfs_free_space), 0,
6850 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6851 if (!btrfs_free_space_cachep)
6856 btrfs_destroy_cachep();
6860 static int btrfs_getattr(struct vfsmount *mnt,
6861 struct dentry *dentry, struct kstat *stat)
6863 struct inode *inode = dentry->d_inode;
6864 generic_fillattr(inode, stat);
6865 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6866 stat->blksize = PAGE_CACHE_SIZE;
6867 stat->blocks = (inode_get_bytes(inode) +
6868 BTRFS_I(inode)->delalloc_bytes) >> 9;
6873 * If a file is moved, it will inherit the cow and compression flags of the new
6876 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6878 struct btrfs_inode *b_dir = BTRFS_I(dir);
6879 struct btrfs_inode *b_inode = BTRFS_I(inode);
6881 if (b_dir->flags & BTRFS_INODE_NODATACOW)
6882 b_inode->flags |= BTRFS_INODE_NODATACOW;
6884 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6886 if (b_dir->flags & BTRFS_INODE_COMPRESS)
6887 b_inode->flags |= BTRFS_INODE_COMPRESS;
6889 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6892 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6893 struct inode *new_dir, struct dentry *new_dentry)
6895 struct btrfs_trans_handle *trans;
6896 struct btrfs_root *root = BTRFS_I(old_dir)->root;
6897 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6898 struct inode *new_inode = new_dentry->d_inode;
6899 struct inode *old_inode = old_dentry->d_inode;
6900 struct timespec ctime = CURRENT_TIME;
6905 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6908 /* we only allow rename subvolume link between subvolumes */
6909 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6912 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6913 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6916 if (S_ISDIR(old_inode->i_mode) && new_inode &&
6917 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6920 * we're using rename to replace one file with another.
6921 * and the replacement file is large. Start IO on it now so
6922 * we don't add too much work to the end of the transaction
6924 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6925 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6926 filemap_flush(old_inode->i_mapping);
6928 /* close the racy window with snapshot create/destroy ioctl */
6929 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6930 down_read(&root->fs_info->subvol_sem);
6932 * We want to reserve the absolute worst case amount of items. So if
6933 * both inodes are subvols and we need to unlink them then that would
6934 * require 4 item modifications, but if they are both normal inodes it
6935 * would require 5 item modifications, so we'll assume their normal
6936 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6937 * should cover the worst case number of items we'll modify.
6939 trans = btrfs_start_transaction(root, 20);
6941 return PTR_ERR(trans);
6943 btrfs_set_trans_block_group(trans, new_dir);
6946 btrfs_record_root_in_trans(trans, dest);
6948 ret = btrfs_set_inode_index(new_dir, &index);
6952 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6953 /* force full log commit if subvolume involved. */
6954 root->fs_info->last_trans_log_full_commit = trans->transid;
6956 ret = btrfs_insert_inode_ref(trans, dest,
6957 new_dentry->d_name.name,
6958 new_dentry->d_name.len,
6960 new_dir->i_ino, index);
6964 * this is an ugly little race, but the rename is required
6965 * to make sure that if we crash, the inode is either at the
6966 * old name or the new one. pinning the log transaction lets
6967 * us make sure we don't allow a log commit to come in after
6968 * we unlink the name but before we add the new name back in.
6970 btrfs_pin_log_trans(root);
6973 * make sure the inode gets flushed if it is replacing
6976 if (new_inode && new_inode->i_size &&
6977 old_inode && S_ISREG(old_inode->i_mode)) {
6978 btrfs_add_ordered_operation(trans, root, old_inode);
6981 old_dir->i_ctime = old_dir->i_mtime = ctime;
6982 new_dir->i_ctime = new_dir->i_mtime = ctime;
6983 old_inode->i_ctime = ctime;
6985 if (old_dentry->d_parent != new_dentry->d_parent)
6986 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6988 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6989 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6990 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6991 old_dentry->d_name.name,
6992 old_dentry->d_name.len);
6994 btrfs_inc_nlink(old_dentry->d_inode);
6995 ret = btrfs_unlink_inode(trans, root, old_dir,
6996 old_dentry->d_inode,
6997 old_dentry->d_name.name,
6998 old_dentry->d_name.len);
7003 new_inode->i_ctime = CURRENT_TIME;
7004 if (unlikely(new_inode->i_ino ==
7005 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7006 root_objectid = BTRFS_I(new_inode)->location.objectid;
7007 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7009 new_dentry->d_name.name,
7010 new_dentry->d_name.len);
7011 BUG_ON(new_inode->i_nlink == 0);
7013 ret = btrfs_unlink_inode(trans, dest, new_dir,
7014 new_dentry->d_inode,
7015 new_dentry->d_name.name,
7016 new_dentry->d_name.len);
7019 if (new_inode->i_nlink == 0) {
7020 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7025 fixup_inode_flags(new_dir, old_inode);
7027 ret = btrfs_add_link(trans, new_dir, old_inode,
7028 new_dentry->d_name.name,
7029 new_dentry->d_name.len, 0, index);
7032 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7033 struct dentry *parent = dget_parent(new_dentry);
7034 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7036 btrfs_end_log_trans(root);
7039 btrfs_end_transaction_throttle(trans, root);
7041 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7042 up_read(&root->fs_info->subvol_sem);
7048 * some fairly slow code that needs optimization. This walks the list
7049 * of all the inodes with pending delalloc and forces them to disk.
7051 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7053 struct list_head *head = &root->fs_info->delalloc_inodes;
7054 struct btrfs_inode *binode;
7055 struct inode *inode;
7057 if (root->fs_info->sb->s_flags & MS_RDONLY)
7060 spin_lock(&root->fs_info->delalloc_lock);
7061 while (!list_empty(head)) {
7062 binode = list_entry(head->next, struct btrfs_inode,
7064 inode = igrab(&binode->vfs_inode);
7066 list_del_init(&binode->delalloc_inodes);
7067 spin_unlock(&root->fs_info->delalloc_lock);
7069 filemap_flush(inode->i_mapping);
7071 btrfs_add_delayed_iput(inode);
7076 spin_lock(&root->fs_info->delalloc_lock);
7078 spin_unlock(&root->fs_info->delalloc_lock);
7080 /* the filemap_flush will queue IO into the worker threads, but
7081 * we have to make sure the IO is actually started and that
7082 * ordered extents get created before we return
7084 atomic_inc(&root->fs_info->async_submit_draining);
7085 while (atomic_read(&root->fs_info->nr_async_submits) ||
7086 atomic_read(&root->fs_info->async_delalloc_pages)) {
7087 wait_event(root->fs_info->async_submit_wait,
7088 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7089 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7091 atomic_dec(&root->fs_info->async_submit_draining);
7095 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7098 struct btrfs_inode *binode;
7099 struct inode *inode = NULL;
7101 spin_lock(&root->fs_info->delalloc_lock);
7102 while (!list_empty(&root->fs_info->delalloc_inodes)) {
7103 binode = list_entry(root->fs_info->delalloc_inodes.next,
7104 struct btrfs_inode, delalloc_inodes);
7105 inode = igrab(&binode->vfs_inode);
7107 list_move_tail(&binode->delalloc_inodes,
7108 &root->fs_info->delalloc_inodes);
7112 list_del_init(&binode->delalloc_inodes);
7113 cond_resched_lock(&root->fs_info->delalloc_lock);
7115 spin_unlock(&root->fs_info->delalloc_lock);
7119 filemap_write_and_wait(inode->i_mapping);
7121 * We have to do this because compression doesn't
7122 * actually set PG_writeback until it submits the pages
7123 * for IO, which happens in an async thread, so we could
7124 * race and not actually wait for any writeback pages
7125 * because they've not been submitted yet. Technically
7126 * this could still be the case for the ordered stuff
7127 * since the async thread may not have started to do its
7128 * work yet. If this becomes the case then we need to
7129 * figure out a way to make sure that in writepage we
7130 * wait for any async pages to be submitted before
7131 * returning so that fdatawait does what its supposed to
7134 btrfs_wait_ordered_range(inode, 0, (u64)-1);
7136 filemap_flush(inode->i_mapping);
7139 btrfs_add_delayed_iput(inode);
7147 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7148 const char *symname)
7150 struct btrfs_trans_handle *trans;
7151 struct btrfs_root *root = BTRFS_I(dir)->root;
7152 struct btrfs_path *path;
7153 struct btrfs_key key;
7154 struct inode *inode = NULL;
7162 struct btrfs_file_extent_item *ei;
7163 struct extent_buffer *leaf;
7164 unsigned long nr = 0;
7166 name_len = strlen(symname) + 1;
7167 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7168 return -ENAMETOOLONG;
7170 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7174 * 2 items for inode item and ref
7175 * 2 items for dir items
7176 * 1 item for xattr if selinux is on
7178 trans = btrfs_start_transaction(root, 5);
7180 return PTR_ERR(trans);
7182 btrfs_set_trans_block_group(trans, dir);
7184 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7185 dentry->d_name.len, dir->i_ino, objectid,
7186 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7188 err = PTR_ERR(inode);
7192 err = btrfs_init_inode_security(trans, inode, dir);
7198 btrfs_set_trans_block_group(trans, inode);
7199 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7203 inode->i_mapping->a_ops = &btrfs_aops;
7204 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7205 inode->i_fop = &btrfs_file_operations;
7206 inode->i_op = &btrfs_file_inode_operations;
7207 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7209 btrfs_update_inode_block_group(trans, inode);
7210 btrfs_update_inode_block_group(trans, dir);
7214 path = btrfs_alloc_path();
7216 key.objectid = inode->i_ino;
7218 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7219 datasize = btrfs_file_extent_calc_inline_size(name_len);
7220 err = btrfs_insert_empty_item(trans, root, path, &key,
7226 leaf = path->nodes[0];
7227 ei = btrfs_item_ptr(leaf, path->slots[0],
7228 struct btrfs_file_extent_item);
7229 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7230 btrfs_set_file_extent_type(leaf, ei,
7231 BTRFS_FILE_EXTENT_INLINE);
7232 btrfs_set_file_extent_encryption(leaf, ei, 0);
7233 btrfs_set_file_extent_compression(leaf, ei, 0);
7234 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7235 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7237 ptr = btrfs_file_extent_inline_start(ei);
7238 write_extent_buffer(leaf, symname, ptr, name_len);
7239 btrfs_mark_buffer_dirty(leaf);
7240 btrfs_free_path(path);
7242 inode->i_op = &btrfs_symlink_inode_operations;
7243 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7244 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7245 inode_set_bytes(inode, name_len);
7246 btrfs_i_size_write(inode, name_len - 1);
7247 err = btrfs_update_inode(trans, root, inode);
7252 nr = trans->blocks_used;
7253 btrfs_end_transaction_throttle(trans, root);
7255 inode_dec_link_count(inode);
7258 btrfs_btree_balance_dirty(root, nr);
7262 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7263 u64 start, u64 num_bytes, u64 min_size,
7264 loff_t actual_len, u64 *alloc_hint,
7265 struct btrfs_trans_handle *trans)
7267 struct btrfs_root *root = BTRFS_I(inode)->root;
7268 struct btrfs_key ins;
7269 u64 cur_offset = start;
7272 bool own_trans = true;
7276 while (num_bytes > 0) {
7278 trans = btrfs_start_transaction(root, 3);
7279 if (IS_ERR(trans)) {
7280 ret = PTR_ERR(trans);
7285 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7286 0, *alloc_hint, (u64)-1, &ins, 1);
7289 btrfs_end_transaction(trans, root);
7293 ret = insert_reserved_file_extent(trans, inode,
7294 cur_offset, ins.objectid,
7295 ins.offset, ins.offset,
7296 ins.offset, 0, 0, 0,
7297 BTRFS_FILE_EXTENT_PREALLOC);
7299 btrfs_drop_extent_cache(inode, cur_offset,
7300 cur_offset + ins.offset -1, 0);
7302 num_bytes -= ins.offset;
7303 cur_offset += ins.offset;
7304 *alloc_hint = ins.objectid + ins.offset;
7306 inode->i_ctime = CURRENT_TIME;
7307 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7308 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7309 (actual_len > inode->i_size) &&
7310 (cur_offset > inode->i_size)) {
7311 if (cur_offset > actual_len)
7312 i_size = actual_len;
7314 i_size = cur_offset;
7315 i_size_write(inode, i_size);
7316 btrfs_ordered_update_i_size(inode, i_size, NULL);
7319 ret = btrfs_update_inode(trans, root, inode);
7323 btrfs_end_transaction(trans, root);
7328 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7329 u64 start, u64 num_bytes, u64 min_size,
7330 loff_t actual_len, u64 *alloc_hint)
7332 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7333 min_size, actual_len, alloc_hint,
7337 int btrfs_prealloc_file_range_trans(struct inode *inode,
7338 struct btrfs_trans_handle *trans, int mode,
7339 u64 start, u64 num_bytes, u64 min_size,
7340 loff_t actual_len, u64 *alloc_hint)
7342 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7343 min_size, actual_len, alloc_hint, trans);
7346 static int btrfs_set_page_dirty(struct page *page)
7348 return __set_page_dirty_nobuffers(page);
7351 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7353 struct btrfs_root *root = BTRFS_I(inode)->root;
7355 if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7357 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7359 return generic_permission(inode, mask, flags, btrfs_check_acl);
7362 static const struct inode_operations btrfs_dir_inode_operations = {
7363 .getattr = btrfs_getattr,
7364 .lookup = btrfs_lookup,
7365 .create = btrfs_create,
7366 .unlink = btrfs_unlink,
7368 .mkdir = btrfs_mkdir,
7369 .rmdir = btrfs_rmdir,
7370 .rename = btrfs_rename,
7371 .symlink = btrfs_symlink,
7372 .setattr = btrfs_setattr,
7373 .mknod = btrfs_mknod,
7374 .setxattr = btrfs_setxattr,
7375 .getxattr = btrfs_getxattr,
7376 .listxattr = btrfs_listxattr,
7377 .removexattr = btrfs_removexattr,
7378 .permission = btrfs_permission,
7380 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7381 .lookup = btrfs_lookup,
7382 .permission = btrfs_permission,
7385 static const struct file_operations btrfs_dir_file_operations = {
7386 .llseek = generic_file_llseek,
7387 .read = generic_read_dir,
7388 .readdir = btrfs_real_readdir,
7389 .unlocked_ioctl = btrfs_ioctl,
7390 #ifdef CONFIG_COMPAT
7391 .compat_ioctl = btrfs_ioctl,
7393 .release = btrfs_release_file,
7394 .fsync = btrfs_sync_file,
7397 static struct extent_io_ops btrfs_extent_io_ops = {
7398 .fill_delalloc = run_delalloc_range,
7399 .submit_bio_hook = btrfs_submit_bio_hook,
7400 .merge_bio_hook = btrfs_merge_bio_hook,
7401 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7402 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7403 .writepage_start_hook = btrfs_writepage_start_hook,
7404 .readpage_io_failed_hook = btrfs_io_failed_hook,
7405 .set_bit_hook = btrfs_set_bit_hook,
7406 .clear_bit_hook = btrfs_clear_bit_hook,
7407 .merge_extent_hook = btrfs_merge_extent_hook,
7408 .split_extent_hook = btrfs_split_extent_hook,
7412 * btrfs doesn't support the bmap operation because swapfiles
7413 * use bmap to make a mapping of extents in the file. They assume
7414 * these extents won't change over the life of the file and they
7415 * use the bmap result to do IO directly to the drive.
7417 * the btrfs bmap call would return logical addresses that aren't
7418 * suitable for IO and they also will change frequently as COW
7419 * operations happen. So, swapfile + btrfs == corruption.
7421 * For now we're avoiding this by dropping bmap.
7423 static const struct address_space_operations btrfs_aops = {
7424 .readpage = btrfs_readpage,
7425 .writepage = btrfs_writepage,
7426 .writepages = btrfs_writepages,
7427 .readpages = btrfs_readpages,
7428 .sync_page = block_sync_page,
7429 .direct_IO = btrfs_direct_IO,
7430 .invalidatepage = btrfs_invalidatepage,
7431 .releasepage = btrfs_releasepage,
7432 .set_page_dirty = btrfs_set_page_dirty,
7433 .error_remove_page = generic_error_remove_page,
7436 static const struct address_space_operations btrfs_symlink_aops = {
7437 .readpage = btrfs_readpage,
7438 .writepage = btrfs_writepage,
7439 .invalidatepage = btrfs_invalidatepage,
7440 .releasepage = btrfs_releasepage,
7443 static const struct inode_operations btrfs_file_inode_operations = {
7444 .getattr = btrfs_getattr,
7445 .setattr = btrfs_setattr,
7446 .setxattr = btrfs_setxattr,
7447 .getxattr = btrfs_getxattr,
7448 .listxattr = btrfs_listxattr,
7449 .removexattr = btrfs_removexattr,
7450 .permission = btrfs_permission,
7451 .fiemap = btrfs_fiemap,
7453 static const struct inode_operations btrfs_special_inode_operations = {
7454 .getattr = btrfs_getattr,
7455 .setattr = btrfs_setattr,
7456 .permission = btrfs_permission,
7457 .setxattr = btrfs_setxattr,
7458 .getxattr = btrfs_getxattr,
7459 .listxattr = btrfs_listxattr,
7460 .removexattr = btrfs_removexattr,
7462 static const struct inode_operations btrfs_symlink_inode_operations = {
7463 .readlink = generic_readlink,
7464 .follow_link = page_follow_link_light,
7465 .put_link = page_put_link,
7466 .getattr = btrfs_getattr,
7467 .permission = btrfs_permission,
7468 .setxattr = btrfs_setxattr,
7469 .getxattr = btrfs_getxattr,
7470 .listxattr = btrfs_listxattr,
7471 .removexattr = btrfs_removexattr,
7474 const struct dentry_operations btrfs_dentry_operations = {
7475 .d_delete = btrfs_dentry_delete,