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>
42 #include "transaction.h"
43 #include "btrfs_inode.h"
45 #include "print-tree.h"
47 #include "ordered-data.h"
50 #include "compression.h"
53 struct btrfs_iget_args {
55 struct btrfs_root *root;
58 static const struct inode_operations btrfs_dir_inode_operations;
59 static const struct inode_operations btrfs_symlink_inode_operations;
60 static const struct inode_operations btrfs_dir_ro_inode_operations;
61 static const struct inode_operations btrfs_special_inode_operations;
62 static const struct inode_operations btrfs_file_inode_operations;
63 static const struct address_space_operations btrfs_aops;
64 static const struct address_space_operations btrfs_symlink_aops;
65 static const struct file_operations btrfs_dir_file_operations;
66 static struct extent_io_ops btrfs_extent_io_ops;
68 static struct kmem_cache *btrfs_inode_cachep;
69 struct kmem_cache *btrfs_trans_handle_cachep;
70 struct kmem_cache *btrfs_transaction_cachep;
71 struct kmem_cache *btrfs_path_cachep;
74 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
75 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
76 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
77 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
78 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
79 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
80 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
81 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
84 static void btrfs_truncate(struct inode *inode);
85 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
86 static noinline int cow_file_range(struct inode *inode,
87 struct page *locked_page,
88 u64 start, u64 end, int *page_started,
89 unsigned long *nr_written, int unlock);
91 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
92 struct inode *inode, struct inode *dir)
96 err = btrfs_init_acl(trans, inode, dir);
98 err = btrfs_xattr_security_init(trans, inode, dir);
103 * this does all the hard work for inserting an inline extent into
104 * the btree. The caller should have done a btrfs_drop_extents so that
105 * no overlapping inline items exist in the btree
107 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
108 struct btrfs_root *root, struct inode *inode,
109 u64 start, size_t size, size_t compressed_size,
110 struct page **compressed_pages)
112 struct btrfs_key key;
113 struct btrfs_path *path;
114 struct extent_buffer *leaf;
115 struct page *page = NULL;
118 struct btrfs_file_extent_item *ei;
121 size_t cur_size = size;
123 unsigned long offset;
124 int use_compress = 0;
126 if (compressed_size && compressed_pages) {
128 cur_size = compressed_size;
131 path = btrfs_alloc_path();
135 path->leave_spinning = 1;
136 btrfs_set_trans_block_group(trans, inode);
138 key.objectid = inode->i_ino;
140 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
141 datasize = btrfs_file_extent_calc_inline_size(cur_size);
143 inode_add_bytes(inode, size);
144 ret = btrfs_insert_empty_item(trans, root, path, &key,
151 leaf = path->nodes[0];
152 ei = btrfs_item_ptr(leaf, path->slots[0],
153 struct btrfs_file_extent_item);
154 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
155 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
156 btrfs_set_file_extent_encryption(leaf, ei, 0);
157 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
158 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
159 ptr = btrfs_file_extent_inline_start(ei);
164 while (compressed_size > 0) {
165 cpage = compressed_pages[i];
166 cur_size = min_t(unsigned long, compressed_size,
169 kaddr = kmap_atomic(cpage, KM_USER0);
170 write_extent_buffer(leaf, kaddr, ptr, cur_size);
171 kunmap_atomic(kaddr, KM_USER0);
175 compressed_size -= cur_size;
177 btrfs_set_file_extent_compression(leaf, ei,
178 BTRFS_COMPRESS_ZLIB);
180 page = find_get_page(inode->i_mapping,
181 start >> PAGE_CACHE_SHIFT);
182 btrfs_set_file_extent_compression(leaf, ei, 0);
183 kaddr = kmap_atomic(page, KM_USER0);
184 offset = start & (PAGE_CACHE_SIZE - 1);
185 write_extent_buffer(leaf, kaddr + offset, ptr, size);
186 kunmap_atomic(kaddr, KM_USER0);
187 page_cache_release(page);
189 btrfs_mark_buffer_dirty(leaf);
190 btrfs_free_path(path);
193 * we're an inline extent, so nobody can
194 * extend the file past i_size without locking
195 * a page we already have locked.
197 * We must do any isize and inode updates
198 * before we unlock the pages. Otherwise we
199 * could end up racing with unlink.
201 BTRFS_I(inode)->disk_i_size = inode->i_size;
202 btrfs_update_inode(trans, root, inode);
206 btrfs_free_path(path);
212 * conditionally insert an inline extent into the file. This
213 * does the checks required to make sure the data is small enough
214 * to fit as an inline extent.
216 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
217 struct btrfs_root *root,
218 struct inode *inode, u64 start, u64 end,
219 size_t compressed_size,
220 struct page **compressed_pages)
222 u64 isize = i_size_read(inode);
223 u64 actual_end = min(end + 1, isize);
224 u64 inline_len = actual_end - start;
225 u64 aligned_end = (end + root->sectorsize - 1) &
226 ~((u64)root->sectorsize - 1);
228 u64 data_len = inline_len;
232 data_len = compressed_size;
235 actual_end >= PAGE_CACHE_SIZE ||
236 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
238 (actual_end & (root->sectorsize - 1)) == 0) ||
240 data_len > root->fs_info->max_inline) {
244 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
248 if (isize > actual_end)
249 inline_len = min_t(u64, isize, actual_end);
250 ret = insert_inline_extent(trans, root, inode, start,
251 inline_len, compressed_size,
254 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
258 struct async_extent {
263 unsigned long nr_pages;
264 struct list_head list;
269 struct btrfs_root *root;
270 struct page *locked_page;
273 struct list_head extents;
274 struct btrfs_work work;
277 static noinline int add_async_extent(struct async_cow *cow,
278 u64 start, u64 ram_size,
281 unsigned long nr_pages)
283 struct async_extent *async_extent;
285 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
286 async_extent->start = start;
287 async_extent->ram_size = ram_size;
288 async_extent->compressed_size = compressed_size;
289 async_extent->pages = pages;
290 async_extent->nr_pages = nr_pages;
291 list_add_tail(&async_extent->list, &cow->extents);
296 * we create compressed extents in two phases. The first
297 * phase compresses a range of pages that have already been
298 * locked (both pages and state bits are locked).
300 * This is done inside an ordered work queue, and the compression
301 * is spread across many cpus. The actual IO submission is step
302 * two, and the ordered work queue takes care of making sure that
303 * happens in the same order things were put onto the queue by
304 * writepages and friends.
306 * If this code finds it can't get good compression, it puts an
307 * entry onto the work queue to write the uncompressed bytes. This
308 * makes sure that both compressed inodes and uncompressed inodes
309 * are written in the same order that pdflush sent them down.
311 static noinline int compress_file_range(struct inode *inode,
312 struct page *locked_page,
314 struct async_cow *async_cow,
317 struct btrfs_root *root = BTRFS_I(inode)->root;
318 struct btrfs_trans_handle *trans;
322 u64 blocksize = root->sectorsize;
324 u64 isize = i_size_read(inode);
326 struct page **pages = NULL;
327 unsigned long nr_pages;
328 unsigned long nr_pages_ret = 0;
329 unsigned long total_compressed = 0;
330 unsigned long total_in = 0;
331 unsigned long max_compressed = 128 * 1024;
332 unsigned long max_uncompressed = 128 * 1024;
338 actual_end = min_t(u64, isize, end + 1);
341 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
342 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
345 * we don't want to send crud past the end of i_size through
346 * compression, that's just a waste of CPU time. So, if the
347 * end of the file is before the start of our current
348 * requested range of bytes, we bail out to the uncompressed
349 * cleanup code that can deal with all of this.
351 * It isn't really the fastest way to fix things, but this is a
352 * very uncommon corner.
354 if (actual_end <= start)
355 goto cleanup_and_bail_uncompressed;
357 total_compressed = actual_end - start;
359 /* we want to make sure that amount of ram required to uncompress
360 * an extent is reasonable, so we limit the total size in ram
361 * of a compressed extent to 128k. This is a crucial number
362 * because it also controls how easily we can spread reads across
363 * cpus for decompression.
365 * We also want to make sure the amount of IO required to do
366 * a random read is reasonably small, so we limit the size of
367 * a compressed extent to 128k.
369 total_compressed = min(total_compressed, max_uncompressed);
370 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
371 num_bytes = max(blocksize, num_bytes);
372 disk_num_bytes = num_bytes;
377 * we do compression for mount -o compress and when the
378 * inode has not been flagged as nocompress. This flag can
379 * change at any time if we discover bad compression ratios.
381 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
382 btrfs_test_opt(root, COMPRESS)) {
384 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
386 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
387 total_compressed, pages,
388 nr_pages, &nr_pages_ret,
394 unsigned long offset = total_compressed &
395 (PAGE_CACHE_SIZE - 1);
396 struct page *page = pages[nr_pages_ret - 1];
399 /* zero the tail end of the last page, we might be
400 * sending it down to disk
403 kaddr = kmap_atomic(page, KM_USER0);
404 memset(kaddr + offset, 0,
405 PAGE_CACHE_SIZE - offset);
406 kunmap_atomic(kaddr, KM_USER0);
412 trans = btrfs_join_transaction(root, 1);
414 btrfs_set_trans_block_group(trans, inode);
416 /* lets try to make an inline extent */
417 if (ret || total_in < (actual_end - start)) {
418 /* we didn't compress the entire range, try
419 * to make an uncompressed inline extent.
421 ret = cow_file_range_inline(trans, root, inode,
422 start, end, 0, NULL);
424 /* try making a compressed inline extent */
425 ret = cow_file_range_inline(trans, root, inode,
427 total_compressed, pages);
431 * inline extent creation worked, we don't need
432 * to create any more async work items. Unlock
433 * and free up our temp pages.
435 extent_clear_unlock_delalloc(inode,
436 &BTRFS_I(inode)->io_tree,
438 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
439 EXTENT_CLEAR_DELALLOC |
440 EXTENT_CLEAR_ACCOUNTING |
441 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
443 btrfs_end_transaction(trans, root);
446 btrfs_end_transaction(trans, root);
451 * we aren't doing an inline extent round the compressed size
452 * up to a block size boundary so the allocator does sane
455 total_compressed = (total_compressed + blocksize - 1) &
459 * one last check to make sure the compression is really a
460 * win, compare the page count read with the blocks on disk
462 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
463 ~(PAGE_CACHE_SIZE - 1);
464 if (total_compressed >= total_in) {
467 disk_num_bytes = total_compressed;
468 num_bytes = total_in;
471 if (!will_compress && pages) {
473 * the compression code ran but failed to make things smaller,
474 * free any pages it allocated and our page pointer array
476 for (i = 0; i < nr_pages_ret; i++) {
477 WARN_ON(pages[i]->mapping);
478 page_cache_release(pages[i]);
482 total_compressed = 0;
485 /* flag the file so we don't compress in the future */
486 if (!btrfs_test_opt(root, FORCE_COMPRESS))
487 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
492 /* the async work queues will take care of doing actual
493 * allocation on disk for these compressed pages,
494 * and will submit them to the elevator.
496 add_async_extent(async_cow, start, num_bytes,
497 total_compressed, pages, nr_pages_ret);
499 if (start + num_bytes < end && start + num_bytes < actual_end) {
506 cleanup_and_bail_uncompressed:
508 * No compression, but we still need to write the pages in
509 * the file we've been given so far. redirty the locked
510 * page if it corresponds to our extent and set things up
511 * for the async work queue to run cow_file_range to do
512 * the normal delalloc dance
514 if (page_offset(locked_page) >= start &&
515 page_offset(locked_page) <= end) {
516 __set_page_dirty_nobuffers(locked_page);
517 /* unlocked later on in the async handlers */
519 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
527 for (i = 0; i < nr_pages_ret; i++) {
528 WARN_ON(pages[i]->mapping);
529 page_cache_release(pages[i]);
537 * phase two of compressed writeback. This is the ordered portion
538 * of the code, which only gets called in the order the work was
539 * queued. We walk all the async extents created by compress_file_range
540 * and send them down to the disk.
542 static noinline int submit_compressed_extents(struct inode *inode,
543 struct async_cow *async_cow)
545 struct async_extent *async_extent;
547 struct btrfs_trans_handle *trans;
548 struct btrfs_key ins;
549 struct extent_map *em;
550 struct btrfs_root *root = BTRFS_I(inode)->root;
551 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
552 struct extent_io_tree *io_tree;
555 if (list_empty(&async_cow->extents))
559 while (!list_empty(&async_cow->extents)) {
560 async_extent = list_entry(async_cow->extents.next,
561 struct async_extent, list);
562 list_del(&async_extent->list);
564 io_tree = &BTRFS_I(inode)->io_tree;
567 /* did the compression code fall back to uncompressed IO? */
568 if (!async_extent->pages) {
569 int page_started = 0;
570 unsigned long nr_written = 0;
572 lock_extent(io_tree, async_extent->start,
573 async_extent->start +
574 async_extent->ram_size - 1, GFP_NOFS);
576 /* allocate blocks */
577 ret = cow_file_range(inode, async_cow->locked_page,
579 async_extent->start +
580 async_extent->ram_size - 1,
581 &page_started, &nr_written, 0);
584 * if page_started, cow_file_range inserted an
585 * inline extent and took care of all the unlocking
586 * and IO for us. Otherwise, we need to submit
587 * all those pages down to the drive.
589 if (!page_started && !ret)
590 extent_write_locked_range(io_tree,
591 inode, async_extent->start,
592 async_extent->start +
593 async_extent->ram_size - 1,
601 lock_extent(io_tree, async_extent->start,
602 async_extent->start + async_extent->ram_size - 1,
605 trans = btrfs_join_transaction(root, 1);
606 ret = btrfs_reserve_extent(trans, root,
607 async_extent->compressed_size,
608 async_extent->compressed_size,
611 btrfs_end_transaction(trans, root);
615 for (i = 0; i < async_extent->nr_pages; i++) {
616 WARN_ON(async_extent->pages[i]->mapping);
617 page_cache_release(async_extent->pages[i]);
619 kfree(async_extent->pages);
620 async_extent->nr_pages = 0;
621 async_extent->pages = NULL;
622 unlock_extent(io_tree, async_extent->start,
623 async_extent->start +
624 async_extent->ram_size - 1, GFP_NOFS);
629 * here we're doing allocation and writeback of the
632 btrfs_drop_extent_cache(inode, async_extent->start,
633 async_extent->start +
634 async_extent->ram_size - 1, 0);
636 em = alloc_extent_map(GFP_NOFS);
637 em->start = async_extent->start;
638 em->len = async_extent->ram_size;
639 em->orig_start = em->start;
641 em->block_start = ins.objectid;
642 em->block_len = ins.offset;
643 em->bdev = root->fs_info->fs_devices->latest_bdev;
644 set_bit(EXTENT_FLAG_PINNED, &em->flags);
645 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
648 write_lock(&em_tree->lock);
649 ret = add_extent_mapping(em_tree, em);
650 write_unlock(&em_tree->lock);
651 if (ret != -EEXIST) {
655 btrfs_drop_extent_cache(inode, async_extent->start,
656 async_extent->start +
657 async_extent->ram_size - 1, 0);
660 ret = btrfs_add_ordered_extent(inode, async_extent->start,
662 async_extent->ram_size,
664 BTRFS_ORDERED_COMPRESSED);
668 * clear dirty, set writeback and unlock the pages.
670 extent_clear_unlock_delalloc(inode,
671 &BTRFS_I(inode)->io_tree,
673 async_extent->start +
674 async_extent->ram_size - 1,
675 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
676 EXTENT_CLEAR_UNLOCK |
677 EXTENT_CLEAR_DELALLOC |
678 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
680 ret = btrfs_submit_compressed_write(inode,
682 async_extent->ram_size,
684 ins.offset, async_extent->pages,
685 async_extent->nr_pages);
688 alloc_hint = ins.objectid + ins.offset;
697 * when extent_io.c finds a delayed allocation range in the file,
698 * the call backs end up in this code. The basic idea is to
699 * allocate extents on disk for the range, and create ordered data structs
700 * in ram to track those extents.
702 * locked_page is the page that writepage had locked already. We use
703 * it to make sure we don't do extra locks or unlocks.
705 * *page_started is set to one if we unlock locked_page and do everything
706 * required to start IO on it. It may be clean and already done with
709 static noinline int cow_file_range(struct inode *inode,
710 struct page *locked_page,
711 u64 start, u64 end, int *page_started,
712 unsigned long *nr_written,
715 struct btrfs_root *root = BTRFS_I(inode)->root;
716 struct btrfs_trans_handle *trans;
719 unsigned long ram_size;
722 u64 blocksize = root->sectorsize;
724 u64 isize = i_size_read(inode);
725 struct btrfs_key ins;
726 struct extent_map *em;
727 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
730 trans = btrfs_join_transaction(root, 1);
732 btrfs_set_trans_block_group(trans, inode);
734 actual_end = min_t(u64, isize, end + 1);
736 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
737 num_bytes = max(blocksize, num_bytes);
738 disk_num_bytes = num_bytes;
742 /* lets try to make an inline extent */
743 ret = cow_file_range_inline(trans, root, inode,
744 start, end, 0, NULL);
746 extent_clear_unlock_delalloc(inode,
747 &BTRFS_I(inode)->io_tree,
749 EXTENT_CLEAR_UNLOCK_PAGE |
750 EXTENT_CLEAR_UNLOCK |
751 EXTENT_CLEAR_DELALLOC |
752 EXTENT_CLEAR_ACCOUNTING |
754 EXTENT_SET_WRITEBACK |
755 EXTENT_END_WRITEBACK);
757 *nr_written = *nr_written +
758 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
765 BUG_ON(disk_num_bytes >
766 btrfs_super_total_bytes(&root->fs_info->super_copy));
769 read_lock(&BTRFS_I(inode)->extent_tree.lock);
770 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
774 * if block start isn't an actual block number then find the
775 * first block in this inode and use that as a hint. If that
776 * block is also bogus then just don't worry about it.
778 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
780 em = search_extent_mapping(em_tree, 0, 0);
781 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
782 alloc_hint = em->block_start;
786 alloc_hint = em->block_start;
790 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
791 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
793 while (disk_num_bytes > 0) {
796 cur_alloc_size = min(disk_num_bytes, root->fs_info->max_extent);
797 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
798 root->sectorsize, 0, alloc_hint,
802 em = alloc_extent_map(GFP_NOFS);
804 em->orig_start = em->start;
805 ram_size = ins.offset;
806 em->len = ins.offset;
808 em->block_start = ins.objectid;
809 em->block_len = ins.offset;
810 em->bdev = root->fs_info->fs_devices->latest_bdev;
811 set_bit(EXTENT_FLAG_PINNED, &em->flags);
814 write_lock(&em_tree->lock);
815 ret = add_extent_mapping(em_tree, em);
816 write_unlock(&em_tree->lock);
817 if (ret != -EEXIST) {
821 btrfs_drop_extent_cache(inode, start,
822 start + ram_size - 1, 0);
825 cur_alloc_size = ins.offset;
826 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
827 ram_size, cur_alloc_size, 0);
830 if (root->root_key.objectid ==
831 BTRFS_DATA_RELOC_TREE_OBJECTID) {
832 ret = btrfs_reloc_clone_csums(inode, start,
837 if (disk_num_bytes < cur_alloc_size)
840 /* we're not doing compressed IO, don't unlock the first
841 * page (which the caller expects to stay locked), don't
842 * clear any dirty bits and don't set any writeback bits
844 * Do set the Private2 bit so we know this page was properly
845 * setup for writepage
847 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
848 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
851 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
852 start, start + ram_size - 1,
854 disk_num_bytes -= cur_alloc_size;
855 num_bytes -= cur_alloc_size;
856 alloc_hint = ins.objectid + ins.offset;
857 start += cur_alloc_size;
861 btrfs_end_transaction(trans, root);
867 * work queue call back to started compression on a file and pages
869 static noinline void async_cow_start(struct btrfs_work *work)
871 struct async_cow *async_cow;
873 async_cow = container_of(work, struct async_cow, work);
875 compress_file_range(async_cow->inode, async_cow->locked_page,
876 async_cow->start, async_cow->end, async_cow,
879 async_cow->inode = NULL;
883 * work queue call back to submit previously compressed pages
885 static noinline void async_cow_submit(struct btrfs_work *work)
887 struct async_cow *async_cow;
888 struct btrfs_root *root;
889 unsigned long nr_pages;
891 async_cow = container_of(work, struct async_cow, work);
893 root = async_cow->root;
894 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
897 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
899 if (atomic_read(&root->fs_info->async_delalloc_pages) <
901 waitqueue_active(&root->fs_info->async_submit_wait))
902 wake_up(&root->fs_info->async_submit_wait);
904 if (async_cow->inode)
905 submit_compressed_extents(async_cow->inode, async_cow);
908 static noinline void async_cow_free(struct btrfs_work *work)
910 struct async_cow *async_cow;
911 async_cow = container_of(work, struct async_cow, work);
915 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
916 u64 start, u64 end, int *page_started,
917 unsigned long *nr_written)
919 struct async_cow *async_cow;
920 struct btrfs_root *root = BTRFS_I(inode)->root;
921 unsigned long nr_pages;
923 int limit = 10 * 1024 * 1042;
925 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
926 1, 0, NULL, GFP_NOFS);
927 while (start < end) {
928 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
929 async_cow->inode = inode;
930 async_cow->root = root;
931 async_cow->locked_page = locked_page;
932 async_cow->start = start;
934 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
937 cur_end = min(end, start + 512 * 1024 - 1);
939 async_cow->end = cur_end;
940 INIT_LIST_HEAD(&async_cow->extents);
942 async_cow->work.func = async_cow_start;
943 async_cow->work.ordered_func = async_cow_submit;
944 async_cow->work.ordered_free = async_cow_free;
945 async_cow->work.flags = 0;
947 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
949 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
951 btrfs_queue_worker(&root->fs_info->delalloc_workers,
954 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
955 wait_event(root->fs_info->async_submit_wait,
956 (atomic_read(&root->fs_info->async_delalloc_pages) <
960 while (atomic_read(&root->fs_info->async_submit_draining) &&
961 atomic_read(&root->fs_info->async_delalloc_pages)) {
962 wait_event(root->fs_info->async_submit_wait,
963 (atomic_read(&root->fs_info->async_delalloc_pages) ==
967 *nr_written += nr_pages;
974 static noinline int csum_exist_in_range(struct btrfs_root *root,
975 u64 bytenr, u64 num_bytes)
978 struct btrfs_ordered_sum *sums;
981 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
982 bytenr + num_bytes - 1, &list);
983 if (ret == 0 && list_empty(&list))
986 while (!list_empty(&list)) {
987 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
988 list_del(&sums->list);
995 * when nowcow writeback call back. This checks for snapshots or COW copies
996 * of the extents that exist in the file, and COWs the file as required.
998 * If no cow copies or snapshots exist, we write directly to the existing
1001 static noinline int run_delalloc_nocow(struct inode *inode,
1002 struct page *locked_page,
1003 u64 start, u64 end, int *page_started, int force,
1004 unsigned long *nr_written)
1006 struct btrfs_root *root = BTRFS_I(inode)->root;
1007 struct btrfs_trans_handle *trans;
1008 struct extent_buffer *leaf;
1009 struct btrfs_path *path;
1010 struct btrfs_file_extent_item *fi;
1011 struct btrfs_key found_key;
1024 path = btrfs_alloc_path();
1026 trans = btrfs_join_transaction(root, 1);
1029 cow_start = (u64)-1;
1032 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1035 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1036 leaf = path->nodes[0];
1037 btrfs_item_key_to_cpu(leaf, &found_key,
1038 path->slots[0] - 1);
1039 if (found_key.objectid == inode->i_ino &&
1040 found_key.type == BTRFS_EXTENT_DATA_KEY)
1045 leaf = path->nodes[0];
1046 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1047 ret = btrfs_next_leaf(root, path);
1052 leaf = path->nodes[0];
1058 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1060 if (found_key.objectid > inode->i_ino ||
1061 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1062 found_key.offset > end)
1065 if (found_key.offset > cur_offset) {
1066 extent_end = found_key.offset;
1071 fi = btrfs_item_ptr(leaf, path->slots[0],
1072 struct btrfs_file_extent_item);
1073 extent_type = btrfs_file_extent_type(leaf, fi);
1075 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1076 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1077 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1078 extent_offset = btrfs_file_extent_offset(leaf, fi);
1079 extent_end = found_key.offset +
1080 btrfs_file_extent_num_bytes(leaf, fi);
1081 if (extent_end <= start) {
1085 if (disk_bytenr == 0)
1087 if (btrfs_file_extent_compression(leaf, fi) ||
1088 btrfs_file_extent_encryption(leaf, fi) ||
1089 btrfs_file_extent_other_encoding(leaf, fi))
1091 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1093 if (btrfs_extent_readonly(root, disk_bytenr))
1095 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1097 extent_offset, disk_bytenr))
1099 disk_bytenr += extent_offset;
1100 disk_bytenr += cur_offset - found_key.offset;
1101 num_bytes = min(end + 1, extent_end) - cur_offset;
1103 * force cow if csum exists in the range.
1104 * this ensure that csum for a given extent are
1105 * either valid or do not exist.
1107 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1110 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1111 extent_end = found_key.offset +
1112 btrfs_file_extent_inline_len(leaf, fi);
1113 extent_end = ALIGN(extent_end, root->sectorsize);
1118 if (extent_end <= start) {
1123 if (cow_start == (u64)-1)
1124 cow_start = cur_offset;
1125 cur_offset = extent_end;
1126 if (cur_offset > end)
1132 btrfs_release_path(root, path);
1133 if (cow_start != (u64)-1) {
1134 ret = cow_file_range(inode, locked_page, cow_start,
1135 found_key.offset - 1, page_started,
1138 cow_start = (u64)-1;
1141 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1142 struct extent_map *em;
1143 struct extent_map_tree *em_tree;
1144 em_tree = &BTRFS_I(inode)->extent_tree;
1145 em = alloc_extent_map(GFP_NOFS);
1146 em->start = cur_offset;
1147 em->orig_start = em->start;
1148 em->len = num_bytes;
1149 em->block_len = num_bytes;
1150 em->block_start = disk_bytenr;
1151 em->bdev = root->fs_info->fs_devices->latest_bdev;
1152 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1154 write_lock(&em_tree->lock);
1155 ret = add_extent_mapping(em_tree, em);
1156 write_unlock(&em_tree->lock);
1157 if (ret != -EEXIST) {
1158 free_extent_map(em);
1161 btrfs_drop_extent_cache(inode, em->start,
1162 em->start + em->len - 1, 0);
1164 type = BTRFS_ORDERED_PREALLOC;
1166 type = BTRFS_ORDERED_NOCOW;
1169 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1170 num_bytes, num_bytes, type);
1173 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1174 cur_offset, cur_offset + num_bytes - 1,
1175 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1176 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1177 EXTENT_SET_PRIVATE2);
1178 cur_offset = extent_end;
1179 if (cur_offset > end)
1182 btrfs_release_path(root, path);
1184 if (cur_offset <= end && cow_start == (u64)-1)
1185 cow_start = cur_offset;
1186 if (cow_start != (u64)-1) {
1187 ret = cow_file_range(inode, locked_page, cow_start, end,
1188 page_started, nr_written, 1);
1192 ret = btrfs_end_transaction(trans, root);
1194 btrfs_free_path(path);
1199 * extent_io.c call back to do delayed allocation processing
1201 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1202 u64 start, u64 end, int *page_started,
1203 unsigned long *nr_written)
1206 struct btrfs_root *root = BTRFS_I(inode)->root;
1208 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1209 ret = run_delalloc_nocow(inode, locked_page, start, end,
1210 page_started, 1, nr_written);
1211 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1212 ret = run_delalloc_nocow(inode, locked_page, start, end,
1213 page_started, 0, nr_written);
1214 else if (!btrfs_test_opt(root, COMPRESS))
1215 ret = cow_file_range(inode, locked_page, start, end,
1216 page_started, nr_written, 1);
1218 ret = cow_file_range_async(inode, locked_page, start, end,
1219 page_started, nr_written);
1223 static int btrfs_split_extent_hook(struct inode *inode,
1224 struct extent_state *orig, u64 split)
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1229 if (!(orig->state & EXTENT_DELALLOC))
1232 size = orig->end - orig->start + 1;
1233 if (size > root->fs_info->max_extent) {
1237 new_size = orig->end - split + 1;
1238 num_extents = div64_u64(size + root->fs_info->max_extent - 1,
1239 root->fs_info->max_extent);
1242 * if we break a large extent up then leave oustanding_extents
1243 * be, since we've already accounted for the large extent.
1245 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1246 root->fs_info->max_extent) < num_extents)
1250 spin_lock(&BTRFS_I(inode)->accounting_lock);
1251 BTRFS_I(inode)->outstanding_extents++;
1252 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1258 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1259 * extents so we can keep track of new extents that are just merged onto old
1260 * extents, such as when we are doing sequential writes, so we can properly
1261 * account for the metadata space we'll need.
1263 static int btrfs_merge_extent_hook(struct inode *inode,
1264 struct extent_state *new,
1265 struct extent_state *other)
1267 struct btrfs_root *root = BTRFS_I(inode)->root;
1268 u64 new_size, old_size;
1271 /* not delalloc, ignore it */
1272 if (!(other->state & EXTENT_DELALLOC))
1275 old_size = other->end - other->start + 1;
1276 if (new->start < other->start)
1277 new_size = other->end - new->start + 1;
1279 new_size = new->end - other->start + 1;
1281 /* we're not bigger than the max, unreserve the space and go */
1282 if (new_size <= root->fs_info->max_extent) {
1283 spin_lock(&BTRFS_I(inode)->accounting_lock);
1284 BTRFS_I(inode)->outstanding_extents--;
1285 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1290 * If we grew by another max_extent, just return, we want to keep that
1293 num_extents = div64_u64(old_size + root->fs_info->max_extent - 1,
1294 root->fs_info->max_extent);
1295 if (div64_u64(new_size + root->fs_info->max_extent - 1,
1296 root->fs_info->max_extent) > num_extents)
1299 spin_lock(&BTRFS_I(inode)->accounting_lock);
1300 BTRFS_I(inode)->outstanding_extents--;
1301 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1307 * extent_io.c set_bit_hook, used to track delayed allocation
1308 * bytes in this file, and to maintain the list of inodes that
1309 * have pending delalloc work to be done.
1311 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1312 unsigned long old, unsigned long bits)
1316 * set_bit and clear bit hooks normally require _irqsave/restore
1317 * but in this case, we are only testeing for the DELALLOC
1318 * bit, which is only set or cleared with irqs on
1320 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1321 struct btrfs_root *root = BTRFS_I(inode)->root;
1323 spin_lock(&BTRFS_I(inode)->accounting_lock);
1324 BTRFS_I(inode)->outstanding_extents++;
1325 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1326 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1327 spin_lock(&root->fs_info->delalloc_lock);
1328 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1329 root->fs_info->delalloc_bytes += end - start + 1;
1330 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1331 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1332 &root->fs_info->delalloc_inodes);
1334 spin_unlock(&root->fs_info->delalloc_lock);
1340 * extent_io.c clear_bit_hook, see set_bit_hook for why
1342 static int btrfs_clear_bit_hook(struct inode *inode,
1343 struct extent_state *state, unsigned long bits)
1346 * set_bit and clear bit hooks normally require _irqsave/restore
1347 * but in this case, we are only testeing for the DELALLOC
1348 * bit, which is only set or cleared with irqs on
1350 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1351 struct btrfs_root *root = BTRFS_I(inode)->root;
1353 if (bits & EXTENT_DO_ACCOUNTING) {
1354 spin_lock(&BTRFS_I(inode)->accounting_lock);
1355 BTRFS_I(inode)->outstanding_extents--;
1356 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1357 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1360 spin_lock(&root->fs_info->delalloc_lock);
1361 if (state->end - state->start + 1 >
1362 root->fs_info->delalloc_bytes) {
1363 printk(KERN_INFO "btrfs warning: delalloc account "
1365 (unsigned long long)
1366 state->end - state->start + 1,
1367 (unsigned long long)
1368 root->fs_info->delalloc_bytes);
1369 btrfs_delalloc_free_space(root, inode, (u64)-1);
1370 root->fs_info->delalloc_bytes = 0;
1371 BTRFS_I(inode)->delalloc_bytes = 0;
1373 btrfs_delalloc_free_space(root, inode,
1376 root->fs_info->delalloc_bytes -= state->end -
1378 BTRFS_I(inode)->delalloc_bytes -= state->end -
1381 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1382 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1383 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1385 spin_unlock(&root->fs_info->delalloc_lock);
1391 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1392 * we don't create bios that span stripes or chunks
1394 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1395 size_t size, struct bio *bio,
1396 unsigned long bio_flags)
1398 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1399 struct btrfs_mapping_tree *map_tree;
1400 u64 logical = (u64)bio->bi_sector << 9;
1405 if (bio_flags & EXTENT_BIO_COMPRESSED)
1408 length = bio->bi_size;
1409 map_tree = &root->fs_info->mapping_tree;
1410 map_length = length;
1411 ret = btrfs_map_block(map_tree, READ, logical,
1412 &map_length, NULL, 0);
1414 if (map_length < length + size)
1420 * in order to insert checksums into the metadata in large chunks,
1421 * we wait until bio submission time. All the pages in the bio are
1422 * checksummed and sums are attached onto the ordered extent record.
1424 * At IO completion time the cums attached on the ordered extent record
1425 * are inserted into the btree
1427 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1428 struct bio *bio, int mirror_num,
1429 unsigned long bio_flags)
1431 struct btrfs_root *root = BTRFS_I(inode)->root;
1434 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1440 * in order to insert checksums into the metadata in large chunks,
1441 * we wait until bio submission time. All the pages in the bio are
1442 * checksummed and sums are attached onto the ordered extent record.
1444 * At IO completion time the cums attached on the ordered extent record
1445 * are inserted into the btree
1447 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1448 int mirror_num, unsigned long bio_flags)
1450 struct btrfs_root *root = BTRFS_I(inode)->root;
1451 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1455 * extent_io.c submission hook. This does the right thing for csum calculation
1456 * on write, or reading the csums from the tree before a read
1458 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1459 int mirror_num, unsigned long bio_flags)
1461 struct btrfs_root *root = BTRFS_I(inode)->root;
1465 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1467 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1470 if (!(rw & (1 << BIO_RW))) {
1471 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1472 return btrfs_submit_compressed_read(inode, bio,
1473 mirror_num, bio_flags);
1474 } else if (!skip_sum)
1475 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1477 } else if (!skip_sum) {
1478 /* csum items have already been cloned */
1479 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1481 /* we're doing a write, do the async checksumming */
1482 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1483 inode, rw, bio, mirror_num,
1484 bio_flags, __btrfs_submit_bio_start,
1485 __btrfs_submit_bio_done);
1489 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1493 * given a list of ordered sums record them in the inode. This happens
1494 * at IO completion time based on sums calculated at bio submission time.
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497 struct inode *inode, u64 file_offset,
1498 struct list_head *list)
1500 struct btrfs_ordered_sum *sum;
1502 btrfs_set_trans_block_group(trans, inode);
1504 list_for_each_entry(sum, list, list) {
1505 btrfs_csum_file_blocks(trans,
1506 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
1513 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup {
1522 struct btrfs_work work;
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 struct btrfs_writepage_fixup *fixup;
1528 struct btrfs_ordered_extent *ordered;
1530 struct inode *inode;
1534 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1538 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1539 ClearPageChecked(page);
1543 inode = page->mapping->host;
1544 page_start = page_offset(page);
1545 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1547 lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1549 /* already ordered? We're done */
1550 if (PagePrivate2(page))
1553 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1555 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
1556 page_end, GFP_NOFS);
1558 btrfs_start_ordered_extent(inode, ordered, 1);
1562 btrfs_set_extent_delalloc(inode, page_start, page_end);
1563 ClearPageChecked(page);
1565 unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
1568 page_cache_release(page);
1572 * There are a few paths in the higher layers of the kernel that directly
1573 * set the page dirty bit without asking the filesystem if it is a
1574 * good idea. This causes problems because we want to make sure COW
1575 * properly happens and the data=ordered rules are followed.
1577 * In our case any range that doesn't have the ORDERED bit set
1578 * hasn't been properly setup for IO. We kick off an async process
1579 * to fix it up. The async helper will wait for ordered extents, set
1580 * the delalloc bit and make it safe to write the page.
1582 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1584 struct inode *inode = page->mapping->host;
1585 struct btrfs_writepage_fixup *fixup;
1586 struct btrfs_root *root = BTRFS_I(inode)->root;
1588 /* this page is properly in the ordered list */
1589 if (TestClearPagePrivate2(page))
1592 if (PageChecked(page))
1595 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1599 SetPageChecked(page);
1600 page_cache_get(page);
1601 fixup->work.func = btrfs_writepage_fixup_worker;
1603 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1607 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1608 struct inode *inode, u64 file_pos,
1609 u64 disk_bytenr, u64 disk_num_bytes,
1610 u64 num_bytes, u64 ram_bytes,
1611 u8 compression, u8 encryption,
1612 u16 other_encoding, int extent_type)
1614 struct btrfs_root *root = BTRFS_I(inode)->root;
1615 struct btrfs_file_extent_item *fi;
1616 struct btrfs_path *path;
1617 struct extent_buffer *leaf;
1618 struct btrfs_key ins;
1622 path = btrfs_alloc_path();
1625 path->leave_spinning = 1;
1628 * we may be replacing one extent in the tree with another.
1629 * The new extent is pinned in the extent map, and we don't want
1630 * to drop it from the cache until it is completely in the btree.
1632 * So, tell btrfs_drop_extents to leave this extent in the cache.
1633 * the caller is expected to unpin it and allow it to be merged
1636 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1640 ins.objectid = inode->i_ino;
1641 ins.offset = file_pos;
1642 ins.type = BTRFS_EXTENT_DATA_KEY;
1643 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1645 leaf = path->nodes[0];
1646 fi = btrfs_item_ptr(leaf, path->slots[0],
1647 struct btrfs_file_extent_item);
1648 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1649 btrfs_set_file_extent_type(leaf, fi, extent_type);
1650 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1651 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1652 btrfs_set_file_extent_offset(leaf, fi, 0);
1653 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1654 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1655 btrfs_set_file_extent_compression(leaf, fi, compression);
1656 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1657 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1659 btrfs_unlock_up_safe(path, 1);
1660 btrfs_set_lock_blocking(leaf);
1662 btrfs_mark_buffer_dirty(leaf);
1664 inode_add_bytes(inode, num_bytes);
1666 ins.objectid = disk_bytenr;
1667 ins.offset = disk_num_bytes;
1668 ins.type = BTRFS_EXTENT_ITEM_KEY;
1669 ret = btrfs_alloc_reserved_file_extent(trans, root,
1670 root->root_key.objectid,
1671 inode->i_ino, file_pos, &ins);
1673 btrfs_free_path(path);
1679 * helper function for btrfs_finish_ordered_io, this
1680 * just reads in some of the csum leaves to prime them into ram
1681 * before we start the transaction. It limits the amount of btree
1682 * reads required while inside the transaction.
1684 static noinline void reada_csum(struct btrfs_root *root,
1685 struct btrfs_path *path,
1686 struct btrfs_ordered_extent *ordered_extent)
1688 struct btrfs_ordered_sum *sum;
1691 sum = list_entry(ordered_extent->list.next, struct btrfs_ordered_sum,
1693 bytenr = sum->sums[0].bytenr;
1696 * we don't care about the results, the point of this search is
1697 * just to get the btree leaves into ram
1699 btrfs_lookup_csum(NULL, root->fs_info->csum_root, path, bytenr, 0);
1702 /* as ordered data IO finishes, this gets called so we can finish
1703 * an ordered extent if the range of bytes in the file it covers are
1706 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1708 struct btrfs_root *root = BTRFS_I(inode)->root;
1709 struct btrfs_trans_handle *trans;
1710 struct btrfs_ordered_extent *ordered_extent = NULL;
1711 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1712 struct btrfs_path *path;
1716 ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
1721 * before we join the transaction, try to do some of our IO.
1722 * This will limit the amount of IO that we have to do with
1723 * the transaction running. We're unlikely to need to do any
1724 * IO if the file extents are new, the disk_i_size checks
1725 * covers the most common case.
1727 if (start < BTRFS_I(inode)->disk_i_size) {
1728 path = btrfs_alloc_path();
1730 ret = btrfs_lookup_file_extent(NULL, root, path,
1733 ordered_extent = btrfs_lookup_ordered_extent(inode,
1735 if (!list_empty(&ordered_extent->list)) {
1736 btrfs_release_path(root, path);
1737 reada_csum(root, path, ordered_extent);
1739 btrfs_free_path(path);
1743 if (!ordered_extent)
1744 ordered_extent = btrfs_lookup_ordered_extent(inode, start);
1745 BUG_ON(!ordered_extent);
1746 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1747 BUG_ON(!list_empty(&ordered_extent->list));
1748 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1750 trans = btrfs_join_transaction(root, 1);
1751 ret = btrfs_update_inode(trans, root, inode);
1753 btrfs_end_transaction(trans, root);
1758 lock_extent(io_tree, ordered_extent->file_offset,
1759 ordered_extent->file_offset + ordered_extent->len - 1,
1762 trans = btrfs_join_transaction(root, 1);
1764 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1766 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1768 ret = btrfs_mark_extent_written(trans, inode,
1769 ordered_extent->file_offset,
1770 ordered_extent->file_offset +
1771 ordered_extent->len);
1774 ret = insert_reserved_file_extent(trans, inode,
1775 ordered_extent->file_offset,
1776 ordered_extent->start,
1777 ordered_extent->disk_len,
1778 ordered_extent->len,
1779 ordered_extent->len,
1781 BTRFS_FILE_EXTENT_REG);
1782 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1783 ordered_extent->file_offset,
1784 ordered_extent->len);
1787 unlock_extent(io_tree, ordered_extent->file_offset,
1788 ordered_extent->file_offset + ordered_extent->len - 1,
1790 add_pending_csums(trans, inode, ordered_extent->file_offset,
1791 &ordered_extent->list);
1793 /* this also removes the ordered extent from the tree */
1794 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1795 ret = btrfs_update_inode(trans, root, inode);
1797 btrfs_end_transaction(trans, root);
1800 btrfs_put_ordered_extent(ordered_extent);
1801 /* once for the tree */
1802 btrfs_put_ordered_extent(ordered_extent);
1807 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1808 struct extent_state *state, int uptodate)
1810 ClearPagePrivate2(page);
1811 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1815 * When IO fails, either with EIO or csum verification fails, we
1816 * try other mirrors that might have a good copy of the data. This
1817 * io_failure_record is used to record state as we go through all the
1818 * mirrors. If another mirror has good data, the page is set up to date
1819 * and things continue. If a good mirror can't be found, the original
1820 * bio end_io callback is called to indicate things have failed.
1822 struct io_failure_record {
1827 unsigned long bio_flags;
1831 static int btrfs_io_failed_hook(struct bio *failed_bio,
1832 struct page *page, u64 start, u64 end,
1833 struct extent_state *state)
1835 struct io_failure_record *failrec = NULL;
1837 struct extent_map *em;
1838 struct inode *inode = page->mapping->host;
1839 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1840 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1847 ret = get_state_private(failure_tree, start, &private);
1849 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1852 failrec->start = start;
1853 failrec->len = end - start + 1;
1854 failrec->last_mirror = 0;
1855 failrec->bio_flags = 0;
1857 read_lock(&em_tree->lock);
1858 em = lookup_extent_mapping(em_tree, start, failrec->len);
1859 if (em->start > start || em->start + em->len < start) {
1860 free_extent_map(em);
1863 read_unlock(&em_tree->lock);
1865 if (!em || IS_ERR(em)) {
1869 logical = start - em->start;
1870 logical = em->block_start + logical;
1871 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1872 logical = em->block_start;
1873 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1875 failrec->logical = logical;
1876 free_extent_map(em);
1877 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1878 EXTENT_DIRTY, GFP_NOFS);
1879 set_state_private(failure_tree, start,
1880 (u64)(unsigned long)failrec);
1882 failrec = (struct io_failure_record *)(unsigned long)private;
1884 num_copies = btrfs_num_copies(
1885 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1886 failrec->logical, failrec->len);
1887 failrec->last_mirror++;
1889 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1890 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1893 if (state && state->start != failrec->start)
1895 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1897 if (!state || failrec->last_mirror > num_copies) {
1898 set_state_private(failure_tree, failrec->start, 0);
1899 clear_extent_bits(failure_tree, failrec->start,
1900 failrec->start + failrec->len - 1,
1901 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1905 bio = bio_alloc(GFP_NOFS, 1);
1906 bio->bi_private = state;
1907 bio->bi_end_io = failed_bio->bi_end_io;
1908 bio->bi_sector = failrec->logical >> 9;
1909 bio->bi_bdev = failed_bio->bi_bdev;
1912 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1913 if (failed_bio->bi_rw & (1 << BIO_RW))
1918 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1919 failrec->last_mirror,
1920 failrec->bio_flags);
1925 * each time an IO finishes, we do a fast check in the IO failure tree
1926 * to see if we need to process or clean up an io_failure_record
1928 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1931 u64 private_failure;
1932 struct io_failure_record *failure;
1936 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1937 (u64)-1, 1, EXTENT_DIRTY)) {
1938 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1939 start, &private_failure);
1941 failure = (struct io_failure_record *)(unsigned long)
1943 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1945 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1947 failure->start + failure->len - 1,
1948 EXTENT_DIRTY | EXTENT_LOCKED,
1957 * when reads are done, we need to check csums to verify the data is correct
1958 * if there's a match, we allow the bio to finish. If not, we go through
1959 * the io_failure_record routines to find good copies
1961 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1962 struct extent_state *state)
1964 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1965 struct inode *inode = page->mapping->host;
1966 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1968 u64 private = ~(u32)0;
1970 struct btrfs_root *root = BTRFS_I(inode)->root;
1973 if (PageChecked(page)) {
1974 ClearPageChecked(page);
1978 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1981 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1982 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1983 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1988 if (state && state->start == start) {
1989 private = state->private;
1992 ret = get_state_private(io_tree, start, &private);
1994 kaddr = kmap_atomic(page, KM_USER0);
1998 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1999 btrfs_csum_final(csum, (char *)&csum);
2000 if (csum != private)
2003 kunmap_atomic(kaddr, KM_USER0);
2005 /* if the io failure tree for this inode is non-empty,
2006 * check to see if we've recovered from a failed IO
2008 btrfs_clean_io_failures(inode, start);
2012 if (printk_ratelimit()) {
2013 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2014 "private %llu\n", page->mapping->host->i_ino,
2015 (unsigned long long)start, csum,
2016 (unsigned long long)private);
2018 memset(kaddr + offset, 1, end - start + 1);
2019 flush_dcache_page(page);
2020 kunmap_atomic(kaddr, KM_USER0);
2026 struct delayed_iput {
2027 struct list_head list;
2028 struct inode *inode;
2031 void btrfs_add_delayed_iput(struct inode *inode)
2033 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2034 struct delayed_iput *delayed;
2036 if (atomic_add_unless(&inode->i_count, -1, 1))
2039 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2040 delayed->inode = inode;
2042 spin_lock(&fs_info->delayed_iput_lock);
2043 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2044 spin_unlock(&fs_info->delayed_iput_lock);
2047 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2050 struct btrfs_fs_info *fs_info = root->fs_info;
2051 struct delayed_iput *delayed;
2054 spin_lock(&fs_info->delayed_iput_lock);
2055 empty = list_empty(&fs_info->delayed_iputs);
2056 spin_unlock(&fs_info->delayed_iput_lock);
2060 down_read(&root->fs_info->cleanup_work_sem);
2061 spin_lock(&fs_info->delayed_iput_lock);
2062 list_splice_init(&fs_info->delayed_iputs, &list);
2063 spin_unlock(&fs_info->delayed_iput_lock);
2065 while (!list_empty(&list)) {
2066 delayed = list_entry(list.next, struct delayed_iput, list);
2067 list_del(&delayed->list);
2068 iput(delayed->inode);
2071 up_read(&root->fs_info->cleanup_work_sem);
2075 * This creates an orphan entry for the given inode in case something goes
2076 * wrong in the middle of an unlink/truncate.
2078 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2083 spin_lock(&root->list_lock);
2085 /* already on the orphan list, we're good */
2086 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2087 spin_unlock(&root->list_lock);
2091 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2093 spin_unlock(&root->list_lock);
2096 * insert an orphan item to track this unlinked/truncated file
2098 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2104 * We have done the truncate/delete so we can go ahead and remove the orphan
2105 * item for this particular inode.
2107 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2109 struct btrfs_root *root = BTRFS_I(inode)->root;
2112 spin_lock(&root->list_lock);
2114 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2115 spin_unlock(&root->list_lock);
2119 list_del_init(&BTRFS_I(inode)->i_orphan);
2121 spin_unlock(&root->list_lock);
2125 spin_unlock(&root->list_lock);
2127 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2133 * this cleans up any orphans that may be left on the list from the last use
2136 void btrfs_orphan_cleanup(struct btrfs_root *root)
2138 struct btrfs_path *path;
2139 struct extent_buffer *leaf;
2140 struct btrfs_item *item;
2141 struct btrfs_key key, found_key;
2142 struct btrfs_trans_handle *trans;
2143 struct inode *inode;
2144 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2146 if (!xchg(&root->clean_orphans, 0))
2149 path = btrfs_alloc_path();
2153 key.objectid = BTRFS_ORPHAN_OBJECTID;
2154 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2155 key.offset = (u64)-1;
2158 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2160 printk(KERN_ERR "Error searching slot for orphan: %d"
2166 * if ret == 0 means we found what we were searching for, which
2167 * is weird, but possible, so only screw with path if we didnt
2168 * find the key and see if we have stuff that matches
2171 if (path->slots[0] == 0)
2176 /* pull out the item */
2177 leaf = path->nodes[0];
2178 item = btrfs_item_nr(leaf, path->slots[0]);
2179 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2181 /* make sure the item matches what we want */
2182 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2184 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2187 /* release the path since we're done with it */
2188 btrfs_release_path(root, path);
2191 * this is where we are basically btrfs_lookup, without the
2192 * crossing root thing. we store the inode number in the
2193 * offset of the orphan item.
2195 found_key.objectid = found_key.offset;
2196 found_key.type = BTRFS_INODE_ITEM_KEY;
2197 found_key.offset = 0;
2198 inode = btrfs_iget(root->fs_info->sb, &found_key, root);
2203 * add this inode to the orphan list so btrfs_orphan_del does
2204 * the proper thing when we hit it
2206 spin_lock(&root->list_lock);
2207 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2208 spin_unlock(&root->list_lock);
2211 * if this is a bad inode, means we actually succeeded in
2212 * removing the inode, but not the orphan record, which means
2213 * we need to manually delete the orphan since iput will just
2214 * do a destroy_inode
2216 if (is_bad_inode(inode)) {
2217 trans = btrfs_start_transaction(root, 1);
2218 btrfs_orphan_del(trans, inode);
2219 btrfs_end_transaction(trans, root);
2224 /* if we have links, this was a truncate, lets do that */
2225 if (inode->i_nlink) {
2227 btrfs_truncate(inode);
2232 /* this will do delete_inode and everything for us */
2237 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2239 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2241 btrfs_free_path(path);
2245 * very simple check to peek ahead in the leaf looking for xattrs. If we
2246 * don't find any xattrs, we know there can't be any acls.
2248 * slot is the slot the inode is in, objectid is the objectid of the inode
2250 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2251 int slot, u64 objectid)
2253 u32 nritems = btrfs_header_nritems(leaf);
2254 struct btrfs_key found_key;
2258 while (slot < nritems) {
2259 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2261 /* we found a different objectid, there must not be acls */
2262 if (found_key.objectid != objectid)
2265 /* we found an xattr, assume we've got an acl */
2266 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2270 * we found a key greater than an xattr key, there can't
2271 * be any acls later on
2273 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2280 * it goes inode, inode backrefs, xattrs, extents,
2281 * so if there are a ton of hard links to an inode there can
2282 * be a lot of backrefs. Don't waste time searching too hard,
2283 * this is just an optimization
2288 /* we hit the end of the leaf before we found an xattr or
2289 * something larger than an xattr. We have to assume the inode
2296 * read an inode from the btree into the in-memory inode
2298 static void btrfs_read_locked_inode(struct inode *inode)
2300 struct btrfs_path *path;
2301 struct extent_buffer *leaf;
2302 struct btrfs_inode_item *inode_item;
2303 struct btrfs_timespec *tspec;
2304 struct btrfs_root *root = BTRFS_I(inode)->root;
2305 struct btrfs_key location;
2307 u64 alloc_group_block;
2311 path = btrfs_alloc_path();
2313 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2315 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2319 leaf = path->nodes[0];
2320 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2321 struct btrfs_inode_item);
2323 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2324 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2325 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2326 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2327 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2329 tspec = btrfs_inode_atime(inode_item);
2330 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2331 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2333 tspec = btrfs_inode_mtime(inode_item);
2334 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2335 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2337 tspec = btrfs_inode_ctime(inode_item);
2338 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2339 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2341 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2342 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2343 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2344 inode->i_generation = BTRFS_I(inode)->generation;
2346 rdev = btrfs_inode_rdev(leaf, inode_item);
2348 BTRFS_I(inode)->index_cnt = (u64)-1;
2349 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2351 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2354 * try to precache a NULL acl entry for files that don't have
2355 * any xattrs or acls
2357 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2359 cache_no_acl(inode);
2361 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2362 alloc_group_block, 0);
2363 btrfs_free_path(path);
2366 switch (inode->i_mode & S_IFMT) {
2368 inode->i_mapping->a_ops = &btrfs_aops;
2369 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2370 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2371 inode->i_fop = &btrfs_file_operations;
2372 inode->i_op = &btrfs_file_inode_operations;
2375 inode->i_fop = &btrfs_dir_file_operations;
2376 if (root == root->fs_info->tree_root)
2377 inode->i_op = &btrfs_dir_ro_inode_operations;
2379 inode->i_op = &btrfs_dir_inode_operations;
2382 inode->i_op = &btrfs_symlink_inode_operations;
2383 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2384 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2387 inode->i_op = &btrfs_special_inode_operations;
2388 init_special_inode(inode, inode->i_mode, rdev);
2392 btrfs_update_iflags(inode);
2396 btrfs_free_path(path);
2397 make_bad_inode(inode);
2401 * given a leaf and an inode, copy the inode fields into the leaf
2403 static void fill_inode_item(struct btrfs_trans_handle *trans,
2404 struct extent_buffer *leaf,
2405 struct btrfs_inode_item *item,
2406 struct inode *inode)
2408 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2409 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2410 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2411 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2412 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2414 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2415 inode->i_atime.tv_sec);
2416 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2417 inode->i_atime.tv_nsec);
2419 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2420 inode->i_mtime.tv_sec);
2421 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2422 inode->i_mtime.tv_nsec);
2424 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2425 inode->i_ctime.tv_sec);
2426 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2427 inode->i_ctime.tv_nsec);
2429 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2430 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2431 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2432 btrfs_set_inode_transid(leaf, item, trans->transid);
2433 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2434 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2435 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2439 * copy everything in the in-memory inode into the btree.
2441 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2442 struct btrfs_root *root, struct inode *inode)
2444 struct btrfs_inode_item *inode_item;
2445 struct btrfs_path *path;
2446 struct extent_buffer *leaf;
2449 path = btrfs_alloc_path();
2451 path->leave_spinning = 1;
2452 ret = btrfs_lookup_inode(trans, root, path,
2453 &BTRFS_I(inode)->location, 1);
2460 btrfs_unlock_up_safe(path, 1);
2461 leaf = path->nodes[0];
2462 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2463 struct btrfs_inode_item);
2465 fill_inode_item(trans, leaf, inode_item, inode);
2466 btrfs_mark_buffer_dirty(leaf);
2467 btrfs_set_inode_last_trans(trans, inode);
2470 btrfs_free_path(path);
2476 * unlink helper that gets used here in inode.c and in the tree logging
2477 * recovery code. It remove a link in a directory with a given name, and
2478 * also drops the back refs in the inode to the directory
2480 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2481 struct btrfs_root *root,
2482 struct inode *dir, struct inode *inode,
2483 const char *name, int name_len)
2485 struct btrfs_path *path;
2487 struct extent_buffer *leaf;
2488 struct btrfs_dir_item *di;
2489 struct btrfs_key key;
2492 path = btrfs_alloc_path();
2498 path->leave_spinning = 1;
2499 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2500 name, name_len, -1);
2509 leaf = path->nodes[0];
2510 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2511 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2514 btrfs_release_path(root, path);
2516 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2518 dir->i_ino, &index);
2520 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2521 "inode %lu parent %lu\n", name_len, name,
2522 inode->i_ino, dir->i_ino);
2526 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2527 index, name, name_len, -1);
2536 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2537 btrfs_release_path(root, path);
2539 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2541 BUG_ON(ret != 0 && ret != -ENOENT);
2543 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2547 btrfs_free_path(path);
2551 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2552 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2553 btrfs_update_inode(trans, root, dir);
2554 btrfs_drop_nlink(inode);
2555 ret = btrfs_update_inode(trans, root, inode);
2560 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2562 struct btrfs_root *root;
2563 struct btrfs_trans_handle *trans;
2564 struct inode *inode = dentry->d_inode;
2566 unsigned long nr = 0;
2568 root = BTRFS_I(dir)->root;
2571 * 5 items for unlink inode
2574 ret = btrfs_reserve_metadata_space(root, 6);
2578 trans = btrfs_start_transaction(root, 1);
2579 if (IS_ERR(trans)) {
2580 btrfs_unreserve_metadata_space(root, 6);
2581 return PTR_ERR(trans);
2584 btrfs_set_trans_block_group(trans, dir);
2586 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2588 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2589 dentry->d_name.name, dentry->d_name.len);
2591 if (inode->i_nlink == 0)
2592 ret = btrfs_orphan_add(trans, inode);
2594 nr = trans->blocks_used;
2596 btrfs_end_transaction_throttle(trans, root);
2597 btrfs_unreserve_metadata_space(root, 6);
2598 btrfs_btree_balance_dirty(root, nr);
2602 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2603 struct btrfs_root *root,
2604 struct inode *dir, u64 objectid,
2605 const char *name, int name_len)
2607 struct btrfs_path *path;
2608 struct extent_buffer *leaf;
2609 struct btrfs_dir_item *di;
2610 struct btrfs_key key;
2614 path = btrfs_alloc_path();
2618 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2619 name, name_len, -1);
2620 BUG_ON(!di || IS_ERR(di));
2622 leaf = path->nodes[0];
2623 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2624 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2625 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2627 btrfs_release_path(root, path);
2629 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2630 objectid, root->root_key.objectid,
2631 dir->i_ino, &index, name, name_len);
2633 BUG_ON(ret != -ENOENT);
2634 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2636 BUG_ON(!di || IS_ERR(di));
2638 leaf = path->nodes[0];
2639 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2640 btrfs_release_path(root, path);
2644 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2645 index, name, name_len, -1);
2646 BUG_ON(!di || IS_ERR(di));
2648 leaf = path->nodes[0];
2649 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2650 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2651 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2653 btrfs_release_path(root, path);
2655 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2656 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2657 ret = btrfs_update_inode(trans, root, dir);
2659 dir->i_sb->s_dirt = 1;
2661 btrfs_free_path(path);
2665 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2667 struct inode *inode = dentry->d_inode;
2670 struct btrfs_root *root = BTRFS_I(dir)->root;
2671 struct btrfs_trans_handle *trans;
2672 unsigned long nr = 0;
2674 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2675 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2678 ret = btrfs_reserve_metadata_space(root, 5);
2682 trans = btrfs_start_transaction(root, 1);
2683 if (IS_ERR(trans)) {
2684 btrfs_unreserve_metadata_space(root, 5);
2685 return PTR_ERR(trans);
2688 btrfs_set_trans_block_group(trans, dir);
2690 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2691 err = btrfs_unlink_subvol(trans, root, dir,
2692 BTRFS_I(inode)->location.objectid,
2693 dentry->d_name.name,
2694 dentry->d_name.len);
2698 err = btrfs_orphan_add(trans, inode);
2702 /* now the directory is empty */
2703 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2704 dentry->d_name.name, dentry->d_name.len);
2706 btrfs_i_size_write(inode, 0);
2708 nr = trans->blocks_used;
2709 ret = btrfs_end_transaction_throttle(trans, root);
2710 btrfs_unreserve_metadata_space(root, 5);
2711 btrfs_btree_balance_dirty(root, nr);
2720 * when truncating bytes in a file, it is possible to avoid reading
2721 * the leaves that contain only checksum items. This can be the
2722 * majority of the IO required to delete a large file, but it must
2723 * be done carefully.
2725 * The keys in the level just above the leaves are checked to make sure
2726 * the lowest key in a given leaf is a csum key, and starts at an offset
2727 * after the new size.
2729 * Then the key for the next leaf is checked to make sure it also has
2730 * a checksum item for the same file. If it does, we know our target leaf
2731 * contains only checksum items, and it can be safely freed without reading
2734 * This is just an optimization targeted at large files. It may do
2735 * nothing. It will return 0 unless things went badly.
2737 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2738 struct btrfs_root *root,
2739 struct btrfs_path *path,
2740 struct inode *inode, u64 new_size)
2742 struct btrfs_key key;
2745 struct btrfs_key found_key;
2746 struct btrfs_key other_key;
2747 struct btrfs_leaf_ref *ref;
2751 path->lowest_level = 1;
2752 key.objectid = inode->i_ino;
2753 key.type = BTRFS_CSUM_ITEM_KEY;
2754 key.offset = new_size;
2756 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2760 if (path->nodes[1] == NULL) {
2765 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2766 nritems = btrfs_header_nritems(path->nodes[1]);
2771 if (path->slots[1] >= nritems)
2774 /* did we find a key greater than anything we want to delete? */
2775 if (found_key.objectid > inode->i_ino ||
2776 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2779 /* we check the next key in the node to make sure the leave contains
2780 * only checksum items. This comparison doesn't work if our
2781 * leaf is the last one in the node
2783 if (path->slots[1] + 1 >= nritems) {
2785 /* search forward from the last key in the node, this
2786 * will bring us into the next node in the tree
2788 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2790 /* unlikely, but we inc below, so check to be safe */
2791 if (found_key.offset == (u64)-1)
2794 /* search_forward needs a path with locks held, do the
2795 * search again for the original key. It is possible
2796 * this will race with a balance and return a path that
2797 * we could modify, but this drop is just an optimization
2798 * and is allowed to miss some leaves.
2800 btrfs_release_path(root, path);
2803 /* setup a max key for search_forward */
2804 other_key.offset = (u64)-1;
2805 other_key.type = key.type;
2806 other_key.objectid = key.objectid;
2808 path->keep_locks = 1;
2809 ret = btrfs_search_forward(root, &found_key, &other_key,
2811 path->keep_locks = 0;
2812 if (ret || found_key.objectid != key.objectid ||
2813 found_key.type != key.type) {
2818 key.offset = found_key.offset;
2819 btrfs_release_path(root, path);
2824 /* we know there's one more slot after us in the tree,
2825 * read that key so we can verify it is also a checksum item
2827 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2829 if (found_key.objectid < inode->i_ino)
2832 if (found_key.type != key.type || found_key.offset < new_size)
2836 * if the key for the next leaf isn't a csum key from this objectid,
2837 * we can't be sure there aren't good items inside this leaf.
2840 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2843 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2844 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2846 * it is safe to delete this leaf, it contains only
2847 * csum items from this inode at an offset >= new_size
2849 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2852 if (root->ref_cows && leaf_gen < trans->transid) {
2853 ref = btrfs_alloc_leaf_ref(root, 0);
2855 ref->root_gen = root->root_key.offset;
2856 ref->bytenr = leaf_start;
2858 ref->generation = leaf_gen;
2861 btrfs_sort_leaf_ref(ref);
2863 ret = btrfs_add_leaf_ref(root, ref, 0);
2865 btrfs_free_leaf_ref(root, ref);
2871 btrfs_release_path(root, path);
2873 if (other_key.objectid == inode->i_ino &&
2874 other_key.type == key.type && other_key.offset > key.offset) {
2875 key.offset = other_key.offset;
2881 /* fixup any changes we've made to the path */
2882 path->lowest_level = 0;
2883 path->keep_locks = 0;
2884 btrfs_release_path(root, path);
2891 * this can truncate away extent items, csum items and directory items.
2892 * It starts at a high offset and removes keys until it can't find
2893 * any higher than new_size
2895 * csum items that cross the new i_size are truncated to the new size
2898 * min_type is the minimum key type to truncate down to. If set to 0, this
2899 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2901 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2902 struct btrfs_root *root,
2903 struct inode *inode,
2904 u64 new_size, u32 min_type)
2906 struct btrfs_path *path;
2907 struct extent_buffer *leaf;
2908 struct btrfs_file_extent_item *fi;
2909 struct btrfs_key key;
2910 struct btrfs_key found_key;
2911 u64 extent_start = 0;
2912 u64 extent_num_bytes = 0;
2913 u64 extent_offset = 0;
2915 u64 mask = root->sectorsize - 1;
2916 u32 found_type = (u8)-1;
2919 int pending_del_nr = 0;
2920 int pending_del_slot = 0;
2921 int extent_type = -1;
2926 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2929 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2931 path = btrfs_alloc_path();
2935 key.objectid = inode->i_ino;
2936 key.offset = (u64)-1;
2940 path->leave_spinning = 1;
2941 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2948 /* there are no items in the tree for us to truncate, we're
2951 if (path->slots[0] == 0)
2958 leaf = path->nodes[0];
2959 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2960 found_type = btrfs_key_type(&found_key);
2963 if (found_key.objectid != inode->i_ino)
2966 if (found_type < min_type)
2969 item_end = found_key.offset;
2970 if (found_type == BTRFS_EXTENT_DATA_KEY) {
2971 fi = btrfs_item_ptr(leaf, path->slots[0],
2972 struct btrfs_file_extent_item);
2973 extent_type = btrfs_file_extent_type(leaf, fi);
2974 encoding = btrfs_file_extent_compression(leaf, fi);
2975 encoding |= btrfs_file_extent_encryption(leaf, fi);
2976 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
2978 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
2980 btrfs_file_extent_num_bytes(leaf, fi);
2981 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
2982 item_end += btrfs_file_extent_inline_len(leaf,
2987 if (found_type > min_type) {
2990 if (item_end < new_size)
2992 if (found_key.offset >= new_size)
2998 /* FIXME, shrink the extent if the ref count is only 1 */
2999 if (found_type != BTRFS_EXTENT_DATA_KEY)
3002 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3004 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3005 if (!del_item && !encoding) {
3006 u64 orig_num_bytes =
3007 btrfs_file_extent_num_bytes(leaf, fi);
3008 extent_num_bytes = new_size -
3009 found_key.offset + root->sectorsize - 1;
3010 extent_num_bytes = extent_num_bytes &
3011 ~((u64)root->sectorsize - 1);
3012 btrfs_set_file_extent_num_bytes(leaf, fi,
3014 num_dec = (orig_num_bytes -
3016 if (root->ref_cows && extent_start != 0)
3017 inode_sub_bytes(inode, num_dec);
3018 btrfs_mark_buffer_dirty(leaf);
3021 btrfs_file_extent_disk_num_bytes(leaf,
3023 extent_offset = found_key.offset -
3024 btrfs_file_extent_offset(leaf, fi);
3026 /* FIXME blocksize != 4096 */
3027 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3028 if (extent_start != 0) {
3031 inode_sub_bytes(inode, num_dec);
3034 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3036 * we can't truncate inline items that have had
3040 btrfs_file_extent_compression(leaf, fi) == 0 &&
3041 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3042 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3043 u32 size = new_size - found_key.offset;
3045 if (root->ref_cows) {
3046 inode_sub_bytes(inode, item_end + 1 -
3050 btrfs_file_extent_calc_inline_size(size);
3051 ret = btrfs_truncate_item(trans, root, path,
3054 } else if (root->ref_cows) {
3055 inode_sub_bytes(inode, item_end + 1 -
3061 if (!pending_del_nr) {
3062 /* no pending yet, add ourselves */
3063 pending_del_slot = path->slots[0];
3065 } else if (pending_del_nr &&
3066 path->slots[0] + 1 == pending_del_slot) {
3067 /* hop on the pending chunk */
3069 pending_del_slot = path->slots[0];
3076 if (found_extent && root->ref_cows) {
3077 btrfs_set_path_blocking(path);
3078 ret = btrfs_free_extent(trans, root, extent_start,
3079 extent_num_bytes, 0,
3080 btrfs_header_owner(leaf),
3081 inode->i_ino, extent_offset);
3085 if (found_type == BTRFS_INODE_ITEM_KEY)
3088 if (path->slots[0] == 0 ||
3089 path->slots[0] != pending_del_slot) {
3090 if (root->ref_cows) {
3094 if (pending_del_nr) {
3095 ret = btrfs_del_items(trans, root, path,
3101 btrfs_release_path(root, path);
3108 if (pending_del_nr) {
3109 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3112 btrfs_free_path(path);
3117 * taken from block_truncate_page, but does cow as it zeros out
3118 * any bytes left in the last page in the file.
3120 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3122 struct inode *inode = mapping->host;
3123 struct btrfs_root *root = BTRFS_I(inode)->root;
3124 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3125 struct btrfs_ordered_extent *ordered;
3127 u32 blocksize = root->sectorsize;
3128 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3129 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3135 if ((offset & (blocksize - 1)) == 0)
3137 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3141 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3147 page = grab_cache_page(mapping, index);
3149 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3150 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3154 page_start = page_offset(page);
3155 page_end = page_start + PAGE_CACHE_SIZE - 1;
3157 if (!PageUptodate(page)) {
3158 ret = btrfs_readpage(NULL, page);
3160 if (page->mapping != mapping) {
3162 page_cache_release(page);
3165 if (!PageUptodate(page)) {
3170 wait_on_page_writeback(page);
3172 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3173 set_page_extent_mapped(page);
3175 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3177 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3179 page_cache_release(page);
3180 btrfs_start_ordered_extent(inode, ordered, 1);
3181 btrfs_put_ordered_extent(ordered);
3185 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
3186 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3189 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
3191 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3196 if (offset != PAGE_CACHE_SIZE) {
3198 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3199 flush_dcache_page(page);
3202 ClearPageChecked(page);
3203 set_page_dirty(page);
3204 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3208 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3209 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3211 page_cache_release(page);
3216 int btrfs_cont_expand(struct inode *inode, loff_t size)
3218 struct btrfs_trans_handle *trans;
3219 struct btrfs_root *root = BTRFS_I(inode)->root;
3220 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3221 struct extent_map *em;
3222 u64 mask = root->sectorsize - 1;
3223 u64 hole_start = (inode->i_size + mask) & ~mask;
3224 u64 block_end = (size + mask) & ~mask;
3230 if (size <= hole_start)
3234 struct btrfs_ordered_extent *ordered;
3235 btrfs_wait_ordered_range(inode, hole_start,
3236 block_end - hole_start);
3237 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3238 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3241 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3242 btrfs_put_ordered_extent(ordered);
3245 cur_offset = hole_start;
3247 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3248 block_end - cur_offset, 0);
3249 BUG_ON(IS_ERR(em) || !em);
3250 last_byte = min(extent_map_end(em), block_end);
3251 last_byte = (last_byte + mask) & ~mask;
3252 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3254 hole_size = last_byte - cur_offset;
3256 err = btrfs_reserve_metadata_space(root, 2);
3260 trans = btrfs_start_transaction(root, 1);
3261 btrfs_set_trans_block_group(trans, inode);
3263 err = btrfs_drop_extents(trans, inode, cur_offset,
3264 cur_offset + hole_size,
3268 err = btrfs_insert_file_extent(trans, root,
3269 inode->i_ino, cur_offset, 0,
3270 0, hole_size, 0, hole_size,
3274 btrfs_drop_extent_cache(inode, hole_start,
3277 btrfs_end_transaction(trans, root);
3278 btrfs_unreserve_metadata_space(root, 2);
3280 free_extent_map(em);
3281 cur_offset = last_byte;
3282 if (cur_offset >= block_end)
3286 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
3290 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3292 struct btrfs_root *root = BTRFS_I(inode)->root;
3293 struct btrfs_trans_handle *trans;
3297 if (attr->ia_size == inode->i_size)
3300 if (attr->ia_size > inode->i_size) {
3301 unsigned long limit;
3302 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3303 if (attr->ia_size > inode->i_sb->s_maxbytes)
3305 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3306 send_sig(SIGXFSZ, current, 0);
3311 ret = btrfs_reserve_metadata_space(root, 1);
3315 trans = btrfs_start_transaction(root, 1);
3316 btrfs_set_trans_block_group(trans, inode);
3318 ret = btrfs_orphan_add(trans, inode);
3321 nr = trans->blocks_used;
3322 btrfs_end_transaction(trans, root);
3323 btrfs_unreserve_metadata_space(root, 1);
3324 btrfs_btree_balance_dirty(root, nr);
3326 if (attr->ia_size > inode->i_size) {
3327 ret = btrfs_cont_expand(inode, attr->ia_size);
3329 btrfs_truncate(inode);
3333 i_size_write(inode, attr->ia_size);
3334 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3336 trans = btrfs_start_transaction(root, 1);
3337 btrfs_set_trans_block_group(trans, inode);
3339 ret = btrfs_update_inode(trans, root, inode);
3341 if (inode->i_nlink > 0) {
3342 ret = btrfs_orphan_del(trans, inode);
3345 nr = trans->blocks_used;
3346 btrfs_end_transaction(trans, root);
3347 btrfs_btree_balance_dirty(root, nr);
3352 * We're truncating a file that used to have good data down to
3353 * zero. Make sure it gets into the ordered flush list so that
3354 * any new writes get down to disk quickly.
3356 if (attr->ia_size == 0)
3357 BTRFS_I(inode)->ordered_data_close = 1;
3359 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3360 ret = vmtruncate(inode, attr->ia_size);
3366 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3368 struct inode *inode = dentry->d_inode;
3371 err = inode_change_ok(inode, attr);
3375 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3376 err = btrfs_setattr_size(inode, attr);
3380 attr->ia_valid &= ~ATTR_SIZE;
3383 err = inode_setattr(inode, attr);
3385 if (!err && ((attr->ia_valid & ATTR_MODE)))
3386 err = btrfs_acl_chmod(inode);
3390 void btrfs_delete_inode(struct inode *inode)
3392 struct btrfs_trans_handle *trans;
3393 struct btrfs_root *root = BTRFS_I(inode)->root;
3397 truncate_inode_pages(&inode->i_data, 0);
3398 if (is_bad_inode(inode)) {
3399 btrfs_orphan_del(NULL, inode);
3402 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3404 if (root->fs_info->log_root_recovering) {
3405 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3409 if (inode->i_nlink > 0) {
3410 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3414 btrfs_i_size_write(inode, 0);
3417 trans = btrfs_start_transaction(root, 1);
3418 btrfs_set_trans_block_group(trans, inode);
3419 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3424 nr = trans->blocks_used;
3425 btrfs_end_transaction(trans, root);
3427 btrfs_btree_balance_dirty(root, nr);
3431 ret = btrfs_orphan_del(trans, inode);
3435 nr = trans->blocks_used;
3436 btrfs_end_transaction(trans, root);
3437 btrfs_btree_balance_dirty(root, nr);
3444 * this returns the key found in the dir entry in the location pointer.
3445 * If no dir entries were found, location->objectid is 0.
3447 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3448 struct btrfs_key *location)
3450 const char *name = dentry->d_name.name;
3451 int namelen = dentry->d_name.len;
3452 struct btrfs_dir_item *di;
3453 struct btrfs_path *path;
3454 struct btrfs_root *root = BTRFS_I(dir)->root;
3457 path = btrfs_alloc_path();
3460 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3465 if (!di || IS_ERR(di))
3468 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3470 btrfs_free_path(path);
3473 location->objectid = 0;
3478 * when we hit a tree root in a directory, the btrfs part of the inode
3479 * needs to be changed to reflect the root directory of the tree root. This
3480 * is kind of like crossing a mount point.
3482 static int fixup_tree_root_location(struct btrfs_root *root,
3484 struct dentry *dentry,
3485 struct btrfs_key *location,
3486 struct btrfs_root **sub_root)
3488 struct btrfs_path *path;
3489 struct btrfs_root *new_root;
3490 struct btrfs_root_ref *ref;
3491 struct extent_buffer *leaf;
3495 path = btrfs_alloc_path();
3502 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3503 BTRFS_I(dir)->root->root_key.objectid,
3504 location->objectid);
3511 leaf = path->nodes[0];
3512 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3513 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3514 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3517 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3518 (unsigned long)(ref + 1),
3519 dentry->d_name.len);
3523 btrfs_release_path(root->fs_info->tree_root, path);
3525 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3526 if (IS_ERR(new_root)) {
3527 err = PTR_ERR(new_root);
3531 if (btrfs_root_refs(&new_root->root_item) == 0) {
3536 *sub_root = new_root;
3537 location->objectid = btrfs_root_dirid(&new_root->root_item);
3538 location->type = BTRFS_INODE_ITEM_KEY;
3539 location->offset = 0;
3542 btrfs_free_path(path);
3546 static void inode_tree_add(struct inode *inode)
3548 struct btrfs_root *root = BTRFS_I(inode)->root;
3549 struct btrfs_inode *entry;
3551 struct rb_node *parent;
3553 p = &root->inode_tree.rb_node;
3556 if (hlist_unhashed(&inode->i_hash))
3559 spin_lock(&root->inode_lock);
3562 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3564 if (inode->i_ino < entry->vfs_inode.i_ino)
3565 p = &parent->rb_left;
3566 else if (inode->i_ino > entry->vfs_inode.i_ino)
3567 p = &parent->rb_right;
3569 WARN_ON(!(entry->vfs_inode.i_state &
3570 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3571 rb_erase(parent, &root->inode_tree);
3572 RB_CLEAR_NODE(parent);
3573 spin_unlock(&root->inode_lock);
3577 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3578 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3579 spin_unlock(&root->inode_lock);
3582 static void inode_tree_del(struct inode *inode)
3584 struct btrfs_root *root = BTRFS_I(inode)->root;
3587 spin_lock(&root->inode_lock);
3588 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3589 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3590 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3591 empty = RB_EMPTY_ROOT(&root->inode_tree);
3593 spin_unlock(&root->inode_lock);
3595 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3596 synchronize_srcu(&root->fs_info->subvol_srcu);
3597 spin_lock(&root->inode_lock);
3598 empty = RB_EMPTY_ROOT(&root->inode_tree);
3599 spin_unlock(&root->inode_lock);
3601 btrfs_add_dead_root(root);
3605 int btrfs_invalidate_inodes(struct btrfs_root *root)
3607 struct rb_node *node;
3608 struct rb_node *prev;
3609 struct btrfs_inode *entry;
3610 struct inode *inode;
3613 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3615 spin_lock(&root->inode_lock);
3617 node = root->inode_tree.rb_node;
3621 entry = rb_entry(node, struct btrfs_inode, rb_node);
3623 if (objectid < entry->vfs_inode.i_ino)
3624 node = node->rb_left;
3625 else if (objectid > entry->vfs_inode.i_ino)
3626 node = node->rb_right;
3632 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3633 if (objectid <= entry->vfs_inode.i_ino) {
3637 prev = rb_next(prev);
3641 entry = rb_entry(node, struct btrfs_inode, rb_node);
3642 objectid = entry->vfs_inode.i_ino + 1;
3643 inode = igrab(&entry->vfs_inode);
3645 spin_unlock(&root->inode_lock);
3646 if (atomic_read(&inode->i_count) > 1)
3647 d_prune_aliases(inode);
3649 * btrfs_drop_inode will remove it from
3650 * the inode cache when its usage count
3655 spin_lock(&root->inode_lock);
3659 if (cond_resched_lock(&root->inode_lock))
3662 node = rb_next(node);
3664 spin_unlock(&root->inode_lock);
3668 static noinline void init_btrfs_i(struct inode *inode)
3670 struct btrfs_inode *bi = BTRFS_I(inode);
3675 bi->last_sub_trans = 0;
3676 bi->logged_trans = 0;
3677 bi->delalloc_bytes = 0;
3678 bi->reserved_bytes = 0;
3679 bi->disk_i_size = 0;
3681 bi->index_cnt = (u64)-1;
3682 bi->last_unlink_trans = 0;
3683 bi->ordered_data_close = 0;
3684 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3685 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3686 inode->i_mapping, GFP_NOFS);
3687 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3688 inode->i_mapping, GFP_NOFS);
3689 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3690 INIT_LIST_HEAD(&BTRFS_I(inode)->ordered_operations);
3691 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3692 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3693 mutex_init(&BTRFS_I(inode)->log_mutex);
3696 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3698 struct btrfs_iget_args *args = p;
3699 inode->i_ino = args->ino;
3700 init_btrfs_i(inode);
3701 BTRFS_I(inode)->root = args->root;
3702 btrfs_set_inode_space_info(args->root, inode);
3706 static int btrfs_find_actor(struct inode *inode, void *opaque)
3708 struct btrfs_iget_args *args = opaque;
3709 return args->ino == inode->i_ino &&
3710 args->root == BTRFS_I(inode)->root;
3713 static struct inode *btrfs_iget_locked(struct super_block *s,
3715 struct btrfs_root *root)
3717 struct inode *inode;
3718 struct btrfs_iget_args args;
3719 args.ino = objectid;
3722 inode = iget5_locked(s, objectid, btrfs_find_actor,
3723 btrfs_init_locked_inode,
3728 /* Get an inode object given its location and corresponding root.
3729 * Returns in *is_new if the inode was read from disk
3731 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3732 struct btrfs_root *root)
3734 struct inode *inode;
3736 inode = btrfs_iget_locked(s, location->objectid, root);
3738 return ERR_PTR(-ENOMEM);
3740 if (inode->i_state & I_NEW) {
3741 BTRFS_I(inode)->root = root;
3742 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3743 btrfs_read_locked_inode(inode);
3745 inode_tree_add(inode);
3746 unlock_new_inode(inode);
3752 static struct inode *new_simple_dir(struct super_block *s,
3753 struct btrfs_key *key,
3754 struct btrfs_root *root)
3756 struct inode *inode = new_inode(s);
3759 return ERR_PTR(-ENOMEM);
3761 init_btrfs_i(inode);
3763 BTRFS_I(inode)->root = root;
3764 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3765 BTRFS_I(inode)->dummy_inode = 1;
3767 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3768 inode->i_op = &simple_dir_inode_operations;
3769 inode->i_fop = &simple_dir_operations;
3770 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3771 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3776 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3778 struct inode *inode;
3779 struct btrfs_root *root = BTRFS_I(dir)->root;
3780 struct btrfs_root *sub_root = root;
3781 struct btrfs_key location;
3785 dentry->d_op = &btrfs_dentry_operations;
3787 if (dentry->d_name.len > BTRFS_NAME_LEN)
3788 return ERR_PTR(-ENAMETOOLONG);
3790 ret = btrfs_inode_by_name(dir, dentry, &location);
3793 return ERR_PTR(ret);
3795 if (location.objectid == 0)
3798 if (location.type == BTRFS_INODE_ITEM_KEY) {
3799 inode = btrfs_iget(dir->i_sb, &location, root);
3803 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3805 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3806 ret = fixup_tree_root_location(root, dir, dentry,
3807 &location, &sub_root);
3810 inode = ERR_PTR(ret);
3812 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3814 inode = btrfs_iget(dir->i_sb, &location, sub_root);
3816 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3818 if (root != sub_root) {
3819 down_read(&root->fs_info->cleanup_work_sem);
3820 if (!(inode->i_sb->s_flags & MS_RDONLY))
3821 btrfs_orphan_cleanup(sub_root);
3822 up_read(&root->fs_info->cleanup_work_sem);
3828 static int btrfs_dentry_delete(struct dentry *dentry)
3830 struct btrfs_root *root;
3832 if (!dentry->d_inode && !IS_ROOT(dentry))
3833 dentry = dentry->d_parent;
3835 if (dentry->d_inode) {
3836 root = BTRFS_I(dentry->d_inode)->root;
3837 if (btrfs_root_refs(&root->root_item) == 0)
3843 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3844 struct nameidata *nd)
3846 struct inode *inode;
3848 inode = btrfs_lookup_dentry(dir, dentry);
3850 return ERR_CAST(inode);
3852 return d_splice_alias(inode, dentry);
3855 static unsigned char btrfs_filetype_table[] = {
3856 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3859 static int btrfs_real_readdir(struct file *filp, void *dirent,
3862 struct inode *inode = filp->f_dentry->d_inode;
3863 struct btrfs_root *root = BTRFS_I(inode)->root;
3864 struct btrfs_item *item;
3865 struct btrfs_dir_item *di;
3866 struct btrfs_key key;
3867 struct btrfs_key found_key;
3868 struct btrfs_path *path;
3871 struct extent_buffer *leaf;
3874 unsigned char d_type;
3879 int key_type = BTRFS_DIR_INDEX_KEY;
3884 /* FIXME, use a real flag for deciding about the key type */
3885 if (root->fs_info->tree_root == root)
3886 key_type = BTRFS_DIR_ITEM_KEY;
3888 /* special case for "." */
3889 if (filp->f_pos == 0) {
3890 over = filldir(dirent, ".", 1,
3897 /* special case for .., just use the back ref */
3898 if (filp->f_pos == 1) {
3899 u64 pino = parent_ino(filp->f_path.dentry);
3900 over = filldir(dirent, "..", 2,
3906 path = btrfs_alloc_path();
3909 btrfs_set_key_type(&key, key_type);
3910 key.offset = filp->f_pos;
3911 key.objectid = inode->i_ino;
3913 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3919 leaf = path->nodes[0];
3920 nritems = btrfs_header_nritems(leaf);
3921 slot = path->slots[0];
3922 if (advance || slot >= nritems) {
3923 if (slot >= nritems - 1) {
3924 ret = btrfs_next_leaf(root, path);
3927 leaf = path->nodes[0];
3928 nritems = btrfs_header_nritems(leaf);
3929 slot = path->slots[0];
3937 item = btrfs_item_nr(leaf, slot);
3938 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3940 if (found_key.objectid != key.objectid)
3942 if (btrfs_key_type(&found_key) != key_type)
3944 if (found_key.offset < filp->f_pos)
3947 filp->f_pos = found_key.offset;
3949 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
3951 di_total = btrfs_item_size(leaf, item);
3953 while (di_cur < di_total) {
3954 struct btrfs_key location;
3956 name_len = btrfs_dir_name_len(leaf, di);
3957 if (name_len <= sizeof(tmp_name)) {
3958 name_ptr = tmp_name;
3960 name_ptr = kmalloc(name_len, GFP_NOFS);
3966 read_extent_buffer(leaf, name_ptr,
3967 (unsigned long)(di + 1), name_len);
3969 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
3970 btrfs_dir_item_key_to_cpu(leaf, di, &location);
3972 /* is this a reference to our own snapshot? If so
3975 if (location.type == BTRFS_ROOT_ITEM_KEY &&
3976 location.objectid == root->root_key.objectid) {
3980 over = filldir(dirent, name_ptr, name_len,
3981 found_key.offset, location.objectid,
3985 if (name_ptr != tmp_name)
3990 di_len = btrfs_dir_name_len(leaf, di) +
3991 btrfs_dir_data_len(leaf, di) + sizeof(*di);
3993 di = (struct btrfs_dir_item *)((char *)di + di_len);
3997 /* Reached end of directory/root. Bump pos past the last item. */
3998 if (key_type == BTRFS_DIR_INDEX_KEY)
4000 * 32-bit glibc will use getdents64, but then strtol -
4001 * so the last number we can serve is this.
4003 filp->f_pos = 0x7fffffff;
4009 btrfs_free_path(path);
4013 int btrfs_write_inode(struct inode *inode, int wait)
4015 struct btrfs_root *root = BTRFS_I(inode)->root;
4016 struct btrfs_trans_handle *trans;
4019 if (root->fs_info->btree_inode == inode)
4023 trans = btrfs_join_transaction(root, 1);
4024 btrfs_set_trans_block_group(trans, inode);
4025 ret = btrfs_commit_transaction(trans, root);
4031 * This is somewhat expensive, updating the tree every time the
4032 * inode changes. But, it is most likely to find the inode in cache.
4033 * FIXME, needs more benchmarking...there are no reasons other than performance
4034 * to keep or drop this code.
4036 void btrfs_dirty_inode(struct inode *inode)
4038 struct btrfs_root *root = BTRFS_I(inode)->root;
4039 struct btrfs_trans_handle *trans;
4041 trans = btrfs_join_transaction(root, 1);
4042 btrfs_set_trans_block_group(trans, inode);
4043 btrfs_update_inode(trans, root, inode);
4044 btrfs_end_transaction(trans, root);
4048 * find the highest existing sequence number in a directory
4049 * and then set the in-memory index_cnt variable to reflect
4050 * free sequence numbers
4052 static int btrfs_set_inode_index_count(struct inode *inode)
4054 struct btrfs_root *root = BTRFS_I(inode)->root;
4055 struct btrfs_key key, found_key;
4056 struct btrfs_path *path;
4057 struct extent_buffer *leaf;
4060 key.objectid = inode->i_ino;
4061 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4062 key.offset = (u64)-1;
4064 path = btrfs_alloc_path();
4068 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4071 /* FIXME: we should be able to handle this */
4077 * MAGIC NUMBER EXPLANATION:
4078 * since we search a directory based on f_pos we have to start at 2
4079 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4080 * else has to start at 2
4082 if (path->slots[0] == 0) {
4083 BTRFS_I(inode)->index_cnt = 2;
4089 leaf = path->nodes[0];
4090 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4092 if (found_key.objectid != inode->i_ino ||
4093 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4094 BTRFS_I(inode)->index_cnt = 2;
4098 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4100 btrfs_free_path(path);
4105 * helper to find a free sequence number in a given directory. This current
4106 * code is very simple, later versions will do smarter things in the btree
4108 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4112 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4113 ret = btrfs_set_inode_index_count(dir);
4118 *index = BTRFS_I(dir)->index_cnt;
4119 BTRFS_I(dir)->index_cnt++;
4124 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4125 struct btrfs_root *root,
4127 const char *name, int name_len,
4128 u64 ref_objectid, u64 objectid,
4129 u64 alloc_hint, int mode, u64 *index)
4131 struct inode *inode;
4132 struct btrfs_inode_item *inode_item;
4133 struct btrfs_key *location;
4134 struct btrfs_path *path;
4135 struct btrfs_inode_ref *ref;
4136 struct btrfs_key key[2];
4142 path = btrfs_alloc_path();
4145 inode = new_inode(root->fs_info->sb);
4147 return ERR_PTR(-ENOMEM);
4150 ret = btrfs_set_inode_index(dir, index);
4153 return ERR_PTR(ret);
4157 * index_cnt is ignored for everything but a dir,
4158 * btrfs_get_inode_index_count has an explanation for the magic
4161 init_btrfs_i(inode);
4162 BTRFS_I(inode)->index_cnt = 2;
4163 BTRFS_I(inode)->root = root;
4164 BTRFS_I(inode)->generation = trans->transid;
4165 btrfs_set_inode_space_info(root, inode);
4171 BTRFS_I(inode)->block_group =
4172 btrfs_find_block_group(root, 0, alloc_hint, owner);
4174 key[0].objectid = objectid;
4175 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4178 key[1].objectid = objectid;
4179 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4180 key[1].offset = ref_objectid;
4182 sizes[0] = sizeof(struct btrfs_inode_item);
4183 sizes[1] = name_len + sizeof(*ref);
4185 path->leave_spinning = 1;
4186 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4190 inode->i_uid = current_fsuid();
4192 if (dir && (dir->i_mode & S_ISGID)) {
4193 inode->i_gid = dir->i_gid;
4197 inode->i_gid = current_fsgid();
4199 inode->i_mode = mode;
4200 inode->i_ino = objectid;
4201 inode_set_bytes(inode, 0);
4202 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4203 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4204 struct btrfs_inode_item);
4205 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4207 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4208 struct btrfs_inode_ref);
4209 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4210 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4211 ptr = (unsigned long)(ref + 1);
4212 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4214 btrfs_mark_buffer_dirty(path->nodes[0]);
4215 btrfs_free_path(path);
4217 location = &BTRFS_I(inode)->location;
4218 location->objectid = objectid;
4219 location->offset = 0;
4220 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4222 btrfs_inherit_iflags(inode, dir);
4224 if ((mode & S_IFREG)) {
4225 if (btrfs_test_opt(root, NODATASUM))
4226 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4227 if (btrfs_test_opt(root, NODATACOW))
4228 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4231 insert_inode_hash(inode);
4232 inode_tree_add(inode);
4236 BTRFS_I(dir)->index_cnt--;
4237 btrfs_free_path(path);
4239 return ERR_PTR(ret);
4242 static inline u8 btrfs_inode_type(struct inode *inode)
4244 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4248 * utility function to add 'inode' into 'parent_inode' with
4249 * a give name and a given sequence number.
4250 * if 'add_backref' is true, also insert a backref from the
4251 * inode to the parent directory.
4253 int btrfs_add_link(struct btrfs_trans_handle *trans,
4254 struct inode *parent_inode, struct inode *inode,
4255 const char *name, int name_len, int add_backref, u64 index)
4258 struct btrfs_key key;
4259 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4261 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4262 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4264 key.objectid = inode->i_ino;
4265 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4269 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4270 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4271 key.objectid, root->root_key.objectid,
4272 parent_inode->i_ino,
4273 index, name, name_len);
4274 } else if (add_backref) {
4275 ret = btrfs_insert_inode_ref(trans, root,
4276 name, name_len, inode->i_ino,
4277 parent_inode->i_ino, index);
4281 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4282 parent_inode->i_ino, &key,
4283 btrfs_inode_type(inode), index);
4286 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4288 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4289 ret = btrfs_update_inode(trans, root, parent_inode);
4294 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4295 struct dentry *dentry, struct inode *inode,
4296 int backref, u64 index)
4298 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4299 inode, dentry->d_name.name,
4300 dentry->d_name.len, backref, index);
4302 d_instantiate(dentry, inode);
4310 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4311 int mode, dev_t rdev)
4313 struct btrfs_trans_handle *trans;
4314 struct btrfs_root *root = BTRFS_I(dir)->root;
4315 struct inode *inode = NULL;
4319 unsigned long nr = 0;
4322 if (!new_valid_dev(rdev))
4326 * 2 for inode item and ref
4328 * 1 for xattr if selinux is on
4330 err = btrfs_reserve_metadata_space(root, 5);
4334 trans = btrfs_start_transaction(root, 1);
4337 btrfs_set_trans_block_group(trans, dir);
4339 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4345 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4347 dentry->d_parent->d_inode->i_ino, objectid,
4348 BTRFS_I(dir)->block_group, mode, &index);
4349 err = PTR_ERR(inode);
4353 err = btrfs_init_inode_security(trans, inode, dir);
4359 btrfs_set_trans_block_group(trans, inode);
4360 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4364 inode->i_op = &btrfs_special_inode_operations;
4365 init_special_inode(inode, inode->i_mode, rdev);
4366 btrfs_update_inode(trans, root, inode);
4368 btrfs_update_inode_block_group(trans, inode);
4369 btrfs_update_inode_block_group(trans, dir);
4371 nr = trans->blocks_used;
4372 btrfs_end_transaction_throttle(trans, root);
4374 btrfs_unreserve_metadata_space(root, 5);
4376 inode_dec_link_count(inode);
4379 btrfs_btree_balance_dirty(root, nr);
4383 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4384 int mode, struct nameidata *nd)
4386 struct btrfs_trans_handle *trans;
4387 struct btrfs_root *root = BTRFS_I(dir)->root;
4388 struct inode *inode = NULL;
4391 unsigned long nr = 0;
4396 * 2 for inode item and ref
4398 * 1 for xattr if selinux is on
4400 err = btrfs_reserve_metadata_space(root, 5);
4404 trans = btrfs_start_transaction(root, 1);
4407 btrfs_set_trans_block_group(trans, dir);
4409 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4415 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4417 dentry->d_parent->d_inode->i_ino,
4418 objectid, BTRFS_I(dir)->block_group, mode,
4420 err = PTR_ERR(inode);
4424 err = btrfs_init_inode_security(trans, inode, dir);
4430 btrfs_set_trans_block_group(trans, inode);
4431 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4435 inode->i_mapping->a_ops = &btrfs_aops;
4436 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4437 inode->i_fop = &btrfs_file_operations;
4438 inode->i_op = &btrfs_file_inode_operations;
4439 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4441 btrfs_update_inode_block_group(trans, inode);
4442 btrfs_update_inode_block_group(trans, dir);
4444 nr = trans->blocks_used;
4445 btrfs_end_transaction_throttle(trans, root);
4447 btrfs_unreserve_metadata_space(root, 5);
4449 inode_dec_link_count(inode);
4452 btrfs_btree_balance_dirty(root, nr);
4456 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4457 struct dentry *dentry)
4459 struct btrfs_trans_handle *trans;
4460 struct btrfs_root *root = BTRFS_I(dir)->root;
4461 struct inode *inode = old_dentry->d_inode;
4463 unsigned long nr = 0;
4467 if (inode->i_nlink == 0)
4470 /* do not allow sys_link's with other subvols of the same device */
4471 if (root->objectid != BTRFS_I(inode)->root->objectid)
4475 * 1 item for inode ref
4476 * 2 items for dir items
4478 err = btrfs_reserve_metadata_space(root, 3);
4482 btrfs_inc_nlink(inode);
4484 err = btrfs_set_inode_index(dir, &index);
4488 trans = btrfs_start_transaction(root, 1);
4490 btrfs_set_trans_block_group(trans, dir);
4491 atomic_inc(&inode->i_count);
4493 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4498 btrfs_update_inode_block_group(trans, dir);
4499 err = btrfs_update_inode(trans, root, inode);
4501 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4504 nr = trans->blocks_used;
4505 btrfs_end_transaction_throttle(trans, root);
4507 btrfs_unreserve_metadata_space(root, 3);
4509 inode_dec_link_count(inode);
4512 btrfs_btree_balance_dirty(root, nr);
4516 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4518 struct inode *inode = NULL;
4519 struct btrfs_trans_handle *trans;
4520 struct btrfs_root *root = BTRFS_I(dir)->root;
4522 int drop_on_err = 0;
4525 unsigned long nr = 1;
4528 * 2 items for inode and ref
4529 * 2 items for dir items
4530 * 1 for xattr if selinux is on
4532 err = btrfs_reserve_metadata_space(root, 5);
4536 trans = btrfs_start_transaction(root, 1);
4541 btrfs_set_trans_block_group(trans, dir);
4543 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
4549 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4551 dentry->d_parent->d_inode->i_ino, objectid,
4552 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4554 if (IS_ERR(inode)) {
4555 err = PTR_ERR(inode);
4561 err = btrfs_init_inode_security(trans, inode, dir);
4565 inode->i_op = &btrfs_dir_inode_operations;
4566 inode->i_fop = &btrfs_dir_file_operations;
4567 btrfs_set_trans_block_group(trans, inode);
4569 btrfs_i_size_write(inode, 0);
4570 err = btrfs_update_inode(trans, root, inode);
4574 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4575 inode, dentry->d_name.name,
4576 dentry->d_name.len, 0, index);
4580 d_instantiate(dentry, inode);
4582 btrfs_update_inode_block_group(trans, inode);
4583 btrfs_update_inode_block_group(trans, dir);
4586 nr = trans->blocks_used;
4587 btrfs_end_transaction_throttle(trans, root);
4590 btrfs_unreserve_metadata_space(root, 5);
4593 btrfs_btree_balance_dirty(root, nr);
4597 /* helper for btfs_get_extent. Given an existing extent in the tree,
4598 * and an extent that you want to insert, deal with overlap and insert
4599 * the new extent into the tree.
4601 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4602 struct extent_map *existing,
4603 struct extent_map *em,
4604 u64 map_start, u64 map_len)
4608 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4609 start_diff = map_start - em->start;
4610 em->start = map_start;
4612 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4613 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4614 em->block_start += start_diff;
4615 em->block_len -= start_diff;
4617 return add_extent_mapping(em_tree, em);
4620 static noinline int uncompress_inline(struct btrfs_path *path,
4621 struct inode *inode, struct page *page,
4622 size_t pg_offset, u64 extent_offset,
4623 struct btrfs_file_extent_item *item)
4626 struct extent_buffer *leaf = path->nodes[0];
4629 unsigned long inline_size;
4632 WARN_ON(pg_offset != 0);
4633 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4634 inline_size = btrfs_file_extent_inline_item_len(leaf,
4635 btrfs_item_nr(leaf, path->slots[0]));
4636 tmp = kmalloc(inline_size, GFP_NOFS);
4637 ptr = btrfs_file_extent_inline_start(item);
4639 read_extent_buffer(leaf, tmp, ptr, inline_size);
4641 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4642 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4643 inline_size, max_size);
4645 char *kaddr = kmap_atomic(page, KM_USER0);
4646 unsigned long copy_size = min_t(u64,
4647 PAGE_CACHE_SIZE - pg_offset,
4648 max_size - extent_offset);
4649 memset(kaddr + pg_offset, 0, copy_size);
4650 kunmap_atomic(kaddr, KM_USER0);
4657 * a bit scary, this does extent mapping from logical file offset to the disk.
4658 * the ugly parts come from merging extents from the disk with the in-ram
4659 * representation. This gets more complex because of the data=ordered code,
4660 * where the in-ram extents might be locked pending data=ordered completion.
4662 * This also copies inline extents directly into the page.
4665 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4666 size_t pg_offset, u64 start, u64 len,
4672 u64 extent_start = 0;
4674 u64 objectid = inode->i_ino;
4676 struct btrfs_path *path = NULL;
4677 struct btrfs_root *root = BTRFS_I(inode)->root;
4678 struct btrfs_file_extent_item *item;
4679 struct extent_buffer *leaf;
4680 struct btrfs_key found_key;
4681 struct extent_map *em = NULL;
4682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4683 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4684 struct btrfs_trans_handle *trans = NULL;
4688 read_lock(&em_tree->lock);
4689 em = lookup_extent_mapping(em_tree, start, len);
4691 em->bdev = root->fs_info->fs_devices->latest_bdev;
4692 read_unlock(&em_tree->lock);
4695 if (em->start > start || em->start + em->len <= start)
4696 free_extent_map(em);
4697 else if (em->block_start == EXTENT_MAP_INLINE && page)
4698 free_extent_map(em);
4702 em = alloc_extent_map(GFP_NOFS);
4707 em->bdev = root->fs_info->fs_devices->latest_bdev;
4708 em->start = EXTENT_MAP_HOLE;
4709 em->orig_start = EXTENT_MAP_HOLE;
4711 em->block_len = (u64)-1;
4714 path = btrfs_alloc_path();
4718 ret = btrfs_lookup_file_extent(trans, root, path,
4719 objectid, start, trans != NULL);
4726 if (path->slots[0] == 0)
4731 leaf = path->nodes[0];
4732 item = btrfs_item_ptr(leaf, path->slots[0],
4733 struct btrfs_file_extent_item);
4734 /* are we inside the extent that was found? */
4735 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4736 found_type = btrfs_key_type(&found_key);
4737 if (found_key.objectid != objectid ||
4738 found_type != BTRFS_EXTENT_DATA_KEY) {
4742 found_type = btrfs_file_extent_type(leaf, item);
4743 extent_start = found_key.offset;
4744 compressed = btrfs_file_extent_compression(leaf, item);
4745 if (found_type == BTRFS_FILE_EXTENT_REG ||
4746 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4747 extent_end = extent_start +
4748 btrfs_file_extent_num_bytes(leaf, item);
4749 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4751 size = btrfs_file_extent_inline_len(leaf, item);
4752 extent_end = (extent_start + size + root->sectorsize - 1) &
4753 ~((u64)root->sectorsize - 1);
4756 if (start >= extent_end) {
4758 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4759 ret = btrfs_next_leaf(root, path);
4766 leaf = path->nodes[0];
4768 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4769 if (found_key.objectid != objectid ||
4770 found_key.type != BTRFS_EXTENT_DATA_KEY)
4772 if (start + len <= found_key.offset)
4775 em->len = found_key.offset - start;
4779 if (found_type == BTRFS_FILE_EXTENT_REG ||
4780 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4781 em->start = extent_start;
4782 em->len = extent_end - extent_start;
4783 em->orig_start = extent_start -
4784 btrfs_file_extent_offset(leaf, item);
4785 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4787 em->block_start = EXTENT_MAP_HOLE;
4791 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4792 em->block_start = bytenr;
4793 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4796 bytenr += btrfs_file_extent_offset(leaf, item);
4797 em->block_start = bytenr;
4798 em->block_len = em->len;
4799 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4800 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4803 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4807 size_t extent_offset;
4810 em->block_start = EXTENT_MAP_INLINE;
4811 if (!page || create) {
4812 em->start = extent_start;
4813 em->len = extent_end - extent_start;
4817 size = btrfs_file_extent_inline_len(leaf, item);
4818 extent_offset = page_offset(page) + pg_offset - extent_start;
4819 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4820 size - extent_offset);
4821 em->start = extent_start + extent_offset;
4822 em->len = (copy_size + root->sectorsize - 1) &
4823 ~((u64)root->sectorsize - 1);
4824 em->orig_start = EXTENT_MAP_INLINE;
4826 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4827 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4828 if (create == 0 && !PageUptodate(page)) {
4829 if (btrfs_file_extent_compression(leaf, item) ==
4830 BTRFS_COMPRESS_ZLIB) {
4831 ret = uncompress_inline(path, inode, page,
4833 extent_offset, item);
4837 read_extent_buffer(leaf, map + pg_offset, ptr,
4839 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4840 memset(map + pg_offset + copy_size, 0,
4841 PAGE_CACHE_SIZE - pg_offset -
4846 flush_dcache_page(page);
4847 } else if (create && PageUptodate(page)) {
4850 free_extent_map(em);
4852 btrfs_release_path(root, path);
4853 trans = btrfs_join_transaction(root, 1);
4857 write_extent_buffer(leaf, map + pg_offset, ptr,
4860 btrfs_mark_buffer_dirty(leaf);
4862 set_extent_uptodate(io_tree, em->start,
4863 extent_map_end(em) - 1, GFP_NOFS);
4866 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4873 em->block_start = EXTENT_MAP_HOLE;
4874 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4876 btrfs_release_path(root, path);
4877 if (em->start > start || extent_map_end(em) <= start) {
4878 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4879 "[%llu %llu]\n", (unsigned long long)em->start,
4880 (unsigned long long)em->len,
4881 (unsigned long long)start,
4882 (unsigned long long)len);
4888 write_lock(&em_tree->lock);
4889 ret = add_extent_mapping(em_tree, em);
4890 /* it is possible that someone inserted the extent into the tree
4891 * while we had the lock dropped. It is also possible that
4892 * an overlapping map exists in the tree
4894 if (ret == -EEXIST) {
4895 struct extent_map *existing;
4899 existing = lookup_extent_mapping(em_tree, start, len);
4900 if (existing && (existing->start > start ||
4901 existing->start + existing->len <= start)) {
4902 free_extent_map(existing);
4906 existing = lookup_extent_mapping(em_tree, em->start,
4909 err = merge_extent_mapping(em_tree, existing,
4912 free_extent_map(existing);
4914 free_extent_map(em);
4919 free_extent_map(em);
4923 free_extent_map(em);
4928 write_unlock(&em_tree->lock);
4931 btrfs_free_path(path);
4933 ret = btrfs_end_transaction(trans, root);
4938 free_extent_map(em);
4939 return ERR_PTR(err);
4944 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4945 const struct iovec *iov, loff_t offset,
4946 unsigned long nr_segs)
4951 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4952 __u64 start, __u64 len)
4954 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4957 int btrfs_readpage(struct file *file, struct page *page)
4959 struct extent_io_tree *tree;
4960 tree = &BTRFS_I(page->mapping->host)->io_tree;
4961 return extent_read_full_page(tree, page, btrfs_get_extent);
4964 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4966 struct extent_io_tree *tree;
4969 if (current->flags & PF_MEMALLOC) {
4970 redirty_page_for_writepage(wbc, page);
4974 tree = &BTRFS_I(page->mapping->host)->io_tree;
4975 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
4978 int btrfs_writepages(struct address_space *mapping,
4979 struct writeback_control *wbc)
4981 struct extent_io_tree *tree;
4983 tree = &BTRFS_I(mapping->host)->io_tree;
4984 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
4988 btrfs_readpages(struct file *file, struct address_space *mapping,
4989 struct list_head *pages, unsigned nr_pages)
4991 struct extent_io_tree *tree;
4992 tree = &BTRFS_I(mapping->host)->io_tree;
4993 return extent_readpages(tree, mapping, pages, nr_pages,
4996 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
4998 struct extent_io_tree *tree;
4999 struct extent_map_tree *map;
5002 tree = &BTRFS_I(page->mapping->host)->io_tree;
5003 map = &BTRFS_I(page->mapping->host)->extent_tree;
5004 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
5006 ClearPagePrivate(page);
5007 set_page_private(page, 0);
5008 page_cache_release(page);
5013 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5015 if (PageWriteback(page) || PageDirty(page))
5017 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5020 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5022 struct extent_io_tree *tree;
5023 struct btrfs_ordered_extent *ordered;
5024 u64 page_start = page_offset(page);
5025 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5029 * we have the page locked, so new writeback can't start,
5030 * and the dirty bit won't be cleared while we are here.
5032 * Wait for IO on this page so that we can safely clear
5033 * the PagePrivate2 bit and do ordered accounting
5035 wait_on_page_writeback(page);
5037 tree = &BTRFS_I(page->mapping->host)->io_tree;
5039 btrfs_releasepage(page, GFP_NOFS);
5042 lock_extent(tree, page_start, page_end, GFP_NOFS);
5043 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5047 * IO on this page will never be started, so we need
5048 * to account for any ordered extents now
5050 clear_extent_bit(tree, page_start, page_end,
5051 EXTENT_DIRTY | EXTENT_DELALLOC |
5052 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5055 * whoever cleared the private bit is responsible
5056 * for the finish_ordered_io
5058 if (TestClearPagePrivate2(page)) {
5059 btrfs_finish_ordered_io(page->mapping->host,
5060 page_start, page_end);
5062 btrfs_put_ordered_extent(ordered);
5063 lock_extent(tree, page_start, page_end, GFP_NOFS);
5065 clear_extent_bit(tree, page_start, page_end,
5066 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5067 EXTENT_DO_ACCOUNTING, 1, 1, NULL, GFP_NOFS);
5068 __btrfs_releasepage(page, GFP_NOFS);
5070 ClearPageChecked(page);
5071 if (PagePrivate(page)) {
5072 ClearPagePrivate(page);
5073 set_page_private(page, 0);
5074 page_cache_release(page);
5079 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5080 * called from a page fault handler when a page is first dirtied. Hence we must
5081 * be careful to check for EOF conditions here. We set the page up correctly
5082 * for a written page which means we get ENOSPC checking when writing into
5083 * holes and correct delalloc and unwritten extent mapping on filesystems that
5084 * support these features.
5086 * We are not allowed to take the i_mutex here so we have to play games to
5087 * protect against truncate races as the page could now be beyond EOF. Because
5088 * vmtruncate() writes the inode size before removing pages, once we have the
5089 * page lock we can determine safely if the page is beyond EOF. If it is not
5090 * beyond EOF, then the page is guaranteed safe against truncation until we
5093 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5095 struct page *page = vmf->page;
5096 struct inode *inode = fdentry(vma->vm_file)->d_inode;
5097 struct btrfs_root *root = BTRFS_I(inode)->root;
5098 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5099 struct btrfs_ordered_extent *ordered;
5101 unsigned long zero_start;
5107 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5111 else /* -ENOSPC, -EIO, etc */
5112 ret = VM_FAULT_SIGBUS;
5116 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5118 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5119 ret = VM_FAULT_SIGBUS;
5123 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5126 size = i_size_read(inode);
5127 page_start = page_offset(page);
5128 page_end = page_start + PAGE_CACHE_SIZE - 1;
5130 if ((page->mapping != inode->i_mapping) ||
5131 (page_start >= size)) {
5132 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5133 /* page got truncated out from underneath us */
5136 wait_on_page_writeback(page);
5138 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
5139 set_page_extent_mapped(page);
5142 * we can't set the delalloc bits if there are pending ordered
5143 * extents. Drop our locks and wait for them to finish
5145 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5147 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5149 btrfs_start_ordered_extent(inode, ordered, 1);
5150 btrfs_put_ordered_extent(ordered);
5155 * XXX - page_mkwrite gets called every time the page is dirtied, even
5156 * if it was already dirty, so for space accounting reasons we need to
5157 * clear any delalloc bits for the range we are fixing to save. There
5158 * is probably a better way to do this, but for now keep consistent with
5159 * prepare_pages in the normal write path.
5161 clear_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
5162 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5165 ret = btrfs_set_extent_delalloc(inode, page_start, page_end);
5167 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5168 ret = VM_FAULT_SIGBUS;
5169 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5174 /* page is wholly or partially inside EOF */
5175 if (page_start + PAGE_CACHE_SIZE > size)
5176 zero_start = size & ~PAGE_CACHE_MASK;
5178 zero_start = PAGE_CACHE_SIZE;
5180 if (zero_start != PAGE_CACHE_SIZE) {
5182 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5183 flush_dcache_page(page);
5186 ClearPageChecked(page);
5187 set_page_dirty(page);
5188 SetPageUptodate(page);
5190 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5191 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5193 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
5196 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5198 return VM_FAULT_LOCKED;
5204 static void btrfs_truncate(struct inode *inode)
5206 struct btrfs_root *root = BTRFS_I(inode)->root;
5208 struct btrfs_trans_handle *trans;
5210 u64 mask = root->sectorsize - 1;
5212 if (!S_ISREG(inode->i_mode)) {
5217 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5221 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5222 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5224 trans = btrfs_start_transaction(root, 1);
5225 btrfs_set_trans_block_group(trans, inode);
5228 * setattr is responsible for setting the ordered_data_close flag,
5229 * but that is only tested during the last file release. That
5230 * could happen well after the next commit, leaving a great big
5231 * window where new writes may get lost if someone chooses to write
5232 * to this file after truncating to zero
5234 * The inode doesn't have any dirty data here, and so if we commit
5235 * this is a noop. If someone immediately starts writing to the inode
5236 * it is very likely we'll catch some of their writes in this
5237 * transaction, and the commit will find this file on the ordered
5238 * data list with good things to send down.
5240 * This is a best effort solution, there is still a window where
5241 * using truncate to replace the contents of the file will
5242 * end up with a zero length file after a crash.
5244 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5245 btrfs_add_ordered_operation(trans, root, inode);
5248 ret = btrfs_truncate_inode_items(trans, root, inode,
5250 BTRFS_EXTENT_DATA_KEY);
5254 ret = btrfs_update_inode(trans, root, inode);
5257 nr = trans->blocks_used;
5258 btrfs_end_transaction(trans, root);
5259 btrfs_btree_balance_dirty(root, nr);
5261 trans = btrfs_start_transaction(root, 1);
5262 btrfs_set_trans_block_group(trans, inode);
5265 if (ret == 0 && inode->i_nlink > 0) {
5266 ret = btrfs_orphan_del(trans, inode);
5270 ret = btrfs_update_inode(trans, root, inode);
5273 nr = trans->blocks_used;
5274 ret = btrfs_end_transaction_throttle(trans, root);
5276 btrfs_btree_balance_dirty(root, nr);
5280 * create a new subvolume directory/inode (helper for the ioctl).
5282 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5283 struct btrfs_root *new_root,
5284 u64 new_dirid, u64 alloc_hint)
5286 struct inode *inode;
5290 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5291 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5293 return PTR_ERR(inode);
5294 inode->i_op = &btrfs_dir_inode_operations;
5295 inode->i_fop = &btrfs_dir_file_operations;
5298 btrfs_i_size_write(inode, 0);
5300 err = btrfs_update_inode(trans, new_root, inode);
5307 /* helper function for file defrag and space balancing. This
5308 * forces readahead on a given range of bytes in an inode
5310 unsigned long btrfs_force_ra(struct address_space *mapping,
5311 struct file_ra_state *ra, struct file *file,
5312 pgoff_t offset, pgoff_t last_index)
5314 pgoff_t req_size = last_index - offset + 1;
5316 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5317 return offset + req_size;
5320 struct inode *btrfs_alloc_inode(struct super_block *sb)
5322 struct btrfs_inode *ei;
5324 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5328 ei->last_sub_trans = 0;
5329 ei->logged_trans = 0;
5330 ei->outstanding_extents = 0;
5331 ei->reserved_extents = 0;
5333 spin_lock_init(&ei->accounting_lock);
5334 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5335 INIT_LIST_HEAD(&ei->i_orphan);
5336 INIT_LIST_HEAD(&ei->ordered_operations);
5337 return &ei->vfs_inode;
5340 void btrfs_destroy_inode(struct inode *inode)
5342 struct btrfs_ordered_extent *ordered;
5343 struct btrfs_root *root = BTRFS_I(inode)->root;
5345 WARN_ON(!list_empty(&inode->i_dentry));
5346 WARN_ON(inode->i_data.nrpages);
5349 * This can happen where we create an inode, but somebody else also
5350 * created the same inode and we need to destroy the one we already
5357 * Make sure we're properly removed from the ordered operation
5361 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5362 spin_lock(&root->fs_info->ordered_extent_lock);
5363 list_del_init(&BTRFS_I(inode)->ordered_operations);
5364 spin_unlock(&root->fs_info->ordered_extent_lock);
5367 spin_lock(&root->list_lock);
5368 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5369 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5371 list_del_init(&BTRFS_I(inode)->i_orphan);
5373 spin_unlock(&root->list_lock);
5376 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5380 printk(KERN_ERR "btrfs found ordered "
5381 "extent %llu %llu on inode cleanup\n",
5382 (unsigned long long)ordered->file_offset,
5383 (unsigned long long)ordered->len);
5384 btrfs_remove_ordered_extent(inode, ordered);
5385 btrfs_put_ordered_extent(ordered);
5386 btrfs_put_ordered_extent(ordered);
5389 inode_tree_del(inode);
5390 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5392 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5395 void btrfs_drop_inode(struct inode *inode)
5397 struct btrfs_root *root = BTRFS_I(inode)->root;
5399 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5400 generic_delete_inode(inode);
5402 generic_drop_inode(inode);
5405 static void init_once(void *foo)
5407 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5409 inode_init_once(&ei->vfs_inode);
5412 void btrfs_destroy_cachep(void)
5414 if (btrfs_inode_cachep)
5415 kmem_cache_destroy(btrfs_inode_cachep);
5416 if (btrfs_trans_handle_cachep)
5417 kmem_cache_destroy(btrfs_trans_handle_cachep);
5418 if (btrfs_transaction_cachep)
5419 kmem_cache_destroy(btrfs_transaction_cachep);
5420 if (btrfs_path_cachep)
5421 kmem_cache_destroy(btrfs_path_cachep);
5424 int btrfs_init_cachep(void)
5426 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5427 sizeof(struct btrfs_inode), 0,
5428 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5429 if (!btrfs_inode_cachep)
5432 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5433 sizeof(struct btrfs_trans_handle), 0,
5434 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5435 if (!btrfs_trans_handle_cachep)
5438 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5439 sizeof(struct btrfs_transaction), 0,
5440 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5441 if (!btrfs_transaction_cachep)
5444 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5445 sizeof(struct btrfs_path), 0,
5446 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5447 if (!btrfs_path_cachep)
5452 btrfs_destroy_cachep();
5456 static int btrfs_getattr(struct vfsmount *mnt,
5457 struct dentry *dentry, struct kstat *stat)
5459 struct inode *inode = dentry->d_inode;
5460 generic_fillattr(inode, stat);
5461 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5462 stat->blksize = PAGE_CACHE_SIZE;
5463 stat->blocks = (inode_get_bytes(inode) +
5464 BTRFS_I(inode)->delalloc_bytes) >> 9;
5468 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5469 struct inode *new_dir, struct dentry *new_dentry)
5471 struct btrfs_trans_handle *trans;
5472 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5473 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5474 struct inode *new_inode = new_dentry->d_inode;
5475 struct inode *old_inode = old_dentry->d_inode;
5476 struct timespec ctime = CURRENT_TIME;
5481 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5484 /* we only allow rename subvolume link between subvolumes */
5485 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5488 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5489 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5492 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5493 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5497 * We want to reserve the absolute worst case amount of items. So if
5498 * both inodes are subvols and we need to unlink them then that would
5499 * require 4 item modifications, but if they are both normal inodes it
5500 * would require 5 item modifications, so we'll assume their normal
5501 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5502 * should cover the worst case number of items we'll modify.
5504 ret = btrfs_reserve_metadata_space(root, 11);
5509 * we're using rename to replace one file with another.
5510 * and the replacement file is large. Start IO on it now so
5511 * we don't add too much work to the end of the transaction
5513 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5514 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5515 filemap_flush(old_inode->i_mapping);
5517 /* close the racy window with snapshot create/destroy ioctl */
5518 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5519 down_read(&root->fs_info->subvol_sem);
5521 trans = btrfs_start_transaction(root, 1);
5522 btrfs_set_trans_block_group(trans, new_dir);
5525 btrfs_record_root_in_trans(trans, dest);
5527 ret = btrfs_set_inode_index(new_dir, &index);
5531 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5532 /* force full log commit if subvolume involved. */
5533 root->fs_info->last_trans_log_full_commit = trans->transid;
5535 ret = btrfs_insert_inode_ref(trans, dest,
5536 new_dentry->d_name.name,
5537 new_dentry->d_name.len,
5539 new_dir->i_ino, index);
5543 * this is an ugly little race, but the rename is required
5544 * to make sure that if we crash, the inode is either at the
5545 * old name or the new one. pinning the log transaction lets
5546 * us make sure we don't allow a log commit to come in after
5547 * we unlink the name but before we add the new name back in.
5549 btrfs_pin_log_trans(root);
5552 * make sure the inode gets flushed if it is replacing
5555 if (new_inode && new_inode->i_size &&
5556 old_inode && S_ISREG(old_inode->i_mode)) {
5557 btrfs_add_ordered_operation(trans, root, old_inode);
5560 old_dir->i_ctime = old_dir->i_mtime = ctime;
5561 new_dir->i_ctime = new_dir->i_mtime = ctime;
5562 old_inode->i_ctime = ctime;
5564 if (old_dentry->d_parent != new_dentry->d_parent)
5565 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5567 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5568 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5569 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5570 old_dentry->d_name.name,
5571 old_dentry->d_name.len);
5573 btrfs_inc_nlink(old_dentry->d_inode);
5574 ret = btrfs_unlink_inode(trans, root, old_dir,
5575 old_dentry->d_inode,
5576 old_dentry->d_name.name,
5577 old_dentry->d_name.len);
5582 new_inode->i_ctime = CURRENT_TIME;
5583 if (unlikely(new_inode->i_ino ==
5584 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5585 root_objectid = BTRFS_I(new_inode)->location.objectid;
5586 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5588 new_dentry->d_name.name,
5589 new_dentry->d_name.len);
5590 BUG_ON(new_inode->i_nlink == 0);
5592 ret = btrfs_unlink_inode(trans, dest, new_dir,
5593 new_dentry->d_inode,
5594 new_dentry->d_name.name,
5595 new_dentry->d_name.len);
5598 if (new_inode->i_nlink == 0) {
5599 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5604 ret = btrfs_add_link(trans, new_dir, old_inode,
5605 new_dentry->d_name.name,
5606 new_dentry->d_name.len, 0, index);
5609 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5610 btrfs_log_new_name(trans, old_inode, old_dir,
5611 new_dentry->d_parent);
5612 btrfs_end_log_trans(root);
5615 btrfs_end_transaction_throttle(trans, root);
5617 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5618 up_read(&root->fs_info->subvol_sem);
5620 btrfs_unreserve_metadata_space(root, 11);
5625 * some fairly slow code that needs optimization. This walks the list
5626 * of all the inodes with pending delalloc and forces them to disk.
5628 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5630 struct list_head *head = &root->fs_info->delalloc_inodes;
5631 struct btrfs_inode *binode;
5632 struct inode *inode;
5634 if (root->fs_info->sb->s_flags & MS_RDONLY)
5637 spin_lock(&root->fs_info->delalloc_lock);
5638 while (!list_empty(head)) {
5639 binode = list_entry(head->next, struct btrfs_inode,
5641 inode = igrab(&binode->vfs_inode);
5643 list_del_init(&binode->delalloc_inodes);
5644 spin_unlock(&root->fs_info->delalloc_lock);
5646 filemap_flush(inode->i_mapping);
5648 btrfs_add_delayed_iput(inode);
5653 spin_lock(&root->fs_info->delalloc_lock);
5655 spin_unlock(&root->fs_info->delalloc_lock);
5657 /* the filemap_flush will queue IO into the worker threads, but
5658 * we have to make sure the IO is actually started and that
5659 * ordered extents get created before we return
5661 atomic_inc(&root->fs_info->async_submit_draining);
5662 while (atomic_read(&root->fs_info->nr_async_submits) ||
5663 atomic_read(&root->fs_info->async_delalloc_pages)) {
5664 wait_event(root->fs_info->async_submit_wait,
5665 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5666 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5668 atomic_dec(&root->fs_info->async_submit_draining);
5672 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5673 const char *symname)
5675 struct btrfs_trans_handle *trans;
5676 struct btrfs_root *root = BTRFS_I(dir)->root;
5677 struct btrfs_path *path;
5678 struct btrfs_key key;
5679 struct inode *inode = NULL;
5687 struct btrfs_file_extent_item *ei;
5688 struct extent_buffer *leaf;
5689 unsigned long nr = 0;
5691 name_len = strlen(symname) + 1;
5692 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5693 return -ENAMETOOLONG;
5696 * 2 items for inode item and ref
5697 * 2 items for dir items
5698 * 1 item for xattr if selinux is on
5700 err = btrfs_reserve_metadata_space(root, 5);
5704 trans = btrfs_start_transaction(root, 1);
5707 btrfs_set_trans_block_group(trans, dir);
5709 err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
5715 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5717 dentry->d_parent->d_inode->i_ino, objectid,
5718 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5720 err = PTR_ERR(inode);
5724 err = btrfs_init_inode_security(trans, inode, dir);
5730 btrfs_set_trans_block_group(trans, inode);
5731 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5735 inode->i_mapping->a_ops = &btrfs_aops;
5736 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5737 inode->i_fop = &btrfs_file_operations;
5738 inode->i_op = &btrfs_file_inode_operations;
5739 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5741 btrfs_update_inode_block_group(trans, inode);
5742 btrfs_update_inode_block_group(trans, dir);
5746 path = btrfs_alloc_path();
5748 key.objectid = inode->i_ino;
5750 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5751 datasize = btrfs_file_extent_calc_inline_size(name_len);
5752 err = btrfs_insert_empty_item(trans, root, path, &key,
5758 leaf = path->nodes[0];
5759 ei = btrfs_item_ptr(leaf, path->slots[0],
5760 struct btrfs_file_extent_item);
5761 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5762 btrfs_set_file_extent_type(leaf, ei,
5763 BTRFS_FILE_EXTENT_INLINE);
5764 btrfs_set_file_extent_encryption(leaf, ei, 0);
5765 btrfs_set_file_extent_compression(leaf, ei, 0);
5766 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5767 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5769 ptr = btrfs_file_extent_inline_start(ei);
5770 write_extent_buffer(leaf, symname, ptr, name_len);
5771 btrfs_mark_buffer_dirty(leaf);
5772 btrfs_free_path(path);
5774 inode->i_op = &btrfs_symlink_inode_operations;
5775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5776 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5777 inode_set_bytes(inode, name_len);
5778 btrfs_i_size_write(inode, name_len - 1);
5779 err = btrfs_update_inode(trans, root, inode);
5784 nr = trans->blocks_used;
5785 btrfs_end_transaction_throttle(trans, root);
5787 btrfs_unreserve_metadata_space(root, 5);
5789 inode_dec_link_count(inode);
5792 btrfs_btree_balance_dirty(root, nr);
5796 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
5797 u64 alloc_hint, int mode, loff_t actual_len)
5799 struct btrfs_trans_handle *trans;
5800 struct btrfs_root *root = BTRFS_I(inode)->root;
5801 struct btrfs_key ins;
5803 u64 cur_offset = start;
5804 u64 num_bytes = end - start;
5808 while (num_bytes > 0) {
5809 alloc_size = min(num_bytes, root->fs_info->max_extent);
5811 trans = btrfs_start_transaction(root, 1);
5813 ret = btrfs_reserve_extent(trans, root, alloc_size,
5814 root->sectorsize, 0, alloc_hint,
5821 ret = btrfs_reserve_metadata_space(root, 3);
5823 btrfs_free_reserved_extent(root, ins.objectid,
5828 ret = insert_reserved_file_extent(trans, inode,
5829 cur_offset, ins.objectid,
5830 ins.offset, ins.offset,
5831 ins.offset, 0, 0, 0,
5832 BTRFS_FILE_EXTENT_PREALLOC);
5834 btrfs_drop_extent_cache(inode, cur_offset,
5835 cur_offset + ins.offset -1, 0);
5837 num_bytes -= ins.offset;
5838 cur_offset += ins.offset;
5839 alloc_hint = ins.objectid + ins.offset;
5841 inode->i_ctime = CURRENT_TIME;
5842 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5843 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5844 cur_offset > inode->i_size) {
5845 if (cur_offset > actual_len)
5846 i_size = actual_len;
5848 i_size = cur_offset;
5849 i_size_write(inode, i_size);
5850 btrfs_ordered_update_i_size(inode, i_size, NULL);
5853 ret = btrfs_update_inode(trans, root, inode);
5856 btrfs_end_transaction(trans, root);
5857 btrfs_unreserve_metadata_space(root, 3);
5862 btrfs_end_transaction(trans, root);
5867 static long btrfs_fallocate(struct inode *inode, int mode,
5868 loff_t offset, loff_t len)
5876 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5877 struct extent_map *em;
5880 alloc_start = offset & ~mask;
5881 alloc_end = (offset + len + mask) & ~mask;
5884 * wait for ordered IO before we have any locks. We'll loop again
5885 * below with the locks held.
5887 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5889 mutex_lock(&inode->i_mutex);
5890 if (alloc_start > inode->i_size) {
5891 ret = btrfs_cont_expand(inode, alloc_start);
5896 ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode,
5897 alloc_end - alloc_start);
5901 locked_end = alloc_end - 1;
5903 struct btrfs_ordered_extent *ordered;
5905 /* the extent lock is ordered inside the running
5908 lock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5910 ordered = btrfs_lookup_first_ordered_extent(inode,
5913 ordered->file_offset + ordered->len > alloc_start &&
5914 ordered->file_offset < alloc_end) {
5915 btrfs_put_ordered_extent(ordered);
5916 unlock_extent(&BTRFS_I(inode)->io_tree,
5917 alloc_start, locked_end, GFP_NOFS);
5919 * we can't wait on the range with the transaction
5920 * running or with the extent lock held
5922 btrfs_wait_ordered_range(inode, alloc_start,
5923 alloc_end - alloc_start);
5926 btrfs_put_ordered_extent(ordered);
5931 cur_offset = alloc_start;
5933 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
5934 alloc_end - cur_offset, 0);
5935 BUG_ON(IS_ERR(em) || !em);
5936 last_byte = min(extent_map_end(em), alloc_end);
5937 last_byte = (last_byte + mask) & ~mask;
5938 if (em->block_start == EXTENT_MAP_HOLE ||
5939 (cur_offset >= inode->i_size &&
5940 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5941 ret = prealloc_file_range(inode,
5942 cur_offset, last_byte,
5943 alloc_hint, mode, offset+len);
5945 free_extent_map(em);
5949 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
5950 alloc_hint = em->block_start;
5951 free_extent_map(em);
5953 cur_offset = last_byte;
5954 if (cur_offset >= alloc_end) {
5959 unlock_extent(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
5962 btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode,
5963 alloc_end - alloc_start);
5965 mutex_unlock(&inode->i_mutex);
5969 static int btrfs_set_page_dirty(struct page *page)
5971 return __set_page_dirty_nobuffers(page);
5974 static int btrfs_permission(struct inode *inode, int mask)
5976 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
5978 return generic_permission(inode, mask, btrfs_check_acl);
5981 static const struct inode_operations btrfs_dir_inode_operations = {
5982 .getattr = btrfs_getattr,
5983 .lookup = btrfs_lookup,
5984 .create = btrfs_create,
5985 .unlink = btrfs_unlink,
5987 .mkdir = btrfs_mkdir,
5988 .rmdir = btrfs_rmdir,
5989 .rename = btrfs_rename,
5990 .symlink = btrfs_symlink,
5991 .setattr = btrfs_setattr,
5992 .mknod = btrfs_mknod,
5993 .setxattr = btrfs_setxattr,
5994 .getxattr = btrfs_getxattr,
5995 .listxattr = btrfs_listxattr,
5996 .removexattr = btrfs_removexattr,
5997 .permission = btrfs_permission,
5999 static const struct inode_operations btrfs_dir_ro_inode_operations = {
6000 .lookup = btrfs_lookup,
6001 .permission = btrfs_permission,
6004 static const struct file_operations btrfs_dir_file_operations = {
6005 .llseek = generic_file_llseek,
6006 .read = generic_read_dir,
6007 .readdir = btrfs_real_readdir,
6008 .unlocked_ioctl = btrfs_ioctl,
6009 #ifdef CONFIG_COMPAT
6010 .compat_ioctl = btrfs_ioctl,
6012 .release = btrfs_release_file,
6013 .fsync = btrfs_sync_file,
6016 static struct extent_io_ops btrfs_extent_io_ops = {
6017 .fill_delalloc = run_delalloc_range,
6018 .submit_bio_hook = btrfs_submit_bio_hook,
6019 .merge_bio_hook = btrfs_merge_bio_hook,
6020 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6021 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6022 .writepage_start_hook = btrfs_writepage_start_hook,
6023 .readpage_io_failed_hook = btrfs_io_failed_hook,
6024 .set_bit_hook = btrfs_set_bit_hook,
6025 .clear_bit_hook = btrfs_clear_bit_hook,
6026 .merge_extent_hook = btrfs_merge_extent_hook,
6027 .split_extent_hook = btrfs_split_extent_hook,
6031 * btrfs doesn't support the bmap operation because swapfiles
6032 * use bmap to make a mapping of extents in the file. They assume
6033 * these extents won't change over the life of the file and they
6034 * use the bmap result to do IO directly to the drive.
6036 * the btrfs bmap call would return logical addresses that aren't
6037 * suitable for IO and they also will change frequently as COW
6038 * operations happen. So, swapfile + btrfs == corruption.
6040 * For now we're avoiding this by dropping bmap.
6042 static const struct address_space_operations btrfs_aops = {
6043 .readpage = btrfs_readpage,
6044 .writepage = btrfs_writepage,
6045 .writepages = btrfs_writepages,
6046 .readpages = btrfs_readpages,
6047 .sync_page = block_sync_page,
6048 .direct_IO = btrfs_direct_IO,
6049 .invalidatepage = btrfs_invalidatepage,
6050 .releasepage = btrfs_releasepage,
6051 .set_page_dirty = btrfs_set_page_dirty,
6052 .error_remove_page = generic_error_remove_page,
6055 static const struct address_space_operations btrfs_symlink_aops = {
6056 .readpage = btrfs_readpage,
6057 .writepage = btrfs_writepage,
6058 .invalidatepage = btrfs_invalidatepage,
6059 .releasepage = btrfs_releasepage,
6062 static const struct inode_operations btrfs_file_inode_operations = {
6063 .truncate = btrfs_truncate,
6064 .getattr = btrfs_getattr,
6065 .setattr = btrfs_setattr,
6066 .setxattr = btrfs_setxattr,
6067 .getxattr = btrfs_getxattr,
6068 .listxattr = btrfs_listxattr,
6069 .removexattr = btrfs_removexattr,
6070 .permission = btrfs_permission,
6071 .fallocate = btrfs_fallocate,
6072 .fiemap = btrfs_fiemap,
6074 static const struct inode_operations btrfs_special_inode_operations = {
6075 .getattr = btrfs_getattr,
6076 .setattr = btrfs_setattr,
6077 .permission = btrfs_permission,
6078 .setxattr = btrfs_setxattr,
6079 .getxattr = btrfs_getxattr,
6080 .listxattr = btrfs_listxattr,
6081 .removexattr = btrfs_removexattr,
6083 static const struct inode_operations btrfs_symlink_inode_operations = {
6084 .readlink = generic_readlink,
6085 .follow_link = page_follow_link_light,
6086 .put_link = page_put_link,
6087 .permission = btrfs_permission,
6088 .setxattr = btrfs_setxattr,
6089 .getxattr = btrfs_getxattr,
6090 .listxattr = btrfs_listxattr,
6091 .removexattr = btrfs_removexattr,
6094 const struct dentry_operations btrfs_dentry_operations = {
6095 .d_delete = btrfs_dentry_delete,