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_delalloc_release_metadata(inode, end + 1 - start);
256 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
260 struct async_extent {
265 unsigned long nr_pages;
266 struct list_head list;
271 struct btrfs_root *root;
272 struct page *locked_page;
275 struct list_head extents;
276 struct btrfs_work work;
279 static noinline int add_async_extent(struct async_cow *cow,
280 u64 start, u64 ram_size,
283 unsigned long nr_pages)
285 struct async_extent *async_extent;
287 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
288 async_extent->start = start;
289 async_extent->ram_size = ram_size;
290 async_extent->compressed_size = compressed_size;
291 async_extent->pages = pages;
292 async_extent->nr_pages = nr_pages;
293 list_add_tail(&async_extent->list, &cow->extents);
298 * we create compressed extents in two phases. The first
299 * phase compresses a range of pages that have already been
300 * locked (both pages and state bits are locked).
302 * This is done inside an ordered work queue, and the compression
303 * is spread across many cpus. The actual IO submission is step
304 * two, and the ordered work queue takes care of making sure that
305 * happens in the same order things were put onto the queue by
306 * writepages and friends.
308 * If this code finds it can't get good compression, it puts an
309 * entry onto the work queue to write the uncompressed bytes. This
310 * makes sure that both compressed inodes and uncompressed inodes
311 * are written in the same order that pdflush sent them down.
313 static noinline int compress_file_range(struct inode *inode,
314 struct page *locked_page,
316 struct async_cow *async_cow,
319 struct btrfs_root *root = BTRFS_I(inode)->root;
320 struct btrfs_trans_handle *trans;
324 u64 blocksize = root->sectorsize;
326 u64 isize = i_size_read(inode);
328 struct page **pages = NULL;
329 unsigned long nr_pages;
330 unsigned long nr_pages_ret = 0;
331 unsigned long total_compressed = 0;
332 unsigned long total_in = 0;
333 unsigned long max_compressed = 128 * 1024;
334 unsigned long max_uncompressed = 128 * 1024;
340 actual_end = min_t(u64, isize, end + 1);
343 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
344 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
347 * we don't want to send crud past the end of i_size through
348 * compression, that's just a waste of CPU time. So, if the
349 * end of the file is before the start of our current
350 * requested range of bytes, we bail out to the uncompressed
351 * cleanup code that can deal with all of this.
353 * It isn't really the fastest way to fix things, but this is a
354 * very uncommon corner.
356 if (actual_end <= start)
357 goto cleanup_and_bail_uncompressed;
359 total_compressed = actual_end - start;
361 /* we want to make sure that amount of ram required to uncompress
362 * an extent is reasonable, so we limit the total size in ram
363 * of a compressed extent to 128k. This is a crucial number
364 * because it also controls how easily we can spread reads across
365 * cpus for decompression.
367 * We also want to make sure the amount of IO required to do
368 * a random read is reasonably small, so we limit the size of
369 * a compressed extent to 128k.
371 total_compressed = min(total_compressed, max_uncompressed);
372 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
373 num_bytes = max(blocksize, num_bytes);
374 disk_num_bytes = num_bytes;
379 * we do compression for mount -o compress and when the
380 * inode has not been flagged as nocompress. This flag can
381 * change at any time if we discover bad compression ratios.
383 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
384 (btrfs_test_opt(root, COMPRESS) ||
385 (BTRFS_I(inode)->force_compress))) {
387 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
389 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
390 total_compressed, pages,
391 nr_pages, &nr_pages_ret,
397 unsigned long offset = total_compressed &
398 (PAGE_CACHE_SIZE - 1);
399 struct page *page = pages[nr_pages_ret - 1];
402 /* zero the tail end of the last page, we might be
403 * sending it down to disk
406 kaddr = kmap_atomic(page, KM_USER0);
407 memset(kaddr + offset, 0,
408 PAGE_CACHE_SIZE - offset);
409 kunmap_atomic(kaddr, KM_USER0);
415 trans = btrfs_join_transaction(root, 1);
417 btrfs_set_trans_block_group(trans, inode);
418 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
420 /* lets try to make an inline extent */
421 if (ret || total_in < (actual_end - start)) {
422 /* we didn't compress the entire range, try
423 * to make an uncompressed inline extent.
425 ret = cow_file_range_inline(trans, root, inode,
426 start, end, 0, NULL);
428 /* try making a compressed inline extent */
429 ret = cow_file_range_inline(trans, root, inode,
431 total_compressed, pages);
435 * inline extent creation worked, we don't need
436 * to create any more async work items. Unlock
437 * and free up our temp pages.
439 extent_clear_unlock_delalloc(inode,
440 &BTRFS_I(inode)->io_tree,
442 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
443 EXTENT_CLEAR_DELALLOC |
444 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
446 btrfs_end_transaction(trans, root);
449 btrfs_end_transaction(trans, root);
454 * we aren't doing an inline extent round the compressed size
455 * up to a block size boundary so the allocator does sane
458 total_compressed = (total_compressed + blocksize - 1) &
462 * one last check to make sure the compression is really a
463 * win, compare the page count read with the blocks on disk
465 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
466 ~(PAGE_CACHE_SIZE - 1);
467 if (total_compressed >= total_in) {
470 disk_num_bytes = total_compressed;
471 num_bytes = total_in;
474 if (!will_compress && pages) {
476 * the compression code ran but failed to make things smaller,
477 * free any pages it allocated and our page pointer array
479 for (i = 0; i < nr_pages_ret; i++) {
480 WARN_ON(pages[i]->mapping);
481 page_cache_release(pages[i]);
485 total_compressed = 0;
488 /* flag the file so we don't compress in the future */
489 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
490 !(BTRFS_I(inode)->force_compress)) {
491 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
497 /* the async work queues will take care of doing actual
498 * allocation on disk for these compressed pages,
499 * and will submit them to the elevator.
501 add_async_extent(async_cow, start, num_bytes,
502 total_compressed, pages, nr_pages_ret);
504 if (start + num_bytes < end && start + num_bytes < actual_end) {
511 cleanup_and_bail_uncompressed:
513 * No compression, but we still need to write the pages in
514 * the file we've been given so far. redirty the locked
515 * page if it corresponds to our extent and set things up
516 * for the async work queue to run cow_file_range to do
517 * the normal delalloc dance
519 if (page_offset(locked_page) >= start &&
520 page_offset(locked_page) <= end) {
521 __set_page_dirty_nobuffers(locked_page);
522 /* unlocked later on in the async handlers */
524 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
532 for (i = 0; i < nr_pages_ret; i++) {
533 WARN_ON(pages[i]->mapping);
534 page_cache_release(pages[i]);
542 * phase two of compressed writeback. This is the ordered portion
543 * of the code, which only gets called in the order the work was
544 * queued. We walk all the async extents created by compress_file_range
545 * and send them down to the disk.
547 static noinline int submit_compressed_extents(struct inode *inode,
548 struct async_cow *async_cow)
550 struct async_extent *async_extent;
552 struct btrfs_trans_handle *trans;
553 struct btrfs_key ins;
554 struct extent_map *em;
555 struct btrfs_root *root = BTRFS_I(inode)->root;
556 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
557 struct extent_io_tree *io_tree;
560 if (list_empty(&async_cow->extents))
564 while (!list_empty(&async_cow->extents)) {
565 async_extent = list_entry(async_cow->extents.next,
566 struct async_extent, list);
567 list_del(&async_extent->list);
569 io_tree = &BTRFS_I(inode)->io_tree;
572 /* did the compression code fall back to uncompressed IO? */
573 if (!async_extent->pages) {
574 int page_started = 0;
575 unsigned long nr_written = 0;
577 lock_extent(io_tree, async_extent->start,
578 async_extent->start +
579 async_extent->ram_size - 1, GFP_NOFS);
581 /* allocate blocks */
582 ret = cow_file_range(inode, async_cow->locked_page,
584 async_extent->start +
585 async_extent->ram_size - 1,
586 &page_started, &nr_written, 0);
589 * if page_started, cow_file_range inserted an
590 * inline extent and took care of all the unlocking
591 * and IO for us. Otherwise, we need to submit
592 * all those pages down to the drive.
594 if (!page_started && !ret)
595 extent_write_locked_range(io_tree,
596 inode, async_extent->start,
597 async_extent->start +
598 async_extent->ram_size - 1,
606 lock_extent(io_tree, async_extent->start,
607 async_extent->start + async_extent->ram_size - 1,
610 trans = btrfs_join_transaction(root, 1);
611 ret = btrfs_reserve_extent(trans, root,
612 async_extent->compressed_size,
613 async_extent->compressed_size,
616 btrfs_end_transaction(trans, root);
620 for (i = 0; i < async_extent->nr_pages; i++) {
621 WARN_ON(async_extent->pages[i]->mapping);
622 page_cache_release(async_extent->pages[i]);
624 kfree(async_extent->pages);
625 async_extent->nr_pages = 0;
626 async_extent->pages = NULL;
627 unlock_extent(io_tree, async_extent->start,
628 async_extent->start +
629 async_extent->ram_size - 1, GFP_NOFS);
634 * here we're doing allocation and writeback of the
637 btrfs_drop_extent_cache(inode, async_extent->start,
638 async_extent->start +
639 async_extent->ram_size - 1, 0);
641 em = alloc_extent_map(GFP_NOFS);
642 em->start = async_extent->start;
643 em->len = async_extent->ram_size;
644 em->orig_start = em->start;
646 em->block_start = ins.objectid;
647 em->block_len = ins.offset;
648 em->bdev = root->fs_info->fs_devices->latest_bdev;
649 set_bit(EXTENT_FLAG_PINNED, &em->flags);
650 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
653 write_lock(&em_tree->lock);
654 ret = add_extent_mapping(em_tree, em);
655 write_unlock(&em_tree->lock);
656 if (ret != -EEXIST) {
660 btrfs_drop_extent_cache(inode, async_extent->start,
661 async_extent->start +
662 async_extent->ram_size - 1, 0);
665 ret = btrfs_add_ordered_extent(inode, async_extent->start,
667 async_extent->ram_size,
669 BTRFS_ORDERED_COMPRESSED);
673 * clear dirty, set writeback and unlock the pages.
675 extent_clear_unlock_delalloc(inode,
676 &BTRFS_I(inode)->io_tree,
678 async_extent->start +
679 async_extent->ram_size - 1,
680 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
681 EXTENT_CLEAR_UNLOCK |
682 EXTENT_CLEAR_DELALLOC |
683 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
685 ret = btrfs_submit_compressed_write(inode,
687 async_extent->ram_size,
689 ins.offset, async_extent->pages,
690 async_extent->nr_pages);
693 alloc_hint = ins.objectid + ins.offset;
702 * when extent_io.c finds a delayed allocation range in the file,
703 * the call backs end up in this code. The basic idea is to
704 * allocate extents on disk for the range, and create ordered data structs
705 * in ram to track those extents.
707 * locked_page is the page that writepage had locked already. We use
708 * it to make sure we don't do extra locks or unlocks.
710 * *page_started is set to one if we unlock locked_page and do everything
711 * required to start IO on it. It may be clean and already done with
714 static noinline int cow_file_range(struct inode *inode,
715 struct page *locked_page,
716 u64 start, u64 end, int *page_started,
717 unsigned long *nr_written,
720 struct btrfs_root *root = BTRFS_I(inode)->root;
721 struct btrfs_trans_handle *trans;
724 unsigned long ram_size;
727 u64 blocksize = root->sectorsize;
729 u64 isize = i_size_read(inode);
730 struct btrfs_key ins;
731 struct extent_map *em;
732 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
735 trans = btrfs_join_transaction(root, 1);
737 btrfs_set_trans_block_group(trans, inode);
738 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
740 actual_end = min_t(u64, isize, end + 1);
742 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
743 num_bytes = max(blocksize, num_bytes);
744 disk_num_bytes = num_bytes;
748 /* lets try to make an inline extent */
749 ret = cow_file_range_inline(trans, root, inode,
750 start, end, 0, NULL);
752 extent_clear_unlock_delalloc(inode,
753 &BTRFS_I(inode)->io_tree,
755 EXTENT_CLEAR_UNLOCK_PAGE |
756 EXTENT_CLEAR_UNLOCK |
757 EXTENT_CLEAR_DELALLOC |
759 EXTENT_SET_WRITEBACK |
760 EXTENT_END_WRITEBACK);
762 *nr_written = *nr_written +
763 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
770 BUG_ON(disk_num_bytes >
771 btrfs_super_total_bytes(&root->fs_info->super_copy));
774 read_lock(&BTRFS_I(inode)->extent_tree.lock);
775 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
779 * if block start isn't an actual block number then find the
780 * first block in this inode and use that as a hint. If that
781 * block is also bogus then just don't worry about it.
783 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
785 em = search_extent_mapping(em_tree, 0, 0);
786 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
787 alloc_hint = em->block_start;
791 alloc_hint = em->block_start;
795 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
796 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
798 while (disk_num_bytes > 0) {
801 cur_alloc_size = disk_num_bytes;
802 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
803 root->sectorsize, 0, alloc_hint,
807 em = alloc_extent_map(GFP_NOFS);
809 em->orig_start = em->start;
810 ram_size = ins.offset;
811 em->len = ins.offset;
813 em->block_start = ins.objectid;
814 em->block_len = ins.offset;
815 em->bdev = root->fs_info->fs_devices->latest_bdev;
816 set_bit(EXTENT_FLAG_PINNED, &em->flags);
819 write_lock(&em_tree->lock);
820 ret = add_extent_mapping(em_tree, em);
821 write_unlock(&em_tree->lock);
822 if (ret != -EEXIST) {
826 btrfs_drop_extent_cache(inode, start,
827 start + ram_size - 1, 0);
830 cur_alloc_size = ins.offset;
831 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
832 ram_size, cur_alloc_size, 0);
835 if (root->root_key.objectid ==
836 BTRFS_DATA_RELOC_TREE_OBJECTID) {
837 ret = btrfs_reloc_clone_csums(inode, start,
842 if (disk_num_bytes < cur_alloc_size)
845 /* we're not doing compressed IO, don't unlock the first
846 * page (which the caller expects to stay locked), don't
847 * clear any dirty bits and don't set any writeback bits
849 * Do set the Private2 bit so we know this page was properly
850 * setup for writepage
852 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
853 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
856 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
857 start, start + ram_size - 1,
859 disk_num_bytes -= cur_alloc_size;
860 num_bytes -= cur_alloc_size;
861 alloc_hint = ins.objectid + ins.offset;
862 start += cur_alloc_size;
866 btrfs_end_transaction(trans, root);
872 * work queue call back to started compression on a file and pages
874 static noinline void async_cow_start(struct btrfs_work *work)
876 struct async_cow *async_cow;
878 async_cow = container_of(work, struct async_cow, work);
880 compress_file_range(async_cow->inode, async_cow->locked_page,
881 async_cow->start, async_cow->end, async_cow,
884 async_cow->inode = NULL;
888 * work queue call back to submit previously compressed pages
890 static noinline void async_cow_submit(struct btrfs_work *work)
892 struct async_cow *async_cow;
893 struct btrfs_root *root;
894 unsigned long nr_pages;
896 async_cow = container_of(work, struct async_cow, work);
898 root = async_cow->root;
899 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
902 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
904 if (atomic_read(&root->fs_info->async_delalloc_pages) <
906 waitqueue_active(&root->fs_info->async_submit_wait))
907 wake_up(&root->fs_info->async_submit_wait);
909 if (async_cow->inode)
910 submit_compressed_extents(async_cow->inode, async_cow);
913 static noinline void async_cow_free(struct btrfs_work *work)
915 struct async_cow *async_cow;
916 async_cow = container_of(work, struct async_cow, work);
920 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
921 u64 start, u64 end, int *page_started,
922 unsigned long *nr_written)
924 struct async_cow *async_cow;
925 struct btrfs_root *root = BTRFS_I(inode)->root;
926 unsigned long nr_pages;
928 int limit = 10 * 1024 * 1042;
930 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
931 1, 0, NULL, GFP_NOFS);
932 while (start < end) {
933 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
934 async_cow->inode = inode;
935 async_cow->root = root;
936 async_cow->locked_page = locked_page;
937 async_cow->start = start;
939 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
942 cur_end = min(end, start + 512 * 1024 - 1);
944 async_cow->end = cur_end;
945 INIT_LIST_HEAD(&async_cow->extents);
947 async_cow->work.func = async_cow_start;
948 async_cow->work.ordered_func = async_cow_submit;
949 async_cow->work.ordered_free = async_cow_free;
950 async_cow->work.flags = 0;
952 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
954 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
956 btrfs_queue_worker(&root->fs_info->delalloc_workers,
959 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
960 wait_event(root->fs_info->async_submit_wait,
961 (atomic_read(&root->fs_info->async_delalloc_pages) <
965 while (atomic_read(&root->fs_info->async_submit_draining) &&
966 atomic_read(&root->fs_info->async_delalloc_pages)) {
967 wait_event(root->fs_info->async_submit_wait,
968 (atomic_read(&root->fs_info->async_delalloc_pages) ==
972 *nr_written += nr_pages;
979 static noinline int csum_exist_in_range(struct btrfs_root *root,
980 u64 bytenr, u64 num_bytes)
983 struct btrfs_ordered_sum *sums;
986 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
987 bytenr + num_bytes - 1, &list);
988 if (ret == 0 && list_empty(&list))
991 while (!list_empty(&list)) {
992 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
993 list_del(&sums->list);
1000 * when nowcow writeback call back. This checks for snapshots or COW copies
1001 * of the extents that exist in the file, and COWs the file as required.
1003 * If no cow copies or snapshots exist, we write directly to the existing
1006 static noinline int run_delalloc_nocow(struct inode *inode,
1007 struct page *locked_page,
1008 u64 start, u64 end, int *page_started, int force,
1009 unsigned long *nr_written)
1011 struct btrfs_root *root = BTRFS_I(inode)->root;
1012 struct btrfs_trans_handle *trans;
1013 struct extent_buffer *leaf;
1014 struct btrfs_path *path;
1015 struct btrfs_file_extent_item *fi;
1016 struct btrfs_key found_key;
1029 path = btrfs_alloc_path();
1031 trans = btrfs_join_transaction(root, 1);
1034 cow_start = (u64)-1;
1037 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1040 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1041 leaf = path->nodes[0];
1042 btrfs_item_key_to_cpu(leaf, &found_key,
1043 path->slots[0] - 1);
1044 if (found_key.objectid == inode->i_ino &&
1045 found_key.type == BTRFS_EXTENT_DATA_KEY)
1050 leaf = path->nodes[0];
1051 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1052 ret = btrfs_next_leaf(root, path);
1057 leaf = path->nodes[0];
1063 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1065 if (found_key.objectid > inode->i_ino ||
1066 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1067 found_key.offset > end)
1070 if (found_key.offset > cur_offset) {
1071 extent_end = found_key.offset;
1076 fi = btrfs_item_ptr(leaf, path->slots[0],
1077 struct btrfs_file_extent_item);
1078 extent_type = btrfs_file_extent_type(leaf, fi);
1080 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1081 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1082 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1083 extent_offset = btrfs_file_extent_offset(leaf, fi);
1084 extent_end = found_key.offset +
1085 btrfs_file_extent_num_bytes(leaf, fi);
1086 if (extent_end <= start) {
1090 if (disk_bytenr == 0)
1092 if (btrfs_file_extent_compression(leaf, fi) ||
1093 btrfs_file_extent_encryption(leaf, fi) ||
1094 btrfs_file_extent_other_encoding(leaf, fi))
1096 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1098 if (btrfs_extent_readonly(root, disk_bytenr))
1100 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1102 extent_offset, disk_bytenr))
1104 disk_bytenr += extent_offset;
1105 disk_bytenr += cur_offset - found_key.offset;
1106 num_bytes = min(end + 1, extent_end) - cur_offset;
1108 * force cow if csum exists in the range.
1109 * this ensure that csum for a given extent are
1110 * either valid or do not exist.
1112 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1115 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1116 extent_end = found_key.offset +
1117 btrfs_file_extent_inline_len(leaf, fi);
1118 extent_end = ALIGN(extent_end, root->sectorsize);
1123 if (extent_end <= start) {
1128 if (cow_start == (u64)-1)
1129 cow_start = cur_offset;
1130 cur_offset = extent_end;
1131 if (cur_offset > end)
1137 btrfs_release_path(root, path);
1138 if (cow_start != (u64)-1) {
1139 ret = cow_file_range(inode, locked_page, cow_start,
1140 found_key.offset - 1, page_started,
1143 cow_start = (u64)-1;
1146 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1147 struct extent_map *em;
1148 struct extent_map_tree *em_tree;
1149 em_tree = &BTRFS_I(inode)->extent_tree;
1150 em = alloc_extent_map(GFP_NOFS);
1151 em->start = cur_offset;
1152 em->orig_start = em->start;
1153 em->len = num_bytes;
1154 em->block_len = num_bytes;
1155 em->block_start = disk_bytenr;
1156 em->bdev = root->fs_info->fs_devices->latest_bdev;
1157 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1159 write_lock(&em_tree->lock);
1160 ret = add_extent_mapping(em_tree, em);
1161 write_unlock(&em_tree->lock);
1162 if (ret != -EEXIST) {
1163 free_extent_map(em);
1166 btrfs_drop_extent_cache(inode, em->start,
1167 em->start + em->len - 1, 0);
1169 type = BTRFS_ORDERED_PREALLOC;
1171 type = BTRFS_ORDERED_NOCOW;
1174 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1175 num_bytes, num_bytes, type);
1178 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1179 cur_offset, cur_offset + num_bytes - 1,
1180 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1181 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1182 EXTENT_SET_PRIVATE2);
1183 cur_offset = extent_end;
1184 if (cur_offset > end)
1187 btrfs_release_path(root, path);
1189 if (cur_offset <= end && cow_start == (u64)-1)
1190 cow_start = cur_offset;
1191 if (cow_start != (u64)-1) {
1192 ret = cow_file_range(inode, locked_page, cow_start, end,
1193 page_started, nr_written, 1);
1197 ret = btrfs_end_transaction(trans, root);
1199 btrfs_free_path(path);
1204 * extent_io.c call back to do delayed allocation processing
1206 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1207 u64 start, u64 end, int *page_started,
1208 unsigned long *nr_written)
1211 struct btrfs_root *root = BTRFS_I(inode)->root;
1213 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1214 ret = run_delalloc_nocow(inode, locked_page, start, end,
1215 page_started, 1, nr_written);
1216 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1217 ret = run_delalloc_nocow(inode, locked_page, start, end,
1218 page_started, 0, nr_written);
1219 else if (!btrfs_test_opt(root, COMPRESS) &&
1220 !(BTRFS_I(inode)->force_compress))
1221 ret = cow_file_range(inode, locked_page, start, end,
1222 page_started, nr_written, 1);
1224 ret = cow_file_range_async(inode, locked_page, start, end,
1225 page_started, nr_written);
1229 static int btrfs_split_extent_hook(struct inode *inode,
1230 struct extent_state *orig, u64 split)
1232 /* not delalloc, ignore it */
1233 if (!(orig->state & EXTENT_DELALLOC))
1236 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1241 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1242 * extents so we can keep track of new extents that are just merged onto old
1243 * extents, such as when we are doing sequential writes, so we can properly
1244 * account for the metadata space we'll need.
1246 static int btrfs_merge_extent_hook(struct inode *inode,
1247 struct extent_state *new,
1248 struct extent_state *other)
1250 /* not delalloc, ignore it */
1251 if (!(other->state & EXTENT_DELALLOC))
1254 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1259 * extent_io.c set_bit_hook, used to track delayed allocation
1260 * bytes in this file, and to maintain the list of inodes that
1261 * have pending delalloc work to be done.
1263 static int btrfs_set_bit_hook(struct inode *inode,
1264 struct extent_state *state, int *bits)
1268 * set_bit and clear bit hooks normally require _irqsave/restore
1269 * but in this case, we are only testeing for the DELALLOC
1270 * bit, which is only set or cleared with irqs on
1272 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1273 struct btrfs_root *root = BTRFS_I(inode)->root;
1274 u64 len = state->end + 1 - state->start;
1276 if (*bits & EXTENT_FIRST_DELALLOC)
1277 *bits &= ~EXTENT_FIRST_DELALLOC;
1279 atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1281 spin_lock(&root->fs_info->delalloc_lock);
1282 BTRFS_I(inode)->delalloc_bytes += len;
1283 root->fs_info->delalloc_bytes += len;
1284 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1285 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1286 &root->fs_info->delalloc_inodes);
1288 spin_unlock(&root->fs_info->delalloc_lock);
1294 * extent_io.c clear_bit_hook, see set_bit_hook for why
1296 static int btrfs_clear_bit_hook(struct inode *inode,
1297 struct extent_state *state, int *bits)
1300 * set_bit and clear bit hooks normally require _irqsave/restore
1301 * but in this case, we are only testeing for the DELALLOC
1302 * bit, which is only set or cleared with irqs on
1304 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1305 struct btrfs_root *root = BTRFS_I(inode)->root;
1306 u64 len = state->end + 1 - state->start;
1308 if (*bits & EXTENT_FIRST_DELALLOC)
1309 *bits &= ~EXTENT_FIRST_DELALLOC;
1310 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1311 atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1313 if (*bits & EXTENT_DO_ACCOUNTING)
1314 btrfs_delalloc_release_metadata(inode, len);
1316 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
1317 btrfs_free_reserved_data_space(inode, len);
1319 spin_lock(&root->fs_info->delalloc_lock);
1320 root->fs_info->delalloc_bytes -= len;
1321 BTRFS_I(inode)->delalloc_bytes -= len;
1323 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1324 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1325 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1327 spin_unlock(&root->fs_info->delalloc_lock);
1333 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1334 * we don't create bios that span stripes or chunks
1336 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1337 size_t size, struct bio *bio,
1338 unsigned long bio_flags)
1340 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1341 struct btrfs_mapping_tree *map_tree;
1342 u64 logical = (u64)bio->bi_sector << 9;
1347 if (bio_flags & EXTENT_BIO_COMPRESSED)
1350 length = bio->bi_size;
1351 map_tree = &root->fs_info->mapping_tree;
1352 map_length = length;
1353 ret = btrfs_map_block(map_tree, READ, logical,
1354 &map_length, NULL, 0);
1356 if (map_length < length + size)
1362 * in order to insert checksums into the metadata in large chunks,
1363 * we wait until bio submission time. All the pages in the bio are
1364 * checksummed and sums are attached onto the ordered extent record.
1366 * At IO completion time the cums attached on the ordered extent record
1367 * are inserted into the btree
1369 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1370 struct bio *bio, int mirror_num,
1371 unsigned long bio_flags)
1373 struct btrfs_root *root = BTRFS_I(inode)->root;
1376 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1382 * in order to insert checksums into the metadata in large chunks,
1383 * we wait until bio submission time. All the pages in the bio are
1384 * checksummed and sums are attached onto the ordered extent record.
1386 * At IO completion time the cums attached on the ordered extent record
1387 * are inserted into the btree
1389 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1390 int mirror_num, unsigned long bio_flags)
1392 struct btrfs_root *root = BTRFS_I(inode)->root;
1393 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1397 * extent_io.c submission hook. This does the right thing for csum calculation
1398 * on write, or reading the csums from the tree before a read
1400 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1401 int mirror_num, unsigned long bio_flags)
1403 struct btrfs_root *root = BTRFS_I(inode)->root;
1407 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1409 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1412 if (!(rw & (1 << BIO_RW))) {
1413 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1414 return btrfs_submit_compressed_read(inode, bio,
1415 mirror_num, bio_flags);
1416 } else if (!skip_sum)
1417 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1419 } else if (!skip_sum) {
1420 /* csum items have already been cloned */
1421 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1423 /* we're doing a write, do the async checksumming */
1424 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1425 inode, rw, bio, mirror_num,
1426 bio_flags, __btrfs_submit_bio_start,
1427 __btrfs_submit_bio_done);
1431 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1435 * given a list of ordered sums record them in the inode. This happens
1436 * at IO completion time based on sums calculated at bio submission time.
1438 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1439 struct inode *inode, u64 file_offset,
1440 struct list_head *list)
1442 struct btrfs_ordered_sum *sum;
1444 btrfs_set_trans_block_group(trans, inode);
1446 list_for_each_entry(sum, list, list) {
1447 btrfs_csum_file_blocks(trans,
1448 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1453 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1454 struct extent_state **cached_state)
1456 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1458 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1459 cached_state, GFP_NOFS);
1462 /* see btrfs_writepage_start_hook for details on why this is required */
1463 struct btrfs_writepage_fixup {
1465 struct btrfs_work work;
1468 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1470 struct btrfs_writepage_fixup *fixup;
1471 struct btrfs_ordered_extent *ordered;
1472 struct extent_state *cached_state = NULL;
1474 struct inode *inode;
1478 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1482 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1483 ClearPageChecked(page);
1487 inode = page->mapping->host;
1488 page_start = page_offset(page);
1489 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1491 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1492 &cached_state, GFP_NOFS);
1494 /* already ordered? We're done */
1495 if (PagePrivate2(page))
1498 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1500 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1501 page_end, &cached_state, GFP_NOFS);
1503 btrfs_start_ordered_extent(inode, ordered, 1);
1508 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1509 ClearPageChecked(page);
1511 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1512 &cached_state, GFP_NOFS);
1515 page_cache_release(page);
1519 * There are a few paths in the higher layers of the kernel that directly
1520 * set the page dirty bit without asking the filesystem if it is a
1521 * good idea. This causes problems because we want to make sure COW
1522 * properly happens and the data=ordered rules are followed.
1524 * In our case any range that doesn't have the ORDERED bit set
1525 * hasn't been properly setup for IO. We kick off an async process
1526 * to fix it up. The async helper will wait for ordered extents, set
1527 * the delalloc bit and make it safe to write the page.
1529 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1531 struct inode *inode = page->mapping->host;
1532 struct btrfs_writepage_fixup *fixup;
1533 struct btrfs_root *root = BTRFS_I(inode)->root;
1535 /* this page is properly in the ordered list */
1536 if (TestClearPagePrivate2(page))
1539 if (PageChecked(page))
1542 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1546 SetPageChecked(page);
1547 page_cache_get(page);
1548 fixup->work.func = btrfs_writepage_fixup_worker;
1550 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1554 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1555 struct inode *inode, u64 file_pos,
1556 u64 disk_bytenr, u64 disk_num_bytes,
1557 u64 num_bytes, u64 ram_bytes,
1558 u8 compression, u8 encryption,
1559 u16 other_encoding, int extent_type)
1561 struct btrfs_root *root = BTRFS_I(inode)->root;
1562 struct btrfs_file_extent_item *fi;
1563 struct btrfs_path *path;
1564 struct extent_buffer *leaf;
1565 struct btrfs_key ins;
1569 path = btrfs_alloc_path();
1572 path->leave_spinning = 1;
1575 * we may be replacing one extent in the tree with another.
1576 * The new extent is pinned in the extent map, and we don't want
1577 * to drop it from the cache until it is completely in the btree.
1579 * So, tell btrfs_drop_extents to leave this extent in the cache.
1580 * the caller is expected to unpin it and allow it to be merged
1583 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1587 ins.objectid = inode->i_ino;
1588 ins.offset = file_pos;
1589 ins.type = BTRFS_EXTENT_DATA_KEY;
1590 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1592 leaf = path->nodes[0];
1593 fi = btrfs_item_ptr(leaf, path->slots[0],
1594 struct btrfs_file_extent_item);
1595 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1596 btrfs_set_file_extent_type(leaf, fi, extent_type);
1597 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1598 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1599 btrfs_set_file_extent_offset(leaf, fi, 0);
1600 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1601 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1602 btrfs_set_file_extent_compression(leaf, fi, compression);
1603 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1604 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1606 btrfs_unlock_up_safe(path, 1);
1607 btrfs_set_lock_blocking(leaf);
1609 btrfs_mark_buffer_dirty(leaf);
1611 inode_add_bytes(inode, num_bytes);
1613 ins.objectid = disk_bytenr;
1614 ins.offset = disk_num_bytes;
1615 ins.type = BTRFS_EXTENT_ITEM_KEY;
1616 ret = btrfs_alloc_reserved_file_extent(trans, root,
1617 root->root_key.objectid,
1618 inode->i_ino, file_pos, &ins);
1620 btrfs_free_path(path);
1626 * helper function for btrfs_finish_ordered_io, this
1627 * just reads in some of the csum leaves to prime them into ram
1628 * before we start the transaction. It limits the amount of btree
1629 * reads required while inside the transaction.
1631 /* as ordered data IO finishes, this gets called so we can finish
1632 * an ordered extent if the range of bytes in the file it covers are
1635 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1637 struct btrfs_root *root = BTRFS_I(inode)->root;
1638 struct btrfs_trans_handle *trans = NULL;
1639 struct btrfs_ordered_extent *ordered_extent = NULL;
1640 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1641 struct extent_state *cached_state = NULL;
1645 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1649 BUG_ON(!ordered_extent);
1651 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1652 BUG_ON(!list_empty(&ordered_extent->list));
1653 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1655 trans = btrfs_join_transaction(root, 1);
1656 btrfs_set_trans_block_group(trans, inode);
1657 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1658 ret = btrfs_update_inode(trans, root, inode);
1664 lock_extent_bits(io_tree, ordered_extent->file_offset,
1665 ordered_extent->file_offset + ordered_extent->len - 1,
1666 0, &cached_state, GFP_NOFS);
1668 trans = btrfs_join_transaction(root, 1);
1669 btrfs_set_trans_block_group(trans, inode);
1670 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1672 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1674 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1676 ret = btrfs_mark_extent_written(trans, inode,
1677 ordered_extent->file_offset,
1678 ordered_extent->file_offset +
1679 ordered_extent->len);
1682 ret = insert_reserved_file_extent(trans, inode,
1683 ordered_extent->file_offset,
1684 ordered_extent->start,
1685 ordered_extent->disk_len,
1686 ordered_extent->len,
1687 ordered_extent->len,
1689 BTRFS_FILE_EXTENT_REG);
1690 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1691 ordered_extent->file_offset,
1692 ordered_extent->len);
1695 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1696 ordered_extent->file_offset +
1697 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1699 add_pending_csums(trans, inode, ordered_extent->file_offset,
1700 &ordered_extent->list);
1702 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1703 ret = btrfs_update_inode(trans, root, inode);
1706 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1708 btrfs_end_transaction(trans, root);
1710 btrfs_put_ordered_extent(ordered_extent);
1711 /* once for the tree */
1712 btrfs_put_ordered_extent(ordered_extent);
1717 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1718 struct extent_state *state, int uptodate)
1720 ClearPagePrivate2(page);
1721 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1725 * When IO fails, either with EIO or csum verification fails, we
1726 * try other mirrors that might have a good copy of the data. This
1727 * io_failure_record is used to record state as we go through all the
1728 * mirrors. If another mirror has good data, the page is set up to date
1729 * and things continue. If a good mirror can't be found, the original
1730 * bio end_io callback is called to indicate things have failed.
1732 struct io_failure_record {
1737 unsigned long bio_flags;
1741 static int btrfs_io_failed_hook(struct bio *failed_bio,
1742 struct page *page, u64 start, u64 end,
1743 struct extent_state *state)
1745 struct io_failure_record *failrec = NULL;
1747 struct extent_map *em;
1748 struct inode *inode = page->mapping->host;
1749 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1750 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1757 ret = get_state_private(failure_tree, start, &private);
1759 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1762 failrec->start = start;
1763 failrec->len = end - start + 1;
1764 failrec->last_mirror = 0;
1765 failrec->bio_flags = 0;
1767 read_lock(&em_tree->lock);
1768 em = lookup_extent_mapping(em_tree, start, failrec->len);
1769 if (em->start > start || em->start + em->len < start) {
1770 free_extent_map(em);
1773 read_unlock(&em_tree->lock);
1775 if (!em || IS_ERR(em)) {
1779 logical = start - em->start;
1780 logical = em->block_start + logical;
1781 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1782 logical = em->block_start;
1783 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1785 failrec->logical = logical;
1786 free_extent_map(em);
1787 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1788 EXTENT_DIRTY, GFP_NOFS);
1789 set_state_private(failure_tree, start,
1790 (u64)(unsigned long)failrec);
1792 failrec = (struct io_failure_record *)(unsigned long)private;
1794 num_copies = btrfs_num_copies(
1795 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1796 failrec->logical, failrec->len);
1797 failrec->last_mirror++;
1799 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1800 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1803 if (state && state->start != failrec->start)
1805 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1807 if (!state || failrec->last_mirror > num_copies) {
1808 set_state_private(failure_tree, failrec->start, 0);
1809 clear_extent_bits(failure_tree, failrec->start,
1810 failrec->start + failrec->len - 1,
1811 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1815 bio = bio_alloc(GFP_NOFS, 1);
1816 bio->bi_private = state;
1817 bio->bi_end_io = failed_bio->bi_end_io;
1818 bio->bi_sector = failrec->logical >> 9;
1819 bio->bi_bdev = failed_bio->bi_bdev;
1822 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1823 if (failed_bio->bi_rw & (1 << BIO_RW))
1828 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1829 failrec->last_mirror,
1830 failrec->bio_flags);
1835 * each time an IO finishes, we do a fast check in the IO failure tree
1836 * to see if we need to process or clean up an io_failure_record
1838 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1841 u64 private_failure;
1842 struct io_failure_record *failure;
1846 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1847 (u64)-1, 1, EXTENT_DIRTY)) {
1848 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1849 start, &private_failure);
1851 failure = (struct io_failure_record *)(unsigned long)
1853 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1855 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1857 failure->start + failure->len - 1,
1858 EXTENT_DIRTY | EXTENT_LOCKED,
1867 * when reads are done, we need to check csums to verify the data is correct
1868 * if there's a match, we allow the bio to finish. If not, we go through
1869 * the io_failure_record routines to find good copies
1871 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1872 struct extent_state *state)
1874 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1875 struct inode *inode = page->mapping->host;
1876 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1878 u64 private = ~(u32)0;
1880 struct btrfs_root *root = BTRFS_I(inode)->root;
1883 if (PageChecked(page)) {
1884 ClearPageChecked(page);
1888 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1891 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1892 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1893 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1898 if (state && state->start == start) {
1899 private = state->private;
1902 ret = get_state_private(io_tree, start, &private);
1904 kaddr = kmap_atomic(page, KM_USER0);
1908 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1909 btrfs_csum_final(csum, (char *)&csum);
1910 if (csum != private)
1913 kunmap_atomic(kaddr, KM_USER0);
1915 /* if the io failure tree for this inode is non-empty,
1916 * check to see if we've recovered from a failed IO
1918 btrfs_clean_io_failures(inode, start);
1922 if (printk_ratelimit()) {
1923 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1924 "private %llu\n", page->mapping->host->i_ino,
1925 (unsigned long long)start, csum,
1926 (unsigned long long)private);
1928 memset(kaddr + offset, 1, end - start + 1);
1929 flush_dcache_page(page);
1930 kunmap_atomic(kaddr, KM_USER0);
1936 struct delayed_iput {
1937 struct list_head list;
1938 struct inode *inode;
1941 void btrfs_add_delayed_iput(struct inode *inode)
1943 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1944 struct delayed_iput *delayed;
1946 if (atomic_add_unless(&inode->i_count, -1, 1))
1949 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1950 delayed->inode = inode;
1952 spin_lock(&fs_info->delayed_iput_lock);
1953 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1954 spin_unlock(&fs_info->delayed_iput_lock);
1957 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1960 struct btrfs_fs_info *fs_info = root->fs_info;
1961 struct delayed_iput *delayed;
1964 spin_lock(&fs_info->delayed_iput_lock);
1965 empty = list_empty(&fs_info->delayed_iputs);
1966 spin_unlock(&fs_info->delayed_iput_lock);
1970 down_read(&root->fs_info->cleanup_work_sem);
1971 spin_lock(&fs_info->delayed_iput_lock);
1972 list_splice_init(&fs_info->delayed_iputs, &list);
1973 spin_unlock(&fs_info->delayed_iput_lock);
1975 while (!list_empty(&list)) {
1976 delayed = list_entry(list.next, struct delayed_iput, list);
1977 list_del(&delayed->list);
1978 iput(delayed->inode);
1981 up_read(&root->fs_info->cleanup_work_sem);
1985 * calculate extra metadata reservation when snapshotting a subvolume
1986 * contains orphan files.
1988 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
1989 struct btrfs_pending_snapshot *pending,
1990 u64 *bytes_to_reserve)
1992 struct btrfs_root *root;
1993 struct btrfs_block_rsv *block_rsv;
1997 root = pending->root;
1998 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2001 block_rsv = root->orphan_block_rsv;
2003 /* orphan block reservation for the snapshot */
2004 num_bytes = block_rsv->size;
2007 * after the snapshot is created, COWing tree blocks may use more
2008 * space than it frees. So we should make sure there is enough
2011 index = trans->transid & 0x1;
2012 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2013 num_bytes += block_rsv->size -
2014 (block_rsv->reserved + block_rsv->freed[index]);
2017 *bytes_to_reserve += num_bytes;
2020 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2021 struct btrfs_pending_snapshot *pending)
2023 struct btrfs_root *root = pending->root;
2024 struct btrfs_root *snap = pending->snap;
2025 struct btrfs_block_rsv *block_rsv;
2030 if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2033 /* refill source subvolume's orphan block reservation */
2034 block_rsv = root->orphan_block_rsv;
2035 index = trans->transid & 0x1;
2036 if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2037 num_bytes = block_rsv->size -
2038 (block_rsv->reserved + block_rsv->freed[index]);
2039 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2040 root->orphan_block_rsv,
2045 /* setup orphan block reservation for the snapshot */
2046 block_rsv = btrfs_alloc_block_rsv(snap);
2049 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2050 snap->orphan_block_rsv = block_rsv;
2052 num_bytes = root->orphan_block_rsv->size;
2053 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2054 block_rsv, num_bytes);
2058 /* insert orphan item for the snapshot */
2059 WARN_ON(!root->orphan_item_inserted);
2060 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2061 snap->root_key.objectid);
2063 snap->orphan_item_inserted = 1;
2067 enum btrfs_orphan_cleanup_state {
2068 ORPHAN_CLEANUP_STARTED = 1,
2069 ORPHAN_CLEANUP_DONE = 2,
2073 * This is called in transaction commmit time. If there are no orphan
2074 * files in the subvolume, it removes orphan item and frees block_rsv
2077 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2078 struct btrfs_root *root)
2082 if (!list_empty(&root->orphan_list) ||
2083 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2086 if (root->orphan_item_inserted &&
2087 btrfs_root_refs(&root->root_item) > 0) {
2088 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2089 root->root_key.objectid);
2091 root->orphan_item_inserted = 0;
2094 if (root->orphan_block_rsv) {
2095 WARN_ON(root->orphan_block_rsv->size > 0);
2096 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2097 root->orphan_block_rsv = NULL;
2102 * This creates an orphan entry for the given inode in case something goes
2103 * wrong in the middle of an unlink/truncate.
2105 * NOTE: caller of this function should reserve 5 units of metadata for
2108 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2110 struct btrfs_root *root = BTRFS_I(inode)->root;
2111 struct btrfs_block_rsv *block_rsv = NULL;
2116 if (!root->orphan_block_rsv) {
2117 block_rsv = btrfs_alloc_block_rsv(root);
2121 spin_lock(&root->orphan_lock);
2122 if (!root->orphan_block_rsv) {
2123 root->orphan_block_rsv = block_rsv;
2124 } else if (block_rsv) {
2125 btrfs_free_block_rsv(root, block_rsv);
2129 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2130 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2133 * For proper ENOSPC handling, we should do orphan
2134 * cleanup when mounting. But this introduces backward
2135 * compatibility issue.
2137 if (!xchg(&root->orphan_item_inserted, 1))
2144 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2147 if (!BTRFS_I(inode)->orphan_meta_reserved) {
2148 BTRFS_I(inode)->orphan_meta_reserved = 1;
2151 spin_unlock(&root->orphan_lock);
2154 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2156 /* grab metadata reservation from transaction handle */
2158 ret = btrfs_orphan_reserve_metadata(trans, inode);
2162 /* insert an orphan item to track this unlinked/truncated file */
2164 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2168 /* insert an orphan item to track subvolume contains orphan files */
2170 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2171 root->root_key.objectid);
2178 * We have done the truncate/delete so we can go ahead and remove the orphan
2179 * item for this particular inode.
2181 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2183 struct btrfs_root *root = BTRFS_I(inode)->root;
2184 int delete_item = 0;
2185 int release_rsv = 0;
2188 spin_lock(&root->orphan_lock);
2189 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2190 list_del_init(&BTRFS_I(inode)->i_orphan);
2194 if (BTRFS_I(inode)->orphan_meta_reserved) {
2195 BTRFS_I(inode)->orphan_meta_reserved = 0;
2198 spin_unlock(&root->orphan_lock);
2200 if (trans && delete_item) {
2201 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2206 btrfs_orphan_release_metadata(inode);
2212 * this cleans up any orphans that may be left on the list from the last use
2215 void btrfs_orphan_cleanup(struct btrfs_root *root)
2217 struct btrfs_path *path;
2218 struct extent_buffer *leaf;
2219 struct btrfs_item *item;
2220 struct btrfs_key key, found_key;
2221 struct btrfs_trans_handle *trans;
2222 struct inode *inode;
2223 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2225 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2228 path = btrfs_alloc_path();
2232 key.objectid = BTRFS_ORPHAN_OBJECTID;
2233 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2234 key.offset = (u64)-1;
2237 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2239 printk(KERN_ERR "Error searching slot for orphan: %d"
2245 * if ret == 0 means we found what we were searching for, which
2246 * is weird, but possible, so only screw with path if we didnt
2247 * find the key and see if we have stuff that matches
2250 if (path->slots[0] == 0)
2255 /* pull out the item */
2256 leaf = path->nodes[0];
2257 item = btrfs_item_nr(leaf, path->slots[0]);
2258 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2260 /* make sure the item matches what we want */
2261 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2263 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2266 /* release the path since we're done with it */
2267 btrfs_release_path(root, path);
2270 * this is where we are basically btrfs_lookup, without the
2271 * crossing root thing. we store the inode number in the
2272 * offset of the orphan item.
2274 found_key.objectid = found_key.offset;
2275 found_key.type = BTRFS_INODE_ITEM_KEY;
2276 found_key.offset = 0;
2277 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2278 BUG_ON(IS_ERR(inode));
2281 * add this inode to the orphan list so btrfs_orphan_del does
2282 * the proper thing when we hit it
2284 spin_lock(&root->orphan_lock);
2285 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2286 spin_unlock(&root->orphan_lock);
2289 * if this is a bad inode, means we actually succeeded in
2290 * removing the inode, but not the orphan record, which means
2291 * we need to manually delete the orphan since iput will just
2292 * do a destroy_inode
2294 if (is_bad_inode(inode)) {
2295 trans = btrfs_start_transaction(root, 0);
2296 btrfs_orphan_del(trans, inode);
2297 btrfs_end_transaction(trans, root);
2302 /* if we have links, this was a truncate, lets do that */
2303 if (inode->i_nlink) {
2305 btrfs_truncate(inode);
2310 /* this will do delete_inode and everything for us */
2313 btrfs_free_path(path);
2315 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2317 if (root->orphan_block_rsv)
2318 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2321 if (root->orphan_block_rsv || root->orphan_item_inserted) {
2322 trans = btrfs_join_transaction(root, 1);
2323 btrfs_end_transaction(trans, root);
2327 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2329 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2333 * very simple check to peek ahead in the leaf looking for xattrs. If we
2334 * don't find any xattrs, we know there can't be any acls.
2336 * slot is the slot the inode is in, objectid is the objectid of the inode
2338 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2339 int slot, u64 objectid)
2341 u32 nritems = btrfs_header_nritems(leaf);
2342 struct btrfs_key found_key;
2346 while (slot < nritems) {
2347 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2349 /* we found a different objectid, there must not be acls */
2350 if (found_key.objectid != objectid)
2353 /* we found an xattr, assume we've got an acl */
2354 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2358 * we found a key greater than an xattr key, there can't
2359 * be any acls later on
2361 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2368 * it goes inode, inode backrefs, xattrs, extents,
2369 * so if there are a ton of hard links to an inode there can
2370 * be a lot of backrefs. Don't waste time searching too hard,
2371 * this is just an optimization
2376 /* we hit the end of the leaf before we found an xattr or
2377 * something larger than an xattr. We have to assume the inode
2384 * read an inode from the btree into the in-memory inode
2386 static void btrfs_read_locked_inode(struct inode *inode)
2388 struct btrfs_path *path;
2389 struct extent_buffer *leaf;
2390 struct btrfs_inode_item *inode_item;
2391 struct btrfs_timespec *tspec;
2392 struct btrfs_root *root = BTRFS_I(inode)->root;
2393 struct btrfs_key location;
2395 u64 alloc_group_block;
2399 path = btrfs_alloc_path();
2401 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2403 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2407 leaf = path->nodes[0];
2408 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2409 struct btrfs_inode_item);
2411 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2412 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2413 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2414 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2415 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2417 tspec = btrfs_inode_atime(inode_item);
2418 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2419 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2421 tspec = btrfs_inode_mtime(inode_item);
2422 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2423 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2425 tspec = btrfs_inode_ctime(inode_item);
2426 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2427 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2429 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2430 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2431 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2432 inode->i_generation = BTRFS_I(inode)->generation;
2434 rdev = btrfs_inode_rdev(leaf, inode_item);
2436 BTRFS_I(inode)->index_cnt = (u64)-1;
2437 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2439 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2442 * try to precache a NULL acl entry for files that don't have
2443 * any xattrs or acls
2445 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2447 cache_no_acl(inode);
2449 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2450 alloc_group_block, 0);
2451 btrfs_free_path(path);
2454 switch (inode->i_mode & S_IFMT) {
2456 inode->i_mapping->a_ops = &btrfs_aops;
2457 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2458 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2459 inode->i_fop = &btrfs_file_operations;
2460 inode->i_op = &btrfs_file_inode_operations;
2463 inode->i_fop = &btrfs_dir_file_operations;
2464 if (root == root->fs_info->tree_root)
2465 inode->i_op = &btrfs_dir_ro_inode_operations;
2467 inode->i_op = &btrfs_dir_inode_operations;
2470 inode->i_op = &btrfs_symlink_inode_operations;
2471 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2472 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2475 inode->i_op = &btrfs_special_inode_operations;
2476 init_special_inode(inode, inode->i_mode, rdev);
2480 btrfs_update_iflags(inode);
2484 btrfs_free_path(path);
2485 make_bad_inode(inode);
2489 * given a leaf and an inode, copy the inode fields into the leaf
2491 static void fill_inode_item(struct btrfs_trans_handle *trans,
2492 struct extent_buffer *leaf,
2493 struct btrfs_inode_item *item,
2494 struct inode *inode)
2496 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2497 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2498 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2499 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2500 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2502 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2503 inode->i_atime.tv_sec);
2504 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2505 inode->i_atime.tv_nsec);
2507 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2508 inode->i_mtime.tv_sec);
2509 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2510 inode->i_mtime.tv_nsec);
2512 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2513 inode->i_ctime.tv_sec);
2514 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2515 inode->i_ctime.tv_nsec);
2517 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2518 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2519 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2520 btrfs_set_inode_transid(leaf, item, trans->transid);
2521 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2522 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2523 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2527 * copy everything in the in-memory inode into the btree.
2529 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2530 struct btrfs_root *root, struct inode *inode)
2532 struct btrfs_inode_item *inode_item;
2533 struct btrfs_path *path;
2534 struct extent_buffer *leaf;
2537 path = btrfs_alloc_path();
2539 path->leave_spinning = 1;
2540 ret = btrfs_lookup_inode(trans, root, path,
2541 &BTRFS_I(inode)->location, 1);
2548 btrfs_unlock_up_safe(path, 1);
2549 leaf = path->nodes[0];
2550 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2551 struct btrfs_inode_item);
2553 fill_inode_item(trans, leaf, inode_item, inode);
2554 btrfs_mark_buffer_dirty(leaf);
2555 btrfs_set_inode_last_trans(trans, inode);
2558 btrfs_free_path(path);
2564 * unlink helper that gets used here in inode.c and in the tree logging
2565 * recovery code. It remove a link in a directory with a given name, and
2566 * also drops the back refs in the inode to the directory
2568 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2569 struct btrfs_root *root,
2570 struct inode *dir, struct inode *inode,
2571 const char *name, int name_len)
2573 struct btrfs_path *path;
2575 struct extent_buffer *leaf;
2576 struct btrfs_dir_item *di;
2577 struct btrfs_key key;
2580 path = btrfs_alloc_path();
2586 path->leave_spinning = 1;
2587 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2588 name, name_len, -1);
2597 leaf = path->nodes[0];
2598 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2599 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2602 btrfs_release_path(root, path);
2604 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2606 dir->i_ino, &index);
2608 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2609 "inode %lu parent %lu\n", name_len, name,
2610 inode->i_ino, dir->i_ino);
2614 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2615 index, name, name_len, -1);
2624 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2625 btrfs_release_path(root, path);
2627 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2629 BUG_ON(ret != 0 && ret != -ENOENT);
2631 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2635 btrfs_free_path(path);
2639 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2640 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2641 btrfs_update_inode(trans, root, dir);
2642 btrfs_drop_nlink(inode);
2643 ret = btrfs_update_inode(trans, root, inode);
2648 /* helper to check if there is any shared block in the path */
2649 static int check_path_shared(struct btrfs_root *root,
2650 struct btrfs_path *path)
2652 struct extent_buffer *eb;
2657 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2658 if (!path->nodes[level])
2660 eb = path->nodes[level];
2661 if (!btrfs_block_can_be_shared(root, eb))
2663 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2672 * helper to start transaction for unlink and rmdir.
2674 * unlink and rmdir are special in btrfs, they do not always free space.
2675 * so in enospc case, we should make sure they will free space before
2676 * allowing them to use the global metadata reservation.
2678 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2679 struct dentry *dentry)
2681 struct btrfs_trans_handle *trans;
2682 struct btrfs_root *root = BTRFS_I(dir)->root;
2683 struct btrfs_path *path;
2684 struct btrfs_inode_ref *ref;
2685 struct btrfs_dir_item *di;
2686 struct inode *inode = dentry->d_inode;
2692 trans = btrfs_start_transaction(root, 10);
2693 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2696 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2697 return ERR_PTR(-ENOSPC);
2699 /* check if there is someone else holds reference */
2700 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2701 return ERR_PTR(-ENOSPC);
2703 if (atomic_read(&inode->i_count) > 2)
2704 return ERR_PTR(-ENOSPC);
2706 if (xchg(&root->fs_info->enospc_unlink, 1))
2707 return ERR_PTR(-ENOSPC);
2709 path = btrfs_alloc_path();
2711 root->fs_info->enospc_unlink = 0;
2712 return ERR_PTR(-ENOMEM);
2715 trans = btrfs_start_transaction(root, 0);
2716 if (IS_ERR(trans)) {
2717 btrfs_free_path(path);
2718 root->fs_info->enospc_unlink = 0;
2722 path->skip_locking = 1;
2723 path->search_commit_root = 1;
2725 ret = btrfs_lookup_inode(trans, root, path,
2726 &BTRFS_I(dir)->location, 0);
2732 if (check_path_shared(root, path))
2737 btrfs_release_path(root, path);
2739 ret = btrfs_lookup_inode(trans, root, path,
2740 &BTRFS_I(inode)->location, 0);
2746 if (check_path_shared(root, path))
2751 btrfs_release_path(root, path);
2753 if (ret == 0 && S_ISREG(inode->i_mode)) {
2754 ret = btrfs_lookup_file_extent(trans, root, path,
2755 inode->i_ino, (u64)-1, 0);
2761 if (check_path_shared(root, path))
2763 btrfs_release_path(root, path);
2771 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2772 dentry->d_name.name, dentry->d_name.len, 0);
2778 if (check_path_shared(root, path))
2784 btrfs_release_path(root, path);
2786 ref = btrfs_lookup_inode_ref(trans, root, path,
2787 dentry->d_name.name, dentry->d_name.len,
2788 inode->i_ino, dir->i_ino, 0);
2794 if (check_path_shared(root, path))
2796 index = btrfs_inode_ref_index(path->nodes[0], ref);
2797 btrfs_release_path(root, path);
2799 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2800 dentry->d_name.name, dentry->d_name.len, 0);
2805 BUG_ON(ret == -ENOENT);
2806 if (check_path_shared(root, path))
2811 btrfs_free_path(path);
2813 btrfs_end_transaction(trans, root);
2814 root->fs_info->enospc_unlink = 0;
2815 return ERR_PTR(err);
2818 trans->block_rsv = &root->fs_info->global_block_rsv;
2822 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2823 struct btrfs_root *root)
2825 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2826 BUG_ON(!root->fs_info->enospc_unlink);
2827 root->fs_info->enospc_unlink = 0;
2829 btrfs_end_transaction_throttle(trans, root);
2832 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2834 struct btrfs_root *root = BTRFS_I(dir)->root;
2835 struct btrfs_trans_handle *trans;
2836 struct inode *inode = dentry->d_inode;
2838 unsigned long nr = 0;
2840 trans = __unlink_start_trans(dir, dentry);
2842 return PTR_ERR(trans);
2844 btrfs_set_trans_block_group(trans, dir);
2846 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2848 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2849 dentry->d_name.name, dentry->d_name.len);
2852 if (inode->i_nlink == 0) {
2853 ret = btrfs_orphan_add(trans, inode);
2857 nr = trans->blocks_used;
2858 __unlink_end_trans(trans, root);
2859 btrfs_btree_balance_dirty(root, nr);
2863 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2864 struct btrfs_root *root,
2865 struct inode *dir, u64 objectid,
2866 const char *name, int name_len)
2868 struct btrfs_path *path;
2869 struct extent_buffer *leaf;
2870 struct btrfs_dir_item *di;
2871 struct btrfs_key key;
2875 path = btrfs_alloc_path();
2879 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2880 name, name_len, -1);
2881 BUG_ON(!di || IS_ERR(di));
2883 leaf = path->nodes[0];
2884 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2885 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2886 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2888 btrfs_release_path(root, path);
2890 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2891 objectid, root->root_key.objectid,
2892 dir->i_ino, &index, name, name_len);
2894 BUG_ON(ret != -ENOENT);
2895 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2897 BUG_ON(!di || IS_ERR(di));
2899 leaf = path->nodes[0];
2900 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2901 btrfs_release_path(root, path);
2905 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2906 index, name, name_len, -1);
2907 BUG_ON(!di || IS_ERR(di));
2909 leaf = path->nodes[0];
2910 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2911 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2912 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2914 btrfs_release_path(root, path);
2916 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2917 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2918 ret = btrfs_update_inode(trans, root, dir);
2920 dir->i_sb->s_dirt = 1;
2922 btrfs_free_path(path);
2926 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2928 struct inode *inode = dentry->d_inode;
2930 struct btrfs_root *root = BTRFS_I(dir)->root;
2931 struct btrfs_trans_handle *trans;
2932 unsigned long nr = 0;
2934 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2935 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2938 trans = __unlink_start_trans(dir, dentry);
2940 return PTR_ERR(trans);
2942 btrfs_set_trans_block_group(trans, dir);
2944 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2945 err = btrfs_unlink_subvol(trans, root, dir,
2946 BTRFS_I(inode)->location.objectid,
2947 dentry->d_name.name,
2948 dentry->d_name.len);
2952 err = btrfs_orphan_add(trans, inode);
2956 /* now the directory is empty */
2957 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2958 dentry->d_name.name, dentry->d_name.len);
2960 btrfs_i_size_write(inode, 0);
2962 nr = trans->blocks_used;
2963 __unlink_end_trans(trans, root);
2964 btrfs_btree_balance_dirty(root, nr);
2971 * when truncating bytes in a file, it is possible to avoid reading
2972 * the leaves that contain only checksum items. This can be the
2973 * majority of the IO required to delete a large file, but it must
2974 * be done carefully.
2976 * The keys in the level just above the leaves are checked to make sure
2977 * the lowest key in a given leaf is a csum key, and starts at an offset
2978 * after the new size.
2980 * Then the key for the next leaf is checked to make sure it also has
2981 * a checksum item for the same file. If it does, we know our target leaf
2982 * contains only checksum items, and it can be safely freed without reading
2985 * This is just an optimization targeted at large files. It may do
2986 * nothing. It will return 0 unless things went badly.
2988 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2989 struct btrfs_root *root,
2990 struct btrfs_path *path,
2991 struct inode *inode, u64 new_size)
2993 struct btrfs_key key;
2996 struct btrfs_key found_key;
2997 struct btrfs_key other_key;
2998 struct btrfs_leaf_ref *ref;
3002 path->lowest_level = 1;
3003 key.objectid = inode->i_ino;
3004 key.type = BTRFS_CSUM_ITEM_KEY;
3005 key.offset = new_size;
3007 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3011 if (path->nodes[1] == NULL) {
3016 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3017 nritems = btrfs_header_nritems(path->nodes[1]);
3022 if (path->slots[1] >= nritems)
3025 /* did we find a key greater than anything we want to delete? */
3026 if (found_key.objectid > inode->i_ino ||
3027 (found_key.objectid == inode->i_ino && found_key.type > key.type))
3030 /* we check the next key in the node to make sure the leave contains
3031 * only checksum items. This comparison doesn't work if our
3032 * leaf is the last one in the node
3034 if (path->slots[1] + 1 >= nritems) {
3036 /* search forward from the last key in the node, this
3037 * will bring us into the next node in the tree
3039 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3041 /* unlikely, but we inc below, so check to be safe */
3042 if (found_key.offset == (u64)-1)
3045 /* search_forward needs a path with locks held, do the
3046 * search again for the original key. It is possible
3047 * this will race with a balance and return a path that
3048 * we could modify, but this drop is just an optimization
3049 * and is allowed to miss some leaves.
3051 btrfs_release_path(root, path);
3054 /* setup a max key for search_forward */
3055 other_key.offset = (u64)-1;
3056 other_key.type = key.type;
3057 other_key.objectid = key.objectid;
3059 path->keep_locks = 1;
3060 ret = btrfs_search_forward(root, &found_key, &other_key,
3062 path->keep_locks = 0;
3063 if (ret || found_key.objectid != key.objectid ||
3064 found_key.type != key.type) {
3069 key.offset = found_key.offset;
3070 btrfs_release_path(root, path);
3075 /* we know there's one more slot after us in the tree,
3076 * read that key so we can verify it is also a checksum item
3078 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3080 if (found_key.objectid < inode->i_ino)
3083 if (found_key.type != key.type || found_key.offset < new_size)
3087 * if the key for the next leaf isn't a csum key from this objectid,
3088 * we can't be sure there aren't good items inside this leaf.
3091 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3094 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3095 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3097 * it is safe to delete this leaf, it contains only
3098 * csum items from this inode at an offset >= new_size
3100 ret = btrfs_del_leaf(trans, root, path, leaf_start);
3103 if (root->ref_cows && leaf_gen < trans->transid) {
3104 ref = btrfs_alloc_leaf_ref(root, 0);
3106 ref->root_gen = root->root_key.offset;
3107 ref->bytenr = leaf_start;
3109 ref->generation = leaf_gen;
3112 btrfs_sort_leaf_ref(ref);
3114 ret = btrfs_add_leaf_ref(root, ref, 0);
3116 btrfs_free_leaf_ref(root, ref);
3122 btrfs_release_path(root, path);
3124 if (other_key.objectid == inode->i_ino &&
3125 other_key.type == key.type && other_key.offset > key.offset) {
3126 key.offset = other_key.offset;
3132 /* fixup any changes we've made to the path */
3133 path->lowest_level = 0;
3134 path->keep_locks = 0;
3135 btrfs_release_path(root, path);
3142 * this can truncate away extent items, csum items and directory items.
3143 * It starts at a high offset and removes keys until it can't find
3144 * any higher than new_size
3146 * csum items that cross the new i_size are truncated to the new size
3149 * min_type is the minimum key type to truncate down to. If set to 0, this
3150 * will kill all the items on this inode, including the INODE_ITEM_KEY.
3152 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3153 struct btrfs_root *root,
3154 struct inode *inode,
3155 u64 new_size, u32 min_type)
3157 struct btrfs_path *path;
3158 struct extent_buffer *leaf;
3159 struct btrfs_file_extent_item *fi;
3160 struct btrfs_key key;
3161 struct btrfs_key found_key;
3162 u64 extent_start = 0;
3163 u64 extent_num_bytes = 0;
3164 u64 extent_offset = 0;
3166 u64 mask = root->sectorsize - 1;
3167 u32 found_type = (u8)-1;
3170 int pending_del_nr = 0;
3171 int pending_del_slot = 0;
3172 int extent_type = -1;
3177 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3180 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3182 path = btrfs_alloc_path();
3186 key.objectid = inode->i_ino;
3187 key.offset = (u64)-1;
3191 path->leave_spinning = 1;
3192 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3199 /* there are no items in the tree for us to truncate, we're
3202 if (path->slots[0] == 0)
3209 leaf = path->nodes[0];
3210 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3211 found_type = btrfs_key_type(&found_key);
3214 if (found_key.objectid != inode->i_ino)
3217 if (found_type < min_type)
3220 item_end = found_key.offset;
3221 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3222 fi = btrfs_item_ptr(leaf, path->slots[0],
3223 struct btrfs_file_extent_item);
3224 extent_type = btrfs_file_extent_type(leaf, fi);
3225 encoding = btrfs_file_extent_compression(leaf, fi);
3226 encoding |= btrfs_file_extent_encryption(leaf, fi);
3227 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3229 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3231 btrfs_file_extent_num_bytes(leaf, fi);
3232 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3233 item_end += btrfs_file_extent_inline_len(leaf,
3238 if (found_type > min_type) {
3241 if (item_end < new_size)
3243 if (found_key.offset >= new_size)
3249 /* FIXME, shrink the extent if the ref count is only 1 */
3250 if (found_type != BTRFS_EXTENT_DATA_KEY)
3253 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3255 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3256 if (!del_item && !encoding) {
3257 u64 orig_num_bytes =
3258 btrfs_file_extent_num_bytes(leaf, fi);
3259 extent_num_bytes = new_size -
3260 found_key.offset + root->sectorsize - 1;
3261 extent_num_bytes = extent_num_bytes &
3262 ~((u64)root->sectorsize - 1);
3263 btrfs_set_file_extent_num_bytes(leaf, fi,
3265 num_dec = (orig_num_bytes -
3267 if (root->ref_cows && extent_start != 0)
3268 inode_sub_bytes(inode, num_dec);
3269 btrfs_mark_buffer_dirty(leaf);
3272 btrfs_file_extent_disk_num_bytes(leaf,
3274 extent_offset = found_key.offset -
3275 btrfs_file_extent_offset(leaf, fi);
3277 /* FIXME blocksize != 4096 */
3278 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3279 if (extent_start != 0) {
3282 inode_sub_bytes(inode, num_dec);
3285 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3287 * we can't truncate inline items that have had
3291 btrfs_file_extent_compression(leaf, fi) == 0 &&
3292 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3293 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3294 u32 size = new_size - found_key.offset;
3296 if (root->ref_cows) {
3297 inode_sub_bytes(inode, item_end + 1 -
3301 btrfs_file_extent_calc_inline_size(size);
3302 ret = btrfs_truncate_item(trans, root, path,
3305 } else if (root->ref_cows) {
3306 inode_sub_bytes(inode, item_end + 1 -
3312 if (!pending_del_nr) {
3313 /* no pending yet, add ourselves */
3314 pending_del_slot = path->slots[0];
3316 } else if (pending_del_nr &&
3317 path->slots[0] + 1 == pending_del_slot) {
3318 /* hop on the pending chunk */
3320 pending_del_slot = path->slots[0];
3327 if (found_extent && root->ref_cows) {
3328 btrfs_set_path_blocking(path);
3329 ret = btrfs_free_extent(trans, root, extent_start,
3330 extent_num_bytes, 0,
3331 btrfs_header_owner(leaf),
3332 inode->i_ino, extent_offset);
3336 if (found_type == BTRFS_INODE_ITEM_KEY)
3339 if (path->slots[0] == 0 ||
3340 path->slots[0] != pending_del_slot) {
3341 if (root->ref_cows) {
3345 if (pending_del_nr) {
3346 ret = btrfs_del_items(trans, root, path,
3352 btrfs_release_path(root, path);
3359 if (pending_del_nr) {
3360 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3364 btrfs_free_path(path);
3369 * taken from block_truncate_page, but does cow as it zeros out
3370 * any bytes left in the last page in the file.
3372 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3374 struct inode *inode = mapping->host;
3375 struct btrfs_root *root = BTRFS_I(inode)->root;
3376 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3377 struct btrfs_ordered_extent *ordered;
3378 struct extent_state *cached_state = NULL;
3380 u32 blocksize = root->sectorsize;
3381 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3382 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3388 if ((offset & (blocksize - 1)) == 0)
3390 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3396 page = grab_cache_page(mapping, index);
3398 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3402 page_start = page_offset(page);
3403 page_end = page_start + PAGE_CACHE_SIZE - 1;
3405 if (!PageUptodate(page)) {
3406 ret = btrfs_readpage(NULL, page);
3408 if (page->mapping != mapping) {
3410 page_cache_release(page);
3413 if (!PageUptodate(page)) {
3418 wait_on_page_writeback(page);
3420 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3422 set_page_extent_mapped(page);
3424 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3426 unlock_extent_cached(io_tree, page_start, page_end,
3427 &cached_state, GFP_NOFS);
3429 page_cache_release(page);
3430 btrfs_start_ordered_extent(inode, ordered, 1);
3431 btrfs_put_ordered_extent(ordered);
3435 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3436 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3437 0, 0, &cached_state, GFP_NOFS);
3439 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3442 unlock_extent_cached(io_tree, page_start, page_end,
3443 &cached_state, GFP_NOFS);
3448 if (offset != PAGE_CACHE_SIZE) {
3450 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3451 flush_dcache_page(page);
3454 ClearPageChecked(page);
3455 set_page_dirty(page);
3456 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3461 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3463 page_cache_release(page);
3468 int btrfs_cont_expand(struct inode *inode, loff_t size)
3470 struct btrfs_trans_handle *trans;
3471 struct btrfs_root *root = BTRFS_I(inode)->root;
3472 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3473 struct extent_map *em = NULL;
3474 struct extent_state *cached_state = NULL;
3475 u64 mask = root->sectorsize - 1;
3476 u64 hole_start = (inode->i_size + mask) & ~mask;
3477 u64 block_end = (size + mask) & ~mask;
3483 if (size <= hole_start)
3487 struct btrfs_ordered_extent *ordered;
3488 btrfs_wait_ordered_range(inode, hole_start,
3489 block_end - hole_start);
3490 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3491 &cached_state, GFP_NOFS);
3492 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3495 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3496 &cached_state, GFP_NOFS);
3497 btrfs_put_ordered_extent(ordered);
3500 cur_offset = hole_start;
3502 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3503 block_end - cur_offset, 0);
3504 BUG_ON(IS_ERR(em) || !em);
3505 last_byte = min(extent_map_end(em), block_end);
3506 last_byte = (last_byte + mask) & ~mask;
3507 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3509 hole_size = last_byte - cur_offset;
3511 trans = btrfs_start_transaction(root, 2);
3512 if (IS_ERR(trans)) {
3513 err = PTR_ERR(trans);
3516 btrfs_set_trans_block_group(trans, inode);
3518 err = btrfs_drop_extents(trans, inode, cur_offset,
3519 cur_offset + hole_size,
3523 err = btrfs_insert_file_extent(trans, root,
3524 inode->i_ino, cur_offset, 0,
3525 0, hole_size, 0, hole_size,
3529 btrfs_drop_extent_cache(inode, hole_start,
3532 btrfs_end_transaction(trans, root);
3534 free_extent_map(em);
3536 cur_offset = last_byte;
3537 if (cur_offset >= block_end)
3541 free_extent_map(em);
3542 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3547 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3549 struct btrfs_root *root = BTRFS_I(inode)->root;
3550 struct btrfs_trans_handle *trans;
3554 if (attr->ia_size == inode->i_size)
3557 if (attr->ia_size > inode->i_size) {
3558 unsigned long limit;
3559 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3560 if (attr->ia_size > inode->i_sb->s_maxbytes)
3562 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3563 send_sig(SIGXFSZ, current, 0);
3568 trans = btrfs_start_transaction(root, 5);
3570 return PTR_ERR(trans);
3572 btrfs_set_trans_block_group(trans, inode);
3574 ret = btrfs_orphan_add(trans, inode);
3577 nr = trans->blocks_used;
3578 btrfs_end_transaction(trans, root);
3579 btrfs_btree_balance_dirty(root, nr);
3581 if (attr->ia_size > inode->i_size) {
3582 ret = btrfs_cont_expand(inode, attr->ia_size);
3584 btrfs_truncate(inode);
3588 i_size_write(inode, attr->ia_size);
3589 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3591 trans = btrfs_start_transaction(root, 0);
3592 BUG_ON(IS_ERR(trans));
3593 btrfs_set_trans_block_group(trans, inode);
3594 trans->block_rsv = root->orphan_block_rsv;
3595 BUG_ON(!trans->block_rsv);
3597 ret = btrfs_update_inode(trans, root, inode);
3599 if (inode->i_nlink > 0) {
3600 ret = btrfs_orphan_del(trans, inode);
3603 nr = trans->blocks_used;
3604 btrfs_end_transaction(trans, root);
3605 btrfs_btree_balance_dirty(root, nr);
3610 * We're truncating a file that used to have good data down to
3611 * zero. Make sure it gets into the ordered flush list so that
3612 * any new writes get down to disk quickly.
3614 if (attr->ia_size == 0)
3615 BTRFS_I(inode)->ordered_data_close = 1;
3617 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3618 ret = vmtruncate(inode, attr->ia_size);
3624 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3626 struct inode *inode = dentry->d_inode;
3629 err = inode_change_ok(inode, attr);
3633 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3634 err = btrfs_setattr_size(inode, attr);
3638 attr->ia_valid &= ~ATTR_SIZE;
3641 err = inode_setattr(inode, attr);
3643 if (!err && ((attr->ia_valid & ATTR_MODE)))
3644 err = btrfs_acl_chmod(inode);
3648 void btrfs_delete_inode(struct inode *inode)
3650 struct btrfs_trans_handle *trans;
3651 struct btrfs_root *root = BTRFS_I(inode)->root;
3655 truncate_inode_pages(&inode->i_data, 0);
3656 if (is_bad_inode(inode)) {
3657 btrfs_orphan_del(NULL, inode);
3660 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3662 if (root->fs_info->log_root_recovering) {
3663 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3667 if (inode->i_nlink > 0) {
3668 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3672 btrfs_i_size_write(inode, 0);
3675 trans = btrfs_start_transaction(root, 0);
3676 BUG_ON(IS_ERR(trans));
3677 btrfs_set_trans_block_group(trans, inode);
3678 trans->block_rsv = root->orphan_block_rsv;
3680 ret = btrfs_block_rsv_check(trans, root,
3681 root->orphan_block_rsv, 0, 5);
3683 BUG_ON(ret != -EAGAIN);
3684 ret = btrfs_commit_transaction(trans, root);
3689 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3693 nr = trans->blocks_used;
3694 btrfs_end_transaction(trans, root);
3696 btrfs_btree_balance_dirty(root, nr);
3701 ret = btrfs_orphan_del(trans, inode);
3705 nr = trans->blocks_used;
3706 btrfs_end_transaction(trans, root);
3707 btrfs_btree_balance_dirty(root, nr);
3714 * this returns the key found in the dir entry in the location pointer.
3715 * If no dir entries were found, location->objectid is 0.
3717 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3718 struct btrfs_key *location)
3720 const char *name = dentry->d_name.name;
3721 int namelen = dentry->d_name.len;
3722 struct btrfs_dir_item *di;
3723 struct btrfs_path *path;
3724 struct btrfs_root *root = BTRFS_I(dir)->root;
3727 path = btrfs_alloc_path();
3730 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3735 if (!di || IS_ERR(di))
3738 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3740 btrfs_free_path(path);
3743 location->objectid = 0;
3748 * when we hit a tree root in a directory, the btrfs part of the inode
3749 * needs to be changed to reflect the root directory of the tree root. This
3750 * is kind of like crossing a mount point.
3752 static int fixup_tree_root_location(struct btrfs_root *root,
3754 struct dentry *dentry,
3755 struct btrfs_key *location,
3756 struct btrfs_root **sub_root)
3758 struct btrfs_path *path;
3759 struct btrfs_root *new_root;
3760 struct btrfs_root_ref *ref;
3761 struct extent_buffer *leaf;
3765 path = btrfs_alloc_path();
3772 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3773 BTRFS_I(dir)->root->root_key.objectid,
3774 location->objectid);
3781 leaf = path->nodes[0];
3782 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3783 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3784 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3787 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3788 (unsigned long)(ref + 1),
3789 dentry->d_name.len);
3793 btrfs_release_path(root->fs_info->tree_root, path);
3795 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3796 if (IS_ERR(new_root)) {
3797 err = PTR_ERR(new_root);
3801 if (btrfs_root_refs(&new_root->root_item) == 0) {
3806 *sub_root = new_root;
3807 location->objectid = btrfs_root_dirid(&new_root->root_item);
3808 location->type = BTRFS_INODE_ITEM_KEY;
3809 location->offset = 0;
3812 btrfs_free_path(path);
3816 static void inode_tree_add(struct inode *inode)
3818 struct btrfs_root *root = BTRFS_I(inode)->root;
3819 struct btrfs_inode *entry;
3821 struct rb_node *parent;
3823 p = &root->inode_tree.rb_node;
3826 if (hlist_unhashed(&inode->i_hash))
3829 spin_lock(&root->inode_lock);
3832 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3834 if (inode->i_ino < entry->vfs_inode.i_ino)
3835 p = &parent->rb_left;
3836 else if (inode->i_ino > entry->vfs_inode.i_ino)
3837 p = &parent->rb_right;
3839 WARN_ON(!(entry->vfs_inode.i_state &
3840 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3841 rb_erase(parent, &root->inode_tree);
3842 RB_CLEAR_NODE(parent);
3843 spin_unlock(&root->inode_lock);
3847 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3848 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3849 spin_unlock(&root->inode_lock);
3852 static void inode_tree_del(struct inode *inode)
3854 struct btrfs_root *root = BTRFS_I(inode)->root;
3857 spin_lock(&root->inode_lock);
3858 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3859 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3860 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3861 empty = RB_EMPTY_ROOT(&root->inode_tree);
3863 spin_unlock(&root->inode_lock);
3865 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3866 synchronize_srcu(&root->fs_info->subvol_srcu);
3867 spin_lock(&root->inode_lock);
3868 empty = RB_EMPTY_ROOT(&root->inode_tree);
3869 spin_unlock(&root->inode_lock);
3871 btrfs_add_dead_root(root);
3875 int btrfs_invalidate_inodes(struct btrfs_root *root)
3877 struct rb_node *node;
3878 struct rb_node *prev;
3879 struct btrfs_inode *entry;
3880 struct inode *inode;
3883 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3885 spin_lock(&root->inode_lock);
3887 node = root->inode_tree.rb_node;
3891 entry = rb_entry(node, struct btrfs_inode, rb_node);
3893 if (objectid < entry->vfs_inode.i_ino)
3894 node = node->rb_left;
3895 else if (objectid > entry->vfs_inode.i_ino)
3896 node = node->rb_right;
3902 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3903 if (objectid <= entry->vfs_inode.i_ino) {
3907 prev = rb_next(prev);
3911 entry = rb_entry(node, struct btrfs_inode, rb_node);
3912 objectid = entry->vfs_inode.i_ino + 1;
3913 inode = igrab(&entry->vfs_inode);
3915 spin_unlock(&root->inode_lock);
3916 if (atomic_read(&inode->i_count) > 1)
3917 d_prune_aliases(inode);
3919 * btrfs_drop_inode will remove it from
3920 * the inode cache when its usage count
3925 spin_lock(&root->inode_lock);
3929 if (cond_resched_lock(&root->inode_lock))
3932 node = rb_next(node);
3934 spin_unlock(&root->inode_lock);
3938 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3940 struct btrfs_iget_args *args = p;
3941 inode->i_ino = args->ino;
3942 BTRFS_I(inode)->root = args->root;
3943 btrfs_set_inode_space_info(args->root, inode);
3947 static int btrfs_find_actor(struct inode *inode, void *opaque)
3949 struct btrfs_iget_args *args = opaque;
3950 return args->ino == inode->i_ino &&
3951 args->root == BTRFS_I(inode)->root;
3954 static struct inode *btrfs_iget_locked(struct super_block *s,
3956 struct btrfs_root *root)
3958 struct inode *inode;
3959 struct btrfs_iget_args args;
3960 args.ino = objectid;
3963 inode = iget5_locked(s, objectid, btrfs_find_actor,
3964 btrfs_init_locked_inode,
3969 /* Get an inode object given its location and corresponding root.
3970 * Returns in *is_new if the inode was read from disk
3972 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3973 struct btrfs_root *root, int *new)
3975 struct inode *inode;
3977 inode = btrfs_iget_locked(s, location->objectid, root);
3979 return ERR_PTR(-ENOMEM);
3981 if (inode->i_state & I_NEW) {
3982 BTRFS_I(inode)->root = root;
3983 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3984 btrfs_read_locked_inode(inode);
3986 inode_tree_add(inode);
3987 unlock_new_inode(inode);
3995 static struct inode *new_simple_dir(struct super_block *s,
3996 struct btrfs_key *key,
3997 struct btrfs_root *root)
3999 struct inode *inode = new_inode(s);
4002 return ERR_PTR(-ENOMEM);
4004 BTRFS_I(inode)->root = root;
4005 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4006 BTRFS_I(inode)->dummy_inode = 1;
4008 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4009 inode->i_op = &simple_dir_inode_operations;
4010 inode->i_fop = &simple_dir_operations;
4011 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4012 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4017 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4019 struct inode *inode;
4020 struct btrfs_root *root = BTRFS_I(dir)->root;
4021 struct btrfs_root *sub_root = root;
4022 struct btrfs_key location;
4026 dentry->d_op = &btrfs_dentry_operations;
4028 if (dentry->d_name.len > BTRFS_NAME_LEN)
4029 return ERR_PTR(-ENAMETOOLONG);
4031 ret = btrfs_inode_by_name(dir, dentry, &location);
4034 return ERR_PTR(ret);
4036 if (location.objectid == 0)
4039 if (location.type == BTRFS_INODE_ITEM_KEY) {
4040 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4044 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4046 index = srcu_read_lock(&root->fs_info->subvol_srcu);
4047 ret = fixup_tree_root_location(root, dir, dentry,
4048 &location, &sub_root);
4051 inode = ERR_PTR(ret);
4053 inode = new_simple_dir(dir->i_sb, &location, sub_root);
4055 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4057 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4059 if (root != sub_root) {
4060 down_read(&root->fs_info->cleanup_work_sem);
4061 if (!(inode->i_sb->s_flags & MS_RDONLY))
4062 btrfs_orphan_cleanup(sub_root);
4063 up_read(&root->fs_info->cleanup_work_sem);
4069 static int btrfs_dentry_delete(struct dentry *dentry)
4071 struct btrfs_root *root;
4073 if (!dentry->d_inode && !IS_ROOT(dentry))
4074 dentry = dentry->d_parent;
4076 if (dentry->d_inode) {
4077 root = BTRFS_I(dentry->d_inode)->root;
4078 if (btrfs_root_refs(&root->root_item) == 0)
4084 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4085 struct nameidata *nd)
4087 struct inode *inode;
4089 inode = btrfs_lookup_dentry(dir, dentry);
4091 return ERR_CAST(inode);
4093 return d_splice_alias(inode, dentry);
4096 static unsigned char btrfs_filetype_table[] = {
4097 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4100 static int btrfs_real_readdir(struct file *filp, void *dirent,
4103 struct inode *inode = filp->f_dentry->d_inode;
4104 struct btrfs_root *root = BTRFS_I(inode)->root;
4105 struct btrfs_item *item;
4106 struct btrfs_dir_item *di;
4107 struct btrfs_key key;
4108 struct btrfs_key found_key;
4109 struct btrfs_path *path;
4112 struct extent_buffer *leaf;
4115 unsigned char d_type;
4120 int key_type = BTRFS_DIR_INDEX_KEY;
4125 /* FIXME, use a real flag for deciding about the key type */
4126 if (root->fs_info->tree_root == root)
4127 key_type = BTRFS_DIR_ITEM_KEY;
4129 /* special case for "." */
4130 if (filp->f_pos == 0) {
4131 over = filldir(dirent, ".", 1,
4138 /* special case for .., just use the back ref */
4139 if (filp->f_pos == 1) {
4140 u64 pino = parent_ino(filp->f_path.dentry);
4141 over = filldir(dirent, "..", 2,
4147 path = btrfs_alloc_path();
4150 btrfs_set_key_type(&key, key_type);
4151 key.offset = filp->f_pos;
4152 key.objectid = inode->i_ino;
4154 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4160 leaf = path->nodes[0];
4161 nritems = btrfs_header_nritems(leaf);
4162 slot = path->slots[0];
4163 if (advance || slot >= nritems) {
4164 if (slot >= nritems - 1) {
4165 ret = btrfs_next_leaf(root, path);
4168 leaf = path->nodes[0];
4169 nritems = btrfs_header_nritems(leaf);
4170 slot = path->slots[0];
4178 item = btrfs_item_nr(leaf, slot);
4179 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4181 if (found_key.objectid != key.objectid)
4183 if (btrfs_key_type(&found_key) != key_type)
4185 if (found_key.offset < filp->f_pos)
4188 filp->f_pos = found_key.offset;
4190 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4192 di_total = btrfs_item_size(leaf, item);
4194 while (di_cur < di_total) {
4195 struct btrfs_key location;
4197 name_len = btrfs_dir_name_len(leaf, di);
4198 if (name_len <= sizeof(tmp_name)) {
4199 name_ptr = tmp_name;
4201 name_ptr = kmalloc(name_len, GFP_NOFS);
4207 read_extent_buffer(leaf, name_ptr,
4208 (unsigned long)(di + 1), name_len);
4210 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4211 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4213 /* is this a reference to our own snapshot? If so
4216 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4217 location.objectid == root->root_key.objectid) {
4221 over = filldir(dirent, name_ptr, name_len,
4222 found_key.offset, location.objectid,
4226 if (name_ptr != tmp_name)
4231 di_len = btrfs_dir_name_len(leaf, di) +
4232 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4234 di = (struct btrfs_dir_item *)((char *)di + di_len);
4238 /* Reached end of directory/root. Bump pos past the last item. */
4239 if (key_type == BTRFS_DIR_INDEX_KEY)
4241 * 32-bit glibc will use getdents64, but then strtol -
4242 * so the last number we can serve is this.
4244 filp->f_pos = 0x7fffffff;
4250 btrfs_free_path(path);
4254 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4256 struct btrfs_root *root = BTRFS_I(inode)->root;
4257 struct btrfs_trans_handle *trans;
4260 if (BTRFS_I(inode)->dummy_inode)
4263 if (wbc->sync_mode == WB_SYNC_ALL) {
4264 trans = btrfs_join_transaction(root, 1);
4265 btrfs_set_trans_block_group(trans, inode);
4266 ret = btrfs_commit_transaction(trans, root);
4272 * This is somewhat expensive, updating the tree every time the
4273 * inode changes. But, it is most likely to find the inode in cache.
4274 * FIXME, needs more benchmarking...there are no reasons other than performance
4275 * to keep or drop this code.
4277 void btrfs_dirty_inode(struct inode *inode)
4279 struct btrfs_root *root = BTRFS_I(inode)->root;
4280 struct btrfs_trans_handle *trans;
4283 if (BTRFS_I(inode)->dummy_inode)
4286 trans = btrfs_join_transaction(root, 1);
4287 btrfs_set_trans_block_group(trans, inode);
4289 ret = btrfs_update_inode(trans, root, inode);
4291 printk(KERN_ERR"btrfs: fail to dirty inode %lu error %d\n",
4294 btrfs_end_transaction(trans, root);
4298 * find the highest existing sequence number in a directory
4299 * and then set the in-memory index_cnt variable to reflect
4300 * free sequence numbers
4302 static int btrfs_set_inode_index_count(struct inode *inode)
4304 struct btrfs_root *root = BTRFS_I(inode)->root;
4305 struct btrfs_key key, found_key;
4306 struct btrfs_path *path;
4307 struct extent_buffer *leaf;
4310 key.objectid = inode->i_ino;
4311 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4312 key.offset = (u64)-1;
4314 path = btrfs_alloc_path();
4318 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4321 /* FIXME: we should be able to handle this */
4327 * MAGIC NUMBER EXPLANATION:
4328 * since we search a directory based on f_pos we have to start at 2
4329 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4330 * else has to start at 2
4332 if (path->slots[0] == 0) {
4333 BTRFS_I(inode)->index_cnt = 2;
4339 leaf = path->nodes[0];
4340 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4342 if (found_key.objectid != inode->i_ino ||
4343 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4344 BTRFS_I(inode)->index_cnt = 2;
4348 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4350 btrfs_free_path(path);
4355 * helper to find a free sequence number in a given directory. This current
4356 * code is very simple, later versions will do smarter things in the btree
4358 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4362 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4363 ret = btrfs_set_inode_index_count(dir);
4368 *index = BTRFS_I(dir)->index_cnt;
4369 BTRFS_I(dir)->index_cnt++;
4374 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4375 struct btrfs_root *root,
4377 const char *name, int name_len,
4378 u64 ref_objectid, u64 objectid,
4379 u64 alloc_hint, int mode, u64 *index)
4381 struct inode *inode;
4382 struct btrfs_inode_item *inode_item;
4383 struct btrfs_key *location;
4384 struct btrfs_path *path;
4385 struct btrfs_inode_ref *ref;
4386 struct btrfs_key key[2];
4392 path = btrfs_alloc_path();
4395 inode = new_inode(root->fs_info->sb);
4397 return ERR_PTR(-ENOMEM);
4400 ret = btrfs_set_inode_index(dir, index);
4403 return ERR_PTR(ret);
4407 * index_cnt is ignored for everything but a dir,
4408 * btrfs_get_inode_index_count has an explanation for the magic
4411 BTRFS_I(inode)->index_cnt = 2;
4412 BTRFS_I(inode)->root = root;
4413 BTRFS_I(inode)->generation = trans->transid;
4414 btrfs_set_inode_space_info(root, inode);
4420 BTRFS_I(inode)->block_group =
4421 btrfs_find_block_group(root, 0, alloc_hint, owner);
4423 key[0].objectid = objectid;
4424 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4427 key[1].objectid = objectid;
4428 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4429 key[1].offset = ref_objectid;
4431 sizes[0] = sizeof(struct btrfs_inode_item);
4432 sizes[1] = name_len + sizeof(*ref);
4434 path->leave_spinning = 1;
4435 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4439 inode->i_uid = current_fsuid();
4441 if (dir && (dir->i_mode & S_ISGID)) {
4442 inode->i_gid = dir->i_gid;
4446 inode->i_gid = current_fsgid();
4448 inode->i_mode = mode;
4449 inode->i_ino = objectid;
4450 inode_set_bytes(inode, 0);
4451 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4452 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4453 struct btrfs_inode_item);
4454 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4456 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4457 struct btrfs_inode_ref);
4458 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4459 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4460 ptr = (unsigned long)(ref + 1);
4461 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4463 btrfs_mark_buffer_dirty(path->nodes[0]);
4464 btrfs_free_path(path);
4466 location = &BTRFS_I(inode)->location;
4467 location->objectid = objectid;
4468 location->offset = 0;
4469 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4471 btrfs_inherit_iflags(inode, dir);
4473 if ((mode & S_IFREG)) {
4474 if (btrfs_test_opt(root, NODATASUM))
4475 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4476 if (btrfs_test_opt(root, NODATACOW))
4477 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4480 insert_inode_hash(inode);
4481 inode_tree_add(inode);
4485 BTRFS_I(dir)->index_cnt--;
4486 btrfs_free_path(path);
4488 return ERR_PTR(ret);
4491 static inline u8 btrfs_inode_type(struct inode *inode)
4493 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4497 * utility function to add 'inode' into 'parent_inode' with
4498 * a give name and a given sequence number.
4499 * if 'add_backref' is true, also insert a backref from the
4500 * inode to the parent directory.
4502 int btrfs_add_link(struct btrfs_trans_handle *trans,
4503 struct inode *parent_inode, struct inode *inode,
4504 const char *name, int name_len, int add_backref, u64 index)
4507 struct btrfs_key key;
4508 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4510 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4511 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4513 key.objectid = inode->i_ino;
4514 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4518 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4519 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4520 key.objectid, root->root_key.objectid,
4521 parent_inode->i_ino,
4522 index, name, name_len);
4523 } else if (add_backref) {
4524 ret = btrfs_insert_inode_ref(trans, root,
4525 name, name_len, inode->i_ino,
4526 parent_inode->i_ino, index);
4530 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4531 parent_inode->i_ino, &key,
4532 btrfs_inode_type(inode), index);
4535 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4537 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4538 ret = btrfs_update_inode(trans, root, parent_inode);
4543 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4544 struct dentry *dentry, struct inode *inode,
4545 int backref, u64 index)
4547 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4548 inode, dentry->d_name.name,
4549 dentry->d_name.len, backref, index);
4551 d_instantiate(dentry, inode);
4559 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4560 int mode, dev_t rdev)
4562 struct btrfs_trans_handle *trans;
4563 struct btrfs_root *root = BTRFS_I(dir)->root;
4564 struct inode *inode = NULL;
4568 unsigned long nr = 0;
4571 if (!new_valid_dev(rdev))
4574 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4579 * 2 for inode item and ref
4581 * 1 for xattr if selinux is on
4583 trans = btrfs_start_transaction(root, 5);
4585 return PTR_ERR(trans);
4587 btrfs_set_trans_block_group(trans, dir);
4589 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4591 dentry->d_parent->d_inode->i_ino, objectid,
4592 BTRFS_I(dir)->block_group, mode, &index);
4593 err = PTR_ERR(inode);
4597 err = btrfs_init_inode_security(trans, inode, dir);
4603 btrfs_set_trans_block_group(trans, inode);
4604 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4608 inode->i_op = &btrfs_special_inode_operations;
4609 init_special_inode(inode, inode->i_mode, rdev);
4610 btrfs_update_inode(trans, root, inode);
4612 btrfs_update_inode_block_group(trans, inode);
4613 btrfs_update_inode_block_group(trans, dir);
4615 nr = trans->blocks_used;
4616 btrfs_end_transaction_throttle(trans, root);
4617 btrfs_btree_balance_dirty(root, nr);
4619 inode_dec_link_count(inode);
4625 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4626 int mode, struct nameidata *nd)
4628 struct btrfs_trans_handle *trans;
4629 struct btrfs_root *root = BTRFS_I(dir)->root;
4630 struct inode *inode = NULL;
4633 unsigned long nr = 0;
4637 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4641 * 2 for inode item and ref
4643 * 1 for xattr if selinux is on
4645 trans = btrfs_start_transaction(root, 5);
4647 return PTR_ERR(trans);
4649 btrfs_set_trans_block_group(trans, dir);
4651 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4653 dentry->d_parent->d_inode->i_ino,
4654 objectid, BTRFS_I(dir)->block_group, mode,
4656 err = PTR_ERR(inode);
4660 err = btrfs_init_inode_security(trans, inode, dir);
4666 btrfs_set_trans_block_group(trans, inode);
4667 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4671 inode->i_mapping->a_ops = &btrfs_aops;
4672 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4673 inode->i_fop = &btrfs_file_operations;
4674 inode->i_op = &btrfs_file_inode_operations;
4675 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4677 btrfs_update_inode_block_group(trans, inode);
4678 btrfs_update_inode_block_group(trans, dir);
4680 nr = trans->blocks_used;
4681 btrfs_end_transaction_throttle(trans, root);
4683 inode_dec_link_count(inode);
4686 btrfs_btree_balance_dirty(root, nr);
4690 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4691 struct dentry *dentry)
4693 struct btrfs_trans_handle *trans;
4694 struct btrfs_root *root = BTRFS_I(dir)->root;
4695 struct inode *inode = old_dentry->d_inode;
4697 unsigned long nr = 0;
4701 if (inode->i_nlink == 0)
4704 /* do not allow sys_link's with other subvols of the same device */
4705 if (root->objectid != BTRFS_I(inode)->root->objectid)
4708 btrfs_inc_nlink(inode);
4710 err = btrfs_set_inode_index(dir, &index);
4715 * 1 item for inode ref
4716 * 2 items for dir items
4718 trans = btrfs_start_transaction(root, 3);
4719 if (IS_ERR(trans)) {
4720 err = PTR_ERR(trans);
4724 btrfs_set_trans_block_group(trans, dir);
4725 atomic_inc(&inode->i_count);
4727 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4732 btrfs_update_inode_block_group(trans, dir);
4733 err = btrfs_update_inode(trans, root, inode);
4735 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4738 nr = trans->blocks_used;
4739 btrfs_end_transaction_throttle(trans, root);
4742 inode_dec_link_count(inode);
4745 btrfs_btree_balance_dirty(root, nr);
4749 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4751 struct inode *inode = NULL;
4752 struct btrfs_trans_handle *trans;
4753 struct btrfs_root *root = BTRFS_I(dir)->root;
4755 int drop_on_err = 0;
4758 unsigned long nr = 1;
4760 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4765 * 2 items for inode and ref
4766 * 2 items for dir items
4767 * 1 for xattr if selinux is on
4769 trans = btrfs_start_transaction(root, 5);
4771 return PTR_ERR(trans);
4772 btrfs_set_trans_block_group(trans, dir);
4774 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4776 dentry->d_parent->d_inode->i_ino, objectid,
4777 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4779 if (IS_ERR(inode)) {
4780 err = PTR_ERR(inode);
4786 err = btrfs_init_inode_security(trans, inode, dir);
4790 inode->i_op = &btrfs_dir_inode_operations;
4791 inode->i_fop = &btrfs_dir_file_operations;
4792 btrfs_set_trans_block_group(trans, inode);
4794 btrfs_i_size_write(inode, 0);
4795 err = btrfs_update_inode(trans, root, inode);
4799 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4800 inode, dentry->d_name.name,
4801 dentry->d_name.len, 0, index);
4805 d_instantiate(dentry, inode);
4807 btrfs_update_inode_block_group(trans, inode);
4808 btrfs_update_inode_block_group(trans, dir);
4811 nr = trans->blocks_used;
4812 btrfs_end_transaction_throttle(trans, root);
4815 btrfs_btree_balance_dirty(root, nr);
4819 /* helper for btfs_get_extent. Given an existing extent in the tree,
4820 * and an extent that you want to insert, deal with overlap and insert
4821 * the new extent into the tree.
4823 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4824 struct extent_map *existing,
4825 struct extent_map *em,
4826 u64 map_start, u64 map_len)
4830 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4831 start_diff = map_start - em->start;
4832 em->start = map_start;
4834 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4835 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4836 em->block_start += start_diff;
4837 em->block_len -= start_diff;
4839 return add_extent_mapping(em_tree, em);
4842 static noinline int uncompress_inline(struct btrfs_path *path,
4843 struct inode *inode, struct page *page,
4844 size_t pg_offset, u64 extent_offset,
4845 struct btrfs_file_extent_item *item)
4848 struct extent_buffer *leaf = path->nodes[0];
4851 unsigned long inline_size;
4854 WARN_ON(pg_offset != 0);
4855 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4856 inline_size = btrfs_file_extent_inline_item_len(leaf,
4857 btrfs_item_nr(leaf, path->slots[0]));
4858 tmp = kmalloc(inline_size, GFP_NOFS);
4859 ptr = btrfs_file_extent_inline_start(item);
4861 read_extent_buffer(leaf, tmp, ptr, inline_size);
4863 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4864 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4865 inline_size, max_size);
4867 char *kaddr = kmap_atomic(page, KM_USER0);
4868 unsigned long copy_size = min_t(u64,
4869 PAGE_CACHE_SIZE - pg_offset,
4870 max_size - extent_offset);
4871 memset(kaddr + pg_offset, 0, copy_size);
4872 kunmap_atomic(kaddr, KM_USER0);
4879 * a bit scary, this does extent mapping from logical file offset to the disk.
4880 * the ugly parts come from merging extents from the disk with the in-ram
4881 * representation. This gets more complex because of the data=ordered code,
4882 * where the in-ram extents might be locked pending data=ordered completion.
4884 * This also copies inline extents directly into the page.
4887 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4888 size_t pg_offset, u64 start, u64 len,
4894 u64 extent_start = 0;
4896 u64 objectid = inode->i_ino;
4898 struct btrfs_path *path = NULL;
4899 struct btrfs_root *root = BTRFS_I(inode)->root;
4900 struct btrfs_file_extent_item *item;
4901 struct extent_buffer *leaf;
4902 struct btrfs_key found_key;
4903 struct extent_map *em = NULL;
4904 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4905 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4906 struct btrfs_trans_handle *trans = NULL;
4910 read_lock(&em_tree->lock);
4911 em = lookup_extent_mapping(em_tree, start, len);
4913 em->bdev = root->fs_info->fs_devices->latest_bdev;
4914 read_unlock(&em_tree->lock);
4917 if (em->start > start || em->start + em->len <= start)
4918 free_extent_map(em);
4919 else if (em->block_start == EXTENT_MAP_INLINE && page)
4920 free_extent_map(em);
4924 em = alloc_extent_map(GFP_NOFS);
4929 em->bdev = root->fs_info->fs_devices->latest_bdev;
4930 em->start = EXTENT_MAP_HOLE;
4931 em->orig_start = EXTENT_MAP_HOLE;
4933 em->block_len = (u64)-1;
4936 path = btrfs_alloc_path();
4940 ret = btrfs_lookup_file_extent(trans, root, path,
4941 objectid, start, trans != NULL);
4948 if (path->slots[0] == 0)
4953 leaf = path->nodes[0];
4954 item = btrfs_item_ptr(leaf, path->slots[0],
4955 struct btrfs_file_extent_item);
4956 /* are we inside the extent that was found? */
4957 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4958 found_type = btrfs_key_type(&found_key);
4959 if (found_key.objectid != objectid ||
4960 found_type != BTRFS_EXTENT_DATA_KEY) {
4964 found_type = btrfs_file_extent_type(leaf, item);
4965 extent_start = found_key.offset;
4966 compressed = btrfs_file_extent_compression(leaf, item);
4967 if (found_type == BTRFS_FILE_EXTENT_REG ||
4968 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4969 extent_end = extent_start +
4970 btrfs_file_extent_num_bytes(leaf, item);
4971 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4973 size = btrfs_file_extent_inline_len(leaf, item);
4974 extent_end = (extent_start + size + root->sectorsize - 1) &
4975 ~((u64)root->sectorsize - 1);
4978 if (start >= extent_end) {
4980 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4981 ret = btrfs_next_leaf(root, path);
4988 leaf = path->nodes[0];
4990 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4991 if (found_key.objectid != objectid ||
4992 found_key.type != BTRFS_EXTENT_DATA_KEY)
4994 if (start + len <= found_key.offset)
4997 em->len = found_key.offset - start;
5001 if (found_type == BTRFS_FILE_EXTENT_REG ||
5002 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5003 em->start = extent_start;
5004 em->len = extent_end - extent_start;
5005 em->orig_start = extent_start -
5006 btrfs_file_extent_offset(leaf, item);
5007 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5009 em->block_start = EXTENT_MAP_HOLE;
5013 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5014 em->block_start = bytenr;
5015 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5018 bytenr += btrfs_file_extent_offset(leaf, item);
5019 em->block_start = bytenr;
5020 em->block_len = em->len;
5021 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5022 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5025 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5029 size_t extent_offset;
5032 em->block_start = EXTENT_MAP_INLINE;
5033 if (!page || create) {
5034 em->start = extent_start;
5035 em->len = extent_end - extent_start;
5039 size = btrfs_file_extent_inline_len(leaf, item);
5040 extent_offset = page_offset(page) + pg_offset - extent_start;
5041 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5042 size - extent_offset);
5043 em->start = extent_start + extent_offset;
5044 em->len = (copy_size + root->sectorsize - 1) &
5045 ~((u64)root->sectorsize - 1);
5046 em->orig_start = EXTENT_MAP_INLINE;
5048 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5049 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5050 if (create == 0 && !PageUptodate(page)) {
5051 if (btrfs_file_extent_compression(leaf, item) ==
5052 BTRFS_COMPRESS_ZLIB) {
5053 ret = uncompress_inline(path, inode, page,
5055 extent_offset, item);
5059 read_extent_buffer(leaf, map + pg_offset, ptr,
5061 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5062 memset(map + pg_offset + copy_size, 0,
5063 PAGE_CACHE_SIZE - pg_offset -
5068 flush_dcache_page(page);
5069 } else if (create && PageUptodate(page)) {
5073 free_extent_map(em);
5075 btrfs_release_path(root, path);
5076 trans = btrfs_join_transaction(root, 1);
5080 write_extent_buffer(leaf, map + pg_offset, ptr,
5083 btrfs_mark_buffer_dirty(leaf);
5085 set_extent_uptodate(io_tree, em->start,
5086 extent_map_end(em) - 1, GFP_NOFS);
5089 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5096 em->block_start = EXTENT_MAP_HOLE;
5097 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5099 btrfs_release_path(root, path);
5100 if (em->start > start || extent_map_end(em) <= start) {
5101 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5102 "[%llu %llu]\n", (unsigned long long)em->start,
5103 (unsigned long long)em->len,
5104 (unsigned long long)start,
5105 (unsigned long long)len);
5111 write_lock(&em_tree->lock);
5112 ret = add_extent_mapping(em_tree, em);
5113 /* it is possible that someone inserted the extent into the tree
5114 * while we had the lock dropped. It is also possible that
5115 * an overlapping map exists in the tree
5117 if (ret == -EEXIST) {
5118 struct extent_map *existing;
5122 existing = lookup_extent_mapping(em_tree, start, len);
5123 if (existing && (existing->start > start ||
5124 existing->start + existing->len <= start)) {
5125 free_extent_map(existing);
5129 existing = lookup_extent_mapping(em_tree, em->start,
5132 err = merge_extent_mapping(em_tree, existing,
5135 free_extent_map(existing);
5137 free_extent_map(em);
5142 free_extent_map(em);
5146 free_extent_map(em);
5151 write_unlock(&em_tree->lock);
5154 btrfs_free_path(path);
5156 ret = btrfs_end_transaction(trans, root);
5161 free_extent_map(em);
5162 return ERR_PTR(err);
5167 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
5168 const struct iovec *iov, loff_t offset,
5169 unsigned long nr_segs)
5174 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
5175 __u64 start, __u64 len)
5177 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
5180 int btrfs_readpage(struct file *file, struct page *page)
5182 struct extent_io_tree *tree;
5183 tree = &BTRFS_I(page->mapping->host)->io_tree;
5184 return extent_read_full_page(tree, page, btrfs_get_extent);
5187 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
5189 struct extent_io_tree *tree;
5192 if (current->flags & PF_MEMALLOC) {
5193 redirty_page_for_writepage(wbc, page);
5197 tree = &BTRFS_I(page->mapping->host)->io_tree;
5198 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
5201 int btrfs_writepages(struct address_space *mapping,
5202 struct writeback_control *wbc)
5204 struct extent_io_tree *tree;
5206 tree = &BTRFS_I(mapping->host)->io_tree;
5207 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
5211 btrfs_readpages(struct file *file, struct address_space *mapping,
5212 struct list_head *pages, unsigned nr_pages)
5214 struct extent_io_tree *tree;
5215 tree = &BTRFS_I(mapping->host)->io_tree;
5216 return extent_readpages(tree, mapping, pages, nr_pages,
5219 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5221 struct extent_io_tree *tree;
5222 struct extent_map_tree *map;
5225 tree = &BTRFS_I(page->mapping->host)->io_tree;
5226 map = &BTRFS_I(page->mapping->host)->extent_tree;
5227 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
5229 ClearPagePrivate(page);
5230 set_page_private(page, 0);
5231 page_cache_release(page);
5236 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5238 if (PageWriteback(page) || PageDirty(page))
5240 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5243 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5245 struct extent_io_tree *tree;
5246 struct btrfs_ordered_extent *ordered;
5247 struct extent_state *cached_state = NULL;
5248 u64 page_start = page_offset(page);
5249 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5253 * we have the page locked, so new writeback can't start,
5254 * and the dirty bit won't be cleared while we are here.
5256 * Wait for IO on this page so that we can safely clear
5257 * the PagePrivate2 bit and do ordered accounting
5259 wait_on_page_writeback(page);
5261 tree = &BTRFS_I(page->mapping->host)->io_tree;
5263 btrfs_releasepage(page, GFP_NOFS);
5266 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5268 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5272 * IO on this page will never be started, so we need
5273 * to account for any ordered extents now
5275 clear_extent_bit(tree, page_start, page_end,
5276 EXTENT_DIRTY | EXTENT_DELALLOC |
5277 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5278 &cached_state, GFP_NOFS);
5280 * whoever cleared the private bit is responsible
5281 * for the finish_ordered_io
5283 if (TestClearPagePrivate2(page)) {
5284 btrfs_finish_ordered_io(page->mapping->host,
5285 page_start, page_end);
5287 btrfs_put_ordered_extent(ordered);
5288 cached_state = NULL;
5289 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5292 clear_extent_bit(tree, page_start, page_end,
5293 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5294 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5295 __btrfs_releasepage(page, GFP_NOFS);
5297 ClearPageChecked(page);
5298 if (PagePrivate(page)) {
5299 ClearPagePrivate(page);
5300 set_page_private(page, 0);
5301 page_cache_release(page);
5306 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5307 * called from a page fault handler when a page is first dirtied. Hence we must
5308 * be careful to check for EOF conditions here. We set the page up correctly
5309 * for a written page which means we get ENOSPC checking when writing into
5310 * holes and correct delalloc and unwritten extent mapping on filesystems that
5311 * support these features.
5313 * We are not allowed to take the i_mutex here so we have to play games to
5314 * protect against truncate races as the page could now be beyond EOF. Because
5315 * vmtruncate() writes the inode size before removing pages, once we have the
5316 * page lock we can determine safely if the page is beyond EOF. If it is not
5317 * beyond EOF, then the page is guaranteed safe against truncation until we
5320 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5322 struct page *page = vmf->page;
5323 struct inode *inode = fdentry(vma->vm_file)->d_inode;
5324 struct btrfs_root *root = BTRFS_I(inode)->root;
5325 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5326 struct btrfs_ordered_extent *ordered;
5327 struct extent_state *cached_state = NULL;
5329 unsigned long zero_start;
5335 ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
5339 else /* -ENOSPC, -EIO, etc */
5340 ret = VM_FAULT_SIGBUS;
5344 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5347 size = i_size_read(inode);
5348 page_start = page_offset(page);
5349 page_end = page_start + PAGE_CACHE_SIZE - 1;
5351 if ((page->mapping != inode->i_mapping) ||
5352 (page_start >= size)) {
5353 /* page got truncated out from underneath us */
5356 wait_on_page_writeback(page);
5358 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
5360 set_page_extent_mapped(page);
5363 * we can't set the delalloc bits if there are pending ordered
5364 * extents. Drop our locks and wait for them to finish
5366 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5368 unlock_extent_cached(io_tree, page_start, page_end,
5369 &cached_state, GFP_NOFS);
5371 btrfs_start_ordered_extent(inode, ordered, 1);
5372 btrfs_put_ordered_extent(ordered);
5377 * XXX - page_mkwrite gets called every time the page is dirtied, even
5378 * if it was already dirty, so for space accounting reasons we need to
5379 * clear any delalloc bits for the range we are fixing to save. There
5380 * is probably a better way to do this, but for now keep consistent with
5381 * prepare_pages in the normal write path.
5383 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5384 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5385 0, 0, &cached_state, GFP_NOFS);
5387 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
5390 unlock_extent_cached(io_tree, page_start, page_end,
5391 &cached_state, GFP_NOFS);
5392 ret = VM_FAULT_SIGBUS;
5397 /* page is wholly or partially inside EOF */
5398 if (page_start + PAGE_CACHE_SIZE > size)
5399 zero_start = size & ~PAGE_CACHE_MASK;
5401 zero_start = PAGE_CACHE_SIZE;
5403 if (zero_start != PAGE_CACHE_SIZE) {
5405 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5406 flush_dcache_page(page);
5409 ClearPageChecked(page);
5410 set_page_dirty(page);
5411 SetPageUptodate(page);
5413 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5414 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5416 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
5420 return VM_FAULT_LOCKED;
5422 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
5427 static void btrfs_truncate(struct inode *inode)
5429 struct btrfs_root *root = BTRFS_I(inode)->root;
5431 struct btrfs_trans_handle *trans;
5433 u64 mask = root->sectorsize - 1;
5435 if (!S_ISREG(inode->i_mode)) {
5440 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5444 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5445 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5447 trans = btrfs_start_transaction(root, 0);
5448 BUG_ON(IS_ERR(trans));
5449 btrfs_set_trans_block_group(trans, inode);
5450 trans->block_rsv = root->orphan_block_rsv;
5453 * setattr is responsible for setting the ordered_data_close flag,
5454 * but that is only tested during the last file release. That
5455 * could happen well after the next commit, leaving a great big
5456 * window where new writes may get lost if someone chooses to write
5457 * to this file after truncating to zero
5459 * The inode doesn't have any dirty data here, and so if we commit
5460 * this is a noop. If someone immediately starts writing to the inode
5461 * it is very likely we'll catch some of their writes in this
5462 * transaction, and the commit will find this file on the ordered
5463 * data list with good things to send down.
5465 * This is a best effort solution, there is still a window where
5466 * using truncate to replace the contents of the file will
5467 * end up with a zero length file after a crash.
5469 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5470 btrfs_add_ordered_operation(trans, root, inode);
5474 trans = btrfs_start_transaction(root, 0);
5475 BUG_ON(IS_ERR(trans));
5476 btrfs_set_trans_block_group(trans, inode);
5477 trans->block_rsv = root->orphan_block_rsv;
5480 ret = btrfs_block_rsv_check(trans, root,
5481 root->orphan_block_rsv, 0, 5);
5483 BUG_ON(ret != -EAGAIN);
5484 ret = btrfs_commit_transaction(trans, root);
5490 ret = btrfs_truncate_inode_items(trans, root, inode,
5492 BTRFS_EXTENT_DATA_KEY);
5496 ret = btrfs_update_inode(trans, root, inode);
5499 nr = trans->blocks_used;
5500 btrfs_end_transaction(trans, root);
5502 btrfs_btree_balance_dirty(root, nr);
5505 if (ret == 0 && inode->i_nlink > 0) {
5506 ret = btrfs_orphan_del(trans, inode);
5510 ret = btrfs_update_inode(trans, root, inode);
5513 nr = trans->blocks_used;
5514 ret = btrfs_end_transaction_throttle(trans, root);
5516 btrfs_btree_balance_dirty(root, nr);
5520 * create a new subvolume directory/inode (helper for the ioctl).
5522 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5523 struct btrfs_root *new_root,
5524 u64 new_dirid, u64 alloc_hint)
5526 struct inode *inode;
5530 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5531 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5533 return PTR_ERR(inode);
5534 inode->i_op = &btrfs_dir_inode_operations;
5535 inode->i_fop = &btrfs_dir_file_operations;
5538 btrfs_i_size_write(inode, 0);
5540 err = btrfs_update_inode(trans, new_root, inode);
5547 /* helper function for file defrag and space balancing. This
5548 * forces readahead on a given range of bytes in an inode
5550 unsigned long btrfs_force_ra(struct address_space *mapping,
5551 struct file_ra_state *ra, struct file *file,
5552 pgoff_t offset, pgoff_t last_index)
5554 pgoff_t req_size = last_index - offset + 1;
5556 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5557 return offset + req_size;
5560 struct inode *btrfs_alloc_inode(struct super_block *sb)
5562 struct btrfs_inode *ei;
5563 struct inode *inode;
5565 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5570 ei->space_info = NULL;
5574 ei->last_sub_trans = 0;
5575 ei->logged_trans = 0;
5576 ei->delalloc_bytes = 0;
5577 ei->reserved_bytes = 0;
5578 ei->disk_i_size = 0;
5580 ei->index_cnt = (u64)-1;
5581 ei->last_unlink_trans = 0;
5583 spin_lock_init(&ei->accounting_lock);
5584 atomic_set(&ei->outstanding_extents, 0);
5585 ei->reserved_extents = 0;
5587 ei->ordered_data_close = 0;
5588 ei->orphan_meta_reserved = 0;
5589 ei->dummy_inode = 0;
5590 ei->force_compress = 0;
5592 inode = &ei->vfs_inode;
5593 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
5594 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
5595 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
5596 mutex_init(&ei->log_mutex);
5597 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5598 INIT_LIST_HEAD(&ei->i_orphan);
5599 INIT_LIST_HEAD(&ei->delalloc_inodes);
5600 INIT_LIST_HEAD(&ei->ordered_operations);
5601 RB_CLEAR_NODE(&ei->rb_node);
5606 void btrfs_destroy_inode(struct inode *inode)
5608 struct btrfs_ordered_extent *ordered;
5609 struct btrfs_root *root = BTRFS_I(inode)->root;
5611 WARN_ON(!list_empty(&inode->i_dentry));
5612 WARN_ON(inode->i_data.nrpages);
5613 WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
5614 WARN_ON(BTRFS_I(inode)->reserved_extents);
5617 * This can happen where we create an inode, but somebody else also
5618 * created the same inode and we need to destroy the one we already
5625 * Make sure we're properly removed from the ordered operation
5629 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5630 spin_lock(&root->fs_info->ordered_extent_lock);
5631 list_del_init(&BTRFS_I(inode)->ordered_operations);
5632 spin_unlock(&root->fs_info->ordered_extent_lock);
5635 spin_lock(&root->orphan_lock);
5636 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5637 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5639 list_del_init(&BTRFS_I(inode)->i_orphan);
5641 spin_unlock(&root->orphan_lock);
5644 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5648 printk(KERN_ERR "btrfs found ordered "
5649 "extent %llu %llu on inode cleanup\n",
5650 (unsigned long long)ordered->file_offset,
5651 (unsigned long long)ordered->len);
5652 btrfs_remove_ordered_extent(inode, ordered);
5653 btrfs_put_ordered_extent(ordered);
5654 btrfs_put_ordered_extent(ordered);
5657 inode_tree_del(inode);
5658 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5660 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5663 void btrfs_drop_inode(struct inode *inode)
5665 struct btrfs_root *root = BTRFS_I(inode)->root;
5666 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5667 generic_delete_inode(inode);
5669 generic_drop_inode(inode);
5672 static void init_once(void *foo)
5674 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5676 inode_init_once(&ei->vfs_inode);
5679 void btrfs_destroy_cachep(void)
5681 if (btrfs_inode_cachep)
5682 kmem_cache_destroy(btrfs_inode_cachep);
5683 if (btrfs_trans_handle_cachep)
5684 kmem_cache_destroy(btrfs_trans_handle_cachep);
5685 if (btrfs_transaction_cachep)
5686 kmem_cache_destroy(btrfs_transaction_cachep);
5687 if (btrfs_path_cachep)
5688 kmem_cache_destroy(btrfs_path_cachep);
5691 int btrfs_init_cachep(void)
5693 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5694 sizeof(struct btrfs_inode), 0,
5695 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5696 if (!btrfs_inode_cachep)
5699 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5700 sizeof(struct btrfs_trans_handle), 0,
5701 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5702 if (!btrfs_trans_handle_cachep)
5705 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5706 sizeof(struct btrfs_transaction), 0,
5707 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5708 if (!btrfs_transaction_cachep)
5711 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5712 sizeof(struct btrfs_path), 0,
5713 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5714 if (!btrfs_path_cachep)
5719 btrfs_destroy_cachep();
5723 static int btrfs_getattr(struct vfsmount *mnt,
5724 struct dentry *dentry, struct kstat *stat)
5726 struct inode *inode = dentry->d_inode;
5727 generic_fillattr(inode, stat);
5728 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5729 stat->blksize = PAGE_CACHE_SIZE;
5730 stat->blocks = (inode_get_bytes(inode) +
5731 BTRFS_I(inode)->delalloc_bytes) >> 9;
5735 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5736 struct inode *new_dir, struct dentry *new_dentry)
5738 struct btrfs_trans_handle *trans;
5739 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5740 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5741 struct inode *new_inode = new_dentry->d_inode;
5742 struct inode *old_inode = old_dentry->d_inode;
5743 struct timespec ctime = CURRENT_TIME;
5748 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5751 /* we only allow rename subvolume link between subvolumes */
5752 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5755 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5756 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5759 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5760 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5763 * we're using rename to replace one file with another.
5764 * and the replacement file is large. Start IO on it now so
5765 * we don't add too much work to the end of the transaction
5767 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5768 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5769 filemap_flush(old_inode->i_mapping);
5771 /* close the racy window with snapshot create/destroy ioctl */
5772 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5773 down_read(&root->fs_info->subvol_sem);
5775 * We want to reserve the absolute worst case amount of items. So if
5776 * both inodes are subvols and we need to unlink them then that would
5777 * require 4 item modifications, but if they are both normal inodes it
5778 * would require 5 item modifications, so we'll assume their normal
5779 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5780 * should cover the worst case number of items we'll modify.
5782 trans = btrfs_start_transaction(root, 20);
5784 return PTR_ERR(trans);
5786 btrfs_set_trans_block_group(trans, new_dir);
5789 btrfs_record_root_in_trans(trans, dest);
5791 ret = btrfs_set_inode_index(new_dir, &index);
5795 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5796 /* force full log commit if subvolume involved. */
5797 root->fs_info->last_trans_log_full_commit = trans->transid;
5799 ret = btrfs_insert_inode_ref(trans, dest,
5800 new_dentry->d_name.name,
5801 new_dentry->d_name.len,
5803 new_dir->i_ino, index);
5807 * this is an ugly little race, but the rename is required
5808 * to make sure that if we crash, the inode is either at the
5809 * old name or the new one. pinning the log transaction lets
5810 * us make sure we don't allow a log commit to come in after
5811 * we unlink the name but before we add the new name back in.
5813 btrfs_pin_log_trans(root);
5816 * make sure the inode gets flushed if it is replacing
5819 if (new_inode && new_inode->i_size &&
5820 old_inode && S_ISREG(old_inode->i_mode)) {
5821 btrfs_add_ordered_operation(trans, root, old_inode);
5824 old_dir->i_ctime = old_dir->i_mtime = ctime;
5825 new_dir->i_ctime = new_dir->i_mtime = ctime;
5826 old_inode->i_ctime = ctime;
5828 if (old_dentry->d_parent != new_dentry->d_parent)
5829 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5831 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5832 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5833 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5834 old_dentry->d_name.name,
5835 old_dentry->d_name.len);
5837 btrfs_inc_nlink(old_dentry->d_inode);
5838 ret = btrfs_unlink_inode(trans, root, old_dir,
5839 old_dentry->d_inode,
5840 old_dentry->d_name.name,
5841 old_dentry->d_name.len);
5846 new_inode->i_ctime = CURRENT_TIME;
5847 if (unlikely(new_inode->i_ino ==
5848 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5849 root_objectid = BTRFS_I(new_inode)->location.objectid;
5850 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5852 new_dentry->d_name.name,
5853 new_dentry->d_name.len);
5854 BUG_ON(new_inode->i_nlink == 0);
5856 ret = btrfs_unlink_inode(trans, dest, new_dir,
5857 new_dentry->d_inode,
5858 new_dentry->d_name.name,
5859 new_dentry->d_name.len);
5862 if (new_inode->i_nlink == 0) {
5863 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5868 ret = btrfs_add_link(trans, new_dir, old_inode,
5869 new_dentry->d_name.name,
5870 new_dentry->d_name.len, 0, index);
5873 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5874 btrfs_log_new_name(trans, old_inode, old_dir,
5875 new_dentry->d_parent);
5876 btrfs_end_log_trans(root);
5879 btrfs_end_transaction_throttle(trans, root);
5881 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5882 up_read(&root->fs_info->subvol_sem);
5888 * some fairly slow code that needs optimization. This walks the list
5889 * of all the inodes with pending delalloc and forces them to disk.
5891 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5893 struct list_head *head = &root->fs_info->delalloc_inodes;
5894 struct btrfs_inode *binode;
5895 struct inode *inode;
5897 if (root->fs_info->sb->s_flags & MS_RDONLY)
5900 spin_lock(&root->fs_info->delalloc_lock);
5901 while (!list_empty(head)) {
5902 binode = list_entry(head->next, struct btrfs_inode,
5904 inode = igrab(&binode->vfs_inode);
5906 list_del_init(&binode->delalloc_inodes);
5907 spin_unlock(&root->fs_info->delalloc_lock);
5909 filemap_flush(inode->i_mapping);
5911 btrfs_add_delayed_iput(inode);
5916 spin_lock(&root->fs_info->delalloc_lock);
5918 spin_unlock(&root->fs_info->delalloc_lock);
5920 /* the filemap_flush will queue IO into the worker threads, but
5921 * we have to make sure the IO is actually started and that
5922 * ordered extents get created before we return
5924 atomic_inc(&root->fs_info->async_submit_draining);
5925 while (atomic_read(&root->fs_info->nr_async_submits) ||
5926 atomic_read(&root->fs_info->async_delalloc_pages)) {
5927 wait_event(root->fs_info->async_submit_wait,
5928 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5929 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5931 atomic_dec(&root->fs_info->async_submit_draining);
5935 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput)
5937 struct btrfs_inode *binode;
5938 struct inode *inode = NULL;
5940 spin_lock(&root->fs_info->delalloc_lock);
5941 while (!list_empty(&root->fs_info->delalloc_inodes)) {
5942 binode = list_entry(root->fs_info->delalloc_inodes.next,
5943 struct btrfs_inode, delalloc_inodes);
5944 inode = igrab(&binode->vfs_inode);
5946 list_move_tail(&binode->delalloc_inodes,
5947 &root->fs_info->delalloc_inodes);
5951 list_del_init(&binode->delalloc_inodes);
5952 cond_resched_lock(&root->fs_info->delalloc_lock);
5954 spin_unlock(&root->fs_info->delalloc_lock);
5957 write_inode_now(inode, 0);
5959 btrfs_add_delayed_iput(inode);
5967 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5968 const char *symname)
5970 struct btrfs_trans_handle *trans;
5971 struct btrfs_root *root = BTRFS_I(dir)->root;
5972 struct btrfs_path *path;
5973 struct btrfs_key key;
5974 struct inode *inode = NULL;
5982 struct btrfs_file_extent_item *ei;
5983 struct extent_buffer *leaf;
5984 unsigned long nr = 0;
5986 name_len = strlen(symname) + 1;
5987 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5988 return -ENAMETOOLONG;
5990 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
5994 * 2 items for inode item and ref
5995 * 2 items for dir items
5996 * 1 item for xattr if selinux is on
5998 trans = btrfs_start_transaction(root, 5);
6000 return PTR_ERR(trans);
6002 btrfs_set_trans_block_group(trans, dir);
6004 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6006 dentry->d_parent->d_inode->i_ino, objectid,
6007 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
6009 err = PTR_ERR(inode);
6013 err = btrfs_init_inode_security(trans, inode, dir);
6019 btrfs_set_trans_block_group(trans, inode);
6020 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
6024 inode->i_mapping->a_ops = &btrfs_aops;
6025 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6026 inode->i_fop = &btrfs_file_operations;
6027 inode->i_op = &btrfs_file_inode_operations;
6028 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6030 btrfs_update_inode_block_group(trans, inode);
6031 btrfs_update_inode_block_group(trans, dir);
6035 path = btrfs_alloc_path();
6037 key.objectid = inode->i_ino;
6039 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
6040 datasize = btrfs_file_extent_calc_inline_size(name_len);
6041 err = btrfs_insert_empty_item(trans, root, path, &key,
6047 leaf = path->nodes[0];
6048 ei = btrfs_item_ptr(leaf, path->slots[0],
6049 struct btrfs_file_extent_item);
6050 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
6051 btrfs_set_file_extent_type(leaf, ei,
6052 BTRFS_FILE_EXTENT_INLINE);
6053 btrfs_set_file_extent_encryption(leaf, ei, 0);
6054 btrfs_set_file_extent_compression(leaf, ei, 0);
6055 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
6056 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
6058 ptr = btrfs_file_extent_inline_start(ei);
6059 write_extent_buffer(leaf, symname, ptr, name_len);
6060 btrfs_mark_buffer_dirty(leaf);
6061 btrfs_free_path(path);
6063 inode->i_op = &btrfs_symlink_inode_operations;
6064 inode->i_mapping->a_ops = &btrfs_symlink_aops;
6065 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6066 inode_set_bytes(inode, name_len);
6067 btrfs_i_size_write(inode, name_len - 1);
6068 err = btrfs_update_inode(trans, root, inode);
6073 nr = trans->blocks_used;
6074 btrfs_end_transaction_throttle(trans, root);
6076 inode_dec_link_count(inode);
6079 btrfs_btree_balance_dirty(root, nr);
6083 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
6084 u64 alloc_hint, int mode, loff_t actual_len)
6086 struct btrfs_trans_handle *trans;
6087 struct btrfs_root *root = BTRFS_I(inode)->root;
6088 struct btrfs_key ins;
6089 u64 cur_offset = start;
6090 u64 num_bytes = end - start;
6094 while (num_bytes > 0) {
6095 trans = btrfs_start_transaction(root, 3);
6096 if (IS_ERR(trans)) {
6097 ret = PTR_ERR(trans);
6101 ret = btrfs_reserve_extent(trans, root, num_bytes,
6102 root->sectorsize, 0, alloc_hint,
6105 btrfs_end_transaction(trans, root);
6109 ret = insert_reserved_file_extent(trans, inode,
6110 cur_offset, ins.objectid,
6111 ins.offset, ins.offset,
6112 ins.offset, 0, 0, 0,
6113 BTRFS_FILE_EXTENT_PREALLOC);
6115 btrfs_drop_extent_cache(inode, cur_offset,
6116 cur_offset + ins.offset -1, 0);
6118 num_bytes -= ins.offset;
6119 cur_offset += ins.offset;
6120 alloc_hint = ins.objectid + ins.offset;
6122 inode->i_ctime = CURRENT_TIME;
6123 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
6124 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
6125 (actual_len > inode->i_size) &&
6126 (cur_offset > inode->i_size)) {
6128 if (cur_offset > actual_len)
6129 i_size = actual_len;
6131 i_size = cur_offset;
6132 i_size_write(inode, i_size);
6133 btrfs_ordered_update_i_size(inode, i_size, NULL);
6136 ret = btrfs_update_inode(trans, root, inode);
6139 btrfs_end_transaction(trans, root);
6144 static long btrfs_fallocate(struct inode *inode, int mode,
6145 loff_t offset, loff_t len)
6147 struct extent_state *cached_state = NULL;
6154 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
6155 struct extent_map *em;
6158 alloc_start = offset & ~mask;
6159 alloc_end = (offset + len + mask) & ~mask;
6162 * wait for ordered IO before we have any locks. We'll loop again
6163 * below with the locks held.
6165 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
6167 mutex_lock(&inode->i_mutex);
6168 if (alloc_start > inode->i_size) {
6169 ret = btrfs_cont_expand(inode, alloc_start);
6174 ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
6178 locked_end = alloc_end - 1;
6180 struct btrfs_ordered_extent *ordered;
6182 /* the extent lock is ordered inside the running
6185 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
6186 locked_end, 0, &cached_state, GFP_NOFS);
6187 ordered = btrfs_lookup_first_ordered_extent(inode,
6190 ordered->file_offset + ordered->len > alloc_start &&
6191 ordered->file_offset < alloc_end) {
6192 btrfs_put_ordered_extent(ordered);
6193 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
6194 alloc_start, locked_end,
6195 &cached_state, GFP_NOFS);
6197 * we can't wait on the range with the transaction
6198 * running or with the extent lock held
6200 btrfs_wait_ordered_range(inode, alloc_start,
6201 alloc_end - alloc_start);
6204 btrfs_put_ordered_extent(ordered);
6209 cur_offset = alloc_start;
6211 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
6212 alloc_end - cur_offset, 0);
6213 BUG_ON(IS_ERR(em) || !em);
6214 last_byte = min(extent_map_end(em), alloc_end);
6215 last_byte = (last_byte + mask) & ~mask;
6216 if (em->block_start == EXTENT_MAP_HOLE ||
6217 (cur_offset >= inode->i_size &&
6218 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6219 ret = prealloc_file_range(inode,
6220 cur_offset, last_byte,
6221 alloc_hint, mode, offset+len);
6223 free_extent_map(em);
6227 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
6228 alloc_hint = em->block_start;
6229 free_extent_map(em);
6231 cur_offset = last_byte;
6232 if (cur_offset >= alloc_end) {
6237 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
6238 &cached_state, GFP_NOFS);
6240 btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
6242 mutex_unlock(&inode->i_mutex);
6246 static int btrfs_set_page_dirty(struct page *page)
6248 return __set_page_dirty_nobuffers(page);
6251 static int btrfs_permission(struct inode *inode, int mask)
6253 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
6255 return generic_permission(inode, mask, btrfs_check_acl);
6258 static const struct inode_operations btrfs_dir_inode_operations = {
6259 .getattr = btrfs_getattr,
6260 .lookup = btrfs_lookup,
6261 .create = btrfs_create,
6262 .unlink = btrfs_unlink,
6264 .mkdir = btrfs_mkdir,
6265 .rmdir = btrfs_rmdir,
6266 .rename = btrfs_rename,
6267 .symlink = btrfs_symlink,
6268 .setattr = btrfs_setattr,
6269 .mknod = btrfs_mknod,
6270 .setxattr = btrfs_setxattr,
6271 .getxattr = btrfs_getxattr,
6272 .listxattr = btrfs_listxattr,
6273 .removexattr = btrfs_removexattr,
6274 .permission = btrfs_permission,
6276 static const struct inode_operations btrfs_dir_ro_inode_operations = {
6277 .lookup = btrfs_lookup,
6278 .permission = btrfs_permission,
6281 static const struct file_operations btrfs_dir_file_operations = {
6282 .llseek = generic_file_llseek,
6283 .read = generic_read_dir,
6284 .readdir = btrfs_real_readdir,
6285 .unlocked_ioctl = btrfs_ioctl,
6286 #ifdef CONFIG_COMPAT
6287 .compat_ioctl = btrfs_ioctl,
6289 .release = btrfs_release_file,
6290 .fsync = btrfs_sync_file,
6293 static struct extent_io_ops btrfs_extent_io_ops = {
6294 .fill_delalloc = run_delalloc_range,
6295 .submit_bio_hook = btrfs_submit_bio_hook,
6296 .merge_bio_hook = btrfs_merge_bio_hook,
6297 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6298 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6299 .writepage_start_hook = btrfs_writepage_start_hook,
6300 .readpage_io_failed_hook = btrfs_io_failed_hook,
6301 .set_bit_hook = btrfs_set_bit_hook,
6302 .clear_bit_hook = btrfs_clear_bit_hook,
6303 .merge_extent_hook = btrfs_merge_extent_hook,
6304 .split_extent_hook = btrfs_split_extent_hook,
6308 * btrfs doesn't support the bmap operation because swapfiles
6309 * use bmap to make a mapping of extents in the file. They assume
6310 * these extents won't change over the life of the file and they
6311 * use the bmap result to do IO directly to the drive.
6313 * the btrfs bmap call would return logical addresses that aren't
6314 * suitable for IO and they also will change frequently as COW
6315 * operations happen. So, swapfile + btrfs == corruption.
6317 * For now we're avoiding this by dropping bmap.
6319 static const struct address_space_operations btrfs_aops = {
6320 .readpage = btrfs_readpage,
6321 .writepage = btrfs_writepage,
6322 .writepages = btrfs_writepages,
6323 .readpages = btrfs_readpages,
6324 .sync_page = block_sync_page,
6325 .direct_IO = btrfs_direct_IO,
6326 .invalidatepage = btrfs_invalidatepage,
6327 .releasepage = btrfs_releasepage,
6328 .set_page_dirty = btrfs_set_page_dirty,
6329 .error_remove_page = generic_error_remove_page,
6332 static const struct address_space_operations btrfs_symlink_aops = {
6333 .readpage = btrfs_readpage,
6334 .writepage = btrfs_writepage,
6335 .invalidatepage = btrfs_invalidatepage,
6336 .releasepage = btrfs_releasepage,
6339 static const struct inode_operations btrfs_file_inode_operations = {
6340 .truncate = btrfs_truncate,
6341 .getattr = btrfs_getattr,
6342 .setattr = btrfs_setattr,
6343 .setxattr = btrfs_setxattr,
6344 .getxattr = btrfs_getxattr,
6345 .listxattr = btrfs_listxattr,
6346 .removexattr = btrfs_removexattr,
6347 .permission = btrfs_permission,
6348 .fallocate = btrfs_fallocate,
6349 .fiemap = btrfs_fiemap,
6351 static const struct inode_operations btrfs_special_inode_operations = {
6352 .getattr = btrfs_getattr,
6353 .setattr = btrfs_setattr,
6354 .permission = btrfs_permission,
6355 .setxattr = btrfs_setxattr,
6356 .getxattr = btrfs_getxattr,
6357 .listxattr = btrfs_listxattr,
6358 .removexattr = btrfs_removexattr,
6360 static const struct inode_operations btrfs_symlink_inode_operations = {
6361 .readlink = generic_readlink,
6362 .follow_link = page_follow_link_light,
6363 .put_link = page_put_link,
6364 .permission = btrfs_permission,
6365 .setxattr = btrfs_setxattr,
6366 .getxattr = btrfs_getxattr,
6367 .listxattr = btrfs_listxattr,
6368 .removexattr = btrfs_removexattr,
6371 const struct dentry_operations btrfs_dentry_operations = {
6372 .d_delete = btrfs_dentry_delete,