2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir)
97 err = btrfs_init_acl(trans, inode, dir);
99 err = btrfs_xattr_security_init(trans, inode, dir);
104 * this does all the hard work for inserting an inline extent into
105 * the btree. The caller should have done a btrfs_drop_extents so that
106 * no overlapping inline items exist in the btree
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root, struct inode *inode,
110 u64 start, size_t size, size_t compressed_size,
111 struct page **compressed_pages)
113 struct btrfs_key key;
114 struct btrfs_path *path;
115 struct extent_buffer *leaf;
116 struct page *page = NULL;
119 struct btrfs_file_extent_item *ei;
122 size_t cur_size = size;
124 unsigned long offset;
125 int use_compress = 0;
127 if (compressed_size && compressed_pages) {
129 cur_size = compressed_size;
132 path = btrfs_alloc_path();
136 path->leave_spinning = 1;
137 btrfs_set_trans_block_group(trans, inode);
139 key.objectid = inode->i_ino;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
165 while (compressed_size > 0) {
166 cpage = compressed_pages[i];
167 cur_size = min_t(unsigned long, compressed_size,
170 kaddr = kmap_atomic(cpage, KM_USER0);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap_atomic(kaddr, KM_USER0);
176 compressed_size -= cur_size;
178 btrfs_set_file_extent_compression(leaf, ei,
179 BTRFS_COMPRESS_ZLIB);
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
194 * we're an inline extent, so nobody can
195 * extend the file past i_size without locking
196 * a page we already have locked.
198 * We must do any isize and inode updates
199 * before we unlock the pages. Otherwise we
200 * could end up racing with unlink.
202 BTRFS_I(inode)->disk_i_size = inode->i_size;
203 btrfs_update_inode(trans, root, inode);
207 btrfs_free_path(path);
213 * conditionally insert an inline extent into the file. This
214 * does the checks required to make sure the data is small enough
215 * to fit as an inline extent.
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218 struct btrfs_root *root,
219 struct inode *inode, u64 start, u64 end,
220 size_t compressed_size,
221 struct page **compressed_pages)
223 u64 isize = i_size_read(inode);
224 u64 actual_end = min(end + 1, isize);
225 u64 inline_len = actual_end - start;
226 u64 aligned_end = (end + root->sectorsize - 1) &
227 ~((u64)root->sectorsize - 1);
229 u64 data_len = inline_len;
233 data_len = compressed_size;
236 actual_end >= PAGE_CACHE_SIZE ||
237 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239 (actual_end & (root->sectorsize - 1)) == 0) ||
241 data_len > root->fs_info->max_inline) {
245 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 if (isize > actual_end)
250 inline_len = min_t(u64, isize, actual_end);
251 ret = insert_inline_extent(trans, root, inode, start,
252 inline_len, compressed_size,
255 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
259 struct async_extent {
264 unsigned long nr_pages;
265 struct list_head list;
270 struct btrfs_root *root;
271 struct page *locked_page;
274 struct list_head extents;
275 struct btrfs_work work;
278 static noinline int add_async_extent(struct async_cow *cow,
279 u64 start, u64 ram_size,
282 unsigned long nr_pages)
284 struct async_extent *async_extent;
286 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
287 async_extent->start = start;
288 async_extent->ram_size = ram_size;
289 async_extent->compressed_size = compressed_size;
290 async_extent->pages = pages;
291 async_extent->nr_pages = nr_pages;
292 list_add_tail(&async_extent->list, &cow->extents);
297 * we create compressed extents in two phases. The first
298 * phase compresses a range of pages that have already been
299 * locked (both pages and state bits are locked).
301 * This is done inside an ordered work queue, and the compression
302 * is spread across many cpus. The actual IO submission is step
303 * two, and the ordered work queue takes care of making sure that
304 * happens in the same order things were put onto the queue by
305 * writepages and friends.
307 * If this code finds it can't get good compression, it puts an
308 * entry onto the work queue to write the uncompressed bytes. This
309 * makes sure that both compressed inodes and uncompressed inodes
310 * are written in the same order that pdflush sent them down.
312 static noinline int compress_file_range(struct inode *inode,
313 struct page *locked_page,
315 struct async_cow *async_cow,
318 struct btrfs_root *root = BTRFS_I(inode)->root;
319 struct btrfs_trans_handle *trans;
323 u64 blocksize = root->sectorsize;
325 u64 isize = i_size_read(inode);
327 struct page **pages = NULL;
328 unsigned long nr_pages;
329 unsigned long nr_pages_ret = 0;
330 unsigned long total_compressed = 0;
331 unsigned long total_in = 0;
332 unsigned long max_compressed = 128 * 1024;
333 unsigned long max_uncompressed = 128 * 1024;
339 actual_end = min_t(u64, isize, end + 1);
342 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
343 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
346 * we don't want to send crud past the end of i_size through
347 * compression, that's just a waste of CPU time. So, if the
348 * end of the file is before the start of our current
349 * requested range of bytes, we bail out to the uncompressed
350 * cleanup code that can deal with all of this.
352 * It isn't really the fastest way to fix things, but this is a
353 * very uncommon corner.
355 if (actual_end <= start)
356 goto cleanup_and_bail_uncompressed;
358 total_compressed = actual_end - start;
360 /* we want to make sure that amount of ram required to uncompress
361 * an extent is reasonable, so we limit the total size in ram
362 * of a compressed extent to 128k. This is a crucial number
363 * because it also controls how easily we can spread reads across
364 * cpus for decompression.
366 * We also want to make sure the amount of IO required to do
367 * a random read is reasonably small, so we limit the size of
368 * a compressed extent to 128k.
370 total_compressed = min(total_compressed, max_uncompressed);
371 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
372 num_bytes = max(blocksize, num_bytes);
373 disk_num_bytes = num_bytes;
378 * we do compression for mount -o compress and when the
379 * inode has not been flagged as nocompress. This flag can
380 * change at any time if we discover bad compression ratios.
382 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
383 (btrfs_test_opt(root, COMPRESS) ||
384 (BTRFS_I(inode)->force_compress))) {
386 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
389 total_compressed, pages,
390 nr_pages, &nr_pages_ret,
396 unsigned long offset = total_compressed &
397 (PAGE_CACHE_SIZE - 1);
398 struct page *page = pages[nr_pages_ret - 1];
401 /* zero the tail end of the last page, we might be
402 * sending it down to disk
405 kaddr = kmap_atomic(page, KM_USER0);
406 memset(kaddr + offset, 0,
407 PAGE_CACHE_SIZE - offset);
408 kunmap_atomic(kaddr, KM_USER0);
414 trans = btrfs_join_transaction(root, 1);
416 btrfs_set_trans_block_group(trans, inode);
418 /* lets try to make an inline extent */
419 if (ret || total_in < (actual_end - start)) {
420 /* we didn't compress the entire range, try
421 * to make an uncompressed inline extent.
423 ret = cow_file_range_inline(trans, root, inode,
424 start, end, 0, NULL);
426 /* try making a compressed inline extent */
427 ret = cow_file_range_inline(trans, root, inode,
429 total_compressed, pages);
433 * inline extent creation worked, we don't need
434 * to create any more async work items. Unlock
435 * and free up our temp pages.
437 extent_clear_unlock_delalloc(inode,
438 &BTRFS_I(inode)->io_tree,
440 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
441 EXTENT_CLEAR_DELALLOC |
442 EXTENT_CLEAR_ACCOUNTING |
443 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
445 btrfs_end_transaction(trans, root);
448 btrfs_end_transaction(trans, root);
453 * we aren't doing an inline extent round the compressed size
454 * up to a block size boundary so the allocator does sane
457 total_compressed = (total_compressed + blocksize - 1) &
461 * one last check to make sure the compression is really a
462 * win, compare the page count read with the blocks on disk
464 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
465 ~(PAGE_CACHE_SIZE - 1);
466 if (total_compressed >= total_in) {
469 disk_num_bytes = total_compressed;
470 num_bytes = total_in;
473 if (!will_compress && pages) {
475 * the compression code ran but failed to make things smaller,
476 * free any pages it allocated and our page pointer array
478 for (i = 0; i < nr_pages_ret; i++) {
479 WARN_ON(pages[i]->mapping);
480 page_cache_release(pages[i]);
484 total_compressed = 0;
487 /* flag the file so we don't compress in the future */
488 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
489 !(BTRFS_I(inode)->force_compress)) {
490 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
496 /* the async work queues will take care of doing actual
497 * allocation on disk for these compressed pages,
498 * and will submit them to the elevator.
500 add_async_extent(async_cow, start, num_bytes,
501 total_compressed, pages, nr_pages_ret);
503 if (start + num_bytes < end && start + num_bytes < actual_end) {
510 cleanup_and_bail_uncompressed:
512 * No compression, but we still need to write the pages in
513 * the file we've been given so far. redirty the locked
514 * page if it corresponds to our extent and set things up
515 * for the async work queue to run cow_file_range to do
516 * the normal delalloc dance
518 if (page_offset(locked_page) >= start &&
519 page_offset(locked_page) <= end) {
520 __set_page_dirty_nobuffers(locked_page);
521 /* unlocked later on in the async handlers */
523 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
531 for (i = 0; i < nr_pages_ret; i++) {
532 WARN_ON(pages[i]->mapping);
533 page_cache_release(pages[i]);
541 * phase two of compressed writeback. This is the ordered portion
542 * of the code, which only gets called in the order the work was
543 * queued. We walk all the async extents created by compress_file_range
544 * and send them down to the disk.
546 static noinline int submit_compressed_extents(struct inode *inode,
547 struct async_cow *async_cow)
549 struct async_extent *async_extent;
551 struct btrfs_trans_handle *trans;
552 struct btrfs_key ins;
553 struct extent_map *em;
554 struct btrfs_root *root = BTRFS_I(inode)->root;
555 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
556 struct extent_io_tree *io_tree;
559 if (list_empty(&async_cow->extents))
563 while (!list_empty(&async_cow->extents)) {
564 async_extent = list_entry(async_cow->extents.next,
565 struct async_extent, list);
566 list_del(&async_extent->list);
568 io_tree = &BTRFS_I(inode)->io_tree;
571 /* did the compression code fall back to uncompressed IO? */
572 if (!async_extent->pages) {
573 int page_started = 0;
574 unsigned long nr_written = 0;
576 lock_extent(io_tree, async_extent->start,
577 async_extent->start +
578 async_extent->ram_size - 1, GFP_NOFS);
580 /* allocate blocks */
581 ret = cow_file_range(inode, async_cow->locked_page,
583 async_extent->start +
584 async_extent->ram_size - 1,
585 &page_started, &nr_written, 0);
588 * if page_started, cow_file_range inserted an
589 * inline extent and took care of all the unlocking
590 * and IO for us. Otherwise, we need to submit
591 * all those pages down to the drive.
593 if (!page_started && !ret)
594 extent_write_locked_range(io_tree,
595 inode, async_extent->start,
596 async_extent->start +
597 async_extent->ram_size - 1,
605 lock_extent(io_tree, async_extent->start,
606 async_extent->start + async_extent->ram_size - 1,
609 trans = btrfs_join_transaction(root, 1);
610 ret = btrfs_reserve_extent(trans, root,
611 async_extent->compressed_size,
612 async_extent->compressed_size,
615 btrfs_end_transaction(trans, root);
619 for (i = 0; i < async_extent->nr_pages; i++) {
620 WARN_ON(async_extent->pages[i]->mapping);
621 page_cache_release(async_extent->pages[i]);
623 kfree(async_extent->pages);
624 async_extent->nr_pages = 0;
625 async_extent->pages = NULL;
626 unlock_extent(io_tree, async_extent->start,
627 async_extent->start +
628 async_extent->ram_size - 1, GFP_NOFS);
633 * here we're doing allocation and writeback of the
636 btrfs_drop_extent_cache(inode, async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1, 0);
640 em = alloc_extent_map(GFP_NOFS);
641 em->start = async_extent->start;
642 em->len = async_extent->ram_size;
643 em->orig_start = em->start;
645 em->block_start = ins.objectid;
646 em->block_len = ins.offset;
647 em->bdev = root->fs_info->fs_devices->latest_bdev;
648 set_bit(EXTENT_FLAG_PINNED, &em->flags);
649 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
652 write_lock(&em_tree->lock);
653 ret = add_extent_mapping(em_tree, em);
654 write_unlock(&em_tree->lock);
655 if (ret != -EEXIST) {
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
664 ret = btrfs_add_ordered_extent(inode, async_extent->start,
666 async_extent->ram_size,
668 BTRFS_ORDERED_COMPRESSED);
672 * clear dirty, set writeback and unlock the pages.
674 extent_clear_unlock_delalloc(inode,
675 &BTRFS_I(inode)->io_tree,
677 async_extent->start +
678 async_extent->ram_size - 1,
679 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
680 EXTENT_CLEAR_UNLOCK |
681 EXTENT_CLEAR_DELALLOC |
682 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
684 ret = btrfs_submit_compressed_write(inode,
686 async_extent->ram_size,
688 ins.offset, async_extent->pages,
689 async_extent->nr_pages);
692 alloc_hint = ins.objectid + ins.offset;
701 * when extent_io.c finds a delayed allocation range in the file,
702 * the call backs end up in this code. The basic idea is to
703 * allocate extents on disk for the range, and create ordered data structs
704 * in ram to track those extents.
706 * locked_page is the page that writepage had locked already. We use
707 * it to make sure we don't do extra locks or unlocks.
709 * *page_started is set to one if we unlock locked_page and do everything
710 * required to start IO on it. It may be clean and already done with
713 static noinline int cow_file_range(struct inode *inode,
714 struct page *locked_page,
715 u64 start, u64 end, int *page_started,
716 unsigned long *nr_written,
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct btrfs_trans_handle *trans;
723 unsigned long ram_size;
726 u64 blocksize = root->sectorsize;
728 u64 isize = i_size_read(inode);
729 struct btrfs_key ins;
730 struct extent_map *em;
731 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
734 trans = btrfs_join_transaction(root, 1);
736 btrfs_set_trans_block_group(trans, inode);
738 actual_end = min_t(u64, isize, end + 1);
740 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
741 num_bytes = max(blocksize, num_bytes);
742 disk_num_bytes = num_bytes;
746 /* lets try to make an inline extent */
747 ret = cow_file_range_inline(trans, root, inode,
748 start, end, 0, NULL);
750 extent_clear_unlock_delalloc(inode,
751 &BTRFS_I(inode)->io_tree,
753 EXTENT_CLEAR_UNLOCK_PAGE |
754 EXTENT_CLEAR_UNLOCK |
755 EXTENT_CLEAR_DELALLOC |
756 EXTENT_CLEAR_ACCOUNTING |
758 EXTENT_SET_WRITEBACK |
759 EXTENT_END_WRITEBACK);
761 *nr_written = *nr_written +
762 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
769 BUG_ON(disk_num_bytes >
770 btrfs_super_total_bytes(&root->fs_info->super_copy));
773 read_lock(&BTRFS_I(inode)->extent_tree.lock);
774 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
778 * if block start isn't an actual block number then find the
779 * first block in this inode and use that as a hint. If that
780 * block is also bogus then just don't worry about it.
782 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
784 em = search_extent_mapping(em_tree, 0, 0);
785 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
786 alloc_hint = em->block_start;
790 alloc_hint = em->block_start;
794 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
795 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
797 while (disk_num_bytes > 0) {
800 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
801 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
802 root->sectorsize, 0, alloc_hint,
806 em = alloc_extent_map(GFP_NOFS);
808 em->orig_start = em->start;
809 ram_size = ins.offset;
810 em->len = ins.offset;
812 em->block_start = ins.objectid;
813 em->block_len = ins.offset;
814 em->bdev = root->fs_info->fs_devices->latest_bdev;
815 set_bit(EXTENT_FLAG_PINNED, &em->flags);
818 write_lock(&em_tree->lock);
819 ret = add_extent_mapping(em_tree, em);
820 write_unlock(&em_tree->lock);
821 if (ret != -EEXIST) {
825 btrfs_drop_extent_cache(inode, start,
826 start + ram_size - 1, 0);
829 cur_alloc_size = ins.offset;
830 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
831 ram_size, cur_alloc_size, 0);
834 if (root->root_key.objectid ==
835 BTRFS_DATA_RELOC_TREE_OBJECTID) {
836 ret = btrfs_reloc_clone_csums(inode, start,
841 if (disk_num_bytes < cur_alloc_size)
844 /* we're not doing compressed IO, don't unlock the first
845 * page (which the caller expects to stay locked), don't
846 * clear any dirty bits and don't set any writeback bits
848 * Do set the Private2 bit so we know this page was properly
849 * setup for writepage
851 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
852 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
855 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
856 start, start + ram_size - 1,
858 disk_num_bytes -= cur_alloc_size;
859 num_bytes -= cur_alloc_size;
860 alloc_hint = ins.objectid + ins.offset;
861 start += cur_alloc_size;
865 btrfs_end_transaction(trans, root);
871 * work queue call back to started compression on a file and pages
873 static noinline void async_cow_start(struct btrfs_work *work)
875 struct async_cow *async_cow;
877 async_cow = container_of(work, struct async_cow, work);
879 compress_file_range(async_cow->inode, async_cow->locked_page,
880 async_cow->start, async_cow->end, async_cow,
883 async_cow->inode = NULL;
887 * work queue call back to submit previously compressed pages
889 static noinline void async_cow_submit(struct btrfs_work *work)
891 struct async_cow *async_cow;
892 struct btrfs_root *root;
893 unsigned long nr_pages;
895 async_cow = container_of(work, struct async_cow, work);
897 root = async_cow->root;
898 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
901 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
903 if (atomic_read(&root->fs_info->async_delalloc_pages) <
905 waitqueue_active(&root->fs_info->async_submit_wait))
906 wake_up(&root->fs_info->async_submit_wait);
908 if (async_cow->inode)
909 submit_compressed_extents(async_cow->inode, async_cow);
912 static noinline void async_cow_free(struct btrfs_work *work)
914 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
919 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written)
923 struct async_cow *async_cow;
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 unsigned long nr_pages;
927 int limit = 10 * 1024 * 1042;
929 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
930 1, 0, NULL, GFP_NOFS);
931 while (start < end) {
932 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
933 async_cow->inode = inode;
934 async_cow->root = root;
935 async_cow->locked_page = locked_page;
936 async_cow->start = start;
938 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
941 cur_end = min(end, start + 512 * 1024 - 1);
943 async_cow->end = cur_end;
944 INIT_LIST_HEAD(&async_cow->extents);
946 async_cow->work.func = async_cow_start;
947 async_cow->work.ordered_func = async_cow_submit;
948 async_cow->work.ordered_free = async_cow_free;
949 async_cow->work.flags = 0;
951 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
953 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
955 btrfs_queue_worker(&root->fs_info->delalloc_workers,
958 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
959 wait_event(root->fs_info->async_submit_wait,
960 (atomic_read(&root->fs_info->async_delalloc_pages) <
964 while (atomic_read(&root->fs_info->async_submit_draining) &&
965 atomic_read(&root->fs_info->async_delalloc_pages)) {
966 wait_event(root->fs_info->async_submit_wait,
967 (atomic_read(&root->fs_info->async_delalloc_pages) ==
971 *nr_written += nr_pages;
978 static noinline int csum_exist_in_range(struct btrfs_root *root,
979 u64 bytenr, u64 num_bytes)
982 struct btrfs_ordered_sum *sums;
985 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
986 bytenr + num_bytes - 1, &list);
987 if (ret == 0 && list_empty(&list))
990 while (!list_empty(&list)) {
991 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
992 list_del(&sums->list);
999 * when nowcow writeback call back. This checks for snapshots or COW copies
1000 * of the extents that exist in the file, and COWs the file as required.
1002 * If no cow copies or snapshots exist, we write directly to the existing
1005 static noinline int run_delalloc_nocow(struct inode *inode,
1006 struct page *locked_page,
1007 u64 start, u64 end, int *page_started, int force,
1008 unsigned long *nr_written)
1010 struct btrfs_root *root = BTRFS_I(inode)->root;
1011 struct btrfs_trans_handle *trans;
1012 struct extent_buffer *leaf;
1013 struct btrfs_path *path;
1014 struct btrfs_file_extent_item *fi;
1015 struct btrfs_key found_key;
1028 path = btrfs_alloc_path();
1030 trans = btrfs_join_transaction(root, 1);
1033 cow_start = (u64)-1;
1036 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1039 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1040 leaf = path->nodes[0];
1041 btrfs_item_key_to_cpu(leaf, &found_key,
1042 path->slots[0] - 1);
1043 if (found_key.objectid == inode->i_ino &&
1044 found_key.type == BTRFS_EXTENT_DATA_KEY)
1049 leaf = path->nodes[0];
1050 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1051 ret = btrfs_next_leaf(root, path);
1056 leaf = path->nodes[0];
1062 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1064 if (found_key.objectid > inode->i_ino ||
1065 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1066 found_key.offset > end)
1069 if (found_key.offset > cur_offset) {
1070 extent_end = found_key.offset;
1075 fi = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_file_extent_item);
1077 extent_type = btrfs_file_extent_type(leaf, fi);
1079 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1080 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1081 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1082 extent_offset = btrfs_file_extent_offset(leaf, fi);
1083 extent_end = found_key.offset +
1084 btrfs_file_extent_num_bytes(leaf, fi);
1085 if (extent_end <= start) {
1089 if (disk_bytenr == 0)
1091 if (btrfs_file_extent_compression(leaf, fi) ||
1092 btrfs_file_extent_encryption(leaf, fi) ||
1093 btrfs_file_extent_other_encoding(leaf, fi))
1095 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1097 if (btrfs_extent_readonly(root, disk_bytenr))
1099 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1101 extent_offset, disk_bytenr))
1103 disk_bytenr += extent_offset;
1104 disk_bytenr += cur_offset - found_key.offset;
1105 num_bytes = min(end + 1, extent_end) - cur_offset;
1107 * force cow if csum exists in the range.
1108 * this ensure that csum for a given extent are
1109 * either valid or do not exist.
1111 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1114 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1115 extent_end = found_key.offset +
1116 btrfs_file_extent_inline_len(leaf, fi);
1117 extent_end = ALIGN(extent_end, root->sectorsize);
1122 if (extent_end <= start) {
1127 if (cow_start == (u64)-1)
1128 cow_start = cur_offset;
1129 cur_offset = extent_end;
1130 if (cur_offset > end)
1136 btrfs_release_path(root, path);
1137 if (cow_start != (u64)-1) {
1138 ret = cow_file_range(inode, locked_page, cow_start,
1139 found_key.offset - 1, page_started,
1142 cow_start = (u64)-1;
1145 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1146 struct extent_map *em;
1147 struct extent_map_tree *em_tree;
1148 em_tree = &BTRFS_I(inode)->extent_tree;
1149 em = alloc_extent_map(GFP_NOFS);
1150 em->start = cur_offset;
1151 em->orig_start = em->start;
1152 em->len = num_bytes;
1153 em->block_len = num_bytes;
1154 em->block_start = disk_bytenr;
1155 em->bdev = root->fs_info->fs_devices->latest_bdev;
1156 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1158 write_lock(&em_tree->lock);
1159 ret = add_extent_mapping(em_tree, em);
1160 write_unlock(&em_tree->lock);
1161 if (ret != -EEXIST) {
1162 free_extent_map(em);
1165 btrfs_drop_extent_cache(inode, em->start,
1166 em->start + em->len - 1, 0);
1168 type = BTRFS_ORDERED_PREALLOC;
1170 type = BTRFS_ORDERED_NOCOW;
1173 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1174 num_bytes, num_bytes, type);
1177 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1178 cur_offset, cur_offset + num_bytes - 1,
1179 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1181 EXTENT_SET_PRIVATE2);
1182 cur_offset = extent_end;
1183 if (cur_offset > end)
1186 btrfs_release_path(root, path);
1188 if (cur_offset <= end && cow_start == (u64)-1)
1189 cow_start = cur_offset;
1190 if (cow_start != (u64)-1) {
1191 ret = cow_file_range(inode, locked_page, cow_start, end,
1192 page_started, nr_written, 1);
1196 ret = btrfs_end_transaction(trans, root);
1198 btrfs_free_path(path);
1203 * extent_io.c call back to do delayed allocation processing
1205 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1206 u64 start, u64 end, int *page_started,
1207 unsigned long *nr_written)
1210 struct btrfs_root *root = BTRFS_I(inode)->root;
1212 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1213 ret = run_delalloc_nocow(inode, locked_page, start, end,
1214 page_started, 1, nr_written);
1215 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1216 ret = run_delalloc_nocow(inode, locked_page, start, end,
1217 page_started, 0, nr_written);
1218 else if (!btrfs_test_opt(root, COMPRESS) &&
1219 !(BTRFS_I(inode)->force_compress))
1220 ret = cow_file_range(inode, locked_page, start, end,
1221 page_started, nr_written, 1);
1223 ret = cow_file_range_async(inode, locked_page, start, end,
1224 page_started, nr_written);
1228 static int btrfs_split_extent_hook(struct inode *inode,
1229 struct extent_state *orig, u64 split)
1231 struct btrfs_root *root = BTRFS_I(inode)->root;
1234 if (!(orig->state & EXTENT_DELALLOC))
1237 size = orig->end - orig->start + 1;
1238 if (size > root->fs_info->max_extent) {
1242 new_size = orig->end - split + 1;
1243 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1244 root->fs_info->max_extent);
1247 * if we break a large extent up then leave oustanding_extents
1248 * be, since we've already accounted for the large extent.
1250 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1251 root->fs_info->max_extent) < num_extents)
1255 spin_lock(&BTRFS_I(inode)->accounting_lock);
1256 BTRFS_I(inode)->outstanding_extents++;
1257 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1263 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1264 * extents so we can keep track of new extents that are just merged onto old
1265 * extents, such as when we are doing sequential writes, so we can properly
1266 * account for the metadata space we'll need.
1268 static int btrfs_merge_extent_hook(struct inode *inode,
1269 struct extent_state *new,
1270 struct extent_state *other)
1272 struct btrfs_root *root = BTRFS_I(inode)->root;
1273 u64 new_size, old_size;
1276 /* not delalloc, ignore it */
1277 if (!(other->state & EXTENT_DELALLOC))
1280 old_size = other->end - other->start + 1;
1281 if (new->start < other->start)
1282 new_size = other->end - new->start + 1;
1284 new_size = new->end - other->start + 1;
1286 /* we're not bigger than the max, unreserve the space and go */
1287 if (new_size <= root->fs_info->max_extent) {
1288 spin_lock(&BTRFS_I(inode)->accounting_lock);
1289 BTRFS_I(inode)->outstanding_extents--;
1290 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1295 * If we grew by another max_extent, just return, we want to keep that
1298 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1299 root->fs_info->max_extent);
1300 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1301 root->fs_info->max_extent) > num_extents)
1304 spin_lock(&BTRFS_I(inode)->accounting_lock);
1305 BTRFS_I(inode)->outstanding_extents--;
1306 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1312 * extent_io.c set_bit_hook, used to track delayed allocation
1313 * bytes in this file, and to maintain the list of inodes that
1314 * have pending delalloc work to be done.
1316 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1317 unsigned long old, unsigned long bits)
1321 * set_bit and clear bit hooks normally require _irqsave/restore
1322 * but in this case, we are only testeing for the DELALLOC
1323 * bit, which is only set or cleared with irqs on
1325 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1326 struct btrfs_root *root = BTRFS_I(inode)->root;
1328 spin_lock(&BTRFS_I(inode)->accounting_lock);
1329 BTRFS_I(inode)->outstanding_extents++;
1330 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1331 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1332 spin_lock(&root->fs_info->delalloc_lock);
1333 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1334 root->fs_info->delalloc_bytes += end - start + 1;
1335 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1336 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1337 &root->fs_info->delalloc_inodes);
1339 spin_unlock(&root->fs_info->delalloc_lock);
1345 * extent_io.c clear_bit_hook, see set_bit_hook for why
1347 static int btrfs_clear_bit_hook(struct inode *inode,
1348 struct extent_state *state, unsigned long bits)
1351 * set_bit and clear bit hooks normally require _irqsave/restore
1352 * but in this case, we are only testeing for the DELALLOC
1353 * bit, which is only set or cleared with irqs on
1355 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1356 struct btrfs_root *root = BTRFS_I(inode)->root;
1358 if (bits & EXTENT_DO_ACCOUNTING) {
1359 spin_lock(&BTRFS_I(inode)->accounting_lock);
1360 BTRFS_I(inode)->outstanding_extents--;
1361 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1362 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1365 spin_lock(&root->fs_info->delalloc_lock);
1366 if (state->end - state->start + 1 >
1367 root->fs_info->delalloc_bytes) {
1368 printk(KERN_INFO "btrfs warning: delalloc account "
1370 (unsigned long long)
1371 state->end - state->start + 1,
1372 (unsigned long long)
1373 root->fs_info->delalloc_bytes);
1374 btrfs_delalloc_free_space(root, inode, (u64)-1);
1375 root->fs_info->delalloc_bytes = 0;
1376 BTRFS_I(inode)->delalloc_bytes = 0;
1378 btrfs_delalloc_free_space(root, inode,
1381 root->fs_info->delalloc_bytes -= state->end -
1383 BTRFS_I(inode)->delalloc_bytes -= state->end -
1386 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1387 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1388 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1390 spin_unlock(&root->fs_info->delalloc_lock);
1396 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1397 * we don't create bios that span stripes or chunks
1399 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1400 size_t size, struct bio *bio,
1401 unsigned long bio_flags)
1403 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1404 struct btrfs_mapping_tree *map_tree;
1405 u64 logical = (u64)bio->bi_sector << 9;
1410 if (bio_flags & EXTENT_BIO_COMPRESSED)
1413 length = bio->bi_size;
1414 map_tree = &root->fs_info->mapping_tree;
1415 map_length = length;
1416 ret = btrfs_map_block(map_tree, READ, logical,
1417 &map_length, NULL, 0);
1419 if (map_length < length + size)
1425 * in order to insert checksums into the metadata in large chunks,
1426 * we wait until bio submission time. All the pages in the bio are
1427 * checksummed and sums are attached onto the ordered extent record.
1429 * At IO completion time the cums attached on the ordered extent record
1430 * are inserted into the btree
1432 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1433 struct bio *bio, int mirror_num,
1434 unsigned long bio_flags)
1436 struct btrfs_root *root = BTRFS_I(inode)->root;
1439 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1445 * in order to insert checksums into the metadata in large chunks,
1446 * we wait until bio submission time. All the pages in the bio are
1447 * checksummed and sums are attached onto the ordered extent record.
1449 * At IO completion time the cums attached on the ordered extent record
1450 * are inserted into the btree
1452 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1453 int mirror_num, unsigned long bio_flags)
1455 struct btrfs_root *root = BTRFS_I(inode)->root;
1456 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1460 * extent_io.c submission hook. This does the right thing for csum calculation
1461 * on write, or reading the csums from the tree before a read
1463 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1464 int mirror_num, unsigned long bio_flags)
1466 struct btrfs_root *root = BTRFS_I(inode)->root;
1470 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1472 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1475 if (!(rw & (1 << BIO_RW))) {
1476 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1477 return btrfs_submit_compressed_read(inode, bio,
1478 mirror_num, bio_flags);
1479 } else if (!skip_sum)
1480 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1482 } else if (!skip_sum) {
1483 /* csum items have already been cloned */
1484 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1486 /* we're doing a write, do the async checksumming */
1487 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1488 inode, rw, bio, mirror_num,
1489 bio_flags, __btrfs_submit_bio_start,
1490 __btrfs_submit_bio_done);
1494 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1498 * given a list of ordered sums record them in the inode. This happens
1499 * at IO completion time based on sums calculated at bio submission time.
1501 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1502 struct inode *inode, u64 file_offset,
1503 struct list_head *list)
1505 struct btrfs_ordered_sum *sum;
1507 btrfs_set_trans_block_group(trans, inode);
1509 list_for_each_entry(sum, list, list) {
1510 btrfs_csum_file_blocks(trans,
1511 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1516 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1517 struct extent_state **cached_state)
1519 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1521 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1522 cached_state, GFP_NOFS);
1525 /* see btrfs_writepage_start_hook for details on why this is required */
1526 struct btrfs_writepage_fixup {
1528 struct btrfs_work work;
1531 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1533 struct btrfs_writepage_fixup *fixup;
1534 struct btrfs_ordered_extent *ordered;
1535 struct extent_state *cached_state = NULL;
1537 struct inode *inode;
1541 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1545 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1546 ClearPageChecked(page);
1550 inode = page->mapping->host;
1551 page_start = page_offset(page);
1552 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1554 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1555 &cached_state, GFP_NOFS);
1557 /* already ordered? We're done */
1558 if (PagePrivate2(page))
1561 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1563 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1564 page_end, &cached_state, GFP_NOFS);
1566 btrfs_start_ordered_extent(inode, ordered, 1);
1570 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1571 ClearPageChecked(page);
1573 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1574 &cached_state, GFP_NOFS);
1577 page_cache_release(page);
1581 * There are a few paths in the higher layers of the kernel that directly
1582 * set the page dirty bit without asking the filesystem if it is a
1583 * good idea. This causes problems because we want to make sure COW
1584 * properly happens and the data=ordered rules are followed.
1586 * In our case any range that doesn't have the ORDERED bit set
1587 * hasn't been properly setup for IO. We kick off an async process
1588 * to fix it up. The async helper will wait for ordered extents, set
1589 * the delalloc bit and make it safe to write the page.
1591 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1593 struct inode *inode = page->mapping->host;
1594 struct btrfs_writepage_fixup *fixup;
1595 struct btrfs_root *root = BTRFS_I(inode)->root;
1597 /* this page is properly in the ordered list */
1598 if (TestClearPagePrivate2(page))
1601 if (PageChecked(page))
1604 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1608 SetPageChecked(page);
1609 page_cache_get(page);
1610 fixup->work.func = btrfs_writepage_fixup_worker;
1612 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1616 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1617 struct inode *inode, u64 file_pos,
1618 u64 disk_bytenr, u64 disk_num_bytes,
1619 u64 num_bytes, u64 ram_bytes,
1620 u8 compression, u8 encryption,
1621 u16 other_encoding, int extent_type)
1623 struct btrfs_root *root = BTRFS_I(inode)->root;
1624 struct btrfs_file_extent_item *fi;
1625 struct btrfs_path *path;
1626 struct extent_buffer *leaf;
1627 struct btrfs_key ins;
1631 path = btrfs_alloc_path();
1634 path->leave_spinning = 1;
1637 * we may be replacing one extent in the tree with another.
1638 * The new extent is pinned in the extent map, and we don't want
1639 * to drop it from the cache until it is completely in the btree.
1641 * So, tell btrfs_drop_extents to leave this extent in the cache.
1642 * the caller is expected to unpin it and allow it to be merged
1645 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1649 ins.objectid = inode->i_ino;
1650 ins.offset = file_pos;
1651 ins.type = BTRFS_EXTENT_DATA_KEY;
1652 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1654 leaf = path->nodes[0];
1655 fi = btrfs_item_ptr(leaf, path->slots[0],
1656 struct btrfs_file_extent_item);
1657 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1658 btrfs_set_file_extent_type(leaf, fi, extent_type);
1659 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1660 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1661 btrfs_set_file_extent_offset(leaf, fi, 0);
1662 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1663 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1664 btrfs_set_file_extent_compression(leaf, fi, compression);
1665 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1666 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1668 btrfs_unlock_up_safe(path, 1);
1669 btrfs_set_lock_blocking(leaf);
1671 btrfs_mark_buffer_dirty(leaf);
1673 inode_add_bytes(inode, num_bytes);
1675 ins.objectid = disk_bytenr;
1676 ins.offset = disk_num_bytes;
1677 ins.type = BTRFS_EXTENT_ITEM_KEY;
1678 ret = btrfs_alloc_reserved_file_extent(trans, root,
1679 root->root_key.objectid,
1680 inode->i_ino, file_pos, &ins);
1682 btrfs_free_path(path);
1688 * helper function for btrfs_finish_ordered_io, this
1689 * just reads in some of the csum leaves to prime them into ram
1690 * before we start the transaction. It limits the amount of btree
1691 * reads required while inside the transaction.
1693 /* as ordered data IO finishes, this gets called so we can finish
1694 * an ordered extent if the range of bytes in the file it covers are
1697 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1699 struct btrfs_root *root = BTRFS_I(inode)->root;
1700 struct btrfs_trans_handle *trans;
1701 struct btrfs_ordered_extent *ordered_extent = NULL;
1702 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1703 struct extent_state *cached_state = NULL;
1707 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1711 BUG_ON(!ordered_extent);
1713 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1714 BUG_ON(!list_empty(&ordered_extent->list));
1715 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1717 trans = btrfs_join_transaction(root, 1);
1718 ret = btrfs_update_inode(trans, root, inode);
1720 btrfs_end_transaction(trans, root);
1725 lock_extent_bits(io_tree, ordered_extent->file_offset,
1726 ordered_extent->file_offset + ordered_extent->len - 1,
1727 0, &cached_state, GFP_NOFS);
1729 trans = btrfs_join_transaction(root, 1);
1731 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1733 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1735 ret = btrfs_mark_extent_written(trans, inode,
1736 ordered_extent->file_offset,
1737 ordered_extent->file_offset +
1738 ordered_extent->len);
1741 ret = insert_reserved_file_extent(trans, inode,
1742 ordered_extent->file_offset,
1743 ordered_extent->start,
1744 ordered_extent->disk_len,
1745 ordered_extent->len,
1746 ordered_extent->len,
1748 BTRFS_FILE_EXTENT_REG);
1749 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1750 ordered_extent->file_offset,
1751 ordered_extent->len);
1754 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1755 ordered_extent->file_offset +
1756 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1758 add_pending_csums(trans, inode, ordered_extent->file_offset,
1759 &ordered_extent->list);
1761 /* this also removes the ordered extent from the tree */
1762 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1763 ret = btrfs_update_inode(trans, root, inode);
1765 btrfs_end_transaction(trans, root);
1768 btrfs_put_ordered_extent(ordered_extent);
1769 /* once for the tree */
1770 btrfs_put_ordered_extent(ordered_extent);
1775 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1776 struct extent_state *state, int uptodate)
1778 ClearPagePrivate2(page);
1779 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1783 * When IO fails, either with EIO or csum verification fails, we
1784 * try other mirrors that might have a good copy of the data. This
1785 * io_failure_record is used to record state as we go through all the
1786 * mirrors. If another mirror has good data, the page is set up to date
1787 * and things continue. If a good mirror can't be found, the original
1788 * bio end_io callback is called to indicate things have failed.
1790 struct io_failure_record {
1795 unsigned long bio_flags;
1799 static int btrfs_io_failed_hook(struct bio *failed_bio,
1800 struct page *page, u64 start, u64 end,
1801 struct extent_state *state)
1803 struct io_failure_record *failrec = NULL;
1805 struct extent_map *em;
1806 struct inode *inode = page->mapping->host;
1807 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1808 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1815 ret = get_state_private(failure_tree, start, &private);
1817 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1820 failrec->start = start;
1821 failrec->len = end - start + 1;
1822 failrec->last_mirror = 0;
1823 failrec->bio_flags = 0;
1825 read_lock(&em_tree->lock);
1826 em = lookup_extent_mapping(em_tree, start, failrec->len);
1827 if (em->start > start || em->start + em->len < start) {
1828 free_extent_map(em);
1831 read_unlock(&em_tree->lock);
1833 if (!em || IS_ERR(em)) {
1837 logical = start - em->start;
1838 logical = em->block_start + logical;
1839 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1840 logical = em->block_start;
1841 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1843 failrec->logical = logical;
1844 free_extent_map(em);
1845 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1846 EXTENT_DIRTY, GFP_NOFS);
1847 set_state_private(failure_tree, start,
1848 (u64)(unsigned long)failrec);
1850 failrec = (struct io_failure_record *)(unsigned long)private;
1852 num_copies = btrfs_num_copies(
1853 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1854 failrec->logical, failrec->len);
1855 failrec->last_mirror++;
1857 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1858 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1861 if (state && state->start != failrec->start)
1863 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1865 if (!state || failrec->last_mirror > num_copies) {
1866 set_state_private(failure_tree, failrec->start, 0);
1867 clear_extent_bits(failure_tree, failrec->start,
1868 failrec->start + failrec->len - 1,
1869 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1873 bio = bio_alloc(GFP_NOFS, 1);
1874 bio->bi_private = state;
1875 bio->bi_end_io = failed_bio->bi_end_io;
1876 bio->bi_sector = failrec->logical >> 9;
1877 bio->bi_bdev = failed_bio->bi_bdev;
1880 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1881 if (failed_bio->bi_rw & (1 << BIO_RW))
1886 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1887 failrec->last_mirror,
1888 failrec->bio_flags);
1893 * each time an IO finishes, we do a fast check in the IO failure tree
1894 * to see if we need to process or clean up an io_failure_record
1896 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1899 u64 private_failure;
1900 struct io_failure_record *failure;
1904 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1905 (u64)-1, 1, EXTENT_DIRTY)) {
1906 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1907 start, &private_failure);
1909 failure = (struct io_failure_record *)(unsigned long)
1911 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1913 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1915 failure->start + failure->len - 1,
1916 EXTENT_DIRTY | EXTENT_LOCKED,
1925 * when reads are done, we need to check csums to verify the data is correct
1926 * if there's a match, we allow the bio to finish. If not, we go through
1927 * the io_failure_record routines to find good copies
1929 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1930 struct extent_state *state)
1932 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1933 struct inode *inode = page->mapping->host;
1934 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1936 u64 private = ~(u32)0;
1938 struct btrfs_root *root = BTRFS_I(inode)->root;
1941 if (PageChecked(page)) {
1942 ClearPageChecked(page);
1946 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1949 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1950 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1951 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1956 if (state && state->start == start) {
1957 private = state->private;
1960 ret = get_state_private(io_tree, start, &private);
1962 kaddr = kmap_atomic(page, KM_USER0);
1966 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1967 btrfs_csum_final(csum, (char *)&csum);
1968 if (csum != private)
1971 kunmap_atomic(kaddr, KM_USER0);
1973 /* if the io failure tree for this inode is non-empty,
1974 * check to see if we've recovered from a failed IO
1976 btrfs_clean_io_failures(inode, start);
1980 if (printk_ratelimit()) {
1981 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1982 "private %llu\n", page->mapping->host->i_ino,
1983 (unsigned long long)start, csum,
1984 (unsigned long long)private);
1986 memset(kaddr + offset, 1, end - start + 1);
1987 flush_dcache_page(page);
1988 kunmap_atomic(kaddr, KM_USER0);
1994 struct delayed_iput {
1995 struct list_head list;
1996 struct inode *inode;
1999 void btrfs_add_delayed_iput(struct inode *inode)
2001 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2002 struct delayed_iput *delayed;
2004 if (atomic_add_unless(&inode->i_count, -1, 1))
2007 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2008 delayed->inode = inode;
2010 spin_lock(&fs_info->delayed_iput_lock);
2011 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2012 spin_unlock(&fs_info->delayed_iput_lock);
2015 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2018 struct btrfs_fs_info *fs_info = root->fs_info;
2019 struct delayed_iput *delayed;
2022 spin_lock(&fs_info->delayed_iput_lock);
2023 empty = list_empty(&fs_info->delayed_iputs);
2024 spin_unlock(&fs_info->delayed_iput_lock);
2028 down_read(&root->fs_info->cleanup_work_sem);
2029 spin_lock(&fs_info->delayed_iput_lock);
2030 list_splice_init(&fs_info->delayed_iputs, &list);
2031 spin_unlock(&fs_info->delayed_iput_lock);
2033 while (!list_empty(&list)) {
2034 delayed = list_entry(list.next, struct delayed_iput, list);
2035 list_del(&delayed->list);
2036 iput(delayed->inode);
2039 up_read(&root->fs_info->cleanup_work_sem);
2043 * This creates an orphan entry for the given inode in case something goes
2044 * wrong in the middle of an unlink/truncate.
2046 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2048 struct btrfs_root *root = BTRFS_I(inode)->root;
2051 spin_lock(&root->list_lock);
2053 /* already on the orphan list, we're good */
2054 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2055 spin_unlock(&root->list_lock);
2059 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2061 spin_unlock(&root->list_lock);
2064 * insert an orphan item to track this unlinked/truncated file
2066 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2072 * We have done the truncate/delete so we can go ahead and remove the orphan
2073 * item for this particular inode.
2075 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2077 struct btrfs_root *root = BTRFS_I(inode)->root;
2080 spin_lock(&root->list_lock);
2082 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2083 spin_unlock(&root->list_lock);
2087 list_del_init(&BTRFS_I(inode)->i_orphan);
2089 spin_unlock(&root->list_lock);
2093 spin_unlock(&root->list_lock);
2095 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2101 * this cleans up any orphans that may be left on the list from the last use
2104 void btrfs_orphan_cleanup(struct btrfs_root *root)
2106 struct btrfs_path *path;
2107 struct extent_buffer *leaf;
2108 struct btrfs_item *item;
2109 struct btrfs_key key, found_key;
2110 struct btrfs_trans_handle *trans;
2111 struct inode *inode;
2112 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2114 if (!xchg(&root->clean_orphans, 0))
2117 path = btrfs_alloc_path();
2121 key.objectid = BTRFS_ORPHAN_OBJECTID;
2122 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2123 key.offset = (u64)-1;
2126 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2128 printk(KERN_ERR "Error searching slot for orphan: %d"
2134 * if ret == 0 means we found what we were searching for, which
2135 * is weird, but possible, so only screw with path if we didnt
2136 * find the key and see if we have stuff that matches
2139 if (path->slots[0] == 0)
2144 /* pull out the item */
2145 leaf = path->nodes[0];
2146 item = btrfs_item_nr(leaf, path->slots[0]);
2147 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2149 /* make sure the item matches what we want */
2150 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2152 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2155 /* release the path since we're done with it */
2156 btrfs_release_path(root, path);
2159 * this is where we are basically btrfs_lookup, without the
2160 * crossing root thing. we store the inode number in the
2161 * offset of the orphan item.
2163 found_key.objectid = found_key.offset;
2164 found_key.type = BTRFS_INODE_ITEM_KEY;
2165 found_key.offset = 0;
2166 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2171 * add this inode to the orphan list so btrfs_orphan_del does
2172 * the proper thing when we hit it
2174 spin_lock(&root->list_lock);
2175 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2176 spin_unlock(&root->list_lock);
2179 * if this is a bad inode, means we actually succeeded in
2180 * removing the inode, but not the orphan record, which means
2181 * we need to manually delete the orphan since iput will just
2182 * do a destroy_inode
2184 if (is_bad_inode(inode)) {
2185 trans = btrfs_start_transaction(root, 1);
2186 btrfs_orphan_del(trans, inode);
2187 btrfs_end_transaction(trans, root);
2192 /* if we have links, this was a truncate, lets do that */
2193 if (inode->i_nlink) {
2195 btrfs_truncate(inode);
2200 /* this will do delete_inode and everything for us */
2205 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2207 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2209 btrfs_free_path(path);
2213 * very simple check to peek ahead in the leaf looking for xattrs. If we
2214 * don't find any xattrs, we know there can't be any acls.
2216 * slot is the slot the inode is in, objectid is the objectid of the inode
2218 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2219 int slot, u64 objectid)
2221 u32 nritems = btrfs_header_nritems(leaf);
2222 struct btrfs_key found_key;
2226 while (slot < nritems) {
2227 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2229 /* we found a different objectid, there must not be acls */
2230 if (found_key.objectid != objectid)
2233 /* we found an xattr, assume we've got an acl */
2234 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2238 * we found a key greater than an xattr key, there can't
2239 * be any acls later on
2241 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2248 * it goes inode, inode backrefs, xattrs, extents,
2249 * so if there are a ton of hard links to an inode there can
2250 * be a lot of backrefs. Don't waste time searching too hard,
2251 * this is just an optimization
2256 /* we hit the end of the leaf before we found an xattr or
2257 * something larger than an xattr. We have to assume the inode
2264 * read an inode from the btree into the in-memory inode
2266 static void btrfs_read_locked_inode(struct inode *inode)
2268 struct btrfs_path *path;
2269 struct extent_buffer *leaf;
2270 struct btrfs_inode_item *inode_item;
2271 struct btrfs_timespec *tspec;
2272 struct btrfs_root *root = BTRFS_I(inode)->root;
2273 struct btrfs_key location;
2275 u64 alloc_group_block;
2279 path = btrfs_alloc_path();
2281 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2283 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2287 leaf = path->nodes[0];
2288 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2289 struct btrfs_inode_item);
2291 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2292 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2293 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2294 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2295 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2297 tspec = btrfs_inode_atime(inode_item);
2298 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2299 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2301 tspec = btrfs_inode_mtime(inode_item);
2302 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2303 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2305 tspec = btrfs_inode_ctime(inode_item);
2306 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2307 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2309 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2310 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2311 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2312 inode->i_generation = BTRFS_I(inode)->generation;
2314 rdev = btrfs_inode_rdev(leaf, inode_item);
2316 BTRFS_I(inode)->index_cnt = (u64)-1;
2317 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2319 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2322 * try to precache a NULL acl entry for files that don't have
2323 * any xattrs or acls
2325 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2327 cache_no_acl(inode);
2329 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2330 alloc_group_block, 0);
2331 btrfs_free_path(path);
2334 switch (inode->i_mode & S_IFMT) {
2336 inode->i_mapping->a_ops = &btrfs_aops;
2337 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2338 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2339 inode->i_fop = &btrfs_file_operations;
2340 inode->i_op = &btrfs_file_inode_operations;
2343 inode->i_fop = &btrfs_dir_file_operations;
2344 if (root == root->fs_info->tree_root)
2345 inode->i_op = &btrfs_dir_ro_inode_operations;
2347 inode->i_op = &btrfs_dir_inode_operations;
2350 inode->i_op = &btrfs_symlink_inode_operations;
2351 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2352 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2355 inode->i_op = &btrfs_special_inode_operations;
2356 init_special_inode(inode, inode->i_mode, rdev);
2360 btrfs_update_iflags(inode);
2364 btrfs_free_path(path);
2365 make_bad_inode(inode);
2369 * given a leaf and an inode, copy the inode fields into the leaf
2371 static void fill_inode_item(struct btrfs_trans_handle *trans,
2372 struct extent_buffer *leaf,
2373 struct btrfs_inode_item *item,
2374 struct inode *inode)
2376 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2377 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2378 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2379 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2380 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2382 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2383 inode->i_atime.tv_sec);
2384 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2385 inode->i_atime.tv_nsec);
2387 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2388 inode->i_mtime.tv_sec);
2389 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2390 inode->i_mtime.tv_nsec);
2392 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2393 inode->i_ctime.tv_sec);
2394 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2395 inode->i_ctime.tv_nsec);
2397 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2398 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2399 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2400 btrfs_set_inode_transid(leaf, item, trans->transid);
2401 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2402 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2403 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2407 * copy everything in the in-memory inode into the btree.
2409 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2410 struct btrfs_root *root, struct inode *inode)
2412 struct btrfs_inode_item *inode_item;
2413 struct btrfs_path *path;
2414 struct extent_buffer *leaf;
2417 path = btrfs_alloc_path();
2419 path->leave_spinning = 1;
2420 ret = btrfs_lookup_inode(trans, root, path,
2421 &BTRFS_I(inode)->location, 1);
2428 btrfs_unlock_up_safe(path, 1);
2429 leaf = path->nodes[0];
2430 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2431 struct btrfs_inode_item);
2433 fill_inode_item(trans, leaf, inode_item, inode);
2434 btrfs_mark_buffer_dirty(leaf);
2435 btrfs_set_inode_last_trans(trans, inode);
2438 btrfs_free_path(path);
2444 * unlink helper that gets used here in inode.c and in the tree logging
2445 * recovery code. It remove a link in a directory with a given name, and
2446 * also drops the back refs in the inode to the directory
2448 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2449 struct btrfs_root *root,
2450 struct inode *dir, struct inode *inode,
2451 const char *name, int name_len)
2453 struct btrfs_path *path;
2455 struct extent_buffer *leaf;
2456 struct btrfs_dir_item *di;
2457 struct btrfs_key key;
2460 path = btrfs_alloc_path();
2466 path->leave_spinning = 1;
2467 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2468 name, name_len, -1);
2477 leaf = path->nodes[0];
2478 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2479 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2482 btrfs_release_path(root, path);
2484 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2486 dir->i_ino, &index);
2488 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2489 "inode %lu parent %lu\n", name_len, name,
2490 inode->i_ino, dir->i_ino);
2494 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2495 index, name, name_len, -1);
2504 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2505 btrfs_release_path(root, path);
2507 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2509 BUG_ON(ret != 0 && ret != -ENOENT);
2511 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2515 btrfs_free_path(path);
2519 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2520 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2521 btrfs_update_inode(trans, root, dir);
2522 btrfs_drop_nlink(inode);
2523 ret = btrfs_update_inode(trans, root, inode);
2528 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2530 struct btrfs_root *root;
2531 struct btrfs_trans_handle *trans;
2532 struct inode *inode = dentry->d_inode;
2534 unsigned long nr = 0;
2536 root = BTRFS_I(dir)->root;
2539 * 5 items for unlink inode
2542 ret = btrfs_reserve_metadata_space(root, 6);
2546 trans = btrfs_start_transaction(root, 1);
2547 if (IS_ERR(trans)) {
2548 btrfs_unreserve_metadata_space(root, 6);
2549 return PTR_ERR(trans);
2552 btrfs_set_trans_block_group(trans, dir);
2554 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2556 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2557 dentry->d_name.name, dentry->d_name.len);
2559 if (inode->i_nlink == 0)
2560 ret = btrfs_orphan_add(trans, inode);
2562 nr = trans->blocks_used;
2564 btrfs_end_transaction_throttle(trans, root);
2565 btrfs_unreserve_metadata_space(root, 6);
2566 btrfs_btree_balance_dirty(root, nr);
2570 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2571 struct btrfs_root *root,
2572 struct inode *dir, u64 objectid,
2573 const char *name, int name_len)
2575 struct btrfs_path *path;
2576 struct extent_buffer *leaf;
2577 struct btrfs_dir_item *di;
2578 struct btrfs_key key;
2582 path = btrfs_alloc_path();
2586 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2587 name, name_len, -1);
2588 BUG_ON(!di || IS_ERR(di));
2590 leaf = path->nodes[0];
2591 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2592 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2593 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2595 btrfs_release_path(root, path);
2597 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2598 objectid, root->root_key.objectid,
2599 dir->i_ino, &index, name, name_len);
2601 BUG_ON(ret != -ENOENT);
2602 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2604 BUG_ON(!di || IS_ERR(di));
2606 leaf = path->nodes[0];
2607 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2608 btrfs_release_path(root, path);
2612 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2613 index, name, name_len, -1);
2614 BUG_ON(!di || IS_ERR(di));
2616 leaf = path->nodes[0];
2617 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2618 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2619 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2621 btrfs_release_path(root, path);
2623 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2624 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2625 ret = btrfs_update_inode(trans, root, dir);
2627 dir->i_sb->s_dirt = 1;
2629 btrfs_free_path(path);
2633 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2635 struct inode *inode = dentry->d_inode;
2638 struct btrfs_root *root = BTRFS_I(dir)->root;
2639 struct btrfs_trans_handle *trans;
2640 unsigned long nr = 0;
2642 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2643 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2646 ret = btrfs_reserve_metadata_space(root, 5);
2650 trans = btrfs_start_transaction(root, 1);
2651 if (IS_ERR(trans)) {
2652 btrfs_unreserve_metadata_space(root, 5);
2653 return PTR_ERR(trans);
2656 btrfs_set_trans_block_group(trans, dir);
2658 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2659 err = btrfs_unlink_subvol(trans, root, dir,
2660 BTRFS_I(inode)->location.objectid,
2661 dentry->d_name.name,
2662 dentry->d_name.len);
2666 err = btrfs_orphan_add(trans, inode);
2670 /* now the directory is empty */
2671 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2672 dentry->d_name.name, dentry->d_name.len);
2674 btrfs_i_size_write(inode, 0);
2676 nr = trans->blocks_used;
2677 ret = btrfs_end_transaction_throttle(trans, root);
2678 btrfs_unreserve_metadata_space(root, 5);
2679 btrfs_btree_balance_dirty(root, nr);
2688 * when truncating bytes in a file, it is possible to avoid reading
2689 * the leaves that contain only checksum items. This can be the
2690 * majority of the IO required to delete a large file, but it must
2691 * be done carefully.
2693 * The keys in the level just above the leaves are checked to make sure
2694 * the lowest key in a given leaf is a csum key, and starts at an offset
2695 * after the new size.
2697 * Then the key for the next leaf is checked to make sure it also has
2698 * a checksum item for the same file. If it does, we know our target leaf
2699 * contains only checksum items, and it can be safely freed without reading
2702 * This is just an optimization targeted at large files. It may do
2703 * nothing. It will return 0 unless things went badly.
2705 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2706 struct btrfs_root *root,
2707 struct btrfs_path *path,
2708 struct inode *inode, u64 new_size)
2710 struct btrfs_key key;
2713 struct btrfs_key found_key;
2714 struct btrfs_key other_key;
2715 struct btrfs_leaf_ref *ref;
2719 path->lowest_level = 1;
2720 key.objectid = inode->i_ino;
2721 key.type = BTRFS_CSUM_ITEM_KEY;
2722 key.offset = new_size;
2724 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2728 if (path->nodes[1] == NULL) {
2733 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2734 nritems = btrfs_header_nritems(path->nodes[1]);
2739 if (path->slots[1] >= nritems)
2742 /* did we find a key greater than anything we want to delete? */
2743 if (found_key.objectid > inode->i_ino ||
2744 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2747 /* we check the next key in the node to make sure the leave contains
2748 * only checksum items. This comparison doesn't work if our
2749 * leaf is the last one in the node
2751 if (path->slots[1] + 1 >= nritems) {
2753 /* search forward from the last key in the node, this
2754 * will bring us into the next node in the tree
2756 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2758 /* unlikely, but we inc below, so check to be safe */
2759 if (found_key.offset == (u64)-1)
2762 /* search_forward needs a path with locks held, do the
2763 * search again for the original key. It is possible
2764 * this will race with a balance and return a path that
2765 * we could modify, but this drop is just an optimization
2766 * and is allowed to miss some leaves.
2768 btrfs_release_path(root, path);
2771 /* setup a max key for search_forward */
2772 other_key.offset = (u64)-1;
2773 other_key.type = key.type;
2774 other_key.objectid = key.objectid;
2776 path->keep_locks = 1;
2777 ret = btrfs_search_forward(root, &found_key, &other_key,
2779 path->keep_locks = 0;
2780 if (ret || found_key.objectid != key.objectid ||
2781 found_key.type != key.type) {
2786 key.offset = found_key.offset;
2787 btrfs_release_path(root, path);
2792 /* we know there's one more slot after us in the tree,
2793 * read that key so we can verify it is also a checksum item
2795 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2797 if (found_key.objectid < inode->i_ino)
2800 if (found_key.type != key.type || found_key.offset < new_size)
2804 * if the key for the next leaf isn't a csum key from this objectid,
2805 * we can't be sure there aren't good items inside this leaf.
2808 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2811 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2812 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2814 * it is safe to delete this leaf, it contains only
2815 * csum items from this inode at an offset >= new_size
2817 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2820 if (root->ref_cows && leaf_gen < trans->transid) {
2821 ref = btrfs_alloc_leaf_ref(root, 0);
2823 ref->root_gen = root->root_key.offset;
2824 ref->bytenr = leaf_start;
2826 ref->generation = leaf_gen;
2829 btrfs_sort_leaf_ref(ref);
2831 ret = btrfs_add_leaf_ref(root, ref, 0);
2833 btrfs_free_leaf_ref(root, ref);
2839 btrfs_release_path(root, path);
2841 if (other_key.objectid == inode->i_ino &&
2842 other_key.type == key.type && other_key.offset > key.offset) {
2843 key.offset = other_key.offset;
2849 /* fixup any changes we've made to the path */
2850 path->lowest_level = 0;
2851 path->keep_locks = 0;
2852 btrfs_release_path(root, path);
2859 * this can truncate away extent items, csum items and directory items.
2860 * It starts at a high offset and removes keys until it can't find
2861 * any higher than new_size
2863 * csum items that cross the new i_size are truncated to the new size
2866 * min_type is the minimum key type to truncate down to. If set to 0, this
2867 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2869 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2870 struct btrfs_root *root,
2871 struct inode *inode,
2872 u64 new_size, u32 min_type)
2874 struct btrfs_path *path;
2875 struct extent_buffer *leaf;
2876 struct btrfs_file_extent_item *fi;
2877 struct btrfs_key key;
2878 struct btrfs_key found_key;
2879 u64 extent_start = 0;
2880 u64 extent_num_bytes = 0;
2881 u64 extent_offset = 0;
2883 u64 mask = root->sectorsize - 1;
2884 u32 found_type = (u8)-1;
2887 int pending_del_nr = 0;
2888 int pending_del_slot = 0;
2889 int extent_type = -1;
2894 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2897 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2899 path = btrfs_alloc_path();
2903 key.objectid = inode->i_ino;
2904 key.offset = (u64)-1;
2908 path->leave_spinning = 1;
2909 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2916 /* there are no items in the tree for us to truncate, we're
2919 if (path->slots[0] == 0)
2926 leaf = path->nodes[0];
2927 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2928 found_type = btrfs_key_type(&found_key);
2931 if (found_key.objectid != inode->i_ino)
2934 if (found_type < min_type)
2937 item_end = found_key.offset;
2938 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2939 fi = btrfs_item_ptr(leaf, path->slots[0],
2940 struct btrfs_file_extent_item);
2941 extent_type = btrfs_file_extent_type(leaf, fi);
2942 encoding = btrfs_file_extent_compression(leaf, fi);
2943 encoding |= btrfs_file_extent_encryption(leaf, fi);
2944 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2946 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2948 btrfs_file_extent_num_bytes(leaf, fi);
2949 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2950 item_end += btrfs_file_extent_inline_len(leaf,
2955 if (found_type > min_type) {
2958 if (item_end < new_size)
2960 if (found_key.offset >= new_size)
2966 /* FIXME, shrink the extent if the ref count is only 1 */
2967 if (found_type != BTRFS_EXTENT_DATA_KEY)
2970 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2972 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
2973 if (!del_item && !encoding) {
2974 u64 orig_num_bytes =
2975 btrfs_file_extent_num_bytes(leaf, fi);
2976 extent_num_bytes = new_size -
2977 found_key.offset + root->sectorsize - 1;
2978 extent_num_bytes = extent_num_bytes &
2979 ~((u64)root->sectorsize - 1);
2980 btrfs_set_file_extent_num_bytes(leaf, fi,
2982 num_dec = (orig_num_bytes -
2984 if (root->ref_cows && extent_start != 0)
2985 inode_sub_bytes(inode, num_dec);
2986 btrfs_mark_buffer_dirty(leaf);
2989 btrfs_file_extent_disk_num_bytes(leaf,
2991 extent_offset = found_key.offset -
2992 btrfs_file_extent_offset(leaf, fi);
2994 /* FIXME blocksize != 4096 */
2995 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
2996 if (extent_start != 0) {
2999 inode_sub_bytes(inode, num_dec);
3002 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3004 * we can't truncate inline items that have had
3008 btrfs_file_extent_compression(leaf, fi) == 0 &&
3009 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3010 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3011 u32 size = new_size - found_key.offset;
3013 if (root->ref_cows) {
3014 inode_sub_bytes(inode, item_end + 1 -
3018 btrfs_file_extent_calc_inline_size(size);
3019 ret = btrfs_truncate_item(trans, root, path,
3022 } else if (root->ref_cows) {
3023 inode_sub_bytes(inode, item_end + 1 -
3029 if (!pending_del_nr) {
3030 /* no pending yet, add ourselves */
3031 pending_del_slot = path->slots[0];
3033 } else if (pending_del_nr &&
3034 path->slots[0] + 1 == pending_del_slot) {
3035 /* hop on the pending chunk */
3037 pending_del_slot = path->slots[0];
3044 if (found_extent && root->ref_cows) {
3045 btrfs_set_path_blocking(path);
3046 ret = btrfs_free_extent(trans, root, extent_start,
3047 extent_num_bytes, 0,
3048 btrfs_header_owner(leaf),
3049 inode->i_ino, extent_offset);
3053 if (found_type == BTRFS_INODE_ITEM_KEY)
3056 if (path->slots[0] == 0 ||
3057 path->slots[0] != pending_del_slot) {
3058 if (root->ref_cows) {
3062 if (pending_del_nr) {
3063 ret = btrfs_del_items(trans, root, path,
3069 btrfs_release_path(root, path);
3076 if (pending_del_nr) {
3077 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3080 btrfs_free_path(path);
3085 * taken from block_truncate_page, but does cow as it zeros out
3086 * any bytes left in the last page in the file.
3088 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3090 struct inode *inode = mapping->host;
3091 struct btrfs_root *root = BTRFS_I(inode)->root;
3092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3093 struct btrfs_ordered_extent *ordered;
3094 struct extent_state *cached_state = NULL;
3096 u32 blocksize = root->sectorsize;
3097 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3098 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3104 if ((offset & (blocksize - 1)) == 0)
3106 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3110 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3116 page = grab_cache_page(mapping, index);
3118 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3119 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3123 page_start = page_offset(page);
3124 page_end = page_start + PAGE_CACHE_SIZE - 1;
3126 if (!PageUptodate(page)) {
3127 ret = btrfs_readpage(NULL, page);
3129 if (page->mapping != mapping) {
3131 page_cache_release(page);
3134 if (!PageUptodate(page)) {
3139 wait_on_page_writeback(page);
3141 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3143 set_page_extent_mapped(page);
3145 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3147 unlock_extent_cached(io_tree, page_start, page_end,
3148 &cached_state, GFP_NOFS);
3150 page_cache_release(page);
3151 btrfs_start_ordered_extent(inode, ordered, 1);
3152 btrfs_put_ordered_extent(ordered);
3156 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3157 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3158 0, 0, &cached_state, GFP_NOFS);
3160 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3163 unlock_extent_cached(io_tree, page_start, page_end,
3164 &cached_state, GFP_NOFS);
3169 if (offset != PAGE_CACHE_SIZE) {
3171 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3172 flush_dcache_page(page);
3175 ClearPageChecked(page);
3176 set_page_dirty(page);
3177 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3182 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3183 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3185 page_cache_release(page);
3190 int btrfs_cont_expand(struct inode *inode, loff_t size)
3192 struct btrfs_trans_handle *trans;
3193 struct btrfs_root *root = BTRFS_I(inode)->root;
3194 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3195 struct extent_map *em;
3196 struct extent_state *cached_state = NULL;
3197 u64 mask = root->sectorsize - 1;
3198 u64 hole_start = (inode->i_size + mask) & ~mask;
3199 u64 block_end = (size + mask) & ~mask;
3205 if (size <= hole_start)
3209 struct btrfs_ordered_extent *ordered;
3210 btrfs_wait_ordered_range(inode, hole_start,
3211 block_end - hole_start);
3212 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3213 &cached_state, GFP_NOFS);
3214 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3217 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3218 &cached_state, GFP_NOFS);
3219 btrfs_put_ordered_extent(ordered);
3222 cur_offset = hole_start;
3224 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3225 block_end - cur_offset, 0);
3226 BUG_ON(IS_ERR(em) || !em);
3227 last_byte = min(extent_map_end(em), block_end);
3228 last_byte = (last_byte + mask) & ~mask;
3229 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3231 hole_size = last_byte - cur_offset;
3233 err = btrfs_reserve_metadata_space(root, 2);
3237 trans = btrfs_start_transaction(root, 1);
3238 btrfs_set_trans_block_group(trans, inode);
3240 err = btrfs_drop_extents(trans, inode, cur_offset,
3241 cur_offset + hole_size,
3245 err = btrfs_insert_file_extent(trans, root,
3246 inode->i_ino, cur_offset, 0,
3247 0, hole_size, 0, hole_size,
3251 btrfs_drop_extent_cache(inode, hole_start,
3254 btrfs_end_transaction(trans, root);
3255 btrfs_unreserve_metadata_space(root, 2);
3257 free_extent_map(em);
3258 cur_offset = last_byte;
3259 if (cur_offset >= block_end)
3263 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3268 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3270 struct btrfs_root *root = BTRFS_I(inode)->root;
3271 struct btrfs_trans_handle *trans;
3275 if (attr->ia_size == inode->i_size)
3278 if (attr->ia_size > inode->i_size) {
3279 unsigned long limit;
3280 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3281 if (attr->ia_size > inode->i_sb->s_maxbytes)
3283 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3284 send_sig(SIGXFSZ, current, 0);
3289 ret = btrfs_reserve_metadata_space(root, 1);
3293 trans = btrfs_start_transaction(root, 1);
3294 btrfs_set_trans_block_group(trans, inode);
3296 ret = btrfs_orphan_add(trans, inode);
3299 nr = trans->blocks_used;
3300 btrfs_end_transaction(trans, root);
3301 btrfs_unreserve_metadata_space(root, 1);
3302 btrfs_btree_balance_dirty(root, nr);
3304 if (attr->ia_size > inode->i_size) {
3305 ret = btrfs_cont_expand(inode, attr->ia_size);
3307 btrfs_truncate(inode);
3311 i_size_write(inode, attr->ia_size);
3312 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3314 trans = btrfs_start_transaction(root, 1);
3315 btrfs_set_trans_block_group(trans, inode);
3317 ret = btrfs_update_inode(trans, root, inode);
3319 if (inode->i_nlink > 0) {
3320 ret = btrfs_orphan_del(trans, inode);
3323 nr = trans->blocks_used;
3324 btrfs_end_transaction(trans, root);
3325 btrfs_btree_balance_dirty(root, nr);
3330 * We're truncating a file that used to have good data down to
3331 * zero. Make sure it gets into the ordered flush list so that
3332 * any new writes get down to disk quickly.
3334 if (attr->ia_size == 0)
3335 BTRFS_I(inode)->ordered_data_close = 1;
3337 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3338 ret = vmtruncate(inode, attr->ia_size);
3344 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3346 struct inode *inode = dentry->d_inode;
3349 err = inode_change_ok(inode, attr);
3353 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3354 err = btrfs_setattr_size(inode, attr);
3358 attr->ia_valid &= ~ATTR_SIZE;
3361 err = inode_setattr(inode, attr);
3363 if (!err && ((attr->ia_valid & ATTR_MODE)))
3364 err = btrfs_acl_chmod(inode);
3368 void btrfs_delete_inode(struct inode *inode)
3370 struct btrfs_trans_handle *trans;
3371 struct btrfs_root *root = BTRFS_I(inode)->root;
3375 truncate_inode_pages(&inode->i_data, 0);
3376 if (is_bad_inode(inode)) {
3377 btrfs_orphan_del(NULL, inode);
3380 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3382 if (root->fs_info->log_root_recovering) {
3383 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3387 if (inode->i_nlink > 0) {
3388 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3392 btrfs_i_size_write(inode, 0);
3395 trans = btrfs_start_transaction(root, 1);
3396 btrfs_set_trans_block_group(trans, inode);
3397 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3402 nr = trans->blocks_used;
3403 btrfs_end_transaction(trans, root);
3405 btrfs_btree_balance_dirty(root, nr);
3409 ret = btrfs_orphan_del(trans, inode);
3413 nr = trans->blocks_used;
3414 btrfs_end_transaction(trans, root);
3415 btrfs_btree_balance_dirty(root, nr);
3422 * this returns the key found in the dir entry in the location pointer.
3423 * If no dir entries were found, location->objectid is 0.
3425 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3426 struct btrfs_key *location)
3428 const char *name = dentry->d_name.name;
3429 int namelen = dentry->d_name.len;
3430 struct btrfs_dir_item *di;
3431 struct btrfs_path *path;
3432 struct btrfs_root *root = BTRFS_I(dir)->root;
3435 path = btrfs_alloc_path();
3438 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3443 if (!di || IS_ERR(di))
3446 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3448 btrfs_free_path(path);
3451 location->objectid = 0;
3456 * when we hit a tree root in a directory, the btrfs part of the inode
3457 * needs to be changed to reflect the root directory of the tree root. This
3458 * is kind of like crossing a mount point.
3460 static int fixup_tree_root_location(struct btrfs_root *root,
3462 struct dentry *dentry,
3463 struct btrfs_key *location,
3464 struct btrfs_root **sub_root)
3466 struct btrfs_path *path;
3467 struct btrfs_root *new_root;
3468 struct btrfs_root_ref *ref;
3469 struct extent_buffer *leaf;
3473 path = btrfs_alloc_path();
3480 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3481 BTRFS_I(dir)->root->root_key.objectid,
3482 location->objectid);
3489 leaf = path->nodes[0];
3490 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3491 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3492 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3495 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3496 (unsigned long)(ref + 1),
3497 dentry->d_name.len);
3501 btrfs_release_path(root->fs_info->tree_root, path);
3503 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3504 if (IS_ERR(new_root)) {
3505 err = PTR_ERR(new_root);
3509 if (btrfs_root_refs(&new_root->root_item) == 0) {
3514 *sub_root = new_root;
3515 location->objectid = btrfs_root_dirid(&new_root->root_item);
3516 location->type = BTRFS_INODE_ITEM_KEY;
3517 location->offset = 0;
3520 btrfs_free_path(path);
3524 static void inode_tree_add(struct inode *inode)
3526 struct btrfs_root *root = BTRFS_I(inode)->root;
3527 struct btrfs_inode *entry;
3529 struct rb_node *parent;
3531 p = &root->inode_tree.rb_node;
3534 if (hlist_unhashed(&inode->i_hash))
3537 spin_lock(&root->inode_lock);
3540 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3542 if (inode->i_ino < entry->vfs_inode.i_ino)
3543 p = &parent->rb_left;
3544 else if (inode->i_ino > entry->vfs_inode.i_ino)
3545 p = &parent->rb_right;
3547 WARN_ON(!(entry->vfs_inode.i_state &
3548 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3549 rb_erase(parent, &root->inode_tree);
3550 RB_CLEAR_NODE(parent);
3551 spin_unlock(&root->inode_lock);
3555 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3556 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3557 spin_unlock(&root->inode_lock);
3560 static void inode_tree_del(struct inode *inode)
3562 struct btrfs_root *root = BTRFS_I(inode)->root;
3565 spin_lock(&root->inode_lock);
3566 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3567 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3568 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3569 empty = RB_EMPTY_ROOT(&root->inode_tree);
3571 spin_unlock(&root->inode_lock);
3573 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3574 synchronize_srcu(&root->fs_info->subvol_srcu);
3575 spin_lock(&root->inode_lock);
3576 empty = RB_EMPTY_ROOT(&root->inode_tree);
3577 spin_unlock(&root->inode_lock);
3579 btrfs_add_dead_root(root);
3583 int btrfs_invalidate_inodes(struct btrfs_root *root)
3585 struct rb_node *node;
3586 struct rb_node *prev;
3587 struct btrfs_inode *entry;
3588 struct inode *inode;
3591 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3593 spin_lock(&root->inode_lock);
3595 node = root->inode_tree.rb_node;
3599 entry = rb_entry(node, struct btrfs_inode, rb_node);
3601 if (objectid < entry->vfs_inode.i_ino)
3602 node = node->rb_left;
3603 else if (objectid > entry->vfs_inode.i_ino)
3604 node = node->rb_right;
3610 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3611 if (objectid <= entry->vfs_inode.i_ino) {
3615 prev = rb_next(prev);
3619 entry = rb_entry(node, struct btrfs_inode, rb_node);
3620 objectid = entry->vfs_inode.i_ino + 1;
3621 inode = igrab(&entry->vfs_inode);
3623 spin_unlock(&root->inode_lock);
3624 if (atomic_read(&inode->i_count) > 1)
3625 d_prune_aliases(inode);
3627 * btrfs_drop_inode will remove it from
3628 * the inode cache when its usage count
3633 spin_lock(&root->inode_lock);
3637 if (cond_resched_lock(&root->inode_lock))
3640 node = rb_next(node);
3642 spin_unlock(&root->inode_lock);
3646 static noinline void init_btrfs_i(struct inode *inode)
3648 struct btrfs_inode *bi = BTRFS_I(inode);
3653 bi->last_sub_trans = 0;
3654 bi->logged_trans = 0;
3655 bi->delalloc_bytes = 0;
3656 bi->reserved_bytes = 0;
3657 bi->disk_i_size = 0;
3659 bi->index_cnt = (u64)-1;
3660 bi->last_unlink_trans = 0;
3661 bi->ordered_data_close = 0;
3662 bi->force_compress = 0;
3663 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3664 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3665 inode->i_mapping, GFP_NOFS);
3666 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3667 inode->i_mapping, GFP_NOFS);
3668 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3669 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3670 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3671 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3672 mutex_init(&BTRFS_I(inode)->log_mutex);
3675 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3677 struct btrfs_iget_args *args = p;
3678 inode->i_ino = args->ino;
3679 init_btrfs_i(inode);
3680 BTRFS_I(inode)->root = args->root;
3681 btrfs_set_inode_space_info(args->root, inode);
3685 static int btrfs_find_actor(struct inode *inode, void *opaque)
3687 struct btrfs_iget_args *args = opaque;
3688 return args->ino == inode->i_ino &&
3689 args->root == BTRFS_I(inode)->root;
3692 static struct inode *btrfs_iget_locked(struct super_block *s,
3694 struct btrfs_root *root)
3696 struct inode *inode;
3697 struct btrfs_iget_args args;
3698 args.ino = objectid;
3701 inode = iget5_locked(s, objectid, btrfs_find_actor,
3702 btrfs_init_locked_inode,
3707 /* Get an inode object given its location and corresponding root.
3708 * Returns in *is_new if the inode was read from disk
3710 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3711 struct btrfs_root *root, int *new)
3713 struct inode *inode;
3715 inode = btrfs_iget_locked(s, location->objectid, root);
3717 return ERR_PTR(-ENOMEM);
3719 if (inode->i_state & I_NEW) {
3720 BTRFS_I(inode)->root = root;
3721 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3722 btrfs_read_locked_inode(inode);
3724 inode_tree_add(inode);
3725 unlock_new_inode(inode);
3733 static struct inode *new_simple_dir(struct super_block *s,
3734 struct btrfs_key *key,
3735 struct btrfs_root *root)
3737 struct inode *inode = new_inode(s);
3740 return ERR_PTR(-ENOMEM);
3742 init_btrfs_i(inode);
3744 BTRFS_I(inode)->root = root;
3745 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3746 BTRFS_I(inode)->dummy_inode = 1;
3748 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3749 inode->i_op = &simple_dir_inode_operations;
3750 inode->i_fop = &simple_dir_operations;
3751 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3752 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3757 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3759 struct inode *inode;
3760 struct btrfs_root *root = BTRFS_I(dir)->root;
3761 struct btrfs_root *sub_root = root;
3762 struct btrfs_key location;
3766 dentry->d_op = &btrfs_dentry_operations;
3768 if (dentry->d_name.len > BTRFS_NAME_LEN)
3769 return ERR_PTR(-ENAMETOOLONG);
3771 ret = btrfs_inode_by_name(dir, dentry, &location);
3774 return ERR_PTR(ret);
3776 if (location.objectid == 0)
3779 if (location.type == BTRFS_INODE_ITEM_KEY) {
3780 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3784 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3786 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3787 ret = fixup_tree_root_location(root, dir, dentry,
3788 &location, &sub_root);
3791 inode = ERR_PTR(ret);
3793 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3795 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3797 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3799 if (root != sub_root) {
3800 down_read(&root->fs_info->cleanup_work_sem);
3801 if (!(inode->i_sb->s_flags & MS_RDONLY))
3802 btrfs_orphan_cleanup(sub_root);
3803 up_read(&root->fs_info->cleanup_work_sem);
3809 static int btrfs_dentry_delete(struct dentry *dentry)
3811 struct btrfs_root *root;
3813 if (!dentry->d_inode && !IS_ROOT(dentry))
3814 dentry = dentry->d_parent;
3816 if (dentry->d_inode) {
3817 root = BTRFS_I(dentry->d_inode)->root;
3818 if (btrfs_root_refs(&root->root_item) == 0)
3824 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3825 struct nameidata *nd)
3827 struct inode *inode;
3829 inode = btrfs_lookup_dentry(dir, dentry);
3831 return ERR_CAST(inode);
3833 return d_splice_alias(inode, dentry);
3836 static unsigned char btrfs_filetype_table[] = {
3837 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3840 static int btrfs_real_readdir(struct file *filp, void *dirent,
3843 struct inode *inode = filp->f_dentry->d_inode;
3844 struct btrfs_root *root = BTRFS_I(inode)->root;
3845 struct btrfs_item *item;
3846 struct btrfs_dir_item *di;
3847 struct btrfs_key key;
3848 struct btrfs_key found_key;
3849 struct btrfs_path *path;
3852 struct extent_buffer *leaf;
3855 unsigned char d_type;
3860 int key_type = BTRFS_DIR_INDEX_KEY;
3865 /* FIXME, use a real flag for deciding about the key type */
3866 if (root->fs_info->tree_root == root)
3867 key_type = BTRFS_DIR_ITEM_KEY;
3869 /* special case for "." */
3870 if (filp->f_pos == 0) {
3871 over = filldir(dirent, ".", 1,
3878 /* special case for .., just use the back ref */
3879 if (filp->f_pos == 1) {
3880 u64 pino = parent_ino(filp->f_path.dentry);
3881 over = filldir(dirent, "..", 2,
3887 path = btrfs_alloc_path();
3890 btrfs_set_key_type(&key, key_type);
3891 key.offset = filp->f_pos;
3892 key.objectid = inode->i_ino;
3894 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3900 leaf = path->nodes[0];
3901 nritems = btrfs_header_nritems(leaf);
3902 slot = path->slots[0];
3903 if (advance || slot >= nritems) {
3904 if (slot >= nritems - 1) {
3905 ret = btrfs_next_leaf(root, path);
3908 leaf = path->nodes[0];
3909 nritems = btrfs_header_nritems(leaf);
3910 slot = path->slots[0];
3918 item = btrfs_item_nr(leaf, slot);
3919 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3921 if (found_key.objectid != key.objectid)
3923 if (btrfs_key_type(&found_key) != key_type)
3925 if (found_key.offset < filp->f_pos)
3928 filp->f_pos = found_key.offset;
3930 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3932 di_total = btrfs_item_size(leaf, item);
3934 while (di_cur < di_total) {
3935 struct btrfs_key location;
3937 name_len = btrfs_dir_name_len(leaf, di);
3938 if (name_len <= sizeof(tmp_name)) {
3939 name_ptr = tmp_name;
3941 name_ptr = kmalloc(name_len, GFP_NOFS);
3947 read_extent_buffer(leaf, name_ptr,
3948 (unsigned long)(di + 1), name_len);
3950 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3951 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3953 /* is this a reference to our own snapshot? If so
3956 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3957 location.objectid == root->root_key.objectid) {
3961 over = filldir(dirent, name_ptr, name_len,
3962 found_key.offset, location.objectid,
3966 if (name_ptr != tmp_name)
3971 di_len = btrfs_dir_name_len(leaf, di) +
3972 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3974 di = (struct btrfs_dir_item *)((char *)di + di_len);
3978 /* Reached end of directory/root. Bump pos past the last item. */
3979 if (key_type == BTRFS_DIR_INDEX_KEY)
3981 * 32-bit glibc will use getdents64, but then strtol -
3982 * so the last number we can serve is this.
3984 filp->f_pos = 0x7fffffff;
3990 btrfs_free_path(path);
3994 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
3996 struct btrfs_root *root = BTRFS_I(inode)->root;
3997 struct btrfs_trans_handle *trans;
4000 if (root->fs_info->btree_inode == inode)
4003 if (wbc->sync_mode == WB_SYNC_ALL) {
4004 trans = btrfs_join_transaction(root, 1);
4005 btrfs_set_trans_block_group(trans, inode);
4006 ret = btrfs_commit_transaction(trans, root);
4012 * This is somewhat expensive, updating the tree every time the
4013 * inode changes. But, it is most likely to find the inode in cache.
4014 * FIXME, needs more benchmarking...there are no reasons other than performance
4015 * to keep or drop this code.
4017 void btrfs_dirty_inode(struct inode *inode)
4019 struct btrfs_root *root = BTRFS_I(inode)->root;
4020 struct btrfs_trans_handle *trans;
4022 trans = btrfs_join_transaction(root, 1);
4023 btrfs_set_trans_block_group(trans, inode);
4024 btrfs_update_inode(trans, root, inode);
4025 btrfs_end_transaction(trans, root);
4029 * find the highest existing sequence number in a directory
4030 * and then set the in-memory index_cnt variable to reflect
4031 * free sequence numbers
4033 static int btrfs_set_inode_index_count(struct inode *inode)
4035 struct btrfs_root *root = BTRFS_I(inode)->root;
4036 struct btrfs_key key, found_key;
4037 struct btrfs_path *path;
4038 struct extent_buffer *leaf;
4041 key.objectid = inode->i_ino;
4042 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4043 key.offset = (u64)-1;
4045 path = btrfs_alloc_path();
4049 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4052 /* FIXME: we should be able to handle this */
4058 * MAGIC NUMBER EXPLANATION:
4059 * since we search a directory based on f_pos we have to start at 2
4060 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4061 * else has to start at 2
4063 if (path->slots[0] == 0) {
4064 BTRFS_I(inode)->index_cnt = 2;
4070 leaf = path->nodes[0];
4071 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4073 if (found_key.objectid != inode->i_ino ||
4074 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4075 BTRFS_I(inode)->index_cnt = 2;
4079 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4081 btrfs_free_path(path);
4086 * helper to find a free sequence number in a given directory. This current
4087 * code is very simple, later versions will do smarter things in the btree
4089 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4093 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4094 ret = btrfs_set_inode_index_count(dir);
4099 *index = BTRFS_I(dir)->index_cnt;
4100 BTRFS_I(dir)->index_cnt++;
4105 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4106 struct btrfs_root *root,
4108 const char *name, int name_len,
4109 u64 ref_objectid, u64 objectid,
4110 u64 alloc_hint, int mode, u64 *index)
4112 struct inode *inode;
4113 struct btrfs_inode_item *inode_item;
4114 struct btrfs_key *location;
4115 struct btrfs_path *path;
4116 struct btrfs_inode_ref *ref;
4117 struct btrfs_key key[2];
4123 path = btrfs_alloc_path();
4126 inode = new_inode(root->fs_info->sb);
4128 return ERR_PTR(-ENOMEM);
4131 ret = btrfs_set_inode_index(dir, index);
4134 return ERR_PTR(ret);
4138 * index_cnt is ignored for everything but a dir,
4139 * btrfs_get_inode_index_count has an explanation for the magic
4142 init_btrfs_i(inode);
4143 BTRFS_I(inode)->index_cnt = 2;
4144 BTRFS_I(inode)->root = root;
4145 BTRFS_I(inode)->generation = trans->transid;
4146 btrfs_set_inode_space_info(root, inode);
4152 BTRFS_I(inode)->block_group =
4153 btrfs_find_block_group(root, 0, alloc_hint, owner);
4155 key[0].objectid = objectid;
4156 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4159 key[1].objectid = objectid;
4160 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4161 key[1].offset = ref_objectid;
4163 sizes[0] = sizeof(struct btrfs_inode_item);
4164 sizes[1] = name_len + sizeof(*ref);
4166 path->leave_spinning = 1;
4167 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4171 inode->i_uid = current_fsuid();
4173 if (dir && (dir->i_mode & S_ISGID)) {
4174 inode->i_gid = dir->i_gid;
4178 inode->i_gid = current_fsgid();
4180 inode->i_mode = mode;
4181 inode->i_ino = objectid;
4182 inode_set_bytes(inode, 0);
4183 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4184 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4185 struct btrfs_inode_item);
4186 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4188 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4189 struct btrfs_inode_ref);
4190 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4191 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4192 ptr = (unsigned long)(ref + 1);
4193 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4195 btrfs_mark_buffer_dirty(path->nodes[0]);
4196 btrfs_free_path(path);
4198 location = &BTRFS_I(inode)->location;
4199 location->objectid = objectid;
4200 location->offset = 0;
4201 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4203 btrfs_inherit_iflags(inode, dir);
4205 if ((mode & S_IFREG)) {
4206 if (btrfs_test_opt(root, NODATASUM))
4207 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4208 if (btrfs_test_opt(root, NODATACOW))
4209 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4212 insert_inode_hash(inode);
4213 inode_tree_add(inode);
4217 BTRFS_I(dir)->index_cnt--;
4218 btrfs_free_path(path);
4220 return ERR_PTR(ret);
4223 static inline u8 btrfs_inode_type(struct inode *inode)
4225 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4229 * utility function to add 'inode' into 'parent_inode' with
4230 * a give name and a given sequence number.
4231 * if 'add_backref' is true, also insert a backref from the
4232 * inode to the parent directory.
4234 int btrfs_add_link(struct btrfs_trans_handle *trans,
4235 struct inode *parent_inode, struct inode *inode,
4236 const char *name, int name_len, int add_backref, u64 index)
4239 struct btrfs_key key;
4240 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4242 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4243 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4245 key.objectid = inode->i_ino;
4246 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4250 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4251 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4252 key.objectid, root->root_key.objectid,
4253 parent_inode->i_ino,
4254 index, name, name_len);
4255 } else if (add_backref) {
4256 ret = btrfs_insert_inode_ref(trans, root,
4257 name, name_len, inode->i_ino,
4258 parent_inode->i_ino, index);
4262 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4263 parent_inode->i_ino, &key,
4264 btrfs_inode_type(inode), index);
4267 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4269 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4270 ret = btrfs_update_inode(trans, root, parent_inode);
4275 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4276 struct dentry *dentry, struct inode *inode,
4277 int backref, u64 index)
4279 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4280 inode, dentry->d_name.name,
4281 dentry->d_name.len, backref, index);
4283 d_instantiate(dentry, inode);
4291 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4292 int mode, dev_t rdev)
4294 struct btrfs_trans_handle *trans;
4295 struct btrfs_root *root = BTRFS_I(dir)->root;
4296 struct inode *inode = NULL;
4300 unsigned long nr = 0;
4303 if (!new_valid_dev(rdev))
4307 * 2 for inode item and ref
4309 * 1 for xattr if selinux is on
4311 err = btrfs_reserve_metadata_space(root, 5);
4315 trans = btrfs_start_transaction(root, 1);
4318 btrfs_set_trans_block_group(trans, dir);
4320 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4326 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4328 dentry->d_parent->d_inode->i_ino, objectid,
4329 BTRFS_I(dir)->block_group, mode, &index);
4330 err = PTR_ERR(inode);
4334 err = btrfs_init_inode_security(trans, inode, dir);
4340 btrfs_set_trans_block_group(trans, inode);
4341 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4345 inode->i_op = &btrfs_special_inode_operations;
4346 init_special_inode(inode, inode->i_mode, rdev);
4347 btrfs_update_inode(trans, root, inode);
4349 btrfs_update_inode_block_group(trans, inode);
4350 btrfs_update_inode_block_group(trans, dir);
4352 nr = trans->blocks_used;
4353 btrfs_end_transaction_throttle(trans, root);
4355 btrfs_unreserve_metadata_space(root, 5);
4357 inode_dec_link_count(inode);
4360 btrfs_btree_balance_dirty(root, nr);
4364 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4365 int mode, struct nameidata *nd)
4367 struct btrfs_trans_handle *trans;
4368 struct btrfs_root *root = BTRFS_I(dir)->root;
4369 struct inode *inode = NULL;
4372 unsigned long nr = 0;
4377 * 2 for inode item and ref
4379 * 1 for xattr if selinux is on
4381 err = btrfs_reserve_metadata_space(root, 5);
4385 trans = btrfs_start_transaction(root, 1);
4388 btrfs_set_trans_block_group(trans, dir);
4390 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4396 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4398 dentry->d_parent->d_inode->i_ino,
4399 objectid, BTRFS_I(dir)->block_group, mode,
4401 err = PTR_ERR(inode);
4405 err = btrfs_init_inode_security(trans, inode, dir);
4411 btrfs_set_trans_block_group(trans, inode);
4412 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4416 inode->i_mapping->a_ops = &btrfs_aops;
4417 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4418 inode->i_fop = &btrfs_file_operations;
4419 inode->i_op = &btrfs_file_inode_operations;
4420 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4422 btrfs_update_inode_block_group(trans, inode);
4423 btrfs_update_inode_block_group(trans, dir);
4425 nr = trans->blocks_used;
4426 btrfs_end_transaction_throttle(trans, root);
4428 btrfs_unreserve_metadata_space(root, 5);
4430 inode_dec_link_count(inode);
4433 btrfs_btree_balance_dirty(root, nr);
4437 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4438 struct dentry *dentry)
4440 struct btrfs_trans_handle *trans;
4441 struct btrfs_root *root = BTRFS_I(dir)->root;
4442 struct inode *inode = old_dentry->d_inode;
4444 unsigned long nr = 0;
4448 if (inode->i_nlink == 0)
4451 /* do not allow sys_link's with other subvols of the same device */
4452 if (root->objectid != BTRFS_I(inode)->root->objectid)
4456 * 1 item for inode ref
4457 * 2 items for dir items
4459 err = btrfs_reserve_metadata_space(root, 3);
4463 btrfs_inc_nlink(inode);
4465 err = btrfs_set_inode_index(dir, &index);
4469 trans = btrfs_start_transaction(root, 1);
4471 btrfs_set_trans_block_group(trans, dir);
4472 atomic_inc(&inode->i_count);
4474 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4479 btrfs_update_inode_block_group(trans, dir);
4480 err = btrfs_update_inode(trans, root, inode);
4482 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4485 nr = trans->blocks_used;
4486 btrfs_end_transaction_throttle(trans, root);
4488 btrfs_unreserve_metadata_space(root, 3);
4490 inode_dec_link_count(inode);
4493 btrfs_btree_balance_dirty(root, nr);
4497 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4499 struct inode *inode = NULL;
4500 struct btrfs_trans_handle *trans;
4501 struct btrfs_root *root = BTRFS_I(dir)->root;
4503 int drop_on_err = 0;
4506 unsigned long nr = 1;
4509 * 2 items for inode and ref
4510 * 2 items for dir items
4511 * 1 for xattr if selinux is on
4513 err = btrfs_reserve_metadata_space(root, 5);
4517 trans = btrfs_start_transaction(root, 1);
4522 btrfs_set_trans_block_group(trans, dir);
4524 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4530 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4532 dentry->d_parent->d_inode->i_ino, objectid,
4533 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4535 if (IS_ERR(inode)) {
4536 err = PTR_ERR(inode);
4542 err = btrfs_init_inode_security(trans, inode, dir);
4546 inode->i_op = &btrfs_dir_inode_operations;
4547 inode->i_fop = &btrfs_dir_file_operations;
4548 btrfs_set_trans_block_group(trans, inode);
4550 btrfs_i_size_write(inode, 0);
4551 err = btrfs_update_inode(trans, root, inode);
4555 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4556 inode, dentry->d_name.name,
4557 dentry->d_name.len, 0, index);
4561 d_instantiate(dentry, inode);
4563 btrfs_update_inode_block_group(trans, inode);
4564 btrfs_update_inode_block_group(trans, dir);
4567 nr = trans->blocks_used;
4568 btrfs_end_transaction_throttle(trans, root);
4571 btrfs_unreserve_metadata_space(root, 5);
4574 btrfs_btree_balance_dirty(root, nr);
4578 /* helper for btfs_get_extent. Given an existing extent in the tree,
4579 * and an extent that you want to insert, deal with overlap and insert
4580 * the new extent into the tree.
4582 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4583 struct extent_map *existing,
4584 struct extent_map *em,
4585 u64 map_start, u64 map_len)
4589 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4590 start_diff = map_start - em->start;
4591 em->start = map_start;
4593 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4594 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4595 em->block_start += start_diff;
4596 em->block_len -= start_diff;
4598 return add_extent_mapping(em_tree, em);
4601 static noinline int uncompress_inline(struct btrfs_path *path,
4602 struct inode *inode, struct page *page,
4603 size_t pg_offset, u64 extent_offset,
4604 struct btrfs_file_extent_item *item)
4607 struct extent_buffer *leaf = path->nodes[0];
4610 unsigned long inline_size;
4613 WARN_ON(pg_offset != 0);
4614 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4615 inline_size = btrfs_file_extent_inline_item_len(leaf,
4616 btrfs_item_nr(leaf, path->slots[0]));
4617 tmp = kmalloc(inline_size, GFP_NOFS);
4618 ptr = btrfs_file_extent_inline_start(item);
4620 read_extent_buffer(leaf, tmp, ptr, inline_size);
4622 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4623 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4624 inline_size, max_size);
4626 char *kaddr = kmap_atomic(page, KM_USER0);
4627 unsigned long copy_size = min_t(u64,
4628 PAGE_CACHE_SIZE - pg_offset,
4629 max_size - extent_offset);
4630 memset(kaddr + pg_offset, 0, copy_size);
4631 kunmap_atomic(kaddr, KM_USER0);
4638 * a bit scary, this does extent mapping from logical file offset to the disk.
4639 * the ugly parts come from merging extents from the disk with the in-ram
4640 * representation. This gets more complex because of the data=ordered code,
4641 * where the in-ram extents might be locked pending data=ordered completion.
4643 * This also copies inline extents directly into the page.
4646 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4647 size_t pg_offset, u64 start, u64 len,
4653 u64 extent_start = 0;
4655 u64 objectid = inode->i_ino;
4657 struct btrfs_path *path = NULL;
4658 struct btrfs_root *root = BTRFS_I(inode)->root;
4659 struct btrfs_file_extent_item *item;
4660 struct extent_buffer *leaf;
4661 struct btrfs_key found_key;
4662 struct extent_map *em = NULL;
4663 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4664 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4665 struct btrfs_trans_handle *trans = NULL;
4669 read_lock(&em_tree->lock);
4670 em = lookup_extent_mapping(em_tree, start, len);
4672 em->bdev = root->fs_info->fs_devices->latest_bdev;
4673 read_unlock(&em_tree->lock);
4676 if (em->start > start || em->start + em->len <= start)
4677 free_extent_map(em);
4678 else if (em->block_start == EXTENT_MAP_INLINE && page)
4679 free_extent_map(em);
4683 em = alloc_extent_map(GFP_NOFS);
4688 em->bdev = root->fs_info->fs_devices->latest_bdev;
4689 em->start = EXTENT_MAP_HOLE;
4690 em->orig_start = EXTENT_MAP_HOLE;
4692 em->block_len = (u64)-1;
4695 path = btrfs_alloc_path();
4699 ret = btrfs_lookup_file_extent(trans, root, path,
4700 objectid, start, trans != NULL);
4707 if (path->slots[0] == 0)
4712 leaf = path->nodes[0];
4713 item = btrfs_item_ptr(leaf, path->slots[0],
4714 struct btrfs_file_extent_item);
4715 /* are we inside the extent that was found? */
4716 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4717 found_type = btrfs_key_type(&found_key);
4718 if (found_key.objectid != objectid ||
4719 found_type != BTRFS_EXTENT_DATA_KEY) {
4723 found_type = btrfs_file_extent_type(leaf, item);
4724 extent_start = found_key.offset;
4725 compressed = btrfs_file_extent_compression(leaf, item);
4726 if (found_type == BTRFS_FILE_EXTENT_REG ||
4727 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4728 extent_end = extent_start +
4729 btrfs_file_extent_num_bytes(leaf, item);
4730 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4732 size = btrfs_file_extent_inline_len(leaf, item);
4733 extent_end = (extent_start + size + root->sectorsize - 1) &
4734 ~((u64)root->sectorsize - 1);
4737 if (start >= extent_end) {
4739 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4740 ret = btrfs_next_leaf(root, path);
4747 leaf = path->nodes[0];
4749 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4750 if (found_key.objectid != objectid ||
4751 found_key.type != BTRFS_EXTENT_DATA_KEY)
4753 if (start + len <= found_key.offset)
4756 em->len = found_key.offset - start;
4760 if (found_type == BTRFS_FILE_EXTENT_REG ||
4761 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4762 em->start = extent_start;
4763 em->len = extent_end - extent_start;
4764 em->orig_start = extent_start -
4765 btrfs_file_extent_offset(leaf, item);
4766 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4768 em->block_start = EXTENT_MAP_HOLE;
4772 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4773 em->block_start = bytenr;
4774 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4777 bytenr += btrfs_file_extent_offset(leaf, item);
4778 em->block_start = bytenr;
4779 em->block_len = em->len;
4780 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4781 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4784 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4788 size_t extent_offset;
4791 em->block_start = EXTENT_MAP_INLINE;
4792 if (!page || create) {
4793 em->start = extent_start;
4794 em->len = extent_end - extent_start;
4798 size = btrfs_file_extent_inline_len(leaf, item);
4799 extent_offset = page_offset(page) + pg_offset - extent_start;
4800 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4801 size - extent_offset);
4802 em->start = extent_start + extent_offset;
4803 em->len = (copy_size + root->sectorsize - 1) &
4804 ~((u64)root->sectorsize - 1);
4805 em->orig_start = EXTENT_MAP_INLINE;
4807 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4808 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4809 if (create == 0 && !PageUptodate(page)) {
4810 if (btrfs_file_extent_compression(leaf, item) ==
4811 BTRFS_COMPRESS_ZLIB) {
4812 ret = uncompress_inline(path, inode, page,
4814 extent_offset, item);
4818 read_extent_buffer(leaf, map + pg_offset, ptr,
4820 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4821 memset(map + pg_offset + copy_size, 0,
4822 PAGE_CACHE_SIZE - pg_offset -
4827 flush_dcache_page(page);
4828 } else if (create && PageUptodate(page)) {
4831 free_extent_map(em);
4833 btrfs_release_path(root, path);
4834 trans = btrfs_join_transaction(root, 1);
4838 write_extent_buffer(leaf, map + pg_offset, ptr,
4841 btrfs_mark_buffer_dirty(leaf);
4843 set_extent_uptodate(io_tree, em->start,
4844 extent_map_end(em) - 1, GFP_NOFS);
4847 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4854 em->block_start = EXTENT_MAP_HOLE;
4855 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4857 btrfs_release_path(root, path);
4858 if (em->start > start || extent_map_end(em) <= start) {
4859 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4860 "[%llu %llu]\n", (unsigned long long)em->start,
4861 (unsigned long long)em->len,
4862 (unsigned long long)start,
4863 (unsigned long long)len);
4869 write_lock(&em_tree->lock);
4870 ret = add_extent_mapping(em_tree, em);
4871 /* it is possible that someone inserted the extent into the tree
4872 * while we had the lock dropped. It is also possible that
4873 * an overlapping map exists in the tree
4875 if (ret == -EEXIST) {
4876 struct extent_map *existing;
4880 existing = lookup_extent_mapping(em_tree, start, len);
4881 if (existing && (existing->start > start ||
4882 existing->start + existing->len <= start)) {
4883 free_extent_map(existing);
4887 existing = lookup_extent_mapping(em_tree, em->start,
4890 err = merge_extent_mapping(em_tree, existing,
4893 free_extent_map(existing);
4895 free_extent_map(em);
4900 free_extent_map(em);
4904 free_extent_map(em);
4909 write_unlock(&em_tree->lock);
4912 btrfs_free_path(path);
4914 ret = btrfs_end_transaction(trans, root);
4919 free_extent_map(em);
4920 return ERR_PTR(err);
4925 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4926 const struct iovec *iov, loff_t offset,
4927 unsigned long nr_segs)
4932 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4933 __u64 start, __u64 len)
4935 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4938 int btrfs_readpage(struct file *file, struct page *page)
4940 struct extent_io_tree *tree;
4941 tree = &BTRFS_I(page->mapping->host)->io_tree;
4942 return extent_read_full_page(tree, page, btrfs_get_extent);
4945 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4947 struct extent_io_tree *tree;
4950 if (current->flags & PF_MEMALLOC) {
4951 redirty_page_for_writepage(wbc, page);
4955 tree = &BTRFS_I(page->mapping->host)->io_tree;
4956 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4959 int btrfs_writepages(struct address_space *mapping,
4960 struct writeback_control *wbc)
4962 struct extent_io_tree *tree;
4964 tree = &BTRFS_I(mapping->host)->io_tree;
4965 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4969 btrfs_readpages(struct file *file, struct address_space *mapping,
4970 struct list_head *pages, unsigned nr_pages)
4972 struct extent_io_tree *tree;
4973 tree = &BTRFS_I(mapping->host)->io_tree;
4974 return extent_readpages(tree, mapping, pages, nr_pages,
4977 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4979 struct extent_io_tree *tree;
4980 struct extent_map_tree *map;
4983 tree = &BTRFS_I(page->mapping->host)->io_tree;
4984 map = &BTRFS_I(page->mapping->host)->extent_tree;
4985 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
4987 ClearPagePrivate(page);
4988 set_page_private(page, 0);
4989 page_cache_release(page);
4994 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4996 if (PageWriteback(page) || PageDirty(page))
4998 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5001 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5003 struct extent_io_tree *tree;
5004 struct btrfs_ordered_extent *ordered;
5005 struct extent_state *cached_state = NULL;
5006 u64 page_start = page_offset(page);
5007 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5011 * we have the page locked, so new writeback can't start,
5012 * and the dirty bit won't be cleared while we are here.
5014 * Wait for IO on this page so that we can safely clear
5015 * the PagePrivate2 bit and do ordered accounting
5017 wait_on_page_writeback(page);
5019 tree = &BTRFS_I(page->mapping->host)->io_tree;
5021 btrfs_releasepage(page, GFP_NOFS);
5024 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5026 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5030 * IO on this page will never be started, so we need
5031 * to account for any ordered extents now
5033 clear_extent_bit(tree, page_start, page_end,
5034 EXTENT_DIRTY | EXTENT_DELALLOC |
5035 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5036 &cached_state, GFP_NOFS);
5038 * whoever cleared the private bit is responsible
5039 * for the finish_ordered_io
5041 if (TestClearPagePrivate2(page)) {
5042 btrfs_finish_ordered_io(page->mapping->host,
5043 page_start, page_end);
5045 btrfs_put_ordered_extent(ordered);
5046 cached_state = NULL;
5047 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5050 clear_extent_bit(tree, page_start, page_end,
5051 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5052 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5053 __btrfs_releasepage(page, GFP_NOFS);
5055 ClearPageChecked(page);
5056 if (PagePrivate(page)) {
5057 ClearPagePrivate(page);
5058 set_page_private(page, 0);
5059 page_cache_release(page);
5064 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5065 * called from a page fault handler when a page is first dirtied. Hence we must
5066 * be careful to check for EOF conditions here. We set the page up correctly
5067 * for a written page which means we get ENOSPC checking when writing into
5068 * holes and correct delalloc and unwritten extent mapping on filesystems that
5069 * support these features.
5071 * We are not allowed to take the i_mutex here so we have to play games to
5072 * protect against truncate races as the page could now be beyond EOF. Because
5073 * vmtruncate() writes the inode size before removing pages, once we have the
5074 * page lock we can determine safely if the page is beyond EOF. If it is not
5075 * beyond EOF, then the page is guaranteed safe against truncation until we
5078 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5080 struct page *page = vmf->page;
5081 struct inode *inode = fdentry(vma->vm_file)->d_inode;
5082 struct btrfs_root *root = BTRFS_I(inode)->root;
5083 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5084 struct btrfs_ordered_extent *ordered;
5085 struct extent_state *cached_state = NULL;
5087 unsigned long zero_start;
5093 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5097 else /* -ENOSPC, -EIO, etc */
5098 ret = VM_FAULT_SIGBUS;
5102 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5104 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5105 ret = VM_FAULT_SIGBUS;
5109 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5112 size = i_size_read(inode);
5113 page_start = page_offset(page);
5114 page_end = page_start + PAGE_CACHE_SIZE - 1;
5116 if ((page->mapping != inode->i_mapping) ||
5117 (page_start >= size)) {
5118 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5119 /* page got truncated out from underneath us */
5122 wait_on_page_writeback(page);
5124 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
5126 set_page_extent_mapped(page);
5129 * we can't set the delalloc bits if there are pending ordered
5130 * extents. Drop our locks and wait for them to finish
5132 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5134 unlock_extent_cached(io_tree, page_start, page_end,
5135 &cached_state, GFP_NOFS);
5137 btrfs_start_ordered_extent(inode, ordered, 1);
5138 btrfs_put_ordered_extent(ordered);
5143 * XXX - page_mkwrite gets called every time the page is dirtied, even
5144 * if it was already dirty, so for space accounting reasons we need to
5145 * clear any delalloc bits for the range we are fixing to save. There
5146 * is probably a better way to do this, but for now keep consistent with
5147 * prepare_pages in the normal write path.
5149 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5150 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5151 0, 0, &cached_state, GFP_NOFS);
5153 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
5156 unlock_extent_cached(io_tree, page_start, page_end,
5157 &cached_state, GFP_NOFS);
5158 ret = VM_FAULT_SIGBUS;
5159 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5164 /* page is wholly or partially inside EOF */
5165 if (page_start + PAGE_CACHE_SIZE > size)
5166 zero_start = size & ~PAGE_CACHE_MASK;
5168 zero_start = PAGE_CACHE_SIZE;
5170 if (zero_start != PAGE_CACHE_SIZE) {
5172 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5173 flush_dcache_page(page);
5176 ClearPageChecked(page);
5177 set_page_dirty(page);
5178 SetPageUptodate(page);
5180 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5181 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5183 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
5186 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5188 return VM_FAULT_LOCKED;
5194 static void btrfs_truncate(struct inode *inode)
5196 struct btrfs_root *root = BTRFS_I(inode)->root;
5198 struct btrfs_trans_handle *trans;
5200 u64 mask = root->sectorsize - 1;
5202 if (!S_ISREG(inode->i_mode)) {
5207 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5211 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5212 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5214 trans = btrfs_start_transaction(root, 1);
5215 btrfs_set_trans_block_group(trans, inode);
5218 * setattr is responsible for setting the ordered_data_close flag,
5219 * but that is only tested during the last file release. That
5220 * could happen well after the next commit, leaving a great big
5221 * window where new writes may get lost if someone chooses to write
5222 * to this file after truncating to zero
5224 * The inode doesn't have any dirty data here, and so if we commit
5225 * this is a noop. If someone immediately starts writing to the inode
5226 * it is very likely we'll catch some of their writes in this
5227 * transaction, and the commit will find this file on the ordered
5228 * data list with good things to send down.
5230 * This is a best effort solution, there is still a window where
5231 * using truncate to replace the contents of the file will
5232 * end up with a zero length file after a crash.
5234 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5235 btrfs_add_ordered_operation(trans, root, inode);
5238 ret = btrfs_truncate_inode_items(trans, root, inode,
5240 BTRFS_EXTENT_DATA_KEY);
5244 ret = btrfs_update_inode(trans, root, inode);
5247 nr = trans->blocks_used;
5248 btrfs_end_transaction(trans, root);
5249 btrfs_btree_balance_dirty(root, nr);
5251 trans = btrfs_start_transaction(root, 1);
5252 btrfs_set_trans_block_group(trans, inode);
5255 if (ret == 0 && inode->i_nlink > 0) {
5256 ret = btrfs_orphan_del(trans, inode);
5260 ret = btrfs_update_inode(trans, root, inode);
5263 nr = trans->blocks_used;
5264 ret = btrfs_end_transaction_throttle(trans, root);
5266 btrfs_btree_balance_dirty(root, nr);
5270 * create a new subvolume directory/inode (helper for the ioctl).
5272 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5273 struct btrfs_root *new_root,
5274 u64 new_dirid, u64 alloc_hint)
5276 struct inode *inode;
5280 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5281 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5283 return PTR_ERR(inode);
5284 inode->i_op = &btrfs_dir_inode_operations;
5285 inode->i_fop = &btrfs_dir_file_operations;
5288 btrfs_i_size_write(inode, 0);
5290 err = btrfs_update_inode(trans, new_root, inode);
5297 /* helper function for file defrag and space balancing. This
5298 * forces readahead on a given range of bytes in an inode
5300 unsigned long btrfs_force_ra(struct address_space *mapping,
5301 struct file_ra_state *ra, struct file *file,
5302 pgoff_t offset, pgoff_t last_index)
5304 pgoff_t req_size = last_index - offset + 1;
5306 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5307 return offset + req_size;
5310 struct inode *btrfs_alloc_inode(struct super_block *sb)
5312 struct btrfs_inode *ei;
5314 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5318 ei->last_sub_trans = 0;
5319 ei->logged_trans = 0;
5320 ei->outstanding_extents = 0;
5321 ei->reserved_extents = 0;
5323 spin_lock_init(&ei->accounting_lock);
5324 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5325 INIT_LIST_HEAD(&ei->i_orphan);
5326 INIT_LIST_HEAD(&ei->ordered_operations);
5327 return &ei->vfs_inode;
5330 void btrfs_destroy_inode(struct inode *inode)
5332 struct btrfs_ordered_extent *ordered;
5333 struct btrfs_root *root = BTRFS_I(inode)->root;
5335 WARN_ON(!list_empty(&inode->i_dentry));
5336 WARN_ON(inode->i_data.nrpages);
5339 * This can happen where we create an inode, but somebody else also
5340 * created the same inode and we need to destroy the one we already
5347 * Make sure we're properly removed from the ordered operation
5351 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5352 spin_lock(&root->fs_info->ordered_extent_lock);
5353 list_del_init(&BTRFS_I(inode)->ordered_operations);
5354 spin_unlock(&root->fs_info->ordered_extent_lock);
5357 spin_lock(&root->list_lock);
5358 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5359 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5361 list_del_init(&BTRFS_I(inode)->i_orphan);
5363 spin_unlock(&root->list_lock);
5366 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5370 printk(KERN_ERR "btrfs found ordered "
5371 "extent %llu %llu on inode cleanup\n",
5372 (unsigned long long)ordered->file_offset,
5373 (unsigned long long)ordered->len);
5374 btrfs_remove_ordered_extent(inode, ordered);
5375 btrfs_put_ordered_extent(ordered);
5376 btrfs_put_ordered_extent(ordered);
5379 inode_tree_del(inode);
5380 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5382 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5385 void btrfs_drop_inode(struct inode *inode)
5387 struct btrfs_root *root = BTRFS_I(inode)->root;
5389 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5390 generic_delete_inode(inode);
5392 generic_drop_inode(inode);
5395 static void init_once(void *foo)
5397 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5399 inode_init_once(&ei->vfs_inode);
5402 void btrfs_destroy_cachep(void)
5404 if (btrfs_inode_cachep)
5405 kmem_cache_destroy(btrfs_inode_cachep);
5406 if (btrfs_trans_handle_cachep)
5407 kmem_cache_destroy(btrfs_trans_handle_cachep);
5408 if (btrfs_transaction_cachep)
5409 kmem_cache_destroy(btrfs_transaction_cachep);
5410 if (btrfs_path_cachep)
5411 kmem_cache_destroy(btrfs_path_cachep);
5414 int btrfs_init_cachep(void)
5416 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5417 sizeof(struct btrfs_inode), 0,
5418 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5419 if (!btrfs_inode_cachep)
5422 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5423 sizeof(struct btrfs_trans_handle), 0,
5424 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5425 if (!btrfs_trans_handle_cachep)
5428 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5429 sizeof(struct btrfs_transaction), 0,
5430 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5431 if (!btrfs_transaction_cachep)
5434 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5435 sizeof(struct btrfs_path), 0,
5436 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5437 if (!btrfs_path_cachep)
5442 btrfs_destroy_cachep();
5446 static int btrfs_getattr(struct vfsmount *mnt,
5447 struct dentry *dentry, struct kstat *stat)
5449 struct inode *inode = dentry->d_inode;
5450 generic_fillattr(inode, stat);
5451 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5452 stat->blksize = PAGE_CACHE_SIZE;
5453 stat->blocks = (inode_get_bytes(inode) +
5454 BTRFS_I(inode)->delalloc_bytes) >> 9;
5458 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5459 struct inode *new_dir, struct dentry *new_dentry)
5461 struct btrfs_trans_handle *trans;
5462 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5463 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5464 struct inode *new_inode = new_dentry->d_inode;
5465 struct inode *old_inode = old_dentry->d_inode;
5466 struct timespec ctime = CURRENT_TIME;
5471 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5474 /* we only allow rename subvolume link between subvolumes */
5475 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5478 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5479 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5482 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5483 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5487 * We want to reserve the absolute worst case amount of items. So if
5488 * both inodes are subvols and we need to unlink them then that would
5489 * require 4 item modifications, but if they are both normal inodes it
5490 * would require 5 item modifications, so we'll assume their normal
5491 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5492 * should cover the worst case number of items we'll modify.
5494 ret = btrfs_reserve_metadata_space(root, 11);
5499 * we're using rename to replace one file with another.
5500 * and the replacement file is large. Start IO on it now so
5501 * we don't add too much work to the end of the transaction
5503 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5504 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5505 filemap_flush(old_inode->i_mapping);
5507 /* close the racy window with snapshot create/destroy ioctl */
5508 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5509 down_read(&root->fs_info->subvol_sem);
5511 trans = btrfs_start_transaction(root, 1);
5512 btrfs_set_trans_block_group(trans, new_dir);
5515 btrfs_record_root_in_trans(trans, dest);
5517 ret = btrfs_set_inode_index(new_dir, &index);
5521 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5522 /* force full log commit if subvolume involved. */
5523 root->fs_info->last_trans_log_full_commit = trans->transid;
5525 ret = btrfs_insert_inode_ref(trans, dest,
5526 new_dentry->d_name.name,
5527 new_dentry->d_name.len,
5529 new_dir->i_ino, index);
5533 * this is an ugly little race, but the rename is required
5534 * to make sure that if we crash, the inode is either at the
5535 * old name or the new one. pinning the log transaction lets
5536 * us make sure we don't allow a log commit to come in after
5537 * we unlink the name but before we add the new name back in.
5539 btrfs_pin_log_trans(root);
5542 * make sure the inode gets flushed if it is replacing
5545 if (new_inode && new_inode->i_size &&
5546 old_inode && S_ISREG(old_inode->i_mode)) {
5547 btrfs_add_ordered_operation(trans, root, old_inode);
5550 old_dir->i_ctime = old_dir->i_mtime = ctime;
5551 new_dir->i_ctime = new_dir->i_mtime = ctime;
5552 old_inode->i_ctime = ctime;
5554 if (old_dentry->d_parent != new_dentry->d_parent)
5555 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5557 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5558 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5559 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5560 old_dentry->d_name.name,
5561 old_dentry->d_name.len);
5563 btrfs_inc_nlink(old_dentry->d_inode);
5564 ret = btrfs_unlink_inode(trans, root, old_dir,
5565 old_dentry->d_inode,
5566 old_dentry->d_name.name,
5567 old_dentry->d_name.len);
5572 new_inode->i_ctime = CURRENT_TIME;
5573 if (unlikely(new_inode->i_ino ==
5574 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5575 root_objectid = BTRFS_I(new_inode)->location.objectid;
5576 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5578 new_dentry->d_name.name,
5579 new_dentry->d_name.len);
5580 BUG_ON(new_inode->i_nlink == 0);
5582 ret = btrfs_unlink_inode(trans, dest, new_dir,
5583 new_dentry->d_inode,
5584 new_dentry->d_name.name,
5585 new_dentry->d_name.len);
5588 if (new_inode->i_nlink == 0) {
5589 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5594 ret = btrfs_add_link(trans, new_dir, old_inode,
5595 new_dentry->d_name.name,
5596 new_dentry->d_name.len, 0, index);
5599 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5600 btrfs_log_new_name(trans, old_inode, old_dir,
5601 new_dentry->d_parent);
5602 btrfs_end_log_trans(root);
5605 btrfs_end_transaction_throttle(trans, root);
5607 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5608 up_read(&root->fs_info->subvol_sem);
5610 btrfs_unreserve_metadata_space(root, 11);
5615 * some fairly slow code that needs optimization. This walks the list
5616 * of all the inodes with pending delalloc and forces them to disk.
5618 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5620 struct list_head *head = &root->fs_info->delalloc_inodes;
5621 struct btrfs_inode *binode;
5622 struct inode *inode;
5624 if (root->fs_info->sb->s_flags & MS_RDONLY)
5627 spin_lock(&root->fs_info->delalloc_lock);
5628 while (!list_empty(head)) {
5629 binode = list_entry(head->next, struct btrfs_inode,
5631 inode = igrab(&binode->vfs_inode);
5633 list_del_init(&binode->delalloc_inodes);
5634 spin_unlock(&root->fs_info->delalloc_lock);
5636 filemap_flush(inode->i_mapping);
5638 btrfs_add_delayed_iput(inode);
5643 spin_lock(&root->fs_info->delalloc_lock);
5645 spin_unlock(&root->fs_info->delalloc_lock);
5647 /* the filemap_flush will queue IO into the worker threads, but
5648 * we have to make sure the IO is actually started and that
5649 * ordered extents get created before we return
5651 atomic_inc(&root->fs_info->async_submit_draining);
5652 while (atomic_read(&root->fs_info->nr_async_submits) ||
5653 atomic_read(&root->fs_info->async_delalloc_pages)) {
5654 wait_event(root->fs_info->async_submit_wait,
5655 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5656 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5658 atomic_dec(&root->fs_info->async_submit_draining);
5662 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5663 const char *symname)
5665 struct btrfs_trans_handle *trans;
5666 struct btrfs_root *root = BTRFS_I(dir)->root;
5667 struct btrfs_path *path;
5668 struct btrfs_key key;
5669 struct inode *inode = NULL;
5677 struct btrfs_file_extent_item *ei;
5678 struct extent_buffer *leaf;
5679 unsigned long nr = 0;
5681 name_len = strlen(symname) + 1;
5682 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5683 return -ENAMETOOLONG;
5686 * 2 items for inode item and ref
5687 * 2 items for dir items
5688 * 1 item for xattr if selinux is on
5690 err = btrfs_reserve_metadata_space(root, 5);
5694 trans = btrfs_start_transaction(root, 1);
5697 btrfs_set_trans_block_group(trans, dir);
5699 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5705 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5707 dentry->d_parent->d_inode->i_ino, objectid,
5708 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5710 err = PTR_ERR(inode);
5714 err = btrfs_init_inode_security(trans, inode, dir);
5720 btrfs_set_trans_block_group(trans, inode);
5721 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5725 inode->i_mapping->a_ops = &btrfs_aops;
5726 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5727 inode->i_fop = &btrfs_file_operations;
5728 inode->i_op = &btrfs_file_inode_operations;
5729 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5731 btrfs_update_inode_block_group(trans, inode);
5732 btrfs_update_inode_block_group(trans, dir);
5736 path = btrfs_alloc_path();
5738 key.objectid = inode->i_ino;
5740 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5741 datasize = btrfs_file_extent_calc_inline_size(name_len);
5742 err = btrfs_insert_empty_item(trans, root, path, &key,
5748 leaf = path->nodes[0];
5749 ei = btrfs_item_ptr(leaf, path->slots[0],
5750 struct btrfs_file_extent_item);
5751 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5752 btrfs_set_file_extent_type(leaf, ei,
5753 BTRFS_FILE_EXTENT_INLINE);
5754 btrfs_set_file_extent_encryption(leaf, ei, 0);
5755 btrfs_set_file_extent_compression(leaf, ei, 0);
5756 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5757 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5759 ptr = btrfs_file_extent_inline_start(ei);
5760 write_extent_buffer(leaf, symname, ptr, name_len);
5761 btrfs_mark_buffer_dirty(leaf);
5762 btrfs_free_path(path);
5764 inode->i_op = &btrfs_symlink_inode_operations;
5765 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5766 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5767 inode_set_bytes(inode, name_len);
5768 btrfs_i_size_write(inode, name_len - 1);
5769 err = btrfs_update_inode(trans, root, inode);
5774 nr = trans->blocks_used;
5775 btrfs_end_transaction_throttle(trans, root);
5777 btrfs_unreserve_metadata_space(root, 5);
5779 inode_dec_link_count(inode);
5782 btrfs_btree_balance_dirty(root, nr);
5786 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
5787 u64 alloc_hint, int mode, loff_t actual_len)
5789 struct btrfs_trans_handle *trans;
5790 struct btrfs_root *root = BTRFS_I(inode)->root;
5791 struct btrfs_key ins;
5793 u64 cur_offset = start;
5794 u64 num_bytes = end - start;
5798 while (num_bytes > 0) {
5799 alloc_size = min(num_bytes, root->fs_info->max_extent);
5801 trans = btrfs_start_transaction(root, 1);
5803 ret = btrfs_reserve_extent(trans, root, alloc_size,
5804 root->sectorsize, 0, alloc_hint,
5811 ret = btrfs_reserve_metadata_space(root, 3);
5813 btrfs_free_reserved_extent(root, ins.objectid,
5818 ret = insert_reserved_file_extent(trans, inode,
5819 cur_offset, ins.objectid,
5820 ins.offset, ins.offset,
5821 ins.offset, 0, 0, 0,
5822 BTRFS_FILE_EXTENT_PREALLOC);
5824 btrfs_drop_extent_cache(inode, cur_offset,
5825 cur_offset + ins.offset -1, 0);
5827 num_bytes -= ins.offset;
5828 cur_offset += ins.offset;
5829 alloc_hint = ins.objectid + ins.offset;
5831 inode->i_ctime = CURRENT_TIME;
5832 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5833 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5834 (actual_len > inode->i_size) &&
5835 (cur_offset > inode->i_size)) {
5837 if (cur_offset > actual_len)
5838 i_size = actual_len;
5840 i_size = cur_offset;
5841 i_size_write(inode, i_size);
5842 btrfs_ordered_update_i_size(inode, i_size, NULL);
5845 ret = btrfs_update_inode(trans, root, inode);
5848 btrfs_end_transaction(trans, root);
5849 btrfs_unreserve_metadata_space(root, 3);
5854 btrfs_end_transaction(trans, root);
5859 static long btrfs_fallocate(struct inode *inode, int mode,
5860 loff_t offset, loff_t len)
5862 struct extent_state *cached_state = NULL;
5869 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5870 struct extent_map *em;
5873 alloc_start = offset & ~mask;
5874 alloc_end = (offset + len + mask) & ~mask;
5877 * wait for ordered IO before we have any locks. We'll loop again
5878 * below with the locks held.
5880 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5882 mutex_lock(&inode->i_mutex);
5883 if (alloc_start > inode->i_size) {
5884 ret = btrfs_cont_expand(inode, alloc_start);
5889 ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode,
5890 alloc_end - alloc_start);
5894 locked_end = alloc_end - 1;
5896 struct btrfs_ordered_extent *ordered;
5898 /* the extent lock is ordered inside the running
5901 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
5902 locked_end, 0, &cached_state, GFP_NOFS);
5903 ordered = btrfs_lookup_first_ordered_extent(inode,
5906 ordered->file_offset + ordered->len > alloc_start &&
5907 ordered->file_offset < alloc_end) {
5908 btrfs_put_ordered_extent(ordered);
5909 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
5910 alloc_start, locked_end,
5911 &cached_state, GFP_NOFS);
5913 * we can't wait on the range with the transaction
5914 * running or with the extent lock held
5916 btrfs_wait_ordered_range(inode, alloc_start,
5917 alloc_end - alloc_start);
5920 btrfs_put_ordered_extent(ordered);
5925 cur_offset = alloc_start;
5927 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5928 alloc_end - cur_offset, 0);
5929 BUG_ON(IS_ERR(em) || !em);
5930 last_byte = min(extent_map_end(em), alloc_end);
5931 last_byte = (last_byte + mask) & ~mask;
5932 if (em->block_start == EXTENT_MAP_HOLE ||
5933 (cur_offset >= inode->i_size &&
5934 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5935 ret = prealloc_file_range(inode,
5936 cur_offset, last_byte,
5937 alloc_hint, mode, offset+len);
5939 free_extent_map(em);
5943 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5944 alloc_hint = em->block_start;
5945 free_extent_map(em);
5947 cur_offset = last_byte;
5948 if (cur_offset >= alloc_end) {
5953 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5954 &cached_state, GFP_NOFS);
5956 btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode,
5957 alloc_end - alloc_start);
5959 mutex_unlock(&inode->i_mutex);
5963 static int btrfs_set_page_dirty(struct page *page)
5965 return __set_page_dirty_nobuffers(page);
5968 static int btrfs_permission(struct inode *inode, int mask)
5970 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5972 return generic_permission(inode, mask, btrfs_check_acl);
5975 static const struct inode_operations btrfs_dir_inode_operations = {
5976 .getattr = btrfs_getattr,
5977 .lookup = btrfs_lookup,
5978 .create = btrfs_create,
5979 .unlink = btrfs_unlink,
5981 .mkdir = btrfs_mkdir,
5982 .rmdir = btrfs_rmdir,
5983 .rename = btrfs_rename,
5984 .symlink = btrfs_symlink,
5985 .setattr = btrfs_setattr,
5986 .mknod = btrfs_mknod,
5987 .setxattr = btrfs_setxattr,
5988 .getxattr = btrfs_getxattr,
5989 .listxattr = btrfs_listxattr,
5990 .removexattr = btrfs_removexattr,
5991 .permission = btrfs_permission,
5993 static const struct inode_operations btrfs_dir_ro_inode_operations = {
5994 .lookup = btrfs_lookup,
5995 .permission = btrfs_permission,
5998 static const struct file_operations btrfs_dir_file_operations = {
5999 .llseek = generic_file_llseek,
6000 .read = generic_read_dir,
6001 .readdir = btrfs_real_readdir,
6002 .unlocked_ioctl = btrfs_ioctl,
6003 #ifdef CONFIG_COMPAT
6004 .compat_ioctl = btrfs_ioctl,
6006 .release = btrfs_release_file,
6007 .fsync = btrfs_sync_file,
6010 static struct extent_io_ops btrfs_extent_io_ops = {
6011 .fill_delalloc = run_delalloc_range,
6012 .submit_bio_hook = btrfs_submit_bio_hook,
6013 .merge_bio_hook = btrfs_merge_bio_hook,
6014 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6015 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6016 .writepage_start_hook = btrfs_writepage_start_hook,
6017 .readpage_io_failed_hook = btrfs_io_failed_hook,
6018 .set_bit_hook = btrfs_set_bit_hook,
6019 .clear_bit_hook = btrfs_clear_bit_hook,
6020 .merge_extent_hook = btrfs_merge_extent_hook,
6021 .split_extent_hook = btrfs_split_extent_hook,
6025 * btrfs doesn't support the bmap operation because swapfiles
6026 * use bmap to make a mapping of extents in the file. They assume
6027 * these extents won't change over the life of the file and they
6028 * use the bmap result to do IO directly to the drive.
6030 * the btrfs bmap call would return logical addresses that aren't
6031 * suitable for IO and they also will change frequently as COW
6032 * operations happen. So, swapfile + btrfs == corruption.
6034 * For now we're avoiding this by dropping bmap.
6036 static const struct address_space_operations btrfs_aops = {
6037 .readpage = btrfs_readpage,
6038 .writepage = btrfs_writepage,
6039 .writepages = btrfs_writepages,
6040 .readpages = btrfs_readpages,
6041 .sync_page = block_sync_page,
6042 .direct_IO = btrfs_direct_IO,
6043 .invalidatepage = btrfs_invalidatepage,
6044 .releasepage = btrfs_releasepage,
6045 .set_page_dirty = btrfs_set_page_dirty,
6046 .error_remove_page = generic_error_remove_page,
6049 static const struct address_space_operations btrfs_symlink_aops = {
6050 .readpage = btrfs_readpage,
6051 .writepage = btrfs_writepage,
6052 .invalidatepage = btrfs_invalidatepage,
6053 .releasepage = btrfs_releasepage,
6056 static const struct inode_operations btrfs_file_inode_operations = {
6057 .truncate = btrfs_truncate,
6058 .getattr = btrfs_getattr,
6059 .setattr = btrfs_setattr,
6060 .setxattr = btrfs_setxattr,
6061 .getxattr = btrfs_getxattr,
6062 .listxattr = btrfs_listxattr,
6063 .removexattr = btrfs_removexattr,
6064 .permission = btrfs_permission,
6065 .fallocate = btrfs_fallocate,
6066 .fiemap = btrfs_fiemap,
6068 static const struct inode_operations btrfs_special_inode_operations = {
6069 .getattr = btrfs_getattr,
6070 .setattr = btrfs_setattr,
6071 .permission = btrfs_permission,
6072 .setxattr = btrfs_setxattr,
6073 .getxattr = btrfs_getxattr,
6074 .listxattr = btrfs_listxattr,
6075 .removexattr = btrfs_removexattr,
6077 static const struct inode_operations btrfs_symlink_inode_operations = {
6078 .readlink = generic_readlink,
6079 .follow_link = page_follow_link_light,
6080 .put_link = page_put_link,
6081 .permission = btrfs_permission,
6082 .setxattr = btrfs_setxattr,
6083 .getxattr = btrfs_getxattr,
6084 .listxattr = btrfs_listxattr,
6085 .removexattr = btrfs_removexattr,
6088 const struct dentry_operations btrfs_dentry_operations = {
6089 .d_delete = btrfs_dentry_delete,