2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/kernel.h>
20 #include <linux/bio.h>
21 #include <linux/buffer_head.h>
22 #include <linux/file.h>
24 #include <linux/pagemap.h>
25 #include <linux/highmem.h>
26 #include <linux/time.h>
27 #include <linux/init.h>
28 #include <linux/string.h>
29 #include <linux/backing-dev.h>
30 #include <linux/mpage.h>
31 #include <linux/swap.h>
32 #include <linux/writeback.h>
33 #include <linux/statfs.h>
34 #include <linux/compat.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/xattr.h>
37 #include <linux/posix_acl.h>
38 #include <linux/falloc.h>
39 #include <linux/slab.h>
40 #include <linux/ratelimit.h>
41 #include <linux/mount.h>
42 #include <linux/btrfs.h>
43 #include <linux/blkdev.h>
44 #include <linux/posix_acl_xattr.h>
45 #include <linux/uio.h>
48 #include "transaction.h"
49 #include "btrfs_inode.h"
50 #include "print-tree.h"
51 #include "ordered-data.h"
55 #include "compression.h"
57 #include "free-space-cache.h"
58 #include "inode-map.h"
64 struct btrfs_iget_args {
65 struct btrfs_key *location;
66 struct btrfs_root *root;
69 struct btrfs_dio_data {
70 u64 outstanding_extents;
72 u64 unsubmitted_oe_range_start;
73 u64 unsubmitted_oe_range_end;
76 static const struct inode_operations btrfs_dir_inode_operations;
77 static const struct inode_operations btrfs_symlink_inode_operations;
78 static const struct inode_operations btrfs_dir_ro_inode_operations;
79 static const struct inode_operations btrfs_special_inode_operations;
80 static const struct inode_operations btrfs_file_inode_operations;
81 static const struct address_space_operations btrfs_aops;
82 static const struct address_space_operations btrfs_symlink_aops;
83 static const struct file_operations btrfs_dir_file_operations;
84 static const struct extent_io_ops btrfs_extent_io_ops;
86 static struct kmem_cache *btrfs_inode_cachep;
87 struct kmem_cache *btrfs_trans_handle_cachep;
88 struct kmem_cache *btrfs_transaction_cachep;
89 struct kmem_cache *btrfs_path_cachep;
90 struct kmem_cache *btrfs_free_space_cachep;
93 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
94 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
95 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
96 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
97 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
98 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
99 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
100 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
103 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
104 static int btrfs_truncate(struct inode *inode);
105 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
106 static noinline int cow_file_range(struct inode *inode,
107 struct page *locked_page,
108 u64 start, u64 end, int *page_started,
109 unsigned long *nr_written, int unlock);
110 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
111 u64 len, u64 orig_start,
112 u64 block_start, u64 block_len,
113 u64 orig_block_len, u64 ram_bytes,
116 static int btrfs_dirty_inode(struct inode *inode);
118 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
119 void btrfs_test_inode_set_ops(struct inode *inode)
121 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
125 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
126 struct inode *inode, struct inode *dir,
127 const struct qstr *qstr)
131 err = btrfs_init_acl(trans, inode, dir);
133 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
138 * this does all the hard work for inserting an inline extent into
139 * the btree. The caller should have done a btrfs_drop_extents so that
140 * no overlapping inline items exist in the btree
142 static int insert_inline_extent(struct btrfs_trans_handle *trans,
143 struct btrfs_path *path, int extent_inserted,
144 struct btrfs_root *root, struct inode *inode,
145 u64 start, size_t size, size_t compressed_size,
147 struct page **compressed_pages)
149 struct extent_buffer *leaf;
150 struct page *page = NULL;
153 struct btrfs_file_extent_item *ei;
156 size_t cur_size = size;
157 unsigned long offset;
159 if (compressed_size && compressed_pages)
160 cur_size = compressed_size;
162 inode_add_bytes(inode, size);
164 if (!extent_inserted) {
165 struct btrfs_key key;
168 key.objectid = btrfs_ino(inode);
170 key.type = BTRFS_EXTENT_DATA_KEY;
172 datasize = btrfs_file_extent_calc_inline_size(cur_size);
173 path->leave_spinning = 1;
174 ret = btrfs_insert_empty_item(trans, root, path, &key,
181 leaf = path->nodes[0];
182 ei = btrfs_item_ptr(leaf, path->slots[0],
183 struct btrfs_file_extent_item);
184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
185 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
186 btrfs_set_file_extent_encryption(leaf, ei, 0);
187 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
188 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
189 ptr = btrfs_file_extent_inline_start(ei);
191 if (compress_type != BTRFS_COMPRESS_NONE) {
194 while (compressed_size > 0) {
195 cpage = compressed_pages[i];
196 cur_size = min_t(unsigned long, compressed_size,
199 kaddr = kmap_atomic(cpage);
200 write_extent_buffer(leaf, kaddr, ptr, cur_size);
201 kunmap_atomic(kaddr);
205 compressed_size -= cur_size;
207 btrfs_set_file_extent_compression(leaf, ei,
210 page = find_get_page(inode->i_mapping,
211 start >> PAGE_CACHE_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_CACHE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
217 page_cache_release(page);
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
249 struct page **compressed_pages)
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
263 data_len = compressed_size;
266 actual_end > PAGE_CACHE_SIZE ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
269 (actual_end & (root->sectorsize - 1)) == 0) ||
271 data_len > root->fs_info->max_inline) {
275 path = btrfs_alloc_path();
279 trans = btrfs_join_transaction(root);
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
297 btrfs_abort_transaction(trans, root, ret);
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
310 } else if (ret == -ENOSPC) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode, 0, PAGE_CACHE_SIZE);
326 btrfs_free_path(path);
327 btrfs_end_transaction(trans, root);
331 struct async_extent {
336 unsigned long nr_pages;
338 struct list_head list;
343 struct btrfs_root *root;
344 struct page *locked_page;
347 struct list_head extents;
348 struct btrfs_work work;
351 static noinline int add_async_extent(struct async_cow *cow,
352 u64 start, u64 ram_size,
355 unsigned long nr_pages,
358 struct async_extent *async_extent;
360 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
361 BUG_ON(!async_extent); /* -ENOMEM */
362 async_extent->start = start;
363 async_extent->ram_size = ram_size;
364 async_extent->compressed_size = compressed_size;
365 async_extent->pages = pages;
366 async_extent->nr_pages = nr_pages;
367 async_extent->compress_type = compress_type;
368 list_add_tail(&async_extent->list, &cow->extents);
372 static inline int inode_need_compress(struct inode *inode)
374 struct btrfs_root *root = BTRFS_I(inode)->root;
377 if (btrfs_test_opt(root, FORCE_COMPRESS))
379 /* bad compression ratios */
380 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
382 if (btrfs_test_opt(root, COMPRESS) ||
383 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
384 BTRFS_I(inode)->force_compress)
390 * we create compressed extents in two phases. The first
391 * phase compresses a range of pages that have already been
392 * locked (both pages and state bits are locked).
394 * This is done inside an ordered work queue, and the compression
395 * is spread across many cpus. The actual IO submission is step
396 * two, and the ordered work queue takes care of making sure that
397 * happens in the same order things were put onto the queue by
398 * writepages and friends.
400 * If this code finds it can't get good compression, it puts an
401 * entry onto the work queue to write the uncompressed bytes. This
402 * makes sure that both compressed inodes and uncompressed inodes
403 * are written in the same order that the flusher thread sent them
406 static noinline void compress_file_range(struct inode *inode,
407 struct page *locked_page,
409 struct async_cow *async_cow,
412 struct btrfs_root *root = BTRFS_I(inode)->root;
414 u64 blocksize = root->sectorsize;
416 u64 isize = i_size_read(inode);
418 struct page **pages = NULL;
419 unsigned long nr_pages;
420 unsigned long nr_pages_ret = 0;
421 unsigned long total_compressed = 0;
422 unsigned long total_in = 0;
423 unsigned long max_compressed = SZ_128K;
424 unsigned long max_uncompressed = SZ_128K;
427 int compress_type = root->fs_info->compress_type;
430 /* if this is a small write inside eof, kick off a defrag */
431 if ((end - start + 1) < SZ_16K &&
432 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
433 btrfs_add_inode_defrag(NULL, inode);
435 actual_end = min_t(u64, isize, end + 1);
438 nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_CACHE_SIZE);
442 * we don't want to send crud past the end of i_size through
443 * compression, that's just a waste of CPU time. So, if the
444 * end of the file is before the start of our current
445 * requested range of bytes, we bail out to the uncompressed
446 * cleanup code that can deal with all of this.
448 * It isn't really the fastest way to fix things, but this is a
449 * very uncommon corner.
451 if (actual_end <= start)
452 goto cleanup_and_bail_uncompressed;
454 total_compressed = actual_end - start;
457 * skip compression for a small file range(<=blocksize) that
458 * isn't an inline extent, since it dosen't save disk space at all.
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode)) {
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode, start, end);
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
516 unsigned long offset = total_compressed &
517 (PAGE_CACHE_SIZE - 1);
518 struct page *page = pages[nr_pages_ret - 1];
521 /* zero the tail end of the last page, we might be
522 * sending it down to disk
525 kaddr = kmap_atomic(page);
526 memset(kaddr + offset, 0,
527 PAGE_CACHE_SIZE - offset);
528 kunmap_atomic(kaddr);
535 /* lets try to make an inline extent */
536 if (ret || total_in < (actual_end - start)) {
537 /* we didn't compress the entire range, try
538 * to make an uncompressed inline extent.
540 ret = cow_file_range_inline(root, inode, start, end,
543 /* try making a compressed inline extent */
544 ret = cow_file_range_inline(root, inode, start, end,
546 compress_type, pages);
549 unsigned long clear_flags = EXTENT_DELALLOC |
551 unsigned long page_error_op;
553 clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
554 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
557 * inline extent creation worked or returned error,
558 * we don't need to create any more async work items.
559 * Unlock and free up our temp pages.
561 extent_clear_unlock_delalloc(inode, start, end, NULL,
562 clear_flags, PAGE_UNLOCK |
573 * we aren't doing an inline extent round the compressed size
574 * up to a block size boundary so the allocator does sane
577 total_compressed = ALIGN(total_compressed, blocksize);
580 * one last check to make sure the compression is really a
581 * win, compare the page count read with the blocks on disk
583 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
584 if (total_compressed >= total_in) {
587 num_bytes = total_in;
590 if (!will_compress && pages) {
592 * the compression code ran but failed to make things smaller,
593 * free any pages it allocated and our page pointer array
595 for (i = 0; i < nr_pages_ret; i++) {
596 WARN_ON(pages[i]->mapping);
597 page_cache_release(pages[i]);
601 total_compressed = 0;
604 /* flag the file so we don't compress in the future */
605 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
606 !(BTRFS_I(inode)->force_compress)) {
607 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
613 /* the async work queues will take care of doing actual
614 * allocation on disk for these compressed pages,
615 * and will submit them to the elevator.
617 add_async_extent(async_cow, start, num_bytes,
618 total_compressed, pages, nr_pages_ret,
621 if (start + num_bytes < end) {
628 cleanup_and_bail_uncompressed:
630 * No compression, but we still need to write the pages in
631 * the file we've been given so far. redirty the locked
632 * page if it corresponds to our extent and set things up
633 * for the async work queue to run cow_file_range to do
634 * the normal delalloc dance
636 if (page_offset(locked_page) >= start &&
637 page_offset(locked_page) <= end) {
638 __set_page_dirty_nobuffers(locked_page);
639 /* unlocked later on in the async handlers */
642 extent_range_redirty_for_io(inode, start, end);
643 add_async_extent(async_cow, start, end - start + 1,
644 0, NULL, 0, BTRFS_COMPRESS_NONE);
651 for (i = 0; i < nr_pages_ret; i++) {
652 WARN_ON(pages[i]->mapping);
653 page_cache_release(pages[i]);
658 static void free_async_extent_pages(struct async_extent *async_extent)
662 if (!async_extent->pages)
665 for (i = 0; i < async_extent->nr_pages; i++) {
666 WARN_ON(async_extent->pages[i]->mapping);
667 page_cache_release(async_extent->pages[i]);
669 kfree(async_extent->pages);
670 async_extent->nr_pages = 0;
671 async_extent->pages = NULL;
675 * phase two of compressed writeback. This is the ordered portion
676 * of the code, which only gets called in the order the work was
677 * queued. We walk all the async extents created by compress_file_range
678 * and send them down to the disk.
680 static noinline void submit_compressed_extents(struct inode *inode,
681 struct async_cow *async_cow)
683 struct async_extent *async_extent;
685 struct btrfs_key ins;
686 struct extent_map *em;
687 struct btrfs_root *root = BTRFS_I(inode)->root;
688 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
689 struct extent_io_tree *io_tree;
693 while (!list_empty(&async_cow->extents)) {
694 async_extent = list_entry(async_cow->extents.next,
695 struct async_extent, list);
696 list_del(&async_extent->list);
698 io_tree = &BTRFS_I(inode)->io_tree;
701 /* did the compression code fall back to uncompressed IO? */
702 if (!async_extent->pages) {
703 int page_started = 0;
704 unsigned long nr_written = 0;
706 lock_extent(io_tree, async_extent->start,
707 async_extent->start +
708 async_extent->ram_size - 1);
710 /* allocate blocks */
711 ret = cow_file_range(inode, async_cow->locked_page,
713 async_extent->start +
714 async_extent->ram_size - 1,
715 &page_started, &nr_written, 0);
720 * if page_started, cow_file_range inserted an
721 * inline extent and took care of all the unlocking
722 * and IO for us. Otherwise, we need to submit
723 * all those pages down to the drive.
725 if (!page_started && !ret)
726 extent_write_locked_range(io_tree,
727 inode, async_extent->start,
728 async_extent->start +
729 async_extent->ram_size - 1,
733 unlock_page(async_cow->locked_page);
739 lock_extent(io_tree, async_extent->start,
740 async_extent->start + async_extent->ram_size - 1);
742 ret = btrfs_reserve_extent(root,
743 async_extent->compressed_size,
744 async_extent->compressed_size,
745 0, alloc_hint, &ins, 1, 1);
747 free_async_extent_pages(async_extent);
749 if (ret == -ENOSPC) {
750 unlock_extent(io_tree, async_extent->start,
751 async_extent->start +
752 async_extent->ram_size - 1);
755 * we need to redirty the pages if we decide to
756 * fallback to uncompressed IO, otherwise we
757 * will not submit these pages down to lower
760 extent_range_redirty_for_io(inode,
762 async_extent->start +
763 async_extent->ram_size - 1);
770 * here we're doing allocation and writeback of the
773 btrfs_drop_extent_cache(inode, async_extent->start,
774 async_extent->start +
775 async_extent->ram_size - 1, 0);
777 em = alloc_extent_map();
780 goto out_free_reserve;
782 em->start = async_extent->start;
783 em->len = async_extent->ram_size;
784 em->orig_start = em->start;
785 em->mod_start = em->start;
786 em->mod_len = em->len;
788 em->block_start = ins.objectid;
789 em->block_len = ins.offset;
790 em->orig_block_len = ins.offset;
791 em->ram_bytes = async_extent->ram_size;
792 em->bdev = root->fs_info->fs_devices->latest_bdev;
793 em->compress_type = async_extent->compress_type;
794 set_bit(EXTENT_FLAG_PINNED, &em->flags);
795 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
799 write_lock(&em_tree->lock);
800 ret = add_extent_mapping(em_tree, em, 1);
801 write_unlock(&em_tree->lock);
802 if (ret != -EEXIST) {
806 btrfs_drop_extent_cache(inode, async_extent->start,
807 async_extent->start +
808 async_extent->ram_size - 1, 0);
812 goto out_free_reserve;
814 ret = btrfs_add_ordered_extent_compress(inode,
817 async_extent->ram_size,
819 BTRFS_ORDERED_COMPRESSED,
820 async_extent->compress_type);
822 btrfs_drop_extent_cache(inode, async_extent->start,
823 async_extent->start +
824 async_extent->ram_size - 1, 0);
825 goto out_free_reserve;
829 * clear dirty, set writeback and unlock the pages.
831 extent_clear_unlock_delalloc(inode, async_extent->start,
832 async_extent->start +
833 async_extent->ram_size - 1,
834 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
835 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
837 ret = btrfs_submit_compressed_write(inode,
839 async_extent->ram_size,
841 ins.offset, async_extent->pages,
842 async_extent->nr_pages);
844 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
845 struct page *p = async_extent->pages[0];
846 const u64 start = async_extent->start;
847 const u64 end = start + async_extent->ram_size - 1;
849 p->mapping = inode->i_mapping;
850 tree->ops->writepage_end_io_hook(p, start, end,
853 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
856 free_async_extent_pages(async_extent);
858 alloc_hint = ins.objectid + ins.offset;
864 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
866 extent_clear_unlock_delalloc(inode, async_extent->start,
867 async_extent->start +
868 async_extent->ram_size - 1,
869 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
870 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
871 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
872 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
874 free_async_extent_pages(async_extent);
879 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
882 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
883 struct extent_map *em;
886 read_lock(&em_tree->lock);
887 em = search_extent_mapping(em_tree, start, num_bytes);
890 * if block start isn't an actual block number then find the
891 * first block in this inode and use that as a hint. If that
892 * block is also bogus then just don't worry about it.
894 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
896 em = search_extent_mapping(em_tree, 0, 0);
897 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
898 alloc_hint = em->block_start;
902 alloc_hint = em->block_start;
906 read_unlock(&em_tree->lock);
912 * when extent_io.c finds a delayed allocation range in the file,
913 * the call backs end up in this code. The basic idea is to
914 * allocate extents on disk for the range, and create ordered data structs
915 * in ram to track those extents.
917 * locked_page is the page that writepage had locked already. We use
918 * it to make sure we don't do extra locks or unlocks.
920 * *page_started is set to one if we unlock locked_page and do everything
921 * required to start IO on it. It may be clean and already done with
924 static noinline int cow_file_range(struct inode *inode,
925 struct page *locked_page,
926 u64 start, u64 end, int *page_started,
927 unsigned long *nr_written,
930 struct btrfs_root *root = BTRFS_I(inode)->root;
933 unsigned long ram_size;
936 u64 blocksize = root->sectorsize;
937 struct btrfs_key ins;
938 struct extent_map *em;
939 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
942 if (btrfs_is_free_space_inode(inode)) {
948 num_bytes = ALIGN(end - start + 1, blocksize);
949 num_bytes = max(blocksize, num_bytes);
950 disk_num_bytes = num_bytes;
952 /* if this is a small write inside eof, kick off defrag */
953 if (num_bytes < SZ_64K &&
954 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
955 btrfs_add_inode_defrag(NULL, inode);
958 /* lets try to make an inline extent */
959 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
962 extent_clear_unlock_delalloc(inode, start, end, NULL,
963 EXTENT_LOCKED | EXTENT_DELALLOC |
964 EXTENT_DEFRAG, PAGE_UNLOCK |
965 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
968 *nr_written = *nr_written +
969 (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
972 } else if (ret < 0) {
977 BUG_ON(disk_num_bytes >
978 btrfs_super_total_bytes(root->fs_info->super_copy));
980 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
981 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
983 while (disk_num_bytes > 0) {
986 cur_alloc_size = disk_num_bytes;
987 ret = btrfs_reserve_extent(root, cur_alloc_size,
988 root->sectorsize, 0, alloc_hint,
993 em = alloc_extent_map();
999 em->orig_start = em->start;
1000 ram_size = ins.offset;
1001 em->len = ins.offset;
1002 em->mod_start = em->start;
1003 em->mod_len = em->len;
1005 em->block_start = ins.objectid;
1006 em->block_len = ins.offset;
1007 em->orig_block_len = ins.offset;
1008 em->ram_bytes = ram_size;
1009 em->bdev = root->fs_info->fs_devices->latest_bdev;
1010 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1011 em->generation = -1;
1014 write_lock(&em_tree->lock);
1015 ret = add_extent_mapping(em_tree, em, 1);
1016 write_unlock(&em_tree->lock);
1017 if (ret != -EEXIST) {
1018 free_extent_map(em);
1021 btrfs_drop_extent_cache(inode, start,
1022 start + ram_size - 1, 0);
1027 cur_alloc_size = ins.offset;
1028 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1029 ram_size, cur_alloc_size, 0);
1031 goto out_drop_extent_cache;
1033 if (root->root_key.objectid ==
1034 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1035 ret = btrfs_reloc_clone_csums(inode, start,
1038 goto out_drop_extent_cache;
1041 if (disk_num_bytes < cur_alloc_size)
1044 /* we're not doing compressed IO, don't unlock the first
1045 * page (which the caller expects to stay locked), don't
1046 * clear any dirty bits and don't set any writeback bits
1048 * Do set the Private2 bit so we know this page was properly
1049 * setup for writepage
1051 op = unlock ? PAGE_UNLOCK : 0;
1052 op |= PAGE_SET_PRIVATE2;
1054 extent_clear_unlock_delalloc(inode, start,
1055 start + ram_size - 1, locked_page,
1056 EXTENT_LOCKED | EXTENT_DELALLOC,
1058 disk_num_bytes -= cur_alloc_size;
1059 num_bytes -= cur_alloc_size;
1060 alloc_hint = ins.objectid + ins.offset;
1061 start += cur_alloc_size;
1066 out_drop_extent_cache:
1067 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1069 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1071 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1072 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1073 EXTENT_DELALLOC | EXTENT_DEFRAG,
1074 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1075 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1080 * work queue call back to started compression on a file and pages
1082 static noinline void async_cow_start(struct btrfs_work *work)
1084 struct async_cow *async_cow;
1086 async_cow = container_of(work, struct async_cow, work);
1088 compress_file_range(async_cow->inode, async_cow->locked_page,
1089 async_cow->start, async_cow->end, async_cow,
1091 if (num_added == 0) {
1092 btrfs_add_delayed_iput(async_cow->inode);
1093 async_cow->inode = NULL;
1098 * work queue call back to submit previously compressed pages
1100 static noinline void async_cow_submit(struct btrfs_work *work)
1102 struct async_cow *async_cow;
1103 struct btrfs_root *root;
1104 unsigned long nr_pages;
1106 async_cow = container_of(work, struct async_cow, work);
1108 root = async_cow->root;
1109 nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1113 * atomic_sub_return implies a barrier for waitqueue_active
1115 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1117 waitqueue_active(&root->fs_info->async_submit_wait))
1118 wake_up(&root->fs_info->async_submit_wait);
1120 if (async_cow->inode)
1121 submit_compressed_extents(async_cow->inode, async_cow);
1124 static noinline void async_cow_free(struct btrfs_work *work)
1126 struct async_cow *async_cow;
1127 async_cow = container_of(work, struct async_cow, work);
1128 if (async_cow->inode)
1129 btrfs_add_delayed_iput(async_cow->inode);
1133 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1134 u64 start, u64 end, int *page_started,
1135 unsigned long *nr_written)
1137 struct async_cow *async_cow;
1138 struct btrfs_root *root = BTRFS_I(inode)->root;
1139 unsigned long nr_pages;
1141 int limit = 10 * SZ_1M;
1143 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1144 1, 0, NULL, GFP_NOFS);
1145 while (start < end) {
1146 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1147 BUG_ON(!async_cow); /* -ENOMEM */
1148 async_cow->inode = igrab(inode);
1149 async_cow->root = root;
1150 async_cow->locked_page = locked_page;
1151 async_cow->start = start;
1153 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1154 !btrfs_test_opt(root, FORCE_COMPRESS))
1157 cur_end = min(end, start + SZ_512K - 1);
1159 async_cow->end = cur_end;
1160 INIT_LIST_HEAD(&async_cow->extents);
1162 btrfs_init_work(&async_cow->work,
1163 btrfs_delalloc_helper,
1164 async_cow_start, async_cow_submit,
1167 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1169 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1171 btrfs_queue_work(root->fs_info->delalloc_workers,
1174 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1175 wait_event(root->fs_info->async_submit_wait,
1176 (atomic_read(&root->fs_info->async_delalloc_pages) <
1180 while (atomic_read(&root->fs_info->async_submit_draining) &&
1181 atomic_read(&root->fs_info->async_delalloc_pages)) {
1182 wait_event(root->fs_info->async_submit_wait,
1183 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1187 *nr_written += nr_pages;
1188 start = cur_end + 1;
1194 static noinline int csum_exist_in_range(struct btrfs_root *root,
1195 u64 bytenr, u64 num_bytes)
1198 struct btrfs_ordered_sum *sums;
1201 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1202 bytenr + num_bytes - 1, &list, 0);
1203 if (ret == 0 && list_empty(&list))
1206 while (!list_empty(&list)) {
1207 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1208 list_del(&sums->list);
1215 * when nowcow writeback call back. This checks for snapshots or COW copies
1216 * of the extents that exist in the file, and COWs the file as required.
1218 * If no cow copies or snapshots exist, we write directly to the existing
1221 static noinline int run_delalloc_nocow(struct inode *inode,
1222 struct page *locked_page,
1223 u64 start, u64 end, int *page_started, int force,
1224 unsigned long *nr_written)
1226 struct btrfs_root *root = BTRFS_I(inode)->root;
1227 struct btrfs_trans_handle *trans;
1228 struct extent_buffer *leaf;
1229 struct btrfs_path *path;
1230 struct btrfs_file_extent_item *fi;
1231 struct btrfs_key found_key;
1246 u64 ino = btrfs_ino(inode);
1248 path = btrfs_alloc_path();
1250 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1251 EXTENT_LOCKED | EXTENT_DELALLOC |
1252 EXTENT_DO_ACCOUNTING |
1253 EXTENT_DEFRAG, PAGE_UNLOCK |
1255 PAGE_SET_WRITEBACK |
1256 PAGE_END_WRITEBACK);
1260 nolock = btrfs_is_free_space_inode(inode);
1263 trans = btrfs_join_transaction_nolock(root);
1265 trans = btrfs_join_transaction(root);
1267 if (IS_ERR(trans)) {
1268 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1269 EXTENT_LOCKED | EXTENT_DELALLOC |
1270 EXTENT_DO_ACCOUNTING |
1271 EXTENT_DEFRAG, PAGE_UNLOCK |
1273 PAGE_SET_WRITEBACK |
1274 PAGE_END_WRITEBACK);
1275 btrfs_free_path(path);
1276 return PTR_ERR(trans);
1279 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1281 cow_start = (u64)-1;
1284 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1288 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1289 leaf = path->nodes[0];
1290 btrfs_item_key_to_cpu(leaf, &found_key,
1291 path->slots[0] - 1);
1292 if (found_key.objectid == ino &&
1293 found_key.type == BTRFS_EXTENT_DATA_KEY)
1298 leaf = path->nodes[0];
1299 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1300 ret = btrfs_next_leaf(root, path);
1305 leaf = path->nodes[0];
1311 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1313 if (found_key.objectid > ino)
1315 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1316 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1320 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1321 found_key.offset > end)
1324 if (found_key.offset > cur_offset) {
1325 extent_end = found_key.offset;
1330 fi = btrfs_item_ptr(leaf, path->slots[0],
1331 struct btrfs_file_extent_item);
1332 extent_type = btrfs_file_extent_type(leaf, fi);
1334 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1335 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1336 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1337 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1338 extent_offset = btrfs_file_extent_offset(leaf, fi);
1339 extent_end = found_key.offset +
1340 btrfs_file_extent_num_bytes(leaf, fi);
1342 btrfs_file_extent_disk_num_bytes(leaf, fi);
1343 if (extent_end <= start) {
1347 if (disk_bytenr == 0)
1349 if (btrfs_file_extent_compression(leaf, fi) ||
1350 btrfs_file_extent_encryption(leaf, fi) ||
1351 btrfs_file_extent_other_encoding(leaf, fi))
1353 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1355 if (btrfs_extent_readonly(root, disk_bytenr))
1357 if (btrfs_cross_ref_exist(trans, root, ino,
1359 extent_offset, disk_bytenr))
1361 disk_bytenr += extent_offset;
1362 disk_bytenr += cur_offset - found_key.offset;
1363 num_bytes = min(end + 1, extent_end) - cur_offset;
1365 * if there are pending snapshots for this root,
1366 * we fall into common COW way.
1369 err = btrfs_start_write_no_snapshoting(root);
1374 * force cow if csum exists in the range.
1375 * this ensure that csum for a given extent are
1376 * either valid or do not exist.
1378 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1381 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1382 extent_end = found_key.offset +
1383 btrfs_file_extent_inline_len(leaf,
1384 path->slots[0], fi);
1385 extent_end = ALIGN(extent_end, root->sectorsize);
1390 if (extent_end <= start) {
1392 if (!nolock && nocow)
1393 btrfs_end_write_no_snapshoting(root);
1397 if (cow_start == (u64)-1)
1398 cow_start = cur_offset;
1399 cur_offset = extent_end;
1400 if (cur_offset > end)
1406 btrfs_release_path(path);
1407 if (cow_start != (u64)-1) {
1408 ret = cow_file_range(inode, locked_page,
1409 cow_start, found_key.offset - 1,
1410 page_started, nr_written, 1);
1412 if (!nolock && nocow)
1413 btrfs_end_write_no_snapshoting(root);
1416 cow_start = (u64)-1;
1419 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1420 struct extent_map *em;
1421 struct extent_map_tree *em_tree;
1422 em_tree = &BTRFS_I(inode)->extent_tree;
1423 em = alloc_extent_map();
1424 BUG_ON(!em); /* -ENOMEM */
1425 em->start = cur_offset;
1426 em->orig_start = found_key.offset - extent_offset;
1427 em->len = num_bytes;
1428 em->block_len = num_bytes;
1429 em->block_start = disk_bytenr;
1430 em->orig_block_len = disk_num_bytes;
1431 em->ram_bytes = ram_bytes;
1432 em->bdev = root->fs_info->fs_devices->latest_bdev;
1433 em->mod_start = em->start;
1434 em->mod_len = em->len;
1435 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1436 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1437 em->generation = -1;
1439 write_lock(&em_tree->lock);
1440 ret = add_extent_mapping(em_tree, em, 1);
1441 write_unlock(&em_tree->lock);
1442 if (ret != -EEXIST) {
1443 free_extent_map(em);
1446 btrfs_drop_extent_cache(inode, em->start,
1447 em->start + em->len - 1, 0);
1449 type = BTRFS_ORDERED_PREALLOC;
1451 type = BTRFS_ORDERED_NOCOW;
1454 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1455 num_bytes, num_bytes, type);
1456 BUG_ON(ret); /* -ENOMEM */
1458 if (root->root_key.objectid ==
1459 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1460 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1463 if (!nolock && nocow)
1464 btrfs_end_write_no_snapshoting(root);
1469 extent_clear_unlock_delalloc(inode, cur_offset,
1470 cur_offset + num_bytes - 1,
1471 locked_page, EXTENT_LOCKED |
1472 EXTENT_DELALLOC, PAGE_UNLOCK |
1474 if (!nolock && nocow)
1475 btrfs_end_write_no_snapshoting(root);
1476 cur_offset = extent_end;
1477 if (cur_offset > end)
1480 btrfs_release_path(path);
1482 if (cur_offset <= end && cow_start == (u64)-1) {
1483 cow_start = cur_offset;
1487 if (cow_start != (u64)-1) {
1488 ret = cow_file_range(inode, locked_page, cow_start, end,
1489 page_started, nr_written, 1);
1495 err = btrfs_end_transaction(trans, root);
1499 if (ret && cur_offset < end)
1500 extent_clear_unlock_delalloc(inode, cur_offset, end,
1501 locked_page, EXTENT_LOCKED |
1502 EXTENT_DELALLOC | EXTENT_DEFRAG |
1503 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1505 PAGE_SET_WRITEBACK |
1506 PAGE_END_WRITEBACK);
1507 btrfs_free_path(path);
1511 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1514 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1515 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1519 * @defrag_bytes is a hint value, no spinlock held here,
1520 * if is not zero, it means the file is defragging.
1521 * Force cow if given extent needs to be defragged.
1523 if (BTRFS_I(inode)->defrag_bytes &&
1524 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1525 EXTENT_DEFRAG, 0, NULL))
1532 * extent_io.c call back to do delayed allocation processing
1534 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1535 u64 start, u64 end, int *page_started,
1536 unsigned long *nr_written)
1539 int force_cow = need_force_cow(inode, start, end);
1541 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1542 ret = run_delalloc_nocow(inode, locked_page, start, end,
1543 page_started, 1, nr_written);
1544 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1545 ret = run_delalloc_nocow(inode, locked_page, start, end,
1546 page_started, 0, nr_written);
1547 } else if (!inode_need_compress(inode)) {
1548 ret = cow_file_range(inode, locked_page, start, end,
1549 page_started, nr_written, 1);
1551 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1552 &BTRFS_I(inode)->runtime_flags);
1553 ret = cow_file_range_async(inode, locked_page, start, end,
1554 page_started, nr_written);
1559 static void btrfs_split_extent_hook(struct inode *inode,
1560 struct extent_state *orig, u64 split)
1564 /* not delalloc, ignore it */
1565 if (!(orig->state & EXTENT_DELALLOC))
1568 size = orig->end - orig->start + 1;
1569 if (size > BTRFS_MAX_EXTENT_SIZE) {
1574 * See the explanation in btrfs_merge_extent_hook, the same
1575 * applies here, just in reverse.
1577 new_size = orig->end - split + 1;
1578 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1579 BTRFS_MAX_EXTENT_SIZE);
1580 new_size = split - orig->start;
1581 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1582 BTRFS_MAX_EXTENT_SIZE);
1583 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1584 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1588 spin_lock(&BTRFS_I(inode)->lock);
1589 BTRFS_I(inode)->outstanding_extents++;
1590 spin_unlock(&BTRFS_I(inode)->lock);
1594 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1595 * extents so we can keep track of new extents that are just merged onto old
1596 * extents, such as when we are doing sequential writes, so we can properly
1597 * account for the metadata space we'll need.
1599 static void btrfs_merge_extent_hook(struct inode *inode,
1600 struct extent_state *new,
1601 struct extent_state *other)
1603 u64 new_size, old_size;
1606 /* not delalloc, ignore it */
1607 if (!(other->state & EXTENT_DELALLOC))
1610 if (new->start > other->start)
1611 new_size = new->end - other->start + 1;
1613 new_size = other->end - new->start + 1;
1615 /* we're not bigger than the max, unreserve the space and go */
1616 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1617 spin_lock(&BTRFS_I(inode)->lock);
1618 BTRFS_I(inode)->outstanding_extents--;
1619 spin_unlock(&BTRFS_I(inode)->lock);
1624 * We have to add up either side to figure out how many extents were
1625 * accounted for before we merged into one big extent. If the number of
1626 * extents we accounted for is <= the amount we need for the new range
1627 * then we can return, otherwise drop. Think of it like this
1631 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1632 * need 2 outstanding extents, on one side we have 1 and the other side
1633 * we have 1 so they are == and we can return. But in this case
1635 * [MAX_SIZE+4k][MAX_SIZE+4k]
1637 * Each range on their own accounts for 2 extents, but merged together
1638 * they are only 3 extents worth of accounting, so we need to drop in
1641 old_size = other->end - other->start + 1;
1642 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1643 BTRFS_MAX_EXTENT_SIZE);
1644 old_size = new->end - new->start + 1;
1645 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1646 BTRFS_MAX_EXTENT_SIZE);
1648 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1649 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1652 spin_lock(&BTRFS_I(inode)->lock);
1653 BTRFS_I(inode)->outstanding_extents--;
1654 spin_unlock(&BTRFS_I(inode)->lock);
1657 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1658 struct inode *inode)
1660 spin_lock(&root->delalloc_lock);
1661 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1662 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1663 &root->delalloc_inodes);
1664 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1665 &BTRFS_I(inode)->runtime_flags);
1666 root->nr_delalloc_inodes++;
1667 if (root->nr_delalloc_inodes == 1) {
1668 spin_lock(&root->fs_info->delalloc_root_lock);
1669 BUG_ON(!list_empty(&root->delalloc_root));
1670 list_add_tail(&root->delalloc_root,
1671 &root->fs_info->delalloc_roots);
1672 spin_unlock(&root->fs_info->delalloc_root_lock);
1675 spin_unlock(&root->delalloc_lock);
1678 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1679 struct inode *inode)
1681 spin_lock(&root->delalloc_lock);
1682 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1683 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1684 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1685 &BTRFS_I(inode)->runtime_flags);
1686 root->nr_delalloc_inodes--;
1687 if (!root->nr_delalloc_inodes) {
1688 spin_lock(&root->fs_info->delalloc_root_lock);
1689 BUG_ON(list_empty(&root->delalloc_root));
1690 list_del_init(&root->delalloc_root);
1691 spin_unlock(&root->fs_info->delalloc_root_lock);
1694 spin_unlock(&root->delalloc_lock);
1698 * extent_io.c set_bit_hook, used to track delayed allocation
1699 * bytes in this file, and to maintain the list of inodes that
1700 * have pending delalloc work to be done.
1702 static void btrfs_set_bit_hook(struct inode *inode,
1703 struct extent_state *state, unsigned *bits)
1706 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1709 * set_bit and clear bit hooks normally require _irqsave/restore
1710 * but in this case, we are only testing for the DELALLOC
1711 * bit, which is only set or cleared with irqs on
1713 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1714 struct btrfs_root *root = BTRFS_I(inode)->root;
1715 u64 len = state->end + 1 - state->start;
1716 bool do_list = !btrfs_is_free_space_inode(inode);
1718 if (*bits & EXTENT_FIRST_DELALLOC) {
1719 *bits &= ~EXTENT_FIRST_DELALLOC;
1721 spin_lock(&BTRFS_I(inode)->lock);
1722 BTRFS_I(inode)->outstanding_extents++;
1723 spin_unlock(&BTRFS_I(inode)->lock);
1726 /* For sanity tests */
1727 if (btrfs_test_is_dummy_root(root))
1730 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1731 root->fs_info->delalloc_batch);
1732 spin_lock(&BTRFS_I(inode)->lock);
1733 BTRFS_I(inode)->delalloc_bytes += len;
1734 if (*bits & EXTENT_DEFRAG)
1735 BTRFS_I(inode)->defrag_bytes += len;
1736 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &BTRFS_I(inode)->runtime_flags))
1738 btrfs_add_delalloc_inodes(root, inode);
1739 spin_unlock(&BTRFS_I(inode)->lock);
1744 * extent_io.c clear_bit_hook, see set_bit_hook for why
1746 static void btrfs_clear_bit_hook(struct inode *inode,
1747 struct extent_state *state,
1750 u64 len = state->end + 1 - state->start;
1751 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1752 BTRFS_MAX_EXTENT_SIZE);
1754 spin_lock(&BTRFS_I(inode)->lock);
1755 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1756 BTRFS_I(inode)->defrag_bytes -= len;
1757 spin_unlock(&BTRFS_I(inode)->lock);
1760 * set_bit and clear bit hooks normally require _irqsave/restore
1761 * but in this case, we are only testing for the DELALLOC
1762 * bit, which is only set or cleared with irqs on
1764 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1765 struct btrfs_root *root = BTRFS_I(inode)->root;
1766 bool do_list = !btrfs_is_free_space_inode(inode);
1768 if (*bits & EXTENT_FIRST_DELALLOC) {
1769 *bits &= ~EXTENT_FIRST_DELALLOC;
1770 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1771 spin_lock(&BTRFS_I(inode)->lock);
1772 BTRFS_I(inode)->outstanding_extents -= num_extents;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1777 * We don't reserve metadata space for space cache inodes so we
1778 * don't need to call dellalloc_release_metadata if there is an
1781 if (*bits & EXTENT_DO_ACCOUNTING &&
1782 root != root->fs_info->tree_root)
1783 btrfs_delalloc_release_metadata(inode, len);
1785 /* For sanity tests. */
1786 if (btrfs_test_is_dummy_root(root))
1789 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1790 && do_list && !(state->state & EXTENT_NORESERVE))
1791 btrfs_free_reserved_data_space_noquota(inode,
1794 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1795 root->fs_info->delalloc_batch);
1796 spin_lock(&BTRFS_I(inode)->lock);
1797 BTRFS_I(inode)->delalloc_bytes -= len;
1798 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1799 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1800 &BTRFS_I(inode)->runtime_flags))
1801 btrfs_del_delalloc_inode(root, inode);
1802 spin_unlock(&BTRFS_I(inode)->lock);
1807 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1808 * we don't create bios that span stripes or chunks
1810 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1811 size_t size, struct bio *bio,
1812 unsigned long bio_flags)
1814 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1815 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1820 if (bio_flags & EXTENT_BIO_COMPRESSED)
1823 length = bio->bi_iter.bi_size;
1824 map_length = length;
1825 ret = btrfs_map_block(root->fs_info, rw, logical,
1826 &map_length, NULL, 0);
1827 /* Will always return 0 with map_multi == NULL */
1829 if (map_length < length + size)
1835 * in order to insert checksums into the metadata in large chunks,
1836 * we wait until bio submission time. All the pages in the bio are
1837 * checksummed and sums are attached onto the ordered extent record.
1839 * At IO completion time the cums attached on the ordered extent record
1840 * are inserted into the btree
1842 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1843 struct bio *bio, int mirror_num,
1844 unsigned long bio_flags,
1847 struct btrfs_root *root = BTRFS_I(inode)->root;
1850 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1851 BUG_ON(ret); /* -ENOMEM */
1856 * in order to insert checksums into the metadata in large chunks,
1857 * we wait until bio submission time. All the pages in the bio are
1858 * checksummed and sums are attached onto the ordered extent record.
1860 * At IO completion time the cums attached on the ordered extent record
1861 * are inserted into the btree
1863 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1864 int mirror_num, unsigned long bio_flags,
1867 struct btrfs_root *root = BTRFS_I(inode)->root;
1870 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1872 bio->bi_error = ret;
1879 * extent_io.c submission hook. This does the right thing for csum calculation
1880 * on write, or reading the csums from the tree before a read
1882 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1883 int mirror_num, unsigned long bio_flags,
1886 struct btrfs_root *root = BTRFS_I(inode)->root;
1887 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1890 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1892 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1894 if (btrfs_is_free_space_inode(inode))
1895 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1897 if (!(rw & REQ_WRITE)) {
1898 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1902 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1903 ret = btrfs_submit_compressed_read(inode, bio,
1907 } else if (!skip_sum) {
1908 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1913 } else if (async && !skip_sum) {
1914 /* csum items have already been cloned */
1915 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1917 /* we're doing a write, do the async checksumming */
1918 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1919 inode, rw, bio, mirror_num,
1920 bio_flags, bio_offset,
1921 __btrfs_submit_bio_start,
1922 __btrfs_submit_bio_done);
1924 } else if (!skip_sum) {
1925 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1931 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1935 bio->bi_error = ret;
1942 * given a list of ordered sums record them in the inode. This happens
1943 * at IO completion time based on sums calculated at bio submission time.
1945 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1946 struct inode *inode, u64 file_offset,
1947 struct list_head *list)
1949 struct btrfs_ordered_sum *sum;
1951 list_for_each_entry(sum, list, list) {
1952 trans->adding_csums = 1;
1953 btrfs_csum_file_blocks(trans,
1954 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1955 trans->adding_csums = 0;
1960 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1961 struct extent_state **cached_state)
1963 WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1964 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1965 cached_state, GFP_NOFS);
1968 /* see btrfs_writepage_start_hook for details on why this is required */
1969 struct btrfs_writepage_fixup {
1971 struct btrfs_work work;
1974 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1976 struct btrfs_writepage_fixup *fixup;
1977 struct btrfs_ordered_extent *ordered;
1978 struct extent_state *cached_state = NULL;
1980 struct inode *inode;
1985 fixup = container_of(work, struct btrfs_writepage_fixup, work);
1989 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1990 ClearPageChecked(page);
1994 inode = page->mapping->host;
1995 page_start = page_offset(page);
1996 page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1998 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2001 /* already ordered? We're done */
2002 if (PagePrivate2(page))
2005 ordered = btrfs_lookup_ordered_extent(inode, page_start);
2007 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2008 page_end, &cached_state, GFP_NOFS);
2010 btrfs_start_ordered_extent(inode, ordered, 1);
2011 btrfs_put_ordered_extent(ordered);
2015 ret = btrfs_delalloc_reserve_space(inode, page_start,
2018 mapping_set_error(page->mapping, ret);
2019 end_extent_writepage(page, ret, page_start, page_end);
2020 ClearPageChecked(page);
2024 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2025 ClearPageChecked(page);
2026 set_page_dirty(page);
2028 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2029 &cached_state, GFP_NOFS);
2032 page_cache_release(page);
2037 * There are a few paths in the higher layers of the kernel that directly
2038 * set the page dirty bit without asking the filesystem if it is a
2039 * good idea. This causes problems because we want to make sure COW
2040 * properly happens and the data=ordered rules are followed.
2042 * In our case any range that doesn't have the ORDERED bit set
2043 * hasn't been properly setup for IO. We kick off an async process
2044 * to fix it up. The async helper will wait for ordered extents, set
2045 * the delalloc bit and make it safe to write the page.
2047 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2049 struct inode *inode = page->mapping->host;
2050 struct btrfs_writepage_fixup *fixup;
2051 struct btrfs_root *root = BTRFS_I(inode)->root;
2053 /* this page is properly in the ordered list */
2054 if (TestClearPagePrivate2(page))
2057 if (PageChecked(page))
2060 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2064 SetPageChecked(page);
2065 page_cache_get(page);
2066 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2067 btrfs_writepage_fixup_worker, NULL, NULL);
2069 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2073 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2074 struct inode *inode, u64 file_pos,
2075 u64 disk_bytenr, u64 disk_num_bytes,
2076 u64 num_bytes, u64 ram_bytes,
2077 u8 compression, u8 encryption,
2078 u16 other_encoding, int extent_type)
2080 struct btrfs_root *root = BTRFS_I(inode)->root;
2081 struct btrfs_file_extent_item *fi;
2082 struct btrfs_path *path;
2083 struct extent_buffer *leaf;
2084 struct btrfs_key ins;
2085 int extent_inserted = 0;
2088 path = btrfs_alloc_path();
2093 * we may be replacing one extent in the tree with another.
2094 * The new extent is pinned in the extent map, and we don't want
2095 * to drop it from the cache until it is completely in the btree.
2097 * So, tell btrfs_drop_extents to leave this extent in the cache.
2098 * the caller is expected to unpin it and allow it to be merged
2101 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2102 file_pos + num_bytes, NULL, 0,
2103 1, sizeof(*fi), &extent_inserted);
2107 if (!extent_inserted) {
2108 ins.objectid = btrfs_ino(inode);
2109 ins.offset = file_pos;
2110 ins.type = BTRFS_EXTENT_DATA_KEY;
2112 path->leave_spinning = 1;
2113 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2118 leaf = path->nodes[0];
2119 fi = btrfs_item_ptr(leaf, path->slots[0],
2120 struct btrfs_file_extent_item);
2121 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2122 btrfs_set_file_extent_type(leaf, fi, extent_type);
2123 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2124 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2125 btrfs_set_file_extent_offset(leaf, fi, 0);
2126 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2127 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2128 btrfs_set_file_extent_compression(leaf, fi, compression);
2129 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2130 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2132 btrfs_mark_buffer_dirty(leaf);
2133 btrfs_release_path(path);
2135 inode_add_bytes(inode, num_bytes);
2137 ins.objectid = disk_bytenr;
2138 ins.offset = disk_num_bytes;
2139 ins.type = BTRFS_EXTENT_ITEM_KEY;
2140 ret = btrfs_alloc_reserved_file_extent(trans, root,
2141 root->root_key.objectid,
2142 btrfs_ino(inode), file_pos,
2145 * Release the reserved range from inode dirty range map, as it is
2146 * already moved into delayed_ref_head
2148 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2150 btrfs_free_path(path);
2155 /* snapshot-aware defrag */
2156 struct sa_defrag_extent_backref {
2157 struct rb_node node;
2158 struct old_sa_defrag_extent *old;
2167 struct old_sa_defrag_extent {
2168 struct list_head list;
2169 struct new_sa_defrag_extent *new;
2178 struct new_sa_defrag_extent {
2179 struct rb_root root;
2180 struct list_head head;
2181 struct btrfs_path *path;
2182 struct inode *inode;
2190 static int backref_comp(struct sa_defrag_extent_backref *b1,
2191 struct sa_defrag_extent_backref *b2)
2193 if (b1->root_id < b2->root_id)
2195 else if (b1->root_id > b2->root_id)
2198 if (b1->inum < b2->inum)
2200 else if (b1->inum > b2->inum)
2203 if (b1->file_pos < b2->file_pos)
2205 else if (b1->file_pos > b2->file_pos)
2209 * [------------------------------] ===> (a range of space)
2210 * |<--->| |<---->| =============> (fs/file tree A)
2211 * |<---------------------------->| ===> (fs/file tree B)
2213 * A range of space can refer to two file extents in one tree while
2214 * refer to only one file extent in another tree.
2216 * So we may process a disk offset more than one time(two extents in A)
2217 * and locate at the same extent(one extent in B), then insert two same
2218 * backrefs(both refer to the extent in B).
2223 static void backref_insert(struct rb_root *root,
2224 struct sa_defrag_extent_backref *backref)
2226 struct rb_node **p = &root->rb_node;
2227 struct rb_node *parent = NULL;
2228 struct sa_defrag_extent_backref *entry;
2233 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2235 ret = backref_comp(backref, entry);
2239 p = &(*p)->rb_right;
2242 rb_link_node(&backref->node, parent, p);
2243 rb_insert_color(&backref->node, root);
2247 * Note the backref might has changed, and in this case we just return 0.
2249 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2252 struct btrfs_file_extent_item *extent;
2253 struct btrfs_fs_info *fs_info;
2254 struct old_sa_defrag_extent *old = ctx;
2255 struct new_sa_defrag_extent *new = old->new;
2256 struct btrfs_path *path = new->path;
2257 struct btrfs_key key;
2258 struct btrfs_root *root;
2259 struct sa_defrag_extent_backref *backref;
2260 struct extent_buffer *leaf;
2261 struct inode *inode = new->inode;
2267 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2268 inum == btrfs_ino(inode))
2271 key.objectid = root_id;
2272 key.type = BTRFS_ROOT_ITEM_KEY;
2273 key.offset = (u64)-1;
2275 fs_info = BTRFS_I(inode)->root->fs_info;
2276 root = btrfs_read_fs_root_no_name(fs_info, &key);
2278 if (PTR_ERR(root) == -ENOENT)
2281 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2282 inum, offset, root_id);
2283 return PTR_ERR(root);
2286 key.objectid = inum;
2287 key.type = BTRFS_EXTENT_DATA_KEY;
2288 if (offset > (u64)-1 << 32)
2291 key.offset = offset;
2293 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2294 if (WARN_ON(ret < 0))
2301 leaf = path->nodes[0];
2302 slot = path->slots[0];
2304 if (slot >= btrfs_header_nritems(leaf)) {
2305 ret = btrfs_next_leaf(root, path);
2308 } else if (ret > 0) {
2317 btrfs_item_key_to_cpu(leaf, &key, slot);
2319 if (key.objectid > inum)
2322 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2325 extent = btrfs_item_ptr(leaf, slot,
2326 struct btrfs_file_extent_item);
2328 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2332 * 'offset' refers to the exact key.offset,
2333 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2334 * (key.offset - extent_offset).
2336 if (key.offset != offset)
2339 extent_offset = btrfs_file_extent_offset(leaf, extent);
2340 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2342 if (extent_offset >= old->extent_offset + old->offset +
2343 old->len || extent_offset + num_bytes <=
2344 old->extent_offset + old->offset)
2349 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2355 backref->root_id = root_id;
2356 backref->inum = inum;
2357 backref->file_pos = offset;
2358 backref->num_bytes = num_bytes;
2359 backref->extent_offset = extent_offset;
2360 backref->generation = btrfs_file_extent_generation(leaf, extent);
2362 backref_insert(&new->root, backref);
2365 btrfs_release_path(path);
2370 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2371 struct new_sa_defrag_extent *new)
2373 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2374 struct old_sa_defrag_extent *old, *tmp;
2379 list_for_each_entry_safe(old, tmp, &new->head, list) {
2380 ret = iterate_inodes_from_logical(old->bytenr +
2381 old->extent_offset, fs_info,
2382 path, record_one_backref,
2384 if (ret < 0 && ret != -ENOENT)
2387 /* no backref to be processed for this extent */
2389 list_del(&old->list);
2394 if (list_empty(&new->head))
2400 static int relink_is_mergable(struct extent_buffer *leaf,
2401 struct btrfs_file_extent_item *fi,
2402 struct new_sa_defrag_extent *new)
2404 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2407 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2410 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2413 if (btrfs_file_extent_encryption(leaf, fi) ||
2414 btrfs_file_extent_other_encoding(leaf, fi))
2421 * Note the backref might has changed, and in this case we just return 0.
2423 static noinline int relink_extent_backref(struct btrfs_path *path,
2424 struct sa_defrag_extent_backref *prev,
2425 struct sa_defrag_extent_backref *backref)
2427 struct btrfs_file_extent_item *extent;
2428 struct btrfs_file_extent_item *item;
2429 struct btrfs_ordered_extent *ordered;
2430 struct btrfs_trans_handle *trans;
2431 struct btrfs_fs_info *fs_info;
2432 struct btrfs_root *root;
2433 struct btrfs_key key;
2434 struct extent_buffer *leaf;
2435 struct old_sa_defrag_extent *old = backref->old;
2436 struct new_sa_defrag_extent *new = old->new;
2437 struct inode *src_inode = new->inode;
2438 struct inode *inode;
2439 struct extent_state *cached = NULL;
2448 if (prev && prev->root_id == backref->root_id &&
2449 prev->inum == backref->inum &&
2450 prev->file_pos + prev->num_bytes == backref->file_pos)
2453 /* step 1: get root */
2454 key.objectid = backref->root_id;
2455 key.type = BTRFS_ROOT_ITEM_KEY;
2456 key.offset = (u64)-1;
2458 fs_info = BTRFS_I(src_inode)->root->fs_info;
2459 index = srcu_read_lock(&fs_info->subvol_srcu);
2461 root = btrfs_read_fs_root_no_name(fs_info, &key);
2463 srcu_read_unlock(&fs_info->subvol_srcu, index);
2464 if (PTR_ERR(root) == -ENOENT)
2466 return PTR_ERR(root);
2469 if (btrfs_root_readonly(root)) {
2470 srcu_read_unlock(&fs_info->subvol_srcu, index);
2474 /* step 2: get inode */
2475 key.objectid = backref->inum;
2476 key.type = BTRFS_INODE_ITEM_KEY;
2479 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2480 if (IS_ERR(inode)) {
2481 srcu_read_unlock(&fs_info->subvol_srcu, index);
2485 srcu_read_unlock(&fs_info->subvol_srcu, index);
2487 /* step 3: relink backref */
2488 lock_start = backref->file_pos;
2489 lock_end = backref->file_pos + backref->num_bytes - 1;
2490 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2493 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2495 btrfs_put_ordered_extent(ordered);
2499 trans = btrfs_join_transaction(root);
2500 if (IS_ERR(trans)) {
2501 ret = PTR_ERR(trans);
2505 key.objectid = backref->inum;
2506 key.type = BTRFS_EXTENT_DATA_KEY;
2507 key.offset = backref->file_pos;
2509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512 } else if (ret > 0) {
2517 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2518 struct btrfs_file_extent_item);
2520 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2521 backref->generation)
2524 btrfs_release_path(path);
2526 start = backref->file_pos;
2527 if (backref->extent_offset < old->extent_offset + old->offset)
2528 start += old->extent_offset + old->offset -
2529 backref->extent_offset;
2531 len = min(backref->extent_offset + backref->num_bytes,
2532 old->extent_offset + old->offset + old->len);
2533 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2535 ret = btrfs_drop_extents(trans, root, inode, start,
2540 key.objectid = btrfs_ino(inode);
2541 key.type = BTRFS_EXTENT_DATA_KEY;
2544 path->leave_spinning = 1;
2546 struct btrfs_file_extent_item *fi;
2548 struct btrfs_key found_key;
2550 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2555 leaf = path->nodes[0];
2556 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2558 fi = btrfs_item_ptr(leaf, path->slots[0],
2559 struct btrfs_file_extent_item);
2560 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2562 if (extent_len + found_key.offset == start &&
2563 relink_is_mergable(leaf, fi, new)) {
2564 btrfs_set_file_extent_num_bytes(leaf, fi,
2566 btrfs_mark_buffer_dirty(leaf);
2567 inode_add_bytes(inode, len);
2573 btrfs_release_path(path);
2578 ret = btrfs_insert_empty_item(trans, root, path, &key,
2581 btrfs_abort_transaction(trans, root, ret);
2585 leaf = path->nodes[0];
2586 item = btrfs_item_ptr(leaf, path->slots[0],
2587 struct btrfs_file_extent_item);
2588 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2589 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2590 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2591 btrfs_set_file_extent_num_bytes(leaf, item, len);
2592 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2593 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2594 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2595 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2596 btrfs_set_file_extent_encryption(leaf, item, 0);
2597 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2599 btrfs_mark_buffer_dirty(leaf);
2600 inode_add_bytes(inode, len);
2601 btrfs_release_path(path);
2603 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2605 backref->root_id, backref->inum,
2606 new->file_pos); /* start - extent_offset */
2608 btrfs_abort_transaction(trans, root, ret);
2614 btrfs_release_path(path);
2615 path->leave_spinning = 0;
2616 btrfs_end_transaction(trans, root);
2618 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2624 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2626 struct old_sa_defrag_extent *old, *tmp;
2631 list_for_each_entry_safe(old, tmp, &new->head, list) {
2637 static void relink_file_extents(struct new_sa_defrag_extent *new)
2639 struct btrfs_path *path;
2640 struct sa_defrag_extent_backref *backref;
2641 struct sa_defrag_extent_backref *prev = NULL;
2642 struct inode *inode;
2643 struct btrfs_root *root;
2644 struct rb_node *node;
2648 root = BTRFS_I(inode)->root;
2650 path = btrfs_alloc_path();
2654 if (!record_extent_backrefs(path, new)) {
2655 btrfs_free_path(path);
2658 btrfs_release_path(path);
2661 node = rb_first(&new->root);
2664 rb_erase(node, &new->root);
2666 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2668 ret = relink_extent_backref(path, prev, backref);
2681 btrfs_free_path(path);
2683 free_sa_defrag_extent(new);
2685 atomic_dec(&root->fs_info->defrag_running);
2686 wake_up(&root->fs_info->transaction_wait);
2689 static struct new_sa_defrag_extent *
2690 record_old_file_extents(struct inode *inode,
2691 struct btrfs_ordered_extent *ordered)
2693 struct btrfs_root *root = BTRFS_I(inode)->root;
2694 struct btrfs_path *path;
2695 struct btrfs_key key;
2696 struct old_sa_defrag_extent *old;
2697 struct new_sa_defrag_extent *new;
2700 new = kmalloc(sizeof(*new), GFP_NOFS);
2705 new->file_pos = ordered->file_offset;
2706 new->len = ordered->len;
2707 new->bytenr = ordered->start;
2708 new->disk_len = ordered->disk_len;
2709 new->compress_type = ordered->compress_type;
2710 new->root = RB_ROOT;
2711 INIT_LIST_HEAD(&new->head);
2713 path = btrfs_alloc_path();
2717 key.objectid = btrfs_ino(inode);
2718 key.type = BTRFS_EXTENT_DATA_KEY;
2719 key.offset = new->file_pos;
2721 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2724 if (ret > 0 && path->slots[0] > 0)
2727 /* find out all the old extents for the file range */
2729 struct btrfs_file_extent_item *extent;
2730 struct extent_buffer *l;
2739 slot = path->slots[0];
2741 if (slot >= btrfs_header_nritems(l)) {
2742 ret = btrfs_next_leaf(root, path);
2750 btrfs_item_key_to_cpu(l, &key, slot);
2752 if (key.objectid != btrfs_ino(inode))
2754 if (key.type != BTRFS_EXTENT_DATA_KEY)
2756 if (key.offset >= new->file_pos + new->len)
2759 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2761 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2762 if (key.offset + num_bytes < new->file_pos)
2765 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2769 extent_offset = btrfs_file_extent_offset(l, extent);
2771 old = kmalloc(sizeof(*old), GFP_NOFS);
2775 offset = max(new->file_pos, key.offset);
2776 end = min(new->file_pos + new->len, key.offset + num_bytes);
2778 old->bytenr = disk_bytenr;
2779 old->extent_offset = extent_offset;
2780 old->offset = offset - key.offset;
2781 old->len = end - offset;
2784 list_add_tail(&old->list, &new->head);
2790 btrfs_free_path(path);
2791 atomic_inc(&root->fs_info->defrag_running);
2796 btrfs_free_path(path);
2798 free_sa_defrag_extent(new);
2802 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2805 struct btrfs_block_group_cache *cache;
2807 cache = btrfs_lookup_block_group(root->fs_info, start);
2810 spin_lock(&cache->lock);
2811 cache->delalloc_bytes -= len;
2812 spin_unlock(&cache->lock);
2814 btrfs_put_block_group(cache);
2817 /* as ordered data IO finishes, this gets called so we can finish
2818 * an ordered extent if the range of bytes in the file it covers are
2821 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2823 struct inode *inode = ordered_extent->inode;
2824 struct btrfs_root *root = BTRFS_I(inode)->root;
2825 struct btrfs_trans_handle *trans = NULL;
2826 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2827 struct extent_state *cached_state = NULL;
2828 struct new_sa_defrag_extent *new = NULL;
2829 int compress_type = 0;
2831 u64 logical_len = ordered_extent->len;
2833 bool truncated = false;
2835 nolock = btrfs_is_free_space_inode(inode);
2837 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2842 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2843 ordered_extent->file_offset +
2844 ordered_extent->len - 1);
2846 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2848 logical_len = ordered_extent->truncated_len;
2849 /* Truncated the entire extent, don't bother adding */
2854 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2855 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2858 * For mwrite(mmap + memset to write) case, we still reserve
2859 * space for NOCOW range.
2860 * As NOCOW won't cause a new delayed ref, just free the space
2862 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2863 ordered_extent->len);
2864 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2866 trans = btrfs_join_transaction_nolock(root);
2868 trans = btrfs_join_transaction(root);
2869 if (IS_ERR(trans)) {
2870 ret = PTR_ERR(trans);
2874 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2875 ret = btrfs_update_inode_fallback(trans, root, inode);
2876 if (ret) /* -ENOMEM or corruption */
2877 btrfs_abort_transaction(trans, root, ret);
2881 lock_extent_bits(io_tree, ordered_extent->file_offset,
2882 ordered_extent->file_offset + ordered_extent->len - 1,
2885 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2886 ordered_extent->file_offset + ordered_extent->len - 1,
2887 EXTENT_DEFRAG, 1, cached_state);
2889 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2890 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2891 /* the inode is shared */
2892 new = record_old_file_extents(inode, ordered_extent);
2894 clear_extent_bit(io_tree, ordered_extent->file_offset,
2895 ordered_extent->file_offset + ordered_extent->len - 1,
2896 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2900 trans = btrfs_join_transaction_nolock(root);
2902 trans = btrfs_join_transaction(root);
2903 if (IS_ERR(trans)) {
2904 ret = PTR_ERR(trans);
2909 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2911 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2912 compress_type = ordered_extent->compress_type;
2913 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2914 BUG_ON(compress_type);
2915 ret = btrfs_mark_extent_written(trans, inode,
2916 ordered_extent->file_offset,
2917 ordered_extent->file_offset +
2920 BUG_ON(root == root->fs_info->tree_root);
2921 ret = insert_reserved_file_extent(trans, inode,
2922 ordered_extent->file_offset,
2923 ordered_extent->start,
2924 ordered_extent->disk_len,
2925 logical_len, logical_len,
2926 compress_type, 0, 0,
2927 BTRFS_FILE_EXTENT_REG);
2929 btrfs_release_delalloc_bytes(root,
2930 ordered_extent->start,
2931 ordered_extent->disk_len);
2933 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2934 ordered_extent->file_offset, ordered_extent->len,
2937 btrfs_abort_transaction(trans, root, ret);
2941 add_pending_csums(trans, inode, ordered_extent->file_offset,
2942 &ordered_extent->list);
2944 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2945 ret = btrfs_update_inode_fallback(trans, root, inode);
2946 if (ret) { /* -ENOMEM or corruption */
2947 btrfs_abort_transaction(trans, root, ret);
2952 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2953 ordered_extent->file_offset +
2954 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2956 if (root != root->fs_info->tree_root)
2957 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2959 btrfs_end_transaction(trans, root);
2961 if (ret || truncated) {
2965 start = ordered_extent->file_offset + logical_len;
2967 start = ordered_extent->file_offset;
2968 end = ordered_extent->file_offset + ordered_extent->len - 1;
2969 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2971 /* Drop the cache for the part of the extent we didn't write. */
2972 btrfs_drop_extent_cache(inode, start, end, 0);
2975 * If the ordered extent had an IOERR or something else went
2976 * wrong we need to return the space for this ordered extent
2977 * back to the allocator. We only free the extent in the
2978 * truncated case if we didn't write out the extent at all.
2980 if ((ret || !logical_len) &&
2981 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2982 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2983 btrfs_free_reserved_extent(root, ordered_extent->start,
2984 ordered_extent->disk_len, 1);
2989 * This needs to be done to make sure anybody waiting knows we are done
2990 * updating everything for this ordered extent.
2992 btrfs_remove_ordered_extent(inode, ordered_extent);
2994 /* for snapshot-aware defrag */
2997 free_sa_defrag_extent(new);
2998 atomic_dec(&root->fs_info->defrag_running);
3000 relink_file_extents(new);
3005 btrfs_put_ordered_extent(ordered_extent);
3006 /* once for the tree */
3007 btrfs_put_ordered_extent(ordered_extent);
3012 static void finish_ordered_fn(struct btrfs_work *work)
3014 struct btrfs_ordered_extent *ordered_extent;
3015 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3016 btrfs_finish_ordered_io(ordered_extent);
3019 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3020 struct extent_state *state, int uptodate)
3022 struct inode *inode = page->mapping->host;
3023 struct btrfs_root *root = BTRFS_I(inode)->root;
3024 struct btrfs_ordered_extent *ordered_extent = NULL;
3025 struct btrfs_workqueue *wq;
3026 btrfs_work_func_t func;
3028 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3030 ClearPagePrivate2(page);
3031 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3032 end - start + 1, uptodate))
3035 if (btrfs_is_free_space_inode(inode)) {
3036 wq = root->fs_info->endio_freespace_worker;
3037 func = btrfs_freespace_write_helper;
3039 wq = root->fs_info->endio_write_workers;
3040 func = btrfs_endio_write_helper;
3043 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3045 btrfs_queue_work(wq, &ordered_extent->work);
3050 static int __readpage_endio_check(struct inode *inode,
3051 struct btrfs_io_bio *io_bio,
3052 int icsum, struct page *page,
3053 int pgoff, u64 start, size_t len)
3059 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3061 kaddr = kmap_atomic(page);
3062 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3063 btrfs_csum_final(csum, (char *)&csum);
3064 if (csum != csum_expected)
3067 kunmap_atomic(kaddr);
3070 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3071 "csum failed ino %llu off %llu csum %u expected csum %u",
3072 btrfs_ino(inode), start, csum, csum_expected);
3073 memset(kaddr + pgoff, 1, len);
3074 flush_dcache_page(page);
3075 kunmap_atomic(kaddr);
3076 if (csum_expected == 0)
3082 * when reads are done, we need to check csums to verify the data is correct
3083 * if there's a match, we allow the bio to finish. If not, the code in
3084 * extent_io.c will try to find good copies for us.
3086 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3087 u64 phy_offset, struct page *page,
3088 u64 start, u64 end, int mirror)
3090 size_t offset = start - page_offset(page);
3091 struct inode *inode = page->mapping->host;
3092 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3093 struct btrfs_root *root = BTRFS_I(inode)->root;
3095 if (PageChecked(page)) {
3096 ClearPageChecked(page);
3100 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3103 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3104 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3105 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3110 phy_offset >>= inode->i_sb->s_blocksize_bits;
3111 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3112 start, (size_t)(end - start + 1));
3115 void btrfs_add_delayed_iput(struct inode *inode)
3117 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3118 struct btrfs_inode *binode = BTRFS_I(inode);
3120 if (atomic_add_unless(&inode->i_count, -1, 1))
3123 spin_lock(&fs_info->delayed_iput_lock);
3124 if (binode->delayed_iput_count == 0) {
3125 ASSERT(list_empty(&binode->delayed_iput));
3126 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3128 binode->delayed_iput_count++;
3130 spin_unlock(&fs_info->delayed_iput_lock);
3133 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3135 struct btrfs_fs_info *fs_info = root->fs_info;
3137 spin_lock(&fs_info->delayed_iput_lock);
3138 while (!list_empty(&fs_info->delayed_iputs)) {
3139 struct btrfs_inode *inode;
3141 inode = list_first_entry(&fs_info->delayed_iputs,
3142 struct btrfs_inode, delayed_iput);
3143 if (inode->delayed_iput_count) {
3144 inode->delayed_iput_count--;
3145 list_move_tail(&inode->delayed_iput,
3146 &fs_info->delayed_iputs);
3148 list_del_init(&inode->delayed_iput);
3150 spin_unlock(&fs_info->delayed_iput_lock);
3151 iput(&inode->vfs_inode);
3152 spin_lock(&fs_info->delayed_iput_lock);
3154 spin_unlock(&fs_info->delayed_iput_lock);
3158 * This is called in transaction commit time. If there are no orphan
3159 * files in the subvolume, it removes orphan item and frees block_rsv
3162 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3163 struct btrfs_root *root)
3165 struct btrfs_block_rsv *block_rsv;
3168 if (atomic_read(&root->orphan_inodes) ||
3169 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3172 spin_lock(&root->orphan_lock);
3173 if (atomic_read(&root->orphan_inodes)) {
3174 spin_unlock(&root->orphan_lock);
3178 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3179 spin_unlock(&root->orphan_lock);
3183 block_rsv = root->orphan_block_rsv;
3184 root->orphan_block_rsv = NULL;
3185 spin_unlock(&root->orphan_lock);
3187 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3188 btrfs_root_refs(&root->root_item) > 0) {
3189 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3190 root->root_key.objectid);
3192 btrfs_abort_transaction(trans, root, ret);
3194 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3199 WARN_ON(block_rsv->size > 0);
3200 btrfs_free_block_rsv(root, block_rsv);
3205 * This creates an orphan entry for the given inode in case something goes
3206 * wrong in the middle of an unlink/truncate.
3208 * NOTE: caller of this function should reserve 5 units of metadata for
3211 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3213 struct btrfs_root *root = BTRFS_I(inode)->root;
3214 struct btrfs_block_rsv *block_rsv = NULL;
3219 if (!root->orphan_block_rsv) {
3220 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3225 spin_lock(&root->orphan_lock);
3226 if (!root->orphan_block_rsv) {
3227 root->orphan_block_rsv = block_rsv;
3228 } else if (block_rsv) {
3229 btrfs_free_block_rsv(root, block_rsv);
3233 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3234 &BTRFS_I(inode)->runtime_flags)) {
3237 * For proper ENOSPC handling, we should do orphan
3238 * cleanup when mounting. But this introduces backward
3239 * compatibility issue.
3241 if (!xchg(&root->orphan_item_inserted, 1))
3247 atomic_inc(&root->orphan_inodes);
3250 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3251 &BTRFS_I(inode)->runtime_flags))
3253 spin_unlock(&root->orphan_lock);
3255 /* grab metadata reservation from transaction handle */
3257 ret = btrfs_orphan_reserve_metadata(trans, inode);
3258 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3261 /* insert an orphan item to track this unlinked/truncated file */
3263 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3265 atomic_dec(&root->orphan_inodes);
3267 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags);
3269 btrfs_orphan_release_metadata(inode);
3271 if (ret != -EEXIST) {
3272 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3273 &BTRFS_I(inode)->runtime_flags);
3274 btrfs_abort_transaction(trans, root, ret);
3281 /* insert an orphan item to track subvolume contains orphan files */
3283 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3284 root->root_key.objectid);
3285 if (ret && ret != -EEXIST) {
3286 btrfs_abort_transaction(trans, root, ret);
3294 * We have done the truncate/delete so we can go ahead and remove the orphan
3295 * item for this particular inode.
3297 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3298 struct inode *inode)
3300 struct btrfs_root *root = BTRFS_I(inode)->root;
3301 int delete_item = 0;
3302 int release_rsv = 0;
3305 spin_lock(&root->orphan_lock);
3306 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3307 &BTRFS_I(inode)->runtime_flags))
3310 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3311 &BTRFS_I(inode)->runtime_flags))
3313 spin_unlock(&root->orphan_lock);
3316 atomic_dec(&root->orphan_inodes);
3318 ret = btrfs_del_orphan_item(trans, root,
3323 btrfs_orphan_release_metadata(inode);
3329 * this cleans up any orphans that may be left on the list from the last use
3332 int btrfs_orphan_cleanup(struct btrfs_root *root)
3334 struct btrfs_path *path;
3335 struct extent_buffer *leaf;
3336 struct btrfs_key key, found_key;
3337 struct btrfs_trans_handle *trans;
3338 struct inode *inode;
3339 u64 last_objectid = 0;
3340 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3342 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3345 path = btrfs_alloc_path();
3350 path->reada = READA_BACK;
3352 key.objectid = BTRFS_ORPHAN_OBJECTID;
3353 key.type = BTRFS_ORPHAN_ITEM_KEY;
3354 key.offset = (u64)-1;
3357 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3362 * if ret == 0 means we found what we were searching for, which
3363 * is weird, but possible, so only screw with path if we didn't
3364 * find the key and see if we have stuff that matches
3368 if (path->slots[0] == 0)
3373 /* pull out the item */
3374 leaf = path->nodes[0];
3375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3377 /* make sure the item matches what we want */
3378 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3380 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3383 /* release the path since we're done with it */
3384 btrfs_release_path(path);
3387 * this is where we are basically btrfs_lookup, without the
3388 * crossing root thing. we store the inode number in the
3389 * offset of the orphan item.
3392 if (found_key.offset == last_objectid) {
3393 btrfs_err(root->fs_info,
3394 "Error removing orphan entry, stopping orphan cleanup");
3399 last_objectid = found_key.offset;
3401 found_key.objectid = found_key.offset;
3402 found_key.type = BTRFS_INODE_ITEM_KEY;
3403 found_key.offset = 0;
3404 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3405 ret = PTR_ERR_OR_ZERO(inode);
3406 if (ret && ret != -ESTALE)
3409 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3410 struct btrfs_root *dead_root;
3411 struct btrfs_fs_info *fs_info = root->fs_info;
3412 int is_dead_root = 0;
3415 * this is an orphan in the tree root. Currently these
3416 * could come from 2 sources:
3417 * a) a snapshot deletion in progress
3418 * b) a free space cache inode
3419 * We need to distinguish those two, as the snapshot
3420 * orphan must not get deleted.
3421 * find_dead_roots already ran before us, so if this
3422 * is a snapshot deletion, we should find the root
3423 * in the dead_roots list
3425 spin_lock(&fs_info->trans_lock);
3426 list_for_each_entry(dead_root, &fs_info->dead_roots,
3428 if (dead_root->root_key.objectid ==
3429 found_key.objectid) {
3434 spin_unlock(&fs_info->trans_lock);
3436 /* prevent this orphan from being found again */
3437 key.offset = found_key.objectid - 1;
3442 * Inode is already gone but the orphan item is still there,
3443 * kill the orphan item.
3445 if (ret == -ESTALE) {
3446 trans = btrfs_start_transaction(root, 1);
3447 if (IS_ERR(trans)) {
3448 ret = PTR_ERR(trans);
3451 btrfs_debug(root->fs_info, "auto deleting %Lu",
3452 found_key.objectid);
3453 ret = btrfs_del_orphan_item(trans, root,
3454 found_key.objectid);
3455 btrfs_end_transaction(trans, root);
3462 * add this inode to the orphan list so btrfs_orphan_del does
3463 * the proper thing when we hit it
3465 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3466 &BTRFS_I(inode)->runtime_flags);
3467 atomic_inc(&root->orphan_inodes);
3469 /* if we have links, this was a truncate, lets do that */
3470 if (inode->i_nlink) {
3471 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3477 /* 1 for the orphan item deletion. */
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
3481 ret = PTR_ERR(trans);
3484 ret = btrfs_orphan_add(trans, inode);
3485 btrfs_end_transaction(trans, root);
3491 ret = btrfs_truncate(inode);
3493 btrfs_orphan_del(NULL, inode);
3498 /* this will do delete_inode and everything for us */
3503 /* release the path since we're done with it */
3504 btrfs_release_path(path);
3506 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3508 if (root->orphan_block_rsv)
3509 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3512 if (root->orphan_block_rsv ||
3513 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3514 trans = btrfs_join_transaction(root);
3516 btrfs_end_transaction(trans, root);
3520 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3522 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3526 btrfs_err(root->fs_info,
3527 "could not do orphan cleanup %d", ret);
3528 btrfs_free_path(path);
3533 * very simple check to peek ahead in the leaf looking for xattrs. If we
3534 * don't find any xattrs, we know there can't be any acls.
3536 * slot is the slot the inode is in, objectid is the objectid of the inode
3538 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3539 int slot, u64 objectid,
3540 int *first_xattr_slot)
3542 u32 nritems = btrfs_header_nritems(leaf);
3543 struct btrfs_key found_key;
3544 static u64 xattr_access = 0;
3545 static u64 xattr_default = 0;
3548 if (!xattr_access) {
3549 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3550 strlen(POSIX_ACL_XATTR_ACCESS));
3551 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3552 strlen(POSIX_ACL_XATTR_DEFAULT));
3556 *first_xattr_slot = -1;
3557 while (slot < nritems) {
3558 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3560 /* we found a different objectid, there must not be acls */
3561 if (found_key.objectid != objectid)
3564 /* we found an xattr, assume we've got an acl */
3565 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3566 if (*first_xattr_slot == -1)
3567 *first_xattr_slot = slot;
3568 if (found_key.offset == xattr_access ||
3569 found_key.offset == xattr_default)
3574 * we found a key greater than an xattr key, there can't
3575 * be any acls later on
3577 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3584 * it goes inode, inode backrefs, xattrs, extents,
3585 * so if there are a ton of hard links to an inode there can
3586 * be a lot of backrefs. Don't waste time searching too hard,
3587 * this is just an optimization
3592 /* we hit the end of the leaf before we found an xattr or
3593 * something larger than an xattr. We have to assume the inode
3596 if (*first_xattr_slot == -1)
3597 *first_xattr_slot = slot;
3602 * read an inode from the btree into the in-memory inode
3604 static void btrfs_read_locked_inode(struct inode *inode)
3606 struct btrfs_path *path;
3607 struct extent_buffer *leaf;
3608 struct btrfs_inode_item *inode_item;
3609 struct btrfs_root *root = BTRFS_I(inode)->root;
3610 struct btrfs_key location;
3615 bool filled = false;
3616 int first_xattr_slot;
3618 ret = btrfs_fill_inode(inode, &rdev);
3622 path = btrfs_alloc_path();
3626 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3628 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3632 leaf = path->nodes[0];
3637 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3638 struct btrfs_inode_item);
3639 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3640 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3641 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3642 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3643 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3645 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3646 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3648 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3649 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3651 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3652 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3654 BTRFS_I(inode)->i_otime.tv_sec =
3655 btrfs_timespec_sec(leaf, &inode_item->otime);
3656 BTRFS_I(inode)->i_otime.tv_nsec =
3657 btrfs_timespec_nsec(leaf, &inode_item->otime);
3659 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3660 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3661 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3663 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3664 inode->i_generation = BTRFS_I(inode)->generation;
3666 rdev = btrfs_inode_rdev(leaf, inode_item);
3668 BTRFS_I(inode)->index_cnt = (u64)-1;
3669 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3673 * If we were modified in the current generation and evicted from memory
3674 * and then re-read we need to do a full sync since we don't have any
3675 * idea about which extents were modified before we were evicted from
3678 * This is required for both inode re-read from disk and delayed inode
3679 * in delayed_nodes_tree.
3681 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3682 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3683 &BTRFS_I(inode)->runtime_flags);
3686 * We don't persist the id of the transaction where an unlink operation
3687 * against the inode was last made. So here we assume the inode might
3688 * have been evicted, and therefore the exact value of last_unlink_trans
3689 * lost, and set it to last_trans to avoid metadata inconsistencies
3690 * between the inode and its parent if the inode is fsync'ed and the log
3691 * replayed. For example, in the scenario:
3694 * ln mydir/foo mydir/bar
3697 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3698 * xfs_io -c fsync mydir/foo
3700 * mount fs, triggers fsync log replay
3702 * We must make sure that when we fsync our inode foo we also log its
3703 * parent inode, otherwise after log replay the parent still has the
3704 * dentry with the "bar" name but our inode foo has a link count of 1
3705 * and doesn't have an inode ref with the name "bar" anymore.
3707 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3708 * but it guarantees correctness at the expense of ocassional full
3709 * transaction commits on fsync if our inode is a directory, or if our
3710 * inode is not a directory, logging its parent unnecessarily.
3712 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3715 if (inode->i_nlink != 1 ||
3716 path->slots[0] >= btrfs_header_nritems(leaf))
3719 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3720 if (location.objectid != btrfs_ino(inode))
3723 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3724 if (location.type == BTRFS_INODE_REF_KEY) {
3725 struct btrfs_inode_ref *ref;
3727 ref = (struct btrfs_inode_ref *)ptr;
3728 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3729 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3730 struct btrfs_inode_extref *extref;
3732 extref = (struct btrfs_inode_extref *)ptr;
3733 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3738 * try to precache a NULL acl entry for files that don't have
3739 * any xattrs or acls
3741 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3742 btrfs_ino(inode), &first_xattr_slot);
3743 if (first_xattr_slot != -1) {
3744 path->slots[0] = first_xattr_slot;
3745 ret = btrfs_load_inode_props(inode, path);
3747 btrfs_err(root->fs_info,
3748 "error loading props for ino %llu (root %llu): %d",
3750 root->root_key.objectid, ret);
3752 btrfs_free_path(path);
3755 cache_no_acl(inode);
3757 switch (inode->i_mode & S_IFMT) {
3759 inode->i_mapping->a_ops = &btrfs_aops;
3760 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3761 inode->i_fop = &btrfs_file_operations;
3762 inode->i_op = &btrfs_file_inode_operations;
3765 inode->i_fop = &btrfs_dir_file_operations;
3766 if (root == root->fs_info->tree_root)
3767 inode->i_op = &btrfs_dir_ro_inode_operations;
3769 inode->i_op = &btrfs_dir_inode_operations;
3772 inode->i_op = &btrfs_symlink_inode_operations;
3773 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3776 inode->i_op = &btrfs_special_inode_operations;
3777 init_special_inode(inode, inode->i_mode, rdev);
3781 btrfs_update_iflags(inode);
3785 btrfs_free_path(path);
3786 make_bad_inode(inode);
3790 * given a leaf and an inode, copy the inode fields into the leaf
3792 static void fill_inode_item(struct btrfs_trans_handle *trans,
3793 struct extent_buffer *leaf,
3794 struct btrfs_inode_item *item,
3795 struct inode *inode)
3797 struct btrfs_map_token token;
3799 btrfs_init_map_token(&token);
3801 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3802 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3803 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3805 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3806 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3808 btrfs_set_token_timespec_sec(leaf, &item->atime,
3809 inode->i_atime.tv_sec, &token);
3810 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3811 inode->i_atime.tv_nsec, &token);
3813 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3814 inode->i_mtime.tv_sec, &token);
3815 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3816 inode->i_mtime.tv_nsec, &token);
3818 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3819 inode->i_ctime.tv_sec, &token);
3820 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3821 inode->i_ctime.tv_nsec, &token);
3823 btrfs_set_token_timespec_sec(leaf, &item->otime,
3824 BTRFS_I(inode)->i_otime.tv_sec, &token);
3825 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3826 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3828 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3830 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3832 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3833 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3834 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3835 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3836 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3840 * copy everything in the in-memory inode into the btree.
3842 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3843 struct btrfs_root *root, struct inode *inode)
3845 struct btrfs_inode_item *inode_item;
3846 struct btrfs_path *path;
3847 struct extent_buffer *leaf;
3850 path = btrfs_alloc_path();
3854 path->leave_spinning = 1;
3855 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3863 leaf = path->nodes[0];
3864 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3865 struct btrfs_inode_item);
3867 fill_inode_item(trans, leaf, inode_item, inode);
3868 btrfs_mark_buffer_dirty(leaf);
3869 btrfs_set_inode_last_trans(trans, inode);
3872 btrfs_free_path(path);
3877 * copy everything in the in-memory inode into the btree.
3879 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, struct inode *inode)
3885 * If the inode is a free space inode, we can deadlock during commit
3886 * if we put it into the delayed code.
3888 * The data relocation inode should also be directly updated
3891 if (!btrfs_is_free_space_inode(inode)
3892 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3893 && !root->fs_info->log_root_recovering) {
3894 btrfs_update_root_times(trans, root);
3896 ret = btrfs_delayed_update_inode(trans, root, inode);
3898 btrfs_set_inode_last_trans(trans, inode);
3902 return btrfs_update_inode_item(trans, root, inode);
3905 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3906 struct btrfs_root *root,
3907 struct inode *inode)
3911 ret = btrfs_update_inode(trans, root, inode);
3913 return btrfs_update_inode_item(trans, root, inode);
3918 * unlink helper that gets used here in inode.c and in the tree logging
3919 * recovery code. It remove a link in a directory with a given name, and
3920 * also drops the back refs in the inode to the directory
3922 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3923 struct btrfs_root *root,
3924 struct inode *dir, struct inode *inode,
3925 const char *name, int name_len)
3927 struct btrfs_path *path;
3929 struct extent_buffer *leaf;
3930 struct btrfs_dir_item *di;
3931 struct btrfs_key key;
3933 u64 ino = btrfs_ino(inode);
3934 u64 dir_ino = btrfs_ino(dir);
3936 path = btrfs_alloc_path();
3942 path->leave_spinning = 1;
3943 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3944 name, name_len, -1);
3953 leaf = path->nodes[0];
3954 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3955 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3958 btrfs_release_path(path);
3961 * If we don't have dir index, we have to get it by looking up
3962 * the inode ref, since we get the inode ref, remove it directly,
3963 * it is unnecessary to do delayed deletion.
3965 * But if we have dir index, needn't search inode ref to get it.
3966 * Since the inode ref is close to the inode item, it is better
3967 * that we delay to delete it, and just do this deletion when
3968 * we update the inode item.
3970 if (BTRFS_I(inode)->dir_index) {
3971 ret = btrfs_delayed_delete_inode_ref(inode);
3973 index = BTRFS_I(inode)->dir_index;
3978 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3981 btrfs_info(root->fs_info,
3982 "failed to delete reference to %.*s, inode %llu parent %llu",
3983 name_len, name, ino, dir_ino);
3984 btrfs_abort_transaction(trans, root, ret);
3988 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3990 btrfs_abort_transaction(trans, root, ret);
3994 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3996 if (ret != 0 && ret != -ENOENT) {
3997 btrfs_abort_transaction(trans, root, ret);
4001 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4006 btrfs_abort_transaction(trans, root, ret);
4008 btrfs_free_path(path);
4012 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4013 inode_inc_iversion(inode);
4014 inode_inc_iversion(dir);
4015 inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4016 ret = btrfs_update_inode(trans, root, dir);
4021 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4022 struct btrfs_root *root,
4023 struct inode *dir, struct inode *inode,
4024 const char *name, int name_len)
4027 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4030 ret = btrfs_update_inode(trans, root, inode);
4036 * helper to start transaction for unlink and rmdir.
4038 * unlink and rmdir are special in btrfs, they do not always free space, so
4039 * if we cannot make our reservations the normal way try and see if there is
4040 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4041 * allow the unlink to occur.
4043 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4045 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 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4057 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4059 struct btrfs_root *root = BTRFS_I(dir)->root;
4060 struct btrfs_trans_handle *trans;
4061 struct inode *inode = d_inode(dentry);
4064 trans = __unlink_start_trans(dir);
4066 return PTR_ERR(trans);
4068 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4070 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4071 dentry->d_name.name, dentry->d_name.len);
4075 if (inode->i_nlink == 0) {
4076 ret = btrfs_orphan_add(trans, inode);
4082 btrfs_end_transaction(trans, root);
4083 btrfs_btree_balance_dirty(root);
4087 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4088 struct btrfs_root *root,
4089 struct inode *dir, u64 objectid,
4090 const char *name, int name_len)
4092 struct btrfs_path *path;
4093 struct extent_buffer *leaf;
4094 struct btrfs_dir_item *di;
4095 struct btrfs_key key;
4098 u64 dir_ino = btrfs_ino(dir);
4100 path = btrfs_alloc_path();
4104 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4105 name, name_len, -1);
4106 if (IS_ERR_OR_NULL(di)) {
4114 leaf = path->nodes[0];
4115 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4116 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4117 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4119 btrfs_abort_transaction(trans, root, ret);
4122 btrfs_release_path(path);
4124 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4125 objectid, root->root_key.objectid,
4126 dir_ino, &index, name, name_len);
4128 if (ret != -ENOENT) {
4129 btrfs_abort_transaction(trans, root, ret);
4132 di = btrfs_search_dir_index_item(root, path, dir_ino,
4134 if (IS_ERR_OR_NULL(di)) {
4139 btrfs_abort_transaction(trans, root, ret);
4143 leaf = path->nodes[0];
4144 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4145 btrfs_release_path(path);
4148 btrfs_release_path(path);
4150 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4152 btrfs_abort_transaction(trans, root, ret);
4156 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4157 inode_inc_iversion(dir);
4158 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4159 ret = btrfs_update_inode_fallback(trans, root, dir);
4161 btrfs_abort_transaction(trans, root, ret);
4163 btrfs_free_path(path);
4167 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4169 struct inode *inode = d_inode(dentry);
4171 struct btrfs_root *root = BTRFS_I(dir)->root;
4172 struct btrfs_trans_handle *trans;
4174 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4176 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4179 trans = __unlink_start_trans(dir);
4181 return PTR_ERR(trans);
4183 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4184 err = btrfs_unlink_subvol(trans, root, dir,
4185 BTRFS_I(inode)->location.objectid,
4186 dentry->d_name.name,
4187 dentry->d_name.len);
4191 err = btrfs_orphan_add(trans, inode);
4195 /* now the directory is empty */
4196 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4197 dentry->d_name.name, dentry->d_name.len);
4199 btrfs_i_size_write(inode, 0);
4201 btrfs_end_transaction(trans, root);
4202 btrfs_btree_balance_dirty(root);
4207 static int truncate_space_check(struct btrfs_trans_handle *trans,
4208 struct btrfs_root *root,
4213 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4214 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4215 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4217 trans->bytes_reserved += bytes_deleted;
4222 static int truncate_inline_extent(struct inode *inode,
4223 struct btrfs_path *path,
4224 struct btrfs_key *found_key,
4228 struct extent_buffer *leaf = path->nodes[0];
4229 int slot = path->slots[0];
4230 struct btrfs_file_extent_item *fi;
4231 u32 size = (u32)(new_size - found_key->offset);
4232 struct btrfs_root *root = BTRFS_I(inode)->root;
4234 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4236 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4237 loff_t offset = new_size;
4238 loff_t page_end = ALIGN(offset, PAGE_CACHE_SIZE);
4241 * Zero out the remaining of the last page of our inline extent,
4242 * instead of directly truncating our inline extent here - that
4243 * would be much more complex (decompressing all the data, then
4244 * compressing the truncated data, which might be bigger than
4245 * the size of the inline extent, resize the extent, etc).
4246 * We release the path because to get the page we might need to
4247 * read the extent item from disk (data not in the page cache).
4249 btrfs_release_path(path);
4250 return btrfs_truncate_block(inode, offset, page_end - offset,
4254 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4255 size = btrfs_file_extent_calc_inline_size(size);
4256 btrfs_truncate_item(root, path, size, 1);
4258 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4259 inode_sub_bytes(inode, item_end + 1 - new_size);
4265 * this can truncate away extent items, csum items and directory items.
4266 * It starts at a high offset and removes keys until it can't find
4267 * any higher than new_size
4269 * csum items that cross the new i_size are truncated to the new size
4272 * min_type is the minimum key type to truncate down to. If set to 0, this
4273 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4275 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4276 struct btrfs_root *root,
4277 struct inode *inode,
4278 u64 new_size, u32 min_type)
4280 struct btrfs_path *path;
4281 struct extent_buffer *leaf;
4282 struct btrfs_file_extent_item *fi;
4283 struct btrfs_key key;
4284 struct btrfs_key found_key;
4285 u64 extent_start = 0;
4286 u64 extent_num_bytes = 0;
4287 u64 extent_offset = 0;
4289 u64 last_size = new_size;
4290 u32 found_type = (u8)-1;
4293 int pending_del_nr = 0;
4294 int pending_del_slot = 0;
4295 int extent_type = -1;
4298 u64 ino = btrfs_ino(inode);
4299 u64 bytes_deleted = 0;
4301 bool should_throttle = 0;
4302 bool should_end = 0;
4304 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4307 * for non-free space inodes and ref cows, we want to back off from
4310 if (!btrfs_is_free_space_inode(inode) &&
4311 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4314 path = btrfs_alloc_path();
4317 path->reada = READA_BACK;
4320 * We want to drop from the next block forward in case this new size is
4321 * not block aligned since we will be keeping the last block of the
4322 * extent just the way it is.
4324 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4325 root == root->fs_info->tree_root)
4326 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4327 root->sectorsize), (u64)-1, 0);
4330 * This function is also used to drop the items in the log tree before
4331 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4332 * it is used to drop the loged items. So we shouldn't kill the delayed
4335 if (min_type == 0 && root == BTRFS_I(inode)->root)
4336 btrfs_kill_delayed_inode_items(inode);
4339 key.offset = (u64)-1;
4344 * with a 16K leaf size and 128MB extents, you can actually queue
4345 * up a huge file in a single leaf. Most of the time that
4346 * bytes_deleted is > 0, it will be huge by the time we get here
4348 if (be_nice && bytes_deleted > SZ_32M) {
4349 if (btrfs_should_end_transaction(trans, root)) {
4356 path->leave_spinning = 1;
4357 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4364 /* there are no items in the tree for us to truncate, we're
4367 if (path->slots[0] == 0)
4374 leaf = path->nodes[0];
4375 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4376 found_type = found_key.type;
4378 if (found_key.objectid != ino)
4381 if (found_type < min_type)
4384 item_end = found_key.offset;
4385 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4386 fi = btrfs_item_ptr(leaf, path->slots[0],
4387 struct btrfs_file_extent_item);
4388 extent_type = btrfs_file_extent_type(leaf, fi);
4389 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4391 btrfs_file_extent_num_bytes(leaf, fi);
4392 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4393 item_end += btrfs_file_extent_inline_len(leaf,
4394 path->slots[0], fi);
4398 if (found_type > min_type) {
4401 if (item_end < new_size)
4403 if (found_key.offset >= new_size)
4409 /* FIXME, shrink the extent if the ref count is only 1 */
4410 if (found_type != BTRFS_EXTENT_DATA_KEY)
4414 last_size = found_key.offset;
4416 last_size = new_size;
4418 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4420 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4422 u64 orig_num_bytes =
4423 btrfs_file_extent_num_bytes(leaf, fi);
4424 extent_num_bytes = ALIGN(new_size -
4427 btrfs_set_file_extent_num_bytes(leaf, fi,
4429 num_dec = (orig_num_bytes -
4431 if (test_bit(BTRFS_ROOT_REF_COWS,
4434 inode_sub_bytes(inode, num_dec);
4435 btrfs_mark_buffer_dirty(leaf);
4438 btrfs_file_extent_disk_num_bytes(leaf,
4440 extent_offset = found_key.offset -
4441 btrfs_file_extent_offset(leaf, fi);
4443 /* FIXME blocksize != 4096 */
4444 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4445 if (extent_start != 0) {
4447 if (test_bit(BTRFS_ROOT_REF_COWS,
4449 inode_sub_bytes(inode, num_dec);
4452 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4454 * we can't truncate inline items that have had
4458 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4459 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4462 * Need to release path in order to truncate a
4463 * compressed extent. So delete any accumulated
4464 * extent items so far.
4466 if (btrfs_file_extent_compression(leaf, fi) !=
4467 BTRFS_COMPRESS_NONE && pending_del_nr) {
4468 err = btrfs_del_items(trans, root, path,
4472 btrfs_abort_transaction(trans,
4480 err = truncate_inline_extent(inode, path,
4485 btrfs_abort_transaction(trans,
4489 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4491 inode_sub_bytes(inode, item_end + 1 - new_size);
4496 if (!pending_del_nr) {
4497 /* no pending yet, add ourselves */
4498 pending_del_slot = path->slots[0];
4500 } else if (pending_del_nr &&
4501 path->slots[0] + 1 == pending_del_slot) {
4502 /* hop on the pending chunk */
4504 pending_del_slot = path->slots[0];
4511 should_throttle = 0;
4514 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4515 root == root->fs_info->tree_root)) {
4516 btrfs_set_path_blocking(path);
4517 bytes_deleted += extent_num_bytes;
4518 ret = btrfs_free_extent(trans, root, extent_start,
4519 extent_num_bytes, 0,
4520 btrfs_header_owner(leaf),
4521 ino, extent_offset);
4523 if (btrfs_should_throttle_delayed_refs(trans, root))
4524 btrfs_async_run_delayed_refs(root,
4525 trans->delayed_ref_updates * 2, 0);
4527 if (truncate_space_check(trans, root,
4528 extent_num_bytes)) {
4531 if (btrfs_should_throttle_delayed_refs(trans,
4533 should_throttle = 1;
4538 if (found_type == BTRFS_INODE_ITEM_KEY)
4541 if (path->slots[0] == 0 ||
4542 path->slots[0] != pending_del_slot ||
4543 should_throttle || should_end) {
4544 if (pending_del_nr) {
4545 ret = btrfs_del_items(trans, root, path,
4549 btrfs_abort_transaction(trans,
4555 btrfs_release_path(path);
4556 if (should_throttle) {
4557 unsigned long updates = trans->delayed_ref_updates;
4559 trans->delayed_ref_updates = 0;
4560 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4566 * if we failed to refill our space rsv, bail out
4567 * and let the transaction restart
4579 if (pending_del_nr) {
4580 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4583 btrfs_abort_transaction(trans, root, ret);
4586 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4587 btrfs_ordered_update_i_size(inode, last_size, NULL);
4589 btrfs_free_path(path);
4591 if (be_nice && bytes_deleted > SZ_32M) {
4592 unsigned long updates = trans->delayed_ref_updates;
4594 trans->delayed_ref_updates = 0;
4595 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4604 * btrfs_truncate_block - read, zero a chunk and write a block
4605 * @inode - inode that we're zeroing
4606 * @from - the offset to start zeroing
4607 * @len - the length to zero, 0 to zero the entire range respective to the
4609 * @front - zero up to the offset instead of from the offset on
4611 * This will find the block for the "from" offset and cow the block and zero the
4612 * part we want to zero. This is used with truncate and hole punching.
4614 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4617 struct address_space *mapping = inode->i_mapping;
4618 struct btrfs_root *root = BTRFS_I(inode)->root;
4619 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4620 struct btrfs_ordered_extent *ordered;
4621 struct extent_state *cached_state = NULL;
4623 u32 blocksize = root->sectorsize;
4624 pgoff_t index = from >> PAGE_CACHE_SHIFT;
4625 unsigned offset = from & (blocksize - 1);
4627 gfp_t mask = btrfs_alloc_write_mask(mapping);
4632 if ((offset & (blocksize - 1)) == 0 &&
4633 (!len || ((len & (blocksize - 1)) == 0)))
4636 ret = btrfs_delalloc_reserve_space(inode,
4637 round_down(from, blocksize), blocksize);
4642 page = find_or_create_page(mapping, index, mask);
4644 btrfs_delalloc_release_space(inode,
4645 round_down(from, blocksize),
4651 block_start = round_down(from, blocksize);
4652 block_end = block_start + blocksize - 1;
4654 if (!PageUptodate(page)) {
4655 ret = btrfs_readpage(NULL, page);
4657 if (page->mapping != mapping) {
4659 page_cache_release(page);
4662 if (!PageUptodate(page)) {
4667 wait_on_page_writeback(page);
4669 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4670 set_page_extent_mapped(page);
4672 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4674 unlock_extent_cached(io_tree, block_start, block_end,
4675 &cached_state, GFP_NOFS);
4677 page_cache_release(page);
4678 btrfs_start_ordered_extent(inode, ordered, 1);
4679 btrfs_put_ordered_extent(ordered);
4683 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4684 EXTENT_DIRTY | EXTENT_DELALLOC |
4685 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4686 0, 0, &cached_state, GFP_NOFS);
4688 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4691 unlock_extent_cached(io_tree, block_start, block_end,
4692 &cached_state, GFP_NOFS);
4696 if (offset != blocksize) {
4698 len = blocksize - offset;
4701 memset(kaddr + (block_start - page_offset(page)),
4704 memset(kaddr + (block_start - page_offset(page)) + offset,
4706 flush_dcache_page(page);
4709 ClearPageChecked(page);
4710 set_page_dirty(page);
4711 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4716 btrfs_delalloc_release_space(inode, block_start,
4719 page_cache_release(page);
4724 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4725 u64 offset, u64 len)
4727 struct btrfs_trans_handle *trans;
4731 * Still need to make sure the inode looks like it's been updated so
4732 * that any holes get logged if we fsync.
4734 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4735 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4736 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4737 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4742 * 1 - for the one we're dropping
4743 * 1 - for the one we're adding
4744 * 1 - for updating the inode.
4746 trans = btrfs_start_transaction(root, 3);
4748 return PTR_ERR(trans);
4750 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4752 btrfs_abort_transaction(trans, root, ret);
4753 btrfs_end_transaction(trans, root);
4757 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4758 0, 0, len, 0, len, 0, 0, 0);
4760 btrfs_abort_transaction(trans, root, ret);
4762 btrfs_update_inode(trans, root, inode);
4763 btrfs_end_transaction(trans, root);
4768 * This function puts in dummy file extents for the area we're creating a hole
4769 * for. So if we are truncating this file to a larger size we need to insert
4770 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4771 * the range between oldsize and size
4773 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4775 struct btrfs_root *root = BTRFS_I(inode)->root;
4776 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4777 struct extent_map *em = NULL;
4778 struct extent_state *cached_state = NULL;
4779 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4780 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4781 u64 block_end = ALIGN(size, root->sectorsize);
4788 * If our size started in the middle of a block we need to zero out the
4789 * rest of the block before we expand the i_size, otherwise we could
4790 * expose stale data.
4792 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4796 if (size <= hole_start)
4800 struct btrfs_ordered_extent *ordered;
4802 lock_extent_bits(io_tree, hole_start, block_end - 1,
4804 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4805 block_end - hole_start);
4808 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4809 &cached_state, GFP_NOFS);
4810 btrfs_start_ordered_extent(inode, ordered, 1);
4811 btrfs_put_ordered_extent(ordered);
4814 cur_offset = hole_start;
4816 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4817 block_end - cur_offset, 0);
4823 last_byte = min(extent_map_end(em), block_end);
4824 last_byte = ALIGN(last_byte , root->sectorsize);
4825 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4826 struct extent_map *hole_em;
4827 hole_size = last_byte - cur_offset;
4829 err = maybe_insert_hole(root, inode, cur_offset,
4833 btrfs_drop_extent_cache(inode, cur_offset,
4834 cur_offset + hole_size - 1, 0);
4835 hole_em = alloc_extent_map();
4837 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4838 &BTRFS_I(inode)->runtime_flags);
4841 hole_em->start = cur_offset;
4842 hole_em->len = hole_size;
4843 hole_em->orig_start = cur_offset;
4845 hole_em->block_start = EXTENT_MAP_HOLE;
4846 hole_em->block_len = 0;
4847 hole_em->orig_block_len = 0;
4848 hole_em->ram_bytes = hole_size;
4849 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4850 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4851 hole_em->generation = root->fs_info->generation;
4854 write_lock(&em_tree->lock);
4855 err = add_extent_mapping(em_tree, hole_em, 1);
4856 write_unlock(&em_tree->lock);
4859 btrfs_drop_extent_cache(inode, cur_offset,
4863 free_extent_map(hole_em);
4866 free_extent_map(em);
4868 cur_offset = last_byte;
4869 if (cur_offset >= block_end)
4872 free_extent_map(em);
4873 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4878 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4880 struct btrfs_root *root = BTRFS_I(inode)->root;
4881 struct btrfs_trans_handle *trans;
4882 loff_t oldsize = i_size_read(inode);
4883 loff_t newsize = attr->ia_size;
4884 int mask = attr->ia_valid;
4888 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4889 * special case where we need to update the times despite not having
4890 * these flags set. For all other operations the VFS set these flags
4891 * explicitly if it wants a timestamp update.
4893 if (newsize != oldsize) {
4894 inode_inc_iversion(inode);
4895 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4896 inode->i_ctime = inode->i_mtime =
4897 current_fs_time(inode->i_sb);
4900 if (newsize > oldsize) {
4901 truncate_pagecache(inode, newsize);
4903 * Don't do an expanding truncate while snapshoting is ongoing.
4904 * This is to ensure the snapshot captures a fully consistent
4905 * state of this file - if the snapshot captures this expanding
4906 * truncation, it must capture all writes that happened before
4909 btrfs_wait_for_snapshot_creation(root);
4910 ret = btrfs_cont_expand(inode, oldsize, newsize);
4912 btrfs_end_write_no_snapshoting(root);
4916 trans = btrfs_start_transaction(root, 1);
4917 if (IS_ERR(trans)) {
4918 btrfs_end_write_no_snapshoting(root);
4919 return PTR_ERR(trans);
4922 i_size_write(inode, newsize);
4923 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4924 ret = btrfs_update_inode(trans, root, inode);
4925 btrfs_end_write_no_snapshoting(root);
4926 btrfs_end_transaction(trans, root);
4930 * We're truncating a file that used to have good data down to
4931 * zero. Make sure it gets into the ordered flush list so that
4932 * any new writes get down to disk quickly.
4935 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4936 &BTRFS_I(inode)->runtime_flags);
4939 * 1 for the orphan item we're going to add
4940 * 1 for the orphan item deletion.
4942 trans = btrfs_start_transaction(root, 2);
4944 return PTR_ERR(trans);
4947 * We need to do this in case we fail at _any_ point during the
4948 * actual truncate. Once we do the truncate_setsize we could
4949 * invalidate pages which forces any outstanding ordered io to
4950 * be instantly completed which will give us extents that need
4951 * to be truncated. If we fail to get an orphan inode down we
4952 * could have left over extents that were never meant to live,
4953 * so we need to garuntee from this point on that everything
4954 * will be consistent.
4956 ret = btrfs_orphan_add(trans, inode);
4957 btrfs_end_transaction(trans, root);
4961 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4962 truncate_setsize(inode, newsize);
4964 /* Disable nonlocked read DIO to avoid the end less truncate */
4965 btrfs_inode_block_unlocked_dio(inode);
4966 inode_dio_wait(inode);
4967 btrfs_inode_resume_unlocked_dio(inode);
4969 ret = btrfs_truncate(inode);
4970 if (ret && inode->i_nlink) {
4974 * failed to truncate, disk_i_size is only adjusted down
4975 * as we remove extents, so it should represent the true
4976 * size of the inode, so reset the in memory size and
4977 * delete our orphan entry.
4979 trans = btrfs_join_transaction(root);
4980 if (IS_ERR(trans)) {
4981 btrfs_orphan_del(NULL, inode);
4984 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4985 err = btrfs_orphan_del(trans, inode);
4987 btrfs_abort_transaction(trans, root, err);
4988 btrfs_end_transaction(trans, root);
4995 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4997 struct inode *inode = d_inode(dentry);
4998 struct btrfs_root *root = BTRFS_I(inode)->root;
5001 if (btrfs_root_readonly(root))
5004 err = inode_change_ok(inode, attr);
5008 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5009 err = btrfs_setsize(inode, attr);
5014 if (attr->ia_valid) {
5015 setattr_copy(inode, attr);
5016 inode_inc_iversion(inode);
5017 err = btrfs_dirty_inode(inode);
5019 if (!err && attr->ia_valid & ATTR_MODE)
5020 err = posix_acl_chmod(inode, inode->i_mode);
5027 * While truncating the inode pages during eviction, we get the VFS calling
5028 * btrfs_invalidatepage() against each page of the inode. This is slow because
5029 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5030 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5031 * extent_state structures over and over, wasting lots of time.
5033 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5034 * those expensive operations on a per page basis and do only the ordered io
5035 * finishing, while we release here the extent_map and extent_state structures,
5036 * without the excessive merging and splitting.
5038 static void evict_inode_truncate_pages(struct inode *inode)
5040 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5041 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5042 struct rb_node *node;
5044 ASSERT(inode->i_state & I_FREEING);
5045 truncate_inode_pages_final(&inode->i_data);
5047 write_lock(&map_tree->lock);
5048 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5049 struct extent_map *em;
5051 node = rb_first(&map_tree->map);
5052 em = rb_entry(node, struct extent_map, rb_node);
5053 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5054 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5055 remove_extent_mapping(map_tree, em);
5056 free_extent_map(em);
5057 if (need_resched()) {
5058 write_unlock(&map_tree->lock);
5060 write_lock(&map_tree->lock);
5063 write_unlock(&map_tree->lock);
5066 * Keep looping until we have no more ranges in the io tree.
5067 * We can have ongoing bios started by readpages (called from readahead)
5068 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5069 * still in progress (unlocked the pages in the bio but did not yet
5070 * unlocked the ranges in the io tree). Therefore this means some
5071 * ranges can still be locked and eviction started because before
5072 * submitting those bios, which are executed by a separate task (work
5073 * queue kthread), inode references (inode->i_count) were not taken
5074 * (which would be dropped in the end io callback of each bio).
5075 * Therefore here we effectively end up waiting for those bios and
5076 * anyone else holding locked ranges without having bumped the inode's
5077 * reference count - if we don't do it, when they access the inode's
5078 * io_tree to unlock a range it may be too late, leading to an
5079 * use-after-free issue.
5081 spin_lock(&io_tree->lock);
5082 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5083 struct extent_state *state;
5084 struct extent_state *cached_state = NULL;
5088 node = rb_first(&io_tree->state);
5089 state = rb_entry(node, struct extent_state, rb_node);
5090 start = state->start;
5092 spin_unlock(&io_tree->lock);
5094 lock_extent_bits(io_tree, start, end, &cached_state);
5097 * If still has DELALLOC flag, the extent didn't reach disk,
5098 * and its reserved space won't be freed by delayed_ref.
5099 * So we need to free its reserved space here.
5100 * (Refer to comment in btrfs_invalidatepage, case 2)
5102 * Note, end is the bytenr of last byte, so we need + 1 here.
5104 if (state->state & EXTENT_DELALLOC)
5105 btrfs_qgroup_free_data(inode, start, end - start + 1);
5107 clear_extent_bit(io_tree, start, end,
5108 EXTENT_LOCKED | EXTENT_DIRTY |
5109 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5110 EXTENT_DEFRAG, 1, 1,
5111 &cached_state, GFP_NOFS);
5114 spin_lock(&io_tree->lock);
5116 spin_unlock(&io_tree->lock);
5119 void btrfs_evict_inode(struct inode *inode)
5121 struct btrfs_trans_handle *trans;
5122 struct btrfs_root *root = BTRFS_I(inode)->root;
5123 struct btrfs_block_rsv *rsv, *global_rsv;
5124 int steal_from_global = 0;
5125 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5128 trace_btrfs_inode_evict(inode);
5130 evict_inode_truncate_pages(inode);
5132 if (inode->i_nlink &&
5133 ((btrfs_root_refs(&root->root_item) != 0 &&
5134 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5135 btrfs_is_free_space_inode(inode)))
5138 if (is_bad_inode(inode)) {
5139 btrfs_orphan_del(NULL, inode);
5142 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5143 if (!special_file(inode->i_mode))
5144 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5146 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5148 if (root->fs_info->log_root_recovering) {
5149 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5150 &BTRFS_I(inode)->runtime_flags));
5154 if (inode->i_nlink > 0) {
5155 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5156 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5160 ret = btrfs_commit_inode_delayed_inode(inode);
5162 btrfs_orphan_del(NULL, inode);
5166 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5168 btrfs_orphan_del(NULL, inode);
5171 rsv->size = min_size;
5173 global_rsv = &root->fs_info->global_block_rsv;
5175 btrfs_i_size_write(inode, 0);
5178 * This is a bit simpler than btrfs_truncate since we've already
5179 * reserved our space for our orphan item in the unlink, so we just
5180 * need to reserve some slack space in case we add bytes and update
5181 * inode item when doing the truncate.
5184 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5185 BTRFS_RESERVE_FLUSH_LIMIT);
5188 * Try and steal from the global reserve since we will
5189 * likely not use this space anyway, we want to try as
5190 * hard as possible to get this to work.
5193 steal_from_global++;
5195 steal_from_global = 0;
5199 * steal_from_global == 0: we reserved stuff, hooray!
5200 * steal_from_global == 1: we didn't reserve stuff, boo!
5201 * steal_from_global == 2: we've committed, still not a lot of
5202 * room but maybe we'll have room in the global reserve this
5204 * steal_from_global == 3: abandon all hope!
5206 if (steal_from_global > 2) {
5207 btrfs_warn(root->fs_info,
5208 "Could not get space for a delete, will truncate on mount %d",
5210 btrfs_orphan_del(NULL, inode);
5211 btrfs_free_block_rsv(root, rsv);
5215 trans = btrfs_join_transaction(root);
5216 if (IS_ERR(trans)) {
5217 btrfs_orphan_del(NULL, inode);
5218 btrfs_free_block_rsv(root, rsv);
5223 * We can't just steal from the global reserve, we need tomake
5224 * sure there is room to do it, if not we need to commit and try
5227 if (steal_from_global) {
5228 if (!btrfs_check_space_for_delayed_refs(trans, root))
5229 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5236 * Couldn't steal from the global reserve, we have too much
5237 * pending stuff built up, commit the transaction and try it
5241 ret = btrfs_commit_transaction(trans, root);
5243 btrfs_orphan_del(NULL, inode);
5244 btrfs_free_block_rsv(root, rsv);
5249 steal_from_global = 0;
5252 trans->block_rsv = rsv;
5254 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5255 if (ret != -ENOSPC && ret != -EAGAIN)
5258 trans->block_rsv = &root->fs_info->trans_block_rsv;
5259 btrfs_end_transaction(trans, root);
5261 btrfs_btree_balance_dirty(root);
5264 btrfs_free_block_rsv(root, rsv);
5267 * Errors here aren't a big deal, it just means we leave orphan items
5268 * in the tree. They will be cleaned up on the next mount.
5271 trans->block_rsv = root->orphan_block_rsv;
5272 btrfs_orphan_del(trans, inode);
5274 btrfs_orphan_del(NULL, inode);
5277 trans->block_rsv = &root->fs_info->trans_block_rsv;
5278 if (!(root == root->fs_info->tree_root ||
5279 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5280 btrfs_return_ino(root, btrfs_ino(inode));
5282 btrfs_end_transaction(trans, root);
5283 btrfs_btree_balance_dirty(root);
5285 btrfs_remove_delayed_node(inode);
5290 * this returns the key found in the dir entry in the location pointer.
5291 * If no dir entries were found, location->objectid is 0.
5293 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5294 struct btrfs_key *location)
5296 const char *name = dentry->d_name.name;
5297 int namelen = dentry->d_name.len;
5298 struct btrfs_dir_item *di;
5299 struct btrfs_path *path;
5300 struct btrfs_root *root = BTRFS_I(dir)->root;
5303 path = btrfs_alloc_path();
5307 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5312 if (IS_ERR_OR_NULL(di))
5315 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5317 btrfs_free_path(path);
5320 location->objectid = 0;
5325 * when we hit a tree root in a directory, the btrfs part of the inode
5326 * needs to be changed to reflect the root directory of the tree root. This
5327 * is kind of like crossing a mount point.
5329 static int fixup_tree_root_location(struct btrfs_root *root,
5331 struct dentry *dentry,
5332 struct btrfs_key *location,
5333 struct btrfs_root **sub_root)
5335 struct btrfs_path *path;
5336 struct btrfs_root *new_root;
5337 struct btrfs_root_ref *ref;
5338 struct extent_buffer *leaf;
5339 struct btrfs_key key;
5343 path = btrfs_alloc_path();
5350 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5351 key.type = BTRFS_ROOT_REF_KEY;
5352 key.offset = location->objectid;
5354 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5362 leaf = path->nodes[0];
5363 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5364 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5365 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5368 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5369 (unsigned long)(ref + 1),
5370 dentry->d_name.len);
5374 btrfs_release_path(path);
5376 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5377 if (IS_ERR(new_root)) {
5378 err = PTR_ERR(new_root);
5382 *sub_root = new_root;
5383 location->objectid = btrfs_root_dirid(&new_root->root_item);
5384 location->type = BTRFS_INODE_ITEM_KEY;
5385 location->offset = 0;
5388 btrfs_free_path(path);
5392 static void inode_tree_add(struct inode *inode)
5394 struct btrfs_root *root = BTRFS_I(inode)->root;
5395 struct btrfs_inode *entry;
5397 struct rb_node *parent;
5398 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5399 u64 ino = btrfs_ino(inode);
5401 if (inode_unhashed(inode))
5404 spin_lock(&root->inode_lock);
5405 p = &root->inode_tree.rb_node;
5408 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5410 if (ino < btrfs_ino(&entry->vfs_inode))
5411 p = &parent->rb_left;
5412 else if (ino > btrfs_ino(&entry->vfs_inode))
5413 p = &parent->rb_right;
5415 WARN_ON(!(entry->vfs_inode.i_state &
5416 (I_WILL_FREE | I_FREEING)));
5417 rb_replace_node(parent, new, &root->inode_tree);
5418 RB_CLEAR_NODE(parent);
5419 spin_unlock(&root->inode_lock);
5423 rb_link_node(new, parent, p);
5424 rb_insert_color(new, &root->inode_tree);
5425 spin_unlock(&root->inode_lock);
5428 static void inode_tree_del(struct inode *inode)
5430 struct btrfs_root *root = BTRFS_I(inode)->root;
5433 spin_lock(&root->inode_lock);
5434 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5435 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5436 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5437 empty = RB_EMPTY_ROOT(&root->inode_tree);
5439 spin_unlock(&root->inode_lock);
5441 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5442 synchronize_srcu(&root->fs_info->subvol_srcu);
5443 spin_lock(&root->inode_lock);
5444 empty = RB_EMPTY_ROOT(&root->inode_tree);
5445 spin_unlock(&root->inode_lock);
5447 btrfs_add_dead_root(root);
5451 void btrfs_invalidate_inodes(struct btrfs_root *root)
5453 struct rb_node *node;
5454 struct rb_node *prev;
5455 struct btrfs_inode *entry;
5456 struct inode *inode;
5459 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5460 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5462 spin_lock(&root->inode_lock);
5464 node = root->inode_tree.rb_node;
5468 entry = rb_entry(node, struct btrfs_inode, rb_node);
5470 if (objectid < btrfs_ino(&entry->vfs_inode))
5471 node = node->rb_left;
5472 else if (objectid > btrfs_ino(&entry->vfs_inode))
5473 node = node->rb_right;
5479 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5480 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5484 prev = rb_next(prev);
5488 entry = rb_entry(node, struct btrfs_inode, rb_node);
5489 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5490 inode = igrab(&entry->vfs_inode);
5492 spin_unlock(&root->inode_lock);
5493 if (atomic_read(&inode->i_count) > 1)
5494 d_prune_aliases(inode);
5496 * btrfs_drop_inode will have it removed from
5497 * the inode cache when its usage count
5502 spin_lock(&root->inode_lock);
5506 if (cond_resched_lock(&root->inode_lock))
5509 node = rb_next(node);
5511 spin_unlock(&root->inode_lock);
5514 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5516 struct btrfs_iget_args *args = p;
5517 inode->i_ino = args->location->objectid;
5518 memcpy(&BTRFS_I(inode)->location, args->location,
5519 sizeof(*args->location));
5520 BTRFS_I(inode)->root = args->root;
5524 static int btrfs_find_actor(struct inode *inode, void *opaque)
5526 struct btrfs_iget_args *args = opaque;
5527 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5528 args->root == BTRFS_I(inode)->root;
5531 static struct inode *btrfs_iget_locked(struct super_block *s,
5532 struct btrfs_key *location,
5533 struct btrfs_root *root)
5535 struct inode *inode;
5536 struct btrfs_iget_args args;
5537 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5539 args.location = location;
5542 inode = iget5_locked(s, hashval, btrfs_find_actor,
5543 btrfs_init_locked_inode,
5548 /* Get an inode object given its location and corresponding root.
5549 * Returns in *is_new if the inode was read from disk
5551 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5552 struct btrfs_root *root, int *new)
5554 struct inode *inode;
5556 inode = btrfs_iget_locked(s, location, root);
5558 return ERR_PTR(-ENOMEM);
5560 if (inode->i_state & I_NEW) {
5561 btrfs_read_locked_inode(inode);
5562 if (!is_bad_inode(inode)) {
5563 inode_tree_add(inode);
5564 unlock_new_inode(inode);
5568 unlock_new_inode(inode);
5570 inode = ERR_PTR(-ESTALE);
5577 static struct inode *new_simple_dir(struct super_block *s,
5578 struct btrfs_key *key,
5579 struct btrfs_root *root)
5581 struct inode *inode = new_inode(s);
5584 return ERR_PTR(-ENOMEM);
5586 BTRFS_I(inode)->root = root;
5587 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5588 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5590 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5591 inode->i_op = &btrfs_dir_ro_inode_operations;
5592 inode->i_fop = &simple_dir_operations;
5593 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5594 inode->i_mtime = CURRENT_TIME;
5595 inode->i_atime = inode->i_mtime;
5596 inode->i_ctime = inode->i_mtime;
5597 BTRFS_I(inode)->i_otime = inode->i_mtime;
5602 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5604 struct inode *inode;
5605 struct btrfs_root *root = BTRFS_I(dir)->root;
5606 struct btrfs_root *sub_root = root;
5607 struct btrfs_key location;
5611 if (dentry->d_name.len > BTRFS_NAME_LEN)
5612 return ERR_PTR(-ENAMETOOLONG);
5614 ret = btrfs_inode_by_name(dir, dentry, &location);
5616 return ERR_PTR(ret);
5618 if (location.objectid == 0)
5619 return ERR_PTR(-ENOENT);
5621 if (location.type == BTRFS_INODE_ITEM_KEY) {
5622 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5626 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5628 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5629 ret = fixup_tree_root_location(root, dir, dentry,
5630 &location, &sub_root);
5633 inode = ERR_PTR(ret);
5635 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5637 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5639 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5641 if (!IS_ERR(inode) && root != sub_root) {
5642 down_read(&root->fs_info->cleanup_work_sem);
5643 if (!(inode->i_sb->s_flags & MS_RDONLY))
5644 ret = btrfs_orphan_cleanup(sub_root);
5645 up_read(&root->fs_info->cleanup_work_sem);
5648 inode = ERR_PTR(ret);
5655 static int btrfs_dentry_delete(const struct dentry *dentry)
5657 struct btrfs_root *root;
5658 struct inode *inode = d_inode(dentry);
5660 if (!inode && !IS_ROOT(dentry))
5661 inode = d_inode(dentry->d_parent);
5664 root = BTRFS_I(inode)->root;
5665 if (btrfs_root_refs(&root->root_item) == 0)
5668 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5674 static void btrfs_dentry_release(struct dentry *dentry)
5676 kfree(dentry->d_fsdata);
5679 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5682 struct inode *inode;
5684 inode = btrfs_lookup_dentry(dir, dentry);
5685 if (IS_ERR(inode)) {
5686 if (PTR_ERR(inode) == -ENOENT)
5689 return ERR_CAST(inode);
5692 return d_splice_alias(inode, dentry);
5695 unsigned char btrfs_filetype_table[] = {
5696 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5699 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5701 struct inode *inode = file_inode(file);
5702 struct btrfs_root *root = BTRFS_I(inode)->root;
5703 struct btrfs_item *item;
5704 struct btrfs_dir_item *di;
5705 struct btrfs_key key;
5706 struct btrfs_key found_key;
5707 struct btrfs_path *path;
5708 struct list_head ins_list;
5709 struct list_head del_list;
5711 struct extent_buffer *leaf;
5713 unsigned char d_type;
5718 int key_type = BTRFS_DIR_INDEX_KEY;
5722 int is_curr = 0; /* ctx->pos points to the current index? */
5724 /* FIXME, use a real flag for deciding about the key type */
5725 if (root->fs_info->tree_root == root)
5726 key_type = BTRFS_DIR_ITEM_KEY;
5728 if (!dir_emit_dots(file, ctx))
5731 path = btrfs_alloc_path();
5735 path->reada = READA_FORWARD;
5737 if (key_type == BTRFS_DIR_INDEX_KEY) {
5738 INIT_LIST_HEAD(&ins_list);
5739 INIT_LIST_HEAD(&del_list);
5740 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5743 key.type = key_type;
5744 key.offset = ctx->pos;
5745 key.objectid = btrfs_ino(inode);
5747 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5752 leaf = path->nodes[0];
5753 slot = path->slots[0];
5754 if (slot >= btrfs_header_nritems(leaf)) {
5755 ret = btrfs_next_leaf(root, path);
5763 item = btrfs_item_nr(slot);
5764 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5766 if (found_key.objectid != key.objectid)
5768 if (found_key.type != key_type)
5770 if (found_key.offset < ctx->pos)
5772 if (key_type == BTRFS_DIR_INDEX_KEY &&
5773 btrfs_should_delete_dir_index(&del_list,
5777 ctx->pos = found_key.offset;
5780 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5782 di_total = btrfs_item_size(leaf, item);
5784 while (di_cur < di_total) {
5785 struct btrfs_key location;
5787 if (verify_dir_item(root, leaf, di))
5790 name_len = btrfs_dir_name_len(leaf, di);
5791 if (name_len <= sizeof(tmp_name)) {
5792 name_ptr = tmp_name;
5794 name_ptr = kmalloc(name_len, GFP_NOFS);
5800 read_extent_buffer(leaf, name_ptr,
5801 (unsigned long)(di + 1), name_len);
5803 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5804 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5807 /* is this a reference to our own snapshot? If so
5810 * In contrast to old kernels, we insert the snapshot's
5811 * dir item and dir index after it has been created, so
5812 * we won't find a reference to our own snapshot. We
5813 * still keep the following code for backward
5816 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5817 location.objectid == root->root_key.objectid) {
5821 over = !dir_emit(ctx, name_ptr, name_len,
5822 location.objectid, d_type);
5825 if (name_ptr != tmp_name)
5830 di_len = btrfs_dir_name_len(leaf, di) +
5831 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5833 di = (struct btrfs_dir_item *)((char *)di + di_len);
5839 if (key_type == BTRFS_DIR_INDEX_KEY) {
5842 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5847 /* Reached end of directory/root. Bump pos past the last item. */
5851 * Stop new entries from being returned after we return the last
5854 * New directory entries are assigned a strictly increasing
5855 * offset. This means that new entries created during readdir
5856 * are *guaranteed* to be seen in the future by that readdir.
5857 * This has broken buggy programs which operate on names as
5858 * they're returned by readdir. Until we re-use freed offsets
5859 * we have this hack to stop new entries from being returned
5860 * under the assumption that they'll never reach this huge
5863 * This is being careful not to overflow 32bit loff_t unless the
5864 * last entry requires it because doing so has broken 32bit apps
5867 if (key_type == BTRFS_DIR_INDEX_KEY) {
5868 if (ctx->pos >= INT_MAX)
5869 ctx->pos = LLONG_MAX;
5876 if (key_type == BTRFS_DIR_INDEX_KEY)
5877 btrfs_put_delayed_items(&ins_list, &del_list);
5878 btrfs_free_path(path);
5882 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5884 struct btrfs_root *root = BTRFS_I(inode)->root;
5885 struct btrfs_trans_handle *trans;
5887 bool nolock = false;
5889 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5892 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5895 if (wbc->sync_mode == WB_SYNC_ALL) {
5897 trans = btrfs_join_transaction_nolock(root);
5899 trans = btrfs_join_transaction(root);
5901 return PTR_ERR(trans);
5902 ret = btrfs_commit_transaction(trans, root);
5908 * This is somewhat expensive, updating the tree every time the
5909 * inode changes. But, it is most likely to find the inode in cache.
5910 * FIXME, needs more benchmarking...there are no reasons other than performance
5911 * to keep or drop this code.
5913 static int btrfs_dirty_inode(struct inode *inode)
5915 struct btrfs_root *root = BTRFS_I(inode)->root;
5916 struct btrfs_trans_handle *trans;
5919 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5922 trans = btrfs_join_transaction(root);
5924 return PTR_ERR(trans);
5926 ret = btrfs_update_inode(trans, root, inode);
5927 if (ret && ret == -ENOSPC) {
5928 /* whoops, lets try again with the full transaction */
5929 btrfs_end_transaction(trans, root);
5930 trans = btrfs_start_transaction(root, 1);
5932 return PTR_ERR(trans);
5934 ret = btrfs_update_inode(trans, root, inode);
5936 btrfs_end_transaction(trans, root);
5937 if (BTRFS_I(inode)->delayed_node)
5938 btrfs_balance_delayed_items(root);
5944 * This is a copy of file_update_time. We need this so we can return error on
5945 * ENOSPC for updating the inode in the case of file write and mmap writes.
5947 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5950 struct btrfs_root *root = BTRFS_I(inode)->root;
5952 if (btrfs_root_readonly(root))
5955 if (flags & S_VERSION)
5956 inode_inc_iversion(inode);
5957 if (flags & S_CTIME)
5958 inode->i_ctime = *now;
5959 if (flags & S_MTIME)
5960 inode->i_mtime = *now;
5961 if (flags & S_ATIME)
5962 inode->i_atime = *now;
5963 return btrfs_dirty_inode(inode);
5967 * find the highest existing sequence number in a directory
5968 * and then set the in-memory index_cnt variable to reflect
5969 * free sequence numbers
5971 static int btrfs_set_inode_index_count(struct inode *inode)
5973 struct btrfs_root *root = BTRFS_I(inode)->root;
5974 struct btrfs_key key, found_key;
5975 struct btrfs_path *path;
5976 struct extent_buffer *leaf;
5979 key.objectid = btrfs_ino(inode);
5980 key.type = BTRFS_DIR_INDEX_KEY;
5981 key.offset = (u64)-1;
5983 path = btrfs_alloc_path();
5987 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5990 /* FIXME: we should be able to handle this */
5996 * MAGIC NUMBER EXPLANATION:
5997 * since we search a directory based on f_pos we have to start at 2
5998 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5999 * else has to start at 2
6001 if (path->slots[0] == 0) {
6002 BTRFS_I(inode)->index_cnt = 2;
6008 leaf = path->nodes[0];
6009 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6011 if (found_key.objectid != btrfs_ino(inode) ||
6012 found_key.type != BTRFS_DIR_INDEX_KEY) {
6013 BTRFS_I(inode)->index_cnt = 2;
6017 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6019 btrfs_free_path(path);
6024 * helper to find a free sequence number in a given directory. This current
6025 * code is very simple, later versions will do smarter things in the btree
6027 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6031 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6032 ret = btrfs_inode_delayed_dir_index_count(dir);
6034 ret = btrfs_set_inode_index_count(dir);
6040 *index = BTRFS_I(dir)->index_cnt;
6041 BTRFS_I(dir)->index_cnt++;
6046 static int btrfs_insert_inode_locked(struct inode *inode)
6048 struct btrfs_iget_args args;
6049 args.location = &BTRFS_I(inode)->location;
6050 args.root = BTRFS_I(inode)->root;
6052 return insert_inode_locked4(inode,
6053 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6054 btrfs_find_actor, &args);
6057 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6058 struct btrfs_root *root,
6060 const char *name, int name_len,
6061 u64 ref_objectid, u64 objectid,
6062 umode_t mode, u64 *index)
6064 struct inode *inode;
6065 struct btrfs_inode_item *inode_item;
6066 struct btrfs_key *location;
6067 struct btrfs_path *path;
6068 struct btrfs_inode_ref *ref;
6069 struct btrfs_key key[2];
6071 int nitems = name ? 2 : 1;
6075 path = btrfs_alloc_path();
6077 return ERR_PTR(-ENOMEM);
6079 inode = new_inode(root->fs_info->sb);
6081 btrfs_free_path(path);
6082 return ERR_PTR(-ENOMEM);
6086 * O_TMPFILE, set link count to 0, so that after this point,
6087 * we fill in an inode item with the correct link count.
6090 set_nlink(inode, 0);
6093 * we have to initialize this early, so we can reclaim the inode
6094 * number if we fail afterwards in this function.
6096 inode->i_ino = objectid;
6099 trace_btrfs_inode_request(dir);
6101 ret = btrfs_set_inode_index(dir, index);
6103 btrfs_free_path(path);
6105 return ERR_PTR(ret);
6111 * index_cnt is ignored for everything but a dir,
6112 * btrfs_get_inode_index_count has an explanation for the magic
6115 BTRFS_I(inode)->index_cnt = 2;
6116 BTRFS_I(inode)->dir_index = *index;
6117 BTRFS_I(inode)->root = root;
6118 BTRFS_I(inode)->generation = trans->transid;
6119 inode->i_generation = BTRFS_I(inode)->generation;
6122 * We could have gotten an inode number from somebody who was fsynced
6123 * and then removed in this same transaction, so let's just set full
6124 * sync since it will be a full sync anyway and this will blow away the
6125 * old info in the log.
6127 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6129 key[0].objectid = objectid;
6130 key[0].type = BTRFS_INODE_ITEM_KEY;
6133 sizes[0] = sizeof(struct btrfs_inode_item);
6137 * Start new inodes with an inode_ref. This is slightly more
6138 * efficient for small numbers of hard links since they will
6139 * be packed into one item. Extended refs will kick in if we
6140 * add more hard links than can fit in the ref item.
6142 key[1].objectid = objectid;
6143 key[1].type = BTRFS_INODE_REF_KEY;
6144 key[1].offset = ref_objectid;
6146 sizes[1] = name_len + sizeof(*ref);
6149 location = &BTRFS_I(inode)->location;
6150 location->objectid = objectid;
6151 location->offset = 0;
6152 location->type = BTRFS_INODE_ITEM_KEY;
6154 ret = btrfs_insert_inode_locked(inode);
6158 path->leave_spinning = 1;
6159 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6163 inode_init_owner(inode, dir, mode);
6164 inode_set_bytes(inode, 0);
6166 inode->i_mtime = CURRENT_TIME;
6167 inode->i_atime = inode->i_mtime;
6168 inode->i_ctime = inode->i_mtime;
6169 BTRFS_I(inode)->i_otime = inode->i_mtime;
6171 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6172 struct btrfs_inode_item);
6173 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6174 sizeof(*inode_item));
6175 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6178 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6179 struct btrfs_inode_ref);
6180 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6181 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6182 ptr = (unsigned long)(ref + 1);
6183 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6186 btrfs_mark_buffer_dirty(path->nodes[0]);
6187 btrfs_free_path(path);
6189 btrfs_inherit_iflags(inode, dir);
6191 if (S_ISREG(mode)) {
6192 if (btrfs_test_opt(root, NODATASUM))
6193 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6194 if (btrfs_test_opt(root, NODATACOW))
6195 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6196 BTRFS_INODE_NODATASUM;
6199 inode_tree_add(inode);
6201 trace_btrfs_inode_new(inode);
6202 btrfs_set_inode_last_trans(trans, inode);
6204 btrfs_update_root_times(trans, root);
6206 ret = btrfs_inode_inherit_props(trans, inode, dir);
6208 btrfs_err(root->fs_info,
6209 "error inheriting props for ino %llu (root %llu): %d",
6210 btrfs_ino(inode), root->root_key.objectid, ret);
6215 unlock_new_inode(inode);
6218 BTRFS_I(dir)->index_cnt--;
6219 btrfs_free_path(path);
6221 return ERR_PTR(ret);
6224 static inline u8 btrfs_inode_type(struct inode *inode)
6226 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6230 * utility function to add 'inode' into 'parent_inode' with
6231 * a give name and a given sequence number.
6232 * if 'add_backref' is true, also insert a backref from the
6233 * inode to the parent directory.
6235 int btrfs_add_link(struct btrfs_trans_handle *trans,
6236 struct inode *parent_inode, struct inode *inode,
6237 const char *name, int name_len, int add_backref, u64 index)
6240 struct btrfs_key key;
6241 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6242 u64 ino = btrfs_ino(inode);
6243 u64 parent_ino = btrfs_ino(parent_inode);
6245 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6246 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6249 key.type = BTRFS_INODE_ITEM_KEY;
6253 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6254 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6255 key.objectid, root->root_key.objectid,
6256 parent_ino, index, name, name_len);
6257 } else if (add_backref) {
6258 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6262 /* Nothing to clean up yet */
6266 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6268 btrfs_inode_type(inode), index);
6269 if (ret == -EEXIST || ret == -EOVERFLOW)
6272 btrfs_abort_transaction(trans, root, ret);
6276 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6278 inode_inc_iversion(parent_inode);
6279 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6280 ret = btrfs_update_inode(trans, root, parent_inode);
6282 btrfs_abort_transaction(trans, root, ret);
6286 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6289 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6290 key.objectid, root->root_key.objectid,
6291 parent_ino, &local_index, name, name_len);
6293 } else if (add_backref) {
6297 err = btrfs_del_inode_ref(trans, root, name, name_len,
6298 ino, parent_ino, &local_index);
6303 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6304 struct inode *dir, struct dentry *dentry,
6305 struct inode *inode, int backref, u64 index)
6307 int err = btrfs_add_link(trans, dir, inode,
6308 dentry->d_name.name, dentry->d_name.len,
6315 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6316 umode_t mode, dev_t rdev)
6318 struct btrfs_trans_handle *trans;
6319 struct btrfs_root *root = BTRFS_I(dir)->root;
6320 struct inode *inode = NULL;
6327 * 2 for inode item and ref
6329 * 1 for xattr if selinux is on
6331 trans = btrfs_start_transaction(root, 5);
6333 return PTR_ERR(trans);
6335 err = btrfs_find_free_ino(root, &objectid);
6339 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6340 dentry->d_name.len, btrfs_ino(dir), objectid,
6342 if (IS_ERR(inode)) {
6343 err = PTR_ERR(inode);
6348 * If the active LSM wants to access the inode during
6349 * d_instantiate it needs these. Smack checks to see
6350 * if the filesystem supports xattrs by looking at the
6353 inode->i_op = &btrfs_special_inode_operations;
6354 init_special_inode(inode, inode->i_mode, rdev);
6356 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6358 goto out_unlock_inode;
6360 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6362 goto out_unlock_inode;
6364 btrfs_update_inode(trans, root, inode);
6365 unlock_new_inode(inode);
6366 d_instantiate(dentry, inode);
6370 btrfs_end_transaction(trans, root);
6371 btrfs_balance_delayed_items(root);
6372 btrfs_btree_balance_dirty(root);
6374 inode_dec_link_count(inode);
6381 unlock_new_inode(inode);
6386 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6387 umode_t mode, bool excl)
6389 struct btrfs_trans_handle *trans;
6390 struct btrfs_root *root = BTRFS_I(dir)->root;
6391 struct inode *inode = NULL;
6392 int drop_inode_on_err = 0;
6398 * 2 for inode item and ref
6400 * 1 for xattr if selinux is on
6402 trans = btrfs_start_transaction(root, 5);
6404 return PTR_ERR(trans);
6406 err = btrfs_find_free_ino(root, &objectid);
6410 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6411 dentry->d_name.len, btrfs_ino(dir), objectid,
6413 if (IS_ERR(inode)) {
6414 err = PTR_ERR(inode);
6417 drop_inode_on_err = 1;
6419 * If the active LSM wants to access the inode during
6420 * d_instantiate it needs these. Smack checks to see
6421 * if the filesystem supports xattrs by looking at the
6424 inode->i_fop = &btrfs_file_operations;
6425 inode->i_op = &btrfs_file_inode_operations;
6426 inode->i_mapping->a_ops = &btrfs_aops;
6428 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6430 goto out_unlock_inode;
6432 err = btrfs_update_inode(trans, root, inode);
6434 goto out_unlock_inode;
6436 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6438 goto out_unlock_inode;
6440 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6441 unlock_new_inode(inode);
6442 d_instantiate(dentry, inode);
6445 btrfs_end_transaction(trans, root);
6446 if (err && drop_inode_on_err) {
6447 inode_dec_link_count(inode);
6450 btrfs_balance_delayed_items(root);
6451 btrfs_btree_balance_dirty(root);
6455 unlock_new_inode(inode);
6460 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6461 struct dentry *dentry)
6463 struct btrfs_trans_handle *trans = NULL;
6464 struct btrfs_root *root = BTRFS_I(dir)->root;
6465 struct inode *inode = d_inode(old_dentry);
6470 /* do not allow sys_link's with other subvols of the same device */
6471 if (root->objectid != BTRFS_I(inode)->root->objectid)
6474 if (inode->i_nlink >= BTRFS_LINK_MAX)
6477 err = btrfs_set_inode_index(dir, &index);
6482 * 2 items for inode and inode ref
6483 * 2 items for dir items
6484 * 1 item for parent inode
6486 trans = btrfs_start_transaction(root, 5);
6487 if (IS_ERR(trans)) {
6488 err = PTR_ERR(trans);
6493 /* There are several dir indexes for this inode, clear the cache. */
6494 BTRFS_I(inode)->dir_index = 0ULL;
6496 inode_inc_iversion(inode);
6497 inode->i_ctime = CURRENT_TIME;
6499 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6501 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6506 struct dentry *parent = dentry->d_parent;
6507 err = btrfs_update_inode(trans, root, inode);
6510 if (inode->i_nlink == 1) {
6512 * If new hard link count is 1, it's a file created
6513 * with open(2) O_TMPFILE flag.
6515 err = btrfs_orphan_del(trans, inode);
6519 d_instantiate(dentry, inode);
6520 btrfs_log_new_name(trans, inode, NULL, parent);
6523 btrfs_balance_delayed_items(root);
6526 btrfs_end_transaction(trans, root);
6528 inode_dec_link_count(inode);
6531 btrfs_btree_balance_dirty(root);
6535 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6537 struct inode *inode = NULL;
6538 struct btrfs_trans_handle *trans;
6539 struct btrfs_root *root = BTRFS_I(dir)->root;
6541 int drop_on_err = 0;
6546 * 2 items for inode and ref
6547 * 2 items for dir items
6548 * 1 for xattr if selinux is on
6550 trans = btrfs_start_transaction(root, 5);
6552 return PTR_ERR(trans);
6554 err = btrfs_find_free_ino(root, &objectid);
6558 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6559 dentry->d_name.len, btrfs_ino(dir), objectid,
6560 S_IFDIR | mode, &index);
6561 if (IS_ERR(inode)) {
6562 err = PTR_ERR(inode);
6567 /* these must be set before we unlock the inode */
6568 inode->i_op = &btrfs_dir_inode_operations;
6569 inode->i_fop = &btrfs_dir_file_operations;
6571 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6573 goto out_fail_inode;
6575 btrfs_i_size_write(inode, 0);
6576 err = btrfs_update_inode(trans, root, inode);
6578 goto out_fail_inode;
6580 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6581 dentry->d_name.len, 0, index);
6583 goto out_fail_inode;
6585 d_instantiate(dentry, inode);
6587 * mkdir is special. We're unlocking after we call d_instantiate
6588 * to avoid a race with nfsd calling d_instantiate.
6590 unlock_new_inode(inode);
6594 btrfs_end_transaction(trans, root);
6596 inode_dec_link_count(inode);
6599 btrfs_balance_delayed_items(root);
6600 btrfs_btree_balance_dirty(root);
6604 unlock_new_inode(inode);
6608 /* Find next extent map of a given extent map, caller needs to ensure locks */
6609 static struct extent_map *next_extent_map(struct extent_map *em)
6611 struct rb_node *next;
6613 next = rb_next(&em->rb_node);
6616 return container_of(next, struct extent_map, rb_node);
6619 static struct extent_map *prev_extent_map(struct extent_map *em)
6621 struct rb_node *prev;
6623 prev = rb_prev(&em->rb_node);
6626 return container_of(prev, struct extent_map, rb_node);
6629 /* helper for btfs_get_extent. Given an existing extent in the tree,
6630 * the existing extent is the nearest extent to map_start,
6631 * and an extent that you want to insert, deal with overlap and insert
6632 * the best fitted new extent into the tree.
6634 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6635 struct extent_map *existing,
6636 struct extent_map *em,
6639 struct extent_map *prev;
6640 struct extent_map *next;
6645 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6647 if (existing->start > map_start) {
6649 prev = prev_extent_map(next);
6652 next = next_extent_map(prev);
6655 start = prev ? extent_map_end(prev) : em->start;
6656 start = max_t(u64, start, em->start);
6657 end = next ? next->start : extent_map_end(em);
6658 end = min_t(u64, end, extent_map_end(em));
6659 start_diff = start - em->start;
6661 em->len = end - start;
6662 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6663 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6664 em->block_start += start_diff;
6665 em->block_len -= start_diff;
6667 return add_extent_mapping(em_tree, em, 0);
6670 static noinline int uncompress_inline(struct btrfs_path *path,
6672 size_t pg_offset, u64 extent_offset,
6673 struct btrfs_file_extent_item *item)
6676 struct extent_buffer *leaf = path->nodes[0];
6679 unsigned long inline_size;
6683 WARN_ON(pg_offset != 0);
6684 compress_type = btrfs_file_extent_compression(leaf, item);
6685 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6686 inline_size = btrfs_file_extent_inline_item_len(leaf,
6687 btrfs_item_nr(path->slots[0]));
6688 tmp = kmalloc(inline_size, GFP_NOFS);
6691 ptr = btrfs_file_extent_inline_start(item);
6693 read_extent_buffer(leaf, tmp, ptr, inline_size);
6695 max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6696 ret = btrfs_decompress(compress_type, tmp, page,
6697 extent_offset, inline_size, max_size);
6703 * a bit scary, this does extent mapping from logical file offset to the disk.
6704 * the ugly parts come from merging extents from the disk with the in-ram
6705 * representation. This gets more complex because of the data=ordered code,
6706 * where the in-ram extents might be locked pending data=ordered completion.
6708 * This also copies inline extents directly into the page.
6711 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6712 size_t pg_offset, u64 start, u64 len,
6717 u64 extent_start = 0;
6719 u64 objectid = btrfs_ino(inode);
6721 struct btrfs_path *path = NULL;
6722 struct btrfs_root *root = BTRFS_I(inode)->root;
6723 struct btrfs_file_extent_item *item;
6724 struct extent_buffer *leaf;
6725 struct btrfs_key found_key;
6726 struct extent_map *em = NULL;
6727 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6728 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6729 struct btrfs_trans_handle *trans = NULL;
6730 const bool new_inline = !page || create;
6733 read_lock(&em_tree->lock);
6734 em = lookup_extent_mapping(em_tree, start, len);
6736 em->bdev = root->fs_info->fs_devices->latest_bdev;
6737 read_unlock(&em_tree->lock);
6740 if (em->start > start || em->start + em->len <= start)
6741 free_extent_map(em);
6742 else if (em->block_start == EXTENT_MAP_INLINE && page)
6743 free_extent_map(em);
6747 em = alloc_extent_map();
6752 em->bdev = root->fs_info->fs_devices->latest_bdev;
6753 em->start = EXTENT_MAP_HOLE;
6754 em->orig_start = EXTENT_MAP_HOLE;
6756 em->block_len = (u64)-1;
6759 path = btrfs_alloc_path();
6765 * Chances are we'll be called again, so go ahead and do
6768 path->reada = READA_FORWARD;
6771 ret = btrfs_lookup_file_extent(trans, root, path,
6772 objectid, start, trans != NULL);
6779 if (path->slots[0] == 0)
6784 leaf = path->nodes[0];
6785 item = btrfs_item_ptr(leaf, path->slots[0],
6786 struct btrfs_file_extent_item);
6787 /* are we inside the extent that was found? */
6788 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6789 found_type = found_key.type;
6790 if (found_key.objectid != objectid ||
6791 found_type != BTRFS_EXTENT_DATA_KEY) {
6793 * If we backup past the first extent we want to move forward
6794 * and see if there is an extent in front of us, otherwise we'll
6795 * say there is a hole for our whole search range which can
6802 found_type = btrfs_file_extent_type(leaf, item);
6803 extent_start = found_key.offset;
6804 if (found_type == BTRFS_FILE_EXTENT_REG ||
6805 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6806 extent_end = extent_start +
6807 btrfs_file_extent_num_bytes(leaf, item);
6808 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6810 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6811 extent_end = ALIGN(extent_start + size, root->sectorsize);
6814 if (start >= extent_end) {
6816 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6817 ret = btrfs_next_leaf(root, path);
6824 leaf = path->nodes[0];
6826 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6827 if (found_key.objectid != objectid ||
6828 found_key.type != BTRFS_EXTENT_DATA_KEY)
6830 if (start + len <= found_key.offset)
6832 if (start > found_key.offset)
6835 em->orig_start = start;
6836 em->len = found_key.offset - start;
6840 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6842 if (found_type == BTRFS_FILE_EXTENT_REG ||
6843 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6845 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6849 size_t extent_offset;
6855 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6856 extent_offset = page_offset(page) + pg_offset - extent_start;
6857 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6858 size - extent_offset);
6859 em->start = extent_start + extent_offset;
6860 em->len = ALIGN(copy_size, root->sectorsize);
6861 em->orig_block_len = em->len;
6862 em->orig_start = em->start;
6863 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6864 if (create == 0 && !PageUptodate(page)) {
6865 if (btrfs_file_extent_compression(leaf, item) !=
6866 BTRFS_COMPRESS_NONE) {
6867 ret = uncompress_inline(path, page, pg_offset,
6868 extent_offset, item);
6875 read_extent_buffer(leaf, map + pg_offset, ptr,
6877 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6878 memset(map + pg_offset + copy_size, 0,
6879 PAGE_CACHE_SIZE - pg_offset -
6884 flush_dcache_page(page);
6885 } else if (create && PageUptodate(page)) {
6889 free_extent_map(em);
6892 btrfs_release_path(path);
6893 trans = btrfs_join_transaction(root);
6896 return ERR_CAST(trans);
6900 write_extent_buffer(leaf, map + pg_offset, ptr,
6903 btrfs_mark_buffer_dirty(leaf);
6905 set_extent_uptodate(io_tree, em->start,
6906 extent_map_end(em) - 1, NULL, GFP_NOFS);
6911 em->orig_start = start;
6914 em->block_start = EXTENT_MAP_HOLE;
6915 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6917 btrfs_release_path(path);
6918 if (em->start > start || extent_map_end(em) <= start) {
6919 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6920 em->start, em->len, start, len);
6926 write_lock(&em_tree->lock);
6927 ret = add_extent_mapping(em_tree, em, 0);
6928 /* it is possible that someone inserted the extent into the tree
6929 * while we had the lock dropped. It is also possible that
6930 * an overlapping map exists in the tree
6932 if (ret == -EEXIST) {
6933 struct extent_map *existing;
6937 existing = search_extent_mapping(em_tree, start, len);
6939 * existing will always be non-NULL, since there must be
6940 * extent causing the -EEXIST.
6942 if (start >= extent_map_end(existing) ||
6943 start <= existing->start) {
6945 * The existing extent map is the one nearest to
6946 * the [start, start + len) range which overlaps
6948 err = merge_extent_mapping(em_tree, existing,
6950 free_extent_map(existing);
6952 free_extent_map(em);
6956 free_extent_map(em);
6961 write_unlock(&em_tree->lock);
6964 trace_btrfs_get_extent(root, em);
6966 btrfs_free_path(path);
6968 ret = btrfs_end_transaction(trans, root);
6973 free_extent_map(em);
6974 return ERR_PTR(err);
6976 BUG_ON(!em); /* Error is always set */
6980 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6981 size_t pg_offset, u64 start, u64 len,
6984 struct extent_map *em;
6985 struct extent_map *hole_em = NULL;
6986 u64 range_start = start;
6992 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6999 * - a pre-alloc extent,
7000 * there might actually be delalloc bytes behind it.
7002 if (em->block_start != EXTENT_MAP_HOLE &&
7003 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7009 /* check to see if we've wrapped (len == -1 or similar) */
7018 /* ok, we didn't find anything, lets look for delalloc */
7019 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7020 end, len, EXTENT_DELALLOC, 1);
7021 found_end = range_start + found;
7022 if (found_end < range_start)
7023 found_end = (u64)-1;
7026 * we didn't find anything useful, return
7027 * the original results from get_extent()
7029 if (range_start > end || found_end <= start) {
7035 /* adjust the range_start to make sure it doesn't
7036 * go backwards from the start they passed in
7038 range_start = max(start, range_start);
7039 found = found_end - range_start;
7042 u64 hole_start = start;
7045 em = alloc_extent_map();
7051 * when btrfs_get_extent can't find anything it
7052 * returns one huge hole
7054 * make sure what it found really fits our range, and
7055 * adjust to make sure it is based on the start from
7059 u64 calc_end = extent_map_end(hole_em);
7061 if (calc_end <= start || (hole_em->start > end)) {
7062 free_extent_map(hole_em);
7065 hole_start = max(hole_em->start, start);
7066 hole_len = calc_end - hole_start;
7070 if (hole_em && range_start > hole_start) {
7071 /* our hole starts before our delalloc, so we
7072 * have to return just the parts of the hole
7073 * that go until the delalloc starts
7075 em->len = min(hole_len,
7076 range_start - hole_start);
7077 em->start = hole_start;
7078 em->orig_start = hole_start;
7080 * don't adjust block start at all,
7081 * it is fixed at EXTENT_MAP_HOLE
7083 em->block_start = hole_em->block_start;
7084 em->block_len = hole_len;
7085 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7086 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7088 em->start = range_start;
7090 em->orig_start = range_start;
7091 em->block_start = EXTENT_MAP_DELALLOC;
7092 em->block_len = found;
7094 } else if (hole_em) {
7099 free_extent_map(hole_em);
7101 free_extent_map(em);
7102 return ERR_PTR(err);
7107 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7110 struct btrfs_root *root = BTRFS_I(inode)->root;
7111 struct extent_map *em;
7112 struct btrfs_key ins;
7116 alloc_hint = get_extent_allocation_hint(inode, start, len);
7117 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7118 alloc_hint, &ins, 1, 1);
7120 return ERR_PTR(ret);
7123 * Create the ordered extent before the extent map. This is to avoid
7124 * races with the fast fsync path that would lead to it logging file
7125 * extent items that point to disk extents that were not yet written to.
7126 * The fast fsync path collects ordered extents into a local list and
7127 * then collects all the new extent maps, so we must create the ordered
7128 * extent first and make sure the fast fsync path collects any new
7129 * ordered extents after collecting new extent maps as well.
7130 * The fsync path simply can not rely on inode_dio_wait() because it
7131 * causes deadlock with AIO.
7133 ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
7134 ins.offset, ins.offset, 0);
7136 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7137 return ERR_PTR(ret);
7140 em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
7141 ins.offset, ins.offset, ins.offset, 0);
7143 struct btrfs_ordered_extent *oe;
7145 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7146 oe = btrfs_lookup_ordered_extent(inode, start);
7150 set_bit(BTRFS_ORDERED_IOERR, &oe->flags);
7151 set_bit(BTRFS_ORDERED_IO_DONE, &oe->flags);
7152 btrfs_remove_ordered_extent(inode, oe);
7153 /* Once for our lookup and once for the ordered extents tree. */
7154 btrfs_put_ordered_extent(oe);
7155 btrfs_put_ordered_extent(oe);
7161 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7162 * block must be cow'd
7164 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7165 u64 *orig_start, u64 *orig_block_len,
7168 struct btrfs_trans_handle *trans;
7169 struct btrfs_path *path;
7171 struct extent_buffer *leaf;
7172 struct btrfs_root *root = BTRFS_I(inode)->root;
7173 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7174 struct btrfs_file_extent_item *fi;
7175 struct btrfs_key key;
7182 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7184 path = btrfs_alloc_path();
7188 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7193 slot = path->slots[0];
7196 /* can't find the item, must cow */
7203 leaf = path->nodes[0];
7204 btrfs_item_key_to_cpu(leaf, &key, slot);
7205 if (key.objectid != btrfs_ino(inode) ||
7206 key.type != BTRFS_EXTENT_DATA_KEY) {
7207 /* not our file or wrong item type, must cow */
7211 if (key.offset > offset) {
7212 /* Wrong offset, must cow */
7216 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7217 found_type = btrfs_file_extent_type(leaf, fi);
7218 if (found_type != BTRFS_FILE_EXTENT_REG &&
7219 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7220 /* not a regular extent, must cow */
7224 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7227 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7228 if (extent_end <= offset)
7231 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7232 if (disk_bytenr == 0)
7235 if (btrfs_file_extent_compression(leaf, fi) ||
7236 btrfs_file_extent_encryption(leaf, fi) ||
7237 btrfs_file_extent_other_encoding(leaf, fi))
7240 backref_offset = btrfs_file_extent_offset(leaf, fi);
7243 *orig_start = key.offset - backref_offset;
7244 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7245 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7248 if (btrfs_extent_readonly(root, disk_bytenr))
7251 num_bytes = min(offset + *len, extent_end) - offset;
7252 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7255 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7256 ret = test_range_bit(io_tree, offset, range_end,
7257 EXTENT_DELALLOC, 0, NULL);
7264 btrfs_release_path(path);
7267 * look for other files referencing this extent, if we
7268 * find any we must cow
7270 trans = btrfs_join_transaction(root);
7271 if (IS_ERR(trans)) {
7276 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7277 key.offset - backref_offset, disk_bytenr);
7278 btrfs_end_transaction(trans, root);
7285 * adjust disk_bytenr and num_bytes to cover just the bytes
7286 * in this extent we are about to write. If there
7287 * are any csums in that range we have to cow in order
7288 * to keep the csums correct
7290 disk_bytenr += backref_offset;
7291 disk_bytenr += offset - key.offset;
7292 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7295 * all of the above have passed, it is safe to overwrite this extent
7301 btrfs_free_path(path);
7305 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7307 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7309 void **pagep = NULL;
7310 struct page *page = NULL;
7314 start_idx = start >> PAGE_CACHE_SHIFT;
7317 * end is the last byte in the last page. end == start is legal
7319 end_idx = end >> PAGE_CACHE_SHIFT;
7323 /* Most of the code in this while loop is lifted from
7324 * find_get_page. It's been modified to begin searching from a
7325 * page and return just the first page found in that range. If the
7326 * found idx is less than or equal to the end idx then we know that
7327 * a page exists. If no pages are found or if those pages are
7328 * outside of the range then we're fine (yay!) */
7329 while (page == NULL &&
7330 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7331 page = radix_tree_deref_slot(pagep);
7332 if (unlikely(!page))
7335 if (radix_tree_exception(page)) {
7336 if (radix_tree_deref_retry(page)) {
7341 * Otherwise, shmem/tmpfs must be storing a swap entry
7342 * here as an exceptional entry: so return it without
7343 * attempting to raise page count.
7346 break; /* TODO: Is this relevant for this use case? */
7349 if (!page_cache_get_speculative(page)) {
7355 * Has the page moved?
7356 * This is part of the lockless pagecache protocol. See
7357 * include/linux/pagemap.h for details.
7359 if (unlikely(page != *pagep)) {
7360 page_cache_release(page);
7366 if (page->index <= end_idx)
7368 page_cache_release(page);
7375 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7376 struct extent_state **cached_state, int writing)
7378 struct btrfs_ordered_extent *ordered;
7382 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7385 * We're concerned with the entire range that we're going to be
7386 * doing DIO to, so we need to make sure theres no ordered
7387 * extents in this range.
7389 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7390 lockend - lockstart + 1);
7393 * We need to make sure there are no buffered pages in this
7394 * range either, we could have raced between the invalidate in
7395 * generic_file_direct_write and locking the extent. The
7396 * invalidate needs to happen so that reads after a write do not
7401 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7404 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7405 cached_state, GFP_NOFS);
7408 btrfs_start_ordered_extent(inode, ordered, 1);
7409 btrfs_put_ordered_extent(ordered);
7412 * We could trigger writeback for this range (and wait
7413 * for it to complete) and then invalidate the pages for
7414 * this range (through invalidate_inode_pages2_range()),
7415 * but that can lead us to a deadlock with a concurrent
7416 * call to readpages() (a buffered read or a defrag call
7417 * triggered a readahead) on a page lock due to an
7418 * ordered dio extent we created before but did not have
7419 * yet a corresponding bio submitted (whence it can not
7420 * complete), which makes readpages() wait for that
7421 * ordered extent to complete while holding a lock on
7434 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7435 u64 len, u64 orig_start,
7436 u64 block_start, u64 block_len,
7437 u64 orig_block_len, u64 ram_bytes,
7440 struct extent_map_tree *em_tree;
7441 struct extent_map *em;
7442 struct btrfs_root *root = BTRFS_I(inode)->root;
7445 em_tree = &BTRFS_I(inode)->extent_tree;
7446 em = alloc_extent_map();
7448 return ERR_PTR(-ENOMEM);
7451 em->orig_start = orig_start;
7452 em->mod_start = start;
7455 em->block_len = block_len;
7456 em->block_start = block_start;
7457 em->bdev = root->fs_info->fs_devices->latest_bdev;
7458 em->orig_block_len = orig_block_len;
7459 em->ram_bytes = ram_bytes;
7460 em->generation = -1;
7461 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7462 if (type == BTRFS_ORDERED_PREALLOC)
7463 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7466 btrfs_drop_extent_cache(inode, em->start,
7467 em->start + em->len - 1, 0);
7468 write_lock(&em_tree->lock);
7469 ret = add_extent_mapping(em_tree, em, 1);
7470 write_unlock(&em_tree->lock);
7471 } while (ret == -EEXIST);
7474 free_extent_map(em);
7475 return ERR_PTR(ret);
7481 static void adjust_dio_outstanding_extents(struct inode *inode,
7482 struct btrfs_dio_data *dio_data,
7485 unsigned num_extents;
7487 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7488 BTRFS_MAX_EXTENT_SIZE);
7490 * If we have an outstanding_extents count still set then we're
7491 * within our reservation, otherwise we need to adjust our inode
7492 * counter appropriately.
7494 if (dio_data->outstanding_extents) {
7495 dio_data->outstanding_extents -= num_extents;
7497 spin_lock(&BTRFS_I(inode)->lock);
7498 BTRFS_I(inode)->outstanding_extents += num_extents;
7499 spin_unlock(&BTRFS_I(inode)->lock);
7503 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7504 struct buffer_head *bh_result, int create)
7506 struct extent_map *em;
7507 struct btrfs_root *root = BTRFS_I(inode)->root;
7508 struct extent_state *cached_state = NULL;
7509 struct btrfs_dio_data *dio_data = NULL;
7510 u64 start = iblock << inode->i_blkbits;
7511 u64 lockstart, lockend;
7512 u64 len = bh_result->b_size;
7513 int unlock_bits = EXTENT_LOCKED;
7517 unlock_bits |= EXTENT_DIRTY;
7519 len = min_t(u64, len, root->sectorsize);
7522 lockend = start + len - 1;
7524 if (current->journal_info) {
7526 * Need to pull our outstanding extents and set journal_info to NULL so
7527 * that anything that needs to check if there's a transction doesn't get
7530 dio_data = current->journal_info;
7531 current->journal_info = NULL;
7535 * If this errors out it's because we couldn't invalidate pagecache for
7536 * this range and we need to fallback to buffered.
7538 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7544 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7551 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7552 * io. INLINE is special, and we could probably kludge it in here, but
7553 * it's still buffered so for safety lets just fall back to the generic
7556 * For COMPRESSED we _have_ to read the entire extent in so we can
7557 * decompress it, so there will be buffering required no matter what we
7558 * do, so go ahead and fallback to buffered.
7560 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7561 * to buffered IO. Don't blame me, this is the price we pay for using
7564 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7565 em->block_start == EXTENT_MAP_INLINE) {
7566 free_extent_map(em);
7571 /* Just a good old fashioned hole, return */
7572 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7573 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7574 free_extent_map(em);
7579 * We don't allocate a new extent in the following cases
7581 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7583 * 2) The extent is marked as PREALLOC. We're good to go here and can
7584 * just use the extent.
7588 len = min(len, em->len - (start - em->start));
7589 lockstart = start + len;
7593 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7594 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7595 em->block_start != EXTENT_MAP_HOLE)) {
7597 u64 block_start, orig_start, orig_block_len, ram_bytes;
7599 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7600 type = BTRFS_ORDERED_PREALLOC;
7602 type = BTRFS_ORDERED_NOCOW;
7603 len = min(len, em->len - (start - em->start));
7604 block_start = em->block_start + (start - em->start);
7606 if (can_nocow_extent(inode, start, &len, &orig_start,
7607 &orig_block_len, &ram_bytes) == 1) {
7608 if (type == BTRFS_ORDERED_PREALLOC) {
7609 free_extent_map(em);
7610 em = create_pinned_em(inode, start, len,
7621 ret = btrfs_add_ordered_extent_dio(inode, start,
7622 block_start, len, len, type);
7624 free_extent_map(em);
7632 * this will cow the extent, reset the len in case we changed
7635 len = bh_result->b_size;
7636 free_extent_map(em);
7637 em = btrfs_new_extent_direct(inode, start, len);
7642 len = min(len, em->len - (start - em->start));
7644 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7646 bh_result->b_size = len;
7647 bh_result->b_bdev = em->bdev;
7648 set_buffer_mapped(bh_result);
7650 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7651 set_buffer_new(bh_result);
7654 * Need to update the i_size under the extent lock so buffered
7655 * readers will get the updated i_size when we unlock.
7657 if (start + len > i_size_read(inode))
7658 i_size_write(inode, start + len);
7660 adjust_dio_outstanding_extents(inode, dio_data, len);
7661 btrfs_free_reserved_data_space(inode, start, len);
7662 WARN_ON(dio_data->reserve < len);
7663 dio_data->reserve -= len;
7664 dio_data->unsubmitted_oe_range_end = start + len;
7665 current->journal_info = dio_data;
7669 * In the case of write we need to clear and unlock the entire range,
7670 * in the case of read we need to unlock only the end area that we
7671 * aren't using if there is any left over space.
7673 if (lockstart < lockend) {
7674 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7675 lockend, unlock_bits, 1, 0,
7676 &cached_state, GFP_NOFS);
7678 free_extent_state(cached_state);
7681 free_extent_map(em);
7686 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7687 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7690 current->journal_info = dio_data;
7692 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7693 * write less data then expected, so that we don't underflow our inode's
7694 * outstanding extents counter.
7696 if (create && dio_data)
7697 adjust_dio_outstanding_extents(inode, dio_data, len);
7702 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7703 int rw, int mirror_num)
7705 struct btrfs_root *root = BTRFS_I(inode)->root;
7708 BUG_ON(rw & REQ_WRITE);
7712 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7713 BTRFS_WQ_ENDIO_DIO_REPAIR);
7717 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7723 static int btrfs_check_dio_repairable(struct inode *inode,
7724 struct bio *failed_bio,
7725 struct io_failure_record *failrec,
7730 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7731 failrec->logical, failrec->len);
7732 if (num_copies == 1) {
7734 * we only have a single copy of the data, so don't bother with
7735 * all the retry and error correction code that follows. no
7736 * matter what the error is, it is very likely to persist.
7738 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7739 num_copies, failrec->this_mirror, failed_mirror);
7743 failrec->failed_mirror = failed_mirror;
7744 failrec->this_mirror++;
7745 if (failrec->this_mirror == failed_mirror)
7746 failrec->this_mirror++;
7748 if (failrec->this_mirror > num_copies) {
7749 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7750 num_copies, failrec->this_mirror, failed_mirror);
7757 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7758 struct page *page, unsigned int pgoff,
7759 u64 start, u64 end, int failed_mirror,
7760 bio_end_io_t *repair_endio, void *repair_arg)
7762 struct io_failure_record *failrec;
7768 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7770 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7774 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7777 free_io_failure(inode, failrec);
7781 if ((failed_bio->bi_vcnt > 1)
7782 || (failed_bio->bi_io_vec->bv_len
7783 > BTRFS_I(inode)->root->sectorsize))
7784 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7786 read_mode = READ_SYNC;
7788 isector = start - btrfs_io_bio(failed_bio)->logical;
7789 isector >>= inode->i_sb->s_blocksize_bits;
7790 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7791 pgoff, isector, repair_endio, repair_arg);
7793 free_io_failure(inode, failrec);
7797 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7798 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7799 read_mode, failrec->this_mirror, failrec->in_validation);
7801 ret = submit_dio_repair_bio(inode, bio, read_mode,
7802 failrec->this_mirror);
7804 free_io_failure(inode, failrec);
7811 struct btrfs_retry_complete {
7812 struct completion done;
7813 struct inode *inode;
7818 static void btrfs_retry_endio_nocsum(struct bio *bio)
7820 struct btrfs_retry_complete *done = bio->bi_private;
7821 struct inode *inode;
7822 struct bio_vec *bvec;
7828 ASSERT(bio->bi_vcnt == 1);
7829 inode = bio->bi_io_vec->bv_page->mapping->host;
7830 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7833 bio_for_each_segment_all(bvec, bio, i)
7834 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7836 complete(&done->done);
7840 static int __btrfs_correct_data_nocsum(struct inode *inode,
7841 struct btrfs_io_bio *io_bio)
7843 struct btrfs_fs_info *fs_info;
7844 struct bio_vec *bvec;
7845 struct btrfs_retry_complete done;
7853 fs_info = BTRFS_I(inode)->root->fs_info;
7854 sectorsize = BTRFS_I(inode)->root->sectorsize;
7856 start = io_bio->logical;
7859 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7860 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7861 pgoff = bvec->bv_offset;
7863 next_block_or_try_again:
7866 init_completion(&done.done);
7868 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7869 pgoff, start, start + sectorsize - 1,
7871 btrfs_retry_endio_nocsum, &done);
7875 wait_for_completion(&done.done);
7877 if (!done.uptodate) {
7878 /* We might have another mirror, so try again */
7879 goto next_block_or_try_again;
7882 start += sectorsize;
7885 pgoff += sectorsize;
7886 goto next_block_or_try_again;
7893 static void btrfs_retry_endio(struct bio *bio)
7895 struct btrfs_retry_complete *done = bio->bi_private;
7896 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7897 struct inode *inode;
7898 struct bio_vec *bvec;
7909 start = done->start;
7911 ASSERT(bio->bi_vcnt == 1);
7912 inode = bio->bi_io_vec->bv_page->mapping->host;
7913 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7915 bio_for_each_segment_all(bvec, bio, i) {
7916 ret = __readpage_endio_check(done->inode, io_bio, i,
7917 bvec->bv_page, bvec->bv_offset,
7918 done->start, bvec->bv_len);
7920 clean_io_failure(done->inode, done->start,
7921 bvec->bv_page, bvec->bv_offset);
7926 done->uptodate = uptodate;
7928 complete(&done->done);
7932 static int __btrfs_subio_endio_read(struct inode *inode,
7933 struct btrfs_io_bio *io_bio, int err)
7935 struct btrfs_fs_info *fs_info;
7936 struct bio_vec *bvec;
7937 struct btrfs_retry_complete done;
7947 fs_info = BTRFS_I(inode)->root->fs_info;
7948 sectorsize = BTRFS_I(inode)->root->sectorsize;
7951 start = io_bio->logical;
7954 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7955 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7957 pgoff = bvec->bv_offset;
7959 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7960 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7961 bvec->bv_page, pgoff, start,
7968 init_completion(&done.done);
7970 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7971 pgoff, start, start + sectorsize - 1,
7973 btrfs_retry_endio, &done);
7979 wait_for_completion(&done.done);
7981 if (!done.uptodate) {
7982 /* We might have another mirror, so try again */
7986 offset += sectorsize;
7987 start += sectorsize;
7992 pgoff += sectorsize;
8000 static int btrfs_subio_endio_read(struct inode *inode,
8001 struct btrfs_io_bio *io_bio, int err)
8003 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8007 return __btrfs_correct_data_nocsum(inode, io_bio);
8011 return __btrfs_subio_endio_read(inode, io_bio, err);
8015 static void btrfs_endio_direct_read(struct bio *bio)
8017 struct btrfs_dio_private *dip = bio->bi_private;
8018 struct inode *inode = dip->inode;
8019 struct bio *dio_bio;
8020 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8021 int err = bio->bi_error;
8023 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8024 err = btrfs_subio_endio_read(inode, io_bio, err);
8026 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8027 dip->logical_offset + dip->bytes - 1);
8028 dio_bio = dip->dio_bio;
8032 dio_end_io(dio_bio, bio->bi_error);
8035 io_bio->end_io(io_bio, err);
8039 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8044 struct btrfs_root *root = BTRFS_I(inode)->root;
8045 struct btrfs_ordered_extent *ordered = NULL;
8046 u64 ordered_offset = offset;
8047 u64 ordered_bytes = bytes;
8051 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8058 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8059 finish_ordered_fn, NULL, NULL);
8060 btrfs_queue_work(root->fs_info->endio_write_workers,
8064 * our bio might span multiple ordered extents. If we haven't
8065 * completed the accounting for the whole dio, go back and try again
8067 if (ordered_offset < offset + bytes) {
8068 ordered_bytes = offset + bytes - ordered_offset;
8074 static void btrfs_endio_direct_write(struct bio *bio)
8076 struct btrfs_dio_private *dip = bio->bi_private;
8077 struct bio *dio_bio = dip->dio_bio;
8079 btrfs_endio_direct_write_update_ordered(dip->inode,
8080 dip->logical_offset,
8086 dio_end_io(dio_bio, bio->bi_error);
8090 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8091 struct bio *bio, int mirror_num,
8092 unsigned long bio_flags, u64 offset)
8095 struct btrfs_root *root = BTRFS_I(inode)->root;
8096 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8097 BUG_ON(ret); /* -ENOMEM */
8101 static void btrfs_end_dio_bio(struct bio *bio)
8103 struct btrfs_dio_private *dip = bio->bi_private;
8104 int err = bio->bi_error;
8107 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8108 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8109 btrfs_ino(dip->inode), bio->bi_rw,
8110 (unsigned long long)bio->bi_iter.bi_sector,
8111 bio->bi_iter.bi_size, err);
8113 if (dip->subio_endio)
8114 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8120 * before atomic variable goto zero, we must make sure
8121 * dip->errors is perceived to be set.
8123 smp_mb__before_atomic();
8126 /* if there are more bios still pending for this dio, just exit */
8127 if (!atomic_dec_and_test(&dip->pending_bios))
8131 bio_io_error(dip->orig_bio);
8133 dip->dio_bio->bi_error = 0;
8134 bio_endio(dip->orig_bio);
8140 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8141 u64 first_sector, gfp_t gfp_flags)
8144 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8146 bio_associate_current(bio);
8150 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8151 struct inode *inode,
8152 struct btrfs_dio_private *dip,
8156 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8157 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8161 * We load all the csum data we need when we submit
8162 * the first bio to reduce the csum tree search and
8165 if (dip->logical_offset == file_offset) {
8166 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8172 if (bio == dip->orig_bio)
8175 file_offset -= dip->logical_offset;
8176 file_offset >>= inode->i_sb->s_blocksize_bits;
8177 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8182 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8183 int rw, u64 file_offset, int skip_sum,
8186 struct btrfs_dio_private *dip = bio->bi_private;
8187 int write = rw & REQ_WRITE;
8188 struct btrfs_root *root = BTRFS_I(inode)->root;
8192 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8197 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8198 BTRFS_WQ_ENDIO_DATA);
8206 if (write && async_submit) {
8207 ret = btrfs_wq_submit_bio(root->fs_info,
8208 inode, rw, bio, 0, 0,
8210 __btrfs_submit_bio_start_direct_io,
8211 __btrfs_submit_bio_done);
8215 * If we aren't doing async submit, calculate the csum of the
8218 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8222 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8228 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8234 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8237 struct inode *inode = dip->inode;
8238 struct btrfs_root *root = BTRFS_I(inode)->root;
8240 struct bio *orig_bio = dip->orig_bio;
8241 struct bio_vec *bvec = orig_bio->bi_io_vec;
8242 u64 start_sector = orig_bio->bi_iter.bi_sector;
8243 u64 file_offset = dip->logical_offset;
8248 int async_submit = 0;
8250 map_length = orig_bio->bi_iter.bi_size;
8251 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8252 &map_length, NULL, 0);
8256 if (map_length >= orig_bio->bi_iter.bi_size) {
8258 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8262 /* async crcs make it difficult to collect full stripe writes. */
8263 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8268 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8272 bio->bi_private = dip;
8273 bio->bi_end_io = btrfs_end_dio_bio;
8274 btrfs_io_bio(bio)->logical = file_offset;
8275 atomic_inc(&dip->pending_bios);
8277 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8278 if (map_length < submit_len + bvec->bv_len ||
8279 bio_add_page(bio, bvec->bv_page, bvec->bv_len,
8280 bvec->bv_offset) < bvec->bv_len) {
8282 * inc the count before we submit the bio so
8283 * we know the end IO handler won't happen before
8284 * we inc the count. Otherwise, the dip might get freed
8285 * before we're done setting it up
8287 atomic_inc(&dip->pending_bios);
8288 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8289 file_offset, skip_sum,
8293 atomic_dec(&dip->pending_bios);
8297 start_sector += submit_len >> 9;
8298 file_offset += submit_len;
8303 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8304 start_sector, GFP_NOFS);
8307 bio->bi_private = dip;
8308 bio->bi_end_io = btrfs_end_dio_bio;
8309 btrfs_io_bio(bio)->logical = file_offset;
8311 map_length = orig_bio->bi_iter.bi_size;
8312 ret = btrfs_map_block(root->fs_info, rw,
8314 &map_length, NULL, 0);
8320 submit_len += bvec->bv_len;
8327 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8336 * before atomic variable goto zero, we must
8337 * make sure dip->errors is perceived to be set.
8339 smp_mb__before_atomic();
8340 if (atomic_dec_and_test(&dip->pending_bios))
8341 bio_io_error(dip->orig_bio);
8343 /* bio_end_io() will handle error, so we needn't return it */
8347 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8348 struct inode *inode, loff_t file_offset)
8350 struct btrfs_dio_private *dip = NULL;
8351 struct bio *io_bio = NULL;
8352 struct btrfs_io_bio *btrfs_bio;
8354 int write = rw & REQ_WRITE;
8357 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8359 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8365 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8371 dip->private = dio_bio->bi_private;
8373 dip->logical_offset = file_offset;
8374 dip->bytes = dio_bio->bi_iter.bi_size;
8375 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8376 io_bio->bi_private = dip;
8377 dip->orig_bio = io_bio;
8378 dip->dio_bio = dio_bio;
8379 atomic_set(&dip->pending_bios, 0);
8380 btrfs_bio = btrfs_io_bio(io_bio);
8381 btrfs_bio->logical = file_offset;
8384 io_bio->bi_end_io = btrfs_endio_direct_write;
8386 io_bio->bi_end_io = btrfs_endio_direct_read;
8387 dip->subio_endio = btrfs_subio_endio_read;
8391 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8392 * even if we fail to submit a bio, because in such case we do the
8393 * corresponding error handling below and it must not be done a second
8394 * time by btrfs_direct_IO().
8397 struct btrfs_dio_data *dio_data = current->journal_info;
8399 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8401 dio_data->unsubmitted_oe_range_start =
8402 dio_data->unsubmitted_oe_range_end;
8405 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8409 if (btrfs_bio->end_io)
8410 btrfs_bio->end_io(btrfs_bio, ret);
8414 * If we arrived here it means either we failed to submit the dip
8415 * or we either failed to clone the dio_bio or failed to allocate the
8416 * dip. If we cloned the dio_bio and allocated the dip, we can just
8417 * call bio_endio against our io_bio so that we get proper resource
8418 * cleanup if we fail to submit the dip, otherwise, we must do the
8419 * same as btrfs_endio_direct_[write|read] because we can't call these
8420 * callbacks - they require an allocated dip and a clone of dio_bio.
8422 if (io_bio && dip) {
8423 io_bio->bi_error = -EIO;
8426 * The end io callbacks free our dip, do the final put on io_bio
8427 * and all the cleanup and final put for dio_bio (through
8434 btrfs_endio_direct_write_update_ordered(inode,
8436 dio_bio->bi_iter.bi_size,
8439 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8440 file_offset + dio_bio->bi_iter.bi_size - 1);
8442 dio_bio->bi_error = -EIO;
8444 * Releases and cleans up our dio_bio, no need to bio_put()
8445 * nor bio_endio()/bio_io_error() against dio_bio.
8447 dio_end_io(dio_bio, ret);
8454 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8455 const struct iov_iter *iter, loff_t offset)
8459 unsigned blocksize_mask = root->sectorsize - 1;
8460 ssize_t retval = -EINVAL;
8462 if (offset & blocksize_mask)
8465 if (iov_iter_alignment(iter) & blocksize_mask)
8468 /* If this is a write we don't need to check anymore */
8469 if (iov_iter_rw(iter) == WRITE)
8472 * Check to make sure we don't have duplicate iov_base's in this
8473 * iovec, if so return EINVAL, otherwise we'll get csum errors
8474 * when reading back.
8476 for (seg = 0; seg < iter->nr_segs; seg++) {
8477 for (i = seg + 1; i < iter->nr_segs; i++) {
8478 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8487 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
8490 struct file *file = iocb->ki_filp;
8491 struct inode *inode = file->f_mapping->host;
8492 struct btrfs_root *root = BTRFS_I(inode)->root;
8493 struct btrfs_dio_data dio_data = { 0 };
8497 bool relock = false;
8500 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8503 inode_dio_begin(inode);
8504 smp_mb__after_atomic();
8507 * The generic stuff only does filemap_write_and_wait_range, which
8508 * isn't enough if we've written compressed pages to this area, so
8509 * we need to flush the dirty pages again to make absolutely sure
8510 * that any outstanding dirty pages are on disk.
8512 count = iov_iter_count(iter);
8513 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8514 &BTRFS_I(inode)->runtime_flags))
8515 filemap_fdatawrite_range(inode->i_mapping, offset,
8516 offset + count - 1);
8518 if (iov_iter_rw(iter) == WRITE) {
8520 * If the write DIO is beyond the EOF, we need update
8521 * the isize, but it is protected by i_mutex. So we can
8522 * not unlock the i_mutex at this case.
8524 if (offset + count <= inode->i_size) {
8525 mutex_unlock(&inode->i_mutex);
8528 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8531 dio_data.outstanding_extents = div64_u64(count +
8532 BTRFS_MAX_EXTENT_SIZE - 1,
8533 BTRFS_MAX_EXTENT_SIZE);
8536 * We need to know how many extents we reserved so that we can
8537 * do the accounting properly if we go over the number we
8538 * originally calculated. Abuse current->journal_info for this.
8540 dio_data.reserve = round_up(count, root->sectorsize);
8541 dio_data.unsubmitted_oe_range_start = (u64)offset;
8542 dio_data.unsubmitted_oe_range_end = (u64)offset;
8543 current->journal_info = &dio_data;
8544 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8545 &BTRFS_I(inode)->runtime_flags)) {
8546 inode_dio_end(inode);
8547 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8551 ret = __blockdev_direct_IO(iocb, inode,
8552 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8553 iter, offset, btrfs_get_blocks_direct, NULL,
8554 btrfs_submit_direct, flags);
8555 if (iov_iter_rw(iter) == WRITE) {
8556 current->journal_info = NULL;
8557 if (ret < 0 && ret != -EIOCBQUEUED) {
8558 if (dio_data.reserve)
8559 btrfs_delalloc_release_space(inode, offset,
8562 * On error we might have left some ordered extents
8563 * without submitting corresponding bios for them, so
8564 * cleanup them up to avoid other tasks getting them
8565 * and waiting for them to complete forever.
8567 if (dio_data.unsubmitted_oe_range_start <
8568 dio_data.unsubmitted_oe_range_end)
8569 btrfs_endio_direct_write_update_ordered(inode,
8570 dio_data.unsubmitted_oe_range_start,
8571 dio_data.unsubmitted_oe_range_end -
8572 dio_data.unsubmitted_oe_range_start,
8574 } else if (ret >= 0 && (size_t)ret < count)
8575 btrfs_delalloc_release_space(inode, offset,
8576 count - (size_t)ret);
8580 inode_dio_end(inode);
8582 mutex_lock(&inode->i_mutex);
8587 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8589 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8590 __u64 start, __u64 len)
8594 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8598 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8601 int btrfs_readpage(struct file *file, struct page *page)
8603 struct extent_io_tree *tree;
8604 tree = &BTRFS_I(page->mapping->host)->io_tree;
8605 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8608 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8610 struct extent_io_tree *tree;
8611 struct inode *inode = page->mapping->host;
8614 if (current->flags & PF_MEMALLOC) {
8615 redirty_page_for_writepage(wbc, page);
8621 * If we are under memory pressure we will call this directly from the
8622 * VM, we need to make sure we have the inode referenced for the ordered
8623 * extent. If not just return like we didn't do anything.
8625 if (!igrab(inode)) {
8626 redirty_page_for_writepage(wbc, page);
8627 return AOP_WRITEPAGE_ACTIVATE;
8629 tree = &BTRFS_I(page->mapping->host)->io_tree;
8630 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8631 btrfs_add_delayed_iput(inode);
8635 static int btrfs_writepages(struct address_space *mapping,
8636 struct writeback_control *wbc)
8638 struct extent_io_tree *tree;
8640 tree = &BTRFS_I(mapping->host)->io_tree;
8641 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8645 btrfs_readpages(struct file *file, struct address_space *mapping,
8646 struct list_head *pages, unsigned nr_pages)
8648 struct extent_io_tree *tree;
8649 tree = &BTRFS_I(mapping->host)->io_tree;
8650 return extent_readpages(tree, mapping, pages, nr_pages,
8653 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8655 struct extent_io_tree *tree;
8656 struct extent_map_tree *map;
8659 tree = &BTRFS_I(page->mapping->host)->io_tree;
8660 map = &BTRFS_I(page->mapping->host)->extent_tree;
8661 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8663 ClearPagePrivate(page);
8664 set_page_private(page, 0);
8665 page_cache_release(page);
8670 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8672 if (PageWriteback(page) || PageDirty(page))
8674 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8677 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8678 unsigned int length)
8680 struct inode *inode = page->mapping->host;
8681 struct extent_io_tree *tree;
8682 struct btrfs_ordered_extent *ordered;
8683 struct extent_state *cached_state = NULL;
8684 u64 page_start = page_offset(page);
8685 u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8686 int inode_evicting = inode->i_state & I_FREEING;
8689 * we have the page locked, so new writeback can't start,
8690 * and the dirty bit won't be cleared while we are here.
8692 * Wait for IO on this page so that we can safely clear
8693 * the PagePrivate2 bit and do ordered accounting
8695 wait_on_page_writeback(page);
8697 tree = &BTRFS_I(inode)->io_tree;
8699 btrfs_releasepage(page, GFP_NOFS);
8703 if (!inode_evicting)
8704 lock_extent_bits(tree, page_start, page_end, &cached_state);
8705 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8708 * IO on this page will never be started, so we need
8709 * to account for any ordered extents now
8711 if (!inode_evicting)
8712 clear_extent_bit(tree, page_start, page_end,
8713 EXTENT_DIRTY | EXTENT_DELALLOC |
8714 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8715 EXTENT_DEFRAG, 1, 0, &cached_state,
8718 * whoever cleared the private bit is responsible
8719 * for the finish_ordered_io
8721 if (TestClearPagePrivate2(page)) {
8722 struct btrfs_ordered_inode_tree *tree;
8725 tree = &BTRFS_I(inode)->ordered_tree;
8727 spin_lock_irq(&tree->lock);
8728 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8729 new_len = page_start - ordered->file_offset;
8730 if (new_len < ordered->truncated_len)
8731 ordered->truncated_len = new_len;
8732 spin_unlock_irq(&tree->lock);
8734 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8736 PAGE_CACHE_SIZE, 1))
8737 btrfs_finish_ordered_io(ordered);
8739 btrfs_put_ordered_extent(ordered);
8740 if (!inode_evicting) {
8741 cached_state = NULL;
8742 lock_extent_bits(tree, page_start, page_end,
8748 * Qgroup reserved space handler
8749 * Page here will be either
8750 * 1) Already written to disk
8751 * In this case, its reserved space is released from data rsv map
8752 * and will be freed by delayed_ref handler finally.
8753 * So even we call qgroup_free_data(), it won't decrease reserved
8755 * 2) Not written to disk
8756 * This means the reserved space should be freed here.
8758 btrfs_qgroup_free_data(inode, page_start, PAGE_CACHE_SIZE);
8759 if (!inode_evicting) {
8760 clear_extent_bit(tree, page_start, page_end,
8761 EXTENT_LOCKED | EXTENT_DIRTY |
8762 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8763 EXTENT_DEFRAG, 1, 1,
8764 &cached_state, GFP_NOFS);
8766 __btrfs_releasepage(page, GFP_NOFS);
8769 ClearPageChecked(page);
8770 if (PagePrivate(page)) {
8771 ClearPagePrivate(page);
8772 set_page_private(page, 0);
8773 page_cache_release(page);
8778 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8779 * called from a page fault handler when a page is first dirtied. Hence we must
8780 * be careful to check for EOF conditions here. We set the page up correctly
8781 * for a written page which means we get ENOSPC checking when writing into
8782 * holes and correct delalloc and unwritten extent mapping on filesystems that
8783 * support these features.
8785 * We are not allowed to take the i_mutex here so we have to play games to
8786 * protect against truncate races as the page could now be beyond EOF. Because
8787 * vmtruncate() writes the inode size before removing pages, once we have the
8788 * page lock we can determine safely if the page is beyond EOF. If it is not
8789 * beyond EOF, then the page is guaranteed safe against truncation until we
8792 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8794 struct page *page = vmf->page;
8795 struct inode *inode = file_inode(vma->vm_file);
8796 struct btrfs_root *root = BTRFS_I(inode)->root;
8797 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8798 struct btrfs_ordered_extent *ordered;
8799 struct extent_state *cached_state = NULL;
8801 unsigned long zero_start;
8808 sb_start_pagefault(inode->i_sb);
8809 page_start = page_offset(page);
8810 page_end = page_start + PAGE_CACHE_SIZE - 1;
8812 ret = btrfs_delalloc_reserve_space(inode, page_start,
8815 ret = file_update_time(vma->vm_file);
8821 else /* -ENOSPC, -EIO, etc */
8822 ret = VM_FAULT_SIGBUS;
8828 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8831 size = i_size_read(inode);
8833 if ((page->mapping != inode->i_mapping) ||
8834 (page_start >= size)) {
8835 /* page got truncated out from underneath us */
8838 wait_on_page_writeback(page);
8840 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8841 set_page_extent_mapped(page);
8844 * we can't set the delalloc bits if there are pending ordered
8845 * extents. Drop our locks and wait for them to finish
8847 ordered = btrfs_lookup_ordered_extent(inode, page_start);
8849 unlock_extent_cached(io_tree, page_start, page_end,
8850 &cached_state, GFP_NOFS);
8852 btrfs_start_ordered_extent(inode, ordered, 1);
8853 btrfs_put_ordered_extent(ordered);
8858 * XXX - page_mkwrite gets called every time the page is dirtied, even
8859 * if it was already dirty, so for space accounting reasons we need to
8860 * clear any delalloc bits for the range we are fixing to save. There
8861 * is probably a better way to do this, but for now keep consistent with
8862 * prepare_pages in the normal write path.
8864 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8865 EXTENT_DIRTY | EXTENT_DELALLOC |
8866 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8867 0, 0, &cached_state, GFP_NOFS);
8869 ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8872 unlock_extent_cached(io_tree, page_start, page_end,
8873 &cached_state, GFP_NOFS);
8874 ret = VM_FAULT_SIGBUS;
8879 /* page is wholly or partially inside EOF */
8880 if (page_start + PAGE_CACHE_SIZE > size)
8881 zero_start = size & ~PAGE_CACHE_MASK;
8883 zero_start = PAGE_CACHE_SIZE;
8885 if (zero_start != PAGE_CACHE_SIZE) {
8887 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8888 flush_dcache_page(page);
8891 ClearPageChecked(page);
8892 set_page_dirty(page);
8893 SetPageUptodate(page);
8895 BTRFS_I(inode)->last_trans = root->fs_info->generation;
8896 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8897 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8899 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8903 sb_end_pagefault(inode->i_sb);
8904 return VM_FAULT_LOCKED;
8908 btrfs_delalloc_release_space(inode, page_start, PAGE_CACHE_SIZE);
8910 sb_end_pagefault(inode->i_sb);
8914 static int btrfs_truncate(struct inode *inode)
8916 struct btrfs_root *root = BTRFS_I(inode)->root;
8917 struct btrfs_block_rsv *rsv;
8920 struct btrfs_trans_handle *trans;
8921 u64 mask = root->sectorsize - 1;
8922 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8924 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8930 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
8931 * 3 things going on here
8933 * 1) We need to reserve space for our orphan item and the space to
8934 * delete our orphan item. Lord knows we don't want to have a dangling
8935 * orphan item because we didn't reserve space to remove it.
8937 * 2) We need to reserve space to update our inode.
8939 * 3) We need to have something to cache all the space that is going to
8940 * be free'd up by the truncate operation, but also have some slack
8941 * space reserved in case it uses space during the truncate (thank you
8942 * very much snapshotting).
8944 * And we need these to all be seperate. The fact is we can use alot of
8945 * space doing the truncate, and we have no earthly idea how much space
8946 * we will use, so we need the truncate reservation to be seperate so it
8947 * doesn't end up using space reserved for updating the inode or
8948 * removing the orphan item. We also need to be able to stop the
8949 * transaction and start a new one, which means we need to be able to
8950 * update the inode several times, and we have no idea of knowing how
8951 * many times that will be, so we can't just reserve 1 item for the
8952 * entirety of the opration, so that has to be done seperately as well.
8953 * Then there is the orphan item, which does indeed need to be held on
8954 * to for the whole operation, and we need nobody to touch this reserved
8955 * space except the orphan code.
8957 * So that leaves us with
8959 * 1) root->orphan_block_rsv - for the orphan deletion.
8960 * 2) rsv - for the truncate reservation, which we will steal from the
8961 * transaction reservation.
8962 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8963 * updating the inode.
8965 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8968 rsv->size = min_size;
8972 * 1 for the truncate slack space
8973 * 1 for updating the inode.
8975 trans = btrfs_start_transaction(root, 2);
8976 if (IS_ERR(trans)) {
8977 err = PTR_ERR(trans);
8981 /* Migrate the slack space for the truncate to our reserve */
8982 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8987 * So if we truncate and then write and fsync we normally would just
8988 * write the extents that changed, which is a problem if we need to
8989 * first truncate that entire inode. So set this flag so we write out
8990 * all of the extents in the inode to the sync log so we're completely
8993 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8994 trans->block_rsv = rsv;
8997 ret = btrfs_truncate_inode_items(trans, root, inode,
8999 BTRFS_EXTENT_DATA_KEY);
9000 if (ret != -ENOSPC && ret != -EAGAIN) {
9005 trans->block_rsv = &root->fs_info->trans_block_rsv;
9006 ret = btrfs_update_inode(trans, root, inode);
9012 btrfs_end_transaction(trans, root);
9013 btrfs_btree_balance_dirty(root);
9015 trans = btrfs_start_transaction(root, 2);
9016 if (IS_ERR(trans)) {
9017 ret = err = PTR_ERR(trans);
9022 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9024 BUG_ON(ret); /* shouldn't happen */
9025 trans->block_rsv = rsv;
9028 if (ret == 0 && inode->i_nlink > 0) {
9029 trans->block_rsv = root->orphan_block_rsv;
9030 ret = btrfs_orphan_del(trans, inode);
9036 trans->block_rsv = &root->fs_info->trans_block_rsv;
9037 ret = btrfs_update_inode(trans, root, inode);
9041 ret = btrfs_end_transaction(trans, root);
9042 btrfs_btree_balance_dirty(root);
9046 btrfs_free_block_rsv(root, rsv);
9055 * create a new subvolume directory/inode (helper for the ioctl).
9057 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9058 struct btrfs_root *new_root,
9059 struct btrfs_root *parent_root,
9062 struct inode *inode;
9066 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9067 new_dirid, new_dirid,
9068 S_IFDIR | (~current_umask() & S_IRWXUGO),
9071 return PTR_ERR(inode);
9072 inode->i_op = &btrfs_dir_inode_operations;
9073 inode->i_fop = &btrfs_dir_file_operations;
9075 set_nlink(inode, 1);
9076 btrfs_i_size_write(inode, 0);
9077 unlock_new_inode(inode);
9079 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9081 btrfs_err(new_root->fs_info,
9082 "error inheriting subvolume %llu properties: %d",
9083 new_root->root_key.objectid, err);
9085 err = btrfs_update_inode(trans, new_root, inode);
9091 struct inode *btrfs_alloc_inode(struct super_block *sb)
9093 struct btrfs_inode *ei;
9094 struct inode *inode;
9096 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9103 ei->last_sub_trans = 0;
9104 ei->logged_trans = 0;
9105 ei->delalloc_bytes = 0;
9106 ei->defrag_bytes = 0;
9107 ei->disk_i_size = 0;
9110 ei->index_cnt = (u64)-1;
9112 ei->last_unlink_trans = 0;
9113 ei->last_log_commit = 0;
9114 ei->delayed_iput_count = 0;
9116 spin_lock_init(&ei->lock);
9117 ei->outstanding_extents = 0;
9118 ei->reserved_extents = 0;
9120 ei->runtime_flags = 0;
9121 ei->force_compress = BTRFS_COMPRESS_NONE;
9123 ei->delayed_node = NULL;
9125 ei->i_otime.tv_sec = 0;
9126 ei->i_otime.tv_nsec = 0;
9128 inode = &ei->vfs_inode;
9129 extent_map_tree_init(&ei->extent_tree);
9130 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9131 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9132 ei->io_tree.track_uptodate = 1;
9133 ei->io_failure_tree.track_uptodate = 1;
9134 atomic_set(&ei->sync_writers, 0);
9135 mutex_init(&ei->log_mutex);
9136 mutex_init(&ei->delalloc_mutex);
9137 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9138 INIT_LIST_HEAD(&ei->delalloc_inodes);
9139 INIT_LIST_HEAD(&ei->delayed_iput);
9140 RB_CLEAR_NODE(&ei->rb_node);
9145 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9146 void btrfs_test_destroy_inode(struct inode *inode)
9148 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9149 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9153 static void btrfs_i_callback(struct rcu_head *head)
9155 struct inode *inode = container_of(head, struct inode, i_rcu);
9156 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9159 void btrfs_destroy_inode(struct inode *inode)
9161 struct btrfs_ordered_extent *ordered;
9162 struct btrfs_root *root = BTRFS_I(inode)->root;
9164 WARN_ON(!hlist_empty(&inode->i_dentry));
9165 WARN_ON(inode->i_data.nrpages);
9166 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9167 WARN_ON(BTRFS_I(inode)->reserved_extents);
9168 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9169 WARN_ON(BTRFS_I(inode)->csum_bytes);
9170 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9173 * This can happen where we create an inode, but somebody else also
9174 * created the same inode and we need to destroy the one we already
9180 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9181 &BTRFS_I(inode)->runtime_flags)) {
9182 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9184 atomic_dec(&root->orphan_inodes);
9188 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9192 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9193 ordered->file_offset, ordered->len);
9194 btrfs_remove_ordered_extent(inode, ordered);
9195 btrfs_put_ordered_extent(ordered);
9196 btrfs_put_ordered_extent(ordered);
9199 btrfs_qgroup_check_reserved_leak(inode);
9200 inode_tree_del(inode);
9201 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9203 call_rcu(&inode->i_rcu, btrfs_i_callback);
9206 int btrfs_drop_inode(struct inode *inode)
9208 struct btrfs_root *root = BTRFS_I(inode)->root;
9213 /* the snap/subvol tree is on deleting */
9214 if (btrfs_root_refs(&root->root_item) == 0)
9217 return generic_drop_inode(inode);
9220 static void init_once(void *foo)
9222 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9224 inode_init_once(&ei->vfs_inode);
9227 void btrfs_destroy_cachep(void)
9230 * Make sure all delayed rcu free inodes are flushed before we
9234 if (btrfs_inode_cachep)
9235 kmem_cache_destroy(btrfs_inode_cachep);
9236 if (btrfs_trans_handle_cachep)
9237 kmem_cache_destroy(btrfs_trans_handle_cachep);
9238 if (btrfs_transaction_cachep)
9239 kmem_cache_destroy(btrfs_transaction_cachep);
9240 if (btrfs_path_cachep)
9241 kmem_cache_destroy(btrfs_path_cachep);
9242 if (btrfs_free_space_cachep)
9243 kmem_cache_destroy(btrfs_free_space_cachep);
9246 int btrfs_init_cachep(void)
9248 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9249 sizeof(struct btrfs_inode), 0,
9250 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
9251 if (!btrfs_inode_cachep)
9254 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9255 sizeof(struct btrfs_trans_handle), 0,
9256 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9257 if (!btrfs_trans_handle_cachep)
9260 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9261 sizeof(struct btrfs_transaction), 0,
9262 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9263 if (!btrfs_transaction_cachep)
9266 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9267 sizeof(struct btrfs_path), 0,
9268 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9269 if (!btrfs_path_cachep)
9272 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9273 sizeof(struct btrfs_free_space), 0,
9274 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9275 if (!btrfs_free_space_cachep)
9280 btrfs_destroy_cachep();
9284 static int btrfs_getattr(struct vfsmount *mnt,
9285 struct dentry *dentry, struct kstat *stat)
9288 struct inode *inode = d_inode(dentry);
9289 u32 blocksize = inode->i_sb->s_blocksize;
9291 generic_fillattr(inode, stat);
9292 stat->dev = BTRFS_I(inode)->root->anon_dev;
9293 stat->blksize = PAGE_CACHE_SIZE;
9295 spin_lock(&BTRFS_I(inode)->lock);
9296 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9297 spin_unlock(&BTRFS_I(inode)->lock);
9298 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9299 ALIGN(delalloc_bytes, blocksize)) >> 9;
9303 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9304 struct inode *new_dir, struct dentry *new_dentry)
9306 struct btrfs_trans_handle *trans;
9307 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9308 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9309 struct inode *new_inode = d_inode(new_dentry);
9310 struct inode *old_inode = d_inode(old_dentry);
9311 struct timespec ctime = CURRENT_TIME;
9315 u64 old_ino = btrfs_ino(old_inode);
9317 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9320 /* we only allow rename subvolume link between subvolumes */
9321 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9324 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9325 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9328 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9329 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9333 /* check for collisions, even if the name isn't there */
9334 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9335 new_dentry->d_name.name,
9336 new_dentry->d_name.len);
9339 if (ret == -EEXIST) {
9341 * eexist without a new_inode */
9342 if (WARN_ON(!new_inode)) {
9346 /* maybe -EOVERFLOW */
9353 * we're using rename to replace one file with another. Start IO on it
9354 * now so we don't add too much work to the end of the transaction
9356 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9357 filemap_flush(old_inode->i_mapping);
9359 /* close the racy window with snapshot create/destroy ioctl */
9360 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9361 down_read(&root->fs_info->subvol_sem);
9363 * We want to reserve the absolute worst case amount of items. So if
9364 * both inodes are subvols and we need to unlink them then that would
9365 * require 4 item modifications, but if they are both normal inodes it
9366 * would require 5 item modifications, so we'll assume their normal
9367 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9368 * should cover the worst case number of items we'll modify.
9370 trans = btrfs_start_transaction(root, 11);
9371 if (IS_ERR(trans)) {
9372 ret = PTR_ERR(trans);
9377 btrfs_record_root_in_trans(trans, dest);
9379 ret = btrfs_set_inode_index(new_dir, &index);
9383 BTRFS_I(old_inode)->dir_index = 0ULL;
9384 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9385 /* force full log commit if subvolume involved. */
9386 btrfs_set_log_full_commit(root->fs_info, trans);
9388 ret = btrfs_insert_inode_ref(trans, dest,
9389 new_dentry->d_name.name,
9390 new_dentry->d_name.len,
9392 btrfs_ino(new_dir), index);
9396 * this is an ugly little race, but the rename is required
9397 * to make sure that if we crash, the inode is either at the
9398 * old name or the new one. pinning the log transaction lets
9399 * us make sure we don't allow a log commit to come in after
9400 * we unlink the name but before we add the new name back in.
9402 btrfs_pin_log_trans(root);
9405 inode_inc_iversion(old_dir);
9406 inode_inc_iversion(new_dir);
9407 inode_inc_iversion(old_inode);
9408 old_dir->i_ctime = old_dir->i_mtime = ctime;
9409 new_dir->i_ctime = new_dir->i_mtime = ctime;
9410 old_inode->i_ctime = ctime;
9412 if (old_dentry->d_parent != new_dentry->d_parent)
9413 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9415 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9416 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9417 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9418 old_dentry->d_name.name,
9419 old_dentry->d_name.len);
9421 ret = __btrfs_unlink_inode(trans, root, old_dir,
9422 d_inode(old_dentry),
9423 old_dentry->d_name.name,
9424 old_dentry->d_name.len);
9426 ret = btrfs_update_inode(trans, root, old_inode);
9429 btrfs_abort_transaction(trans, root, ret);
9434 inode_inc_iversion(new_inode);
9435 new_inode->i_ctime = CURRENT_TIME;
9436 if (unlikely(btrfs_ino(new_inode) ==
9437 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9438 root_objectid = BTRFS_I(new_inode)->location.objectid;
9439 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9441 new_dentry->d_name.name,
9442 new_dentry->d_name.len);
9443 BUG_ON(new_inode->i_nlink == 0);
9445 ret = btrfs_unlink_inode(trans, dest, new_dir,
9446 d_inode(new_dentry),
9447 new_dentry->d_name.name,
9448 new_dentry->d_name.len);
9450 if (!ret && new_inode->i_nlink == 0)
9451 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9453 btrfs_abort_transaction(trans, root, ret);
9458 ret = btrfs_add_link(trans, new_dir, old_inode,
9459 new_dentry->d_name.name,
9460 new_dentry->d_name.len, 0, index);
9462 btrfs_abort_transaction(trans, root, ret);
9466 if (old_inode->i_nlink == 1)
9467 BTRFS_I(old_inode)->dir_index = index;
9469 if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9470 struct dentry *parent = new_dentry->d_parent;
9471 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9472 btrfs_end_log_trans(root);
9475 btrfs_end_transaction(trans, root);
9477 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9478 up_read(&root->fs_info->subvol_sem);
9483 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9484 struct inode *new_dir, struct dentry *new_dentry,
9487 if (flags & ~RENAME_NOREPLACE)
9490 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9493 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9495 struct btrfs_delalloc_work *delalloc_work;
9496 struct inode *inode;
9498 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9500 inode = delalloc_work->inode;
9501 filemap_flush(inode->i_mapping);
9502 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9503 &BTRFS_I(inode)->runtime_flags))
9504 filemap_flush(inode->i_mapping);
9506 if (delalloc_work->delay_iput)
9507 btrfs_add_delayed_iput(inode);
9510 complete(&delalloc_work->completion);
9513 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9516 struct btrfs_delalloc_work *work;
9518 work = kmalloc(sizeof(*work), GFP_NOFS);
9522 init_completion(&work->completion);
9523 INIT_LIST_HEAD(&work->list);
9524 work->inode = inode;
9525 work->delay_iput = delay_iput;
9526 WARN_ON_ONCE(!inode);
9527 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9528 btrfs_run_delalloc_work, NULL, NULL);
9533 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9535 wait_for_completion(&work->completion);
9540 * some fairly slow code that needs optimization. This walks the list
9541 * of all the inodes with pending delalloc and forces them to disk.
9543 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9546 struct btrfs_inode *binode;
9547 struct inode *inode;
9548 struct btrfs_delalloc_work *work, *next;
9549 struct list_head works;
9550 struct list_head splice;
9553 INIT_LIST_HEAD(&works);
9554 INIT_LIST_HEAD(&splice);
9556 mutex_lock(&root->delalloc_mutex);
9557 spin_lock(&root->delalloc_lock);
9558 list_splice_init(&root->delalloc_inodes, &splice);
9559 while (!list_empty(&splice)) {
9560 binode = list_entry(splice.next, struct btrfs_inode,
9563 list_move_tail(&binode->delalloc_inodes,
9564 &root->delalloc_inodes);
9565 inode = igrab(&binode->vfs_inode);
9567 cond_resched_lock(&root->delalloc_lock);
9570 spin_unlock(&root->delalloc_lock);
9572 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9575 btrfs_add_delayed_iput(inode);
9581 list_add_tail(&work->list, &works);
9582 btrfs_queue_work(root->fs_info->flush_workers,
9585 if (nr != -1 && ret >= nr)
9588 spin_lock(&root->delalloc_lock);
9590 spin_unlock(&root->delalloc_lock);
9593 list_for_each_entry_safe(work, next, &works, list) {
9594 list_del_init(&work->list);
9595 btrfs_wait_and_free_delalloc_work(work);
9598 if (!list_empty_careful(&splice)) {
9599 spin_lock(&root->delalloc_lock);
9600 list_splice_tail(&splice, &root->delalloc_inodes);
9601 spin_unlock(&root->delalloc_lock);
9603 mutex_unlock(&root->delalloc_mutex);
9607 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9611 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9614 ret = __start_delalloc_inodes(root, delay_iput, -1);
9618 * the filemap_flush will queue IO into the worker threads, but
9619 * we have to make sure the IO is actually started and that
9620 * ordered extents get created before we return
9622 atomic_inc(&root->fs_info->async_submit_draining);
9623 while (atomic_read(&root->fs_info->nr_async_submits) ||
9624 atomic_read(&root->fs_info->async_delalloc_pages)) {
9625 wait_event(root->fs_info->async_submit_wait,
9626 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9627 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9629 atomic_dec(&root->fs_info->async_submit_draining);
9633 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9636 struct btrfs_root *root;
9637 struct list_head splice;
9640 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9643 INIT_LIST_HEAD(&splice);
9645 mutex_lock(&fs_info->delalloc_root_mutex);
9646 spin_lock(&fs_info->delalloc_root_lock);
9647 list_splice_init(&fs_info->delalloc_roots, &splice);
9648 while (!list_empty(&splice) && nr) {
9649 root = list_first_entry(&splice, struct btrfs_root,
9651 root = btrfs_grab_fs_root(root);
9653 list_move_tail(&root->delalloc_root,
9654 &fs_info->delalloc_roots);
9655 spin_unlock(&fs_info->delalloc_root_lock);
9657 ret = __start_delalloc_inodes(root, delay_iput, nr);
9658 btrfs_put_fs_root(root);
9666 spin_lock(&fs_info->delalloc_root_lock);
9668 spin_unlock(&fs_info->delalloc_root_lock);
9671 atomic_inc(&fs_info->async_submit_draining);
9672 while (atomic_read(&fs_info->nr_async_submits) ||
9673 atomic_read(&fs_info->async_delalloc_pages)) {
9674 wait_event(fs_info->async_submit_wait,
9675 (atomic_read(&fs_info->nr_async_submits) == 0 &&
9676 atomic_read(&fs_info->async_delalloc_pages) == 0));
9678 atomic_dec(&fs_info->async_submit_draining);
9680 if (!list_empty_careful(&splice)) {
9681 spin_lock(&fs_info->delalloc_root_lock);
9682 list_splice_tail(&splice, &fs_info->delalloc_roots);
9683 spin_unlock(&fs_info->delalloc_root_lock);
9685 mutex_unlock(&fs_info->delalloc_root_mutex);
9689 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9690 const char *symname)
9692 struct btrfs_trans_handle *trans;
9693 struct btrfs_root *root = BTRFS_I(dir)->root;
9694 struct btrfs_path *path;
9695 struct btrfs_key key;
9696 struct inode *inode = NULL;
9704 struct btrfs_file_extent_item *ei;
9705 struct extent_buffer *leaf;
9707 name_len = strlen(symname);
9708 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9709 return -ENAMETOOLONG;
9712 * 2 items for inode item and ref
9713 * 2 items for dir items
9714 * 1 item for updating parent inode item
9715 * 1 item for the inline extent item
9716 * 1 item for xattr if selinux is on
9718 trans = btrfs_start_transaction(root, 7);
9720 return PTR_ERR(trans);
9722 err = btrfs_find_free_ino(root, &objectid);
9726 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9727 dentry->d_name.len, btrfs_ino(dir), objectid,
9728 S_IFLNK|S_IRWXUGO, &index);
9729 if (IS_ERR(inode)) {
9730 err = PTR_ERR(inode);
9735 * If the active LSM wants to access the inode during
9736 * d_instantiate it needs these. Smack checks to see
9737 * if the filesystem supports xattrs by looking at the
9740 inode->i_fop = &btrfs_file_operations;
9741 inode->i_op = &btrfs_file_inode_operations;
9742 inode->i_mapping->a_ops = &btrfs_aops;
9743 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9745 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9747 goto out_unlock_inode;
9749 path = btrfs_alloc_path();
9752 goto out_unlock_inode;
9754 key.objectid = btrfs_ino(inode);
9756 key.type = BTRFS_EXTENT_DATA_KEY;
9757 datasize = btrfs_file_extent_calc_inline_size(name_len);
9758 err = btrfs_insert_empty_item(trans, root, path, &key,
9761 btrfs_free_path(path);
9762 goto out_unlock_inode;
9764 leaf = path->nodes[0];
9765 ei = btrfs_item_ptr(leaf, path->slots[0],
9766 struct btrfs_file_extent_item);
9767 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9768 btrfs_set_file_extent_type(leaf, ei,
9769 BTRFS_FILE_EXTENT_INLINE);
9770 btrfs_set_file_extent_encryption(leaf, ei, 0);
9771 btrfs_set_file_extent_compression(leaf, ei, 0);
9772 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9773 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9775 ptr = btrfs_file_extent_inline_start(ei);
9776 write_extent_buffer(leaf, symname, ptr, name_len);
9777 btrfs_mark_buffer_dirty(leaf);
9778 btrfs_free_path(path);
9780 inode->i_op = &btrfs_symlink_inode_operations;
9781 inode->i_mapping->a_ops = &btrfs_symlink_aops;
9782 inode_set_bytes(inode, name_len);
9783 btrfs_i_size_write(inode, name_len);
9784 err = btrfs_update_inode(trans, root, inode);
9786 * Last step, add directory indexes for our symlink inode. This is the
9787 * last step to avoid extra cleanup of these indexes if an error happens
9791 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9794 goto out_unlock_inode;
9797 unlock_new_inode(inode);
9798 d_instantiate(dentry, inode);
9801 btrfs_end_transaction(trans, root);
9803 inode_dec_link_count(inode);
9806 btrfs_btree_balance_dirty(root);
9811 unlock_new_inode(inode);
9815 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9816 u64 start, u64 num_bytes, u64 min_size,
9817 loff_t actual_len, u64 *alloc_hint,
9818 struct btrfs_trans_handle *trans)
9820 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9821 struct extent_map *em;
9822 struct btrfs_root *root = BTRFS_I(inode)->root;
9823 struct btrfs_key ins;
9824 u64 cur_offset = start;
9827 u64 last_alloc = (u64)-1;
9829 bool own_trans = true;
9833 while (num_bytes > 0) {
9835 trans = btrfs_start_transaction(root, 3);
9836 if (IS_ERR(trans)) {
9837 ret = PTR_ERR(trans);
9842 cur_bytes = min_t(u64, num_bytes, SZ_256M);
9843 cur_bytes = max(cur_bytes, min_size);
9845 * If we are severely fragmented we could end up with really
9846 * small allocations, so if the allocator is returning small
9847 * chunks lets make its job easier by only searching for those
9850 cur_bytes = min(cur_bytes, last_alloc);
9851 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9852 *alloc_hint, &ins, 1, 0);
9855 btrfs_end_transaction(trans, root);
9859 last_alloc = ins.offset;
9860 ret = insert_reserved_file_extent(trans, inode,
9861 cur_offset, ins.objectid,
9862 ins.offset, ins.offset,
9863 ins.offset, 0, 0, 0,
9864 BTRFS_FILE_EXTENT_PREALLOC);
9866 btrfs_free_reserved_extent(root, ins.objectid,
9868 btrfs_abort_transaction(trans, root, ret);
9870 btrfs_end_transaction(trans, root);
9874 btrfs_drop_extent_cache(inode, cur_offset,
9875 cur_offset + ins.offset -1, 0);
9877 em = alloc_extent_map();
9879 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9880 &BTRFS_I(inode)->runtime_flags);
9884 em->start = cur_offset;
9885 em->orig_start = cur_offset;
9886 em->len = ins.offset;
9887 em->block_start = ins.objectid;
9888 em->block_len = ins.offset;
9889 em->orig_block_len = ins.offset;
9890 em->ram_bytes = ins.offset;
9891 em->bdev = root->fs_info->fs_devices->latest_bdev;
9892 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9893 em->generation = trans->transid;
9896 write_lock(&em_tree->lock);
9897 ret = add_extent_mapping(em_tree, em, 1);
9898 write_unlock(&em_tree->lock);
9901 btrfs_drop_extent_cache(inode, cur_offset,
9902 cur_offset + ins.offset - 1,
9905 free_extent_map(em);
9907 num_bytes -= ins.offset;
9908 cur_offset += ins.offset;
9909 *alloc_hint = ins.objectid + ins.offset;
9911 inode_inc_iversion(inode);
9912 inode->i_ctime = CURRENT_TIME;
9913 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9914 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9915 (actual_len > inode->i_size) &&
9916 (cur_offset > inode->i_size)) {
9917 if (cur_offset > actual_len)
9918 i_size = actual_len;
9920 i_size = cur_offset;
9921 i_size_write(inode, i_size);
9922 btrfs_ordered_update_i_size(inode, i_size, NULL);
9925 ret = btrfs_update_inode(trans, root, inode);
9928 btrfs_abort_transaction(trans, root, ret);
9930 btrfs_end_transaction(trans, root);
9935 btrfs_end_transaction(trans, root);
9940 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9941 u64 start, u64 num_bytes, u64 min_size,
9942 loff_t actual_len, u64 *alloc_hint)
9944 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9945 min_size, actual_len, alloc_hint,
9949 int btrfs_prealloc_file_range_trans(struct inode *inode,
9950 struct btrfs_trans_handle *trans, int mode,
9951 u64 start, u64 num_bytes, u64 min_size,
9952 loff_t actual_len, u64 *alloc_hint)
9954 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9955 min_size, actual_len, alloc_hint, trans);
9958 static int btrfs_set_page_dirty(struct page *page)
9960 return __set_page_dirty_nobuffers(page);
9963 static int btrfs_permission(struct inode *inode, int mask)
9965 struct btrfs_root *root = BTRFS_I(inode)->root;
9966 umode_t mode = inode->i_mode;
9968 if (mask & MAY_WRITE &&
9969 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9970 if (btrfs_root_readonly(root))
9972 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9975 return generic_permission(inode, mask);
9978 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9980 struct btrfs_trans_handle *trans;
9981 struct btrfs_root *root = BTRFS_I(dir)->root;
9982 struct inode *inode = NULL;
9988 * 5 units required for adding orphan entry
9990 trans = btrfs_start_transaction(root, 5);
9992 return PTR_ERR(trans);
9994 ret = btrfs_find_free_ino(root, &objectid);
9998 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9999 btrfs_ino(dir), objectid, mode, &index);
10000 if (IS_ERR(inode)) {
10001 ret = PTR_ERR(inode);
10006 inode->i_fop = &btrfs_file_operations;
10007 inode->i_op = &btrfs_file_inode_operations;
10009 inode->i_mapping->a_ops = &btrfs_aops;
10010 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10012 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10016 ret = btrfs_update_inode(trans, root, inode);
10019 ret = btrfs_orphan_add(trans, inode);
10024 * We set number of links to 0 in btrfs_new_inode(), and here we set
10025 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10028 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10030 set_nlink(inode, 1);
10031 unlock_new_inode(inode);
10032 d_tmpfile(dentry, inode);
10033 mark_inode_dirty(inode);
10036 btrfs_end_transaction(trans, root);
10039 btrfs_balance_delayed_items(root);
10040 btrfs_btree_balance_dirty(root);
10044 unlock_new_inode(inode);
10049 /* Inspired by filemap_check_errors() */
10050 int btrfs_inode_check_errors(struct inode *inode)
10054 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10055 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10057 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10058 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10064 static const struct inode_operations btrfs_dir_inode_operations = {
10065 .getattr = btrfs_getattr,
10066 .lookup = btrfs_lookup,
10067 .create = btrfs_create,
10068 .unlink = btrfs_unlink,
10069 .link = btrfs_link,
10070 .mkdir = btrfs_mkdir,
10071 .rmdir = btrfs_rmdir,
10072 .rename2 = btrfs_rename2,
10073 .symlink = btrfs_symlink,
10074 .setattr = btrfs_setattr,
10075 .mknod = btrfs_mknod,
10076 .setxattr = btrfs_setxattr,
10077 .getxattr = btrfs_getxattr,
10078 .listxattr = btrfs_listxattr,
10079 .removexattr = btrfs_removexattr,
10080 .permission = btrfs_permission,
10081 .get_acl = btrfs_get_acl,
10082 .set_acl = btrfs_set_acl,
10083 .update_time = btrfs_update_time,
10084 .tmpfile = btrfs_tmpfile,
10086 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10087 .lookup = btrfs_lookup,
10088 .permission = btrfs_permission,
10089 .get_acl = btrfs_get_acl,
10090 .set_acl = btrfs_set_acl,
10091 .update_time = btrfs_update_time,
10094 static const struct file_operations btrfs_dir_file_operations = {
10095 .llseek = generic_file_llseek,
10096 .read = generic_read_dir,
10097 .iterate = btrfs_real_readdir,
10098 .unlocked_ioctl = btrfs_ioctl,
10099 #ifdef CONFIG_COMPAT
10100 .compat_ioctl = btrfs_ioctl,
10102 .release = btrfs_release_file,
10103 .fsync = btrfs_sync_file,
10106 static const struct extent_io_ops btrfs_extent_io_ops = {
10107 .fill_delalloc = run_delalloc_range,
10108 .submit_bio_hook = btrfs_submit_bio_hook,
10109 .merge_bio_hook = btrfs_merge_bio_hook,
10110 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10111 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10112 .writepage_start_hook = btrfs_writepage_start_hook,
10113 .set_bit_hook = btrfs_set_bit_hook,
10114 .clear_bit_hook = btrfs_clear_bit_hook,
10115 .merge_extent_hook = btrfs_merge_extent_hook,
10116 .split_extent_hook = btrfs_split_extent_hook,
10120 * btrfs doesn't support the bmap operation because swapfiles
10121 * use bmap to make a mapping of extents in the file. They assume
10122 * these extents won't change over the life of the file and they
10123 * use the bmap result to do IO directly to the drive.
10125 * the btrfs bmap call would return logical addresses that aren't
10126 * suitable for IO and they also will change frequently as COW
10127 * operations happen. So, swapfile + btrfs == corruption.
10129 * For now we're avoiding this by dropping bmap.
10131 static const struct address_space_operations btrfs_aops = {
10132 .readpage = btrfs_readpage,
10133 .writepage = btrfs_writepage,
10134 .writepages = btrfs_writepages,
10135 .readpages = btrfs_readpages,
10136 .direct_IO = btrfs_direct_IO,
10137 .invalidatepage = btrfs_invalidatepage,
10138 .releasepage = btrfs_releasepage,
10139 .set_page_dirty = btrfs_set_page_dirty,
10140 .error_remove_page = generic_error_remove_page,
10143 static const struct address_space_operations btrfs_symlink_aops = {
10144 .readpage = btrfs_readpage,
10145 .writepage = btrfs_writepage,
10146 .invalidatepage = btrfs_invalidatepage,
10147 .releasepage = btrfs_releasepage,
10150 static const struct inode_operations btrfs_file_inode_operations = {
10151 .getattr = btrfs_getattr,
10152 .setattr = btrfs_setattr,
10153 .setxattr = btrfs_setxattr,
10154 .getxattr = btrfs_getxattr,
10155 .listxattr = btrfs_listxattr,
10156 .removexattr = btrfs_removexattr,
10157 .permission = btrfs_permission,
10158 .fiemap = btrfs_fiemap,
10159 .get_acl = btrfs_get_acl,
10160 .set_acl = btrfs_set_acl,
10161 .update_time = btrfs_update_time,
10163 static const struct inode_operations btrfs_special_inode_operations = {
10164 .getattr = btrfs_getattr,
10165 .setattr = btrfs_setattr,
10166 .permission = btrfs_permission,
10167 .setxattr = btrfs_setxattr,
10168 .getxattr = btrfs_getxattr,
10169 .listxattr = btrfs_listxattr,
10170 .removexattr = btrfs_removexattr,
10171 .get_acl = btrfs_get_acl,
10172 .set_acl = btrfs_set_acl,
10173 .update_time = btrfs_update_time,
10175 static const struct inode_operations btrfs_symlink_inode_operations = {
10176 .readlink = generic_readlink,
10177 .follow_link = page_follow_link_light,
10178 .put_link = page_put_link,
10179 .getattr = btrfs_getattr,
10180 .setattr = btrfs_setattr,
10181 .permission = btrfs_permission,
10182 .setxattr = btrfs_setxattr,
10183 .getxattr = btrfs_getxattr,
10184 .listxattr = btrfs_listxattr,
10185 .removexattr = btrfs_removexattr,
10186 .update_time = btrfs_update_time,
10189 const struct dentry_operations btrfs_dentry_operations = {
10190 .d_delete = btrfs_dentry_delete,
10191 .d_release = btrfs_dentry_release,