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_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
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 > root->sectorsize ||
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_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_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_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 &
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,
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_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);
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);
658 static void free_async_extent_pages(struct async_extent *async_extent)
662 if (!async_extent->pages)
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 put_page(async_extent->pages[i]);
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
683 struct async_extent *async_extent;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
698 io_tree = &BTRFS_I(inode)->io_tree;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &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;
827 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
830 * clear dirty, set writeback and unlock the pages.
832 extent_clear_unlock_delalloc(inode, async_extent->start,
833 async_extent->start +
834 async_extent->ram_size - 1,
835 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
836 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
838 ret = btrfs_submit_compressed_write(inode,
840 async_extent->ram_size,
842 ins.offset, async_extent->pages,
843 async_extent->nr_pages);
845 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 struct page *p = async_extent->pages[0];
847 const u64 start = async_extent->start;
848 const u64 end = start + async_extent->ram_size - 1;
850 p->mapping = inode->i_mapping;
851 tree->ops->writepage_end_io_hook(p, start, end,
854 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
857 free_async_extent_pages(async_extent);
859 alloc_hint = ins.objectid + ins.offset;
865 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
866 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
868 extent_clear_unlock_delalloc(inode, async_extent->start,
869 async_extent->start +
870 async_extent->ram_size - 1,
871 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
872 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
873 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
874 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
876 free_async_extent_pages(async_extent);
881 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
884 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885 struct extent_map *em;
888 read_lock(&em_tree->lock);
889 em = search_extent_mapping(em_tree, start, num_bytes);
892 * if block start isn't an actual block number then find the
893 * first block in this inode and use that as a hint. If that
894 * block is also bogus then just don't worry about it.
896 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
898 em = search_extent_mapping(em_tree, 0, 0);
899 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
900 alloc_hint = em->block_start;
904 alloc_hint = em->block_start;
908 read_unlock(&em_tree->lock);
914 * when extent_io.c finds a delayed allocation range in the file,
915 * the call backs end up in this code. The basic idea is to
916 * allocate extents on disk for the range, and create ordered data structs
917 * in ram to track those extents.
919 * locked_page is the page that writepage had locked already. We use
920 * it to make sure we don't do extra locks or unlocks.
922 * *page_started is set to one if we unlock locked_page and do everything
923 * required to start IO on it. It may be clean and already done with
926 static noinline int cow_file_range(struct inode *inode,
927 struct page *locked_page,
928 u64 start, u64 end, int *page_started,
929 unsigned long *nr_written,
932 struct btrfs_root *root = BTRFS_I(inode)->root;
935 unsigned long ram_size;
938 u64 blocksize = root->sectorsize;
939 struct btrfs_key ins;
940 struct extent_map *em;
941 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
944 if (btrfs_is_free_space_inode(inode)) {
950 num_bytes = ALIGN(end - start + 1, blocksize);
951 num_bytes = max(blocksize, num_bytes);
952 disk_num_bytes = num_bytes;
954 /* if this is a small write inside eof, kick off defrag */
955 if (num_bytes < SZ_64K &&
956 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
957 btrfs_add_inode_defrag(NULL, inode);
960 /* lets try to make an inline extent */
961 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
964 extent_clear_unlock_delalloc(inode, start, end, NULL,
965 EXTENT_LOCKED | EXTENT_DELALLOC |
966 EXTENT_DEFRAG, PAGE_UNLOCK |
967 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
970 *nr_written = *nr_written +
971 (end - start + PAGE_SIZE) / PAGE_SIZE;
974 } else if (ret < 0) {
979 BUG_ON(disk_num_bytes >
980 btrfs_super_total_bytes(root->fs_info->super_copy));
982 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
983 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
985 while (disk_num_bytes > 0) {
988 cur_alloc_size = disk_num_bytes;
989 ret = btrfs_reserve_extent(root, cur_alloc_size,
990 root->sectorsize, 0, alloc_hint,
995 em = alloc_extent_map();
1001 em->orig_start = em->start;
1002 ram_size = ins.offset;
1003 em->len = ins.offset;
1004 em->mod_start = em->start;
1005 em->mod_len = em->len;
1007 em->block_start = ins.objectid;
1008 em->block_len = ins.offset;
1009 em->orig_block_len = ins.offset;
1010 em->ram_bytes = ram_size;
1011 em->bdev = root->fs_info->fs_devices->latest_bdev;
1012 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1013 em->generation = -1;
1016 write_lock(&em_tree->lock);
1017 ret = add_extent_mapping(em_tree, em, 1);
1018 write_unlock(&em_tree->lock);
1019 if (ret != -EEXIST) {
1020 free_extent_map(em);
1023 btrfs_drop_extent_cache(inode, start,
1024 start + ram_size - 1, 0);
1029 cur_alloc_size = ins.offset;
1030 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1031 ram_size, cur_alloc_size, 0);
1033 goto out_drop_extent_cache;
1035 if (root->root_key.objectid ==
1036 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1037 ret = btrfs_reloc_clone_csums(inode, start,
1040 goto out_drop_extent_cache;
1043 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1045 if (disk_num_bytes < cur_alloc_size)
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1055 op = unlock ? PAGE_UNLOCK : 0;
1056 op |= PAGE_SET_PRIVATE2;
1058 extent_clear_unlock_delalloc(inode, start,
1059 start + ram_size - 1, locked_page,
1060 EXTENT_LOCKED | EXTENT_DELALLOC,
1062 disk_num_bytes -= cur_alloc_size;
1063 num_bytes -= cur_alloc_size;
1064 alloc_hint = ins.objectid + ins.offset;
1065 start += cur_alloc_size;
1070 out_drop_extent_cache:
1071 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1073 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1074 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1076 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1078 EXTENT_DELALLOC | EXTENT_DEFRAG,
1079 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1080 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1085 * work queue call back to started compression on a file and pages
1087 static noinline void async_cow_start(struct btrfs_work *work)
1089 struct async_cow *async_cow;
1091 async_cow = container_of(work, struct async_cow, work);
1093 compress_file_range(async_cow->inode, async_cow->locked_page,
1094 async_cow->start, async_cow->end, async_cow,
1096 if (num_added == 0) {
1097 btrfs_add_delayed_iput(async_cow->inode);
1098 async_cow->inode = NULL;
1103 * work queue call back to submit previously compressed pages
1105 static noinline void async_cow_submit(struct btrfs_work *work)
1107 struct async_cow *async_cow;
1108 struct btrfs_root *root;
1109 unsigned long nr_pages;
1111 async_cow = container_of(work, struct async_cow, work);
1113 root = async_cow->root;
1114 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1118 * atomic_sub_return implies a barrier for waitqueue_active
1120 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1122 waitqueue_active(&root->fs_info->async_submit_wait))
1123 wake_up(&root->fs_info->async_submit_wait);
1125 if (async_cow->inode)
1126 submit_compressed_extents(async_cow->inode, async_cow);
1129 static noinline void async_cow_free(struct btrfs_work *work)
1131 struct async_cow *async_cow;
1132 async_cow = container_of(work, struct async_cow, work);
1133 if (async_cow->inode)
1134 btrfs_add_delayed_iput(async_cow->inode);
1138 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1139 u64 start, u64 end, int *page_started,
1140 unsigned long *nr_written)
1142 struct async_cow *async_cow;
1143 struct btrfs_root *root = BTRFS_I(inode)->root;
1144 unsigned long nr_pages;
1146 int limit = 10 * SZ_1M;
1148 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1149 1, 0, NULL, GFP_NOFS);
1150 while (start < end) {
1151 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1152 BUG_ON(!async_cow); /* -ENOMEM */
1153 async_cow->inode = igrab(inode);
1154 async_cow->root = root;
1155 async_cow->locked_page = locked_page;
1156 async_cow->start = start;
1158 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1159 !btrfs_test_opt(root, FORCE_COMPRESS))
1162 cur_end = min(end, start + SZ_512K - 1);
1164 async_cow->end = cur_end;
1165 INIT_LIST_HEAD(&async_cow->extents);
1167 btrfs_init_work(&async_cow->work,
1168 btrfs_delalloc_helper,
1169 async_cow_start, async_cow_submit,
1172 nr_pages = (cur_end - start + PAGE_SIZE) >>
1174 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1176 btrfs_queue_work(root->fs_info->delalloc_workers,
1179 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1180 wait_event(root->fs_info->async_submit_wait,
1181 (atomic_read(&root->fs_info->async_delalloc_pages) <
1185 while (atomic_read(&root->fs_info->async_submit_draining) &&
1186 atomic_read(&root->fs_info->async_delalloc_pages)) {
1187 wait_event(root->fs_info->async_submit_wait,
1188 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1192 *nr_written += nr_pages;
1193 start = cur_end + 1;
1199 static noinline int csum_exist_in_range(struct btrfs_root *root,
1200 u64 bytenr, u64 num_bytes)
1203 struct btrfs_ordered_sum *sums;
1206 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1207 bytenr + num_bytes - 1, &list, 0);
1208 if (ret == 0 && list_empty(&list))
1211 while (!list_empty(&list)) {
1212 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1213 list_del(&sums->list);
1220 * when nowcow writeback call back. This checks for snapshots or COW copies
1221 * of the extents that exist in the file, and COWs the file as required.
1223 * If no cow copies or snapshots exist, we write directly to the existing
1226 static noinline int run_delalloc_nocow(struct inode *inode,
1227 struct page *locked_page,
1228 u64 start, u64 end, int *page_started, int force,
1229 unsigned long *nr_written)
1231 struct btrfs_root *root = BTRFS_I(inode)->root;
1232 struct btrfs_trans_handle *trans;
1233 struct extent_buffer *leaf;
1234 struct btrfs_path *path;
1235 struct btrfs_file_extent_item *fi;
1236 struct btrfs_key found_key;
1251 u64 ino = btrfs_ino(inode);
1253 path = btrfs_alloc_path();
1255 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1256 EXTENT_LOCKED | EXTENT_DELALLOC |
1257 EXTENT_DO_ACCOUNTING |
1258 EXTENT_DEFRAG, PAGE_UNLOCK |
1260 PAGE_SET_WRITEBACK |
1261 PAGE_END_WRITEBACK);
1265 nolock = btrfs_is_free_space_inode(inode);
1268 trans = btrfs_join_transaction_nolock(root);
1270 trans = btrfs_join_transaction(root);
1272 if (IS_ERR(trans)) {
1273 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1274 EXTENT_LOCKED | EXTENT_DELALLOC |
1275 EXTENT_DO_ACCOUNTING |
1276 EXTENT_DEFRAG, PAGE_UNLOCK |
1278 PAGE_SET_WRITEBACK |
1279 PAGE_END_WRITEBACK);
1280 btrfs_free_path(path);
1281 return PTR_ERR(trans);
1284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1286 cow_start = (u64)-1;
1289 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1293 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1294 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key,
1296 path->slots[0] - 1);
1297 if (found_key.objectid == ino &&
1298 found_key.type == BTRFS_EXTENT_DATA_KEY)
1303 leaf = path->nodes[0];
1304 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1305 ret = btrfs_next_leaf(root, path);
1310 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1318 if (found_key.objectid > ino)
1320 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1321 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1325 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1326 found_key.offset > end)
1329 if (found_key.offset > cur_offset) {
1330 extent_end = found_key.offset;
1335 fi = btrfs_item_ptr(leaf, path->slots[0],
1336 struct btrfs_file_extent_item);
1337 extent_type = btrfs_file_extent_type(leaf, fi);
1339 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1340 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1341 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1342 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1343 extent_offset = btrfs_file_extent_offset(leaf, fi);
1344 extent_end = found_key.offset +
1345 btrfs_file_extent_num_bytes(leaf, fi);
1347 btrfs_file_extent_disk_num_bytes(leaf, fi);
1348 if (extent_end <= start) {
1352 if (disk_bytenr == 0)
1354 if (btrfs_file_extent_compression(leaf, fi) ||
1355 btrfs_file_extent_encryption(leaf, fi) ||
1356 btrfs_file_extent_other_encoding(leaf, fi))
1358 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1360 if (btrfs_extent_readonly(root, disk_bytenr))
1362 if (btrfs_cross_ref_exist(trans, root, ino,
1364 extent_offset, disk_bytenr))
1366 disk_bytenr += extent_offset;
1367 disk_bytenr += cur_offset - found_key.offset;
1368 num_bytes = min(end + 1, extent_end) - cur_offset;
1370 * if there are pending snapshots for this root,
1371 * we fall into common COW way.
1374 err = btrfs_start_write_no_snapshoting(root);
1379 * force cow if csum exists in the range.
1380 * this ensure that csum for a given extent are
1381 * either valid or do not exist.
1383 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1386 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1387 extent_end = found_key.offset +
1388 btrfs_file_extent_inline_len(leaf,
1389 path->slots[0], fi);
1390 extent_end = ALIGN(extent_end, root->sectorsize);
1395 if (extent_end <= start) {
1397 if (!nolock && nocow)
1398 btrfs_end_write_no_snapshoting(root);
1402 if (cow_start == (u64)-1)
1403 cow_start = cur_offset;
1404 cur_offset = extent_end;
1405 if (cur_offset > end)
1411 btrfs_release_path(path);
1412 if (cow_start != (u64)-1) {
1413 ret = cow_file_range(inode, locked_page,
1414 cow_start, found_key.offset - 1,
1415 page_started, nr_written, 1);
1417 if (!nolock && nocow)
1418 btrfs_end_write_no_snapshoting(root);
1421 cow_start = (u64)-1;
1424 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1425 struct extent_map *em;
1426 struct extent_map_tree *em_tree;
1427 em_tree = &BTRFS_I(inode)->extent_tree;
1428 em = alloc_extent_map();
1429 BUG_ON(!em); /* -ENOMEM */
1430 em->start = cur_offset;
1431 em->orig_start = found_key.offset - extent_offset;
1432 em->len = num_bytes;
1433 em->block_len = num_bytes;
1434 em->block_start = disk_bytenr;
1435 em->orig_block_len = disk_num_bytes;
1436 em->ram_bytes = ram_bytes;
1437 em->bdev = root->fs_info->fs_devices->latest_bdev;
1438 em->mod_start = em->start;
1439 em->mod_len = em->len;
1440 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1441 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1442 em->generation = -1;
1444 write_lock(&em_tree->lock);
1445 ret = add_extent_mapping(em_tree, em, 1);
1446 write_unlock(&em_tree->lock);
1447 if (ret != -EEXIST) {
1448 free_extent_map(em);
1451 btrfs_drop_extent_cache(inode, em->start,
1452 em->start + em->len - 1, 0);
1454 type = BTRFS_ORDERED_PREALLOC;
1456 type = BTRFS_ORDERED_NOCOW;
1459 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1460 num_bytes, num_bytes, type);
1461 BUG_ON(ret); /* -ENOMEM */
1463 if (root->root_key.objectid ==
1464 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1465 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1468 if (!nolock && nocow)
1469 btrfs_end_write_no_snapshoting(root);
1474 extent_clear_unlock_delalloc(inode, cur_offset,
1475 cur_offset + num_bytes - 1,
1476 locked_page, EXTENT_LOCKED |
1477 EXTENT_DELALLOC, PAGE_UNLOCK |
1479 if (!nolock && nocow)
1480 btrfs_end_write_no_snapshoting(root);
1481 cur_offset = extent_end;
1482 if (cur_offset > end)
1485 btrfs_release_path(path);
1487 if (cur_offset <= end && cow_start == (u64)-1) {
1488 cow_start = cur_offset;
1492 if (cow_start != (u64)-1) {
1493 ret = cow_file_range(inode, locked_page, cow_start, end,
1494 page_started, nr_written, 1);
1500 err = btrfs_end_transaction(trans, root);
1504 if (ret && cur_offset < end)
1505 extent_clear_unlock_delalloc(inode, cur_offset, end,
1506 locked_page, EXTENT_LOCKED |
1507 EXTENT_DELALLOC | EXTENT_DEFRAG |
1508 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1510 PAGE_SET_WRITEBACK |
1511 PAGE_END_WRITEBACK);
1512 btrfs_free_path(path);
1516 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1519 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1520 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1524 * @defrag_bytes is a hint value, no spinlock held here,
1525 * if is not zero, it means the file is defragging.
1526 * Force cow if given extent needs to be defragged.
1528 if (BTRFS_I(inode)->defrag_bytes &&
1529 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1530 EXTENT_DEFRAG, 0, NULL))
1537 * extent_io.c call back to do delayed allocation processing
1539 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1540 u64 start, u64 end, int *page_started,
1541 unsigned long *nr_written)
1544 int force_cow = need_force_cow(inode, start, end);
1546 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1547 ret = run_delalloc_nocow(inode, locked_page, start, end,
1548 page_started, 1, nr_written);
1549 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1550 ret = run_delalloc_nocow(inode, locked_page, start, end,
1551 page_started, 0, nr_written);
1552 } else if (!inode_need_compress(inode)) {
1553 ret = cow_file_range(inode, locked_page, start, end,
1554 page_started, nr_written, 1);
1556 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1557 &BTRFS_I(inode)->runtime_flags);
1558 ret = cow_file_range_async(inode, locked_page, start, end,
1559 page_started, nr_written);
1564 static void btrfs_split_extent_hook(struct inode *inode,
1565 struct extent_state *orig, u64 split)
1569 /* not delalloc, ignore it */
1570 if (!(orig->state & EXTENT_DELALLOC))
1573 size = orig->end - orig->start + 1;
1574 if (size > BTRFS_MAX_EXTENT_SIZE) {
1579 * See the explanation in btrfs_merge_extent_hook, the same
1580 * applies here, just in reverse.
1582 new_size = orig->end - split + 1;
1583 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE);
1585 new_size = split - orig->start;
1586 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1587 BTRFS_MAX_EXTENT_SIZE);
1588 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1589 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1593 spin_lock(&BTRFS_I(inode)->lock);
1594 BTRFS_I(inode)->outstanding_extents++;
1595 spin_unlock(&BTRFS_I(inode)->lock);
1599 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1600 * extents so we can keep track of new extents that are just merged onto old
1601 * extents, such as when we are doing sequential writes, so we can properly
1602 * account for the metadata space we'll need.
1604 static void btrfs_merge_extent_hook(struct inode *inode,
1605 struct extent_state *new,
1606 struct extent_state *other)
1608 u64 new_size, old_size;
1611 /* not delalloc, ignore it */
1612 if (!(other->state & EXTENT_DELALLOC))
1615 if (new->start > other->start)
1616 new_size = new->end - other->start + 1;
1618 new_size = other->end - new->start + 1;
1620 /* we're not bigger than the max, unreserve the space and go */
1621 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1622 spin_lock(&BTRFS_I(inode)->lock);
1623 BTRFS_I(inode)->outstanding_extents--;
1624 spin_unlock(&BTRFS_I(inode)->lock);
1629 * We have to add up either side to figure out how many extents were
1630 * accounted for before we merged into one big extent. If the number of
1631 * extents we accounted for is <= the amount we need for the new range
1632 * then we can return, otherwise drop. Think of it like this
1636 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1637 * need 2 outstanding extents, on one side we have 1 and the other side
1638 * we have 1 so they are == and we can return. But in this case
1640 * [MAX_SIZE+4k][MAX_SIZE+4k]
1642 * Each range on their own accounts for 2 extents, but merged together
1643 * they are only 3 extents worth of accounting, so we need to drop in
1646 old_size = other->end - other->start + 1;
1647 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1648 BTRFS_MAX_EXTENT_SIZE);
1649 old_size = new->end - new->start + 1;
1650 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1651 BTRFS_MAX_EXTENT_SIZE);
1653 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1654 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1657 spin_lock(&BTRFS_I(inode)->lock);
1658 BTRFS_I(inode)->outstanding_extents--;
1659 spin_unlock(&BTRFS_I(inode)->lock);
1662 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1663 struct inode *inode)
1665 spin_lock(&root->delalloc_lock);
1666 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1667 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1668 &root->delalloc_inodes);
1669 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1670 &BTRFS_I(inode)->runtime_flags);
1671 root->nr_delalloc_inodes++;
1672 if (root->nr_delalloc_inodes == 1) {
1673 spin_lock(&root->fs_info->delalloc_root_lock);
1674 BUG_ON(!list_empty(&root->delalloc_root));
1675 list_add_tail(&root->delalloc_root,
1676 &root->fs_info->delalloc_roots);
1677 spin_unlock(&root->fs_info->delalloc_root_lock);
1680 spin_unlock(&root->delalloc_lock);
1683 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1684 struct inode *inode)
1686 spin_lock(&root->delalloc_lock);
1687 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1688 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1689 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1690 &BTRFS_I(inode)->runtime_flags);
1691 root->nr_delalloc_inodes--;
1692 if (!root->nr_delalloc_inodes) {
1693 spin_lock(&root->fs_info->delalloc_root_lock);
1694 BUG_ON(list_empty(&root->delalloc_root));
1695 list_del_init(&root->delalloc_root);
1696 spin_unlock(&root->fs_info->delalloc_root_lock);
1699 spin_unlock(&root->delalloc_lock);
1703 * extent_io.c set_bit_hook, used to track delayed allocation
1704 * bytes in this file, and to maintain the list of inodes that
1705 * have pending delalloc work to be done.
1707 static void btrfs_set_bit_hook(struct inode *inode,
1708 struct extent_state *state, unsigned *bits)
1711 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1714 * set_bit and clear bit hooks normally require _irqsave/restore
1715 * but in this case, we are only testing for the DELALLOC
1716 * bit, which is only set or cleared with irqs on
1718 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1719 struct btrfs_root *root = BTRFS_I(inode)->root;
1720 u64 len = state->end + 1 - state->start;
1721 bool do_list = !btrfs_is_free_space_inode(inode);
1723 if (*bits & EXTENT_FIRST_DELALLOC) {
1724 *bits &= ~EXTENT_FIRST_DELALLOC;
1726 spin_lock(&BTRFS_I(inode)->lock);
1727 BTRFS_I(inode)->outstanding_extents++;
1728 spin_unlock(&BTRFS_I(inode)->lock);
1731 /* For sanity tests */
1732 if (btrfs_test_is_dummy_root(root))
1735 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1736 root->fs_info->delalloc_batch);
1737 spin_lock(&BTRFS_I(inode)->lock);
1738 BTRFS_I(inode)->delalloc_bytes += len;
1739 if (*bits & EXTENT_DEFRAG)
1740 BTRFS_I(inode)->defrag_bytes += len;
1741 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1742 &BTRFS_I(inode)->runtime_flags))
1743 btrfs_add_delalloc_inodes(root, inode);
1744 spin_unlock(&BTRFS_I(inode)->lock);
1749 * extent_io.c clear_bit_hook, see set_bit_hook for why
1751 static void btrfs_clear_bit_hook(struct inode *inode,
1752 struct extent_state *state,
1755 u64 len = state->end + 1 - state->start;
1756 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1757 BTRFS_MAX_EXTENT_SIZE);
1759 spin_lock(&BTRFS_I(inode)->lock);
1760 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1761 BTRFS_I(inode)->defrag_bytes -= len;
1762 spin_unlock(&BTRFS_I(inode)->lock);
1765 * set_bit and clear bit hooks normally require _irqsave/restore
1766 * but in this case, we are only testing for the DELALLOC
1767 * bit, which is only set or cleared with irqs on
1769 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1770 struct btrfs_root *root = BTRFS_I(inode)->root;
1771 bool do_list = !btrfs_is_free_space_inode(inode);
1773 if (*bits & EXTENT_FIRST_DELALLOC) {
1774 *bits &= ~EXTENT_FIRST_DELALLOC;
1775 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1776 spin_lock(&BTRFS_I(inode)->lock);
1777 BTRFS_I(inode)->outstanding_extents -= num_extents;
1778 spin_unlock(&BTRFS_I(inode)->lock);
1782 * We don't reserve metadata space for space cache inodes so we
1783 * don't need to call dellalloc_release_metadata if there is an
1786 if (*bits & EXTENT_DO_ACCOUNTING &&
1787 root != root->fs_info->tree_root)
1788 btrfs_delalloc_release_metadata(inode, len);
1790 /* For sanity tests. */
1791 if (btrfs_test_is_dummy_root(root))
1794 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1795 && do_list && !(state->state & EXTENT_NORESERVE))
1796 btrfs_free_reserved_data_space_noquota(inode,
1799 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1800 root->fs_info->delalloc_batch);
1801 spin_lock(&BTRFS_I(inode)->lock);
1802 BTRFS_I(inode)->delalloc_bytes -= len;
1803 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1804 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1805 &BTRFS_I(inode)->runtime_flags))
1806 btrfs_del_delalloc_inode(root, inode);
1807 spin_unlock(&BTRFS_I(inode)->lock);
1812 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1813 * we don't create bios that span stripes or chunks
1815 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1816 size_t size, struct bio *bio,
1817 unsigned long bio_flags)
1819 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1820 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1825 if (bio_flags & EXTENT_BIO_COMPRESSED)
1828 length = bio->bi_iter.bi_size;
1829 map_length = length;
1830 ret = btrfs_map_block(root->fs_info, rw, logical,
1831 &map_length, NULL, 0);
1832 /* Will always return 0 with map_multi == NULL */
1834 if (map_length < length + size)
1840 * in order to insert checksums into the metadata in large chunks,
1841 * we wait until bio submission time. All the pages in the bio are
1842 * checksummed and sums are attached onto the ordered extent record.
1844 * At IO completion time the cums attached on the ordered extent record
1845 * are inserted into the btree
1847 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1848 struct bio *bio, int mirror_num,
1849 unsigned long bio_flags,
1852 struct btrfs_root *root = BTRFS_I(inode)->root;
1855 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1856 BUG_ON(ret); /* -ENOMEM */
1861 * in order to insert checksums into the metadata in large chunks,
1862 * we wait until bio submission time. All the pages in the bio are
1863 * checksummed and sums are attached onto the ordered extent record.
1865 * At IO completion time the cums attached on the ordered extent record
1866 * are inserted into the btree
1868 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1869 int mirror_num, unsigned long bio_flags,
1872 struct btrfs_root *root = BTRFS_I(inode)->root;
1875 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1877 bio->bi_error = ret;
1884 * extent_io.c submission hook. This does the right thing for csum calculation
1885 * on write, or reading the csums from the tree before a read
1887 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1888 int mirror_num, unsigned long bio_flags,
1891 struct btrfs_root *root = BTRFS_I(inode)->root;
1892 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1895 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1897 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1899 if (btrfs_is_free_space_inode(inode))
1900 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1902 if (!(rw & REQ_WRITE)) {
1903 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1907 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1908 ret = btrfs_submit_compressed_read(inode, bio,
1912 } else if (!skip_sum) {
1913 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1918 } else if (async && !skip_sum) {
1919 /* csum items have already been cloned */
1920 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1922 /* we're doing a write, do the async checksumming */
1923 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1924 inode, rw, bio, mirror_num,
1925 bio_flags, bio_offset,
1926 __btrfs_submit_bio_start,
1927 __btrfs_submit_bio_done);
1929 } else if (!skip_sum) {
1930 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1936 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1940 bio->bi_error = ret;
1947 * given a list of ordered sums record them in the inode. This happens
1948 * at IO completion time based on sums calculated at bio submission time.
1950 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1951 struct inode *inode, u64 file_offset,
1952 struct list_head *list)
1954 struct btrfs_ordered_sum *sum;
1956 list_for_each_entry(sum, list, list) {
1957 trans->adding_csums = 1;
1958 btrfs_csum_file_blocks(trans,
1959 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1960 trans->adding_csums = 0;
1965 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1966 struct extent_state **cached_state)
1968 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1969 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1970 cached_state, GFP_NOFS);
1973 /* see btrfs_writepage_start_hook for details on why this is required */
1974 struct btrfs_writepage_fixup {
1976 struct btrfs_work work;
1979 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1981 struct btrfs_writepage_fixup *fixup;
1982 struct btrfs_ordered_extent *ordered;
1983 struct extent_state *cached_state = NULL;
1985 struct inode *inode;
1990 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1994 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1995 ClearPageChecked(page);
1999 inode = page->mapping->host;
2000 page_start = page_offset(page);
2001 page_end = page_offset(page) + PAGE_SIZE - 1;
2003 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2006 /* already ordered? We're done */
2007 if (PagePrivate2(page))
2010 ordered = btrfs_lookup_ordered_range(inode, page_start,
2013 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2014 page_end, &cached_state, GFP_NOFS);
2016 btrfs_start_ordered_extent(inode, ordered, 1);
2017 btrfs_put_ordered_extent(ordered);
2021 ret = btrfs_delalloc_reserve_space(inode, page_start,
2024 mapping_set_error(page->mapping, ret);
2025 end_extent_writepage(page, ret, page_start, page_end);
2026 ClearPageChecked(page);
2030 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2031 ClearPageChecked(page);
2032 set_page_dirty(page);
2034 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2035 &cached_state, GFP_NOFS);
2043 * There are a few paths in the higher layers of the kernel that directly
2044 * set the page dirty bit without asking the filesystem if it is a
2045 * good idea. This causes problems because we want to make sure COW
2046 * properly happens and the data=ordered rules are followed.
2048 * In our case any range that doesn't have the ORDERED bit set
2049 * hasn't been properly setup for IO. We kick off an async process
2050 * to fix it up. The async helper will wait for ordered extents, set
2051 * the delalloc bit and make it safe to write the page.
2053 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2055 struct inode *inode = page->mapping->host;
2056 struct btrfs_writepage_fixup *fixup;
2057 struct btrfs_root *root = BTRFS_I(inode)->root;
2059 /* this page is properly in the ordered list */
2060 if (TestClearPagePrivate2(page))
2063 if (PageChecked(page))
2066 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2070 SetPageChecked(page);
2072 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2073 btrfs_writepage_fixup_worker, NULL, NULL);
2075 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2079 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2080 struct inode *inode, u64 file_pos,
2081 u64 disk_bytenr, u64 disk_num_bytes,
2082 u64 num_bytes, u64 ram_bytes,
2083 u8 compression, u8 encryption,
2084 u16 other_encoding, int extent_type)
2086 struct btrfs_root *root = BTRFS_I(inode)->root;
2087 struct btrfs_file_extent_item *fi;
2088 struct btrfs_path *path;
2089 struct extent_buffer *leaf;
2090 struct btrfs_key ins;
2091 int extent_inserted = 0;
2094 path = btrfs_alloc_path();
2099 * we may be replacing one extent in the tree with another.
2100 * The new extent is pinned in the extent map, and we don't want
2101 * to drop it from the cache until it is completely in the btree.
2103 * So, tell btrfs_drop_extents to leave this extent in the cache.
2104 * the caller is expected to unpin it and allow it to be merged
2107 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2108 file_pos + num_bytes, NULL, 0,
2109 1, sizeof(*fi), &extent_inserted);
2113 if (!extent_inserted) {
2114 ins.objectid = btrfs_ino(inode);
2115 ins.offset = file_pos;
2116 ins.type = BTRFS_EXTENT_DATA_KEY;
2118 path->leave_spinning = 1;
2119 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2124 leaf = path->nodes[0];
2125 fi = btrfs_item_ptr(leaf, path->slots[0],
2126 struct btrfs_file_extent_item);
2127 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2128 btrfs_set_file_extent_type(leaf, fi, extent_type);
2129 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2130 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2131 btrfs_set_file_extent_offset(leaf, fi, 0);
2132 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2133 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2134 btrfs_set_file_extent_compression(leaf, fi, compression);
2135 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2136 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2138 btrfs_mark_buffer_dirty(leaf);
2139 btrfs_release_path(path);
2141 inode_add_bytes(inode, num_bytes);
2143 ins.objectid = disk_bytenr;
2144 ins.offset = disk_num_bytes;
2145 ins.type = BTRFS_EXTENT_ITEM_KEY;
2146 ret = btrfs_alloc_reserved_file_extent(trans, root,
2147 root->root_key.objectid,
2148 btrfs_ino(inode), file_pos,
2151 * Release the reserved range from inode dirty range map, as it is
2152 * already moved into delayed_ref_head
2154 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2156 btrfs_free_path(path);
2161 /* snapshot-aware defrag */
2162 struct sa_defrag_extent_backref {
2163 struct rb_node node;
2164 struct old_sa_defrag_extent *old;
2173 struct old_sa_defrag_extent {
2174 struct list_head list;
2175 struct new_sa_defrag_extent *new;
2184 struct new_sa_defrag_extent {
2185 struct rb_root root;
2186 struct list_head head;
2187 struct btrfs_path *path;
2188 struct inode *inode;
2196 static int backref_comp(struct sa_defrag_extent_backref *b1,
2197 struct sa_defrag_extent_backref *b2)
2199 if (b1->root_id < b2->root_id)
2201 else if (b1->root_id > b2->root_id)
2204 if (b1->inum < b2->inum)
2206 else if (b1->inum > b2->inum)
2209 if (b1->file_pos < b2->file_pos)
2211 else if (b1->file_pos > b2->file_pos)
2215 * [------------------------------] ===> (a range of space)
2216 * |<--->| |<---->| =============> (fs/file tree A)
2217 * |<---------------------------->| ===> (fs/file tree B)
2219 * A range of space can refer to two file extents in one tree while
2220 * refer to only one file extent in another tree.
2222 * So we may process a disk offset more than one time(two extents in A)
2223 * and locate at the same extent(one extent in B), then insert two same
2224 * backrefs(both refer to the extent in B).
2229 static void backref_insert(struct rb_root *root,
2230 struct sa_defrag_extent_backref *backref)
2232 struct rb_node **p = &root->rb_node;
2233 struct rb_node *parent = NULL;
2234 struct sa_defrag_extent_backref *entry;
2239 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2241 ret = backref_comp(backref, entry);
2245 p = &(*p)->rb_right;
2248 rb_link_node(&backref->node, parent, p);
2249 rb_insert_color(&backref->node, root);
2253 * Note the backref might has changed, and in this case we just return 0.
2255 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2258 struct btrfs_file_extent_item *extent;
2259 struct btrfs_fs_info *fs_info;
2260 struct old_sa_defrag_extent *old = ctx;
2261 struct new_sa_defrag_extent *new = old->new;
2262 struct btrfs_path *path = new->path;
2263 struct btrfs_key key;
2264 struct btrfs_root *root;
2265 struct sa_defrag_extent_backref *backref;
2266 struct extent_buffer *leaf;
2267 struct inode *inode = new->inode;
2273 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2274 inum == btrfs_ino(inode))
2277 key.objectid = root_id;
2278 key.type = BTRFS_ROOT_ITEM_KEY;
2279 key.offset = (u64)-1;
2281 fs_info = BTRFS_I(inode)->root->fs_info;
2282 root = btrfs_read_fs_root_no_name(fs_info, &key);
2284 if (PTR_ERR(root) == -ENOENT)
2287 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2288 inum, offset, root_id);
2289 return PTR_ERR(root);
2292 key.objectid = inum;
2293 key.type = BTRFS_EXTENT_DATA_KEY;
2294 if (offset > (u64)-1 << 32)
2297 key.offset = offset;
2299 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2300 if (WARN_ON(ret < 0))
2307 leaf = path->nodes[0];
2308 slot = path->slots[0];
2310 if (slot >= btrfs_header_nritems(leaf)) {
2311 ret = btrfs_next_leaf(root, path);
2314 } else if (ret > 0) {
2323 btrfs_item_key_to_cpu(leaf, &key, slot);
2325 if (key.objectid > inum)
2328 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2331 extent = btrfs_item_ptr(leaf, slot,
2332 struct btrfs_file_extent_item);
2334 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2338 * 'offset' refers to the exact key.offset,
2339 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2340 * (key.offset - extent_offset).
2342 if (key.offset != offset)
2345 extent_offset = btrfs_file_extent_offset(leaf, extent);
2346 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2348 if (extent_offset >= old->extent_offset + old->offset +
2349 old->len || extent_offset + num_bytes <=
2350 old->extent_offset + old->offset)
2355 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2361 backref->root_id = root_id;
2362 backref->inum = inum;
2363 backref->file_pos = offset;
2364 backref->num_bytes = num_bytes;
2365 backref->extent_offset = extent_offset;
2366 backref->generation = btrfs_file_extent_generation(leaf, extent);
2368 backref_insert(&new->root, backref);
2371 btrfs_release_path(path);
2376 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2377 struct new_sa_defrag_extent *new)
2379 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2380 struct old_sa_defrag_extent *old, *tmp;
2385 list_for_each_entry_safe(old, tmp, &new->head, list) {
2386 ret = iterate_inodes_from_logical(old->bytenr +
2387 old->extent_offset, fs_info,
2388 path, record_one_backref,
2390 if (ret < 0 && ret != -ENOENT)
2393 /* no backref to be processed for this extent */
2395 list_del(&old->list);
2400 if (list_empty(&new->head))
2406 static int relink_is_mergable(struct extent_buffer *leaf,
2407 struct btrfs_file_extent_item *fi,
2408 struct new_sa_defrag_extent *new)
2410 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2413 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2416 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2419 if (btrfs_file_extent_encryption(leaf, fi) ||
2420 btrfs_file_extent_other_encoding(leaf, fi))
2427 * Note the backref might has changed, and in this case we just return 0.
2429 static noinline int relink_extent_backref(struct btrfs_path *path,
2430 struct sa_defrag_extent_backref *prev,
2431 struct sa_defrag_extent_backref *backref)
2433 struct btrfs_file_extent_item *extent;
2434 struct btrfs_file_extent_item *item;
2435 struct btrfs_ordered_extent *ordered;
2436 struct btrfs_trans_handle *trans;
2437 struct btrfs_fs_info *fs_info;
2438 struct btrfs_root *root;
2439 struct btrfs_key key;
2440 struct extent_buffer *leaf;
2441 struct old_sa_defrag_extent *old = backref->old;
2442 struct new_sa_defrag_extent *new = old->new;
2443 struct inode *src_inode = new->inode;
2444 struct inode *inode;
2445 struct extent_state *cached = NULL;
2454 if (prev && prev->root_id == backref->root_id &&
2455 prev->inum == backref->inum &&
2456 prev->file_pos + prev->num_bytes == backref->file_pos)
2459 /* step 1: get root */
2460 key.objectid = backref->root_id;
2461 key.type = BTRFS_ROOT_ITEM_KEY;
2462 key.offset = (u64)-1;
2464 fs_info = BTRFS_I(src_inode)->root->fs_info;
2465 index = srcu_read_lock(&fs_info->subvol_srcu);
2467 root = btrfs_read_fs_root_no_name(fs_info, &key);
2469 srcu_read_unlock(&fs_info->subvol_srcu, index);
2470 if (PTR_ERR(root) == -ENOENT)
2472 return PTR_ERR(root);
2475 if (btrfs_root_readonly(root)) {
2476 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480 /* step 2: get inode */
2481 key.objectid = backref->inum;
2482 key.type = BTRFS_INODE_ITEM_KEY;
2485 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2486 if (IS_ERR(inode)) {
2487 srcu_read_unlock(&fs_info->subvol_srcu, index);
2491 srcu_read_unlock(&fs_info->subvol_srcu, index);
2493 /* step 3: relink backref */
2494 lock_start = backref->file_pos;
2495 lock_end = backref->file_pos + backref->num_bytes - 1;
2496 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2499 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2501 btrfs_put_ordered_extent(ordered);
2505 trans = btrfs_join_transaction(root);
2506 if (IS_ERR(trans)) {
2507 ret = PTR_ERR(trans);
2511 key.objectid = backref->inum;
2512 key.type = BTRFS_EXTENT_DATA_KEY;
2513 key.offset = backref->file_pos;
2515 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2518 } else if (ret > 0) {
2523 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2524 struct btrfs_file_extent_item);
2526 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2527 backref->generation)
2530 btrfs_release_path(path);
2532 start = backref->file_pos;
2533 if (backref->extent_offset < old->extent_offset + old->offset)
2534 start += old->extent_offset + old->offset -
2535 backref->extent_offset;
2537 len = min(backref->extent_offset + backref->num_bytes,
2538 old->extent_offset + old->offset + old->len);
2539 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2541 ret = btrfs_drop_extents(trans, root, inode, start,
2546 key.objectid = btrfs_ino(inode);
2547 key.type = BTRFS_EXTENT_DATA_KEY;
2550 path->leave_spinning = 1;
2552 struct btrfs_file_extent_item *fi;
2554 struct btrfs_key found_key;
2556 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2561 leaf = path->nodes[0];
2562 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2564 fi = btrfs_item_ptr(leaf, path->slots[0],
2565 struct btrfs_file_extent_item);
2566 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2568 if (extent_len + found_key.offset == start &&
2569 relink_is_mergable(leaf, fi, new)) {
2570 btrfs_set_file_extent_num_bytes(leaf, fi,
2572 btrfs_mark_buffer_dirty(leaf);
2573 inode_add_bytes(inode, len);
2579 btrfs_release_path(path);
2584 ret = btrfs_insert_empty_item(trans, root, path, &key,
2587 btrfs_abort_transaction(trans, root, ret);
2591 leaf = path->nodes[0];
2592 item = btrfs_item_ptr(leaf, path->slots[0],
2593 struct btrfs_file_extent_item);
2594 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2595 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2596 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2597 btrfs_set_file_extent_num_bytes(leaf, item, len);
2598 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2599 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2600 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2601 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2602 btrfs_set_file_extent_encryption(leaf, item, 0);
2603 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2605 btrfs_mark_buffer_dirty(leaf);
2606 inode_add_bytes(inode, len);
2607 btrfs_release_path(path);
2609 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2611 backref->root_id, backref->inum,
2612 new->file_pos); /* start - extent_offset */
2614 btrfs_abort_transaction(trans, root, ret);
2620 btrfs_release_path(path);
2621 path->leave_spinning = 0;
2622 btrfs_end_transaction(trans, root);
2624 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2630 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2632 struct old_sa_defrag_extent *old, *tmp;
2637 list_for_each_entry_safe(old, tmp, &new->head, list) {
2643 static void relink_file_extents(struct new_sa_defrag_extent *new)
2645 struct btrfs_path *path;
2646 struct sa_defrag_extent_backref *backref;
2647 struct sa_defrag_extent_backref *prev = NULL;
2648 struct inode *inode;
2649 struct btrfs_root *root;
2650 struct rb_node *node;
2654 root = BTRFS_I(inode)->root;
2656 path = btrfs_alloc_path();
2660 if (!record_extent_backrefs(path, new)) {
2661 btrfs_free_path(path);
2664 btrfs_release_path(path);
2667 node = rb_first(&new->root);
2670 rb_erase(node, &new->root);
2672 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2674 ret = relink_extent_backref(path, prev, backref);
2687 btrfs_free_path(path);
2689 free_sa_defrag_extent(new);
2691 atomic_dec(&root->fs_info->defrag_running);
2692 wake_up(&root->fs_info->transaction_wait);
2695 static struct new_sa_defrag_extent *
2696 record_old_file_extents(struct inode *inode,
2697 struct btrfs_ordered_extent *ordered)
2699 struct btrfs_root *root = BTRFS_I(inode)->root;
2700 struct btrfs_path *path;
2701 struct btrfs_key key;
2702 struct old_sa_defrag_extent *old;
2703 struct new_sa_defrag_extent *new;
2706 new = kmalloc(sizeof(*new), GFP_NOFS);
2711 new->file_pos = ordered->file_offset;
2712 new->len = ordered->len;
2713 new->bytenr = ordered->start;
2714 new->disk_len = ordered->disk_len;
2715 new->compress_type = ordered->compress_type;
2716 new->root = RB_ROOT;
2717 INIT_LIST_HEAD(&new->head);
2719 path = btrfs_alloc_path();
2723 key.objectid = btrfs_ino(inode);
2724 key.type = BTRFS_EXTENT_DATA_KEY;
2725 key.offset = new->file_pos;
2727 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2730 if (ret > 0 && path->slots[0] > 0)
2733 /* find out all the old extents for the file range */
2735 struct btrfs_file_extent_item *extent;
2736 struct extent_buffer *l;
2745 slot = path->slots[0];
2747 if (slot >= btrfs_header_nritems(l)) {
2748 ret = btrfs_next_leaf(root, path);
2756 btrfs_item_key_to_cpu(l, &key, slot);
2758 if (key.objectid != btrfs_ino(inode))
2760 if (key.type != BTRFS_EXTENT_DATA_KEY)
2762 if (key.offset >= new->file_pos + new->len)
2765 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2767 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2768 if (key.offset + num_bytes < new->file_pos)
2771 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2775 extent_offset = btrfs_file_extent_offset(l, extent);
2777 old = kmalloc(sizeof(*old), GFP_NOFS);
2781 offset = max(new->file_pos, key.offset);
2782 end = min(new->file_pos + new->len, key.offset + num_bytes);
2784 old->bytenr = disk_bytenr;
2785 old->extent_offset = extent_offset;
2786 old->offset = offset - key.offset;
2787 old->len = end - offset;
2790 list_add_tail(&old->list, &new->head);
2796 btrfs_free_path(path);
2797 atomic_inc(&root->fs_info->defrag_running);
2802 btrfs_free_path(path);
2804 free_sa_defrag_extent(new);
2808 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2811 struct btrfs_block_group_cache *cache;
2813 cache = btrfs_lookup_block_group(root->fs_info, start);
2816 spin_lock(&cache->lock);
2817 cache->delalloc_bytes -= len;
2818 spin_unlock(&cache->lock);
2820 btrfs_put_block_group(cache);
2823 /* as ordered data IO finishes, this gets called so we can finish
2824 * an ordered extent if the range of bytes in the file it covers are
2827 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2829 struct inode *inode = ordered_extent->inode;
2830 struct btrfs_root *root = BTRFS_I(inode)->root;
2831 struct btrfs_trans_handle *trans = NULL;
2832 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2833 struct extent_state *cached_state = NULL;
2834 struct new_sa_defrag_extent *new = NULL;
2835 int compress_type = 0;
2837 u64 logical_len = ordered_extent->len;
2839 bool truncated = false;
2841 nolock = btrfs_is_free_space_inode(inode);
2843 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2848 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2849 ordered_extent->file_offset +
2850 ordered_extent->len - 1);
2852 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2854 logical_len = ordered_extent->truncated_len;
2855 /* Truncated the entire extent, don't bother adding */
2860 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2861 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2864 * For mwrite(mmap + memset to write) case, we still reserve
2865 * space for NOCOW range.
2866 * As NOCOW won't cause a new delayed ref, just free the space
2868 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2869 ordered_extent->len);
2870 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2872 trans = btrfs_join_transaction_nolock(root);
2874 trans = btrfs_join_transaction(root);
2875 if (IS_ERR(trans)) {
2876 ret = PTR_ERR(trans);
2880 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2881 ret = btrfs_update_inode_fallback(trans, root, inode);
2882 if (ret) /* -ENOMEM or corruption */
2883 btrfs_abort_transaction(trans, root, ret);
2887 lock_extent_bits(io_tree, ordered_extent->file_offset,
2888 ordered_extent->file_offset + ordered_extent->len - 1,
2891 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2892 ordered_extent->file_offset + ordered_extent->len - 1,
2893 EXTENT_DEFRAG, 1, cached_state);
2895 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2896 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2897 /* the inode is shared */
2898 new = record_old_file_extents(inode, ordered_extent);
2900 clear_extent_bit(io_tree, ordered_extent->file_offset,
2901 ordered_extent->file_offset + ordered_extent->len - 1,
2902 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2906 trans = btrfs_join_transaction_nolock(root);
2908 trans = btrfs_join_transaction(root);
2909 if (IS_ERR(trans)) {
2910 ret = PTR_ERR(trans);
2915 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2917 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2918 compress_type = ordered_extent->compress_type;
2919 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2920 BUG_ON(compress_type);
2921 ret = btrfs_mark_extent_written(trans, inode,
2922 ordered_extent->file_offset,
2923 ordered_extent->file_offset +
2926 BUG_ON(root == root->fs_info->tree_root);
2927 ret = insert_reserved_file_extent(trans, inode,
2928 ordered_extent->file_offset,
2929 ordered_extent->start,
2930 ordered_extent->disk_len,
2931 logical_len, logical_len,
2932 compress_type, 0, 0,
2933 BTRFS_FILE_EXTENT_REG);
2935 btrfs_release_delalloc_bytes(root,
2936 ordered_extent->start,
2937 ordered_extent->disk_len);
2939 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2940 ordered_extent->file_offset, ordered_extent->len,
2943 btrfs_abort_transaction(trans, root, ret);
2947 add_pending_csums(trans, inode, ordered_extent->file_offset,
2948 &ordered_extent->list);
2950 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2951 ret = btrfs_update_inode_fallback(trans, root, inode);
2952 if (ret) { /* -ENOMEM or corruption */
2953 btrfs_abort_transaction(trans, root, ret);
2958 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2959 ordered_extent->file_offset +
2960 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2962 if (root != root->fs_info->tree_root)
2963 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2965 btrfs_end_transaction(trans, root);
2967 if (ret || truncated) {
2971 start = ordered_extent->file_offset + logical_len;
2973 start = ordered_extent->file_offset;
2974 end = ordered_extent->file_offset + ordered_extent->len - 1;
2975 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2977 /* Drop the cache for the part of the extent we didn't write. */
2978 btrfs_drop_extent_cache(inode, start, end, 0);
2981 * If the ordered extent had an IOERR or something else went
2982 * wrong we need to return the space for this ordered extent
2983 * back to the allocator. We only free the extent in the
2984 * truncated case if we didn't write out the extent at all.
2986 if ((ret || !logical_len) &&
2987 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2988 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2989 btrfs_free_reserved_extent(root, ordered_extent->start,
2990 ordered_extent->disk_len, 1);
2995 * This needs to be done to make sure anybody waiting knows we are done
2996 * updating everything for this ordered extent.
2998 btrfs_remove_ordered_extent(inode, ordered_extent);
3000 /* for snapshot-aware defrag */
3003 free_sa_defrag_extent(new);
3004 atomic_dec(&root->fs_info->defrag_running);
3006 relink_file_extents(new);
3011 btrfs_put_ordered_extent(ordered_extent);
3012 /* once for the tree */
3013 btrfs_put_ordered_extent(ordered_extent);
3018 static void finish_ordered_fn(struct btrfs_work *work)
3020 struct btrfs_ordered_extent *ordered_extent;
3021 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3022 btrfs_finish_ordered_io(ordered_extent);
3025 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3026 struct extent_state *state, int uptodate)
3028 struct inode *inode = page->mapping->host;
3029 struct btrfs_root *root = BTRFS_I(inode)->root;
3030 struct btrfs_ordered_extent *ordered_extent = NULL;
3031 struct btrfs_workqueue *wq;
3032 btrfs_work_func_t func;
3034 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3036 ClearPagePrivate2(page);
3037 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3038 end - start + 1, uptodate))
3041 if (btrfs_is_free_space_inode(inode)) {
3042 wq = root->fs_info->endio_freespace_worker;
3043 func = btrfs_freespace_write_helper;
3045 wq = root->fs_info->endio_write_workers;
3046 func = btrfs_endio_write_helper;
3049 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3051 btrfs_queue_work(wq, &ordered_extent->work);
3056 static int __readpage_endio_check(struct inode *inode,
3057 struct btrfs_io_bio *io_bio,
3058 int icsum, struct page *page,
3059 int pgoff, u64 start, size_t len)
3065 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3067 kaddr = kmap_atomic(page);
3068 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3069 btrfs_csum_final(csum, (char *)&csum);
3070 if (csum != csum_expected)
3073 kunmap_atomic(kaddr);
3076 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3077 "csum failed ino %llu off %llu csum %u expected csum %u",
3078 btrfs_ino(inode), start, csum, csum_expected);
3079 memset(kaddr + pgoff, 1, len);
3080 flush_dcache_page(page);
3081 kunmap_atomic(kaddr);
3082 if (csum_expected == 0)
3088 * when reads are done, we need to check csums to verify the data is correct
3089 * if there's a match, we allow the bio to finish. If not, the code in
3090 * extent_io.c will try to find good copies for us.
3092 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3093 u64 phy_offset, struct page *page,
3094 u64 start, u64 end, int mirror)
3096 size_t offset = start - page_offset(page);
3097 struct inode *inode = page->mapping->host;
3098 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3099 struct btrfs_root *root = BTRFS_I(inode)->root;
3101 if (PageChecked(page)) {
3102 ClearPageChecked(page);
3106 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3109 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3110 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3111 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3116 phy_offset >>= inode->i_sb->s_blocksize_bits;
3117 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3118 start, (size_t)(end - start + 1));
3121 void btrfs_add_delayed_iput(struct inode *inode)
3123 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3124 struct btrfs_inode *binode = BTRFS_I(inode);
3126 if (atomic_add_unless(&inode->i_count, -1, 1))
3129 spin_lock(&fs_info->delayed_iput_lock);
3130 if (binode->delayed_iput_count == 0) {
3131 ASSERT(list_empty(&binode->delayed_iput));
3132 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3134 binode->delayed_iput_count++;
3136 spin_unlock(&fs_info->delayed_iput_lock);
3139 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3141 struct btrfs_fs_info *fs_info = root->fs_info;
3143 spin_lock(&fs_info->delayed_iput_lock);
3144 while (!list_empty(&fs_info->delayed_iputs)) {
3145 struct btrfs_inode *inode;
3147 inode = list_first_entry(&fs_info->delayed_iputs,
3148 struct btrfs_inode, delayed_iput);
3149 if (inode->delayed_iput_count) {
3150 inode->delayed_iput_count--;
3151 list_move_tail(&inode->delayed_iput,
3152 &fs_info->delayed_iputs);
3154 list_del_init(&inode->delayed_iput);
3156 spin_unlock(&fs_info->delayed_iput_lock);
3157 iput(&inode->vfs_inode);
3158 spin_lock(&fs_info->delayed_iput_lock);
3160 spin_unlock(&fs_info->delayed_iput_lock);
3164 * This is called in transaction commit time. If there are no orphan
3165 * files in the subvolume, it removes orphan item and frees block_rsv
3168 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3169 struct btrfs_root *root)
3171 struct btrfs_block_rsv *block_rsv;
3174 if (atomic_read(&root->orphan_inodes) ||
3175 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3178 spin_lock(&root->orphan_lock);
3179 if (atomic_read(&root->orphan_inodes)) {
3180 spin_unlock(&root->orphan_lock);
3184 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3185 spin_unlock(&root->orphan_lock);
3189 block_rsv = root->orphan_block_rsv;
3190 root->orphan_block_rsv = NULL;
3191 spin_unlock(&root->orphan_lock);
3193 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3194 btrfs_root_refs(&root->root_item) > 0) {
3195 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3196 root->root_key.objectid);
3198 btrfs_abort_transaction(trans, root, ret);
3200 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3205 WARN_ON(block_rsv->size > 0);
3206 btrfs_free_block_rsv(root, block_rsv);
3211 * This creates an orphan entry for the given inode in case something goes
3212 * wrong in the middle of an unlink/truncate.
3214 * NOTE: caller of this function should reserve 5 units of metadata for
3217 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3219 struct btrfs_root *root = BTRFS_I(inode)->root;
3220 struct btrfs_block_rsv *block_rsv = NULL;
3225 if (!root->orphan_block_rsv) {
3226 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3231 spin_lock(&root->orphan_lock);
3232 if (!root->orphan_block_rsv) {
3233 root->orphan_block_rsv = block_rsv;
3234 } else if (block_rsv) {
3235 btrfs_free_block_rsv(root, block_rsv);
3239 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3240 &BTRFS_I(inode)->runtime_flags)) {
3243 * For proper ENOSPC handling, we should do orphan
3244 * cleanup when mounting. But this introduces backward
3245 * compatibility issue.
3247 if (!xchg(&root->orphan_item_inserted, 1))
3253 atomic_inc(&root->orphan_inodes);
3256 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3257 &BTRFS_I(inode)->runtime_flags))
3259 spin_unlock(&root->orphan_lock);
3261 /* grab metadata reservation from transaction handle */
3263 ret = btrfs_orphan_reserve_metadata(trans, inode);
3264 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3267 /* insert an orphan item to track this unlinked/truncated file */
3269 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3271 atomic_dec(&root->orphan_inodes);
3273 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3274 &BTRFS_I(inode)->runtime_flags);
3275 btrfs_orphan_release_metadata(inode);
3277 if (ret != -EEXIST) {
3278 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3279 &BTRFS_I(inode)->runtime_flags);
3280 btrfs_abort_transaction(trans, root, ret);
3287 /* insert an orphan item to track subvolume contains orphan files */
3289 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3290 root->root_key.objectid);
3291 if (ret && ret != -EEXIST) {
3292 btrfs_abort_transaction(trans, root, ret);
3300 * We have done the truncate/delete so we can go ahead and remove the orphan
3301 * item for this particular inode.
3303 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3304 struct inode *inode)
3306 struct btrfs_root *root = BTRFS_I(inode)->root;
3307 int delete_item = 0;
3308 int release_rsv = 0;
3311 spin_lock(&root->orphan_lock);
3312 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3313 &BTRFS_I(inode)->runtime_flags))
3316 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3317 &BTRFS_I(inode)->runtime_flags))
3319 spin_unlock(&root->orphan_lock);
3322 atomic_dec(&root->orphan_inodes);
3324 ret = btrfs_del_orphan_item(trans, root,
3329 btrfs_orphan_release_metadata(inode);
3335 * this cleans up any orphans that may be left on the list from the last use
3338 int btrfs_orphan_cleanup(struct btrfs_root *root)
3340 struct btrfs_path *path;
3341 struct extent_buffer *leaf;
3342 struct btrfs_key key, found_key;
3343 struct btrfs_trans_handle *trans;
3344 struct inode *inode;
3345 u64 last_objectid = 0;
3346 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3348 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3351 path = btrfs_alloc_path();
3356 path->reada = READA_BACK;
3358 key.objectid = BTRFS_ORPHAN_OBJECTID;
3359 key.type = BTRFS_ORPHAN_ITEM_KEY;
3360 key.offset = (u64)-1;
3363 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3368 * if ret == 0 means we found what we were searching for, which
3369 * is weird, but possible, so only screw with path if we didn't
3370 * find the key and see if we have stuff that matches
3374 if (path->slots[0] == 0)
3379 /* pull out the item */
3380 leaf = path->nodes[0];
3381 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3383 /* make sure the item matches what we want */
3384 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3386 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3389 /* release the path since we're done with it */
3390 btrfs_release_path(path);
3393 * this is where we are basically btrfs_lookup, without the
3394 * crossing root thing. we store the inode number in the
3395 * offset of the orphan item.
3398 if (found_key.offset == last_objectid) {
3399 btrfs_err(root->fs_info,
3400 "Error removing orphan entry, stopping orphan cleanup");
3405 last_objectid = found_key.offset;
3407 found_key.objectid = found_key.offset;
3408 found_key.type = BTRFS_INODE_ITEM_KEY;
3409 found_key.offset = 0;
3410 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3411 ret = PTR_ERR_OR_ZERO(inode);
3412 if (ret && ret != -ESTALE)
3415 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3416 struct btrfs_root *dead_root;
3417 struct btrfs_fs_info *fs_info = root->fs_info;
3418 int is_dead_root = 0;
3421 * this is an orphan in the tree root. Currently these
3422 * could come from 2 sources:
3423 * a) a snapshot deletion in progress
3424 * b) a free space cache inode
3425 * We need to distinguish those two, as the snapshot
3426 * orphan must not get deleted.
3427 * find_dead_roots already ran before us, so if this
3428 * is a snapshot deletion, we should find the root
3429 * in the dead_roots list
3431 spin_lock(&fs_info->trans_lock);
3432 list_for_each_entry(dead_root, &fs_info->dead_roots,
3434 if (dead_root->root_key.objectid ==
3435 found_key.objectid) {
3440 spin_unlock(&fs_info->trans_lock);
3442 /* prevent this orphan from being found again */
3443 key.offset = found_key.objectid - 1;
3448 * Inode is already gone but the orphan item is still there,
3449 * kill the orphan item.
3451 if (ret == -ESTALE) {
3452 trans = btrfs_start_transaction(root, 1);
3453 if (IS_ERR(trans)) {
3454 ret = PTR_ERR(trans);
3457 btrfs_debug(root->fs_info, "auto deleting %Lu",
3458 found_key.objectid);
3459 ret = btrfs_del_orphan_item(trans, root,
3460 found_key.objectid);
3461 btrfs_end_transaction(trans, root);
3468 * add this inode to the orphan list so btrfs_orphan_del does
3469 * the proper thing when we hit it
3471 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3472 &BTRFS_I(inode)->runtime_flags);
3473 atomic_inc(&root->orphan_inodes);
3475 /* if we have links, this was a truncate, lets do that */
3476 if (inode->i_nlink) {
3477 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3483 /* 1 for the orphan item deletion. */
3484 trans = btrfs_start_transaction(root, 1);
3485 if (IS_ERR(trans)) {
3487 ret = PTR_ERR(trans);
3490 ret = btrfs_orphan_add(trans, inode);
3491 btrfs_end_transaction(trans, root);
3497 ret = btrfs_truncate(inode);
3499 btrfs_orphan_del(NULL, inode);
3504 /* this will do delete_inode and everything for us */
3509 /* release the path since we're done with it */
3510 btrfs_release_path(path);
3512 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3514 if (root->orphan_block_rsv)
3515 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3518 if (root->orphan_block_rsv ||
3519 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3520 trans = btrfs_join_transaction(root);
3522 btrfs_end_transaction(trans, root);
3526 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3528 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3532 btrfs_err(root->fs_info,
3533 "could not do orphan cleanup %d", ret);
3534 btrfs_free_path(path);
3539 * very simple check to peek ahead in the leaf looking for xattrs. If we
3540 * don't find any xattrs, we know there can't be any acls.
3542 * slot is the slot the inode is in, objectid is the objectid of the inode
3544 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3545 int slot, u64 objectid,
3546 int *first_xattr_slot)
3548 u32 nritems = btrfs_header_nritems(leaf);
3549 struct btrfs_key found_key;
3550 static u64 xattr_access = 0;
3551 static u64 xattr_default = 0;
3554 if (!xattr_access) {
3555 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3556 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3557 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3558 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3562 *first_xattr_slot = -1;
3563 while (slot < nritems) {
3564 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3566 /* we found a different objectid, there must not be acls */
3567 if (found_key.objectid != objectid)
3570 /* we found an xattr, assume we've got an acl */
3571 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3572 if (*first_xattr_slot == -1)
3573 *first_xattr_slot = slot;
3574 if (found_key.offset == xattr_access ||
3575 found_key.offset == xattr_default)
3580 * we found a key greater than an xattr key, there can't
3581 * be any acls later on
3583 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3590 * it goes inode, inode backrefs, xattrs, extents,
3591 * so if there are a ton of hard links to an inode there can
3592 * be a lot of backrefs. Don't waste time searching too hard,
3593 * this is just an optimization
3598 /* we hit the end of the leaf before we found an xattr or
3599 * something larger than an xattr. We have to assume the inode
3602 if (*first_xattr_slot == -1)
3603 *first_xattr_slot = slot;
3608 * read an inode from the btree into the in-memory inode
3610 static void btrfs_read_locked_inode(struct inode *inode)
3612 struct btrfs_path *path;
3613 struct extent_buffer *leaf;
3614 struct btrfs_inode_item *inode_item;
3615 struct btrfs_root *root = BTRFS_I(inode)->root;
3616 struct btrfs_key location;
3621 bool filled = false;
3622 int first_xattr_slot;
3624 ret = btrfs_fill_inode(inode, &rdev);
3628 path = btrfs_alloc_path();
3632 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3634 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3638 leaf = path->nodes[0];
3643 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3644 struct btrfs_inode_item);
3645 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3646 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3647 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3648 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3649 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3651 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3652 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3654 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3655 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3657 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3658 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3660 BTRFS_I(inode)->i_otime.tv_sec =
3661 btrfs_timespec_sec(leaf, &inode_item->otime);
3662 BTRFS_I(inode)->i_otime.tv_nsec =
3663 btrfs_timespec_nsec(leaf, &inode_item->otime);
3665 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3666 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3667 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3669 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3670 inode->i_generation = BTRFS_I(inode)->generation;
3672 rdev = btrfs_inode_rdev(leaf, inode_item);
3674 BTRFS_I(inode)->index_cnt = (u64)-1;
3675 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3679 * If we were modified in the current generation and evicted from memory
3680 * and then re-read we need to do a full sync since we don't have any
3681 * idea about which extents were modified before we were evicted from
3684 * This is required for both inode re-read from disk and delayed inode
3685 * in delayed_nodes_tree.
3687 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3688 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3689 &BTRFS_I(inode)->runtime_flags);
3692 * We don't persist the id of the transaction where an unlink operation
3693 * against the inode was last made. So here we assume the inode might
3694 * have been evicted, and therefore the exact value of last_unlink_trans
3695 * lost, and set it to last_trans to avoid metadata inconsistencies
3696 * between the inode and its parent if the inode is fsync'ed and the log
3697 * replayed. For example, in the scenario:
3700 * ln mydir/foo mydir/bar
3703 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3704 * xfs_io -c fsync mydir/foo
3706 * mount fs, triggers fsync log replay
3708 * We must make sure that when we fsync our inode foo we also log its
3709 * parent inode, otherwise after log replay the parent still has the
3710 * dentry with the "bar" name but our inode foo has a link count of 1
3711 * and doesn't have an inode ref with the name "bar" anymore.
3713 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3714 * but it guarantees correctness at the expense of ocassional full
3715 * transaction commits on fsync if our inode is a directory, or if our
3716 * inode is not a directory, logging its parent unnecessarily.
3718 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3721 if (inode->i_nlink != 1 ||
3722 path->slots[0] >= btrfs_header_nritems(leaf))
3725 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3726 if (location.objectid != btrfs_ino(inode))
3729 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3730 if (location.type == BTRFS_INODE_REF_KEY) {
3731 struct btrfs_inode_ref *ref;
3733 ref = (struct btrfs_inode_ref *)ptr;
3734 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3735 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3736 struct btrfs_inode_extref *extref;
3738 extref = (struct btrfs_inode_extref *)ptr;
3739 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3744 * try to precache a NULL acl entry for files that don't have
3745 * any xattrs or acls
3747 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3748 btrfs_ino(inode), &first_xattr_slot);
3749 if (first_xattr_slot != -1) {
3750 path->slots[0] = first_xattr_slot;
3751 ret = btrfs_load_inode_props(inode, path);
3753 btrfs_err(root->fs_info,
3754 "error loading props for ino %llu (root %llu): %d",
3756 root->root_key.objectid, ret);
3758 btrfs_free_path(path);
3761 cache_no_acl(inode);
3763 switch (inode->i_mode & S_IFMT) {
3765 inode->i_mapping->a_ops = &btrfs_aops;
3766 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3767 inode->i_fop = &btrfs_file_operations;
3768 inode->i_op = &btrfs_file_inode_operations;
3771 inode->i_fop = &btrfs_dir_file_operations;
3772 if (root == root->fs_info->tree_root)
3773 inode->i_op = &btrfs_dir_ro_inode_operations;
3775 inode->i_op = &btrfs_dir_inode_operations;
3778 inode->i_op = &btrfs_symlink_inode_operations;
3779 inode_nohighmem(inode);
3780 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3783 inode->i_op = &btrfs_special_inode_operations;
3784 init_special_inode(inode, inode->i_mode, rdev);
3788 btrfs_update_iflags(inode);
3792 btrfs_free_path(path);
3793 make_bad_inode(inode);
3797 * given a leaf and an inode, copy the inode fields into the leaf
3799 static void fill_inode_item(struct btrfs_trans_handle *trans,
3800 struct extent_buffer *leaf,
3801 struct btrfs_inode_item *item,
3802 struct inode *inode)
3804 struct btrfs_map_token token;
3806 btrfs_init_map_token(&token);
3808 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3809 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3810 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3812 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3813 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3815 btrfs_set_token_timespec_sec(leaf, &item->atime,
3816 inode->i_atime.tv_sec, &token);
3817 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3818 inode->i_atime.tv_nsec, &token);
3820 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3821 inode->i_mtime.tv_sec, &token);
3822 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3823 inode->i_mtime.tv_nsec, &token);
3825 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3826 inode->i_ctime.tv_sec, &token);
3827 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3828 inode->i_ctime.tv_nsec, &token);
3830 btrfs_set_token_timespec_sec(leaf, &item->otime,
3831 BTRFS_I(inode)->i_otime.tv_sec, &token);
3832 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3833 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3835 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3837 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3839 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3840 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3841 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3842 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3843 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3847 * copy everything in the in-memory inode into the btree.
3849 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3850 struct btrfs_root *root, struct inode *inode)
3852 struct btrfs_inode_item *inode_item;
3853 struct btrfs_path *path;
3854 struct extent_buffer *leaf;
3857 path = btrfs_alloc_path();
3861 path->leave_spinning = 1;
3862 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3870 leaf = path->nodes[0];
3871 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3872 struct btrfs_inode_item);
3874 fill_inode_item(trans, leaf, inode_item, inode);
3875 btrfs_mark_buffer_dirty(leaf);
3876 btrfs_set_inode_last_trans(trans, inode);
3879 btrfs_free_path(path);
3884 * copy everything in the in-memory inode into the btree.
3886 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3887 struct btrfs_root *root, struct inode *inode)
3892 * If the inode is a free space inode, we can deadlock during commit
3893 * if we put it into the delayed code.
3895 * The data relocation inode should also be directly updated
3898 if (!btrfs_is_free_space_inode(inode)
3899 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3900 && !root->fs_info->log_root_recovering) {
3901 btrfs_update_root_times(trans, root);
3903 ret = btrfs_delayed_update_inode(trans, root, inode);
3905 btrfs_set_inode_last_trans(trans, inode);
3909 return btrfs_update_inode_item(trans, root, inode);
3912 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3913 struct btrfs_root *root,
3914 struct inode *inode)
3918 ret = btrfs_update_inode(trans, root, inode);
3920 return btrfs_update_inode_item(trans, root, inode);
3925 * unlink helper that gets used here in inode.c and in the tree logging
3926 * recovery code. It remove a link in a directory with a given name, and
3927 * also drops the back refs in the inode to the directory
3929 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3930 struct btrfs_root *root,
3931 struct inode *dir, struct inode *inode,
3932 const char *name, int name_len)
3934 struct btrfs_path *path;
3936 struct extent_buffer *leaf;
3937 struct btrfs_dir_item *di;
3938 struct btrfs_key key;
3940 u64 ino = btrfs_ino(inode);
3941 u64 dir_ino = btrfs_ino(dir);
3943 path = btrfs_alloc_path();
3949 path->leave_spinning = 1;
3950 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3951 name, name_len, -1);
3960 leaf = path->nodes[0];
3961 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3962 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3965 btrfs_release_path(path);
3968 * If we don't have dir index, we have to get it by looking up
3969 * the inode ref, since we get the inode ref, remove it directly,
3970 * it is unnecessary to do delayed deletion.
3972 * But if we have dir index, needn't search inode ref to get it.
3973 * Since the inode ref is close to the inode item, it is better
3974 * that we delay to delete it, and just do this deletion when
3975 * we update the inode item.
3977 if (BTRFS_I(inode)->dir_index) {
3978 ret = btrfs_delayed_delete_inode_ref(inode);
3980 index = BTRFS_I(inode)->dir_index;
3985 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3988 btrfs_info(root->fs_info,
3989 "failed to delete reference to %.*s, inode %llu parent %llu",
3990 name_len, name, ino, dir_ino);
3991 btrfs_abort_transaction(trans, root, ret);
3995 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3997 btrfs_abort_transaction(trans, root, ret);
4001 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4003 if (ret != 0 && ret != -ENOENT) {
4004 btrfs_abort_transaction(trans, root, ret);
4008 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4013 btrfs_abort_transaction(trans, root, ret);
4015 btrfs_free_path(path);
4019 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4020 inode_inc_iversion(inode);
4021 inode_inc_iversion(dir);
4022 inode->i_ctime = dir->i_mtime =
4023 dir->i_ctime = current_fs_time(inode->i_sb);
4024 ret = btrfs_update_inode(trans, root, dir);
4029 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4030 struct btrfs_root *root,
4031 struct inode *dir, struct inode *inode,
4032 const char *name, int name_len)
4035 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4038 ret = btrfs_update_inode(trans, root, inode);
4044 * helper to start transaction for unlink and rmdir.
4046 * unlink and rmdir are special in btrfs, they do not always free space, so
4047 * if we cannot make our reservations the normal way try and see if there is
4048 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4049 * allow the unlink to occur.
4051 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4053 struct btrfs_root *root = BTRFS_I(dir)->root;
4056 * 1 for the possible orphan item
4057 * 1 for the dir item
4058 * 1 for the dir index
4059 * 1 for the inode ref
4062 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4065 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4067 struct btrfs_root *root = BTRFS_I(dir)->root;
4068 struct btrfs_trans_handle *trans;
4069 struct inode *inode = d_inode(dentry);
4072 trans = __unlink_start_trans(dir);
4074 return PTR_ERR(trans);
4076 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4078 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4079 dentry->d_name.name, dentry->d_name.len);
4083 if (inode->i_nlink == 0) {
4084 ret = btrfs_orphan_add(trans, inode);
4090 btrfs_end_transaction(trans, root);
4091 btrfs_btree_balance_dirty(root);
4095 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4096 struct btrfs_root *root,
4097 struct inode *dir, u64 objectid,
4098 const char *name, int name_len)
4100 struct btrfs_path *path;
4101 struct extent_buffer *leaf;
4102 struct btrfs_dir_item *di;
4103 struct btrfs_key key;
4106 u64 dir_ino = btrfs_ino(dir);
4108 path = btrfs_alloc_path();
4112 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4113 name, name_len, -1);
4114 if (IS_ERR_OR_NULL(di)) {
4122 leaf = path->nodes[0];
4123 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4124 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4125 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4127 btrfs_abort_transaction(trans, root, ret);
4130 btrfs_release_path(path);
4132 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4133 objectid, root->root_key.objectid,
4134 dir_ino, &index, name, name_len);
4136 if (ret != -ENOENT) {
4137 btrfs_abort_transaction(trans, root, ret);
4140 di = btrfs_search_dir_index_item(root, path, dir_ino,
4142 if (IS_ERR_OR_NULL(di)) {
4147 btrfs_abort_transaction(trans, root, ret);
4151 leaf = path->nodes[0];
4152 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4153 btrfs_release_path(path);
4156 btrfs_release_path(path);
4158 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4160 btrfs_abort_transaction(trans, root, ret);
4164 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4165 inode_inc_iversion(dir);
4166 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4167 ret = btrfs_update_inode_fallback(trans, root, dir);
4169 btrfs_abort_transaction(trans, root, ret);
4171 btrfs_free_path(path);
4175 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4177 struct inode *inode = d_inode(dentry);
4179 struct btrfs_root *root = BTRFS_I(dir)->root;
4180 struct btrfs_trans_handle *trans;
4182 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4184 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4187 trans = __unlink_start_trans(dir);
4189 return PTR_ERR(trans);
4191 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4192 err = btrfs_unlink_subvol(trans, root, dir,
4193 BTRFS_I(inode)->location.objectid,
4194 dentry->d_name.name,
4195 dentry->d_name.len);
4199 err = btrfs_orphan_add(trans, inode);
4203 /* now the directory is empty */
4204 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4205 dentry->d_name.name, dentry->d_name.len);
4207 btrfs_i_size_write(inode, 0);
4209 btrfs_end_transaction(trans, root);
4210 btrfs_btree_balance_dirty(root);
4215 static int truncate_space_check(struct btrfs_trans_handle *trans,
4216 struct btrfs_root *root,
4222 * This is only used to apply pressure to the enospc system, we don't
4223 * intend to use this reservation at all.
4225 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4226 bytes_deleted *= root->nodesize;
4227 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4228 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4230 trace_btrfs_space_reservation(root->fs_info, "transaction",
4233 trans->bytes_reserved += bytes_deleted;
4239 static int truncate_inline_extent(struct inode *inode,
4240 struct btrfs_path *path,
4241 struct btrfs_key *found_key,
4245 struct extent_buffer *leaf = path->nodes[0];
4246 int slot = path->slots[0];
4247 struct btrfs_file_extent_item *fi;
4248 u32 size = (u32)(new_size - found_key->offset);
4249 struct btrfs_root *root = BTRFS_I(inode)->root;
4251 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4253 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4254 loff_t offset = new_size;
4255 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4258 * Zero out the remaining of the last page of our inline extent,
4259 * instead of directly truncating our inline extent here - that
4260 * would be much more complex (decompressing all the data, then
4261 * compressing the truncated data, which might be bigger than
4262 * the size of the inline extent, resize the extent, etc).
4263 * We release the path because to get the page we might need to
4264 * read the extent item from disk (data not in the page cache).
4266 btrfs_release_path(path);
4267 return btrfs_truncate_block(inode, offset, page_end - offset,
4271 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4272 size = btrfs_file_extent_calc_inline_size(size);
4273 btrfs_truncate_item(root, path, size, 1);
4275 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4276 inode_sub_bytes(inode, item_end + 1 - new_size);
4282 * this can truncate away extent items, csum items and directory items.
4283 * It starts at a high offset and removes keys until it can't find
4284 * any higher than new_size
4286 * csum items that cross the new i_size are truncated to the new size
4289 * min_type is the minimum key type to truncate down to. If set to 0, this
4290 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4292 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4293 struct btrfs_root *root,
4294 struct inode *inode,
4295 u64 new_size, u32 min_type)
4297 struct btrfs_path *path;
4298 struct extent_buffer *leaf;
4299 struct btrfs_file_extent_item *fi;
4300 struct btrfs_key key;
4301 struct btrfs_key found_key;
4302 u64 extent_start = 0;
4303 u64 extent_num_bytes = 0;
4304 u64 extent_offset = 0;
4306 u64 last_size = new_size;
4307 u32 found_type = (u8)-1;
4310 int pending_del_nr = 0;
4311 int pending_del_slot = 0;
4312 int extent_type = -1;
4315 u64 ino = btrfs_ino(inode);
4316 u64 bytes_deleted = 0;
4318 bool should_throttle = 0;
4319 bool should_end = 0;
4321 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4324 * for non-free space inodes and ref cows, we want to back off from
4327 if (!btrfs_is_free_space_inode(inode) &&
4328 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4331 path = btrfs_alloc_path();
4334 path->reada = READA_BACK;
4337 * We want to drop from the next block forward in case this new size is
4338 * not block aligned since we will be keeping the last block of the
4339 * extent just the way it is.
4341 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4342 root == root->fs_info->tree_root)
4343 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4344 root->sectorsize), (u64)-1, 0);
4347 * This function is also used to drop the items in the log tree before
4348 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4349 * it is used to drop the loged items. So we shouldn't kill the delayed
4352 if (min_type == 0 && root == BTRFS_I(inode)->root)
4353 btrfs_kill_delayed_inode_items(inode);
4356 key.offset = (u64)-1;
4361 * with a 16K leaf size and 128MB extents, you can actually queue
4362 * up a huge file in a single leaf. Most of the time that
4363 * bytes_deleted is > 0, it will be huge by the time we get here
4365 if (be_nice && bytes_deleted > SZ_32M) {
4366 if (btrfs_should_end_transaction(trans, root)) {
4373 path->leave_spinning = 1;
4374 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4381 /* there are no items in the tree for us to truncate, we're
4384 if (path->slots[0] == 0)
4391 leaf = path->nodes[0];
4392 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4393 found_type = found_key.type;
4395 if (found_key.objectid != ino)
4398 if (found_type < min_type)
4401 item_end = found_key.offset;
4402 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4403 fi = btrfs_item_ptr(leaf, path->slots[0],
4404 struct btrfs_file_extent_item);
4405 extent_type = btrfs_file_extent_type(leaf, fi);
4406 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4408 btrfs_file_extent_num_bytes(leaf, fi);
4409 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4410 item_end += btrfs_file_extent_inline_len(leaf,
4411 path->slots[0], fi);
4415 if (found_type > min_type) {
4418 if (item_end < new_size)
4420 if (found_key.offset >= new_size)
4426 /* FIXME, shrink the extent if the ref count is only 1 */
4427 if (found_type != BTRFS_EXTENT_DATA_KEY)
4431 last_size = found_key.offset;
4433 last_size = new_size;
4435 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4437 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4439 u64 orig_num_bytes =
4440 btrfs_file_extent_num_bytes(leaf, fi);
4441 extent_num_bytes = ALIGN(new_size -
4444 btrfs_set_file_extent_num_bytes(leaf, fi,
4446 num_dec = (orig_num_bytes -
4448 if (test_bit(BTRFS_ROOT_REF_COWS,
4451 inode_sub_bytes(inode, num_dec);
4452 btrfs_mark_buffer_dirty(leaf);
4455 btrfs_file_extent_disk_num_bytes(leaf,
4457 extent_offset = found_key.offset -
4458 btrfs_file_extent_offset(leaf, fi);
4460 /* FIXME blocksize != 4096 */
4461 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4462 if (extent_start != 0) {
4464 if (test_bit(BTRFS_ROOT_REF_COWS,
4466 inode_sub_bytes(inode, num_dec);
4469 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4471 * we can't truncate inline items that have had
4475 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4476 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4479 * Need to release path in order to truncate a
4480 * compressed extent. So delete any accumulated
4481 * extent items so far.
4483 if (btrfs_file_extent_compression(leaf, fi) !=
4484 BTRFS_COMPRESS_NONE && pending_del_nr) {
4485 err = btrfs_del_items(trans, root, path,
4489 btrfs_abort_transaction(trans,
4497 err = truncate_inline_extent(inode, path,
4502 btrfs_abort_transaction(trans,
4506 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4508 inode_sub_bytes(inode, item_end + 1 - new_size);
4513 if (!pending_del_nr) {
4514 /* no pending yet, add ourselves */
4515 pending_del_slot = path->slots[0];
4517 } else if (pending_del_nr &&
4518 path->slots[0] + 1 == pending_del_slot) {
4519 /* hop on the pending chunk */
4521 pending_del_slot = path->slots[0];
4528 should_throttle = 0;
4531 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4532 root == root->fs_info->tree_root)) {
4533 btrfs_set_path_blocking(path);
4534 bytes_deleted += extent_num_bytes;
4535 ret = btrfs_free_extent(trans, root, extent_start,
4536 extent_num_bytes, 0,
4537 btrfs_header_owner(leaf),
4538 ino, extent_offset);
4540 if (btrfs_should_throttle_delayed_refs(trans, root))
4541 btrfs_async_run_delayed_refs(root,
4542 trans->delayed_ref_updates * 2, 0);
4544 if (truncate_space_check(trans, root,
4545 extent_num_bytes)) {
4548 if (btrfs_should_throttle_delayed_refs(trans,
4550 should_throttle = 1;
4555 if (found_type == BTRFS_INODE_ITEM_KEY)
4558 if (path->slots[0] == 0 ||
4559 path->slots[0] != pending_del_slot ||
4560 should_throttle || should_end) {
4561 if (pending_del_nr) {
4562 ret = btrfs_del_items(trans, root, path,
4566 btrfs_abort_transaction(trans,
4572 btrfs_release_path(path);
4573 if (should_throttle) {
4574 unsigned long updates = trans->delayed_ref_updates;
4576 trans->delayed_ref_updates = 0;
4577 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4583 * if we failed to refill our space rsv, bail out
4584 * and let the transaction restart
4596 if (pending_del_nr) {
4597 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4600 btrfs_abort_transaction(trans, root, ret);
4603 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4604 btrfs_ordered_update_i_size(inode, last_size, NULL);
4606 btrfs_free_path(path);
4608 if (be_nice && bytes_deleted > SZ_32M) {
4609 unsigned long updates = trans->delayed_ref_updates;
4611 trans->delayed_ref_updates = 0;
4612 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4621 * btrfs_truncate_block - read, zero a chunk and write a block
4622 * @inode - inode that we're zeroing
4623 * @from - the offset to start zeroing
4624 * @len - the length to zero, 0 to zero the entire range respective to the
4626 * @front - zero up to the offset instead of from the offset on
4628 * This will find the block for the "from" offset and cow the block and zero the
4629 * part we want to zero. This is used with truncate and hole punching.
4631 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4634 struct address_space *mapping = inode->i_mapping;
4635 struct btrfs_root *root = BTRFS_I(inode)->root;
4636 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4637 struct btrfs_ordered_extent *ordered;
4638 struct extent_state *cached_state = NULL;
4640 u32 blocksize = root->sectorsize;
4641 pgoff_t index = from >> PAGE_SHIFT;
4642 unsigned offset = from & (blocksize - 1);
4644 gfp_t mask = btrfs_alloc_write_mask(mapping);
4649 if ((offset & (blocksize - 1)) == 0 &&
4650 (!len || ((len & (blocksize - 1)) == 0)))
4653 ret = btrfs_delalloc_reserve_space(inode,
4654 round_down(from, blocksize), blocksize);
4659 page = find_or_create_page(mapping, index, mask);
4661 btrfs_delalloc_release_space(inode,
4662 round_down(from, blocksize),
4668 block_start = round_down(from, blocksize);
4669 block_end = block_start + blocksize - 1;
4671 if (!PageUptodate(page)) {
4672 ret = btrfs_readpage(NULL, page);
4674 if (page->mapping != mapping) {
4679 if (!PageUptodate(page)) {
4684 wait_on_page_writeback(page);
4686 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4687 set_page_extent_mapped(page);
4689 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4691 unlock_extent_cached(io_tree, block_start, block_end,
4692 &cached_state, GFP_NOFS);
4695 btrfs_start_ordered_extent(inode, ordered, 1);
4696 btrfs_put_ordered_extent(ordered);
4700 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4701 EXTENT_DIRTY | EXTENT_DELALLOC |
4702 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4703 0, 0, &cached_state, GFP_NOFS);
4705 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4708 unlock_extent_cached(io_tree, block_start, block_end,
4709 &cached_state, GFP_NOFS);
4713 if (offset != blocksize) {
4715 len = blocksize - offset;
4718 memset(kaddr + (block_start - page_offset(page)),
4721 memset(kaddr + (block_start - page_offset(page)) + offset,
4723 flush_dcache_page(page);
4726 ClearPageChecked(page);
4727 set_page_dirty(page);
4728 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4733 btrfs_delalloc_release_space(inode, block_start,
4741 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4742 u64 offset, u64 len)
4744 struct btrfs_trans_handle *trans;
4748 * Still need to make sure the inode looks like it's been updated so
4749 * that any holes get logged if we fsync.
4751 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4752 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4753 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4754 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4759 * 1 - for the one we're dropping
4760 * 1 - for the one we're adding
4761 * 1 - for updating the inode.
4763 trans = btrfs_start_transaction(root, 3);
4765 return PTR_ERR(trans);
4767 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4769 btrfs_abort_transaction(trans, root, ret);
4770 btrfs_end_transaction(trans, root);
4774 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4775 0, 0, len, 0, len, 0, 0, 0);
4777 btrfs_abort_transaction(trans, root, ret);
4779 btrfs_update_inode(trans, root, inode);
4780 btrfs_end_transaction(trans, root);
4785 * This function puts in dummy file extents for the area we're creating a hole
4786 * for. So if we are truncating this file to a larger size we need to insert
4787 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4788 * the range between oldsize and size
4790 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4792 struct btrfs_root *root = BTRFS_I(inode)->root;
4793 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4794 struct extent_map *em = NULL;
4795 struct extent_state *cached_state = NULL;
4796 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4797 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4798 u64 block_end = ALIGN(size, root->sectorsize);
4805 * If our size started in the middle of a block we need to zero out the
4806 * rest of the block before we expand the i_size, otherwise we could
4807 * expose stale data.
4809 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4813 if (size <= hole_start)
4817 struct btrfs_ordered_extent *ordered;
4819 lock_extent_bits(io_tree, hole_start, block_end - 1,
4821 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4822 block_end - hole_start);
4825 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4826 &cached_state, GFP_NOFS);
4827 btrfs_start_ordered_extent(inode, ordered, 1);
4828 btrfs_put_ordered_extent(ordered);
4831 cur_offset = hole_start;
4833 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4834 block_end - cur_offset, 0);
4840 last_byte = min(extent_map_end(em), block_end);
4841 last_byte = ALIGN(last_byte , root->sectorsize);
4842 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4843 struct extent_map *hole_em;
4844 hole_size = last_byte - cur_offset;
4846 err = maybe_insert_hole(root, inode, cur_offset,
4850 btrfs_drop_extent_cache(inode, cur_offset,
4851 cur_offset + hole_size - 1, 0);
4852 hole_em = alloc_extent_map();
4854 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4855 &BTRFS_I(inode)->runtime_flags);
4858 hole_em->start = cur_offset;
4859 hole_em->len = hole_size;
4860 hole_em->orig_start = cur_offset;
4862 hole_em->block_start = EXTENT_MAP_HOLE;
4863 hole_em->block_len = 0;
4864 hole_em->orig_block_len = 0;
4865 hole_em->ram_bytes = hole_size;
4866 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4867 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4868 hole_em->generation = root->fs_info->generation;
4871 write_lock(&em_tree->lock);
4872 err = add_extent_mapping(em_tree, hole_em, 1);
4873 write_unlock(&em_tree->lock);
4876 btrfs_drop_extent_cache(inode, cur_offset,
4880 free_extent_map(hole_em);
4883 free_extent_map(em);
4885 cur_offset = last_byte;
4886 if (cur_offset >= block_end)
4889 free_extent_map(em);
4890 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4895 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4897 struct btrfs_root *root = BTRFS_I(inode)->root;
4898 struct btrfs_trans_handle *trans;
4899 loff_t oldsize = i_size_read(inode);
4900 loff_t newsize = attr->ia_size;
4901 int mask = attr->ia_valid;
4905 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4906 * special case where we need to update the times despite not having
4907 * these flags set. For all other operations the VFS set these flags
4908 * explicitly if it wants a timestamp update.
4910 if (newsize != oldsize) {
4911 inode_inc_iversion(inode);
4912 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4913 inode->i_ctime = inode->i_mtime =
4914 current_fs_time(inode->i_sb);
4917 if (newsize > oldsize) {
4919 * Don't do an expanding truncate while snapshoting is ongoing.
4920 * This is to ensure the snapshot captures a fully consistent
4921 * state of this file - if the snapshot captures this expanding
4922 * truncation, it must capture all writes that happened before
4925 btrfs_wait_for_snapshot_creation(root);
4926 ret = btrfs_cont_expand(inode, oldsize, newsize);
4928 btrfs_end_write_no_snapshoting(root);
4932 trans = btrfs_start_transaction(root, 1);
4933 if (IS_ERR(trans)) {
4934 btrfs_end_write_no_snapshoting(root);
4935 return PTR_ERR(trans);
4938 i_size_write(inode, newsize);
4939 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4940 pagecache_isize_extended(inode, oldsize, newsize);
4941 ret = btrfs_update_inode(trans, root, inode);
4942 btrfs_end_write_no_snapshoting(root);
4943 btrfs_end_transaction(trans, root);
4947 * We're truncating a file that used to have good data down to
4948 * zero. Make sure it gets into the ordered flush list so that
4949 * any new writes get down to disk quickly.
4952 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4953 &BTRFS_I(inode)->runtime_flags);
4956 * 1 for the orphan item we're going to add
4957 * 1 for the orphan item deletion.
4959 trans = btrfs_start_transaction(root, 2);
4961 return PTR_ERR(trans);
4964 * We need to do this in case we fail at _any_ point during the
4965 * actual truncate. Once we do the truncate_setsize we could
4966 * invalidate pages which forces any outstanding ordered io to
4967 * be instantly completed which will give us extents that need
4968 * to be truncated. If we fail to get an orphan inode down we
4969 * could have left over extents that were never meant to live,
4970 * so we need to garuntee from this point on that everything
4971 * will be consistent.
4973 ret = btrfs_orphan_add(trans, inode);
4974 btrfs_end_transaction(trans, root);
4978 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4979 truncate_setsize(inode, newsize);
4981 /* Disable nonlocked read DIO to avoid the end less truncate */
4982 btrfs_inode_block_unlocked_dio(inode);
4983 inode_dio_wait(inode);
4984 btrfs_inode_resume_unlocked_dio(inode);
4986 ret = btrfs_truncate(inode);
4987 if (ret && inode->i_nlink) {
4991 * failed to truncate, disk_i_size is only adjusted down
4992 * as we remove extents, so it should represent the true
4993 * size of the inode, so reset the in memory size and
4994 * delete our orphan entry.
4996 trans = btrfs_join_transaction(root);
4997 if (IS_ERR(trans)) {
4998 btrfs_orphan_del(NULL, inode);
5001 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5002 err = btrfs_orphan_del(trans, inode);
5004 btrfs_abort_transaction(trans, root, err);
5005 btrfs_end_transaction(trans, root);
5012 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5014 struct inode *inode = d_inode(dentry);
5015 struct btrfs_root *root = BTRFS_I(inode)->root;
5018 if (btrfs_root_readonly(root))
5021 err = inode_change_ok(inode, attr);
5025 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5026 err = btrfs_setsize(inode, attr);
5031 if (attr->ia_valid) {
5032 setattr_copy(inode, attr);
5033 inode_inc_iversion(inode);
5034 err = btrfs_dirty_inode(inode);
5036 if (!err && attr->ia_valid & ATTR_MODE)
5037 err = posix_acl_chmod(inode, inode->i_mode);
5044 * While truncating the inode pages during eviction, we get the VFS calling
5045 * btrfs_invalidatepage() against each page of the inode. This is slow because
5046 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5047 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5048 * extent_state structures over and over, wasting lots of time.
5050 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5051 * those expensive operations on a per page basis and do only the ordered io
5052 * finishing, while we release here the extent_map and extent_state structures,
5053 * without the excessive merging and splitting.
5055 static void evict_inode_truncate_pages(struct inode *inode)
5057 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5058 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5059 struct rb_node *node;
5061 ASSERT(inode->i_state & I_FREEING);
5062 truncate_inode_pages_final(&inode->i_data);
5064 write_lock(&map_tree->lock);
5065 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5066 struct extent_map *em;
5068 node = rb_first(&map_tree->map);
5069 em = rb_entry(node, struct extent_map, rb_node);
5070 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5071 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5072 remove_extent_mapping(map_tree, em);
5073 free_extent_map(em);
5074 if (need_resched()) {
5075 write_unlock(&map_tree->lock);
5077 write_lock(&map_tree->lock);
5080 write_unlock(&map_tree->lock);
5083 * Keep looping until we have no more ranges in the io tree.
5084 * We can have ongoing bios started by readpages (called from readahead)
5085 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5086 * still in progress (unlocked the pages in the bio but did not yet
5087 * unlocked the ranges in the io tree). Therefore this means some
5088 * ranges can still be locked and eviction started because before
5089 * submitting those bios, which are executed by a separate task (work
5090 * queue kthread), inode references (inode->i_count) were not taken
5091 * (which would be dropped in the end io callback of each bio).
5092 * Therefore here we effectively end up waiting for those bios and
5093 * anyone else holding locked ranges without having bumped the inode's
5094 * reference count - if we don't do it, when they access the inode's
5095 * io_tree to unlock a range it may be too late, leading to an
5096 * use-after-free issue.
5098 spin_lock(&io_tree->lock);
5099 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5100 struct extent_state *state;
5101 struct extent_state *cached_state = NULL;
5105 node = rb_first(&io_tree->state);
5106 state = rb_entry(node, struct extent_state, rb_node);
5107 start = state->start;
5109 spin_unlock(&io_tree->lock);
5111 lock_extent_bits(io_tree, start, end, &cached_state);
5114 * If still has DELALLOC flag, the extent didn't reach disk,
5115 * and its reserved space won't be freed by delayed_ref.
5116 * So we need to free its reserved space here.
5117 * (Refer to comment in btrfs_invalidatepage, case 2)
5119 * Note, end is the bytenr of last byte, so we need + 1 here.
5121 if (state->state & EXTENT_DELALLOC)
5122 btrfs_qgroup_free_data(inode, start, end - start + 1);
5124 clear_extent_bit(io_tree, start, end,
5125 EXTENT_LOCKED | EXTENT_DIRTY |
5126 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5127 EXTENT_DEFRAG, 1, 1,
5128 &cached_state, GFP_NOFS);
5131 spin_lock(&io_tree->lock);
5133 spin_unlock(&io_tree->lock);
5136 void btrfs_evict_inode(struct inode *inode)
5138 struct btrfs_trans_handle *trans;
5139 struct btrfs_root *root = BTRFS_I(inode)->root;
5140 struct btrfs_block_rsv *rsv, *global_rsv;
5141 int steal_from_global = 0;
5142 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5145 trace_btrfs_inode_evict(inode);
5147 evict_inode_truncate_pages(inode);
5149 if (inode->i_nlink &&
5150 ((btrfs_root_refs(&root->root_item) != 0 &&
5151 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5152 btrfs_is_free_space_inode(inode)))
5155 if (is_bad_inode(inode)) {
5156 btrfs_orphan_del(NULL, inode);
5159 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5160 if (!special_file(inode->i_mode))
5161 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5163 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5165 if (root->fs_info->log_root_recovering) {
5166 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5167 &BTRFS_I(inode)->runtime_flags));
5171 if (inode->i_nlink > 0) {
5172 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5173 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5177 ret = btrfs_commit_inode_delayed_inode(inode);
5179 btrfs_orphan_del(NULL, inode);
5183 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5185 btrfs_orphan_del(NULL, inode);
5188 rsv->size = min_size;
5190 global_rsv = &root->fs_info->global_block_rsv;
5192 btrfs_i_size_write(inode, 0);
5195 * This is a bit simpler than btrfs_truncate since we've already
5196 * reserved our space for our orphan item in the unlink, so we just
5197 * need to reserve some slack space in case we add bytes and update
5198 * inode item when doing the truncate.
5201 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5202 BTRFS_RESERVE_FLUSH_LIMIT);
5205 * Try and steal from the global reserve since we will
5206 * likely not use this space anyway, we want to try as
5207 * hard as possible to get this to work.
5210 steal_from_global++;
5212 steal_from_global = 0;
5216 * steal_from_global == 0: we reserved stuff, hooray!
5217 * steal_from_global == 1: we didn't reserve stuff, boo!
5218 * steal_from_global == 2: we've committed, still not a lot of
5219 * room but maybe we'll have room in the global reserve this
5221 * steal_from_global == 3: abandon all hope!
5223 if (steal_from_global > 2) {
5224 btrfs_warn(root->fs_info,
5225 "Could not get space for a delete, will truncate on mount %d",
5227 btrfs_orphan_del(NULL, inode);
5228 btrfs_free_block_rsv(root, rsv);
5232 trans = btrfs_join_transaction(root);
5233 if (IS_ERR(trans)) {
5234 btrfs_orphan_del(NULL, inode);
5235 btrfs_free_block_rsv(root, rsv);
5240 * We can't just steal from the global reserve, we need tomake
5241 * sure there is room to do it, if not we need to commit and try
5244 if (steal_from_global) {
5245 if (!btrfs_check_space_for_delayed_refs(trans, root))
5246 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5253 * Couldn't steal from the global reserve, we have too much
5254 * pending stuff built up, commit the transaction and try it
5258 ret = btrfs_commit_transaction(trans, root);
5260 btrfs_orphan_del(NULL, inode);
5261 btrfs_free_block_rsv(root, rsv);
5266 steal_from_global = 0;
5269 trans->block_rsv = rsv;
5271 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5272 if (ret != -ENOSPC && ret != -EAGAIN)
5275 trans->block_rsv = &root->fs_info->trans_block_rsv;
5276 btrfs_end_transaction(trans, root);
5278 btrfs_btree_balance_dirty(root);
5281 btrfs_free_block_rsv(root, rsv);
5284 * Errors here aren't a big deal, it just means we leave orphan items
5285 * in the tree. They will be cleaned up on the next mount.
5288 trans->block_rsv = root->orphan_block_rsv;
5289 btrfs_orphan_del(trans, inode);
5291 btrfs_orphan_del(NULL, inode);
5294 trans->block_rsv = &root->fs_info->trans_block_rsv;
5295 if (!(root == root->fs_info->tree_root ||
5296 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5297 btrfs_return_ino(root, btrfs_ino(inode));
5299 btrfs_end_transaction(trans, root);
5300 btrfs_btree_balance_dirty(root);
5302 btrfs_remove_delayed_node(inode);
5307 * this returns the key found in the dir entry in the location pointer.
5308 * If no dir entries were found, location->objectid is 0.
5310 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5311 struct btrfs_key *location)
5313 const char *name = dentry->d_name.name;
5314 int namelen = dentry->d_name.len;
5315 struct btrfs_dir_item *di;
5316 struct btrfs_path *path;
5317 struct btrfs_root *root = BTRFS_I(dir)->root;
5320 path = btrfs_alloc_path();
5324 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5329 if (IS_ERR_OR_NULL(di))
5332 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5334 btrfs_free_path(path);
5337 location->objectid = 0;
5342 * when we hit a tree root in a directory, the btrfs part of the inode
5343 * needs to be changed to reflect the root directory of the tree root. This
5344 * is kind of like crossing a mount point.
5346 static int fixup_tree_root_location(struct btrfs_root *root,
5348 struct dentry *dentry,
5349 struct btrfs_key *location,
5350 struct btrfs_root **sub_root)
5352 struct btrfs_path *path;
5353 struct btrfs_root *new_root;
5354 struct btrfs_root_ref *ref;
5355 struct extent_buffer *leaf;
5356 struct btrfs_key key;
5360 path = btrfs_alloc_path();
5367 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5368 key.type = BTRFS_ROOT_REF_KEY;
5369 key.offset = location->objectid;
5371 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5379 leaf = path->nodes[0];
5380 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5381 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5382 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5385 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5386 (unsigned long)(ref + 1),
5387 dentry->d_name.len);
5391 btrfs_release_path(path);
5393 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5394 if (IS_ERR(new_root)) {
5395 err = PTR_ERR(new_root);
5399 *sub_root = new_root;
5400 location->objectid = btrfs_root_dirid(&new_root->root_item);
5401 location->type = BTRFS_INODE_ITEM_KEY;
5402 location->offset = 0;
5405 btrfs_free_path(path);
5409 static void inode_tree_add(struct inode *inode)
5411 struct btrfs_root *root = BTRFS_I(inode)->root;
5412 struct btrfs_inode *entry;
5414 struct rb_node *parent;
5415 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5416 u64 ino = btrfs_ino(inode);
5418 if (inode_unhashed(inode))
5421 spin_lock(&root->inode_lock);
5422 p = &root->inode_tree.rb_node;
5425 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5427 if (ino < btrfs_ino(&entry->vfs_inode))
5428 p = &parent->rb_left;
5429 else if (ino > btrfs_ino(&entry->vfs_inode))
5430 p = &parent->rb_right;
5432 WARN_ON(!(entry->vfs_inode.i_state &
5433 (I_WILL_FREE | I_FREEING)));
5434 rb_replace_node(parent, new, &root->inode_tree);
5435 RB_CLEAR_NODE(parent);
5436 spin_unlock(&root->inode_lock);
5440 rb_link_node(new, parent, p);
5441 rb_insert_color(new, &root->inode_tree);
5442 spin_unlock(&root->inode_lock);
5445 static void inode_tree_del(struct inode *inode)
5447 struct btrfs_root *root = BTRFS_I(inode)->root;
5450 spin_lock(&root->inode_lock);
5451 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5452 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5453 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5454 empty = RB_EMPTY_ROOT(&root->inode_tree);
5456 spin_unlock(&root->inode_lock);
5458 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5459 synchronize_srcu(&root->fs_info->subvol_srcu);
5460 spin_lock(&root->inode_lock);
5461 empty = RB_EMPTY_ROOT(&root->inode_tree);
5462 spin_unlock(&root->inode_lock);
5464 btrfs_add_dead_root(root);
5468 void btrfs_invalidate_inodes(struct btrfs_root *root)
5470 struct rb_node *node;
5471 struct rb_node *prev;
5472 struct btrfs_inode *entry;
5473 struct inode *inode;
5476 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5477 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5479 spin_lock(&root->inode_lock);
5481 node = root->inode_tree.rb_node;
5485 entry = rb_entry(node, struct btrfs_inode, rb_node);
5487 if (objectid < btrfs_ino(&entry->vfs_inode))
5488 node = node->rb_left;
5489 else if (objectid > btrfs_ino(&entry->vfs_inode))
5490 node = node->rb_right;
5496 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5497 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5501 prev = rb_next(prev);
5505 entry = rb_entry(node, struct btrfs_inode, rb_node);
5506 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5507 inode = igrab(&entry->vfs_inode);
5509 spin_unlock(&root->inode_lock);
5510 if (atomic_read(&inode->i_count) > 1)
5511 d_prune_aliases(inode);
5513 * btrfs_drop_inode will have it removed from
5514 * the inode cache when its usage count
5519 spin_lock(&root->inode_lock);
5523 if (cond_resched_lock(&root->inode_lock))
5526 node = rb_next(node);
5528 spin_unlock(&root->inode_lock);
5531 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5533 struct btrfs_iget_args *args = p;
5534 inode->i_ino = args->location->objectid;
5535 memcpy(&BTRFS_I(inode)->location, args->location,
5536 sizeof(*args->location));
5537 BTRFS_I(inode)->root = args->root;
5541 static int btrfs_find_actor(struct inode *inode, void *opaque)
5543 struct btrfs_iget_args *args = opaque;
5544 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5545 args->root == BTRFS_I(inode)->root;
5548 static struct inode *btrfs_iget_locked(struct super_block *s,
5549 struct btrfs_key *location,
5550 struct btrfs_root *root)
5552 struct inode *inode;
5553 struct btrfs_iget_args args;
5554 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5556 args.location = location;
5559 inode = iget5_locked(s, hashval, btrfs_find_actor,
5560 btrfs_init_locked_inode,
5565 /* Get an inode object given its location and corresponding root.
5566 * Returns in *is_new if the inode was read from disk
5568 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5569 struct btrfs_root *root, int *new)
5571 struct inode *inode;
5573 inode = btrfs_iget_locked(s, location, root);
5575 return ERR_PTR(-ENOMEM);
5577 if (inode->i_state & I_NEW) {
5578 btrfs_read_locked_inode(inode);
5579 if (!is_bad_inode(inode)) {
5580 inode_tree_add(inode);
5581 unlock_new_inode(inode);
5585 unlock_new_inode(inode);
5587 inode = ERR_PTR(-ESTALE);
5594 static struct inode *new_simple_dir(struct super_block *s,
5595 struct btrfs_key *key,
5596 struct btrfs_root *root)
5598 struct inode *inode = new_inode(s);
5601 return ERR_PTR(-ENOMEM);
5603 BTRFS_I(inode)->root = root;
5604 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5605 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5607 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5608 inode->i_op = &btrfs_dir_ro_inode_operations;
5609 inode->i_fop = &simple_dir_operations;
5610 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5611 inode->i_mtime = current_fs_time(inode->i_sb);
5612 inode->i_atime = inode->i_mtime;
5613 inode->i_ctime = inode->i_mtime;
5614 BTRFS_I(inode)->i_otime = inode->i_mtime;
5619 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5621 struct inode *inode;
5622 struct btrfs_root *root = BTRFS_I(dir)->root;
5623 struct btrfs_root *sub_root = root;
5624 struct btrfs_key location;
5628 if (dentry->d_name.len > BTRFS_NAME_LEN)
5629 return ERR_PTR(-ENAMETOOLONG);
5631 ret = btrfs_inode_by_name(dir, dentry, &location);
5633 return ERR_PTR(ret);
5635 if (location.objectid == 0)
5636 return ERR_PTR(-ENOENT);
5638 if (location.type == BTRFS_INODE_ITEM_KEY) {
5639 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5643 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5645 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5646 ret = fixup_tree_root_location(root, dir, dentry,
5647 &location, &sub_root);
5650 inode = ERR_PTR(ret);
5652 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5654 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5656 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5658 if (!IS_ERR(inode) && root != sub_root) {
5659 down_read(&root->fs_info->cleanup_work_sem);
5660 if (!(inode->i_sb->s_flags & MS_RDONLY))
5661 ret = btrfs_orphan_cleanup(sub_root);
5662 up_read(&root->fs_info->cleanup_work_sem);
5665 inode = ERR_PTR(ret);
5672 static int btrfs_dentry_delete(const struct dentry *dentry)
5674 struct btrfs_root *root;
5675 struct inode *inode = d_inode(dentry);
5677 if (!inode && !IS_ROOT(dentry))
5678 inode = d_inode(dentry->d_parent);
5681 root = BTRFS_I(inode)->root;
5682 if (btrfs_root_refs(&root->root_item) == 0)
5685 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5691 static void btrfs_dentry_release(struct dentry *dentry)
5693 kfree(dentry->d_fsdata);
5696 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5699 struct inode *inode;
5701 inode = btrfs_lookup_dentry(dir, dentry);
5702 if (IS_ERR(inode)) {
5703 if (PTR_ERR(inode) == -ENOENT)
5706 return ERR_CAST(inode);
5709 return d_splice_alias(inode, dentry);
5712 unsigned char btrfs_filetype_table[] = {
5713 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5716 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5718 struct inode *inode = file_inode(file);
5719 struct btrfs_root *root = BTRFS_I(inode)->root;
5720 struct btrfs_item *item;
5721 struct btrfs_dir_item *di;
5722 struct btrfs_key key;
5723 struct btrfs_key found_key;
5724 struct btrfs_path *path;
5725 struct list_head ins_list;
5726 struct list_head del_list;
5728 struct extent_buffer *leaf;
5730 unsigned char d_type;
5735 int key_type = BTRFS_DIR_INDEX_KEY;
5739 int is_curr = 0; /* ctx->pos points to the current index? */
5742 /* FIXME, use a real flag for deciding about the key type */
5743 if (root->fs_info->tree_root == root)
5744 key_type = BTRFS_DIR_ITEM_KEY;
5746 if (!dir_emit_dots(file, ctx))
5749 path = btrfs_alloc_path();
5753 path->reada = READA_FORWARD;
5755 if (key_type == BTRFS_DIR_INDEX_KEY) {
5756 INIT_LIST_HEAD(&ins_list);
5757 INIT_LIST_HEAD(&del_list);
5758 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5761 key.type = key_type;
5762 key.offset = ctx->pos;
5763 key.objectid = btrfs_ino(inode);
5765 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5771 leaf = path->nodes[0];
5772 slot = path->slots[0];
5773 if (slot >= btrfs_header_nritems(leaf)) {
5774 ret = btrfs_next_leaf(root, path);
5782 item = btrfs_item_nr(slot);
5783 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5785 if (found_key.objectid != key.objectid)
5787 if (found_key.type != key_type)
5789 if (found_key.offset < ctx->pos)
5791 if (key_type == BTRFS_DIR_INDEX_KEY &&
5792 btrfs_should_delete_dir_index(&del_list,
5796 ctx->pos = found_key.offset;
5799 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5801 di_total = btrfs_item_size(leaf, item);
5803 while (di_cur < di_total) {
5804 struct btrfs_key location;
5806 if (verify_dir_item(root, leaf, di))
5809 name_len = btrfs_dir_name_len(leaf, di);
5810 if (name_len <= sizeof(tmp_name)) {
5811 name_ptr = tmp_name;
5813 name_ptr = kmalloc(name_len, GFP_KERNEL);
5819 read_extent_buffer(leaf, name_ptr,
5820 (unsigned long)(di + 1), name_len);
5822 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5823 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5826 /* is this a reference to our own snapshot? If so
5829 * In contrast to old kernels, we insert the snapshot's
5830 * dir item and dir index after it has been created, so
5831 * we won't find a reference to our own snapshot. We
5832 * still keep the following code for backward
5835 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5836 location.objectid == root->root_key.objectid) {
5840 over = !dir_emit(ctx, name_ptr, name_len,
5841 location.objectid, d_type);
5844 if (name_ptr != tmp_name)
5850 di_len = btrfs_dir_name_len(leaf, di) +
5851 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5853 di = (struct btrfs_dir_item *)((char *)di + di_len);
5859 if (key_type == BTRFS_DIR_INDEX_KEY) {
5862 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5868 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5869 * it was was set to the termination value in previous call. We assume
5870 * that "." and ".." were emitted if we reach this point and set the
5871 * termination value as well for an empty directory.
5873 if (ctx->pos > 2 && !emitted)
5876 /* Reached end of directory/root. Bump pos past the last item. */
5880 * Stop new entries from being returned after we return the last
5883 * New directory entries are assigned a strictly increasing
5884 * offset. This means that new entries created during readdir
5885 * are *guaranteed* to be seen in the future by that readdir.
5886 * This has broken buggy programs which operate on names as
5887 * they're returned by readdir. Until we re-use freed offsets
5888 * we have this hack to stop new entries from being returned
5889 * under the assumption that they'll never reach this huge
5892 * This is being careful not to overflow 32bit loff_t unless the
5893 * last entry requires it because doing so has broken 32bit apps
5896 if (key_type == BTRFS_DIR_INDEX_KEY) {
5897 if (ctx->pos >= INT_MAX)
5898 ctx->pos = LLONG_MAX;
5905 if (key_type == BTRFS_DIR_INDEX_KEY)
5906 btrfs_put_delayed_items(&ins_list, &del_list);
5907 btrfs_free_path(path);
5911 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5913 struct btrfs_root *root = BTRFS_I(inode)->root;
5914 struct btrfs_trans_handle *trans;
5916 bool nolock = false;
5918 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5921 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5924 if (wbc->sync_mode == WB_SYNC_ALL) {
5926 trans = btrfs_join_transaction_nolock(root);
5928 trans = btrfs_join_transaction(root);
5930 return PTR_ERR(trans);
5931 ret = btrfs_commit_transaction(trans, root);
5937 * This is somewhat expensive, updating the tree every time the
5938 * inode changes. But, it is most likely to find the inode in cache.
5939 * FIXME, needs more benchmarking...there are no reasons other than performance
5940 * to keep or drop this code.
5942 static int btrfs_dirty_inode(struct inode *inode)
5944 struct btrfs_root *root = BTRFS_I(inode)->root;
5945 struct btrfs_trans_handle *trans;
5948 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5951 trans = btrfs_join_transaction(root);
5953 return PTR_ERR(trans);
5955 ret = btrfs_update_inode(trans, root, inode);
5956 if (ret && ret == -ENOSPC) {
5957 /* whoops, lets try again with the full transaction */
5958 btrfs_end_transaction(trans, root);
5959 trans = btrfs_start_transaction(root, 1);
5961 return PTR_ERR(trans);
5963 ret = btrfs_update_inode(trans, root, inode);
5965 btrfs_end_transaction(trans, root);
5966 if (BTRFS_I(inode)->delayed_node)
5967 btrfs_balance_delayed_items(root);
5973 * This is a copy of file_update_time. We need this so we can return error on
5974 * ENOSPC for updating the inode in the case of file write and mmap writes.
5976 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5979 struct btrfs_root *root = BTRFS_I(inode)->root;
5981 if (btrfs_root_readonly(root))
5984 if (flags & S_VERSION)
5985 inode_inc_iversion(inode);
5986 if (flags & S_CTIME)
5987 inode->i_ctime = *now;
5988 if (flags & S_MTIME)
5989 inode->i_mtime = *now;
5990 if (flags & S_ATIME)
5991 inode->i_atime = *now;
5992 return btrfs_dirty_inode(inode);
5996 * find the highest existing sequence number in a directory
5997 * and then set the in-memory index_cnt variable to reflect
5998 * free sequence numbers
6000 static int btrfs_set_inode_index_count(struct inode *inode)
6002 struct btrfs_root *root = BTRFS_I(inode)->root;
6003 struct btrfs_key key, found_key;
6004 struct btrfs_path *path;
6005 struct extent_buffer *leaf;
6008 key.objectid = btrfs_ino(inode);
6009 key.type = BTRFS_DIR_INDEX_KEY;
6010 key.offset = (u64)-1;
6012 path = btrfs_alloc_path();
6016 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6019 /* FIXME: we should be able to handle this */
6025 * MAGIC NUMBER EXPLANATION:
6026 * since we search a directory based on f_pos we have to start at 2
6027 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6028 * else has to start at 2
6030 if (path->slots[0] == 0) {
6031 BTRFS_I(inode)->index_cnt = 2;
6037 leaf = path->nodes[0];
6038 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6040 if (found_key.objectid != btrfs_ino(inode) ||
6041 found_key.type != BTRFS_DIR_INDEX_KEY) {
6042 BTRFS_I(inode)->index_cnt = 2;
6046 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6048 btrfs_free_path(path);
6053 * helper to find a free sequence number in a given directory. This current
6054 * code is very simple, later versions will do smarter things in the btree
6056 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6060 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6061 ret = btrfs_inode_delayed_dir_index_count(dir);
6063 ret = btrfs_set_inode_index_count(dir);
6069 *index = BTRFS_I(dir)->index_cnt;
6070 BTRFS_I(dir)->index_cnt++;
6075 static int btrfs_insert_inode_locked(struct inode *inode)
6077 struct btrfs_iget_args args;
6078 args.location = &BTRFS_I(inode)->location;
6079 args.root = BTRFS_I(inode)->root;
6081 return insert_inode_locked4(inode,
6082 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6083 btrfs_find_actor, &args);
6086 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6087 struct btrfs_root *root,
6089 const char *name, int name_len,
6090 u64 ref_objectid, u64 objectid,
6091 umode_t mode, u64 *index)
6093 struct inode *inode;
6094 struct btrfs_inode_item *inode_item;
6095 struct btrfs_key *location;
6096 struct btrfs_path *path;
6097 struct btrfs_inode_ref *ref;
6098 struct btrfs_key key[2];
6100 int nitems = name ? 2 : 1;
6104 path = btrfs_alloc_path();
6106 return ERR_PTR(-ENOMEM);
6108 inode = new_inode(root->fs_info->sb);
6110 btrfs_free_path(path);
6111 return ERR_PTR(-ENOMEM);
6115 * O_TMPFILE, set link count to 0, so that after this point,
6116 * we fill in an inode item with the correct link count.
6119 set_nlink(inode, 0);
6122 * we have to initialize this early, so we can reclaim the inode
6123 * number if we fail afterwards in this function.
6125 inode->i_ino = objectid;
6128 trace_btrfs_inode_request(dir);
6130 ret = btrfs_set_inode_index(dir, index);
6132 btrfs_free_path(path);
6134 return ERR_PTR(ret);
6140 * index_cnt is ignored for everything but a dir,
6141 * btrfs_get_inode_index_count has an explanation for the magic
6144 BTRFS_I(inode)->index_cnt = 2;
6145 BTRFS_I(inode)->dir_index = *index;
6146 BTRFS_I(inode)->root = root;
6147 BTRFS_I(inode)->generation = trans->transid;
6148 inode->i_generation = BTRFS_I(inode)->generation;
6151 * We could have gotten an inode number from somebody who was fsynced
6152 * and then removed in this same transaction, so let's just set full
6153 * sync since it will be a full sync anyway and this will blow away the
6154 * old info in the log.
6156 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6158 key[0].objectid = objectid;
6159 key[0].type = BTRFS_INODE_ITEM_KEY;
6162 sizes[0] = sizeof(struct btrfs_inode_item);
6166 * Start new inodes with an inode_ref. This is slightly more
6167 * efficient for small numbers of hard links since they will
6168 * be packed into one item. Extended refs will kick in if we
6169 * add more hard links than can fit in the ref item.
6171 key[1].objectid = objectid;
6172 key[1].type = BTRFS_INODE_REF_KEY;
6173 key[1].offset = ref_objectid;
6175 sizes[1] = name_len + sizeof(*ref);
6178 location = &BTRFS_I(inode)->location;
6179 location->objectid = objectid;
6180 location->offset = 0;
6181 location->type = BTRFS_INODE_ITEM_KEY;
6183 ret = btrfs_insert_inode_locked(inode);
6187 path->leave_spinning = 1;
6188 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6192 inode_init_owner(inode, dir, mode);
6193 inode_set_bytes(inode, 0);
6195 inode->i_mtime = current_fs_time(inode->i_sb);
6196 inode->i_atime = inode->i_mtime;
6197 inode->i_ctime = inode->i_mtime;
6198 BTRFS_I(inode)->i_otime = inode->i_mtime;
6200 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6201 struct btrfs_inode_item);
6202 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6203 sizeof(*inode_item));
6204 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6207 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6208 struct btrfs_inode_ref);
6209 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6210 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6211 ptr = (unsigned long)(ref + 1);
6212 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6215 btrfs_mark_buffer_dirty(path->nodes[0]);
6216 btrfs_free_path(path);
6218 btrfs_inherit_iflags(inode, dir);
6220 if (S_ISREG(mode)) {
6221 if (btrfs_test_opt(root, NODATASUM))
6222 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6223 if (btrfs_test_opt(root, NODATACOW))
6224 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6225 BTRFS_INODE_NODATASUM;
6228 inode_tree_add(inode);
6230 trace_btrfs_inode_new(inode);
6231 btrfs_set_inode_last_trans(trans, inode);
6233 btrfs_update_root_times(trans, root);
6235 ret = btrfs_inode_inherit_props(trans, inode, dir);
6237 btrfs_err(root->fs_info,
6238 "error inheriting props for ino %llu (root %llu): %d",
6239 btrfs_ino(inode), root->root_key.objectid, ret);
6244 unlock_new_inode(inode);
6247 BTRFS_I(dir)->index_cnt--;
6248 btrfs_free_path(path);
6250 return ERR_PTR(ret);
6253 static inline u8 btrfs_inode_type(struct inode *inode)
6255 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6259 * utility function to add 'inode' into 'parent_inode' with
6260 * a give name and a given sequence number.
6261 * if 'add_backref' is true, also insert a backref from the
6262 * inode to the parent directory.
6264 int btrfs_add_link(struct btrfs_trans_handle *trans,
6265 struct inode *parent_inode, struct inode *inode,
6266 const char *name, int name_len, int add_backref, u64 index)
6269 struct btrfs_key key;
6270 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6271 u64 ino = btrfs_ino(inode);
6272 u64 parent_ino = btrfs_ino(parent_inode);
6274 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6275 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6278 key.type = BTRFS_INODE_ITEM_KEY;
6282 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6283 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6284 key.objectid, root->root_key.objectid,
6285 parent_ino, index, name, name_len);
6286 } else if (add_backref) {
6287 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6291 /* Nothing to clean up yet */
6295 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6297 btrfs_inode_type(inode), index);
6298 if (ret == -EEXIST || ret == -EOVERFLOW)
6301 btrfs_abort_transaction(trans, root, ret);
6305 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6307 inode_inc_iversion(parent_inode);
6308 parent_inode->i_mtime = parent_inode->i_ctime =
6309 current_fs_time(parent_inode->i_sb);
6310 ret = btrfs_update_inode(trans, root, parent_inode);
6312 btrfs_abort_transaction(trans, root, ret);
6316 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6319 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6320 key.objectid, root->root_key.objectid,
6321 parent_ino, &local_index, name, name_len);
6323 } else if (add_backref) {
6327 err = btrfs_del_inode_ref(trans, root, name, name_len,
6328 ino, parent_ino, &local_index);
6333 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6334 struct inode *dir, struct dentry *dentry,
6335 struct inode *inode, int backref, u64 index)
6337 int err = btrfs_add_link(trans, dir, inode,
6338 dentry->d_name.name, dentry->d_name.len,
6345 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6346 umode_t mode, dev_t rdev)
6348 struct btrfs_trans_handle *trans;
6349 struct btrfs_root *root = BTRFS_I(dir)->root;
6350 struct inode *inode = NULL;
6357 * 2 for inode item and ref
6359 * 1 for xattr if selinux is on
6361 trans = btrfs_start_transaction(root, 5);
6363 return PTR_ERR(trans);
6365 err = btrfs_find_free_ino(root, &objectid);
6369 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6370 dentry->d_name.len, btrfs_ino(dir), objectid,
6372 if (IS_ERR(inode)) {
6373 err = PTR_ERR(inode);
6378 * If the active LSM wants to access the inode during
6379 * d_instantiate it needs these. Smack checks to see
6380 * if the filesystem supports xattrs by looking at the
6383 inode->i_op = &btrfs_special_inode_operations;
6384 init_special_inode(inode, inode->i_mode, rdev);
6386 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6388 goto out_unlock_inode;
6390 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6392 goto out_unlock_inode;
6394 btrfs_update_inode(trans, root, inode);
6395 unlock_new_inode(inode);
6396 d_instantiate(dentry, inode);
6400 btrfs_end_transaction(trans, root);
6401 btrfs_balance_delayed_items(root);
6402 btrfs_btree_balance_dirty(root);
6404 inode_dec_link_count(inode);
6411 unlock_new_inode(inode);
6416 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6417 umode_t mode, bool excl)
6419 struct btrfs_trans_handle *trans;
6420 struct btrfs_root *root = BTRFS_I(dir)->root;
6421 struct inode *inode = NULL;
6422 int drop_inode_on_err = 0;
6428 * 2 for inode item and ref
6430 * 1 for xattr if selinux is on
6432 trans = btrfs_start_transaction(root, 5);
6434 return PTR_ERR(trans);
6436 err = btrfs_find_free_ino(root, &objectid);
6440 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6441 dentry->d_name.len, btrfs_ino(dir), objectid,
6443 if (IS_ERR(inode)) {
6444 err = PTR_ERR(inode);
6447 drop_inode_on_err = 1;
6449 * If the active LSM wants to access the inode during
6450 * d_instantiate it needs these. Smack checks to see
6451 * if the filesystem supports xattrs by looking at the
6454 inode->i_fop = &btrfs_file_operations;
6455 inode->i_op = &btrfs_file_inode_operations;
6456 inode->i_mapping->a_ops = &btrfs_aops;
6458 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6460 goto out_unlock_inode;
6462 err = btrfs_update_inode(trans, root, inode);
6464 goto out_unlock_inode;
6466 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6468 goto out_unlock_inode;
6470 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6471 unlock_new_inode(inode);
6472 d_instantiate(dentry, inode);
6475 btrfs_end_transaction(trans, root);
6476 if (err && drop_inode_on_err) {
6477 inode_dec_link_count(inode);
6480 btrfs_balance_delayed_items(root);
6481 btrfs_btree_balance_dirty(root);
6485 unlock_new_inode(inode);
6490 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6491 struct dentry *dentry)
6493 struct btrfs_trans_handle *trans = NULL;
6494 struct btrfs_root *root = BTRFS_I(dir)->root;
6495 struct inode *inode = d_inode(old_dentry);
6500 /* do not allow sys_link's with other subvols of the same device */
6501 if (root->objectid != BTRFS_I(inode)->root->objectid)
6504 if (inode->i_nlink >= BTRFS_LINK_MAX)
6507 err = btrfs_set_inode_index(dir, &index);
6512 * 2 items for inode and inode ref
6513 * 2 items for dir items
6514 * 1 item for parent inode
6516 trans = btrfs_start_transaction(root, 5);
6517 if (IS_ERR(trans)) {
6518 err = PTR_ERR(trans);
6523 /* There are several dir indexes for this inode, clear the cache. */
6524 BTRFS_I(inode)->dir_index = 0ULL;
6526 inode_inc_iversion(inode);
6527 inode->i_ctime = current_fs_time(inode->i_sb);
6529 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6531 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6536 struct dentry *parent = dentry->d_parent;
6537 err = btrfs_update_inode(trans, root, inode);
6540 if (inode->i_nlink == 1) {
6542 * If new hard link count is 1, it's a file created
6543 * with open(2) O_TMPFILE flag.
6545 err = btrfs_orphan_del(trans, inode);
6549 d_instantiate(dentry, inode);
6550 btrfs_log_new_name(trans, inode, NULL, parent);
6553 btrfs_balance_delayed_items(root);
6556 btrfs_end_transaction(trans, root);
6558 inode_dec_link_count(inode);
6561 btrfs_btree_balance_dirty(root);
6565 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6567 struct inode *inode = NULL;
6568 struct btrfs_trans_handle *trans;
6569 struct btrfs_root *root = BTRFS_I(dir)->root;
6571 int drop_on_err = 0;
6576 * 2 items for inode and ref
6577 * 2 items for dir items
6578 * 1 for xattr if selinux is on
6580 trans = btrfs_start_transaction(root, 5);
6582 return PTR_ERR(trans);
6584 err = btrfs_find_free_ino(root, &objectid);
6588 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6589 dentry->d_name.len, btrfs_ino(dir), objectid,
6590 S_IFDIR | mode, &index);
6591 if (IS_ERR(inode)) {
6592 err = PTR_ERR(inode);
6597 /* these must be set before we unlock the inode */
6598 inode->i_op = &btrfs_dir_inode_operations;
6599 inode->i_fop = &btrfs_dir_file_operations;
6601 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6603 goto out_fail_inode;
6605 btrfs_i_size_write(inode, 0);
6606 err = btrfs_update_inode(trans, root, inode);
6608 goto out_fail_inode;
6610 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6611 dentry->d_name.len, 0, index);
6613 goto out_fail_inode;
6615 d_instantiate(dentry, inode);
6617 * mkdir is special. We're unlocking after we call d_instantiate
6618 * to avoid a race with nfsd calling d_instantiate.
6620 unlock_new_inode(inode);
6624 btrfs_end_transaction(trans, root);
6626 inode_dec_link_count(inode);
6629 btrfs_balance_delayed_items(root);
6630 btrfs_btree_balance_dirty(root);
6634 unlock_new_inode(inode);
6638 /* Find next extent map of a given extent map, caller needs to ensure locks */
6639 static struct extent_map *next_extent_map(struct extent_map *em)
6641 struct rb_node *next;
6643 next = rb_next(&em->rb_node);
6646 return container_of(next, struct extent_map, rb_node);
6649 static struct extent_map *prev_extent_map(struct extent_map *em)
6651 struct rb_node *prev;
6653 prev = rb_prev(&em->rb_node);
6656 return container_of(prev, struct extent_map, rb_node);
6659 /* helper for btfs_get_extent. Given an existing extent in the tree,
6660 * the existing extent is the nearest extent to map_start,
6661 * and an extent that you want to insert, deal with overlap and insert
6662 * the best fitted new extent into the tree.
6664 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6665 struct extent_map *existing,
6666 struct extent_map *em,
6669 struct extent_map *prev;
6670 struct extent_map *next;
6675 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6677 if (existing->start > map_start) {
6679 prev = prev_extent_map(next);
6682 next = next_extent_map(prev);
6685 start = prev ? extent_map_end(prev) : em->start;
6686 start = max_t(u64, start, em->start);
6687 end = next ? next->start : extent_map_end(em);
6688 end = min_t(u64, end, extent_map_end(em));
6689 start_diff = start - em->start;
6691 em->len = end - start;
6692 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6693 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6694 em->block_start += start_diff;
6695 em->block_len -= start_diff;
6697 return add_extent_mapping(em_tree, em, 0);
6700 static noinline int uncompress_inline(struct btrfs_path *path,
6702 size_t pg_offset, u64 extent_offset,
6703 struct btrfs_file_extent_item *item)
6706 struct extent_buffer *leaf = path->nodes[0];
6709 unsigned long inline_size;
6713 WARN_ON(pg_offset != 0);
6714 compress_type = btrfs_file_extent_compression(leaf, item);
6715 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6716 inline_size = btrfs_file_extent_inline_item_len(leaf,
6717 btrfs_item_nr(path->slots[0]));
6718 tmp = kmalloc(inline_size, GFP_NOFS);
6721 ptr = btrfs_file_extent_inline_start(item);
6723 read_extent_buffer(leaf, tmp, ptr, inline_size);
6725 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6726 ret = btrfs_decompress(compress_type, tmp, page,
6727 extent_offset, inline_size, max_size);
6733 * a bit scary, this does extent mapping from logical file offset to the disk.
6734 * the ugly parts come from merging extents from the disk with the in-ram
6735 * representation. This gets more complex because of the data=ordered code,
6736 * where the in-ram extents might be locked pending data=ordered completion.
6738 * This also copies inline extents directly into the page.
6741 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6742 size_t pg_offset, u64 start, u64 len,
6747 u64 extent_start = 0;
6749 u64 objectid = btrfs_ino(inode);
6751 struct btrfs_path *path = NULL;
6752 struct btrfs_root *root = BTRFS_I(inode)->root;
6753 struct btrfs_file_extent_item *item;
6754 struct extent_buffer *leaf;
6755 struct btrfs_key found_key;
6756 struct extent_map *em = NULL;
6757 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6758 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6759 struct btrfs_trans_handle *trans = NULL;
6760 const bool new_inline = !page || create;
6763 read_lock(&em_tree->lock);
6764 em = lookup_extent_mapping(em_tree, start, len);
6766 em->bdev = root->fs_info->fs_devices->latest_bdev;
6767 read_unlock(&em_tree->lock);
6770 if (em->start > start || em->start + em->len <= start)
6771 free_extent_map(em);
6772 else if (em->block_start == EXTENT_MAP_INLINE && page)
6773 free_extent_map(em);
6777 em = alloc_extent_map();
6782 em->bdev = root->fs_info->fs_devices->latest_bdev;
6783 em->start = EXTENT_MAP_HOLE;
6784 em->orig_start = EXTENT_MAP_HOLE;
6786 em->block_len = (u64)-1;
6789 path = btrfs_alloc_path();
6795 * Chances are we'll be called again, so go ahead and do
6798 path->reada = READA_FORWARD;
6801 ret = btrfs_lookup_file_extent(trans, root, path,
6802 objectid, start, trans != NULL);
6809 if (path->slots[0] == 0)
6814 leaf = path->nodes[0];
6815 item = btrfs_item_ptr(leaf, path->slots[0],
6816 struct btrfs_file_extent_item);
6817 /* are we inside the extent that was found? */
6818 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6819 found_type = found_key.type;
6820 if (found_key.objectid != objectid ||
6821 found_type != BTRFS_EXTENT_DATA_KEY) {
6823 * If we backup past the first extent we want to move forward
6824 * and see if there is an extent in front of us, otherwise we'll
6825 * say there is a hole for our whole search range which can
6832 found_type = btrfs_file_extent_type(leaf, item);
6833 extent_start = found_key.offset;
6834 if (found_type == BTRFS_FILE_EXTENT_REG ||
6835 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6836 extent_end = extent_start +
6837 btrfs_file_extent_num_bytes(leaf, item);
6838 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6840 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6841 extent_end = ALIGN(extent_start + size, root->sectorsize);
6844 if (start >= extent_end) {
6846 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6847 ret = btrfs_next_leaf(root, path);
6854 leaf = path->nodes[0];
6856 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6857 if (found_key.objectid != objectid ||
6858 found_key.type != BTRFS_EXTENT_DATA_KEY)
6860 if (start + len <= found_key.offset)
6862 if (start > found_key.offset)
6865 em->orig_start = start;
6866 em->len = found_key.offset - start;
6870 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6872 if (found_type == BTRFS_FILE_EXTENT_REG ||
6873 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6875 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6879 size_t extent_offset;
6885 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6886 extent_offset = page_offset(page) + pg_offset - extent_start;
6887 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6888 size - extent_offset);
6889 em->start = extent_start + extent_offset;
6890 em->len = ALIGN(copy_size, root->sectorsize);
6891 em->orig_block_len = em->len;
6892 em->orig_start = em->start;
6893 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6894 if (create == 0 && !PageUptodate(page)) {
6895 if (btrfs_file_extent_compression(leaf, item) !=
6896 BTRFS_COMPRESS_NONE) {
6897 ret = uncompress_inline(path, page, pg_offset,
6898 extent_offset, item);
6905 read_extent_buffer(leaf, map + pg_offset, ptr,
6907 if (pg_offset + copy_size < PAGE_SIZE) {
6908 memset(map + pg_offset + copy_size, 0,
6909 PAGE_SIZE - pg_offset -
6914 flush_dcache_page(page);
6915 } else if (create && PageUptodate(page)) {
6919 free_extent_map(em);
6922 btrfs_release_path(path);
6923 trans = btrfs_join_transaction(root);
6926 return ERR_CAST(trans);
6930 write_extent_buffer(leaf, map + pg_offset, ptr,
6933 btrfs_mark_buffer_dirty(leaf);
6935 set_extent_uptodate(io_tree, em->start,
6936 extent_map_end(em) - 1, NULL, GFP_NOFS);
6941 em->orig_start = start;
6944 em->block_start = EXTENT_MAP_HOLE;
6945 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6947 btrfs_release_path(path);
6948 if (em->start > start || extent_map_end(em) <= start) {
6949 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6950 em->start, em->len, start, len);
6956 write_lock(&em_tree->lock);
6957 ret = add_extent_mapping(em_tree, em, 0);
6958 /* it is possible that someone inserted the extent into the tree
6959 * while we had the lock dropped. It is also possible that
6960 * an overlapping map exists in the tree
6962 if (ret == -EEXIST) {
6963 struct extent_map *existing;
6967 existing = search_extent_mapping(em_tree, start, len);
6969 * existing will always be non-NULL, since there must be
6970 * extent causing the -EEXIST.
6972 if (start >= extent_map_end(existing) ||
6973 start <= existing->start) {
6975 * The existing extent map is the one nearest to
6976 * the [start, start + len) range which overlaps
6978 err = merge_extent_mapping(em_tree, existing,
6980 free_extent_map(existing);
6982 free_extent_map(em);
6986 free_extent_map(em);
6991 write_unlock(&em_tree->lock);
6994 trace_btrfs_get_extent(root, em);
6996 btrfs_free_path(path);
6998 ret = btrfs_end_transaction(trans, root);
7003 free_extent_map(em);
7004 return ERR_PTR(err);
7006 BUG_ON(!em); /* Error is always set */
7010 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7011 size_t pg_offset, u64 start, u64 len,
7014 struct extent_map *em;
7015 struct extent_map *hole_em = NULL;
7016 u64 range_start = start;
7022 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7029 * - a pre-alloc extent,
7030 * there might actually be delalloc bytes behind it.
7032 if (em->block_start != EXTENT_MAP_HOLE &&
7033 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7039 /* check to see if we've wrapped (len == -1 or similar) */
7048 /* ok, we didn't find anything, lets look for delalloc */
7049 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7050 end, len, EXTENT_DELALLOC, 1);
7051 found_end = range_start + found;
7052 if (found_end < range_start)
7053 found_end = (u64)-1;
7056 * we didn't find anything useful, return
7057 * the original results from get_extent()
7059 if (range_start > end || found_end <= start) {
7065 /* adjust the range_start to make sure it doesn't
7066 * go backwards from the start they passed in
7068 range_start = max(start, range_start);
7069 found = found_end - range_start;
7072 u64 hole_start = start;
7075 em = alloc_extent_map();
7081 * when btrfs_get_extent can't find anything it
7082 * returns one huge hole
7084 * make sure what it found really fits our range, and
7085 * adjust to make sure it is based on the start from
7089 u64 calc_end = extent_map_end(hole_em);
7091 if (calc_end <= start || (hole_em->start > end)) {
7092 free_extent_map(hole_em);
7095 hole_start = max(hole_em->start, start);
7096 hole_len = calc_end - hole_start;
7100 if (hole_em && range_start > hole_start) {
7101 /* our hole starts before our delalloc, so we
7102 * have to return just the parts of the hole
7103 * that go until the delalloc starts
7105 em->len = min(hole_len,
7106 range_start - hole_start);
7107 em->start = hole_start;
7108 em->orig_start = hole_start;
7110 * don't adjust block start at all,
7111 * it is fixed at EXTENT_MAP_HOLE
7113 em->block_start = hole_em->block_start;
7114 em->block_len = hole_len;
7115 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7116 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7118 em->start = range_start;
7120 em->orig_start = range_start;
7121 em->block_start = EXTENT_MAP_DELALLOC;
7122 em->block_len = found;
7124 } else if (hole_em) {
7129 free_extent_map(hole_em);
7131 free_extent_map(em);
7132 return ERR_PTR(err);
7137 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7140 struct btrfs_root *root = BTRFS_I(inode)->root;
7141 struct extent_map *em;
7142 struct btrfs_key ins;
7146 alloc_hint = get_extent_allocation_hint(inode, start, len);
7147 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7148 alloc_hint, &ins, 1, 1);
7150 return ERR_PTR(ret);
7153 * Create the ordered extent before the extent map. This is to avoid
7154 * races with the fast fsync path that would lead to it logging file
7155 * extent items that point to disk extents that were not yet written to.
7156 * The fast fsync path collects ordered extents into a local list and
7157 * then collects all the new extent maps, so we must create the ordered
7158 * extent first and make sure the fast fsync path collects any new
7159 * ordered extents after collecting new extent maps as well.
7160 * The fsync path simply can not rely on inode_dio_wait() because it
7161 * causes deadlock with AIO.
7163 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7164 ins.offset, ins.offset, 0);
7166 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7167 return ERR_PTR(ret);
7170 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7172 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7173 ins.offset, ins.offset, ins.offset, 0);
7175 struct btrfs_ordered_extent *oe;
7177 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7178 oe = btrfs_lookup_ordered_extent(inode, start);
7182 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7183 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7184 btrfs_remove_ordered_extent(inode, oe);
7185 /* Once for our lookup and once for the ordered extents tree. */
7186 btrfs_put_ordered_extent(oe);
7187 btrfs_put_ordered_extent(oe);
7193 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7194 * block must be cow'd
7196 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7197 u64 *orig_start, u64 *orig_block_len,
7200 struct btrfs_trans_handle *trans;
7201 struct btrfs_path *path;
7203 struct extent_buffer *leaf;
7204 struct btrfs_root *root = BTRFS_I(inode)->root;
7205 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7206 struct btrfs_file_extent_item *fi;
7207 struct btrfs_key key;
7214 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7216 path = btrfs_alloc_path();
7220 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7225 slot = path->slots[0];
7228 /* can't find the item, must cow */
7235 leaf = path->nodes[0];
7236 btrfs_item_key_to_cpu(leaf, &key, slot);
7237 if (key.objectid != btrfs_ino(inode) ||
7238 key.type != BTRFS_EXTENT_DATA_KEY) {
7239 /* not our file or wrong item type, must cow */
7243 if (key.offset > offset) {
7244 /* Wrong offset, must cow */
7248 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7249 found_type = btrfs_file_extent_type(leaf, fi);
7250 if (found_type != BTRFS_FILE_EXTENT_REG &&
7251 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7252 /* not a regular extent, must cow */
7256 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7259 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7260 if (extent_end <= offset)
7263 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7264 if (disk_bytenr == 0)
7267 if (btrfs_file_extent_compression(leaf, fi) ||
7268 btrfs_file_extent_encryption(leaf, fi) ||
7269 btrfs_file_extent_other_encoding(leaf, fi))
7272 backref_offset = btrfs_file_extent_offset(leaf, fi);
7275 *orig_start = key.offset - backref_offset;
7276 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7277 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7280 if (btrfs_extent_readonly(root, disk_bytenr))
7283 num_bytes = min(offset + *len, extent_end) - offset;
7284 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7287 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7288 ret = test_range_bit(io_tree, offset, range_end,
7289 EXTENT_DELALLOC, 0, NULL);
7296 btrfs_release_path(path);
7299 * look for other files referencing this extent, if we
7300 * find any we must cow
7302 trans = btrfs_join_transaction(root);
7303 if (IS_ERR(trans)) {
7308 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7309 key.offset - backref_offset, disk_bytenr);
7310 btrfs_end_transaction(trans, root);
7317 * adjust disk_bytenr and num_bytes to cover just the bytes
7318 * in this extent we are about to write. If there
7319 * are any csums in that range we have to cow in order
7320 * to keep the csums correct
7322 disk_bytenr += backref_offset;
7323 disk_bytenr += offset - key.offset;
7324 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7327 * all of the above have passed, it is safe to overwrite this extent
7333 btrfs_free_path(path);
7337 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7339 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7341 void **pagep = NULL;
7342 struct page *page = NULL;
7346 start_idx = start >> PAGE_SHIFT;
7349 * end is the last byte in the last page. end == start is legal
7351 end_idx = end >> PAGE_SHIFT;
7355 /* Most of the code in this while loop is lifted from
7356 * find_get_page. It's been modified to begin searching from a
7357 * page and return just the first page found in that range. If the
7358 * found idx is less than or equal to the end idx then we know that
7359 * a page exists. If no pages are found or if those pages are
7360 * outside of the range then we're fine (yay!) */
7361 while (page == NULL &&
7362 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7363 page = radix_tree_deref_slot(pagep);
7364 if (unlikely(!page))
7367 if (radix_tree_exception(page)) {
7368 if (radix_tree_deref_retry(page)) {
7373 * Otherwise, shmem/tmpfs must be storing a swap entry
7374 * here as an exceptional entry: so return it without
7375 * attempting to raise page count.
7378 break; /* TODO: Is this relevant for this use case? */
7381 if (!page_cache_get_speculative(page)) {
7387 * Has the page moved?
7388 * This is part of the lockless pagecache protocol. See
7389 * include/linux/pagemap.h for details.
7391 if (unlikely(page != *pagep)) {
7398 if (page->index <= end_idx)
7407 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7408 struct extent_state **cached_state, int writing)
7410 struct btrfs_ordered_extent *ordered;
7414 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7417 * We're concerned with the entire range that we're going to be
7418 * doing DIO to, so we need to make sure theres no ordered
7419 * extents in this range.
7421 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7422 lockend - lockstart + 1);
7425 * We need to make sure there are no buffered pages in this
7426 * range either, we could have raced between the invalidate in
7427 * generic_file_direct_write and locking the extent. The
7428 * invalidate needs to happen so that reads after a write do not
7433 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7436 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7437 cached_state, GFP_NOFS);
7441 * If we are doing a DIO read and the ordered extent we
7442 * found is for a buffered write, we can not wait for it
7443 * to complete and retry, because if we do so we can
7444 * deadlock with concurrent buffered writes on page
7445 * locks. This happens only if our DIO read covers more
7446 * than one extent map, if at this point has already
7447 * created an ordered extent for a previous extent map
7448 * and locked its range in the inode's io tree, and a
7449 * concurrent write against that previous extent map's
7450 * range and this range started (we unlock the ranges
7451 * in the io tree only when the bios complete and
7452 * buffered writes always lock pages before attempting
7453 * to lock range in the io tree).
7456 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7457 btrfs_start_ordered_extent(inode, ordered, 1);
7460 btrfs_put_ordered_extent(ordered);
7463 * We could trigger writeback for this range (and wait
7464 * for it to complete) and then invalidate the pages for
7465 * this range (through invalidate_inode_pages2_range()),
7466 * but that can lead us to a deadlock with a concurrent
7467 * call to readpages() (a buffered read or a defrag call
7468 * triggered a readahead) on a page lock due to an
7469 * ordered dio extent we created before but did not have
7470 * yet a corresponding bio submitted (whence it can not
7471 * complete), which makes readpages() wait for that
7472 * ordered extent to complete while holding a lock on
7487 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7488 u64 len, u64 orig_start,
7489 u64 block_start, u64 block_len,
7490 u64 orig_block_len, u64 ram_bytes,
7493 struct extent_map_tree *em_tree;
7494 struct extent_map *em;
7495 struct btrfs_root *root = BTRFS_I(inode)->root;
7498 em_tree = &BTRFS_I(inode)->extent_tree;
7499 em = alloc_extent_map();
7501 return ERR_PTR(-ENOMEM);
7504 em->orig_start = orig_start;
7505 em->mod_start = start;
7508 em->block_len = block_len;
7509 em->block_start = block_start;
7510 em->bdev = root->fs_info->fs_devices->latest_bdev;
7511 em->orig_block_len = orig_block_len;
7512 em->ram_bytes = ram_bytes;
7513 em->generation = -1;
7514 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7515 if (type == BTRFS_ORDERED_PREALLOC)
7516 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7519 btrfs_drop_extent_cache(inode, em->start,
7520 em->start + em->len - 1, 0);
7521 write_lock(&em_tree->lock);
7522 ret = add_extent_mapping(em_tree, em, 1);
7523 write_unlock(&em_tree->lock);
7524 } while (ret == -EEXIST);
7527 free_extent_map(em);
7528 return ERR_PTR(ret);
7534 static void adjust_dio_outstanding_extents(struct inode *inode,
7535 struct btrfs_dio_data *dio_data,
7538 unsigned num_extents;
7540 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7541 BTRFS_MAX_EXTENT_SIZE);
7543 * If we have an outstanding_extents count still set then we're
7544 * within our reservation, otherwise we need to adjust our inode
7545 * counter appropriately.
7547 if (dio_data->outstanding_extents) {
7548 dio_data->outstanding_extents -= num_extents;
7550 spin_lock(&BTRFS_I(inode)->lock);
7551 BTRFS_I(inode)->outstanding_extents += num_extents;
7552 spin_unlock(&BTRFS_I(inode)->lock);
7556 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7557 struct buffer_head *bh_result, int create)
7559 struct extent_map *em;
7560 struct btrfs_root *root = BTRFS_I(inode)->root;
7561 struct extent_state *cached_state = NULL;
7562 struct btrfs_dio_data *dio_data = NULL;
7563 u64 start = iblock << inode->i_blkbits;
7564 u64 lockstart, lockend;
7565 u64 len = bh_result->b_size;
7566 int unlock_bits = EXTENT_LOCKED;
7570 unlock_bits |= EXTENT_DIRTY;
7572 len = min_t(u64, len, root->sectorsize);
7575 lockend = start + len - 1;
7577 if (current->journal_info) {
7579 * Need to pull our outstanding extents and set journal_info to NULL so
7580 * that anything that needs to check if there's a transction doesn't get
7583 dio_data = current->journal_info;
7584 current->journal_info = NULL;
7588 * If this errors out it's because we couldn't invalidate pagecache for
7589 * this range and we need to fallback to buffered.
7591 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7597 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7604 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7605 * io. INLINE is special, and we could probably kludge it in here, but
7606 * it's still buffered so for safety lets just fall back to the generic
7609 * For COMPRESSED we _have_ to read the entire extent in so we can
7610 * decompress it, so there will be buffering required no matter what we
7611 * do, so go ahead and fallback to buffered.
7613 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7614 * to buffered IO. Don't blame me, this is the price we pay for using
7617 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7618 em->block_start == EXTENT_MAP_INLINE) {
7619 free_extent_map(em);
7624 /* Just a good old fashioned hole, return */
7625 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7626 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7627 free_extent_map(em);
7632 * We don't allocate a new extent in the following cases
7634 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7636 * 2) The extent is marked as PREALLOC. We're good to go here and can
7637 * just use the extent.
7641 len = min(len, em->len - (start - em->start));
7642 lockstart = start + len;
7646 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7647 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7648 em->block_start != EXTENT_MAP_HOLE)) {
7650 u64 block_start, orig_start, orig_block_len, ram_bytes;
7652 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7653 type = BTRFS_ORDERED_PREALLOC;
7655 type = BTRFS_ORDERED_NOCOW;
7656 len = min(len, em->len - (start - em->start));
7657 block_start = em->block_start + (start - em->start);
7659 if (can_nocow_extent(inode, start, &len, &orig_start,
7660 &orig_block_len, &ram_bytes) == 1) {
7661 if (type == BTRFS_ORDERED_PREALLOC) {
7662 free_extent_map(em);
7663 em = create_pinned_em(inode, start, len,
7674 ret = btrfs_add_ordered_extent_dio(inode, start,
7675 block_start, len, len, type);
7677 free_extent_map(em);
7685 * this will cow the extent, reset the len in case we changed
7688 len = bh_result->b_size;
7689 free_extent_map(em);
7690 em = btrfs_new_extent_direct(inode, start, len);
7695 len = min(len, em->len - (start - em->start));
7697 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7699 bh_result->b_size = len;
7700 bh_result->b_bdev = em->bdev;
7701 set_buffer_mapped(bh_result);
7703 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7704 set_buffer_new(bh_result);
7707 * Need to update the i_size under the extent lock so buffered
7708 * readers will get the updated i_size when we unlock.
7710 if (start + len > i_size_read(inode))
7711 i_size_write(inode, start + len);
7713 adjust_dio_outstanding_extents(inode, dio_data, len);
7714 btrfs_free_reserved_data_space(inode, start, len);
7715 WARN_ON(dio_data->reserve < len);
7716 dio_data->reserve -= len;
7717 dio_data->unsubmitted_oe_range_end = start + len;
7718 current->journal_info = dio_data;
7722 * In the case of write we need to clear and unlock the entire range,
7723 * in the case of read we need to unlock only the end area that we
7724 * aren't using if there is any left over space.
7726 if (lockstart < lockend) {
7727 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7728 lockend, unlock_bits, 1, 0,
7729 &cached_state, GFP_NOFS);
7731 free_extent_state(cached_state);
7734 free_extent_map(em);
7739 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7740 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7743 current->journal_info = dio_data;
7745 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7746 * write less data then expected, so that we don't underflow our inode's
7747 * outstanding extents counter.
7749 if (create && dio_data)
7750 adjust_dio_outstanding_extents(inode, dio_data, len);
7755 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7756 int rw, int mirror_num)
7758 struct btrfs_root *root = BTRFS_I(inode)->root;
7761 BUG_ON(rw & REQ_WRITE);
7765 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7766 BTRFS_WQ_ENDIO_DIO_REPAIR);
7770 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7776 static int btrfs_check_dio_repairable(struct inode *inode,
7777 struct bio *failed_bio,
7778 struct io_failure_record *failrec,
7783 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7784 failrec->logical, failrec->len);
7785 if (num_copies == 1) {
7787 * we only have a single copy of the data, so don't bother with
7788 * all the retry and error correction code that follows. no
7789 * matter what the error is, it is very likely to persist.
7791 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7792 num_copies, failrec->this_mirror, failed_mirror);
7796 failrec->failed_mirror = failed_mirror;
7797 failrec->this_mirror++;
7798 if (failrec->this_mirror == failed_mirror)
7799 failrec->this_mirror++;
7801 if (failrec->this_mirror > num_copies) {
7802 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7803 num_copies, failrec->this_mirror, failed_mirror);
7810 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7811 struct page *page, unsigned int pgoff,
7812 u64 start, u64 end, int failed_mirror,
7813 bio_end_io_t *repair_endio, void *repair_arg)
7815 struct io_failure_record *failrec;
7821 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7823 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7827 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7830 free_io_failure(inode, failrec);
7834 if ((failed_bio->bi_vcnt > 1)
7835 || (failed_bio->bi_io_vec->bv_len
7836 > BTRFS_I(inode)->root->sectorsize))
7837 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7839 read_mode = READ_SYNC;
7841 isector = start - btrfs_io_bio(failed_bio)->logical;
7842 isector >>= inode->i_sb->s_blocksize_bits;
7843 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7844 pgoff, isector, repair_endio, repair_arg);
7846 free_io_failure(inode, failrec);
7850 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7851 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7852 read_mode, failrec->this_mirror, failrec->in_validation);
7854 ret = submit_dio_repair_bio(inode, bio, read_mode,
7855 failrec->this_mirror);
7857 free_io_failure(inode, failrec);
7864 struct btrfs_retry_complete {
7865 struct completion done;
7866 struct inode *inode;
7871 static void btrfs_retry_endio_nocsum(struct bio *bio)
7873 struct btrfs_retry_complete *done = bio->bi_private;
7874 struct inode *inode;
7875 struct bio_vec *bvec;
7881 ASSERT(bio->bi_vcnt == 1);
7882 inode = bio->bi_io_vec->bv_page->mapping->host;
7883 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7886 bio_for_each_segment_all(bvec, bio, i)
7887 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7889 complete(&done->done);
7893 static int __btrfs_correct_data_nocsum(struct inode *inode,
7894 struct btrfs_io_bio *io_bio)
7896 struct btrfs_fs_info *fs_info;
7897 struct bio_vec *bvec;
7898 struct btrfs_retry_complete done;
7906 fs_info = BTRFS_I(inode)->root->fs_info;
7907 sectorsize = BTRFS_I(inode)->root->sectorsize;
7909 start = io_bio->logical;
7912 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7913 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7914 pgoff = bvec->bv_offset;
7916 next_block_or_try_again:
7919 init_completion(&done.done);
7921 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7922 pgoff, start, start + sectorsize - 1,
7924 btrfs_retry_endio_nocsum, &done);
7928 wait_for_completion(&done.done);
7930 if (!done.uptodate) {
7931 /* We might have another mirror, so try again */
7932 goto next_block_or_try_again;
7935 start += sectorsize;
7938 pgoff += sectorsize;
7939 goto next_block_or_try_again;
7946 static void btrfs_retry_endio(struct bio *bio)
7948 struct btrfs_retry_complete *done = bio->bi_private;
7949 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7950 struct inode *inode;
7951 struct bio_vec *bvec;
7962 start = done->start;
7964 ASSERT(bio->bi_vcnt == 1);
7965 inode = bio->bi_io_vec->bv_page->mapping->host;
7966 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7968 bio_for_each_segment_all(bvec, bio, i) {
7969 ret = __readpage_endio_check(done->inode, io_bio, i,
7970 bvec->bv_page, bvec->bv_offset,
7971 done->start, bvec->bv_len);
7973 clean_io_failure(done->inode, done->start,
7974 bvec->bv_page, bvec->bv_offset);
7979 done->uptodate = uptodate;
7981 complete(&done->done);
7985 static int __btrfs_subio_endio_read(struct inode *inode,
7986 struct btrfs_io_bio *io_bio, int err)
7988 struct btrfs_fs_info *fs_info;
7989 struct bio_vec *bvec;
7990 struct btrfs_retry_complete done;
8000 fs_info = BTRFS_I(inode)->root->fs_info;
8001 sectorsize = BTRFS_I(inode)->root->sectorsize;
8004 start = io_bio->logical;
8007 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8008 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8010 pgoff = bvec->bv_offset;
8012 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8013 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8014 bvec->bv_page, pgoff, start,
8021 init_completion(&done.done);
8023 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8024 pgoff, start, start + sectorsize - 1,
8026 btrfs_retry_endio, &done);
8032 wait_for_completion(&done.done);
8034 if (!done.uptodate) {
8035 /* We might have another mirror, so try again */
8039 offset += sectorsize;
8040 start += sectorsize;
8045 pgoff += sectorsize;
8053 static int btrfs_subio_endio_read(struct inode *inode,
8054 struct btrfs_io_bio *io_bio, int err)
8056 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8060 return __btrfs_correct_data_nocsum(inode, io_bio);
8064 return __btrfs_subio_endio_read(inode, io_bio, err);
8068 static void btrfs_endio_direct_read(struct bio *bio)
8070 struct btrfs_dio_private *dip = bio->bi_private;
8071 struct inode *inode = dip->inode;
8072 struct bio *dio_bio;
8073 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8074 int err = bio->bi_error;
8076 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8077 err = btrfs_subio_endio_read(inode, io_bio, err);
8079 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8080 dip->logical_offset + dip->bytes - 1);
8081 dio_bio = dip->dio_bio;
8085 dio_bio->bi_error = bio->bi_error;
8086 dio_end_io(dio_bio, bio->bi_error);
8089 io_bio->end_io(io_bio, err);
8093 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8098 struct btrfs_root *root = BTRFS_I(inode)->root;
8099 struct btrfs_ordered_extent *ordered = NULL;
8100 u64 ordered_offset = offset;
8101 u64 ordered_bytes = bytes;
8105 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8112 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8113 finish_ordered_fn, NULL, NULL);
8114 btrfs_queue_work(root->fs_info->endio_write_workers,
8118 * our bio might span multiple ordered extents. If we haven't
8119 * completed the accounting for the whole dio, go back and try again
8121 if (ordered_offset < offset + bytes) {
8122 ordered_bytes = offset + bytes - ordered_offset;
8128 static void btrfs_endio_direct_write(struct bio *bio)
8130 struct btrfs_dio_private *dip = bio->bi_private;
8131 struct bio *dio_bio = dip->dio_bio;
8133 btrfs_endio_direct_write_update_ordered(dip->inode,
8134 dip->logical_offset,
8140 dio_bio->bi_error = bio->bi_error;
8141 dio_end_io(dio_bio, bio->bi_error);
8145 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8146 struct bio *bio, int mirror_num,
8147 unsigned long bio_flags, u64 offset)
8150 struct btrfs_root *root = BTRFS_I(inode)->root;
8151 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8152 BUG_ON(ret); /* -ENOMEM */
8156 static void btrfs_end_dio_bio(struct bio *bio)
8158 struct btrfs_dio_private *dip = bio->bi_private;
8159 int err = bio->bi_error;
8162 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8163 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8164 btrfs_ino(dip->inode), bio->bi_rw,
8165 (unsigned long long)bio->bi_iter.bi_sector,
8166 bio->bi_iter.bi_size, err);
8168 if (dip->subio_endio)
8169 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8175 * before atomic variable goto zero, we must make sure
8176 * dip->errors is perceived to be set.
8178 smp_mb__before_atomic();
8181 /* if there are more bios still pending for this dio, just exit */
8182 if (!atomic_dec_and_test(&dip->pending_bios))
8186 bio_io_error(dip->orig_bio);
8188 dip->dio_bio->bi_error = 0;
8189 bio_endio(dip->orig_bio);
8195 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8196 u64 first_sector, gfp_t gfp_flags)
8199 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8201 bio_associate_current(bio);
8205 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8206 struct inode *inode,
8207 struct btrfs_dio_private *dip,
8211 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8212 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8216 * We load all the csum data we need when we submit
8217 * the first bio to reduce the csum tree search and
8220 if (dip->logical_offset == file_offset) {
8221 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8227 if (bio == dip->orig_bio)
8230 file_offset -= dip->logical_offset;
8231 file_offset >>= inode->i_sb->s_blocksize_bits;
8232 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8237 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8238 int rw, u64 file_offset, int skip_sum,
8241 struct btrfs_dio_private *dip = bio->bi_private;
8242 int write = rw & REQ_WRITE;
8243 struct btrfs_root *root = BTRFS_I(inode)->root;
8247 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8252 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8253 BTRFS_WQ_ENDIO_DATA);
8261 if (write && async_submit) {
8262 ret = btrfs_wq_submit_bio(root->fs_info,
8263 inode, rw, bio, 0, 0,
8265 __btrfs_submit_bio_start_direct_io,
8266 __btrfs_submit_bio_done);
8270 * If we aren't doing async submit, calculate the csum of the
8273 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8277 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8283 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8289 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8292 struct inode *inode = dip->inode;
8293 struct btrfs_root *root = BTRFS_I(inode)->root;
8295 struct bio *orig_bio = dip->orig_bio;
8296 struct bio_vec *bvec = orig_bio->bi_io_vec;
8297 u64 start_sector = orig_bio->bi_iter.bi_sector;
8298 u64 file_offset = dip->logical_offset;
8301 u32 blocksize = root->sectorsize;
8302 int async_submit = 0;
8307 map_length = orig_bio->bi_iter.bi_size;
8308 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8309 &map_length, NULL, 0);
8313 if (map_length >= orig_bio->bi_iter.bi_size) {
8315 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8319 /* async crcs make it difficult to collect full stripe writes. */
8320 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8325 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8329 bio->bi_private = dip;
8330 bio->bi_end_io = btrfs_end_dio_bio;
8331 btrfs_io_bio(bio)->logical = file_offset;
8332 atomic_inc(&dip->pending_bios);
8334 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8335 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8338 if (unlikely(map_length < submit_len + blocksize ||
8339 bio_add_page(bio, bvec->bv_page, blocksize,
8340 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8342 * inc the count before we submit the bio so
8343 * we know the end IO handler won't happen before
8344 * we inc the count. Otherwise, the dip might get freed
8345 * before we're done setting it up
8347 atomic_inc(&dip->pending_bios);
8348 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8349 file_offset, skip_sum,
8353 atomic_dec(&dip->pending_bios);
8357 start_sector += submit_len >> 9;
8358 file_offset += submit_len;
8362 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8363 start_sector, GFP_NOFS);
8366 bio->bi_private = dip;
8367 bio->bi_end_io = btrfs_end_dio_bio;
8368 btrfs_io_bio(bio)->logical = file_offset;
8370 map_length = orig_bio->bi_iter.bi_size;
8371 ret = btrfs_map_block(root->fs_info, rw,
8373 &map_length, NULL, 0);
8381 submit_len += blocksize;
8391 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8400 * before atomic variable goto zero, we must
8401 * make sure dip->errors is perceived to be set.
8403 smp_mb__before_atomic();
8404 if (atomic_dec_and_test(&dip->pending_bios))
8405 bio_io_error(dip->orig_bio);
8407 /* bio_end_io() will handle error, so we needn't return it */
8411 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8412 struct inode *inode, loff_t file_offset)
8414 struct btrfs_dio_private *dip = NULL;
8415 struct bio *io_bio = NULL;
8416 struct btrfs_io_bio *btrfs_bio;
8418 int write = rw & REQ_WRITE;
8421 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8423 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8429 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8435 dip->private = dio_bio->bi_private;
8437 dip->logical_offset = file_offset;
8438 dip->bytes = dio_bio->bi_iter.bi_size;
8439 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8440 io_bio->bi_private = dip;
8441 dip->orig_bio = io_bio;
8442 dip->dio_bio = dio_bio;
8443 atomic_set(&dip->pending_bios, 0);
8444 btrfs_bio = btrfs_io_bio(io_bio);
8445 btrfs_bio->logical = file_offset;
8448 io_bio->bi_end_io = btrfs_endio_direct_write;
8450 io_bio->bi_end_io = btrfs_endio_direct_read;
8451 dip->subio_endio = btrfs_subio_endio_read;
8455 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8456 * even if we fail to submit a bio, because in such case we do the
8457 * corresponding error handling below and it must not be done a second
8458 * time by btrfs_direct_IO().
8461 struct btrfs_dio_data *dio_data = current->journal_info;
8463 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8465 dio_data->unsubmitted_oe_range_start =
8466 dio_data->unsubmitted_oe_range_end;
8469 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8473 if (btrfs_bio->end_io)
8474 btrfs_bio->end_io(btrfs_bio, ret);
8478 * If we arrived here it means either we failed to submit the dip
8479 * or we either failed to clone the dio_bio or failed to allocate the
8480 * dip. If we cloned the dio_bio and allocated the dip, we can just
8481 * call bio_endio against our io_bio so that we get proper resource
8482 * cleanup if we fail to submit the dip, otherwise, we must do the
8483 * same as btrfs_endio_direct_[write|read] because we can't call these
8484 * callbacks - they require an allocated dip and a clone of dio_bio.
8486 if (io_bio && dip) {
8487 io_bio->bi_error = -EIO;
8490 * The end io callbacks free our dip, do the final put on io_bio
8491 * and all the cleanup and final put for dio_bio (through
8498 btrfs_endio_direct_write_update_ordered(inode,
8500 dio_bio->bi_iter.bi_size,
8503 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8504 file_offset + dio_bio->bi_iter.bi_size - 1);
8506 dio_bio->bi_error = -EIO;
8508 * Releases and cleans up our dio_bio, no need to bio_put()
8509 * nor bio_endio()/bio_io_error() against dio_bio.
8511 dio_end_io(dio_bio, ret);
8518 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8519 const struct iov_iter *iter, loff_t offset)
8523 unsigned blocksize_mask = root->sectorsize - 1;
8524 ssize_t retval = -EINVAL;
8526 if (offset & blocksize_mask)
8529 if (iov_iter_alignment(iter) & blocksize_mask)
8532 /* If this is a write we don't need to check anymore */
8533 if (iov_iter_rw(iter) == WRITE)
8536 * Check to make sure we don't have duplicate iov_base's in this
8537 * iovec, if so return EINVAL, otherwise we'll get csum errors
8538 * when reading back.
8540 for (seg = 0; seg < iter->nr_segs; seg++) {
8541 for (i = seg + 1; i < iter->nr_segs; i++) {
8542 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8551 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8554 struct file *file = iocb->ki_filp;
8555 struct inode *inode = file->f_mapping->host;
8556 struct btrfs_root *root = BTRFS_I(inode)->root;
8557 struct btrfs_dio_data dio_data = { 0 };
8561 bool relock = false;
8564 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8567 inode_dio_begin(inode);
8568 smp_mb__after_atomic();
8571 * The generic stuff only does filemap_write_and_wait_range, which
8572 * isn't enough if we've written compressed pages to this area, so
8573 * we need to flush the dirty pages again to make absolutely sure
8574 * that any outstanding dirty pages are on disk.
8576 count = iov_iter_count(iter);
8577 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8578 &BTRFS_I(inode)->runtime_flags))
8579 filemap_fdatawrite_range(inode->i_mapping, offset,
8580 offset + count - 1);
8582 if (iov_iter_rw(iter) == WRITE) {
8584 * If the write DIO is beyond the EOF, we need update
8585 * the isize, but it is protected by i_mutex. So we can
8586 * not unlock the i_mutex at this case.
8588 if (offset + count <= inode->i_size) {
8589 inode_unlock(inode);
8592 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8595 dio_data.outstanding_extents = div64_u64(count +
8596 BTRFS_MAX_EXTENT_SIZE - 1,
8597 BTRFS_MAX_EXTENT_SIZE);
8600 * We need to know how many extents we reserved so that we can
8601 * do the accounting properly if we go over the number we
8602 * originally calculated. Abuse current->journal_info for this.
8604 dio_data.reserve = round_up(count, root->sectorsize);
8605 dio_data.unsubmitted_oe_range_start = (u64)offset;
8606 dio_data.unsubmitted_oe_range_end = (u64)offset;
8607 current->journal_info = &dio_data;
8608 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8609 &BTRFS_I(inode)->runtime_flags)) {
8610 inode_dio_end(inode);
8611 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8615 ret = __blockdev_direct_IO(iocb, inode,
8616 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8617 iter, offset, btrfs_get_blocks_direct, NULL,
8618 btrfs_submit_direct, flags);
8619 if (iov_iter_rw(iter) == WRITE) {
8620 current->journal_info = NULL;
8621 if (ret < 0 && ret != -EIOCBQUEUED) {
8622 if (dio_data.reserve)
8623 btrfs_delalloc_release_space(inode, offset,
8626 * On error we might have left some ordered extents
8627 * without submitting corresponding bios for them, so
8628 * cleanup them up to avoid other tasks getting them
8629 * and waiting for them to complete forever.
8631 if (dio_data.unsubmitted_oe_range_start <
8632 dio_data.unsubmitted_oe_range_end)
8633 btrfs_endio_direct_write_update_ordered(inode,
8634 dio_data.unsubmitted_oe_range_start,
8635 dio_data.unsubmitted_oe_range_end -
8636 dio_data.unsubmitted_oe_range_start,
8638 } else if (ret >= 0 && (size_t)ret < count)
8639 btrfs_delalloc_release_space(inode, offset,
8640 count - (size_t)ret);
8644 inode_dio_end(inode);
8651 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8653 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8654 __u64 start, __u64 len)
8658 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8662 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8665 int btrfs_readpage(struct file *file, struct page *page)
8667 struct extent_io_tree *tree;
8668 tree = &BTRFS_I(page->mapping->host)->io_tree;
8669 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8672 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8674 struct extent_io_tree *tree;
8675 struct inode *inode = page->mapping->host;
8678 if (current->flags & PF_MEMALLOC) {
8679 redirty_page_for_writepage(wbc, page);
8685 * If we are under memory pressure we will call this directly from the
8686 * VM, we need to make sure we have the inode referenced for the ordered
8687 * extent. If not just return like we didn't do anything.
8689 if (!igrab(inode)) {
8690 redirty_page_for_writepage(wbc, page);
8691 return AOP_WRITEPAGE_ACTIVATE;
8693 tree = &BTRFS_I(page->mapping->host)->io_tree;
8694 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8695 btrfs_add_delayed_iput(inode);
8699 static int btrfs_writepages(struct address_space *mapping,
8700 struct writeback_control *wbc)
8702 struct extent_io_tree *tree;
8704 tree = &BTRFS_I(mapping->host)->io_tree;
8705 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8709 btrfs_readpages(struct file *file, struct address_space *mapping,
8710 struct list_head *pages, unsigned nr_pages)
8712 struct extent_io_tree *tree;
8713 tree = &BTRFS_I(mapping->host)->io_tree;
8714 return extent_readpages(tree, mapping, pages, nr_pages,
8717 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8719 struct extent_io_tree *tree;
8720 struct extent_map_tree *map;
8723 tree = &BTRFS_I(page->mapping->host)->io_tree;
8724 map = &BTRFS_I(page->mapping->host)->extent_tree;
8725 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8727 ClearPagePrivate(page);
8728 set_page_private(page, 0);
8734 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8736 if (PageWriteback(page) || PageDirty(page))
8738 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8741 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8742 unsigned int length)
8744 struct inode *inode = page->mapping->host;
8745 struct extent_io_tree *tree;
8746 struct btrfs_ordered_extent *ordered;
8747 struct extent_state *cached_state = NULL;
8748 u64 page_start = page_offset(page);
8749 u64 page_end = page_start + PAGE_SIZE - 1;
8752 int inode_evicting = inode->i_state & I_FREEING;
8755 * we have the page locked, so new writeback can't start,
8756 * and the dirty bit won't be cleared while we are here.
8758 * Wait for IO on this page so that we can safely clear
8759 * the PagePrivate2 bit and do ordered accounting
8761 wait_on_page_writeback(page);
8763 tree = &BTRFS_I(inode)->io_tree;
8765 btrfs_releasepage(page, GFP_NOFS);
8769 if (!inode_evicting)
8770 lock_extent_bits(tree, page_start, page_end, &cached_state);
8773 ordered = btrfs_lookup_ordered_range(inode, start,
8774 page_end - start + 1);
8776 end = min(page_end, ordered->file_offset + ordered->len - 1);
8778 * IO on this page will never be started, so we need
8779 * to account for any ordered extents now
8781 if (!inode_evicting)
8782 clear_extent_bit(tree, start, end,
8783 EXTENT_DIRTY | EXTENT_DELALLOC |
8784 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8785 EXTENT_DEFRAG, 1, 0, &cached_state,
8788 * whoever cleared the private bit is responsible
8789 * for the finish_ordered_io
8791 if (TestClearPagePrivate2(page)) {
8792 struct btrfs_ordered_inode_tree *tree;
8795 tree = &BTRFS_I(inode)->ordered_tree;
8797 spin_lock_irq(&tree->lock);
8798 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8799 new_len = start - ordered->file_offset;
8800 if (new_len < ordered->truncated_len)
8801 ordered->truncated_len = new_len;
8802 spin_unlock_irq(&tree->lock);
8804 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8806 end - start + 1, 1))
8807 btrfs_finish_ordered_io(ordered);
8809 btrfs_put_ordered_extent(ordered);
8810 if (!inode_evicting) {
8811 cached_state = NULL;
8812 lock_extent_bits(tree, start, end,
8817 if (start < page_end)
8822 * Qgroup reserved space handler
8823 * Page here will be either
8824 * 1) Already written to disk
8825 * In this case, its reserved space is released from data rsv map
8826 * and will be freed by delayed_ref handler finally.
8827 * So even we call qgroup_free_data(), it won't decrease reserved
8829 * 2) Not written to disk
8830 * This means the reserved space should be freed here.
8832 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8833 if (!inode_evicting) {
8834 clear_extent_bit(tree, page_start, page_end,
8835 EXTENT_LOCKED | EXTENT_DIRTY |
8836 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8837 EXTENT_DEFRAG, 1, 1,
8838 &cached_state, GFP_NOFS);
8840 __btrfs_releasepage(page, GFP_NOFS);
8843 ClearPageChecked(page);
8844 if (PagePrivate(page)) {
8845 ClearPagePrivate(page);
8846 set_page_private(page, 0);
8852 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8853 * called from a page fault handler when a page is first dirtied. Hence we must
8854 * be careful to check for EOF conditions here. We set the page up correctly
8855 * for a written page which means we get ENOSPC checking when writing into
8856 * holes and correct delalloc and unwritten extent mapping on filesystems that
8857 * support these features.
8859 * We are not allowed to take the i_mutex here so we have to play games to
8860 * protect against truncate races as the page could now be beyond EOF. Because
8861 * vmtruncate() writes the inode size before removing pages, once we have the
8862 * page lock we can determine safely if the page is beyond EOF. If it is not
8863 * beyond EOF, then the page is guaranteed safe against truncation until we
8866 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8868 struct page *page = vmf->page;
8869 struct inode *inode = file_inode(vma->vm_file);
8870 struct btrfs_root *root = BTRFS_I(inode)->root;
8871 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8872 struct btrfs_ordered_extent *ordered;
8873 struct extent_state *cached_state = NULL;
8875 unsigned long zero_start;
8884 reserved_space = PAGE_SIZE;
8886 sb_start_pagefault(inode->i_sb);
8887 page_start = page_offset(page);
8888 page_end = page_start + PAGE_SIZE - 1;
8892 * Reserving delalloc space after obtaining the page lock can lead to
8893 * deadlock. For example, if a dirty page is locked by this function
8894 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8895 * dirty page write out, then the btrfs_writepage() function could
8896 * end up waiting indefinitely to get a lock on the page currently
8897 * being processed by btrfs_page_mkwrite() function.
8899 ret = btrfs_delalloc_reserve_space(inode, page_start,
8902 ret = file_update_time(vma->vm_file);
8908 else /* -ENOSPC, -EIO, etc */
8909 ret = VM_FAULT_SIGBUS;
8915 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8918 size = i_size_read(inode);
8920 if ((page->mapping != inode->i_mapping) ||
8921 (page_start >= size)) {
8922 /* page got truncated out from underneath us */
8925 wait_on_page_writeback(page);
8927 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8928 set_page_extent_mapped(page);
8931 * we can't set the delalloc bits if there are pending ordered
8932 * extents. Drop our locks and wait for them to finish
8934 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8936 unlock_extent_cached(io_tree, page_start, page_end,
8937 &cached_state, GFP_NOFS);
8939 btrfs_start_ordered_extent(inode, ordered, 1);
8940 btrfs_put_ordered_extent(ordered);
8944 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8945 reserved_space = round_up(size - page_start, root->sectorsize);
8946 if (reserved_space < PAGE_SIZE) {
8947 end = page_start + reserved_space - 1;
8948 spin_lock(&BTRFS_I(inode)->lock);
8949 BTRFS_I(inode)->outstanding_extents++;
8950 spin_unlock(&BTRFS_I(inode)->lock);
8951 btrfs_delalloc_release_space(inode, page_start,
8952 PAGE_SIZE - reserved_space);
8957 * XXX - page_mkwrite gets called every time the page is dirtied, even
8958 * if it was already dirty, so for space accounting reasons we need to
8959 * clear any delalloc bits for the range we are fixing to save. There
8960 * is probably a better way to do this, but for now keep consistent with
8961 * prepare_pages in the normal write path.
8963 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8964 EXTENT_DIRTY | EXTENT_DELALLOC |
8965 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8966 0, 0, &cached_state, GFP_NOFS);
8968 ret = btrfs_set_extent_delalloc(inode, page_start, end,
8971 unlock_extent_cached(io_tree, page_start, page_end,
8972 &cached_state, GFP_NOFS);
8973 ret = VM_FAULT_SIGBUS;
8978 /* page is wholly or partially inside EOF */
8979 if (page_start + PAGE_SIZE > size)
8980 zero_start = size & ~PAGE_MASK;
8982 zero_start = PAGE_SIZE;
8984 if (zero_start != PAGE_SIZE) {
8986 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8987 flush_dcache_page(page);
8990 ClearPageChecked(page);
8991 set_page_dirty(page);
8992 SetPageUptodate(page);
8994 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8995 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8996 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8998 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9002 sb_end_pagefault(inode->i_sb);
9003 return VM_FAULT_LOCKED;
9007 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9009 sb_end_pagefault(inode->i_sb);
9013 static int btrfs_truncate(struct inode *inode)
9015 struct btrfs_root *root = BTRFS_I(inode)->root;
9016 struct btrfs_block_rsv *rsv;
9019 struct btrfs_trans_handle *trans;
9020 u64 mask = root->sectorsize - 1;
9021 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9023 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9029 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
9030 * 3 things going on here
9032 * 1) We need to reserve space for our orphan item and the space to
9033 * delete our orphan item. Lord knows we don't want to have a dangling
9034 * orphan item because we didn't reserve space to remove it.
9036 * 2) We need to reserve space to update our inode.
9038 * 3) We need to have something to cache all the space that is going to
9039 * be free'd up by the truncate operation, but also have some slack
9040 * space reserved in case it uses space during the truncate (thank you
9041 * very much snapshotting).
9043 * And we need these to all be seperate. The fact is we can use alot of
9044 * space doing the truncate, and we have no earthly idea how much space
9045 * we will use, so we need the truncate reservation to be seperate so it
9046 * doesn't end up using space reserved for updating the inode or
9047 * removing the orphan item. We also need to be able to stop the
9048 * transaction and start a new one, which means we need to be able to
9049 * update the inode several times, and we have no idea of knowing how
9050 * many times that will be, so we can't just reserve 1 item for the
9051 * entirety of the opration, so that has to be done seperately as well.
9052 * Then there is the orphan item, which does indeed need to be held on
9053 * to for the whole operation, and we need nobody to touch this reserved
9054 * space except the orphan code.
9056 * So that leaves us with
9058 * 1) root->orphan_block_rsv - for the orphan deletion.
9059 * 2) rsv - for the truncate reservation, which we will steal from the
9060 * transaction reservation.
9061 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9062 * updating the inode.
9064 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9067 rsv->size = min_size;
9071 * 1 for the truncate slack space
9072 * 1 for updating the inode.
9074 trans = btrfs_start_transaction(root, 2);
9075 if (IS_ERR(trans)) {
9076 err = PTR_ERR(trans);
9080 /* Migrate the slack space for the truncate to our reserve */
9081 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9086 * So if we truncate and then write and fsync we normally would just
9087 * write the extents that changed, which is a problem if we need to
9088 * first truncate that entire inode. So set this flag so we write out
9089 * all of the extents in the inode to the sync log so we're completely
9092 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9093 trans->block_rsv = rsv;
9096 ret = btrfs_truncate_inode_items(trans, root, inode,
9098 BTRFS_EXTENT_DATA_KEY);
9099 if (ret != -ENOSPC && ret != -EAGAIN) {
9104 trans->block_rsv = &root->fs_info->trans_block_rsv;
9105 ret = btrfs_update_inode(trans, root, inode);
9111 btrfs_end_transaction(trans, root);
9112 btrfs_btree_balance_dirty(root);
9114 trans = btrfs_start_transaction(root, 2);
9115 if (IS_ERR(trans)) {
9116 ret = err = PTR_ERR(trans);
9121 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9123 BUG_ON(ret); /* shouldn't happen */
9124 trans->block_rsv = rsv;
9127 if (ret == 0 && inode->i_nlink > 0) {
9128 trans->block_rsv = root->orphan_block_rsv;
9129 ret = btrfs_orphan_del(trans, inode);
9135 trans->block_rsv = &root->fs_info->trans_block_rsv;
9136 ret = btrfs_update_inode(trans, root, inode);
9140 ret = btrfs_end_transaction(trans, root);
9141 btrfs_btree_balance_dirty(root);
9145 btrfs_free_block_rsv(root, rsv);
9154 * create a new subvolume directory/inode (helper for the ioctl).
9156 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9157 struct btrfs_root *new_root,
9158 struct btrfs_root *parent_root,
9161 struct inode *inode;
9165 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9166 new_dirid, new_dirid,
9167 S_IFDIR | (~current_umask() & S_IRWXUGO),
9170 return PTR_ERR(inode);
9171 inode->i_op = &btrfs_dir_inode_operations;
9172 inode->i_fop = &btrfs_dir_file_operations;
9174 set_nlink(inode, 1);
9175 btrfs_i_size_write(inode, 0);
9176 unlock_new_inode(inode);
9178 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9180 btrfs_err(new_root->fs_info,
9181 "error inheriting subvolume %llu properties: %d",
9182 new_root->root_key.objectid, err);
9184 err = btrfs_update_inode(trans, new_root, inode);
9190 struct inode *btrfs_alloc_inode(struct super_block *sb)
9192 struct btrfs_inode *ei;
9193 struct inode *inode;
9195 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9202 ei->last_sub_trans = 0;
9203 ei->logged_trans = 0;
9204 ei->delalloc_bytes = 0;
9205 ei->defrag_bytes = 0;
9206 ei->disk_i_size = 0;
9209 ei->index_cnt = (u64)-1;
9211 ei->last_unlink_trans = 0;
9212 ei->last_log_commit = 0;
9213 ei->delayed_iput_count = 0;
9215 spin_lock_init(&ei->lock);
9216 ei->outstanding_extents = 0;
9217 ei->reserved_extents = 0;
9219 ei->runtime_flags = 0;
9220 ei->force_compress = BTRFS_COMPRESS_NONE;
9222 ei->delayed_node = NULL;
9224 ei->i_otime.tv_sec = 0;
9225 ei->i_otime.tv_nsec = 0;
9227 inode = &ei->vfs_inode;
9228 extent_map_tree_init(&ei->extent_tree);
9229 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9230 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9231 ei->io_tree.track_uptodate = 1;
9232 ei->io_failure_tree.track_uptodate = 1;
9233 atomic_set(&ei->sync_writers, 0);
9234 mutex_init(&ei->log_mutex);
9235 mutex_init(&ei->delalloc_mutex);
9236 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9237 INIT_LIST_HEAD(&ei->delalloc_inodes);
9238 INIT_LIST_HEAD(&ei->delayed_iput);
9239 RB_CLEAR_NODE(&ei->rb_node);
9244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9245 void btrfs_test_destroy_inode(struct inode *inode)
9247 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9248 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9252 static void btrfs_i_callback(struct rcu_head *head)
9254 struct inode *inode = container_of(head, struct inode, i_rcu);
9255 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9258 void btrfs_destroy_inode(struct inode *inode)
9260 struct btrfs_ordered_extent *ordered;
9261 struct btrfs_root *root = BTRFS_I(inode)->root;
9263 WARN_ON(!hlist_empty(&inode->i_dentry));
9264 WARN_ON(inode->i_data.nrpages);
9265 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9266 WARN_ON(BTRFS_I(inode)->reserved_extents);
9267 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9268 WARN_ON(BTRFS_I(inode)->csum_bytes);
9269 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9272 * This can happen where we create an inode, but somebody else also
9273 * created the same inode and we need to destroy the one we already
9279 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9280 &BTRFS_I(inode)->runtime_flags)) {
9281 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9283 atomic_dec(&root->orphan_inodes);
9287 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9291 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9292 ordered->file_offset, ordered->len);
9293 btrfs_remove_ordered_extent(inode, ordered);
9294 btrfs_put_ordered_extent(ordered);
9295 btrfs_put_ordered_extent(ordered);
9298 btrfs_qgroup_check_reserved_leak(inode);
9299 inode_tree_del(inode);
9300 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9302 call_rcu(&inode->i_rcu, btrfs_i_callback);
9305 int btrfs_drop_inode(struct inode *inode)
9307 struct btrfs_root *root = BTRFS_I(inode)->root;
9312 /* the snap/subvol tree is on deleting */
9313 if (btrfs_root_refs(&root->root_item) == 0)
9316 return generic_drop_inode(inode);
9319 static void init_once(void *foo)
9321 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9323 inode_init_once(&ei->vfs_inode);
9326 void btrfs_destroy_cachep(void)
9329 * Make sure all delayed rcu free inodes are flushed before we
9333 kmem_cache_destroy(btrfs_inode_cachep);
9334 kmem_cache_destroy(btrfs_trans_handle_cachep);
9335 kmem_cache_destroy(btrfs_transaction_cachep);
9336 kmem_cache_destroy(btrfs_path_cachep);
9337 kmem_cache_destroy(btrfs_free_space_cachep);
9340 int btrfs_init_cachep(void)
9342 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9343 sizeof(struct btrfs_inode), 0,
9344 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9346 if (!btrfs_inode_cachep)
9349 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9350 sizeof(struct btrfs_trans_handle), 0,
9351 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9352 if (!btrfs_trans_handle_cachep)
9355 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9356 sizeof(struct btrfs_transaction), 0,
9357 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9358 if (!btrfs_transaction_cachep)
9361 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9362 sizeof(struct btrfs_path), 0,
9363 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9364 if (!btrfs_path_cachep)
9367 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9368 sizeof(struct btrfs_free_space), 0,
9369 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9370 if (!btrfs_free_space_cachep)
9375 btrfs_destroy_cachep();
9379 static int btrfs_getattr(struct vfsmount *mnt,
9380 struct dentry *dentry, struct kstat *stat)
9383 struct inode *inode = d_inode(dentry);
9384 u32 blocksize = inode->i_sb->s_blocksize;
9386 generic_fillattr(inode, stat);
9387 stat->dev = BTRFS_I(inode)->root->anon_dev;
9389 spin_lock(&BTRFS_I(inode)->lock);
9390 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9391 spin_unlock(&BTRFS_I(inode)->lock);
9392 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9393 ALIGN(delalloc_bytes, blocksize)) >> 9;
9397 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9398 struct inode *new_dir, struct dentry *new_dentry)
9400 struct btrfs_trans_handle *trans;
9401 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9402 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9403 struct inode *new_inode = d_inode(new_dentry);
9404 struct inode *old_inode = d_inode(old_dentry);
9408 u64 old_ino = btrfs_ino(old_inode);
9410 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9413 /* we only allow rename subvolume link between subvolumes */
9414 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9417 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9418 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9421 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9422 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9426 /* check for collisions, even if the name isn't there */
9427 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9428 new_dentry->d_name.name,
9429 new_dentry->d_name.len);
9432 if (ret == -EEXIST) {
9434 * eexist without a new_inode */
9435 if (WARN_ON(!new_inode)) {
9439 /* maybe -EOVERFLOW */
9446 * we're using rename to replace one file with another. Start IO on it
9447 * now so we don't add too much work to the end of the transaction
9449 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9450 filemap_flush(old_inode->i_mapping);
9452 /* close the racy window with snapshot create/destroy ioctl */
9453 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9454 down_read(&root->fs_info->subvol_sem);
9456 * We want to reserve the absolute worst case amount of items. So if
9457 * both inodes are subvols and we need to unlink them then that would
9458 * require 4 item modifications, but if they are both normal inodes it
9459 * would require 5 item modifications, so we'll assume their normal
9460 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9461 * should cover the worst case number of items we'll modify.
9463 trans = btrfs_start_transaction(root, 11);
9464 if (IS_ERR(trans)) {
9465 ret = PTR_ERR(trans);
9470 btrfs_record_root_in_trans(trans, dest);
9472 ret = btrfs_set_inode_index(new_dir, &index);
9476 BTRFS_I(old_inode)->dir_index = 0ULL;
9477 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9478 /* force full log commit if subvolume involved. */
9479 btrfs_set_log_full_commit(root->fs_info, trans);
9481 ret = btrfs_insert_inode_ref(trans, dest,
9482 new_dentry->d_name.name,
9483 new_dentry->d_name.len,
9485 btrfs_ino(new_dir), index);
9489 * this is an ugly little race, but the rename is required
9490 * to make sure that if we crash, the inode is either at the
9491 * old name or the new one. pinning the log transaction lets
9492 * us make sure we don't allow a log commit to come in after
9493 * we unlink the name but before we add the new name back in.
9495 btrfs_pin_log_trans(root);
9498 inode_inc_iversion(old_dir);
9499 inode_inc_iversion(new_dir);
9500 inode_inc_iversion(old_inode);
9501 old_dir->i_ctime = old_dir->i_mtime =
9502 new_dir->i_ctime = new_dir->i_mtime =
9503 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9505 if (old_dentry->d_parent != new_dentry->d_parent)
9506 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9508 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9509 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9510 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9511 old_dentry->d_name.name,
9512 old_dentry->d_name.len);
9514 ret = __btrfs_unlink_inode(trans, root, old_dir,
9515 d_inode(old_dentry),
9516 old_dentry->d_name.name,
9517 old_dentry->d_name.len);
9519 ret = btrfs_update_inode(trans, root, old_inode);
9522 btrfs_abort_transaction(trans, root, ret);
9527 inode_inc_iversion(new_inode);
9528 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9529 if (unlikely(btrfs_ino(new_inode) ==
9530 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9531 root_objectid = BTRFS_I(new_inode)->location.objectid;
9532 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9534 new_dentry->d_name.name,
9535 new_dentry->d_name.len);
9536 BUG_ON(new_inode->i_nlink == 0);
9538 ret = btrfs_unlink_inode(trans, dest, new_dir,
9539 d_inode(new_dentry),
9540 new_dentry->d_name.name,
9541 new_dentry->d_name.len);
9543 if (!ret && new_inode->i_nlink == 0)
9544 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9546 btrfs_abort_transaction(trans, root, ret);
9551 ret = btrfs_add_link(trans, new_dir, old_inode,
9552 new_dentry->d_name.name,
9553 new_dentry->d_name.len, 0, index);
9555 btrfs_abort_transaction(trans, root, ret);
9559 if (old_inode->i_nlink == 1)
9560 BTRFS_I(old_inode)->dir_index = index;
9562 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9563 struct dentry *parent = new_dentry->d_parent;
9564 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9565 btrfs_end_log_trans(root);
9568 btrfs_end_transaction(trans, root);
9570 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9571 up_read(&root->fs_info->subvol_sem);
9576 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9577 struct inode *new_dir, struct dentry *new_dentry,
9580 if (flags & ~RENAME_NOREPLACE)
9583 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9586 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9588 struct btrfs_delalloc_work *delalloc_work;
9589 struct inode *inode;
9591 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9593 inode = delalloc_work->inode;
9594 filemap_flush(inode->i_mapping);
9595 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9596 &BTRFS_I(inode)->runtime_flags))
9597 filemap_flush(inode->i_mapping);
9599 if (delalloc_work->delay_iput)
9600 btrfs_add_delayed_iput(inode);
9603 complete(&delalloc_work->completion);
9606 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9609 struct btrfs_delalloc_work *work;
9611 work = kmalloc(sizeof(*work), GFP_NOFS);
9615 init_completion(&work->completion);
9616 INIT_LIST_HEAD(&work->list);
9617 work->inode = inode;
9618 work->delay_iput = delay_iput;
9619 WARN_ON_ONCE(!inode);
9620 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9621 btrfs_run_delalloc_work, NULL, NULL);
9626 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9628 wait_for_completion(&work->completion);
9633 * some fairly slow code that needs optimization. This walks the list
9634 * of all the inodes with pending delalloc and forces them to disk.
9636 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9639 struct btrfs_inode *binode;
9640 struct inode *inode;
9641 struct btrfs_delalloc_work *work, *next;
9642 struct list_head works;
9643 struct list_head splice;
9646 INIT_LIST_HEAD(&works);
9647 INIT_LIST_HEAD(&splice);
9649 mutex_lock(&root->delalloc_mutex);
9650 spin_lock(&root->delalloc_lock);
9651 list_splice_init(&root->delalloc_inodes, &splice);
9652 while (!list_empty(&splice)) {
9653 binode = list_entry(splice.next, struct btrfs_inode,
9656 list_move_tail(&binode->delalloc_inodes,
9657 &root->delalloc_inodes);
9658 inode = igrab(&binode->vfs_inode);
9660 cond_resched_lock(&root->delalloc_lock);
9663 spin_unlock(&root->delalloc_lock);
9665 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9668 btrfs_add_delayed_iput(inode);
9674 list_add_tail(&work->list, &works);
9675 btrfs_queue_work(root->fs_info->flush_workers,
9678 if (nr != -1 && ret >= nr)
9681 spin_lock(&root->delalloc_lock);
9683 spin_unlock(&root->delalloc_lock);
9686 list_for_each_entry_safe(work, next, &works, list) {
9687 list_del_init(&work->list);
9688 btrfs_wait_and_free_delalloc_work(work);
9691 if (!list_empty_careful(&splice)) {
9692 spin_lock(&root->delalloc_lock);
9693 list_splice_tail(&splice, &root->delalloc_inodes);
9694 spin_unlock(&root->delalloc_lock);
9696 mutex_unlock(&root->delalloc_mutex);
9700 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9704 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9707 ret = __start_delalloc_inodes(root, delay_iput, -1);
9711 * the filemap_flush will queue IO into the worker threads, but
9712 * we have to make sure the IO is actually started and that
9713 * ordered extents get created before we return
9715 atomic_inc(&root->fs_info->async_submit_draining);
9716 while (atomic_read(&root->fs_info->nr_async_submits) ||
9717 atomic_read(&root->fs_info->async_delalloc_pages)) {
9718 wait_event(root->fs_info->async_submit_wait,
9719 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9720 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9722 atomic_dec(&root->fs_info->async_submit_draining);
9726 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9729 struct btrfs_root *root;
9730 struct list_head splice;
9733 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9736 INIT_LIST_HEAD(&splice);
9738 mutex_lock(&fs_info->delalloc_root_mutex);
9739 spin_lock(&fs_info->delalloc_root_lock);
9740 list_splice_init(&fs_info->delalloc_roots, &splice);
9741 while (!list_empty(&splice) && nr) {
9742 root = list_first_entry(&splice, struct btrfs_root,
9744 root = btrfs_grab_fs_root(root);
9746 list_move_tail(&root->delalloc_root,
9747 &fs_info->delalloc_roots);
9748 spin_unlock(&fs_info->delalloc_root_lock);
9750 ret = __start_delalloc_inodes(root, delay_iput, nr);
9751 btrfs_put_fs_root(root);
9759 spin_lock(&fs_info->delalloc_root_lock);
9761 spin_unlock(&fs_info->delalloc_root_lock);
9764 atomic_inc(&fs_info->async_submit_draining);
9765 while (atomic_read(&fs_info->nr_async_submits) ||
9766 atomic_read(&fs_info->async_delalloc_pages)) {
9767 wait_event(fs_info->async_submit_wait,
9768 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9769 atomic_read(&fs_info->async_delalloc_pages) == 0));
9771 atomic_dec(&fs_info->async_submit_draining);
9773 if (!list_empty_careful(&splice)) {
9774 spin_lock(&fs_info->delalloc_root_lock);
9775 list_splice_tail(&splice, &fs_info->delalloc_roots);
9776 spin_unlock(&fs_info->delalloc_root_lock);
9778 mutex_unlock(&fs_info->delalloc_root_mutex);
9782 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9783 const char *symname)
9785 struct btrfs_trans_handle *trans;
9786 struct btrfs_root *root = BTRFS_I(dir)->root;
9787 struct btrfs_path *path;
9788 struct btrfs_key key;
9789 struct inode *inode = NULL;
9797 struct btrfs_file_extent_item *ei;
9798 struct extent_buffer *leaf;
9800 name_len = strlen(symname);
9801 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9802 return -ENAMETOOLONG;
9805 * 2 items for inode item and ref
9806 * 2 items for dir items
9807 * 1 item for updating parent inode item
9808 * 1 item for the inline extent item
9809 * 1 item for xattr if selinux is on
9811 trans = btrfs_start_transaction(root, 7);
9813 return PTR_ERR(trans);
9815 err = btrfs_find_free_ino(root, &objectid);
9819 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9820 dentry->d_name.len, btrfs_ino(dir), objectid,
9821 S_IFLNK|S_IRWXUGO, &index);
9822 if (IS_ERR(inode)) {
9823 err = PTR_ERR(inode);
9828 * If the active LSM wants to access the inode during
9829 * d_instantiate it needs these. Smack checks to see
9830 * if the filesystem supports xattrs by looking at the
9833 inode->i_fop = &btrfs_file_operations;
9834 inode->i_op = &btrfs_file_inode_operations;
9835 inode->i_mapping->a_ops = &btrfs_aops;
9836 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9838 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9840 goto out_unlock_inode;
9842 path = btrfs_alloc_path();
9845 goto out_unlock_inode;
9847 key.objectid = btrfs_ino(inode);
9849 key.type = BTRFS_EXTENT_DATA_KEY;
9850 datasize = btrfs_file_extent_calc_inline_size(name_len);
9851 err = btrfs_insert_empty_item(trans, root, path, &key,
9854 btrfs_free_path(path);
9855 goto out_unlock_inode;
9857 leaf = path->nodes[0];
9858 ei = btrfs_item_ptr(leaf, path->slots[0],
9859 struct btrfs_file_extent_item);
9860 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9861 btrfs_set_file_extent_type(leaf, ei,
9862 BTRFS_FILE_EXTENT_INLINE);
9863 btrfs_set_file_extent_encryption(leaf, ei, 0);
9864 btrfs_set_file_extent_compression(leaf, ei, 0);
9865 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9866 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9868 ptr = btrfs_file_extent_inline_start(ei);
9869 write_extent_buffer(leaf, symname, ptr, name_len);
9870 btrfs_mark_buffer_dirty(leaf);
9871 btrfs_free_path(path);
9873 inode->i_op = &btrfs_symlink_inode_operations;
9874 inode_nohighmem(inode);
9875 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9876 inode_set_bytes(inode, name_len);
9877 btrfs_i_size_write(inode, name_len);
9878 err = btrfs_update_inode(trans, root, inode);
9880 * Last step, add directory indexes for our symlink inode. This is the
9881 * last step to avoid extra cleanup of these indexes if an error happens
9885 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9888 goto out_unlock_inode;
9891 unlock_new_inode(inode);
9892 d_instantiate(dentry, inode);
9895 btrfs_end_transaction(trans, root);
9897 inode_dec_link_count(inode);
9900 btrfs_btree_balance_dirty(root);
9905 unlock_new_inode(inode);
9909 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9910 u64 start, u64 num_bytes, u64 min_size,
9911 loff_t actual_len, u64 *alloc_hint,
9912 struct btrfs_trans_handle *trans)
9914 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9915 struct extent_map *em;
9916 struct btrfs_root *root = BTRFS_I(inode)->root;
9917 struct btrfs_key ins;
9918 u64 cur_offset = start;
9921 u64 last_alloc = (u64)-1;
9923 bool own_trans = true;
9927 while (num_bytes > 0) {
9929 trans = btrfs_start_transaction(root, 3);
9930 if (IS_ERR(trans)) {
9931 ret = PTR_ERR(trans);
9936 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9937 cur_bytes = max(cur_bytes, min_size);
9939 * If we are severely fragmented we could end up with really
9940 * small allocations, so if the allocator is returning small
9941 * chunks lets make its job easier by only searching for those
9944 cur_bytes = min(cur_bytes, last_alloc);
9945 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9946 *alloc_hint, &ins, 1, 0);
9949 btrfs_end_transaction(trans, root);
9952 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
9954 last_alloc = ins.offset;
9955 ret = insert_reserved_file_extent(trans, inode,
9956 cur_offset, ins.objectid,
9957 ins.offset, ins.offset,
9958 ins.offset, 0, 0, 0,
9959 BTRFS_FILE_EXTENT_PREALLOC);
9961 btrfs_free_reserved_extent(root, ins.objectid,
9963 btrfs_abort_transaction(trans, root, ret);
9965 btrfs_end_transaction(trans, root);
9969 btrfs_drop_extent_cache(inode, cur_offset,
9970 cur_offset + ins.offset -1, 0);
9972 em = alloc_extent_map();
9974 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9975 &BTRFS_I(inode)->runtime_flags);
9979 em->start = cur_offset;
9980 em->orig_start = cur_offset;
9981 em->len = ins.offset;
9982 em->block_start = ins.objectid;
9983 em->block_len = ins.offset;
9984 em->orig_block_len = ins.offset;
9985 em->ram_bytes = ins.offset;
9986 em->bdev = root->fs_info->fs_devices->latest_bdev;
9987 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9988 em->generation = trans->transid;
9991 write_lock(&em_tree->lock);
9992 ret = add_extent_mapping(em_tree, em, 1);
9993 write_unlock(&em_tree->lock);
9996 btrfs_drop_extent_cache(inode, cur_offset,
9997 cur_offset + ins.offset - 1,
10000 free_extent_map(em);
10002 num_bytes -= ins.offset;
10003 cur_offset += ins.offset;
10004 *alloc_hint = ins.objectid + ins.offset;
10006 inode_inc_iversion(inode);
10007 inode->i_ctime = current_fs_time(inode->i_sb);
10008 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10009 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10010 (actual_len > inode->i_size) &&
10011 (cur_offset > inode->i_size)) {
10012 if (cur_offset > actual_len)
10013 i_size = actual_len;
10015 i_size = cur_offset;
10016 i_size_write(inode, i_size);
10017 btrfs_ordered_update_i_size(inode, i_size, NULL);
10020 ret = btrfs_update_inode(trans, root, inode);
10023 btrfs_abort_transaction(trans, root, ret);
10025 btrfs_end_transaction(trans, root);
10030 btrfs_end_transaction(trans, root);
10035 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10036 u64 start, u64 num_bytes, u64 min_size,
10037 loff_t actual_len, u64 *alloc_hint)
10039 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10040 min_size, actual_len, alloc_hint,
10044 int btrfs_prealloc_file_range_trans(struct inode *inode,
10045 struct btrfs_trans_handle *trans, int mode,
10046 u64 start, u64 num_bytes, u64 min_size,
10047 loff_t actual_len, u64 *alloc_hint)
10049 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10050 min_size, actual_len, alloc_hint, trans);
10053 static int btrfs_set_page_dirty(struct page *page)
10055 return __set_page_dirty_nobuffers(page);
10058 static int btrfs_permission(struct inode *inode, int mask)
10060 struct btrfs_root *root = BTRFS_I(inode)->root;
10061 umode_t mode = inode->i_mode;
10063 if (mask & MAY_WRITE &&
10064 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10065 if (btrfs_root_readonly(root))
10067 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10070 return generic_permission(inode, mask);
10073 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10075 struct btrfs_trans_handle *trans;
10076 struct btrfs_root *root = BTRFS_I(dir)->root;
10077 struct inode *inode = NULL;
10083 * 5 units required for adding orphan entry
10085 trans = btrfs_start_transaction(root, 5);
10087 return PTR_ERR(trans);
10089 ret = btrfs_find_free_ino(root, &objectid);
10093 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10094 btrfs_ino(dir), objectid, mode, &index);
10095 if (IS_ERR(inode)) {
10096 ret = PTR_ERR(inode);
10101 inode->i_fop = &btrfs_file_operations;
10102 inode->i_op = &btrfs_file_inode_operations;
10104 inode->i_mapping->a_ops = &btrfs_aops;
10105 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10107 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10111 ret = btrfs_update_inode(trans, root, inode);
10114 ret = btrfs_orphan_add(trans, inode);
10119 * We set number of links to 0 in btrfs_new_inode(), and here we set
10120 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10123 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10125 set_nlink(inode, 1);
10126 unlock_new_inode(inode);
10127 d_tmpfile(dentry, inode);
10128 mark_inode_dirty(inode);
10131 btrfs_end_transaction(trans, root);
10134 btrfs_balance_delayed_items(root);
10135 btrfs_btree_balance_dirty(root);
10139 unlock_new_inode(inode);
10144 /* Inspired by filemap_check_errors() */
10145 int btrfs_inode_check_errors(struct inode *inode)
10149 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10150 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10152 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10153 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10159 static const struct inode_operations btrfs_dir_inode_operations = {
10160 .getattr = btrfs_getattr,
10161 .lookup = btrfs_lookup,
10162 .create = btrfs_create,
10163 .unlink = btrfs_unlink,
10164 .link = btrfs_link,
10165 .mkdir = btrfs_mkdir,
10166 .rmdir = btrfs_rmdir,
10167 .rename2 = btrfs_rename2,
10168 .symlink = btrfs_symlink,
10169 .setattr = btrfs_setattr,
10170 .mknod = btrfs_mknod,
10171 .setxattr = btrfs_setxattr,
10172 .getxattr = generic_getxattr,
10173 .listxattr = btrfs_listxattr,
10174 .removexattr = btrfs_removexattr,
10175 .permission = btrfs_permission,
10176 .get_acl = btrfs_get_acl,
10177 .set_acl = btrfs_set_acl,
10178 .update_time = btrfs_update_time,
10179 .tmpfile = btrfs_tmpfile,
10181 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10182 .lookup = btrfs_lookup,
10183 .permission = btrfs_permission,
10184 .get_acl = btrfs_get_acl,
10185 .set_acl = btrfs_set_acl,
10186 .update_time = btrfs_update_time,
10189 static const struct file_operations btrfs_dir_file_operations = {
10190 .llseek = generic_file_llseek,
10191 .read = generic_read_dir,
10192 .iterate = btrfs_real_readdir,
10193 .unlocked_ioctl = btrfs_ioctl,
10194 #ifdef CONFIG_COMPAT
10195 .compat_ioctl = btrfs_ioctl,
10197 .release = btrfs_release_file,
10198 .fsync = btrfs_sync_file,
10201 static const struct extent_io_ops btrfs_extent_io_ops = {
10202 .fill_delalloc = run_delalloc_range,
10203 .submit_bio_hook = btrfs_submit_bio_hook,
10204 .merge_bio_hook = btrfs_merge_bio_hook,
10205 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10206 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10207 .writepage_start_hook = btrfs_writepage_start_hook,
10208 .set_bit_hook = btrfs_set_bit_hook,
10209 .clear_bit_hook = btrfs_clear_bit_hook,
10210 .merge_extent_hook = btrfs_merge_extent_hook,
10211 .split_extent_hook = btrfs_split_extent_hook,
10215 * btrfs doesn't support the bmap operation because swapfiles
10216 * use bmap to make a mapping of extents in the file. They assume
10217 * these extents won't change over the life of the file and they
10218 * use the bmap result to do IO directly to the drive.
10220 * the btrfs bmap call would return logical addresses that aren't
10221 * suitable for IO and they also will change frequently as COW
10222 * operations happen. So, swapfile + btrfs == corruption.
10224 * For now we're avoiding this by dropping bmap.
10226 static const struct address_space_operations btrfs_aops = {
10227 .readpage = btrfs_readpage,
10228 .writepage = btrfs_writepage,
10229 .writepages = btrfs_writepages,
10230 .readpages = btrfs_readpages,
10231 .direct_IO = btrfs_direct_IO,
10232 .invalidatepage = btrfs_invalidatepage,
10233 .releasepage = btrfs_releasepage,
10234 .set_page_dirty = btrfs_set_page_dirty,
10235 .error_remove_page = generic_error_remove_page,
10238 static const struct address_space_operations btrfs_symlink_aops = {
10239 .readpage = btrfs_readpage,
10240 .writepage = btrfs_writepage,
10241 .invalidatepage = btrfs_invalidatepage,
10242 .releasepage = btrfs_releasepage,
10245 static const struct inode_operations btrfs_file_inode_operations = {
10246 .getattr = btrfs_getattr,
10247 .setattr = btrfs_setattr,
10248 .setxattr = btrfs_setxattr,
10249 .getxattr = generic_getxattr,
10250 .listxattr = btrfs_listxattr,
10251 .removexattr = btrfs_removexattr,
10252 .permission = btrfs_permission,
10253 .fiemap = btrfs_fiemap,
10254 .get_acl = btrfs_get_acl,
10255 .set_acl = btrfs_set_acl,
10256 .update_time = btrfs_update_time,
10258 static const struct inode_operations btrfs_special_inode_operations = {
10259 .getattr = btrfs_getattr,
10260 .setattr = btrfs_setattr,
10261 .permission = btrfs_permission,
10262 .setxattr = btrfs_setxattr,
10263 .getxattr = generic_getxattr,
10264 .listxattr = btrfs_listxattr,
10265 .removexattr = btrfs_removexattr,
10266 .get_acl = btrfs_get_acl,
10267 .set_acl = btrfs_set_acl,
10268 .update_time = btrfs_update_time,
10270 static const struct inode_operations btrfs_symlink_inode_operations = {
10271 .readlink = generic_readlink,
10272 .get_link = page_get_link,
10273 .getattr = btrfs_getattr,
10274 .setattr = btrfs_setattr,
10275 .permission = btrfs_permission,
10276 .setxattr = btrfs_setxattr,
10277 .getxattr = generic_getxattr,
10278 .listxattr = btrfs_listxattr,
10279 .removexattr = btrfs_removexattr,
10280 .update_time = btrfs_update_time,
10283 const struct dentry_operations btrfs_dentry_operations = {
10284 .d_delete = btrfs_dentry_delete,
10285 .d_release = btrfs_dentry_release,