2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
65 struct btrfs_iget_args {
66 struct btrfs_key *location;
67 struct btrfs_root *root;
70 struct btrfs_dio_data {
71 u64 outstanding_extents;
73 u64 unsubmitted_oe_range_start;
74 u64 unsubmitted_oe_range_end;
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_transaction_cachep;
90 struct kmem_cache *btrfs_path_cachep;
91 struct kmem_cache *btrfs_free_space_cachep;
94 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
95 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
96 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
97 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
98 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
99 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
100 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
101 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
104 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
105 static int btrfs_truncate(struct inode *inode);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
107 static noinline int cow_file_range(struct inode *inode,
108 struct page *locked_page,
109 u64 start, u64 end, u64 delalloc_end,
110 int *page_started, unsigned long *nr_written,
111 int unlock, struct btrfs_dedupe_hash *hash);
112 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
113 u64 len, u64 orig_start,
114 u64 block_start, u64 block_len,
115 u64 orig_block_len, u64 ram_bytes,
118 static int btrfs_dirty_inode(struct inode *inode);
120 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
121 void btrfs_test_inode_set_ops(struct inode *inode)
123 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
127 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
128 struct inode *inode, struct inode *dir,
129 const struct qstr *qstr)
133 err = btrfs_init_acl(trans, inode, dir);
135 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
140 * this does all the hard work for inserting an inline extent into
141 * the btree. The caller should have done a btrfs_drop_extents so that
142 * no overlapping inline items exist in the btree
144 static int insert_inline_extent(struct btrfs_trans_handle *trans,
145 struct btrfs_path *path, int extent_inserted,
146 struct btrfs_root *root, struct inode *inode,
147 u64 start, size_t size, size_t compressed_size,
149 struct page **compressed_pages)
151 struct extent_buffer *leaf;
152 struct page *page = NULL;
155 struct btrfs_file_extent_item *ei;
158 size_t cur_size = size;
159 unsigned long offset;
161 if (compressed_size && compressed_pages)
162 cur_size = compressed_size;
164 inode_add_bytes(inode, size);
166 if (!extent_inserted) {
167 struct btrfs_key key;
170 key.objectid = btrfs_ino(inode);
172 key.type = BTRFS_EXTENT_DATA_KEY;
174 datasize = btrfs_file_extent_calc_inline_size(cur_size);
175 path->leave_spinning = 1;
176 ret = btrfs_insert_empty_item(trans, root, path, &key,
183 leaf = path->nodes[0];
184 ei = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_file_extent_item);
186 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
187 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
188 btrfs_set_file_extent_encryption(leaf, ei, 0);
189 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
190 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
191 ptr = btrfs_file_extent_inline_start(ei);
193 if (compress_type != BTRFS_COMPRESS_NONE) {
196 while (compressed_size > 0) {
197 cpage = compressed_pages[i];
198 cur_size = min_t(unsigned long, compressed_size,
201 kaddr = kmap_atomic(cpage);
202 write_extent_buffer(leaf, kaddr, ptr, cur_size);
203 kunmap_atomic(kaddr);
207 compressed_size -= cur_size;
209 btrfs_set_file_extent_compression(leaf, ei,
212 page = find_get_page(inode->i_mapping,
213 start >> PAGE_SHIFT);
214 btrfs_set_file_extent_compression(leaf, ei, 0);
215 kaddr = kmap_atomic(page);
216 offset = start & (PAGE_SIZE - 1);
217 write_extent_buffer(leaf, kaddr + offset, ptr, size);
218 kunmap_atomic(kaddr);
221 btrfs_mark_buffer_dirty(leaf);
222 btrfs_release_path(path);
225 * we're an inline extent, so nobody can
226 * extend the file past i_size without locking
227 * a page we already have locked.
229 * We must do any isize and inode updates
230 * before we unlock the pages. Otherwise we
231 * could end up racing with unlink.
233 BTRFS_I(inode)->disk_i_size = inode->i_size;
234 ret = btrfs_update_inode(trans, root, inode);
243 * conditionally insert an inline extent into the file. This
244 * does the checks required to make sure the data is small enough
245 * to fit as an inline extent.
247 static noinline int cow_file_range_inline(struct btrfs_root *root,
248 struct inode *inode, u64 start,
249 u64 end, size_t compressed_size,
251 struct page **compressed_pages)
253 struct btrfs_trans_handle *trans;
254 u64 isize = i_size_read(inode);
255 u64 actual_end = min(end + 1, isize);
256 u64 inline_len = actual_end - start;
257 u64 aligned_end = ALIGN(end, root->sectorsize);
258 u64 data_len = inline_len;
260 struct btrfs_path *path;
261 int extent_inserted = 0;
262 u32 extent_item_size;
265 data_len = compressed_size;
268 actual_end > root->sectorsize ||
269 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
271 (actual_end & (root->sectorsize - 1)) == 0) ||
273 data_len > root->fs_info->max_inline) {
277 path = btrfs_alloc_path();
281 trans = btrfs_join_transaction(root);
283 btrfs_free_path(path);
284 return PTR_ERR(trans);
286 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
288 if (compressed_size && compressed_pages)
289 extent_item_size = btrfs_file_extent_calc_inline_size(
292 extent_item_size = btrfs_file_extent_calc_inline_size(
295 ret = __btrfs_drop_extents(trans, root, inode, path,
296 start, aligned_end, NULL,
297 1, 1, extent_item_size, &extent_inserted);
299 btrfs_abort_transaction(trans, root, ret);
303 if (isize > actual_end)
304 inline_len = min_t(u64, isize, actual_end);
305 ret = insert_inline_extent(trans, path, extent_inserted,
307 inline_len, compressed_size,
308 compress_type, compressed_pages);
309 if (ret && ret != -ENOSPC) {
310 btrfs_abort_transaction(trans, root, ret);
312 } else if (ret == -ENOSPC) {
317 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
318 btrfs_delalloc_release_metadata(inode, end + 1 - start);
319 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
322 * Don't forget to free the reserved space, as for inlined extent
323 * it won't count as data extent, free them directly here.
324 * And at reserve time, it's always aligned to page size, so
325 * just free one page here.
327 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
328 btrfs_free_path(path);
329 btrfs_end_transaction(trans, root);
333 struct async_extent {
338 unsigned long nr_pages;
340 struct list_head list;
345 struct btrfs_root *root;
346 struct page *locked_page;
349 struct list_head extents;
350 struct btrfs_work work;
353 static noinline int add_async_extent(struct async_cow *cow,
354 u64 start, u64 ram_size,
357 unsigned long nr_pages,
360 struct async_extent *async_extent;
362 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
363 BUG_ON(!async_extent); /* -ENOMEM */
364 async_extent->start = start;
365 async_extent->ram_size = ram_size;
366 async_extent->compressed_size = compressed_size;
367 async_extent->pages = pages;
368 async_extent->nr_pages = nr_pages;
369 async_extent->compress_type = compress_type;
370 list_add_tail(&async_extent->list, &cow->extents);
374 static inline int inode_need_compress(struct inode *inode)
376 struct btrfs_root *root = BTRFS_I(inode)->root;
379 if (btrfs_test_opt(root->fs_info, FORCE_COMPRESS))
381 /* bad compression ratios */
382 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
384 if (btrfs_test_opt(root->fs_info, COMPRESS) ||
385 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
386 BTRFS_I(inode)->force_compress)
392 * we create compressed extents in two phases. The first
393 * phase compresses a range of pages that have already been
394 * locked (both pages and state bits are locked).
396 * This is done inside an ordered work queue, and the compression
397 * is spread across many cpus. The actual IO submission is step
398 * two, and the ordered work queue takes care of making sure that
399 * happens in the same order things were put onto the queue by
400 * writepages and friends.
402 * If this code finds it can't get good compression, it puts an
403 * entry onto the work queue to write the uncompressed bytes. This
404 * makes sure that both compressed inodes and uncompressed inodes
405 * are written in the same order that the flusher thread sent them
408 static noinline void compress_file_range(struct inode *inode,
409 struct page *locked_page,
411 struct async_cow *async_cow,
414 struct btrfs_root *root = BTRFS_I(inode)->root;
416 u64 blocksize = root->sectorsize;
418 u64 isize = i_size_read(inode);
420 struct page **pages = NULL;
421 unsigned long nr_pages;
422 unsigned long nr_pages_ret = 0;
423 unsigned long total_compressed = 0;
424 unsigned long total_in = 0;
425 unsigned long max_compressed = SZ_128K;
426 unsigned long max_uncompressed = SZ_128K;
429 int compress_type = root->fs_info->compress_type;
432 /* if this is a small write inside eof, kick off a defrag */
433 if ((end - start + 1) < SZ_16K &&
434 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
435 btrfs_add_inode_defrag(NULL, inode);
437 actual_end = min_t(u64, isize, end + 1);
440 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
441 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_SIZE);
444 * we don't want to send crud past the end of i_size through
445 * compression, that's just a waste of CPU time. So, if the
446 * end of the file is before the start of our current
447 * requested range of bytes, we bail out to the uncompressed
448 * cleanup code that can deal with all of this.
450 * It isn't really the fastest way to fix things, but this is a
451 * very uncommon corner.
453 if (actual_end <= start)
454 goto cleanup_and_bail_uncompressed;
456 total_compressed = actual_end - start;
459 * skip compression for a small file range(<=blocksize) that
460 * isn't an inline extent, since it doesn't save disk space at all.
462 if (total_compressed <= blocksize &&
463 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
464 goto cleanup_and_bail_uncompressed;
466 /* we want to make sure that amount of ram required to uncompress
467 * an extent is reasonable, so we limit the total size in ram
468 * of a compressed extent to 128k. This is a crucial number
469 * because it also controls how easily we can spread reads across
470 * cpus for decompression.
472 * We also want to make sure the amount of IO required to do
473 * a random read is reasonably small, so we limit the size of
474 * a compressed extent to 128k.
476 total_compressed = min(total_compressed, max_uncompressed);
477 num_bytes = ALIGN(end - start + 1, blocksize);
478 num_bytes = max(blocksize, num_bytes);
483 * we do compression for mount -o compress and when the
484 * inode has not been flagged as nocompress. This flag can
485 * change at any time if we discover bad compression ratios.
487 if (inode_need_compress(inode)) {
489 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
491 /* just bail out to the uncompressed code */
495 if (BTRFS_I(inode)->force_compress)
496 compress_type = BTRFS_I(inode)->force_compress;
499 * we need to call clear_page_dirty_for_io on each
500 * page in the range. Otherwise applications with the file
501 * mmap'd can wander in and change the page contents while
502 * we are compressing them.
504 * If the compression fails for any reason, we set the pages
505 * dirty again later on.
507 extent_range_clear_dirty_for_io(inode, start, end);
509 ret = btrfs_compress_pages(compress_type,
510 inode->i_mapping, start,
511 total_compressed, pages,
512 nr_pages, &nr_pages_ret,
518 unsigned long offset = total_compressed &
520 struct page *page = pages[nr_pages_ret - 1];
523 /* zero the tail end of the last page, we might be
524 * sending it down to disk
527 kaddr = kmap_atomic(page);
528 memset(kaddr + offset, 0,
530 kunmap_atomic(kaddr);
537 /* lets try to make an inline extent */
538 if (ret || total_in < (actual_end - start)) {
539 /* we didn't compress the entire range, try
540 * to make an uncompressed inline extent.
542 ret = cow_file_range_inline(root, inode, start, end,
545 /* try making a compressed inline extent */
546 ret = cow_file_range_inline(root, inode, start, end,
548 compress_type, pages);
551 unsigned long clear_flags = EXTENT_DELALLOC |
553 unsigned long page_error_op;
555 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
556 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
559 * inline extent creation worked or returned error,
560 * we don't need to create any more async work items.
561 * Unlock and free up our temp pages.
563 extent_clear_unlock_delalloc(inode, start, end, NULL,
564 clear_flags, PAGE_UNLOCK |
575 * we aren't doing an inline extent round the compressed size
576 * up to a block size boundary so the allocator does sane
579 total_compressed = ALIGN(total_compressed, blocksize);
582 * one last check to make sure the compression is really a
583 * win, compare the page count read with the blocks on disk
585 total_in = ALIGN(total_in, PAGE_SIZE);
586 if (total_compressed >= total_in) {
589 num_bytes = total_in;
593 * The async work queues will take care of doing actual
594 * allocation on disk for these compressed pages, and
595 * will submit them to the elevator.
597 add_async_extent(async_cow, start, num_bytes,
598 total_compressed, pages, nr_pages_ret,
601 if (start + num_bytes < end) {
612 * the compression code ran but failed to make things smaller,
613 * free any pages it allocated and our page pointer array
615 for (i = 0; i < nr_pages_ret; i++) {
616 WARN_ON(pages[i]->mapping);
621 total_compressed = 0;
624 /* flag the file so we don't compress in the future */
625 if (!btrfs_test_opt(root->fs_info, FORCE_COMPRESS) &&
626 !(BTRFS_I(inode)->force_compress)) {
627 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
630 cleanup_and_bail_uncompressed:
632 * No compression, but we still need to write the pages in the file
633 * we've been given so far. redirty the locked page if it corresponds
634 * to our extent and set things up for the async work queue to run
635 * cow_file_range to do the normal delalloc dance.
637 if (page_offset(locked_page) >= start &&
638 page_offset(locked_page) <= end)
639 __set_page_dirty_nobuffers(locked_page);
640 /* unlocked later on in the async handlers */
643 extent_range_redirty_for_io(inode, start, end);
644 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
645 BTRFS_COMPRESS_NONE);
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
658 static void free_async_extent_pages(struct async_extent *async_extent)
662 if (!async_extent->pages)
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 put_page(async_extent->pages[i]);
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
683 struct async_extent *async_extent;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
698 io_tree = &BTRFS_I(inode)->io_tree;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 async_extent->start +
716 async_extent->ram_size - 1,
717 &page_started, &nr_written, 0,
723 * if page_started, cow_file_range inserted an
724 * inline extent and took care of all the unlocking
725 * and IO for us. Otherwise, we need to submit
726 * all those pages down to the drive.
728 if (!page_started && !ret)
729 extent_write_locked_range(io_tree,
730 inode, async_extent->start,
731 async_extent->start +
732 async_extent->ram_size - 1,
736 unlock_page(async_cow->locked_page);
742 lock_extent(io_tree, async_extent->start,
743 async_extent->start + async_extent->ram_size - 1);
745 ret = btrfs_reserve_extent(root,
746 async_extent->compressed_size,
747 async_extent->compressed_size,
748 0, alloc_hint, &ins, 1, 1);
750 free_async_extent_pages(async_extent);
752 if (ret == -ENOSPC) {
753 unlock_extent(io_tree, async_extent->start,
754 async_extent->start +
755 async_extent->ram_size - 1);
758 * we need to redirty the pages if we decide to
759 * fallback to uncompressed IO, otherwise we
760 * will not submit these pages down to lower
763 extent_range_redirty_for_io(inode,
765 async_extent->start +
766 async_extent->ram_size - 1);
773 * here we're doing allocation and writeback of the
776 btrfs_drop_extent_cache(inode, async_extent->start,
777 async_extent->start +
778 async_extent->ram_size - 1, 0);
780 em = alloc_extent_map();
783 goto out_free_reserve;
785 em->start = async_extent->start;
786 em->len = async_extent->ram_size;
787 em->orig_start = em->start;
788 em->mod_start = em->start;
789 em->mod_len = em->len;
791 em->block_start = ins.objectid;
792 em->block_len = ins.offset;
793 em->orig_block_len = ins.offset;
794 em->ram_bytes = async_extent->ram_size;
795 em->bdev = root->fs_info->fs_devices->latest_bdev;
796 em->compress_type = async_extent->compress_type;
797 set_bit(EXTENT_FLAG_PINNED, &em->flags);
798 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
802 write_lock(&em_tree->lock);
803 ret = add_extent_mapping(em_tree, em, 1);
804 write_unlock(&em_tree->lock);
805 if (ret != -EEXIST) {
809 btrfs_drop_extent_cache(inode, async_extent->start,
810 async_extent->start +
811 async_extent->ram_size - 1, 0);
815 goto out_free_reserve;
817 ret = btrfs_add_ordered_extent_compress(inode,
820 async_extent->ram_size,
822 BTRFS_ORDERED_COMPRESSED,
823 async_extent->compress_type);
825 btrfs_drop_extent_cache(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1, 0);
828 goto out_free_reserve;
830 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
833 * clear dirty, set writeback and unlock the pages.
835 extent_clear_unlock_delalloc(inode, async_extent->start,
836 async_extent->start +
837 async_extent->ram_size - 1,
838 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
839 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
841 ret = btrfs_submit_compressed_write(inode,
843 async_extent->ram_size,
845 ins.offset, async_extent->pages,
846 async_extent->nr_pages);
848 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
849 struct page *p = async_extent->pages[0];
850 const u64 start = async_extent->start;
851 const u64 end = start + async_extent->ram_size - 1;
853 p->mapping = inode->i_mapping;
854 tree->ops->writepage_end_io_hook(p, start, end,
857 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
860 free_async_extent_pages(async_extent);
862 alloc_hint = ins.objectid + ins.offset;
868 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
869 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
871 extent_clear_unlock_delalloc(inode, async_extent->start,
872 async_extent->start +
873 async_extent->ram_size - 1,
874 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
875 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
876 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
877 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
879 free_async_extent_pages(async_extent);
884 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
887 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
888 struct extent_map *em;
891 read_lock(&em_tree->lock);
892 em = search_extent_mapping(em_tree, start, num_bytes);
895 * if block start isn't an actual block number then find the
896 * first block in this inode and use that as a hint. If that
897 * block is also bogus then just don't worry about it.
899 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
901 em = search_extent_mapping(em_tree, 0, 0);
902 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
903 alloc_hint = em->block_start;
907 alloc_hint = em->block_start;
911 read_unlock(&em_tree->lock);
917 * when extent_io.c finds a delayed allocation range in the file,
918 * the call backs end up in this code. The basic idea is to
919 * allocate extents on disk for the range, and create ordered data structs
920 * in ram to track those extents.
922 * locked_page is the page that writepage had locked already. We use
923 * it to make sure we don't do extra locks or unlocks.
925 * *page_started is set to one if we unlock locked_page and do everything
926 * required to start IO on it. It may be clean and already done with
929 static noinline int cow_file_range(struct inode *inode,
930 struct page *locked_page,
931 u64 start, u64 end, u64 delalloc_end,
932 int *page_started, unsigned long *nr_written,
933 int unlock, struct btrfs_dedupe_hash *hash)
935 struct btrfs_root *root = BTRFS_I(inode)->root;
938 unsigned long ram_size;
941 u64 blocksize = root->sectorsize;
942 struct btrfs_key ins;
943 struct extent_map *em;
944 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
947 if (btrfs_is_free_space_inode(inode)) {
953 num_bytes = ALIGN(end - start + 1, blocksize);
954 num_bytes = max(blocksize, num_bytes);
955 disk_num_bytes = num_bytes;
957 /* if this is a small write inside eof, kick off defrag */
958 if (num_bytes < SZ_64K &&
959 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
960 btrfs_add_inode_defrag(NULL, inode);
963 /* lets try to make an inline extent */
964 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
967 extent_clear_unlock_delalloc(inode, start, end, NULL,
968 EXTENT_LOCKED | EXTENT_DELALLOC |
969 EXTENT_DEFRAG, PAGE_UNLOCK |
970 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
973 *nr_written = *nr_written +
974 (end - start + PAGE_SIZE) / PAGE_SIZE;
977 } else if (ret < 0) {
982 BUG_ON(disk_num_bytes >
983 btrfs_super_total_bytes(root->fs_info->super_copy));
985 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
986 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
988 while (disk_num_bytes > 0) {
991 cur_alloc_size = disk_num_bytes;
992 ret = btrfs_reserve_extent(root, cur_alloc_size,
993 root->sectorsize, 0, alloc_hint,
998 em = alloc_extent_map();
1004 em->orig_start = em->start;
1005 ram_size = ins.offset;
1006 em->len = ins.offset;
1007 em->mod_start = em->start;
1008 em->mod_len = em->len;
1010 em->block_start = ins.objectid;
1011 em->block_len = ins.offset;
1012 em->orig_block_len = ins.offset;
1013 em->ram_bytes = ram_size;
1014 em->bdev = root->fs_info->fs_devices->latest_bdev;
1015 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1016 em->generation = -1;
1019 write_lock(&em_tree->lock);
1020 ret = add_extent_mapping(em_tree, em, 1);
1021 write_unlock(&em_tree->lock);
1022 if (ret != -EEXIST) {
1023 free_extent_map(em);
1026 btrfs_drop_extent_cache(inode, start,
1027 start + ram_size - 1, 0);
1032 cur_alloc_size = ins.offset;
1033 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1034 ram_size, cur_alloc_size, 0);
1036 goto out_drop_extent_cache;
1038 if (root->root_key.objectid ==
1039 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1040 ret = btrfs_reloc_clone_csums(inode, start,
1043 goto out_drop_extent_cache;
1046 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1048 if (disk_num_bytes < cur_alloc_size)
1051 /* we're not doing compressed IO, don't unlock the first
1052 * page (which the caller expects to stay locked), don't
1053 * clear any dirty bits and don't set any writeback bits
1055 * Do set the Private2 bit so we know this page was properly
1056 * setup for writepage
1058 op = unlock ? PAGE_UNLOCK : 0;
1059 op |= PAGE_SET_PRIVATE2;
1061 extent_clear_unlock_delalloc(inode, start,
1062 start + ram_size - 1, locked_page,
1063 EXTENT_LOCKED | EXTENT_DELALLOC,
1065 disk_num_bytes -= cur_alloc_size;
1066 num_bytes -= cur_alloc_size;
1067 alloc_hint = ins.objectid + ins.offset;
1068 start += cur_alloc_size;
1073 out_drop_extent_cache:
1074 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1076 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1077 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1079 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1080 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1081 EXTENT_DELALLOC | EXTENT_DEFRAG,
1082 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1083 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1088 * work queue call back to started compression on a file and pages
1090 static noinline void async_cow_start(struct btrfs_work *work)
1092 struct async_cow *async_cow;
1094 async_cow = container_of(work, struct async_cow, work);
1096 compress_file_range(async_cow->inode, async_cow->locked_page,
1097 async_cow->start, async_cow->end, async_cow,
1099 if (num_added == 0) {
1100 btrfs_add_delayed_iput(async_cow->inode);
1101 async_cow->inode = NULL;
1106 * work queue call back to submit previously compressed pages
1108 static noinline void async_cow_submit(struct btrfs_work *work)
1110 struct async_cow *async_cow;
1111 struct btrfs_root *root;
1112 unsigned long nr_pages;
1114 async_cow = container_of(work, struct async_cow, work);
1116 root = async_cow->root;
1117 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1121 * atomic_sub_return implies a barrier for waitqueue_active
1123 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1125 waitqueue_active(&root->fs_info->async_submit_wait))
1126 wake_up(&root->fs_info->async_submit_wait);
1128 if (async_cow->inode)
1129 submit_compressed_extents(async_cow->inode, async_cow);
1132 static noinline void async_cow_free(struct btrfs_work *work)
1134 struct async_cow *async_cow;
1135 async_cow = container_of(work, struct async_cow, work);
1136 if (async_cow->inode)
1137 btrfs_add_delayed_iput(async_cow->inode);
1141 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1142 u64 start, u64 end, int *page_started,
1143 unsigned long *nr_written)
1145 struct async_cow *async_cow;
1146 struct btrfs_root *root = BTRFS_I(inode)->root;
1147 unsigned long nr_pages;
1149 int limit = 10 * SZ_1M;
1151 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1152 1, 0, NULL, GFP_NOFS);
1153 while (start < end) {
1154 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1155 BUG_ON(!async_cow); /* -ENOMEM */
1156 async_cow->inode = igrab(inode);
1157 async_cow->root = root;
1158 async_cow->locked_page = locked_page;
1159 async_cow->start = start;
1161 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1162 !btrfs_test_opt(root->fs_info, FORCE_COMPRESS))
1165 cur_end = min(end, start + SZ_512K - 1);
1167 async_cow->end = cur_end;
1168 INIT_LIST_HEAD(&async_cow->extents);
1170 btrfs_init_work(&async_cow->work,
1171 btrfs_delalloc_helper,
1172 async_cow_start, async_cow_submit,
1175 nr_pages = (cur_end - start + PAGE_SIZE) >>
1177 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1179 btrfs_queue_work(root->fs_info->delalloc_workers,
1182 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1183 wait_event(root->fs_info->async_submit_wait,
1184 (atomic_read(&root->fs_info->async_delalloc_pages) <
1188 while (atomic_read(&root->fs_info->async_submit_draining) &&
1189 atomic_read(&root->fs_info->async_delalloc_pages)) {
1190 wait_event(root->fs_info->async_submit_wait,
1191 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1195 *nr_written += nr_pages;
1196 start = cur_end + 1;
1202 static noinline int csum_exist_in_range(struct btrfs_root *root,
1203 u64 bytenr, u64 num_bytes)
1206 struct btrfs_ordered_sum *sums;
1209 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1210 bytenr + num_bytes - 1, &list, 0);
1211 if (ret == 0 && list_empty(&list))
1214 while (!list_empty(&list)) {
1215 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1216 list_del(&sums->list);
1223 * when nowcow writeback call back. This checks for snapshots or COW copies
1224 * of the extents that exist in the file, and COWs the file as required.
1226 * If no cow copies or snapshots exist, we write directly to the existing
1229 static noinline int run_delalloc_nocow(struct inode *inode,
1230 struct page *locked_page,
1231 u64 start, u64 end, int *page_started, int force,
1232 unsigned long *nr_written)
1234 struct btrfs_root *root = BTRFS_I(inode)->root;
1235 struct btrfs_trans_handle *trans;
1236 struct extent_buffer *leaf;
1237 struct btrfs_path *path;
1238 struct btrfs_file_extent_item *fi;
1239 struct btrfs_key found_key;
1254 u64 ino = btrfs_ino(inode);
1256 path = btrfs_alloc_path();
1258 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1259 EXTENT_LOCKED | EXTENT_DELALLOC |
1260 EXTENT_DO_ACCOUNTING |
1261 EXTENT_DEFRAG, PAGE_UNLOCK |
1263 PAGE_SET_WRITEBACK |
1264 PAGE_END_WRITEBACK);
1268 nolock = btrfs_is_free_space_inode(inode);
1271 trans = btrfs_join_transaction_nolock(root);
1273 trans = btrfs_join_transaction(root);
1275 if (IS_ERR(trans)) {
1276 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1277 EXTENT_LOCKED | EXTENT_DELALLOC |
1278 EXTENT_DO_ACCOUNTING |
1279 EXTENT_DEFRAG, PAGE_UNLOCK |
1281 PAGE_SET_WRITEBACK |
1282 PAGE_END_WRITEBACK);
1283 btrfs_free_path(path);
1284 return PTR_ERR(trans);
1287 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1289 cow_start = (u64)-1;
1292 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1296 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1297 leaf = path->nodes[0];
1298 btrfs_item_key_to_cpu(leaf, &found_key,
1299 path->slots[0] - 1);
1300 if (found_key.objectid == ino &&
1301 found_key.type == BTRFS_EXTENT_DATA_KEY)
1306 leaf = path->nodes[0];
1307 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1308 ret = btrfs_next_leaf(root, path);
1313 leaf = path->nodes[0];
1319 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1321 if (found_key.objectid > ino)
1323 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1324 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1328 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1329 found_key.offset > end)
1332 if (found_key.offset > cur_offset) {
1333 extent_end = found_key.offset;
1338 fi = btrfs_item_ptr(leaf, path->slots[0],
1339 struct btrfs_file_extent_item);
1340 extent_type = btrfs_file_extent_type(leaf, fi);
1342 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1343 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1344 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1345 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1346 extent_offset = btrfs_file_extent_offset(leaf, fi);
1347 extent_end = found_key.offset +
1348 btrfs_file_extent_num_bytes(leaf, fi);
1350 btrfs_file_extent_disk_num_bytes(leaf, fi);
1351 if (extent_end <= start) {
1355 if (disk_bytenr == 0)
1357 if (btrfs_file_extent_compression(leaf, fi) ||
1358 btrfs_file_extent_encryption(leaf, fi) ||
1359 btrfs_file_extent_other_encoding(leaf, fi))
1361 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1363 if (btrfs_extent_readonly(root, disk_bytenr))
1365 if (btrfs_cross_ref_exist(trans, root, ino,
1367 extent_offset, disk_bytenr))
1369 disk_bytenr += extent_offset;
1370 disk_bytenr += cur_offset - found_key.offset;
1371 num_bytes = min(end + 1, extent_end) - cur_offset;
1373 * if there are pending snapshots for this root,
1374 * we fall into common COW way.
1377 err = btrfs_start_write_no_snapshoting(root);
1382 * force cow if csum exists in the range.
1383 * this ensure that csum for a given extent are
1384 * either valid or do not exist.
1386 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1388 if (!btrfs_inc_nocow_writers(root->fs_info,
1392 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1393 extent_end = found_key.offset +
1394 btrfs_file_extent_inline_len(leaf,
1395 path->slots[0], fi);
1396 extent_end = ALIGN(extent_end, root->sectorsize);
1401 if (extent_end <= start) {
1403 if (!nolock && nocow)
1404 btrfs_end_write_no_snapshoting(root);
1406 btrfs_dec_nocow_writers(root->fs_info,
1411 if (cow_start == (u64)-1)
1412 cow_start = cur_offset;
1413 cur_offset = extent_end;
1414 if (cur_offset > end)
1420 btrfs_release_path(path);
1421 if (cow_start != (u64)-1) {
1422 ret = cow_file_range(inode, locked_page,
1423 cow_start, found_key.offset - 1,
1424 end, page_started, nr_written, 1,
1427 if (!nolock && nocow)
1428 btrfs_end_write_no_snapshoting(root);
1430 btrfs_dec_nocow_writers(root->fs_info,
1434 cow_start = (u64)-1;
1437 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1438 struct extent_map *em;
1439 struct extent_map_tree *em_tree;
1440 em_tree = &BTRFS_I(inode)->extent_tree;
1441 em = alloc_extent_map();
1442 BUG_ON(!em); /* -ENOMEM */
1443 em->start = cur_offset;
1444 em->orig_start = found_key.offset - extent_offset;
1445 em->len = num_bytes;
1446 em->block_len = num_bytes;
1447 em->block_start = disk_bytenr;
1448 em->orig_block_len = disk_num_bytes;
1449 em->ram_bytes = ram_bytes;
1450 em->bdev = root->fs_info->fs_devices->latest_bdev;
1451 em->mod_start = em->start;
1452 em->mod_len = em->len;
1453 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1454 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1455 em->generation = -1;
1457 write_lock(&em_tree->lock);
1458 ret = add_extent_mapping(em_tree, em, 1);
1459 write_unlock(&em_tree->lock);
1460 if (ret != -EEXIST) {
1461 free_extent_map(em);
1464 btrfs_drop_extent_cache(inode, em->start,
1465 em->start + em->len - 1, 0);
1467 type = BTRFS_ORDERED_PREALLOC;
1469 type = BTRFS_ORDERED_NOCOW;
1472 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1473 num_bytes, num_bytes, type);
1475 btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1476 BUG_ON(ret); /* -ENOMEM */
1478 if (root->root_key.objectid ==
1479 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1480 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1483 if (!nolock && nocow)
1484 btrfs_end_write_no_snapshoting(root);
1489 extent_clear_unlock_delalloc(inode, cur_offset,
1490 cur_offset + num_bytes - 1,
1491 locked_page, EXTENT_LOCKED |
1492 EXTENT_DELALLOC, PAGE_UNLOCK |
1494 if (!nolock && nocow)
1495 btrfs_end_write_no_snapshoting(root);
1496 cur_offset = extent_end;
1497 if (cur_offset > end)
1500 btrfs_release_path(path);
1502 if (cur_offset <= end && cow_start == (u64)-1) {
1503 cow_start = cur_offset;
1507 if (cow_start != (u64)-1) {
1508 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1509 page_started, nr_written, 1, NULL);
1515 err = btrfs_end_transaction(trans, root);
1519 if (ret && cur_offset < end)
1520 extent_clear_unlock_delalloc(inode, cur_offset, end,
1521 locked_page, EXTENT_LOCKED |
1522 EXTENT_DELALLOC | EXTENT_DEFRAG |
1523 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1525 PAGE_SET_WRITEBACK |
1526 PAGE_END_WRITEBACK);
1527 btrfs_free_path(path);
1531 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1534 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1535 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1539 * @defrag_bytes is a hint value, no spinlock held here,
1540 * if is not zero, it means the file is defragging.
1541 * Force cow if given extent needs to be defragged.
1543 if (BTRFS_I(inode)->defrag_bytes &&
1544 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1545 EXTENT_DEFRAG, 0, NULL))
1552 * extent_io.c call back to do delayed allocation processing
1554 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1555 u64 start, u64 end, int *page_started,
1556 unsigned long *nr_written)
1559 int force_cow = need_force_cow(inode, start, end);
1561 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1562 ret = run_delalloc_nocow(inode, locked_page, start, end,
1563 page_started, 1, nr_written);
1564 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1565 ret = run_delalloc_nocow(inode, locked_page, start, end,
1566 page_started, 0, nr_written);
1567 } else if (!inode_need_compress(inode)) {
1568 ret = cow_file_range(inode, locked_page, start, end, end,
1569 page_started, nr_written, 1, NULL);
1571 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1572 &BTRFS_I(inode)->runtime_flags);
1573 ret = cow_file_range_async(inode, locked_page, start, end,
1574 page_started, nr_written);
1579 static void btrfs_split_extent_hook(struct inode *inode,
1580 struct extent_state *orig, u64 split)
1584 /* not delalloc, ignore it */
1585 if (!(orig->state & EXTENT_DELALLOC))
1588 size = orig->end - orig->start + 1;
1589 if (size > BTRFS_MAX_EXTENT_SIZE) {
1594 * See the explanation in btrfs_merge_extent_hook, the same
1595 * applies here, just in reverse.
1597 new_size = orig->end - split + 1;
1598 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1599 BTRFS_MAX_EXTENT_SIZE);
1600 new_size = split - orig->start;
1601 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1602 BTRFS_MAX_EXTENT_SIZE);
1603 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1604 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1608 spin_lock(&BTRFS_I(inode)->lock);
1609 BTRFS_I(inode)->outstanding_extents++;
1610 spin_unlock(&BTRFS_I(inode)->lock);
1614 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1615 * extents so we can keep track of new extents that are just merged onto old
1616 * extents, such as when we are doing sequential writes, so we can properly
1617 * account for the metadata space we'll need.
1619 static void btrfs_merge_extent_hook(struct inode *inode,
1620 struct extent_state *new,
1621 struct extent_state *other)
1623 u64 new_size, old_size;
1626 /* not delalloc, ignore it */
1627 if (!(other->state & EXTENT_DELALLOC))
1630 if (new->start > other->start)
1631 new_size = new->end - other->start + 1;
1633 new_size = other->end - new->start + 1;
1635 /* we're not bigger than the max, unreserve the space and go */
1636 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1637 spin_lock(&BTRFS_I(inode)->lock);
1638 BTRFS_I(inode)->outstanding_extents--;
1639 spin_unlock(&BTRFS_I(inode)->lock);
1644 * We have to add up either side to figure out how many extents were
1645 * accounted for before we merged into one big extent. If the number of
1646 * extents we accounted for is <= the amount we need for the new range
1647 * then we can return, otherwise drop. Think of it like this
1651 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1652 * need 2 outstanding extents, on one side we have 1 and the other side
1653 * we have 1 so they are == and we can return. But in this case
1655 * [MAX_SIZE+4k][MAX_SIZE+4k]
1657 * Each range on their own accounts for 2 extents, but merged together
1658 * they are only 3 extents worth of accounting, so we need to drop in
1661 old_size = other->end - other->start + 1;
1662 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1663 BTRFS_MAX_EXTENT_SIZE);
1664 old_size = new->end - new->start + 1;
1665 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1666 BTRFS_MAX_EXTENT_SIZE);
1668 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1669 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1672 spin_lock(&BTRFS_I(inode)->lock);
1673 BTRFS_I(inode)->outstanding_extents--;
1674 spin_unlock(&BTRFS_I(inode)->lock);
1677 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1678 struct inode *inode)
1680 spin_lock(&root->delalloc_lock);
1681 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1682 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1683 &root->delalloc_inodes);
1684 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes++;
1687 if (root->nr_delalloc_inodes == 1) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(!list_empty(&root->delalloc_root));
1690 list_add_tail(&root->delalloc_root,
1691 &root->fs_info->delalloc_roots);
1692 spin_unlock(&root->fs_info->delalloc_root_lock);
1695 spin_unlock(&root->delalloc_lock);
1698 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1699 struct inode *inode)
1701 spin_lock(&root->delalloc_lock);
1702 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1703 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1704 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1705 &BTRFS_I(inode)->runtime_flags);
1706 root->nr_delalloc_inodes--;
1707 if (!root->nr_delalloc_inodes) {
1708 spin_lock(&root->fs_info->delalloc_root_lock);
1709 BUG_ON(list_empty(&root->delalloc_root));
1710 list_del_init(&root->delalloc_root);
1711 spin_unlock(&root->fs_info->delalloc_root_lock);
1714 spin_unlock(&root->delalloc_lock);
1718 * extent_io.c set_bit_hook, used to track delayed allocation
1719 * bytes in this file, and to maintain the list of inodes that
1720 * have pending delalloc work to be done.
1722 static void btrfs_set_bit_hook(struct inode *inode,
1723 struct extent_state *state, unsigned *bits)
1726 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1729 * set_bit and clear bit hooks normally require _irqsave/restore
1730 * but in this case, we are only testing for the DELALLOC
1731 * bit, which is only set or cleared with irqs on
1733 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1734 struct btrfs_root *root = BTRFS_I(inode)->root;
1735 u64 len = state->end + 1 - state->start;
1736 bool do_list = !btrfs_is_free_space_inode(inode);
1738 if (*bits & EXTENT_FIRST_DELALLOC) {
1739 *bits &= ~EXTENT_FIRST_DELALLOC;
1741 spin_lock(&BTRFS_I(inode)->lock);
1742 BTRFS_I(inode)->outstanding_extents++;
1743 spin_unlock(&BTRFS_I(inode)->lock);
1746 /* For sanity tests */
1747 if (btrfs_test_is_dummy_root(root))
1750 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1751 root->fs_info->delalloc_batch);
1752 spin_lock(&BTRFS_I(inode)->lock);
1753 BTRFS_I(inode)->delalloc_bytes += len;
1754 if (*bits & EXTENT_DEFRAG)
1755 BTRFS_I(inode)->defrag_bytes += len;
1756 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1757 &BTRFS_I(inode)->runtime_flags))
1758 btrfs_add_delalloc_inodes(root, inode);
1759 spin_unlock(&BTRFS_I(inode)->lock);
1764 * extent_io.c clear_bit_hook, see set_bit_hook for why
1766 static void btrfs_clear_bit_hook(struct inode *inode,
1767 struct extent_state *state,
1770 u64 len = state->end + 1 - state->start;
1771 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1772 BTRFS_MAX_EXTENT_SIZE);
1774 spin_lock(&BTRFS_I(inode)->lock);
1775 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1776 BTRFS_I(inode)->defrag_bytes -= len;
1777 spin_unlock(&BTRFS_I(inode)->lock);
1780 * set_bit and clear bit hooks normally require _irqsave/restore
1781 * but in this case, we are only testing for the DELALLOC
1782 * bit, which is only set or cleared with irqs on
1784 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1785 struct btrfs_root *root = BTRFS_I(inode)->root;
1786 bool do_list = !btrfs_is_free_space_inode(inode);
1788 if (*bits & EXTENT_FIRST_DELALLOC) {
1789 *bits &= ~EXTENT_FIRST_DELALLOC;
1790 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1791 spin_lock(&BTRFS_I(inode)->lock);
1792 BTRFS_I(inode)->outstanding_extents -= num_extents;
1793 spin_unlock(&BTRFS_I(inode)->lock);
1797 * We don't reserve metadata space for space cache inodes so we
1798 * don't need to call dellalloc_release_metadata if there is an
1801 if (*bits & EXTENT_DO_ACCOUNTING &&
1802 root != root->fs_info->tree_root)
1803 btrfs_delalloc_release_metadata(inode, len);
1805 /* For sanity tests. */
1806 if (btrfs_test_is_dummy_root(root))
1809 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1810 && do_list && !(state->state & EXTENT_NORESERVE))
1811 btrfs_free_reserved_data_space_noquota(inode,
1814 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1815 root->fs_info->delalloc_batch);
1816 spin_lock(&BTRFS_I(inode)->lock);
1817 BTRFS_I(inode)->delalloc_bytes -= len;
1818 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1819 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1820 &BTRFS_I(inode)->runtime_flags))
1821 btrfs_del_delalloc_inode(root, inode);
1822 spin_unlock(&BTRFS_I(inode)->lock);
1827 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1828 * we don't create bios that span stripes or chunks
1830 * return 1 if page cannot be merged to bio
1831 * return 0 if page can be merged to bio
1832 * return error otherwise
1834 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1835 size_t size, struct bio *bio,
1836 unsigned long bio_flags)
1838 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1839 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1844 if (bio_flags & EXTENT_BIO_COMPRESSED)
1847 length = bio->bi_iter.bi_size;
1848 map_length = length;
1849 ret = btrfs_map_block(root->fs_info, rw, logical,
1850 &map_length, NULL, 0);
1853 if (map_length < length + size)
1859 * in order to insert checksums into the metadata in large chunks,
1860 * we wait until bio submission time. All the pages in the bio are
1861 * checksummed and sums are attached onto the ordered extent record.
1863 * At IO completion time the cums attached on the ordered extent record
1864 * are inserted into the btree
1866 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1867 struct bio *bio, int mirror_num,
1868 unsigned long bio_flags,
1871 struct btrfs_root *root = BTRFS_I(inode)->root;
1874 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1875 BUG_ON(ret); /* -ENOMEM */
1880 * in order to insert checksums into the metadata in large chunks,
1881 * we wait until bio submission time. All the pages in the bio are
1882 * checksummed and sums are attached onto the ordered extent record.
1884 * At IO completion time the cums attached on the ordered extent record
1885 * are inserted into the btree
1887 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1888 int mirror_num, unsigned long bio_flags,
1891 struct btrfs_root *root = BTRFS_I(inode)->root;
1894 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1896 bio->bi_error = ret;
1903 * extent_io.c submission hook. This does the right thing for csum calculation
1904 * on write, or reading the csums from the tree before a read
1906 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1907 int mirror_num, unsigned long bio_flags,
1910 struct btrfs_root *root = BTRFS_I(inode)->root;
1911 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1914 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1916 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1918 if (btrfs_is_free_space_inode(inode))
1919 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1921 if (!(rw & REQ_WRITE)) {
1922 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1926 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1927 ret = btrfs_submit_compressed_read(inode, bio,
1931 } else if (!skip_sum) {
1932 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1937 } else if (async && !skip_sum) {
1938 /* csum items have already been cloned */
1939 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1941 /* we're doing a write, do the async checksumming */
1942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1943 inode, rw, bio, mirror_num,
1944 bio_flags, bio_offset,
1945 __btrfs_submit_bio_start,
1946 __btrfs_submit_bio_done);
1948 } else if (!skip_sum) {
1949 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1955 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1959 bio->bi_error = ret;
1966 * given a list of ordered sums record them in the inode. This happens
1967 * at IO completion time based on sums calculated at bio submission time.
1969 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1970 struct inode *inode, u64 file_offset,
1971 struct list_head *list)
1973 struct btrfs_ordered_sum *sum;
1975 list_for_each_entry(sum, list, list) {
1976 trans->adding_csums = 1;
1977 btrfs_csum_file_blocks(trans,
1978 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1979 trans->adding_csums = 0;
1984 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1985 struct extent_state **cached_state)
1987 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1988 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1992 /* see btrfs_writepage_start_hook for details on why this is required */
1993 struct btrfs_writepage_fixup {
1995 struct btrfs_work work;
1998 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2000 struct btrfs_writepage_fixup *fixup;
2001 struct btrfs_ordered_extent *ordered;
2002 struct extent_state *cached_state = NULL;
2004 struct inode *inode;
2009 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2013 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2014 ClearPageChecked(page);
2018 inode = page->mapping->host;
2019 page_start = page_offset(page);
2020 page_end = page_offset(page) + PAGE_SIZE - 1;
2022 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2025 /* already ordered? We're done */
2026 if (PagePrivate2(page))
2029 ordered = btrfs_lookup_ordered_range(inode, page_start,
2032 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2033 page_end, &cached_state, GFP_NOFS);
2035 btrfs_start_ordered_extent(inode, ordered, 1);
2036 btrfs_put_ordered_extent(ordered);
2040 ret = btrfs_delalloc_reserve_space(inode, page_start,
2043 mapping_set_error(page->mapping, ret);
2044 end_extent_writepage(page, ret, page_start, page_end);
2045 ClearPageChecked(page);
2049 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2050 ClearPageChecked(page);
2051 set_page_dirty(page);
2053 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2054 &cached_state, GFP_NOFS);
2062 * There are a few paths in the higher layers of the kernel that directly
2063 * set the page dirty bit without asking the filesystem if it is a
2064 * good idea. This causes problems because we want to make sure COW
2065 * properly happens and the data=ordered rules are followed.
2067 * In our case any range that doesn't have the ORDERED bit set
2068 * hasn't been properly setup for IO. We kick off an async process
2069 * to fix it up. The async helper will wait for ordered extents, set
2070 * the delalloc bit and make it safe to write the page.
2072 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2074 struct inode *inode = page->mapping->host;
2075 struct btrfs_writepage_fixup *fixup;
2076 struct btrfs_root *root = BTRFS_I(inode)->root;
2078 /* this page is properly in the ordered list */
2079 if (TestClearPagePrivate2(page))
2082 if (PageChecked(page))
2085 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2089 SetPageChecked(page);
2091 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2092 btrfs_writepage_fixup_worker, NULL, NULL);
2094 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2098 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2099 struct inode *inode, u64 file_pos,
2100 u64 disk_bytenr, u64 disk_num_bytes,
2101 u64 num_bytes, u64 ram_bytes,
2102 u8 compression, u8 encryption,
2103 u16 other_encoding, int extent_type)
2105 struct btrfs_root *root = BTRFS_I(inode)->root;
2106 struct btrfs_file_extent_item *fi;
2107 struct btrfs_path *path;
2108 struct extent_buffer *leaf;
2109 struct btrfs_key ins;
2110 int extent_inserted = 0;
2113 path = btrfs_alloc_path();
2118 * we may be replacing one extent in the tree with another.
2119 * The new extent is pinned in the extent map, and we don't want
2120 * to drop it from the cache until it is completely in the btree.
2122 * So, tell btrfs_drop_extents to leave this extent in the cache.
2123 * the caller is expected to unpin it and allow it to be merged
2126 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2127 file_pos + num_bytes, NULL, 0,
2128 1, sizeof(*fi), &extent_inserted);
2132 if (!extent_inserted) {
2133 ins.objectid = btrfs_ino(inode);
2134 ins.offset = file_pos;
2135 ins.type = BTRFS_EXTENT_DATA_KEY;
2137 path->leave_spinning = 1;
2138 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2143 leaf = path->nodes[0];
2144 fi = btrfs_item_ptr(leaf, path->slots[0],
2145 struct btrfs_file_extent_item);
2146 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2147 btrfs_set_file_extent_type(leaf, fi, extent_type);
2148 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2149 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2150 btrfs_set_file_extent_offset(leaf, fi, 0);
2151 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2152 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2153 btrfs_set_file_extent_compression(leaf, fi, compression);
2154 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2155 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2160 inode_add_bytes(inode, num_bytes);
2162 ins.objectid = disk_bytenr;
2163 ins.offset = disk_num_bytes;
2164 ins.type = BTRFS_EXTENT_ITEM_KEY;
2165 ret = btrfs_alloc_reserved_file_extent(trans, root,
2166 root->root_key.objectid,
2167 btrfs_ino(inode), file_pos,
2170 * Release the reserved range from inode dirty range map, as it is
2171 * already moved into delayed_ref_head
2173 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2175 btrfs_free_path(path);
2180 /* snapshot-aware defrag */
2181 struct sa_defrag_extent_backref {
2182 struct rb_node node;
2183 struct old_sa_defrag_extent *old;
2192 struct old_sa_defrag_extent {
2193 struct list_head list;
2194 struct new_sa_defrag_extent *new;
2203 struct new_sa_defrag_extent {
2204 struct rb_root root;
2205 struct list_head head;
2206 struct btrfs_path *path;
2207 struct inode *inode;
2215 static int backref_comp(struct sa_defrag_extent_backref *b1,
2216 struct sa_defrag_extent_backref *b2)
2218 if (b1->root_id < b2->root_id)
2220 else if (b1->root_id > b2->root_id)
2223 if (b1->inum < b2->inum)
2225 else if (b1->inum > b2->inum)
2228 if (b1->file_pos < b2->file_pos)
2230 else if (b1->file_pos > b2->file_pos)
2234 * [------------------------------] ===> (a range of space)
2235 * |<--->| |<---->| =============> (fs/file tree A)
2236 * |<---------------------------->| ===> (fs/file tree B)
2238 * A range of space can refer to two file extents in one tree while
2239 * refer to only one file extent in another tree.
2241 * So we may process a disk offset more than one time(two extents in A)
2242 * and locate at the same extent(one extent in B), then insert two same
2243 * backrefs(both refer to the extent in B).
2248 static void backref_insert(struct rb_root *root,
2249 struct sa_defrag_extent_backref *backref)
2251 struct rb_node **p = &root->rb_node;
2252 struct rb_node *parent = NULL;
2253 struct sa_defrag_extent_backref *entry;
2258 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2260 ret = backref_comp(backref, entry);
2264 p = &(*p)->rb_right;
2267 rb_link_node(&backref->node, parent, p);
2268 rb_insert_color(&backref->node, root);
2272 * Note the backref might has changed, and in this case we just return 0.
2274 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2277 struct btrfs_file_extent_item *extent;
2278 struct btrfs_fs_info *fs_info;
2279 struct old_sa_defrag_extent *old = ctx;
2280 struct new_sa_defrag_extent *new = old->new;
2281 struct btrfs_path *path = new->path;
2282 struct btrfs_key key;
2283 struct btrfs_root *root;
2284 struct sa_defrag_extent_backref *backref;
2285 struct extent_buffer *leaf;
2286 struct inode *inode = new->inode;
2292 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2293 inum == btrfs_ino(inode))
2296 key.objectid = root_id;
2297 key.type = BTRFS_ROOT_ITEM_KEY;
2298 key.offset = (u64)-1;
2300 fs_info = BTRFS_I(inode)->root->fs_info;
2301 root = btrfs_read_fs_root_no_name(fs_info, &key);
2303 if (PTR_ERR(root) == -ENOENT)
2306 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2307 inum, offset, root_id);
2308 return PTR_ERR(root);
2311 key.objectid = inum;
2312 key.type = BTRFS_EXTENT_DATA_KEY;
2313 if (offset > (u64)-1 << 32)
2316 key.offset = offset;
2318 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2319 if (WARN_ON(ret < 0))
2326 leaf = path->nodes[0];
2327 slot = path->slots[0];
2329 if (slot >= btrfs_header_nritems(leaf)) {
2330 ret = btrfs_next_leaf(root, path);
2333 } else if (ret > 0) {
2342 btrfs_item_key_to_cpu(leaf, &key, slot);
2344 if (key.objectid > inum)
2347 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2350 extent = btrfs_item_ptr(leaf, slot,
2351 struct btrfs_file_extent_item);
2353 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2357 * 'offset' refers to the exact key.offset,
2358 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2359 * (key.offset - extent_offset).
2361 if (key.offset != offset)
2364 extent_offset = btrfs_file_extent_offset(leaf, extent);
2365 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2367 if (extent_offset >= old->extent_offset + old->offset +
2368 old->len || extent_offset + num_bytes <=
2369 old->extent_offset + old->offset)
2374 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2380 backref->root_id = root_id;
2381 backref->inum = inum;
2382 backref->file_pos = offset;
2383 backref->num_bytes = num_bytes;
2384 backref->extent_offset = extent_offset;
2385 backref->generation = btrfs_file_extent_generation(leaf, extent);
2387 backref_insert(&new->root, backref);
2390 btrfs_release_path(path);
2395 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2396 struct new_sa_defrag_extent *new)
2398 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2399 struct old_sa_defrag_extent *old, *tmp;
2404 list_for_each_entry_safe(old, tmp, &new->head, list) {
2405 ret = iterate_inodes_from_logical(old->bytenr +
2406 old->extent_offset, fs_info,
2407 path, record_one_backref,
2409 if (ret < 0 && ret != -ENOENT)
2412 /* no backref to be processed for this extent */
2414 list_del(&old->list);
2419 if (list_empty(&new->head))
2425 static int relink_is_mergable(struct extent_buffer *leaf,
2426 struct btrfs_file_extent_item *fi,
2427 struct new_sa_defrag_extent *new)
2429 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2432 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2435 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2438 if (btrfs_file_extent_encryption(leaf, fi) ||
2439 btrfs_file_extent_other_encoding(leaf, fi))
2446 * Note the backref might has changed, and in this case we just return 0.
2448 static noinline int relink_extent_backref(struct btrfs_path *path,
2449 struct sa_defrag_extent_backref *prev,
2450 struct sa_defrag_extent_backref *backref)
2452 struct btrfs_file_extent_item *extent;
2453 struct btrfs_file_extent_item *item;
2454 struct btrfs_ordered_extent *ordered;
2455 struct btrfs_trans_handle *trans;
2456 struct btrfs_fs_info *fs_info;
2457 struct btrfs_root *root;
2458 struct btrfs_key key;
2459 struct extent_buffer *leaf;
2460 struct old_sa_defrag_extent *old = backref->old;
2461 struct new_sa_defrag_extent *new = old->new;
2462 struct inode *src_inode = new->inode;
2463 struct inode *inode;
2464 struct extent_state *cached = NULL;
2473 if (prev && prev->root_id == backref->root_id &&
2474 prev->inum == backref->inum &&
2475 prev->file_pos + prev->num_bytes == backref->file_pos)
2478 /* step 1: get root */
2479 key.objectid = backref->root_id;
2480 key.type = BTRFS_ROOT_ITEM_KEY;
2481 key.offset = (u64)-1;
2483 fs_info = BTRFS_I(src_inode)->root->fs_info;
2484 index = srcu_read_lock(&fs_info->subvol_srcu);
2486 root = btrfs_read_fs_root_no_name(fs_info, &key);
2488 srcu_read_unlock(&fs_info->subvol_srcu, index);
2489 if (PTR_ERR(root) == -ENOENT)
2491 return PTR_ERR(root);
2494 if (btrfs_root_readonly(root)) {
2495 srcu_read_unlock(&fs_info->subvol_srcu, index);
2499 /* step 2: get inode */
2500 key.objectid = backref->inum;
2501 key.type = BTRFS_INODE_ITEM_KEY;
2504 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2505 if (IS_ERR(inode)) {
2506 srcu_read_unlock(&fs_info->subvol_srcu, index);
2510 srcu_read_unlock(&fs_info->subvol_srcu, index);
2512 /* step 3: relink backref */
2513 lock_start = backref->file_pos;
2514 lock_end = backref->file_pos + backref->num_bytes - 1;
2515 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2518 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2520 btrfs_put_ordered_extent(ordered);
2524 trans = btrfs_join_transaction(root);
2525 if (IS_ERR(trans)) {
2526 ret = PTR_ERR(trans);
2530 key.objectid = backref->inum;
2531 key.type = BTRFS_EXTENT_DATA_KEY;
2532 key.offset = backref->file_pos;
2534 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2537 } else if (ret > 0) {
2542 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2543 struct btrfs_file_extent_item);
2545 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2546 backref->generation)
2549 btrfs_release_path(path);
2551 start = backref->file_pos;
2552 if (backref->extent_offset < old->extent_offset + old->offset)
2553 start += old->extent_offset + old->offset -
2554 backref->extent_offset;
2556 len = min(backref->extent_offset + backref->num_bytes,
2557 old->extent_offset + old->offset + old->len);
2558 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2560 ret = btrfs_drop_extents(trans, root, inode, start,
2565 key.objectid = btrfs_ino(inode);
2566 key.type = BTRFS_EXTENT_DATA_KEY;
2569 path->leave_spinning = 1;
2571 struct btrfs_file_extent_item *fi;
2573 struct btrfs_key found_key;
2575 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2580 leaf = path->nodes[0];
2581 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2583 fi = btrfs_item_ptr(leaf, path->slots[0],
2584 struct btrfs_file_extent_item);
2585 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2587 if (extent_len + found_key.offset == start &&
2588 relink_is_mergable(leaf, fi, new)) {
2589 btrfs_set_file_extent_num_bytes(leaf, fi,
2591 btrfs_mark_buffer_dirty(leaf);
2592 inode_add_bytes(inode, len);
2598 btrfs_release_path(path);
2603 ret = btrfs_insert_empty_item(trans, root, path, &key,
2606 btrfs_abort_transaction(trans, root, ret);
2610 leaf = path->nodes[0];
2611 item = btrfs_item_ptr(leaf, path->slots[0],
2612 struct btrfs_file_extent_item);
2613 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2614 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2615 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2616 btrfs_set_file_extent_num_bytes(leaf, item, len);
2617 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2618 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2619 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2620 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2621 btrfs_set_file_extent_encryption(leaf, item, 0);
2622 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2624 btrfs_mark_buffer_dirty(leaf);
2625 inode_add_bytes(inode, len);
2626 btrfs_release_path(path);
2628 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2630 backref->root_id, backref->inum,
2631 new->file_pos); /* start - extent_offset */
2633 btrfs_abort_transaction(trans, root, ret);
2639 btrfs_release_path(path);
2640 path->leave_spinning = 0;
2641 btrfs_end_transaction(trans, root);
2643 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2649 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2651 struct old_sa_defrag_extent *old, *tmp;
2656 list_for_each_entry_safe(old, tmp, &new->head, list) {
2662 static void relink_file_extents(struct new_sa_defrag_extent *new)
2664 struct btrfs_path *path;
2665 struct sa_defrag_extent_backref *backref;
2666 struct sa_defrag_extent_backref *prev = NULL;
2667 struct inode *inode;
2668 struct btrfs_root *root;
2669 struct rb_node *node;
2673 root = BTRFS_I(inode)->root;
2675 path = btrfs_alloc_path();
2679 if (!record_extent_backrefs(path, new)) {
2680 btrfs_free_path(path);
2683 btrfs_release_path(path);
2686 node = rb_first(&new->root);
2689 rb_erase(node, &new->root);
2691 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2693 ret = relink_extent_backref(path, prev, backref);
2706 btrfs_free_path(path);
2708 free_sa_defrag_extent(new);
2710 atomic_dec(&root->fs_info->defrag_running);
2711 wake_up(&root->fs_info->transaction_wait);
2714 static struct new_sa_defrag_extent *
2715 record_old_file_extents(struct inode *inode,
2716 struct btrfs_ordered_extent *ordered)
2718 struct btrfs_root *root = BTRFS_I(inode)->root;
2719 struct btrfs_path *path;
2720 struct btrfs_key key;
2721 struct old_sa_defrag_extent *old;
2722 struct new_sa_defrag_extent *new;
2725 new = kmalloc(sizeof(*new), GFP_NOFS);
2730 new->file_pos = ordered->file_offset;
2731 new->len = ordered->len;
2732 new->bytenr = ordered->start;
2733 new->disk_len = ordered->disk_len;
2734 new->compress_type = ordered->compress_type;
2735 new->root = RB_ROOT;
2736 INIT_LIST_HEAD(&new->head);
2738 path = btrfs_alloc_path();
2742 key.objectid = btrfs_ino(inode);
2743 key.type = BTRFS_EXTENT_DATA_KEY;
2744 key.offset = new->file_pos;
2746 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2749 if (ret > 0 && path->slots[0] > 0)
2752 /* find out all the old extents for the file range */
2754 struct btrfs_file_extent_item *extent;
2755 struct extent_buffer *l;
2764 slot = path->slots[0];
2766 if (slot >= btrfs_header_nritems(l)) {
2767 ret = btrfs_next_leaf(root, path);
2775 btrfs_item_key_to_cpu(l, &key, slot);
2777 if (key.objectid != btrfs_ino(inode))
2779 if (key.type != BTRFS_EXTENT_DATA_KEY)
2781 if (key.offset >= new->file_pos + new->len)
2784 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2786 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2787 if (key.offset + num_bytes < new->file_pos)
2790 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2794 extent_offset = btrfs_file_extent_offset(l, extent);
2796 old = kmalloc(sizeof(*old), GFP_NOFS);
2800 offset = max(new->file_pos, key.offset);
2801 end = min(new->file_pos + new->len, key.offset + num_bytes);
2803 old->bytenr = disk_bytenr;
2804 old->extent_offset = extent_offset;
2805 old->offset = offset - key.offset;
2806 old->len = end - offset;
2809 list_add_tail(&old->list, &new->head);
2815 btrfs_free_path(path);
2816 atomic_inc(&root->fs_info->defrag_running);
2821 btrfs_free_path(path);
2823 free_sa_defrag_extent(new);
2827 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2830 struct btrfs_block_group_cache *cache;
2832 cache = btrfs_lookup_block_group(root->fs_info, start);
2835 spin_lock(&cache->lock);
2836 cache->delalloc_bytes -= len;
2837 spin_unlock(&cache->lock);
2839 btrfs_put_block_group(cache);
2842 /* as ordered data IO finishes, this gets called so we can finish
2843 * an ordered extent if the range of bytes in the file it covers are
2846 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2848 struct inode *inode = ordered_extent->inode;
2849 struct btrfs_root *root = BTRFS_I(inode)->root;
2850 struct btrfs_trans_handle *trans = NULL;
2851 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2852 struct extent_state *cached_state = NULL;
2853 struct new_sa_defrag_extent *new = NULL;
2854 int compress_type = 0;
2856 u64 logical_len = ordered_extent->len;
2858 bool truncated = false;
2860 nolock = btrfs_is_free_space_inode(inode);
2862 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2867 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2868 ordered_extent->file_offset +
2869 ordered_extent->len - 1);
2871 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2873 logical_len = ordered_extent->truncated_len;
2874 /* Truncated the entire extent, don't bother adding */
2879 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2880 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2883 * For mwrite(mmap + memset to write) case, we still reserve
2884 * space for NOCOW range.
2885 * As NOCOW won't cause a new delayed ref, just free the space
2887 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2888 ordered_extent->len);
2889 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2891 trans = btrfs_join_transaction_nolock(root);
2893 trans = btrfs_join_transaction(root);
2894 if (IS_ERR(trans)) {
2895 ret = PTR_ERR(trans);
2899 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2900 ret = btrfs_update_inode_fallback(trans, root, inode);
2901 if (ret) /* -ENOMEM or corruption */
2902 btrfs_abort_transaction(trans, root, ret);
2906 lock_extent_bits(io_tree, ordered_extent->file_offset,
2907 ordered_extent->file_offset + ordered_extent->len - 1,
2910 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2911 ordered_extent->file_offset + ordered_extent->len - 1,
2912 EXTENT_DEFRAG, 1, cached_state);
2914 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2915 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2916 /* the inode is shared */
2917 new = record_old_file_extents(inode, ordered_extent);
2919 clear_extent_bit(io_tree, ordered_extent->file_offset,
2920 ordered_extent->file_offset + ordered_extent->len - 1,
2921 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2925 trans = btrfs_join_transaction_nolock(root);
2927 trans = btrfs_join_transaction(root);
2928 if (IS_ERR(trans)) {
2929 ret = PTR_ERR(trans);
2934 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2936 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2937 compress_type = ordered_extent->compress_type;
2938 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2939 BUG_ON(compress_type);
2940 ret = btrfs_mark_extent_written(trans, inode,
2941 ordered_extent->file_offset,
2942 ordered_extent->file_offset +
2945 BUG_ON(root == root->fs_info->tree_root);
2946 ret = insert_reserved_file_extent(trans, inode,
2947 ordered_extent->file_offset,
2948 ordered_extent->start,
2949 ordered_extent->disk_len,
2950 logical_len, logical_len,
2951 compress_type, 0, 0,
2952 BTRFS_FILE_EXTENT_REG);
2954 btrfs_release_delalloc_bytes(root,
2955 ordered_extent->start,
2956 ordered_extent->disk_len);
2958 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2959 ordered_extent->file_offset, ordered_extent->len,
2962 btrfs_abort_transaction(trans, root, ret);
2966 add_pending_csums(trans, inode, ordered_extent->file_offset,
2967 &ordered_extent->list);
2969 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2970 ret = btrfs_update_inode_fallback(trans, root, inode);
2971 if (ret) { /* -ENOMEM or corruption */
2972 btrfs_abort_transaction(trans, root, ret);
2977 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2978 ordered_extent->file_offset +
2979 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2981 if (root != root->fs_info->tree_root)
2982 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2984 btrfs_end_transaction(trans, root);
2986 if (ret || truncated) {
2990 start = ordered_extent->file_offset + logical_len;
2992 start = ordered_extent->file_offset;
2993 end = ordered_extent->file_offset + ordered_extent->len - 1;
2994 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2996 /* Drop the cache for the part of the extent we didn't write. */
2997 btrfs_drop_extent_cache(inode, start, end, 0);
3000 * If the ordered extent had an IOERR or something else went
3001 * wrong we need to return the space for this ordered extent
3002 * back to the allocator. We only free the extent in the
3003 * truncated case if we didn't write out the extent at all.
3005 if ((ret || !logical_len) &&
3006 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3007 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3008 btrfs_free_reserved_extent(root, ordered_extent->start,
3009 ordered_extent->disk_len, 1);
3014 * This needs to be done to make sure anybody waiting knows we are done
3015 * updating everything for this ordered extent.
3017 btrfs_remove_ordered_extent(inode, ordered_extent);
3019 /* for snapshot-aware defrag */
3022 free_sa_defrag_extent(new);
3023 atomic_dec(&root->fs_info->defrag_running);
3025 relink_file_extents(new);
3030 btrfs_put_ordered_extent(ordered_extent);
3031 /* once for the tree */
3032 btrfs_put_ordered_extent(ordered_extent);
3037 static void finish_ordered_fn(struct btrfs_work *work)
3039 struct btrfs_ordered_extent *ordered_extent;
3040 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3041 btrfs_finish_ordered_io(ordered_extent);
3044 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3045 struct extent_state *state, int uptodate)
3047 struct inode *inode = page->mapping->host;
3048 struct btrfs_root *root = BTRFS_I(inode)->root;
3049 struct btrfs_ordered_extent *ordered_extent = NULL;
3050 struct btrfs_workqueue *wq;
3051 btrfs_work_func_t func;
3053 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3055 ClearPagePrivate2(page);
3056 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3057 end - start + 1, uptodate))
3060 if (btrfs_is_free_space_inode(inode)) {
3061 wq = root->fs_info->endio_freespace_worker;
3062 func = btrfs_freespace_write_helper;
3064 wq = root->fs_info->endio_write_workers;
3065 func = btrfs_endio_write_helper;
3068 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3070 btrfs_queue_work(wq, &ordered_extent->work);
3075 static int __readpage_endio_check(struct inode *inode,
3076 struct btrfs_io_bio *io_bio,
3077 int icsum, struct page *page,
3078 int pgoff, u64 start, size_t len)
3084 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3086 kaddr = kmap_atomic(page);
3087 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3088 btrfs_csum_final(csum, (char *)&csum);
3089 if (csum != csum_expected)
3092 kunmap_atomic(kaddr);
3095 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3096 "csum failed ino %llu off %llu csum %u expected csum %u",
3097 btrfs_ino(inode), start, csum, csum_expected);
3098 memset(kaddr + pgoff, 1, len);
3099 flush_dcache_page(page);
3100 kunmap_atomic(kaddr);
3101 if (csum_expected == 0)
3107 * when reads are done, we need to check csums to verify the data is correct
3108 * if there's a match, we allow the bio to finish. If not, the code in
3109 * extent_io.c will try to find good copies for us.
3111 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3112 u64 phy_offset, struct page *page,
3113 u64 start, u64 end, int mirror)
3115 size_t offset = start - page_offset(page);
3116 struct inode *inode = page->mapping->host;
3117 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3118 struct btrfs_root *root = BTRFS_I(inode)->root;
3120 if (PageChecked(page)) {
3121 ClearPageChecked(page);
3125 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3128 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3129 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3130 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3134 phy_offset >>= inode->i_sb->s_blocksize_bits;
3135 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3136 start, (size_t)(end - start + 1));
3139 void btrfs_add_delayed_iput(struct inode *inode)
3141 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3142 struct btrfs_inode *binode = BTRFS_I(inode);
3144 if (atomic_add_unless(&inode->i_count, -1, 1))
3147 spin_lock(&fs_info->delayed_iput_lock);
3148 if (binode->delayed_iput_count == 0) {
3149 ASSERT(list_empty(&binode->delayed_iput));
3150 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3152 binode->delayed_iput_count++;
3154 spin_unlock(&fs_info->delayed_iput_lock);
3157 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3159 struct btrfs_fs_info *fs_info = root->fs_info;
3161 spin_lock(&fs_info->delayed_iput_lock);
3162 while (!list_empty(&fs_info->delayed_iputs)) {
3163 struct btrfs_inode *inode;
3165 inode = list_first_entry(&fs_info->delayed_iputs,
3166 struct btrfs_inode, delayed_iput);
3167 if (inode->delayed_iput_count) {
3168 inode->delayed_iput_count--;
3169 list_move_tail(&inode->delayed_iput,
3170 &fs_info->delayed_iputs);
3172 list_del_init(&inode->delayed_iput);
3174 spin_unlock(&fs_info->delayed_iput_lock);
3175 iput(&inode->vfs_inode);
3176 spin_lock(&fs_info->delayed_iput_lock);
3178 spin_unlock(&fs_info->delayed_iput_lock);
3182 * This is called in transaction commit time. If there are no orphan
3183 * files in the subvolume, it removes orphan item and frees block_rsv
3186 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3187 struct btrfs_root *root)
3189 struct btrfs_block_rsv *block_rsv;
3192 if (atomic_read(&root->orphan_inodes) ||
3193 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3196 spin_lock(&root->orphan_lock);
3197 if (atomic_read(&root->orphan_inodes)) {
3198 spin_unlock(&root->orphan_lock);
3202 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3203 spin_unlock(&root->orphan_lock);
3207 block_rsv = root->orphan_block_rsv;
3208 root->orphan_block_rsv = NULL;
3209 spin_unlock(&root->orphan_lock);
3211 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3212 btrfs_root_refs(&root->root_item) > 0) {
3213 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3214 root->root_key.objectid);
3216 btrfs_abort_transaction(trans, root, ret);
3218 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3223 WARN_ON(block_rsv->size > 0);
3224 btrfs_free_block_rsv(root, block_rsv);
3229 * This creates an orphan entry for the given inode in case something goes
3230 * wrong in the middle of an unlink/truncate.
3232 * NOTE: caller of this function should reserve 5 units of metadata for
3235 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3237 struct btrfs_root *root = BTRFS_I(inode)->root;
3238 struct btrfs_block_rsv *block_rsv = NULL;
3243 if (!root->orphan_block_rsv) {
3244 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3249 spin_lock(&root->orphan_lock);
3250 if (!root->orphan_block_rsv) {
3251 root->orphan_block_rsv = block_rsv;
3252 } else if (block_rsv) {
3253 btrfs_free_block_rsv(root, block_rsv);
3257 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3258 &BTRFS_I(inode)->runtime_flags)) {
3261 * For proper ENOSPC handling, we should do orphan
3262 * cleanup when mounting. But this introduces backward
3263 * compatibility issue.
3265 if (!xchg(&root->orphan_item_inserted, 1))
3271 atomic_inc(&root->orphan_inodes);
3274 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3275 &BTRFS_I(inode)->runtime_flags))
3277 spin_unlock(&root->orphan_lock);
3279 /* grab metadata reservation from transaction handle */
3281 ret = btrfs_orphan_reserve_metadata(trans, inode);
3284 atomic_dec(&root->orphan_inodes);
3285 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3286 &BTRFS_I(inode)->runtime_flags);
3288 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3289 &BTRFS_I(inode)->runtime_flags);
3294 /* insert an orphan item to track this unlinked/truncated file */
3296 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3298 atomic_dec(&root->orphan_inodes);
3300 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3301 &BTRFS_I(inode)->runtime_flags);
3302 btrfs_orphan_release_metadata(inode);
3304 if (ret != -EEXIST) {
3305 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3306 &BTRFS_I(inode)->runtime_flags);
3307 btrfs_abort_transaction(trans, root, ret);
3314 /* insert an orphan item to track subvolume contains orphan files */
3316 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3317 root->root_key.objectid);
3318 if (ret && ret != -EEXIST) {
3319 btrfs_abort_transaction(trans, root, ret);
3327 * We have done the truncate/delete so we can go ahead and remove the orphan
3328 * item for this particular inode.
3330 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3331 struct inode *inode)
3333 struct btrfs_root *root = BTRFS_I(inode)->root;
3334 int delete_item = 0;
3335 int release_rsv = 0;
3338 spin_lock(&root->orphan_lock);
3339 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3340 &BTRFS_I(inode)->runtime_flags))
3343 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3344 &BTRFS_I(inode)->runtime_flags))
3346 spin_unlock(&root->orphan_lock);
3349 atomic_dec(&root->orphan_inodes);
3351 ret = btrfs_del_orphan_item(trans, root,
3356 btrfs_orphan_release_metadata(inode);
3362 * this cleans up any orphans that may be left on the list from the last use
3365 int btrfs_orphan_cleanup(struct btrfs_root *root)
3367 struct btrfs_path *path;
3368 struct extent_buffer *leaf;
3369 struct btrfs_key key, found_key;
3370 struct btrfs_trans_handle *trans;
3371 struct inode *inode;
3372 u64 last_objectid = 0;
3373 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3375 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3378 path = btrfs_alloc_path();
3383 path->reada = READA_BACK;
3385 key.objectid = BTRFS_ORPHAN_OBJECTID;
3386 key.type = BTRFS_ORPHAN_ITEM_KEY;
3387 key.offset = (u64)-1;
3390 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3395 * if ret == 0 means we found what we were searching for, which
3396 * is weird, but possible, so only screw with path if we didn't
3397 * find the key and see if we have stuff that matches
3401 if (path->slots[0] == 0)
3406 /* pull out the item */
3407 leaf = path->nodes[0];
3408 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3410 /* make sure the item matches what we want */
3411 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3413 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3416 /* release the path since we're done with it */
3417 btrfs_release_path(path);
3420 * this is where we are basically btrfs_lookup, without the
3421 * crossing root thing. we store the inode number in the
3422 * offset of the orphan item.
3425 if (found_key.offset == last_objectid) {
3426 btrfs_err(root->fs_info,
3427 "Error removing orphan entry, stopping orphan cleanup");
3432 last_objectid = found_key.offset;
3434 found_key.objectid = found_key.offset;
3435 found_key.type = BTRFS_INODE_ITEM_KEY;
3436 found_key.offset = 0;
3437 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3438 ret = PTR_ERR_OR_ZERO(inode);
3439 if (ret && ret != -ESTALE)
3442 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3443 struct btrfs_root *dead_root;
3444 struct btrfs_fs_info *fs_info = root->fs_info;
3445 int is_dead_root = 0;
3448 * this is an orphan in the tree root. Currently these
3449 * could come from 2 sources:
3450 * a) a snapshot deletion in progress
3451 * b) a free space cache inode
3452 * We need to distinguish those two, as the snapshot
3453 * orphan must not get deleted.
3454 * find_dead_roots already ran before us, so if this
3455 * is a snapshot deletion, we should find the root
3456 * in the dead_roots list
3458 spin_lock(&fs_info->trans_lock);
3459 list_for_each_entry(dead_root, &fs_info->dead_roots,
3461 if (dead_root->root_key.objectid ==
3462 found_key.objectid) {
3467 spin_unlock(&fs_info->trans_lock);
3469 /* prevent this orphan from being found again */
3470 key.offset = found_key.objectid - 1;
3475 * Inode is already gone but the orphan item is still there,
3476 * kill the orphan item.
3478 if (ret == -ESTALE) {
3479 trans = btrfs_start_transaction(root, 1);
3480 if (IS_ERR(trans)) {
3481 ret = PTR_ERR(trans);
3484 btrfs_debug(root->fs_info, "auto deleting %Lu",
3485 found_key.objectid);
3486 ret = btrfs_del_orphan_item(trans, root,
3487 found_key.objectid);
3488 btrfs_end_transaction(trans, root);
3495 * add this inode to the orphan list so btrfs_orphan_del does
3496 * the proper thing when we hit it
3498 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3499 &BTRFS_I(inode)->runtime_flags);
3500 atomic_inc(&root->orphan_inodes);
3502 /* if we have links, this was a truncate, lets do that */
3503 if (inode->i_nlink) {
3504 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3510 /* 1 for the orphan item deletion. */
3511 trans = btrfs_start_transaction(root, 1);
3512 if (IS_ERR(trans)) {
3514 ret = PTR_ERR(trans);
3517 ret = btrfs_orphan_add(trans, inode);
3518 btrfs_end_transaction(trans, root);
3524 ret = btrfs_truncate(inode);
3526 btrfs_orphan_del(NULL, inode);
3531 /* this will do delete_inode and everything for us */
3536 /* release the path since we're done with it */
3537 btrfs_release_path(path);
3539 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3541 if (root->orphan_block_rsv)
3542 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3545 if (root->orphan_block_rsv ||
3546 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3547 trans = btrfs_join_transaction(root);
3549 btrfs_end_transaction(trans, root);
3553 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3555 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3559 btrfs_err(root->fs_info,
3560 "could not do orphan cleanup %d", ret);
3561 btrfs_free_path(path);
3566 * very simple check to peek ahead in the leaf looking for xattrs. If we
3567 * don't find any xattrs, we know there can't be any acls.
3569 * slot is the slot the inode is in, objectid is the objectid of the inode
3571 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3572 int slot, u64 objectid,
3573 int *first_xattr_slot)
3575 u32 nritems = btrfs_header_nritems(leaf);
3576 struct btrfs_key found_key;
3577 static u64 xattr_access = 0;
3578 static u64 xattr_default = 0;
3581 if (!xattr_access) {
3582 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3583 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3584 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3585 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3589 *first_xattr_slot = -1;
3590 while (slot < nritems) {
3591 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3593 /* we found a different objectid, there must not be acls */
3594 if (found_key.objectid != objectid)
3597 /* we found an xattr, assume we've got an acl */
3598 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3599 if (*first_xattr_slot == -1)
3600 *first_xattr_slot = slot;
3601 if (found_key.offset == xattr_access ||
3602 found_key.offset == xattr_default)
3607 * we found a key greater than an xattr key, there can't
3608 * be any acls later on
3610 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3617 * it goes inode, inode backrefs, xattrs, extents,
3618 * so if there are a ton of hard links to an inode there can
3619 * be a lot of backrefs. Don't waste time searching too hard,
3620 * this is just an optimization
3625 /* we hit the end of the leaf before we found an xattr or
3626 * something larger than an xattr. We have to assume the inode
3629 if (*first_xattr_slot == -1)
3630 *first_xattr_slot = slot;
3635 * read an inode from the btree into the in-memory inode
3637 static void btrfs_read_locked_inode(struct inode *inode)
3639 struct btrfs_path *path;
3640 struct extent_buffer *leaf;
3641 struct btrfs_inode_item *inode_item;
3642 struct btrfs_root *root = BTRFS_I(inode)->root;
3643 struct btrfs_key location;
3648 bool filled = false;
3649 int first_xattr_slot;
3651 ret = btrfs_fill_inode(inode, &rdev);
3655 path = btrfs_alloc_path();
3659 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3661 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3665 leaf = path->nodes[0];
3670 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3671 struct btrfs_inode_item);
3672 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3673 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3674 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3675 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3676 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3678 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3679 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3681 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3682 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3684 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3685 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3687 BTRFS_I(inode)->i_otime.tv_sec =
3688 btrfs_timespec_sec(leaf, &inode_item->otime);
3689 BTRFS_I(inode)->i_otime.tv_nsec =
3690 btrfs_timespec_nsec(leaf, &inode_item->otime);
3692 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3693 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3694 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3696 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3697 inode->i_generation = BTRFS_I(inode)->generation;
3699 rdev = btrfs_inode_rdev(leaf, inode_item);
3701 BTRFS_I(inode)->index_cnt = (u64)-1;
3702 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3706 * If we were modified in the current generation and evicted from memory
3707 * and then re-read we need to do a full sync since we don't have any
3708 * idea about which extents were modified before we were evicted from
3711 * This is required for both inode re-read from disk and delayed inode
3712 * in delayed_nodes_tree.
3714 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3715 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3716 &BTRFS_I(inode)->runtime_flags);
3719 * We don't persist the id of the transaction where an unlink operation
3720 * against the inode was last made. So here we assume the inode might
3721 * have been evicted, and therefore the exact value of last_unlink_trans
3722 * lost, and set it to last_trans to avoid metadata inconsistencies
3723 * between the inode and its parent if the inode is fsync'ed and the log
3724 * replayed. For example, in the scenario:
3727 * ln mydir/foo mydir/bar
3730 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3731 * xfs_io -c fsync mydir/foo
3733 * mount fs, triggers fsync log replay
3735 * We must make sure that when we fsync our inode foo we also log its
3736 * parent inode, otherwise after log replay the parent still has the
3737 * dentry with the "bar" name but our inode foo has a link count of 1
3738 * and doesn't have an inode ref with the name "bar" anymore.
3740 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3741 * but it guarantees correctness at the expense of occasional full
3742 * transaction commits on fsync if our inode is a directory, or if our
3743 * inode is not a directory, logging its parent unnecessarily.
3745 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3748 if (inode->i_nlink != 1 ||
3749 path->slots[0] >= btrfs_header_nritems(leaf))
3752 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3753 if (location.objectid != btrfs_ino(inode))
3756 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3757 if (location.type == BTRFS_INODE_REF_KEY) {
3758 struct btrfs_inode_ref *ref;
3760 ref = (struct btrfs_inode_ref *)ptr;
3761 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3762 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3763 struct btrfs_inode_extref *extref;
3765 extref = (struct btrfs_inode_extref *)ptr;
3766 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3771 * try to precache a NULL acl entry for files that don't have
3772 * any xattrs or acls
3774 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3775 btrfs_ino(inode), &first_xattr_slot);
3776 if (first_xattr_slot != -1) {
3777 path->slots[0] = first_xattr_slot;
3778 ret = btrfs_load_inode_props(inode, path);
3780 btrfs_err(root->fs_info,
3781 "error loading props for ino %llu (root %llu): %d",
3783 root->root_key.objectid, ret);
3785 btrfs_free_path(path);
3788 cache_no_acl(inode);
3790 switch (inode->i_mode & S_IFMT) {
3792 inode->i_mapping->a_ops = &btrfs_aops;
3793 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3794 inode->i_fop = &btrfs_file_operations;
3795 inode->i_op = &btrfs_file_inode_operations;
3798 inode->i_fop = &btrfs_dir_file_operations;
3799 if (root == root->fs_info->tree_root)
3800 inode->i_op = &btrfs_dir_ro_inode_operations;
3802 inode->i_op = &btrfs_dir_inode_operations;
3805 inode->i_op = &btrfs_symlink_inode_operations;
3806 inode_nohighmem(inode);
3807 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3810 inode->i_op = &btrfs_special_inode_operations;
3811 init_special_inode(inode, inode->i_mode, rdev);
3815 btrfs_update_iflags(inode);
3819 btrfs_free_path(path);
3820 make_bad_inode(inode);
3824 * given a leaf and an inode, copy the inode fields into the leaf
3826 static void fill_inode_item(struct btrfs_trans_handle *trans,
3827 struct extent_buffer *leaf,
3828 struct btrfs_inode_item *item,
3829 struct inode *inode)
3831 struct btrfs_map_token token;
3833 btrfs_init_map_token(&token);
3835 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3836 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3837 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3839 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3840 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3842 btrfs_set_token_timespec_sec(leaf, &item->atime,
3843 inode->i_atime.tv_sec, &token);
3844 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3845 inode->i_atime.tv_nsec, &token);
3847 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3848 inode->i_mtime.tv_sec, &token);
3849 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3850 inode->i_mtime.tv_nsec, &token);
3852 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3853 inode->i_ctime.tv_sec, &token);
3854 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3855 inode->i_ctime.tv_nsec, &token);
3857 btrfs_set_token_timespec_sec(leaf, &item->otime,
3858 BTRFS_I(inode)->i_otime.tv_sec, &token);
3859 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3860 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3862 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3864 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3866 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3867 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3868 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3869 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3870 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3874 * copy everything in the in-memory inode into the btree.
3876 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3877 struct btrfs_root *root, struct inode *inode)
3879 struct btrfs_inode_item *inode_item;
3880 struct btrfs_path *path;
3881 struct extent_buffer *leaf;
3884 path = btrfs_alloc_path();
3888 path->leave_spinning = 1;
3889 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3897 leaf = path->nodes[0];
3898 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3899 struct btrfs_inode_item);
3901 fill_inode_item(trans, leaf, inode_item, inode);
3902 btrfs_mark_buffer_dirty(leaf);
3903 btrfs_set_inode_last_trans(trans, inode);
3906 btrfs_free_path(path);
3911 * copy everything in the in-memory inode into the btree.
3913 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3914 struct btrfs_root *root, struct inode *inode)
3919 * If the inode is a free space inode, we can deadlock during commit
3920 * if we put it into the delayed code.
3922 * The data relocation inode should also be directly updated
3925 if (!btrfs_is_free_space_inode(inode)
3926 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3927 && !root->fs_info->log_root_recovering) {
3928 btrfs_update_root_times(trans, root);
3930 ret = btrfs_delayed_update_inode(trans, root, inode);
3932 btrfs_set_inode_last_trans(trans, inode);
3936 return btrfs_update_inode_item(trans, root, inode);
3939 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3940 struct btrfs_root *root,
3941 struct inode *inode)
3945 ret = btrfs_update_inode(trans, root, inode);
3947 return btrfs_update_inode_item(trans, root, inode);
3952 * unlink helper that gets used here in inode.c and in the tree logging
3953 * recovery code. It remove a link in a directory with a given name, and
3954 * also drops the back refs in the inode to the directory
3956 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3957 struct btrfs_root *root,
3958 struct inode *dir, struct inode *inode,
3959 const char *name, int name_len)
3961 struct btrfs_path *path;
3963 struct extent_buffer *leaf;
3964 struct btrfs_dir_item *di;
3965 struct btrfs_key key;
3967 u64 ino = btrfs_ino(inode);
3968 u64 dir_ino = btrfs_ino(dir);
3970 path = btrfs_alloc_path();
3976 path->leave_spinning = 1;
3977 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3978 name, name_len, -1);
3987 leaf = path->nodes[0];
3988 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3989 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3992 btrfs_release_path(path);
3995 * If we don't have dir index, we have to get it by looking up
3996 * the inode ref, since we get the inode ref, remove it directly,
3997 * it is unnecessary to do delayed deletion.
3999 * But if we have dir index, needn't search inode ref to get it.
4000 * Since the inode ref is close to the inode item, it is better
4001 * that we delay to delete it, and just do this deletion when
4002 * we update the inode item.
4004 if (BTRFS_I(inode)->dir_index) {
4005 ret = btrfs_delayed_delete_inode_ref(inode);
4007 index = BTRFS_I(inode)->dir_index;
4012 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4015 btrfs_info(root->fs_info,
4016 "failed to delete reference to %.*s, inode %llu parent %llu",
4017 name_len, name, ino, dir_ino);
4018 btrfs_abort_transaction(trans, root, ret);
4022 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4024 btrfs_abort_transaction(trans, root, ret);
4028 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4030 if (ret != 0 && ret != -ENOENT) {
4031 btrfs_abort_transaction(trans, root, ret);
4035 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4040 btrfs_abort_transaction(trans, root, ret);
4042 btrfs_free_path(path);
4046 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4047 inode_inc_iversion(inode);
4048 inode_inc_iversion(dir);
4049 inode->i_ctime = dir->i_mtime =
4050 dir->i_ctime = current_fs_time(inode->i_sb);
4051 ret = btrfs_update_inode(trans, root, dir);
4056 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4057 struct btrfs_root *root,
4058 struct inode *dir, struct inode *inode,
4059 const char *name, int name_len)
4062 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4065 ret = btrfs_update_inode(trans, root, inode);
4071 * helper to start transaction for unlink and rmdir.
4073 * unlink and rmdir are special in btrfs, they do not always free space, so
4074 * if we cannot make our reservations the normal way try and see if there is
4075 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4076 * allow the unlink to occur.
4078 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4080 struct btrfs_root *root = BTRFS_I(dir)->root;
4083 * 1 for the possible orphan item
4084 * 1 for the dir item
4085 * 1 for the dir index
4086 * 1 for the inode ref
4089 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4092 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4094 struct btrfs_root *root = BTRFS_I(dir)->root;
4095 struct btrfs_trans_handle *trans;
4096 struct inode *inode = d_inode(dentry);
4099 trans = __unlink_start_trans(dir);
4101 return PTR_ERR(trans);
4103 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4105 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4106 dentry->d_name.name, dentry->d_name.len);
4110 if (inode->i_nlink == 0) {
4111 ret = btrfs_orphan_add(trans, inode);
4117 btrfs_end_transaction(trans, root);
4118 btrfs_btree_balance_dirty(root);
4122 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4123 struct btrfs_root *root,
4124 struct inode *dir, u64 objectid,
4125 const char *name, int name_len)
4127 struct btrfs_path *path;
4128 struct extent_buffer *leaf;
4129 struct btrfs_dir_item *di;
4130 struct btrfs_key key;
4133 u64 dir_ino = btrfs_ino(dir);
4135 path = btrfs_alloc_path();
4139 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4140 name, name_len, -1);
4141 if (IS_ERR_OR_NULL(di)) {
4149 leaf = path->nodes[0];
4150 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4151 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4152 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4154 btrfs_abort_transaction(trans, root, ret);
4157 btrfs_release_path(path);
4159 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4160 objectid, root->root_key.objectid,
4161 dir_ino, &index, name, name_len);
4163 if (ret != -ENOENT) {
4164 btrfs_abort_transaction(trans, root, ret);
4167 di = btrfs_search_dir_index_item(root, path, dir_ino,
4169 if (IS_ERR_OR_NULL(di)) {
4174 btrfs_abort_transaction(trans, root, ret);
4178 leaf = path->nodes[0];
4179 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4180 btrfs_release_path(path);
4183 btrfs_release_path(path);
4185 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4187 btrfs_abort_transaction(trans, root, ret);
4191 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4192 inode_inc_iversion(dir);
4193 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4194 ret = btrfs_update_inode_fallback(trans, root, dir);
4196 btrfs_abort_transaction(trans, root, ret);
4198 btrfs_free_path(path);
4202 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4204 struct inode *inode = d_inode(dentry);
4206 struct btrfs_root *root = BTRFS_I(dir)->root;
4207 struct btrfs_trans_handle *trans;
4209 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4211 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4214 trans = __unlink_start_trans(dir);
4216 return PTR_ERR(trans);
4218 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4219 err = btrfs_unlink_subvol(trans, root, dir,
4220 BTRFS_I(inode)->location.objectid,
4221 dentry->d_name.name,
4222 dentry->d_name.len);
4226 err = btrfs_orphan_add(trans, inode);
4230 /* now the directory is empty */
4231 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4232 dentry->d_name.name, dentry->d_name.len);
4234 btrfs_i_size_write(inode, 0);
4236 btrfs_end_transaction(trans, root);
4237 btrfs_btree_balance_dirty(root);
4242 static int truncate_space_check(struct btrfs_trans_handle *trans,
4243 struct btrfs_root *root,
4249 * This is only used to apply pressure to the enospc system, we don't
4250 * intend to use this reservation at all.
4252 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4253 bytes_deleted *= root->nodesize;
4254 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4255 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4257 trace_btrfs_space_reservation(root->fs_info, "transaction",
4260 trans->bytes_reserved += bytes_deleted;
4266 static int truncate_inline_extent(struct inode *inode,
4267 struct btrfs_path *path,
4268 struct btrfs_key *found_key,
4272 struct extent_buffer *leaf = path->nodes[0];
4273 int slot = path->slots[0];
4274 struct btrfs_file_extent_item *fi;
4275 u32 size = (u32)(new_size - found_key->offset);
4276 struct btrfs_root *root = BTRFS_I(inode)->root;
4278 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4280 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4281 loff_t offset = new_size;
4282 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4285 * Zero out the remaining of the last page of our inline extent,
4286 * instead of directly truncating our inline extent here - that
4287 * would be much more complex (decompressing all the data, then
4288 * compressing the truncated data, which might be bigger than
4289 * the size of the inline extent, resize the extent, etc).
4290 * We release the path because to get the page we might need to
4291 * read the extent item from disk (data not in the page cache).
4293 btrfs_release_path(path);
4294 return btrfs_truncate_block(inode, offset, page_end - offset,
4298 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4299 size = btrfs_file_extent_calc_inline_size(size);
4300 btrfs_truncate_item(root, path, size, 1);
4302 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4303 inode_sub_bytes(inode, item_end + 1 - new_size);
4309 * this can truncate away extent items, csum items and directory items.
4310 * It starts at a high offset and removes keys until it can't find
4311 * any higher than new_size
4313 * csum items that cross the new i_size are truncated to the new size
4316 * min_type is the minimum key type to truncate down to. If set to 0, this
4317 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4319 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4320 struct btrfs_root *root,
4321 struct inode *inode,
4322 u64 new_size, u32 min_type)
4324 struct btrfs_path *path;
4325 struct extent_buffer *leaf;
4326 struct btrfs_file_extent_item *fi;
4327 struct btrfs_key key;
4328 struct btrfs_key found_key;
4329 u64 extent_start = 0;
4330 u64 extent_num_bytes = 0;
4331 u64 extent_offset = 0;
4333 u64 last_size = new_size;
4334 u32 found_type = (u8)-1;
4337 int pending_del_nr = 0;
4338 int pending_del_slot = 0;
4339 int extent_type = -1;
4342 u64 ino = btrfs_ino(inode);
4343 u64 bytes_deleted = 0;
4345 bool should_throttle = 0;
4346 bool should_end = 0;
4348 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4351 * for non-free space inodes and ref cows, we want to back off from
4354 if (!btrfs_is_free_space_inode(inode) &&
4355 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4358 path = btrfs_alloc_path();
4361 path->reada = READA_BACK;
4364 * We want to drop from the next block forward in case this new size is
4365 * not block aligned since we will be keeping the last block of the
4366 * extent just the way it is.
4368 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4369 root == root->fs_info->tree_root)
4370 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4371 root->sectorsize), (u64)-1, 0);
4374 * This function is also used to drop the items in the log tree before
4375 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4376 * it is used to drop the loged items. So we shouldn't kill the delayed
4379 if (min_type == 0 && root == BTRFS_I(inode)->root)
4380 btrfs_kill_delayed_inode_items(inode);
4383 key.offset = (u64)-1;
4388 * with a 16K leaf size and 128MB extents, you can actually queue
4389 * up a huge file in a single leaf. Most of the time that
4390 * bytes_deleted is > 0, it will be huge by the time we get here
4392 if (be_nice && bytes_deleted > SZ_32M) {
4393 if (btrfs_should_end_transaction(trans, root)) {
4400 path->leave_spinning = 1;
4401 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4408 /* there are no items in the tree for us to truncate, we're
4411 if (path->slots[0] == 0)
4418 leaf = path->nodes[0];
4419 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4420 found_type = found_key.type;
4422 if (found_key.objectid != ino)
4425 if (found_type < min_type)
4428 item_end = found_key.offset;
4429 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4430 fi = btrfs_item_ptr(leaf, path->slots[0],
4431 struct btrfs_file_extent_item);
4432 extent_type = btrfs_file_extent_type(leaf, fi);
4433 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4435 btrfs_file_extent_num_bytes(leaf, fi);
4436 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4437 item_end += btrfs_file_extent_inline_len(leaf,
4438 path->slots[0], fi);
4442 if (found_type > min_type) {
4445 if (item_end < new_size)
4447 if (found_key.offset >= new_size)
4453 /* FIXME, shrink the extent if the ref count is only 1 */
4454 if (found_type != BTRFS_EXTENT_DATA_KEY)
4458 last_size = found_key.offset;
4460 last_size = new_size;
4462 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4464 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4466 u64 orig_num_bytes =
4467 btrfs_file_extent_num_bytes(leaf, fi);
4468 extent_num_bytes = ALIGN(new_size -
4471 btrfs_set_file_extent_num_bytes(leaf, fi,
4473 num_dec = (orig_num_bytes -
4475 if (test_bit(BTRFS_ROOT_REF_COWS,
4478 inode_sub_bytes(inode, num_dec);
4479 btrfs_mark_buffer_dirty(leaf);
4482 btrfs_file_extent_disk_num_bytes(leaf,
4484 extent_offset = found_key.offset -
4485 btrfs_file_extent_offset(leaf, fi);
4487 /* FIXME blocksize != 4096 */
4488 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4489 if (extent_start != 0) {
4491 if (test_bit(BTRFS_ROOT_REF_COWS,
4493 inode_sub_bytes(inode, num_dec);
4496 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4498 * we can't truncate inline items that have had
4502 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4503 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4506 * Need to release path in order to truncate a
4507 * compressed extent. So delete any accumulated
4508 * extent items so far.
4510 if (btrfs_file_extent_compression(leaf, fi) !=
4511 BTRFS_COMPRESS_NONE && pending_del_nr) {
4512 err = btrfs_del_items(trans, root, path,
4516 btrfs_abort_transaction(trans,
4524 err = truncate_inline_extent(inode, path,
4529 btrfs_abort_transaction(trans,
4533 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4535 inode_sub_bytes(inode, item_end + 1 - new_size);
4540 if (!pending_del_nr) {
4541 /* no pending yet, add ourselves */
4542 pending_del_slot = path->slots[0];
4544 } else if (pending_del_nr &&
4545 path->slots[0] + 1 == pending_del_slot) {
4546 /* hop on the pending chunk */
4548 pending_del_slot = path->slots[0];
4555 should_throttle = 0;
4558 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4559 root == root->fs_info->tree_root)) {
4560 btrfs_set_path_blocking(path);
4561 bytes_deleted += extent_num_bytes;
4562 ret = btrfs_free_extent(trans, root, extent_start,
4563 extent_num_bytes, 0,
4564 btrfs_header_owner(leaf),
4565 ino, extent_offset);
4567 if (btrfs_should_throttle_delayed_refs(trans, root))
4568 btrfs_async_run_delayed_refs(root,
4570 trans->delayed_ref_updates * 2, 0);
4572 if (truncate_space_check(trans, root,
4573 extent_num_bytes)) {
4576 if (btrfs_should_throttle_delayed_refs(trans,
4578 should_throttle = 1;
4583 if (found_type == BTRFS_INODE_ITEM_KEY)
4586 if (path->slots[0] == 0 ||
4587 path->slots[0] != pending_del_slot ||
4588 should_throttle || should_end) {
4589 if (pending_del_nr) {
4590 ret = btrfs_del_items(trans, root, path,
4594 btrfs_abort_transaction(trans,
4600 btrfs_release_path(path);
4601 if (should_throttle) {
4602 unsigned long updates = trans->delayed_ref_updates;
4604 trans->delayed_ref_updates = 0;
4605 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4611 * if we failed to refill our space rsv, bail out
4612 * and let the transaction restart
4624 if (pending_del_nr) {
4625 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4628 btrfs_abort_transaction(trans, root, ret);
4631 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4632 btrfs_ordered_update_i_size(inode, last_size, NULL);
4634 btrfs_free_path(path);
4636 if (be_nice && bytes_deleted > SZ_32M) {
4637 unsigned long updates = trans->delayed_ref_updates;
4639 trans->delayed_ref_updates = 0;
4640 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4649 * btrfs_truncate_block - read, zero a chunk and write a block
4650 * @inode - inode that we're zeroing
4651 * @from - the offset to start zeroing
4652 * @len - the length to zero, 0 to zero the entire range respective to the
4654 * @front - zero up to the offset instead of from the offset on
4656 * This will find the block for the "from" offset and cow the block and zero the
4657 * part we want to zero. This is used with truncate and hole punching.
4659 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4662 struct address_space *mapping = inode->i_mapping;
4663 struct btrfs_root *root = BTRFS_I(inode)->root;
4664 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4665 struct btrfs_ordered_extent *ordered;
4666 struct extent_state *cached_state = NULL;
4668 u32 blocksize = root->sectorsize;
4669 pgoff_t index = from >> PAGE_SHIFT;
4670 unsigned offset = from & (blocksize - 1);
4672 gfp_t mask = btrfs_alloc_write_mask(mapping);
4677 if ((offset & (blocksize - 1)) == 0 &&
4678 (!len || ((len & (blocksize - 1)) == 0)))
4681 ret = btrfs_delalloc_reserve_space(inode,
4682 round_down(from, blocksize), blocksize);
4687 page = find_or_create_page(mapping, index, mask);
4689 btrfs_delalloc_release_space(inode,
4690 round_down(from, blocksize),
4696 block_start = round_down(from, blocksize);
4697 block_end = block_start + blocksize - 1;
4699 if (!PageUptodate(page)) {
4700 ret = btrfs_readpage(NULL, page);
4702 if (page->mapping != mapping) {
4707 if (!PageUptodate(page)) {
4712 wait_on_page_writeback(page);
4714 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4715 set_page_extent_mapped(page);
4717 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4719 unlock_extent_cached(io_tree, block_start, block_end,
4720 &cached_state, GFP_NOFS);
4723 btrfs_start_ordered_extent(inode, ordered, 1);
4724 btrfs_put_ordered_extent(ordered);
4728 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4729 EXTENT_DIRTY | EXTENT_DELALLOC |
4730 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4731 0, 0, &cached_state, GFP_NOFS);
4733 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4736 unlock_extent_cached(io_tree, block_start, block_end,
4737 &cached_state, GFP_NOFS);
4741 if (offset != blocksize) {
4743 len = blocksize - offset;
4746 memset(kaddr + (block_start - page_offset(page)),
4749 memset(kaddr + (block_start - page_offset(page)) + offset,
4751 flush_dcache_page(page);
4754 ClearPageChecked(page);
4755 set_page_dirty(page);
4756 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4761 btrfs_delalloc_release_space(inode, block_start,
4769 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4770 u64 offset, u64 len)
4772 struct btrfs_trans_handle *trans;
4776 * Still need to make sure the inode looks like it's been updated so
4777 * that any holes get logged if we fsync.
4779 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4780 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4781 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4782 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4787 * 1 - for the one we're dropping
4788 * 1 - for the one we're adding
4789 * 1 - for updating the inode.
4791 trans = btrfs_start_transaction(root, 3);
4793 return PTR_ERR(trans);
4795 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4797 btrfs_abort_transaction(trans, root, ret);
4798 btrfs_end_transaction(trans, root);
4802 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4803 0, 0, len, 0, len, 0, 0, 0);
4805 btrfs_abort_transaction(trans, root, ret);
4807 btrfs_update_inode(trans, root, inode);
4808 btrfs_end_transaction(trans, root);
4813 * This function puts in dummy file extents for the area we're creating a hole
4814 * for. So if we are truncating this file to a larger size we need to insert
4815 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4816 * the range between oldsize and size
4818 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4820 struct btrfs_root *root = BTRFS_I(inode)->root;
4821 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4822 struct extent_map *em = NULL;
4823 struct extent_state *cached_state = NULL;
4824 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4825 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4826 u64 block_end = ALIGN(size, root->sectorsize);
4833 * If our size started in the middle of a block we need to zero out the
4834 * rest of the block before we expand the i_size, otherwise we could
4835 * expose stale data.
4837 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4841 if (size <= hole_start)
4845 struct btrfs_ordered_extent *ordered;
4847 lock_extent_bits(io_tree, hole_start, block_end - 1,
4849 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4850 block_end - hole_start);
4853 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4854 &cached_state, GFP_NOFS);
4855 btrfs_start_ordered_extent(inode, ordered, 1);
4856 btrfs_put_ordered_extent(ordered);
4859 cur_offset = hole_start;
4861 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4862 block_end - cur_offset, 0);
4868 last_byte = min(extent_map_end(em), block_end);
4869 last_byte = ALIGN(last_byte , root->sectorsize);
4870 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4871 struct extent_map *hole_em;
4872 hole_size = last_byte - cur_offset;
4874 err = maybe_insert_hole(root, inode, cur_offset,
4878 btrfs_drop_extent_cache(inode, cur_offset,
4879 cur_offset + hole_size - 1, 0);
4880 hole_em = alloc_extent_map();
4882 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4883 &BTRFS_I(inode)->runtime_flags);
4886 hole_em->start = cur_offset;
4887 hole_em->len = hole_size;
4888 hole_em->orig_start = cur_offset;
4890 hole_em->block_start = EXTENT_MAP_HOLE;
4891 hole_em->block_len = 0;
4892 hole_em->orig_block_len = 0;
4893 hole_em->ram_bytes = hole_size;
4894 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4895 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4896 hole_em->generation = root->fs_info->generation;
4899 write_lock(&em_tree->lock);
4900 err = add_extent_mapping(em_tree, hole_em, 1);
4901 write_unlock(&em_tree->lock);
4904 btrfs_drop_extent_cache(inode, cur_offset,
4908 free_extent_map(hole_em);
4911 free_extent_map(em);
4913 cur_offset = last_byte;
4914 if (cur_offset >= block_end)
4917 free_extent_map(em);
4918 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4923 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4925 struct btrfs_root *root = BTRFS_I(inode)->root;
4926 struct btrfs_trans_handle *trans;
4927 loff_t oldsize = i_size_read(inode);
4928 loff_t newsize = attr->ia_size;
4929 int mask = attr->ia_valid;
4933 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4934 * special case where we need to update the times despite not having
4935 * these flags set. For all other operations the VFS set these flags
4936 * explicitly if it wants a timestamp update.
4938 if (newsize != oldsize) {
4939 inode_inc_iversion(inode);
4940 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4941 inode->i_ctime = inode->i_mtime =
4942 current_fs_time(inode->i_sb);
4945 if (newsize > oldsize) {
4947 * Don't do an expanding truncate while snapshoting is ongoing.
4948 * This is to ensure the snapshot captures a fully consistent
4949 * state of this file - if the snapshot captures this expanding
4950 * truncation, it must capture all writes that happened before
4953 btrfs_wait_for_snapshot_creation(root);
4954 ret = btrfs_cont_expand(inode, oldsize, newsize);
4956 btrfs_end_write_no_snapshoting(root);
4960 trans = btrfs_start_transaction(root, 1);
4961 if (IS_ERR(trans)) {
4962 btrfs_end_write_no_snapshoting(root);
4963 return PTR_ERR(trans);
4966 i_size_write(inode, newsize);
4967 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4968 pagecache_isize_extended(inode, oldsize, newsize);
4969 ret = btrfs_update_inode(trans, root, inode);
4970 btrfs_end_write_no_snapshoting(root);
4971 btrfs_end_transaction(trans, root);
4975 * We're truncating a file that used to have good data down to
4976 * zero. Make sure it gets into the ordered flush list so that
4977 * any new writes get down to disk quickly.
4980 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4981 &BTRFS_I(inode)->runtime_flags);
4984 * 1 for the orphan item we're going to add
4985 * 1 for the orphan item deletion.
4987 trans = btrfs_start_transaction(root, 2);
4989 return PTR_ERR(trans);
4992 * We need to do this in case we fail at _any_ point during the
4993 * actual truncate. Once we do the truncate_setsize we could
4994 * invalidate pages which forces any outstanding ordered io to
4995 * be instantly completed which will give us extents that need
4996 * to be truncated. If we fail to get an orphan inode down we
4997 * could have left over extents that were never meant to live,
4998 * so we need to guarantee from this point on that everything
4999 * will be consistent.
5001 ret = btrfs_orphan_add(trans, inode);
5002 btrfs_end_transaction(trans, root);
5006 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5007 truncate_setsize(inode, newsize);
5009 /* Disable nonlocked read DIO to avoid the end less truncate */
5010 btrfs_inode_block_unlocked_dio(inode);
5011 inode_dio_wait(inode);
5012 btrfs_inode_resume_unlocked_dio(inode);
5014 ret = btrfs_truncate(inode);
5015 if (ret && inode->i_nlink) {
5019 * failed to truncate, disk_i_size is only adjusted down
5020 * as we remove extents, so it should represent the true
5021 * size of the inode, so reset the in memory size and
5022 * delete our orphan entry.
5024 trans = btrfs_join_transaction(root);
5025 if (IS_ERR(trans)) {
5026 btrfs_orphan_del(NULL, inode);
5029 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5030 err = btrfs_orphan_del(trans, inode);
5032 btrfs_abort_transaction(trans, root, err);
5033 btrfs_end_transaction(trans, root);
5040 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5042 struct inode *inode = d_inode(dentry);
5043 struct btrfs_root *root = BTRFS_I(inode)->root;
5046 if (btrfs_root_readonly(root))
5049 err = inode_change_ok(inode, attr);
5053 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5054 err = btrfs_setsize(inode, attr);
5059 if (attr->ia_valid) {
5060 setattr_copy(inode, attr);
5061 inode_inc_iversion(inode);
5062 err = btrfs_dirty_inode(inode);
5064 if (!err && attr->ia_valid & ATTR_MODE)
5065 err = posix_acl_chmod(inode, inode->i_mode);
5072 * While truncating the inode pages during eviction, we get the VFS calling
5073 * btrfs_invalidatepage() against each page of the inode. This is slow because
5074 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5075 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5076 * extent_state structures over and over, wasting lots of time.
5078 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5079 * those expensive operations on a per page basis and do only the ordered io
5080 * finishing, while we release here the extent_map and extent_state structures,
5081 * without the excessive merging and splitting.
5083 static void evict_inode_truncate_pages(struct inode *inode)
5085 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5086 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5087 struct rb_node *node;
5089 ASSERT(inode->i_state & I_FREEING);
5090 truncate_inode_pages_final(&inode->i_data);
5092 write_lock(&map_tree->lock);
5093 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5094 struct extent_map *em;
5096 node = rb_first(&map_tree->map);
5097 em = rb_entry(node, struct extent_map, rb_node);
5098 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5099 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5100 remove_extent_mapping(map_tree, em);
5101 free_extent_map(em);
5102 if (need_resched()) {
5103 write_unlock(&map_tree->lock);
5105 write_lock(&map_tree->lock);
5108 write_unlock(&map_tree->lock);
5111 * Keep looping until we have no more ranges in the io tree.
5112 * We can have ongoing bios started by readpages (called from readahead)
5113 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5114 * still in progress (unlocked the pages in the bio but did not yet
5115 * unlocked the ranges in the io tree). Therefore this means some
5116 * ranges can still be locked and eviction started because before
5117 * submitting those bios, which are executed by a separate task (work
5118 * queue kthread), inode references (inode->i_count) were not taken
5119 * (which would be dropped in the end io callback of each bio).
5120 * Therefore here we effectively end up waiting for those bios and
5121 * anyone else holding locked ranges without having bumped the inode's
5122 * reference count - if we don't do it, when they access the inode's
5123 * io_tree to unlock a range it may be too late, leading to an
5124 * use-after-free issue.
5126 spin_lock(&io_tree->lock);
5127 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5128 struct extent_state *state;
5129 struct extent_state *cached_state = NULL;
5133 node = rb_first(&io_tree->state);
5134 state = rb_entry(node, struct extent_state, rb_node);
5135 start = state->start;
5137 spin_unlock(&io_tree->lock);
5139 lock_extent_bits(io_tree, start, end, &cached_state);
5142 * If still has DELALLOC flag, the extent didn't reach disk,
5143 * and its reserved space won't be freed by delayed_ref.
5144 * So we need to free its reserved space here.
5145 * (Refer to comment in btrfs_invalidatepage, case 2)
5147 * Note, end is the bytenr of last byte, so we need + 1 here.
5149 if (state->state & EXTENT_DELALLOC)
5150 btrfs_qgroup_free_data(inode, start, end - start + 1);
5152 clear_extent_bit(io_tree, start, end,
5153 EXTENT_LOCKED | EXTENT_DIRTY |
5154 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5155 EXTENT_DEFRAG, 1, 1,
5156 &cached_state, GFP_NOFS);
5159 spin_lock(&io_tree->lock);
5161 spin_unlock(&io_tree->lock);
5164 void btrfs_evict_inode(struct inode *inode)
5166 struct btrfs_trans_handle *trans;
5167 struct btrfs_root *root = BTRFS_I(inode)->root;
5168 struct btrfs_block_rsv *rsv, *global_rsv;
5169 int steal_from_global = 0;
5173 trace_btrfs_inode_evict(inode);
5176 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5180 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5182 evict_inode_truncate_pages(inode);
5184 if (inode->i_nlink &&
5185 ((btrfs_root_refs(&root->root_item) != 0 &&
5186 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5187 btrfs_is_free_space_inode(inode)))
5190 if (is_bad_inode(inode)) {
5191 btrfs_orphan_del(NULL, inode);
5194 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5195 if (!special_file(inode->i_mode))
5196 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5198 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5200 if (root->fs_info->log_root_recovering) {
5201 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5202 &BTRFS_I(inode)->runtime_flags));
5206 if (inode->i_nlink > 0) {
5207 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5208 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5212 ret = btrfs_commit_inode_delayed_inode(inode);
5214 btrfs_orphan_del(NULL, inode);
5218 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5220 btrfs_orphan_del(NULL, inode);
5223 rsv->size = min_size;
5225 global_rsv = &root->fs_info->global_block_rsv;
5227 btrfs_i_size_write(inode, 0);
5230 * This is a bit simpler than btrfs_truncate since we've already
5231 * reserved our space for our orphan item in the unlink, so we just
5232 * need to reserve some slack space in case we add bytes and update
5233 * inode item when doing the truncate.
5236 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5237 BTRFS_RESERVE_FLUSH_LIMIT);
5240 * Try and steal from the global reserve since we will
5241 * likely not use this space anyway, we want to try as
5242 * hard as possible to get this to work.
5245 steal_from_global++;
5247 steal_from_global = 0;
5251 * steal_from_global == 0: we reserved stuff, hooray!
5252 * steal_from_global == 1: we didn't reserve stuff, boo!
5253 * steal_from_global == 2: we've committed, still not a lot of
5254 * room but maybe we'll have room in the global reserve this
5256 * steal_from_global == 3: abandon all hope!
5258 if (steal_from_global > 2) {
5259 btrfs_warn(root->fs_info,
5260 "Could not get space for a delete, will truncate on mount %d",
5262 btrfs_orphan_del(NULL, inode);
5263 btrfs_free_block_rsv(root, rsv);
5267 trans = btrfs_join_transaction(root);
5268 if (IS_ERR(trans)) {
5269 btrfs_orphan_del(NULL, inode);
5270 btrfs_free_block_rsv(root, rsv);
5275 * We can't just steal from the global reserve, we need to make
5276 * sure there is room to do it, if not we need to commit and try
5279 if (steal_from_global) {
5280 if (!btrfs_check_space_for_delayed_refs(trans, root))
5281 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5288 * Couldn't steal from the global reserve, we have too much
5289 * pending stuff built up, commit the transaction and try it
5293 ret = btrfs_commit_transaction(trans, root);
5295 btrfs_orphan_del(NULL, inode);
5296 btrfs_free_block_rsv(root, rsv);
5301 steal_from_global = 0;
5304 trans->block_rsv = rsv;
5306 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5307 if (ret != -ENOSPC && ret != -EAGAIN)
5310 trans->block_rsv = &root->fs_info->trans_block_rsv;
5311 btrfs_end_transaction(trans, root);
5313 btrfs_btree_balance_dirty(root);
5316 btrfs_free_block_rsv(root, rsv);
5319 * Errors here aren't a big deal, it just means we leave orphan items
5320 * in the tree. They will be cleaned up on the next mount.
5323 trans->block_rsv = root->orphan_block_rsv;
5324 btrfs_orphan_del(trans, inode);
5326 btrfs_orphan_del(NULL, inode);
5329 trans->block_rsv = &root->fs_info->trans_block_rsv;
5330 if (!(root == root->fs_info->tree_root ||
5331 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5332 btrfs_return_ino(root, btrfs_ino(inode));
5334 btrfs_end_transaction(trans, root);
5335 btrfs_btree_balance_dirty(root);
5337 btrfs_remove_delayed_node(inode);
5342 * this returns the key found in the dir entry in the location pointer.
5343 * If no dir entries were found, location->objectid is 0.
5345 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5346 struct btrfs_key *location)
5348 const char *name = dentry->d_name.name;
5349 int namelen = dentry->d_name.len;
5350 struct btrfs_dir_item *di;
5351 struct btrfs_path *path;
5352 struct btrfs_root *root = BTRFS_I(dir)->root;
5355 path = btrfs_alloc_path();
5359 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5364 if (IS_ERR_OR_NULL(di))
5367 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5369 btrfs_free_path(path);
5372 location->objectid = 0;
5377 * when we hit a tree root in a directory, the btrfs part of the inode
5378 * needs to be changed to reflect the root directory of the tree root. This
5379 * is kind of like crossing a mount point.
5381 static int fixup_tree_root_location(struct btrfs_root *root,
5383 struct dentry *dentry,
5384 struct btrfs_key *location,
5385 struct btrfs_root **sub_root)
5387 struct btrfs_path *path;
5388 struct btrfs_root *new_root;
5389 struct btrfs_root_ref *ref;
5390 struct extent_buffer *leaf;
5391 struct btrfs_key key;
5395 path = btrfs_alloc_path();
5402 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5403 key.type = BTRFS_ROOT_REF_KEY;
5404 key.offset = location->objectid;
5406 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5414 leaf = path->nodes[0];
5415 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5416 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5417 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5420 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5421 (unsigned long)(ref + 1),
5422 dentry->d_name.len);
5426 btrfs_release_path(path);
5428 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5429 if (IS_ERR(new_root)) {
5430 err = PTR_ERR(new_root);
5434 *sub_root = new_root;
5435 location->objectid = btrfs_root_dirid(&new_root->root_item);
5436 location->type = BTRFS_INODE_ITEM_KEY;
5437 location->offset = 0;
5440 btrfs_free_path(path);
5444 static void inode_tree_add(struct inode *inode)
5446 struct btrfs_root *root = BTRFS_I(inode)->root;
5447 struct btrfs_inode *entry;
5449 struct rb_node *parent;
5450 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5451 u64 ino = btrfs_ino(inode);
5453 if (inode_unhashed(inode))
5456 spin_lock(&root->inode_lock);
5457 p = &root->inode_tree.rb_node;
5460 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5462 if (ino < btrfs_ino(&entry->vfs_inode))
5463 p = &parent->rb_left;
5464 else if (ino > btrfs_ino(&entry->vfs_inode))
5465 p = &parent->rb_right;
5467 WARN_ON(!(entry->vfs_inode.i_state &
5468 (I_WILL_FREE | I_FREEING)));
5469 rb_replace_node(parent, new, &root->inode_tree);
5470 RB_CLEAR_NODE(parent);
5471 spin_unlock(&root->inode_lock);
5475 rb_link_node(new, parent, p);
5476 rb_insert_color(new, &root->inode_tree);
5477 spin_unlock(&root->inode_lock);
5480 static void inode_tree_del(struct inode *inode)
5482 struct btrfs_root *root = BTRFS_I(inode)->root;
5485 spin_lock(&root->inode_lock);
5486 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5487 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5488 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5489 empty = RB_EMPTY_ROOT(&root->inode_tree);
5491 spin_unlock(&root->inode_lock);
5493 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5494 synchronize_srcu(&root->fs_info->subvol_srcu);
5495 spin_lock(&root->inode_lock);
5496 empty = RB_EMPTY_ROOT(&root->inode_tree);
5497 spin_unlock(&root->inode_lock);
5499 btrfs_add_dead_root(root);
5503 void btrfs_invalidate_inodes(struct btrfs_root *root)
5505 struct rb_node *node;
5506 struct rb_node *prev;
5507 struct btrfs_inode *entry;
5508 struct inode *inode;
5511 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5512 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5514 spin_lock(&root->inode_lock);
5516 node = root->inode_tree.rb_node;
5520 entry = rb_entry(node, struct btrfs_inode, rb_node);
5522 if (objectid < btrfs_ino(&entry->vfs_inode))
5523 node = node->rb_left;
5524 else if (objectid > btrfs_ino(&entry->vfs_inode))
5525 node = node->rb_right;
5531 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5532 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5536 prev = rb_next(prev);
5540 entry = rb_entry(node, struct btrfs_inode, rb_node);
5541 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5542 inode = igrab(&entry->vfs_inode);
5544 spin_unlock(&root->inode_lock);
5545 if (atomic_read(&inode->i_count) > 1)
5546 d_prune_aliases(inode);
5548 * btrfs_drop_inode will have it removed from
5549 * the inode cache when its usage count
5554 spin_lock(&root->inode_lock);
5558 if (cond_resched_lock(&root->inode_lock))
5561 node = rb_next(node);
5563 spin_unlock(&root->inode_lock);
5566 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5568 struct btrfs_iget_args *args = p;
5569 inode->i_ino = args->location->objectid;
5570 memcpy(&BTRFS_I(inode)->location, args->location,
5571 sizeof(*args->location));
5572 BTRFS_I(inode)->root = args->root;
5576 static int btrfs_find_actor(struct inode *inode, void *opaque)
5578 struct btrfs_iget_args *args = opaque;
5579 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5580 args->root == BTRFS_I(inode)->root;
5583 static struct inode *btrfs_iget_locked(struct super_block *s,
5584 struct btrfs_key *location,
5585 struct btrfs_root *root)
5587 struct inode *inode;
5588 struct btrfs_iget_args args;
5589 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5591 args.location = location;
5594 inode = iget5_locked(s, hashval, btrfs_find_actor,
5595 btrfs_init_locked_inode,
5600 /* Get an inode object given its location and corresponding root.
5601 * Returns in *is_new if the inode was read from disk
5603 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5604 struct btrfs_root *root, int *new)
5606 struct inode *inode;
5608 inode = btrfs_iget_locked(s, location, root);
5610 return ERR_PTR(-ENOMEM);
5612 if (inode->i_state & I_NEW) {
5613 btrfs_read_locked_inode(inode);
5614 if (!is_bad_inode(inode)) {
5615 inode_tree_add(inode);
5616 unlock_new_inode(inode);
5620 unlock_new_inode(inode);
5622 inode = ERR_PTR(-ESTALE);
5629 static struct inode *new_simple_dir(struct super_block *s,
5630 struct btrfs_key *key,
5631 struct btrfs_root *root)
5633 struct inode *inode = new_inode(s);
5636 return ERR_PTR(-ENOMEM);
5638 BTRFS_I(inode)->root = root;
5639 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5640 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5642 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5643 inode->i_op = &btrfs_dir_ro_inode_operations;
5644 inode->i_fop = &simple_dir_operations;
5645 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5646 inode->i_mtime = current_fs_time(inode->i_sb);
5647 inode->i_atime = inode->i_mtime;
5648 inode->i_ctime = inode->i_mtime;
5649 BTRFS_I(inode)->i_otime = inode->i_mtime;
5654 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5656 struct inode *inode;
5657 struct btrfs_root *root = BTRFS_I(dir)->root;
5658 struct btrfs_root *sub_root = root;
5659 struct btrfs_key location;
5663 if (dentry->d_name.len > BTRFS_NAME_LEN)
5664 return ERR_PTR(-ENAMETOOLONG);
5666 ret = btrfs_inode_by_name(dir, dentry, &location);
5668 return ERR_PTR(ret);
5670 if (location.objectid == 0)
5671 return ERR_PTR(-ENOENT);
5673 if (location.type == BTRFS_INODE_ITEM_KEY) {
5674 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5678 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5680 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5681 ret = fixup_tree_root_location(root, dir, dentry,
5682 &location, &sub_root);
5685 inode = ERR_PTR(ret);
5687 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5689 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5691 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5693 if (!IS_ERR(inode) && root != sub_root) {
5694 down_read(&root->fs_info->cleanup_work_sem);
5695 if (!(inode->i_sb->s_flags & MS_RDONLY))
5696 ret = btrfs_orphan_cleanup(sub_root);
5697 up_read(&root->fs_info->cleanup_work_sem);
5700 inode = ERR_PTR(ret);
5707 static int btrfs_dentry_delete(const struct dentry *dentry)
5709 struct btrfs_root *root;
5710 struct inode *inode = d_inode(dentry);
5712 if (!inode && !IS_ROOT(dentry))
5713 inode = d_inode(dentry->d_parent);
5716 root = BTRFS_I(inode)->root;
5717 if (btrfs_root_refs(&root->root_item) == 0)
5720 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5726 static void btrfs_dentry_release(struct dentry *dentry)
5728 kfree(dentry->d_fsdata);
5731 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5734 struct inode *inode;
5736 inode = btrfs_lookup_dentry(dir, dentry);
5737 if (IS_ERR(inode)) {
5738 if (PTR_ERR(inode) == -ENOENT)
5741 return ERR_CAST(inode);
5744 return d_splice_alias(inode, dentry);
5747 unsigned char btrfs_filetype_table[] = {
5748 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5751 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5753 struct inode *inode = file_inode(file);
5754 struct btrfs_root *root = BTRFS_I(inode)->root;
5755 struct btrfs_item *item;
5756 struct btrfs_dir_item *di;
5757 struct btrfs_key key;
5758 struct btrfs_key found_key;
5759 struct btrfs_path *path;
5760 struct list_head ins_list;
5761 struct list_head del_list;
5763 struct extent_buffer *leaf;
5765 unsigned char d_type;
5770 int key_type = BTRFS_DIR_INDEX_KEY;
5774 int is_curr = 0; /* ctx->pos points to the current index? */
5778 /* FIXME, use a real flag for deciding about the key type */
5779 if (root->fs_info->tree_root == root)
5780 key_type = BTRFS_DIR_ITEM_KEY;
5782 if (!dir_emit_dots(file, ctx))
5785 path = btrfs_alloc_path();
5789 path->reada = READA_FORWARD;
5791 if (key_type == BTRFS_DIR_INDEX_KEY) {
5792 INIT_LIST_HEAD(&ins_list);
5793 INIT_LIST_HEAD(&del_list);
5794 put = btrfs_readdir_get_delayed_items(inode, &ins_list,
5798 key.type = key_type;
5799 key.offset = ctx->pos;
5800 key.objectid = btrfs_ino(inode);
5802 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5808 leaf = path->nodes[0];
5809 slot = path->slots[0];
5810 if (slot >= btrfs_header_nritems(leaf)) {
5811 ret = btrfs_next_leaf(root, path);
5819 item = btrfs_item_nr(slot);
5820 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5822 if (found_key.objectid != key.objectid)
5824 if (found_key.type != key_type)
5826 if (found_key.offset < ctx->pos)
5828 if (key_type == BTRFS_DIR_INDEX_KEY &&
5829 btrfs_should_delete_dir_index(&del_list,
5833 ctx->pos = found_key.offset;
5836 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5838 di_total = btrfs_item_size(leaf, item);
5840 while (di_cur < di_total) {
5841 struct btrfs_key location;
5843 if (verify_dir_item(root, leaf, di))
5846 name_len = btrfs_dir_name_len(leaf, di);
5847 if (name_len <= sizeof(tmp_name)) {
5848 name_ptr = tmp_name;
5850 name_ptr = kmalloc(name_len, GFP_KERNEL);
5856 read_extent_buffer(leaf, name_ptr,
5857 (unsigned long)(di + 1), name_len);
5859 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5860 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5863 /* is this a reference to our own snapshot? If so
5866 * In contrast to old kernels, we insert the snapshot's
5867 * dir item and dir index after it has been created, so
5868 * we won't find a reference to our own snapshot. We
5869 * still keep the following code for backward
5872 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5873 location.objectid == root->root_key.objectid) {
5877 over = !dir_emit(ctx, name_ptr, name_len,
5878 location.objectid, d_type);
5881 if (name_ptr != tmp_name)
5887 di_len = btrfs_dir_name_len(leaf, di) +
5888 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5890 di = (struct btrfs_dir_item *)((char *)di + di_len);
5896 if (key_type == BTRFS_DIR_INDEX_KEY) {
5899 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5905 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5906 * it was was set to the termination value in previous call. We assume
5907 * that "." and ".." were emitted if we reach this point and set the
5908 * termination value as well for an empty directory.
5910 if (ctx->pos > 2 && !emitted)
5913 /* Reached end of directory/root. Bump pos past the last item. */
5917 * Stop new entries from being returned after we return the last
5920 * New directory entries are assigned a strictly increasing
5921 * offset. This means that new entries created during readdir
5922 * are *guaranteed* to be seen in the future by that readdir.
5923 * This has broken buggy programs which operate on names as
5924 * they're returned by readdir. Until we re-use freed offsets
5925 * we have this hack to stop new entries from being returned
5926 * under the assumption that they'll never reach this huge
5929 * This is being careful not to overflow 32bit loff_t unless the
5930 * last entry requires it because doing so has broken 32bit apps
5933 if (key_type == BTRFS_DIR_INDEX_KEY) {
5934 if (ctx->pos >= INT_MAX)
5935 ctx->pos = LLONG_MAX;
5943 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5944 btrfs_free_path(path);
5948 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5950 struct btrfs_root *root = BTRFS_I(inode)->root;
5951 struct btrfs_trans_handle *trans;
5953 bool nolock = false;
5955 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5958 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5961 if (wbc->sync_mode == WB_SYNC_ALL) {
5963 trans = btrfs_join_transaction_nolock(root);
5965 trans = btrfs_join_transaction(root);
5967 return PTR_ERR(trans);
5968 ret = btrfs_commit_transaction(trans, root);
5974 * This is somewhat expensive, updating the tree every time the
5975 * inode changes. But, it is most likely to find the inode in cache.
5976 * FIXME, needs more benchmarking...there are no reasons other than performance
5977 * to keep or drop this code.
5979 static int btrfs_dirty_inode(struct inode *inode)
5981 struct btrfs_root *root = BTRFS_I(inode)->root;
5982 struct btrfs_trans_handle *trans;
5985 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5988 trans = btrfs_join_transaction(root);
5990 return PTR_ERR(trans);
5992 ret = btrfs_update_inode(trans, root, inode);
5993 if (ret && ret == -ENOSPC) {
5994 /* whoops, lets try again with the full transaction */
5995 btrfs_end_transaction(trans, root);
5996 trans = btrfs_start_transaction(root, 1);
5998 return PTR_ERR(trans);
6000 ret = btrfs_update_inode(trans, root, inode);
6002 btrfs_end_transaction(trans, root);
6003 if (BTRFS_I(inode)->delayed_node)
6004 btrfs_balance_delayed_items(root);
6010 * This is a copy of file_update_time. We need this so we can return error on
6011 * ENOSPC for updating the inode in the case of file write and mmap writes.
6013 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6016 struct btrfs_root *root = BTRFS_I(inode)->root;
6018 if (btrfs_root_readonly(root))
6021 if (flags & S_VERSION)
6022 inode_inc_iversion(inode);
6023 if (flags & S_CTIME)
6024 inode->i_ctime = *now;
6025 if (flags & S_MTIME)
6026 inode->i_mtime = *now;
6027 if (flags & S_ATIME)
6028 inode->i_atime = *now;
6029 return btrfs_dirty_inode(inode);
6033 * find the highest existing sequence number in a directory
6034 * and then set the in-memory index_cnt variable to reflect
6035 * free sequence numbers
6037 static int btrfs_set_inode_index_count(struct inode *inode)
6039 struct btrfs_root *root = BTRFS_I(inode)->root;
6040 struct btrfs_key key, found_key;
6041 struct btrfs_path *path;
6042 struct extent_buffer *leaf;
6045 key.objectid = btrfs_ino(inode);
6046 key.type = BTRFS_DIR_INDEX_KEY;
6047 key.offset = (u64)-1;
6049 path = btrfs_alloc_path();
6053 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6056 /* FIXME: we should be able to handle this */
6062 * MAGIC NUMBER EXPLANATION:
6063 * since we search a directory based on f_pos we have to start at 2
6064 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6065 * else has to start at 2
6067 if (path->slots[0] == 0) {
6068 BTRFS_I(inode)->index_cnt = 2;
6074 leaf = path->nodes[0];
6075 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6077 if (found_key.objectid != btrfs_ino(inode) ||
6078 found_key.type != BTRFS_DIR_INDEX_KEY) {
6079 BTRFS_I(inode)->index_cnt = 2;
6083 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6085 btrfs_free_path(path);
6090 * helper to find a free sequence number in a given directory. This current
6091 * code is very simple, later versions will do smarter things in the btree
6093 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6097 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6098 ret = btrfs_inode_delayed_dir_index_count(dir);
6100 ret = btrfs_set_inode_index_count(dir);
6106 *index = BTRFS_I(dir)->index_cnt;
6107 BTRFS_I(dir)->index_cnt++;
6112 static int btrfs_insert_inode_locked(struct inode *inode)
6114 struct btrfs_iget_args args;
6115 args.location = &BTRFS_I(inode)->location;
6116 args.root = BTRFS_I(inode)->root;
6118 return insert_inode_locked4(inode,
6119 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6120 btrfs_find_actor, &args);
6123 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6124 struct btrfs_root *root,
6126 const char *name, int name_len,
6127 u64 ref_objectid, u64 objectid,
6128 umode_t mode, u64 *index)
6130 struct inode *inode;
6131 struct btrfs_inode_item *inode_item;
6132 struct btrfs_key *location;
6133 struct btrfs_path *path;
6134 struct btrfs_inode_ref *ref;
6135 struct btrfs_key key[2];
6137 int nitems = name ? 2 : 1;
6141 path = btrfs_alloc_path();
6143 return ERR_PTR(-ENOMEM);
6145 inode = new_inode(root->fs_info->sb);
6147 btrfs_free_path(path);
6148 return ERR_PTR(-ENOMEM);
6152 * O_TMPFILE, set link count to 0, so that after this point,
6153 * we fill in an inode item with the correct link count.
6156 set_nlink(inode, 0);
6159 * we have to initialize this early, so we can reclaim the inode
6160 * number if we fail afterwards in this function.
6162 inode->i_ino = objectid;
6165 trace_btrfs_inode_request(dir);
6167 ret = btrfs_set_inode_index(dir, index);
6169 btrfs_free_path(path);
6171 return ERR_PTR(ret);
6177 * index_cnt is ignored for everything but a dir,
6178 * btrfs_get_inode_index_count has an explanation for the magic
6181 BTRFS_I(inode)->index_cnt = 2;
6182 BTRFS_I(inode)->dir_index = *index;
6183 BTRFS_I(inode)->root = root;
6184 BTRFS_I(inode)->generation = trans->transid;
6185 inode->i_generation = BTRFS_I(inode)->generation;
6188 * We could have gotten an inode number from somebody who was fsynced
6189 * and then removed in this same transaction, so let's just set full
6190 * sync since it will be a full sync anyway and this will blow away the
6191 * old info in the log.
6193 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6195 key[0].objectid = objectid;
6196 key[0].type = BTRFS_INODE_ITEM_KEY;
6199 sizes[0] = sizeof(struct btrfs_inode_item);
6203 * Start new inodes with an inode_ref. This is slightly more
6204 * efficient for small numbers of hard links since they will
6205 * be packed into one item. Extended refs will kick in if we
6206 * add more hard links than can fit in the ref item.
6208 key[1].objectid = objectid;
6209 key[1].type = BTRFS_INODE_REF_KEY;
6210 key[1].offset = ref_objectid;
6212 sizes[1] = name_len + sizeof(*ref);
6215 location = &BTRFS_I(inode)->location;
6216 location->objectid = objectid;
6217 location->offset = 0;
6218 location->type = BTRFS_INODE_ITEM_KEY;
6220 ret = btrfs_insert_inode_locked(inode);
6224 path->leave_spinning = 1;
6225 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6229 inode_init_owner(inode, dir, mode);
6230 inode_set_bytes(inode, 0);
6232 inode->i_mtime = current_fs_time(inode->i_sb);
6233 inode->i_atime = inode->i_mtime;
6234 inode->i_ctime = inode->i_mtime;
6235 BTRFS_I(inode)->i_otime = inode->i_mtime;
6237 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6238 struct btrfs_inode_item);
6239 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6240 sizeof(*inode_item));
6241 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6244 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6245 struct btrfs_inode_ref);
6246 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6247 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6248 ptr = (unsigned long)(ref + 1);
6249 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6252 btrfs_mark_buffer_dirty(path->nodes[0]);
6253 btrfs_free_path(path);
6255 btrfs_inherit_iflags(inode, dir);
6257 if (S_ISREG(mode)) {
6258 if (btrfs_test_opt(root->fs_info, NODATASUM))
6259 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6260 if (btrfs_test_opt(root->fs_info, NODATACOW))
6261 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6262 BTRFS_INODE_NODATASUM;
6265 inode_tree_add(inode);
6267 trace_btrfs_inode_new(inode);
6268 btrfs_set_inode_last_trans(trans, inode);
6270 btrfs_update_root_times(trans, root);
6272 ret = btrfs_inode_inherit_props(trans, inode, dir);
6274 btrfs_err(root->fs_info,
6275 "error inheriting props for ino %llu (root %llu): %d",
6276 btrfs_ino(inode), root->root_key.objectid, ret);
6281 unlock_new_inode(inode);
6284 BTRFS_I(dir)->index_cnt--;
6285 btrfs_free_path(path);
6287 return ERR_PTR(ret);
6290 static inline u8 btrfs_inode_type(struct inode *inode)
6292 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6296 * utility function to add 'inode' into 'parent_inode' with
6297 * a give name and a given sequence number.
6298 * if 'add_backref' is true, also insert a backref from the
6299 * inode to the parent directory.
6301 int btrfs_add_link(struct btrfs_trans_handle *trans,
6302 struct inode *parent_inode, struct inode *inode,
6303 const char *name, int name_len, int add_backref, u64 index)
6306 struct btrfs_key key;
6307 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6308 u64 ino = btrfs_ino(inode);
6309 u64 parent_ino = btrfs_ino(parent_inode);
6311 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6312 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6315 key.type = BTRFS_INODE_ITEM_KEY;
6319 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6320 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6321 key.objectid, root->root_key.objectid,
6322 parent_ino, index, name, name_len);
6323 } else if (add_backref) {
6324 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6328 /* Nothing to clean up yet */
6332 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6334 btrfs_inode_type(inode), index);
6335 if (ret == -EEXIST || ret == -EOVERFLOW)
6338 btrfs_abort_transaction(trans, root, ret);
6342 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6344 inode_inc_iversion(parent_inode);
6345 parent_inode->i_mtime = parent_inode->i_ctime =
6346 current_fs_time(parent_inode->i_sb);
6347 ret = btrfs_update_inode(trans, root, parent_inode);
6349 btrfs_abort_transaction(trans, root, ret);
6353 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6356 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6357 key.objectid, root->root_key.objectid,
6358 parent_ino, &local_index, name, name_len);
6360 } else if (add_backref) {
6364 err = btrfs_del_inode_ref(trans, root, name, name_len,
6365 ino, parent_ino, &local_index);
6370 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6371 struct inode *dir, struct dentry *dentry,
6372 struct inode *inode, int backref, u64 index)
6374 int err = btrfs_add_link(trans, dir, inode,
6375 dentry->d_name.name, dentry->d_name.len,
6382 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6383 umode_t mode, dev_t rdev)
6385 struct btrfs_trans_handle *trans;
6386 struct btrfs_root *root = BTRFS_I(dir)->root;
6387 struct inode *inode = NULL;
6394 * 2 for inode item and ref
6396 * 1 for xattr if selinux is on
6398 trans = btrfs_start_transaction(root, 5);
6400 return PTR_ERR(trans);
6402 err = btrfs_find_free_ino(root, &objectid);
6406 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6407 dentry->d_name.len, btrfs_ino(dir), objectid,
6409 if (IS_ERR(inode)) {
6410 err = PTR_ERR(inode);
6415 * If the active LSM wants to access the inode during
6416 * d_instantiate it needs these. Smack checks to see
6417 * if the filesystem supports xattrs by looking at the
6420 inode->i_op = &btrfs_special_inode_operations;
6421 init_special_inode(inode, inode->i_mode, rdev);
6423 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6425 goto out_unlock_inode;
6427 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6429 goto out_unlock_inode;
6431 btrfs_update_inode(trans, root, inode);
6432 unlock_new_inode(inode);
6433 d_instantiate(dentry, inode);
6437 btrfs_end_transaction(trans, root);
6438 btrfs_balance_delayed_items(root);
6439 btrfs_btree_balance_dirty(root);
6441 inode_dec_link_count(inode);
6448 unlock_new_inode(inode);
6453 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6454 umode_t mode, bool excl)
6456 struct btrfs_trans_handle *trans;
6457 struct btrfs_root *root = BTRFS_I(dir)->root;
6458 struct inode *inode = NULL;
6459 int drop_inode_on_err = 0;
6465 * 2 for inode item and ref
6467 * 1 for xattr if selinux is on
6469 trans = btrfs_start_transaction(root, 5);
6471 return PTR_ERR(trans);
6473 err = btrfs_find_free_ino(root, &objectid);
6477 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6478 dentry->d_name.len, btrfs_ino(dir), objectid,
6480 if (IS_ERR(inode)) {
6481 err = PTR_ERR(inode);
6484 drop_inode_on_err = 1;
6486 * If the active LSM wants to access the inode during
6487 * d_instantiate it needs these. Smack checks to see
6488 * if the filesystem supports xattrs by looking at the
6491 inode->i_fop = &btrfs_file_operations;
6492 inode->i_op = &btrfs_file_inode_operations;
6493 inode->i_mapping->a_ops = &btrfs_aops;
6495 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6497 goto out_unlock_inode;
6499 err = btrfs_update_inode(trans, root, inode);
6501 goto out_unlock_inode;
6503 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6505 goto out_unlock_inode;
6507 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6508 unlock_new_inode(inode);
6509 d_instantiate(dentry, inode);
6512 btrfs_end_transaction(trans, root);
6513 if (err && drop_inode_on_err) {
6514 inode_dec_link_count(inode);
6517 btrfs_balance_delayed_items(root);
6518 btrfs_btree_balance_dirty(root);
6522 unlock_new_inode(inode);
6527 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6528 struct dentry *dentry)
6530 struct btrfs_trans_handle *trans = NULL;
6531 struct btrfs_root *root = BTRFS_I(dir)->root;
6532 struct inode *inode = d_inode(old_dentry);
6537 /* do not allow sys_link's with other subvols of the same device */
6538 if (root->objectid != BTRFS_I(inode)->root->objectid)
6541 if (inode->i_nlink >= BTRFS_LINK_MAX)
6544 err = btrfs_set_inode_index(dir, &index);
6549 * 2 items for inode and inode ref
6550 * 2 items for dir items
6551 * 1 item for parent inode
6553 trans = btrfs_start_transaction(root, 5);
6554 if (IS_ERR(trans)) {
6555 err = PTR_ERR(trans);
6560 /* There are several dir indexes for this inode, clear the cache. */
6561 BTRFS_I(inode)->dir_index = 0ULL;
6563 inode_inc_iversion(inode);
6564 inode->i_ctime = current_fs_time(inode->i_sb);
6566 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6568 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6573 struct dentry *parent = dentry->d_parent;
6574 err = btrfs_update_inode(trans, root, inode);
6577 if (inode->i_nlink == 1) {
6579 * If new hard link count is 1, it's a file created
6580 * with open(2) O_TMPFILE flag.
6582 err = btrfs_orphan_del(trans, inode);
6586 d_instantiate(dentry, inode);
6587 btrfs_log_new_name(trans, inode, NULL, parent);
6590 btrfs_balance_delayed_items(root);
6593 btrfs_end_transaction(trans, root);
6595 inode_dec_link_count(inode);
6598 btrfs_btree_balance_dirty(root);
6602 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6604 struct inode *inode = NULL;
6605 struct btrfs_trans_handle *trans;
6606 struct btrfs_root *root = BTRFS_I(dir)->root;
6608 int drop_on_err = 0;
6613 * 2 items for inode and ref
6614 * 2 items for dir items
6615 * 1 for xattr if selinux is on
6617 trans = btrfs_start_transaction(root, 5);
6619 return PTR_ERR(trans);
6621 err = btrfs_find_free_ino(root, &objectid);
6625 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6626 dentry->d_name.len, btrfs_ino(dir), objectid,
6627 S_IFDIR | mode, &index);
6628 if (IS_ERR(inode)) {
6629 err = PTR_ERR(inode);
6634 /* these must be set before we unlock the inode */
6635 inode->i_op = &btrfs_dir_inode_operations;
6636 inode->i_fop = &btrfs_dir_file_operations;
6638 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6640 goto out_fail_inode;
6642 btrfs_i_size_write(inode, 0);
6643 err = btrfs_update_inode(trans, root, inode);
6645 goto out_fail_inode;
6647 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6648 dentry->d_name.len, 0, index);
6650 goto out_fail_inode;
6652 d_instantiate(dentry, inode);
6654 * mkdir is special. We're unlocking after we call d_instantiate
6655 * to avoid a race with nfsd calling d_instantiate.
6657 unlock_new_inode(inode);
6661 btrfs_end_transaction(trans, root);
6663 inode_dec_link_count(inode);
6666 btrfs_balance_delayed_items(root);
6667 btrfs_btree_balance_dirty(root);
6671 unlock_new_inode(inode);
6675 /* Find next extent map of a given extent map, caller needs to ensure locks */
6676 static struct extent_map *next_extent_map(struct extent_map *em)
6678 struct rb_node *next;
6680 next = rb_next(&em->rb_node);
6683 return container_of(next, struct extent_map, rb_node);
6686 static struct extent_map *prev_extent_map(struct extent_map *em)
6688 struct rb_node *prev;
6690 prev = rb_prev(&em->rb_node);
6693 return container_of(prev, struct extent_map, rb_node);
6696 /* helper for btfs_get_extent. Given an existing extent in the tree,
6697 * the existing extent is the nearest extent to map_start,
6698 * and an extent that you want to insert, deal with overlap and insert
6699 * the best fitted new extent into the tree.
6701 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6702 struct extent_map *existing,
6703 struct extent_map *em,
6706 struct extent_map *prev;
6707 struct extent_map *next;
6712 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6714 if (existing->start > map_start) {
6716 prev = prev_extent_map(next);
6719 next = next_extent_map(prev);
6722 start = prev ? extent_map_end(prev) : em->start;
6723 start = max_t(u64, start, em->start);
6724 end = next ? next->start : extent_map_end(em);
6725 end = min_t(u64, end, extent_map_end(em));
6726 start_diff = start - em->start;
6728 em->len = end - start;
6729 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6730 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6731 em->block_start += start_diff;
6732 em->block_len -= start_diff;
6734 return add_extent_mapping(em_tree, em, 0);
6737 static noinline int uncompress_inline(struct btrfs_path *path,
6739 size_t pg_offset, u64 extent_offset,
6740 struct btrfs_file_extent_item *item)
6743 struct extent_buffer *leaf = path->nodes[0];
6746 unsigned long inline_size;
6750 WARN_ON(pg_offset != 0);
6751 compress_type = btrfs_file_extent_compression(leaf, item);
6752 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6753 inline_size = btrfs_file_extent_inline_item_len(leaf,
6754 btrfs_item_nr(path->slots[0]));
6755 tmp = kmalloc(inline_size, GFP_NOFS);
6758 ptr = btrfs_file_extent_inline_start(item);
6760 read_extent_buffer(leaf, tmp, ptr, inline_size);
6762 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6763 ret = btrfs_decompress(compress_type, tmp, page,
6764 extent_offset, inline_size, max_size);
6770 * a bit scary, this does extent mapping from logical file offset to the disk.
6771 * the ugly parts come from merging extents from the disk with the in-ram
6772 * representation. This gets more complex because of the data=ordered code,
6773 * where the in-ram extents might be locked pending data=ordered completion.
6775 * This also copies inline extents directly into the page.
6778 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6779 size_t pg_offset, u64 start, u64 len,
6784 u64 extent_start = 0;
6786 u64 objectid = btrfs_ino(inode);
6788 struct btrfs_path *path = NULL;
6789 struct btrfs_root *root = BTRFS_I(inode)->root;
6790 struct btrfs_file_extent_item *item;
6791 struct extent_buffer *leaf;
6792 struct btrfs_key found_key;
6793 struct extent_map *em = NULL;
6794 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6795 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6796 struct btrfs_trans_handle *trans = NULL;
6797 const bool new_inline = !page || create;
6800 read_lock(&em_tree->lock);
6801 em = lookup_extent_mapping(em_tree, start, len);
6803 em->bdev = root->fs_info->fs_devices->latest_bdev;
6804 read_unlock(&em_tree->lock);
6807 if (em->start > start || em->start + em->len <= start)
6808 free_extent_map(em);
6809 else if (em->block_start == EXTENT_MAP_INLINE && page)
6810 free_extent_map(em);
6814 em = alloc_extent_map();
6819 em->bdev = root->fs_info->fs_devices->latest_bdev;
6820 em->start = EXTENT_MAP_HOLE;
6821 em->orig_start = EXTENT_MAP_HOLE;
6823 em->block_len = (u64)-1;
6826 path = btrfs_alloc_path();
6832 * Chances are we'll be called again, so go ahead and do
6835 path->reada = READA_FORWARD;
6838 ret = btrfs_lookup_file_extent(trans, root, path,
6839 objectid, start, trans != NULL);
6846 if (path->slots[0] == 0)
6851 leaf = path->nodes[0];
6852 item = btrfs_item_ptr(leaf, path->slots[0],
6853 struct btrfs_file_extent_item);
6854 /* are we inside the extent that was found? */
6855 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6856 found_type = found_key.type;
6857 if (found_key.objectid != objectid ||
6858 found_type != BTRFS_EXTENT_DATA_KEY) {
6860 * If we backup past the first extent we want to move forward
6861 * and see if there is an extent in front of us, otherwise we'll
6862 * say there is a hole for our whole search range which can
6869 found_type = btrfs_file_extent_type(leaf, item);
6870 extent_start = found_key.offset;
6871 if (found_type == BTRFS_FILE_EXTENT_REG ||
6872 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6873 extent_end = extent_start +
6874 btrfs_file_extent_num_bytes(leaf, item);
6875 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6877 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6878 extent_end = ALIGN(extent_start + size, root->sectorsize);
6881 if (start >= extent_end) {
6883 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6884 ret = btrfs_next_leaf(root, path);
6891 leaf = path->nodes[0];
6893 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6894 if (found_key.objectid != objectid ||
6895 found_key.type != BTRFS_EXTENT_DATA_KEY)
6897 if (start + len <= found_key.offset)
6899 if (start > found_key.offset)
6902 em->orig_start = start;
6903 em->len = found_key.offset - start;
6907 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6909 if (found_type == BTRFS_FILE_EXTENT_REG ||
6910 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6912 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6916 size_t extent_offset;
6922 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6923 extent_offset = page_offset(page) + pg_offset - extent_start;
6924 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6925 size - extent_offset);
6926 em->start = extent_start + extent_offset;
6927 em->len = ALIGN(copy_size, root->sectorsize);
6928 em->orig_block_len = em->len;
6929 em->orig_start = em->start;
6930 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6931 if (create == 0 && !PageUptodate(page)) {
6932 if (btrfs_file_extent_compression(leaf, item) !=
6933 BTRFS_COMPRESS_NONE) {
6934 ret = uncompress_inline(path, page, pg_offset,
6935 extent_offset, item);
6942 read_extent_buffer(leaf, map + pg_offset, ptr,
6944 if (pg_offset + copy_size < PAGE_SIZE) {
6945 memset(map + pg_offset + copy_size, 0,
6946 PAGE_SIZE - pg_offset -
6951 flush_dcache_page(page);
6952 } else if (create && PageUptodate(page)) {
6956 free_extent_map(em);
6959 btrfs_release_path(path);
6960 trans = btrfs_join_transaction(root);
6963 return ERR_CAST(trans);
6967 write_extent_buffer(leaf, map + pg_offset, ptr,
6970 btrfs_mark_buffer_dirty(leaf);
6972 set_extent_uptodate(io_tree, em->start,
6973 extent_map_end(em) - 1, NULL, GFP_NOFS);
6978 em->orig_start = start;
6981 em->block_start = EXTENT_MAP_HOLE;
6982 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6984 btrfs_release_path(path);
6985 if (em->start > start || extent_map_end(em) <= start) {
6986 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6987 em->start, em->len, start, len);
6993 write_lock(&em_tree->lock);
6994 ret = add_extent_mapping(em_tree, em, 0);
6995 /* it is possible that someone inserted the extent into the tree
6996 * while we had the lock dropped. It is also possible that
6997 * an overlapping map exists in the tree
6999 if (ret == -EEXIST) {
7000 struct extent_map *existing;
7004 existing = search_extent_mapping(em_tree, start, len);
7006 * existing will always be non-NULL, since there must be
7007 * extent causing the -EEXIST.
7009 if (existing->start == em->start &&
7010 extent_map_end(existing) == extent_map_end(em) &&
7011 em->block_start == existing->block_start) {
7013 * these two extents are the same, it happens
7014 * with inlines especially
7016 free_extent_map(em);
7020 } else if (start >= extent_map_end(existing) ||
7021 start <= existing->start) {
7023 * The existing extent map is the one nearest to
7024 * the [start, start + len) range which overlaps
7026 err = merge_extent_mapping(em_tree, existing,
7028 free_extent_map(existing);
7030 free_extent_map(em);
7034 free_extent_map(em);
7039 write_unlock(&em_tree->lock);
7042 trace_btrfs_get_extent(root, em);
7044 btrfs_free_path(path);
7046 ret = btrfs_end_transaction(trans, root);
7051 free_extent_map(em);
7052 return ERR_PTR(err);
7054 BUG_ON(!em); /* Error is always set */
7058 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7059 size_t pg_offset, u64 start, u64 len,
7062 struct extent_map *em;
7063 struct extent_map *hole_em = NULL;
7064 u64 range_start = start;
7070 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7077 * - a pre-alloc extent,
7078 * there might actually be delalloc bytes behind it.
7080 if (em->block_start != EXTENT_MAP_HOLE &&
7081 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7087 /* check to see if we've wrapped (len == -1 or similar) */
7096 /* ok, we didn't find anything, lets look for delalloc */
7097 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7098 end, len, EXTENT_DELALLOC, 1);
7099 found_end = range_start + found;
7100 if (found_end < range_start)
7101 found_end = (u64)-1;
7104 * we didn't find anything useful, return
7105 * the original results from get_extent()
7107 if (range_start > end || found_end <= start) {
7113 /* adjust the range_start to make sure it doesn't
7114 * go backwards from the start they passed in
7116 range_start = max(start, range_start);
7117 found = found_end - range_start;
7120 u64 hole_start = start;
7123 em = alloc_extent_map();
7129 * when btrfs_get_extent can't find anything it
7130 * returns one huge hole
7132 * make sure what it found really fits our range, and
7133 * adjust to make sure it is based on the start from
7137 u64 calc_end = extent_map_end(hole_em);
7139 if (calc_end <= start || (hole_em->start > end)) {
7140 free_extent_map(hole_em);
7143 hole_start = max(hole_em->start, start);
7144 hole_len = calc_end - hole_start;
7148 if (hole_em && range_start > hole_start) {
7149 /* our hole starts before our delalloc, so we
7150 * have to return just the parts of the hole
7151 * that go until the delalloc starts
7153 em->len = min(hole_len,
7154 range_start - hole_start);
7155 em->start = hole_start;
7156 em->orig_start = hole_start;
7158 * don't adjust block start at all,
7159 * it is fixed at EXTENT_MAP_HOLE
7161 em->block_start = hole_em->block_start;
7162 em->block_len = hole_len;
7163 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7164 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7166 em->start = range_start;
7168 em->orig_start = range_start;
7169 em->block_start = EXTENT_MAP_DELALLOC;
7170 em->block_len = found;
7172 } else if (hole_em) {
7177 free_extent_map(hole_em);
7179 free_extent_map(em);
7180 return ERR_PTR(err);
7185 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7188 const u64 orig_start,
7189 const u64 block_start,
7190 const u64 block_len,
7191 const u64 orig_block_len,
7192 const u64 ram_bytes,
7195 struct extent_map *em = NULL;
7198 down_read(&BTRFS_I(inode)->dio_sem);
7199 if (type != BTRFS_ORDERED_NOCOW) {
7200 em = create_pinned_em(inode, start, len, orig_start,
7201 block_start, block_len, orig_block_len,
7206 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7207 len, block_len, type);
7210 free_extent_map(em);
7211 btrfs_drop_extent_cache(inode, start,
7212 start + len - 1, 0);
7217 up_read(&BTRFS_I(inode)->dio_sem);
7222 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7225 struct btrfs_root *root = BTRFS_I(inode)->root;
7226 struct extent_map *em;
7227 struct btrfs_key ins;
7231 alloc_hint = get_extent_allocation_hint(inode, start, len);
7232 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7233 alloc_hint, &ins, 1, 1);
7235 return ERR_PTR(ret);
7237 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7238 ins.objectid, ins.offset, ins.offset,
7240 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7242 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7248 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7249 * block must be cow'd
7251 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7252 u64 *orig_start, u64 *orig_block_len,
7255 struct btrfs_trans_handle *trans;
7256 struct btrfs_path *path;
7258 struct extent_buffer *leaf;
7259 struct btrfs_root *root = BTRFS_I(inode)->root;
7260 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7261 struct btrfs_file_extent_item *fi;
7262 struct btrfs_key key;
7269 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7271 path = btrfs_alloc_path();
7275 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7280 slot = path->slots[0];
7283 /* can't find the item, must cow */
7290 leaf = path->nodes[0];
7291 btrfs_item_key_to_cpu(leaf, &key, slot);
7292 if (key.objectid != btrfs_ino(inode) ||
7293 key.type != BTRFS_EXTENT_DATA_KEY) {
7294 /* not our file or wrong item type, must cow */
7298 if (key.offset > offset) {
7299 /* Wrong offset, must cow */
7303 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7304 found_type = btrfs_file_extent_type(leaf, fi);
7305 if (found_type != BTRFS_FILE_EXTENT_REG &&
7306 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7307 /* not a regular extent, must cow */
7311 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7314 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7315 if (extent_end <= offset)
7318 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7319 if (disk_bytenr == 0)
7322 if (btrfs_file_extent_compression(leaf, fi) ||
7323 btrfs_file_extent_encryption(leaf, fi) ||
7324 btrfs_file_extent_other_encoding(leaf, fi))
7327 backref_offset = btrfs_file_extent_offset(leaf, fi);
7330 *orig_start = key.offset - backref_offset;
7331 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7332 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7335 if (btrfs_extent_readonly(root, disk_bytenr))
7338 num_bytes = min(offset + *len, extent_end) - offset;
7339 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7342 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7343 ret = test_range_bit(io_tree, offset, range_end,
7344 EXTENT_DELALLOC, 0, NULL);
7351 btrfs_release_path(path);
7354 * look for other files referencing this extent, if we
7355 * find any we must cow
7357 trans = btrfs_join_transaction(root);
7358 if (IS_ERR(trans)) {
7363 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7364 key.offset - backref_offset, disk_bytenr);
7365 btrfs_end_transaction(trans, root);
7372 * adjust disk_bytenr and num_bytes to cover just the bytes
7373 * in this extent we are about to write. If there
7374 * are any csums in that range we have to cow in order
7375 * to keep the csums correct
7377 disk_bytenr += backref_offset;
7378 disk_bytenr += offset - key.offset;
7379 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7382 * all of the above have passed, it is safe to overwrite this extent
7388 btrfs_free_path(path);
7392 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7394 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7396 void **pagep = NULL;
7397 struct page *page = NULL;
7401 start_idx = start >> PAGE_SHIFT;
7404 * end is the last byte in the last page. end == start is legal
7406 end_idx = end >> PAGE_SHIFT;
7410 /* Most of the code in this while loop is lifted from
7411 * find_get_page. It's been modified to begin searching from a
7412 * page and return just the first page found in that range. If the
7413 * found idx is less than or equal to the end idx then we know that
7414 * a page exists. If no pages are found or if those pages are
7415 * outside of the range then we're fine (yay!) */
7416 while (page == NULL &&
7417 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7418 page = radix_tree_deref_slot(pagep);
7419 if (unlikely(!page))
7422 if (radix_tree_exception(page)) {
7423 if (radix_tree_deref_retry(page)) {
7428 * Otherwise, shmem/tmpfs must be storing a swap entry
7429 * here as an exceptional entry: so return it without
7430 * attempting to raise page count.
7433 break; /* TODO: Is this relevant for this use case? */
7436 if (!page_cache_get_speculative(page)) {
7442 * Has the page moved?
7443 * This is part of the lockless pagecache protocol. See
7444 * include/linux/pagemap.h for details.
7446 if (unlikely(page != *pagep)) {
7453 if (page->index <= end_idx)
7462 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7463 struct extent_state **cached_state, int writing)
7465 struct btrfs_ordered_extent *ordered;
7469 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7472 * We're concerned with the entire range that we're going to be
7473 * doing DIO to, so we need to make sure there's no ordered
7474 * extents in this range.
7476 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7477 lockend - lockstart + 1);
7480 * We need to make sure there are no buffered pages in this
7481 * range either, we could have raced between the invalidate in
7482 * generic_file_direct_write and locking the extent. The
7483 * invalidate needs to happen so that reads after a write do not
7488 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7491 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7492 cached_state, GFP_NOFS);
7496 * If we are doing a DIO read and the ordered extent we
7497 * found is for a buffered write, we can not wait for it
7498 * to complete and retry, because if we do so we can
7499 * deadlock with concurrent buffered writes on page
7500 * locks. This happens only if our DIO read covers more
7501 * than one extent map, if at this point has already
7502 * created an ordered extent for a previous extent map
7503 * and locked its range in the inode's io tree, and a
7504 * concurrent write against that previous extent map's
7505 * range and this range started (we unlock the ranges
7506 * in the io tree only when the bios complete and
7507 * buffered writes always lock pages before attempting
7508 * to lock range in the io tree).
7511 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7512 btrfs_start_ordered_extent(inode, ordered, 1);
7515 btrfs_put_ordered_extent(ordered);
7518 * We could trigger writeback for this range (and wait
7519 * for it to complete) and then invalidate the pages for
7520 * this range (through invalidate_inode_pages2_range()),
7521 * but that can lead us to a deadlock with a concurrent
7522 * call to readpages() (a buffered read or a defrag call
7523 * triggered a readahead) on a page lock due to an
7524 * ordered dio extent we created before but did not have
7525 * yet a corresponding bio submitted (whence it can not
7526 * complete), which makes readpages() wait for that
7527 * ordered extent to complete while holding a lock on
7542 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7543 u64 len, u64 orig_start,
7544 u64 block_start, u64 block_len,
7545 u64 orig_block_len, u64 ram_bytes,
7548 struct extent_map_tree *em_tree;
7549 struct extent_map *em;
7550 struct btrfs_root *root = BTRFS_I(inode)->root;
7553 em_tree = &BTRFS_I(inode)->extent_tree;
7554 em = alloc_extent_map();
7556 return ERR_PTR(-ENOMEM);
7559 em->orig_start = orig_start;
7560 em->mod_start = start;
7563 em->block_len = block_len;
7564 em->block_start = block_start;
7565 em->bdev = root->fs_info->fs_devices->latest_bdev;
7566 em->orig_block_len = orig_block_len;
7567 em->ram_bytes = ram_bytes;
7568 em->generation = -1;
7569 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7570 if (type == BTRFS_ORDERED_PREALLOC)
7571 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7574 btrfs_drop_extent_cache(inode, em->start,
7575 em->start + em->len - 1, 0);
7576 write_lock(&em_tree->lock);
7577 ret = add_extent_mapping(em_tree, em, 1);
7578 write_unlock(&em_tree->lock);
7579 } while (ret == -EEXIST);
7582 free_extent_map(em);
7583 return ERR_PTR(ret);
7589 static void adjust_dio_outstanding_extents(struct inode *inode,
7590 struct btrfs_dio_data *dio_data,
7593 unsigned num_extents;
7595 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7596 BTRFS_MAX_EXTENT_SIZE);
7598 * If we have an outstanding_extents count still set then we're
7599 * within our reservation, otherwise we need to adjust our inode
7600 * counter appropriately.
7602 if (dio_data->outstanding_extents) {
7603 dio_data->outstanding_extents -= num_extents;
7605 spin_lock(&BTRFS_I(inode)->lock);
7606 BTRFS_I(inode)->outstanding_extents += num_extents;
7607 spin_unlock(&BTRFS_I(inode)->lock);
7611 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7612 struct buffer_head *bh_result, int create)
7614 struct extent_map *em;
7615 struct btrfs_root *root = BTRFS_I(inode)->root;
7616 struct extent_state *cached_state = NULL;
7617 struct btrfs_dio_data *dio_data = NULL;
7618 u64 start = iblock << inode->i_blkbits;
7619 u64 lockstart, lockend;
7620 u64 len = bh_result->b_size;
7621 int unlock_bits = EXTENT_LOCKED;
7625 unlock_bits |= EXTENT_DIRTY;
7627 len = min_t(u64, len, root->sectorsize);
7630 lockend = start + len - 1;
7632 if (current->journal_info) {
7634 * Need to pull our outstanding extents and set journal_info to NULL so
7635 * that anything that needs to check if there's a transaction doesn't get
7638 dio_data = current->journal_info;
7639 current->journal_info = NULL;
7643 * If this errors out it's because we couldn't invalidate pagecache for
7644 * this range and we need to fallback to buffered.
7646 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7652 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7659 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7660 * io. INLINE is special, and we could probably kludge it in here, but
7661 * it's still buffered so for safety lets just fall back to the generic
7664 * For COMPRESSED we _have_ to read the entire extent in so we can
7665 * decompress it, so there will be buffering required no matter what we
7666 * do, so go ahead and fallback to buffered.
7668 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7669 * to buffered IO. Don't blame me, this is the price we pay for using
7672 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7673 em->block_start == EXTENT_MAP_INLINE) {
7674 free_extent_map(em);
7679 /* Just a good old fashioned hole, return */
7680 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7681 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7682 free_extent_map(em);
7687 * We don't allocate a new extent in the following cases
7689 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7691 * 2) The extent is marked as PREALLOC. We're good to go here and can
7692 * just use the extent.
7696 len = min(len, em->len - (start - em->start));
7697 lockstart = start + len;
7701 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7702 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7703 em->block_start != EXTENT_MAP_HOLE)) {
7705 u64 block_start, orig_start, orig_block_len, ram_bytes;
7707 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7708 type = BTRFS_ORDERED_PREALLOC;
7710 type = BTRFS_ORDERED_NOCOW;
7711 len = min(len, em->len - (start - em->start));
7712 block_start = em->block_start + (start - em->start);
7714 if (can_nocow_extent(inode, start, &len, &orig_start,
7715 &orig_block_len, &ram_bytes) == 1 &&
7716 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7717 struct extent_map *em2;
7719 em2 = btrfs_create_dio_extent(inode, start, len,
7720 orig_start, block_start,
7721 len, orig_block_len,
7723 btrfs_dec_nocow_writers(root->fs_info, block_start);
7724 if (type == BTRFS_ORDERED_PREALLOC) {
7725 free_extent_map(em);
7728 if (em2 && IS_ERR(em2)) {
7737 * this will cow the extent, reset the len in case we changed
7740 len = bh_result->b_size;
7741 free_extent_map(em);
7742 em = btrfs_new_extent_direct(inode, start, len);
7747 len = min(len, em->len - (start - em->start));
7749 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7751 bh_result->b_size = len;
7752 bh_result->b_bdev = em->bdev;
7753 set_buffer_mapped(bh_result);
7755 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7756 set_buffer_new(bh_result);
7759 * Need to update the i_size under the extent lock so buffered
7760 * readers will get the updated i_size when we unlock.
7762 if (start + len > i_size_read(inode))
7763 i_size_write(inode, start + len);
7765 adjust_dio_outstanding_extents(inode, dio_data, len);
7766 btrfs_free_reserved_data_space(inode, start, len);
7767 WARN_ON(dio_data->reserve < len);
7768 dio_data->reserve -= len;
7769 dio_data->unsubmitted_oe_range_end = start + len;
7770 current->journal_info = dio_data;
7774 * In the case of write we need to clear and unlock the entire range,
7775 * in the case of read we need to unlock only the end area that we
7776 * aren't using if there is any left over space.
7778 if (lockstart < lockend) {
7779 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7780 lockend, unlock_bits, 1, 0,
7781 &cached_state, GFP_NOFS);
7783 free_extent_state(cached_state);
7786 free_extent_map(em);
7791 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7792 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7795 current->journal_info = dio_data;
7797 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7798 * write less data then expected, so that we don't underflow our inode's
7799 * outstanding extents counter.
7801 if (create && dio_data)
7802 adjust_dio_outstanding_extents(inode, dio_data, len);
7807 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7808 int rw, int mirror_num)
7810 struct btrfs_root *root = BTRFS_I(inode)->root;
7813 BUG_ON(rw & REQ_WRITE);
7817 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7818 BTRFS_WQ_ENDIO_DIO_REPAIR);
7822 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7828 static int btrfs_check_dio_repairable(struct inode *inode,
7829 struct bio *failed_bio,
7830 struct io_failure_record *failrec,
7835 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7836 failrec->logical, failrec->len);
7837 if (num_copies == 1) {
7839 * we only have a single copy of the data, so don't bother with
7840 * all the retry and error correction code that follows. no
7841 * matter what the error is, it is very likely to persist.
7843 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7844 num_copies, failrec->this_mirror, failed_mirror);
7848 failrec->failed_mirror = failed_mirror;
7849 failrec->this_mirror++;
7850 if (failrec->this_mirror == failed_mirror)
7851 failrec->this_mirror++;
7853 if (failrec->this_mirror > num_copies) {
7854 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7855 num_copies, failrec->this_mirror, failed_mirror);
7862 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7863 struct page *page, unsigned int pgoff,
7864 u64 start, u64 end, int failed_mirror,
7865 bio_end_io_t *repair_endio, void *repair_arg)
7867 struct io_failure_record *failrec;
7873 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7875 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7879 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7882 free_io_failure(inode, failrec);
7886 if ((failed_bio->bi_vcnt > 1)
7887 || (failed_bio->bi_io_vec->bv_len
7888 > BTRFS_I(inode)->root->sectorsize))
7889 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7891 read_mode = READ_SYNC;
7893 isector = start - btrfs_io_bio(failed_bio)->logical;
7894 isector >>= inode->i_sb->s_blocksize_bits;
7895 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7896 pgoff, isector, repair_endio, repair_arg);
7898 free_io_failure(inode, failrec);
7902 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7903 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7904 read_mode, failrec->this_mirror, failrec->in_validation);
7906 ret = submit_dio_repair_bio(inode, bio, read_mode,
7907 failrec->this_mirror);
7909 free_io_failure(inode, failrec);
7916 struct btrfs_retry_complete {
7917 struct completion done;
7918 struct inode *inode;
7923 static void btrfs_retry_endio_nocsum(struct bio *bio)
7925 struct btrfs_retry_complete *done = bio->bi_private;
7926 struct inode *inode;
7927 struct bio_vec *bvec;
7933 ASSERT(bio->bi_vcnt == 1);
7934 inode = bio->bi_io_vec->bv_page->mapping->host;
7935 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7938 bio_for_each_segment_all(bvec, bio, i)
7939 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7941 complete(&done->done);
7945 static int __btrfs_correct_data_nocsum(struct inode *inode,
7946 struct btrfs_io_bio *io_bio)
7948 struct btrfs_fs_info *fs_info;
7949 struct bio_vec *bvec;
7950 struct btrfs_retry_complete done;
7958 fs_info = BTRFS_I(inode)->root->fs_info;
7959 sectorsize = BTRFS_I(inode)->root->sectorsize;
7961 start = io_bio->logical;
7964 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7965 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7966 pgoff = bvec->bv_offset;
7968 next_block_or_try_again:
7971 init_completion(&done.done);
7973 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7974 pgoff, start, start + sectorsize - 1,
7976 btrfs_retry_endio_nocsum, &done);
7980 wait_for_completion(&done.done);
7982 if (!done.uptodate) {
7983 /* We might have another mirror, so try again */
7984 goto next_block_or_try_again;
7987 start += sectorsize;
7990 pgoff += sectorsize;
7991 goto next_block_or_try_again;
7998 static void btrfs_retry_endio(struct bio *bio)
8000 struct btrfs_retry_complete *done = bio->bi_private;
8001 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8002 struct inode *inode;
8003 struct bio_vec *bvec;
8014 start = done->start;
8016 ASSERT(bio->bi_vcnt == 1);
8017 inode = bio->bi_io_vec->bv_page->mapping->host;
8018 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
8020 bio_for_each_segment_all(bvec, bio, i) {
8021 ret = __readpage_endio_check(done->inode, io_bio, i,
8022 bvec->bv_page, bvec->bv_offset,
8023 done->start, bvec->bv_len);
8025 clean_io_failure(done->inode, done->start,
8026 bvec->bv_page, bvec->bv_offset);
8031 done->uptodate = uptodate;
8033 complete(&done->done);
8037 static int __btrfs_subio_endio_read(struct inode *inode,
8038 struct btrfs_io_bio *io_bio, int err)
8040 struct btrfs_fs_info *fs_info;
8041 struct bio_vec *bvec;
8042 struct btrfs_retry_complete done;
8052 fs_info = BTRFS_I(inode)->root->fs_info;
8053 sectorsize = BTRFS_I(inode)->root->sectorsize;
8056 start = io_bio->logical;
8059 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8060 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8062 pgoff = bvec->bv_offset;
8064 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8065 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8066 bvec->bv_page, pgoff, start,
8073 init_completion(&done.done);
8075 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8076 pgoff, start, start + sectorsize - 1,
8078 btrfs_retry_endio, &done);
8084 wait_for_completion(&done.done);
8086 if (!done.uptodate) {
8087 /* We might have another mirror, so try again */
8091 offset += sectorsize;
8092 start += sectorsize;
8097 pgoff += sectorsize;
8105 static int btrfs_subio_endio_read(struct inode *inode,
8106 struct btrfs_io_bio *io_bio, int err)
8108 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8112 return __btrfs_correct_data_nocsum(inode, io_bio);
8116 return __btrfs_subio_endio_read(inode, io_bio, err);
8120 static void btrfs_endio_direct_read(struct bio *bio)
8122 struct btrfs_dio_private *dip = bio->bi_private;
8123 struct inode *inode = dip->inode;
8124 struct bio *dio_bio;
8125 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8126 int err = bio->bi_error;
8128 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8129 err = btrfs_subio_endio_read(inode, io_bio, err);
8131 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8132 dip->logical_offset + dip->bytes - 1);
8133 dio_bio = dip->dio_bio;
8137 dio_bio->bi_error = bio->bi_error;
8138 dio_end_io(dio_bio, bio->bi_error);
8141 io_bio->end_io(io_bio, err);
8145 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8150 struct btrfs_root *root = BTRFS_I(inode)->root;
8151 struct btrfs_ordered_extent *ordered = NULL;
8152 u64 ordered_offset = offset;
8153 u64 ordered_bytes = bytes;
8157 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8164 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8165 finish_ordered_fn, NULL, NULL);
8166 btrfs_queue_work(root->fs_info->endio_write_workers,
8170 * our bio might span multiple ordered extents. If we haven't
8171 * completed the accounting for the whole dio, go back and try again
8173 if (ordered_offset < offset + bytes) {
8174 ordered_bytes = offset + bytes - ordered_offset;
8180 static void btrfs_endio_direct_write(struct bio *bio)
8182 struct btrfs_dio_private *dip = bio->bi_private;
8183 struct bio *dio_bio = dip->dio_bio;
8185 btrfs_endio_direct_write_update_ordered(dip->inode,
8186 dip->logical_offset,
8192 dio_bio->bi_error = bio->bi_error;
8193 dio_end_io(dio_bio, bio->bi_error);
8197 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8198 struct bio *bio, int mirror_num,
8199 unsigned long bio_flags, u64 offset)
8202 struct btrfs_root *root = BTRFS_I(inode)->root;
8203 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8204 BUG_ON(ret); /* -ENOMEM */
8208 static void btrfs_end_dio_bio(struct bio *bio)
8210 struct btrfs_dio_private *dip = bio->bi_private;
8211 int err = bio->bi_error;
8214 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8215 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8216 btrfs_ino(dip->inode), bio->bi_rw,
8217 (unsigned long long)bio->bi_iter.bi_sector,
8218 bio->bi_iter.bi_size, err);
8220 if (dip->subio_endio)
8221 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8227 * before atomic variable goto zero, we must make sure
8228 * dip->errors is perceived to be set.
8230 smp_mb__before_atomic();
8233 /* if there are more bios still pending for this dio, just exit */
8234 if (!atomic_dec_and_test(&dip->pending_bios))
8238 bio_io_error(dip->orig_bio);
8240 dip->dio_bio->bi_error = 0;
8241 bio_endio(dip->orig_bio);
8247 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8248 u64 first_sector, gfp_t gfp_flags)
8251 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8253 bio_associate_current(bio);
8257 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8258 struct inode *inode,
8259 struct btrfs_dio_private *dip,
8263 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8264 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8268 * We load all the csum data we need when we submit
8269 * the first bio to reduce the csum tree search and
8272 if (dip->logical_offset == file_offset) {
8273 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8279 if (bio == dip->orig_bio)
8282 file_offset -= dip->logical_offset;
8283 file_offset >>= inode->i_sb->s_blocksize_bits;
8284 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8289 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8290 int rw, u64 file_offset, int skip_sum,
8293 struct btrfs_dio_private *dip = bio->bi_private;
8294 int write = rw & REQ_WRITE;
8295 struct btrfs_root *root = BTRFS_I(inode)->root;
8299 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8304 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8305 BTRFS_WQ_ENDIO_DATA);
8313 if (write && async_submit) {
8314 ret = btrfs_wq_submit_bio(root->fs_info,
8315 inode, rw, bio, 0, 0,
8317 __btrfs_submit_bio_start_direct_io,
8318 __btrfs_submit_bio_done);
8322 * If we aren't doing async submit, calculate the csum of the
8325 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8329 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8335 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8341 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8344 struct inode *inode = dip->inode;
8345 struct btrfs_root *root = BTRFS_I(inode)->root;
8347 struct bio *orig_bio = dip->orig_bio;
8348 struct bio_vec *bvec = orig_bio->bi_io_vec;
8349 u64 start_sector = orig_bio->bi_iter.bi_sector;
8350 u64 file_offset = dip->logical_offset;
8353 u32 blocksize = root->sectorsize;
8354 int async_submit = 0;
8359 map_length = orig_bio->bi_iter.bi_size;
8360 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8361 &map_length, NULL, 0);
8365 if (map_length >= orig_bio->bi_iter.bi_size) {
8367 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8371 /* async crcs make it difficult to collect full stripe writes. */
8372 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8377 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8381 bio->bi_private = dip;
8382 bio->bi_end_io = btrfs_end_dio_bio;
8383 btrfs_io_bio(bio)->logical = file_offset;
8384 atomic_inc(&dip->pending_bios);
8386 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8387 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8390 if (unlikely(map_length < submit_len + blocksize ||
8391 bio_add_page(bio, bvec->bv_page, blocksize,
8392 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8394 * inc the count before we submit the bio so
8395 * we know the end IO handler won't happen before
8396 * we inc the count. Otherwise, the dip might get freed
8397 * before we're done setting it up
8399 atomic_inc(&dip->pending_bios);
8400 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8401 file_offset, skip_sum,
8405 atomic_dec(&dip->pending_bios);
8409 start_sector += submit_len >> 9;
8410 file_offset += submit_len;
8414 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8415 start_sector, GFP_NOFS);
8418 bio->bi_private = dip;
8419 bio->bi_end_io = btrfs_end_dio_bio;
8420 btrfs_io_bio(bio)->logical = file_offset;
8422 map_length = orig_bio->bi_iter.bi_size;
8423 ret = btrfs_map_block(root->fs_info, rw,
8425 &map_length, NULL, 0);
8433 submit_len += blocksize;
8443 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8452 * before atomic variable goto zero, we must
8453 * make sure dip->errors is perceived to be set.
8455 smp_mb__before_atomic();
8456 if (atomic_dec_and_test(&dip->pending_bios))
8457 bio_io_error(dip->orig_bio);
8459 /* bio_end_io() will handle error, so we needn't return it */
8463 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8464 struct inode *inode, loff_t file_offset)
8466 struct btrfs_dio_private *dip = NULL;
8467 struct bio *io_bio = NULL;
8468 struct btrfs_io_bio *btrfs_bio;
8470 int write = rw & REQ_WRITE;
8473 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8475 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8481 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8487 dip->private = dio_bio->bi_private;
8489 dip->logical_offset = file_offset;
8490 dip->bytes = dio_bio->bi_iter.bi_size;
8491 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8492 io_bio->bi_private = dip;
8493 dip->orig_bio = io_bio;
8494 dip->dio_bio = dio_bio;
8495 atomic_set(&dip->pending_bios, 0);
8496 btrfs_bio = btrfs_io_bio(io_bio);
8497 btrfs_bio->logical = file_offset;
8500 io_bio->bi_end_io = btrfs_endio_direct_write;
8502 io_bio->bi_end_io = btrfs_endio_direct_read;
8503 dip->subio_endio = btrfs_subio_endio_read;
8507 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8508 * even if we fail to submit a bio, because in such case we do the
8509 * corresponding error handling below and it must not be done a second
8510 * time by btrfs_direct_IO().
8513 struct btrfs_dio_data *dio_data = current->journal_info;
8515 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8517 dio_data->unsubmitted_oe_range_start =
8518 dio_data->unsubmitted_oe_range_end;
8521 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8525 if (btrfs_bio->end_io)
8526 btrfs_bio->end_io(btrfs_bio, ret);
8530 * If we arrived here it means either we failed to submit the dip
8531 * or we either failed to clone the dio_bio or failed to allocate the
8532 * dip. If we cloned the dio_bio and allocated the dip, we can just
8533 * call bio_endio against our io_bio so that we get proper resource
8534 * cleanup if we fail to submit the dip, otherwise, we must do the
8535 * same as btrfs_endio_direct_[write|read] because we can't call these
8536 * callbacks - they require an allocated dip and a clone of dio_bio.
8538 if (io_bio && dip) {
8539 io_bio->bi_error = -EIO;
8542 * The end io callbacks free our dip, do the final put on io_bio
8543 * and all the cleanup and final put for dio_bio (through
8550 btrfs_endio_direct_write_update_ordered(inode,
8552 dio_bio->bi_iter.bi_size,
8555 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8556 file_offset + dio_bio->bi_iter.bi_size - 1);
8558 dio_bio->bi_error = -EIO;
8560 * Releases and cleans up our dio_bio, no need to bio_put()
8561 * nor bio_endio()/bio_io_error() against dio_bio.
8563 dio_end_io(dio_bio, ret);
8570 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8571 const struct iov_iter *iter, loff_t offset)
8575 unsigned blocksize_mask = root->sectorsize - 1;
8576 ssize_t retval = -EINVAL;
8578 if (offset & blocksize_mask)
8581 if (iov_iter_alignment(iter) & blocksize_mask)
8584 /* If this is a write we don't need to check anymore */
8585 if (iov_iter_rw(iter) == WRITE)
8588 * Check to make sure we don't have duplicate iov_base's in this
8589 * iovec, if so return EINVAL, otherwise we'll get csum errors
8590 * when reading back.
8592 for (seg = 0; seg < iter->nr_segs; seg++) {
8593 for (i = seg + 1; i < iter->nr_segs; i++) {
8594 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8603 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8605 struct file *file = iocb->ki_filp;
8606 struct inode *inode = file->f_mapping->host;
8607 struct btrfs_root *root = BTRFS_I(inode)->root;
8608 struct btrfs_dio_data dio_data = { 0 };
8609 loff_t offset = iocb->ki_pos;
8613 bool relock = false;
8616 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8619 inode_dio_begin(inode);
8620 smp_mb__after_atomic();
8623 * The generic stuff only does filemap_write_and_wait_range, which
8624 * isn't enough if we've written compressed pages to this area, so
8625 * we need to flush the dirty pages again to make absolutely sure
8626 * that any outstanding dirty pages are on disk.
8628 count = iov_iter_count(iter);
8629 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8630 &BTRFS_I(inode)->runtime_flags))
8631 filemap_fdatawrite_range(inode->i_mapping, offset,
8632 offset + count - 1);
8634 if (iov_iter_rw(iter) == WRITE) {
8636 * If the write DIO is beyond the EOF, we need update
8637 * the isize, but it is protected by i_mutex. So we can
8638 * not unlock the i_mutex at this case.
8640 if (offset + count <= inode->i_size) {
8641 inode_unlock(inode);
8644 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8647 dio_data.outstanding_extents = div64_u64(count +
8648 BTRFS_MAX_EXTENT_SIZE - 1,
8649 BTRFS_MAX_EXTENT_SIZE);
8652 * We need to know how many extents we reserved so that we can
8653 * do the accounting properly if we go over the number we
8654 * originally calculated. Abuse current->journal_info for this.
8656 dio_data.reserve = round_up(count, root->sectorsize);
8657 dio_data.unsubmitted_oe_range_start = (u64)offset;
8658 dio_data.unsubmitted_oe_range_end = (u64)offset;
8659 current->journal_info = &dio_data;
8660 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8661 &BTRFS_I(inode)->runtime_flags)) {
8662 inode_dio_end(inode);
8663 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8667 ret = __blockdev_direct_IO(iocb, inode,
8668 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8669 iter, btrfs_get_blocks_direct, NULL,
8670 btrfs_submit_direct, flags);
8671 if (iov_iter_rw(iter) == WRITE) {
8672 current->journal_info = NULL;
8673 if (ret < 0 && ret != -EIOCBQUEUED) {
8674 if (dio_data.reserve)
8675 btrfs_delalloc_release_space(inode, offset,
8678 * On error we might have left some ordered extents
8679 * without submitting corresponding bios for them, so
8680 * cleanup them up to avoid other tasks getting them
8681 * and waiting for them to complete forever.
8683 if (dio_data.unsubmitted_oe_range_start <
8684 dio_data.unsubmitted_oe_range_end)
8685 btrfs_endio_direct_write_update_ordered(inode,
8686 dio_data.unsubmitted_oe_range_start,
8687 dio_data.unsubmitted_oe_range_end -
8688 dio_data.unsubmitted_oe_range_start,
8690 } else if (ret >= 0 && (size_t)ret < count)
8691 btrfs_delalloc_release_space(inode, offset,
8692 count - (size_t)ret);
8696 inode_dio_end(inode);
8703 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8705 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8706 __u64 start, __u64 len)
8710 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8714 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8717 int btrfs_readpage(struct file *file, struct page *page)
8719 struct extent_io_tree *tree;
8720 tree = &BTRFS_I(page->mapping->host)->io_tree;
8721 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8724 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8726 struct extent_io_tree *tree;
8727 struct inode *inode = page->mapping->host;
8730 if (current->flags & PF_MEMALLOC) {
8731 redirty_page_for_writepage(wbc, page);
8737 * If we are under memory pressure we will call this directly from the
8738 * VM, we need to make sure we have the inode referenced for the ordered
8739 * extent. If not just return like we didn't do anything.
8741 if (!igrab(inode)) {
8742 redirty_page_for_writepage(wbc, page);
8743 return AOP_WRITEPAGE_ACTIVATE;
8745 tree = &BTRFS_I(page->mapping->host)->io_tree;
8746 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8747 btrfs_add_delayed_iput(inode);
8751 static int btrfs_writepages(struct address_space *mapping,
8752 struct writeback_control *wbc)
8754 struct extent_io_tree *tree;
8756 tree = &BTRFS_I(mapping->host)->io_tree;
8757 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8761 btrfs_readpages(struct file *file, struct address_space *mapping,
8762 struct list_head *pages, unsigned nr_pages)
8764 struct extent_io_tree *tree;
8765 tree = &BTRFS_I(mapping->host)->io_tree;
8766 return extent_readpages(tree, mapping, pages, nr_pages,
8769 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8771 struct extent_io_tree *tree;
8772 struct extent_map_tree *map;
8775 tree = &BTRFS_I(page->mapping->host)->io_tree;
8776 map = &BTRFS_I(page->mapping->host)->extent_tree;
8777 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8779 ClearPagePrivate(page);
8780 set_page_private(page, 0);
8786 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8788 if (PageWriteback(page) || PageDirty(page))
8790 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8793 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8794 unsigned int length)
8796 struct inode *inode = page->mapping->host;
8797 struct extent_io_tree *tree;
8798 struct btrfs_ordered_extent *ordered;
8799 struct extent_state *cached_state = NULL;
8800 u64 page_start = page_offset(page);
8801 u64 page_end = page_start + PAGE_SIZE - 1;
8804 int inode_evicting = inode->i_state & I_FREEING;
8807 * we have the page locked, so new writeback can't start,
8808 * and the dirty bit won't be cleared while we are here.
8810 * Wait for IO on this page so that we can safely clear
8811 * the PagePrivate2 bit and do ordered accounting
8813 wait_on_page_writeback(page);
8815 tree = &BTRFS_I(inode)->io_tree;
8817 btrfs_releasepage(page, GFP_NOFS);
8821 if (!inode_evicting)
8822 lock_extent_bits(tree, page_start, page_end, &cached_state);
8825 ordered = btrfs_lookup_ordered_range(inode, start,
8826 page_end - start + 1);
8828 end = min(page_end, ordered->file_offset + ordered->len - 1);
8830 * IO on this page will never be started, so we need
8831 * to account for any ordered extents now
8833 if (!inode_evicting)
8834 clear_extent_bit(tree, start, end,
8835 EXTENT_DIRTY | EXTENT_DELALLOC |
8836 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8837 EXTENT_DEFRAG, 1, 0, &cached_state,
8840 * whoever cleared the private bit is responsible
8841 * for the finish_ordered_io
8843 if (TestClearPagePrivate2(page)) {
8844 struct btrfs_ordered_inode_tree *tree;
8847 tree = &BTRFS_I(inode)->ordered_tree;
8849 spin_lock_irq(&tree->lock);
8850 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8851 new_len = start - ordered->file_offset;
8852 if (new_len < ordered->truncated_len)
8853 ordered->truncated_len = new_len;
8854 spin_unlock_irq(&tree->lock);
8856 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8858 end - start + 1, 1))
8859 btrfs_finish_ordered_io(ordered);
8861 btrfs_put_ordered_extent(ordered);
8862 if (!inode_evicting) {
8863 cached_state = NULL;
8864 lock_extent_bits(tree, start, end,
8869 if (start < page_end)
8874 * Qgroup reserved space handler
8875 * Page here will be either
8876 * 1) Already written to disk
8877 * In this case, its reserved space is released from data rsv map
8878 * and will be freed by delayed_ref handler finally.
8879 * So even we call qgroup_free_data(), it won't decrease reserved
8881 * 2) Not written to disk
8882 * This means the reserved space should be freed here.
8884 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8885 if (!inode_evicting) {
8886 clear_extent_bit(tree, page_start, page_end,
8887 EXTENT_LOCKED | EXTENT_DIRTY |
8888 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8889 EXTENT_DEFRAG, 1, 1,
8890 &cached_state, GFP_NOFS);
8892 __btrfs_releasepage(page, GFP_NOFS);
8895 ClearPageChecked(page);
8896 if (PagePrivate(page)) {
8897 ClearPagePrivate(page);
8898 set_page_private(page, 0);
8904 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8905 * called from a page fault handler when a page is first dirtied. Hence we must
8906 * be careful to check for EOF conditions here. We set the page up correctly
8907 * for a written page which means we get ENOSPC checking when writing into
8908 * holes and correct delalloc and unwritten extent mapping on filesystems that
8909 * support these features.
8911 * We are not allowed to take the i_mutex here so we have to play games to
8912 * protect against truncate races as the page could now be beyond EOF. Because
8913 * vmtruncate() writes the inode size before removing pages, once we have the
8914 * page lock we can determine safely if the page is beyond EOF. If it is not
8915 * beyond EOF, then the page is guaranteed safe against truncation until we
8918 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8920 struct page *page = vmf->page;
8921 struct inode *inode = file_inode(vma->vm_file);
8922 struct btrfs_root *root = BTRFS_I(inode)->root;
8923 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8924 struct btrfs_ordered_extent *ordered;
8925 struct extent_state *cached_state = NULL;
8927 unsigned long zero_start;
8936 reserved_space = PAGE_SIZE;
8938 sb_start_pagefault(inode->i_sb);
8939 page_start = page_offset(page);
8940 page_end = page_start + PAGE_SIZE - 1;
8944 * Reserving delalloc space after obtaining the page lock can lead to
8945 * deadlock. For example, if a dirty page is locked by this function
8946 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8947 * dirty page write out, then the btrfs_writepage() function could
8948 * end up waiting indefinitely to get a lock on the page currently
8949 * being processed by btrfs_page_mkwrite() function.
8951 ret = btrfs_delalloc_reserve_space(inode, page_start,
8954 ret = file_update_time(vma->vm_file);
8960 else /* -ENOSPC, -EIO, etc */
8961 ret = VM_FAULT_SIGBUS;
8967 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8970 size = i_size_read(inode);
8972 if ((page->mapping != inode->i_mapping) ||
8973 (page_start >= size)) {
8974 /* page got truncated out from underneath us */
8977 wait_on_page_writeback(page);
8979 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8980 set_page_extent_mapped(page);
8983 * we can't set the delalloc bits if there are pending ordered
8984 * extents. Drop our locks and wait for them to finish
8986 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8988 unlock_extent_cached(io_tree, page_start, page_end,
8989 &cached_state, GFP_NOFS);
8991 btrfs_start_ordered_extent(inode, ordered, 1);
8992 btrfs_put_ordered_extent(ordered);
8996 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8997 reserved_space = round_up(size - page_start, root->sectorsize);
8998 if (reserved_space < PAGE_SIZE) {
8999 end = page_start + reserved_space - 1;
9000 spin_lock(&BTRFS_I(inode)->lock);
9001 BTRFS_I(inode)->outstanding_extents++;
9002 spin_unlock(&BTRFS_I(inode)->lock);
9003 btrfs_delalloc_release_space(inode, page_start,
9004 PAGE_SIZE - reserved_space);
9009 * XXX - page_mkwrite gets called every time the page is dirtied, even
9010 * if it was already dirty, so for space accounting reasons we need to
9011 * clear any delalloc bits for the range we are fixing to save. There
9012 * is probably a better way to do this, but for now keep consistent with
9013 * prepare_pages in the normal write path.
9015 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9016 EXTENT_DIRTY | EXTENT_DELALLOC |
9017 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9018 0, 0, &cached_state, GFP_NOFS);
9020 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9023 unlock_extent_cached(io_tree, page_start, page_end,
9024 &cached_state, GFP_NOFS);
9025 ret = VM_FAULT_SIGBUS;
9030 /* page is wholly or partially inside EOF */
9031 if (page_start + PAGE_SIZE > size)
9032 zero_start = size & ~PAGE_MASK;
9034 zero_start = PAGE_SIZE;
9036 if (zero_start != PAGE_SIZE) {
9038 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9039 flush_dcache_page(page);
9042 ClearPageChecked(page);
9043 set_page_dirty(page);
9044 SetPageUptodate(page);
9046 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9047 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9048 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9050 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9054 sb_end_pagefault(inode->i_sb);
9055 return VM_FAULT_LOCKED;
9059 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9061 sb_end_pagefault(inode->i_sb);
9065 static int btrfs_truncate(struct inode *inode)
9067 struct btrfs_root *root = BTRFS_I(inode)->root;
9068 struct btrfs_block_rsv *rsv;
9071 struct btrfs_trans_handle *trans;
9072 u64 mask = root->sectorsize - 1;
9073 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9075 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9081 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9082 * 3 things going on here
9084 * 1) We need to reserve space for our orphan item and the space to
9085 * delete our orphan item. Lord knows we don't want to have a dangling
9086 * orphan item because we didn't reserve space to remove it.
9088 * 2) We need to reserve space to update our inode.
9090 * 3) We need to have something to cache all the space that is going to
9091 * be free'd up by the truncate operation, but also have some slack
9092 * space reserved in case it uses space during the truncate (thank you
9093 * very much snapshotting).
9095 * And we need these to all be separate. The fact is we can use a lot of
9096 * space doing the truncate, and we have no earthly idea how much space
9097 * we will use, so we need the truncate reservation to be separate so it
9098 * doesn't end up using space reserved for updating the inode or
9099 * removing the orphan item. We also need to be able to stop the
9100 * transaction and start a new one, which means we need to be able to
9101 * update the inode several times, and we have no idea of knowing how
9102 * many times that will be, so we can't just reserve 1 item for the
9103 * entirety of the operation, so that has to be done separately as well.
9104 * Then there is the orphan item, which does indeed need to be held on
9105 * to for the whole operation, and we need nobody to touch this reserved
9106 * space except the orphan code.
9108 * So that leaves us with
9110 * 1) root->orphan_block_rsv - for the orphan deletion.
9111 * 2) rsv - for the truncate reservation, which we will steal from the
9112 * transaction reservation.
9113 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9114 * updating the inode.
9116 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9119 rsv->size = min_size;
9123 * 1 for the truncate slack space
9124 * 1 for updating the inode.
9126 trans = btrfs_start_transaction(root, 2);
9127 if (IS_ERR(trans)) {
9128 err = PTR_ERR(trans);
9132 /* Migrate the slack space for the truncate to our reserve */
9133 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9138 * So if we truncate and then write and fsync we normally would just
9139 * write the extents that changed, which is a problem if we need to
9140 * first truncate that entire inode. So set this flag so we write out
9141 * all of the extents in the inode to the sync log so we're completely
9144 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9145 trans->block_rsv = rsv;
9148 ret = btrfs_truncate_inode_items(trans, root, inode,
9150 BTRFS_EXTENT_DATA_KEY);
9151 if (ret != -ENOSPC && ret != -EAGAIN) {
9156 trans->block_rsv = &root->fs_info->trans_block_rsv;
9157 ret = btrfs_update_inode(trans, root, inode);
9163 btrfs_end_transaction(trans, root);
9164 btrfs_btree_balance_dirty(root);
9166 trans = btrfs_start_transaction(root, 2);
9167 if (IS_ERR(trans)) {
9168 ret = err = PTR_ERR(trans);
9173 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9175 BUG_ON(ret); /* shouldn't happen */
9176 trans->block_rsv = rsv;
9179 if (ret == 0 && inode->i_nlink > 0) {
9180 trans->block_rsv = root->orphan_block_rsv;
9181 ret = btrfs_orphan_del(trans, inode);
9187 trans->block_rsv = &root->fs_info->trans_block_rsv;
9188 ret = btrfs_update_inode(trans, root, inode);
9192 ret = btrfs_end_transaction(trans, root);
9193 btrfs_btree_balance_dirty(root);
9196 btrfs_free_block_rsv(root, rsv);
9205 * create a new subvolume directory/inode (helper for the ioctl).
9207 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9208 struct btrfs_root *new_root,
9209 struct btrfs_root *parent_root,
9212 struct inode *inode;
9216 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9217 new_dirid, new_dirid,
9218 S_IFDIR | (~current_umask() & S_IRWXUGO),
9221 return PTR_ERR(inode);
9222 inode->i_op = &btrfs_dir_inode_operations;
9223 inode->i_fop = &btrfs_dir_file_operations;
9225 set_nlink(inode, 1);
9226 btrfs_i_size_write(inode, 0);
9227 unlock_new_inode(inode);
9229 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9231 btrfs_err(new_root->fs_info,
9232 "error inheriting subvolume %llu properties: %d",
9233 new_root->root_key.objectid, err);
9235 err = btrfs_update_inode(trans, new_root, inode);
9241 struct inode *btrfs_alloc_inode(struct super_block *sb)
9243 struct btrfs_inode *ei;
9244 struct inode *inode;
9246 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9253 ei->last_sub_trans = 0;
9254 ei->logged_trans = 0;
9255 ei->delalloc_bytes = 0;
9256 ei->defrag_bytes = 0;
9257 ei->disk_i_size = 0;
9260 ei->index_cnt = (u64)-1;
9262 ei->last_unlink_trans = 0;
9263 ei->last_log_commit = 0;
9264 ei->delayed_iput_count = 0;
9266 spin_lock_init(&ei->lock);
9267 ei->outstanding_extents = 0;
9268 ei->reserved_extents = 0;
9270 ei->runtime_flags = 0;
9271 ei->force_compress = BTRFS_COMPRESS_NONE;
9273 ei->delayed_node = NULL;
9275 ei->i_otime.tv_sec = 0;
9276 ei->i_otime.tv_nsec = 0;
9278 inode = &ei->vfs_inode;
9279 extent_map_tree_init(&ei->extent_tree);
9280 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9281 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9282 ei->io_tree.track_uptodate = 1;
9283 ei->io_failure_tree.track_uptodate = 1;
9284 atomic_set(&ei->sync_writers, 0);
9285 mutex_init(&ei->log_mutex);
9286 mutex_init(&ei->delalloc_mutex);
9287 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9288 INIT_LIST_HEAD(&ei->delalloc_inodes);
9289 INIT_LIST_HEAD(&ei->delayed_iput);
9290 RB_CLEAR_NODE(&ei->rb_node);
9291 init_rwsem(&ei->dio_sem);
9296 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9297 void btrfs_test_destroy_inode(struct inode *inode)
9299 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9300 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9304 static void btrfs_i_callback(struct rcu_head *head)
9306 struct inode *inode = container_of(head, struct inode, i_rcu);
9307 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9310 void btrfs_destroy_inode(struct inode *inode)
9312 struct btrfs_ordered_extent *ordered;
9313 struct btrfs_root *root = BTRFS_I(inode)->root;
9315 WARN_ON(!hlist_empty(&inode->i_dentry));
9316 WARN_ON(inode->i_data.nrpages);
9317 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9318 WARN_ON(BTRFS_I(inode)->reserved_extents);
9319 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9320 WARN_ON(BTRFS_I(inode)->csum_bytes);
9321 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9324 * This can happen where we create an inode, but somebody else also
9325 * created the same inode and we need to destroy the one we already
9331 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9332 &BTRFS_I(inode)->runtime_flags)) {
9333 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9335 atomic_dec(&root->orphan_inodes);
9339 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9343 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9344 ordered->file_offset, ordered->len);
9345 btrfs_remove_ordered_extent(inode, ordered);
9346 btrfs_put_ordered_extent(ordered);
9347 btrfs_put_ordered_extent(ordered);
9350 btrfs_qgroup_check_reserved_leak(inode);
9351 inode_tree_del(inode);
9352 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9354 call_rcu(&inode->i_rcu, btrfs_i_callback);
9357 int btrfs_drop_inode(struct inode *inode)
9359 struct btrfs_root *root = BTRFS_I(inode)->root;
9364 /* the snap/subvol tree is on deleting */
9365 if (btrfs_root_refs(&root->root_item) == 0)
9368 return generic_drop_inode(inode);
9371 static void init_once(void *foo)
9373 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9375 inode_init_once(&ei->vfs_inode);
9378 void btrfs_destroy_cachep(void)
9381 * Make sure all delayed rcu free inodes are flushed before we
9385 kmem_cache_destroy(btrfs_inode_cachep);
9386 kmem_cache_destroy(btrfs_trans_handle_cachep);
9387 kmem_cache_destroy(btrfs_transaction_cachep);
9388 kmem_cache_destroy(btrfs_path_cachep);
9389 kmem_cache_destroy(btrfs_free_space_cachep);
9392 int btrfs_init_cachep(void)
9394 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9395 sizeof(struct btrfs_inode), 0,
9396 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9398 if (!btrfs_inode_cachep)
9401 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9402 sizeof(struct btrfs_trans_handle), 0,
9403 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9404 if (!btrfs_trans_handle_cachep)
9407 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9408 sizeof(struct btrfs_transaction), 0,
9409 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9410 if (!btrfs_transaction_cachep)
9413 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9414 sizeof(struct btrfs_path), 0,
9415 SLAB_MEM_SPREAD, NULL);
9416 if (!btrfs_path_cachep)
9419 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9420 sizeof(struct btrfs_free_space), 0,
9421 SLAB_MEM_SPREAD, NULL);
9422 if (!btrfs_free_space_cachep)
9427 btrfs_destroy_cachep();
9431 static int btrfs_getattr(struct vfsmount *mnt,
9432 struct dentry *dentry, struct kstat *stat)
9435 struct inode *inode = d_inode(dentry);
9436 u32 blocksize = inode->i_sb->s_blocksize;
9438 generic_fillattr(inode, stat);
9439 stat->dev = BTRFS_I(inode)->root->anon_dev;
9441 spin_lock(&BTRFS_I(inode)->lock);
9442 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9443 spin_unlock(&BTRFS_I(inode)->lock);
9444 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9445 ALIGN(delalloc_bytes, blocksize)) >> 9;
9449 static int btrfs_rename_exchange(struct inode *old_dir,
9450 struct dentry *old_dentry,
9451 struct inode *new_dir,
9452 struct dentry *new_dentry)
9454 struct btrfs_trans_handle *trans;
9455 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9456 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9457 struct inode *new_inode = new_dentry->d_inode;
9458 struct inode *old_inode = old_dentry->d_inode;
9459 struct timespec ctime = CURRENT_TIME;
9460 struct dentry *parent;
9461 u64 old_ino = btrfs_ino(old_inode);
9462 u64 new_ino = btrfs_ino(new_inode);
9467 bool root_log_pinned = false;
9468 bool dest_log_pinned = false;
9470 /* we only allow rename subvolume link between subvolumes */
9471 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9474 /* close the race window with snapshot create/destroy ioctl */
9475 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9476 down_read(&root->fs_info->subvol_sem);
9477 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9478 down_read(&dest->fs_info->subvol_sem);
9481 * We want to reserve the absolute worst case amount of items. So if
9482 * both inodes are subvols and we need to unlink them then that would
9483 * require 4 item modifications, but if they are both normal inodes it
9484 * would require 5 item modifications, so we'll assume their normal
9485 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9486 * should cover the worst case number of items we'll modify.
9488 trans = btrfs_start_transaction(root, 12);
9489 if (IS_ERR(trans)) {
9490 ret = PTR_ERR(trans);
9495 * We need to find a free sequence number both in the source and
9496 * in the destination directory for the exchange.
9498 ret = btrfs_set_inode_index(new_dir, &old_idx);
9501 ret = btrfs_set_inode_index(old_dir, &new_idx);
9505 BTRFS_I(old_inode)->dir_index = 0ULL;
9506 BTRFS_I(new_inode)->dir_index = 0ULL;
9508 /* Reference for the source. */
9509 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9510 /* force full log commit if subvolume involved. */
9511 btrfs_set_log_full_commit(root->fs_info, trans);
9513 btrfs_pin_log_trans(root);
9514 root_log_pinned = true;
9515 ret = btrfs_insert_inode_ref(trans, dest,
9516 new_dentry->d_name.name,
9517 new_dentry->d_name.len,
9519 btrfs_ino(new_dir), old_idx);
9524 /* And now for the dest. */
9525 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9526 /* force full log commit if subvolume involved. */
9527 btrfs_set_log_full_commit(dest->fs_info, trans);
9529 btrfs_pin_log_trans(dest);
9530 dest_log_pinned = true;
9531 ret = btrfs_insert_inode_ref(trans, root,
9532 old_dentry->d_name.name,
9533 old_dentry->d_name.len,
9535 btrfs_ino(old_dir), new_idx);
9540 /* Update inode version and ctime/mtime. */
9541 inode_inc_iversion(old_dir);
9542 inode_inc_iversion(new_dir);
9543 inode_inc_iversion(old_inode);
9544 inode_inc_iversion(new_inode);
9545 old_dir->i_ctime = old_dir->i_mtime = ctime;
9546 new_dir->i_ctime = new_dir->i_mtime = ctime;
9547 old_inode->i_ctime = ctime;
9548 new_inode->i_ctime = ctime;
9550 if (old_dentry->d_parent != new_dentry->d_parent) {
9551 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9552 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9555 /* src is a subvolume */
9556 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9557 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9558 ret = btrfs_unlink_subvol(trans, root, old_dir,
9560 old_dentry->d_name.name,
9561 old_dentry->d_name.len);
9562 } else { /* src is an inode */
9563 ret = __btrfs_unlink_inode(trans, root, old_dir,
9564 old_dentry->d_inode,
9565 old_dentry->d_name.name,
9566 old_dentry->d_name.len);
9568 ret = btrfs_update_inode(trans, root, old_inode);
9571 btrfs_abort_transaction(trans, root, ret);
9575 /* dest is a subvolume */
9576 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9577 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9578 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9580 new_dentry->d_name.name,
9581 new_dentry->d_name.len);
9582 } else { /* dest is an inode */
9583 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9584 new_dentry->d_inode,
9585 new_dentry->d_name.name,
9586 new_dentry->d_name.len);
9588 ret = btrfs_update_inode(trans, dest, new_inode);
9591 btrfs_abort_transaction(trans, root, ret);
9595 ret = btrfs_add_link(trans, new_dir, old_inode,
9596 new_dentry->d_name.name,
9597 new_dentry->d_name.len, 0, old_idx);
9599 btrfs_abort_transaction(trans, root, ret);
9603 ret = btrfs_add_link(trans, old_dir, new_inode,
9604 old_dentry->d_name.name,
9605 old_dentry->d_name.len, 0, new_idx);
9607 btrfs_abort_transaction(trans, root, ret);
9611 if (old_inode->i_nlink == 1)
9612 BTRFS_I(old_inode)->dir_index = old_idx;
9613 if (new_inode->i_nlink == 1)
9614 BTRFS_I(new_inode)->dir_index = new_idx;
9616 if (root_log_pinned) {
9617 parent = new_dentry->d_parent;
9618 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9619 btrfs_end_log_trans(root);
9620 root_log_pinned = false;
9622 if (dest_log_pinned) {
9623 parent = old_dentry->d_parent;
9624 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9625 btrfs_end_log_trans(dest);
9626 dest_log_pinned = false;
9630 * If we have pinned a log and an error happened, we unpin tasks
9631 * trying to sync the log and force them to fallback to a transaction
9632 * commit if the log currently contains any of the inodes involved in
9633 * this rename operation (to ensure we do not persist a log with an
9634 * inconsistent state for any of these inodes or leading to any
9635 * inconsistencies when replayed). If the transaction was aborted, the
9636 * abortion reason is propagated to userspace when attempting to commit
9637 * the transaction. If the log does not contain any of these inodes, we
9638 * allow the tasks to sync it.
9640 if (ret && (root_log_pinned || dest_log_pinned)) {
9641 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9642 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9643 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9645 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9646 btrfs_set_log_full_commit(root->fs_info, trans);
9648 if (root_log_pinned) {
9649 btrfs_end_log_trans(root);
9650 root_log_pinned = false;
9652 if (dest_log_pinned) {
9653 btrfs_end_log_trans(dest);
9654 dest_log_pinned = false;
9657 ret = btrfs_end_transaction(trans, root);
9659 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9660 up_read(&dest->fs_info->subvol_sem);
9661 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9662 up_read(&root->fs_info->subvol_sem);
9667 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9668 struct btrfs_root *root,
9670 struct dentry *dentry)
9673 struct inode *inode;
9677 ret = btrfs_find_free_ino(root, &objectid);
9681 inode = btrfs_new_inode(trans, root, dir,
9682 dentry->d_name.name,
9686 S_IFCHR | WHITEOUT_MODE,
9689 if (IS_ERR(inode)) {
9690 ret = PTR_ERR(inode);
9694 inode->i_op = &btrfs_special_inode_operations;
9695 init_special_inode(inode, inode->i_mode,
9698 ret = btrfs_init_inode_security(trans, inode, dir,
9703 ret = btrfs_add_nondir(trans, dir, dentry,
9708 ret = btrfs_update_inode(trans, root, inode);
9710 unlock_new_inode(inode);
9712 inode_dec_link_count(inode);
9718 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9719 struct inode *new_dir, struct dentry *new_dentry,
9722 struct btrfs_trans_handle *trans;
9723 unsigned int trans_num_items;
9724 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9725 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9726 struct inode *new_inode = d_inode(new_dentry);
9727 struct inode *old_inode = d_inode(old_dentry);
9731 u64 old_ino = btrfs_ino(old_inode);
9732 bool log_pinned = false;
9734 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9737 /* we only allow rename subvolume link between subvolumes */
9738 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9741 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9742 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9745 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9746 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9750 /* check for collisions, even if the name isn't there */
9751 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9752 new_dentry->d_name.name,
9753 new_dentry->d_name.len);
9756 if (ret == -EEXIST) {
9758 * eexist without a new_inode */
9759 if (WARN_ON(!new_inode)) {
9763 /* maybe -EOVERFLOW */
9770 * we're using rename to replace one file with another. Start IO on it
9771 * now so we don't add too much work to the end of the transaction
9773 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9774 filemap_flush(old_inode->i_mapping);
9776 /* close the racy window with snapshot create/destroy ioctl */
9777 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9778 down_read(&root->fs_info->subvol_sem);
9780 * We want to reserve the absolute worst case amount of items. So if
9781 * both inodes are subvols and we need to unlink them then that would
9782 * require 4 item modifications, but if they are both normal inodes it
9783 * would require 5 item modifications, so we'll assume they are normal
9784 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9785 * should cover the worst case number of items we'll modify.
9786 * If our rename has the whiteout flag, we need more 5 units for the
9787 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9788 * when selinux is enabled).
9790 trans_num_items = 11;
9791 if (flags & RENAME_WHITEOUT)
9792 trans_num_items += 5;
9793 trans = btrfs_start_transaction(root, trans_num_items);
9794 if (IS_ERR(trans)) {
9795 ret = PTR_ERR(trans);
9800 btrfs_record_root_in_trans(trans, dest);
9802 ret = btrfs_set_inode_index(new_dir, &index);
9806 BTRFS_I(old_inode)->dir_index = 0ULL;
9807 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9808 /* force full log commit if subvolume involved. */
9809 btrfs_set_log_full_commit(root->fs_info, trans);
9811 btrfs_pin_log_trans(root);
9813 ret = btrfs_insert_inode_ref(trans, dest,
9814 new_dentry->d_name.name,
9815 new_dentry->d_name.len,
9817 btrfs_ino(new_dir), index);
9822 inode_inc_iversion(old_dir);
9823 inode_inc_iversion(new_dir);
9824 inode_inc_iversion(old_inode);
9825 old_dir->i_ctime = old_dir->i_mtime =
9826 new_dir->i_ctime = new_dir->i_mtime =
9827 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9829 if (old_dentry->d_parent != new_dentry->d_parent)
9830 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9832 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9833 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9834 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9835 old_dentry->d_name.name,
9836 old_dentry->d_name.len);
9838 ret = __btrfs_unlink_inode(trans, root, old_dir,
9839 d_inode(old_dentry),
9840 old_dentry->d_name.name,
9841 old_dentry->d_name.len);
9843 ret = btrfs_update_inode(trans, root, old_inode);
9846 btrfs_abort_transaction(trans, root, ret);
9851 inode_inc_iversion(new_inode);
9852 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9853 if (unlikely(btrfs_ino(new_inode) ==
9854 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9855 root_objectid = BTRFS_I(new_inode)->location.objectid;
9856 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9858 new_dentry->d_name.name,
9859 new_dentry->d_name.len);
9860 BUG_ON(new_inode->i_nlink == 0);
9862 ret = btrfs_unlink_inode(trans, dest, new_dir,
9863 d_inode(new_dentry),
9864 new_dentry->d_name.name,
9865 new_dentry->d_name.len);
9867 if (!ret && new_inode->i_nlink == 0)
9868 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9870 btrfs_abort_transaction(trans, root, ret);
9875 ret = btrfs_add_link(trans, new_dir, old_inode,
9876 new_dentry->d_name.name,
9877 new_dentry->d_name.len, 0, index);
9879 btrfs_abort_transaction(trans, root, ret);
9883 if (old_inode->i_nlink == 1)
9884 BTRFS_I(old_inode)->dir_index = index;
9887 struct dentry *parent = new_dentry->d_parent;
9889 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9890 btrfs_end_log_trans(root);
9894 if (flags & RENAME_WHITEOUT) {
9895 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9899 btrfs_abort_transaction(trans, root, ret);
9905 * If we have pinned the log and an error happened, we unpin tasks
9906 * trying to sync the log and force them to fallback to a transaction
9907 * commit if the log currently contains any of the inodes involved in
9908 * this rename operation (to ensure we do not persist a log with an
9909 * inconsistent state for any of these inodes or leading to any
9910 * inconsistencies when replayed). If the transaction was aborted, the
9911 * abortion reason is propagated to userspace when attempting to commit
9912 * the transaction. If the log does not contain any of these inodes, we
9913 * allow the tasks to sync it.
9915 if (ret && log_pinned) {
9916 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9917 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9918 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9920 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9921 btrfs_set_log_full_commit(root->fs_info, trans);
9923 btrfs_end_log_trans(root);
9926 btrfs_end_transaction(trans, root);
9928 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9929 up_read(&root->fs_info->subvol_sem);
9934 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9935 struct inode *new_dir, struct dentry *new_dentry,
9938 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9941 if (flags & RENAME_EXCHANGE)
9942 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9945 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9948 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9950 struct btrfs_delalloc_work *delalloc_work;
9951 struct inode *inode;
9953 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9955 inode = delalloc_work->inode;
9956 filemap_flush(inode->i_mapping);
9957 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9958 &BTRFS_I(inode)->runtime_flags))
9959 filemap_flush(inode->i_mapping);
9961 if (delalloc_work->delay_iput)
9962 btrfs_add_delayed_iput(inode);
9965 complete(&delalloc_work->completion);
9968 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9971 struct btrfs_delalloc_work *work;
9973 work = kmalloc(sizeof(*work), GFP_NOFS);
9977 init_completion(&work->completion);
9978 INIT_LIST_HEAD(&work->list);
9979 work->inode = inode;
9980 work->delay_iput = delay_iput;
9981 WARN_ON_ONCE(!inode);
9982 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9983 btrfs_run_delalloc_work, NULL, NULL);
9988 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9990 wait_for_completion(&work->completion);
9995 * some fairly slow code that needs optimization. This walks the list
9996 * of all the inodes with pending delalloc and forces them to disk.
9998 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10001 struct btrfs_inode *binode;
10002 struct inode *inode;
10003 struct btrfs_delalloc_work *work, *next;
10004 struct list_head works;
10005 struct list_head splice;
10008 INIT_LIST_HEAD(&works);
10009 INIT_LIST_HEAD(&splice);
10011 mutex_lock(&root->delalloc_mutex);
10012 spin_lock(&root->delalloc_lock);
10013 list_splice_init(&root->delalloc_inodes, &splice);
10014 while (!list_empty(&splice)) {
10015 binode = list_entry(splice.next, struct btrfs_inode,
10018 list_move_tail(&binode->delalloc_inodes,
10019 &root->delalloc_inodes);
10020 inode = igrab(&binode->vfs_inode);
10022 cond_resched_lock(&root->delalloc_lock);
10025 spin_unlock(&root->delalloc_lock);
10027 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10030 btrfs_add_delayed_iput(inode);
10036 list_add_tail(&work->list, &works);
10037 btrfs_queue_work(root->fs_info->flush_workers,
10040 if (nr != -1 && ret >= nr)
10043 spin_lock(&root->delalloc_lock);
10045 spin_unlock(&root->delalloc_lock);
10048 list_for_each_entry_safe(work, next, &works, list) {
10049 list_del_init(&work->list);
10050 btrfs_wait_and_free_delalloc_work(work);
10053 if (!list_empty_careful(&splice)) {
10054 spin_lock(&root->delalloc_lock);
10055 list_splice_tail(&splice, &root->delalloc_inodes);
10056 spin_unlock(&root->delalloc_lock);
10058 mutex_unlock(&root->delalloc_mutex);
10062 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10066 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10069 ret = __start_delalloc_inodes(root, delay_iput, -1);
10073 * the filemap_flush will queue IO into the worker threads, but
10074 * we have to make sure the IO is actually started and that
10075 * ordered extents get created before we return
10077 atomic_inc(&root->fs_info->async_submit_draining);
10078 while (atomic_read(&root->fs_info->nr_async_submits) ||
10079 atomic_read(&root->fs_info->async_delalloc_pages)) {
10080 wait_event(root->fs_info->async_submit_wait,
10081 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10082 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10084 atomic_dec(&root->fs_info->async_submit_draining);
10088 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10091 struct btrfs_root *root;
10092 struct list_head splice;
10095 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10098 INIT_LIST_HEAD(&splice);
10100 mutex_lock(&fs_info->delalloc_root_mutex);
10101 spin_lock(&fs_info->delalloc_root_lock);
10102 list_splice_init(&fs_info->delalloc_roots, &splice);
10103 while (!list_empty(&splice) && nr) {
10104 root = list_first_entry(&splice, struct btrfs_root,
10106 root = btrfs_grab_fs_root(root);
10108 list_move_tail(&root->delalloc_root,
10109 &fs_info->delalloc_roots);
10110 spin_unlock(&fs_info->delalloc_root_lock);
10112 ret = __start_delalloc_inodes(root, delay_iput, nr);
10113 btrfs_put_fs_root(root);
10121 spin_lock(&fs_info->delalloc_root_lock);
10123 spin_unlock(&fs_info->delalloc_root_lock);
10126 atomic_inc(&fs_info->async_submit_draining);
10127 while (atomic_read(&fs_info->nr_async_submits) ||
10128 atomic_read(&fs_info->async_delalloc_pages)) {
10129 wait_event(fs_info->async_submit_wait,
10130 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10131 atomic_read(&fs_info->async_delalloc_pages) == 0));
10133 atomic_dec(&fs_info->async_submit_draining);
10135 if (!list_empty_careful(&splice)) {
10136 spin_lock(&fs_info->delalloc_root_lock);
10137 list_splice_tail(&splice, &fs_info->delalloc_roots);
10138 spin_unlock(&fs_info->delalloc_root_lock);
10140 mutex_unlock(&fs_info->delalloc_root_mutex);
10144 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10145 const char *symname)
10147 struct btrfs_trans_handle *trans;
10148 struct btrfs_root *root = BTRFS_I(dir)->root;
10149 struct btrfs_path *path;
10150 struct btrfs_key key;
10151 struct inode *inode = NULL;
10153 int drop_inode = 0;
10159 struct btrfs_file_extent_item *ei;
10160 struct extent_buffer *leaf;
10162 name_len = strlen(symname);
10163 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10164 return -ENAMETOOLONG;
10167 * 2 items for inode item and ref
10168 * 2 items for dir items
10169 * 1 item for updating parent inode item
10170 * 1 item for the inline extent item
10171 * 1 item for xattr if selinux is on
10173 trans = btrfs_start_transaction(root, 7);
10175 return PTR_ERR(trans);
10177 err = btrfs_find_free_ino(root, &objectid);
10181 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10182 dentry->d_name.len, btrfs_ino(dir), objectid,
10183 S_IFLNK|S_IRWXUGO, &index);
10184 if (IS_ERR(inode)) {
10185 err = PTR_ERR(inode);
10190 * If the active LSM wants to access the inode during
10191 * d_instantiate it needs these. Smack checks to see
10192 * if the filesystem supports xattrs by looking at the
10195 inode->i_fop = &btrfs_file_operations;
10196 inode->i_op = &btrfs_file_inode_operations;
10197 inode->i_mapping->a_ops = &btrfs_aops;
10198 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10200 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10202 goto out_unlock_inode;
10204 path = btrfs_alloc_path();
10207 goto out_unlock_inode;
10209 key.objectid = btrfs_ino(inode);
10211 key.type = BTRFS_EXTENT_DATA_KEY;
10212 datasize = btrfs_file_extent_calc_inline_size(name_len);
10213 err = btrfs_insert_empty_item(trans, root, path, &key,
10216 btrfs_free_path(path);
10217 goto out_unlock_inode;
10219 leaf = path->nodes[0];
10220 ei = btrfs_item_ptr(leaf, path->slots[0],
10221 struct btrfs_file_extent_item);
10222 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10223 btrfs_set_file_extent_type(leaf, ei,
10224 BTRFS_FILE_EXTENT_INLINE);
10225 btrfs_set_file_extent_encryption(leaf, ei, 0);
10226 btrfs_set_file_extent_compression(leaf, ei, 0);
10227 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10228 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10230 ptr = btrfs_file_extent_inline_start(ei);
10231 write_extent_buffer(leaf, symname, ptr, name_len);
10232 btrfs_mark_buffer_dirty(leaf);
10233 btrfs_free_path(path);
10235 inode->i_op = &btrfs_symlink_inode_operations;
10236 inode_nohighmem(inode);
10237 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10238 inode_set_bytes(inode, name_len);
10239 btrfs_i_size_write(inode, name_len);
10240 err = btrfs_update_inode(trans, root, inode);
10242 * Last step, add directory indexes for our symlink inode. This is the
10243 * last step to avoid extra cleanup of these indexes if an error happens
10247 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10250 goto out_unlock_inode;
10253 unlock_new_inode(inode);
10254 d_instantiate(dentry, inode);
10257 btrfs_end_transaction(trans, root);
10259 inode_dec_link_count(inode);
10262 btrfs_btree_balance_dirty(root);
10267 unlock_new_inode(inode);
10271 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10272 u64 start, u64 num_bytes, u64 min_size,
10273 loff_t actual_len, u64 *alloc_hint,
10274 struct btrfs_trans_handle *trans)
10276 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10277 struct extent_map *em;
10278 struct btrfs_root *root = BTRFS_I(inode)->root;
10279 struct btrfs_key ins;
10280 u64 cur_offset = start;
10283 u64 last_alloc = (u64)-1;
10285 bool own_trans = true;
10289 while (num_bytes > 0) {
10291 trans = btrfs_start_transaction(root, 3);
10292 if (IS_ERR(trans)) {
10293 ret = PTR_ERR(trans);
10298 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10299 cur_bytes = max(cur_bytes, min_size);
10301 * If we are severely fragmented we could end up with really
10302 * small allocations, so if the allocator is returning small
10303 * chunks lets make its job easier by only searching for those
10306 cur_bytes = min(cur_bytes, last_alloc);
10307 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10308 *alloc_hint, &ins, 1, 0);
10311 btrfs_end_transaction(trans, root);
10314 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10316 last_alloc = ins.offset;
10317 ret = insert_reserved_file_extent(trans, inode,
10318 cur_offset, ins.objectid,
10319 ins.offset, ins.offset,
10320 ins.offset, 0, 0, 0,
10321 BTRFS_FILE_EXTENT_PREALLOC);
10323 btrfs_free_reserved_extent(root, ins.objectid,
10325 btrfs_abort_transaction(trans, root, ret);
10327 btrfs_end_transaction(trans, root);
10331 btrfs_drop_extent_cache(inode, cur_offset,
10332 cur_offset + ins.offset -1, 0);
10334 em = alloc_extent_map();
10336 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10337 &BTRFS_I(inode)->runtime_flags);
10341 em->start = cur_offset;
10342 em->orig_start = cur_offset;
10343 em->len = ins.offset;
10344 em->block_start = ins.objectid;
10345 em->block_len = ins.offset;
10346 em->orig_block_len = ins.offset;
10347 em->ram_bytes = ins.offset;
10348 em->bdev = root->fs_info->fs_devices->latest_bdev;
10349 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10350 em->generation = trans->transid;
10353 write_lock(&em_tree->lock);
10354 ret = add_extent_mapping(em_tree, em, 1);
10355 write_unlock(&em_tree->lock);
10356 if (ret != -EEXIST)
10358 btrfs_drop_extent_cache(inode, cur_offset,
10359 cur_offset + ins.offset - 1,
10362 free_extent_map(em);
10364 num_bytes -= ins.offset;
10365 cur_offset += ins.offset;
10366 *alloc_hint = ins.objectid + ins.offset;
10368 inode_inc_iversion(inode);
10369 inode->i_ctime = current_fs_time(inode->i_sb);
10370 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10371 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10372 (actual_len > inode->i_size) &&
10373 (cur_offset > inode->i_size)) {
10374 if (cur_offset > actual_len)
10375 i_size = actual_len;
10377 i_size = cur_offset;
10378 i_size_write(inode, i_size);
10379 btrfs_ordered_update_i_size(inode, i_size, NULL);
10382 ret = btrfs_update_inode(trans, root, inode);
10385 btrfs_abort_transaction(trans, root, ret);
10387 btrfs_end_transaction(trans, root);
10392 btrfs_end_transaction(trans, root);
10397 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10398 u64 start, u64 num_bytes, u64 min_size,
10399 loff_t actual_len, u64 *alloc_hint)
10401 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10402 min_size, actual_len, alloc_hint,
10406 int btrfs_prealloc_file_range_trans(struct inode *inode,
10407 struct btrfs_trans_handle *trans, int mode,
10408 u64 start, u64 num_bytes, u64 min_size,
10409 loff_t actual_len, u64 *alloc_hint)
10411 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10412 min_size, actual_len, alloc_hint, trans);
10415 static int btrfs_set_page_dirty(struct page *page)
10417 return __set_page_dirty_nobuffers(page);
10420 static int btrfs_permission(struct inode *inode, int mask)
10422 struct btrfs_root *root = BTRFS_I(inode)->root;
10423 umode_t mode = inode->i_mode;
10425 if (mask & MAY_WRITE &&
10426 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10427 if (btrfs_root_readonly(root))
10429 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10432 return generic_permission(inode, mask);
10435 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10437 struct btrfs_trans_handle *trans;
10438 struct btrfs_root *root = BTRFS_I(dir)->root;
10439 struct inode *inode = NULL;
10445 * 5 units required for adding orphan entry
10447 trans = btrfs_start_transaction(root, 5);
10449 return PTR_ERR(trans);
10451 ret = btrfs_find_free_ino(root, &objectid);
10455 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10456 btrfs_ino(dir), objectid, mode, &index);
10457 if (IS_ERR(inode)) {
10458 ret = PTR_ERR(inode);
10463 inode->i_fop = &btrfs_file_operations;
10464 inode->i_op = &btrfs_file_inode_operations;
10466 inode->i_mapping->a_ops = &btrfs_aops;
10467 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10469 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10473 ret = btrfs_update_inode(trans, root, inode);
10476 ret = btrfs_orphan_add(trans, inode);
10481 * We set number of links to 0 in btrfs_new_inode(), and here we set
10482 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10485 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10487 set_nlink(inode, 1);
10488 unlock_new_inode(inode);
10489 d_tmpfile(dentry, inode);
10490 mark_inode_dirty(inode);
10493 btrfs_end_transaction(trans, root);
10496 btrfs_balance_delayed_items(root);
10497 btrfs_btree_balance_dirty(root);
10501 unlock_new_inode(inode);
10506 /* Inspired by filemap_check_errors() */
10507 int btrfs_inode_check_errors(struct inode *inode)
10511 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10512 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10514 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10515 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10521 static const struct inode_operations btrfs_dir_inode_operations = {
10522 .getattr = btrfs_getattr,
10523 .lookup = btrfs_lookup,
10524 .create = btrfs_create,
10525 .unlink = btrfs_unlink,
10526 .link = btrfs_link,
10527 .mkdir = btrfs_mkdir,
10528 .rmdir = btrfs_rmdir,
10529 .rename2 = btrfs_rename2,
10530 .symlink = btrfs_symlink,
10531 .setattr = btrfs_setattr,
10532 .mknod = btrfs_mknod,
10533 .setxattr = generic_setxattr,
10534 .getxattr = generic_getxattr,
10535 .listxattr = btrfs_listxattr,
10536 .removexattr = generic_removexattr,
10537 .permission = btrfs_permission,
10538 .get_acl = btrfs_get_acl,
10539 .set_acl = btrfs_set_acl,
10540 .update_time = btrfs_update_time,
10541 .tmpfile = btrfs_tmpfile,
10543 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10544 .lookup = btrfs_lookup,
10545 .permission = btrfs_permission,
10546 .get_acl = btrfs_get_acl,
10547 .set_acl = btrfs_set_acl,
10548 .update_time = btrfs_update_time,
10551 static const struct file_operations btrfs_dir_file_operations = {
10552 .llseek = generic_file_llseek,
10553 .read = generic_read_dir,
10554 .iterate_shared = btrfs_real_readdir,
10555 .unlocked_ioctl = btrfs_ioctl,
10556 #ifdef CONFIG_COMPAT
10557 .compat_ioctl = btrfs_compat_ioctl,
10559 .release = btrfs_release_file,
10560 .fsync = btrfs_sync_file,
10563 static const struct extent_io_ops btrfs_extent_io_ops = {
10564 .fill_delalloc = run_delalloc_range,
10565 .submit_bio_hook = btrfs_submit_bio_hook,
10566 .merge_bio_hook = btrfs_merge_bio_hook,
10567 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10568 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10569 .writepage_start_hook = btrfs_writepage_start_hook,
10570 .set_bit_hook = btrfs_set_bit_hook,
10571 .clear_bit_hook = btrfs_clear_bit_hook,
10572 .merge_extent_hook = btrfs_merge_extent_hook,
10573 .split_extent_hook = btrfs_split_extent_hook,
10577 * btrfs doesn't support the bmap operation because swapfiles
10578 * use bmap to make a mapping of extents in the file. They assume
10579 * these extents won't change over the life of the file and they
10580 * use the bmap result to do IO directly to the drive.
10582 * the btrfs bmap call would return logical addresses that aren't
10583 * suitable for IO and they also will change frequently as COW
10584 * operations happen. So, swapfile + btrfs == corruption.
10586 * For now we're avoiding this by dropping bmap.
10588 static const struct address_space_operations btrfs_aops = {
10589 .readpage = btrfs_readpage,
10590 .writepage = btrfs_writepage,
10591 .writepages = btrfs_writepages,
10592 .readpages = btrfs_readpages,
10593 .direct_IO = btrfs_direct_IO,
10594 .invalidatepage = btrfs_invalidatepage,
10595 .releasepage = btrfs_releasepage,
10596 .set_page_dirty = btrfs_set_page_dirty,
10597 .error_remove_page = generic_error_remove_page,
10600 static const struct address_space_operations btrfs_symlink_aops = {
10601 .readpage = btrfs_readpage,
10602 .writepage = btrfs_writepage,
10603 .invalidatepage = btrfs_invalidatepage,
10604 .releasepage = btrfs_releasepage,
10607 static const struct inode_operations btrfs_file_inode_operations = {
10608 .getattr = btrfs_getattr,
10609 .setattr = btrfs_setattr,
10610 .setxattr = generic_setxattr,
10611 .getxattr = generic_getxattr,
10612 .listxattr = btrfs_listxattr,
10613 .removexattr = generic_removexattr,
10614 .permission = btrfs_permission,
10615 .fiemap = btrfs_fiemap,
10616 .get_acl = btrfs_get_acl,
10617 .set_acl = btrfs_set_acl,
10618 .update_time = btrfs_update_time,
10620 static const struct inode_operations btrfs_special_inode_operations = {
10621 .getattr = btrfs_getattr,
10622 .setattr = btrfs_setattr,
10623 .permission = btrfs_permission,
10624 .setxattr = generic_setxattr,
10625 .getxattr = generic_getxattr,
10626 .listxattr = btrfs_listxattr,
10627 .removexattr = generic_removexattr,
10628 .get_acl = btrfs_get_acl,
10629 .set_acl = btrfs_set_acl,
10630 .update_time = btrfs_update_time,
10632 static const struct inode_operations btrfs_symlink_inode_operations = {
10633 .readlink = generic_readlink,
10634 .get_link = page_get_link,
10635 .getattr = btrfs_getattr,
10636 .setattr = btrfs_setattr,
10637 .permission = btrfs_permission,
10638 .setxattr = generic_setxattr,
10639 .getxattr = generic_getxattr,
10640 .listxattr = btrfs_listxattr,
10641 .removexattr = generic_removexattr,
10642 .update_time = btrfs_update_time,
10645 const struct dentry_operations btrfs_dentry_operations = {
10646 .d_delete = btrfs_dentry_delete,
10647 .d_release = btrfs_dentry_release,