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 dosen't save disk space at all.
460 if (total_compressed <= blocksize &&
461 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
462 goto cleanup_and_bail_uncompressed;
464 /* we want to make sure that amount of ram required to uncompress
465 * an extent is reasonable, so we limit the total size in ram
466 * of a compressed extent to 128k. This is a crucial number
467 * because it also controls how easily we can spread reads across
468 * cpus for decompression.
470 * We also want to make sure the amount of IO required to do
471 * a random read is reasonably small, so we limit the size of
472 * a compressed extent to 128k.
474 total_compressed = min(total_compressed, max_uncompressed);
475 num_bytes = ALIGN(end - start + 1, blocksize);
476 num_bytes = max(blocksize, num_bytes);
481 * we do compression for mount -o compress and when the
482 * inode has not been flagged as nocompress. This flag can
483 * change at any time if we discover bad compression ratios.
485 if (inode_need_compress(inode)) {
487 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
489 /* just bail out to the uncompressed code */
493 if (BTRFS_I(inode)->force_compress)
494 compress_type = BTRFS_I(inode)->force_compress;
497 * we need to call clear_page_dirty_for_io on each
498 * page in the range. Otherwise applications with the file
499 * mmap'd can wander in and change the page contents while
500 * we are compressing them.
502 * If the compression fails for any reason, we set the pages
503 * dirty again later on.
505 extent_range_clear_dirty_for_io(inode, start, end);
507 ret = btrfs_compress_pages(compress_type,
508 inode->i_mapping, start,
509 total_compressed, pages,
510 nr_pages, &nr_pages_ret,
516 unsigned long offset = total_compressed &
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(int rw, 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, rw, 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, int rw,
1859 struct bio *bio, int mirror_num,
1860 unsigned long bio_flags,
1863 struct btrfs_root *root = BTRFS_I(inode)->root;
1866 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1867 BUG_ON(ret); /* -ENOMEM */
1872 * in order to insert checksums into the metadata in large chunks,
1873 * we wait until bio submission time. All the pages in the bio are
1874 * checksummed and sums are attached onto the ordered extent record.
1876 * At IO completion time the cums attached on the ordered extent record
1877 * are inserted into the btree
1879 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1880 int mirror_num, unsigned long bio_flags,
1883 struct btrfs_root *root = BTRFS_I(inode)->root;
1886 ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1888 bio->bi_error = ret;
1895 * extent_io.c submission hook. This does the right thing for csum calculation
1896 * on write, or reading the csums from the tree before a read
1898 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1899 int mirror_num, unsigned long bio_flags,
1902 struct btrfs_root *root = BTRFS_I(inode)->root;
1903 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1906 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1908 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1910 if (btrfs_is_free_space_inode(inode))
1911 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1913 if (!(rw & REQ_WRITE)) {
1914 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1918 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1919 ret = btrfs_submit_compressed_read(inode, bio,
1923 } else if (!skip_sum) {
1924 ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1929 } else if (async && !skip_sum) {
1930 /* csum items have already been cloned */
1931 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1933 /* we're doing a write, do the async checksumming */
1934 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1935 inode, rw, bio, mirror_num,
1936 bio_flags, bio_offset,
1937 __btrfs_submit_bio_start,
1938 __btrfs_submit_bio_done);
1940 } else if (!skip_sum) {
1941 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1947 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1951 bio->bi_error = ret;
1958 * given a list of ordered sums record them in the inode. This happens
1959 * at IO completion time based on sums calculated at bio submission time.
1961 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1962 struct inode *inode, u64 file_offset,
1963 struct list_head *list)
1965 struct btrfs_ordered_sum *sum;
1967 list_for_each_entry(sum, list, list) {
1968 trans->adding_csums = 1;
1969 btrfs_csum_file_blocks(trans,
1970 BTRFS_I(inode)->root->fs_info->csum_root, sum);
1971 trans->adding_csums = 0;
1976 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1977 struct extent_state **cached_state)
1979 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
1980 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1981 cached_state, GFP_NOFS);
1984 /* see btrfs_writepage_start_hook for details on why this is required */
1985 struct btrfs_writepage_fixup {
1987 struct btrfs_work work;
1990 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1992 struct btrfs_writepage_fixup *fixup;
1993 struct btrfs_ordered_extent *ordered;
1994 struct extent_state *cached_state = NULL;
1996 struct inode *inode;
2001 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2005 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2006 ClearPageChecked(page);
2010 inode = page->mapping->host;
2011 page_start = page_offset(page);
2012 page_end = page_offset(page) + PAGE_SIZE - 1;
2014 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2017 /* already ordered? We're done */
2018 if (PagePrivate2(page))
2021 ordered = btrfs_lookup_ordered_range(inode, page_start,
2024 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2025 page_end, &cached_state, GFP_NOFS);
2027 btrfs_start_ordered_extent(inode, ordered, 1);
2028 btrfs_put_ordered_extent(ordered);
2032 ret = btrfs_delalloc_reserve_space(inode, page_start,
2035 mapping_set_error(page->mapping, ret);
2036 end_extent_writepage(page, ret, page_start, page_end);
2037 ClearPageChecked(page);
2041 btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
2042 ClearPageChecked(page);
2043 set_page_dirty(page);
2045 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2046 &cached_state, GFP_NOFS);
2054 * There are a few paths in the higher layers of the kernel that directly
2055 * set the page dirty bit without asking the filesystem if it is a
2056 * good idea. This causes problems because we want to make sure COW
2057 * properly happens and the data=ordered rules are followed.
2059 * In our case any range that doesn't have the ORDERED bit set
2060 * hasn't been properly setup for IO. We kick off an async process
2061 * to fix it up. The async helper will wait for ordered extents, set
2062 * the delalloc bit and make it safe to write the page.
2064 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2066 struct inode *inode = page->mapping->host;
2067 struct btrfs_writepage_fixup *fixup;
2068 struct btrfs_root *root = BTRFS_I(inode)->root;
2070 /* this page is properly in the ordered list */
2071 if (TestClearPagePrivate2(page))
2074 if (PageChecked(page))
2077 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2081 SetPageChecked(page);
2083 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2084 btrfs_writepage_fixup_worker, NULL, NULL);
2086 btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2090 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2091 struct inode *inode, u64 file_pos,
2092 u64 disk_bytenr, u64 disk_num_bytes,
2093 u64 num_bytes, u64 ram_bytes,
2094 u8 compression, u8 encryption,
2095 u16 other_encoding, int extent_type)
2097 struct btrfs_root *root = BTRFS_I(inode)->root;
2098 struct btrfs_file_extent_item *fi;
2099 struct btrfs_path *path;
2100 struct extent_buffer *leaf;
2101 struct btrfs_key ins;
2102 int extent_inserted = 0;
2105 path = btrfs_alloc_path();
2110 * we may be replacing one extent in the tree with another.
2111 * The new extent is pinned in the extent map, and we don't want
2112 * to drop it from the cache until it is completely in the btree.
2114 * So, tell btrfs_drop_extents to leave this extent in the cache.
2115 * the caller is expected to unpin it and allow it to be merged
2118 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2119 file_pos + num_bytes, NULL, 0,
2120 1, sizeof(*fi), &extent_inserted);
2124 if (!extent_inserted) {
2125 ins.objectid = btrfs_ino(inode);
2126 ins.offset = file_pos;
2127 ins.type = BTRFS_EXTENT_DATA_KEY;
2129 path->leave_spinning = 1;
2130 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2135 leaf = path->nodes[0];
2136 fi = btrfs_item_ptr(leaf, path->slots[0],
2137 struct btrfs_file_extent_item);
2138 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2139 btrfs_set_file_extent_type(leaf, fi, extent_type);
2140 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2141 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2142 btrfs_set_file_extent_offset(leaf, fi, 0);
2143 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2144 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2145 btrfs_set_file_extent_compression(leaf, fi, compression);
2146 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2147 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2149 btrfs_mark_buffer_dirty(leaf);
2150 btrfs_release_path(path);
2152 inode_add_bytes(inode, num_bytes);
2154 ins.objectid = disk_bytenr;
2155 ins.offset = disk_num_bytes;
2156 ins.type = BTRFS_EXTENT_ITEM_KEY;
2157 ret = btrfs_alloc_reserved_file_extent(trans, root,
2158 root->root_key.objectid,
2159 btrfs_ino(inode), file_pos,
2162 * Release the reserved range from inode dirty range map, as it is
2163 * already moved into delayed_ref_head
2165 btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2167 btrfs_free_path(path);
2172 /* snapshot-aware defrag */
2173 struct sa_defrag_extent_backref {
2174 struct rb_node node;
2175 struct old_sa_defrag_extent *old;
2184 struct old_sa_defrag_extent {
2185 struct list_head list;
2186 struct new_sa_defrag_extent *new;
2195 struct new_sa_defrag_extent {
2196 struct rb_root root;
2197 struct list_head head;
2198 struct btrfs_path *path;
2199 struct inode *inode;
2207 static int backref_comp(struct sa_defrag_extent_backref *b1,
2208 struct sa_defrag_extent_backref *b2)
2210 if (b1->root_id < b2->root_id)
2212 else if (b1->root_id > b2->root_id)
2215 if (b1->inum < b2->inum)
2217 else if (b1->inum > b2->inum)
2220 if (b1->file_pos < b2->file_pos)
2222 else if (b1->file_pos > b2->file_pos)
2226 * [------------------------------] ===> (a range of space)
2227 * |<--->| |<---->| =============> (fs/file tree A)
2228 * |<---------------------------->| ===> (fs/file tree B)
2230 * A range of space can refer to two file extents in one tree while
2231 * refer to only one file extent in another tree.
2233 * So we may process a disk offset more than one time(two extents in A)
2234 * and locate at the same extent(one extent in B), then insert two same
2235 * backrefs(both refer to the extent in B).
2240 static void backref_insert(struct rb_root *root,
2241 struct sa_defrag_extent_backref *backref)
2243 struct rb_node **p = &root->rb_node;
2244 struct rb_node *parent = NULL;
2245 struct sa_defrag_extent_backref *entry;
2250 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2252 ret = backref_comp(backref, entry);
2256 p = &(*p)->rb_right;
2259 rb_link_node(&backref->node, parent, p);
2260 rb_insert_color(&backref->node, root);
2264 * Note the backref might has changed, and in this case we just return 0.
2266 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2269 struct btrfs_file_extent_item *extent;
2270 struct btrfs_fs_info *fs_info;
2271 struct old_sa_defrag_extent *old = ctx;
2272 struct new_sa_defrag_extent *new = old->new;
2273 struct btrfs_path *path = new->path;
2274 struct btrfs_key key;
2275 struct btrfs_root *root;
2276 struct sa_defrag_extent_backref *backref;
2277 struct extent_buffer *leaf;
2278 struct inode *inode = new->inode;
2284 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2285 inum == btrfs_ino(inode))
2288 key.objectid = root_id;
2289 key.type = BTRFS_ROOT_ITEM_KEY;
2290 key.offset = (u64)-1;
2292 fs_info = BTRFS_I(inode)->root->fs_info;
2293 root = btrfs_read_fs_root_no_name(fs_info, &key);
2295 if (PTR_ERR(root) == -ENOENT)
2298 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2299 inum, offset, root_id);
2300 return PTR_ERR(root);
2303 key.objectid = inum;
2304 key.type = BTRFS_EXTENT_DATA_KEY;
2305 if (offset > (u64)-1 << 32)
2308 key.offset = offset;
2310 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2311 if (WARN_ON(ret < 0))
2318 leaf = path->nodes[0];
2319 slot = path->slots[0];
2321 if (slot >= btrfs_header_nritems(leaf)) {
2322 ret = btrfs_next_leaf(root, path);
2325 } else if (ret > 0) {
2334 btrfs_item_key_to_cpu(leaf, &key, slot);
2336 if (key.objectid > inum)
2339 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2342 extent = btrfs_item_ptr(leaf, slot,
2343 struct btrfs_file_extent_item);
2345 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2349 * 'offset' refers to the exact key.offset,
2350 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2351 * (key.offset - extent_offset).
2353 if (key.offset != offset)
2356 extent_offset = btrfs_file_extent_offset(leaf, extent);
2357 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2359 if (extent_offset >= old->extent_offset + old->offset +
2360 old->len || extent_offset + num_bytes <=
2361 old->extent_offset + old->offset)
2366 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2372 backref->root_id = root_id;
2373 backref->inum = inum;
2374 backref->file_pos = offset;
2375 backref->num_bytes = num_bytes;
2376 backref->extent_offset = extent_offset;
2377 backref->generation = btrfs_file_extent_generation(leaf, extent);
2379 backref_insert(&new->root, backref);
2382 btrfs_release_path(path);
2387 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2388 struct new_sa_defrag_extent *new)
2390 struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2391 struct old_sa_defrag_extent *old, *tmp;
2396 list_for_each_entry_safe(old, tmp, &new->head, list) {
2397 ret = iterate_inodes_from_logical(old->bytenr +
2398 old->extent_offset, fs_info,
2399 path, record_one_backref,
2401 if (ret < 0 && ret != -ENOENT)
2404 /* no backref to be processed for this extent */
2406 list_del(&old->list);
2411 if (list_empty(&new->head))
2417 static int relink_is_mergable(struct extent_buffer *leaf,
2418 struct btrfs_file_extent_item *fi,
2419 struct new_sa_defrag_extent *new)
2421 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2424 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2427 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2430 if (btrfs_file_extent_encryption(leaf, fi) ||
2431 btrfs_file_extent_other_encoding(leaf, fi))
2438 * Note the backref might has changed, and in this case we just return 0.
2440 static noinline int relink_extent_backref(struct btrfs_path *path,
2441 struct sa_defrag_extent_backref *prev,
2442 struct sa_defrag_extent_backref *backref)
2444 struct btrfs_file_extent_item *extent;
2445 struct btrfs_file_extent_item *item;
2446 struct btrfs_ordered_extent *ordered;
2447 struct btrfs_trans_handle *trans;
2448 struct btrfs_fs_info *fs_info;
2449 struct btrfs_root *root;
2450 struct btrfs_key key;
2451 struct extent_buffer *leaf;
2452 struct old_sa_defrag_extent *old = backref->old;
2453 struct new_sa_defrag_extent *new = old->new;
2454 struct inode *src_inode = new->inode;
2455 struct inode *inode;
2456 struct extent_state *cached = NULL;
2465 if (prev && prev->root_id == backref->root_id &&
2466 prev->inum == backref->inum &&
2467 prev->file_pos + prev->num_bytes == backref->file_pos)
2470 /* step 1: get root */
2471 key.objectid = backref->root_id;
2472 key.type = BTRFS_ROOT_ITEM_KEY;
2473 key.offset = (u64)-1;
2475 fs_info = BTRFS_I(src_inode)->root->fs_info;
2476 index = srcu_read_lock(&fs_info->subvol_srcu);
2478 root = btrfs_read_fs_root_no_name(fs_info, &key);
2480 srcu_read_unlock(&fs_info->subvol_srcu, index);
2481 if (PTR_ERR(root) == -ENOENT)
2483 return PTR_ERR(root);
2486 if (btrfs_root_readonly(root)) {
2487 srcu_read_unlock(&fs_info->subvol_srcu, index);
2491 /* step 2: get inode */
2492 key.objectid = backref->inum;
2493 key.type = BTRFS_INODE_ITEM_KEY;
2496 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2497 if (IS_ERR(inode)) {
2498 srcu_read_unlock(&fs_info->subvol_srcu, index);
2502 srcu_read_unlock(&fs_info->subvol_srcu, index);
2504 /* step 3: relink backref */
2505 lock_start = backref->file_pos;
2506 lock_end = backref->file_pos + backref->num_bytes - 1;
2507 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2510 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2512 btrfs_put_ordered_extent(ordered);
2516 trans = btrfs_join_transaction(root);
2517 if (IS_ERR(trans)) {
2518 ret = PTR_ERR(trans);
2522 key.objectid = backref->inum;
2523 key.type = BTRFS_EXTENT_DATA_KEY;
2524 key.offset = backref->file_pos;
2526 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2529 } else if (ret > 0) {
2534 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2535 struct btrfs_file_extent_item);
2537 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2538 backref->generation)
2541 btrfs_release_path(path);
2543 start = backref->file_pos;
2544 if (backref->extent_offset < old->extent_offset + old->offset)
2545 start += old->extent_offset + old->offset -
2546 backref->extent_offset;
2548 len = min(backref->extent_offset + backref->num_bytes,
2549 old->extent_offset + old->offset + old->len);
2550 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2552 ret = btrfs_drop_extents(trans, root, inode, start,
2557 key.objectid = btrfs_ino(inode);
2558 key.type = BTRFS_EXTENT_DATA_KEY;
2561 path->leave_spinning = 1;
2563 struct btrfs_file_extent_item *fi;
2565 struct btrfs_key found_key;
2567 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2572 leaf = path->nodes[0];
2573 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2575 fi = btrfs_item_ptr(leaf, path->slots[0],
2576 struct btrfs_file_extent_item);
2577 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2579 if (extent_len + found_key.offset == start &&
2580 relink_is_mergable(leaf, fi, new)) {
2581 btrfs_set_file_extent_num_bytes(leaf, fi,
2583 btrfs_mark_buffer_dirty(leaf);
2584 inode_add_bytes(inode, len);
2590 btrfs_release_path(path);
2595 ret = btrfs_insert_empty_item(trans, root, path, &key,
2598 btrfs_abort_transaction(trans, root, ret);
2602 leaf = path->nodes[0];
2603 item = btrfs_item_ptr(leaf, path->slots[0],
2604 struct btrfs_file_extent_item);
2605 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2606 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2607 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2608 btrfs_set_file_extent_num_bytes(leaf, item, len);
2609 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2610 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2611 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2612 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2613 btrfs_set_file_extent_encryption(leaf, item, 0);
2614 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2616 btrfs_mark_buffer_dirty(leaf);
2617 inode_add_bytes(inode, len);
2618 btrfs_release_path(path);
2620 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2622 backref->root_id, backref->inum,
2623 new->file_pos); /* start - extent_offset */
2625 btrfs_abort_transaction(trans, root, ret);
2631 btrfs_release_path(path);
2632 path->leave_spinning = 0;
2633 btrfs_end_transaction(trans, root);
2635 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2641 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2643 struct old_sa_defrag_extent *old, *tmp;
2648 list_for_each_entry_safe(old, tmp, &new->head, list) {
2654 static void relink_file_extents(struct new_sa_defrag_extent *new)
2656 struct btrfs_path *path;
2657 struct sa_defrag_extent_backref *backref;
2658 struct sa_defrag_extent_backref *prev = NULL;
2659 struct inode *inode;
2660 struct btrfs_root *root;
2661 struct rb_node *node;
2665 root = BTRFS_I(inode)->root;
2667 path = btrfs_alloc_path();
2671 if (!record_extent_backrefs(path, new)) {
2672 btrfs_free_path(path);
2675 btrfs_release_path(path);
2678 node = rb_first(&new->root);
2681 rb_erase(node, &new->root);
2683 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2685 ret = relink_extent_backref(path, prev, backref);
2698 btrfs_free_path(path);
2700 free_sa_defrag_extent(new);
2702 atomic_dec(&root->fs_info->defrag_running);
2703 wake_up(&root->fs_info->transaction_wait);
2706 static struct new_sa_defrag_extent *
2707 record_old_file_extents(struct inode *inode,
2708 struct btrfs_ordered_extent *ordered)
2710 struct btrfs_root *root = BTRFS_I(inode)->root;
2711 struct btrfs_path *path;
2712 struct btrfs_key key;
2713 struct old_sa_defrag_extent *old;
2714 struct new_sa_defrag_extent *new;
2717 new = kmalloc(sizeof(*new), GFP_NOFS);
2722 new->file_pos = ordered->file_offset;
2723 new->len = ordered->len;
2724 new->bytenr = ordered->start;
2725 new->disk_len = ordered->disk_len;
2726 new->compress_type = ordered->compress_type;
2727 new->root = RB_ROOT;
2728 INIT_LIST_HEAD(&new->head);
2730 path = btrfs_alloc_path();
2734 key.objectid = btrfs_ino(inode);
2735 key.type = BTRFS_EXTENT_DATA_KEY;
2736 key.offset = new->file_pos;
2738 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2741 if (ret > 0 && path->slots[0] > 0)
2744 /* find out all the old extents for the file range */
2746 struct btrfs_file_extent_item *extent;
2747 struct extent_buffer *l;
2756 slot = path->slots[0];
2758 if (slot >= btrfs_header_nritems(l)) {
2759 ret = btrfs_next_leaf(root, path);
2767 btrfs_item_key_to_cpu(l, &key, slot);
2769 if (key.objectid != btrfs_ino(inode))
2771 if (key.type != BTRFS_EXTENT_DATA_KEY)
2773 if (key.offset >= new->file_pos + new->len)
2776 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2778 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2779 if (key.offset + num_bytes < new->file_pos)
2782 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2786 extent_offset = btrfs_file_extent_offset(l, extent);
2788 old = kmalloc(sizeof(*old), GFP_NOFS);
2792 offset = max(new->file_pos, key.offset);
2793 end = min(new->file_pos + new->len, key.offset + num_bytes);
2795 old->bytenr = disk_bytenr;
2796 old->extent_offset = extent_offset;
2797 old->offset = offset - key.offset;
2798 old->len = end - offset;
2801 list_add_tail(&old->list, &new->head);
2807 btrfs_free_path(path);
2808 atomic_inc(&root->fs_info->defrag_running);
2813 btrfs_free_path(path);
2815 free_sa_defrag_extent(new);
2819 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2822 struct btrfs_block_group_cache *cache;
2824 cache = btrfs_lookup_block_group(root->fs_info, start);
2827 spin_lock(&cache->lock);
2828 cache->delalloc_bytes -= len;
2829 spin_unlock(&cache->lock);
2831 btrfs_put_block_group(cache);
2834 /* as ordered data IO finishes, this gets called so we can finish
2835 * an ordered extent if the range of bytes in the file it covers are
2838 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2840 struct inode *inode = ordered_extent->inode;
2841 struct btrfs_root *root = BTRFS_I(inode)->root;
2842 struct btrfs_trans_handle *trans = NULL;
2843 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2844 struct extent_state *cached_state = NULL;
2845 struct new_sa_defrag_extent *new = NULL;
2846 int compress_type = 0;
2848 u64 logical_len = ordered_extent->len;
2850 bool truncated = false;
2852 nolock = btrfs_is_free_space_inode(inode);
2854 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2859 btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2860 ordered_extent->file_offset +
2861 ordered_extent->len - 1);
2863 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2865 logical_len = ordered_extent->truncated_len;
2866 /* Truncated the entire extent, don't bother adding */
2871 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2872 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2875 * For mwrite(mmap + memset to write) case, we still reserve
2876 * space for NOCOW range.
2877 * As NOCOW won't cause a new delayed ref, just free the space
2879 btrfs_qgroup_free_data(inode, ordered_extent->file_offset,
2880 ordered_extent->len);
2881 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2883 trans = btrfs_join_transaction_nolock(root);
2885 trans = btrfs_join_transaction(root);
2886 if (IS_ERR(trans)) {
2887 ret = PTR_ERR(trans);
2891 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2892 ret = btrfs_update_inode_fallback(trans, root, inode);
2893 if (ret) /* -ENOMEM or corruption */
2894 btrfs_abort_transaction(trans, root, ret);
2898 lock_extent_bits(io_tree, ordered_extent->file_offset,
2899 ordered_extent->file_offset + ordered_extent->len - 1,
2902 ret = test_range_bit(io_tree, ordered_extent->file_offset,
2903 ordered_extent->file_offset + ordered_extent->len - 1,
2904 EXTENT_DEFRAG, 1, cached_state);
2906 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2907 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2908 /* the inode is shared */
2909 new = record_old_file_extents(inode, ordered_extent);
2911 clear_extent_bit(io_tree, ordered_extent->file_offset,
2912 ordered_extent->file_offset + ordered_extent->len - 1,
2913 EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2917 trans = btrfs_join_transaction_nolock(root);
2919 trans = btrfs_join_transaction(root);
2920 if (IS_ERR(trans)) {
2921 ret = PTR_ERR(trans);
2926 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2928 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2929 compress_type = ordered_extent->compress_type;
2930 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2931 BUG_ON(compress_type);
2932 ret = btrfs_mark_extent_written(trans, inode,
2933 ordered_extent->file_offset,
2934 ordered_extent->file_offset +
2937 BUG_ON(root == root->fs_info->tree_root);
2938 ret = insert_reserved_file_extent(trans, inode,
2939 ordered_extent->file_offset,
2940 ordered_extent->start,
2941 ordered_extent->disk_len,
2942 logical_len, logical_len,
2943 compress_type, 0, 0,
2944 BTRFS_FILE_EXTENT_REG);
2946 btrfs_release_delalloc_bytes(root,
2947 ordered_extent->start,
2948 ordered_extent->disk_len);
2950 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2951 ordered_extent->file_offset, ordered_extent->len,
2954 btrfs_abort_transaction(trans, root, ret);
2958 add_pending_csums(trans, inode, ordered_extent->file_offset,
2959 &ordered_extent->list);
2961 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2962 ret = btrfs_update_inode_fallback(trans, root, inode);
2963 if (ret) { /* -ENOMEM or corruption */
2964 btrfs_abort_transaction(trans, root, ret);
2969 unlock_extent_cached(io_tree, ordered_extent->file_offset,
2970 ordered_extent->file_offset +
2971 ordered_extent->len - 1, &cached_state, GFP_NOFS);
2973 if (root != root->fs_info->tree_root)
2974 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2976 btrfs_end_transaction(trans, root);
2978 if (ret || truncated) {
2982 start = ordered_extent->file_offset + logical_len;
2984 start = ordered_extent->file_offset;
2985 end = ordered_extent->file_offset + ordered_extent->len - 1;
2986 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2988 /* Drop the cache for the part of the extent we didn't write. */
2989 btrfs_drop_extent_cache(inode, start, end, 0);
2992 * If the ordered extent had an IOERR or something else went
2993 * wrong we need to return the space for this ordered extent
2994 * back to the allocator. We only free the extent in the
2995 * truncated case if we didn't write out the extent at all.
2997 if ((ret || !logical_len) &&
2998 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2999 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3000 btrfs_free_reserved_extent(root, ordered_extent->start,
3001 ordered_extent->disk_len, 1);
3006 * This needs to be done to make sure anybody waiting knows we are done
3007 * updating everything for this ordered extent.
3009 btrfs_remove_ordered_extent(inode, ordered_extent);
3011 /* for snapshot-aware defrag */
3014 free_sa_defrag_extent(new);
3015 atomic_dec(&root->fs_info->defrag_running);
3017 relink_file_extents(new);
3022 btrfs_put_ordered_extent(ordered_extent);
3023 /* once for the tree */
3024 btrfs_put_ordered_extent(ordered_extent);
3029 static void finish_ordered_fn(struct btrfs_work *work)
3031 struct btrfs_ordered_extent *ordered_extent;
3032 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3033 btrfs_finish_ordered_io(ordered_extent);
3036 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3037 struct extent_state *state, int uptodate)
3039 struct inode *inode = page->mapping->host;
3040 struct btrfs_root *root = BTRFS_I(inode)->root;
3041 struct btrfs_ordered_extent *ordered_extent = NULL;
3042 struct btrfs_workqueue *wq;
3043 btrfs_work_func_t func;
3045 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3047 ClearPagePrivate2(page);
3048 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3049 end - start + 1, uptodate))
3052 if (btrfs_is_free_space_inode(inode)) {
3053 wq = root->fs_info->endio_freespace_worker;
3054 func = btrfs_freespace_write_helper;
3056 wq = root->fs_info->endio_write_workers;
3057 func = btrfs_endio_write_helper;
3060 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3062 btrfs_queue_work(wq, &ordered_extent->work);
3067 static int __readpage_endio_check(struct inode *inode,
3068 struct btrfs_io_bio *io_bio,
3069 int icsum, struct page *page,
3070 int pgoff, u64 start, size_t len)
3076 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3078 kaddr = kmap_atomic(page);
3079 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3080 btrfs_csum_final(csum, (char *)&csum);
3081 if (csum != csum_expected)
3084 kunmap_atomic(kaddr);
3087 btrfs_warn_rl(BTRFS_I(inode)->root->fs_info,
3088 "csum failed ino %llu off %llu csum %u expected csum %u",
3089 btrfs_ino(inode), start, csum, csum_expected);
3090 memset(kaddr + pgoff, 1, len);
3091 flush_dcache_page(page);
3092 kunmap_atomic(kaddr);
3093 if (csum_expected == 0)
3099 * when reads are done, we need to check csums to verify the data is correct
3100 * if there's a match, we allow the bio to finish. If not, the code in
3101 * extent_io.c will try to find good copies for us.
3103 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3104 u64 phy_offset, struct page *page,
3105 u64 start, u64 end, int mirror)
3107 size_t offset = start - page_offset(page);
3108 struct inode *inode = page->mapping->host;
3109 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3110 struct btrfs_root *root = BTRFS_I(inode)->root;
3112 if (PageChecked(page)) {
3113 ClearPageChecked(page);
3117 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3120 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3121 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3122 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3127 phy_offset >>= inode->i_sb->s_blocksize_bits;
3128 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3129 start, (size_t)(end - start + 1));
3132 void btrfs_add_delayed_iput(struct inode *inode)
3134 struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3135 struct btrfs_inode *binode = BTRFS_I(inode);
3137 if (atomic_add_unless(&inode->i_count, -1, 1))
3140 spin_lock(&fs_info->delayed_iput_lock);
3141 if (binode->delayed_iput_count == 0) {
3142 ASSERT(list_empty(&binode->delayed_iput));
3143 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3145 binode->delayed_iput_count++;
3147 spin_unlock(&fs_info->delayed_iput_lock);
3150 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3152 struct btrfs_fs_info *fs_info = root->fs_info;
3154 spin_lock(&fs_info->delayed_iput_lock);
3155 while (!list_empty(&fs_info->delayed_iputs)) {
3156 struct btrfs_inode *inode;
3158 inode = list_first_entry(&fs_info->delayed_iputs,
3159 struct btrfs_inode, delayed_iput);
3160 if (inode->delayed_iput_count) {
3161 inode->delayed_iput_count--;
3162 list_move_tail(&inode->delayed_iput,
3163 &fs_info->delayed_iputs);
3165 list_del_init(&inode->delayed_iput);
3167 spin_unlock(&fs_info->delayed_iput_lock);
3168 iput(&inode->vfs_inode);
3169 spin_lock(&fs_info->delayed_iput_lock);
3171 spin_unlock(&fs_info->delayed_iput_lock);
3175 * This is called in transaction commit time. If there are no orphan
3176 * files in the subvolume, it removes orphan item and frees block_rsv
3179 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3180 struct btrfs_root *root)
3182 struct btrfs_block_rsv *block_rsv;
3185 if (atomic_read(&root->orphan_inodes) ||
3186 root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3189 spin_lock(&root->orphan_lock);
3190 if (atomic_read(&root->orphan_inodes)) {
3191 spin_unlock(&root->orphan_lock);
3195 if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3196 spin_unlock(&root->orphan_lock);
3200 block_rsv = root->orphan_block_rsv;
3201 root->orphan_block_rsv = NULL;
3202 spin_unlock(&root->orphan_lock);
3204 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3205 btrfs_root_refs(&root->root_item) > 0) {
3206 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3207 root->root_key.objectid);
3209 btrfs_abort_transaction(trans, root, ret);
3211 clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3216 WARN_ON(block_rsv->size > 0);
3217 btrfs_free_block_rsv(root, block_rsv);
3222 * This creates an orphan entry for the given inode in case something goes
3223 * wrong in the middle of an unlink/truncate.
3225 * NOTE: caller of this function should reserve 5 units of metadata for
3228 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3230 struct btrfs_root *root = BTRFS_I(inode)->root;
3231 struct btrfs_block_rsv *block_rsv = NULL;
3236 if (!root->orphan_block_rsv) {
3237 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3242 spin_lock(&root->orphan_lock);
3243 if (!root->orphan_block_rsv) {
3244 root->orphan_block_rsv = block_rsv;
3245 } else if (block_rsv) {
3246 btrfs_free_block_rsv(root, block_rsv);
3250 if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3251 &BTRFS_I(inode)->runtime_flags)) {
3254 * For proper ENOSPC handling, we should do orphan
3255 * cleanup when mounting. But this introduces backward
3256 * compatibility issue.
3258 if (!xchg(&root->orphan_item_inserted, 1))
3264 atomic_inc(&root->orphan_inodes);
3267 if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3268 &BTRFS_I(inode)->runtime_flags))
3270 spin_unlock(&root->orphan_lock);
3272 /* grab metadata reservation from transaction handle */
3274 ret = btrfs_orphan_reserve_metadata(trans, inode);
3275 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3278 /* insert an orphan item to track this unlinked/truncated file */
3280 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3282 atomic_dec(&root->orphan_inodes);
3284 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3285 &BTRFS_I(inode)->runtime_flags);
3286 btrfs_orphan_release_metadata(inode);
3288 if (ret != -EEXIST) {
3289 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3290 &BTRFS_I(inode)->runtime_flags);
3291 btrfs_abort_transaction(trans, root, ret);
3298 /* insert an orphan item to track subvolume contains orphan files */
3300 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3301 root->root_key.objectid);
3302 if (ret && ret != -EEXIST) {
3303 btrfs_abort_transaction(trans, root, ret);
3311 * We have done the truncate/delete so we can go ahead and remove the orphan
3312 * item for this particular inode.
3314 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3315 struct inode *inode)
3317 struct btrfs_root *root = BTRFS_I(inode)->root;
3318 int delete_item = 0;
3319 int release_rsv = 0;
3322 spin_lock(&root->orphan_lock);
3323 if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3324 &BTRFS_I(inode)->runtime_flags))
3327 if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3328 &BTRFS_I(inode)->runtime_flags))
3330 spin_unlock(&root->orphan_lock);
3333 atomic_dec(&root->orphan_inodes);
3335 ret = btrfs_del_orphan_item(trans, root,
3340 btrfs_orphan_release_metadata(inode);
3346 * this cleans up any orphans that may be left on the list from the last use
3349 int btrfs_orphan_cleanup(struct btrfs_root *root)
3351 struct btrfs_path *path;
3352 struct extent_buffer *leaf;
3353 struct btrfs_key key, found_key;
3354 struct btrfs_trans_handle *trans;
3355 struct inode *inode;
3356 u64 last_objectid = 0;
3357 int ret = 0, nr_unlink = 0, nr_truncate = 0;
3359 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3362 path = btrfs_alloc_path();
3367 path->reada = READA_BACK;
3369 key.objectid = BTRFS_ORPHAN_OBJECTID;
3370 key.type = BTRFS_ORPHAN_ITEM_KEY;
3371 key.offset = (u64)-1;
3374 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3379 * if ret == 0 means we found what we were searching for, which
3380 * is weird, but possible, so only screw with path if we didn't
3381 * find the key and see if we have stuff that matches
3385 if (path->slots[0] == 0)
3390 /* pull out the item */
3391 leaf = path->nodes[0];
3392 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3394 /* make sure the item matches what we want */
3395 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3397 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3400 /* release the path since we're done with it */
3401 btrfs_release_path(path);
3404 * this is where we are basically btrfs_lookup, without the
3405 * crossing root thing. we store the inode number in the
3406 * offset of the orphan item.
3409 if (found_key.offset == last_objectid) {
3410 btrfs_err(root->fs_info,
3411 "Error removing orphan entry, stopping orphan cleanup");
3416 last_objectid = found_key.offset;
3418 found_key.objectid = found_key.offset;
3419 found_key.type = BTRFS_INODE_ITEM_KEY;
3420 found_key.offset = 0;
3421 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3422 ret = PTR_ERR_OR_ZERO(inode);
3423 if (ret && ret != -ESTALE)
3426 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3427 struct btrfs_root *dead_root;
3428 struct btrfs_fs_info *fs_info = root->fs_info;
3429 int is_dead_root = 0;
3432 * this is an orphan in the tree root. Currently these
3433 * could come from 2 sources:
3434 * a) a snapshot deletion in progress
3435 * b) a free space cache inode
3436 * We need to distinguish those two, as the snapshot
3437 * orphan must not get deleted.
3438 * find_dead_roots already ran before us, so if this
3439 * is a snapshot deletion, we should find the root
3440 * in the dead_roots list
3442 spin_lock(&fs_info->trans_lock);
3443 list_for_each_entry(dead_root, &fs_info->dead_roots,
3445 if (dead_root->root_key.objectid ==
3446 found_key.objectid) {
3451 spin_unlock(&fs_info->trans_lock);
3453 /* prevent this orphan from being found again */
3454 key.offset = found_key.objectid - 1;
3459 * Inode is already gone but the orphan item is still there,
3460 * kill the orphan item.
3462 if (ret == -ESTALE) {
3463 trans = btrfs_start_transaction(root, 1);
3464 if (IS_ERR(trans)) {
3465 ret = PTR_ERR(trans);
3468 btrfs_debug(root->fs_info, "auto deleting %Lu",
3469 found_key.objectid);
3470 ret = btrfs_del_orphan_item(trans, root,
3471 found_key.objectid);
3472 btrfs_end_transaction(trans, root);
3479 * add this inode to the orphan list so btrfs_orphan_del does
3480 * the proper thing when we hit it
3482 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3483 &BTRFS_I(inode)->runtime_flags);
3484 atomic_inc(&root->orphan_inodes);
3486 /* if we have links, this was a truncate, lets do that */
3487 if (inode->i_nlink) {
3488 if (WARN_ON(!S_ISREG(inode->i_mode))) {
3494 /* 1 for the orphan item deletion. */
3495 trans = btrfs_start_transaction(root, 1);
3496 if (IS_ERR(trans)) {
3498 ret = PTR_ERR(trans);
3501 ret = btrfs_orphan_add(trans, inode);
3502 btrfs_end_transaction(trans, root);
3508 ret = btrfs_truncate(inode);
3510 btrfs_orphan_del(NULL, inode);
3515 /* this will do delete_inode and everything for us */
3520 /* release the path since we're done with it */
3521 btrfs_release_path(path);
3523 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3525 if (root->orphan_block_rsv)
3526 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3529 if (root->orphan_block_rsv ||
3530 test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3531 trans = btrfs_join_transaction(root);
3533 btrfs_end_transaction(trans, root);
3537 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3539 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3543 btrfs_err(root->fs_info,
3544 "could not do orphan cleanup %d", ret);
3545 btrfs_free_path(path);
3550 * very simple check to peek ahead in the leaf looking for xattrs. If we
3551 * don't find any xattrs, we know there can't be any acls.
3553 * slot is the slot the inode is in, objectid is the objectid of the inode
3555 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3556 int slot, u64 objectid,
3557 int *first_xattr_slot)
3559 u32 nritems = btrfs_header_nritems(leaf);
3560 struct btrfs_key found_key;
3561 static u64 xattr_access = 0;
3562 static u64 xattr_default = 0;
3565 if (!xattr_access) {
3566 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3567 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3568 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3569 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3573 *first_xattr_slot = -1;
3574 while (slot < nritems) {
3575 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3577 /* we found a different objectid, there must not be acls */
3578 if (found_key.objectid != objectid)
3581 /* we found an xattr, assume we've got an acl */
3582 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3583 if (*first_xattr_slot == -1)
3584 *first_xattr_slot = slot;
3585 if (found_key.offset == xattr_access ||
3586 found_key.offset == xattr_default)
3591 * we found a key greater than an xattr key, there can't
3592 * be any acls later on
3594 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3601 * it goes inode, inode backrefs, xattrs, extents,
3602 * so if there are a ton of hard links to an inode there can
3603 * be a lot of backrefs. Don't waste time searching too hard,
3604 * this is just an optimization
3609 /* we hit the end of the leaf before we found an xattr or
3610 * something larger than an xattr. We have to assume the inode
3613 if (*first_xattr_slot == -1)
3614 *first_xattr_slot = slot;
3619 * read an inode from the btree into the in-memory inode
3621 static void btrfs_read_locked_inode(struct inode *inode)
3623 struct btrfs_path *path;
3624 struct extent_buffer *leaf;
3625 struct btrfs_inode_item *inode_item;
3626 struct btrfs_root *root = BTRFS_I(inode)->root;
3627 struct btrfs_key location;
3632 bool filled = false;
3633 int first_xattr_slot;
3635 ret = btrfs_fill_inode(inode, &rdev);
3639 path = btrfs_alloc_path();
3643 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3645 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3649 leaf = path->nodes[0];
3654 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3655 struct btrfs_inode_item);
3656 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3657 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3658 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3659 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3660 btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3662 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3663 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3665 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3666 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3668 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3669 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3671 BTRFS_I(inode)->i_otime.tv_sec =
3672 btrfs_timespec_sec(leaf, &inode_item->otime);
3673 BTRFS_I(inode)->i_otime.tv_nsec =
3674 btrfs_timespec_nsec(leaf, &inode_item->otime);
3676 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3677 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3678 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3680 inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3681 inode->i_generation = BTRFS_I(inode)->generation;
3683 rdev = btrfs_inode_rdev(leaf, inode_item);
3685 BTRFS_I(inode)->index_cnt = (u64)-1;
3686 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3690 * If we were modified in the current generation and evicted from memory
3691 * and then re-read we need to do a full sync since we don't have any
3692 * idea about which extents were modified before we were evicted from
3695 * This is required for both inode re-read from disk and delayed inode
3696 * in delayed_nodes_tree.
3698 if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3699 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3700 &BTRFS_I(inode)->runtime_flags);
3703 * We don't persist the id of the transaction where an unlink operation
3704 * against the inode was last made. So here we assume the inode might
3705 * have been evicted, and therefore the exact value of last_unlink_trans
3706 * lost, and set it to last_trans to avoid metadata inconsistencies
3707 * between the inode and its parent if the inode is fsync'ed and the log
3708 * replayed. For example, in the scenario:
3711 * ln mydir/foo mydir/bar
3714 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3715 * xfs_io -c fsync mydir/foo
3717 * mount fs, triggers fsync log replay
3719 * We must make sure that when we fsync our inode foo we also log its
3720 * parent inode, otherwise after log replay the parent still has the
3721 * dentry with the "bar" name but our inode foo has a link count of 1
3722 * and doesn't have an inode ref with the name "bar" anymore.
3724 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3725 * but it guarantees correctness at the expense of ocassional full
3726 * transaction commits on fsync if our inode is a directory, or if our
3727 * inode is not a directory, logging its parent unnecessarily.
3729 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3732 if (inode->i_nlink != 1 ||
3733 path->slots[0] >= btrfs_header_nritems(leaf))
3736 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3737 if (location.objectid != btrfs_ino(inode))
3740 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3741 if (location.type == BTRFS_INODE_REF_KEY) {
3742 struct btrfs_inode_ref *ref;
3744 ref = (struct btrfs_inode_ref *)ptr;
3745 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3746 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3747 struct btrfs_inode_extref *extref;
3749 extref = (struct btrfs_inode_extref *)ptr;
3750 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3755 * try to precache a NULL acl entry for files that don't have
3756 * any xattrs or acls
3758 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3759 btrfs_ino(inode), &first_xattr_slot);
3760 if (first_xattr_slot != -1) {
3761 path->slots[0] = first_xattr_slot;
3762 ret = btrfs_load_inode_props(inode, path);
3764 btrfs_err(root->fs_info,
3765 "error loading props for ino %llu (root %llu): %d",
3767 root->root_key.objectid, ret);
3769 btrfs_free_path(path);
3772 cache_no_acl(inode);
3774 switch (inode->i_mode & S_IFMT) {
3776 inode->i_mapping->a_ops = &btrfs_aops;
3777 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3778 inode->i_fop = &btrfs_file_operations;
3779 inode->i_op = &btrfs_file_inode_operations;
3782 inode->i_fop = &btrfs_dir_file_operations;
3783 if (root == root->fs_info->tree_root)
3784 inode->i_op = &btrfs_dir_ro_inode_operations;
3786 inode->i_op = &btrfs_dir_inode_operations;
3789 inode->i_op = &btrfs_symlink_inode_operations;
3790 inode_nohighmem(inode);
3791 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3794 inode->i_op = &btrfs_special_inode_operations;
3795 init_special_inode(inode, inode->i_mode, rdev);
3799 btrfs_update_iflags(inode);
3803 btrfs_free_path(path);
3804 make_bad_inode(inode);
3808 * given a leaf and an inode, copy the inode fields into the leaf
3810 static void fill_inode_item(struct btrfs_trans_handle *trans,
3811 struct extent_buffer *leaf,
3812 struct btrfs_inode_item *item,
3813 struct inode *inode)
3815 struct btrfs_map_token token;
3817 btrfs_init_map_token(&token);
3819 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3820 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3821 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3823 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3824 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3826 btrfs_set_token_timespec_sec(leaf, &item->atime,
3827 inode->i_atime.tv_sec, &token);
3828 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3829 inode->i_atime.tv_nsec, &token);
3831 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3832 inode->i_mtime.tv_sec, &token);
3833 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3834 inode->i_mtime.tv_nsec, &token);
3836 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3837 inode->i_ctime.tv_sec, &token);
3838 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3839 inode->i_ctime.tv_nsec, &token);
3841 btrfs_set_token_timespec_sec(leaf, &item->otime,
3842 BTRFS_I(inode)->i_otime.tv_sec, &token);
3843 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3844 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3846 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3848 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3850 btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3851 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3852 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3853 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3854 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3858 * copy everything in the in-memory inode into the btree.
3860 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3861 struct btrfs_root *root, struct inode *inode)
3863 struct btrfs_inode_item *inode_item;
3864 struct btrfs_path *path;
3865 struct extent_buffer *leaf;
3868 path = btrfs_alloc_path();
3872 path->leave_spinning = 1;
3873 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3881 leaf = path->nodes[0];
3882 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3883 struct btrfs_inode_item);
3885 fill_inode_item(trans, leaf, inode_item, inode);
3886 btrfs_mark_buffer_dirty(leaf);
3887 btrfs_set_inode_last_trans(trans, inode);
3890 btrfs_free_path(path);
3895 * copy everything in the in-memory inode into the btree.
3897 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3898 struct btrfs_root *root, struct inode *inode)
3903 * If the inode is a free space inode, we can deadlock during commit
3904 * if we put it into the delayed code.
3906 * The data relocation inode should also be directly updated
3909 if (!btrfs_is_free_space_inode(inode)
3910 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3911 && !root->fs_info->log_root_recovering) {
3912 btrfs_update_root_times(trans, root);
3914 ret = btrfs_delayed_update_inode(trans, root, inode);
3916 btrfs_set_inode_last_trans(trans, inode);
3920 return btrfs_update_inode_item(trans, root, inode);
3923 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3924 struct btrfs_root *root,
3925 struct inode *inode)
3929 ret = btrfs_update_inode(trans, root, inode);
3931 return btrfs_update_inode_item(trans, root, inode);
3936 * unlink helper that gets used here in inode.c and in the tree logging
3937 * recovery code. It remove a link in a directory with a given name, and
3938 * also drops the back refs in the inode to the directory
3940 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3941 struct btrfs_root *root,
3942 struct inode *dir, struct inode *inode,
3943 const char *name, int name_len)
3945 struct btrfs_path *path;
3947 struct extent_buffer *leaf;
3948 struct btrfs_dir_item *di;
3949 struct btrfs_key key;
3951 u64 ino = btrfs_ino(inode);
3952 u64 dir_ino = btrfs_ino(dir);
3954 path = btrfs_alloc_path();
3960 path->leave_spinning = 1;
3961 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3962 name, name_len, -1);
3971 leaf = path->nodes[0];
3972 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3973 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3976 btrfs_release_path(path);
3979 * If we don't have dir index, we have to get it by looking up
3980 * the inode ref, since we get the inode ref, remove it directly,
3981 * it is unnecessary to do delayed deletion.
3983 * But if we have dir index, needn't search inode ref to get it.
3984 * Since the inode ref is close to the inode item, it is better
3985 * that we delay to delete it, and just do this deletion when
3986 * we update the inode item.
3988 if (BTRFS_I(inode)->dir_index) {
3989 ret = btrfs_delayed_delete_inode_ref(inode);
3991 index = BTRFS_I(inode)->dir_index;
3996 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3999 btrfs_info(root->fs_info,
4000 "failed to delete reference to %.*s, inode %llu parent %llu",
4001 name_len, name, ino, dir_ino);
4002 btrfs_abort_transaction(trans, root, ret);
4006 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4008 btrfs_abort_transaction(trans, root, ret);
4012 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
4014 if (ret != 0 && ret != -ENOENT) {
4015 btrfs_abort_transaction(trans, root, ret);
4019 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
4024 btrfs_abort_transaction(trans, root, ret);
4026 btrfs_free_path(path);
4030 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4031 inode_inc_iversion(inode);
4032 inode_inc_iversion(dir);
4033 inode->i_ctime = dir->i_mtime =
4034 dir->i_ctime = current_fs_time(inode->i_sb);
4035 ret = btrfs_update_inode(trans, root, dir);
4040 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4041 struct btrfs_root *root,
4042 struct inode *dir, struct inode *inode,
4043 const char *name, int name_len)
4046 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4049 ret = btrfs_update_inode(trans, root, inode);
4055 * helper to start transaction for unlink and rmdir.
4057 * unlink and rmdir are special in btrfs, they do not always free space, so
4058 * if we cannot make our reservations the normal way try and see if there is
4059 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4060 * allow the unlink to occur.
4062 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4064 struct btrfs_root *root = BTRFS_I(dir)->root;
4067 * 1 for the possible orphan item
4068 * 1 for the dir item
4069 * 1 for the dir index
4070 * 1 for the inode ref
4073 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4076 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4078 struct btrfs_root *root = BTRFS_I(dir)->root;
4079 struct btrfs_trans_handle *trans;
4080 struct inode *inode = d_inode(dentry);
4083 trans = __unlink_start_trans(dir);
4085 return PTR_ERR(trans);
4087 btrfs_record_unlink_dir(trans, dir, d_inode(dentry), 0);
4089 ret = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4090 dentry->d_name.name, dentry->d_name.len);
4094 if (inode->i_nlink == 0) {
4095 ret = btrfs_orphan_add(trans, inode);
4101 btrfs_end_transaction(trans, root);
4102 btrfs_btree_balance_dirty(root);
4106 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4107 struct btrfs_root *root,
4108 struct inode *dir, u64 objectid,
4109 const char *name, int name_len)
4111 struct btrfs_path *path;
4112 struct extent_buffer *leaf;
4113 struct btrfs_dir_item *di;
4114 struct btrfs_key key;
4117 u64 dir_ino = btrfs_ino(dir);
4119 path = btrfs_alloc_path();
4123 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4124 name, name_len, -1);
4125 if (IS_ERR_OR_NULL(di)) {
4133 leaf = path->nodes[0];
4134 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4135 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4136 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4138 btrfs_abort_transaction(trans, root, ret);
4141 btrfs_release_path(path);
4143 ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4144 objectid, root->root_key.objectid,
4145 dir_ino, &index, name, name_len);
4147 if (ret != -ENOENT) {
4148 btrfs_abort_transaction(trans, root, ret);
4151 di = btrfs_search_dir_index_item(root, path, dir_ino,
4153 if (IS_ERR_OR_NULL(di)) {
4158 btrfs_abort_transaction(trans, root, ret);
4162 leaf = path->nodes[0];
4163 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4164 btrfs_release_path(path);
4167 btrfs_release_path(path);
4169 ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4171 btrfs_abort_transaction(trans, root, ret);
4175 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4176 inode_inc_iversion(dir);
4177 dir->i_mtime = dir->i_ctime = current_fs_time(dir->i_sb);
4178 ret = btrfs_update_inode_fallback(trans, root, dir);
4180 btrfs_abort_transaction(trans, root, ret);
4182 btrfs_free_path(path);
4186 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4188 struct inode *inode = d_inode(dentry);
4190 struct btrfs_root *root = BTRFS_I(dir)->root;
4191 struct btrfs_trans_handle *trans;
4193 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4195 if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4198 trans = __unlink_start_trans(dir);
4200 return PTR_ERR(trans);
4202 if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4203 err = btrfs_unlink_subvol(trans, root, dir,
4204 BTRFS_I(inode)->location.objectid,
4205 dentry->d_name.name,
4206 dentry->d_name.len);
4210 err = btrfs_orphan_add(trans, inode);
4214 /* now the directory is empty */
4215 err = btrfs_unlink_inode(trans, root, dir, d_inode(dentry),
4216 dentry->d_name.name, dentry->d_name.len);
4218 btrfs_i_size_write(inode, 0);
4220 btrfs_end_transaction(trans, root);
4221 btrfs_btree_balance_dirty(root);
4226 static int truncate_space_check(struct btrfs_trans_handle *trans,
4227 struct btrfs_root *root,
4233 * This is only used to apply pressure to the enospc system, we don't
4234 * intend to use this reservation at all.
4236 bytes_deleted = btrfs_csum_bytes_to_leaves(root, bytes_deleted);
4237 bytes_deleted *= root->nodesize;
4238 ret = btrfs_block_rsv_add(root, &root->fs_info->trans_block_rsv,
4239 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4241 trace_btrfs_space_reservation(root->fs_info, "transaction",
4244 trans->bytes_reserved += bytes_deleted;
4250 static int truncate_inline_extent(struct inode *inode,
4251 struct btrfs_path *path,
4252 struct btrfs_key *found_key,
4256 struct extent_buffer *leaf = path->nodes[0];
4257 int slot = path->slots[0];
4258 struct btrfs_file_extent_item *fi;
4259 u32 size = (u32)(new_size - found_key->offset);
4260 struct btrfs_root *root = BTRFS_I(inode)->root;
4262 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
4264 if (btrfs_file_extent_compression(leaf, fi) != BTRFS_COMPRESS_NONE) {
4265 loff_t offset = new_size;
4266 loff_t page_end = ALIGN(offset, PAGE_SIZE);
4269 * Zero out the remaining of the last page of our inline extent,
4270 * instead of directly truncating our inline extent here - that
4271 * would be much more complex (decompressing all the data, then
4272 * compressing the truncated data, which might be bigger than
4273 * the size of the inline extent, resize the extent, etc).
4274 * We release the path because to get the page we might need to
4275 * read the extent item from disk (data not in the page cache).
4277 btrfs_release_path(path);
4278 return btrfs_truncate_block(inode, offset, page_end - offset,
4282 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4283 size = btrfs_file_extent_calc_inline_size(size);
4284 btrfs_truncate_item(root, path, size, 1);
4286 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4287 inode_sub_bytes(inode, item_end + 1 - new_size);
4293 * this can truncate away extent items, csum items and directory items.
4294 * It starts at a high offset and removes keys until it can't find
4295 * any higher than new_size
4297 * csum items that cross the new i_size are truncated to the new size
4300 * min_type is the minimum key type to truncate down to. If set to 0, this
4301 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4303 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4304 struct btrfs_root *root,
4305 struct inode *inode,
4306 u64 new_size, u32 min_type)
4308 struct btrfs_path *path;
4309 struct extent_buffer *leaf;
4310 struct btrfs_file_extent_item *fi;
4311 struct btrfs_key key;
4312 struct btrfs_key found_key;
4313 u64 extent_start = 0;
4314 u64 extent_num_bytes = 0;
4315 u64 extent_offset = 0;
4317 u64 last_size = new_size;
4318 u32 found_type = (u8)-1;
4321 int pending_del_nr = 0;
4322 int pending_del_slot = 0;
4323 int extent_type = -1;
4326 u64 ino = btrfs_ino(inode);
4327 u64 bytes_deleted = 0;
4329 bool should_throttle = 0;
4330 bool should_end = 0;
4332 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4335 * for non-free space inodes and ref cows, we want to back off from
4338 if (!btrfs_is_free_space_inode(inode) &&
4339 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4342 path = btrfs_alloc_path();
4345 path->reada = READA_BACK;
4348 * We want to drop from the next block forward in case this new size is
4349 * not block aligned since we will be keeping the last block of the
4350 * extent just the way it is.
4352 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4353 root == root->fs_info->tree_root)
4354 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4355 root->sectorsize), (u64)-1, 0);
4358 * This function is also used to drop the items in the log tree before
4359 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4360 * it is used to drop the loged items. So we shouldn't kill the delayed
4363 if (min_type == 0 && root == BTRFS_I(inode)->root)
4364 btrfs_kill_delayed_inode_items(inode);
4367 key.offset = (u64)-1;
4372 * with a 16K leaf size and 128MB extents, you can actually queue
4373 * up a huge file in a single leaf. Most of the time that
4374 * bytes_deleted is > 0, it will be huge by the time we get here
4376 if (be_nice && bytes_deleted > SZ_32M) {
4377 if (btrfs_should_end_transaction(trans, root)) {
4384 path->leave_spinning = 1;
4385 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4392 /* there are no items in the tree for us to truncate, we're
4395 if (path->slots[0] == 0)
4402 leaf = path->nodes[0];
4403 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4404 found_type = found_key.type;
4406 if (found_key.objectid != ino)
4409 if (found_type < min_type)
4412 item_end = found_key.offset;
4413 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4414 fi = btrfs_item_ptr(leaf, path->slots[0],
4415 struct btrfs_file_extent_item);
4416 extent_type = btrfs_file_extent_type(leaf, fi);
4417 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4419 btrfs_file_extent_num_bytes(leaf, fi);
4420 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4421 item_end += btrfs_file_extent_inline_len(leaf,
4422 path->slots[0], fi);
4426 if (found_type > min_type) {
4429 if (item_end < new_size)
4431 if (found_key.offset >= new_size)
4437 /* FIXME, shrink the extent if the ref count is only 1 */
4438 if (found_type != BTRFS_EXTENT_DATA_KEY)
4442 last_size = found_key.offset;
4444 last_size = new_size;
4446 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4448 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4450 u64 orig_num_bytes =
4451 btrfs_file_extent_num_bytes(leaf, fi);
4452 extent_num_bytes = ALIGN(new_size -
4455 btrfs_set_file_extent_num_bytes(leaf, fi,
4457 num_dec = (orig_num_bytes -
4459 if (test_bit(BTRFS_ROOT_REF_COWS,
4462 inode_sub_bytes(inode, num_dec);
4463 btrfs_mark_buffer_dirty(leaf);
4466 btrfs_file_extent_disk_num_bytes(leaf,
4468 extent_offset = found_key.offset -
4469 btrfs_file_extent_offset(leaf, fi);
4471 /* FIXME blocksize != 4096 */
4472 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4473 if (extent_start != 0) {
4475 if (test_bit(BTRFS_ROOT_REF_COWS,
4477 inode_sub_bytes(inode, num_dec);
4480 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4482 * we can't truncate inline items that have had
4486 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4487 btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4490 * Need to release path in order to truncate a
4491 * compressed extent. So delete any accumulated
4492 * extent items so far.
4494 if (btrfs_file_extent_compression(leaf, fi) !=
4495 BTRFS_COMPRESS_NONE && pending_del_nr) {
4496 err = btrfs_del_items(trans, root, path,
4500 btrfs_abort_transaction(trans,
4508 err = truncate_inline_extent(inode, path,
4513 btrfs_abort_transaction(trans,
4517 } else if (test_bit(BTRFS_ROOT_REF_COWS,
4519 inode_sub_bytes(inode, item_end + 1 - new_size);
4524 if (!pending_del_nr) {
4525 /* no pending yet, add ourselves */
4526 pending_del_slot = path->slots[0];
4528 } else if (pending_del_nr &&
4529 path->slots[0] + 1 == pending_del_slot) {
4530 /* hop on the pending chunk */
4532 pending_del_slot = path->slots[0];
4539 should_throttle = 0;
4542 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4543 root == root->fs_info->tree_root)) {
4544 btrfs_set_path_blocking(path);
4545 bytes_deleted += extent_num_bytes;
4546 ret = btrfs_free_extent(trans, root, extent_start,
4547 extent_num_bytes, 0,
4548 btrfs_header_owner(leaf),
4549 ino, extent_offset);
4551 if (btrfs_should_throttle_delayed_refs(trans, root))
4552 btrfs_async_run_delayed_refs(root,
4553 trans->delayed_ref_updates * 2, 0);
4555 if (truncate_space_check(trans, root,
4556 extent_num_bytes)) {
4559 if (btrfs_should_throttle_delayed_refs(trans,
4561 should_throttle = 1;
4566 if (found_type == BTRFS_INODE_ITEM_KEY)
4569 if (path->slots[0] == 0 ||
4570 path->slots[0] != pending_del_slot ||
4571 should_throttle || should_end) {
4572 if (pending_del_nr) {
4573 ret = btrfs_del_items(trans, root, path,
4577 btrfs_abort_transaction(trans,
4583 btrfs_release_path(path);
4584 if (should_throttle) {
4585 unsigned long updates = trans->delayed_ref_updates;
4587 trans->delayed_ref_updates = 0;
4588 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4594 * if we failed to refill our space rsv, bail out
4595 * and let the transaction restart
4607 if (pending_del_nr) {
4608 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4611 btrfs_abort_transaction(trans, root, ret);
4614 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4615 btrfs_ordered_update_i_size(inode, last_size, NULL);
4617 btrfs_free_path(path);
4619 if (be_nice && bytes_deleted > SZ_32M) {
4620 unsigned long updates = trans->delayed_ref_updates;
4622 trans->delayed_ref_updates = 0;
4623 ret = btrfs_run_delayed_refs(trans, root, updates * 2);
4632 * btrfs_truncate_block - read, zero a chunk and write a block
4633 * @inode - inode that we're zeroing
4634 * @from - the offset to start zeroing
4635 * @len - the length to zero, 0 to zero the entire range respective to the
4637 * @front - zero up to the offset instead of from the offset on
4639 * This will find the block for the "from" offset and cow the block and zero the
4640 * part we want to zero. This is used with truncate and hole punching.
4642 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4645 struct address_space *mapping = inode->i_mapping;
4646 struct btrfs_root *root = BTRFS_I(inode)->root;
4647 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4648 struct btrfs_ordered_extent *ordered;
4649 struct extent_state *cached_state = NULL;
4651 u32 blocksize = root->sectorsize;
4652 pgoff_t index = from >> PAGE_SHIFT;
4653 unsigned offset = from & (blocksize - 1);
4655 gfp_t mask = btrfs_alloc_write_mask(mapping);
4660 if ((offset & (blocksize - 1)) == 0 &&
4661 (!len || ((len & (blocksize - 1)) == 0)))
4664 ret = btrfs_delalloc_reserve_space(inode,
4665 round_down(from, blocksize), blocksize);
4670 page = find_or_create_page(mapping, index, mask);
4672 btrfs_delalloc_release_space(inode,
4673 round_down(from, blocksize),
4679 block_start = round_down(from, blocksize);
4680 block_end = block_start + blocksize - 1;
4682 if (!PageUptodate(page)) {
4683 ret = btrfs_readpage(NULL, page);
4685 if (page->mapping != mapping) {
4690 if (!PageUptodate(page)) {
4695 wait_on_page_writeback(page);
4697 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4698 set_page_extent_mapped(page);
4700 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4702 unlock_extent_cached(io_tree, block_start, block_end,
4703 &cached_state, GFP_NOFS);
4706 btrfs_start_ordered_extent(inode, ordered, 1);
4707 btrfs_put_ordered_extent(ordered);
4711 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4712 EXTENT_DIRTY | EXTENT_DELALLOC |
4713 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4714 0, 0, &cached_state, GFP_NOFS);
4716 ret = btrfs_set_extent_delalloc(inode, block_start, block_end,
4719 unlock_extent_cached(io_tree, block_start, block_end,
4720 &cached_state, GFP_NOFS);
4724 if (offset != blocksize) {
4726 len = blocksize - offset;
4729 memset(kaddr + (block_start - page_offset(page)),
4732 memset(kaddr + (block_start - page_offset(page)) + offset,
4734 flush_dcache_page(page);
4737 ClearPageChecked(page);
4738 set_page_dirty(page);
4739 unlock_extent_cached(io_tree, block_start, block_end, &cached_state,
4744 btrfs_delalloc_release_space(inode, block_start,
4752 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4753 u64 offset, u64 len)
4755 struct btrfs_trans_handle *trans;
4759 * Still need to make sure the inode looks like it's been updated so
4760 * that any holes get logged if we fsync.
4762 if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4763 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4764 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4765 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4770 * 1 - for the one we're dropping
4771 * 1 - for the one we're adding
4772 * 1 - for updating the inode.
4774 trans = btrfs_start_transaction(root, 3);
4776 return PTR_ERR(trans);
4778 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4780 btrfs_abort_transaction(trans, root, ret);
4781 btrfs_end_transaction(trans, root);
4785 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4786 0, 0, len, 0, len, 0, 0, 0);
4788 btrfs_abort_transaction(trans, root, ret);
4790 btrfs_update_inode(trans, root, inode);
4791 btrfs_end_transaction(trans, root);
4796 * This function puts in dummy file extents for the area we're creating a hole
4797 * for. So if we are truncating this file to a larger size we need to insert
4798 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4799 * the range between oldsize and size
4801 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4803 struct btrfs_root *root = BTRFS_I(inode)->root;
4804 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4805 struct extent_map *em = NULL;
4806 struct extent_state *cached_state = NULL;
4807 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4808 u64 hole_start = ALIGN(oldsize, root->sectorsize);
4809 u64 block_end = ALIGN(size, root->sectorsize);
4816 * If our size started in the middle of a block we need to zero out the
4817 * rest of the block before we expand the i_size, otherwise we could
4818 * expose stale data.
4820 err = btrfs_truncate_block(inode, oldsize, 0, 0);
4824 if (size <= hole_start)
4828 struct btrfs_ordered_extent *ordered;
4830 lock_extent_bits(io_tree, hole_start, block_end - 1,
4832 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4833 block_end - hole_start);
4836 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4837 &cached_state, GFP_NOFS);
4838 btrfs_start_ordered_extent(inode, ordered, 1);
4839 btrfs_put_ordered_extent(ordered);
4842 cur_offset = hole_start;
4844 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4845 block_end - cur_offset, 0);
4851 last_byte = min(extent_map_end(em), block_end);
4852 last_byte = ALIGN(last_byte , root->sectorsize);
4853 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4854 struct extent_map *hole_em;
4855 hole_size = last_byte - cur_offset;
4857 err = maybe_insert_hole(root, inode, cur_offset,
4861 btrfs_drop_extent_cache(inode, cur_offset,
4862 cur_offset + hole_size - 1, 0);
4863 hole_em = alloc_extent_map();
4865 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4866 &BTRFS_I(inode)->runtime_flags);
4869 hole_em->start = cur_offset;
4870 hole_em->len = hole_size;
4871 hole_em->orig_start = cur_offset;
4873 hole_em->block_start = EXTENT_MAP_HOLE;
4874 hole_em->block_len = 0;
4875 hole_em->orig_block_len = 0;
4876 hole_em->ram_bytes = hole_size;
4877 hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4878 hole_em->compress_type = BTRFS_COMPRESS_NONE;
4879 hole_em->generation = root->fs_info->generation;
4882 write_lock(&em_tree->lock);
4883 err = add_extent_mapping(em_tree, hole_em, 1);
4884 write_unlock(&em_tree->lock);
4887 btrfs_drop_extent_cache(inode, cur_offset,
4891 free_extent_map(hole_em);
4894 free_extent_map(em);
4896 cur_offset = last_byte;
4897 if (cur_offset >= block_end)
4900 free_extent_map(em);
4901 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4906 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4908 struct btrfs_root *root = BTRFS_I(inode)->root;
4909 struct btrfs_trans_handle *trans;
4910 loff_t oldsize = i_size_read(inode);
4911 loff_t newsize = attr->ia_size;
4912 int mask = attr->ia_valid;
4916 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4917 * special case where we need to update the times despite not having
4918 * these flags set. For all other operations the VFS set these flags
4919 * explicitly if it wants a timestamp update.
4921 if (newsize != oldsize) {
4922 inode_inc_iversion(inode);
4923 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4924 inode->i_ctime = inode->i_mtime =
4925 current_fs_time(inode->i_sb);
4928 if (newsize > oldsize) {
4930 * Don't do an expanding truncate while snapshoting is ongoing.
4931 * This is to ensure the snapshot captures a fully consistent
4932 * state of this file - if the snapshot captures this expanding
4933 * truncation, it must capture all writes that happened before
4936 btrfs_wait_for_snapshot_creation(root);
4937 ret = btrfs_cont_expand(inode, oldsize, newsize);
4939 btrfs_end_write_no_snapshoting(root);
4943 trans = btrfs_start_transaction(root, 1);
4944 if (IS_ERR(trans)) {
4945 btrfs_end_write_no_snapshoting(root);
4946 return PTR_ERR(trans);
4949 i_size_write(inode, newsize);
4950 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4951 pagecache_isize_extended(inode, oldsize, newsize);
4952 ret = btrfs_update_inode(trans, root, inode);
4953 btrfs_end_write_no_snapshoting(root);
4954 btrfs_end_transaction(trans, root);
4958 * We're truncating a file that used to have good data down to
4959 * zero. Make sure it gets into the ordered flush list so that
4960 * any new writes get down to disk quickly.
4963 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4964 &BTRFS_I(inode)->runtime_flags);
4967 * 1 for the orphan item we're going to add
4968 * 1 for the orphan item deletion.
4970 trans = btrfs_start_transaction(root, 2);
4972 return PTR_ERR(trans);
4975 * We need to do this in case we fail at _any_ point during the
4976 * actual truncate. Once we do the truncate_setsize we could
4977 * invalidate pages which forces any outstanding ordered io to
4978 * be instantly completed which will give us extents that need
4979 * to be truncated. If we fail to get an orphan inode down we
4980 * could have left over extents that were never meant to live,
4981 * so we need to garuntee from this point on that everything
4982 * will be consistent.
4984 ret = btrfs_orphan_add(trans, inode);
4985 btrfs_end_transaction(trans, root);
4989 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4990 truncate_setsize(inode, newsize);
4992 /* Disable nonlocked read DIO to avoid the end less truncate */
4993 btrfs_inode_block_unlocked_dio(inode);
4994 inode_dio_wait(inode);
4995 btrfs_inode_resume_unlocked_dio(inode);
4997 ret = btrfs_truncate(inode);
4998 if (ret && inode->i_nlink) {
5002 * failed to truncate, disk_i_size is only adjusted down
5003 * as we remove extents, so it should represent the true
5004 * size of the inode, so reset the in memory size and
5005 * delete our orphan entry.
5007 trans = btrfs_join_transaction(root);
5008 if (IS_ERR(trans)) {
5009 btrfs_orphan_del(NULL, inode);
5012 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5013 err = btrfs_orphan_del(trans, inode);
5015 btrfs_abort_transaction(trans, root, err);
5016 btrfs_end_transaction(trans, root);
5023 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5025 struct inode *inode = d_inode(dentry);
5026 struct btrfs_root *root = BTRFS_I(inode)->root;
5029 if (btrfs_root_readonly(root))
5032 err = inode_change_ok(inode, attr);
5036 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5037 err = btrfs_setsize(inode, attr);
5042 if (attr->ia_valid) {
5043 setattr_copy(inode, attr);
5044 inode_inc_iversion(inode);
5045 err = btrfs_dirty_inode(inode);
5047 if (!err && attr->ia_valid & ATTR_MODE)
5048 err = posix_acl_chmod(inode, inode->i_mode);
5055 * While truncating the inode pages during eviction, we get the VFS calling
5056 * btrfs_invalidatepage() against each page of the inode. This is slow because
5057 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5058 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5059 * extent_state structures over and over, wasting lots of time.
5061 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5062 * those expensive operations on a per page basis and do only the ordered io
5063 * finishing, while we release here the extent_map and extent_state structures,
5064 * without the excessive merging and splitting.
5066 static void evict_inode_truncate_pages(struct inode *inode)
5068 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5069 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5070 struct rb_node *node;
5072 ASSERT(inode->i_state & I_FREEING);
5073 truncate_inode_pages_final(&inode->i_data);
5075 write_lock(&map_tree->lock);
5076 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5077 struct extent_map *em;
5079 node = rb_first(&map_tree->map);
5080 em = rb_entry(node, struct extent_map, rb_node);
5081 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5082 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5083 remove_extent_mapping(map_tree, em);
5084 free_extent_map(em);
5085 if (need_resched()) {
5086 write_unlock(&map_tree->lock);
5088 write_lock(&map_tree->lock);
5091 write_unlock(&map_tree->lock);
5094 * Keep looping until we have no more ranges in the io tree.
5095 * We can have ongoing bios started by readpages (called from readahead)
5096 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5097 * still in progress (unlocked the pages in the bio but did not yet
5098 * unlocked the ranges in the io tree). Therefore this means some
5099 * ranges can still be locked and eviction started because before
5100 * submitting those bios, which are executed by a separate task (work
5101 * queue kthread), inode references (inode->i_count) were not taken
5102 * (which would be dropped in the end io callback of each bio).
5103 * Therefore here we effectively end up waiting for those bios and
5104 * anyone else holding locked ranges without having bumped the inode's
5105 * reference count - if we don't do it, when they access the inode's
5106 * io_tree to unlock a range it may be too late, leading to an
5107 * use-after-free issue.
5109 spin_lock(&io_tree->lock);
5110 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5111 struct extent_state *state;
5112 struct extent_state *cached_state = NULL;
5116 node = rb_first(&io_tree->state);
5117 state = rb_entry(node, struct extent_state, rb_node);
5118 start = state->start;
5120 spin_unlock(&io_tree->lock);
5122 lock_extent_bits(io_tree, start, end, &cached_state);
5125 * If still has DELALLOC flag, the extent didn't reach disk,
5126 * and its reserved space won't be freed by delayed_ref.
5127 * So we need to free its reserved space here.
5128 * (Refer to comment in btrfs_invalidatepage, case 2)
5130 * Note, end is the bytenr of last byte, so we need + 1 here.
5132 if (state->state & EXTENT_DELALLOC)
5133 btrfs_qgroup_free_data(inode, start, end - start + 1);
5135 clear_extent_bit(io_tree, start, end,
5136 EXTENT_LOCKED | EXTENT_DIRTY |
5137 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5138 EXTENT_DEFRAG, 1, 1,
5139 &cached_state, GFP_NOFS);
5142 spin_lock(&io_tree->lock);
5144 spin_unlock(&io_tree->lock);
5147 void btrfs_evict_inode(struct inode *inode)
5149 struct btrfs_trans_handle *trans;
5150 struct btrfs_root *root = BTRFS_I(inode)->root;
5151 struct btrfs_block_rsv *rsv, *global_rsv;
5152 int steal_from_global = 0;
5153 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
5156 trace_btrfs_inode_evict(inode);
5158 evict_inode_truncate_pages(inode);
5160 if (inode->i_nlink &&
5161 ((btrfs_root_refs(&root->root_item) != 0 &&
5162 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5163 btrfs_is_free_space_inode(inode)))
5166 if (is_bad_inode(inode)) {
5167 btrfs_orphan_del(NULL, inode);
5170 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5171 if (!special_file(inode->i_mode))
5172 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5174 btrfs_free_io_failure_record(inode, 0, (u64)-1);
5176 if (root->fs_info->log_root_recovering) {
5177 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
5178 &BTRFS_I(inode)->runtime_flags));
5182 if (inode->i_nlink > 0) {
5183 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5184 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5188 ret = btrfs_commit_inode_delayed_inode(inode);
5190 btrfs_orphan_del(NULL, inode);
5194 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
5196 btrfs_orphan_del(NULL, inode);
5199 rsv->size = min_size;
5201 global_rsv = &root->fs_info->global_block_rsv;
5203 btrfs_i_size_write(inode, 0);
5206 * This is a bit simpler than btrfs_truncate since we've already
5207 * reserved our space for our orphan item in the unlink, so we just
5208 * need to reserve some slack space in case we add bytes and update
5209 * inode item when doing the truncate.
5212 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5213 BTRFS_RESERVE_FLUSH_LIMIT);
5216 * Try and steal from the global reserve since we will
5217 * likely not use this space anyway, we want to try as
5218 * hard as possible to get this to work.
5221 steal_from_global++;
5223 steal_from_global = 0;
5227 * steal_from_global == 0: we reserved stuff, hooray!
5228 * steal_from_global == 1: we didn't reserve stuff, boo!
5229 * steal_from_global == 2: we've committed, still not a lot of
5230 * room but maybe we'll have room in the global reserve this
5232 * steal_from_global == 3: abandon all hope!
5234 if (steal_from_global > 2) {
5235 btrfs_warn(root->fs_info,
5236 "Could not get space for a delete, will truncate on mount %d",
5238 btrfs_orphan_del(NULL, inode);
5239 btrfs_free_block_rsv(root, rsv);
5243 trans = btrfs_join_transaction(root);
5244 if (IS_ERR(trans)) {
5245 btrfs_orphan_del(NULL, inode);
5246 btrfs_free_block_rsv(root, rsv);
5251 * We can't just steal from the global reserve, we need tomake
5252 * sure there is room to do it, if not we need to commit and try
5255 if (steal_from_global) {
5256 if (!btrfs_check_space_for_delayed_refs(trans, root))
5257 ret = btrfs_block_rsv_migrate(global_rsv, rsv,
5264 * Couldn't steal from the global reserve, we have too much
5265 * pending stuff built up, commit the transaction and try it
5269 ret = btrfs_commit_transaction(trans, root);
5271 btrfs_orphan_del(NULL, inode);
5272 btrfs_free_block_rsv(root, rsv);
5277 steal_from_global = 0;
5280 trans->block_rsv = rsv;
5282 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5283 if (ret != -ENOSPC && ret != -EAGAIN)
5286 trans->block_rsv = &root->fs_info->trans_block_rsv;
5287 btrfs_end_transaction(trans, root);
5289 btrfs_btree_balance_dirty(root);
5292 btrfs_free_block_rsv(root, rsv);
5295 * Errors here aren't a big deal, it just means we leave orphan items
5296 * in the tree. They will be cleaned up on the next mount.
5299 trans->block_rsv = root->orphan_block_rsv;
5300 btrfs_orphan_del(trans, inode);
5302 btrfs_orphan_del(NULL, inode);
5305 trans->block_rsv = &root->fs_info->trans_block_rsv;
5306 if (!(root == root->fs_info->tree_root ||
5307 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5308 btrfs_return_ino(root, btrfs_ino(inode));
5310 btrfs_end_transaction(trans, root);
5311 btrfs_btree_balance_dirty(root);
5313 btrfs_remove_delayed_node(inode);
5318 * this returns the key found in the dir entry in the location pointer.
5319 * If no dir entries were found, location->objectid is 0.
5321 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5322 struct btrfs_key *location)
5324 const char *name = dentry->d_name.name;
5325 int namelen = dentry->d_name.len;
5326 struct btrfs_dir_item *di;
5327 struct btrfs_path *path;
5328 struct btrfs_root *root = BTRFS_I(dir)->root;
5331 path = btrfs_alloc_path();
5335 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5340 if (IS_ERR_OR_NULL(di))
5343 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5345 btrfs_free_path(path);
5348 location->objectid = 0;
5353 * when we hit a tree root in a directory, the btrfs part of the inode
5354 * needs to be changed to reflect the root directory of the tree root. This
5355 * is kind of like crossing a mount point.
5357 static int fixup_tree_root_location(struct btrfs_root *root,
5359 struct dentry *dentry,
5360 struct btrfs_key *location,
5361 struct btrfs_root **sub_root)
5363 struct btrfs_path *path;
5364 struct btrfs_root *new_root;
5365 struct btrfs_root_ref *ref;
5366 struct extent_buffer *leaf;
5367 struct btrfs_key key;
5371 path = btrfs_alloc_path();
5378 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5379 key.type = BTRFS_ROOT_REF_KEY;
5380 key.offset = location->objectid;
5382 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5390 leaf = path->nodes[0];
5391 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5392 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5393 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5396 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5397 (unsigned long)(ref + 1),
5398 dentry->d_name.len);
5402 btrfs_release_path(path);
5404 new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5405 if (IS_ERR(new_root)) {
5406 err = PTR_ERR(new_root);
5410 *sub_root = new_root;
5411 location->objectid = btrfs_root_dirid(&new_root->root_item);
5412 location->type = BTRFS_INODE_ITEM_KEY;
5413 location->offset = 0;
5416 btrfs_free_path(path);
5420 static void inode_tree_add(struct inode *inode)
5422 struct btrfs_root *root = BTRFS_I(inode)->root;
5423 struct btrfs_inode *entry;
5425 struct rb_node *parent;
5426 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5427 u64 ino = btrfs_ino(inode);
5429 if (inode_unhashed(inode))
5432 spin_lock(&root->inode_lock);
5433 p = &root->inode_tree.rb_node;
5436 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5438 if (ino < btrfs_ino(&entry->vfs_inode))
5439 p = &parent->rb_left;
5440 else if (ino > btrfs_ino(&entry->vfs_inode))
5441 p = &parent->rb_right;
5443 WARN_ON(!(entry->vfs_inode.i_state &
5444 (I_WILL_FREE | I_FREEING)));
5445 rb_replace_node(parent, new, &root->inode_tree);
5446 RB_CLEAR_NODE(parent);
5447 spin_unlock(&root->inode_lock);
5451 rb_link_node(new, parent, p);
5452 rb_insert_color(new, &root->inode_tree);
5453 spin_unlock(&root->inode_lock);
5456 static void inode_tree_del(struct inode *inode)
5458 struct btrfs_root *root = BTRFS_I(inode)->root;
5461 spin_lock(&root->inode_lock);
5462 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5463 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5464 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5465 empty = RB_EMPTY_ROOT(&root->inode_tree);
5467 spin_unlock(&root->inode_lock);
5469 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5470 synchronize_srcu(&root->fs_info->subvol_srcu);
5471 spin_lock(&root->inode_lock);
5472 empty = RB_EMPTY_ROOT(&root->inode_tree);
5473 spin_unlock(&root->inode_lock);
5475 btrfs_add_dead_root(root);
5479 void btrfs_invalidate_inodes(struct btrfs_root *root)
5481 struct rb_node *node;
5482 struct rb_node *prev;
5483 struct btrfs_inode *entry;
5484 struct inode *inode;
5487 if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5488 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5490 spin_lock(&root->inode_lock);
5492 node = root->inode_tree.rb_node;
5496 entry = rb_entry(node, struct btrfs_inode, rb_node);
5498 if (objectid < btrfs_ino(&entry->vfs_inode))
5499 node = node->rb_left;
5500 else if (objectid > btrfs_ino(&entry->vfs_inode))
5501 node = node->rb_right;
5507 entry = rb_entry(prev, struct btrfs_inode, rb_node);
5508 if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5512 prev = rb_next(prev);
5516 entry = rb_entry(node, struct btrfs_inode, rb_node);
5517 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5518 inode = igrab(&entry->vfs_inode);
5520 spin_unlock(&root->inode_lock);
5521 if (atomic_read(&inode->i_count) > 1)
5522 d_prune_aliases(inode);
5524 * btrfs_drop_inode will have it removed from
5525 * the inode cache when its usage count
5530 spin_lock(&root->inode_lock);
5534 if (cond_resched_lock(&root->inode_lock))
5537 node = rb_next(node);
5539 spin_unlock(&root->inode_lock);
5542 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5544 struct btrfs_iget_args *args = p;
5545 inode->i_ino = args->location->objectid;
5546 memcpy(&BTRFS_I(inode)->location, args->location,
5547 sizeof(*args->location));
5548 BTRFS_I(inode)->root = args->root;
5552 static int btrfs_find_actor(struct inode *inode, void *opaque)
5554 struct btrfs_iget_args *args = opaque;
5555 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5556 args->root == BTRFS_I(inode)->root;
5559 static struct inode *btrfs_iget_locked(struct super_block *s,
5560 struct btrfs_key *location,
5561 struct btrfs_root *root)
5563 struct inode *inode;
5564 struct btrfs_iget_args args;
5565 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5567 args.location = location;
5570 inode = iget5_locked(s, hashval, btrfs_find_actor,
5571 btrfs_init_locked_inode,
5576 /* Get an inode object given its location and corresponding root.
5577 * Returns in *is_new if the inode was read from disk
5579 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5580 struct btrfs_root *root, int *new)
5582 struct inode *inode;
5584 inode = btrfs_iget_locked(s, location, root);
5586 return ERR_PTR(-ENOMEM);
5588 if (inode->i_state & I_NEW) {
5589 btrfs_read_locked_inode(inode);
5590 if (!is_bad_inode(inode)) {
5591 inode_tree_add(inode);
5592 unlock_new_inode(inode);
5596 unlock_new_inode(inode);
5598 inode = ERR_PTR(-ESTALE);
5605 static struct inode *new_simple_dir(struct super_block *s,
5606 struct btrfs_key *key,
5607 struct btrfs_root *root)
5609 struct inode *inode = new_inode(s);
5612 return ERR_PTR(-ENOMEM);
5614 BTRFS_I(inode)->root = root;
5615 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5616 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5618 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5619 inode->i_op = &btrfs_dir_ro_inode_operations;
5620 inode->i_fop = &simple_dir_operations;
5621 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5622 inode->i_mtime = current_fs_time(inode->i_sb);
5623 inode->i_atime = inode->i_mtime;
5624 inode->i_ctime = inode->i_mtime;
5625 BTRFS_I(inode)->i_otime = inode->i_mtime;
5630 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5632 struct inode *inode;
5633 struct btrfs_root *root = BTRFS_I(dir)->root;
5634 struct btrfs_root *sub_root = root;
5635 struct btrfs_key location;
5639 if (dentry->d_name.len > BTRFS_NAME_LEN)
5640 return ERR_PTR(-ENAMETOOLONG);
5642 ret = btrfs_inode_by_name(dir, dentry, &location);
5644 return ERR_PTR(ret);
5646 if (location.objectid == 0)
5647 return ERR_PTR(-ENOENT);
5649 if (location.type == BTRFS_INODE_ITEM_KEY) {
5650 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5654 BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5656 index = srcu_read_lock(&root->fs_info->subvol_srcu);
5657 ret = fixup_tree_root_location(root, dir, dentry,
5658 &location, &sub_root);
5661 inode = ERR_PTR(ret);
5663 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5665 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5667 srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5669 if (!IS_ERR(inode) && root != sub_root) {
5670 down_read(&root->fs_info->cleanup_work_sem);
5671 if (!(inode->i_sb->s_flags & MS_RDONLY))
5672 ret = btrfs_orphan_cleanup(sub_root);
5673 up_read(&root->fs_info->cleanup_work_sem);
5676 inode = ERR_PTR(ret);
5683 static int btrfs_dentry_delete(const struct dentry *dentry)
5685 struct btrfs_root *root;
5686 struct inode *inode = d_inode(dentry);
5688 if (!inode && !IS_ROOT(dentry))
5689 inode = d_inode(dentry->d_parent);
5692 root = BTRFS_I(inode)->root;
5693 if (btrfs_root_refs(&root->root_item) == 0)
5696 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5702 static void btrfs_dentry_release(struct dentry *dentry)
5704 kfree(dentry->d_fsdata);
5707 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5710 struct inode *inode;
5712 inode = btrfs_lookup_dentry(dir, dentry);
5713 if (IS_ERR(inode)) {
5714 if (PTR_ERR(inode) == -ENOENT)
5717 return ERR_CAST(inode);
5720 return d_splice_alias(inode, dentry);
5723 unsigned char btrfs_filetype_table[] = {
5724 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5727 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5729 struct inode *inode = file_inode(file);
5730 struct btrfs_root *root = BTRFS_I(inode)->root;
5731 struct btrfs_item *item;
5732 struct btrfs_dir_item *di;
5733 struct btrfs_key key;
5734 struct btrfs_key found_key;
5735 struct btrfs_path *path;
5736 struct list_head ins_list;
5737 struct list_head del_list;
5739 struct extent_buffer *leaf;
5741 unsigned char d_type;
5746 int key_type = BTRFS_DIR_INDEX_KEY;
5750 int is_curr = 0; /* ctx->pos points to the current index? */
5753 /* FIXME, use a real flag for deciding about the key type */
5754 if (root->fs_info->tree_root == root)
5755 key_type = BTRFS_DIR_ITEM_KEY;
5757 if (!dir_emit_dots(file, ctx))
5760 path = btrfs_alloc_path();
5764 path->reada = READA_FORWARD;
5766 if (key_type == BTRFS_DIR_INDEX_KEY) {
5767 INIT_LIST_HEAD(&ins_list);
5768 INIT_LIST_HEAD(&del_list);
5769 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5772 key.type = key_type;
5773 key.offset = ctx->pos;
5774 key.objectid = btrfs_ino(inode);
5776 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5782 leaf = path->nodes[0];
5783 slot = path->slots[0];
5784 if (slot >= btrfs_header_nritems(leaf)) {
5785 ret = btrfs_next_leaf(root, path);
5793 item = btrfs_item_nr(slot);
5794 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5796 if (found_key.objectid != key.objectid)
5798 if (found_key.type != key_type)
5800 if (found_key.offset < ctx->pos)
5802 if (key_type == BTRFS_DIR_INDEX_KEY &&
5803 btrfs_should_delete_dir_index(&del_list,
5807 ctx->pos = found_key.offset;
5810 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5812 di_total = btrfs_item_size(leaf, item);
5814 while (di_cur < di_total) {
5815 struct btrfs_key location;
5817 if (verify_dir_item(root, leaf, di))
5820 name_len = btrfs_dir_name_len(leaf, di);
5821 if (name_len <= sizeof(tmp_name)) {
5822 name_ptr = tmp_name;
5824 name_ptr = kmalloc(name_len, GFP_KERNEL);
5830 read_extent_buffer(leaf, name_ptr,
5831 (unsigned long)(di + 1), name_len);
5833 d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5834 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5837 /* is this a reference to our own snapshot? If so
5840 * In contrast to old kernels, we insert the snapshot's
5841 * dir item and dir index after it has been created, so
5842 * we won't find a reference to our own snapshot. We
5843 * still keep the following code for backward
5846 if (location.type == BTRFS_ROOT_ITEM_KEY &&
5847 location.objectid == root->root_key.objectid) {
5851 over = !dir_emit(ctx, name_ptr, name_len,
5852 location.objectid, d_type);
5855 if (name_ptr != tmp_name)
5861 di_len = btrfs_dir_name_len(leaf, di) +
5862 btrfs_dir_data_len(leaf, di) + sizeof(*di);
5864 di = (struct btrfs_dir_item *)((char *)di + di_len);
5870 if (key_type == BTRFS_DIR_INDEX_KEY) {
5873 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list, &emitted);
5879 * If we haven't emitted any dir entry, we must not touch ctx->pos as
5880 * it was was set to the termination value in previous call. We assume
5881 * that "." and ".." were emitted if we reach this point and set the
5882 * termination value as well for an empty directory.
5884 if (ctx->pos > 2 && !emitted)
5887 /* Reached end of directory/root. Bump pos past the last item. */
5891 * Stop new entries from being returned after we return the last
5894 * New directory entries are assigned a strictly increasing
5895 * offset. This means that new entries created during readdir
5896 * are *guaranteed* to be seen in the future by that readdir.
5897 * This has broken buggy programs which operate on names as
5898 * they're returned by readdir. Until we re-use freed offsets
5899 * we have this hack to stop new entries from being returned
5900 * under the assumption that they'll never reach this huge
5903 * This is being careful not to overflow 32bit loff_t unless the
5904 * last entry requires it because doing so has broken 32bit apps
5907 if (key_type == BTRFS_DIR_INDEX_KEY) {
5908 if (ctx->pos >= INT_MAX)
5909 ctx->pos = LLONG_MAX;
5916 if (key_type == BTRFS_DIR_INDEX_KEY)
5917 btrfs_put_delayed_items(&ins_list, &del_list);
5918 btrfs_free_path(path);
5922 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5924 struct btrfs_root *root = BTRFS_I(inode)->root;
5925 struct btrfs_trans_handle *trans;
5927 bool nolock = false;
5929 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5932 if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5935 if (wbc->sync_mode == WB_SYNC_ALL) {
5937 trans = btrfs_join_transaction_nolock(root);
5939 trans = btrfs_join_transaction(root);
5941 return PTR_ERR(trans);
5942 ret = btrfs_commit_transaction(trans, root);
5948 * This is somewhat expensive, updating the tree every time the
5949 * inode changes. But, it is most likely to find the inode in cache.
5950 * FIXME, needs more benchmarking...there are no reasons other than performance
5951 * to keep or drop this code.
5953 static int btrfs_dirty_inode(struct inode *inode)
5955 struct btrfs_root *root = BTRFS_I(inode)->root;
5956 struct btrfs_trans_handle *trans;
5959 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5962 trans = btrfs_join_transaction(root);
5964 return PTR_ERR(trans);
5966 ret = btrfs_update_inode(trans, root, inode);
5967 if (ret && ret == -ENOSPC) {
5968 /* whoops, lets try again with the full transaction */
5969 btrfs_end_transaction(trans, root);
5970 trans = btrfs_start_transaction(root, 1);
5972 return PTR_ERR(trans);
5974 ret = btrfs_update_inode(trans, root, inode);
5976 btrfs_end_transaction(trans, root);
5977 if (BTRFS_I(inode)->delayed_node)
5978 btrfs_balance_delayed_items(root);
5984 * This is a copy of file_update_time. We need this so we can return error on
5985 * ENOSPC for updating the inode in the case of file write and mmap writes.
5987 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5990 struct btrfs_root *root = BTRFS_I(inode)->root;
5992 if (btrfs_root_readonly(root))
5995 if (flags & S_VERSION)
5996 inode_inc_iversion(inode);
5997 if (flags & S_CTIME)
5998 inode->i_ctime = *now;
5999 if (flags & S_MTIME)
6000 inode->i_mtime = *now;
6001 if (flags & S_ATIME)
6002 inode->i_atime = *now;
6003 return btrfs_dirty_inode(inode);
6007 * find the highest existing sequence number in a directory
6008 * and then set the in-memory index_cnt variable to reflect
6009 * free sequence numbers
6011 static int btrfs_set_inode_index_count(struct inode *inode)
6013 struct btrfs_root *root = BTRFS_I(inode)->root;
6014 struct btrfs_key key, found_key;
6015 struct btrfs_path *path;
6016 struct extent_buffer *leaf;
6019 key.objectid = btrfs_ino(inode);
6020 key.type = BTRFS_DIR_INDEX_KEY;
6021 key.offset = (u64)-1;
6023 path = btrfs_alloc_path();
6027 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6030 /* FIXME: we should be able to handle this */
6036 * MAGIC NUMBER EXPLANATION:
6037 * since we search a directory based on f_pos we have to start at 2
6038 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6039 * else has to start at 2
6041 if (path->slots[0] == 0) {
6042 BTRFS_I(inode)->index_cnt = 2;
6048 leaf = path->nodes[0];
6049 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6051 if (found_key.objectid != btrfs_ino(inode) ||
6052 found_key.type != BTRFS_DIR_INDEX_KEY) {
6053 BTRFS_I(inode)->index_cnt = 2;
6057 BTRFS_I(inode)->index_cnt = found_key.offset + 1;
6059 btrfs_free_path(path);
6064 * helper to find a free sequence number in a given directory. This current
6065 * code is very simple, later versions will do smarter things in the btree
6067 int btrfs_set_inode_index(struct inode *dir, u64 *index)
6071 if (BTRFS_I(dir)->index_cnt == (u64)-1) {
6072 ret = btrfs_inode_delayed_dir_index_count(dir);
6074 ret = btrfs_set_inode_index_count(dir);
6080 *index = BTRFS_I(dir)->index_cnt;
6081 BTRFS_I(dir)->index_cnt++;
6086 static int btrfs_insert_inode_locked(struct inode *inode)
6088 struct btrfs_iget_args args;
6089 args.location = &BTRFS_I(inode)->location;
6090 args.root = BTRFS_I(inode)->root;
6092 return insert_inode_locked4(inode,
6093 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6094 btrfs_find_actor, &args);
6097 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6098 struct btrfs_root *root,
6100 const char *name, int name_len,
6101 u64 ref_objectid, u64 objectid,
6102 umode_t mode, u64 *index)
6104 struct inode *inode;
6105 struct btrfs_inode_item *inode_item;
6106 struct btrfs_key *location;
6107 struct btrfs_path *path;
6108 struct btrfs_inode_ref *ref;
6109 struct btrfs_key key[2];
6111 int nitems = name ? 2 : 1;
6115 path = btrfs_alloc_path();
6117 return ERR_PTR(-ENOMEM);
6119 inode = new_inode(root->fs_info->sb);
6121 btrfs_free_path(path);
6122 return ERR_PTR(-ENOMEM);
6126 * O_TMPFILE, set link count to 0, so that after this point,
6127 * we fill in an inode item with the correct link count.
6130 set_nlink(inode, 0);
6133 * we have to initialize this early, so we can reclaim the inode
6134 * number if we fail afterwards in this function.
6136 inode->i_ino = objectid;
6139 trace_btrfs_inode_request(dir);
6141 ret = btrfs_set_inode_index(dir, index);
6143 btrfs_free_path(path);
6145 return ERR_PTR(ret);
6151 * index_cnt is ignored for everything but a dir,
6152 * btrfs_get_inode_index_count has an explanation for the magic
6155 BTRFS_I(inode)->index_cnt = 2;
6156 BTRFS_I(inode)->dir_index = *index;
6157 BTRFS_I(inode)->root = root;
6158 BTRFS_I(inode)->generation = trans->transid;
6159 inode->i_generation = BTRFS_I(inode)->generation;
6162 * We could have gotten an inode number from somebody who was fsynced
6163 * and then removed in this same transaction, so let's just set full
6164 * sync since it will be a full sync anyway and this will blow away the
6165 * old info in the log.
6167 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6169 key[0].objectid = objectid;
6170 key[0].type = BTRFS_INODE_ITEM_KEY;
6173 sizes[0] = sizeof(struct btrfs_inode_item);
6177 * Start new inodes with an inode_ref. This is slightly more
6178 * efficient for small numbers of hard links since they will
6179 * be packed into one item. Extended refs will kick in if we
6180 * add more hard links than can fit in the ref item.
6182 key[1].objectid = objectid;
6183 key[1].type = BTRFS_INODE_REF_KEY;
6184 key[1].offset = ref_objectid;
6186 sizes[1] = name_len + sizeof(*ref);
6189 location = &BTRFS_I(inode)->location;
6190 location->objectid = objectid;
6191 location->offset = 0;
6192 location->type = BTRFS_INODE_ITEM_KEY;
6194 ret = btrfs_insert_inode_locked(inode);
6198 path->leave_spinning = 1;
6199 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6203 inode_init_owner(inode, dir, mode);
6204 inode_set_bytes(inode, 0);
6206 inode->i_mtime = current_fs_time(inode->i_sb);
6207 inode->i_atime = inode->i_mtime;
6208 inode->i_ctime = inode->i_mtime;
6209 BTRFS_I(inode)->i_otime = inode->i_mtime;
6211 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6212 struct btrfs_inode_item);
6213 memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
6214 sizeof(*inode_item));
6215 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6218 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6219 struct btrfs_inode_ref);
6220 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6221 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6222 ptr = (unsigned long)(ref + 1);
6223 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6226 btrfs_mark_buffer_dirty(path->nodes[0]);
6227 btrfs_free_path(path);
6229 btrfs_inherit_iflags(inode, dir);
6231 if (S_ISREG(mode)) {
6232 if (btrfs_test_opt(root, NODATASUM))
6233 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6234 if (btrfs_test_opt(root, NODATACOW))
6235 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6236 BTRFS_INODE_NODATASUM;
6239 inode_tree_add(inode);
6241 trace_btrfs_inode_new(inode);
6242 btrfs_set_inode_last_trans(trans, inode);
6244 btrfs_update_root_times(trans, root);
6246 ret = btrfs_inode_inherit_props(trans, inode, dir);
6248 btrfs_err(root->fs_info,
6249 "error inheriting props for ino %llu (root %llu): %d",
6250 btrfs_ino(inode), root->root_key.objectid, ret);
6255 unlock_new_inode(inode);
6258 BTRFS_I(dir)->index_cnt--;
6259 btrfs_free_path(path);
6261 return ERR_PTR(ret);
6264 static inline u8 btrfs_inode_type(struct inode *inode)
6266 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6270 * utility function to add 'inode' into 'parent_inode' with
6271 * a give name and a given sequence number.
6272 * if 'add_backref' is true, also insert a backref from the
6273 * inode to the parent directory.
6275 int btrfs_add_link(struct btrfs_trans_handle *trans,
6276 struct inode *parent_inode, struct inode *inode,
6277 const char *name, int name_len, int add_backref, u64 index)
6280 struct btrfs_key key;
6281 struct btrfs_root *root = BTRFS_I(parent_inode)->root;
6282 u64 ino = btrfs_ino(inode);
6283 u64 parent_ino = btrfs_ino(parent_inode);
6285 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6286 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
6289 key.type = BTRFS_INODE_ITEM_KEY;
6293 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6294 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
6295 key.objectid, root->root_key.objectid,
6296 parent_ino, index, name, name_len);
6297 } else if (add_backref) {
6298 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6302 /* Nothing to clean up yet */
6306 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6308 btrfs_inode_type(inode), index);
6309 if (ret == -EEXIST || ret == -EOVERFLOW)
6312 btrfs_abort_transaction(trans, root, ret);
6316 btrfs_i_size_write(parent_inode, parent_inode->i_size +
6318 inode_inc_iversion(parent_inode);
6319 parent_inode->i_mtime = parent_inode->i_ctime =
6320 current_fs_time(parent_inode->i_sb);
6321 ret = btrfs_update_inode(trans, root, parent_inode);
6323 btrfs_abort_transaction(trans, root, ret);
6327 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6330 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6331 key.objectid, root->root_key.objectid,
6332 parent_ino, &local_index, name, name_len);
6334 } else if (add_backref) {
6338 err = btrfs_del_inode_ref(trans, root, name, name_len,
6339 ino, parent_ino, &local_index);
6344 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6345 struct inode *dir, struct dentry *dentry,
6346 struct inode *inode, int backref, u64 index)
6348 int err = btrfs_add_link(trans, dir, inode,
6349 dentry->d_name.name, dentry->d_name.len,
6356 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6357 umode_t mode, dev_t rdev)
6359 struct btrfs_trans_handle *trans;
6360 struct btrfs_root *root = BTRFS_I(dir)->root;
6361 struct inode *inode = NULL;
6368 * 2 for inode item and ref
6370 * 1 for xattr if selinux is on
6372 trans = btrfs_start_transaction(root, 5);
6374 return PTR_ERR(trans);
6376 err = btrfs_find_free_ino(root, &objectid);
6380 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6381 dentry->d_name.len, btrfs_ino(dir), objectid,
6383 if (IS_ERR(inode)) {
6384 err = PTR_ERR(inode);
6389 * If the active LSM wants to access the inode during
6390 * d_instantiate it needs these. Smack checks to see
6391 * if the filesystem supports xattrs by looking at the
6394 inode->i_op = &btrfs_special_inode_operations;
6395 init_special_inode(inode, inode->i_mode, rdev);
6397 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6399 goto out_unlock_inode;
6401 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6403 goto out_unlock_inode;
6405 btrfs_update_inode(trans, root, inode);
6406 unlock_new_inode(inode);
6407 d_instantiate(dentry, inode);
6411 btrfs_end_transaction(trans, root);
6412 btrfs_balance_delayed_items(root);
6413 btrfs_btree_balance_dirty(root);
6415 inode_dec_link_count(inode);
6422 unlock_new_inode(inode);
6427 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6428 umode_t mode, bool excl)
6430 struct btrfs_trans_handle *trans;
6431 struct btrfs_root *root = BTRFS_I(dir)->root;
6432 struct inode *inode = NULL;
6433 int drop_inode_on_err = 0;
6439 * 2 for inode item and ref
6441 * 1 for xattr if selinux is on
6443 trans = btrfs_start_transaction(root, 5);
6445 return PTR_ERR(trans);
6447 err = btrfs_find_free_ino(root, &objectid);
6451 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6452 dentry->d_name.len, btrfs_ino(dir), objectid,
6454 if (IS_ERR(inode)) {
6455 err = PTR_ERR(inode);
6458 drop_inode_on_err = 1;
6460 * If the active LSM wants to access the inode during
6461 * d_instantiate it needs these. Smack checks to see
6462 * if the filesystem supports xattrs by looking at the
6465 inode->i_fop = &btrfs_file_operations;
6466 inode->i_op = &btrfs_file_inode_operations;
6467 inode->i_mapping->a_ops = &btrfs_aops;
6469 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6471 goto out_unlock_inode;
6473 err = btrfs_update_inode(trans, root, inode);
6475 goto out_unlock_inode;
6477 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6479 goto out_unlock_inode;
6481 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6482 unlock_new_inode(inode);
6483 d_instantiate(dentry, inode);
6486 btrfs_end_transaction(trans, root);
6487 if (err && drop_inode_on_err) {
6488 inode_dec_link_count(inode);
6491 btrfs_balance_delayed_items(root);
6492 btrfs_btree_balance_dirty(root);
6496 unlock_new_inode(inode);
6501 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6502 struct dentry *dentry)
6504 struct btrfs_trans_handle *trans = NULL;
6505 struct btrfs_root *root = BTRFS_I(dir)->root;
6506 struct inode *inode = d_inode(old_dentry);
6511 /* do not allow sys_link's with other subvols of the same device */
6512 if (root->objectid != BTRFS_I(inode)->root->objectid)
6515 if (inode->i_nlink >= BTRFS_LINK_MAX)
6518 err = btrfs_set_inode_index(dir, &index);
6523 * 2 items for inode and inode ref
6524 * 2 items for dir items
6525 * 1 item for parent inode
6527 trans = btrfs_start_transaction(root, 5);
6528 if (IS_ERR(trans)) {
6529 err = PTR_ERR(trans);
6534 /* There are several dir indexes for this inode, clear the cache. */
6535 BTRFS_I(inode)->dir_index = 0ULL;
6537 inode_inc_iversion(inode);
6538 inode->i_ctime = current_fs_time(inode->i_sb);
6540 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6542 err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6547 struct dentry *parent = dentry->d_parent;
6548 err = btrfs_update_inode(trans, root, inode);
6551 if (inode->i_nlink == 1) {
6553 * If new hard link count is 1, it's a file created
6554 * with open(2) O_TMPFILE flag.
6556 err = btrfs_orphan_del(trans, inode);
6560 d_instantiate(dentry, inode);
6561 btrfs_log_new_name(trans, inode, NULL, parent);
6564 btrfs_balance_delayed_items(root);
6567 btrfs_end_transaction(trans, root);
6569 inode_dec_link_count(inode);
6572 btrfs_btree_balance_dirty(root);
6576 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6578 struct inode *inode = NULL;
6579 struct btrfs_trans_handle *trans;
6580 struct btrfs_root *root = BTRFS_I(dir)->root;
6582 int drop_on_err = 0;
6587 * 2 items for inode and ref
6588 * 2 items for dir items
6589 * 1 for xattr if selinux is on
6591 trans = btrfs_start_transaction(root, 5);
6593 return PTR_ERR(trans);
6595 err = btrfs_find_free_ino(root, &objectid);
6599 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6600 dentry->d_name.len, btrfs_ino(dir), objectid,
6601 S_IFDIR | mode, &index);
6602 if (IS_ERR(inode)) {
6603 err = PTR_ERR(inode);
6608 /* these must be set before we unlock the inode */
6609 inode->i_op = &btrfs_dir_inode_operations;
6610 inode->i_fop = &btrfs_dir_file_operations;
6612 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6614 goto out_fail_inode;
6616 btrfs_i_size_write(inode, 0);
6617 err = btrfs_update_inode(trans, root, inode);
6619 goto out_fail_inode;
6621 err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6622 dentry->d_name.len, 0, index);
6624 goto out_fail_inode;
6626 d_instantiate(dentry, inode);
6628 * mkdir is special. We're unlocking after we call d_instantiate
6629 * to avoid a race with nfsd calling d_instantiate.
6631 unlock_new_inode(inode);
6635 btrfs_end_transaction(trans, root);
6637 inode_dec_link_count(inode);
6640 btrfs_balance_delayed_items(root);
6641 btrfs_btree_balance_dirty(root);
6645 unlock_new_inode(inode);
6649 /* Find next extent map of a given extent map, caller needs to ensure locks */
6650 static struct extent_map *next_extent_map(struct extent_map *em)
6652 struct rb_node *next;
6654 next = rb_next(&em->rb_node);
6657 return container_of(next, struct extent_map, rb_node);
6660 static struct extent_map *prev_extent_map(struct extent_map *em)
6662 struct rb_node *prev;
6664 prev = rb_prev(&em->rb_node);
6667 return container_of(prev, struct extent_map, rb_node);
6670 /* helper for btfs_get_extent. Given an existing extent in the tree,
6671 * the existing extent is the nearest extent to map_start,
6672 * and an extent that you want to insert, deal with overlap and insert
6673 * the best fitted new extent into the tree.
6675 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6676 struct extent_map *existing,
6677 struct extent_map *em,
6680 struct extent_map *prev;
6681 struct extent_map *next;
6686 BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6688 if (existing->start > map_start) {
6690 prev = prev_extent_map(next);
6693 next = next_extent_map(prev);
6696 start = prev ? extent_map_end(prev) : em->start;
6697 start = max_t(u64, start, em->start);
6698 end = next ? next->start : extent_map_end(em);
6699 end = min_t(u64, end, extent_map_end(em));
6700 start_diff = start - em->start;
6702 em->len = end - start;
6703 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6704 !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6705 em->block_start += start_diff;
6706 em->block_len -= start_diff;
6708 return add_extent_mapping(em_tree, em, 0);
6711 static noinline int uncompress_inline(struct btrfs_path *path,
6713 size_t pg_offset, u64 extent_offset,
6714 struct btrfs_file_extent_item *item)
6717 struct extent_buffer *leaf = path->nodes[0];
6720 unsigned long inline_size;
6724 WARN_ON(pg_offset != 0);
6725 compress_type = btrfs_file_extent_compression(leaf, item);
6726 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6727 inline_size = btrfs_file_extent_inline_item_len(leaf,
6728 btrfs_item_nr(path->slots[0]));
6729 tmp = kmalloc(inline_size, GFP_NOFS);
6732 ptr = btrfs_file_extent_inline_start(item);
6734 read_extent_buffer(leaf, tmp, ptr, inline_size);
6736 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6737 ret = btrfs_decompress(compress_type, tmp, page,
6738 extent_offset, inline_size, max_size);
6744 * a bit scary, this does extent mapping from logical file offset to the disk.
6745 * the ugly parts come from merging extents from the disk with the in-ram
6746 * representation. This gets more complex because of the data=ordered code,
6747 * where the in-ram extents might be locked pending data=ordered completion.
6749 * This also copies inline extents directly into the page.
6752 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6753 size_t pg_offset, u64 start, u64 len,
6758 u64 extent_start = 0;
6760 u64 objectid = btrfs_ino(inode);
6762 struct btrfs_path *path = NULL;
6763 struct btrfs_root *root = BTRFS_I(inode)->root;
6764 struct btrfs_file_extent_item *item;
6765 struct extent_buffer *leaf;
6766 struct btrfs_key found_key;
6767 struct extent_map *em = NULL;
6768 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6769 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6770 struct btrfs_trans_handle *trans = NULL;
6771 const bool new_inline = !page || create;
6774 read_lock(&em_tree->lock);
6775 em = lookup_extent_mapping(em_tree, start, len);
6777 em->bdev = root->fs_info->fs_devices->latest_bdev;
6778 read_unlock(&em_tree->lock);
6781 if (em->start > start || em->start + em->len <= start)
6782 free_extent_map(em);
6783 else if (em->block_start == EXTENT_MAP_INLINE && page)
6784 free_extent_map(em);
6788 em = alloc_extent_map();
6793 em->bdev = root->fs_info->fs_devices->latest_bdev;
6794 em->start = EXTENT_MAP_HOLE;
6795 em->orig_start = EXTENT_MAP_HOLE;
6797 em->block_len = (u64)-1;
6800 path = btrfs_alloc_path();
6806 * Chances are we'll be called again, so go ahead and do
6809 path->reada = READA_FORWARD;
6812 ret = btrfs_lookup_file_extent(trans, root, path,
6813 objectid, start, trans != NULL);
6820 if (path->slots[0] == 0)
6825 leaf = path->nodes[0];
6826 item = btrfs_item_ptr(leaf, path->slots[0],
6827 struct btrfs_file_extent_item);
6828 /* are we inside the extent that was found? */
6829 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6830 found_type = found_key.type;
6831 if (found_key.objectid != objectid ||
6832 found_type != BTRFS_EXTENT_DATA_KEY) {
6834 * If we backup past the first extent we want to move forward
6835 * and see if there is an extent in front of us, otherwise we'll
6836 * say there is a hole for our whole search range which can
6843 found_type = btrfs_file_extent_type(leaf, item);
6844 extent_start = found_key.offset;
6845 if (found_type == BTRFS_FILE_EXTENT_REG ||
6846 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6847 extent_end = extent_start +
6848 btrfs_file_extent_num_bytes(leaf, item);
6849 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6851 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6852 extent_end = ALIGN(extent_start + size, root->sectorsize);
6855 if (start >= extent_end) {
6857 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6858 ret = btrfs_next_leaf(root, path);
6865 leaf = path->nodes[0];
6867 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6868 if (found_key.objectid != objectid ||
6869 found_key.type != BTRFS_EXTENT_DATA_KEY)
6871 if (start + len <= found_key.offset)
6873 if (start > found_key.offset)
6876 em->orig_start = start;
6877 em->len = found_key.offset - start;
6881 btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6883 if (found_type == BTRFS_FILE_EXTENT_REG ||
6884 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6886 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6890 size_t extent_offset;
6896 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6897 extent_offset = page_offset(page) + pg_offset - extent_start;
6898 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6899 size - extent_offset);
6900 em->start = extent_start + extent_offset;
6901 em->len = ALIGN(copy_size, root->sectorsize);
6902 em->orig_block_len = em->len;
6903 em->orig_start = em->start;
6904 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6905 if (create == 0 && !PageUptodate(page)) {
6906 if (btrfs_file_extent_compression(leaf, item) !=
6907 BTRFS_COMPRESS_NONE) {
6908 ret = uncompress_inline(path, page, pg_offset,
6909 extent_offset, item);
6916 read_extent_buffer(leaf, map + pg_offset, ptr,
6918 if (pg_offset + copy_size < PAGE_SIZE) {
6919 memset(map + pg_offset + copy_size, 0,
6920 PAGE_SIZE - pg_offset -
6925 flush_dcache_page(page);
6926 } else if (create && PageUptodate(page)) {
6930 free_extent_map(em);
6933 btrfs_release_path(path);
6934 trans = btrfs_join_transaction(root);
6937 return ERR_CAST(trans);
6941 write_extent_buffer(leaf, map + pg_offset, ptr,
6944 btrfs_mark_buffer_dirty(leaf);
6946 set_extent_uptodate(io_tree, em->start,
6947 extent_map_end(em) - 1, NULL, GFP_NOFS);
6952 em->orig_start = start;
6955 em->block_start = EXTENT_MAP_HOLE;
6956 set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6958 btrfs_release_path(path);
6959 if (em->start > start || extent_map_end(em) <= start) {
6960 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6961 em->start, em->len, start, len);
6967 write_lock(&em_tree->lock);
6968 ret = add_extent_mapping(em_tree, em, 0);
6969 /* it is possible that someone inserted the extent into the tree
6970 * while we had the lock dropped. It is also possible that
6971 * an overlapping map exists in the tree
6973 if (ret == -EEXIST) {
6974 struct extent_map *existing;
6978 existing = search_extent_mapping(em_tree, start, len);
6980 * existing will always be non-NULL, since there must be
6981 * extent causing the -EEXIST.
6983 if (start >= extent_map_end(existing) ||
6984 start <= existing->start) {
6986 * The existing extent map is the one nearest to
6987 * the [start, start + len) range which overlaps
6989 err = merge_extent_mapping(em_tree, existing,
6991 free_extent_map(existing);
6993 free_extent_map(em);
6997 free_extent_map(em);
7002 write_unlock(&em_tree->lock);
7005 trace_btrfs_get_extent(root, em);
7007 btrfs_free_path(path);
7009 ret = btrfs_end_transaction(trans, root);
7014 free_extent_map(em);
7015 return ERR_PTR(err);
7017 BUG_ON(!em); /* Error is always set */
7021 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
7022 size_t pg_offset, u64 start, u64 len,
7025 struct extent_map *em;
7026 struct extent_map *hole_em = NULL;
7027 u64 range_start = start;
7033 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7040 * - a pre-alloc extent,
7041 * there might actually be delalloc bytes behind it.
7043 if (em->block_start != EXTENT_MAP_HOLE &&
7044 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7050 /* check to see if we've wrapped (len == -1 or similar) */
7059 /* ok, we didn't find anything, lets look for delalloc */
7060 found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
7061 end, len, EXTENT_DELALLOC, 1);
7062 found_end = range_start + found;
7063 if (found_end < range_start)
7064 found_end = (u64)-1;
7067 * we didn't find anything useful, return
7068 * the original results from get_extent()
7070 if (range_start > end || found_end <= start) {
7076 /* adjust the range_start to make sure it doesn't
7077 * go backwards from the start they passed in
7079 range_start = max(start, range_start);
7080 found = found_end - range_start;
7083 u64 hole_start = start;
7086 em = alloc_extent_map();
7092 * when btrfs_get_extent can't find anything it
7093 * returns one huge hole
7095 * make sure what it found really fits our range, and
7096 * adjust to make sure it is based on the start from
7100 u64 calc_end = extent_map_end(hole_em);
7102 if (calc_end <= start || (hole_em->start > end)) {
7103 free_extent_map(hole_em);
7106 hole_start = max(hole_em->start, start);
7107 hole_len = calc_end - hole_start;
7111 if (hole_em && range_start > hole_start) {
7112 /* our hole starts before our delalloc, so we
7113 * have to return just the parts of the hole
7114 * that go until the delalloc starts
7116 em->len = min(hole_len,
7117 range_start - hole_start);
7118 em->start = hole_start;
7119 em->orig_start = hole_start;
7121 * don't adjust block start at all,
7122 * it is fixed at EXTENT_MAP_HOLE
7124 em->block_start = hole_em->block_start;
7125 em->block_len = hole_len;
7126 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7127 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7129 em->start = range_start;
7131 em->orig_start = range_start;
7132 em->block_start = EXTENT_MAP_DELALLOC;
7133 em->block_len = found;
7135 } else if (hole_em) {
7140 free_extent_map(hole_em);
7142 free_extent_map(em);
7143 return ERR_PTR(err);
7148 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7151 const u64 orig_start,
7152 const u64 block_start,
7153 const u64 block_len,
7154 const u64 orig_block_len,
7155 const u64 ram_bytes,
7158 struct extent_map *em = NULL;
7161 down_read(&BTRFS_I(inode)->dio_sem);
7162 if (type != BTRFS_ORDERED_NOCOW) {
7163 em = create_pinned_em(inode, start, len, orig_start,
7164 block_start, block_len, orig_block_len,
7169 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7170 len, block_len, type);
7173 free_extent_map(em);
7174 btrfs_drop_extent_cache(inode, start,
7175 start + len - 1, 0);
7180 up_read(&BTRFS_I(inode)->dio_sem);
7185 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7188 struct btrfs_root *root = BTRFS_I(inode)->root;
7189 struct extent_map *em;
7190 struct btrfs_key ins;
7194 alloc_hint = get_extent_allocation_hint(inode, start, len);
7195 ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
7196 alloc_hint, &ins, 1, 1);
7198 return ERR_PTR(ret);
7200 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7201 ins.objectid, ins.offset, ins.offset,
7203 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
7205 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
7211 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7212 * block must be cow'd
7214 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7215 u64 *orig_start, u64 *orig_block_len,
7218 struct btrfs_trans_handle *trans;
7219 struct btrfs_path *path;
7221 struct extent_buffer *leaf;
7222 struct btrfs_root *root = BTRFS_I(inode)->root;
7223 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7224 struct btrfs_file_extent_item *fi;
7225 struct btrfs_key key;
7232 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7234 path = btrfs_alloc_path();
7238 ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
7243 slot = path->slots[0];
7246 /* can't find the item, must cow */
7253 leaf = path->nodes[0];
7254 btrfs_item_key_to_cpu(leaf, &key, slot);
7255 if (key.objectid != btrfs_ino(inode) ||
7256 key.type != BTRFS_EXTENT_DATA_KEY) {
7257 /* not our file or wrong item type, must cow */
7261 if (key.offset > offset) {
7262 /* Wrong offset, must cow */
7266 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7267 found_type = btrfs_file_extent_type(leaf, fi);
7268 if (found_type != BTRFS_FILE_EXTENT_REG &&
7269 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7270 /* not a regular extent, must cow */
7274 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7277 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7278 if (extent_end <= offset)
7281 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7282 if (disk_bytenr == 0)
7285 if (btrfs_file_extent_compression(leaf, fi) ||
7286 btrfs_file_extent_encryption(leaf, fi) ||
7287 btrfs_file_extent_other_encoding(leaf, fi))
7290 backref_offset = btrfs_file_extent_offset(leaf, fi);
7293 *orig_start = key.offset - backref_offset;
7294 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7295 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7298 if (btrfs_extent_readonly(root, disk_bytenr))
7301 num_bytes = min(offset + *len, extent_end) - offset;
7302 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7305 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
7306 ret = test_range_bit(io_tree, offset, range_end,
7307 EXTENT_DELALLOC, 0, NULL);
7314 btrfs_release_path(path);
7317 * look for other files referencing this extent, if we
7318 * find any we must cow
7320 trans = btrfs_join_transaction(root);
7321 if (IS_ERR(trans)) {
7326 ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
7327 key.offset - backref_offset, disk_bytenr);
7328 btrfs_end_transaction(trans, root);
7335 * adjust disk_bytenr and num_bytes to cover just the bytes
7336 * in this extent we are about to write. If there
7337 * are any csums in that range we have to cow in order
7338 * to keep the csums correct
7340 disk_bytenr += backref_offset;
7341 disk_bytenr += offset - key.offset;
7342 if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7345 * all of the above have passed, it is safe to overwrite this extent
7351 btrfs_free_path(path);
7355 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7357 struct radix_tree_root *root = &inode->i_mapping->page_tree;
7359 void **pagep = NULL;
7360 struct page *page = NULL;
7364 start_idx = start >> PAGE_SHIFT;
7367 * end is the last byte in the last page. end == start is legal
7369 end_idx = end >> PAGE_SHIFT;
7373 /* Most of the code in this while loop is lifted from
7374 * find_get_page. It's been modified to begin searching from a
7375 * page and return just the first page found in that range. If the
7376 * found idx is less than or equal to the end idx then we know that
7377 * a page exists. If no pages are found or if those pages are
7378 * outside of the range then we're fine (yay!) */
7379 while (page == NULL &&
7380 radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7381 page = radix_tree_deref_slot(pagep);
7382 if (unlikely(!page))
7385 if (radix_tree_exception(page)) {
7386 if (radix_tree_deref_retry(page)) {
7391 * Otherwise, shmem/tmpfs must be storing a swap entry
7392 * here as an exceptional entry: so return it without
7393 * attempting to raise page count.
7396 break; /* TODO: Is this relevant for this use case? */
7399 if (!page_cache_get_speculative(page)) {
7405 * Has the page moved?
7406 * This is part of the lockless pagecache protocol. See
7407 * include/linux/pagemap.h for details.
7409 if (unlikely(page != *pagep)) {
7416 if (page->index <= end_idx)
7425 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7426 struct extent_state **cached_state, int writing)
7428 struct btrfs_ordered_extent *ordered;
7432 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7435 * We're concerned with the entire range that we're going to be
7436 * doing DIO to, so we need to make sure theres no ordered
7437 * extents in this range.
7439 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7440 lockend - lockstart + 1);
7443 * We need to make sure there are no buffered pages in this
7444 * range either, we could have raced between the invalidate in
7445 * generic_file_direct_write and locking the extent. The
7446 * invalidate needs to happen so that reads after a write do not
7451 !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7454 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7455 cached_state, GFP_NOFS);
7459 * If we are doing a DIO read and the ordered extent we
7460 * found is for a buffered write, we can not wait for it
7461 * to complete and retry, because if we do so we can
7462 * deadlock with concurrent buffered writes on page
7463 * locks. This happens only if our DIO read covers more
7464 * than one extent map, if at this point has already
7465 * created an ordered extent for a previous extent map
7466 * and locked its range in the inode's io tree, and a
7467 * concurrent write against that previous extent map's
7468 * range and this range started (we unlock the ranges
7469 * in the io tree only when the bios complete and
7470 * buffered writes always lock pages before attempting
7471 * to lock range in the io tree).
7474 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7475 btrfs_start_ordered_extent(inode, ordered, 1);
7478 btrfs_put_ordered_extent(ordered);
7481 * We could trigger writeback for this range (and wait
7482 * for it to complete) and then invalidate the pages for
7483 * this range (through invalidate_inode_pages2_range()),
7484 * but that can lead us to a deadlock with a concurrent
7485 * call to readpages() (a buffered read or a defrag call
7486 * triggered a readahead) on a page lock due to an
7487 * ordered dio extent we created before but did not have
7488 * yet a corresponding bio submitted (whence it can not
7489 * complete), which makes readpages() wait for that
7490 * ordered extent to complete while holding a lock on
7505 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7506 u64 len, u64 orig_start,
7507 u64 block_start, u64 block_len,
7508 u64 orig_block_len, u64 ram_bytes,
7511 struct extent_map_tree *em_tree;
7512 struct extent_map *em;
7513 struct btrfs_root *root = BTRFS_I(inode)->root;
7516 em_tree = &BTRFS_I(inode)->extent_tree;
7517 em = alloc_extent_map();
7519 return ERR_PTR(-ENOMEM);
7522 em->orig_start = orig_start;
7523 em->mod_start = start;
7526 em->block_len = block_len;
7527 em->block_start = block_start;
7528 em->bdev = root->fs_info->fs_devices->latest_bdev;
7529 em->orig_block_len = orig_block_len;
7530 em->ram_bytes = ram_bytes;
7531 em->generation = -1;
7532 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7533 if (type == BTRFS_ORDERED_PREALLOC)
7534 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7537 btrfs_drop_extent_cache(inode, em->start,
7538 em->start + em->len - 1, 0);
7539 write_lock(&em_tree->lock);
7540 ret = add_extent_mapping(em_tree, em, 1);
7541 write_unlock(&em_tree->lock);
7542 } while (ret == -EEXIST);
7545 free_extent_map(em);
7546 return ERR_PTR(ret);
7552 static void adjust_dio_outstanding_extents(struct inode *inode,
7553 struct btrfs_dio_data *dio_data,
7556 unsigned num_extents;
7558 num_extents = (unsigned) div64_u64(len + BTRFS_MAX_EXTENT_SIZE - 1,
7559 BTRFS_MAX_EXTENT_SIZE);
7561 * If we have an outstanding_extents count still set then we're
7562 * within our reservation, otherwise we need to adjust our inode
7563 * counter appropriately.
7565 if (dio_data->outstanding_extents) {
7566 dio_data->outstanding_extents -= num_extents;
7568 spin_lock(&BTRFS_I(inode)->lock);
7569 BTRFS_I(inode)->outstanding_extents += num_extents;
7570 spin_unlock(&BTRFS_I(inode)->lock);
7574 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7575 struct buffer_head *bh_result, int create)
7577 struct extent_map *em;
7578 struct btrfs_root *root = BTRFS_I(inode)->root;
7579 struct extent_state *cached_state = NULL;
7580 struct btrfs_dio_data *dio_data = NULL;
7581 u64 start = iblock << inode->i_blkbits;
7582 u64 lockstart, lockend;
7583 u64 len = bh_result->b_size;
7584 int unlock_bits = EXTENT_LOCKED;
7588 unlock_bits |= EXTENT_DIRTY;
7590 len = min_t(u64, len, root->sectorsize);
7593 lockend = start + len - 1;
7595 if (current->journal_info) {
7597 * Need to pull our outstanding extents and set journal_info to NULL so
7598 * that anything that needs to check if there's a transction doesn't get
7601 dio_data = current->journal_info;
7602 current->journal_info = NULL;
7606 * If this errors out it's because we couldn't invalidate pagecache for
7607 * this range and we need to fallback to buffered.
7609 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7615 em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7622 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7623 * io. INLINE is special, and we could probably kludge it in here, but
7624 * it's still buffered so for safety lets just fall back to the generic
7627 * For COMPRESSED we _have_ to read the entire extent in so we can
7628 * decompress it, so there will be buffering required no matter what we
7629 * do, so go ahead and fallback to buffered.
7631 * We return -ENOTBLK because thats what makes DIO go ahead and go back
7632 * to buffered IO. Don't blame me, this is the price we pay for using
7635 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7636 em->block_start == EXTENT_MAP_INLINE) {
7637 free_extent_map(em);
7642 /* Just a good old fashioned hole, return */
7643 if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7644 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7645 free_extent_map(em);
7650 * We don't allocate a new extent in the following cases
7652 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7654 * 2) The extent is marked as PREALLOC. We're good to go here and can
7655 * just use the extent.
7659 len = min(len, em->len - (start - em->start));
7660 lockstart = start + len;
7664 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7665 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7666 em->block_start != EXTENT_MAP_HOLE)) {
7668 u64 block_start, orig_start, orig_block_len, ram_bytes;
7670 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7671 type = BTRFS_ORDERED_PREALLOC;
7673 type = BTRFS_ORDERED_NOCOW;
7674 len = min(len, em->len - (start - em->start));
7675 block_start = em->block_start + (start - em->start);
7677 if (can_nocow_extent(inode, start, &len, &orig_start,
7678 &orig_block_len, &ram_bytes) == 1 &&
7679 btrfs_inc_nocow_writers(root->fs_info, block_start)) {
7680 struct extent_map *em2;
7682 em2 = btrfs_create_dio_extent(inode, start, len,
7683 orig_start, block_start,
7684 len, orig_block_len,
7686 btrfs_dec_nocow_writers(root->fs_info, block_start);
7687 if (type == BTRFS_ORDERED_PREALLOC) {
7688 free_extent_map(em);
7691 if (em2 && IS_ERR(em2)) {
7700 * this will cow the extent, reset the len in case we changed
7703 len = bh_result->b_size;
7704 free_extent_map(em);
7705 em = btrfs_new_extent_direct(inode, start, len);
7710 len = min(len, em->len - (start - em->start));
7712 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7714 bh_result->b_size = len;
7715 bh_result->b_bdev = em->bdev;
7716 set_buffer_mapped(bh_result);
7718 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7719 set_buffer_new(bh_result);
7722 * Need to update the i_size under the extent lock so buffered
7723 * readers will get the updated i_size when we unlock.
7725 if (start + len > i_size_read(inode))
7726 i_size_write(inode, start + len);
7728 adjust_dio_outstanding_extents(inode, dio_data, len);
7729 btrfs_free_reserved_data_space(inode, start, len);
7730 WARN_ON(dio_data->reserve < len);
7731 dio_data->reserve -= len;
7732 dio_data->unsubmitted_oe_range_end = start + len;
7733 current->journal_info = dio_data;
7737 * In the case of write we need to clear and unlock the entire range,
7738 * in the case of read we need to unlock only the end area that we
7739 * aren't using if there is any left over space.
7741 if (lockstart < lockend) {
7742 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7743 lockend, unlock_bits, 1, 0,
7744 &cached_state, GFP_NOFS);
7746 free_extent_state(cached_state);
7749 free_extent_map(em);
7754 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7755 unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7758 current->journal_info = dio_data;
7760 * Compensate the delalloc release we do in btrfs_direct_IO() when we
7761 * write less data then expected, so that we don't underflow our inode's
7762 * outstanding extents counter.
7764 if (create && dio_data)
7765 adjust_dio_outstanding_extents(inode, dio_data, len);
7770 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7771 int rw, int mirror_num)
7773 struct btrfs_root *root = BTRFS_I(inode)->root;
7776 BUG_ON(rw & REQ_WRITE);
7780 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7781 BTRFS_WQ_ENDIO_DIO_REPAIR);
7785 ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7791 static int btrfs_check_dio_repairable(struct inode *inode,
7792 struct bio *failed_bio,
7793 struct io_failure_record *failrec,
7798 num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7799 failrec->logical, failrec->len);
7800 if (num_copies == 1) {
7802 * we only have a single copy of the data, so don't bother with
7803 * all the retry and error correction code that follows. no
7804 * matter what the error is, it is very likely to persist.
7806 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7807 num_copies, failrec->this_mirror, failed_mirror);
7811 failrec->failed_mirror = failed_mirror;
7812 failrec->this_mirror++;
7813 if (failrec->this_mirror == failed_mirror)
7814 failrec->this_mirror++;
7816 if (failrec->this_mirror > num_copies) {
7817 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7818 num_copies, failrec->this_mirror, failed_mirror);
7825 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7826 struct page *page, unsigned int pgoff,
7827 u64 start, u64 end, int failed_mirror,
7828 bio_end_io_t *repair_endio, void *repair_arg)
7830 struct io_failure_record *failrec;
7836 BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7838 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7842 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7845 free_io_failure(inode, failrec);
7849 if ((failed_bio->bi_vcnt > 1)
7850 || (failed_bio->bi_io_vec->bv_len
7851 > BTRFS_I(inode)->root->sectorsize))
7852 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7854 read_mode = READ_SYNC;
7856 isector = start - btrfs_io_bio(failed_bio)->logical;
7857 isector >>= inode->i_sb->s_blocksize_bits;
7858 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7859 pgoff, isector, repair_endio, repair_arg);
7861 free_io_failure(inode, failrec);
7865 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7866 "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7867 read_mode, failrec->this_mirror, failrec->in_validation);
7869 ret = submit_dio_repair_bio(inode, bio, read_mode,
7870 failrec->this_mirror);
7872 free_io_failure(inode, failrec);
7879 struct btrfs_retry_complete {
7880 struct completion done;
7881 struct inode *inode;
7886 static void btrfs_retry_endio_nocsum(struct bio *bio)
7888 struct btrfs_retry_complete *done = bio->bi_private;
7889 struct inode *inode;
7890 struct bio_vec *bvec;
7896 ASSERT(bio->bi_vcnt == 1);
7897 inode = bio->bi_io_vec->bv_page->mapping->host;
7898 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7901 bio_for_each_segment_all(bvec, bio, i)
7902 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7904 complete(&done->done);
7908 static int __btrfs_correct_data_nocsum(struct inode *inode,
7909 struct btrfs_io_bio *io_bio)
7911 struct btrfs_fs_info *fs_info;
7912 struct bio_vec *bvec;
7913 struct btrfs_retry_complete done;
7921 fs_info = BTRFS_I(inode)->root->fs_info;
7922 sectorsize = BTRFS_I(inode)->root->sectorsize;
7924 start = io_bio->logical;
7927 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7928 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
7929 pgoff = bvec->bv_offset;
7931 next_block_or_try_again:
7934 init_completion(&done.done);
7936 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
7937 pgoff, start, start + sectorsize - 1,
7939 btrfs_retry_endio_nocsum, &done);
7943 wait_for_completion(&done.done);
7945 if (!done.uptodate) {
7946 /* We might have another mirror, so try again */
7947 goto next_block_or_try_again;
7950 start += sectorsize;
7953 pgoff += sectorsize;
7954 goto next_block_or_try_again;
7961 static void btrfs_retry_endio(struct bio *bio)
7963 struct btrfs_retry_complete *done = bio->bi_private;
7964 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7965 struct inode *inode;
7966 struct bio_vec *bvec;
7977 start = done->start;
7979 ASSERT(bio->bi_vcnt == 1);
7980 inode = bio->bi_io_vec->bv_page->mapping->host;
7981 ASSERT(bio->bi_io_vec->bv_len == BTRFS_I(inode)->root->sectorsize);
7983 bio_for_each_segment_all(bvec, bio, i) {
7984 ret = __readpage_endio_check(done->inode, io_bio, i,
7985 bvec->bv_page, bvec->bv_offset,
7986 done->start, bvec->bv_len);
7988 clean_io_failure(done->inode, done->start,
7989 bvec->bv_page, bvec->bv_offset);
7994 done->uptodate = uptodate;
7996 complete(&done->done);
8000 static int __btrfs_subio_endio_read(struct inode *inode,
8001 struct btrfs_io_bio *io_bio, int err)
8003 struct btrfs_fs_info *fs_info;
8004 struct bio_vec *bvec;
8005 struct btrfs_retry_complete done;
8015 fs_info = BTRFS_I(inode)->root->fs_info;
8016 sectorsize = BTRFS_I(inode)->root->sectorsize;
8019 start = io_bio->logical;
8022 bio_for_each_segment_all(bvec, &io_bio->bio, i) {
8023 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec->bv_len);
8025 pgoff = bvec->bv_offset;
8027 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8028 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8029 bvec->bv_page, pgoff, start,
8036 init_completion(&done.done);
8038 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page,
8039 pgoff, start, start + sectorsize - 1,
8041 btrfs_retry_endio, &done);
8047 wait_for_completion(&done.done);
8049 if (!done.uptodate) {
8050 /* We might have another mirror, so try again */
8054 offset += sectorsize;
8055 start += sectorsize;
8060 pgoff += sectorsize;
8068 static int btrfs_subio_endio_read(struct inode *inode,
8069 struct btrfs_io_bio *io_bio, int err)
8071 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8075 return __btrfs_correct_data_nocsum(inode, io_bio);
8079 return __btrfs_subio_endio_read(inode, io_bio, err);
8083 static void btrfs_endio_direct_read(struct bio *bio)
8085 struct btrfs_dio_private *dip = bio->bi_private;
8086 struct inode *inode = dip->inode;
8087 struct bio *dio_bio;
8088 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8089 int err = bio->bi_error;
8091 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8092 err = btrfs_subio_endio_read(inode, io_bio, err);
8094 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8095 dip->logical_offset + dip->bytes - 1);
8096 dio_bio = dip->dio_bio;
8100 dio_bio->bi_error = bio->bi_error;
8101 dio_end_io(dio_bio, bio->bi_error);
8104 io_bio->end_io(io_bio, err);
8108 static void btrfs_endio_direct_write_update_ordered(struct inode *inode,
8113 struct btrfs_root *root = BTRFS_I(inode)->root;
8114 struct btrfs_ordered_extent *ordered = NULL;
8115 u64 ordered_offset = offset;
8116 u64 ordered_bytes = bytes;
8120 ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
8127 btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
8128 finish_ordered_fn, NULL, NULL);
8129 btrfs_queue_work(root->fs_info->endio_write_workers,
8133 * our bio might span multiple ordered extents. If we haven't
8134 * completed the accounting for the whole dio, go back and try again
8136 if (ordered_offset < offset + bytes) {
8137 ordered_bytes = offset + bytes - ordered_offset;
8143 static void btrfs_endio_direct_write(struct bio *bio)
8145 struct btrfs_dio_private *dip = bio->bi_private;
8146 struct bio *dio_bio = dip->dio_bio;
8148 btrfs_endio_direct_write_update_ordered(dip->inode,
8149 dip->logical_offset,
8155 dio_bio->bi_error = bio->bi_error;
8156 dio_end_io(dio_bio, bio->bi_error);
8160 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
8161 struct bio *bio, int mirror_num,
8162 unsigned long bio_flags, u64 offset)
8165 struct btrfs_root *root = BTRFS_I(inode)->root;
8166 ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
8167 BUG_ON(ret); /* -ENOMEM */
8171 static void btrfs_end_dio_bio(struct bio *bio)
8173 struct btrfs_dio_private *dip = bio->bi_private;
8174 int err = bio->bi_error;
8177 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8178 "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
8179 btrfs_ino(dip->inode), bio->bi_rw,
8180 (unsigned long long)bio->bi_iter.bi_sector,
8181 bio->bi_iter.bi_size, err);
8183 if (dip->subio_endio)
8184 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8190 * before atomic variable goto zero, we must make sure
8191 * dip->errors is perceived to be set.
8193 smp_mb__before_atomic();
8196 /* if there are more bios still pending for this dio, just exit */
8197 if (!atomic_dec_and_test(&dip->pending_bios))
8201 bio_io_error(dip->orig_bio);
8203 dip->dio_bio->bi_error = 0;
8204 bio_endio(dip->orig_bio);
8210 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
8211 u64 first_sector, gfp_t gfp_flags)
8214 bio = btrfs_bio_alloc(bdev, first_sector, BIO_MAX_PAGES, gfp_flags);
8216 bio_associate_current(bio);
8220 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
8221 struct inode *inode,
8222 struct btrfs_dio_private *dip,
8226 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8227 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8231 * We load all the csum data we need when we submit
8232 * the first bio to reduce the csum tree search and
8235 if (dip->logical_offset == file_offset) {
8236 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
8242 if (bio == dip->orig_bio)
8245 file_offset -= dip->logical_offset;
8246 file_offset >>= inode->i_sb->s_blocksize_bits;
8247 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8252 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
8253 int rw, u64 file_offset, int skip_sum,
8256 struct btrfs_dio_private *dip = bio->bi_private;
8257 int write = rw & REQ_WRITE;
8258 struct btrfs_root *root = BTRFS_I(inode)->root;
8262 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8267 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
8268 BTRFS_WQ_ENDIO_DATA);
8276 if (write && async_submit) {
8277 ret = btrfs_wq_submit_bio(root->fs_info,
8278 inode, rw, bio, 0, 0,
8280 __btrfs_submit_bio_start_direct_io,
8281 __btrfs_submit_bio_done);
8285 * If we aren't doing async submit, calculate the csum of the
8288 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
8292 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
8298 ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
8304 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
8307 struct inode *inode = dip->inode;
8308 struct btrfs_root *root = BTRFS_I(inode)->root;
8310 struct bio *orig_bio = dip->orig_bio;
8311 struct bio_vec *bvec = orig_bio->bi_io_vec;
8312 u64 start_sector = orig_bio->bi_iter.bi_sector;
8313 u64 file_offset = dip->logical_offset;
8316 u32 blocksize = root->sectorsize;
8317 int async_submit = 0;
8322 map_length = orig_bio->bi_iter.bi_size;
8323 ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
8324 &map_length, NULL, 0);
8328 if (map_length >= orig_bio->bi_iter.bi_size) {
8330 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8334 /* async crcs make it difficult to collect full stripe writes. */
8335 if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8340 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
8344 bio->bi_private = dip;
8345 bio->bi_end_io = btrfs_end_dio_bio;
8346 btrfs_io_bio(bio)->logical = file_offset;
8347 atomic_inc(&dip->pending_bios);
8349 while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
8350 nr_sectors = BTRFS_BYTES_TO_BLKS(root->fs_info, bvec->bv_len);
8353 if (unlikely(map_length < submit_len + blocksize ||
8354 bio_add_page(bio, bvec->bv_page, blocksize,
8355 bvec->bv_offset + (i * blocksize)) < blocksize)) {
8357 * inc the count before we submit the bio so
8358 * we know the end IO handler won't happen before
8359 * we inc the count. Otherwise, the dip might get freed
8360 * before we're done setting it up
8362 atomic_inc(&dip->pending_bios);
8363 ret = __btrfs_submit_dio_bio(bio, inode, rw,
8364 file_offset, skip_sum,
8368 atomic_dec(&dip->pending_bios);
8372 start_sector += submit_len >> 9;
8373 file_offset += submit_len;
8377 bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
8378 start_sector, GFP_NOFS);
8381 bio->bi_private = dip;
8382 bio->bi_end_io = btrfs_end_dio_bio;
8383 btrfs_io_bio(bio)->logical = file_offset;
8385 map_length = orig_bio->bi_iter.bi_size;
8386 ret = btrfs_map_block(root->fs_info, rw,
8388 &map_length, NULL, 0);
8396 submit_len += blocksize;
8406 ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
8415 * before atomic variable goto zero, we must
8416 * make sure dip->errors is perceived to be set.
8418 smp_mb__before_atomic();
8419 if (atomic_dec_and_test(&dip->pending_bios))
8420 bio_io_error(dip->orig_bio);
8422 /* bio_end_io() will handle error, so we needn't return it */
8426 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
8427 struct inode *inode, loff_t file_offset)
8429 struct btrfs_dio_private *dip = NULL;
8430 struct bio *io_bio = NULL;
8431 struct btrfs_io_bio *btrfs_bio;
8433 int write = rw & REQ_WRITE;
8436 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8438 io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
8444 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8450 dip->private = dio_bio->bi_private;
8452 dip->logical_offset = file_offset;
8453 dip->bytes = dio_bio->bi_iter.bi_size;
8454 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8455 io_bio->bi_private = dip;
8456 dip->orig_bio = io_bio;
8457 dip->dio_bio = dio_bio;
8458 atomic_set(&dip->pending_bios, 0);
8459 btrfs_bio = btrfs_io_bio(io_bio);
8460 btrfs_bio->logical = file_offset;
8463 io_bio->bi_end_io = btrfs_endio_direct_write;
8465 io_bio->bi_end_io = btrfs_endio_direct_read;
8466 dip->subio_endio = btrfs_subio_endio_read;
8470 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8471 * even if we fail to submit a bio, because in such case we do the
8472 * corresponding error handling below and it must not be done a second
8473 * time by btrfs_direct_IO().
8476 struct btrfs_dio_data *dio_data = current->journal_info;
8478 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8480 dio_data->unsubmitted_oe_range_start =
8481 dio_data->unsubmitted_oe_range_end;
8484 ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8488 if (btrfs_bio->end_io)
8489 btrfs_bio->end_io(btrfs_bio, ret);
8493 * If we arrived here it means either we failed to submit the dip
8494 * or we either failed to clone the dio_bio or failed to allocate the
8495 * dip. If we cloned the dio_bio and allocated the dip, we can just
8496 * call bio_endio against our io_bio so that we get proper resource
8497 * cleanup if we fail to submit the dip, otherwise, we must do the
8498 * same as btrfs_endio_direct_[write|read] because we can't call these
8499 * callbacks - they require an allocated dip and a clone of dio_bio.
8501 if (io_bio && dip) {
8502 io_bio->bi_error = -EIO;
8505 * The end io callbacks free our dip, do the final put on io_bio
8506 * and all the cleanup and final put for dio_bio (through
8513 btrfs_endio_direct_write_update_ordered(inode,
8515 dio_bio->bi_iter.bi_size,
8518 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8519 file_offset + dio_bio->bi_iter.bi_size - 1);
8521 dio_bio->bi_error = -EIO;
8523 * Releases and cleans up our dio_bio, no need to bio_put()
8524 * nor bio_endio()/bio_io_error() against dio_bio.
8526 dio_end_io(dio_bio, ret);
8533 static ssize_t check_direct_IO(struct btrfs_root *root, struct kiocb *iocb,
8534 const struct iov_iter *iter, loff_t offset)
8538 unsigned blocksize_mask = root->sectorsize - 1;
8539 ssize_t retval = -EINVAL;
8541 if (offset & blocksize_mask)
8544 if (iov_iter_alignment(iter) & blocksize_mask)
8547 /* If this is a write we don't need to check anymore */
8548 if (iov_iter_rw(iter) == WRITE)
8551 * Check to make sure we don't have duplicate iov_base's in this
8552 * iovec, if so return EINVAL, otherwise we'll get csum errors
8553 * when reading back.
8555 for (seg = 0; seg < iter->nr_segs; seg++) {
8556 for (i = seg + 1; i < iter->nr_segs; i++) {
8557 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8566 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8568 struct file *file = iocb->ki_filp;
8569 struct inode *inode = file->f_mapping->host;
8570 struct btrfs_root *root = BTRFS_I(inode)->root;
8571 struct btrfs_dio_data dio_data = { 0 };
8572 loff_t offset = iocb->ki_pos;
8576 bool relock = false;
8579 if (check_direct_IO(BTRFS_I(inode)->root, iocb, iter, offset))
8582 inode_dio_begin(inode);
8583 smp_mb__after_atomic();
8586 * The generic stuff only does filemap_write_and_wait_range, which
8587 * isn't enough if we've written compressed pages to this area, so
8588 * we need to flush the dirty pages again to make absolutely sure
8589 * that any outstanding dirty pages are on disk.
8591 count = iov_iter_count(iter);
8592 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8593 &BTRFS_I(inode)->runtime_flags))
8594 filemap_fdatawrite_range(inode->i_mapping, offset,
8595 offset + count - 1);
8597 if (iov_iter_rw(iter) == WRITE) {
8599 * If the write DIO is beyond the EOF, we need update
8600 * the isize, but it is protected by i_mutex. So we can
8601 * not unlock the i_mutex at this case.
8603 if (offset + count <= inode->i_size) {
8604 inode_unlock(inode);
8607 ret = btrfs_delalloc_reserve_space(inode, offset, count);
8610 dio_data.outstanding_extents = div64_u64(count +
8611 BTRFS_MAX_EXTENT_SIZE - 1,
8612 BTRFS_MAX_EXTENT_SIZE);
8615 * We need to know how many extents we reserved so that we can
8616 * do the accounting properly if we go over the number we
8617 * originally calculated. Abuse current->journal_info for this.
8619 dio_data.reserve = round_up(count, root->sectorsize);
8620 dio_data.unsubmitted_oe_range_start = (u64)offset;
8621 dio_data.unsubmitted_oe_range_end = (u64)offset;
8622 current->journal_info = &dio_data;
8623 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8624 &BTRFS_I(inode)->runtime_flags)) {
8625 inode_dio_end(inode);
8626 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8630 ret = __blockdev_direct_IO(iocb, inode,
8631 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8632 iter, btrfs_get_blocks_direct, NULL,
8633 btrfs_submit_direct, flags);
8634 if (iov_iter_rw(iter) == WRITE) {
8635 current->journal_info = NULL;
8636 if (ret < 0 && ret != -EIOCBQUEUED) {
8637 if (dio_data.reserve)
8638 btrfs_delalloc_release_space(inode, offset,
8641 * On error we might have left some ordered extents
8642 * without submitting corresponding bios for them, so
8643 * cleanup them up to avoid other tasks getting them
8644 * and waiting for them to complete forever.
8646 if (dio_data.unsubmitted_oe_range_start <
8647 dio_data.unsubmitted_oe_range_end)
8648 btrfs_endio_direct_write_update_ordered(inode,
8649 dio_data.unsubmitted_oe_range_start,
8650 dio_data.unsubmitted_oe_range_end -
8651 dio_data.unsubmitted_oe_range_start,
8653 } else if (ret >= 0 && (size_t)ret < count)
8654 btrfs_delalloc_release_space(inode, offset,
8655 count - (size_t)ret);
8659 inode_dio_end(inode);
8666 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8668 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8669 __u64 start, __u64 len)
8673 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8677 return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8680 int btrfs_readpage(struct file *file, struct page *page)
8682 struct extent_io_tree *tree;
8683 tree = &BTRFS_I(page->mapping->host)->io_tree;
8684 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8687 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8689 struct extent_io_tree *tree;
8690 struct inode *inode = page->mapping->host;
8693 if (current->flags & PF_MEMALLOC) {
8694 redirty_page_for_writepage(wbc, page);
8700 * If we are under memory pressure we will call this directly from the
8701 * VM, we need to make sure we have the inode referenced for the ordered
8702 * extent. If not just return like we didn't do anything.
8704 if (!igrab(inode)) {
8705 redirty_page_for_writepage(wbc, page);
8706 return AOP_WRITEPAGE_ACTIVATE;
8708 tree = &BTRFS_I(page->mapping->host)->io_tree;
8709 ret = extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8710 btrfs_add_delayed_iput(inode);
8714 static int btrfs_writepages(struct address_space *mapping,
8715 struct writeback_control *wbc)
8717 struct extent_io_tree *tree;
8719 tree = &BTRFS_I(mapping->host)->io_tree;
8720 return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8724 btrfs_readpages(struct file *file, struct address_space *mapping,
8725 struct list_head *pages, unsigned nr_pages)
8727 struct extent_io_tree *tree;
8728 tree = &BTRFS_I(mapping->host)->io_tree;
8729 return extent_readpages(tree, mapping, pages, nr_pages,
8732 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8734 struct extent_io_tree *tree;
8735 struct extent_map_tree *map;
8738 tree = &BTRFS_I(page->mapping->host)->io_tree;
8739 map = &BTRFS_I(page->mapping->host)->extent_tree;
8740 ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8742 ClearPagePrivate(page);
8743 set_page_private(page, 0);
8749 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8751 if (PageWriteback(page) || PageDirty(page))
8753 return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8756 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8757 unsigned int length)
8759 struct inode *inode = page->mapping->host;
8760 struct extent_io_tree *tree;
8761 struct btrfs_ordered_extent *ordered;
8762 struct extent_state *cached_state = NULL;
8763 u64 page_start = page_offset(page);
8764 u64 page_end = page_start + PAGE_SIZE - 1;
8767 int inode_evicting = inode->i_state & I_FREEING;
8770 * we have the page locked, so new writeback can't start,
8771 * and the dirty bit won't be cleared while we are here.
8773 * Wait for IO on this page so that we can safely clear
8774 * the PagePrivate2 bit and do ordered accounting
8776 wait_on_page_writeback(page);
8778 tree = &BTRFS_I(inode)->io_tree;
8780 btrfs_releasepage(page, GFP_NOFS);
8784 if (!inode_evicting)
8785 lock_extent_bits(tree, page_start, page_end, &cached_state);
8788 ordered = btrfs_lookup_ordered_range(inode, start,
8789 page_end - start + 1);
8791 end = min(page_end, ordered->file_offset + ordered->len - 1);
8793 * IO on this page will never be started, so we need
8794 * to account for any ordered extents now
8796 if (!inode_evicting)
8797 clear_extent_bit(tree, start, end,
8798 EXTENT_DIRTY | EXTENT_DELALLOC |
8799 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8800 EXTENT_DEFRAG, 1, 0, &cached_state,
8803 * whoever cleared the private bit is responsible
8804 * for the finish_ordered_io
8806 if (TestClearPagePrivate2(page)) {
8807 struct btrfs_ordered_inode_tree *tree;
8810 tree = &BTRFS_I(inode)->ordered_tree;
8812 spin_lock_irq(&tree->lock);
8813 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8814 new_len = start - ordered->file_offset;
8815 if (new_len < ordered->truncated_len)
8816 ordered->truncated_len = new_len;
8817 spin_unlock_irq(&tree->lock);
8819 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8821 end - start + 1, 1))
8822 btrfs_finish_ordered_io(ordered);
8824 btrfs_put_ordered_extent(ordered);
8825 if (!inode_evicting) {
8826 cached_state = NULL;
8827 lock_extent_bits(tree, start, end,
8832 if (start < page_end)
8837 * Qgroup reserved space handler
8838 * Page here will be either
8839 * 1) Already written to disk
8840 * In this case, its reserved space is released from data rsv map
8841 * and will be freed by delayed_ref handler finally.
8842 * So even we call qgroup_free_data(), it won't decrease reserved
8844 * 2) Not written to disk
8845 * This means the reserved space should be freed here.
8847 btrfs_qgroup_free_data(inode, page_start, PAGE_SIZE);
8848 if (!inode_evicting) {
8849 clear_extent_bit(tree, page_start, page_end,
8850 EXTENT_LOCKED | EXTENT_DIRTY |
8851 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8852 EXTENT_DEFRAG, 1, 1,
8853 &cached_state, GFP_NOFS);
8855 __btrfs_releasepage(page, GFP_NOFS);
8858 ClearPageChecked(page);
8859 if (PagePrivate(page)) {
8860 ClearPagePrivate(page);
8861 set_page_private(page, 0);
8867 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8868 * called from a page fault handler when a page is first dirtied. Hence we must
8869 * be careful to check for EOF conditions here. We set the page up correctly
8870 * for a written page which means we get ENOSPC checking when writing into
8871 * holes and correct delalloc and unwritten extent mapping on filesystems that
8872 * support these features.
8874 * We are not allowed to take the i_mutex here so we have to play games to
8875 * protect against truncate races as the page could now be beyond EOF. Because
8876 * vmtruncate() writes the inode size before removing pages, once we have the
8877 * page lock we can determine safely if the page is beyond EOF. If it is not
8878 * beyond EOF, then the page is guaranteed safe against truncation until we
8881 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8883 struct page *page = vmf->page;
8884 struct inode *inode = file_inode(vma->vm_file);
8885 struct btrfs_root *root = BTRFS_I(inode)->root;
8886 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8887 struct btrfs_ordered_extent *ordered;
8888 struct extent_state *cached_state = NULL;
8890 unsigned long zero_start;
8899 reserved_space = PAGE_SIZE;
8901 sb_start_pagefault(inode->i_sb);
8902 page_start = page_offset(page);
8903 page_end = page_start + PAGE_SIZE - 1;
8907 * Reserving delalloc space after obtaining the page lock can lead to
8908 * deadlock. For example, if a dirty page is locked by this function
8909 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8910 * dirty page write out, then the btrfs_writepage() function could
8911 * end up waiting indefinitely to get a lock on the page currently
8912 * being processed by btrfs_page_mkwrite() function.
8914 ret = btrfs_delalloc_reserve_space(inode, page_start,
8917 ret = file_update_time(vma->vm_file);
8923 else /* -ENOSPC, -EIO, etc */
8924 ret = VM_FAULT_SIGBUS;
8930 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8933 size = i_size_read(inode);
8935 if ((page->mapping != inode->i_mapping) ||
8936 (page_start >= size)) {
8937 /* page got truncated out from underneath us */
8940 wait_on_page_writeback(page);
8942 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8943 set_page_extent_mapped(page);
8946 * we can't set the delalloc bits if there are pending ordered
8947 * extents. Drop our locks and wait for them to finish
8949 ordered = btrfs_lookup_ordered_range(inode, page_start, page_end);
8951 unlock_extent_cached(io_tree, page_start, page_end,
8952 &cached_state, GFP_NOFS);
8954 btrfs_start_ordered_extent(inode, ordered, 1);
8955 btrfs_put_ordered_extent(ordered);
8959 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8960 reserved_space = round_up(size - page_start, root->sectorsize);
8961 if (reserved_space < PAGE_SIZE) {
8962 end = page_start + reserved_space - 1;
8963 spin_lock(&BTRFS_I(inode)->lock);
8964 BTRFS_I(inode)->outstanding_extents++;
8965 spin_unlock(&BTRFS_I(inode)->lock);
8966 btrfs_delalloc_release_space(inode, page_start,
8967 PAGE_SIZE - reserved_space);
8972 * XXX - page_mkwrite gets called every time the page is dirtied, even
8973 * if it was already dirty, so for space accounting reasons we need to
8974 * clear any delalloc bits for the range we are fixing to save. There
8975 * is probably a better way to do this, but for now keep consistent with
8976 * prepare_pages in the normal write path.
8978 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8979 EXTENT_DIRTY | EXTENT_DELALLOC |
8980 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8981 0, 0, &cached_state, GFP_NOFS);
8983 ret = btrfs_set_extent_delalloc(inode, page_start, end,
8986 unlock_extent_cached(io_tree, page_start, page_end,
8987 &cached_state, GFP_NOFS);
8988 ret = VM_FAULT_SIGBUS;
8993 /* page is wholly or partially inside EOF */
8994 if (page_start + PAGE_SIZE > size)
8995 zero_start = size & ~PAGE_MASK;
8997 zero_start = PAGE_SIZE;
8999 if (zero_start != PAGE_SIZE) {
9001 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
9002 flush_dcache_page(page);
9005 ClearPageChecked(page);
9006 set_page_dirty(page);
9007 SetPageUptodate(page);
9009 BTRFS_I(inode)->last_trans = root->fs_info->generation;
9010 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9011 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9013 unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
9017 sb_end_pagefault(inode->i_sb);
9018 return VM_FAULT_LOCKED;
9022 btrfs_delalloc_release_space(inode, page_start, reserved_space);
9024 sb_end_pagefault(inode->i_sb);
9028 static int btrfs_truncate(struct inode *inode)
9030 struct btrfs_root *root = BTRFS_I(inode)->root;
9031 struct btrfs_block_rsv *rsv;
9034 struct btrfs_trans_handle *trans;
9035 u64 mask = root->sectorsize - 1;
9036 u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
9038 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9044 * Yes ladies and gentelment, this is indeed ugly. The fact is we have
9045 * 3 things going on here
9047 * 1) We need to reserve space for our orphan item and the space to
9048 * delete our orphan item. Lord knows we don't want to have a dangling
9049 * orphan item because we didn't reserve space to remove it.
9051 * 2) We need to reserve space to update our inode.
9053 * 3) We need to have something to cache all the space that is going to
9054 * be free'd up by the truncate operation, but also have some slack
9055 * space reserved in case it uses space during the truncate (thank you
9056 * very much snapshotting).
9058 * And we need these to all be seperate. The fact is we can use alot of
9059 * space doing the truncate, and we have no earthly idea how much space
9060 * we will use, so we need the truncate reservation to be seperate so it
9061 * doesn't end up using space reserved for updating the inode or
9062 * removing the orphan item. We also need to be able to stop the
9063 * transaction and start a new one, which means we need to be able to
9064 * update the inode several times, and we have no idea of knowing how
9065 * many times that will be, so we can't just reserve 1 item for the
9066 * entirety of the opration, so that has to be done seperately as well.
9067 * Then there is the orphan item, which does indeed need to be held on
9068 * to for the whole operation, and we need nobody to touch this reserved
9069 * space except the orphan code.
9071 * So that leaves us with
9073 * 1) root->orphan_block_rsv - for the orphan deletion.
9074 * 2) rsv - for the truncate reservation, which we will steal from the
9075 * transaction reservation.
9076 * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
9077 * updating the inode.
9079 rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
9082 rsv->size = min_size;
9086 * 1 for the truncate slack space
9087 * 1 for updating the inode.
9089 trans = btrfs_start_transaction(root, 2);
9090 if (IS_ERR(trans)) {
9091 err = PTR_ERR(trans);
9095 /* Migrate the slack space for the truncate to our reserve */
9096 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
9101 * So if we truncate and then write and fsync we normally would just
9102 * write the extents that changed, which is a problem if we need to
9103 * first truncate that entire inode. So set this flag so we write out
9104 * all of the extents in the inode to the sync log so we're completely
9107 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9108 trans->block_rsv = rsv;
9111 ret = btrfs_truncate_inode_items(trans, root, inode,
9113 BTRFS_EXTENT_DATA_KEY);
9114 if (ret != -ENOSPC && ret != -EAGAIN) {
9119 trans->block_rsv = &root->fs_info->trans_block_rsv;
9120 ret = btrfs_update_inode(trans, root, inode);
9126 btrfs_end_transaction(trans, root);
9127 btrfs_btree_balance_dirty(root);
9129 trans = btrfs_start_transaction(root, 2);
9130 if (IS_ERR(trans)) {
9131 ret = err = PTR_ERR(trans);
9136 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
9138 BUG_ON(ret); /* shouldn't happen */
9139 trans->block_rsv = rsv;
9142 if (ret == 0 && inode->i_nlink > 0) {
9143 trans->block_rsv = root->orphan_block_rsv;
9144 ret = btrfs_orphan_del(trans, inode);
9150 trans->block_rsv = &root->fs_info->trans_block_rsv;
9151 ret = btrfs_update_inode(trans, root, inode);
9155 ret = btrfs_end_transaction(trans, root);
9156 btrfs_btree_balance_dirty(root);
9160 btrfs_free_block_rsv(root, rsv);
9169 * create a new subvolume directory/inode (helper for the ioctl).
9171 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9172 struct btrfs_root *new_root,
9173 struct btrfs_root *parent_root,
9176 struct inode *inode;
9180 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9181 new_dirid, new_dirid,
9182 S_IFDIR | (~current_umask() & S_IRWXUGO),
9185 return PTR_ERR(inode);
9186 inode->i_op = &btrfs_dir_inode_operations;
9187 inode->i_fop = &btrfs_dir_file_operations;
9189 set_nlink(inode, 1);
9190 btrfs_i_size_write(inode, 0);
9191 unlock_new_inode(inode);
9193 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9195 btrfs_err(new_root->fs_info,
9196 "error inheriting subvolume %llu properties: %d",
9197 new_root->root_key.objectid, err);
9199 err = btrfs_update_inode(trans, new_root, inode);
9205 struct inode *btrfs_alloc_inode(struct super_block *sb)
9207 struct btrfs_inode *ei;
9208 struct inode *inode;
9210 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
9217 ei->last_sub_trans = 0;
9218 ei->logged_trans = 0;
9219 ei->delalloc_bytes = 0;
9220 ei->defrag_bytes = 0;
9221 ei->disk_i_size = 0;
9224 ei->index_cnt = (u64)-1;
9226 ei->last_unlink_trans = 0;
9227 ei->last_log_commit = 0;
9228 ei->delayed_iput_count = 0;
9230 spin_lock_init(&ei->lock);
9231 ei->outstanding_extents = 0;
9232 ei->reserved_extents = 0;
9234 ei->runtime_flags = 0;
9235 ei->force_compress = BTRFS_COMPRESS_NONE;
9237 ei->delayed_node = NULL;
9239 ei->i_otime.tv_sec = 0;
9240 ei->i_otime.tv_nsec = 0;
9242 inode = &ei->vfs_inode;
9243 extent_map_tree_init(&ei->extent_tree);
9244 extent_io_tree_init(&ei->io_tree, &inode->i_data);
9245 extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
9246 ei->io_tree.track_uptodate = 1;
9247 ei->io_failure_tree.track_uptodate = 1;
9248 atomic_set(&ei->sync_writers, 0);
9249 mutex_init(&ei->log_mutex);
9250 mutex_init(&ei->delalloc_mutex);
9251 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9252 INIT_LIST_HEAD(&ei->delalloc_inodes);
9253 INIT_LIST_HEAD(&ei->delayed_iput);
9254 RB_CLEAR_NODE(&ei->rb_node);
9255 init_rwsem(&ei->dio_sem);
9260 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9261 void btrfs_test_destroy_inode(struct inode *inode)
9263 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9264 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9268 static void btrfs_i_callback(struct rcu_head *head)
9270 struct inode *inode = container_of(head, struct inode, i_rcu);
9271 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9274 void btrfs_destroy_inode(struct inode *inode)
9276 struct btrfs_ordered_extent *ordered;
9277 struct btrfs_root *root = BTRFS_I(inode)->root;
9279 WARN_ON(!hlist_empty(&inode->i_dentry));
9280 WARN_ON(inode->i_data.nrpages);
9281 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9282 WARN_ON(BTRFS_I(inode)->reserved_extents);
9283 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9284 WARN_ON(BTRFS_I(inode)->csum_bytes);
9285 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9288 * This can happen where we create an inode, but somebody else also
9289 * created the same inode and we need to destroy the one we already
9295 if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
9296 &BTRFS_I(inode)->runtime_flags)) {
9297 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
9299 atomic_dec(&root->orphan_inodes);
9303 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9307 btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
9308 ordered->file_offset, ordered->len);
9309 btrfs_remove_ordered_extent(inode, ordered);
9310 btrfs_put_ordered_extent(ordered);
9311 btrfs_put_ordered_extent(ordered);
9314 btrfs_qgroup_check_reserved_leak(inode);
9315 inode_tree_del(inode);
9316 btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
9318 call_rcu(&inode->i_rcu, btrfs_i_callback);
9321 int btrfs_drop_inode(struct inode *inode)
9323 struct btrfs_root *root = BTRFS_I(inode)->root;
9328 /* the snap/subvol tree is on deleting */
9329 if (btrfs_root_refs(&root->root_item) == 0)
9332 return generic_drop_inode(inode);
9335 static void init_once(void *foo)
9337 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9339 inode_init_once(&ei->vfs_inode);
9342 void btrfs_destroy_cachep(void)
9345 * Make sure all delayed rcu free inodes are flushed before we
9349 kmem_cache_destroy(btrfs_inode_cachep);
9350 kmem_cache_destroy(btrfs_trans_handle_cachep);
9351 kmem_cache_destroy(btrfs_transaction_cachep);
9352 kmem_cache_destroy(btrfs_path_cachep);
9353 kmem_cache_destroy(btrfs_free_space_cachep);
9356 int btrfs_init_cachep(void)
9358 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9359 sizeof(struct btrfs_inode), 0,
9360 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9362 if (!btrfs_inode_cachep)
9365 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9366 sizeof(struct btrfs_trans_handle), 0,
9367 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9368 if (!btrfs_trans_handle_cachep)
9371 btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
9372 sizeof(struct btrfs_transaction), 0,
9373 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9374 if (!btrfs_transaction_cachep)
9377 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9378 sizeof(struct btrfs_path), 0,
9379 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9380 if (!btrfs_path_cachep)
9383 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9384 sizeof(struct btrfs_free_space), 0,
9385 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
9386 if (!btrfs_free_space_cachep)
9391 btrfs_destroy_cachep();
9395 static int btrfs_getattr(struct vfsmount *mnt,
9396 struct dentry *dentry, struct kstat *stat)
9399 struct inode *inode = d_inode(dentry);
9400 u32 blocksize = inode->i_sb->s_blocksize;
9402 generic_fillattr(inode, stat);
9403 stat->dev = BTRFS_I(inode)->root->anon_dev;
9405 spin_lock(&BTRFS_I(inode)->lock);
9406 delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
9407 spin_unlock(&BTRFS_I(inode)->lock);
9408 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9409 ALIGN(delalloc_bytes, blocksize)) >> 9;
9413 static int btrfs_rename_exchange(struct inode *old_dir,
9414 struct dentry *old_dentry,
9415 struct inode *new_dir,
9416 struct dentry *new_dentry)
9418 struct btrfs_trans_handle *trans;
9419 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9420 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9421 struct inode *new_inode = new_dentry->d_inode;
9422 struct inode *old_inode = old_dentry->d_inode;
9423 struct timespec ctime = CURRENT_TIME;
9424 struct dentry *parent;
9425 u64 old_ino = btrfs_ino(old_inode);
9426 u64 new_ino = btrfs_ino(new_inode);
9431 bool root_log_pinned = false;
9432 bool dest_log_pinned = false;
9434 /* we only allow rename subvolume link between subvolumes */
9435 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9438 /* close the race window with snapshot create/destroy ioctl */
9439 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9440 down_read(&root->fs_info->subvol_sem);
9441 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9442 down_read(&dest->fs_info->subvol_sem);
9445 * We want to reserve the absolute worst case amount of items. So if
9446 * both inodes are subvols and we need to unlink them then that would
9447 * require 4 item modifications, but if they are both normal inodes it
9448 * would require 5 item modifications, so we'll assume their normal
9449 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9450 * should cover the worst case number of items we'll modify.
9452 trans = btrfs_start_transaction(root, 12);
9453 if (IS_ERR(trans)) {
9454 ret = PTR_ERR(trans);
9459 * We need to find a free sequence number both in the source and
9460 * in the destination directory for the exchange.
9462 ret = btrfs_set_inode_index(new_dir, &old_idx);
9465 ret = btrfs_set_inode_index(old_dir, &new_idx);
9469 BTRFS_I(old_inode)->dir_index = 0ULL;
9470 BTRFS_I(new_inode)->dir_index = 0ULL;
9472 /* Reference for the source. */
9473 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9474 /* force full log commit if subvolume involved. */
9475 btrfs_set_log_full_commit(root->fs_info, trans);
9477 btrfs_pin_log_trans(root);
9478 root_log_pinned = true;
9479 ret = btrfs_insert_inode_ref(trans, dest,
9480 new_dentry->d_name.name,
9481 new_dentry->d_name.len,
9483 btrfs_ino(new_dir), old_idx);
9488 /* And now for the dest. */
9489 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9490 /* force full log commit if subvolume involved. */
9491 btrfs_set_log_full_commit(dest->fs_info, trans);
9493 btrfs_pin_log_trans(dest);
9494 dest_log_pinned = true;
9495 ret = btrfs_insert_inode_ref(trans, root,
9496 old_dentry->d_name.name,
9497 old_dentry->d_name.len,
9499 btrfs_ino(old_dir), new_idx);
9504 /* Update inode version and ctime/mtime. */
9505 inode_inc_iversion(old_dir);
9506 inode_inc_iversion(new_dir);
9507 inode_inc_iversion(old_inode);
9508 inode_inc_iversion(new_inode);
9509 old_dir->i_ctime = old_dir->i_mtime = ctime;
9510 new_dir->i_ctime = new_dir->i_mtime = ctime;
9511 old_inode->i_ctime = ctime;
9512 new_inode->i_ctime = ctime;
9514 if (old_dentry->d_parent != new_dentry->d_parent) {
9515 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9516 btrfs_record_unlink_dir(trans, new_dir, new_inode, 1);
9519 /* src is a subvolume */
9520 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9521 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9522 ret = btrfs_unlink_subvol(trans, root, old_dir,
9524 old_dentry->d_name.name,
9525 old_dentry->d_name.len);
9526 } else { /* src is an inode */
9527 ret = __btrfs_unlink_inode(trans, root, old_dir,
9528 old_dentry->d_inode,
9529 old_dentry->d_name.name,
9530 old_dentry->d_name.len);
9532 ret = btrfs_update_inode(trans, root, old_inode);
9535 btrfs_abort_transaction(trans, root, ret);
9539 /* dest is a subvolume */
9540 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9541 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9542 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9544 new_dentry->d_name.name,
9545 new_dentry->d_name.len);
9546 } else { /* dest is an inode */
9547 ret = __btrfs_unlink_inode(trans, dest, new_dir,
9548 new_dentry->d_inode,
9549 new_dentry->d_name.name,
9550 new_dentry->d_name.len);
9552 ret = btrfs_update_inode(trans, dest, new_inode);
9555 btrfs_abort_transaction(trans, root, ret);
9559 ret = btrfs_add_link(trans, new_dir, old_inode,
9560 new_dentry->d_name.name,
9561 new_dentry->d_name.len, 0, old_idx);
9563 btrfs_abort_transaction(trans, root, ret);
9567 ret = btrfs_add_link(trans, old_dir, new_inode,
9568 old_dentry->d_name.name,
9569 old_dentry->d_name.len, 0, new_idx);
9571 btrfs_abort_transaction(trans, root, ret);
9575 if (old_inode->i_nlink == 1)
9576 BTRFS_I(old_inode)->dir_index = old_idx;
9577 if (new_inode->i_nlink == 1)
9578 BTRFS_I(new_inode)->dir_index = new_idx;
9580 if (root_log_pinned) {
9581 parent = new_dentry->d_parent;
9582 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9583 btrfs_end_log_trans(root);
9584 root_log_pinned = false;
9586 if (dest_log_pinned) {
9587 parent = old_dentry->d_parent;
9588 btrfs_log_new_name(trans, new_inode, new_dir, parent);
9589 btrfs_end_log_trans(dest);
9590 dest_log_pinned = false;
9594 * If we have pinned a log and an error happened, we unpin tasks
9595 * trying to sync the log and force them to fallback to a transaction
9596 * commit if the log currently contains any of the inodes involved in
9597 * this rename operation (to ensure we do not persist a log with an
9598 * inconsistent state for any of these inodes or leading to any
9599 * inconsistencies when replayed). If the transaction was aborted, the
9600 * abortion reason is propagated to userspace when attempting to commit
9601 * the transaction. If the log does not contain any of these inodes, we
9602 * allow the tasks to sync it.
9604 if (ret && (root_log_pinned || dest_log_pinned)) {
9605 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9606 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9607 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9609 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9610 btrfs_set_log_full_commit(root->fs_info, trans);
9612 if (root_log_pinned) {
9613 btrfs_end_log_trans(root);
9614 root_log_pinned = false;
9616 if (dest_log_pinned) {
9617 btrfs_end_log_trans(dest);
9618 dest_log_pinned = false;
9621 ret = btrfs_end_transaction(trans, root);
9623 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9624 up_read(&dest->fs_info->subvol_sem);
9625 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9626 up_read(&root->fs_info->subvol_sem);
9631 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9632 struct btrfs_root *root,
9634 struct dentry *dentry)
9637 struct inode *inode;
9641 ret = btrfs_find_free_ino(root, &objectid);
9645 inode = btrfs_new_inode(trans, root, dir,
9646 dentry->d_name.name,
9650 S_IFCHR | WHITEOUT_MODE,
9653 if (IS_ERR(inode)) {
9654 ret = PTR_ERR(inode);
9658 inode->i_op = &btrfs_special_inode_operations;
9659 init_special_inode(inode, inode->i_mode,
9662 ret = btrfs_init_inode_security(trans, inode, dir,
9667 ret = btrfs_add_nondir(trans, dir, dentry,
9672 ret = btrfs_update_inode(trans, root, inode);
9674 unlock_new_inode(inode);
9676 inode_dec_link_count(inode);
9682 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9683 struct inode *new_dir, struct dentry *new_dentry,
9686 struct btrfs_trans_handle *trans;
9687 unsigned int trans_num_items;
9688 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9689 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9690 struct inode *new_inode = d_inode(new_dentry);
9691 struct inode *old_inode = d_inode(old_dentry);
9695 u64 old_ino = btrfs_ino(old_inode);
9696 bool log_pinned = false;
9698 if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9701 /* we only allow rename subvolume link between subvolumes */
9702 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9705 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9706 (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
9709 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9710 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9714 /* check for collisions, even if the name isn't there */
9715 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9716 new_dentry->d_name.name,
9717 new_dentry->d_name.len);
9720 if (ret == -EEXIST) {
9722 * eexist without a new_inode */
9723 if (WARN_ON(!new_inode)) {
9727 /* maybe -EOVERFLOW */
9734 * we're using rename to replace one file with another. Start IO on it
9735 * now so we don't add too much work to the end of the transaction
9737 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9738 filemap_flush(old_inode->i_mapping);
9740 /* close the racy window with snapshot create/destroy ioctl */
9741 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9742 down_read(&root->fs_info->subvol_sem);
9744 * We want to reserve the absolute worst case amount of items. So if
9745 * both inodes are subvols and we need to unlink them then that would
9746 * require 4 item modifications, but if they are both normal inodes it
9747 * would require 5 item modifications, so we'll assume they are normal
9748 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9749 * should cover the worst case number of items we'll modify.
9750 * If our rename has the whiteout flag, we need more 5 units for the
9751 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9752 * when selinux is enabled).
9754 trans_num_items = 11;
9755 if (flags & RENAME_WHITEOUT)
9756 trans_num_items += 5;
9757 trans = btrfs_start_transaction(root, trans_num_items);
9758 if (IS_ERR(trans)) {
9759 ret = PTR_ERR(trans);
9764 btrfs_record_root_in_trans(trans, dest);
9766 ret = btrfs_set_inode_index(new_dir, &index);
9770 BTRFS_I(old_inode)->dir_index = 0ULL;
9771 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9772 /* force full log commit if subvolume involved. */
9773 btrfs_set_log_full_commit(root->fs_info, trans);
9775 btrfs_pin_log_trans(root);
9777 ret = btrfs_insert_inode_ref(trans, dest,
9778 new_dentry->d_name.name,
9779 new_dentry->d_name.len,
9781 btrfs_ino(new_dir), index);
9786 inode_inc_iversion(old_dir);
9787 inode_inc_iversion(new_dir);
9788 inode_inc_iversion(old_inode);
9789 old_dir->i_ctime = old_dir->i_mtime =
9790 new_dir->i_ctime = new_dir->i_mtime =
9791 old_inode->i_ctime = current_fs_time(old_dir->i_sb);
9793 if (old_dentry->d_parent != new_dentry->d_parent)
9794 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
9796 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9797 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9798 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9799 old_dentry->d_name.name,
9800 old_dentry->d_name.len);
9802 ret = __btrfs_unlink_inode(trans, root, old_dir,
9803 d_inode(old_dentry),
9804 old_dentry->d_name.name,
9805 old_dentry->d_name.len);
9807 ret = btrfs_update_inode(trans, root, old_inode);
9810 btrfs_abort_transaction(trans, root, ret);
9815 inode_inc_iversion(new_inode);
9816 new_inode->i_ctime = current_fs_time(new_inode->i_sb);
9817 if (unlikely(btrfs_ino(new_inode) ==
9818 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9819 root_objectid = BTRFS_I(new_inode)->location.objectid;
9820 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9822 new_dentry->d_name.name,
9823 new_dentry->d_name.len);
9824 BUG_ON(new_inode->i_nlink == 0);
9826 ret = btrfs_unlink_inode(trans, dest, new_dir,
9827 d_inode(new_dentry),
9828 new_dentry->d_name.name,
9829 new_dentry->d_name.len);
9831 if (!ret && new_inode->i_nlink == 0)
9832 ret = btrfs_orphan_add(trans, d_inode(new_dentry));
9834 btrfs_abort_transaction(trans, root, ret);
9839 ret = btrfs_add_link(trans, new_dir, old_inode,
9840 new_dentry->d_name.name,
9841 new_dentry->d_name.len, 0, index);
9843 btrfs_abort_transaction(trans, root, ret);
9847 if (old_inode->i_nlink == 1)
9848 BTRFS_I(old_inode)->dir_index = index;
9851 struct dentry *parent = new_dentry->d_parent;
9853 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9854 btrfs_end_log_trans(root);
9858 if (flags & RENAME_WHITEOUT) {
9859 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9863 btrfs_abort_transaction(trans, root, ret);
9869 * If we have pinned the log and an error happened, we unpin tasks
9870 * trying to sync the log and force them to fallback to a transaction
9871 * commit if the log currently contains any of the inodes involved in
9872 * this rename operation (to ensure we do not persist a log with an
9873 * inconsistent state for any of these inodes or leading to any
9874 * inconsistencies when replayed). If the transaction was aborted, the
9875 * abortion reason is propagated to userspace when attempting to commit
9876 * the transaction. If the log does not contain any of these inodes, we
9877 * allow the tasks to sync it.
9879 if (ret && log_pinned) {
9880 if (btrfs_inode_in_log(old_dir, root->fs_info->generation) ||
9881 btrfs_inode_in_log(new_dir, root->fs_info->generation) ||
9882 btrfs_inode_in_log(old_inode, root->fs_info->generation) ||
9884 btrfs_inode_in_log(new_inode, root->fs_info->generation)))
9885 btrfs_set_log_full_commit(root->fs_info, trans);
9887 btrfs_end_log_trans(root);
9890 btrfs_end_transaction(trans, root);
9892 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9893 up_read(&root->fs_info->subvol_sem);
9898 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9899 struct inode *new_dir, struct dentry *new_dentry,
9902 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9905 if (flags & RENAME_EXCHANGE)
9906 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9909 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9912 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9914 struct btrfs_delalloc_work *delalloc_work;
9915 struct inode *inode;
9917 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9919 inode = delalloc_work->inode;
9920 filemap_flush(inode->i_mapping);
9921 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9922 &BTRFS_I(inode)->runtime_flags))
9923 filemap_flush(inode->i_mapping);
9925 if (delalloc_work->delay_iput)
9926 btrfs_add_delayed_iput(inode);
9929 complete(&delalloc_work->completion);
9932 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9935 struct btrfs_delalloc_work *work;
9937 work = kmalloc(sizeof(*work), GFP_NOFS);
9941 init_completion(&work->completion);
9942 INIT_LIST_HEAD(&work->list);
9943 work->inode = inode;
9944 work->delay_iput = delay_iput;
9945 WARN_ON_ONCE(!inode);
9946 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9947 btrfs_run_delalloc_work, NULL, NULL);
9952 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9954 wait_for_completion(&work->completion);
9959 * some fairly slow code that needs optimization. This walks the list
9960 * of all the inodes with pending delalloc and forces them to disk.
9962 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9965 struct btrfs_inode *binode;
9966 struct inode *inode;
9967 struct btrfs_delalloc_work *work, *next;
9968 struct list_head works;
9969 struct list_head splice;
9972 INIT_LIST_HEAD(&works);
9973 INIT_LIST_HEAD(&splice);
9975 mutex_lock(&root->delalloc_mutex);
9976 spin_lock(&root->delalloc_lock);
9977 list_splice_init(&root->delalloc_inodes, &splice);
9978 while (!list_empty(&splice)) {
9979 binode = list_entry(splice.next, struct btrfs_inode,
9982 list_move_tail(&binode->delalloc_inodes,
9983 &root->delalloc_inodes);
9984 inode = igrab(&binode->vfs_inode);
9986 cond_resched_lock(&root->delalloc_lock);
9989 spin_unlock(&root->delalloc_lock);
9991 work = btrfs_alloc_delalloc_work(inode, delay_iput);
9994 btrfs_add_delayed_iput(inode);
10000 list_add_tail(&work->list, &works);
10001 btrfs_queue_work(root->fs_info->flush_workers,
10004 if (nr != -1 && ret >= nr)
10007 spin_lock(&root->delalloc_lock);
10009 spin_unlock(&root->delalloc_lock);
10012 list_for_each_entry_safe(work, next, &works, list) {
10013 list_del_init(&work->list);
10014 btrfs_wait_and_free_delalloc_work(work);
10017 if (!list_empty_careful(&splice)) {
10018 spin_lock(&root->delalloc_lock);
10019 list_splice_tail(&splice, &root->delalloc_inodes);
10020 spin_unlock(&root->delalloc_lock);
10022 mutex_unlock(&root->delalloc_mutex);
10026 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
10030 if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
10033 ret = __start_delalloc_inodes(root, delay_iput, -1);
10037 * the filemap_flush will queue IO into the worker threads, but
10038 * we have to make sure the IO is actually started and that
10039 * ordered extents get created before we return
10041 atomic_inc(&root->fs_info->async_submit_draining);
10042 while (atomic_read(&root->fs_info->nr_async_submits) ||
10043 atomic_read(&root->fs_info->async_delalloc_pages)) {
10044 wait_event(root->fs_info->async_submit_wait,
10045 (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
10046 atomic_read(&root->fs_info->async_delalloc_pages) == 0));
10048 atomic_dec(&root->fs_info->async_submit_draining);
10052 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
10055 struct btrfs_root *root;
10056 struct list_head splice;
10059 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10062 INIT_LIST_HEAD(&splice);
10064 mutex_lock(&fs_info->delalloc_root_mutex);
10065 spin_lock(&fs_info->delalloc_root_lock);
10066 list_splice_init(&fs_info->delalloc_roots, &splice);
10067 while (!list_empty(&splice) && nr) {
10068 root = list_first_entry(&splice, struct btrfs_root,
10070 root = btrfs_grab_fs_root(root);
10072 list_move_tail(&root->delalloc_root,
10073 &fs_info->delalloc_roots);
10074 spin_unlock(&fs_info->delalloc_root_lock);
10076 ret = __start_delalloc_inodes(root, delay_iput, nr);
10077 btrfs_put_fs_root(root);
10085 spin_lock(&fs_info->delalloc_root_lock);
10087 spin_unlock(&fs_info->delalloc_root_lock);
10090 atomic_inc(&fs_info->async_submit_draining);
10091 while (atomic_read(&fs_info->nr_async_submits) ||
10092 atomic_read(&fs_info->async_delalloc_pages)) {
10093 wait_event(fs_info->async_submit_wait,
10094 (atomic_read(&fs_info->nr_async_submits) == 0 &&
10095 atomic_read(&fs_info->async_delalloc_pages) == 0));
10097 atomic_dec(&fs_info->async_submit_draining);
10099 if (!list_empty_careful(&splice)) {
10100 spin_lock(&fs_info->delalloc_root_lock);
10101 list_splice_tail(&splice, &fs_info->delalloc_roots);
10102 spin_unlock(&fs_info->delalloc_root_lock);
10104 mutex_unlock(&fs_info->delalloc_root_mutex);
10108 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10109 const char *symname)
10111 struct btrfs_trans_handle *trans;
10112 struct btrfs_root *root = BTRFS_I(dir)->root;
10113 struct btrfs_path *path;
10114 struct btrfs_key key;
10115 struct inode *inode = NULL;
10117 int drop_inode = 0;
10123 struct btrfs_file_extent_item *ei;
10124 struct extent_buffer *leaf;
10126 name_len = strlen(symname);
10127 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
10128 return -ENAMETOOLONG;
10131 * 2 items for inode item and ref
10132 * 2 items for dir items
10133 * 1 item for updating parent inode item
10134 * 1 item for the inline extent item
10135 * 1 item for xattr if selinux is on
10137 trans = btrfs_start_transaction(root, 7);
10139 return PTR_ERR(trans);
10141 err = btrfs_find_free_ino(root, &objectid);
10145 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10146 dentry->d_name.len, btrfs_ino(dir), objectid,
10147 S_IFLNK|S_IRWXUGO, &index);
10148 if (IS_ERR(inode)) {
10149 err = PTR_ERR(inode);
10154 * If the active LSM wants to access the inode during
10155 * d_instantiate it needs these. Smack checks to see
10156 * if the filesystem supports xattrs by looking at the
10159 inode->i_fop = &btrfs_file_operations;
10160 inode->i_op = &btrfs_file_inode_operations;
10161 inode->i_mapping->a_ops = &btrfs_aops;
10162 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10164 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10166 goto out_unlock_inode;
10168 path = btrfs_alloc_path();
10171 goto out_unlock_inode;
10173 key.objectid = btrfs_ino(inode);
10175 key.type = BTRFS_EXTENT_DATA_KEY;
10176 datasize = btrfs_file_extent_calc_inline_size(name_len);
10177 err = btrfs_insert_empty_item(trans, root, path, &key,
10180 btrfs_free_path(path);
10181 goto out_unlock_inode;
10183 leaf = path->nodes[0];
10184 ei = btrfs_item_ptr(leaf, path->slots[0],
10185 struct btrfs_file_extent_item);
10186 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10187 btrfs_set_file_extent_type(leaf, ei,
10188 BTRFS_FILE_EXTENT_INLINE);
10189 btrfs_set_file_extent_encryption(leaf, ei, 0);
10190 btrfs_set_file_extent_compression(leaf, ei, 0);
10191 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10192 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10194 ptr = btrfs_file_extent_inline_start(ei);
10195 write_extent_buffer(leaf, symname, ptr, name_len);
10196 btrfs_mark_buffer_dirty(leaf);
10197 btrfs_free_path(path);
10199 inode->i_op = &btrfs_symlink_inode_operations;
10200 inode_nohighmem(inode);
10201 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10202 inode_set_bytes(inode, name_len);
10203 btrfs_i_size_write(inode, name_len);
10204 err = btrfs_update_inode(trans, root, inode);
10206 * Last step, add directory indexes for our symlink inode. This is the
10207 * last step to avoid extra cleanup of these indexes if an error happens
10211 err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
10214 goto out_unlock_inode;
10217 unlock_new_inode(inode);
10218 d_instantiate(dentry, inode);
10221 btrfs_end_transaction(trans, root);
10223 inode_dec_link_count(inode);
10226 btrfs_btree_balance_dirty(root);
10231 unlock_new_inode(inode);
10235 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10236 u64 start, u64 num_bytes, u64 min_size,
10237 loff_t actual_len, u64 *alloc_hint,
10238 struct btrfs_trans_handle *trans)
10240 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10241 struct extent_map *em;
10242 struct btrfs_root *root = BTRFS_I(inode)->root;
10243 struct btrfs_key ins;
10244 u64 cur_offset = start;
10247 u64 last_alloc = (u64)-1;
10249 bool own_trans = true;
10253 while (num_bytes > 0) {
10255 trans = btrfs_start_transaction(root, 3);
10256 if (IS_ERR(trans)) {
10257 ret = PTR_ERR(trans);
10262 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10263 cur_bytes = max(cur_bytes, min_size);
10265 * If we are severely fragmented we could end up with really
10266 * small allocations, so if the allocator is returning small
10267 * chunks lets make its job easier by only searching for those
10270 cur_bytes = min(cur_bytes, last_alloc);
10271 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
10272 *alloc_hint, &ins, 1, 0);
10275 btrfs_end_transaction(trans, root);
10278 btrfs_dec_block_group_reservations(root->fs_info, ins.objectid);
10280 last_alloc = ins.offset;
10281 ret = insert_reserved_file_extent(trans, inode,
10282 cur_offset, ins.objectid,
10283 ins.offset, ins.offset,
10284 ins.offset, 0, 0, 0,
10285 BTRFS_FILE_EXTENT_PREALLOC);
10287 btrfs_free_reserved_extent(root, ins.objectid,
10289 btrfs_abort_transaction(trans, root, ret);
10291 btrfs_end_transaction(trans, root);
10295 btrfs_drop_extent_cache(inode, cur_offset,
10296 cur_offset + ins.offset -1, 0);
10298 em = alloc_extent_map();
10300 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10301 &BTRFS_I(inode)->runtime_flags);
10305 em->start = cur_offset;
10306 em->orig_start = cur_offset;
10307 em->len = ins.offset;
10308 em->block_start = ins.objectid;
10309 em->block_len = ins.offset;
10310 em->orig_block_len = ins.offset;
10311 em->ram_bytes = ins.offset;
10312 em->bdev = root->fs_info->fs_devices->latest_bdev;
10313 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10314 em->generation = trans->transid;
10317 write_lock(&em_tree->lock);
10318 ret = add_extent_mapping(em_tree, em, 1);
10319 write_unlock(&em_tree->lock);
10320 if (ret != -EEXIST)
10322 btrfs_drop_extent_cache(inode, cur_offset,
10323 cur_offset + ins.offset - 1,
10326 free_extent_map(em);
10328 num_bytes -= ins.offset;
10329 cur_offset += ins.offset;
10330 *alloc_hint = ins.objectid + ins.offset;
10332 inode_inc_iversion(inode);
10333 inode->i_ctime = current_fs_time(inode->i_sb);
10334 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10335 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10336 (actual_len > inode->i_size) &&
10337 (cur_offset > inode->i_size)) {
10338 if (cur_offset > actual_len)
10339 i_size = actual_len;
10341 i_size = cur_offset;
10342 i_size_write(inode, i_size);
10343 btrfs_ordered_update_i_size(inode, i_size, NULL);
10346 ret = btrfs_update_inode(trans, root, inode);
10349 btrfs_abort_transaction(trans, root, ret);
10351 btrfs_end_transaction(trans, root);
10356 btrfs_end_transaction(trans, root);
10361 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10362 u64 start, u64 num_bytes, u64 min_size,
10363 loff_t actual_len, u64 *alloc_hint)
10365 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10366 min_size, actual_len, alloc_hint,
10370 int btrfs_prealloc_file_range_trans(struct inode *inode,
10371 struct btrfs_trans_handle *trans, 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, trans);
10379 static int btrfs_set_page_dirty(struct page *page)
10381 return __set_page_dirty_nobuffers(page);
10384 static int btrfs_permission(struct inode *inode, int mask)
10386 struct btrfs_root *root = BTRFS_I(inode)->root;
10387 umode_t mode = inode->i_mode;
10389 if (mask & MAY_WRITE &&
10390 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10391 if (btrfs_root_readonly(root))
10393 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10396 return generic_permission(inode, mask);
10399 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10401 struct btrfs_trans_handle *trans;
10402 struct btrfs_root *root = BTRFS_I(dir)->root;
10403 struct inode *inode = NULL;
10409 * 5 units required for adding orphan entry
10411 trans = btrfs_start_transaction(root, 5);
10413 return PTR_ERR(trans);
10415 ret = btrfs_find_free_ino(root, &objectid);
10419 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10420 btrfs_ino(dir), objectid, mode, &index);
10421 if (IS_ERR(inode)) {
10422 ret = PTR_ERR(inode);
10427 inode->i_fop = &btrfs_file_operations;
10428 inode->i_op = &btrfs_file_inode_operations;
10430 inode->i_mapping->a_ops = &btrfs_aops;
10431 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10433 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10437 ret = btrfs_update_inode(trans, root, inode);
10440 ret = btrfs_orphan_add(trans, inode);
10445 * We set number of links to 0 in btrfs_new_inode(), and here we set
10446 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10449 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10451 set_nlink(inode, 1);
10452 unlock_new_inode(inode);
10453 d_tmpfile(dentry, inode);
10454 mark_inode_dirty(inode);
10457 btrfs_end_transaction(trans, root);
10460 btrfs_balance_delayed_items(root);
10461 btrfs_btree_balance_dirty(root);
10465 unlock_new_inode(inode);
10470 /* Inspired by filemap_check_errors() */
10471 int btrfs_inode_check_errors(struct inode *inode)
10475 if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
10476 test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
10478 if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
10479 test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
10485 static const struct inode_operations btrfs_dir_inode_operations = {
10486 .getattr = btrfs_getattr,
10487 .lookup = btrfs_lookup,
10488 .create = btrfs_create,
10489 .unlink = btrfs_unlink,
10490 .link = btrfs_link,
10491 .mkdir = btrfs_mkdir,
10492 .rmdir = btrfs_rmdir,
10493 .rename2 = btrfs_rename2,
10494 .symlink = btrfs_symlink,
10495 .setattr = btrfs_setattr,
10496 .mknod = btrfs_mknod,
10497 .setxattr = generic_setxattr,
10498 .getxattr = generic_getxattr,
10499 .listxattr = btrfs_listxattr,
10500 .removexattr = generic_removexattr,
10501 .permission = btrfs_permission,
10502 .get_acl = btrfs_get_acl,
10503 .set_acl = btrfs_set_acl,
10504 .update_time = btrfs_update_time,
10505 .tmpfile = btrfs_tmpfile,
10507 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10508 .lookup = btrfs_lookup,
10509 .permission = btrfs_permission,
10510 .get_acl = btrfs_get_acl,
10511 .set_acl = btrfs_set_acl,
10512 .update_time = btrfs_update_time,
10515 static const struct file_operations btrfs_dir_file_operations = {
10516 .llseek = generic_file_llseek,
10517 .read = generic_read_dir,
10518 .iterate = btrfs_real_readdir,
10519 .unlocked_ioctl = btrfs_ioctl,
10520 #ifdef CONFIG_COMPAT
10521 .compat_ioctl = btrfs_compat_ioctl,
10523 .release = btrfs_release_file,
10524 .fsync = btrfs_sync_file,
10527 static const struct extent_io_ops btrfs_extent_io_ops = {
10528 .fill_delalloc = run_delalloc_range,
10529 .submit_bio_hook = btrfs_submit_bio_hook,
10530 .merge_bio_hook = btrfs_merge_bio_hook,
10531 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10532 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10533 .writepage_start_hook = btrfs_writepage_start_hook,
10534 .set_bit_hook = btrfs_set_bit_hook,
10535 .clear_bit_hook = btrfs_clear_bit_hook,
10536 .merge_extent_hook = btrfs_merge_extent_hook,
10537 .split_extent_hook = btrfs_split_extent_hook,
10541 * btrfs doesn't support the bmap operation because swapfiles
10542 * use bmap to make a mapping of extents in the file. They assume
10543 * these extents won't change over the life of the file and they
10544 * use the bmap result to do IO directly to the drive.
10546 * the btrfs bmap call would return logical addresses that aren't
10547 * suitable for IO and they also will change frequently as COW
10548 * operations happen. So, swapfile + btrfs == corruption.
10550 * For now we're avoiding this by dropping bmap.
10552 static const struct address_space_operations btrfs_aops = {
10553 .readpage = btrfs_readpage,
10554 .writepage = btrfs_writepage,
10555 .writepages = btrfs_writepages,
10556 .readpages = btrfs_readpages,
10557 .direct_IO = btrfs_direct_IO,
10558 .invalidatepage = btrfs_invalidatepage,
10559 .releasepage = btrfs_releasepage,
10560 .set_page_dirty = btrfs_set_page_dirty,
10561 .error_remove_page = generic_error_remove_page,
10564 static const struct address_space_operations btrfs_symlink_aops = {
10565 .readpage = btrfs_readpage,
10566 .writepage = btrfs_writepage,
10567 .invalidatepage = btrfs_invalidatepage,
10568 .releasepage = btrfs_releasepage,
10571 static const struct inode_operations btrfs_file_inode_operations = {
10572 .getattr = btrfs_getattr,
10573 .setattr = btrfs_setattr,
10574 .setxattr = generic_setxattr,
10575 .getxattr = generic_getxattr,
10576 .listxattr = btrfs_listxattr,
10577 .removexattr = generic_removexattr,
10578 .permission = btrfs_permission,
10579 .fiemap = btrfs_fiemap,
10580 .get_acl = btrfs_get_acl,
10581 .set_acl = btrfs_set_acl,
10582 .update_time = btrfs_update_time,
10584 static const struct inode_operations btrfs_special_inode_operations = {
10585 .getattr = btrfs_getattr,
10586 .setattr = btrfs_setattr,
10587 .permission = btrfs_permission,
10588 .setxattr = generic_setxattr,
10589 .getxattr = generic_getxattr,
10590 .listxattr = btrfs_listxattr,
10591 .removexattr = generic_removexattr,
10592 .get_acl = btrfs_get_acl,
10593 .set_acl = btrfs_set_acl,
10594 .update_time = btrfs_update_time,
10596 static const struct inode_operations btrfs_symlink_inode_operations = {
10597 .readlink = generic_readlink,
10598 .get_link = page_get_link,
10599 .getattr = btrfs_getattr,
10600 .setattr = btrfs_setattr,
10601 .permission = btrfs_permission,
10602 .setxattr = generic_setxattr,
10603 .getxattr = generic_getxattr,
10604 .listxattr = btrfs_listxattr,
10605 .removexattr = generic_removexattr,
10606 .update_time = btrfs_update_time,
10609 const struct dentry_operations btrfs_dentry_operations = {
10610 .d_delete = btrfs_dentry_delete,
10611 .d_release = btrfs_dentry_release,