2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
116 static int btrfs_dirty_inode(struct inode *inode);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
131 err = btrfs_init_acl(trans, inode, dir);
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
147 struct page **compressed_pages)
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
153 struct btrfs_file_extent_item *ei;
156 size_t cur_size = size;
157 unsigned long offset;
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
162 inode_add_bytes(inode, size);
164 if (!extent_inserted) {
165 struct btrfs_key key;
168 key.objectid = btrfs_ino(inode);
170 key.type = BTRFS_EXTENT_DATA_KEY;
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
191 if (compress_type != BTRFS_COMPRESS_NONE) {
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
205 compressed_size -= cur_size;
207 btrfs_set_file_extent_compression(leaf, ei,
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_CACHE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_CACHE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 page_cache_release(page);
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
249 struct page **compressed_pages)
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
263 data_len = compressed_size;
266 actual_end > root->sectorsize ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
269 (actual_end & (root->sectorsize - 1)) == 0) ||
271 data_len > root->fs_info->max_inline) {
275 path = btrfs_alloc_path();
279 trans = btrfs_join_transaction(root);
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
297 btrfs_abort_transaction(trans, root, ret);
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
310 } else if (ret == -ENOSPC) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
331 struct async_extent {
336 unsigned long nr_pages;
338 struct list_head list;
343 struct btrfs_root *root;
344 struct page *locked_page;
347 struct list_head extents;
348 struct btrfs_work work;
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
355 unsigned long nr_pages,
358 struct async_extent *async_extent;
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
372 static inline int inode_need_compress(struct inode *inode)
374 struct btrfs_root *root = BTRFS_I(inode)->root;
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
409 struct async_cow *async_cow,
412 struct btrfs_root *root = BTRFS_I(inode)->root;
414 u64 blocksize = root->sectorsize;
416 u64 isize = i_size_read(inode);
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
427 int compress_type = root->fs_info->compress_type;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
435 actual_end = min_t(u64, isize, end + 1);
438 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_SIZE);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
454 total_compressed = actual_end - start;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode)) {
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode, start, end);
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
516 unsigned long offset = total_compressed &
517 (PAGE_CACHE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
527 PAGE_CACHE_SIZE - offset);
528 kunmap_atomic(kaddr);
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret = cow_file_range_inline(root, inode, start, end,
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
546 compress_type, pages);
549 unsigned long clear_flags = EXTENT_DELALLOC |
551 unsigned long page_error_op;
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed = ALIGN(total_compressed, blocksize);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
584 if (total_compressed >= total_in) {
587 num_bytes = total_in;
590 if (!will_compress && pages) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
597 page_cache_release(pages[i]);
601 total_compressed = 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
621 if (start + num_bytes < end) {
628 cleanup_and_bail_uncompressed:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 page_cache_release(pages[i]);
658 static void free_async_extent_pages(struct async_extent *async_extent)
662 if (!async_extent->pages)
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 page_cache_release(async_extent->pages[i]);
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
683 struct async_extent *async_extent;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
698 io_tree = &BTRFS_I(inode)->io_tree;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
733 unlock_page(async_cow->locked_page);
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
747 free_async_extent_pages(async_extent);
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode,
762 async_extent->start +
763 async_extent->ram_size - 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
777 em = alloc_extent_map();
780 goto out_free_reserve;
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
812 goto out_free_reserve;
814 ret = btrfs_add_ordered_extent_compress(inode,
817 async_extent->ram_size,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
829 * clear dirty, set writeback and unlock the pages.
831 extent_clear_unlock_delalloc(inode, async_extent->start,
832 async_extent->start +
833 async_extent->ram_size - 1,
834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
837 ret = btrfs_submit_compressed_write(inode,
839 async_extent->ram_size,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
856 free_async_extent_pages(async_extent);
858 alloc_hint = ins.objectid + ins.offset;
864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
866 extent_clear_unlock_delalloc(inode, async_extent->start,
867 async_extent->start +
868 async_extent->ram_size - 1,
869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
874 free_async_extent_pages(async_extent);
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
902 alloc_hint = em->block_start;
906 read_unlock(&em_tree->lock);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
930 struct btrfs_root *root = BTRFS_I(inode)->root;
933 unsigned long ram_size;
936 u64 blocksize = root->sectorsize;
937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942 if (btrfs_is_free_space_inode(inode)) {
948 num_bytes = ALIGN(end - start + 1, blocksize);
949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes < SZ_64K &&
954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955 btrfs_add_inode_defrag(NULL, inode);
958 /* lets try to make an inline extent */
959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
964 EXTENT_DEFRAG, PAGE_UNLOCK |
965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968 *nr_written = *nr_written +
969 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
972 } else if (ret < 0) {
977 BUG_ON(disk_num_bytes >
978 btrfs_super_total_bytes(root->fs_info->super_copy));
980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
983 while (disk_num_bytes > 0) {
986 cur_alloc_size = disk_num_bytes;
987 ret = btrfs_reserve_extent(root, cur_alloc_size,
988 root->sectorsize, 0, alloc_hint,
993 em = alloc_extent_map();
999 em->orig_start = em->start;
1000 ram_size = ins.offset;
1001 em->len = ins.offset;
1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
1005 em->block_start = ins.objectid;
1006 em->block_len = ins.offset;
1007 em->orig_block_len = ins.offset;
1008 em->ram_bytes = ram_size;
1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011 em->generation = -1;
1014 write_lock(&em_tree->lock);
1015 ret = add_extent_mapping(em_tree, em, 1);
1016 write_unlock(&em_tree->lock);
1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1021 btrfs_drop_extent_cache(inode, start,
1022 start + ram_size - 1, 0);
1027 cur_alloc_size = ins.offset;
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1031 goto out_drop_extent_cache;
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1038 goto out_drop_extent_cache;
1041 if (disk_num_bytes < cur_alloc_size)
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
1054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1058 disk_num_bytes -= cur_alloc_size;
1059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
1066 out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1080 * work queue call back to started compression on a file and pages
1082 static noinline void async_cow_start(struct btrfs_work *work)
1084 struct async_cow *async_cow;
1086 async_cow = container_of(work, struct async_cow, work);
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1091 if (num_added == 0) {
1092 btrfs_add_delayed_iput(async_cow->inode);
1093 async_cow->inode = NULL;
1098 * work queue call back to submit previously compressed pages
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1106 async_cow = container_of(work, struct async_cow, work);
1108 root = async_cow->root;
1109 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1113 * atomic_sub_return implies a barrier for waitqueue_active
1115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1120 if (async_cow->inode)
1121 submit_compressed_extents(async_cow->inode, async_cow);
1124 static noinline void async_cow_free(struct btrfs_work *work)
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
1128 if (async_cow->inode)
1129 btrfs_add_delayed_iput(async_cow->inode);
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1141 int limit = 10 * SZ_1M;
1143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
1145 while (start < end) {
1146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147 BUG_ON(!async_cow); /* -ENOMEM */
1148 async_cow->inode = igrab(inode);
1149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
1157 cur_end = min(end, start + SZ_512K - 1);
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1167 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1171 btrfs_queue_work(root->fs_info->delalloc_workers,
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1180 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195 u64 bytenr, u64 num_bytes)
1198 struct btrfs_ordered_sum *sums;
1201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202 bytenr + num_bytes - 1, &list, 0);
1203 if (ret == 0 && list_empty(&list))
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1218 * If no cow copies or snapshots exist, we write directly to the existing
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
1223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1227 struct btrfs_trans_handle *trans;
1228 struct extent_buffer *leaf;
1229 struct btrfs_path *path;
1230 struct btrfs_file_extent_item *fi;
1231 struct btrfs_key found_key;
1246 u64 ino = btrfs_ino(inode);
1248 path = btrfs_alloc_path();
1250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
1252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
1260 nolock = btrfs_is_free_space_inode(inode);
1263 trans = btrfs_join_transaction_nolock(root);
1265 trans = btrfs_join_transaction(root);
1267 if (IS_ERR(trans)) {
1268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
1270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
1275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1281 cow_start = (u64)-1;
1284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
1292 if (found_key.objectid == ino &&
1293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
1305 leaf = path->nodes[0];
1311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1313 if (found_key.objectid > ino)
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321 found_key.offset > end)
1324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
1330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
1334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338 extent_offset = btrfs_file_extent_offset(leaf, fi);
1339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343 if (extent_end <= start) {
1347 if (disk_bytenr == 0)
1349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1355 if (btrfs_extent_readonly(root, disk_bytenr))
1357 if (btrfs_cross_ref_exist(trans, root, ino,
1359 extent_offset, disk_bytenr))
1361 disk_bytenr += extent_offset;
1362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1369 err = btrfs_start_write_no_snapshoting(root);
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
1383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
1385 extent_end = ALIGN(extent_end, root->sectorsize);
1390 if (extent_end <= start) {
1392 if (!nolock && nocow)
1393 btrfs_end_write_no_snapshoting(root);
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1406 btrfs_release_path(path);
1407 if (cow_start != (u64)-1) {
1408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
1412 if (!nolock && nocow)
1413 btrfs_end_write_no_snapshoting(root);
1416 cow_start = (u64)-1;
1419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
1423 em = alloc_extent_map();
1424 BUG_ON(!em); /* -ENOMEM */
1425 em->start = cur_offset;
1426 em->orig_start = found_key.offset - extent_offset;
1427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
1430 em->orig_block_len = disk_num_bytes;
1431 em->ram_bytes = ram_bytes;
1432 em->bdev = root->fs_info->fs_devices->latest_bdev;
1433 em->mod_start = em->start;
1434 em->mod_len = em->len;
1435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437 em->generation = -1;
1439 write_lock(&em_tree->lock);
1440 ret = add_extent_mapping(em_tree, em, 1);
1441 write_unlock(&em_tree->lock);
1442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1449 type = BTRFS_ORDERED_PREALLOC;
1451 type = BTRFS_ORDERED_NOCOW;
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455 num_bytes, num_bytes, type);
1456 BUG_ON(ret); /* -ENOMEM */
1458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1463 if (!nolock && nocow)
1464 btrfs_end_write_no_snapshoting(root);
1469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1474 if (!nolock && nocow)
1475 btrfs_end_write_no_snapshoting(root);
1476 cur_offset = extent_end;
1477 if (cur_offset > end)
1480 btrfs_release_path(path);
1482 if (cur_offset <= end && cow_start == (u64)-1) {
1483 cow_start = cur_offset;
1487 if (cow_start != (u64)-1) {
1488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
1495 err = btrfs_end_transaction(trans, root);
1499 if (ret && cur_offset < end)
1500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
1502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
1507 btrfs_free_path(path);
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1532 * extent_io.c call back to do delayed allocation processing
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
1539 int force_cow = need_force_cow(inode, start, end);
1541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543 page_started, 1, nr_written);
1544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546 page_started, 0, nr_written);
1547 } else if (!inode_need_compress(inode)) {
1548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
1553 ret = cow_file_range_async(inode, locked_page, start, end,
1554 page_started, nr_written);
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
1564 /* not delalloc, ignore it */
1565 if (!(orig->state & EXTENT_DELALLOC))
1568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1577 new_size = orig->end - split + 1;
1578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579 BTRFS_MAX_EXTENT_SIZE);
1580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
1603 u64 new_size, old_size;
1606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
1610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1613 new_size = other->end - new->start + 1;
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1641 old_size = other->end - other->start + 1;
1642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
1644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1675 spin_unlock(&root->delalloc_lock);
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1694 spin_unlock(&root->delalloc_lock);
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703 struct extent_state *state, unsigned *bits)
1706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714 struct btrfs_root *root = BTRFS_I(inode)->root;
1715 u64 len = state->end + 1 - state->start;
1716 bool do_list = !btrfs_is_free_space_inode(inode);
1718 if (*bits & EXTENT_FIRST_DELALLOC) {
1719 *bits &= ~EXTENT_FIRST_DELALLOC;
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 BTRFS_I(inode)->delalloc_bytes += len;
1734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
1736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
1739 spin_unlock(&BTRFS_I(inode)->lock);
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747 struct extent_state *state,
1750 u64 len = state->end + 1 - state->start;
1751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
1766 bool do_list = !btrfs_is_free_space_inode(inode);
1768 if (*bits & EXTENT_FIRST_DELALLOC) {
1769 *bits &= ~EXTENT_FIRST_DELALLOC;
1770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
1772 BTRFS_I(inode)->outstanding_extents -= num_extents;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
1783 btrfs_delalloc_release_metadata(inode, len);
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list && !(state->state & EXTENT_NORESERVE))
1791 btrfs_free_reserved_data_space_noquota(inode,
1794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
1796 spin_lock(&BTRFS_I(inode)->lock);
1797 BTRFS_I(inode)->delalloc_bytes -= len;
1798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
1802 spin_unlock(&BTRFS_I(inode)->lock);
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1823 length = bio->bi_iter.bi_size;
1824 map_length = length;
1825 ret = btrfs_map_block(root->fs_info, rw, logical,
1826 &map_length, NULL, 0);
1827 /* Will always return 0 with map_multi == NULL */
1829 if (map_length < length + size)
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
1844 unsigned long bio_flags,
1847 struct btrfs_root *root = BTRFS_I(inode)->root;
1850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851 BUG_ON(ret); /* -ENOMEM */
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864 int mirror_num, unsigned long bio_flags,
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1872 bio->bi_error = ret;
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883 int mirror_num, unsigned long bio_flags,
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
1887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1894 if (btrfs_is_free_space_inode(inode))
1895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1897 if (!(rw & REQ_WRITE)) {
1898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903 ret = btrfs_submit_compressed_read(inode, bio,
1907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1913 } else if (async && !skip_sum) {
1914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1917 /* we're doing a write, do the async checksumming */
1918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919 inode, rw, bio, mirror_num,
1920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
1922 __btrfs_submit_bio_done);
1924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1935 bio->bi_error = ret;
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1949 struct btrfs_ordered_sum *sum;
1951 list_for_each_entry(sum, list, list) {
1952 trans->adding_csums = 1;
1953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955 trans->adding_csums = 0;
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
1963 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965 cached_state, GFP_NOFS);
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1971 struct btrfs_work work;
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
1978 struct extent_state *cached_state = NULL;
1980 struct inode *inode;
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
1996 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page))
2005 ordered = btrfs_lookup_ordered_range(inode, page_start,
2008 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2009 page_end, &cached_state, GFP_NOFS);
2011 btrfs_start_ordered_extent(inode, ordered, 1);
2012 btrfs_put_ordered_extent(ordered);
2016 ret = btrfs_delalloc_reserve_space(inode, page_start,
2019 mapping_set_error(page->mapping, ret);
2020 end_extent_writepage(page, ret, page_start, page_end);
2021 ClearPageChecked(page);
2025 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2026 ClearPageChecked(page);
2027 set_page_dirty(page);
2029 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2030 &cached_state, GFP_NOFS);
2033 page_cache_release(page);
2038 * There are a few paths in the higher layers of the kernel that directly
2039 * set the page dirty bit without asking the filesystem if it is a
2040 * good idea. This causes problems because we want to make sure COW
2041 * properly happens and the data=ordered rules are followed.
2043 * In our case any range that doesn't have the ORDERED bit set
2044 * hasn't been properly setup for IO. We kick off an async process
2045 * to fix it up. The async helper will wait for ordered extents, set
2046 * the delalloc bit and make it safe to write the page.
2048 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2050 struct inode *inode = page->mapping->host;
2051 struct btrfs_writepage_fixup *fixup;
2052 struct btrfs_root *root = BTRFS_I(inode)->root;
2054 /* this page is properly in the ordered list */
2055 if (TestClearPagePrivate2(page))
2058 if (PageChecked(page))
2061 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2065 SetPageChecked(page);
2066 page_cache_get(page);
2067 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2068 btrfs_writepage_fixup_worker, NULL, NULL);
2070 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2074 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2075 struct inode *inode, u64 file_pos,
2076 u64 disk_bytenr, u64 disk_num_bytes,
2077 u64 num_bytes, u64 ram_bytes,
2078 u8 compression, u8 encryption,
2079 u16 other_encoding, int extent_type)
2081 struct btrfs_root *root = BTRFS_I(inode)->root;
2082 struct btrfs_file_extent_item *fi;
2083 struct btrfs_path *path;
2084 struct extent_buffer *leaf;
2085 struct btrfs_key ins;
2086 int extent_inserted = 0;
2089 path = btrfs_alloc_path();
2094 * we may be replacing one extent in the tree with another.
2095 * The new extent is pinned in the extent map, and we don't want
2096 * to drop it from the cache until it is completely in the btree.
2098 * So, tell btrfs_drop_extents to leave this extent in the cache.
2099 * the caller is expected to unpin it and allow it to be merged
2102 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2103 file_pos + num_bytes, NULL, 0,
2104 1, sizeof(*fi), &extent_inserted);
2108 if (!extent_inserted) {
2109 ins.objectid = btrfs_ino(inode);
2110 ins.offset = file_pos;
2111 ins.type = BTRFS_EXTENT_DATA_KEY;
2113 path->leave_spinning = 1;
2114 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2119 leaf = path->nodes[0];
2120 fi = btrfs_item_ptr(leaf, path->slots[0],
2121 struct btrfs_file_extent_item);
2122 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2123 btrfs_set_file_extent_type(leaf, fi, extent_type);
2124 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2125 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2126 btrfs_set_file_extent_offset(leaf, fi, 0);
2127 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2128 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2129 btrfs_set_file_extent_compression(leaf, fi, compression);
2130 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2131 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2133 btrfs_mark_buffer_dirty(leaf);
2134 btrfs_release_path(path);
2136 inode_add_bytes(inode, num_bytes);
2138 ins.objectid = disk_bytenr;
2139 ins.offset = disk_num_bytes;
2140 ins.type = BTRFS_EXTENT_ITEM_KEY;
2141 ret = btrfs_alloc_reserved_file_extent(trans, root,
2142 root->root_key.objectid,
2143 btrfs_ino(inode), file_pos,
2146 * Release the reserved range from inode dirty range map, as it is
2147 * already moved into delayed_ref_head
2149 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2151 btrfs_free_path(path);
2156 /* snapshot-aware defrag */
2157 struct sa_defrag_extent_backref {
2158 struct rb_node node;
2159 struct old_sa_defrag_extent *old;
2168 struct old_sa_defrag_extent {
2169 struct list_head list;
2170 struct new_sa_defrag_extent *new;
2179 struct new_sa_defrag_extent {
2180 struct rb_root root;
2181 struct list_head head;
2182 struct btrfs_path *path;
2183 struct inode *inode;
2191 static int backref_comp(struct sa_defrag_extent_backref *b1,
2192 struct sa_defrag_extent_backref *b2)
2194 if (b1->root_id < b2->root_id)
2196 else if (b1->root_id > b2->root_id)
2199 if (b1->inum < b2->inum)
2201 else if (b1->inum > b2->inum)
2204 if (b1->file_pos < b2->file_pos)
2206 else if (b1->file_pos > b2->file_pos)
2210 * [------------------------------] ===> (a range of space)
2211 * |<--->| |<---->| =============> (fs/file tree A)
2212 * |<---------------------------->| ===> (fs/file tree B)
2214 * A range of space can refer to two file extents in one tree while
2215 * refer to only one file extent in another tree.
2217 * So we may process a disk offset more than one time(two extents in A)
2218 * and locate at the same extent(one extent in B), then insert two same
2219 * backrefs(both refer to the extent in B).
2224 static void backref_insert(struct rb_root *root,
2225 struct sa_defrag_extent_backref *backref)
2227 struct rb_node **p = &root->rb_node;
2228 struct rb_node *parent = NULL;
2229 struct sa_defrag_extent_backref *entry;
2234 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2236 ret = backref_comp(backref, entry);
2240 p = &(*p)->rb_right;
2243 rb_link_node(&backref->node, parent, p);
2244 rb_insert_color(&backref->node, root);
2248 * Note the backref might has changed, and in this case we just return 0.
2250 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2253 struct btrfs_file_extent_item *extent;
2254 struct btrfs_fs_info *fs_info;
2255 struct old_sa_defrag_extent *old = ctx;
2256 struct new_sa_defrag_extent *new = old->new;
2257 struct btrfs_path *path = new->path;
2258 struct btrfs_key key;
2259 struct btrfs_root *root;
2260 struct sa_defrag_extent_backref *backref;
2261 struct extent_buffer *leaf;
2262 struct inode *inode = new->inode;
2268 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2269 inum == btrfs_ino(inode))
2272 key.objectid = root_id;
2273 key.type = BTRFS_ROOT_ITEM_KEY;
2274 key.offset = (u64)-1;
2276 fs_info = BTRFS_I(inode)->root->fs_info;
2277 root = btrfs_read_fs_root_no_name(fs_info, &key);
2279 if (PTR_ERR(root) == -ENOENT)
2282 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2283 inum, offset, root_id);
2284 return PTR_ERR(root);
2287 key.objectid = inum;
2288 key.type = BTRFS_EXTENT_DATA_KEY;
2289 if (offset > (u64)-1 << 32)
2292 key.offset = offset;
2294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2295 if (WARN_ON(ret < 0))
2302 leaf = path->nodes[0];
2303 slot = path->slots[0];
2305 if (slot >= btrfs_header_nritems(leaf)) {
2306 ret = btrfs_next_leaf(root, path);
2309 } else if (ret > 0) {
2318 btrfs_item_key_to_cpu(leaf, &key, slot);
2320 if (key.objectid > inum)
2323 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2326 extent = btrfs_item_ptr(leaf, slot,
2327 struct btrfs_file_extent_item);
2329 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2333 * 'offset' refers to the exact key.offset,
2334 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2335 * (key.offset - extent_offset).
2337 if (key.offset != offset)
2340 extent_offset = btrfs_file_extent_offset(leaf, extent);
2341 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2343 if (extent_offset >= old->extent_offset + old->offset +
2344 old->len || extent_offset + num_bytes <=
2345 old->extent_offset + old->offset)
2350 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2356 backref->root_id = root_id;
2357 backref->inum = inum;
2358 backref->file_pos = offset;
2359 backref->num_bytes = num_bytes;
2360 backref->extent_offset = extent_offset;
2361 backref->generation = btrfs_file_extent_generation(leaf, extent);
2363 backref_insert(&new->root, backref);
2366 btrfs_release_path(path);
2371 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2372 struct new_sa_defrag_extent *new)
2374 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2375 struct old_sa_defrag_extent *old, *tmp;
2380 list_for_each_entry_safe(old, tmp, &new->head, list) {
2381 ret = iterate_inodes_from_logical(old->bytenr +
2382 old->extent_offset, fs_info,
2383 path, record_one_backref,
2385 if (ret < 0 && ret != -ENOENT)
2388 /* no backref to be processed for this extent */
2390 list_del(&old->list);
2395 if (list_empty(&new->head))
2401 static int relink_is_mergable(struct extent_buffer *leaf,
2402 struct btrfs_file_extent_item *fi,
2403 struct new_sa_defrag_extent *new)
2405 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2408 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2411 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2414 if (btrfs_file_extent_encryption(leaf, fi) ||
2415 btrfs_file_extent_other_encoding(leaf, fi))
2422 * Note the backref might has changed, and in this case we just return 0.
2424 static noinline int relink_extent_backref(struct btrfs_path *path,
2425 struct sa_defrag_extent_backref *prev,
2426 struct sa_defrag_extent_backref *backref)
2428 struct btrfs_file_extent_item *extent;
2429 struct btrfs_file_extent_item *item;
2430 struct btrfs_ordered_extent *ordered;
2431 struct btrfs_trans_handle *trans;
2432 struct btrfs_fs_info *fs_info;
2433 struct btrfs_root *root;
2434 struct btrfs_key key;
2435 struct extent_buffer *leaf;
2436 struct old_sa_defrag_extent *old = backref->old;
2437 struct new_sa_defrag_extent *new = old->new;
2438 struct inode *src_inode = new->inode;
2439 struct inode *inode;
2440 struct extent_state *cached = NULL;
2449 if (prev && prev->root_id == backref->root_id &&
2450 prev->inum == backref->inum &&
2451 prev->file_pos + prev->num_bytes == backref->file_pos)
2454 /* step 1: get root */
2455 key.objectid = backref->root_id;
2456 key.type = BTRFS_ROOT_ITEM_KEY;
2457 key.offset = (u64)-1;
2459 fs_info = BTRFS_I(src_inode)->root->fs_info;
2460 index = srcu_read_lock(&fs_info->subvol_srcu);
2462 root = btrfs_read_fs_root_no_name(fs_info, &key);
2464 srcu_read_unlock(&fs_info->subvol_srcu, index);
2465 if (PTR_ERR(root) == -ENOENT)
2467 return PTR_ERR(root);
2470 if (btrfs_root_readonly(root)) {
2471 srcu_read_unlock(&fs_info->subvol_srcu, index);
2475 /* step 2: get inode */
2476 key.objectid = backref->inum;
2477 key.type = BTRFS_INODE_ITEM_KEY;
2480 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2481 if (IS_ERR(inode)) {
2482 srcu_read_unlock(&fs_info->subvol_srcu, index);
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2488 /* step 3: relink backref */
2489 lock_start = backref->file_pos;
2490 lock_end = backref->file_pos + backref->num_bytes - 1;
2491 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2494 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2496 btrfs_put_ordered_extent(ordered);
2500 trans = btrfs_join_transaction(root);
2501 if (IS_ERR(trans)) {
2502 ret = PTR_ERR(trans);
2506 key.objectid = backref->inum;
2507 key.type = BTRFS_EXTENT_DATA_KEY;
2508 key.offset = backref->file_pos;
2510 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2513 } else if (ret > 0) {
2518 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2519 struct btrfs_file_extent_item);
2521 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2522 backref->generation)
2525 btrfs_release_path(path);
2527 start = backref->file_pos;
2528 if (backref->extent_offset < old->extent_offset + old->offset)
2529 start += old->extent_offset + old->offset -
2530 backref->extent_offset;
2532 len = min(backref->extent_offset + backref->num_bytes,
2533 old->extent_offset + old->offset + old->len);
2534 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2536 ret = btrfs_drop_extents(trans, root, inode, start,
2541 key.objectid = btrfs_ino(inode);
2542 key.type = BTRFS_EXTENT_DATA_KEY;
2545 path->leave_spinning = 1;
2547 struct btrfs_file_extent_item *fi;
2549 struct btrfs_key found_key;
2551 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2556 leaf = path->nodes[0];
2557 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2559 fi = btrfs_item_ptr(leaf, path->slots[0],
2560 struct btrfs_file_extent_item);
2561 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2563 if (extent_len + found_key.offset == start &&
2564 relink_is_mergable(leaf, fi, new)) {
2565 btrfs_set_file_extent_num_bytes(leaf, fi,
2567 btrfs_mark_buffer_dirty(leaf);
2568 inode_add_bytes(inode, len);
2574 btrfs_release_path(path);
2579 ret = btrfs_insert_empty_item(trans, root, path, &key,
2582 btrfs_abort_transaction(trans, root, ret);
2586 leaf = path->nodes[0];
2587 item = btrfs_item_ptr(leaf, path->slots[0],
2588 struct btrfs_file_extent_item);
2589 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2590 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2591 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2592 btrfs_set_file_extent_num_bytes(leaf, item, len);
2593 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2594 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2595 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2596 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2597 btrfs_set_file_extent_encryption(leaf, item, 0);
2598 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2600 btrfs_mark_buffer_dirty(leaf);
2601 inode_add_bytes(inode, len);
2602 btrfs_release_path(path);
2604 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2606 backref->root_id, backref->inum,
2607 new->file_pos); /* start - extent_offset */
2609 btrfs_abort_transaction(trans, root, ret);
2615 btrfs_release_path(path);
2616 path->leave_spinning = 0;
2617 btrfs_end_transaction(trans, root);
2619 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2625 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2627 struct old_sa_defrag_extent *old, *tmp;
2632 list_for_each_entry_safe(old, tmp, &new->head, list) {
2638 static void relink_file_extents(struct new_sa_defrag_extent *new)
2640 struct btrfs_path *path;
2641 struct sa_defrag_extent_backref *backref;
2642 struct sa_defrag_extent_backref *prev = NULL;
2643 struct inode *inode;
2644 struct btrfs_root *root;
2645 struct rb_node *node;
2649 root = BTRFS_I(inode)->root;
2651 path = btrfs_alloc_path();
2655 if (!record_extent_backrefs(path, new)) {
2656 btrfs_free_path(path);
2659 btrfs_release_path(path);
2662 node = rb_first(&new->root);
2665 rb_erase(node, &new->root);
2667 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2669 ret = relink_extent_backref(path, prev, backref);
2682 btrfs_free_path(path);
2684 free_sa_defrag_extent(new);
2686 atomic_dec(&root->fs_info->defrag_running);
2687 wake_up(&root->fs_info->transaction_wait);
2690 static struct new_sa_defrag_extent *
2691 record_old_file_extents(struct inode *inode,
2692 struct btrfs_ordered_extent *ordered)
2694 struct btrfs_root *root = BTRFS_I(inode)->root;
2695 struct btrfs_path *path;
2696 struct btrfs_key key;
2697 struct old_sa_defrag_extent *old;
2698 struct new_sa_defrag_extent *new;
2701 new = kmalloc(sizeof(*new), GFP_NOFS);
2706 new->file_pos = ordered->file_offset;
2707 new->len = ordered->len;
2708 new->bytenr = ordered->start;
2709 new->disk_len = ordered->disk_len;
2710 new->compress_type = ordered->compress_type;
2711 new->root = RB_ROOT;
2712 INIT_LIST_HEAD(&new->head);
2714 path = btrfs_alloc_path();
2718 key.objectid = btrfs_ino(inode);
2719 key.type = BTRFS_EXTENT_DATA_KEY;
2720 key.offset = new->file_pos;
2722 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2725 if (ret > 0 && path->slots[0] > 0)
2728 /* find out all the old extents for the file range */
2730 struct btrfs_file_extent_item *extent;
2731 struct extent_buffer *l;
2740 slot = path->slots[0];
2742 if (slot >= btrfs_header_nritems(l)) {
2743 ret = btrfs_next_leaf(root, path);
2751 btrfs_item_key_to_cpu(l, &key, slot);
2753 if (key.objectid != btrfs_ino(inode))
2755 if (key.type != BTRFS_EXTENT_DATA_KEY)
2757 if (key.offset >= new->file_pos + new->len)
2760 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2762 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2763 if (key.offset + num_bytes < new->file_pos)
2766 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2770 extent_offset = btrfs_file_extent_offset(l, extent);
2772 old = kmalloc(sizeof(*old), GFP_NOFS);
2776 offset = max(new->file_pos, key.offset);
2777 end = min(new->file_pos + new->len, key.offset + num_bytes);
2779 old->bytenr = disk_bytenr;
2780 old->extent_offset = extent_offset;
2781 old->offset = offset - key.offset;
2782 old->len = end - offset;
2785 list_add_tail(&old->list, &new->head);
2791 btrfs_free_path(path);
2792 atomic_inc(&root->fs_info->defrag_running);
2797 btrfs_free_path(path);
2799 free_sa_defrag_extent(new);
2803 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2806 struct btrfs_block_group_cache *cache;
2808 cache = btrfs_lookup_block_group(root->fs_info, start);
2811 spin_lock(&cache->lock);
2812 cache->delalloc_bytes -= len;
2813 spin_unlock(&cache->lock);
2815 btrfs_put_block_group(cache);
2818 /* as ordered data IO finishes, this gets called so we can finish
2819 * an ordered extent if the range of bytes in the file it covers are
2822 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2824 struct inode *inode = ordered_extent->inode;
2825 struct btrfs_root *root = BTRFS_I(inode)->root;
2826 struct btrfs_trans_handle *trans = NULL;
2827 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2828 struct extent_state *cached_state = NULL;
2829 struct new_sa_defrag_extent *new = NULL;
2830 int compress_type = 0;
2832 u64 logical_len = ordered_extent->len;
2834 bool truncated = false;
2836 nolock = btrfs_is_free_space_inode(inode);
2838 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2843 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2844 ordered_extent->file_offset +
2845 ordered_extent->len - 1);
2847 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2849 logical_len = ordered_extent->truncated_len;
2850 /* Truncated the entire extent, don't bother adding */
2855 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2856 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2859 * For mwrite(mmap + memset to write) case, we still reserve
2860 * space for NOCOW range.
2861 * As NOCOW won't cause a new delayed ref, just free the space
2863 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2864 ordered_extent->len);
2865 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2867 trans = btrfs_join_transaction_nolock(root);
2869 trans = btrfs_join_transaction(root);
2870 if (IS_ERR(trans)) {
2871 ret = PTR_ERR(trans);
2875 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2876 ret = btrfs_update_inode_fallback(trans, root, inode);
2877 if (ret) /* -ENOMEM or corruption */
2878 btrfs_abort_transaction(trans, root, ret);
2882 lock_extent_bits(io_tree, ordered_extent->file_offset,
2883 ordered_extent->file_offset + ordered_extent->len - 1,
2886 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2887 ordered_extent->file_offset + ordered_extent->len - 1,
2888 EXTENT_DEFRAG, 1, cached_state);
2890 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2891 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2892 /* the inode is shared */
2893 new = record_old_file_extents(inode, ordered_extent);
2895 clear_extent_bit(io_tree, ordered_extent->file_offset,
2896 ordered_extent->file_offset + ordered_extent->len - 1,
2897 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2901 trans = btrfs_join_transaction_nolock(root);
2903 trans = btrfs_join_transaction(root);
2904 if (IS_ERR(trans)) {
2905 ret = PTR_ERR(trans);
2910 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2912 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2913 compress_type = ordered_extent->compress_type;
2914 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2915 BUG_ON(compress_type);
2916 ret = btrfs_mark_extent_written(trans, inode,
2917 ordered_extent->file_offset,
2918 ordered_extent->file_offset +
2921 BUG_ON(root == root->fs_info->tree_root);
2922 ret = insert_reserved_file_extent(trans, inode,
2923 ordered_extent->file_offset,
2924 ordered_extent->start,
2925 ordered_extent->disk_len,
2926 logical_len, logical_len,
2927 compress_type, 0, 0,
2928 BTRFS_FILE_EXTENT_REG);
2930 btrfs_release_delalloc_bytes(root,
2931 ordered_extent->start,
2932 ordered_extent->disk_len);
2934 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2935 ordered_extent->file_offset, ordered_extent->len,
2938 btrfs_abort_transaction(trans, root, ret);
2942 add_pending_csums(trans, inode, ordered_extent->file_offset,
2943 &ordered_extent->list);
2945 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2946 ret = btrfs_update_inode_fallback(trans, root, inode);
2947 if (ret) { /* -ENOMEM or corruption */
2948 btrfs_abort_transaction(trans, root, ret);
2953 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2954 ordered_extent->file_offset +
2955 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2957 if (root != root->fs_info->tree_root)
2958 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2960 btrfs_end_transaction(trans, root);
2962 if (ret || truncated) {
2966 start = ordered_extent->file_offset + logical_len;
2968 start = ordered_extent->file_offset;
2969 end = ordered_extent->file_offset + ordered_extent->len - 1;
2970 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2972 /* Drop the cache for the part of the extent we didn't write. */
2973 btrfs_drop_extent_cache(inode, start, end, 0);
2976 * If the ordered extent had an IOERR or something else went
2977 * wrong we need to return the space for this ordered extent
2978 * back to the allocator. We only free the extent in the
2979 * truncated case if we didn't write out the extent at all.
2981 if ((ret || !logical_len) &&
2982 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2983 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2984 btrfs_free_reserved_extent(root, ordered_extent->start,
2985 ordered_extent->disk_len, 1);
2990 * This needs to be done to make sure anybody waiting knows we are done
2991 * updating everything for this ordered extent.
2993 btrfs_remove_ordered_extent(inode, ordered_extent);
2995 /* for snapshot-aware defrag */
2998 free_sa_defrag_extent(new);
2999 atomic_dec(&root->fs_info->defrag_running);
3001 relink_file_extents(new);
3006 btrfs_put_ordered_extent(ordered_extent);
3007 /* once for the tree */
3008 btrfs_put_ordered_extent(ordered_extent);
3013 static void finish_ordered_fn(struct btrfs_work *work)
3015 struct btrfs_ordered_extent *ordered_extent;
3016 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3017 btrfs_finish_ordered_io(ordered_extent);
3020 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3021 struct extent_state *state, int uptodate)
3023 struct inode *inode = page->mapping->host;
3024 struct btrfs_root *root = BTRFS_I(inode)->root;
3025 struct btrfs_ordered_extent *ordered_extent = NULL;
3026 struct btrfs_workqueue *wq;
3027 btrfs_work_func_t func;
3029 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3031 ClearPagePrivate2(page);
3032 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3033 end - start + 1, uptodate))
3036 if (btrfs_is_free_space_inode(inode)) {
3037 wq = root->fs_info->endio_freespace_worker;
3038 func = btrfs_freespace_write_helper;
3040 wq = root->fs_info->endio_write_workers;
3041 func = btrfs_endio_write_helper;
3044 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3046 btrfs_queue_work(wq, &ordered_extent->work);
3051 static int __readpage_endio_check(struct inode *inode,
3052 struct btrfs_io_bio *io_bio,
3053 int icsum, struct page *page,
3054 int pgoff, u64 start, size_t len)
3060 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3062 kaddr = kmap_atomic(page);
3063 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3064 btrfs_csum_final(csum, (char *)&csum);
3065 if (csum != csum_expected)
3068 kunmap_atomic(kaddr);
3071 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3072 "csum failed ino %llu off %llu csum %u expected csum %u",
3073 btrfs_ino(inode), start, csum, csum_expected);
3074 memset(kaddr + pgoff, 1, len);
3075 flush_dcache_page(page);
3076 kunmap_atomic(kaddr);
3077 if (csum_expected == 0)
3083 * when reads are done, we need to check csums to verify the data is correct
3084 * if there's a match, we allow the bio to finish. If not, the code in
3085 * extent_io.c will try to find good copies for us.
3087 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3088 u64 phy_offset, struct page *page,
3089 u64 start, u64 end, int mirror)
3091 size_t offset = start - page_offset(page);
3092 struct inode *inode = page->mapping->host;
3093 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3094 struct btrfs_root *root = BTRFS_I(inode)->root;
3096 if (PageChecked(page)) {
3097 ClearPageChecked(page);
3101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3104 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3105 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3106 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3111 phy_offset >>= inode->i_sb->s_blocksize_bits;
3112 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3113 start, (size_t)(end - start + 1));
3116 void btrfs_add_delayed_iput(struct inode *inode)
3118 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3119 struct btrfs_inode *binode = BTRFS_I(inode);
3121 if (atomic_add_unless(&inode->i_count, -1, 1))
3124 spin_lock(&fs_info->delayed_iput_lock);
3125 if (binode->delayed_iput_count == 0) {
3126 ASSERT(list_empty(&binode->delayed_iput));
3127 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3129 binode->delayed_iput_count++;
3131 spin_unlock(&fs_info->delayed_iput_lock);
3134 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3136 struct btrfs_fs_info *fs_info = root->fs_info;
3138 spin_lock(&fs_info->delayed_iput_lock);
3139 while (!list_empty(&fs_info->delayed_iputs)) {
3140 struct btrfs_inode *inode;
3142 inode = list_first_entry(&fs_info->delayed_iputs,
3143 struct btrfs_inode, delayed_iput);
3144 if (inode->delayed_iput_count) {
3145 inode->delayed_iput_count--;
3146 list_move_tail(&inode->delayed_iput,
3147 &fs_info->delayed_iputs);
3149 list_del_init(&inode->delayed_iput);
3151 spin_unlock(&fs_info->delayed_iput_lock);
3152 iput(&inode->vfs_inode);
3153 spin_lock(&fs_info->delayed_iput_lock);
3155 spin_unlock(&fs_info->delayed_iput_lock);
3159 * This is called in transaction commit time. If there are no orphan
3160 * files in the subvolume, it removes orphan item and frees block_rsv
3163 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3164 struct btrfs_root *root)
3166 struct btrfs_block_rsv *block_rsv;
3169 if (atomic_read(&root->orphan_inodes) ||
3170 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3173 spin_lock(&root->orphan_lock);
3174 if (atomic_read(&root->orphan_inodes)) {
3175 spin_unlock(&root->orphan_lock);
3179 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3180 spin_unlock(&root->orphan_lock);
3184 block_rsv = root->orphan_block_rsv;
3185 root->orphan_block_rsv = NULL;
3186 spin_unlock(&root->orphan_lock);
3188 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3189 btrfs_root_refs(&root->root_item) > 0) {
3190 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3191 root->root_key.objectid);
3193 btrfs_abort_transaction(trans, root, ret);
3195 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3200 WARN_ON(block_rsv->size > 0);
3201 btrfs_free_block_rsv(root, block_rsv);
3206 * This creates an orphan entry for the given inode in case something goes
3207 * wrong in the middle of an unlink/truncate.
3209 * NOTE: caller of this function should reserve 5 units of metadata for
3212 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3214 struct btrfs_root *root = BTRFS_I(inode)->root;
3215 struct btrfs_block_rsv *block_rsv = NULL;
3220 if (!root->orphan_block_rsv) {
3221 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3226 spin_lock(&root->orphan_lock);
3227 if (!root->orphan_block_rsv) {
3228 root->orphan_block_rsv = block_rsv;
3229 } else if (block_rsv) {
3230 btrfs_free_block_rsv(root, block_rsv);
3234 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3235 &BTRFS_I(inode)->runtime_flags)) {
3238 * For proper ENOSPC handling, we should do orphan
3239 * cleanup when mounting. But this introduces backward
3240 * compatibility issue.
3242 if (!xchg(&root->orphan_item_inserted, 1))
3248 atomic_inc(&root->orphan_inodes);
3251 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3252 &BTRFS_I(inode)->runtime_flags))
3254 spin_unlock(&root->orphan_lock);
3256 /* grab metadata reservation from transaction handle */
3258 ret = btrfs_orphan_reserve_metadata(trans, inode);
3259 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3262 /* insert an orphan item to track this unlinked/truncated file */
3264 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3266 atomic_dec(&root->orphan_inodes);
3268 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3269 &BTRFS_I(inode)->runtime_flags);
3270 btrfs_orphan_release_metadata(inode);
3272 if (ret != -EEXIST) {
3273 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3274 &BTRFS_I(inode)->runtime_flags);
3275 btrfs_abort_transaction(trans, root, ret);
3282 /* insert an orphan item to track subvolume contains orphan files */
3284 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3285 root->root_key.objectid);
3286 if (ret && ret != -EEXIST) {
3287 btrfs_abort_transaction(trans, root, ret);
3295 * We have done the truncate/delete so we can go ahead and remove the orphan
3296 * item for this particular inode.
3298 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3299 struct inode *inode)
3301 struct btrfs_root *root = BTRFS_I(inode)->root;
3302 int delete_item = 0;
3303 int release_rsv = 0;
3306 spin_lock(&root->orphan_lock);
3307 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3308 &BTRFS_I(inode)->runtime_flags))
3311 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3312 &BTRFS_I(inode)->runtime_flags))
3314 spin_unlock(&root->orphan_lock);
3317 atomic_dec(&root->orphan_inodes);
3319 ret = btrfs_del_orphan_item(trans, root,
3324 btrfs_orphan_release_metadata(inode);
3330 * this cleans up any orphans that may be left on the list from the last use
3333 int btrfs_orphan_cleanup(struct btrfs_root *root)
3335 struct btrfs_path *path;
3336 struct extent_buffer *leaf;
3337 struct btrfs_key key, found_key;
3338 struct btrfs_trans_handle *trans;
3339 struct inode *inode;
3340 u64 last_objectid = 0;
3341 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3343 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3346 path = btrfs_alloc_path();
3351 path->reada = READA_BACK;
3353 key.objectid = BTRFS_ORPHAN_OBJECTID;
3354 key.type = BTRFS_ORPHAN_ITEM_KEY;
3355 key.offset = (u64)-1;
3358 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3363 * if ret == 0 means we found what we were searching for, which
3364 * is weird, but possible, so only screw with path if we didn't
3365 * find the key and see if we have stuff that matches
3369 if (path->slots[0] == 0)
3374 /* pull out the item */
3375 leaf = path->nodes[0];
3376 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3378 /* make sure the item matches what we want */
3379 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3381 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3384 /* release the path since we're done with it */
3385 btrfs_release_path(path);
3388 * this is where we are basically btrfs_lookup, without the
3389 * crossing root thing. we store the inode number in the
3390 * offset of the orphan item.
3393 if (found_key.offset == last_objectid) {
3394 btrfs_err(root->fs_info,
3395 "Error removing orphan entry, stopping orphan cleanup");
3400 last_objectid = found_key.offset;
3402 found_key.objectid = found_key.offset;
3403 found_key.type = BTRFS_INODE_ITEM_KEY;
3404 found_key.offset = 0;
3405 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3406 ret = PTR_ERR_OR_ZERO(inode);
3407 if (ret && ret != -ESTALE)
3410 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3411 struct btrfs_root *dead_root;
3412 struct btrfs_fs_info *fs_info = root->fs_info;
3413 int is_dead_root = 0;
3416 * this is an orphan in the tree root. Currently these
3417 * could come from 2 sources:
3418 * a) a snapshot deletion in progress
3419 * b) a free space cache inode
3420 * We need to distinguish those two, as the snapshot
3421 * orphan must not get deleted.
3422 * find_dead_roots already ran before us, so if this
3423 * is a snapshot deletion, we should find the root
3424 * in the dead_roots list
3426 spin_lock(&fs_info->trans_lock);
3427 list_for_each_entry(dead_root, &fs_info->dead_roots,
3429 if (dead_root->root_key.objectid ==
3430 found_key.objectid) {
3435 spin_unlock(&fs_info->trans_lock);
3437 /* prevent this orphan from being found again */
3438 key.offset = found_key.objectid - 1;
3443 * Inode is already gone but the orphan item is still there,
3444 * kill the orphan item.
3446 if (ret == -ESTALE) {
3447 trans = btrfs_start_transaction(root, 1);
3448 if (IS_ERR(trans)) {
3449 ret = PTR_ERR(trans);
3452 btrfs_debug(root->fs_info, "auto deleting %Lu",
3453 found_key.objectid);
3454 ret = btrfs_del_orphan_item(trans, root,
3455 found_key.objectid);
3456 btrfs_end_transaction(trans, root);
3463 * add this inode to the orphan list so btrfs_orphan_del does
3464 * the proper thing when we hit it
3466 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3467 &BTRFS_I(inode)->runtime_flags);
3468 atomic_inc(&root->orphan_inodes);
3470 /* if we have links, this was a truncate, lets do that */
3471 if (inode->i_nlink) {
3472 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3478 /* 1 for the orphan item deletion. */
3479 trans = btrfs_start_transaction(root, 1);
3480 if (IS_ERR(trans)) {
3482 ret = PTR_ERR(trans);
3485 ret = btrfs_orphan_add(trans, inode);
3486 btrfs_end_transaction(trans, root);
3492 ret = btrfs_truncate(inode);
3494 btrfs_orphan_del(NULL, inode);
3499 /* this will do delete_inode and everything for us */
3504 /* release the path since we're done with it */
3505 btrfs_release_path(path);
3507 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3509 if (root->orphan_block_rsv)
3510 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3513 if (root->orphan_block_rsv ||
3514 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3515 trans = btrfs_join_transaction(root);
3517 btrfs_end_transaction(trans, root);
3521 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3523 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3527 btrfs_err(root->fs_info,
3528 "could not do orphan cleanup %d", ret);
3529 btrfs_free_path(path);
3534 * very simple check to peek ahead in the leaf looking for xattrs. If we
3535 * don't find any xattrs, we know there can't be any acls.
3537 * slot is the slot the inode is in, objectid is the objectid of the inode
3539 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3540 int slot, u64 objectid,
3541 int *first_xattr_slot)
3543 u32 nritems = btrfs_header_nritems(leaf);
3544 struct btrfs_key found_key;
3545 static u64 xattr_access = 0;
3546 static u64 xattr_default = 0;
3549 if (!xattr_access) {
3550 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3551 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3552 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3553 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3557 *first_xattr_slot = -1;
3558 while (slot < nritems) {
3559 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3561 /* we found a different objectid, there must not be acls */
3562 if (found_key.objectid != objectid)
3565 /* we found an xattr, assume we've got an acl */
3566 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3567 if (*first_xattr_slot == -1)
3568 *first_xattr_slot = slot;
3569 if (found_key.offset == xattr_access ||
3570 found_key.offset == xattr_default)
3575 * we found a key greater than an xattr key, there can't
3576 * be any acls later on
3578 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3585 * it goes inode, inode backrefs, xattrs, extents,
3586 * so if there are a ton of hard links to an inode there can
3587 * be a lot of backrefs. Don't waste time searching too hard,
3588 * this is just an optimization
3593 /* we hit the end of the leaf before we found an xattr or
3594 * something larger than an xattr. We have to assume the inode
3597 if (*first_xattr_slot == -1)
3598 *first_xattr_slot = slot;
3603 * read an inode from the btree into the in-memory inode
3605 static void btrfs_read_locked_inode(struct inode *inode)
3607 struct btrfs_path *path;
3608 struct extent_buffer *leaf;
3609 struct btrfs_inode_item *inode_item;
3610 struct btrfs_root *root = BTRFS_I(inode)->root;
3611 struct btrfs_key location;
3616 bool filled = false;
3617 int first_xattr_slot;
3619 ret = btrfs_fill_inode(inode, &rdev);
3623 path = btrfs_alloc_path();
3627 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3629 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3633 leaf = path->nodes[0];
3638 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3639 struct btrfs_inode_item);
3640 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3641 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3642 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3643 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3644 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3646 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3647 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3649 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3650 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3652 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3653 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3655 BTRFS_I(inode)->i_otime.tv_sec =
3656 btrfs_timespec_sec(leaf, &inode_item->otime);
3657 BTRFS_I(inode)->i_otime.tv_nsec =
3658 btrfs_timespec_nsec(leaf, &inode_item->otime);
3660 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3661 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3662 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3664 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3665 inode->i_generation = BTRFS_I(inode)->generation;
3667 rdev = btrfs_inode_rdev(leaf, inode_item);
3669 BTRFS_I(inode)->index_cnt = (u64)-1;
3670 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3674 * If we were modified in the current generation and evicted from memory
3675 * and then re-read we need to do a full sync since we don't have any
3676 * idea about which extents were modified before we were evicted from
3679 * This is required for both inode re-read from disk and delayed inode
3680 * in delayed_nodes_tree.
3682 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3683 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3684 &BTRFS_I(inode)->runtime_flags);
3687 * We don't persist the id of the transaction where an unlink operation
3688 * against the inode was last made. So here we assume the inode might
3689 * have been evicted, and therefore the exact value of last_unlink_trans
3690 * lost, and set it to last_trans to avoid metadata inconsistencies
3691 * between the inode and its parent if the inode is fsync'ed and the log
3692 * replayed. For example, in the scenario:
3695 * ln mydir/foo mydir/bar
3698 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3699 * xfs_io -c fsync mydir/foo
3701 * mount fs, triggers fsync log replay
3703 * We must make sure that when we fsync our inode foo we also log its
3704 * parent inode, otherwise after log replay the parent still has the
3705 * dentry with the "bar" name but our inode foo has a link count of 1
3706 * and doesn't have an inode ref with the name "bar" anymore.
3708 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3709 * but it guarantees correctness at the expense of ocassional full
3710 * transaction commits on fsync if our inode is a directory, or if our
3711 * inode is not a directory, logging its parent unnecessarily.
3713 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3716 if (inode->i_nlink != 1 ||
3717 path->slots[0] >= btrfs_header_nritems(leaf))
3720 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3721 if (location.objectid != btrfs_ino(inode))
3724 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3725 if (location.type == BTRFS_INODE_REF_KEY) {
3726 struct btrfs_inode_ref *ref;
3728 ref = (struct btrfs_inode_ref *)ptr;
3729 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3730 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3731 struct btrfs_inode_extref *extref;
3733 extref = (struct btrfs_inode_extref *)ptr;
3734 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3739 * try to precache a NULL acl entry for files that don't have
3740 * any xattrs or acls
3742 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3743 btrfs_ino(inode), &first_xattr_slot);
3744 if (first_xattr_slot != -1) {
3745 path->slots[0] = first_xattr_slot;
3746 ret = btrfs_load_inode_props(inode, path);
3748 btrfs_err(root->fs_info,
3749 "error loading props for ino %llu (root %llu): %d",
3751 root->root_key.objectid, ret);
3753 btrfs_free_path(path);
3756 cache_no_acl(inode);
3758 switch (inode->i_mode & S_IFMT) {
3760 inode->i_mapping->a_ops = &btrfs_aops;
3761 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3762 inode->i_fop = &btrfs_file_operations;
3763 inode->i_op = &btrfs_file_inode_operations;
3766 inode->i_fop = &btrfs_dir_file_operations;
3767 if (root == root->fs_info->tree_root)
3768 inode->i_op = &btrfs_dir_ro_inode_operations;
3770 inode->i_op = &btrfs_dir_inode_operations;
3773 inode->i_op = &btrfs_symlink_inode_operations;
3774 inode_nohighmem(inode);
3775 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3778 inode->i_op = &btrfs_special_inode_operations;
3779 init_special_inode(inode, inode->i_mode, rdev);
3783 btrfs_update_iflags(inode);
3787 btrfs_free_path(path);
3788 make_bad_inode(inode);
3792 * given a leaf and an inode, copy the inode fields into the leaf
3794 static void fill_inode_item(struct btrfs_trans_handle *trans,
3795 struct extent_buffer *leaf,
3796 struct btrfs_inode_item *item,
3797 struct inode *inode)
3799 struct btrfs_map_token token;
3801 btrfs_init_map_token(&token);
3803 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3804 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3805 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3807 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3808 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3810 btrfs_set_token_timespec_sec(leaf, &item->atime,
3811 inode->i_atime.tv_sec, &token);
3812 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3813 inode->i_atime.tv_nsec, &token);
3815 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3816 inode->i_mtime.tv_sec, &token);
3817 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3818 inode->i_mtime.tv_nsec, &token);
3820 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3821 inode->i_ctime.tv_sec, &token);
3822 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3823 inode->i_ctime.tv_nsec, &token);
3825 btrfs_set_token_timespec_sec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_sec, &token);
3827 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3828 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3830 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3832 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3834 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3835 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3836 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3837 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3838 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3842 * copy everything in the in-memory inode into the btree.
3844 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3845 struct btrfs_root *root, struct inode *inode)
3847 struct btrfs_inode_item *inode_item;
3848 struct btrfs_path *path;
3849 struct extent_buffer *leaf;
3852 path = btrfs_alloc_path();
3856 path->leave_spinning = 1;
3857 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3865 leaf = path->nodes[0];
3866 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3867 struct btrfs_inode_item);
3869 fill_inode_item(trans, leaf, inode_item, inode);
3870 btrfs_mark_buffer_dirty(leaf);
3871 btrfs_set_inode_last_trans(trans, inode);
3874 btrfs_free_path(path);
3879 * copy everything in the in-memory inode into the btree.
3881 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3882 struct btrfs_root *root, struct inode *inode)
3887 * If the inode is a free space inode, we can deadlock during commit
3888 * if we put it into the delayed code.
3890 * The data relocation inode should also be directly updated
3893 if (!btrfs_is_free_space_inode(inode)
3894 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3895 && !root->fs_info->log_root_recovering) {
3896 btrfs_update_root_times(trans, root);
3898 ret = btrfs_delayed_update_inode(trans, root, inode);
3900 btrfs_set_inode_last_trans(trans, inode);
3904 return btrfs_update_inode_item(trans, root, inode);
3907 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *root,
3909 struct inode *inode)
3913 ret = btrfs_update_inode(trans, root, inode);
3915 return btrfs_update_inode_item(trans, root, inode);
3920 * unlink helper that gets used here in inode.c and in the tree logging
3921 * recovery code. It remove a link in a directory with a given name, and
3922 * also drops the back refs in the inode to the directory
3924 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3925 struct btrfs_root *root,
3926 struct inode *dir, struct inode *inode,
3927 const char *name, int name_len)
3929 struct btrfs_path *path;
3931 struct extent_buffer *leaf;
3932 struct btrfs_dir_item *di;
3933 struct btrfs_key key;
3935 u64 ino = btrfs_ino(inode);
3936 u64 dir_ino = btrfs_ino(dir);
3938 path = btrfs_alloc_path();
3944 path->leave_spinning = 1;
3945 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3946 name, name_len, -1);
3955 leaf = path->nodes[0];
3956 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3957 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3960 btrfs_release_path(path);
3963 * If we don't have dir index, we have to get it by looking up
3964 * the inode ref, since we get the inode ref, remove it directly,
3965 * it is unnecessary to do delayed deletion.
3967 * But if we have dir index, needn't search inode ref to get it.
3968 * Since the inode ref is close to the inode item, it is better
3969 * that we delay to delete it, and just do this deletion when
3970 * we update the inode item.
3972 if (BTRFS_I(inode)->dir_index) {
3973 ret = btrfs_delayed_delete_inode_ref(inode);
3975 index = BTRFS_I(inode)->dir_index;
3980 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3983 btrfs_info(root->fs_info,
3984 "failed to delete reference to %.*s, inode %llu parent %llu",
3985 name_len, name, ino, dir_ino);
3986 btrfs_abort_transaction(trans, root, ret);
3990 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3992 btrfs_abort_transaction(trans, root, ret);
3996 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3998 if (ret != 0 && ret != -ENOENT) {
3999 btrfs_abort_transaction(trans, root, ret);
4003 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4008 btrfs_abort_transaction(trans, root, ret);
4010 btrfs_free_path(path);
4014 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4015 inode_inc_iversion(inode);
4016 inode_inc_iversion(dir);
4017 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4018 ret = btrfs_update_inode(trans, root, dir);
4023 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4024 struct btrfs_root *root,
4025 struct inode *dir, struct inode *inode,
4026 const char *name, int name_len)
4029 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4032 ret = btrfs_update_inode(trans, root, inode);
4038 * helper to start transaction for unlink and rmdir.
4040 * unlink and rmdir are special in btrfs, they do not always free space, so
4041 * if we cannot make our reservations the normal way try and see if there is
4042 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4043 * allow the unlink to occur.
4045 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4047 struct btrfs_root *root = BTRFS_I(dir)->root;
4050 * 1 for the possible orphan item
4051 * 1 for the dir item
4052 * 1 for the dir index
4053 * 1 for the inode ref
4056 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4059 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4061 struct btrfs_root *root = BTRFS_I(dir)->root;
4062 struct btrfs_trans_handle *trans;
4063 struct inode *inode = d_inode(dentry);
4066 trans = __unlink_start_trans(dir);
4068 return PTR_ERR(trans);
4070 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4072 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4073 dentry->d_name.name, dentry->d_name.len);
4077 if (inode->i_nlink == 0) {
4078 ret = btrfs_orphan_add(trans, inode);
4084 btrfs_end_transaction(trans, root);
4085 btrfs_btree_balance_dirty(root);
4089 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4090 struct btrfs_root *root,
4091 struct inode *dir, u64 objectid,
4092 const char *name, int name_len)
4094 struct btrfs_path *path;
4095 struct extent_buffer *leaf;
4096 struct btrfs_dir_item *di;
4097 struct btrfs_key key;
4100 u64 dir_ino = btrfs_ino(dir);
4102 path = btrfs_alloc_path();
4106 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4107 name, name_len, -1);
4108 if (IS_ERR_OR_NULL(di)) {
4116 leaf = path->nodes[0];
4117 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4118 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4119 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4121 btrfs_abort_transaction(trans, root, ret);
4124 btrfs_release_path(path);
4126 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4127 objectid, root->root_key.objectid,
4128 dir_ino, &index, name, name_len);
4130 if (ret != -ENOENT) {
4131 btrfs_abort_transaction(trans, root, ret);
4134 di = btrfs_search_dir_index_item(root, path, dir_ino,
4136 if (IS_ERR_OR_NULL(di)) {
4141 btrfs_abort_transaction(trans, root, ret);
4145 leaf = path->nodes[0];
4146 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4147 btrfs_release_path(path);
4150 btrfs_release_path(path);
4152 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4154 btrfs_abort_transaction(trans, root, ret);
4158 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4159 inode_inc_iversion(dir);
4160 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4161 ret = btrfs_update_inode_fallback(trans, root, dir);
4163 btrfs_abort_transaction(trans, root, ret);
4165 btrfs_free_path(path);
4169 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4171 struct inode *inode = d_inode(dentry);
4173 struct btrfs_root *root = BTRFS_I(dir)->root;
4174 struct btrfs_trans_handle *trans;
4176 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4178 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4181 trans = __unlink_start_trans(dir);
4183 return PTR_ERR(trans);
4185 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4186 err = btrfs_unlink_subvol(trans, root, dir,
4187 BTRFS_I(inode)->location.objectid,
4188 dentry->d_name.name,
4189 dentry->d_name.len);
4193 err = btrfs_orphan_add(trans, inode);
4197 /* now the directory is empty */
4198 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4199 dentry->d_name.name, dentry->d_name.len);
4201 btrfs_i_size_write(inode, 0);
4203 btrfs_end_transaction(trans, root);
4204 btrfs_btree_balance_dirty(root);
4209 static int truncate_space_check(struct btrfs_trans_handle *trans,
4210 struct btrfs_root *root,
4215 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4216 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4217 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4219 trans->bytes_reserved += bytes_deleted;
4224 static int truncate_inline_extent(struct inode *inode,
4225 struct btrfs_path *path,
4226 struct btrfs_key *found_key,
4230 struct extent_buffer *leaf = path->nodes[0];
4231 int slot = path->slots[0];
4232 struct btrfs_file_extent_item *fi;
4233 u32 size = (u32)(new_size - found_key->offset);
4234 struct btrfs_root *root = BTRFS_I(inode)->root;
4236 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4238 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4239 loff_t offset = new_size;
4240 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4243 * Zero out the remaining of the last page of our inline extent,
4244 * instead of directly truncating our inline extent here - that
4245 * would be much more complex (decompressing all the data, then
4246 * compressing the truncated data, which might be bigger than
4247 * the size of the inline extent, resize the extent, etc).
4248 * We release the path because to get the page we might need to
4249 * read the extent item from disk (data not in the page cache).
4251 btrfs_release_path(path);
4252 return btrfs_truncate_block(inode, offset, page_end - offset,
4256 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4257 size = btrfs_file_extent_calc_inline_size(size);
4258 btrfs_truncate_item(root, path, size, 1);
4260 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4261 inode_sub_bytes(inode, item_end + 1 - new_size);
4267 * this can truncate away extent items, csum items and directory items.
4268 * It starts at a high offset and removes keys until it can't find
4269 * any higher than new_size
4271 * csum items that cross the new i_size are truncated to the new size
4274 * min_type is the minimum key type to truncate down to. If set to 0, this
4275 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4277 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4278 struct btrfs_root *root,
4279 struct inode *inode,
4280 u64 new_size, u32 min_type)
4282 struct btrfs_path *path;
4283 struct extent_buffer *leaf;
4284 struct btrfs_file_extent_item *fi;
4285 struct btrfs_key key;
4286 struct btrfs_key found_key;
4287 u64 extent_start = 0;
4288 u64 extent_num_bytes = 0;
4289 u64 extent_offset = 0;
4291 u64 last_size = new_size;
4292 u32 found_type = (u8)-1;
4295 int pending_del_nr = 0;
4296 int pending_del_slot = 0;
4297 int extent_type = -1;
4300 u64 ino = btrfs_ino(inode);
4301 u64 bytes_deleted = 0;
4303 bool should_throttle = 0;
4304 bool should_end = 0;
4306 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4309 * for non-free space inodes and ref cows, we want to back off from
4312 if (!btrfs_is_free_space_inode(inode) &&
4313 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4316 path = btrfs_alloc_path();
4319 path->reada = READA_BACK;
4322 * We want to drop from the next block forward in case this new size is
4323 * not block aligned since we will be keeping the last block of the
4324 * extent just the way it is.
4326 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4327 root == root->fs_info->tree_root)
4328 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4329 root->sectorsize), (u64)-1, 0);
4332 * This function is also used to drop the items in the log tree before
4333 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4334 * it is used to drop the loged items. So we shouldn't kill the delayed
4337 if (min_type == 0 && root == BTRFS_I(inode)->root)
4338 btrfs_kill_delayed_inode_items(inode);
4341 key.offset = (u64)-1;
4346 * with a 16K leaf size and 128MB extents, you can actually queue
4347 * up a huge file in a single leaf. Most of the time that
4348 * bytes_deleted is > 0, it will be huge by the time we get here
4350 if (be_nice && bytes_deleted > SZ_32M) {
4351 if (btrfs_should_end_transaction(trans, root)) {
4358 path->leave_spinning = 1;
4359 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4366 /* there are no items in the tree for us to truncate, we're
4369 if (path->slots[0] == 0)
4376 leaf = path->nodes[0];
4377 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4378 found_type = found_key.type;
4380 if (found_key.objectid != ino)
4383 if (found_type < min_type)
4386 item_end = found_key.offset;
4387 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4388 fi = btrfs_item_ptr(leaf, path->slots[0],
4389 struct btrfs_file_extent_item);
4390 extent_type = btrfs_file_extent_type(leaf, fi);
4391 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4393 btrfs_file_extent_num_bytes(leaf, fi);
4394 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4395 item_end += btrfs_file_extent_inline_len(leaf,
4396 path->slots[0], fi);
4400 if (found_type > min_type) {
4403 if (item_end < new_size)
4405 if (found_key.offset >= new_size)
4411 /* FIXME, shrink the extent if the ref count is only 1 */
4412 if (found_type != BTRFS_EXTENT_DATA_KEY)
4416 last_size = found_key.offset;
4418 last_size = new_size;
4420 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4422 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4424 u64 orig_num_bytes =
4425 btrfs_file_extent_num_bytes(leaf, fi);
4426 extent_num_bytes = ALIGN(new_size -
4429 btrfs_set_file_extent_num_bytes(leaf, fi,
4431 num_dec = (orig_num_bytes -
4433 if (test_bit(BTRFS_ROOT_REF_COWS,
4436 inode_sub_bytes(inode, num_dec);
4437 btrfs_mark_buffer_dirty(leaf);
4440 btrfs_file_extent_disk_num_bytes(leaf,
4442 extent_offset = found_key.offset -
4443 btrfs_file_extent_offset(leaf, fi);
4445 /* FIXME blocksize != 4096 */
4446 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4447 if (extent_start != 0) {
4449 if (test_bit(BTRFS_ROOT_REF_COWS,
4451 inode_sub_bytes(inode, num_dec);
4454 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4456 * we can't truncate inline items that have had
4460 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4461 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4464 * Need to release path in order to truncate a
4465 * compressed extent. So delete any accumulated
4466 * extent items so far.
4468 if (btrfs_file_extent_compression(leaf, fi) !=
4469 BTRFS_COMPRESS_NONE && pending_del_nr) {
4470 err = btrfs_del_items(trans, root, path,
4474 btrfs_abort_transaction(trans,
4482 err = truncate_inline_extent(inode, path,
4487 btrfs_abort_transaction(trans,
4491 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4493 inode_sub_bytes(inode, item_end + 1 - new_size);
4498 if (!pending_del_nr) {
4499 /* no pending yet, add ourselves */
4500 pending_del_slot = path->slots[0];
4502 } else if (pending_del_nr &&
4503 path->slots[0] + 1 == pending_del_slot) {
4504 /* hop on the pending chunk */
4506 pending_del_slot = path->slots[0];
4513 should_throttle = 0;
4516 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4517 root == root->fs_info->tree_root)) {
4518 btrfs_set_path_blocking(path);
4519 bytes_deleted += extent_num_bytes;
4520 ret = btrfs_free_extent(trans, root, extent_start,
4521 extent_num_bytes, 0,
4522 btrfs_header_owner(leaf),
4523 ino, extent_offset);
4525 if (btrfs_should_throttle_delayed_refs(trans, root))
4526 btrfs_async_run_delayed_refs(root,
4527 trans->delayed_ref_updates * 2, 0);
4529 if (truncate_space_check(trans, root,
4530 extent_num_bytes)) {
4533 if (btrfs_should_throttle_delayed_refs(trans,
4535 should_throttle = 1;
4540 if (found_type == BTRFS_INODE_ITEM_KEY)
4543 if (path->slots[0] == 0 ||
4544 path->slots[0] != pending_del_slot ||
4545 should_throttle || should_end) {
4546 if (pending_del_nr) {
4547 ret = btrfs_del_items(trans, root, path,
4551 btrfs_abort_transaction(trans,
4557 btrfs_release_path(path);
4558 if (should_throttle) {
4559 unsigned long updates = trans->delayed_ref_updates;
4561 trans->delayed_ref_updates = 0;
4562 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4568 * if we failed to refill our space rsv, bail out
4569 * and let the transaction restart
4581 if (pending_del_nr) {
4582 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4585 btrfs_abort_transaction(trans, root, ret);
4588 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4589 btrfs_ordered_update_i_size(inode, last_size, NULL);
4591 btrfs_free_path(path);
4593 if (be_nice && bytes_deleted > SZ_32M) {
4594 unsigned long updates = trans->delayed_ref_updates;
4596 trans->delayed_ref_updates = 0;
4597 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4606 * btrfs_truncate_block - read, zero a chunk and write a block
4607 * @inode - inode that we're zeroing
4608 * @from - the offset to start zeroing
4609 * @len - the length to zero, 0 to zero the entire range respective to the
4611 * @front - zero up to the offset instead of from the offset on
4613 * This will find the block for the "from" offset and cow the block and zero the
4614 * part we want to zero. This is used with truncate and hole punching.
4616 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4619 struct address_space *mapping = inode->i_mapping;
4620 struct btrfs_root *root = BTRFS_I(inode)->root;
4621 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4622 struct btrfs_ordered_extent *ordered;
4623 struct extent_state *cached_state = NULL;
4625 u32 blocksize = root->sectorsize;
4626 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4627 unsigned offset = from & (blocksize - 1);
4629 gfp_t mask = btrfs_alloc_write_mask(mapping);
4634 if ((offset & (blocksize - 1)) == 0 &&
4635 (!len || ((len & (blocksize - 1)) == 0)))
4638 ret = btrfs_delalloc_reserve_space(inode,
4639 round_down(from, blocksize), blocksize);
4644 page = find_or_create_page(mapping, index, mask);
4646 btrfs_delalloc_release_space(inode,
4647 round_down(from, blocksize),
4653 block_start = round_down(from, blocksize);
4654 block_end = block_start + blocksize - 1;
4656 if (!PageUptodate(page)) {
4657 ret = btrfs_readpage(NULL, page);
4659 if (page->mapping != mapping) {
4661 page_cache_release(page);
4664 if (!PageUptodate(page)) {
4669 wait_on_page_writeback(page);
4671 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4672 set_page_extent_mapped(page);
4674 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4676 unlock_extent_cached(io_tree, block_start, block_end,
4677 &cached_state, GFP_NOFS);
4679 page_cache_release(page);
4680 btrfs_start_ordered_extent(inode, ordered, 1);
4681 btrfs_put_ordered_extent(ordered);
4685 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4686 EXTENT_DIRTY | EXTENT_DELALLOC |
4687 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4688 0, 0, &cached_state, GFP_NOFS);
4690 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4693 unlock_extent_cached(io_tree, block_start, block_end,
4694 &cached_state, GFP_NOFS);
4698 if (offset != blocksize) {
4700 len = blocksize - offset;
4703 memset(kaddr + (block_start - page_offset(page)),
4706 memset(kaddr + (block_start - page_offset(page)) + offset,
4708 flush_dcache_page(page);
4711 ClearPageChecked(page);
4712 set_page_dirty(page);
4713 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4718 btrfs_delalloc_release_space(inode, block_start,
4721 page_cache_release(page);
4726 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4727 u64 offset, u64 len)
4729 struct btrfs_trans_handle *trans;
4733 * Still need to make sure the inode looks like it's been updated so
4734 * that any holes get logged if we fsync.
4736 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4737 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4738 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4739 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4744 * 1 - for the one we're dropping
4745 * 1 - for the one we're adding
4746 * 1 - for updating the inode.
4748 trans = btrfs_start_transaction(root, 3);
4750 return PTR_ERR(trans);
4752 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4754 btrfs_abort_transaction(trans, root, ret);
4755 btrfs_end_transaction(trans, root);
4759 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4760 0, 0, len, 0, len, 0, 0, 0);
4762 btrfs_abort_transaction(trans, root, ret);
4764 btrfs_update_inode(trans, root, inode);
4765 btrfs_end_transaction(trans, root);
4770 * This function puts in dummy file extents for the area we're creating a hole
4771 * for. So if we are truncating this file to a larger size we need to insert
4772 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4773 * the range between oldsize and size
4775 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4777 struct btrfs_root *root = BTRFS_I(inode)->root;
4778 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4779 struct extent_map *em = NULL;
4780 struct extent_state *cached_state = NULL;
4781 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4782 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4783 u64 block_end = ALIGN(size, root->sectorsize);
4790 * If our size started in the middle of a block we need to zero out the
4791 * rest of the block before we expand the i_size, otherwise we could
4792 * expose stale data.
4794 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4798 if (size <= hole_start)
4802 struct btrfs_ordered_extent *ordered;
4804 lock_extent_bits(io_tree, hole_start, block_end - 1,
4806 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4807 block_end - hole_start);
4810 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4811 &cached_state, GFP_NOFS);
4812 btrfs_start_ordered_extent(inode, ordered, 1);
4813 btrfs_put_ordered_extent(ordered);
4816 cur_offset = hole_start;
4818 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4819 block_end - cur_offset, 0);
4825 last_byte = min(extent_map_end(em), block_end);
4826 last_byte = ALIGN(last_byte , root->sectorsize);
4827 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4828 struct extent_map *hole_em;
4829 hole_size = last_byte - cur_offset;
4831 err = maybe_insert_hole(root, inode, cur_offset,
4835 btrfs_drop_extent_cache(inode, cur_offset,
4836 cur_offset + hole_size - 1, 0);
4837 hole_em = alloc_extent_map();
4839 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4840 &BTRFS_I(inode)->runtime_flags);
4843 hole_em->start = cur_offset;
4844 hole_em->len = hole_size;
4845 hole_em->orig_start = cur_offset;
4847 hole_em->block_start = EXTENT_MAP_HOLE;
4848 hole_em->block_len = 0;
4849 hole_em->orig_block_len = 0;
4850 hole_em->ram_bytes = hole_size;
4851 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4852 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4853 hole_em->generation = root->fs_info->generation;
4856 write_lock(&em_tree->lock);
4857 err = add_extent_mapping(em_tree, hole_em, 1);
4858 write_unlock(&em_tree->lock);
4861 btrfs_drop_extent_cache(inode, cur_offset,
4865 free_extent_map(hole_em);
4868 free_extent_map(em);
4870 cur_offset = last_byte;
4871 if (cur_offset >= block_end)
4874 free_extent_map(em);
4875 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4880 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4882 struct btrfs_root *root = BTRFS_I(inode)->root;
4883 struct btrfs_trans_handle *trans;
4884 loff_t oldsize = i_size_read(inode);
4885 loff_t newsize = attr->ia_size;
4886 int mask = attr->ia_valid;
4890 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4891 * special case where we need to update the times despite not having
4892 * these flags set. For all other operations the VFS set these flags
4893 * explicitly if it wants a timestamp update.
4895 if (newsize != oldsize) {
4896 inode_inc_iversion(inode);
4897 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4898 inode->i_ctime = inode->i_mtime =
4899 current_fs_time(inode->i_sb);
4902 if (newsize > oldsize) {
4904 * Don't do an expanding truncate while snapshoting is ongoing.
4905 * This is to ensure the snapshot captures a fully consistent
4906 * state of this file - if the snapshot captures this expanding
4907 * truncation, it must capture all writes that happened before
4910 btrfs_wait_for_snapshot_creation(root);
4911 ret = btrfs_cont_expand(inode, oldsize, newsize);
4913 btrfs_end_write_no_snapshoting(root);
4917 trans = btrfs_start_transaction(root, 1);
4918 if (IS_ERR(trans)) {
4919 btrfs_end_write_no_snapshoting(root);
4920 return PTR_ERR(trans);
4923 i_size_write(inode, newsize);
4924 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4925 pagecache_isize_extended(inode, oldsize, newsize);
4926 ret = btrfs_update_inode(trans, root, inode);
4927 btrfs_end_write_no_snapshoting(root);
4928 btrfs_end_transaction(trans, root);
4932 * We're truncating a file that used to have good data down to
4933 * zero. Make sure it gets into the ordered flush list so that
4934 * any new writes get down to disk quickly.
4937 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4938 &BTRFS_I(inode)->runtime_flags);
4941 * 1 for the orphan item we're going to add
4942 * 1 for the orphan item deletion.
4944 trans = btrfs_start_transaction(root, 2);
4946 return PTR_ERR(trans);
4949 * We need to do this in case we fail at _any_ point during the
4950 * actual truncate. Once we do the truncate_setsize we could
4951 * invalidate pages which forces any outstanding ordered io to
4952 * be instantly completed which will give us extents that need
4953 * to be truncated. If we fail to get an orphan inode down we
4954 * could have left over extents that were never meant to live,
4955 * so we need to garuntee from this point on that everything
4956 * will be consistent.
4958 ret = btrfs_orphan_add(trans, inode);
4959 btrfs_end_transaction(trans, root);
4963 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4964 truncate_setsize(inode, newsize);
4966 /* Disable nonlocked read DIO to avoid the end less truncate */
4967 btrfs_inode_block_unlocked_dio(inode);
4968 inode_dio_wait(inode);
4969 btrfs_inode_resume_unlocked_dio(inode);
4971 ret = btrfs_truncate(inode);
4972 if (ret && inode->i_nlink) {
4976 * failed to truncate, disk_i_size is only adjusted down
4977 * as we remove extents, so it should represent the true
4978 * size of the inode, so reset the in memory size and
4979 * delete our orphan entry.
4981 trans = btrfs_join_transaction(root);
4982 if (IS_ERR(trans)) {
4983 btrfs_orphan_del(NULL, inode);
4986 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4987 err = btrfs_orphan_del(trans, inode);
4989 btrfs_abort_transaction(trans, root, err);
4990 btrfs_end_transaction(trans, root);
4997 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4999 struct inode *inode = d_inode(dentry);
5000 struct btrfs_root *root = BTRFS_I(inode)->root;
5003 if (btrfs_root_readonly(root))
5006 err = inode_change_ok(inode, attr);
5010 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5011 err = btrfs_setsize(inode, attr);
5016 if (attr->ia_valid) {
5017 setattr_copy(inode, attr);
5018 inode_inc_iversion(inode);
5019 err = btrfs_dirty_inode(inode);
5021 if (!err && attr->ia_valid & ATTR_MODE)
5022 err = posix_acl_chmod(inode, inode->i_mode);
5029 * While truncating the inode pages during eviction, we get the VFS calling
5030 * btrfs_invalidatepage() against each page of the inode. This is slow because
5031 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5032 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5033 * extent_state structures over and over, wasting lots of time.
5035 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5036 * those expensive operations on a per page basis and do only the ordered io
5037 * finishing, while we release here the extent_map and extent_state structures,
5038 * without the excessive merging and splitting.
5040 static void evict_inode_truncate_pages(struct inode *inode)
5042 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5043 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5044 struct rb_node *node;
5046 ASSERT(inode->i_state & I_FREEING);
5047 truncate_inode_pages_final(&inode->i_data);
5049 write_lock(&map_tree->lock);
5050 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5051 struct extent_map *em;
5053 node = rb_first(&map_tree->map);
5054 em = rb_entry(node, struct extent_map, rb_node);
5055 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5056 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5057 remove_extent_mapping(map_tree, em);
5058 free_extent_map(em);
5059 if (need_resched()) {
5060 write_unlock(&map_tree->lock);
5062 write_lock(&map_tree->lock);
5065 write_unlock(&map_tree->lock);
5068 * Keep looping until we have no more ranges in the io tree.
5069 * We can have ongoing bios started by readpages (called from readahead)
5070 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5071 * still in progress (unlocked the pages in the bio but did not yet
5072 * unlocked the ranges in the io tree). Therefore this means some
5073 * ranges can still be locked and eviction started because before
5074 * submitting those bios, which are executed by a separate task (work
5075 * queue kthread), inode references (inode->i_count) were not taken
5076 * (which would be dropped in the end io callback of each bio).
5077 * Therefore here we effectively end up waiting for those bios and
5078 * anyone else holding locked ranges without having bumped the inode's
5079 * reference count - if we don't do it, when they access the inode's
5080 * io_tree to unlock a range it may be too late, leading to an
5081 * use-after-free issue.
5083 spin_lock(&io_tree->lock);
5084 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5085 struct extent_state *state;
5086 struct extent_state *cached_state = NULL;
5090 node = rb_first(&io_tree->state);
5091 state = rb_entry(node, struct extent_state, rb_node);
5092 start = state->start;
5094 spin_unlock(&io_tree->lock);
5096 lock_extent_bits(io_tree, start, end, &cached_state);
5099 * If still has DELALLOC flag, the extent didn't reach disk,
5100 * and its reserved space won't be freed by delayed_ref.
5101 * So we need to free its reserved space here.
5102 * (Refer to comment in btrfs_invalidatepage, case 2)
5104 * Note, end is the bytenr of last byte, so we need + 1 here.
5106 if (state->state & EXTENT_DELALLOC)
5107 btrfs_qgroup_free_data(inode, start, end - start + 1);
5109 clear_extent_bit(io_tree, start, end,
5110 EXTENT_LOCKED | EXTENT_DIRTY |
5111 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5112 EXTENT_DEFRAG, 1, 1,
5113 &cached_state, GFP_NOFS);
5116 spin_lock(&io_tree->lock);
5118 spin_unlock(&io_tree->lock);
5121 void btrfs_evict_inode(struct inode *inode)
5123 struct btrfs_trans_handle *trans;
5124 struct btrfs_root *root = BTRFS_I(inode)->root;
5125 struct btrfs_block_rsv *rsv, *global_rsv;
5126 int steal_from_global = 0;
5127 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5130 trace_btrfs_inode_evict(inode);
5132 evict_inode_truncate_pages(inode);
5134 if (inode->i_nlink &&
5135 ((btrfs_root_refs(&root->root_item) != 0 &&
5136 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5137 btrfs_is_free_space_inode(inode)))
5140 if (is_bad_inode(inode)) {
5141 btrfs_orphan_del(NULL, inode);
5144 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5145 if (!special_file(inode->i_mode))
5146 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5148 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5150 if (root->fs_info->log_root_recovering) {
5151 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5152 &BTRFS_I(inode)->runtime_flags));
5156 if (inode->i_nlink > 0) {
5157 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5158 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5162 ret = btrfs_commit_inode_delayed_inode(inode);
5164 btrfs_orphan_del(NULL, inode);
5168 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5170 btrfs_orphan_del(NULL, inode);
5173 rsv->size = min_size;
5175 global_rsv = &root->fs_info->global_block_rsv;
5177 btrfs_i_size_write(inode, 0);
5180 * This is a bit simpler than btrfs_truncate since we've already
5181 * reserved our space for our orphan item in the unlink, so we just
5182 * need to reserve some slack space in case we add bytes and update
5183 * inode item when doing the truncate.
5186 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5187 BTRFS_RESERVE_FLUSH_LIMIT);
5190 * Try and steal from the global reserve since we will
5191 * likely not use this space anyway, we want to try as
5192 * hard as possible to get this to work.
5195 steal_from_global++;
5197 steal_from_global = 0;
5201 * steal_from_global == 0: we reserved stuff, hooray!
5202 * steal_from_global == 1: we didn't reserve stuff, boo!
5203 * steal_from_global == 2: we've committed, still not a lot of
5204 * room but maybe we'll have room in the global reserve this
5206 * steal_from_global == 3: abandon all hope!
5208 if (steal_from_global > 2) {
5209 btrfs_warn(root->fs_info,
5210 "Could not get space for a delete, will truncate on mount %d",
5212 btrfs_orphan_del(NULL, inode);
5213 btrfs_free_block_rsv(root, rsv);
5217 trans = btrfs_join_transaction(root);
5218 if (IS_ERR(trans)) {
5219 btrfs_orphan_del(NULL, inode);
5220 btrfs_free_block_rsv(root, rsv);
5225 * We can't just steal from the global reserve, we need tomake
5226 * sure there is room to do it, if not we need to commit and try
5229 if (steal_from_global) {
5230 if (!btrfs_check_space_for_delayed_refs(trans, root))
5231 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5238 * Couldn't steal from the global reserve, we have too much
5239 * pending stuff built up, commit the transaction and try it
5243 ret = btrfs_commit_transaction(trans, root);
5245 btrfs_orphan_del(NULL, inode);
5246 btrfs_free_block_rsv(root, rsv);
5251 steal_from_global = 0;
5254 trans->block_rsv = rsv;
5256 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5257 if (ret != -ENOSPC && ret != -EAGAIN)
5260 trans->block_rsv = &root->fs_info->trans_block_rsv;
5261 btrfs_end_transaction(trans, root);
5263 btrfs_btree_balance_dirty(root);
5266 btrfs_free_block_rsv(root, rsv);
5269 * Errors here aren't a big deal, it just means we leave orphan items
5270 * in the tree. They will be cleaned up on the next mount.
5273 trans->block_rsv = root->orphan_block_rsv;
5274 btrfs_orphan_del(trans, inode);
5276 btrfs_orphan_del(NULL, inode);
5279 trans->block_rsv = &root->fs_info->trans_block_rsv;
5280 if (!(root == root->fs_info->tree_root ||
5281 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5282 btrfs_return_ino(root, btrfs_ino(inode));
5284 btrfs_end_transaction(trans, root);
5285 btrfs_btree_balance_dirty(root);
5287 btrfs_remove_delayed_node(inode);
5292 * this returns the key found in the dir entry in the location pointer.
5293 * If no dir entries were found, location->objectid is 0.
5295 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5296 struct btrfs_key *location)
5298 const char *name = dentry->d_name.name;
5299 int namelen = dentry->d_name.len;
5300 struct btrfs_dir_item *di;
5301 struct btrfs_path *path;
5302 struct btrfs_root *root = BTRFS_I(dir)->root;
5305 path = btrfs_alloc_path();
5309 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5314 if (IS_ERR_OR_NULL(di))
5317 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5319 btrfs_free_path(path);
5322 location->objectid = 0;
5327 * when we hit a tree root in a directory, the btrfs part of the inode
5328 * needs to be changed to reflect the root directory of the tree root. This
5329 * is kind of like crossing a mount point.
5331 static int fixup_tree_root_location(struct btrfs_root *root,
5333 struct dentry *dentry,
5334 struct btrfs_key *location,
5335 struct btrfs_root **sub_root)
5337 struct btrfs_path *path;
5338 struct btrfs_root *new_root;
5339 struct btrfs_root_ref *ref;
5340 struct extent_buffer *leaf;
5341 struct btrfs_key key;
5345 path = btrfs_alloc_path();
5352 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5353 key.type = BTRFS_ROOT_REF_KEY;
5354 key.offset = location->objectid;
5356 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5364 leaf = path->nodes[0];
5365 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5366 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5367 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5370 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5371 (unsigned long)(ref + 1),
5372 dentry->d_name.len);
5376 btrfs_release_path(path);
5378 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5379 if (IS_ERR(new_root)) {
5380 err = PTR_ERR(new_root);
5384 *sub_root = new_root;
5385 location->objectid = btrfs_root_dirid(&new_root->root_item);
5386 location->type = BTRFS_INODE_ITEM_KEY;
5387 location->offset = 0;
5390 btrfs_free_path(path);
5394 static void inode_tree_add(struct inode *inode)
5396 struct btrfs_root *root = BTRFS_I(inode)->root;
5397 struct btrfs_inode *entry;
5399 struct rb_node *parent;
5400 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5401 u64 ino = btrfs_ino(inode);
5403 if (inode_unhashed(inode))
5406 spin_lock(&root->inode_lock);
5407 p = &root->inode_tree.rb_node;
5410 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5412 if (ino < btrfs_ino(&entry->vfs_inode))
5413 p = &parent->rb_left;
5414 else if (ino > btrfs_ino(&entry->vfs_inode))
5415 p = &parent->rb_right;
5417 WARN_ON(!(entry->vfs_inode.i_state &
5418 (I_WILL_FREE | I_FREEING)));
5419 rb_replace_node(parent, new, &root->inode_tree);
5420 RB_CLEAR_NODE(parent);
5421 spin_unlock(&root->inode_lock);
5425 rb_link_node(new, parent, p);
5426 rb_insert_color(new, &root->inode_tree);
5427 spin_unlock(&root->inode_lock);
5430 static void inode_tree_del(struct inode *inode)
5432 struct btrfs_root *root = BTRFS_I(inode)->root;
5435 spin_lock(&root->inode_lock);
5436 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5437 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5438 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5439 empty = RB_EMPTY_ROOT(&root->inode_tree);
5441 spin_unlock(&root->inode_lock);
5443 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5444 synchronize_srcu(&root->fs_info->subvol_srcu);
5445 spin_lock(&root->inode_lock);
5446 empty = RB_EMPTY_ROOT(&root->inode_tree);
5447 spin_unlock(&root->inode_lock);
5449 btrfs_add_dead_root(root);
5453 void btrfs_invalidate_inodes(struct btrfs_root *root)
5455 struct rb_node *node;
5456 struct rb_node *prev;
5457 struct btrfs_inode *entry;
5458 struct inode *inode;
5461 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5462 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5464 spin_lock(&root->inode_lock);
5466 node = root->inode_tree.rb_node;
5470 entry = rb_entry(node, struct btrfs_inode, rb_node);
5472 if (objectid < btrfs_ino(&entry->vfs_inode))
5473 node = node->rb_left;
5474 else if (objectid > btrfs_ino(&entry->vfs_inode))
5475 node = node->rb_right;
5481 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5482 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5486 prev = rb_next(prev);
5490 entry = rb_entry(node, struct btrfs_inode, rb_node);
5491 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5492 inode = igrab(&entry->vfs_inode);
5494 spin_unlock(&root->inode_lock);
5495 if (atomic_read(&inode->i_count) > 1)
5496 d_prune_aliases(inode);
5498 * btrfs_drop_inode will have it removed from
5499 * the inode cache when its usage count
5504 spin_lock(&root->inode_lock);
5508 if (cond_resched_lock(&root->inode_lock))
5511 node = rb_next(node);
5513 spin_unlock(&root->inode_lock);
5516 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5518 struct btrfs_iget_args *args = p;
5519 inode->i_ino = args->location->objectid;
5520 memcpy(&BTRFS_I(inode)->location, args->location,
5521 sizeof(*args->location));
5522 BTRFS_I(inode)->root = args->root;
5526 static int btrfs_find_actor(struct inode *inode, void *opaque)
5528 struct btrfs_iget_args *args = opaque;
5529 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5530 args->root == BTRFS_I(inode)->root;
5533 static struct inode *btrfs_iget_locked(struct super_block *s,
5534 struct btrfs_key *location,
5535 struct btrfs_root *root)
5537 struct inode *inode;
5538 struct btrfs_iget_args args;
5539 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5541 args.location = location;
5544 inode = iget5_locked(s, hashval, btrfs_find_actor,
5545 btrfs_init_locked_inode,
5550 /* Get an inode object given its location and corresponding root.
5551 * Returns in *is_new if the inode was read from disk
5553 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5554 struct btrfs_root *root, int *new)
5556 struct inode *inode;
5558 inode = btrfs_iget_locked(s, location, root);
5560 return ERR_PTR(-ENOMEM);
5562 if (inode->i_state & I_NEW) {
5563 btrfs_read_locked_inode(inode);
5564 if (!is_bad_inode(inode)) {
5565 inode_tree_add(inode);
5566 unlock_new_inode(inode);
5570 unlock_new_inode(inode);
5572 inode = ERR_PTR(-ESTALE);
5579 static struct inode *new_simple_dir(struct super_block *s,
5580 struct btrfs_key *key,
5581 struct btrfs_root *root)
5583 struct inode *inode = new_inode(s);
5586 return ERR_PTR(-ENOMEM);
5588 BTRFS_I(inode)->root = root;
5589 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5590 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5592 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5593 inode->i_op = &btrfs_dir_ro_inode_operations;
5594 inode->i_fop = &simple_dir_operations;
5595 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5596 inode->i_mtime = CURRENT_TIME;
5597 inode->i_atime = inode->i_mtime;
5598 inode->i_ctime = inode->i_mtime;
5599 BTRFS_I(inode)->i_otime = inode->i_mtime;
5604 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5606 struct inode *inode;
5607 struct btrfs_root *root = BTRFS_I(dir)->root;
5608 struct btrfs_root *sub_root = root;
5609 struct btrfs_key location;
5613 if (dentry->d_name.len > BTRFS_NAME_LEN)
5614 return ERR_PTR(-ENAMETOOLONG);
5616 ret = btrfs_inode_by_name(dir, dentry, &location);
5618 return ERR_PTR(ret);
5620 if (location.objectid == 0)
5621 return ERR_PTR(-ENOENT);
5623 if (location.type == BTRFS_INODE_ITEM_KEY) {
5624 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5628 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5630 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5631 ret = fixup_tree_root_location(root, dir, dentry,
5632 &location, &sub_root);
5635 inode = ERR_PTR(ret);
5637 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5639 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5641 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5643 if (!IS_ERR(inode) && root != sub_root) {
5644 down_read(&root->fs_info->cleanup_work_sem);
5645 if (!(inode->i_sb->s_flags & MS_RDONLY))
5646 ret = btrfs_orphan_cleanup(sub_root);
5647 up_read(&root->fs_info->cleanup_work_sem);
5650 inode = ERR_PTR(ret);
5657 static int btrfs_dentry_delete(const struct dentry *dentry)
5659 struct btrfs_root *root;
5660 struct inode *inode = d_inode(dentry);
5662 if (!inode && !IS_ROOT(dentry))
5663 inode = d_inode(dentry->d_parent);
5666 root = BTRFS_I(inode)->root;
5667 if (btrfs_root_refs(&root->root_item) == 0)
5670 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5676 static void btrfs_dentry_release(struct dentry *dentry)
5678 kfree(dentry->d_fsdata);
5681 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5684 struct inode *inode;
5686 inode = btrfs_lookup_dentry(dir, dentry);
5687 if (IS_ERR(inode)) {
5688 if (PTR_ERR(inode) == -ENOENT)
5691 return ERR_CAST(inode);
5694 return d_splice_alias(inode, dentry);
5697 unsigned char btrfs_filetype_table[] = {
5698 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5701 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5703 struct inode *inode = file_inode(file);
5704 struct btrfs_root *root = BTRFS_I(inode)->root;
5705 struct btrfs_item *item;
5706 struct btrfs_dir_item *di;
5707 struct btrfs_key key;
5708 struct btrfs_key found_key;
5709 struct btrfs_path *path;
5710 struct list_head ins_list;
5711 struct list_head del_list;
5713 struct extent_buffer *leaf;
5715 unsigned char d_type;
5720 int key_type = BTRFS_DIR_INDEX_KEY;
5724 int is_curr = 0; /* ctx->pos points to the current index? */
5726 /* FIXME, use a real flag for deciding about the key type */
5727 if (root->fs_info->tree_root == root)
5728 key_type = BTRFS_DIR_ITEM_KEY;
5730 if (!dir_emit_dots(file, ctx))
5733 path = btrfs_alloc_path();
5737 path->reada = READA_FORWARD;
5739 if (key_type == BTRFS_DIR_INDEX_KEY) {
5740 INIT_LIST_HEAD(&ins_list);
5741 INIT_LIST_HEAD(&del_list);
5742 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5745 key.type = key_type;
5746 key.offset = ctx->pos;
5747 key.objectid = btrfs_ino(inode);
5749 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5754 leaf = path->nodes[0];
5755 slot = path->slots[0];
5756 if (slot >= btrfs_header_nritems(leaf)) {
5757 ret = btrfs_next_leaf(root, path);
5765 item = btrfs_item_nr(slot);
5766 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5768 if (found_key.objectid != key.objectid)
5770 if (found_key.type != key_type)
5772 if (found_key.offset < ctx->pos)
5774 if (key_type == BTRFS_DIR_INDEX_KEY &&
5775 btrfs_should_delete_dir_index(&del_list,
5779 ctx->pos = found_key.offset;
5782 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5784 di_total = btrfs_item_size(leaf, item);
5786 while (di_cur < di_total) {
5787 struct btrfs_key location;
5789 if (verify_dir_item(root, leaf, di))
5792 name_len = btrfs_dir_name_len(leaf, di);
5793 if (name_len <= sizeof(tmp_name)) {
5794 name_ptr = tmp_name;
5796 name_ptr = kmalloc(name_len, GFP_NOFS);
5802 read_extent_buffer(leaf, name_ptr,
5803 (unsigned long)(di + 1), name_len);
5805 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5806 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5809 /* is this a reference to our own snapshot? If so
5812 * In contrast to old kernels, we insert the snapshot's
5813 * dir item and dir index after it has been created, so
5814 * we won't find a reference to our own snapshot. We
5815 * still keep the following code for backward
5818 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5819 location.objectid == root->root_key.objectid) {
5823 over = !dir_emit(ctx, name_ptr, name_len,
5824 location.objectid, d_type);
5827 if (name_ptr != tmp_name)
5832 di_len = btrfs_dir_name_len(leaf, di) +
5833 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5835 di = (struct btrfs_dir_item *)((char *)di + di_len);
5841 if (key_type == BTRFS_DIR_INDEX_KEY) {
5844 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5849 /* Reached end of directory/root. Bump pos past the last item. */
5853 * Stop new entries from being returned after we return the last
5856 * New directory entries are assigned a strictly increasing
5857 * offset. This means that new entries created during readdir
5858 * are *guaranteed* to be seen in the future by that readdir.
5859 * This has broken buggy programs which operate on names as
5860 * they're returned by readdir. Until we re-use freed offsets
5861 * we have this hack to stop new entries from being returned
5862 * under the assumption that they'll never reach this huge
5865 * This is being careful not to overflow 32bit loff_t unless the
5866 * last entry requires it because doing so has broken 32bit apps
5869 if (key_type == BTRFS_DIR_INDEX_KEY) {
5870 if (ctx->pos >= INT_MAX)
5871 ctx->pos = LLONG_MAX;
5878 if (key_type == BTRFS_DIR_INDEX_KEY)
5879 btrfs_put_delayed_items(&ins_list, &del_list);
5880 btrfs_free_path(path);
5884 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5886 struct btrfs_root *root = BTRFS_I(inode)->root;
5887 struct btrfs_trans_handle *trans;
5889 bool nolock = false;
5891 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5894 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5897 if (wbc->sync_mode == WB_SYNC_ALL) {
5899 trans = btrfs_join_transaction_nolock(root);
5901 trans = btrfs_join_transaction(root);
5903 return PTR_ERR(trans);
5904 ret = btrfs_commit_transaction(trans, root);
5910 * This is somewhat expensive, updating the tree every time the
5911 * inode changes. But, it is most likely to find the inode in cache.
5912 * FIXME, needs more benchmarking...there are no reasons other than performance
5913 * to keep or drop this code.
5915 static int btrfs_dirty_inode(struct inode *inode)
5917 struct btrfs_root *root = BTRFS_I(inode)->root;
5918 struct btrfs_trans_handle *trans;
5921 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5924 trans = btrfs_join_transaction(root);
5926 return PTR_ERR(trans);
5928 ret = btrfs_update_inode(trans, root, inode);
5929 if (ret && ret == -ENOSPC) {
5930 /* whoops, lets try again with the full transaction */
5931 btrfs_end_transaction(trans, root);
5932 trans = btrfs_start_transaction(root, 1);
5934 return PTR_ERR(trans);
5936 ret = btrfs_update_inode(trans, root, inode);
5938 btrfs_end_transaction(trans, root);
5939 if (BTRFS_I(inode)->delayed_node)
5940 btrfs_balance_delayed_items(root);
5946 * This is a copy of file_update_time. We need this so we can return error on
5947 * ENOSPC for updating the inode in the case of file write and mmap writes.
5949 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5952 struct btrfs_root *root = BTRFS_I(inode)->root;
5954 if (btrfs_root_readonly(root))
5957 if (flags & S_VERSION)
5958 inode_inc_iversion(inode);
5959 if (flags & S_CTIME)
5960 inode->i_ctime = *now;
5961 if (flags & S_MTIME)
5962 inode->i_mtime = *now;
5963 if (flags & S_ATIME)
5964 inode->i_atime = *now;
5965 return btrfs_dirty_inode(inode);
5969 * find the highest existing sequence number in a directory
5970 * and then set the in-memory index_cnt variable to reflect
5971 * free sequence numbers
5973 static int btrfs_set_inode_index_count(struct inode *inode)
5975 struct btrfs_root *root = BTRFS_I(inode)->root;
5976 struct btrfs_key key, found_key;
5977 struct btrfs_path *path;
5978 struct extent_buffer *leaf;
5981 key.objectid = btrfs_ino(inode);
5982 key.type = BTRFS_DIR_INDEX_KEY;
5983 key.offset = (u64)-1;
5985 path = btrfs_alloc_path();
5989 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5992 /* FIXME: we should be able to handle this */
5998 * MAGIC NUMBER EXPLANATION:
5999 * since we search a directory based on f_pos we have to start at 2
6000 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6001 * else has to start at 2
6003 if (path->slots[0] == 0) {
6004 BTRFS_I(inode)->index_cnt = 2;
6010 leaf = path->nodes[0];
6011 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6013 if (found_key.objectid != btrfs_ino(inode) ||
6014 found_key.type != BTRFS_DIR_INDEX_KEY) {
6015 BTRFS_I(inode)->index_cnt = 2;
6019 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6021 btrfs_free_path(path);
6026 * helper to find a free sequence number in a given directory. This current
6027 * code is very simple, later versions will do smarter things in the btree
6029 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6033 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6034 ret = btrfs_inode_delayed_dir_index_count(dir);
6036 ret = btrfs_set_inode_index_count(dir);
6042 *index = BTRFS_I(dir)->index_cnt;
6043 BTRFS_I(dir)->index_cnt++;
6048 static int btrfs_insert_inode_locked(struct inode *inode)
6050 struct btrfs_iget_args args;
6051 args.location = &BTRFS_I(inode)->location;
6052 args.root = BTRFS_I(inode)->root;
6054 return insert_inode_locked4(inode,
6055 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6056 btrfs_find_actor, &args);
6059 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6060 struct btrfs_root *root,
6062 const char *name, int name_len,
6063 u64 ref_objectid, u64 objectid,
6064 umode_t mode, u64 *index)
6066 struct inode *inode;
6067 struct btrfs_inode_item *inode_item;
6068 struct btrfs_key *location;
6069 struct btrfs_path *path;
6070 struct btrfs_inode_ref *ref;
6071 struct btrfs_key key[2];
6073 int nitems = name ? 2 : 1;
6077 path = btrfs_alloc_path();
6079 return ERR_PTR(-ENOMEM);
6081 inode = new_inode(root->fs_info->sb);
6083 btrfs_free_path(path);
6084 return ERR_PTR(-ENOMEM);
6088 * O_TMPFILE, set link count to 0, so that after this point,
6089 * we fill in an inode item with the correct link count.
6092 set_nlink(inode, 0);
6095 * we have to initialize this early, so we can reclaim the inode
6096 * number if we fail afterwards in this function.
6098 inode->i_ino = objectid;
6101 trace_btrfs_inode_request(dir);
6103 ret = btrfs_set_inode_index(dir, index);
6105 btrfs_free_path(path);
6107 return ERR_PTR(ret);
6113 * index_cnt is ignored for everything but a dir,
6114 * btrfs_get_inode_index_count has an explanation for the magic
6117 BTRFS_I(inode)->index_cnt = 2;
6118 BTRFS_I(inode)->dir_index = *index;
6119 BTRFS_I(inode)->root = root;
6120 BTRFS_I(inode)->generation = trans->transid;
6121 inode->i_generation = BTRFS_I(inode)->generation;
6124 * We could have gotten an inode number from somebody who was fsynced
6125 * and then removed in this same transaction, so let's just set full
6126 * sync since it will be a full sync anyway and this will blow away the
6127 * old info in the log.
6129 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6131 key[0].objectid = objectid;
6132 key[0].type = BTRFS_INODE_ITEM_KEY;
6135 sizes[0] = sizeof(struct btrfs_inode_item);
6139 * Start new inodes with an inode_ref. This is slightly more
6140 * efficient for small numbers of hard links since they will
6141 * be packed into one item. Extended refs will kick in if we
6142 * add more hard links than can fit in the ref item.
6144 key[1].objectid = objectid;
6145 key[1].type = BTRFS_INODE_REF_KEY;
6146 key[1].offset = ref_objectid;
6148 sizes[1] = name_len + sizeof(*ref);
6151 location = &BTRFS_I(inode)->location;
6152 location->objectid = objectid;
6153 location->offset = 0;
6154 location->type = BTRFS_INODE_ITEM_KEY;
6156 ret = btrfs_insert_inode_locked(inode);
6160 path->leave_spinning = 1;
6161 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6165 inode_init_owner(inode, dir, mode);
6166 inode_set_bytes(inode, 0);
6168 inode->i_mtime = CURRENT_TIME;
6169 inode->i_atime = inode->i_mtime;
6170 inode->i_ctime = inode->i_mtime;
6171 BTRFS_I(inode)->i_otime = inode->i_mtime;
6173 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6174 struct btrfs_inode_item);
6175 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6176 sizeof(*inode_item));
6177 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6180 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6181 struct btrfs_inode_ref);
6182 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6183 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6184 ptr = (unsigned long)(ref + 1);
6185 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6188 btrfs_mark_buffer_dirty(path->nodes[0]);
6189 btrfs_free_path(path);
6191 btrfs_inherit_iflags(inode, dir);
6193 if (S_ISREG(mode)) {
6194 if (btrfs_test_opt(root, NODATASUM))
6195 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6196 if (btrfs_test_opt(root, NODATACOW))
6197 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6198 BTRFS_INODE_NODATASUM;
6201 inode_tree_add(inode);
6203 trace_btrfs_inode_new(inode);
6204 btrfs_set_inode_last_trans(trans, inode);
6206 btrfs_update_root_times(trans, root);
6208 ret = btrfs_inode_inherit_props(trans, inode, dir);
6210 btrfs_err(root->fs_info,
6211 "error inheriting props for ino %llu (root %llu): %d",
6212 btrfs_ino(inode), root->root_key.objectid, ret);
6217 unlock_new_inode(inode);
6220 BTRFS_I(dir)->index_cnt--;
6221 btrfs_free_path(path);
6223 return ERR_PTR(ret);
6226 static inline u8 btrfs_inode_type(struct inode *inode)
6228 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6232 * utility function to add 'inode' into 'parent_inode' with
6233 * a give name and a given sequence number.
6234 * if 'add_backref' is true, also insert a backref from the
6235 * inode to the parent directory.
6237 int btrfs_add_link(struct btrfs_trans_handle *trans,
6238 struct inode *parent_inode, struct inode *inode,
6239 const char *name, int name_len, int add_backref, u64 index)
6242 struct btrfs_key key;
6243 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6244 u64 ino = btrfs_ino(inode);
6245 u64 parent_ino = btrfs_ino(parent_inode);
6247 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6248 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6251 key.type = BTRFS_INODE_ITEM_KEY;
6255 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6256 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6257 key.objectid, root->root_key.objectid,
6258 parent_ino, index, name, name_len);
6259 } else if (add_backref) {
6260 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6264 /* Nothing to clean up yet */
6268 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6270 btrfs_inode_type(inode), index);
6271 if (ret == -EEXIST || ret == -EOVERFLOW)
6274 btrfs_abort_transaction(trans, root, ret);
6278 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6280 inode_inc_iversion(parent_inode);
6281 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6282 ret = btrfs_update_inode(trans, root, parent_inode);
6284 btrfs_abort_transaction(trans, root, ret);
6288 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6291 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6292 key.objectid, root->root_key.objectid,
6293 parent_ino, &local_index, name, name_len);
6295 } else if (add_backref) {
6299 err = btrfs_del_inode_ref(trans, root, name, name_len,
6300 ino, parent_ino, &local_index);
6305 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6306 struct inode *dir, struct dentry *dentry,
6307 struct inode *inode, int backref, u64 index)
6309 int err = btrfs_add_link(trans, dir, inode,
6310 dentry->d_name.name, dentry->d_name.len,
6317 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6318 umode_t mode, dev_t rdev)
6320 struct btrfs_trans_handle *trans;
6321 struct btrfs_root *root = BTRFS_I(dir)->root;
6322 struct inode *inode = NULL;
6329 * 2 for inode item and ref
6331 * 1 for xattr if selinux is on
6333 trans = btrfs_start_transaction(root, 5);
6335 return PTR_ERR(trans);
6337 err = btrfs_find_free_ino(root, &objectid);
6341 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6342 dentry->d_name.len, btrfs_ino(dir), objectid,
6344 if (IS_ERR(inode)) {
6345 err = PTR_ERR(inode);
6350 * If the active LSM wants to access the inode during
6351 * d_instantiate it needs these. Smack checks to see
6352 * if the filesystem supports xattrs by looking at the
6355 inode->i_op = &btrfs_special_inode_operations;
6356 init_special_inode(inode, inode->i_mode, rdev);
6358 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6360 goto out_unlock_inode;
6362 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6364 goto out_unlock_inode;
6366 btrfs_update_inode(trans, root, inode);
6367 unlock_new_inode(inode);
6368 d_instantiate(dentry, inode);
6372 btrfs_end_transaction(trans, root);
6373 btrfs_balance_delayed_items(root);
6374 btrfs_btree_balance_dirty(root);
6376 inode_dec_link_count(inode);
6383 unlock_new_inode(inode);
6388 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6389 umode_t mode, bool excl)
6391 struct btrfs_trans_handle *trans;
6392 struct btrfs_root *root = BTRFS_I(dir)->root;
6393 struct inode *inode = NULL;
6394 int drop_inode_on_err = 0;
6400 * 2 for inode item and ref
6402 * 1 for xattr if selinux is on
6404 trans = btrfs_start_transaction(root, 5);
6406 return PTR_ERR(trans);
6408 err = btrfs_find_free_ino(root, &objectid);
6412 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6413 dentry->d_name.len, btrfs_ino(dir), objectid,
6415 if (IS_ERR(inode)) {
6416 err = PTR_ERR(inode);
6419 drop_inode_on_err = 1;
6421 * If the active LSM wants to access the inode during
6422 * d_instantiate it needs these. Smack checks to see
6423 * if the filesystem supports xattrs by looking at the
6426 inode->i_fop = &btrfs_file_operations;
6427 inode->i_op = &btrfs_file_inode_operations;
6428 inode->i_mapping->a_ops = &btrfs_aops;
6430 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6432 goto out_unlock_inode;
6434 err = btrfs_update_inode(trans, root, inode);
6436 goto out_unlock_inode;
6438 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6440 goto out_unlock_inode;
6442 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6443 unlock_new_inode(inode);
6444 d_instantiate(dentry, inode);
6447 btrfs_end_transaction(trans, root);
6448 if (err && drop_inode_on_err) {
6449 inode_dec_link_count(inode);
6452 btrfs_balance_delayed_items(root);
6453 btrfs_btree_balance_dirty(root);
6457 unlock_new_inode(inode);
6462 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6463 struct dentry *dentry)
6465 struct btrfs_trans_handle *trans = NULL;
6466 struct btrfs_root *root = BTRFS_I(dir)->root;
6467 struct inode *inode = d_inode(old_dentry);
6472 /* do not allow sys_link's with other subvols of the same device */
6473 if (root->objectid != BTRFS_I(inode)->root->objectid)
6476 if (inode->i_nlink >= BTRFS_LINK_MAX)
6479 err = btrfs_set_inode_index(dir, &index);
6484 * 2 items for inode and inode ref
6485 * 2 items for dir items
6486 * 1 item for parent inode
6488 trans = btrfs_start_transaction(root, 5);
6489 if (IS_ERR(trans)) {
6490 err = PTR_ERR(trans);
6495 /* There are several dir indexes for this inode, clear the cache. */
6496 BTRFS_I(inode)->dir_index = 0ULL;
6498 inode_inc_iversion(inode);
6499 inode->i_ctime = CURRENT_TIME;
6501 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6503 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6508 struct dentry *parent = dentry->d_parent;
6509 err = btrfs_update_inode(trans, root, inode);
6512 if (inode->i_nlink == 1) {
6514 * If new hard link count is 1, it's a file created
6515 * with open(2) O_TMPFILE flag.
6517 err = btrfs_orphan_del(trans, inode);
6521 d_instantiate(dentry, inode);
6522 btrfs_log_new_name(trans, inode, NULL, parent);
6525 btrfs_balance_delayed_items(root);
6528 btrfs_end_transaction(trans, root);
6530 inode_dec_link_count(inode);
6533 btrfs_btree_balance_dirty(root);
6537 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6539 struct inode *inode = NULL;
6540 struct btrfs_trans_handle *trans;
6541 struct btrfs_root *root = BTRFS_I(dir)->root;
6543 int drop_on_err = 0;
6548 * 2 items for inode and ref
6549 * 2 items for dir items
6550 * 1 for xattr if selinux is on
6552 trans = btrfs_start_transaction(root, 5);
6554 return PTR_ERR(trans);
6556 err = btrfs_find_free_ino(root, &objectid);
6560 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6561 dentry->d_name.len, btrfs_ino(dir), objectid,
6562 S_IFDIR | mode, &index);
6563 if (IS_ERR(inode)) {
6564 err = PTR_ERR(inode);
6569 /* these must be set before we unlock the inode */
6570 inode->i_op = &btrfs_dir_inode_operations;
6571 inode->i_fop = &btrfs_dir_file_operations;
6573 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6575 goto out_fail_inode;
6577 btrfs_i_size_write(inode, 0);
6578 err = btrfs_update_inode(trans, root, inode);
6580 goto out_fail_inode;
6582 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6583 dentry->d_name.len, 0, index);
6585 goto out_fail_inode;
6587 d_instantiate(dentry, inode);
6589 * mkdir is special. We're unlocking after we call d_instantiate
6590 * to avoid a race with nfsd calling d_instantiate.
6592 unlock_new_inode(inode);
6596 btrfs_end_transaction(trans, root);
6598 inode_dec_link_count(inode);
6601 btrfs_balance_delayed_items(root);
6602 btrfs_btree_balance_dirty(root);
6606 unlock_new_inode(inode);
6610 /* Find next extent map of a given extent map, caller needs to ensure locks */
6611 static struct extent_map *next_extent_map(struct extent_map *em)
6613 struct rb_node *next;
6615 next = rb_next(&em->rb_node);
6618 return container_of(next, struct extent_map, rb_node);
6621 static struct extent_map *prev_extent_map(struct extent_map *em)
6623 struct rb_node *prev;
6625 prev = rb_prev(&em->rb_node);
6628 return container_of(prev, struct extent_map, rb_node);
6631 /* helper for btfs_get_extent. Given an existing extent in the tree,
6632 * the existing extent is the nearest extent to map_start,
6633 * and an extent that you want to insert, deal with overlap and insert
6634 * the best fitted new extent into the tree.
6636 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6637 struct extent_map *existing,
6638 struct extent_map *em,
6641 struct extent_map *prev;
6642 struct extent_map *next;
6647 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6649 if (existing->start > map_start) {
6651 prev = prev_extent_map(next);
6654 next = next_extent_map(prev);
6657 start = prev ? extent_map_end(prev) : em->start;
6658 start = max_t(u64, start, em->start);
6659 end = next ? next->start : extent_map_end(em);
6660 end = min_t(u64, end, extent_map_end(em));
6661 start_diff = start - em->start;
6663 em->len = end - start;
6664 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6665 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6666 em->block_start += start_diff;
6667 em->block_len -= start_diff;
6669 return add_extent_mapping(em_tree, em, 0);
6672 static noinline int uncompress_inline(struct btrfs_path *path,
6674 size_t pg_offset, u64 extent_offset,
6675 struct btrfs_file_extent_item *item)
6678 struct extent_buffer *leaf = path->nodes[0];
6681 unsigned long inline_size;
6685 WARN_ON(pg_offset != 0);
6686 compress_type = btrfs_file_extent_compression(leaf, item);
6687 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6688 inline_size = btrfs_file_extent_inline_item_len(leaf,
6689 btrfs_item_nr(path->slots[0]));
6690 tmp = kmalloc(inline_size, GFP_NOFS);
6693 ptr = btrfs_file_extent_inline_start(item);
6695 read_extent_buffer(leaf, tmp, ptr, inline_size);
6697 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6698 ret = btrfs_decompress(compress_type, tmp, page,
6699 extent_offset, inline_size, max_size);
6705 * a bit scary, this does extent mapping from logical file offset to the disk.
6706 * the ugly parts come from merging extents from the disk with the in-ram
6707 * representation. This gets more complex because of the data=ordered code,
6708 * where the in-ram extents might be locked pending data=ordered completion.
6710 * This also copies inline extents directly into the page.
6713 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6714 size_t pg_offset, u64 start, u64 len,
6719 u64 extent_start = 0;
6721 u64 objectid = btrfs_ino(inode);
6723 struct btrfs_path *path = NULL;
6724 struct btrfs_root *root = BTRFS_I(inode)->root;
6725 struct btrfs_file_extent_item *item;
6726 struct extent_buffer *leaf;
6727 struct btrfs_key found_key;
6728 struct extent_map *em = NULL;
6729 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6730 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6731 struct btrfs_trans_handle *trans = NULL;
6732 const bool new_inline = !page || create;
6735 read_lock(&em_tree->lock);
6736 em = lookup_extent_mapping(em_tree, start, len);
6738 em->bdev = root->fs_info->fs_devices->latest_bdev;
6739 read_unlock(&em_tree->lock);
6742 if (em->start > start || em->start + em->len <= start)
6743 free_extent_map(em);
6744 else if (em->block_start == EXTENT_MAP_INLINE && page)
6745 free_extent_map(em);
6749 em = alloc_extent_map();
6754 em->bdev = root->fs_info->fs_devices->latest_bdev;
6755 em->start = EXTENT_MAP_HOLE;
6756 em->orig_start = EXTENT_MAP_HOLE;
6758 em->block_len = (u64)-1;
6761 path = btrfs_alloc_path();
6767 * Chances are we'll be called again, so go ahead and do
6770 path->reada = READA_FORWARD;
6773 ret = btrfs_lookup_file_extent(trans, root, path,
6774 objectid, start, trans != NULL);
6781 if (path->slots[0] == 0)
6786 leaf = path->nodes[0];
6787 item = btrfs_item_ptr(leaf, path->slots[0],
6788 struct btrfs_file_extent_item);
6789 /* are we inside the extent that was found? */
6790 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6791 found_type = found_key.type;
6792 if (found_key.objectid != objectid ||
6793 found_type != BTRFS_EXTENT_DATA_KEY) {
6795 * If we backup past the first extent we want to move forward
6796 * and see if there is an extent in front of us, otherwise we'll
6797 * say there is a hole for our whole search range which can
6804 found_type = btrfs_file_extent_type(leaf, item);
6805 extent_start = found_key.offset;
6806 if (found_type == BTRFS_FILE_EXTENT_REG ||
6807 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6808 extent_end = extent_start +
6809 btrfs_file_extent_num_bytes(leaf, item);
6810 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6812 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6813 extent_end = ALIGN(extent_start + size, root->sectorsize);
6816 if (start >= extent_end) {
6818 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6819 ret = btrfs_next_leaf(root, path);
6826 leaf = path->nodes[0];
6828 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6829 if (found_key.objectid != objectid ||
6830 found_key.type != BTRFS_EXTENT_DATA_KEY)
6832 if (start + len <= found_key.offset)
6834 if (start > found_key.offset)
6837 em->orig_start = start;
6838 em->len = found_key.offset - start;
6842 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6844 if (found_type == BTRFS_FILE_EXTENT_REG ||
6845 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6847 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6851 size_t extent_offset;
6857 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6858 extent_offset = page_offset(page) + pg_offset - extent_start;
6859 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6860 size - extent_offset);
6861 em->start = extent_start + extent_offset;
6862 em->len = ALIGN(copy_size, root->sectorsize);
6863 em->orig_block_len = em->len;
6864 em->orig_start = em->start;
6865 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6866 if (create == 0 && !PageUptodate(page)) {
6867 if (btrfs_file_extent_compression(leaf, item) !=
6868 BTRFS_COMPRESS_NONE) {
6869 ret = uncompress_inline(path, page, pg_offset,
6870 extent_offset, item);
6877 read_extent_buffer(leaf, map + pg_offset, ptr,
6879 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6880 memset(map + pg_offset + copy_size, 0,
6881 PAGE_CACHE_SIZE - pg_offset -
6886 flush_dcache_page(page);
6887 } else if (create && PageUptodate(page)) {
6891 free_extent_map(em);
6894 btrfs_release_path(path);
6895 trans = btrfs_join_transaction(root);
6898 return ERR_CAST(trans);
6902 write_extent_buffer(leaf, map + pg_offset, ptr,
6905 btrfs_mark_buffer_dirty(leaf);
6907 set_extent_uptodate(io_tree, em->start,
6908 extent_map_end(em) - 1, NULL, GFP_NOFS);
6913 em->orig_start = start;
6916 em->block_start = EXTENT_MAP_HOLE;
6917 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6919 btrfs_release_path(path);
6920 if (em->start > start || extent_map_end(em) <= start) {
6921 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6922 em->start, em->len, start, len);
6928 write_lock(&em_tree->lock);
6929 ret = add_extent_mapping(em_tree, em, 0);
6930 /* it is possible that someone inserted the extent into the tree
6931 * while we had the lock dropped. It is also possible that
6932 * an overlapping map exists in the tree
6934 if (ret == -EEXIST) {
6935 struct extent_map *existing;
6939 existing = search_extent_mapping(em_tree, start, len);
6941 * existing will always be non-NULL, since there must be
6942 * extent causing the -EEXIST.
6944 if (start >= extent_map_end(existing) ||
6945 start <= existing->start) {
6947 * The existing extent map is the one nearest to
6948 * the [start, start + len) range which overlaps
6950 err = merge_extent_mapping(em_tree, existing,
6952 free_extent_map(existing);
6954 free_extent_map(em);
6958 free_extent_map(em);
6963 write_unlock(&em_tree->lock);
6966 trace_btrfs_get_extent(root, em);
6968 btrfs_free_path(path);
6970 ret = btrfs_end_transaction(trans, root);
6975 free_extent_map(em);
6976 return ERR_PTR(err);
6978 BUG_ON(!em); /* Error is always set */
6982 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6983 size_t pg_offset, u64 start, u64 len,
6986 struct extent_map *em;
6987 struct extent_map *hole_em = NULL;
6988 u64 range_start = start;
6994 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7001 * - a pre-alloc extent,
7002 * there might actually be delalloc bytes behind it.
7004 if (em->block_start != EXTENT_MAP_HOLE &&
7005 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7011 /* check to see if we've wrapped (len == -1 or similar) */
7020 /* ok, we didn't find anything, lets look for delalloc */
7021 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7022 end, len, EXTENT_DELALLOC, 1);
7023 found_end = range_start + found;
7024 if (found_end < range_start)
7025 found_end = (u64)-1;
7028 * we didn't find anything useful, return
7029 * the original results from get_extent()
7031 if (range_start > end || found_end <= start) {
7037 /* adjust the range_start to make sure it doesn't
7038 * go backwards from the start they passed in
7040 range_start = max(start, range_start);
7041 found = found_end - range_start;
7044 u64 hole_start = start;
7047 em = alloc_extent_map();
7053 * when btrfs_get_extent can't find anything it
7054 * returns one huge hole
7056 * make sure what it found really fits our range, and
7057 * adjust to make sure it is based on the start from
7061 u64 calc_end = extent_map_end(hole_em);
7063 if (calc_end <= start || (hole_em->start > end)) {
7064 free_extent_map(hole_em);
7067 hole_start = max(hole_em->start, start);
7068 hole_len = calc_end - hole_start;
7072 if (hole_em && range_start > hole_start) {
7073 /* our hole starts before our delalloc, so we
7074 * have to return just the parts of the hole
7075 * that go until the delalloc starts
7077 em->len = min(hole_len,
7078 range_start - hole_start);
7079 em->start = hole_start;
7080 em->orig_start = hole_start;
7082 * don't adjust block start at all,
7083 * it is fixed at EXTENT_MAP_HOLE
7085 em->block_start = hole_em->block_start;
7086 em->block_len = hole_len;
7087 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7088 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7090 em->start = range_start;
7092 em->orig_start = range_start;
7093 em->block_start = EXTENT_MAP_DELALLOC;
7094 em->block_len = found;
7096 } else if (hole_em) {
7101 free_extent_map(hole_em);
7103 free_extent_map(em);
7104 return ERR_PTR(err);
7109 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7112 struct btrfs_root *root = BTRFS_I(inode)->root;
7113 struct extent_map *em;
7114 struct btrfs_key ins;
7118 alloc_hint = get_extent_allocation_hint(inode, start, len);
7119 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7120 alloc_hint, &ins, 1, 1);
7122 return ERR_PTR(ret);
7125 * Create the ordered extent before the extent map. This is to avoid
7126 * races with the fast fsync path that would lead to it logging file
7127 * extent items that point to disk extents that were not yet written to.
7128 * The fast fsync path collects ordered extents into a local list and
7129 * then collects all the new extent maps, so we must create the ordered
7130 * extent first and make sure the fast fsync path collects any new
7131 * ordered extents after collecting new extent maps as well.
7132 * The fsync path simply can not rely on inode_dio_wait() because it
7133 * causes deadlock with AIO.
7135 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7136 ins.offset, ins.offset, 0);
7138 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7139 return ERR_PTR(ret);
7142 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7143 ins.offset, ins.offset, ins.offset, 0);
7145 struct btrfs_ordered_extent *oe;
7147 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7148 oe = btrfs_lookup_ordered_extent(inode, start);
7152 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7153 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7154 btrfs_remove_ordered_extent(inode, oe);
7155 /* Once for our lookup and once for the ordered extents tree. */
7156 btrfs_put_ordered_extent(oe);
7157 btrfs_put_ordered_extent(oe);
7163 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7164 * block must be cow'd
7166 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7167 u64 *orig_start, u64 *orig_block_len,
7170 struct btrfs_trans_handle *trans;
7171 struct btrfs_path *path;
7173 struct extent_buffer *leaf;
7174 struct btrfs_root *root = BTRFS_I(inode)->root;
7175 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7176 struct btrfs_file_extent_item *fi;
7177 struct btrfs_key key;
7184 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7186 path = btrfs_alloc_path();
7190 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7195 slot = path->slots[0];
7198 /* can't find the item, must cow */
7205 leaf = path->nodes[0];
7206 btrfs_item_key_to_cpu(leaf, &key, slot);
7207 if (key.objectid != btrfs_ino(inode) ||
7208 key.type != BTRFS_EXTENT_DATA_KEY) {
7209 /* not our file or wrong item type, must cow */
7213 if (key.offset > offset) {
7214 /* Wrong offset, must cow */
7218 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7219 found_type = btrfs_file_extent_type(leaf, fi);
7220 if (found_type != BTRFS_FILE_EXTENT_REG &&
7221 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7222 /* not a regular extent, must cow */
7226 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7229 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7230 if (extent_end <= offset)
7233 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7234 if (disk_bytenr == 0)
7237 if (btrfs_file_extent_compression(leaf, fi) ||
7238 btrfs_file_extent_encryption(leaf, fi) ||
7239 btrfs_file_extent_other_encoding(leaf, fi))
7242 backref_offset = btrfs_file_extent_offset(leaf, fi);
7245 *orig_start = key.offset - backref_offset;
7246 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7247 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7250 if (btrfs_extent_readonly(root, disk_bytenr))
7253 num_bytes = min(offset + *len, extent_end) - offset;
7254 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7257 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7258 ret = test_range_bit(io_tree, offset, range_end,
7259 EXTENT_DELALLOC, 0, NULL);
7266 btrfs_release_path(path);
7269 * look for other files referencing this extent, if we
7270 * find any we must cow
7272 trans = btrfs_join_transaction(root);
7273 if (IS_ERR(trans)) {
7278 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7279 key.offset - backref_offset, disk_bytenr);
7280 btrfs_end_transaction(trans, root);
7287 * adjust disk_bytenr and num_bytes to cover just the bytes
7288 * in this extent we are about to write. If there
7289 * are any csums in that range we have to cow in order
7290 * to keep the csums correct
7292 disk_bytenr += backref_offset;
7293 disk_bytenr += offset - key.offset;
7294 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7297 * all of the above have passed, it is safe to overwrite this extent
7303 btrfs_free_path(path);
7307 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7309 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7311 void **pagep = NULL;
7312 struct page *page = NULL;
7316 start_idx = start >> PAGE_CACHE_SHIFT;
7319 * end is the last byte in the last page. end == start is legal
7321 end_idx = end >> PAGE_CACHE_SHIFT;
7325 /* Most of the code in this while loop is lifted from
7326 * find_get_page. It's been modified to begin searching from a
7327 * page and return just the first page found in that range. If the
7328 * found idx is less than or equal to the end idx then we know that
7329 * a page exists. If no pages are found or if those pages are
7330 * outside of the range then we're fine (yay!) */
7331 while (page == NULL &&
7332 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7333 page = radix_tree_deref_slot(pagep);
7334 if (unlikely(!page))
7337 if (radix_tree_exception(page)) {
7338 if (radix_tree_deref_retry(page)) {
7343 * Otherwise, shmem/tmpfs must be storing a swap entry
7344 * here as an exceptional entry: so return it without
7345 * attempting to raise page count.
7348 break; /* TODO: Is this relevant for this use case? */
7351 if (!page_cache_get_speculative(page)) {
7357 * Has the page moved?
7358 * This is part of the lockless pagecache protocol. See
7359 * include/linux/pagemap.h for details.
7361 if (unlikely(page != *pagep)) {
7362 page_cache_release(page);
7368 if (page->index <= end_idx)
7370 page_cache_release(page);
7377 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7378 struct extent_state **cached_state, int writing)
7380 struct btrfs_ordered_extent *ordered;
7384 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7387 * We're concerned with the entire range that we're going to be
7388 * doing DIO to, so we need to make sure theres no ordered
7389 * extents in this range.
7391 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7392 lockend - lockstart + 1);
7395 * We need to make sure there are no buffered pages in this
7396 * range either, we could have raced between the invalidate in
7397 * generic_file_direct_write and locking the extent. The
7398 * invalidate needs to happen so that reads after a write do not
7403 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7406 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7407 cached_state, GFP_NOFS);
7410 btrfs_start_ordered_extent(inode, ordered, 1);
7411 btrfs_put_ordered_extent(ordered);
7414 * We could trigger writeback for this range (and wait
7415 * for it to complete) and then invalidate the pages for
7416 * this range (through invalidate_inode_pages2_range()),
7417 * but that can lead us to a deadlock with a concurrent
7418 * call to readpages() (a buffered read or a defrag call
7419 * triggered a readahead) on a page lock due to an
7420 * ordered dio extent we created before but did not have
7421 * yet a corresponding bio submitted (whence it can not
7422 * complete), which makes readpages() wait for that
7423 * ordered extent to complete while holding a lock on
7436 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7437 u64 len, u64 orig_start,
7438 u64 block_start, u64 block_len,
7439 u64 orig_block_len, u64 ram_bytes,
7442 struct extent_map_tree *em_tree;
7443 struct extent_map *em;
7444 struct btrfs_root *root = BTRFS_I(inode)->root;
7447 em_tree = &BTRFS_I(inode)->extent_tree;
7448 em = alloc_extent_map();
7450 return ERR_PTR(-ENOMEM);
7453 em->orig_start = orig_start;
7454 em->mod_start = start;
7457 em->block_len = block_len;
7458 em->block_start = block_start;
7459 em->bdev = root->fs_info->fs_devices->latest_bdev;
7460 em->orig_block_len = orig_block_len;
7461 em->ram_bytes = ram_bytes;
7462 em->generation = -1;
7463 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7464 if (type == BTRFS_ORDERED_PREALLOC)
7465 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7468 btrfs_drop_extent_cache(inode, em->start,
7469 em->start + em->len - 1, 0);
7470 write_lock(&em_tree->lock);
7471 ret = add_extent_mapping(em_tree, em, 1);
7472 write_unlock(&em_tree->lock);
7473 } while (ret == -EEXIST);
7476 free_extent_map(em);
7477 return ERR_PTR(ret);
7483 static void adjust_dio_outstanding_extents(struct inode *inode,
7484 struct btrfs_dio_data *dio_data,
7487 unsigned num_extents;
7489 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7490 BTRFS_MAX_EXTENT_SIZE);
7492 * If we have an outstanding_extents count still set then we're
7493 * within our reservation, otherwise we need to adjust our inode
7494 * counter appropriately.
7496 if (dio_data->outstanding_extents) {
7497 dio_data->outstanding_extents -= num_extents;
7499 spin_lock(&BTRFS_I(inode)->lock);
7500 BTRFS_I(inode)->outstanding_extents += num_extents;
7501 spin_unlock(&BTRFS_I(inode)->lock);
7505 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7506 struct buffer_head *bh_result, int create)
7508 struct extent_map *em;
7509 struct btrfs_root *root = BTRFS_I(inode)->root;
7510 struct extent_state *cached_state = NULL;
7511 struct btrfs_dio_data *dio_data = NULL;
7512 u64 start = iblock << inode->i_blkbits;
7513 u64 lockstart, lockend;
7514 u64 len = bh_result->b_size;
7515 int unlock_bits = EXTENT_LOCKED;
7519 unlock_bits |= EXTENT_DIRTY;
7521 len = min_t(u64, len, root->sectorsize);
7524 lockend = start + len - 1;
7526 if (current->journal_info) {
7528 * Need to pull our outstanding extents and set journal_info to NULL so
7529 * that anything that needs to check if there's a transction doesn't get
7532 dio_data = current->journal_info;
7533 current->journal_info = NULL;
7537 * If this errors out it's because we couldn't invalidate pagecache for
7538 * this range and we need to fallback to buffered.
7540 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7546 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7553 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7554 * io. INLINE is special, and we could probably kludge it in here, but
7555 * it's still buffered so for safety lets just fall back to the generic
7558 * For COMPRESSED we _have_ to read the entire extent in so we can
7559 * decompress it, so there will be buffering required no matter what we
7560 * do, so go ahead and fallback to buffered.
7562 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7563 * to buffered IO. Don't blame me, this is the price we pay for using
7566 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7567 em->block_start == EXTENT_MAP_INLINE) {
7568 free_extent_map(em);
7573 /* Just a good old fashioned hole, return */
7574 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7575 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7576 free_extent_map(em);
7581 * We don't allocate a new extent in the following cases
7583 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7585 * 2) The extent is marked as PREALLOC. We're good to go here and can
7586 * just use the extent.
7590 len = min(len, em->len - (start - em->start));
7591 lockstart = start + len;
7595 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7596 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7597 em->block_start != EXTENT_MAP_HOLE)) {
7599 u64 block_start, orig_start, orig_block_len, ram_bytes;
7601 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7602 type = BTRFS_ORDERED_PREALLOC;
7604 type = BTRFS_ORDERED_NOCOW;
7605 len = min(len, em->len - (start - em->start));
7606 block_start = em->block_start + (start - em->start);
7608 if (can_nocow_extent(inode, start, &len, &orig_start,
7609 &orig_block_len, &ram_bytes) == 1) {
7610 if (type == BTRFS_ORDERED_PREALLOC) {
7611 free_extent_map(em);
7612 em = create_pinned_em(inode, start, len,
7623 ret = btrfs_add_ordered_extent_dio(inode, start,
7624 block_start, len, len, type);
7626 free_extent_map(em);
7634 * this will cow the extent, reset the len in case we changed
7637 len = bh_result->b_size;
7638 free_extent_map(em);
7639 em = btrfs_new_extent_direct(inode, start, len);
7644 len = min(len, em->len - (start - em->start));
7646 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7648 bh_result->b_size = len;
7649 bh_result->b_bdev = em->bdev;
7650 set_buffer_mapped(bh_result);
7652 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7653 set_buffer_new(bh_result);
7656 * Need to update the i_size under the extent lock so buffered
7657 * readers will get the updated i_size when we unlock.
7659 if (start + len > i_size_read(inode))
7660 i_size_write(inode, start + len);
7662 adjust_dio_outstanding_extents(inode, dio_data, len);
7663 btrfs_free_reserved_data_space(inode, start, len);
7664 WARN_ON(dio_data->reserve < len);
7665 dio_data->reserve -= len;
7666 dio_data->unsubmitted_oe_range_end = start + len;
7667 current->journal_info = dio_data;
7671 * In the case of write we need to clear and unlock the entire range,
7672 * in the case of read we need to unlock only the end area that we
7673 * aren't using if there is any left over space.
7675 if (lockstart < lockend) {
7676 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7677 lockend, unlock_bits, 1, 0,
7678 &cached_state, GFP_NOFS);
7680 free_extent_state(cached_state);
7683 free_extent_map(em);
7688 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7689 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7692 current->journal_info = dio_data;
7694 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7695 * write less data then expected, so that we don't underflow our inode's
7696 * outstanding extents counter.
7698 if (create && dio_data)
7699 adjust_dio_outstanding_extents(inode, dio_data, len);
7704 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7705 int rw, int mirror_num)
7707 struct btrfs_root *root = BTRFS_I(inode)->root;
7710 BUG_ON(rw & REQ_WRITE);
7714 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7715 BTRFS_WQ_ENDIO_DIO_REPAIR);
7719 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7725 static int btrfs_check_dio_repairable(struct inode *inode,
7726 struct bio *failed_bio,
7727 struct io_failure_record *failrec,
7732 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7733 failrec->logical, failrec->len);
7734 if (num_copies == 1) {
7736 * we only have a single copy of the data, so don't bother with
7737 * all the retry and error correction code that follows. no
7738 * matter what the error is, it is very likely to persist.
7740 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7741 num_copies, failrec->this_mirror, failed_mirror);
7745 failrec->failed_mirror = failed_mirror;
7746 failrec->this_mirror++;
7747 if (failrec->this_mirror == failed_mirror)
7748 failrec->this_mirror++;
7750 if (failrec->this_mirror > num_copies) {
7751 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7752 num_copies, failrec->this_mirror, failed_mirror);
7759 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7760 struct page *page, unsigned int pgoff,
7761 u64 start, u64 end, int failed_mirror,
7762 bio_end_io_t *repair_endio, void *repair_arg)
7764 struct io_failure_record *failrec;
7770 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7772 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7776 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7779 free_io_failure(inode, failrec);
7783 if ((failed_bio->bi_vcnt > 1)
7784 || (failed_bio->bi_io_vec->bv_len
7785 > BTRFS_I(inode)->root->sectorsize))
7786 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7788 read_mode = READ_SYNC;
7790 isector = start - btrfs_io_bio(failed_bio)->logical;
7791 isector >>= inode->i_sb->s_blocksize_bits;
7792 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7793 pgoff, isector, repair_endio, repair_arg);
7795 free_io_failure(inode, failrec);
7799 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7800 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7801 read_mode, failrec->this_mirror, failrec->in_validation);
7803 ret = submit_dio_repair_bio(inode, bio, read_mode,
7804 failrec->this_mirror);
7806 free_io_failure(inode, failrec);
7813 struct btrfs_retry_complete {
7814 struct completion done;
7815 struct inode *inode;
7820 static void btrfs_retry_endio_nocsum(struct bio *bio)
7822 struct btrfs_retry_complete *done = bio->bi_private;
7823 struct inode *inode;
7824 struct bio_vec *bvec;
7830 ASSERT(bio->bi_vcnt == 1);
7831 inode = bio->bi_io_vec->bv_page->mapping->host;
7832 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7835 bio_for_each_segment_all(bvec, bio, i)
7836 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7838 complete(&done->done);
7842 static int __btrfs_correct_data_nocsum(struct inode *inode,
7843 struct btrfs_io_bio *io_bio)
7845 struct btrfs_fs_info *fs_info;
7846 struct bio_vec *bvec;
7847 struct btrfs_retry_complete done;
7855 fs_info = BTRFS_I(inode)->root->fs_info;
7856 sectorsize = BTRFS_I(inode)->root->sectorsize;
7858 start = io_bio->logical;
7861 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7862 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7863 pgoff = bvec->bv_offset;
7865 next_block_or_try_again:
7868 init_completion(&done.done);
7870 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7871 pgoff, start, start + sectorsize - 1,
7873 btrfs_retry_endio_nocsum, &done);
7877 wait_for_completion(&done.done);
7879 if (!done.uptodate) {
7880 /* We might have another mirror, so try again */
7881 goto next_block_or_try_again;
7884 start += sectorsize;
7887 pgoff += sectorsize;
7888 goto next_block_or_try_again;
7895 static void btrfs_retry_endio(struct bio *bio)
7897 struct btrfs_retry_complete *done = bio->bi_private;
7898 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7899 struct inode *inode;
7900 struct bio_vec *bvec;
7911 start = done->start;
7913 ASSERT(bio->bi_vcnt == 1);
7914 inode = bio->bi_io_vec->bv_page->mapping->host;
7915 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7917 bio_for_each_segment_all(bvec, bio, i) {
7918 ret = __readpage_endio_check(done->inode, io_bio, i,
7919 bvec->bv_page, bvec->bv_offset,
7920 done->start, bvec->bv_len);
7922 clean_io_failure(done->inode, done->start,
7923 bvec->bv_page, bvec->bv_offset);
7928 done->uptodate = uptodate;
7930 complete(&done->done);
7934 static int __btrfs_subio_endio_read(struct inode *inode,
7935 struct btrfs_io_bio *io_bio, int err)
7937 struct btrfs_fs_info *fs_info;
7938 struct bio_vec *bvec;
7939 struct btrfs_retry_complete done;
7949 fs_info = BTRFS_I(inode)->root->fs_info;
7950 sectorsize = BTRFS_I(inode)->root->sectorsize;
7953 start = io_bio->logical;
7956 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7957 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7959 pgoff = bvec->bv_offset;
7961 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7962 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7963 bvec->bv_page, pgoff, start,
7970 init_completion(&done.done);
7972 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7973 pgoff, start, start + sectorsize - 1,
7975 btrfs_retry_endio, &done);
7981 wait_for_completion(&done.done);
7983 if (!done.uptodate) {
7984 /* We might have another mirror, so try again */
7988 offset += sectorsize;
7989 start += sectorsize;
7994 pgoff += sectorsize;
8002 static int btrfs_subio_endio_read(struct inode *inode,
8003 struct btrfs_io_bio *io_bio, int err)
8005 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8009 return __btrfs_correct_data_nocsum(inode, io_bio);
8013 return __btrfs_subio_endio_read(inode, io_bio, err);
8017 static void btrfs_endio_direct_read(struct bio *bio)
8019 struct btrfs_dio_private *dip = bio->bi_private;
8020 struct inode *inode = dip->inode;
8021 struct bio *dio_bio;
8022 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8023 int err = bio->bi_error;
8025 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8026 err = btrfs_subio_endio_read(inode, io_bio, err);
8028 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8029 dip->logical_offset + dip->bytes - 1);
8030 dio_bio = dip->dio_bio;
8034 dio_end_io(dio_bio, bio->bi_error);
8037 io_bio->end_io(io_bio, err);
8041 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8046 struct btrfs_root *root = BTRFS_I(inode)->root;
8047 struct btrfs_ordered_extent *ordered = NULL;
8048 u64 ordered_offset = offset;
8049 u64 ordered_bytes = bytes;
8053 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8060 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8061 finish_ordered_fn, NULL, NULL);
8062 btrfs_queue_work(root->fs_info->endio_write_workers,
8066 * our bio might span multiple ordered extents. If we haven't
8067 * completed the accounting for the whole dio, go back and try again
8069 if (ordered_offset < offset + bytes) {
8070 ordered_bytes = offset + bytes - ordered_offset;
8076 static void btrfs_endio_direct_write(struct bio *bio)
8078 struct btrfs_dio_private *dip = bio->bi_private;
8079 struct bio *dio_bio = dip->dio_bio;
8081 btrfs_endio_direct_write_update_ordered(dip->inode,
8082 dip->logical_offset,
8088 dio_end_io(dio_bio, bio->bi_error);
8092 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8093 struct bio *bio, int mirror_num,
8094 unsigned long bio_flags, u64 offset)
8097 struct btrfs_root *root = BTRFS_I(inode)->root;
8098 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8099 BUG_ON(ret); /* -ENOMEM */
8103 static void btrfs_end_dio_bio(struct bio *bio)
8105 struct btrfs_dio_private *dip = bio->bi_private;
8106 int err = bio->bi_error;
8109 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8110 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8111 btrfs_ino(dip->inode), bio->bi_rw,
8112 (unsigned long long)bio->bi_iter.bi_sector,
8113 bio->bi_iter.bi_size, err);
8115 if (dip->subio_endio)
8116 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8122 * before atomic variable goto zero, we must make sure
8123 * dip->errors is perceived to be set.
8125 smp_mb__before_atomic();
8128 /* if there are more bios still pending for this dio, just exit */
8129 if (!atomic_dec_and_test(&dip->pending_bios))
8133 bio_io_error(dip->orig_bio);
8135 dip->dio_bio->bi_error = 0;
8136 bio_endio(dip->orig_bio);
8142 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8143 u64 first_sector, gfp_t gfp_flags)
8146 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8148 bio_associate_current(bio);
8152 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8153 struct inode *inode,
8154 struct btrfs_dio_private *dip,
8158 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8159 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8163 * We load all the csum data we need when we submit
8164 * the first bio to reduce the csum tree search and
8167 if (dip->logical_offset == file_offset) {
8168 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8174 if (bio == dip->orig_bio)
8177 file_offset -= dip->logical_offset;
8178 file_offset >>= inode->i_sb->s_blocksize_bits;
8179 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8184 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8185 int rw, u64 file_offset, int skip_sum,
8188 struct btrfs_dio_private *dip = bio->bi_private;
8189 int write = rw & REQ_WRITE;
8190 struct btrfs_root *root = BTRFS_I(inode)->root;
8194 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8199 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8200 BTRFS_WQ_ENDIO_DATA);
8208 if (write && async_submit) {
8209 ret = btrfs_wq_submit_bio(root->fs_info,
8210 inode, rw, bio, 0, 0,
8212 __btrfs_submit_bio_start_direct_io,
8213 __btrfs_submit_bio_done);
8217 * If we aren't doing async submit, calculate the csum of the
8220 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8224 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8230 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8236 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8239 struct inode *inode = dip->inode;
8240 struct btrfs_root *root = BTRFS_I(inode)->root;
8242 struct bio *orig_bio = dip->orig_bio;
8243 struct bio_vec *bvec = orig_bio->bi_io_vec;
8244 u64 start_sector = orig_bio->bi_iter.bi_sector;
8245 u64 file_offset = dip->logical_offset;
8248 u32 blocksize = root->sectorsize;
8249 int async_submit = 0;
8254 map_length = orig_bio->bi_iter.bi_size;
8255 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8256 &map_length, NULL, 0);
8260 if (map_length >= orig_bio->bi_iter.bi_size) {
8262 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8266 /* async crcs make it difficult to collect full stripe writes. */
8267 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8272 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8276 bio->bi_private = dip;
8277 bio->bi_end_io = btrfs_end_dio_bio;
8278 btrfs_io_bio(bio)->logical = file_offset;
8279 atomic_inc(&dip->pending_bios);
8281 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8282 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8285 if (unlikely(map_length < submit_len + blocksize ||
8286 bio_add_page(bio, bvec->bv_page, blocksize,
8287 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8289 * inc the count before we submit the bio so
8290 * we know the end IO handler won't happen before
8291 * we inc the count. Otherwise, the dip might get freed
8292 * before we're done setting it up
8294 atomic_inc(&dip->pending_bios);
8295 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8296 file_offset, skip_sum,
8300 atomic_dec(&dip->pending_bios);
8304 start_sector += submit_len >> 9;
8305 file_offset += submit_len;
8309 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8310 start_sector, GFP_NOFS);
8313 bio->bi_private = dip;
8314 bio->bi_end_io = btrfs_end_dio_bio;
8315 btrfs_io_bio(bio)->logical = file_offset;
8317 map_length = orig_bio->bi_iter.bi_size;
8318 ret = btrfs_map_block(root->fs_info, rw,
8320 &map_length, NULL, 0);
8328 submit_len += blocksize;
8338 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8347 * before atomic variable goto zero, we must
8348 * make sure dip->errors is perceived to be set.
8350 smp_mb__before_atomic();
8351 if (atomic_dec_and_test(&dip->pending_bios))
8352 bio_io_error(dip->orig_bio);
8354 /* bio_end_io() will handle error, so we needn't return it */
8358 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8359 struct inode *inode, loff_t file_offset)
8361 struct btrfs_dio_private *dip = NULL;
8362 struct bio *io_bio = NULL;
8363 struct btrfs_io_bio *btrfs_bio;
8365 int write = rw & REQ_WRITE;
8368 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8370 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8376 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8382 dip->private = dio_bio->bi_private;
8384 dip->logical_offset = file_offset;
8385 dip->bytes = dio_bio->bi_iter.bi_size;
8386 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8387 io_bio->bi_private = dip;
8388 dip->orig_bio = io_bio;
8389 dip->dio_bio = dio_bio;
8390 atomic_set(&dip->pending_bios, 0);
8391 btrfs_bio = btrfs_io_bio(io_bio);
8392 btrfs_bio->logical = file_offset;
8395 io_bio->bi_end_io = btrfs_endio_direct_write;
8397 io_bio->bi_end_io = btrfs_endio_direct_read;
8398 dip->subio_endio = btrfs_subio_endio_read;
8402 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8403 * even if we fail to submit a bio, because in such case we do the
8404 * corresponding error handling below and it must not be done a second
8405 * time by btrfs_direct_IO().
8408 struct btrfs_dio_data *dio_data = current->journal_info;
8410 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8412 dio_data->unsubmitted_oe_range_start =
8413 dio_data->unsubmitted_oe_range_end;
8416 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8420 if (btrfs_bio->end_io)
8421 btrfs_bio->end_io(btrfs_bio, ret);
8425 * If we arrived here it means either we failed to submit the dip
8426 * or we either failed to clone the dio_bio or failed to allocate the
8427 * dip. If we cloned the dio_bio and allocated the dip, we can just
8428 * call bio_endio against our io_bio so that we get proper resource
8429 * cleanup if we fail to submit the dip, otherwise, we must do the
8430 * same as btrfs_endio_direct_[write|read] because we can't call these
8431 * callbacks - they require an allocated dip and a clone of dio_bio.
8433 if (io_bio && dip) {
8434 io_bio->bi_error = -EIO;
8437 * The end io callbacks free our dip, do the final put on io_bio
8438 * and all the cleanup and final put for dio_bio (through
8445 btrfs_endio_direct_write_update_ordered(inode,
8447 dio_bio->bi_iter.bi_size,
8450 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8451 file_offset + dio_bio->bi_iter.bi_size - 1);
8453 dio_bio->bi_error = -EIO;
8455 * Releases and cleans up our dio_bio, no need to bio_put()
8456 * nor bio_endio()/bio_io_error() against dio_bio.
8458 dio_end_io(dio_bio, ret);
8465 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8466 const struct iov_iter *iter, loff_t offset)
8470 unsigned blocksize_mask = root->sectorsize - 1;
8471 ssize_t retval = -EINVAL;
8473 if (offset & blocksize_mask)
8476 if (iov_iter_alignment(iter) & blocksize_mask)
8479 /* If this is a write we don't need to check anymore */
8480 if (iov_iter_rw(iter) == WRITE)
8483 * Check to make sure we don't have duplicate iov_base's in this
8484 * iovec, if so return EINVAL, otherwise we'll get csum errors
8485 * when reading back.
8487 for (seg = 0; seg < iter->nr_segs; seg++) {
8488 for (i = seg + 1; i < iter->nr_segs; i++) {
8489 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8498 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8501 struct file *file = iocb->ki_filp;
8502 struct inode *inode = file->f_mapping->host;
8503 struct btrfs_root *root = BTRFS_I(inode)->root;
8504 struct btrfs_dio_data dio_data = { 0 };
8508 bool relock = false;
8511 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8514 inode_dio_begin(inode);
8515 smp_mb__after_atomic();
8518 * The generic stuff only does filemap_write_and_wait_range, which
8519 * isn't enough if we've written compressed pages to this area, so
8520 * we need to flush the dirty pages again to make absolutely sure
8521 * that any outstanding dirty pages are on disk.
8523 count = iov_iter_count(iter);
8524 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8525 &BTRFS_I(inode)->runtime_flags))
8526 filemap_fdatawrite_range(inode->i_mapping, offset,
8527 offset + count - 1);
8529 if (iov_iter_rw(iter) == WRITE) {
8531 * If the write DIO is beyond the EOF, we need update
8532 * the isize, but it is protected by i_mutex. So we can
8533 * not unlock the i_mutex at this case.
8535 if (offset + count <= inode->i_size) {
8536 inode_unlock(inode);
8539 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8542 dio_data.outstanding_extents = div64_u64(count +
8543 BTRFS_MAX_EXTENT_SIZE - 1,
8544 BTRFS_MAX_EXTENT_SIZE);
8547 * We need to know how many extents we reserved so that we can
8548 * do the accounting properly if we go over the number we
8549 * originally calculated. Abuse current->journal_info for this.
8551 dio_data.reserve = round_up(count, root->sectorsize);
8552 dio_data.unsubmitted_oe_range_start = (u64)offset;
8553 dio_data.unsubmitted_oe_range_end = (u64)offset;
8554 current->journal_info = &dio_data;
8555 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8556 &BTRFS_I(inode)->runtime_flags)) {
8557 inode_dio_end(inode);
8558 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8562 ret = __blockdev_direct_IO(iocb, inode,
8563 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8564 iter, offset, btrfs_get_blocks_direct, NULL,
8565 btrfs_submit_direct, flags);
8566 if (iov_iter_rw(iter) == WRITE) {
8567 current->journal_info = NULL;
8568 if (ret < 0 && ret != -EIOCBQUEUED) {
8569 if (dio_data.reserve)
8570 btrfs_delalloc_release_space(inode, offset,
8573 * On error we might have left some ordered extents
8574 * without submitting corresponding bios for them, so
8575 * cleanup them up to avoid other tasks getting them
8576 * and waiting for them to complete forever.
8578 if (dio_data.unsubmitted_oe_range_start <
8579 dio_data.unsubmitted_oe_range_end)
8580 btrfs_endio_direct_write_update_ordered(inode,
8581 dio_data.unsubmitted_oe_range_start,
8582 dio_data.unsubmitted_oe_range_end -
8583 dio_data.unsubmitted_oe_range_start,
8585 } else if (ret >= 0 && (size_t)ret < count)
8586 btrfs_delalloc_release_space(inode, offset,
8587 count - (size_t)ret);
8591 inode_dio_end(inode);
8598 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8600 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8601 __u64 start, __u64 len)
8605 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8609 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8612 int btrfs_readpage(struct file *file, struct page *page)
8614 struct extent_io_tree *tree;
8615 tree = &BTRFS_I(page->mapping->host)->io_tree;
8616 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8619 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8621 struct extent_io_tree *tree;
8622 struct inode *inode = page->mapping->host;
8625 if (current->flags & PF_MEMALLOC) {
8626 redirty_page_for_writepage(wbc, page);
8632 * If we are under memory pressure we will call this directly from the
8633 * VM, we need to make sure we have the inode referenced for the ordered
8634 * extent. If not just return like we didn't do anything.
8636 if (!igrab(inode)) {
8637 redirty_page_for_writepage(wbc, page);
8638 return AOP_WRITEPAGE_ACTIVATE;
8640 tree = &BTRFS_I(page->mapping->host)->io_tree;
8641 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8642 btrfs_add_delayed_iput(inode);
8646 static int btrfs_writepages(struct address_space *mapping,
8647 struct writeback_control *wbc)
8649 struct extent_io_tree *tree;
8651 tree = &BTRFS_I(mapping->host)->io_tree;
8652 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8656 btrfs_readpages(struct file *file, struct address_space *mapping,
8657 struct list_head *pages, unsigned nr_pages)
8659 struct extent_io_tree *tree;
8660 tree = &BTRFS_I(mapping->host)->io_tree;
8661 return extent_readpages(tree, mapping, pages, nr_pages,
8664 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8666 struct extent_io_tree *tree;
8667 struct extent_map_tree *map;
8670 tree = &BTRFS_I(page->mapping->host)->io_tree;
8671 map = &BTRFS_I(page->mapping->host)->extent_tree;
8672 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8674 ClearPagePrivate(page);
8675 set_page_private(page, 0);
8676 page_cache_release(page);
8681 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8683 if (PageWriteback(page) || PageDirty(page))
8685 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8688 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8689 unsigned int length)
8691 struct inode *inode = page->mapping->host;
8692 struct extent_io_tree *tree;
8693 struct btrfs_ordered_extent *ordered;
8694 struct extent_state *cached_state = NULL;
8695 u64 page_start = page_offset(page);
8696 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8699 int inode_evicting = inode->i_state & I_FREEING;
8702 * we have the page locked, so new writeback can't start,
8703 * and the dirty bit won't be cleared while we are here.
8705 * Wait for IO on this page so that we can safely clear
8706 * the PagePrivate2 bit and do ordered accounting
8708 wait_on_page_writeback(page);
8710 tree = &BTRFS_I(inode)->io_tree;
8712 btrfs_releasepage(page, GFP_NOFS);
8716 if (!inode_evicting)
8717 lock_extent_bits(tree, page_start, page_end, &cached_state);
8720 ordered = btrfs_lookup_ordered_range(inode, start,
8721 page_end - start + 1);
8723 end = min(page_end, ordered->file_offset + ordered->len - 1);
8725 * IO on this page will never be started, so we need
8726 * to account for any ordered extents now
8728 if (!inode_evicting)
8729 clear_extent_bit(tree, start, end,
8730 EXTENT_DIRTY | EXTENT_DELALLOC |
8731 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8732 EXTENT_DEFRAG, 1, 0, &cached_state,
8735 * whoever cleared the private bit is responsible
8736 * for the finish_ordered_io
8738 if (TestClearPagePrivate2(page)) {
8739 struct btrfs_ordered_inode_tree *tree;
8742 tree = &BTRFS_I(inode)->ordered_tree;
8744 spin_lock_irq(&tree->lock);
8745 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8746 new_len = start - ordered->file_offset;
8747 if (new_len < ordered->truncated_len)
8748 ordered->truncated_len = new_len;
8749 spin_unlock_irq(&tree->lock);
8751 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8753 end - start + 1, 1))
8754 btrfs_finish_ordered_io(ordered);
8756 btrfs_put_ordered_extent(ordered);
8757 if (!inode_evicting) {
8758 cached_state = NULL;
8759 lock_extent_bits(tree, start, end,
8764 if (start < page_end)
8769 * Qgroup reserved space handler
8770 * Page here will be either
8771 * 1) Already written to disk
8772 * In this case, its reserved space is released from data rsv map
8773 * and will be freed by delayed_ref handler finally.
8774 * So even we call qgroup_free_data(), it won't decrease reserved
8776 * 2) Not written to disk
8777 * This means the reserved space should be freed here.
8779 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8780 if (!inode_evicting) {
8781 clear_extent_bit(tree, page_start, page_end,
8782 EXTENT_LOCKED | EXTENT_DIRTY |
8783 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8784 EXTENT_DEFRAG, 1, 1,
8785 &cached_state, GFP_NOFS);
8787 __btrfs_releasepage(page, GFP_NOFS);
8790 ClearPageChecked(page);
8791 if (PagePrivate(page)) {
8792 ClearPagePrivate(page);
8793 set_page_private(page, 0);
8794 page_cache_release(page);
8799 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8800 * called from a page fault handler when a page is first dirtied. Hence we must
8801 * be careful to check for EOF conditions here. We set the page up correctly
8802 * for a written page which means we get ENOSPC checking when writing into
8803 * holes and correct delalloc and unwritten extent mapping on filesystems that
8804 * support these features.
8806 * We are not allowed to take the i_mutex here so we have to play games to
8807 * protect against truncate races as the page could now be beyond EOF. Because
8808 * vmtruncate() writes the inode size before removing pages, once we have the
8809 * page lock we can determine safely if the page is beyond EOF. If it is not
8810 * beyond EOF, then the page is guaranteed safe against truncation until we
8813 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8815 struct page *page = vmf->page;
8816 struct inode *inode = file_inode(vma->vm_file);
8817 struct btrfs_root *root = BTRFS_I(inode)->root;
8818 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8819 struct btrfs_ordered_extent *ordered;
8820 struct extent_state *cached_state = NULL;
8822 unsigned long zero_start;
8831 reserved_space = PAGE_CACHE_SIZE;
8833 sb_start_pagefault(inode->i_sb);
8834 page_start = page_offset(page);
8835 page_end = page_start + PAGE_CACHE_SIZE - 1;
8839 * Reserving delalloc space after obtaining the page lock can lead to
8840 * deadlock. For example, if a dirty page is locked by this function
8841 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8842 * dirty page write out, then the btrfs_writepage() function could
8843 * end up waiting indefinitely to get a lock on the page currently
8844 * being processed by btrfs_page_mkwrite() function.
8846 ret = btrfs_delalloc_reserve_space(inode, page_start,
8849 ret = file_update_time(vma->vm_file);
8855 else /* -ENOSPC, -EIO, etc */
8856 ret = VM_FAULT_SIGBUS;
8862 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8865 size = i_size_read(inode);
8867 if ((page->mapping != inode->i_mapping) ||
8868 (page_start >= size)) {
8869 /* page got truncated out from underneath us */
8872 wait_on_page_writeback(page);
8874 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8875 set_page_extent_mapped(page);
8878 * we can't set the delalloc bits if there are pending ordered
8879 * extents. Drop our locks and wait for them to finish
8881 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8883 unlock_extent_cached(io_tree, page_start, page_end,
8884 &cached_state, GFP_NOFS);
8886 btrfs_start_ordered_extent(inode, ordered, 1);
8887 btrfs_put_ordered_extent(ordered);
8891 if (page->index == ((size - 1) >> PAGE_CACHE_SHIFT)) {
8892 reserved_space = round_up(size - page_start, root->sectorsize);
8893 if (reserved_space < PAGE_CACHE_SIZE) {
8894 end = page_start + reserved_space - 1;
8895 spin_lock(&BTRFS_I(inode)->lock);
8896 BTRFS_I(inode)->outstanding_extents++;
8897 spin_unlock(&BTRFS_I(inode)->lock);
8898 btrfs_delalloc_release_space(inode, page_start,
8899 PAGE_CACHE_SIZE - reserved_space);
8904 * XXX - page_mkwrite gets called every time the page is dirtied, even
8905 * if it was already dirty, so for space accounting reasons we need to
8906 * clear any delalloc bits for the range we are fixing to save. There
8907 * is probably a better way to do this, but for now keep consistent with
8908 * prepare_pages in the normal write path.
8910 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8911 EXTENT_DIRTY | EXTENT_DELALLOC |
8912 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8913 0, 0, &cached_state, GFP_NOFS);
8915 ret = btrfs_set_extent_delalloc(inode, page_start, end,
8918 unlock_extent_cached(io_tree, page_start, page_end,
8919 &cached_state, GFP_NOFS);
8920 ret = VM_FAULT_SIGBUS;
8925 /* page is wholly or partially inside EOF */
8926 if (page_start + PAGE_CACHE_SIZE > size)
8927 zero_start = size & ~PAGE_CACHE_MASK;
8929 zero_start = PAGE_CACHE_SIZE;
8931 if (zero_start != PAGE_CACHE_SIZE) {
8933 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8934 flush_dcache_page(page);
8937 ClearPageChecked(page);
8938 set_page_dirty(page);
8939 SetPageUptodate(page);
8941 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8942 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8943 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8945 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8949 sb_end_pagefault(inode->i_sb);
8950 return VM_FAULT_LOCKED;
8954 btrfs_delalloc_release_space(inode, page_start, reserved_space);
8956 sb_end_pagefault(inode->i_sb);
8960 static int btrfs_truncate(struct inode *inode)
8962 struct btrfs_root *root = BTRFS_I(inode)->root;
8963 struct btrfs_block_rsv *rsv;
8966 struct btrfs_trans_handle *trans;
8967 u64 mask = root->sectorsize - 1;
8968 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8970 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8976 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8977 * 3 things going on here
8979 * 1) We need to reserve space for our orphan item and the space to
8980 * delete our orphan item. Lord knows we don't want to have a dangling
8981 * orphan item because we didn't reserve space to remove it.
8983 * 2) We need to reserve space to update our inode.
8985 * 3) We need to have something to cache all the space that is going to
8986 * be free'd up by the truncate operation, but also have some slack
8987 * space reserved in case it uses space during the truncate (thank you
8988 * very much snapshotting).
8990 * And we need these to all be seperate. The fact is we can use alot of
8991 * space doing the truncate, and we have no earthly idea how much space
8992 * we will use, so we need the truncate reservation to be seperate so it
8993 * doesn't end up using space reserved for updating the inode or
8994 * removing the orphan item. We also need to be able to stop the
8995 * transaction and start a new one, which means we need to be able to
8996 * update the inode several times, and we have no idea of knowing how
8997 * many times that will be, so we can't just reserve 1 item for the
8998 * entirety of the opration, so that has to be done seperately as well.
8999 * Then there is the orphan item, which does indeed need to be held on
9000 * to for the whole operation, and we need nobody to touch this reserved
9001 * space except the orphan code.
9003 * So that leaves us with
9005 * 1) root->orphan_block_rsv - for the orphan deletion.
9006 * 2) rsv - for the truncate reservation, which we will steal from the
9007 * transaction reservation.
9008 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9009 * updating the inode.
9011 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9014 rsv->size = min_size;
9018 * 1 for the truncate slack space
9019 * 1 for updating the inode.
9021 trans = btrfs_start_transaction(root, 2);
9022 if (IS_ERR(trans)) {
9023 err = PTR_ERR(trans);
9027 /* Migrate the slack space for the truncate to our reserve */
9028 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9033 * So if we truncate and then write and fsync we normally would just
9034 * write the extents that changed, which is a problem if we need to
9035 * first truncate that entire inode. So set this flag so we write out
9036 * all of the extents in the inode to the sync log so we're completely
9039 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9040 trans->block_rsv = rsv;
9043 ret = btrfs_truncate_inode_items(trans, root, inode,
9045 BTRFS_EXTENT_DATA_KEY);
9046 if (ret != -ENOSPC && ret != -EAGAIN) {
9051 trans->block_rsv = &root->fs_info->trans_block_rsv;
9052 ret = btrfs_update_inode(trans, root, inode);
9058 btrfs_end_transaction(trans, root);
9059 btrfs_btree_balance_dirty(root);
9061 trans = btrfs_start_transaction(root, 2);
9062 if (IS_ERR(trans)) {
9063 ret = err = PTR_ERR(trans);
9068 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9070 BUG_ON(ret); /* shouldn't happen */
9071 trans->block_rsv = rsv;
9074 if (ret == 0 && inode->i_nlink > 0) {
9075 trans->block_rsv = root->orphan_block_rsv;
9076 ret = btrfs_orphan_del(trans, inode);
9082 trans->block_rsv = &root->fs_info->trans_block_rsv;
9083 ret = btrfs_update_inode(trans, root, inode);
9087 ret = btrfs_end_transaction(trans, root);
9088 btrfs_btree_balance_dirty(root);
9092 btrfs_free_block_rsv(root, rsv);
9101 * create a new subvolume directory/inode (helper for the ioctl).
9103 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9104 struct btrfs_root *new_root,
9105 struct btrfs_root *parent_root,
9108 struct inode *inode;
9112 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9113 new_dirid, new_dirid,
9114 S_IFDIR | (~current_umask() & S_IRWXUGO),
9117 return PTR_ERR(inode);
9118 inode->i_op = &btrfs_dir_inode_operations;
9119 inode->i_fop = &btrfs_dir_file_operations;
9121 set_nlink(inode, 1);
9122 btrfs_i_size_write(inode, 0);
9123 unlock_new_inode(inode);
9125 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9127 btrfs_err(new_root->fs_info,
9128 "error inheriting subvolume %llu properties: %d",
9129 new_root->root_key.objectid, err);
9131 err = btrfs_update_inode(trans, new_root, inode);
9137 struct inode *btrfs_alloc_inode(struct super_block *sb)
9139 struct btrfs_inode *ei;
9140 struct inode *inode;
9142 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9149 ei->last_sub_trans = 0;
9150 ei->logged_trans = 0;
9151 ei->delalloc_bytes = 0;
9152 ei->defrag_bytes = 0;
9153 ei->disk_i_size = 0;
9156 ei->index_cnt = (u64)-1;
9158 ei->last_unlink_trans = 0;
9159 ei->last_log_commit = 0;
9160 ei->delayed_iput_count = 0;
9162 spin_lock_init(&ei->lock);
9163 ei->outstanding_extents = 0;
9164 ei->reserved_extents = 0;
9166 ei->runtime_flags = 0;
9167 ei->force_compress = BTRFS_COMPRESS_NONE;
9169 ei->delayed_node = NULL;
9171 ei->i_otime.tv_sec = 0;
9172 ei->i_otime.tv_nsec = 0;
9174 inode = &ei->vfs_inode;
9175 extent_map_tree_init(&ei->extent_tree);
9176 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9177 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9178 ei->io_tree.track_uptodate = 1;
9179 ei->io_failure_tree.track_uptodate = 1;
9180 atomic_set(&ei->sync_writers, 0);
9181 mutex_init(&ei->log_mutex);
9182 mutex_init(&ei->delalloc_mutex);
9183 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9184 INIT_LIST_HEAD(&ei->delalloc_inodes);
9185 INIT_LIST_HEAD(&ei->delayed_iput);
9186 RB_CLEAR_NODE(&ei->rb_node);
9191 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9192 void btrfs_test_destroy_inode(struct inode *inode)
9194 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9195 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9199 static void btrfs_i_callback(struct rcu_head *head)
9201 struct inode *inode = container_of(head, struct inode, i_rcu);
9202 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9205 void btrfs_destroy_inode(struct inode *inode)
9207 struct btrfs_ordered_extent *ordered;
9208 struct btrfs_root *root = BTRFS_I(inode)->root;
9210 WARN_ON(!hlist_empty(&inode->i_dentry));
9211 WARN_ON(inode->i_data.nrpages);
9212 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9213 WARN_ON(BTRFS_I(inode)->reserved_extents);
9214 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9215 WARN_ON(BTRFS_I(inode)->csum_bytes);
9216 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9219 * This can happen where we create an inode, but somebody else also
9220 * created the same inode and we need to destroy the one we already
9226 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9227 &BTRFS_I(inode)->runtime_flags)) {
9228 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9230 atomic_dec(&root->orphan_inodes);
9234 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9238 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9239 ordered->file_offset, ordered->len);
9240 btrfs_remove_ordered_extent(inode, ordered);
9241 btrfs_put_ordered_extent(ordered);
9242 btrfs_put_ordered_extent(ordered);
9245 btrfs_qgroup_check_reserved_leak(inode);
9246 inode_tree_del(inode);
9247 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9249 call_rcu(&inode->i_rcu, btrfs_i_callback);
9252 int btrfs_drop_inode(struct inode *inode)
9254 struct btrfs_root *root = BTRFS_I(inode)->root;
9259 /* the snap/subvol tree is on deleting */
9260 if (btrfs_root_refs(&root->root_item) == 0)
9263 return generic_drop_inode(inode);
9266 static void init_once(void *foo)
9268 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9270 inode_init_once(&ei->vfs_inode);
9273 void btrfs_destroy_cachep(void)
9276 * Make sure all delayed rcu free inodes are flushed before we
9280 if (btrfs_inode_cachep)
9281 kmem_cache_destroy(btrfs_inode_cachep);
9282 if (btrfs_trans_handle_cachep)
9283 kmem_cache_destroy(btrfs_trans_handle_cachep);
9284 if (btrfs_transaction_cachep)
9285 kmem_cache_destroy(btrfs_transaction_cachep);
9286 if (btrfs_path_cachep)
9287 kmem_cache_destroy(btrfs_path_cachep);
9288 if (btrfs_free_space_cachep)
9289 kmem_cache_destroy(btrfs_free_space_cachep);
9292 int btrfs_init_cachep(void)
9294 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9295 sizeof(struct btrfs_inode), 0,
9296 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9298 if (!btrfs_inode_cachep)
9301 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9302 sizeof(struct btrfs_trans_handle), 0,
9303 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9304 if (!btrfs_trans_handle_cachep)
9307 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9308 sizeof(struct btrfs_transaction), 0,
9309 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9310 if (!btrfs_transaction_cachep)
9313 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9314 sizeof(struct btrfs_path), 0,
9315 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9316 if (!btrfs_path_cachep)
9319 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9320 sizeof(struct btrfs_free_space), 0,
9321 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9322 if (!btrfs_free_space_cachep)
9327 btrfs_destroy_cachep();
9331 static int btrfs_getattr(struct vfsmount *mnt,
9332 struct dentry *dentry, struct kstat *stat)
9335 struct inode *inode = d_inode(dentry);
9336 u32 blocksize = inode->i_sb->s_blocksize;
9338 generic_fillattr(inode, stat);
9339 stat->dev = BTRFS_I(inode)->root->anon_dev;
9341 spin_lock(&BTRFS_I(inode)->lock);
9342 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9343 spin_unlock(&BTRFS_I(inode)->lock);
9344 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9345 ALIGN(delalloc_bytes, blocksize)) >> 9;
9349 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9350 struct inode *new_dir, struct dentry *new_dentry)
9352 struct btrfs_trans_handle *trans;
9353 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9354 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9355 struct inode *new_inode = d_inode(new_dentry);
9356 struct inode *old_inode = d_inode(old_dentry);
9357 struct timespec ctime = CURRENT_TIME;
9361 u64 old_ino = btrfs_ino(old_inode);
9363 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9366 /* we only allow rename subvolume link between subvolumes */
9367 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9370 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9371 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9374 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9375 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9379 /* check for collisions, even if the name isn't there */
9380 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9381 new_dentry->d_name.name,
9382 new_dentry->d_name.len);
9385 if (ret == -EEXIST) {
9387 * eexist without a new_inode */
9388 if (WARN_ON(!new_inode)) {
9392 /* maybe -EOVERFLOW */
9399 * we're using rename to replace one file with another. Start IO on it
9400 * now so we don't add too much work to the end of the transaction
9402 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9403 filemap_flush(old_inode->i_mapping);
9405 /* close the racy window with snapshot create/destroy ioctl */
9406 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9407 down_read(&root->fs_info->subvol_sem);
9409 * We want to reserve the absolute worst case amount of items. So if
9410 * both inodes are subvols and we need to unlink them then that would
9411 * require 4 item modifications, but if they are both normal inodes it
9412 * would require 5 item modifications, so we'll assume their normal
9413 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9414 * should cover the worst case number of items we'll modify.
9416 trans = btrfs_start_transaction(root, 11);
9417 if (IS_ERR(trans)) {
9418 ret = PTR_ERR(trans);
9423 btrfs_record_root_in_trans(trans, dest);
9425 ret = btrfs_set_inode_index(new_dir, &index);
9429 BTRFS_I(old_inode)->dir_index = 0ULL;
9430 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9431 /* force full log commit if subvolume involved. */
9432 btrfs_set_log_full_commit(root->fs_info, trans);
9434 ret = btrfs_insert_inode_ref(trans, dest,
9435 new_dentry->d_name.name,
9436 new_dentry->d_name.len,
9438 btrfs_ino(new_dir), index);
9442 * this is an ugly little race, but the rename is required
9443 * to make sure that if we crash, the inode is either at the
9444 * old name or the new one. pinning the log transaction lets
9445 * us make sure we don't allow a log commit to come in after
9446 * we unlink the name but before we add the new name back in.
9448 btrfs_pin_log_trans(root);
9451 inode_inc_iversion(old_dir);
9452 inode_inc_iversion(new_dir);
9453 inode_inc_iversion(old_inode);
9454 old_dir->i_ctime = old_dir->i_mtime = ctime;
9455 new_dir->i_ctime = new_dir->i_mtime = ctime;
9456 old_inode->i_ctime = ctime;
9458 if (old_dentry->d_parent != new_dentry->d_parent)
9459 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9461 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9462 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9463 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9464 old_dentry->d_name.name,
9465 old_dentry->d_name.len);
9467 ret = __btrfs_unlink_inode(trans, root, old_dir,
9468 d_inode(old_dentry),
9469 old_dentry->d_name.name,
9470 old_dentry->d_name.len);
9472 ret = btrfs_update_inode(trans, root, old_inode);
9475 btrfs_abort_transaction(trans, root, ret);
9480 inode_inc_iversion(new_inode);
9481 new_inode->i_ctime = CURRENT_TIME;
9482 if (unlikely(btrfs_ino(new_inode) ==
9483 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9484 root_objectid = BTRFS_I(new_inode)->location.objectid;
9485 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9487 new_dentry->d_name.name,
9488 new_dentry->d_name.len);
9489 BUG_ON(new_inode->i_nlink == 0);
9491 ret = btrfs_unlink_inode(trans, dest, new_dir,
9492 d_inode(new_dentry),
9493 new_dentry->d_name.name,
9494 new_dentry->d_name.len);
9496 if (!ret && new_inode->i_nlink == 0)
9497 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9499 btrfs_abort_transaction(trans, root, ret);
9504 ret = btrfs_add_link(trans, new_dir, old_inode,
9505 new_dentry->d_name.name,
9506 new_dentry->d_name.len, 0, index);
9508 btrfs_abort_transaction(trans, root, ret);
9512 if (old_inode->i_nlink == 1)
9513 BTRFS_I(old_inode)->dir_index = index;
9515 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9516 struct dentry *parent = new_dentry->d_parent;
9517 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9518 btrfs_end_log_trans(root);
9521 btrfs_end_transaction(trans, root);
9523 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9524 up_read(&root->fs_info->subvol_sem);
9529 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9530 struct inode *new_dir, struct dentry *new_dentry,
9533 if (flags & ~RENAME_NOREPLACE)
9536 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9539 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9541 struct btrfs_delalloc_work *delalloc_work;
9542 struct inode *inode;
9544 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9546 inode = delalloc_work->inode;
9547 filemap_flush(inode->i_mapping);
9548 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9549 &BTRFS_I(inode)->runtime_flags))
9550 filemap_flush(inode->i_mapping);
9552 if (delalloc_work->delay_iput)
9553 btrfs_add_delayed_iput(inode);
9556 complete(&delalloc_work->completion);
9559 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9562 struct btrfs_delalloc_work *work;
9564 work = kmalloc(sizeof(*work), GFP_NOFS);
9568 init_completion(&work->completion);
9569 INIT_LIST_HEAD(&work->list);
9570 work->inode = inode;
9571 work->delay_iput = delay_iput;
9572 WARN_ON_ONCE(!inode);
9573 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9574 btrfs_run_delalloc_work, NULL, NULL);
9579 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9581 wait_for_completion(&work->completion);
9586 * some fairly slow code that needs optimization. This walks the list
9587 * of all the inodes with pending delalloc and forces them to disk.
9589 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9592 struct btrfs_inode *binode;
9593 struct inode *inode;
9594 struct btrfs_delalloc_work *work, *next;
9595 struct list_head works;
9596 struct list_head splice;
9599 INIT_LIST_HEAD(&works);
9600 INIT_LIST_HEAD(&splice);
9602 mutex_lock(&root->delalloc_mutex);
9603 spin_lock(&root->delalloc_lock);
9604 list_splice_init(&root->delalloc_inodes, &splice);
9605 while (!list_empty(&splice)) {
9606 binode = list_entry(splice.next, struct btrfs_inode,
9609 list_move_tail(&binode->delalloc_inodes,
9610 &root->delalloc_inodes);
9611 inode = igrab(&binode->vfs_inode);
9613 cond_resched_lock(&root->delalloc_lock);
9616 spin_unlock(&root->delalloc_lock);
9618 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9621 btrfs_add_delayed_iput(inode);
9627 list_add_tail(&work->list, &works);
9628 btrfs_queue_work(root->fs_info->flush_workers,
9631 if (nr != -1 && ret >= nr)
9634 spin_lock(&root->delalloc_lock);
9636 spin_unlock(&root->delalloc_lock);
9639 list_for_each_entry_safe(work, next, &works, list) {
9640 list_del_init(&work->list);
9641 btrfs_wait_and_free_delalloc_work(work);
9644 if (!list_empty_careful(&splice)) {
9645 spin_lock(&root->delalloc_lock);
9646 list_splice_tail(&splice, &root->delalloc_inodes);
9647 spin_unlock(&root->delalloc_lock);
9649 mutex_unlock(&root->delalloc_mutex);
9653 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9657 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9660 ret = __start_delalloc_inodes(root, delay_iput, -1);
9664 * the filemap_flush will queue IO into the worker threads, but
9665 * we have to make sure the IO is actually started and that
9666 * ordered extents get created before we return
9668 atomic_inc(&root->fs_info->async_submit_draining);
9669 while (atomic_read(&root->fs_info->nr_async_submits) ||
9670 atomic_read(&root->fs_info->async_delalloc_pages)) {
9671 wait_event(root->fs_info->async_submit_wait,
9672 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9673 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9675 atomic_dec(&root->fs_info->async_submit_draining);
9679 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9682 struct btrfs_root *root;
9683 struct list_head splice;
9686 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9689 INIT_LIST_HEAD(&splice);
9691 mutex_lock(&fs_info->delalloc_root_mutex);
9692 spin_lock(&fs_info->delalloc_root_lock);
9693 list_splice_init(&fs_info->delalloc_roots, &splice);
9694 while (!list_empty(&splice) && nr) {
9695 root = list_first_entry(&splice, struct btrfs_root,
9697 root = btrfs_grab_fs_root(root);
9699 list_move_tail(&root->delalloc_root,
9700 &fs_info->delalloc_roots);
9701 spin_unlock(&fs_info->delalloc_root_lock);
9703 ret = __start_delalloc_inodes(root, delay_iput, nr);
9704 btrfs_put_fs_root(root);
9712 spin_lock(&fs_info->delalloc_root_lock);
9714 spin_unlock(&fs_info->delalloc_root_lock);
9717 atomic_inc(&fs_info->async_submit_draining);
9718 while (atomic_read(&fs_info->nr_async_submits) ||
9719 atomic_read(&fs_info->async_delalloc_pages)) {
9720 wait_event(fs_info->async_submit_wait,
9721 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9722 atomic_read(&fs_info->async_delalloc_pages) == 0));
9724 atomic_dec(&fs_info->async_submit_draining);
9726 if (!list_empty_careful(&splice)) {
9727 spin_lock(&fs_info->delalloc_root_lock);
9728 list_splice_tail(&splice, &fs_info->delalloc_roots);
9729 spin_unlock(&fs_info->delalloc_root_lock);
9731 mutex_unlock(&fs_info->delalloc_root_mutex);
9735 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9736 const char *symname)
9738 struct btrfs_trans_handle *trans;
9739 struct btrfs_root *root = BTRFS_I(dir)->root;
9740 struct btrfs_path *path;
9741 struct btrfs_key key;
9742 struct inode *inode = NULL;
9750 struct btrfs_file_extent_item *ei;
9751 struct extent_buffer *leaf;
9753 name_len = strlen(symname);
9754 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9755 return -ENAMETOOLONG;
9758 * 2 items for inode item and ref
9759 * 2 items for dir items
9760 * 1 item for updating parent inode item
9761 * 1 item for the inline extent item
9762 * 1 item for xattr if selinux is on
9764 trans = btrfs_start_transaction(root, 7);
9766 return PTR_ERR(trans);
9768 err = btrfs_find_free_ino(root, &objectid);
9772 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9773 dentry->d_name.len, btrfs_ino(dir), objectid,
9774 S_IFLNK|S_IRWXUGO, &index);
9775 if (IS_ERR(inode)) {
9776 err = PTR_ERR(inode);
9781 * If the active LSM wants to access the inode during
9782 * d_instantiate it needs these. Smack checks to see
9783 * if the filesystem supports xattrs by looking at the
9786 inode->i_fop = &btrfs_file_operations;
9787 inode->i_op = &btrfs_file_inode_operations;
9788 inode->i_mapping->a_ops = &btrfs_aops;
9789 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9791 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9793 goto out_unlock_inode;
9795 path = btrfs_alloc_path();
9798 goto out_unlock_inode;
9800 key.objectid = btrfs_ino(inode);
9802 key.type = BTRFS_EXTENT_DATA_KEY;
9803 datasize = btrfs_file_extent_calc_inline_size(name_len);
9804 err = btrfs_insert_empty_item(trans, root, path, &key,
9807 btrfs_free_path(path);
9808 goto out_unlock_inode;
9810 leaf = path->nodes[0];
9811 ei = btrfs_item_ptr(leaf, path->slots[0],
9812 struct btrfs_file_extent_item);
9813 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9814 btrfs_set_file_extent_type(leaf, ei,
9815 BTRFS_FILE_EXTENT_INLINE);
9816 btrfs_set_file_extent_encryption(leaf, ei, 0);
9817 btrfs_set_file_extent_compression(leaf, ei, 0);
9818 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9819 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9821 ptr = btrfs_file_extent_inline_start(ei);
9822 write_extent_buffer(leaf, symname, ptr, name_len);
9823 btrfs_mark_buffer_dirty(leaf);
9824 btrfs_free_path(path);
9826 inode->i_op = &btrfs_symlink_inode_operations;
9827 inode_nohighmem(inode);
9828 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9829 inode_set_bytes(inode, name_len);
9830 btrfs_i_size_write(inode, name_len);
9831 err = btrfs_update_inode(trans, root, inode);
9833 * Last step, add directory indexes for our symlink inode. This is the
9834 * last step to avoid extra cleanup of these indexes if an error happens
9838 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9841 goto out_unlock_inode;
9844 unlock_new_inode(inode);
9845 d_instantiate(dentry, inode);
9848 btrfs_end_transaction(trans, root);
9850 inode_dec_link_count(inode);
9853 btrfs_btree_balance_dirty(root);
9858 unlock_new_inode(inode);
9862 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9863 u64 start, u64 num_bytes, u64 min_size,
9864 loff_t actual_len, u64 *alloc_hint,
9865 struct btrfs_trans_handle *trans)
9867 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9868 struct extent_map *em;
9869 struct btrfs_root *root = BTRFS_I(inode)->root;
9870 struct btrfs_key ins;
9871 u64 cur_offset = start;
9874 u64 last_alloc = (u64)-1;
9876 bool own_trans = true;
9880 while (num_bytes > 0) {
9882 trans = btrfs_start_transaction(root, 3);
9883 if (IS_ERR(trans)) {
9884 ret = PTR_ERR(trans);
9889 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9890 cur_bytes = max(cur_bytes, min_size);
9892 * If we are severely fragmented we could end up with really
9893 * small allocations, so if the allocator is returning small
9894 * chunks lets make its job easier by only searching for those
9897 cur_bytes = min(cur_bytes, last_alloc);
9898 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9899 *alloc_hint, &ins, 1, 0);
9902 btrfs_end_transaction(trans, root);
9906 last_alloc = ins.offset;
9907 ret = insert_reserved_file_extent(trans, inode,
9908 cur_offset, ins.objectid,
9909 ins.offset, ins.offset,
9910 ins.offset, 0, 0, 0,
9911 BTRFS_FILE_EXTENT_PREALLOC);
9913 btrfs_free_reserved_extent(root, ins.objectid,
9915 btrfs_abort_transaction(trans, root, ret);
9917 btrfs_end_transaction(trans, root);
9921 btrfs_drop_extent_cache(inode, cur_offset,
9922 cur_offset + ins.offset -1, 0);
9924 em = alloc_extent_map();
9926 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9927 &BTRFS_I(inode)->runtime_flags);
9931 em->start = cur_offset;
9932 em->orig_start = cur_offset;
9933 em->len = ins.offset;
9934 em->block_start = ins.objectid;
9935 em->block_len = ins.offset;
9936 em->orig_block_len = ins.offset;
9937 em->ram_bytes = ins.offset;
9938 em->bdev = root->fs_info->fs_devices->latest_bdev;
9939 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9940 em->generation = trans->transid;
9943 write_lock(&em_tree->lock);
9944 ret = add_extent_mapping(em_tree, em, 1);
9945 write_unlock(&em_tree->lock);
9948 btrfs_drop_extent_cache(inode, cur_offset,
9949 cur_offset + ins.offset - 1,
9952 free_extent_map(em);
9954 num_bytes -= ins.offset;
9955 cur_offset += ins.offset;
9956 *alloc_hint = ins.objectid + ins.offset;
9958 inode_inc_iversion(inode);
9959 inode->i_ctime = CURRENT_TIME;
9960 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9961 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9962 (actual_len > inode->i_size) &&
9963 (cur_offset > inode->i_size)) {
9964 if (cur_offset > actual_len)
9965 i_size = actual_len;
9967 i_size = cur_offset;
9968 i_size_write(inode, i_size);
9969 btrfs_ordered_update_i_size(inode, i_size, NULL);
9972 ret = btrfs_update_inode(trans, root, inode);
9975 btrfs_abort_transaction(trans, root, ret);
9977 btrfs_end_transaction(trans, root);
9982 btrfs_end_transaction(trans, root);
9987 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9988 u64 start, u64 num_bytes, u64 min_size,
9989 loff_t actual_len, u64 *alloc_hint)
9991 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9992 min_size, actual_len, alloc_hint,
9996 int btrfs_prealloc_file_range_trans(struct inode *inode,
9997 struct btrfs_trans_handle *trans, int mode,
9998 u64 start, u64 num_bytes, u64 min_size,
9999 loff_t actual_len, u64 *alloc_hint)
10001 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10002 min_size, actual_len, alloc_hint, trans);
10005 static int btrfs_set_page_dirty(struct page *page)
10007 return __set_page_dirty_nobuffers(page);
10010 static int btrfs_permission(struct inode *inode, int mask)
10012 struct btrfs_root *root = BTRFS_I(inode)->root;
10013 umode_t mode = inode->i_mode;
10015 if (mask & MAY_WRITE &&
10016 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10017 if (btrfs_root_readonly(root))
10019 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10022 return generic_permission(inode, mask);
10025 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10027 struct btrfs_trans_handle *trans;
10028 struct btrfs_root *root = BTRFS_I(dir)->root;
10029 struct inode *inode = NULL;
10035 * 5 units required for adding orphan entry
10037 trans = btrfs_start_transaction(root, 5);
10039 return PTR_ERR(trans);
10041 ret = btrfs_find_free_ino(root, &objectid);
10045 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10046 btrfs_ino(dir), objectid, mode, &index);
10047 if (IS_ERR(inode)) {
10048 ret = PTR_ERR(inode);
10053 inode->i_fop = &btrfs_file_operations;
10054 inode->i_op = &btrfs_file_inode_operations;
10056 inode->i_mapping->a_ops = &btrfs_aops;
10057 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10059 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10063 ret = btrfs_update_inode(trans, root, inode);
10066 ret = btrfs_orphan_add(trans, inode);
10071 * We set number of links to 0 in btrfs_new_inode(), and here we set
10072 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10075 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10077 set_nlink(inode, 1);
10078 unlock_new_inode(inode);
10079 d_tmpfile(dentry, inode);
10080 mark_inode_dirty(inode);
10083 btrfs_end_transaction(trans, root);
10086 btrfs_balance_delayed_items(root);
10087 btrfs_btree_balance_dirty(root);
10091 unlock_new_inode(inode);
10096 /* Inspired by filemap_check_errors() */
10097 int btrfs_inode_check_errors(struct inode *inode)
10101 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10102 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10104 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10105 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10111 static const struct inode_operations btrfs_dir_inode_operations = {
10112 .getattr = btrfs_getattr,
10113 .lookup = btrfs_lookup,
10114 .create = btrfs_create,
10115 .unlink = btrfs_unlink,
10116 .link = btrfs_link,
10117 .mkdir = btrfs_mkdir,
10118 .rmdir = btrfs_rmdir,
10119 .rename2 = btrfs_rename2,
10120 .symlink = btrfs_symlink,
10121 .setattr = btrfs_setattr,
10122 .mknod = btrfs_mknod,
10123 .setxattr = btrfs_setxattr,
10124 .getxattr = generic_getxattr,
10125 .listxattr = btrfs_listxattr,
10126 .removexattr = btrfs_removexattr,
10127 .permission = btrfs_permission,
10128 .get_acl = btrfs_get_acl,
10129 .set_acl = btrfs_set_acl,
10130 .update_time = btrfs_update_time,
10131 .tmpfile = btrfs_tmpfile,
10133 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10134 .lookup = btrfs_lookup,
10135 .permission = btrfs_permission,
10136 .get_acl = btrfs_get_acl,
10137 .set_acl = btrfs_set_acl,
10138 .update_time = btrfs_update_time,
10141 static const struct file_operations btrfs_dir_file_operations = {
10142 .llseek = generic_file_llseek,
10143 .read = generic_read_dir,
10144 .iterate = btrfs_real_readdir,
10145 .unlocked_ioctl = btrfs_ioctl,
10146 #ifdef CONFIG_COMPAT
10147 .compat_ioctl = btrfs_ioctl,
10149 .release = btrfs_release_file,
10150 .fsync = btrfs_sync_file,
10153 static const struct extent_io_ops btrfs_extent_io_ops = {
10154 .fill_delalloc = run_delalloc_range,
10155 .submit_bio_hook = btrfs_submit_bio_hook,
10156 .merge_bio_hook = btrfs_merge_bio_hook,
10157 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10158 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10159 .writepage_start_hook = btrfs_writepage_start_hook,
10160 .set_bit_hook = btrfs_set_bit_hook,
10161 .clear_bit_hook = btrfs_clear_bit_hook,
10162 .merge_extent_hook = btrfs_merge_extent_hook,
10163 .split_extent_hook = btrfs_split_extent_hook,
10167 * btrfs doesn't support the bmap operation because swapfiles
10168 * use bmap to make a mapping of extents in the file. They assume
10169 * these extents won't change over the life of the file and they
10170 * use the bmap result to do IO directly to the drive.
10172 * the btrfs bmap call would return logical addresses that aren't
10173 * suitable for IO and they also will change frequently as COW
10174 * operations happen. So, swapfile + btrfs == corruption.
10176 * For now we're avoiding this by dropping bmap.
10178 static const struct address_space_operations btrfs_aops = {
10179 .readpage = btrfs_readpage,
10180 .writepage = btrfs_writepage,
10181 .writepages = btrfs_writepages,
10182 .readpages = btrfs_readpages,
10183 .direct_IO = btrfs_direct_IO,
10184 .invalidatepage = btrfs_invalidatepage,
10185 .releasepage = btrfs_releasepage,
10186 .set_page_dirty = btrfs_set_page_dirty,
10187 .error_remove_page = generic_error_remove_page,
10190 static const struct address_space_operations btrfs_symlink_aops = {
10191 .readpage = btrfs_readpage,
10192 .writepage = btrfs_writepage,
10193 .invalidatepage = btrfs_invalidatepage,
10194 .releasepage = btrfs_releasepage,
10197 static const struct inode_operations btrfs_file_inode_operations = {
10198 .getattr = btrfs_getattr,
10199 .setattr = btrfs_setattr,
10200 .setxattr = btrfs_setxattr,
10201 .getxattr = generic_getxattr,
10202 .listxattr = btrfs_listxattr,
10203 .removexattr = btrfs_removexattr,
10204 .permission = btrfs_permission,
10205 .fiemap = btrfs_fiemap,
10206 .get_acl = btrfs_get_acl,
10207 .set_acl = btrfs_set_acl,
10208 .update_time = btrfs_update_time,
10210 static const struct inode_operations btrfs_special_inode_operations = {
10211 .getattr = btrfs_getattr,
10212 .setattr = btrfs_setattr,
10213 .permission = btrfs_permission,
10214 .setxattr = btrfs_setxattr,
10215 .getxattr = generic_getxattr,
10216 .listxattr = btrfs_listxattr,
10217 .removexattr = btrfs_removexattr,
10218 .get_acl = btrfs_get_acl,
10219 .set_acl = btrfs_set_acl,
10220 .update_time = btrfs_update_time,
10222 static const struct inode_operations btrfs_symlink_inode_operations = {
10223 .readlink = generic_readlink,
10224 .get_link = page_get_link,
10225 .getattr = btrfs_getattr,
10226 .setattr = btrfs_setattr,
10227 .permission = btrfs_permission,
10228 .setxattr = btrfs_setxattr,
10229 .getxattr = generic_getxattr,
10230 .listxattr = btrfs_listxattr,
10231 .removexattr = btrfs_removexattr,
10232 .update_time = btrfs_update_time,
10235 const struct dentry_operations btrfs_dentry_operations = {
10236 .d_delete = btrfs_dentry_delete,
10237 .d_release = btrfs_dentry_release,