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 static const struct inode_operations btrfs_dir_inode_operations;
70 static const struct inode_operations btrfs_symlink_inode_operations;
71 static const struct inode_operations btrfs_dir_ro_inode_operations;
72 static const struct inode_operations btrfs_special_inode_operations;
73 static const struct inode_operations btrfs_file_inode_operations;
74 static const struct address_space_operations btrfs_aops;
75 static const struct address_space_operations btrfs_symlink_aops;
76 static const struct file_operations btrfs_dir_file_operations;
77 static struct extent_io_ops btrfs_extent_io_ops;
79 static struct kmem_cache *btrfs_inode_cachep;
80 static struct kmem_cache *btrfs_delalloc_work_cachep;
81 struct kmem_cache *btrfs_trans_handle_cachep;
82 struct kmem_cache *btrfs_transaction_cachep;
83 struct kmem_cache *btrfs_path_cachep;
84 struct kmem_cache *btrfs_free_space_cachep;
87 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
88 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
89 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
90 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
91 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
92 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
93 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
94 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
97 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
98 static int btrfs_truncate(struct inode *inode);
99 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
100 static noinline int cow_file_range(struct inode *inode,
101 struct page *locked_page,
102 u64 start, u64 end, int *page_started,
103 unsigned long *nr_written, int unlock);
104 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
105 u64 len, u64 orig_start,
106 u64 block_start, u64 block_len,
107 u64 orig_block_len, u64 ram_bytes,
110 static int btrfs_dirty_inode(struct inode *inode);
112 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
113 void btrfs_test_inode_set_ops(struct inode *inode)
115 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
119 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
120 struct inode *inode, struct inode *dir,
121 const struct qstr *qstr)
125 err = btrfs_init_acl(trans, inode, dir);
127 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
132 * this does all the hard work for inserting an inline extent into
133 * the btree. The caller should have done a btrfs_drop_extents so that
134 * no overlapping inline items exist in the btree
136 static int insert_inline_extent(struct btrfs_trans_handle *trans,
137 struct btrfs_path *path, int extent_inserted,
138 struct btrfs_root *root, struct inode *inode,
139 u64 start, size_t size, size_t compressed_size,
141 struct page **compressed_pages)
143 struct extent_buffer *leaf;
144 struct page *page = NULL;
147 struct btrfs_file_extent_item *ei;
150 size_t cur_size = size;
151 unsigned long offset;
153 if (compressed_size && compressed_pages)
154 cur_size = compressed_size;
156 inode_add_bytes(inode, size);
158 if (!extent_inserted) {
159 struct btrfs_key key;
162 key.objectid = btrfs_ino(inode);
164 key.type = BTRFS_EXTENT_DATA_KEY;
166 datasize = btrfs_file_extent_calc_inline_size(cur_size);
167 path->leave_spinning = 1;
168 ret = btrfs_insert_empty_item(trans, root, path, &key,
175 leaf = path->nodes[0];
176 ei = btrfs_item_ptr(leaf, path->slots[0],
177 struct btrfs_file_extent_item);
178 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
179 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
180 btrfs_set_file_extent_encryption(leaf, ei, 0);
181 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
182 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
183 ptr = btrfs_file_extent_inline_start(ei);
185 if (compress_type != BTRFS_COMPRESS_NONE) {
188 while (compressed_size > 0) {
189 cpage = compressed_pages[i];
190 cur_size = min_t(unsigned long, compressed_size,
193 kaddr = kmap_atomic(cpage);
194 write_extent_buffer(leaf, kaddr, ptr, cur_size);
195 kunmap_atomic(kaddr);
199 compressed_size -= cur_size;
201 btrfs_set_file_extent_compression(leaf, ei,
204 page = find_get_page(inode->i_mapping,
205 start >> PAGE_CACHE_SHIFT);
206 btrfs_set_file_extent_compression(leaf, ei, 0);
207 kaddr = kmap_atomic(page);
208 offset = start & (PAGE_CACHE_SIZE - 1);
209 write_extent_buffer(leaf, kaddr + offset, ptr, size);
210 kunmap_atomic(kaddr);
211 page_cache_release(page);
213 btrfs_mark_buffer_dirty(leaf);
214 btrfs_release_path(path);
217 * we're an inline extent, so nobody can
218 * extend the file past i_size without locking
219 * a page we already have locked.
221 * We must do any isize and inode updates
222 * before we unlock the pages. Otherwise we
223 * could end up racing with unlink.
225 BTRFS_I(inode)->disk_i_size = inode->i_size;
226 ret = btrfs_update_inode(trans, root, inode);
235 * conditionally insert an inline extent into the file. This
236 * does the checks required to make sure the data is small enough
237 * to fit as an inline extent.
239 static noinline int cow_file_range_inline(struct btrfs_root *root,
240 struct inode *inode, u64 start,
241 u64 end, size_t compressed_size,
243 struct page **compressed_pages)
245 struct btrfs_trans_handle *trans;
246 u64 isize = i_size_read(inode);
247 u64 actual_end = min(end + 1, isize);
248 u64 inline_len = actual_end - start;
249 u64 aligned_end = ALIGN(end, root->sectorsize);
250 u64 data_len = inline_len;
252 struct btrfs_path *path;
253 int extent_inserted = 0;
254 u32 extent_item_size;
257 data_len = compressed_size;
260 actual_end > PAGE_CACHE_SIZE ||
261 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
263 (actual_end & (root->sectorsize - 1)) == 0) ||
265 data_len > root->fs_info->max_inline) {
269 path = btrfs_alloc_path();
273 trans = btrfs_join_transaction(root);
275 btrfs_free_path(path);
276 return PTR_ERR(trans);
278 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
280 if (compressed_size && compressed_pages)
281 extent_item_size = btrfs_file_extent_calc_inline_size(
284 extent_item_size = btrfs_file_extent_calc_inline_size(
287 ret = __btrfs_drop_extents(trans, root, inode, path,
288 start, aligned_end, NULL,
289 1, 1, extent_item_size, &extent_inserted);
291 btrfs_abort_transaction(trans, root, ret);
295 if (isize > actual_end)
296 inline_len = min_t(u64, isize, actual_end);
297 ret = insert_inline_extent(trans, path, extent_inserted,
299 inline_len, compressed_size,
300 compress_type, compressed_pages);
301 if (ret && ret != -ENOSPC) {
302 btrfs_abort_transaction(trans, root, ret);
304 } else if (ret == -ENOSPC) {
309 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
310 btrfs_delalloc_release_metadata(inode, end + 1 - start);
311 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
314 * Don't forget to free the reserved space, as for inlined extent
315 * it won't count as data extent, free them directly here.
316 * And at reserve time, it's always aligned to page size, so
317 * just free one page here.
319 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
320 btrfs_free_path(path);
321 btrfs_end_transaction(trans, root);
325 struct async_extent {
330 unsigned long nr_pages;
332 struct list_head list;
337 struct btrfs_root *root;
338 struct page *locked_page;
341 struct list_head extents;
342 struct btrfs_work work;
345 static noinline int add_async_extent(struct async_cow *cow,
346 u64 start, u64 ram_size,
349 unsigned long nr_pages,
352 struct async_extent *async_extent;
354 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
355 BUG_ON(!async_extent); /* -ENOMEM */
356 async_extent->start = start;
357 async_extent->ram_size = ram_size;
358 async_extent->compressed_size = compressed_size;
359 async_extent->pages = pages;
360 async_extent->nr_pages = nr_pages;
361 async_extent->compress_type = compress_type;
362 list_add_tail(&async_extent->list, &cow->extents);
366 static inline int inode_need_compress(struct inode *inode)
368 struct btrfs_root *root = BTRFS_I(inode)->root;
371 if (btrfs_test_opt(root, FORCE_COMPRESS))
373 /* bad compression ratios */
374 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
376 if (btrfs_test_opt(root, COMPRESS) ||
377 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
378 BTRFS_I(inode)->force_compress)
384 * we create compressed extents in two phases. The first
385 * phase compresses a range of pages that have already been
386 * locked (both pages and state bits are locked).
388 * This is done inside an ordered work queue, and the compression
389 * is spread across many cpus. The actual IO submission is step
390 * two, and the ordered work queue takes care of making sure that
391 * happens in the same order things were put onto the queue by
392 * writepages and friends.
394 * If this code finds it can't get good compression, it puts an
395 * entry onto the work queue to write the uncompressed bytes. This
396 * makes sure that both compressed inodes and uncompressed inodes
397 * are written in the same order that the flusher thread sent them
400 static noinline void compress_file_range(struct inode *inode,
401 struct page *locked_page,
403 struct async_cow *async_cow,
406 struct btrfs_root *root = BTRFS_I(inode)->root;
408 u64 blocksize = root->sectorsize;
410 u64 isize = i_size_read(inode);
412 struct page **pages = NULL;
413 unsigned long nr_pages;
414 unsigned long nr_pages_ret = 0;
415 unsigned long total_compressed = 0;
416 unsigned long total_in = 0;
417 unsigned long max_compressed = 128 * 1024;
418 unsigned long max_uncompressed = 128 * 1024;
421 int compress_type = root->fs_info->compress_type;
424 /* if this is a small write inside eof, kick off a defrag */
425 if ((end - start + 1) < 16 * 1024 &&
426 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
427 btrfs_add_inode_defrag(NULL, inode);
429 actual_end = min_t(u64, isize, end + 1);
432 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
433 nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
436 * we don't want to send crud past the end of i_size through
437 * compression, that's just a waste of CPU time. So, if the
438 * end of the file is before the start of our current
439 * requested range of bytes, we bail out to the uncompressed
440 * cleanup code that can deal with all of this.
442 * It isn't really the fastest way to fix things, but this is a
443 * very uncommon corner.
445 if (actual_end <= start)
446 goto cleanup_and_bail_uncompressed;
448 total_compressed = actual_end - start;
451 * skip compression for a small file range(<=blocksize) that
452 * isn't an inline extent, since it dosen't save disk space at all.
454 if (total_compressed <= blocksize &&
455 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
456 goto cleanup_and_bail_uncompressed;
458 /* we want to make sure that amount of ram required to uncompress
459 * an extent is reasonable, so we limit the total size in ram
460 * of a compressed extent to 128k. This is a crucial number
461 * because it also controls how easily we can spread reads across
462 * cpus for decompression.
464 * We also want to make sure the amount of IO required to do
465 * a random read is reasonably small, so we limit the size of
466 * a compressed extent to 128k.
468 total_compressed = min(total_compressed, max_uncompressed);
469 num_bytes = ALIGN(end - start + 1, blocksize);
470 num_bytes = max(blocksize, num_bytes);
475 * we do compression for mount -o compress and when the
476 * inode has not been flagged as nocompress. This flag can
477 * change at any time if we discover bad compression ratios.
479 if (inode_need_compress(inode)) {
481 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
483 /* just bail out to the uncompressed code */
487 if (BTRFS_I(inode)->force_compress)
488 compress_type = BTRFS_I(inode)->force_compress;
491 * we need to call clear_page_dirty_for_io on each
492 * page in the range. Otherwise applications with the file
493 * mmap'd can wander in and change the page contents while
494 * we are compressing them.
496 * If the compression fails for any reason, we set the pages
497 * dirty again later on.
499 extent_range_clear_dirty_for_io(inode, start, end);
501 ret = btrfs_compress_pages(compress_type,
502 inode->i_mapping, start,
503 total_compressed, pages,
504 nr_pages, &nr_pages_ret,
510 unsigned long offset = total_compressed &
511 (PAGE_CACHE_SIZE - 1);
512 struct page *page = pages[nr_pages_ret - 1];
515 /* zero the tail end of the last page, we might be
516 * sending it down to disk
519 kaddr = kmap_atomic(page);
520 memset(kaddr + offset, 0,
521 PAGE_CACHE_SIZE - offset);
522 kunmap_atomic(kaddr);
529 /* lets try to make an inline extent */
530 if (ret || total_in < (actual_end - start)) {
531 /* we didn't compress the entire range, try
532 * to make an uncompressed inline extent.
534 ret = cow_file_range_inline(root, inode, start, end,
537 /* try making a compressed inline extent */
538 ret = cow_file_range_inline(root, inode, start, end,
540 compress_type, pages);
543 unsigned long clear_flags = EXTENT_DELALLOC |
545 unsigned long page_error_op;
547 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
548 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
551 * inline extent creation worked or returned error,
552 * we don't need to create any more async work items.
553 * Unlock and free up our temp pages.
555 extent_clear_unlock_delalloc(inode, start, end, NULL,
556 clear_flags, PAGE_UNLOCK |
567 * we aren't doing an inline extent round the compressed size
568 * up to a block size boundary so the allocator does sane
571 total_compressed = ALIGN(total_compressed, blocksize);
574 * one last check to make sure the compression is really a
575 * win, compare the page count read with the blocks on disk
577 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
578 if (total_compressed >= total_in) {
581 num_bytes = total_in;
584 if (!will_compress && pages) {
586 * the compression code ran but failed to make things smaller,
587 * free any pages it allocated and our page pointer array
589 for (i = 0; i < nr_pages_ret; i++) {
590 WARN_ON(pages[i]->mapping);
591 page_cache_release(pages[i]);
595 total_compressed = 0;
598 /* flag the file so we don't compress in the future */
599 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
600 !(BTRFS_I(inode)->force_compress)) {
601 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
607 /* the async work queues will take care of doing actual
608 * allocation on disk for these compressed pages,
609 * and will submit them to the elevator.
611 add_async_extent(async_cow, start, num_bytes,
612 total_compressed, pages, nr_pages_ret,
615 if (start + num_bytes < end) {
622 cleanup_and_bail_uncompressed:
624 * No compression, but we still need to write the pages in
625 * the file we've been given so far. redirty the locked
626 * page if it corresponds to our extent and set things up
627 * for the async work queue to run cow_file_range to do
628 * the normal delalloc dance
630 if (page_offset(locked_page) >= start &&
631 page_offset(locked_page) <= end) {
632 __set_page_dirty_nobuffers(locked_page);
633 /* unlocked later on in the async handlers */
636 extent_range_redirty_for_io(inode, start, end);
637 add_async_extent(async_cow, start, end - start + 1,
638 0, NULL, 0, BTRFS_COMPRESS_NONE);
645 for (i = 0; i < nr_pages_ret; i++) {
646 WARN_ON(pages[i]->mapping);
647 page_cache_release(pages[i]);
652 static void free_async_extent_pages(struct async_extent *async_extent)
656 if (!async_extent->pages)
659 for (i = 0; i < async_extent->nr_pages; i++) {
660 WARN_ON(async_extent->pages[i]->mapping);
661 page_cache_release(async_extent->pages[i]);
663 kfree(async_extent->pages);
664 async_extent->nr_pages = 0;
665 async_extent->pages = NULL;
669 * phase two of compressed writeback. This is the ordered portion
670 * of the code, which only gets called in the order the work was
671 * queued. We walk all the async extents created by compress_file_range
672 * and send them down to the disk.
674 static noinline void submit_compressed_extents(struct inode *inode,
675 struct async_cow *async_cow)
677 struct async_extent *async_extent;
679 struct btrfs_key ins;
680 struct extent_map *em;
681 struct btrfs_root *root = BTRFS_I(inode)->root;
682 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
683 struct extent_io_tree *io_tree;
687 while (!list_empty(&async_cow->extents)) {
688 async_extent = list_entry(async_cow->extents.next,
689 struct async_extent, list);
690 list_del(&async_extent->list);
692 io_tree = &BTRFS_I(inode)->io_tree;
695 /* did the compression code fall back to uncompressed IO? */
696 if (!async_extent->pages) {
697 int page_started = 0;
698 unsigned long nr_written = 0;
700 lock_extent(io_tree, async_extent->start,
701 async_extent->start +
702 async_extent->ram_size - 1);
704 /* allocate blocks */
705 ret = cow_file_range(inode, async_cow->locked_page,
707 async_extent->start +
708 async_extent->ram_size - 1,
709 &page_started, &nr_written, 0);
714 * if page_started, cow_file_range inserted an
715 * inline extent and took care of all the unlocking
716 * and IO for us. Otherwise, we need to submit
717 * all those pages down to the drive.
719 if (!page_started && !ret)
720 extent_write_locked_range(io_tree,
721 inode, async_extent->start,
722 async_extent->start +
723 async_extent->ram_size - 1,
727 unlock_page(async_cow->locked_page);
733 lock_extent(io_tree, async_extent->start,
734 async_extent->start + async_extent->ram_size - 1);
736 ret = btrfs_reserve_extent(root,
737 async_extent->compressed_size,
738 async_extent->compressed_size,
739 0, alloc_hint, &ins, 1, 1);
741 free_async_extent_pages(async_extent);
743 if (ret == -ENOSPC) {
744 unlock_extent(io_tree, async_extent->start,
745 async_extent->start +
746 async_extent->ram_size - 1);
749 * we need to redirty the pages if we decide to
750 * fallback to uncompressed IO, otherwise we
751 * will not submit these pages down to lower
754 extent_range_redirty_for_io(inode,
756 async_extent->start +
757 async_extent->ram_size - 1);
764 * here we're doing allocation and writeback of the
767 btrfs_drop_extent_cache(inode, async_extent->start,
768 async_extent->start +
769 async_extent->ram_size - 1, 0);
771 em = alloc_extent_map();
774 goto out_free_reserve;
776 em->start = async_extent->start;
777 em->len = async_extent->ram_size;
778 em->orig_start = em->start;
779 em->mod_start = em->start;
780 em->mod_len = em->len;
782 em->block_start = ins.objectid;
783 em->block_len = ins.offset;
784 em->orig_block_len = ins.offset;
785 em->ram_bytes = async_extent->ram_size;
786 em->bdev = root->fs_info->fs_devices->latest_bdev;
787 em->compress_type = async_extent->compress_type;
788 set_bit(EXTENT_FLAG_PINNED, &em->flags);
789 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
793 write_lock(&em_tree->lock);
794 ret = add_extent_mapping(em_tree, em, 1);
795 write_unlock(&em_tree->lock);
796 if (ret != -EEXIST) {
800 btrfs_drop_extent_cache(inode, async_extent->start,
801 async_extent->start +
802 async_extent->ram_size - 1, 0);
806 goto out_free_reserve;
808 ret = btrfs_add_ordered_extent_compress(inode,
811 async_extent->ram_size,
813 BTRFS_ORDERED_COMPRESSED,
814 async_extent->compress_type);
816 btrfs_drop_extent_cache(inode, async_extent->start,
817 async_extent->start +
818 async_extent->ram_size - 1, 0);
819 goto out_free_reserve;
823 * clear dirty, set writeback and unlock the pages.
825 extent_clear_unlock_delalloc(inode, async_extent->start,
826 async_extent->start +
827 async_extent->ram_size - 1,
828 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
829 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
831 ret = btrfs_submit_compressed_write(inode,
833 async_extent->ram_size,
835 ins.offset, async_extent->pages,
836 async_extent->nr_pages);
838 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
839 struct page *p = async_extent->pages[0];
840 const u64 start = async_extent->start;
841 const u64 end = start + async_extent->ram_size - 1;
843 p->mapping = inode->i_mapping;
844 tree->ops->writepage_end_io_hook(p, start, end,
847 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
850 free_async_extent_pages(async_extent);
852 alloc_hint = ins.objectid + ins.offset;
858 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
860 extent_clear_unlock_delalloc(inode, async_extent->start,
861 async_extent->start +
862 async_extent->ram_size - 1,
863 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
864 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
865 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
866 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
868 free_async_extent_pages(async_extent);
873 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
876 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
877 struct extent_map *em;
880 read_lock(&em_tree->lock);
881 em = search_extent_mapping(em_tree, start, num_bytes);
884 * if block start isn't an actual block number then find the
885 * first block in this inode and use that as a hint. If that
886 * block is also bogus then just don't worry about it.
888 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
890 em = search_extent_mapping(em_tree, 0, 0);
891 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
892 alloc_hint = em->block_start;
896 alloc_hint = em->block_start;
900 read_unlock(&em_tree->lock);
906 * when extent_io.c finds a delayed allocation range in the file,
907 * the call backs end up in this code. The basic idea is to
908 * allocate extents on disk for the range, and create ordered data structs
909 * in ram to track those extents.
911 * locked_page is the page that writepage had locked already. We use
912 * it to make sure we don't do extra locks or unlocks.
914 * *page_started is set to one if we unlock locked_page and do everything
915 * required to start IO on it. It may be clean and already done with
918 static noinline int cow_file_range(struct inode *inode,
919 struct page *locked_page,
920 u64 start, u64 end, int *page_started,
921 unsigned long *nr_written,
924 struct btrfs_root *root = BTRFS_I(inode)->root;
927 unsigned long ram_size;
930 u64 blocksize = root->sectorsize;
931 struct btrfs_key ins;
932 struct extent_map *em;
933 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
936 if (btrfs_is_free_space_inode(inode)) {
942 num_bytes = ALIGN(end - start + 1, blocksize);
943 num_bytes = max(blocksize, num_bytes);
944 disk_num_bytes = num_bytes;
946 /* if this is a small write inside eof, kick off defrag */
947 if (num_bytes < 64 * 1024 &&
948 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
949 btrfs_add_inode_defrag(NULL, inode);
952 /* lets try to make an inline extent */
953 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
956 extent_clear_unlock_delalloc(inode, start, end, NULL,
957 EXTENT_LOCKED | EXTENT_DELALLOC |
958 EXTENT_DEFRAG, PAGE_UNLOCK |
959 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
962 *nr_written = *nr_written +
963 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
966 } else if (ret < 0) {
971 BUG_ON(disk_num_bytes >
972 btrfs_super_total_bytes(root->fs_info->super_copy));
974 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
975 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
977 while (disk_num_bytes > 0) {
980 cur_alloc_size = disk_num_bytes;
981 ret = btrfs_reserve_extent(root, cur_alloc_size,
982 root->sectorsize, 0, alloc_hint,
987 em = alloc_extent_map();
993 em->orig_start = em->start;
994 ram_size = ins.offset;
995 em->len = ins.offset;
996 em->mod_start = em->start;
997 em->mod_len = em->len;
999 em->block_start = ins.objectid;
1000 em->block_len = ins.offset;
1001 em->orig_block_len = ins.offset;
1002 em->ram_bytes = ram_size;
1003 em->bdev = root->fs_info->fs_devices->latest_bdev;
1004 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1005 em->generation = -1;
1008 write_lock(&em_tree->lock);
1009 ret = add_extent_mapping(em_tree, em, 1);
1010 write_unlock(&em_tree->lock);
1011 if (ret != -EEXIST) {
1012 free_extent_map(em);
1015 btrfs_drop_extent_cache(inode, start,
1016 start + ram_size - 1, 0);
1021 cur_alloc_size = ins.offset;
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1025 goto out_drop_extent_cache;
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1032 goto out_drop_extent_cache;
1035 if (disk_num_bytes < cur_alloc_size)
1038 /* we're not doing compressed IO, don't unlock the first
1039 * page (which the caller expects to stay locked), don't
1040 * clear any dirty bits and don't set any writeback bits
1042 * Do set the Private2 bit so we know this page was properly
1043 * setup for writepage
1045 op = unlock ? PAGE_UNLOCK : 0;
1046 op |= PAGE_SET_PRIVATE2;
1048 extent_clear_unlock_delalloc(inode, start,
1049 start + ram_size - 1, locked_page,
1050 EXTENT_LOCKED | EXTENT_DELALLOC,
1052 disk_num_bytes -= cur_alloc_size;
1053 num_bytes -= cur_alloc_size;
1054 alloc_hint = ins.objectid + ins.offset;
1055 start += cur_alloc_size;
1060 out_drop_extent_cache:
1061 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1063 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1065 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1066 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1067 EXTENT_DELALLOC | EXTENT_DEFRAG,
1068 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1069 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1074 * work queue call back to started compression on a file and pages
1076 static noinline void async_cow_start(struct btrfs_work *work)
1078 struct async_cow *async_cow;
1080 async_cow = container_of(work, struct async_cow, work);
1082 compress_file_range(async_cow->inode, async_cow->locked_page,
1083 async_cow->start, async_cow->end, async_cow,
1085 if (num_added == 0) {
1086 btrfs_add_delayed_iput(async_cow->inode);
1087 async_cow->inode = NULL;
1092 * work queue call back to submit previously compressed pages
1094 static noinline void async_cow_submit(struct btrfs_work *work)
1096 struct async_cow *async_cow;
1097 struct btrfs_root *root;
1098 unsigned long nr_pages;
1100 async_cow = container_of(work, struct async_cow, work);
1102 root = async_cow->root;
1103 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1107 * atomic_sub_return implies a barrier for waitqueue_active
1109 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1111 waitqueue_active(&root->fs_info->async_submit_wait))
1112 wake_up(&root->fs_info->async_submit_wait);
1114 if (async_cow->inode)
1115 submit_compressed_extents(async_cow->inode, async_cow);
1118 static noinline void async_cow_free(struct btrfs_work *work)
1120 struct async_cow *async_cow;
1121 async_cow = container_of(work, struct async_cow, work);
1122 if (async_cow->inode)
1123 btrfs_add_delayed_iput(async_cow->inode);
1127 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1128 u64 start, u64 end, int *page_started,
1129 unsigned long *nr_written)
1131 struct async_cow *async_cow;
1132 struct btrfs_root *root = BTRFS_I(inode)->root;
1133 unsigned long nr_pages;
1135 int limit = 10 * 1024 * 1024;
1137 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1138 1, 0, NULL, GFP_NOFS);
1139 while (start < end) {
1140 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1141 BUG_ON(!async_cow); /* -ENOMEM */
1142 async_cow->inode = igrab(inode);
1143 async_cow->root = root;
1144 async_cow->locked_page = locked_page;
1145 async_cow->start = start;
1147 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1148 !btrfs_test_opt(root, FORCE_COMPRESS))
1151 cur_end = min(end, start + 512 * 1024 - 1);
1153 async_cow->end = cur_end;
1154 INIT_LIST_HEAD(&async_cow->extents);
1156 btrfs_init_work(&async_cow->work,
1157 btrfs_delalloc_helper,
1158 async_cow_start, async_cow_submit,
1161 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1163 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1165 btrfs_queue_work(root->fs_info->delalloc_workers,
1168 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1169 wait_event(root->fs_info->async_submit_wait,
1170 (atomic_read(&root->fs_info->async_delalloc_pages) <
1174 while (atomic_read(&root->fs_info->async_submit_draining) &&
1175 atomic_read(&root->fs_info->async_delalloc_pages)) {
1176 wait_event(root->fs_info->async_submit_wait,
1177 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1181 *nr_written += nr_pages;
1182 start = cur_end + 1;
1188 static noinline int csum_exist_in_range(struct btrfs_root *root,
1189 u64 bytenr, u64 num_bytes)
1192 struct btrfs_ordered_sum *sums;
1195 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1196 bytenr + num_bytes - 1, &list, 0);
1197 if (ret == 0 && list_empty(&list))
1200 while (!list_empty(&list)) {
1201 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1202 list_del(&sums->list);
1209 * when nowcow writeback call back. This checks for snapshots or COW copies
1210 * of the extents that exist in the file, and COWs the file as required.
1212 * If no cow copies or snapshots exist, we write directly to the existing
1215 static noinline int run_delalloc_nocow(struct inode *inode,
1216 struct page *locked_page,
1217 u64 start, u64 end, int *page_started, int force,
1218 unsigned long *nr_written)
1220 struct btrfs_root *root = BTRFS_I(inode)->root;
1221 struct btrfs_trans_handle *trans;
1222 struct extent_buffer *leaf;
1223 struct btrfs_path *path;
1224 struct btrfs_file_extent_item *fi;
1225 struct btrfs_key found_key;
1240 u64 ino = btrfs_ino(inode);
1242 path = btrfs_alloc_path();
1244 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1245 EXTENT_LOCKED | EXTENT_DELALLOC |
1246 EXTENT_DO_ACCOUNTING |
1247 EXTENT_DEFRAG, PAGE_UNLOCK |
1249 PAGE_SET_WRITEBACK |
1250 PAGE_END_WRITEBACK);
1254 nolock = btrfs_is_free_space_inode(inode);
1257 trans = btrfs_join_transaction_nolock(root);
1259 trans = btrfs_join_transaction(root);
1261 if (IS_ERR(trans)) {
1262 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1263 EXTENT_LOCKED | EXTENT_DELALLOC |
1264 EXTENT_DO_ACCOUNTING |
1265 EXTENT_DEFRAG, PAGE_UNLOCK |
1267 PAGE_SET_WRITEBACK |
1268 PAGE_END_WRITEBACK);
1269 btrfs_free_path(path);
1270 return PTR_ERR(trans);
1273 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1275 cow_start = (u64)-1;
1278 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1282 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1283 leaf = path->nodes[0];
1284 btrfs_item_key_to_cpu(leaf, &found_key,
1285 path->slots[0] - 1);
1286 if (found_key.objectid == ino &&
1287 found_key.type == BTRFS_EXTENT_DATA_KEY)
1292 leaf = path->nodes[0];
1293 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1294 ret = btrfs_next_leaf(root, path);
1299 leaf = path->nodes[0];
1305 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1307 if (found_key.objectid > ino ||
1308 found_key.type > BTRFS_EXTENT_DATA_KEY ||
1309 found_key.offset > end)
1312 if (found_key.offset > cur_offset) {
1313 extent_end = found_key.offset;
1318 fi = btrfs_item_ptr(leaf, path->slots[0],
1319 struct btrfs_file_extent_item);
1320 extent_type = btrfs_file_extent_type(leaf, fi);
1322 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1323 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1324 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1325 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1326 extent_offset = btrfs_file_extent_offset(leaf, fi);
1327 extent_end = found_key.offset +
1328 btrfs_file_extent_num_bytes(leaf, fi);
1330 btrfs_file_extent_disk_num_bytes(leaf, fi);
1331 if (extent_end <= start) {
1335 if (disk_bytenr == 0)
1337 if (btrfs_file_extent_compression(leaf, fi) ||
1338 btrfs_file_extent_encryption(leaf, fi) ||
1339 btrfs_file_extent_other_encoding(leaf, fi))
1341 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1343 if (btrfs_extent_readonly(root, disk_bytenr))
1345 if (btrfs_cross_ref_exist(trans, root, ino,
1347 extent_offset, disk_bytenr))
1349 disk_bytenr += extent_offset;
1350 disk_bytenr += cur_offset - found_key.offset;
1351 num_bytes = min(end + 1, extent_end) - cur_offset;
1353 * if there are pending snapshots for this root,
1354 * we fall into common COW way.
1357 err = btrfs_start_write_no_snapshoting(root);
1362 * force cow if csum exists in the range.
1363 * this ensure that csum for a given extent are
1364 * either valid or do not exist.
1366 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1369 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1370 extent_end = found_key.offset +
1371 btrfs_file_extent_inline_len(leaf,
1372 path->slots[0], fi);
1373 extent_end = ALIGN(extent_end, root->sectorsize);
1378 if (extent_end <= start) {
1380 if (!nolock && nocow)
1381 btrfs_end_write_no_snapshoting(root);
1385 if (cow_start == (u64)-1)
1386 cow_start = cur_offset;
1387 cur_offset = extent_end;
1388 if (cur_offset > end)
1394 btrfs_release_path(path);
1395 if (cow_start != (u64)-1) {
1396 ret = cow_file_range(inode, locked_page,
1397 cow_start, found_key.offset - 1,
1398 page_started, nr_written, 1);
1400 if (!nolock && nocow)
1401 btrfs_end_write_no_snapshoting(root);
1404 cow_start = (u64)-1;
1407 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1408 struct extent_map *em;
1409 struct extent_map_tree *em_tree;
1410 em_tree = &BTRFS_I(inode)->extent_tree;
1411 em = alloc_extent_map();
1412 BUG_ON(!em); /* -ENOMEM */
1413 em->start = cur_offset;
1414 em->orig_start = found_key.offset - extent_offset;
1415 em->len = num_bytes;
1416 em->block_len = num_bytes;
1417 em->block_start = disk_bytenr;
1418 em->orig_block_len = disk_num_bytes;
1419 em->ram_bytes = ram_bytes;
1420 em->bdev = root->fs_info->fs_devices->latest_bdev;
1421 em->mod_start = em->start;
1422 em->mod_len = em->len;
1423 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1424 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1425 em->generation = -1;
1427 write_lock(&em_tree->lock);
1428 ret = add_extent_mapping(em_tree, em, 1);
1429 write_unlock(&em_tree->lock);
1430 if (ret != -EEXIST) {
1431 free_extent_map(em);
1434 btrfs_drop_extent_cache(inode, em->start,
1435 em->start + em->len - 1, 0);
1437 type = BTRFS_ORDERED_PREALLOC;
1439 type = BTRFS_ORDERED_NOCOW;
1442 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1443 num_bytes, num_bytes, type);
1444 BUG_ON(ret); /* -ENOMEM */
1446 if (root->root_key.objectid ==
1447 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1448 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1451 if (!nolock && nocow)
1452 btrfs_end_write_no_snapshoting(root);
1457 extent_clear_unlock_delalloc(inode, cur_offset,
1458 cur_offset + num_bytes - 1,
1459 locked_page, EXTENT_LOCKED |
1460 EXTENT_DELALLOC, PAGE_UNLOCK |
1462 if (!nolock && nocow)
1463 btrfs_end_write_no_snapshoting(root);
1464 cur_offset = extent_end;
1465 if (cur_offset > end)
1468 btrfs_release_path(path);
1470 if (cur_offset <= end && cow_start == (u64)-1) {
1471 cow_start = cur_offset;
1475 if (cow_start != (u64)-1) {
1476 ret = cow_file_range(inode, locked_page, cow_start, end,
1477 page_started, nr_written, 1);
1483 err = btrfs_end_transaction(trans, root);
1487 if (ret && cur_offset < end)
1488 extent_clear_unlock_delalloc(inode, cur_offset, end,
1489 locked_page, EXTENT_LOCKED |
1490 EXTENT_DELALLOC | EXTENT_DEFRAG |
1491 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1493 PAGE_SET_WRITEBACK |
1494 PAGE_END_WRITEBACK);
1495 btrfs_free_path(path);
1499 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1502 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1503 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1507 * @defrag_bytes is a hint value, no spinlock held here,
1508 * if is not zero, it means the file is defragging.
1509 * Force cow if given extent needs to be defragged.
1511 if (BTRFS_I(inode)->defrag_bytes &&
1512 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1513 EXTENT_DEFRAG, 0, NULL))
1520 * extent_io.c call back to do delayed allocation processing
1522 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1523 u64 start, u64 end, int *page_started,
1524 unsigned long *nr_written)
1527 int force_cow = need_force_cow(inode, start, end);
1529 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1530 ret = run_delalloc_nocow(inode, locked_page, start, end,
1531 page_started, 1, nr_written);
1532 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1533 ret = run_delalloc_nocow(inode, locked_page, start, end,
1534 page_started, 0, nr_written);
1535 } else if (!inode_need_compress(inode)) {
1536 ret = cow_file_range(inode, locked_page, start, end,
1537 page_started, nr_written, 1);
1539 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1540 &BTRFS_I(inode)->runtime_flags);
1541 ret = cow_file_range_async(inode, locked_page, start, end,
1542 page_started, nr_written);
1547 static void btrfs_split_extent_hook(struct inode *inode,
1548 struct extent_state *orig, u64 split)
1552 /* not delalloc, ignore it */
1553 if (!(orig->state & EXTENT_DELALLOC))
1556 size = orig->end - orig->start + 1;
1557 if (size > BTRFS_MAX_EXTENT_SIZE) {
1562 * See the explanation in btrfs_merge_extent_hook, the same
1563 * applies here, just in reverse.
1565 new_size = orig->end - split + 1;
1566 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1567 BTRFS_MAX_EXTENT_SIZE);
1568 new_size = split - orig->start;
1569 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1570 BTRFS_MAX_EXTENT_SIZE);
1571 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1572 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1576 spin_lock(&BTRFS_I(inode)->lock);
1577 BTRFS_I(inode)->outstanding_extents++;
1578 spin_unlock(&BTRFS_I(inode)->lock);
1582 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1583 * extents so we can keep track of new extents that are just merged onto old
1584 * extents, such as when we are doing sequential writes, so we can properly
1585 * account for the metadata space we'll need.
1587 static void btrfs_merge_extent_hook(struct inode *inode,
1588 struct extent_state *new,
1589 struct extent_state *other)
1591 u64 new_size, old_size;
1594 /* not delalloc, ignore it */
1595 if (!(other->state & EXTENT_DELALLOC))
1598 if (new->start > other->start)
1599 new_size = new->end - other->start + 1;
1601 new_size = other->end - new->start + 1;
1603 /* we're not bigger than the max, unreserve the space and go */
1604 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1605 spin_lock(&BTRFS_I(inode)->lock);
1606 BTRFS_I(inode)->outstanding_extents--;
1607 spin_unlock(&BTRFS_I(inode)->lock);
1612 * We have to add up either side to figure out how many extents were
1613 * accounted for before we merged into one big extent. If the number of
1614 * extents we accounted for is <= the amount we need for the new range
1615 * then we can return, otherwise drop. Think of it like this
1619 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1620 * need 2 outstanding extents, on one side we have 1 and the other side
1621 * we have 1 so they are == and we can return. But in this case
1623 * [MAX_SIZE+4k][MAX_SIZE+4k]
1625 * Each range on their own accounts for 2 extents, but merged together
1626 * they are only 3 extents worth of accounting, so we need to drop in
1629 old_size = other->end - other->start + 1;
1630 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1631 BTRFS_MAX_EXTENT_SIZE);
1632 old_size = new->end - new->start + 1;
1633 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1634 BTRFS_MAX_EXTENT_SIZE);
1636 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1637 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1640 spin_lock(&BTRFS_I(inode)->lock);
1641 BTRFS_I(inode)->outstanding_extents--;
1642 spin_unlock(&BTRFS_I(inode)->lock);
1645 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1646 struct inode *inode)
1648 spin_lock(&root->delalloc_lock);
1649 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1650 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1651 &root->delalloc_inodes);
1652 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1653 &BTRFS_I(inode)->runtime_flags);
1654 root->nr_delalloc_inodes++;
1655 if (root->nr_delalloc_inodes == 1) {
1656 spin_lock(&root->fs_info->delalloc_root_lock);
1657 BUG_ON(!list_empty(&root->delalloc_root));
1658 list_add_tail(&root->delalloc_root,
1659 &root->fs_info->delalloc_roots);
1660 spin_unlock(&root->fs_info->delalloc_root_lock);
1663 spin_unlock(&root->delalloc_lock);
1666 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1667 struct inode *inode)
1669 spin_lock(&root->delalloc_lock);
1670 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1671 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1672 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1673 &BTRFS_I(inode)->runtime_flags);
1674 root->nr_delalloc_inodes--;
1675 if (!root->nr_delalloc_inodes) {
1676 spin_lock(&root->fs_info->delalloc_root_lock);
1677 BUG_ON(list_empty(&root->delalloc_root));
1678 list_del_init(&root->delalloc_root);
1679 spin_unlock(&root->fs_info->delalloc_root_lock);
1682 spin_unlock(&root->delalloc_lock);
1686 * extent_io.c set_bit_hook, used to track delayed allocation
1687 * bytes in this file, and to maintain the list of inodes that
1688 * have pending delalloc work to be done.
1690 static void btrfs_set_bit_hook(struct inode *inode,
1691 struct extent_state *state, unsigned *bits)
1694 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1697 * set_bit and clear bit hooks normally require _irqsave/restore
1698 * but in this case, we are only testing for the DELALLOC
1699 * bit, which is only set or cleared with irqs on
1701 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1702 struct btrfs_root *root = BTRFS_I(inode)->root;
1703 u64 len = state->end + 1 - state->start;
1704 bool do_list = !btrfs_is_free_space_inode(inode);
1706 if (*bits & EXTENT_FIRST_DELALLOC) {
1707 *bits &= ~EXTENT_FIRST_DELALLOC;
1709 spin_lock(&BTRFS_I(inode)->lock);
1710 BTRFS_I(inode)->outstanding_extents++;
1711 spin_unlock(&BTRFS_I(inode)->lock);
1714 /* For sanity tests */
1715 if (btrfs_test_is_dummy_root(root))
1718 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1719 root->fs_info->delalloc_batch);
1720 spin_lock(&BTRFS_I(inode)->lock);
1721 BTRFS_I(inode)->delalloc_bytes += len;
1722 if (*bits & EXTENT_DEFRAG)
1723 BTRFS_I(inode)->defrag_bytes += len;
1724 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1725 &BTRFS_I(inode)->runtime_flags))
1726 btrfs_add_delalloc_inodes(root, inode);
1727 spin_unlock(&BTRFS_I(inode)->lock);
1732 * extent_io.c clear_bit_hook, see set_bit_hook for why
1734 static void btrfs_clear_bit_hook(struct inode *inode,
1735 struct extent_state *state,
1738 u64 len = state->end + 1 - state->start;
1739 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1740 BTRFS_MAX_EXTENT_SIZE);
1742 spin_lock(&BTRFS_I(inode)->lock);
1743 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1744 BTRFS_I(inode)->defrag_bytes -= len;
1745 spin_unlock(&BTRFS_I(inode)->lock);
1748 * set_bit and clear bit hooks normally require _irqsave/restore
1749 * but in this case, we are only testing for the DELALLOC
1750 * bit, which is only set or cleared with irqs on
1752 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1753 struct btrfs_root *root = BTRFS_I(inode)->root;
1754 bool do_list = !btrfs_is_free_space_inode(inode);
1756 if (*bits & EXTENT_FIRST_DELALLOC) {
1757 *bits &= ~EXTENT_FIRST_DELALLOC;
1758 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1759 spin_lock(&BTRFS_I(inode)->lock);
1760 BTRFS_I(inode)->outstanding_extents -= num_extents;
1761 spin_unlock(&BTRFS_I(inode)->lock);
1765 * We don't reserve metadata space for space cache inodes so we
1766 * don't need to call dellalloc_release_metadata if there is an
1769 if (*bits & EXTENT_DO_ACCOUNTING &&
1770 root != root->fs_info->tree_root)
1771 btrfs_delalloc_release_metadata(inode, len);
1773 /* For sanity tests. */
1774 if (btrfs_test_is_dummy_root(root))
1777 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1778 && do_list && !(state->state & EXTENT_NORESERVE))
1779 btrfs_free_reserved_data_space_noquota(inode,
1782 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1783 root->fs_info->delalloc_batch);
1784 spin_lock(&BTRFS_I(inode)->lock);
1785 BTRFS_I(inode)->delalloc_bytes -= len;
1786 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1787 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1788 &BTRFS_I(inode)->runtime_flags))
1789 btrfs_del_delalloc_inode(root, inode);
1790 spin_unlock(&BTRFS_I(inode)->lock);
1795 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1796 * we don't create bios that span stripes or chunks
1798 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1799 size_t size, struct bio *bio,
1800 unsigned long bio_flags)
1802 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1803 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1808 if (bio_flags & EXTENT_BIO_COMPRESSED)
1811 length = bio->bi_iter.bi_size;
1812 map_length = length;
1813 ret = btrfs_map_block(root->fs_info, rw, logical,
1814 &map_length, NULL, 0);
1815 /* Will always return 0 with map_multi == NULL */
1817 if (map_length < length + size)
1823 * in order to insert checksums into the metadata in large chunks,
1824 * we wait until bio submission time. All the pages in the bio are
1825 * checksummed and sums are attached onto the ordered extent record.
1827 * At IO completion time the cums attached on the ordered extent record
1828 * are inserted into the btree
1830 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1831 struct bio *bio, int mirror_num,
1832 unsigned long bio_flags,
1835 struct btrfs_root *root = BTRFS_I(inode)->root;
1838 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1839 BUG_ON(ret); /* -ENOMEM */
1844 * in order to insert checksums into the metadata in large chunks,
1845 * we wait until bio submission time. All the pages in the bio are
1846 * checksummed and sums are attached onto the ordered extent record.
1848 * At IO completion time the cums attached on the ordered extent record
1849 * are inserted into the btree
1851 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1852 int mirror_num, unsigned long bio_flags,
1855 struct btrfs_root *root = BTRFS_I(inode)->root;
1858 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1860 bio->bi_error = ret;
1867 * extent_io.c submission hook. This does the right thing for csum calculation
1868 * on write, or reading the csums from the tree before a read
1870 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1871 int mirror_num, unsigned long bio_flags,
1874 struct btrfs_root *root = BTRFS_I(inode)->root;
1875 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1878 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1880 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1882 if (btrfs_is_free_space_inode(inode))
1883 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1885 if (!(rw & REQ_WRITE)) {
1886 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1890 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1891 ret = btrfs_submit_compressed_read(inode, bio,
1895 } else if (!skip_sum) {
1896 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1901 } else if (async && !skip_sum) {
1902 /* csum items have already been cloned */
1903 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1905 /* we're doing a write, do the async checksumming */
1906 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1907 inode, rw, bio, mirror_num,
1908 bio_flags, bio_offset,
1909 __btrfs_submit_bio_start,
1910 __btrfs_submit_bio_done);
1912 } else if (!skip_sum) {
1913 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1919 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1923 bio->bi_error = ret;
1930 * given a list of ordered sums record them in the inode. This happens
1931 * at IO completion time based on sums calculated at bio submission time.
1933 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1934 struct inode *inode, u64 file_offset,
1935 struct list_head *list)
1937 struct btrfs_ordered_sum *sum;
1939 list_for_each_entry(sum, list, list) {
1940 trans->adding_csums = 1;
1941 btrfs_csum_file_blocks(trans,
1942 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1943 trans->adding_csums = 0;
1948 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1949 struct extent_state **cached_state)
1951 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1952 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1953 cached_state, GFP_NOFS);
1956 /* see btrfs_writepage_start_hook for details on why this is required */
1957 struct btrfs_writepage_fixup {
1959 struct btrfs_work work;
1962 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1964 struct btrfs_writepage_fixup *fixup;
1965 struct btrfs_ordered_extent *ordered;
1966 struct extent_state *cached_state = NULL;
1968 struct inode *inode;
1973 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1977 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1978 ClearPageChecked(page);
1982 inode = page->mapping->host;
1983 page_start = page_offset(page);
1984 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1986 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1989 /* already ordered? We're done */
1990 if (PagePrivate2(page))
1993 ordered = btrfs_lookup_ordered_extent(inode, page_start);
1995 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1996 page_end, &cached_state, GFP_NOFS);
1998 btrfs_start_ordered_extent(inode, ordered, 1);
1999 btrfs_put_ordered_extent(ordered);
2003 ret = btrfs_delalloc_reserve_space(inode, page_start,
2006 mapping_set_error(page->mapping, ret);
2007 end_extent_writepage(page, ret, page_start, page_end);
2008 ClearPageChecked(page);
2012 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2013 ClearPageChecked(page);
2014 set_page_dirty(page);
2016 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2017 &cached_state, GFP_NOFS);
2020 page_cache_release(page);
2025 * There are a few paths in the higher layers of the kernel that directly
2026 * set the page dirty bit without asking the filesystem if it is a
2027 * good idea. This causes problems because we want to make sure COW
2028 * properly happens and the data=ordered rules are followed.
2030 * In our case any range that doesn't have the ORDERED bit set
2031 * hasn't been properly setup for IO. We kick off an async process
2032 * to fix it up. The async helper will wait for ordered extents, set
2033 * the delalloc bit and make it safe to write the page.
2035 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2037 struct inode *inode = page->mapping->host;
2038 struct btrfs_writepage_fixup *fixup;
2039 struct btrfs_root *root = BTRFS_I(inode)->root;
2041 /* this page is properly in the ordered list */
2042 if (TestClearPagePrivate2(page))
2045 if (PageChecked(page))
2048 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2052 SetPageChecked(page);
2053 page_cache_get(page);
2054 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2055 btrfs_writepage_fixup_worker, NULL, NULL);
2057 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2061 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2062 struct inode *inode, u64 file_pos,
2063 u64 disk_bytenr, u64 disk_num_bytes,
2064 u64 num_bytes, u64 ram_bytes,
2065 u8 compression, u8 encryption,
2066 u16 other_encoding, int extent_type)
2068 struct btrfs_root *root = BTRFS_I(inode)->root;
2069 struct btrfs_file_extent_item *fi;
2070 struct btrfs_path *path;
2071 struct extent_buffer *leaf;
2072 struct btrfs_key ins;
2073 int extent_inserted = 0;
2076 path = btrfs_alloc_path();
2081 * we may be replacing one extent in the tree with another.
2082 * The new extent is pinned in the extent map, and we don't want
2083 * to drop it from the cache until it is completely in the btree.
2085 * So, tell btrfs_drop_extents to leave this extent in the cache.
2086 * the caller is expected to unpin it and allow it to be merged
2089 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2090 file_pos + num_bytes, NULL, 0,
2091 1, sizeof(*fi), &extent_inserted);
2095 if (!extent_inserted) {
2096 ins.objectid = btrfs_ino(inode);
2097 ins.offset = file_pos;
2098 ins.type = BTRFS_EXTENT_DATA_KEY;
2100 path->leave_spinning = 1;
2101 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2106 leaf = path->nodes[0];
2107 fi = btrfs_item_ptr(leaf, path->slots[0],
2108 struct btrfs_file_extent_item);
2109 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2110 btrfs_set_file_extent_type(leaf, fi, extent_type);
2111 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2112 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2113 btrfs_set_file_extent_offset(leaf, fi, 0);
2114 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2115 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2116 btrfs_set_file_extent_compression(leaf, fi, compression);
2117 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2118 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2120 btrfs_mark_buffer_dirty(leaf);
2121 btrfs_release_path(path);
2123 inode_add_bytes(inode, num_bytes);
2125 ins.objectid = disk_bytenr;
2126 ins.offset = disk_num_bytes;
2127 ins.type = BTRFS_EXTENT_ITEM_KEY;
2128 ret = btrfs_alloc_reserved_file_extent(trans, root,
2129 root->root_key.objectid,
2130 btrfs_ino(inode), file_pos,
2133 * Release the reserved range from inode dirty range map, as it is
2134 * already moved into delayed_ref_head
2136 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2138 btrfs_free_path(path);
2143 /* snapshot-aware defrag */
2144 struct sa_defrag_extent_backref {
2145 struct rb_node node;
2146 struct old_sa_defrag_extent *old;
2155 struct old_sa_defrag_extent {
2156 struct list_head list;
2157 struct new_sa_defrag_extent *new;
2166 struct new_sa_defrag_extent {
2167 struct rb_root root;
2168 struct list_head head;
2169 struct btrfs_path *path;
2170 struct inode *inode;
2178 static int backref_comp(struct sa_defrag_extent_backref *b1,
2179 struct sa_defrag_extent_backref *b2)
2181 if (b1->root_id < b2->root_id)
2183 else if (b1->root_id > b2->root_id)
2186 if (b1->inum < b2->inum)
2188 else if (b1->inum > b2->inum)
2191 if (b1->file_pos < b2->file_pos)
2193 else if (b1->file_pos > b2->file_pos)
2197 * [------------------------------] ===> (a range of space)
2198 * |<--->| |<---->| =============> (fs/file tree A)
2199 * |<---------------------------->| ===> (fs/file tree B)
2201 * A range of space can refer to two file extents in one tree while
2202 * refer to only one file extent in another tree.
2204 * So we may process a disk offset more than one time(two extents in A)
2205 * and locate at the same extent(one extent in B), then insert two same
2206 * backrefs(both refer to the extent in B).
2211 static void backref_insert(struct rb_root *root,
2212 struct sa_defrag_extent_backref *backref)
2214 struct rb_node **p = &root->rb_node;
2215 struct rb_node *parent = NULL;
2216 struct sa_defrag_extent_backref *entry;
2221 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2223 ret = backref_comp(backref, entry);
2227 p = &(*p)->rb_right;
2230 rb_link_node(&backref->node, parent, p);
2231 rb_insert_color(&backref->node, root);
2235 * Note the backref might has changed, and in this case we just return 0.
2237 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2240 struct btrfs_file_extent_item *extent;
2241 struct btrfs_fs_info *fs_info;
2242 struct old_sa_defrag_extent *old = ctx;
2243 struct new_sa_defrag_extent *new = old->new;
2244 struct btrfs_path *path = new->path;
2245 struct btrfs_key key;
2246 struct btrfs_root *root;
2247 struct sa_defrag_extent_backref *backref;
2248 struct extent_buffer *leaf;
2249 struct inode *inode = new->inode;
2255 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2256 inum == btrfs_ino(inode))
2259 key.objectid = root_id;
2260 key.type = BTRFS_ROOT_ITEM_KEY;
2261 key.offset = (u64)-1;
2263 fs_info = BTRFS_I(inode)->root->fs_info;
2264 root = btrfs_read_fs_root_no_name(fs_info, &key);
2266 if (PTR_ERR(root) == -ENOENT)
2269 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2270 inum, offset, root_id);
2271 return PTR_ERR(root);
2274 key.objectid = inum;
2275 key.type = BTRFS_EXTENT_DATA_KEY;
2276 if (offset > (u64)-1 << 32)
2279 key.offset = offset;
2281 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2282 if (WARN_ON(ret < 0))
2289 leaf = path->nodes[0];
2290 slot = path->slots[0];
2292 if (slot >= btrfs_header_nritems(leaf)) {
2293 ret = btrfs_next_leaf(root, path);
2296 } else if (ret > 0) {
2305 btrfs_item_key_to_cpu(leaf, &key, slot);
2307 if (key.objectid > inum)
2310 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2313 extent = btrfs_item_ptr(leaf, slot,
2314 struct btrfs_file_extent_item);
2316 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2320 * 'offset' refers to the exact key.offset,
2321 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2322 * (key.offset - extent_offset).
2324 if (key.offset != offset)
2327 extent_offset = btrfs_file_extent_offset(leaf, extent);
2328 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2330 if (extent_offset >= old->extent_offset + old->offset +
2331 old->len || extent_offset + num_bytes <=
2332 old->extent_offset + old->offset)
2337 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2343 backref->root_id = root_id;
2344 backref->inum = inum;
2345 backref->file_pos = offset;
2346 backref->num_bytes = num_bytes;
2347 backref->extent_offset = extent_offset;
2348 backref->generation = btrfs_file_extent_generation(leaf, extent);
2350 backref_insert(&new->root, backref);
2353 btrfs_release_path(path);
2358 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2359 struct new_sa_defrag_extent *new)
2361 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2362 struct old_sa_defrag_extent *old, *tmp;
2367 list_for_each_entry_safe(old, tmp, &new->head, list) {
2368 ret = iterate_inodes_from_logical(old->bytenr +
2369 old->extent_offset, fs_info,
2370 path, record_one_backref,
2372 if (ret < 0 && ret != -ENOENT)
2375 /* no backref to be processed for this extent */
2377 list_del(&old->list);
2382 if (list_empty(&new->head))
2388 static int relink_is_mergable(struct extent_buffer *leaf,
2389 struct btrfs_file_extent_item *fi,
2390 struct new_sa_defrag_extent *new)
2392 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2395 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2398 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2401 if (btrfs_file_extent_encryption(leaf, fi) ||
2402 btrfs_file_extent_other_encoding(leaf, fi))
2409 * Note the backref might has changed, and in this case we just return 0.
2411 static noinline int relink_extent_backref(struct btrfs_path *path,
2412 struct sa_defrag_extent_backref *prev,
2413 struct sa_defrag_extent_backref *backref)
2415 struct btrfs_file_extent_item *extent;
2416 struct btrfs_file_extent_item *item;
2417 struct btrfs_ordered_extent *ordered;
2418 struct btrfs_trans_handle *trans;
2419 struct btrfs_fs_info *fs_info;
2420 struct btrfs_root *root;
2421 struct btrfs_key key;
2422 struct extent_buffer *leaf;
2423 struct old_sa_defrag_extent *old = backref->old;
2424 struct new_sa_defrag_extent *new = old->new;
2425 struct inode *src_inode = new->inode;
2426 struct inode *inode;
2427 struct extent_state *cached = NULL;
2436 if (prev && prev->root_id == backref->root_id &&
2437 prev->inum == backref->inum &&
2438 prev->file_pos + prev->num_bytes == backref->file_pos)
2441 /* step 1: get root */
2442 key.objectid = backref->root_id;
2443 key.type = BTRFS_ROOT_ITEM_KEY;
2444 key.offset = (u64)-1;
2446 fs_info = BTRFS_I(src_inode)->root->fs_info;
2447 index = srcu_read_lock(&fs_info->subvol_srcu);
2449 root = btrfs_read_fs_root_no_name(fs_info, &key);
2451 srcu_read_unlock(&fs_info->subvol_srcu, index);
2452 if (PTR_ERR(root) == -ENOENT)
2454 return PTR_ERR(root);
2457 if (btrfs_root_readonly(root)) {
2458 srcu_read_unlock(&fs_info->subvol_srcu, index);
2462 /* step 2: get inode */
2463 key.objectid = backref->inum;
2464 key.type = BTRFS_INODE_ITEM_KEY;
2467 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2468 if (IS_ERR(inode)) {
2469 srcu_read_unlock(&fs_info->subvol_srcu, index);
2473 srcu_read_unlock(&fs_info->subvol_srcu, index);
2475 /* step 3: relink backref */
2476 lock_start = backref->file_pos;
2477 lock_end = backref->file_pos + backref->num_bytes - 1;
2478 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2481 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2483 btrfs_put_ordered_extent(ordered);
2487 trans = btrfs_join_transaction(root);
2488 if (IS_ERR(trans)) {
2489 ret = PTR_ERR(trans);
2493 key.objectid = backref->inum;
2494 key.type = BTRFS_EXTENT_DATA_KEY;
2495 key.offset = backref->file_pos;
2497 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2500 } else if (ret > 0) {
2505 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2506 struct btrfs_file_extent_item);
2508 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2509 backref->generation)
2512 btrfs_release_path(path);
2514 start = backref->file_pos;
2515 if (backref->extent_offset < old->extent_offset + old->offset)
2516 start += old->extent_offset + old->offset -
2517 backref->extent_offset;
2519 len = min(backref->extent_offset + backref->num_bytes,
2520 old->extent_offset + old->offset + old->len);
2521 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2523 ret = btrfs_drop_extents(trans, root, inode, start,
2528 key.objectid = btrfs_ino(inode);
2529 key.type = BTRFS_EXTENT_DATA_KEY;
2532 path->leave_spinning = 1;
2534 struct btrfs_file_extent_item *fi;
2536 struct btrfs_key found_key;
2538 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2543 leaf = path->nodes[0];
2544 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2546 fi = btrfs_item_ptr(leaf, path->slots[0],
2547 struct btrfs_file_extent_item);
2548 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2550 if (extent_len + found_key.offset == start &&
2551 relink_is_mergable(leaf, fi, new)) {
2552 btrfs_set_file_extent_num_bytes(leaf, fi,
2554 btrfs_mark_buffer_dirty(leaf);
2555 inode_add_bytes(inode, len);
2561 btrfs_release_path(path);
2566 ret = btrfs_insert_empty_item(trans, root, path, &key,
2569 btrfs_abort_transaction(trans, root, ret);
2573 leaf = path->nodes[0];
2574 item = btrfs_item_ptr(leaf, path->slots[0],
2575 struct btrfs_file_extent_item);
2576 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2577 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2578 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2579 btrfs_set_file_extent_num_bytes(leaf, item, len);
2580 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2581 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2582 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2583 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2584 btrfs_set_file_extent_encryption(leaf, item, 0);
2585 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2587 btrfs_mark_buffer_dirty(leaf);
2588 inode_add_bytes(inode, len);
2589 btrfs_release_path(path);
2591 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2593 backref->root_id, backref->inum,
2594 new->file_pos); /* start - extent_offset */
2596 btrfs_abort_transaction(trans, root, ret);
2602 btrfs_release_path(path);
2603 path->leave_spinning = 0;
2604 btrfs_end_transaction(trans, root);
2606 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2612 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2614 struct old_sa_defrag_extent *old, *tmp;
2619 list_for_each_entry_safe(old, tmp, &new->head, list) {
2625 static void relink_file_extents(struct new_sa_defrag_extent *new)
2627 struct btrfs_path *path;
2628 struct sa_defrag_extent_backref *backref;
2629 struct sa_defrag_extent_backref *prev = NULL;
2630 struct inode *inode;
2631 struct btrfs_root *root;
2632 struct rb_node *node;
2636 root = BTRFS_I(inode)->root;
2638 path = btrfs_alloc_path();
2642 if (!record_extent_backrefs(path, new)) {
2643 btrfs_free_path(path);
2646 btrfs_release_path(path);
2649 node = rb_first(&new->root);
2652 rb_erase(node, &new->root);
2654 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2656 ret = relink_extent_backref(path, prev, backref);
2669 btrfs_free_path(path);
2671 free_sa_defrag_extent(new);
2673 atomic_dec(&root->fs_info->defrag_running);
2674 wake_up(&root->fs_info->transaction_wait);
2677 static struct new_sa_defrag_extent *
2678 record_old_file_extents(struct inode *inode,
2679 struct btrfs_ordered_extent *ordered)
2681 struct btrfs_root *root = BTRFS_I(inode)->root;
2682 struct btrfs_path *path;
2683 struct btrfs_key key;
2684 struct old_sa_defrag_extent *old;
2685 struct new_sa_defrag_extent *new;
2688 new = kmalloc(sizeof(*new), GFP_NOFS);
2693 new->file_pos = ordered->file_offset;
2694 new->len = ordered->len;
2695 new->bytenr = ordered->start;
2696 new->disk_len = ordered->disk_len;
2697 new->compress_type = ordered->compress_type;
2698 new->root = RB_ROOT;
2699 INIT_LIST_HEAD(&new->head);
2701 path = btrfs_alloc_path();
2705 key.objectid = btrfs_ino(inode);
2706 key.type = BTRFS_EXTENT_DATA_KEY;
2707 key.offset = new->file_pos;
2709 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2712 if (ret > 0 && path->slots[0] > 0)
2715 /* find out all the old extents for the file range */
2717 struct btrfs_file_extent_item *extent;
2718 struct extent_buffer *l;
2727 slot = path->slots[0];
2729 if (slot >= btrfs_header_nritems(l)) {
2730 ret = btrfs_next_leaf(root, path);
2738 btrfs_item_key_to_cpu(l, &key, slot);
2740 if (key.objectid != btrfs_ino(inode))
2742 if (key.type != BTRFS_EXTENT_DATA_KEY)
2744 if (key.offset >= new->file_pos + new->len)
2747 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2749 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2750 if (key.offset + num_bytes < new->file_pos)
2753 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2757 extent_offset = btrfs_file_extent_offset(l, extent);
2759 old = kmalloc(sizeof(*old), GFP_NOFS);
2763 offset = max(new->file_pos, key.offset);
2764 end = min(new->file_pos + new->len, key.offset + num_bytes);
2766 old->bytenr = disk_bytenr;
2767 old->extent_offset = extent_offset;
2768 old->offset = offset - key.offset;
2769 old->len = end - offset;
2772 list_add_tail(&old->list, &new->head);
2778 btrfs_free_path(path);
2779 atomic_inc(&root->fs_info->defrag_running);
2784 btrfs_free_path(path);
2786 free_sa_defrag_extent(new);
2790 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2793 struct btrfs_block_group_cache *cache;
2795 cache = btrfs_lookup_block_group(root->fs_info, start);
2798 spin_lock(&cache->lock);
2799 cache->delalloc_bytes -= len;
2800 spin_unlock(&cache->lock);
2802 btrfs_put_block_group(cache);
2805 /* as ordered data IO finishes, this gets called so we can finish
2806 * an ordered extent if the range of bytes in the file it covers are
2809 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2811 struct inode *inode = ordered_extent->inode;
2812 struct btrfs_root *root = BTRFS_I(inode)->root;
2813 struct btrfs_trans_handle *trans = NULL;
2814 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2815 struct extent_state *cached_state = NULL;
2816 struct new_sa_defrag_extent *new = NULL;
2817 int compress_type = 0;
2819 u64 logical_len = ordered_extent->len;
2821 bool truncated = false;
2823 nolock = btrfs_is_free_space_inode(inode);
2825 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2830 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2831 ordered_extent->file_offset +
2832 ordered_extent->len - 1);
2834 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2836 logical_len = ordered_extent->truncated_len;
2837 /* Truncated the entire extent, don't bother adding */
2842 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2843 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2846 * For mwrite(mmap + memset to write) case, we still reserve
2847 * space for NOCOW range.
2848 * As NOCOW won't cause a new delayed ref, just free the space
2850 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2851 ordered_extent->len);
2852 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2854 trans = btrfs_join_transaction_nolock(root);
2856 trans = btrfs_join_transaction(root);
2857 if (IS_ERR(trans)) {
2858 ret = PTR_ERR(trans);
2862 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2863 ret = btrfs_update_inode_fallback(trans, root, inode);
2864 if (ret) /* -ENOMEM or corruption */
2865 btrfs_abort_transaction(trans, root, ret);
2869 lock_extent_bits(io_tree, ordered_extent->file_offset,
2870 ordered_extent->file_offset + ordered_extent->len - 1,
2873 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2874 ordered_extent->file_offset + ordered_extent->len - 1,
2875 EXTENT_DEFRAG, 1, cached_state);
2877 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2878 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2879 /* the inode is shared */
2880 new = record_old_file_extents(inode, ordered_extent);
2882 clear_extent_bit(io_tree, ordered_extent->file_offset,
2883 ordered_extent->file_offset + ordered_extent->len - 1,
2884 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2888 trans = btrfs_join_transaction_nolock(root);
2890 trans = btrfs_join_transaction(root);
2891 if (IS_ERR(trans)) {
2892 ret = PTR_ERR(trans);
2897 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2899 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2900 compress_type = ordered_extent->compress_type;
2901 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2902 BUG_ON(compress_type);
2903 ret = btrfs_mark_extent_written(trans, inode,
2904 ordered_extent->file_offset,
2905 ordered_extent->file_offset +
2908 BUG_ON(root == root->fs_info->tree_root);
2909 ret = insert_reserved_file_extent(trans, inode,
2910 ordered_extent->file_offset,
2911 ordered_extent->start,
2912 ordered_extent->disk_len,
2913 logical_len, logical_len,
2914 compress_type, 0, 0,
2915 BTRFS_FILE_EXTENT_REG);
2917 btrfs_release_delalloc_bytes(root,
2918 ordered_extent->start,
2919 ordered_extent->disk_len);
2921 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2922 ordered_extent->file_offset, ordered_extent->len,
2925 btrfs_abort_transaction(trans, root, ret);
2929 add_pending_csums(trans, inode, ordered_extent->file_offset,
2930 &ordered_extent->list);
2932 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2933 ret = btrfs_update_inode_fallback(trans, root, inode);
2934 if (ret) { /* -ENOMEM or corruption */
2935 btrfs_abort_transaction(trans, root, ret);
2940 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2941 ordered_extent->file_offset +
2942 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2944 if (root != root->fs_info->tree_root)
2945 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2947 btrfs_end_transaction(trans, root);
2949 if (ret || truncated) {
2953 start = ordered_extent->file_offset + logical_len;
2955 start = ordered_extent->file_offset;
2956 end = ordered_extent->file_offset + ordered_extent->len - 1;
2957 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2959 /* Drop the cache for the part of the extent we didn't write. */
2960 btrfs_drop_extent_cache(inode, start, end, 0);
2963 * If the ordered extent had an IOERR or something else went
2964 * wrong we need to return the space for this ordered extent
2965 * back to the allocator. We only free the extent in the
2966 * truncated case if we didn't write out the extent at all.
2968 if ((ret || !logical_len) &&
2969 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2970 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2971 btrfs_free_reserved_extent(root, ordered_extent->start,
2972 ordered_extent->disk_len, 1);
2977 * This needs to be done to make sure anybody waiting knows we are done
2978 * updating everything for this ordered extent.
2980 btrfs_remove_ordered_extent(inode, ordered_extent);
2982 /* for snapshot-aware defrag */
2985 free_sa_defrag_extent(new);
2986 atomic_dec(&root->fs_info->defrag_running);
2988 relink_file_extents(new);
2993 btrfs_put_ordered_extent(ordered_extent);
2994 /* once for the tree */
2995 btrfs_put_ordered_extent(ordered_extent);
3000 static void finish_ordered_fn(struct btrfs_work *work)
3002 struct btrfs_ordered_extent *ordered_extent;
3003 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3004 btrfs_finish_ordered_io(ordered_extent);
3007 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3008 struct extent_state *state, int uptodate)
3010 struct inode *inode = page->mapping->host;
3011 struct btrfs_root *root = BTRFS_I(inode)->root;
3012 struct btrfs_ordered_extent *ordered_extent = NULL;
3013 struct btrfs_workqueue *wq;
3014 btrfs_work_func_t func;
3016 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3018 ClearPagePrivate2(page);
3019 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3020 end - start + 1, uptodate))
3023 if (btrfs_is_free_space_inode(inode)) {
3024 wq = root->fs_info->endio_freespace_worker;
3025 func = btrfs_freespace_write_helper;
3027 wq = root->fs_info->endio_write_workers;
3028 func = btrfs_endio_write_helper;
3031 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3033 btrfs_queue_work(wq, &ordered_extent->work);
3038 static int __readpage_endio_check(struct inode *inode,
3039 struct btrfs_io_bio *io_bio,
3040 int icsum, struct page *page,
3041 int pgoff, u64 start, size_t len)
3047 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3049 kaddr = kmap_atomic(page);
3050 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3051 btrfs_csum_final(csum, (char *)&csum);
3052 if (csum != csum_expected)
3055 kunmap_atomic(kaddr);
3058 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3059 "csum failed ino %llu off %llu csum %u expected csum %u",
3060 btrfs_ino(inode), start, csum, csum_expected);
3061 memset(kaddr + pgoff, 1, len);
3062 flush_dcache_page(page);
3063 kunmap_atomic(kaddr);
3064 if (csum_expected == 0)
3070 * when reads are done, we need to check csums to verify the data is correct
3071 * if there's a match, we allow the bio to finish. If not, the code in
3072 * extent_io.c will try to find good copies for us.
3074 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3075 u64 phy_offset, struct page *page,
3076 u64 start, u64 end, int mirror)
3078 size_t offset = start - page_offset(page);
3079 struct inode *inode = page->mapping->host;
3080 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3081 struct btrfs_root *root = BTRFS_I(inode)->root;
3083 if (PageChecked(page)) {
3084 ClearPageChecked(page);
3088 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3091 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3092 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3093 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3098 phy_offset >>= inode->i_sb->s_blocksize_bits;
3099 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3100 start, (size_t)(end - start + 1));
3103 struct delayed_iput {
3104 struct list_head list;
3105 struct inode *inode;
3108 /* JDM: If this is fs-wide, why can't we add a pointer to
3109 * btrfs_inode instead and avoid the allocation? */
3110 void btrfs_add_delayed_iput(struct inode *inode)
3112 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3113 struct delayed_iput *delayed;
3115 if (atomic_add_unless(&inode->i_count, -1, 1))
3118 delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3119 delayed->inode = inode;
3121 spin_lock(&fs_info->delayed_iput_lock);
3122 list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3123 spin_unlock(&fs_info->delayed_iput_lock);
3126 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3129 struct btrfs_fs_info *fs_info = root->fs_info;
3130 struct delayed_iput *delayed;
3133 spin_lock(&fs_info->delayed_iput_lock);
3134 empty = list_empty(&fs_info->delayed_iputs);
3135 spin_unlock(&fs_info->delayed_iput_lock);
3139 down_read(&fs_info->delayed_iput_sem);
3141 spin_lock(&fs_info->delayed_iput_lock);
3142 list_splice_init(&fs_info->delayed_iputs, &list);
3143 spin_unlock(&fs_info->delayed_iput_lock);
3145 while (!list_empty(&list)) {
3146 delayed = list_entry(list.next, struct delayed_iput, list);
3147 list_del(&delayed->list);
3148 iput(delayed->inode);
3152 up_read(&root->fs_info->delayed_iput_sem);
3156 * This is called in transaction commit time. If there are no orphan
3157 * files in the subvolume, it removes orphan item and frees block_rsv
3160 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root)
3163 struct btrfs_block_rsv *block_rsv;
3166 if (atomic_read(&root->orphan_inodes) ||
3167 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3170 spin_lock(&root->orphan_lock);
3171 if (atomic_read(&root->orphan_inodes)) {
3172 spin_unlock(&root->orphan_lock);
3176 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3177 spin_unlock(&root->orphan_lock);
3181 block_rsv = root->orphan_block_rsv;
3182 root->orphan_block_rsv = NULL;
3183 spin_unlock(&root->orphan_lock);
3185 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3186 btrfs_root_refs(&root->root_item) > 0) {
3187 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3188 root->root_key.objectid);
3190 btrfs_abort_transaction(trans, root, ret);
3192 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3197 WARN_ON(block_rsv->size > 0);
3198 btrfs_free_block_rsv(root, block_rsv);
3203 * This creates an orphan entry for the given inode in case something goes
3204 * wrong in the middle of an unlink/truncate.
3206 * NOTE: caller of this function should reserve 5 units of metadata for
3209 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3211 struct btrfs_root *root = BTRFS_I(inode)->root;
3212 struct btrfs_block_rsv *block_rsv = NULL;
3217 if (!root->orphan_block_rsv) {
3218 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3223 spin_lock(&root->orphan_lock);
3224 if (!root->orphan_block_rsv) {
3225 root->orphan_block_rsv = block_rsv;
3226 } else if (block_rsv) {
3227 btrfs_free_block_rsv(root, block_rsv);
3231 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3232 &BTRFS_I(inode)->runtime_flags)) {
3235 * For proper ENOSPC handling, we should do orphan
3236 * cleanup when mounting. But this introduces backward
3237 * compatibility issue.
3239 if (!xchg(&root->orphan_item_inserted, 1))
3245 atomic_inc(&root->orphan_inodes);
3248 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3249 &BTRFS_I(inode)->runtime_flags))
3251 spin_unlock(&root->orphan_lock);
3253 /* grab metadata reservation from transaction handle */
3255 ret = btrfs_orphan_reserve_metadata(trans, inode);
3256 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3259 /* insert an orphan item to track this unlinked/truncated file */
3261 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3263 atomic_dec(&root->orphan_inodes);
3265 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3266 &BTRFS_I(inode)->runtime_flags);
3267 btrfs_orphan_release_metadata(inode);
3269 if (ret != -EEXIST) {
3270 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3271 &BTRFS_I(inode)->runtime_flags);
3272 btrfs_abort_transaction(trans, root, ret);
3279 /* insert an orphan item to track subvolume contains orphan files */
3281 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3282 root->root_key.objectid);
3283 if (ret && ret != -EEXIST) {
3284 btrfs_abort_transaction(trans, root, ret);
3292 * We have done the truncate/delete so we can go ahead and remove the orphan
3293 * item for this particular inode.
3295 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3296 struct inode *inode)
3298 struct btrfs_root *root = BTRFS_I(inode)->root;
3299 int delete_item = 0;
3300 int release_rsv = 0;
3303 spin_lock(&root->orphan_lock);
3304 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3305 &BTRFS_I(inode)->runtime_flags))
3308 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3309 &BTRFS_I(inode)->runtime_flags))
3311 spin_unlock(&root->orphan_lock);
3314 atomic_dec(&root->orphan_inodes);
3316 ret = btrfs_del_orphan_item(trans, root,
3321 btrfs_orphan_release_metadata(inode);
3327 * this cleans up any orphans that may be left on the list from the last use
3330 int btrfs_orphan_cleanup(struct btrfs_root *root)
3332 struct btrfs_path *path;
3333 struct extent_buffer *leaf;
3334 struct btrfs_key key, found_key;
3335 struct btrfs_trans_handle *trans;
3336 struct inode *inode;
3337 u64 last_objectid = 0;
3338 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3340 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3343 path = btrfs_alloc_path();
3350 key.objectid = BTRFS_ORPHAN_OBJECTID;
3351 key.type = BTRFS_ORPHAN_ITEM_KEY;
3352 key.offset = (u64)-1;
3355 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3360 * if ret == 0 means we found what we were searching for, which
3361 * is weird, but possible, so only screw with path if we didn't
3362 * find the key and see if we have stuff that matches
3366 if (path->slots[0] == 0)
3371 /* pull out the item */
3372 leaf = path->nodes[0];
3373 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3375 /* make sure the item matches what we want */
3376 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3378 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3381 /* release the path since we're done with it */
3382 btrfs_release_path(path);
3385 * this is where we are basically btrfs_lookup, without the
3386 * crossing root thing. we store the inode number in the
3387 * offset of the orphan item.
3390 if (found_key.offset == last_objectid) {
3391 btrfs_err(root->fs_info,
3392 "Error removing orphan entry, stopping orphan cleanup");
3397 last_objectid = found_key.offset;
3399 found_key.objectid = found_key.offset;
3400 found_key.type = BTRFS_INODE_ITEM_KEY;
3401 found_key.offset = 0;
3402 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3403 ret = PTR_ERR_OR_ZERO(inode);
3404 if (ret && ret != -ESTALE)
3407 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3408 struct btrfs_root *dead_root;
3409 struct btrfs_fs_info *fs_info = root->fs_info;
3410 int is_dead_root = 0;
3413 * this is an orphan in the tree root. Currently these
3414 * could come from 2 sources:
3415 * a) a snapshot deletion in progress
3416 * b) a free space cache inode
3417 * We need to distinguish those two, as the snapshot
3418 * orphan must not get deleted.
3419 * find_dead_roots already ran before us, so if this
3420 * is a snapshot deletion, we should find the root
3421 * in the dead_roots list
3423 spin_lock(&fs_info->trans_lock);
3424 list_for_each_entry(dead_root, &fs_info->dead_roots,
3426 if (dead_root->root_key.objectid ==
3427 found_key.objectid) {
3432 spin_unlock(&fs_info->trans_lock);
3434 /* prevent this orphan from being found again */
3435 key.offset = found_key.objectid - 1;
3440 * Inode is already gone but the orphan item is still there,
3441 * kill the orphan item.
3443 if (ret == -ESTALE) {
3444 trans = btrfs_start_transaction(root, 1);
3445 if (IS_ERR(trans)) {
3446 ret = PTR_ERR(trans);
3449 btrfs_debug(root->fs_info, "auto deleting %Lu",
3450 found_key.objectid);
3451 ret = btrfs_del_orphan_item(trans, root,
3452 found_key.objectid);
3453 btrfs_end_transaction(trans, root);
3460 * add this inode to the orphan list so btrfs_orphan_del does
3461 * the proper thing when we hit it
3463 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3464 &BTRFS_I(inode)->runtime_flags);
3465 atomic_inc(&root->orphan_inodes);
3467 /* if we have links, this was a truncate, lets do that */
3468 if (inode->i_nlink) {
3469 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3475 /* 1 for the orphan item deletion. */
3476 trans = btrfs_start_transaction(root, 1);
3477 if (IS_ERR(trans)) {
3479 ret = PTR_ERR(trans);
3482 ret = btrfs_orphan_add(trans, inode);
3483 btrfs_end_transaction(trans, root);
3489 ret = btrfs_truncate(inode);
3491 btrfs_orphan_del(NULL, inode);
3496 /* this will do delete_inode and everything for us */
3501 /* release the path since we're done with it */
3502 btrfs_release_path(path);
3504 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3506 if (root->orphan_block_rsv)
3507 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3510 if (root->orphan_block_rsv ||
3511 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3512 trans = btrfs_join_transaction(root);
3514 btrfs_end_transaction(trans, root);
3518 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3520 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3524 btrfs_err(root->fs_info,
3525 "could not do orphan cleanup %d", ret);
3526 btrfs_free_path(path);
3531 * very simple check to peek ahead in the leaf looking for xattrs. If we
3532 * don't find any xattrs, we know there can't be any acls.
3534 * slot is the slot the inode is in, objectid is the objectid of the inode
3536 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3537 int slot, u64 objectid,
3538 int *first_xattr_slot)
3540 u32 nritems = btrfs_header_nritems(leaf);
3541 struct btrfs_key found_key;
3542 static u64 xattr_access = 0;
3543 static u64 xattr_default = 0;
3546 if (!xattr_access) {
3547 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3548 strlen(POSIX_ACL_XATTR_ACCESS));
3549 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3550 strlen(POSIX_ACL_XATTR_DEFAULT));
3554 *first_xattr_slot = -1;
3555 while (slot < nritems) {
3556 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3558 /* we found a different objectid, there must not be acls */
3559 if (found_key.objectid != objectid)
3562 /* we found an xattr, assume we've got an acl */
3563 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3564 if (*first_xattr_slot == -1)
3565 *first_xattr_slot = slot;
3566 if (found_key.offset == xattr_access ||
3567 found_key.offset == xattr_default)
3572 * we found a key greater than an xattr key, there can't
3573 * be any acls later on
3575 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3582 * it goes inode, inode backrefs, xattrs, extents,
3583 * so if there are a ton of hard links to an inode there can
3584 * be a lot of backrefs. Don't waste time searching too hard,
3585 * this is just an optimization
3590 /* we hit the end of the leaf before we found an xattr or
3591 * something larger than an xattr. We have to assume the inode
3594 if (*first_xattr_slot == -1)
3595 *first_xattr_slot = slot;
3600 * read an inode from the btree into the in-memory inode
3602 static void btrfs_read_locked_inode(struct inode *inode)
3604 struct btrfs_path *path;
3605 struct extent_buffer *leaf;
3606 struct btrfs_inode_item *inode_item;
3607 struct btrfs_root *root = BTRFS_I(inode)->root;
3608 struct btrfs_key location;
3613 bool filled = false;
3614 int first_xattr_slot;
3616 ret = btrfs_fill_inode(inode, &rdev);
3620 path = btrfs_alloc_path();
3624 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3626 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3630 leaf = path->nodes[0];
3635 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3636 struct btrfs_inode_item);
3637 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3638 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3639 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3640 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3641 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3643 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3644 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3646 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3647 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3649 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3650 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3652 BTRFS_I(inode)->i_otime.tv_sec =
3653 btrfs_timespec_sec(leaf, &inode_item->otime);
3654 BTRFS_I(inode)->i_otime.tv_nsec =
3655 btrfs_timespec_nsec(leaf, &inode_item->otime);
3657 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3658 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3659 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3661 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3662 inode->i_generation = BTRFS_I(inode)->generation;
3664 rdev = btrfs_inode_rdev(leaf, inode_item);
3666 BTRFS_I(inode)->index_cnt = (u64)-1;
3667 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3671 * If we were modified in the current generation and evicted from memory
3672 * and then re-read we need to do a full sync since we don't have any
3673 * idea about which extents were modified before we were evicted from
3676 * This is required for both inode re-read from disk and delayed inode
3677 * in delayed_nodes_tree.
3679 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3680 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3681 &BTRFS_I(inode)->runtime_flags);
3684 * We don't persist the id of the transaction where an unlink operation
3685 * against the inode was last made. So here we assume the inode might
3686 * have been evicted, and therefore the exact value of last_unlink_trans
3687 * lost, and set it to last_trans to avoid metadata inconsistencies
3688 * between the inode and its parent if the inode is fsync'ed and the log
3689 * replayed. For example, in the scenario:
3692 * ln mydir/foo mydir/bar
3695 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3696 * xfs_io -c fsync mydir/foo
3698 * mount fs, triggers fsync log replay
3700 * We must make sure that when we fsync our inode foo we also log its
3701 * parent inode, otherwise after log replay the parent still has the
3702 * dentry with the "bar" name but our inode foo has a link count of 1
3703 * and doesn't have an inode ref with the name "bar" anymore.
3705 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3706 * but it guarantees correctness at the expense of ocassional full
3707 * transaction commits on fsync if our inode is a directory, or if our
3708 * inode is not a directory, logging its parent unnecessarily.
3710 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3713 if (inode->i_nlink != 1 ||
3714 path->slots[0] >= btrfs_header_nritems(leaf))
3717 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3718 if (location.objectid != btrfs_ino(inode))
3721 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3722 if (location.type == BTRFS_INODE_REF_KEY) {
3723 struct btrfs_inode_ref *ref;
3725 ref = (struct btrfs_inode_ref *)ptr;
3726 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3727 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3728 struct btrfs_inode_extref *extref;
3730 extref = (struct btrfs_inode_extref *)ptr;
3731 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3736 * try to precache a NULL acl entry for files that don't have
3737 * any xattrs or acls
3739 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3740 btrfs_ino(inode), &first_xattr_slot);
3741 if (first_xattr_slot != -1) {
3742 path->slots[0] = first_xattr_slot;
3743 ret = btrfs_load_inode_props(inode, path);
3745 btrfs_err(root->fs_info,
3746 "error loading props for ino %llu (root %llu): %d",
3748 root->root_key.objectid, ret);
3750 btrfs_free_path(path);
3753 cache_no_acl(inode);
3755 switch (inode->i_mode & S_IFMT) {
3757 inode->i_mapping->a_ops = &btrfs_aops;
3758 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3759 inode->i_fop = &btrfs_file_operations;
3760 inode->i_op = &btrfs_file_inode_operations;
3763 inode->i_fop = &btrfs_dir_file_operations;
3764 if (root == root->fs_info->tree_root)
3765 inode->i_op = &btrfs_dir_ro_inode_operations;
3767 inode->i_op = &btrfs_dir_inode_operations;
3770 inode->i_op = &btrfs_symlink_inode_operations;
3771 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3774 inode->i_op = &btrfs_special_inode_operations;
3775 init_special_inode(inode, inode->i_mode, rdev);
3779 btrfs_update_iflags(inode);
3783 btrfs_free_path(path);
3784 make_bad_inode(inode);
3788 * given a leaf and an inode, copy the inode fields into the leaf
3790 static void fill_inode_item(struct btrfs_trans_handle *trans,
3791 struct extent_buffer *leaf,
3792 struct btrfs_inode_item *item,
3793 struct inode *inode)
3795 struct btrfs_map_token token;
3797 btrfs_init_map_token(&token);
3799 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3800 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3801 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3803 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3804 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3806 btrfs_set_token_timespec_sec(leaf, &item->atime,
3807 inode->i_atime.tv_sec, &token);
3808 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3809 inode->i_atime.tv_nsec, &token);
3811 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3812 inode->i_mtime.tv_sec, &token);
3813 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3814 inode->i_mtime.tv_nsec, &token);
3816 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3817 inode->i_ctime.tv_sec, &token);
3818 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3819 inode->i_ctime.tv_nsec, &token);
3821 btrfs_set_token_timespec_sec(leaf, &item->otime,
3822 BTRFS_I(inode)->i_otime.tv_sec, &token);
3823 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3824 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3826 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3828 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3830 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3831 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3832 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3833 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3834 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3838 * copy everything in the in-memory inode into the btree.
3840 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3841 struct btrfs_root *root, struct inode *inode)
3843 struct btrfs_inode_item *inode_item;
3844 struct btrfs_path *path;
3845 struct extent_buffer *leaf;
3848 path = btrfs_alloc_path();
3852 path->leave_spinning = 1;
3853 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3861 leaf = path->nodes[0];
3862 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3863 struct btrfs_inode_item);
3865 fill_inode_item(trans, leaf, inode_item, inode);
3866 btrfs_mark_buffer_dirty(leaf);
3867 btrfs_set_inode_last_trans(trans, inode);
3870 btrfs_free_path(path);
3875 * copy everything in the in-memory inode into the btree.
3877 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3878 struct btrfs_root *root, struct inode *inode)
3883 * If the inode is a free space inode, we can deadlock during commit
3884 * if we put it into the delayed code.
3886 * The data relocation inode should also be directly updated
3889 if (!btrfs_is_free_space_inode(inode)
3890 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3891 && !root->fs_info->log_root_recovering) {
3892 btrfs_update_root_times(trans, root);
3894 ret = btrfs_delayed_update_inode(trans, root, inode);
3896 btrfs_set_inode_last_trans(trans, inode);
3900 return btrfs_update_inode_item(trans, root, inode);
3903 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3904 struct btrfs_root *root,
3905 struct inode *inode)
3909 ret = btrfs_update_inode(trans, root, inode);
3911 return btrfs_update_inode_item(trans, root, inode);
3916 * unlink helper that gets used here in inode.c and in the tree logging
3917 * recovery code. It remove a link in a directory with a given name, and
3918 * also drops the back refs in the inode to the directory
3920 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3921 struct btrfs_root *root,
3922 struct inode *dir, struct inode *inode,
3923 const char *name, int name_len)
3925 struct btrfs_path *path;
3927 struct extent_buffer *leaf;
3928 struct btrfs_dir_item *di;
3929 struct btrfs_key key;
3931 u64 ino = btrfs_ino(inode);
3932 u64 dir_ino = btrfs_ino(dir);
3934 path = btrfs_alloc_path();
3940 path->leave_spinning = 1;
3941 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3942 name, name_len, -1);
3951 leaf = path->nodes[0];
3952 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3953 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3956 btrfs_release_path(path);
3959 * If we don't have dir index, we have to get it by looking up
3960 * the inode ref, since we get the inode ref, remove it directly,
3961 * it is unnecessary to do delayed deletion.
3963 * But if we have dir index, needn't search inode ref to get it.
3964 * Since the inode ref is close to the inode item, it is better
3965 * that we delay to delete it, and just do this deletion when
3966 * we update the inode item.
3968 if (BTRFS_I(inode)->dir_index) {
3969 ret = btrfs_delayed_delete_inode_ref(inode);
3971 index = BTRFS_I(inode)->dir_index;
3976 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3979 btrfs_info(root->fs_info,
3980 "failed to delete reference to %.*s, inode %llu parent %llu",
3981 name_len, name, ino, dir_ino);
3982 btrfs_abort_transaction(trans, root, ret);
3986 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3988 btrfs_abort_transaction(trans, root, ret);
3992 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3994 if (ret != 0 && ret != -ENOENT) {
3995 btrfs_abort_transaction(trans, root, ret);
3999 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4004 btrfs_abort_transaction(trans, root, ret);
4006 btrfs_free_path(path);
4010 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4011 inode_inc_iversion(inode);
4012 inode_inc_iversion(dir);
4013 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4014 ret = btrfs_update_inode(trans, root, dir);
4019 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4020 struct btrfs_root *root,
4021 struct inode *dir, struct inode *inode,
4022 const char *name, int name_len)
4025 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4028 ret = btrfs_update_inode(trans, root, inode);
4034 * helper to start transaction for unlink and rmdir.
4036 * unlink and rmdir are special in btrfs, they do not always free space, so
4037 * if we cannot make our reservations the normal way try and see if there is
4038 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4039 * allow the unlink to occur.
4041 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4043 struct btrfs_trans_handle *trans;
4044 struct btrfs_root *root = BTRFS_I(dir)->root;
4048 * 1 for the possible orphan item
4049 * 1 for the dir item
4050 * 1 for the dir index
4051 * 1 for the inode ref
4054 trans = btrfs_start_transaction(root, 5);
4055 if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
4058 if (PTR_ERR(trans) == -ENOSPC) {
4059 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
4061 trans = btrfs_start_transaction(root, 0);
4064 ret = btrfs_cond_migrate_bytes(root->fs_info,
4065 &root->fs_info->trans_block_rsv,
4068 btrfs_end_transaction(trans, root);
4069 return ERR_PTR(ret);
4071 trans->block_rsv = &root->fs_info->trans_block_rsv;
4072 trans->bytes_reserved = num_bytes;
4077 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4079 struct btrfs_root *root = BTRFS_I(dir)->root;
4080 struct btrfs_trans_handle *trans;
4081 struct inode *inode = d_inode(dentry);
4084 trans = __unlink_start_trans(dir);
4086 return PTR_ERR(trans);
4088 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4090 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4091 dentry->d_name.name, dentry->d_name.len);
4095 if (inode->i_nlink == 0) {
4096 ret = btrfs_orphan_add(trans, inode);
4102 btrfs_end_transaction(trans, root);
4103 btrfs_btree_balance_dirty(root);
4107 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4108 struct btrfs_root *root,
4109 struct inode *dir, u64 objectid,
4110 const char *name, int name_len)
4112 struct btrfs_path *path;
4113 struct extent_buffer *leaf;
4114 struct btrfs_dir_item *di;
4115 struct btrfs_key key;
4118 u64 dir_ino = btrfs_ino(dir);
4120 path = btrfs_alloc_path();
4124 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4125 name, name_len, -1);
4126 if (IS_ERR_OR_NULL(di)) {
4134 leaf = path->nodes[0];
4135 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4136 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4137 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4139 btrfs_abort_transaction(trans, root, ret);
4142 btrfs_release_path(path);
4144 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4145 objectid, root->root_key.objectid,
4146 dir_ino, &index, name, name_len);
4148 if (ret != -ENOENT) {
4149 btrfs_abort_transaction(trans, root, ret);
4152 di = btrfs_search_dir_index_item(root, path, dir_ino,
4154 if (IS_ERR_OR_NULL(di)) {
4159 btrfs_abort_transaction(trans, root, ret);
4163 leaf = path->nodes[0];
4164 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4165 btrfs_release_path(path);
4168 btrfs_release_path(path);
4170 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4172 btrfs_abort_transaction(trans, root, ret);
4176 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4177 inode_inc_iversion(dir);
4178 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4179 ret = btrfs_update_inode_fallback(trans, root, dir);
4181 btrfs_abort_transaction(trans, root, ret);
4183 btrfs_free_path(path);
4187 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4189 struct inode *inode = d_inode(dentry);
4191 struct btrfs_root *root = BTRFS_I(dir)->root;
4192 struct btrfs_trans_handle *trans;
4194 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4196 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4199 trans = __unlink_start_trans(dir);
4201 return PTR_ERR(trans);
4203 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4204 err = btrfs_unlink_subvol(trans, root, dir,
4205 BTRFS_I(inode)->location.objectid,
4206 dentry->d_name.name,
4207 dentry->d_name.len);
4211 err = btrfs_orphan_add(trans, inode);
4215 /* now the directory is empty */
4216 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4217 dentry->d_name.name, dentry->d_name.len);
4219 btrfs_i_size_write(inode, 0);
4221 btrfs_end_transaction(trans, root);
4222 btrfs_btree_balance_dirty(root);
4227 static int truncate_space_check(struct btrfs_trans_handle *trans,
4228 struct btrfs_root *root,
4233 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4234 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4235 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4237 trans->bytes_reserved += bytes_deleted;
4242 static int truncate_inline_extent(struct inode *inode,
4243 struct btrfs_path *path,
4244 struct btrfs_key *found_key,
4248 struct extent_buffer *leaf = path->nodes[0];
4249 int slot = path->slots[0];
4250 struct btrfs_file_extent_item *fi;
4251 u32 size = (u32)(new_size - found_key->offset);
4252 struct btrfs_root *root = BTRFS_I(inode)->root;
4254 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4256 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4257 loff_t offset = new_size;
4258 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4261 * Zero out the remaining of the last page of our inline extent,
4262 * instead of directly truncating our inline extent here - that
4263 * would be much more complex (decompressing all the data, then
4264 * compressing the truncated data, which might be bigger than
4265 * the size of the inline extent, resize the extent, etc).
4266 * We release the path because to get the page we might need to
4267 * read the extent item from disk (data not in the page cache).
4269 btrfs_release_path(path);
4270 return btrfs_truncate_page(inode, offset, page_end - offset, 0);
4273 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4274 size = btrfs_file_extent_calc_inline_size(size);
4275 btrfs_truncate_item(root, path, size, 1);
4277 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4278 inode_sub_bytes(inode, item_end + 1 - new_size);
4284 * this can truncate away extent items, csum items and directory items.
4285 * It starts at a high offset and removes keys until it can't find
4286 * any higher than new_size
4288 * csum items that cross the new i_size are truncated to the new size
4291 * min_type is the minimum key type to truncate down to. If set to 0, this
4292 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4294 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4295 struct btrfs_root *root,
4296 struct inode *inode,
4297 u64 new_size, u32 min_type)
4299 struct btrfs_path *path;
4300 struct extent_buffer *leaf;
4301 struct btrfs_file_extent_item *fi;
4302 struct btrfs_key key;
4303 struct btrfs_key found_key;
4304 u64 extent_start = 0;
4305 u64 extent_num_bytes = 0;
4306 u64 extent_offset = 0;
4308 u64 last_size = new_size;
4309 u32 found_type = (u8)-1;
4312 int pending_del_nr = 0;
4313 int pending_del_slot = 0;
4314 int extent_type = -1;
4317 u64 ino = btrfs_ino(inode);
4318 u64 bytes_deleted = 0;
4320 bool should_throttle = 0;
4321 bool should_end = 0;
4323 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4326 * for non-free space inodes and ref cows, we want to back off from
4329 if (!btrfs_is_free_space_inode(inode) &&
4330 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4333 path = btrfs_alloc_path();
4339 * We want to drop from the next block forward in case this new size is
4340 * not block aligned since we will be keeping the last block of the
4341 * extent just the way it is.
4343 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4344 root == root->fs_info->tree_root)
4345 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4346 root->sectorsize), (u64)-1, 0);
4349 * This function is also used to drop the items in the log tree before
4350 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4351 * it is used to drop the loged items. So we shouldn't kill the delayed
4354 if (min_type == 0 && root == BTRFS_I(inode)->root)
4355 btrfs_kill_delayed_inode_items(inode);
4358 key.offset = (u64)-1;
4363 * with a 16K leaf size and 128MB extents, you can actually queue
4364 * up a huge file in a single leaf. Most of the time that
4365 * bytes_deleted is > 0, it will be huge by the time we get here
4367 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4368 if (btrfs_should_end_transaction(trans, root)) {
4375 path->leave_spinning = 1;
4376 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4383 /* there are no items in the tree for us to truncate, we're
4386 if (path->slots[0] == 0)
4393 leaf = path->nodes[0];
4394 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4395 found_type = found_key.type;
4397 if (found_key.objectid != ino)
4400 if (found_type < min_type)
4403 item_end = found_key.offset;
4404 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4405 fi = btrfs_item_ptr(leaf, path->slots[0],
4406 struct btrfs_file_extent_item);
4407 extent_type = btrfs_file_extent_type(leaf, fi);
4408 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4410 btrfs_file_extent_num_bytes(leaf, fi);
4411 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4412 item_end += btrfs_file_extent_inline_len(leaf,
4413 path->slots[0], fi);
4417 if (found_type > min_type) {
4420 if (item_end < new_size)
4422 if (found_key.offset >= new_size)
4428 /* FIXME, shrink the extent if the ref count is only 1 */
4429 if (found_type != BTRFS_EXTENT_DATA_KEY)
4433 last_size = found_key.offset;
4435 last_size = new_size;
4437 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4439 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4441 u64 orig_num_bytes =
4442 btrfs_file_extent_num_bytes(leaf, fi);
4443 extent_num_bytes = ALIGN(new_size -
4446 btrfs_set_file_extent_num_bytes(leaf, fi,
4448 num_dec = (orig_num_bytes -
4450 if (test_bit(BTRFS_ROOT_REF_COWS,
4453 inode_sub_bytes(inode, num_dec);
4454 btrfs_mark_buffer_dirty(leaf);
4457 btrfs_file_extent_disk_num_bytes(leaf,
4459 extent_offset = found_key.offset -
4460 btrfs_file_extent_offset(leaf, fi);
4462 /* FIXME blocksize != 4096 */
4463 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4464 if (extent_start != 0) {
4466 if (test_bit(BTRFS_ROOT_REF_COWS,
4468 inode_sub_bytes(inode, num_dec);
4471 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4473 * we can't truncate inline items that have had
4477 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4478 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4481 * Need to release path in order to truncate a
4482 * compressed extent. So delete any accumulated
4483 * extent items so far.
4485 if (btrfs_file_extent_compression(leaf, fi) !=
4486 BTRFS_COMPRESS_NONE && pending_del_nr) {
4487 err = btrfs_del_items(trans, root, path,
4491 btrfs_abort_transaction(trans,
4499 err = truncate_inline_extent(inode, path,
4504 btrfs_abort_transaction(trans,
4508 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4510 inode_sub_bytes(inode, item_end + 1 - new_size);
4515 if (!pending_del_nr) {
4516 /* no pending yet, add ourselves */
4517 pending_del_slot = path->slots[0];
4519 } else if (pending_del_nr &&
4520 path->slots[0] + 1 == pending_del_slot) {
4521 /* hop on the pending chunk */
4523 pending_del_slot = path->slots[0];
4530 should_throttle = 0;
4533 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4534 root == root->fs_info->tree_root)) {
4535 btrfs_set_path_blocking(path);
4536 bytes_deleted += extent_num_bytes;
4537 ret = btrfs_free_extent(trans, root, extent_start,
4538 extent_num_bytes, 0,
4539 btrfs_header_owner(leaf),
4540 ino, extent_offset);
4542 if (btrfs_should_throttle_delayed_refs(trans, root))
4543 btrfs_async_run_delayed_refs(root,
4544 trans->delayed_ref_updates * 2, 0);
4546 if (truncate_space_check(trans, root,
4547 extent_num_bytes)) {
4550 if (btrfs_should_throttle_delayed_refs(trans,
4552 should_throttle = 1;
4557 if (found_type == BTRFS_INODE_ITEM_KEY)
4560 if (path->slots[0] == 0 ||
4561 path->slots[0] != pending_del_slot ||
4562 should_throttle || should_end) {
4563 if (pending_del_nr) {
4564 ret = btrfs_del_items(trans, root, path,
4568 btrfs_abort_transaction(trans,
4574 btrfs_release_path(path);
4575 if (should_throttle) {
4576 unsigned long updates = trans->delayed_ref_updates;
4578 trans->delayed_ref_updates = 0;
4579 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4585 * if we failed to refill our space rsv, bail out
4586 * and let the transaction restart
4598 if (pending_del_nr) {
4599 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4602 btrfs_abort_transaction(trans, root, ret);
4605 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4606 btrfs_ordered_update_i_size(inode, last_size, NULL);
4608 btrfs_free_path(path);
4610 if (be_nice && bytes_deleted > 32 * 1024 * 1024) {
4611 unsigned long updates = trans->delayed_ref_updates;
4613 trans->delayed_ref_updates = 0;
4614 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4623 * btrfs_truncate_page - read, zero a chunk and write a page
4624 * @inode - inode that we're zeroing
4625 * @from - the offset to start zeroing
4626 * @len - the length to zero, 0 to zero the entire range respective to the
4628 * @front - zero up to the offset instead of from the offset on
4630 * This will find the page for the "from" offset and cow the page and zero the
4631 * part we want to zero. This is used with truncate and hole punching.
4633 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4636 struct address_space *mapping = inode->i_mapping;
4637 struct btrfs_root *root = BTRFS_I(inode)->root;
4638 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4639 struct btrfs_ordered_extent *ordered;
4640 struct extent_state *cached_state = NULL;
4642 u32 blocksize = root->sectorsize;
4643 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4644 unsigned offset = from & (PAGE_CACHE_SIZE-1);
4646 gfp_t mask = btrfs_alloc_write_mask(mapping);
4651 if ((offset & (blocksize - 1)) == 0 &&
4652 (!len || ((len & (blocksize - 1)) == 0)))
4654 ret = btrfs_delalloc_reserve_space(inode,
4655 round_down(from, PAGE_CACHE_SIZE), PAGE_CACHE_SIZE);
4660 page = find_or_create_page(mapping, index, mask);
4662 btrfs_delalloc_release_space(inode,
4663 round_down(from, PAGE_CACHE_SIZE),
4669 page_start = page_offset(page);
4670 page_end = page_start + PAGE_CACHE_SIZE - 1;
4672 if (!PageUptodate(page)) {
4673 ret = btrfs_readpage(NULL, page);
4675 if (page->mapping != mapping) {
4677 page_cache_release(page);
4680 if (!PageUptodate(page)) {
4685 wait_on_page_writeback(page);
4687 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4688 set_page_extent_mapped(page);
4690 ordered = btrfs_lookup_ordered_extent(inode, page_start);
4692 unlock_extent_cached(io_tree, page_start, page_end,
4693 &cached_state, GFP_NOFS);
4695 page_cache_release(page);
4696 btrfs_start_ordered_extent(inode, ordered, 1);
4697 btrfs_put_ordered_extent(ordered);
4701 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4702 EXTENT_DIRTY | EXTENT_DELALLOC |
4703 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4704 0, 0, &cached_state, GFP_NOFS);
4706 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4709 unlock_extent_cached(io_tree, page_start, page_end,
4710 &cached_state, GFP_NOFS);
4714 if (offset != PAGE_CACHE_SIZE) {
4716 len = PAGE_CACHE_SIZE - offset;
4719 memset(kaddr, 0, offset);
4721 memset(kaddr + offset, 0, len);
4722 flush_dcache_page(page);
4725 ClearPageChecked(page);
4726 set_page_dirty(page);
4727 unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4732 btrfs_delalloc_release_space(inode, page_start,
4735 page_cache_release(page);
4740 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4741 u64 offset, u64 len)
4743 struct btrfs_trans_handle *trans;
4747 * Still need to make sure the inode looks like it's been updated so
4748 * that any holes get logged if we fsync.
4750 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4751 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4752 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4753 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4758 * 1 - for the one we're dropping
4759 * 1 - for the one we're adding
4760 * 1 - for updating the inode.
4762 trans = btrfs_start_transaction(root, 3);
4764 return PTR_ERR(trans);
4766 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4768 btrfs_abort_transaction(trans, root, ret);
4769 btrfs_end_transaction(trans, root);
4773 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4774 0, 0, len, 0, len, 0, 0, 0);
4776 btrfs_abort_transaction(trans, root, ret);
4778 btrfs_update_inode(trans, root, inode);
4779 btrfs_end_transaction(trans, root);
4784 * This function puts in dummy file extents for the area we're creating a hole
4785 * for. So if we are truncating this file to a larger size we need to insert
4786 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4787 * the range between oldsize and size
4789 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4791 struct btrfs_root *root = BTRFS_I(inode)->root;
4792 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4793 struct extent_map *em = NULL;
4794 struct extent_state *cached_state = NULL;
4795 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4796 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4797 u64 block_end = ALIGN(size, root->sectorsize);
4804 * If our size started in the middle of a page we need to zero out the
4805 * rest of the page before we expand the i_size, otherwise we could
4806 * expose stale data.
4808 err = btrfs_truncate_page(inode, oldsize, 0, 0);
4812 if (size <= hole_start)
4816 struct btrfs_ordered_extent *ordered;
4818 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4820 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4821 block_end - hole_start);
4824 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4825 &cached_state, GFP_NOFS);
4826 btrfs_start_ordered_extent(inode, ordered, 1);
4827 btrfs_put_ordered_extent(ordered);
4830 cur_offset = hole_start;
4832 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4833 block_end - cur_offset, 0);
4839 last_byte = min(extent_map_end(em), block_end);
4840 last_byte = ALIGN(last_byte , root->sectorsize);
4841 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4842 struct extent_map *hole_em;
4843 hole_size = last_byte - cur_offset;
4845 err = maybe_insert_hole(root, inode, cur_offset,
4849 btrfs_drop_extent_cache(inode, cur_offset,
4850 cur_offset + hole_size - 1, 0);
4851 hole_em = alloc_extent_map();
4853 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4854 &BTRFS_I(inode)->runtime_flags);
4857 hole_em->start = cur_offset;
4858 hole_em->len = hole_size;
4859 hole_em->orig_start = cur_offset;
4861 hole_em->block_start = EXTENT_MAP_HOLE;
4862 hole_em->block_len = 0;
4863 hole_em->orig_block_len = 0;
4864 hole_em->ram_bytes = hole_size;
4865 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4866 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4867 hole_em->generation = root->fs_info->generation;
4870 write_lock(&em_tree->lock);
4871 err = add_extent_mapping(em_tree, hole_em, 1);
4872 write_unlock(&em_tree->lock);
4875 btrfs_drop_extent_cache(inode, cur_offset,
4879 free_extent_map(hole_em);
4882 free_extent_map(em);
4884 cur_offset = last_byte;
4885 if (cur_offset >= block_end)
4888 free_extent_map(em);
4889 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4894 static int wait_snapshoting_atomic_t(atomic_t *a)
4900 static void wait_for_snapshot_creation(struct btrfs_root *root)
4905 ret = btrfs_start_write_no_snapshoting(root);
4908 wait_on_atomic_t(&root->will_be_snapshoted,
4909 wait_snapshoting_atomic_t,
4910 TASK_UNINTERRUPTIBLE);
4914 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4916 struct btrfs_root *root = BTRFS_I(inode)->root;
4917 struct btrfs_trans_handle *trans;
4918 loff_t oldsize = i_size_read(inode);
4919 loff_t newsize = attr->ia_size;
4920 int mask = attr->ia_valid;
4924 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4925 * special case where we need to update the times despite not having
4926 * these flags set. For all other operations the VFS set these flags
4927 * explicitly if it wants a timestamp update.
4929 if (newsize != oldsize) {
4930 inode_inc_iversion(inode);
4931 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4932 inode->i_ctime = inode->i_mtime =
4933 current_fs_time(inode->i_sb);
4936 if (newsize > oldsize) {
4937 truncate_pagecache(inode, newsize);
4939 * Don't do an expanding truncate while snapshoting is ongoing.
4940 * This is to ensure the snapshot captures a fully consistent
4941 * state of this file - if the snapshot captures this expanding
4942 * truncation, it must capture all writes that happened before
4945 wait_for_snapshot_creation(root);
4946 ret = btrfs_cont_expand(inode, oldsize, newsize);
4948 btrfs_end_write_no_snapshoting(root);
4952 trans = btrfs_start_transaction(root, 1);
4953 if (IS_ERR(trans)) {
4954 btrfs_end_write_no_snapshoting(root);
4955 return PTR_ERR(trans);
4958 i_size_write(inode, newsize);
4959 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4960 ret = btrfs_update_inode(trans, root, inode);
4961 btrfs_end_write_no_snapshoting(root);
4962 btrfs_end_transaction(trans, root);
4966 * We're truncating a file that used to have good data down to
4967 * zero. Make sure it gets into the ordered flush list so that
4968 * any new writes get down to disk quickly.
4971 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4972 &BTRFS_I(inode)->runtime_flags);
4975 * 1 for the orphan item we're going to add
4976 * 1 for the orphan item deletion.
4978 trans = btrfs_start_transaction(root, 2);
4980 return PTR_ERR(trans);
4983 * We need to do this in case we fail at _any_ point during the
4984 * actual truncate. Once we do the truncate_setsize we could
4985 * invalidate pages which forces any outstanding ordered io to
4986 * be instantly completed which will give us extents that need
4987 * to be truncated. If we fail to get an orphan inode down we
4988 * could have left over extents that were never meant to live,
4989 * so we need to garuntee from this point on that everything
4990 * will be consistent.
4992 ret = btrfs_orphan_add(trans, inode);
4993 btrfs_end_transaction(trans, root);
4997 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4998 truncate_setsize(inode, newsize);
5000 /* Disable nonlocked read DIO to avoid the end less truncate */
5001 btrfs_inode_block_unlocked_dio(inode);
5002 inode_dio_wait(inode);
5003 btrfs_inode_resume_unlocked_dio(inode);
5005 ret = btrfs_truncate(inode);
5006 if (ret && inode->i_nlink) {
5010 * failed to truncate, disk_i_size is only adjusted down
5011 * as we remove extents, so it should represent the true
5012 * size of the inode, so reset the in memory size and
5013 * delete our orphan entry.
5015 trans = btrfs_join_transaction(root);
5016 if (IS_ERR(trans)) {
5017 btrfs_orphan_del(NULL, inode);
5020 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5021 err = btrfs_orphan_del(trans, inode);
5023 btrfs_abort_transaction(trans, root, err);
5024 btrfs_end_transaction(trans, root);
5031 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5033 struct inode *inode = d_inode(dentry);
5034 struct btrfs_root *root = BTRFS_I(inode)->root;
5037 if (btrfs_root_readonly(root))
5040 err = inode_change_ok(inode, attr);
5044 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5045 err = btrfs_setsize(inode, attr);
5050 if (attr->ia_valid) {
5051 setattr_copy(inode, attr);
5052 inode_inc_iversion(inode);
5053 err = btrfs_dirty_inode(inode);
5055 if (!err && attr->ia_valid & ATTR_MODE)
5056 err = posix_acl_chmod(inode, inode->i_mode);
5063 * While truncating the inode pages during eviction, we get the VFS calling
5064 * btrfs_invalidatepage() against each page of the inode. This is slow because
5065 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5066 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5067 * extent_state structures over and over, wasting lots of time.
5069 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5070 * those expensive operations on a per page basis and do only the ordered io
5071 * finishing, while we release here the extent_map and extent_state structures,
5072 * without the excessive merging and splitting.
5074 static void evict_inode_truncate_pages(struct inode *inode)
5076 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5077 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5078 struct rb_node *node;
5080 ASSERT(inode->i_state & I_FREEING);
5081 truncate_inode_pages_final(&inode->i_data);
5083 write_lock(&map_tree->lock);
5084 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5085 struct extent_map *em;
5087 node = rb_first(&map_tree->map);
5088 em = rb_entry(node, struct extent_map, rb_node);
5089 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5090 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5091 remove_extent_mapping(map_tree, em);
5092 free_extent_map(em);
5093 if (need_resched()) {
5094 write_unlock(&map_tree->lock);
5096 write_lock(&map_tree->lock);
5099 write_unlock(&map_tree->lock);
5102 * Keep looping until we have no more ranges in the io tree.
5103 * We can have ongoing bios started by readpages (called from readahead)
5104 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5105 * still in progress (unlocked the pages in the bio but did not yet
5106 * unlocked the ranges in the io tree). Therefore this means some
5107 * ranges can still be locked and eviction started because before
5108 * submitting those bios, which are executed by a separate task (work
5109 * queue kthread), inode references (inode->i_count) were not taken
5110 * (which would be dropped in the end io callback of each bio).
5111 * Therefore here we effectively end up waiting for those bios and
5112 * anyone else holding locked ranges without having bumped the inode's
5113 * reference count - if we don't do it, when they access the inode's
5114 * io_tree to unlock a range it may be too late, leading to an
5115 * use-after-free issue.
5117 spin_lock(&io_tree->lock);
5118 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5119 struct extent_state *state;
5120 struct extent_state *cached_state = NULL;
5124 node = rb_first(&io_tree->state);
5125 state = rb_entry(node, struct extent_state, rb_node);
5126 start = state->start;
5128 spin_unlock(&io_tree->lock);
5130 lock_extent_bits(io_tree, start, end, 0, &cached_state);
5133 * If still has DELALLOC flag, the extent didn't reach disk,
5134 * and its reserved space won't be freed by delayed_ref.
5135 * So we need to free its reserved space here.
5136 * (Refer to comment in btrfs_invalidatepage, case 2)
5138 * Note, end is the bytenr of last byte, so we need + 1 here.
5140 if (state->state & EXTENT_DELALLOC)
5141 btrfs_qgroup_free_data(inode, start, end - start + 1);
5143 clear_extent_bit(io_tree, start, end,
5144 EXTENT_LOCKED | EXTENT_DIRTY |
5145 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5146 EXTENT_DEFRAG, 1, 1,
5147 &cached_state, GFP_NOFS);
5150 spin_lock(&io_tree->lock);
5152 spin_unlock(&io_tree->lock);
5155 void btrfs_evict_inode(struct inode *inode)
5157 struct btrfs_trans_handle *trans;
5158 struct btrfs_root *root = BTRFS_I(inode)->root;
5159 struct btrfs_block_rsv *rsv, *global_rsv;
5160 int steal_from_global = 0;
5161 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5164 trace_btrfs_inode_evict(inode);
5166 evict_inode_truncate_pages(inode);
5168 if (inode->i_nlink &&
5169 ((btrfs_root_refs(&root->root_item) != 0 &&
5170 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5171 btrfs_is_free_space_inode(inode)))
5174 if (is_bad_inode(inode)) {
5175 btrfs_orphan_del(NULL, inode);
5178 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5179 if (!special_file(inode->i_mode))
5180 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5182 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5184 if (root->fs_info->log_root_recovering) {
5185 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5186 &BTRFS_I(inode)->runtime_flags));
5190 if (inode->i_nlink > 0) {
5191 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5192 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5196 ret = btrfs_commit_inode_delayed_inode(inode);
5198 btrfs_orphan_del(NULL, inode);
5202 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5204 btrfs_orphan_del(NULL, inode);
5207 rsv->size = min_size;
5209 global_rsv = &root->fs_info->global_block_rsv;
5211 btrfs_i_size_write(inode, 0);
5214 * This is a bit simpler than btrfs_truncate since we've already
5215 * reserved our space for our orphan item in the unlink, so we just
5216 * need to reserve some slack space in case we add bytes and update
5217 * inode item when doing the truncate.
5220 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5221 BTRFS_RESERVE_FLUSH_LIMIT);
5224 * Try and steal from the global reserve since we will
5225 * likely not use this space anyway, we want to try as
5226 * hard as possible to get this to work.
5229 steal_from_global++;
5231 steal_from_global = 0;
5235 * steal_from_global == 0: we reserved stuff, hooray!
5236 * steal_from_global == 1: we didn't reserve stuff, boo!
5237 * steal_from_global == 2: we've committed, still not a lot of
5238 * room but maybe we'll have room in the global reserve this
5240 * steal_from_global == 3: abandon all hope!
5242 if (steal_from_global > 2) {
5243 btrfs_warn(root->fs_info,
5244 "Could not get space for a delete, will truncate on mount %d",
5246 btrfs_orphan_del(NULL, inode);
5247 btrfs_free_block_rsv(root, rsv);
5251 trans = btrfs_join_transaction(root);
5252 if (IS_ERR(trans)) {
5253 btrfs_orphan_del(NULL, inode);
5254 btrfs_free_block_rsv(root, rsv);
5259 * We can't just steal from the global reserve, we need tomake
5260 * sure there is room to do it, if not we need to commit and try
5263 if (steal_from_global) {
5264 if (!btrfs_check_space_for_delayed_refs(trans, root))
5265 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5272 * Couldn't steal from the global reserve, we have too much
5273 * pending stuff built up, commit the transaction and try it
5277 ret = btrfs_commit_transaction(trans, root);
5279 btrfs_orphan_del(NULL, inode);
5280 btrfs_free_block_rsv(root, rsv);
5285 steal_from_global = 0;
5288 trans->block_rsv = rsv;
5290 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5291 if (ret != -ENOSPC && ret != -EAGAIN)
5294 trans->block_rsv = &root->fs_info->trans_block_rsv;
5295 btrfs_end_transaction(trans, root);
5297 btrfs_btree_balance_dirty(root);
5300 btrfs_free_block_rsv(root, rsv);
5303 * Errors here aren't a big deal, it just means we leave orphan items
5304 * in the tree. They will be cleaned up on the next mount.
5307 trans->block_rsv = root->orphan_block_rsv;
5308 btrfs_orphan_del(trans, inode);
5310 btrfs_orphan_del(NULL, inode);
5313 trans->block_rsv = &root->fs_info->trans_block_rsv;
5314 if (!(root == root->fs_info->tree_root ||
5315 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5316 btrfs_return_ino(root, btrfs_ino(inode));
5318 btrfs_end_transaction(trans, root);
5319 btrfs_btree_balance_dirty(root);
5321 btrfs_remove_delayed_node(inode);
5327 * this returns the key found in the dir entry in the location pointer.
5328 * If no dir entries were found, location->objectid is 0.
5330 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5331 struct btrfs_key *location)
5333 const char *name = dentry->d_name.name;
5334 int namelen = dentry->d_name.len;
5335 struct btrfs_dir_item *di;
5336 struct btrfs_path *path;
5337 struct btrfs_root *root = BTRFS_I(dir)->root;
5340 path = btrfs_alloc_path();
5344 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5349 if (IS_ERR_OR_NULL(di))
5352 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5354 btrfs_free_path(path);
5357 location->objectid = 0;
5362 * when we hit a tree root in a directory, the btrfs part of the inode
5363 * needs to be changed to reflect the root directory of the tree root. This
5364 * is kind of like crossing a mount point.
5366 static int fixup_tree_root_location(struct btrfs_root *root,
5368 struct dentry *dentry,
5369 struct btrfs_key *location,
5370 struct btrfs_root **sub_root)
5372 struct btrfs_path *path;
5373 struct btrfs_root *new_root;
5374 struct btrfs_root_ref *ref;
5375 struct extent_buffer *leaf;
5376 struct btrfs_key key;
5380 path = btrfs_alloc_path();
5387 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5388 key.type = BTRFS_ROOT_REF_KEY;
5389 key.offset = location->objectid;
5391 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5399 leaf = path->nodes[0];
5400 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5401 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5402 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5405 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5406 (unsigned long)(ref + 1),
5407 dentry->d_name.len);
5411 btrfs_release_path(path);
5413 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5414 if (IS_ERR(new_root)) {
5415 err = PTR_ERR(new_root);
5419 *sub_root = new_root;
5420 location->objectid = btrfs_root_dirid(&new_root->root_item);
5421 location->type = BTRFS_INODE_ITEM_KEY;
5422 location->offset = 0;
5425 btrfs_free_path(path);
5429 static void inode_tree_add(struct inode *inode)
5431 struct btrfs_root *root = BTRFS_I(inode)->root;
5432 struct btrfs_inode *entry;
5434 struct rb_node *parent;
5435 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5436 u64 ino = btrfs_ino(inode);
5438 if (inode_unhashed(inode))
5441 spin_lock(&root->inode_lock);
5442 p = &root->inode_tree.rb_node;
5445 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5447 if (ino < btrfs_ino(&entry->vfs_inode))
5448 p = &parent->rb_left;
5449 else if (ino > btrfs_ino(&entry->vfs_inode))
5450 p = &parent->rb_right;
5452 WARN_ON(!(entry->vfs_inode.i_state &
5453 (I_WILL_FREE | I_FREEING)));
5454 rb_replace_node(parent, new, &root->inode_tree);
5455 RB_CLEAR_NODE(parent);
5456 spin_unlock(&root->inode_lock);
5460 rb_link_node(new, parent, p);
5461 rb_insert_color(new, &root->inode_tree);
5462 spin_unlock(&root->inode_lock);
5465 static void inode_tree_del(struct inode *inode)
5467 struct btrfs_root *root = BTRFS_I(inode)->root;
5470 spin_lock(&root->inode_lock);
5471 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5472 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5473 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5474 empty = RB_EMPTY_ROOT(&root->inode_tree);
5476 spin_unlock(&root->inode_lock);
5478 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5479 synchronize_srcu(&root->fs_info->subvol_srcu);
5480 spin_lock(&root->inode_lock);
5481 empty = RB_EMPTY_ROOT(&root->inode_tree);
5482 spin_unlock(&root->inode_lock);
5484 btrfs_add_dead_root(root);
5488 void btrfs_invalidate_inodes(struct btrfs_root *root)
5490 struct rb_node *node;
5491 struct rb_node *prev;
5492 struct btrfs_inode *entry;
5493 struct inode *inode;
5496 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5497 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5499 spin_lock(&root->inode_lock);
5501 node = root->inode_tree.rb_node;
5505 entry = rb_entry(node, struct btrfs_inode, rb_node);
5507 if (objectid < btrfs_ino(&entry->vfs_inode))
5508 node = node->rb_left;
5509 else if (objectid > btrfs_ino(&entry->vfs_inode))
5510 node = node->rb_right;
5516 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5517 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5521 prev = rb_next(prev);
5525 entry = rb_entry(node, struct btrfs_inode, rb_node);
5526 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5527 inode = igrab(&entry->vfs_inode);
5529 spin_unlock(&root->inode_lock);
5530 if (atomic_read(&inode->i_count) > 1)
5531 d_prune_aliases(inode);
5533 * btrfs_drop_inode will have it removed from
5534 * the inode cache when its usage count
5539 spin_lock(&root->inode_lock);
5543 if (cond_resched_lock(&root->inode_lock))
5546 node = rb_next(node);
5548 spin_unlock(&root->inode_lock);
5551 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5553 struct btrfs_iget_args *args = p;
5554 inode->i_ino = args->location->objectid;
5555 memcpy(&BTRFS_I(inode)->location, args->location,
5556 sizeof(*args->location));
5557 BTRFS_I(inode)->root = args->root;
5561 static int btrfs_find_actor(struct inode *inode, void *opaque)
5563 struct btrfs_iget_args *args = opaque;
5564 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5565 args->root == BTRFS_I(inode)->root;
5568 static struct inode *btrfs_iget_locked(struct super_block *s,
5569 struct btrfs_key *location,
5570 struct btrfs_root *root)
5572 struct inode *inode;
5573 struct btrfs_iget_args args;
5574 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5576 args.location = location;
5579 inode = iget5_locked(s, hashval, btrfs_find_actor,
5580 btrfs_init_locked_inode,
5585 /* Get an inode object given its location and corresponding root.
5586 * Returns in *is_new if the inode was read from disk
5588 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5589 struct btrfs_root *root, int *new)
5591 struct inode *inode;
5593 inode = btrfs_iget_locked(s, location, root);
5595 return ERR_PTR(-ENOMEM);
5597 if (inode->i_state & I_NEW) {
5598 btrfs_read_locked_inode(inode);
5599 if (!is_bad_inode(inode)) {
5600 inode_tree_add(inode);
5601 unlock_new_inode(inode);
5605 unlock_new_inode(inode);
5607 inode = ERR_PTR(-ESTALE);
5614 static struct inode *new_simple_dir(struct super_block *s,
5615 struct btrfs_key *key,
5616 struct btrfs_root *root)
5618 struct inode *inode = new_inode(s);
5621 return ERR_PTR(-ENOMEM);
5623 BTRFS_I(inode)->root = root;
5624 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5625 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5627 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5628 inode->i_op = &btrfs_dir_ro_inode_operations;
5629 inode->i_fop = &simple_dir_operations;
5630 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5631 inode->i_mtime = CURRENT_TIME;
5632 inode->i_atime = inode->i_mtime;
5633 inode->i_ctime = inode->i_mtime;
5634 BTRFS_I(inode)->i_otime = inode->i_mtime;
5639 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5641 struct inode *inode;
5642 struct btrfs_root *root = BTRFS_I(dir)->root;
5643 struct btrfs_root *sub_root = root;
5644 struct btrfs_key location;
5648 if (dentry->d_name.len > BTRFS_NAME_LEN)
5649 return ERR_PTR(-ENAMETOOLONG);
5651 ret = btrfs_inode_by_name(dir, dentry, &location);
5653 return ERR_PTR(ret);
5655 if (location.objectid == 0)
5656 return ERR_PTR(-ENOENT);
5658 if (location.type == BTRFS_INODE_ITEM_KEY) {
5659 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5663 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5665 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5666 ret = fixup_tree_root_location(root, dir, dentry,
5667 &location, &sub_root);
5670 inode = ERR_PTR(ret);
5672 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5674 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5676 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5678 if (!IS_ERR(inode) && root != sub_root) {
5679 down_read(&root->fs_info->cleanup_work_sem);
5680 if (!(inode->i_sb->s_flags & MS_RDONLY))
5681 ret = btrfs_orphan_cleanup(sub_root);
5682 up_read(&root->fs_info->cleanup_work_sem);
5685 inode = ERR_PTR(ret);
5692 static int btrfs_dentry_delete(const struct dentry *dentry)
5694 struct btrfs_root *root;
5695 struct inode *inode = d_inode(dentry);
5697 if (!inode && !IS_ROOT(dentry))
5698 inode = d_inode(dentry->d_parent);
5701 root = BTRFS_I(inode)->root;
5702 if (btrfs_root_refs(&root->root_item) == 0)
5705 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5711 static void btrfs_dentry_release(struct dentry *dentry)
5713 kfree(dentry->d_fsdata);
5716 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5719 struct inode *inode;
5721 inode = btrfs_lookup_dentry(dir, dentry);
5722 if (IS_ERR(inode)) {
5723 if (PTR_ERR(inode) == -ENOENT)
5726 return ERR_CAST(inode);
5729 return d_splice_alias(inode, dentry);
5732 unsigned char btrfs_filetype_table[] = {
5733 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5736 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5738 struct inode *inode = file_inode(file);
5739 struct btrfs_root *root = BTRFS_I(inode)->root;
5740 struct btrfs_item *item;
5741 struct btrfs_dir_item *di;
5742 struct btrfs_key key;
5743 struct btrfs_key found_key;
5744 struct btrfs_path *path;
5745 struct list_head ins_list;
5746 struct list_head del_list;
5748 struct extent_buffer *leaf;
5750 unsigned char d_type;
5755 int key_type = BTRFS_DIR_INDEX_KEY;
5759 int is_curr = 0; /* ctx->pos points to the current index? */
5761 /* FIXME, use a real flag for deciding about the key type */
5762 if (root->fs_info->tree_root == root)
5763 key_type = BTRFS_DIR_ITEM_KEY;
5765 if (!dir_emit_dots(file, ctx))
5768 path = btrfs_alloc_path();
5774 if (key_type == BTRFS_DIR_INDEX_KEY) {
5775 INIT_LIST_HEAD(&ins_list);
5776 INIT_LIST_HEAD(&del_list);
5777 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5780 key.type = key_type;
5781 key.offset = ctx->pos;
5782 key.objectid = btrfs_ino(inode);
5784 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5789 leaf = path->nodes[0];
5790 slot = path->slots[0];
5791 if (slot >= btrfs_header_nritems(leaf)) {
5792 ret = btrfs_next_leaf(root, path);
5800 item = btrfs_item_nr(slot);
5801 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5803 if (found_key.objectid != key.objectid)
5805 if (found_key.type != key_type)
5807 if (found_key.offset < ctx->pos)
5809 if (key_type == BTRFS_DIR_INDEX_KEY &&
5810 btrfs_should_delete_dir_index(&del_list,
5814 ctx->pos = found_key.offset;
5817 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5819 di_total = btrfs_item_size(leaf, item);
5821 while (di_cur < di_total) {
5822 struct btrfs_key location;
5824 if (verify_dir_item(root, leaf, di))
5827 name_len = btrfs_dir_name_len(leaf, di);
5828 if (name_len <= sizeof(tmp_name)) {
5829 name_ptr = tmp_name;
5831 name_ptr = kmalloc(name_len, GFP_NOFS);
5837 read_extent_buffer(leaf, name_ptr,
5838 (unsigned long)(di + 1), name_len);
5840 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5841 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5844 /* is this a reference to our own snapshot? If so
5847 * In contrast to old kernels, we insert the snapshot's
5848 * dir item and dir index after it has been created, so
5849 * we won't find a reference to our own snapshot. We
5850 * still keep the following code for backward
5853 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5854 location.objectid == root->root_key.objectid) {
5858 over = !dir_emit(ctx, name_ptr, name_len,
5859 location.objectid, d_type);
5862 if (name_ptr != tmp_name)
5867 di_len = btrfs_dir_name_len(leaf, di) +
5868 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5870 di = (struct btrfs_dir_item *)((char *)di + di_len);
5876 if (key_type == BTRFS_DIR_INDEX_KEY) {
5879 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5884 /* Reached end of directory/root. Bump pos past the last item. */
5888 * Stop new entries from being returned after we return the last
5891 * New directory entries are assigned a strictly increasing
5892 * offset. This means that new entries created during readdir
5893 * are *guaranteed* to be seen in the future by that readdir.
5894 * This has broken buggy programs which operate on names as
5895 * they're returned by readdir. Until we re-use freed offsets
5896 * we have this hack to stop new entries from being returned
5897 * under the assumption that they'll never reach this huge
5900 * This is being careful not to overflow 32bit loff_t unless the
5901 * last entry requires it because doing so has broken 32bit apps
5904 if (key_type == BTRFS_DIR_INDEX_KEY) {
5905 if (ctx->pos >= INT_MAX)
5906 ctx->pos = LLONG_MAX;
5913 if (key_type == BTRFS_DIR_INDEX_KEY)
5914 btrfs_put_delayed_items(&ins_list, &del_list);
5915 btrfs_free_path(path);
5919 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5921 struct btrfs_root *root = BTRFS_I(inode)->root;
5922 struct btrfs_trans_handle *trans;
5924 bool nolock = false;
5926 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5929 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5932 if (wbc->sync_mode == WB_SYNC_ALL) {
5934 trans = btrfs_join_transaction_nolock(root);
5936 trans = btrfs_join_transaction(root);
5938 return PTR_ERR(trans);
5939 ret = btrfs_commit_transaction(trans, root);
5945 * This is somewhat expensive, updating the tree every time the
5946 * inode changes. But, it is most likely to find the inode in cache.
5947 * FIXME, needs more benchmarking...there are no reasons other than performance
5948 * to keep or drop this code.
5950 static int btrfs_dirty_inode(struct inode *inode)
5952 struct btrfs_root *root = BTRFS_I(inode)->root;
5953 struct btrfs_trans_handle *trans;
5956 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5959 trans = btrfs_join_transaction(root);
5961 return PTR_ERR(trans);
5963 ret = btrfs_update_inode(trans, root, inode);
5964 if (ret && ret == -ENOSPC) {
5965 /* whoops, lets try again with the full transaction */
5966 btrfs_end_transaction(trans, root);
5967 trans = btrfs_start_transaction(root, 1);
5969 return PTR_ERR(trans);
5971 ret = btrfs_update_inode(trans, root, inode);
5973 btrfs_end_transaction(trans, root);
5974 if (BTRFS_I(inode)->delayed_node)
5975 btrfs_balance_delayed_items(root);
5981 * This is a copy of file_update_time. We need this so we can return error on
5982 * ENOSPC for updating the inode in the case of file write and mmap writes.
5984 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5987 struct btrfs_root *root = BTRFS_I(inode)->root;
5989 if (btrfs_root_readonly(root))
5992 if (flags & S_VERSION)
5993 inode_inc_iversion(inode);
5994 if (flags & S_CTIME)
5995 inode->i_ctime = *now;
5996 if (flags & S_MTIME)
5997 inode->i_mtime = *now;
5998 if (flags & S_ATIME)
5999 inode->i_atime = *now;
6000 return btrfs_dirty_inode(inode);
6004 * find the highest existing sequence number in a directory
6005 * and then set the in-memory index_cnt variable to reflect
6006 * free sequence numbers
6008 static int btrfs_set_inode_index_count(struct inode *inode)
6010 struct btrfs_root *root = BTRFS_I(inode)->root;
6011 struct btrfs_key key, found_key;
6012 struct btrfs_path *path;
6013 struct extent_buffer *leaf;
6016 key.objectid = btrfs_ino(inode);
6017 key.type = BTRFS_DIR_INDEX_KEY;
6018 key.offset = (u64)-1;
6020 path = btrfs_alloc_path();
6024 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6027 /* FIXME: we should be able to handle this */
6033 * MAGIC NUMBER EXPLANATION:
6034 * since we search a directory based on f_pos we have to start at 2
6035 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6036 * else has to start at 2
6038 if (path->slots[0] == 0) {
6039 BTRFS_I(inode)->index_cnt = 2;
6045 leaf = path->nodes[0];
6046 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6048 if (found_key.objectid != btrfs_ino(inode) ||
6049 found_key.type != BTRFS_DIR_INDEX_KEY) {
6050 BTRFS_I(inode)->index_cnt = 2;
6054 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6056 btrfs_free_path(path);
6061 * helper to find a free sequence number in a given directory. This current
6062 * code is very simple, later versions will do smarter things in the btree
6064 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6068 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6069 ret = btrfs_inode_delayed_dir_index_count(dir);
6071 ret = btrfs_set_inode_index_count(dir);
6077 *index = BTRFS_I(dir)->index_cnt;
6078 BTRFS_I(dir)->index_cnt++;
6083 static int btrfs_insert_inode_locked(struct inode *inode)
6085 struct btrfs_iget_args args;
6086 args.location = &BTRFS_I(inode)->location;
6087 args.root = BTRFS_I(inode)->root;
6089 return insert_inode_locked4(inode,
6090 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6091 btrfs_find_actor, &args);
6094 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6095 struct btrfs_root *root,
6097 const char *name, int name_len,
6098 u64 ref_objectid, u64 objectid,
6099 umode_t mode, u64 *index)
6101 struct inode *inode;
6102 struct btrfs_inode_item *inode_item;
6103 struct btrfs_key *location;
6104 struct btrfs_path *path;
6105 struct btrfs_inode_ref *ref;
6106 struct btrfs_key key[2];
6108 int nitems = name ? 2 : 1;
6112 path = btrfs_alloc_path();
6114 return ERR_PTR(-ENOMEM);
6116 inode = new_inode(root->fs_info->sb);
6118 btrfs_free_path(path);
6119 return ERR_PTR(-ENOMEM);
6123 * O_TMPFILE, set link count to 0, so that after this point,
6124 * we fill in an inode item with the correct link count.
6127 set_nlink(inode, 0);
6130 * we have to initialize this early, so we can reclaim the inode
6131 * number if we fail afterwards in this function.
6133 inode->i_ino = objectid;
6136 trace_btrfs_inode_request(dir);
6138 ret = btrfs_set_inode_index(dir, index);
6140 btrfs_free_path(path);
6142 return ERR_PTR(ret);
6148 * index_cnt is ignored for everything but a dir,
6149 * btrfs_get_inode_index_count has an explanation for the magic
6152 BTRFS_I(inode)->index_cnt = 2;
6153 BTRFS_I(inode)->dir_index = *index;
6154 BTRFS_I(inode)->root = root;
6155 BTRFS_I(inode)->generation = trans->transid;
6156 inode->i_generation = BTRFS_I(inode)->generation;
6159 * We could have gotten an inode number from somebody who was fsynced
6160 * and then removed in this same transaction, so let's just set full
6161 * sync since it will be a full sync anyway and this will blow away the
6162 * old info in the log.
6164 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6166 key[0].objectid = objectid;
6167 key[0].type = BTRFS_INODE_ITEM_KEY;
6170 sizes[0] = sizeof(struct btrfs_inode_item);
6174 * Start new inodes with an inode_ref. This is slightly more
6175 * efficient for small numbers of hard links since they will
6176 * be packed into one item. Extended refs will kick in if we
6177 * add more hard links than can fit in the ref item.
6179 key[1].objectid = objectid;
6180 key[1].type = BTRFS_INODE_REF_KEY;
6181 key[1].offset = ref_objectid;
6183 sizes[1] = name_len + sizeof(*ref);
6186 location = &BTRFS_I(inode)->location;
6187 location->objectid = objectid;
6188 location->offset = 0;
6189 location->type = BTRFS_INODE_ITEM_KEY;
6191 ret = btrfs_insert_inode_locked(inode);
6195 path->leave_spinning = 1;
6196 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6200 inode_init_owner(inode, dir, mode);
6201 inode_set_bytes(inode, 0);
6203 inode->i_mtime = CURRENT_TIME;
6204 inode->i_atime = inode->i_mtime;
6205 inode->i_ctime = inode->i_mtime;
6206 BTRFS_I(inode)->i_otime = inode->i_mtime;
6208 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6209 struct btrfs_inode_item);
6210 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6211 sizeof(*inode_item));
6212 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6215 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6216 struct btrfs_inode_ref);
6217 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6218 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6219 ptr = (unsigned long)(ref + 1);
6220 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6223 btrfs_mark_buffer_dirty(path->nodes[0]);
6224 btrfs_free_path(path);
6226 btrfs_inherit_iflags(inode, dir);
6228 if (S_ISREG(mode)) {
6229 if (btrfs_test_opt(root, NODATASUM))
6230 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6231 if (btrfs_test_opt(root, NODATACOW))
6232 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6233 BTRFS_INODE_NODATASUM;
6236 inode_tree_add(inode);
6238 trace_btrfs_inode_new(inode);
6239 btrfs_set_inode_last_trans(trans, inode);
6241 btrfs_update_root_times(trans, root);
6243 ret = btrfs_inode_inherit_props(trans, inode, dir);
6245 btrfs_err(root->fs_info,
6246 "error inheriting props for ino %llu (root %llu): %d",
6247 btrfs_ino(inode), root->root_key.objectid, ret);
6252 unlock_new_inode(inode);
6255 BTRFS_I(dir)->index_cnt--;
6256 btrfs_free_path(path);
6258 return ERR_PTR(ret);
6261 static inline u8 btrfs_inode_type(struct inode *inode)
6263 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6267 * utility function to add 'inode' into 'parent_inode' with
6268 * a give name and a given sequence number.
6269 * if 'add_backref' is true, also insert a backref from the
6270 * inode to the parent directory.
6272 int btrfs_add_link(struct btrfs_trans_handle *trans,
6273 struct inode *parent_inode, struct inode *inode,
6274 const char *name, int name_len, int add_backref, u64 index)
6277 struct btrfs_key key;
6278 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6279 u64 ino = btrfs_ino(inode);
6280 u64 parent_ino = btrfs_ino(parent_inode);
6282 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6283 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6286 key.type = BTRFS_INODE_ITEM_KEY;
6290 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6291 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6292 key.objectid, root->root_key.objectid,
6293 parent_ino, index, name, name_len);
6294 } else if (add_backref) {
6295 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6299 /* Nothing to clean up yet */
6303 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6305 btrfs_inode_type(inode), index);
6306 if (ret == -EEXIST || ret == -EOVERFLOW)
6309 btrfs_abort_transaction(trans, root, ret);
6313 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6315 inode_inc_iversion(parent_inode);
6316 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6317 ret = btrfs_update_inode(trans, root, parent_inode);
6319 btrfs_abort_transaction(trans, root, ret);
6323 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6326 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6327 key.objectid, root->root_key.objectid,
6328 parent_ino, &local_index, name, name_len);
6330 } else if (add_backref) {
6334 err = btrfs_del_inode_ref(trans, root, name, name_len,
6335 ino, parent_ino, &local_index);
6340 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6341 struct inode *dir, struct dentry *dentry,
6342 struct inode *inode, int backref, u64 index)
6344 int err = btrfs_add_link(trans, dir, inode,
6345 dentry->d_name.name, dentry->d_name.len,
6352 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6353 umode_t mode, dev_t rdev)
6355 struct btrfs_trans_handle *trans;
6356 struct btrfs_root *root = BTRFS_I(dir)->root;
6357 struct inode *inode = NULL;
6363 if (!new_valid_dev(rdev))
6367 * 2 for inode item and ref
6369 * 1 for xattr if selinux is on
6371 trans = btrfs_start_transaction(root, 5);
6373 return PTR_ERR(trans);
6375 err = btrfs_find_free_ino(root, &objectid);
6379 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6380 dentry->d_name.len, btrfs_ino(dir), objectid,
6382 if (IS_ERR(inode)) {
6383 err = PTR_ERR(inode);
6388 * If the active LSM wants to access the inode during
6389 * d_instantiate it needs these. Smack checks to see
6390 * if the filesystem supports xattrs by looking at the
6393 inode->i_op = &btrfs_special_inode_operations;
6394 init_special_inode(inode, inode->i_mode, rdev);
6396 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6398 goto out_unlock_inode;
6400 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6402 goto out_unlock_inode;
6404 btrfs_update_inode(trans, root, inode);
6405 unlock_new_inode(inode);
6406 d_instantiate(dentry, inode);
6410 btrfs_end_transaction(trans, root);
6411 btrfs_balance_delayed_items(root);
6412 btrfs_btree_balance_dirty(root);
6414 inode_dec_link_count(inode);
6421 unlock_new_inode(inode);
6426 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6427 umode_t mode, bool excl)
6429 struct btrfs_trans_handle *trans;
6430 struct btrfs_root *root = BTRFS_I(dir)->root;
6431 struct inode *inode = NULL;
6432 int drop_inode_on_err = 0;
6438 * 2 for inode item and ref
6440 * 1 for xattr if selinux is on
6442 trans = btrfs_start_transaction(root, 5);
6444 return PTR_ERR(trans);
6446 err = btrfs_find_free_ino(root, &objectid);
6450 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6451 dentry->d_name.len, btrfs_ino(dir), objectid,
6453 if (IS_ERR(inode)) {
6454 err = PTR_ERR(inode);
6457 drop_inode_on_err = 1;
6459 * If the active LSM wants to access the inode during
6460 * d_instantiate it needs these. Smack checks to see
6461 * if the filesystem supports xattrs by looking at the
6464 inode->i_fop = &btrfs_file_operations;
6465 inode->i_op = &btrfs_file_inode_operations;
6466 inode->i_mapping->a_ops = &btrfs_aops;
6468 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6470 goto out_unlock_inode;
6472 err = btrfs_update_inode(trans, root, inode);
6474 goto out_unlock_inode;
6476 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6478 goto out_unlock_inode;
6480 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6481 unlock_new_inode(inode);
6482 d_instantiate(dentry, inode);
6485 btrfs_end_transaction(trans, root);
6486 if (err && drop_inode_on_err) {
6487 inode_dec_link_count(inode);
6490 btrfs_balance_delayed_items(root);
6491 btrfs_btree_balance_dirty(root);
6495 unlock_new_inode(inode);
6500 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6501 struct dentry *dentry)
6503 struct btrfs_trans_handle *trans;
6504 struct btrfs_root *root = BTRFS_I(dir)->root;
6505 struct inode *inode = d_inode(old_dentry);
6510 /* do not allow sys_link's with other subvols of the same device */
6511 if (root->objectid != BTRFS_I(inode)->root->objectid)
6514 if (inode->i_nlink >= BTRFS_LINK_MAX)
6517 err = btrfs_set_inode_index(dir, &index);
6522 * 2 items for inode and inode ref
6523 * 2 items for dir items
6524 * 1 item for parent inode
6526 trans = btrfs_start_transaction(root, 5);
6527 if (IS_ERR(trans)) {
6528 err = PTR_ERR(trans);
6532 /* There are several dir indexes for this inode, clear the cache. */
6533 BTRFS_I(inode)->dir_index = 0ULL;
6535 inode_inc_iversion(inode);
6536 inode->i_ctime = CURRENT_TIME;
6538 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6540 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6545 struct dentry *parent = dentry->d_parent;
6546 err = btrfs_update_inode(trans, root, inode);
6549 if (inode->i_nlink == 1) {
6551 * If new hard link count is 1, it's a file created
6552 * with open(2) O_TMPFILE flag.
6554 err = btrfs_orphan_del(trans, inode);
6558 d_instantiate(dentry, inode);
6559 btrfs_log_new_name(trans, inode, NULL, parent);
6562 btrfs_end_transaction(trans, root);
6563 btrfs_balance_delayed_items(root);
6566 inode_dec_link_count(inode);
6569 btrfs_btree_balance_dirty(root);
6573 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6575 struct inode *inode = NULL;
6576 struct btrfs_trans_handle *trans;
6577 struct btrfs_root *root = BTRFS_I(dir)->root;
6579 int drop_on_err = 0;
6584 * 2 items for inode and ref
6585 * 2 items for dir items
6586 * 1 for xattr if selinux is on
6588 trans = btrfs_start_transaction(root, 5);
6590 return PTR_ERR(trans);
6592 err = btrfs_find_free_ino(root, &objectid);
6596 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6597 dentry->d_name.len, btrfs_ino(dir), objectid,
6598 S_IFDIR | mode, &index);
6599 if (IS_ERR(inode)) {
6600 err = PTR_ERR(inode);
6605 /* these must be set before we unlock the inode */
6606 inode->i_op = &btrfs_dir_inode_operations;
6607 inode->i_fop = &btrfs_dir_file_operations;
6609 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6611 goto out_fail_inode;
6613 btrfs_i_size_write(inode, 0);
6614 err = btrfs_update_inode(trans, root, inode);
6616 goto out_fail_inode;
6618 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6619 dentry->d_name.len, 0, index);
6621 goto out_fail_inode;
6623 d_instantiate(dentry, inode);
6625 * mkdir is special. We're unlocking after we call d_instantiate
6626 * to avoid a race with nfsd calling d_instantiate.
6628 unlock_new_inode(inode);
6632 btrfs_end_transaction(trans, root);
6634 inode_dec_link_count(inode);
6637 btrfs_balance_delayed_items(root);
6638 btrfs_btree_balance_dirty(root);
6642 unlock_new_inode(inode);
6646 /* Find next extent map of a given extent map, caller needs to ensure locks */
6647 static struct extent_map *next_extent_map(struct extent_map *em)
6649 struct rb_node *next;
6651 next = rb_next(&em->rb_node);
6654 return container_of(next, struct extent_map, rb_node);
6657 static struct extent_map *prev_extent_map(struct extent_map *em)
6659 struct rb_node *prev;
6661 prev = rb_prev(&em->rb_node);
6664 return container_of(prev, struct extent_map, rb_node);
6667 /* helper for btfs_get_extent. Given an existing extent in the tree,
6668 * the existing extent is the nearest extent to map_start,
6669 * and an extent that you want to insert, deal with overlap and insert
6670 * the best fitted new extent into the tree.
6672 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6673 struct extent_map *existing,
6674 struct extent_map *em,
6677 struct extent_map *prev;
6678 struct extent_map *next;
6683 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6685 if (existing->start > map_start) {
6687 prev = prev_extent_map(next);
6690 next = next_extent_map(prev);
6693 start = prev ? extent_map_end(prev) : em->start;
6694 start = max_t(u64, start, em->start);
6695 end = next ? next->start : extent_map_end(em);
6696 end = min_t(u64, end, extent_map_end(em));
6697 start_diff = start - em->start;
6699 em->len = end - start;
6700 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6701 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6702 em->block_start += start_diff;
6703 em->block_len -= start_diff;
6705 return add_extent_mapping(em_tree, em, 0);
6708 static noinline int uncompress_inline(struct btrfs_path *path,
6709 struct inode *inode, struct page *page,
6710 size_t pg_offset, u64 extent_offset,
6711 struct btrfs_file_extent_item *item)
6714 struct extent_buffer *leaf = path->nodes[0];
6717 unsigned long inline_size;
6721 WARN_ON(pg_offset != 0);
6722 compress_type = btrfs_file_extent_compression(leaf, item);
6723 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6724 inline_size = btrfs_file_extent_inline_item_len(leaf,
6725 btrfs_item_nr(path->slots[0]));
6726 tmp = kmalloc(inline_size, GFP_NOFS);
6729 ptr = btrfs_file_extent_inline_start(item);
6731 read_extent_buffer(leaf, tmp, ptr, inline_size);
6733 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6734 ret = btrfs_decompress(compress_type, tmp, page,
6735 extent_offset, inline_size, max_size);
6741 * a bit scary, this does extent mapping from logical file offset to the disk.
6742 * the ugly parts come from merging extents from the disk with the in-ram
6743 * representation. This gets more complex because of the data=ordered code,
6744 * where the in-ram extents might be locked pending data=ordered completion.
6746 * This also copies inline extents directly into the page.
6749 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6750 size_t pg_offset, u64 start, u64 len,
6755 u64 extent_start = 0;
6757 u64 objectid = btrfs_ino(inode);
6759 struct btrfs_path *path = NULL;
6760 struct btrfs_root *root = BTRFS_I(inode)->root;
6761 struct btrfs_file_extent_item *item;
6762 struct extent_buffer *leaf;
6763 struct btrfs_key found_key;
6764 struct extent_map *em = NULL;
6765 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6766 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6767 struct btrfs_trans_handle *trans = NULL;
6768 const bool new_inline = !page || create;
6771 read_lock(&em_tree->lock);
6772 em = lookup_extent_mapping(em_tree, start, len);
6774 em->bdev = root->fs_info->fs_devices->latest_bdev;
6775 read_unlock(&em_tree->lock);
6778 if (em->start > start || em->start + em->len <= start)
6779 free_extent_map(em);
6780 else if (em->block_start == EXTENT_MAP_INLINE && page)
6781 free_extent_map(em);
6785 em = alloc_extent_map();
6790 em->bdev = root->fs_info->fs_devices->latest_bdev;
6791 em->start = EXTENT_MAP_HOLE;
6792 em->orig_start = EXTENT_MAP_HOLE;
6794 em->block_len = (u64)-1;
6797 path = btrfs_alloc_path();
6803 * Chances are we'll be called again, so go ahead and do
6809 ret = btrfs_lookup_file_extent(trans, root, path,
6810 objectid, start, trans != NULL);
6817 if (path->slots[0] == 0)
6822 leaf = path->nodes[0];
6823 item = btrfs_item_ptr(leaf, path->slots[0],
6824 struct btrfs_file_extent_item);
6825 /* are we inside the extent that was found? */
6826 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6827 found_type = found_key.type;
6828 if (found_key.objectid != objectid ||
6829 found_type != BTRFS_EXTENT_DATA_KEY) {
6831 * If we backup past the first extent we want to move forward
6832 * and see if there is an extent in front of us, otherwise we'll
6833 * say there is a hole for our whole search range which can
6840 found_type = btrfs_file_extent_type(leaf, item);
6841 extent_start = found_key.offset;
6842 if (found_type == BTRFS_FILE_EXTENT_REG ||
6843 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6844 extent_end = extent_start +
6845 btrfs_file_extent_num_bytes(leaf, item);
6846 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6848 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6849 extent_end = ALIGN(extent_start + size, root->sectorsize);
6852 if (start >= extent_end) {
6854 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6855 ret = btrfs_next_leaf(root, path);
6862 leaf = path->nodes[0];
6864 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6865 if (found_key.objectid != objectid ||
6866 found_key.type != BTRFS_EXTENT_DATA_KEY)
6868 if (start + len <= found_key.offset)
6870 if (start > found_key.offset)
6873 em->orig_start = start;
6874 em->len = found_key.offset - start;
6878 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6880 if (found_type == BTRFS_FILE_EXTENT_REG ||
6881 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6883 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6887 size_t extent_offset;
6893 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6894 extent_offset = page_offset(page) + pg_offset - extent_start;
6895 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6896 size - extent_offset);
6897 em->start = extent_start + extent_offset;
6898 em->len = ALIGN(copy_size, root->sectorsize);
6899 em->orig_block_len = em->len;
6900 em->orig_start = em->start;
6901 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6902 if (create == 0 && !PageUptodate(page)) {
6903 if (btrfs_file_extent_compression(leaf, item) !=
6904 BTRFS_COMPRESS_NONE) {
6905 ret = uncompress_inline(path, inode, page,
6907 extent_offset, item);
6914 read_extent_buffer(leaf, map + pg_offset, ptr,
6916 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6917 memset(map + pg_offset + copy_size, 0,
6918 PAGE_CACHE_SIZE - pg_offset -
6923 flush_dcache_page(page);
6924 } else if (create && PageUptodate(page)) {
6928 free_extent_map(em);
6931 btrfs_release_path(path);
6932 trans = btrfs_join_transaction(root);
6935 return ERR_CAST(trans);
6939 write_extent_buffer(leaf, map + pg_offset, ptr,
6942 btrfs_mark_buffer_dirty(leaf);
6944 set_extent_uptodate(io_tree, em->start,
6945 extent_map_end(em) - 1, NULL, GFP_NOFS);
6950 em->orig_start = start;
6953 em->block_start = EXTENT_MAP_HOLE;
6954 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6956 btrfs_release_path(path);
6957 if (em->start > start || extent_map_end(em) <= start) {
6958 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6959 em->start, em->len, start, len);
6965 write_lock(&em_tree->lock);
6966 ret = add_extent_mapping(em_tree, em, 0);
6967 /* it is possible that someone inserted the extent into the tree
6968 * while we had the lock dropped. It is also possible that
6969 * an overlapping map exists in the tree
6971 if (ret == -EEXIST) {
6972 struct extent_map *existing;
6976 existing = search_extent_mapping(em_tree, start, len);
6978 * existing will always be non-NULL, since there must be
6979 * extent causing the -EEXIST.
6981 if (start >= extent_map_end(existing) ||
6982 start <= existing->start) {
6984 * The existing extent map is the one nearest to
6985 * the [start, start + len) range which overlaps
6987 err = merge_extent_mapping(em_tree, existing,
6989 free_extent_map(existing);
6991 free_extent_map(em);
6995 free_extent_map(em);
7000 write_unlock(&em_tree->lock);
7003 trace_btrfs_get_extent(root, em);
7005 btrfs_free_path(path);
7007 ret = btrfs_end_transaction(trans, root);
7012 free_extent_map(em);
7013 return ERR_PTR(err);
7015 BUG_ON(!em); /* Error is always set */
7019 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7020 size_t pg_offset, u64 start, u64 len,
7023 struct extent_map *em;
7024 struct extent_map *hole_em = NULL;
7025 u64 range_start = start;
7031 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7038 * - a pre-alloc extent,
7039 * there might actually be delalloc bytes behind it.
7041 if (em->block_start != EXTENT_MAP_HOLE &&
7042 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7048 /* check to see if we've wrapped (len == -1 or similar) */
7057 /* ok, we didn't find anything, lets look for delalloc */
7058 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7059 end, len, EXTENT_DELALLOC, 1);
7060 found_end = range_start + found;
7061 if (found_end < range_start)
7062 found_end = (u64)-1;
7065 * we didn't find anything useful, return
7066 * the original results from get_extent()
7068 if (range_start > end || found_end <= start) {
7074 /* adjust the range_start to make sure it doesn't
7075 * go backwards from the start they passed in
7077 range_start = max(start, range_start);
7078 found = found_end - range_start;
7081 u64 hole_start = start;
7084 em = alloc_extent_map();
7090 * when btrfs_get_extent can't find anything it
7091 * returns one huge hole
7093 * make sure what it found really fits our range, and
7094 * adjust to make sure it is based on the start from
7098 u64 calc_end = extent_map_end(hole_em);
7100 if (calc_end <= start || (hole_em->start > end)) {
7101 free_extent_map(hole_em);
7104 hole_start = max(hole_em->start, start);
7105 hole_len = calc_end - hole_start;
7109 if (hole_em && range_start > hole_start) {
7110 /* our hole starts before our delalloc, so we
7111 * have to return just the parts of the hole
7112 * that go until the delalloc starts
7114 em->len = min(hole_len,
7115 range_start - hole_start);
7116 em->start = hole_start;
7117 em->orig_start = hole_start;
7119 * don't adjust block start at all,
7120 * it is fixed at EXTENT_MAP_HOLE
7122 em->block_start = hole_em->block_start;
7123 em->block_len = hole_len;
7124 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7125 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7127 em->start = range_start;
7129 em->orig_start = range_start;
7130 em->block_start = EXTENT_MAP_DELALLOC;
7131 em->block_len = found;
7133 } else if (hole_em) {
7138 free_extent_map(hole_em);
7140 free_extent_map(em);
7141 return ERR_PTR(err);
7146 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7149 struct btrfs_root *root = BTRFS_I(inode)->root;
7150 struct extent_map *em;
7151 struct btrfs_key ins;
7155 alloc_hint = get_extent_allocation_hint(inode, start, len);
7156 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7157 alloc_hint, &ins, 1, 1);
7159 return ERR_PTR(ret);
7161 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7162 ins.offset, ins.offset, ins.offset, 0);
7164 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7168 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7169 ins.offset, ins.offset, 0);
7171 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7172 free_extent_map(em);
7173 return ERR_PTR(ret);
7180 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7181 * block must be cow'd
7183 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7184 u64 *orig_start, u64 *orig_block_len,
7187 struct btrfs_trans_handle *trans;
7188 struct btrfs_path *path;
7190 struct extent_buffer *leaf;
7191 struct btrfs_root *root = BTRFS_I(inode)->root;
7192 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7193 struct btrfs_file_extent_item *fi;
7194 struct btrfs_key key;
7201 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7203 path = btrfs_alloc_path();
7207 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7212 slot = path->slots[0];
7215 /* can't find the item, must cow */
7222 leaf = path->nodes[0];
7223 btrfs_item_key_to_cpu(leaf, &key, slot);
7224 if (key.objectid != btrfs_ino(inode) ||
7225 key.type != BTRFS_EXTENT_DATA_KEY) {
7226 /* not our file or wrong item type, must cow */
7230 if (key.offset > offset) {
7231 /* Wrong offset, must cow */
7235 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7236 found_type = btrfs_file_extent_type(leaf, fi);
7237 if (found_type != BTRFS_FILE_EXTENT_REG &&
7238 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7239 /* not a regular extent, must cow */
7243 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7246 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7247 if (extent_end <= offset)
7250 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7251 if (disk_bytenr == 0)
7254 if (btrfs_file_extent_compression(leaf, fi) ||
7255 btrfs_file_extent_encryption(leaf, fi) ||
7256 btrfs_file_extent_other_encoding(leaf, fi))
7259 backref_offset = btrfs_file_extent_offset(leaf, fi);
7262 *orig_start = key.offset - backref_offset;
7263 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7264 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7267 if (btrfs_extent_readonly(root, disk_bytenr))
7270 num_bytes = min(offset + *len, extent_end) - offset;
7271 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7274 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7275 ret = test_range_bit(io_tree, offset, range_end,
7276 EXTENT_DELALLOC, 0, NULL);
7283 btrfs_release_path(path);
7286 * look for other files referencing this extent, if we
7287 * find any we must cow
7289 trans = btrfs_join_transaction(root);
7290 if (IS_ERR(trans)) {
7295 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7296 key.offset - backref_offset, disk_bytenr);
7297 btrfs_end_transaction(trans, root);
7304 * adjust disk_bytenr and num_bytes to cover just the bytes
7305 * in this extent we are about to write. If there
7306 * are any csums in that range we have to cow in order
7307 * to keep the csums correct
7309 disk_bytenr += backref_offset;
7310 disk_bytenr += offset - key.offset;
7311 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7314 * all of the above have passed, it is safe to overwrite this extent
7320 btrfs_free_path(path);
7324 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7326 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7328 void **pagep = NULL;
7329 struct page *page = NULL;
7333 start_idx = start >> PAGE_CACHE_SHIFT;
7336 * end is the last byte in the last page. end == start is legal
7338 end_idx = end >> PAGE_CACHE_SHIFT;
7342 /* Most of the code in this while loop is lifted from
7343 * find_get_page. It's been modified to begin searching from a
7344 * page and return just the first page found in that range. If the
7345 * found idx is less than or equal to the end idx then we know that
7346 * a page exists. If no pages are found or if those pages are
7347 * outside of the range then we're fine (yay!) */
7348 while (page == NULL &&
7349 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7350 page = radix_tree_deref_slot(pagep);
7351 if (unlikely(!page))
7354 if (radix_tree_exception(page)) {
7355 if (radix_tree_deref_retry(page)) {
7360 * Otherwise, shmem/tmpfs must be storing a swap entry
7361 * here as an exceptional entry: so return it without
7362 * attempting to raise page count.
7365 break; /* TODO: Is this relevant for this use case? */
7368 if (!page_cache_get_speculative(page)) {
7374 * Has the page moved?
7375 * This is part of the lockless pagecache protocol. See
7376 * include/linux/pagemap.h for details.
7378 if (unlikely(page != *pagep)) {
7379 page_cache_release(page);
7385 if (page->index <= end_idx)
7387 page_cache_release(page);
7394 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7395 struct extent_state **cached_state, int writing)
7397 struct btrfs_ordered_extent *ordered;
7401 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7404 * We're concerned with the entire range that we're going to be
7405 * doing DIO to, so we need to make sure theres no ordered
7406 * extents in this range.
7408 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7409 lockend - lockstart + 1);
7412 * We need to make sure there are no buffered pages in this
7413 * range either, we could have raced between the invalidate in
7414 * generic_file_direct_write and locking the extent. The
7415 * invalidate needs to happen so that reads after a write do not
7420 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7423 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7424 cached_state, GFP_NOFS);
7427 btrfs_start_ordered_extent(inode, ordered, 1);
7428 btrfs_put_ordered_extent(ordered);
7430 /* Screw you mmap */
7431 ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7434 ret = filemap_fdatawait_range(inode->i_mapping,
7441 * If we found a page that couldn't be invalidated just
7442 * fall back to buffered.
7444 ret = invalidate_inode_pages2_range(inode->i_mapping,
7445 lockstart >> PAGE_CACHE_SHIFT,
7446 lockend >> PAGE_CACHE_SHIFT);
7457 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7458 u64 len, u64 orig_start,
7459 u64 block_start, u64 block_len,
7460 u64 orig_block_len, u64 ram_bytes,
7463 struct extent_map_tree *em_tree;
7464 struct extent_map *em;
7465 struct btrfs_root *root = BTRFS_I(inode)->root;
7468 em_tree = &BTRFS_I(inode)->extent_tree;
7469 em = alloc_extent_map();
7471 return ERR_PTR(-ENOMEM);
7474 em->orig_start = orig_start;
7475 em->mod_start = start;
7478 em->block_len = block_len;
7479 em->block_start = block_start;
7480 em->bdev = root->fs_info->fs_devices->latest_bdev;
7481 em->orig_block_len = orig_block_len;
7482 em->ram_bytes = ram_bytes;
7483 em->generation = -1;
7484 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7485 if (type == BTRFS_ORDERED_PREALLOC)
7486 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7489 btrfs_drop_extent_cache(inode, em->start,
7490 em->start + em->len - 1, 0);
7491 write_lock(&em_tree->lock);
7492 ret = add_extent_mapping(em_tree, em, 1);
7493 write_unlock(&em_tree->lock);
7494 } while (ret == -EEXIST);
7497 free_extent_map(em);
7498 return ERR_PTR(ret);
7504 struct btrfs_dio_data {
7505 u64 outstanding_extents;
7509 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7510 struct buffer_head *bh_result, int create)
7512 struct extent_map *em;
7513 struct btrfs_root *root = BTRFS_I(inode)->root;
7514 struct extent_state *cached_state = NULL;
7515 struct btrfs_dio_data *dio_data = NULL;
7516 u64 start = iblock << inode->i_blkbits;
7517 u64 lockstart, lockend;
7518 u64 len = bh_result->b_size;
7519 int unlock_bits = EXTENT_LOCKED;
7523 unlock_bits |= EXTENT_DIRTY;
7525 len = min_t(u64, len, root->sectorsize);
7528 lockend = start + len - 1;
7530 if (current->journal_info) {
7532 * Need to pull our outstanding extents and set journal_info to NULL so
7533 * that anything that needs to check if there's a transction doesn't get
7536 dio_data = current->journal_info;
7537 current->journal_info = NULL;
7541 * If this errors out it's because we couldn't invalidate pagecache for
7542 * this range and we need to fallback to buffered.
7544 if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7547 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7554 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7555 * io. INLINE is special, and we could probably kludge it in here, but
7556 * it's still buffered so for safety lets just fall back to the generic
7559 * For COMPRESSED we _have_ to read the entire extent in so we can
7560 * decompress it, so there will be buffering required no matter what we
7561 * do, so go ahead and fallback to buffered.
7563 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7564 * to buffered IO. Don't blame me, this is the price we pay for using
7567 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7568 em->block_start == EXTENT_MAP_INLINE) {
7569 free_extent_map(em);
7574 /* Just a good old fashioned hole, return */
7575 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7576 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7577 free_extent_map(em);
7582 * We don't allocate a new extent in the following cases
7584 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7586 * 2) The extent is marked as PREALLOC. We're good to go here and can
7587 * just use the extent.
7591 len = min(len, em->len - (start - em->start));
7592 lockstart = start + len;
7596 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7597 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7598 em->block_start != EXTENT_MAP_HOLE)) {
7600 u64 block_start, orig_start, orig_block_len, ram_bytes;
7602 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7603 type = BTRFS_ORDERED_PREALLOC;
7605 type = BTRFS_ORDERED_NOCOW;
7606 len = min(len, em->len - (start - em->start));
7607 block_start = em->block_start + (start - em->start);
7609 if (can_nocow_extent(inode, start, &len, &orig_start,
7610 &orig_block_len, &ram_bytes) == 1) {
7611 if (type == BTRFS_ORDERED_PREALLOC) {
7612 free_extent_map(em);
7613 em = create_pinned_em(inode, start, len,
7624 ret = btrfs_add_ordered_extent_dio(inode, start,
7625 block_start, len, len, type);
7627 free_extent_map(em);
7635 * this will cow the extent, reset the len in case we changed
7638 len = bh_result->b_size;
7639 free_extent_map(em);
7640 em = btrfs_new_extent_direct(inode, start, len);
7645 len = min(len, em->len - (start - em->start));
7647 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7649 bh_result->b_size = len;
7650 bh_result->b_bdev = em->bdev;
7651 set_buffer_mapped(bh_result);
7653 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7654 set_buffer_new(bh_result);
7657 * Need to update the i_size under the extent lock so buffered
7658 * readers will get the updated i_size when we unlock.
7660 if (start + len > i_size_read(inode))
7661 i_size_write(inode, start + len);
7664 * If we have an outstanding_extents count still set then we're
7665 * within our reservation, otherwise we need to adjust our inode
7666 * counter appropriately.
7668 if (dio_data->outstanding_extents) {
7669 (dio_data->outstanding_extents)--;
7671 spin_lock(&BTRFS_I(inode)->lock);
7672 BTRFS_I(inode)->outstanding_extents++;
7673 spin_unlock(&BTRFS_I(inode)->lock);
7676 btrfs_free_reserved_data_space(inode, start, len);
7677 WARN_ON(dio_data->reserve < len);
7678 dio_data->reserve -= len;
7679 current->journal_info = dio_data;
7683 * In the case of write we need to clear and unlock the entire range,
7684 * in the case of read we need to unlock only the end area that we
7685 * aren't using if there is any left over space.
7687 if (lockstart < lockend) {
7688 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7689 lockend, unlock_bits, 1, 0,
7690 &cached_state, GFP_NOFS);
7692 free_extent_state(cached_state);
7695 free_extent_map(em);
7700 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7701 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7703 current->journal_info = dio_data;
7707 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7708 int rw, int mirror_num)
7710 struct btrfs_root *root = BTRFS_I(inode)->root;
7713 BUG_ON(rw & REQ_WRITE);
7717 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7718 BTRFS_WQ_ENDIO_DIO_REPAIR);
7722 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7728 static int btrfs_check_dio_repairable(struct inode *inode,
7729 struct bio *failed_bio,
7730 struct io_failure_record *failrec,
7735 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7736 failrec->logical, failrec->len);
7737 if (num_copies == 1) {
7739 * we only have a single copy of the data, so don't bother with
7740 * all the retry and error correction code that follows. no
7741 * matter what the error is, it is very likely to persist.
7743 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7744 num_copies, failrec->this_mirror, failed_mirror);
7748 failrec->failed_mirror = failed_mirror;
7749 failrec->this_mirror++;
7750 if (failrec->this_mirror == failed_mirror)
7751 failrec->this_mirror++;
7753 if (failrec->this_mirror > num_copies) {
7754 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7755 num_copies, failrec->this_mirror, failed_mirror);
7762 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7763 struct page *page, u64 start, u64 end,
7764 int failed_mirror, bio_end_io_t *repair_endio,
7767 struct io_failure_record *failrec;
7773 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7775 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7779 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7782 free_io_failure(inode, failrec);
7786 if (failed_bio->bi_vcnt > 1)
7787 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7789 read_mode = READ_SYNC;
7791 isector = start - btrfs_io_bio(failed_bio)->logical;
7792 isector >>= inode->i_sb->s_blocksize_bits;
7793 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7794 0, isector, repair_endio, repair_arg);
7796 free_io_failure(inode, failrec);
7800 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7801 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7802 read_mode, failrec->this_mirror, failrec->in_validation);
7804 ret = submit_dio_repair_bio(inode, bio, read_mode,
7805 failrec->this_mirror);
7807 free_io_failure(inode, failrec);
7814 struct btrfs_retry_complete {
7815 struct completion done;
7816 struct inode *inode;
7821 static void btrfs_retry_endio_nocsum(struct bio *bio)
7823 struct btrfs_retry_complete *done = bio->bi_private;
7824 struct bio_vec *bvec;
7831 bio_for_each_segment_all(bvec, bio, i)
7832 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7834 complete(&done->done);
7838 static int __btrfs_correct_data_nocsum(struct inode *inode,
7839 struct btrfs_io_bio *io_bio)
7841 struct bio_vec *bvec;
7842 struct btrfs_retry_complete done;
7847 start = io_bio->logical;
7850 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7854 init_completion(&done.done);
7856 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7857 start + bvec->bv_len - 1,
7859 btrfs_retry_endio_nocsum, &done);
7863 wait_for_completion(&done.done);
7865 if (!done.uptodate) {
7866 /* We might have another mirror, so try again */
7870 start += bvec->bv_len;
7876 static void btrfs_retry_endio(struct bio *bio)
7878 struct btrfs_retry_complete *done = bio->bi_private;
7879 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7880 struct bio_vec *bvec;
7889 bio_for_each_segment_all(bvec, bio, i) {
7890 ret = __readpage_endio_check(done->inode, io_bio, i,
7892 done->start, bvec->bv_len);
7894 clean_io_failure(done->inode, done->start,
7900 done->uptodate = uptodate;
7902 complete(&done->done);
7906 static int __btrfs_subio_endio_read(struct inode *inode,
7907 struct btrfs_io_bio *io_bio, int err)
7909 struct bio_vec *bvec;
7910 struct btrfs_retry_complete done;
7917 start = io_bio->logical;
7920 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7921 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7922 0, start, bvec->bv_len);
7928 init_completion(&done.done);
7930 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7931 start + bvec->bv_len - 1,
7933 btrfs_retry_endio, &done);
7939 wait_for_completion(&done.done);
7941 if (!done.uptodate) {
7942 /* We might have another mirror, so try again */
7946 offset += bvec->bv_len;
7947 start += bvec->bv_len;
7953 static int btrfs_subio_endio_read(struct inode *inode,
7954 struct btrfs_io_bio *io_bio, int err)
7956 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7960 return __btrfs_correct_data_nocsum(inode, io_bio);
7964 return __btrfs_subio_endio_read(inode, io_bio, err);
7968 static void btrfs_endio_direct_read(struct bio *bio)
7970 struct btrfs_dio_private *dip = bio->bi_private;
7971 struct inode *inode = dip->inode;
7972 struct bio *dio_bio;
7973 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7974 int err = bio->bi_error;
7976 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7977 err = btrfs_subio_endio_read(inode, io_bio, err);
7979 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7980 dip->logical_offset + dip->bytes - 1);
7981 dio_bio = dip->dio_bio;
7985 dio_end_io(dio_bio, bio->bi_error);
7988 io_bio->end_io(io_bio, err);
7992 static void btrfs_endio_direct_write(struct bio *bio)
7994 struct btrfs_dio_private *dip = bio->bi_private;
7995 struct inode *inode = dip->inode;
7996 struct btrfs_root *root = BTRFS_I(inode)->root;
7997 struct btrfs_ordered_extent *ordered = NULL;
7998 u64 ordered_offset = dip->logical_offset;
7999 u64 ordered_bytes = dip->bytes;
8000 struct bio *dio_bio;
8004 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8011 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8012 finish_ordered_fn, NULL, NULL);
8013 btrfs_queue_work(root->fs_info->endio_write_workers,
8017 * our bio might span multiple ordered extents. If we haven't
8018 * completed the accounting for the whole dio, go back and try again
8020 if (ordered_offset < dip->logical_offset + dip->bytes) {
8021 ordered_bytes = dip->logical_offset + dip->bytes -
8026 dio_bio = dip->dio_bio;
8030 dio_end_io(dio_bio, bio->bi_error);
8034 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8035 struct bio *bio, int mirror_num,
8036 unsigned long bio_flags, u64 offset)
8039 struct btrfs_root *root = BTRFS_I(inode)->root;
8040 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8041 BUG_ON(ret); /* -ENOMEM */
8045 static void btrfs_end_dio_bio(struct bio *bio)
8047 struct btrfs_dio_private *dip = bio->bi_private;
8048 int err = bio->bi_error;
8051 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8052 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8053 btrfs_ino(dip->inode), bio->bi_rw,
8054 (unsigned long long)bio->bi_iter.bi_sector,
8055 bio->bi_iter.bi_size, err);
8057 if (dip->subio_endio)
8058 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8064 * before atomic variable goto zero, we must make sure
8065 * dip->errors is perceived to be set.
8067 smp_mb__before_atomic();
8070 /* if there are more bios still pending for this dio, just exit */
8071 if (!atomic_dec_and_test(&dip->pending_bios))
8075 bio_io_error(dip->orig_bio);
8077 dip->dio_bio->bi_error = 0;
8078 bio_endio(dip->orig_bio);
8084 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8085 u64 first_sector, gfp_t gfp_flags)
8088 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8090 bio_associate_current(bio);
8094 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8095 struct inode *inode,
8096 struct btrfs_dio_private *dip,
8100 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8101 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8105 * We load all the csum data we need when we submit
8106 * the first bio to reduce the csum tree search and
8109 if (dip->logical_offset == file_offset) {
8110 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8116 if (bio == dip->orig_bio)
8119 file_offset -= dip->logical_offset;
8120 file_offset >>= inode->i_sb->s_blocksize_bits;
8121 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8126 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8127 int rw, u64 file_offset, int skip_sum,
8130 struct btrfs_dio_private *dip = bio->bi_private;
8131 int write = rw & REQ_WRITE;
8132 struct btrfs_root *root = BTRFS_I(inode)->root;
8136 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8141 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8142 BTRFS_WQ_ENDIO_DATA);
8150 if (write && async_submit) {
8151 ret = btrfs_wq_submit_bio(root->fs_info,
8152 inode, rw, bio, 0, 0,
8154 __btrfs_submit_bio_start_direct_io,
8155 __btrfs_submit_bio_done);
8159 * If we aren't doing async submit, calculate the csum of the
8162 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8166 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8172 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8178 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8181 struct inode *inode = dip->inode;
8182 struct btrfs_root *root = BTRFS_I(inode)->root;
8184 struct bio *orig_bio = dip->orig_bio;
8185 struct bio_vec *bvec = orig_bio->bi_io_vec;
8186 u64 start_sector = orig_bio->bi_iter.bi_sector;
8187 u64 file_offset = dip->logical_offset;
8192 int async_submit = 0;
8194 map_length = orig_bio->bi_iter.bi_size;
8195 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8196 &map_length, NULL, 0);
8200 if (map_length >= orig_bio->bi_iter.bi_size) {
8202 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8206 /* async crcs make it difficult to collect full stripe writes. */
8207 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8212 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8216 bio->bi_private = dip;
8217 bio->bi_end_io = btrfs_end_dio_bio;
8218 btrfs_io_bio(bio)->logical = file_offset;
8219 atomic_inc(&dip->pending_bios);
8221 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8222 if (map_length < submit_len + bvec->bv_len ||
8223 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8224 bvec->bv_offset) < bvec->bv_len) {
8226 * inc the count before we submit the bio so
8227 * we know the end IO handler won't happen before
8228 * we inc the count. Otherwise, the dip might get freed
8229 * before we're done setting it up
8231 atomic_inc(&dip->pending_bios);
8232 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8233 file_offset, skip_sum,
8237 atomic_dec(&dip->pending_bios);
8241 start_sector += submit_len >> 9;
8242 file_offset += submit_len;
8247 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8248 start_sector, GFP_NOFS);
8251 bio->bi_private = dip;
8252 bio->bi_end_io = btrfs_end_dio_bio;
8253 btrfs_io_bio(bio)->logical = file_offset;
8255 map_length = orig_bio->bi_iter.bi_size;
8256 ret = btrfs_map_block(root->fs_info, rw,
8258 &map_length, NULL, 0);
8264 submit_len += bvec->bv_len;
8271 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8280 * before atomic variable goto zero, we must
8281 * make sure dip->errors is perceived to be set.
8283 smp_mb__before_atomic();
8284 if (atomic_dec_and_test(&dip->pending_bios))
8285 bio_io_error(dip->orig_bio);
8287 /* bio_end_io() will handle error, so we needn't return it */
8291 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8292 struct inode *inode, loff_t file_offset)
8294 struct btrfs_dio_private *dip = NULL;
8295 struct bio *io_bio = NULL;
8296 struct btrfs_io_bio *btrfs_bio;
8298 int write = rw & REQ_WRITE;
8301 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8303 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8309 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8315 dip->private = dio_bio->bi_private;
8317 dip->logical_offset = file_offset;
8318 dip->bytes = dio_bio->bi_iter.bi_size;
8319 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8320 io_bio->bi_private = dip;
8321 dip->orig_bio = io_bio;
8322 dip->dio_bio = dio_bio;
8323 atomic_set(&dip->pending_bios, 0);
8324 btrfs_bio = btrfs_io_bio(io_bio);
8325 btrfs_bio->logical = file_offset;
8328 io_bio->bi_end_io = btrfs_endio_direct_write;
8330 io_bio->bi_end_io = btrfs_endio_direct_read;
8331 dip->subio_endio = btrfs_subio_endio_read;
8334 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8338 if (btrfs_bio->end_io)
8339 btrfs_bio->end_io(btrfs_bio, ret);
8343 * If we arrived here it means either we failed to submit the dip
8344 * or we either failed to clone the dio_bio or failed to allocate the
8345 * dip. If we cloned the dio_bio and allocated the dip, we can just
8346 * call bio_endio against our io_bio so that we get proper resource
8347 * cleanup if we fail to submit the dip, otherwise, we must do the
8348 * same as btrfs_endio_direct_[write|read] because we can't call these
8349 * callbacks - they require an allocated dip and a clone of dio_bio.
8351 if (io_bio && dip) {
8352 io_bio->bi_error = -EIO;
8355 * The end io callbacks free our dip, do the final put on io_bio
8356 * and all the cleanup and final put for dio_bio (through
8363 struct btrfs_ordered_extent *ordered;
8365 ordered = btrfs_lookup_ordered_extent(inode,
8367 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
8369 * Decrements our ref on the ordered extent and removes
8370 * the ordered extent from the inode's ordered tree,
8371 * doing all the proper resource cleanup such as for the
8372 * reserved space and waking up any waiters for this
8373 * ordered extent (through btrfs_remove_ordered_extent).
8375 btrfs_finish_ordered_io(ordered);
8377 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8378 file_offset + dio_bio->bi_iter.bi_size - 1);
8380 dio_bio->bi_error = -EIO;
8382 * Releases and cleans up our dio_bio, no need to bio_put()
8383 * nor bio_endio()/bio_io_error() against dio_bio.
8385 dio_end_io(dio_bio, ret);
8392 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8393 const struct iov_iter *iter, loff_t offset)
8397 unsigned blocksize_mask = root->sectorsize - 1;
8398 ssize_t retval = -EINVAL;
8400 if (offset & blocksize_mask)
8403 if (iov_iter_alignment(iter) & blocksize_mask)
8406 /* If this is a write we don't need to check anymore */
8407 if (iov_iter_rw(iter) == WRITE)
8410 * Check to make sure we don't have duplicate iov_base's in this
8411 * iovec, if so return EINVAL, otherwise we'll get csum errors
8412 * when reading back.
8414 for (seg = 0; seg < iter->nr_segs; seg++) {
8415 for (i = seg + 1; i < iter->nr_segs; i++) {
8416 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8425 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8428 struct file *file = iocb->ki_filp;
8429 struct inode *inode = file->f_mapping->host;
8430 struct btrfs_root *root = BTRFS_I(inode)->root;
8431 struct btrfs_dio_data dio_data = { 0 };
8435 bool relock = false;
8438 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8441 inode_dio_begin(inode);
8442 smp_mb__after_atomic();
8445 * The generic stuff only does filemap_write_and_wait_range, which
8446 * isn't enough if we've written compressed pages to this area, so
8447 * we need to flush the dirty pages again to make absolutely sure
8448 * that any outstanding dirty pages are on disk.
8450 count = iov_iter_count(iter);
8451 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8452 &BTRFS_I(inode)->runtime_flags))
8453 filemap_fdatawrite_range(inode->i_mapping, offset,
8454 offset + count - 1);
8456 if (iov_iter_rw(iter) == WRITE) {
8458 * If the write DIO is beyond the EOF, we need update
8459 * the isize, but it is protected by i_mutex. So we can
8460 * not unlock the i_mutex at this case.
8462 if (offset + count <= inode->i_size) {
8463 mutex_unlock(&inode->i_mutex);
8466 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8469 dio_data.outstanding_extents = div64_u64(count +
8470 BTRFS_MAX_EXTENT_SIZE - 1,
8471 BTRFS_MAX_EXTENT_SIZE);
8474 * We need to know how many extents we reserved so that we can
8475 * do the accounting properly if we go over the number we
8476 * originally calculated. Abuse current->journal_info for this.
8478 dio_data.reserve = round_up(count, root->sectorsize);
8479 current->journal_info = &dio_data;
8480 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8481 &BTRFS_I(inode)->runtime_flags)) {
8482 inode_dio_end(inode);
8483 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8487 ret = __blockdev_direct_IO(iocb, inode,
8488 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8489 iter, offset, btrfs_get_blocks_direct, NULL,
8490 btrfs_submit_direct, flags);
8491 if (iov_iter_rw(iter) == WRITE) {
8492 current->journal_info = NULL;
8493 if (ret < 0 && ret != -EIOCBQUEUED) {
8494 if (dio_data.reserve)
8495 btrfs_delalloc_release_space(inode, offset,
8497 } else if (ret >= 0 && (size_t)ret < count)
8498 btrfs_delalloc_release_space(inode, offset,
8499 count - (size_t)ret);
8503 inode_dio_end(inode);
8505 mutex_lock(&inode->i_mutex);
8510 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8512 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8513 __u64 start, __u64 len)
8517 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8521 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8524 int btrfs_readpage(struct file *file, struct page *page)
8526 struct extent_io_tree *tree;
8527 tree = &BTRFS_I(page->mapping->host)->io_tree;
8528 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8531 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8533 struct extent_io_tree *tree;
8536 if (current->flags & PF_MEMALLOC) {
8537 redirty_page_for_writepage(wbc, page);
8541 tree = &BTRFS_I(page->mapping->host)->io_tree;
8542 return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8545 static int btrfs_writepages(struct address_space *mapping,
8546 struct writeback_control *wbc)
8548 struct extent_io_tree *tree;
8550 tree = &BTRFS_I(mapping->host)->io_tree;
8551 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8555 btrfs_readpages(struct file *file, struct address_space *mapping,
8556 struct list_head *pages, unsigned nr_pages)
8558 struct extent_io_tree *tree;
8559 tree = &BTRFS_I(mapping->host)->io_tree;
8560 return extent_readpages(tree, mapping, pages, nr_pages,
8563 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8565 struct extent_io_tree *tree;
8566 struct extent_map_tree *map;
8569 tree = &BTRFS_I(page->mapping->host)->io_tree;
8570 map = &BTRFS_I(page->mapping->host)->extent_tree;
8571 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8573 ClearPagePrivate(page);
8574 set_page_private(page, 0);
8575 page_cache_release(page);
8580 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8582 if (PageWriteback(page) || PageDirty(page))
8584 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8587 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8588 unsigned int length)
8590 struct inode *inode = page->mapping->host;
8591 struct extent_io_tree *tree;
8592 struct btrfs_ordered_extent *ordered;
8593 struct extent_state *cached_state = NULL;
8594 u64 page_start = page_offset(page);
8595 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8596 int inode_evicting = inode->i_state & I_FREEING;
8599 * we have the page locked, so new writeback can't start,
8600 * and the dirty bit won't be cleared while we are here.
8602 * Wait for IO on this page so that we can safely clear
8603 * the PagePrivate2 bit and do ordered accounting
8605 wait_on_page_writeback(page);
8607 tree = &BTRFS_I(inode)->io_tree;
8609 btrfs_releasepage(page, GFP_NOFS);
8613 if (!inode_evicting)
8614 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8615 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8618 * IO on this page will never be started, so we need
8619 * to account for any ordered extents now
8621 if (!inode_evicting)
8622 clear_extent_bit(tree, page_start, page_end,
8623 EXTENT_DIRTY | EXTENT_DELALLOC |
8624 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8625 EXTENT_DEFRAG, 1, 0, &cached_state,
8628 * whoever cleared the private bit is responsible
8629 * for the finish_ordered_io
8631 if (TestClearPagePrivate2(page)) {
8632 struct btrfs_ordered_inode_tree *tree;
8635 tree = &BTRFS_I(inode)->ordered_tree;
8637 spin_lock_irq(&tree->lock);
8638 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8639 new_len = page_start - ordered->file_offset;
8640 if (new_len < ordered->truncated_len)
8641 ordered->truncated_len = new_len;
8642 spin_unlock_irq(&tree->lock);
8644 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8646 PAGE_CACHE_SIZE, 1))
8647 btrfs_finish_ordered_io(ordered);
8649 btrfs_put_ordered_extent(ordered);
8650 if (!inode_evicting) {
8651 cached_state = NULL;
8652 lock_extent_bits(tree, page_start, page_end, 0,
8658 * Qgroup reserved space handler
8659 * Page here will be either
8660 * 1) Already written to disk
8661 * In this case, its reserved space is released from data rsv map
8662 * and will be freed by delayed_ref handler finally.
8663 * So even we call qgroup_free_data(), it won't decrease reserved
8665 * 2) Not written to disk
8666 * This means the reserved space should be freed here.
8668 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8669 if (!inode_evicting) {
8670 clear_extent_bit(tree, page_start, page_end,
8671 EXTENT_LOCKED | EXTENT_DIRTY |
8672 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8673 EXTENT_DEFRAG, 1, 1,
8674 &cached_state, GFP_NOFS);
8676 __btrfs_releasepage(page, GFP_NOFS);
8679 ClearPageChecked(page);
8680 if (PagePrivate(page)) {
8681 ClearPagePrivate(page);
8682 set_page_private(page, 0);
8683 page_cache_release(page);
8688 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8689 * called from a page fault handler when a page is first dirtied. Hence we must
8690 * be careful to check for EOF conditions here. We set the page up correctly
8691 * for a written page which means we get ENOSPC checking when writing into
8692 * holes and correct delalloc and unwritten extent mapping on filesystems that
8693 * support these features.
8695 * We are not allowed to take the i_mutex here so we have to play games to
8696 * protect against truncate races as the page could now be beyond EOF. Because
8697 * vmtruncate() writes the inode size before removing pages, once we have the
8698 * page lock we can determine safely if the page is beyond EOF. If it is not
8699 * beyond EOF, then the page is guaranteed safe against truncation until we
8702 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8704 struct page *page = vmf->page;
8705 struct inode *inode = file_inode(vma->vm_file);
8706 struct btrfs_root *root = BTRFS_I(inode)->root;
8707 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8708 struct btrfs_ordered_extent *ordered;
8709 struct extent_state *cached_state = NULL;
8711 unsigned long zero_start;
8718 sb_start_pagefault(inode->i_sb);
8719 page_start = page_offset(page);
8720 page_end = page_start + PAGE_CACHE_SIZE - 1;
8722 ret = btrfs_delalloc_reserve_space(inode, page_start,
8725 ret = file_update_time(vma->vm_file);
8731 else /* -ENOSPC, -EIO, etc */
8732 ret = VM_FAULT_SIGBUS;
8738 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8741 size = i_size_read(inode);
8743 if ((page->mapping != inode->i_mapping) ||
8744 (page_start >= size)) {
8745 /* page got truncated out from underneath us */
8748 wait_on_page_writeback(page);
8750 lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8751 set_page_extent_mapped(page);
8754 * we can't set the delalloc bits if there are pending ordered
8755 * extents. Drop our locks and wait for them to finish
8757 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8759 unlock_extent_cached(io_tree, page_start, page_end,
8760 &cached_state, GFP_NOFS);
8762 btrfs_start_ordered_extent(inode, ordered, 1);
8763 btrfs_put_ordered_extent(ordered);
8768 * XXX - page_mkwrite gets called every time the page is dirtied, even
8769 * if it was already dirty, so for space accounting reasons we need to
8770 * clear any delalloc bits for the range we are fixing to save. There
8771 * is probably a better way to do this, but for now keep consistent with
8772 * prepare_pages in the normal write path.
8774 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8775 EXTENT_DIRTY | EXTENT_DELALLOC |
8776 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8777 0, 0, &cached_state, GFP_NOFS);
8779 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8782 unlock_extent_cached(io_tree, page_start, page_end,
8783 &cached_state, GFP_NOFS);
8784 ret = VM_FAULT_SIGBUS;
8789 /* page is wholly or partially inside EOF */
8790 if (page_start + PAGE_CACHE_SIZE > size)
8791 zero_start = size & ~PAGE_CACHE_MASK;
8793 zero_start = PAGE_CACHE_SIZE;
8795 if (zero_start != PAGE_CACHE_SIZE) {
8797 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8798 flush_dcache_page(page);
8801 ClearPageChecked(page);
8802 set_page_dirty(page);
8803 SetPageUptodate(page);
8805 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8806 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8807 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8809 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8813 sb_end_pagefault(inode->i_sb);
8814 return VM_FAULT_LOCKED;
8818 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8820 sb_end_pagefault(inode->i_sb);
8824 static int btrfs_truncate(struct inode *inode)
8826 struct btrfs_root *root = BTRFS_I(inode)->root;
8827 struct btrfs_block_rsv *rsv;
8830 struct btrfs_trans_handle *trans;
8831 u64 mask = root->sectorsize - 1;
8832 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8834 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8840 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8841 * 3 things going on here
8843 * 1) We need to reserve space for our orphan item and the space to
8844 * delete our orphan item. Lord knows we don't want to have a dangling
8845 * orphan item because we didn't reserve space to remove it.
8847 * 2) We need to reserve space to update our inode.
8849 * 3) We need to have something to cache all the space that is going to
8850 * be free'd up by the truncate operation, but also have some slack
8851 * space reserved in case it uses space during the truncate (thank you
8852 * very much snapshotting).
8854 * And we need these to all be seperate. The fact is we can use alot of
8855 * space doing the truncate, and we have no earthly idea how much space
8856 * we will use, so we need the truncate reservation to be seperate so it
8857 * doesn't end up using space reserved for updating the inode or
8858 * removing the orphan item. We also need to be able to stop the
8859 * transaction and start a new one, which means we need to be able to
8860 * update the inode several times, and we have no idea of knowing how
8861 * many times that will be, so we can't just reserve 1 item for the
8862 * entirety of the opration, so that has to be done seperately as well.
8863 * Then there is the orphan item, which does indeed need to be held on
8864 * to for the whole operation, and we need nobody to touch this reserved
8865 * space except the orphan code.
8867 * So that leaves us with
8869 * 1) root->orphan_block_rsv - for the orphan deletion.
8870 * 2) rsv - for the truncate reservation, which we will steal from the
8871 * transaction reservation.
8872 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8873 * updating the inode.
8875 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8878 rsv->size = min_size;
8882 * 1 for the truncate slack space
8883 * 1 for updating the inode.
8885 trans = btrfs_start_transaction(root, 2);
8886 if (IS_ERR(trans)) {
8887 err = PTR_ERR(trans);
8891 /* Migrate the slack space for the truncate to our reserve */
8892 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8897 * So if we truncate and then write and fsync we normally would just
8898 * write the extents that changed, which is a problem if we need to
8899 * first truncate that entire inode. So set this flag so we write out
8900 * all of the extents in the inode to the sync log so we're completely
8903 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8904 trans->block_rsv = rsv;
8907 ret = btrfs_truncate_inode_items(trans, root, inode,
8909 BTRFS_EXTENT_DATA_KEY);
8910 if (ret != -ENOSPC && ret != -EAGAIN) {
8915 trans->block_rsv = &root->fs_info->trans_block_rsv;
8916 ret = btrfs_update_inode(trans, root, inode);
8922 btrfs_end_transaction(trans, root);
8923 btrfs_btree_balance_dirty(root);
8925 trans = btrfs_start_transaction(root, 2);
8926 if (IS_ERR(trans)) {
8927 ret = err = PTR_ERR(trans);
8932 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8934 BUG_ON(ret); /* shouldn't happen */
8935 trans->block_rsv = rsv;
8938 if (ret == 0 && inode->i_nlink > 0) {
8939 trans->block_rsv = root->orphan_block_rsv;
8940 ret = btrfs_orphan_del(trans, inode);
8946 trans->block_rsv = &root->fs_info->trans_block_rsv;
8947 ret = btrfs_update_inode(trans, root, inode);
8951 ret = btrfs_end_transaction(trans, root);
8952 btrfs_btree_balance_dirty(root);
8956 btrfs_free_block_rsv(root, rsv);
8965 * create a new subvolume directory/inode (helper for the ioctl).
8967 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8968 struct btrfs_root *new_root,
8969 struct btrfs_root *parent_root,
8972 struct inode *inode;
8976 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8977 new_dirid, new_dirid,
8978 S_IFDIR | (~current_umask() & S_IRWXUGO),
8981 return PTR_ERR(inode);
8982 inode->i_op = &btrfs_dir_inode_operations;
8983 inode->i_fop = &btrfs_dir_file_operations;
8985 set_nlink(inode, 1);
8986 btrfs_i_size_write(inode, 0);
8987 unlock_new_inode(inode);
8989 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8991 btrfs_err(new_root->fs_info,
8992 "error inheriting subvolume %llu properties: %d",
8993 new_root->root_key.objectid, err);
8995 err = btrfs_update_inode(trans, new_root, inode);
9001 struct inode *btrfs_alloc_inode(struct super_block *sb)
9003 struct btrfs_inode *ei;
9004 struct inode *inode;
9006 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9013 ei->last_sub_trans = 0;
9014 ei->logged_trans = 0;
9015 ei->delalloc_bytes = 0;
9016 ei->defrag_bytes = 0;
9017 ei->disk_i_size = 0;
9020 ei->index_cnt = (u64)-1;
9022 ei->last_unlink_trans = 0;
9023 ei->last_log_commit = 0;
9025 spin_lock_init(&ei->lock);
9026 ei->outstanding_extents = 0;
9027 ei->reserved_extents = 0;
9029 ei->runtime_flags = 0;
9030 ei->force_compress = BTRFS_COMPRESS_NONE;
9032 ei->delayed_node = NULL;
9034 ei->i_otime.tv_sec = 0;
9035 ei->i_otime.tv_nsec = 0;
9037 inode = &ei->vfs_inode;
9038 extent_map_tree_init(&ei->extent_tree);
9039 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9040 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9041 ei->io_tree.track_uptodate = 1;
9042 ei->io_failure_tree.track_uptodate = 1;
9043 atomic_set(&ei->sync_writers, 0);
9044 mutex_init(&ei->log_mutex);
9045 mutex_init(&ei->delalloc_mutex);
9046 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9047 INIT_LIST_HEAD(&ei->delalloc_inodes);
9048 RB_CLEAR_NODE(&ei->rb_node);
9053 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9054 void btrfs_test_destroy_inode(struct inode *inode)
9056 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9057 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9061 static void btrfs_i_callback(struct rcu_head *head)
9063 struct inode *inode = container_of(head, struct inode, i_rcu);
9064 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9067 void btrfs_destroy_inode(struct inode *inode)
9069 struct btrfs_ordered_extent *ordered;
9070 struct btrfs_root *root = BTRFS_I(inode)->root;
9072 WARN_ON(!hlist_empty(&inode->i_dentry));
9073 WARN_ON(inode->i_data.nrpages);
9074 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9075 WARN_ON(BTRFS_I(inode)->reserved_extents);
9076 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9077 WARN_ON(BTRFS_I(inode)->csum_bytes);
9078 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9081 * This can happen where we create an inode, but somebody else also
9082 * created the same inode and we need to destroy the one we already
9088 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9089 &BTRFS_I(inode)->runtime_flags)) {
9090 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9092 atomic_dec(&root->orphan_inodes);
9096 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9100 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9101 ordered->file_offset, ordered->len);
9102 btrfs_remove_ordered_extent(inode, ordered);
9103 btrfs_put_ordered_extent(ordered);
9104 btrfs_put_ordered_extent(ordered);
9107 btrfs_qgroup_check_reserved_leak(inode);
9108 inode_tree_del(inode);
9109 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9111 call_rcu(&inode->i_rcu, btrfs_i_callback);
9114 int btrfs_drop_inode(struct inode *inode)
9116 struct btrfs_root *root = BTRFS_I(inode)->root;
9121 /* the snap/subvol tree is on deleting */
9122 if (btrfs_root_refs(&root->root_item) == 0)
9125 return generic_drop_inode(inode);
9128 static void init_once(void *foo)
9130 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9132 inode_init_once(&ei->vfs_inode);
9135 void btrfs_destroy_cachep(void)
9138 * Make sure all delayed rcu free inodes are flushed before we
9142 if (btrfs_inode_cachep)
9143 kmem_cache_destroy(btrfs_inode_cachep);
9144 if (btrfs_trans_handle_cachep)
9145 kmem_cache_destroy(btrfs_trans_handle_cachep);
9146 if (btrfs_transaction_cachep)
9147 kmem_cache_destroy(btrfs_transaction_cachep);
9148 if (btrfs_path_cachep)
9149 kmem_cache_destroy(btrfs_path_cachep);
9150 if (btrfs_free_space_cachep)
9151 kmem_cache_destroy(btrfs_free_space_cachep);
9152 if (btrfs_delalloc_work_cachep)
9153 kmem_cache_destroy(btrfs_delalloc_work_cachep);
9156 int btrfs_init_cachep(void)
9158 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9159 sizeof(struct btrfs_inode), 0,
9160 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9161 if (!btrfs_inode_cachep)
9164 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9165 sizeof(struct btrfs_trans_handle), 0,
9166 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9167 if (!btrfs_trans_handle_cachep)
9170 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9171 sizeof(struct btrfs_transaction), 0,
9172 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9173 if (!btrfs_transaction_cachep)
9176 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9177 sizeof(struct btrfs_path), 0,
9178 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9179 if (!btrfs_path_cachep)
9182 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9183 sizeof(struct btrfs_free_space), 0,
9184 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9185 if (!btrfs_free_space_cachep)
9188 btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
9189 sizeof(struct btrfs_delalloc_work), 0,
9190 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
9192 if (!btrfs_delalloc_work_cachep)
9197 btrfs_destroy_cachep();
9201 static int btrfs_getattr(struct vfsmount *mnt,
9202 struct dentry *dentry, struct kstat *stat)
9205 struct inode *inode = d_inode(dentry);
9206 u32 blocksize = inode->i_sb->s_blocksize;
9208 generic_fillattr(inode, stat);
9209 stat->dev = BTRFS_I(inode)->root->anon_dev;
9210 stat->blksize = PAGE_CACHE_SIZE;
9212 spin_lock(&BTRFS_I(inode)->lock);
9213 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9214 spin_unlock(&BTRFS_I(inode)->lock);
9215 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9216 ALIGN(delalloc_bytes, blocksize)) >> 9;
9220 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9221 struct inode *new_dir, struct dentry *new_dentry)
9223 struct btrfs_trans_handle *trans;
9224 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9225 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9226 struct inode *new_inode = d_inode(new_dentry);
9227 struct inode *old_inode = d_inode(old_dentry);
9228 struct timespec ctime = CURRENT_TIME;
9232 u64 old_ino = btrfs_ino(old_inode);
9234 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9237 /* we only allow rename subvolume link between subvolumes */
9238 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9241 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9242 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9245 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9246 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9250 /* check for collisions, even if the name isn't there */
9251 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9252 new_dentry->d_name.name,
9253 new_dentry->d_name.len);
9256 if (ret == -EEXIST) {
9258 * eexist without a new_inode */
9259 if (WARN_ON(!new_inode)) {
9263 /* maybe -EOVERFLOW */
9270 * we're using rename to replace one file with another. Start IO on it
9271 * now so we don't add too much work to the end of the transaction
9273 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9274 filemap_flush(old_inode->i_mapping);
9276 /* close the racy window with snapshot create/destroy ioctl */
9277 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9278 down_read(&root->fs_info->subvol_sem);
9280 * We want to reserve the absolute worst case amount of items. So if
9281 * both inodes are subvols and we need to unlink them then that would
9282 * require 4 item modifications, but if they are both normal inodes it
9283 * would require 5 item modifications, so we'll assume their normal
9284 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9285 * should cover the worst case number of items we'll modify.
9287 trans = btrfs_start_transaction(root, 11);
9288 if (IS_ERR(trans)) {
9289 ret = PTR_ERR(trans);
9294 btrfs_record_root_in_trans(trans, dest);
9296 ret = btrfs_set_inode_index(new_dir, &index);
9300 BTRFS_I(old_inode)->dir_index = 0ULL;
9301 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9302 /* force full log commit if subvolume involved. */
9303 btrfs_set_log_full_commit(root->fs_info, trans);
9305 ret = btrfs_insert_inode_ref(trans, dest,
9306 new_dentry->d_name.name,
9307 new_dentry->d_name.len,
9309 btrfs_ino(new_dir), index);
9313 * this is an ugly little race, but the rename is required
9314 * to make sure that if we crash, the inode is either at the
9315 * old name or the new one. pinning the log transaction lets
9316 * us make sure we don't allow a log commit to come in after
9317 * we unlink the name but before we add the new name back in.
9319 btrfs_pin_log_trans(root);
9322 inode_inc_iversion(old_dir);
9323 inode_inc_iversion(new_dir);
9324 inode_inc_iversion(old_inode);
9325 old_dir->i_ctime = old_dir->i_mtime = ctime;
9326 new_dir->i_ctime = new_dir->i_mtime = ctime;
9327 old_inode->i_ctime = ctime;
9329 if (old_dentry->d_parent != new_dentry->d_parent)
9330 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9332 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9333 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9334 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9335 old_dentry->d_name.name,
9336 old_dentry->d_name.len);
9338 ret = __btrfs_unlink_inode(trans, root, old_dir,
9339 d_inode(old_dentry),
9340 old_dentry->d_name.name,
9341 old_dentry->d_name.len);
9343 ret = btrfs_update_inode(trans, root, old_inode);
9346 btrfs_abort_transaction(trans, root, ret);
9351 inode_inc_iversion(new_inode);
9352 new_inode->i_ctime = CURRENT_TIME;
9353 if (unlikely(btrfs_ino(new_inode) ==
9354 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9355 root_objectid = BTRFS_I(new_inode)->location.objectid;
9356 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9358 new_dentry->d_name.name,
9359 new_dentry->d_name.len);
9360 BUG_ON(new_inode->i_nlink == 0);
9362 ret = btrfs_unlink_inode(trans, dest, new_dir,
9363 d_inode(new_dentry),
9364 new_dentry->d_name.name,
9365 new_dentry->d_name.len);
9367 if (!ret && new_inode->i_nlink == 0)
9368 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9370 btrfs_abort_transaction(trans, root, ret);
9375 ret = btrfs_add_link(trans, new_dir, old_inode,
9376 new_dentry->d_name.name,
9377 new_dentry->d_name.len, 0, index);
9379 btrfs_abort_transaction(trans, root, ret);
9383 if (old_inode->i_nlink == 1)
9384 BTRFS_I(old_inode)->dir_index = index;
9386 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9387 struct dentry *parent = new_dentry->d_parent;
9388 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9389 btrfs_end_log_trans(root);
9392 btrfs_end_transaction(trans, root);
9394 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9395 up_read(&root->fs_info->subvol_sem);
9400 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9401 struct inode *new_dir, struct dentry *new_dentry,
9404 if (flags & ~RENAME_NOREPLACE)
9407 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9410 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9412 struct btrfs_delalloc_work *delalloc_work;
9413 struct inode *inode;
9415 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9417 inode = delalloc_work->inode;
9418 if (delalloc_work->wait) {
9419 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9421 filemap_flush(inode->i_mapping);
9422 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9423 &BTRFS_I(inode)->runtime_flags))
9424 filemap_flush(inode->i_mapping);
9427 if (delalloc_work->delay_iput)
9428 btrfs_add_delayed_iput(inode);
9431 complete(&delalloc_work->completion);
9434 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9435 int wait, int delay_iput)
9437 struct btrfs_delalloc_work *work;
9439 work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9443 init_completion(&work->completion);
9444 INIT_LIST_HEAD(&work->list);
9445 work->inode = inode;
9447 work->delay_iput = delay_iput;
9448 WARN_ON_ONCE(!inode);
9449 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9450 btrfs_run_delalloc_work, NULL, NULL);
9455 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9457 wait_for_completion(&work->completion);
9458 kmem_cache_free(btrfs_delalloc_work_cachep, work);
9462 * some fairly slow code that needs optimization. This walks the list
9463 * of all the inodes with pending delalloc and forces them to disk.
9465 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9468 struct btrfs_inode *binode;
9469 struct inode *inode;
9470 struct btrfs_delalloc_work *work, *next;
9471 struct list_head works;
9472 struct list_head splice;
9475 INIT_LIST_HEAD(&works);
9476 INIT_LIST_HEAD(&splice);
9478 mutex_lock(&root->delalloc_mutex);
9479 spin_lock(&root->delalloc_lock);
9480 list_splice_init(&root->delalloc_inodes, &splice);
9481 while (!list_empty(&splice)) {
9482 binode = list_entry(splice.next, struct btrfs_inode,
9485 list_move_tail(&binode->delalloc_inodes,
9486 &root->delalloc_inodes);
9487 inode = igrab(&binode->vfs_inode);
9489 cond_resched_lock(&root->delalloc_lock);
9492 spin_unlock(&root->delalloc_lock);
9494 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9497 btrfs_add_delayed_iput(inode);
9503 list_add_tail(&work->list, &works);
9504 btrfs_queue_work(root->fs_info->flush_workers,
9507 if (nr != -1 && ret >= nr)
9510 spin_lock(&root->delalloc_lock);
9512 spin_unlock(&root->delalloc_lock);
9515 list_for_each_entry_safe(work, next, &works, list) {
9516 list_del_init(&work->list);
9517 btrfs_wait_and_free_delalloc_work(work);
9520 if (!list_empty_careful(&splice)) {
9521 spin_lock(&root->delalloc_lock);
9522 list_splice_tail(&splice, &root->delalloc_inodes);
9523 spin_unlock(&root->delalloc_lock);
9525 mutex_unlock(&root->delalloc_mutex);
9529 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9533 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9536 ret = __start_delalloc_inodes(root, delay_iput, -1);
9540 * the filemap_flush will queue IO into the worker threads, but
9541 * we have to make sure the IO is actually started and that
9542 * ordered extents get created before we return
9544 atomic_inc(&root->fs_info->async_submit_draining);
9545 while (atomic_read(&root->fs_info->nr_async_submits) ||
9546 atomic_read(&root->fs_info->async_delalloc_pages)) {
9547 wait_event(root->fs_info->async_submit_wait,
9548 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9549 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9551 atomic_dec(&root->fs_info->async_submit_draining);
9555 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9558 struct btrfs_root *root;
9559 struct list_head splice;
9562 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9565 INIT_LIST_HEAD(&splice);
9567 mutex_lock(&fs_info->delalloc_root_mutex);
9568 spin_lock(&fs_info->delalloc_root_lock);
9569 list_splice_init(&fs_info->delalloc_roots, &splice);
9570 while (!list_empty(&splice) && nr) {
9571 root = list_first_entry(&splice, struct btrfs_root,
9573 root = btrfs_grab_fs_root(root);
9575 list_move_tail(&root->delalloc_root,
9576 &fs_info->delalloc_roots);
9577 spin_unlock(&fs_info->delalloc_root_lock);
9579 ret = __start_delalloc_inodes(root, delay_iput, nr);
9580 btrfs_put_fs_root(root);
9588 spin_lock(&fs_info->delalloc_root_lock);
9590 spin_unlock(&fs_info->delalloc_root_lock);
9593 atomic_inc(&fs_info->async_submit_draining);
9594 while (atomic_read(&fs_info->nr_async_submits) ||
9595 atomic_read(&fs_info->async_delalloc_pages)) {
9596 wait_event(fs_info->async_submit_wait,
9597 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9598 atomic_read(&fs_info->async_delalloc_pages) == 0));
9600 atomic_dec(&fs_info->async_submit_draining);
9602 if (!list_empty_careful(&splice)) {
9603 spin_lock(&fs_info->delalloc_root_lock);
9604 list_splice_tail(&splice, &fs_info->delalloc_roots);
9605 spin_unlock(&fs_info->delalloc_root_lock);
9607 mutex_unlock(&fs_info->delalloc_root_mutex);
9611 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9612 const char *symname)
9614 struct btrfs_trans_handle *trans;
9615 struct btrfs_root *root = BTRFS_I(dir)->root;
9616 struct btrfs_path *path;
9617 struct btrfs_key key;
9618 struct inode *inode = NULL;
9626 struct btrfs_file_extent_item *ei;
9627 struct extent_buffer *leaf;
9629 name_len = strlen(symname);
9630 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9631 return -ENAMETOOLONG;
9634 * 2 items for inode item and ref
9635 * 2 items for dir items
9636 * 1 item for xattr if selinux is on
9638 trans = btrfs_start_transaction(root, 5);
9640 return PTR_ERR(trans);
9642 err = btrfs_find_free_ino(root, &objectid);
9646 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9647 dentry->d_name.len, btrfs_ino(dir), objectid,
9648 S_IFLNK|S_IRWXUGO, &index);
9649 if (IS_ERR(inode)) {
9650 err = PTR_ERR(inode);
9655 * If the active LSM wants to access the inode during
9656 * d_instantiate it needs these. Smack checks to see
9657 * if the filesystem supports xattrs by looking at the
9660 inode->i_fop = &btrfs_file_operations;
9661 inode->i_op = &btrfs_file_inode_operations;
9662 inode->i_mapping->a_ops = &btrfs_aops;
9663 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9665 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9667 goto out_unlock_inode;
9669 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9671 goto out_unlock_inode;
9673 path = btrfs_alloc_path();
9676 goto out_unlock_inode;
9678 key.objectid = btrfs_ino(inode);
9680 key.type = BTRFS_EXTENT_DATA_KEY;
9681 datasize = btrfs_file_extent_calc_inline_size(name_len);
9682 err = btrfs_insert_empty_item(trans, root, path, &key,
9685 btrfs_free_path(path);
9686 goto out_unlock_inode;
9688 leaf = path->nodes[0];
9689 ei = btrfs_item_ptr(leaf, path->slots[0],
9690 struct btrfs_file_extent_item);
9691 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9692 btrfs_set_file_extent_type(leaf, ei,
9693 BTRFS_FILE_EXTENT_INLINE);
9694 btrfs_set_file_extent_encryption(leaf, ei, 0);
9695 btrfs_set_file_extent_compression(leaf, ei, 0);
9696 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9697 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9699 ptr = btrfs_file_extent_inline_start(ei);
9700 write_extent_buffer(leaf, symname, ptr, name_len);
9701 btrfs_mark_buffer_dirty(leaf);
9702 btrfs_free_path(path);
9704 inode->i_op = &btrfs_symlink_inode_operations;
9705 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9706 inode_set_bytes(inode, name_len);
9707 btrfs_i_size_write(inode, name_len);
9708 err = btrfs_update_inode(trans, root, inode);
9711 goto out_unlock_inode;
9714 unlock_new_inode(inode);
9715 d_instantiate(dentry, inode);
9718 btrfs_end_transaction(trans, root);
9720 inode_dec_link_count(inode);
9723 btrfs_btree_balance_dirty(root);
9728 unlock_new_inode(inode);
9732 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9733 u64 start, u64 num_bytes, u64 min_size,
9734 loff_t actual_len, u64 *alloc_hint,
9735 struct btrfs_trans_handle *trans)
9737 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9738 struct extent_map *em;
9739 struct btrfs_root *root = BTRFS_I(inode)->root;
9740 struct btrfs_key ins;
9741 u64 cur_offset = start;
9744 u64 last_alloc = (u64)-1;
9746 bool own_trans = true;
9750 while (num_bytes > 0) {
9752 trans = btrfs_start_transaction(root, 3);
9753 if (IS_ERR(trans)) {
9754 ret = PTR_ERR(trans);
9759 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9760 cur_bytes = max(cur_bytes, min_size);
9762 * If we are severely fragmented we could end up with really
9763 * small allocations, so if the allocator is returning small
9764 * chunks lets make its job easier by only searching for those
9767 cur_bytes = min(cur_bytes, last_alloc);
9768 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9769 *alloc_hint, &ins, 1, 0);
9772 btrfs_end_transaction(trans, root);
9776 last_alloc = ins.offset;
9777 ret = insert_reserved_file_extent(trans, inode,
9778 cur_offset, ins.objectid,
9779 ins.offset, ins.offset,
9780 ins.offset, 0, 0, 0,
9781 BTRFS_FILE_EXTENT_PREALLOC);
9783 btrfs_free_reserved_extent(root, ins.objectid,
9785 btrfs_abort_transaction(trans, root, ret);
9787 btrfs_end_transaction(trans, root);
9791 btrfs_drop_extent_cache(inode, cur_offset,
9792 cur_offset + ins.offset -1, 0);
9794 em = alloc_extent_map();
9796 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9797 &BTRFS_I(inode)->runtime_flags);
9801 em->start = cur_offset;
9802 em->orig_start = cur_offset;
9803 em->len = ins.offset;
9804 em->block_start = ins.objectid;
9805 em->block_len = ins.offset;
9806 em->orig_block_len = ins.offset;
9807 em->ram_bytes = ins.offset;
9808 em->bdev = root->fs_info->fs_devices->latest_bdev;
9809 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9810 em->generation = trans->transid;
9813 write_lock(&em_tree->lock);
9814 ret = add_extent_mapping(em_tree, em, 1);
9815 write_unlock(&em_tree->lock);
9818 btrfs_drop_extent_cache(inode, cur_offset,
9819 cur_offset + ins.offset - 1,
9822 free_extent_map(em);
9824 num_bytes -= ins.offset;
9825 cur_offset += ins.offset;
9826 *alloc_hint = ins.objectid + ins.offset;
9828 inode_inc_iversion(inode);
9829 inode->i_ctime = CURRENT_TIME;
9830 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9831 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9832 (actual_len > inode->i_size) &&
9833 (cur_offset > inode->i_size)) {
9834 if (cur_offset > actual_len)
9835 i_size = actual_len;
9837 i_size = cur_offset;
9838 i_size_write(inode, i_size);
9839 btrfs_ordered_update_i_size(inode, i_size, NULL);
9842 ret = btrfs_update_inode(trans, root, inode);
9845 btrfs_abort_transaction(trans, root, ret);
9847 btrfs_end_transaction(trans, root);
9852 btrfs_end_transaction(trans, root);
9857 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9858 u64 start, u64 num_bytes, u64 min_size,
9859 loff_t actual_len, u64 *alloc_hint)
9861 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9862 min_size, actual_len, alloc_hint,
9866 int btrfs_prealloc_file_range_trans(struct inode *inode,
9867 struct btrfs_trans_handle *trans, int mode,
9868 u64 start, u64 num_bytes, u64 min_size,
9869 loff_t actual_len, u64 *alloc_hint)
9871 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9872 min_size, actual_len, alloc_hint, trans);
9875 static int btrfs_set_page_dirty(struct page *page)
9877 return __set_page_dirty_nobuffers(page);
9880 static int btrfs_permission(struct inode *inode, int mask)
9882 struct btrfs_root *root = BTRFS_I(inode)->root;
9883 umode_t mode = inode->i_mode;
9885 if (mask & MAY_WRITE &&
9886 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9887 if (btrfs_root_readonly(root))
9889 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9892 return generic_permission(inode, mask);
9895 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9897 struct btrfs_trans_handle *trans;
9898 struct btrfs_root *root = BTRFS_I(dir)->root;
9899 struct inode *inode = NULL;
9905 * 5 units required for adding orphan entry
9907 trans = btrfs_start_transaction(root, 5);
9909 return PTR_ERR(trans);
9911 ret = btrfs_find_free_ino(root, &objectid);
9915 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9916 btrfs_ino(dir), objectid, mode, &index);
9917 if (IS_ERR(inode)) {
9918 ret = PTR_ERR(inode);
9923 inode->i_fop = &btrfs_file_operations;
9924 inode->i_op = &btrfs_file_inode_operations;
9926 inode->i_mapping->a_ops = &btrfs_aops;
9927 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9929 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9933 ret = btrfs_update_inode(trans, root, inode);
9936 ret = btrfs_orphan_add(trans, inode);
9941 * We set number of links to 0 in btrfs_new_inode(), and here we set
9942 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9945 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9947 set_nlink(inode, 1);
9948 unlock_new_inode(inode);
9949 d_tmpfile(dentry, inode);
9950 mark_inode_dirty(inode);
9953 btrfs_end_transaction(trans, root);
9956 btrfs_balance_delayed_items(root);
9957 btrfs_btree_balance_dirty(root);
9961 unlock_new_inode(inode);
9966 /* Inspired by filemap_check_errors() */
9967 int btrfs_inode_check_errors(struct inode *inode)
9971 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9972 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9974 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9975 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9981 static const struct inode_operations btrfs_dir_inode_operations = {
9982 .getattr = btrfs_getattr,
9983 .lookup = btrfs_lookup,
9984 .create = btrfs_create,
9985 .unlink = btrfs_unlink,
9987 .mkdir = btrfs_mkdir,
9988 .rmdir = btrfs_rmdir,
9989 .rename2 = btrfs_rename2,
9990 .symlink = btrfs_symlink,
9991 .setattr = btrfs_setattr,
9992 .mknod = btrfs_mknod,
9993 .setxattr = btrfs_setxattr,
9994 .getxattr = btrfs_getxattr,
9995 .listxattr = btrfs_listxattr,
9996 .removexattr = btrfs_removexattr,
9997 .permission = btrfs_permission,
9998 .get_acl = btrfs_get_acl,
9999 .set_acl = btrfs_set_acl,
10000 .update_time = btrfs_update_time,
10001 .tmpfile = btrfs_tmpfile,
10003 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10004 .lookup = btrfs_lookup,
10005 .permission = btrfs_permission,
10006 .get_acl = btrfs_get_acl,
10007 .set_acl = btrfs_set_acl,
10008 .update_time = btrfs_update_time,
10011 static const struct file_operations btrfs_dir_file_operations = {
10012 .llseek = generic_file_llseek,
10013 .read = generic_read_dir,
10014 .iterate = btrfs_real_readdir,
10015 .unlocked_ioctl = btrfs_ioctl,
10016 #ifdef CONFIG_COMPAT
10017 .compat_ioctl = btrfs_ioctl,
10019 .release = btrfs_release_file,
10020 .fsync = btrfs_sync_file,
10023 static struct extent_io_ops btrfs_extent_io_ops = {
10024 .fill_delalloc = run_delalloc_range,
10025 .submit_bio_hook = btrfs_submit_bio_hook,
10026 .merge_bio_hook = btrfs_merge_bio_hook,
10027 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10028 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10029 .writepage_start_hook = btrfs_writepage_start_hook,
10030 .set_bit_hook = btrfs_set_bit_hook,
10031 .clear_bit_hook = btrfs_clear_bit_hook,
10032 .merge_extent_hook = btrfs_merge_extent_hook,
10033 .split_extent_hook = btrfs_split_extent_hook,
10037 * btrfs doesn't support the bmap operation because swapfiles
10038 * use bmap to make a mapping of extents in the file. They assume
10039 * these extents won't change over the life of the file and they
10040 * use the bmap result to do IO directly to the drive.
10042 * the btrfs bmap call would return logical addresses that aren't
10043 * suitable for IO and they also will change frequently as COW
10044 * operations happen. So, swapfile + btrfs == corruption.
10046 * For now we're avoiding this by dropping bmap.
10048 static const struct address_space_operations btrfs_aops = {
10049 .readpage = btrfs_readpage,
10050 .writepage = btrfs_writepage,
10051 .writepages = btrfs_writepages,
10052 .readpages = btrfs_readpages,
10053 .direct_IO = btrfs_direct_IO,
10054 .invalidatepage = btrfs_invalidatepage,
10055 .releasepage = btrfs_releasepage,
10056 .set_page_dirty = btrfs_set_page_dirty,
10057 .error_remove_page = generic_error_remove_page,
10060 static const struct address_space_operations btrfs_symlink_aops = {
10061 .readpage = btrfs_readpage,
10062 .writepage = btrfs_writepage,
10063 .invalidatepage = btrfs_invalidatepage,
10064 .releasepage = btrfs_releasepage,
10067 static const struct inode_operations btrfs_file_inode_operations = {
10068 .getattr = btrfs_getattr,
10069 .setattr = btrfs_setattr,
10070 .setxattr = btrfs_setxattr,
10071 .getxattr = btrfs_getxattr,
10072 .listxattr = btrfs_listxattr,
10073 .removexattr = btrfs_removexattr,
10074 .permission = btrfs_permission,
10075 .fiemap = btrfs_fiemap,
10076 .get_acl = btrfs_get_acl,
10077 .set_acl = btrfs_set_acl,
10078 .update_time = btrfs_update_time,
10080 static const struct inode_operations btrfs_special_inode_operations = {
10081 .getattr = btrfs_getattr,
10082 .setattr = btrfs_setattr,
10083 .permission = btrfs_permission,
10084 .setxattr = btrfs_setxattr,
10085 .getxattr = btrfs_getxattr,
10086 .listxattr = btrfs_listxattr,
10087 .removexattr = btrfs_removexattr,
10088 .get_acl = btrfs_get_acl,
10089 .set_acl = btrfs_set_acl,
10090 .update_time = btrfs_update_time,
10092 static const struct inode_operations btrfs_symlink_inode_operations = {
10093 .readlink = generic_readlink,
10094 .follow_link = page_follow_link_light,
10095 .put_link = page_put_link,
10096 .getattr = btrfs_getattr,
10097 .setattr = btrfs_setattr,
10098 .permission = btrfs_permission,
10099 .setxattr = btrfs_setxattr,
10100 .getxattr = btrfs_getxattr,
10101 .listxattr = btrfs_listxattr,
10102 .removexattr = btrfs_removexattr,
10103 .update_time = btrfs_update_time,
10106 const struct dentry_operations btrfs_dentry_operations = {
10107 .d_delete = btrfs_dentry_delete,
10108 .d_release = btrfs_dentry_release,