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"
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
116 static int btrfs_dirty_inode(struct inode *inode);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
131 err = btrfs_init_acl(trans, inode, dir);
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
147 struct page **compressed_pages)
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
153 struct btrfs_file_extent_item *ei;
156 size_t cur_size = size;
157 unsigned long offset;
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
162 inode_add_bytes(inode, size);
164 if (!extent_inserted) {
165 struct btrfs_key key;
168 key.objectid = btrfs_ino(inode);
170 key.type = BTRFS_EXTENT_DATA_KEY;
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
191 if (compress_type != BTRFS_COMPRESS_NONE) {
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
205 compressed_size -= cur_size;
207 btrfs_set_file_extent_compression(leaf, ei,
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_CACHE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_CACHE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 page_cache_release(page);
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
249 struct page **compressed_pages)
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
263 data_len = compressed_size;
266 actual_end > PAGE_CACHE_SIZE ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
269 (actual_end & (root->sectorsize - 1)) == 0) ||
271 data_len > root->fs_info->max_inline) {
275 path = btrfs_alloc_path();
279 trans = btrfs_join_transaction(root);
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
297 btrfs_abort_transaction(trans, root, ret);
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
310 } else if (ret == -ENOSPC) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
331 struct async_extent {
336 unsigned long nr_pages;
338 struct list_head list;
343 struct btrfs_root *root;
344 struct page *locked_page;
347 struct list_head extents;
348 struct btrfs_work work;
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
355 unsigned long nr_pages,
358 struct async_extent *async_extent;
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
372 static inline int inode_need_compress(struct inode *inode)
374 struct btrfs_root *root = BTRFS_I(inode)->root;
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
409 struct async_cow *async_cow,
412 struct btrfs_root *root = BTRFS_I(inode)->root;
414 u64 blocksize = root->sectorsize;
416 u64 isize = i_size_read(inode);
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
427 int compress_type = root->fs_info->compress_type;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
435 actual_end = min_t(u64, isize, end + 1);
438 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_SIZE);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
454 total_compressed = actual_end - start;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode)) {
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode, start, end);
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
516 unsigned long offset = total_compressed &
517 (PAGE_CACHE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
527 PAGE_CACHE_SIZE - offset);
528 kunmap_atomic(kaddr);
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret = cow_file_range_inline(root, inode, start, end,
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
546 compress_type, pages);
549 unsigned long clear_flags = EXTENT_DELALLOC |
551 unsigned long page_error_op;
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed = ALIGN(total_compressed, blocksize);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
584 if (total_compressed >= total_in) {
587 num_bytes = total_in;
590 if (!will_compress && pages) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
597 page_cache_release(pages[i]);
601 total_compressed = 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
621 if (start + num_bytes < end) {
628 cleanup_and_bail_uncompressed:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 page_cache_release(pages[i]);
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 page_cache_release(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 &page_started, &nr_written, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
733 unlock_page(async_cow->locked_page);
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
747 free_async_extent_pages(async_extent);
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode,
762 async_extent->start +
763 async_extent->ram_size - 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
777 em = alloc_extent_map();
780 goto out_free_reserve;
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
812 goto out_free_reserve;
814 ret = btrfs_add_ordered_extent_compress(inode,
817 async_extent->ram_size,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
829 * clear dirty, set writeback and unlock the pages.
831 extent_clear_unlock_delalloc(inode, async_extent->start,
832 async_extent->start +
833 async_extent->ram_size - 1,
834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
837 ret = btrfs_submit_compressed_write(inode,
839 async_extent->ram_size,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
856 free_async_extent_pages(async_extent);
858 alloc_hint = ins.objectid + ins.offset;
864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
866 extent_clear_unlock_delalloc(inode, async_extent->start,
867 async_extent->start +
868 async_extent->ram_size - 1,
869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
874 free_async_extent_pages(async_extent);
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
902 alloc_hint = em->block_start;
906 read_unlock(&em_tree->lock);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
930 struct btrfs_root *root = BTRFS_I(inode)->root;
933 unsigned long ram_size;
936 u64 blocksize = root->sectorsize;
937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942 if (btrfs_is_free_space_inode(inode)) {
948 num_bytes = ALIGN(end - start + 1, blocksize);
949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes < SZ_64K &&
954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955 btrfs_add_inode_defrag(NULL, inode);
958 /* lets try to make an inline extent */
959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
964 EXTENT_DEFRAG, PAGE_UNLOCK |
965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968 *nr_written = *nr_written +
969 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
972 } else if (ret < 0) {
977 BUG_ON(disk_num_bytes >
978 btrfs_super_total_bytes(root->fs_info->super_copy));
980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
983 while (disk_num_bytes > 0) {
986 cur_alloc_size = disk_num_bytes;
987 ret = btrfs_reserve_extent(root, cur_alloc_size,
988 root->sectorsize, 0, alloc_hint,
993 em = alloc_extent_map();
999 em->orig_start = em->start;
1000 ram_size = ins.offset;
1001 em->len = ins.offset;
1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
1005 em->block_start = ins.objectid;
1006 em->block_len = ins.offset;
1007 em->orig_block_len = ins.offset;
1008 em->ram_bytes = ram_size;
1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011 em->generation = -1;
1014 write_lock(&em_tree->lock);
1015 ret = add_extent_mapping(em_tree, em, 1);
1016 write_unlock(&em_tree->lock);
1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1021 btrfs_drop_extent_cache(inode, start,
1022 start + ram_size - 1, 0);
1027 cur_alloc_size = ins.offset;
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1031 goto out_drop_extent_cache;
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1038 goto out_drop_extent_cache;
1041 if (disk_num_bytes < cur_alloc_size)
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
1054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1058 disk_num_bytes -= cur_alloc_size;
1059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
1066 out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1080 * work queue call back to started compression on a file and pages
1082 static noinline void async_cow_start(struct btrfs_work *work)
1084 struct async_cow *async_cow;
1086 async_cow = container_of(work, struct async_cow, work);
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1091 if (num_added == 0) {
1092 btrfs_add_delayed_iput(async_cow->inode);
1093 async_cow->inode = NULL;
1098 * work queue call back to submit previously compressed pages
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1106 async_cow = container_of(work, struct async_cow, work);
1108 root = async_cow->root;
1109 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1113 * atomic_sub_return implies a barrier for waitqueue_active
1115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1120 if (async_cow->inode)
1121 submit_compressed_extents(async_cow->inode, async_cow);
1124 static noinline void async_cow_free(struct btrfs_work *work)
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
1128 if (async_cow->inode)
1129 btrfs_add_delayed_iput(async_cow->inode);
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1141 int limit = 10 * SZ_1M;
1143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
1145 while (start < end) {
1146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147 BUG_ON(!async_cow); /* -ENOMEM */
1148 async_cow->inode = igrab(inode);
1149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
1157 cur_end = min(end, start + SZ_512K - 1);
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1167 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1171 btrfs_queue_work(root->fs_info->delalloc_workers,
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1180 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195 u64 bytenr, u64 num_bytes)
1198 struct btrfs_ordered_sum *sums;
1201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202 bytenr + num_bytes - 1, &list, 0);
1203 if (ret == 0 && list_empty(&list))
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1218 * If no cow copies or snapshots exist, we write directly to the existing
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
1223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1227 struct btrfs_trans_handle *trans;
1228 struct extent_buffer *leaf;
1229 struct btrfs_path *path;
1230 struct btrfs_file_extent_item *fi;
1231 struct btrfs_key found_key;
1246 u64 ino = btrfs_ino(inode);
1248 path = btrfs_alloc_path();
1250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
1252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
1260 nolock = btrfs_is_free_space_inode(inode);
1263 trans = btrfs_join_transaction_nolock(root);
1265 trans = btrfs_join_transaction(root);
1267 if (IS_ERR(trans)) {
1268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
1270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
1275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1281 cow_start = (u64)-1;
1284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
1292 if (found_key.objectid == ino &&
1293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
1305 leaf = path->nodes[0];
1311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1313 if (found_key.objectid > ino)
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321 found_key.offset > end)
1324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
1330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
1334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338 extent_offset = btrfs_file_extent_offset(leaf, fi);
1339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343 if (extent_end <= start) {
1347 if (disk_bytenr == 0)
1349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1355 if (btrfs_extent_readonly(root, disk_bytenr))
1357 if (btrfs_cross_ref_exist(trans, root, ino,
1359 extent_offset, disk_bytenr))
1361 disk_bytenr += extent_offset;
1362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1369 err = btrfs_start_write_no_snapshoting(root);
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
1383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
1385 extent_end = ALIGN(extent_end, root->sectorsize);
1390 if (extent_end <= start) {
1392 if (!nolock && nocow)
1393 btrfs_end_write_no_snapshoting(root);
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1406 btrfs_release_path(path);
1407 if (cow_start != (u64)-1) {
1408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
1412 if (!nolock && nocow)
1413 btrfs_end_write_no_snapshoting(root);
1416 cow_start = (u64)-1;
1419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
1423 em = alloc_extent_map();
1424 BUG_ON(!em); /* -ENOMEM */
1425 em->start = cur_offset;
1426 em->orig_start = found_key.offset - extent_offset;
1427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
1430 em->orig_block_len = disk_num_bytes;
1431 em->ram_bytes = ram_bytes;
1432 em->bdev = root->fs_info->fs_devices->latest_bdev;
1433 em->mod_start = em->start;
1434 em->mod_len = em->len;
1435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437 em->generation = -1;
1439 write_lock(&em_tree->lock);
1440 ret = add_extent_mapping(em_tree, em, 1);
1441 write_unlock(&em_tree->lock);
1442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1449 type = BTRFS_ORDERED_PREALLOC;
1451 type = BTRFS_ORDERED_NOCOW;
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455 num_bytes, num_bytes, type);
1456 BUG_ON(ret); /* -ENOMEM */
1458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1463 if (!nolock && nocow)
1464 btrfs_end_write_no_snapshoting(root);
1469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1474 if (!nolock && nocow)
1475 btrfs_end_write_no_snapshoting(root);
1476 cur_offset = extent_end;
1477 if (cur_offset > end)
1480 btrfs_release_path(path);
1482 if (cur_offset <= end && cow_start == (u64)-1) {
1483 cow_start = cur_offset;
1487 if (cow_start != (u64)-1) {
1488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
1495 err = btrfs_end_transaction(trans, root);
1499 if (ret && cur_offset < end)
1500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
1502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
1507 btrfs_free_path(path);
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1532 * extent_io.c call back to do delayed allocation processing
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
1539 int force_cow = need_force_cow(inode, start, end);
1541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543 page_started, 1, nr_written);
1544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546 page_started, 0, nr_written);
1547 } else if (!inode_need_compress(inode)) {
1548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
1553 ret = cow_file_range_async(inode, locked_page, start, end,
1554 page_started, nr_written);
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
1564 /* not delalloc, ignore it */
1565 if (!(orig->state & EXTENT_DELALLOC))
1568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1577 new_size = orig->end - split + 1;
1578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579 BTRFS_MAX_EXTENT_SIZE);
1580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
1603 u64 new_size, old_size;
1606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
1610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1613 new_size = other->end - new->start + 1;
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1641 old_size = other->end - other->start + 1;
1642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
1644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1675 spin_unlock(&root->delalloc_lock);
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1694 spin_unlock(&root->delalloc_lock);
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703 struct extent_state *state, unsigned *bits)
1706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714 struct btrfs_root *root = BTRFS_I(inode)->root;
1715 u64 len = state->end + 1 - state->start;
1716 bool do_list = !btrfs_is_free_space_inode(inode);
1718 if (*bits & EXTENT_FIRST_DELALLOC) {
1719 *bits &= ~EXTENT_FIRST_DELALLOC;
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 BTRFS_I(inode)->delalloc_bytes += len;
1734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
1736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
1739 spin_unlock(&BTRFS_I(inode)->lock);
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747 struct extent_state *state,
1750 u64 len = state->end + 1 - state->start;
1751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
1766 bool do_list = !btrfs_is_free_space_inode(inode);
1768 if (*bits & EXTENT_FIRST_DELALLOC) {
1769 *bits &= ~EXTENT_FIRST_DELALLOC;
1770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
1772 BTRFS_I(inode)->outstanding_extents -= num_extents;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
1783 btrfs_delalloc_release_metadata(inode, len);
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list && !(state->state & EXTENT_NORESERVE))
1791 btrfs_free_reserved_data_space_noquota(inode,
1794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
1796 spin_lock(&BTRFS_I(inode)->lock);
1797 BTRFS_I(inode)->delalloc_bytes -= len;
1798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
1802 spin_unlock(&BTRFS_I(inode)->lock);
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1823 length = bio->bi_iter.bi_size;
1824 map_length = length;
1825 ret = btrfs_map_block(root->fs_info, rw, logical,
1826 &map_length, NULL, 0);
1827 /* Will always return 0 with map_multi == NULL */
1829 if (map_length < length + size)
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
1844 unsigned long bio_flags,
1847 struct btrfs_root *root = BTRFS_I(inode)->root;
1850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851 BUG_ON(ret); /* -ENOMEM */
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864 int mirror_num, unsigned long bio_flags,
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1872 bio->bi_error = ret;
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883 int mirror_num, unsigned long bio_flags,
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
1887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1894 if (btrfs_is_free_space_inode(inode))
1895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1897 if (!(rw & REQ_WRITE)) {
1898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903 ret = btrfs_submit_compressed_read(inode, bio,
1907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1913 } else if (async && !skip_sum) {
1914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1917 /* we're doing a write, do the async checksumming */
1918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919 inode, rw, bio, mirror_num,
1920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
1922 __btrfs_submit_bio_done);
1924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1935 bio->bi_error = ret;
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1949 struct btrfs_ordered_sum *sum;
1951 list_for_each_entry(sum, list, list) {
1952 trans->adding_csums = 1;
1953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955 trans->adding_csums = 0;
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
1963 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965 cached_state, GFP_NOFS);
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1971 struct btrfs_work work;
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
1978 struct extent_state *cached_state = NULL;
1980 struct inode *inode;
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
1996 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page))
2005 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2008 page_end, &cached_state, GFP_NOFS);
2010 btrfs_start_ordered_extent(inode, ordered, 1);
2011 btrfs_put_ordered_extent(ordered);
2015 ret = btrfs_delalloc_reserve_space(inode, page_start,
2018 mapping_set_error(page->mapping, ret);
2019 end_extent_writepage(page, ret, page_start, page_end);
2020 ClearPageChecked(page);
2024 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2025 ClearPageChecked(page);
2026 set_page_dirty(page);
2028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2029 &cached_state, GFP_NOFS);
2032 page_cache_release(page);
2037 * There are a few paths in the higher layers of the kernel that directly
2038 * set the page dirty bit without asking the filesystem if it is a
2039 * good idea. This causes problems because we want to make sure COW
2040 * properly happens and the data=ordered rules are followed.
2042 * In our case any range that doesn't have the ORDERED bit set
2043 * hasn't been properly setup for IO. We kick off an async process
2044 * to fix it up. The async helper will wait for ordered extents, set
2045 * the delalloc bit and make it safe to write the page.
2047 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2049 struct inode *inode = page->mapping->host;
2050 struct btrfs_writepage_fixup *fixup;
2051 struct btrfs_root *root = BTRFS_I(inode)->root;
2053 /* this page is properly in the ordered list */
2054 if (TestClearPagePrivate2(page))
2057 if (PageChecked(page))
2060 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2064 SetPageChecked(page);
2065 page_cache_get(page);
2066 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2067 btrfs_writepage_fixup_worker, NULL, NULL);
2069 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2073 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2074 struct inode *inode, u64 file_pos,
2075 u64 disk_bytenr, u64 disk_num_bytes,
2076 u64 num_bytes, u64 ram_bytes,
2077 u8 compression, u8 encryption,
2078 u16 other_encoding, int extent_type)
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2081 struct btrfs_file_extent_item *fi;
2082 struct btrfs_path *path;
2083 struct extent_buffer *leaf;
2084 struct btrfs_key ins;
2085 int extent_inserted = 0;
2088 path = btrfs_alloc_path();
2093 * we may be replacing one extent in the tree with another.
2094 * The new extent is pinned in the extent map, and we don't want
2095 * to drop it from the cache until it is completely in the btree.
2097 * So, tell btrfs_drop_extents to leave this extent in the cache.
2098 * the caller is expected to unpin it and allow it to be merged
2101 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2102 file_pos + num_bytes, NULL, 0,
2103 1, sizeof(*fi), &extent_inserted);
2107 if (!extent_inserted) {
2108 ins.objectid = btrfs_ino(inode);
2109 ins.offset = file_pos;
2110 ins.type = BTRFS_EXTENT_DATA_KEY;
2112 path->leave_spinning = 1;
2113 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2118 leaf = path->nodes[0];
2119 fi = btrfs_item_ptr(leaf, path->slots[0],
2120 struct btrfs_file_extent_item);
2121 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2122 btrfs_set_file_extent_type(leaf, fi, extent_type);
2123 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2124 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2125 btrfs_set_file_extent_offset(leaf, fi, 0);
2126 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2127 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2128 btrfs_set_file_extent_compression(leaf, fi, compression);
2129 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2130 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2132 btrfs_mark_buffer_dirty(leaf);
2133 btrfs_release_path(path);
2135 inode_add_bytes(inode, num_bytes);
2137 ins.objectid = disk_bytenr;
2138 ins.offset = disk_num_bytes;
2139 ins.type = BTRFS_EXTENT_ITEM_KEY;
2140 ret = btrfs_alloc_reserved_file_extent(trans, root,
2141 root->root_key.objectid,
2142 btrfs_ino(inode), file_pos,
2145 * Release the reserved range from inode dirty range map, as it is
2146 * already moved into delayed_ref_head
2148 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2150 btrfs_free_path(path);
2155 /* snapshot-aware defrag */
2156 struct sa_defrag_extent_backref {
2157 struct rb_node node;
2158 struct old_sa_defrag_extent *old;
2167 struct old_sa_defrag_extent {
2168 struct list_head list;
2169 struct new_sa_defrag_extent *new;
2178 struct new_sa_defrag_extent {
2179 struct rb_root root;
2180 struct list_head head;
2181 struct btrfs_path *path;
2182 struct inode *inode;
2190 static int backref_comp(struct sa_defrag_extent_backref *b1,
2191 struct sa_defrag_extent_backref *b2)
2193 if (b1->root_id < b2->root_id)
2195 else if (b1->root_id > b2->root_id)
2198 if (b1->inum < b2->inum)
2200 else if (b1->inum > b2->inum)
2203 if (b1->file_pos < b2->file_pos)
2205 else if (b1->file_pos > b2->file_pos)
2209 * [------------------------------] ===> (a range of space)
2210 * |<--->| |<---->| =============> (fs/file tree A)
2211 * |<---------------------------->| ===> (fs/file tree B)
2213 * A range of space can refer to two file extents in one tree while
2214 * refer to only one file extent in another tree.
2216 * So we may process a disk offset more than one time(two extents in A)
2217 * and locate at the same extent(one extent in B), then insert two same
2218 * backrefs(both refer to the extent in B).
2223 static void backref_insert(struct rb_root *root,
2224 struct sa_defrag_extent_backref *backref)
2226 struct rb_node **p = &root->rb_node;
2227 struct rb_node *parent = NULL;
2228 struct sa_defrag_extent_backref *entry;
2233 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235 ret = backref_comp(backref, entry);
2239 p = &(*p)->rb_right;
2242 rb_link_node(&backref->node, parent, p);
2243 rb_insert_color(&backref->node, root);
2247 * Note the backref might has changed, and in this case we just return 0.
2249 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2252 struct btrfs_file_extent_item *extent;
2253 struct btrfs_fs_info *fs_info;
2254 struct old_sa_defrag_extent *old = ctx;
2255 struct new_sa_defrag_extent *new = old->new;
2256 struct btrfs_path *path = new->path;
2257 struct btrfs_key key;
2258 struct btrfs_root *root;
2259 struct sa_defrag_extent_backref *backref;
2260 struct extent_buffer *leaf;
2261 struct inode *inode = new->inode;
2267 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2268 inum == btrfs_ino(inode))
2271 key.objectid = root_id;
2272 key.type = BTRFS_ROOT_ITEM_KEY;
2273 key.offset = (u64)-1;
2275 fs_info = BTRFS_I(inode)->root->fs_info;
2276 root = btrfs_read_fs_root_no_name(fs_info, &key);
2278 if (PTR_ERR(root) == -ENOENT)
2281 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282 inum, offset, root_id);
2283 return PTR_ERR(root);
2286 key.objectid = inum;
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2288 if (offset > (u64)-1 << 32)
2291 key.offset = offset;
2293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2294 if (WARN_ON(ret < 0))
2301 leaf = path->nodes[0];
2302 slot = path->slots[0];
2304 if (slot >= btrfs_header_nritems(leaf)) {
2305 ret = btrfs_next_leaf(root, path);
2308 } else if (ret > 0) {
2317 btrfs_item_key_to_cpu(leaf, &key, slot);
2319 if (key.objectid > inum)
2322 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2325 extent = btrfs_item_ptr(leaf, slot,
2326 struct btrfs_file_extent_item);
2328 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2332 * 'offset' refers to the exact key.offset,
2333 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334 * (key.offset - extent_offset).
2336 if (key.offset != offset)
2339 extent_offset = btrfs_file_extent_offset(leaf, extent);
2340 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2342 if (extent_offset >= old->extent_offset + old->offset +
2343 old->len || extent_offset + num_bytes <=
2344 old->extent_offset + old->offset)
2349 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2355 backref->root_id = root_id;
2356 backref->inum = inum;
2357 backref->file_pos = offset;
2358 backref->num_bytes = num_bytes;
2359 backref->extent_offset = extent_offset;
2360 backref->generation = btrfs_file_extent_generation(leaf, extent);
2362 backref_insert(&new->root, backref);
2365 btrfs_release_path(path);
2370 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2371 struct new_sa_defrag_extent *new)
2373 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2374 struct old_sa_defrag_extent *old, *tmp;
2379 list_for_each_entry_safe(old, tmp, &new->head, list) {
2380 ret = iterate_inodes_from_logical(old->bytenr +
2381 old->extent_offset, fs_info,
2382 path, record_one_backref,
2384 if (ret < 0 && ret != -ENOENT)
2387 /* no backref to be processed for this extent */
2389 list_del(&old->list);
2394 if (list_empty(&new->head))
2400 static int relink_is_mergable(struct extent_buffer *leaf,
2401 struct btrfs_file_extent_item *fi,
2402 struct new_sa_defrag_extent *new)
2404 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2407 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2410 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2413 if (btrfs_file_extent_encryption(leaf, fi) ||
2414 btrfs_file_extent_other_encoding(leaf, fi))
2421 * Note the backref might has changed, and in this case we just return 0.
2423 static noinline int relink_extent_backref(struct btrfs_path *path,
2424 struct sa_defrag_extent_backref *prev,
2425 struct sa_defrag_extent_backref *backref)
2427 struct btrfs_file_extent_item *extent;
2428 struct btrfs_file_extent_item *item;
2429 struct btrfs_ordered_extent *ordered;
2430 struct btrfs_trans_handle *trans;
2431 struct btrfs_fs_info *fs_info;
2432 struct btrfs_root *root;
2433 struct btrfs_key key;
2434 struct extent_buffer *leaf;
2435 struct old_sa_defrag_extent *old = backref->old;
2436 struct new_sa_defrag_extent *new = old->new;
2437 struct inode *src_inode = new->inode;
2438 struct inode *inode;
2439 struct extent_state *cached = NULL;
2448 if (prev && prev->root_id == backref->root_id &&
2449 prev->inum == backref->inum &&
2450 prev->file_pos + prev->num_bytes == backref->file_pos)
2453 /* step 1: get root */
2454 key.objectid = backref->root_id;
2455 key.type = BTRFS_ROOT_ITEM_KEY;
2456 key.offset = (u64)-1;
2458 fs_info = BTRFS_I(src_inode)->root->fs_info;
2459 index = srcu_read_lock(&fs_info->subvol_srcu);
2461 root = btrfs_read_fs_root_no_name(fs_info, &key);
2463 srcu_read_unlock(&fs_info->subvol_srcu, index);
2464 if (PTR_ERR(root) == -ENOENT)
2466 return PTR_ERR(root);
2469 if (btrfs_root_readonly(root)) {
2470 srcu_read_unlock(&fs_info->subvol_srcu, index);
2474 /* step 2: get inode */
2475 key.objectid = backref->inum;
2476 key.type = BTRFS_INODE_ITEM_KEY;
2479 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2480 if (IS_ERR(inode)) {
2481 srcu_read_unlock(&fs_info->subvol_srcu, index);
2485 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487 /* step 3: relink backref */
2488 lock_start = backref->file_pos;
2489 lock_end = backref->file_pos + backref->num_bytes - 1;
2490 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2493 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495 btrfs_put_ordered_extent(ordered);
2499 trans = btrfs_join_transaction(root);
2500 if (IS_ERR(trans)) {
2501 ret = PTR_ERR(trans);
2505 key.objectid = backref->inum;
2506 key.type = BTRFS_EXTENT_DATA_KEY;
2507 key.offset = backref->file_pos;
2509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512 } else if (ret > 0) {
2517 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2518 struct btrfs_file_extent_item);
2520 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2521 backref->generation)
2524 btrfs_release_path(path);
2526 start = backref->file_pos;
2527 if (backref->extent_offset < old->extent_offset + old->offset)
2528 start += old->extent_offset + old->offset -
2529 backref->extent_offset;
2531 len = min(backref->extent_offset + backref->num_bytes,
2532 old->extent_offset + old->offset + old->len);
2533 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535 ret = btrfs_drop_extents(trans, root, inode, start,
2540 key.objectid = btrfs_ino(inode);
2541 key.type = BTRFS_EXTENT_DATA_KEY;
2544 path->leave_spinning = 1;
2546 struct btrfs_file_extent_item *fi;
2548 struct btrfs_key found_key;
2550 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2555 leaf = path->nodes[0];
2556 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558 fi = btrfs_item_ptr(leaf, path->slots[0],
2559 struct btrfs_file_extent_item);
2560 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562 if (extent_len + found_key.offset == start &&
2563 relink_is_mergable(leaf, fi, new)) {
2564 btrfs_set_file_extent_num_bytes(leaf, fi,
2566 btrfs_mark_buffer_dirty(leaf);
2567 inode_add_bytes(inode, len);
2573 btrfs_release_path(path);
2578 ret = btrfs_insert_empty_item(trans, root, path, &key,
2581 btrfs_abort_transaction(trans, root, ret);
2585 leaf = path->nodes[0];
2586 item = btrfs_item_ptr(leaf, path->slots[0],
2587 struct btrfs_file_extent_item);
2588 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2589 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2590 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2591 btrfs_set_file_extent_num_bytes(leaf, item, len);
2592 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2593 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2594 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2595 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2596 btrfs_set_file_extent_encryption(leaf, item, 0);
2597 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599 btrfs_mark_buffer_dirty(leaf);
2600 inode_add_bytes(inode, len);
2601 btrfs_release_path(path);
2603 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605 backref->root_id, backref->inum,
2606 new->file_pos); /* start - extent_offset */
2608 btrfs_abort_transaction(trans, root, ret);
2614 btrfs_release_path(path);
2615 path->leave_spinning = 0;
2616 btrfs_end_transaction(trans, root);
2618 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2624 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626 struct old_sa_defrag_extent *old, *tmp;
2631 list_for_each_entry_safe(old, tmp, &new->head, list) {
2637 static void relink_file_extents(struct new_sa_defrag_extent *new)
2639 struct btrfs_path *path;
2640 struct sa_defrag_extent_backref *backref;
2641 struct sa_defrag_extent_backref *prev = NULL;
2642 struct inode *inode;
2643 struct btrfs_root *root;
2644 struct rb_node *node;
2648 root = BTRFS_I(inode)->root;
2650 path = btrfs_alloc_path();
2654 if (!record_extent_backrefs(path, new)) {
2655 btrfs_free_path(path);
2658 btrfs_release_path(path);
2661 node = rb_first(&new->root);
2664 rb_erase(node, &new->root);
2666 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668 ret = relink_extent_backref(path, prev, backref);
2681 btrfs_free_path(path);
2683 free_sa_defrag_extent(new);
2685 atomic_dec(&root->fs_info->defrag_running);
2686 wake_up(&root->fs_info->transaction_wait);
2689 static struct new_sa_defrag_extent *
2690 record_old_file_extents(struct inode *inode,
2691 struct btrfs_ordered_extent *ordered)
2693 struct btrfs_root *root = BTRFS_I(inode)->root;
2694 struct btrfs_path *path;
2695 struct btrfs_key key;
2696 struct old_sa_defrag_extent *old;
2697 struct new_sa_defrag_extent *new;
2700 new = kmalloc(sizeof(*new), GFP_NOFS);
2705 new->file_pos = ordered->file_offset;
2706 new->len = ordered->len;
2707 new->bytenr = ordered->start;
2708 new->disk_len = ordered->disk_len;
2709 new->compress_type = ordered->compress_type;
2710 new->root = RB_ROOT;
2711 INIT_LIST_HEAD(&new->head);
2713 path = btrfs_alloc_path();
2717 key.objectid = btrfs_ino(inode);
2718 key.type = BTRFS_EXTENT_DATA_KEY;
2719 key.offset = new->file_pos;
2721 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2724 if (ret > 0 && path->slots[0] > 0)
2727 /* find out all the old extents for the file range */
2729 struct btrfs_file_extent_item *extent;
2730 struct extent_buffer *l;
2739 slot = path->slots[0];
2741 if (slot >= btrfs_header_nritems(l)) {
2742 ret = btrfs_next_leaf(root, path);
2750 btrfs_item_key_to_cpu(l, &key, slot);
2752 if (key.objectid != btrfs_ino(inode))
2754 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756 if (key.offset >= new->file_pos + new->len)
2759 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2762 if (key.offset + num_bytes < new->file_pos)
2765 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2769 extent_offset = btrfs_file_extent_offset(l, extent);
2771 old = kmalloc(sizeof(*old), GFP_NOFS);
2775 offset = max(new->file_pos, key.offset);
2776 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778 old->bytenr = disk_bytenr;
2779 old->extent_offset = extent_offset;
2780 old->offset = offset - key.offset;
2781 old->len = end - offset;
2784 list_add_tail(&old->list, &new->head);
2790 btrfs_free_path(path);
2791 atomic_inc(&root->fs_info->defrag_running);
2796 btrfs_free_path(path);
2798 free_sa_defrag_extent(new);
2802 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2805 struct btrfs_block_group_cache *cache;
2807 cache = btrfs_lookup_block_group(root->fs_info, start);
2810 spin_lock(&cache->lock);
2811 cache->delalloc_bytes -= len;
2812 spin_unlock(&cache->lock);
2814 btrfs_put_block_group(cache);
2817 /* as ordered data IO finishes, this gets called so we can finish
2818 * an ordered extent if the range of bytes in the file it covers are
2821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2823 struct inode *inode = ordered_extent->inode;
2824 struct btrfs_root *root = BTRFS_I(inode)->root;
2825 struct btrfs_trans_handle *trans = NULL;
2826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2827 struct extent_state *cached_state = NULL;
2828 struct new_sa_defrag_extent *new = NULL;
2829 int compress_type = 0;
2831 u64 logical_len = ordered_extent->len;
2833 bool truncated = false;
2835 nolock = btrfs_is_free_space_inode(inode);
2837 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2842 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2843 ordered_extent->file_offset +
2844 ordered_extent->len - 1);
2846 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848 logical_len = ordered_extent->truncated_len;
2849 /* Truncated the entire extent, don't bother adding */
2854 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2855 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2858 * For mwrite(mmap + memset to write) case, we still reserve
2859 * space for NOCOW range.
2860 * As NOCOW won't cause a new delayed ref, just free the space
2862 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2863 ordered_extent->len);
2864 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866 trans = btrfs_join_transaction_nolock(root);
2868 trans = btrfs_join_transaction(root);
2869 if (IS_ERR(trans)) {
2870 ret = PTR_ERR(trans);
2874 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2875 ret = btrfs_update_inode_fallback(trans, root, inode);
2876 if (ret) /* -ENOMEM or corruption */
2877 btrfs_abort_transaction(trans, root, ret);
2881 lock_extent_bits(io_tree, ordered_extent->file_offset,
2882 ordered_extent->file_offset + ordered_extent->len - 1,
2885 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2886 ordered_extent->file_offset + ordered_extent->len - 1,
2887 EXTENT_DEFRAG, 1, cached_state);
2889 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2890 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2891 /* the inode is shared */
2892 new = record_old_file_extents(inode, ordered_extent);
2894 clear_extent_bit(io_tree, ordered_extent->file_offset,
2895 ordered_extent->file_offset + ordered_extent->len - 1,
2896 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2900 trans = btrfs_join_transaction_nolock(root);
2902 trans = btrfs_join_transaction(root);
2903 if (IS_ERR(trans)) {
2904 ret = PTR_ERR(trans);
2909 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2911 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2912 compress_type = ordered_extent->compress_type;
2913 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2914 BUG_ON(compress_type);
2915 ret = btrfs_mark_extent_written(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->file_offset +
2920 BUG_ON(root == root->fs_info->tree_root);
2921 ret = insert_reserved_file_extent(trans, inode,
2922 ordered_extent->file_offset,
2923 ordered_extent->start,
2924 ordered_extent->disk_len,
2925 logical_len, logical_len,
2926 compress_type, 0, 0,
2927 BTRFS_FILE_EXTENT_REG);
2929 btrfs_release_delalloc_bytes(root,
2930 ordered_extent->start,
2931 ordered_extent->disk_len);
2933 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2934 ordered_extent->file_offset, ordered_extent->len,
2937 btrfs_abort_transaction(trans, root, ret);
2941 add_pending_csums(trans, inode, ordered_extent->file_offset,
2942 &ordered_extent->list);
2944 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2945 ret = btrfs_update_inode_fallback(trans, root, inode);
2946 if (ret) { /* -ENOMEM or corruption */
2947 btrfs_abort_transaction(trans, root, ret);
2952 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2953 ordered_extent->file_offset +
2954 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2956 if (root != root->fs_info->tree_root)
2957 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2959 btrfs_end_transaction(trans, root);
2961 if (ret || truncated) {
2965 start = ordered_extent->file_offset + logical_len;
2967 start = ordered_extent->file_offset;
2968 end = ordered_extent->file_offset + ordered_extent->len - 1;
2969 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971 /* Drop the cache for the part of the extent we didn't write. */
2972 btrfs_drop_extent_cache(inode, start, end, 0);
2975 * If the ordered extent had an IOERR or something else went
2976 * wrong we need to return the space for this ordered extent
2977 * back to the allocator. We only free the extent in the
2978 * truncated case if we didn't write out the extent at all.
2980 if ((ret || !logical_len) &&
2981 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2982 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2983 btrfs_free_reserved_extent(root, ordered_extent->start,
2984 ordered_extent->disk_len, 1);
2989 * This needs to be done to make sure anybody waiting knows we are done
2990 * updating everything for this ordered extent.
2992 btrfs_remove_ordered_extent(inode, ordered_extent);
2994 /* for snapshot-aware defrag */
2997 free_sa_defrag_extent(new);
2998 atomic_dec(&root->fs_info->defrag_running);
3000 relink_file_extents(new);
3005 btrfs_put_ordered_extent(ordered_extent);
3006 /* once for the tree */
3007 btrfs_put_ordered_extent(ordered_extent);
3012 static void finish_ordered_fn(struct btrfs_work *work)
3014 struct btrfs_ordered_extent *ordered_extent;
3015 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3016 btrfs_finish_ordered_io(ordered_extent);
3019 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3020 struct extent_state *state, int uptodate)
3022 struct inode *inode = page->mapping->host;
3023 struct btrfs_root *root = BTRFS_I(inode)->root;
3024 struct btrfs_ordered_extent *ordered_extent = NULL;
3025 struct btrfs_workqueue *wq;
3026 btrfs_work_func_t func;
3028 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030 ClearPagePrivate2(page);
3031 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3032 end - start + 1, uptodate))
3035 if (btrfs_is_free_space_inode(inode)) {
3036 wq = root->fs_info->endio_freespace_worker;
3037 func = btrfs_freespace_write_helper;
3039 wq = root->fs_info->endio_write_workers;
3040 func = btrfs_endio_write_helper;
3043 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045 btrfs_queue_work(wq, &ordered_extent->work);
3050 static int __readpage_endio_check(struct inode *inode,
3051 struct btrfs_io_bio *io_bio,
3052 int icsum, struct page *page,
3053 int pgoff, u64 start, size_t len)
3059 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061 kaddr = kmap_atomic(page);
3062 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3063 btrfs_csum_final(csum, (char *)&csum);
3064 if (csum != csum_expected)
3067 kunmap_atomic(kaddr);
3070 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3071 "csum failed ino %llu off %llu csum %u expected csum %u",
3072 btrfs_ino(inode), start, csum, csum_expected);
3073 memset(kaddr + pgoff, 1, len);
3074 flush_dcache_page(page);
3075 kunmap_atomic(kaddr);
3076 if (csum_expected == 0)
3082 * when reads are done, we need to check csums to verify the data is correct
3083 * if there's a match, we allow the bio to finish. If not, the code in
3084 * extent_io.c will try to find good copies for us.
3086 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3087 u64 phy_offset, struct page *page,
3088 u64 start, u64 end, int mirror)
3090 size_t offset = start - page_offset(page);
3091 struct inode *inode = page->mapping->host;
3092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3093 struct btrfs_root *root = BTRFS_I(inode)->root;
3095 if (PageChecked(page)) {
3096 ClearPageChecked(page);
3100 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3103 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3104 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3105 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3110 phy_offset >>= inode->i_sb->s_blocksize_bits;
3111 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112 start, (size_t)(end - start + 1));
3115 void btrfs_add_delayed_iput(struct inode *inode)
3117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3118 struct btrfs_inode *binode = BTRFS_I(inode);
3120 if (atomic_add_unless(&inode->i_count, -1, 1))
3123 spin_lock(&fs_info->delayed_iput_lock);
3124 if (binode->delayed_iput_count == 0) {
3125 ASSERT(list_empty(&binode->delayed_iput));
3126 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3128 binode->delayed_iput_count++;
3130 spin_unlock(&fs_info->delayed_iput_lock);
3133 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3137 spin_lock(&fs_info->delayed_iput_lock);
3138 while (!list_empty(&fs_info->delayed_iputs)) {
3139 struct btrfs_inode *inode;
3141 inode = list_first_entry(&fs_info->delayed_iputs,
3142 struct btrfs_inode, delayed_iput);
3143 if (inode->delayed_iput_count) {
3144 inode->delayed_iput_count--;
3145 list_move_tail(&inode->delayed_iput,
3146 &fs_info->delayed_iputs);
3148 list_del_init(&inode->delayed_iput);
3150 spin_unlock(&fs_info->delayed_iput_lock);
3151 iput(&inode->vfs_inode);
3152 spin_lock(&fs_info->delayed_iput_lock);
3154 spin_unlock(&fs_info->delayed_iput_lock);
3158 * This is called in transaction commit time. If there are no orphan
3159 * files in the subvolume, it removes orphan item and frees block_rsv
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root)
3165 struct btrfs_block_rsv *block_rsv;
3168 if (atomic_read(&root->orphan_inodes) ||
3169 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3172 spin_lock(&root->orphan_lock);
3173 if (atomic_read(&root->orphan_inodes)) {
3174 spin_unlock(&root->orphan_lock);
3178 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179 spin_unlock(&root->orphan_lock);
3183 block_rsv = root->orphan_block_rsv;
3184 root->orphan_block_rsv = NULL;
3185 spin_unlock(&root->orphan_lock);
3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3188 btrfs_root_refs(&root->root_item) > 0) {
3189 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190 root->root_key.objectid);
3192 btrfs_abort_transaction(trans, root, ret);
3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3199 WARN_ON(block_rsv->size > 0);
3200 btrfs_free_block_rsv(root, block_rsv);
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3211 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3213 struct btrfs_root *root = BTRFS_I(inode)->root;
3214 struct btrfs_block_rsv *block_rsv = NULL;
3219 if (!root->orphan_block_rsv) {
3220 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3225 spin_lock(&root->orphan_lock);
3226 if (!root->orphan_block_rsv) {
3227 root->orphan_block_rsv = block_rsv;
3228 } else if (block_rsv) {
3229 btrfs_free_block_rsv(root, block_rsv);
3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234 &BTRFS_I(inode)->runtime_flags)) {
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3241 if (!xchg(&root->orphan_item_inserted, 1))
3247 atomic_inc(&root->orphan_inodes);
3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251 &BTRFS_I(inode)->runtime_flags))
3253 spin_unlock(&root->orphan_lock);
3255 /* grab metadata reservation from transaction handle */
3257 ret = btrfs_orphan_reserve_metadata(trans, inode);
3258 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3261 /* insert an orphan item to track this unlinked/truncated file */
3263 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3265 atomic_dec(&root->orphan_inodes);
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags);
3269 btrfs_orphan_release_metadata(inode);
3271 if (ret != -EEXIST) {
3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &BTRFS_I(inode)->runtime_flags);
3274 btrfs_abort_transaction(trans, root, ret);
3281 /* insert an orphan item to track subvolume contains orphan files */
3283 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284 root->root_key.objectid);
3285 if (ret && ret != -EEXIST) {
3286 btrfs_abort_transaction(trans, root, ret);
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3297 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298 struct inode *inode)
3300 struct btrfs_root *root = BTRFS_I(inode)->root;
3301 int delete_item = 0;
3302 int release_rsv = 0;
3305 spin_lock(&root->orphan_lock);
3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307 &BTRFS_I(inode)->runtime_flags))
3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311 &BTRFS_I(inode)->runtime_flags))
3313 spin_unlock(&root->orphan_lock);
3316 atomic_dec(&root->orphan_inodes);
3318 ret = btrfs_del_orphan_item(trans, root,
3323 btrfs_orphan_release_metadata(inode);
3329 * this cleans up any orphans that may be left on the list from the last use
3332 int btrfs_orphan_cleanup(struct btrfs_root *root)
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
3336 struct btrfs_key key, found_key;
3337 struct btrfs_trans_handle *trans;
3338 struct inode *inode;
3339 u64 last_objectid = 0;
3340 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3342 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3345 path = btrfs_alloc_path();
3350 path->reada = READA_BACK;
3352 key.objectid = BTRFS_ORPHAN_OBJECTID;
3353 key.type = BTRFS_ORPHAN_ITEM_KEY;
3354 key.offset = (u64)-1;
3357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3362 * if ret == 0 means we found what we were searching for, which
3363 * is weird, but possible, so only screw with path if we didn't
3364 * find the key and see if we have stuff that matches
3368 if (path->slots[0] == 0)
3373 /* pull out the item */
3374 leaf = path->nodes[0];
3375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3377 /* make sure the item matches what we want */
3378 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3380 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3383 /* release the path since we're done with it */
3384 btrfs_release_path(path);
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3392 if (found_key.offset == last_objectid) {
3393 btrfs_err(root->fs_info,
3394 "Error removing orphan entry, stopping orphan cleanup");
3399 last_objectid = found_key.offset;
3401 found_key.objectid = found_key.offset;
3402 found_key.type = BTRFS_INODE_ITEM_KEY;
3403 found_key.offset = 0;
3404 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3405 ret = PTR_ERR_OR_ZERO(inode);
3406 if (ret && ret != -ESTALE)
3409 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410 struct btrfs_root *dead_root;
3411 struct btrfs_fs_info *fs_info = root->fs_info;
3412 int is_dead_root = 0;
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3425 spin_lock(&fs_info->trans_lock);
3426 list_for_each_entry(dead_root, &fs_info->dead_roots,
3428 if (dead_root->root_key.objectid ==
3429 found_key.objectid) {
3434 spin_unlock(&fs_info->trans_lock);
3436 /* prevent this orphan from being found again */
3437 key.offset = found_key.objectid - 1;
3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
3445 if (ret == -ESTALE) {
3446 trans = btrfs_start_transaction(root, 1);
3447 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3451 btrfs_debug(root->fs_info, "auto deleting %Lu",
3452 found_key.objectid);
3453 ret = btrfs_del_orphan_item(trans, root,
3454 found_key.objectid);
3455 btrfs_end_transaction(trans, root);
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466 &BTRFS_I(inode)->runtime_flags);
3467 atomic_inc(&root->orphan_inodes);
3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode->i_nlink) {
3471 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3477 /* 1 for the orphan item deletion. */
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
3481 ret = PTR_ERR(trans);
3484 ret = btrfs_orphan_add(trans, inode);
3485 btrfs_end_transaction(trans, root);
3491 ret = btrfs_truncate(inode);
3493 btrfs_orphan_del(NULL, inode);
3498 /* this will do delete_inode and everything for us */
3503 /* release the path since we're done with it */
3504 btrfs_release_path(path);
3506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3508 if (root->orphan_block_rsv)
3509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3512 if (root->orphan_block_rsv ||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3514 trans = btrfs_join_transaction(root);
3516 btrfs_end_transaction(trans, root);
3520 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3522 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3526 btrfs_err(root->fs_info,
3527 "could not do orphan cleanup %d", ret);
3528 btrfs_free_path(path);
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3538 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3539 int slot, u64 objectid,
3540 int *first_xattr_slot)
3542 u32 nritems = btrfs_header_nritems(leaf);
3543 struct btrfs_key found_key;
3544 static u64 xattr_access = 0;
3545 static u64 xattr_default = 0;
3548 if (!xattr_access) {
3549 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3550 strlen(POSIX_ACL_XATTR_ACCESS));
3551 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3552 strlen(POSIX_ACL_XATTR_DEFAULT));
3556 *first_xattr_slot = -1;
3557 while (slot < nritems) {
3558 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key.objectid != objectid)
3564 /* we found an xattr, assume we've got an acl */
3565 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3566 if (*first_xattr_slot == -1)
3567 *first_xattr_slot = slot;
3568 if (found_key.offset == xattr_access ||
3569 found_key.offset == xattr_default)
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3577 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3596 if (*first_xattr_slot == -1)
3597 *first_xattr_slot = slot;
3602 * read an inode from the btree into the in-memory inode
3604 static void btrfs_read_locked_inode(struct inode *inode)
3606 struct btrfs_path *path;
3607 struct extent_buffer *leaf;
3608 struct btrfs_inode_item *inode_item;
3609 struct btrfs_root *root = BTRFS_I(inode)->root;
3610 struct btrfs_key location;
3615 bool filled = false;
3616 int first_xattr_slot;
3618 ret = btrfs_fill_inode(inode, &rdev);
3622 path = btrfs_alloc_path();
3626 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3628 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3632 leaf = path->nodes[0];
3637 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638 struct btrfs_inode_item);
3639 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3640 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3641 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3643 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3645 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3648 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3651 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3654 BTRFS_I(inode)->i_otime.tv_sec =
3655 btrfs_timespec_sec(leaf, &inode_item->otime);
3656 BTRFS_I(inode)->i_otime.tv_nsec =
3657 btrfs_timespec_nsec(leaf, &inode_item->otime);
3659 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3660 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3661 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3663 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664 inode->i_generation = BTRFS_I(inode)->generation;
3666 rdev = btrfs_inode_rdev(leaf, inode_item);
3668 BTRFS_I(inode)->index_cnt = (u64)-1;
3669 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
3681 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683 &BTRFS_I(inode)->runtime_flags);
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3694 * ln mydir/foo mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3700 * mount fs, triggers fsync log replay
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3712 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3715 if (inode->i_nlink != 1 ||
3716 path->slots[0] >= btrfs_header_nritems(leaf))
3719 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720 if (location.objectid != btrfs_ino(inode))
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 if (location.type == BTRFS_INODE_REF_KEY) {
3725 struct btrfs_inode_ref *ref;
3727 ref = (struct btrfs_inode_ref *)ptr;
3728 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730 struct btrfs_inode_extref *extref;
3732 extref = (struct btrfs_inode_extref *)ptr;
3733 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3741 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3742 btrfs_ino(inode), &first_xattr_slot);
3743 if (first_xattr_slot != -1) {
3744 path->slots[0] = first_xattr_slot;
3745 ret = btrfs_load_inode_props(inode, path);
3747 btrfs_err(root->fs_info,
3748 "error loading props for ino %llu (root %llu): %d",
3750 root->root_key.objectid, ret);
3752 btrfs_free_path(path);
3755 cache_no_acl(inode);
3757 switch (inode->i_mode & S_IFMT) {
3759 inode->i_mapping->a_ops = &btrfs_aops;
3760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3761 inode->i_fop = &btrfs_file_operations;
3762 inode->i_op = &btrfs_file_inode_operations;
3765 inode->i_fop = &btrfs_dir_file_operations;
3766 if (root == root->fs_info->tree_root)
3767 inode->i_op = &btrfs_dir_ro_inode_operations;
3769 inode->i_op = &btrfs_dir_inode_operations;
3772 inode->i_op = &btrfs_symlink_inode_operations;
3773 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3776 inode->i_op = &btrfs_special_inode_operations;
3777 init_special_inode(inode, inode->i_mode, rdev);
3781 btrfs_update_iflags(inode);
3785 btrfs_free_path(path);
3786 make_bad_inode(inode);
3790 * given a leaf and an inode, copy the inode fields into the leaf
3792 static void fill_inode_item(struct btrfs_trans_handle *trans,
3793 struct extent_buffer *leaf,
3794 struct btrfs_inode_item *item,
3795 struct inode *inode)
3797 struct btrfs_map_token token;
3799 btrfs_init_map_token(&token);
3801 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3802 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3803 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3806 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3808 btrfs_set_token_timespec_sec(leaf, &item->atime,
3809 inode->i_atime.tv_sec, &token);
3810 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3811 inode->i_atime.tv_nsec, &token);
3813 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3814 inode->i_mtime.tv_sec, &token);
3815 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3816 inode->i_mtime.tv_nsec, &token);
3818 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3819 inode->i_ctime.tv_sec, &token);
3820 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3821 inode->i_ctime.tv_nsec, &token);
3823 btrfs_set_token_timespec_sec(leaf, &item->otime,
3824 BTRFS_I(inode)->i_otime.tv_sec, &token);
3825 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3833 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3834 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3835 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3836 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3840 * copy everything in the in-memory inode into the btree.
3842 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3843 struct btrfs_root *root, struct inode *inode)
3845 struct btrfs_inode_item *inode_item;
3846 struct btrfs_path *path;
3847 struct extent_buffer *leaf;
3850 path = btrfs_alloc_path();
3854 path->leave_spinning = 1;
3855 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3863 leaf = path->nodes[0];
3864 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3865 struct btrfs_inode_item);
3867 fill_inode_item(trans, leaf, inode_item, inode);
3868 btrfs_mark_buffer_dirty(leaf);
3869 btrfs_set_inode_last_trans(trans, inode);
3872 btrfs_free_path(path);
3877 * copy everything in the in-memory inode into the btree.
3879 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, struct inode *inode)
3885 * If the inode is a free space inode, we can deadlock during commit
3886 * if we put it into the delayed code.
3888 * The data relocation inode should also be directly updated
3891 if (!btrfs_is_free_space_inode(inode)
3892 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3893 && !root->fs_info->log_root_recovering) {
3894 btrfs_update_root_times(trans, root);
3896 ret = btrfs_delayed_update_inode(trans, root, inode);
3898 btrfs_set_inode_last_trans(trans, inode);
3902 return btrfs_update_inode_item(trans, root, inode);
3905 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct inode *inode)
3911 ret = btrfs_update_inode(trans, root, inode);
3913 return btrfs_update_inode_item(trans, root, inode);
3918 * unlink helper that gets used here in inode.c and in the tree logging
3919 * recovery code. It remove a link in a directory with a given name, and
3920 * also drops the back refs in the inode to the directory
3922 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3923 struct btrfs_root *root,
3924 struct inode *dir, struct inode *inode,
3925 const char *name, int name_len)
3927 struct btrfs_path *path;
3929 struct extent_buffer *leaf;
3930 struct btrfs_dir_item *di;
3931 struct btrfs_key key;
3933 u64 ino = btrfs_ino(inode);
3934 u64 dir_ino = btrfs_ino(dir);
3936 path = btrfs_alloc_path();
3942 path->leave_spinning = 1;
3943 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3944 name, name_len, -1);
3953 leaf = path->nodes[0];
3954 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3955 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3958 btrfs_release_path(path);
3961 * If we don't have dir index, we have to get it by looking up
3962 * the inode ref, since we get the inode ref, remove it directly,
3963 * it is unnecessary to do delayed deletion.
3965 * But if we have dir index, needn't search inode ref to get it.
3966 * Since the inode ref is close to the inode item, it is better
3967 * that we delay to delete it, and just do this deletion when
3968 * we update the inode item.
3970 if (BTRFS_I(inode)->dir_index) {
3971 ret = btrfs_delayed_delete_inode_ref(inode);
3973 index = BTRFS_I(inode)->dir_index;
3978 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3981 btrfs_info(root->fs_info,
3982 "failed to delete reference to %.*s, inode %llu parent %llu",
3983 name_len, name, ino, dir_ino);
3984 btrfs_abort_transaction(trans, root, ret);
3988 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3990 btrfs_abort_transaction(trans, root, ret);
3994 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3996 if (ret != 0 && ret != -ENOENT) {
3997 btrfs_abort_transaction(trans, root, ret);
4001 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4006 btrfs_abort_transaction(trans, root, ret);
4008 btrfs_free_path(path);
4012 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4013 inode_inc_iversion(inode);
4014 inode_inc_iversion(dir);
4015 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4016 ret = btrfs_update_inode(trans, root, dir);
4021 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4022 struct btrfs_root *root,
4023 struct inode *dir, struct inode *inode,
4024 const char *name, int name_len)
4027 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4030 ret = btrfs_update_inode(trans, root, inode);
4036 * helper to start transaction for unlink and rmdir.
4038 * unlink and rmdir are special in btrfs, they do not always free space, so
4039 * if we cannot make our reservations the normal way try and see if there is
4040 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4041 * allow the unlink to occur.
4043 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4045 struct btrfs_root *root = BTRFS_I(dir)->root;
4048 * 1 for the possible orphan item
4049 * 1 for the dir item
4050 * 1 for the dir index
4051 * 1 for the inode ref
4054 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4057 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4059 struct btrfs_root *root = BTRFS_I(dir)->root;
4060 struct btrfs_trans_handle *trans;
4061 struct inode *inode = d_inode(dentry);
4064 trans = __unlink_start_trans(dir);
4066 return PTR_ERR(trans);
4068 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4070 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4071 dentry->d_name.name, dentry->d_name.len);
4075 if (inode->i_nlink == 0) {
4076 ret = btrfs_orphan_add(trans, inode);
4082 btrfs_end_transaction(trans, root);
4083 btrfs_btree_balance_dirty(root);
4087 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root,
4089 struct inode *dir, u64 objectid,
4090 const char *name, int name_len)
4092 struct btrfs_path *path;
4093 struct extent_buffer *leaf;
4094 struct btrfs_dir_item *di;
4095 struct btrfs_key key;
4098 u64 dir_ino = btrfs_ino(dir);
4100 path = btrfs_alloc_path();
4104 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4105 name, name_len, -1);
4106 if (IS_ERR_OR_NULL(di)) {
4114 leaf = path->nodes[0];
4115 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4116 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4117 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119 btrfs_abort_transaction(trans, root, ret);
4122 btrfs_release_path(path);
4124 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4125 objectid, root->root_key.objectid,
4126 dir_ino, &index, name, name_len);
4128 if (ret != -ENOENT) {
4129 btrfs_abort_transaction(trans, root, ret);
4132 di = btrfs_search_dir_index_item(root, path, dir_ino,
4134 if (IS_ERR_OR_NULL(di)) {
4139 btrfs_abort_transaction(trans, root, ret);
4143 leaf = path->nodes[0];
4144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4145 btrfs_release_path(path);
4148 btrfs_release_path(path);
4150 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4152 btrfs_abort_transaction(trans, root, ret);
4156 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4157 inode_inc_iversion(dir);
4158 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4159 ret = btrfs_update_inode_fallback(trans, root, dir);
4161 btrfs_abort_transaction(trans, root, ret);
4163 btrfs_free_path(path);
4167 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4169 struct inode *inode = d_inode(dentry);
4171 struct btrfs_root *root = BTRFS_I(dir)->root;
4172 struct btrfs_trans_handle *trans;
4174 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4176 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4179 trans = __unlink_start_trans(dir);
4181 return PTR_ERR(trans);
4183 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4184 err = btrfs_unlink_subvol(trans, root, dir,
4185 BTRFS_I(inode)->location.objectid,
4186 dentry->d_name.name,
4187 dentry->d_name.len);
4191 err = btrfs_orphan_add(trans, inode);
4195 /* now the directory is empty */
4196 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4197 dentry->d_name.name, dentry->d_name.len);
4199 btrfs_i_size_write(inode, 0);
4201 btrfs_end_transaction(trans, root);
4202 btrfs_btree_balance_dirty(root);
4207 static int truncate_space_check(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4213 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4214 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4215 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4217 trans->bytes_reserved += bytes_deleted;
4222 static int truncate_inline_extent(struct inode *inode,
4223 struct btrfs_path *path,
4224 struct btrfs_key *found_key,
4228 struct extent_buffer *leaf = path->nodes[0];
4229 int slot = path->slots[0];
4230 struct btrfs_file_extent_item *fi;
4231 u32 size = (u32)(new_size - found_key->offset);
4232 struct btrfs_root *root = BTRFS_I(inode)->root;
4234 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4236 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4237 loff_t offset = new_size;
4238 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4241 * Zero out the remaining of the last page of our inline extent,
4242 * instead of directly truncating our inline extent here - that
4243 * would be much more complex (decompressing all the data, then
4244 * compressing the truncated data, which might be bigger than
4245 * the size of the inline extent, resize the extent, etc).
4246 * We release the path because to get the page we might need to
4247 * read the extent item from disk (data not in the page cache).
4249 btrfs_release_path(path);
4250 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4253 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4254 size = btrfs_file_extent_calc_inline_size(size);
4255 btrfs_truncate_item(root, path, size, 1);
4257 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4258 inode_sub_bytes(inode, item_end + 1 - new_size);
4264 * this can truncate away extent items, csum items and directory items.
4265 * It starts at a high offset and removes keys until it can't find
4266 * any higher than new_size
4268 * csum items that cross the new i_size are truncated to the new size
4271 * min_type is the minimum key type to truncate down to. If set to 0, this
4272 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4274 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4275 struct btrfs_root *root,
4276 struct inode *inode,
4277 u64 new_size, u32 min_type)
4279 struct btrfs_path *path;
4280 struct extent_buffer *leaf;
4281 struct btrfs_file_extent_item *fi;
4282 struct btrfs_key key;
4283 struct btrfs_key found_key;
4284 u64 extent_start = 0;
4285 u64 extent_num_bytes = 0;
4286 u64 extent_offset = 0;
4288 u64 last_size = new_size;
4289 u32 found_type = (u8)-1;
4292 int pending_del_nr = 0;
4293 int pending_del_slot = 0;
4294 int extent_type = -1;
4297 u64 ino = btrfs_ino(inode);
4298 u64 bytes_deleted = 0;
4300 bool should_throttle = 0;
4301 bool should_end = 0;
4303 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4306 * for non-free space inodes and ref cows, we want to back off from
4309 if (!btrfs_is_free_space_inode(inode) &&
4310 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4313 path = btrfs_alloc_path();
4316 path->reada = READA_BACK;
4319 * We want to drop from the next block forward in case this new size is
4320 * not block aligned since we will be keeping the last block of the
4321 * extent just the way it is.
4323 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4324 root == root->fs_info->tree_root)
4325 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4326 root->sectorsize), (u64)-1, 0);
4329 * This function is also used to drop the items in the log tree before
4330 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4331 * it is used to drop the loged items. So we shouldn't kill the delayed
4334 if (min_type == 0 && root == BTRFS_I(inode)->root)
4335 btrfs_kill_delayed_inode_items(inode);
4338 key.offset = (u64)-1;
4343 * with a 16K leaf size and 128MB extents, you can actually queue
4344 * up a huge file in a single leaf. Most of the time that
4345 * bytes_deleted is > 0, it will be huge by the time we get here
4347 if (be_nice && bytes_deleted > SZ_32M) {
4348 if (btrfs_should_end_transaction(trans, root)) {
4355 path->leave_spinning = 1;
4356 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4363 /* there are no items in the tree for us to truncate, we're
4366 if (path->slots[0] == 0)
4373 leaf = path->nodes[0];
4374 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4375 found_type = found_key.type;
4377 if (found_key.objectid != ino)
4380 if (found_type < min_type)
4383 item_end = found_key.offset;
4384 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4385 fi = btrfs_item_ptr(leaf, path->slots[0],
4386 struct btrfs_file_extent_item);
4387 extent_type = btrfs_file_extent_type(leaf, fi);
4388 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4390 btrfs_file_extent_num_bytes(leaf, fi);
4391 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4392 item_end += btrfs_file_extent_inline_len(leaf,
4393 path->slots[0], fi);
4397 if (found_type > min_type) {
4400 if (item_end < new_size)
4402 if (found_key.offset >= new_size)
4408 /* FIXME, shrink the extent if the ref count is only 1 */
4409 if (found_type != BTRFS_EXTENT_DATA_KEY)
4413 last_size = found_key.offset;
4415 last_size = new_size;
4417 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4419 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4421 u64 orig_num_bytes =
4422 btrfs_file_extent_num_bytes(leaf, fi);
4423 extent_num_bytes = ALIGN(new_size -
4426 btrfs_set_file_extent_num_bytes(leaf, fi,
4428 num_dec = (orig_num_bytes -
4430 if (test_bit(BTRFS_ROOT_REF_COWS,
4433 inode_sub_bytes(inode, num_dec);
4434 btrfs_mark_buffer_dirty(leaf);
4437 btrfs_file_extent_disk_num_bytes(leaf,
4439 extent_offset = found_key.offset -
4440 btrfs_file_extent_offset(leaf, fi);
4442 /* FIXME blocksize != 4096 */
4443 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4444 if (extent_start != 0) {
4446 if (test_bit(BTRFS_ROOT_REF_COWS,
4448 inode_sub_bytes(inode, num_dec);
4451 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4453 * we can't truncate inline items that have had
4457 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4458 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4461 * Need to release path in order to truncate a
4462 * compressed extent. So delete any accumulated
4463 * extent items so far.
4465 if (btrfs_file_extent_compression(leaf, fi) !=
4466 BTRFS_COMPRESS_NONE && pending_del_nr) {
4467 err = btrfs_del_items(trans, root, path,
4471 btrfs_abort_transaction(trans,
4479 err = truncate_inline_extent(inode, path,
4484 btrfs_abort_transaction(trans,
4488 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4490 inode_sub_bytes(inode, item_end + 1 - new_size);
4495 if (!pending_del_nr) {
4496 /* no pending yet, add ourselves */
4497 pending_del_slot = path->slots[0];
4499 } else if (pending_del_nr &&
4500 path->slots[0] + 1 == pending_del_slot) {
4501 /* hop on the pending chunk */
4503 pending_del_slot = path->slots[0];
4510 should_throttle = 0;
4513 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4514 root == root->fs_info->tree_root)) {
4515 btrfs_set_path_blocking(path);
4516 bytes_deleted += extent_num_bytes;
4517 ret = btrfs_free_extent(trans, root, extent_start,
4518 extent_num_bytes, 0,
4519 btrfs_header_owner(leaf),
4520 ino, extent_offset);
4522 if (btrfs_should_throttle_delayed_refs(trans, root))
4523 btrfs_async_run_delayed_refs(root,
4524 trans->delayed_ref_updates * 2, 0);
4526 if (truncate_space_check(trans, root,
4527 extent_num_bytes)) {
4530 if (btrfs_should_throttle_delayed_refs(trans,
4532 should_throttle = 1;
4537 if (found_type == BTRFS_INODE_ITEM_KEY)
4540 if (path->slots[0] == 0 ||
4541 path->slots[0] != pending_del_slot ||
4542 should_throttle || should_end) {
4543 if (pending_del_nr) {
4544 ret = btrfs_del_items(trans, root, path,
4548 btrfs_abort_transaction(trans,
4554 btrfs_release_path(path);
4555 if (should_throttle) {
4556 unsigned long updates = trans->delayed_ref_updates;
4558 trans->delayed_ref_updates = 0;
4559 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4565 * if we failed to refill our space rsv, bail out
4566 * and let the transaction restart
4578 if (pending_del_nr) {
4579 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4582 btrfs_abort_transaction(trans, root, ret);
4585 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4586 btrfs_ordered_update_i_size(inode, last_size, NULL);
4588 btrfs_free_path(path);
4590 if (be_nice && bytes_deleted > SZ_32M) {
4591 unsigned long updates = trans->delayed_ref_updates;
4593 trans->delayed_ref_updates = 0;
4594 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4603 * btrfs_truncate_page - read, zero a chunk and write a page
4604 * @inode - inode that we're zeroing
4605 * @from - the offset to start zeroing
4606 * @len - the length to zero, 0 to zero the entire range respective to the
4608 * @front - zero up to the offset instead of from the offset on
4610 * This will find the page for the "from" offset and cow the page and zero the
4611 * part we want to zero. This is used with truncate and hole punching.
4613 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4616 struct address_space *mapping = inode->i_mapping;
4617 struct btrfs_root *root = BTRFS_I(inode)->root;
4618 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4619 struct btrfs_ordered_extent *ordered;
4620 struct extent_state *cached_state = NULL;
4622 u32 blocksize = root->sectorsize;
4623 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4624 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4626 gfp_t mask = btrfs_alloc_write_mask(mapping);
4631 if ((offset & (blocksize - 1)) == 0 &&
4632 (!len || ((len & (blocksize - 1)) == 0)))
4634 ret = btrfs_delalloc_reserve_space(inode,
4635 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4640 page = find_or_create_page(mapping, index, mask);
4642 btrfs_delalloc_release_space(inode,
4643 round_down(from, PAGE_CACHE_SIZE),
4649 page_start = page_offset(page);
4650 page_end = page_start + PAGE_CACHE_SIZE - 1;
4652 if (!PageUptodate(page)) {
4653 ret = btrfs_readpage(NULL, page);
4655 if (page->mapping != mapping) {
4657 page_cache_release(page);
4660 if (!PageUptodate(page)) {
4665 wait_on_page_writeback(page);
4667 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
4668 set_page_extent_mapped(page);
4670 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4672 unlock_extent_cached(io_tree, page_start, page_end,
4673 &cached_state, GFP_NOFS);
4675 page_cache_release(page);
4676 btrfs_start_ordered_extent(inode, ordered, 1);
4677 btrfs_put_ordered_extent(ordered);
4681 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4682 EXTENT_DIRTY | EXTENT_DELALLOC |
4683 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4684 0, 0, &cached_state, GFP_NOFS);
4686 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4689 unlock_extent_cached(io_tree, page_start, page_end,
4690 &cached_state, GFP_NOFS);
4694 if (offset != PAGE_CACHE_SIZE) {
4696 len = PAGE_CACHE_SIZE - offset;
4699 memset(kaddr, 0, offset);
4701 memset(kaddr + offset, 0, len);
4702 flush_dcache_page(page);
4705 ClearPageChecked(page);
4706 set_page_dirty(page);
4707 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4712 btrfs_delalloc_release_space(inode, page_start,
4715 page_cache_release(page);
4720 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4721 u64 offset, u64 len)
4723 struct btrfs_trans_handle *trans;
4727 * Still need to make sure the inode looks like it's been updated so
4728 * that any holes get logged if we fsync.
4730 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4731 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4732 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4733 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4738 * 1 - for the one we're dropping
4739 * 1 - for the one we're adding
4740 * 1 - for updating the inode.
4742 trans = btrfs_start_transaction(root, 3);
4744 return PTR_ERR(trans);
4746 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4748 btrfs_abort_transaction(trans, root, ret);
4749 btrfs_end_transaction(trans, root);
4753 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4754 0, 0, len, 0, len, 0, 0, 0);
4756 btrfs_abort_transaction(trans, root, ret);
4758 btrfs_update_inode(trans, root, inode);
4759 btrfs_end_transaction(trans, root);
4764 * This function puts in dummy file extents for the area we're creating a hole
4765 * for. So if we are truncating this file to a larger size we need to insert
4766 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4767 * the range between oldsize and size
4769 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4771 struct btrfs_root *root = BTRFS_I(inode)->root;
4772 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4773 struct extent_map *em = NULL;
4774 struct extent_state *cached_state = NULL;
4775 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4776 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4777 u64 block_end = ALIGN(size, root->sectorsize);
4784 * If our size started in the middle of a page we need to zero out the
4785 * rest of the page before we expand the i_size, otherwise we could
4786 * expose stale data.
4788 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4792 if (size <= hole_start)
4796 struct btrfs_ordered_extent *ordered;
4798 lock_extent_bits(io_tree, hole_start, block_end - 1,
4800 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4801 block_end - hole_start);
4804 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4805 &cached_state, GFP_NOFS);
4806 btrfs_start_ordered_extent(inode, ordered, 1);
4807 btrfs_put_ordered_extent(ordered);
4810 cur_offset = hole_start;
4812 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4813 block_end - cur_offset, 0);
4819 last_byte = min(extent_map_end(em), block_end);
4820 last_byte = ALIGN(last_byte , root->sectorsize);
4821 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4822 struct extent_map *hole_em;
4823 hole_size = last_byte - cur_offset;
4825 err = maybe_insert_hole(root, inode, cur_offset,
4829 btrfs_drop_extent_cache(inode, cur_offset,
4830 cur_offset + hole_size - 1, 0);
4831 hole_em = alloc_extent_map();
4833 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4834 &BTRFS_I(inode)->runtime_flags);
4837 hole_em->start = cur_offset;
4838 hole_em->len = hole_size;
4839 hole_em->orig_start = cur_offset;
4841 hole_em->block_start = EXTENT_MAP_HOLE;
4842 hole_em->block_len = 0;
4843 hole_em->orig_block_len = 0;
4844 hole_em->ram_bytes = hole_size;
4845 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4846 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4847 hole_em->generation = root->fs_info->generation;
4850 write_lock(&em_tree->lock);
4851 err = add_extent_mapping(em_tree, hole_em, 1);
4852 write_unlock(&em_tree->lock);
4855 btrfs_drop_extent_cache(inode, cur_offset,
4859 free_extent_map(hole_em);
4862 free_extent_map(em);
4864 cur_offset = last_byte;
4865 if (cur_offset >= block_end)
4868 free_extent_map(em);
4869 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4874 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4876 struct btrfs_root *root = BTRFS_I(inode)->root;
4877 struct btrfs_trans_handle *trans;
4878 loff_t oldsize = i_size_read(inode);
4879 loff_t newsize = attr->ia_size;
4880 int mask = attr->ia_valid;
4884 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4885 * special case where we need to update the times despite not having
4886 * these flags set. For all other operations the VFS set these flags
4887 * explicitly if it wants a timestamp update.
4889 if (newsize != oldsize) {
4890 inode_inc_iversion(inode);
4891 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4892 inode->i_ctime = inode->i_mtime =
4893 current_fs_time(inode->i_sb);
4896 if (newsize > oldsize) {
4897 truncate_pagecache(inode, newsize);
4899 * Don't do an expanding truncate while snapshoting is ongoing.
4900 * This is to ensure the snapshot captures a fully consistent
4901 * state of this file - if the snapshot captures this expanding
4902 * truncation, it must capture all writes that happened before
4905 btrfs_wait_for_snapshot_creation(root);
4906 ret = btrfs_cont_expand(inode, oldsize, newsize);
4908 btrfs_end_write_no_snapshoting(root);
4912 trans = btrfs_start_transaction(root, 1);
4913 if (IS_ERR(trans)) {
4914 btrfs_end_write_no_snapshoting(root);
4915 return PTR_ERR(trans);
4918 i_size_write(inode, newsize);
4919 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4920 ret = btrfs_update_inode(trans, root, inode);
4921 btrfs_end_write_no_snapshoting(root);
4922 btrfs_end_transaction(trans, root);
4926 * We're truncating a file that used to have good data down to
4927 * zero. Make sure it gets into the ordered flush list so that
4928 * any new writes get down to disk quickly.
4931 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4932 &BTRFS_I(inode)->runtime_flags);
4935 * 1 for the orphan item we're going to add
4936 * 1 for the orphan item deletion.
4938 trans = btrfs_start_transaction(root, 2);
4940 return PTR_ERR(trans);
4943 * We need to do this in case we fail at _any_ point during the
4944 * actual truncate. Once we do the truncate_setsize we could
4945 * invalidate pages which forces any outstanding ordered io to
4946 * be instantly completed which will give us extents that need
4947 * to be truncated. If we fail to get an orphan inode down we
4948 * could have left over extents that were never meant to live,
4949 * so we need to garuntee from this point on that everything
4950 * will be consistent.
4952 ret = btrfs_orphan_add(trans, inode);
4953 btrfs_end_transaction(trans, root);
4957 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4958 truncate_setsize(inode, newsize);
4960 /* Disable nonlocked read DIO to avoid the end less truncate */
4961 btrfs_inode_block_unlocked_dio(inode);
4962 inode_dio_wait(inode);
4963 btrfs_inode_resume_unlocked_dio(inode);
4965 ret = btrfs_truncate(inode);
4966 if (ret && inode->i_nlink) {
4970 * failed to truncate, disk_i_size is only adjusted down
4971 * as we remove extents, so it should represent the true
4972 * size of the inode, so reset the in memory size and
4973 * delete our orphan entry.
4975 trans = btrfs_join_transaction(root);
4976 if (IS_ERR(trans)) {
4977 btrfs_orphan_del(NULL, inode);
4980 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4981 err = btrfs_orphan_del(trans, inode);
4983 btrfs_abort_transaction(trans, root, err);
4984 btrfs_end_transaction(trans, root);
4991 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4993 struct inode *inode = d_inode(dentry);
4994 struct btrfs_root *root = BTRFS_I(inode)->root;
4997 if (btrfs_root_readonly(root))
5000 err = inode_change_ok(inode, attr);
5004 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5005 err = btrfs_setsize(inode, attr);
5010 if (attr->ia_valid) {
5011 setattr_copy(inode, attr);
5012 inode_inc_iversion(inode);
5013 err = btrfs_dirty_inode(inode);
5015 if (!err && attr->ia_valid & ATTR_MODE)
5016 err = posix_acl_chmod(inode, inode->i_mode);
5023 * While truncating the inode pages during eviction, we get the VFS calling
5024 * btrfs_invalidatepage() against each page of the inode. This is slow because
5025 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5026 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5027 * extent_state structures over and over, wasting lots of time.
5029 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5030 * those expensive operations on a per page basis and do only the ordered io
5031 * finishing, while we release here the extent_map and extent_state structures,
5032 * without the excessive merging and splitting.
5034 static void evict_inode_truncate_pages(struct inode *inode)
5036 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5037 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5038 struct rb_node *node;
5040 ASSERT(inode->i_state & I_FREEING);
5041 truncate_inode_pages_final(&inode->i_data);
5043 write_lock(&map_tree->lock);
5044 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5045 struct extent_map *em;
5047 node = rb_first(&map_tree->map);
5048 em = rb_entry(node, struct extent_map, rb_node);
5049 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5050 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5051 remove_extent_mapping(map_tree, em);
5052 free_extent_map(em);
5053 if (need_resched()) {
5054 write_unlock(&map_tree->lock);
5056 write_lock(&map_tree->lock);
5059 write_unlock(&map_tree->lock);
5062 * Keep looping until we have no more ranges in the io tree.
5063 * We can have ongoing bios started by readpages (called from readahead)
5064 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5065 * still in progress (unlocked the pages in the bio but did not yet
5066 * unlocked the ranges in the io tree). Therefore this means some
5067 * ranges can still be locked and eviction started because before
5068 * submitting those bios, which are executed by a separate task (work
5069 * queue kthread), inode references (inode->i_count) were not taken
5070 * (which would be dropped in the end io callback of each bio).
5071 * Therefore here we effectively end up waiting for those bios and
5072 * anyone else holding locked ranges without having bumped the inode's
5073 * reference count - if we don't do it, when they access the inode's
5074 * io_tree to unlock a range it may be too late, leading to an
5075 * use-after-free issue.
5077 spin_lock(&io_tree->lock);
5078 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5079 struct extent_state *state;
5080 struct extent_state *cached_state = NULL;
5084 node = rb_first(&io_tree->state);
5085 state = rb_entry(node, struct extent_state, rb_node);
5086 start = state->start;
5088 spin_unlock(&io_tree->lock);
5090 lock_extent_bits(io_tree, start, end, &cached_state);
5093 * If still has DELALLOC flag, the extent didn't reach disk,
5094 * and its reserved space won't be freed by delayed_ref.
5095 * So we need to free its reserved space here.
5096 * (Refer to comment in btrfs_invalidatepage, case 2)
5098 * Note, end is the bytenr of last byte, so we need + 1 here.
5100 if (state->state & EXTENT_DELALLOC)
5101 btrfs_qgroup_free_data(inode, start, end - start + 1);
5103 clear_extent_bit(io_tree, start, end,
5104 EXTENT_LOCKED | EXTENT_DIRTY |
5105 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5106 EXTENT_DEFRAG, 1, 1,
5107 &cached_state, GFP_NOFS);
5110 spin_lock(&io_tree->lock);
5112 spin_unlock(&io_tree->lock);
5115 void btrfs_evict_inode(struct inode *inode)
5117 struct btrfs_trans_handle *trans;
5118 struct btrfs_root *root = BTRFS_I(inode)->root;
5119 struct btrfs_block_rsv *rsv, *global_rsv;
5120 int steal_from_global = 0;
5121 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5124 trace_btrfs_inode_evict(inode);
5126 evict_inode_truncate_pages(inode);
5128 if (inode->i_nlink &&
5129 ((btrfs_root_refs(&root->root_item) != 0 &&
5130 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5131 btrfs_is_free_space_inode(inode)))
5134 if (is_bad_inode(inode)) {
5135 btrfs_orphan_del(NULL, inode);
5138 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5139 if (!special_file(inode->i_mode))
5140 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5142 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5144 if (root->fs_info->log_root_recovering) {
5145 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5146 &BTRFS_I(inode)->runtime_flags));
5150 if (inode->i_nlink > 0) {
5151 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5152 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5156 ret = btrfs_commit_inode_delayed_inode(inode);
5158 btrfs_orphan_del(NULL, inode);
5162 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5164 btrfs_orphan_del(NULL, inode);
5167 rsv->size = min_size;
5169 global_rsv = &root->fs_info->global_block_rsv;
5171 btrfs_i_size_write(inode, 0);
5174 * This is a bit simpler than btrfs_truncate since we've already
5175 * reserved our space for our orphan item in the unlink, so we just
5176 * need to reserve some slack space in case we add bytes and update
5177 * inode item when doing the truncate.
5180 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5181 BTRFS_RESERVE_FLUSH_LIMIT);
5184 * Try and steal from the global reserve since we will
5185 * likely not use this space anyway, we want to try as
5186 * hard as possible to get this to work.
5189 steal_from_global++;
5191 steal_from_global = 0;
5195 * steal_from_global == 0: we reserved stuff, hooray!
5196 * steal_from_global == 1: we didn't reserve stuff, boo!
5197 * steal_from_global == 2: we've committed, still not a lot of
5198 * room but maybe we'll have room in the global reserve this
5200 * steal_from_global == 3: abandon all hope!
5202 if (steal_from_global > 2) {
5203 btrfs_warn(root->fs_info,
5204 "Could not get space for a delete, will truncate on mount %d",
5206 btrfs_orphan_del(NULL, inode);
5207 btrfs_free_block_rsv(root, rsv);
5211 trans = btrfs_join_transaction(root);
5212 if (IS_ERR(trans)) {
5213 btrfs_orphan_del(NULL, inode);
5214 btrfs_free_block_rsv(root, rsv);
5219 * We can't just steal from the global reserve, we need tomake
5220 * sure there is room to do it, if not we need to commit and try
5223 if (steal_from_global) {
5224 if (!btrfs_check_space_for_delayed_refs(trans, root))
5225 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5232 * Couldn't steal from the global reserve, we have too much
5233 * pending stuff built up, commit the transaction and try it
5237 ret = btrfs_commit_transaction(trans, root);
5239 btrfs_orphan_del(NULL, inode);
5240 btrfs_free_block_rsv(root, rsv);
5245 steal_from_global = 0;
5248 trans->block_rsv = rsv;
5250 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5251 if (ret != -ENOSPC && ret != -EAGAIN)
5254 trans->block_rsv = &root->fs_info->trans_block_rsv;
5255 btrfs_end_transaction(trans, root);
5257 btrfs_btree_balance_dirty(root);
5260 btrfs_free_block_rsv(root, rsv);
5263 * Errors here aren't a big deal, it just means we leave orphan items
5264 * in the tree. They will be cleaned up on the next mount.
5267 trans->block_rsv = root->orphan_block_rsv;
5268 btrfs_orphan_del(trans, inode);
5270 btrfs_orphan_del(NULL, inode);
5273 trans->block_rsv = &root->fs_info->trans_block_rsv;
5274 if (!(root == root->fs_info->tree_root ||
5275 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5276 btrfs_return_ino(root, btrfs_ino(inode));
5278 btrfs_end_transaction(trans, root);
5279 btrfs_btree_balance_dirty(root);
5281 btrfs_remove_delayed_node(inode);
5286 * this returns the key found in the dir entry in the location pointer.
5287 * If no dir entries were found, location->objectid is 0.
5289 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5290 struct btrfs_key *location)
5292 const char *name = dentry->d_name.name;
5293 int namelen = dentry->d_name.len;
5294 struct btrfs_dir_item *di;
5295 struct btrfs_path *path;
5296 struct btrfs_root *root = BTRFS_I(dir)->root;
5299 path = btrfs_alloc_path();
5303 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5308 if (IS_ERR_OR_NULL(di))
5311 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5313 btrfs_free_path(path);
5316 location->objectid = 0;
5321 * when we hit a tree root in a directory, the btrfs part of the inode
5322 * needs to be changed to reflect the root directory of the tree root. This
5323 * is kind of like crossing a mount point.
5325 static int fixup_tree_root_location(struct btrfs_root *root,
5327 struct dentry *dentry,
5328 struct btrfs_key *location,
5329 struct btrfs_root **sub_root)
5331 struct btrfs_path *path;
5332 struct btrfs_root *new_root;
5333 struct btrfs_root_ref *ref;
5334 struct extent_buffer *leaf;
5335 struct btrfs_key key;
5339 path = btrfs_alloc_path();
5346 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5347 key.type = BTRFS_ROOT_REF_KEY;
5348 key.offset = location->objectid;
5350 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5358 leaf = path->nodes[0];
5359 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5360 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5361 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5364 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5365 (unsigned long)(ref + 1),
5366 dentry->d_name.len);
5370 btrfs_release_path(path);
5372 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5373 if (IS_ERR(new_root)) {
5374 err = PTR_ERR(new_root);
5378 *sub_root = new_root;
5379 location->objectid = btrfs_root_dirid(&new_root->root_item);
5380 location->type = BTRFS_INODE_ITEM_KEY;
5381 location->offset = 0;
5384 btrfs_free_path(path);
5388 static void inode_tree_add(struct inode *inode)
5390 struct btrfs_root *root = BTRFS_I(inode)->root;
5391 struct btrfs_inode *entry;
5393 struct rb_node *parent;
5394 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5395 u64 ino = btrfs_ino(inode);
5397 if (inode_unhashed(inode))
5400 spin_lock(&root->inode_lock);
5401 p = &root->inode_tree.rb_node;
5404 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5406 if (ino < btrfs_ino(&entry->vfs_inode))
5407 p = &parent->rb_left;
5408 else if (ino > btrfs_ino(&entry->vfs_inode))
5409 p = &parent->rb_right;
5411 WARN_ON(!(entry->vfs_inode.i_state &
5412 (I_WILL_FREE | I_FREEING)));
5413 rb_replace_node(parent, new, &root->inode_tree);
5414 RB_CLEAR_NODE(parent);
5415 spin_unlock(&root->inode_lock);
5419 rb_link_node(new, parent, p);
5420 rb_insert_color(new, &root->inode_tree);
5421 spin_unlock(&root->inode_lock);
5424 static void inode_tree_del(struct inode *inode)
5426 struct btrfs_root *root = BTRFS_I(inode)->root;
5429 spin_lock(&root->inode_lock);
5430 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5431 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5432 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5433 empty = RB_EMPTY_ROOT(&root->inode_tree);
5435 spin_unlock(&root->inode_lock);
5437 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5438 synchronize_srcu(&root->fs_info->subvol_srcu);
5439 spin_lock(&root->inode_lock);
5440 empty = RB_EMPTY_ROOT(&root->inode_tree);
5441 spin_unlock(&root->inode_lock);
5443 btrfs_add_dead_root(root);
5447 void btrfs_invalidate_inodes(struct btrfs_root *root)
5449 struct rb_node *node;
5450 struct rb_node *prev;
5451 struct btrfs_inode *entry;
5452 struct inode *inode;
5455 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5456 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5458 spin_lock(&root->inode_lock);
5460 node = root->inode_tree.rb_node;
5464 entry = rb_entry(node, struct btrfs_inode, rb_node);
5466 if (objectid < btrfs_ino(&entry->vfs_inode))
5467 node = node->rb_left;
5468 else if (objectid > btrfs_ino(&entry->vfs_inode))
5469 node = node->rb_right;
5475 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5476 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5480 prev = rb_next(prev);
5484 entry = rb_entry(node, struct btrfs_inode, rb_node);
5485 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5486 inode = igrab(&entry->vfs_inode);
5488 spin_unlock(&root->inode_lock);
5489 if (atomic_read(&inode->i_count) > 1)
5490 d_prune_aliases(inode);
5492 * btrfs_drop_inode will have it removed from
5493 * the inode cache when its usage count
5498 spin_lock(&root->inode_lock);
5502 if (cond_resched_lock(&root->inode_lock))
5505 node = rb_next(node);
5507 spin_unlock(&root->inode_lock);
5510 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5512 struct btrfs_iget_args *args = p;
5513 inode->i_ino = args->location->objectid;
5514 memcpy(&BTRFS_I(inode)->location, args->location,
5515 sizeof(*args->location));
5516 BTRFS_I(inode)->root = args->root;
5520 static int btrfs_find_actor(struct inode *inode, void *opaque)
5522 struct btrfs_iget_args *args = opaque;
5523 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5524 args->root == BTRFS_I(inode)->root;
5527 static struct inode *btrfs_iget_locked(struct super_block *s,
5528 struct btrfs_key *location,
5529 struct btrfs_root *root)
5531 struct inode *inode;
5532 struct btrfs_iget_args args;
5533 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5535 args.location = location;
5538 inode = iget5_locked(s, hashval, btrfs_find_actor,
5539 btrfs_init_locked_inode,
5544 /* Get an inode object given its location and corresponding root.
5545 * Returns in *is_new if the inode was read from disk
5547 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5548 struct btrfs_root *root, int *new)
5550 struct inode *inode;
5552 inode = btrfs_iget_locked(s, location, root);
5554 return ERR_PTR(-ENOMEM);
5556 if (inode->i_state & I_NEW) {
5557 btrfs_read_locked_inode(inode);
5558 if (!is_bad_inode(inode)) {
5559 inode_tree_add(inode);
5560 unlock_new_inode(inode);
5564 unlock_new_inode(inode);
5566 inode = ERR_PTR(-ESTALE);
5573 static struct inode *new_simple_dir(struct super_block *s,
5574 struct btrfs_key *key,
5575 struct btrfs_root *root)
5577 struct inode *inode = new_inode(s);
5580 return ERR_PTR(-ENOMEM);
5582 BTRFS_I(inode)->root = root;
5583 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5584 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5586 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5587 inode->i_op = &btrfs_dir_ro_inode_operations;
5588 inode->i_fop = &simple_dir_operations;
5589 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5590 inode->i_mtime = CURRENT_TIME;
5591 inode->i_atime = inode->i_mtime;
5592 inode->i_ctime = inode->i_mtime;
5593 BTRFS_I(inode)->i_otime = inode->i_mtime;
5598 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5600 struct inode *inode;
5601 struct btrfs_root *root = BTRFS_I(dir)->root;
5602 struct btrfs_root *sub_root = root;
5603 struct btrfs_key location;
5607 if (dentry->d_name.len > BTRFS_NAME_LEN)
5608 return ERR_PTR(-ENAMETOOLONG);
5610 ret = btrfs_inode_by_name(dir, dentry, &location);
5612 return ERR_PTR(ret);
5614 if (location.objectid == 0)
5615 return ERR_PTR(-ENOENT);
5617 if (location.type == BTRFS_INODE_ITEM_KEY) {
5618 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5622 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5624 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5625 ret = fixup_tree_root_location(root, dir, dentry,
5626 &location, &sub_root);
5629 inode = ERR_PTR(ret);
5631 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5633 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5635 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5637 if (!IS_ERR(inode) && root != sub_root) {
5638 down_read(&root->fs_info->cleanup_work_sem);
5639 if (!(inode->i_sb->s_flags & MS_RDONLY))
5640 ret = btrfs_orphan_cleanup(sub_root);
5641 up_read(&root->fs_info->cleanup_work_sem);
5644 inode = ERR_PTR(ret);
5651 static int btrfs_dentry_delete(const struct dentry *dentry)
5653 struct btrfs_root *root;
5654 struct inode *inode = d_inode(dentry);
5656 if (!inode && !IS_ROOT(dentry))
5657 inode = d_inode(dentry->d_parent);
5660 root = BTRFS_I(inode)->root;
5661 if (btrfs_root_refs(&root->root_item) == 0)
5664 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5670 static void btrfs_dentry_release(struct dentry *dentry)
5672 kfree(dentry->d_fsdata);
5675 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5678 struct inode *inode;
5680 inode = btrfs_lookup_dentry(dir, dentry);
5681 if (IS_ERR(inode)) {
5682 if (PTR_ERR(inode) == -ENOENT)
5685 return ERR_CAST(inode);
5688 return d_splice_alias(inode, dentry);
5691 unsigned char btrfs_filetype_table[] = {
5692 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5695 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5697 struct inode *inode = file_inode(file);
5698 struct btrfs_root *root = BTRFS_I(inode)->root;
5699 struct btrfs_item *item;
5700 struct btrfs_dir_item *di;
5701 struct btrfs_key key;
5702 struct btrfs_key found_key;
5703 struct btrfs_path *path;
5704 struct list_head ins_list;
5705 struct list_head del_list;
5707 struct extent_buffer *leaf;
5709 unsigned char d_type;
5714 int key_type = BTRFS_DIR_INDEX_KEY;
5718 int is_curr = 0; /* ctx->pos points to the current index? */
5720 /* FIXME, use a real flag for deciding about the key type */
5721 if (root->fs_info->tree_root == root)
5722 key_type = BTRFS_DIR_ITEM_KEY;
5724 if (!dir_emit_dots(file, ctx))
5727 path = btrfs_alloc_path();
5731 path->reada = READA_FORWARD;
5733 if (key_type == BTRFS_DIR_INDEX_KEY) {
5734 INIT_LIST_HEAD(&ins_list);
5735 INIT_LIST_HEAD(&del_list);
5736 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5739 key.type = key_type;
5740 key.offset = ctx->pos;
5741 key.objectid = btrfs_ino(inode);
5743 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5748 leaf = path->nodes[0];
5749 slot = path->slots[0];
5750 if (slot >= btrfs_header_nritems(leaf)) {
5751 ret = btrfs_next_leaf(root, path);
5759 item = btrfs_item_nr(slot);
5760 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5762 if (found_key.objectid != key.objectid)
5764 if (found_key.type != key_type)
5766 if (found_key.offset < ctx->pos)
5768 if (key_type == BTRFS_DIR_INDEX_KEY &&
5769 btrfs_should_delete_dir_index(&del_list,
5773 ctx->pos = found_key.offset;
5776 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5778 di_total = btrfs_item_size(leaf, item);
5780 while (di_cur < di_total) {
5781 struct btrfs_key location;
5783 if (verify_dir_item(root, leaf, di))
5786 name_len = btrfs_dir_name_len(leaf, di);
5787 if (name_len <= sizeof(tmp_name)) {
5788 name_ptr = tmp_name;
5790 name_ptr = kmalloc(name_len, GFP_NOFS);
5796 read_extent_buffer(leaf, name_ptr,
5797 (unsigned long)(di + 1), name_len);
5799 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5800 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5803 /* is this a reference to our own snapshot? If so
5806 * In contrast to old kernels, we insert the snapshot's
5807 * dir item and dir index after it has been created, so
5808 * we won't find a reference to our own snapshot. We
5809 * still keep the following code for backward
5812 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5813 location.objectid == root->root_key.objectid) {
5817 over = !dir_emit(ctx, name_ptr, name_len,
5818 location.objectid, d_type);
5821 if (name_ptr != tmp_name)
5826 di_len = btrfs_dir_name_len(leaf, di) +
5827 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5829 di = (struct btrfs_dir_item *)((char *)di + di_len);
5835 if (key_type == BTRFS_DIR_INDEX_KEY) {
5838 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5843 /* Reached end of directory/root. Bump pos past the last item. */
5847 * Stop new entries from being returned after we return the last
5850 * New directory entries are assigned a strictly increasing
5851 * offset. This means that new entries created during readdir
5852 * are *guaranteed* to be seen in the future by that readdir.
5853 * This has broken buggy programs which operate on names as
5854 * they're returned by readdir. Until we re-use freed offsets
5855 * we have this hack to stop new entries from being returned
5856 * under the assumption that they'll never reach this huge
5859 * This is being careful not to overflow 32bit loff_t unless the
5860 * last entry requires it because doing so has broken 32bit apps
5863 if (key_type == BTRFS_DIR_INDEX_KEY) {
5864 if (ctx->pos >= INT_MAX)
5865 ctx->pos = LLONG_MAX;
5872 if (key_type == BTRFS_DIR_INDEX_KEY)
5873 btrfs_put_delayed_items(&ins_list, &del_list);
5874 btrfs_free_path(path);
5878 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5880 struct btrfs_root *root = BTRFS_I(inode)->root;
5881 struct btrfs_trans_handle *trans;
5883 bool nolock = false;
5885 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5888 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5891 if (wbc->sync_mode == WB_SYNC_ALL) {
5893 trans = btrfs_join_transaction_nolock(root);
5895 trans = btrfs_join_transaction(root);
5897 return PTR_ERR(trans);
5898 ret = btrfs_commit_transaction(trans, root);
5904 * This is somewhat expensive, updating the tree every time the
5905 * inode changes. But, it is most likely to find the inode in cache.
5906 * FIXME, needs more benchmarking...there are no reasons other than performance
5907 * to keep or drop this code.
5909 static int btrfs_dirty_inode(struct inode *inode)
5911 struct btrfs_root *root = BTRFS_I(inode)->root;
5912 struct btrfs_trans_handle *trans;
5915 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5918 trans = btrfs_join_transaction(root);
5920 return PTR_ERR(trans);
5922 ret = btrfs_update_inode(trans, root, inode);
5923 if (ret && ret == -ENOSPC) {
5924 /* whoops, lets try again with the full transaction */
5925 btrfs_end_transaction(trans, root);
5926 trans = btrfs_start_transaction(root, 1);
5928 return PTR_ERR(trans);
5930 ret = btrfs_update_inode(trans, root, inode);
5932 btrfs_end_transaction(trans, root);
5933 if (BTRFS_I(inode)->delayed_node)
5934 btrfs_balance_delayed_items(root);
5940 * This is a copy of file_update_time. We need this so we can return error on
5941 * ENOSPC for updating the inode in the case of file write and mmap writes.
5943 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5946 struct btrfs_root *root = BTRFS_I(inode)->root;
5948 if (btrfs_root_readonly(root))
5951 if (flags & S_VERSION)
5952 inode_inc_iversion(inode);
5953 if (flags & S_CTIME)
5954 inode->i_ctime = *now;
5955 if (flags & S_MTIME)
5956 inode->i_mtime = *now;
5957 if (flags & S_ATIME)
5958 inode->i_atime = *now;
5959 return btrfs_dirty_inode(inode);
5963 * find the highest existing sequence number in a directory
5964 * and then set the in-memory index_cnt variable to reflect
5965 * free sequence numbers
5967 static int btrfs_set_inode_index_count(struct inode *inode)
5969 struct btrfs_root *root = BTRFS_I(inode)->root;
5970 struct btrfs_key key, found_key;
5971 struct btrfs_path *path;
5972 struct extent_buffer *leaf;
5975 key.objectid = btrfs_ino(inode);
5976 key.type = BTRFS_DIR_INDEX_KEY;
5977 key.offset = (u64)-1;
5979 path = btrfs_alloc_path();
5983 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5986 /* FIXME: we should be able to handle this */
5992 * MAGIC NUMBER EXPLANATION:
5993 * since we search a directory based on f_pos we have to start at 2
5994 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5995 * else has to start at 2
5997 if (path->slots[0] == 0) {
5998 BTRFS_I(inode)->index_cnt = 2;
6004 leaf = path->nodes[0];
6005 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6007 if (found_key.objectid != btrfs_ino(inode) ||
6008 found_key.type != BTRFS_DIR_INDEX_KEY) {
6009 BTRFS_I(inode)->index_cnt = 2;
6013 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6015 btrfs_free_path(path);
6020 * helper to find a free sequence number in a given directory. This current
6021 * code is very simple, later versions will do smarter things in the btree
6023 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6027 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6028 ret = btrfs_inode_delayed_dir_index_count(dir);
6030 ret = btrfs_set_inode_index_count(dir);
6036 *index = BTRFS_I(dir)->index_cnt;
6037 BTRFS_I(dir)->index_cnt++;
6042 static int btrfs_insert_inode_locked(struct inode *inode)
6044 struct btrfs_iget_args args;
6045 args.location = &BTRFS_I(inode)->location;
6046 args.root = BTRFS_I(inode)->root;
6048 return insert_inode_locked4(inode,
6049 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6050 btrfs_find_actor, &args);
6053 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6054 struct btrfs_root *root,
6056 const char *name, int name_len,
6057 u64 ref_objectid, u64 objectid,
6058 umode_t mode, u64 *index)
6060 struct inode *inode;
6061 struct btrfs_inode_item *inode_item;
6062 struct btrfs_key *location;
6063 struct btrfs_path *path;
6064 struct btrfs_inode_ref *ref;
6065 struct btrfs_key key[2];
6067 int nitems = name ? 2 : 1;
6071 path = btrfs_alloc_path();
6073 return ERR_PTR(-ENOMEM);
6075 inode = new_inode(root->fs_info->sb);
6077 btrfs_free_path(path);
6078 return ERR_PTR(-ENOMEM);
6082 * O_TMPFILE, set link count to 0, so that after this point,
6083 * we fill in an inode item with the correct link count.
6086 set_nlink(inode, 0);
6089 * we have to initialize this early, so we can reclaim the inode
6090 * number if we fail afterwards in this function.
6092 inode->i_ino = objectid;
6095 trace_btrfs_inode_request(dir);
6097 ret = btrfs_set_inode_index(dir, index);
6099 btrfs_free_path(path);
6101 return ERR_PTR(ret);
6107 * index_cnt is ignored for everything but a dir,
6108 * btrfs_get_inode_index_count has an explanation for the magic
6111 BTRFS_I(inode)->index_cnt = 2;
6112 BTRFS_I(inode)->dir_index = *index;
6113 BTRFS_I(inode)->root = root;
6114 BTRFS_I(inode)->generation = trans->transid;
6115 inode->i_generation = BTRFS_I(inode)->generation;
6118 * We could have gotten an inode number from somebody who was fsynced
6119 * and then removed in this same transaction, so let's just set full
6120 * sync since it will be a full sync anyway and this will blow away the
6121 * old info in the log.
6123 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6125 key[0].objectid = objectid;
6126 key[0].type = BTRFS_INODE_ITEM_KEY;
6129 sizes[0] = sizeof(struct btrfs_inode_item);
6133 * Start new inodes with an inode_ref. This is slightly more
6134 * efficient for small numbers of hard links since they will
6135 * be packed into one item. Extended refs will kick in if we
6136 * add more hard links than can fit in the ref item.
6138 key[1].objectid = objectid;
6139 key[1].type = BTRFS_INODE_REF_KEY;
6140 key[1].offset = ref_objectid;
6142 sizes[1] = name_len + sizeof(*ref);
6145 location = &BTRFS_I(inode)->location;
6146 location->objectid = objectid;
6147 location->offset = 0;
6148 location->type = BTRFS_INODE_ITEM_KEY;
6150 ret = btrfs_insert_inode_locked(inode);
6154 path->leave_spinning = 1;
6155 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6159 inode_init_owner(inode, dir, mode);
6160 inode_set_bytes(inode, 0);
6162 inode->i_mtime = CURRENT_TIME;
6163 inode->i_atime = inode->i_mtime;
6164 inode->i_ctime = inode->i_mtime;
6165 BTRFS_I(inode)->i_otime = inode->i_mtime;
6167 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6168 struct btrfs_inode_item);
6169 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6170 sizeof(*inode_item));
6171 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6174 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6175 struct btrfs_inode_ref);
6176 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6177 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6178 ptr = (unsigned long)(ref + 1);
6179 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6182 btrfs_mark_buffer_dirty(path->nodes[0]);
6183 btrfs_free_path(path);
6185 btrfs_inherit_iflags(inode, dir);
6187 if (S_ISREG(mode)) {
6188 if (btrfs_test_opt(root, NODATASUM))
6189 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6190 if (btrfs_test_opt(root, NODATACOW))
6191 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6192 BTRFS_INODE_NODATASUM;
6195 inode_tree_add(inode);
6197 trace_btrfs_inode_new(inode);
6198 btrfs_set_inode_last_trans(trans, inode);
6200 btrfs_update_root_times(trans, root);
6202 ret = btrfs_inode_inherit_props(trans, inode, dir);
6204 btrfs_err(root->fs_info,
6205 "error inheriting props for ino %llu (root %llu): %d",
6206 btrfs_ino(inode), root->root_key.objectid, ret);
6211 unlock_new_inode(inode);
6214 BTRFS_I(dir)->index_cnt--;
6215 btrfs_free_path(path);
6217 return ERR_PTR(ret);
6220 static inline u8 btrfs_inode_type(struct inode *inode)
6222 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6226 * utility function to add 'inode' into 'parent_inode' with
6227 * a give name and a given sequence number.
6228 * if 'add_backref' is true, also insert a backref from the
6229 * inode to the parent directory.
6231 int btrfs_add_link(struct btrfs_trans_handle *trans,
6232 struct inode *parent_inode, struct inode *inode,
6233 const char *name, int name_len, int add_backref, u64 index)
6236 struct btrfs_key key;
6237 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6238 u64 ino = btrfs_ino(inode);
6239 u64 parent_ino = btrfs_ino(parent_inode);
6241 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6242 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6245 key.type = BTRFS_INODE_ITEM_KEY;
6249 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6250 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6251 key.objectid, root->root_key.objectid,
6252 parent_ino, index, name, name_len);
6253 } else if (add_backref) {
6254 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6258 /* Nothing to clean up yet */
6262 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6264 btrfs_inode_type(inode), index);
6265 if (ret == -EEXIST || ret == -EOVERFLOW)
6268 btrfs_abort_transaction(trans, root, ret);
6272 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6274 inode_inc_iversion(parent_inode);
6275 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6276 ret = btrfs_update_inode(trans, root, parent_inode);
6278 btrfs_abort_transaction(trans, root, ret);
6282 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6285 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6286 key.objectid, root->root_key.objectid,
6287 parent_ino, &local_index, name, name_len);
6289 } else if (add_backref) {
6293 err = btrfs_del_inode_ref(trans, root, name, name_len,
6294 ino, parent_ino, &local_index);
6299 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6300 struct inode *dir, struct dentry *dentry,
6301 struct inode *inode, int backref, u64 index)
6303 int err = btrfs_add_link(trans, dir, inode,
6304 dentry->d_name.name, dentry->d_name.len,
6311 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6312 umode_t mode, dev_t rdev)
6314 struct btrfs_trans_handle *trans;
6315 struct btrfs_root *root = BTRFS_I(dir)->root;
6316 struct inode *inode = NULL;
6323 * 2 for inode item and ref
6325 * 1 for xattr if selinux is on
6327 trans = btrfs_start_transaction(root, 5);
6329 return PTR_ERR(trans);
6331 err = btrfs_find_free_ino(root, &objectid);
6335 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6336 dentry->d_name.len, btrfs_ino(dir), objectid,
6338 if (IS_ERR(inode)) {
6339 err = PTR_ERR(inode);
6344 * If the active LSM wants to access the inode during
6345 * d_instantiate it needs these. Smack checks to see
6346 * if the filesystem supports xattrs by looking at the
6349 inode->i_op = &btrfs_special_inode_operations;
6350 init_special_inode(inode, inode->i_mode, rdev);
6352 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6354 goto out_unlock_inode;
6356 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6358 goto out_unlock_inode;
6360 btrfs_update_inode(trans, root, inode);
6361 unlock_new_inode(inode);
6362 d_instantiate(dentry, inode);
6366 btrfs_end_transaction(trans, root);
6367 btrfs_balance_delayed_items(root);
6368 btrfs_btree_balance_dirty(root);
6370 inode_dec_link_count(inode);
6377 unlock_new_inode(inode);
6382 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6383 umode_t mode, bool excl)
6385 struct btrfs_trans_handle *trans;
6386 struct btrfs_root *root = BTRFS_I(dir)->root;
6387 struct inode *inode = NULL;
6388 int drop_inode_on_err = 0;
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);
6413 drop_inode_on_err = 1;
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_fop = &btrfs_file_operations;
6421 inode->i_op = &btrfs_file_inode_operations;
6422 inode->i_mapping->a_ops = &btrfs_aops;
6424 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6426 goto out_unlock_inode;
6428 err = btrfs_update_inode(trans, root, inode);
6430 goto out_unlock_inode;
6432 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6434 goto out_unlock_inode;
6436 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6437 unlock_new_inode(inode);
6438 d_instantiate(dentry, inode);
6441 btrfs_end_transaction(trans, root);
6442 if (err && drop_inode_on_err) {
6443 inode_dec_link_count(inode);
6446 btrfs_balance_delayed_items(root);
6447 btrfs_btree_balance_dirty(root);
6451 unlock_new_inode(inode);
6456 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6457 struct dentry *dentry)
6459 struct btrfs_trans_handle *trans = NULL;
6460 struct btrfs_root *root = BTRFS_I(dir)->root;
6461 struct inode *inode = d_inode(old_dentry);
6466 /* do not allow sys_link's with other subvols of the same device */
6467 if (root->objectid != BTRFS_I(inode)->root->objectid)
6470 if (inode->i_nlink >= BTRFS_LINK_MAX)
6473 err = btrfs_set_inode_index(dir, &index);
6478 * 2 items for inode and inode ref
6479 * 2 items for dir items
6480 * 1 item for parent inode
6482 trans = btrfs_start_transaction(root, 5);
6483 if (IS_ERR(trans)) {
6484 err = PTR_ERR(trans);
6489 /* There are several dir indexes for this inode, clear the cache. */
6490 BTRFS_I(inode)->dir_index = 0ULL;
6492 inode_inc_iversion(inode);
6493 inode->i_ctime = CURRENT_TIME;
6495 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6497 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6502 struct dentry *parent = dentry->d_parent;
6503 err = btrfs_update_inode(trans, root, inode);
6506 if (inode->i_nlink == 1) {
6508 * If new hard link count is 1, it's a file created
6509 * with open(2) O_TMPFILE flag.
6511 err = btrfs_orphan_del(trans, inode);
6515 d_instantiate(dentry, inode);
6516 btrfs_log_new_name(trans, inode, NULL, parent);
6519 btrfs_balance_delayed_items(root);
6522 btrfs_end_transaction(trans, root);
6524 inode_dec_link_count(inode);
6527 btrfs_btree_balance_dirty(root);
6531 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6533 struct inode *inode = NULL;
6534 struct btrfs_trans_handle *trans;
6535 struct btrfs_root *root = BTRFS_I(dir)->root;
6537 int drop_on_err = 0;
6542 * 2 items for inode and ref
6543 * 2 items for dir items
6544 * 1 for xattr if selinux is on
6546 trans = btrfs_start_transaction(root, 5);
6548 return PTR_ERR(trans);
6550 err = btrfs_find_free_ino(root, &objectid);
6554 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6555 dentry->d_name.len, btrfs_ino(dir), objectid,
6556 S_IFDIR | mode, &index);
6557 if (IS_ERR(inode)) {
6558 err = PTR_ERR(inode);
6563 /* these must be set before we unlock the inode */
6564 inode->i_op = &btrfs_dir_inode_operations;
6565 inode->i_fop = &btrfs_dir_file_operations;
6567 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6569 goto out_fail_inode;
6571 btrfs_i_size_write(inode, 0);
6572 err = btrfs_update_inode(trans, root, inode);
6574 goto out_fail_inode;
6576 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6577 dentry->d_name.len, 0, index);
6579 goto out_fail_inode;
6581 d_instantiate(dentry, inode);
6583 * mkdir is special. We're unlocking after we call d_instantiate
6584 * to avoid a race with nfsd calling d_instantiate.
6586 unlock_new_inode(inode);
6590 btrfs_end_transaction(trans, root);
6592 inode_dec_link_count(inode);
6595 btrfs_balance_delayed_items(root);
6596 btrfs_btree_balance_dirty(root);
6600 unlock_new_inode(inode);
6604 /* Find next extent map of a given extent map, caller needs to ensure locks */
6605 static struct extent_map *next_extent_map(struct extent_map *em)
6607 struct rb_node *next;
6609 next = rb_next(&em->rb_node);
6612 return container_of(next, struct extent_map, rb_node);
6615 static struct extent_map *prev_extent_map(struct extent_map *em)
6617 struct rb_node *prev;
6619 prev = rb_prev(&em->rb_node);
6622 return container_of(prev, struct extent_map, rb_node);
6625 /* helper for btfs_get_extent. Given an existing extent in the tree,
6626 * the existing extent is the nearest extent to map_start,
6627 * and an extent that you want to insert, deal with overlap and insert
6628 * the best fitted new extent into the tree.
6630 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6631 struct extent_map *existing,
6632 struct extent_map *em,
6635 struct extent_map *prev;
6636 struct extent_map *next;
6641 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6643 if (existing->start > map_start) {
6645 prev = prev_extent_map(next);
6648 next = next_extent_map(prev);
6651 start = prev ? extent_map_end(prev) : em->start;
6652 start = max_t(u64, start, em->start);
6653 end = next ? next->start : extent_map_end(em);
6654 end = min_t(u64, end, extent_map_end(em));
6655 start_diff = start - em->start;
6657 em->len = end - start;
6658 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6659 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6660 em->block_start += start_diff;
6661 em->block_len -= start_diff;
6663 return add_extent_mapping(em_tree, em, 0);
6666 static noinline int uncompress_inline(struct btrfs_path *path,
6668 size_t pg_offset, u64 extent_offset,
6669 struct btrfs_file_extent_item *item)
6672 struct extent_buffer *leaf = path->nodes[0];
6675 unsigned long inline_size;
6679 WARN_ON(pg_offset != 0);
6680 compress_type = btrfs_file_extent_compression(leaf, item);
6681 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6682 inline_size = btrfs_file_extent_inline_item_len(leaf,
6683 btrfs_item_nr(path->slots[0]));
6684 tmp = kmalloc(inline_size, GFP_NOFS);
6687 ptr = btrfs_file_extent_inline_start(item);
6689 read_extent_buffer(leaf, tmp, ptr, inline_size);
6691 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6692 ret = btrfs_decompress(compress_type, tmp, page,
6693 extent_offset, inline_size, max_size);
6699 * a bit scary, this does extent mapping from logical file offset to the disk.
6700 * the ugly parts come from merging extents from the disk with the in-ram
6701 * representation. This gets more complex because of the data=ordered code,
6702 * where the in-ram extents might be locked pending data=ordered completion.
6704 * This also copies inline extents directly into the page.
6707 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6708 size_t pg_offset, u64 start, u64 len,
6713 u64 extent_start = 0;
6715 u64 objectid = btrfs_ino(inode);
6717 struct btrfs_path *path = NULL;
6718 struct btrfs_root *root = BTRFS_I(inode)->root;
6719 struct btrfs_file_extent_item *item;
6720 struct extent_buffer *leaf;
6721 struct btrfs_key found_key;
6722 struct extent_map *em = NULL;
6723 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6724 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6725 struct btrfs_trans_handle *trans = NULL;
6726 const bool new_inline = !page || create;
6729 read_lock(&em_tree->lock);
6730 em = lookup_extent_mapping(em_tree, start, len);
6732 em->bdev = root->fs_info->fs_devices->latest_bdev;
6733 read_unlock(&em_tree->lock);
6736 if (em->start > start || em->start + em->len <= start)
6737 free_extent_map(em);
6738 else if (em->block_start == EXTENT_MAP_INLINE && page)
6739 free_extent_map(em);
6743 em = alloc_extent_map();
6748 em->bdev = root->fs_info->fs_devices->latest_bdev;
6749 em->start = EXTENT_MAP_HOLE;
6750 em->orig_start = EXTENT_MAP_HOLE;
6752 em->block_len = (u64)-1;
6755 path = btrfs_alloc_path();
6761 * Chances are we'll be called again, so go ahead and do
6764 path->reada = READA_FORWARD;
6767 ret = btrfs_lookup_file_extent(trans, root, path,
6768 objectid, start, trans != NULL);
6775 if (path->slots[0] == 0)
6780 leaf = path->nodes[0];
6781 item = btrfs_item_ptr(leaf, path->slots[0],
6782 struct btrfs_file_extent_item);
6783 /* are we inside the extent that was found? */
6784 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6785 found_type = found_key.type;
6786 if (found_key.objectid != objectid ||
6787 found_type != BTRFS_EXTENT_DATA_KEY) {
6789 * If we backup past the first extent we want to move forward
6790 * and see if there is an extent in front of us, otherwise we'll
6791 * say there is a hole for our whole search range which can
6798 found_type = btrfs_file_extent_type(leaf, item);
6799 extent_start = found_key.offset;
6800 if (found_type == BTRFS_FILE_EXTENT_REG ||
6801 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6802 extent_end = extent_start +
6803 btrfs_file_extent_num_bytes(leaf, item);
6804 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6806 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6807 extent_end = ALIGN(extent_start + size, root->sectorsize);
6810 if (start >= extent_end) {
6812 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6813 ret = btrfs_next_leaf(root, path);
6820 leaf = path->nodes[0];
6822 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6823 if (found_key.objectid != objectid ||
6824 found_key.type != BTRFS_EXTENT_DATA_KEY)
6826 if (start + len <= found_key.offset)
6828 if (start > found_key.offset)
6831 em->orig_start = start;
6832 em->len = found_key.offset - start;
6836 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6838 if (found_type == BTRFS_FILE_EXTENT_REG ||
6839 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6841 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6845 size_t extent_offset;
6851 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6852 extent_offset = page_offset(page) + pg_offset - extent_start;
6853 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6854 size - extent_offset);
6855 em->start = extent_start + extent_offset;
6856 em->len = ALIGN(copy_size, root->sectorsize);
6857 em->orig_block_len = em->len;
6858 em->orig_start = em->start;
6859 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6860 if (create == 0 && !PageUptodate(page)) {
6861 if (btrfs_file_extent_compression(leaf, item) !=
6862 BTRFS_COMPRESS_NONE) {
6863 ret = uncompress_inline(path, page, pg_offset,
6864 extent_offset, item);
6871 read_extent_buffer(leaf, map + pg_offset, ptr,
6873 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6874 memset(map + pg_offset + copy_size, 0,
6875 PAGE_CACHE_SIZE - pg_offset -
6880 flush_dcache_page(page);
6881 } else if (create && PageUptodate(page)) {
6885 free_extent_map(em);
6888 btrfs_release_path(path);
6889 trans = btrfs_join_transaction(root);
6892 return ERR_CAST(trans);
6896 write_extent_buffer(leaf, map + pg_offset, ptr,
6899 btrfs_mark_buffer_dirty(leaf);
6901 set_extent_uptodate(io_tree, em->start,
6902 extent_map_end(em) - 1, NULL, GFP_NOFS);
6907 em->orig_start = start;
6910 em->block_start = EXTENT_MAP_HOLE;
6911 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6913 btrfs_release_path(path);
6914 if (em->start > start || extent_map_end(em) <= start) {
6915 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6916 em->start, em->len, start, len);
6922 write_lock(&em_tree->lock);
6923 ret = add_extent_mapping(em_tree, em, 0);
6924 /* it is possible that someone inserted the extent into the tree
6925 * while we had the lock dropped. It is also possible that
6926 * an overlapping map exists in the tree
6928 if (ret == -EEXIST) {
6929 struct extent_map *existing;
6933 existing = search_extent_mapping(em_tree, start, len);
6935 * existing will always be non-NULL, since there must be
6936 * extent causing the -EEXIST.
6938 if (start >= extent_map_end(existing) ||
6939 start <= existing->start) {
6941 * The existing extent map is the one nearest to
6942 * the [start, start + len) range which overlaps
6944 err = merge_extent_mapping(em_tree, existing,
6946 free_extent_map(existing);
6948 free_extent_map(em);
6952 free_extent_map(em);
6957 write_unlock(&em_tree->lock);
6960 trace_btrfs_get_extent(root, em);
6962 btrfs_free_path(path);
6964 ret = btrfs_end_transaction(trans, root);
6969 free_extent_map(em);
6970 return ERR_PTR(err);
6972 BUG_ON(!em); /* Error is always set */
6976 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6977 size_t pg_offset, u64 start, u64 len,
6980 struct extent_map *em;
6981 struct extent_map *hole_em = NULL;
6982 u64 range_start = start;
6988 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6995 * - a pre-alloc extent,
6996 * there might actually be delalloc bytes behind it.
6998 if (em->block_start != EXTENT_MAP_HOLE &&
6999 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7005 /* check to see if we've wrapped (len == -1 or similar) */
7014 /* ok, we didn't find anything, lets look for delalloc */
7015 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7016 end, len, EXTENT_DELALLOC, 1);
7017 found_end = range_start + found;
7018 if (found_end < range_start)
7019 found_end = (u64)-1;
7022 * we didn't find anything useful, return
7023 * the original results from get_extent()
7025 if (range_start > end || found_end <= start) {
7031 /* adjust the range_start to make sure it doesn't
7032 * go backwards from the start they passed in
7034 range_start = max(start, range_start);
7035 found = found_end - range_start;
7038 u64 hole_start = start;
7041 em = alloc_extent_map();
7047 * when btrfs_get_extent can't find anything it
7048 * returns one huge hole
7050 * make sure what it found really fits our range, and
7051 * adjust to make sure it is based on the start from
7055 u64 calc_end = extent_map_end(hole_em);
7057 if (calc_end <= start || (hole_em->start > end)) {
7058 free_extent_map(hole_em);
7061 hole_start = max(hole_em->start, start);
7062 hole_len = calc_end - hole_start;
7066 if (hole_em && range_start > hole_start) {
7067 /* our hole starts before our delalloc, so we
7068 * have to return just the parts of the hole
7069 * that go until the delalloc starts
7071 em->len = min(hole_len,
7072 range_start - hole_start);
7073 em->start = hole_start;
7074 em->orig_start = hole_start;
7076 * don't adjust block start at all,
7077 * it is fixed at EXTENT_MAP_HOLE
7079 em->block_start = hole_em->block_start;
7080 em->block_len = hole_len;
7081 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7082 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7084 em->start = range_start;
7086 em->orig_start = range_start;
7087 em->block_start = EXTENT_MAP_DELALLOC;
7088 em->block_len = found;
7090 } else if (hole_em) {
7095 free_extent_map(hole_em);
7097 free_extent_map(em);
7098 return ERR_PTR(err);
7103 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7106 struct btrfs_root *root = BTRFS_I(inode)->root;
7107 struct extent_map *em;
7108 struct btrfs_key ins;
7112 alloc_hint = get_extent_allocation_hint(inode, start, len);
7113 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7114 alloc_hint, &ins, 1, 1);
7116 return ERR_PTR(ret);
7118 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7119 ins.offset, ins.offset, ins.offset, 0);
7121 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7125 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7126 ins.offset, ins.offset, 0);
7128 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7129 free_extent_map(em);
7130 return ERR_PTR(ret);
7137 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7138 * block must be cow'd
7140 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7141 u64 *orig_start, u64 *orig_block_len,
7144 struct btrfs_trans_handle *trans;
7145 struct btrfs_path *path;
7147 struct extent_buffer *leaf;
7148 struct btrfs_root *root = BTRFS_I(inode)->root;
7149 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7150 struct btrfs_file_extent_item *fi;
7151 struct btrfs_key key;
7158 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7160 path = btrfs_alloc_path();
7164 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7169 slot = path->slots[0];
7172 /* can't find the item, must cow */
7179 leaf = path->nodes[0];
7180 btrfs_item_key_to_cpu(leaf, &key, slot);
7181 if (key.objectid != btrfs_ino(inode) ||
7182 key.type != BTRFS_EXTENT_DATA_KEY) {
7183 /* not our file or wrong item type, must cow */
7187 if (key.offset > offset) {
7188 /* Wrong offset, must cow */
7192 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7193 found_type = btrfs_file_extent_type(leaf, fi);
7194 if (found_type != BTRFS_FILE_EXTENT_REG &&
7195 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7196 /* not a regular extent, must cow */
7200 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7203 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7204 if (extent_end <= offset)
7207 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7208 if (disk_bytenr == 0)
7211 if (btrfs_file_extent_compression(leaf, fi) ||
7212 btrfs_file_extent_encryption(leaf, fi) ||
7213 btrfs_file_extent_other_encoding(leaf, fi))
7216 backref_offset = btrfs_file_extent_offset(leaf, fi);
7219 *orig_start = key.offset - backref_offset;
7220 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7221 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7224 if (btrfs_extent_readonly(root, disk_bytenr))
7227 num_bytes = min(offset + *len, extent_end) - offset;
7228 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7231 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7232 ret = test_range_bit(io_tree, offset, range_end,
7233 EXTENT_DELALLOC, 0, NULL);
7240 btrfs_release_path(path);
7243 * look for other files referencing this extent, if we
7244 * find any we must cow
7246 trans = btrfs_join_transaction(root);
7247 if (IS_ERR(trans)) {
7252 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7253 key.offset - backref_offset, disk_bytenr);
7254 btrfs_end_transaction(trans, root);
7261 * adjust disk_bytenr and num_bytes to cover just the bytes
7262 * in this extent we are about to write. If there
7263 * are any csums in that range we have to cow in order
7264 * to keep the csums correct
7266 disk_bytenr += backref_offset;
7267 disk_bytenr += offset - key.offset;
7268 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7271 * all of the above have passed, it is safe to overwrite this extent
7277 btrfs_free_path(path);
7281 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7283 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7285 void **pagep = NULL;
7286 struct page *page = NULL;
7290 start_idx = start >> PAGE_CACHE_SHIFT;
7293 * end is the last byte in the last page. end == start is legal
7295 end_idx = end >> PAGE_CACHE_SHIFT;
7299 /* Most of the code in this while loop is lifted from
7300 * find_get_page. It's been modified to begin searching from a
7301 * page and return just the first page found in that range. If the
7302 * found idx is less than or equal to the end idx then we know that
7303 * a page exists. If no pages are found or if those pages are
7304 * outside of the range then we're fine (yay!) */
7305 while (page == NULL &&
7306 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7307 page = radix_tree_deref_slot(pagep);
7308 if (unlikely(!page))
7311 if (radix_tree_exception(page)) {
7312 if (radix_tree_deref_retry(page)) {
7317 * Otherwise, shmem/tmpfs must be storing a swap entry
7318 * here as an exceptional entry: so return it without
7319 * attempting to raise page count.
7322 break; /* TODO: Is this relevant for this use case? */
7325 if (!page_cache_get_speculative(page)) {
7331 * Has the page moved?
7332 * This is part of the lockless pagecache protocol. See
7333 * include/linux/pagemap.h for details.
7335 if (unlikely(page != *pagep)) {
7336 page_cache_release(page);
7342 if (page->index <= end_idx)
7344 page_cache_release(page);
7351 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7352 struct extent_state **cached_state, int writing)
7354 struct btrfs_ordered_extent *ordered;
7358 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7361 * We're concerned with the entire range that we're going to be
7362 * doing DIO to, so we need to make sure theres no ordered
7363 * extents in this range.
7365 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7366 lockend - lockstart + 1);
7369 * We need to make sure there are no buffered pages in this
7370 * range either, we could have raced between the invalidate in
7371 * generic_file_direct_write and locking the extent. The
7372 * invalidate needs to happen so that reads after a write do not
7377 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7380 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7381 cached_state, GFP_NOFS);
7384 btrfs_start_ordered_extent(inode, ordered, 1);
7385 btrfs_put_ordered_extent(ordered);
7388 * We could trigger writeback for this range (and wait
7389 * for it to complete) and then invalidate the pages for
7390 * this range (through invalidate_inode_pages2_range()),
7391 * but that can lead us to a deadlock with a concurrent
7392 * call to readpages() (a buffered read or a defrag call
7393 * triggered a readahead) on a page lock due to an
7394 * ordered dio extent we created before but did not have
7395 * yet a corresponding bio submitted (whence it can not
7396 * complete), which makes readpages() wait for that
7397 * ordered extent to complete while holding a lock on
7410 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7411 u64 len, u64 orig_start,
7412 u64 block_start, u64 block_len,
7413 u64 orig_block_len, u64 ram_bytes,
7416 struct extent_map_tree *em_tree;
7417 struct extent_map *em;
7418 struct btrfs_root *root = BTRFS_I(inode)->root;
7421 em_tree = &BTRFS_I(inode)->extent_tree;
7422 em = alloc_extent_map();
7424 return ERR_PTR(-ENOMEM);
7427 em->orig_start = orig_start;
7428 em->mod_start = start;
7431 em->block_len = block_len;
7432 em->block_start = block_start;
7433 em->bdev = root->fs_info->fs_devices->latest_bdev;
7434 em->orig_block_len = orig_block_len;
7435 em->ram_bytes = ram_bytes;
7436 em->generation = -1;
7437 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7438 if (type == BTRFS_ORDERED_PREALLOC)
7439 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7442 btrfs_drop_extent_cache(inode, em->start,
7443 em->start + em->len - 1, 0);
7444 write_lock(&em_tree->lock);
7445 ret = add_extent_mapping(em_tree, em, 1);
7446 write_unlock(&em_tree->lock);
7447 } while (ret == -EEXIST);
7450 free_extent_map(em);
7451 return ERR_PTR(ret);
7457 static void adjust_dio_outstanding_extents(struct inode *inode,
7458 struct btrfs_dio_data *dio_data,
7461 unsigned num_extents;
7463 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7464 BTRFS_MAX_EXTENT_SIZE);
7466 * If we have an outstanding_extents count still set then we're
7467 * within our reservation, otherwise we need to adjust our inode
7468 * counter appropriately.
7470 if (dio_data->outstanding_extents) {
7471 dio_data->outstanding_extents -= num_extents;
7473 spin_lock(&BTRFS_I(inode)->lock);
7474 BTRFS_I(inode)->outstanding_extents += num_extents;
7475 spin_unlock(&BTRFS_I(inode)->lock);
7479 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7480 struct buffer_head *bh_result, int create)
7482 struct extent_map *em;
7483 struct btrfs_root *root = BTRFS_I(inode)->root;
7484 struct extent_state *cached_state = NULL;
7485 struct btrfs_dio_data *dio_data = NULL;
7486 u64 start = iblock << inode->i_blkbits;
7487 u64 lockstart, lockend;
7488 u64 len = bh_result->b_size;
7489 int unlock_bits = EXTENT_LOCKED;
7493 unlock_bits |= EXTENT_DIRTY;
7495 len = min_t(u64, len, root->sectorsize);
7498 lockend = start + len - 1;
7500 if (current->journal_info) {
7502 * Need to pull our outstanding extents and set journal_info to NULL so
7503 * that anything that needs to check if there's a transction doesn't get
7506 dio_data = current->journal_info;
7507 current->journal_info = NULL;
7511 * If this errors out it's because we couldn't invalidate pagecache for
7512 * this range and we need to fallback to buffered.
7514 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7520 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7527 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7528 * io. INLINE is special, and we could probably kludge it in here, but
7529 * it's still buffered so for safety lets just fall back to the generic
7532 * For COMPRESSED we _have_ to read the entire extent in so we can
7533 * decompress it, so there will be buffering required no matter what we
7534 * do, so go ahead and fallback to buffered.
7536 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7537 * to buffered IO. Don't blame me, this is the price we pay for using
7540 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7541 em->block_start == EXTENT_MAP_INLINE) {
7542 free_extent_map(em);
7547 /* Just a good old fashioned hole, return */
7548 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7549 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7550 free_extent_map(em);
7555 * We don't allocate a new extent in the following cases
7557 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7559 * 2) The extent is marked as PREALLOC. We're good to go here and can
7560 * just use the extent.
7564 len = min(len, em->len - (start - em->start));
7565 lockstart = start + len;
7569 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7570 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7571 em->block_start != EXTENT_MAP_HOLE)) {
7573 u64 block_start, orig_start, orig_block_len, ram_bytes;
7575 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7576 type = BTRFS_ORDERED_PREALLOC;
7578 type = BTRFS_ORDERED_NOCOW;
7579 len = min(len, em->len - (start - em->start));
7580 block_start = em->block_start + (start - em->start);
7582 if (can_nocow_extent(inode, start, &len, &orig_start,
7583 &orig_block_len, &ram_bytes) == 1) {
7584 if (type == BTRFS_ORDERED_PREALLOC) {
7585 free_extent_map(em);
7586 em = create_pinned_em(inode, start, len,
7597 ret = btrfs_add_ordered_extent_dio(inode, start,
7598 block_start, len, len, type);
7600 free_extent_map(em);
7608 * this will cow the extent, reset the len in case we changed
7611 len = bh_result->b_size;
7612 free_extent_map(em);
7613 em = btrfs_new_extent_direct(inode, start, len);
7618 len = min(len, em->len - (start - em->start));
7620 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7622 bh_result->b_size = len;
7623 bh_result->b_bdev = em->bdev;
7624 set_buffer_mapped(bh_result);
7626 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7627 set_buffer_new(bh_result);
7630 * Need to update the i_size under the extent lock so buffered
7631 * readers will get the updated i_size when we unlock.
7633 if (start + len > i_size_read(inode))
7634 i_size_write(inode, start + len);
7636 adjust_dio_outstanding_extents(inode, dio_data, len);
7637 btrfs_free_reserved_data_space(inode, start, len);
7638 WARN_ON(dio_data->reserve < len);
7639 dio_data->reserve -= len;
7640 dio_data->unsubmitted_oe_range_end = start + len;
7641 current->journal_info = dio_data;
7645 * In the case of write we need to clear and unlock the entire range,
7646 * in the case of read we need to unlock only the end area that we
7647 * aren't using if there is any left over space.
7649 if (lockstart < lockend) {
7650 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7651 lockend, unlock_bits, 1, 0,
7652 &cached_state, GFP_NOFS);
7654 free_extent_state(cached_state);
7657 free_extent_map(em);
7662 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7663 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7666 current->journal_info = dio_data;
7668 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7669 * write less data then expected, so that we don't underflow our inode's
7670 * outstanding extents counter.
7672 if (create && dio_data)
7673 adjust_dio_outstanding_extents(inode, dio_data, len);
7678 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7679 int rw, int mirror_num)
7681 struct btrfs_root *root = BTRFS_I(inode)->root;
7684 BUG_ON(rw & REQ_WRITE);
7688 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7689 BTRFS_WQ_ENDIO_DIO_REPAIR);
7693 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7699 static int btrfs_check_dio_repairable(struct inode *inode,
7700 struct bio *failed_bio,
7701 struct io_failure_record *failrec,
7706 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7707 failrec->logical, failrec->len);
7708 if (num_copies == 1) {
7710 * we only have a single copy of the data, so don't bother with
7711 * all the retry and error correction code that follows. no
7712 * matter what the error is, it is very likely to persist.
7714 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7715 num_copies, failrec->this_mirror, failed_mirror);
7719 failrec->failed_mirror = failed_mirror;
7720 failrec->this_mirror++;
7721 if (failrec->this_mirror == failed_mirror)
7722 failrec->this_mirror++;
7724 if (failrec->this_mirror > num_copies) {
7725 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7726 num_copies, failrec->this_mirror, failed_mirror);
7733 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7734 struct page *page, u64 start, u64 end,
7735 int failed_mirror, bio_end_io_t *repair_endio,
7738 struct io_failure_record *failrec;
7744 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7746 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7750 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7753 free_io_failure(inode, failrec);
7757 if (failed_bio->bi_vcnt > 1)
7758 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7760 read_mode = READ_SYNC;
7762 isector = start - btrfs_io_bio(failed_bio)->logical;
7763 isector >>= inode->i_sb->s_blocksize_bits;
7764 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7765 0, isector, repair_endio, repair_arg);
7767 free_io_failure(inode, failrec);
7771 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7772 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7773 read_mode, failrec->this_mirror, failrec->in_validation);
7775 ret = submit_dio_repair_bio(inode, bio, read_mode,
7776 failrec->this_mirror);
7778 free_io_failure(inode, failrec);
7785 struct btrfs_retry_complete {
7786 struct completion done;
7787 struct inode *inode;
7792 static void btrfs_retry_endio_nocsum(struct bio *bio)
7794 struct btrfs_retry_complete *done = bio->bi_private;
7795 struct bio_vec *bvec;
7802 bio_for_each_segment_all(bvec, bio, i)
7803 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7805 complete(&done->done);
7809 static int __btrfs_correct_data_nocsum(struct inode *inode,
7810 struct btrfs_io_bio *io_bio)
7812 struct bio_vec *bvec;
7813 struct btrfs_retry_complete done;
7818 start = io_bio->logical;
7821 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7825 init_completion(&done.done);
7827 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7828 start + bvec->bv_len - 1,
7830 btrfs_retry_endio_nocsum, &done);
7834 wait_for_completion(&done.done);
7836 if (!done.uptodate) {
7837 /* We might have another mirror, so try again */
7841 start += bvec->bv_len;
7847 static void btrfs_retry_endio(struct bio *bio)
7849 struct btrfs_retry_complete *done = bio->bi_private;
7850 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7851 struct bio_vec *bvec;
7860 bio_for_each_segment_all(bvec, bio, i) {
7861 ret = __readpage_endio_check(done->inode, io_bio, i,
7863 done->start, bvec->bv_len);
7865 clean_io_failure(done->inode, done->start,
7871 done->uptodate = uptodate;
7873 complete(&done->done);
7877 static int __btrfs_subio_endio_read(struct inode *inode,
7878 struct btrfs_io_bio *io_bio, int err)
7880 struct bio_vec *bvec;
7881 struct btrfs_retry_complete done;
7888 start = io_bio->logical;
7891 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7892 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7893 0, start, bvec->bv_len);
7899 init_completion(&done.done);
7901 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7902 start + bvec->bv_len - 1,
7904 btrfs_retry_endio, &done);
7910 wait_for_completion(&done.done);
7912 if (!done.uptodate) {
7913 /* We might have another mirror, so try again */
7917 offset += bvec->bv_len;
7918 start += bvec->bv_len;
7924 static int btrfs_subio_endio_read(struct inode *inode,
7925 struct btrfs_io_bio *io_bio, int err)
7927 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7931 return __btrfs_correct_data_nocsum(inode, io_bio);
7935 return __btrfs_subio_endio_read(inode, io_bio, err);
7939 static void btrfs_endio_direct_read(struct bio *bio)
7941 struct btrfs_dio_private *dip = bio->bi_private;
7942 struct inode *inode = dip->inode;
7943 struct bio *dio_bio;
7944 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7945 int err = bio->bi_error;
7947 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7948 err = btrfs_subio_endio_read(inode, io_bio, err);
7950 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7951 dip->logical_offset + dip->bytes - 1);
7952 dio_bio = dip->dio_bio;
7956 dio_end_io(dio_bio, bio->bi_error);
7959 io_bio->end_io(io_bio, err);
7963 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
7968 struct btrfs_root *root = BTRFS_I(inode)->root;
7969 struct btrfs_ordered_extent *ordered = NULL;
7970 u64 ordered_offset = offset;
7971 u64 ordered_bytes = bytes;
7975 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7982 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7983 finish_ordered_fn, NULL, NULL);
7984 btrfs_queue_work(root->fs_info->endio_write_workers,
7988 * our bio might span multiple ordered extents. If we haven't
7989 * completed the accounting for the whole dio, go back and try again
7991 if (ordered_offset < offset + bytes) {
7992 ordered_bytes = offset + bytes - ordered_offset;
7998 static void btrfs_endio_direct_write(struct bio *bio)
8000 struct btrfs_dio_private *dip = bio->bi_private;
8001 struct bio *dio_bio = dip->dio_bio;
8003 btrfs_endio_direct_write_update_ordered(dip->inode,
8004 dip->logical_offset,
8010 dio_end_io(dio_bio, bio->bi_error);
8014 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8015 struct bio *bio, int mirror_num,
8016 unsigned long bio_flags, u64 offset)
8019 struct btrfs_root *root = BTRFS_I(inode)->root;
8020 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8021 BUG_ON(ret); /* -ENOMEM */
8025 static void btrfs_end_dio_bio(struct bio *bio)
8027 struct btrfs_dio_private *dip = bio->bi_private;
8028 int err = bio->bi_error;
8031 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8032 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8033 btrfs_ino(dip->inode), bio->bi_rw,
8034 (unsigned long long)bio->bi_iter.bi_sector,
8035 bio->bi_iter.bi_size, err);
8037 if (dip->subio_endio)
8038 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8044 * before atomic variable goto zero, we must make sure
8045 * dip->errors is perceived to be set.
8047 smp_mb__before_atomic();
8050 /* if there are more bios still pending for this dio, just exit */
8051 if (!atomic_dec_and_test(&dip->pending_bios))
8055 bio_io_error(dip->orig_bio);
8057 dip->dio_bio->bi_error = 0;
8058 bio_endio(dip->orig_bio);
8064 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8065 u64 first_sector, gfp_t gfp_flags)
8068 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8070 bio_associate_current(bio);
8074 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8075 struct inode *inode,
8076 struct btrfs_dio_private *dip,
8080 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8081 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8085 * We load all the csum data we need when we submit
8086 * the first bio to reduce the csum tree search and
8089 if (dip->logical_offset == file_offset) {
8090 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8096 if (bio == dip->orig_bio)
8099 file_offset -= dip->logical_offset;
8100 file_offset >>= inode->i_sb->s_blocksize_bits;
8101 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8106 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8107 int rw, u64 file_offset, int skip_sum,
8110 struct btrfs_dio_private *dip = bio->bi_private;
8111 int write = rw & REQ_WRITE;
8112 struct btrfs_root *root = BTRFS_I(inode)->root;
8116 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8121 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8122 BTRFS_WQ_ENDIO_DATA);
8130 if (write && async_submit) {
8131 ret = btrfs_wq_submit_bio(root->fs_info,
8132 inode, rw, bio, 0, 0,
8134 __btrfs_submit_bio_start_direct_io,
8135 __btrfs_submit_bio_done);
8139 * If we aren't doing async submit, calculate the csum of the
8142 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8146 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8152 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8158 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8161 struct inode *inode = dip->inode;
8162 struct btrfs_root *root = BTRFS_I(inode)->root;
8164 struct bio *orig_bio = dip->orig_bio;
8165 struct bio_vec *bvec = orig_bio->bi_io_vec;
8166 u64 start_sector = orig_bio->bi_iter.bi_sector;
8167 u64 file_offset = dip->logical_offset;
8172 int async_submit = 0;
8174 map_length = orig_bio->bi_iter.bi_size;
8175 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8176 &map_length, NULL, 0);
8180 if (map_length >= orig_bio->bi_iter.bi_size) {
8182 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8186 /* async crcs make it difficult to collect full stripe writes. */
8187 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8192 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8196 bio->bi_private = dip;
8197 bio->bi_end_io = btrfs_end_dio_bio;
8198 btrfs_io_bio(bio)->logical = file_offset;
8199 atomic_inc(&dip->pending_bios);
8201 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8202 if (map_length < submit_len + bvec->bv_len ||
8203 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8204 bvec->bv_offset) < bvec->bv_len) {
8206 * inc the count before we submit the bio so
8207 * we know the end IO handler won't happen before
8208 * we inc the count. Otherwise, the dip might get freed
8209 * before we're done setting it up
8211 atomic_inc(&dip->pending_bios);
8212 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8213 file_offset, skip_sum,
8217 atomic_dec(&dip->pending_bios);
8221 start_sector += submit_len >> 9;
8222 file_offset += submit_len;
8227 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8228 start_sector, GFP_NOFS);
8231 bio->bi_private = dip;
8232 bio->bi_end_io = btrfs_end_dio_bio;
8233 btrfs_io_bio(bio)->logical = file_offset;
8235 map_length = orig_bio->bi_iter.bi_size;
8236 ret = btrfs_map_block(root->fs_info, rw,
8238 &map_length, NULL, 0);
8244 submit_len += bvec->bv_len;
8251 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8260 * before atomic variable goto zero, we must
8261 * make sure dip->errors is perceived to be set.
8263 smp_mb__before_atomic();
8264 if (atomic_dec_and_test(&dip->pending_bios))
8265 bio_io_error(dip->orig_bio);
8267 /* bio_end_io() will handle error, so we needn't return it */
8271 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8272 struct inode *inode, loff_t file_offset)
8274 struct btrfs_dio_private *dip = NULL;
8275 struct bio *io_bio = NULL;
8276 struct btrfs_io_bio *btrfs_bio;
8278 int write = rw & REQ_WRITE;
8281 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8283 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8289 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8295 dip->private = dio_bio->bi_private;
8297 dip->logical_offset = file_offset;
8298 dip->bytes = dio_bio->bi_iter.bi_size;
8299 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8300 io_bio->bi_private = dip;
8301 dip->orig_bio = io_bio;
8302 dip->dio_bio = dio_bio;
8303 atomic_set(&dip->pending_bios, 0);
8304 btrfs_bio = btrfs_io_bio(io_bio);
8305 btrfs_bio->logical = file_offset;
8308 io_bio->bi_end_io = btrfs_endio_direct_write;
8310 io_bio->bi_end_io = btrfs_endio_direct_read;
8311 dip->subio_endio = btrfs_subio_endio_read;
8315 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8316 * even if we fail to submit a bio, because in such case we do the
8317 * corresponding error handling below and it must not be done a second
8318 * time by btrfs_direct_IO().
8321 struct btrfs_dio_data *dio_data = current->journal_info;
8323 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8325 dio_data->unsubmitted_oe_range_start =
8326 dio_data->unsubmitted_oe_range_end;
8329 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8333 if (btrfs_bio->end_io)
8334 btrfs_bio->end_io(btrfs_bio, ret);
8338 * If we arrived here it means either we failed to submit the dip
8339 * or we either failed to clone the dio_bio or failed to allocate the
8340 * dip. If we cloned the dio_bio and allocated the dip, we can just
8341 * call bio_endio against our io_bio so that we get proper resource
8342 * cleanup if we fail to submit the dip, otherwise, we must do the
8343 * same as btrfs_endio_direct_[write|read] because we can't call these
8344 * callbacks - they require an allocated dip and a clone of dio_bio.
8346 if (io_bio && dip) {
8347 io_bio->bi_error = -EIO;
8350 * The end io callbacks free our dip, do the final put on io_bio
8351 * and all the cleanup and final put for dio_bio (through
8358 btrfs_endio_direct_write_update_ordered(inode,
8360 dio_bio->bi_iter.bi_size,
8363 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8364 file_offset + dio_bio->bi_iter.bi_size - 1);
8366 dio_bio->bi_error = -EIO;
8368 * Releases and cleans up our dio_bio, no need to bio_put()
8369 * nor bio_endio()/bio_io_error() against dio_bio.
8371 dio_end_io(dio_bio, ret);
8378 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8379 const struct iov_iter *iter, loff_t offset)
8383 unsigned blocksize_mask = root->sectorsize - 1;
8384 ssize_t retval = -EINVAL;
8386 if (offset & blocksize_mask)
8389 if (iov_iter_alignment(iter) & blocksize_mask)
8392 /* If this is a write we don't need to check anymore */
8393 if (iov_iter_rw(iter) == WRITE)
8396 * Check to make sure we don't have duplicate iov_base's in this
8397 * iovec, if so return EINVAL, otherwise we'll get csum errors
8398 * when reading back.
8400 for (seg = 0; seg < iter->nr_segs; seg++) {
8401 for (i = seg + 1; i < iter->nr_segs; i++) {
8402 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8411 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8414 struct file *file = iocb->ki_filp;
8415 struct inode *inode = file->f_mapping->host;
8416 struct btrfs_root *root = BTRFS_I(inode)->root;
8417 struct btrfs_dio_data dio_data = { 0 };
8421 bool relock = false;
8424 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8427 inode_dio_begin(inode);
8428 smp_mb__after_atomic();
8431 * The generic stuff only does filemap_write_and_wait_range, which
8432 * isn't enough if we've written compressed pages to this area, so
8433 * we need to flush the dirty pages again to make absolutely sure
8434 * that any outstanding dirty pages are on disk.
8436 count = iov_iter_count(iter);
8437 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8438 &BTRFS_I(inode)->runtime_flags))
8439 filemap_fdatawrite_range(inode->i_mapping, offset,
8440 offset + count - 1);
8442 if (iov_iter_rw(iter) == WRITE) {
8444 * If the write DIO is beyond the EOF, we need update
8445 * the isize, but it is protected by i_mutex. So we can
8446 * not unlock the i_mutex at this case.
8448 if (offset + count <= inode->i_size) {
8449 mutex_unlock(&inode->i_mutex);
8452 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8455 dio_data.outstanding_extents = div64_u64(count +
8456 BTRFS_MAX_EXTENT_SIZE - 1,
8457 BTRFS_MAX_EXTENT_SIZE);
8460 * We need to know how many extents we reserved so that we can
8461 * do the accounting properly if we go over the number we
8462 * originally calculated. Abuse current->journal_info for this.
8464 dio_data.reserve = round_up(count, root->sectorsize);
8465 dio_data.unsubmitted_oe_range_start = (u64)offset;
8466 dio_data.unsubmitted_oe_range_end = (u64)offset;
8467 current->journal_info = &dio_data;
8468 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8469 &BTRFS_I(inode)->runtime_flags)) {
8470 inode_dio_end(inode);
8471 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8475 ret = __blockdev_direct_IO(iocb, inode,
8476 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8477 iter, offset, btrfs_get_blocks_direct, NULL,
8478 btrfs_submit_direct, flags);
8479 if (iov_iter_rw(iter) == WRITE) {
8480 current->journal_info = NULL;
8481 if (ret < 0 && ret != -EIOCBQUEUED) {
8482 if (dio_data.reserve)
8483 btrfs_delalloc_release_space(inode, offset,
8486 * On error we might have left some ordered extents
8487 * without submitting corresponding bios for them, so
8488 * cleanup them up to avoid other tasks getting them
8489 * and waiting for them to complete forever.
8491 if (dio_data.unsubmitted_oe_range_start <
8492 dio_data.unsubmitted_oe_range_end)
8493 btrfs_endio_direct_write_update_ordered(inode,
8494 dio_data.unsubmitted_oe_range_start,
8495 dio_data.unsubmitted_oe_range_end -
8496 dio_data.unsubmitted_oe_range_start,
8498 } else if (ret >= 0 && (size_t)ret < count)
8499 btrfs_delalloc_release_space(inode, offset,
8500 count - (size_t)ret);
8504 inode_dio_end(inode);
8506 mutex_lock(&inode->i_mutex);
8511 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8513 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8514 __u64 start, __u64 len)
8518 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8522 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8525 int btrfs_readpage(struct file *file, struct page *page)
8527 struct extent_io_tree *tree;
8528 tree = &BTRFS_I(page->mapping->host)->io_tree;
8529 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8532 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8534 struct extent_io_tree *tree;
8535 struct inode *inode = page->mapping->host;
8538 if (current->flags & PF_MEMALLOC) {
8539 redirty_page_for_writepage(wbc, page);
8545 * If we are under memory pressure we will call this directly from the
8546 * VM, we need to make sure we have the inode referenced for the ordered
8547 * extent. If not just return like we didn't do anything.
8549 if (!igrab(inode)) {
8550 redirty_page_for_writepage(wbc, page);
8551 return AOP_WRITEPAGE_ACTIVATE;
8553 tree = &BTRFS_I(page->mapping->host)->io_tree;
8554 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8555 btrfs_add_delayed_iput(inode);
8559 static int btrfs_writepages(struct address_space *mapping,
8560 struct writeback_control *wbc)
8562 struct extent_io_tree *tree;
8564 tree = &BTRFS_I(mapping->host)->io_tree;
8565 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8569 btrfs_readpages(struct file *file, struct address_space *mapping,
8570 struct list_head *pages, unsigned nr_pages)
8572 struct extent_io_tree *tree;
8573 tree = &BTRFS_I(mapping->host)->io_tree;
8574 return extent_readpages(tree, mapping, pages, nr_pages,
8577 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8579 struct extent_io_tree *tree;
8580 struct extent_map_tree *map;
8583 tree = &BTRFS_I(page->mapping->host)->io_tree;
8584 map = &BTRFS_I(page->mapping->host)->extent_tree;
8585 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8587 ClearPagePrivate(page);
8588 set_page_private(page, 0);
8589 page_cache_release(page);
8594 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8596 if (PageWriteback(page) || PageDirty(page))
8598 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8601 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8602 unsigned int length)
8604 struct inode *inode = page->mapping->host;
8605 struct extent_io_tree *tree;
8606 struct btrfs_ordered_extent *ordered;
8607 struct extent_state *cached_state = NULL;
8608 u64 page_start = page_offset(page);
8609 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8610 int inode_evicting = inode->i_state & I_FREEING;
8613 * we have the page locked, so new writeback can't start,
8614 * and the dirty bit won't be cleared while we are here.
8616 * Wait for IO on this page so that we can safely clear
8617 * the PagePrivate2 bit and do ordered accounting
8619 wait_on_page_writeback(page);
8621 tree = &BTRFS_I(inode)->io_tree;
8623 btrfs_releasepage(page, GFP_NOFS);
8627 if (!inode_evicting)
8628 lock_extent_bits(tree, page_start, page_end, &cached_state);
8629 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8632 * IO on this page will never be started, so we need
8633 * to account for any ordered extents now
8635 if (!inode_evicting)
8636 clear_extent_bit(tree, page_start, page_end,
8637 EXTENT_DIRTY | EXTENT_DELALLOC |
8638 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8639 EXTENT_DEFRAG, 1, 0, &cached_state,
8642 * whoever cleared the private bit is responsible
8643 * for the finish_ordered_io
8645 if (TestClearPagePrivate2(page)) {
8646 struct btrfs_ordered_inode_tree *tree;
8649 tree = &BTRFS_I(inode)->ordered_tree;
8651 spin_lock_irq(&tree->lock);
8652 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8653 new_len = page_start - ordered->file_offset;
8654 if (new_len < ordered->truncated_len)
8655 ordered->truncated_len = new_len;
8656 spin_unlock_irq(&tree->lock);
8658 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8660 PAGE_CACHE_SIZE, 1))
8661 btrfs_finish_ordered_io(ordered);
8663 btrfs_put_ordered_extent(ordered);
8664 if (!inode_evicting) {
8665 cached_state = NULL;
8666 lock_extent_bits(tree, page_start, page_end,
8672 * Qgroup reserved space handler
8673 * Page here will be either
8674 * 1) Already written to disk
8675 * In this case, its reserved space is released from data rsv map
8676 * and will be freed by delayed_ref handler finally.
8677 * So even we call qgroup_free_data(), it won't decrease reserved
8679 * 2) Not written to disk
8680 * This means the reserved space should be freed here.
8682 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8683 if (!inode_evicting) {
8684 clear_extent_bit(tree, page_start, page_end,
8685 EXTENT_LOCKED | EXTENT_DIRTY |
8686 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8687 EXTENT_DEFRAG, 1, 1,
8688 &cached_state, GFP_NOFS);
8690 __btrfs_releasepage(page, GFP_NOFS);
8693 ClearPageChecked(page);
8694 if (PagePrivate(page)) {
8695 ClearPagePrivate(page);
8696 set_page_private(page, 0);
8697 page_cache_release(page);
8702 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8703 * called from a page fault handler when a page is first dirtied. Hence we must
8704 * be careful to check for EOF conditions here. We set the page up correctly
8705 * for a written page which means we get ENOSPC checking when writing into
8706 * holes and correct delalloc and unwritten extent mapping on filesystems that
8707 * support these features.
8709 * We are not allowed to take the i_mutex here so we have to play games to
8710 * protect against truncate races as the page could now be beyond EOF. Because
8711 * vmtruncate() writes the inode size before removing pages, once we have the
8712 * page lock we can determine safely if the page is beyond EOF. If it is not
8713 * beyond EOF, then the page is guaranteed safe against truncation until we
8716 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8718 struct page *page = vmf->page;
8719 struct inode *inode = file_inode(vma->vm_file);
8720 struct btrfs_root *root = BTRFS_I(inode)->root;
8721 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8722 struct btrfs_ordered_extent *ordered;
8723 struct extent_state *cached_state = NULL;
8725 unsigned long zero_start;
8732 sb_start_pagefault(inode->i_sb);
8733 page_start = page_offset(page);
8734 page_end = page_start + PAGE_CACHE_SIZE - 1;
8736 ret = btrfs_delalloc_reserve_space(inode, page_start,
8739 ret = file_update_time(vma->vm_file);
8745 else /* -ENOSPC, -EIO, etc */
8746 ret = VM_FAULT_SIGBUS;
8752 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8755 size = i_size_read(inode);
8757 if ((page->mapping != inode->i_mapping) ||
8758 (page_start >= size)) {
8759 /* page got truncated out from underneath us */
8762 wait_on_page_writeback(page);
8764 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8765 set_page_extent_mapped(page);
8768 * we can't set the delalloc bits if there are pending ordered
8769 * extents. Drop our locks and wait for them to finish
8771 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8773 unlock_extent_cached(io_tree, page_start, page_end,
8774 &cached_state, GFP_NOFS);
8776 btrfs_start_ordered_extent(inode, ordered, 1);
8777 btrfs_put_ordered_extent(ordered);
8782 * XXX - page_mkwrite gets called every time the page is dirtied, even
8783 * if it was already dirty, so for space accounting reasons we need to
8784 * clear any delalloc bits for the range we are fixing to save. There
8785 * is probably a better way to do this, but for now keep consistent with
8786 * prepare_pages in the normal write path.
8788 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8789 EXTENT_DIRTY | EXTENT_DELALLOC |
8790 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8791 0, 0, &cached_state, GFP_NOFS);
8793 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8796 unlock_extent_cached(io_tree, page_start, page_end,
8797 &cached_state, GFP_NOFS);
8798 ret = VM_FAULT_SIGBUS;
8803 /* page is wholly or partially inside EOF */
8804 if (page_start + PAGE_CACHE_SIZE > size)
8805 zero_start = size & ~PAGE_CACHE_MASK;
8807 zero_start = PAGE_CACHE_SIZE;
8809 if (zero_start != PAGE_CACHE_SIZE) {
8811 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8812 flush_dcache_page(page);
8815 ClearPageChecked(page);
8816 set_page_dirty(page);
8817 SetPageUptodate(page);
8819 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8820 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8821 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8823 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8827 sb_end_pagefault(inode->i_sb);
8828 return VM_FAULT_LOCKED;
8832 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8834 sb_end_pagefault(inode->i_sb);
8838 static int btrfs_truncate(struct inode *inode)
8840 struct btrfs_root *root = BTRFS_I(inode)->root;
8841 struct btrfs_block_rsv *rsv;
8844 struct btrfs_trans_handle *trans;
8845 u64 mask = root->sectorsize - 1;
8846 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8848 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8854 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8855 * 3 things going on here
8857 * 1) We need to reserve space for our orphan item and the space to
8858 * delete our orphan item. Lord knows we don't want to have a dangling
8859 * orphan item because we didn't reserve space to remove it.
8861 * 2) We need to reserve space to update our inode.
8863 * 3) We need to have something to cache all the space that is going to
8864 * be free'd up by the truncate operation, but also have some slack
8865 * space reserved in case it uses space during the truncate (thank you
8866 * very much snapshotting).
8868 * And we need these to all be seperate. The fact is we can use alot of
8869 * space doing the truncate, and we have no earthly idea how much space
8870 * we will use, so we need the truncate reservation to be seperate so it
8871 * doesn't end up using space reserved for updating the inode or
8872 * removing the orphan item. We also need to be able to stop the
8873 * transaction and start a new one, which means we need to be able to
8874 * update the inode several times, and we have no idea of knowing how
8875 * many times that will be, so we can't just reserve 1 item for the
8876 * entirety of the opration, so that has to be done seperately as well.
8877 * Then there is the orphan item, which does indeed need to be held on
8878 * to for the whole operation, and we need nobody to touch this reserved
8879 * space except the orphan code.
8881 * So that leaves us with
8883 * 1) root->orphan_block_rsv - for the orphan deletion.
8884 * 2) rsv - for the truncate reservation, which we will steal from the
8885 * transaction reservation.
8886 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8887 * updating the inode.
8889 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8892 rsv->size = min_size;
8896 * 1 for the truncate slack space
8897 * 1 for updating the inode.
8899 trans = btrfs_start_transaction(root, 2);
8900 if (IS_ERR(trans)) {
8901 err = PTR_ERR(trans);
8905 /* Migrate the slack space for the truncate to our reserve */
8906 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8911 * So if we truncate and then write and fsync we normally would just
8912 * write the extents that changed, which is a problem if we need to
8913 * first truncate that entire inode. So set this flag so we write out
8914 * all of the extents in the inode to the sync log so we're completely
8917 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8918 trans->block_rsv = rsv;
8921 ret = btrfs_truncate_inode_items(trans, root, inode,
8923 BTRFS_EXTENT_DATA_KEY);
8924 if (ret != -ENOSPC && ret != -EAGAIN) {
8929 trans->block_rsv = &root->fs_info->trans_block_rsv;
8930 ret = btrfs_update_inode(trans, root, inode);
8936 btrfs_end_transaction(trans, root);
8937 btrfs_btree_balance_dirty(root);
8939 trans = btrfs_start_transaction(root, 2);
8940 if (IS_ERR(trans)) {
8941 ret = err = PTR_ERR(trans);
8946 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8948 BUG_ON(ret); /* shouldn't happen */
8949 trans->block_rsv = rsv;
8952 if (ret == 0 && inode->i_nlink > 0) {
8953 trans->block_rsv = root->orphan_block_rsv;
8954 ret = btrfs_orphan_del(trans, inode);
8960 trans->block_rsv = &root->fs_info->trans_block_rsv;
8961 ret = btrfs_update_inode(trans, root, inode);
8965 ret = btrfs_end_transaction(trans, root);
8966 btrfs_btree_balance_dirty(root);
8970 btrfs_free_block_rsv(root, rsv);
8979 * create a new subvolume directory/inode (helper for the ioctl).
8981 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8982 struct btrfs_root *new_root,
8983 struct btrfs_root *parent_root,
8986 struct inode *inode;
8990 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8991 new_dirid, new_dirid,
8992 S_IFDIR | (~current_umask() & S_IRWXUGO),
8995 return PTR_ERR(inode);
8996 inode->i_op = &btrfs_dir_inode_operations;
8997 inode->i_fop = &btrfs_dir_file_operations;
8999 set_nlink(inode, 1);
9000 btrfs_i_size_write(inode, 0);
9001 unlock_new_inode(inode);
9003 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9005 btrfs_err(new_root->fs_info,
9006 "error inheriting subvolume %llu properties: %d",
9007 new_root->root_key.objectid, err);
9009 err = btrfs_update_inode(trans, new_root, inode);
9015 struct inode *btrfs_alloc_inode(struct super_block *sb)
9017 struct btrfs_inode *ei;
9018 struct inode *inode;
9020 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9027 ei->last_sub_trans = 0;
9028 ei->logged_trans = 0;
9029 ei->delalloc_bytes = 0;
9030 ei->defrag_bytes = 0;
9031 ei->disk_i_size = 0;
9034 ei->index_cnt = (u64)-1;
9036 ei->last_unlink_trans = 0;
9037 ei->last_log_commit = 0;
9038 ei->delayed_iput_count = 0;
9040 spin_lock_init(&ei->lock);
9041 ei->outstanding_extents = 0;
9042 ei->reserved_extents = 0;
9044 ei->runtime_flags = 0;
9045 ei->force_compress = BTRFS_COMPRESS_NONE;
9047 ei->delayed_node = NULL;
9049 ei->i_otime.tv_sec = 0;
9050 ei->i_otime.tv_nsec = 0;
9052 inode = &ei->vfs_inode;
9053 extent_map_tree_init(&ei->extent_tree);
9054 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9055 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9056 ei->io_tree.track_uptodate = 1;
9057 ei->io_failure_tree.track_uptodate = 1;
9058 atomic_set(&ei->sync_writers, 0);
9059 mutex_init(&ei->log_mutex);
9060 mutex_init(&ei->delalloc_mutex);
9061 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9062 INIT_LIST_HEAD(&ei->delalloc_inodes);
9063 INIT_LIST_HEAD(&ei->delayed_iput);
9064 RB_CLEAR_NODE(&ei->rb_node);
9069 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9070 void btrfs_test_destroy_inode(struct inode *inode)
9072 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9073 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9077 static void btrfs_i_callback(struct rcu_head *head)
9079 struct inode *inode = container_of(head, struct inode, i_rcu);
9080 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9083 void btrfs_destroy_inode(struct inode *inode)
9085 struct btrfs_ordered_extent *ordered;
9086 struct btrfs_root *root = BTRFS_I(inode)->root;
9088 WARN_ON(!hlist_empty(&inode->i_dentry));
9089 WARN_ON(inode->i_data.nrpages);
9090 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9091 WARN_ON(BTRFS_I(inode)->reserved_extents);
9092 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9093 WARN_ON(BTRFS_I(inode)->csum_bytes);
9094 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9097 * This can happen where we create an inode, but somebody else also
9098 * created the same inode and we need to destroy the one we already
9104 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9105 &BTRFS_I(inode)->runtime_flags)) {
9106 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9108 atomic_dec(&root->orphan_inodes);
9112 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9116 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9117 ordered->file_offset, ordered->len);
9118 btrfs_remove_ordered_extent(inode, ordered);
9119 btrfs_put_ordered_extent(ordered);
9120 btrfs_put_ordered_extent(ordered);
9123 btrfs_qgroup_check_reserved_leak(inode);
9124 inode_tree_del(inode);
9125 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9127 call_rcu(&inode->i_rcu, btrfs_i_callback);
9130 int btrfs_drop_inode(struct inode *inode)
9132 struct btrfs_root *root = BTRFS_I(inode)->root;
9137 /* the snap/subvol tree is on deleting */
9138 if (btrfs_root_refs(&root->root_item) == 0)
9141 return generic_drop_inode(inode);
9144 static void init_once(void *foo)
9146 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9148 inode_init_once(&ei->vfs_inode);
9151 void btrfs_destroy_cachep(void)
9154 * Make sure all delayed rcu free inodes are flushed before we
9158 if (btrfs_inode_cachep)
9159 kmem_cache_destroy(btrfs_inode_cachep);
9160 if (btrfs_trans_handle_cachep)
9161 kmem_cache_destroy(btrfs_trans_handle_cachep);
9162 if (btrfs_transaction_cachep)
9163 kmem_cache_destroy(btrfs_transaction_cachep);
9164 if (btrfs_path_cachep)
9165 kmem_cache_destroy(btrfs_path_cachep);
9166 if (btrfs_free_space_cachep)
9167 kmem_cache_destroy(btrfs_free_space_cachep);
9170 int btrfs_init_cachep(void)
9172 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9173 sizeof(struct btrfs_inode), 0,
9174 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9175 if (!btrfs_inode_cachep)
9178 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9179 sizeof(struct btrfs_trans_handle), 0,
9180 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9181 if (!btrfs_trans_handle_cachep)
9184 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9185 sizeof(struct btrfs_transaction), 0,
9186 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9187 if (!btrfs_transaction_cachep)
9190 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9191 sizeof(struct btrfs_path), 0,
9192 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9193 if (!btrfs_path_cachep)
9196 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9197 sizeof(struct btrfs_free_space), 0,
9198 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9199 if (!btrfs_free_space_cachep)
9204 btrfs_destroy_cachep();
9208 static int btrfs_getattr(struct vfsmount *mnt,
9209 struct dentry *dentry, struct kstat *stat)
9212 struct inode *inode = d_inode(dentry);
9213 u32 blocksize = inode->i_sb->s_blocksize;
9215 generic_fillattr(inode, stat);
9216 stat->dev = BTRFS_I(inode)->root->anon_dev;
9217 stat->blksize = PAGE_CACHE_SIZE;
9219 spin_lock(&BTRFS_I(inode)->lock);
9220 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9221 spin_unlock(&BTRFS_I(inode)->lock);
9222 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9223 ALIGN(delalloc_bytes, blocksize)) >> 9;
9227 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9228 struct inode *new_dir, struct dentry *new_dentry)
9230 struct btrfs_trans_handle *trans;
9231 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9232 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9233 struct inode *new_inode = d_inode(new_dentry);
9234 struct inode *old_inode = d_inode(old_dentry);
9235 struct timespec ctime = CURRENT_TIME;
9239 u64 old_ino = btrfs_ino(old_inode);
9241 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9244 /* we only allow rename subvolume link between subvolumes */
9245 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9248 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9249 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9252 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9253 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9257 /* check for collisions, even if the name isn't there */
9258 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9259 new_dentry->d_name.name,
9260 new_dentry->d_name.len);
9263 if (ret == -EEXIST) {
9265 * eexist without a new_inode */
9266 if (WARN_ON(!new_inode)) {
9270 /* maybe -EOVERFLOW */
9277 * we're using rename to replace one file with another. Start IO on it
9278 * now so we don't add too much work to the end of the transaction
9280 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9281 filemap_flush(old_inode->i_mapping);
9283 /* close the racy window with snapshot create/destroy ioctl */
9284 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9285 down_read(&root->fs_info->subvol_sem);
9287 * We want to reserve the absolute worst case amount of items. So if
9288 * both inodes are subvols and we need to unlink them then that would
9289 * require 4 item modifications, but if they are both normal inodes it
9290 * would require 5 item modifications, so we'll assume their normal
9291 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9292 * should cover the worst case number of items we'll modify.
9294 trans = btrfs_start_transaction(root, 11);
9295 if (IS_ERR(trans)) {
9296 ret = PTR_ERR(trans);
9301 btrfs_record_root_in_trans(trans, dest);
9303 ret = btrfs_set_inode_index(new_dir, &index);
9307 BTRFS_I(old_inode)->dir_index = 0ULL;
9308 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9309 /* force full log commit if subvolume involved. */
9310 btrfs_set_log_full_commit(root->fs_info, trans);
9312 ret = btrfs_insert_inode_ref(trans, dest,
9313 new_dentry->d_name.name,
9314 new_dentry->d_name.len,
9316 btrfs_ino(new_dir), index);
9320 * this is an ugly little race, but the rename is required
9321 * to make sure that if we crash, the inode is either at the
9322 * old name or the new one. pinning the log transaction lets
9323 * us make sure we don't allow a log commit to come in after
9324 * we unlink the name but before we add the new name back in.
9326 btrfs_pin_log_trans(root);
9329 inode_inc_iversion(old_dir);
9330 inode_inc_iversion(new_dir);
9331 inode_inc_iversion(old_inode);
9332 old_dir->i_ctime = old_dir->i_mtime = ctime;
9333 new_dir->i_ctime = new_dir->i_mtime = ctime;
9334 old_inode->i_ctime = ctime;
9336 if (old_dentry->d_parent != new_dentry->d_parent)
9337 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9339 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9340 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9341 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9342 old_dentry->d_name.name,
9343 old_dentry->d_name.len);
9345 ret = __btrfs_unlink_inode(trans, root, old_dir,
9346 d_inode(old_dentry),
9347 old_dentry->d_name.name,
9348 old_dentry->d_name.len);
9350 ret = btrfs_update_inode(trans, root, old_inode);
9353 btrfs_abort_transaction(trans, root, ret);
9358 inode_inc_iversion(new_inode);
9359 new_inode->i_ctime = CURRENT_TIME;
9360 if (unlikely(btrfs_ino(new_inode) ==
9361 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9362 root_objectid = BTRFS_I(new_inode)->location.objectid;
9363 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9365 new_dentry->d_name.name,
9366 new_dentry->d_name.len);
9367 BUG_ON(new_inode->i_nlink == 0);
9369 ret = btrfs_unlink_inode(trans, dest, new_dir,
9370 d_inode(new_dentry),
9371 new_dentry->d_name.name,
9372 new_dentry->d_name.len);
9374 if (!ret && new_inode->i_nlink == 0)
9375 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9377 btrfs_abort_transaction(trans, root, ret);
9382 ret = btrfs_add_link(trans, new_dir, old_inode,
9383 new_dentry->d_name.name,
9384 new_dentry->d_name.len, 0, index);
9386 btrfs_abort_transaction(trans, root, ret);
9390 if (old_inode->i_nlink == 1)
9391 BTRFS_I(old_inode)->dir_index = index;
9393 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9394 struct dentry *parent = new_dentry->d_parent;
9395 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9396 btrfs_end_log_trans(root);
9399 btrfs_end_transaction(trans, root);
9401 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9402 up_read(&root->fs_info->subvol_sem);
9407 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9408 struct inode *new_dir, struct dentry *new_dentry,
9411 if (flags & ~RENAME_NOREPLACE)
9414 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9417 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9419 struct btrfs_delalloc_work *delalloc_work;
9420 struct inode *inode;
9422 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9424 inode = delalloc_work->inode;
9425 filemap_flush(inode->i_mapping);
9426 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9427 &BTRFS_I(inode)->runtime_flags))
9428 filemap_flush(inode->i_mapping);
9430 if (delalloc_work->delay_iput)
9431 btrfs_add_delayed_iput(inode);
9434 complete(&delalloc_work->completion);
9437 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9440 struct btrfs_delalloc_work *work;
9442 work = kmalloc(sizeof(*work), GFP_NOFS);
9446 init_completion(&work->completion);
9447 INIT_LIST_HEAD(&work->list);
9448 work->inode = inode;
9449 work->delay_iput = delay_iput;
9450 WARN_ON_ONCE(!inode);
9451 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9452 btrfs_run_delalloc_work, NULL, NULL);
9457 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9459 wait_for_completion(&work->completion);
9464 * some fairly slow code that needs optimization. This walks the list
9465 * of all the inodes with pending delalloc and forces them to disk.
9467 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9470 struct btrfs_inode *binode;
9471 struct inode *inode;
9472 struct btrfs_delalloc_work *work, *next;
9473 struct list_head works;
9474 struct list_head splice;
9477 INIT_LIST_HEAD(&works);
9478 INIT_LIST_HEAD(&splice);
9480 mutex_lock(&root->delalloc_mutex);
9481 spin_lock(&root->delalloc_lock);
9482 list_splice_init(&root->delalloc_inodes, &splice);
9483 while (!list_empty(&splice)) {
9484 binode = list_entry(splice.next, struct btrfs_inode,
9487 list_move_tail(&binode->delalloc_inodes,
9488 &root->delalloc_inodes);
9489 inode = igrab(&binode->vfs_inode);
9491 cond_resched_lock(&root->delalloc_lock);
9494 spin_unlock(&root->delalloc_lock);
9496 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9499 btrfs_add_delayed_iput(inode);
9505 list_add_tail(&work->list, &works);
9506 btrfs_queue_work(root->fs_info->flush_workers,
9509 if (nr != -1 && ret >= nr)
9512 spin_lock(&root->delalloc_lock);
9514 spin_unlock(&root->delalloc_lock);
9517 list_for_each_entry_safe(work, next, &works, list) {
9518 list_del_init(&work->list);
9519 btrfs_wait_and_free_delalloc_work(work);
9522 if (!list_empty_careful(&splice)) {
9523 spin_lock(&root->delalloc_lock);
9524 list_splice_tail(&splice, &root->delalloc_inodes);
9525 spin_unlock(&root->delalloc_lock);
9527 mutex_unlock(&root->delalloc_mutex);
9531 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9535 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9538 ret = __start_delalloc_inodes(root, delay_iput, -1);
9542 * the filemap_flush will queue IO into the worker threads, but
9543 * we have to make sure the IO is actually started and that
9544 * ordered extents get created before we return
9546 atomic_inc(&root->fs_info->async_submit_draining);
9547 while (atomic_read(&root->fs_info->nr_async_submits) ||
9548 atomic_read(&root->fs_info->async_delalloc_pages)) {
9549 wait_event(root->fs_info->async_submit_wait,
9550 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9551 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9553 atomic_dec(&root->fs_info->async_submit_draining);
9557 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9560 struct btrfs_root *root;
9561 struct list_head splice;
9564 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9567 INIT_LIST_HEAD(&splice);
9569 mutex_lock(&fs_info->delalloc_root_mutex);
9570 spin_lock(&fs_info->delalloc_root_lock);
9571 list_splice_init(&fs_info->delalloc_roots, &splice);
9572 while (!list_empty(&splice) && nr) {
9573 root = list_first_entry(&splice, struct btrfs_root,
9575 root = btrfs_grab_fs_root(root);
9577 list_move_tail(&root->delalloc_root,
9578 &fs_info->delalloc_roots);
9579 spin_unlock(&fs_info->delalloc_root_lock);
9581 ret = __start_delalloc_inodes(root, delay_iput, nr);
9582 btrfs_put_fs_root(root);
9590 spin_lock(&fs_info->delalloc_root_lock);
9592 spin_unlock(&fs_info->delalloc_root_lock);
9595 atomic_inc(&fs_info->async_submit_draining);
9596 while (atomic_read(&fs_info->nr_async_submits) ||
9597 atomic_read(&fs_info->async_delalloc_pages)) {
9598 wait_event(fs_info->async_submit_wait,
9599 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9600 atomic_read(&fs_info->async_delalloc_pages) == 0));
9602 atomic_dec(&fs_info->async_submit_draining);
9604 if (!list_empty_careful(&splice)) {
9605 spin_lock(&fs_info->delalloc_root_lock);
9606 list_splice_tail(&splice, &fs_info->delalloc_roots);
9607 spin_unlock(&fs_info->delalloc_root_lock);
9609 mutex_unlock(&fs_info->delalloc_root_mutex);
9613 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9614 const char *symname)
9616 struct btrfs_trans_handle *trans;
9617 struct btrfs_root *root = BTRFS_I(dir)->root;
9618 struct btrfs_path *path;
9619 struct btrfs_key key;
9620 struct inode *inode = NULL;
9628 struct btrfs_file_extent_item *ei;
9629 struct extent_buffer *leaf;
9631 name_len = strlen(symname);
9632 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9633 return -ENAMETOOLONG;
9636 * 2 items for inode item and ref
9637 * 2 items for dir items
9638 * 1 item for updating parent inode item
9639 * 1 item for the inline extent item
9640 * 1 item for xattr if selinux is on
9642 trans = btrfs_start_transaction(root, 7);
9644 return PTR_ERR(trans);
9646 err = btrfs_find_free_ino(root, &objectid);
9650 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9651 dentry->d_name.len, btrfs_ino(dir), objectid,
9652 S_IFLNK|S_IRWXUGO, &index);
9653 if (IS_ERR(inode)) {
9654 err = PTR_ERR(inode);
9659 * If the active LSM wants to access the inode during
9660 * d_instantiate it needs these. Smack checks to see
9661 * if the filesystem supports xattrs by looking at the
9664 inode->i_fop = &btrfs_file_operations;
9665 inode->i_op = &btrfs_file_inode_operations;
9666 inode->i_mapping->a_ops = &btrfs_aops;
9667 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9669 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9671 goto out_unlock_inode;
9673 path = btrfs_alloc_path();
9676 goto out_unlock_inode;
9678 key.objectid = btrfs_ino(inode);
9680 key.type = BTRFS_EXTENT_DATA_KEY;
9681 datasize = btrfs_file_extent_calc_inline_size(name_len);
9682 err = btrfs_insert_empty_item(trans, root, path, &key,
9685 btrfs_free_path(path);
9686 goto out_unlock_inode;
9688 leaf = path->nodes[0];
9689 ei = btrfs_item_ptr(leaf, path->slots[0],
9690 struct btrfs_file_extent_item);
9691 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9692 btrfs_set_file_extent_type(leaf, ei,
9693 BTRFS_FILE_EXTENT_INLINE);
9694 btrfs_set_file_extent_encryption(leaf, ei, 0);
9695 btrfs_set_file_extent_compression(leaf, ei, 0);
9696 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9697 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9699 ptr = btrfs_file_extent_inline_start(ei);
9700 write_extent_buffer(leaf, symname, ptr, name_len);
9701 btrfs_mark_buffer_dirty(leaf);
9702 btrfs_free_path(path);
9704 inode->i_op = &btrfs_symlink_inode_operations;
9705 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9706 inode_set_bytes(inode, name_len);
9707 btrfs_i_size_write(inode, name_len);
9708 err = btrfs_update_inode(trans, root, inode);
9710 * Last step, add directory indexes for our symlink inode. This is the
9711 * last step to avoid extra cleanup of these indexes if an error happens
9715 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9718 goto out_unlock_inode;
9721 unlock_new_inode(inode);
9722 d_instantiate(dentry, inode);
9725 btrfs_end_transaction(trans, root);
9727 inode_dec_link_count(inode);
9730 btrfs_btree_balance_dirty(root);
9735 unlock_new_inode(inode);
9739 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9740 u64 start, u64 num_bytes, u64 min_size,
9741 loff_t actual_len, u64 *alloc_hint,
9742 struct btrfs_trans_handle *trans)
9744 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9745 struct extent_map *em;
9746 struct btrfs_root *root = BTRFS_I(inode)->root;
9747 struct btrfs_key ins;
9748 u64 cur_offset = start;
9751 u64 last_alloc = (u64)-1;
9753 bool own_trans = true;
9757 while (num_bytes > 0) {
9759 trans = btrfs_start_transaction(root, 3);
9760 if (IS_ERR(trans)) {
9761 ret = PTR_ERR(trans);
9766 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9767 cur_bytes = max(cur_bytes, min_size);
9769 * If we are severely fragmented we could end up with really
9770 * small allocations, so if the allocator is returning small
9771 * chunks lets make its job easier by only searching for those
9774 cur_bytes = min(cur_bytes, last_alloc);
9775 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9776 *alloc_hint, &ins, 1, 0);
9779 btrfs_end_transaction(trans, root);
9783 last_alloc = ins.offset;
9784 ret = insert_reserved_file_extent(trans, inode,
9785 cur_offset, ins.objectid,
9786 ins.offset, ins.offset,
9787 ins.offset, 0, 0, 0,
9788 BTRFS_FILE_EXTENT_PREALLOC);
9790 btrfs_free_reserved_extent(root, ins.objectid,
9792 btrfs_abort_transaction(trans, root, ret);
9794 btrfs_end_transaction(trans, root);
9798 btrfs_drop_extent_cache(inode, cur_offset,
9799 cur_offset + ins.offset -1, 0);
9801 em = alloc_extent_map();
9803 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9804 &BTRFS_I(inode)->runtime_flags);
9808 em->start = cur_offset;
9809 em->orig_start = cur_offset;
9810 em->len = ins.offset;
9811 em->block_start = ins.objectid;
9812 em->block_len = ins.offset;
9813 em->orig_block_len = ins.offset;
9814 em->ram_bytes = ins.offset;
9815 em->bdev = root->fs_info->fs_devices->latest_bdev;
9816 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9817 em->generation = trans->transid;
9820 write_lock(&em_tree->lock);
9821 ret = add_extent_mapping(em_tree, em, 1);
9822 write_unlock(&em_tree->lock);
9825 btrfs_drop_extent_cache(inode, cur_offset,
9826 cur_offset + ins.offset - 1,
9829 free_extent_map(em);
9831 num_bytes -= ins.offset;
9832 cur_offset += ins.offset;
9833 *alloc_hint = ins.objectid + ins.offset;
9835 inode_inc_iversion(inode);
9836 inode->i_ctime = CURRENT_TIME;
9837 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9838 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9839 (actual_len > inode->i_size) &&
9840 (cur_offset > inode->i_size)) {
9841 if (cur_offset > actual_len)
9842 i_size = actual_len;
9844 i_size = cur_offset;
9845 i_size_write(inode, i_size);
9846 btrfs_ordered_update_i_size(inode, i_size, NULL);
9849 ret = btrfs_update_inode(trans, root, inode);
9852 btrfs_abort_transaction(trans, root, ret);
9854 btrfs_end_transaction(trans, root);
9859 btrfs_end_transaction(trans, root);
9864 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9865 u64 start, u64 num_bytes, u64 min_size,
9866 loff_t actual_len, u64 *alloc_hint)
9868 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9869 min_size, actual_len, alloc_hint,
9873 int btrfs_prealloc_file_range_trans(struct inode *inode,
9874 struct btrfs_trans_handle *trans, int mode,
9875 u64 start, u64 num_bytes, u64 min_size,
9876 loff_t actual_len, u64 *alloc_hint)
9878 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9879 min_size, actual_len, alloc_hint, trans);
9882 static int btrfs_set_page_dirty(struct page *page)
9884 return __set_page_dirty_nobuffers(page);
9887 static int btrfs_permission(struct inode *inode, int mask)
9889 struct btrfs_root *root = BTRFS_I(inode)->root;
9890 umode_t mode = inode->i_mode;
9892 if (mask & MAY_WRITE &&
9893 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9894 if (btrfs_root_readonly(root))
9896 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9899 return generic_permission(inode, mask);
9902 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9904 struct btrfs_trans_handle *trans;
9905 struct btrfs_root *root = BTRFS_I(dir)->root;
9906 struct inode *inode = NULL;
9912 * 5 units required for adding orphan entry
9914 trans = btrfs_start_transaction(root, 5);
9916 return PTR_ERR(trans);
9918 ret = btrfs_find_free_ino(root, &objectid);
9922 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9923 btrfs_ino(dir), objectid, mode, &index);
9924 if (IS_ERR(inode)) {
9925 ret = PTR_ERR(inode);
9930 inode->i_fop = &btrfs_file_operations;
9931 inode->i_op = &btrfs_file_inode_operations;
9933 inode->i_mapping->a_ops = &btrfs_aops;
9934 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9936 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9940 ret = btrfs_update_inode(trans, root, inode);
9943 ret = btrfs_orphan_add(trans, inode);
9948 * We set number of links to 0 in btrfs_new_inode(), and here we set
9949 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9952 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9954 set_nlink(inode, 1);
9955 unlock_new_inode(inode);
9956 d_tmpfile(dentry, inode);
9957 mark_inode_dirty(inode);
9960 btrfs_end_transaction(trans, root);
9963 btrfs_balance_delayed_items(root);
9964 btrfs_btree_balance_dirty(root);
9968 unlock_new_inode(inode);
9973 /* Inspired by filemap_check_errors() */
9974 int btrfs_inode_check_errors(struct inode *inode)
9978 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9979 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9981 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9982 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9988 static const struct inode_operations btrfs_dir_inode_operations = {
9989 .getattr = btrfs_getattr,
9990 .lookup = btrfs_lookup,
9991 .create = btrfs_create,
9992 .unlink = btrfs_unlink,
9994 .mkdir = btrfs_mkdir,
9995 .rmdir = btrfs_rmdir,
9996 .rename2 = btrfs_rename2,
9997 .symlink = btrfs_symlink,
9998 .setattr = btrfs_setattr,
9999 .mknod = btrfs_mknod,
10000 .setxattr = btrfs_setxattr,
10001 .getxattr = btrfs_getxattr,
10002 .listxattr = btrfs_listxattr,
10003 .removexattr = btrfs_removexattr,
10004 .permission = btrfs_permission,
10005 .get_acl = btrfs_get_acl,
10006 .set_acl = btrfs_set_acl,
10007 .update_time = btrfs_update_time,
10008 .tmpfile = btrfs_tmpfile,
10010 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10011 .lookup = btrfs_lookup,
10012 .permission = btrfs_permission,
10013 .get_acl = btrfs_get_acl,
10014 .set_acl = btrfs_set_acl,
10015 .update_time = btrfs_update_time,
10018 static const struct file_operations btrfs_dir_file_operations = {
10019 .llseek = generic_file_llseek,
10020 .read = generic_read_dir,
10021 .iterate = btrfs_real_readdir,
10022 .unlocked_ioctl = btrfs_ioctl,
10023 #ifdef CONFIG_COMPAT
10024 .compat_ioctl = btrfs_ioctl,
10026 .release = btrfs_release_file,
10027 .fsync = btrfs_sync_file,
10030 static const struct extent_io_ops btrfs_extent_io_ops = {
10031 .fill_delalloc = run_delalloc_range,
10032 .submit_bio_hook = btrfs_submit_bio_hook,
10033 .merge_bio_hook = btrfs_merge_bio_hook,
10034 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10035 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10036 .writepage_start_hook = btrfs_writepage_start_hook,
10037 .set_bit_hook = btrfs_set_bit_hook,
10038 .clear_bit_hook = btrfs_clear_bit_hook,
10039 .merge_extent_hook = btrfs_merge_extent_hook,
10040 .split_extent_hook = btrfs_split_extent_hook,
10044 * btrfs doesn't support the bmap operation because swapfiles
10045 * use bmap to make a mapping of extents in the file. They assume
10046 * these extents won't change over the life of the file and they
10047 * use the bmap result to do IO directly to the drive.
10049 * the btrfs bmap call would return logical addresses that aren't
10050 * suitable for IO and they also will change frequently as COW
10051 * operations happen. So, swapfile + btrfs == corruption.
10053 * For now we're avoiding this by dropping bmap.
10055 static const struct address_space_operations btrfs_aops = {
10056 .readpage = btrfs_readpage,
10057 .writepage = btrfs_writepage,
10058 .writepages = btrfs_writepages,
10059 .readpages = btrfs_readpages,
10060 .direct_IO = btrfs_direct_IO,
10061 .invalidatepage = btrfs_invalidatepage,
10062 .releasepage = btrfs_releasepage,
10063 .set_page_dirty = btrfs_set_page_dirty,
10064 .error_remove_page = generic_error_remove_page,
10067 static const struct address_space_operations btrfs_symlink_aops = {
10068 .readpage = btrfs_readpage,
10069 .writepage = btrfs_writepage,
10070 .invalidatepage = btrfs_invalidatepage,
10071 .releasepage = btrfs_releasepage,
10074 static const struct inode_operations btrfs_file_inode_operations = {
10075 .getattr = btrfs_getattr,
10076 .setattr = btrfs_setattr,
10077 .setxattr = btrfs_setxattr,
10078 .getxattr = btrfs_getxattr,
10079 .listxattr = btrfs_listxattr,
10080 .removexattr = btrfs_removexattr,
10081 .permission = btrfs_permission,
10082 .fiemap = btrfs_fiemap,
10083 .get_acl = btrfs_get_acl,
10084 .set_acl = btrfs_set_acl,
10085 .update_time = btrfs_update_time,
10087 static const struct inode_operations btrfs_special_inode_operations = {
10088 .getattr = btrfs_getattr,
10089 .setattr = btrfs_setattr,
10090 .permission = btrfs_permission,
10091 .setxattr = btrfs_setxattr,
10092 .getxattr = btrfs_getxattr,
10093 .listxattr = btrfs_listxattr,
10094 .removexattr = btrfs_removexattr,
10095 .get_acl = btrfs_get_acl,
10096 .set_acl = btrfs_set_acl,
10097 .update_time = btrfs_update_time,
10099 static const struct inode_operations btrfs_symlink_inode_operations = {
10100 .readlink = generic_readlink,
10101 .follow_link = page_follow_link_light,
10102 .put_link = page_put_link,
10103 .getattr = btrfs_getattr,
10104 .setattr = btrfs_setattr,
10105 .permission = btrfs_permission,
10106 .setxattr = btrfs_setxattr,
10107 .getxattr = btrfs_getxattr,
10108 .listxattr = btrfs_listxattr,
10109 .removexattr = btrfs_removexattr,
10110 .update_time = btrfs_update_time,
10113 const struct dentry_operations btrfs_dentry_operations = {
10114 .d_delete = btrfs_dentry_delete,
10115 .d_release = btrfs_dentry_release,