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 void __endio_write_update_ordered(struct inode *inode,
119 const u64 offset, const u64 bytes,
120 const bool uptodate);
123 * Cleanup all submitted ordered extents in specified range to handle errors
124 * from the fill_dellaloc() callback.
126 * NOTE: caller must ensure that when an error happens, it can not call
127 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
128 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
129 * to be released, which we want to happen only when finishing the ordered
130 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
131 * fill_delalloc() callback already does proper cleanup for the first page of
132 * the range, that is, it invokes the callback writepage_end_io_hook() for the
133 * range of the first page.
135 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
139 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
140 bytes - PAGE_SIZE, false);
143 static int btrfs_dirty_inode(struct inode *inode);
145 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
146 void btrfs_test_inode_set_ops(struct inode *inode)
148 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
152 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
153 struct inode *inode, struct inode *dir,
154 const struct qstr *qstr)
158 err = btrfs_init_acl(trans, inode, dir);
160 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
165 * this does all the hard work for inserting an inline extent into
166 * the btree. The caller should have done a btrfs_drop_extents so that
167 * no overlapping inline items exist in the btree
169 static int insert_inline_extent(struct btrfs_trans_handle *trans,
170 struct btrfs_path *path, int extent_inserted,
171 struct btrfs_root *root, struct inode *inode,
172 u64 start, size_t size, size_t compressed_size,
174 struct page **compressed_pages)
176 struct extent_buffer *leaf;
177 struct page *page = NULL;
180 struct btrfs_file_extent_item *ei;
183 size_t cur_size = size;
184 unsigned long offset;
186 if (compressed_size && compressed_pages)
187 cur_size = compressed_size;
189 inode_add_bytes(inode, size);
191 if (!extent_inserted) {
192 struct btrfs_key key;
195 key.objectid = btrfs_ino(BTRFS_I(inode));
197 key.type = BTRFS_EXTENT_DATA_KEY;
199 datasize = btrfs_file_extent_calc_inline_size(cur_size);
200 path->leave_spinning = 1;
201 ret = btrfs_insert_empty_item(trans, root, path, &key,
208 leaf = path->nodes[0];
209 ei = btrfs_item_ptr(leaf, path->slots[0],
210 struct btrfs_file_extent_item);
211 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
212 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
213 btrfs_set_file_extent_encryption(leaf, ei, 0);
214 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
215 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
216 ptr = btrfs_file_extent_inline_start(ei);
218 if (compress_type != BTRFS_COMPRESS_NONE) {
221 while (compressed_size > 0) {
222 cpage = compressed_pages[i];
223 cur_size = min_t(unsigned long, compressed_size,
226 kaddr = kmap_atomic(cpage);
227 write_extent_buffer(leaf, kaddr, ptr, cur_size);
228 kunmap_atomic(kaddr);
232 compressed_size -= cur_size;
234 btrfs_set_file_extent_compression(leaf, ei,
237 page = find_get_page(inode->i_mapping,
238 start >> PAGE_SHIFT);
239 btrfs_set_file_extent_compression(leaf, ei, 0);
240 kaddr = kmap_atomic(page);
241 offset = start & (PAGE_SIZE - 1);
242 write_extent_buffer(leaf, kaddr + offset, ptr, size);
243 kunmap_atomic(kaddr);
246 btrfs_mark_buffer_dirty(leaf);
247 btrfs_release_path(path);
250 * we're an inline extent, so nobody can
251 * extend the file past i_size without locking
252 * a page we already have locked.
254 * We must do any isize and inode updates
255 * before we unlock the pages. Otherwise we
256 * could end up racing with unlink.
258 BTRFS_I(inode)->disk_i_size = inode->i_size;
259 ret = btrfs_update_inode(trans, root, inode);
268 * conditionally insert an inline extent into the file. This
269 * does the checks required to make sure the data is small enough
270 * to fit as an inline extent.
272 static noinline int cow_file_range_inline(struct btrfs_root *root,
273 struct inode *inode, u64 start,
274 u64 end, size_t compressed_size,
276 struct page **compressed_pages)
278 struct btrfs_fs_info *fs_info = root->fs_info;
279 struct btrfs_trans_handle *trans;
280 u64 isize = i_size_read(inode);
281 u64 actual_end = min(end + 1, isize);
282 u64 inline_len = actual_end - start;
283 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
284 u64 data_len = inline_len;
286 struct btrfs_path *path;
287 int extent_inserted = 0;
288 u32 extent_item_size;
291 data_len = compressed_size;
294 actual_end > fs_info->sectorsize ||
295 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
297 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
299 data_len > fs_info->max_inline) {
303 path = btrfs_alloc_path();
307 trans = btrfs_join_transaction(root);
309 btrfs_free_path(path);
310 return PTR_ERR(trans);
312 trans->block_rsv = &fs_info->delalloc_block_rsv;
314 if (compressed_size && compressed_pages)
315 extent_item_size = btrfs_file_extent_calc_inline_size(
318 extent_item_size = btrfs_file_extent_calc_inline_size(
321 ret = __btrfs_drop_extents(trans, root, inode, path,
322 start, aligned_end, NULL,
323 1, 1, extent_item_size, &extent_inserted);
325 btrfs_abort_transaction(trans, ret);
329 if (isize > actual_end)
330 inline_len = min_t(u64, isize, actual_end);
331 ret = insert_inline_extent(trans, path, extent_inserted,
333 inline_len, compressed_size,
334 compress_type, compressed_pages);
335 if (ret && ret != -ENOSPC) {
336 btrfs_abort_transaction(trans, ret);
338 } else if (ret == -ENOSPC) {
343 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
344 btrfs_delalloc_release_metadata(BTRFS_I(inode), end + 1 - start);
345 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
348 * Don't forget to free the reserved space, as for inlined extent
349 * it won't count as data extent, free them directly here.
350 * And at reserve time, it's always aligned to page size, so
351 * just free one page here.
353 btrfs_qgroup_free_data(inode, 0, PAGE_SIZE);
354 btrfs_free_path(path);
355 btrfs_end_transaction(trans);
359 struct async_extent {
364 unsigned long nr_pages;
366 struct list_head list;
371 struct btrfs_root *root;
372 struct page *locked_page;
375 struct list_head extents;
376 struct btrfs_work work;
379 static noinline int add_async_extent(struct async_cow *cow,
380 u64 start, u64 ram_size,
383 unsigned long nr_pages,
386 struct async_extent *async_extent;
388 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
389 BUG_ON(!async_extent); /* -ENOMEM */
390 async_extent->start = start;
391 async_extent->ram_size = ram_size;
392 async_extent->compressed_size = compressed_size;
393 async_extent->pages = pages;
394 async_extent->nr_pages = nr_pages;
395 async_extent->compress_type = compress_type;
396 list_add_tail(&async_extent->list, &cow->extents);
400 static inline int inode_need_compress(struct inode *inode)
402 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
405 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
407 /* bad compression ratios */
408 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
410 if (btrfs_test_opt(fs_info, COMPRESS) ||
411 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
412 BTRFS_I(inode)->force_compress)
417 static inline void inode_should_defrag(struct btrfs_inode *inode,
418 u64 start, u64 end, u64 num_bytes, u64 small_write)
420 /* If this is a small write inside eof, kick off a defrag */
421 if (num_bytes < small_write &&
422 (start > 0 || end + 1 < inode->disk_i_size))
423 btrfs_add_inode_defrag(NULL, inode);
427 * we create compressed extents in two phases. The first
428 * phase compresses a range of pages that have already been
429 * locked (both pages and state bits are locked).
431 * This is done inside an ordered work queue, and the compression
432 * is spread across many cpus. The actual IO submission is step
433 * two, and the ordered work queue takes care of making sure that
434 * happens in the same order things were put onto the queue by
435 * writepages and friends.
437 * If this code finds it can't get good compression, it puts an
438 * entry onto the work queue to write the uncompressed bytes. This
439 * makes sure that both compressed inodes and uncompressed inodes
440 * are written in the same order that the flusher thread sent them
443 static noinline void compress_file_range(struct inode *inode,
444 struct page *locked_page,
446 struct async_cow *async_cow,
449 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
450 struct btrfs_root *root = BTRFS_I(inode)->root;
452 u64 blocksize = fs_info->sectorsize;
454 u64 isize = i_size_read(inode);
456 struct page **pages = NULL;
457 unsigned long nr_pages;
458 unsigned long total_compressed = 0;
459 unsigned long total_in = 0;
462 int compress_type = fs_info->compress_type;
465 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
468 actual_end = min_t(u64, isize, end + 1);
471 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
472 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
473 nr_pages = min_t(unsigned long, nr_pages,
474 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
477 * we don't want to send crud past the end of i_size through
478 * compression, that's just a waste of CPU time. So, if the
479 * end of the file is before the start of our current
480 * requested range of bytes, we bail out to the uncompressed
481 * cleanup code that can deal with all of this.
483 * It isn't really the fastest way to fix things, but this is a
484 * very uncommon corner.
486 if (actual_end <= start)
487 goto cleanup_and_bail_uncompressed;
489 total_compressed = actual_end - start;
492 * skip compression for a small file range(<=blocksize) that
493 * isn't an inline extent, since it doesn't save disk space at all.
495 if (total_compressed <= blocksize &&
496 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
497 goto cleanup_and_bail_uncompressed;
499 total_compressed = min_t(unsigned long, total_compressed,
500 BTRFS_MAX_UNCOMPRESSED);
501 num_bytes = ALIGN(end - start + 1, blocksize);
502 num_bytes = max(blocksize, num_bytes);
507 * we do compression for mount -o compress and when the
508 * inode has not been flagged as nocompress. This flag can
509 * change at any time if we discover bad compression ratios.
511 if (inode_need_compress(inode)) {
513 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
515 /* just bail out to the uncompressed code */
519 if (BTRFS_I(inode)->force_compress)
520 compress_type = BTRFS_I(inode)->force_compress;
523 * we need to call clear_page_dirty_for_io on each
524 * page in the range. Otherwise applications with the file
525 * mmap'd can wander in and change the page contents while
526 * we are compressing them.
528 * If the compression fails for any reason, we set the pages
529 * dirty again later on.
531 extent_range_clear_dirty_for_io(inode, start, end);
533 ret = btrfs_compress_pages(compress_type,
534 inode->i_mapping, start,
541 unsigned long offset = total_compressed &
543 struct page *page = pages[nr_pages - 1];
546 /* zero the tail end of the last page, we might be
547 * sending it down to disk
550 kaddr = kmap_atomic(page);
551 memset(kaddr + offset, 0,
553 kunmap_atomic(kaddr);
560 /* lets try to make an inline extent */
561 if (ret || total_in < (actual_end - start)) {
562 /* we didn't compress the entire range, try
563 * to make an uncompressed inline extent.
565 ret = cow_file_range_inline(root, inode, start, end,
566 0, BTRFS_COMPRESS_NONE, NULL);
568 /* try making a compressed inline extent */
569 ret = cow_file_range_inline(root, inode, start, end,
571 compress_type, pages);
574 unsigned long clear_flags = EXTENT_DELALLOC |
575 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG;
576 unsigned long page_error_op;
578 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
579 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
582 * inline extent creation worked or returned error,
583 * we don't need to create any more async work items.
584 * Unlock and free up our temp pages.
586 extent_clear_unlock_delalloc(inode, start, end, end,
594 btrfs_free_reserved_data_space_noquota(inode,
603 * we aren't doing an inline extent round the compressed size
604 * up to a block size boundary so the allocator does sane
607 total_compressed = ALIGN(total_compressed, blocksize);
610 * one last check to make sure the compression is really a
611 * win, compare the page count read with the blocks on disk
613 total_in = ALIGN(total_in, PAGE_SIZE);
614 if (total_compressed >= total_in) {
617 num_bytes = total_in;
621 * The async work queues will take care of doing actual
622 * allocation on disk for these compressed pages, and
623 * will submit them to the elevator.
625 add_async_extent(async_cow, start, num_bytes,
626 total_compressed, pages, nr_pages,
629 if (start + num_bytes < end) {
640 * the compression code ran but failed to make things smaller,
641 * free any pages it allocated and our page pointer array
643 for (i = 0; i < nr_pages; i++) {
644 WARN_ON(pages[i]->mapping);
649 total_compressed = 0;
652 /* flag the file so we don't compress in the future */
653 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
654 !(BTRFS_I(inode)->force_compress)) {
655 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
658 cleanup_and_bail_uncompressed:
660 * No compression, but we still need to write the pages in the file
661 * we've been given so far. redirty the locked page if it corresponds
662 * to our extent and set things up for the async work queue to run
663 * cow_file_range to do the normal delalloc dance.
665 if (page_offset(locked_page) >= start &&
666 page_offset(locked_page) <= end)
667 __set_page_dirty_nobuffers(locked_page);
668 /* unlocked later on in the async handlers */
671 extent_range_redirty_for_io(inode, start, end);
672 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
673 BTRFS_COMPRESS_NONE);
679 for (i = 0; i < nr_pages; i++) {
680 WARN_ON(pages[i]->mapping);
686 static void free_async_extent_pages(struct async_extent *async_extent)
690 if (!async_extent->pages)
693 for (i = 0; i < async_extent->nr_pages; i++) {
694 WARN_ON(async_extent->pages[i]->mapping);
695 put_page(async_extent->pages[i]);
697 kfree(async_extent->pages);
698 async_extent->nr_pages = 0;
699 async_extent->pages = NULL;
703 * phase two of compressed writeback. This is the ordered portion
704 * of the code, which only gets called in the order the work was
705 * queued. We walk all the async extents created by compress_file_range
706 * and send them down to the disk.
708 static noinline void submit_compressed_extents(struct inode *inode,
709 struct async_cow *async_cow)
711 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
712 struct async_extent *async_extent;
714 struct btrfs_key ins;
715 struct extent_map *em;
716 struct btrfs_root *root = BTRFS_I(inode)->root;
717 struct extent_io_tree *io_tree;
721 while (!list_empty(&async_cow->extents)) {
722 async_extent = list_entry(async_cow->extents.next,
723 struct async_extent, list);
724 list_del(&async_extent->list);
726 io_tree = &BTRFS_I(inode)->io_tree;
729 /* did the compression code fall back to uncompressed IO? */
730 if (!async_extent->pages) {
731 int page_started = 0;
732 unsigned long nr_written = 0;
734 lock_extent(io_tree, async_extent->start,
735 async_extent->start +
736 async_extent->ram_size - 1);
738 /* allocate blocks */
739 ret = cow_file_range(inode, async_cow->locked_page,
741 async_extent->start +
742 async_extent->ram_size - 1,
743 async_extent->start +
744 async_extent->ram_size - 1,
745 &page_started, &nr_written, 0,
751 * if page_started, cow_file_range inserted an
752 * inline extent and took care of all the unlocking
753 * and IO for us. Otherwise, we need to submit
754 * all those pages down to the drive.
756 if (!page_started && !ret)
757 extent_write_locked_range(io_tree,
758 inode, async_extent->start,
759 async_extent->start +
760 async_extent->ram_size - 1,
764 unlock_page(async_cow->locked_page);
770 lock_extent(io_tree, async_extent->start,
771 async_extent->start + async_extent->ram_size - 1);
773 ret = btrfs_reserve_extent(root, async_extent->ram_size,
774 async_extent->compressed_size,
775 async_extent->compressed_size,
776 0, alloc_hint, &ins, 1, 1);
778 free_async_extent_pages(async_extent);
780 if (ret == -ENOSPC) {
781 unlock_extent(io_tree, async_extent->start,
782 async_extent->start +
783 async_extent->ram_size - 1);
786 * we need to redirty the pages if we decide to
787 * fallback to uncompressed IO, otherwise we
788 * will not submit these pages down to lower
791 extent_range_redirty_for_io(inode,
793 async_extent->start +
794 async_extent->ram_size - 1);
801 * here we're doing allocation and writeback of the
804 em = create_io_em(inode, async_extent->start,
805 async_extent->ram_size, /* len */
806 async_extent->start, /* orig_start */
807 ins.objectid, /* block_start */
808 ins.offset, /* block_len */
809 ins.offset, /* orig_block_len */
810 async_extent->ram_size, /* ram_bytes */
811 async_extent->compress_type,
812 BTRFS_ORDERED_COMPRESSED);
814 /* ret value is not necessary due to void function */
815 goto out_free_reserve;
818 ret = btrfs_add_ordered_extent_compress(inode,
821 async_extent->ram_size,
823 BTRFS_ORDERED_COMPRESSED,
824 async_extent->compress_type);
826 btrfs_drop_extent_cache(BTRFS_I(inode),
828 async_extent->start +
829 async_extent->ram_size - 1, 0);
830 goto out_free_reserve;
832 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
835 * clear dirty, set writeback and unlock the pages.
837 extent_clear_unlock_delalloc(inode, async_extent->start,
838 async_extent->start +
839 async_extent->ram_size - 1,
840 async_extent->start +
841 async_extent->ram_size - 1,
842 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
843 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
845 ret = btrfs_submit_compressed_write(inode,
847 async_extent->ram_size,
849 ins.offset, async_extent->pages,
850 async_extent->nr_pages);
852 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
853 struct page *p = async_extent->pages[0];
854 const u64 start = async_extent->start;
855 const u64 end = start + async_extent->ram_size - 1;
857 p->mapping = inode->i_mapping;
858 tree->ops->writepage_end_io_hook(p, start, end,
861 extent_clear_unlock_delalloc(inode, start, end, end,
865 free_async_extent_pages(async_extent);
867 alloc_hint = ins.objectid + ins.offset;
873 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
874 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
876 extent_clear_unlock_delalloc(inode, async_extent->start,
877 async_extent->start +
878 async_extent->ram_size - 1,
879 async_extent->start +
880 async_extent->ram_size - 1,
881 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
882 EXTENT_DELALLOC_NEW |
883 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
884 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
885 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
887 free_async_extent_pages(async_extent);
892 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
895 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
896 struct extent_map *em;
899 read_lock(&em_tree->lock);
900 em = search_extent_mapping(em_tree, start, num_bytes);
903 * if block start isn't an actual block number then find the
904 * first block in this inode and use that as a hint. If that
905 * block is also bogus then just don't worry about it.
907 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
909 em = search_extent_mapping(em_tree, 0, 0);
910 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
911 alloc_hint = em->block_start;
915 alloc_hint = em->block_start;
919 read_unlock(&em_tree->lock);
925 * when extent_io.c finds a delayed allocation range in the file,
926 * the call backs end up in this code. The basic idea is to
927 * allocate extents on disk for the range, and create ordered data structs
928 * in ram to track those extents.
930 * locked_page is the page that writepage had locked already. We use
931 * it to make sure we don't do extra locks or unlocks.
933 * *page_started is set to one if we unlock locked_page and do everything
934 * required to start IO on it. It may be clean and already done with
937 static noinline int cow_file_range(struct inode *inode,
938 struct page *locked_page,
939 u64 start, u64 end, u64 delalloc_end,
940 int *page_started, unsigned long *nr_written,
941 int unlock, struct btrfs_dedupe_hash *hash)
943 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
944 struct btrfs_root *root = BTRFS_I(inode)->root;
947 unsigned long ram_size;
949 u64 cur_alloc_size = 0;
950 u64 blocksize = fs_info->sectorsize;
951 struct btrfs_key ins;
952 struct extent_map *em;
954 unsigned long page_ops;
955 bool extent_reserved = false;
958 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
964 num_bytes = ALIGN(end - start + 1, blocksize);
965 num_bytes = max(blocksize, num_bytes);
966 disk_num_bytes = num_bytes;
968 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
971 /* lets try to make an inline extent */
972 ret = cow_file_range_inline(root, inode, start, end, 0,
973 BTRFS_COMPRESS_NONE, NULL);
975 extent_clear_unlock_delalloc(inode, start, end,
977 EXTENT_LOCKED | EXTENT_DELALLOC |
978 EXTENT_DELALLOC_NEW |
979 EXTENT_DEFRAG, PAGE_UNLOCK |
980 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
982 btrfs_free_reserved_data_space_noquota(inode, start,
984 *nr_written = *nr_written +
985 (end - start + PAGE_SIZE) / PAGE_SIZE;
988 } else if (ret < 0) {
993 BUG_ON(disk_num_bytes >
994 btrfs_super_total_bytes(fs_info->super_copy));
996 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
997 btrfs_drop_extent_cache(BTRFS_I(inode), start,
998 start + num_bytes - 1, 0);
1000 while (disk_num_bytes > 0) {
1001 cur_alloc_size = disk_num_bytes;
1002 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1003 fs_info->sectorsize, 0, alloc_hint,
1007 cur_alloc_size = ins.offset;
1008 extent_reserved = true;
1010 ram_size = ins.offset;
1011 em = create_io_em(inode, start, ins.offset, /* len */
1012 start, /* orig_start */
1013 ins.objectid, /* block_start */
1014 ins.offset, /* block_len */
1015 ins.offset, /* orig_block_len */
1016 ram_size, /* ram_bytes */
1017 BTRFS_COMPRESS_NONE, /* compress_type */
1018 BTRFS_ORDERED_REGULAR /* type */);
1021 free_extent_map(em);
1023 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1024 ram_size, cur_alloc_size, 0);
1026 goto out_drop_extent_cache;
1028 if (root->root_key.objectid ==
1029 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1030 ret = btrfs_reloc_clone_csums(inode, start,
1033 * Only drop cache here, and process as normal.
1035 * We must not allow extent_clear_unlock_delalloc()
1036 * at out_unlock label to free meta of this ordered
1037 * extent, as its meta should be freed by
1038 * btrfs_finish_ordered_io().
1040 * So we must continue until @start is increased to
1041 * skip current ordered extent.
1044 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1045 start + ram_size - 1, 0);
1048 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1050 /* we're not doing compressed IO, don't unlock the first
1051 * page (which the caller expects to stay locked), don't
1052 * clear any dirty bits and don't set any writeback bits
1054 * Do set the Private2 bit so we know this page was properly
1055 * setup for writepage
1057 page_ops = unlock ? PAGE_UNLOCK : 0;
1058 page_ops |= PAGE_SET_PRIVATE2;
1060 extent_clear_unlock_delalloc(inode, start,
1061 start + ram_size - 1,
1062 delalloc_end, locked_page,
1063 EXTENT_LOCKED | EXTENT_DELALLOC,
1065 if (disk_num_bytes < cur_alloc_size)
1068 disk_num_bytes -= cur_alloc_size;
1069 num_bytes -= cur_alloc_size;
1070 alloc_hint = ins.objectid + ins.offset;
1071 start += cur_alloc_size;
1072 extent_reserved = false;
1075 * btrfs_reloc_clone_csums() error, since start is increased
1076 * extent_clear_unlock_delalloc() at out_unlock label won't
1077 * free metadata of current ordered extent, we're OK to exit.
1085 out_drop_extent_cache:
1086 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1088 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1089 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1091 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1092 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1093 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1096 * If we reserved an extent for our delalloc range (or a subrange) and
1097 * failed to create the respective ordered extent, then it means that
1098 * when we reserved the extent we decremented the extent's size from
1099 * the data space_info's bytes_may_use counter and incremented the
1100 * space_info's bytes_reserved counter by the same amount. We must make
1101 * sure extent_clear_unlock_delalloc() does not try to decrement again
1102 * the data space_info's bytes_may_use counter, therefore we do not pass
1103 * it the flag EXTENT_CLEAR_DATA_RESV.
1105 if (extent_reserved) {
1106 extent_clear_unlock_delalloc(inode, start,
1107 start + cur_alloc_size,
1108 start + cur_alloc_size,
1112 start += cur_alloc_size;
1116 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1118 clear_bits | EXTENT_CLEAR_DATA_RESV,
1124 * work queue call back to started compression on a file and pages
1126 static noinline void async_cow_start(struct btrfs_work *work)
1128 struct async_cow *async_cow;
1130 async_cow = container_of(work, struct async_cow, work);
1132 compress_file_range(async_cow->inode, async_cow->locked_page,
1133 async_cow->start, async_cow->end, async_cow,
1135 if (num_added == 0) {
1136 btrfs_add_delayed_iput(async_cow->inode);
1137 async_cow->inode = NULL;
1142 * work queue call back to submit previously compressed pages
1144 static noinline void async_cow_submit(struct btrfs_work *work)
1146 struct btrfs_fs_info *fs_info;
1147 struct async_cow *async_cow;
1148 struct btrfs_root *root;
1149 unsigned long nr_pages;
1151 async_cow = container_of(work, struct async_cow, work);
1153 root = async_cow->root;
1154 fs_info = root->fs_info;
1155 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1159 * atomic_sub_return implies a barrier for waitqueue_active
1161 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1163 waitqueue_active(&fs_info->async_submit_wait))
1164 wake_up(&fs_info->async_submit_wait);
1166 if (async_cow->inode)
1167 submit_compressed_extents(async_cow->inode, async_cow);
1170 static noinline void async_cow_free(struct btrfs_work *work)
1172 struct async_cow *async_cow;
1173 async_cow = container_of(work, struct async_cow, work);
1174 if (async_cow->inode)
1175 btrfs_add_delayed_iput(async_cow->inode);
1179 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1180 u64 start, u64 end, int *page_started,
1181 unsigned long *nr_written)
1183 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1184 struct async_cow *async_cow;
1185 struct btrfs_root *root = BTRFS_I(inode)->root;
1186 unsigned long nr_pages;
1189 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1190 1, 0, NULL, GFP_NOFS);
1191 while (start < end) {
1192 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1193 BUG_ON(!async_cow); /* -ENOMEM */
1194 async_cow->inode = igrab(inode);
1195 async_cow->root = root;
1196 async_cow->locked_page = locked_page;
1197 async_cow->start = start;
1199 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1200 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1203 cur_end = min(end, start + SZ_512K - 1);
1205 async_cow->end = cur_end;
1206 INIT_LIST_HEAD(&async_cow->extents);
1208 btrfs_init_work(&async_cow->work,
1209 btrfs_delalloc_helper,
1210 async_cow_start, async_cow_submit,
1213 nr_pages = (cur_end - start + PAGE_SIZE) >>
1215 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1217 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1219 while (atomic_read(&fs_info->async_submit_draining) &&
1220 atomic_read(&fs_info->async_delalloc_pages)) {
1221 wait_event(fs_info->async_submit_wait,
1222 (atomic_read(&fs_info->async_delalloc_pages) ==
1226 *nr_written += nr_pages;
1227 start = cur_end + 1;
1233 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1234 u64 bytenr, u64 num_bytes)
1237 struct btrfs_ordered_sum *sums;
1240 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1241 bytenr + num_bytes - 1, &list, 0);
1242 if (ret == 0 && list_empty(&list))
1245 while (!list_empty(&list)) {
1246 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1247 list_del(&sums->list);
1254 * when nowcow writeback call back. This checks for snapshots or COW copies
1255 * of the extents that exist in the file, and COWs the file as required.
1257 * If no cow copies or snapshots exist, we write directly to the existing
1260 static noinline int run_delalloc_nocow(struct inode *inode,
1261 struct page *locked_page,
1262 u64 start, u64 end, int *page_started, int force,
1263 unsigned long *nr_written)
1265 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1266 struct btrfs_root *root = BTRFS_I(inode)->root;
1267 struct extent_buffer *leaf;
1268 struct btrfs_path *path;
1269 struct btrfs_file_extent_item *fi;
1270 struct btrfs_key found_key;
1271 struct extent_map *em;
1286 u64 ino = btrfs_ino(BTRFS_I(inode));
1288 path = btrfs_alloc_path();
1290 extent_clear_unlock_delalloc(inode, start, end, end,
1292 EXTENT_LOCKED | EXTENT_DELALLOC |
1293 EXTENT_DO_ACCOUNTING |
1294 EXTENT_DEFRAG, PAGE_UNLOCK |
1296 PAGE_SET_WRITEBACK |
1297 PAGE_END_WRITEBACK);
1301 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1303 cow_start = (u64)-1;
1306 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1310 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1311 leaf = path->nodes[0];
1312 btrfs_item_key_to_cpu(leaf, &found_key,
1313 path->slots[0] - 1);
1314 if (found_key.objectid == ino &&
1315 found_key.type == BTRFS_EXTENT_DATA_KEY)
1320 leaf = path->nodes[0];
1321 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1322 ret = btrfs_next_leaf(root, path);
1327 leaf = path->nodes[0];
1333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1335 if (found_key.objectid > ino)
1337 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1338 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1342 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1343 found_key.offset > end)
1346 if (found_key.offset > cur_offset) {
1347 extent_end = found_key.offset;
1352 fi = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_file_extent_item);
1354 extent_type = btrfs_file_extent_type(leaf, fi);
1356 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1357 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1358 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1359 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1360 extent_offset = btrfs_file_extent_offset(leaf, fi);
1361 extent_end = found_key.offset +
1362 btrfs_file_extent_num_bytes(leaf, fi);
1364 btrfs_file_extent_disk_num_bytes(leaf, fi);
1365 if (extent_end <= start) {
1369 if (disk_bytenr == 0)
1371 if (btrfs_file_extent_compression(leaf, fi) ||
1372 btrfs_file_extent_encryption(leaf, fi) ||
1373 btrfs_file_extent_other_encoding(leaf, fi))
1375 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1377 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1379 if (btrfs_cross_ref_exist(root, ino,
1381 extent_offset, disk_bytenr))
1383 disk_bytenr += extent_offset;
1384 disk_bytenr += cur_offset - found_key.offset;
1385 num_bytes = min(end + 1, extent_end) - cur_offset;
1387 * if there are pending snapshots for this root,
1388 * we fall into common COW way.
1391 err = btrfs_start_write_no_snapshoting(root);
1396 * force cow if csum exists in the range.
1397 * this ensure that csum for a given extent are
1398 * either valid or do not exist.
1400 if (csum_exist_in_range(fs_info, disk_bytenr,
1403 btrfs_end_write_no_snapshoting(root);
1406 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1408 btrfs_end_write_no_snapshoting(root);
1412 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1413 extent_end = found_key.offset +
1414 btrfs_file_extent_inline_len(leaf,
1415 path->slots[0], fi);
1416 extent_end = ALIGN(extent_end,
1417 fs_info->sectorsize);
1422 if (extent_end <= start) {
1424 if (!nolock && nocow)
1425 btrfs_end_write_no_snapshoting(root);
1427 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1431 if (cow_start == (u64)-1)
1432 cow_start = cur_offset;
1433 cur_offset = extent_end;
1434 if (cur_offset > end)
1440 btrfs_release_path(path);
1441 if (cow_start != (u64)-1) {
1442 ret = cow_file_range(inode, locked_page,
1443 cow_start, found_key.offset - 1,
1444 end, page_started, nr_written, 1,
1447 if (!nolock && nocow)
1448 btrfs_end_write_no_snapshoting(root);
1450 btrfs_dec_nocow_writers(fs_info,
1454 cow_start = (u64)-1;
1457 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1458 u64 orig_start = found_key.offset - extent_offset;
1460 em = create_io_em(inode, cur_offset, num_bytes,
1462 disk_bytenr, /* block_start */
1463 num_bytes, /* block_len */
1464 disk_num_bytes, /* orig_block_len */
1465 ram_bytes, BTRFS_COMPRESS_NONE,
1466 BTRFS_ORDERED_PREALLOC);
1468 if (!nolock && nocow)
1469 btrfs_end_write_no_snapshoting(root);
1471 btrfs_dec_nocow_writers(fs_info,
1476 free_extent_map(em);
1479 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1480 type = BTRFS_ORDERED_PREALLOC;
1482 type = BTRFS_ORDERED_NOCOW;
1485 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1486 num_bytes, num_bytes, type);
1488 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1489 BUG_ON(ret); /* -ENOMEM */
1491 if (root->root_key.objectid ==
1492 BTRFS_DATA_RELOC_TREE_OBJECTID)
1494 * Error handled later, as we must prevent
1495 * extent_clear_unlock_delalloc() in error handler
1496 * from freeing metadata of created ordered extent.
1498 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1501 extent_clear_unlock_delalloc(inode, cur_offset,
1502 cur_offset + num_bytes - 1, end,
1503 locked_page, EXTENT_LOCKED |
1505 EXTENT_CLEAR_DATA_RESV,
1506 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1508 if (!nolock && nocow)
1509 btrfs_end_write_no_snapshoting(root);
1510 cur_offset = extent_end;
1513 * btrfs_reloc_clone_csums() error, now we're OK to call error
1514 * handler, as metadata for created ordered extent will only
1515 * be freed by btrfs_finish_ordered_io().
1519 if (cur_offset > end)
1522 btrfs_release_path(path);
1524 if (cur_offset <= end && cow_start == (u64)-1) {
1525 cow_start = cur_offset;
1529 if (cow_start != (u64)-1) {
1530 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1531 page_started, nr_written, 1, NULL);
1537 if (ret && cur_offset < end)
1538 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1539 locked_page, EXTENT_LOCKED |
1540 EXTENT_DELALLOC | EXTENT_DEFRAG |
1541 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1543 PAGE_SET_WRITEBACK |
1544 PAGE_END_WRITEBACK);
1545 btrfs_free_path(path);
1549 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1552 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1553 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1557 * @defrag_bytes is a hint value, no spinlock held here,
1558 * if is not zero, it means the file is defragging.
1559 * Force cow if given extent needs to be defragged.
1561 if (BTRFS_I(inode)->defrag_bytes &&
1562 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1563 EXTENT_DEFRAG, 0, NULL))
1570 * extent_io.c call back to do delayed allocation processing
1572 static int run_delalloc_range(void *private_data, struct page *locked_page,
1573 u64 start, u64 end, int *page_started,
1574 unsigned long *nr_written)
1576 struct inode *inode = private_data;
1578 int force_cow = need_force_cow(inode, start, end);
1580 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1581 ret = run_delalloc_nocow(inode, locked_page, start, end,
1582 page_started, 1, nr_written);
1583 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1584 ret = run_delalloc_nocow(inode, locked_page, start, end,
1585 page_started, 0, nr_written);
1586 } else if (!inode_need_compress(inode)) {
1587 ret = cow_file_range(inode, locked_page, start, end, end,
1588 page_started, nr_written, 1, NULL);
1590 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1591 &BTRFS_I(inode)->runtime_flags);
1592 ret = cow_file_range_async(inode, locked_page, start, end,
1593 page_started, nr_written);
1596 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1600 static void btrfs_split_extent_hook(void *private_data,
1601 struct extent_state *orig, u64 split)
1603 struct inode *inode = private_data;
1606 /* not delalloc, ignore it */
1607 if (!(orig->state & EXTENT_DELALLOC))
1610 size = orig->end - orig->start + 1;
1611 if (size > BTRFS_MAX_EXTENT_SIZE) {
1616 * See the explanation in btrfs_merge_extent_hook, the same
1617 * applies here, just in reverse.
1619 new_size = orig->end - split + 1;
1620 num_extents = count_max_extents(new_size);
1621 new_size = split - orig->start;
1622 num_extents += count_max_extents(new_size);
1623 if (count_max_extents(size) >= num_extents)
1627 spin_lock(&BTRFS_I(inode)->lock);
1628 BTRFS_I(inode)->outstanding_extents++;
1629 spin_unlock(&BTRFS_I(inode)->lock);
1633 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1634 * extents so we can keep track of new extents that are just merged onto old
1635 * extents, such as when we are doing sequential writes, so we can properly
1636 * account for the metadata space we'll need.
1638 static void btrfs_merge_extent_hook(void *private_data,
1639 struct extent_state *new,
1640 struct extent_state *other)
1642 struct inode *inode = private_data;
1643 u64 new_size, old_size;
1646 /* not delalloc, ignore it */
1647 if (!(other->state & EXTENT_DELALLOC))
1650 if (new->start > other->start)
1651 new_size = new->end - other->start + 1;
1653 new_size = other->end - new->start + 1;
1655 /* we're not bigger than the max, unreserve the space and go */
1656 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1657 spin_lock(&BTRFS_I(inode)->lock);
1658 BTRFS_I(inode)->outstanding_extents--;
1659 spin_unlock(&BTRFS_I(inode)->lock);
1664 * We have to add up either side to figure out how many extents were
1665 * accounted for before we merged into one big extent. If the number of
1666 * extents we accounted for is <= the amount we need for the new range
1667 * then we can return, otherwise drop. Think of it like this
1671 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1672 * need 2 outstanding extents, on one side we have 1 and the other side
1673 * we have 1 so they are == and we can return. But in this case
1675 * [MAX_SIZE+4k][MAX_SIZE+4k]
1677 * Each range on their own accounts for 2 extents, but merged together
1678 * they are only 3 extents worth of accounting, so we need to drop in
1681 old_size = other->end - other->start + 1;
1682 num_extents = count_max_extents(old_size);
1683 old_size = new->end - new->start + 1;
1684 num_extents += count_max_extents(old_size);
1685 if (count_max_extents(new_size) >= num_extents)
1688 spin_lock(&BTRFS_I(inode)->lock);
1689 BTRFS_I(inode)->outstanding_extents--;
1690 spin_unlock(&BTRFS_I(inode)->lock);
1693 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1694 struct inode *inode)
1696 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1698 spin_lock(&root->delalloc_lock);
1699 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1700 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1701 &root->delalloc_inodes);
1702 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1703 &BTRFS_I(inode)->runtime_flags);
1704 root->nr_delalloc_inodes++;
1705 if (root->nr_delalloc_inodes == 1) {
1706 spin_lock(&fs_info->delalloc_root_lock);
1707 BUG_ON(!list_empty(&root->delalloc_root));
1708 list_add_tail(&root->delalloc_root,
1709 &fs_info->delalloc_roots);
1710 spin_unlock(&fs_info->delalloc_root_lock);
1713 spin_unlock(&root->delalloc_lock);
1716 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1717 struct btrfs_inode *inode)
1719 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1721 spin_lock(&root->delalloc_lock);
1722 if (!list_empty(&inode->delalloc_inodes)) {
1723 list_del_init(&inode->delalloc_inodes);
1724 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1725 &inode->runtime_flags);
1726 root->nr_delalloc_inodes--;
1727 if (!root->nr_delalloc_inodes) {
1728 spin_lock(&fs_info->delalloc_root_lock);
1729 BUG_ON(list_empty(&root->delalloc_root));
1730 list_del_init(&root->delalloc_root);
1731 spin_unlock(&fs_info->delalloc_root_lock);
1734 spin_unlock(&root->delalloc_lock);
1738 * extent_io.c set_bit_hook, used to track delayed allocation
1739 * bytes in this file, and to maintain the list of inodes that
1740 * have pending delalloc work to be done.
1742 static void btrfs_set_bit_hook(void *private_data,
1743 struct extent_state *state, unsigned *bits)
1745 struct inode *inode = private_data;
1747 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1749 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1752 * set_bit and clear bit hooks normally require _irqsave/restore
1753 * but in this case, we are only testing for the DELALLOC
1754 * bit, which is only set or cleared with irqs on
1756 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1757 struct btrfs_root *root = BTRFS_I(inode)->root;
1758 u64 len = state->end + 1 - state->start;
1759 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1761 if (*bits & EXTENT_FIRST_DELALLOC) {
1762 *bits &= ~EXTENT_FIRST_DELALLOC;
1764 spin_lock(&BTRFS_I(inode)->lock);
1765 BTRFS_I(inode)->outstanding_extents++;
1766 spin_unlock(&BTRFS_I(inode)->lock);
1769 /* For sanity tests */
1770 if (btrfs_is_testing(fs_info))
1773 __percpu_counter_add(&fs_info->delalloc_bytes, len,
1774 fs_info->delalloc_batch);
1775 spin_lock(&BTRFS_I(inode)->lock);
1776 BTRFS_I(inode)->delalloc_bytes += len;
1777 if (*bits & EXTENT_DEFRAG)
1778 BTRFS_I(inode)->defrag_bytes += len;
1779 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1780 &BTRFS_I(inode)->runtime_flags))
1781 btrfs_add_delalloc_inodes(root, inode);
1782 spin_unlock(&BTRFS_I(inode)->lock);
1785 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1786 (*bits & EXTENT_DELALLOC_NEW)) {
1787 spin_lock(&BTRFS_I(inode)->lock);
1788 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1790 spin_unlock(&BTRFS_I(inode)->lock);
1795 * extent_io.c clear_bit_hook, see set_bit_hook for why
1797 static void btrfs_clear_bit_hook(void *private_data,
1798 struct extent_state *state,
1801 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1802 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1803 u64 len = state->end + 1 - state->start;
1804 u32 num_extents = count_max_extents(len);
1806 spin_lock(&inode->lock);
1807 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1808 inode->defrag_bytes -= len;
1809 spin_unlock(&inode->lock);
1812 * set_bit and clear bit hooks normally require _irqsave/restore
1813 * but in this case, we are only testing for the DELALLOC
1814 * bit, which is only set or cleared with irqs on
1816 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1817 struct btrfs_root *root = inode->root;
1818 bool do_list = !btrfs_is_free_space_inode(inode);
1820 if (*bits & EXTENT_FIRST_DELALLOC) {
1821 *bits &= ~EXTENT_FIRST_DELALLOC;
1822 } else if (!(*bits & EXTENT_CLEAR_META_RESV)) {
1823 spin_lock(&inode->lock);
1824 inode->outstanding_extents -= num_extents;
1825 spin_unlock(&inode->lock);
1829 * We don't reserve metadata space for space cache inodes so we
1830 * don't need to call dellalloc_release_metadata if there is an
1833 if (*bits & EXTENT_CLEAR_META_RESV &&
1834 root != fs_info->tree_root)
1835 btrfs_delalloc_release_metadata(inode, len);
1837 /* For sanity tests. */
1838 if (btrfs_is_testing(fs_info))
1841 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1842 do_list && !(state->state & EXTENT_NORESERVE) &&
1843 (*bits & EXTENT_CLEAR_DATA_RESV))
1844 btrfs_free_reserved_data_space_noquota(
1848 __percpu_counter_add(&fs_info->delalloc_bytes, -len,
1849 fs_info->delalloc_batch);
1850 spin_lock(&inode->lock);
1851 inode->delalloc_bytes -= len;
1852 if (do_list && inode->delalloc_bytes == 0 &&
1853 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1854 &inode->runtime_flags))
1855 btrfs_del_delalloc_inode(root, inode);
1856 spin_unlock(&inode->lock);
1859 if ((state->state & EXTENT_DELALLOC_NEW) &&
1860 (*bits & EXTENT_DELALLOC_NEW)) {
1861 spin_lock(&inode->lock);
1862 ASSERT(inode->new_delalloc_bytes >= len);
1863 inode->new_delalloc_bytes -= len;
1864 spin_unlock(&inode->lock);
1869 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1870 * we don't create bios that span stripes or chunks
1872 * return 1 if page cannot be merged to bio
1873 * return 0 if page can be merged to bio
1874 * return error otherwise
1876 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1877 size_t size, struct bio *bio,
1878 unsigned long bio_flags)
1880 struct inode *inode = page->mapping->host;
1881 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1882 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1887 if (bio_flags & EXTENT_BIO_COMPRESSED)
1890 length = bio->bi_iter.bi_size;
1891 map_length = length;
1892 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1896 if (map_length < length + size)
1902 * in order to insert checksums into the metadata in large chunks,
1903 * we wait until bio submission time. All the pages in the bio are
1904 * checksummed and sums are attached onto the ordered extent record.
1906 * At IO completion time the cums attached on the ordered extent record
1907 * are inserted into the btree
1909 static int __btrfs_submit_bio_start(void *private_data, struct bio *bio,
1910 int mirror_num, unsigned long bio_flags,
1913 struct inode *inode = private_data;
1916 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1917 BUG_ON(ret); /* -ENOMEM */
1922 * in order to insert checksums into the metadata in large chunks,
1923 * we wait until bio submission time. All the pages in the bio are
1924 * checksummed and sums are attached onto the ordered extent record.
1926 * At IO completion time the cums attached on the ordered extent record
1927 * are inserted into the btree
1929 static int __btrfs_submit_bio_done(void *private_data, struct bio *bio,
1930 int mirror_num, unsigned long bio_flags,
1933 struct inode *inode = private_data;
1934 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1937 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1939 bio->bi_error = ret;
1946 * extent_io.c submission hook. This does the right thing for csum calculation
1947 * on write, or reading the csums from the tree before a read
1949 static int btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1950 int mirror_num, unsigned long bio_flags,
1953 struct inode *inode = private_data;
1954 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1955 struct btrfs_root *root = BTRFS_I(inode)->root;
1956 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1959 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1961 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1963 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1964 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1966 if (bio_op(bio) != REQ_OP_WRITE) {
1967 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1971 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1972 ret = btrfs_submit_compressed_read(inode, bio,
1976 } else if (!skip_sum) {
1977 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
1982 } else if (async && !skip_sum) {
1983 /* csum items have already been cloned */
1984 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1986 /* we're doing a write, do the async checksumming */
1987 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
1989 __btrfs_submit_bio_start,
1990 __btrfs_submit_bio_done);
1992 } else if (!skip_sum) {
1993 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1999 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2003 bio->bi_error = ret;
2010 * given a list of ordered sums record them in the inode. This happens
2011 * at IO completion time based on sums calculated at bio submission time.
2013 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2014 struct inode *inode, struct list_head *list)
2016 struct btrfs_ordered_sum *sum;
2018 list_for_each_entry(sum, list, list) {
2019 trans->adding_csums = 1;
2020 btrfs_csum_file_blocks(trans,
2021 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2022 trans->adding_csums = 0;
2027 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2028 struct extent_state **cached_state, int dedupe)
2030 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2031 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2035 /* see btrfs_writepage_start_hook for details on why this is required */
2036 struct btrfs_writepage_fixup {
2038 struct btrfs_work work;
2041 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2043 struct btrfs_writepage_fixup *fixup;
2044 struct btrfs_ordered_extent *ordered;
2045 struct extent_state *cached_state = NULL;
2047 struct inode *inode;
2052 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2056 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2057 ClearPageChecked(page);
2061 inode = page->mapping->host;
2062 page_start = page_offset(page);
2063 page_end = page_offset(page) + PAGE_SIZE - 1;
2065 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2068 /* already ordered? We're done */
2069 if (PagePrivate2(page))
2072 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2075 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2076 page_end, &cached_state, GFP_NOFS);
2078 btrfs_start_ordered_extent(inode, ordered, 1);
2079 btrfs_put_ordered_extent(ordered);
2083 ret = btrfs_delalloc_reserve_space(inode, page_start,
2086 mapping_set_error(page->mapping, ret);
2087 end_extent_writepage(page, ret, page_start, page_end);
2088 ClearPageChecked(page);
2092 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state,
2094 ClearPageChecked(page);
2095 set_page_dirty(page);
2097 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2098 &cached_state, GFP_NOFS);
2106 * There are a few paths in the higher layers of the kernel that directly
2107 * set the page dirty bit without asking the filesystem if it is a
2108 * good idea. This causes problems because we want to make sure COW
2109 * properly happens and the data=ordered rules are followed.
2111 * In our case any range that doesn't have the ORDERED bit set
2112 * hasn't been properly setup for IO. We kick off an async process
2113 * to fix it up. The async helper will wait for ordered extents, set
2114 * the delalloc bit and make it safe to write the page.
2116 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2118 struct inode *inode = page->mapping->host;
2119 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2120 struct btrfs_writepage_fixup *fixup;
2122 /* this page is properly in the ordered list */
2123 if (TestClearPagePrivate2(page))
2126 if (PageChecked(page))
2129 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2133 SetPageChecked(page);
2135 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2136 btrfs_writepage_fixup_worker, NULL, NULL);
2138 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2142 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2143 struct inode *inode, u64 file_pos,
2144 u64 disk_bytenr, u64 disk_num_bytes,
2145 u64 num_bytes, u64 ram_bytes,
2146 u8 compression, u8 encryption,
2147 u16 other_encoding, int extent_type)
2149 struct btrfs_root *root = BTRFS_I(inode)->root;
2150 struct btrfs_file_extent_item *fi;
2151 struct btrfs_path *path;
2152 struct extent_buffer *leaf;
2153 struct btrfs_key ins;
2154 int extent_inserted = 0;
2157 path = btrfs_alloc_path();
2162 * we may be replacing one extent in the tree with another.
2163 * The new extent is pinned in the extent map, and we don't want
2164 * to drop it from the cache until it is completely in the btree.
2166 * So, tell btrfs_drop_extents to leave this extent in the cache.
2167 * the caller is expected to unpin it and allow it to be merged
2170 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2171 file_pos + num_bytes, NULL, 0,
2172 1, sizeof(*fi), &extent_inserted);
2176 if (!extent_inserted) {
2177 ins.objectid = btrfs_ino(BTRFS_I(inode));
2178 ins.offset = file_pos;
2179 ins.type = BTRFS_EXTENT_DATA_KEY;
2181 path->leave_spinning = 1;
2182 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2187 leaf = path->nodes[0];
2188 fi = btrfs_item_ptr(leaf, path->slots[0],
2189 struct btrfs_file_extent_item);
2190 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2191 btrfs_set_file_extent_type(leaf, fi, extent_type);
2192 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2193 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2194 btrfs_set_file_extent_offset(leaf, fi, 0);
2195 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2196 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2197 btrfs_set_file_extent_compression(leaf, fi, compression);
2198 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2199 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2201 btrfs_mark_buffer_dirty(leaf);
2202 btrfs_release_path(path);
2204 inode_add_bytes(inode, num_bytes);
2206 ins.objectid = disk_bytenr;
2207 ins.offset = disk_num_bytes;
2208 ins.type = BTRFS_EXTENT_ITEM_KEY;
2209 ret = btrfs_alloc_reserved_file_extent(trans, root->root_key.objectid,
2210 btrfs_ino(BTRFS_I(inode)), file_pos, ram_bytes, &ins);
2212 * Release the reserved range from inode dirty range map, as it is
2213 * already moved into delayed_ref_head
2215 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2217 btrfs_free_path(path);
2222 /* snapshot-aware defrag */
2223 struct sa_defrag_extent_backref {
2224 struct rb_node node;
2225 struct old_sa_defrag_extent *old;
2234 struct old_sa_defrag_extent {
2235 struct list_head list;
2236 struct new_sa_defrag_extent *new;
2245 struct new_sa_defrag_extent {
2246 struct rb_root root;
2247 struct list_head head;
2248 struct btrfs_path *path;
2249 struct inode *inode;
2257 static int backref_comp(struct sa_defrag_extent_backref *b1,
2258 struct sa_defrag_extent_backref *b2)
2260 if (b1->root_id < b2->root_id)
2262 else if (b1->root_id > b2->root_id)
2265 if (b1->inum < b2->inum)
2267 else if (b1->inum > b2->inum)
2270 if (b1->file_pos < b2->file_pos)
2272 else if (b1->file_pos > b2->file_pos)
2276 * [------------------------------] ===> (a range of space)
2277 * |<--->| |<---->| =============> (fs/file tree A)
2278 * |<---------------------------->| ===> (fs/file tree B)
2280 * A range of space can refer to two file extents in one tree while
2281 * refer to only one file extent in another tree.
2283 * So we may process a disk offset more than one time(two extents in A)
2284 * and locate at the same extent(one extent in B), then insert two same
2285 * backrefs(both refer to the extent in B).
2290 static void backref_insert(struct rb_root *root,
2291 struct sa_defrag_extent_backref *backref)
2293 struct rb_node **p = &root->rb_node;
2294 struct rb_node *parent = NULL;
2295 struct sa_defrag_extent_backref *entry;
2300 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2302 ret = backref_comp(backref, entry);
2306 p = &(*p)->rb_right;
2309 rb_link_node(&backref->node, parent, p);
2310 rb_insert_color(&backref->node, root);
2314 * Note the backref might has changed, and in this case we just return 0.
2316 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2319 struct btrfs_file_extent_item *extent;
2320 struct old_sa_defrag_extent *old = ctx;
2321 struct new_sa_defrag_extent *new = old->new;
2322 struct btrfs_path *path = new->path;
2323 struct btrfs_key key;
2324 struct btrfs_root *root;
2325 struct sa_defrag_extent_backref *backref;
2326 struct extent_buffer *leaf;
2327 struct inode *inode = new->inode;
2328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2334 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2335 inum == btrfs_ino(BTRFS_I(inode)))
2338 key.objectid = root_id;
2339 key.type = BTRFS_ROOT_ITEM_KEY;
2340 key.offset = (u64)-1;
2342 root = btrfs_read_fs_root_no_name(fs_info, &key);
2344 if (PTR_ERR(root) == -ENOENT)
2347 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2348 inum, offset, root_id);
2349 return PTR_ERR(root);
2352 key.objectid = inum;
2353 key.type = BTRFS_EXTENT_DATA_KEY;
2354 if (offset > (u64)-1 << 32)
2357 key.offset = offset;
2359 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2360 if (WARN_ON(ret < 0))
2367 leaf = path->nodes[0];
2368 slot = path->slots[0];
2370 if (slot >= btrfs_header_nritems(leaf)) {
2371 ret = btrfs_next_leaf(root, path);
2374 } else if (ret > 0) {
2383 btrfs_item_key_to_cpu(leaf, &key, slot);
2385 if (key.objectid > inum)
2388 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2391 extent = btrfs_item_ptr(leaf, slot,
2392 struct btrfs_file_extent_item);
2394 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2398 * 'offset' refers to the exact key.offset,
2399 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2400 * (key.offset - extent_offset).
2402 if (key.offset != offset)
2405 extent_offset = btrfs_file_extent_offset(leaf, extent);
2406 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2408 if (extent_offset >= old->extent_offset + old->offset +
2409 old->len || extent_offset + num_bytes <=
2410 old->extent_offset + old->offset)
2415 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2421 backref->root_id = root_id;
2422 backref->inum = inum;
2423 backref->file_pos = offset;
2424 backref->num_bytes = num_bytes;
2425 backref->extent_offset = extent_offset;
2426 backref->generation = btrfs_file_extent_generation(leaf, extent);
2428 backref_insert(&new->root, backref);
2431 btrfs_release_path(path);
2436 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2437 struct new_sa_defrag_extent *new)
2439 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2440 struct old_sa_defrag_extent *old, *tmp;
2445 list_for_each_entry_safe(old, tmp, &new->head, list) {
2446 ret = iterate_inodes_from_logical(old->bytenr +
2447 old->extent_offset, fs_info,
2448 path, record_one_backref,
2450 if (ret < 0 && ret != -ENOENT)
2453 /* no backref to be processed for this extent */
2455 list_del(&old->list);
2460 if (list_empty(&new->head))
2466 static int relink_is_mergable(struct extent_buffer *leaf,
2467 struct btrfs_file_extent_item *fi,
2468 struct new_sa_defrag_extent *new)
2470 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2473 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2476 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2479 if (btrfs_file_extent_encryption(leaf, fi) ||
2480 btrfs_file_extent_other_encoding(leaf, fi))
2487 * Note the backref might has changed, and in this case we just return 0.
2489 static noinline int relink_extent_backref(struct btrfs_path *path,
2490 struct sa_defrag_extent_backref *prev,
2491 struct sa_defrag_extent_backref *backref)
2493 struct btrfs_file_extent_item *extent;
2494 struct btrfs_file_extent_item *item;
2495 struct btrfs_ordered_extent *ordered;
2496 struct btrfs_trans_handle *trans;
2497 struct btrfs_root *root;
2498 struct btrfs_key key;
2499 struct extent_buffer *leaf;
2500 struct old_sa_defrag_extent *old = backref->old;
2501 struct new_sa_defrag_extent *new = old->new;
2502 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2503 struct inode *inode;
2504 struct extent_state *cached = NULL;
2513 if (prev && prev->root_id == backref->root_id &&
2514 prev->inum == backref->inum &&
2515 prev->file_pos + prev->num_bytes == backref->file_pos)
2518 /* step 1: get root */
2519 key.objectid = backref->root_id;
2520 key.type = BTRFS_ROOT_ITEM_KEY;
2521 key.offset = (u64)-1;
2523 index = srcu_read_lock(&fs_info->subvol_srcu);
2525 root = btrfs_read_fs_root_no_name(fs_info, &key);
2527 srcu_read_unlock(&fs_info->subvol_srcu, index);
2528 if (PTR_ERR(root) == -ENOENT)
2530 return PTR_ERR(root);
2533 if (btrfs_root_readonly(root)) {
2534 srcu_read_unlock(&fs_info->subvol_srcu, index);
2538 /* step 2: get inode */
2539 key.objectid = backref->inum;
2540 key.type = BTRFS_INODE_ITEM_KEY;
2543 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2544 if (IS_ERR(inode)) {
2545 srcu_read_unlock(&fs_info->subvol_srcu, index);
2549 srcu_read_unlock(&fs_info->subvol_srcu, index);
2551 /* step 3: relink backref */
2552 lock_start = backref->file_pos;
2553 lock_end = backref->file_pos + backref->num_bytes - 1;
2554 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2557 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2559 btrfs_put_ordered_extent(ordered);
2563 trans = btrfs_join_transaction(root);
2564 if (IS_ERR(trans)) {
2565 ret = PTR_ERR(trans);
2569 key.objectid = backref->inum;
2570 key.type = BTRFS_EXTENT_DATA_KEY;
2571 key.offset = backref->file_pos;
2573 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2576 } else if (ret > 0) {
2581 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2582 struct btrfs_file_extent_item);
2584 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2585 backref->generation)
2588 btrfs_release_path(path);
2590 start = backref->file_pos;
2591 if (backref->extent_offset < old->extent_offset + old->offset)
2592 start += old->extent_offset + old->offset -
2593 backref->extent_offset;
2595 len = min(backref->extent_offset + backref->num_bytes,
2596 old->extent_offset + old->offset + old->len);
2597 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2599 ret = btrfs_drop_extents(trans, root, inode, start,
2604 key.objectid = btrfs_ino(BTRFS_I(inode));
2605 key.type = BTRFS_EXTENT_DATA_KEY;
2608 path->leave_spinning = 1;
2610 struct btrfs_file_extent_item *fi;
2612 struct btrfs_key found_key;
2614 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2619 leaf = path->nodes[0];
2620 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2622 fi = btrfs_item_ptr(leaf, path->slots[0],
2623 struct btrfs_file_extent_item);
2624 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2626 if (extent_len + found_key.offset == start &&
2627 relink_is_mergable(leaf, fi, new)) {
2628 btrfs_set_file_extent_num_bytes(leaf, fi,
2630 btrfs_mark_buffer_dirty(leaf);
2631 inode_add_bytes(inode, len);
2637 btrfs_release_path(path);
2642 ret = btrfs_insert_empty_item(trans, root, path, &key,
2645 btrfs_abort_transaction(trans, ret);
2649 leaf = path->nodes[0];
2650 item = btrfs_item_ptr(leaf, path->slots[0],
2651 struct btrfs_file_extent_item);
2652 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2653 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2654 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2655 btrfs_set_file_extent_num_bytes(leaf, item, len);
2656 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2657 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2658 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2659 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2660 btrfs_set_file_extent_encryption(leaf, item, 0);
2661 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2663 btrfs_mark_buffer_dirty(leaf);
2664 inode_add_bytes(inode, len);
2665 btrfs_release_path(path);
2667 ret = btrfs_inc_extent_ref(trans, fs_info, new->bytenr,
2669 backref->root_id, backref->inum,
2670 new->file_pos); /* start - extent_offset */
2672 btrfs_abort_transaction(trans, ret);
2678 btrfs_release_path(path);
2679 path->leave_spinning = 0;
2680 btrfs_end_transaction(trans);
2682 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2688 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2690 struct old_sa_defrag_extent *old, *tmp;
2695 list_for_each_entry_safe(old, tmp, &new->head, list) {
2701 static void relink_file_extents(struct new_sa_defrag_extent *new)
2703 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2704 struct btrfs_path *path;
2705 struct sa_defrag_extent_backref *backref;
2706 struct sa_defrag_extent_backref *prev = NULL;
2707 struct inode *inode;
2708 struct btrfs_root *root;
2709 struct rb_node *node;
2713 root = BTRFS_I(inode)->root;
2715 path = btrfs_alloc_path();
2719 if (!record_extent_backrefs(path, new)) {
2720 btrfs_free_path(path);
2723 btrfs_release_path(path);
2726 node = rb_first(&new->root);
2729 rb_erase(node, &new->root);
2731 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2733 ret = relink_extent_backref(path, prev, backref);
2746 btrfs_free_path(path);
2748 free_sa_defrag_extent(new);
2750 atomic_dec(&fs_info->defrag_running);
2751 wake_up(&fs_info->transaction_wait);
2754 static struct new_sa_defrag_extent *
2755 record_old_file_extents(struct inode *inode,
2756 struct btrfs_ordered_extent *ordered)
2758 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2759 struct btrfs_root *root = BTRFS_I(inode)->root;
2760 struct btrfs_path *path;
2761 struct btrfs_key key;
2762 struct old_sa_defrag_extent *old;
2763 struct new_sa_defrag_extent *new;
2766 new = kmalloc(sizeof(*new), GFP_NOFS);
2771 new->file_pos = ordered->file_offset;
2772 new->len = ordered->len;
2773 new->bytenr = ordered->start;
2774 new->disk_len = ordered->disk_len;
2775 new->compress_type = ordered->compress_type;
2776 new->root = RB_ROOT;
2777 INIT_LIST_HEAD(&new->head);
2779 path = btrfs_alloc_path();
2783 key.objectid = btrfs_ino(BTRFS_I(inode));
2784 key.type = BTRFS_EXTENT_DATA_KEY;
2785 key.offset = new->file_pos;
2787 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2790 if (ret > 0 && path->slots[0] > 0)
2793 /* find out all the old extents for the file range */
2795 struct btrfs_file_extent_item *extent;
2796 struct extent_buffer *l;
2805 slot = path->slots[0];
2807 if (slot >= btrfs_header_nritems(l)) {
2808 ret = btrfs_next_leaf(root, path);
2816 btrfs_item_key_to_cpu(l, &key, slot);
2818 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2820 if (key.type != BTRFS_EXTENT_DATA_KEY)
2822 if (key.offset >= new->file_pos + new->len)
2825 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2827 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2828 if (key.offset + num_bytes < new->file_pos)
2831 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2835 extent_offset = btrfs_file_extent_offset(l, extent);
2837 old = kmalloc(sizeof(*old), GFP_NOFS);
2841 offset = max(new->file_pos, key.offset);
2842 end = min(new->file_pos + new->len, key.offset + num_bytes);
2844 old->bytenr = disk_bytenr;
2845 old->extent_offset = extent_offset;
2846 old->offset = offset - key.offset;
2847 old->len = end - offset;
2850 list_add_tail(&old->list, &new->head);
2856 btrfs_free_path(path);
2857 atomic_inc(&fs_info->defrag_running);
2862 btrfs_free_path(path);
2864 free_sa_defrag_extent(new);
2868 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2871 struct btrfs_block_group_cache *cache;
2873 cache = btrfs_lookup_block_group(fs_info, start);
2876 spin_lock(&cache->lock);
2877 cache->delalloc_bytes -= len;
2878 spin_unlock(&cache->lock);
2880 btrfs_put_block_group(cache);
2883 /* as ordered data IO finishes, this gets called so we can finish
2884 * an ordered extent if the range of bytes in the file it covers are
2887 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2889 struct inode *inode = ordered_extent->inode;
2890 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2891 struct btrfs_root *root = BTRFS_I(inode)->root;
2892 struct btrfs_trans_handle *trans = NULL;
2893 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2894 struct extent_state *cached_state = NULL;
2895 struct new_sa_defrag_extent *new = NULL;
2896 int compress_type = 0;
2898 u64 logical_len = ordered_extent->len;
2900 bool truncated = false;
2901 bool range_locked = false;
2902 bool clear_new_delalloc_bytes = false;
2904 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2905 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2906 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2907 clear_new_delalloc_bytes = true;
2909 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2911 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2916 btrfs_free_io_failure_record(BTRFS_I(inode),
2917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
2919 ordered_extent->len - 1);
2921 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2923 logical_len = ordered_extent->truncated_len;
2924 /* Truncated the entire extent, don't bother adding */
2929 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2930 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2933 * For mwrite(mmap + memset to write) case, we still reserve
2934 * space for NOCOW range.
2935 * As NOCOW won't cause a new delayed ref, just free the space
2937 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2938 ordered_extent->len);
2939 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2941 trans = btrfs_join_transaction_nolock(root);
2943 trans = btrfs_join_transaction(root);
2944 if (IS_ERR(trans)) {
2945 ret = PTR_ERR(trans);
2949 trans->block_rsv = &fs_info->delalloc_block_rsv;
2950 ret = btrfs_update_inode_fallback(trans, root, inode);
2951 if (ret) /* -ENOMEM or corruption */
2952 btrfs_abort_transaction(trans, ret);
2956 range_locked = true;
2957 lock_extent_bits(io_tree, ordered_extent->file_offset,
2958 ordered_extent->file_offset + ordered_extent->len - 1,
2961 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2962 ordered_extent->file_offset + ordered_extent->len - 1,
2963 EXTENT_DEFRAG, 0, cached_state);
2965 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2966 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2967 /* the inode is shared */
2968 new = record_old_file_extents(inode, ordered_extent);
2970 clear_extent_bit(io_tree, ordered_extent->file_offset,
2971 ordered_extent->file_offset + ordered_extent->len - 1,
2972 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2976 trans = btrfs_join_transaction_nolock(root);
2978 trans = btrfs_join_transaction(root);
2979 if (IS_ERR(trans)) {
2980 ret = PTR_ERR(trans);
2985 trans->block_rsv = &fs_info->delalloc_block_rsv;
2987 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2988 compress_type = ordered_extent->compress_type;
2989 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2990 BUG_ON(compress_type);
2991 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
2992 ordered_extent->file_offset,
2993 ordered_extent->file_offset +
2996 BUG_ON(root == fs_info->tree_root);
2997 ret = insert_reserved_file_extent(trans, inode,
2998 ordered_extent->file_offset,
2999 ordered_extent->start,
3000 ordered_extent->disk_len,
3001 logical_len, logical_len,
3002 compress_type, 0, 0,
3003 BTRFS_FILE_EXTENT_REG);
3005 btrfs_release_delalloc_bytes(fs_info,
3006 ordered_extent->start,
3007 ordered_extent->disk_len);
3009 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3010 ordered_extent->file_offset, ordered_extent->len,
3013 btrfs_abort_transaction(trans, ret);
3017 add_pending_csums(trans, inode, &ordered_extent->list);
3019 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3020 ret = btrfs_update_inode_fallback(trans, root, inode);
3021 if (ret) { /* -ENOMEM or corruption */
3022 btrfs_abort_transaction(trans, ret);
3027 if (range_locked || clear_new_delalloc_bytes) {
3028 unsigned int clear_bits = 0;
3031 clear_bits |= EXTENT_LOCKED;
3032 if (clear_new_delalloc_bytes)
3033 clear_bits |= EXTENT_DELALLOC_NEW;
3034 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3035 ordered_extent->file_offset,
3036 ordered_extent->file_offset +
3037 ordered_extent->len - 1,
3039 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3040 0, &cached_state, GFP_NOFS);
3043 if (root != fs_info->tree_root)
3044 btrfs_delalloc_release_metadata(BTRFS_I(inode),
3045 ordered_extent->len);
3047 btrfs_end_transaction(trans);
3049 if (ret || truncated) {
3053 start = ordered_extent->file_offset + logical_len;
3055 start = ordered_extent->file_offset;
3056 end = ordered_extent->file_offset + ordered_extent->len - 1;
3057 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
3059 /* Drop the cache for the part of the extent we didn't write. */
3060 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3063 * If the ordered extent had an IOERR or something else went
3064 * wrong we need to return the space for this ordered extent
3065 * back to the allocator. We only free the extent in the
3066 * truncated case if we didn't write out the extent at all.
3068 if ((ret || !logical_len) &&
3069 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3070 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3071 btrfs_free_reserved_extent(fs_info,
3072 ordered_extent->start,
3073 ordered_extent->disk_len, 1);
3078 * This needs to be done to make sure anybody waiting knows we are done
3079 * updating everything for this ordered extent.
3081 btrfs_remove_ordered_extent(inode, ordered_extent);
3083 /* for snapshot-aware defrag */
3086 free_sa_defrag_extent(new);
3087 atomic_dec(&fs_info->defrag_running);
3089 relink_file_extents(new);
3094 btrfs_put_ordered_extent(ordered_extent);
3095 /* once for the tree */
3096 btrfs_put_ordered_extent(ordered_extent);
3101 static void finish_ordered_fn(struct btrfs_work *work)
3103 struct btrfs_ordered_extent *ordered_extent;
3104 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3105 btrfs_finish_ordered_io(ordered_extent);
3108 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3109 struct extent_state *state, int uptodate)
3111 struct inode *inode = page->mapping->host;
3112 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3113 struct btrfs_ordered_extent *ordered_extent = NULL;
3114 struct btrfs_workqueue *wq;
3115 btrfs_work_func_t func;
3117 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3119 ClearPagePrivate2(page);
3120 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3121 end - start + 1, uptodate))
3124 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3125 wq = fs_info->endio_freespace_worker;
3126 func = btrfs_freespace_write_helper;
3128 wq = fs_info->endio_write_workers;
3129 func = btrfs_endio_write_helper;
3132 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3134 btrfs_queue_work(wq, &ordered_extent->work);
3137 static int __readpage_endio_check(struct inode *inode,
3138 struct btrfs_io_bio *io_bio,
3139 int icsum, struct page *page,
3140 int pgoff, u64 start, size_t len)
3146 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3148 kaddr = kmap_atomic(page);
3149 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3150 btrfs_csum_final(csum, (u8 *)&csum);
3151 if (csum != csum_expected)
3154 kunmap_atomic(kaddr);
3157 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3158 io_bio->mirror_num);
3159 memset(kaddr + pgoff, 1, len);
3160 flush_dcache_page(page);
3161 kunmap_atomic(kaddr);
3162 if (csum_expected == 0)
3168 * when reads are done, we need to check csums to verify the data is correct
3169 * if there's a match, we allow the bio to finish. If not, the code in
3170 * extent_io.c will try to find good copies for us.
3172 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3173 u64 phy_offset, struct page *page,
3174 u64 start, u64 end, int mirror)
3176 size_t offset = start - page_offset(page);
3177 struct inode *inode = page->mapping->host;
3178 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3179 struct btrfs_root *root = BTRFS_I(inode)->root;
3181 if (PageChecked(page)) {
3182 ClearPageChecked(page);
3186 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3189 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3190 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3191 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3195 phy_offset >>= inode->i_sb->s_blocksize_bits;
3196 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3197 start, (size_t)(end - start + 1));
3200 void btrfs_add_delayed_iput(struct inode *inode)
3202 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3203 struct btrfs_inode *binode = BTRFS_I(inode);
3205 if (atomic_add_unless(&inode->i_count, -1, 1))
3208 spin_lock(&fs_info->delayed_iput_lock);
3209 if (binode->delayed_iput_count == 0) {
3210 ASSERT(list_empty(&binode->delayed_iput));
3211 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3213 binode->delayed_iput_count++;
3215 spin_unlock(&fs_info->delayed_iput_lock);
3218 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3221 spin_lock(&fs_info->delayed_iput_lock);
3222 while (!list_empty(&fs_info->delayed_iputs)) {
3223 struct btrfs_inode *inode;
3225 inode = list_first_entry(&fs_info->delayed_iputs,
3226 struct btrfs_inode, delayed_iput);
3227 if (inode->delayed_iput_count) {
3228 inode->delayed_iput_count--;
3229 list_move_tail(&inode->delayed_iput,
3230 &fs_info->delayed_iputs);
3232 list_del_init(&inode->delayed_iput);
3234 spin_unlock(&fs_info->delayed_iput_lock);
3235 iput(&inode->vfs_inode);
3236 spin_lock(&fs_info->delayed_iput_lock);
3238 spin_unlock(&fs_info->delayed_iput_lock);
3242 * This is called in transaction commit time. If there are no orphan
3243 * files in the subvolume, it removes orphan item and frees block_rsv
3246 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3247 struct btrfs_root *root)
3249 struct btrfs_fs_info *fs_info = root->fs_info;
3250 struct btrfs_block_rsv *block_rsv;
3253 if (atomic_read(&root->orphan_inodes) ||
3254 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3257 spin_lock(&root->orphan_lock);
3258 if (atomic_read(&root->orphan_inodes)) {
3259 spin_unlock(&root->orphan_lock);
3263 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3264 spin_unlock(&root->orphan_lock);
3268 block_rsv = root->orphan_block_rsv;
3269 root->orphan_block_rsv = NULL;
3270 spin_unlock(&root->orphan_lock);
3272 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3273 btrfs_root_refs(&root->root_item) > 0) {
3274 ret = btrfs_del_orphan_item(trans, fs_info->tree_root,
3275 root->root_key.objectid);
3277 btrfs_abort_transaction(trans, ret);
3279 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3284 WARN_ON(block_rsv->size > 0);
3285 btrfs_free_block_rsv(fs_info, block_rsv);
3290 * This creates an orphan entry for the given inode in case something goes
3291 * wrong in the middle of an unlink/truncate.
3293 * NOTE: caller of this function should reserve 5 units of metadata for
3296 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3297 struct btrfs_inode *inode)
3299 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
3300 struct btrfs_root *root = inode->root;
3301 struct btrfs_block_rsv *block_rsv = NULL;
3306 if (!root->orphan_block_rsv) {
3307 block_rsv = btrfs_alloc_block_rsv(fs_info,
3308 BTRFS_BLOCK_RSV_TEMP);
3313 spin_lock(&root->orphan_lock);
3314 if (!root->orphan_block_rsv) {
3315 root->orphan_block_rsv = block_rsv;
3316 } else if (block_rsv) {
3317 btrfs_free_block_rsv(fs_info, block_rsv);
3321 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3322 &inode->runtime_flags)) {
3325 * For proper ENOSPC handling, we should do orphan
3326 * cleanup when mounting. But this introduces backward
3327 * compatibility issue.
3329 if (!xchg(&root->orphan_item_inserted, 1))
3335 atomic_inc(&root->orphan_inodes);
3338 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3339 &inode->runtime_flags))
3341 spin_unlock(&root->orphan_lock);
3343 /* grab metadata reservation from transaction handle */
3345 ret = btrfs_orphan_reserve_metadata(trans, inode);
3348 atomic_dec(&root->orphan_inodes);
3349 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3350 &inode->runtime_flags);
3352 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3353 &inode->runtime_flags);
3358 /* insert an orphan item to track this unlinked/truncated file */
3360 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3362 atomic_dec(&root->orphan_inodes);
3364 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3365 &inode->runtime_flags);
3366 btrfs_orphan_release_metadata(inode);
3368 if (ret != -EEXIST) {
3369 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3370 &inode->runtime_flags);
3371 btrfs_abort_transaction(trans, ret);
3378 /* insert an orphan item to track subvolume contains orphan files */
3380 ret = btrfs_insert_orphan_item(trans, fs_info->tree_root,
3381 root->root_key.objectid);
3382 if (ret && ret != -EEXIST) {
3383 btrfs_abort_transaction(trans, ret);
3391 * We have done the truncate/delete so we can go ahead and remove the orphan
3392 * item for this particular inode.
3394 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3395 struct btrfs_inode *inode)
3397 struct btrfs_root *root = inode->root;
3398 int delete_item = 0;
3399 int release_rsv = 0;
3402 spin_lock(&root->orphan_lock);
3403 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3404 &inode->runtime_flags))
3407 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3408 &inode->runtime_flags))
3410 spin_unlock(&root->orphan_lock);
3413 atomic_dec(&root->orphan_inodes);
3415 ret = btrfs_del_orphan_item(trans, root,
3420 btrfs_orphan_release_metadata(inode);
3426 * this cleans up any orphans that may be left on the list from the last use
3429 int btrfs_orphan_cleanup(struct btrfs_root *root)
3431 struct btrfs_fs_info *fs_info = root->fs_info;
3432 struct btrfs_path *path;
3433 struct extent_buffer *leaf;
3434 struct btrfs_key key, found_key;
3435 struct btrfs_trans_handle *trans;
3436 struct inode *inode;
3437 u64 last_objectid = 0;
3438 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3440 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3443 path = btrfs_alloc_path();
3448 path->reada = READA_BACK;
3450 key.objectid = BTRFS_ORPHAN_OBJECTID;
3451 key.type = BTRFS_ORPHAN_ITEM_KEY;
3452 key.offset = (u64)-1;
3455 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3460 * if ret == 0 means we found what we were searching for, which
3461 * is weird, but possible, so only screw with path if we didn't
3462 * find the key and see if we have stuff that matches
3466 if (path->slots[0] == 0)
3471 /* pull out the item */
3472 leaf = path->nodes[0];
3473 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3475 /* make sure the item matches what we want */
3476 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3478 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3481 /* release the path since we're done with it */
3482 btrfs_release_path(path);
3485 * this is where we are basically btrfs_lookup, without the
3486 * crossing root thing. we store the inode number in the
3487 * offset of the orphan item.
3490 if (found_key.offset == last_objectid) {
3492 "Error removing orphan entry, stopping orphan cleanup");
3497 last_objectid = found_key.offset;
3499 found_key.objectid = found_key.offset;
3500 found_key.type = BTRFS_INODE_ITEM_KEY;
3501 found_key.offset = 0;
3502 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3503 ret = PTR_ERR_OR_ZERO(inode);
3504 if (ret && ret != -ENOENT)
3507 if (ret == -ENOENT && root == fs_info->tree_root) {
3508 struct btrfs_root *dead_root;
3509 struct btrfs_fs_info *fs_info = root->fs_info;
3510 int is_dead_root = 0;
3513 * this is an orphan in the tree root. Currently these
3514 * could come from 2 sources:
3515 * a) a snapshot deletion in progress
3516 * b) a free space cache inode
3517 * We need to distinguish those two, as the snapshot
3518 * orphan must not get deleted.
3519 * find_dead_roots already ran before us, so if this
3520 * is a snapshot deletion, we should find the root
3521 * in the dead_roots list
3523 spin_lock(&fs_info->trans_lock);
3524 list_for_each_entry(dead_root, &fs_info->dead_roots,
3526 if (dead_root->root_key.objectid ==
3527 found_key.objectid) {
3532 spin_unlock(&fs_info->trans_lock);
3534 /* prevent this orphan from being found again */
3535 key.offset = found_key.objectid - 1;
3540 * Inode is already gone but the orphan item is still there,
3541 * kill the orphan item.
3543 if (ret == -ENOENT) {
3544 trans = btrfs_start_transaction(root, 1);
3545 if (IS_ERR(trans)) {
3546 ret = PTR_ERR(trans);
3549 btrfs_debug(fs_info, "auto deleting %Lu",
3550 found_key.objectid);
3551 ret = btrfs_del_orphan_item(trans, root,
3552 found_key.objectid);
3553 btrfs_end_transaction(trans);
3560 * add this inode to the orphan list so btrfs_orphan_del does
3561 * the proper thing when we hit it
3563 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3564 &BTRFS_I(inode)->runtime_flags);
3565 atomic_inc(&root->orphan_inodes);
3567 /* if we have links, this was a truncate, lets do that */
3568 if (inode->i_nlink) {
3569 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3575 /* 1 for the orphan item deletion. */
3576 trans = btrfs_start_transaction(root, 1);
3577 if (IS_ERR(trans)) {
3579 ret = PTR_ERR(trans);
3582 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3583 btrfs_end_transaction(trans);
3589 ret = btrfs_truncate(inode);
3591 btrfs_orphan_del(NULL, BTRFS_I(inode));
3596 /* this will do delete_inode and everything for us */
3601 /* release the path since we're done with it */
3602 btrfs_release_path(path);
3604 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3606 if (root->orphan_block_rsv)
3607 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv,
3610 if (root->orphan_block_rsv ||
3611 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3612 trans = btrfs_join_transaction(root);
3614 btrfs_end_transaction(trans);
3618 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3620 btrfs_debug(fs_info, "truncated %d orphans", nr_truncate);
3624 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3625 btrfs_free_path(path);
3630 * very simple check to peek ahead in the leaf looking for xattrs. If we
3631 * don't find any xattrs, we know there can't be any acls.
3633 * slot is the slot the inode is in, objectid is the objectid of the inode
3635 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3636 int slot, u64 objectid,
3637 int *first_xattr_slot)
3639 u32 nritems = btrfs_header_nritems(leaf);
3640 struct btrfs_key found_key;
3641 static u64 xattr_access = 0;
3642 static u64 xattr_default = 0;
3645 if (!xattr_access) {
3646 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3647 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3648 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3649 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3653 *first_xattr_slot = -1;
3654 while (slot < nritems) {
3655 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3657 /* we found a different objectid, there must not be acls */
3658 if (found_key.objectid != objectid)
3661 /* we found an xattr, assume we've got an acl */
3662 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3663 if (*first_xattr_slot == -1)
3664 *first_xattr_slot = slot;
3665 if (found_key.offset == xattr_access ||
3666 found_key.offset == xattr_default)
3671 * we found a key greater than an xattr key, there can't
3672 * be any acls later on
3674 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3681 * it goes inode, inode backrefs, xattrs, extents,
3682 * so if there are a ton of hard links to an inode there can
3683 * be a lot of backrefs. Don't waste time searching too hard,
3684 * this is just an optimization
3689 /* we hit the end of the leaf before we found an xattr or
3690 * something larger than an xattr. We have to assume the inode
3693 if (*first_xattr_slot == -1)
3694 *first_xattr_slot = slot;
3699 * read an inode from the btree into the in-memory inode
3701 static int btrfs_read_locked_inode(struct inode *inode)
3703 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3704 struct btrfs_path *path;
3705 struct extent_buffer *leaf;
3706 struct btrfs_inode_item *inode_item;
3707 struct btrfs_root *root = BTRFS_I(inode)->root;
3708 struct btrfs_key location;
3713 bool filled = false;
3714 int first_xattr_slot;
3716 ret = btrfs_fill_inode(inode, &rdev);
3720 path = btrfs_alloc_path();
3726 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3728 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3735 leaf = path->nodes[0];
3740 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3741 struct btrfs_inode_item);
3742 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3743 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3744 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3745 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3746 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3748 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3749 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3751 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3752 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3754 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3755 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3757 BTRFS_I(inode)->i_otime.tv_sec =
3758 btrfs_timespec_sec(leaf, &inode_item->otime);
3759 BTRFS_I(inode)->i_otime.tv_nsec =
3760 btrfs_timespec_nsec(leaf, &inode_item->otime);
3762 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3763 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3764 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3766 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3767 inode->i_generation = BTRFS_I(inode)->generation;
3769 rdev = btrfs_inode_rdev(leaf, inode_item);
3771 BTRFS_I(inode)->index_cnt = (u64)-1;
3772 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3776 * If we were modified in the current generation and evicted from memory
3777 * and then re-read we need to do a full sync since we don't have any
3778 * idea about which extents were modified before we were evicted from
3781 * This is required for both inode re-read from disk and delayed inode
3782 * in delayed_nodes_tree.
3784 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3785 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3786 &BTRFS_I(inode)->runtime_flags);
3789 * We don't persist the id of the transaction where an unlink operation
3790 * against the inode was last made. So here we assume the inode might
3791 * have been evicted, and therefore the exact value of last_unlink_trans
3792 * lost, and set it to last_trans to avoid metadata inconsistencies
3793 * between the inode and its parent if the inode is fsync'ed and the log
3794 * replayed. For example, in the scenario:
3797 * ln mydir/foo mydir/bar
3800 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3801 * xfs_io -c fsync mydir/foo
3803 * mount fs, triggers fsync log replay
3805 * We must make sure that when we fsync our inode foo we also log its
3806 * parent inode, otherwise after log replay the parent still has the
3807 * dentry with the "bar" name but our inode foo has a link count of 1
3808 * and doesn't have an inode ref with the name "bar" anymore.
3810 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3811 * but it guarantees correctness at the expense of occasional full
3812 * transaction commits on fsync if our inode is a directory, or if our
3813 * inode is not a directory, logging its parent unnecessarily.
3815 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3818 if (inode->i_nlink != 1 ||
3819 path->slots[0] >= btrfs_header_nritems(leaf))
3822 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3823 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3826 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3827 if (location.type == BTRFS_INODE_REF_KEY) {
3828 struct btrfs_inode_ref *ref;
3830 ref = (struct btrfs_inode_ref *)ptr;
3831 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3832 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3833 struct btrfs_inode_extref *extref;
3835 extref = (struct btrfs_inode_extref *)ptr;
3836 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3841 * try to precache a NULL acl entry for files that don't have
3842 * any xattrs or acls
3844 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3845 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3846 if (first_xattr_slot != -1) {
3847 path->slots[0] = first_xattr_slot;
3848 ret = btrfs_load_inode_props(inode, path);
3851 "error loading props for ino %llu (root %llu): %d",
3852 btrfs_ino(BTRFS_I(inode)),
3853 root->root_key.objectid, ret);
3855 btrfs_free_path(path);
3858 cache_no_acl(inode);
3860 switch (inode->i_mode & S_IFMT) {
3862 inode->i_mapping->a_ops = &btrfs_aops;
3863 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3864 inode->i_fop = &btrfs_file_operations;
3865 inode->i_op = &btrfs_file_inode_operations;
3868 inode->i_fop = &btrfs_dir_file_operations;
3869 inode->i_op = &btrfs_dir_inode_operations;
3872 inode->i_op = &btrfs_symlink_inode_operations;
3873 inode_nohighmem(inode);
3874 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3877 inode->i_op = &btrfs_special_inode_operations;
3878 init_special_inode(inode, inode->i_mode, rdev);
3882 btrfs_update_iflags(inode);
3886 btrfs_free_path(path);
3887 make_bad_inode(inode);
3892 * given a leaf and an inode, copy the inode fields into the leaf
3894 static void fill_inode_item(struct btrfs_trans_handle *trans,
3895 struct extent_buffer *leaf,
3896 struct btrfs_inode_item *item,
3897 struct inode *inode)
3899 struct btrfs_map_token token;
3901 btrfs_init_map_token(&token);
3903 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3904 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3905 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3907 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3908 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3910 btrfs_set_token_timespec_sec(leaf, &item->atime,
3911 inode->i_atime.tv_sec, &token);
3912 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3913 inode->i_atime.tv_nsec, &token);
3915 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3916 inode->i_mtime.tv_sec, &token);
3917 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3918 inode->i_mtime.tv_nsec, &token);
3920 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3921 inode->i_ctime.tv_sec, &token);
3922 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3923 inode->i_ctime.tv_nsec, &token);
3925 btrfs_set_token_timespec_sec(leaf, &item->otime,
3926 BTRFS_I(inode)->i_otime.tv_sec, &token);
3927 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3928 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3930 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3932 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3934 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3935 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3936 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3937 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3938 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3942 * copy everything in the in-memory inode into the btree.
3944 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3945 struct btrfs_root *root, struct inode *inode)
3947 struct btrfs_inode_item *inode_item;
3948 struct btrfs_path *path;
3949 struct extent_buffer *leaf;
3952 path = btrfs_alloc_path();
3956 path->leave_spinning = 1;
3957 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3965 leaf = path->nodes[0];
3966 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3967 struct btrfs_inode_item);
3969 fill_inode_item(trans, leaf, inode_item, inode);
3970 btrfs_mark_buffer_dirty(leaf);
3971 btrfs_set_inode_last_trans(trans, inode);
3974 btrfs_free_path(path);
3979 * copy everything in the in-memory inode into the btree.
3981 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3982 struct btrfs_root *root, struct inode *inode)
3984 struct btrfs_fs_info *fs_info = root->fs_info;
3988 * If the inode is a free space inode, we can deadlock during commit
3989 * if we put it into the delayed code.
3991 * The data relocation inode should also be directly updated
3994 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3995 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3996 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3997 btrfs_update_root_times(trans, root);
3999 ret = btrfs_delayed_update_inode(trans, root, inode);
4001 btrfs_set_inode_last_trans(trans, inode);
4005 return btrfs_update_inode_item(trans, root, inode);
4008 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
4009 struct btrfs_root *root,
4010 struct inode *inode)
4014 ret = btrfs_update_inode(trans, root, inode);
4016 return btrfs_update_inode_item(trans, root, inode);
4021 * unlink helper that gets used here in inode.c and in the tree logging
4022 * recovery code. It remove a link in a directory with a given name, and
4023 * also drops the back refs in the inode to the directory
4025 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4026 struct btrfs_root *root,
4027 struct btrfs_inode *dir,
4028 struct btrfs_inode *inode,
4029 const char *name, int name_len)
4031 struct btrfs_fs_info *fs_info = root->fs_info;
4032 struct btrfs_path *path;
4034 struct extent_buffer *leaf;
4035 struct btrfs_dir_item *di;
4036 struct btrfs_key key;
4038 u64 ino = btrfs_ino(inode);
4039 u64 dir_ino = btrfs_ino(dir);
4041 path = btrfs_alloc_path();
4047 path->leave_spinning = 1;
4048 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4049 name, name_len, -1);
4058 leaf = path->nodes[0];
4059 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4060 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4063 btrfs_release_path(path);
4066 * If we don't have dir index, we have to get it by looking up
4067 * the inode ref, since we get the inode ref, remove it directly,
4068 * it is unnecessary to do delayed deletion.
4070 * But if we have dir index, needn't search inode ref to get it.
4071 * Since the inode ref is close to the inode item, it is better
4072 * that we delay to delete it, and just do this deletion when
4073 * we update the inode item.
4075 if (inode->dir_index) {
4076 ret = btrfs_delayed_delete_inode_ref(inode);
4078 index = inode->dir_index;
4083 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
4087 "failed to delete reference to %.*s, inode %llu parent %llu",
4088 name_len, name, ino, dir_ino);
4089 btrfs_abort_transaction(trans, ret);
4093 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
4095 btrfs_abort_transaction(trans, ret);
4099 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
4101 if (ret != 0 && ret != -ENOENT) {
4102 btrfs_abort_transaction(trans, ret);
4106 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4111 btrfs_abort_transaction(trans, ret);
4113 btrfs_free_path(path);
4117 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4118 inode_inc_iversion(&inode->vfs_inode);
4119 inode_inc_iversion(&dir->vfs_inode);
4120 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4121 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4122 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4127 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4128 struct btrfs_root *root,
4129 struct btrfs_inode *dir, struct btrfs_inode *inode,
4130 const char *name, int name_len)
4133 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4135 drop_nlink(&inode->vfs_inode);
4136 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4142 * helper to start transaction for unlink and rmdir.
4144 * unlink and rmdir are special in btrfs, they do not always free space, so
4145 * if we cannot make our reservations the normal way try and see if there is
4146 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4147 * allow the unlink to occur.
4149 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4151 struct btrfs_root *root = BTRFS_I(dir)->root;
4154 * 1 for the possible orphan item
4155 * 1 for the dir item
4156 * 1 for the dir index
4157 * 1 for the inode ref
4160 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4163 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4165 struct btrfs_root *root = BTRFS_I(dir)->root;
4166 struct btrfs_trans_handle *trans;
4167 struct inode *inode = d_inode(dentry);
4170 trans = __unlink_start_trans(dir);
4172 return PTR_ERR(trans);
4174 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4177 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4178 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4179 dentry->d_name.len);
4183 if (inode->i_nlink == 0) {
4184 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4190 btrfs_end_transaction(trans);
4191 btrfs_btree_balance_dirty(root->fs_info);
4195 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4196 struct btrfs_root *root,
4197 struct inode *dir, u64 objectid,
4198 const char *name, int name_len)
4200 struct btrfs_fs_info *fs_info = root->fs_info;
4201 struct btrfs_path *path;
4202 struct extent_buffer *leaf;
4203 struct btrfs_dir_item *di;
4204 struct btrfs_key key;
4207 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4209 path = btrfs_alloc_path();
4213 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4214 name, name_len, -1);
4215 if (IS_ERR_OR_NULL(di)) {
4223 leaf = path->nodes[0];
4224 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4225 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4226 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4228 btrfs_abort_transaction(trans, ret);
4231 btrfs_release_path(path);
4233 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4234 root->root_key.objectid, dir_ino,
4235 &index, name, name_len);
4237 if (ret != -ENOENT) {
4238 btrfs_abort_transaction(trans, ret);
4241 di = btrfs_search_dir_index_item(root, path, dir_ino,
4243 if (IS_ERR_OR_NULL(di)) {
4248 btrfs_abort_transaction(trans, ret);
4252 leaf = path->nodes[0];
4253 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4254 btrfs_release_path(path);
4257 btrfs_release_path(path);
4259 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4261 btrfs_abort_transaction(trans, ret);
4265 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4266 inode_inc_iversion(dir);
4267 dir->i_mtime = dir->i_ctime = current_time(dir);
4268 ret = btrfs_update_inode_fallback(trans, root, dir);
4270 btrfs_abort_transaction(trans, ret);
4272 btrfs_free_path(path);
4276 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4278 struct inode *inode = d_inode(dentry);
4280 struct btrfs_root *root = BTRFS_I(dir)->root;
4281 struct btrfs_trans_handle *trans;
4282 u64 last_unlink_trans;
4284 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4286 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4289 trans = __unlink_start_trans(dir);
4291 return PTR_ERR(trans);
4293 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4294 err = btrfs_unlink_subvol(trans, root, dir,
4295 BTRFS_I(inode)->location.objectid,
4296 dentry->d_name.name,
4297 dentry->d_name.len);
4301 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4305 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4307 /* now the directory is empty */
4308 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4309 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4310 dentry->d_name.len);
4312 btrfs_i_size_write(BTRFS_I(inode), 0);
4314 * Propagate the last_unlink_trans value of the deleted dir to
4315 * its parent directory. This is to prevent an unrecoverable
4316 * log tree in the case we do something like this:
4318 * 2) create snapshot under dir foo
4319 * 3) delete the snapshot
4322 * 6) fsync foo or some file inside foo
4324 if (last_unlink_trans >= trans->transid)
4325 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4328 btrfs_end_transaction(trans);
4329 btrfs_btree_balance_dirty(root->fs_info);
4334 static int truncate_space_check(struct btrfs_trans_handle *trans,
4335 struct btrfs_root *root,
4338 struct btrfs_fs_info *fs_info = root->fs_info;
4342 * This is only used to apply pressure to the enospc system, we don't
4343 * intend to use this reservation at all.
4345 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4346 bytes_deleted *= fs_info->nodesize;
4347 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4348 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4350 trace_btrfs_space_reservation(fs_info, "transaction",
4353 trans->bytes_reserved += bytes_deleted;
4359 static int truncate_inline_extent(struct inode *inode,
4360 struct btrfs_path *path,
4361 struct btrfs_key *found_key,
4365 struct extent_buffer *leaf = path->nodes[0];
4366 int slot = path->slots[0];
4367 struct btrfs_file_extent_item *fi;
4368 u32 size = (u32)(new_size - found_key->offset);
4369 struct btrfs_root *root = BTRFS_I(inode)->root;
4371 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4373 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4374 loff_t offset = new_size;
4375 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4378 * Zero out the remaining of the last page of our inline extent,
4379 * instead of directly truncating our inline extent here - that
4380 * would be much more complex (decompressing all the data, then
4381 * compressing the truncated data, which might be bigger than
4382 * the size of the inline extent, resize the extent, etc).
4383 * We release the path because to get the page we might need to
4384 * read the extent item from disk (data not in the page cache).
4386 btrfs_release_path(path);
4387 return btrfs_truncate_block(inode, offset, page_end - offset,
4391 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4392 size = btrfs_file_extent_calc_inline_size(size);
4393 btrfs_truncate_item(root->fs_info, path, size, 1);
4395 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4396 inode_sub_bytes(inode, item_end + 1 - new_size);
4402 * this can truncate away extent items, csum items and directory items.
4403 * It starts at a high offset and removes keys until it can't find
4404 * any higher than new_size
4406 * csum items that cross the new i_size are truncated to the new size
4409 * min_type is the minimum key type to truncate down to. If set to 0, this
4410 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4412 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4413 struct btrfs_root *root,
4414 struct inode *inode,
4415 u64 new_size, u32 min_type)
4417 struct btrfs_fs_info *fs_info = root->fs_info;
4418 struct btrfs_path *path;
4419 struct extent_buffer *leaf;
4420 struct btrfs_file_extent_item *fi;
4421 struct btrfs_key key;
4422 struct btrfs_key found_key;
4423 u64 extent_start = 0;
4424 u64 extent_num_bytes = 0;
4425 u64 extent_offset = 0;
4427 u64 last_size = new_size;
4428 u32 found_type = (u8)-1;
4431 int pending_del_nr = 0;
4432 int pending_del_slot = 0;
4433 int extent_type = -1;
4436 u64 ino = btrfs_ino(BTRFS_I(inode));
4437 u64 bytes_deleted = 0;
4439 bool should_throttle = 0;
4440 bool should_end = 0;
4442 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4445 * for non-free space inodes and ref cows, we want to back off from
4448 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4449 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4452 path = btrfs_alloc_path();
4455 path->reada = READA_BACK;
4458 * We want to drop from the next block forward in case this new size is
4459 * not block aligned since we will be keeping the last block of the
4460 * extent just the way it is.
4462 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4463 root == fs_info->tree_root)
4464 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4465 fs_info->sectorsize),
4469 * This function is also used to drop the items in the log tree before
4470 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4471 * it is used to drop the loged items. So we shouldn't kill the delayed
4474 if (min_type == 0 && root == BTRFS_I(inode)->root)
4475 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4478 key.offset = (u64)-1;
4483 * with a 16K leaf size and 128MB extents, you can actually queue
4484 * up a huge file in a single leaf. Most of the time that
4485 * bytes_deleted is > 0, it will be huge by the time we get here
4487 if (be_nice && bytes_deleted > SZ_32M) {
4488 if (btrfs_should_end_transaction(trans)) {
4495 path->leave_spinning = 1;
4496 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4503 /* there are no items in the tree for us to truncate, we're
4506 if (path->slots[0] == 0)
4513 leaf = path->nodes[0];
4514 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4515 found_type = found_key.type;
4517 if (found_key.objectid != ino)
4520 if (found_type < min_type)
4523 item_end = found_key.offset;
4524 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4525 fi = btrfs_item_ptr(leaf, path->slots[0],
4526 struct btrfs_file_extent_item);
4527 extent_type = btrfs_file_extent_type(leaf, fi);
4528 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4530 btrfs_file_extent_num_bytes(leaf, fi);
4532 trace_btrfs_truncate_show_fi_regular(
4533 BTRFS_I(inode), leaf, fi,
4535 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4536 item_end += btrfs_file_extent_inline_len(leaf,
4537 path->slots[0], fi);
4539 trace_btrfs_truncate_show_fi_inline(
4540 BTRFS_I(inode), leaf, fi, path->slots[0],
4545 if (found_type > min_type) {
4548 if (item_end < new_size)
4550 if (found_key.offset >= new_size)
4556 /* FIXME, shrink the extent if the ref count is only 1 */
4557 if (found_type != BTRFS_EXTENT_DATA_KEY)
4561 last_size = found_key.offset;
4563 last_size = new_size;
4565 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4567 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4569 u64 orig_num_bytes =
4570 btrfs_file_extent_num_bytes(leaf, fi);
4571 extent_num_bytes = ALIGN(new_size -
4573 fs_info->sectorsize);
4574 btrfs_set_file_extent_num_bytes(leaf, fi,
4576 num_dec = (orig_num_bytes -
4578 if (test_bit(BTRFS_ROOT_REF_COWS,
4581 inode_sub_bytes(inode, num_dec);
4582 btrfs_mark_buffer_dirty(leaf);
4585 btrfs_file_extent_disk_num_bytes(leaf,
4587 extent_offset = found_key.offset -
4588 btrfs_file_extent_offset(leaf, fi);
4590 /* FIXME blocksize != 4096 */
4591 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4592 if (extent_start != 0) {
4594 if (test_bit(BTRFS_ROOT_REF_COWS,
4596 inode_sub_bytes(inode, num_dec);
4599 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4601 * we can't truncate inline items that have had
4605 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4606 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4609 * Need to release path in order to truncate a
4610 * compressed extent. So delete any accumulated
4611 * extent items so far.
4613 if (btrfs_file_extent_compression(leaf, fi) !=
4614 BTRFS_COMPRESS_NONE && pending_del_nr) {
4615 err = btrfs_del_items(trans, root, path,
4619 btrfs_abort_transaction(trans,
4626 err = truncate_inline_extent(inode, path,
4631 btrfs_abort_transaction(trans, err);
4634 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4636 inode_sub_bytes(inode, item_end + 1 - new_size);
4641 if (!pending_del_nr) {
4642 /* no pending yet, add ourselves */
4643 pending_del_slot = path->slots[0];
4645 } else if (pending_del_nr &&
4646 path->slots[0] + 1 == pending_del_slot) {
4647 /* hop on the pending chunk */
4649 pending_del_slot = path->slots[0];
4656 should_throttle = 0;
4659 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4660 root == fs_info->tree_root)) {
4661 btrfs_set_path_blocking(path);
4662 bytes_deleted += extent_num_bytes;
4663 ret = btrfs_free_extent(trans, fs_info, extent_start,
4664 extent_num_bytes, 0,
4665 btrfs_header_owner(leaf),
4666 ino, extent_offset);
4668 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4669 btrfs_async_run_delayed_refs(fs_info,
4670 trans->delayed_ref_updates * 2,
4673 if (truncate_space_check(trans, root,
4674 extent_num_bytes)) {
4677 if (btrfs_should_throttle_delayed_refs(trans,
4679 should_throttle = 1;
4683 if (found_type == BTRFS_INODE_ITEM_KEY)
4686 if (path->slots[0] == 0 ||
4687 path->slots[0] != pending_del_slot ||
4688 should_throttle || should_end) {
4689 if (pending_del_nr) {
4690 ret = btrfs_del_items(trans, root, path,
4694 btrfs_abort_transaction(trans, ret);
4699 btrfs_release_path(path);
4700 if (should_throttle) {
4701 unsigned long updates = trans->delayed_ref_updates;
4703 trans->delayed_ref_updates = 0;
4704 ret = btrfs_run_delayed_refs(trans,
4712 * if we failed to refill our space rsv, bail out
4713 * and let the transaction restart
4725 if (pending_del_nr) {
4726 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4729 btrfs_abort_transaction(trans, ret);
4732 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4733 ASSERT(last_size >= new_size);
4734 if (!err && last_size > new_size)
4735 last_size = new_size;
4736 btrfs_ordered_update_i_size(inode, last_size, NULL);
4739 btrfs_free_path(path);
4741 if (be_nice && bytes_deleted > SZ_32M) {
4742 unsigned long updates = trans->delayed_ref_updates;
4744 trans->delayed_ref_updates = 0;
4745 ret = btrfs_run_delayed_refs(trans, fs_info,
4755 * btrfs_truncate_block - read, zero a chunk and write a block
4756 * @inode - inode that we're zeroing
4757 * @from - the offset to start zeroing
4758 * @len - the length to zero, 0 to zero the entire range respective to the
4760 * @front - zero up to the offset instead of from the offset on
4762 * This will find the block for the "from" offset and cow the block and zero the
4763 * part we want to zero. This is used with truncate and hole punching.
4765 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4768 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4769 struct address_space *mapping = inode->i_mapping;
4770 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4771 struct btrfs_ordered_extent *ordered;
4772 struct extent_state *cached_state = NULL;
4774 u32 blocksize = fs_info->sectorsize;
4775 pgoff_t index = from >> PAGE_SHIFT;
4776 unsigned offset = from & (blocksize - 1);
4778 gfp_t mask = btrfs_alloc_write_mask(mapping);
4783 if ((offset & (blocksize - 1)) == 0 &&
4784 (!len || ((len & (blocksize - 1)) == 0)))
4787 ret = btrfs_delalloc_reserve_space(inode,
4788 round_down(from, blocksize), blocksize);
4793 page = find_or_create_page(mapping, index, mask);
4795 btrfs_delalloc_release_space(inode,
4796 round_down(from, blocksize),
4802 block_start = round_down(from, blocksize);
4803 block_end = block_start + blocksize - 1;
4805 if (!PageUptodate(page)) {
4806 ret = btrfs_readpage(NULL, page);
4808 if (page->mapping != mapping) {
4813 if (!PageUptodate(page)) {
4818 wait_on_page_writeback(page);
4820 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4821 set_page_extent_mapped(page);
4823 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4825 unlock_extent_cached(io_tree, block_start, block_end,
4826 &cached_state, GFP_NOFS);
4829 btrfs_start_ordered_extent(inode, ordered, 1);
4830 btrfs_put_ordered_extent(ordered);
4834 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4835 EXTENT_DIRTY | EXTENT_DELALLOC |
4836 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4837 0, 0, &cached_state, GFP_NOFS);
4839 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4842 unlock_extent_cached(io_tree, block_start, block_end,
4843 &cached_state, GFP_NOFS);
4847 if (offset != blocksize) {
4849 len = blocksize - offset;
4852 memset(kaddr + (block_start - page_offset(page)),
4855 memset(kaddr + (block_start - page_offset(page)) + offset,
4857 flush_dcache_page(page);
4860 ClearPageChecked(page);
4861 set_page_dirty(page);
4862 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4867 btrfs_delalloc_release_space(inode, block_start,
4875 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4876 u64 offset, u64 len)
4878 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4879 struct btrfs_trans_handle *trans;
4883 * Still need to make sure the inode looks like it's been updated so
4884 * that any holes get logged if we fsync.
4886 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4887 BTRFS_I(inode)->last_trans = fs_info->generation;
4888 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4889 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4894 * 1 - for the one we're dropping
4895 * 1 - for the one we're adding
4896 * 1 - for updating the inode.
4898 trans = btrfs_start_transaction(root, 3);
4900 return PTR_ERR(trans);
4902 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4904 btrfs_abort_transaction(trans, ret);
4905 btrfs_end_transaction(trans);
4909 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4910 offset, 0, 0, len, 0, len, 0, 0, 0);
4912 btrfs_abort_transaction(trans, ret);
4914 btrfs_update_inode(trans, root, inode);
4915 btrfs_end_transaction(trans);
4920 * This function puts in dummy file extents for the area we're creating a hole
4921 * for. So if we are truncating this file to a larger size we need to insert
4922 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4923 * the range between oldsize and size
4925 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4927 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4928 struct btrfs_root *root = BTRFS_I(inode)->root;
4929 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4930 struct extent_map *em = NULL;
4931 struct extent_state *cached_state = NULL;
4932 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4933 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4934 u64 block_end = ALIGN(size, fs_info->sectorsize);
4941 * If our size started in the middle of a block we need to zero out the
4942 * rest of the block before we expand the i_size, otherwise we could
4943 * expose stale data.
4945 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4949 if (size <= hole_start)
4953 struct btrfs_ordered_extent *ordered;
4955 lock_extent_bits(io_tree, hole_start, block_end - 1,
4957 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
4958 block_end - hole_start);
4961 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4962 &cached_state, GFP_NOFS);
4963 btrfs_start_ordered_extent(inode, ordered, 1);
4964 btrfs_put_ordered_extent(ordered);
4967 cur_offset = hole_start;
4969 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
4970 block_end - cur_offset, 0);
4976 last_byte = min(extent_map_end(em), block_end);
4977 last_byte = ALIGN(last_byte, fs_info->sectorsize);
4978 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4979 struct extent_map *hole_em;
4980 hole_size = last_byte - cur_offset;
4982 err = maybe_insert_hole(root, inode, cur_offset,
4986 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
4987 cur_offset + hole_size - 1, 0);
4988 hole_em = alloc_extent_map();
4990 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4991 &BTRFS_I(inode)->runtime_flags);
4994 hole_em->start = cur_offset;
4995 hole_em->len = hole_size;
4996 hole_em->orig_start = cur_offset;
4998 hole_em->block_start = EXTENT_MAP_HOLE;
4999 hole_em->block_len = 0;
5000 hole_em->orig_block_len = 0;
5001 hole_em->ram_bytes = hole_size;
5002 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5003 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5004 hole_em->generation = fs_info->generation;
5007 write_lock(&em_tree->lock);
5008 err = add_extent_mapping(em_tree, hole_em, 1);
5009 write_unlock(&em_tree->lock);
5012 btrfs_drop_extent_cache(BTRFS_I(inode),
5017 free_extent_map(hole_em);
5020 free_extent_map(em);
5022 cur_offset = last_byte;
5023 if (cur_offset >= block_end)
5026 free_extent_map(em);
5027 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
5032 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5035 struct btrfs_trans_handle *trans;
5036 loff_t oldsize = i_size_read(inode);
5037 loff_t newsize = attr->ia_size;
5038 int mask = attr->ia_valid;
5042 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5043 * special case where we need to update the times despite not having
5044 * these flags set. For all other operations the VFS set these flags
5045 * explicitly if it wants a timestamp update.
5047 if (newsize != oldsize) {
5048 inode_inc_iversion(inode);
5049 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5050 inode->i_ctime = inode->i_mtime =
5051 current_time(inode);
5054 if (newsize > oldsize) {
5056 * Don't do an expanding truncate while snapshoting is ongoing.
5057 * This is to ensure the snapshot captures a fully consistent
5058 * state of this file - if the snapshot captures this expanding
5059 * truncation, it must capture all writes that happened before
5062 btrfs_wait_for_snapshot_creation(root);
5063 ret = btrfs_cont_expand(inode, oldsize, newsize);
5065 btrfs_end_write_no_snapshoting(root);
5069 trans = btrfs_start_transaction(root, 1);
5070 if (IS_ERR(trans)) {
5071 btrfs_end_write_no_snapshoting(root);
5072 return PTR_ERR(trans);
5075 i_size_write(inode, newsize);
5076 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5077 pagecache_isize_extended(inode, oldsize, newsize);
5078 ret = btrfs_update_inode(trans, root, inode);
5079 btrfs_end_write_no_snapshoting(root);
5080 btrfs_end_transaction(trans);
5084 * We're truncating a file that used to have good data down to
5085 * zero. Make sure it gets into the ordered flush list so that
5086 * any new writes get down to disk quickly.
5089 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5090 &BTRFS_I(inode)->runtime_flags);
5093 * 1 for the orphan item we're going to add
5094 * 1 for the orphan item deletion.
5096 trans = btrfs_start_transaction(root, 2);
5098 return PTR_ERR(trans);
5101 * We need to do this in case we fail at _any_ point during the
5102 * actual truncate. Once we do the truncate_setsize we could
5103 * invalidate pages which forces any outstanding ordered io to
5104 * be instantly completed which will give us extents that need
5105 * to be truncated. If we fail to get an orphan inode down we
5106 * could have left over extents that were never meant to live,
5107 * so we need to guarantee from this point on that everything
5108 * will be consistent.
5110 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
5111 btrfs_end_transaction(trans);
5115 /* we don't support swapfiles, so vmtruncate shouldn't fail */
5116 truncate_setsize(inode, newsize);
5118 /* Disable nonlocked read DIO to avoid the end less truncate */
5119 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5120 inode_dio_wait(inode);
5121 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5123 ret = btrfs_truncate(inode);
5124 if (ret && inode->i_nlink) {
5127 /* To get a stable disk_i_size */
5128 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5130 btrfs_orphan_del(NULL, BTRFS_I(inode));
5135 * failed to truncate, disk_i_size is only adjusted down
5136 * as we remove extents, so it should represent the true
5137 * size of the inode, so reset the in memory size and
5138 * delete our orphan entry.
5140 trans = btrfs_join_transaction(root);
5141 if (IS_ERR(trans)) {
5142 btrfs_orphan_del(NULL, BTRFS_I(inode));
5145 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5146 err = btrfs_orphan_del(trans, BTRFS_I(inode));
5148 btrfs_abort_transaction(trans, err);
5149 btrfs_end_transaction(trans);
5156 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5158 struct inode *inode = d_inode(dentry);
5159 struct btrfs_root *root = BTRFS_I(inode)->root;
5162 if (btrfs_root_readonly(root))
5165 err = setattr_prepare(dentry, attr);
5169 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5170 err = btrfs_setsize(inode, attr);
5175 if (attr->ia_valid) {
5176 setattr_copy(inode, attr);
5177 inode_inc_iversion(inode);
5178 err = btrfs_dirty_inode(inode);
5180 if (!err && attr->ia_valid & ATTR_MODE)
5181 err = posix_acl_chmod(inode, inode->i_mode);
5188 * While truncating the inode pages during eviction, we get the VFS calling
5189 * btrfs_invalidatepage() against each page of the inode. This is slow because
5190 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5191 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5192 * extent_state structures over and over, wasting lots of time.
5194 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5195 * those expensive operations on a per page basis and do only the ordered io
5196 * finishing, while we release here the extent_map and extent_state structures,
5197 * without the excessive merging and splitting.
5199 static void evict_inode_truncate_pages(struct inode *inode)
5201 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5202 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5203 struct rb_node *node;
5205 ASSERT(inode->i_state & I_FREEING);
5206 truncate_inode_pages_final(&inode->i_data);
5208 write_lock(&map_tree->lock);
5209 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5210 struct extent_map *em;
5212 node = rb_first(&map_tree->map);
5213 em = rb_entry(node, struct extent_map, rb_node);
5214 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5215 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5216 remove_extent_mapping(map_tree, em);
5217 free_extent_map(em);
5218 if (need_resched()) {
5219 write_unlock(&map_tree->lock);
5221 write_lock(&map_tree->lock);
5224 write_unlock(&map_tree->lock);
5227 * Keep looping until we have no more ranges in the io tree.
5228 * We can have ongoing bios started by readpages (called from readahead)
5229 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5230 * still in progress (unlocked the pages in the bio but did not yet
5231 * unlocked the ranges in the io tree). Therefore this means some
5232 * ranges can still be locked and eviction started because before
5233 * submitting those bios, which are executed by a separate task (work
5234 * queue kthread), inode references (inode->i_count) were not taken
5235 * (which would be dropped in the end io callback of each bio).
5236 * Therefore here we effectively end up waiting for those bios and
5237 * anyone else holding locked ranges without having bumped the inode's
5238 * reference count - if we don't do it, when they access the inode's
5239 * io_tree to unlock a range it may be too late, leading to an
5240 * use-after-free issue.
5242 spin_lock(&io_tree->lock);
5243 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5244 struct extent_state *state;
5245 struct extent_state *cached_state = NULL;
5249 node = rb_first(&io_tree->state);
5250 state = rb_entry(node, struct extent_state, rb_node);
5251 start = state->start;
5253 spin_unlock(&io_tree->lock);
5255 lock_extent_bits(io_tree, start, end, &cached_state);
5258 * If still has DELALLOC flag, the extent didn't reach disk,
5259 * and its reserved space won't be freed by delayed_ref.
5260 * So we need to free its reserved space here.
5261 * (Refer to comment in btrfs_invalidatepage, case 2)
5263 * Note, end is the bytenr of last byte, so we need + 1 here.
5265 if (state->state & EXTENT_DELALLOC)
5266 btrfs_qgroup_free_data(inode, start, end - start + 1);
5268 clear_extent_bit(io_tree, start, end,
5269 EXTENT_LOCKED | EXTENT_DIRTY |
5270 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5271 EXTENT_DEFRAG, 1, 1,
5272 &cached_state, GFP_NOFS);
5275 spin_lock(&io_tree->lock);
5277 spin_unlock(&io_tree->lock);
5280 void btrfs_evict_inode(struct inode *inode)
5282 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5283 struct btrfs_trans_handle *trans;
5284 struct btrfs_root *root = BTRFS_I(inode)->root;
5285 struct btrfs_block_rsv *rsv, *global_rsv;
5286 int steal_from_global = 0;
5290 trace_btrfs_inode_evict(inode);
5293 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
5297 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5299 evict_inode_truncate_pages(inode);
5301 if (inode->i_nlink &&
5302 ((btrfs_root_refs(&root->root_item) != 0 &&
5303 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5304 btrfs_is_free_space_inode(BTRFS_I(inode))))
5307 if (is_bad_inode(inode)) {
5308 btrfs_orphan_del(NULL, BTRFS_I(inode));
5311 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5312 if (!special_file(inode->i_mode))
5313 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5315 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5317 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
5318 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5319 &BTRFS_I(inode)->runtime_flags));
5323 if (inode->i_nlink > 0) {
5324 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5325 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5329 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5331 btrfs_orphan_del(NULL, BTRFS_I(inode));
5335 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5337 btrfs_orphan_del(NULL, BTRFS_I(inode));
5340 rsv->size = min_size;
5342 global_rsv = &fs_info->global_block_rsv;
5344 btrfs_i_size_write(BTRFS_I(inode), 0);
5347 * This is a bit simpler than btrfs_truncate since we've already
5348 * reserved our space for our orphan item in the unlink, so we just
5349 * need to reserve some slack space in case we add bytes and update
5350 * inode item when doing the truncate.
5353 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5354 BTRFS_RESERVE_FLUSH_LIMIT);
5357 * Try and steal from the global reserve since we will
5358 * likely not use this space anyway, we want to try as
5359 * hard as possible to get this to work.
5362 steal_from_global++;
5364 steal_from_global = 0;
5368 * steal_from_global == 0: we reserved stuff, hooray!
5369 * steal_from_global == 1: we didn't reserve stuff, boo!
5370 * steal_from_global == 2: we've committed, still not a lot of
5371 * room but maybe we'll have room in the global reserve this
5373 * steal_from_global == 3: abandon all hope!
5375 if (steal_from_global > 2) {
5377 "Could not get space for a delete, will truncate on mount %d",
5379 btrfs_orphan_del(NULL, BTRFS_I(inode));
5380 btrfs_free_block_rsv(fs_info, rsv);
5384 trans = btrfs_join_transaction(root);
5385 if (IS_ERR(trans)) {
5386 btrfs_orphan_del(NULL, BTRFS_I(inode));
5387 btrfs_free_block_rsv(fs_info, rsv);
5392 * We can't just steal from the global reserve, we need to make
5393 * sure there is room to do it, if not we need to commit and try
5396 if (steal_from_global) {
5397 if (!btrfs_check_space_for_delayed_refs(trans, fs_info))
5398 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5405 * Couldn't steal from the global reserve, we have too much
5406 * pending stuff built up, commit the transaction and try it
5410 ret = btrfs_commit_transaction(trans);
5412 btrfs_orphan_del(NULL, BTRFS_I(inode));
5413 btrfs_free_block_rsv(fs_info, rsv);
5418 steal_from_global = 0;
5421 trans->block_rsv = rsv;
5423 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5424 if (ret != -ENOSPC && ret != -EAGAIN)
5427 trans->block_rsv = &fs_info->trans_block_rsv;
5428 btrfs_end_transaction(trans);
5430 btrfs_btree_balance_dirty(fs_info);
5433 btrfs_free_block_rsv(fs_info, rsv);
5436 * Errors here aren't a big deal, it just means we leave orphan items
5437 * in the tree. They will be cleaned up on the next mount.
5440 trans->block_rsv = root->orphan_block_rsv;
5441 btrfs_orphan_del(trans, BTRFS_I(inode));
5443 btrfs_orphan_del(NULL, BTRFS_I(inode));
5446 trans->block_rsv = &fs_info->trans_block_rsv;
5447 if (!(root == fs_info->tree_root ||
5448 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5449 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5451 btrfs_end_transaction(trans);
5452 btrfs_btree_balance_dirty(fs_info);
5454 btrfs_remove_delayed_node(BTRFS_I(inode));
5459 * this returns the key found in the dir entry in the location pointer.
5460 * If no dir entries were found, location->objectid is 0.
5462 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5463 struct btrfs_key *location)
5465 const char *name = dentry->d_name.name;
5466 int namelen = dentry->d_name.len;
5467 struct btrfs_dir_item *di;
5468 struct btrfs_path *path;
5469 struct btrfs_root *root = BTRFS_I(dir)->root;
5472 path = btrfs_alloc_path();
5476 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5481 if (IS_ERR_OR_NULL(di))
5484 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5486 btrfs_free_path(path);
5489 location->objectid = 0;
5494 * when we hit a tree root in a directory, the btrfs part of the inode
5495 * needs to be changed to reflect the root directory of the tree root. This
5496 * is kind of like crossing a mount point.
5498 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5500 struct dentry *dentry,
5501 struct btrfs_key *location,
5502 struct btrfs_root **sub_root)
5504 struct btrfs_path *path;
5505 struct btrfs_root *new_root;
5506 struct btrfs_root_ref *ref;
5507 struct extent_buffer *leaf;
5508 struct btrfs_key key;
5512 path = btrfs_alloc_path();
5519 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5520 key.type = BTRFS_ROOT_REF_KEY;
5521 key.offset = location->objectid;
5523 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5530 leaf = path->nodes[0];
5531 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5532 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5533 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5536 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5537 (unsigned long)(ref + 1),
5538 dentry->d_name.len);
5542 btrfs_release_path(path);
5544 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5545 if (IS_ERR(new_root)) {
5546 err = PTR_ERR(new_root);
5550 *sub_root = new_root;
5551 location->objectid = btrfs_root_dirid(&new_root->root_item);
5552 location->type = BTRFS_INODE_ITEM_KEY;
5553 location->offset = 0;
5556 btrfs_free_path(path);
5560 static void inode_tree_add(struct inode *inode)
5562 struct btrfs_root *root = BTRFS_I(inode)->root;
5563 struct btrfs_inode *entry;
5565 struct rb_node *parent;
5566 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5567 u64 ino = btrfs_ino(BTRFS_I(inode));
5569 if (inode_unhashed(inode))
5572 spin_lock(&root->inode_lock);
5573 p = &root->inode_tree.rb_node;
5576 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5578 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5579 p = &parent->rb_left;
5580 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5581 p = &parent->rb_right;
5583 WARN_ON(!(entry->vfs_inode.i_state &
5584 (I_WILL_FREE | I_FREEING)));
5585 rb_replace_node(parent, new, &root->inode_tree);
5586 RB_CLEAR_NODE(parent);
5587 spin_unlock(&root->inode_lock);
5591 rb_link_node(new, parent, p);
5592 rb_insert_color(new, &root->inode_tree);
5593 spin_unlock(&root->inode_lock);
5596 static void inode_tree_del(struct inode *inode)
5598 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5599 struct btrfs_root *root = BTRFS_I(inode)->root;
5602 spin_lock(&root->inode_lock);
5603 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5604 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5605 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5606 empty = RB_EMPTY_ROOT(&root->inode_tree);
5608 spin_unlock(&root->inode_lock);
5610 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5611 synchronize_srcu(&fs_info->subvol_srcu);
5612 spin_lock(&root->inode_lock);
5613 empty = RB_EMPTY_ROOT(&root->inode_tree);
5614 spin_unlock(&root->inode_lock);
5616 btrfs_add_dead_root(root);
5620 void btrfs_invalidate_inodes(struct btrfs_root *root)
5622 struct btrfs_fs_info *fs_info = root->fs_info;
5623 struct rb_node *node;
5624 struct rb_node *prev;
5625 struct btrfs_inode *entry;
5626 struct inode *inode;
5629 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
5630 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5632 spin_lock(&root->inode_lock);
5634 node = root->inode_tree.rb_node;
5638 entry = rb_entry(node, struct btrfs_inode, rb_node);
5640 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5641 node = node->rb_left;
5642 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5643 node = node->rb_right;
5649 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5650 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
5654 prev = rb_next(prev);
5658 entry = rb_entry(node, struct btrfs_inode, rb_node);
5659 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
5660 inode = igrab(&entry->vfs_inode);
5662 spin_unlock(&root->inode_lock);
5663 if (atomic_read(&inode->i_count) > 1)
5664 d_prune_aliases(inode);
5666 * btrfs_drop_inode will have it removed from
5667 * the inode cache when its usage count
5672 spin_lock(&root->inode_lock);
5676 if (cond_resched_lock(&root->inode_lock))
5679 node = rb_next(node);
5681 spin_unlock(&root->inode_lock);
5684 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5686 struct btrfs_iget_args *args = p;
5687 inode->i_ino = args->location->objectid;
5688 memcpy(&BTRFS_I(inode)->location, args->location,
5689 sizeof(*args->location));
5690 BTRFS_I(inode)->root = args->root;
5694 static int btrfs_find_actor(struct inode *inode, void *opaque)
5696 struct btrfs_iget_args *args = opaque;
5697 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5698 args->root == BTRFS_I(inode)->root;
5701 static struct inode *btrfs_iget_locked(struct super_block *s,
5702 struct btrfs_key *location,
5703 struct btrfs_root *root)
5705 struct inode *inode;
5706 struct btrfs_iget_args args;
5707 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5709 args.location = location;
5712 inode = iget5_locked(s, hashval, btrfs_find_actor,
5713 btrfs_init_locked_inode,
5718 /* Get an inode object given its location and corresponding root.
5719 * Returns in *is_new if the inode was read from disk
5721 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5722 struct btrfs_root *root, int *new)
5724 struct inode *inode;
5726 inode = btrfs_iget_locked(s, location, root);
5728 return ERR_PTR(-ENOMEM);
5730 if (inode->i_state & I_NEW) {
5733 ret = btrfs_read_locked_inode(inode);
5734 if (!is_bad_inode(inode)) {
5735 inode_tree_add(inode);
5736 unlock_new_inode(inode);
5740 unlock_new_inode(inode);
5743 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5750 static struct inode *new_simple_dir(struct super_block *s,
5751 struct btrfs_key *key,
5752 struct btrfs_root *root)
5754 struct inode *inode = new_inode(s);
5757 return ERR_PTR(-ENOMEM);
5759 BTRFS_I(inode)->root = root;
5760 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5761 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5763 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5764 inode->i_op = &btrfs_dir_ro_inode_operations;
5765 inode->i_opflags &= ~IOP_XATTR;
5766 inode->i_fop = &simple_dir_operations;
5767 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5768 inode->i_mtime = current_time(inode);
5769 inode->i_atime = inode->i_mtime;
5770 inode->i_ctime = inode->i_mtime;
5771 BTRFS_I(inode)->i_otime = inode->i_mtime;
5776 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5778 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5779 struct inode *inode;
5780 struct btrfs_root *root = BTRFS_I(dir)->root;
5781 struct btrfs_root *sub_root = root;
5782 struct btrfs_key location;
5786 if (dentry->d_name.len > BTRFS_NAME_LEN)
5787 return ERR_PTR(-ENAMETOOLONG);
5789 ret = btrfs_inode_by_name(dir, dentry, &location);
5791 return ERR_PTR(ret);
5793 if (location.objectid == 0)
5794 return ERR_PTR(-ENOENT);
5796 if (location.type == BTRFS_INODE_ITEM_KEY) {
5797 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5801 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5803 index = srcu_read_lock(&fs_info->subvol_srcu);
5804 ret = fixup_tree_root_location(fs_info, dir, dentry,
5805 &location, &sub_root);
5808 inode = ERR_PTR(ret);
5810 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5812 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5814 srcu_read_unlock(&fs_info->subvol_srcu, index);
5816 if (!IS_ERR(inode) && root != sub_root) {
5817 down_read(&fs_info->cleanup_work_sem);
5818 if (!(inode->i_sb->s_flags & MS_RDONLY))
5819 ret = btrfs_orphan_cleanup(sub_root);
5820 up_read(&fs_info->cleanup_work_sem);
5823 inode = ERR_PTR(ret);
5830 static int btrfs_dentry_delete(const struct dentry *dentry)
5832 struct btrfs_root *root;
5833 struct inode *inode = d_inode(dentry);
5835 if (!inode && !IS_ROOT(dentry))
5836 inode = d_inode(dentry->d_parent);
5839 root = BTRFS_I(inode)->root;
5840 if (btrfs_root_refs(&root->root_item) == 0)
5843 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5849 static void btrfs_dentry_release(struct dentry *dentry)
5851 kfree(dentry->d_fsdata);
5854 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5857 struct inode *inode;
5859 inode = btrfs_lookup_dentry(dir, dentry);
5860 if (IS_ERR(inode)) {
5861 if (PTR_ERR(inode) == -ENOENT)
5864 return ERR_CAST(inode);
5867 return d_splice_alias(inode, dentry);
5870 unsigned char btrfs_filetype_table[] = {
5871 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5874 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5876 struct inode *inode = file_inode(file);
5877 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5878 struct btrfs_root *root = BTRFS_I(inode)->root;
5879 struct btrfs_dir_item *di;
5880 struct btrfs_key key;
5881 struct btrfs_key found_key;
5882 struct btrfs_path *path;
5883 struct list_head ins_list;
5884 struct list_head del_list;
5886 struct extent_buffer *leaf;
5888 unsigned char d_type;
5894 struct btrfs_key location;
5896 if (!dir_emit_dots(file, ctx))
5899 path = btrfs_alloc_path();
5903 path->reada = READA_FORWARD;
5905 INIT_LIST_HEAD(&ins_list);
5906 INIT_LIST_HEAD(&del_list);
5907 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5909 key.type = BTRFS_DIR_INDEX_KEY;
5910 key.offset = ctx->pos;
5911 key.objectid = btrfs_ino(BTRFS_I(inode));
5913 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5918 leaf = path->nodes[0];
5919 slot = path->slots[0];
5920 if (slot >= btrfs_header_nritems(leaf)) {
5921 ret = btrfs_next_leaf(root, path);
5929 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5931 if (found_key.objectid != key.objectid)
5933 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5935 if (found_key.offset < ctx->pos)
5937 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5940 ctx->pos = found_key.offset;
5942 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5943 if (verify_dir_item(fs_info, leaf, di))
5946 name_len = btrfs_dir_name_len(leaf, di);
5947 if (name_len <= sizeof(tmp_name)) {
5948 name_ptr = tmp_name;
5950 name_ptr = kmalloc(name_len, GFP_KERNEL);
5956 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5959 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5960 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5962 over = !dir_emit(ctx, name_ptr, name_len, location.objectid,
5965 if (name_ptr != tmp_name)
5975 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5980 * Stop new entries from being returned after we return the last
5983 * New directory entries are assigned a strictly increasing
5984 * offset. This means that new entries created during readdir
5985 * are *guaranteed* to be seen in the future by that readdir.
5986 * This has broken buggy programs which operate on names as
5987 * they're returned by readdir. Until we re-use freed offsets
5988 * we have this hack to stop new entries from being returned
5989 * under the assumption that they'll never reach this huge
5992 * This is being careful not to overflow 32bit loff_t unless the
5993 * last entry requires it because doing so has broken 32bit apps
5996 if (ctx->pos >= INT_MAX)
5997 ctx->pos = LLONG_MAX;
6004 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6005 btrfs_free_path(path);
6009 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6011 struct btrfs_root *root = BTRFS_I(inode)->root;
6012 struct btrfs_trans_handle *trans;
6014 bool nolock = false;
6016 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6019 if (btrfs_fs_closing(root->fs_info) &&
6020 btrfs_is_free_space_inode(BTRFS_I(inode)))
6023 if (wbc->sync_mode == WB_SYNC_ALL) {
6025 trans = btrfs_join_transaction_nolock(root);
6027 trans = btrfs_join_transaction(root);
6029 return PTR_ERR(trans);
6030 ret = btrfs_commit_transaction(trans);
6036 * This is somewhat expensive, updating the tree every time the
6037 * inode changes. But, it is most likely to find the inode in cache.
6038 * FIXME, needs more benchmarking...there are no reasons other than performance
6039 * to keep or drop this code.
6041 static int btrfs_dirty_inode(struct inode *inode)
6043 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6044 struct btrfs_root *root = BTRFS_I(inode)->root;
6045 struct btrfs_trans_handle *trans;
6048 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6051 trans = btrfs_join_transaction(root);
6053 return PTR_ERR(trans);
6055 ret = btrfs_update_inode(trans, root, inode);
6056 if (ret && ret == -ENOSPC) {
6057 /* whoops, lets try again with the full transaction */
6058 btrfs_end_transaction(trans);
6059 trans = btrfs_start_transaction(root, 1);
6061 return PTR_ERR(trans);
6063 ret = btrfs_update_inode(trans, root, inode);
6065 btrfs_end_transaction(trans);
6066 if (BTRFS_I(inode)->delayed_node)
6067 btrfs_balance_delayed_items(fs_info);
6073 * This is a copy of file_update_time. We need this so we can return error on
6074 * ENOSPC for updating the inode in the case of file write and mmap writes.
6076 static int btrfs_update_time(struct inode *inode, struct timespec *now,
6079 struct btrfs_root *root = BTRFS_I(inode)->root;
6081 if (btrfs_root_readonly(root))
6084 if (flags & S_VERSION)
6085 inode_inc_iversion(inode);
6086 if (flags & S_CTIME)
6087 inode->i_ctime = *now;
6088 if (flags & S_MTIME)
6089 inode->i_mtime = *now;
6090 if (flags & S_ATIME)
6091 inode->i_atime = *now;
6092 return btrfs_dirty_inode(inode);
6096 * find the highest existing sequence number in a directory
6097 * and then set the in-memory index_cnt variable to reflect
6098 * free sequence numbers
6100 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6102 struct btrfs_root *root = inode->root;
6103 struct btrfs_key key, found_key;
6104 struct btrfs_path *path;
6105 struct extent_buffer *leaf;
6108 key.objectid = btrfs_ino(inode);
6109 key.type = BTRFS_DIR_INDEX_KEY;
6110 key.offset = (u64)-1;
6112 path = btrfs_alloc_path();
6116 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6119 /* FIXME: we should be able to handle this */
6125 * MAGIC NUMBER EXPLANATION:
6126 * since we search a directory based on f_pos we have to start at 2
6127 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6128 * else has to start at 2
6130 if (path->slots[0] == 0) {
6131 inode->index_cnt = 2;
6137 leaf = path->nodes[0];
6138 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6140 if (found_key.objectid != btrfs_ino(inode) ||
6141 found_key.type != BTRFS_DIR_INDEX_KEY) {
6142 inode->index_cnt = 2;
6146 inode->index_cnt = found_key.offset + 1;
6148 btrfs_free_path(path);
6153 * helper to find a free sequence number in a given directory. This current
6154 * code is very simple, later versions will do smarter things in the btree
6156 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6160 if (dir->index_cnt == (u64)-1) {
6161 ret = btrfs_inode_delayed_dir_index_count(dir);
6163 ret = btrfs_set_inode_index_count(dir);
6169 *index = dir->index_cnt;
6175 static int btrfs_insert_inode_locked(struct inode *inode)
6177 struct btrfs_iget_args args;
6178 args.location = &BTRFS_I(inode)->location;
6179 args.root = BTRFS_I(inode)->root;
6181 return insert_inode_locked4(inode,
6182 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6183 btrfs_find_actor, &args);
6186 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root,
6189 const char *name, int name_len,
6190 u64 ref_objectid, u64 objectid,
6191 umode_t mode, u64 *index)
6193 struct btrfs_fs_info *fs_info = root->fs_info;
6194 struct inode *inode;
6195 struct btrfs_inode_item *inode_item;
6196 struct btrfs_key *location;
6197 struct btrfs_path *path;
6198 struct btrfs_inode_ref *ref;
6199 struct btrfs_key key[2];
6201 int nitems = name ? 2 : 1;
6205 path = btrfs_alloc_path();
6207 return ERR_PTR(-ENOMEM);
6209 inode = new_inode(fs_info->sb);
6211 btrfs_free_path(path);
6212 return ERR_PTR(-ENOMEM);
6216 * O_TMPFILE, set link count to 0, so that after this point,
6217 * we fill in an inode item with the correct link count.
6220 set_nlink(inode, 0);
6223 * we have to initialize this early, so we can reclaim the inode
6224 * number if we fail afterwards in this function.
6226 inode->i_ino = objectid;
6229 trace_btrfs_inode_request(dir);
6231 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6233 btrfs_free_path(path);
6235 return ERR_PTR(ret);
6241 * index_cnt is ignored for everything but a dir,
6242 * btrfs_get_inode_index_count has an explanation for the magic
6245 BTRFS_I(inode)->index_cnt = 2;
6246 BTRFS_I(inode)->dir_index = *index;
6247 BTRFS_I(inode)->root = root;
6248 BTRFS_I(inode)->generation = trans->transid;
6249 inode->i_generation = BTRFS_I(inode)->generation;
6252 * We could have gotten an inode number from somebody who was fsynced
6253 * and then removed in this same transaction, so let's just set full
6254 * sync since it will be a full sync anyway and this will blow away the
6255 * old info in the log.
6257 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6259 key[0].objectid = objectid;
6260 key[0].type = BTRFS_INODE_ITEM_KEY;
6263 sizes[0] = sizeof(struct btrfs_inode_item);
6267 * Start new inodes with an inode_ref. This is slightly more
6268 * efficient for small numbers of hard links since they will
6269 * be packed into one item. Extended refs will kick in if we
6270 * add more hard links than can fit in the ref item.
6272 key[1].objectid = objectid;
6273 key[1].type = BTRFS_INODE_REF_KEY;
6274 key[1].offset = ref_objectid;
6276 sizes[1] = name_len + sizeof(*ref);
6279 location = &BTRFS_I(inode)->location;
6280 location->objectid = objectid;
6281 location->offset = 0;
6282 location->type = BTRFS_INODE_ITEM_KEY;
6284 ret = btrfs_insert_inode_locked(inode);
6288 path->leave_spinning = 1;
6289 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6293 inode_init_owner(inode, dir, mode);
6294 inode_set_bytes(inode, 0);
6296 inode->i_mtime = current_time(inode);
6297 inode->i_atime = inode->i_mtime;
6298 inode->i_ctime = inode->i_mtime;
6299 BTRFS_I(inode)->i_otime = inode->i_mtime;
6301 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6302 struct btrfs_inode_item);
6303 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6304 sizeof(*inode_item));
6305 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6308 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6309 struct btrfs_inode_ref);
6310 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6311 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6312 ptr = (unsigned long)(ref + 1);
6313 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6316 btrfs_mark_buffer_dirty(path->nodes[0]);
6317 btrfs_free_path(path);
6319 btrfs_inherit_iflags(inode, dir);
6321 if (S_ISREG(mode)) {
6322 if (btrfs_test_opt(fs_info, NODATASUM))
6323 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6324 if (btrfs_test_opt(fs_info, NODATACOW))
6325 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6326 BTRFS_INODE_NODATASUM;
6329 inode_tree_add(inode);
6331 trace_btrfs_inode_new(inode);
6332 btrfs_set_inode_last_trans(trans, inode);
6334 btrfs_update_root_times(trans, root);
6336 ret = btrfs_inode_inherit_props(trans, inode, dir);
6339 "error inheriting props for ino %llu (root %llu): %d",
6340 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6345 unlock_new_inode(inode);
6348 BTRFS_I(dir)->index_cnt--;
6349 btrfs_free_path(path);
6351 return ERR_PTR(ret);
6354 static inline u8 btrfs_inode_type(struct inode *inode)
6356 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6360 * utility function to add 'inode' into 'parent_inode' with
6361 * a give name and a given sequence number.
6362 * if 'add_backref' is true, also insert a backref from the
6363 * inode to the parent directory.
6365 int btrfs_add_link(struct btrfs_trans_handle *trans,
6366 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6367 const char *name, int name_len, int add_backref, u64 index)
6369 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6371 struct btrfs_key key;
6372 struct btrfs_root *root = parent_inode->root;
6373 u64 ino = btrfs_ino(inode);
6374 u64 parent_ino = btrfs_ino(parent_inode);
6376 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6377 memcpy(&key, &inode->root->root_key, sizeof(key));
6380 key.type = BTRFS_INODE_ITEM_KEY;
6384 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6385 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6386 root->root_key.objectid, parent_ino,
6387 index, name, name_len);
6388 } else if (add_backref) {
6389 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6393 /* Nothing to clean up yet */
6397 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6399 btrfs_inode_type(&inode->vfs_inode), index);
6400 if (ret == -EEXIST || ret == -EOVERFLOW)
6403 btrfs_abort_transaction(trans, ret);
6407 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6409 inode_inc_iversion(&parent_inode->vfs_inode);
6410 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6411 current_time(&parent_inode->vfs_inode);
6412 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6414 btrfs_abort_transaction(trans, ret);
6418 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6421 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6422 root->root_key.objectid, parent_ino,
6423 &local_index, name, name_len);
6425 } else if (add_backref) {
6429 err = btrfs_del_inode_ref(trans, root, name, name_len,
6430 ino, parent_ino, &local_index);
6435 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6436 struct btrfs_inode *dir, struct dentry *dentry,
6437 struct btrfs_inode *inode, int backref, u64 index)
6439 int err = btrfs_add_link(trans, dir, inode,
6440 dentry->d_name.name, dentry->d_name.len,
6447 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6448 umode_t mode, dev_t rdev)
6450 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6451 struct btrfs_trans_handle *trans;
6452 struct btrfs_root *root = BTRFS_I(dir)->root;
6453 struct inode *inode = NULL;
6460 * 2 for inode item and ref
6462 * 1 for xattr if selinux is on
6464 trans = btrfs_start_transaction(root, 5);
6466 return PTR_ERR(trans);
6468 err = btrfs_find_free_ino(root, &objectid);
6472 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6473 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6475 if (IS_ERR(inode)) {
6476 err = PTR_ERR(inode);
6481 * If the active LSM wants to access the inode during
6482 * d_instantiate it needs these. Smack checks to see
6483 * if the filesystem supports xattrs by looking at the
6486 inode->i_op = &btrfs_special_inode_operations;
6487 init_special_inode(inode, inode->i_mode, rdev);
6489 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6491 goto out_unlock_inode;
6493 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6496 goto out_unlock_inode;
6498 btrfs_update_inode(trans, root, inode);
6499 unlock_new_inode(inode);
6500 d_instantiate(dentry, inode);
6504 btrfs_end_transaction(trans);
6505 btrfs_balance_delayed_items(fs_info);
6506 btrfs_btree_balance_dirty(fs_info);
6508 inode_dec_link_count(inode);
6515 unlock_new_inode(inode);
6520 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6521 umode_t mode, bool excl)
6523 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6524 struct btrfs_trans_handle *trans;
6525 struct btrfs_root *root = BTRFS_I(dir)->root;
6526 struct inode *inode = NULL;
6527 int drop_inode_on_err = 0;
6533 * 2 for inode item and ref
6535 * 1 for xattr if selinux is on
6537 trans = btrfs_start_transaction(root, 5);
6539 return PTR_ERR(trans);
6541 err = btrfs_find_free_ino(root, &objectid);
6545 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6546 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6548 if (IS_ERR(inode)) {
6549 err = PTR_ERR(inode);
6552 drop_inode_on_err = 1;
6554 * If the active LSM wants to access the inode during
6555 * d_instantiate it needs these. Smack checks to see
6556 * if the filesystem supports xattrs by looking at the
6559 inode->i_fop = &btrfs_file_operations;
6560 inode->i_op = &btrfs_file_inode_operations;
6561 inode->i_mapping->a_ops = &btrfs_aops;
6563 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6565 goto out_unlock_inode;
6567 err = btrfs_update_inode(trans, root, inode);
6569 goto out_unlock_inode;
6571 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6574 goto out_unlock_inode;
6576 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6577 unlock_new_inode(inode);
6578 d_instantiate(dentry, inode);
6581 btrfs_end_transaction(trans);
6582 if (err && drop_inode_on_err) {
6583 inode_dec_link_count(inode);
6586 btrfs_balance_delayed_items(fs_info);
6587 btrfs_btree_balance_dirty(fs_info);
6591 unlock_new_inode(inode);
6596 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6597 struct dentry *dentry)
6599 struct btrfs_trans_handle *trans = NULL;
6600 struct btrfs_root *root = BTRFS_I(dir)->root;
6601 struct inode *inode = d_inode(old_dentry);
6602 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6607 /* do not allow sys_link's with other subvols of the same device */
6608 if (root->objectid != BTRFS_I(inode)->root->objectid)
6611 if (inode->i_nlink >= BTRFS_LINK_MAX)
6614 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6619 * 2 items for inode and inode ref
6620 * 2 items for dir items
6621 * 1 item for parent inode
6623 trans = btrfs_start_transaction(root, 5);
6624 if (IS_ERR(trans)) {
6625 err = PTR_ERR(trans);
6630 /* There are several dir indexes for this inode, clear the cache. */
6631 BTRFS_I(inode)->dir_index = 0ULL;
6633 inode_inc_iversion(inode);
6634 inode->i_ctime = current_time(inode);
6636 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6638 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6644 struct dentry *parent = dentry->d_parent;
6645 err = btrfs_update_inode(trans, root, inode);
6648 if (inode->i_nlink == 1) {
6650 * If new hard link count is 1, it's a file created
6651 * with open(2) O_TMPFILE flag.
6653 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6657 d_instantiate(dentry, inode);
6658 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6661 btrfs_balance_delayed_items(fs_info);
6664 btrfs_end_transaction(trans);
6666 inode_dec_link_count(inode);
6669 btrfs_btree_balance_dirty(fs_info);
6673 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6675 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6676 struct inode *inode = NULL;
6677 struct btrfs_trans_handle *trans;
6678 struct btrfs_root *root = BTRFS_I(dir)->root;
6680 int drop_on_err = 0;
6685 * 2 items for inode and ref
6686 * 2 items for dir items
6687 * 1 for xattr if selinux is on
6689 trans = btrfs_start_transaction(root, 5);
6691 return PTR_ERR(trans);
6693 err = btrfs_find_free_ino(root, &objectid);
6697 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6698 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6699 S_IFDIR | mode, &index);
6700 if (IS_ERR(inode)) {
6701 err = PTR_ERR(inode);
6706 /* these must be set before we unlock the inode */
6707 inode->i_op = &btrfs_dir_inode_operations;
6708 inode->i_fop = &btrfs_dir_file_operations;
6710 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6712 goto out_fail_inode;
6714 btrfs_i_size_write(BTRFS_I(inode), 0);
6715 err = btrfs_update_inode(trans, root, inode);
6717 goto out_fail_inode;
6719 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6720 dentry->d_name.name,
6721 dentry->d_name.len, 0, index);
6723 goto out_fail_inode;
6725 d_instantiate(dentry, inode);
6727 * mkdir is special. We're unlocking after we call d_instantiate
6728 * to avoid a race with nfsd calling d_instantiate.
6730 unlock_new_inode(inode);
6734 btrfs_end_transaction(trans);
6736 inode_dec_link_count(inode);
6739 btrfs_balance_delayed_items(fs_info);
6740 btrfs_btree_balance_dirty(fs_info);
6744 unlock_new_inode(inode);
6748 /* Find next extent map of a given extent map, caller needs to ensure locks */
6749 static struct extent_map *next_extent_map(struct extent_map *em)
6751 struct rb_node *next;
6753 next = rb_next(&em->rb_node);
6756 return container_of(next, struct extent_map, rb_node);
6759 static struct extent_map *prev_extent_map(struct extent_map *em)
6761 struct rb_node *prev;
6763 prev = rb_prev(&em->rb_node);
6766 return container_of(prev, struct extent_map, rb_node);
6769 /* helper for btfs_get_extent. Given an existing extent in the tree,
6770 * the existing extent is the nearest extent to map_start,
6771 * and an extent that you want to insert, deal with overlap and insert
6772 * the best fitted new extent into the tree.
6774 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6775 struct extent_map *existing,
6776 struct extent_map *em,
6779 struct extent_map *prev;
6780 struct extent_map *next;
6785 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6787 if (existing->start > map_start) {
6789 prev = prev_extent_map(next);
6792 next = next_extent_map(prev);
6795 start = prev ? extent_map_end(prev) : em->start;
6796 start = max_t(u64, start, em->start);
6797 end = next ? next->start : extent_map_end(em);
6798 end = min_t(u64, end, extent_map_end(em));
6799 start_diff = start - em->start;
6801 em->len = end - start;
6802 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6803 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6804 em->block_start += start_diff;
6805 em->block_len -= start_diff;
6807 return add_extent_mapping(em_tree, em, 0);
6810 static noinline int uncompress_inline(struct btrfs_path *path,
6812 size_t pg_offset, u64 extent_offset,
6813 struct btrfs_file_extent_item *item)
6816 struct extent_buffer *leaf = path->nodes[0];
6819 unsigned long inline_size;
6823 WARN_ON(pg_offset != 0);
6824 compress_type = btrfs_file_extent_compression(leaf, item);
6825 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6826 inline_size = btrfs_file_extent_inline_item_len(leaf,
6827 btrfs_item_nr(path->slots[0]));
6828 tmp = kmalloc(inline_size, GFP_NOFS);
6831 ptr = btrfs_file_extent_inline_start(item);
6833 read_extent_buffer(leaf, tmp, ptr, inline_size);
6835 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6836 ret = btrfs_decompress(compress_type, tmp, page,
6837 extent_offset, inline_size, max_size);
6840 * decompression code contains a memset to fill in any space between the end
6841 * of the uncompressed data and the end of max_size in case the decompressed
6842 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6843 * the end of an inline extent and the beginning of the next block, so we
6844 * cover that region here.
6847 if (max_size + pg_offset < PAGE_SIZE) {
6848 char *map = kmap(page);
6849 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6857 * a bit scary, this does extent mapping from logical file offset to the disk.
6858 * the ugly parts come from merging extents from the disk with the in-ram
6859 * representation. This gets more complex because of the data=ordered code,
6860 * where the in-ram extents might be locked pending data=ordered completion.
6862 * This also copies inline extents directly into the page.
6864 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6866 size_t pg_offset, u64 start, u64 len,
6869 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6872 u64 extent_start = 0;
6874 u64 objectid = btrfs_ino(inode);
6876 struct btrfs_path *path = NULL;
6877 struct btrfs_root *root = inode->root;
6878 struct btrfs_file_extent_item *item;
6879 struct extent_buffer *leaf;
6880 struct btrfs_key found_key;
6881 struct extent_map *em = NULL;
6882 struct extent_map_tree *em_tree = &inode->extent_tree;
6883 struct extent_io_tree *io_tree = &inode->io_tree;
6884 struct btrfs_trans_handle *trans = NULL;
6885 const bool new_inline = !page || create;
6888 read_lock(&em_tree->lock);
6889 em = lookup_extent_mapping(em_tree, start, len);
6891 em->bdev = fs_info->fs_devices->latest_bdev;
6892 read_unlock(&em_tree->lock);
6895 if (em->start > start || em->start + em->len <= start)
6896 free_extent_map(em);
6897 else if (em->block_start == EXTENT_MAP_INLINE && page)
6898 free_extent_map(em);
6902 em = alloc_extent_map();
6907 em->bdev = fs_info->fs_devices->latest_bdev;
6908 em->start = EXTENT_MAP_HOLE;
6909 em->orig_start = EXTENT_MAP_HOLE;
6911 em->block_len = (u64)-1;
6914 path = btrfs_alloc_path();
6920 * Chances are we'll be called again, so go ahead and do
6923 path->reada = READA_FORWARD;
6926 ret = btrfs_lookup_file_extent(trans, root, path,
6927 objectid, start, trans != NULL);
6934 if (path->slots[0] == 0)
6939 leaf = path->nodes[0];
6940 item = btrfs_item_ptr(leaf, path->slots[0],
6941 struct btrfs_file_extent_item);
6942 /* are we inside the extent that was found? */
6943 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6944 found_type = found_key.type;
6945 if (found_key.objectid != objectid ||
6946 found_type != BTRFS_EXTENT_DATA_KEY) {
6948 * If we backup past the first extent we want to move forward
6949 * and see if there is an extent in front of us, otherwise we'll
6950 * say there is a hole for our whole search range which can
6957 found_type = btrfs_file_extent_type(leaf, item);
6958 extent_start = found_key.offset;
6959 if (found_type == BTRFS_FILE_EXTENT_REG ||
6960 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6961 extent_end = extent_start +
6962 btrfs_file_extent_num_bytes(leaf, item);
6964 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6966 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6968 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6969 extent_end = ALIGN(extent_start + size,
6970 fs_info->sectorsize);
6972 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6977 if (start >= extent_end) {
6979 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6980 ret = btrfs_next_leaf(root, path);
6987 leaf = path->nodes[0];
6989 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6990 if (found_key.objectid != objectid ||
6991 found_key.type != BTRFS_EXTENT_DATA_KEY)
6993 if (start + len <= found_key.offset)
6995 if (start > found_key.offset)
6998 em->orig_start = start;
6999 em->len = found_key.offset - start;
7003 btrfs_extent_item_to_extent_map(inode, path, item,
7006 if (found_type == BTRFS_FILE_EXTENT_REG ||
7007 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7009 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
7013 size_t extent_offset;
7019 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
7020 extent_offset = page_offset(page) + pg_offset - extent_start;
7021 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7022 size - extent_offset);
7023 em->start = extent_start + extent_offset;
7024 em->len = ALIGN(copy_size, fs_info->sectorsize);
7025 em->orig_block_len = em->len;
7026 em->orig_start = em->start;
7027 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7028 if (create == 0 && !PageUptodate(page)) {
7029 if (btrfs_file_extent_compression(leaf, item) !=
7030 BTRFS_COMPRESS_NONE) {
7031 ret = uncompress_inline(path, page, pg_offset,
7032 extent_offset, item);
7039 read_extent_buffer(leaf, map + pg_offset, ptr,
7041 if (pg_offset + copy_size < PAGE_SIZE) {
7042 memset(map + pg_offset + copy_size, 0,
7043 PAGE_SIZE - pg_offset -
7048 flush_dcache_page(page);
7049 } else if (create && PageUptodate(page)) {
7053 free_extent_map(em);
7056 btrfs_release_path(path);
7057 trans = btrfs_join_transaction(root);
7060 return ERR_CAST(trans);
7064 write_extent_buffer(leaf, map + pg_offset, ptr,
7067 btrfs_mark_buffer_dirty(leaf);
7069 set_extent_uptodate(io_tree, em->start,
7070 extent_map_end(em) - 1, NULL, GFP_NOFS);
7075 em->orig_start = start;
7078 em->block_start = EXTENT_MAP_HOLE;
7079 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
7081 btrfs_release_path(path);
7082 if (em->start > start || extent_map_end(em) <= start) {
7084 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7085 em->start, em->len, start, len);
7091 write_lock(&em_tree->lock);
7092 ret = add_extent_mapping(em_tree, em, 0);
7093 /* it is possible that someone inserted the extent into the tree
7094 * while we had the lock dropped. It is also possible that
7095 * an overlapping map exists in the tree
7097 if (ret == -EEXIST) {
7098 struct extent_map *existing;
7102 existing = search_extent_mapping(em_tree, start, len);
7104 * existing will always be non-NULL, since there must be
7105 * extent causing the -EEXIST.
7107 if (existing->start == em->start &&
7108 extent_map_end(existing) >= extent_map_end(em) &&
7109 em->block_start == existing->block_start) {
7111 * The existing extent map already encompasses the
7112 * entire extent map we tried to add.
7114 free_extent_map(em);
7118 } else if (start >= extent_map_end(existing) ||
7119 start <= existing->start) {
7121 * The existing extent map is the one nearest to
7122 * the [start, start + len) range which overlaps
7124 err = merge_extent_mapping(em_tree, existing,
7126 free_extent_map(existing);
7128 free_extent_map(em);
7132 free_extent_map(em);
7137 write_unlock(&em_tree->lock);
7140 trace_btrfs_get_extent(root, inode, em);
7142 btrfs_free_path(path);
7144 ret = btrfs_end_transaction(trans);
7149 free_extent_map(em);
7150 return ERR_PTR(err);
7152 BUG_ON(!em); /* Error is always set */
7156 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7158 size_t pg_offset, u64 start, u64 len,
7161 struct extent_map *em;
7162 struct extent_map *hole_em = NULL;
7163 u64 range_start = start;
7169 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7173 * If our em maps to:
7175 * - a pre-alloc extent,
7176 * there might actually be delalloc bytes behind it.
7178 if (em->block_start != EXTENT_MAP_HOLE &&
7179 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7184 /* check to see if we've wrapped (len == -1 or similar) */
7193 /* ok, we didn't find anything, lets look for delalloc */
7194 found = count_range_bits(&inode->io_tree, &range_start,
7195 end, len, EXTENT_DELALLOC, 1);
7196 found_end = range_start + found;
7197 if (found_end < range_start)
7198 found_end = (u64)-1;
7201 * we didn't find anything useful, return
7202 * the original results from get_extent()
7204 if (range_start > end || found_end <= start) {
7210 /* adjust the range_start to make sure it doesn't
7211 * go backwards from the start they passed in
7213 range_start = max(start, range_start);
7214 found = found_end - range_start;
7217 u64 hole_start = start;
7220 em = alloc_extent_map();
7226 * when btrfs_get_extent can't find anything it
7227 * returns one huge hole
7229 * make sure what it found really fits our range, and
7230 * adjust to make sure it is based on the start from
7234 u64 calc_end = extent_map_end(hole_em);
7236 if (calc_end <= start || (hole_em->start > end)) {
7237 free_extent_map(hole_em);
7240 hole_start = max(hole_em->start, start);
7241 hole_len = calc_end - hole_start;
7245 if (hole_em && range_start > hole_start) {
7246 /* our hole starts before our delalloc, so we
7247 * have to return just the parts of the hole
7248 * that go until the delalloc starts
7250 em->len = min(hole_len,
7251 range_start - hole_start);
7252 em->start = hole_start;
7253 em->orig_start = hole_start;
7255 * don't adjust block start at all,
7256 * it is fixed at EXTENT_MAP_HOLE
7258 em->block_start = hole_em->block_start;
7259 em->block_len = hole_len;
7260 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7261 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7263 em->start = range_start;
7265 em->orig_start = range_start;
7266 em->block_start = EXTENT_MAP_DELALLOC;
7267 em->block_len = found;
7269 } else if (hole_em) {
7274 free_extent_map(hole_em);
7276 free_extent_map(em);
7277 return ERR_PTR(err);
7282 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7285 const u64 orig_start,
7286 const u64 block_start,
7287 const u64 block_len,
7288 const u64 orig_block_len,
7289 const u64 ram_bytes,
7292 struct extent_map *em = NULL;
7295 if (type != BTRFS_ORDERED_NOCOW) {
7296 em = create_io_em(inode, start, len, orig_start,
7297 block_start, block_len, orig_block_len,
7299 BTRFS_COMPRESS_NONE, /* compress_type */
7304 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7305 len, block_len, type);
7308 free_extent_map(em);
7309 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7310 start + len - 1, 0);
7319 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7322 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7323 struct btrfs_root *root = BTRFS_I(inode)->root;
7324 struct extent_map *em;
7325 struct btrfs_key ins;
7329 alloc_hint = get_extent_allocation_hint(inode, start, len);
7330 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7331 0, alloc_hint, &ins, 1, 1);
7333 return ERR_PTR(ret);
7335 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7336 ins.objectid, ins.offset, ins.offset,
7337 ins.offset, BTRFS_ORDERED_REGULAR);
7338 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7340 btrfs_free_reserved_extent(fs_info, ins.objectid,
7347 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7348 * block must be cow'd
7350 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7351 u64 *orig_start, u64 *orig_block_len,
7354 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7355 struct btrfs_path *path;
7357 struct extent_buffer *leaf;
7358 struct btrfs_root *root = BTRFS_I(inode)->root;
7359 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7360 struct btrfs_file_extent_item *fi;
7361 struct btrfs_key key;
7368 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7370 path = btrfs_alloc_path();
7374 ret = btrfs_lookup_file_extent(NULL, root, path,
7375 btrfs_ino(BTRFS_I(inode)), offset, 0);
7379 slot = path->slots[0];
7382 /* can't find the item, must cow */
7389 leaf = path->nodes[0];
7390 btrfs_item_key_to_cpu(leaf, &key, slot);
7391 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7392 key.type != BTRFS_EXTENT_DATA_KEY) {
7393 /* not our file or wrong item type, must cow */
7397 if (key.offset > offset) {
7398 /* Wrong offset, must cow */
7402 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7403 found_type = btrfs_file_extent_type(leaf, fi);
7404 if (found_type != BTRFS_FILE_EXTENT_REG &&
7405 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7406 /* not a regular extent, must cow */
7410 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7413 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7414 if (extent_end <= offset)
7417 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7418 if (disk_bytenr == 0)
7421 if (btrfs_file_extent_compression(leaf, fi) ||
7422 btrfs_file_extent_encryption(leaf, fi) ||
7423 btrfs_file_extent_other_encoding(leaf, fi))
7426 backref_offset = btrfs_file_extent_offset(leaf, fi);
7429 *orig_start = key.offset - backref_offset;
7430 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7431 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7434 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7437 num_bytes = min(offset + *len, extent_end) - offset;
7438 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7441 range_end = round_up(offset + num_bytes,
7442 root->fs_info->sectorsize) - 1;
7443 ret = test_range_bit(io_tree, offset, range_end,
7444 EXTENT_DELALLOC, 0, NULL);
7451 btrfs_release_path(path);
7454 * look for other files referencing this extent, if we
7455 * find any we must cow
7458 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7459 key.offset - backref_offset, disk_bytenr);
7466 * adjust disk_bytenr and num_bytes to cover just the bytes
7467 * in this extent we are about to write. If there
7468 * are any csums in that range we have to cow in order
7469 * to keep the csums correct
7471 disk_bytenr += backref_offset;
7472 disk_bytenr += offset - key.offset;
7473 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7476 * all of the above have passed, it is safe to overwrite this extent
7482 btrfs_free_path(path);
7486 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7488 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7490 void **pagep = NULL;
7491 struct page *page = NULL;
7492 unsigned long start_idx;
7493 unsigned long end_idx;
7495 start_idx = start >> PAGE_SHIFT;
7498 * end is the last byte in the last page. end == start is legal
7500 end_idx = end >> PAGE_SHIFT;
7504 /* Most of the code in this while loop is lifted from
7505 * find_get_page. It's been modified to begin searching from a
7506 * page and return just the first page found in that range. If the
7507 * found idx is less than or equal to the end idx then we know that
7508 * a page exists. If no pages are found or if those pages are
7509 * outside of the range then we're fine (yay!) */
7510 while (page == NULL &&
7511 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7512 page = radix_tree_deref_slot(pagep);
7513 if (unlikely(!page))
7516 if (radix_tree_exception(page)) {
7517 if (radix_tree_deref_retry(page)) {
7522 * Otherwise, shmem/tmpfs must be storing a swap entry
7523 * here as an exceptional entry: so return it without
7524 * attempting to raise page count.
7527 break; /* TODO: Is this relevant for this use case? */
7530 if (!page_cache_get_speculative(page)) {
7536 * Has the page moved?
7537 * This is part of the lockless pagecache protocol. See
7538 * include/linux/pagemap.h for details.
7540 if (unlikely(page != *pagep)) {
7547 if (page->index <= end_idx)
7556 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7557 struct extent_state **cached_state, int writing)
7559 struct btrfs_ordered_extent *ordered;
7563 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7566 * We're concerned with the entire range that we're going to be
7567 * doing DIO to, so we need to make sure there's no ordered
7568 * extents in this range.
7570 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7571 lockend - lockstart + 1);
7574 * We need to make sure there are no buffered pages in this
7575 * range either, we could have raced between the invalidate in
7576 * generic_file_direct_write and locking the extent. The
7577 * invalidate needs to happen so that reads after a write do not
7582 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7585 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7586 cached_state, GFP_NOFS);
7590 * If we are doing a DIO read and the ordered extent we
7591 * found is for a buffered write, we can not wait for it
7592 * to complete and retry, because if we do so we can
7593 * deadlock with concurrent buffered writes on page
7594 * locks. This happens only if our DIO read covers more
7595 * than one extent map, if at this point has already
7596 * created an ordered extent for a previous extent map
7597 * and locked its range in the inode's io tree, and a
7598 * concurrent write against that previous extent map's
7599 * range and this range started (we unlock the ranges
7600 * in the io tree only when the bios complete and
7601 * buffered writes always lock pages before attempting
7602 * to lock range in the io tree).
7605 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7606 btrfs_start_ordered_extent(inode, ordered, 1);
7609 btrfs_put_ordered_extent(ordered);
7612 * We could trigger writeback for this range (and wait
7613 * for it to complete) and then invalidate the pages for
7614 * this range (through invalidate_inode_pages2_range()),
7615 * but that can lead us to a deadlock with a concurrent
7616 * call to readpages() (a buffered read or a defrag call
7617 * triggered a readahead) on a page lock due to an
7618 * ordered dio extent we created before but did not have
7619 * yet a corresponding bio submitted (whence it can not
7620 * complete), which makes readpages() wait for that
7621 * ordered extent to complete while holding a lock on
7636 /* The callers of this must take lock_extent() */
7637 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7638 u64 orig_start, u64 block_start,
7639 u64 block_len, u64 orig_block_len,
7640 u64 ram_bytes, int compress_type,
7643 struct extent_map_tree *em_tree;
7644 struct extent_map *em;
7645 struct btrfs_root *root = BTRFS_I(inode)->root;
7648 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7649 type == BTRFS_ORDERED_COMPRESSED ||
7650 type == BTRFS_ORDERED_NOCOW ||
7651 type == BTRFS_ORDERED_REGULAR);
7653 em_tree = &BTRFS_I(inode)->extent_tree;
7654 em = alloc_extent_map();
7656 return ERR_PTR(-ENOMEM);
7659 em->orig_start = orig_start;
7661 em->block_len = block_len;
7662 em->block_start = block_start;
7663 em->bdev = root->fs_info->fs_devices->latest_bdev;
7664 em->orig_block_len = orig_block_len;
7665 em->ram_bytes = ram_bytes;
7666 em->generation = -1;
7667 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7668 if (type == BTRFS_ORDERED_PREALLOC) {
7669 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7670 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7671 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7672 em->compress_type = compress_type;
7676 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7677 em->start + em->len - 1, 0);
7678 write_lock(&em_tree->lock);
7679 ret = add_extent_mapping(em_tree, em, 1);
7680 write_unlock(&em_tree->lock);
7682 * The caller has taken lock_extent(), who could race with us
7685 } while (ret == -EEXIST);
7688 free_extent_map(em);
7689 return ERR_PTR(ret);
7692 /* em got 2 refs now, callers needs to do free_extent_map once. */
7696 static void adjust_dio_outstanding_extents(struct inode *inode,
7697 struct btrfs_dio_data *dio_data,
7700 unsigned num_extents = count_max_extents(len);
7703 * If we have an outstanding_extents count still set then we're
7704 * within our reservation, otherwise we need to adjust our inode
7705 * counter appropriately.
7707 if (dio_data->outstanding_extents >= num_extents) {
7708 dio_data->outstanding_extents -= num_extents;
7711 * If dio write length has been split due to no large enough
7712 * contiguous space, we need to compensate our inode counter
7715 u64 num_needed = num_extents - dio_data->outstanding_extents;
7717 spin_lock(&BTRFS_I(inode)->lock);
7718 BTRFS_I(inode)->outstanding_extents += num_needed;
7719 spin_unlock(&BTRFS_I(inode)->lock);
7723 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7724 struct buffer_head *bh_result, int create)
7726 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7727 struct extent_map *em;
7728 struct extent_state *cached_state = NULL;
7729 struct btrfs_dio_data *dio_data = NULL;
7730 u64 start = iblock << inode->i_blkbits;
7731 u64 lockstart, lockend;
7732 u64 len = bh_result->b_size;
7733 int unlock_bits = EXTENT_LOCKED;
7737 unlock_bits |= EXTENT_DIRTY;
7739 len = min_t(u64, len, fs_info->sectorsize);
7742 lockend = start + len - 1;
7744 if (current->journal_info) {
7746 * Need to pull our outstanding extents and set journal_info to NULL so
7747 * that anything that needs to check if there's a transaction doesn't get
7750 dio_data = current->journal_info;
7751 current->journal_info = NULL;
7755 * If this errors out it's because we couldn't invalidate pagecache for
7756 * this range and we need to fallback to buffered.
7758 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7764 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7771 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7772 * io. INLINE is special, and we could probably kludge it in here, but
7773 * it's still buffered so for safety lets just fall back to the generic
7776 * For COMPRESSED we _have_ to read the entire extent in so we can
7777 * decompress it, so there will be buffering required no matter what we
7778 * do, so go ahead and fallback to buffered.
7780 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7781 * to buffered IO. Don't blame me, this is the price we pay for using
7784 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7785 em->block_start == EXTENT_MAP_INLINE) {
7786 free_extent_map(em);
7791 /* Just a good old fashioned hole, return */
7792 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7793 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7794 free_extent_map(em);
7799 * We don't allocate a new extent in the following cases
7801 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7803 * 2) The extent is marked as PREALLOC. We're good to go here and can
7804 * just use the extent.
7808 len = min(len, em->len - (start - em->start));
7809 lockstart = start + len;
7813 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7814 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7815 em->block_start != EXTENT_MAP_HOLE)) {
7817 u64 block_start, orig_start, orig_block_len, ram_bytes;
7819 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7820 type = BTRFS_ORDERED_PREALLOC;
7822 type = BTRFS_ORDERED_NOCOW;
7823 len = min(len, em->len - (start - em->start));
7824 block_start = em->block_start + (start - em->start);
7826 if (can_nocow_extent(inode, start, &len, &orig_start,
7827 &orig_block_len, &ram_bytes) == 1 &&
7828 btrfs_inc_nocow_writers(fs_info, block_start)) {
7829 struct extent_map *em2;
7831 em2 = btrfs_create_dio_extent(inode, start, len,
7832 orig_start, block_start,
7833 len, orig_block_len,
7835 btrfs_dec_nocow_writers(fs_info, block_start);
7836 if (type == BTRFS_ORDERED_PREALLOC) {
7837 free_extent_map(em);
7840 if (em2 && IS_ERR(em2)) {
7845 * For inode marked NODATACOW or extent marked PREALLOC,
7846 * use the existing or preallocated extent, so does not
7847 * need to adjust btrfs_space_info's bytes_may_use.
7849 btrfs_free_reserved_data_space_noquota(inode,
7856 * this will cow the extent, reset the len in case we changed
7859 len = bh_result->b_size;
7860 free_extent_map(em);
7861 em = btrfs_new_extent_direct(inode, start, len);
7866 len = min(len, em->len - (start - em->start));
7868 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7870 bh_result->b_size = len;
7871 bh_result->b_bdev = em->bdev;
7872 set_buffer_mapped(bh_result);
7874 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7875 set_buffer_new(bh_result);
7878 * Need to update the i_size under the extent lock so buffered
7879 * readers will get the updated i_size when we unlock.
7881 if (!dio_data->overwrite && start + len > i_size_read(inode))
7882 i_size_write(inode, start + len);
7884 adjust_dio_outstanding_extents(inode, dio_data, len);
7885 WARN_ON(dio_data->reserve < len);
7886 dio_data->reserve -= len;
7887 dio_data->unsubmitted_oe_range_end = start + len;
7888 current->journal_info = dio_data;
7892 * In the case of write we need to clear and unlock the entire range,
7893 * in the case of read we need to unlock only the end area that we
7894 * aren't using if there is any left over space.
7896 if (lockstart < lockend) {
7897 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7898 lockend, unlock_bits, 1, 0,
7899 &cached_state, GFP_NOFS);
7901 free_extent_state(cached_state);
7904 free_extent_map(em);
7909 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7910 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7913 current->journal_info = dio_data;
7915 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7916 * write less data then expected, so that we don't underflow our inode's
7917 * outstanding extents counter.
7919 if (create && dio_data)
7920 adjust_dio_outstanding_extents(inode, dio_data, len);
7925 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7931 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7935 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7939 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7945 static int btrfs_check_dio_repairable(struct inode *inode,
7946 struct bio *failed_bio,
7947 struct io_failure_record *failrec,
7950 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7953 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7954 if (num_copies == 1) {
7956 * we only have a single copy of the data, so don't bother with
7957 * all the retry and error correction code that follows. no
7958 * matter what the error is, it is very likely to persist.
7960 btrfs_debug(fs_info,
7961 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7962 num_copies, failrec->this_mirror, failed_mirror);
7966 failrec->failed_mirror = failed_mirror;
7967 failrec->this_mirror++;
7968 if (failrec->this_mirror == failed_mirror)
7969 failrec->this_mirror++;
7971 if (failrec->this_mirror > num_copies) {
7972 btrfs_debug(fs_info,
7973 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7974 num_copies, failrec->this_mirror, failed_mirror);
7981 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7982 struct page *page, unsigned int pgoff,
7983 u64 start, u64 end, int failed_mirror,
7984 bio_end_io_t *repair_endio, void *repair_arg)
7986 struct io_failure_record *failrec;
7987 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7988 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7995 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7997 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
8001 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
8004 free_io_failure(failure_tree, io_tree, failrec);
8008 segs = bio_segments(failed_bio);
8010 (failed_bio->bi_io_vec->bv_len > btrfs_inode_sectorsize(inode)))
8011 read_mode |= REQ_FAILFAST_DEV;
8013 isector = start - btrfs_io_bio(failed_bio)->logical;
8014 isector >>= inode->i_sb->s_blocksize_bits;
8015 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
8016 pgoff, isector, repair_endio, repair_arg);
8018 free_io_failure(failure_tree, io_tree, failrec);
8021 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
8023 btrfs_debug(BTRFS_I(inode)->root->fs_info,
8024 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
8025 read_mode, failrec->this_mirror, failrec->in_validation);
8027 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
8029 free_io_failure(failure_tree, io_tree, failrec);
8036 struct btrfs_retry_complete {
8037 struct completion done;
8038 struct inode *inode;
8043 static void btrfs_retry_endio_nocsum(struct bio *bio)
8045 struct btrfs_retry_complete *done = bio->bi_private;
8046 struct inode *inode = done->inode;
8047 struct bio_vec *bvec;
8048 struct extent_io_tree *io_tree, *failure_tree;
8054 ASSERT(bio->bi_vcnt == 1);
8055 io_tree = &BTRFS_I(inode)->io_tree;
8056 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8057 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(inode));
8060 bio_for_each_segment_all(bvec, bio, i)
8061 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
8062 io_tree, done->start, bvec->bv_page,
8063 btrfs_ino(BTRFS_I(inode)), 0);
8065 complete(&done->done);
8069 static int __btrfs_correct_data_nocsum(struct inode *inode,
8070 struct btrfs_io_bio *io_bio)
8072 struct btrfs_fs_info *fs_info;
8073 struct bio_vec bvec;
8074 struct bvec_iter iter;
8075 struct btrfs_retry_complete done;
8083 fs_info = BTRFS_I(inode)->root->fs_info;
8084 sectorsize = fs_info->sectorsize;
8086 start = io_bio->logical;
8088 io_bio->bio.bi_iter = io_bio->iter;
8090 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8091 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8092 pgoff = bvec.bv_offset;
8094 next_block_or_try_again:
8097 init_completion(&done.done);
8099 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8100 pgoff, start, start + sectorsize - 1,
8102 btrfs_retry_endio_nocsum, &done);
8108 wait_for_completion(&done.done);
8110 if (!done.uptodate) {
8111 /* We might have another mirror, so try again */
8112 goto next_block_or_try_again;
8116 start += sectorsize;
8120 pgoff += sectorsize;
8121 ASSERT(pgoff < PAGE_SIZE);
8122 goto next_block_or_try_again;
8129 static void btrfs_retry_endio(struct bio *bio)
8131 struct btrfs_retry_complete *done = bio->bi_private;
8132 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8133 struct extent_io_tree *io_tree, *failure_tree;
8134 struct inode *inode = done->inode;
8135 struct bio_vec *bvec;
8145 ASSERT(bio->bi_vcnt == 1);
8146 ASSERT(bio->bi_io_vec->bv_len == btrfs_inode_sectorsize(done->inode));
8148 io_tree = &BTRFS_I(inode)->io_tree;
8149 failure_tree = &BTRFS_I(inode)->io_failure_tree;
8151 bio_for_each_segment_all(bvec, bio, i) {
8152 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8153 bvec->bv_offset, done->start,
8156 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8157 failure_tree, io_tree, done->start,
8159 btrfs_ino(BTRFS_I(inode)),
8165 done->uptodate = uptodate;
8167 complete(&done->done);
8171 static int __btrfs_subio_endio_read(struct inode *inode,
8172 struct btrfs_io_bio *io_bio, int err)
8174 struct btrfs_fs_info *fs_info;
8175 struct bio_vec bvec;
8176 struct bvec_iter iter;
8177 struct btrfs_retry_complete done;
8186 fs_info = BTRFS_I(inode)->root->fs_info;
8187 sectorsize = fs_info->sectorsize;
8190 start = io_bio->logical;
8192 io_bio->bio.bi_iter = io_bio->iter;
8194 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8195 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8197 pgoff = bvec.bv_offset;
8199 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8200 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8201 bvec.bv_page, pgoff, start,
8208 init_completion(&done.done);
8210 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8211 pgoff, start, start + sectorsize - 1,
8213 btrfs_retry_endio, &done);
8219 wait_for_completion(&done.done);
8221 if (!done.uptodate) {
8222 /* We might have another mirror, so try again */
8226 offset += sectorsize;
8227 start += sectorsize;
8233 pgoff += sectorsize;
8234 ASSERT(pgoff < PAGE_SIZE);
8242 static int btrfs_subio_endio_read(struct inode *inode,
8243 struct btrfs_io_bio *io_bio, int err)
8245 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8249 return __btrfs_correct_data_nocsum(inode, io_bio);
8253 return __btrfs_subio_endio_read(inode, io_bio, err);
8257 static void btrfs_endio_direct_read(struct bio *bio)
8259 struct btrfs_dio_private *dip = bio->bi_private;
8260 struct inode *inode = dip->inode;
8261 struct bio *dio_bio;
8262 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8263 int err = bio->bi_error;
8265 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8266 err = btrfs_subio_endio_read(inode, io_bio, err);
8268 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8269 dip->logical_offset + dip->bytes - 1);
8270 dio_bio = dip->dio_bio;
8274 dio_bio->bi_error = bio->bi_error;
8275 dio_end_io(dio_bio, bio->bi_error);
8278 io_bio->end_io(io_bio, err);
8282 static void __endio_write_update_ordered(struct inode *inode,
8283 const u64 offset, const u64 bytes,
8284 const bool uptodate)
8286 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8287 struct btrfs_ordered_extent *ordered = NULL;
8288 struct btrfs_workqueue *wq;
8289 btrfs_work_func_t func;
8290 u64 ordered_offset = offset;
8291 u64 ordered_bytes = bytes;
8294 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8295 wq = fs_info->endio_freespace_worker;
8296 func = btrfs_freespace_write_helper;
8298 wq = fs_info->endio_write_workers;
8299 func = btrfs_endio_write_helper;
8303 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8310 btrfs_init_work(&ordered->work, func, finish_ordered_fn, NULL, NULL);
8311 btrfs_queue_work(wq, &ordered->work);
8314 * our bio might span multiple ordered extents. If we haven't
8315 * completed the accounting for the whole dio, go back and try again
8317 if (ordered_offset < offset + bytes) {
8318 ordered_bytes = offset + bytes - ordered_offset;
8324 static void btrfs_endio_direct_write(struct bio *bio)
8326 struct btrfs_dio_private *dip = bio->bi_private;
8327 struct bio *dio_bio = dip->dio_bio;
8329 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8330 dip->bytes, !bio->bi_error);
8334 dio_bio->bi_error = bio->bi_error;
8335 dio_end_io(dio_bio, bio->bi_error);
8339 static int __btrfs_submit_bio_start_direct_io(void *private_data,
8340 struct bio *bio, int mirror_num,
8341 unsigned long bio_flags, u64 offset)
8343 struct inode *inode = private_data;
8345 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8346 BUG_ON(ret); /* -ENOMEM */
8350 static void btrfs_end_dio_bio(struct bio *bio)
8352 struct btrfs_dio_private *dip = bio->bi_private;
8353 int err = bio->bi_error;
8356 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8357 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8358 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8360 (unsigned long long)bio->bi_iter.bi_sector,
8361 bio->bi_iter.bi_size, err);
8363 if (dip->subio_endio)
8364 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8370 * before atomic variable goto zero, we must make sure
8371 * dip->errors is perceived to be set.
8373 smp_mb__before_atomic();
8376 /* if there are more bios still pending for this dio, just exit */
8377 if (!atomic_dec_and_test(&dip->pending_bios))
8381 bio_io_error(dip->orig_bio);
8383 dip->dio_bio->bi_error = 0;
8384 bio_endio(dip->orig_bio);
8390 static inline int btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8391 struct btrfs_dio_private *dip,
8395 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8396 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8400 * We load all the csum data we need when we submit
8401 * the first bio to reduce the csum tree search and
8404 if (dip->logical_offset == file_offset) {
8405 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8411 if (bio == dip->orig_bio)
8414 file_offset -= dip->logical_offset;
8415 file_offset >>= inode->i_sb->s_blocksize_bits;
8416 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8421 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8422 u64 file_offset, int skip_sum,
8425 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8426 struct btrfs_dio_private *dip = bio->bi_private;
8427 bool write = bio_op(bio) == REQ_OP_WRITE;
8431 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8436 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8444 if (write && async_submit) {
8445 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8447 __btrfs_submit_bio_start_direct_io,
8448 __btrfs_submit_bio_done);
8452 * If we aren't doing async submit, calculate the csum of the
8455 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8459 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8465 ret = btrfs_map_bio(fs_info, bio, 0, async_submit);
8471 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8474 struct inode *inode = dip->inode;
8475 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8477 struct bio *orig_bio = dip->orig_bio;
8478 u64 start_sector = orig_bio->bi_iter.bi_sector;
8479 u64 file_offset = dip->logical_offset;
8481 int async_submit = 0;
8483 int clone_offset = 0;
8487 map_length = orig_bio->bi_iter.bi_size;
8488 submit_len = map_length;
8489 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8490 &map_length, NULL, 0);
8494 if (map_length >= submit_len) {
8496 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8500 /* async crcs make it difficult to collect full stripe writes. */
8501 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8507 ASSERT(map_length <= INT_MAX);
8508 atomic_inc(&dip->pending_bios);
8510 clone_len = min_t(int, submit_len, map_length);
8513 * This will never fail as it's passing GPF_NOFS and
8514 * the allocation is backed by btrfs_bioset.
8516 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8518 bio->bi_private = dip;
8519 bio->bi_end_io = btrfs_end_dio_bio;
8520 btrfs_io_bio(bio)->logical = file_offset;
8522 ASSERT(submit_len >= clone_len);
8523 submit_len -= clone_len;
8524 if (submit_len == 0)
8528 * Increase the count before we submit the bio so we know
8529 * the end IO handler won't happen before we increase the
8530 * count. Otherwise, the dip might get freed before we're
8531 * done setting it up.
8533 atomic_inc(&dip->pending_bios);
8535 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8539 atomic_dec(&dip->pending_bios);
8543 clone_offset += clone_len;
8544 start_sector += clone_len >> 9;
8545 file_offset += clone_len;
8547 map_length = submit_len;
8548 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8549 start_sector << 9, &map_length, NULL, 0);
8552 } while (submit_len > 0);
8555 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8564 * before atomic variable goto zero, we must
8565 * make sure dip->errors is perceived to be set.
8567 smp_mb__before_atomic();
8568 if (atomic_dec_and_test(&dip->pending_bios))
8569 bio_io_error(dip->orig_bio);
8571 /* bio_end_io() will handle error, so we needn't return it */
8575 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8578 struct btrfs_dio_private *dip = NULL;
8579 struct bio *bio = NULL;
8580 struct btrfs_io_bio *io_bio;
8582 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8585 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8587 bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8593 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8599 dip->private = dio_bio->bi_private;
8601 dip->logical_offset = file_offset;
8602 dip->bytes = dio_bio->bi_iter.bi_size;
8603 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8604 bio->bi_private = dip;
8605 dip->orig_bio = bio;
8606 dip->dio_bio = dio_bio;
8607 atomic_set(&dip->pending_bios, 0);
8608 io_bio = btrfs_io_bio(bio);
8609 io_bio->logical = file_offset;
8612 bio->bi_end_io = btrfs_endio_direct_write;
8614 bio->bi_end_io = btrfs_endio_direct_read;
8615 dip->subio_endio = btrfs_subio_endio_read;
8619 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8620 * even if we fail to submit a bio, because in such case we do the
8621 * corresponding error handling below and it must not be done a second
8622 * time by btrfs_direct_IO().
8625 struct btrfs_dio_data *dio_data = current->journal_info;
8627 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8629 dio_data->unsubmitted_oe_range_start =
8630 dio_data->unsubmitted_oe_range_end;
8633 ret = btrfs_submit_direct_hook(dip, skip_sum);
8638 io_bio->end_io(io_bio, ret);
8642 * If we arrived here it means either we failed to submit the dip
8643 * or we either failed to clone the dio_bio or failed to allocate the
8644 * dip. If we cloned the dio_bio and allocated the dip, we can just
8645 * call bio_endio against our io_bio so that we get proper resource
8646 * cleanup if we fail to submit the dip, otherwise, we must do the
8647 * same as btrfs_endio_direct_[write|read] because we can't call these
8648 * callbacks - they require an allocated dip and a clone of dio_bio.
8651 bio->bi_error = -EIO;
8654 * The end io callbacks free our dip, do the final put on bio
8655 * and all the cleanup and final put for dio_bio (through
8662 __endio_write_update_ordered(inode,
8664 dio_bio->bi_iter.bi_size,
8667 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8668 file_offset + dio_bio->bi_iter.bi_size - 1);
8670 dio_bio->bi_error = -EIO;
8672 * Releases and cleans up our dio_bio, no need to bio_put()
8673 * nor bio_endio()/bio_io_error() against dio_bio.
8675 dio_end_io(dio_bio, ret);
8682 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8684 const struct iov_iter *iter, loff_t offset)
8688 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8689 ssize_t retval = -EINVAL;
8691 if (offset & blocksize_mask)
8694 if (iov_iter_alignment(iter) & blocksize_mask)
8697 /* If this is a write we don't need to check anymore */
8698 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8701 * Check to make sure we don't have duplicate iov_base's in this
8702 * iovec, if so return EINVAL, otherwise we'll get csum errors
8703 * when reading back.
8705 for (seg = 0; seg < iter->nr_segs; seg++) {
8706 for (i = seg + 1; i < iter->nr_segs; i++) {
8707 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8716 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8718 struct file *file = iocb->ki_filp;
8719 struct inode *inode = file->f_mapping->host;
8720 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8721 struct btrfs_dio_data dio_data = { 0 };
8722 loff_t offset = iocb->ki_pos;
8726 bool relock = false;
8729 if (check_direct_IO(fs_info, iocb, iter, offset))
8732 inode_dio_begin(inode);
8733 smp_mb__after_atomic();
8736 * The generic stuff only does filemap_write_and_wait_range, which
8737 * isn't enough if we've written compressed pages to this area, so
8738 * we need to flush the dirty pages again to make absolutely sure
8739 * that any outstanding dirty pages are on disk.
8741 count = iov_iter_count(iter);
8742 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8743 &BTRFS_I(inode)->runtime_flags))
8744 filemap_fdatawrite_range(inode->i_mapping, offset,
8745 offset + count - 1);
8747 if (iov_iter_rw(iter) == WRITE) {
8749 * If the write DIO is beyond the EOF, we need update
8750 * the isize, but it is protected by i_mutex. So we can
8751 * not unlock the i_mutex at this case.
8753 if (offset + count <= inode->i_size) {
8754 dio_data.overwrite = 1;
8755 inode_unlock(inode);
8758 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8761 dio_data.outstanding_extents = count_max_extents(count);
8764 * We need to know how many extents we reserved so that we can
8765 * do the accounting properly if we go over the number we
8766 * originally calculated. Abuse current->journal_info for this.
8768 dio_data.reserve = round_up(count,
8769 fs_info->sectorsize);
8770 dio_data.unsubmitted_oe_range_start = (u64)offset;
8771 dio_data.unsubmitted_oe_range_end = (u64)offset;
8772 current->journal_info = &dio_data;
8773 down_read(&BTRFS_I(inode)->dio_sem);
8774 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8775 &BTRFS_I(inode)->runtime_flags)) {
8776 inode_dio_end(inode);
8777 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8781 ret = __blockdev_direct_IO(iocb, inode,
8782 fs_info->fs_devices->latest_bdev,
8783 iter, btrfs_get_blocks_direct, NULL,
8784 btrfs_submit_direct, flags);
8785 if (iov_iter_rw(iter) == WRITE) {
8786 up_read(&BTRFS_I(inode)->dio_sem);
8787 current->journal_info = NULL;
8788 if (ret < 0 && ret != -EIOCBQUEUED) {
8789 if (dio_data.reserve)
8790 btrfs_delalloc_release_space(inode, offset,
8793 * On error we might have left some ordered extents
8794 * without submitting corresponding bios for them, so
8795 * cleanup them up to avoid other tasks getting them
8796 * and waiting for them to complete forever.
8798 if (dio_data.unsubmitted_oe_range_start <
8799 dio_data.unsubmitted_oe_range_end)
8800 __endio_write_update_ordered(inode,
8801 dio_data.unsubmitted_oe_range_start,
8802 dio_data.unsubmitted_oe_range_end -
8803 dio_data.unsubmitted_oe_range_start,
8805 } else if (ret >= 0 && (size_t)ret < count)
8806 btrfs_delalloc_release_space(inode, offset,
8807 count - (size_t)ret);
8811 inode_dio_end(inode);
8818 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8820 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8821 __u64 start, __u64 len)
8825 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8829 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8832 int btrfs_readpage(struct file *file, struct page *page)
8834 struct extent_io_tree *tree;
8835 tree = &BTRFS_I(page->mapping->host)->io_tree;
8836 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8839 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8841 struct extent_io_tree *tree;
8842 struct inode *inode = page->mapping->host;
8845 if (current->flags & PF_MEMALLOC) {
8846 redirty_page_for_writepage(wbc, page);
8852 * If we are under memory pressure we will call this directly from the
8853 * VM, we need to make sure we have the inode referenced for the ordered
8854 * extent. If not just return like we didn't do anything.
8856 if (!igrab(inode)) {
8857 redirty_page_for_writepage(wbc, page);
8858 return AOP_WRITEPAGE_ACTIVATE;
8860 tree = &BTRFS_I(page->mapping->host)->io_tree;
8861 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8862 btrfs_add_delayed_iput(inode);
8866 static int btrfs_writepages(struct address_space *mapping,
8867 struct writeback_control *wbc)
8869 struct extent_io_tree *tree;
8871 tree = &BTRFS_I(mapping->host)->io_tree;
8872 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8876 btrfs_readpages(struct file *file, struct address_space *mapping,
8877 struct list_head *pages, unsigned nr_pages)
8879 struct extent_io_tree *tree;
8880 tree = &BTRFS_I(mapping->host)->io_tree;
8881 return extent_readpages(tree, mapping, pages, nr_pages,
8884 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8886 struct extent_io_tree *tree;
8887 struct extent_map_tree *map;
8890 tree = &BTRFS_I(page->mapping->host)->io_tree;
8891 map = &BTRFS_I(page->mapping->host)->extent_tree;
8892 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8894 ClearPagePrivate(page);
8895 set_page_private(page, 0);
8901 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8903 if (PageWriteback(page) || PageDirty(page))
8905 return __btrfs_releasepage(page, gfp_flags);
8908 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8909 unsigned int length)
8911 struct inode *inode = page->mapping->host;
8912 struct extent_io_tree *tree;
8913 struct btrfs_ordered_extent *ordered;
8914 struct extent_state *cached_state = NULL;
8915 u64 page_start = page_offset(page);
8916 u64 page_end = page_start + PAGE_SIZE - 1;
8919 int inode_evicting = inode->i_state & I_FREEING;
8922 * we have the page locked, so new writeback can't start,
8923 * and the dirty bit won't be cleared while we are here.
8925 * Wait for IO on this page so that we can safely clear
8926 * the PagePrivate2 bit and do ordered accounting
8928 wait_on_page_writeback(page);
8930 tree = &BTRFS_I(inode)->io_tree;
8932 btrfs_releasepage(page, GFP_NOFS);
8936 if (!inode_evicting)
8937 lock_extent_bits(tree, page_start, page_end, &cached_state);
8940 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8941 page_end - start + 1);
8943 end = min(page_end, ordered->file_offset + ordered->len - 1);
8945 * IO on this page will never be started, so we need
8946 * to account for any ordered extents now
8948 if (!inode_evicting)
8949 clear_extent_bit(tree, start, end,
8950 EXTENT_DIRTY | EXTENT_DELALLOC |
8951 EXTENT_DELALLOC_NEW |
8952 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8953 EXTENT_DEFRAG, 1, 0, &cached_state,
8956 * whoever cleared the private bit is responsible
8957 * for the finish_ordered_io
8959 if (TestClearPagePrivate2(page)) {
8960 struct btrfs_ordered_inode_tree *tree;
8963 tree = &BTRFS_I(inode)->ordered_tree;
8965 spin_lock_irq(&tree->lock);
8966 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8967 new_len = start - ordered->file_offset;
8968 if (new_len < ordered->truncated_len)
8969 ordered->truncated_len = new_len;
8970 spin_unlock_irq(&tree->lock);
8972 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8974 end - start + 1, 1))
8975 btrfs_finish_ordered_io(ordered);
8977 btrfs_put_ordered_extent(ordered);
8978 if (!inode_evicting) {
8979 cached_state = NULL;
8980 lock_extent_bits(tree, start, end,
8985 if (start < page_end)
8990 * Qgroup reserved space handler
8991 * Page here will be either
8992 * 1) Already written to disk
8993 * In this case, its reserved space is released from data rsv map
8994 * and will be freed by delayed_ref handler finally.
8995 * So even we call qgroup_free_data(), it won't decrease reserved
8997 * 2) Not written to disk
8998 * This means the reserved space should be freed here. However,
8999 * if a truncate invalidates the page (by clearing PageDirty)
9000 * and the page is accounted for while allocating extent
9001 * in btrfs_check_data_free_space() we let delayed_ref to
9002 * free the entire extent.
9004 if (PageDirty(page))
9005 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
9006 if (!inode_evicting) {
9007 clear_extent_bit(tree, page_start, page_end,
9008 EXTENT_LOCKED | EXTENT_DIRTY |
9009 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
9010 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
9011 &cached_state, GFP_NOFS);
9013 __btrfs_releasepage(page, GFP_NOFS);
9016 ClearPageChecked(page);
9017 if (PagePrivate(page)) {
9018 ClearPagePrivate(page);
9019 set_page_private(page, 0);
9025 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
9026 * called from a page fault handler when a page is first dirtied. Hence we must
9027 * be careful to check for EOF conditions here. We set the page up correctly
9028 * for a written page which means we get ENOSPC checking when writing into
9029 * holes and correct delalloc and unwritten extent mapping on filesystems that
9030 * support these features.
9032 * We are not allowed to take the i_mutex here so we have to play games to
9033 * protect against truncate races as the page could now be beyond EOF. Because
9034 * vmtruncate() writes the inode size before removing pages, once we have the
9035 * page lock we can determine safely if the page is beyond EOF. If it is not
9036 * beyond EOF, then the page is guaranteed safe against truncation until we
9039 int btrfs_page_mkwrite(struct vm_fault *vmf)
9041 struct page *page = vmf->page;
9042 struct inode *inode = file_inode(vmf->vma->vm_file);
9043 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9044 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
9045 struct btrfs_ordered_extent *ordered;
9046 struct extent_state *cached_state = NULL;
9048 unsigned long zero_start;
9057 reserved_space = PAGE_SIZE;
9059 sb_start_pagefault(inode->i_sb);
9060 page_start = page_offset(page);
9061 page_end = page_start + PAGE_SIZE - 1;
9065 * Reserving delalloc space after obtaining the page lock can lead to
9066 * deadlock. For example, if a dirty page is locked by this function
9067 * and the call to btrfs_delalloc_reserve_space() ends up triggering
9068 * dirty page write out, then the btrfs_writepage() function could
9069 * end up waiting indefinitely to get a lock on the page currently
9070 * being processed by btrfs_page_mkwrite() function.
9072 ret = btrfs_delalloc_reserve_space(inode, page_start,
9075 ret = file_update_time(vmf->vma->vm_file);
9081 else /* -ENOSPC, -EIO, etc */
9082 ret = VM_FAULT_SIGBUS;
9088 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
9091 size = i_size_read(inode);
9093 if ((page->mapping != inode->i_mapping) ||
9094 (page_start >= size)) {
9095 /* page got truncated out from underneath us */
9098 wait_on_page_writeback(page);
9100 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
9101 set_page_extent_mapped(page);
9104 * we can't set the delalloc bits if there are pending ordered
9105 * extents. Drop our locks and wait for them to finish
9107 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
9110 unlock_extent_cached(io_tree, page_start, page_end,
9111 &cached_state, GFP_NOFS);
9113 btrfs_start_ordered_extent(inode, ordered, 1);
9114 btrfs_put_ordered_extent(ordered);
9118 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
9119 reserved_space = round_up(size - page_start,
9120 fs_info->sectorsize);
9121 if (reserved_space < PAGE_SIZE) {
9122 end = page_start + reserved_space - 1;
9123 spin_lock(&BTRFS_I(inode)->lock);
9124 BTRFS_I(inode)->outstanding_extents++;
9125 spin_unlock(&BTRFS_I(inode)->lock);
9126 btrfs_delalloc_release_space(inode, page_start,
9127 PAGE_SIZE - reserved_space);
9132 * page_mkwrite gets called when the page is firstly dirtied after it's
9133 * faulted in, but write(2) could also dirty a page and set delalloc
9134 * bits, thus in this case for space account reason, we still need to
9135 * clear any delalloc bits within this page range since we have to
9136 * reserve data&meta space before lock_page() (see above comments).
9138 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
9139 EXTENT_DIRTY | EXTENT_DELALLOC |
9140 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
9141 0, 0, &cached_state, GFP_NOFS);
9143 ret = btrfs_set_extent_delalloc(inode, page_start, end,
9146 unlock_extent_cached(io_tree, page_start, page_end,
9147 &cached_state, GFP_NOFS);
9148 ret = VM_FAULT_SIGBUS;
9153 /* page is wholly or partially inside EOF */
9154 if (page_start + PAGE_SIZE > size)
9155 zero_start = size & ~PAGE_MASK;
9157 zero_start = PAGE_SIZE;
9159 if (zero_start != PAGE_SIZE) {
9161 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9162 flush_dcache_page(page);
9165 ClearPageChecked(page);
9166 set_page_dirty(page);
9167 SetPageUptodate(page);
9169 BTRFS_I(inode)->last_trans = fs_info->generation;
9170 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9171 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9173 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9177 sb_end_pagefault(inode->i_sb);
9178 return VM_FAULT_LOCKED;
9182 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9184 sb_end_pagefault(inode->i_sb);
9188 static int btrfs_truncate(struct inode *inode)
9190 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9191 struct btrfs_root *root = BTRFS_I(inode)->root;
9192 struct btrfs_block_rsv *rsv;
9195 struct btrfs_trans_handle *trans;
9196 u64 mask = fs_info->sectorsize - 1;
9197 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9199 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9205 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9206 * 3 things going on here
9208 * 1) We need to reserve space for our orphan item and the space to
9209 * delete our orphan item. Lord knows we don't want to have a dangling
9210 * orphan item because we didn't reserve space to remove it.
9212 * 2) We need to reserve space to update our inode.
9214 * 3) We need to have something to cache all the space that is going to
9215 * be free'd up by the truncate operation, but also have some slack
9216 * space reserved in case it uses space during the truncate (thank you
9217 * very much snapshotting).
9219 * And we need these to all be separate. The fact is we can use a lot of
9220 * space doing the truncate, and we have no earthly idea how much space
9221 * we will use, so we need the truncate reservation to be separate so it
9222 * doesn't end up using space reserved for updating the inode or
9223 * removing the orphan item. We also need to be able to stop the
9224 * transaction and start a new one, which means we need to be able to
9225 * update the inode several times, and we have no idea of knowing how
9226 * many times that will be, so we can't just reserve 1 item for the
9227 * entirety of the operation, so that has to be done separately as well.
9228 * Then there is the orphan item, which does indeed need to be held on
9229 * to for the whole operation, and we need nobody to touch this reserved
9230 * space except the orphan code.
9232 * So that leaves us with
9234 * 1) root->orphan_block_rsv - for the orphan deletion.
9235 * 2) rsv - for the truncate reservation, which we will steal from the
9236 * transaction reservation.
9237 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9238 * updating the inode.
9240 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9243 rsv->size = min_size;
9247 * 1 for the truncate slack space
9248 * 1 for updating the inode.
9250 trans = btrfs_start_transaction(root, 2);
9251 if (IS_ERR(trans)) {
9252 err = PTR_ERR(trans);
9256 /* Migrate the slack space for the truncate to our reserve */
9257 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9262 * So if we truncate and then write and fsync we normally would just
9263 * write the extents that changed, which is a problem if we need to
9264 * first truncate that entire inode. So set this flag so we write out
9265 * all of the extents in the inode to the sync log so we're completely
9268 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9269 trans->block_rsv = rsv;
9272 ret = btrfs_truncate_inode_items(trans, root, inode,
9274 BTRFS_EXTENT_DATA_KEY);
9275 if (ret != -ENOSPC && ret != -EAGAIN) {
9280 trans->block_rsv = &fs_info->trans_block_rsv;
9281 ret = btrfs_update_inode(trans, root, inode);
9287 btrfs_end_transaction(trans);
9288 btrfs_btree_balance_dirty(fs_info);
9290 trans = btrfs_start_transaction(root, 2);
9291 if (IS_ERR(trans)) {
9292 ret = err = PTR_ERR(trans);
9297 btrfs_block_rsv_release(fs_info, rsv, -1);
9298 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9300 BUG_ON(ret); /* shouldn't happen */
9301 trans->block_rsv = rsv;
9304 if (ret == 0 && inode->i_nlink > 0) {
9305 trans->block_rsv = root->orphan_block_rsv;
9306 ret = btrfs_orphan_del(trans, BTRFS_I(inode));
9312 trans->block_rsv = &fs_info->trans_block_rsv;
9313 ret = btrfs_update_inode(trans, root, inode);
9317 ret = btrfs_end_transaction(trans);
9318 btrfs_btree_balance_dirty(fs_info);
9321 btrfs_free_block_rsv(fs_info, rsv);
9330 * create a new subvolume directory/inode (helper for the ioctl).
9332 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9333 struct btrfs_root *new_root,
9334 struct btrfs_root *parent_root,
9337 struct inode *inode;
9341 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9342 new_dirid, new_dirid,
9343 S_IFDIR | (~current_umask() & S_IRWXUGO),
9346 return PTR_ERR(inode);
9347 inode->i_op = &btrfs_dir_inode_operations;
9348 inode->i_fop = &btrfs_dir_file_operations;
9350 set_nlink(inode, 1);
9351 btrfs_i_size_write(BTRFS_I(inode), 0);
9352 unlock_new_inode(inode);
9354 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9356 btrfs_err(new_root->fs_info,
9357 "error inheriting subvolume %llu properties: %d",
9358 new_root->root_key.objectid, err);
9360 err = btrfs_update_inode(trans, new_root, inode);
9366 struct inode *btrfs_alloc_inode(struct super_block *sb)
9368 struct btrfs_inode *ei;
9369 struct inode *inode;
9371 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9378 ei->last_sub_trans = 0;
9379 ei->logged_trans = 0;
9380 ei->delalloc_bytes = 0;
9381 ei->new_delalloc_bytes = 0;
9382 ei->defrag_bytes = 0;
9383 ei->disk_i_size = 0;
9386 ei->index_cnt = (u64)-1;
9388 ei->last_unlink_trans = 0;
9389 ei->last_log_commit = 0;
9390 ei->delayed_iput_count = 0;
9392 spin_lock_init(&ei->lock);
9393 ei->outstanding_extents = 0;
9394 ei->reserved_extents = 0;
9396 ei->runtime_flags = 0;
9397 ei->force_compress = BTRFS_COMPRESS_NONE;
9399 ei->delayed_node = NULL;
9401 ei->i_otime.tv_sec = 0;
9402 ei->i_otime.tv_nsec = 0;
9404 inode = &ei->vfs_inode;
9405 extent_map_tree_init(&ei->extent_tree);
9406 extent_io_tree_init(&ei->io_tree, inode);
9407 extent_io_tree_init(&ei->io_failure_tree, inode);
9408 ei->io_tree.track_uptodate = 1;
9409 ei->io_failure_tree.track_uptodate = 1;
9410 atomic_set(&ei->sync_writers, 0);
9411 mutex_init(&ei->log_mutex);
9412 mutex_init(&ei->delalloc_mutex);
9413 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9414 INIT_LIST_HEAD(&ei->delalloc_inodes);
9415 INIT_LIST_HEAD(&ei->delayed_iput);
9416 RB_CLEAR_NODE(&ei->rb_node);
9417 init_rwsem(&ei->dio_sem);
9422 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9423 void btrfs_test_destroy_inode(struct inode *inode)
9425 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9426 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9430 static void btrfs_i_callback(struct rcu_head *head)
9432 struct inode *inode = container_of(head, struct inode, i_rcu);
9433 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9436 void btrfs_destroy_inode(struct inode *inode)
9438 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9439 struct btrfs_ordered_extent *ordered;
9440 struct btrfs_root *root = BTRFS_I(inode)->root;
9442 WARN_ON(!hlist_empty(&inode->i_dentry));
9443 WARN_ON(inode->i_data.nrpages);
9444 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9445 WARN_ON(BTRFS_I(inode)->reserved_extents);
9446 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9447 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9448 WARN_ON(BTRFS_I(inode)->csum_bytes);
9449 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9452 * This can happen where we create an inode, but somebody else also
9453 * created the same inode and we need to destroy the one we already
9459 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9460 &BTRFS_I(inode)->runtime_flags)) {
9461 btrfs_info(fs_info, "inode %llu still on the orphan list",
9462 btrfs_ino(BTRFS_I(inode)));
9463 atomic_dec(&root->orphan_inodes);
9467 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9472 "found ordered extent %llu %llu on inode cleanup",
9473 ordered->file_offset, ordered->len);
9474 btrfs_remove_ordered_extent(inode, ordered);
9475 btrfs_put_ordered_extent(ordered);
9476 btrfs_put_ordered_extent(ordered);
9479 btrfs_qgroup_check_reserved_leak(inode);
9480 inode_tree_del(inode);
9481 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9483 call_rcu(&inode->i_rcu, btrfs_i_callback);
9486 int btrfs_drop_inode(struct inode *inode)
9488 struct btrfs_root *root = BTRFS_I(inode)->root;
9493 /* the snap/subvol tree is on deleting */
9494 if (btrfs_root_refs(&root->root_item) == 0)
9497 return generic_drop_inode(inode);
9500 static void init_once(void *foo)
9502 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9504 inode_init_once(&ei->vfs_inode);
9507 void btrfs_destroy_cachep(void)
9510 * Make sure all delayed rcu free inodes are flushed before we
9514 kmem_cache_destroy(btrfs_inode_cachep);
9515 kmem_cache_destroy(btrfs_trans_handle_cachep);
9516 kmem_cache_destroy(btrfs_transaction_cachep);
9517 kmem_cache_destroy(btrfs_path_cachep);
9518 kmem_cache_destroy(btrfs_free_space_cachep);
9521 int btrfs_init_cachep(void)
9523 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9524 sizeof(struct btrfs_inode), 0,
9525 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9527 if (!btrfs_inode_cachep)
9530 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9531 sizeof(struct btrfs_trans_handle), 0,
9532 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9533 if (!btrfs_trans_handle_cachep)
9536 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9537 sizeof(struct btrfs_transaction), 0,
9538 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9539 if (!btrfs_transaction_cachep)
9542 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9543 sizeof(struct btrfs_path), 0,
9544 SLAB_MEM_SPREAD, NULL);
9545 if (!btrfs_path_cachep)
9548 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9549 sizeof(struct btrfs_free_space), 0,
9550 SLAB_MEM_SPREAD, NULL);
9551 if (!btrfs_free_space_cachep)
9556 btrfs_destroy_cachep();
9560 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9561 u32 request_mask, unsigned int flags)
9564 struct inode *inode = d_inode(path->dentry);
9565 u32 blocksize = inode->i_sb->s_blocksize;
9567 generic_fillattr(inode, stat);
9568 stat->dev = BTRFS_I(inode)->root->anon_dev;
9570 spin_lock(&BTRFS_I(inode)->lock);
9571 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9572 spin_unlock(&BTRFS_I(inode)->lock);
9573 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9574 ALIGN(delalloc_bytes, blocksize)) >> 9;
9578 static int btrfs_rename_exchange(struct inode *old_dir,
9579 struct dentry *old_dentry,
9580 struct inode *new_dir,
9581 struct dentry *new_dentry)
9583 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9584 struct btrfs_trans_handle *trans;
9585 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9586 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9587 struct inode *new_inode = new_dentry->d_inode;
9588 struct inode *old_inode = old_dentry->d_inode;
9589 struct timespec ctime = current_time(old_inode);
9590 struct dentry *parent;
9591 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9592 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9597 bool root_log_pinned = false;
9598 bool dest_log_pinned = false;
9600 /* we only allow rename subvolume link between subvolumes */
9601 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9604 /* close the race window with snapshot create/destroy ioctl */
9605 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9606 down_read(&fs_info->subvol_sem);
9607 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9608 down_read(&fs_info->subvol_sem);
9611 * We want to reserve the absolute worst case amount of items. So if
9612 * both inodes are subvols and we need to unlink them then that would
9613 * require 4 item modifications, but if they are both normal inodes it
9614 * would require 5 item modifications, so we'll assume their normal
9615 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9616 * should cover the worst case number of items we'll modify.
9618 trans = btrfs_start_transaction(root, 12);
9619 if (IS_ERR(trans)) {
9620 ret = PTR_ERR(trans);
9625 * We need to find a free sequence number both in the source and
9626 * in the destination directory for the exchange.
9628 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9631 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9635 BTRFS_I(old_inode)->dir_index = 0ULL;
9636 BTRFS_I(new_inode)->dir_index = 0ULL;
9638 /* Reference for the source. */
9639 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9640 /* force full log commit if subvolume involved. */
9641 btrfs_set_log_full_commit(fs_info, trans);
9643 btrfs_pin_log_trans(root);
9644 root_log_pinned = true;
9645 ret = btrfs_insert_inode_ref(trans, dest,
9646 new_dentry->d_name.name,
9647 new_dentry->d_name.len,
9649 btrfs_ino(BTRFS_I(new_dir)),
9655 /* And now for the dest. */
9656 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9657 /* force full log commit if subvolume involved. */
9658 btrfs_set_log_full_commit(fs_info, trans);
9660 btrfs_pin_log_trans(dest);
9661 dest_log_pinned = true;
9662 ret = btrfs_insert_inode_ref(trans, root,
9663 old_dentry->d_name.name,
9664 old_dentry->d_name.len,
9666 btrfs_ino(BTRFS_I(old_dir)),
9672 /* Update inode version and ctime/mtime. */
9673 inode_inc_iversion(old_dir);
9674 inode_inc_iversion(new_dir);
9675 inode_inc_iversion(old_inode);
9676 inode_inc_iversion(new_inode);
9677 old_dir->i_ctime = old_dir->i_mtime = ctime;
9678 new_dir->i_ctime = new_dir->i_mtime = ctime;
9679 old_inode->i_ctime = ctime;
9680 new_inode->i_ctime = ctime;
9682 if (old_dentry->d_parent != new_dentry->d_parent) {
9683 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9684 BTRFS_I(old_inode), 1);
9685 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9686 BTRFS_I(new_inode), 1);
9689 /* src is a subvolume */
9690 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9691 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9692 ret = btrfs_unlink_subvol(trans, root, old_dir,
9694 old_dentry->d_name.name,
9695 old_dentry->d_name.len);
9696 } else { /* src is an inode */
9697 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9698 BTRFS_I(old_dentry->d_inode),
9699 old_dentry->d_name.name,
9700 old_dentry->d_name.len);
9702 ret = btrfs_update_inode(trans, root, old_inode);
9705 btrfs_abort_transaction(trans, ret);
9709 /* dest is a subvolume */
9710 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9711 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9712 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9714 new_dentry->d_name.name,
9715 new_dentry->d_name.len);
9716 } else { /* dest is an inode */
9717 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9718 BTRFS_I(new_dentry->d_inode),
9719 new_dentry->d_name.name,
9720 new_dentry->d_name.len);
9722 ret = btrfs_update_inode(trans, dest, new_inode);
9725 btrfs_abort_transaction(trans, ret);
9729 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9730 new_dentry->d_name.name,
9731 new_dentry->d_name.len, 0, old_idx);
9733 btrfs_abort_transaction(trans, ret);
9737 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9738 old_dentry->d_name.name,
9739 old_dentry->d_name.len, 0, new_idx);
9741 btrfs_abort_transaction(trans, ret);
9745 if (old_inode->i_nlink == 1)
9746 BTRFS_I(old_inode)->dir_index = old_idx;
9747 if (new_inode->i_nlink == 1)
9748 BTRFS_I(new_inode)->dir_index = new_idx;
9750 if (root_log_pinned) {
9751 parent = new_dentry->d_parent;
9752 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9754 btrfs_end_log_trans(root);
9755 root_log_pinned = false;
9757 if (dest_log_pinned) {
9758 parent = old_dentry->d_parent;
9759 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9761 btrfs_end_log_trans(dest);
9762 dest_log_pinned = false;
9766 * If we have pinned a log and an error happened, we unpin tasks
9767 * trying to sync the log and force them to fallback to a transaction
9768 * commit if the log currently contains any of the inodes involved in
9769 * this rename operation (to ensure we do not persist a log with an
9770 * inconsistent state for any of these inodes or leading to any
9771 * inconsistencies when replayed). If the transaction was aborted, the
9772 * abortion reason is propagated to userspace when attempting to commit
9773 * the transaction. If the log does not contain any of these inodes, we
9774 * allow the tasks to sync it.
9776 if (ret && (root_log_pinned || dest_log_pinned)) {
9777 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9778 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9779 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9781 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9782 btrfs_set_log_full_commit(fs_info, trans);
9784 if (root_log_pinned) {
9785 btrfs_end_log_trans(root);
9786 root_log_pinned = false;
9788 if (dest_log_pinned) {
9789 btrfs_end_log_trans(dest);
9790 dest_log_pinned = false;
9793 ret = btrfs_end_transaction(trans);
9795 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9796 up_read(&fs_info->subvol_sem);
9797 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9798 up_read(&fs_info->subvol_sem);
9803 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9804 struct btrfs_root *root,
9806 struct dentry *dentry)
9809 struct inode *inode;
9813 ret = btrfs_find_free_ino(root, &objectid);
9817 inode = btrfs_new_inode(trans, root, dir,
9818 dentry->d_name.name,
9820 btrfs_ino(BTRFS_I(dir)),
9822 S_IFCHR | WHITEOUT_MODE,
9825 if (IS_ERR(inode)) {
9826 ret = PTR_ERR(inode);
9830 inode->i_op = &btrfs_special_inode_operations;
9831 init_special_inode(inode, inode->i_mode,
9834 ret = btrfs_init_inode_security(trans, inode, dir,
9839 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9840 BTRFS_I(inode), 0, index);
9844 ret = btrfs_update_inode(trans, root, inode);
9846 unlock_new_inode(inode);
9848 inode_dec_link_count(inode);
9854 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9855 struct inode *new_dir, struct dentry *new_dentry,
9858 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9859 struct btrfs_trans_handle *trans;
9860 unsigned int trans_num_items;
9861 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9862 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9863 struct inode *new_inode = d_inode(new_dentry);
9864 struct inode *old_inode = d_inode(old_dentry);
9868 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9869 bool log_pinned = false;
9871 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9874 /* we only allow rename subvolume link between subvolumes */
9875 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9878 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9879 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9882 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9883 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9887 /* check for collisions, even if the name isn't there */
9888 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9889 new_dentry->d_name.name,
9890 new_dentry->d_name.len);
9893 if (ret == -EEXIST) {
9895 * eexist without a new_inode */
9896 if (WARN_ON(!new_inode)) {
9900 /* maybe -EOVERFLOW */
9907 * we're using rename to replace one file with another. Start IO on it
9908 * now so we don't add too much work to the end of the transaction
9910 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9911 filemap_flush(old_inode->i_mapping);
9913 /* close the racy window with snapshot create/destroy ioctl */
9914 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9915 down_read(&fs_info->subvol_sem);
9917 * We want to reserve the absolute worst case amount of items. So if
9918 * both inodes are subvols and we need to unlink them then that would
9919 * require 4 item modifications, but if they are both normal inodes it
9920 * would require 5 item modifications, so we'll assume they are normal
9921 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9922 * should cover the worst case number of items we'll modify.
9923 * If our rename has the whiteout flag, we need more 5 units for the
9924 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9925 * when selinux is enabled).
9927 trans_num_items = 11;
9928 if (flags & RENAME_WHITEOUT)
9929 trans_num_items += 5;
9930 trans = btrfs_start_transaction(root, trans_num_items);
9931 if (IS_ERR(trans)) {
9932 ret = PTR_ERR(trans);
9937 btrfs_record_root_in_trans(trans, dest);
9939 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9943 BTRFS_I(old_inode)->dir_index = 0ULL;
9944 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9945 /* force full log commit if subvolume involved. */
9946 btrfs_set_log_full_commit(fs_info, trans);
9948 btrfs_pin_log_trans(root);
9950 ret = btrfs_insert_inode_ref(trans, dest,
9951 new_dentry->d_name.name,
9952 new_dentry->d_name.len,
9954 btrfs_ino(BTRFS_I(new_dir)), index);
9959 inode_inc_iversion(old_dir);
9960 inode_inc_iversion(new_dir);
9961 inode_inc_iversion(old_inode);
9962 old_dir->i_ctime = old_dir->i_mtime =
9963 new_dir->i_ctime = new_dir->i_mtime =
9964 old_inode->i_ctime = current_time(old_dir);
9966 if (old_dentry->d_parent != new_dentry->d_parent)
9967 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9968 BTRFS_I(old_inode), 1);
9970 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9971 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9972 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9973 old_dentry->d_name.name,
9974 old_dentry->d_name.len);
9976 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9977 BTRFS_I(d_inode(old_dentry)),
9978 old_dentry->d_name.name,
9979 old_dentry->d_name.len);
9981 ret = btrfs_update_inode(trans, root, old_inode);
9984 btrfs_abort_transaction(trans, ret);
9989 inode_inc_iversion(new_inode);
9990 new_inode->i_ctime = current_time(new_inode);
9991 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9992 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9993 root_objectid = BTRFS_I(new_inode)->location.objectid;
9994 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9996 new_dentry->d_name.name,
9997 new_dentry->d_name.len);
9998 BUG_ON(new_inode->i_nlink == 0);
10000 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
10001 BTRFS_I(d_inode(new_dentry)),
10002 new_dentry->d_name.name,
10003 new_dentry->d_name.len);
10005 if (!ret && new_inode->i_nlink == 0)
10006 ret = btrfs_orphan_add(trans,
10007 BTRFS_I(d_inode(new_dentry)));
10009 btrfs_abort_transaction(trans, ret);
10014 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
10015 new_dentry->d_name.name,
10016 new_dentry->d_name.len, 0, index);
10018 btrfs_abort_transaction(trans, ret);
10022 if (old_inode->i_nlink == 1)
10023 BTRFS_I(old_inode)->dir_index = index;
10026 struct dentry *parent = new_dentry->d_parent;
10028 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
10030 btrfs_end_log_trans(root);
10031 log_pinned = false;
10034 if (flags & RENAME_WHITEOUT) {
10035 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
10039 btrfs_abort_transaction(trans, ret);
10045 * If we have pinned the log and an error happened, we unpin tasks
10046 * trying to sync the log and force them to fallback to a transaction
10047 * commit if the log currently contains any of the inodes involved in
10048 * this rename operation (to ensure we do not persist a log with an
10049 * inconsistent state for any of these inodes or leading to any
10050 * inconsistencies when replayed). If the transaction was aborted, the
10051 * abortion reason is propagated to userspace when attempting to commit
10052 * the transaction. If the log does not contain any of these inodes, we
10053 * allow the tasks to sync it.
10055 if (ret && log_pinned) {
10056 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
10057 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
10058 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
10060 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
10061 btrfs_set_log_full_commit(fs_info, trans);
10063 btrfs_end_log_trans(root);
10064 log_pinned = false;
10066 btrfs_end_transaction(trans);
10068 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
10069 up_read(&fs_info->subvol_sem);
10074 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
10075 struct inode *new_dir, struct dentry *new_dentry,
10076 unsigned int flags)
10078 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
10081 if (flags & RENAME_EXCHANGE)
10082 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
10085 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
10088 static void btrfs_run_delalloc_work(struct btrfs_work *work)
10090 struct btrfs_delalloc_work *delalloc_work;
10091 struct inode *inode;
10093 delalloc_work = container_of(work, struct btrfs_delalloc_work,
10095 inode = delalloc_work->inode;
10096 filemap_flush(inode->i_mapping);
10097 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
10098 &BTRFS_I(inode)->runtime_flags))
10099 filemap_flush(inode->i_mapping);
10101 if (delalloc_work->delay_iput)
10102 btrfs_add_delayed_iput(inode);
10105 complete(&delalloc_work->completion);
10108 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
10111 struct btrfs_delalloc_work *work;
10113 work = kmalloc(sizeof(*work), GFP_NOFS);
10117 init_completion(&work->completion);
10118 INIT_LIST_HEAD(&work->list);
10119 work->inode = inode;
10120 work->delay_iput = delay_iput;
10121 WARN_ON_ONCE(!inode);
10122 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
10123 btrfs_run_delalloc_work, NULL, NULL);
10128 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
10130 wait_for_completion(&work->completion);
10135 * some fairly slow code that needs optimization. This walks the list
10136 * of all the inodes with pending delalloc and forces them to disk.
10138 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
10141 struct btrfs_inode *binode;
10142 struct inode *inode;
10143 struct btrfs_delalloc_work *work, *next;
10144 struct list_head works;
10145 struct list_head splice;
10148 INIT_LIST_HEAD(&works);
10149 INIT_LIST_HEAD(&splice);
10151 mutex_lock(&root->delalloc_mutex);
10152 spin_lock(&root->delalloc_lock);
10153 list_splice_init(&root->delalloc_inodes, &splice);
10154 while (!list_empty(&splice)) {
10155 binode = list_entry(splice.next, struct btrfs_inode,
10158 list_move_tail(&binode->delalloc_inodes,
10159 &root->delalloc_inodes);
10160 inode = igrab(&binode->vfs_inode);
10162 cond_resched_lock(&root->delalloc_lock);
10165 spin_unlock(&root->delalloc_lock);
10167 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10170 btrfs_add_delayed_iput(inode);
10176 list_add_tail(&work->list, &works);
10177 btrfs_queue_work(root->fs_info->flush_workers,
10180 if (nr != -1 && ret >= nr)
10183 spin_lock(&root->delalloc_lock);
10185 spin_unlock(&root->delalloc_lock);
10188 list_for_each_entry_safe(work, next, &works, list) {
10189 list_del_init(&work->list);
10190 btrfs_wait_and_free_delalloc_work(work);
10193 if (!list_empty_careful(&splice)) {
10194 spin_lock(&root->delalloc_lock);
10195 list_splice_tail(&splice, &root->delalloc_inodes);
10196 spin_unlock(&root->delalloc_lock);
10198 mutex_unlock(&root->delalloc_mutex);
10202 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10204 struct btrfs_fs_info *fs_info = root->fs_info;
10207 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10210 ret = __start_delalloc_inodes(root, delay_iput, -1);
10214 * the filemap_flush will queue IO into the worker threads, but
10215 * we have to make sure the IO is actually started and that
10216 * ordered extents get created before we return
10218 atomic_inc(&fs_info->async_submit_draining);
10219 while (atomic_read(&fs_info->nr_async_submits) ||
10220 atomic_read(&fs_info->async_delalloc_pages)) {
10221 wait_event(fs_info->async_submit_wait,
10222 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10223 atomic_read(&fs_info->async_delalloc_pages) == 0));
10225 atomic_dec(&fs_info->async_submit_draining);
10229 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10232 struct btrfs_root *root;
10233 struct list_head splice;
10236 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10239 INIT_LIST_HEAD(&splice);
10241 mutex_lock(&fs_info->delalloc_root_mutex);
10242 spin_lock(&fs_info->delalloc_root_lock);
10243 list_splice_init(&fs_info->delalloc_roots, &splice);
10244 while (!list_empty(&splice) && nr) {
10245 root = list_first_entry(&splice, struct btrfs_root,
10247 root = btrfs_grab_fs_root(root);
10249 list_move_tail(&root->delalloc_root,
10250 &fs_info->delalloc_roots);
10251 spin_unlock(&fs_info->delalloc_root_lock);
10253 ret = __start_delalloc_inodes(root, delay_iput, nr);
10254 btrfs_put_fs_root(root);
10262 spin_lock(&fs_info->delalloc_root_lock);
10264 spin_unlock(&fs_info->delalloc_root_lock);
10267 atomic_inc(&fs_info->async_submit_draining);
10268 while (atomic_read(&fs_info->nr_async_submits) ||
10269 atomic_read(&fs_info->async_delalloc_pages)) {
10270 wait_event(fs_info->async_submit_wait,
10271 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10272 atomic_read(&fs_info->async_delalloc_pages) == 0));
10274 atomic_dec(&fs_info->async_submit_draining);
10276 if (!list_empty_careful(&splice)) {
10277 spin_lock(&fs_info->delalloc_root_lock);
10278 list_splice_tail(&splice, &fs_info->delalloc_roots);
10279 spin_unlock(&fs_info->delalloc_root_lock);
10281 mutex_unlock(&fs_info->delalloc_root_mutex);
10285 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10286 const char *symname)
10288 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10289 struct btrfs_trans_handle *trans;
10290 struct btrfs_root *root = BTRFS_I(dir)->root;
10291 struct btrfs_path *path;
10292 struct btrfs_key key;
10293 struct inode *inode = NULL;
10295 int drop_inode = 0;
10301 struct btrfs_file_extent_item *ei;
10302 struct extent_buffer *leaf;
10304 name_len = strlen(symname);
10305 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10306 return -ENAMETOOLONG;
10309 * 2 items for inode item and ref
10310 * 2 items for dir items
10311 * 1 item for updating parent inode item
10312 * 1 item for the inline extent item
10313 * 1 item for xattr if selinux is on
10315 trans = btrfs_start_transaction(root, 7);
10317 return PTR_ERR(trans);
10319 err = btrfs_find_free_ino(root, &objectid);
10323 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10324 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10325 objectid, S_IFLNK|S_IRWXUGO, &index);
10326 if (IS_ERR(inode)) {
10327 err = PTR_ERR(inode);
10332 * If the active LSM wants to access the inode during
10333 * d_instantiate it needs these. Smack checks to see
10334 * if the filesystem supports xattrs by looking at the
10337 inode->i_fop = &btrfs_file_operations;
10338 inode->i_op = &btrfs_file_inode_operations;
10339 inode->i_mapping->a_ops = &btrfs_aops;
10340 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10342 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10344 goto out_unlock_inode;
10346 path = btrfs_alloc_path();
10349 goto out_unlock_inode;
10351 key.objectid = btrfs_ino(BTRFS_I(inode));
10353 key.type = BTRFS_EXTENT_DATA_KEY;
10354 datasize = btrfs_file_extent_calc_inline_size(name_len);
10355 err = btrfs_insert_empty_item(trans, root, path, &key,
10358 btrfs_free_path(path);
10359 goto out_unlock_inode;
10361 leaf = path->nodes[0];
10362 ei = btrfs_item_ptr(leaf, path->slots[0],
10363 struct btrfs_file_extent_item);
10364 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10365 btrfs_set_file_extent_type(leaf, ei,
10366 BTRFS_FILE_EXTENT_INLINE);
10367 btrfs_set_file_extent_encryption(leaf, ei, 0);
10368 btrfs_set_file_extent_compression(leaf, ei, 0);
10369 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10370 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10372 ptr = btrfs_file_extent_inline_start(ei);
10373 write_extent_buffer(leaf, symname, ptr, name_len);
10374 btrfs_mark_buffer_dirty(leaf);
10375 btrfs_free_path(path);
10377 inode->i_op = &btrfs_symlink_inode_operations;
10378 inode_nohighmem(inode);
10379 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10380 inode_set_bytes(inode, name_len);
10381 btrfs_i_size_write(BTRFS_I(inode), name_len);
10382 err = btrfs_update_inode(trans, root, inode);
10384 * Last step, add directory indexes for our symlink inode. This is the
10385 * last step to avoid extra cleanup of these indexes if an error happens
10389 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10390 BTRFS_I(inode), 0, index);
10393 goto out_unlock_inode;
10396 unlock_new_inode(inode);
10397 d_instantiate(dentry, inode);
10400 btrfs_end_transaction(trans);
10402 inode_dec_link_count(inode);
10405 btrfs_btree_balance_dirty(fs_info);
10410 unlock_new_inode(inode);
10414 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10415 u64 start, u64 num_bytes, u64 min_size,
10416 loff_t actual_len, u64 *alloc_hint,
10417 struct btrfs_trans_handle *trans)
10419 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10420 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10421 struct extent_map *em;
10422 struct btrfs_root *root = BTRFS_I(inode)->root;
10423 struct btrfs_key ins;
10424 u64 cur_offset = start;
10427 u64 last_alloc = (u64)-1;
10429 bool own_trans = true;
10430 u64 end = start + num_bytes - 1;
10434 while (num_bytes > 0) {
10436 trans = btrfs_start_transaction(root, 3);
10437 if (IS_ERR(trans)) {
10438 ret = PTR_ERR(trans);
10443 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10444 cur_bytes = max(cur_bytes, min_size);
10446 * If we are severely fragmented we could end up with really
10447 * small allocations, so if the allocator is returning small
10448 * chunks lets make its job easier by only searching for those
10451 cur_bytes = min(cur_bytes, last_alloc);
10452 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10453 min_size, 0, *alloc_hint, &ins, 1, 0);
10456 btrfs_end_transaction(trans);
10459 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10461 last_alloc = ins.offset;
10462 ret = insert_reserved_file_extent(trans, inode,
10463 cur_offset, ins.objectid,
10464 ins.offset, ins.offset,
10465 ins.offset, 0, 0, 0,
10466 BTRFS_FILE_EXTENT_PREALLOC);
10468 btrfs_free_reserved_extent(fs_info, ins.objectid,
10470 btrfs_abort_transaction(trans, ret);
10472 btrfs_end_transaction(trans);
10476 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10477 cur_offset + ins.offset -1, 0);
10479 em = alloc_extent_map();
10481 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10482 &BTRFS_I(inode)->runtime_flags);
10486 em->start = cur_offset;
10487 em->orig_start = cur_offset;
10488 em->len = ins.offset;
10489 em->block_start = ins.objectid;
10490 em->block_len = ins.offset;
10491 em->orig_block_len = ins.offset;
10492 em->ram_bytes = ins.offset;
10493 em->bdev = fs_info->fs_devices->latest_bdev;
10494 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10495 em->generation = trans->transid;
10498 write_lock(&em_tree->lock);
10499 ret = add_extent_mapping(em_tree, em, 1);
10500 write_unlock(&em_tree->lock);
10501 if (ret != -EEXIST)
10503 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10504 cur_offset + ins.offset - 1,
10507 free_extent_map(em);
10509 num_bytes -= ins.offset;
10510 cur_offset += ins.offset;
10511 *alloc_hint = ins.objectid + ins.offset;
10513 inode_inc_iversion(inode);
10514 inode->i_ctime = current_time(inode);
10515 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10516 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10517 (actual_len > inode->i_size) &&
10518 (cur_offset > inode->i_size)) {
10519 if (cur_offset > actual_len)
10520 i_size = actual_len;
10522 i_size = cur_offset;
10523 i_size_write(inode, i_size);
10524 btrfs_ordered_update_i_size(inode, i_size, NULL);
10527 ret = btrfs_update_inode(trans, root, inode);
10530 btrfs_abort_transaction(trans, ret);
10532 btrfs_end_transaction(trans);
10537 btrfs_end_transaction(trans);
10539 if (cur_offset < end)
10540 btrfs_free_reserved_data_space(inode, cur_offset,
10541 end - cur_offset + 1);
10545 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10546 u64 start, u64 num_bytes, u64 min_size,
10547 loff_t actual_len, u64 *alloc_hint)
10549 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10550 min_size, actual_len, alloc_hint,
10554 int btrfs_prealloc_file_range_trans(struct inode *inode,
10555 struct btrfs_trans_handle *trans, int mode,
10556 u64 start, u64 num_bytes, u64 min_size,
10557 loff_t actual_len, u64 *alloc_hint)
10559 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10560 min_size, actual_len, alloc_hint, trans);
10563 static int btrfs_set_page_dirty(struct page *page)
10565 return __set_page_dirty_nobuffers(page);
10568 static int btrfs_permission(struct inode *inode, int mask)
10570 struct btrfs_root *root = BTRFS_I(inode)->root;
10571 umode_t mode = inode->i_mode;
10573 if (mask & MAY_WRITE &&
10574 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10575 if (btrfs_root_readonly(root))
10577 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10580 return generic_permission(inode, mask);
10583 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10585 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10586 struct btrfs_trans_handle *trans;
10587 struct btrfs_root *root = BTRFS_I(dir)->root;
10588 struct inode *inode = NULL;
10594 * 5 units required for adding orphan entry
10596 trans = btrfs_start_transaction(root, 5);
10598 return PTR_ERR(trans);
10600 ret = btrfs_find_free_ino(root, &objectid);
10604 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10605 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10606 if (IS_ERR(inode)) {
10607 ret = PTR_ERR(inode);
10612 inode->i_fop = &btrfs_file_operations;
10613 inode->i_op = &btrfs_file_inode_operations;
10615 inode->i_mapping->a_ops = &btrfs_aops;
10616 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10618 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10622 ret = btrfs_update_inode(trans, root, inode);
10625 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10630 * We set number of links to 0 in btrfs_new_inode(), and here we set
10631 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10634 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10636 set_nlink(inode, 1);
10637 unlock_new_inode(inode);
10638 d_tmpfile(dentry, inode);
10639 mark_inode_dirty(inode);
10642 btrfs_end_transaction(trans);
10645 btrfs_balance_delayed_items(fs_info);
10646 btrfs_btree_balance_dirty(fs_info);
10650 unlock_new_inode(inode);
10655 __attribute__((const))
10656 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10661 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10663 struct inode *inode = private_data;
10664 return btrfs_sb(inode->i_sb);
10667 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10668 u64 start, u64 end)
10670 struct inode *inode = private_data;
10673 isize = i_size_read(inode);
10674 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10675 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10676 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10677 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10681 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10683 struct inode *inode = private_data;
10684 unsigned long index = start >> PAGE_SHIFT;
10685 unsigned long end_index = end >> PAGE_SHIFT;
10688 while (index <= end_index) {
10689 page = find_get_page(inode->i_mapping, index);
10690 ASSERT(page); /* Pages should be in the extent_io_tree */
10691 set_page_writeback(page);
10697 static const struct inode_operations btrfs_dir_inode_operations = {
10698 .getattr = btrfs_getattr,
10699 .lookup = btrfs_lookup,
10700 .create = btrfs_create,
10701 .unlink = btrfs_unlink,
10702 .link = btrfs_link,
10703 .mkdir = btrfs_mkdir,
10704 .rmdir = btrfs_rmdir,
10705 .rename = btrfs_rename2,
10706 .symlink = btrfs_symlink,
10707 .setattr = btrfs_setattr,
10708 .mknod = btrfs_mknod,
10709 .listxattr = btrfs_listxattr,
10710 .permission = btrfs_permission,
10711 .get_acl = btrfs_get_acl,
10712 .set_acl = btrfs_set_acl,
10713 .update_time = btrfs_update_time,
10714 .tmpfile = btrfs_tmpfile,
10716 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10717 .lookup = btrfs_lookup,
10718 .permission = btrfs_permission,
10719 .update_time = btrfs_update_time,
10722 static const struct file_operations btrfs_dir_file_operations = {
10723 .llseek = generic_file_llseek,
10724 .read = generic_read_dir,
10725 .iterate_shared = btrfs_real_readdir,
10726 .unlocked_ioctl = btrfs_ioctl,
10727 #ifdef CONFIG_COMPAT
10728 .compat_ioctl = btrfs_compat_ioctl,
10730 .release = btrfs_release_file,
10731 .fsync = btrfs_sync_file,
10734 static const struct extent_io_ops btrfs_extent_io_ops = {
10735 /* mandatory callbacks */
10736 .submit_bio_hook = btrfs_submit_bio_hook,
10737 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10738 .merge_bio_hook = btrfs_merge_bio_hook,
10739 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10740 .tree_fs_info = iotree_fs_info,
10741 .set_range_writeback = btrfs_set_range_writeback,
10743 /* optional callbacks */
10744 .fill_delalloc = run_delalloc_range,
10745 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10746 .writepage_start_hook = btrfs_writepage_start_hook,
10747 .set_bit_hook = btrfs_set_bit_hook,
10748 .clear_bit_hook = btrfs_clear_bit_hook,
10749 .merge_extent_hook = btrfs_merge_extent_hook,
10750 .split_extent_hook = btrfs_split_extent_hook,
10751 .check_extent_io_range = btrfs_check_extent_io_range,
10755 * btrfs doesn't support the bmap operation because swapfiles
10756 * use bmap to make a mapping of extents in the file. They assume
10757 * these extents won't change over the life of the file and they
10758 * use the bmap result to do IO directly to the drive.
10760 * the btrfs bmap call would return logical addresses that aren't
10761 * suitable for IO and they also will change frequently as COW
10762 * operations happen. So, swapfile + btrfs == corruption.
10764 * For now we're avoiding this by dropping bmap.
10766 static const struct address_space_operations btrfs_aops = {
10767 .readpage = btrfs_readpage,
10768 .writepage = btrfs_writepage,
10769 .writepages = btrfs_writepages,
10770 .readpages = btrfs_readpages,
10771 .direct_IO = btrfs_direct_IO,
10772 .invalidatepage = btrfs_invalidatepage,
10773 .releasepage = btrfs_releasepage,
10774 .set_page_dirty = btrfs_set_page_dirty,
10775 .error_remove_page = generic_error_remove_page,
10778 static const struct address_space_operations btrfs_symlink_aops = {
10779 .readpage = btrfs_readpage,
10780 .writepage = btrfs_writepage,
10781 .invalidatepage = btrfs_invalidatepage,
10782 .releasepage = btrfs_releasepage,
10785 static const struct inode_operations btrfs_file_inode_operations = {
10786 .getattr = btrfs_getattr,
10787 .setattr = btrfs_setattr,
10788 .listxattr = btrfs_listxattr,
10789 .permission = btrfs_permission,
10790 .fiemap = btrfs_fiemap,
10791 .get_acl = btrfs_get_acl,
10792 .set_acl = btrfs_set_acl,
10793 .update_time = btrfs_update_time,
10795 static const struct inode_operations btrfs_special_inode_operations = {
10796 .getattr = btrfs_getattr,
10797 .setattr = btrfs_setattr,
10798 .permission = btrfs_permission,
10799 .listxattr = btrfs_listxattr,
10800 .get_acl = btrfs_get_acl,
10801 .set_acl = btrfs_set_acl,
10802 .update_time = btrfs_update_time,
10804 static const struct inode_operations btrfs_symlink_inode_operations = {
10805 .get_link = page_get_link,
10806 .getattr = btrfs_getattr,
10807 .setattr = btrfs_setattr,
10808 .permission = btrfs_permission,
10809 .listxattr = btrfs_listxattr,
10810 .update_time = btrfs_update_time,
10813 const struct dentry_operations btrfs_dentry_operations = {
10814 .d_delete = btrfs_dentry_delete,
10815 .d_release = btrfs_dentry_release,