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/compat.h>
34 #include <linux/bit_spinlock.h>
35 #include <linux/xattr.h>
36 #include <linux/posix_acl.h>
37 #include <linux/falloc.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
40 #include <linux/mount.h>
41 #include <linux/btrfs.h>
42 #include <linux/blkdev.h>
43 #include <linux/posix_acl_xattr.h>
44 #include <linux/uio.h>
47 #include "transaction.h"
48 #include "btrfs_inode.h"
49 #include "print-tree.h"
50 #include "ordered-data.h"
54 #include "compression.h"
56 #include "free-space-cache.h"
57 #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;
77 static const struct inode_operations btrfs_dir_inode_operations;
78 static const struct inode_operations btrfs_symlink_inode_operations;
79 static const struct inode_operations btrfs_dir_ro_inode_operations;
80 static const struct inode_operations btrfs_special_inode_operations;
81 static const struct inode_operations btrfs_file_inode_operations;
82 static const struct address_space_operations btrfs_aops;
83 static const struct address_space_operations btrfs_symlink_aops;
84 static const struct file_operations btrfs_dir_file_operations;
85 static const struct extent_io_ops btrfs_extent_io_ops;
87 static struct kmem_cache *btrfs_inode_cachep;
88 struct kmem_cache *btrfs_trans_handle_cachep;
89 struct kmem_cache *btrfs_transaction_cachep;
90 struct kmem_cache *btrfs_path_cachep;
91 struct kmem_cache *btrfs_free_space_cachep;
94 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
95 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
96 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
97 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
98 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
99 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
100 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
101 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
104 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
105 static int btrfs_truncate(struct inode *inode);
106 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
107 static noinline int cow_file_range(struct inode *inode,
108 struct page *locked_page,
109 u64 start, u64 end, u64 delalloc_end,
110 int *page_started, unsigned long *nr_written,
111 int unlock, struct btrfs_dedupe_hash *hash);
112 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
113 u64 orig_start, u64 block_start,
114 u64 block_len, u64 orig_block_len,
115 u64 ram_bytes, int compress_type,
118 static int btrfs_dirty_inode(struct inode *inode);
120 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
121 void btrfs_test_inode_set_ops(struct inode *inode)
123 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
127 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
128 struct inode *inode, struct inode *dir,
129 const struct qstr *qstr)
133 err = btrfs_init_acl(trans, inode, dir);
135 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
140 * this does all the hard work for inserting an inline extent into
141 * the btree. The caller should have done a btrfs_drop_extents so that
142 * no overlapping inline items exist in the btree
144 static int insert_inline_extent(struct btrfs_trans_handle *trans,
145 struct btrfs_path *path, int extent_inserted,
146 struct btrfs_root *root, struct inode *inode,
147 u64 start, size_t size, size_t compressed_size,
149 struct page **compressed_pages)
151 struct extent_buffer *leaf;
152 struct page *page = NULL;
155 struct btrfs_file_extent_item *ei;
158 size_t cur_size = size;
159 unsigned long offset;
161 if (compressed_size && compressed_pages)
162 cur_size = compressed_size;
164 inode_add_bytes(inode, size);
166 if (!extent_inserted) {
167 struct btrfs_key key;
170 key.objectid = btrfs_ino(BTRFS_I(inode));
172 key.type = BTRFS_EXTENT_DATA_KEY;
174 datasize = btrfs_file_extent_calc_inline_size(cur_size);
175 path->leave_spinning = 1;
176 ret = btrfs_insert_empty_item(trans, root, path, &key,
183 leaf = path->nodes[0];
184 ei = btrfs_item_ptr(leaf, path->slots[0],
185 struct btrfs_file_extent_item);
186 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
187 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
188 btrfs_set_file_extent_encryption(leaf, ei, 0);
189 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
190 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
191 ptr = btrfs_file_extent_inline_start(ei);
193 if (compress_type != BTRFS_COMPRESS_NONE) {
196 while (compressed_size > 0) {
197 cpage = compressed_pages[i];
198 cur_size = min_t(unsigned long, compressed_size,
201 kaddr = kmap_atomic(cpage);
202 write_extent_buffer(leaf, kaddr, ptr, cur_size);
203 kunmap_atomic(kaddr);
207 compressed_size -= cur_size;
209 btrfs_set_file_extent_compression(leaf, ei,
212 page = find_get_page(inode->i_mapping,
213 start >> PAGE_SHIFT);
214 btrfs_set_file_extent_compression(leaf, ei, 0);
215 kaddr = kmap_atomic(page);
216 offset = start & (PAGE_SIZE - 1);
217 write_extent_buffer(leaf, kaddr + offset, ptr, size);
218 kunmap_atomic(kaddr);
221 btrfs_mark_buffer_dirty(leaf);
222 btrfs_release_path(path);
225 * we're an inline extent, so nobody can
226 * extend the file past i_size without locking
227 * a page we already have locked.
229 * We must do any isize and inode updates
230 * before we unlock the pages. Otherwise we
231 * could end up racing with unlink.
233 BTRFS_I(inode)->disk_i_size = inode->i_size;
234 ret = btrfs_update_inode(trans, root, inode);
243 * conditionally insert an inline extent into the file. This
244 * does the checks required to make sure the data is small enough
245 * to fit as an inline extent.
247 static noinline int cow_file_range_inline(struct btrfs_root *root,
248 struct inode *inode, u64 start,
249 u64 end, size_t compressed_size,
251 struct page **compressed_pages)
253 struct btrfs_fs_info *fs_info = root->fs_info;
254 struct btrfs_trans_handle *trans;
255 u64 isize = i_size_read(inode);
256 u64 actual_end = min(end + 1, isize);
257 u64 inline_len = actual_end - start;
258 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
259 u64 data_len = inline_len;
261 struct btrfs_path *path;
262 int extent_inserted = 0;
263 u32 extent_item_size;
266 data_len = compressed_size;
269 actual_end > fs_info->sectorsize ||
270 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
272 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
274 data_len > fs_info->max_inline) {
278 path = btrfs_alloc_path();
282 trans = btrfs_join_transaction(root);
284 btrfs_free_path(path);
285 return PTR_ERR(trans);
287 trans->block_rsv = &fs_info->delalloc_block_rsv;
289 if (compressed_size && compressed_pages)
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 extent_item_size = btrfs_file_extent_calc_inline_size(
296 ret = __btrfs_drop_extents(trans, root, inode, path,
297 start, aligned_end, NULL,
298 1, 1, extent_item_size, &extent_inserted);
300 btrfs_abort_transaction(trans, ret);
304 if (isize > actual_end)
305 inline_len = min_t(u64, isize, actual_end);
306 ret = insert_inline_extent(trans, path, extent_inserted,
308 inline_len, compressed_size,
309 compress_type, compressed_pages);
310 if (ret && ret != -ENOSPC) {
311 btrfs_abort_transaction(trans, ret);
313 } else if (ret == -ENOSPC) {
318 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
319 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
320 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
323 * Don't forget to free the reserved space, as for inlined extent
324 * it won't count as data extent, free them directly here.
325 * And at reserve time, it's always aligned to page size, so
326 * just free one page here.
328 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
329 btrfs_free_path(path);
330 btrfs_end_transaction(trans);
334 struct async_extent {
339 unsigned long nr_pages;
341 struct list_head list;
346 struct btrfs_root *root;
347 struct page *locked_page;
350 struct list_head extents;
351 struct btrfs_work work;
354 static noinline int add_async_extent(struct async_cow *cow,
355 u64 start, u64 ram_size,
358 unsigned long nr_pages,
361 struct async_extent *async_extent;
363 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
364 BUG_ON(!async_extent); /* -ENOMEM */
365 async_extent->start = start;
366 async_extent->ram_size = ram_size;
367 async_extent->compressed_size = compressed_size;
368 async_extent->pages = pages;
369 async_extent->nr_pages = nr_pages;
370 async_extent->compress_type = compress_type;
371 list_add_tail(&async_extent->list, &cow->extents);
375 static inline int inode_need_compress(struct inode *inode)
377 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
380 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
382 /* bad compression ratios */
383 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
385 if (btrfs_test_opt(fs_info, COMPRESS) ||
386 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
387 BTRFS_I(inode)->force_compress)
392 static inline void inode_should_defrag(struct btrfs_inode *inode,
393 u64 start, u64 end, u64 num_bytes, u64 small_write)
395 /* If this is a small write inside eof, kick off a defrag */
396 if (num_bytes < small_write &&
397 (start > 0 || end + 1 < inode->disk_i_size))
398 btrfs_add_inode_defrag(NULL, inode);
402 * we create compressed extents in two phases. The first
403 * phase compresses a range of pages that have already been
404 * locked (both pages and state bits are locked).
406 * This is done inside an ordered work queue, and the compression
407 * is spread across many cpus. The actual IO submission is step
408 * two, and the ordered work queue takes care of making sure that
409 * happens in the same order things were put onto the queue by
410 * writepages and friends.
412 * If this code finds it can't get good compression, it puts an
413 * entry onto the work queue to write the uncompressed bytes. This
414 * makes sure that both compressed inodes and uncompressed inodes
415 * are written in the same order that the flusher thread sent them
418 static noinline void compress_file_range(struct inode *inode,
419 struct page *locked_page,
421 struct async_cow *async_cow,
424 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
425 struct btrfs_root *root = BTRFS_I(inode)->root;
427 u64 blocksize = fs_info->sectorsize;
429 u64 isize = i_size_read(inode);
431 struct page **pages = NULL;
432 unsigned long nr_pages;
433 unsigned long total_compressed = 0;
434 unsigned long total_in = 0;
437 int compress_type = fs_info->compress_type;
440 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
443 actual_end = min_t(u64, isize, end + 1);
446 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
447 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
448 nr_pages = min_t(unsigned long, nr_pages,
449 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
452 * we don't want to send crud past the end of i_size through
453 * compression, that's just a waste of CPU time. So, if the
454 * end of the file is before the start of our current
455 * requested range of bytes, we bail out to the uncompressed
456 * cleanup code that can deal with all of this.
458 * It isn't really the fastest way to fix things, but this is a
459 * very uncommon corner.
461 if (actual_end <= start)
462 goto cleanup_and_bail_uncompressed;
464 total_compressed = actual_end - start;
467 * skip compression for a small file range(<=blocksize) that
468 * isn't an inline extent, since it doesn't save disk space at all.
470 if (total_compressed <= blocksize &&
471 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
472 goto cleanup_and_bail_uncompressed;
474 total_compressed = min_t(unsigned long, total_compressed,
475 BTRFS_MAX_UNCOMPRESSED);
476 num_bytes = ALIGN(end - start + 1, blocksize);
477 num_bytes = max(blocksize, num_bytes);
482 * we do compression for mount -o compress and when the
483 * inode has not been flagged as nocompress. This flag can
484 * change at any time if we discover bad compression ratios.
486 if (inode_need_compress(inode)) {
488 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
490 /* just bail out to the uncompressed code */
494 if (BTRFS_I(inode)->force_compress)
495 compress_type = BTRFS_I(inode)->force_compress;
498 * we need to call clear_page_dirty_for_io on each
499 * page in the range. Otherwise applications with the file
500 * mmap'd can wander in and change the page contents while
501 * we are compressing them.
503 * If the compression fails for any reason, we set the pages
504 * dirty again later on.
506 extent_range_clear_dirty_for_io(inode, start, end);
508 ret = btrfs_compress_pages(compress_type,
509 inode->i_mapping, start,
516 unsigned long offset = total_compressed &
518 struct page *page = pages[nr_pages - 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,
541 0, BTRFS_COMPRESS_NONE, NULL);
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, end,
568 btrfs_free_reserved_data_space_noquota(inode, start,
576 * we aren't doing an inline extent round the compressed size
577 * up to a block size boundary so the allocator does sane
580 total_compressed = ALIGN(total_compressed, blocksize);
583 * one last check to make sure the compression is really a
584 * win, compare the page count read with the blocks on disk
586 total_in = ALIGN(total_in, PAGE_SIZE);
587 if (total_compressed >= total_in) {
590 num_bytes = total_in;
594 * The async work queues will take care of doing actual
595 * allocation on disk for these compressed pages, and
596 * will submit them to the elevator.
598 add_async_extent(async_cow, start, num_bytes,
599 total_compressed, pages, nr_pages,
602 if (start + num_bytes < end) {
613 * the compression code ran but failed to make things smaller,
614 * free any pages it allocated and our page pointer array
616 for (i = 0; i < nr_pages; i++) {
617 WARN_ON(pages[i]->mapping);
622 total_compressed = 0;
625 /* flag the file so we don't compress in the future */
626 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
627 !(BTRFS_I(inode)->force_compress)) {
628 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
631 cleanup_and_bail_uncompressed:
633 * No compression, but we still need to write the pages in the file
634 * we've been given so far. redirty the locked page if it corresponds
635 * to our extent and set things up for the async work queue to run
636 * cow_file_range to do the normal delalloc dance.
638 if (page_offset(locked_page) >= start &&
639 page_offset(locked_page) <= end)
640 __set_page_dirty_nobuffers(locked_page);
641 /* unlocked later on in the async handlers */
644 extent_range_redirty_for_io(inode, start, end);
645 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
646 BTRFS_COMPRESS_NONE);
652 for (i = 0; i < nr_pages; i++) {
653 WARN_ON(pages[i]->mapping);
659 static void free_async_extent_pages(struct async_extent *async_extent)
663 if (!async_extent->pages)
666 for (i = 0; i < async_extent->nr_pages; i++) {
667 WARN_ON(async_extent->pages[i]->mapping);
668 put_page(async_extent->pages[i]);
670 kfree(async_extent->pages);
671 async_extent->nr_pages = 0;
672 async_extent->pages = NULL;
676 * phase two of compressed writeback. This is the ordered portion
677 * of the code, which only gets called in the order the work was
678 * queued. We walk all the async extents created by compress_file_range
679 * and send them down to the disk.
681 static noinline void submit_compressed_extents(struct inode *inode,
682 struct async_cow *async_cow)
684 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
685 struct async_extent *async_extent;
687 struct btrfs_key ins;
688 struct extent_map *em;
689 struct btrfs_root *root = BTRFS_I(inode)->root;
690 struct extent_io_tree *io_tree;
694 while (!list_empty(&async_cow->extents)) {
695 async_extent = list_entry(async_cow->extents.next,
696 struct async_extent, list);
697 list_del(&async_extent->list);
699 io_tree = &BTRFS_I(inode)->io_tree;
702 /* did the compression code fall back to uncompressed IO? */
703 if (!async_extent->pages) {
704 int page_started = 0;
705 unsigned long nr_written = 0;
707 lock_extent(io_tree, async_extent->start,
708 async_extent->start +
709 async_extent->ram_size - 1);
711 /* allocate blocks */
712 ret = cow_file_range(inode, async_cow->locked_page,
714 async_extent->start +
715 async_extent->ram_size - 1,
716 async_extent->start +
717 async_extent->ram_size - 1,
718 &page_started, &nr_written, 0,
724 * if page_started, cow_file_range inserted an
725 * inline extent and took care of all the unlocking
726 * and IO for us. Otherwise, we need to submit
727 * all those pages down to the drive.
729 if (!page_started && !ret)
730 extent_write_locked_range(io_tree,
731 inode, async_extent->start,
732 async_extent->start +
733 async_extent->ram_size - 1,
737 unlock_page(async_cow->locked_page);
743 lock_extent(io_tree, async_extent->start,
744 async_extent->start + async_extent->ram_size - 1);
746 ret = btrfs_reserve_extent(root, async_extent->ram_size,
747 async_extent->compressed_size,
748 async_extent->compressed_size,
749 0, alloc_hint, &ins, 1, 1);
751 free_async_extent_pages(async_extent);
753 if (ret == -ENOSPC) {
754 unlock_extent(io_tree, async_extent->start,
755 async_extent->start +
756 async_extent->ram_size - 1);
759 * we need to redirty the pages if we decide to
760 * fallback to uncompressed IO, otherwise we
761 * will not submit these pages down to lower
764 extent_range_redirty_for_io(inode,
766 async_extent->start +
767 async_extent->ram_size - 1);
774 * here we're doing allocation and writeback of the
777 em = create_io_em(inode, async_extent->start,
778 async_extent->ram_size, /* len */
779 async_extent->start, /* orig_start */
780 ins.objectid, /* block_start */
781 ins.offset, /* block_len */
782 ins.offset, /* orig_block_len */
783 async_extent->ram_size, /* ram_bytes */
784 async_extent->compress_type,
785 BTRFS_ORDERED_COMPRESSED);
787 /* ret value is not necessary due to void function */
788 goto out_free_reserve;
791 ret = btrfs_add_ordered_extent_compress(inode,
794 async_extent->ram_size,
796 BTRFS_ORDERED_COMPRESSED,
797 async_extent->compress_type);
799 btrfs_drop_extent_cache(BTRFS_I(inode),
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
803 goto out_free_reserve;
805 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
808 * clear dirty, set writeback and unlock the pages.
810 extent_clear_unlock_delalloc(inode, async_extent->start,
811 async_extent->start +
812 async_extent->ram_size - 1,
813 async_extent->start +
814 async_extent->ram_size - 1,
815 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
816 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
818 ret = btrfs_submit_compressed_write(inode,
820 async_extent->ram_size,
822 ins.offset, async_extent->pages,
823 async_extent->nr_pages);
825 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
826 struct page *p = async_extent->pages[0];
827 const u64 start = async_extent->start;
828 const u64 end = start + async_extent->ram_size - 1;
830 p->mapping = inode->i_mapping;
831 tree->ops->writepage_end_io_hook(p, start, end,
834 extent_clear_unlock_delalloc(inode, start, end, end,
838 free_async_extent_pages(async_extent);
840 alloc_hint = ins.objectid + ins.offset;
846 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
847 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
849 extent_clear_unlock_delalloc(inode, async_extent->start,
850 async_extent->start +
851 async_extent->ram_size - 1,
852 async_extent->start +
853 async_extent->ram_size - 1,
854 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
855 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
856 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
857 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
859 free_async_extent_pages(async_extent);
864 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
867 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
868 struct extent_map *em;
871 read_lock(&em_tree->lock);
872 em = search_extent_mapping(em_tree, start, num_bytes);
875 * if block start isn't an actual block number then find the
876 * first block in this inode and use that as a hint. If that
877 * block is also bogus then just don't worry about it.
879 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
881 em = search_extent_mapping(em_tree, 0, 0);
882 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
883 alloc_hint = em->block_start;
887 alloc_hint = em->block_start;
891 read_unlock(&em_tree->lock);
897 * when extent_io.c finds a delayed allocation range in the file,
898 * the call backs end up in this code. The basic idea is to
899 * allocate extents on disk for the range, and create ordered data structs
900 * in ram to track those extents.
902 * locked_page is the page that writepage had locked already. We use
903 * it to make sure we don't do extra locks or unlocks.
905 * *page_started is set to one if we unlock locked_page and do everything
906 * required to start IO on it. It may be clean and already done with
909 static noinline int cow_file_range(struct inode *inode,
910 struct page *locked_page,
911 u64 start, u64 end, u64 delalloc_end,
912 int *page_started, unsigned long *nr_written,
913 int unlock, struct btrfs_dedupe_hash *hash)
915 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
916 struct btrfs_root *root = BTRFS_I(inode)->root;
919 unsigned long ram_size;
922 u64 blocksize = fs_info->sectorsize;
923 struct btrfs_key ins;
924 struct extent_map *em;
927 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
933 num_bytes = ALIGN(end - start + 1, blocksize);
934 num_bytes = max(blocksize, num_bytes);
935 disk_num_bytes = num_bytes;
937 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
940 /* lets try to make an inline extent */
941 ret = cow_file_range_inline(root, inode, start, end, 0,
942 BTRFS_COMPRESS_NONE, NULL);
944 extent_clear_unlock_delalloc(inode, start, end,
946 EXTENT_LOCKED | EXTENT_DELALLOC |
947 EXTENT_DEFRAG, PAGE_UNLOCK |
948 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
950 btrfs_free_reserved_data_space_noquota(inode, start,
952 *nr_written = *nr_written +
953 (end - start + PAGE_SIZE) / PAGE_SIZE;
956 } else if (ret < 0) {
961 BUG_ON(disk_num_bytes >
962 btrfs_super_total_bytes(fs_info->super_copy));
964 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
965 btrfs_drop_extent_cache(BTRFS_I(inode), start,
966 start + num_bytes - 1, 0);
968 while (disk_num_bytes > 0) {
971 cur_alloc_size = disk_num_bytes;
972 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
973 fs_info->sectorsize, 0, alloc_hint,
978 ram_size = ins.offset;
979 em = create_io_em(inode, start, ins.offset, /* len */
980 start, /* orig_start */
981 ins.objectid, /* block_start */
982 ins.offset, /* block_len */
983 ins.offset, /* orig_block_len */
984 ram_size, /* ram_bytes */
985 BTRFS_COMPRESS_NONE, /* compress_type */
986 BTRFS_ORDERED_REGULAR /* type */);
991 cur_alloc_size = ins.offset;
992 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
993 ram_size, cur_alloc_size, 0);
995 goto out_drop_extent_cache;
997 if (root->root_key.objectid ==
998 BTRFS_DATA_RELOC_TREE_OBJECTID) {
999 ret = btrfs_reloc_clone_csums(inode, start,
1002 goto out_drop_extent_cache;
1005 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1007 if (disk_num_bytes < cur_alloc_size)
1010 /* we're not doing compressed IO, don't unlock the first
1011 * page (which the caller expects to stay locked), don't
1012 * clear any dirty bits and don't set any writeback bits
1014 * Do set the Private2 bit so we know this page was properly
1015 * setup for writepage
1017 op = unlock ? PAGE_UNLOCK : 0;
1018 op |= PAGE_SET_PRIVATE2;
1020 extent_clear_unlock_delalloc(inode, start,
1021 start + ram_size - 1,
1022 delalloc_end, locked_page,
1023 EXTENT_LOCKED | EXTENT_DELALLOC,
1025 disk_num_bytes -= cur_alloc_size;
1026 num_bytes -= cur_alloc_size;
1027 alloc_hint = ins.objectid + ins.offset;
1028 start += cur_alloc_size;
1033 out_drop_extent_cache:
1034 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1036 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1037 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1039 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1041 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1042 EXTENT_DELALLOC | EXTENT_DEFRAG,
1043 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1044 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1049 * work queue call back to started compression on a file and pages
1051 static noinline void async_cow_start(struct btrfs_work *work)
1053 struct async_cow *async_cow;
1055 async_cow = container_of(work, struct async_cow, work);
1057 compress_file_range(async_cow->inode, async_cow->locked_page,
1058 async_cow->start, async_cow->end, async_cow,
1060 if (num_added == 0) {
1061 btrfs_add_delayed_iput(async_cow->inode);
1062 async_cow->inode = NULL;
1067 * work queue call back to submit previously compressed pages
1069 static noinline void async_cow_submit(struct btrfs_work *work)
1071 struct btrfs_fs_info *fs_info;
1072 struct async_cow *async_cow;
1073 struct btrfs_root *root;
1074 unsigned long nr_pages;
1076 async_cow = container_of(work, struct async_cow, work);
1078 root = async_cow->root;
1079 fs_info = root->fs_info;
1080 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1084 * atomic_sub_return implies a barrier for waitqueue_active
1086 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1088 waitqueue_active(&fs_info->async_submit_wait))
1089 wake_up(&fs_info->async_submit_wait);
1091 if (async_cow->inode)
1092 submit_compressed_extents(async_cow->inode, async_cow);
1095 static noinline void async_cow_free(struct btrfs_work *work)
1097 struct async_cow *async_cow;
1098 async_cow = container_of(work, struct async_cow, work);
1099 if (async_cow->inode)
1100 btrfs_add_delayed_iput(async_cow->inode);
1104 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1105 u64 start, u64 end, int *page_started,
1106 unsigned long *nr_written)
1108 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1109 struct async_cow *async_cow;
1110 struct btrfs_root *root = BTRFS_I(inode)->root;
1111 unsigned long nr_pages;
1114 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1115 1, 0, NULL, GFP_NOFS);
1116 while (start < end) {
1117 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1118 BUG_ON(!async_cow); /* -ENOMEM */
1119 async_cow->inode = igrab(inode);
1120 async_cow->root = root;
1121 async_cow->locked_page = locked_page;
1122 async_cow->start = start;
1124 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1125 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1128 cur_end = min(end, start + SZ_512K - 1);
1130 async_cow->end = cur_end;
1131 INIT_LIST_HEAD(&async_cow->extents);
1133 btrfs_init_work(&async_cow->work,
1134 btrfs_delalloc_helper,
1135 async_cow_start, async_cow_submit,
1138 nr_pages = (cur_end - start + PAGE_SIZE) >>
1140 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1142 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1144 while (atomic_read(&fs_info->async_submit_draining) &&
1145 atomic_read(&fs_info->async_delalloc_pages)) {
1146 wait_event(fs_info->async_submit_wait,
1147 (atomic_read(&fs_info->async_delalloc_pages) ==
1151 *nr_written += nr_pages;
1152 start = cur_end + 1;
1158 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1159 u64 bytenr, u64 num_bytes)
1162 struct btrfs_ordered_sum *sums;
1165 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1166 bytenr + num_bytes - 1, &list, 0);
1167 if (ret == 0 && list_empty(&list))
1170 while (!list_empty(&list)) {
1171 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1172 list_del(&sums->list);
1179 * when nowcow writeback call back. This checks for snapshots or COW copies
1180 * of the extents that exist in the file, and COWs the file as required.
1182 * If no cow copies or snapshots exist, we write directly to the existing
1185 static noinline int run_delalloc_nocow(struct inode *inode,
1186 struct page *locked_page,
1187 u64 start, u64 end, int *page_started, int force,
1188 unsigned long *nr_written)
1190 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1191 struct btrfs_root *root = BTRFS_I(inode)->root;
1192 struct extent_buffer *leaf;
1193 struct btrfs_path *path;
1194 struct btrfs_file_extent_item *fi;
1195 struct btrfs_key found_key;
1196 struct extent_map *em;
1211 u64 ino = btrfs_ino(BTRFS_I(inode));
1213 path = btrfs_alloc_path();
1215 extent_clear_unlock_delalloc(inode, start, end, end,
1217 EXTENT_LOCKED | EXTENT_DELALLOC |
1218 EXTENT_DO_ACCOUNTING |
1219 EXTENT_DEFRAG, PAGE_UNLOCK |
1221 PAGE_SET_WRITEBACK |
1222 PAGE_END_WRITEBACK);
1226 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1228 cow_start = (u64)-1;
1231 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1235 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1236 leaf = path->nodes[0];
1237 btrfs_item_key_to_cpu(leaf, &found_key,
1238 path->slots[0] - 1);
1239 if (found_key.objectid == ino &&
1240 found_key.type == BTRFS_EXTENT_DATA_KEY)
1245 leaf = path->nodes[0];
1246 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1247 ret = btrfs_next_leaf(root, path);
1252 leaf = path->nodes[0];
1258 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1260 if (found_key.objectid > ino)
1262 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1263 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1267 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1268 found_key.offset > end)
1271 if (found_key.offset > cur_offset) {
1272 extent_end = found_key.offset;
1277 fi = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_file_extent_item);
1279 extent_type = btrfs_file_extent_type(leaf, fi);
1281 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1282 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1283 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1284 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1285 extent_offset = btrfs_file_extent_offset(leaf, fi);
1286 extent_end = found_key.offset +
1287 btrfs_file_extent_num_bytes(leaf, fi);
1289 btrfs_file_extent_disk_num_bytes(leaf, fi);
1290 if (extent_end <= start) {
1294 if (disk_bytenr == 0)
1296 if (btrfs_file_extent_compression(leaf, fi) ||
1297 btrfs_file_extent_encryption(leaf, fi) ||
1298 btrfs_file_extent_other_encoding(leaf, fi))
1300 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1302 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1304 if (btrfs_cross_ref_exist(root, ino,
1306 extent_offset, disk_bytenr))
1308 disk_bytenr += extent_offset;
1309 disk_bytenr += cur_offset - found_key.offset;
1310 num_bytes = min(end + 1, extent_end) - cur_offset;
1312 * if there are pending snapshots for this root,
1313 * we fall into common COW way.
1316 err = btrfs_start_write_no_snapshoting(root);
1321 * force cow if csum exists in the range.
1322 * this ensure that csum for a given extent are
1323 * either valid or do not exist.
1325 if (csum_exist_in_range(fs_info, disk_bytenr,
1328 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1331 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1332 extent_end = found_key.offset +
1333 btrfs_file_extent_inline_len(leaf,
1334 path->slots[0], fi);
1335 extent_end = ALIGN(extent_end,
1336 fs_info->sectorsize);
1341 if (extent_end <= start) {
1343 if (!nolock && nocow)
1344 btrfs_end_write_no_snapshoting(root);
1346 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1350 if (cow_start == (u64)-1)
1351 cow_start = cur_offset;
1352 cur_offset = extent_end;
1353 if (cur_offset > end)
1359 btrfs_release_path(path);
1360 if (cow_start != (u64)-1) {
1361 ret = cow_file_range(inode, locked_page,
1362 cow_start, found_key.offset - 1,
1363 end, page_started, nr_written, 1,
1366 if (!nolock && nocow)
1367 btrfs_end_write_no_snapshoting(root);
1369 btrfs_dec_nocow_writers(fs_info,
1373 cow_start = (u64)-1;
1376 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1377 u64 orig_start = found_key.offset - extent_offset;
1379 em = create_io_em(inode, cur_offset, num_bytes,
1381 disk_bytenr, /* block_start */
1382 num_bytes, /* block_len */
1383 disk_num_bytes, /* orig_block_len */
1384 ram_bytes, BTRFS_COMPRESS_NONE,
1385 BTRFS_ORDERED_PREALLOC);
1387 if (!nolock && nocow)
1388 btrfs_end_write_no_snapshoting(root);
1390 btrfs_dec_nocow_writers(fs_info,
1395 free_extent_map(em);
1398 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1399 type = BTRFS_ORDERED_PREALLOC;
1401 type = BTRFS_ORDERED_NOCOW;
1404 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1405 num_bytes, num_bytes, type);
1407 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1408 BUG_ON(ret); /* -ENOMEM */
1410 if (root->root_key.objectid ==
1411 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1412 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1415 if (!nolock && nocow)
1416 btrfs_end_write_no_snapshoting(root);
1421 extent_clear_unlock_delalloc(inode, cur_offset,
1422 cur_offset + num_bytes - 1, end,
1423 locked_page, EXTENT_LOCKED |
1425 EXTENT_CLEAR_DATA_RESV,
1426 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1428 if (!nolock && nocow)
1429 btrfs_end_write_no_snapshoting(root);
1430 cur_offset = extent_end;
1431 if (cur_offset > end)
1434 btrfs_release_path(path);
1436 if (cur_offset <= end && cow_start == (u64)-1) {
1437 cow_start = cur_offset;
1441 if (cow_start != (u64)-1) {
1442 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1443 page_started, nr_written, 1, NULL);
1449 if (ret && cur_offset < end)
1450 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1451 locked_page, EXTENT_LOCKED |
1452 EXTENT_DELALLOC | EXTENT_DEFRAG |
1453 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1455 PAGE_SET_WRITEBACK |
1456 PAGE_END_WRITEBACK);
1457 btrfs_free_path(path);
1461 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1464 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1465 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1469 * @defrag_bytes is a hint value, no spinlock held here,
1470 * if is not zero, it means the file is defragging.
1471 * Force cow if given extent needs to be defragged.
1473 if (BTRFS_I(inode)->defrag_bytes &&
1474 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1475 EXTENT_DEFRAG, 0, NULL))
1482 * extent_io.c call back to do delayed allocation processing
1484 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1485 u64 start, u64 end, int *page_started,
1486 unsigned long *nr_written)
1489 int force_cow = need_force_cow(inode, start, end);
1491 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1492 ret = run_delalloc_nocow(inode, locked_page, start, end,
1493 page_started, 1, nr_written);
1494 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1495 ret = run_delalloc_nocow(inode, locked_page, start, end,
1496 page_started, 0, nr_written);
1497 } else if (!inode_need_compress(inode)) {
1498 ret = cow_file_range(inode, locked_page, start, end, end,
1499 page_started, nr_written, 1, NULL);
1501 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1502 &BTRFS_I(inode)->runtime_flags);
1503 ret = cow_file_range_async(inode, locked_page, start, end,
1504 page_started, nr_written);
1509 static void btrfs_split_extent_hook(struct inode *inode,
1510 struct extent_state *orig, u64 split)
1514 /* not delalloc, ignore it */
1515 if (!(orig->state & EXTENT_DELALLOC))
1518 size = orig->end - orig->start + 1;
1519 if (size > BTRFS_MAX_EXTENT_SIZE) {
1524 * See the explanation in btrfs_merge_extent_hook, the same
1525 * applies here, just in reverse.
1527 new_size = orig->end - split + 1;
1528 num_extents = count_max_extents(new_size);
1529 new_size = split - orig->start;
1530 num_extents += count_max_extents(new_size);
1531 if (count_max_extents(size) >= num_extents)
1535 spin_lock(&BTRFS_I(inode)->lock);
1536 BTRFS_I(inode)->outstanding_extents++;
1537 spin_unlock(&BTRFS_I(inode)->lock);
1541 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1542 * extents so we can keep track of new extents that are just merged onto old
1543 * extents, such as when we are doing sequential writes, so we can properly
1544 * account for the metadata space we'll need.
1546 static void btrfs_merge_extent_hook(struct inode *inode,
1547 struct extent_state *new,
1548 struct extent_state *other)
1550 u64 new_size, old_size;
1553 /* not delalloc, ignore it */
1554 if (!(other->state & EXTENT_DELALLOC))
1557 if (new->start > other->start)
1558 new_size = new->end - other->start + 1;
1560 new_size = other->end - new->start + 1;
1562 /* we're not bigger than the max, unreserve the space and go */
1563 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1564 spin_lock(&BTRFS_I(inode)->lock);
1565 BTRFS_I(inode)->outstanding_extents--;
1566 spin_unlock(&BTRFS_I(inode)->lock);
1571 * We have to add up either side to figure out how many extents were
1572 * accounted for before we merged into one big extent. If the number of
1573 * extents we accounted for is <= the amount we need for the new range
1574 * then we can return, otherwise drop. Think of it like this
1578 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1579 * need 2 outstanding extents, on one side we have 1 and the other side
1580 * we have 1 so they are == and we can return. But in this case
1582 * [MAX_SIZE+4k][MAX_SIZE+4k]
1584 * Each range on their own accounts for 2 extents, but merged together
1585 * they are only 3 extents worth of accounting, so we need to drop in
1588 old_size = other->end - other->start + 1;
1589 num_extents = count_max_extents(old_size);
1590 old_size = new->end - new->start + 1;
1591 num_extents += count_max_extents(old_size);
1592 if (count_max_extents(new_size) >= num_extents)
1595 spin_lock(&BTRFS_I(inode)->lock);
1596 BTRFS_I(inode)->outstanding_extents--;
1597 spin_unlock(&BTRFS_I(inode)->lock);
1600 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1601 struct inode *inode)
1603 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1605 spin_lock(&root->delalloc_lock);
1606 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1607 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1608 &root->delalloc_inodes);
1609 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1610 &BTRFS_I(inode)->runtime_flags);
1611 root->nr_delalloc_inodes++;
1612 if (root->nr_delalloc_inodes == 1) {
1613 spin_lock(&fs_info->delalloc_root_lock);
1614 BUG_ON(!list_empty(&root->delalloc_root));
1615 list_add_tail(&root->delalloc_root,
1616 &fs_info->delalloc_roots);
1617 spin_unlock(&fs_info->delalloc_root_lock);
1620 spin_unlock(&root->delalloc_lock);
1623 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1624 struct btrfs_inode *inode)
1626 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1628 spin_lock(&root->delalloc_lock);
1629 if (!list_empty(&inode->delalloc_inodes)) {
1630 list_del_init(&inode->delalloc_inodes);
1631 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1632 &inode->runtime_flags);
1633 root->nr_delalloc_inodes--;
1634 if (!root->nr_delalloc_inodes) {
1635 spin_lock(&fs_info->delalloc_root_lock);
1636 BUG_ON(list_empty(&root->delalloc_root));
1637 list_del_init(&root->delalloc_root);
1638 spin_unlock(&fs_info->delalloc_root_lock);
1641 spin_unlock(&root->delalloc_lock);
1645 * extent_io.c set_bit_hook, used to track delayed allocation
1646 * bytes in this file, and to maintain the list of inodes that
1647 * have pending delalloc work to be done.
1649 static void btrfs_set_bit_hook(struct inode *inode,
1650 struct extent_state *state, unsigned *bits)
1653 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1655 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1658 * set_bit and clear bit hooks normally require _irqsave/restore
1659 * but in this case, we are only testing for the DELALLOC
1660 * bit, which is only set or cleared with irqs on
1662 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1663 struct btrfs_root *root = BTRFS_I(inode)->root;
1664 u64 len = state->end + 1 - state->start;
1665 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1667 if (*bits & EXTENT_FIRST_DELALLOC) {
1668 *bits &= ~EXTENT_FIRST_DELALLOC;
1670 spin_lock(&BTRFS_I(inode)->lock);
1671 BTRFS_I(inode)->outstanding_extents++;
1672 spin_unlock(&BTRFS_I(inode)->lock);
1675 /* For sanity tests */
1676 if (btrfs_is_testing(fs_info))
1679 __percpu_counter_add(&fs_info->delalloc_bytes, len,
1680 fs_info->delalloc_batch);
1681 spin_lock(&BTRFS_I(inode)->lock);
1682 BTRFS_I(inode)->delalloc_bytes += len;
1683 if (*bits & EXTENT_DEFRAG)
1684 BTRFS_I(inode)->defrag_bytes += len;
1685 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1686 &BTRFS_I(inode)->runtime_flags))
1687 btrfs_add_delalloc_inodes(root, inode);
1688 spin_unlock(&BTRFS_I(inode)->lock);
1693 * extent_io.c clear_bit_hook, see set_bit_hook for why
1695 static void btrfs_clear_bit_hook(struct btrfs_inode *inode,
1696 struct extent_state *state,
1699 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1700 u64 len = state->end + 1 - state->start;
1701 u32 num_extents = count_max_extents(len);
1703 spin_lock(&inode->lock);
1704 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1705 inode->defrag_bytes -= len;
1706 spin_unlock(&inode->lock);
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714 struct btrfs_root *root = inode->root;
1715 bool do_list = !btrfs_is_free_space_inode(inode);
1717 if (*bits & EXTENT_FIRST_DELALLOC) {
1718 *bits &= ~EXTENT_FIRST_DELALLOC;
1719 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1720 spin_lock(&inode->lock);
1721 inode->outstanding_extents -= num_extents;
1722 spin_unlock(&inode->lock);
1726 * We don't reserve metadata space for space cache inodes so we
1727 * don't need to call dellalloc_release_metadata if there is an
1730 if (*bits & EXTENT_DO_ACCOUNTING &&
1731 root != fs_info->tree_root)
1732 btrfs_delalloc_release_metadata(inode, len);
1734 /* For sanity tests. */
1735 if (btrfs_is_testing(fs_info))
1738 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1739 && do_list && !(state->state & EXTENT_NORESERVE)
1740 && (*bits & (EXTENT_DO_ACCOUNTING |
1741 EXTENT_CLEAR_DATA_RESV)))
1742 btrfs_free_reserved_data_space_noquota(
1746 __percpu_counter_add(&fs_info->delalloc_bytes, -len,
1747 fs_info->delalloc_batch);
1748 spin_lock(&inode->lock);
1749 inode->delalloc_bytes -= len;
1750 if (do_list && inode->delalloc_bytes == 0 &&
1751 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1752 &inode->runtime_flags))
1753 btrfs_del_delalloc_inode(root, inode);
1754 spin_unlock(&inode->lock);
1759 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1760 * we don't create bios that span stripes or chunks
1762 * return 1 if page cannot be merged to bio
1763 * return 0 if page can be merged to bio
1764 * return error otherwise
1766 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1767 size_t size, struct bio *bio,
1768 unsigned long bio_flags)
1770 struct inode *inode = page->mapping->host;
1771 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1772 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1777 if (bio_flags & EXTENT_BIO_COMPRESSED)
1780 length = bio->bi_iter.bi_size;
1781 map_length = length;
1782 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1786 if (map_length < length + size)
1792 * in order to insert checksums into the metadata in large chunks,
1793 * we wait until bio submission time. All the pages in the bio are
1794 * checksummed and sums are attached onto the ordered extent record.
1796 * At IO completion time the cums attached on the ordered extent record
1797 * are inserted into the btree
1799 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1800 int mirror_num, unsigned long bio_flags,
1805 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1806 BUG_ON(ret); /* -ENOMEM */
1811 * in order to insert checksums into the metadata in large chunks,
1812 * we wait until bio submission time. All the pages in the bio are
1813 * checksummed and sums are attached onto the ordered extent record.
1815 * At IO completion time the cums attached on the ordered extent record
1816 * are inserted into the btree
1818 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1819 int mirror_num, unsigned long bio_flags,
1822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1825 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1827 bio->bi_error = ret;
1834 * extent_io.c submission hook. This does the right thing for csum calculation
1835 * on write, or reading the csums from the tree before a read
1837 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1838 int mirror_num, unsigned long bio_flags,
1841 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1842 struct btrfs_root *root = BTRFS_I(inode)->root;
1843 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1846 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1848 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1850 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1851 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1853 if (bio_op(bio) != REQ_OP_WRITE) {
1854 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1858 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1859 ret = btrfs_submit_compressed_read(inode, bio,
1863 } else if (!skip_sum) {
1864 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1869 } else if (async && !skip_sum) {
1870 /* csum items have already been cloned */
1871 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1873 /* we're doing a write, do the async checksumming */
1874 ret = btrfs_wq_submit_bio(fs_info, inode, bio, mirror_num,
1875 bio_flags, bio_offset,
1876 __btrfs_submit_bio_start,
1877 __btrfs_submit_bio_done);
1879 } else if (!skip_sum) {
1880 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1886 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
1890 bio->bi_error = ret;
1897 * given a list of ordered sums record them in the inode. This happens
1898 * at IO completion time based on sums calculated at bio submission time.
1900 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1901 struct inode *inode, struct list_head *list)
1903 struct btrfs_ordered_sum *sum;
1905 list_for_each_entry(sum, list, list) {
1906 trans->adding_csums = 1;
1907 btrfs_csum_file_blocks(trans,
1908 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1909 trans->adding_csums = 0;
1914 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1915 struct extent_state **cached_state, int dedupe)
1917 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1918 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1922 /* see btrfs_writepage_start_hook for details on why this is required */
1923 struct btrfs_writepage_fixup {
1925 struct btrfs_work work;
1928 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1930 struct btrfs_writepage_fixup *fixup;
1931 struct btrfs_ordered_extent *ordered;
1932 struct extent_state *cached_state = NULL;
1934 struct inode *inode;
1939 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1943 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1944 ClearPageChecked(page);
1948 inode = page->mapping->host;
1949 page_start = page_offset(page);
1950 page_end = page_offset(page) + PAGE_SIZE - 1;
1952 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
1955 /* already ordered? We're done */
1956 if (PagePrivate2(page))
1959 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
1962 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1963 page_end, &cached_state, GFP_NOFS);
1965 btrfs_start_ordered_extent(inode, ordered, 1);
1966 btrfs_put_ordered_extent(ordered);
1970 ret = btrfs_delalloc_reserve_space(inode, page_start,
1973 mapping_set_error(page->mapping, ret);
1974 end_extent_writepage(page, ret, page_start, page_end);
1975 ClearPageChecked(page);
1979 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
1981 ClearPageChecked(page);
1982 set_page_dirty(page);
1984 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1985 &cached_state, GFP_NOFS);
1993 * There are a few paths in the higher layers of the kernel that directly
1994 * set the page dirty bit without asking the filesystem if it is a
1995 * good idea. This causes problems because we want to make sure COW
1996 * properly happens and the data=ordered rules are followed.
1998 * In our case any range that doesn't have the ORDERED bit set
1999 * hasn't been properly setup for IO. We kick off an async process
2000 * to fix it up. The async helper will wait for ordered extents, set
2001 * the delalloc bit and make it safe to write the page.
2003 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2005 struct inode *inode = page->mapping->host;
2006 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2007 struct btrfs_writepage_fixup *fixup;
2009 /* this page is properly in the ordered list */
2010 if (TestClearPagePrivate2(page))
2013 if (PageChecked(page))
2016 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2020 SetPageChecked(page);
2022 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2023 btrfs_writepage_fixup_worker, NULL, NULL);
2025 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2029 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2030 struct inode *inode, u64 file_pos,
2031 u64 disk_bytenr, u64 disk_num_bytes,
2032 u64 num_bytes, u64 ram_bytes,
2033 u8 compression, u8 encryption,
2034 u16 other_encoding, int extent_type)
2036 struct btrfs_root *root = BTRFS_I(inode)->root;
2037 struct btrfs_file_extent_item *fi;
2038 struct btrfs_path *path;
2039 struct extent_buffer *leaf;
2040 struct btrfs_key ins;
2041 int extent_inserted = 0;
2044 path = btrfs_alloc_path();
2049 * we may be replacing one extent in the tree with another.
2050 * The new extent is pinned in the extent map, and we don't want
2051 * to drop it from the cache until it is completely in the btree.
2053 * So, tell btrfs_drop_extents to leave this extent in the cache.
2054 * the caller is expected to unpin it and allow it to be merged
2057 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2058 file_pos + num_bytes, NULL, 0,
2059 1, sizeof(*fi), &extent_inserted);
2063 if (!extent_inserted) {
2064 ins.objectid = btrfs_ino(BTRFS_I(inode));
2065 ins.offset = file_pos;
2066 ins.type = BTRFS_EXTENT_DATA_KEY;
2068 path->leave_spinning = 1;
2069 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2074 leaf = path->nodes[0];
2075 fi = btrfs_item_ptr(leaf, path->slots[0],
2076 struct btrfs_file_extent_item);
2077 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2078 btrfs_set_file_extent_type(leaf, fi, extent_type);
2079 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2080 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2081 btrfs_set_file_extent_offset(leaf, fi, 0);
2082 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2083 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2084 btrfs_set_file_extent_compression(leaf, fi, compression);
2085 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2086 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2088 btrfs_mark_buffer_dirty(leaf);
2089 btrfs_release_path(path);
2091 inode_add_bytes(inode, num_bytes);
2093 ins.objectid = disk_bytenr;
2094 ins.offset = disk_num_bytes;
2095 ins.type = BTRFS_EXTENT_ITEM_KEY;
2096 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2097 btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins);
2099 * Release the reserved range from inode dirty range map, as it is
2100 * already moved into delayed_ref_head
2102 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2104 btrfs_free_path(path);
2109 /* snapshot-aware defrag */
2110 struct sa_defrag_extent_backref {
2111 struct rb_node node;
2112 struct old_sa_defrag_extent *old;
2121 struct old_sa_defrag_extent {
2122 struct list_head list;
2123 struct new_sa_defrag_extent *new;
2132 struct new_sa_defrag_extent {
2133 struct rb_root root;
2134 struct list_head head;
2135 struct btrfs_path *path;
2136 struct inode *inode;
2144 static int backref_comp(struct sa_defrag_extent_backref *b1,
2145 struct sa_defrag_extent_backref *b2)
2147 if (b1->root_id < b2->root_id)
2149 else if (b1->root_id > b2->root_id)
2152 if (b1->inum < b2->inum)
2154 else if (b1->inum > b2->inum)
2157 if (b1->file_pos < b2->file_pos)
2159 else if (b1->file_pos > b2->file_pos)
2163 * [------------------------------] ===> (a range of space)
2164 * |<--->| |<---->| =============> (fs/file tree A)
2165 * |<---------------------------->| ===> (fs/file tree B)
2167 * A range of space can refer to two file extents in one tree while
2168 * refer to only one file extent in another tree.
2170 * So we may process a disk offset more than one time(two extents in A)
2171 * and locate at the same extent(one extent in B), then insert two same
2172 * backrefs(both refer to the extent in B).
2177 static void backref_insert(struct rb_root *root,
2178 struct sa_defrag_extent_backref *backref)
2180 struct rb_node **p = &root->rb_node;
2181 struct rb_node *parent = NULL;
2182 struct sa_defrag_extent_backref *entry;
2187 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2189 ret = backref_comp(backref, entry);
2193 p = &(*p)->rb_right;
2196 rb_link_node(&backref->node, parent, p);
2197 rb_insert_color(&backref->node, root);
2201 * Note the backref might has changed, and in this case we just return 0.
2203 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2206 struct btrfs_file_extent_item *extent;
2207 struct old_sa_defrag_extent *old = ctx;
2208 struct new_sa_defrag_extent *new = old->new;
2209 struct btrfs_path *path = new->path;
2210 struct btrfs_key key;
2211 struct btrfs_root *root;
2212 struct sa_defrag_extent_backref *backref;
2213 struct extent_buffer *leaf;
2214 struct inode *inode = new->inode;
2215 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2221 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2222 inum == btrfs_ino(BTRFS_I(inode)))
2225 key.objectid = root_id;
2226 key.type = BTRFS_ROOT_ITEM_KEY;
2227 key.offset = (u64)-1;
2229 root = btrfs_read_fs_root_no_name(fs_info, &key);
2231 if (PTR_ERR(root) == -ENOENT)
2234 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2235 inum, offset, root_id);
2236 return PTR_ERR(root);
2239 key.objectid = inum;
2240 key.type = BTRFS_EXTENT_DATA_KEY;
2241 if (offset > (u64)-1 << 32)
2244 key.offset = offset;
2246 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2247 if (WARN_ON(ret < 0))
2254 leaf = path->nodes[0];
2255 slot = path->slots[0];
2257 if (slot >= btrfs_header_nritems(leaf)) {
2258 ret = btrfs_next_leaf(root, path);
2261 } else if (ret > 0) {
2270 btrfs_item_key_to_cpu(leaf, &key, slot);
2272 if (key.objectid > inum)
2275 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2278 extent = btrfs_item_ptr(leaf, slot,
2279 struct btrfs_file_extent_item);
2281 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2285 * 'offset' refers to the exact key.offset,
2286 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2287 * (key.offset - extent_offset).
2289 if (key.offset != offset)
2292 extent_offset = btrfs_file_extent_offset(leaf, extent);
2293 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2295 if (extent_offset >= old->extent_offset + old->offset +
2296 old->len || extent_offset + num_bytes <=
2297 old->extent_offset + old->offset)
2302 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2308 backref->root_id = root_id;
2309 backref->inum = inum;
2310 backref->file_pos = offset;
2311 backref->num_bytes = num_bytes;
2312 backref->extent_offset = extent_offset;
2313 backref->generation = btrfs_file_extent_generation(leaf, extent);
2315 backref_insert(&new->root, backref);
2318 btrfs_release_path(path);
2323 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2324 struct new_sa_defrag_extent *new)
2326 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2327 struct old_sa_defrag_extent *old, *tmp;
2332 list_for_each_entry_safe(old, tmp, &new->head, list) {
2333 ret = iterate_inodes_from_logical(old->bytenr +
2334 old->extent_offset, fs_info,
2335 path, record_one_backref,
2337 if (ret < 0 && ret != -ENOENT)
2340 /* no backref to be processed for this extent */
2342 list_del(&old->list);
2347 if (list_empty(&new->head))
2353 static int relink_is_mergable(struct extent_buffer *leaf,
2354 struct btrfs_file_extent_item *fi,
2355 struct new_sa_defrag_extent *new)
2357 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2360 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2363 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2366 if (btrfs_file_extent_encryption(leaf, fi) ||
2367 btrfs_file_extent_other_encoding(leaf, fi))
2374 * Note the backref might has changed, and in this case we just return 0.
2376 static noinline int relink_extent_backref(struct btrfs_path *path,
2377 struct sa_defrag_extent_backref *prev,
2378 struct sa_defrag_extent_backref *backref)
2380 struct btrfs_file_extent_item *extent;
2381 struct btrfs_file_extent_item *item;
2382 struct btrfs_ordered_extent *ordered;
2383 struct btrfs_trans_handle *trans;
2384 struct btrfs_root *root;
2385 struct btrfs_key key;
2386 struct extent_buffer *leaf;
2387 struct old_sa_defrag_extent *old = backref->old;
2388 struct new_sa_defrag_extent *new = old->new;
2389 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2390 struct inode *inode;
2391 struct extent_state *cached = NULL;
2400 if (prev && prev->root_id == backref->root_id &&
2401 prev->inum == backref->inum &&
2402 prev->file_pos + prev->num_bytes == backref->file_pos)
2405 /* step 1: get root */
2406 key.objectid = backref->root_id;
2407 key.type = BTRFS_ROOT_ITEM_KEY;
2408 key.offset = (u64)-1;
2410 index = srcu_read_lock(&fs_info->subvol_srcu);
2412 root = btrfs_read_fs_root_no_name(fs_info, &key);
2414 srcu_read_unlock(&fs_info->subvol_srcu, index);
2415 if (PTR_ERR(root) == -ENOENT)
2417 return PTR_ERR(root);
2420 if (btrfs_root_readonly(root)) {
2421 srcu_read_unlock(&fs_info->subvol_srcu, index);
2425 /* step 2: get inode */
2426 key.objectid = backref->inum;
2427 key.type = BTRFS_INODE_ITEM_KEY;
2430 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2431 if (IS_ERR(inode)) {
2432 srcu_read_unlock(&fs_info->subvol_srcu, index);
2436 srcu_read_unlock(&fs_info->subvol_srcu, index);
2438 /* step 3: relink backref */
2439 lock_start = backref->file_pos;
2440 lock_end = backref->file_pos + backref->num_bytes - 1;
2441 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2444 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2446 btrfs_put_ordered_extent(ordered);
2450 trans = btrfs_join_transaction(root);
2451 if (IS_ERR(trans)) {
2452 ret = PTR_ERR(trans);
2456 key.objectid = backref->inum;
2457 key.type = BTRFS_EXTENT_DATA_KEY;
2458 key.offset = backref->file_pos;
2460 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2463 } else if (ret > 0) {
2468 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2469 struct btrfs_file_extent_item);
2471 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2472 backref->generation)
2475 btrfs_release_path(path);
2477 start = backref->file_pos;
2478 if (backref->extent_offset < old->extent_offset + old->offset)
2479 start += old->extent_offset + old->offset -
2480 backref->extent_offset;
2482 len = min(backref->extent_offset + backref->num_bytes,
2483 old->extent_offset + old->offset + old->len);
2484 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2486 ret = btrfs_drop_extents(trans, root, inode, start,
2491 key.objectid = btrfs_ino(BTRFS_I(inode));
2492 key.type = BTRFS_EXTENT_DATA_KEY;
2495 path->leave_spinning = 1;
2497 struct btrfs_file_extent_item *fi;
2499 struct btrfs_key found_key;
2501 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2506 leaf = path->nodes[0];
2507 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2509 fi = btrfs_item_ptr(leaf, path->slots[0],
2510 struct btrfs_file_extent_item);
2511 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2513 if (extent_len + found_key.offset == start &&
2514 relink_is_mergable(leaf, fi, new)) {
2515 btrfs_set_file_extent_num_bytes(leaf, fi,
2517 btrfs_mark_buffer_dirty(leaf);
2518 inode_add_bytes(inode, len);
2524 btrfs_release_path(path);
2529 ret = btrfs_insert_empty_item(trans, root, path, &key,
2532 btrfs_abort_transaction(trans, ret);
2536 leaf = path->nodes[0];
2537 item = btrfs_item_ptr(leaf, path->slots[0],
2538 struct btrfs_file_extent_item);
2539 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2540 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2541 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2542 btrfs_set_file_extent_num_bytes(leaf, item, len);
2543 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2544 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2545 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2546 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2547 btrfs_set_file_extent_encryption(leaf, item, 0);
2548 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2550 btrfs_mark_buffer_dirty(leaf);
2551 inode_add_bytes(inode, len);
2552 btrfs_release_path(path);
2554 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2556 backref->root_id, backref->inum,
2557 new->file_pos); /* start - extent_offset */
2559 btrfs_abort_transaction(trans, ret);
2565 btrfs_release_path(path);
2566 path->leave_spinning = 0;
2567 btrfs_end_transaction(trans);
2569 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2575 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2577 struct old_sa_defrag_extent *old, *tmp;
2582 list_for_each_entry_safe(old, tmp, &new->head, list) {
2588 static void relink_file_extents(struct new_sa_defrag_extent *new)
2590 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2591 struct btrfs_path *path;
2592 struct sa_defrag_extent_backref *backref;
2593 struct sa_defrag_extent_backref *prev = NULL;
2594 struct inode *inode;
2595 struct btrfs_root *root;
2596 struct rb_node *node;
2600 root = BTRFS_I(inode)->root;
2602 path = btrfs_alloc_path();
2606 if (!record_extent_backrefs(path, new)) {
2607 btrfs_free_path(path);
2610 btrfs_release_path(path);
2613 node = rb_first(&new->root);
2616 rb_erase(node, &new->root);
2618 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2620 ret = relink_extent_backref(path, prev, backref);
2633 btrfs_free_path(path);
2635 free_sa_defrag_extent(new);
2637 atomic_dec(&fs_info->defrag_running);
2638 wake_up(&fs_info->transaction_wait);
2641 static struct new_sa_defrag_extent *
2642 record_old_file_extents(struct inode *inode,
2643 struct btrfs_ordered_extent *ordered)
2645 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2646 struct btrfs_root *root = BTRFS_I(inode)->root;
2647 struct btrfs_path *path;
2648 struct btrfs_key key;
2649 struct old_sa_defrag_extent *old;
2650 struct new_sa_defrag_extent *new;
2653 new = kmalloc(sizeof(*new), GFP_NOFS);
2658 new->file_pos = ordered->file_offset;
2659 new->len = ordered->len;
2660 new->bytenr = ordered->start;
2661 new->disk_len = ordered->disk_len;
2662 new->compress_type = ordered->compress_type;
2663 new->root = RB_ROOT;
2664 INIT_LIST_HEAD(&new->head);
2666 path = btrfs_alloc_path();
2670 key.objectid = btrfs_ino(BTRFS_I(inode));
2671 key.type = BTRFS_EXTENT_DATA_KEY;
2672 key.offset = new->file_pos;
2674 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2677 if (ret > 0 && path->slots[0] > 0)
2680 /* find out all the old extents for the file range */
2682 struct btrfs_file_extent_item *extent;
2683 struct extent_buffer *l;
2692 slot = path->slots[0];
2694 if (slot >= btrfs_header_nritems(l)) {
2695 ret = btrfs_next_leaf(root, path);
2703 btrfs_item_key_to_cpu(l, &key, slot);
2705 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2707 if (key.type != BTRFS_EXTENT_DATA_KEY)
2709 if (key.offset >= new->file_pos + new->len)
2712 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2714 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2715 if (key.offset + num_bytes < new->file_pos)
2718 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2722 extent_offset = btrfs_file_extent_offset(l, extent);
2724 old = kmalloc(sizeof(*old), GFP_NOFS);
2728 offset = max(new->file_pos, key.offset);
2729 end = min(new->file_pos + new->len, key.offset + num_bytes);
2731 old->bytenr = disk_bytenr;
2732 old->extent_offset = extent_offset;
2733 old->offset = offset - key.offset;
2734 old->len = end - offset;
2737 list_add_tail(&old->list, &new->head);
2743 btrfs_free_path(path);
2744 atomic_inc(&fs_info->defrag_running);
2749 btrfs_free_path(path);
2751 free_sa_defrag_extent(new);
2755 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2758 struct btrfs_block_group_cache *cache;
2760 cache = btrfs_lookup_block_group(fs_info, start);
2763 spin_lock(&cache->lock);
2764 cache->delalloc_bytes -= len;
2765 spin_unlock(&cache->lock);
2767 btrfs_put_block_group(cache);
2770 /* as ordered data IO finishes, this gets called so we can finish
2771 * an ordered extent if the range of bytes in the file it covers are
2774 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2776 struct inode *inode = ordered_extent->inode;
2777 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2778 struct btrfs_root *root = BTRFS_I(inode)->root;
2779 struct btrfs_trans_handle *trans = NULL;
2780 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2781 struct extent_state *cached_state = NULL;
2782 struct new_sa_defrag_extent *new = NULL;
2783 int compress_type = 0;
2785 u64 logical_len = ordered_extent->len;
2787 bool truncated = false;
2789 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2791 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2796 btrfs_free_io_failure_record(BTRFS_I(inode),
2797 ordered_extent->file_offset,
2798 ordered_extent->file_offset +
2799 ordered_extent->len - 1);
2801 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2803 logical_len = ordered_extent->truncated_len;
2804 /* Truncated the entire extent, don't bother adding */
2809 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2810 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2813 * For mwrite(mmap + memset to write) case, we still reserve
2814 * space for NOCOW range.
2815 * As NOCOW won't cause a new delayed ref, just free the space
2817 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2818 ordered_extent->len);
2819 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2821 trans = btrfs_join_transaction_nolock(root);
2823 trans = btrfs_join_transaction(root);
2824 if (IS_ERR(trans)) {
2825 ret = PTR_ERR(trans);
2829 trans->block_rsv = &fs_info->delalloc_block_rsv;
2830 ret = btrfs_update_inode_fallback(trans, root, inode);
2831 if (ret) /* -ENOMEM or corruption */
2832 btrfs_abort_transaction(trans, ret);
2836 lock_extent_bits(io_tree, ordered_extent->file_offset,
2837 ordered_extent->file_offset + ordered_extent->len - 1,
2840 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2841 ordered_extent->file_offset + ordered_extent->len - 1,
2842 EXTENT_DEFRAG, 1, cached_state);
2844 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2845 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2846 /* the inode is shared */
2847 new = record_old_file_extents(inode, ordered_extent);
2849 clear_extent_bit(io_tree, ordered_extent->file_offset,
2850 ordered_extent->file_offset + ordered_extent->len - 1,
2851 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2855 trans = btrfs_join_transaction_nolock(root);
2857 trans = btrfs_join_transaction(root);
2858 if (IS_ERR(trans)) {
2859 ret = PTR_ERR(trans);
2864 trans->block_rsv = &fs_info->delalloc_block_rsv;
2866 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2867 compress_type = ordered_extent->compress_type;
2868 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2869 BUG_ON(compress_type);
2870 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2871 ordered_extent->file_offset,
2872 ordered_extent->file_offset +
2875 BUG_ON(root == fs_info->tree_root);
2876 ret = insert_reserved_file_extent(trans, inode,
2877 ordered_extent->file_offset,
2878 ordered_extent->start,
2879 ordered_extent->disk_len,
2880 logical_len, logical_len,
2881 compress_type, 0, 0,
2882 BTRFS_FILE_EXTENT_REG);
2884 btrfs_release_delalloc_bytes(fs_info,
2885 ordered_extent->start,
2886 ordered_extent->disk_len);
2888 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2889 ordered_extent->file_offset, ordered_extent->len,
2892 btrfs_abort_transaction(trans, ret);
2896 add_pending_csums(trans, inode, &ordered_extent->list);
2898 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2899 ret = btrfs_update_inode_fallback(trans, root, inode);
2900 if (ret) { /* -ENOMEM or corruption */
2901 btrfs_abort_transaction(trans, ret);
2906 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2907 ordered_extent->file_offset +
2908 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2910 if (root != fs_info->tree_root)
2911 btrfs_delalloc_release_metadata(BTRFS_I(inode),
2912 ordered_extent->len);
2914 btrfs_end_transaction(trans);
2916 if (ret || truncated) {
2920 start = ordered_extent->file_offset + logical_len;
2922 start = ordered_extent->file_offset;
2923 end = ordered_extent->file_offset + ordered_extent->len - 1;
2924 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2926 /* Drop the cache for the part of the extent we didn't write. */
2927 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
2930 * If the ordered extent had an IOERR or something else went
2931 * wrong we need to return the space for this ordered extent
2932 * back to the allocator. We only free the extent in the
2933 * truncated case if we didn't write out the extent at all.
2935 if ((ret || !logical_len) &&
2936 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2937 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2938 btrfs_free_reserved_extent(fs_info,
2939 ordered_extent->start,
2940 ordered_extent->disk_len, 1);
2945 * This needs to be done to make sure anybody waiting knows we are done
2946 * updating everything for this ordered extent.
2948 btrfs_remove_ordered_extent(inode, ordered_extent);
2950 /* for snapshot-aware defrag */
2953 free_sa_defrag_extent(new);
2954 atomic_dec(&fs_info->defrag_running);
2956 relink_file_extents(new);
2961 btrfs_put_ordered_extent(ordered_extent);
2962 /* once for the tree */
2963 btrfs_put_ordered_extent(ordered_extent);
2968 static void finish_ordered_fn(struct btrfs_work *work)
2970 struct btrfs_ordered_extent *ordered_extent;
2971 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2972 btrfs_finish_ordered_io(ordered_extent);
2975 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2976 struct extent_state *state, int uptodate)
2978 struct inode *inode = page->mapping->host;
2979 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2980 struct btrfs_ordered_extent *ordered_extent = NULL;
2981 struct btrfs_workqueue *wq;
2982 btrfs_work_func_t func;
2984 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2986 ClearPagePrivate2(page);
2987 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2988 end - start + 1, uptodate))
2991 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
2992 wq = fs_info->endio_freespace_worker;
2993 func = btrfs_freespace_write_helper;
2995 wq = fs_info->endio_write_workers;
2996 func = btrfs_endio_write_helper;
2999 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3001 btrfs_queue_work(wq, &ordered_extent->work);
3006 static int __readpage_endio_check(struct inode *inode,
3007 struct btrfs_io_bio *io_bio,
3008 int icsum, struct page *page,
3009 int pgoff, u64 start, size_t len)
3015 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3017 kaddr = kmap_atomic(page);
3018 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3019 btrfs_csum_final(csum, (u8 *)&csum);
3020 if (csum != csum_expected)
3023 kunmap_atomic(kaddr);
3026 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3027 io_bio->mirror_num);
3028 memset(kaddr + pgoff, 1, len);
3029 flush_dcache_page(page);
3030 kunmap_atomic(kaddr);
3031 if (csum_expected == 0)
3037 * when reads are done, we need to check csums to verify the data is correct
3038 * if there's a match, we allow the bio to finish. If not, the code in
3039 * extent_io.c will try to find good copies for us.
3041 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3042 u64 phy_offset, struct page *page,
3043 u64 start, u64 end, int mirror)
3045 size_t offset = start - page_offset(page);
3046 struct inode *inode = page->mapping->host;
3047 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3048 struct btrfs_root *root = BTRFS_I(inode)->root;
3050 if (PageChecked(page)) {
3051 ClearPageChecked(page);
3055 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3058 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3059 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3060 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3064 phy_offset >>= inode->i_sb->s_blocksize_bits;
3065 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3066 start, (size_t)(end - start + 1));
3069 void btrfs_add_delayed_iput(struct inode *inode)
3071 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3072 struct btrfs_inode *binode = BTRFS_I(inode);
3074 if (atomic_add_unless(&inode->i_count, -1, 1))
3077 spin_lock(&fs_info->delayed_iput_lock);
3078 if (binode->delayed_iput_count == 0) {
3079 ASSERT(list_empty(&binode->delayed_iput));
3080 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3082 binode->delayed_iput_count++;
3084 spin_unlock(&fs_info->delayed_iput_lock);
3087 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3090 spin_lock(&fs_info->delayed_iput_lock);
3091 while (!list_empty(&fs_info->delayed_iputs)) {
3092 struct btrfs_inode *inode;
3094 inode = list_first_entry(&fs_info->delayed_iputs,
3095 struct btrfs_inode, delayed_iput);
3096 if (inode->delayed_iput_count) {
3097 inode->delayed_iput_count--;
3098 list_move_tail(&inode->delayed_iput,
3099 &fs_info->delayed_iputs);
3101 list_del_init(&inode->delayed_iput);
3103 spin_unlock(&fs_info->delayed_iput_lock);
3104 iput(&inode->vfs_inode);
3105 spin_lock(&fs_info->delayed_iput_lock);
3107 spin_unlock(&fs_info->delayed_iput_lock);
3111 * This is called in transaction commit time. If there are no orphan
3112 * files in the subvolume, it removes orphan item and frees block_rsv
3115 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3116 struct btrfs_root *root)
3118 struct btrfs_fs_info *fs_info = root->fs_info;
3119 struct btrfs_block_rsv *block_rsv;
3122 if (atomic_read(&root->orphan_inodes) ||
3123 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3126 spin_lock(&root->orphan_lock);
3127 if (atomic_read(&root->orphan_inodes)) {
3128 spin_unlock(&root->orphan_lock);
3132 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3133 spin_unlock(&root->orphan_lock);
3137 block_rsv = root->orphan_block_rsv;
3138 root->orphan_block_rsv = NULL;
3139 spin_unlock(&root->orphan_lock);
3141 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3142 btrfs_root_refs(&root->root_item) > 0) {
3143 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3144 root->root_key.objectid);
3146 btrfs_abort_transaction(trans, ret);
3148 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3153 WARN_ON(block_rsv->size > 0);
3154 btrfs_free_block_rsv(fs_info, block_rsv);
3159 * This creates an orphan entry for the given inode in case something goes
3160 * wrong in the middle of an unlink/truncate.
3162 * NOTE: caller of this function should reserve 5 units of metadata for
3165 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3166 struct btrfs_inode *inode)
3168 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3169 struct btrfs_root *root = inode->root;
3170 struct btrfs_block_rsv *block_rsv = NULL;
3175 if (!root->orphan_block_rsv) {
3176 block_rsv = btrfs_alloc_block_rsv(fs_info,
3177 BTRFS_BLOCK_RSV_TEMP);
3182 spin_lock(&root->orphan_lock);
3183 if (!root->orphan_block_rsv) {
3184 root->orphan_block_rsv = block_rsv;
3185 } else if (block_rsv) {
3186 btrfs_free_block_rsv(fs_info, block_rsv);
3190 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3191 &inode->runtime_flags)) {
3194 * For proper ENOSPC handling, we should do orphan
3195 * cleanup when mounting. But this introduces backward
3196 * compatibility issue.
3198 if (!xchg(&root->orphan_item_inserted, 1))
3204 atomic_inc(&root->orphan_inodes);
3207 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3208 &inode->runtime_flags))
3210 spin_unlock(&root->orphan_lock);
3212 /* grab metadata reservation from transaction handle */
3214 ret = btrfs_orphan_reserve_metadata(trans, inode);
3217 atomic_dec(&root->orphan_inodes);
3218 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3219 &inode->runtime_flags);
3221 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3222 &inode->runtime_flags);
3227 /* insert an orphan item to track this unlinked/truncated file */
3229 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3231 atomic_dec(&root->orphan_inodes);
3233 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3234 &inode->runtime_flags);
3235 btrfs_orphan_release_metadata(inode);
3237 if (ret != -EEXIST) {
3238 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3239 &inode->runtime_flags);
3240 btrfs_abort_transaction(trans, ret);
3247 /* insert an orphan item to track subvolume contains orphan files */
3249 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3250 root->root_key.objectid);
3251 if (ret && ret != -EEXIST) {
3252 btrfs_abort_transaction(trans, ret);
3260 * We have done the truncate/delete so we can go ahead and remove the orphan
3261 * item for this particular inode.
3263 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3264 struct btrfs_inode *inode)
3266 struct btrfs_root *root = inode->root;
3267 int delete_item = 0;
3268 int release_rsv = 0;
3271 spin_lock(&root->orphan_lock);
3272 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &inode->runtime_flags))
3276 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3277 &inode->runtime_flags))
3279 spin_unlock(&root->orphan_lock);
3282 atomic_dec(&root->orphan_inodes);
3284 ret = btrfs_del_orphan_item(trans, root,
3289 btrfs_orphan_release_metadata(inode);
3295 * this cleans up any orphans that may be left on the list from the last use
3298 int btrfs_orphan_cleanup(struct btrfs_root *root)
3300 struct btrfs_fs_info *fs_info = root->fs_info;
3301 struct btrfs_path *path;
3302 struct extent_buffer *leaf;
3303 struct btrfs_key key, found_key;
3304 struct btrfs_trans_handle *trans;
3305 struct inode *inode;
3306 u64 last_objectid = 0;
3307 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3309 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3312 path = btrfs_alloc_path();
3317 path->reada = READA_BACK;
3319 key.objectid = BTRFS_ORPHAN_OBJECTID;
3320 key.type = BTRFS_ORPHAN_ITEM_KEY;
3321 key.offset = (u64)-1;
3324 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3329 * if ret == 0 means we found what we were searching for, which
3330 * is weird, but possible, so only screw with path if we didn't
3331 * find the key and see if we have stuff that matches
3335 if (path->slots[0] == 0)
3340 /* pull out the item */
3341 leaf = path->nodes[0];
3342 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3344 /* make sure the item matches what we want */
3345 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3347 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3350 /* release the path since we're done with it */
3351 btrfs_release_path(path);
3354 * this is where we are basically btrfs_lookup, without the
3355 * crossing root thing. we store the inode number in the
3356 * offset of the orphan item.
3359 if (found_key.offset == last_objectid) {
3361 "Error removing orphan entry, stopping orphan cleanup");
3366 last_objectid = found_key.offset;
3368 found_key.objectid = found_key.offset;
3369 found_key.type = BTRFS_INODE_ITEM_KEY;
3370 found_key.offset = 0;
3371 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3372 ret = PTR_ERR_OR_ZERO(inode);
3373 if (ret && ret != -ENOENT)
3376 if (ret == -ENOENT && root == fs_info->tree_root) {
3377 struct btrfs_root *dead_root;
3378 struct btrfs_fs_info *fs_info = root->fs_info;
3379 int is_dead_root = 0;
3382 * this is an orphan in the tree root. Currently these
3383 * could come from 2 sources:
3384 * a) a snapshot deletion in progress
3385 * b) a free space cache inode
3386 * We need to distinguish those two, as the snapshot
3387 * orphan must not get deleted.
3388 * find_dead_roots already ran before us, so if this
3389 * is a snapshot deletion, we should find the root
3390 * in the dead_roots list
3392 spin_lock(&fs_info->trans_lock);
3393 list_for_each_entry(dead_root, &fs_info->dead_roots,
3395 if (dead_root->root_key.objectid ==
3396 found_key.objectid) {
3401 spin_unlock(&fs_info->trans_lock);
3403 /* prevent this orphan from being found again */
3404 key.offset = found_key.objectid - 1;
3409 * Inode is already gone but the orphan item is still there,
3410 * kill the orphan item.
3412 if (ret == -ENOENT) {
3413 trans = btrfs_start_transaction(root, 1);
3414 if (IS_ERR(trans)) {
3415 ret = PTR_ERR(trans);
3418 btrfs_debug(fs_info, "auto deleting %Lu",
3419 found_key.objectid);
3420 ret = btrfs_del_orphan_item(trans, root,
3421 found_key.objectid);
3422 btrfs_end_transaction(trans);
3429 * add this inode to the orphan list so btrfs_orphan_del does
3430 * the proper thing when we hit it
3432 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3433 &BTRFS_I(inode)->runtime_flags);
3434 atomic_inc(&root->orphan_inodes);
3436 /* if we have links, this was a truncate, lets do that */
3437 if (inode->i_nlink) {
3438 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3444 /* 1 for the orphan item deletion. */
3445 trans = btrfs_start_transaction(root, 1);
3446 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3451 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3452 btrfs_end_transaction(trans);
3458 ret = btrfs_truncate(inode);
3460 btrfs_orphan_del(NULL, BTRFS_I(inode));
3465 /* this will do delete_inode and everything for us */
3470 /* release the path since we're done with it */
3471 btrfs_release_path(path);
3473 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3475 if (root->orphan_block_rsv)
3476 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3479 if (root->orphan_block_rsv ||
3480 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3481 trans = btrfs_join_transaction(root);
3483 btrfs_end_transaction(trans);
3487 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3489 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3493 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3494 btrfs_free_path(path);
3499 * very simple check to peek ahead in the leaf looking for xattrs. If we
3500 * don't find any xattrs, we know there can't be any acls.
3502 * slot is the slot the inode is in, objectid is the objectid of the inode
3504 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3505 int slot, u64 objectid,
3506 int *first_xattr_slot)
3508 u32 nritems = btrfs_header_nritems(leaf);
3509 struct btrfs_key found_key;
3510 static u64 xattr_access = 0;
3511 static u64 xattr_default = 0;
3514 if (!xattr_access) {
3515 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3516 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3517 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3518 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3522 *first_xattr_slot = -1;
3523 while (slot < nritems) {
3524 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3526 /* we found a different objectid, there must not be acls */
3527 if (found_key.objectid != objectid)
3530 /* we found an xattr, assume we've got an acl */
3531 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3532 if (*first_xattr_slot == -1)
3533 *first_xattr_slot = slot;
3534 if (found_key.offset == xattr_access ||
3535 found_key.offset == xattr_default)
3540 * we found a key greater than an xattr key, there can't
3541 * be any acls later on
3543 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3550 * it goes inode, inode backrefs, xattrs, extents,
3551 * so if there are a ton of hard links to an inode there can
3552 * be a lot of backrefs. Don't waste time searching too hard,
3553 * this is just an optimization
3558 /* we hit the end of the leaf before we found an xattr or
3559 * something larger than an xattr. We have to assume the inode
3562 if (*first_xattr_slot == -1)
3563 *first_xattr_slot = slot;
3568 * read an inode from the btree into the in-memory inode
3570 static int btrfs_read_locked_inode(struct inode *inode)
3572 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3573 struct btrfs_path *path;
3574 struct extent_buffer *leaf;
3575 struct btrfs_inode_item *inode_item;
3576 struct btrfs_root *root = BTRFS_I(inode)->root;
3577 struct btrfs_key location;
3582 bool filled = false;
3583 int first_xattr_slot;
3585 ret = btrfs_fill_inode(inode, &rdev);
3589 path = btrfs_alloc_path();
3595 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3597 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3604 leaf = path->nodes[0];
3609 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3610 struct btrfs_inode_item);
3611 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3612 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3613 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3614 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3615 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3617 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3618 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3620 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3621 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3623 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3624 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3626 BTRFS_I(inode)->i_otime.tv_sec =
3627 btrfs_timespec_sec(leaf, &inode_item->otime);
3628 BTRFS_I(inode)->i_otime.tv_nsec =
3629 btrfs_timespec_nsec(leaf, &inode_item->otime);
3631 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3632 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3633 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3635 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3636 inode->i_generation = BTRFS_I(inode)->generation;
3638 rdev = btrfs_inode_rdev(leaf, inode_item);
3640 BTRFS_I(inode)->index_cnt = (u64)-1;
3641 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3645 * If we were modified in the current generation and evicted from memory
3646 * and then re-read we need to do a full sync since we don't have any
3647 * idea about which extents were modified before we were evicted from
3650 * This is required for both inode re-read from disk and delayed inode
3651 * in delayed_nodes_tree.
3653 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3654 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3655 &BTRFS_I(inode)->runtime_flags);
3658 * We don't persist the id of the transaction where an unlink operation
3659 * against the inode was last made. So here we assume the inode might
3660 * have been evicted, and therefore the exact value of last_unlink_trans
3661 * lost, and set it to last_trans to avoid metadata inconsistencies
3662 * between the inode and its parent if the inode is fsync'ed and the log
3663 * replayed. For example, in the scenario:
3666 * ln mydir/foo mydir/bar
3669 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3670 * xfs_io -c fsync mydir/foo
3672 * mount fs, triggers fsync log replay
3674 * We must make sure that when we fsync our inode foo we also log its
3675 * parent inode, otherwise after log replay the parent still has the
3676 * dentry with the "bar" name but our inode foo has a link count of 1
3677 * and doesn't have an inode ref with the name "bar" anymore.
3679 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3680 * but it guarantees correctness at the expense of occasional full
3681 * transaction commits on fsync if our inode is a directory, or if our
3682 * inode is not a directory, logging its parent unnecessarily.
3684 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3687 if (inode->i_nlink != 1 ||
3688 path->slots[0] >= btrfs_header_nritems(leaf))
3691 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3692 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3695 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3696 if (location.type == BTRFS_INODE_REF_KEY) {
3697 struct btrfs_inode_ref *ref;
3699 ref = (struct btrfs_inode_ref *)ptr;
3700 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3701 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3702 struct btrfs_inode_extref *extref;
3704 extref = (struct btrfs_inode_extref *)ptr;
3705 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3710 * try to precache a NULL acl entry for files that don't have
3711 * any xattrs or acls
3713 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3714 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3715 if (first_xattr_slot != -1) {
3716 path->slots[0] = first_xattr_slot;
3717 ret = btrfs_load_inode_props(inode, path);
3720 "error loading props for ino %llu (root %llu): %d",
3721 btrfs_ino(BTRFS_I(inode)),
3722 root->root_key.objectid, ret);
3724 btrfs_free_path(path);
3727 cache_no_acl(inode);
3729 switch (inode->i_mode & S_IFMT) {
3731 inode->i_mapping->a_ops = &btrfs_aops;
3732 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3733 inode->i_fop = &btrfs_file_operations;
3734 inode->i_op = &btrfs_file_inode_operations;
3737 inode->i_fop = &btrfs_dir_file_operations;
3738 inode->i_op = &btrfs_dir_inode_operations;
3741 inode->i_op = &btrfs_symlink_inode_operations;
3742 inode_nohighmem(inode);
3743 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3746 inode->i_op = &btrfs_special_inode_operations;
3747 init_special_inode(inode, inode->i_mode, rdev);
3751 btrfs_update_iflags(inode);
3755 btrfs_free_path(path);
3756 make_bad_inode(inode);
3761 * given a leaf and an inode, copy the inode fields into the leaf
3763 static void fill_inode_item(struct btrfs_trans_handle *trans,
3764 struct extent_buffer *leaf,
3765 struct btrfs_inode_item *item,
3766 struct inode *inode)
3768 struct btrfs_map_token token;
3770 btrfs_init_map_token(&token);
3772 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3773 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3774 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3776 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3777 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3779 btrfs_set_token_timespec_sec(leaf, &item->atime,
3780 inode->i_atime.tv_sec, &token);
3781 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3782 inode->i_atime.tv_nsec, &token);
3784 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3785 inode->i_mtime.tv_sec, &token);
3786 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3787 inode->i_mtime.tv_nsec, &token);
3789 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3790 inode->i_ctime.tv_sec, &token);
3791 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3792 inode->i_ctime.tv_nsec, &token);
3794 btrfs_set_token_timespec_sec(leaf, &item->otime,
3795 BTRFS_I(inode)->i_otime.tv_sec, &token);
3796 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3797 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3799 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3801 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3803 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3804 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3805 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3806 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3807 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3811 * copy everything in the in-memory inode into the btree.
3813 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3814 struct btrfs_root *root, struct inode *inode)
3816 struct btrfs_inode_item *inode_item;
3817 struct btrfs_path *path;
3818 struct extent_buffer *leaf;
3821 path = btrfs_alloc_path();
3825 path->leave_spinning = 1;
3826 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3834 leaf = path->nodes[0];
3835 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3836 struct btrfs_inode_item);
3838 fill_inode_item(trans, leaf, inode_item, inode);
3839 btrfs_mark_buffer_dirty(leaf);
3840 btrfs_set_inode_last_trans(trans, inode);
3843 btrfs_free_path(path);
3848 * copy everything in the in-memory inode into the btree.
3850 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3851 struct btrfs_root *root, struct inode *inode)
3853 struct btrfs_fs_info *fs_info = root->fs_info;
3857 * If the inode is a free space inode, we can deadlock during commit
3858 * if we put it into the delayed code.
3860 * The data relocation inode should also be directly updated
3863 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3864 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3865 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3866 btrfs_update_root_times(trans, root);
3868 ret = btrfs_delayed_update_inode(trans, root, inode);
3870 btrfs_set_inode_last_trans(trans, inode);
3874 return btrfs_update_inode_item(trans, root, inode);
3877 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3878 struct btrfs_root *root,
3879 struct inode *inode)
3883 ret = btrfs_update_inode(trans, root, inode);
3885 return btrfs_update_inode_item(trans, root, inode);
3890 * unlink helper that gets used here in inode.c and in the tree logging
3891 * recovery code. It remove a link in a directory with a given name, and
3892 * also drops the back refs in the inode to the directory
3894 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3895 struct btrfs_root *root,
3896 struct btrfs_inode *dir,
3897 struct btrfs_inode *inode,
3898 const char *name, int name_len)
3900 struct btrfs_fs_info *fs_info = root->fs_info;
3901 struct btrfs_path *path;
3903 struct extent_buffer *leaf;
3904 struct btrfs_dir_item *di;
3905 struct btrfs_key key;
3907 u64 ino = btrfs_ino(inode);
3908 u64 dir_ino = btrfs_ino(dir);
3910 path = btrfs_alloc_path();
3916 path->leave_spinning = 1;
3917 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3918 name, name_len, -1);
3927 leaf = path->nodes[0];
3928 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3929 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3932 btrfs_release_path(path);
3935 * If we don't have dir index, we have to get it by looking up
3936 * the inode ref, since we get the inode ref, remove it directly,
3937 * it is unnecessary to do delayed deletion.
3939 * But if we have dir index, needn't search inode ref to get it.
3940 * Since the inode ref is close to the inode item, it is better
3941 * that we delay to delete it, and just do this deletion when
3942 * we update the inode item.
3944 if (inode->dir_index) {
3945 ret = btrfs_delayed_delete_inode_ref(inode);
3947 index = inode->dir_index;
3952 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3956 "failed to delete reference to %.*s, inode %llu parent %llu",
3957 name_len, name, ino, dir_ino);
3958 btrfs_abort_transaction(trans, ret);
3962 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
3964 btrfs_abort_transaction(trans, ret);
3968 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3970 if (ret != 0 && ret != -ENOENT) {
3971 btrfs_abort_transaction(trans, ret);
3975 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3980 btrfs_abort_transaction(trans, ret);
3982 btrfs_free_path(path);
3986 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3987 inode_inc_iversion(&inode->vfs_inode);
3988 inode_inc_iversion(&dir->vfs_inode);
3989 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3990 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3991 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3996 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3997 struct btrfs_root *root,
3998 struct btrfs_inode *dir, struct btrfs_inode *inode,
3999 const char *name, int name_len)
4002 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4004 drop_nlink(&inode->vfs_inode);
4005 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4011 * helper to start transaction for unlink and rmdir.
4013 * unlink and rmdir are special in btrfs, they do not always free space, so
4014 * if we cannot make our reservations the normal way try and see if there is
4015 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4016 * allow the unlink to occur.
4018 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4020 struct btrfs_root *root = BTRFS_I(dir)->root;
4023 * 1 for the possible orphan item
4024 * 1 for the dir item
4025 * 1 for the dir index
4026 * 1 for the inode ref
4029 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4032 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4034 struct btrfs_root *root = BTRFS_I(dir)->root;
4035 struct btrfs_trans_handle *trans;
4036 struct inode *inode = d_inode(dentry);
4039 trans = __unlink_start_trans(dir);
4041 return PTR_ERR(trans);
4043 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4046 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4047 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4048 dentry->d_name.len);
4052 if (inode->i_nlink == 0) {
4053 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4059 btrfs_end_transaction(trans);
4060 btrfs_btree_balance_dirty(root->fs_info);
4064 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4065 struct btrfs_root *root,
4066 struct inode *dir, u64 objectid,
4067 const char *name, int name_len)
4069 struct btrfs_fs_info *fs_info = root->fs_info;
4070 struct btrfs_path *path;
4071 struct extent_buffer *leaf;
4072 struct btrfs_dir_item *di;
4073 struct btrfs_key key;
4076 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4078 path = btrfs_alloc_path();
4082 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4083 name, name_len, -1);
4084 if (IS_ERR_OR_NULL(di)) {
4092 leaf = path->nodes[0];
4093 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4094 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4095 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4097 btrfs_abort_transaction(trans, ret);
4100 btrfs_release_path(path);
4102 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4103 root->root_key.objectid, dir_ino,
4104 &index, name, name_len);
4106 if (ret != -ENOENT) {
4107 btrfs_abort_transaction(trans, ret);
4110 di = btrfs_search_dir_index_item(root, path, dir_ino,
4112 if (IS_ERR_OR_NULL(di)) {
4117 btrfs_abort_transaction(trans, ret);
4121 leaf = path->nodes[0];
4122 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4123 btrfs_release_path(path);
4126 btrfs_release_path(path);
4128 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4130 btrfs_abort_transaction(trans, ret);
4134 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4135 inode_inc_iversion(dir);
4136 dir->i_mtime = dir->i_ctime = current_time(dir);
4137 ret = btrfs_update_inode_fallback(trans, root, dir);
4139 btrfs_abort_transaction(trans, ret);
4141 btrfs_free_path(path);
4145 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4147 struct inode *inode = d_inode(dentry);
4149 struct btrfs_root *root = BTRFS_I(dir)->root;
4150 struct btrfs_trans_handle *trans;
4151 u64 last_unlink_trans;
4153 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4155 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4158 trans = __unlink_start_trans(dir);
4160 return PTR_ERR(trans);
4162 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4163 err = btrfs_unlink_subvol(trans, root, dir,
4164 BTRFS_I(inode)->location.objectid,
4165 dentry->d_name.name,
4166 dentry->d_name.len);
4170 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4174 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4176 /* now the directory is empty */
4177 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4178 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4179 dentry->d_name.len);
4181 btrfs_i_size_write(BTRFS_I(inode), 0);
4183 * Propagate the last_unlink_trans value of the deleted dir to
4184 * its parent directory. This is to prevent an unrecoverable
4185 * log tree in the case we do something like this:
4187 * 2) create snapshot under dir foo
4188 * 3) delete the snapshot
4191 * 6) fsync foo or some file inside foo
4193 if (last_unlink_trans >= trans->transid)
4194 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4197 btrfs_end_transaction(trans);
4198 btrfs_btree_balance_dirty(root->fs_info);
4203 static int truncate_space_check(struct btrfs_trans_handle *trans,
4204 struct btrfs_root *root,
4207 struct btrfs_fs_info *fs_info = root->fs_info;
4211 * This is only used to apply pressure to the enospc system, we don't
4212 * intend to use this reservation at all.
4214 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4215 bytes_deleted *= fs_info->nodesize;
4216 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4217 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4219 trace_btrfs_space_reservation(fs_info, "transaction",
4222 trans->bytes_reserved += bytes_deleted;
4228 static int truncate_inline_extent(struct inode *inode,
4229 struct btrfs_path *path,
4230 struct btrfs_key *found_key,
4234 struct extent_buffer *leaf = path->nodes[0];
4235 int slot = path->slots[0];
4236 struct btrfs_file_extent_item *fi;
4237 u32 size = (u32)(new_size - found_key->offset);
4238 struct btrfs_root *root = BTRFS_I(inode)->root;
4240 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4242 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4243 loff_t offset = new_size;
4244 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4247 * Zero out the remaining of the last page of our inline extent,
4248 * instead of directly truncating our inline extent here - that
4249 * would be much more complex (decompressing all the data, then
4250 * compressing the truncated data, which might be bigger than
4251 * the size of the inline extent, resize the extent, etc).
4252 * We release the path because to get the page we might need to
4253 * read the extent item from disk (data not in the page cache).
4255 btrfs_release_path(path);
4256 return btrfs_truncate_block(inode, offset, page_end - offset,
4260 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4261 size = btrfs_file_extent_calc_inline_size(size);
4262 btrfs_truncate_item(root->fs_info, path, size, 1);
4264 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4265 inode_sub_bytes(inode, item_end + 1 - new_size);
4271 * this can truncate away extent items, csum items and directory items.
4272 * It starts at a high offset and removes keys until it can't find
4273 * any higher than new_size
4275 * csum items that cross the new i_size are truncated to the new size
4278 * min_type is the minimum key type to truncate down to. If set to 0, this
4279 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4281 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4282 struct btrfs_root *root,
4283 struct inode *inode,
4284 u64 new_size, u32 min_type)
4286 struct btrfs_fs_info *fs_info = root->fs_info;
4287 struct btrfs_path *path;
4288 struct extent_buffer *leaf;
4289 struct btrfs_file_extent_item *fi;
4290 struct btrfs_key key;
4291 struct btrfs_key found_key;
4292 u64 extent_start = 0;
4293 u64 extent_num_bytes = 0;
4294 u64 extent_offset = 0;
4296 u64 last_size = new_size;
4297 u32 found_type = (u8)-1;
4300 int pending_del_nr = 0;
4301 int pending_del_slot = 0;
4302 int extent_type = -1;
4305 u64 ino = btrfs_ino(BTRFS_I(inode));
4306 u64 bytes_deleted = 0;
4308 bool should_throttle = 0;
4309 bool should_end = 0;
4311 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4314 * for non-free space inodes and ref cows, we want to back off from
4317 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4318 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4321 path = btrfs_alloc_path();
4324 path->reada = READA_BACK;
4327 * We want to drop from the next block forward in case this new size is
4328 * not block aligned since we will be keeping the last block of the
4329 * extent just the way it is.
4331 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4332 root == fs_info->tree_root)
4333 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4334 fs_info->sectorsize),
4338 * This function is also used to drop the items in the log tree before
4339 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4340 * it is used to drop the loged items. So we shouldn't kill the delayed
4343 if (min_type == 0 && root == BTRFS_I(inode)->root)
4344 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4347 key.offset = (u64)-1;
4352 * with a 16K leaf size and 128MB extents, you can actually queue
4353 * up a huge file in a single leaf. Most of the time that
4354 * bytes_deleted is > 0, it will be huge by the time we get here
4356 if (be_nice && bytes_deleted > SZ_32M) {
4357 if (btrfs_should_end_transaction(trans)) {
4364 path->leave_spinning = 1;
4365 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4372 /* there are no items in the tree for us to truncate, we're
4375 if (path->slots[0] == 0)
4382 leaf = path->nodes[0];
4383 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4384 found_type = found_key.type;
4386 if (found_key.objectid != ino)
4389 if (found_type < min_type)
4392 item_end = found_key.offset;
4393 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4394 fi = btrfs_item_ptr(leaf, path->slots[0],
4395 struct btrfs_file_extent_item);
4396 extent_type = btrfs_file_extent_type(leaf, fi);
4397 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4399 btrfs_file_extent_num_bytes(leaf, fi);
4400 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4401 item_end += btrfs_file_extent_inline_len(leaf,
4402 path->slots[0], fi);
4406 if (found_type > min_type) {
4409 if (item_end < new_size) {
4411 * With NO_HOLES mode, for the following mapping
4413 * [0-4k][hole][8k-12k]
4415 * if truncating isize down to 6k, it ends up
4418 if (btrfs_fs_incompat(root->fs_info, NO_HOLES))
4419 last_size = new_size;
4422 if (found_key.offset >= new_size)
4428 /* FIXME, shrink the extent if the ref count is only 1 */
4429 if (found_type != BTRFS_EXTENT_DATA_KEY)
4433 last_size = found_key.offset;
4435 last_size = new_size;
4437 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4439 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4441 u64 orig_num_bytes =
4442 btrfs_file_extent_num_bytes(leaf, fi);
4443 extent_num_bytes = ALIGN(new_size -
4445 fs_info->sectorsize);
4446 btrfs_set_file_extent_num_bytes(leaf, fi,
4448 num_dec = (orig_num_bytes -
4450 if (test_bit(BTRFS_ROOT_REF_COWS,
4453 inode_sub_bytes(inode, num_dec);
4454 btrfs_mark_buffer_dirty(leaf);
4457 btrfs_file_extent_disk_num_bytes(leaf,
4459 extent_offset = found_key.offset -
4460 btrfs_file_extent_offset(leaf, fi);
4462 /* FIXME blocksize != 4096 */
4463 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4464 if (extent_start != 0) {
4466 if (test_bit(BTRFS_ROOT_REF_COWS,
4468 inode_sub_bytes(inode, num_dec);
4471 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4473 * we can't truncate inline items that have had
4477 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4478 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4481 * Need to release path in order to truncate a
4482 * compressed extent. So delete any accumulated
4483 * extent items so far.
4485 if (btrfs_file_extent_compression(leaf, fi) !=
4486 BTRFS_COMPRESS_NONE && pending_del_nr) {
4487 err = btrfs_del_items(trans, root, path,
4491 btrfs_abort_transaction(trans,
4498 err = truncate_inline_extent(inode, path,
4503 btrfs_abort_transaction(trans, err);
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 == fs_info->tree_root)) {
4533 btrfs_set_path_blocking(path);
4534 bytes_deleted += extent_num_bytes;
4535 ret = btrfs_free_extent(trans, fs_info, extent_start,
4536 extent_num_bytes, 0,
4537 btrfs_header_owner(leaf),
4538 ino, extent_offset);
4540 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4541 btrfs_async_run_delayed_refs(fs_info,
4542 trans->delayed_ref_updates * 2,
4545 if (truncate_space_check(trans, root,
4546 extent_num_bytes)) {
4549 if (btrfs_should_throttle_delayed_refs(trans,
4551 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, ret);
4571 btrfs_release_path(path);
4572 if (should_throttle) {
4573 unsigned long updates = trans->delayed_ref_updates;
4575 trans->delayed_ref_updates = 0;
4576 ret = btrfs_run_delayed_refs(trans,
4584 * if we failed to refill our space rsv, bail out
4585 * and let the transaction restart
4597 if (pending_del_nr) {
4598 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4601 btrfs_abort_transaction(trans, ret);
4604 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4605 btrfs_ordered_update_i_size(inode, last_size, NULL);
4607 btrfs_free_path(path);
4610 /* only inline file may have last_size != new_size */
4611 if (new_size >= fs_info->sectorsize ||
4612 new_size > fs_info->max_inline)
4613 ASSERT(last_size == new_size);
4616 if (be_nice && bytes_deleted > SZ_32M) {
4617 unsigned long updates = trans->delayed_ref_updates;
4619 trans->delayed_ref_updates = 0;
4620 ret = btrfs_run_delayed_refs(trans, fs_info,
4630 * btrfs_truncate_block - read, zero a chunk and write a block
4631 * @inode - inode that we're zeroing
4632 * @from - the offset to start zeroing
4633 * @len - the length to zero, 0 to zero the entire range respective to the
4635 * @front - zero up to the offset instead of from the offset on
4637 * This will find the block for the "from" offset and cow the block and zero the
4638 * part we want to zero. This is used with truncate and hole punching.
4640 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4643 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4644 struct address_space *mapping = inode->i_mapping;
4645 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4646 struct btrfs_ordered_extent *ordered;
4647 struct extent_state *cached_state = NULL;
4649 u32 blocksize = fs_info->sectorsize;
4650 pgoff_t index = from >> PAGE_SHIFT;
4651 unsigned offset = from & (blocksize - 1);
4653 gfp_t mask = btrfs_alloc_write_mask(mapping);
4658 if ((offset & (blocksize - 1)) == 0 &&
4659 (!len || ((len & (blocksize - 1)) == 0)))
4662 ret = btrfs_delalloc_reserve_space(inode,
4663 round_down(from, blocksize), blocksize);
4668 page = find_or_create_page(mapping, index, mask);
4670 btrfs_delalloc_release_space(inode,
4671 round_down(from, blocksize),
4677 block_start = round_down(from, blocksize);
4678 block_end = block_start + blocksize - 1;
4680 if (!PageUptodate(page)) {
4681 ret = btrfs_readpage(NULL, page);
4683 if (page->mapping != mapping) {
4688 if (!PageUptodate(page)) {
4693 wait_on_page_writeback(page);
4695 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4696 set_page_extent_mapped(page);
4698 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4700 unlock_extent_cached(io_tree, block_start, block_end,
4701 &cached_state, GFP_NOFS);
4704 btrfs_start_ordered_extent(inode, ordered, 1);
4705 btrfs_put_ordered_extent(ordered);
4709 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4710 EXTENT_DIRTY | EXTENT_DELALLOC |
4711 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4712 0, 0, &cached_state, GFP_NOFS);
4714 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4717 unlock_extent_cached(io_tree, block_start, block_end,
4718 &cached_state, GFP_NOFS);
4722 if (offset != blocksize) {
4724 len = blocksize - offset;
4727 memset(kaddr + (block_start - page_offset(page)),
4730 memset(kaddr + (block_start - page_offset(page)) + offset,
4732 flush_dcache_page(page);
4735 ClearPageChecked(page);
4736 set_page_dirty(page);
4737 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4742 btrfs_delalloc_release_space(inode, block_start,
4750 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4751 u64 offset, u64 len)
4753 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4754 struct btrfs_trans_handle *trans;
4758 * Still need to make sure the inode looks like it's been updated so
4759 * that any holes get logged if we fsync.
4761 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4762 BTRFS_I(inode)->last_trans = fs_info->generation;
4763 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4764 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4769 * 1 - for the one we're dropping
4770 * 1 - for the one we're adding
4771 * 1 - for updating the inode.
4773 trans = btrfs_start_transaction(root, 3);
4775 return PTR_ERR(trans);
4777 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4779 btrfs_abort_transaction(trans, ret);
4780 btrfs_end_transaction(trans);
4784 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4785 offset, 0, 0, len, 0, len, 0, 0, 0);
4787 btrfs_abort_transaction(trans, ret);
4789 btrfs_update_inode(trans, root, inode);
4790 btrfs_end_transaction(trans);
4795 * This function puts in dummy file extents for the area we're creating a hole
4796 * for. So if we are truncating this file to a larger size we need to insert
4797 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4798 * the range between oldsize and size
4800 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4802 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4803 struct btrfs_root *root = BTRFS_I(inode)->root;
4804 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4805 struct extent_map *em = NULL;
4806 struct extent_state *cached_state = NULL;
4807 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4808 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4809 u64 block_end = ALIGN(size, fs_info->sectorsize);
4816 * If our size started in the middle of a block we need to zero out the
4817 * rest of the block before we expand the i_size, otherwise we could
4818 * expose stale data.
4820 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4824 if (size <= hole_start)
4828 struct btrfs_ordered_extent *ordered;
4830 lock_extent_bits(io_tree, hole_start, block_end - 1,
4832 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4833 block_end - hole_start);
4836 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4837 &cached_state, GFP_NOFS);
4838 btrfs_start_ordered_extent(inode, ordered, 1);
4839 btrfs_put_ordered_extent(ordered);
4842 cur_offset = hole_start;
4844 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4845 block_end - cur_offset, 0);
4851 last_byte = min(extent_map_end(em), block_end);
4852 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4853 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4854 struct extent_map *hole_em;
4855 hole_size = last_byte - cur_offset;
4857 err = maybe_insert_hole(root, inode, cur_offset,
4861 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4862 cur_offset + hole_size - 1, 0);
4863 hole_em = alloc_extent_map();
4865 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4866 &BTRFS_I(inode)->runtime_flags);
4869 hole_em->start = cur_offset;
4870 hole_em->len = hole_size;
4871 hole_em->orig_start = cur_offset;
4873 hole_em->block_start = EXTENT_MAP_HOLE;
4874 hole_em->block_len = 0;
4875 hole_em->orig_block_len = 0;
4876 hole_em->ram_bytes = hole_size;
4877 hole_em->bdev = fs_info->fs_devices->latest_bdev;
4878 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4879 hole_em->generation = fs_info->generation;
4882 write_lock(&em_tree->lock);
4883 err = add_extent_mapping(em_tree, hole_em, 1);
4884 write_unlock(&em_tree->lock);
4887 btrfs_drop_extent_cache(BTRFS_I(inode),
4892 free_extent_map(hole_em);
4895 free_extent_map(em);
4897 cur_offset = last_byte;
4898 if (cur_offset >= block_end)
4901 free_extent_map(em);
4902 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4907 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4909 struct btrfs_root *root = BTRFS_I(inode)->root;
4910 struct btrfs_trans_handle *trans;
4911 loff_t oldsize = i_size_read(inode);
4912 loff_t newsize = attr->ia_size;
4913 int mask = attr->ia_valid;
4917 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4918 * special case where we need to update the times despite not having
4919 * these flags set. For all other operations the VFS set these flags
4920 * explicitly if it wants a timestamp update.
4922 if (newsize != oldsize) {
4923 inode_inc_iversion(inode);
4924 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4925 inode->i_ctime = inode->i_mtime =
4926 current_time(inode);
4929 if (newsize > oldsize) {
4931 * Don't do an expanding truncate while snapshoting is ongoing.
4932 * This is to ensure the snapshot captures a fully consistent
4933 * state of this file - if the snapshot captures this expanding
4934 * truncation, it must capture all writes that happened before
4937 btrfs_wait_for_snapshot_creation(root);
4938 ret = btrfs_cont_expand(inode, oldsize, newsize);
4940 btrfs_end_write_no_snapshoting(root);
4944 trans = btrfs_start_transaction(root, 1);
4945 if (IS_ERR(trans)) {
4946 btrfs_end_write_no_snapshoting(root);
4947 return PTR_ERR(trans);
4950 i_size_write(inode, newsize);
4951 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4952 pagecache_isize_extended(inode, oldsize, newsize);
4953 ret = btrfs_update_inode(trans, root, inode);
4954 btrfs_end_write_no_snapshoting(root);
4955 btrfs_end_transaction(trans);
4959 * We're truncating a file that used to have good data down to
4960 * zero. Make sure it gets into the ordered flush list so that
4961 * any new writes get down to disk quickly.
4964 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4965 &BTRFS_I(inode)->runtime_flags);
4968 * 1 for the orphan item we're going to add
4969 * 1 for the orphan item deletion.
4971 trans = btrfs_start_transaction(root, 2);
4973 return PTR_ERR(trans);
4976 * We need to do this in case we fail at _any_ point during the
4977 * actual truncate. Once we do the truncate_setsize we could
4978 * invalidate pages which forces any outstanding ordered io to
4979 * be instantly completed which will give us extents that need
4980 * to be truncated. If we fail to get an orphan inode down we
4981 * could have left over extents that were never meant to live,
4982 * so we need to guarantee from this point on that everything
4983 * will be consistent.
4985 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4986 btrfs_end_transaction(trans);
4990 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4991 truncate_setsize(inode, newsize);
4993 /* Disable nonlocked read DIO to avoid the end less truncate */
4994 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
4995 inode_dio_wait(inode);
4996 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
4998 ret = btrfs_truncate(inode);
4999 if (ret && inode->i_nlink) {
5002 /* To get a stable disk_i_size */
5003 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5005 btrfs_orphan_del(NULL, BTRFS_I(inode));
5010 * failed to truncate, disk_i_size is only adjusted down
5011 * as we remove extents, so it should represent the true
5012 * size of the inode, so reset the in memory size and
5013 * delete our orphan entry.
5015 trans = btrfs_join_transaction(root);
5016 if (IS_ERR(trans)) {
5017 btrfs_orphan_del(NULL, BTRFS_I(inode));
5020 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5021 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5023 btrfs_abort_transaction(trans, err);
5024 btrfs_end_transaction(trans);
5031 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5033 struct inode *inode = d_inode(dentry);
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5037 if (btrfs_root_readonly(root))
5040 err = setattr_prepare(dentry, attr);
5044 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5045 err = btrfs_setsize(inode, attr);
5050 if (attr->ia_valid) {
5051 setattr_copy(inode, attr);
5052 inode_inc_iversion(inode);
5053 err = btrfs_dirty_inode(inode);
5055 if (!err && attr->ia_valid & ATTR_MODE)
5056 err = posix_acl_chmod(inode, inode->i_mode);
5063 * While truncating the inode pages during eviction, we get the VFS calling
5064 * btrfs_invalidatepage() against each page of the inode. This is slow because
5065 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5066 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5067 * extent_state structures over and over, wasting lots of time.
5069 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5070 * those expensive operations on a per page basis and do only the ordered io
5071 * finishing, while we release here the extent_map and extent_state structures,
5072 * without the excessive merging and splitting.
5074 static void evict_inode_truncate_pages(struct inode *inode)
5076 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5077 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5078 struct rb_node *node;
5080 ASSERT(inode->i_state & I_FREEING);
5081 truncate_inode_pages_final(&inode->i_data);
5083 write_lock(&map_tree->lock);
5084 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5085 struct extent_map *em;
5087 node = rb_first(&map_tree->map);
5088 em = rb_entry(node, struct extent_map, rb_node);
5089 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5090 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5091 remove_extent_mapping(map_tree, em);
5092 free_extent_map(em);
5093 if (need_resched()) {
5094 write_unlock(&map_tree->lock);
5096 write_lock(&map_tree->lock);
5099 write_unlock(&map_tree->lock);
5102 * Keep looping until we have no more ranges in the io tree.
5103 * We can have ongoing bios started by readpages (called from readahead)
5104 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5105 * still in progress (unlocked the pages in the bio but did not yet
5106 * unlocked the ranges in the io tree). Therefore this means some
5107 * ranges can still be locked and eviction started because before
5108 * submitting those bios, which are executed by a separate task (work
5109 * queue kthread), inode references (inode->i_count) were not taken
5110 * (which would be dropped in the end io callback of each bio).
5111 * Therefore here we effectively end up waiting for those bios and
5112 * anyone else holding locked ranges without having bumped the inode's
5113 * reference count - if we don't do it, when they access the inode's
5114 * io_tree to unlock a range it may be too late, leading to an
5115 * use-after-free issue.
5117 spin_lock(&io_tree->lock);
5118 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5119 struct extent_state *state;
5120 struct extent_state *cached_state = NULL;
5124 node = rb_first(&io_tree->state);
5125 state = rb_entry(node, struct extent_state, rb_node);
5126 start = state->start;
5128 spin_unlock(&io_tree->lock);
5130 lock_extent_bits(io_tree, start, end, &cached_state);
5133 * If still has DELALLOC flag, the extent didn't reach disk,
5134 * and its reserved space won't be freed by delayed_ref.
5135 * So we need to free its reserved space here.
5136 * (Refer to comment in btrfs_invalidatepage, case 2)
5138 * Note, end is the bytenr of last byte, so we need + 1 here.
5140 if (state->state & EXTENT_DELALLOC)
5141 btrfs_qgroup_free_data(inode, start, end - start + 1);
5143 clear_extent_bit(io_tree, start, end,
5144 EXTENT_LOCKED | EXTENT_DIRTY |
5145 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5146 EXTENT_DEFRAG, 1, 1,
5147 &cached_state, GFP_NOFS);
5150 spin_lock(&io_tree->lock);
5152 spin_unlock(&io_tree->lock);
5155 void btrfs_evict_inode(struct inode *inode)
5157 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5158 struct btrfs_trans_handle *trans;
5159 struct btrfs_root *root = BTRFS_I(inode)->root;
5160 struct btrfs_block_rsv *rsv, *global_rsv;
5161 int steal_from_global = 0;
5165 trace_btrfs_inode_evict(inode);
5168 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5172 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5174 evict_inode_truncate_pages(inode);
5176 if (inode->i_nlink &&
5177 ((btrfs_root_refs(&root->root_item) != 0 &&
5178 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5179 btrfs_is_free_space_inode(BTRFS_I(inode))))
5182 if (is_bad_inode(inode)) {
5183 btrfs_orphan_del(NULL, BTRFS_I(inode));
5186 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5187 if (!special_file(inode->i_mode))
5188 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5190 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5192 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5193 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5194 &BTRFS_I(inode)->runtime_flags));
5198 if (inode->i_nlink > 0) {
5199 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5200 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5204 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5206 btrfs_orphan_del(NULL, BTRFS_I(inode));
5210 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5212 btrfs_orphan_del(NULL, BTRFS_I(inode));
5215 rsv->size = min_size;
5217 global_rsv = &fs_info->global_block_rsv;
5219 btrfs_i_size_write(BTRFS_I(inode), 0);
5222 * This is a bit simpler than btrfs_truncate since we've already
5223 * reserved our space for our orphan item in the unlink, so we just
5224 * need to reserve some slack space in case we add bytes and update
5225 * inode item when doing the truncate.
5228 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5229 BTRFS_RESERVE_FLUSH_LIMIT);
5232 * Try and steal from the global reserve since we will
5233 * likely not use this space anyway, we want to try as
5234 * hard as possible to get this to work.
5237 steal_from_global++;
5239 steal_from_global = 0;
5243 * steal_from_global == 0: we reserved stuff, hooray!
5244 * steal_from_global == 1: we didn't reserve stuff, boo!
5245 * steal_from_global == 2: we've committed, still not a lot of
5246 * room but maybe we'll have room in the global reserve this
5248 * steal_from_global == 3: abandon all hope!
5250 if (steal_from_global > 2) {
5252 "Could not get space for a delete, will truncate on mount %d",
5254 btrfs_orphan_del(NULL, BTRFS_I(inode));
5255 btrfs_free_block_rsv(fs_info, rsv);
5259 trans = btrfs_join_transaction(root);
5260 if (IS_ERR(trans)) {
5261 btrfs_orphan_del(NULL, BTRFS_I(inode));
5262 btrfs_free_block_rsv(fs_info, rsv);
5267 * We can't just steal from the global reserve, we need to make
5268 * sure there is room to do it, if not we need to commit and try
5271 if (steal_from_global) {
5272 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5273 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5280 * Couldn't steal from the global reserve, we have too much
5281 * pending stuff built up, commit the transaction and try it
5285 ret = btrfs_commit_transaction(trans);
5287 btrfs_orphan_del(NULL, BTRFS_I(inode));
5288 btrfs_free_block_rsv(fs_info, rsv);
5293 steal_from_global = 0;
5296 trans->block_rsv = rsv;
5298 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5299 if (ret != -ENOSPC && ret != -EAGAIN)
5302 trans->block_rsv = &fs_info->trans_block_rsv;
5303 btrfs_end_transaction(trans);
5305 btrfs_btree_balance_dirty(fs_info);
5308 btrfs_free_block_rsv(fs_info, rsv);
5311 * Errors here aren't a big deal, it just means we leave orphan items
5312 * in the tree. They will be cleaned up on the next mount.
5315 trans->block_rsv = root->orphan_block_rsv;
5316 btrfs_orphan_del(trans, BTRFS_I(inode));
5318 btrfs_orphan_del(NULL, BTRFS_I(inode));
5321 trans->block_rsv = &fs_info->trans_block_rsv;
5322 if (!(root == fs_info->tree_root ||
5323 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5324 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5326 btrfs_end_transaction(trans);
5327 btrfs_btree_balance_dirty(fs_info);
5329 btrfs_remove_delayed_node(BTRFS_I(inode));
5334 * this returns the key found in the dir entry in the location pointer.
5335 * If no dir entries were found, location->objectid is 0.
5337 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5338 struct btrfs_key *location)
5340 const char *name = dentry->d_name.name;
5341 int namelen = dentry->d_name.len;
5342 struct btrfs_dir_item *di;
5343 struct btrfs_path *path;
5344 struct btrfs_root *root = BTRFS_I(dir)->root;
5347 path = btrfs_alloc_path();
5351 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5356 if (IS_ERR_OR_NULL(di))
5359 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5361 btrfs_free_path(path);
5364 location->objectid = 0;
5369 * when we hit a tree root in a directory, the btrfs part of the inode
5370 * needs to be changed to reflect the root directory of the tree root. This
5371 * is kind of like crossing a mount point.
5373 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5375 struct dentry *dentry,
5376 struct btrfs_key *location,
5377 struct btrfs_root **sub_root)
5379 struct btrfs_path *path;
5380 struct btrfs_root *new_root;
5381 struct btrfs_root_ref *ref;
5382 struct extent_buffer *leaf;
5383 struct btrfs_key key;
5387 path = btrfs_alloc_path();
5394 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5395 key.type = BTRFS_ROOT_REF_KEY;
5396 key.offset = location->objectid;
5398 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5405 leaf = path->nodes[0];
5406 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5407 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5408 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5411 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5412 (unsigned long)(ref + 1),
5413 dentry->d_name.len);
5417 btrfs_release_path(path);
5419 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5420 if (IS_ERR(new_root)) {
5421 err = PTR_ERR(new_root);
5425 *sub_root = new_root;
5426 location->objectid = btrfs_root_dirid(&new_root->root_item);
5427 location->type = BTRFS_INODE_ITEM_KEY;
5428 location->offset = 0;
5431 btrfs_free_path(path);
5435 static void inode_tree_add(struct inode *inode)
5437 struct btrfs_root *root = BTRFS_I(inode)->root;
5438 struct btrfs_inode *entry;
5440 struct rb_node *parent;
5441 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5442 u64 ino = btrfs_ino(BTRFS_I(inode));
5444 if (inode_unhashed(inode))
5447 spin_lock(&root->inode_lock);
5448 p = &root->inode_tree.rb_node;
5451 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5453 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5454 p = &parent->rb_left;
5455 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5456 p = &parent->rb_right;
5458 WARN_ON(!(entry->vfs_inode.i_state &
5459 (I_WILL_FREE | I_FREEING)));
5460 rb_replace_node(parent, new, &root->inode_tree);
5461 RB_CLEAR_NODE(parent);
5462 spin_unlock(&root->inode_lock);
5466 rb_link_node(new, parent, p);
5467 rb_insert_color(new, &root->inode_tree);
5468 spin_unlock(&root->inode_lock);
5471 static void inode_tree_del(struct inode *inode)
5473 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5474 struct btrfs_root *root = BTRFS_I(inode)->root;
5477 spin_lock(&root->inode_lock);
5478 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5479 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5480 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5481 empty = RB_EMPTY_ROOT(&root->inode_tree);
5483 spin_unlock(&root->inode_lock);
5485 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5486 synchronize_srcu(&fs_info->subvol_srcu);
5487 spin_lock(&root->inode_lock);
5488 empty = RB_EMPTY_ROOT(&root->inode_tree);
5489 spin_unlock(&root->inode_lock);
5491 btrfs_add_dead_root(root);
5495 void btrfs_invalidate_inodes(struct btrfs_root *root)
5497 struct btrfs_fs_info *fs_info = root->fs_info;
5498 struct rb_node *node;
5499 struct rb_node *prev;
5500 struct btrfs_inode *entry;
5501 struct inode *inode;
5504 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5505 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5507 spin_lock(&root->inode_lock);
5509 node = root->inode_tree.rb_node;
5513 entry = rb_entry(node, struct btrfs_inode, rb_node);
5515 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5516 node = node->rb_left;
5517 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5518 node = node->rb_right;
5524 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5525 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5529 prev = rb_next(prev);
5533 entry = rb_entry(node, struct btrfs_inode, rb_node);
5534 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5535 inode = igrab(&entry->vfs_inode);
5537 spin_unlock(&root->inode_lock);
5538 if (atomic_read(&inode->i_count) > 1)
5539 d_prune_aliases(inode);
5541 * btrfs_drop_inode will have it removed from
5542 * the inode cache when its usage count
5547 spin_lock(&root->inode_lock);
5551 if (cond_resched_lock(&root->inode_lock))
5554 node = rb_next(node);
5556 spin_unlock(&root->inode_lock);
5559 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5561 struct btrfs_iget_args *args = p;
5562 inode->i_ino = args->location->objectid;
5563 memcpy(&BTRFS_I(inode)->location, args->location,
5564 sizeof(*args->location));
5565 BTRFS_I(inode)->root = args->root;
5569 static int btrfs_find_actor(struct inode *inode, void *opaque)
5571 struct btrfs_iget_args *args = opaque;
5572 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5573 args->root == BTRFS_I(inode)->root;
5576 static struct inode *btrfs_iget_locked(struct super_block *s,
5577 struct btrfs_key *location,
5578 struct btrfs_root *root)
5580 struct inode *inode;
5581 struct btrfs_iget_args args;
5582 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5584 args.location = location;
5587 inode = iget5_locked(s, hashval, btrfs_find_actor,
5588 btrfs_init_locked_inode,
5593 /* Get an inode object given its location and corresponding root.
5594 * Returns in *is_new if the inode was read from disk
5596 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5597 struct btrfs_root *root, int *new)
5599 struct inode *inode;
5601 inode = btrfs_iget_locked(s, location, root);
5603 return ERR_PTR(-ENOMEM);
5605 if (inode->i_state & I_NEW) {
5608 ret = btrfs_read_locked_inode(inode);
5609 if (!is_bad_inode(inode)) {
5610 inode_tree_add(inode);
5611 unlock_new_inode(inode);
5615 unlock_new_inode(inode);
5618 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5625 static struct inode *new_simple_dir(struct super_block *s,
5626 struct btrfs_key *key,
5627 struct btrfs_root *root)
5629 struct inode *inode = new_inode(s);
5632 return ERR_PTR(-ENOMEM);
5634 BTRFS_I(inode)->root = root;
5635 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5636 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5638 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5639 inode->i_op = &btrfs_dir_ro_inode_operations;
5640 inode->i_opflags &= ~IOP_XATTR;
5641 inode->i_fop = &simple_dir_operations;
5642 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5643 inode->i_mtime = current_time(inode);
5644 inode->i_atime = inode->i_mtime;
5645 inode->i_ctime = inode->i_mtime;
5646 BTRFS_I(inode)->i_otime = inode->i_mtime;
5651 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5653 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5654 struct inode *inode;
5655 struct btrfs_root *root = BTRFS_I(dir)->root;
5656 struct btrfs_root *sub_root = root;
5657 struct btrfs_key location;
5661 if (dentry->d_name.len > BTRFS_NAME_LEN)
5662 return ERR_PTR(-ENAMETOOLONG);
5664 ret = btrfs_inode_by_name(dir, dentry, &location);
5666 return ERR_PTR(ret);
5668 if (location.objectid == 0)
5669 return ERR_PTR(-ENOENT);
5671 if (location.type == BTRFS_INODE_ITEM_KEY) {
5672 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5676 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5678 index = srcu_read_lock(&fs_info->subvol_srcu);
5679 ret = fixup_tree_root_location(fs_info, dir, dentry,
5680 &location, &sub_root);
5683 inode = ERR_PTR(ret);
5685 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5687 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5689 srcu_read_unlock(&fs_info->subvol_srcu, index);
5691 if (!IS_ERR(inode) && root != sub_root) {
5692 down_read(&fs_info->cleanup_work_sem);
5693 if (!(inode->i_sb->s_flags & MS_RDONLY))
5694 ret = btrfs_orphan_cleanup(sub_root);
5695 up_read(&fs_info->cleanup_work_sem);
5698 inode = ERR_PTR(ret);
5705 static int btrfs_dentry_delete(const struct dentry *dentry)
5707 struct btrfs_root *root;
5708 struct inode *inode = d_inode(dentry);
5710 if (!inode && !IS_ROOT(dentry))
5711 inode = d_inode(dentry->d_parent);
5714 root = BTRFS_I(inode)->root;
5715 if (btrfs_root_refs(&root->root_item) == 0)
5718 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5724 static void btrfs_dentry_release(struct dentry *dentry)
5726 kfree(dentry->d_fsdata);
5729 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5732 struct inode *inode;
5734 inode = btrfs_lookup_dentry(dir, dentry);
5735 if (IS_ERR(inode)) {
5736 if (PTR_ERR(inode) == -ENOENT)
5739 return ERR_CAST(inode);
5742 return d_splice_alias(inode, dentry);
5745 unsigned char btrfs_filetype_table[] = {
5746 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5749 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5751 struct inode *inode = file_inode(file);
5752 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5753 struct btrfs_root *root = BTRFS_I(inode)->root;
5754 struct btrfs_item *item;
5755 struct btrfs_dir_item *di;
5756 struct btrfs_key key;
5757 struct btrfs_key found_key;
5758 struct btrfs_path *path;
5759 struct list_head ins_list;
5760 struct list_head del_list;
5762 struct extent_buffer *leaf;
5764 unsigned char d_type;
5770 struct btrfs_key location;
5772 if (!dir_emit_dots(file, ctx))
5775 path = btrfs_alloc_path();
5779 path->reada = READA_FORWARD;
5781 INIT_LIST_HEAD(&ins_list);
5782 INIT_LIST_HEAD(&del_list);
5783 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5785 key.type = BTRFS_DIR_INDEX_KEY;
5786 key.offset = ctx->pos;
5787 key.objectid = btrfs_ino(BTRFS_I(inode));
5789 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5794 leaf = path->nodes[0];
5795 slot = path->slots[0];
5796 if (slot >= btrfs_header_nritems(leaf)) {
5797 ret = btrfs_next_leaf(root, path);
5805 item = btrfs_item_nr(slot);
5806 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5808 if (found_key.objectid != key.objectid)
5810 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5812 if (found_key.offset < ctx->pos)
5814 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5817 ctx->pos = found_key.offset;
5819 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5820 if (verify_dir_item(fs_info, leaf, di))
5823 name_len = btrfs_dir_name_len(leaf, di);
5824 if (name_len <= sizeof(tmp_name)) {
5825 name_ptr = tmp_name;
5827 name_ptr = kmalloc(name_len, GFP_KERNEL);
5833 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5836 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5837 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5839 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5842 if (name_ptr != tmp_name)
5852 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5857 * Stop new entries from being returned after we return the last
5860 * New directory entries are assigned a strictly increasing
5861 * offset. This means that new entries created during readdir
5862 * are *guaranteed* to be seen in the future by that readdir.
5863 * This has broken buggy programs which operate on names as
5864 * they're returned by readdir. Until we re-use freed offsets
5865 * we have this hack to stop new entries from being returned
5866 * under the assumption that they'll never reach this huge
5869 * This is being careful not to overflow 32bit loff_t unless the
5870 * last entry requires it because doing so has broken 32bit apps
5873 if (ctx->pos >= INT_MAX)
5874 ctx->pos = LLONG_MAX;
5881 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5882 btrfs_free_path(path);
5886 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5888 struct btrfs_root *root = BTRFS_I(inode)->root;
5889 struct btrfs_trans_handle *trans;
5891 bool nolock = false;
5893 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5896 if (btrfs_fs_closing(root->fs_info) &&
5897 btrfs_is_free_space_inode(BTRFS_I(inode)))
5900 if (wbc->sync_mode == WB_SYNC_ALL) {
5902 trans = btrfs_join_transaction_nolock(root);
5904 trans = btrfs_join_transaction(root);
5906 return PTR_ERR(trans);
5907 ret = btrfs_commit_transaction(trans);
5913 * This is somewhat expensive, updating the tree every time the
5914 * inode changes. But, it is most likely to find the inode in cache.
5915 * FIXME, needs more benchmarking...there are no reasons other than performance
5916 * to keep or drop this code.
5918 static int btrfs_dirty_inode(struct inode *inode)
5920 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5921 struct btrfs_root *root = BTRFS_I(inode)->root;
5922 struct btrfs_trans_handle *trans;
5925 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5928 trans = btrfs_join_transaction(root);
5930 return PTR_ERR(trans);
5932 ret = btrfs_update_inode(trans, root, inode);
5933 if (ret && ret == -ENOSPC) {
5934 /* whoops, lets try again with the full transaction */
5935 btrfs_end_transaction(trans);
5936 trans = btrfs_start_transaction(root, 1);
5938 return PTR_ERR(trans);
5940 ret = btrfs_update_inode(trans, root, inode);
5942 btrfs_end_transaction(trans);
5943 if (BTRFS_I(inode)->delayed_node)
5944 btrfs_balance_delayed_items(fs_info);
5950 * This is a copy of file_update_time. We need this so we can return error on
5951 * ENOSPC for updating the inode in the case of file write and mmap writes.
5953 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5956 struct btrfs_root *root = BTRFS_I(inode)->root;
5958 if (btrfs_root_readonly(root))
5961 if (flags & S_VERSION)
5962 inode_inc_iversion(inode);
5963 if (flags & S_CTIME)
5964 inode->i_ctime = *now;
5965 if (flags & S_MTIME)
5966 inode->i_mtime = *now;
5967 if (flags & S_ATIME)
5968 inode->i_atime = *now;
5969 return btrfs_dirty_inode(inode);
5973 * find the highest existing sequence number in a directory
5974 * and then set the in-memory index_cnt variable to reflect
5975 * free sequence numbers
5977 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
5979 struct btrfs_root *root = inode->root;
5980 struct btrfs_key key, found_key;
5981 struct btrfs_path *path;
5982 struct extent_buffer *leaf;
5985 key.objectid = btrfs_ino(inode);
5986 key.type = BTRFS_DIR_INDEX_KEY;
5987 key.offset = (u64)-1;
5989 path = btrfs_alloc_path();
5993 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5996 /* FIXME: we should be able to handle this */
6002 * MAGIC NUMBER EXPLANATION:
6003 * since we search a directory based on f_pos we have to start at 2
6004 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6005 * else has to start at 2
6007 if (path->slots[0] == 0) {
6008 inode->index_cnt = 2;
6014 leaf = path->nodes[0];
6015 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6017 if (found_key.objectid != btrfs_ino(inode) ||
6018 found_key.type != BTRFS_DIR_INDEX_KEY) {
6019 inode->index_cnt = 2;
6023 inode->index_cnt = found_key.offset + 1;
6025 btrfs_free_path(path);
6030 * helper to find a free sequence number in a given directory. This current
6031 * code is very simple, later versions will do smarter things in the btree
6033 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6037 if (dir->index_cnt == (u64)-1) {
6038 ret = btrfs_inode_delayed_dir_index_count(dir);
6040 ret = btrfs_set_inode_index_count(dir);
6046 *index = dir->index_cnt;
6052 static int btrfs_insert_inode_locked(struct inode *inode)
6054 struct btrfs_iget_args args;
6055 args.location = &BTRFS_I(inode)->location;
6056 args.root = BTRFS_I(inode)->root;
6058 return insert_inode_locked4(inode,
6059 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6060 btrfs_find_actor, &args);
6063 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6064 struct btrfs_root *root,
6066 const char *name, int name_len,
6067 u64 ref_objectid, u64 objectid,
6068 umode_t mode, u64 *index)
6070 struct btrfs_fs_info *fs_info = root->fs_info;
6071 struct inode *inode;
6072 struct btrfs_inode_item *inode_item;
6073 struct btrfs_key *location;
6074 struct btrfs_path *path;
6075 struct btrfs_inode_ref *ref;
6076 struct btrfs_key key[2];
6078 int nitems = name ? 2 : 1;
6082 path = btrfs_alloc_path();
6084 return ERR_PTR(-ENOMEM);
6086 inode = new_inode(fs_info->sb);
6088 btrfs_free_path(path);
6089 return ERR_PTR(-ENOMEM);
6093 * O_TMPFILE, set link count to 0, so that after this point,
6094 * we fill in an inode item with the correct link count.
6097 set_nlink(inode, 0);
6100 * we have to initialize this early, so we can reclaim the inode
6101 * number if we fail afterwards in this function.
6103 inode->i_ino = objectid;
6106 trace_btrfs_inode_request(dir);
6108 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6110 btrfs_free_path(path);
6112 return ERR_PTR(ret);
6118 * index_cnt is ignored for everything but a dir,
6119 * btrfs_get_inode_index_count has an explanation for the magic
6122 BTRFS_I(inode)->index_cnt = 2;
6123 BTRFS_I(inode)->dir_index = *index;
6124 BTRFS_I(inode)->root = root;
6125 BTRFS_I(inode)->generation = trans->transid;
6126 inode->i_generation = BTRFS_I(inode)->generation;
6129 * We could have gotten an inode number from somebody who was fsynced
6130 * and then removed in this same transaction, so let's just set full
6131 * sync since it will be a full sync anyway and this will blow away the
6132 * old info in the log.
6134 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6136 key[0].objectid = objectid;
6137 key[0].type = BTRFS_INODE_ITEM_KEY;
6140 sizes[0] = sizeof(struct btrfs_inode_item);
6144 * Start new inodes with an inode_ref. This is slightly more
6145 * efficient for small numbers of hard links since they will
6146 * be packed into one item. Extended refs will kick in if we
6147 * add more hard links than can fit in the ref item.
6149 key[1].objectid = objectid;
6150 key[1].type = BTRFS_INODE_REF_KEY;
6151 key[1].offset = ref_objectid;
6153 sizes[1] = name_len + sizeof(*ref);
6156 location = &BTRFS_I(inode)->location;
6157 location->objectid = objectid;
6158 location->offset = 0;
6159 location->type = BTRFS_INODE_ITEM_KEY;
6161 ret = btrfs_insert_inode_locked(inode);
6165 path->leave_spinning = 1;
6166 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6170 inode_init_owner(inode, dir, mode);
6171 inode_set_bytes(inode, 0);
6173 inode->i_mtime = current_time(inode);
6174 inode->i_atime = inode->i_mtime;
6175 inode->i_ctime = inode->i_mtime;
6176 BTRFS_I(inode)->i_otime = inode->i_mtime;
6178 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6179 struct btrfs_inode_item);
6180 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6181 sizeof(*inode_item));
6182 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6185 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6186 struct btrfs_inode_ref);
6187 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6188 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6189 ptr = (unsigned long)(ref + 1);
6190 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6193 btrfs_mark_buffer_dirty(path->nodes[0]);
6194 btrfs_free_path(path);
6196 btrfs_inherit_iflags(inode, dir);
6198 if (S_ISREG(mode)) {
6199 if (btrfs_test_opt(fs_info, NODATASUM))
6200 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6201 if (btrfs_test_opt(fs_info, NODATACOW))
6202 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6203 BTRFS_INODE_NODATASUM;
6206 inode_tree_add(inode);
6208 trace_btrfs_inode_new(inode);
6209 btrfs_set_inode_last_trans(trans, inode);
6211 btrfs_update_root_times(trans, root);
6213 ret = btrfs_inode_inherit_props(trans, inode, dir);
6216 "error inheriting props for ino %llu (root %llu): %d",
6217 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6222 unlock_new_inode(inode);
6225 BTRFS_I(dir)->index_cnt--;
6226 btrfs_free_path(path);
6228 return ERR_PTR(ret);
6231 static inline u8 btrfs_inode_type(struct inode *inode)
6233 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6237 * utility function to add 'inode' into 'parent_inode' with
6238 * a give name and a given sequence number.
6239 * if 'add_backref' is true, also insert a backref from the
6240 * inode to the parent directory.
6242 int btrfs_add_link(struct btrfs_trans_handle *trans,
6243 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6244 const char *name, int name_len, int add_backref, u64 index)
6246 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6248 struct btrfs_key key;
6249 struct btrfs_root *root = parent_inode->root;
6250 u64 ino = btrfs_ino(inode);
6251 u64 parent_ino = btrfs_ino(parent_inode);
6253 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6254 memcpy(&key, &inode->root->root_key, sizeof(key));
6257 key.type = BTRFS_INODE_ITEM_KEY;
6261 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6262 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6263 root->root_key.objectid, parent_ino,
6264 index, name, name_len);
6265 } else if (add_backref) {
6266 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6270 /* Nothing to clean up yet */
6274 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6276 btrfs_inode_type(&inode->vfs_inode), index);
6277 if (ret == -EEXIST || ret == -EOVERFLOW)
6280 btrfs_abort_transaction(trans, ret);
6284 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6286 inode_inc_iversion(&parent_inode->vfs_inode);
6287 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6288 current_time(&parent_inode->vfs_inode);
6289 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6291 btrfs_abort_transaction(trans, ret);
6295 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6298 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6299 root->root_key.objectid, parent_ino,
6300 &local_index, name, name_len);
6302 } else if (add_backref) {
6306 err = btrfs_del_inode_ref(trans, root, name, name_len,
6307 ino, parent_ino, &local_index);
6312 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6313 struct btrfs_inode *dir, struct dentry *dentry,
6314 struct btrfs_inode *inode, int backref, u64 index)
6316 int err = btrfs_add_link(trans, dir, inode,
6317 dentry->d_name.name, dentry->d_name.len,
6324 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6325 umode_t mode, dev_t rdev)
6327 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6328 struct btrfs_trans_handle *trans;
6329 struct btrfs_root *root = BTRFS_I(dir)->root;
6330 struct inode *inode = NULL;
6337 * 2 for inode item and ref
6339 * 1 for xattr if selinux is on
6341 trans = btrfs_start_transaction(root, 5);
6343 return PTR_ERR(trans);
6345 err = btrfs_find_free_ino(root, &objectid);
6349 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6350 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6352 if (IS_ERR(inode)) {
6353 err = PTR_ERR(inode);
6358 * If the active LSM wants to access the inode during
6359 * d_instantiate it needs these. Smack checks to see
6360 * if the filesystem supports xattrs by looking at the
6363 inode->i_op = &btrfs_special_inode_operations;
6364 init_special_inode(inode, inode->i_mode, rdev);
6366 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6368 goto out_unlock_inode;
6370 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6373 goto out_unlock_inode;
6375 btrfs_update_inode(trans, root, inode);
6376 unlock_new_inode(inode);
6377 d_instantiate(dentry, inode);
6381 btrfs_end_transaction(trans);
6382 btrfs_balance_delayed_items(fs_info);
6383 btrfs_btree_balance_dirty(fs_info);
6385 inode_dec_link_count(inode);
6392 unlock_new_inode(inode);
6397 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6398 umode_t mode, bool excl)
6400 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6401 struct btrfs_trans_handle *trans;
6402 struct btrfs_root *root = BTRFS_I(dir)->root;
6403 struct inode *inode = NULL;
6404 int drop_inode_on_err = 0;
6410 * 2 for inode item and ref
6412 * 1 for xattr if selinux is on
6414 trans = btrfs_start_transaction(root, 5);
6416 return PTR_ERR(trans);
6418 err = btrfs_find_free_ino(root, &objectid);
6422 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6423 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6425 if (IS_ERR(inode)) {
6426 err = PTR_ERR(inode);
6429 drop_inode_on_err = 1;
6431 * If the active LSM wants to access the inode during
6432 * d_instantiate it needs these. Smack checks to see
6433 * if the filesystem supports xattrs by looking at the
6436 inode->i_fop = &btrfs_file_operations;
6437 inode->i_op = &btrfs_file_inode_operations;
6438 inode->i_mapping->a_ops = &btrfs_aops;
6440 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6442 goto out_unlock_inode;
6444 err = btrfs_update_inode(trans, root, inode);
6446 goto out_unlock_inode;
6448 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6451 goto out_unlock_inode;
6453 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6454 unlock_new_inode(inode);
6455 d_instantiate(dentry, inode);
6458 btrfs_end_transaction(trans);
6459 if (err && drop_inode_on_err) {
6460 inode_dec_link_count(inode);
6463 btrfs_balance_delayed_items(fs_info);
6464 btrfs_btree_balance_dirty(fs_info);
6468 unlock_new_inode(inode);
6473 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6474 struct dentry *dentry)
6476 struct btrfs_trans_handle *trans = NULL;
6477 struct btrfs_root *root = BTRFS_I(dir)->root;
6478 struct inode *inode = d_inode(old_dentry);
6479 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6484 /* do not allow sys_link's with other subvols of the same device */
6485 if (root->objectid != BTRFS_I(inode)->root->objectid)
6488 if (inode->i_nlink >= BTRFS_LINK_MAX)
6491 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6496 * 2 items for inode and inode ref
6497 * 2 items for dir items
6498 * 1 item for parent inode
6500 trans = btrfs_start_transaction(root, 5);
6501 if (IS_ERR(trans)) {
6502 err = PTR_ERR(trans);
6507 /* There are several dir indexes for this inode, clear the cache. */
6508 BTRFS_I(inode)->dir_index = 0ULL;
6510 inode_inc_iversion(inode);
6511 inode->i_ctime = current_time(inode);
6513 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6515 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6521 struct dentry *parent = dentry->d_parent;
6522 err = btrfs_update_inode(trans, root, inode);
6525 if (inode->i_nlink == 1) {
6527 * If new hard link count is 1, it's a file created
6528 * with open(2) O_TMPFILE flag.
6530 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6534 d_instantiate(dentry, inode);
6535 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6538 btrfs_balance_delayed_items(fs_info);
6541 btrfs_end_transaction(trans);
6543 inode_dec_link_count(inode);
6546 btrfs_btree_balance_dirty(fs_info);
6550 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6552 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6553 struct inode *inode = NULL;
6554 struct btrfs_trans_handle *trans;
6555 struct btrfs_root *root = BTRFS_I(dir)->root;
6557 int drop_on_err = 0;
6562 * 2 items for inode and ref
6563 * 2 items for dir items
6564 * 1 for xattr if selinux is on
6566 trans = btrfs_start_transaction(root, 5);
6568 return PTR_ERR(trans);
6570 err = btrfs_find_free_ino(root, &objectid);
6574 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6575 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6576 S_IFDIR | mode, &index);
6577 if (IS_ERR(inode)) {
6578 err = PTR_ERR(inode);
6583 /* these must be set before we unlock the inode */
6584 inode->i_op = &btrfs_dir_inode_operations;
6585 inode->i_fop = &btrfs_dir_file_operations;
6587 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6589 goto out_fail_inode;
6591 btrfs_i_size_write(BTRFS_I(inode), 0);
6592 err = btrfs_update_inode(trans, root, inode);
6594 goto out_fail_inode;
6596 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6597 dentry->d_name.name,
6598 dentry->d_name.len, 0, index);
6600 goto out_fail_inode;
6602 d_instantiate(dentry, inode);
6604 * mkdir is special. We're unlocking after we call d_instantiate
6605 * to avoid a race with nfsd calling d_instantiate.
6607 unlock_new_inode(inode);
6611 btrfs_end_transaction(trans);
6613 inode_dec_link_count(inode);
6616 btrfs_balance_delayed_items(fs_info);
6617 btrfs_btree_balance_dirty(fs_info);
6621 unlock_new_inode(inode);
6625 /* Find next extent map of a given extent map, caller needs to ensure locks */
6626 static struct extent_map *next_extent_map(struct extent_map *em)
6628 struct rb_node *next;
6630 next = rb_next(&em->rb_node);
6633 return container_of(next, struct extent_map, rb_node);
6636 static struct extent_map *prev_extent_map(struct extent_map *em)
6638 struct rb_node *prev;
6640 prev = rb_prev(&em->rb_node);
6643 return container_of(prev, struct extent_map, rb_node);
6646 /* helper for btfs_get_extent. Given an existing extent in the tree,
6647 * the existing extent is the nearest extent to map_start,
6648 * and an extent that you want to insert, deal with overlap and insert
6649 * the best fitted new extent into the tree.
6651 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6652 struct extent_map *existing,
6653 struct extent_map *em,
6656 struct extent_map *prev;
6657 struct extent_map *next;
6662 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6664 if (existing->start > map_start) {
6666 prev = prev_extent_map(next);
6669 next = next_extent_map(prev);
6672 start = prev ? extent_map_end(prev) : em->start;
6673 start = max_t(u64, start, em->start);
6674 end = next ? next->start : extent_map_end(em);
6675 end = min_t(u64, end, extent_map_end(em));
6676 start_diff = start - em->start;
6678 em->len = end - start;
6679 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6680 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6681 em->block_start += start_diff;
6682 em->block_len -= start_diff;
6684 return add_extent_mapping(em_tree, em, 0);
6687 static noinline int uncompress_inline(struct btrfs_path *path,
6689 size_t pg_offset, u64 extent_offset,
6690 struct btrfs_file_extent_item *item)
6693 struct extent_buffer *leaf = path->nodes[0];
6696 unsigned long inline_size;
6700 WARN_ON(pg_offset != 0);
6701 compress_type = btrfs_file_extent_compression(leaf, item);
6702 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6703 inline_size = btrfs_file_extent_inline_item_len(leaf,
6704 btrfs_item_nr(path->slots[0]));
6705 tmp = kmalloc(inline_size, GFP_NOFS);
6708 ptr = btrfs_file_extent_inline_start(item);
6710 read_extent_buffer(leaf, tmp, ptr, inline_size);
6712 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6713 ret = btrfs_decompress(compress_type, tmp, page,
6714 extent_offset, inline_size, max_size);
6720 * a bit scary, this does extent mapping from logical file offset to the disk.
6721 * the ugly parts come from merging extents from the disk with the in-ram
6722 * representation. This gets more complex because of the data=ordered code,
6723 * where the in-ram extents might be locked pending data=ordered completion.
6725 * This also copies inline extents directly into the page.
6728 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6730 size_t pg_offset, u64 start, u64 len,
6733 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6736 u64 extent_start = 0;
6738 u64 objectid = btrfs_ino(inode);
6740 struct btrfs_path *path = NULL;
6741 struct btrfs_root *root = inode->root;
6742 struct btrfs_file_extent_item *item;
6743 struct extent_buffer *leaf;
6744 struct btrfs_key found_key;
6745 struct extent_map *em = NULL;
6746 struct extent_map_tree *em_tree = &inode->extent_tree;
6747 struct extent_io_tree *io_tree = &inode->io_tree;
6748 struct btrfs_trans_handle *trans = NULL;
6749 const bool new_inline = !page || create;
6752 read_lock(&em_tree->lock);
6753 em = lookup_extent_mapping(em_tree, start, len);
6755 em->bdev = fs_info->fs_devices->latest_bdev;
6756 read_unlock(&em_tree->lock);
6759 if (em->start > start || em->start + em->len <= start)
6760 free_extent_map(em);
6761 else if (em->block_start == EXTENT_MAP_INLINE && page)
6762 free_extent_map(em);
6766 em = alloc_extent_map();
6771 em->bdev = fs_info->fs_devices->latest_bdev;
6772 em->start = EXTENT_MAP_HOLE;
6773 em->orig_start = EXTENT_MAP_HOLE;
6775 em->block_len = (u64)-1;
6778 path = btrfs_alloc_path();
6784 * Chances are we'll be called again, so go ahead and do
6787 path->reada = READA_FORWARD;
6790 ret = btrfs_lookup_file_extent(trans, root, path,
6791 objectid, start, trans != NULL);
6798 if (path->slots[0] == 0)
6803 leaf = path->nodes[0];
6804 item = btrfs_item_ptr(leaf, path->slots[0],
6805 struct btrfs_file_extent_item);
6806 /* are we inside the extent that was found? */
6807 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6808 found_type = found_key.type;
6809 if (found_key.objectid != objectid ||
6810 found_type != BTRFS_EXTENT_DATA_KEY) {
6812 * If we backup past the first extent we want to move forward
6813 * and see if there is an extent in front of us, otherwise we'll
6814 * say there is a hole for our whole search range which can
6821 found_type = btrfs_file_extent_type(leaf, item);
6822 extent_start = found_key.offset;
6823 if (found_type == BTRFS_FILE_EXTENT_REG ||
6824 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6825 extent_end = extent_start +
6826 btrfs_file_extent_num_bytes(leaf, item);
6827 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6829 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6830 extent_end = ALIGN(extent_start + size,
6831 fs_info->sectorsize);
6834 if (start >= extent_end) {
6836 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6837 ret = btrfs_next_leaf(root, path);
6844 leaf = path->nodes[0];
6846 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6847 if (found_key.objectid != objectid ||
6848 found_key.type != BTRFS_EXTENT_DATA_KEY)
6850 if (start + len <= found_key.offset)
6852 if (start > found_key.offset)
6855 em->orig_start = start;
6856 em->len = found_key.offset - start;
6860 btrfs_extent_item_to_extent_map(inode, path, item,
6863 if (found_type == BTRFS_FILE_EXTENT_REG ||
6864 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6866 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6870 size_t extent_offset;
6876 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6877 extent_offset = page_offset(page) + pg_offset - extent_start;
6878 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6879 size - extent_offset);
6880 em->start = extent_start + extent_offset;
6881 em->len = ALIGN(copy_size, fs_info->sectorsize);
6882 em->orig_block_len = em->len;
6883 em->orig_start = em->start;
6884 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6885 if (create == 0 && !PageUptodate(page)) {
6886 if (btrfs_file_extent_compression(leaf, item) !=
6887 BTRFS_COMPRESS_NONE) {
6888 ret = uncompress_inline(path, page, pg_offset,
6889 extent_offset, item);
6896 read_extent_buffer(leaf, map + pg_offset, ptr,
6898 if (pg_offset + copy_size < PAGE_SIZE) {
6899 memset(map + pg_offset + copy_size, 0,
6900 PAGE_SIZE - pg_offset -
6905 flush_dcache_page(page);
6906 } else if (create && PageUptodate(page)) {
6910 free_extent_map(em);
6913 btrfs_release_path(path);
6914 trans = btrfs_join_transaction(root);
6917 return ERR_CAST(trans);
6921 write_extent_buffer(leaf, map + pg_offset, ptr,
6924 btrfs_mark_buffer_dirty(leaf);
6926 set_extent_uptodate(io_tree, em->start,
6927 extent_map_end(em) - 1, NULL, GFP_NOFS);
6932 em->orig_start = start;
6935 em->block_start = EXTENT_MAP_HOLE;
6936 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6938 btrfs_release_path(path);
6939 if (em->start > start || extent_map_end(em) <= start) {
6941 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6942 em->start, em->len, start, len);
6948 write_lock(&em_tree->lock);
6949 ret = add_extent_mapping(em_tree, em, 0);
6950 /* it is possible that someone inserted the extent into the tree
6951 * while we had the lock dropped. It is also possible that
6952 * an overlapping map exists in the tree
6954 if (ret == -EEXIST) {
6955 struct extent_map *existing;
6959 existing = search_extent_mapping(em_tree, start, len);
6961 * existing will always be non-NULL, since there must be
6962 * extent causing the -EEXIST.
6964 if (existing->start == em->start &&
6965 extent_map_end(existing) >= extent_map_end(em) &&
6966 em->block_start == existing->block_start) {
6968 * The existing extent map already encompasses the
6969 * entire extent map we tried to add.
6971 free_extent_map(em);
6975 } else if (start >= extent_map_end(existing) ||
6976 start <= existing->start) {
6978 * The existing extent map is the one nearest to
6979 * the [start, start + len) range which overlaps
6981 err = merge_extent_mapping(em_tree, existing,
6983 free_extent_map(existing);
6985 free_extent_map(em);
6989 free_extent_map(em);
6994 write_unlock(&em_tree->lock);
6997 trace_btrfs_get_extent(root, inode, em);
6999 btrfs_free_path(path);
7001 ret = btrfs_end_transaction(trans);
7006 free_extent_map(em);
7007 return ERR_PTR(err);
7009 BUG_ON(!em); /* Error is always set */
7013 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7015 size_t pg_offset, u64 start, u64 len,
7018 struct extent_map *em;
7019 struct extent_map *hole_em = NULL;
7020 u64 range_start = start;
7026 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7033 * - a pre-alloc extent,
7034 * there might actually be delalloc bytes behind it.
7036 if (em->block_start != EXTENT_MAP_HOLE &&
7037 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7043 /* check to see if we've wrapped (len == -1 or similar) */
7052 /* ok, we didn't find anything, lets look for delalloc */
7053 found = count_range_bits(&inode->io_tree, &range_start,
7054 end, len, EXTENT_DELALLOC, 1);
7055 found_end = range_start + found;
7056 if (found_end < range_start)
7057 found_end = (u64)-1;
7060 * we didn't find anything useful, return
7061 * the original results from get_extent()
7063 if (range_start > end || found_end <= start) {
7069 /* adjust the range_start to make sure it doesn't
7070 * go backwards from the start they passed in
7072 range_start = max(start, range_start);
7073 found = found_end - range_start;
7076 u64 hole_start = start;
7079 em = alloc_extent_map();
7085 * when btrfs_get_extent can't find anything it
7086 * returns one huge hole
7088 * make sure what it found really fits our range, and
7089 * adjust to make sure it is based on the start from
7093 u64 calc_end = extent_map_end(hole_em);
7095 if (calc_end <= start || (hole_em->start > end)) {
7096 free_extent_map(hole_em);
7099 hole_start = max(hole_em->start, start);
7100 hole_len = calc_end - hole_start;
7104 if (hole_em && range_start > hole_start) {
7105 /* our hole starts before our delalloc, so we
7106 * have to return just the parts of the hole
7107 * that go until the delalloc starts
7109 em->len = min(hole_len,
7110 range_start - hole_start);
7111 em->start = hole_start;
7112 em->orig_start = hole_start;
7114 * don't adjust block start at all,
7115 * it is fixed at EXTENT_MAP_HOLE
7117 em->block_start = hole_em->block_start;
7118 em->block_len = hole_len;
7119 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7120 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7122 em->start = range_start;
7124 em->orig_start = range_start;
7125 em->block_start = EXTENT_MAP_DELALLOC;
7126 em->block_len = found;
7128 } else if (hole_em) {
7133 free_extent_map(hole_em);
7135 free_extent_map(em);
7136 return ERR_PTR(err);
7141 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7144 const u64 orig_start,
7145 const u64 block_start,
7146 const u64 block_len,
7147 const u64 orig_block_len,
7148 const u64 ram_bytes,
7151 struct extent_map *em = NULL;
7154 if (type != BTRFS_ORDERED_NOCOW) {
7155 em = create_io_em(inode, start, len, orig_start,
7156 block_start, block_len, orig_block_len,
7158 BTRFS_COMPRESS_NONE, /* compress_type */
7163 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7164 len, block_len, type);
7167 free_extent_map(em);
7168 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7169 start + len - 1, 0);
7178 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7181 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7182 struct btrfs_root *root = BTRFS_I(inode)->root;
7183 struct extent_map *em;
7184 struct btrfs_key ins;
7188 alloc_hint = get_extent_allocation_hint(inode, start, len);
7189 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7190 0, alloc_hint, &ins, 1, 1);
7192 return ERR_PTR(ret);
7194 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7195 ins.objectid, ins.offset, ins.offset,
7196 ins.offset, BTRFS_ORDERED_REGULAR);
7197 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7199 btrfs_free_reserved_extent(fs_info, ins.objectid,
7206 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7207 * block must be cow'd
7209 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7210 u64 *orig_start, u64 *orig_block_len,
7213 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7214 struct btrfs_path *path;
7216 struct extent_buffer *leaf;
7217 struct btrfs_root *root = BTRFS_I(inode)->root;
7218 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7219 struct btrfs_file_extent_item *fi;
7220 struct btrfs_key key;
7227 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7229 path = btrfs_alloc_path();
7233 ret = btrfs_lookup_file_extent(NULL, root, path,
7234 btrfs_ino(BTRFS_I(inode)), offset, 0);
7238 slot = path->slots[0];
7241 /* can't find the item, must cow */
7248 leaf = path->nodes[0];
7249 btrfs_item_key_to_cpu(leaf, &key, slot);
7250 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7251 key.type != BTRFS_EXTENT_DATA_KEY) {
7252 /* not our file or wrong item type, must cow */
7256 if (key.offset > offset) {
7257 /* Wrong offset, must cow */
7261 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7262 found_type = btrfs_file_extent_type(leaf, fi);
7263 if (found_type != BTRFS_FILE_EXTENT_REG &&
7264 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7265 /* not a regular extent, must cow */
7269 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7272 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7273 if (extent_end <= offset)
7276 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7277 if (disk_bytenr == 0)
7280 if (btrfs_file_extent_compression(leaf, fi) ||
7281 btrfs_file_extent_encryption(leaf, fi) ||
7282 btrfs_file_extent_other_encoding(leaf, fi))
7285 backref_offset = btrfs_file_extent_offset(leaf, fi);
7288 *orig_start = key.offset - backref_offset;
7289 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7290 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7293 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7296 num_bytes = min(offset + *len, extent_end) - offset;
7297 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7300 range_end = round_up(offset + num_bytes,
7301 root->fs_info->sectorsize) - 1;
7302 ret = test_range_bit(io_tree, offset, range_end,
7303 EXTENT_DELALLOC, 0, NULL);
7310 btrfs_release_path(path);
7313 * look for other files referencing this extent, if we
7314 * find any we must cow
7317 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7318 key.offset - backref_offset, disk_bytenr);
7325 * adjust disk_bytenr and num_bytes to cover just the bytes
7326 * in this extent we are about to write. If there
7327 * are any csums in that range we have to cow in order
7328 * to keep the csums correct
7330 disk_bytenr += backref_offset;
7331 disk_bytenr += offset - key.offset;
7332 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7335 * all of the above have passed, it is safe to overwrite this extent
7341 btrfs_free_path(path);
7345 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7347 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7349 void **pagep = NULL;
7350 struct page *page = NULL;
7354 start_idx = start >> PAGE_SHIFT;
7357 * end is the last byte in the last page. end == start is legal
7359 end_idx = end >> PAGE_SHIFT;
7363 /* Most of the code in this while loop is lifted from
7364 * find_get_page. It's been modified to begin searching from a
7365 * page and return just the first page found in that range. If the
7366 * found idx is less than or equal to the end idx then we know that
7367 * a page exists. If no pages are found or if those pages are
7368 * outside of the range then we're fine (yay!) */
7369 while (page == NULL &&
7370 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7371 page = radix_tree_deref_slot(pagep);
7372 if (unlikely(!page))
7375 if (radix_tree_exception(page)) {
7376 if (radix_tree_deref_retry(page)) {
7381 * Otherwise, shmem/tmpfs must be storing a swap entry
7382 * here as an exceptional entry: so return it without
7383 * attempting to raise page count.
7386 break; /* TODO: Is this relevant for this use case? */
7389 if (!page_cache_get_speculative(page)) {
7395 * Has the page moved?
7396 * This is part of the lockless pagecache protocol. See
7397 * include/linux/pagemap.h for details.
7399 if (unlikely(page != *pagep)) {
7406 if (page->index <= end_idx)
7415 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7416 struct extent_state **cached_state, int writing)
7418 struct btrfs_ordered_extent *ordered;
7422 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7425 * We're concerned with the entire range that we're going to be
7426 * doing DIO to, so we need to make sure there's no ordered
7427 * extents in this range.
7429 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7430 lockend - lockstart + 1);
7433 * We need to make sure there are no buffered pages in this
7434 * range either, we could have raced between the invalidate in
7435 * generic_file_direct_write and locking the extent. The
7436 * invalidate needs to happen so that reads after a write do not
7441 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7444 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7445 cached_state, GFP_NOFS);
7449 * If we are doing a DIO read and the ordered extent we
7450 * found is for a buffered write, we can not wait for it
7451 * to complete and retry, because if we do so we can
7452 * deadlock with concurrent buffered writes on page
7453 * locks. This happens only if our DIO read covers more
7454 * than one extent map, if at this point has already
7455 * created an ordered extent for a previous extent map
7456 * and locked its range in the inode's io tree, and a
7457 * concurrent write against that previous extent map's
7458 * range and this range started (we unlock the ranges
7459 * in the io tree only when the bios complete and
7460 * buffered writes always lock pages before attempting
7461 * to lock range in the io tree).
7464 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7465 btrfs_start_ordered_extent(inode, ordered, 1);
7468 btrfs_put_ordered_extent(ordered);
7471 * We could trigger writeback for this range (and wait
7472 * for it to complete) and then invalidate the pages for
7473 * this range (through invalidate_inode_pages2_range()),
7474 * but that can lead us to a deadlock with a concurrent
7475 * call to readpages() (a buffered read or a defrag call
7476 * triggered a readahead) on a page lock due to an
7477 * ordered dio extent we created before but did not have
7478 * yet a corresponding bio submitted (whence it can not
7479 * complete), which makes readpages() wait for that
7480 * ordered extent to complete while holding a lock on
7495 /* The callers of this must take lock_extent() */
7496 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7497 u64 orig_start, u64 block_start,
7498 u64 block_len, u64 orig_block_len,
7499 u64 ram_bytes, int compress_type,
7502 struct extent_map_tree *em_tree;
7503 struct extent_map *em;
7504 struct btrfs_root *root = BTRFS_I(inode)->root;
7507 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7508 type == BTRFS_ORDERED_COMPRESSED ||
7509 type == BTRFS_ORDERED_NOCOW ||
7510 type == BTRFS_ORDERED_REGULAR);
7512 em_tree = &BTRFS_I(inode)->extent_tree;
7513 em = alloc_extent_map();
7515 return ERR_PTR(-ENOMEM);
7518 em->orig_start = orig_start;
7520 em->block_len = block_len;
7521 em->block_start = block_start;
7522 em->bdev = root->fs_info->fs_devices->latest_bdev;
7523 em->orig_block_len = orig_block_len;
7524 em->ram_bytes = ram_bytes;
7525 em->generation = -1;
7526 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7527 if (type == BTRFS_ORDERED_PREALLOC) {
7528 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7529 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7530 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7531 em->compress_type = compress_type;
7535 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7536 em->start + em->len - 1, 0);
7537 write_lock(&em_tree->lock);
7538 ret = add_extent_mapping(em_tree, em, 1);
7539 write_unlock(&em_tree->lock);
7541 * The caller has taken lock_extent(), who could race with us
7544 } while (ret == -EEXIST);
7547 free_extent_map(em);
7548 return ERR_PTR(ret);
7551 /* em got 2 refs now, callers needs to do free_extent_map once. */
7555 static void adjust_dio_outstanding_extents(struct inode *inode,
7556 struct btrfs_dio_data *dio_data,
7559 unsigned num_extents = count_max_extents(len);
7562 * If we have an outstanding_extents count still set then we're
7563 * within our reservation, otherwise we need to adjust our inode
7564 * counter appropriately.
7566 if (dio_data->outstanding_extents >= num_extents) {
7567 dio_data->outstanding_extents -= num_extents;
7570 * If dio write length has been split due to no large enough
7571 * contiguous space, we need to compensate our inode counter
7574 u64 num_needed = num_extents - dio_data->outstanding_extents;
7576 spin_lock(&BTRFS_I(inode)->lock);
7577 BTRFS_I(inode)->outstanding_extents += num_needed;
7578 spin_unlock(&BTRFS_I(inode)->lock);
7582 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7583 struct buffer_head *bh_result, int create)
7585 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7586 struct extent_map *em;
7587 struct extent_state *cached_state = NULL;
7588 struct btrfs_dio_data *dio_data = NULL;
7589 u64 start = iblock << inode->i_blkbits;
7590 u64 lockstart, lockend;
7591 u64 len = bh_result->b_size;
7592 int unlock_bits = EXTENT_LOCKED;
7596 unlock_bits |= EXTENT_DIRTY;
7598 len = min_t(u64, len, fs_info->sectorsize);
7601 lockend = start + len - 1;
7603 if (current->journal_info) {
7605 * Need to pull our outstanding extents and set journal_info to NULL so
7606 * that anything that needs to check if there's a transaction doesn't get
7609 dio_data = current->journal_info;
7610 current->journal_info = NULL;
7614 * If this errors out it's because we couldn't invalidate pagecache for
7615 * this range and we need to fallback to buffered.
7617 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7623 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7630 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7631 * io. INLINE is special, and we could probably kludge it in here, but
7632 * it's still buffered so for safety lets just fall back to the generic
7635 * For COMPRESSED we _have_ to read the entire extent in so we can
7636 * decompress it, so there will be buffering required no matter what we
7637 * do, so go ahead and fallback to buffered.
7639 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7640 * to buffered IO. Don't blame me, this is the price we pay for using
7643 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7644 em->block_start == EXTENT_MAP_INLINE) {
7645 free_extent_map(em);
7650 /* Just a good old fashioned hole, return */
7651 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7652 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7653 free_extent_map(em);
7658 * We don't allocate a new extent in the following cases
7660 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7662 * 2) The extent is marked as PREALLOC. We're good to go here and can
7663 * just use the extent.
7667 len = min(len, em->len - (start - em->start));
7668 lockstart = start + len;
7672 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7673 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7674 em->block_start != EXTENT_MAP_HOLE)) {
7676 u64 block_start, orig_start, orig_block_len, ram_bytes;
7678 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7679 type = BTRFS_ORDERED_PREALLOC;
7681 type = BTRFS_ORDERED_NOCOW;
7682 len = min(len, em->len - (start - em->start));
7683 block_start = em->block_start + (start - em->start);
7685 if (can_nocow_extent(inode, start, &len, &orig_start,
7686 &orig_block_len, &ram_bytes) == 1 &&
7687 btrfs_inc_nocow_writers(fs_info, block_start)) {
7688 struct extent_map *em2;
7690 em2 = btrfs_create_dio_extent(inode, start, len,
7691 orig_start, block_start,
7692 len, orig_block_len,
7694 btrfs_dec_nocow_writers(fs_info, block_start);
7695 if (type == BTRFS_ORDERED_PREALLOC) {
7696 free_extent_map(em);
7699 if (em2 && IS_ERR(em2)) {
7704 * For inode marked NODATACOW or extent marked PREALLOC,
7705 * use the existing or preallocated extent, so does not
7706 * need to adjust btrfs_space_info's bytes_may_use.
7708 btrfs_free_reserved_data_space_noquota(inode,
7715 * this will cow the extent, reset the len in case we changed
7718 len = bh_result->b_size;
7719 free_extent_map(em);
7720 em = btrfs_new_extent_direct(inode, start, len);
7725 len = min(len, em->len - (start - em->start));
7727 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7729 bh_result->b_size = len;
7730 bh_result->b_bdev = em->bdev;
7731 set_buffer_mapped(bh_result);
7733 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7734 set_buffer_new(bh_result);
7737 * Need to update the i_size under the extent lock so buffered
7738 * readers will get the updated i_size when we unlock.
7740 if (!dio_data->overwrite && start + len > i_size_read(inode))
7741 i_size_write(inode, start + len);
7743 adjust_dio_outstanding_extents(inode, dio_data, len);
7744 WARN_ON(dio_data->reserve < len);
7745 dio_data->reserve -= len;
7746 dio_data->unsubmitted_oe_range_end = start + len;
7747 current->journal_info = dio_data;
7751 * In the case of write we need to clear and unlock the entire range,
7752 * in the case of read we need to unlock only the end area that we
7753 * aren't using if there is any left over space.
7755 if (lockstart < lockend) {
7756 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7757 lockend, unlock_bits, 1, 0,
7758 &cached_state, GFP_NOFS);
7760 free_extent_state(cached_state);
7763 free_extent_map(em);
7768 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7769 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7772 current->journal_info = dio_data;
7774 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7775 * write less data then expected, so that we don't underflow our inode's
7776 * outstanding extents counter.
7778 if (create && dio_data)
7779 adjust_dio_outstanding_extents(inode, dio_data, len);
7784 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7787 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7790 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7794 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7798 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7804 static int btrfs_check_dio_repairable(struct inode *inode,
7805 struct bio *failed_bio,
7806 struct io_failure_record *failrec,
7809 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7812 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7813 if (num_copies == 1) {
7815 * we only have a single copy of the data, so don't bother with
7816 * all the retry and error correction code that follows. no
7817 * matter what the error is, it is very likely to persist.
7819 btrfs_debug(fs_info,
7820 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7821 num_copies, failrec->this_mirror, failed_mirror);
7825 failrec->failed_mirror = failed_mirror;
7826 failrec->this_mirror++;
7827 if (failrec->this_mirror == failed_mirror)
7828 failrec->this_mirror++;
7830 if (failrec->this_mirror > num_copies) {
7831 btrfs_debug(fs_info,
7832 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7833 num_copies, failrec->this_mirror, failed_mirror);
7840 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7841 struct page *page, unsigned int pgoff,
7842 u64 start, u64 end, int failed_mirror,
7843 bio_end_io_t *repair_endio, void *repair_arg)
7845 struct io_failure_record *failrec;
7851 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7853 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7857 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7860 free_io_failure(BTRFS_I(inode), failrec);
7864 if ((failed_bio->bi_vcnt > 1)
7865 || (failed_bio->bi_io_vec->bv_len
7866 > btrfs_inode_sectorsize(inode)))
7867 read_mode |= REQ_FAILFAST_DEV;
7869 isector = start - btrfs_io_bio(failed_bio)->logical;
7870 isector >>= inode->i_sb->s_blocksize_bits;
7871 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7872 pgoff, isector, repair_endio, repair_arg);
7874 free_io_failure(BTRFS_I(inode), failrec);
7877 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7879 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7880 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7881 read_mode, failrec->this_mirror, failrec->in_validation);
7883 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7885 free_io_failure(BTRFS_I(inode), failrec);
7892 struct btrfs_retry_complete {
7893 struct completion done;
7894 struct inode *inode;
7899 static void btrfs_retry_endio_nocsum(struct bio *bio)
7901 struct btrfs_retry_complete *done = bio->bi_private;
7902 struct inode *inode;
7903 struct bio_vec *bvec;
7909 ASSERT(bio->bi_vcnt == 1);
7910 inode = bio->bi_io_vec->bv_page->mapping->host;
7911 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7914 bio_for_each_segment_all(bvec, bio, i)
7915 clean_io_failure(BTRFS_I(done->inode), done->start, bvec->bv_page, 0);
7917 complete(&done->done);
7921 static int __btrfs_correct_data_nocsum(struct inode *inode,
7922 struct btrfs_io_bio *io_bio)
7924 struct btrfs_fs_info *fs_info;
7925 struct bio_vec *bvec;
7926 struct btrfs_retry_complete done;
7934 fs_info = BTRFS_I(inode)->root->fs_info;
7935 sectorsize = fs_info->sectorsize;
7937 start = io_bio->logical;
7940 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7941 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7942 pgoff = bvec->bv_offset;
7944 next_block_or_try_again:
7947 init_completion(&done.done);
7949 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7950 pgoff, start, start + sectorsize - 1,
7952 btrfs_retry_endio_nocsum, &done);
7956 wait_for_completion(&done.done);
7958 if (!done.uptodate) {
7959 /* We might have another mirror, so try again */
7960 goto next_block_or_try_again;
7963 start += sectorsize;
7966 pgoff += sectorsize;
7967 goto next_block_or_try_again;
7974 static void btrfs_retry_endio(struct bio *bio)
7976 struct btrfs_retry_complete *done = bio->bi_private;
7977 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7978 struct inode *inode;
7979 struct bio_vec *bvec;
7990 start = done->start;
7992 ASSERT(bio->bi_vcnt == 1);
7993 inode = bio->bi_io_vec->bv_page->mapping->host;
7994 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
7996 bio_for_each_segment_all(bvec, bio, i) {
7997 ret = __readpage_endio_check(done->inode, io_bio, i,
7998 bvec->bv_page, bvec->bv_offset,
7999 done->start, bvec->bv_len);
8001 clean_io_failure(BTRFS_I(done->inode), done->start,
8002 bvec->bv_page, bvec->bv_offset);
8007 done->uptodate = uptodate;
8009 complete(&done->done);
8013 static int __btrfs_subio_endio_read(struct inode *inode,
8014 struct btrfs_io_bio *io_bio, int err)
8016 struct btrfs_fs_info *fs_info;
8017 struct bio_vec *bvec;
8018 struct btrfs_retry_complete done;
8028 fs_info = BTRFS_I(inode)->root->fs_info;
8029 sectorsize = fs_info->sectorsize;
8032 start = io_bio->logical;
8035 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8036 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8038 pgoff = bvec->bv_offset;
8040 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8041 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8042 bvec->bv_page, pgoff, start,
8049 init_completion(&done.done);
8051 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8052 pgoff, start, start + sectorsize - 1,
8054 btrfs_retry_endio, &done);
8060 wait_for_completion(&done.done);
8062 if (!done.uptodate) {
8063 /* We might have another mirror, so try again */
8067 offset += sectorsize;
8068 start += sectorsize;
8073 pgoff += sectorsize;
8081 static int btrfs_subio_endio_read(struct inode *inode,
8082 struct btrfs_io_bio *io_bio, int err)
8084 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8088 return __btrfs_correct_data_nocsum(inode, io_bio);
8092 return __btrfs_subio_endio_read(inode, io_bio, err);
8096 static void btrfs_endio_direct_read(struct bio *bio)
8098 struct btrfs_dio_private *dip = bio->bi_private;
8099 struct inode *inode = dip->inode;
8100 struct bio *dio_bio;
8101 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8102 int err = bio->bi_error;
8104 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8105 err = btrfs_subio_endio_read(inode, io_bio, err);
8107 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8108 dip->logical_offset + dip->bytes - 1);
8109 dio_bio = dip->dio_bio;
8113 dio_bio->bi_error = bio->bi_error;
8114 dio_end_io(dio_bio, bio->bi_error);
8117 io_bio->end_io(io_bio, err);
8121 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8126 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8127 struct btrfs_ordered_extent *ordered = NULL;
8128 u64 ordered_offset = offset;
8129 u64 ordered_bytes = bytes;
8133 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8140 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8141 finish_ordered_fn, NULL, NULL);
8142 btrfs_queue_work(fs_info->endio_write_workers, &ordered->work);
8145 * our bio might span multiple ordered extents. If we haven't
8146 * completed the accounting for the whole dio, go back and try again
8148 if (ordered_offset < offset + bytes) {
8149 ordered_bytes = offset + bytes - ordered_offset;
8155 static void btrfs_endio_direct_write(struct bio *bio)
8157 struct btrfs_dio_private *dip = bio->bi_private;
8158 struct bio *dio_bio = dip->dio_bio;
8160 btrfs_endio_direct_write_update_ordered(dip->inode,
8161 dip->logical_offset,
8167 dio_bio->bi_error = bio->bi_error;
8168 dio_end_io(dio_bio, bio->bi_error);
8172 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8173 struct bio *bio, int mirror_num,
8174 unsigned long bio_flags, u64 offset)
8177 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8178 BUG_ON(ret); /* -ENOMEM */
8182 static void btrfs_end_dio_bio(struct bio *bio)
8184 struct btrfs_dio_private *dip = bio->bi_private;
8185 int err = bio->bi_error;
8188 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8189 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8190 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8192 (unsigned long long)bio->bi_iter.bi_sector,
8193 bio->bi_iter.bi_size, err);
8195 if (dip->subio_endio)
8196 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8202 * before atomic variable goto zero, we must make sure
8203 * dip->errors is perceived to be set.
8205 smp_mb__before_atomic();
8208 /* if there are more bios still pending for this dio, just exit */
8209 if (!atomic_dec_and_test(&dip->pending_bios))
8213 bio_io_error(dip->orig_bio);
8215 dip->dio_bio->bi_error = 0;
8216 bio_endio(dip->orig_bio);
8222 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8223 u64 first_sector, gfp_t gfp_flags)
8226 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8228 bio_associate_current(bio);
8232 static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8233 struct btrfs_dio_private *dip,
8237 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8238 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8242 * We load all the csum data we need when we submit
8243 * the first bio to reduce the csum tree search and
8246 if (dip->logical_offset == file_offset) {
8247 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8253 if (bio == dip->orig_bio)
8256 file_offset -= dip->logical_offset;
8257 file_offset >>= inode->i_sb->s_blocksize_bits;
8258 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8263 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8264 u64 file_offset, int skip_sum,
8267 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8268 struct btrfs_dio_private *dip = bio->bi_private;
8269 bool write = bio_op(bio) == REQ_OP_WRITE;
8273 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8278 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8286 if (write && async_submit) {
8287 ret = btrfs_wq_submit_bio(fs_info, inode, bio, 0, 0,
8289 __btrfs_submit_bio_start_direct_io,
8290 __btrfs_submit_bio_done);
8294 * If we aren't doing async submit, calculate the csum of the
8297 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8301 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8307 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8313 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8316 struct inode *inode = dip->inode;
8317 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8318 struct btrfs_root *root = BTRFS_I(inode)->root;
8320 struct bio *orig_bio = dip->orig_bio;
8321 struct bio_vec *bvec;
8322 u64 start_sector = orig_bio->bi_iter.bi_sector;
8323 u64 file_offset = dip->logical_offset;
8326 u32 blocksize = fs_info->sectorsize;
8327 int async_submit = 0;
8332 map_length = orig_bio->bi_iter.bi_size;
8333 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8334 &map_length, NULL, 0);
8338 if (map_length >= orig_bio->bi_iter.bi_size) {
8340 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8344 /* async crcs make it difficult to collect full stripe writes. */
8345 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8350 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8354 bio->bi_opf = orig_bio->bi_opf;
8355 bio->bi_private = dip;
8356 bio->bi_end_io = btrfs_end_dio_bio;
8357 btrfs_io_bio(bio)->logical = file_offset;
8358 atomic_inc(&dip->pending_bios);
8360 bio_for_each_segment_all(bvec, orig_bio, j) {
8361 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8364 if (unlikely(map_length < submit_len + blocksize ||
8365 bio_add_page(bio, bvec->bv_page, blocksize,
8366 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8368 * inc the count before we submit the bio so
8369 * we know the end IO handler won't happen before
8370 * we inc the count. Otherwise, the dip might get freed
8371 * before we're done setting it up
8373 atomic_inc(&dip->pending_bios);
8374 ret = __btrfs_submit_dio_bio(bio, inode,
8375 file_offset, skip_sum,
8379 atomic_dec(&dip->pending_bios);
8383 start_sector += submit_len >> 9;
8384 file_offset += submit_len;
8388 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8389 start_sector, GFP_NOFS);
8392 bio->bi_opf = orig_bio->bi_opf;
8393 bio->bi_private = dip;
8394 bio->bi_end_io = btrfs_end_dio_bio;
8395 btrfs_io_bio(bio)->logical = file_offset;
8397 map_length = orig_bio->bi_iter.bi_size;
8398 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8400 &map_length, NULL, 0);
8408 submit_len += blocksize;
8417 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8426 * before atomic variable goto zero, we must
8427 * make sure dip->errors is perceived to be set.
8429 smp_mb__before_atomic();
8430 if (atomic_dec_and_test(&dip->pending_bios))
8431 bio_io_error(dip->orig_bio);
8433 /* bio_end_io() will handle error, so we needn't return it */
8437 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8440 struct btrfs_dio_private *dip = NULL;
8441 struct bio *io_bio = NULL;
8442 struct btrfs_io_bio *btrfs_bio;
8444 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8447 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8449 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8455 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8461 dip->private = dio_bio->bi_private;
8463 dip->logical_offset = file_offset;
8464 dip->bytes = dio_bio->bi_iter.bi_size;
8465 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8466 io_bio->bi_private = dip;
8467 dip->orig_bio = io_bio;
8468 dip->dio_bio = dio_bio;
8469 atomic_set(&dip->pending_bios, 0);
8470 btrfs_bio = btrfs_io_bio(io_bio);
8471 btrfs_bio->logical = file_offset;
8474 io_bio->bi_end_io = btrfs_endio_direct_write;
8476 io_bio->bi_end_io = btrfs_endio_direct_read;
8477 dip->subio_endio = btrfs_subio_endio_read;
8481 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8482 * even if we fail to submit a bio, because in such case we do the
8483 * corresponding error handling below and it must not be done a second
8484 * time by btrfs_direct_IO().
8487 struct btrfs_dio_data *dio_data = current->journal_info;
8489 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8491 dio_data->unsubmitted_oe_range_start =
8492 dio_data->unsubmitted_oe_range_end;
8495 ret = btrfs_submit_direct_hook(dip, skip_sum);
8499 if (btrfs_bio->end_io)
8500 btrfs_bio->end_io(btrfs_bio, ret);
8504 * If we arrived here it means either we failed to submit the dip
8505 * or we either failed to clone the dio_bio or failed to allocate the
8506 * dip. If we cloned the dio_bio and allocated the dip, we can just
8507 * call bio_endio against our io_bio so that we get proper resource
8508 * cleanup if we fail to submit the dip, otherwise, we must do the
8509 * same as btrfs_endio_direct_[write|read] because we can't call these
8510 * callbacks - they require an allocated dip and a clone of dio_bio.
8512 if (io_bio && dip) {
8513 io_bio->bi_error = -EIO;
8516 * The end io callbacks free our dip, do the final put on io_bio
8517 * and all the cleanup and final put for dio_bio (through
8524 btrfs_endio_direct_write_update_ordered(inode,
8526 dio_bio->bi_iter.bi_size,
8529 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8530 file_offset + dio_bio->bi_iter.bi_size - 1);
8532 dio_bio->bi_error = -EIO;
8534 * Releases and cleans up our dio_bio, no need to bio_put()
8535 * nor bio_endio()/bio_io_error() against dio_bio.
8537 dio_end_io(dio_bio, ret);
8544 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8546 const struct iov_iter *iter, loff_t offset)
8550 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8551 ssize_t retval = -EINVAL;
8553 if (offset & blocksize_mask)
8556 if (iov_iter_alignment(iter) & blocksize_mask)
8559 /* If this is a write we don't need to check anymore */
8560 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8563 * Check to make sure we don't have duplicate iov_base's in this
8564 * iovec, if so return EINVAL, otherwise we'll get csum errors
8565 * when reading back.
8567 for (seg = 0; seg < iter->nr_segs; seg++) {
8568 for (i = seg + 1; i < iter->nr_segs; i++) {
8569 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8578 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8580 struct file *file = iocb->ki_filp;
8581 struct inode *inode = file->f_mapping->host;
8582 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8583 struct btrfs_dio_data dio_data = { 0 };
8584 loff_t offset = iocb->ki_pos;
8588 bool relock = false;
8591 if (check_direct_IO(fs_info, iocb, iter, offset))
8594 inode_dio_begin(inode);
8595 smp_mb__after_atomic();
8598 * The generic stuff only does filemap_write_and_wait_range, which
8599 * isn't enough if we've written compressed pages to this area, so
8600 * we need to flush the dirty pages again to make absolutely sure
8601 * that any outstanding dirty pages are on disk.
8603 count = iov_iter_count(iter);
8604 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8605 &BTRFS_I(inode)->runtime_flags))
8606 filemap_fdatawrite_range(inode->i_mapping, offset,
8607 offset + count - 1);
8609 if (iov_iter_rw(iter) == WRITE) {
8611 * If the write DIO is beyond the EOF, we need update
8612 * the isize, but it is protected by i_mutex. So we can
8613 * not unlock the i_mutex at this case.
8615 if (offset + count <= inode->i_size) {
8616 dio_data.overwrite = 1;
8617 inode_unlock(inode);
8620 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8623 dio_data.outstanding_extents = count_max_extents(count);
8626 * We need to know how many extents we reserved so that we can
8627 * do the accounting properly if we go over the number we
8628 * originally calculated. Abuse current->journal_info for this.
8630 dio_data.reserve = round_up(count,
8631 fs_info->sectorsize);
8632 dio_data.unsubmitted_oe_range_start = (u64)offset;
8633 dio_data.unsubmitted_oe_range_end = (u64)offset;
8634 current->journal_info = &dio_data;
8635 down_read(&BTRFS_I(inode)->dio_sem);
8636 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8637 &BTRFS_I(inode)->runtime_flags)) {
8638 inode_dio_end(inode);
8639 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8643 ret = __blockdev_direct_IO(iocb, inode,
8644 fs_info->fs_devices->latest_bdev,
8645 iter, btrfs_get_blocks_direct, NULL,
8646 btrfs_submit_direct, flags);
8647 if (iov_iter_rw(iter) == WRITE) {
8648 up_read(&BTRFS_I(inode)->dio_sem);
8649 current->journal_info = NULL;
8650 if (ret < 0 && ret != -EIOCBQUEUED) {
8651 if (dio_data.reserve)
8652 btrfs_delalloc_release_space(inode, offset,
8655 * On error we might have left some ordered extents
8656 * without submitting corresponding bios for them, so
8657 * cleanup them up to avoid other tasks getting them
8658 * and waiting for them to complete forever.
8660 if (dio_data.unsubmitted_oe_range_start <
8661 dio_data.unsubmitted_oe_range_end)
8662 btrfs_endio_direct_write_update_ordered(inode,
8663 dio_data.unsubmitted_oe_range_start,
8664 dio_data.unsubmitted_oe_range_end -
8665 dio_data.unsubmitted_oe_range_start,
8667 } else if (ret >= 0 && (size_t)ret < count)
8668 btrfs_delalloc_release_space(inode, offset,
8669 count - (size_t)ret);
8673 inode_dio_end(inode);
8680 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8682 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8683 __u64 start, __u64 len)
8687 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8691 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8694 int btrfs_readpage(struct file *file, struct page *page)
8696 struct extent_io_tree *tree;
8697 tree = &BTRFS_I(page->mapping->host)->io_tree;
8698 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8701 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8703 struct extent_io_tree *tree;
8704 struct inode *inode = page->mapping->host;
8707 if (current->flags & PF_MEMALLOC) {
8708 redirty_page_for_writepage(wbc, page);
8714 * If we are under memory pressure we will call this directly from the
8715 * VM, we need to make sure we have the inode referenced for the ordered
8716 * extent. If not just return like we didn't do anything.
8718 if (!igrab(inode)) {
8719 redirty_page_for_writepage(wbc, page);
8720 return AOP_WRITEPAGE_ACTIVATE;
8722 tree = &BTRFS_I(page->mapping->host)->io_tree;
8723 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8724 btrfs_add_delayed_iput(inode);
8728 static int btrfs_writepages(struct address_space *mapping,
8729 struct writeback_control *wbc)
8731 struct extent_io_tree *tree;
8733 tree = &BTRFS_I(mapping->host)->io_tree;
8734 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8738 btrfs_readpages(struct file *file, struct address_space *mapping,
8739 struct list_head *pages, unsigned nr_pages)
8741 struct extent_io_tree *tree;
8742 tree = &BTRFS_I(mapping->host)->io_tree;
8743 return extent_readpages(tree, mapping, pages, nr_pages,
8746 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8748 struct extent_io_tree *tree;
8749 struct extent_map_tree *map;
8752 tree = &BTRFS_I(page->mapping->host)->io_tree;
8753 map = &BTRFS_I(page->mapping->host)->extent_tree;
8754 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8756 ClearPagePrivate(page);
8757 set_page_private(page, 0);
8763 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8765 if (PageWriteback(page) || PageDirty(page))
8767 return __btrfs_releasepage(page, gfp_flags);
8770 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8771 unsigned int length)
8773 struct inode *inode = page->mapping->host;
8774 struct extent_io_tree *tree;
8775 struct btrfs_ordered_extent *ordered;
8776 struct extent_state *cached_state = NULL;
8777 u64 page_start = page_offset(page);
8778 u64 page_end = page_start + PAGE_SIZE - 1;
8781 int inode_evicting = inode->i_state & I_FREEING;
8784 * we have the page locked, so new writeback can't start,
8785 * and the dirty bit won't be cleared while we are here.
8787 * Wait for IO on this page so that we can safely clear
8788 * the PagePrivate2 bit and do ordered accounting
8790 wait_on_page_writeback(page);
8792 tree = &BTRFS_I(inode)->io_tree;
8794 btrfs_releasepage(page, GFP_NOFS);
8798 if (!inode_evicting)
8799 lock_extent_bits(tree, page_start, page_end, &cached_state);
8802 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8803 page_end - start + 1);
8805 end = min(page_end, ordered->file_offset + ordered->len - 1);
8807 * IO on this page will never be started, so we need
8808 * to account for any ordered extents now
8810 if (!inode_evicting)
8811 clear_extent_bit(tree, start, end,
8812 EXTENT_DIRTY | EXTENT_DELALLOC |
8813 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8814 EXTENT_DEFRAG, 1, 0, &cached_state,
8817 * whoever cleared the private bit is responsible
8818 * for the finish_ordered_io
8820 if (TestClearPagePrivate2(page)) {
8821 struct btrfs_ordered_inode_tree *tree;
8824 tree = &BTRFS_I(inode)->ordered_tree;
8826 spin_lock_irq(&tree->lock);
8827 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8828 new_len = start - ordered->file_offset;
8829 if (new_len < ordered->truncated_len)
8830 ordered->truncated_len = new_len;
8831 spin_unlock_irq(&tree->lock);
8833 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8835 end - start + 1, 1))
8836 btrfs_finish_ordered_io(ordered);
8838 btrfs_put_ordered_extent(ordered);
8839 if (!inode_evicting) {
8840 cached_state = NULL;
8841 lock_extent_bits(tree, start, end,
8846 if (start < page_end)
8851 * Qgroup reserved space handler
8852 * Page here will be either
8853 * 1) Already written to disk
8854 * In this case, its reserved space is released from data rsv map
8855 * and will be freed by delayed_ref handler finally.
8856 * So even we call qgroup_free_data(), it won't decrease reserved
8858 * 2) Not written to disk
8859 * This means the reserved space should be freed here. However,
8860 * if a truncate invalidates the page (by clearing PageDirty)
8861 * and the page is accounted for while allocating extent
8862 * in btrfs_check_data_free_space() we let delayed_ref to
8863 * free the entire extent.
8865 if (PageDirty(page))
8866 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8867 if (!inode_evicting) {
8868 clear_extent_bit(tree, page_start, page_end,
8869 EXTENT_LOCKED | EXTENT_DIRTY |
8870 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8871 EXTENT_DEFRAG, 1, 1,
8872 &cached_state, GFP_NOFS);
8874 __btrfs_releasepage(page, GFP_NOFS);
8877 ClearPageChecked(page);
8878 if (PagePrivate(page)) {
8879 ClearPagePrivate(page);
8880 set_page_private(page, 0);
8886 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8887 * called from a page fault handler when a page is first dirtied. Hence we must
8888 * be careful to check for EOF conditions here. We set the page up correctly
8889 * for a written page which means we get ENOSPC checking when writing into
8890 * holes and correct delalloc and unwritten extent mapping on filesystems that
8891 * support these features.
8893 * We are not allowed to take the i_mutex here so we have to play games to
8894 * protect against truncate races as the page could now be beyond EOF. Because
8895 * vmtruncate() writes the inode size before removing pages, once we have the
8896 * page lock we can determine safely if the page is beyond EOF. If it is not
8897 * beyond EOF, then the page is guaranteed safe against truncation until we
8900 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8902 struct page *page = vmf->page;
8903 struct inode *inode = file_inode(vma->vm_file);
8904 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8905 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8906 struct btrfs_ordered_extent *ordered;
8907 struct extent_state *cached_state = NULL;
8909 unsigned long zero_start;
8918 reserved_space = PAGE_SIZE;
8920 sb_start_pagefault(inode->i_sb);
8921 page_start = page_offset(page);
8922 page_end = page_start + PAGE_SIZE - 1;
8926 * Reserving delalloc space after obtaining the page lock can lead to
8927 * deadlock. For example, if a dirty page is locked by this function
8928 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8929 * dirty page write out, then the btrfs_writepage() function could
8930 * end up waiting indefinitely to get a lock on the page currently
8931 * being processed by btrfs_page_mkwrite() function.
8933 ret = btrfs_delalloc_reserve_space(inode, page_start,
8936 ret = file_update_time(vma->vm_file);
8942 else /* -ENOSPC, -EIO, etc */
8943 ret = VM_FAULT_SIGBUS;
8949 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8952 size = i_size_read(inode);
8954 if ((page->mapping != inode->i_mapping) ||
8955 (page_start >= size)) {
8956 /* page got truncated out from underneath us */
8959 wait_on_page_writeback(page);
8961 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8962 set_page_extent_mapped(page);
8965 * we can't set the delalloc bits if there are pending ordered
8966 * extents. Drop our locks and wait for them to finish
8968 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8971 unlock_extent_cached(io_tree, page_start, page_end,
8972 &cached_state, GFP_NOFS);
8974 btrfs_start_ordered_extent(inode, ordered, 1);
8975 btrfs_put_ordered_extent(ordered);
8979 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8980 reserved_space = round_up(size - page_start,
8981 fs_info->sectorsize);
8982 if (reserved_space < PAGE_SIZE) {
8983 end = page_start + reserved_space - 1;
8984 spin_lock(&BTRFS_I(inode)->lock);
8985 BTRFS_I(inode)->outstanding_extents++;
8986 spin_unlock(&BTRFS_I(inode)->lock);
8987 btrfs_delalloc_release_space(inode, page_start,
8988 PAGE_SIZE - reserved_space);
8993 * page_mkwrite gets called when the page is firstly dirtied after it's
8994 * faulted in, but write(2) could also dirty a page and set delalloc
8995 * bits, thus in this case for space account reason, we still need to
8996 * clear any delalloc bits within this page range since we have to
8997 * reserve data&meta space before lock_page() (see above comments).
8999 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9000 EXTENT_DIRTY | EXTENT_DELALLOC |
9001 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9002 0, 0, &cached_state, GFP_NOFS);
9004 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9007 unlock_extent_cached(io_tree, page_start, page_end,
9008 &cached_state, GFP_NOFS);
9009 ret = VM_FAULT_SIGBUS;
9014 /* page is wholly or partially inside EOF */
9015 if (page_start + PAGE_SIZE > size)
9016 zero_start = size & ~PAGE_MASK;
9018 zero_start = PAGE_SIZE;
9020 if (zero_start != PAGE_SIZE) {
9022 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9023 flush_dcache_page(page);
9026 ClearPageChecked(page);
9027 set_page_dirty(page);
9028 SetPageUptodate(page);
9030 BTRFS_I(inode)->last_trans = fs_info->generation;
9031 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9032 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9034 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9038 sb_end_pagefault(inode->i_sb);
9039 return VM_FAULT_LOCKED;
9043 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9045 sb_end_pagefault(inode->i_sb);
9049 static int btrfs_truncate(struct inode *inode)
9051 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9052 struct btrfs_root *root = BTRFS_I(inode)->root;
9053 struct btrfs_block_rsv *rsv;
9056 struct btrfs_trans_handle *trans;
9057 u64 mask = fs_info->sectorsize - 1;
9058 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9060 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9066 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9067 * 3 things going on here
9069 * 1) We need to reserve space for our orphan item and the space to
9070 * delete our orphan item. Lord knows we don't want to have a dangling
9071 * orphan item because we didn't reserve space to remove it.
9073 * 2) We need to reserve space to update our inode.
9075 * 3) We need to have something to cache all the space that is going to
9076 * be free'd up by the truncate operation, but also have some slack
9077 * space reserved in case it uses space during the truncate (thank you
9078 * very much snapshotting).
9080 * And we need these to all be separate. The fact is we can use a lot of
9081 * space doing the truncate, and we have no earthly idea how much space
9082 * we will use, so we need the truncate reservation to be separate so it
9083 * doesn't end up using space reserved for updating the inode or
9084 * removing the orphan item. We also need to be able to stop the
9085 * transaction and start a new one, which means we need to be able to
9086 * update the inode several times, and we have no idea of knowing how
9087 * many times that will be, so we can't just reserve 1 item for the
9088 * entirety of the operation, so that has to be done separately as well.
9089 * Then there is the orphan item, which does indeed need to be held on
9090 * to for the whole operation, and we need nobody to touch this reserved
9091 * space except the orphan code.
9093 * So that leaves us with
9095 * 1) root->orphan_block_rsv - for the orphan deletion.
9096 * 2) rsv - for the truncate reservation, which we will steal from the
9097 * transaction reservation.
9098 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9099 * updating the inode.
9101 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9104 rsv->size = min_size;
9108 * 1 for the truncate slack space
9109 * 1 for updating the inode.
9111 trans = btrfs_start_transaction(root, 2);
9112 if (IS_ERR(trans)) {
9113 err = PTR_ERR(trans);
9117 /* Migrate the slack space for the truncate to our reserve */
9118 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9123 * So if we truncate and then write and fsync we normally would just
9124 * write the extents that changed, which is a problem if we need to
9125 * first truncate that entire inode. So set this flag so we write out
9126 * all of the extents in the inode to the sync log so we're completely
9129 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9130 trans->block_rsv = rsv;
9133 ret = btrfs_truncate_inode_items(trans, root, inode,
9135 BTRFS_EXTENT_DATA_KEY);
9136 if (ret != -ENOSPC && ret != -EAGAIN) {
9141 trans->block_rsv = &fs_info->trans_block_rsv;
9142 ret = btrfs_update_inode(trans, root, inode);
9148 btrfs_end_transaction(trans);
9149 btrfs_btree_balance_dirty(fs_info);
9151 trans = btrfs_start_transaction(root, 2);
9152 if (IS_ERR(trans)) {
9153 ret = err = PTR_ERR(trans);
9158 btrfs_block_rsv_release(fs_info, rsv, -1);
9159 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9161 BUG_ON(ret); /* shouldn't happen */
9162 trans->block_rsv = rsv;
9165 if (ret == 0 && inode->i_nlink > 0) {
9166 trans->block_rsv = root->orphan_block_rsv;
9167 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9173 trans->block_rsv = &fs_info->trans_block_rsv;
9174 ret = btrfs_update_inode(trans, root, inode);
9178 ret = btrfs_end_transaction(trans);
9179 btrfs_btree_balance_dirty(fs_info);
9182 btrfs_free_block_rsv(fs_info, rsv);
9191 * create a new subvolume directory/inode (helper for the ioctl).
9193 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9194 struct btrfs_root *new_root,
9195 struct btrfs_root *parent_root,
9198 struct inode *inode;
9202 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9203 new_dirid, new_dirid,
9204 S_IFDIR | (~current_umask() & S_IRWXUGO),
9207 return PTR_ERR(inode);
9208 inode->i_op = &btrfs_dir_inode_operations;
9209 inode->i_fop = &btrfs_dir_file_operations;
9211 set_nlink(inode, 1);
9212 btrfs_i_size_write(BTRFS_I(inode), 0);
9213 unlock_new_inode(inode);
9215 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9217 btrfs_err(new_root->fs_info,
9218 "error inheriting subvolume %llu properties: %d",
9219 new_root->root_key.objectid, err);
9221 err = btrfs_update_inode(trans, new_root, inode);
9227 struct inode *btrfs_alloc_inode(struct super_block *sb)
9229 struct btrfs_inode *ei;
9230 struct inode *inode;
9232 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9239 ei->last_sub_trans = 0;
9240 ei->logged_trans = 0;
9241 ei->delalloc_bytes = 0;
9242 ei->defrag_bytes = 0;
9243 ei->disk_i_size = 0;
9246 ei->index_cnt = (u64)-1;
9248 ei->last_unlink_trans = 0;
9249 ei->last_log_commit = 0;
9250 ei->delayed_iput_count = 0;
9252 spin_lock_init(&ei->lock);
9253 ei->outstanding_extents = 0;
9254 ei->reserved_extents = 0;
9256 ei->runtime_flags = 0;
9257 ei->force_compress = BTRFS_COMPRESS_NONE;
9259 ei->delayed_node = NULL;
9261 ei->i_otime.tv_sec = 0;
9262 ei->i_otime.tv_nsec = 0;
9264 inode = &ei->vfs_inode;
9265 extent_map_tree_init(&ei->extent_tree);
9266 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9267 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9268 ei->io_tree.track_uptodate = 1;
9269 ei->io_failure_tree.track_uptodate = 1;
9270 atomic_set(&ei->sync_writers, 0);
9271 mutex_init(&ei->log_mutex);
9272 mutex_init(&ei->delalloc_mutex);
9273 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9274 INIT_LIST_HEAD(&ei->delalloc_inodes);
9275 INIT_LIST_HEAD(&ei->delayed_iput);
9276 RB_CLEAR_NODE(&ei->rb_node);
9277 init_rwsem(&ei->dio_sem);
9282 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9283 void btrfs_test_destroy_inode(struct inode *inode)
9285 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9286 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9290 static void btrfs_i_callback(struct rcu_head *head)
9292 struct inode *inode = container_of(head, struct inode, i_rcu);
9293 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9296 void btrfs_destroy_inode(struct inode *inode)
9298 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9299 struct btrfs_ordered_extent *ordered;
9300 struct btrfs_root *root = BTRFS_I(inode)->root;
9302 WARN_ON(!hlist_empty(&inode->i_dentry));
9303 WARN_ON(inode->i_data.nrpages);
9304 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9305 WARN_ON(BTRFS_I(inode)->reserved_extents);
9306 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9307 WARN_ON(BTRFS_I(inode)->csum_bytes);
9308 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9311 * This can happen where we create an inode, but somebody else also
9312 * created the same inode and we need to destroy the one we already
9318 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9319 &BTRFS_I(inode)->runtime_flags)) {
9320 btrfs_info(fs_info, "inode %llu still on the orphan list",
9321 btrfs_ino(BTRFS_I(inode)));
9322 atomic_dec(&root->orphan_inodes);
9326 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9331 "found ordered extent %llu %llu on inode cleanup",
9332 ordered->file_offset, ordered->len);
9333 btrfs_remove_ordered_extent(inode, ordered);
9334 btrfs_put_ordered_extent(ordered);
9335 btrfs_put_ordered_extent(ordered);
9338 btrfs_qgroup_check_reserved_leak(inode);
9339 inode_tree_del(inode);
9340 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9342 call_rcu(&inode->i_rcu, btrfs_i_callback);
9345 int btrfs_drop_inode(struct inode *inode)
9347 struct btrfs_root *root = BTRFS_I(inode)->root;
9352 /* the snap/subvol tree is on deleting */
9353 if (btrfs_root_refs(&root->root_item) == 0)
9356 return generic_drop_inode(inode);
9359 static void init_once(void *foo)
9361 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9363 inode_init_once(&ei->vfs_inode);
9366 void btrfs_destroy_cachep(void)
9369 * Make sure all delayed rcu free inodes are flushed before we
9373 kmem_cache_destroy(btrfs_inode_cachep);
9374 kmem_cache_destroy(btrfs_trans_handle_cachep);
9375 kmem_cache_destroy(btrfs_transaction_cachep);
9376 kmem_cache_destroy(btrfs_path_cachep);
9377 kmem_cache_destroy(btrfs_free_space_cachep);
9380 int btrfs_init_cachep(void)
9382 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9383 sizeof(struct btrfs_inode), 0,
9384 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9386 if (!btrfs_inode_cachep)
9389 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9390 sizeof(struct btrfs_trans_handle), 0,
9391 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9392 if (!btrfs_trans_handle_cachep)
9395 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9396 sizeof(struct btrfs_transaction), 0,
9397 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9398 if (!btrfs_transaction_cachep)
9401 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9402 sizeof(struct btrfs_path), 0,
9403 SLAB_MEM_SPREAD, NULL);
9404 if (!btrfs_path_cachep)
9407 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9408 sizeof(struct btrfs_free_space), 0,
9409 SLAB_MEM_SPREAD, NULL);
9410 if (!btrfs_free_space_cachep)
9415 btrfs_destroy_cachep();
9419 static int btrfs_getattr(struct vfsmount *mnt,
9420 struct dentry *dentry, struct kstat *stat)
9423 struct inode *inode = d_inode(dentry);
9424 u32 blocksize = inode->i_sb->s_blocksize;
9426 generic_fillattr(inode, stat);
9427 stat->dev = BTRFS_I(inode)->root->anon_dev;
9429 spin_lock(&BTRFS_I(inode)->lock);
9430 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9431 spin_unlock(&BTRFS_I(inode)->lock);
9432 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9433 ALIGN(delalloc_bytes, blocksize)) >> 9;
9437 static int btrfs_rename_exchange(struct inode *old_dir,
9438 struct dentry *old_dentry,
9439 struct inode *new_dir,
9440 struct dentry *new_dentry)
9442 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9443 struct btrfs_trans_handle *trans;
9444 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9445 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9446 struct inode *new_inode = new_dentry->d_inode;
9447 struct inode *old_inode = old_dentry->d_inode;
9448 struct timespec ctime = current_time(old_inode);
9449 struct dentry *parent;
9450 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9451 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9456 bool root_log_pinned = false;
9457 bool dest_log_pinned = false;
9459 /* we only allow rename subvolume link between subvolumes */
9460 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9463 /* close the race window with snapshot create/destroy ioctl */
9464 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9465 down_read(&fs_info->subvol_sem);
9466 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9467 down_read(&fs_info->subvol_sem);
9470 * We want to reserve the absolute worst case amount of items. So if
9471 * both inodes are subvols and we need to unlink them then that would
9472 * require 4 item modifications, but if they are both normal inodes it
9473 * would require 5 item modifications, so we'll assume their normal
9474 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9475 * should cover the worst case number of items we'll modify.
9477 trans = btrfs_start_transaction(root, 12);
9478 if (IS_ERR(trans)) {
9479 ret = PTR_ERR(trans);
9484 * We need to find a free sequence number both in the source and
9485 * in the destination directory for the exchange.
9487 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9490 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9494 BTRFS_I(old_inode)->dir_index = 0ULL;
9495 BTRFS_I(new_inode)->dir_index = 0ULL;
9497 /* Reference for the source. */
9498 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9499 /* force full log commit if subvolume involved. */
9500 btrfs_set_log_full_commit(fs_info, trans);
9502 btrfs_pin_log_trans(root);
9503 root_log_pinned = true;
9504 ret = btrfs_insert_inode_ref(trans, dest,
9505 new_dentry->d_name.name,
9506 new_dentry->d_name.len,
9508 btrfs_ino(BTRFS_I(new_dir)),
9514 /* And now for the dest. */
9515 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9516 /* force full log commit if subvolume involved. */
9517 btrfs_set_log_full_commit(fs_info, trans);
9519 btrfs_pin_log_trans(dest);
9520 dest_log_pinned = true;
9521 ret = btrfs_insert_inode_ref(trans, root,
9522 old_dentry->d_name.name,
9523 old_dentry->d_name.len,
9525 btrfs_ino(BTRFS_I(old_dir)),
9531 /* Update inode version and ctime/mtime. */
9532 inode_inc_iversion(old_dir);
9533 inode_inc_iversion(new_dir);
9534 inode_inc_iversion(old_inode);
9535 inode_inc_iversion(new_inode);
9536 old_dir->i_ctime = old_dir->i_mtime = ctime;
9537 new_dir->i_ctime = new_dir->i_mtime = ctime;
9538 old_inode->i_ctime = ctime;
9539 new_inode->i_ctime = ctime;
9541 if (old_dentry->d_parent != new_dentry->d_parent) {
9542 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9543 BTRFS_I(old_inode), 1);
9544 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9545 BTRFS_I(new_inode), 1);
9548 /* src is a subvolume */
9549 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9550 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9551 ret = btrfs_unlink_subvol(trans, root, old_dir,
9553 old_dentry->d_name.name,
9554 old_dentry->d_name.len);
9555 } else { /* src is an inode */
9556 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9557 BTRFS_I(old_dentry->d_inode),
9558 old_dentry->d_name.name,
9559 old_dentry->d_name.len);
9561 ret = btrfs_update_inode(trans, root, old_inode);
9564 btrfs_abort_transaction(trans, ret);
9568 /* dest is a subvolume */
9569 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9570 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9571 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9573 new_dentry->d_name.name,
9574 new_dentry->d_name.len);
9575 } else { /* dest is an inode */
9576 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9577 BTRFS_I(new_dentry->d_inode),
9578 new_dentry->d_name.name,
9579 new_dentry->d_name.len);
9581 ret = btrfs_update_inode(trans, dest, new_inode);
9584 btrfs_abort_transaction(trans, ret);
9588 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9589 new_dentry->d_name.name,
9590 new_dentry->d_name.len, 0, old_idx);
9592 btrfs_abort_transaction(trans, ret);
9596 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9597 old_dentry->d_name.name,
9598 old_dentry->d_name.len, 0, new_idx);
9600 btrfs_abort_transaction(trans, ret);
9604 if (old_inode->i_nlink == 1)
9605 BTRFS_I(old_inode)->dir_index = old_idx;
9606 if (new_inode->i_nlink == 1)
9607 BTRFS_I(new_inode)->dir_index = new_idx;
9609 if (root_log_pinned) {
9610 parent = new_dentry->d_parent;
9611 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9613 btrfs_end_log_trans(root);
9614 root_log_pinned = false;
9616 if (dest_log_pinned) {
9617 parent = old_dentry->d_parent;
9618 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9620 btrfs_end_log_trans(dest);
9621 dest_log_pinned = false;
9625 * If we have pinned a log and an error happened, we unpin tasks
9626 * trying to sync the log and force them to fallback to a transaction
9627 * commit if the log currently contains any of the inodes involved in
9628 * this rename operation (to ensure we do not persist a log with an
9629 * inconsistent state for any of these inodes or leading to any
9630 * inconsistencies when replayed). If the transaction was aborted, the
9631 * abortion reason is propagated to userspace when attempting to commit
9632 * the transaction. If the log does not contain any of these inodes, we
9633 * allow the tasks to sync it.
9635 if (ret && (root_log_pinned || dest_log_pinned)) {
9636 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9637 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9638 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9640 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9641 btrfs_set_log_full_commit(fs_info, trans);
9643 if (root_log_pinned) {
9644 btrfs_end_log_trans(root);
9645 root_log_pinned = false;
9647 if (dest_log_pinned) {
9648 btrfs_end_log_trans(dest);
9649 dest_log_pinned = false;
9652 ret = btrfs_end_transaction(trans);
9654 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9655 up_read(&fs_info->subvol_sem);
9656 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9657 up_read(&fs_info->subvol_sem);
9662 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9663 struct btrfs_root *root,
9665 struct dentry *dentry)
9668 struct inode *inode;
9672 ret = btrfs_find_free_ino(root, &objectid);
9676 inode = btrfs_new_inode(trans, root, dir,
9677 dentry->d_name.name,
9679 btrfs_ino(BTRFS_I(dir)),
9681 S_IFCHR | WHITEOUT_MODE,
9684 if (IS_ERR(inode)) {
9685 ret = PTR_ERR(inode);
9689 inode->i_op = &btrfs_special_inode_operations;
9690 init_special_inode(inode, inode->i_mode,
9693 ret = btrfs_init_inode_security(trans, inode, dir,
9698 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9699 BTRFS_I(inode), 0, index);
9703 ret = btrfs_update_inode(trans, root, inode);
9705 unlock_new_inode(inode);
9707 inode_dec_link_count(inode);
9713 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9714 struct inode *new_dir, struct dentry *new_dentry,
9717 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9718 struct btrfs_trans_handle *trans;
9719 unsigned int trans_num_items;
9720 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9721 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9722 struct inode *new_inode = d_inode(new_dentry);
9723 struct inode *old_inode = d_inode(old_dentry);
9727 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9728 bool log_pinned = false;
9730 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9733 /* we only allow rename subvolume link between subvolumes */
9734 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9737 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9738 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9741 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9742 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9746 /* check for collisions, even if the name isn't there */
9747 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9748 new_dentry->d_name.name,
9749 new_dentry->d_name.len);
9752 if (ret == -EEXIST) {
9754 * eexist without a new_inode */
9755 if (WARN_ON(!new_inode)) {
9759 /* maybe -EOVERFLOW */
9766 * we're using rename to replace one file with another. Start IO on it
9767 * now so we don't add too much work to the end of the transaction
9769 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9770 filemap_flush(old_inode->i_mapping);
9772 /* close the racy window with snapshot create/destroy ioctl */
9773 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9774 down_read(&fs_info->subvol_sem);
9776 * We want to reserve the absolute worst case amount of items. So if
9777 * both inodes are subvols and we need to unlink them then that would
9778 * require 4 item modifications, but if they are both normal inodes it
9779 * would require 5 item modifications, so we'll assume they are normal
9780 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9781 * should cover the worst case number of items we'll modify.
9782 * If our rename has the whiteout flag, we need more 5 units for the
9783 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9784 * when selinux is enabled).
9786 trans_num_items = 11;
9787 if (flags & RENAME_WHITEOUT)
9788 trans_num_items += 5;
9789 trans = btrfs_start_transaction(root, trans_num_items);
9790 if (IS_ERR(trans)) {
9791 ret = PTR_ERR(trans);
9796 btrfs_record_root_in_trans(trans, dest);
9798 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9802 BTRFS_I(old_inode)->dir_index = 0ULL;
9803 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9804 /* force full log commit if subvolume involved. */
9805 btrfs_set_log_full_commit(fs_info, trans);
9807 btrfs_pin_log_trans(root);
9809 ret = btrfs_insert_inode_ref(trans, dest,
9810 new_dentry->d_name.name,
9811 new_dentry->d_name.len,
9813 btrfs_ino(BTRFS_I(new_dir)), index);
9818 inode_inc_iversion(old_dir);
9819 inode_inc_iversion(new_dir);
9820 inode_inc_iversion(old_inode);
9821 old_dir->i_ctime = old_dir->i_mtime =
9822 new_dir->i_ctime = new_dir->i_mtime =
9823 old_inode->i_ctime = current_time(old_dir);
9825 if (old_dentry->d_parent != new_dentry->d_parent)
9826 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9827 BTRFS_I(old_inode), 1);
9829 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9830 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9831 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9832 old_dentry->d_name.name,
9833 old_dentry->d_name.len);
9835 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9836 BTRFS_I(d_inode(old_dentry)),
9837 old_dentry->d_name.name,
9838 old_dentry->d_name.len);
9840 ret = btrfs_update_inode(trans, root, old_inode);
9843 btrfs_abort_transaction(trans, ret);
9848 inode_inc_iversion(new_inode);
9849 new_inode->i_ctime = current_time(new_inode);
9850 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9851 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9852 root_objectid = BTRFS_I(new_inode)->location.objectid;
9853 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9855 new_dentry->d_name.name,
9856 new_dentry->d_name.len);
9857 BUG_ON(new_inode->i_nlink == 0);
9859 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9860 BTRFS_I(d_inode(new_dentry)),
9861 new_dentry->d_name.name,
9862 new_dentry->d_name.len);
9864 if (!ret && new_inode->i_nlink == 0)
9865 ret = btrfs_orphan_add(trans,
9866 BTRFS_I(d_inode(new_dentry)));
9868 btrfs_abort_transaction(trans, ret);
9873 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9874 new_dentry->d_name.name,
9875 new_dentry->d_name.len, 0, index);
9877 btrfs_abort_transaction(trans, ret);
9881 if (old_inode->i_nlink == 1)
9882 BTRFS_I(old_inode)->dir_index = index;
9885 struct dentry *parent = new_dentry->d_parent;
9887 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9889 btrfs_end_log_trans(root);
9893 if (flags & RENAME_WHITEOUT) {
9894 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9898 btrfs_abort_transaction(trans, ret);
9904 * If we have pinned the log and an error happened, we unpin tasks
9905 * trying to sync the log and force them to fallback to a transaction
9906 * commit if the log currently contains any of the inodes involved in
9907 * this rename operation (to ensure we do not persist a log with an
9908 * inconsistent state for any of these inodes or leading to any
9909 * inconsistencies when replayed). If the transaction was aborted, the
9910 * abortion reason is propagated to userspace when attempting to commit
9911 * the transaction. If the log does not contain any of these inodes, we
9912 * allow the tasks to sync it.
9914 if (ret && log_pinned) {
9915 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9916 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9917 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9919 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9920 btrfs_set_log_full_commit(fs_info, trans);
9922 btrfs_end_log_trans(root);
9925 btrfs_end_transaction(trans);
9927 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9928 up_read(&fs_info->subvol_sem);
9933 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9934 struct inode *new_dir, struct dentry *new_dentry,
9937 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9940 if (flags & RENAME_EXCHANGE)
9941 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9944 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9947 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9949 struct btrfs_delalloc_work *delalloc_work;
9950 struct inode *inode;
9952 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9954 inode = delalloc_work->inode;
9955 filemap_flush(inode->i_mapping);
9956 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9957 &BTRFS_I(inode)->runtime_flags))
9958 filemap_flush(inode->i_mapping);
9960 if (delalloc_work->delay_iput)
9961 btrfs_add_delayed_iput(inode);
9964 complete(&delalloc_work->completion);
9967 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9970 struct btrfs_delalloc_work *work;
9972 work = kmalloc(sizeof(*work), GFP_NOFS);
9976 init_completion(&work->completion);
9977 INIT_LIST_HEAD(&work->list);
9978 work->inode = inode;
9979 work->delay_iput = delay_iput;
9980 WARN_ON_ONCE(!inode);
9981 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9982 btrfs_run_delalloc_work, NULL, NULL);
9987 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9989 wait_for_completion(&work->completion);
9994 * some fairly slow code that needs optimization. This walks the list
9995 * of all the inodes with pending delalloc and forces them to disk.
9997 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10000 struct btrfs_inode *binode;
10001 struct inode *inode;
10002 struct btrfs_delalloc_work *work, *next;
10003 struct list_head works;
10004 struct list_head splice;
10007 INIT_LIST_HEAD(&works);
10008 INIT_LIST_HEAD(&splice);
10010 mutex_lock(&root->delalloc_mutex);
10011 spin_lock(&root->delalloc_lock);
10012 list_splice_init(&root->delalloc_inodes, &splice);
10013 while (!list_empty(&splice)) {
10014 binode = list_entry(splice.next, struct btrfs_inode,
10017 list_move_tail(&binode->delalloc_inodes,
10018 &root->delalloc_inodes);
10019 inode = igrab(&binode->vfs_inode);
10021 cond_resched_lock(&root->delalloc_lock);
10024 spin_unlock(&root->delalloc_lock);
10026 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10029 btrfs_add_delayed_iput(inode);
10035 list_add_tail(&work->list, &works);
10036 btrfs_queue_work(root->fs_info->flush_workers,
10039 if (nr != -1 && ret >= nr)
10042 spin_lock(&root->delalloc_lock);
10044 spin_unlock(&root->delalloc_lock);
10047 list_for_each_entry_safe(work, next, &works, list) {
10048 list_del_init(&work->list);
10049 btrfs_wait_and_free_delalloc_work(work);
10052 if (!list_empty_careful(&splice)) {
10053 spin_lock(&root->delalloc_lock);
10054 list_splice_tail(&splice, &root->delalloc_inodes);
10055 spin_unlock(&root->delalloc_lock);
10057 mutex_unlock(&root->delalloc_mutex);
10061 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10063 struct btrfs_fs_info *fs_info = root->fs_info;
10066 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10069 ret = __start_delalloc_inodes(root, delay_iput, -1);
10073 * the filemap_flush will queue IO into the worker threads, but
10074 * we have to make sure the IO is actually started and that
10075 * ordered extents get created before we return
10077 atomic_inc(&fs_info->async_submit_draining);
10078 while (atomic_read(&fs_info->nr_async_submits) ||
10079 atomic_read(&fs_info->async_delalloc_pages)) {
10080 wait_event(fs_info->async_submit_wait,
10081 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10082 atomic_read(&fs_info->async_delalloc_pages) == 0));
10084 atomic_dec(&fs_info->async_submit_draining);
10088 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10091 struct btrfs_root *root;
10092 struct list_head splice;
10095 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10098 INIT_LIST_HEAD(&splice);
10100 mutex_lock(&fs_info->delalloc_root_mutex);
10101 spin_lock(&fs_info->delalloc_root_lock);
10102 list_splice_init(&fs_info->delalloc_roots, &splice);
10103 while (!list_empty(&splice) && nr) {
10104 root = list_first_entry(&splice, struct btrfs_root,
10106 root = btrfs_grab_fs_root(root);
10108 list_move_tail(&root->delalloc_root,
10109 &fs_info->delalloc_roots);
10110 spin_unlock(&fs_info->delalloc_root_lock);
10112 ret = __start_delalloc_inodes(root, delay_iput, nr);
10113 btrfs_put_fs_root(root);
10121 spin_lock(&fs_info->delalloc_root_lock);
10123 spin_unlock(&fs_info->delalloc_root_lock);
10126 atomic_inc(&fs_info->async_submit_draining);
10127 while (atomic_read(&fs_info->nr_async_submits) ||
10128 atomic_read(&fs_info->async_delalloc_pages)) {
10129 wait_event(fs_info->async_submit_wait,
10130 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10131 atomic_read(&fs_info->async_delalloc_pages) == 0));
10133 atomic_dec(&fs_info->async_submit_draining);
10135 if (!list_empty_careful(&splice)) {
10136 spin_lock(&fs_info->delalloc_root_lock);
10137 list_splice_tail(&splice, &fs_info->delalloc_roots);
10138 spin_unlock(&fs_info->delalloc_root_lock);
10140 mutex_unlock(&fs_info->delalloc_root_mutex);
10144 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10145 const char *symname)
10147 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10148 struct btrfs_trans_handle *trans;
10149 struct btrfs_root *root = BTRFS_I(dir)->root;
10150 struct btrfs_path *path;
10151 struct btrfs_key key;
10152 struct inode *inode = NULL;
10154 int drop_inode = 0;
10160 struct btrfs_file_extent_item *ei;
10161 struct extent_buffer *leaf;
10163 name_len = strlen(symname);
10164 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10165 return -ENAMETOOLONG;
10168 * 2 items for inode item and ref
10169 * 2 items for dir items
10170 * 1 item for updating parent inode item
10171 * 1 item for the inline extent item
10172 * 1 item for xattr if selinux is on
10174 trans = btrfs_start_transaction(root, 7);
10176 return PTR_ERR(trans);
10178 err = btrfs_find_free_ino(root, &objectid);
10182 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10183 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10184 objectid, S_IFLNK|S_IRWXUGO, &index);
10185 if (IS_ERR(inode)) {
10186 err = PTR_ERR(inode);
10191 * If the active LSM wants to access the inode during
10192 * d_instantiate it needs these. Smack checks to see
10193 * if the filesystem supports xattrs by looking at the
10196 inode->i_fop = &btrfs_file_operations;
10197 inode->i_op = &btrfs_file_inode_operations;
10198 inode->i_mapping->a_ops = &btrfs_aops;
10199 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10201 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10203 goto out_unlock_inode;
10205 path = btrfs_alloc_path();
10208 goto out_unlock_inode;
10210 key.objectid = btrfs_ino(BTRFS_I(inode));
10212 key.type = BTRFS_EXTENT_DATA_KEY;
10213 datasize = btrfs_file_extent_calc_inline_size(name_len);
10214 err = btrfs_insert_empty_item(trans, root, path, &key,
10217 btrfs_free_path(path);
10218 goto out_unlock_inode;
10220 leaf = path->nodes[0];
10221 ei = btrfs_item_ptr(leaf, path->slots[0],
10222 struct btrfs_file_extent_item);
10223 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10224 btrfs_set_file_extent_type(leaf, ei,
10225 BTRFS_FILE_EXTENT_INLINE);
10226 btrfs_set_file_extent_encryption(leaf, ei, 0);
10227 btrfs_set_file_extent_compression(leaf, ei, 0);
10228 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10229 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10231 ptr = btrfs_file_extent_inline_start(ei);
10232 write_extent_buffer(leaf, symname, ptr, name_len);
10233 btrfs_mark_buffer_dirty(leaf);
10234 btrfs_free_path(path);
10236 inode->i_op = &btrfs_symlink_inode_operations;
10237 inode_nohighmem(inode);
10238 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10239 inode_set_bytes(inode, name_len);
10240 btrfs_i_size_write(BTRFS_I(inode), name_len);
10241 err = btrfs_update_inode(trans, root, inode);
10243 * Last step, add directory indexes for our symlink inode. This is the
10244 * last step to avoid extra cleanup of these indexes if an error happens
10248 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10249 BTRFS_I(inode), 0, index);
10252 goto out_unlock_inode;
10255 unlock_new_inode(inode);
10256 d_instantiate(dentry, inode);
10259 btrfs_end_transaction(trans);
10261 inode_dec_link_count(inode);
10264 btrfs_btree_balance_dirty(fs_info);
10269 unlock_new_inode(inode);
10273 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10274 u64 start, u64 num_bytes, u64 min_size,
10275 loff_t actual_len, u64 *alloc_hint,
10276 struct btrfs_trans_handle *trans)
10278 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10279 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10280 struct extent_map *em;
10281 struct btrfs_root *root = BTRFS_I(inode)->root;
10282 struct btrfs_key ins;
10283 u64 cur_offset = start;
10286 u64 last_alloc = (u64)-1;
10288 bool own_trans = true;
10289 u64 end = start + num_bytes - 1;
10293 while (num_bytes > 0) {
10295 trans = btrfs_start_transaction(root, 3);
10296 if (IS_ERR(trans)) {
10297 ret = PTR_ERR(trans);
10302 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10303 cur_bytes = max(cur_bytes, min_size);
10305 * If we are severely fragmented we could end up with really
10306 * small allocations, so if the allocator is returning small
10307 * chunks lets make its job easier by only searching for those
10310 cur_bytes = min(cur_bytes, last_alloc);
10311 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10312 min_size, 0, *alloc_hint, &ins, 1, 0);
10315 btrfs_end_transaction(trans);
10318 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10320 last_alloc = ins.offset;
10321 ret = insert_reserved_file_extent(trans, inode,
10322 cur_offset, ins.objectid,
10323 ins.offset, ins.offset,
10324 ins.offset, 0, 0, 0,
10325 BTRFS_FILE_EXTENT_PREALLOC);
10327 btrfs_free_reserved_extent(fs_info, ins.objectid,
10329 btrfs_abort_transaction(trans, ret);
10331 btrfs_end_transaction(trans);
10335 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10336 cur_offset + ins.offset -1, 0);
10338 em = alloc_extent_map();
10340 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10341 &BTRFS_I(inode)->runtime_flags);
10345 em->start = cur_offset;
10346 em->orig_start = cur_offset;
10347 em->len = ins.offset;
10348 em->block_start = ins.objectid;
10349 em->block_len = ins.offset;
10350 em->orig_block_len = ins.offset;
10351 em->ram_bytes = ins.offset;
10352 em->bdev = fs_info->fs_devices->latest_bdev;
10353 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10354 em->generation = trans->transid;
10357 write_lock(&em_tree->lock);
10358 ret = add_extent_mapping(em_tree, em, 1);
10359 write_unlock(&em_tree->lock);
10360 if (ret != -EEXIST)
10362 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10363 cur_offset + ins.offset - 1,
10366 free_extent_map(em);
10368 num_bytes -= ins.offset;
10369 cur_offset += ins.offset;
10370 *alloc_hint = ins.objectid + ins.offset;
10372 inode_inc_iversion(inode);
10373 inode->i_ctime = current_time(inode);
10374 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10375 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10376 (actual_len > inode->i_size) &&
10377 (cur_offset > inode->i_size)) {
10378 if (cur_offset > actual_len)
10379 i_size = actual_len;
10381 i_size = cur_offset;
10382 i_size_write(inode, i_size);
10383 btrfs_ordered_update_i_size(inode, i_size, NULL);
10386 ret = btrfs_update_inode(trans, root, inode);
10389 btrfs_abort_transaction(trans, ret);
10391 btrfs_end_transaction(trans);
10396 btrfs_end_transaction(trans);
10398 if (cur_offset < end)
10399 btrfs_free_reserved_data_space(inode, cur_offset,
10400 end - cur_offset + 1);
10404 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10405 u64 start, u64 num_bytes, u64 min_size,
10406 loff_t actual_len, u64 *alloc_hint)
10408 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10409 min_size, actual_len, alloc_hint,
10413 int btrfs_prealloc_file_range_trans(struct inode *inode,
10414 struct btrfs_trans_handle *trans, int mode,
10415 u64 start, u64 num_bytes, u64 min_size,
10416 loff_t actual_len, u64 *alloc_hint)
10418 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10419 min_size, actual_len, alloc_hint, trans);
10422 static int btrfs_set_page_dirty(struct page *page)
10424 return __set_page_dirty_nobuffers(page);
10427 static int btrfs_permission(struct inode *inode, int mask)
10429 struct btrfs_root *root = BTRFS_I(inode)->root;
10430 umode_t mode = inode->i_mode;
10432 if (mask & MAY_WRITE &&
10433 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10434 if (btrfs_root_readonly(root))
10436 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10439 return generic_permission(inode, mask);
10442 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10444 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10445 struct btrfs_trans_handle *trans;
10446 struct btrfs_root *root = BTRFS_I(dir)->root;
10447 struct inode *inode = NULL;
10453 * 5 units required for adding orphan entry
10455 trans = btrfs_start_transaction(root, 5);
10457 return PTR_ERR(trans);
10459 ret = btrfs_find_free_ino(root, &objectid);
10463 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10464 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10465 if (IS_ERR(inode)) {
10466 ret = PTR_ERR(inode);
10471 inode->i_fop = &btrfs_file_operations;
10472 inode->i_op = &btrfs_file_inode_operations;
10474 inode->i_mapping->a_ops = &btrfs_aops;
10475 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10477 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10481 ret = btrfs_update_inode(trans, root, inode);
10484 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10489 * We set number of links to 0 in btrfs_new_inode(), and here we set
10490 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10493 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10495 set_nlink(inode, 1);
10496 unlock_new_inode(inode);
10497 d_tmpfile(dentry, inode);
10498 mark_inode_dirty(inode);
10501 btrfs_end_transaction(trans);
10504 btrfs_balance_delayed_items(fs_info);
10505 btrfs_btree_balance_dirty(fs_info);
10509 unlock_new_inode(inode);
10514 static const struct inode_operations btrfs_dir_inode_operations = {
10515 .getattr = btrfs_getattr,
10516 .lookup = btrfs_lookup,
10517 .create = btrfs_create,
10518 .unlink = btrfs_unlink,
10519 .link = btrfs_link,
10520 .mkdir = btrfs_mkdir,
10521 .rmdir = btrfs_rmdir,
10522 .rename = btrfs_rename2,
10523 .symlink = btrfs_symlink,
10524 .setattr = btrfs_setattr,
10525 .mknod = btrfs_mknod,
10526 .listxattr = btrfs_listxattr,
10527 .permission = btrfs_permission,
10528 .get_acl = btrfs_get_acl,
10529 .set_acl = btrfs_set_acl,
10530 .update_time = btrfs_update_time,
10531 .tmpfile = btrfs_tmpfile,
10533 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10534 .lookup = btrfs_lookup,
10535 .permission = btrfs_permission,
10536 .update_time = btrfs_update_time,
10539 static const struct file_operations btrfs_dir_file_operations = {
10540 .llseek = generic_file_llseek,
10541 .read = generic_read_dir,
10542 .iterate_shared = btrfs_real_readdir,
10543 .unlocked_ioctl = btrfs_ioctl,
10544 #ifdef CONFIG_COMPAT
10545 .compat_ioctl = btrfs_compat_ioctl,
10547 .release = btrfs_release_file,
10548 .fsync = btrfs_sync_file,
10551 static const struct extent_io_ops btrfs_extent_io_ops = {
10552 .fill_delalloc = run_delalloc_range,
10553 .submit_bio_hook = btrfs_submit_bio_hook,
10554 .merge_bio_hook = btrfs_merge_bio_hook,
10555 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10556 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10557 .writepage_start_hook = btrfs_writepage_start_hook,
10558 .set_bit_hook = btrfs_set_bit_hook,
10559 .clear_bit_hook = btrfs_clear_bit_hook,
10560 .merge_extent_hook = btrfs_merge_extent_hook,
10561 .split_extent_hook = btrfs_split_extent_hook,
10565 * btrfs doesn't support the bmap operation because swapfiles
10566 * use bmap to make a mapping of extents in the file. They assume
10567 * these extents won't change over the life of the file and they
10568 * use the bmap result to do IO directly to the drive.
10570 * the btrfs bmap call would return logical addresses that aren't
10571 * suitable for IO and they also will change frequently as COW
10572 * operations happen. So, swapfile + btrfs == corruption.
10574 * For now we're avoiding this by dropping bmap.
10576 static const struct address_space_operations btrfs_aops = {
10577 .readpage = btrfs_readpage,
10578 .writepage = btrfs_writepage,
10579 .writepages = btrfs_writepages,
10580 .readpages = btrfs_readpages,
10581 .direct_IO = btrfs_direct_IO,
10582 .invalidatepage = btrfs_invalidatepage,
10583 .releasepage = btrfs_releasepage,
10584 .set_page_dirty = btrfs_set_page_dirty,
10585 .error_remove_page = generic_error_remove_page,
10588 static const struct address_space_operations btrfs_symlink_aops = {
10589 .readpage = btrfs_readpage,
10590 .writepage = btrfs_writepage,
10591 .invalidatepage = btrfs_invalidatepage,
10592 .releasepage = btrfs_releasepage,
10595 static const struct inode_operations btrfs_file_inode_operations = {
10596 .getattr = btrfs_getattr,
10597 .setattr = btrfs_setattr,
10598 .listxattr = btrfs_listxattr,
10599 .permission = btrfs_permission,
10600 .fiemap = btrfs_fiemap,
10601 .get_acl = btrfs_get_acl,
10602 .set_acl = btrfs_set_acl,
10603 .update_time = btrfs_update_time,
10605 static const struct inode_operations btrfs_special_inode_operations = {
10606 .getattr = btrfs_getattr,
10607 .setattr = btrfs_setattr,
10608 .permission = btrfs_permission,
10609 .listxattr = btrfs_listxattr,
10610 .get_acl = btrfs_get_acl,
10611 .set_acl = btrfs_set_acl,
10612 .update_time = btrfs_update_time,
10614 static const struct inode_operations btrfs_symlink_inode_operations = {
10615 .get_link = page_get_link,
10616 .getattr = btrfs_getattr,
10617 .setattr = btrfs_setattr,
10618 .permission = btrfs_permission,
10619 .listxattr = btrfs_listxattr,
10620 .update_time = btrfs_update_time,
10623 const struct dentry_operations btrfs_dentry_operations = {
10624 .d_delete = btrfs_dentry_delete,
10625 .d_release = btrfs_dentry_release,