2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
43 #include "transaction.h"
44 #include "btrfs_inode.h"
46 #include "print-tree.h"
48 #include "ordered-data.h"
51 #include "compression.h"
54 struct btrfs_iget_args {
56 struct btrfs_root *root;
59 static const struct inode_operations btrfs_dir_inode_operations;
60 static const struct inode_operations btrfs_symlink_inode_operations;
61 static const struct inode_operations btrfs_dir_ro_inode_operations;
62 static const struct inode_operations btrfs_special_inode_operations;
63 static const struct inode_operations btrfs_file_inode_operations;
64 static const struct address_space_operations btrfs_aops;
65 static const struct address_space_operations btrfs_symlink_aops;
66 static const struct file_operations btrfs_dir_file_operations;
67 static struct extent_io_ops btrfs_extent_io_ops;
69 static struct kmem_cache *btrfs_inode_cachep;
70 struct kmem_cache *btrfs_trans_handle_cachep;
71 struct kmem_cache *btrfs_transaction_cachep;
72 struct kmem_cache *btrfs_path_cachep;
75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
76 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
77 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
78 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
79 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
80 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
81 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
82 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
85 static void btrfs_truncate(struct inode *inode);
86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
87 static noinline int cow_file_range(struct inode *inode,
88 struct page *locked_page,
89 u64 start, u64 end, int *page_started,
90 unsigned long *nr_written, int unlock);
92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
93 struct inode *inode, struct inode *dir)
97 err = btrfs_init_acl(trans, inode, dir);
99 err = btrfs_xattr_security_init(trans, inode, dir);
104 * this does all the hard work for inserting an inline extent into
105 * the btree. The caller should have done a btrfs_drop_extents so that
106 * no overlapping inline items exist in the btree
108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
109 struct btrfs_root *root, struct inode *inode,
110 u64 start, size_t size, size_t compressed_size,
111 struct page **compressed_pages)
113 struct btrfs_key key;
114 struct btrfs_path *path;
115 struct extent_buffer *leaf;
116 struct page *page = NULL;
119 struct btrfs_file_extent_item *ei;
122 size_t cur_size = size;
124 unsigned long offset;
125 int use_compress = 0;
127 if (compressed_size && compressed_pages) {
129 cur_size = compressed_size;
132 path = btrfs_alloc_path();
136 path->leave_spinning = 1;
137 btrfs_set_trans_block_group(trans, inode);
139 key.objectid = inode->i_ino;
141 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
142 datasize = btrfs_file_extent_calc_inline_size(cur_size);
144 inode_add_bytes(inode, size);
145 ret = btrfs_insert_empty_item(trans, root, path, &key,
152 leaf = path->nodes[0];
153 ei = btrfs_item_ptr(leaf, path->slots[0],
154 struct btrfs_file_extent_item);
155 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
156 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
157 btrfs_set_file_extent_encryption(leaf, ei, 0);
158 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
159 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
160 ptr = btrfs_file_extent_inline_start(ei);
165 while (compressed_size > 0) {
166 cpage = compressed_pages[i];
167 cur_size = min_t(unsigned long, compressed_size,
170 kaddr = kmap_atomic(cpage, KM_USER0);
171 write_extent_buffer(leaf, kaddr, ptr, cur_size);
172 kunmap_atomic(kaddr, KM_USER0);
176 compressed_size -= cur_size;
178 btrfs_set_file_extent_compression(leaf, ei,
179 BTRFS_COMPRESS_ZLIB);
181 page = find_get_page(inode->i_mapping,
182 start >> PAGE_CACHE_SHIFT);
183 btrfs_set_file_extent_compression(leaf, ei, 0);
184 kaddr = kmap_atomic(page, KM_USER0);
185 offset = start & (PAGE_CACHE_SIZE - 1);
186 write_extent_buffer(leaf, kaddr + offset, ptr, size);
187 kunmap_atomic(kaddr, KM_USER0);
188 page_cache_release(page);
190 btrfs_mark_buffer_dirty(leaf);
191 btrfs_free_path(path);
194 * we're an inline extent, so nobody can
195 * extend the file past i_size without locking
196 * a page we already have locked.
198 * We must do any isize and inode updates
199 * before we unlock the pages. Otherwise we
200 * could end up racing with unlink.
202 BTRFS_I(inode)->disk_i_size = inode->i_size;
203 btrfs_update_inode(trans, root, inode);
207 btrfs_free_path(path);
213 * conditionally insert an inline extent into the file. This
214 * does the checks required to make sure the data is small enough
215 * to fit as an inline extent.
217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
218 struct btrfs_root *root,
219 struct inode *inode, u64 start, u64 end,
220 size_t compressed_size,
221 struct page **compressed_pages)
223 u64 isize = i_size_read(inode);
224 u64 actual_end = min(end + 1, isize);
225 u64 inline_len = actual_end - start;
226 u64 aligned_end = (end + root->sectorsize - 1) &
227 ~((u64)root->sectorsize - 1);
229 u64 data_len = inline_len;
233 data_len = compressed_size;
236 actual_end >= PAGE_CACHE_SIZE ||
237 data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
239 (actual_end & (root->sectorsize - 1)) == 0) ||
241 data_len > root->fs_info->max_inline) {
245 ret = btrfs_drop_extents(trans, inode, start, aligned_end,
249 if (isize > actual_end)
250 inline_len = min_t(u64, isize, actual_end);
251 ret = insert_inline_extent(trans, root, inode, start,
252 inline_len, compressed_size,
255 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
259 struct async_extent {
264 unsigned long nr_pages;
265 struct list_head list;
270 struct btrfs_root *root;
271 struct page *locked_page;
274 struct list_head extents;
275 struct btrfs_work work;
278 static noinline int add_async_extent(struct async_cow *cow,
279 u64 start, u64 ram_size,
282 unsigned long nr_pages)
284 struct async_extent *async_extent;
286 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
287 async_extent->start = start;
288 async_extent->ram_size = ram_size;
289 async_extent->compressed_size = compressed_size;
290 async_extent->pages = pages;
291 async_extent->nr_pages = nr_pages;
292 list_add_tail(&async_extent->list, &cow->extents);
297 * we create compressed extents in two phases. The first
298 * phase compresses a range of pages that have already been
299 * locked (both pages and state bits are locked).
301 * This is done inside an ordered work queue, and the compression
302 * is spread across many cpus. The actual IO submission is step
303 * two, and the ordered work queue takes care of making sure that
304 * happens in the same order things were put onto the queue by
305 * writepages and friends.
307 * If this code finds it can't get good compression, it puts an
308 * entry onto the work queue to write the uncompressed bytes. This
309 * makes sure that both compressed inodes and uncompressed inodes
310 * are written in the same order that pdflush sent them down.
312 static noinline int compress_file_range(struct inode *inode,
313 struct page *locked_page,
315 struct async_cow *async_cow,
318 struct btrfs_root *root = BTRFS_I(inode)->root;
319 struct btrfs_trans_handle *trans;
323 u64 blocksize = root->sectorsize;
325 u64 isize = i_size_read(inode);
327 struct page **pages = NULL;
328 unsigned long nr_pages;
329 unsigned long nr_pages_ret = 0;
330 unsigned long total_compressed = 0;
331 unsigned long total_in = 0;
332 unsigned long max_compressed = 128 * 1024;
333 unsigned long max_uncompressed = 128 * 1024;
339 actual_end = min_t(u64, isize, end + 1);
342 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
343 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
346 * we don't want to send crud past the end of i_size through
347 * compression, that's just a waste of CPU time. So, if the
348 * end of the file is before the start of our current
349 * requested range of bytes, we bail out to the uncompressed
350 * cleanup code that can deal with all of this.
352 * It isn't really the fastest way to fix things, but this is a
353 * very uncommon corner.
355 if (actual_end <= start)
356 goto cleanup_and_bail_uncompressed;
358 total_compressed = actual_end - start;
360 /* we want to make sure that amount of ram required to uncompress
361 * an extent is reasonable, so we limit the total size in ram
362 * of a compressed extent to 128k. This is a crucial number
363 * because it also controls how easily we can spread reads across
364 * cpus for decompression.
366 * We also want to make sure the amount of IO required to do
367 * a random read is reasonably small, so we limit the size of
368 * a compressed extent to 128k.
370 total_compressed = min(total_compressed, max_uncompressed);
371 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
372 num_bytes = max(blocksize, num_bytes);
373 disk_num_bytes = num_bytes;
378 * we do compression for mount -o compress and when the
379 * inode has not been flagged as nocompress. This flag can
380 * change at any time if we discover bad compression ratios.
382 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
383 (btrfs_test_opt(root, COMPRESS) ||
384 (BTRFS_I(inode)->force_compress))) {
386 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
388 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
389 total_compressed, pages,
390 nr_pages, &nr_pages_ret,
396 unsigned long offset = total_compressed &
397 (PAGE_CACHE_SIZE - 1);
398 struct page *page = pages[nr_pages_ret - 1];
401 /* zero the tail end of the last page, we might be
402 * sending it down to disk
405 kaddr = kmap_atomic(page, KM_USER0);
406 memset(kaddr + offset, 0,
407 PAGE_CACHE_SIZE - offset);
408 kunmap_atomic(kaddr, KM_USER0);
414 trans = btrfs_join_transaction(root, 1);
416 btrfs_set_trans_block_group(trans, inode);
418 /* lets try to make an inline extent */
419 if (ret || total_in < (actual_end - start)) {
420 /* we didn't compress the entire range, try
421 * to make an uncompressed inline extent.
423 ret = cow_file_range_inline(trans, root, inode,
424 start, end, 0, NULL);
426 /* try making a compressed inline extent */
427 ret = cow_file_range_inline(trans, root, inode,
429 total_compressed, pages);
433 * inline extent creation worked, we don't need
434 * to create any more async work items. Unlock
435 * and free up our temp pages.
437 extent_clear_unlock_delalloc(inode,
438 &BTRFS_I(inode)->io_tree,
440 EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
441 EXTENT_CLEAR_DELALLOC |
442 EXTENT_CLEAR_ACCOUNTING |
443 EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
445 btrfs_end_transaction(trans, root);
448 btrfs_end_transaction(trans, root);
453 * we aren't doing an inline extent round the compressed size
454 * up to a block size boundary so the allocator does sane
457 total_compressed = (total_compressed + blocksize - 1) &
461 * one last check to make sure the compression is really a
462 * win, compare the page count read with the blocks on disk
464 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
465 ~(PAGE_CACHE_SIZE - 1);
466 if (total_compressed >= total_in) {
469 disk_num_bytes = total_compressed;
470 num_bytes = total_in;
473 if (!will_compress && pages) {
475 * the compression code ran but failed to make things smaller,
476 * free any pages it allocated and our page pointer array
478 for (i = 0; i < nr_pages_ret; i++) {
479 WARN_ON(pages[i]->mapping);
480 page_cache_release(pages[i]);
484 total_compressed = 0;
487 /* flag the file so we don't compress in the future */
488 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
489 !(BTRFS_I(inode)->force_compress)) {
490 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
496 /* the async work queues will take care of doing actual
497 * allocation on disk for these compressed pages,
498 * and will submit them to the elevator.
500 add_async_extent(async_cow, start, num_bytes,
501 total_compressed, pages, nr_pages_ret);
503 if (start + num_bytes < end && start + num_bytes < actual_end) {
510 cleanup_and_bail_uncompressed:
512 * No compression, but we still need to write the pages in
513 * the file we've been given so far. redirty the locked
514 * page if it corresponds to our extent and set things up
515 * for the async work queue to run cow_file_range to do
516 * the normal delalloc dance
518 if (page_offset(locked_page) >= start &&
519 page_offset(locked_page) <= end) {
520 __set_page_dirty_nobuffers(locked_page);
521 /* unlocked later on in the async handlers */
523 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
531 for (i = 0; i < nr_pages_ret; i++) {
532 WARN_ON(pages[i]->mapping);
533 page_cache_release(pages[i]);
541 * phase two of compressed writeback. This is the ordered portion
542 * of the code, which only gets called in the order the work was
543 * queued. We walk all the async extents created by compress_file_range
544 * and send them down to the disk.
546 static noinline int submit_compressed_extents(struct inode *inode,
547 struct async_cow *async_cow)
549 struct async_extent *async_extent;
551 struct btrfs_trans_handle *trans;
552 struct btrfs_key ins;
553 struct extent_map *em;
554 struct btrfs_root *root = BTRFS_I(inode)->root;
555 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
556 struct extent_io_tree *io_tree;
559 if (list_empty(&async_cow->extents))
563 while (!list_empty(&async_cow->extents)) {
564 async_extent = list_entry(async_cow->extents.next,
565 struct async_extent, list);
566 list_del(&async_extent->list);
568 io_tree = &BTRFS_I(inode)->io_tree;
571 /* did the compression code fall back to uncompressed IO? */
572 if (!async_extent->pages) {
573 int page_started = 0;
574 unsigned long nr_written = 0;
576 lock_extent(io_tree, async_extent->start,
577 async_extent->start +
578 async_extent->ram_size - 1, GFP_NOFS);
580 /* allocate blocks */
581 ret = cow_file_range(inode, async_cow->locked_page,
583 async_extent->start +
584 async_extent->ram_size - 1,
585 &page_started, &nr_written, 0);
588 * if page_started, cow_file_range inserted an
589 * inline extent and took care of all the unlocking
590 * and IO for us. Otherwise, we need to submit
591 * all those pages down to the drive.
593 if (!page_started && !ret)
594 extent_write_locked_range(io_tree,
595 inode, async_extent->start,
596 async_extent->start +
597 async_extent->ram_size - 1,
605 lock_extent(io_tree, async_extent->start,
606 async_extent->start + async_extent->ram_size - 1,
609 trans = btrfs_join_transaction(root, 1);
610 ret = btrfs_reserve_extent(trans, root,
611 async_extent->compressed_size,
612 async_extent->compressed_size,
615 btrfs_end_transaction(trans, root);
619 for (i = 0; i < async_extent->nr_pages; i++) {
620 WARN_ON(async_extent->pages[i]->mapping);
621 page_cache_release(async_extent->pages[i]);
623 kfree(async_extent->pages);
624 async_extent->nr_pages = 0;
625 async_extent->pages = NULL;
626 unlock_extent(io_tree, async_extent->start,
627 async_extent->start +
628 async_extent->ram_size - 1, GFP_NOFS);
633 * here we're doing allocation and writeback of the
636 btrfs_drop_extent_cache(inode, async_extent->start,
637 async_extent->start +
638 async_extent->ram_size - 1, 0);
640 em = alloc_extent_map(GFP_NOFS);
641 em->start = async_extent->start;
642 em->len = async_extent->ram_size;
643 em->orig_start = em->start;
645 em->block_start = ins.objectid;
646 em->block_len = ins.offset;
647 em->bdev = root->fs_info->fs_devices->latest_bdev;
648 set_bit(EXTENT_FLAG_PINNED, &em->flags);
649 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
652 write_lock(&em_tree->lock);
653 ret = add_extent_mapping(em_tree, em);
654 write_unlock(&em_tree->lock);
655 if (ret != -EEXIST) {
659 btrfs_drop_extent_cache(inode, async_extent->start,
660 async_extent->start +
661 async_extent->ram_size - 1, 0);
664 ret = btrfs_add_ordered_extent(inode, async_extent->start,
666 async_extent->ram_size,
668 BTRFS_ORDERED_COMPRESSED);
672 * clear dirty, set writeback and unlock the pages.
674 extent_clear_unlock_delalloc(inode,
675 &BTRFS_I(inode)->io_tree,
677 async_extent->start +
678 async_extent->ram_size - 1,
679 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
680 EXTENT_CLEAR_UNLOCK |
681 EXTENT_CLEAR_DELALLOC |
682 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
684 ret = btrfs_submit_compressed_write(inode,
686 async_extent->ram_size,
688 ins.offset, async_extent->pages,
689 async_extent->nr_pages);
692 alloc_hint = ins.objectid + ins.offset;
701 * when extent_io.c finds a delayed allocation range in the file,
702 * the call backs end up in this code. The basic idea is to
703 * allocate extents on disk for the range, and create ordered data structs
704 * in ram to track those extents.
706 * locked_page is the page that writepage had locked already. We use
707 * it to make sure we don't do extra locks or unlocks.
709 * *page_started is set to one if we unlock locked_page and do everything
710 * required to start IO on it. It may be clean and already done with
713 static noinline int cow_file_range(struct inode *inode,
714 struct page *locked_page,
715 u64 start, u64 end, int *page_started,
716 unsigned long *nr_written,
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct btrfs_trans_handle *trans;
723 unsigned long ram_size;
726 u64 blocksize = root->sectorsize;
728 u64 isize = i_size_read(inode);
729 struct btrfs_key ins;
730 struct extent_map *em;
731 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
734 trans = btrfs_join_transaction(root, 1);
736 btrfs_set_trans_block_group(trans, inode);
738 actual_end = min_t(u64, isize, end + 1);
740 num_bytes = (end - start + blocksize) & ~(blocksize - 1);
741 num_bytes = max(blocksize, num_bytes);
742 disk_num_bytes = num_bytes;
746 /* lets try to make an inline extent */
747 ret = cow_file_range_inline(trans, root, inode,
748 start, end, 0, NULL);
750 extent_clear_unlock_delalloc(inode,
751 &BTRFS_I(inode)->io_tree,
753 EXTENT_CLEAR_UNLOCK_PAGE |
754 EXTENT_CLEAR_UNLOCK |
755 EXTENT_CLEAR_DELALLOC |
756 EXTENT_CLEAR_ACCOUNTING |
758 EXTENT_SET_WRITEBACK |
759 EXTENT_END_WRITEBACK);
761 *nr_written = *nr_written +
762 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
769 BUG_ON(disk_num_bytes >
770 btrfs_super_total_bytes(&root->fs_info->super_copy));
773 read_lock(&BTRFS_I(inode)->extent_tree.lock);
774 em = search_extent_mapping(&BTRFS_I(inode)->extent_tree,
778 * if block start isn't an actual block number then find the
779 * first block in this inode and use that as a hint. If that
780 * block is also bogus then just don't worry about it.
782 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
784 em = search_extent_mapping(em_tree, 0, 0);
785 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
786 alloc_hint = em->block_start;
790 alloc_hint = em->block_start;
794 read_unlock(&BTRFS_I(inode)->extent_tree.lock);
795 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
797 while (disk_num_bytes > 0) {
800 cur_alloc_size = disk_num_bytes;
801 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
802 root->sectorsize, 0, alloc_hint,
806 em = alloc_extent_map(GFP_NOFS);
808 em->orig_start = em->start;
809 ram_size = ins.offset;
810 em->len = ins.offset;
812 em->block_start = ins.objectid;
813 em->block_len = ins.offset;
814 em->bdev = root->fs_info->fs_devices->latest_bdev;
815 set_bit(EXTENT_FLAG_PINNED, &em->flags);
818 write_lock(&em_tree->lock);
819 ret = add_extent_mapping(em_tree, em);
820 write_unlock(&em_tree->lock);
821 if (ret != -EEXIST) {
825 btrfs_drop_extent_cache(inode, start,
826 start + ram_size - 1, 0);
829 cur_alloc_size = ins.offset;
830 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
831 ram_size, cur_alloc_size, 0);
834 if (root->root_key.objectid ==
835 BTRFS_DATA_RELOC_TREE_OBJECTID) {
836 ret = btrfs_reloc_clone_csums(inode, start,
841 if (disk_num_bytes < cur_alloc_size)
844 /* we're not doing compressed IO, don't unlock the first
845 * page (which the caller expects to stay locked), don't
846 * clear any dirty bits and don't set any writeback bits
848 * Do set the Private2 bit so we know this page was properly
849 * setup for writepage
851 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
852 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
855 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
856 start, start + ram_size - 1,
858 disk_num_bytes -= cur_alloc_size;
859 num_bytes -= cur_alloc_size;
860 alloc_hint = ins.objectid + ins.offset;
861 start += cur_alloc_size;
865 btrfs_end_transaction(trans, root);
871 * work queue call back to started compression on a file and pages
873 static noinline void async_cow_start(struct btrfs_work *work)
875 struct async_cow *async_cow;
877 async_cow = container_of(work, struct async_cow, work);
879 compress_file_range(async_cow->inode, async_cow->locked_page,
880 async_cow->start, async_cow->end, async_cow,
883 async_cow->inode = NULL;
887 * work queue call back to submit previously compressed pages
889 static noinline void async_cow_submit(struct btrfs_work *work)
891 struct async_cow *async_cow;
892 struct btrfs_root *root;
893 unsigned long nr_pages;
895 async_cow = container_of(work, struct async_cow, work);
897 root = async_cow->root;
898 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
901 atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
903 if (atomic_read(&root->fs_info->async_delalloc_pages) <
905 waitqueue_active(&root->fs_info->async_submit_wait))
906 wake_up(&root->fs_info->async_submit_wait);
908 if (async_cow->inode)
909 submit_compressed_extents(async_cow->inode, async_cow);
912 static noinline void async_cow_free(struct btrfs_work *work)
914 struct async_cow *async_cow;
915 async_cow = container_of(work, struct async_cow, work);
919 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written)
923 struct async_cow *async_cow;
924 struct btrfs_root *root = BTRFS_I(inode)->root;
925 unsigned long nr_pages;
927 int limit = 10 * 1024 * 1042;
929 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
930 1, 0, NULL, GFP_NOFS);
931 while (start < end) {
932 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
933 async_cow->inode = inode;
934 async_cow->root = root;
935 async_cow->locked_page = locked_page;
936 async_cow->start = start;
938 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
941 cur_end = min(end, start + 512 * 1024 - 1);
943 async_cow->end = cur_end;
944 INIT_LIST_HEAD(&async_cow->extents);
946 async_cow->work.func = async_cow_start;
947 async_cow->work.ordered_func = async_cow_submit;
948 async_cow->work.ordered_free = async_cow_free;
949 async_cow->work.flags = 0;
951 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
953 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
955 btrfs_queue_worker(&root->fs_info->delalloc_workers,
958 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
959 wait_event(root->fs_info->async_submit_wait,
960 (atomic_read(&root->fs_info->async_delalloc_pages) <
964 while (atomic_read(&root->fs_info->async_submit_draining) &&
965 atomic_read(&root->fs_info->async_delalloc_pages)) {
966 wait_event(root->fs_info->async_submit_wait,
967 (atomic_read(&root->fs_info->async_delalloc_pages) ==
971 *nr_written += nr_pages;
978 static noinline int csum_exist_in_range(struct btrfs_root *root,
979 u64 bytenr, u64 num_bytes)
982 struct btrfs_ordered_sum *sums;
985 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
986 bytenr + num_bytes - 1, &list);
987 if (ret == 0 && list_empty(&list))
990 while (!list_empty(&list)) {
991 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
992 list_del(&sums->list);
999 * when nowcow writeback call back. This checks for snapshots or COW copies
1000 * of the extents that exist in the file, and COWs the file as required.
1002 * If no cow copies or snapshots exist, we write directly to the existing
1005 static noinline int run_delalloc_nocow(struct inode *inode,
1006 struct page *locked_page,
1007 u64 start, u64 end, int *page_started, int force,
1008 unsigned long *nr_written)
1010 struct btrfs_root *root = BTRFS_I(inode)->root;
1011 struct btrfs_trans_handle *trans;
1012 struct extent_buffer *leaf;
1013 struct btrfs_path *path;
1014 struct btrfs_file_extent_item *fi;
1015 struct btrfs_key found_key;
1028 path = btrfs_alloc_path();
1030 trans = btrfs_join_transaction(root, 1);
1033 cow_start = (u64)-1;
1036 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1039 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1040 leaf = path->nodes[0];
1041 btrfs_item_key_to_cpu(leaf, &found_key,
1042 path->slots[0] - 1);
1043 if (found_key.objectid == inode->i_ino &&
1044 found_key.type == BTRFS_EXTENT_DATA_KEY)
1049 leaf = path->nodes[0];
1050 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1051 ret = btrfs_next_leaf(root, path);
1056 leaf = path->nodes[0];
1062 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1064 if (found_key.objectid > inode->i_ino ||
1065 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1066 found_key.offset > end)
1069 if (found_key.offset > cur_offset) {
1070 extent_end = found_key.offset;
1075 fi = btrfs_item_ptr(leaf, path->slots[0],
1076 struct btrfs_file_extent_item);
1077 extent_type = btrfs_file_extent_type(leaf, fi);
1079 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1080 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1081 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1082 extent_offset = btrfs_file_extent_offset(leaf, fi);
1083 extent_end = found_key.offset +
1084 btrfs_file_extent_num_bytes(leaf, fi);
1085 if (extent_end <= start) {
1089 if (disk_bytenr == 0)
1091 if (btrfs_file_extent_compression(leaf, fi) ||
1092 btrfs_file_extent_encryption(leaf, fi) ||
1093 btrfs_file_extent_other_encoding(leaf, fi))
1095 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1097 if (btrfs_extent_readonly(root, disk_bytenr))
1099 if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1101 extent_offset, disk_bytenr))
1103 disk_bytenr += extent_offset;
1104 disk_bytenr += cur_offset - found_key.offset;
1105 num_bytes = min(end + 1, extent_end) - cur_offset;
1107 * force cow if csum exists in the range.
1108 * this ensure that csum for a given extent are
1109 * either valid or do not exist.
1111 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1114 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1115 extent_end = found_key.offset +
1116 btrfs_file_extent_inline_len(leaf, fi);
1117 extent_end = ALIGN(extent_end, root->sectorsize);
1122 if (extent_end <= start) {
1127 if (cow_start == (u64)-1)
1128 cow_start = cur_offset;
1129 cur_offset = extent_end;
1130 if (cur_offset > end)
1136 btrfs_release_path(root, path);
1137 if (cow_start != (u64)-1) {
1138 ret = cow_file_range(inode, locked_page, cow_start,
1139 found_key.offset - 1, page_started,
1142 cow_start = (u64)-1;
1145 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1146 struct extent_map *em;
1147 struct extent_map_tree *em_tree;
1148 em_tree = &BTRFS_I(inode)->extent_tree;
1149 em = alloc_extent_map(GFP_NOFS);
1150 em->start = cur_offset;
1151 em->orig_start = em->start;
1152 em->len = num_bytes;
1153 em->block_len = num_bytes;
1154 em->block_start = disk_bytenr;
1155 em->bdev = root->fs_info->fs_devices->latest_bdev;
1156 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1158 write_lock(&em_tree->lock);
1159 ret = add_extent_mapping(em_tree, em);
1160 write_unlock(&em_tree->lock);
1161 if (ret != -EEXIST) {
1162 free_extent_map(em);
1165 btrfs_drop_extent_cache(inode, em->start,
1166 em->start + em->len - 1, 0);
1168 type = BTRFS_ORDERED_PREALLOC;
1170 type = BTRFS_ORDERED_NOCOW;
1173 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1174 num_bytes, num_bytes, type);
1177 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1178 cur_offset, cur_offset + num_bytes - 1,
1179 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1180 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1181 EXTENT_SET_PRIVATE2);
1182 cur_offset = extent_end;
1183 if (cur_offset > end)
1186 btrfs_release_path(root, path);
1188 if (cur_offset <= end && cow_start == (u64)-1)
1189 cow_start = cur_offset;
1190 if (cow_start != (u64)-1) {
1191 ret = cow_file_range(inode, locked_page, cow_start, end,
1192 page_started, nr_written, 1);
1196 ret = btrfs_end_transaction(trans, root);
1198 btrfs_free_path(path);
1203 * extent_io.c call back to do delayed allocation processing
1205 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1206 u64 start, u64 end, int *page_started,
1207 unsigned long *nr_written)
1210 struct btrfs_root *root = BTRFS_I(inode)->root;
1212 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1213 ret = run_delalloc_nocow(inode, locked_page, start, end,
1214 page_started, 1, nr_written);
1215 else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1216 ret = run_delalloc_nocow(inode, locked_page, start, end,
1217 page_started, 0, nr_written);
1218 else if (!btrfs_test_opt(root, COMPRESS) &&
1219 !(BTRFS_I(inode)->force_compress))
1220 ret = cow_file_range(inode, locked_page, start, end,
1221 page_started, nr_written, 1);
1223 ret = cow_file_range_async(inode, locked_page, start, end,
1224 page_started, nr_written);
1228 static int btrfs_split_extent_hook(struct inode *inode,
1229 struct extent_state *orig, u64 split)
1231 if (!(orig->state & EXTENT_DELALLOC))
1234 spin_lock(&BTRFS_I(inode)->accounting_lock);
1235 BTRFS_I(inode)->outstanding_extents++;
1236 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1242 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1243 * extents so we can keep track of new extents that are just merged onto old
1244 * extents, such as when we are doing sequential writes, so we can properly
1245 * account for the metadata space we'll need.
1247 static int btrfs_merge_extent_hook(struct inode *inode,
1248 struct extent_state *new,
1249 struct extent_state *other)
1251 /* not delalloc, ignore it */
1252 if (!(other->state & EXTENT_DELALLOC))
1255 spin_lock(&BTRFS_I(inode)->accounting_lock);
1256 BTRFS_I(inode)->outstanding_extents--;
1257 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1263 * extent_io.c set_bit_hook, used to track delayed allocation
1264 * bytes in this file, and to maintain the list of inodes that
1265 * have pending delalloc work to be done.
1267 static int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
1268 unsigned long old, unsigned long bits)
1272 * set_bit and clear bit hooks normally require _irqsave/restore
1273 * but in this case, we are only testeing for the DELALLOC
1274 * bit, which is only set or cleared with irqs on
1276 if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1277 struct btrfs_root *root = BTRFS_I(inode)->root;
1279 spin_lock(&BTRFS_I(inode)->accounting_lock);
1280 BTRFS_I(inode)->outstanding_extents++;
1281 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1282 btrfs_delalloc_reserve_space(root, inode, end - start + 1);
1284 spin_lock(&root->fs_info->delalloc_lock);
1285 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
1286 root->fs_info->delalloc_bytes += end - start + 1;
1287 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1288 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1289 &root->fs_info->delalloc_inodes);
1291 spin_unlock(&root->fs_info->delalloc_lock);
1297 * extent_io.c clear_bit_hook, see set_bit_hook for why
1299 static int btrfs_clear_bit_hook(struct inode *inode,
1300 struct extent_state *state, unsigned long bits)
1303 * set_bit and clear bit hooks normally require _irqsave/restore
1304 * but in this case, we are only testeing for the DELALLOC
1305 * bit, which is only set or cleared with irqs on
1307 if ((state->state & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
1308 struct btrfs_root *root = BTRFS_I(inode)->root;
1310 if (bits & EXTENT_DO_ACCOUNTING) {
1311 spin_lock(&BTRFS_I(inode)->accounting_lock);
1312 WARN_ON(!BTRFS_I(inode)->outstanding_extents);
1313 BTRFS_I(inode)->outstanding_extents--;
1314 spin_unlock(&BTRFS_I(inode)->accounting_lock);
1315 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
1318 spin_lock(&root->fs_info->delalloc_lock);
1319 if (state->end - state->start + 1 >
1320 root->fs_info->delalloc_bytes) {
1321 printk(KERN_INFO "btrfs warning: delalloc account "
1323 (unsigned long long)
1324 state->end - state->start + 1,
1325 (unsigned long long)
1326 root->fs_info->delalloc_bytes);
1327 btrfs_delalloc_free_space(root, inode, (u64)-1);
1328 root->fs_info->delalloc_bytes = 0;
1329 BTRFS_I(inode)->delalloc_bytes = 0;
1331 btrfs_delalloc_free_space(root, inode,
1334 root->fs_info->delalloc_bytes -= state->end -
1336 BTRFS_I(inode)->delalloc_bytes -= state->end -
1339 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
1340 !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1341 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1343 spin_unlock(&root->fs_info->delalloc_lock);
1349 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1350 * we don't create bios that span stripes or chunks
1352 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1353 size_t size, struct bio *bio,
1354 unsigned long bio_flags)
1356 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1357 struct btrfs_mapping_tree *map_tree;
1358 u64 logical = (u64)bio->bi_sector << 9;
1363 if (bio_flags & EXTENT_BIO_COMPRESSED)
1366 length = bio->bi_size;
1367 map_tree = &root->fs_info->mapping_tree;
1368 map_length = length;
1369 ret = btrfs_map_block(map_tree, READ, logical,
1370 &map_length, NULL, 0);
1372 if (map_length < length + size)
1378 * in order to insert checksums into the metadata in large chunks,
1379 * we wait until bio submission time. All the pages in the bio are
1380 * checksummed and sums are attached onto the ordered extent record.
1382 * At IO completion time the cums attached on the ordered extent record
1383 * are inserted into the btree
1385 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1386 struct bio *bio, int mirror_num,
1387 unsigned long bio_flags)
1389 struct btrfs_root *root = BTRFS_I(inode)->root;
1392 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1398 * in order to insert checksums into the metadata in large chunks,
1399 * we wait until bio submission time. All the pages in the bio are
1400 * checksummed and sums are attached onto the ordered extent record.
1402 * At IO completion time the cums attached on the ordered extent record
1403 * are inserted into the btree
1405 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1406 int mirror_num, unsigned long bio_flags)
1408 struct btrfs_root *root = BTRFS_I(inode)->root;
1409 return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1413 * extent_io.c submission hook. This does the right thing for csum calculation
1414 * on write, or reading the csums from the tree before a read
1416 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1417 int mirror_num, unsigned long bio_flags)
1419 struct btrfs_root *root = BTRFS_I(inode)->root;
1423 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1425 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1428 if (!(rw & (1 << BIO_RW))) {
1429 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1430 return btrfs_submit_compressed_read(inode, bio,
1431 mirror_num, bio_flags);
1432 } else if (!skip_sum)
1433 btrfs_lookup_bio_sums(root, inode, bio, NULL);
1435 } else if (!skip_sum) {
1436 /* csum items have already been cloned */
1437 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1439 /* we're doing a write, do the async checksumming */
1440 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1441 inode, rw, bio, mirror_num,
1442 bio_flags, __btrfs_submit_bio_start,
1443 __btrfs_submit_bio_done);
1447 return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1451 * given a list of ordered sums record them in the inode. This happens
1452 * at IO completion time based on sums calculated at bio submission time.
1454 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1455 struct inode *inode, u64 file_offset,
1456 struct list_head *list)
1458 struct btrfs_ordered_sum *sum;
1460 btrfs_set_trans_block_group(trans, inode);
1462 list_for_each_entry(sum, list, list) {
1463 btrfs_csum_file_blocks(trans,
1464 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1469 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1470 struct extent_state **cached_state)
1472 if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1474 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1475 cached_state, GFP_NOFS);
1478 /* see btrfs_writepage_start_hook for details on why this is required */
1479 struct btrfs_writepage_fixup {
1481 struct btrfs_work work;
1484 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1486 struct btrfs_writepage_fixup *fixup;
1487 struct btrfs_ordered_extent *ordered;
1488 struct extent_state *cached_state = NULL;
1490 struct inode *inode;
1494 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1498 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1499 ClearPageChecked(page);
1503 inode = page->mapping->host;
1504 page_start = page_offset(page);
1505 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1507 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1508 &cached_state, GFP_NOFS);
1510 /* already ordered? We're done */
1511 if (PagePrivate2(page))
1514 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1516 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1517 page_end, &cached_state, GFP_NOFS);
1519 btrfs_start_ordered_extent(inode, ordered, 1);
1523 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1524 ClearPageChecked(page);
1526 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1527 &cached_state, GFP_NOFS);
1530 page_cache_release(page);
1534 * There are a few paths in the higher layers of the kernel that directly
1535 * set the page dirty bit without asking the filesystem if it is a
1536 * good idea. This causes problems because we want to make sure COW
1537 * properly happens and the data=ordered rules are followed.
1539 * In our case any range that doesn't have the ORDERED bit set
1540 * hasn't been properly setup for IO. We kick off an async process
1541 * to fix it up. The async helper will wait for ordered extents, set
1542 * the delalloc bit and make it safe to write the page.
1544 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1546 struct inode *inode = page->mapping->host;
1547 struct btrfs_writepage_fixup *fixup;
1548 struct btrfs_root *root = BTRFS_I(inode)->root;
1550 /* this page is properly in the ordered list */
1551 if (TestClearPagePrivate2(page))
1554 if (PageChecked(page))
1557 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1561 SetPageChecked(page);
1562 page_cache_get(page);
1563 fixup->work.func = btrfs_writepage_fixup_worker;
1565 btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1569 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1570 struct inode *inode, u64 file_pos,
1571 u64 disk_bytenr, u64 disk_num_bytes,
1572 u64 num_bytes, u64 ram_bytes,
1573 u8 compression, u8 encryption,
1574 u16 other_encoding, int extent_type)
1576 struct btrfs_root *root = BTRFS_I(inode)->root;
1577 struct btrfs_file_extent_item *fi;
1578 struct btrfs_path *path;
1579 struct extent_buffer *leaf;
1580 struct btrfs_key ins;
1584 path = btrfs_alloc_path();
1587 path->leave_spinning = 1;
1590 * we may be replacing one extent in the tree with another.
1591 * The new extent is pinned in the extent map, and we don't want
1592 * to drop it from the cache until it is completely in the btree.
1594 * So, tell btrfs_drop_extents to leave this extent in the cache.
1595 * the caller is expected to unpin it and allow it to be merged
1598 ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1602 ins.objectid = inode->i_ino;
1603 ins.offset = file_pos;
1604 ins.type = BTRFS_EXTENT_DATA_KEY;
1605 ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1607 leaf = path->nodes[0];
1608 fi = btrfs_item_ptr(leaf, path->slots[0],
1609 struct btrfs_file_extent_item);
1610 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1611 btrfs_set_file_extent_type(leaf, fi, extent_type);
1612 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1613 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1614 btrfs_set_file_extent_offset(leaf, fi, 0);
1615 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1616 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1617 btrfs_set_file_extent_compression(leaf, fi, compression);
1618 btrfs_set_file_extent_encryption(leaf, fi, encryption);
1619 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1621 btrfs_unlock_up_safe(path, 1);
1622 btrfs_set_lock_blocking(leaf);
1624 btrfs_mark_buffer_dirty(leaf);
1626 inode_add_bytes(inode, num_bytes);
1628 ins.objectid = disk_bytenr;
1629 ins.offset = disk_num_bytes;
1630 ins.type = BTRFS_EXTENT_ITEM_KEY;
1631 ret = btrfs_alloc_reserved_file_extent(trans, root,
1632 root->root_key.objectid,
1633 inode->i_ino, file_pos, &ins);
1635 btrfs_free_path(path);
1641 * helper function for btrfs_finish_ordered_io, this
1642 * just reads in some of the csum leaves to prime them into ram
1643 * before we start the transaction. It limits the amount of btree
1644 * reads required while inside the transaction.
1646 /* as ordered data IO finishes, this gets called so we can finish
1647 * an ordered extent if the range of bytes in the file it covers are
1650 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1652 struct btrfs_root *root = BTRFS_I(inode)->root;
1653 struct btrfs_trans_handle *trans;
1654 struct btrfs_ordered_extent *ordered_extent = NULL;
1655 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1656 struct extent_state *cached_state = NULL;
1660 ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1664 BUG_ON(!ordered_extent);
1666 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1667 BUG_ON(!list_empty(&ordered_extent->list));
1668 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1670 trans = btrfs_join_transaction(root, 1);
1671 ret = btrfs_update_inode(trans, root, inode);
1673 btrfs_end_transaction(trans, root);
1678 lock_extent_bits(io_tree, ordered_extent->file_offset,
1679 ordered_extent->file_offset + ordered_extent->len - 1,
1680 0, &cached_state, GFP_NOFS);
1682 trans = btrfs_join_transaction(root, 1);
1684 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1686 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1688 ret = btrfs_mark_extent_written(trans, inode,
1689 ordered_extent->file_offset,
1690 ordered_extent->file_offset +
1691 ordered_extent->len);
1694 ret = insert_reserved_file_extent(trans, inode,
1695 ordered_extent->file_offset,
1696 ordered_extent->start,
1697 ordered_extent->disk_len,
1698 ordered_extent->len,
1699 ordered_extent->len,
1701 BTRFS_FILE_EXTENT_REG);
1702 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1703 ordered_extent->file_offset,
1704 ordered_extent->len);
1707 unlock_extent_cached(io_tree, ordered_extent->file_offset,
1708 ordered_extent->file_offset +
1709 ordered_extent->len - 1, &cached_state, GFP_NOFS);
1711 add_pending_csums(trans, inode, ordered_extent->file_offset,
1712 &ordered_extent->list);
1714 /* this also removes the ordered extent from the tree */
1715 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716 ret = btrfs_update_inode(trans, root, inode);
1718 btrfs_end_transaction(trans, root);
1721 btrfs_put_ordered_extent(ordered_extent);
1722 /* once for the tree */
1723 btrfs_put_ordered_extent(ordered_extent);
1728 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1729 struct extent_state *state, int uptodate)
1731 ClearPagePrivate2(page);
1732 return btrfs_finish_ordered_io(page->mapping->host, start, end);
1736 * When IO fails, either with EIO or csum verification fails, we
1737 * try other mirrors that might have a good copy of the data. This
1738 * io_failure_record is used to record state as we go through all the
1739 * mirrors. If another mirror has good data, the page is set up to date
1740 * and things continue. If a good mirror can't be found, the original
1741 * bio end_io callback is called to indicate things have failed.
1743 struct io_failure_record {
1748 unsigned long bio_flags;
1752 static int btrfs_io_failed_hook(struct bio *failed_bio,
1753 struct page *page, u64 start, u64 end,
1754 struct extent_state *state)
1756 struct io_failure_record *failrec = NULL;
1758 struct extent_map *em;
1759 struct inode *inode = page->mapping->host;
1760 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1761 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1768 ret = get_state_private(failure_tree, start, &private);
1770 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1773 failrec->start = start;
1774 failrec->len = end - start + 1;
1775 failrec->last_mirror = 0;
1776 failrec->bio_flags = 0;
1778 read_lock(&em_tree->lock);
1779 em = lookup_extent_mapping(em_tree, start, failrec->len);
1780 if (em->start > start || em->start + em->len < start) {
1781 free_extent_map(em);
1784 read_unlock(&em_tree->lock);
1786 if (!em || IS_ERR(em)) {
1790 logical = start - em->start;
1791 logical = em->block_start + logical;
1792 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1793 logical = em->block_start;
1794 failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1796 failrec->logical = logical;
1797 free_extent_map(em);
1798 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1799 EXTENT_DIRTY, GFP_NOFS);
1800 set_state_private(failure_tree, start,
1801 (u64)(unsigned long)failrec);
1803 failrec = (struct io_failure_record *)(unsigned long)private;
1805 num_copies = btrfs_num_copies(
1806 &BTRFS_I(inode)->root->fs_info->mapping_tree,
1807 failrec->logical, failrec->len);
1808 failrec->last_mirror++;
1810 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1811 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1814 if (state && state->start != failrec->start)
1816 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1818 if (!state || failrec->last_mirror > num_copies) {
1819 set_state_private(failure_tree, failrec->start, 0);
1820 clear_extent_bits(failure_tree, failrec->start,
1821 failrec->start + failrec->len - 1,
1822 EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1826 bio = bio_alloc(GFP_NOFS, 1);
1827 bio->bi_private = state;
1828 bio->bi_end_io = failed_bio->bi_end_io;
1829 bio->bi_sector = failrec->logical >> 9;
1830 bio->bi_bdev = failed_bio->bi_bdev;
1833 bio_add_page(bio, page, failrec->len, start - page_offset(page));
1834 if (failed_bio->bi_rw & (1 << BIO_RW))
1839 BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1840 failrec->last_mirror,
1841 failrec->bio_flags);
1846 * each time an IO finishes, we do a fast check in the IO failure tree
1847 * to see if we need to process or clean up an io_failure_record
1849 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1852 u64 private_failure;
1853 struct io_failure_record *failure;
1857 if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1858 (u64)-1, 1, EXTENT_DIRTY)) {
1859 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1860 start, &private_failure);
1862 failure = (struct io_failure_record *)(unsigned long)
1864 set_state_private(&BTRFS_I(inode)->io_failure_tree,
1866 clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1868 failure->start + failure->len - 1,
1869 EXTENT_DIRTY | EXTENT_LOCKED,
1878 * when reads are done, we need to check csums to verify the data is correct
1879 * if there's a match, we allow the bio to finish. If not, we go through
1880 * the io_failure_record routines to find good copies
1882 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1883 struct extent_state *state)
1885 size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1886 struct inode *inode = page->mapping->host;
1887 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1889 u64 private = ~(u32)0;
1891 struct btrfs_root *root = BTRFS_I(inode)->root;
1894 if (PageChecked(page)) {
1895 ClearPageChecked(page);
1899 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1902 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1903 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1904 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1909 if (state && state->start == start) {
1910 private = state->private;
1913 ret = get_state_private(io_tree, start, &private);
1915 kaddr = kmap_atomic(page, KM_USER0);
1919 csum = btrfs_csum_data(root, kaddr + offset, csum, end - start + 1);
1920 btrfs_csum_final(csum, (char *)&csum);
1921 if (csum != private)
1924 kunmap_atomic(kaddr, KM_USER0);
1926 /* if the io failure tree for this inode is non-empty,
1927 * check to see if we've recovered from a failed IO
1929 btrfs_clean_io_failures(inode, start);
1933 if (printk_ratelimit()) {
1934 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1935 "private %llu\n", page->mapping->host->i_ino,
1936 (unsigned long long)start, csum,
1937 (unsigned long long)private);
1939 memset(kaddr + offset, 1, end - start + 1);
1940 flush_dcache_page(page);
1941 kunmap_atomic(kaddr, KM_USER0);
1947 struct delayed_iput {
1948 struct list_head list;
1949 struct inode *inode;
1952 void btrfs_add_delayed_iput(struct inode *inode)
1954 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1955 struct delayed_iput *delayed;
1957 if (atomic_add_unless(&inode->i_count, -1, 1))
1960 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1961 delayed->inode = inode;
1963 spin_lock(&fs_info->delayed_iput_lock);
1964 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1965 spin_unlock(&fs_info->delayed_iput_lock);
1968 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1971 struct btrfs_fs_info *fs_info = root->fs_info;
1972 struct delayed_iput *delayed;
1975 spin_lock(&fs_info->delayed_iput_lock);
1976 empty = list_empty(&fs_info->delayed_iputs);
1977 spin_unlock(&fs_info->delayed_iput_lock);
1981 down_read(&root->fs_info->cleanup_work_sem);
1982 spin_lock(&fs_info->delayed_iput_lock);
1983 list_splice_init(&fs_info->delayed_iputs, &list);
1984 spin_unlock(&fs_info->delayed_iput_lock);
1986 while (!list_empty(&list)) {
1987 delayed = list_entry(list.next, struct delayed_iput, list);
1988 list_del(&delayed->list);
1989 iput(delayed->inode);
1992 up_read(&root->fs_info->cleanup_work_sem);
1996 * This creates an orphan entry for the given inode in case something goes
1997 * wrong in the middle of an unlink/truncate.
1999 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2001 struct btrfs_root *root = BTRFS_I(inode)->root;
2004 spin_lock(&root->list_lock);
2006 /* already on the orphan list, we're good */
2007 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2008 spin_unlock(&root->list_lock);
2012 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2014 spin_unlock(&root->list_lock);
2017 * insert an orphan item to track this unlinked/truncated file
2019 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2025 * We have done the truncate/delete so we can go ahead and remove the orphan
2026 * item for this particular inode.
2028 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2030 struct btrfs_root *root = BTRFS_I(inode)->root;
2033 spin_lock(&root->list_lock);
2035 if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2036 spin_unlock(&root->list_lock);
2040 list_del_init(&BTRFS_I(inode)->i_orphan);
2042 spin_unlock(&root->list_lock);
2046 spin_unlock(&root->list_lock);
2048 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2054 * this cleans up any orphans that may be left on the list from the last use
2057 void btrfs_orphan_cleanup(struct btrfs_root *root)
2059 struct btrfs_path *path;
2060 struct extent_buffer *leaf;
2061 struct btrfs_item *item;
2062 struct btrfs_key key, found_key;
2063 struct btrfs_trans_handle *trans;
2064 struct inode *inode;
2065 int ret = 0, nr_unlink = 0, nr_truncate = 0;
2067 if (!xchg(&root->clean_orphans, 0))
2070 path = btrfs_alloc_path();
2074 key.objectid = BTRFS_ORPHAN_OBJECTID;
2075 btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2076 key.offset = (u64)-1;
2079 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2081 printk(KERN_ERR "Error searching slot for orphan: %d"
2087 * if ret == 0 means we found what we were searching for, which
2088 * is weird, but possible, so only screw with path if we didnt
2089 * find the key and see if we have stuff that matches
2092 if (path->slots[0] == 0)
2097 /* pull out the item */
2098 leaf = path->nodes[0];
2099 item = btrfs_item_nr(leaf, path->slots[0]);
2100 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2102 /* make sure the item matches what we want */
2103 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2105 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2108 /* release the path since we're done with it */
2109 btrfs_release_path(root, path);
2112 * this is where we are basically btrfs_lookup, without the
2113 * crossing root thing. we store the inode number in the
2114 * offset of the orphan item.
2116 found_key.objectid = found_key.offset;
2117 found_key.type = BTRFS_INODE_ITEM_KEY;
2118 found_key.offset = 0;
2119 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2124 * add this inode to the orphan list so btrfs_orphan_del does
2125 * the proper thing when we hit it
2127 spin_lock(&root->list_lock);
2128 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2129 spin_unlock(&root->list_lock);
2132 * if this is a bad inode, means we actually succeeded in
2133 * removing the inode, but not the orphan record, which means
2134 * we need to manually delete the orphan since iput will just
2135 * do a destroy_inode
2137 if (is_bad_inode(inode)) {
2138 trans = btrfs_start_transaction(root, 0);
2139 btrfs_orphan_del(trans, inode);
2140 btrfs_end_transaction(trans, root);
2145 /* if we have links, this was a truncate, lets do that */
2146 if (inode->i_nlink) {
2148 btrfs_truncate(inode);
2153 /* this will do delete_inode and everything for us */
2158 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2160 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2162 btrfs_free_path(path);
2166 * very simple check to peek ahead in the leaf looking for xattrs. If we
2167 * don't find any xattrs, we know there can't be any acls.
2169 * slot is the slot the inode is in, objectid is the objectid of the inode
2171 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2172 int slot, u64 objectid)
2174 u32 nritems = btrfs_header_nritems(leaf);
2175 struct btrfs_key found_key;
2179 while (slot < nritems) {
2180 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2182 /* we found a different objectid, there must not be acls */
2183 if (found_key.objectid != objectid)
2186 /* we found an xattr, assume we've got an acl */
2187 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2191 * we found a key greater than an xattr key, there can't
2192 * be any acls later on
2194 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2201 * it goes inode, inode backrefs, xattrs, extents,
2202 * so if there are a ton of hard links to an inode there can
2203 * be a lot of backrefs. Don't waste time searching too hard,
2204 * this is just an optimization
2209 /* we hit the end of the leaf before we found an xattr or
2210 * something larger than an xattr. We have to assume the inode
2217 * read an inode from the btree into the in-memory inode
2219 static void btrfs_read_locked_inode(struct inode *inode)
2221 struct btrfs_path *path;
2222 struct extent_buffer *leaf;
2223 struct btrfs_inode_item *inode_item;
2224 struct btrfs_timespec *tspec;
2225 struct btrfs_root *root = BTRFS_I(inode)->root;
2226 struct btrfs_key location;
2228 u64 alloc_group_block;
2232 path = btrfs_alloc_path();
2234 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2236 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2240 leaf = path->nodes[0];
2241 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2242 struct btrfs_inode_item);
2244 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2245 inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2246 inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2247 inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2248 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2250 tspec = btrfs_inode_atime(inode_item);
2251 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2252 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2254 tspec = btrfs_inode_mtime(inode_item);
2255 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2256 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2258 tspec = btrfs_inode_ctime(inode_item);
2259 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2260 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2262 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2263 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2264 BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2265 inode->i_generation = BTRFS_I(inode)->generation;
2267 rdev = btrfs_inode_rdev(leaf, inode_item);
2269 BTRFS_I(inode)->index_cnt = (u64)-1;
2270 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2272 alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2275 * try to precache a NULL acl entry for files that don't have
2276 * any xattrs or acls
2278 maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2280 cache_no_acl(inode);
2282 BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2283 alloc_group_block, 0);
2284 btrfs_free_path(path);
2287 switch (inode->i_mode & S_IFMT) {
2289 inode->i_mapping->a_ops = &btrfs_aops;
2290 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2291 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2292 inode->i_fop = &btrfs_file_operations;
2293 inode->i_op = &btrfs_file_inode_operations;
2296 inode->i_fop = &btrfs_dir_file_operations;
2297 if (root == root->fs_info->tree_root)
2298 inode->i_op = &btrfs_dir_ro_inode_operations;
2300 inode->i_op = &btrfs_dir_inode_operations;
2303 inode->i_op = &btrfs_symlink_inode_operations;
2304 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2305 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2308 inode->i_op = &btrfs_special_inode_operations;
2309 init_special_inode(inode, inode->i_mode, rdev);
2313 btrfs_update_iflags(inode);
2317 btrfs_free_path(path);
2318 make_bad_inode(inode);
2322 * given a leaf and an inode, copy the inode fields into the leaf
2324 static void fill_inode_item(struct btrfs_trans_handle *trans,
2325 struct extent_buffer *leaf,
2326 struct btrfs_inode_item *item,
2327 struct inode *inode)
2329 btrfs_set_inode_uid(leaf, item, inode->i_uid);
2330 btrfs_set_inode_gid(leaf, item, inode->i_gid);
2331 btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2332 btrfs_set_inode_mode(leaf, item, inode->i_mode);
2333 btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2335 btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2336 inode->i_atime.tv_sec);
2337 btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2338 inode->i_atime.tv_nsec);
2340 btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2341 inode->i_mtime.tv_sec);
2342 btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2343 inode->i_mtime.tv_nsec);
2345 btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2346 inode->i_ctime.tv_sec);
2347 btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2348 inode->i_ctime.tv_nsec);
2350 btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2351 btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2352 btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2353 btrfs_set_inode_transid(leaf, item, trans->transid);
2354 btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2355 btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2356 btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2360 * copy everything in the in-memory inode into the btree.
2362 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *root, struct inode *inode)
2365 struct btrfs_inode_item *inode_item;
2366 struct btrfs_path *path;
2367 struct extent_buffer *leaf;
2370 path = btrfs_alloc_path();
2372 path->leave_spinning = 1;
2373 ret = btrfs_lookup_inode(trans, root, path,
2374 &BTRFS_I(inode)->location, 1);
2381 btrfs_unlock_up_safe(path, 1);
2382 leaf = path->nodes[0];
2383 inode_item = btrfs_item_ptr(leaf, path->slots[0],
2384 struct btrfs_inode_item);
2386 fill_inode_item(trans, leaf, inode_item, inode);
2387 btrfs_mark_buffer_dirty(leaf);
2388 btrfs_set_inode_last_trans(trans, inode);
2391 btrfs_free_path(path);
2397 * unlink helper that gets used here in inode.c and in the tree logging
2398 * recovery code. It remove a link in a directory with a given name, and
2399 * also drops the back refs in the inode to the directory
2401 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 struct inode *dir, struct inode *inode,
2404 const char *name, int name_len)
2406 struct btrfs_path *path;
2408 struct extent_buffer *leaf;
2409 struct btrfs_dir_item *di;
2410 struct btrfs_key key;
2413 path = btrfs_alloc_path();
2419 path->leave_spinning = 1;
2420 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2421 name, name_len, -1);
2430 leaf = path->nodes[0];
2431 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2432 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2435 btrfs_release_path(root, path);
2437 ret = btrfs_del_inode_ref(trans, root, name, name_len,
2439 dir->i_ino, &index);
2441 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2442 "inode %lu parent %lu\n", name_len, name,
2443 inode->i_ino, dir->i_ino);
2447 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2448 index, name, name_len, -1);
2457 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2458 btrfs_release_path(root, path);
2460 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2462 BUG_ON(ret != 0 && ret != -ENOENT);
2464 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2468 btrfs_free_path(path);
2472 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2473 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2474 btrfs_update_inode(trans, root, dir);
2475 btrfs_drop_nlink(inode);
2476 ret = btrfs_update_inode(trans, root, inode);
2481 /* helper to check if there is any shared block in the path */
2482 static int check_path_shared(struct btrfs_root *root,
2483 struct btrfs_path *path)
2485 struct extent_buffer *eb;
2490 for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2491 if (!path->nodes[level])
2493 eb = path->nodes[level];
2494 if (!btrfs_block_can_be_shared(root, eb))
2496 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2505 * helper to start transaction for unlink and rmdir.
2507 * unlink and rmdir are special in btrfs, they do not always free space.
2508 * so in enospc case, we should make sure they will free space before
2509 * allowing them to use the global metadata reservation.
2511 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2512 struct dentry *dentry)
2514 struct btrfs_trans_handle *trans;
2515 struct btrfs_root *root = BTRFS_I(dir)->root;
2516 struct btrfs_path *path;
2517 struct btrfs_inode_ref *ref;
2518 struct btrfs_dir_item *di;
2519 struct inode *inode = dentry->d_inode;
2525 trans = btrfs_start_transaction(root, 10);
2526 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2529 if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2530 return ERR_PTR(-ENOSPC);
2532 /* check if there is someone else holds reference */
2533 if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2534 return ERR_PTR(-ENOSPC);
2536 if (atomic_read(&inode->i_count) > 2)
2537 return ERR_PTR(-ENOSPC);
2539 if (xchg(&root->fs_info->enospc_unlink, 1))
2540 return ERR_PTR(-ENOSPC);
2542 path = btrfs_alloc_path();
2544 root->fs_info->enospc_unlink = 0;
2545 return ERR_PTR(-ENOMEM);
2548 trans = btrfs_start_transaction(root, 0);
2549 if (IS_ERR(trans)) {
2550 btrfs_free_path(path);
2551 root->fs_info->enospc_unlink = 0;
2555 path->skip_locking = 1;
2556 path->search_commit_root = 1;
2558 ret = btrfs_lookup_inode(trans, root, path,
2559 &BTRFS_I(dir)->location, 0);
2565 if (check_path_shared(root, path))
2570 btrfs_release_path(root, path);
2572 ret = btrfs_lookup_inode(trans, root, path,
2573 &BTRFS_I(inode)->location, 0);
2579 if (check_path_shared(root, path))
2584 btrfs_release_path(root, path);
2586 if (ret == 0 && S_ISREG(inode->i_mode)) {
2587 ret = btrfs_lookup_file_extent(trans, root, path,
2588 inode->i_ino, (u64)-1, 0);
2594 if (check_path_shared(root, path))
2596 btrfs_release_path(root, path);
2604 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2605 dentry->d_name.name, dentry->d_name.len, 0);
2611 if (check_path_shared(root, path))
2617 btrfs_release_path(root, path);
2619 ref = btrfs_lookup_inode_ref(trans, root, path,
2620 dentry->d_name.name, dentry->d_name.len,
2621 inode->i_ino, dir->i_ino, 0);
2627 if (check_path_shared(root, path))
2629 index = btrfs_inode_ref_index(path->nodes[0], ref);
2630 btrfs_release_path(root, path);
2632 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2633 dentry->d_name.name, dentry->d_name.len, 0);
2638 BUG_ON(ret == -ENOENT);
2639 if (check_path_shared(root, path))
2644 btrfs_free_path(path);
2646 btrfs_end_transaction(trans, root);
2647 root->fs_info->enospc_unlink = 0;
2648 return ERR_PTR(err);
2651 trans->block_rsv = &root->fs_info->global_block_rsv;
2655 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2656 struct btrfs_root *root)
2658 if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2659 BUG_ON(!root->fs_info->enospc_unlink);
2660 root->fs_info->enospc_unlink = 0;
2662 btrfs_end_transaction_throttle(trans, root);
2665 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2667 struct btrfs_root *root = BTRFS_I(dir)->root;
2668 struct btrfs_trans_handle *trans;
2669 struct inode *inode = dentry->d_inode;
2671 unsigned long nr = 0;
2673 trans = __unlink_start_trans(dir, dentry);
2675 return PTR_ERR(trans);
2677 btrfs_set_trans_block_group(trans, dir);
2679 btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2681 ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2682 dentry->d_name.name, dentry->d_name.len);
2685 if (inode->i_nlink == 0) {
2686 ret = btrfs_orphan_add(trans, inode);
2690 nr = trans->blocks_used;
2691 __unlink_end_trans(trans, root);
2692 btrfs_btree_balance_dirty(root, nr);
2696 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2697 struct btrfs_root *root,
2698 struct inode *dir, u64 objectid,
2699 const char *name, int name_len)
2701 struct btrfs_path *path;
2702 struct extent_buffer *leaf;
2703 struct btrfs_dir_item *di;
2704 struct btrfs_key key;
2708 path = btrfs_alloc_path();
2712 di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2713 name, name_len, -1);
2714 BUG_ON(!di || IS_ERR(di));
2716 leaf = path->nodes[0];
2717 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2718 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2719 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2721 btrfs_release_path(root, path);
2723 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2724 objectid, root->root_key.objectid,
2725 dir->i_ino, &index, name, name_len);
2727 BUG_ON(ret != -ENOENT);
2728 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2730 BUG_ON(!di || IS_ERR(di));
2732 leaf = path->nodes[0];
2733 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2734 btrfs_release_path(root, path);
2738 di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2739 index, name, name_len, -1);
2740 BUG_ON(!di || IS_ERR(di));
2742 leaf = path->nodes[0];
2743 btrfs_dir_item_key_to_cpu(leaf, di, &key);
2744 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2745 ret = btrfs_delete_one_dir_name(trans, root, path, di);
2747 btrfs_release_path(root, path);
2749 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2750 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2751 ret = btrfs_update_inode(trans, root, dir);
2753 dir->i_sb->s_dirt = 1;
2755 btrfs_free_path(path);
2759 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2761 struct inode *inode = dentry->d_inode;
2763 struct btrfs_root *root = BTRFS_I(dir)->root;
2764 struct btrfs_trans_handle *trans;
2765 unsigned long nr = 0;
2767 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2768 inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2771 trans = __unlink_start_trans(dir, dentry);
2773 return PTR_ERR(trans);
2775 btrfs_set_trans_block_group(trans, dir);
2777 if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2778 err = btrfs_unlink_subvol(trans, root, dir,
2779 BTRFS_I(inode)->location.objectid,
2780 dentry->d_name.name,
2781 dentry->d_name.len);
2785 err = btrfs_orphan_add(trans, inode);
2789 /* now the directory is empty */
2790 err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2791 dentry->d_name.name, dentry->d_name.len);
2793 btrfs_i_size_write(inode, 0);
2795 nr = trans->blocks_used;
2796 __unlink_end_trans(trans, root);
2797 btrfs_btree_balance_dirty(root, nr);
2804 * when truncating bytes in a file, it is possible to avoid reading
2805 * the leaves that contain only checksum items. This can be the
2806 * majority of the IO required to delete a large file, but it must
2807 * be done carefully.
2809 * The keys in the level just above the leaves are checked to make sure
2810 * the lowest key in a given leaf is a csum key, and starts at an offset
2811 * after the new size.
2813 * Then the key for the next leaf is checked to make sure it also has
2814 * a checksum item for the same file. If it does, we know our target leaf
2815 * contains only checksum items, and it can be safely freed without reading
2818 * This is just an optimization targeted at large files. It may do
2819 * nothing. It will return 0 unless things went badly.
2821 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
2822 struct btrfs_root *root,
2823 struct btrfs_path *path,
2824 struct inode *inode, u64 new_size)
2826 struct btrfs_key key;
2829 struct btrfs_key found_key;
2830 struct btrfs_key other_key;
2831 struct btrfs_leaf_ref *ref;
2835 path->lowest_level = 1;
2836 key.objectid = inode->i_ino;
2837 key.type = BTRFS_CSUM_ITEM_KEY;
2838 key.offset = new_size;
2840 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2844 if (path->nodes[1] == NULL) {
2849 btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
2850 nritems = btrfs_header_nritems(path->nodes[1]);
2855 if (path->slots[1] >= nritems)
2858 /* did we find a key greater than anything we want to delete? */
2859 if (found_key.objectid > inode->i_ino ||
2860 (found_key.objectid == inode->i_ino && found_key.type > key.type))
2863 /* we check the next key in the node to make sure the leave contains
2864 * only checksum items. This comparison doesn't work if our
2865 * leaf is the last one in the node
2867 if (path->slots[1] + 1 >= nritems) {
2869 /* search forward from the last key in the node, this
2870 * will bring us into the next node in the tree
2872 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
2874 /* unlikely, but we inc below, so check to be safe */
2875 if (found_key.offset == (u64)-1)
2878 /* search_forward needs a path with locks held, do the
2879 * search again for the original key. It is possible
2880 * this will race with a balance and return a path that
2881 * we could modify, but this drop is just an optimization
2882 * and is allowed to miss some leaves.
2884 btrfs_release_path(root, path);
2887 /* setup a max key for search_forward */
2888 other_key.offset = (u64)-1;
2889 other_key.type = key.type;
2890 other_key.objectid = key.objectid;
2892 path->keep_locks = 1;
2893 ret = btrfs_search_forward(root, &found_key, &other_key,
2895 path->keep_locks = 0;
2896 if (ret || found_key.objectid != key.objectid ||
2897 found_key.type != key.type) {
2902 key.offset = found_key.offset;
2903 btrfs_release_path(root, path);
2908 /* we know there's one more slot after us in the tree,
2909 * read that key so we can verify it is also a checksum item
2911 btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
2913 if (found_key.objectid < inode->i_ino)
2916 if (found_key.type != key.type || found_key.offset < new_size)
2920 * if the key for the next leaf isn't a csum key from this objectid,
2921 * we can't be sure there aren't good items inside this leaf.
2924 if (other_key.objectid != inode->i_ino || other_key.type != key.type)
2927 leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
2928 leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
2930 * it is safe to delete this leaf, it contains only
2931 * csum items from this inode at an offset >= new_size
2933 ret = btrfs_del_leaf(trans, root, path, leaf_start);
2936 if (root->ref_cows && leaf_gen < trans->transid) {
2937 ref = btrfs_alloc_leaf_ref(root, 0);
2939 ref->root_gen = root->root_key.offset;
2940 ref->bytenr = leaf_start;
2942 ref->generation = leaf_gen;
2945 btrfs_sort_leaf_ref(ref);
2947 ret = btrfs_add_leaf_ref(root, ref, 0);
2949 btrfs_free_leaf_ref(root, ref);
2955 btrfs_release_path(root, path);
2957 if (other_key.objectid == inode->i_ino &&
2958 other_key.type == key.type && other_key.offset > key.offset) {
2959 key.offset = other_key.offset;
2965 /* fixup any changes we've made to the path */
2966 path->lowest_level = 0;
2967 path->keep_locks = 0;
2968 btrfs_release_path(root, path);
2975 * this can truncate away extent items, csum items and directory items.
2976 * It starts at a high offset and removes keys until it can't find
2977 * any higher than new_size
2979 * csum items that cross the new i_size are truncated to the new size
2982 * min_type is the minimum key type to truncate down to. If set to 0, this
2983 * will kill all the items on this inode, including the INODE_ITEM_KEY.
2985 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2986 struct btrfs_root *root,
2987 struct inode *inode,
2988 u64 new_size, u32 min_type)
2990 struct btrfs_path *path;
2991 struct extent_buffer *leaf;
2992 struct btrfs_file_extent_item *fi;
2993 struct btrfs_key key;
2994 struct btrfs_key found_key;
2995 u64 extent_start = 0;
2996 u64 extent_num_bytes = 0;
2997 u64 extent_offset = 0;
2999 u64 mask = root->sectorsize - 1;
3000 u32 found_type = (u8)-1;
3003 int pending_del_nr = 0;
3004 int pending_del_slot = 0;
3005 int extent_type = -1;
3010 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3013 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3015 path = btrfs_alloc_path();
3019 key.objectid = inode->i_ino;
3020 key.offset = (u64)-1;
3024 path->leave_spinning = 1;
3025 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3032 /* there are no items in the tree for us to truncate, we're
3035 if (path->slots[0] == 0)
3042 leaf = path->nodes[0];
3043 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3044 found_type = btrfs_key_type(&found_key);
3047 if (found_key.objectid != inode->i_ino)
3050 if (found_type < min_type)
3053 item_end = found_key.offset;
3054 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3055 fi = btrfs_item_ptr(leaf, path->slots[0],
3056 struct btrfs_file_extent_item);
3057 extent_type = btrfs_file_extent_type(leaf, fi);
3058 encoding = btrfs_file_extent_compression(leaf, fi);
3059 encoding |= btrfs_file_extent_encryption(leaf, fi);
3060 encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3062 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3064 btrfs_file_extent_num_bytes(leaf, fi);
3065 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3066 item_end += btrfs_file_extent_inline_len(leaf,
3071 if (found_type > min_type) {
3074 if (item_end < new_size)
3076 if (found_key.offset >= new_size)
3082 /* FIXME, shrink the extent if the ref count is only 1 */
3083 if (found_type != BTRFS_EXTENT_DATA_KEY)
3086 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3088 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3089 if (!del_item && !encoding) {
3090 u64 orig_num_bytes =
3091 btrfs_file_extent_num_bytes(leaf, fi);
3092 extent_num_bytes = new_size -
3093 found_key.offset + root->sectorsize - 1;
3094 extent_num_bytes = extent_num_bytes &
3095 ~((u64)root->sectorsize - 1);
3096 btrfs_set_file_extent_num_bytes(leaf, fi,
3098 num_dec = (orig_num_bytes -
3100 if (root->ref_cows && extent_start != 0)
3101 inode_sub_bytes(inode, num_dec);
3102 btrfs_mark_buffer_dirty(leaf);
3105 btrfs_file_extent_disk_num_bytes(leaf,
3107 extent_offset = found_key.offset -
3108 btrfs_file_extent_offset(leaf, fi);
3110 /* FIXME blocksize != 4096 */
3111 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3112 if (extent_start != 0) {
3115 inode_sub_bytes(inode, num_dec);
3118 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3120 * we can't truncate inline items that have had
3124 btrfs_file_extent_compression(leaf, fi) == 0 &&
3125 btrfs_file_extent_encryption(leaf, fi) == 0 &&
3126 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3127 u32 size = new_size - found_key.offset;
3129 if (root->ref_cows) {
3130 inode_sub_bytes(inode, item_end + 1 -
3134 btrfs_file_extent_calc_inline_size(size);
3135 ret = btrfs_truncate_item(trans, root, path,
3138 } else if (root->ref_cows) {
3139 inode_sub_bytes(inode, item_end + 1 -
3145 if (!pending_del_nr) {
3146 /* no pending yet, add ourselves */
3147 pending_del_slot = path->slots[0];
3149 } else if (pending_del_nr &&
3150 path->slots[0] + 1 == pending_del_slot) {
3151 /* hop on the pending chunk */
3153 pending_del_slot = path->slots[0];
3160 if (found_extent && root->ref_cows) {
3161 btrfs_set_path_blocking(path);
3162 ret = btrfs_free_extent(trans, root, extent_start,
3163 extent_num_bytes, 0,
3164 btrfs_header_owner(leaf),
3165 inode->i_ino, extent_offset);
3169 if (found_type == BTRFS_INODE_ITEM_KEY)
3172 if (path->slots[0] == 0 ||
3173 path->slots[0] != pending_del_slot) {
3174 if (root->ref_cows) {
3178 if (pending_del_nr) {
3179 ret = btrfs_del_items(trans, root, path,
3185 btrfs_release_path(root, path);
3192 if (pending_del_nr) {
3193 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3196 btrfs_free_path(path);
3201 * taken from block_truncate_page, but does cow as it zeros out
3202 * any bytes left in the last page in the file.
3204 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3206 struct inode *inode = mapping->host;
3207 struct btrfs_root *root = BTRFS_I(inode)->root;
3208 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3209 struct btrfs_ordered_extent *ordered;
3210 struct extent_state *cached_state = NULL;
3212 u32 blocksize = root->sectorsize;
3213 pgoff_t index = from >> PAGE_CACHE_SHIFT;
3214 unsigned offset = from & (PAGE_CACHE_SIZE-1);
3220 if ((offset & (blocksize - 1)) == 0)
3222 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
3226 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
3232 page = grab_cache_page(mapping, index);
3234 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3235 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3239 page_start = page_offset(page);
3240 page_end = page_start + PAGE_CACHE_SIZE - 1;
3242 if (!PageUptodate(page)) {
3243 ret = btrfs_readpage(NULL, page);
3245 if (page->mapping != mapping) {
3247 page_cache_release(page);
3250 if (!PageUptodate(page)) {
3255 wait_on_page_writeback(page);
3257 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3259 set_page_extent_mapped(page);
3261 ordered = btrfs_lookup_ordered_extent(inode, page_start);
3263 unlock_extent_cached(io_tree, page_start, page_end,
3264 &cached_state, GFP_NOFS);
3266 page_cache_release(page);
3267 btrfs_start_ordered_extent(inode, ordered, 1);
3268 btrfs_put_ordered_extent(ordered);
3272 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3273 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3274 0, 0, &cached_state, GFP_NOFS);
3276 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3279 unlock_extent_cached(io_tree, page_start, page_end,
3280 &cached_state, GFP_NOFS);
3285 if (offset != PAGE_CACHE_SIZE) {
3287 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3288 flush_dcache_page(page);
3291 ClearPageChecked(page);
3292 set_page_dirty(page);
3293 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3298 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
3299 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
3301 page_cache_release(page);
3306 int btrfs_cont_expand(struct inode *inode, loff_t size)
3308 struct btrfs_trans_handle *trans;
3309 struct btrfs_root *root = BTRFS_I(inode)->root;
3310 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3311 struct extent_map *em = NULL;
3312 struct extent_state *cached_state = NULL;
3313 u64 mask = root->sectorsize - 1;
3314 u64 hole_start = (inode->i_size + mask) & ~mask;
3315 u64 block_end = (size + mask) & ~mask;
3321 if (size <= hole_start)
3325 struct btrfs_ordered_extent *ordered;
3326 btrfs_wait_ordered_range(inode, hole_start,
3327 block_end - hole_start);
3328 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3329 &cached_state, GFP_NOFS);
3330 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3333 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3334 &cached_state, GFP_NOFS);
3335 btrfs_put_ordered_extent(ordered);
3338 cur_offset = hole_start;
3340 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3341 block_end - cur_offset, 0);
3342 BUG_ON(IS_ERR(em) || !em);
3343 last_byte = min(extent_map_end(em), block_end);
3344 last_byte = (last_byte + mask) & ~mask;
3345 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3347 hole_size = last_byte - cur_offset;
3349 trans = btrfs_start_transaction(root, 2);
3350 if (IS_ERR(trans)) {
3351 err = PTR_ERR(trans);
3354 btrfs_set_trans_block_group(trans, inode);
3356 err = btrfs_drop_extents(trans, inode, cur_offset,
3357 cur_offset + hole_size,
3361 err = btrfs_insert_file_extent(trans, root,
3362 inode->i_ino, cur_offset, 0,
3363 0, hole_size, 0, hole_size,
3367 btrfs_drop_extent_cache(inode, hole_start,
3370 btrfs_end_transaction(trans, root);
3372 free_extent_map(em);
3374 cur_offset = last_byte;
3375 if (cur_offset >= block_end)
3379 free_extent_map(em);
3380 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3385 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3387 struct btrfs_root *root = BTRFS_I(inode)->root;
3388 struct btrfs_trans_handle *trans;
3392 if (attr->ia_size == inode->i_size)
3395 if (attr->ia_size > inode->i_size) {
3396 unsigned long limit;
3397 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3398 if (attr->ia_size > inode->i_sb->s_maxbytes)
3400 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3401 send_sig(SIGXFSZ, current, 0);
3406 trans = btrfs_start_transaction(root, 1);
3407 btrfs_set_trans_block_group(trans, inode);
3409 ret = btrfs_orphan_add(trans, inode);
3412 nr = trans->blocks_used;
3413 btrfs_end_transaction(trans, root);
3414 btrfs_btree_balance_dirty(root, nr);
3416 if (attr->ia_size > inode->i_size) {
3417 ret = btrfs_cont_expand(inode, attr->ia_size);
3419 btrfs_truncate(inode);
3423 i_size_write(inode, attr->ia_size);
3424 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3426 trans = btrfs_start_transaction(root, 1);
3427 btrfs_set_trans_block_group(trans, inode);
3429 ret = btrfs_update_inode(trans, root, inode);
3431 if (inode->i_nlink > 0) {
3432 ret = btrfs_orphan_del(trans, inode);
3435 nr = trans->blocks_used;
3436 btrfs_end_transaction(trans, root);
3437 btrfs_btree_balance_dirty(root, nr);
3442 * We're truncating a file that used to have good data down to
3443 * zero. Make sure it gets into the ordered flush list so that
3444 * any new writes get down to disk quickly.
3446 if (attr->ia_size == 0)
3447 BTRFS_I(inode)->ordered_data_close = 1;
3449 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3450 ret = vmtruncate(inode, attr->ia_size);
3456 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3458 struct inode *inode = dentry->d_inode;
3461 err = inode_change_ok(inode, attr);
3465 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3466 err = btrfs_setattr_size(inode, attr);
3470 attr->ia_valid &= ~ATTR_SIZE;
3473 err = inode_setattr(inode, attr);
3475 if (!err && ((attr->ia_valid & ATTR_MODE)))
3476 err = btrfs_acl_chmod(inode);
3480 void btrfs_delete_inode(struct inode *inode)
3482 struct btrfs_trans_handle *trans;
3483 struct btrfs_root *root = BTRFS_I(inode)->root;
3487 truncate_inode_pages(&inode->i_data, 0);
3488 if (is_bad_inode(inode)) {
3489 btrfs_orphan_del(NULL, inode);
3492 btrfs_wait_ordered_range(inode, 0, (u64)-1);
3494 if (root->fs_info->log_root_recovering) {
3495 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3499 if (inode->i_nlink > 0) {
3500 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3504 btrfs_i_size_write(inode, 0);
3507 trans = btrfs_start_transaction(root, 1);
3508 btrfs_set_trans_block_group(trans, inode);
3509 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3514 nr = trans->blocks_used;
3515 btrfs_end_transaction(trans, root);
3517 btrfs_btree_balance_dirty(root, nr);
3521 ret = btrfs_orphan_del(trans, inode);
3525 nr = trans->blocks_used;
3526 btrfs_end_transaction(trans, root);
3527 btrfs_btree_balance_dirty(root, nr);
3534 * this returns the key found in the dir entry in the location pointer.
3535 * If no dir entries were found, location->objectid is 0.
3537 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3538 struct btrfs_key *location)
3540 const char *name = dentry->d_name.name;
3541 int namelen = dentry->d_name.len;
3542 struct btrfs_dir_item *di;
3543 struct btrfs_path *path;
3544 struct btrfs_root *root = BTRFS_I(dir)->root;
3547 path = btrfs_alloc_path();
3550 di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3555 if (!di || IS_ERR(di))
3558 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3560 btrfs_free_path(path);
3563 location->objectid = 0;
3568 * when we hit a tree root in a directory, the btrfs part of the inode
3569 * needs to be changed to reflect the root directory of the tree root. This
3570 * is kind of like crossing a mount point.
3572 static int fixup_tree_root_location(struct btrfs_root *root,
3574 struct dentry *dentry,
3575 struct btrfs_key *location,
3576 struct btrfs_root **sub_root)
3578 struct btrfs_path *path;
3579 struct btrfs_root *new_root;
3580 struct btrfs_root_ref *ref;
3581 struct extent_buffer *leaf;
3585 path = btrfs_alloc_path();
3592 ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3593 BTRFS_I(dir)->root->root_key.objectid,
3594 location->objectid);
3601 leaf = path->nodes[0];
3602 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3603 if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3604 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3607 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3608 (unsigned long)(ref + 1),
3609 dentry->d_name.len);
3613 btrfs_release_path(root->fs_info->tree_root, path);
3615 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3616 if (IS_ERR(new_root)) {
3617 err = PTR_ERR(new_root);
3621 if (btrfs_root_refs(&new_root->root_item) == 0) {
3626 *sub_root = new_root;
3627 location->objectid = btrfs_root_dirid(&new_root->root_item);
3628 location->type = BTRFS_INODE_ITEM_KEY;
3629 location->offset = 0;
3632 btrfs_free_path(path);
3636 static void inode_tree_add(struct inode *inode)
3638 struct btrfs_root *root = BTRFS_I(inode)->root;
3639 struct btrfs_inode *entry;
3641 struct rb_node *parent;
3643 p = &root->inode_tree.rb_node;
3646 if (hlist_unhashed(&inode->i_hash))
3649 spin_lock(&root->inode_lock);
3652 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3654 if (inode->i_ino < entry->vfs_inode.i_ino)
3655 p = &parent->rb_left;
3656 else if (inode->i_ino > entry->vfs_inode.i_ino)
3657 p = &parent->rb_right;
3659 WARN_ON(!(entry->vfs_inode.i_state &
3660 (I_WILL_FREE | I_FREEING | I_CLEAR)));
3661 rb_erase(parent, &root->inode_tree);
3662 RB_CLEAR_NODE(parent);
3663 spin_unlock(&root->inode_lock);
3667 rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3668 rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3669 spin_unlock(&root->inode_lock);
3672 static void inode_tree_del(struct inode *inode)
3674 struct btrfs_root *root = BTRFS_I(inode)->root;
3677 spin_lock(&root->inode_lock);
3678 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3679 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3680 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3681 empty = RB_EMPTY_ROOT(&root->inode_tree);
3683 spin_unlock(&root->inode_lock);
3685 if (empty && btrfs_root_refs(&root->root_item) == 0) {
3686 synchronize_srcu(&root->fs_info->subvol_srcu);
3687 spin_lock(&root->inode_lock);
3688 empty = RB_EMPTY_ROOT(&root->inode_tree);
3689 spin_unlock(&root->inode_lock);
3691 btrfs_add_dead_root(root);
3695 int btrfs_invalidate_inodes(struct btrfs_root *root)
3697 struct rb_node *node;
3698 struct rb_node *prev;
3699 struct btrfs_inode *entry;
3700 struct inode *inode;
3703 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3705 spin_lock(&root->inode_lock);
3707 node = root->inode_tree.rb_node;
3711 entry = rb_entry(node, struct btrfs_inode, rb_node);
3713 if (objectid < entry->vfs_inode.i_ino)
3714 node = node->rb_left;
3715 else if (objectid > entry->vfs_inode.i_ino)
3716 node = node->rb_right;
3722 entry = rb_entry(prev, struct btrfs_inode, rb_node);
3723 if (objectid <= entry->vfs_inode.i_ino) {
3727 prev = rb_next(prev);
3731 entry = rb_entry(node, struct btrfs_inode, rb_node);
3732 objectid = entry->vfs_inode.i_ino + 1;
3733 inode = igrab(&entry->vfs_inode);
3735 spin_unlock(&root->inode_lock);
3736 if (atomic_read(&inode->i_count) > 1)
3737 d_prune_aliases(inode);
3739 * btrfs_drop_inode will remove it from
3740 * the inode cache when its usage count
3745 spin_lock(&root->inode_lock);
3749 if (cond_resched_lock(&root->inode_lock))
3752 node = rb_next(node);
3754 spin_unlock(&root->inode_lock);
3758 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3760 struct btrfs_iget_args *args = p;
3761 inode->i_ino = args->ino;
3762 BTRFS_I(inode)->root = args->root;
3763 btrfs_set_inode_space_info(args->root, inode);
3767 static int btrfs_find_actor(struct inode *inode, void *opaque)
3769 struct btrfs_iget_args *args = opaque;
3770 return args->ino == inode->i_ino &&
3771 args->root == BTRFS_I(inode)->root;
3774 static struct inode *btrfs_iget_locked(struct super_block *s,
3776 struct btrfs_root *root)
3778 struct inode *inode;
3779 struct btrfs_iget_args args;
3780 args.ino = objectid;
3783 inode = iget5_locked(s, objectid, btrfs_find_actor,
3784 btrfs_init_locked_inode,
3789 /* Get an inode object given its location and corresponding root.
3790 * Returns in *is_new if the inode was read from disk
3792 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3793 struct btrfs_root *root, int *new)
3795 struct inode *inode;
3797 inode = btrfs_iget_locked(s, location->objectid, root);
3799 return ERR_PTR(-ENOMEM);
3801 if (inode->i_state & I_NEW) {
3802 BTRFS_I(inode)->root = root;
3803 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3804 btrfs_read_locked_inode(inode);
3806 inode_tree_add(inode);
3807 unlock_new_inode(inode);
3815 static struct inode *new_simple_dir(struct super_block *s,
3816 struct btrfs_key *key,
3817 struct btrfs_root *root)
3819 struct inode *inode = new_inode(s);
3822 return ERR_PTR(-ENOMEM);
3824 BTRFS_I(inode)->root = root;
3825 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3826 BTRFS_I(inode)->dummy_inode = 1;
3828 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3829 inode->i_op = &simple_dir_inode_operations;
3830 inode->i_fop = &simple_dir_operations;
3831 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3832 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3837 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3839 struct inode *inode;
3840 struct btrfs_root *root = BTRFS_I(dir)->root;
3841 struct btrfs_root *sub_root = root;
3842 struct btrfs_key location;
3846 dentry->d_op = &btrfs_dentry_operations;
3848 if (dentry->d_name.len > BTRFS_NAME_LEN)
3849 return ERR_PTR(-ENAMETOOLONG);
3851 ret = btrfs_inode_by_name(dir, dentry, &location);
3854 return ERR_PTR(ret);
3856 if (location.objectid == 0)
3859 if (location.type == BTRFS_INODE_ITEM_KEY) {
3860 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3864 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3866 index = srcu_read_lock(&root->fs_info->subvol_srcu);
3867 ret = fixup_tree_root_location(root, dir, dentry,
3868 &location, &sub_root);
3871 inode = ERR_PTR(ret);
3873 inode = new_simple_dir(dir->i_sb, &location, sub_root);
3875 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3877 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3879 if (root != sub_root) {
3880 down_read(&root->fs_info->cleanup_work_sem);
3881 if (!(inode->i_sb->s_flags & MS_RDONLY))
3882 btrfs_orphan_cleanup(sub_root);
3883 up_read(&root->fs_info->cleanup_work_sem);
3889 static int btrfs_dentry_delete(struct dentry *dentry)
3891 struct btrfs_root *root;
3893 if (!dentry->d_inode && !IS_ROOT(dentry))
3894 dentry = dentry->d_parent;
3896 if (dentry->d_inode) {
3897 root = BTRFS_I(dentry->d_inode)->root;
3898 if (btrfs_root_refs(&root->root_item) == 0)
3904 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3905 struct nameidata *nd)
3907 struct inode *inode;
3909 inode = btrfs_lookup_dentry(dir, dentry);
3911 return ERR_CAST(inode);
3913 return d_splice_alias(inode, dentry);
3916 static unsigned char btrfs_filetype_table[] = {
3917 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3920 static int btrfs_real_readdir(struct file *filp, void *dirent,
3923 struct inode *inode = filp->f_dentry->d_inode;
3924 struct btrfs_root *root = BTRFS_I(inode)->root;
3925 struct btrfs_item *item;
3926 struct btrfs_dir_item *di;
3927 struct btrfs_key key;
3928 struct btrfs_key found_key;
3929 struct btrfs_path *path;
3932 struct extent_buffer *leaf;
3935 unsigned char d_type;
3940 int key_type = BTRFS_DIR_INDEX_KEY;
3945 /* FIXME, use a real flag for deciding about the key type */
3946 if (root->fs_info->tree_root == root)
3947 key_type = BTRFS_DIR_ITEM_KEY;
3949 /* special case for "." */
3950 if (filp->f_pos == 0) {
3951 over = filldir(dirent, ".", 1,
3958 /* special case for .., just use the back ref */
3959 if (filp->f_pos == 1) {
3960 u64 pino = parent_ino(filp->f_path.dentry);
3961 over = filldir(dirent, "..", 2,
3967 path = btrfs_alloc_path();
3970 btrfs_set_key_type(&key, key_type);
3971 key.offset = filp->f_pos;
3972 key.objectid = inode->i_ino;
3974 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3980 leaf = path->nodes[0];
3981 nritems = btrfs_header_nritems(leaf);
3982 slot = path->slots[0];
3983 if (advance || slot >= nritems) {
3984 if (slot >= nritems - 1) {
3985 ret = btrfs_next_leaf(root, path);
3988 leaf = path->nodes[0];
3989 nritems = btrfs_header_nritems(leaf);
3990 slot = path->slots[0];
3998 item = btrfs_item_nr(leaf, slot);
3999 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4001 if (found_key.objectid != key.objectid)
4003 if (btrfs_key_type(&found_key) != key_type)
4005 if (found_key.offset < filp->f_pos)
4008 filp->f_pos = found_key.offset;
4010 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4012 di_total = btrfs_item_size(leaf, item);
4014 while (di_cur < di_total) {
4015 struct btrfs_key location;
4017 name_len = btrfs_dir_name_len(leaf, di);
4018 if (name_len <= sizeof(tmp_name)) {
4019 name_ptr = tmp_name;
4021 name_ptr = kmalloc(name_len, GFP_NOFS);
4027 read_extent_buffer(leaf, name_ptr,
4028 (unsigned long)(di + 1), name_len);
4030 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4031 btrfs_dir_item_key_to_cpu(leaf, di, &location);
4033 /* is this a reference to our own snapshot? If so
4036 if (location.type == BTRFS_ROOT_ITEM_KEY &&
4037 location.objectid == root->root_key.objectid) {
4041 over = filldir(dirent, name_ptr, name_len,
4042 found_key.offset, location.objectid,
4046 if (name_ptr != tmp_name)
4051 di_len = btrfs_dir_name_len(leaf, di) +
4052 btrfs_dir_data_len(leaf, di) + sizeof(*di);
4054 di = (struct btrfs_dir_item *)((char *)di + di_len);
4058 /* Reached end of directory/root. Bump pos past the last item. */
4059 if (key_type == BTRFS_DIR_INDEX_KEY)
4061 * 32-bit glibc will use getdents64, but then strtol -
4062 * so the last number we can serve is this.
4064 filp->f_pos = 0x7fffffff;
4070 btrfs_free_path(path);
4074 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4076 struct btrfs_root *root = BTRFS_I(inode)->root;
4077 struct btrfs_trans_handle *trans;
4080 if (root->fs_info->btree_inode == inode)
4083 if (wbc->sync_mode == WB_SYNC_ALL) {
4084 trans = btrfs_join_transaction(root, 1);
4085 btrfs_set_trans_block_group(trans, inode);
4086 ret = btrfs_commit_transaction(trans, root);
4092 * This is somewhat expensive, updating the tree every time the
4093 * inode changes. But, it is most likely to find the inode in cache.
4094 * FIXME, needs more benchmarking...there are no reasons other than performance
4095 * to keep or drop this code.
4097 void btrfs_dirty_inode(struct inode *inode)
4099 struct btrfs_root *root = BTRFS_I(inode)->root;
4100 struct btrfs_trans_handle *trans;
4102 trans = btrfs_join_transaction(root, 1);
4103 btrfs_set_trans_block_group(trans, inode);
4104 btrfs_update_inode(trans, root, inode);
4105 btrfs_end_transaction(trans, root);
4109 * find the highest existing sequence number in a directory
4110 * and then set the in-memory index_cnt variable to reflect
4111 * free sequence numbers
4113 static int btrfs_set_inode_index_count(struct inode *inode)
4115 struct btrfs_root *root = BTRFS_I(inode)->root;
4116 struct btrfs_key key, found_key;
4117 struct btrfs_path *path;
4118 struct extent_buffer *leaf;
4121 key.objectid = inode->i_ino;
4122 btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4123 key.offset = (u64)-1;
4125 path = btrfs_alloc_path();
4129 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4132 /* FIXME: we should be able to handle this */
4138 * MAGIC NUMBER EXPLANATION:
4139 * since we search a directory based on f_pos we have to start at 2
4140 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4141 * else has to start at 2
4143 if (path->slots[0] == 0) {
4144 BTRFS_I(inode)->index_cnt = 2;
4150 leaf = path->nodes[0];
4151 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4153 if (found_key.objectid != inode->i_ino ||
4154 btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4155 BTRFS_I(inode)->index_cnt = 2;
4159 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4161 btrfs_free_path(path);
4166 * helper to find a free sequence number in a given directory. This current
4167 * code is very simple, later versions will do smarter things in the btree
4169 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4173 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4174 ret = btrfs_set_inode_index_count(dir);
4179 *index = BTRFS_I(dir)->index_cnt;
4180 BTRFS_I(dir)->index_cnt++;
4185 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4186 struct btrfs_root *root,
4188 const char *name, int name_len,
4189 u64 ref_objectid, u64 objectid,
4190 u64 alloc_hint, int mode, u64 *index)
4192 struct inode *inode;
4193 struct btrfs_inode_item *inode_item;
4194 struct btrfs_key *location;
4195 struct btrfs_path *path;
4196 struct btrfs_inode_ref *ref;
4197 struct btrfs_key key[2];
4203 path = btrfs_alloc_path();
4206 inode = new_inode(root->fs_info->sb);
4208 return ERR_PTR(-ENOMEM);
4211 ret = btrfs_set_inode_index(dir, index);
4214 return ERR_PTR(ret);
4218 * index_cnt is ignored for everything but a dir,
4219 * btrfs_get_inode_index_count has an explanation for the magic
4222 BTRFS_I(inode)->index_cnt = 2;
4223 BTRFS_I(inode)->root = root;
4224 BTRFS_I(inode)->generation = trans->transid;
4225 btrfs_set_inode_space_info(root, inode);
4231 BTRFS_I(inode)->block_group =
4232 btrfs_find_block_group(root, 0, alloc_hint, owner);
4234 key[0].objectid = objectid;
4235 btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4238 key[1].objectid = objectid;
4239 btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4240 key[1].offset = ref_objectid;
4242 sizes[0] = sizeof(struct btrfs_inode_item);
4243 sizes[1] = name_len + sizeof(*ref);
4245 path->leave_spinning = 1;
4246 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4250 inode->i_uid = current_fsuid();
4252 if (dir && (dir->i_mode & S_ISGID)) {
4253 inode->i_gid = dir->i_gid;
4257 inode->i_gid = current_fsgid();
4259 inode->i_mode = mode;
4260 inode->i_ino = objectid;
4261 inode_set_bytes(inode, 0);
4262 inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4263 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4264 struct btrfs_inode_item);
4265 fill_inode_item(trans, path->nodes[0], inode_item, inode);
4267 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4268 struct btrfs_inode_ref);
4269 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4270 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4271 ptr = (unsigned long)(ref + 1);
4272 write_extent_buffer(path->nodes[0], name, ptr, name_len);
4274 btrfs_mark_buffer_dirty(path->nodes[0]);
4275 btrfs_free_path(path);
4277 location = &BTRFS_I(inode)->location;
4278 location->objectid = objectid;
4279 location->offset = 0;
4280 btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4282 btrfs_inherit_iflags(inode, dir);
4284 if ((mode & S_IFREG)) {
4285 if (btrfs_test_opt(root, NODATASUM))
4286 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4287 if (btrfs_test_opt(root, NODATACOW))
4288 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4291 insert_inode_hash(inode);
4292 inode_tree_add(inode);
4296 BTRFS_I(dir)->index_cnt--;
4297 btrfs_free_path(path);
4299 return ERR_PTR(ret);
4302 static inline u8 btrfs_inode_type(struct inode *inode)
4304 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4308 * utility function to add 'inode' into 'parent_inode' with
4309 * a give name and a given sequence number.
4310 * if 'add_backref' is true, also insert a backref from the
4311 * inode to the parent directory.
4313 int btrfs_add_link(struct btrfs_trans_handle *trans,
4314 struct inode *parent_inode, struct inode *inode,
4315 const char *name, int name_len, int add_backref, u64 index)
4318 struct btrfs_key key;
4319 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4321 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4322 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4324 key.objectid = inode->i_ino;
4325 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4329 if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4330 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4331 key.objectid, root->root_key.objectid,
4332 parent_inode->i_ino,
4333 index, name, name_len);
4334 } else if (add_backref) {
4335 ret = btrfs_insert_inode_ref(trans, root,
4336 name, name_len, inode->i_ino,
4337 parent_inode->i_ino, index);
4341 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4342 parent_inode->i_ino, &key,
4343 btrfs_inode_type(inode), index);
4346 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4348 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4349 ret = btrfs_update_inode(trans, root, parent_inode);
4354 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4355 struct dentry *dentry, struct inode *inode,
4356 int backref, u64 index)
4358 int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4359 inode, dentry->d_name.name,
4360 dentry->d_name.len, backref, index);
4362 d_instantiate(dentry, inode);
4370 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4371 int mode, dev_t rdev)
4373 struct btrfs_trans_handle *trans;
4374 struct btrfs_root *root = BTRFS_I(dir)->root;
4375 struct inode *inode = NULL;
4379 unsigned long nr = 0;
4382 if (!new_valid_dev(rdev))
4385 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4390 * 2 for inode item and ref
4392 * 1 for xattr if selinux is on
4394 trans = btrfs_start_transaction(root, 5);
4396 return PTR_ERR(trans);
4398 btrfs_set_trans_block_group(trans, dir);
4400 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4402 dentry->d_parent->d_inode->i_ino, objectid,
4403 BTRFS_I(dir)->block_group, mode, &index);
4404 err = PTR_ERR(inode);
4408 err = btrfs_init_inode_security(trans, inode, dir);
4414 btrfs_set_trans_block_group(trans, inode);
4415 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4419 inode->i_op = &btrfs_special_inode_operations;
4420 init_special_inode(inode, inode->i_mode, rdev);
4421 btrfs_update_inode(trans, root, inode);
4423 btrfs_update_inode_block_group(trans, inode);
4424 btrfs_update_inode_block_group(trans, dir);
4426 nr = trans->blocks_used;
4427 btrfs_end_transaction_throttle(trans, root);
4428 btrfs_btree_balance_dirty(root, nr);
4430 inode_dec_link_count(inode);
4436 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4437 int mode, struct nameidata *nd)
4439 struct btrfs_trans_handle *trans;
4440 struct btrfs_root *root = BTRFS_I(dir)->root;
4441 struct inode *inode = NULL;
4444 unsigned long nr = 0;
4448 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4452 * 2 for inode item and ref
4454 * 1 for xattr if selinux is on
4456 trans = btrfs_start_transaction(root, 5);
4458 return PTR_ERR(trans);
4460 btrfs_set_trans_block_group(trans, dir);
4462 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4464 dentry->d_parent->d_inode->i_ino,
4465 objectid, BTRFS_I(dir)->block_group, mode,
4467 err = PTR_ERR(inode);
4471 err = btrfs_init_inode_security(trans, inode, dir);
4477 btrfs_set_trans_block_group(trans, inode);
4478 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
4482 inode->i_mapping->a_ops = &btrfs_aops;
4483 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4484 inode->i_fop = &btrfs_file_operations;
4485 inode->i_op = &btrfs_file_inode_operations;
4486 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4488 btrfs_update_inode_block_group(trans, inode);
4489 btrfs_update_inode_block_group(trans, dir);
4491 nr = trans->blocks_used;
4492 btrfs_end_transaction_throttle(trans, root);
4494 inode_dec_link_count(inode);
4497 btrfs_btree_balance_dirty(root, nr);
4501 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4502 struct dentry *dentry)
4504 struct btrfs_trans_handle *trans;
4505 struct btrfs_root *root = BTRFS_I(dir)->root;
4506 struct inode *inode = old_dentry->d_inode;
4508 unsigned long nr = 0;
4512 if (inode->i_nlink == 0)
4515 /* do not allow sys_link's with other subvols of the same device */
4516 if (root->objectid != BTRFS_I(inode)->root->objectid)
4519 btrfs_inc_nlink(inode);
4521 err = btrfs_set_inode_index(dir, &index);
4526 * 1 item for inode ref
4527 * 2 items for dir items
4529 trans = btrfs_start_transaction(root, 3);
4530 if (IS_ERR(trans)) {
4531 err = PTR_ERR(trans);
4535 btrfs_set_trans_block_group(trans, dir);
4536 atomic_inc(&inode->i_count);
4538 err = btrfs_add_nondir(trans, dentry, inode, 1, index);
4543 btrfs_update_inode_block_group(trans, dir);
4544 err = btrfs_update_inode(trans, root, inode);
4546 btrfs_log_new_name(trans, inode, NULL, dentry->d_parent);
4549 nr = trans->blocks_used;
4550 btrfs_end_transaction_throttle(trans, root);
4553 inode_dec_link_count(inode);
4556 btrfs_btree_balance_dirty(root, nr);
4560 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4562 struct inode *inode = NULL;
4563 struct btrfs_trans_handle *trans;
4564 struct btrfs_root *root = BTRFS_I(dir)->root;
4566 int drop_on_err = 0;
4569 unsigned long nr = 1;
4571 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4576 * 2 items for inode and ref
4577 * 2 items for dir items
4578 * 1 for xattr if selinux is on
4580 trans = btrfs_start_transaction(root, 5);
4582 return PTR_ERR(trans);
4583 btrfs_set_trans_block_group(trans, dir);
4585 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4587 dentry->d_parent->d_inode->i_ino, objectid,
4588 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4590 if (IS_ERR(inode)) {
4591 err = PTR_ERR(inode);
4597 err = btrfs_init_inode_security(trans, inode, dir);
4601 inode->i_op = &btrfs_dir_inode_operations;
4602 inode->i_fop = &btrfs_dir_file_operations;
4603 btrfs_set_trans_block_group(trans, inode);
4605 btrfs_i_size_write(inode, 0);
4606 err = btrfs_update_inode(trans, root, inode);
4610 err = btrfs_add_link(trans, dentry->d_parent->d_inode,
4611 inode, dentry->d_name.name,
4612 dentry->d_name.len, 0, index);
4616 d_instantiate(dentry, inode);
4618 btrfs_update_inode_block_group(trans, inode);
4619 btrfs_update_inode_block_group(trans, dir);
4622 nr = trans->blocks_used;
4623 btrfs_end_transaction_throttle(trans, root);
4626 btrfs_btree_balance_dirty(root, nr);
4630 /* helper for btfs_get_extent. Given an existing extent in the tree,
4631 * and an extent that you want to insert, deal with overlap and insert
4632 * the new extent into the tree.
4634 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4635 struct extent_map *existing,
4636 struct extent_map *em,
4637 u64 map_start, u64 map_len)
4641 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4642 start_diff = map_start - em->start;
4643 em->start = map_start;
4645 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4646 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4647 em->block_start += start_diff;
4648 em->block_len -= start_diff;
4650 return add_extent_mapping(em_tree, em);
4653 static noinline int uncompress_inline(struct btrfs_path *path,
4654 struct inode *inode, struct page *page,
4655 size_t pg_offset, u64 extent_offset,
4656 struct btrfs_file_extent_item *item)
4659 struct extent_buffer *leaf = path->nodes[0];
4662 unsigned long inline_size;
4665 WARN_ON(pg_offset != 0);
4666 max_size = btrfs_file_extent_ram_bytes(leaf, item);
4667 inline_size = btrfs_file_extent_inline_item_len(leaf,
4668 btrfs_item_nr(leaf, path->slots[0]));
4669 tmp = kmalloc(inline_size, GFP_NOFS);
4670 ptr = btrfs_file_extent_inline_start(item);
4672 read_extent_buffer(leaf, tmp, ptr, inline_size);
4674 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4675 ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4676 inline_size, max_size);
4678 char *kaddr = kmap_atomic(page, KM_USER0);
4679 unsigned long copy_size = min_t(u64,
4680 PAGE_CACHE_SIZE - pg_offset,
4681 max_size - extent_offset);
4682 memset(kaddr + pg_offset, 0, copy_size);
4683 kunmap_atomic(kaddr, KM_USER0);
4690 * a bit scary, this does extent mapping from logical file offset to the disk.
4691 * the ugly parts come from merging extents from the disk with the in-ram
4692 * representation. This gets more complex because of the data=ordered code,
4693 * where the in-ram extents might be locked pending data=ordered completion.
4695 * This also copies inline extents directly into the page.
4698 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4699 size_t pg_offset, u64 start, u64 len,
4705 u64 extent_start = 0;
4707 u64 objectid = inode->i_ino;
4709 struct btrfs_path *path = NULL;
4710 struct btrfs_root *root = BTRFS_I(inode)->root;
4711 struct btrfs_file_extent_item *item;
4712 struct extent_buffer *leaf;
4713 struct btrfs_key found_key;
4714 struct extent_map *em = NULL;
4715 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4716 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4717 struct btrfs_trans_handle *trans = NULL;
4721 read_lock(&em_tree->lock);
4722 em = lookup_extent_mapping(em_tree, start, len);
4724 em->bdev = root->fs_info->fs_devices->latest_bdev;
4725 read_unlock(&em_tree->lock);
4728 if (em->start > start || em->start + em->len <= start)
4729 free_extent_map(em);
4730 else if (em->block_start == EXTENT_MAP_INLINE && page)
4731 free_extent_map(em);
4735 em = alloc_extent_map(GFP_NOFS);
4740 em->bdev = root->fs_info->fs_devices->latest_bdev;
4741 em->start = EXTENT_MAP_HOLE;
4742 em->orig_start = EXTENT_MAP_HOLE;
4744 em->block_len = (u64)-1;
4747 path = btrfs_alloc_path();
4751 ret = btrfs_lookup_file_extent(trans, root, path,
4752 objectid, start, trans != NULL);
4759 if (path->slots[0] == 0)
4764 leaf = path->nodes[0];
4765 item = btrfs_item_ptr(leaf, path->slots[0],
4766 struct btrfs_file_extent_item);
4767 /* are we inside the extent that was found? */
4768 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4769 found_type = btrfs_key_type(&found_key);
4770 if (found_key.objectid != objectid ||
4771 found_type != BTRFS_EXTENT_DATA_KEY) {
4775 found_type = btrfs_file_extent_type(leaf, item);
4776 extent_start = found_key.offset;
4777 compressed = btrfs_file_extent_compression(leaf, item);
4778 if (found_type == BTRFS_FILE_EXTENT_REG ||
4779 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4780 extent_end = extent_start +
4781 btrfs_file_extent_num_bytes(leaf, item);
4782 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4784 size = btrfs_file_extent_inline_len(leaf, item);
4785 extent_end = (extent_start + size + root->sectorsize - 1) &
4786 ~((u64)root->sectorsize - 1);
4789 if (start >= extent_end) {
4791 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4792 ret = btrfs_next_leaf(root, path);
4799 leaf = path->nodes[0];
4801 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4802 if (found_key.objectid != objectid ||
4803 found_key.type != BTRFS_EXTENT_DATA_KEY)
4805 if (start + len <= found_key.offset)
4808 em->len = found_key.offset - start;
4812 if (found_type == BTRFS_FILE_EXTENT_REG ||
4813 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4814 em->start = extent_start;
4815 em->len = extent_end - extent_start;
4816 em->orig_start = extent_start -
4817 btrfs_file_extent_offset(leaf, item);
4818 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4820 em->block_start = EXTENT_MAP_HOLE;
4824 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4825 em->block_start = bytenr;
4826 em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4829 bytenr += btrfs_file_extent_offset(leaf, item);
4830 em->block_start = bytenr;
4831 em->block_len = em->len;
4832 if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4833 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4836 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4840 size_t extent_offset;
4843 em->block_start = EXTENT_MAP_INLINE;
4844 if (!page || create) {
4845 em->start = extent_start;
4846 em->len = extent_end - extent_start;
4850 size = btrfs_file_extent_inline_len(leaf, item);
4851 extent_offset = page_offset(page) + pg_offset - extent_start;
4852 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4853 size - extent_offset);
4854 em->start = extent_start + extent_offset;
4855 em->len = (copy_size + root->sectorsize - 1) &
4856 ~((u64)root->sectorsize - 1);
4857 em->orig_start = EXTENT_MAP_INLINE;
4859 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4860 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4861 if (create == 0 && !PageUptodate(page)) {
4862 if (btrfs_file_extent_compression(leaf, item) ==
4863 BTRFS_COMPRESS_ZLIB) {
4864 ret = uncompress_inline(path, inode, page,
4866 extent_offset, item);
4870 read_extent_buffer(leaf, map + pg_offset, ptr,
4872 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4873 memset(map + pg_offset + copy_size, 0,
4874 PAGE_CACHE_SIZE - pg_offset -
4879 flush_dcache_page(page);
4880 } else if (create && PageUptodate(page)) {
4883 free_extent_map(em);
4885 btrfs_release_path(root, path);
4886 trans = btrfs_join_transaction(root, 1);
4890 write_extent_buffer(leaf, map + pg_offset, ptr,
4893 btrfs_mark_buffer_dirty(leaf);
4895 set_extent_uptodate(io_tree, em->start,
4896 extent_map_end(em) - 1, GFP_NOFS);
4899 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
4906 em->block_start = EXTENT_MAP_HOLE;
4907 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
4909 btrfs_release_path(root, path);
4910 if (em->start > start || extent_map_end(em) <= start) {
4911 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
4912 "[%llu %llu]\n", (unsigned long long)em->start,
4913 (unsigned long long)em->len,
4914 (unsigned long long)start,
4915 (unsigned long long)len);
4921 write_lock(&em_tree->lock);
4922 ret = add_extent_mapping(em_tree, em);
4923 /* it is possible that someone inserted the extent into the tree
4924 * while we had the lock dropped. It is also possible that
4925 * an overlapping map exists in the tree
4927 if (ret == -EEXIST) {
4928 struct extent_map *existing;
4932 existing = lookup_extent_mapping(em_tree, start, len);
4933 if (existing && (existing->start > start ||
4934 existing->start + existing->len <= start)) {
4935 free_extent_map(existing);
4939 existing = lookup_extent_mapping(em_tree, em->start,
4942 err = merge_extent_mapping(em_tree, existing,
4945 free_extent_map(existing);
4947 free_extent_map(em);
4952 free_extent_map(em);
4956 free_extent_map(em);
4961 write_unlock(&em_tree->lock);
4964 btrfs_free_path(path);
4966 ret = btrfs_end_transaction(trans, root);
4971 free_extent_map(em);
4972 return ERR_PTR(err);
4977 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
4978 const struct iovec *iov, loff_t offset,
4979 unsigned long nr_segs)
4984 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
4985 __u64 start, __u64 len)
4987 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
4990 int btrfs_readpage(struct file *file, struct page *page)
4992 struct extent_io_tree *tree;
4993 tree = &BTRFS_I(page->mapping->host)->io_tree;
4994 return extent_read_full_page(tree, page, btrfs_get_extent);
4997 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
4999 struct extent_io_tree *tree;
5002 if (current->flags & PF_MEMALLOC) {
5003 redirty_page_for_writepage(wbc, page);
5007 tree = &BTRFS_I(page->mapping->host)->io_tree;
5008 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
5011 int btrfs_writepages(struct address_space *mapping,
5012 struct writeback_control *wbc)
5014 struct extent_io_tree *tree;
5016 tree = &BTRFS_I(mapping->host)->io_tree;
5017 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
5021 btrfs_readpages(struct file *file, struct address_space *mapping,
5022 struct list_head *pages, unsigned nr_pages)
5024 struct extent_io_tree *tree;
5025 tree = &BTRFS_I(mapping->host)->io_tree;
5026 return extent_readpages(tree, mapping, pages, nr_pages,
5029 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5031 struct extent_io_tree *tree;
5032 struct extent_map_tree *map;
5035 tree = &BTRFS_I(page->mapping->host)->io_tree;
5036 map = &BTRFS_I(page->mapping->host)->extent_tree;
5037 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
5039 ClearPagePrivate(page);
5040 set_page_private(page, 0);
5041 page_cache_release(page);
5046 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
5048 if (PageWriteback(page) || PageDirty(page))
5050 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
5053 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
5055 struct extent_io_tree *tree;
5056 struct btrfs_ordered_extent *ordered;
5057 struct extent_state *cached_state = NULL;
5058 u64 page_start = page_offset(page);
5059 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
5063 * we have the page locked, so new writeback can't start,
5064 * and the dirty bit won't be cleared while we are here.
5066 * Wait for IO on this page so that we can safely clear
5067 * the PagePrivate2 bit and do ordered accounting
5069 wait_on_page_writeback(page);
5071 tree = &BTRFS_I(page->mapping->host)->io_tree;
5073 btrfs_releasepage(page, GFP_NOFS);
5076 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5078 ordered = btrfs_lookup_ordered_extent(page->mapping->host,
5082 * IO on this page will never be started, so we need
5083 * to account for any ordered extents now
5085 clear_extent_bit(tree, page_start, page_end,
5086 EXTENT_DIRTY | EXTENT_DELALLOC |
5087 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
5088 &cached_state, GFP_NOFS);
5090 * whoever cleared the private bit is responsible
5091 * for the finish_ordered_io
5093 if (TestClearPagePrivate2(page)) {
5094 btrfs_finish_ordered_io(page->mapping->host,
5095 page_start, page_end);
5097 btrfs_put_ordered_extent(ordered);
5098 cached_state = NULL;
5099 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
5102 clear_extent_bit(tree, page_start, page_end,
5103 EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
5104 EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
5105 __btrfs_releasepage(page, GFP_NOFS);
5107 ClearPageChecked(page);
5108 if (PagePrivate(page)) {
5109 ClearPagePrivate(page);
5110 set_page_private(page, 0);
5111 page_cache_release(page);
5116 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
5117 * called from a page fault handler when a page is first dirtied. Hence we must
5118 * be careful to check for EOF conditions here. We set the page up correctly
5119 * for a written page which means we get ENOSPC checking when writing into
5120 * holes and correct delalloc and unwritten extent mapping on filesystems that
5121 * support these features.
5123 * We are not allowed to take the i_mutex here so we have to play games to
5124 * protect against truncate races as the page could now be beyond EOF. Because
5125 * vmtruncate() writes the inode size before removing pages, once we have the
5126 * page lock we can determine safely if the page is beyond EOF. If it is not
5127 * beyond EOF, then the page is guaranteed safe against truncation until we
5130 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
5132 struct page *page = vmf->page;
5133 struct inode *inode = fdentry(vma->vm_file)->d_inode;
5134 struct btrfs_root *root = BTRFS_I(inode)->root;
5135 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5136 struct btrfs_ordered_extent *ordered;
5137 struct extent_state *cached_state = NULL;
5139 unsigned long zero_start;
5145 ret = btrfs_check_data_free_space(root, inode, PAGE_CACHE_SIZE);
5149 else /* -ENOSPC, -EIO, etc */
5150 ret = VM_FAULT_SIGBUS;
5154 ret = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
5156 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5157 ret = VM_FAULT_SIGBUS;
5161 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
5164 size = i_size_read(inode);
5165 page_start = page_offset(page);
5166 page_end = page_start + PAGE_CACHE_SIZE - 1;
5168 if ((page->mapping != inode->i_mapping) ||
5169 (page_start >= size)) {
5170 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5171 /* page got truncated out from underneath us */
5174 wait_on_page_writeback(page);
5176 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
5178 set_page_extent_mapped(page);
5181 * we can't set the delalloc bits if there are pending ordered
5182 * extents. Drop our locks and wait for them to finish
5184 ordered = btrfs_lookup_ordered_extent(inode, page_start);
5186 unlock_extent_cached(io_tree, page_start, page_end,
5187 &cached_state, GFP_NOFS);
5189 btrfs_start_ordered_extent(inode, ordered, 1);
5190 btrfs_put_ordered_extent(ordered);
5195 * XXX - page_mkwrite gets called every time the page is dirtied, even
5196 * if it was already dirty, so for space accounting reasons we need to
5197 * clear any delalloc bits for the range we are fixing to save. There
5198 * is probably a better way to do this, but for now keep consistent with
5199 * prepare_pages in the normal write path.
5201 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
5202 EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
5203 0, 0, &cached_state, GFP_NOFS);
5205 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
5208 unlock_extent_cached(io_tree, page_start, page_end,
5209 &cached_state, GFP_NOFS);
5210 ret = VM_FAULT_SIGBUS;
5211 btrfs_free_reserved_data_space(root, inode, PAGE_CACHE_SIZE);
5216 /* page is wholly or partially inside EOF */
5217 if (page_start + PAGE_CACHE_SIZE > size)
5218 zero_start = size & ~PAGE_CACHE_MASK;
5220 zero_start = PAGE_CACHE_SIZE;
5222 if (zero_start != PAGE_CACHE_SIZE) {
5224 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
5225 flush_dcache_page(page);
5228 ClearPageChecked(page);
5229 set_page_dirty(page);
5230 SetPageUptodate(page);
5232 BTRFS_I(inode)->last_trans = root->fs_info->generation;
5233 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
5235 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
5238 btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
5240 return VM_FAULT_LOCKED;
5246 static void btrfs_truncate(struct inode *inode)
5248 struct btrfs_root *root = BTRFS_I(inode)->root;
5250 struct btrfs_trans_handle *trans;
5252 u64 mask = root->sectorsize - 1;
5254 if (!S_ISREG(inode->i_mode)) {
5259 ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
5263 btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
5264 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
5266 trans = btrfs_start_transaction(root, 1);
5267 btrfs_set_trans_block_group(trans, inode);
5270 * setattr is responsible for setting the ordered_data_close flag,
5271 * but that is only tested during the last file release. That
5272 * could happen well after the next commit, leaving a great big
5273 * window where new writes may get lost if someone chooses to write
5274 * to this file after truncating to zero
5276 * The inode doesn't have any dirty data here, and so if we commit
5277 * this is a noop. If someone immediately starts writing to the inode
5278 * it is very likely we'll catch some of their writes in this
5279 * transaction, and the commit will find this file on the ordered
5280 * data list with good things to send down.
5282 * This is a best effort solution, there is still a window where
5283 * using truncate to replace the contents of the file will
5284 * end up with a zero length file after a crash.
5286 if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
5287 btrfs_add_ordered_operation(trans, root, inode);
5290 ret = btrfs_truncate_inode_items(trans, root, inode,
5292 BTRFS_EXTENT_DATA_KEY);
5296 ret = btrfs_update_inode(trans, root, inode);
5299 nr = trans->blocks_used;
5300 btrfs_end_transaction(trans, root);
5301 btrfs_btree_balance_dirty(root, nr);
5303 trans = btrfs_start_transaction(root, 1);
5304 btrfs_set_trans_block_group(trans, inode);
5307 if (ret == 0 && inode->i_nlink > 0) {
5308 ret = btrfs_orphan_del(trans, inode);
5312 ret = btrfs_update_inode(trans, root, inode);
5315 nr = trans->blocks_used;
5316 ret = btrfs_end_transaction_throttle(trans, root);
5318 btrfs_btree_balance_dirty(root, nr);
5322 * create a new subvolume directory/inode (helper for the ioctl).
5324 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
5325 struct btrfs_root *new_root,
5326 u64 new_dirid, u64 alloc_hint)
5328 struct inode *inode;
5332 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
5333 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
5335 return PTR_ERR(inode);
5336 inode->i_op = &btrfs_dir_inode_operations;
5337 inode->i_fop = &btrfs_dir_file_operations;
5340 btrfs_i_size_write(inode, 0);
5342 err = btrfs_update_inode(trans, new_root, inode);
5349 /* helper function for file defrag and space balancing. This
5350 * forces readahead on a given range of bytes in an inode
5352 unsigned long btrfs_force_ra(struct address_space *mapping,
5353 struct file_ra_state *ra, struct file *file,
5354 pgoff_t offset, pgoff_t last_index)
5356 pgoff_t req_size = last_index - offset + 1;
5358 page_cache_sync_readahead(mapping, ra, file, offset, req_size);
5359 return offset + req_size;
5362 struct inode *btrfs_alloc_inode(struct super_block *sb)
5364 struct btrfs_inode *ei;
5365 struct inode *inode;
5367 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
5372 ei->space_info = NULL;
5376 ei->last_sub_trans = 0;
5377 ei->logged_trans = 0;
5378 ei->delalloc_bytes = 0;
5379 ei->reserved_bytes = 0;
5380 ei->disk_i_size = 0;
5382 ei->index_cnt = (u64)-1;
5383 ei->last_unlink_trans = 0;
5385 spin_lock_init(&ei->accounting_lock);
5386 ei->outstanding_extents = 0;
5387 ei->reserved_extents = 0;
5389 ei->ordered_data_close = 0;
5390 ei->dummy_inode = 0;
5391 ei->force_compress = 0;
5393 inode = &ei->vfs_inode;
5394 extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
5395 extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
5396 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
5397 mutex_init(&ei->log_mutex);
5398 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
5399 INIT_LIST_HEAD(&ei->i_orphan);
5400 INIT_LIST_HEAD(&ei->delalloc_inodes);
5401 INIT_LIST_HEAD(&ei->ordered_operations);
5402 RB_CLEAR_NODE(&ei->rb_node);
5407 void btrfs_destroy_inode(struct inode *inode)
5409 struct btrfs_ordered_extent *ordered;
5410 struct btrfs_root *root = BTRFS_I(inode)->root;
5412 WARN_ON(!list_empty(&inode->i_dentry));
5413 WARN_ON(inode->i_data.nrpages);
5416 * This can happen where we create an inode, but somebody else also
5417 * created the same inode and we need to destroy the one we already
5424 * Make sure we're properly removed from the ordered operation
5428 if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
5429 spin_lock(&root->fs_info->ordered_extent_lock);
5430 list_del_init(&BTRFS_I(inode)->ordered_operations);
5431 spin_unlock(&root->fs_info->ordered_extent_lock);
5434 spin_lock(&root->list_lock);
5435 if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
5436 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
5438 list_del_init(&BTRFS_I(inode)->i_orphan);
5440 spin_unlock(&root->list_lock);
5443 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
5447 printk(KERN_ERR "btrfs found ordered "
5448 "extent %llu %llu on inode cleanup\n",
5449 (unsigned long long)ordered->file_offset,
5450 (unsigned long long)ordered->len);
5451 btrfs_remove_ordered_extent(inode, ordered);
5452 btrfs_put_ordered_extent(ordered);
5453 btrfs_put_ordered_extent(ordered);
5456 inode_tree_del(inode);
5457 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
5459 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5462 void btrfs_drop_inode(struct inode *inode)
5464 struct btrfs_root *root = BTRFS_I(inode)->root;
5465 if (inode->i_nlink > 0 && btrfs_root_refs(&root->root_item) == 0)
5466 generic_delete_inode(inode);
5468 generic_drop_inode(inode);
5471 static void init_once(void *foo)
5473 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
5475 inode_init_once(&ei->vfs_inode);
5478 void btrfs_destroy_cachep(void)
5480 if (btrfs_inode_cachep)
5481 kmem_cache_destroy(btrfs_inode_cachep);
5482 if (btrfs_trans_handle_cachep)
5483 kmem_cache_destroy(btrfs_trans_handle_cachep);
5484 if (btrfs_transaction_cachep)
5485 kmem_cache_destroy(btrfs_transaction_cachep);
5486 if (btrfs_path_cachep)
5487 kmem_cache_destroy(btrfs_path_cachep);
5490 int btrfs_init_cachep(void)
5492 btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
5493 sizeof(struct btrfs_inode), 0,
5494 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
5495 if (!btrfs_inode_cachep)
5498 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
5499 sizeof(struct btrfs_trans_handle), 0,
5500 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5501 if (!btrfs_trans_handle_cachep)
5504 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
5505 sizeof(struct btrfs_transaction), 0,
5506 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5507 if (!btrfs_transaction_cachep)
5510 btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
5511 sizeof(struct btrfs_path), 0,
5512 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
5513 if (!btrfs_path_cachep)
5518 btrfs_destroy_cachep();
5522 static int btrfs_getattr(struct vfsmount *mnt,
5523 struct dentry *dentry, struct kstat *stat)
5525 struct inode *inode = dentry->d_inode;
5526 generic_fillattr(inode, stat);
5527 stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
5528 stat->blksize = PAGE_CACHE_SIZE;
5529 stat->blocks = (inode_get_bytes(inode) +
5530 BTRFS_I(inode)->delalloc_bytes) >> 9;
5534 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
5535 struct inode *new_dir, struct dentry *new_dentry)
5537 struct btrfs_trans_handle *trans;
5538 struct btrfs_root *root = BTRFS_I(old_dir)->root;
5539 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
5540 struct inode *new_inode = new_dentry->d_inode;
5541 struct inode *old_inode = old_dentry->d_inode;
5542 struct timespec ctime = CURRENT_TIME;
5547 if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5550 /* we only allow rename subvolume link between subvolumes */
5551 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
5554 if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
5555 (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
5558 if (S_ISDIR(old_inode->i_mode) && new_inode &&
5559 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
5562 * we're using rename to replace one file with another.
5563 * and the replacement file is large. Start IO on it now so
5564 * we don't add too much work to the end of the transaction
5566 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
5567 old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
5568 filemap_flush(old_inode->i_mapping);
5570 /* close the racy window with snapshot create/destroy ioctl */
5571 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5572 down_read(&root->fs_info->subvol_sem);
5574 * We want to reserve the absolute worst case amount of items. So if
5575 * both inodes are subvols and we need to unlink them then that would
5576 * require 4 item modifications, but if they are both normal inodes it
5577 * would require 5 item modifications, so we'll assume their normal
5578 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
5579 * should cover the worst case number of items we'll modify.
5581 trans = btrfs_start_transaction(root, 20);
5583 return PTR_ERR(trans);
5585 btrfs_set_trans_block_group(trans, new_dir);
5588 btrfs_record_root_in_trans(trans, dest);
5590 ret = btrfs_set_inode_index(new_dir, &index);
5594 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5595 /* force full log commit if subvolume involved. */
5596 root->fs_info->last_trans_log_full_commit = trans->transid;
5598 ret = btrfs_insert_inode_ref(trans, dest,
5599 new_dentry->d_name.name,
5600 new_dentry->d_name.len,
5602 new_dir->i_ino, index);
5606 * this is an ugly little race, but the rename is required
5607 * to make sure that if we crash, the inode is either at the
5608 * old name or the new one. pinning the log transaction lets
5609 * us make sure we don't allow a log commit to come in after
5610 * we unlink the name but before we add the new name back in.
5612 btrfs_pin_log_trans(root);
5615 * make sure the inode gets flushed if it is replacing
5618 if (new_inode && new_inode->i_size &&
5619 old_inode && S_ISREG(old_inode->i_mode)) {
5620 btrfs_add_ordered_operation(trans, root, old_inode);
5623 old_dir->i_ctime = old_dir->i_mtime = ctime;
5624 new_dir->i_ctime = new_dir->i_mtime = ctime;
5625 old_inode->i_ctime = ctime;
5627 if (old_dentry->d_parent != new_dentry->d_parent)
5628 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
5630 if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
5631 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
5632 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
5633 old_dentry->d_name.name,
5634 old_dentry->d_name.len);
5636 btrfs_inc_nlink(old_dentry->d_inode);
5637 ret = btrfs_unlink_inode(trans, root, old_dir,
5638 old_dentry->d_inode,
5639 old_dentry->d_name.name,
5640 old_dentry->d_name.len);
5645 new_inode->i_ctime = CURRENT_TIME;
5646 if (unlikely(new_inode->i_ino ==
5647 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
5648 root_objectid = BTRFS_I(new_inode)->location.objectid;
5649 ret = btrfs_unlink_subvol(trans, dest, new_dir,
5651 new_dentry->d_name.name,
5652 new_dentry->d_name.len);
5653 BUG_ON(new_inode->i_nlink == 0);
5655 ret = btrfs_unlink_inode(trans, dest, new_dir,
5656 new_dentry->d_inode,
5657 new_dentry->d_name.name,
5658 new_dentry->d_name.len);
5661 if (new_inode->i_nlink == 0) {
5662 ret = btrfs_orphan_add(trans, new_dentry->d_inode);
5667 ret = btrfs_add_link(trans, new_dir, old_inode,
5668 new_dentry->d_name.name,
5669 new_dentry->d_name.len, 0, index);
5672 if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
5673 btrfs_log_new_name(trans, old_inode, old_dir,
5674 new_dentry->d_parent);
5675 btrfs_end_log_trans(root);
5678 btrfs_end_transaction_throttle(trans, root);
5680 if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
5681 up_read(&root->fs_info->subvol_sem);
5687 * some fairly slow code that needs optimization. This walks the list
5688 * of all the inodes with pending delalloc and forces them to disk.
5690 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
5692 struct list_head *head = &root->fs_info->delalloc_inodes;
5693 struct btrfs_inode *binode;
5694 struct inode *inode;
5696 if (root->fs_info->sb->s_flags & MS_RDONLY)
5699 spin_lock(&root->fs_info->delalloc_lock);
5700 while (!list_empty(head)) {
5701 binode = list_entry(head->next, struct btrfs_inode,
5703 inode = igrab(&binode->vfs_inode);
5705 list_del_init(&binode->delalloc_inodes);
5706 spin_unlock(&root->fs_info->delalloc_lock);
5708 filemap_flush(inode->i_mapping);
5710 btrfs_add_delayed_iput(inode);
5715 spin_lock(&root->fs_info->delalloc_lock);
5717 spin_unlock(&root->fs_info->delalloc_lock);
5719 /* the filemap_flush will queue IO into the worker threads, but
5720 * we have to make sure the IO is actually started and that
5721 * ordered extents get created before we return
5723 atomic_inc(&root->fs_info->async_submit_draining);
5724 while (atomic_read(&root->fs_info->nr_async_submits) ||
5725 atomic_read(&root->fs_info->async_delalloc_pages)) {
5726 wait_event(root->fs_info->async_submit_wait,
5727 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
5728 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
5730 atomic_dec(&root->fs_info->async_submit_draining);
5734 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput)
5736 struct btrfs_inode *binode;
5737 struct inode *inode = NULL;
5739 spin_lock(&root->fs_info->delalloc_lock);
5740 while (!list_empty(&root->fs_info->delalloc_inodes)) {
5741 binode = list_entry(root->fs_info->delalloc_inodes.next,
5742 struct btrfs_inode, delalloc_inodes);
5743 inode = igrab(&binode->vfs_inode);
5745 list_move_tail(&binode->delalloc_inodes,
5746 &root->fs_info->delalloc_inodes);
5750 list_del_init(&binode->delalloc_inodes);
5751 cond_resched_lock(&root->fs_info->delalloc_lock);
5753 spin_unlock(&root->fs_info->delalloc_lock);
5756 write_inode_now(inode, 0);
5758 btrfs_add_delayed_iput(inode);
5766 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
5767 const char *symname)
5769 struct btrfs_trans_handle *trans;
5770 struct btrfs_root *root = BTRFS_I(dir)->root;
5771 struct btrfs_path *path;
5772 struct btrfs_key key;
5773 struct inode *inode = NULL;
5781 struct btrfs_file_extent_item *ei;
5782 struct extent_buffer *leaf;
5783 unsigned long nr = 0;
5785 name_len = strlen(symname) + 1;
5786 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
5787 return -ENAMETOOLONG;
5789 err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
5793 * 2 items for inode item and ref
5794 * 2 items for dir items
5795 * 1 item for xattr if selinux is on
5797 trans = btrfs_start_transaction(root, 5);
5799 return PTR_ERR(trans);
5801 btrfs_set_trans_block_group(trans, dir);
5803 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5805 dentry->d_parent->d_inode->i_ino, objectid,
5806 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
5808 err = PTR_ERR(inode);
5812 err = btrfs_init_inode_security(trans, inode, dir);
5818 btrfs_set_trans_block_group(trans, inode);
5819 err = btrfs_add_nondir(trans, dentry, inode, 0, index);
5823 inode->i_mapping->a_ops = &btrfs_aops;
5824 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5825 inode->i_fop = &btrfs_file_operations;
5826 inode->i_op = &btrfs_file_inode_operations;
5827 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5829 btrfs_update_inode_block_group(trans, inode);
5830 btrfs_update_inode_block_group(trans, dir);
5834 path = btrfs_alloc_path();
5836 key.objectid = inode->i_ino;
5838 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
5839 datasize = btrfs_file_extent_calc_inline_size(name_len);
5840 err = btrfs_insert_empty_item(trans, root, path, &key,
5846 leaf = path->nodes[0];
5847 ei = btrfs_item_ptr(leaf, path->slots[0],
5848 struct btrfs_file_extent_item);
5849 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
5850 btrfs_set_file_extent_type(leaf, ei,
5851 BTRFS_FILE_EXTENT_INLINE);
5852 btrfs_set_file_extent_encryption(leaf, ei, 0);
5853 btrfs_set_file_extent_compression(leaf, ei, 0);
5854 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
5855 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
5857 ptr = btrfs_file_extent_inline_start(ei);
5858 write_extent_buffer(leaf, symname, ptr, name_len);
5859 btrfs_mark_buffer_dirty(leaf);
5860 btrfs_free_path(path);
5862 inode->i_op = &btrfs_symlink_inode_operations;
5863 inode->i_mapping->a_ops = &btrfs_symlink_aops;
5864 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5865 inode_set_bytes(inode, name_len);
5866 btrfs_i_size_write(inode, name_len - 1);
5867 err = btrfs_update_inode(trans, root, inode);
5872 nr = trans->blocks_used;
5873 btrfs_end_transaction_throttle(trans, root);
5875 inode_dec_link_count(inode);
5878 btrfs_btree_balance_dirty(root, nr);
5882 static int prealloc_file_range(struct inode *inode, u64 start, u64 end,
5883 u64 alloc_hint, int mode, loff_t actual_len)
5885 struct btrfs_trans_handle *trans;
5886 struct btrfs_root *root = BTRFS_I(inode)->root;
5887 struct btrfs_key ins;
5888 u64 cur_offset = start;
5889 u64 num_bytes = end - start;
5893 while (num_bytes > 0) {
5894 trans = btrfs_start_transaction(root, 3);
5895 if (IS_ERR(trans)) {
5896 ret = PTR_ERR(trans);
5900 ret = btrfs_reserve_extent(trans, root, num_bytes,
5901 root->sectorsize, 0, alloc_hint,
5904 btrfs_end_transaction(trans, root);
5908 ret = insert_reserved_file_extent(trans, inode,
5909 cur_offset, ins.objectid,
5910 ins.offset, ins.offset,
5911 ins.offset, 0, 0, 0,
5912 BTRFS_FILE_EXTENT_PREALLOC);
5914 btrfs_drop_extent_cache(inode, cur_offset,
5915 cur_offset + ins.offset -1, 0);
5917 num_bytes -= ins.offset;
5918 cur_offset += ins.offset;
5919 alloc_hint = ins.objectid + ins.offset;
5921 inode->i_ctime = CURRENT_TIME;
5922 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
5923 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
5924 (actual_len > inode->i_size) &&
5925 (cur_offset > inode->i_size)) {
5927 if (cur_offset > actual_len)
5928 i_size = actual_len;
5930 i_size = cur_offset;
5931 i_size_write(inode, i_size);
5932 btrfs_ordered_update_i_size(inode, i_size, NULL);
5935 ret = btrfs_update_inode(trans, root, inode);
5938 btrfs_end_transaction(trans, root);
5943 static long btrfs_fallocate(struct inode *inode, int mode,
5944 loff_t offset, loff_t len)
5946 struct extent_state *cached_state = NULL;
5953 u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
5954 struct extent_map *em;
5957 alloc_start = offset & ~mask;
5958 alloc_end = (offset + len + mask) & ~mask;
5961 * wait for ordered IO before we have any locks. We'll loop again
5962 * below with the locks held.
5964 btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
5966 mutex_lock(&inode->i_mutex);
5967 if (alloc_start > inode->i_size) {
5968 ret = btrfs_cont_expand(inode, alloc_start);
5973 ret = btrfs_check_data_free_space(BTRFS_I(inode)->root, inode,
5974 alloc_end - alloc_start);
5978 locked_end = alloc_end - 1;
5980 struct btrfs_ordered_extent *ordered;
5982 /* the extent lock is ordered inside the running
5985 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
5986 locked_end, 0, &cached_state, GFP_NOFS);
5987 ordered = btrfs_lookup_first_ordered_extent(inode,
5990 ordered->file_offset + ordered->len > alloc_start &&
5991 ordered->file_offset < alloc_end) {
5992 btrfs_put_ordered_extent(ordered);
5993 unlock_extent_cached(&BTRFS_I(inode)->io_tree,
5994 alloc_start, locked_end,
5995 &cached_state, GFP_NOFS);
5997 * we can't wait on the range with the transaction
5998 * running or with the extent lock held
6000 btrfs_wait_ordered_range(inode, alloc_start,
6001 alloc_end - alloc_start);
6004 btrfs_put_ordered_extent(ordered);
6009 cur_offset = alloc_start;
6011 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
6012 alloc_end - cur_offset, 0);
6013 BUG_ON(IS_ERR(em) || !em);
6014 last_byte = min(extent_map_end(em), alloc_end);
6015 last_byte = (last_byte + mask) & ~mask;
6016 if (em->block_start == EXTENT_MAP_HOLE ||
6017 (cur_offset >= inode->i_size &&
6018 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6019 ret = prealloc_file_range(inode,
6020 cur_offset, last_byte,
6021 alloc_hint, mode, offset+len);
6023 free_extent_map(em);
6027 if (em->block_start <= EXTENT_MAP_LAST_BYTE)
6028 alloc_hint = em->block_start;
6029 free_extent_map(em);
6031 cur_offset = last_byte;
6032 if (cur_offset >= alloc_end) {
6037 unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
6038 &cached_state, GFP_NOFS);
6040 btrfs_free_reserved_data_space(BTRFS_I(inode)->root, inode,
6041 alloc_end - alloc_start);
6043 mutex_unlock(&inode->i_mutex);
6047 static int btrfs_set_page_dirty(struct page *page)
6049 return __set_page_dirty_nobuffers(page);
6052 static int btrfs_permission(struct inode *inode, int mask)
6054 if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
6056 return generic_permission(inode, mask, btrfs_check_acl);
6059 static const struct inode_operations btrfs_dir_inode_operations = {
6060 .getattr = btrfs_getattr,
6061 .lookup = btrfs_lookup,
6062 .create = btrfs_create,
6063 .unlink = btrfs_unlink,
6065 .mkdir = btrfs_mkdir,
6066 .rmdir = btrfs_rmdir,
6067 .rename = btrfs_rename,
6068 .symlink = btrfs_symlink,
6069 .setattr = btrfs_setattr,
6070 .mknod = btrfs_mknod,
6071 .setxattr = btrfs_setxattr,
6072 .getxattr = btrfs_getxattr,
6073 .listxattr = btrfs_listxattr,
6074 .removexattr = btrfs_removexattr,
6075 .permission = btrfs_permission,
6077 static const struct inode_operations btrfs_dir_ro_inode_operations = {
6078 .lookup = btrfs_lookup,
6079 .permission = btrfs_permission,
6082 static const struct file_operations btrfs_dir_file_operations = {
6083 .llseek = generic_file_llseek,
6084 .read = generic_read_dir,
6085 .readdir = btrfs_real_readdir,
6086 .unlocked_ioctl = btrfs_ioctl,
6087 #ifdef CONFIG_COMPAT
6088 .compat_ioctl = btrfs_ioctl,
6090 .release = btrfs_release_file,
6091 .fsync = btrfs_sync_file,
6094 static struct extent_io_ops btrfs_extent_io_ops = {
6095 .fill_delalloc = run_delalloc_range,
6096 .submit_bio_hook = btrfs_submit_bio_hook,
6097 .merge_bio_hook = btrfs_merge_bio_hook,
6098 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
6099 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
6100 .writepage_start_hook = btrfs_writepage_start_hook,
6101 .readpage_io_failed_hook = btrfs_io_failed_hook,
6102 .set_bit_hook = btrfs_set_bit_hook,
6103 .clear_bit_hook = btrfs_clear_bit_hook,
6104 .merge_extent_hook = btrfs_merge_extent_hook,
6105 .split_extent_hook = btrfs_split_extent_hook,
6109 * btrfs doesn't support the bmap operation because swapfiles
6110 * use bmap to make a mapping of extents in the file. They assume
6111 * these extents won't change over the life of the file and they
6112 * use the bmap result to do IO directly to the drive.
6114 * the btrfs bmap call would return logical addresses that aren't
6115 * suitable for IO and they also will change frequently as COW
6116 * operations happen. So, swapfile + btrfs == corruption.
6118 * For now we're avoiding this by dropping bmap.
6120 static const struct address_space_operations btrfs_aops = {
6121 .readpage = btrfs_readpage,
6122 .writepage = btrfs_writepage,
6123 .writepages = btrfs_writepages,
6124 .readpages = btrfs_readpages,
6125 .sync_page = block_sync_page,
6126 .direct_IO = btrfs_direct_IO,
6127 .invalidatepage = btrfs_invalidatepage,
6128 .releasepage = btrfs_releasepage,
6129 .set_page_dirty = btrfs_set_page_dirty,
6130 .error_remove_page = generic_error_remove_page,
6133 static const struct address_space_operations btrfs_symlink_aops = {
6134 .readpage = btrfs_readpage,
6135 .writepage = btrfs_writepage,
6136 .invalidatepage = btrfs_invalidatepage,
6137 .releasepage = btrfs_releasepage,
6140 static const struct inode_operations btrfs_file_inode_operations = {
6141 .truncate = btrfs_truncate,
6142 .getattr = btrfs_getattr,
6143 .setattr = btrfs_setattr,
6144 .setxattr = btrfs_setxattr,
6145 .getxattr = btrfs_getxattr,
6146 .listxattr = btrfs_listxattr,
6147 .removexattr = btrfs_removexattr,
6148 .permission = btrfs_permission,
6149 .fallocate = btrfs_fallocate,
6150 .fiemap = btrfs_fiemap,
6152 static const struct inode_operations btrfs_special_inode_operations = {
6153 .getattr = btrfs_getattr,
6154 .setattr = btrfs_setattr,
6155 .permission = btrfs_permission,
6156 .setxattr = btrfs_setxattr,
6157 .getxattr = btrfs_getxattr,
6158 .listxattr = btrfs_listxattr,
6159 .removexattr = btrfs_removexattr,
6161 static const struct inode_operations btrfs_symlink_inode_operations = {
6162 .readlink = generic_readlink,
6163 .follow_link = page_follow_link_light,
6164 .put_link = page_put_link,
6165 .permission = btrfs_permission,
6166 .setxattr = btrfs_setxattr,
6167 .getxattr = btrfs_getxattr,
6168 .listxattr = btrfs_listxattr,
6169 .removexattr = btrfs_removexattr,
6172 const struct dentry_operations btrfs_dentry_operations = {
6173 .d_delete = btrfs_dentry_delete,