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>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
45 #include "transaction.h"
46 #include "btrfs_inode.h"
48 #include "print-tree.h"
49 #include "ordered-data.h"
53 #include "compression.h"
55 #include "free-space-cache.h"
56 #include "inode-map.h"
58 struct btrfs_iget_args {
60 struct btrfs_root *root;
63 static const struct inode_operations btrfs_dir_inode_operations;
64 static const struct inode_operations btrfs_symlink_inode_operations;
65 static const struct inode_operations btrfs_dir_ro_inode_operations;
66 static const struct inode_operations btrfs_special_inode_operations;
67 static const struct inode_operations btrfs_file_inode_operations;
68 static const struct address_space_operations btrfs_aops;
69 static const struct address_space_operations btrfs_symlink_aops;
70 static const struct file_operations btrfs_dir_file_operations;
71 static struct extent_io_ops btrfs_extent_io_ops;
73 static struct kmem_cache *btrfs_inode_cachep;
74 struct kmem_cache *btrfs_trans_handle_cachep;
75 struct kmem_cache *btrfs_transaction_cachep;
76 struct kmem_cache *btrfs_path_cachep;
77 struct kmem_cache *btrfs_free_space_cachep;
80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
81 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
82 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
83 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
84 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
85 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
86 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
87 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
91 static int btrfs_truncate(struct inode *inode);
92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
93 static noinline int cow_file_range(struct inode *inode,
94 struct page *locked_page,
95 u64 start, u64 end, int *page_started,
96 unsigned long *nr_written, int unlock);
97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
98 struct btrfs_root *root, struct inode *inode);
100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
101 struct inode *inode, struct inode *dir,
102 const struct qstr *qstr)
106 err = btrfs_init_acl(trans, inode, dir);
108 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
113 * this does all the hard work for inserting an inline extent into
114 * the btree. The caller should have done a btrfs_drop_extents so that
115 * no overlapping inline items exist in the btree
117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
118 struct btrfs_root *root, struct inode *inode,
119 u64 start, size_t size, size_t compressed_size,
121 struct page **compressed_pages)
123 struct btrfs_key key;
124 struct btrfs_path *path;
125 struct extent_buffer *leaf;
126 struct page *page = NULL;
129 struct btrfs_file_extent_item *ei;
132 size_t cur_size = size;
134 unsigned long offset;
136 if (compressed_size && compressed_pages)
137 cur_size = compressed_size;
139 path = btrfs_alloc_path();
143 path->leave_spinning = 1;
145 key.objectid = btrfs_ino(inode);
147 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
148 datasize = btrfs_file_extent_calc_inline_size(cur_size);
150 inode_add_bytes(inode, size);
151 ret = btrfs_insert_empty_item(trans, root, path, &key,
157 leaf = path->nodes[0];
158 ei = btrfs_item_ptr(leaf, path->slots[0],
159 struct btrfs_file_extent_item);
160 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
161 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
162 btrfs_set_file_extent_encryption(leaf, ei, 0);
163 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
164 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
165 ptr = btrfs_file_extent_inline_start(ei);
167 if (compress_type != BTRFS_COMPRESS_NONE) {
170 while (compressed_size > 0) {
171 cpage = compressed_pages[i];
172 cur_size = min_t(unsigned long, compressed_size,
175 kaddr = kmap_atomic(cpage);
176 write_extent_buffer(leaf, kaddr, ptr, cur_size);
177 kunmap_atomic(kaddr);
181 compressed_size -= cur_size;
183 btrfs_set_file_extent_compression(leaf, ei,
186 page = find_get_page(inode->i_mapping,
187 start >> PAGE_CACHE_SHIFT);
188 btrfs_set_file_extent_compression(leaf, ei, 0);
189 kaddr = kmap_atomic(page);
190 offset = start & (PAGE_CACHE_SIZE - 1);
191 write_extent_buffer(leaf, kaddr + offset, ptr, size);
192 kunmap_atomic(kaddr);
193 page_cache_release(page);
195 btrfs_mark_buffer_dirty(leaf);
196 btrfs_free_path(path);
199 * we're an inline extent, so nobody can
200 * extend the file past i_size without locking
201 * a page we already have locked.
203 * We must do any isize and inode updates
204 * before we unlock the pages. Otherwise we
205 * could end up racing with unlink.
207 BTRFS_I(inode)->disk_i_size = inode->i_size;
208 ret = btrfs_update_inode(trans, root, inode);
212 btrfs_free_path(path);
218 * conditionally insert an inline extent into the file. This
219 * does the checks required to make sure the data is small enough
220 * to fit as an inline extent.
222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
223 struct btrfs_root *root,
224 struct inode *inode, u64 start, u64 end,
225 size_t compressed_size, int compress_type,
226 struct page **compressed_pages)
228 u64 isize = i_size_read(inode);
229 u64 actual_end = min(end + 1, isize);
230 u64 inline_len = actual_end - start;
231 u64 aligned_end = (end + root->sectorsize - 1) &
232 ~((u64)root->sectorsize - 1);
234 u64 data_len = inline_len;
238 data_len = compressed_size;
241 actual_end >= PAGE_CACHE_SIZE ||
242 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
244 (actual_end & (root->sectorsize - 1)) == 0) ||
246 data_len > root->fs_info->max_inline) {
250 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
255 if (isize > actual_end)
256 inline_len = min_t(u64, isize, actual_end);
257 ret = insert_inline_extent(trans, root, inode, start,
258 inline_len, compressed_size,
259 compress_type, compressed_pages);
260 if (ret && ret != -ENOSPC) {
261 btrfs_abort_transaction(trans, root, ret);
263 } else if (ret == -ENOSPC) {
267 btrfs_delalloc_release_metadata(inode, end + 1 - start);
268 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
272 struct async_extent {
277 unsigned long nr_pages;
279 struct list_head list;
284 struct btrfs_root *root;
285 struct page *locked_page;
288 struct list_head extents;
289 struct btrfs_work work;
292 static noinline int add_async_extent(struct async_cow *cow,
293 u64 start, u64 ram_size,
296 unsigned long nr_pages,
299 struct async_extent *async_extent;
301 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
302 BUG_ON(!async_extent); /* -ENOMEM */
303 async_extent->start = start;
304 async_extent->ram_size = ram_size;
305 async_extent->compressed_size = compressed_size;
306 async_extent->pages = pages;
307 async_extent->nr_pages = nr_pages;
308 async_extent->compress_type = compress_type;
309 list_add_tail(&async_extent->list, &cow->extents);
314 * we create compressed extents in two phases. The first
315 * phase compresses a range of pages that have already been
316 * locked (both pages and state bits are locked).
318 * This is done inside an ordered work queue, and the compression
319 * is spread across many cpus. The actual IO submission is step
320 * two, and the ordered work queue takes care of making sure that
321 * happens in the same order things were put onto the queue by
322 * writepages and friends.
324 * If this code finds it can't get good compression, it puts an
325 * entry onto the work queue to write the uncompressed bytes. This
326 * makes sure that both compressed inodes and uncompressed inodes
327 * are written in the same order that pdflush sent them down.
329 static noinline int compress_file_range(struct inode *inode,
330 struct page *locked_page,
332 struct async_cow *async_cow,
335 struct btrfs_root *root = BTRFS_I(inode)->root;
336 struct btrfs_trans_handle *trans;
338 u64 blocksize = root->sectorsize;
340 u64 isize = i_size_read(inode);
342 struct page **pages = NULL;
343 unsigned long nr_pages;
344 unsigned long nr_pages_ret = 0;
345 unsigned long total_compressed = 0;
346 unsigned long total_in = 0;
347 unsigned long max_compressed = 128 * 1024;
348 unsigned long max_uncompressed = 128 * 1024;
351 int compress_type = root->fs_info->compress_type;
353 /* if this is a small write inside eof, kick off a defrag */
354 if ((end - start + 1) < 16 * 1024 &&
355 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
356 btrfs_add_inode_defrag(NULL, inode);
358 actual_end = min_t(u64, isize, end + 1);
361 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
362 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
365 * we don't want to send crud past the end of i_size through
366 * compression, that's just a waste of CPU time. So, if the
367 * end of the file is before the start of our current
368 * requested range of bytes, we bail out to the uncompressed
369 * cleanup code that can deal with all of this.
371 * It isn't really the fastest way to fix things, but this is a
372 * very uncommon corner.
374 if (actual_end <= start)
375 goto cleanup_and_bail_uncompressed;
377 total_compressed = actual_end - start;
379 /* we want to make sure that amount of ram required to uncompress
380 * an extent is reasonable, so we limit the total size in ram
381 * of a compressed extent to 128k. This is a crucial number
382 * because it also controls how easily we can spread reads across
383 * cpus for decompression.
385 * We also want to make sure the amount of IO required to do
386 * a random read is reasonably small, so we limit the size of
387 * a compressed extent to 128k.
389 total_compressed = min(total_compressed, max_uncompressed);
390 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
391 num_bytes = max(blocksize, num_bytes);
396 * we do compression for mount -o compress and when the
397 * inode has not been flagged as nocompress. This flag can
398 * change at any time if we discover bad compression ratios.
400 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
401 (btrfs_test_opt(root, COMPRESS) ||
402 (BTRFS_I(inode)->force_compress) ||
403 (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
405 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
407 /* just bail out to the uncompressed code */
411 if (BTRFS_I(inode)->force_compress)
412 compress_type = BTRFS_I(inode)->force_compress;
414 ret = btrfs_compress_pages(compress_type,
415 inode->i_mapping, start,
416 total_compressed, pages,
417 nr_pages, &nr_pages_ret,
423 unsigned long offset = total_compressed &
424 (PAGE_CACHE_SIZE - 1);
425 struct page *page = pages[nr_pages_ret - 1];
428 /* zero the tail end of the last page, we might be
429 * sending it down to disk
432 kaddr = kmap_atomic(page);
433 memset(kaddr + offset, 0,
434 PAGE_CACHE_SIZE - offset);
435 kunmap_atomic(kaddr);
442 trans = btrfs_join_transaction(root);
444 ret = PTR_ERR(trans);
446 goto cleanup_and_out;
448 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
450 /* lets try to make an inline extent */
451 if (ret || total_in < (actual_end - start)) {
452 /* we didn't compress the entire range, try
453 * to make an uncompressed inline extent.
455 ret = cow_file_range_inline(trans, root, inode,
456 start, end, 0, 0, NULL);
458 /* try making a compressed inline extent */
459 ret = cow_file_range_inline(trans, root, inode,
462 compress_type, pages);
466 * inline extent creation worked or returned error,
467 * we don't need to create any more async work items.
468 * Unlock and free up our temp pages.
470 extent_clear_unlock_delalloc(inode,
471 &BTRFS_I(inode)->io_tree,
473 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
474 EXTENT_CLEAR_DELALLOC |
475 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
477 btrfs_end_transaction(trans, root);
480 btrfs_end_transaction(trans, root);
485 * we aren't doing an inline extent round the compressed size
486 * up to a block size boundary so the allocator does sane
489 total_compressed = (total_compressed + blocksize - 1) &
493 * one last check to make sure the compression is really a
494 * win, compare the page count read with the blocks on disk
496 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
497 ~(PAGE_CACHE_SIZE - 1);
498 if (total_compressed >= total_in) {
501 num_bytes = total_in;
504 if (!will_compress && pages) {
506 * the compression code ran but failed to make things smaller,
507 * free any pages it allocated and our page pointer array
509 for (i = 0; i < nr_pages_ret; i++) {
510 WARN_ON(pages[i]->mapping);
511 page_cache_release(pages[i]);
515 total_compressed = 0;
518 /* flag the file so we don't compress in the future */
519 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
520 !(BTRFS_I(inode)->force_compress)) {
521 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
527 /* the async work queues will take care of doing actual
528 * allocation on disk for these compressed pages,
529 * and will submit them to the elevator.
531 add_async_extent(async_cow, start, num_bytes,
532 total_compressed, pages, nr_pages_ret,
535 if (start + num_bytes < end) {
542 cleanup_and_bail_uncompressed:
544 * No compression, but we still need to write the pages in
545 * the file we've been given so far. redirty the locked
546 * page if it corresponds to our extent and set things up
547 * for the async work queue to run cow_file_range to do
548 * the normal delalloc dance
550 if (page_offset(locked_page) >= start &&
551 page_offset(locked_page) <= end) {
552 __set_page_dirty_nobuffers(locked_page);
553 /* unlocked later on in the async handlers */
555 add_async_extent(async_cow, start, end - start + 1,
556 0, NULL, 0, BTRFS_COMPRESS_NONE);
564 for (i = 0; i < nr_pages_ret; i++) {
565 WARN_ON(pages[i]->mapping);
566 page_cache_release(pages[i]);
573 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
575 EXTENT_CLEAR_UNLOCK_PAGE |
577 EXTENT_CLEAR_DELALLOC |
578 EXTENT_SET_WRITEBACK |
579 EXTENT_END_WRITEBACK);
580 if (!trans || IS_ERR(trans))
581 btrfs_error(root->fs_info, ret, "Failed to join transaction");
583 btrfs_abort_transaction(trans, root, ret);
588 * phase two of compressed writeback. This is the ordered portion
589 * of the code, which only gets called in the order the work was
590 * queued. We walk all the async extents created by compress_file_range
591 * and send them down to the disk.
593 static noinline int submit_compressed_extents(struct inode *inode,
594 struct async_cow *async_cow)
596 struct async_extent *async_extent;
598 struct btrfs_trans_handle *trans;
599 struct btrfs_key ins;
600 struct extent_map *em;
601 struct btrfs_root *root = BTRFS_I(inode)->root;
602 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
603 struct extent_io_tree *io_tree;
606 if (list_empty(&async_cow->extents))
610 while (!list_empty(&async_cow->extents)) {
611 async_extent = list_entry(async_cow->extents.next,
612 struct async_extent, list);
613 list_del(&async_extent->list);
615 io_tree = &BTRFS_I(inode)->io_tree;
618 /* did the compression code fall back to uncompressed IO? */
619 if (!async_extent->pages) {
620 int page_started = 0;
621 unsigned long nr_written = 0;
623 lock_extent(io_tree, async_extent->start,
624 async_extent->start +
625 async_extent->ram_size - 1);
627 /* allocate blocks */
628 ret = cow_file_range(inode, async_cow->locked_page,
630 async_extent->start +
631 async_extent->ram_size - 1,
632 &page_started, &nr_written, 0);
637 * if page_started, cow_file_range inserted an
638 * inline extent and took care of all the unlocking
639 * and IO for us. Otherwise, we need to submit
640 * all those pages down to the drive.
642 if (!page_started && !ret)
643 extent_write_locked_range(io_tree,
644 inode, async_extent->start,
645 async_extent->start +
646 async_extent->ram_size - 1,
654 lock_extent(io_tree, async_extent->start,
655 async_extent->start + async_extent->ram_size - 1);
657 trans = btrfs_join_transaction(root);
659 ret = PTR_ERR(trans);
661 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
662 ret = btrfs_reserve_extent(trans, root,
663 async_extent->compressed_size,
664 async_extent->compressed_size,
665 0, alloc_hint, &ins, 1);
667 btrfs_abort_transaction(trans, root, ret);
668 btrfs_end_transaction(trans, root);
673 for (i = 0; i < async_extent->nr_pages; i++) {
674 WARN_ON(async_extent->pages[i]->mapping);
675 page_cache_release(async_extent->pages[i]);
677 kfree(async_extent->pages);
678 async_extent->nr_pages = 0;
679 async_extent->pages = NULL;
680 unlock_extent(io_tree, async_extent->start,
681 async_extent->start +
682 async_extent->ram_size - 1);
685 goto out_free; /* JDM: Requeue? */
689 * here we're doing allocation and writeback of the
692 btrfs_drop_extent_cache(inode, async_extent->start,
693 async_extent->start +
694 async_extent->ram_size - 1, 0);
696 em = alloc_extent_map();
697 BUG_ON(!em); /* -ENOMEM */
698 em->start = async_extent->start;
699 em->len = async_extent->ram_size;
700 em->orig_start = em->start;
702 em->block_start = ins.objectid;
703 em->block_len = ins.offset;
704 em->bdev = root->fs_info->fs_devices->latest_bdev;
705 em->compress_type = async_extent->compress_type;
706 set_bit(EXTENT_FLAG_PINNED, &em->flags);
707 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
710 write_lock(&em_tree->lock);
711 ret = add_extent_mapping(em_tree, em);
712 write_unlock(&em_tree->lock);
713 if (ret != -EEXIST) {
717 btrfs_drop_extent_cache(inode, async_extent->start,
718 async_extent->start +
719 async_extent->ram_size - 1, 0);
722 ret = btrfs_add_ordered_extent_compress(inode,
725 async_extent->ram_size,
727 BTRFS_ORDERED_COMPRESSED,
728 async_extent->compress_type);
729 BUG_ON(ret); /* -ENOMEM */
732 * clear dirty, set writeback and unlock the pages.
734 extent_clear_unlock_delalloc(inode,
735 &BTRFS_I(inode)->io_tree,
737 async_extent->start +
738 async_extent->ram_size - 1,
739 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
740 EXTENT_CLEAR_UNLOCK |
741 EXTENT_CLEAR_DELALLOC |
742 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
744 ret = btrfs_submit_compressed_write(inode,
746 async_extent->ram_size,
748 ins.offset, async_extent->pages,
749 async_extent->nr_pages);
751 BUG_ON(ret); /* -ENOMEM */
752 alloc_hint = ins.objectid + ins.offset;
764 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
767 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
768 struct extent_map *em;
771 read_lock(&em_tree->lock);
772 em = search_extent_mapping(em_tree, start, num_bytes);
775 * if block start isn't an actual block number then find the
776 * first block in this inode and use that as a hint. If that
777 * block is also bogus then just don't worry about it.
779 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
781 em = search_extent_mapping(em_tree, 0, 0);
782 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
783 alloc_hint = em->block_start;
787 alloc_hint = em->block_start;
791 read_unlock(&em_tree->lock);
797 * when extent_io.c finds a delayed allocation range in the file,
798 * the call backs end up in this code. The basic idea is to
799 * allocate extents on disk for the range, and create ordered data structs
800 * in ram to track those extents.
802 * locked_page is the page that writepage had locked already. We use
803 * it to make sure we don't do extra locks or unlocks.
805 * *page_started is set to one if we unlock locked_page and do everything
806 * required to start IO on it. It may be clean and already done with
809 static noinline int cow_file_range(struct inode *inode,
810 struct page *locked_page,
811 u64 start, u64 end, int *page_started,
812 unsigned long *nr_written,
815 struct btrfs_root *root = BTRFS_I(inode)->root;
816 struct btrfs_trans_handle *trans;
819 unsigned long ram_size;
822 u64 blocksize = root->sectorsize;
823 struct btrfs_key ins;
824 struct extent_map *em;
825 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
828 BUG_ON(btrfs_is_free_space_inode(root, inode));
829 trans = btrfs_join_transaction(root);
831 extent_clear_unlock_delalloc(inode,
832 &BTRFS_I(inode)->io_tree,
833 start, end, locked_page,
834 EXTENT_CLEAR_UNLOCK_PAGE |
835 EXTENT_CLEAR_UNLOCK |
836 EXTENT_CLEAR_DELALLOC |
838 EXTENT_SET_WRITEBACK |
839 EXTENT_END_WRITEBACK);
840 return PTR_ERR(trans);
842 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
844 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
845 num_bytes = max(blocksize, num_bytes);
846 disk_num_bytes = num_bytes;
849 /* if this is a small write inside eof, kick off defrag */
850 if (num_bytes < 64 * 1024 &&
851 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
852 btrfs_add_inode_defrag(trans, inode);
855 /* lets try to make an inline extent */
856 ret = cow_file_range_inline(trans, root, inode,
857 start, end, 0, 0, NULL);
859 extent_clear_unlock_delalloc(inode,
860 &BTRFS_I(inode)->io_tree,
862 EXTENT_CLEAR_UNLOCK_PAGE |
863 EXTENT_CLEAR_UNLOCK |
864 EXTENT_CLEAR_DELALLOC |
866 EXTENT_SET_WRITEBACK |
867 EXTENT_END_WRITEBACK);
869 *nr_written = *nr_written +
870 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
873 } else if (ret < 0) {
874 btrfs_abort_transaction(trans, root, ret);
879 BUG_ON(disk_num_bytes >
880 btrfs_super_total_bytes(root->fs_info->super_copy));
882 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
883 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
885 while (disk_num_bytes > 0) {
888 cur_alloc_size = disk_num_bytes;
889 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
890 root->sectorsize, 0, alloc_hint,
893 btrfs_abort_transaction(trans, root, ret);
897 em = alloc_extent_map();
898 BUG_ON(!em); /* -ENOMEM */
900 em->orig_start = em->start;
901 ram_size = ins.offset;
902 em->len = ins.offset;
904 em->block_start = ins.objectid;
905 em->block_len = ins.offset;
906 em->bdev = root->fs_info->fs_devices->latest_bdev;
907 set_bit(EXTENT_FLAG_PINNED, &em->flags);
910 write_lock(&em_tree->lock);
911 ret = add_extent_mapping(em_tree, em);
912 write_unlock(&em_tree->lock);
913 if (ret != -EEXIST) {
917 btrfs_drop_extent_cache(inode, start,
918 start + ram_size - 1, 0);
921 cur_alloc_size = ins.offset;
922 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
923 ram_size, cur_alloc_size, 0);
924 BUG_ON(ret); /* -ENOMEM */
926 if (root->root_key.objectid ==
927 BTRFS_DATA_RELOC_TREE_OBJECTID) {
928 ret = btrfs_reloc_clone_csums(inode, start,
931 btrfs_abort_transaction(trans, root, ret);
936 if (disk_num_bytes < cur_alloc_size)
939 /* we're not doing compressed IO, don't unlock the first
940 * page (which the caller expects to stay locked), don't
941 * clear any dirty bits and don't set any writeback bits
943 * Do set the Private2 bit so we know this page was properly
944 * setup for writepage
946 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
947 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
950 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
951 start, start + ram_size - 1,
953 disk_num_bytes -= cur_alloc_size;
954 num_bytes -= cur_alloc_size;
955 alloc_hint = ins.objectid + ins.offset;
956 start += cur_alloc_size;
960 btrfs_end_transaction(trans, root);
964 extent_clear_unlock_delalloc(inode,
965 &BTRFS_I(inode)->io_tree,
966 start, end, locked_page,
967 EXTENT_CLEAR_UNLOCK_PAGE |
968 EXTENT_CLEAR_UNLOCK |
969 EXTENT_CLEAR_DELALLOC |
971 EXTENT_SET_WRITEBACK |
972 EXTENT_END_WRITEBACK);
978 * work queue call back to started compression on a file and pages
980 static noinline void async_cow_start(struct btrfs_work *work)
982 struct async_cow *async_cow;
984 async_cow = container_of(work, struct async_cow, work);
986 compress_file_range(async_cow->inode, async_cow->locked_page,
987 async_cow->start, async_cow->end, async_cow,
990 async_cow->inode = NULL;
994 * work queue call back to submit previously compressed pages
996 static noinline void async_cow_submit(struct btrfs_work *work)
998 struct async_cow *async_cow;
999 struct btrfs_root *root;
1000 unsigned long nr_pages;
1002 async_cow = container_of(work, struct async_cow, work);
1004 root = async_cow->root;
1005 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1008 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1010 if (atomic_read(&root->fs_info->async_delalloc_pages) <
1012 waitqueue_active(&root->fs_info->async_submit_wait))
1013 wake_up(&root->fs_info->async_submit_wait);
1015 if (async_cow->inode)
1016 submit_compressed_extents(async_cow->inode, async_cow);
1019 static noinline void async_cow_free(struct btrfs_work *work)
1021 struct async_cow *async_cow;
1022 async_cow = container_of(work, struct async_cow, work);
1026 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1027 u64 start, u64 end, int *page_started,
1028 unsigned long *nr_written)
1030 struct async_cow *async_cow;
1031 struct btrfs_root *root = BTRFS_I(inode)->root;
1032 unsigned long nr_pages;
1034 int limit = 10 * 1024 * 1042;
1036 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1037 1, 0, NULL, GFP_NOFS);
1038 while (start < end) {
1039 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1040 BUG_ON(!async_cow); /* -ENOMEM */
1041 async_cow->inode = inode;
1042 async_cow->root = root;
1043 async_cow->locked_page = locked_page;
1044 async_cow->start = start;
1046 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1049 cur_end = min(end, start + 512 * 1024 - 1);
1051 async_cow->end = cur_end;
1052 INIT_LIST_HEAD(&async_cow->extents);
1054 async_cow->work.func = async_cow_start;
1055 async_cow->work.ordered_func = async_cow_submit;
1056 async_cow->work.ordered_free = async_cow_free;
1057 async_cow->work.flags = 0;
1059 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1061 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1063 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1066 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1067 wait_event(root->fs_info->async_submit_wait,
1068 (atomic_read(&root->fs_info->async_delalloc_pages) <
1072 while (atomic_read(&root->fs_info->async_submit_draining) &&
1073 atomic_read(&root->fs_info->async_delalloc_pages)) {
1074 wait_event(root->fs_info->async_submit_wait,
1075 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1079 *nr_written += nr_pages;
1080 start = cur_end + 1;
1086 static noinline int csum_exist_in_range(struct btrfs_root *root,
1087 u64 bytenr, u64 num_bytes)
1090 struct btrfs_ordered_sum *sums;
1093 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1094 bytenr + num_bytes - 1, &list, 0);
1095 if (ret == 0 && list_empty(&list))
1098 while (!list_empty(&list)) {
1099 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1100 list_del(&sums->list);
1107 * when nowcow writeback call back. This checks for snapshots or COW copies
1108 * of the extents that exist in the file, and COWs the file as required.
1110 * If no cow copies or snapshots exist, we write directly to the existing
1113 static noinline int run_delalloc_nocow(struct inode *inode,
1114 struct page *locked_page,
1115 u64 start, u64 end, int *page_started, int force,
1116 unsigned long *nr_written)
1118 struct btrfs_root *root = BTRFS_I(inode)->root;
1119 struct btrfs_trans_handle *trans;
1120 struct extent_buffer *leaf;
1121 struct btrfs_path *path;
1122 struct btrfs_file_extent_item *fi;
1123 struct btrfs_key found_key;
1136 u64 ino = btrfs_ino(inode);
1138 path = btrfs_alloc_path();
1142 nolock = btrfs_is_free_space_inode(root, inode);
1145 trans = btrfs_join_transaction_nolock(root);
1147 trans = btrfs_join_transaction(root);
1149 if (IS_ERR(trans)) {
1150 btrfs_free_path(path);
1151 return PTR_ERR(trans);
1154 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1156 cow_start = (u64)-1;
1159 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1162 btrfs_abort_transaction(trans, root, ret);
1165 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1166 leaf = path->nodes[0];
1167 btrfs_item_key_to_cpu(leaf, &found_key,
1168 path->slots[0] - 1);
1169 if (found_key.objectid == ino &&
1170 found_key.type == BTRFS_EXTENT_DATA_KEY)
1175 leaf = path->nodes[0];
1176 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1177 ret = btrfs_next_leaf(root, path);
1179 btrfs_abort_transaction(trans, root, ret);
1184 leaf = path->nodes[0];
1190 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1192 if (found_key.objectid > ino ||
1193 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1194 found_key.offset > end)
1197 if (found_key.offset > cur_offset) {
1198 extent_end = found_key.offset;
1203 fi = btrfs_item_ptr(leaf, path->slots[0],
1204 struct btrfs_file_extent_item);
1205 extent_type = btrfs_file_extent_type(leaf, fi);
1207 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1208 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1209 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1210 extent_offset = btrfs_file_extent_offset(leaf, fi);
1211 extent_end = found_key.offset +
1212 btrfs_file_extent_num_bytes(leaf, fi);
1213 if (extent_end <= start) {
1217 if (disk_bytenr == 0)
1219 if (btrfs_file_extent_compression(leaf, fi) ||
1220 btrfs_file_extent_encryption(leaf, fi) ||
1221 btrfs_file_extent_other_encoding(leaf, fi))
1223 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1225 if (btrfs_extent_readonly(root, disk_bytenr))
1227 if (btrfs_cross_ref_exist(trans, root, ino,
1229 extent_offset, disk_bytenr))
1231 disk_bytenr += extent_offset;
1232 disk_bytenr += cur_offset - found_key.offset;
1233 num_bytes = min(end + 1, extent_end) - cur_offset;
1235 * force cow if csum exists in the range.
1236 * this ensure that csum for a given extent are
1237 * either valid or do not exist.
1239 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1242 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1243 extent_end = found_key.offset +
1244 btrfs_file_extent_inline_len(leaf, fi);
1245 extent_end = ALIGN(extent_end, root->sectorsize);
1250 if (extent_end <= start) {
1255 if (cow_start == (u64)-1)
1256 cow_start = cur_offset;
1257 cur_offset = extent_end;
1258 if (cur_offset > end)
1264 btrfs_release_path(path);
1265 if (cow_start != (u64)-1) {
1266 ret = cow_file_range(inode, locked_page, cow_start,
1267 found_key.offset - 1, page_started,
1270 btrfs_abort_transaction(trans, root, ret);
1273 cow_start = (u64)-1;
1276 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1277 struct extent_map *em;
1278 struct extent_map_tree *em_tree;
1279 em_tree = &BTRFS_I(inode)->extent_tree;
1280 em = alloc_extent_map();
1281 BUG_ON(!em); /* -ENOMEM */
1282 em->start = cur_offset;
1283 em->orig_start = em->start;
1284 em->len = num_bytes;
1285 em->block_len = num_bytes;
1286 em->block_start = disk_bytenr;
1287 em->bdev = root->fs_info->fs_devices->latest_bdev;
1288 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1290 write_lock(&em_tree->lock);
1291 ret = add_extent_mapping(em_tree, em);
1292 write_unlock(&em_tree->lock);
1293 if (ret != -EEXIST) {
1294 free_extent_map(em);
1297 btrfs_drop_extent_cache(inode, em->start,
1298 em->start + em->len - 1, 0);
1300 type = BTRFS_ORDERED_PREALLOC;
1302 type = BTRFS_ORDERED_NOCOW;
1305 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1306 num_bytes, num_bytes, type);
1307 BUG_ON(ret); /* -ENOMEM */
1309 if (root->root_key.objectid ==
1310 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1311 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1314 btrfs_abort_transaction(trans, root, ret);
1319 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1320 cur_offset, cur_offset + num_bytes - 1,
1321 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1322 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1323 EXTENT_SET_PRIVATE2);
1324 cur_offset = extent_end;
1325 if (cur_offset > end)
1328 btrfs_release_path(path);
1330 if (cur_offset <= end && cow_start == (u64)-1)
1331 cow_start = cur_offset;
1332 if (cow_start != (u64)-1) {
1333 ret = cow_file_range(inode, locked_page, cow_start, end,
1334 page_started, nr_written, 1);
1336 btrfs_abort_transaction(trans, root, ret);
1343 err = btrfs_end_transaction_nolock(trans, root);
1345 err = btrfs_end_transaction(trans, root);
1350 btrfs_free_path(path);
1355 * extent_io.c call back to do delayed allocation processing
1357 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1358 u64 start, u64 end, int *page_started,
1359 unsigned long *nr_written)
1362 struct btrfs_root *root = BTRFS_I(inode)->root;
1364 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1365 ret = run_delalloc_nocow(inode, locked_page, start, end,
1366 page_started, 1, nr_written);
1367 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1368 ret = run_delalloc_nocow(inode, locked_page, start, end,
1369 page_started, 0, nr_written);
1370 else if (!btrfs_test_opt(root, COMPRESS) &&
1371 !(BTRFS_I(inode)->force_compress) &&
1372 !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1373 ret = cow_file_range(inode, locked_page, start, end,
1374 page_started, nr_written, 1);
1376 ret = cow_file_range_async(inode, locked_page, start, end,
1377 page_started, nr_written);
1381 static void btrfs_split_extent_hook(struct inode *inode,
1382 struct extent_state *orig, u64 split)
1384 /* not delalloc, ignore it */
1385 if (!(orig->state & EXTENT_DELALLOC))
1388 spin_lock(&BTRFS_I(inode)->lock);
1389 BTRFS_I(inode)->outstanding_extents++;
1390 spin_unlock(&BTRFS_I(inode)->lock);
1394 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1395 * extents so we can keep track of new extents that are just merged onto old
1396 * extents, such as when we are doing sequential writes, so we can properly
1397 * account for the metadata space we'll need.
1399 static void btrfs_merge_extent_hook(struct inode *inode,
1400 struct extent_state *new,
1401 struct extent_state *other)
1403 /* not delalloc, ignore it */
1404 if (!(other->state & EXTENT_DELALLOC))
1407 spin_lock(&BTRFS_I(inode)->lock);
1408 BTRFS_I(inode)->outstanding_extents--;
1409 spin_unlock(&BTRFS_I(inode)->lock);
1413 * extent_io.c set_bit_hook, used to track delayed allocation
1414 * bytes in this file, and to maintain the list of inodes that
1415 * have pending delalloc work to be done.
1417 static void btrfs_set_bit_hook(struct inode *inode,
1418 struct extent_state *state, int *bits)
1422 * set_bit and clear bit hooks normally require _irqsave/restore
1423 * but in this case, we are only testing for the DELALLOC
1424 * bit, which is only set or cleared with irqs on
1426 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1427 struct btrfs_root *root = BTRFS_I(inode)->root;
1428 u64 len = state->end + 1 - state->start;
1429 bool do_list = !btrfs_is_free_space_inode(root, inode);
1431 if (*bits & EXTENT_FIRST_DELALLOC) {
1432 *bits &= ~EXTENT_FIRST_DELALLOC;
1434 spin_lock(&BTRFS_I(inode)->lock);
1435 BTRFS_I(inode)->outstanding_extents++;
1436 spin_unlock(&BTRFS_I(inode)->lock);
1439 spin_lock(&root->fs_info->delalloc_lock);
1440 BTRFS_I(inode)->delalloc_bytes += len;
1441 root->fs_info->delalloc_bytes += len;
1442 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1443 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1444 &root->fs_info->delalloc_inodes);
1446 spin_unlock(&root->fs_info->delalloc_lock);
1451 * extent_io.c clear_bit_hook, see set_bit_hook for why
1453 static void btrfs_clear_bit_hook(struct inode *inode,
1454 struct extent_state *state, int *bits)
1457 * set_bit and clear bit hooks normally require _irqsave/restore
1458 * but in this case, we are only testing for the DELALLOC
1459 * bit, which is only set or cleared with irqs on
1461 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1462 struct btrfs_root *root = BTRFS_I(inode)->root;
1463 u64 len = state->end + 1 - state->start;
1464 bool do_list = !btrfs_is_free_space_inode(root, inode);
1466 if (*bits & EXTENT_FIRST_DELALLOC) {
1467 *bits &= ~EXTENT_FIRST_DELALLOC;
1468 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1469 spin_lock(&BTRFS_I(inode)->lock);
1470 BTRFS_I(inode)->outstanding_extents--;
1471 spin_unlock(&BTRFS_I(inode)->lock);
1474 if (*bits & EXTENT_DO_ACCOUNTING)
1475 btrfs_delalloc_release_metadata(inode, len);
1477 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1479 btrfs_free_reserved_data_space(inode, len);
1481 spin_lock(&root->fs_info->delalloc_lock);
1482 root->fs_info->delalloc_bytes -= len;
1483 BTRFS_I(inode)->delalloc_bytes -= len;
1485 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1486 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1487 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1489 spin_unlock(&root->fs_info->delalloc_lock);
1494 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1495 * we don't create bios that span stripes or chunks
1497 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1498 size_t size, struct bio *bio,
1499 unsigned long bio_flags)
1501 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1502 struct btrfs_mapping_tree *map_tree;
1503 u64 logical = (u64)bio->bi_sector << 9;
1508 if (bio_flags & EXTENT_BIO_COMPRESSED)
1511 length = bio->bi_size;
1512 map_tree = &root->fs_info->mapping_tree;
1513 map_length = length;
1514 ret = btrfs_map_block(map_tree, READ, logical,
1515 &map_length, NULL, 0);
1516 /* Will always return 0 or 1 with map_multi == NULL */
1518 if (map_length < length + size)
1524 * in order to insert checksums into the metadata in large chunks,
1525 * we wait until bio submission time. All the pages in the bio are
1526 * checksummed and sums are attached onto the ordered extent record.
1528 * At IO completion time the cums attached on the ordered extent record
1529 * are inserted into the btree
1531 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1532 struct bio *bio, int mirror_num,
1533 unsigned long bio_flags,
1536 struct btrfs_root *root = BTRFS_I(inode)->root;
1539 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1540 BUG_ON(ret); /* -ENOMEM */
1545 * in order to insert checksums into the metadata in large chunks,
1546 * we wait until bio submission time. All the pages in the bio are
1547 * checksummed and sums are attached onto the ordered extent record.
1549 * At IO completion time the cums attached on the ordered extent record
1550 * are inserted into the btree
1552 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1553 int mirror_num, unsigned long bio_flags,
1556 struct btrfs_root *root = BTRFS_I(inode)->root;
1557 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1561 * extent_io.c submission hook. This does the right thing for csum calculation
1562 * on write, or reading the csums from the tree before a read
1564 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1565 int mirror_num, unsigned long bio_flags,
1568 struct btrfs_root *root = BTRFS_I(inode)->root;
1573 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1575 if (btrfs_is_free_space_inode(root, inode))
1578 if (!(rw & REQ_WRITE)) {
1579 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1583 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1584 return btrfs_submit_compressed_read(inode, bio,
1585 mirror_num, bio_flags);
1586 } else if (!skip_sum) {
1587 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1592 } else if (!skip_sum) {
1593 /* csum items have already been cloned */
1594 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1596 /* we're doing a write, do the async checksumming */
1597 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1598 inode, rw, bio, mirror_num,
1599 bio_flags, bio_offset,
1600 __btrfs_submit_bio_start,
1601 __btrfs_submit_bio_done);
1605 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1609 * given a list of ordered sums record them in the inode. This happens
1610 * at IO completion time based on sums calculated at bio submission time.
1612 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1613 struct inode *inode, u64 file_offset,
1614 struct list_head *list)
1616 struct btrfs_ordered_sum *sum;
1618 list_for_each_entry(sum, list, list) {
1619 btrfs_csum_file_blocks(trans,
1620 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1625 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1626 struct extent_state **cached_state)
1628 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1630 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1631 cached_state, GFP_NOFS);
1634 /* see btrfs_writepage_start_hook for details on why this is required */
1635 struct btrfs_writepage_fixup {
1637 struct btrfs_work work;
1640 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1642 struct btrfs_writepage_fixup *fixup;
1643 struct btrfs_ordered_extent *ordered;
1644 struct extent_state *cached_state = NULL;
1646 struct inode *inode;
1651 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1655 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1656 ClearPageChecked(page);
1660 inode = page->mapping->host;
1661 page_start = page_offset(page);
1662 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1664 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1667 /* already ordered? We're done */
1668 if (PagePrivate2(page))
1671 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1673 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1674 page_end, &cached_state, GFP_NOFS);
1676 btrfs_start_ordered_extent(inode, ordered, 1);
1677 btrfs_put_ordered_extent(ordered);
1681 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1683 mapping_set_error(page->mapping, ret);
1684 end_extent_writepage(page, ret, page_start, page_end);
1685 ClearPageChecked(page);
1689 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1690 ClearPageChecked(page);
1691 set_page_dirty(page);
1693 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1694 &cached_state, GFP_NOFS);
1697 page_cache_release(page);
1702 * There are a few paths in the higher layers of the kernel that directly
1703 * set the page dirty bit without asking the filesystem if it is a
1704 * good idea. This causes problems because we want to make sure COW
1705 * properly happens and the data=ordered rules are followed.
1707 * In our case any range that doesn't have the ORDERED bit set
1708 * hasn't been properly setup for IO. We kick off an async process
1709 * to fix it up. The async helper will wait for ordered extents, set
1710 * the delalloc bit and make it safe to write the page.
1712 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1714 struct inode *inode = page->mapping->host;
1715 struct btrfs_writepage_fixup *fixup;
1716 struct btrfs_root *root = BTRFS_I(inode)->root;
1718 /* this page is properly in the ordered list */
1719 if (TestClearPagePrivate2(page))
1722 if (PageChecked(page))
1725 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1729 SetPageChecked(page);
1730 page_cache_get(page);
1731 fixup->work.func = btrfs_writepage_fixup_worker;
1733 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1737 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1738 struct inode *inode, u64 file_pos,
1739 u64 disk_bytenr, u64 disk_num_bytes,
1740 u64 num_bytes, u64 ram_bytes,
1741 u8 compression, u8 encryption,
1742 u16 other_encoding, int extent_type)
1744 struct btrfs_root *root = BTRFS_I(inode)->root;
1745 struct btrfs_file_extent_item *fi;
1746 struct btrfs_path *path;
1747 struct extent_buffer *leaf;
1748 struct btrfs_key ins;
1752 path = btrfs_alloc_path();
1756 path->leave_spinning = 1;
1759 * we may be replacing one extent in the tree with another.
1760 * The new extent is pinned in the extent map, and we don't want
1761 * to drop it from the cache until it is completely in the btree.
1763 * So, tell btrfs_drop_extents to leave this extent in the cache.
1764 * the caller is expected to unpin it and allow it to be merged
1767 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1772 ins.objectid = btrfs_ino(inode);
1773 ins.offset = file_pos;
1774 ins.type = BTRFS_EXTENT_DATA_KEY;
1775 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1778 leaf = path->nodes[0];
1779 fi = btrfs_item_ptr(leaf, path->slots[0],
1780 struct btrfs_file_extent_item);
1781 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1782 btrfs_set_file_extent_type(leaf, fi, extent_type);
1783 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1784 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1785 btrfs_set_file_extent_offset(leaf, fi, 0);
1786 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1787 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1788 btrfs_set_file_extent_compression(leaf, fi, compression);
1789 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1790 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1792 btrfs_unlock_up_safe(path, 1);
1793 btrfs_set_lock_blocking(leaf);
1795 btrfs_mark_buffer_dirty(leaf);
1797 inode_add_bytes(inode, num_bytes);
1799 ins.objectid = disk_bytenr;
1800 ins.offset = disk_num_bytes;
1801 ins.type = BTRFS_EXTENT_ITEM_KEY;
1802 ret = btrfs_alloc_reserved_file_extent(trans, root,
1803 root->root_key.objectid,
1804 btrfs_ino(inode), file_pos, &ins);
1806 btrfs_free_path(path);
1812 * helper function for btrfs_finish_ordered_io, this
1813 * just reads in some of the csum leaves to prime them into ram
1814 * before we start the transaction. It limits the amount of btree
1815 * reads required while inside the transaction.
1817 /* as ordered data IO finishes, this gets called so we can finish
1818 * an ordered extent if the range of bytes in the file it covers are
1821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1823 struct inode *inode = ordered_extent->inode;
1824 struct btrfs_root *root = BTRFS_I(inode)->root;
1825 struct btrfs_trans_handle *trans = NULL;
1826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1827 struct extent_state *cached_state = NULL;
1828 int compress_type = 0;
1832 nolock = btrfs_is_free_space_inode(root, inode);
1834 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1839 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1840 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1841 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1844 trans = btrfs_join_transaction_nolock(root);
1846 trans = btrfs_join_transaction(root);
1848 return PTR_ERR(trans);
1849 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1850 ret = btrfs_update_inode_fallback(trans, root, inode);
1851 if (ret) /* -ENOMEM or corruption */
1852 btrfs_abort_transaction(trans, root, ret);
1857 lock_extent_bits(io_tree, ordered_extent->file_offset,
1858 ordered_extent->file_offset + ordered_extent->len - 1,
1862 trans = btrfs_join_transaction_nolock(root);
1864 trans = btrfs_join_transaction(root);
1865 if (IS_ERR(trans)) {
1866 ret = PTR_ERR(trans);
1870 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1872 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1873 compress_type = ordered_extent->compress_type;
1874 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1875 BUG_ON(compress_type);
1876 ret = btrfs_mark_extent_written(trans, inode,
1877 ordered_extent->file_offset,
1878 ordered_extent->file_offset +
1879 ordered_extent->len);
1881 BUG_ON(root == root->fs_info->tree_root);
1882 ret = insert_reserved_file_extent(trans, inode,
1883 ordered_extent->file_offset,
1884 ordered_extent->start,
1885 ordered_extent->disk_len,
1886 ordered_extent->len,
1887 ordered_extent->len,
1888 compress_type, 0, 0,
1889 BTRFS_FILE_EXTENT_REG);
1890 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1891 ordered_extent->file_offset,
1892 ordered_extent->len);
1896 btrfs_abort_transaction(trans, root, ret);
1900 add_pending_csums(trans, inode, ordered_extent->file_offset,
1901 &ordered_extent->list);
1903 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1904 if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1905 ret = btrfs_update_inode_fallback(trans, root, inode);
1906 if (ret) { /* -ENOMEM or corruption */
1907 btrfs_abort_transaction(trans, root, ret);
1913 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1914 ordered_extent->file_offset +
1915 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1917 if (root != root->fs_info->tree_root)
1918 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1921 btrfs_end_transaction_nolock(trans, root);
1923 btrfs_end_transaction(trans, root);
1927 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1928 ordered_extent->file_offset +
1929 ordered_extent->len - 1, NULL, GFP_NOFS);
1932 * This needs to be dont to make sure anybody waiting knows we are done
1933 * upating everything for this ordered extent.
1935 btrfs_remove_ordered_extent(inode, ordered_extent);
1938 btrfs_put_ordered_extent(ordered_extent);
1939 /* once for the tree */
1940 btrfs_put_ordered_extent(ordered_extent);
1945 static void finish_ordered_fn(struct btrfs_work *work)
1947 struct btrfs_ordered_extent *ordered_extent;
1948 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1949 btrfs_finish_ordered_io(ordered_extent);
1952 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1953 struct extent_state *state, int uptodate)
1955 struct inode *inode = page->mapping->host;
1956 struct btrfs_root *root = BTRFS_I(inode)->root;
1957 struct btrfs_ordered_extent *ordered_extent = NULL;
1958 struct btrfs_workers *workers;
1960 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1962 ClearPagePrivate2(page);
1963 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1964 end - start + 1, uptodate))
1967 ordered_extent->work.func = finish_ordered_fn;
1968 ordered_extent->work.flags = 0;
1970 if (btrfs_is_free_space_inode(root, inode))
1971 workers = &root->fs_info->endio_freespace_worker;
1973 workers = &root->fs_info->endio_write_workers;
1974 btrfs_queue_worker(workers, &ordered_extent->work);
1980 * when reads are done, we need to check csums to verify the data is correct
1981 * if there's a match, we allow the bio to finish. If not, the code in
1982 * extent_io.c will try to find good copies for us.
1984 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1985 struct extent_state *state, int mirror)
1987 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1988 struct inode *inode = page->mapping->host;
1989 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1991 u64 private = ~(u32)0;
1993 struct btrfs_root *root = BTRFS_I(inode)->root;
1996 if (PageChecked(page)) {
1997 ClearPageChecked(page);
2001 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2004 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2005 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2006 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2011 if (state && state->start == start) {
2012 private = state->private;
2015 ret = get_state_private(io_tree, start, &private);
2017 kaddr = kmap_atomic(page);
2021 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
2022 btrfs_csum_final(csum, (char *)&csum);
2023 if (csum != private)
2026 kunmap_atomic(kaddr);
2031 printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2033 (unsigned long long)btrfs_ino(page->mapping->host),
2034 (unsigned long long)start, csum,
2035 (unsigned long long)private);
2036 memset(kaddr + offset, 1, end - start + 1);
2037 flush_dcache_page(page);
2038 kunmap_atomic(kaddr);
2044 struct delayed_iput {
2045 struct list_head list;
2046 struct inode *inode;
2049 /* JDM: If this is fs-wide, why can't we add a pointer to
2050 * btrfs_inode instead and avoid the allocation? */
2051 void btrfs_add_delayed_iput(struct inode *inode)
2053 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2054 struct delayed_iput *delayed;
2056 if (atomic_add_unless(&inode->i_count, -1, 1))
2059 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2060 delayed->inode = inode;
2062 spin_lock(&fs_info->delayed_iput_lock);
2063 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2064 spin_unlock(&fs_info->delayed_iput_lock);
2067 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2070 struct btrfs_fs_info *fs_info = root->fs_info;
2071 struct delayed_iput *delayed;
2074 spin_lock(&fs_info->delayed_iput_lock);
2075 empty = list_empty(&fs_info->delayed_iputs);
2076 spin_unlock(&fs_info->delayed_iput_lock);
2080 down_read(&root->fs_info->cleanup_work_sem);
2081 spin_lock(&fs_info->delayed_iput_lock);
2082 list_splice_init(&fs_info->delayed_iputs, &list);
2083 spin_unlock(&fs_info->delayed_iput_lock);
2085 while (!list_empty(&list)) {
2086 delayed = list_entry(list.next, struct delayed_iput, list);
2087 list_del(&delayed->list);
2088 iput(delayed->inode);
2091 up_read(&root->fs_info->cleanup_work_sem);
2094 enum btrfs_orphan_cleanup_state {
2095 ORPHAN_CLEANUP_STARTED = 1,
2096 ORPHAN_CLEANUP_DONE = 2,
2100 * This is called in transaction commit time. If there are no orphan
2101 * files in the subvolume, it removes orphan item and frees block_rsv
2104 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root)
2107 struct btrfs_block_rsv *block_rsv;
2110 if (atomic_read(&root->orphan_inodes) ||
2111 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2114 spin_lock(&root->orphan_lock);
2115 if (atomic_read(&root->orphan_inodes)) {
2116 spin_unlock(&root->orphan_lock);
2120 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2121 spin_unlock(&root->orphan_lock);
2125 block_rsv = root->orphan_block_rsv;
2126 root->orphan_block_rsv = NULL;
2127 spin_unlock(&root->orphan_lock);
2129 if (root->orphan_item_inserted &&
2130 btrfs_root_refs(&root->root_item) > 0) {
2131 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2132 root->root_key.objectid);
2134 root->orphan_item_inserted = 0;
2138 WARN_ON(block_rsv->size > 0);
2139 btrfs_free_block_rsv(root, block_rsv);
2144 * This creates an orphan entry for the given inode in case something goes
2145 * wrong in the middle of an unlink/truncate.
2147 * NOTE: caller of this function should reserve 5 units of metadata for
2150 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2152 struct btrfs_root *root = BTRFS_I(inode)->root;
2153 struct btrfs_block_rsv *block_rsv = NULL;
2158 if (!root->orphan_block_rsv) {
2159 block_rsv = btrfs_alloc_block_rsv(root);
2164 spin_lock(&root->orphan_lock);
2165 if (!root->orphan_block_rsv) {
2166 root->orphan_block_rsv = block_rsv;
2167 } else if (block_rsv) {
2168 btrfs_free_block_rsv(root, block_rsv);
2172 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2173 &BTRFS_I(inode)->runtime_flags)) {
2176 * For proper ENOSPC handling, we should do orphan
2177 * cleanup when mounting. But this introduces backward
2178 * compatibility issue.
2180 if (!xchg(&root->orphan_item_inserted, 1))
2186 atomic_dec(&root->orphan_inodes);
2189 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2190 &BTRFS_I(inode)->runtime_flags))
2192 spin_unlock(&root->orphan_lock);
2194 /* grab metadata reservation from transaction handle */
2196 ret = btrfs_orphan_reserve_metadata(trans, inode);
2197 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2200 /* insert an orphan item to track this unlinked/truncated file */
2202 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2203 if (ret && ret != -EEXIST) {
2204 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2205 &BTRFS_I(inode)->runtime_flags);
2206 btrfs_abort_transaction(trans, root, ret);
2212 /* insert an orphan item to track subvolume contains orphan files */
2214 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2215 root->root_key.objectid);
2216 if (ret && ret != -EEXIST) {
2217 btrfs_abort_transaction(trans, root, ret);
2225 * We have done the truncate/delete so we can go ahead and remove the orphan
2226 * item for this particular inode.
2228 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2230 struct btrfs_root *root = BTRFS_I(inode)->root;
2231 int delete_item = 0;
2232 int release_rsv = 0;
2235 spin_lock(&root->orphan_lock);
2236 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2237 &BTRFS_I(inode)->runtime_flags))
2240 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2241 &BTRFS_I(inode)->runtime_flags))
2243 spin_unlock(&root->orphan_lock);
2245 if (trans && delete_item) {
2246 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2247 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2251 btrfs_orphan_release_metadata(inode);
2252 atomic_dec(&root->orphan_inodes);
2259 * this cleans up any orphans that may be left on the list from the last use
2262 int btrfs_orphan_cleanup(struct btrfs_root *root)
2264 struct btrfs_path *path;
2265 struct extent_buffer *leaf;
2266 struct btrfs_key key, found_key;
2267 struct btrfs_trans_handle *trans;
2268 struct inode *inode;
2269 u64 last_objectid = 0;
2270 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2272 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2275 path = btrfs_alloc_path();
2282 key.objectid = BTRFS_ORPHAN_OBJECTID;
2283 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2284 key.offset = (u64)-1;
2287 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2292 * if ret == 0 means we found what we were searching for, which
2293 * is weird, but possible, so only screw with path if we didn't
2294 * find the key and see if we have stuff that matches
2298 if (path->slots[0] == 0)
2303 /* pull out the item */
2304 leaf = path->nodes[0];
2305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2307 /* make sure the item matches what we want */
2308 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2310 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2313 /* release the path since we're done with it */
2314 btrfs_release_path(path);
2317 * this is where we are basically btrfs_lookup, without the
2318 * crossing root thing. we store the inode number in the
2319 * offset of the orphan item.
2322 if (found_key.offset == last_objectid) {
2323 printk(KERN_ERR "btrfs: Error removing orphan entry, "
2324 "stopping orphan cleanup\n");
2329 last_objectid = found_key.offset;
2331 found_key.objectid = found_key.offset;
2332 found_key.type = BTRFS_INODE_ITEM_KEY;
2333 found_key.offset = 0;
2334 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2335 ret = PTR_RET(inode);
2336 if (ret && ret != -ESTALE)
2339 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2340 struct btrfs_root *dead_root;
2341 struct btrfs_fs_info *fs_info = root->fs_info;
2342 int is_dead_root = 0;
2345 * this is an orphan in the tree root. Currently these
2346 * could come from 2 sources:
2347 * a) a snapshot deletion in progress
2348 * b) a free space cache inode
2349 * We need to distinguish those two, as the snapshot
2350 * orphan must not get deleted.
2351 * find_dead_roots already ran before us, so if this
2352 * is a snapshot deletion, we should find the root
2353 * in the dead_roots list
2355 spin_lock(&fs_info->trans_lock);
2356 list_for_each_entry(dead_root, &fs_info->dead_roots,
2358 if (dead_root->root_key.objectid ==
2359 found_key.objectid) {
2364 spin_unlock(&fs_info->trans_lock);
2366 /* prevent this orphan from being found again */
2367 key.offset = found_key.objectid - 1;
2372 * Inode is already gone but the orphan item is still there,
2373 * kill the orphan item.
2375 if (ret == -ESTALE) {
2376 trans = btrfs_start_transaction(root, 1);
2377 if (IS_ERR(trans)) {
2378 ret = PTR_ERR(trans);
2381 printk(KERN_ERR "auto deleting %Lu\n",
2382 found_key.objectid);
2383 ret = btrfs_del_orphan_item(trans, root,
2384 found_key.objectid);
2385 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2386 btrfs_end_transaction(trans, root);
2391 * add this inode to the orphan list so btrfs_orphan_del does
2392 * the proper thing when we hit it
2394 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2395 &BTRFS_I(inode)->runtime_flags);
2397 /* if we have links, this was a truncate, lets do that */
2398 if (inode->i_nlink) {
2399 if (!S_ISREG(inode->i_mode)) {
2405 ret = btrfs_truncate(inode);
2410 /* this will do delete_inode and everything for us */
2415 /* release the path since we're done with it */
2416 btrfs_release_path(path);
2418 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2420 if (root->orphan_block_rsv)
2421 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2424 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2425 trans = btrfs_join_transaction(root);
2427 btrfs_end_transaction(trans, root);
2431 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2433 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2437 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2438 btrfs_free_path(path);
2443 * very simple check to peek ahead in the leaf looking for xattrs. If we
2444 * don't find any xattrs, we know there can't be any acls.
2446 * slot is the slot the inode is in, objectid is the objectid of the inode
2448 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2449 int slot, u64 objectid)
2451 u32 nritems = btrfs_header_nritems(leaf);
2452 struct btrfs_key found_key;
2456 while (slot < nritems) {
2457 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2459 /* we found a different objectid, there must not be acls */
2460 if (found_key.objectid != objectid)
2463 /* we found an xattr, assume we've got an acl */
2464 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2468 * we found a key greater than an xattr key, there can't
2469 * be any acls later on
2471 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2478 * it goes inode, inode backrefs, xattrs, extents,
2479 * so if there are a ton of hard links to an inode there can
2480 * be a lot of backrefs. Don't waste time searching too hard,
2481 * this is just an optimization
2486 /* we hit the end of the leaf before we found an xattr or
2487 * something larger than an xattr. We have to assume the inode
2494 * read an inode from the btree into the in-memory inode
2496 static void btrfs_read_locked_inode(struct inode *inode)
2498 struct btrfs_path *path;
2499 struct extent_buffer *leaf;
2500 struct btrfs_inode_item *inode_item;
2501 struct btrfs_timespec *tspec;
2502 struct btrfs_root *root = BTRFS_I(inode)->root;
2503 struct btrfs_key location;
2507 bool filled = false;
2509 ret = btrfs_fill_inode(inode, &rdev);
2513 path = btrfs_alloc_path();
2517 path->leave_spinning = 1;
2518 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2520 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2524 leaf = path->nodes[0];
2529 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2530 struct btrfs_inode_item);
2531 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2532 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2533 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2534 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2535 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2537 tspec = btrfs_inode_atime(inode_item);
2538 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2539 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2541 tspec = btrfs_inode_mtime(inode_item);
2542 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2543 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2545 tspec = btrfs_inode_ctime(inode_item);
2546 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2547 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2549 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2550 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2551 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2552 inode->i_generation = BTRFS_I(inode)->generation;
2554 rdev = btrfs_inode_rdev(leaf, inode_item);
2556 BTRFS_I(inode)->index_cnt = (u64)-1;
2557 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2560 * try to precache a NULL acl entry for files that don't have
2561 * any xattrs or acls
2563 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2566 cache_no_acl(inode);
2568 btrfs_free_path(path);
2570 switch (inode->i_mode & S_IFMT) {
2572 inode->i_mapping->a_ops = &btrfs_aops;
2573 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2575 inode->i_fop = &btrfs_file_operations;
2576 inode->i_op = &btrfs_file_inode_operations;
2579 inode->i_fop = &btrfs_dir_file_operations;
2580 if (root == root->fs_info->tree_root)
2581 inode->i_op = &btrfs_dir_ro_inode_operations;
2583 inode->i_op = &btrfs_dir_inode_operations;
2586 inode->i_op = &btrfs_symlink_inode_operations;
2587 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2588 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2591 inode->i_op = &btrfs_special_inode_operations;
2592 init_special_inode(inode, inode->i_mode, rdev);
2596 btrfs_update_iflags(inode);
2600 btrfs_free_path(path);
2601 make_bad_inode(inode);
2605 * given a leaf and an inode, copy the inode fields into the leaf
2607 static void fill_inode_item(struct btrfs_trans_handle *trans,
2608 struct extent_buffer *leaf,
2609 struct btrfs_inode_item *item,
2610 struct inode *inode)
2612 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2613 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2614 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2615 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2616 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2618 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2619 inode->i_atime.tv_sec);
2620 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2621 inode->i_atime.tv_nsec);
2623 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2624 inode->i_mtime.tv_sec);
2625 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2626 inode->i_mtime.tv_nsec);
2628 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2629 inode->i_ctime.tv_sec);
2630 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2631 inode->i_ctime.tv_nsec);
2633 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2634 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2635 btrfs_set_inode_sequence(leaf, item, inode->i_version);
2636 btrfs_set_inode_transid(leaf, item, trans->transid);
2637 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2638 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2639 btrfs_set_inode_block_group(leaf, item, 0);
2643 * copy everything in the in-memory inode into the btree.
2645 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2646 struct btrfs_root *root, struct inode *inode)
2648 struct btrfs_inode_item *inode_item;
2649 struct btrfs_path *path;
2650 struct extent_buffer *leaf;
2653 path = btrfs_alloc_path();
2657 path->leave_spinning = 1;
2658 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2666 btrfs_unlock_up_safe(path, 1);
2667 leaf = path->nodes[0];
2668 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2669 struct btrfs_inode_item);
2671 fill_inode_item(trans, leaf, inode_item, inode);
2672 btrfs_mark_buffer_dirty(leaf);
2673 btrfs_set_inode_last_trans(trans, inode);
2676 btrfs_free_path(path);
2681 * copy everything in the in-memory inode into the btree.
2683 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2684 struct btrfs_root *root, struct inode *inode)
2689 * If the inode is a free space inode, we can deadlock during commit
2690 * if we put it into the delayed code.
2692 * The data relocation inode should also be directly updated
2695 if (!btrfs_is_free_space_inode(root, inode)
2696 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2697 ret = btrfs_delayed_update_inode(trans, root, inode);
2699 btrfs_set_inode_last_trans(trans, inode);
2703 return btrfs_update_inode_item(trans, root, inode);
2706 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2707 struct btrfs_root *root, struct inode *inode)
2711 ret = btrfs_update_inode(trans, root, inode);
2713 return btrfs_update_inode_item(trans, root, inode);
2718 * unlink helper that gets used here in inode.c and in the tree logging
2719 * recovery code. It remove a link in a directory with a given name, and
2720 * also drops the back refs in the inode to the directory
2722 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2723 struct btrfs_root *root,
2724 struct inode *dir, struct inode *inode,
2725 const char *name, int name_len)
2727 struct btrfs_path *path;
2729 struct extent_buffer *leaf;
2730 struct btrfs_dir_item *di;
2731 struct btrfs_key key;
2733 u64 ino = btrfs_ino(inode);
2734 u64 dir_ino = btrfs_ino(dir);
2736 path = btrfs_alloc_path();
2742 path->leave_spinning = 1;
2743 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2744 name, name_len, -1);
2753 leaf = path->nodes[0];
2754 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2755 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2758 btrfs_release_path(path);
2760 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2763 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2764 "inode %llu parent %llu\n", name_len, name,
2765 (unsigned long long)ino, (unsigned long long)dir_ino);
2766 btrfs_abort_transaction(trans, root, ret);
2770 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2772 btrfs_abort_transaction(trans, root, ret);
2776 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2778 if (ret != 0 && ret != -ENOENT) {
2779 btrfs_abort_transaction(trans, root, ret);
2783 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2788 btrfs_free_path(path);
2792 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2793 inode_inc_iversion(inode);
2794 inode_inc_iversion(dir);
2795 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2796 btrfs_update_inode(trans, root, dir);
2801 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2802 struct btrfs_root *root,
2803 struct inode *dir, struct inode *inode,
2804 const char *name, int name_len)
2807 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2809 btrfs_drop_nlink(inode);
2810 ret = btrfs_update_inode(trans, root, inode);
2816 /* helper to check if there is any shared block in the path */
2817 static int check_path_shared(struct btrfs_root *root,
2818 struct btrfs_path *path)
2820 struct extent_buffer *eb;
2824 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2827 if (!path->nodes[level])
2829 eb = path->nodes[level];
2830 if (!btrfs_block_can_be_shared(root, eb))
2832 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2841 * helper to start transaction for unlink and rmdir.
2843 * unlink and rmdir are special in btrfs, they do not always free space.
2844 * so in enospc case, we should make sure they will free space before
2845 * allowing them to use the global metadata reservation.
2847 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2848 struct dentry *dentry)
2850 struct btrfs_trans_handle *trans;
2851 struct btrfs_root *root = BTRFS_I(dir)->root;
2852 struct btrfs_path *path;
2853 struct btrfs_inode_ref *ref;
2854 struct btrfs_dir_item *di;
2855 struct inode *inode = dentry->d_inode;
2860 u64 ino = btrfs_ino(inode);
2861 u64 dir_ino = btrfs_ino(dir);
2864 * 1 for the possible orphan item
2865 * 1 for the dir item
2866 * 1 for the dir index
2867 * 1 for the inode ref
2868 * 1 for the inode ref in the tree log
2869 * 2 for the dir entries in the log
2872 trans = btrfs_start_transaction(root, 8);
2873 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2876 if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2877 return ERR_PTR(-ENOSPC);
2879 /* check if there is someone else holds reference */
2880 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2881 return ERR_PTR(-ENOSPC);
2883 if (atomic_read(&inode->i_count) > 2)
2884 return ERR_PTR(-ENOSPC);
2886 if (xchg(&root->fs_info->enospc_unlink, 1))
2887 return ERR_PTR(-ENOSPC);
2889 path = btrfs_alloc_path();
2891 root->fs_info->enospc_unlink = 0;
2892 return ERR_PTR(-ENOMEM);
2895 /* 1 for the orphan item */
2896 trans = btrfs_start_transaction(root, 1);
2897 if (IS_ERR(trans)) {
2898 btrfs_free_path(path);
2899 root->fs_info->enospc_unlink = 0;
2903 path->skip_locking = 1;
2904 path->search_commit_root = 1;
2906 ret = btrfs_lookup_inode(trans, root, path,
2907 &BTRFS_I(dir)->location, 0);
2913 if (check_path_shared(root, path))
2918 btrfs_release_path(path);
2920 ret = btrfs_lookup_inode(trans, root, path,
2921 &BTRFS_I(inode)->location, 0);
2927 if (check_path_shared(root, path))
2932 btrfs_release_path(path);
2934 if (ret == 0 && S_ISREG(inode->i_mode)) {
2935 ret = btrfs_lookup_file_extent(trans, root, path,
2941 BUG_ON(ret == 0); /* Corruption */
2942 if (check_path_shared(root, path))
2944 btrfs_release_path(path);
2952 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2953 dentry->d_name.name, dentry->d_name.len, 0);
2959 if (check_path_shared(root, path))
2965 btrfs_release_path(path);
2967 ref = btrfs_lookup_inode_ref(trans, root, path,
2968 dentry->d_name.name, dentry->d_name.len,
2974 BUG_ON(!ref); /* Logic error */
2975 if (check_path_shared(root, path))
2977 index = btrfs_inode_ref_index(path->nodes[0], ref);
2978 btrfs_release_path(path);
2981 * This is a commit root search, if we can lookup inode item and other
2982 * relative items in the commit root, it means the transaction of
2983 * dir/file creation has been committed, and the dir index item that we
2984 * delay to insert has also been inserted into the commit root. So
2985 * we needn't worry about the delayed insertion of the dir index item
2988 di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2989 dentry->d_name.name, dentry->d_name.len, 0);
2994 BUG_ON(ret == -ENOENT);
2995 if (check_path_shared(root, path))
3000 btrfs_free_path(path);
3001 /* Migrate the orphan reservation over */
3003 err = btrfs_block_rsv_migrate(trans->block_rsv,
3004 &root->fs_info->global_block_rsv,
3005 trans->bytes_reserved);
3008 btrfs_end_transaction(trans, root);
3009 root->fs_info->enospc_unlink = 0;
3010 return ERR_PTR(err);
3013 trans->block_rsv = &root->fs_info->global_block_rsv;
3017 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3018 struct btrfs_root *root)
3020 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
3021 btrfs_block_rsv_release(root, trans->block_rsv,
3022 trans->bytes_reserved);
3023 trans->block_rsv = &root->fs_info->trans_block_rsv;
3024 BUG_ON(!root->fs_info->enospc_unlink);
3025 root->fs_info->enospc_unlink = 0;
3027 btrfs_end_transaction(trans, root);
3030 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3032 struct btrfs_root *root = BTRFS_I(dir)->root;
3033 struct btrfs_trans_handle *trans;
3034 struct inode *inode = dentry->d_inode;
3036 unsigned long nr = 0;
3038 trans = __unlink_start_trans(dir, dentry);
3040 return PTR_ERR(trans);
3042 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3044 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3045 dentry->d_name.name, dentry->d_name.len);
3049 if (inode->i_nlink == 0) {
3050 ret = btrfs_orphan_add(trans, inode);
3056 nr = trans->blocks_used;
3057 __unlink_end_trans(trans, root);
3058 btrfs_btree_balance_dirty(root, nr);
3062 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3063 struct btrfs_root *root,
3064 struct inode *dir, u64 objectid,
3065 const char *name, int name_len)
3067 struct btrfs_path *path;
3068 struct extent_buffer *leaf;
3069 struct btrfs_dir_item *di;
3070 struct btrfs_key key;
3073 u64 dir_ino = btrfs_ino(dir);
3075 path = btrfs_alloc_path();
3079 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3080 name, name_len, -1);
3081 if (IS_ERR_OR_NULL(di)) {
3089 leaf = path->nodes[0];
3090 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3091 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3092 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3094 btrfs_abort_transaction(trans, root, ret);
3097 btrfs_release_path(path);
3099 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3100 objectid, root->root_key.objectid,
3101 dir_ino, &index, name, name_len);
3103 if (ret != -ENOENT) {
3104 btrfs_abort_transaction(trans, root, ret);
3107 di = btrfs_search_dir_index_item(root, path, dir_ino,
3109 if (IS_ERR_OR_NULL(di)) {
3114 btrfs_abort_transaction(trans, root, ret);
3118 leaf = path->nodes[0];
3119 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3120 btrfs_release_path(path);
3123 btrfs_release_path(path);
3125 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3127 btrfs_abort_transaction(trans, root, ret);
3131 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3132 inode_inc_iversion(dir);
3133 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3134 ret = btrfs_update_inode(trans, root, dir);
3136 btrfs_abort_transaction(trans, root, ret);
3138 btrfs_free_path(path);
3142 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3144 struct inode *inode = dentry->d_inode;
3146 struct btrfs_root *root = BTRFS_I(dir)->root;
3147 struct btrfs_trans_handle *trans;
3148 unsigned long nr = 0;
3150 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3151 btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3154 trans = __unlink_start_trans(dir, dentry);
3156 return PTR_ERR(trans);
3158 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3159 err = btrfs_unlink_subvol(trans, root, dir,
3160 BTRFS_I(inode)->location.objectid,
3161 dentry->d_name.name,
3162 dentry->d_name.len);
3166 err = btrfs_orphan_add(trans, inode);
3170 /* now the directory is empty */
3171 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3172 dentry->d_name.name, dentry->d_name.len);
3174 btrfs_i_size_write(inode, 0);
3176 nr = trans->blocks_used;
3177 __unlink_end_trans(trans, root);
3178 btrfs_btree_balance_dirty(root, nr);
3184 * this can truncate away extent items, csum items and directory items.
3185 * It starts at a high offset and removes keys until it can't find
3186 * any higher than new_size
3188 * csum items that cross the new i_size are truncated to the new size
3191 * min_type is the minimum key type to truncate down to. If set to 0, this
3192 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3194 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3195 struct btrfs_root *root,
3196 struct inode *inode,
3197 u64 new_size, u32 min_type)
3199 struct btrfs_path *path;
3200 struct extent_buffer *leaf;
3201 struct btrfs_file_extent_item *fi;
3202 struct btrfs_key key;
3203 struct btrfs_key found_key;
3204 u64 extent_start = 0;
3205 u64 extent_num_bytes = 0;
3206 u64 extent_offset = 0;
3208 u64 mask = root->sectorsize - 1;
3209 u32 found_type = (u8)-1;
3212 int pending_del_nr = 0;
3213 int pending_del_slot = 0;
3214 int extent_type = -1;
3217 u64 ino = btrfs_ino(inode);
3219 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3221 path = btrfs_alloc_path();
3226 if (root->ref_cows || root == root->fs_info->tree_root)
3227 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3230 * This function is also used to drop the items in the log tree before
3231 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3232 * it is used to drop the loged items. So we shouldn't kill the delayed
3235 if (min_type == 0 && root == BTRFS_I(inode)->root)
3236 btrfs_kill_delayed_inode_items(inode);
3239 key.offset = (u64)-1;
3243 path->leave_spinning = 1;
3244 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3251 /* there are no items in the tree for us to truncate, we're
3254 if (path->slots[0] == 0)
3261 leaf = path->nodes[0];
3262 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3263 found_type = btrfs_key_type(&found_key);
3265 if (found_key.objectid != ino)
3268 if (found_type < min_type)
3271 item_end = found_key.offset;
3272 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3273 fi = btrfs_item_ptr(leaf, path->slots[0],
3274 struct btrfs_file_extent_item);
3275 extent_type = btrfs_file_extent_type(leaf, fi);
3276 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3278 btrfs_file_extent_num_bytes(leaf, fi);
3279 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3280 item_end += btrfs_file_extent_inline_len(leaf,
3285 if (found_type > min_type) {
3288 if (item_end < new_size)
3290 if (found_key.offset >= new_size)
3296 /* FIXME, shrink the extent if the ref count is only 1 */
3297 if (found_type != BTRFS_EXTENT_DATA_KEY)
3300 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3302 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3304 u64 orig_num_bytes =
3305 btrfs_file_extent_num_bytes(leaf, fi);
3306 extent_num_bytes = new_size -
3307 found_key.offset + root->sectorsize - 1;
3308 extent_num_bytes = extent_num_bytes &
3309 ~((u64)root->sectorsize - 1);
3310 btrfs_set_file_extent_num_bytes(leaf, fi,
3312 num_dec = (orig_num_bytes -
3314 if (root->ref_cows && extent_start != 0)
3315 inode_sub_bytes(inode, num_dec);
3316 btrfs_mark_buffer_dirty(leaf);
3319 btrfs_file_extent_disk_num_bytes(leaf,
3321 extent_offset = found_key.offset -
3322 btrfs_file_extent_offset(leaf, fi);
3324 /* FIXME blocksize != 4096 */
3325 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3326 if (extent_start != 0) {
3329 inode_sub_bytes(inode, num_dec);
3332 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3334 * we can't truncate inline items that have had
3338 btrfs_file_extent_compression(leaf, fi) == 0 &&
3339 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3340 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3341 u32 size = new_size - found_key.offset;
3343 if (root->ref_cows) {
3344 inode_sub_bytes(inode, item_end + 1 -
3348 btrfs_file_extent_calc_inline_size(size);
3349 btrfs_truncate_item(trans, root, path,
3351 } else if (root->ref_cows) {
3352 inode_sub_bytes(inode, item_end + 1 -
3358 if (!pending_del_nr) {
3359 /* no pending yet, add ourselves */
3360 pending_del_slot = path->slots[0];
3362 } else if (pending_del_nr &&
3363 path->slots[0] + 1 == pending_del_slot) {
3364 /* hop on the pending chunk */
3366 pending_del_slot = path->slots[0];
3373 if (found_extent && (root->ref_cows ||
3374 root == root->fs_info->tree_root)) {
3375 btrfs_set_path_blocking(path);
3376 ret = btrfs_free_extent(trans, root, extent_start,
3377 extent_num_bytes, 0,
3378 btrfs_header_owner(leaf),
3379 ino, extent_offset, 0);
3383 if (found_type == BTRFS_INODE_ITEM_KEY)
3386 if (path->slots[0] == 0 ||
3387 path->slots[0] != pending_del_slot) {
3388 if (root->ref_cows &&
3389 BTRFS_I(inode)->location.objectid !=
3390 BTRFS_FREE_INO_OBJECTID) {
3394 if (pending_del_nr) {
3395 ret = btrfs_del_items(trans, root, path,
3399 btrfs_abort_transaction(trans,
3405 btrfs_release_path(path);
3412 if (pending_del_nr) {
3413 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3416 btrfs_abort_transaction(trans, root, ret);
3419 btrfs_free_path(path);
3424 * taken from block_truncate_page, but does cow as it zeros out
3425 * any bytes left in the last page in the file.
3427 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3429 struct inode *inode = mapping->host;
3430 struct btrfs_root *root = BTRFS_I(inode)->root;
3431 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3432 struct btrfs_ordered_extent *ordered;
3433 struct extent_state *cached_state = NULL;
3435 u32 blocksize = root->sectorsize;
3436 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3437 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3439 gfp_t mask = btrfs_alloc_write_mask(mapping);
3444 if ((offset & (blocksize - 1)) == 0)
3446 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3452 page = find_or_create_page(mapping, index, mask);
3454 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3458 page_start = page_offset(page);
3459 page_end = page_start + PAGE_CACHE_SIZE - 1;
3461 if (!PageUptodate(page)) {
3462 ret = btrfs_readpage(NULL, page);
3464 if (page->mapping != mapping) {
3466 page_cache_release(page);
3469 if (!PageUptodate(page)) {
3474 wait_on_page_writeback(page);
3476 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3477 set_page_extent_mapped(page);
3479 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3481 unlock_extent_cached(io_tree, page_start, page_end,
3482 &cached_state, GFP_NOFS);
3484 page_cache_release(page);
3485 btrfs_start_ordered_extent(inode, ordered, 1);
3486 btrfs_put_ordered_extent(ordered);
3490 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3491 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3492 0, 0, &cached_state, GFP_NOFS);
3494 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3497 unlock_extent_cached(io_tree, page_start, page_end,
3498 &cached_state, GFP_NOFS);
3503 if (offset != PAGE_CACHE_SIZE) {
3505 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3506 flush_dcache_page(page);
3509 ClearPageChecked(page);
3510 set_page_dirty(page);
3511 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3516 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3518 page_cache_release(page);
3524 * This function puts in dummy file extents for the area we're creating a hole
3525 * for. So if we are truncating this file to a larger size we need to insert
3526 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3527 * the range between oldsize and size
3529 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3531 struct btrfs_trans_handle *trans;
3532 struct btrfs_root *root = BTRFS_I(inode)->root;
3533 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3534 struct extent_map *em = NULL;
3535 struct extent_state *cached_state = NULL;
3536 u64 mask = root->sectorsize - 1;
3537 u64 hole_start = (oldsize + mask) & ~mask;
3538 u64 block_end = (size + mask) & ~mask;
3544 if (size <= hole_start)
3548 struct btrfs_ordered_extent *ordered;
3549 btrfs_wait_ordered_range(inode, hole_start,
3550 block_end - hole_start);
3551 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3553 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3556 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3557 &cached_state, GFP_NOFS);
3558 btrfs_put_ordered_extent(ordered);
3561 cur_offset = hole_start;
3563 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3564 block_end - cur_offset, 0);
3569 last_byte = min(extent_map_end(em), block_end);
3570 last_byte = (last_byte + mask) & ~mask;
3571 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3573 hole_size = last_byte - cur_offset;
3575 trans = btrfs_start_transaction(root, 3);
3576 if (IS_ERR(trans)) {
3577 err = PTR_ERR(trans);
3581 err = btrfs_drop_extents(trans, inode, cur_offset,
3582 cur_offset + hole_size,
3585 btrfs_abort_transaction(trans, root, err);
3586 btrfs_end_transaction(trans, root);
3590 err = btrfs_insert_file_extent(trans, root,
3591 btrfs_ino(inode), cur_offset, 0,
3592 0, hole_size, 0, hole_size,
3595 btrfs_abort_transaction(trans, root, err);
3596 btrfs_end_transaction(trans, root);
3600 btrfs_drop_extent_cache(inode, hole_start,
3603 btrfs_update_inode(trans, root, inode);
3604 btrfs_end_transaction(trans, root);
3606 free_extent_map(em);
3608 cur_offset = last_byte;
3609 if (cur_offset >= block_end)
3613 free_extent_map(em);
3614 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3619 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3621 struct btrfs_root *root = BTRFS_I(inode)->root;
3622 struct btrfs_trans_handle *trans;
3623 loff_t oldsize = i_size_read(inode);
3626 if (newsize == oldsize)
3629 if (newsize > oldsize) {
3630 truncate_pagecache(inode, oldsize, newsize);
3631 ret = btrfs_cont_expand(inode, oldsize, newsize);
3635 trans = btrfs_start_transaction(root, 1);
3637 return PTR_ERR(trans);
3639 i_size_write(inode, newsize);
3640 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3641 ret = btrfs_update_inode(trans, root, inode);
3642 btrfs_end_transaction(trans, root);
3646 * We're truncating a file that used to have good data down to
3647 * zero. Make sure it gets into the ordered flush list so that
3648 * any new writes get down to disk quickly.
3651 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3652 &BTRFS_I(inode)->runtime_flags);
3654 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3655 truncate_setsize(inode, newsize);
3656 ret = btrfs_truncate(inode);
3662 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3664 struct inode *inode = dentry->d_inode;
3665 struct btrfs_root *root = BTRFS_I(inode)->root;
3668 if (btrfs_root_readonly(root))
3671 err = inode_change_ok(inode, attr);
3675 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3676 err = btrfs_setsize(inode, attr->ia_size);
3681 if (attr->ia_valid) {
3682 setattr_copy(inode, attr);
3683 inode_inc_iversion(inode);
3684 err = btrfs_dirty_inode(inode);
3686 if (!err && attr->ia_valid & ATTR_MODE)
3687 err = btrfs_acl_chmod(inode);
3693 void btrfs_evict_inode(struct inode *inode)
3695 struct btrfs_trans_handle *trans;
3696 struct btrfs_root *root = BTRFS_I(inode)->root;
3697 struct btrfs_block_rsv *rsv, *global_rsv;
3698 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3702 trace_btrfs_inode_evict(inode);
3704 truncate_inode_pages(&inode->i_data, 0);
3705 if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3706 btrfs_is_free_space_inode(root, inode)))
3709 if (is_bad_inode(inode)) {
3710 btrfs_orphan_del(NULL, inode);
3713 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3714 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3716 if (root->fs_info->log_root_recovering) {
3717 BUG_ON(!test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3718 &BTRFS_I(inode)->runtime_flags));
3722 if (inode->i_nlink > 0) {
3723 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3727 rsv = btrfs_alloc_block_rsv(root);
3729 btrfs_orphan_del(NULL, inode);
3732 rsv->size = min_size;
3733 global_rsv = &root->fs_info->global_block_rsv;
3735 btrfs_i_size_write(inode, 0);
3738 * This is a bit simpler than btrfs_truncate since
3740 * 1) We've already reserved our space for our orphan item in the
3742 * 2) We're going to delete the inode item, so we don't need to update
3745 * So we just need to reserve some slack space in case we add bytes when
3746 * doing the truncate.
3749 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3752 * Try and steal from the global reserve since we will
3753 * likely not use this space anyway, we want to try as
3754 * hard as possible to get this to work.
3757 ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3760 printk(KERN_WARNING "Could not get space for a "
3761 "delete, will truncate on mount %d\n", ret);
3762 btrfs_orphan_del(NULL, inode);
3763 btrfs_free_block_rsv(root, rsv);
3767 trans = btrfs_start_transaction(root, 0);
3768 if (IS_ERR(trans)) {
3769 btrfs_orphan_del(NULL, inode);
3770 btrfs_free_block_rsv(root, rsv);
3774 trans->block_rsv = rsv;
3776 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3780 nr = trans->blocks_used;
3781 btrfs_end_transaction(trans, root);
3783 btrfs_btree_balance_dirty(root, nr);
3786 btrfs_free_block_rsv(root, rsv);
3789 trans->block_rsv = root->orphan_block_rsv;
3790 ret = btrfs_orphan_del(trans, inode);
3794 trans->block_rsv = &root->fs_info->trans_block_rsv;
3795 if (!(root == root->fs_info->tree_root ||
3796 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3797 btrfs_return_ino(root, btrfs_ino(inode));
3799 nr = trans->blocks_used;
3800 btrfs_end_transaction(trans, root);
3801 btrfs_btree_balance_dirty(root, nr);
3803 end_writeback(inode);
3808 * this returns the key found in the dir entry in the location pointer.
3809 * If no dir entries were found, location->objectid is 0.
3811 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3812 struct btrfs_key *location)
3814 const char *name = dentry->d_name.name;
3815 int namelen = dentry->d_name.len;
3816 struct btrfs_dir_item *di;
3817 struct btrfs_path *path;
3818 struct btrfs_root *root = BTRFS_I(dir)->root;
3821 path = btrfs_alloc_path();
3825 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3830 if (IS_ERR_OR_NULL(di))
3833 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3835 btrfs_free_path(path);
3838 location->objectid = 0;
3843 * when we hit a tree root in a directory, the btrfs part of the inode
3844 * needs to be changed to reflect the root directory of the tree root. This
3845 * is kind of like crossing a mount point.
3847 static int fixup_tree_root_location(struct btrfs_root *root,
3849 struct dentry *dentry,
3850 struct btrfs_key *location,
3851 struct btrfs_root **sub_root)
3853 struct btrfs_path *path;
3854 struct btrfs_root *new_root;
3855 struct btrfs_root_ref *ref;
3856 struct extent_buffer *leaf;
3860 path = btrfs_alloc_path();
3867 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3868 BTRFS_I(dir)->root->root_key.objectid,
3869 location->objectid);
3876 leaf = path->nodes[0];
3877 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3878 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3879 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3882 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3883 (unsigned long)(ref + 1),
3884 dentry->d_name.len);
3888 btrfs_release_path(path);
3890 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3891 if (IS_ERR(new_root)) {
3892 err = PTR_ERR(new_root);
3896 if (btrfs_root_refs(&new_root->root_item) == 0) {
3901 *sub_root = new_root;
3902 location->objectid = btrfs_root_dirid(&new_root->root_item);
3903 location->type = BTRFS_INODE_ITEM_KEY;
3904 location->offset = 0;
3907 btrfs_free_path(path);
3911 static void inode_tree_add(struct inode *inode)
3913 struct btrfs_root *root = BTRFS_I(inode)->root;
3914 struct btrfs_inode *entry;
3916 struct rb_node *parent;
3917 u64 ino = btrfs_ino(inode);
3919 p = &root->inode_tree.rb_node;
3922 if (inode_unhashed(inode))
3925 spin_lock(&root->inode_lock);
3928 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3930 if (ino < btrfs_ino(&entry->vfs_inode))
3931 p = &parent->rb_left;
3932 else if (ino > btrfs_ino(&entry->vfs_inode))
3933 p = &parent->rb_right;
3935 WARN_ON(!(entry->vfs_inode.i_state &
3936 (I_WILL_FREE | I_FREEING)));
3937 rb_erase(parent, &root->inode_tree);
3938 RB_CLEAR_NODE(parent);
3939 spin_unlock(&root->inode_lock);
3943 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3944 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3945 spin_unlock(&root->inode_lock);
3948 static void inode_tree_del(struct inode *inode)
3950 struct btrfs_root *root = BTRFS_I(inode)->root;
3953 spin_lock(&root->inode_lock);
3954 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3955 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3956 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3957 empty = RB_EMPTY_ROOT(&root->inode_tree);
3959 spin_unlock(&root->inode_lock);
3962 * Free space cache has inodes in the tree root, but the tree root has a
3963 * root_refs of 0, so this could end up dropping the tree root as a
3964 * snapshot, so we need the extra !root->fs_info->tree_root check to
3965 * make sure we don't drop it.
3967 if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3968 root != root->fs_info->tree_root) {
3969 synchronize_srcu(&root->fs_info->subvol_srcu);
3970 spin_lock(&root->inode_lock);
3971 empty = RB_EMPTY_ROOT(&root->inode_tree);
3972 spin_unlock(&root->inode_lock);
3974 btrfs_add_dead_root(root);
3978 void btrfs_invalidate_inodes(struct btrfs_root *root)
3980 struct rb_node *node;
3981 struct rb_node *prev;
3982 struct btrfs_inode *entry;
3983 struct inode *inode;
3986 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3988 spin_lock(&root->inode_lock);
3990 node = root->inode_tree.rb_node;
3994 entry = rb_entry(node, struct btrfs_inode, rb_node);
3996 if (objectid < btrfs_ino(&entry->vfs_inode))
3997 node = node->rb_left;
3998 else if (objectid > btrfs_ino(&entry->vfs_inode))
3999 node = node->rb_right;
4005 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4006 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4010 prev = rb_next(prev);
4014 entry = rb_entry(node, struct btrfs_inode, rb_node);
4015 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4016 inode = igrab(&entry->vfs_inode);
4018 spin_unlock(&root->inode_lock);
4019 if (atomic_read(&inode->i_count) > 1)
4020 d_prune_aliases(inode);
4022 * btrfs_drop_inode will have it removed from
4023 * the inode cache when its usage count
4028 spin_lock(&root->inode_lock);
4032 if (cond_resched_lock(&root->inode_lock))
4035 node = rb_next(node);
4037 spin_unlock(&root->inode_lock);
4040 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4042 struct btrfs_iget_args *args = p;
4043 inode->i_ino = args->ino;
4044 BTRFS_I(inode)->root = args->root;
4045 btrfs_set_inode_space_info(args->root, inode);
4049 static int btrfs_find_actor(struct inode *inode, void *opaque)
4051 struct btrfs_iget_args *args = opaque;
4052 return args->ino == btrfs_ino(inode) &&
4053 args->root == BTRFS_I(inode)->root;
4056 static struct inode *btrfs_iget_locked(struct super_block *s,
4058 struct btrfs_root *root)
4060 struct inode *inode;
4061 struct btrfs_iget_args args;
4062 args.ino = objectid;
4065 inode = iget5_locked(s, objectid, btrfs_find_actor,
4066 btrfs_init_locked_inode,
4071 /* Get an inode object given its location and corresponding root.
4072 * Returns in *is_new if the inode was read from disk
4074 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4075 struct btrfs_root *root, int *new)
4077 struct inode *inode;
4079 inode = btrfs_iget_locked(s, location->objectid, root);
4081 return ERR_PTR(-ENOMEM);
4083 if (inode->i_state & I_NEW) {
4084 BTRFS_I(inode)->root = root;
4085 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4086 btrfs_read_locked_inode(inode);
4087 if (!is_bad_inode(inode)) {
4088 inode_tree_add(inode);
4089 unlock_new_inode(inode);
4093 unlock_new_inode(inode);
4095 inode = ERR_PTR(-ESTALE);
4102 static struct inode *new_simple_dir(struct super_block *s,
4103 struct btrfs_key *key,
4104 struct btrfs_root *root)
4106 struct inode *inode = new_inode(s);
4109 return ERR_PTR(-ENOMEM);
4111 BTRFS_I(inode)->root = root;
4112 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4113 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4115 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4116 inode->i_op = &btrfs_dir_ro_inode_operations;
4117 inode->i_fop = &simple_dir_operations;
4118 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4119 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4124 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4126 struct inode *inode;
4127 struct btrfs_root *root = BTRFS_I(dir)->root;
4128 struct btrfs_root *sub_root = root;
4129 struct btrfs_key location;
4133 if (dentry->d_name.len > BTRFS_NAME_LEN)
4134 return ERR_PTR(-ENAMETOOLONG);
4136 if (unlikely(d_need_lookup(dentry))) {
4137 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4138 kfree(dentry->d_fsdata);
4139 dentry->d_fsdata = NULL;
4140 /* This thing is hashed, drop it for now */
4143 ret = btrfs_inode_by_name(dir, dentry, &location);
4147 return ERR_PTR(ret);
4149 if (location.objectid == 0)
4152 if (location.type == BTRFS_INODE_ITEM_KEY) {
4153 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4157 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4159 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4160 ret = fixup_tree_root_location(root, dir, dentry,
4161 &location, &sub_root);
4164 inode = ERR_PTR(ret);
4166 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4168 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4170 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4172 if (!IS_ERR(inode) && root != sub_root) {
4173 down_read(&root->fs_info->cleanup_work_sem);
4174 if (!(inode->i_sb->s_flags & MS_RDONLY))
4175 ret = btrfs_orphan_cleanup(sub_root);
4176 up_read(&root->fs_info->cleanup_work_sem);
4178 inode = ERR_PTR(ret);
4184 static int btrfs_dentry_delete(const struct dentry *dentry)
4186 struct btrfs_root *root;
4187 struct inode *inode = dentry->d_inode;
4189 if (!inode && !IS_ROOT(dentry))
4190 inode = dentry->d_parent->d_inode;
4193 root = BTRFS_I(inode)->root;
4194 if (btrfs_root_refs(&root->root_item) == 0)
4197 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4203 static void btrfs_dentry_release(struct dentry *dentry)
4205 if (dentry->d_fsdata)
4206 kfree(dentry->d_fsdata);
4209 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4210 struct nameidata *nd)
4214 ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4215 if (unlikely(d_need_lookup(dentry))) {
4216 spin_lock(&dentry->d_lock);
4217 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4218 spin_unlock(&dentry->d_lock);
4223 unsigned char btrfs_filetype_table[] = {
4224 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4227 static int btrfs_real_readdir(struct file *filp, void *dirent,
4230 struct inode *inode = filp->f_dentry->d_inode;
4231 struct btrfs_root *root = BTRFS_I(inode)->root;
4232 struct btrfs_item *item;
4233 struct btrfs_dir_item *di;
4234 struct btrfs_key key;
4235 struct btrfs_key found_key;
4236 struct btrfs_path *path;
4237 struct list_head ins_list;
4238 struct list_head del_list;
4240 struct extent_buffer *leaf;
4242 unsigned char d_type;
4247 int key_type = BTRFS_DIR_INDEX_KEY;
4251 int is_curr = 0; /* filp->f_pos points to the current index? */
4253 /* FIXME, use a real flag for deciding about the key type */
4254 if (root->fs_info->tree_root == root)
4255 key_type = BTRFS_DIR_ITEM_KEY;
4257 /* special case for "." */
4258 if (filp->f_pos == 0) {
4259 over = filldir(dirent, ".", 1,
4260 filp->f_pos, btrfs_ino(inode), DT_DIR);
4265 /* special case for .., just use the back ref */
4266 if (filp->f_pos == 1) {
4267 u64 pino = parent_ino(filp->f_path.dentry);
4268 over = filldir(dirent, "..", 2,
4269 filp->f_pos, pino, DT_DIR);
4274 path = btrfs_alloc_path();
4280 if (key_type == BTRFS_DIR_INDEX_KEY) {
4281 INIT_LIST_HEAD(&ins_list);
4282 INIT_LIST_HEAD(&del_list);
4283 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4286 btrfs_set_key_type(&key, key_type);
4287 key.offset = filp->f_pos;
4288 key.objectid = btrfs_ino(inode);
4290 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4295 leaf = path->nodes[0];
4296 slot = path->slots[0];
4297 if (slot >= btrfs_header_nritems(leaf)) {
4298 ret = btrfs_next_leaf(root, path);
4306 item = btrfs_item_nr(leaf, slot);
4307 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4309 if (found_key.objectid != key.objectid)
4311 if (btrfs_key_type(&found_key) != key_type)
4313 if (found_key.offset < filp->f_pos)
4315 if (key_type == BTRFS_DIR_INDEX_KEY &&
4316 btrfs_should_delete_dir_index(&del_list,
4320 filp->f_pos = found_key.offset;
4323 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4325 di_total = btrfs_item_size(leaf, item);
4327 while (di_cur < di_total) {
4328 struct btrfs_key location;
4330 if (verify_dir_item(root, leaf, di))
4333 name_len = btrfs_dir_name_len(leaf, di);
4334 if (name_len <= sizeof(tmp_name)) {
4335 name_ptr = tmp_name;
4337 name_ptr = kmalloc(name_len, GFP_NOFS);
4343 read_extent_buffer(leaf, name_ptr,
4344 (unsigned long)(di + 1), name_len);
4346 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4347 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4350 /* is this a reference to our own snapshot? If so
4353 * In contrast to old kernels, we insert the snapshot's
4354 * dir item and dir index after it has been created, so
4355 * we won't find a reference to our own snapshot. We
4356 * still keep the following code for backward
4359 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4360 location.objectid == root->root_key.objectid) {
4364 over = filldir(dirent, name_ptr, name_len,
4365 found_key.offset, location.objectid,
4369 if (name_ptr != tmp_name)
4374 di_len = btrfs_dir_name_len(leaf, di) +
4375 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4377 di = (struct btrfs_dir_item *)((char *)di + di_len);
4383 if (key_type == BTRFS_DIR_INDEX_KEY) {
4386 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4392 /* Reached end of directory/root. Bump pos past the last item. */
4393 if (key_type == BTRFS_DIR_INDEX_KEY)
4395 * 32-bit glibc will use getdents64, but then strtol -
4396 * so the last number we can serve is this.
4398 filp->f_pos = 0x7fffffff;
4404 if (key_type == BTRFS_DIR_INDEX_KEY)
4405 btrfs_put_delayed_items(&ins_list, &del_list);
4406 btrfs_free_path(path);
4410 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4412 struct btrfs_root *root = BTRFS_I(inode)->root;
4413 struct btrfs_trans_handle *trans;
4415 bool nolock = false;
4417 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4420 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4423 if (wbc->sync_mode == WB_SYNC_ALL) {
4425 trans = btrfs_join_transaction_nolock(root);
4427 trans = btrfs_join_transaction(root);
4429 return PTR_ERR(trans);
4431 ret = btrfs_end_transaction_nolock(trans, root);
4433 ret = btrfs_commit_transaction(trans, root);
4439 * This is somewhat expensive, updating the tree every time the
4440 * inode changes. But, it is most likely to find the inode in cache.
4441 * FIXME, needs more benchmarking...there are no reasons other than performance
4442 * to keep or drop this code.
4444 int btrfs_dirty_inode(struct inode *inode)
4446 struct btrfs_root *root = BTRFS_I(inode)->root;
4447 struct btrfs_trans_handle *trans;
4450 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4453 trans = btrfs_join_transaction(root);
4455 return PTR_ERR(trans);
4457 ret = btrfs_update_inode(trans, root, inode);
4458 if (ret && ret == -ENOSPC) {
4459 /* whoops, lets try again with the full transaction */
4460 btrfs_end_transaction(trans, root);
4461 trans = btrfs_start_transaction(root, 1);
4463 return PTR_ERR(trans);
4465 ret = btrfs_update_inode(trans, root, inode);
4467 btrfs_end_transaction(trans, root);
4468 if (BTRFS_I(inode)->delayed_node)
4469 btrfs_balance_delayed_items(root);
4475 * This is a copy of file_update_time. We need this so we can return error on
4476 * ENOSPC for updating the inode in the case of file write and mmap writes.
4478 int btrfs_update_time(struct file *file)
4480 struct inode *inode = file->f_path.dentry->d_inode;
4481 struct timespec now;
4483 enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4485 /* First try to exhaust all avenues to not sync */
4486 if (IS_NOCMTIME(inode))
4489 now = current_fs_time(inode->i_sb);
4490 if (!timespec_equal(&inode->i_mtime, &now))
4493 if (!timespec_equal(&inode->i_ctime, &now))
4496 if (IS_I_VERSION(inode))
4497 sync_it |= S_VERSION;
4502 /* Finally allowed to write? Takes lock. */
4503 if (mnt_want_write_file(file))
4506 /* Only change inode inside the lock region */
4507 if (sync_it & S_VERSION)
4508 inode_inc_iversion(inode);
4509 if (sync_it & S_CTIME)
4510 inode->i_ctime = now;
4511 if (sync_it & S_MTIME)
4512 inode->i_mtime = now;
4513 ret = btrfs_dirty_inode(inode);
4515 mark_inode_dirty_sync(inode);
4516 mnt_drop_write(file->f_path.mnt);
4521 * find the highest existing sequence number in a directory
4522 * and then set the in-memory index_cnt variable to reflect
4523 * free sequence numbers
4525 static int btrfs_set_inode_index_count(struct inode *inode)
4527 struct btrfs_root *root = BTRFS_I(inode)->root;
4528 struct btrfs_key key, found_key;
4529 struct btrfs_path *path;
4530 struct extent_buffer *leaf;
4533 key.objectid = btrfs_ino(inode);
4534 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4535 key.offset = (u64)-1;
4537 path = btrfs_alloc_path();
4541 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4544 /* FIXME: we should be able to handle this */
4550 * MAGIC NUMBER EXPLANATION:
4551 * since we search a directory based on f_pos we have to start at 2
4552 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4553 * else has to start at 2
4555 if (path->slots[0] == 0) {
4556 BTRFS_I(inode)->index_cnt = 2;
4562 leaf = path->nodes[0];
4563 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4565 if (found_key.objectid != btrfs_ino(inode) ||
4566 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4567 BTRFS_I(inode)->index_cnt = 2;
4571 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4573 btrfs_free_path(path);
4578 * helper to find a free sequence number in a given directory. This current
4579 * code is very simple, later versions will do smarter things in the btree
4581 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4585 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4586 ret = btrfs_inode_delayed_dir_index_count(dir);
4588 ret = btrfs_set_inode_index_count(dir);
4594 *index = BTRFS_I(dir)->index_cnt;
4595 BTRFS_I(dir)->index_cnt++;
4600 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4601 struct btrfs_root *root,
4603 const char *name, int name_len,
4604 u64 ref_objectid, u64 objectid,
4605 umode_t mode, u64 *index)
4607 struct inode *inode;
4608 struct btrfs_inode_item *inode_item;
4609 struct btrfs_key *location;
4610 struct btrfs_path *path;
4611 struct btrfs_inode_ref *ref;
4612 struct btrfs_key key[2];
4618 path = btrfs_alloc_path();
4620 return ERR_PTR(-ENOMEM);
4622 inode = new_inode(root->fs_info->sb);
4624 btrfs_free_path(path);
4625 return ERR_PTR(-ENOMEM);
4629 * we have to initialize this early, so we can reclaim the inode
4630 * number if we fail afterwards in this function.
4632 inode->i_ino = objectid;
4635 trace_btrfs_inode_request(dir);
4637 ret = btrfs_set_inode_index(dir, index);
4639 btrfs_free_path(path);
4641 return ERR_PTR(ret);
4645 * index_cnt is ignored for everything but a dir,
4646 * btrfs_get_inode_index_count has an explanation for the magic
4649 BTRFS_I(inode)->index_cnt = 2;
4650 BTRFS_I(inode)->root = root;
4651 BTRFS_I(inode)->generation = trans->transid;
4652 inode->i_generation = BTRFS_I(inode)->generation;
4653 btrfs_set_inode_space_info(root, inode);
4660 key[0].objectid = objectid;
4661 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4664 key[1].objectid = objectid;
4665 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4666 key[1].offset = ref_objectid;
4668 sizes[0] = sizeof(struct btrfs_inode_item);
4669 sizes[1] = name_len + sizeof(*ref);
4671 path->leave_spinning = 1;
4672 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4676 inode_init_owner(inode, dir, mode);
4677 inode_set_bytes(inode, 0);
4678 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4679 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4680 struct btrfs_inode_item);
4681 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4683 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4684 struct btrfs_inode_ref);
4685 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4686 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4687 ptr = (unsigned long)(ref + 1);
4688 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4690 btrfs_mark_buffer_dirty(path->nodes[0]);
4691 btrfs_free_path(path);
4693 location = &BTRFS_I(inode)->location;
4694 location->objectid = objectid;
4695 location->offset = 0;
4696 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4698 btrfs_inherit_iflags(inode, dir);
4700 if (S_ISREG(mode)) {
4701 if (btrfs_test_opt(root, NODATASUM))
4702 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4703 if (btrfs_test_opt(root, NODATACOW) ||
4704 (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4705 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4708 insert_inode_hash(inode);
4709 inode_tree_add(inode);
4711 trace_btrfs_inode_new(inode);
4712 btrfs_set_inode_last_trans(trans, inode);
4717 BTRFS_I(dir)->index_cnt--;
4718 btrfs_free_path(path);
4720 return ERR_PTR(ret);
4723 static inline u8 btrfs_inode_type(struct inode *inode)
4725 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4729 * utility function to add 'inode' into 'parent_inode' with
4730 * a give name and a given sequence number.
4731 * if 'add_backref' is true, also insert a backref from the
4732 * inode to the parent directory.
4734 int btrfs_add_link(struct btrfs_trans_handle *trans,
4735 struct inode *parent_inode, struct inode *inode,
4736 const char *name, int name_len, int add_backref, u64 index)
4739 struct btrfs_key key;
4740 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4741 u64 ino = btrfs_ino(inode);
4742 u64 parent_ino = btrfs_ino(parent_inode);
4744 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4745 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4748 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4752 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4753 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4754 key.objectid, root->root_key.objectid,
4755 parent_ino, index, name, name_len);
4756 } else if (add_backref) {
4757 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4761 /* Nothing to clean up yet */
4765 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4767 btrfs_inode_type(inode), index);
4771 btrfs_abort_transaction(trans, root, ret);
4775 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4777 inode_inc_iversion(parent_inode);
4778 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4779 ret = btrfs_update_inode(trans, root, parent_inode);
4781 btrfs_abort_transaction(trans, root, ret);
4785 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4788 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4789 key.objectid, root->root_key.objectid,
4790 parent_ino, &local_index, name, name_len);
4792 } else if (add_backref) {
4796 err = btrfs_del_inode_ref(trans, root, name, name_len,
4797 ino, parent_ino, &local_index);
4802 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4803 struct inode *dir, struct dentry *dentry,
4804 struct inode *inode, int backref, u64 index)
4806 int err = btrfs_add_link(trans, dir, inode,
4807 dentry->d_name.name, dentry->d_name.len,
4814 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4815 umode_t mode, dev_t rdev)
4817 struct btrfs_trans_handle *trans;
4818 struct btrfs_root *root = BTRFS_I(dir)->root;
4819 struct inode *inode = NULL;
4823 unsigned long nr = 0;
4826 if (!new_valid_dev(rdev))
4830 * 2 for inode item and ref
4832 * 1 for xattr if selinux is on
4834 trans = btrfs_start_transaction(root, 5);
4836 return PTR_ERR(trans);
4838 err = btrfs_find_free_ino(root, &objectid);
4842 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4843 dentry->d_name.len, btrfs_ino(dir), objectid,
4845 if (IS_ERR(inode)) {
4846 err = PTR_ERR(inode);
4850 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4857 * If the active LSM wants to access the inode during
4858 * d_instantiate it needs these. Smack checks to see
4859 * if the filesystem supports xattrs by looking at the
4863 inode->i_op = &btrfs_special_inode_operations;
4864 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4868 init_special_inode(inode, inode->i_mode, rdev);
4869 btrfs_update_inode(trans, root, inode);
4870 d_instantiate(dentry, inode);
4873 nr = trans->blocks_used;
4874 btrfs_end_transaction(trans, root);
4875 btrfs_btree_balance_dirty(root, nr);
4877 inode_dec_link_count(inode);
4883 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4884 umode_t mode, struct nameidata *nd)
4886 struct btrfs_trans_handle *trans;
4887 struct btrfs_root *root = BTRFS_I(dir)->root;
4888 struct inode *inode = NULL;
4891 unsigned long nr = 0;
4896 * 2 for inode item and ref
4898 * 1 for xattr if selinux is on
4900 trans = btrfs_start_transaction(root, 5);
4902 return PTR_ERR(trans);
4904 err = btrfs_find_free_ino(root, &objectid);
4908 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4909 dentry->d_name.len, btrfs_ino(dir), objectid,
4911 if (IS_ERR(inode)) {
4912 err = PTR_ERR(inode);
4916 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4923 * If the active LSM wants to access the inode during
4924 * d_instantiate it needs these. Smack checks to see
4925 * if the filesystem supports xattrs by looking at the
4928 inode->i_fop = &btrfs_file_operations;
4929 inode->i_op = &btrfs_file_inode_operations;
4931 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4935 inode->i_mapping->a_ops = &btrfs_aops;
4936 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4937 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4938 d_instantiate(dentry, inode);
4941 nr = trans->blocks_used;
4942 btrfs_end_transaction(trans, root);
4944 inode_dec_link_count(inode);
4947 btrfs_btree_balance_dirty(root, nr);
4951 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4952 struct dentry *dentry)
4954 struct btrfs_trans_handle *trans;
4955 struct btrfs_root *root = BTRFS_I(dir)->root;
4956 struct inode *inode = old_dentry->d_inode;
4958 unsigned long nr = 0;
4962 /* do not allow sys_link's with other subvols of the same device */
4963 if (root->objectid != BTRFS_I(inode)->root->objectid)
4966 if (inode->i_nlink == ~0U)
4969 err = btrfs_set_inode_index(dir, &index);
4974 * 2 items for inode and inode ref
4975 * 2 items for dir items
4976 * 1 item for parent inode
4978 trans = btrfs_start_transaction(root, 5);
4979 if (IS_ERR(trans)) {
4980 err = PTR_ERR(trans);
4984 btrfs_inc_nlink(inode);
4985 inode_inc_iversion(inode);
4986 inode->i_ctime = CURRENT_TIME;
4989 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4994 struct dentry *parent = dentry->d_parent;
4995 err = btrfs_update_inode(trans, root, inode);
4998 d_instantiate(dentry, inode);
4999 btrfs_log_new_name(trans, inode, NULL, parent);
5002 nr = trans->blocks_used;
5003 btrfs_end_transaction(trans, root);
5006 inode_dec_link_count(inode);
5009 btrfs_btree_balance_dirty(root, nr);
5013 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5015 struct inode *inode = NULL;
5016 struct btrfs_trans_handle *trans;
5017 struct btrfs_root *root = BTRFS_I(dir)->root;
5019 int drop_on_err = 0;
5022 unsigned long nr = 1;
5025 * 2 items for inode and ref
5026 * 2 items for dir items
5027 * 1 for xattr if selinux is on
5029 trans = btrfs_start_transaction(root, 5);
5031 return PTR_ERR(trans);
5033 err = btrfs_find_free_ino(root, &objectid);
5037 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5038 dentry->d_name.len, btrfs_ino(dir), objectid,
5039 S_IFDIR | mode, &index);
5040 if (IS_ERR(inode)) {
5041 err = PTR_ERR(inode);
5047 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5051 inode->i_op = &btrfs_dir_inode_operations;
5052 inode->i_fop = &btrfs_dir_file_operations;
5054 btrfs_i_size_write(inode, 0);
5055 err = btrfs_update_inode(trans, root, inode);
5059 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5060 dentry->d_name.len, 0, index);
5064 d_instantiate(dentry, inode);
5068 nr = trans->blocks_used;
5069 btrfs_end_transaction(trans, root);
5072 btrfs_btree_balance_dirty(root, nr);
5076 /* helper for btfs_get_extent. Given an existing extent in the tree,
5077 * and an extent that you want to insert, deal with overlap and insert
5078 * the new extent into the tree.
5080 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5081 struct extent_map *existing,
5082 struct extent_map *em,
5083 u64 map_start, u64 map_len)
5087 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5088 start_diff = map_start - em->start;
5089 em->start = map_start;
5091 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5092 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5093 em->block_start += start_diff;
5094 em->block_len -= start_diff;
5096 return add_extent_mapping(em_tree, em);
5099 static noinline int uncompress_inline(struct btrfs_path *path,
5100 struct inode *inode, struct page *page,
5101 size_t pg_offset, u64 extent_offset,
5102 struct btrfs_file_extent_item *item)
5105 struct extent_buffer *leaf = path->nodes[0];
5108 unsigned long inline_size;
5112 WARN_ON(pg_offset != 0);
5113 compress_type = btrfs_file_extent_compression(leaf, item);
5114 max_size = btrfs_file_extent_ram_bytes(leaf, item);
5115 inline_size = btrfs_file_extent_inline_item_len(leaf,
5116 btrfs_item_nr(leaf, path->slots[0]));
5117 tmp = kmalloc(inline_size, GFP_NOFS);
5120 ptr = btrfs_file_extent_inline_start(item);
5122 read_extent_buffer(leaf, tmp, ptr, inline_size);
5124 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5125 ret = btrfs_decompress(compress_type, tmp, page,
5126 extent_offset, inline_size, max_size);
5128 char *kaddr = kmap_atomic(page);
5129 unsigned long copy_size = min_t(u64,
5130 PAGE_CACHE_SIZE - pg_offset,
5131 max_size - extent_offset);
5132 memset(kaddr + pg_offset, 0, copy_size);
5133 kunmap_atomic(kaddr);
5140 * a bit scary, this does extent mapping from logical file offset to the disk.
5141 * the ugly parts come from merging extents from the disk with the in-ram
5142 * representation. This gets more complex because of the data=ordered code,
5143 * where the in-ram extents might be locked pending data=ordered completion.
5145 * This also copies inline extents directly into the page.
5148 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5149 size_t pg_offset, u64 start, u64 len,
5155 u64 extent_start = 0;
5157 u64 objectid = btrfs_ino(inode);
5159 struct btrfs_path *path = NULL;
5160 struct btrfs_root *root = BTRFS_I(inode)->root;
5161 struct btrfs_file_extent_item *item;
5162 struct extent_buffer *leaf;
5163 struct btrfs_key found_key;
5164 struct extent_map *em = NULL;
5165 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5166 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5167 struct btrfs_trans_handle *trans = NULL;
5171 read_lock(&em_tree->lock);
5172 em = lookup_extent_mapping(em_tree, start, len);
5174 em->bdev = root->fs_info->fs_devices->latest_bdev;
5175 read_unlock(&em_tree->lock);
5178 if (em->start > start || em->start + em->len <= start)
5179 free_extent_map(em);
5180 else if (em->block_start == EXTENT_MAP_INLINE && page)
5181 free_extent_map(em);
5185 em = alloc_extent_map();
5190 em->bdev = root->fs_info->fs_devices->latest_bdev;
5191 em->start = EXTENT_MAP_HOLE;
5192 em->orig_start = EXTENT_MAP_HOLE;
5194 em->block_len = (u64)-1;
5197 path = btrfs_alloc_path();
5203 * Chances are we'll be called again, so go ahead and do
5209 ret = btrfs_lookup_file_extent(trans, root, path,
5210 objectid, start, trans != NULL);
5217 if (path->slots[0] == 0)
5222 leaf = path->nodes[0];
5223 item = btrfs_item_ptr(leaf, path->slots[0],
5224 struct btrfs_file_extent_item);
5225 /* are we inside the extent that was found? */
5226 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5227 found_type = btrfs_key_type(&found_key);
5228 if (found_key.objectid != objectid ||
5229 found_type != BTRFS_EXTENT_DATA_KEY) {
5233 found_type = btrfs_file_extent_type(leaf, item);
5234 extent_start = found_key.offset;
5235 compress_type = btrfs_file_extent_compression(leaf, item);
5236 if (found_type == BTRFS_FILE_EXTENT_REG ||
5237 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5238 extent_end = extent_start +
5239 btrfs_file_extent_num_bytes(leaf, item);
5240 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5242 size = btrfs_file_extent_inline_len(leaf, item);
5243 extent_end = (extent_start + size + root->sectorsize - 1) &
5244 ~((u64)root->sectorsize - 1);
5247 if (start >= extent_end) {
5249 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5250 ret = btrfs_next_leaf(root, path);
5257 leaf = path->nodes[0];
5259 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5260 if (found_key.objectid != objectid ||
5261 found_key.type != BTRFS_EXTENT_DATA_KEY)
5263 if (start + len <= found_key.offset)
5266 em->len = found_key.offset - start;
5270 if (found_type == BTRFS_FILE_EXTENT_REG ||
5271 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5272 em->start = extent_start;
5273 em->len = extent_end - extent_start;
5274 em->orig_start = extent_start -
5275 btrfs_file_extent_offset(leaf, item);
5276 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5278 em->block_start = EXTENT_MAP_HOLE;
5281 if (compress_type != BTRFS_COMPRESS_NONE) {
5282 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5283 em->compress_type = compress_type;
5284 em->block_start = bytenr;
5285 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5288 bytenr += btrfs_file_extent_offset(leaf, item);
5289 em->block_start = bytenr;
5290 em->block_len = em->len;
5291 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5292 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5295 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5299 size_t extent_offset;
5302 em->block_start = EXTENT_MAP_INLINE;
5303 if (!page || create) {
5304 em->start = extent_start;
5305 em->len = extent_end - extent_start;
5309 size = btrfs_file_extent_inline_len(leaf, item);
5310 extent_offset = page_offset(page) + pg_offset - extent_start;
5311 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5312 size - extent_offset);
5313 em->start = extent_start + extent_offset;
5314 em->len = (copy_size + root->sectorsize - 1) &
5315 ~((u64)root->sectorsize - 1);
5316 em->orig_start = EXTENT_MAP_INLINE;
5317 if (compress_type) {
5318 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5319 em->compress_type = compress_type;
5321 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5322 if (create == 0 && !PageUptodate(page)) {
5323 if (btrfs_file_extent_compression(leaf, item) !=
5324 BTRFS_COMPRESS_NONE) {
5325 ret = uncompress_inline(path, inode, page,
5327 extent_offset, item);
5328 BUG_ON(ret); /* -ENOMEM */
5331 read_extent_buffer(leaf, map + pg_offset, ptr,
5333 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5334 memset(map + pg_offset + copy_size, 0,
5335 PAGE_CACHE_SIZE - pg_offset -
5340 flush_dcache_page(page);
5341 } else if (create && PageUptodate(page)) {
5345 free_extent_map(em);
5348 btrfs_release_path(path);
5349 trans = btrfs_join_transaction(root);
5352 return ERR_CAST(trans);
5356 write_extent_buffer(leaf, map + pg_offset, ptr,
5359 btrfs_mark_buffer_dirty(leaf);
5361 set_extent_uptodate(io_tree, em->start,
5362 extent_map_end(em) - 1, NULL, GFP_NOFS);
5365 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5372 em->block_start = EXTENT_MAP_HOLE;
5373 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5375 btrfs_release_path(path);
5376 if (em->start > start || extent_map_end(em) <= start) {
5377 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5378 "[%llu %llu]\n", (unsigned long long)em->start,
5379 (unsigned long long)em->len,
5380 (unsigned long long)start,
5381 (unsigned long long)len);
5387 write_lock(&em_tree->lock);
5388 ret = add_extent_mapping(em_tree, em);
5389 /* it is possible that someone inserted the extent into the tree
5390 * while we had the lock dropped. It is also possible that
5391 * an overlapping map exists in the tree
5393 if (ret == -EEXIST) {
5394 struct extent_map *existing;
5398 existing = lookup_extent_mapping(em_tree, start, len);
5399 if (existing && (existing->start > start ||
5400 existing->start + existing->len <= start)) {
5401 free_extent_map(existing);
5405 existing = lookup_extent_mapping(em_tree, em->start,
5408 err = merge_extent_mapping(em_tree, existing,
5411 free_extent_map(existing);
5413 free_extent_map(em);
5418 free_extent_map(em);
5422 free_extent_map(em);
5427 write_unlock(&em_tree->lock);
5430 trace_btrfs_get_extent(root, em);
5433 btrfs_free_path(path);
5435 ret = btrfs_end_transaction(trans, root);
5440 free_extent_map(em);
5441 return ERR_PTR(err);
5443 BUG_ON(!em); /* Error is always set */
5447 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5448 size_t pg_offset, u64 start, u64 len,
5451 struct extent_map *em;
5452 struct extent_map *hole_em = NULL;
5453 u64 range_start = start;
5459 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5464 * if our em maps to a hole, there might
5465 * actually be delalloc bytes behind it
5467 if (em->block_start != EXTENT_MAP_HOLE)
5473 /* check to see if we've wrapped (len == -1 or similar) */
5482 /* ok, we didn't find anything, lets look for delalloc */
5483 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5484 end, len, EXTENT_DELALLOC, 1);
5485 found_end = range_start + found;
5486 if (found_end < range_start)
5487 found_end = (u64)-1;
5490 * we didn't find anything useful, return
5491 * the original results from get_extent()
5493 if (range_start > end || found_end <= start) {
5499 /* adjust the range_start to make sure it doesn't
5500 * go backwards from the start they passed in
5502 range_start = max(start,range_start);
5503 found = found_end - range_start;
5506 u64 hole_start = start;
5509 em = alloc_extent_map();
5515 * when btrfs_get_extent can't find anything it
5516 * returns one huge hole
5518 * make sure what it found really fits our range, and
5519 * adjust to make sure it is based on the start from
5523 u64 calc_end = extent_map_end(hole_em);
5525 if (calc_end <= start || (hole_em->start > end)) {
5526 free_extent_map(hole_em);
5529 hole_start = max(hole_em->start, start);
5530 hole_len = calc_end - hole_start;
5534 if (hole_em && range_start > hole_start) {
5535 /* our hole starts before our delalloc, so we
5536 * have to return just the parts of the hole
5537 * that go until the delalloc starts
5539 em->len = min(hole_len,
5540 range_start - hole_start);
5541 em->start = hole_start;
5542 em->orig_start = hole_start;
5544 * don't adjust block start at all,
5545 * it is fixed at EXTENT_MAP_HOLE
5547 em->block_start = hole_em->block_start;
5548 em->block_len = hole_len;
5550 em->start = range_start;
5552 em->orig_start = range_start;
5553 em->block_start = EXTENT_MAP_DELALLOC;
5554 em->block_len = found;
5556 } else if (hole_em) {
5561 free_extent_map(hole_em);
5563 free_extent_map(em);
5564 return ERR_PTR(err);
5569 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5570 struct extent_map *em,
5573 struct btrfs_root *root = BTRFS_I(inode)->root;
5574 struct btrfs_trans_handle *trans;
5575 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5576 struct btrfs_key ins;
5579 bool insert = false;
5582 * Ok if the extent map we looked up is a hole and is for the exact
5583 * range we want, there is no reason to allocate a new one, however if
5584 * it is not right then we need to free this one and drop the cache for
5587 if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5589 free_extent_map(em);
5592 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5595 trans = btrfs_join_transaction(root);
5597 return ERR_CAST(trans);
5599 if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5600 btrfs_add_inode_defrag(trans, inode);
5602 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5604 alloc_hint = get_extent_allocation_hint(inode, start, len);
5605 ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5606 alloc_hint, &ins, 1);
5613 em = alloc_extent_map();
5615 em = ERR_PTR(-ENOMEM);
5621 em->orig_start = em->start;
5622 em->len = ins.offset;
5624 em->block_start = ins.objectid;
5625 em->block_len = ins.offset;
5626 em->bdev = root->fs_info->fs_devices->latest_bdev;
5629 * We need to do this because if we're using the original em we searched
5630 * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5633 set_bit(EXTENT_FLAG_PINNED, &em->flags);
5636 write_lock(&em_tree->lock);
5637 ret = add_extent_mapping(em_tree, em);
5638 write_unlock(&em_tree->lock);
5641 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5644 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5645 ins.offset, ins.offset, 0);
5647 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5651 btrfs_end_transaction(trans, root);
5656 * returns 1 when the nocow is safe, < 1 on error, 0 if the
5657 * block must be cow'd
5659 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5660 struct inode *inode, u64 offset, u64 len)
5662 struct btrfs_path *path;
5664 struct extent_buffer *leaf;
5665 struct btrfs_root *root = BTRFS_I(inode)->root;
5666 struct btrfs_file_extent_item *fi;
5667 struct btrfs_key key;
5675 path = btrfs_alloc_path();
5679 ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5684 slot = path->slots[0];
5687 /* can't find the item, must cow */
5694 leaf = path->nodes[0];
5695 btrfs_item_key_to_cpu(leaf, &key, slot);
5696 if (key.objectid != btrfs_ino(inode) ||
5697 key.type != BTRFS_EXTENT_DATA_KEY) {
5698 /* not our file or wrong item type, must cow */
5702 if (key.offset > offset) {
5703 /* Wrong offset, must cow */
5707 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5708 found_type = btrfs_file_extent_type(leaf, fi);
5709 if (found_type != BTRFS_FILE_EXTENT_REG &&
5710 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5711 /* not a regular extent, must cow */
5714 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5715 backref_offset = btrfs_file_extent_offset(leaf, fi);
5717 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5718 if (extent_end < offset + len) {
5719 /* extent doesn't include our full range, must cow */
5723 if (btrfs_extent_readonly(root, disk_bytenr))
5727 * look for other files referencing this extent, if we
5728 * find any we must cow
5730 if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5731 key.offset - backref_offset, disk_bytenr))
5735 * adjust disk_bytenr and num_bytes to cover just the bytes
5736 * in this extent we are about to write. If there
5737 * are any csums in that range we have to cow in order
5738 * to keep the csums correct
5740 disk_bytenr += backref_offset;
5741 disk_bytenr += offset - key.offset;
5742 num_bytes = min(offset + len, extent_end) - offset;
5743 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5746 * all of the above have passed, it is safe to overwrite this extent
5751 btrfs_free_path(path);
5755 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5756 struct buffer_head *bh_result, int create)
5758 struct extent_map *em;
5759 struct btrfs_root *root = BTRFS_I(inode)->root;
5760 u64 start = iblock << inode->i_blkbits;
5761 u64 len = bh_result->b_size;
5762 struct btrfs_trans_handle *trans;
5764 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5769 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5770 * io. INLINE is special, and we could probably kludge it in here, but
5771 * it's still buffered so for safety lets just fall back to the generic
5774 * For COMPRESSED we _have_ to read the entire extent in so we can
5775 * decompress it, so there will be buffering required no matter what we
5776 * do, so go ahead and fallback to buffered.
5778 * We return -ENOTBLK because thats what makes DIO go ahead and go back
5779 * to buffered IO. Don't blame me, this is the price we pay for using
5782 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5783 em->block_start == EXTENT_MAP_INLINE) {
5784 free_extent_map(em);
5788 /* Just a good old fashioned hole, return */
5789 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5790 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5791 free_extent_map(em);
5792 /* DIO will do one hole at a time, so just unlock a sector */
5793 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5794 start + root->sectorsize - 1);
5799 * We don't allocate a new extent in the following cases
5801 * 1) The inode is marked as NODATACOW. In this case we'll just use the
5803 * 2) The extent is marked as PREALLOC. We're good to go here and can
5804 * just use the extent.
5808 len = em->len - (start - em->start);
5812 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5813 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5814 em->block_start != EXTENT_MAP_HOLE)) {
5819 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5820 type = BTRFS_ORDERED_PREALLOC;
5822 type = BTRFS_ORDERED_NOCOW;
5823 len = min(len, em->len - (start - em->start));
5824 block_start = em->block_start + (start - em->start);
5827 * we're not going to log anything, but we do need
5828 * to make sure the current transaction stays open
5829 * while we look for nocow cross refs
5831 trans = btrfs_join_transaction(root);
5835 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5836 ret = btrfs_add_ordered_extent_dio(inode, start,
5837 block_start, len, len, type);
5838 btrfs_end_transaction(trans, root);
5840 free_extent_map(em);
5845 btrfs_end_transaction(trans, root);
5849 * this will cow the extent, reset the len in case we changed
5852 len = bh_result->b_size;
5853 em = btrfs_new_extent_direct(inode, em, start, len);
5856 len = min(len, em->len - (start - em->start));
5858 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5859 EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5862 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5864 bh_result->b_size = len;
5865 bh_result->b_bdev = em->bdev;
5866 set_buffer_mapped(bh_result);
5867 if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5868 set_buffer_new(bh_result);
5870 free_extent_map(em);
5875 struct btrfs_dio_private {
5876 struct inode *inode;
5883 /* number of bios pending for this dio */
5884 atomic_t pending_bios;
5889 struct bio *orig_bio;
5892 static void btrfs_endio_direct_read(struct bio *bio, int err)
5894 struct btrfs_dio_private *dip = bio->bi_private;
5895 struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5896 struct bio_vec *bvec = bio->bi_io_vec;
5897 struct inode *inode = dip->inode;
5898 struct btrfs_root *root = BTRFS_I(inode)->root;
5900 u32 *private = dip->csums;
5902 start = dip->logical_offset;
5904 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5905 struct page *page = bvec->bv_page;
5908 unsigned long flags;
5910 local_irq_save(flags);
5911 kaddr = kmap_atomic(page);
5912 csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5913 csum, bvec->bv_len);
5914 btrfs_csum_final(csum, (char *)&csum);
5915 kunmap_atomic(kaddr);
5916 local_irq_restore(flags);
5918 flush_dcache_page(bvec->bv_page);
5919 if (csum != *private) {
5920 printk(KERN_ERR "btrfs csum failed ino %llu off"
5921 " %llu csum %u private %u\n",
5922 (unsigned long long)btrfs_ino(inode),
5923 (unsigned long long)start,
5929 start += bvec->bv_len;
5932 } while (bvec <= bvec_end);
5934 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5935 dip->logical_offset + dip->bytes - 1);
5936 bio->bi_private = dip->private;
5941 /* If we had a csum failure make sure to clear the uptodate flag */
5943 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5944 dio_end_io(bio, err);
5947 static void btrfs_endio_direct_write(struct bio *bio, int err)
5949 struct btrfs_dio_private *dip = bio->bi_private;
5950 struct inode *inode = dip->inode;
5951 struct btrfs_root *root = BTRFS_I(inode)->root;
5952 struct btrfs_ordered_extent *ordered = NULL;
5953 u64 ordered_offset = dip->logical_offset;
5954 u64 ordered_bytes = dip->bytes;
5960 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5962 ordered_bytes, !err);
5966 ordered->work.func = finish_ordered_fn;
5967 ordered->work.flags = 0;
5968 btrfs_queue_worker(&root->fs_info->endio_write_workers,
5972 * our bio might span multiple ordered extents. If we haven't
5973 * completed the accounting for the whole dio, go back and try again
5975 if (ordered_offset < dip->logical_offset + dip->bytes) {
5976 ordered_bytes = dip->logical_offset + dip->bytes -
5982 bio->bi_private = dip->private;
5986 /* If we had an error make sure to clear the uptodate flag */
5988 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5989 dio_end_io(bio, err);
5992 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5993 struct bio *bio, int mirror_num,
5994 unsigned long bio_flags, u64 offset)
5997 struct btrfs_root *root = BTRFS_I(inode)->root;
5998 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5999 BUG_ON(ret); /* -ENOMEM */
6003 static void btrfs_end_dio_bio(struct bio *bio, int err)
6005 struct btrfs_dio_private *dip = bio->bi_private;
6008 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6009 "sector %#Lx len %u err no %d\n",
6010 (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6011 (unsigned long long)bio->bi_sector, bio->bi_size, err);
6015 * before atomic variable goto zero, we must make sure
6016 * dip->errors is perceived to be set.
6018 smp_mb__before_atomic_dec();
6021 /* if there are more bios still pending for this dio, just exit */
6022 if (!atomic_dec_and_test(&dip->pending_bios))
6026 bio_io_error(dip->orig_bio);
6028 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6029 bio_endio(dip->orig_bio, 0);
6035 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6036 u64 first_sector, gfp_t gfp_flags)
6038 int nr_vecs = bio_get_nr_vecs(bdev);
6039 return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6042 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6043 int rw, u64 file_offset, int skip_sum,
6044 u32 *csums, int async_submit)
6046 int write = rw & REQ_WRITE;
6047 struct btrfs_root *root = BTRFS_I(inode)->root;
6053 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6061 if (write && async_submit) {
6062 ret = btrfs_wq_submit_bio(root->fs_info,
6063 inode, rw, bio, 0, 0,
6065 __btrfs_submit_bio_start_direct_io,
6066 __btrfs_submit_bio_done);
6070 * If we aren't doing async submit, calculate the csum of the
6073 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6076 } else if (!skip_sum) {
6077 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6078 file_offset, csums);
6084 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6090 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6093 struct inode *inode = dip->inode;
6094 struct btrfs_root *root = BTRFS_I(inode)->root;
6095 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6097 struct bio *orig_bio = dip->orig_bio;
6098 struct bio_vec *bvec = orig_bio->bi_io_vec;
6099 u64 start_sector = orig_bio->bi_sector;
6100 u64 file_offset = dip->logical_offset;
6104 u32 *csums = dip->csums;
6106 int async_submit = 0;
6107 int write = rw & REQ_WRITE;
6109 map_length = orig_bio->bi_size;
6110 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6111 &map_length, NULL, 0);
6117 if (map_length >= orig_bio->bi_size) {
6123 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6126 bio->bi_private = dip;
6127 bio->bi_end_io = btrfs_end_dio_bio;
6128 atomic_inc(&dip->pending_bios);
6130 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6131 if (unlikely(map_length < submit_len + bvec->bv_len ||
6132 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6133 bvec->bv_offset) < bvec->bv_len)) {
6135 * inc the count before we submit the bio so
6136 * we know the end IO handler won't happen before
6137 * we inc the count. Otherwise, the dip might get freed
6138 * before we're done setting it up
6140 atomic_inc(&dip->pending_bios);
6141 ret = __btrfs_submit_dio_bio(bio, inode, rw,
6142 file_offset, skip_sum,
6143 csums, async_submit);
6146 atomic_dec(&dip->pending_bios);
6150 /* Write's use the ordered csums */
6151 if (!write && !skip_sum)
6152 csums = csums + nr_pages;
6153 start_sector += submit_len >> 9;
6154 file_offset += submit_len;
6159 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6160 start_sector, GFP_NOFS);
6163 bio->bi_private = dip;
6164 bio->bi_end_io = btrfs_end_dio_bio;
6166 map_length = orig_bio->bi_size;
6167 ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6168 &map_length, NULL, 0);
6174 submit_len += bvec->bv_len;
6181 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6182 csums, async_submit);
6190 * before atomic variable goto zero, we must
6191 * make sure dip->errors is perceived to be set.
6193 smp_mb__before_atomic_dec();
6194 if (atomic_dec_and_test(&dip->pending_bios))
6195 bio_io_error(dip->orig_bio);
6197 /* bio_end_io() will handle error, so we needn't return it */
6201 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6204 struct btrfs_root *root = BTRFS_I(inode)->root;
6205 struct btrfs_dio_private *dip;
6206 struct bio_vec *bvec = bio->bi_io_vec;
6208 int write = rw & REQ_WRITE;
6211 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6213 dip = kmalloc(sizeof(*dip), GFP_NOFS);
6220 /* Write's use the ordered csum stuff, so we don't need dip->csums */
6221 if (!write && !skip_sum) {
6222 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6230 dip->private = bio->bi_private;
6232 dip->logical_offset = file_offset;
6236 dip->bytes += bvec->bv_len;
6238 } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6240 dip->disk_bytenr = (u64)bio->bi_sector << 9;
6241 bio->bi_private = dip;
6243 dip->orig_bio = bio;
6244 atomic_set(&dip->pending_bios, 0);
6247 bio->bi_end_io = btrfs_endio_direct_write;
6249 bio->bi_end_io = btrfs_endio_direct_read;
6251 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6256 * If this is a write, we need to clean up the reserved space and kill
6257 * the ordered extent.
6260 struct btrfs_ordered_extent *ordered;
6261 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6262 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6263 !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6264 btrfs_free_reserved_extent(root, ordered->start,
6266 btrfs_put_ordered_extent(ordered);
6267 btrfs_put_ordered_extent(ordered);
6269 bio_endio(bio, ret);
6272 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6273 const struct iovec *iov, loff_t offset,
6274 unsigned long nr_segs)
6280 unsigned blocksize_mask = root->sectorsize - 1;
6281 ssize_t retval = -EINVAL;
6282 loff_t end = offset;
6284 if (offset & blocksize_mask)
6287 /* Check the memory alignment. Blocks cannot straddle pages */
6288 for (seg = 0; seg < nr_segs; seg++) {
6289 addr = (unsigned long)iov[seg].iov_base;
6290 size = iov[seg].iov_len;
6292 if ((addr & blocksize_mask) || (size & blocksize_mask))
6295 /* If this is a write we don't need to check anymore */
6300 * Check to make sure we don't have duplicate iov_base's in this
6301 * iovec, if so return EINVAL, otherwise we'll get csum errors
6302 * when reading back.
6304 for (i = seg + 1; i < nr_segs; i++) {
6305 if (iov[seg].iov_base == iov[i].iov_base)
6313 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6314 const struct iovec *iov, loff_t offset,
6315 unsigned long nr_segs)
6317 struct file *file = iocb->ki_filp;
6318 struct inode *inode = file->f_mapping->host;
6319 struct btrfs_ordered_extent *ordered;
6320 struct extent_state *cached_state = NULL;
6321 u64 lockstart, lockend;
6323 int writing = rw & WRITE;
6325 size_t count = iov_length(iov, nr_segs);
6327 if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6333 lockend = offset + count - 1;
6336 ret = btrfs_delalloc_reserve_space(inode, count);
6342 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6345 * We're concerned with the entire range that we're going to be
6346 * doing DIO to, so we need to make sure theres no ordered
6347 * extents in this range.
6349 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6350 lockend - lockstart + 1);
6353 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6354 &cached_state, GFP_NOFS);
6355 btrfs_start_ordered_extent(inode, ordered, 1);
6356 btrfs_put_ordered_extent(ordered);
6361 * we don't use btrfs_set_extent_delalloc because we don't want
6362 * the dirty or uptodate bits
6365 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6366 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6367 EXTENT_DELALLOC, NULL, &cached_state,
6370 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6371 lockend, EXTENT_LOCKED | write_bits,
6372 1, 0, &cached_state, GFP_NOFS);
6377 free_extent_state(cached_state);
6378 cached_state = NULL;
6380 ret = __blockdev_direct_IO(rw, iocb, inode,
6381 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6382 iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6383 btrfs_submit_direct, 0);
6385 if (ret < 0 && ret != -EIOCBQUEUED) {
6386 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6387 offset + iov_length(iov, nr_segs) - 1,
6388 EXTENT_LOCKED | write_bits, 1, 0,
6389 &cached_state, GFP_NOFS);
6390 } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6392 * We're falling back to buffered, unlock the section we didn't
6395 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6396 offset + iov_length(iov, nr_segs) - 1,
6397 EXTENT_LOCKED | write_bits, 1, 0,
6398 &cached_state, GFP_NOFS);
6401 free_extent_state(cached_state);
6405 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6406 __u64 start, __u64 len)
6408 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6411 int btrfs_readpage(struct file *file, struct page *page)
6413 struct extent_io_tree *tree;
6414 tree = &BTRFS_I(page->mapping->host)->io_tree;
6415 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6418 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6420 struct extent_io_tree *tree;
6423 if (current->flags & PF_MEMALLOC) {
6424 redirty_page_for_writepage(wbc, page);
6428 tree = &BTRFS_I(page->mapping->host)->io_tree;
6429 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6432 int btrfs_writepages(struct address_space *mapping,
6433 struct writeback_control *wbc)
6435 struct extent_io_tree *tree;
6437 tree = &BTRFS_I(mapping->host)->io_tree;
6438 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6442 btrfs_readpages(struct file *file, struct address_space *mapping,
6443 struct list_head *pages, unsigned nr_pages)
6445 struct extent_io_tree *tree;
6446 tree = &BTRFS_I(mapping->host)->io_tree;
6447 return extent_readpages(tree, mapping, pages, nr_pages,
6450 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6452 struct extent_io_tree *tree;
6453 struct extent_map_tree *map;
6456 tree = &BTRFS_I(page->mapping->host)->io_tree;
6457 map = &BTRFS_I(page->mapping->host)->extent_tree;
6458 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6460 ClearPagePrivate(page);
6461 set_page_private(page, 0);
6462 page_cache_release(page);
6467 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6469 if (PageWriteback(page) || PageDirty(page))
6471 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6474 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6476 struct inode *inode = page->mapping->host;
6477 struct extent_io_tree *tree;
6478 struct btrfs_ordered_extent *ordered;
6479 struct extent_state *cached_state = NULL;
6480 u64 page_start = page_offset(page);
6481 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6484 * we have the page locked, so new writeback can't start,
6485 * and the dirty bit won't be cleared while we are here.
6487 * Wait for IO on this page so that we can safely clear
6488 * the PagePrivate2 bit and do ordered accounting
6490 wait_on_page_writeback(page);
6492 tree = &BTRFS_I(inode)->io_tree;
6494 btrfs_releasepage(page, GFP_NOFS);
6497 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6498 ordered = btrfs_lookup_ordered_extent(inode,
6502 * IO on this page will never be started, so we need
6503 * to account for any ordered extents now
6505 clear_extent_bit(tree, page_start, page_end,
6506 EXTENT_DIRTY | EXTENT_DELALLOC |
6507 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6508 &cached_state, GFP_NOFS);
6510 * whoever cleared the private bit is responsible
6511 * for the finish_ordered_io
6513 if (TestClearPagePrivate2(page) &&
6514 btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6515 PAGE_CACHE_SIZE, 1)) {
6516 btrfs_finish_ordered_io(ordered);
6518 btrfs_put_ordered_extent(ordered);
6519 cached_state = NULL;
6520 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6522 clear_extent_bit(tree, page_start, page_end,
6523 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6524 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6525 __btrfs_releasepage(page, GFP_NOFS);
6527 ClearPageChecked(page);
6528 if (PagePrivate(page)) {
6529 ClearPagePrivate(page);
6530 set_page_private(page, 0);
6531 page_cache_release(page);
6536 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6537 * called from a page fault handler when a page is first dirtied. Hence we must
6538 * be careful to check for EOF conditions here. We set the page up correctly
6539 * for a written page which means we get ENOSPC checking when writing into
6540 * holes and correct delalloc and unwritten extent mapping on filesystems that
6541 * support these features.
6543 * We are not allowed to take the i_mutex here so we have to play games to
6544 * protect against truncate races as the page could now be beyond EOF. Because
6545 * vmtruncate() writes the inode size before removing pages, once we have the
6546 * page lock we can determine safely if the page is beyond EOF. If it is not
6547 * beyond EOF, then the page is guaranteed safe against truncation until we
6550 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6552 struct page *page = vmf->page;
6553 struct inode *inode = fdentry(vma->vm_file)->d_inode;
6554 struct btrfs_root *root = BTRFS_I(inode)->root;
6555 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6556 struct btrfs_ordered_extent *ordered;
6557 struct extent_state *cached_state = NULL;
6559 unsigned long zero_start;
6566 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6568 ret = btrfs_update_time(vma->vm_file);
6574 else /* -ENOSPC, -EIO, etc */
6575 ret = VM_FAULT_SIGBUS;
6581 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6584 size = i_size_read(inode);
6585 page_start = page_offset(page);
6586 page_end = page_start + PAGE_CACHE_SIZE - 1;
6588 if ((page->mapping != inode->i_mapping) ||
6589 (page_start >= size)) {
6590 /* page got truncated out from underneath us */
6593 wait_on_page_writeback(page);
6595 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6596 set_page_extent_mapped(page);
6599 * we can't set the delalloc bits if there are pending ordered
6600 * extents. Drop our locks and wait for them to finish
6602 ordered = btrfs_lookup_ordered_extent(inode, page_start);
6604 unlock_extent_cached(io_tree, page_start, page_end,
6605 &cached_state, GFP_NOFS);
6607 btrfs_start_ordered_extent(inode, ordered, 1);
6608 btrfs_put_ordered_extent(ordered);
6613 * XXX - page_mkwrite gets called every time the page is dirtied, even
6614 * if it was already dirty, so for space accounting reasons we need to
6615 * clear any delalloc bits for the range we are fixing to save. There
6616 * is probably a better way to do this, but for now keep consistent with
6617 * prepare_pages in the normal write path.
6619 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6620 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6621 0, 0, &cached_state, GFP_NOFS);
6623 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6626 unlock_extent_cached(io_tree, page_start, page_end,
6627 &cached_state, GFP_NOFS);
6628 ret = VM_FAULT_SIGBUS;
6633 /* page is wholly or partially inside EOF */
6634 if (page_start + PAGE_CACHE_SIZE > size)
6635 zero_start = size & ~PAGE_CACHE_MASK;
6637 zero_start = PAGE_CACHE_SIZE;
6639 if (zero_start != PAGE_CACHE_SIZE) {
6641 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6642 flush_dcache_page(page);
6645 ClearPageChecked(page);
6646 set_page_dirty(page);
6647 SetPageUptodate(page);
6649 BTRFS_I(inode)->last_trans = root->fs_info->generation;
6650 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6652 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6656 return VM_FAULT_LOCKED;
6659 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6664 static int btrfs_truncate(struct inode *inode)
6666 struct btrfs_root *root = BTRFS_I(inode)->root;
6667 struct btrfs_block_rsv *rsv;
6670 struct btrfs_trans_handle *trans;
6672 u64 mask = root->sectorsize - 1;
6673 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6675 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6679 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6680 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6683 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
6684 * 3 things going on here
6686 * 1) We need to reserve space for our orphan item and the space to
6687 * delete our orphan item. Lord knows we don't want to have a dangling
6688 * orphan item because we didn't reserve space to remove it.
6690 * 2) We need to reserve space to update our inode.
6692 * 3) We need to have something to cache all the space that is going to
6693 * be free'd up by the truncate operation, but also have some slack
6694 * space reserved in case it uses space during the truncate (thank you
6695 * very much snapshotting).
6697 * And we need these to all be seperate. The fact is we can use alot of
6698 * space doing the truncate, and we have no earthly idea how much space
6699 * we will use, so we need the truncate reservation to be seperate so it
6700 * doesn't end up using space reserved for updating the inode or
6701 * removing the orphan item. We also need to be able to stop the
6702 * transaction and start a new one, which means we need to be able to
6703 * update the inode several times, and we have no idea of knowing how
6704 * many times that will be, so we can't just reserve 1 item for the
6705 * entirety of the opration, so that has to be done seperately as well.
6706 * Then there is the orphan item, which does indeed need to be held on
6707 * to for the whole operation, and we need nobody to touch this reserved
6708 * space except the orphan code.
6710 * So that leaves us with
6712 * 1) root->orphan_block_rsv - for the orphan deletion.
6713 * 2) rsv - for the truncate reservation, which we will steal from the
6714 * transaction reservation.
6715 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6716 * updating the inode.
6718 rsv = btrfs_alloc_block_rsv(root);
6721 rsv->size = min_size;
6724 * 1 for the truncate slack space
6725 * 1 for the orphan item we're going to add
6726 * 1 for the orphan item deletion
6727 * 1 for updating the inode.
6729 trans = btrfs_start_transaction(root, 4);
6730 if (IS_ERR(trans)) {
6731 err = PTR_ERR(trans);
6735 /* Migrate the slack space for the truncate to our reserve */
6736 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6740 ret = btrfs_orphan_add(trans, inode);
6742 btrfs_end_transaction(trans, root);
6747 * setattr is responsible for setting the ordered_data_close flag,
6748 * but that is only tested during the last file release. That
6749 * could happen well after the next commit, leaving a great big
6750 * window where new writes may get lost if someone chooses to write
6751 * to this file after truncating to zero
6753 * The inode doesn't have any dirty data here, and so if we commit
6754 * this is a noop. If someone immediately starts writing to the inode
6755 * it is very likely we'll catch some of their writes in this
6756 * transaction, and the commit will find this file on the ordered
6757 * data list with good things to send down.
6759 * This is a best effort solution, there is still a window where
6760 * using truncate to replace the contents of the file will
6761 * end up with a zero length file after a crash.
6763 if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6764 &BTRFS_I(inode)->runtime_flags))
6765 btrfs_add_ordered_operation(trans, root, inode);
6768 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6771 * This can only happen with the original transaction we
6772 * started above, every other time we shouldn't have a
6773 * transaction started yet.
6782 /* Just need the 1 for updating the inode */
6783 trans = btrfs_start_transaction(root, 1);
6784 if (IS_ERR(trans)) {
6785 ret = err = PTR_ERR(trans);
6791 trans->block_rsv = rsv;
6793 ret = btrfs_truncate_inode_items(trans, root, inode,
6795 BTRFS_EXTENT_DATA_KEY);
6796 if (ret != -EAGAIN) {
6801 trans->block_rsv = &root->fs_info->trans_block_rsv;
6802 ret = btrfs_update_inode(trans, root, inode);
6808 nr = trans->blocks_used;
6809 btrfs_end_transaction(trans, root);
6811 btrfs_btree_balance_dirty(root, nr);
6814 if (ret == 0 && inode->i_nlink > 0) {
6815 trans->block_rsv = root->orphan_block_rsv;
6816 ret = btrfs_orphan_del(trans, inode);
6819 } else if (ret && inode->i_nlink > 0) {
6821 * Failed to do the truncate, remove us from the in memory
6824 ret = btrfs_orphan_del(NULL, inode);
6828 trans->block_rsv = &root->fs_info->trans_block_rsv;
6829 ret = btrfs_update_inode(trans, root, inode);
6833 nr = trans->blocks_used;
6834 ret = btrfs_end_transaction(trans, root);
6835 btrfs_btree_balance_dirty(root, nr);
6839 btrfs_free_block_rsv(root, rsv);
6848 * create a new subvolume directory/inode (helper for the ioctl).
6850 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6851 struct btrfs_root *new_root, u64 new_dirid)
6853 struct inode *inode;
6857 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6858 new_dirid, new_dirid,
6859 S_IFDIR | (~current_umask() & S_IRWXUGO),
6862 return PTR_ERR(inode);
6863 inode->i_op = &btrfs_dir_inode_operations;
6864 inode->i_fop = &btrfs_dir_file_operations;
6866 set_nlink(inode, 1);
6867 btrfs_i_size_write(inode, 0);
6869 err = btrfs_update_inode(trans, new_root, inode);
6875 struct inode *btrfs_alloc_inode(struct super_block *sb)
6877 struct btrfs_inode *ei;
6878 struct inode *inode;
6880 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6885 ei->space_info = NULL;
6888 ei->last_sub_trans = 0;
6889 ei->logged_trans = 0;
6890 ei->delalloc_bytes = 0;
6891 ei->disk_i_size = 0;
6894 ei->index_cnt = (u64)-1;
6895 ei->last_unlink_trans = 0;
6897 spin_lock_init(&ei->lock);
6898 ei->outstanding_extents = 0;
6899 ei->reserved_extents = 0;
6901 ei->runtime_flags = 0;
6902 ei->force_compress = BTRFS_COMPRESS_NONE;
6904 ei->delayed_node = NULL;
6906 inode = &ei->vfs_inode;
6907 extent_map_tree_init(&ei->extent_tree);
6908 extent_io_tree_init(&ei->io_tree, &inode->i_data);
6909 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6910 ei->io_tree.track_uptodate = 1;
6911 ei->io_failure_tree.track_uptodate = 1;
6912 mutex_init(&ei->log_mutex);
6913 mutex_init(&ei->delalloc_mutex);
6914 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6915 INIT_LIST_HEAD(&ei->delalloc_inodes);
6916 INIT_LIST_HEAD(&ei->ordered_operations);
6917 RB_CLEAR_NODE(&ei->rb_node);
6922 static void btrfs_i_callback(struct rcu_head *head)
6924 struct inode *inode = container_of(head, struct inode, i_rcu);
6925 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6928 void btrfs_destroy_inode(struct inode *inode)
6930 struct btrfs_ordered_extent *ordered;
6931 struct btrfs_root *root = BTRFS_I(inode)->root;
6933 WARN_ON(!list_empty(&inode->i_dentry));
6934 WARN_ON(inode->i_data.nrpages);
6935 WARN_ON(BTRFS_I(inode)->outstanding_extents);
6936 WARN_ON(BTRFS_I(inode)->reserved_extents);
6937 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6938 WARN_ON(BTRFS_I(inode)->csum_bytes);
6941 * This can happen where we create an inode, but somebody else also
6942 * created the same inode and we need to destroy the one we already
6949 * Make sure we're properly removed from the ordered operation
6953 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6954 spin_lock(&root->fs_info->ordered_extent_lock);
6955 list_del_init(&BTRFS_I(inode)->ordered_operations);
6956 spin_unlock(&root->fs_info->ordered_extent_lock);
6959 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
6960 &BTRFS_I(inode)->runtime_flags)) {
6961 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6962 (unsigned long long)btrfs_ino(inode));
6963 atomic_dec(&root->orphan_inodes);
6967 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6971 printk(KERN_ERR "btrfs found ordered "
6972 "extent %llu %llu on inode cleanup\n",
6973 (unsigned long long)ordered->file_offset,
6974 (unsigned long long)ordered->len);
6975 btrfs_remove_ordered_extent(inode, ordered);
6976 btrfs_put_ordered_extent(ordered);
6977 btrfs_put_ordered_extent(ordered);
6980 inode_tree_del(inode);
6981 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6983 btrfs_remove_delayed_node(inode);
6984 call_rcu(&inode->i_rcu, btrfs_i_callback);
6987 int btrfs_drop_inode(struct inode *inode)
6989 struct btrfs_root *root = BTRFS_I(inode)->root;
6991 if (btrfs_root_refs(&root->root_item) == 0 &&
6992 !btrfs_is_free_space_inode(root, inode))
6995 return generic_drop_inode(inode);
6998 static void init_once(void *foo)
7000 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7002 inode_init_once(&ei->vfs_inode);
7005 void btrfs_destroy_cachep(void)
7007 if (btrfs_inode_cachep)
7008 kmem_cache_destroy(btrfs_inode_cachep);
7009 if (btrfs_trans_handle_cachep)
7010 kmem_cache_destroy(btrfs_trans_handle_cachep);
7011 if (btrfs_transaction_cachep)
7012 kmem_cache_destroy(btrfs_transaction_cachep);
7013 if (btrfs_path_cachep)
7014 kmem_cache_destroy(btrfs_path_cachep);
7015 if (btrfs_free_space_cachep)
7016 kmem_cache_destroy(btrfs_free_space_cachep);
7019 int btrfs_init_cachep(void)
7021 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7022 sizeof(struct btrfs_inode), 0,
7023 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7024 if (!btrfs_inode_cachep)
7027 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7028 sizeof(struct btrfs_trans_handle), 0,
7029 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7030 if (!btrfs_trans_handle_cachep)
7033 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7034 sizeof(struct btrfs_transaction), 0,
7035 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7036 if (!btrfs_transaction_cachep)
7039 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7040 sizeof(struct btrfs_path), 0,
7041 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7042 if (!btrfs_path_cachep)
7045 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7046 sizeof(struct btrfs_free_space), 0,
7047 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7048 if (!btrfs_free_space_cachep)
7053 btrfs_destroy_cachep();
7057 static int btrfs_getattr(struct vfsmount *mnt,
7058 struct dentry *dentry, struct kstat *stat)
7060 struct inode *inode = dentry->d_inode;
7061 u32 blocksize = inode->i_sb->s_blocksize;
7063 generic_fillattr(inode, stat);
7064 stat->dev = BTRFS_I(inode)->root->anon_dev;
7065 stat->blksize = PAGE_CACHE_SIZE;
7066 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7067 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7072 * If a file is moved, it will inherit the cow and compression flags of the new
7075 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7077 struct btrfs_inode *b_dir = BTRFS_I(dir);
7078 struct btrfs_inode *b_inode = BTRFS_I(inode);
7080 if (b_dir->flags & BTRFS_INODE_NODATACOW)
7081 b_inode->flags |= BTRFS_INODE_NODATACOW;
7083 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7085 if (b_dir->flags & BTRFS_INODE_COMPRESS)
7086 b_inode->flags |= BTRFS_INODE_COMPRESS;
7088 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7091 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7092 struct inode *new_dir, struct dentry *new_dentry)
7094 struct btrfs_trans_handle *trans;
7095 struct btrfs_root *root = BTRFS_I(old_dir)->root;
7096 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7097 struct inode *new_inode = new_dentry->d_inode;
7098 struct inode *old_inode = old_dentry->d_inode;
7099 struct timespec ctime = CURRENT_TIME;
7103 u64 old_ino = btrfs_ino(old_inode);
7105 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7108 /* we only allow rename subvolume link between subvolumes */
7109 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7112 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7113 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7116 if (S_ISDIR(old_inode->i_mode) && new_inode &&
7117 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7120 * we're using rename to replace one file with another.
7121 * and the replacement file is large. Start IO on it now so
7122 * we don't add too much work to the end of the transaction
7124 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7125 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7126 filemap_flush(old_inode->i_mapping);
7128 /* close the racy window with snapshot create/destroy ioctl */
7129 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7130 down_read(&root->fs_info->subvol_sem);
7132 * We want to reserve the absolute worst case amount of items. So if
7133 * both inodes are subvols and we need to unlink them then that would
7134 * require 4 item modifications, but if they are both normal inodes it
7135 * would require 5 item modifications, so we'll assume their normal
7136 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7137 * should cover the worst case number of items we'll modify.
7139 trans = btrfs_start_transaction(root, 20);
7140 if (IS_ERR(trans)) {
7141 ret = PTR_ERR(trans);
7146 btrfs_record_root_in_trans(trans, dest);
7148 ret = btrfs_set_inode_index(new_dir, &index);
7152 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7153 /* force full log commit if subvolume involved. */
7154 root->fs_info->last_trans_log_full_commit = trans->transid;
7156 ret = btrfs_insert_inode_ref(trans, dest,
7157 new_dentry->d_name.name,
7158 new_dentry->d_name.len,
7160 btrfs_ino(new_dir), index);
7164 * this is an ugly little race, but the rename is required
7165 * to make sure that if we crash, the inode is either at the
7166 * old name or the new one. pinning the log transaction lets
7167 * us make sure we don't allow a log commit to come in after
7168 * we unlink the name but before we add the new name back in.
7170 btrfs_pin_log_trans(root);
7173 * make sure the inode gets flushed if it is replacing
7176 if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7177 btrfs_add_ordered_operation(trans, root, old_inode);
7179 inode_inc_iversion(old_dir);
7180 inode_inc_iversion(new_dir);
7181 inode_inc_iversion(old_inode);
7182 old_dir->i_ctime = old_dir->i_mtime = ctime;
7183 new_dir->i_ctime = new_dir->i_mtime = ctime;
7184 old_inode->i_ctime = ctime;
7186 if (old_dentry->d_parent != new_dentry->d_parent)
7187 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7189 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7190 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7191 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7192 old_dentry->d_name.name,
7193 old_dentry->d_name.len);
7195 ret = __btrfs_unlink_inode(trans, root, old_dir,
7196 old_dentry->d_inode,
7197 old_dentry->d_name.name,
7198 old_dentry->d_name.len);
7200 ret = btrfs_update_inode(trans, root, old_inode);
7203 btrfs_abort_transaction(trans, root, ret);
7208 inode_inc_iversion(new_inode);
7209 new_inode->i_ctime = CURRENT_TIME;
7210 if (unlikely(btrfs_ino(new_inode) ==
7211 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7212 root_objectid = BTRFS_I(new_inode)->location.objectid;
7213 ret = btrfs_unlink_subvol(trans, dest, new_dir,
7215 new_dentry->d_name.name,
7216 new_dentry->d_name.len);
7217 BUG_ON(new_inode->i_nlink == 0);
7219 ret = btrfs_unlink_inode(trans, dest, new_dir,
7220 new_dentry->d_inode,
7221 new_dentry->d_name.name,
7222 new_dentry->d_name.len);
7224 if (!ret && new_inode->i_nlink == 0) {
7225 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7229 btrfs_abort_transaction(trans, root, ret);
7234 fixup_inode_flags(new_dir, old_inode);
7236 ret = btrfs_add_link(trans, new_dir, old_inode,
7237 new_dentry->d_name.name,
7238 new_dentry->d_name.len, 0, index);
7240 btrfs_abort_transaction(trans, root, ret);
7244 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7245 struct dentry *parent = new_dentry->d_parent;
7246 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7247 btrfs_end_log_trans(root);
7250 btrfs_end_transaction(trans, root);
7252 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7253 up_read(&root->fs_info->subvol_sem);
7259 * some fairly slow code that needs optimization. This walks the list
7260 * of all the inodes with pending delalloc and forces them to disk.
7262 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7264 struct list_head *head = &root->fs_info->delalloc_inodes;
7265 struct btrfs_inode *binode;
7266 struct inode *inode;
7268 if (root->fs_info->sb->s_flags & MS_RDONLY)
7271 spin_lock(&root->fs_info->delalloc_lock);
7272 while (!list_empty(head)) {
7273 binode = list_entry(head->next, struct btrfs_inode,
7275 inode = igrab(&binode->vfs_inode);
7277 list_del_init(&binode->delalloc_inodes);
7278 spin_unlock(&root->fs_info->delalloc_lock);
7280 filemap_flush(inode->i_mapping);
7282 btrfs_add_delayed_iput(inode);
7287 spin_lock(&root->fs_info->delalloc_lock);
7289 spin_unlock(&root->fs_info->delalloc_lock);
7291 /* the filemap_flush will queue IO into the worker threads, but
7292 * we have to make sure the IO is actually started and that
7293 * ordered extents get created before we return
7295 atomic_inc(&root->fs_info->async_submit_draining);
7296 while (atomic_read(&root->fs_info->nr_async_submits) ||
7297 atomic_read(&root->fs_info->async_delalloc_pages)) {
7298 wait_event(root->fs_info->async_submit_wait,
7299 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7300 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7302 atomic_dec(&root->fs_info->async_submit_draining);
7306 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7307 const char *symname)
7309 struct btrfs_trans_handle *trans;
7310 struct btrfs_root *root = BTRFS_I(dir)->root;
7311 struct btrfs_path *path;
7312 struct btrfs_key key;
7313 struct inode *inode = NULL;
7321 struct btrfs_file_extent_item *ei;
7322 struct extent_buffer *leaf;
7323 unsigned long nr = 0;
7325 name_len = strlen(symname) + 1;
7326 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7327 return -ENAMETOOLONG;
7330 * 2 items for inode item and ref
7331 * 2 items for dir items
7332 * 1 item for xattr if selinux is on
7334 trans = btrfs_start_transaction(root, 5);
7336 return PTR_ERR(trans);
7338 err = btrfs_find_free_ino(root, &objectid);
7342 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7343 dentry->d_name.len, btrfs_ino(dir), objectid,
7344 S_IFLNK|S_IRWXUGO, &index);
7345 if (IS_ERR(inode)) {
7346 err = PTR_ERR(inode);
7350 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7357 * If the active LSM wants to access the inode during
7358 * d_instantiate it needs these. Smack checks to see
7359 * if the filesystem supports xattrs by looking at the
7362 inode->i_fop = &btrfs_file_operations;
7363 inode->i_op = &btrfs_file_inode_operations;
7365 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7369 inode->i_mapping->a_ops = &btrfs_aops;
7370 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7371 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7376 path = btrfs_alloc_path();
7382 key.objectid = btrfs_ino(inode);
7384 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7385 datasize = btrfs_file_extent_calc_inline_size(name_len);
7386 err = btrfs_insert_empty_item(trans, root, path, &key,
7390 btrfs_free_path(path);
7393 leaf = path->nodes[0];
7394 ei = btrfs_item_ptr(leaf, path->slots[0],
7395 struct btrfs_file_extent_item);
7396 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7397 btrfs_set_file_extent_type(leaf, ei,
7398 BTRFS_FILE_EXTENT_INLINE);
7399 btrfs_set_file_extent_encryption(leaf, ei, 0);
7400 btrfs_set_file_extent_compression(leaf, ei, 0);
7401 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7402 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7404 ptr = btrfs_file_extent_inline_start(ei);
7405 write_extent_buffer(leaf, symname, ptr, name_len);
7406 btrfs_mark_buffer_dirty(leaf);
7407 btrfs_free_path(path);
7409 inode->i_op = &btrfs_symlink_inode_operations;
7410 inode->i_mapping->a_ops = &btrfs_symlink_aops;
7411 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7412 inode_set_bytes(inode, name_len);
7413 btrfs_i_size_write(inode, name_len - 1);
7414 err = btrfs_update_inode(trans, root, inode);
7420 d_instantiate(dentry, inode);
7421 nr = trans->blocks_used;
7422 btrfs_end_transaction(trans, root);
7424 inode_dec_link_count(inode);
7427 btrfs_btree_balance_dirty(root, nr);
7431 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7432 u64 start, u64 num_bytes, u64 min_size,
7433 loff_t actual_len, u64 *alloc_hint,
7434 struct btrfs_trans_handle *trans)
7436 struct btrfs_root *root = BTRFS_I(inode)->root;
7437 struct btrfs_key ins;
7438 u64 cur_offset = start;
7441 bool own_trans = true;
7445 while (num_bytes > 0) {
7447 trans = btrfs_start_transaction(root, 3);
7448 if (IS_ERR(trans)) {
7449 ret = PTR_ERR(trans);
7454 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7455 0, *alloc_hint, &ins, 1);
7458 btrfs_end_transaction(trans, root);
7462 ret = insert_reserved_file_extent(trans, inode,
7463 cur_offset, ins.objectid,
7464 ins.offset, ins.offset,
7465 ins.offset, 0, 0, 0,
7466 BTRFS_FILE_EXTENT_PREALLOC);
7468 btrfs_abort_transaction(trans, root, ret);
7470 btrfs_end_transaction(trans, root);
7473 btrfs_drop_extent_cache(inode, cur_offset,
7474 cur_offset + ins.offset -1, 0);
7476 num_bytes -= ins.offset;
7477 cur_offset += ins.offset;
7478 *alloc_hint = ins.objectid + ins.offset;
7480 inode_inc_iversion(inode);
7481 inode->i_ctime = CURRENT_TIME;
7482 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7483 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7484 (actual_len > inode->i_size) &&
7485 (cur_offset > inode->i_size)) {
7486 if (cur_offset > actual_len)
7487 i_size = actual_len;
7489 i_size = cur_offset;
7490 i_size_write(inode, i_size);
7491 btrfs_ordered_update_i_size(inode, i_size, NULL);
7494 ret = btrfs_update_inode(trans, root, inode);
7497 btrfs_abort_transaction(trans, root, ret);
7499 btrfs_end_transaction(trans, root);
7504 btrfs_end_transaction(trans, root);
7509 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7510 u64 start, u64 num_bytes, u64 min_size,
7511 loff_t actual_len, u64 *alloc_hint)
7513 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7514 min_size, actual_len, alloc_hint,
7518 int btrfs_prealloc_file_range_trans(struct inode *inode,
7519 struct btrfs_trans_handle *trans, int mode,
7520 u64 start, u64 num_bytes, u64 min_size,
7521 loff_t actual_len, u64 *alloc_hint)
7523 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7524 min_size, actual_len, alloc_hint, trans);
7527 static int btrfs_set_page_dirty(struct page *page)
7529 return __set_page_dirty_nobuffers(page);
7532 static int btrfs_permission(struct inode *inode, int mask)
7534 struct btrfs_root *root = BTRFS_I(inode)->root;
7535 umode_t mode = inode->i_mode;
7537 if (mask & MAY_WRITE &&
7538 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7539 if (btrfs_root_readonly(root))
7541 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7544 return generic_permission(inode, mask);
7547 static const struct inode_operations btrfs_dir_inode_operations = {
7548 .getattr = btrfs_getattr,
7549 .lookup = btrfs_lookup,
7550 .create = btrfs_create,
7551 .unlink = btrfs_unlink,
7553 .mkdir = btrfs_mkdir,
7554 .rmdir = btrfs_rmdir,
7555 .rename = btrfs_rename,
7556 .symlink = btrfs_symlink,
7557 .setattr = btrfs_setattr,
7558 .mknod = btrfs_mknod,
7559 .setxattr = btrfs_setxattr,
7560 .getxattr = btrfs_getxattr,
7561 .listxattr = btrfs_listxattr,
7562 .removexattr = btrfs_removexattr,
7563 .permission = btrfs_permission,
7564 .get_acl = btrfs_get_acl,
7566 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7567 .lookup = btrfs_lookup,
7568 .permission = btrfs_permission,
7569 .get_acl = btrfs_get_acl,
7572 static const struct file_operations btrfs_dir_file_operations = {
7573 .llseek = generic_file_llseek,
7574 .read = generic_read_dir,
7575 .readdir = btrfs_real_readdir,
7576 .unlocked_ioctl = btrfs_ioctl,
7577 #ifdef CONFIG_COMPAT
7578 .compat_ioctl = btrfs_ioctl,
7580 .release = btrfs_release_file,
7581 .fsync = btrfs_sync_file,
7584 static struct extent_io_ops btrfs_extent_io_ops = {
7585 .fill_delalloc = run_delalloc_range,
7586 .submit_bio_hook = btrfs_submit_bio_hook,
7587 .merge_bio_hook = btrfs_merge_bio_hook,
7588 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7589 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7590 .writepage_start_hook = btrfs_writepage_start_hook,
7591 .set_bit_hook = btrfs_set_bit_hook,
7592 .clear_bit_hook = btrfs_clear_bit_hook,
7593 .merge_extent_hook = btrfs_merge_extent_hook,
7594 .split_extent_hook = btrfs_split_extent_hook,
7598 * btrfs doesn't support the bmap operation because swapfiles
7599 * use bmap to make a mapping of extents in the file. They assume
7600 * these extents won't change over the life of the file and they
7601 * use the bmap result to do IO directly to the drive.
7603 * the btrfs bmap call would return logical addresses that aren't
7604 * suitable for IO and they also will change frequently as COW
7605 * operations happen. So, swapfile + btrfs == corruption.
7607 * For now we're avoiding this by dropping bmap.
7609 static const struct address_space_operations btrfs_aops = {
7610 .readpage = btrfs_readpage,
7611 .writepage = btrfs_writepage,
7612 .writepages = btrfs_writepages,
7613 .readpages = btrfs_readpages,
7614 .direct_IO = btrfs_direct_IO,
7615 .invalidatepage = btrfs_invalidatepage,
7616 .releasepage = btrfs_releasepage,
7617 .set_page_dirty = btrfs_set_page_dirty,
7618 .error_remove_page = generic_error_remove_page,
7621 static const struct address_space_operations btrfs_symlink_aops = {
7622 .readpage = btrfs_readpage,
7623 .writepage = btrfs_writepage,
7624 .invalidatepage = btrfs_invalidatepage,
7625 .releasepage = btrfs_releasepage,
7628 static const struct inode_operations btrfs_file_inode_operations = {
7629 .getattr = btrfs_getattr,
7630 .setattr = btrfs_setattr,
7631 .setxattr = btrfs_setxattr,
7632 .getxattr = btrfs_getxattr,
7633 .listxattr = btrfs_listxattr,
7634 .removexattr = btrfs_removexattr,
7635 .permission = btrfs_permission,
7636 .fiemap = btrfs_fiemap,
7637 .get_acl = btrfs_get_acl,
7639 static const struct inode_operations btrfs_special_inode_operations = {
7640 .getattr = btrfs_getattr,
7641 .setattr = btrfs_setattr,
7642 .permission = btrfs_permission,
7643 .setxattr = btrfs_setxattr,
7644 .getxattr = btrfs_getxattr,
7645 .listxattr = btrfs_listxattr,
7646 .removexattr = btrfs_removexattr,
7647 .get_acl = btrfs_get_acl,
7649 static const struct inode_operations btrfs_symlink_inode_operations = {
7650 .readlink = generic_readlink,
7651 .follow_link = page_follow_link_light,
7652 .put_link = page_put_link,
7653 .getattr = btrfs_getattr,
7654 .setattr = btrfs_setattr,
7655 .permission = btrfs_permission,
7656 .setxattr = btrfs_setxattr,
7657 .getxattr = btrfs_getxattr,
7658 .listxattr = btrfs_listxattr,
7659 .removexattr = btrfs_removexattr,
7660 .get_acl = btrfs_get_acl,
7663 const struct dentry_operations btrfs_dentry_operations = {
7664 .d_delete = btrfs_dentry_delete,
7665 .d_release = btrfs_dentry_release,