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_SHIFT);
212 btrfs_set_file_extent_compression(leaf, ei, 0);
213 kaddr = kmap_atomic(page);
214 offset = start & (PAGE_SIZE - 1);
215 write_extent_buffer(leaf, kaddr + offset, ptr, size);
216 kunmap_atomic(kaddr);
219 btrfs_mark_buffer_dirty(leaf);
220 btrfs_release_path(path);
223 * we're an inline extent, so nobody can
224 * extend the file past i_size without locking
225 * a page we already have locked.
227 * We must do any isize and inode updates
228 * before we unlock the pages. Otherwise we
229 * could end up racing with unlink.
231 BTRFS_I(inode)->disk_i_size = inode->i_size;
232 ret = btrfs_update_inode(trans, root, inode);
241 * conditionally insert an inline extent into the file. This
242 * does the checks required to make sure the data is small enough
243 * to fit as an inline extent.
245 static noinline int cow_file_range_inline(struct btrfs_root *root,
246 struct inode *inode, u64 start,
247 u64 end, size_t compressed_size,
249 struct page **compressed_pages)
251 struct btrfs_trans_handle *trans;
252 u64 isize = i_size_read(inode);
253 u64 actual_end = min(end + 1, isize);
254 u64 inline_len = actual_end - start;
255 u64 aligned_end = ALIGN(end, root->sectorsize);
256 u64 data_len = inline_len;
258 struct btrfs_path *path;
259 int extent_inserted = 0;
260 u32 extent_item_size;
263 data_len = compressed_size;
266 actual_end > root->sectorsize ||
267 data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
269 (actual_end & (root->sectorsize - 1)) == 0) ||
271 data_len > root->fs_info->max_inline) {
275 path = btrfs_alloc_path();
279 trans = btrfs_join_transaction(root);
281 btrfs_free_path(path);
282 return PTR_ERR(trans);
284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
286 if (compressed_size && compressed_pages)
287 extent_item_size = btrfs_file_extent_calc_inline_size(
290 extent_item_size = btrfs_file_extent_calc_inline_size(
293 ret = __btrfs_drop_extents(trans, root, inode, path,
294 start, aligned_end, NULL,
295 1, 1, extent_item_size, &extent_inserted);
297 btrfs_abort_transaction(trans, root, ret);
301 if (isize > actual_end)
302 inline_len = min_t(u64, isize, actual_end);
303 ret = insert_inline_extent(trans, path, extent_inserted,
305 inline_len, compressed_size,
306 compress_type, compressed_pages);
307 if (ret && ret != -ENOSPC) {
308 btrfs_abort_transaction(trans, root, ret);
310 } else if (ret == -ENOSPC) {
315 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
316 btrfs_delalloc_release_metadata(inode, end + 1 - start);
317 btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
320 * Don't forget to free the reserved space, as for inlined extent
321 * it won't count as data extent, free them directly here.
322 * And at reserve time, it's always aligned to page size, so
323 * just free one page here.
325 btrfs_qgroup_free_data(inode, 0, PAGE_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_SHIFT) - (start >> PAGE_SHIFT) + 1;
439 nr_pages = min_t(unsigned long, nr_pages, SZ_128K / PAGE_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 doesn'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 &
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,
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_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);
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);
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 put_page(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;
827 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
830 * clear dirty, set writeback and unlock the pages.
832 extent_clear_unlock_delalloc(inode, async_extent->start,
833 async_extent->start +
834 async_extent->ram_size - 1,
835 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
836 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
838 ret = btrfs_submit_compressed_write(inode,
840 async_extent->ram_size,
842 ins.offset, async_extent->pages,
843 async_extent->nr_pages);
845 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
846 struct page *p = async_extent->pages[0];
847 const u64 start = async_extent->start;
848 const u64 end = start + async_extent->ram_size - 1;
850 p->mapping = inode->i_mapping;
851 tree->ops->writepage_end_io_hook(p, start, end,
854 extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
857 free_async_extent_pages(async_extent);
859 alloc_hint = ins.objectid + ins.offset;
865 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
866 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
868 extent_clear_unlock_delalloc(inode, async_extent->start,
869 async_extent->start +
870 async_extent->ram_size - 1,
871 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
872 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
873 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
874 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
876 free_async_extent_pages(async_extent);
881 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
884 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
885 struct extent_map *em;
888 read_lock(&em_tree->lock);
889 em = search_extent_mapping(em_tree, start, num_bytes);
892 * if block start isn't an actual block number then find the
893 * first block in this inode and use that as a hint. If that
894 * block is also bogus then just don't worry about it.
896 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
898 em = search_extent_mapping(em_tree, 0, 0);
899 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
900 alloc_hint = em->block_start;
904 alloc_hint = em->block_start;
908 read_unlock(&em_tree->lock);
914 * when extent_io.c finds a delayed allocation range in the file,
915 * the call backs end up in this code. The basic idea is to
916 * allocate extents on disk for the range, and create ordered data structs
917 * in ram to track those extents.
919 * locked_page is the page that writepage had locked already. We use
920 * it to make sure we don't do extra locks or unlocks.
922 * *page_started is set to one if we unlock locked_page and do everything
923 * required to start IO on it. It may be clean and already done with
926 static noinline int cow_file_range(struct inode *inode,
927 struct page *locked_page,
928 u64 start, u64 end, int *page_started,
929 unsigned long *nr_written,
932 struct btrfs_root *root = BTRFS_I(inode)->root;
935 unsigned long ram_size;
938 u64 blocksize = root->sectorsize;
939 struct btrfs_key ins;
940 struct extent_map *em;
941 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
944 if (btrfs_is_free_space_inode(inode)) {
950 num_bytes = ALIGN(end - start + 1, blocksize);
951 num_bytes = max(blocksize, num_bytes);
952 disk_num_bytes = num_bytes;
954 /* if this is a small write inside eof, kick off defrag */
955 if (num_bytes < SZ_64K &&
956 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
957 btrfs_add_inode_defrag(NULL, inode);
960 /* lets try to make an inline extent */
961 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
964 extent_clear_unlock_delalloc(inode, start, end, NULL,
965 EXTENT_LOCKED | EXTENT_DELALLOC |
966 EXTENT_DEFRAG, PAGE_UNLOCK |
967 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
970 *nr_written = *nr_written +
971 (end - start + PAGE_SIZE) / PAGE_SIZE;
974 } else if (ret < 0) {
979 BUG_ON(disk_num_bytes >
980 btrfs_super_total_bytes(root->fs_info->super_copy));
982 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
983 btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
985 while (disk_num_bytes > 0) {
988 cur_alloc_size = disk_num_bytes;
989 ret = btrfs_reserve_extent(root, cur_alloc_size,
990 root->sectorsize, 0, alloc_hint,
995 em = alloc_extent_map();
1001 em->orig_start = em->start;
1002 ram_size = ins.offset;
1003 em->len = ins.offset;
1004 em->mod_start = em->start;
1005 em->mod_len = em->len;
1007 em->block_start = ins.objectid;
1008 em->block_len = ins.offset;
1009 em->orig_block_len = ins.offset;
1010 em->ram_bytes = ram_size;
1011 em->bdev = root->fs_info->fs_devices->latest_bdev;
1012 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1013 em->generation = -1;
1016 write_lock(&em_tree->lock);
1017 ret = add_extent_mapping(em_tree, em, 1);
1018 write_unlock(&em_tree->lock);
1019 if (ret != -EEXIST) {
1020 free_extent_map(em);
1023 btrfs_drop_extent_cache(inode, start,
1024 start + ram_size - 1, 0);
1029 cur_alloc_size = ins.offset;
1030 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1031 ram_size, cur_alloc_size, 0);
1033 goto out_drop_extent_cache;
1035 if (root->root_key.objectid ==
1036 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1037 ret = btrfs_reloc_clone_csums(inode, start,
1040 goto out_drop_extent_cache;
1043 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1045 if (disk_num_bytes < cur_alloc_size)
1048 /* we're not doing compressed IO, don't unlock the first
1049 * page (which the caller expects to stay locked), don't
1050 * clear any dirty bits and don't set any writeback bits
1052 * Do set the Private2 bit so we know this page was properly
1053 * setup for writepage
1055 op = unlock ? PAGE_UNLOCK : 0;
1056 op |= PAGE_SET_PRIVATE2;
1058 extent_clear_unlock_delalloc(inode, start,
1059 start + ram_size - 1, locked_page,
1060 EXTENT_LOCKED | EXTENT_DELALLOC,
1062 disk_num_bytes -= cur_alloc_size;
1063 num_bytes -= cur_alloc_size;
1064 alloc_hint = ins.objectid + ins.offset;
1065 start += cur_alloc_size;
1070 out_drop_extent_cache:
1071 btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1073 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
1074 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1076 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1077 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1078 EXTENT_DELALLOC | EXTENT_DEFRAG,
1079 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1080 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1085 * work queue call back to started compression on a file and pages
1087 static noinline void async_cow_start(struct btrfs_work *work)
1089 struct async_cow *async_cow;
1091 async_cow = container_of(work, struct async_cow, work);
1093 compress_file_range(async_cow->inode, async_cow->locked_page,
1094 async_cow->start, async_cow->end, async_cow,
1096 if (num_added == 0) {
1097 btrfs_add_delayed_iput(async_cow->inode);
1098 async_cow->inode = NULL;
1103 * work queue call back to submit previously compressed pages
1105 static noinline void async_cow_submit(struct btrfs_work *work)
1107 struct async_cow *async_cow;
1108 struct btrfs_root *root;
1109 unsigned long nr_pages;
1111 async_cow = container_of(work, struct async_cow, work);
1113 root = async_cow->root;
1114 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1118 * atomic_sub_return implies a barrier for waitqueue_active
1120 if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1122 waitqueue_active(&root->fs_info->async_submit_wait))
1123 wake_up(&root->fs_info->async_submit_wait);
1125 if (async_cow->inode)
1126 submit_compressed_extents(async_cow->inode, async_cow);
1129 static noinline void async_cow_free(struct btrfs_work *work)
1131 struct async_cow *async_cow;
1132 async_cow = container_of(work, struct async_cow, work);
1133 if (async_cow->inode)
1134 btrfs_add_delayed_iput(async_cow->inode);
1138 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1139 u64 start, u64 end, int *page_started,
1140 unsigned long *nr_written)
1142 struct async_cow *async_cow;
1143 struct btrfs_root *root = BTRFS_I(inode)->root;
1144 unsigned long nr_pages;
1146 int limit = 10 * SZ_1M;
1148 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1149 1, 0, NULL, GFP_NOFS);
1150 while (start < end) {
1151 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1152 BUG_ON(!async_cow); /* -ENOMEM */
1153 async_cow->inode = igrab(inode);
1154 async_cow->root = root;
1155 async_cow->locked_page = locked_page;
1156 async_cow->start = start;
1158 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1159 !btrfs_test_opt(root, FORCE_COMPRESS))
1162 cur_end = min(end, start + SZ_512K - 1);
1164 async_cow->end = cur_end;
1165 INIT_LIST_HEAD(&async_cow->extents);
1167 btrfs_init_work(&async_cow->work,
1168 btrfs_delalloc_helper,
1169 async_cow_start, async_cow_submit,
1172 nr_pages = (cur_end - start + PAGE_SIZE) >>
1174 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1176 btrfs_queue_work(root->fs_info->delalloc_workers,
1179 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1180 wait_event(root->fs_info->async_submit_wait,
1181 (atomic_read(&root->fs_info->async_delalloc_pages) <
1185 while (atomic_read(&root->fs_info->async_submit_draining) &&
1186 atomic_read(&root->fs_info->async_delalloc_pages)) {
1187 wait_event(root->fs_info->async_submit_wait,
1188 (atomic_read(&root->fs_info->async_delalloc_pages) ==
1192 *nr_written += nr_pages;
1193 start = cur_end + 1;
1199 static noinline int csum_exist_in_range(struct btrfs_root *root,
1200 u64 bytenr, u64 num_bytes)
1203 struct btrfs_ordered_sum *sums;
1206 ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1207 bytenr + num_bytes - 1, &list, 0);
1208 if (ret == 0 && list_empty(&list))
1211 while (!list_empty(&list)) {
1212 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1213 list_del(&sums->list);
1220 * when nowcow writeback call back. This checks for snapshots or COW copies
1221 * of the extents that exist in the file, and COWs the file as required.
1223 * If no cow copies or snapshots exist, we write directly to the existing
1226 static noinline int run_delalloc_nocow(struct inode *inode,
1227 struct page *locked_page,
1228 u64 start, u64 end, int *page_started, int force,
1229 unsigned long *nr_written)
1231 struct btrfs_root *root = BTRFS_I(inode)->root;
1232 struct btrfs_trans_handle *trans;
1233 struct extent_buffer *leaf;
1234 struct btrfs_path *path;
1235 struct btrfs_file_extent_item *fi;
1236 struct btrfs_key found_key;
1251 u64 ino = btrfs_ino(inode);
1253 path = btrfs_alloc_path();
1255 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1256 EXTENT_LOCKED | EXTENT_DELALLOC |
1257 EXTENT_DO_ACCOUNTING |
1258 EXTENT_DEFRAG, PAGE_UNLOCK |
1260 PAGE_SET_WRITEBACK |
1261 PAGE_END_WRITEBACK);
1265 nolock = btrfs_is_free_space_inode(inode);
1268 trans = btrfs_join_transaction_nolock(root);
1270 trans = btrfs_join_transaction(root);
1272 if (IS_ERR(trans)) {
1273 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1274 EXTENT_LOCKED | EXTENT_DELALLOC |
1275 EXTENT_DO_ACCOUNTING |
1276 EXTENT_DEFRAG, PAGE_UNLOCK |
1278 PAGE_SET_WRITEBACK |
1279 PAGE_END_WRITEBACK);
1280 btrfs_free_path(path);
1281 return PTR_ERR(trans);
1284 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1286 cow_start = (u64)-1;
1289 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1293 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1294 leaf = path->nodes[0];
1295 btrfs_item_key_to_cpu(leaf, &found_key,
1296 path->slots[0] - 1);
1297 if (found_key.objectid == ino &&
1298 found_key.type == BTRFS_EXTENT_DATA_KEY)
1303 leaf = path->nodes[0];
1304 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1305 ret = btrfs_next_leaf(root, path);
1310 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1318 if (found_key.objectid > ino)
1320 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1321 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1325 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1326 found_key.offset > end)
1329 if (found_key.offset > cur_offset) {
1330 extent_end = found_key.offset;
1335 fi = btrfs_item_ptr(leaf, path->slots[0],
1336 struct btrfs_file_extent_item);
1337 extent_type = btrfs_file_extent_type(leaf, fi);
1339 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1340 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1341 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1342 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1343 extent_offset = btrfs_file_extent_offset(leaf, fi);
1344 extent_end = found_key.offset +
1345 btrfs_file_extent_num_bytes(leaf, fi);
1347 btrfs_file_extent_disk_num_bytes(leaf, fi);
1348 if (extent_end <= start) {
1352 if (disk_bytenr == 0)
1354 if (btrfs_file_extent_compression(leaf, fi) ||
1355 btrfs_file_extent_encryption(leaf, fi) ||
1356 btrfs_file_extent_other_encoding(leaf, fi))
1358 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1360 if (btrfs_extent_readonly(root, disk_bytenr))
1362 if (btrfs_cross_ref_exist(trans, root, ino,
1364 extent_offset, disk_bytenr))
1366 disk_bytenr += extent_offset;
1367 disk_bytenr += cur_offset - found_key.offset;
1368 num_bytes = min(end + 1, extent_end) - cur_offset;
1370 * if there are pending snapshots for this root,
1371 * we fall into common COW way.
1374 err = btrfs_start_write_no_snapshoting(root);
1379 * force cow if csum exists in the range.
1380 * this ensure that csum for a given extent are
1381 * either valid or do not exist.
1383 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1385 if (!btrfs_inc_nocow_writers(root->fs_info,
1389 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1390 extent_end = found_key.offset +
1391 btrfs_file_extent_inline_len(leaf,
1392 path->slots[0], fi);
1393 extent_end = ALIGN(extent_end, root->sectorsize);
1398 if (extent_end <= start) {
1400 if (!nolock && nocow)
1401 btrfs_end_write_no_snapshoting(root);
1403 btrfs_dec_nocow_writers(root->fs_info,
1408 if (cow_start == (u64)-1)
1409 cow_start = cur_offset;
1410 cur_offset = extent_end;
1411 if (cur_offset > end)
1417 btrfs_release_path(path);
1418 if (cow_start != (u64)-1) {
1419 ret = cow_file_range(inode, locked_page,
1420 cow_start, found_key.offset - 1,
1421 page_started, nr_written, 1);
1423 if (!nolock && nocow)
1424 btrfs_end_write_no_snapshoting(root);
1426 btrfs_dec_nocow_writers(root->fs_info,
1430 cow_start = (u64)-1;
1433 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1434 struct extent_map *em;
1435 struct extent_map_tree *em_tree;
1436 em_tree = &BTRFS_I(inode)->extent_tree;
1437 em = alloc_extent_map();
1438 BUG_ON(!em); /* -ENOMEM */
1439 em->start = cur_offset;
1440 em->orig_start = found_key.offset - extent_offset;
1441 em->len = num_bytes;
1442 em->block_len = num_bytes;
1443 em->block_start = disk_bytenr;
1444 em->orig_block_len = disk_num_bytes;
1445 em->ram_bytes = ram_bytes;
1446 em->bdev = root->fs_info->fs_devices->latest_bdev;
1447 em->mod_start = em->start;
1448 em->mod_len = em->len;
1449 set_bit(EXTENT_FLAG_PINNED, &em->flags);
1450 set_bit(EXTENT_FLAG_FILLING, &em->flags);
1451 em->generation = -1;
1453 write_lock(&em_tree->lock);
1454 ret = add_extent_mapping(em_tree, em, 1);
1455 write_unlock(&em_tree->lock);
1456 if (ret != -EEXIST) {
1457 free_extent_map(em);
1460 btrfs_drop_extent_cache(inode, em->start,
1461 em->start + em->len - 1, 0);
1463 type = BTRFS_ORDERED_PREALLOC;
1465 type = BTRFS_ORDERED_NOCOW;
1468 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1469 num_bytes, num_bytes, type);
1471 btrfs_dec_nocow_writers(root->fs_info, disk_bytenr);
1472 BUG_ON(ret); /* -ENOMEM */
1474 if (root->root_key.objectid ==
1475 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1476 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1479 if (!nolock && nocow)
1480 btrfs_end_write_no_snapshoting(root);
1485 extent_clear_unlock_delalloc(inode, cur_offset,
1486 cur_offset + num_bytes - 1,
1487 locked_page, EXTENT_LOCKED |
1488 EXTENT_DELALLOC, PAGE_UNLOCK |
1490 if (!nolock && nocow)
1491 btrfs_end_write_no_snapshoting(root);
1492 cur_offset = extent_end;
1493 if (cur_offset > end)
1496 btrfs_release_path(path);
1498 if (cur_offset <= end && cow_start == (u64)-1) {
1499 cow_start = cur_offset;
1503 if (cow_start != (u64)-1) {
1504 ret = cow_file_range(inode, locked_page, cow_start, end,
1505 page_started, nr_written, 1);
1511 err = btrfs_end_transaction(trans, root);
1515 if (ret && cur_offset < end)
1516 extent_clear_unlock_delalloc(inode, cur_offset, end,
1517 locked_page, EXTENT_LOCKED |
1518 EXTENT_DELALLOC | EXTENT_DEFRAG |
1519 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1521 PAGE_SET_WRITEBACK |
1522 PAGE_END_WRITEBACK);
1523 btrfs_free_path(path);
1527 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1530 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1531 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1535 * @defrag_bytes is a hint value, no spinlock held here,
1536 * if is not zero, it means the file is defragging.
1537 * Force cow if given extent needs to be defragged.
1539 if (BTRFS_I(inode)->defrag_bytes &&
1540 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1541 EXTENT_DEFRAG, 0, NULL))
1548 * extent_io.c call back to do delayed allocation processing
1550 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1551 u64 start, u64 end, int *page_started,
1552 unsigned long *nr_written)
1555 int force_cow = need_force_cow(inode, start, end);
1557 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1558 ret = run_delalloc_nocow(inode, locked_page, start, end,
1559 page_started, 1, nr_written);
1560 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1561 ret = run_delalloc_nocow(inode, locked_page, start, end,
1562 page_started, 0, nr_written);
1563 } else if (!inode_need_compress(inode)) {
1564 ret = cow_file_range(inode, locked_page, start, end,
1565 page_started, nr_written, 1);
1567 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1568 &BTRFS_I(inode)->runtime_flags);
1569 ret = cow_file_range_async(inode, locked_page, start, end,
1570 page_started, nr_written);
1575 static void btrfs_split_extent_hook(struct inode *inode,
1576 struct extent_state *orig, u64 split)
1580 /* not delalloc, ignore it */
1581 if (!(orig->state & EXTENT_DELALLOC))
1584 size = orig->end - orig->start + 1;
1585 if (size > BTRFS_MAX_EXTENT_SIZE) {
1590 * See the explanation in btrfs_merge_extent_hook, the same
1591 * applies here, just in reverse.
1593 new_size = orig->end - split + 1;
1594 num_extents = div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1595 BTRFS_MAX_EXTENT_SIZE);
1596 new_size = split - orig->start;
1597 num_extents += div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1598 BTRFS_MAX_EXTENT_SIZE);
1599 if (div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1600 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1604 spin_lock(&BTRFS_I(inode)->lock);
1605 BTRFS_I(inode)->outstanding_extents++;
1606 spin_unlock(&BTRFS_I(inode)->lock);
1610 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1611 * extents so we can keep track of new extents that are just merged onto old
1612 * extents, such as when we are doing sequential writes, so we can properly
1613 * account for the metadata space we'll need.
1615 static void btrfs_merge_extent_hook(struct inode *inode,
1616 struct extent_state *new,
1617 struct extent_state *other)
1619 u64 new_size, old_size;
1622 /* not delalloc, ignore it */
1623 if (!(other->state & EXTENT_DELALLOC))
1626 if (new->start > other->start)
1627 new_size = new->end - other->start + 1;
1629 new_size = other->end - new->start + 1;
1631 /* we're not bigger than the max, unreserve the space and go */
1632 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1633 spin_lock(&BTRFS_I(inode)->lock);
1634 BTRFS_I(inode)->outstanding_extents--;
1635 spin_unlock(&BTRFS_I(inode)->lock);
1640 * We have to add up either side to figure out how many extents were
1641 * accounted for before we merged into one big extent. If the number of
1642 * extents we accounted for is <= the amount we need for the new range
1643 * then we can return, otherwise drop. Think of it like this
1647 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1648 * need 2 outstanding extents, on one side we have 1 and the other side
1649 * we have 1 so they are == and we can return. But in this case
1651 * [MAX_SIZE+4k][MAX_SIZE+4k]
1653 * Each range on their own accounts for 2 extents, but merged together
1654 * they are only 3 extents worth of accounting, so we need to drop in
1657 old_size = other->end - other->start + 1;
1658 num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1659 BTRFS_MAX_EXTENT_SIZE);
1660 old_size = new->end - new->start + 1;
1661 num_extents += div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1662 BTRFS_MAX_EXTENT_SIZE);
1664 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1665 BTRFS_MAX_EXTENT_SIZE) >= num_extents)
1668 spin_lock(&BTRFS_I(inode)->lock);
1669 BTRFS_I(inode)->outstanding_extents--;
1670 spin_unlock(&BTRFS_I(inode)->lock);
1673 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1674 struct inode *inode)
1676 spin_lock(&root->delalloc_lock);
1677 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1678 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1679 &root->delalloc_inodes);
1680 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1681 &BTRFS_I(inode)->runtime_flags);
1682 root->nr_delalloc_inodes++;
1683 if (root->nr_delalloc_inodes == 1) {
1684 spin_lock(&root->fs_info->delalloc_root_lock);
1685 BUG_ON(!list_empty(&root->delalloc_root));
1686 list_add_tail(&root->delalloc_root,
1687 &root->fs_info->delalloc_roots);
1688 spin_unlock(&root->fs_info->delalloc_root_lock);
1691 spin_unlock(&root->delalloc_lock);
1694 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1695 struct inode *inode)
1697 spin_lock(&root->delalloc_lock);
1698 if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1699 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1700 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1701 &BTRFS_I(inode)->runtime_flags);
1702 root->nr_delalloc_inodes--;
1703 if (!root->nr_delalloc_inodes) {
1704 spin_lock(&root->fs_info->delalloc_root_lock);
1705 BUG_ON(list_empty(&root->delalloc_root));
1706 list_del_init(&root->delalloc_root);
1707 spin_unlock(&root->fs_info->delalloc_root_lock);
1710 spin_unlock(&root->delalloc_lock);
1714 * extent_io.c set_bit_hook, used to track delayed allocation
1715 * bytes in this file, and to maintain the list of inodes that
1716 * have pending delalloc work to be done.
1718 static void btrfs_set_bit_hook(struct inode *inode,
1719 struct extent_state *state, unsigned *bits)
1722 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1725 * set_bit and clear bit hooks normally require _irqsave/restore
1726 * but in this case, we are only testing for the DELALLOC
1727 * bit, which is only set or cleared with irqs on
1729 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1730 struct btrfs_root *root = BTRFS_I(inode)->root;
1731 u64 len = state->end + 1 - state->start;
1732 bool do_list = !btrfs_is_free_space_inode(inode);
1734 if (*bits & EXTENT_FIRST_DELALLOC) {
1735 *bits &= ~EXTENT_FIRST_DELALLOC;
1737 spin_lock(&BTRFS_I(inode)->lock);
1738 BTRFS_I(inode)->outstanding_extents++;
1739 spin_unlock(&BTRFS_I(inode)->lock);
1742 /* For sanity tests */
1743 if (btrfs_test_is_dummy_root(root))
1746 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1747 root->fs_info->delalloc_batch);
1748 spin_lock(&BTRFS_I(inode)->lock);
1749 BTRFS_I(inode)->delalloc_bytes += len;
1750 if (*bits & EXTENT_DEFRAG)
1751 BTRFS_I(inode)->defrag_bytes += len;
1752 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1753 &BTRFS_I(inode)->runtime_flags))
1754 btrfs_add_delalloc_inodes(root, inode);
1755 spin_unlock(&BTRFS_I(inode)->lock);
1760 * extent_io.c clear_bit_hook, see set_bit_hook for why
1762 static void btrfs_clear_bit_hook(struct inode *inode,
1763 struct extent_state *state,
1766 u64 len = state->end + 1 - state->start;
1767 u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1768 BTRFS_MAX_EXTENT_SIZE);
1770 spin_lock(&BTRFS_I(inode)->lock);
1771 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1772 BTRFS_I(inode)->defrag_bytes -= len;
1773 spin_unlock(&BTRFS_I(inode)->lock);
1776 * set_bit and clear bit hooks normally require _irqsave/restore
1777 * but in this case, we are only testing for the DELALLOC
1778 * bit, which is only set or cleared with irqs on
1780 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1781 struct btrfs_root *root = BTRFS_I(inode)->root;
1782 bool do_list = !btrfs_is_free_space_inode(inode);
1784 if (*bits & EXTENT_FIRST_DELALLOC) {
1785 *bits &= ~EXTENT_FIRST_DELALLOC;
1786 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1787 spin_lock(&BTRFS_I(inode)->lock);
1788 BTRFS_I(inode)->outstanding_extents -= num_extents;
1789 spin_unlock(&BTRFS_I(inode)->lock);
1793 * We don't reserve metadata space for space cache inodes so we
1794 * don't need to call dellalloc_release_metadata if there is an
1797 if (*bits & EXTENT_DO_ACCOUNTING &&
1798 root != root->fs_info->tree_root)
1799 btrfs_delalloc_release_metadata(inode, len);
1801 /* For sanity tests. */
1802 if (btrfs_test_is_dummy_root(root))
1805 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1806 && do_list && !(state->state & EXTENT_NORESERVE))
1807 btrfs_free_reserved_data_space_noquota(inode,
1810 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1811 root->fs_info->delalloc_batch);
1812 spin_lock(&BTRFS_I(inode)->lock);
1813 BTRFS_I(inode)->delalloc_bytes -= len;
1814 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1815 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1816 &BTRFS_I(inode)->runtime_flags))
1817 btrfs_del_delalloc_inode(root, inode);
1818 spin_unlock(&BTRFS_I(inode)->lock);
1823 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1824 * we don't create bios that span stripes or chunks
1826 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1827 size_t size, struct bio *bio,
1828 unsigned long bio_flags)
1830 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1831 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1836 if (bio_flags & EXTENT_BIO_COMPRESSED)
1839 length = bio->bi_iter.bi_size;
1840 map_length = length;
1841 ret = btrfs_map_block(root->fs_info, bio_op(bio), logical,
1842 &map_length, NULL, 0);
1843 /* Will always return 0 with map_multi == NULL */
1845 if (map_length < length + size)
1851 * in order to insert checksums into the metadata in large chunks,
1852 * we wait until bio submission time. All the pages in the bio are
1853 * checksummed and sums are attached onto the ordered extent record.
1855 * At IO completion time the cums attached on the ordered extent record
1856 * are inserted into the btree
1858 static int __btrfs_submit_bio_start(struct inode *inode, struct bio *bio,
1859 int mirror_num, unsigned long bio_flags,
1862 struct btrfs_root *root = BTRFS_I(inode)->root;
1865 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1866 BUG_ON(ret); /* -ENOMEM */
1871 * in order to insert checksums into the metadata in large chunks,
1872 * we wait until bio submission time. All the pages in the bio are
1873 * checksummed and sums are attached onto the ordered extent record.
1875 * At IO completion time the cums attached on the ordered extent record
1876 * are inserted into the btree
1878 static int __btrfs_submit_bio_done(struct inode *inode, struct bio *bio,
1879 int mirror_num, unsigned long bio_flags,
1882 struct btrfs_root *root = BTRFS_I(inode)->root;
1885 ret = btrfs_map_bio(root, bio, mirror_num, 1);
1887 bio->bi_error = ret;
1894 * extent_io.c submission hook. This does the right thing for csum calculation
1895 * on write, or reading the csums from the tree before a read
1897 static int btrfs_submit_bio_hook(struct inode *inode, struct bio *bio,
1898 int mirror_num, unsigned long bio_flags,
1901 struct btrfs_root *root = BTRFS_I(inode)->root;
1902 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1905 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1907 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1909 if (btrfs_is_free_space_inode(inode))
1910 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1912 if (bio_op(bio) != REQ_OP_WRITE) {
1913 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1917 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1918 ret = btrfs_submit_compressed_read(inode, bio,
1922 } else if (!skip_sum) {
1923 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1928 } else if (async && !skip_sum) {
1929 /* csum items have already been cloned */
1930 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1932 /* we're doing a write, do the async checksumming */
1933 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1934 inode, bio, mirror_num,
1935 bio_flags, bio_offset,
1936 __btrfs_submit_bio_start,
1937 __btrfs_submit_bio_done);
1939 } else if (!skip_sum) {
1940 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1946 ret = btrfs_map_bio(root, bio, mirror_num, 0);
1950 bio->bi_error = ret;
1957 * given a list of ordered sums record them in the inode. This happens
1958 * at IO completion time based on sums calculated at bio submission time.
1960 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1961 struct inode *inode, u64 file_offset,
1962 struct list_head *list)
1964 struct btrfs_ordered_sum *sum;
1966 list_for_each_entry(sum, list, list) {
1967 trans->adding_csums = 1;
1968 btrfs_csum_file_blocks(trans,
1969 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1970 trans->adding_csums = 0;
1975 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1976 struct extent_state **cached_state)
1978 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1979 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1983 /* see btrfs_writepage_start_hook for details on why this is required */
1984 struct btrfs_writepage_fixup {
1986 struct btrfs_work work;
1989 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1991 struct btrfs_writepage_fixup *fixup;
1992 struct btrfs_ordered_extent *ordered;
1993 struct extent_state *cached_state = NULL;
1995 struct inode *inode;
2000 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2004 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2005 ClearPageChecked(page);
2009 inode = page->mapping->host;
2010 page_start = page_offset(page);
2011 page_end = page_offset(page) + PAGE_SIZE - 1;
2013 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2016 /* already ordered? We're done */
2017 if (PagePrivate2(page))
2020 ordered = btrfs_lookup_ordered_range(inode, page_start,
2023 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2024 page_end, &cached_state, GFP_NOFS);
2026 btrfs_start_ordered_extent(inode, ordered, 1);
2027 btrfs_put_ordered_extent(ordered);
2031 ret = btrfs_delalloc_reserve_space(inode, page_start,
2034 mapping_set_error(page->mapping, ret);
2035 end_extent_writepage(page, ret, page_start, page_end);
2036 ClearPageChecked(page);
2040 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2041 ClearPageChecked(page);
2042 set_page_dirty(page);
2044 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2045 &cached_state, GFP_NOFS);
2053 * There are a few paths in the higher layers of the kernel that directly
2054 * set the page dirty bit without asking the filesystem if it is a
2055 * good idea. This causes problems because we want to make sure COW
2056 * properly happens and the data=ordered rules are followed.
2058 * In our case any range that doesn't have the ORDERED bit set
2059 * hasn't been properly setup for IO. We kick off an async process
2060 * to fix it up. The async helper will wait for ordered extents, set
2061 * the delalloc bit and make it safe to write the page.
2063 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2065 struct inode *inode = page->mapping->host;
2066 struct btrfs_writepage_fixup *fixup;
2067 struct btrfs_root *root = BTRFS_I(inode)->root;
2069 /* this page is properly in the ordered list */
2070 if (TestClearPagePrivate2(page))
2073 if (PageChecked(page))
2076 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2080 SetPageChecked(page);
2082 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2083 btrfs_writepage_fixup_worker, NULL, NULL);
2085 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2089 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2090 struct inode *inode, u64 file_pos,
2091 u64 disk_bytenr, u64 disk_num_bytes,
2092 u64 num_bytes, u64 ram_bytes,
2093 u8 compression, u8 encryption,
2094 u16 other_encoding, int extent_type)
2096 struct btrfs_root *root = BTRFS_I(inode)->root;
2097 struct btrfs_file_extent_item *fi;
2098 struct btrfs_path *path;
2099 struct extent_buffer *leaf;
2100 struct btrfs_key ins;
2101 int extent_inserted = 0;
2104 path = btrfs_alloc_path();
2109 * we may be replacing one extent in the tree with another.
2110 * The new extent is pinned in the extent map, and we don't want
2111 * to drop it from the cache until it is completely in the btree.
2113 * So, tell btrfs_drop_extents to leave this extent in the cache.
2114 * the caller is expected to unpin it and allow it to be merged
2117 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2118 file_pos + num_bytes, NULL, 0,
2119 1, sizeof(*fi), &extent_inserted);
2123 if (!extent_inserted) {
2124 ins.objectid = btrfs_ino(inode);
2125 ins.offset = file_pos;
2126 ins.type = BTRFS_EXTENT_DATA_KEY;
2128 path->leave_spinning = 1;
2129 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2134 leaf = path->nodes[0];
2135 fi = btrfs_item_ptr(leaf, path->slots[0],
2136 struct btrfs_file_extent_item);
2137 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2138 btrfs_set_file_extent_type(leaf, fi, extent_type);
2139 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2140 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2141 btrfs_set_file_extent_offset(leaf, fi, 0);
2142 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2143 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2144 btrfs_set_file_extent_compression(leaf, fi, compression);
2145 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2146 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2148 btrfs_mark_buffer_dirty(leaf);
2149 btrfs_release_path(path);
2151 inode_add_bytes(inode, num_bytes);
2153 ins.objectid = disk_bytenr;
2154 ins.offset = disk_num_bytes;
2155 ins.type = BTRFS_EXTENT_ITEM_KEY;
2156 ret = btrfs_alloc_reserved_file_extent(trans, root,
2157 root->root_key.objectid,
2158 btrfs_ino(inode), file_pos,
2161 * Release the reserved range from inode dirty range map, as it is
2162 * already moved into delayed_ref_head
2164 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2166 btrfs_free_path(path);
2171 /* snapshot-aware defrag */
2172 struct sa_defrag_extent_backref {
2173 struct rb_node node;
2174 struct old_sa_defrag_extent *old;
2183 struct old_sa_defrag_extent {
2184 struct list_head list;
2185 struct new_sa_defrag_extent *new;
2194 struct new_sa_defrag_extent {
2195 struct rb_root root;
2196 struct list_head head;
2197 struct btrfs_path *path;
2198 struct inode *inode;
2206 static int backref_comp(struct sa_defrag_extent_backref *b1,
2207 struct sa_defrag_extent_backref *b2)
2209 if (b1->root_id < b2->root_id)
2211 else if (b1->root_id > b2->root_id)
2214 if (b1->inum < b2->inum)
2216 else if (b1->inum > b2->inum)
2219 if (b1->file_pos < b2->file_pos)
2221 else if (b1->file_pos > b2->file_pos)
2225 * [------------------------------] ===> (a range of space)
2226 * |<--->| |<---->| =============> (fs/file tree A)
2227 * |<---------------------------->| ===> (fs/file tree B)
2229 * A range of space can refer to two file extents in one tree while
2230 * refer to only one file extent in another tree.
2232 * So we may process a disk offset more than one time(two extents in A)
2233 * and locate at the same extent(one extent in B), then insert two same
2234 * backrefs(both refer to the extent in B).
2239 static void backref_insert(struct rb_root *root,
2240 struct sa_defrag_extent_backref *backref)
2242 struct rb_node **p = &root->rb_node;
2243 struct rb_node *parent = NULL;
2244 struct sa_defrag_extent_backref *entry;
2249 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2251 ret = backref_comp(backref, entry);
2255 p = &(*p)->rb_right;
2258 rb_link_node(&backref->node, parent, p);
2259 rb_insert_color(&backref->node, root);
2263 * Note the backref might has changed, and in this case we just return 0.
2265 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2268 struct btrfs_file_extent_item *extent;
2269 struct btrfs_fs_info *fs_info;
2270 struct old_sa_defrag_extent *old = ctx;
2271 struct new_sa_defrag_extent *new = old->new;
2272 struct btrfs_path *path = new->path;
2273 struct btrfs_key key;
2274 struct btrfs_root *root;
2275 struct sa_defrag_extent_backref *backref;
2276 struct extent_buffer *leaf;
2277 struct inode *inode = new->inode;
2283 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2284 inum == btrfs_ino(inode))
2287 key.objectid = root_id;
2288 key.type = BTRFS_ROOT_ITEM_KEY;
2289 key.offset = (u64)-1;
2291 fs_info = BTRFS_I(inode)->root->fs_info;
2292 root = btrfs_read_fs_root_no_name(fs_info, &key);
2294 if (PTR_ERR(root) == -ENOENT)
2297 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2298 inum, offset, root_id);
2299 return PTR_ERR(root);
2302 key.objectid = inum;
2303 key.type = BTRFS_EXTENT_DATA_KEY;
2304 if (offset > (u64)-1 << 32)
2307 key.offset = offset;
2309 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2310 if (WARN_ON(ret < 0))
2317 leaf = path->nodes[0];
2318 slot = path->slots[0];
2320 if (slot >= btrfs_header_nritems(leaf)) {
2321 ret = btrfs_next_leaf(root, path);
2324 } else if (ret > 0) {
2333 btrfs_item_key_to_cpu(leaf, &key, slot);
2335 if (key.objectid > inum)
2338 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2341 extent = btrfs_item_ptr(leaf, slot,
2342 struct btrfs_file_extent_item);
2344 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2348 * 'offset' refers to the exact key.offset,
2349 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2350 * (key.offset - extent_offset).
2352 if (key.offset != offset)
2355 extent_offset = btrfs_file_extent_offset(leaf, extent);
2356 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2358 if (extent_offset >= old->extent_offset + old->offset +
2359 old->len || extent_offset + num_bytes <=
2360 old->extent_offset + old->offset)
2365 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2371 backref->root_id = root_id;
2372 backref->inum = inum;
2373 backref->file_pos = offset;
2374 backref->num_bytes = num_bytes;
2375 backref->extent_offset = extent_offset;
2376 backref->generation = btrfs_file_extent_generation(leaf, extent);
2378 backref_insert(&new->root, backref);
2381 btrfs_release_path(path);
2386 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2387 struct new_sa_defrag_extent *new)
2389 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2390 struct old_sa_defrag_extent *old, *tmp;
2395 list_for_each_entry_safe(old, tmp, &new->head, list) {
2396 ret = iterate_inodes_from_logical(old->bytenr +
2397 old->extent_offset, fs_info,
2398 path, record_one_backref,
2400 if (ret < 0 && ret != -ENOENT)
2403 /* no backref to be processed for this extent */
2405 list_del(&old->list);
2410 if (list_empty(&new->head))
2416 static int relink_is_mergable(struct extent_buffer *leaf,
2417 struct btrfs_file_extent_item *fi,
2418 struct new_sa_defrag_extent *new)
2420 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2423 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2426 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2429 if (btrfs_file_extent_encryption(leaf, fi) ||
2430 btrfs_file_extent_other_encoding(leaf, fi))
2437 * Note the backref might has changed, and in this case we just return 0.
2439 static noinline int relink_extent_backref(struct btrfs_path *path,
2440 struct sa_defrag_extent_backref *prev,
2441 struct sa_defrag_extent_backref *backref)
2443 struct btrfs_file_extent_item *extent;
2444 struct btrfs_file_extent_item *item;
2445 struct btrfs_ordered_extent *ordered;
2446 struct btrfs_trans_handle *trans;
2447 struct btrfs_fs_info *fs_info;
2448 struct btrfs_root *root;
2449 struct btrfs_key key;
2450 struct extent_buffer *leaf;
2451 struct old_sa_defrag_extent *old = backref->old;
2452 struct new_sa_defrag_extent *new = old->new;
2453 struct inode *src_inode = new->inode;
2454 struct inode *inode;
2455 struct extent_state *cached = NULL;
2464 if (prev && prev->root_id == backref->root_id &&
2465 prev->inum == backref->inum &&
2466 prev->file_pos + prev->num_bytes == backref->file_pos)
2469 /* step 1: get root */
2470 key.objectid = backref->root_id;
2471 key.type = BTRFS_ROOT_ITEM_KEY;
2472 key.offset = (u64)-1;
2474 fs_info = BTRFS_I(src_inode)->root->fs_info;
2475 index = srcu_read_lock(&fs_info->subvol_srcu);
2477 root = btrfs_read_fs_root_no_name(fs_info, &key);
2479 srcu_read_unlock(&fs_info->subvol_srcu, index);
2480 if (PTR_ERR(root) == -ENOENT)
2482 return PTR_ERR(root);
2485 if (btrfs_root_readonly(root)) {
2486 srcu_read_unlock(&fs_info->subvol_srcu, index);
2490 /* step 2: get inode */
2491 key.objectid = backref->inum;
2492 key.type = BTRFS_INODE_ITEM_KEY;
2495 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2496 if (IS_ERR(inode)) {
2497 srcu_read_unlock(&fs_info->subvol_srcu, index);
2501 srcu_read_unlock(&fs_info->subvol_srcu, index);
2503 /* step 3: relink backref */
2504 lock_start = backref->file_pos;
2505 lock_end = backref->file_pos + backref->num_bytes - 1;
2506 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2509 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2511 btrfs_put_ordered_extent(ordered);
2515 trans = btrfs_join_transaction(root);
2516 if (IS_ERR(trans)) {
2517 ret = PTR_ERR(trans);
2521 key.objectid = backref->inum;
2522 key.type = BTRFS_EXTENT_DATA_KEY;
2523 key.offset = backref->file_pos;
2525 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2528 } else if (ret > 0) {
2533 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2534 struct btrfs_file_extent_item);
2536 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2537 backref->generation)
2540 btrfs_release_path(path);
2542 start = backref->file_pos;
2543 if (backref->extent_offset < old->extent_offset + old->offset)
2544 start += old->extent_offset + old->offset -
2545 backref->extent_offset;
2547 len = min(backref->extent_offset + backref->num_bytes,
2548 old->extent_offset + old->offset + old->len);
2549 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2551 ret = btrfs_drop_extents(trans, root, inode, start,
2556 key.objectid = btrfs_ino(inode);
2557 key.type = BTRFS_EXTENT_DATA_KEY;
2560 path->leave_spinning = 1;
2562 struct btrfs_file_extent_item *fi;
2564 struct btrfs_key found_key;
2566 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2571 leaf = path->nodes[0];
2572 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2574 fi = btrfs_item_ptr(leaf, path->slots[0],
2575 struct btrfs_file_extent_item);
2576 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2578 if (extent_len + found_key.offset == start &&
2579 relink_is_mergable(leaf, fi, new)) {
2580 btrfs_set_file_extent_num_bytes(leaf, fi,
2582 btrfs_mark_buffer_dirty(leaf);
2583 inode_add_bytes(inode, len);
2589 btrfs_release_path(path);
2594 ret = btrfs_insert_empty_item(trans, root, path, &key,
2597 btrfs_abort_transaction(trans, root, ret);
2601 leaf = path->nodes[0];
2602 item = btrfs_item_ptr(leaf, path->slots[0],
2603 struct btrfs_file_extent_item);
2604 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2605 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2606 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2607 btrfs_set_file_extent_num_bytes(leaf, item, len);
2608 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2609 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2610 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2611 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2612 btrfs_set_file_extent_encryption(leaf, item, 0);
2613 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2615 btrfs_mark_buffer_dirty(leaf);
2616 inode_add_bytes(inode, len);
2617 btrfs_release_path(path);
2619 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2621 backref->root_id, backref->inum,
2622 new->file_pos); /* start - extent_offset */
2624 btrfs_abort_transaction(trans, root, ret);
2630 btrfs_release_path(path);
2631 path->leave_spinning = 0;
2632 btrfs_end_transaction(trans, root);
2634 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2640 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2642 struct old_sa_defrag_extent *old, *tmp;
2647 list_for_each_entry_safe(old, tmp, &new->head, list) {
2653 static void relink_file_extents(struct new_sa_defrag_extent *new)
2655 struct btrfs_path *path;
2656 struct sa_defrag_extent_backref *backref;
2657 struct sa_defrag_extent_backref *prev = NULL;
2658 struct inode *inode;
2659 struct btrfs_root *root;
2660 struct rb_node *node;
2664 root = BTRFS_I(inode)->root;
2666 path = btrfs_alloc_path();
2670 if (!record_extent_backrefs(path, new)) {
2671 btrfs_free_path(path);
2674 btrfs_release_path(path);
2677 node = rb_first(&new->root);
2680 rb_erase(node, &new->root);
2682 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2684 ret = relink_extent_backref(path, prev, backref);
2697 btrfs_free_path(path);
2699 free_sa_defrag_extent(new);
2701 atomic_dec(&root->fs_info->defrag_running);
2702 wake_up(&root->fs_info->transaction_wait);
2705 static struct new_sa_defrag_extent *
2706 record_old_file_extents(struct inode *inode,
2707 struct btrfs_ordered_extent *ordered)
2709 struct btrfs_root *root = BTRFS_I(inode)->root;
2710 struct btrfs_path *path;
2711 struct btrfs_key key;
2712 struct old_sa_defrag_extent *old;
2713 struct new_sa_defrag_extent *new;
2716 new = kmalloc(sizeof(*new), GFP_NOFS);
2721 new->file_pos = ordered->file_offset;
2722 new->len = ordered->len;
2723 new->bytenr = ordered->start;
2724 new->disk_len = ordered->disk_len;
2725 new->compress_type = ordered->compress_type;
2726 new->root = RB_ROOT;
2727 INIT_LIST_HEAD(&new->head);
2729 path = btrfs_alloc_path();
2733 key.objectid = btrfs_ino(inode);
2734 key.type = BTRFS_EXTENT_DATA_KEY;
2735 key.offset = new->file_pos;
2737 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2740 if (ret > 0 && path->slots[0] > 0)
2743 /* find out all the old extents for the file range */
2745 struct btrfs_file_extent_item *extent;
2746 struct extent_buffer *l;
2755 slot = path->slots[0];
2757 if (slot >= btrfs_header_nritems(l)) {
2758 ret = btrfs_next_leaf(root, path);
2766 btrfs_item_key_to_cpu(l, &key, slot);
2768 if (key.objectid != btrfs_ino(inode))
2770 if (key.type != BTRFS_EXTENT_DATA_KEY)
2772 if (key.offset >= new->file_pos + new->len)
2775 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2777 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2778 if (key.offset + num_bytes < new->file_pos)
2781 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2785 extent_offset = btrfs_file_extent_offset(l, extent);
2787 old = kmalloc(sizeof(*old), GFP_NOFS);
2791 offset = max(new->file_pos, key.offset);
2792 end = min(new->file_pos + new->len, key.offset + num_bytes);
2794 old->bytenr = disk_bytenr;
2795 old->extent_offset = extent_offset;
2796 old->offset = offset - key.offset;
2797 old->len = end - offset;
2800 list_add_tail(&old->list, &new->head);
2806 btrfs_free_path(path);
2807 atomic_inc(&root->fs_info->defrag_running);
2812 btrfs_free_path(path);
2814 free_sa_defrag_extent(new);
2818 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2821 struct btrfs_block_group_cache *cache;
2823 cache = btrfs_lookup_block_group(root->fs_info, start);
2826 spin_lock(&cache->lock);
2827 cache->delalloc_bytes -= len;
2828 spin_unlock(&cache->lock);
2830 btrfs_put_block_group(cache);
2833 /* as ordered data IO finishes, this gets called so we can finish
2834 * an ordered extent if the range of bytes in the file it covers are
2837 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2839 struct inode *inode = ordered_extent->inode;
2840 struct btrfs_root *root = BTRFS_I(inode)->root;
2841 struct btrfs_trans_handle *trans = NULL;
2842 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2843 struct extent_state *cached_state = NULL;
2844 struct new_sa_defrag_extent *new = NULL;
2845 int compress_type = 0;
2847 u64 logical_len = ordered_extent->len;
2849 bool truncated = false;
2851 nolock = btrfs_is_free_space_inode(inode);
2853 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2858 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2859 ordered_extent->file_offset +
2860 ordered_extent->len - 1);
2862 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2864 logical_len = ordered_extent->truncated_len;
2865 /* Truncated the entire extent, don't bother adding */
2870 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2871 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2874 * For mwrite(mmap + memset to write) case, we still reserve
2875 * space for NOCOW range.
2876 * As NOCOW won't cause a new delayed ref, just free the space
2878 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2879 ordered_extent->len);
2880 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2882 trans = btrfs_join_transaction_nolock(root);
2884 trans = btrfs_join_transaction(root);
2885 if (IS_ERR(trans)) {
2886 ret = PTR_ERR(trans);
2890 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2891 ret = btrfs_update_inode_fallback(trans, root, inode);
2892 if (ret) /* -ENOMEM or corruption */
2893 btrfs_abort_transaction(trans, root, ret);
2897 lock_extent_bits(io_tree, ordered_extent->file_offset,
2898 ordered_extent->file_offset + ordered_extent->len - 1,
2901 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2902 ordered_extent->file_offset + ordered_extent->len - 1,
2903 EXTENT_DEFRAG, 1, cached_state);
2905 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2906 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2907 /* the inode is shared */
2908 new = record_old_file_extents(inode, ordered_extent);
2910 clear_extent_bit(io_tree, ordered_extent->file_offset,
2911 ordered_extent->file_offset + ordered_extent->len - 1,
2912 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2916 trans = btrfs_join_transaction_nolock(root);
2918 trans = btrfs_join_transaction(root);
2919 if (IS_ERR(trans)) {
2920 ret = PTR_ERR(trans);
2925 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2927 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2928 compress_type = ordered_extent->compress_type;
2929 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2930 BUG_ON(compress_type);
2931 ret = btrfs_mark_extent_written(trans, inode,
2932 ordered_extent->file_offset,
2933 ordered_extent->file_offset +
2936 BUG_ON(root == root->fs_info->tree_root);
2937 ret = insert_reserved_file_extent(trans, inode,
2938 ordered_extent->file_offset,
2939 ordered_extent->start,
2940 ordered_extent->disk_len,
2941 logical_len, logical_len,
2942 compress_type, 0, 0,
2943 BTRFS_FILE_EXTENT_REG);
2945 btrfs_release_delalloc_bytes(root,
2946 ordered_extent->start,
2947 ordered_extent->disk_len);
2949 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2950 ordered_extent->file_offset, ordered_extent->len,
2953 btrfs_abort_transaction(trans, root, ret);
2957 add_pending_csums(trans, inode, ordered_extent->file_offset,
2958 &ordered_extent->list);
2960 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2961 ret = btrfs_update_inode_fallback(trans, root, inode);
2962 if (ret) { /* -ENOMEM or corruption */
2963 btrfs_abort_transaction(trans, root, ret);
2968 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2969 ordered_extent->file_offset +
2970 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2972 if (root != root->fs_info->tree_root)
2973 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2975 btrfs_end_transaction(trans, root);
2977 if (ret || truncated) {
2981 start = ordered_extent->file_offset + logical_len;
2983 start = ordered_extent->file_offset;
2984 end = ordered_extent->file_offset + ordered_extent->len - 1;
2985 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2987 /* Drop the cache for the part of the extent we didn't write. */
2988 btrfs_drop_extent_cache(inode, start, end, 0);
2991 * If the ordered extent had an IOERR or something else went
2992 * wrong we need to return the space for this ordered extent
2993 * back to the allocator. We only free the extent in the
2994 * truncated case if we didn't write out the extent at all.
2996 if ((ret || !logical_len) &&
2997 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2998 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2999 btrfs_free_reserved_extent(root, ordered_extent->start,
3000 ordered_extent->disk_len, 1);
3005 * This needs to be done to make sure anybody waiting knows we are done
3006 * updating everything for this ordered extent.
3008 btrfs_remove_ordered_extent(inode, ordered_extent);
3010 /* for snapshot-aware defrag */
3013 free_sa_defrag_extent(new);
3014 atomic_dec(&root->fs_info->defrag_running);
3016 relink_file_extents(new);
3021 btrfs_put_ordered_extent(ordered_extent);
3022 /* once for the tree */
3023 btrfs_put_ordered_extent(ordered_extent);
3028 static void finish_ordered_fn(struct btrfs_work *work)
3030 struct btrfs_ordered_extent *ordered_extent;
3031 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3032 btrfs_finish_ordered_io(ordered_extent);
3035 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3036 struct extent_state *state, int uptodate)
3038 struct inode *inode = page->mapping->host;
3039 struct btrfs_root *root = BTRFS_I(inode)->root;
3040 struct btrfs_ordered_extent *ordered_extent = NULL;
3041 struct btrfs_workqueue *wq;
3042 btrfs_work_func_t func;
3044 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3046 ClearPagePrivate2(page);
3047 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3048 end - start + 1, uptodate))
3051 if (btrfs_is_free_space_inode(inode)) {
3052 wq = root->fs_info->endio_freespace_worker;
3053 func = btrfs_freespace_write_helper;
3055 wq = root->fs_info->endio_write_workers;
3056 func = btrfs_endio_write_helper;
3059 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3061 btrfs_queue_work(wq, &ordered_extent->work);
3066 static int __readpage_endio_check(struct inode *inode,
3067 struct btrfs_io_bio *io_bio,
3068 int icsum, struct page *page,
3069 int pgoff, u64 start, size_t len)
3075 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3077 kaddr = kmap_atomic(page);
3078 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3079 btrfs_csum_final(csum, (char *)&csum);
3080 if (csum != csum_expected)
3083 kunmap_atomic(kaddr);
3086 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3087 "csum failed ino %llu off %llu csum %u expected csum %u",
3088 btrfs_ino(inode), start, csum, csum_expected);
3089 memset(kaddr + pgoff, 1, len);
3090 flush_dcache_page(page);
3091 kunmap_atomic(kaddr);
3092 if (csum_expected == 0)
3098 * when reads are done, we need to check csums to verify the data is correct
3099 * if there's a match, we allow the bio to finish. If not, the code in
3100 * extent_io.c will try to find good copies for us.
3102 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3103 u64 phy_offset, struct page *page,
3104 u64 start, u64 end, int mirror)
3106 size_t offset = start - page_offset(page);
3107 struct inode *inode = page->mapping->host;
3108 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3109 struct btrfs_root *root = BTRFS_I(inode)->root;
3111 if (PageChecked(page)) {
3112 ClearPageChecked(page);
3116 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3119 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3120 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3121 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3125 phy_offset >>= inode->i_sb->s_blocksize_bits;
3126 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3127 start, (size_t)(end - start + 1));
3130 void btrfs_add_delayed_iput(struct inode *inode)
3132 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3133 struct btrfs_inode *binode = BTRFS_I(inode);
3135 if (atomic_add_unless(&inode->i_count, -1, 1))
3138 spin_lock(&fs_info->delayed_iput_lock);
3139 if (binode->delayed_iput_count == 0) {
3140 ASSERT(list_empty(&binode->delayed_iput));
3141 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3143 binode->delayed_iput_count++;
3145 spin_unlock(&fs_info->delayed_iput_lock);
3148 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3150 struct btrfs_fs_info *fs_info = root->fs_info;
3152 spin_lock(&fs_info->delayed_iput_lock);
3153 while (!list_empty(&fs_info->delayed_iputs)) {
3154 struct btrfs_inode *inode;
3156 inode = list_first_entry(&fs_info->delayed_iputs,
3157 struct btrfs_inode, delayed_iput);
3158 if (inode->delayed_iput_count) {
3159 inode->delayed_iput_count--;
3160 list_move_tail(&inode->delayed_iput,
3161 &fs_info->delayed_iputs);
3163 list_del_init(&inode->delayed_iput);
3165 spin_unlock(&fs_info->delayed_iput_lock);
3166 iput(&inode->vfs_inode);
3167 spin_lock(&fs_info->delayed_iput_lock);
3169 spin_unlock(&fs_info->delayed_iput_lock);
3173 * This is called in transaction commit time. If there are no orphan
3174 * files in the subvolume, it removes orphan item and frees block_rsv
3177 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3178 struct btrfs_root *root)
3180 struct btrfs_block_rsv *block_rsv;
3183 if (atomic_read(&root->orphan_inodes) ||
3184 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3187 spin_lock(&root->orphan_lock);
3188 if (atomic_read(&root->orphan_inodes)) {
3189 spin_unlock(&root->orphan_lock);
3193 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3194 spin_unlock(&root->orphan_lock);
3198 block_rsv = root->orphan_block_rsv;
3199 root->orphan_block_rsv = NULL;
3200 spin_unlock(&root->orphan_lock);
3202 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3203 btrfs_root_refs(&root->root_item) > 0) {
3204 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3205 root->root_key.objectid);
3207 btrfs_abort_transaction(trans, root, ret);
3209 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3214 WARN_ON(block_rsv->size > 0);
3215 btrfs_free_block_rsv(root, block_rsv);
3220 * This creates an orphan entry for the given inode in case something goes
3221 * wrong in the middle of an unlink/truncate.
3223 * NOTE: caller of this function should reserve 5 units of metadata for
3226 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3228 struct btrfs_root *root = BTRFS_I(inode)->root;
3229 struct btrfs_block_rsv *block_rsv = NULL;
3234 if (!root->orphan_block_rsv) {
3235 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3240 spin_lock(&root->orphan_lock);
3241 if (!root->orphan_block_rsv) {
3242 root->orphan_block_rsv = block_rsv;
3243 } else if (block_rsv) {
3244 btrfs_free_block_rsv(root, block_rsv);
3248 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3249 &BTRFS_I(inode)->runtime_flags)) {
3252 * For proper ENOSPC handling, we should do orphan
3253 * cleanup when mounting. But this introduces backward
3254 * compatibility issue.
3256 if (!xchg(&root->orphan_item_inserted, 1))
3262 atomic_inc(&root->orphan_inodes);
3265 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3266 &BTRFS_I(inode)->runtime_flags))
3268 spin_unlock(&root->orphan_lock);
3270 /* grab metadata reservation from transaction handle */
3272 ret = btrfs_orphan_reserve_metadata(trans, inode);
3273 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3276 /* insert an orphan item to track this unlinked/truncated file */
3278 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3280 atomic_dec(&root->orphan_inodes);
3282 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3283 &BTRFS_I(inode)->runtime_flags);
3284 btrfs_orphan_release_metadata(inode);
3286 if (ret != -EEXIST) {
3287 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3288 &BTRFS_I(inode)->runtime_flags);
3289 btrfs_abort_transaction(trans, root, ret);
3296 /* insert an orphan item to track subvolume contains orphan files */
3298 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3299 root->root_key.objectid);
3300 if (ret && ret != -EEXIST) {
3301 btrfs_abort_transaction(trans, root, ret);
3309 * We have done the truncate/delete so we can go ahead and remove the orphan
3310 * item for this particular inode.
3312 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3313 struct inode *inode)
3315 struct btrfs_root *root = BTRFS_I(inode)->root;
3316 int delete_item = 0;
3317 int release_rsv = 0;
3320 spin_lock(&root->orphan_lock);
3321 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3322 &BTRFS_I(inode)->runtime_flags))
3325 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3326 &BTRFS_I(inode)->runtime_flags))
3328 spin_unlock(&root->orphan_lock);
3331 atomic_dec(&root->orphan_inodes);
3333 ret = btrfs_del_orphan_item(trans, root,
3338 btrfs_orphan_release_metadata(inode);
3344 * this cleans up any orphans that may be left on the list from the last use
3347 int btrfs_orphan_cleanup(struct btrfs_root *root)
3349 struct btrfs_path *path;
3350 struct extent_buffer *leaf;
3351 struct btrfs_key key, found_key;
3352 struct btrfs_trans_handle *trans;
3353 struct inode *inode;
3354 u64 last_objectid = 0;
3355 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3357 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3360 path = btrfs_alloc_path();
3365 path->reada = READA_BACK;
3367 key.objectid = BTRFS_ORPHAN_OBJECTID;
3368 key.type = BTRFS_ORPHAN_ITEM_KEY;
3369 key.offset = (u64)-1;
3372 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3377 * if ret == 0 means we found what we were searching for, which
3378 * is weird, but possible, so only screw with path if we didn't
3379 * find the key and see if we have stuff that matches
3383 if (path->slots[0] == 0)
3388 /* pull out the item */
3389 leaf = path->nodes[0];
3390 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3392 /* make sure the item matches what we want */
3393 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3395 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3398 /* release the path since we're done with it */
3399 btrfs_release_path(path);
3402 * this is where we are basically btrfs_lookup, without the
3403 * crossing root thing. we store the inode number in the
3404 * offset of the orphan item.
3407 if (found_key.offset == last_objectid) {
3408 btrfs_err(root->fs_info,
3409 "Error removing orphan entry, stopping orphan cleanup");
3414 last_objectid = found_key.offset;
3416 found_key.objectid = found_key.offset;
3417 found_key.type = BTRFS_INODE_ITEM_KEY;
3418 found_key.offset = 0;
3419 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3420 ret = PTR_ERR_OR_ZERO(inode);
3421 if (ret && ret != -ESTALE)
3424 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3425 struct btrfs_root *dead_root;
3426 struct btrfs_fs_info *fs_info = root->fs_info;
3427 int is_dead_root = 0;
3430 * this is an orphan in the tree root. Currently these
3431 * could come from 2 sources:
3432 * a) a snapshot deletion in progress
3433 * b) a free space cache inode
3434 * We need to distinguish those two, as the snapshot
3435 * orphan must not get deleted.
3436 * find_dead_roots already ran before us, so if this
3437 * is a snapshot deletion, we should find the root
3438 * in the dead_roots list
3440 spin_lock(&fs_info->trans_lock);
3441 list_for_each_entry(dead_root, &fs_info->dead_roots,
3443 if (dead_root->root_key.objectid ==
3444 found_key.objectid) {
3449 spin_unlock(&fs_info->trans_lock);
3451 /* prevent this orphan from being found again */
3452 key.offset = found_key.objectid - 1;
3457 * Inode is already gone but the orphan item is still there,
3458 * kill the orphan item.
3460 if (ret == -ESTALE) {
3461 trans = btrfs_start_transaction(root, 1);
3462 if (IS_ERR(trans)) {
3463 ret = PTR_ERR(trans);
3466 btrfs_debug(root->fs_info, "auto deleting %Lu",
3467 found_key.objectid);
3468 ret = btrfs_del_orphan_item(trans, root,
3469 found_key.objectid);
3470 btrfs_end_transaction(trans, root);
3477 * add this inode to the orphan list so btrfs_orphan_del does
3478 * the proper thing when we hit it
3480 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3481 &BTRFS_I(inode)->runtime_flags);
3482 atomic_inc(&root->orphan_inodes);
3484 /* if we have links, this was a truncate, lets do that */
3485 if (inode->i_nlink) {
3486 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3492 /* 1 for the orphan item deletion. */
3493 trans = btrfs_start_transaction(root, 1);
3494 if (IS_ERR(trans)) {
3496 ret = PTR_ERR(trans);
3499 ret = btrfs_orphan_add(trans, inode);
3500 btrfs_end_transaction(trans, root);
3506 ret = btrfs_truncate(inode);
3508 btrfs_orphan_del(NULL, inode);
3513 /* this will do delete_inode and everything for us */
3518 /* release the path since we're done with it */
3519 btrfs_release_path(path);
3521 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3523 if (root->orphan_block_rsv)
3524 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3527 if (root->orphan_block_rsv ||
3528 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3529 trans = btrfs_join_transaction(root);
3531 btrfs_end_transaction(trans, root);
3535 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3537 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3541 btrfs_err(root->fs_info,
3542 "could not do orphan cleanup %d", ret);
3543 btrfs_free_path(path);
3548 * very simple check to peek ahead in the leaf looking for xattrs. If we
3549 * don't find any xattrs, we know there can't be any acls.
3551 * slot is the slot the inode is in, objectid is the objectid of the inode
3553 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3554 int slot, u64 objectid,
3555 int *first_xattr_slot)
3557 u32 nritems = btrfs_header_nritems(leaf);
3558 struct btrfs_key found_key;
3559 static u64 xattr_access = 0;
3560 static u64 xattr_default = 0;
3563 if (!xattr_access) {
3564 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3565 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3566 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3567 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3571 *first_xattr_slot = -1;
3572 while (slot < nritems) {
3573 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3575 /* we found a different objectid, there must not be acls */
3576 if (found_key.objectid != objectid)
3579 /* we found an xattr, assume we've got an acl */
3580 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3581 if (*first_xattr_slot == -1)
3582 *first_xattr_slot = slot;
3583 if (found_key.offset == xattr_access ||
3584 found_key.offset == xattr_default)
3589 * we found a key greater than an xattr key, there can't
3590 * be any acls later on
3592 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3599 * it goes inode, inode backrefs, xattrs, extents,
3600 * so if there are a ton of hard links to an inode there can
3601 * be a lot of backrefs. Don't waste time searching too hard,
3602 * this is just an optimization
3607 /* we hit the end of the leaf before we found an xattr or
3608 * something larger than an xattr. We have to assume the inode
3611 if (*first_xattr_slot == -1)
3612 *first_xattr_slot = slot;
3617 * read an inode from the btree into the in-memory inode
3619 static void btrfs_read_locked_inode(struct inode *inode)
3621 struct btrfs_path *path;
3622 struct extent_buffer *leaf;
3623 struct btrfs_inode_item *inode_item;
3624 struct btrfs_root *root = BTRFS_I(inode)->root;
3625 struct btrfs_key location;
3630 bool filled = false;
3631 int first_xattr_slot;
3633 ret = btrfs_fill_inode(inode, &rdev);
3637 path = btrfs_alloc_path();
3641 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3643 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3647 leaf = path->nodes[0];
3652 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3653 struct btrfs_inode_item);
3654 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3655 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3656 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3657 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3658 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3660 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3661 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3663 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3664 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3666 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3667 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3669 BTRFS_I(inode)->i_otime.tv_sec =
3670 btrfs_timespec_sec(leaf, &inode_item->otime);
3671 BTRFS_I(inode)->i_otime.tv_nsec =
3672 btrfs_timespec_nsec(leaf, &inode_item->otime);
3674 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3675 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3676 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3678 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3679 inode->i_generation = BTRFS_I(inode)->generation;
3681 rdev = btrfs_inode_rdev(leaf, inode_item);
3683 BTRFS_I(inode)->index_cnt = (u64)-1;
3684 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3688 * If we were modified in the current generation and evicted from memory
3689 * and then re-read we need to do a full sync since we don't have any
3690 * idea about which extents were modified before we were evicted from
3693 * This is required for both inode re-read from disk and delayed inode
3694 * in delayed_nodes_tree.
3696 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3697 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3698 &BTRFS_I(inode)->runtime_flags);
3701 * We don't persist the id of the transaction where an unlink operation
3702 * against the inode was last made. So here we assume the inode might
3703 * have been evicted, and therefore the exact value of last_unlink_trans
3704 * lost, and set it to last_trans to avoid metadata inconsistencies
3705 * between the inode and its parent if the inode is fsync'ed and the log
3706 * replayed. For example, in the scenario:
3709 * ln mydir/foo mydir/bar
3712 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3713 * xfs_io -c fsync mydir/foo
3715 * mount fs, triggers fsync log replay
3717 * We must make sure that when we fsync our inode foo we also log its
3718 * parent inode, otherwise after log replay the parent still has the
3719 * dentry with the "bar" name but our inode foo has a link count of 1
3720 * and doesn't have an inode ref with the name "bar" anymore.
3722 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3723 * but it guarantees correctness at the expense of occasional full
3724 * transaction commits on fsync if our inode is a directory, or if our
3725 * inode is not a directory, logging its parent unnecessarily.
3727 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3730 if (inode->i_nlink != 1 ||
3731 path->slots[0] >= btrfs_header_nritems(leaf))
3734 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3735 if (location.objectid != btrfs_ino(inode))
3738 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3739 if (location.type == BTRFS_INODE_REF_KEY) {
3740 struct btrfs_inode_ref *ref;
3742 ref = (struct btrfs_inode_ref *)ptr;
3743 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3744 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3745 struct btrfs_inode_extref *extref;
3747 extref = (struct btrfs_inode_extref *)ptr;
3748 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3753 * try to precache a NULL acl entry for files that don't have
3754 * any xattrs or acls
3756 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3757 btrfs_ino(inode), &first_xattr_slot);
3758 if (first_xattr_slot != -1) {
3759 path->slots[0] = first_xattr_slot;
3760 ret = btrfs_load_inode_props(inode, path);
3762 btrfs_err(root->fs_info,
3763 "error loading props for ino %llu (root %llu): %d",
3765 root->root_key.objectid, ret);
3767 btrfs_free_path(path);
3770 cache_no_acl(inode);
3772 switch (inode->i_mode & S_IFMT) {
3774 inode->i_mapping->a_ops = &btrfs_aops;
3775 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3776 inode->i_fop = &btrfs_file_operations;
3777 inode->i_op = &btrfs_file_inode_operations;
3780 inode->i_fop = &btrfs_dir_file_operations;
3781 if (root == root->fs_info->tree_root)
3782 inode->i_op = &btrfs_dir_ro_inode_operations;
3784 inode->i_op = &btrfs_dir_inode_operations;
3787 inode->i_op = &btrfs_symlink_inode_operations;
3788 inode_nohighmem(inode);
3789 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3792 inode->i_op = &btrfs_special_inode_operations;
3793 init_special_inode(inode, inode->i_mode, rdev);
3797 btrfs_update_iflags(inode);
3801 btrfs_free_path(path);
3802 make_bad_inode(inode);
3806 * given a leaf and an inode, copy the inode fields into the leaf
3808 static void fill_inode_item(struct btrfs_trans_handle *trans,
3809 struct extent_buffer *leaf,
3810 struct btrfs_inode_item *item,
3811 struct inode *inode)
3813 struct btrfs_map_token token;
3815 btrfs_init_map_token(&token);
3817 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3818 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3819 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3821 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3822 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3824 btrfs_set_token_timespec_sec(leaf, &item->atime,
3825 inode->i_atime.tv_sec, &token);
3826 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3827 inode->i_atime.tv_nsec, &token);
3829 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3830 inode->i_mtime.tv_sec, &token);
3831 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3832 inode->i_mtime.tv_nsec, &token);
3834 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3835 inode->i_ctime.tv_sec, &token);
3836 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3837 inode->i_ctime.tv_nsec, &token);
3839 btrfs_set_token_timespec_sec(leaf, &item->otime,
3840 BTRFS_I(inode)->i_otime.tv_sec, &token);
3841 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3842 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3844 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3846 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3848 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3849 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3850 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3851 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3852 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3856 * copy everything in the in-memory inode into the btree.
3858 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3859 struct btrfs_root *root, struct inode *inode)
3861 struct btrfs_inode_item *inode_item;
3862 struct btrfs_path *path;
3863 struct extent_buffer *leaf;
3866 path = btrfs_alloc_path();
3870 path->leave_spinning = 1;
3871 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3879 leaf = path->nodes[0];
3880 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3881 struct btrfs_inode_item);
3883 fill_inode_item(trans, leaf, inode_item, inode);
3884 btrfs_mark_buffer_dirty(leaf);
3885 btrfs_set_inode_last_trans(trans, inode);
3888 btrfs_free_path(path);
3893 * copy everything in the in-memory inode into the btree.
3895 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3896 struct btrfs_root *root, struct inode *inode)
3901 * If the inode is a free space inode, we can deadlock during commit
3902 * if we put it into the delayed code.
3904 * The data relocation inode should also be directly updated
3907 if (!btrfs_is_free_space_inode(inode)
3908 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3909 && !root->fs_info->log_root_recovering) {
3910 btrfs_update_root_times(trans, root);
3912 ret = btrfs_delayed_update_inode(trans, root, inode);
3914 btrfs_set_inode_last_trans(trans, inode);
3918 return btrfs_update_inode_item(trans, root, inode);
3921 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3922 struct btrfs_root *root,
3923 struct inode *inode)
3927 ret = btrfs_update_inode(trans, root, inode);
3929 return btrfs_update_inode_item(trans, root, inode);
3934 * unlink helper that gets used here in inode.c and in the tree logging
3935 * recovery code. It remove a link in a directory with a given name, and
3936 * also drops the back refs in the inode to the directory
3938 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3939 struct btrfs_root *root,
3940 struct inode *dir, struct inode *inode,
3941 const char *name, int name_len)
3943 struct btrfs_path *path;
3945 struct extent_buffer *leaf;
3946 struct btrfs_dir_item *di;
3947 struct btrfs_key key;
3949 u64 ino = btrfs_ino(inode);
3950 u64 dir_ino = btrfs_ino(dir);
3952 path = btrfs_alloc_path();
3958 path->leave_spinning = 1;
3959 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3960 name, name_len, -1);
3969 leaf = path->nodes[0];
3970 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3971 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3974 btrfs_release_path(path);
3977 * If we don't have dir index, we have to get it by looking up
3978 * the inode ref, since we get the inode ref, remove it directly,
3979 * it is unnecessary to do delayed deletion.
3981 * But if we have dir index, needn't search inode ref to get it.
3982 * Since the inode ref is close to the inode item, it is better
3983 * that we delay to delete it, and just do this deletion when
3984 * we update the inode item.
3986 if (BTRFS_I(inode)->dir_index) {
3987 ret = btrfs_delayed_delete_inode_ref(inode);
3989 index = BTRFS_I(inode)->dir_index;
3994 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3997 btrfs_info(root->fs_info,
3998 "failed to delete reference to %.*s, inode %llu parent %llu",
3999 name_len, name, ino, dir_ino);
4000 btrfs_abort_transaction(trans, root, ret);
4004 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4006 btrfs_abort_transaction(trans, root, ret);
4010 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4012 if (ret != 0 && ret != -ENOENT) {
4013 btrfs_abort_transaction(trans, root, ret);
4017 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4022 btrfs_abort_transaction(trans, root, ret);
4024 btrfs_free_path(path);
4028 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4029 inode_inc_iversion(inode);
4030 inode_inc_iversion(dir);
4031 inode->i_ctime = dir->i_mtime =
4032 dir->i_ctime = current_fs_time(inode->i_sb);
4033 ret = btrfs_update_inode(trans, root, dir);
4038 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4039 struct btrfs_root *root,
4040 struct inode *dir, struct inode *inode,
4041 const char *name, int name_len)
4044 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4047 ret = btrfs_update_inode(trans, root, inode);
4053 * helper to start transaction for unlink and rmdir.
4055 * unlink and rmdir are special in btrfs, they do not always free space, so
4056 * if we cannot make our reservations the normal way try and see if there is
4057 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4058 * allow the unlink to occur.
4060 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4062 struct btrfs_root *root = BTRFS_I(dir)->root;
4065 * 1 for the possible orphan item
4066 * 1 for the dir item
4067 * 1 for the dir index
4068 * 1 for the inode ref
4071 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4074 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4076 struct btrfs_root *root = BTRFS_I(dir)->root;
4077 struct btrfs_trans_handle *trans;
4078 struct inode *inode = d_inode(dentry);
4081 trans = __unlink_start_trans(dir);
4083 return PTR_ERR(trans);
4085 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4087 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4088 dentry->d_name.name, dentry->d_name.len);
4092 if (inode->i_nlink == 0) {
4093 ret = btrfs_orphan_add(trans, inode);
4099 btrfs_end_transaction(trans, root);
4100 btrfs_btree_balance_dirty(root);
4104 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4105 struct btrfs_root *root,
4106 struct inode *dir, u64 objectid,
4107 const char *name, int name_len)
4109 struct btrfs_path *path;
4110 struct extent_buffer *leaf;
4111 struct btrfs_dir_item *di;
4112 struct btrfs_key key;
4115 u64 dir_ino = btrfs_ino(dir);
4117 path = btrfs_alloc_path();
4121 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4122 name, name_len, -1);
4123 if (IS_ERR_OR_NULL(di)) {
4131 leaf = path->nodes[0];
4132 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4133 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4134 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4136 btrfs_abort_transaction(trans, root, ret);
4139 btrfs_release_path(path);
4141 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4142 objectid, root->root_key.objectid,
4143 dir_ino, &index, name, name_len);
4145 if (ret != -ENOENT) {
4146 btrfs_abort_transaction(trans, root, ret);
4149 di = btrfs_search_dir_index_item(root, path, dir_ino,
4151 if (IS_ERR_OR_NULL(di)) {
4156 btrfs_abort_transaction(trans, root, ret);
4160 leaf = path->nodes[0];
4161 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4162 btrfs_release_path(path);
4165 btrfs_release_path(path);
4167 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4169 btrfs_abort_transaction(trans, root, ret);
4173 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4174 inode_inc_iversion(dir);
4175 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4176 ret = btrfs_update_inode_fallback(trans, root, dir);
4178 btrfs_abort_transaction(trans, root, ret);
4180 btrfs_free_path(path);
4184 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4186 struct inode *inode = d_inode(dentry);
4188 struct btrfs_root *root = BTRFS_I(dir)->root;
4189 struct btrfs_trans_handle *trans;
4191 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4193 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4196 trans = __unlink_start_trans(dir);
4198 return PTR_ERR(trans);
4200 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4201 err = btrfs_unlink_subvol(trans, root, dir,
4202 BTRFS_I(inode)->location.objectid,
4203 dentry->d_name.name,
4204 dentry->d_name.len);
4208 err = btrfs_orphan_add(trans, inode);
4212 /* now the directory is empty */
4213 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4214 dentry->d_name.name, dentry->d_name.len);
4216 btrfs_i_size_write(inode, 0);
4218 btrfs_end_transaction(trans, root);
4219 btrfs_btree_balance_dirty(root);
4224 static int truncate_space_check(struct btrfs_trans_handle *trans,
4225 struct btrfs_root *root,
4231 * This is only used to apply pressure to the enospc system, we don't
4232 * intend to use this reservation at all.
4234 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4235 bytes_deleted *= root->nodesize;
4236 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4237 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4239 trace_btrfs_space_reservation(root->fs_info, "transaction",
4242 trans->bytes_reserved += bytes_deleted;
4248 static int truncate_inline_extent(struct inode *inode,
4249 struct btrfs_path *path,
4250 struct btrfs_key *found_key,
4254 struct extent_buffer *leaf = path->nodes[0];
4255 int slot = path->slots[0];
4256 struct btrfs_file_extent_item *fi;
4257 u32 size = (u32)(new_size - found_key->offset);
4258 struct btrfs_root *root = BTRFS_I(inode)->root;
4260 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4262 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4263 loff_t offset = new_size;
4264 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4267 * Zero out the remaining of the last page of our inline extent,
4268 * instead of directly truncating our inline extent here - that
4269 * would be much more complex (decompressing all the data, then
4270 * compressing the truncated data, which might be bigger than
4271 * the size of the inline extent, resize the extent, etc).
4272 * We release the path because to get the page we might need to
4273 * read the extent item from disk (data not in the page cache).
4275 btrfs_release_path(path);
4276 return btrfs_truncate_block(inode, offset, page_end - offset,
4280 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4281 size = btrfs_file_extent_calc_inline_size(size);
4282 btrfs_truncate_item(root, path, size, 1);
4284 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4285 inode_sub_bytes(inode, item_end + 1 - new_size);
4291 * this can truncate away extent items, csum items and directory items.
4292 * It starts at a high offset and removes keys until it can't find
4293 * any higher than new_size
4295 * csum items that cross the new i_size are truncated to the new size
4298 * min_type is the minimum key type to truncate down to. If set to 0, this
4299 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4301 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4302 struct btrfs_root *root,
4303 struct inode *inode,
4304 u64 new_size, u32 min_type)
4306 struct btrfs_path *path;
4307 struct extent_buffer *leaf;
4308 struct btrfs_file_extent_item *fi;
4309 struct btrfs_key key;
4310 struct btrfs_key found_key;
4311 u64 extent_start = 0;
4312 u64 extent_num_bytes = 0;
4313 u64 extent_offset = 0;
4315 u64 last_size = new_size;
4316 u32 found_type = (u8)-1;
4319 int pending_del_nr = 0;
4320 int pending_del_slot = 0;
4321 int extent_type = -1;
4324 u64 ino = btrfs_ino(inode);
4325 u64 bytes_deleted = 0;
4327 bool should_throttle = 0;
4328 bool should_end = 0;
4330 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4333 * for non-free space inodes and ref cows, we want to back off from
4336 if (!btrfs_is_free_space_inode(inode) &&
4337 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4340 path = btrfs_alloc_path();
4343 path->reada = READA_BACK;
4346 * We want to drop from the next block forward in case this new size is
4347 * not block aligned since we will be keeping the last block of the
4348 * extent just the way it is.
4350 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4351 root == root->fs_info->tree_root)
4352 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4353 root->sectorsize), (u64)-1, 0);
4356 * This function is also used to drop the items in the log tree before
4357 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4358 * it is used to drop the loged items. So we shouldn't kill the delayed
4361 if (min_type == 0 && root == BTRFS_I(inode)->root)
4362 btrfs_kill_delayed_inode_items(inode);
4365 key.offset = (u64)-1;
4370 * with a 16K leaf size and 128MB extents, you can actually queue
4371 * up a huge file in a single leaf. Most of the time that
4372 * bytes_deleted is > 0, it will be huge by the time we get here
4374 if (be_nice && bytes_deleted > SZ_32M) {
4375 if (btrfs_should_end_transaction(trans, root)) {
4382 path->leave_spinning = 1;
4383 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4390 /* there are no items in the tree for us to truncate, we're
4393 if (path->slots[0] == 0)
4400 leaf = path->nodes[0];
4401 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4402 found_type = found_key.type;
4404 if (found_key.objectid != ino)
4407 if (found_type < min_type)
4410 item_end = found_key.offset;
4411 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4412 fi = btrfs_item_ptr(leaf, path->slots[0],
4413 struct btrfs_file_extent_item);
4414 extent_type = btrfs_file_extent_type(leaf, fi);
4415 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4417 btrfs_file_extent_num_bytes(leaf, fi);
4418 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4419 item_end += btrfs_file_extent_inline_len(leaf,
4420 path->slots[0], fi);
4424 if (found_type > min_type) {
4427 if (item_end < new_size)
4429 if (found_key.offset >= new_size)
4435 /* FIXME, shrink the extent if the ref count is only 1 */
4436 if (found_type != BTRFS_EXTENT_DATA_KEY)
4440 last_size = found_key.offset;
4442 last_size = new_size;
4444 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4446 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4448 u64 orig_num_bytes =
4449 btrfs_file_extent_num_bytes(leaf, fi);
4450 extent_num_bytes = ALIGN(new_size -
4453 btrfs_set_file_extent_num_bytes(leaf, fi,
4455 num_dec = (orig_num_bytes -
4457 if (test_bit(BTRFS_ROOT_REF_COWS,
4460 inode_sub_bytes(inode, num_dec);
4461 btrfs_mark_buffer_dirty(leaf);
4464 btrfs_file_extent_disk_num_bytes(leaf,
4466 extent_offset = found_key.offset -
4467 btrfs_file_extent_offset(leaf, fi);
4469 /* FIXME blocksize != 4096 */
4470 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4471 if (extent_start != 0) {
4473 if (test_bit(BTRFS_ROOT_REF_COWS,
4475 inode_sub_bytes(inode, num_dec);
4478 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4480 * we can't truncate inline items that have had
4484 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4485 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4488 * Need to release path in order to truncate a
4489 * compressed extent. So delete any accumulated
4490 * extent items so far.
4492 if (btrfs_file_extent_compression(leaf, fi) !=
4493 BTRFS_COMPRESS_NONE && pending_del_nr) {
4494 err = btrfs_del_items(trans, root, path,
4498 btrfs_abort_transaction(trans,
4506 err = truncate_inline_extent(inode, path,
4511 btrfs_abort_transaction(trans,
4515 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4517 inode_sub_bytes(inode, item_end + 1 - new_size);
4522 if (!pending_del_nr) {
4523 /* no pending yet, add ourselves */
4524 pending_del_slot = path->slots[0];
4526 } else if (pending_del_nr &&
4527 path->slots[0] + 1 == pending_del_slot) {
4528 /* hop on the pending chunk */
4530 pending_del_slot = path->slots[0];
4537 should_throttle = 0;
4540 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4541 root == root->fs_info->tree_root)) {
4542 btrfs_set_path_blocking(path);
4543 bytes_deleted += extent_num_bytes;
4544 ret = btrfs_free_extent(trans, root, extent_start,
4545 extent_num_bytes, 0,
4546 btrfs_header_owner(leaf),
4547 ino, extent_offset);
4549 if (btrfs_should_throttle_delayed_refs(trans, root))
4550 btrfs_async_run_delayed_refs(root,
4551 trans->delayed_ref_updates * 2, 0);
4553 if (truncate_space_check(trans, root,
4554 extent_num_bytes)) {
4557 if (btrfs_should_throttle_delayed_refs(trans,
4559 should_throttle = 1;
4564 if (found_type == BTRFS_INODE_ITEM_KEY)
4567 if (path->slots[0] == 0 ||
4568 path->slots[0] != pending_del_slot ||
4569 should_throttle || should_end) {
4570 if (pending_del_nr) {
4571 ret = btrfs_del_items(trans, root, path,
4575 btrfs_abort_transaction(trans,
4581 btrfs_release_path(path);
4582 if (should_throttle) {
4583 unsigned long updates = trans->delayed_ref_updates;
4585 trans->delayed_ref_updates = 0;
4586 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4592 * if we failed to refill our space rsv, bail out
4593 * and let the transaction restart
4605 if (pending_del_nr) {
4606 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4609 btrfs_abort_transaction(trans, root, ret);
4612 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4613 btrfs_ordered_update_i_size(inode, last_size, NULL);
4615 btrfs_free_path(path);
4617 if (be_nice && bytes_deleted > SZ_32M) {
4618 unsigned long updates = trans->delayed_ref_updates;
4620 trans->delayed_ref_updates = 0;
4621 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4630 * btrfs_truncate_block - read, zero a chunk and write a block
4631 * @inode - inode that we're zeroing
4632 * @from - the offset to start zeroing
4633 * @len - the length to zero, 0 to zero the entire range respective to the
4635 * @front - zero up to the offset instead of from the offset on
4637 * This will find the block for the "from" offset and cow the block and zero the
4638 * part we want to zero. This is used with truncate and hole punching.
4640 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4643 struct address_space *mapping = inode->i_mapping;
4644 struct btrfs_root *root = BTRFS_I(inode)->root;
4645 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4646 struct btrfs_ordered_extent *ordered;
4647 struct extent_state *cached_state = NULL;
4649 u32 blocksize = root->sectorsize;
4650 pgoff_t index = from >> PAGE_SHIFT;
4651 unsigned offset = from & (blocksize - 1);
4653 gfp_t mask = btrfs_alloc_write_mask(mapping);
4658 if ((offset & (blocksize - 1)) == 0 &&
4659 (!len || ((len & (blocksize - 1)) == 0)))
4662 ret = btrfs_delalloc_reserve_space(inode,
4663 round_down(from, blocksize), blocksize);
4668 page = find_or_create_page(mapping, index, mask);
4670 btrfs_delalloc_release_space(inode,
4671 round_down(from, blocksize),
4677 block_start = round_down(from, blocksize);
4678 block_end = block_start + blocksize - 1;
4680 if (!PageUptodate(page)) {
4681 ret = btrfs_readpage(NULL, page);
4683 if (page->mapping != mapping) {
4688 if (!PageUptodate(page)) {
4693 wait_on_page_writeback(page);
4695 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4696 set_page_extent_mapped(page);
4698 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4700 unlock_extent_cached(io_tree, block_start, block_end,
4701 &cached_state, GFP_NOFS);
4704 btrfs_start_ordered_extent(inode, ordered, 1);
4705 btrfs_put_ordered_extent(ordered);
4709 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4710 EXTENT_DIRTY | EXTENT_DELALLOC |
4711 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4712 0, 0, &cached_state, GFP_NOFS);
4714 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4717 unlock_extent_cached(io_tree, block_start, block_end,
4718 &cached_state, GFP_NOFS);
4722 if (offset != blocksize) {
4724 len = blocksize - offset;
4727 memset(kaddr + (block_start - page_offset(page)),
4730 memset(kaddr + (block_start - page_offset(page)) + offset,
4732 flush_dcache_page(page);
4735 ClearPageChecked(page);
4736 set_page_dirty(page);
4737 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4742 btrfs_delalloc_release_space(inode, block_start,
4750 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4751 u64 offset, u64 len)
4753 struct btrfs_trans_handle *trans;
4757 * Still need to make sure the inode looks like it's been updated so
4758 * that any holes get logged if we fsync.
4760 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4761 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4762 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4763 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4768 * 1 - for the one we're dropping
4769 * 1 - for the one we're adding
4770 * 1 - for updating the inode.
4772 trans = btrfs_start_transaction(root, 3);
4774 return PTR_ERR(trans);
4776 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4778 btrfs_abort_transaction(trans, root, ret);
4779 btrfs_end_transaction(trans, root);
4783 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4784 0, 0, len, 0, len, 0, 0, 0);
4786 btrfs_abort_transaction(trans, root, ret);
4788 btrfs_update_inode(trans, root, inode);
4789 btrfs_end_transaction(trans, root);
4794 * This function puts in dummy file extents for the area we're creating a hole
4795 * for. So if we are truncating this file to a larger size we need to insert
4796 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4797 * the range between oldsize and size
4799 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4801 struct btrfs_root *root = BTRFS_I(inode)->root;
4802 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4803 struct extent_map *em = NULL;
4804 struct extent_state *cached_state = NULL;
4805 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4806 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4807 u64 block_end = ALIGN(size, root->sectorsize);
4814 * If our size started in the middle of a block we need to zero out the
4815 * rest of the block before we expand the i_size, otherwise we could
4816 * expose stale data.
4818 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4822 if (size <= hole_start)
4826 struct btrfs_ordered_extent *ordered;
4828 lock_extent_bits(io_tree, hole_start, block_end - 1,
4830 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4831 block_end - hole_start);
4834 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4835 &cached_state, GFP_NOFS);
4836 btrfs_start_ordered_extent(inode, ordered, 1);
4837 btrfs_put_ordered_extent(ordered);
4840 cur_offset = hole_start;
4842 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4843 block_end - cur_offset, 0);
4849 last_byte = min(extent_map_end(em), block_end);
4850 last_byte = ALIGN(last_byte , root->sectorsize);
4851 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4852 struct extent_map *hole_em;
4853 hole_size = last_byte - cur_offset;
4855 err = maybe_insert_hole(root, inode, cur_offset,
4859 btrfs_drop_extent_cache(inode, cur_offset,
4860 cur_offset + hole_size - 1, 0);
4861 hole_em = alloc_extent_map();
4863 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4864 &BTRFS_I(inode)->runtime_flags);
4867 hole_em->start = cur_offset;
4868 hole_em->len = hole_size;
4869 hole_em->orig_start = cur_offset;
4871 hole_em->block_start = EXTENT_MAP_HOLE;
4872 hole_em->block_len = 0;
4873 hole_em->orig_block_len = 0;
4874 hole_em->ram_bytes = hole_size;
4875 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4876 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4877 hole_em->generation = root->fs_info->generation;
4880 write_lock(&em_tree->lock);
4881 err = add_extent_mapping(em_tree, hole_em, 1);
4882 write_unlock(&em_tree->lock);
4885 btrfs_drop_extent_cache(inode, cur_offset,
4889 free_extent_map(hole_em);
4892 free_extent_map(em);
4894 cur_offset = last_byte;
4895 if (cur_offset >= block_end)
4898 free_extent_map(em);
4899 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4904 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4906 struct btrfs_root *root = BTRFS_I(inode)->root;
4907 struct btrfs_trans_handle *trans;
4908 loff_t oldsize = i_size_read(inode);
4909 loff_t newsize = attr->ia_size;
4910 int mask = attr->ia_valid;
4914 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4915 * special case where we need to update the times despite not having
4916 * these flags set. For all other operations the VFS set these flags
4917 * explicitly if it wants a timestamp update.
4919 if (newsize != oldsize) {
4920 inode_inc_iversion(inode);
4921 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4922 inode->i_ctime = inode->i_mtime =
4923 current_fs_time(inode->i_sb);
4926 if (newsize > oldsize) {
4928 * Don't do an expanding truncate while snapshoting is ongoing.
4929 * This is to ensure the snapshot captures a fully consistent
4930 * state of this file - if the snapshot captures this expanding
4931 * truncation, it must capture all writes that happened before
4934 btrfs_wait_for_snapshot_creation(root);
4935 ret = btrfs_cont_expand(inode, oldsize, newsize);
4937 btrfs_end_write_no_snapshoting(root);
4941 trans = btrfs_start_transaction(root, 1);
4942 if (IS_ERR(trans)) {
4943 btrfs_end_write_no_snapshoting(root);
4944 return PTR_ERR(trans);
4947 i_size_write(inode, newsize);
4948 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4949 pagecache_isize_extended(inode, oldsize, newsize);
4950 ret = btrfs_update_inode(trans, root, inode);
4951 btrfs_end_write_no_snapshoting(root);
4952 btrfs_end_transaction(trans, root);
4956 * We're truncating a file that used to have good data down to
4957 * zero. Make sure it gets into the ordered flush list so that
4958 * any new writes get down to disk quickly.
4961 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4962 &BTRFS_I(inode)->runtime_flags);
4965 * 1 for the orphan item we're going to add
4966 * 1 for the orphan item deletion.
4968 trans = btrfs_start_transaction(root, 2);
4970 return PTR_ERR(trans);
4973 * We need to do this in case we fail at _any_ point during the
4974 * actual truncate. Once we do the truncate_setsize we could
4975 * invalidate pages which forces any outstanding ordered io to
4976 * be instantly completed which will give us extents that need
4977 * to be truncated. If we fail to get an orphan inode down we
4978 * could have left over extents that were never meant to live,
4979 * so we need to guarantee from this point on that everything
4980 * will be consistent.
4982 ret = btrfs_orphan_add(trans, inode);
4983 btrfs_end_transaction(trans, root);
4987 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4988 truncate_setsize(inode, newsize);
4990 /* Disable nonlocked read DIO to avoid the end less truncate */
4991 btrfs_inode_block_unlocked_dio(inode);
4992 inode_dio_wait(inode);
4993 btrfs_inode_resume_unlocked_dio(inode);
4995 ret = btrfs_truncate(inode);
4996 if (ret && inode->i_nlink) {
5000 * failed to truncate, disk_i_size is only adjusted down
5001 * as we remove extents, so it should represent the true
5002 * size of the inode, so reset the in memory size and
5003 * delete our orphan entry.
5005 trans = btrfs_join_transaction(root);
5006 if (IS_ERR(trans)) {
5007 btrfs_orphan_del(NULL, inode);
5010 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5011 err = btrfs_orphan_del(trans, inode);
5013 btrfs_abort_transaction(trans, root, err);
5014 btrfs_end_transaction(trans, root);
5021 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5023 struct inode *inode = d_inode(dentry);
5024 struct btrfs_root *root = BTRFS_I(inode)->root;
5027 if (btrfs_root_readonly(root))
5030 err = inode_change_ok(inode, attr);
5034 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5035 err = btrfs_setsize(inode, attr);
5040 if (attr->ia_valid) {
5041 setattr_copy(inode, attr);
5042 inode_inc_iversion(inode);
5043 err = btrfs_dirty_inode(inode);
5045 if (!err && attr->ia_valid & ATTR_MODE)
5046 err = posix_acl_chmod(inode, inode->i_mode);
5053 * While truncating the inode pages during eviction, we get the VFS calling
5054 * btrfs_invalidatepage() against each page of the inode. This is slow because
5055 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5056 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5057 * extent_state structures over and over, wasting lots of time.
5059 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5060 * those expensive operations on a per page basis and do only the ordered io
5061 * finishing, while we release here the extent_map and extent_state structures,
5062 * without the excessive merging and splitting.
5064 static void evict_inode_truncate_pages(struct inode *inode)
5066 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5067 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5068 struct rb_node *node;
5070 ASSERT(inode->i_state & I_FREEING);
5071 truncate_inode_pages_final(&inode->i_data);
5073 write_lock(&map_tree->lock);
5074 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5075 struct extent_map *em;
5077 node = rb_first(&map_tree->map);
5078 em = rb_entry(node, struct extent_map, rb_node);
5079 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5080 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5081 remove_extent_mapping(map_tree, em);
5082 free_extent_map(em);
5083 if (need_resched()) {
5084 write_unlock(&map_tree->lock);
5086 write_lock(&map_tree->lock);
5089 write_unlock(&map_tree->lock);
5092 * Keep looping until we have no more ranges in the io tree.
5093 * We can have ongoing bios started by readpages (called from readahead)
5094 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5095 * still in progress (unlocked the pages in the bio but did not yet
5096 * unlocked the ranges in the io tree). Therefore this means some
5097 * ranges can still be locked and eviction started because before
5098 * submitting those bios, which are executed by a separate task (work
5099 * queue kthread), inode references (inode->i_count) were not taken
5100 * (which would be dropped in the end io callback of each bio).
5101 * Therefore here we effectively end up waiting for those bios and
5102 * anyone else holding locked ranges without having bumped the inode's
5103 * reference count - if we don't do it, when they access the inode's
5104 * io_tree to unlock a range it may be too late, leading to an
5105 * use-after-free issue.
5107 spin_lock(&io_tree->lock);
5108 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5109 struct extent_state *state;
5110 struct extent_state *cached_state = NULL;
5114 node = rb_first(&io_tree->state);
5115 state = rb_entry(node, struct extent_state, rb_node);
5116 start = state->start;
5118 spin_unlock(&io_tree->lock);
5120 lock_extent_bits(io_tree, start, end, &cached_state);
5123 * If still has DELALLOC flag, the extent didn't reach disk,
5124 * and its reserved space won't be freed by delayed_ref.
5125 * So we need to free its reserved space here.
5126 * (Refer to comment in btrfs_invalidatepage, case 2)
5128 * Note, end is the bytenr of last byte, so we need + 1 here.
5130 if (state->state & EXTENT_DELALLOC)
5131 btrfs_qgroup_free_data(inode, start, end - start + 1);
5133 clear_extent_bit(io_tree, start, end,
5134 EXTENT_LOCKED | EXTENT_DIRTY |
5135 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5136 EXTENT_DEFRAG, 1, 1,
5137 &cached_state, GFP_NOFS);
5140 spin_lock(&io_tree->lock);
5142 spin_unlock(&io_tree->lock);
5145 void btrfs_evict_inode(struct inode *inode)
5147 struct btrfs_trans_handle *trans;
5148 struct btrfs_root *root = BTRFS_I(inode)->root;
5149 struct btrfs_block_rsv *rsv, *global_rsv;
5150 int steal_from_global = 0;
5151 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5154 trace_btrfs_inode_evict(inode);
5156 evict_inode_truncate_pages(inode);
5158 if (inode->i_nlink &&
5159 ((btrfs_root_refs(&root->root_item) != 0 &&
5160 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5161 btrfs_is_free_space_inode(inode)))
5164 if (is_bad_inode(inode)) {
5165 btrfs_orphan_del(NULL, inode);
5168 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5169 if (!special_file(inode->i_mode))
5170 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5172 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5174 if (root->fs_info->log_root_recovering) {
5175 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5176 &BTRFS_I(inode)->runtime_flags));
5180 if (inode->i_nlink > 0) {
5181 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5182 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5186 ret = btrfs_commit_inode_delayed_inode(inode);
5188 btrfs_orphan_del(NULL, inode);
5192 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5194 btrfs_orphan_del(NULL, inode);
5197 rsv->size = min_size;
5199 global_rsv = &root->fs_info->global_block_rsv;
5201 btrfs_i_size_write(inode, 0);
5204 * This is a bit simpler than btrfs_truncate since we've already
5205 * reserved our space for our orphan item in the unlink, so we just
5206 * need to reserve some slack space in case we add bytes and update
5207 * inode item when doing the truncate.
5210 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5211 BTRFS_RESERVE_FLUSH_LIMIT);
5214 * Try and steal from the global reserve since we will
5215 * likely not use this space anyway, we want to try as
5216 * hard as possible to get this to work.
5219 steal_from_global++;
5221 steal_from_global = 0;
5225 * steal_from_global == 0: we reserved stuff, hooray!
5226 * steal_from_global == 1: we didn't reserve stuff, boo!
5227 * steal_from_global == 2: we've committed, still not a lot of
5228 * room but maybe we'll have room in the global reserve this
5230 * steal_from_global == 3: abandon all hope!
5232 if (steal_from_global > 2) {
5233 btrfs_warn(root->fs_info,
5234 "Could not get space for a delete, will truncate on mount %d",
5236 btrfs_orphan_del(NULL, inode);
5237 btrfs_free_block_rsv(root, rsv);
5241 trans = btrfs_join_transaction(root);
5242 if (IS_ERR(trans)) {
5243 btrfs_orphan_del(NULL, inode);
5244 btrfs_free_block_rsv(root, rsv);
5249 * We can't just steal from the global reserve, we need to make
5250 * sure there is room to do it, if not we need to commit and try
5253 if (steal_from_global) {
5254 if (!btrfs_check_space_for_delayed_refs(trans, root))
5255 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5262 * Couldn't steal from the global reserve, we have too much
5263 * pending stuff built up, commit the transaction and try it
5267 ret = btrfs_commit_transaction(trans, root);
5269 btrfs_orphan_del(NULL, inode);
5270 btrfs_free_block_rsv(root, rsv);
5275 steal_from_global = 0;
5278 trans->block_rsv = rsv;
5280 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5281 if (ret != -ENOSPC && ret != -EAGAIN)
5284 trans->block_rsv = &root->fs_info->trans_block_rsv;
5285 btrfs_end_transaction(trans, root);
5287 btrfs_btree_balance_dirty(root);
5290 btrfs_free_block_rsv(root, rsv);
5293 * Errors here aren't a big deal, it just means we leave orphan items
5294 * in the tree. They will be cleaned up on the next mount.
5297 trans->block_rsv = root->orphan_block_rsv;
5298 btrfs_orphan_del(trans, inode);
5300 btrfs_orphan_del(NULL, inode);
5303 trans->block_rsv = &root->fs_info->trans_block_rsv;
5304 if (!(root == root->fs_info->tree_root ||
5305 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5306 btrfs_return_ino(root, btrfs_ino(inode));
5308 btrfs_end_transaction(trans, root);
5309 btrfs_btree_balance_dirty(root);
5311 btrfs_remove_delayed_node(inode);
5316 * this returns the key found in the dir entry in the location pointer.
5317 * If no dir entries were found, location->objectid is 0.
5319 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5320 struct btrfs_key *location)
5322 const char *name = dentry->d_name.name;
5323 int namelen = dentry->d_name.len;
5324 struct btrfs_dir_item *di;
5325 struct btrfs_path *path;
5326 struct btrfs_root *root = BTRFS_I(dir)->root;
5329 path = btrfs_alloc_path();
5333 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5338 if (IS_ERR_OR_NULL(di))
5341 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5343 btrfs_free_path(path);
5346 location->objectid = 0;
5351 * when we hit a tree root in a directory, the btrfs part of the inode
5352 * needs to be changed to reflect the root directory of the tree root. This
5353 * is kind of like crossing a mount point.
5355 static int fixup_tree_root_location(struct btrfs_root *root,
5357 struct dentry *dentry,
5358 struct btrfs_key *location,
5359 struct btrfs_root **sub_root)
5361 struct btrfs_path *path;
5362 struct btrfs_root *new_root;
5363 struct btrfs_root_ref *ref;
5364 struct extent_buffer *leaf;
5365 struct btrfs_key key;
5369 path = btrfs_alloc_path();
5376 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5377 key.type = BTRFS_ROOT_REF_KEY;
5378 key.offset = location->objectid;
5380 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5388 leaf = path->nodes[0];
5389 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5390 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5391 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5394 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5395 (unsigned long)(ref + 1),
5396 dentry->d_name.len);
5400 btrfs_release_path(path);
5402 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5403 if (IS_ERR(new_root)) {
5404 err = PTR_ERR(new_root);
5408 *sub_root = new_root;
5409 location->objectid = btrfs_root_dirid(&new_root->root_item);
5410 location->type = BTRFS_INODE_ITEM_KEY;
5411 location->offset = 0;
5414 btrfs_free_path(path);
5418 static void inode_tree_add(struct inode *inode)
5420 struct btrfs_root *root = BTRFS_I(inode)->root;
5421 struct btrfs_inode *entry;
5423 struct rb_node *parent;
5424 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5425 u64 ino = btrfs_ino(inode);
5427 if (inode_unhashed(inode))
5430 spin_lock(&root->inode_lock);
5431 p = &root->inode_tree.rb_node;
5434 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5436 if (ino < btrfs_ino(&entry->vfs_inode))
5437 p = &parent->rb_left;
5438 else if (ino > btrfs_ino(&entry->vfs_inode))
5439 p = &parent->rb_right;
5441 WARN_ON(!(entry->vfs_inode.i_state &
5442 (I_WILL_FREE | I_FREEING)));
5443 rb_replace_node(parent, new, &root->inode_tree);
5444 RB_CLEAR_NODE(parent);
5445 spin_unlock(&root->inode_lock);
5449 rb_link_node(new, parent, p);
5450 rb_insert_color(new, &root->inode_tree);
5451 spin_unlock(&root->inode_lock);
5454 static void inode_tree_del(struct inode *inode)
5456 struct btrfs_root *root = BTRFS_I(inode)->root;
5459 spin_lock(&root->inode_lock);
5460 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5461 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5462 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5463 empty = RB_EMPTY_ROOT(&root->inode_tree);
5465 spin_unlock(&root->inode_lock);
5467 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5468 synchronize_srcu(&root->fs_info->subvol_srcu);
5469 spin_lock(&root->inode_lock);
5470 empty = RB_EMPTY_ROOT(&root->inode_tree);
5471 spin_unlock(&root->inode_lock);
5473 btrfs_add_dead_root(root);
5477 void btrfs_invalidate_inodes(struct btrfs_root *root)
5479 struct rb_node *node;
5480 struct rb_node *prev;
5481 struct btrfs_inode *entry;
5482 struct inode *inode;
5485 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5486 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5488 spin_lock(&root->inode_lock);
5490 node = root->inode_tree.rb_node;
5494 entry = rb_entry(node, struct btrfs_inode, rb_node);
5496 if (objectid < btrfs_ino(&entry->vfs_inode))
5497 node = node->rb_left;
5498 else if (objectid > btrfs_ino(&entry->vfs_inode))
5499 node = node->rb_right;
5505 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5506 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5510 prev = rb_next(prev);
5514 entry = rb_entry(node, struct btrfs_inode, rb_node);
5515 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5516 inode = igrab(&entry->vfs_inode);
5518 spin_unlock(&root->inode_lock);
5519 if (atomic_read(&inode->i_count) > 1)
5520 d_prune_aliases(inode);
5522 * btrfs_drop_inode will have it removed from
5523 * the inode cache when its usage count
5528 spin_lock(&root->inode_lock);
5532 if (cond_resched_lock(&root->inode_lock))
5535 node = rb_next(node);
5537 spin_unlock(&root->inode_lock);
5540 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5542 struct btrfs_iget_args *args = p;
5543 inode->i_ino = args->location->objectid;
5544 memcpy(&BTRFS_I(inode)->location, args->location,
5545 sizeof(*args->location));
5546 BTRFS_I(inode)->root = args->root;
5550 static int btrfs_find_actor(struct inode *inode, void *opaque)
5552 struct btrfs_iget_args *args = opaque;
5553 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5554 args->root == BTRFS_I(inode)->root;
5557 static struct inode *btrfs_iget_locked(struct super_block *s,
5558 struct btrfs_key *location,
5559 struct btrfs_root *root)
5561 struct inode *inode;
5562 struct btrfs_iget_args args;
5563 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5565 args.location = location;
5568 inode = iget5_locked(s, hashval, btrfs_find_actor,
5569 btrfs_init_locked_inode,
5574 /* Get an inode object given its location and corresponding root.
5575 * Returns in *is_new if the inode was read from disk
5577 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5578 struct btrfs_root *root, int *new)
5580 struct inode *inode;
5582 inode = btrfs_iget_locked(s, location, root);
5584 return ERR_PTR(-ENOMEM);
5586 if (inode->i_state & I_NEW) {
5587 btrfs_read_locked_inode(inode);
5588 if (!is_bad_inode(inode)) {
5589 inode_tree_add(inode);
5590 unlock_new_inode(inode);
5594 unlock_new_inode(inode);
5596 inode = ERR_PTR(-ESTALE);
5603 static struct inode *new_simple_dir(struct super_block *s,
5604 struct btrfs_key *key,
5605 struct btrfs_root *root)
5607 struct inode *inode = new_inode(s);
5610 return ERR_PTR(-ENOMEM);
5612 BTRFS_I(inode)->root = root;
5613 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5614 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5616 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5617 inode->i_op = &btrfs_dir_ro_inode_operations;
5618 inode->i_fop = &simple_dir_operations;
5619 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5620 inode->i_mtime = current_fs_time(inode->i_sb);
5621 inode->i_atime = inode->i_mtime;
5622 inode->i_ctime = inode->i_mtime;
5623 BTRFS_I(inode)->i_otime = inode->i_mtime;
5628 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5630 struct inode *inode;
5631 struct btrfs_root *root = BTRFS_I(dir)->root;
5632 struct btrfs_root *sub_root = root;
5633 struct btrfs_key location;
5637 if (dentry->d_name.len > BTRFS_NAME_LEN)
5638 return ERR_PTR(-ENAMETOOLONG);
5640 ret = btrfs_inode_by_name(dir, dentry, &location);
5642 return ERR_PTR(ret);
5644 if (location.objectid == 0)
5645 return ERR_PTR(-ENOENT);
5647 if (location.type == BTRFS_INODE_ITEM_KEY) {
5648 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5652 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5654 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5655 ret = fixup_tree_root_location(root, dir, dentry,
5656 &location, &sub_root);
5659 inode = ERR_PTR(ret);
5661 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5663 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5665 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5667 if (!IS_ERR(inode) && root != sub_root) {
5668 down_read(&root->fs_info->cleanup_work_sem);
5669 if (!(inode->i_sb->s_flags & MS_RDONLY))
5670 ret = btrfs_orphan_cleanup(sub_root);
5671 up_read(&root->fs_info->cleanup_work_sem);
5674 inode = ERR_PTR(ret);
5681 static int btrfs_dentry_delete(const struct dentry *dentry)
5683 struct btrfs_root *root;
5684 struct inode *inode = d_inode(dentry);
5686 if (!inode && !IS_ROOT(dentry))
5687 inode = d_inode(dentry->d_parent);
5690 root = BTRFS_I(inode)->root;
5691 if (btrfs_root_refs(&root->root_item) == 0)
5694 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5700 static void btrfs_dentry_release(struct dentry *dentry)
5702 kfree(dentry->d_fsdata);
5705 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5708 struct inode *inode;
5710 inode = btrfs_lookup_dentry(dir, dentry);
5711 if (IS_ERR(inode)) {
5712 if (PTR_ERR(inode) == -ENOENT)
5715 return ERR_CAST(inode);
5718 return d_splice_alias(inode, dentry);
5721 unsigned char btrfs_filetype_table[] = {
5722 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5725 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5727 struct inode *inode = file_inode(file);
5728 struct btrfs_root *root = BTRFS_I(inode)->root;
5729 struct btrfs_item *item;
5730 struct btrfs_dir_item *di;
5731 struct btrfs_key key;
5732 struct btrfs_key found_key;
5733 struct btrfs_path *path;
5734 struct list_head ins_list;
5735 struct list_head del_list;
5737 struct extent_buffer *leaf;
5739 unsigned char d_type;
5744 int key_type = BTRFS_DIR_INDEX_KEY;
5748 int is_curr = 0; /* ctx->pos points to the current index? */
5751 /* FIXME, use a real flag for deciding about the key type */
5752 if (root->fs_info->tree_root == root)
5753 key_type = BTRFS_DIR_ITEM_KEY;
5755 if (!dir_emit_dots(file, ctx))
5758 path = btrfs_alloc_path();
5762 path->reada = READA_FORWARD;
5764 if (key_type == BTRFS_DIR_INDEX_KEY) {
5765 INIT_LIST_HEAD(&ins_list);
5766 INIT_LIST_HEAD(&del_list);
5767 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5770 key.type = key_type;
5771 key.offset = ctx->pos;
5772 key.objectid = btrfs_ino(inode);
5774 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5780 leaf = path->nodes[0];
5781 slot = path->slots[0];
5782 if (slot >= btrfs_header_nritems(leaf)) {
5783 ret = btrfs_next_leaf(root, path);
5791 item = btrfs_item_nr(slot);
5792 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5794 if (found_key.objectid != key.objectid)
5796 if (found_key.type != key_type)
5798 if (found_key.offset < ctx->pos)
5800 if (key_type == BTRFS_DIR_INDEX_KEY &&
5801 btrfs_should_delete_dir_index(&del_list,
5805 ctx->pos = found_key.offset;
5808 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5810 di_total = btrfs_item_size(leaf, item);
5812 while (di_cur < di_total) {
5813 struct btrfs_key location;
5815 if (verify_dir_item(root, leaf, di))
5818 name_len = btrfs_dir_name_len(leaf, di);
5819 if (name_len <= sizeof(tmp_name)) {
5820 name_ptr = tmp_name;
5822 name_ptr = kmalloc(name_len, GFP_KERNEL);
5828 read_extent_buffer(leaf, name_ptr,
5829 (unsigned long)(di + 1), name_len);
5831 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5832 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5835 /* is this a reference to our own snapshot? If so
5838 * In contrast to old kernels, we insert the snapshot's
5839 * dir item and dir index after it has been created, so
5840 * we won't find a reference to our own snapshot. We
5841 * still keep the following code for backward
5844 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5845 location.objectid == root->root_key.objectid) {
5849 over = !dir_emit(ctx, name_ptr, name_len,
5850 location.objectid, d_type);
5853 if (name_ptr != tmp_name)
5859 di_len = btrfs_dir_name_len(leaf, di) +
5860 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5862 di = (struct btrfs_dir_item *)((char *)di + di_len);
5868 if (key_type == BTRFS_DIR_INDEX_KEY) {
5871 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5877 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5878 * it was was set to the termination value in previous call. We assume
5879 * that "." and ".." were emitted if we reach this point and set the
5880 * termination value as well for an empty directory.
5882 if (ctx->pos > 2 && !emitted)
5885 /* Reached end of directory/root. Bump pos past the last item. */
5889 * Stop new entries from being returned after we return the last
5892 * New directory entries are assigned a strictly increasing
5893 * offset. This means that new entries created during readdir
5894 * are *guaranteed* to be seen in the future by that readdir.
5895 * This has broken buggy programs which operate on names as
5896 * they're returned by readdir. Until we re-use freed offsets
5897 * we have this hack to stop new entries from being returned
5898 * under the assumption that they'll never reach this huge
5901 * This is being careful not to overflow 32bit loff_t unless the
5902 * last entry requires it because doing so has broken 32bit apps
5905 if (key_type == BTRFS_DIR_INDEX_KEY) {
5906 if (ctx->pos >= INT_MAX)
5907 ctx->pos = LLONG_MAX;
5914 if (key_type == BTRFS_DIR_INDEX_KEY)
5915 btrfs_put_delayed_items(&ins_list, &del_list);
5916 btrfs_free_path(path);
5920 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5922 struct btrfs_root *root = BTRFS_I(inode)->root;
5923 struct btrfs_trans_handle *trans;
5925 bool nolock = false;
5927 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5930 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5933 if (wbc->sync_mode == WB_SYNC_ALL) {
5935 trans = btrfs_join_transaction_nolock(root);
5937 trans = btrfs_join_transaction(root);
5939 return PTR_ERR(trans);
5940 ret = btrfs_commit_transaction(trans, root);
5946 * This is somewhat expensive, updating the tree every time the
5947 * inode changes. But, it is most likely to find the inode in cache.
5948 * FIXME, needs more benchmarking...there are no reasons other than performance
5949 * to keep or drop this code.
5951 static int btrfs_dirty_inode(struct inode *inode)
5953 struct btrfs_root *root = BTRFS_I(inode)->root;
5954 struct btrfs_trans_handle *trans;
5957 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5960 trans = btrfs_join_transaction(root);
5962 return PTR_ERR(trans);
5964 ret = btrfs_update_inode(trans, root, inode);
5965 if (ret && ret == -ENOSPC) {
5966 /* whoops, lets try again with the full transaction */
5967 btrfs_end_transaction(trans, root);
5968 trans = btrfs_start_transaction(root, 1);
5970 return PTR_ERR(trans);
5972 ret = btrfs_update_inode(trans, root, inode);
5974 btrfs_end_transaction(trans, root);
5975 if (BTRFS_I(inode)->delayed_node)
5976 btrfs_balance_delayed_items(root);
5982 * This is a copy of file_update_time. We need this so we can return error on
5983 * ENOSPC for updating the inode in the case of file write and mmap writes.
5985 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5988 struct btrfs_root *root = BTRFS_I(inode)->root;
5990 if (btrfs_root_readonly(root))
5993 if (flags & S_VERSION)
5994 inode_inc_iversion(inode);
5995 if (flags & S_CTIME)
5996 inode->i_ctime = *now;
5997 if (flags & S_MTIME)
5998 inode->i_mtime = *now;
5999 if (flags & S_ATIME)
6000 inode->i_atime = *now;
6001 return btrfs_dirty_inode(inode);
6005 * find the highest existing sequence number in a directory
6006 * and then set the in-memory index_cnt variable to reflect
6007 * free sequence numbers
6009 static int btrfs_set_inode_index_count(struct inode *inode)
6011 struct btrfs_root *root = BTRFS_I(inode)->root;
6012 struct btrfs_key key, found_key;
6013 struct btrfs_path *path;
6014 struct extent_buffer *leaf;
6017 key.objectid = btrfs_ino(inode);
6018 key.type = BTRFS_DIR_INDEX_KEY;
6019 key.offset = (u64)-1;
6021 path = btrfs_alloc_path();
6025 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6028 /* FIXME: we should be able to handle this */
6034 * MAGIC NUMBER EXPLANATION:
6035 * since we search a directory based on f_pos we have to start at 2
6036 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6037 * else has to start at 2
6039 if (path->slots[0] == 0) {
6040 BTRFS_I(inode)->index_cnt = 2;
6046 leaf = path->nodes[0];
6047 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6049 if (found_key.objectid != btrfs_ino(inode) ||
6050 found_key.type != BTRFS_DIR_INDEX_KEY) {
6051 BTRFS_I(inode)->index_cnt = 2;
6055 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6057 btrfs_free_path(path);
6062 * helper to find a free sequence number in a given directory. This current
6063 * code is very simple, later versions will do smarter things in the btree
6065 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6069 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6070 ret = btrfs_inode_delayed_dir_index_count(dir);
6072 ret = btrfs_set_inode_index_count(dir);
6078 *index = BTRFS_I(dir)->index_cnt;
6079 BTRFS_I(dir)->index_cnt++;
6084 static int btrfs_insert_inode_locked(struct inode *inode)
6086 struct btrfs_iget_args args;
6087 args.location = &BTRFS_I(inode)->location;
6088 args.root = BTRFS_I(inode)->root;
6090 return insert_inode_locked4(inode,
6091 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6092 btrfs_find_actor, &args);
6095 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6096 struct btrfs_root *root,
6098 const char *name, int name_len,
6099 u64 ref_objectid, u64 objectid,
6100 umode_t mode, u64 *index)
6102 struct inode *inode;
6103 struct btrfs_inode_item *inode_item;
6104 struct btrfs_key *location;
6105 struct btrfs_path *path;
6106 struct btrfs_inode_ref *ref;
6107 struct btrfs_key key[2];
6109 int nitems = name ? 2 : 1;
6113 path = btrfs_alloc_path();
6115 return ERR_PTR(-ENOMEM);
6117 inode = new_inode(root->fs_info->sb);
6119 btrfs_free_path(path);
6120 return ERR_PTR(-ENOMEM);
6124 * O_TMPFILE, set link count to 0, so that after this point,
6125 * we fill in an inode item with the correct link count.
6128 set_nlink(inode, 0);
6131 * we have to initialize this early, so we can reclaim the inode
6132 * number if we fail afterwards in this function.
6134 inode->i_ino = objectid;
6137 trace_btrfs_inode_request(dir);
6139 ret = btrfs_set_inode_index(dir, index);
6141 btrfs_free_path(path);
6143 return ERR_PTR(ret);
6149 * index_cnt is ignored for everything but a dir,
6150 * btrfs_get_inode_index_count has an explanation for the magic
6153 BTRFS_I(inode)->index_cnt = 2;
6154 BTRFS_I(inode)->dir_index = *index;
6155 BTRFS_I(inode)->root = root;
6156 BTRFS_I(inode)->generation = trans->transid;
6157 inode->i_generation = BTRFS_I(inode)->generation;
6160 * We could have gotten an inode number from somebody who was fsynced
6161 * and then removed in this same transaction, so let's just set full
6162 * sync since it will be a full sync anyway and this will blow away the
6163 * old info in the log.
6165 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6167 key[0].objectid = objectid;
6168 key[0].type = BTRFS_INODE_ITEM_KEY;
6171 sizes[0] = sizeof(struct btrfs_inode_item);
6175 * Start new inodes with an inode_ref. This is slightly more
6176 * efficient for small numbers of hard links since they will
6177 * be packed into one item. Extended refs will kick in if we
6178 * add more hard links than can fit in the ref item.
6180 key[1].objectid = objectid;
6181 key[1].type = BTRFS_INODE_REF_KEY;
6182 key[1].offset = ref_objectid;
6184 sizes[1] = name_len + sizeof(*ref);
6187 location = &BTRFS_I(inode)->location;
6188 location->objectid = objectid;
6189 location->offset = 0;
6190 location->type = BTRFS_INODE_ITEM_KEY;
6192 ret = btrfs_insert_inode_locked(inode);
6196 path->leave_spinning = 1;
6197 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6201 inode_init_owner(inode, dir, mode);
6202 inode_set_bytes(inode, 0);
6204 inode->i_mtime = current_fs_time(inode->i_sb);
6205 inode->i_atime = inode->i_mtime;
6206 inode->i_ctime = inode->i_mtime;
6207 BTRFS_I(inode)->i_otime = inode->i_mtime;
6209 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6210 struct btrfs_inode_item);
6211 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6212 sizeof(*inode_item));
6213 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6216 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6217 struct btrfs_inode_ref);
6218 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6219 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6220 ptr = (unsigned long)(ref + 1);
6221 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6224 btrfs_mark_buffer_dirty(path->nodes[0]);
6225 btrfs_free_path(path);
6227 btrfs_inherit_iflags(inode, dir);
6229 if (S_ISREG(mode)) {
6230 if (btrfs_test_opt(root, NODATASUM))
6231 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6232 if (btrfs_test_opt(root, NODATACOW))
6233 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6234 BTRFS_INODE_NODATASUM;
6237 inode_tree_add(inode);
6239 trace_btrfs_inode_new(inode);
6240 btrfs_set_inode_last_trans(trans, inode);
6242 btrfs_update_root_times(trans, root);
6244 ret = btrfs_inode_inherit_props(trans, inode, dir);
6246 btrfs_err(root->fs_info,
6247 "error inheriting props for ino %llu (root %llu): %d",
6248 btrfs_ino(inode), root->root_key.objectid, ret);
6253 unlock_new_inode(inode);
6256 BTRFS_I(dir)->index_cnt--;
6257 btrfs_free_path(path);
6259 return ERR_PTR(ret);
6262 static inline u8 btrfs_inode_type(struct inode *inode)
6264 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6268 * utility function to add 'inode' into 'parent_inode' with
6269 * a give name and a given sequence number.
6270 * if 'add_backref' is true, also insert a backref from the
6271 * inode to the parent directory.
6273 int btrfs_add_link(struct btrfs_trans_handle *trans,
6274 struct inode *parent_inode, struct inode *inode,
6275 const char *name, int name_len, int add_backref, u64 index)
6278 struct btrfs_key key;
6279 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6280 u64 ino = btrfs_ino(inode);
6281 u64 parent_ino = btrfs_ino(parent_inode);
6283 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6284 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6287 key.type = BTRFS_INODE_ITEM_KEY;
6291 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6292 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6293 key.objectid, root->root_key.objectid,
6294 parent_ino, index, name, name_len);
6295 } else if (add_backref) {
6296 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6300 /* Nothing to clean up yet */
6304 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6306 btrfs_inode_type(inode), index);
6307 if (ret == -EEXIST || ret == -EOVERFLOW)
6310 btrfs_abort_transaction(trans, root, ret);
6314 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6316 inode_inc_iversion(parent_inode);
6317 parent_inode->i_mtime = parent_inode->i_ctime =
6318 current_fs_time(parent_inode->i_sb);
6319 ret = btrfs_update_inode(trans, root, parent_inode);
6321 btrfs_abort_transaction(trans, root, ret);
6325 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6328 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6329 key.objectid, root->root_key.objectid,
6330 parent_ino, &local_index, name, name_len);
6332 } else if (add_backref) {
6336 err = btrfs_del_inode_ref(trans, root, name, name_len,
6337 ino, parent_ino, &local_index);
6342 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6343 struct inode *dir, struct dentry *dentry,
6344 struct inode *inode, int backref, u64 index)
6346 int err = btrfs_add_link(trans, dir, inode,
6347 dentry->d_name.name, dentry->d_name.len,
6354 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6355 umode_t mode, dev_t rdev)
6357 struct btrfs_trans_handle *trans;
6358 struct btrfs_root *root = BTRFS_I(dir)->root;
6359 struct inode *inode = NULL;
6366 * 2 for inode item and ref
6368 * 1 for xattr if selinux is on
6370 trans = btrfs_start_transaction(root, 5);
6372 return PTR_ERR(trans);
6374 err = btrfs_find_free_ino(root, &objectid);
6378 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6379 dentry->d_name.len, btrfs_ino(dir), objectid,
6381 if (IS_ERR(inode)) {
6382 err = PTR_ERR(inode);
6387 * If the active LSM wants to access the inode during
6388 * d_instantiate it needs these. Smack checks to see
6389 * if the filesystem supports xattrs by looking at the
6392 inode->i_op = &btrfs_special_inode_operations;
6393 init_special_inode(inode, inode->i_mode, rdev);
6395 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6397 goto out_unlock_inode;
6399 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6401 goto out_unlock_inode;
6403 btrfs_update_inode(trans, root, inode);
6404 unlock_new_inode(inode);
6405 d_instantiate(dentry, inode);
6409 btrfs_end_transaction(trans, root);
6410 btrfs_balance_delayed_items(root);
6411 btrfs_btree_balance_dirty(root);
6413 inode_dec_link_count(inode);
6420 unlock_new_inode(inode);
6425 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6426 umode_t mode, bool excl)
6428 struct btrfs_trans_handle *trans;
6429 struct btrfs_root *root = BTRFS_I(dir)->root;
6430 struct inode *inode = NULL;
6431 int drop_inode_on_err = 0;
6437 * 2 for inode item and ref
6439 * 1 for xattr if selinux is on
6441 trans = btrfs_start_transaction(root, 5);
6443 return PTR_ERR(trans);
6445 err = btrfs_find_free_ino(root, &objectid);
6449 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6450 dentry->d_name.len, btrfs_ino(dir), objectid,
6452 if (IS_ERR(inode)) {
6453 err = PTR_ERR(inode);
6456 drop_inode_on_err = 1;
6458 * If the active LSM wants to access the inode during
6459 * d_instantiate it needs these. Smack checks to see
6460 * if the filesystem supports xattrs by looking at the
6463 inode->i_fop = &btrfs_file_operations;
6464 inode->i_op = &btrfs_file_inode_operations;
6465 inode->i_mapping->a_ops = &btrfs_aops;
6467 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6469 goto out_unlock_inode;
6471 err = btrfs_update_inode(trans, root, inode);
6473 goto out_unlock_inode;
6475 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6477 goto out_unlock_inode;
6479 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6480 unlock_new_inode(inode);
6481 d_instantiate(dentry, inode);
6484 btrfs_end_transaction(trans, root);
6485 if (err && drop_inode_on_err) {
6486 inode_dec_link_count(inode);
6489 btrfs_balance_delayed_items(root);
6490 btrfs_btree_balance_dirty(root);
6494 unlock_new_inode(inode);
6499 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6500 struct dentry *dentry)
6502 struct btrfs_trans_handle *trans = NULL;
6503 struct btrfs_root *root = BTRFS_I(dir)->root;
6504 struct inode *inode = d_inode(old_dentry);
6509 /* do not allow sys_link's with other subvols of the same device */
6510 if (root->objectid != BTRFS_I(inode)->root->objectid)
6513 if (inode->i_nlink >= BTRFS_LINK_MAX)
6516 err = btrfs_set_inode_index(dir, &index);
6521 * 2 items for inode and inode ref
6522 * 2 items for dir items
6523 * 1 item for parent inode
6525 trans = btrfs_start_transaction(root, 5);
6526 if (IS_ERR(trans)) {
6527 err = PTR_ERR(trans);
6532 /* There are several dir indexes for this inode, clear the cache. */
6533 BTRFS_I(inode)->dir_index = 0ULL;
6535 inode_inc_iversion(inode);
6536 inode->i_ctime = current_fs_time(inode->i_sb);
6538 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6540 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6545 struct dentry *parent = dentry->d_parent;
6546 err = btrfs_update_inode(trans, root, inode);
6549 if (inode->i_nlink == 1) {
6551 * If new hard link count is 1, it's a file created
6552 * with open(2) O_TMPFILE flag.
6554 err = btrfs_orphan_del(trans, inode);
6558 d_instantiate(dentry, inode);
6559 btrfs_log_new_name(trans, inode, NULL, parent);
6562 btrfs_balance_delayed_items(root);
6565 btrfs_end_transaction(trans, root);
6567 inode_dec_link_count(inode);
6570 btrfs_btree_balance_dirty(root);
6574 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6576 struct inode *inode = NULL;
6577 struct btrfs_trans_handle *trans;
6578 struct btrfs_root *root = BTRFS_I(dir)->root;
6580 int drop_on_err = 0;
6585 * 2 items for inode and ref
6586 * 2 items for dir items
6587 * 1 for xattr if selinux is on
6589 trans = btrfs_start_transaction(root, 5);
6591 return PTR_ERR(trans);
6593 err = btrfs_find_free_ino(root, &objectid);
6597 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6598 dentry->d_name.len, btrfs_ino(dir), objectid,
6599 S_IFDIR | mode, &index);
6600 if (IS_ERR(inode)) {
6601 err = PTR_ERR(inode);
6606 /* these must be set before we unlock the inode */
6607 inode->i_op = &btrfs_dir_inode_operations;
6608 inode->i_fop = &btrfs_dir_file_operations;
6610 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6612 goto out_fail_inode;
6614 btrfs_i_size_write(inode, 0);
6615 err = btrfs_update_inode(trans, root, inode);
6617 goto out_fail_inode;
6619 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6620 dentry->d_name.len, 0, index);
6622 goto out_fail_inode;
6624 d_instantiate(dentry, inode);
6626 * mkdir is special. We're unlocking after we call d_instantiate
6627 * to avoid a race with nfsd calling d_instantiate.
6629 unlock_new_inode(inode);
6633 btrfs_end_transaction(trans, root);
6635 inode_dec_link_count(inode);
6638 btrfs_balance_delayed_items(root);
6639 btrfs_btree_balance_dirty(root);
6643 unlock_new_inode(inode);
6647 /* Find next extent map of a given extent map, caller needs to ensure locks */
6648 static struct extent_map *next_extent_map(struct extent_map *em)
6650 struct rb_node *next;
6652 next = rb_next(&em->rb_node);
6655 return container_of(next, struct extent_map, rb_node);
6658 static struct extent_map *prev_extent_map(struct extent_map *em)
6660 struct rb_node *prev;
6662 prev = rb_prev(&em->rb_node);
6665 return container_of(prev, struct extent_map, rb_node);
6668 /* helper for btfs_get_extent. Given an existing extent in the tree,
6669 * the existing extent is the nearest extent to map_start,
6670 * and an extent that you want to insert, deal with overlap and insert
6671 * the best fitted new extent into the tree.
6673 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6674 struct extent_map *existing,
6675 struct extent_map *em,
6678 struct extent_map *prev;
6679 struct extent_map *next;
6684 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6686 if (existing->start > map_start) {
6688 prev = prev_extent_map(next);
6691 next = next_extent_map(prev);
6694 start = prev ? extent_map_end(prev) : em->start;
6695 start = max_t(u64, start, em->start);
6696 end = next ? next->start : extent_map_end(em);
6697 end = min_t(u64, end, extent_map_end(em));
6698 start_diff = start - em->start;
6700 em->len = end - start;
6701 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6702 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6703 em->block_start += start_diff;
6704 em->block_len -= start_diff;
6706 return add_extent_mapping(em_tree, em, 0);
6709 static noinline int uncompress_inline(struct btrfs_path *path,
6711 size_t pg_offset, u64 extent_offset,
6712 struct btrfs_file_extent_item *item)
6715 struct extent_buffer *leaf = path->nodes[0];
6718 unsigned long inline_size;
6722 WARN_ON(pg_offset != 0);
6723 compress_type = btrfs_file_extent_compression(leaf, item);
6724 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6725 inline_size = btrfs_file_extent_inline_item_len(leaf,
6726 btrfs_item_nr(path->slots[0]));
6727 tmp = kmalloc(inline_size, GFP_NOFS);
6730 ptr = btrfs_file_extent_inline_start(item);
6732 read_extent_buffer(leaf, tmp, ptr, inline_size);
6734 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6735 ret = btrfs_decompress(compress_type, tmp, page,
6736 extent_offset, inline_size, max_size);
6742 * a bit scary, this does extent mapping from logical file offset to the disk.
6743 * the ugly parts come from merging extents from the disk with the in-ram
6744 * representation. This gets more complex because of the data=ordered code,
6745 * where the in-ram extents might be locked pending data=ordered completion.
6747 * This also copies inline extents directly into the page.
6750 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6751 size_t pg_offset, u64 start, u64 len,
6756 u64 extent_start = 0;
6758 u64 objectid = btrfs_ino(inode);
6760 struct btrfs_path *path = NULL;
6761 struct btrfs_root *root = BTRFS_I(inode)->root;
6762 struct btrfs_file_extent_item *item;
6763 struct extent_buffer *leaf;
6764 struct btrfs_key found_key;
6765 struct extent_map *em = NULL;
6766 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6767 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6768 struct btrfs_trans_handle *trans = NULL;
6769 const bool new_inline = !page || create;
6772 read_lock(&em_tree->lock);
6773 em = lookup_extent_mapping(em_tree, start, len);
6775 em->bdev = root->fs_info->fs_devices->latest_bdev;
6776 read_unlock(&em_tree->lock);
6779 if (em->start > start || em->start + em->len <= start)
6780 free_extent_map(em);
6781 else if (em->block_start == EXTENT_MAP_INLINE && page)
6782 free_extent_map(em);
6786 em = alloc_extent_map();
6791 em->bdev = root->fs_info->fs_devices->latest_bdev;
6792 em->start = EXTENT_MAP_HOLE;
6793 em->orig_start = EXTENT_MAP_HOLE;
6795 em->block_len = (u64)-1;
6798 path = btrfs_alloc_path();
6804 * Chances are we'll be called again, so go ahead and do
6807 path->reada = READA_FORWARD;
6810 ret = btrfs_lookup_file_extent(trans, root, path,
6811 objectid, start, trans != NULL);
6818 if (path->slots[0] == 0)
6823 leaf = path->nodes[0];
6824 item = btrfs_item_ptr(leaf, path->slots[0],
6825 struct btrfs_file_extent_item);
6826 /* are we inside the extent that was found? */
6827 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6828 found_type = found_key.type;
6829 if (found_key.objectid != objectid ||
6830 found_type != BTRFS_EXTENT_DATA_KEY) {
6832 * If we backup past the first extent we want to move forward
6833 * and see if there is an extent in front of us, otherwise we'll
6834 * say there is a hole for our whole search range which can
6841 found_type = btrfs_file_extent_type(leaf, item);
6842 extent_start = found_key.offset;
6843 if (found_type == BTRFS_FILE_EXTENT_REG ||
6844 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6845 extent_end = extent_start +
6846 btrfs_file_extent_num_bytes(leaf, item);
6847 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6849 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6850 extent_end = ALIGN(extent_start + size, root->sectorsize);
6853 if (start >= extent_end) {
6855 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6856 ret = btrfs_next_leaf(root, path);
6863 leaf = path->nodes[0];
6865 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6866 if (found_key.objectid != objectid ||
6867 found_key.type != BTRFS_EXTENT_DATA_KEY)
6869 if (start + len <= found_key.offset)
6871 if (start > found_key.offset)
6874 em->orig_start = start;
6875 em->len = found_key.offset - start;
6879 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6881 if (found_type == BTRFS_FILE_EXTENT_REG ||
6882 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6884 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6888 size_t extent_offset;
6894 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6895 extent_offset = page_offset(page) + pg_offset - extent_start;
6896 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6897 size - extent_offset);
6898 em->start = extent_start + extent_offset;
6899 em->len = ALIGN(copy_size, root->sectorsize);
6900 em->orig_block_len = em->len;
6901 em->orig_start = em->start;
6902 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6903 if (create == 0 && !PageUptodate(page)) {
6904 if (btrfs_file_extent_compression(leaf, item) !=
6905 BTRFS_COMPRESS_NONE) {
6906 ret = uncompress_inline(path, page, pg_offset,
6907 extent_offset, item);
6914 read_extent_buffer(leaf, map + pg_offset, ptr,
6916 if (pg_offset + copy_size < PAGE_SIZE) {
6917 memset(map + pg_offset + copy_size, 0,
6918 PAGE_SIZE - pg_offset -
6923 flush_dcache_page(page);
6924 } else if (create && PageUptodate(page)) {
6928 free_extent_map(em);
6931 btrfs_release_path(path);
6932 trans = btrfs_join_transaction(root);
6935 return ERR_CAST(trans);
6939 write_extent_buffer(leaf, map + pg_offset, ptr,
6942 btrfs_mark_buffer_dirty(leaf);
6944 set_extent_uptodate(io_tree, em->start,
6945 extent_map_end(em) - 1, NULL, GFP_NOFS);
6950 em->orig_start = start;
6953 em->block_start = EXTENT_MAP_HOLE;
6954 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6956 btrfs_release_path(path);
6957 if (em->start > start || extent_map_end(em) <= start) {
6958 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6959 em->start, em->len, start, len);
6965 write_lock(&em_tree->lock);
6966 ret = add_extent_mapping(em_tree, em, 0);
6967 /* it is possible that someone inserted the extent into the tree
6968 * while we had the lock dropped. It is also possible that
6969 * an overlapping map exists in the tree
6971 if (ret == -EEXIST) {
6972 struct extent_map *existing;
6976 existing = search_extent_mapping(em_tree, start, len);
6978 * existing will always be non-NULL, since there must be
6979 * extent causing the -EEXIST.
6981 if (existing->start == em->start &&
6982 extent_map_end(existing) == extent_map_end(em) &&
6983 em->block_start == existing->block_start) {
6985 * these two extents are the same, it happens
6986 * with inlines especially
6988 free_extent_map(em);
6992 } else if (start >= extent_map_end(existing) ||
6993 start <= existing->start) {
6995 * The existing extent map is the one nearest to
6996 * the [start, start + len) range which overlaps
6998 err = merge_extent_mapping(em_tree, existing,
7000 free_extent_map(existing);
7002 free_extent_map(em);
7006 free_extent_map(em);
7011 write_unlock(&em_tree->lock);
7014 trace_btrfs_get_extent(root, em);
7016 btrfs_free_path(path);
7018 ret = btrfs_end_transaction(trans, root);
7023 free_extent_map(em);
7024 return ERR_PTR(err);
7026 BUG_ON(!em); /* Error is always set */
7030 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7031 size_t pg_offset, u64 start, u64 len,
7034 struct extent_map *em;
7035 struct extent_map *hole_em = NULL;
7036 u64 range_start = start;
7042 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7049 * - a pre-alloc extent,
7050 * there might actually be delalloc bytes behind it.
7052 if (em->block_start != EXTENT_MAP_HOLE &&
7053 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7059 /* check to see if we've wrapped (len == -1 or similar) */
7068 /* ok, we didn't find anything, lets look for delalloc */
7069 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7070 end, len, EXTENT_DELALLOC, 1);
7071 found_end = range_start + found;
7072 if (found_end < range_start)
7073 found_end = (u64)-1;
7076 * we didn't find anything useful, return
7077 * the original results from get_extent()
7079 if (range_start > end || found_end <= start) {
7085 /* adjust the range_start to make sure it doesn't
7086 * go backwards from the start they passed in
7088 range_start = max(start, range_start);
7089 found = found_end - range_start;
7092 u64 hole_start = start;
7095 em = alloc_extent_map();
7101 * when btrfs_get_extent can't find anything it
7102 * returns one huge hole
7104 * make sure what it found really fits our range, and
7105 * adjust to make sure it is based on the start from
7109 u64 calc_end = extent_map_end(hole_em);
7111 if (calc_end <= start || (hole_em->start > end)) {
7112 free_extent_map(hole_em);
7115 hole_start = max(hole_em->start, start);
7116 hole_len = calc_end - hole_start;
7120 if (hole_em && range_start > hole_start) {
7121 /* our hole starts before our delalloc, so we
7122 * have to return just the parts of the hole
7123 * that go until the delalloc starts
7125 em->len = min(hole_len,
7126 range_start - hole_start);
7127 em->start = hole_start;
7128 em->orig_start = hole_start;
7130 * don't adjust block start at all,
7131 * it is fixed at EXTENT_MAP_HOLE
7133 em->block_start = hole_em->block_start;
7134 em->block_len = hole_len;
7135 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7136 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7138 em->start = range_start;
7140 em->orig_start = range_start;
7141 em->block_start = EXTENT_MAP_DELALLOC;
7142 em->block_len = found;
7144 } else if (hole_em) {
7149 free_extent_map(hole_em);
7151 free_extent_map(em);
7152 return ERR_PTR(err);
7157 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7160 const u64 orig_start,
7161 const u64 block_start,
7162 const u64 block_len,
7163 const u64 orig_block_len,
7164 const u64 ram_bytes,
7167 struct extent_map *em = NULL;
7170 down_read(&BTRFS_I(inode)->dio_sem);
7171 if (type != BTRFS_ORDERED_NOCOW) {
7172 em = create_pinned_em(inode, start, len, orig_start,
7173 block_start, block_len, orig_block_len,
7178 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7179 len, block_len, type);
7182 free_extent_map(em);
7183 btrfs_drop_extent_cache(inode, start,
7184 start + len - 1, 0);
7189 up_read(&BTRFS_I(inode)->dio_sem);
7194 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7197 struct btrfs_root *root = BTRFS_I(inode)->root;
7198 struct extent_map *em;
7199 struct btrfs_key ins;
7203 alloc_hint = get_extent_allocation_hint(inode, start, len);
7204 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7205 alloc_hint, &ins, 1, 1);
7207 return ERR_PTR(ret);
7209 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7210 ins.objectid, ins.offset, ins.offset,
7212 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7214 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7220 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7221 * block must be cow'd
7223 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7224 u64 *orig_start, u64 *orig_block_len,
7227 struct btrfs_trans_handle *trans;
7228 struct btrfs_path *path;
7230 struct extent_buffer *leaf;
7231 struct btrfs_root *root = BTRFS_I(inode)->root;
7232 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7233 struct btrfs_file_extent_item *fi;
7234 struct btrfs_key key;
7241 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7243 path = btrfs_alloc_path();
7247 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7252 slot = path->slots[0];
7255 /* can't find the item, must cow */
7262 leaf = path->nodes[0];
7263 btrfs_item_key_to_cpu(leaf, &key, slot);
7264 if (key.objectid != btrfs_ino(inode) ||
7265 key.type != BTRFS_EXTENT_DATA_KEY) {
7266 /* not our file or wrong item type, must cow */
7270 if (key.offset > offset) {
7271 /* Wrong offset, must cow */
7275 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7276 found_type = btrfs_file_extent_type(leaf, fi);
7277 if (found_type != BTRFS_FILE_EXTENT_REG &&
7278 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7279 /* not a regular extent, must cow */
7283 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7286 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7287 if (extent_end <= offset)
7290 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7291 if (disk_bytenr == 0)
7294 if (btrfs_file_extent_compression(leaf, fi) ||
7295 btrfs_file_extent_encryption(leaf, fi) ||
7296 btrfs_file_extent_other_encoding(leaf, fi))
7299 backref_offset = btrfs_file_extent_offset(leaf, fi);
7302 *orig_start = key.offset - backref_offset;
7303 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7304 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7307 if (btrfs_extent_readonly(root, disk_bytenr))
7310 num_bytes = min(offset + *len, extent_end) - offset;
7311 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7314 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7315 ret = test_range_bit(io_tree, offset, range_end,
7316 EXTENT_DELALLOC, 0, NULL);
7323 btrfs_release_path(path);
7326 * look for other files referencing this extent, if we
7327 * find any we must cow
7329 trans = btrfs_join_transaction(root);
7330 if (IS_ERR(trans)) {
7335 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7336 key.offset - backref_offset, disk_bytenr);
7337 btrfs_end_transaction(trans, root);
7344 * adjust disk_bytenr and num_bytes to cover just the bytes
7345 * in this extent we are about to write. If there
7346 * are any csums in that range we have to cow in order
7347 * to keep the csums correct
7349 disk_bytenr += backref_offset;
7350 disk_bytenr += offset - key.offset;
7351 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7354 * all of the above have passed, it is safe to overwrite this extent
7360 btrfs_free_path(path);
7364 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7366 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7368 void **pagep = NULL;
7369 struct page *page = NULL;
7373 start_idx = start >> PAGE_SHIFT;
7376 * end is the last byte in the last page. end == start is legal
7378 end_idx = end >> PAGE_SHIFT;
7382 /* Most of the code in this while loop is lifted from
7383 * find_get_page. It's been modified to begin searching from a
7384 * page and return just the first page found in that range. If the
7385 * found idx is less than or equal to the end idx then we know that
7386 * a page exists. If no pages are found or if those pages are
7387 * outside of the range then we're fine (yay!) */
7388 while (page == NULL &&
7389 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7390 page = radix_tree_deref_slot(pagep);
7391 if (unlikely(!page))
7394 if (radix_tree_exception(page)) {
7395 if (radix_tree_deref_retry(page)) {
7400 * Otherwise, shmem/tmpfs must be storing a swap entry
7401 * here as an exceptional entry: so return it without
7402 * attempting to raise page count.
7405 break; /* TODO: Is this relevant for this use case? */
7408 if (!page_cache_get_speculative(page)) {
7414 * Has the page moved?
7415 * This is part of the lockless pagecache protocol. See
7416 * include/linux/pagemap.h for details.
7418 if (unlikely(page != *pagep)) {
7425 if (page->index <= end_idx)
7434 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7435 struct extent_state **cached_state, int writing)
7437 struct btrfs_ordered_extent *ordered;
7441 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7444 * We're concerned with the entire range that we're going to be
7445 * doing DIO to, so we need to make sure there's no ordered
7446 * extents in this range.
7448 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7449 lockend - lockstart + 1);
7452 * We need to make sure there are no buffered pages in this
7453 * range either, we could have raced between the invalidate in
7454 * generic_file_direct_write and locking the extent. The
7455 * invalidate needs to happen so that reads after a write do not
7460 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7463 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7464 cached_state, GFP_NOFS);
7468 * If we are doing a DIO read and the ordered extent we
7469 * found is for a buffered write, we can not wait for it
7470 * to complete and retry, because if we do so we can
7471 * deadlock with concurrent buffered writes on page
7472 * locks. This happens only if our DIO read covers more
7473 * than one extent map, if at this point has already
7474 * created an ordered extent for a previous extent map
7475 * and locked its range in the inode's io tree, and a
7476 * concurrent write against that previous extent map's
7477 * range and this range started (we unlock the ranges
7478 * in the io tree only when the bios complete and
7479 * buffered writes always lock pages before attempting
7480 * to lock range in the io tree).
7483 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7484 btrfs_start_ordered_extent(inode, ordered, 1);
7487 btrfs_put_ordered_extent(ordered);
7490 * We could trigger writeback for this range (and wait
7491 * for it to complete) and then invalidate the pages for
7492 * this range (through invalidate_inode_pages2_range()),
7493 * but that can lead us to a deadlock with a concurrent
7494 * call to readpages() (a buffered read or a defrag call
7495 * triggered a readahead) on a page lock due to an
7496 * ordered dio extent we created before but did not have
7497 * yet a corresponding bio submitted (whence it can not
7498 * complete), which makes readpages() wait for that
7499 * ordered extent to complete while holding a lock on
7514 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7515 u64 len, u64 orig_start,
7516 u64 block_start, u64 block_len,
7517 u64 orig_block_len, u64 ram_bytes,
7520 struct extent_map_tree *em_tree;
7521 struct extent_map *em;
7522 struct btrfs_root *root = BTRFS_I(inode)->root;
7525 em_tree = &BTRFS_I(inode)->extent_tree;
7526 em = alloc_extent_map();
7528 return ERR_PTR(-ENOMEM);
7531 em->orig_start = orig_start;
7532 em->mod_start = start;
7535 em->block_len = block_len;
7536 em->block_start = block_start;
7537 em->bdev = root->fs_info->fs_devices->latest_bdev;
7538 em->orig_block_len = orig_block_len;
7539 em->ram_bytes = ram_bytes;
7540 em->generation = -1;
7541 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7542 if (type == BTRFS_ORDERED_PREALLOC)
7543 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7546 btrfs_drop_extent_cache(inode, em->start,
7547 em->start + em->len - 1, 0);
7548 write_lock(&em_tree->lock);
7549 ret = add_extent_mapping(em_tree, em, 1);
7550 write_unlock(&em_tree->lock);
7551 } while (ret == -EEXIST);
7554 free_extent_map(em);
7555 return ERR_PTR(ret);
7561 static void adjust_dio_outstanding_extents(struct inode *inode,
7562 struct btrfs_dio_data *dio_data,
7565 unsigned num_extents;
7567 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7568 BTRFS_MAX_EXTENT_SIZE);
7570 * If we have an outstanding_extents count still set then we're
7571 * within our reservation, otherwise we need to adjust our inode
7572 * counter appropriately.
7574 if (dio_data->outstanding_extents) {
7575 dio_data->outstanding_extents -= num_extents;
7577 spin_lock(&BTRFS_I(inode)->lock);
7578 BTRFS_I(inode)->outstanding_extents += num_extents;
7579 spin_unlock(&BTRFS_I(inode)->lock);
7583 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7584 struct buffer_head *bh_result, int create)
7586 struct extent_map *em;
7587 struct btrfs_root *root = BTRFS_I(inode)->root;
7588 struct extent_state *cached_state = NULL;
7589 struct btrfs_dio_data *dio_data = NULL;
7590 u64 start = iblock << inode->i_blkbits;
7591 u64 lockstart, lockend;
7592 u64 len = bh_result->b_size;
7593 int unlock_bits = EXTENT_LOCKED;
7597 unlock_bits |= EXTENT_DIRTY;
7599 len = min_t(u64, len, root->sectorsize);
7602 lockend = start + len - 1;
7604 if (current->journal_info) {
7606 * Need to pull our outstanding extents and set journal_info to NULL so
7607 * that anything that needs to check if there's a transaction doesn't get
7610 dio_data = current->journal_info;
7611 current->journal_info = NULL;
7615 * If this errors out it's because we couldn't invalidate pagecache for
7616 * this range and we need to fallback to buffered.
7618 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7624 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7631 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7632 * io. INLINE is special, and we could probably kludge it in here, but
7633 * it's still buffered so for safety lets just fall back to the generic
7636 * For COMPRESSED we _have_ to read the entire extent in so we can
7637 * decompress it, so there will be buffering required no matter what we
7638 * do, so go ahead and fallback to buffered.
7640 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7641 * to buffered IO. Don't blame me, this is the price we pay for using
7644 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7645 em->block_start == EXTENT_MAP_INLINE) {
7646 free_extent_map(em);
7651 /* Just a good old fashioned hole, return */
7652 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7653 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7654 free_extent_map(em);
7659 * We don't allocate a new extent in the following cases
7661 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7663 * 2) The extent is marked as PREALLOC. We're good to go here and can
7664 * just use the extent.
7668 len = min(len, em->len - (start - em->start));
7669 lockstart = start + len;
7673 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7674 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7675 em->block_start != EXTENT_MAP_HOLE)) {
7677 u64 block_start, orig_start, orig_block_len, ram_bytes;
7679 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7680 type = BTRFS_ORDERED_PREALLOC;
7682 type = BTRFS_ORDERED_NOCOW;
7683 len = min(len, em->len - (start - em->start));
7684 block_start = em->block_start + (start - em->start);
7686 if (can_nocow_extent(inode, start, &len, &orig_start,
7687 &orig_block_len, &ram_bytes) == 1 &&
7688 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7689 struct extent_map *em2;
7691 em2 = btrfs_create_dio_extent(inode, start, len,
7692 orig_start, block_start,
7693 len, orig_block_len,
7695 btrfs_dec_nocow_writers(root->fs_info, block_start);
7696 if (type == BTRFS_ORDERED_PREALLOC) {
7697 free_extent_map(em);
7700 if (em2 && IS_ERR(em2)) {
7709 * this will cow the extent, reset the len in case we changed
7712 len = bh_result->b_size;
7713 free_extent_map(em);
7714 em = btrfs_new_extent_direct(inode, start, len);
7719 len = min(len, em->len - (start - em->start));
7721 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7723 bh_result->b_size = len;
7724 bh_result->b_bdev = em->bdev;
7725 set_buffer_mapped(bh_result);
7727 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7728 set_buffer_new(bh_result);
7731 * Need to update the i_size under the extent lock so buffered
7732 * readers will get the updated i_size when we unlock.
7734 if (start + len > i_size_read(inode))
7735 i_size_write(inode, start + len);
7737 adjust_dio_outstanding_extents(inode, dio_data, len);
7738 btrfs_free_reserved_data_space(inode, start, len);
7739 WARN_ON(dio_data->reserve < len);
7740 dio_data->reserve -= len;
7741 dio_data->unsubmitted_oe_range_end = start + len;
7742 current->journal_info = dio_data;
7746 * In the case of write we need to clear and unlock the entire range,
7747 * in the case of read we need to unlock only the end area that we
7748 * aren't using if there is any left over space.
7750 if (lockstart < lockend) {
7751 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7752 lockend, unlock_bits, 1, 0,
7753 &cached_state, GFP_NOFS);
7755 free_extent_state(cached_state);
7758 free_extent_map(em);
7763 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7764 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7767 current->journal_info = dio_data;
7769 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7770 * write less data then expected, so that we don't underflow our inode's
7771 * outstanding extents counter.
7773 if (create && dio_data)
7774 adjust_dio_outstanding_extents(inode, dio_data, len);
7779 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7782 struct btrfs_root *root = BTRFS_I(inode)->root;
7785 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7789 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7790 BTRFS_WQ_ENDIO_DIO_REPAIR);
7794 ret = btrfs_map_bio(root, bio, mirror_num, 0);
7800 static int btrfs_check_dio_repairable(struct inode *inode,
7801 struct bio *failed_bio,
7802 struct io_failure_record *failrec,
7807 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7808 failrec->logical, failrec->len);
7809 if (num_copies == 1) {
7811 * we only have a single copy of the data, so don't bother with
7812 * all the retry and error correction code that follows. no
7813 * matter what the error is, it is very likely to persist.
7815 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7816 num_copies, failrec->this_mirror, failed_mirror);
7820 failrec->failed_mirror = failed_mirror;
7821 failrec->this_mirror++;
7822 if (failrec->this_mirror == failed_mirror)
7823 failrec->this_mirror++;
7825 if (failrec->this_mirror > num_copies) {
7826 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7827 num_copies, failrec->this_mirror, failed_mirror);
7834 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7835 struct page *page, unsigned int pgoff,
7836 u64 start, u64 end, int failed_mirror,
7837 bio_end_io_t *repair_endio, void *repair_arg)
7839 struct io_failure_record *failrec;
7845 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7847 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7851 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7854 free_io_failure(inode, failrec);
7858 if ((failed_bio->bi_vcnt > 1)
7859 || (failed_bio->bi_io_vec->bv_len
7860 > BTRFS_I(inode)->root->sectorsize))
7861 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7863 read_mode = READ_SYNC;
7865 isector = start - btrfs_io_bio(failed_bio)->logical;
7866 isector >>= inode->i_sb->s_blocksize_bits;
7867 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7868 pgoff, isector, repair_endio, repair_arg);
7870 free_io_failure(inode, failrec);
7873 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7875 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7876 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7877 read_mode, failrec->this_mirror, failrec->in_validation);
7879 ret = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7881 free_io_failure(inode, failrec);
7888 struct btrfs_retry_complete {
7889 struct completion done;
7890 struct inode *inode;
7895 static void btrfs_retry_endio_nocsum(struct bio *bio)
7897 struct btrfs_retry_complete *done = bio->bi_private;
7898 struct inode *inode;
7899 struct bio_vec *bvec;
7905 ASSERT(bio->bi_vcnt == 1);
7906 inode = bio->bi_io_vec->bv_page->mapping->host;
7907 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7910 bio_for_each_segment_all(bvec, bio, i)
7911 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7913 complete(&done->done);
7917 static int __btrfs_correct_data_nocsum(struct inode *inode,
7918 struct btrfs_io_bio *io_bio)
7920 struct btrfs_fs_info *fs_info;
7921 struct bio_vec *bvec;
7922 struct btrfs_retry_complete done;
7930 fs_info = BTRFS_I(inode)->root->fs_info;
7931 sectorsize = BTRFS_I(inode)->root->sectorsize;
7933 start = io_bio->logical;
7936 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7937 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7938 pgoff = bvec->bv_offset;
7940 next_block_or_try_again:
7943 init_completion(&done.done);
7945 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7946 pgoff, start, start + sectorsize - 1,
7948 btrfs_retry_endio_nocsum, &done);
7952 wait_for_completion(&done.done);
7954 if (!done.uptodate) {
7955 /* We might have another mirror, so try again */
7956 goto next_block_or_try_again;
7959 start += sectorsize;
7962 pgoff += sectorsize;
7963 goto next_block_or_try_again;
7970 static void btrfs_retry_endio(struct bio *bio)
7972 struct btrfs_retry_complete *done = bio->bi_private;
7973 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7974 struct inode *inode;
7975 struct bio_vec *bvec;
7986 start = done->start;
7988 ASSERT(bio->bi_vcnt == 1);
7989 inode = bio->bi_io_vec->bv_page->mapping->host;
7990 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7992 bio_for_each_segment_all(bvec, bio, i) {
7993 ret = __readpage_endio_check(done->inode, io_bio, i,
7994 bvec->bv_page, bvec->bv_offset,
7995 done->start, bvec->bv_len);
7997 clean_io_failure(done->inode, done->start,
7998 bvec->bv_page, bvec->bv_offset);
8003 done->uptodate = uptodate;
8005 complete(&done->done);
8009 static int __btrfs_subio_endio_read(struct inode *inode,
8010 struct btrfs_io_bio *io_bio, int err)
8012 struct btrfs_fs_info *fs_info;
8013 struct bio_vec *bvec;
8014 struct btrfs_retry_complete done;
8024 fs_info = BTRFS_I(inode)->root->fs_info;
8025 sectorsize = BTRFS_I(inode)->root->sectorsize;
8028 start = io_bio->logical;
8031 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8032 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8034 pgoff = bvec->bv_offset;
8036 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8037 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8038 bvec->bv_page, pgoff, start,
8045 init_completion(&done.done);
8047 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8048 pgoff, start, start + sectorsize - 1,
8050 btrfs_retry_endio, &done);
8056 wait_for_completion(&done.done);
8058 if (!done.uptodate) {
8059 /* We might have another mirror, so try again */
8063 offset += sectorsize;
8064 start += sectorsize;
8069 pgoff += sectorsize;
8077 static int btrfs_subio_endio_read(struct inode *inode,
8078 struct btrfs_io_bio *io_bio, int err)
8080 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8084 return __btrfs_correct_data_nocsum(inode, io_bio);
8088 return __btrfs_subio_endio_read(inode, io_bio, err);
8092 static void btrfs_endio_direct_read(struct bio *bio)
8094 struct btrfs_dio_private *dip = bio->bi_private;
8095 struct inode *inode = dip->inode;
8096 struct bio *dio_bio;
8097 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8098 int err = bio->bi_error;
8100 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8101 err = btrfs_subio_endio_read(inode, io_bio, err);
8103 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8104 dip->logical_offset + dip->bytes - 1);
8105 dio_bio = dip->dio_bio;
8109 dio_bio->bi_error = bio->bi_error;
8110 dio_end_io(dio_bio, bio->bi_error);
8113 io_bio->end_io(io_bio, err);
8117 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8122 struct btrfs_root *root = BTRFS_I(inode)->root;
8123 struct btrfs_ordered_extent *ordered = NULL;
8124 u64 ordered_offset = offset;
8125 u64 ordered_bytes = bytes;
8129 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8136 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8137 finish_ordered_fn, NULL, NULL);
8138 btrfs_queue_work(root->fs_info->endio_write_workers,
8142 * our bio might span multiple ordered extents. If we haven't
8143 * completed the accounting for the whole dio, go back and try again
8145 if (ordered_offset < offset + bytes) {
8146 ordered_bytes = offset + bytes - ordered_offset;
8152 static void btrfs_endio_direct_write(struct bio *bio)
8154 struct btrfs_dio_private *dip = bio->bi_private;
8155 struct bio *dio_bio = dip->dio_bio;
8157 btrfs_endio_direct_write_update_ordered(dip->inode,
8158 dip->logical_offset,
8164 dio_bio->bi_error = bio->bi_error;
8165 dio_end_io(dio_bio, bio->bi_error);
8169 static int __btrfs_submit_bio_start_direct_io(struct inode *inode,
8170 struct bio *bio, int mirror_num,
8171 unsigned long bio_flags, u64 offset)
8174 struct btrfs_root *root = BTRFS_I(inode)->root;
8175 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8176 BUG_ON(ret); /* -ENOMEM */
8180 static void btrfs_end_dio_bio(struct bio *bio)
8182 struct btrfs_dio_private *dip = bio->bi_private;
8183 int err = bio->bi_error;
8186 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8187 "direct IO failed ino %llu rw %d,%lu sector %#Lx len %u err no %d",
8188 btrfs_ino(dip->inode), bio_op(bio), bio->bi_rw,
8189 (unsigned long long)bio->bi_iter.bi_sector,
8190 bio->bi_iter.bi_size, err);
8192 if (dip->subio_endio)
8193 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8199 * before atomic variable goto zero, we must make sure
8200 * dip->errors is perceived to be set.
8202 smp_mb__before_atomic();
8205 /* if there are more bios still pending for this dio, just exit */
8206 if (!atomic_dec_and_test(&dip->pending_bios))
8210 bio_io_error(dip->orig_bio);
8212 dip->dio_bio->bi_error = 0;
8213 bio_endio(dip->orig_bio);
8219 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8220 u64 first_sector, gfp_t gfp_flags)
8223 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8225 bio_associate_current(bio);
8229 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8230 struct inode *inode,
8231 struct btrfs_dio_private *dip,
8235 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8236 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8240 * We load all the csum data we need when we submit
8241 * the first bio to reduce the csum tree search and
8244 if (dip->logical_offset == file_offset) {
8245 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8251 if (bio == dip->orig_bio)
8254 file_offset -= dip->logical_offset;
8255 file_offset >>= inode->i_sb->s_blocksize_bits;
8256 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8261 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8262 u64 file_offset, int skip_sum,
8265 struct btrfs_dio_private *dip = bio->bi_private;
8266 bool write = bio_op(bio) == REQ_OP_WRITE;
8267 struct btrfs_root *root = BTRFS_I(inode)->root;
8271 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8276 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8277 BTRFS_WQ_ENDIO_DATA);
8285 if (write && async_submit) {
8286 ret = btrfs_wq_submit_bio(root->fs_info,
8287 inode, bio, 0, 0, file_offset,
8288 __btrfs_submit_bio_start_direct_io,
8289 __btrfs_submit_bio_done);
8293 * If we aren't doing async submit, calculate the csum of the
8296 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8300 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8306 ret = btrfs_map_bio(root, bio, 0, async_submit);
8312 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip,
8315 struct inode *inode = dip->inode;
8316 struct btrfs_root *root = BTRFS_I(inode)->root;
8318 struct bio *orig_bio = dip->orig_bio;
8319 struct bio_vec *bvec = orig_bio->bi_io_vec;
8320 u64 start_sector = orig_bio->bi_iter.bi_sector;
8321 u64 file_offset = dip->logical_offset;
8324 u32 blocksize = root->sectorsize;
8325 int async_submit = 0;
8330 map_length = orig_bio->bi_iter.bi_size;
8331 ret = btrfs_map_block(root->fs_info, bio_op(orig_bio),
8332 start_sector << 9, &map_length, NULL, 0);
8336 if (map_length >= orig_bio->bi_iter.bi_size) {
8338 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8342 /* async crcs make it difficult to collect full stripe writes. */
8343 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8348 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8352 bio_set_op_attrs(bio, bio_op(orig_bio), orig_bio->bi_rw);
8353 bio->bi_private = dip;
8354 bio->bi_end_io = btrfs_end_dio_bio;
8355 btrfs_io_bio(bio)->logical = file_offset;
8356 atomic_inc(&dip->pending_bios);
8358 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8359 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8362 if (unlikely(map_length < submit_len + blocksize ||
8363 bio_add_page(bio, bvec->bv_page, blocksize,
8364 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8366 * inc the count before we submit the bio so
8367 * we know the end IO handler won't happen before
8368 * we inc the count. Otherwise, the dip might get freed
8369 * before we're done setting it up
8371 atomic_inc(&dip->pending_bios);
8372 ret = __btrfs_submit_dio_bio(bio, inode,
8373 file_offset, skip_sum,
8377 atomic_dec(&dip->pending_bios);
8381 start_sector += submit_len >> 9;
8382 file_offset += submit_len;
8386 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8387 start_sector, GFP_NOFS);
8390 bio_set_op_attrs(bio, bio_op(orig_bio), orig_bio->bi_rw);
8391 bio->bi_private = dip;
8392 bio->bi_end_io = btrfs_end_dio_bio;
8393 btrfs_io_bio(bio)->logical = file_offset;
8395 map_length = orig_bio->bi_iter.bi_size;
8396 ret = btrfs_map_block(root->fs_info, bio_op(orig_bio),
8398 &map_length, NULL, 0);
8406 submit_len += blocksize;
8416 ret = __btrfs_submit_dio_bio(bio, inode, file_offset, skip_sum,
8425 * before atomic variable goto zero, we must
8426 * make sure dip->errors is perceived to be set.
8428 smp_mb__before_atomic();
8429 if (atomic_dec_and_test(&dip->pending_bios))
8430 bio_io_error(dip->orig_bio);
8432 /* bio_end_io() will handle error, so we needn't return it */
8436 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8439 struct btrfs_dio_private *dip = NULL;
8440 struct bio *io_bio = NULL;
8441 struct btrfs_io_bio *btrfs_bio;
8443 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8446 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8448 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8454 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8460 dip->private = dio_bio->bi_private;
8462 dip->logical_offset = file_offset;
8463 dip->bytes = dio_bio->bi_iter.bi_size;
8464 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8465 io_bio->bi_private = dip;
8466 dip->orig_bio = io_bio;
8467 dip->dio_bio = dio_bio;
8468 atomic_set(&dip->pending_bios, 0);
8469 btrfs_bio = btrfs_io_bio(io_bio);
8470 btrfs_bio->logical = file_offset;
8473 io_bio->bi_end_io = btrfs_endio_direct_write;
8475 io_bio->bi_end_io = btrfs_endio_direct_read;
8476 dip->subio_endio = btrfs_subio_endio_read;
8480 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8481 * even if we fail to submit a bio, because in such case we do the
8482 * corresponding error handling below and it must not be done a second
8483 * time by btrfs_direct_IO().
8486 struct btrfs_dio_data *dio_data = current->journal_info;
8488 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8490 dio_data->unsubmitted_oe_range_start =
8491 dio_data->unsubmitted_oe_range_end;
8494 ret = btrfs_submit_direct_hook(dip, skip_sum);
8498 if (btrfs_bio->end_io)
8499 btrfs_bio->end_io(btrfs_bio, ret);
8503 * If we arrived here it means either we failed to submit the dip
8504 * or we either failed to clone the dio_bio or failed to allocate the
8505 * dip. If we cloned the dio_bio and allocated the dip, we can just
8506 * call bio_endio against our io_bio so that we get proper resource
8507 * cleanup if we fail to submit the dip, otherwise, we must do the
8508 * same as btrfs_endio_direct_[write|read] because we can't call these
8509 * callbacks - they require an allocated dip and a clone of dio_bio.
8511 if (io_bio && dip) {
8512 io_bio->bi_error = -EIO;
8515 * The end io callbacks free our dip, do the final put on io_bio
8516 * and all the cleanup and final put for dio_bio (through
8523 btrfs_endio_direct_write_update_ordered(inode,
8525 dio_bio->bi_iter.bi_size,
8528 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8529 file_offset + dio_bio->bi_iter.bi_size - 1);
8531 dio_bio->bi_error = -EIO;
8533 * Releases and cleans up our dio_bio, no need to bio_put()
8534 * nor bio_endio()/bio_io_error() against dio_bio.
8536 dio_end_io(dio_bio, ret);
8543 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8544 const struct iov_iter *iter, loff_t offset)
8548 unsigned blocksize_mask = root->sectorsize - 1;
8549 ssize_t retval = -EINVAL;
8551 if (offset & blocksize_mask)
8554 if (iov_iter_alignment(iter) & blocksize_mask)
8557 /* If this is a write we don't need to check anymore */
8558 if (iov_iter_rw(iter) == WRITE)
8561 * Check to make sure we don't have duplicate iov_base's in this
8562 * iovec, if so return EINVAL, otherwise we'll get csum errors
8563 * when reading back.
8565 for (seg = 0; seg < iter->nr_segs; seg++) {
8566 for (i = seg + 1; i < iter->nr_segs; i++) {
8567 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8576 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8578 struct file *file = iocb->ki_filp;
8579 struct inode *inode = file->f_mapping->host;
8580 struct btrfs_root *root = BTRFS_I(inode)->root;
8581 struct btrfs_dio_data dio_data = { 0 };
8582 loff_t offset = iocb->ki_pos;
8586 bool relock = false;
8589 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8592 inode_dio_begin(inode);
8593 smp_mb__after_atomic();
8596 * The generic stuff only does filemap_write_and_wait_range, which
8597 * isn't enough if we've written compressed pages to this area, so
8598 * we need to flush the dirty pages again to make absolutely sure
8599 * that any outstanding dirty pages are on disk.
8601 count = iov_iter_count(iter);
8602 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8603 &BTRFS_I(inode)->runtime_flags))
8604 filemap_fdatawrite_range(inode->i_mapping, offset,
8605 offset + count - 1);
8607 if (iov_iter_rw(iter) == WRITE) {
8609 * If the write DIO is beyond the EOF, we need update
8610 * the isize, but it is protected by i_mutex. So we can
8611 * not unlock the i_mutex at this case.
8613 if (offset + count <= inode->i_size) {
8614 inode_unlock(inode);
8617 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8620 dio_data.outstanding_extents = div64_u64(count +
8621 BTRFS_MAX_EXTENT_SIZE - 1,
8622 BTRFS_MAX_EXTENT_SIZE);
8625 * We need to know how many extents we reserved so that we can
8626 * do the accounting properly if we go over the number we
8627 * originally calculated. Abuse current->journal_info for this.
8629 dio_data.reserve = round_up(count, root->sectorsize);
8630 dio_data.unsubmitted_oe_range_start = (u64)offset;
8631 dio_data.unsubmitted_oe_range_end = (u64)offset;
8632 current->journal_info = &dio_data;
8633 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8634 &BTRFS_I(inode)->runtime_flags)) {
8635 inode_dio_end(inode);
8636 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8640 ret = __blockdev_direct_IO(iocb, inode,
8641 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8642 iter, btrfs_get_blocks_direct, NULL,
8643 btrfs_submit_direct, flags);
8644 if (iov_iter_rw(iter) == WRITE) {
8645 current->journal_info = NULL;
8646 if (ret < 0 && ret != -EIOCBQUEUED) {
8647 if (dio_data.reserve)
8648 btrfs_delalloc_release_space(inode, offset,
8651 * On error we might have left some ordered extents
8652 * without submitting corresponding bios for them, so
8653 * cleanup them up to avoid other tasks getting them
8654 * and waiting for them to complete forever.
8656 if (dio_data.unsubmitted_oe_range_start <
8657 dio_data.unsubmitted_oe_range_end)
8658 btrfs_endio_direct_write_update_ordered(inode,
8659 dio_data.unsubmitted_oe_range_start,
8660 dio_data.unsubmitted_oe_range_end -
8661 dio_data.unsubmitted_oe_range_start,
8663 } else if (ret >= 0 && (size_t)ret < count)
8664 btrfs_delalloc_release_space(inode, offset,
8665 count - (size_t)ret);
8669 inode_dio_end(inode);
8676 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8678 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8679 __u64 start, __u64 len)
8683 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8687 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8690 int btrfs_readpage(struct file *file, struct page *page)
8692 struct extent_io_tree *tree;
8693 tree = &BTRFS_I(page->mapping->host)->io_tree;
8694 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8697 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8699 struct extent_io_tree *tree;
8700 struct inode *inode = page->mapping->host;
8703 if (current->flags & PF_MEMALLOC) {
8704 redirty_page_for_writepage(wbc, page);
8710 * If we are under memory pressure we will call this directly from the
8711 * VM, we need to make sure we have the inode referenced for the ordered
8712 * extent. If not just return like we didn't do anything.
8714 if (!igrab(inode)) {
8715 redirty_page_for_writepage(wbc, page);
8716 return AOP_WRITEPAGE_ACTIVATE;
8718 tree = &BTRFS_I(page->mapping->host)->io_tree;
8719 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8720 btrfs_add_delayed_iput(inode);
8724 static int btrfs_writepages(struct address_space *mapping,
8725 struct writeback_control *wbc)
8727 struct extent_io_tree *tree;
8729 tree = &BTRFS_I(mapping->host)->io_tree;
8730 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8734 btrfs_readpages(struct file *file, struct address_space *mapping,
8735 struct list_head *pages, unsigned nr_pages)
8737 struct extent_io_tree *tree;
8738 tree = &BTRFS_I(mapping->host)->io_tree;
8739 return extent_readpages(tree, mapping, pages, nr_pages,
8742 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8744 struct extent_io_tree *tree;
8745 struct extent_map_tree *map;
8748 tree = &BTRFS_I(page->mapping->host)->io_tree;
8749 map = &BTRFS_I(page->mapping->host)->extent_tree;
8750 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8752 ClearPagePrivate(page);
8753 set_page_private(page, 0);
8759 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8761 if (PageWriteback(page) || PageDirty(page))
8763 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8766 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8767 unsigned int length)
8769 struct inode *inode = page->mapping->host;
8770 struct extent_io_tree *tree;
8771 struct btrfs_ordered_extent *ordered;
8772 struct extent_state *cached_state = NULL;
8773 u64 page_start = page_offset(page);
8774 u64 page_end = page_start + PAGE_SIZE - 1;
8777 int inode_evicting = inode->i_state & I_FREEING;
8780 * we have the page locked, so new writeback can't start,
8781 * and the dirty bit won't be cleared while we are here.
8783 * Wait for IO on this page so that we can safely clear
8784 * the PagePrivate2 bit and do ordered accounting
8786 wait_on_page_writeback(page);
8788 tree = &BTRFS_I(inode)->io_tree;
8790 btrfs_releasepage(page, GFP_NOFS);
8794 if (!inode_evicting)
8795 lock_extent_bits(tree, page_start, page_end, &cached_state);
8798 ordered = btrfs_lookup_ordered_range(inode, start,
8799 page_end - start + 1);
8801 end = min(page_end, ordered->file_offset + ordered->len - 1);
8803 * IO on this page will never be started, so we need
8804 * to account for any ordered extents now
8806 if (!inode_evicting)
8807 clear_extent_bit(tree, start, end,
8808 EXTENT_DIRTY | EXTENT_DELALLOC |
8809 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8810 EXTENT_DEFRAG, 1, 0, &cached_state,
8813 * whoever cleared the private bit is responsible
8814 * for the finish_ordered_io
8816 if (TestClearPagePrivate2(page)) {
8817 struct btrfs_ordered_inode_tree *tree;
8820 tree = &BTRFS_I(inode)->ordered_tree;
8822 spin_lock_irq(&tree->lock);
8823 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8824 new_len = start - ordered->file_offset;
8825 if (new_len < ordered->truncated_len)
8826 ordered->truncated_len = new_len;
8827 spin_unlock_irq(&tree->lock);
8829 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8831 end - start + 1, 1))
8832 btrfs_finish_ordered_io(ordered);
8834 btrfs_put_ordered_extent(ordered);
8835 if (!inode_evicting) {
8836 cached_state = NULL;
8837 lock_extent_bits(tree, start, end,
8842 if (start < page_end)
8847 * Qgroup reserved space handler
8848 * Page here will be either
8849 * 1) Already written to disk
8850 * In this case, its reserved space is released from data rsv map
8851 * and will be freed by delayed_ref handler finally.
8852 * So even we call qgroup_free_data(), it won't decrease reserved
8854 * 2) Not written to disk
8855 * This means the reserved space should be freed here.
8857 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8858 if (!inode_evicting) {
8859 clear_extent_bit(tree, page_start, page_end,
8860 EXTENT_LOCKED | EXTENT_DIRTY |
8861 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8862 EXTENT_DEFRAG, 1, 1,
8863 &cached_state, GFP_NOFS);
8865 __btrfs_releasepage(page, GFP_NOFS);
8868 ClearPageChecked(page);
8869 if (PagePrivate(page)) {
8870 ClearPagePrivate(page);
8871 set_page_private(page, 0);
8877 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8878 * called from a page fault handler when a page is first dirtied. Hence we must
8879 * be careful to check for EOF conditions here. We set the page up correctly
8880 * for a written page which means we get ENOSPC checking when writing into
8881 * holes and correct delalloc and unwritten extent mapping on filesystems that
8882 * support these features.
8884 * We are not allowed to take the i_mutex here so we have to play games to
8885 * protect against truncate races as the page could now be beyond EOF. Because
8886 * vmtruncate() writes the inode size before removing pages, once we have the
8887 * page lock we can determine safely if the page is beyond EOF. If it is not
8888 * beyond EOF, then the page is guaranteed safe against truncation until we
8891 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8893 struct page *page = vmf->page;
8894 struct inode *inode = file_inode(vma->vm_file);
8895 struct btrfs_root *root = BTRFS_I(inode)->root;
8896 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8897 struct btrfs_ordered_extent *ordered;
8898 struct extent_state *cached_state = NULL;
8900 unsigned long zero_start;
8909 reserved_space = PAGE_SIZE;
8911 sb_start_pagefault(inode->i_sb);
8912 page_start = page_offset(page);
8913 page_end = page_start + PAGE_SIZE - 1;
8917 * Reserving delalloc space after obtaining the page lock can lead to
8918 * deadlock. For example, if a dirty page is locked by this function
8919 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8920 * dirty page write out, then the btrfs_writepage() function could
8921 * end up waiting indefinitely to get a lock on the page currently
8922 * being processed by btrfs_page_mkwrite() function.
8924 ret = btrfs_delalloc_reserve_space(inode, page_start,
8927 ret = file_update_time(vma->vm_file);
8933 else /* -ENOSPC, -EIO, etc */
8934 ret = VM_FAULT_SIGBUS;
8940 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8943 size = i_size_read(inode);
8945 if ((page->mapping != inode->i_mapping) ||
8946 (page_start >= size)) {
8947 /* page got truncated out from underneath us */
8950 wait_on_page_writeback(page);
8952 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8953 set_page_extent_mapped(page);
8956 * we can't set the delalloc bits if there are pending ordered
8957 * extents. Drop our locks and wait for them to finish
8959 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8961 unlock_extent_cached(io_tree, page_start, page_end,
8962 &cached_state, GFP_NOFS);
8964 btrfs_start_ordered_extent(inode, ordered, 1);
8965 btrfs_put_ordered_extent(ordered);
8969 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8970 reserved_space = round_up(size - page_start, root->sectorsize);
8971 if (reserved_space < PAGE_SIZE) {
8972 end = page_start + reserved_space - 1;
8973 spin_lock(&BTRFS_I(inode)->lock);
8974 BTRFS_I(inode)->outstanding_extents++;
8975 spin_unlock(&BTRFS_I(inode)->lock);
8976 btrfs_delalloc_release_space(inode, page_start,
8977 PAGE_SIZE - reserved_space);
8982 * XXX - page_mkwrite gets called every time the page is dirtied, even
8983 * if it was already dirty, so for space accounting reasons we need to
8984 * clear any delalloc bits for the range we are fixing to save. There
8985 * is probably a better way to do this, but for now keep consistent with
8986 * prepare_pages in the normal write path.
8988 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8989 EXTENT_DIRTY | EXTENT_DELALLOC |
8990 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8991 0, 0, &cached_state, GFP_NOFS);
8993 ret = btrfs_set_extent_delalloc(inode, page_start, end,
8996 unlock_extent_cached(io_tree, page_start, page_end,
8997 &cached_state, GFP_NOFS);
8998 ret = VM_FAULT_SIGBUS;
9003 /* page is wholly or partially inside EOF */
9004 if (page_start + PAGE_SIZE > size)
9005 zero_start = size & ~PAGE_MASK;
9007 zero_start = PAGE_SIZE;
9009 if (zero_start != PAGE_SIZE) {
9011 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9012 flush_dcache_page(page);
9015 ClearPageChecked(page);
9016 set_page_dirty(page);
9017 SetPageUptodate(page);
9019 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9020 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9021 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9023 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9027 sb_end_pagefault(inode->i_sb);
9028 return VM_FAULT_LOCKED;
9032 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9034 sb_end_pagefault(inode->i_sb);
9038 static int btrfs_truncate(struct inode *inode)
9040 struct btrfs_root *root = BTRFS_I(inode)->root;
9041 struct btrfs_block_rsv *rsv;
9044 struct btrfs_trans_handle *trans;
9045 u64 mask = root->sectorsize - 1;
9046 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9048 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9054 * Yes ladies and gentlemen, this is indeed ugly. The fact is we have
9055 * 3 things going on here
9057 * 1) We need to reserve space for our orphan item and the space to
9058 * delete our orphan item. Lord knows we don't want to have a dangling
9059 * orphan item because we didn't reserve space to remove it.
9061 * 2) We need to reserve space to update our inode.
9063 * 3) We need to have something to cache all the space that is going to
9064 * be free'd up by the truncate operation, but also have some slack
9065 * space reserved in case it uses space during the truncate (thank you
9066 * very much snapshotting).
9068 * And we need these to all be separate. The fact is we can use a lot of
9069 * space doing the truncate, and we have no earthly idea how much space
9070 * we will use, so we need the truncate reservation to be separate so it
9071 * doesn't end up using space reserved for updating the inode or
9072 * removing the orphan item. We also need to be able to stop the
9073 * transaction and start a new one, which means we need to be able to
9074 * update the inode several times, and we have no idea of knowing how
9075 * many times that will be, so we can't just reserve 1 item for the
9076 * entirety of the operation, so that has to be done separately as well.
9077 * Then there is the orphan item, which does indeed need to be held on
9078 * to for the whole operation, and we need nobody to touch this reserved
9079 * space except the orphan code.
9081 * So that leaves us with
9083 * 1) root->orphan_block_rsv - for the orphan deletion.
9084 * 2) rsv - for the truncate reservation, which we will steal from the
9085 * transaction reservation.
9086 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9087 * updating the inode.
9089 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9092 rsv->size = min_size;
9096 * 1 for the truncate slack space
9097 * 1 for updating the inode.
9099 trans = btrfs_start_transaction(root, 2);
9100 if (IS_ERR(trans)) {
9101 err = PTR_ERR(trans);
9105 /* Migrate the slack space for the truncate to our reserve */
9106 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9111 * So if we truncate and then write and fsync we normally would just
9112 * write the extents that changed, which is a problem if we need to
9113 * first truncate that entire inode. So set this flag so we write out
9114 * all of the extents in the inode to the sync log so we're completely
9117 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9118 trans->block_rsv = rsv;
9121 ret = btrfs_truncate_inode_items(trans, root, inode,
9123 BTRFS_EXTENT_DATA_KEY);
9124 if (ret != -ENOSPC && ret != -EAGAIN) {
9129 trans->block_rsv = &root->fs_info->trans_block_rsv;
9130 ret = btrfs_update_inode(trans, root, inode);
9136 btrfs_end_transaction(trans, root);
9137 btrfs_btree_balance_dirty(root);
9139 trans = btrfs_start_transaction(root, 2);
9140 if (IS_ERR(trans)) {
9141 ret = err = PTR_ERR(trans);
9146 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9148 BUG_ON(ret); /* shouldn't happen */
9149 trans->block_rsv = rsv;
9152 if (ret == 0 && inode->i_nlink > 0) {
9153 trans->block_rsv = root->orphan_block_rsv;
9154 ret = btrfs_orphan_del(trans, inode);
9160 trans->block_rsv = &root->fs_info->trans_block_rsv;
9161 ret = btrfs_update_inode(trans, root, inode);
9165 ret = btrfs_end_transaction(trans, root);
9166 btrfs_btree_balance_dirty(root);
9170 btrfs_free_block_rsv(root, rsv);
9179 * create a new subvolume directory/inode (helper for the ioctl).
9181 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9182 struct btrfs_root *new_root,
9183 struct btrfs_root *parent_root,
9186 struct inode *inode;
9190 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9191 new_dirid, new_dirid,
9192 S_IFDIR | (~current_umask() & S_IRWXUGO),
9195 return PTR_ERR(inode);
9196 inode->i_op = &btrfs_dir_inode_operations;
9197 inode->i_fop = &btrfs_dir_file_operations;
9199 set_nlink(inode, 1);
9200 btrfs_i_size_write(inode, 0);
9201 unlock_new_inode(inode);
9203 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9205 btrfs_err(new_root->fs_info,
9206 "error inheriting subvolume %llu properties: %d",
9207 new_root->root_key.objectid, err);
9209 err = btrfs_update_inode(trans, new_root, inode);
9215 struct inode *btrfs_alloc_inode(struct super_block *sb)
9217 struct btrfs_inode *ei;
9218 struct inode *inode;
9220 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9227 ei->last_sub_trans = 0;
9228 ei->logged_trans = 0;
9229 ei->delalloc_bytes = 0;
9230 ei->defrag_bytes = 0;
9231 ei->disk_i_size = 0;
9234 ei->index_cnt = (u64)-1;
9236 ei->last_unlink_trans = 0;
9237 ei->last_log_commit = 0;
9238 ei->delayed_iput_count = 0;
9240 spin_lock_init(&ei->lock);
9241 ei->outstanding_extents = 0;
9242 ei->reserved_extents = 0;
9244 ei->runtime_flags = 0;
9245 ei->force_compress = BTRFS_COMPRESS_NONE;
9247 ei->delayed_node = NULL;
9249 ei->i_otime.tv_sec = 0;
9250 ei->i_otime.tv_nsec = 0;
9252 inode = &ei->vfs_inode;
9253 extent_map_tree_init(&ei->extent_tree);
9254 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9255 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9256 ei->io_tree.track_uptodate = 1;
9257 ei->io_failure_tree.track_uptodate = 1;
9258 atomic_set(&ei->sync_writers, 0);
9259 mutex_init(&ei->log_mutex);
9260 mutex_init(&ei->delalloc_mutex);
9261 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9262 INIT_LIST_HEAD(&ei->delalloc_inodes);
9263 INIT_LIST_HEAD(&ei->delayed_iput);
9264 RB_CLEAR_NODE(&ei->rb_node);
9265 init_rwsem(&ei->dio_sem);
9270 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9271 void btrfs_test_destroy_inode(struct inode *inode)
9273 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9274 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9278 static void btrfs_i_callback(struct rcu_head *head)
9280 struct inode *inode = container_of(head, struct inode, i_rcu);
9281 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9284 void btrfs_destroy_inode(struct inode *inode)
9286 struct btrfs_ordered_extent *ordered;
9287 struct btrfs_root *root = BTRFS_I(inode)->root;
9289 WARN_ON(!hlist_empty(&inode->i_dentry));
9290 WARN_ON(inode->i_data.nrpages);
9291 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9292 WARN_ON(BTRFS_I(inode)->reserved_extents);
9293 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9294 WARN_ON(BTRFS_I(inode)->csum_bytes);
9295 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9298 * This can happen where we create an inode, but somebody else also
9299 * created the same inode and we need to destroy the one we already
9305 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9306 &BTRFS_I(inode)->runtime_flags)) {
9307 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9309 atomic_dec(&root->orphan_inodes);
9313 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9317 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9318 ordered->file_offset, ordered->len);
9319 btrfs_remove_ordered_extent(inode, ordered);
9320 btrfs_put_ordered_extent(ordered);
9321 btrfs_put_ordered_extent(ordered);
9324 btrfs_qgroup_check_reserved_leak(inode);
9325 inode_tree_del(inode);
9326 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9328 call_rcu(&inode->i_rcu, btrfs_i_callback);
9331 int btrfs_drop_inode(struct inode *inode)
9333 struct btrfs_root *root = BTRFS_I(inode)->root;
9338 /* the snap/subvol tree is on deleting */
9339 if (btrfs_root_refs(&root->root_item) == 0)
9342 return generic_drop_inode(inode);
9345 static void init_once(void *foo)
9347 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9349 inode_init_once(&ei->vfs_inode);
9352 void btrfs_destroy_cachep(void)
9355 * Make sure all delayed rcu free inodes are flushed before we
9359 kmem_cache_destroy(btrfs_inode_cachep);
9360 kmem_cache_destroy(btrfs_trans_handle_cachep);
9361 kmem_cache_destroy(btrfs_transaction_cachep);
9362 kmem_cache_destroy(btrfs_path_cachep);
9363 kmem_cache_destroy(btrfs_free_space_cachep);
9366 int btrfs_init_cachep(void)
9368 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9369 sizeof(struct btrfs_inode), 0,
9370 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9372 if (!btrfs_inode_cachep)
9375 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9376 sizeof(struct btrfs_trans_handle), 0,
9377 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9378 if (!btrfs_trans_handle_cachep)
9381 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9382 sizeof(struct btrfs_transaction), 0,
9383 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9384 if (!btrfs_transaction_cachep)
9387 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9388 sizeof(struct btrfs_path), 0,
9389 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9390 if (!btrfs_path_cachep)
9393 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9394 sizeof(struct btrfs_free_space), 0,
9395 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9396 if (!btrfs_free_space_cachep)
9401 btrfs_destroy_cachep();
9405 static int btrfs_getattr(struct vfsmount *mnt,
9406 struct dentry *dentry, struct kstat *stat)
9409 struct inode *inode = d_inode(dentry);
9410 u32 blocksize = inode->i_sb->s_blocksize;
9412 generic_fillattr(inode, stat);
9413 stat->dev = BTRFS_I(inode)->root->anon_dev;
9415 spin_lock(&BTRFS_I(inode)->lock);
9416 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9417 spin_unlock(&BTRFS_I(inode)->lock);
9418 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9419 ALIGN(delalloc_bytes, blocksize)) >> 9;
9423 static int btrfs_rename_exchange(struct inode *old_dir,
9424 struct dentry *old_dentry,
9425 struct inode *new_dir,
9426 struct dentry *new_dentry)
9428 struct btrfs_trans_handle *trans;
9429 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9430 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9431 struct inode *new_inode = new_dentry->d_inode;
9432 struct inode *old_inode = old_dentry->d_inode;
9433 struct timespec ctime = CURRENT_TIME;
9434 struct dentry *parent;
9435 u64 old_ino = btrfs_ino(old_inode);
9436 u64 new_ino = btrfs_ino(new_inode);
9441 bool root_log_pinned = false;
9442 bool dest_log_pinned = false;
9444 /* we only allow rename subvolume link between subvolumes */
9445 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9448 /* close the race window with snapshot create/destroy ioctl */
9449 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9450 down_read(&root->fs_info->subvol_sem);
9451 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9452 down_read(&dest->fs_info->subvol_sem);
9455 * We want to reserve the absolute worst case amount of items. So if
9456 * both inodes are subvols and we need to unlink them then that would
9457 * require 4 item modifications, but if they are both normal inodes it
9458 * would require 5 item modifications, so we'll assume their normal
9459 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9460 * should cover the worst case number of items we'll modify.
9462 trans = btrfs_start_transaction(root, 12);
9463 if (IS_ERR(trans)) {
9464 ret = PTR_ERR(trans);
9469 * We need to find a free sequence number both in the source and
9470 * in the destination directory for the exchange.
9472 ret = btrfs_set_inode_index(new_dir, &old_idx);
9475 ret = btrfs_set_inode_index(old_dir, &new_idx);
9479 BTRFS_I(old_inode)->dir_index = 0ULL;
9480 BTRFS_I(new_inode)->dir_index = 0ULL;
9482 /* Reference for the source. */
9483 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9484 /* force full log commit if subvolume involved. */
9485 btrfs_set_log_full_commit(root->fs_info, trans);
9487 btrfs_pin_log_trans(root);
9488 root_log_pinned = true;
9489 ret = btrfs_insert_inode_ref(trans, dest,
9490 new_dentry->d_name.name,
9491 new_dentry->d_name.len,
9493 btrfs_ino(new_dir), old_idx);
9498 /* And now for the dest. */
9499 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9500 /* force full log commit if subvolume involved. */
9501 btrfs_set_log_full_commit(dest->fs_info, trans);
9503 btrfs_pin_log_trans(dest);
9504 dest_log_pinned = true;
9505 ret = btrfs_insert_inode_ref(trans, root,
9506 old_dentry->d_name.name,
9507 old_dentry->d_name.len,
9509 btrfs_ino(old_dir), new_idx);
9514 /* Update inode version and ctime/mtime. */
9515 inode_inc_iversion(old_dir);
9516 inode_inc_iversion(new_dir);
9517 inode_inc_iversion(old_inode);
9518 inode_inc_iversion(new_inode);
9519 old_dir->i_ctime = old_dir->i_mtime = ctime;
9520 new_dir->i_ctime = new_dir->i_mtime = ctime;
9521 old_inode->i_ctime = ctime;
9522 new_inode->i_ctime = ctime;
9524 if (old_dentry->d_parent != new_dentry->d_parent) {
9525 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9526 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9529 /* src is a subvolume */
9530 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9531 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9532 ret = btrfs_unlink_subvol(trans, root, old_dir,
9534 old_dentry->d_name.name,
9535 old_dentry->d_name.len);
9536 } else { /* src is an inode */
9537 ret = __btrfs_unlink_inode(trans, root, old_dir,
9538 old_dentry->d_inode,
9539 old_dentry->d_name.name,
9540 old_dentry->d_name.len);
9542 ret = btrfs_update_inode(trans, root, old_inode);
9545 btrfs_abort_transaction(trans, root, ret);
9549 /* dest is a subvolume */
9550 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9551 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9552 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9554 new_dentry->d_name.name,
9555 new_dentry->d_name.len);
9556 } else { /* dest is an inode */
9557 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9558 new_dentry->d_inode,
9559 new_dentry->d_name.name,
9560 new_dentry->d_name.len);
9562 ret = btrfs_update_inode(trans, dest, new_inode);
9565 btrfs_abort_transaction(trans, root, ret);
9569 ret = btrfs_add_link(trans, new_dir, old_inode,
9570 new_dentry->d_name.name,
9571 new_dentry->d_name.len, 0, old_idx);
9573 btrfs_abort_transaction(trans, root, ret);
9577 ret = btrfs_add_link(trans, old_dir, new_inode,
9578 old_dentry->d_name.name,
9579 old_dentry->d_name.len, 0, new_idx);
9581 btrfs_abort_transaction(trans, root, ret);
9585 if (old_inode->i_nlink == 1)
9586 BTRFS_I(old_inode)->dir_index = old_idx;
9587 if (new_inode->i_nlink == 1)
9588 BTRFS_I(new_inode)->dir_index = new_idx;
9590 if (root_log_pinned) {
9591 parent = new_dentry->d_parent;
9592 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9593 btrfs_end_log_trans(root);
9594 root_log_pinned = false;
9596 if (dest_log_pinned) {
9597 parent = old_dentry->d_parent;
9598 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9599 btrfs_end_log_trans(dest);
9600 dest_log_pinned = false;
9604 * If we have pinned a log and an error happened, we unpin tasks
9605 * trying to sync the log and force them to fallback to a transaction
9606 * commit if the log currently contains any of the inodes involved in
9607 * this rename operation (to ensure we do not persist a log with an
9608 * inconsistent state for any of these inodes or leading to any
9609 * inconsistencies when replayed). If the transaction was aborted, the
9610 * abortion reason is propagated to userspace when attempting to commit
9611 * the transaction. If the log does not contain any of these inodes, we
9612 * allow the tasks to sync it.
9614 if (ret && (root_log_pinned || dest_log_pinned)) {
9615 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9616 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9617 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9619 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9620 btrfs_set_log_full_commit(root->fs_info, trans);
9622 if (root_log_pinned) {
9623 btrfs_end_log_trans(root);
9624 root_log_pinned = false;
9626 if (dest_log_pinned) {
9627 btrfs_end_log_trans(dest);
9628 dest_log_pinned = false;
9631 ret = btrfs_end_transaction(trans, root);
9633 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9634 up_read(&dest->fs_info->subvol_sem);
9635 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9636 up_read(&root->fs_info->subvol_sem);
9641 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9642 struct btrfs_root *root,
9644 struct dentry *dentry)
9647 struct inode *inode;
9651 ret = btrfs_find_free_ino(root, &objectid);
9655 inode = btrfs_new_inode(trans, root, dir,
9656 dentry->d_name.name,
9660 S_IFCHR | WHITEOUT_MODE,
9663 if (IS_ERR(inode)) {
9664 ret = PTR_ERR(inode);
9668 inode->i_op = &btrfs_special_inode_operations;
9669 init_special_inode(inode, inode->i_mode,
9672 ret = btrfs_init_inode_security(trans, inode, dir,
9677 ret = btrfs_add_nondir(trans, dir, dentry,
9682 ret = btrfs_update_inode(trans, root, inode);
9684 unlock_new_inode(inode);
9686 inode_dec_link_count(inode);
9692 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9693 struct inode *new_dir, struct dentry *new_dentry,
9696 struct btrfs_trans_handle *trans;
9697 unsigned int trans_num_items;
9698 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9699 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9700 struct inode *new_inode = d_inode(new_dentry);
9701 struct inode *old_inode = d_inode(old_dentry);
9705 u64 old_ino = btrfs_ino(old_inode);
9706 bool log_pinned = false;
9708 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9711 /* we only allow rename subvolume link between subvolumes */
9712 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9715 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9716 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9719 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9720 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9724 /* check for collisions, even if the name isn't there */
9725 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9726 new_dentry->d_name.name,
9727 new_dentry->d_name.len);
9730 if (ret == -EEXIST) {
9732 * eexist without a new_inode */
9733 if (WARN_ON(!new_inode)) {
9737 /* maybe -EOVERFLOW */
9744 * we're using rename to replace one file with another. Start IO on it
9745 * now so we don't add too much work to the end of the transaction
9747 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9748 filemap_flush(old_inode->i_mapping);
9750 /* close the racy window with snapshot create/destroy ioctl */
9751 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9752 down_read(&root->fs_info->subvol_sem);
9754 * We want to reserve the absolute worst case amount of items. So if
9755 * both inodes are subvols and we need to unlink them then that would
9756 * require 4 item modifications, but if they are both normal inodes it
9757 * would require 5 item modifications, so we'll assume they are normal
9758 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9759 * should cover the worst case number of items we'll modify.
9760 * If our rename has the whiteout flag, we need more 5 units for the
9761 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9762 * when selinux is enabled).
9764 trans_num_items = 11;
9765 if (flags & RENAME_WHITEOUT)
9766 trans_num_items += 5;
9767 trans = btrfs_start_transaction(root, trans_num_items);
9768 if (IS_ERR(trans)) {
9769 ret = PTR_ERR(trans);
9774 btrfs_record_root_in_trans(trans, dest);
9776 ret = btrfs_set_inode_index(new_dir, &index);
9780 BTRFS_I(old_inode)->dir_index = 0ULL;
9781 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9782 /* force full log commit if subvolume involved. */
9783 btrfs_set_log_full_commit(root->fs_info, trans);
9785 btrfs_pin_log_trans(root);
9787 ret = btrfs_insert_inode_ref(trans, dest,
9788 new_dentry->d_name.name,
9789 new_dentry->d_name.len,
9791 btrfs_ino(new_dir), index);
9796 inode_inc_iversion(old_dir);
9797 inode_inc_iversion(new_dir);
9798 inode_inc_iversion(old_inode);
9799 old_dir->i_ctime = old_dir->i_mtime =
9800 new_dir->i_ctime = new_dir->i_mtime =
9801 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9803 if (old_dentry->d_parent != new_dentry->d_parent)
9804 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9806 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9807 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9808 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9809 old_dentry->d_name.name,
9810 old_dentry->d_name.len);
9812 ret = __btrfs_unlink_inode(trans, root, old_dir,
9813 d_inode(old_dentry),
9814 old_dentry->d_name.name,
9815 old_dentry->d_name.len);
9817 ret = btrfs_update_inode(trans, root, old_inode);
9820 btrfs_abort_transaction(trans, root, ret);
9825 inode_inc_iversion(new_inode);
9826 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9827 if (unlikely(btrfs_ino(new_inode) ==
9828 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9829 root_objectid = BTRFS_I(new_inode)->location.objectid;
9830 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9832 new_dentry->d_name.name,
9833 new_dentry->d_name.len);
9834 BUG_ON(new_inode->i_nlink == 0);
9836 ret = btrfs_unlink_inode(trans, dest, new_dir,
9837 d_inode(new_dentry),
9838 new_dentry->d_name.name,
9839 new_dentry->d_name.len);
9841 if (!ret && new_inode->i_nlink == 0)
9842 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9844 btrfs_abort_transaction(trans, root, ret);
9849 ret = btrfs_add_link(trans, new_dir, old_inode,
9850 new_dentry->d_name.name,
9851 new_dentry->d_name.len, 0, index);
9853 btrfs_abort_transaction(trans, root, ret);
9857 if (old_inode->i_nlink == 1)
9858 BTRFS_I(old_inode)->dir_index = index;
9861 struct dentry *parent = new_dentry->d_parent;
9863 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9864 btrfs_end_log_trans(root);
9868 if (flags & RENAME_WHITEOUT) {
9869 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9873 btrfs_abort_transaction(trans, root, ret);
9879 * If we have pinned the log and an error happened, we unpin tasks
9880 * trying to sync the log and force them to fallback to a transaction
9881 * commit if the log currently contains any of the inodes involved in
9882 * this rename operation (to ensure we do not persist a log with an
9883 * inconsistent state for any of these inodes or leading to any
9884 * inconsistencies when replayed). If the transaction was aborted, the
9885 * abortion reason is propagated to userspace when attempting to commit
9886 * the transaction. If the log does not contain any of these inodes, we
9887 * allow the tasks to sync it.
9889 if (ret && log_pinned) {
9890 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9891 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9892 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9894 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9895 btrfs_set_log_full_commit(root->fs_info, trans);
9897 btrfs_end_log_trans(root);
9900 btrfs_end_transaction(trans, root);
9902 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9903 up_read(&root->fs_info->subvol_sem);
9908 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9909 struct inode *new_dir, struct dentry *new_dentry,
9912 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9915 if (flags & RENAME_EXCHANGE)
9916 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9919 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9922 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9924 struct btrfs_delalloc_work *delalloc_work;
9925 struct inode *inode;
9927 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9929 inode = delalloc_work->inode;
9930 filemap_flush(inode->i_mapping);
9931 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9932 &BTRFS_I(inode)->runtime_flags))
9933 filemap_flush(inode->i_mapping);
9935 if (delalloc_work->delay_iput)
9936 btrfs_add_delayed_iput(inode);
9939 complete(&delalloc_work->completion);
9942 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9945 struct btrfs_delalloc_work *work;
9947 work = kmalloc(sizeof(*work), GFP_NOFS);
9951 init_completion(&work->completion);
9952 INIT_LIST_HEAD(&work->list);
9953 work->inode = inode;
9954 work->delay_iput = delay_iput;
9955 WARN_ON_ONCE(!inode);
9956 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9957 btrfs_run_delalloc_work, NULL, NULL);
9962 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9964 wait_for_completion(&work->completion);
9969 * some fairly slow code that needs optimization. This walks the list
9970 * of all the inodes with pending delalloc and forces them to disk.
9972 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9975 struct btrfs_inode *binode;
9976 struct inode *inode;
9977 struct btrfs_delalloc_work *work, *next;
9978 struct list_head works;
9979 struct list_head splice;
9982 INIT_LIST_HEAD(&works);
9983 INIT_LIST_HEAD(&splice);
9985 mutex_lock(&root->delalloc_mutex);
9986 spin_lock(&root->delalloc_lock);
9987 list_splice_init(&root->delalloc_inodes, &splice);
9988 while (!list_empty(&splice)) {
9989 binode = list_entry(splice.next, struct btrfs_inode,
9992 list_move_tail(&binode->delalloc_inodes,
9993 &root->delalloc_inodes);
9994 inode = igrab(&binode->vfs_inode);
9996 cond_resched_lock(&root->delalloc_lock);
9999 spin_unlock(&root->delalloc_lock);
10001 work = btrfs_alloc_delalloc_work(inode, delay_iput);
10004 btrfs_add_delayed_iput(inode);
10010 list_add_tail(&work->list, &works);
10011 btrfs_queue_work(root->fs_info->flush_workers,
10014 if (nr != -1 && ret >= nr)
10017 spin_lock(&root->delalloc_lock);
10019 spin_unlock(&root->delalloc_lock);
10022 list_for_each_entry_safe(work, next, &works, list) {
10023 list_del_init(&work->list);
10024 btrfs_wait_and_free_delalloc_work(work);
10027 if (!list_empty_careful(&splice)) {
10028 spin_lock(&root->delalloc_lock);
10029 list_splice_tail(&splice, &root->delalloc_inodes);
10030 spin_unlock(&root->delalloc_lock);
10032 mutex_unlock(&root->delalloc_mutex);
10036 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10040 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10043 ret = __start_delalloc_inodes(root, delay_iput, -1);
10047 * the filemap_flush will queue IO into the worker threads, but
10048 * we have to make sure the IO is actually started and that
10049 * ordered extents get created before we return
10051 atomic_inc(&root->fs_info->async_submit_draining);
10052 while (atomic_read(&root->fs_info->nr_async_submits) ||
10053 atomic_read(&root->fs_info->async_delalloc_pages)) {
10054 wait_event(root->fs_info->async_submit_wait,
10055 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10056 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10058 atomic_dec(&root->fs_info->async_submit_draining);
10062 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10065 struct btrfs_root *root;
10066 struct list_head splice;
10069 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10072 INIT_LIST_HEAD(&splice);
10074 mutex_lock(&fs_info->delalloc_root_mutex);
10075 spin_lock(&fs_info->delalloc_root_lock);
10076 list_splice_init(&fs_info->delalloc_roots, &splice);
10077 while (!list_empty(&splice) && nr) {
10078 root = list_first_entry(&splice, struct btrfs_root,
10080 root = btrfs_grab_fs_root(root);
10082 list_move_tail(&root->delalloc_root,
10083 &fs_info->delalloc_roots);
10084 spin_unlock(&fs_info->delalloc_root_lock);
10086 ret = __start_delalloc_inodes(root, delay_iput, nr);
10087 btrfs_put_fs_root(root);
10095 spin_lock(&fs_info->delalloc_root_lock);
10097 spin_unlock(&fs_info->delalloc_root_lock);
10100 atomic_inc(&fs_info->async_submit_draining);
10101 while (atomic_read(&fs_info->nr_async_submits) ||
10102 atomic_read(&fs_info->async_delalloc_pages)) {
10103 wait_event(fs_info->async_submit_wait,
10104 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10105 atomic_read(&fs_info->async_delalloc_pages) == 0));
10107 atomic_dec(&fs_info->async_submit_draining);
10109 if (!list_empty_careful(&splice)) {
10110 spin_lock(&fs_info->delalloc_root_lock);
10111 list_splice_tail(&splice, &fs_info->delalloc_roots);
10112 spin_unlock(&fs_info->delalloc_root_lock);
10114 mutex_unlock(&fs_info->delalloc_root_mutex);
10118 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10119 const char *symname)
10121 struct btrfs_trans_handle *trans;
10122 struct btrfs_root *root = BTRFS_I(dir)->root;
10123 struct btrfs_path *path;
10124 struct btrfs_key key;
10125 struct inode *inode = NULL;
10127 int drop_inode = 0;
10133 struct btrfs_file_extent_item *ei;
10134 struct extent_buffer *leaf;
10136 name_len = strlen(symname);
10137 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10138 return -ENAMETOOLONG;
10141 * 2 items for inode item and ref
10142 * 2 items for dir items
10143 * 1 item for updating parent inode item
10144 * 1 item for the inline extent item
10145 * 1 item for xattr if selinux is on
10147 trans = btrfs_start_transaction(root, 7);
10149 return PTR_ERR(trans);
10151 err = btrfs_find_free_ino(root, &objectid);
10155 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10156 dentry->d_name.len, btrfs_ino(dir), objectid,
10157 S_IFLNK|S_IRWXUGO, &index);
10158 if (IS_ERR(inode)) {
10159 err = PTR_ERR(inode);
10164 * If the active LSM wants to access the inode during
10165 * d_instantiate it needs these. Smack checks to see
10166 * if the filesystem supports xattrs by looking at the
10169 inode->i_fop = &btrfs_file_operations;
10170 inode->i_op = &btrfs_file_inode_operations;
10171 inode->i_mapping->a_ops = &btrfs_aops;
10172 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10174 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10176 goto out_unlock_inode;
10178 path = btrfs_alloc_path();
10181 goto out_unlock_inode;
10183 key.objectid = btrfs_ino(inode);
10185 key.type = BTRFS_EXTENT_DATA_KEY;
10186 datasize = btrfs_file_extent_calc_inline_size(name_len);
10187 err = btrfs_insert_empty_item(trans, root, path, &key,
10190 btrfs_free_path(path);
10191 goto out_unlock_inode;
10193 leaf = path->nodes[0];
10194 ei = btrfs_item_ptr(leaf, path->slots[0],
10195 struct btrfs_file_extent_item);
10196 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10197 btrfs_set_file_extent_type(leaf, ei,
10198 BTRFS_FILE_EXTENT_INLINE);
10199 btrfs_set_file_extent_encryption(leaf, ei, 0);
10200 btrfs_set_file_extent_compression(leaf, ei, 0);
10201 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10202 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10204 ptr = btrfs_file_extent_inline_start(ei);
10205 write_extent_buffer(leaf, symname, ptr, name_len);
10206 btrfs_mark_buffer_dirty(leaf);
10207 btrfs_free_path(path);
10209 inode->i_op = &btrfs_symlink_inode_operations;
10210 inode_nohighmem(inode);
10211 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10212 inode_set_bytes(inode, name_len);
10213 btrfs_i_size_write(inode, name_len);
10214 err = btrfs_update_inode(trans, root, inode);
10216 * Last step, add directory indexes for our symlink inode. This is the
10217 * last step to avoid extra cleanup of these indexes if an error happens
10221 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10224 goto out_unlock_inode;
10227 unlock_new_inode(inode);
10228 d_instantiate(dentry, inode);
10231 btrfs_end_transaction(trans, root);
10233 inode_dec_link_count(inode);
10236 btrfs_btree_balance_dirty(root);
10241 unlock_new_inode(inode);
10245 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10246 u64 start, u64 num_bytes, u64 min_size,
10247 loff_t actual_len, u64 *alloc_hint,
10248 struct btrfs_trans_handle *trans)
10250 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10251 struct extent_map *em;
10252 struct btrfs_root *root = BTRFS_I(inode)->root;
10253 struct btrfs_key ins;
10254 u64 cur_offset = start;
10257 u64 last_alloc = (u64)-1;
10259 bool own_trans = true;
10263 while (num_bytes > 0) {
10265 trans = btrfs_start_transaction(root, 3);
10266 if (IS_ERR(trans)) {
10267 ret = PTR_ERR(trans);
10272 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10273 cur_bytes = max(cur_bytes, min_size);
10275 * If we are severely fragmented we could end up with really
10276 * small allocations, so if the allocator is returning small
10277 * chunks lets make its job easier by only searching for those
10280 cur_bytes = min(cur_bytes, last_alloc);
10281 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10282 *alloc_hint, &ins, 1, 0);
10285 btrfs_end_transaction(trans, root);
10288 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10290 last_alloc = ins.offset;
10291 ret = insert_reserved_file_extent(trans, inode,
10292 cur_offset, ins.objectid,
10293 ins.offset, ins.offset,
10294 ins.offset, 0, 0, 0,
10295 BTRFS_FILE_EXTENT_PREALLOC);
10297 btrfs_free_reserved_extent(root, ins.objectid,
10299 btrfs_abort_transaction(trans, root, ret);
10301 btrfs_end_transaction(trans, root);
10305 btrfs_drop_extent_cache(inode, cur_offset,
10306 cur_offset + ins.offset -1, 0);
10308 em = alloc_extent_map();
10310 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10311 &BTRFS_I(inode)->runtime_flags);
10315 em->start = cur_offset;
10316 em->orig_start = cur_offset;
10317 em->len = ins.offset;
10318 em->block_start = ins.objectid;
10319 em->block_len = ins.offset;
10320 em->orig_block_len = ins.offset;
10321 em->ram_bytes = ins.offset;
10322 em->bdev = root->fs_info->fs_devices->latest_bdev;
10323 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10324 em->generation = trans->transid;
10327 write_lock(&em_tree->lock);
10328 ret = add_extent_mapping(em_tree, em, 1);
10329 write_unlock(&em_tree->lock);
10330 if (ret != -EEXIST)
10332 btrfs_drop_extent_cache(inode, cur_offset,
10333 cur_offset + ins.offset - 1,
10336 free_extent_map(em);
10338 num_bytes -= ins.offset;
10339 cur_offset += ins.offset;
10340 *alloc_hint = ins.objectid + ins.offset;
10342 inode_inc_iversion(inode);
10343 inode->i_ctime = current_fs_time(inode->i_sb);
10344 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10345 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10346 (actual_len > inode->i_size) &&
10347 (cur_offset > inode->i_size)) {
10348 if (cur_offset > actual_len)
10349 i_size = actual_len;
10351 i_size = cur_offset;
10352 i_size_write(inode, i_size);
10353 btrfs_ordered_update_i_size(inode, i_size, NULL);
10356 ret = btrfs_update_inode(trans, root, inode);
10359 btrfs_abort_transaction(trans, root, ret);
10361 btrfs_end_transaction(trans, root);
10366 btrfs_end_transaction(trans, root);
10371 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10372 u64 start, u64 num_bytes, u64 min_size,
10373 loff_t actual_len, u64 *alloc_hint)
10375 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10376 min_size, actual_len, alloc_hint,
10380 int btrfs_prealloc_file_range_trans(struct inode *inode,
10381 struct btrfs_trans_handle *trans, int mode,
10382 u64 start, u64 num_bytes, u64 min_size,
10383 loff_t actual_len, u64 *alloc_hint)
10385 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10386 min_size, actual_len, alloc_hint, trans);
10389 static int btrfs_set_page_dirty(struct page *page)
10391 return __set_page_dirty_nobuffers(page);
10394 static int btrfs_permission(struct inode *inode, int mask)
10396 struct btrfs_root *root = BTRFS_I(inode)->root;
10397 umode_t mode = inode->i_mode;
10399 if (mask & MAY_WRITE &&
10400 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10401 if (btrfs_root_readonly(root))
10403 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10406 return generic_permission(inode, mask);
10409 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10411 struct btrfs_trans_handle *trans;
10412 struct btrfs_root *root = BTRFS_I(dir)->root;
10413 struct inode *inode = NULL;
10419 * 5 units required for adding orphan entry
10421 trans = btrfs_start_transaction(root, 5);
10423 return PTR_ERR(trans);
10425 ret = btrfs_find_free_ino(root, &objectid);
10429 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10430 btrfs_ino(dir), objectid, mode, &index);
10431 if (IS_ERR(inode)) {
10432 ret = PTR_ERR(inode);
10437 inode->i_fop = &btrfs_file_operations;
10438 inode->i_op = &btrfs_file_inode_operations;
10440 inode->i_mapping->a_ops = &btrfs_aops;
10441 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10443 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10447 ret = btrfs_update_inode(trans, root, inode);
10450 ret = btrfs_orphan_add(trans, inode);
10455 * We set number of links to 0 in btrfs_new_inode(), and here we set
10456 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10459 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10461 set_nlink(inode, 1);
10462 unlock_new_inode(inode);
10463 d_tmpfile(dentry, inode);
10464 mark_inode_dirty(inode);
10467 btrfs_end_transaction(trans, root);
10470 btrfs_balance_delayed_items(root);
10471 btrfs_btree_balance_dirty(root);
10475 unlock_new_inode(inode);
10480 /* Inspired by filemap_check_errors() */
10481 int btrfs_inode_check_errors(struct inode *inode)
10485 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10486 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10488 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10489 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10495 static const struct inode_operations btrfs_dir_inode_operations = {
10496 .getattr = btrfs_getattr,
10497 .lookup = btrfs_lookup,
10498 .create = btrfs_create,
10499 .unlink = btrfs_unlink,
10500 .link = btrfs_link,
10501 .mkdir = btrfs_mkdir,
10502 .rmdir = btrfs_rmdir,
10503 .rename2 = btrfs_rename2,
10504 .symlink = btrfs_symlink,
10505 .setattr = btrfs_setattr,
10506 .mknod = btrfs_mknod,
10507 .setxattr = generic_setxattr,
10508 .getxattr = generic_getxattr,
10509 .listxattr = btrfs_listxattr,
10510 .removexattr = generic_removexattr,
10511 .permission = btrfs_permission,
10512 .get_acl = btrfs_get_acl,
10513 .set_acl = btrfs_set_acl,
10514 .update_time = btrfs_update_time,
10515 .tmpfile = btrfs_tmpfile,
10517 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10518 .lookup = btrfs_lookup,
10519 .permission = btrfs_permission,
10520 .get_acl = btrfs_get_acl,
10521 .set_acl = btrfs_set_acl,
10522 .update_time = btrfs_update_time,
10525 static const struct file_operations btrfs_dir_file_operations = {
10526 .llseek = generic_file_llseek,
10527 .read = generic_read_dir,
10528 .iterate = btrfs_real_readdir,
10529 .unlocked_ioctl = btrfs_ioctl,
10530 #ifdef CONFIG_COMPAT
10531 .compat_ioctl = btrfs_compat_ioctl,
10533 .release = btrfs_release_file,
10534 .fsync = btrfs_sync_file,
10537 static const struct extent_io_ops btrfs_extent_io_ops = {
10538 .fill_delalloc = run_delalloc_range,
10539 .submit_bio_hook = btrfs_submit_bio_hook,
10540 .merge_bio_hook = btrfs_merge_bio_hook,
10541 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10542 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10543 .writepage_start_hook = btrfs_writepage_start_hook,
10544 .set_bit_hook = btrfs_set_bit_hook,
10545 .clear_bit_hook = btrfs_clear_bit_hook,
10546 .merge_extent_hook = btrfs_merge_extent_hook,
10547 .split_extent_hook = btrfs_split_extent_hook,
10551 * btrfs doesn't support the bmap operation because swapfiles
10552 * use bmap to make a mapping of extents in the file. They assume
10553 * these extents won't change over the life of the file and they
10554 * use the bmap result to do IO directly to the drive.
10556 * the btrfs bmap call would return logical addresses that aren't
10557 * suitable for IO and they also will change frequently as COW
10558 * operations happen. So, swapfile + btrfs == corruption.
10560 * For now we're avoiding this by dropping bmap.
10562 static const struct address_space_operations btrfs_aops = {
10563 .readpage = btrfs_readpage,
10564 .writepage = btrfs_writepage,
10565 .writepages = btrfs_writepages,
10566 .readpages = btrfs_readpages,
10567 .direct_IO = btrfs_direct_IO,
10568 .invalidatepage = btrfs_invalidatepage,
10569 .releasepage = btrfs_releasepage,
10570 .set_page_dirty = btrfs_set_page_dirty,
10571 .error_remove_page = generic_error_remove_page,
10574 static const struct address_space_operations btrfs_symlink_aops = {
10575 .readpage = btrfs_readpage,
10576 .writepage = btrfs_writepage,
10577 .invalidatepage = btrfs_invalidatepage,
10578 .releasepage = btrfs_releasepage,
10581 static const struct inode_operations btrfs_file_inode_operations = {
10582 .getattr = btrfs_getattr,
10583 .setattr = btrfs_setattr,
10584 .setxattr = generic_setxattr,
10585 .getxattr = generic_getxattr,
10586 .listxattr = btrfs_listxattr,
10587 .removexattr = generic_removexattr,
10588 .permission = btrfs_permission,
10589 .fiemap = btrfs_fiemap,
10590 .get_acl = btrfs_get_acl,
10591 .set_acl = btrfs_set_acl,
10592 .update_time = btrfs_update_time,
10594 static const struct inode_operations btrfs_special_inode_operations = {
10595 .getattr = btrfs_getattr,
10596 .setattr = btrfs_setattr,
10597 .permission = btrfs_permission,
10598 .setxattr = generic_setxattr,
10599 .getxattr = generic_getxattr,
10600 .listxattr = btrfs_listxattr,
10601 .removexattr = generic_removexattr,
10602 .get_acl = btrfs_get_acl,
10603 .set_acl = btrfs_set_acl,
10604 .update_time = btrfs_update_time,
10606 static const struct inode_operations btrfs_symlink_inode_operations = {
10607 .readlink = generic_readlink,
10608 .get_link = page_get_link,
10609 .getattr = btrfs_getattr,
10610 .setattr = btrfs_setattr,
10611 .permission = btrfs_permission,
10612 .setxattr = generic_setxattr,
10613 .getxattr = generic_getxattr,
10614 .listxattr = btrfs_listxattr,
10615 .removexattr = generic_removexattr,
10616 .update_time = btrfs_update_time,
10619 const struct dentry_operations btrfs_dentry_operations = {
10620 .d_delete = btrfs_dentry_delete,
10621 .d_release = btrfs_dentry_release,