1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/kernel.h>
8 #include <linux/buffer_head.h>
9 #include <linux/file.h>
11 #include <linux/pagemap.h>
12 #include <linux/highmem.h>
13 #include <linux/time.h>
14 #include <linux/init.h>
15 #include <linux/string.h>
16 #include <linux/backing-dev.h>
17 #include <linux/writeback.h>
18 #include <linux/compat.h>
19 #include <linux/xattr.h>
20 #include <linux/posix_acl.h>
21 #include <linux/falloc.h>
22 #include <linux/slab.h>
23 #include <linux/ratelimit.h>
24 #include <linux/btrfs.h>
25 #include <linux/blkdev.h>
26 #include <linux/posix_acl_xattr.h>
27 #include <linux/uio.h>
28 #include <linux/magic.h>
29 #include <linux/iversion.h>
30 #include <asm/unaligned.h>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "print-tree.h"
36 #include "ordered-data.h"
40 #include "compression.h"
42 #include "free-space-cache.h"
43 #include "inode-map.h"
49 struct btrfs_iget_args {
50 struct btrfs_key *location;
51 struct btrfs_root *root;
54 struct btrfs_dio_data {
56 u64 unsubmitted_oe_range_start;
57 u64 unsubmitted_oe_range_end;
61 static const struct inode_operations btrfs_dir_inode_operations;
62 static const struct inode_operations btrfs_symlink_inode_operations;
63 static const struct inode_operations btrfs_dir_ro_inode_operations;
64 static const struct inode_operations btrfs_special_inode_operations;
65 static const struct inode_operations btrfs_file_inode_operations;
66 static const struct address_space_operations btrfs_aops;
67 static const struct file_operations btrfs_dir_file_operations;
68 static const struct extent_io_ops btrfs_extent_io_ops;
70 static struct kmem_cache *btrfs_inode_cachep;
71 struct kmem_cache *btrfs_trans_handle_cachep;
72 struct kmem_cache *btrfs_path_cachep;
73 struct kmem_cache *btrfs_free_space_cachep;
76 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
77 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
78 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
79 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
80 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
81 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
82 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
83 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
86 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
87 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
88 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
89 static noinline int cow_file_range(struct inode *inode,
90 struct page *locked_page,
91 u64 start, u64 end, u64 delalloc_end,
92 int *page_started, unsigned long *nr_written,
93 int unlock, struct btrfs_dedupe_hash *hash);
94 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
95 u64 orig_start, u64 block_start,
96 u64 block_len, u64 orig_block_len,
97 u64 ram_bytes, int compress_type,
100 static void __endio_write_update_ordered(struct inode *inode,
101 const u64 offset, const u64 bytes,
102 const bool uptodate);
105 * Cleanup all submitted ordered extents in specified range to handle errors
106 * from the fill_dellaloc() callback.
108 * NOTE: caller must ensure that when an error happens, it can not call
109 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
110 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
111 * to be released, which we want to happen only when finishing the ordered
112 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
113 * fill_delalloc() callback already does proper cleanup for the first page of
114 * the range, that is, it invokes the callback writepage_end_io_hook() for the
115 * range of the first page.
117 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
121 unsigned long index = offset >> PAGE_SHIFT;
122 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
125 while (index <= end_index) {
126 page = find_get_page(inode->i_mapping, index);
130 ClearPagePrivate2(page);
133 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
134 bytes - PAGE_SIZE, false);
137 static int btrfs_dirty_inode(struct inode *inode);
139 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
140 void btrfs_test_inode_set_ops(struct inode *inode)
142 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
146 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
147 struct inode *inode, struct inode *dir,
148 const struct qstr *qstr)
152 err = btrfs_init_acl(trans, inode, dir);
154 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
159 * this does all the hard work for inserting an inline extent into
160 * the btree. The caller should have done a btrfs_drop_extents so that
161 * no overlapping inline items exist in the btree
163 static int insert_inline_extent(struct btrfs_trans_handle *trans,
164 struct btrfs_path *path, int extent_inserted,
165 struct btrfs_root *root, struct inode *inode,
166 u64 start, size_t size, size_t compressed_size,
168 struct page **compressed_pages)
170 struct extent_buffer *leaf;
171 struct page *page = NULL;
174 struct btrfs_file_extent_item *ei;
176 size_t cur_size = size;
177 unsigned long offset;
179 if (compressed_size && compressed_pages)
180 cur_size = compressed_size;
182 inode_add_bytes(inode, size);
184 if (!extent_inserted) {
185 struct btrfs_key key;
188 key.objectid = btrfs_ino(BTRFS_I(inode));
190 key.type = BTRFS_EXTENT_DATA_KEY;
192 datasize = btrfs_file_extent_calc_inline_size(cur_size);
193 path->leave_spinning = 1;
194 ret = btrfs_insert_empty_item(trans, root, path, &key,
199 leaf = path->nodes[0];
200 ei = btrfs_item_ptr(leaf, path->slots[0],
201 struct btrfs_file_extent_item);
202 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
203 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
204 btrfs_set_file_extent_encryption(leaf, ei, 0);
205 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
206 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
207 ptr = btrfs_file_extent_inline_start(ei);
209 if (compress_type != BTRFS_COMPRESS_NONE) {
212 while (compressed_size > 0) {
213 cpage = compressed_pages[i];
214 cur_size = min_t(unsigned long, compressed_size,
217 kaddr = kmap_atomic(cpage);
218 write_extent_buffer(leaf, kaddr, ptr, cur_size);
219 kunmap_atomic(kaddr);
223 compressed_size -= cur_size;
225 btrfs_set_file_extent_compression(leaf, ei,
228 page = find_get_page(inode->i_mapping,
229 start >> PAGE_SHIFT);
230 btrfs_set_file_extent_compression(leaf, ei, 0);
231 kaddr = kmap_atomic(page);
232 offset = start & (PAGE_SIZE - 1);
233 write_extent_buffer(leaf, kaddr + offset, ptr, size);
234 kunmap_atomic(kaddr);
237 btrfs_mark_buffer_dirty(leaf);
238 btrfs_release_path(path);
241 * we're an inline extent, so nobody can
242 * extend the file past i_size without locking
243 * a page we already have locked.
245 * We must do any isize and inode updates
246 * before we unlock the pages. Otherwise we
247 * could end up racing with unlink.
249 BTRFS_I(inode)->disk_i_size = inode->i_size;
250 ret = btrfs_update_inode(trans, root, inode);
258 * conditionally insert an inline extent into the file. This
259 * does the checks required to make sure the data is small enough
260 * to fit as an inline extent.
262 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
263 u64 end, size_t compressed_size,
265 struct page **compressed_pages)
267 struct btrfs_root *root = BTRFS_I(inode)->root;
268 struct btrfs_fs_info *fs_info = root->fs_info;
269 struct btrfs_trans_handle *trans;
270 u64 isize = i_size_read(inode);
271 u64 actual_end = min(end + 1, isize);
272 u64 inline_len = actual_end - start;
273 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
274 u64 data_len = inline_len;
276 struct btrfs_path *path;
277 int extent_inserted = 0;
278 u32 extent_item_size;
281 data_len = compressed_size;
284 actual_end > fs_info->sectorsize ||
285 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
287 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
289 data_len > fs_info->max_inline) {
293 path = btrfs_alloc_path();
297 trans = btrfs_join_transaction(root);
299 btrfs_free_path(path);
300 return PTR_ERR(trans);
302 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
304 if (compressed_size && compressed_pages)
305 extent_item_size = btrfs_file_extent_calc_inline_size(
308 extent_item_size = btrfs_file_extent_calc_inline_size(
311 ret = __btrfs_drop_extents(trans, root, inode, path,
312 start, aligned_end, NULL,
313 1, 1, extent_item_size, &extent_inserted);
315 btrfs_abort_transaction(trans, ret);
319 if (isize > actual_end)
320 inline_len = min_t(u64, isize, actual_end);
321 ret = insert_inline_extent(trans, path, extent_inserted,
323 inline_len, compressed_size,
324 compress_type, compressed_pages);
325 if (ret && ret != -ENOSPC) {
326 btrfs_abort_transaction(trans, ret);
328 } else if (ret == -ENOSPC) {
333 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
334 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
337 * Don't forget to free the reserved space, as for inlined extent
338 * it won't count as data extent, free them directly here.
339 * And at reserve time, it's always aligned to page size, so
340 * just free one page here.
342 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
343 btrfs_free_path(path);
344 btrfs_end_transaction(trans);
348 struct async_extent {
353 unsigned long nr_pages;
355 struct list_head list;
360 struct btrfs_root *root;
361 struct page *locked_page;
364 unsigned int write_flags;
365 struct list_head extents;
366 struct btrfs_work work;
369 static noinline int add_async_extent(struct async_cow *cow,
370 u64 start, u64 ram_size,
373 unsigned long nr_pages,
376 struct async_extent *async_extent;
378 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
379 BUG_ON(!async_extent); /* -ENOMEM */
380 async_extent->start = start;
381 async_extent->ram_size = ram_size;
382 async_extent->compressed_size = compressed_size;
383 async_extent->pages = pages;
384 async_extent->nr_pages = nr_pages;
385 async_extent->compress_type = compress_type;
386 list_add_tail(&async_extent->list, &cow->extents);
390 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
392 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
395 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
398 if (BTRFS_I(inode)->defrag_compress)
400 /* bad compression ratios */
401 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
403 if (btrfs_test_opt(fs_info, COMPRESS) ||
404 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
405 BTRFS_I(inode)->prop_compress)
406 return btrfs_compress_heuristic(inode, start, end);
410 static inline void inode_should_defrag(struct btrfs_inode *inode,
411 u64 start, u64 end, u64 num_bytes, u64 small_write)
413 /* If this is a small write inside eof, kick off a defrag */
414 if (num_bytes < small_write &&
415 (start > 0 || end + 1 < inode->disk_i_size))
416 btrfs_add_inode_defrag(NULL, inode);
420 * we create compressed extents in two phases. The first
421 * phase compresses a range of pages that have already been
422 * locked (both pages and state bits are locked).
424 * This is done inside an ordered work queue, and the compression
425 * is spread across many cpus. The actual IO submission is step
426 * two, and the ordered work queue takes care of making sure that
427 * happens in the same order things were put onto the queue by
428 * writepages and friends.
430 * If this code finds it can't get good compression, it puts an
431 * entry onto the work queue to write the uncompressed bytes. This
432 * makes sure that both compressed inodes and uncompressed inodes
433 * are written in the same order that the flusher thread sent them
436 static noinline void compress_file_range(struct inode *inode,
437 struct page *locked_page,
439 struct async_cow *async_cow,
442 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
443 u64 blocksize = fs_info->sectorsize;
445 u64 isize = i_size_read(inode);
447 struct page **pages = NULL;
448 unsigned long nr_pages;
449 unsigned long total_compressed = 0;
450 unsigned long total_in = 0;
453 int compress_type = fs_info->compress_type;
456 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
459 actual_end = min_t(u64, isize, end + 1);
462 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
463 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
464 nr_pages = min_t(unsigned long, nr_pages,
465 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
468 * we don't want to send crud past the end of i_size through
469 * compression, that's just a waste of CPU time. So, if the
470 * end of the file is before the start of our current
471 * requested range of bytes, we bail out to the uncompressed
472 * cleanup code that can deal with all of this.
474 * It isn't really the fastest way to fix things, but this is a
475 * very uncommon corner.
477 if (actual_end <= start)
478 goto cleanup_and_bail_uncompressed;
480 total_compressed = actual_end - start;
483 * skip compression for a small file range(<=blocksize) that
484 * isn't an inline extent, since it doesn't save disk space at all.
486 if (total_compressed <= blocksize &&
487 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
488 goto cleanup_and_bail_uncompressed;
490 total_compressed = min_t(unsigned long, total_compressed,
491 BTRFS_MAX_UNCOMPRESSED);
496 * we do compression for mount -o compress and when the
497 * inode has not been flagged as nocompress. This flag can
498 * change at any time if we discover bad compression ratios.
500 if (inode_need_compress(inode, start, end)) {
502 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
504 /* just bail out to the uncompressed code */
508 if (BTRFS_I(inode)->defrag_compress)
509 compress_type = BTRFS_I(inode)->defrag_compress;
510 else if (BTRFS_I(inode)->prop_compress)
511 compress_type = BTRFS_I(inode)->prop_compress;
514 * we need to call clear_page_dirty_for_io on each
515 * page in the range. Otherwise applications with the file
516 * mmap'd can wander in and change the page contents while
517 * we are compressing them.
519 * If the compression fails for any reason, we set the pages
520 * dirty again later on.
522 * Note that the remaining part is redirtied, the start pointer
523 * has moved, the end is the original one.
526 extent_range_clear_dirty_for_io(inode, start, end);
530 /* Compression level is applied here and only here */
531 ret = btrfs_compress_pages(
532 compress_type | (fs_info->compress_level << 4),
533 inode->i_mapping, start,
540 unsigned long offset = total_compressed &
542 struct page *page = pages[nr_pages - 1];
545 /* zero the tail end of the last page, we might be
546 * sending it down to disk
549 kaddr = kmap_atomic(page);
550 memset(kaddr + offset, 0,
552 kunmap_atomic(kaddr);
559 /* lets try to make an inline extent */
560 if (ret || total_in < actual_end) {
561 /* we didn't compress the entire range, try
562 * to make an uncompressed inline extent.
564 ret = cow_file_range_inline(inode, start, end, 0,
565 BTRFS_COMPRESS_NONE, NULL);
567 /* try making a compressed inline extent */
568 ret = cow_file_range_inline(inode, start, end,
570 compress_type, pages);
573 unsigned long clear_flags = EXTENT_DELALLOC |
574 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
575 EXTENT_DO_ACCOUNTING;
576 unsigned long page_error_op;
578 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
581 * inline extent creation worked or returned error,
582 * we don't need to create any more async work items.
583 * Unlock and free up our temp pages.
585 * We use DO_ACCOUNTING here because we need the
586 * delalloc_release_metadata to be done _after_ we drop
587 * our outstanding extent for clearing delalloc for this
590 extent_clear_unlock_delalloc(inode, start, end, end,
603 * we aren't doing an inline extent round the compressed size
604 * up to a block size boundary so the allocator does sane
607 total_compressed = ALIGN(total_compressed, blocksize);
610 * one last check to make sure the compression is really a
611 * win, compare the page count read with the blocks on disk,
612 * compression must free at least one sector size
614 total_in = ALIGN(total_in, PAGE_SIZE);
615 if (total_compressed + blocksize <= total_in) {
619 * The async work queues will take care of doing actual
620 * allocation on disk for these compressed pages, and
621 * will submit them to the elevator.
623 add_async_extent(async_cow, start, total_in,
624 total_compressed, pages, nr_pages,
627 if (start + total_in < end) {
638 * the compression code ran but failed to make things smaller,
639 * free any pages it allocated and our page pointer array
641 for (i = 0; i < nr_pages; i++) {
642 WARN_ON(pages[i]->mapping);
647 total_compressed = 0;
650 /* flag the file so we don't compress in the future */
651 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
652 !(BTRFS_I(inode)->prop_compress)) {
653 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
656 cleanup_and_bail_uncompressed:
658 * No compression, but we still need to write the pages in the file
659 * we've been given so far. redirty the locked page if it corresponds
660 * to our extent and set things up for the async work queue to run
661 * cow_file_range to do the normal delalloc dance.
663 if (page_offset(locked_page) >= start &&
664 page_offset(locked_page) <= end)
665 __set_page_dirty_nobuffers(locked_page);
666 /* unlocked later on in the async handlers */
669 extent_range_redirty_for_io(inode, start, end);
670 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
671 BTRFS_COMPRESS_NONE);
677 for (i = 0; i < nr_pages; i++) {
678 WARN_ON(pages[i]->mapping);
684 static void free_async_extent_pages(struct async_extent *async_extent)
688 if (!async_extent->pages)
691 for (i = 0; i < async_extent->nr_pages; i++) {
692 WARN_ON(async_extent->pages[i]->mapping);
693 put_page(async_extent->pages[i]);
695 kfree(async_extent->pages);
696 async_extent->nr_pages = 0;
697 async_extent->pages = NULL;
701 * phase two of compressed writeback. This is the ordered portion
702 * of the code, which only gets called in the order the work was
703 * queued. We walk all the async extents created by compress_file_range
704 * and send them down to the disk.
706 static noinline void submit_compressed_extents(struct inode *inode,
707 struct async_cow *async_cow)
709 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
710 struct async_extent *async_extent;
712 struct btrfs_key ins;
713 struct extent_map *em;
714 struct btrfs_root *root = BTRFS_I(inode)->root;
715 struct extent_io_tree *io_tree;
719 while (!list_empty(&async_cow->extents)) {
720 async_extent = list_entry(async_cow->extents.next,
721 struct async_extent, list);
722 list_del(&async_extent->list);
724 io_tree = &BTRFS_I(inode)->io_tree;
727 /* did the compression code fall back to uncompressed IO? */
728 if (!async_extent->pages) {
729 int page_started = 0;
730 unsigned long nr_written = 0;
732 lock_extent(io_tree, async_extent->start,
733 async_extent->start +
734 async_extent->ram_size - 1);
736 /* allocate blocks */
737 ret = cow_file_range(inode, async_cow->locked_page,
739 async_extent->start +
740 async_extent->ram_size - 1,
741 async_extent->start +
742 async_extent->ram_size - 1,
743 &page_started, &nr_written, 0,
749 * if page_started, cow_file_range inserted an
750 * inline extent and took care of all the unlocking
751 * and IO for us. Otherwise, we need to submit
752 * all those pages down to the drive.
754 if (!page_started && !ret)
755 extent_write_locked_range(inode,
757 async_extent->start +
758 async_extent->ram_size - 1,
761 unlock_page(async_cow->locked_page);
767 lock_extent(io_tree, async_extent->start,
768 async_extent->start + async_extent->ram_size - 1);
770 ret = btrfs_reserve_extent(root, async_extent->ram_size,
771 async_extent->compressed_size,
772 async_extent->compressed_size,
773 0, alloc_hint, &ins, 1, 1);
775 free_async_extent_pages(async_extent);
777 if (ret == -ENOSPC) {
778 unlock_extent(io_tree, async_extent->start,
779 async_extent->start +
780 async_extent->ram_size - 1);
783 * we need to redirty the pages if we decide to
784 * fallback to uncompressed IO, otherwise we
785 * will not submit these pages down to lower
788 extent_range_redirty_for_io(inode,
790 async_extent->start +
791 async_extent->ram_size - 1);
798 * here we're doing allocation and writeback of the
801 em = create_io_em(inode, async_extent->start,
802 async_extent->ram_size, /* len */
803 async_extent->start, /* orig_start */
804 ins.objectid, /* block_start */
805 ins.offset, /* block_len */
806 ins.offset, /* orig_block_len */
807 async_extent->ram_size, /* ram_bytes */
808 async_extent->compress_type,
809 BTRFS_ORDERED_COMPRESSED);
811 /* ret value is not necessary due to void function */
812 goto out_free_reserve;
815 ret = btrfs_add_ordered_extent_compress(inode,
818 async_extent->ram_size,
820 BTRFS_ORDERED_COMPRESSED,
821 async_extent->compress_type);
823 btrfs_drop_extent_cache(BTRFS_I(inode),
825 async_extent->start +
826 async_extent->ram_size - 1, 0);
827 goto out_free_reserve;
829 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
832 * clear dirty, set writeback and unlock the pages.
834 extent_clear_unlock_delalloc(inode, async_extent->start,
835 async_extent->start +
836 async_extent->ram_size - 1,
837 async_extent->start +
838 async_extent->ram_size - 1,
839 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
840 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
842 if (btrfs_submit_compressed_write(inode,
844 async_extent->ram_size,
846 ins.offset, async_extent->pages,
847 async_extent->nr_pages,
848 async_cow->write_flags)) {
849 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
850 struct page *p = async_extent->pages[0];
851 const u64 start = async_extent->start;
852 const u64 end = start + async_extent->ram_size - 1;
854 p->mapping = inode->i_mapping;
855 tree->ops->writepage_end_io_hook(p, start, end,
858 extent_clear_unlock_delalloc(inode, start, end, end,
862 free_async_extent_pages(async_extent);
864 alloc_hint = ins.objectid + ins.offset;
870 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
871 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
873 extent_clear_unlock_delalloc(inode, async_extent->start,
874 async_extent->start +
875 async_extent->ram_size - 1,
876 async_extent->start +
877 async_extent->ram_size - 1,
878 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
879 EXTENT_DELALLOC_NEW |
880 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
881 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
882 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
884 free_async_extent_pages(async_extent);
889 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
892 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
893 struct extent_map *em;
896 read_lock(&em_tree->lock);
897 em = search_extent_mapping(em_tree, start, num_bytes);
900 * if block start isn't an actual block number then find the
901 * first block in this inode and use that as a hint. If that
902 * block is also bogus then just don't worry about it.
904 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
906 em = search_extent_mapping(em_tree, 0, 0);
907 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
908 alloc_hint = em->block_start;
912 alloc_hint = em->block_start;
916 read_unlock(&em_tree->lock);
922 * when extent_io.c finds a delayed allocation range in the file,
923 * the call backs end up in this code. The basic idea is to
924 * allocate extents on disk for the range, and create ordered data structs
925 * in ram to track those extents.
927 * locked_page is the page that writepage had locked already. We use
928 * it to make sure we don't do extra locks or unlocks.
930 * *page_started is set to one if we unlock locked_page and do everything
931 * required to start IO on it. It may be clean and already done with
934 static noinline int cow_file_range(struct inode *inode,
935 struct page *locked_page,
936 u64 start, u64 end, u64 delalloc_end,
937 int *page_started, unsigned long *nr_written,
938 int unlock, struct btrfs_dedupe_hash *hash)
940 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
941 struct btrfs_root *root = BTRFS_I(inode)->root;
944 unsigned long ram_size;
945 u64 cur_alloc_size = 0;
946 u64 blocksize = fs_info->sectorsize;
947 struct btrfs_key ins;
948 struct extent_map *em;
950 unsigned long page_ops;
951 bool extent_reserved = false;
954 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
960 num_bytes = ALIGN(end - start + 1, blocksize);
961 num_bytes = max(blocksize, num_bytes);
962 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
964 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
967 /* lets try to make an inline extent */
968 ret = cow_file_range_inline(inode, start, end, 0,
969 BTRFS_COMPRESS_NONE, NULL);
972 * We use DO_ACCOUNTING here because we need the
973 * delalloc_release_metadata to be run _after_ we drop
974 * our outstanding extent for clearing delalloc for this
977 extent_clear_unlock_delalloc(inode, start, end,
979 EXTENT_LOCKED | EXTENT_DELALLOC |
980 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
981 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
982 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
984 *nr_written = *nr_written +
985 (end - start + PAGE_SIZE) / PAGE_SIZE;
988 } else if (ret < 0) {
993 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
994 btrfs_drop_extent_cache(BTRFS_I(inode), start,
995 start + num_bytes - 1, 0);
997 while (num_bytes > 0) {
998 cur_alloc_size = num_bytes;
999 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1000 fs_info->sectorsize, 0, alloc_hint,
1004 cur_alloc_size = ins.offset;
1005 extent_reserved = true;
1007 ram_size = ins.offset;
1008 em = create_io_em(inode, start, ins.offset, /* len */
1009 start, /* orig_start */
1010 ins.objectid, /* block_start */
1011 ins.offset, /* block_len */
1012 ins.offset, /* orig_block_len */
1013 ram_size, /* ram_bytes */
1014 BTRFS_COMPRESS_NONE, /* compress_type */
1015 BTRFS_ORDERED_REGULAR /* type */);
1020 free_extent_map(em);
1022 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1023 ram_size, cur_alloc_size, 0);
1025 goto out_drop_extent_cache;
1027 if (root->root_key.objectid ==
1028 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1029 ret = btrfs_reloc_clone_csums(inode, start,
1032 * Only drop cache here, and process as normal.
1034 * We must not allow extent_clear_unlock_delalloc()
1035 * at out_unlock label to free meta of this ordered
1036 * extent, as its meta should be freed by
1037 * btrfs_finish_ordered_io().
1039 * So we must continue until @start is increased to
1040 * skip current ordered extent.
1043 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1044 start + ram_size - 1, 0);
1047 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1049 /* we're not doing compressed IO, don't unlock the first
1050 * page (which the caller expects to stay locked), don't
1051 * clear any dirty bits and don't set any writeback bits
1053 * Do set the Private2 bit so we know this page was properly
1054 * setup for writepage
1056 page_ops = unlock ? PAGE_UNLOCK : 0;
1057 page_ops |= PAGE_SET_PRIVATE2;
1059 extent_clear_unlock_delalloc(inode, start,
1060 start + ram_size - 1,
1061 delalloc_end, locked_page,
1062 EXTENT_LOCKED | EXTENT_DELALLOC,
1064 if (num_bytes < cur_alloc_size)
1067 num_bytes -= cur_alloc_size;
1068 alloc_hint = ins.objectid + ins.offset;
1069 start += cur_alloc_size;
1070 extent_reserved = false;
1073 * btrfs_reloc_clone_csums() error, since start is increased
1074 * extent_clear_unlock_delalloc() at out_unlock label won't
1075 * free metadata of current ordered extent, we're OK to exit.
1083 out_drop_extent_cache:
1084 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1086 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1087 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1089 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1090 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1091 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1094 * If we reserved an extent for our delalloc range (or a subrange) and
1095 * failed to create the respective ordered extent, then it means that
1096 * when we reserved the extent we decremented the extent's size from
1097 * the data space_info's bytes_may_use counter and incremented the
1098 * space_info's bytes_reserved counter by the same amount. We must make
1099 * sure extent_clear_unlock_delalloc() does not try to decrement again
1100 * the data space_info's bytes_may_use counter, therefore we do not pass
1101 * it the flag EXTENT_CLEAR_DATA_RESV.
1103 if (extent_reserved) {
1104 extent_clear_unlock_delalloc(inode, start,
1105 start + cur_alloc_size,
1106 start + cur_alloc_size,
1110 start += cur_alloc_size;
1114 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1116 clear_bits | EXTENT_CLEAR_DATA_RESV,
1122 * work queue call back to started compression on a file and pages
1124 static noinline void async_cow_start(struct btrfs_work *work)
1126 struct async_cow *async_cow;
1128 async_cow = container_of(work, struct async_cow, work);
1130 compress_file_range(async_cow->inode, async_cow->locked_page,
1131 async_cow->start, async_cow->end, async_cow,
1133 if (num_added == 0) {
1134 btrfs_add_delayed_iput(async_cow->inode);
1135 async_cow->inode = NULL;
1140 * work queue call back to submit previously compressed pages
1142 static noinline void async_cow_submit(struct btrfs_work *work)
1144 struct btrfs_fs_info *fs_info;
1145 struct async_cow *async_cow;
1146 struct btrfs_root *root;
1147 unsigned long nr_pages;
1149 async_cow = container_of(work, struct async_cow, work);
1151 root = async_cow->root;
1152 fs_info = root->fs_info;
1153 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1156 /* atomic_sub_return implies a barrier */
1157 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1159 cond_wake_up_nomb(&fs_info->async_submit_wait);
1161 if (async_cow->inode)
1162 submit_compressed_extents(async_cow->inode, async_cow);
1165 static noinline void async_cow_free(struct btrfs_work *work)
1167 struct async_cow *async_cow;
1168 async_cow = container_of(work, struct async_cow, work);
1169 if (async_cow->inode)
1170 btrfs_add_delayed_iput(async_cow->inode);
1174 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1175 u64 start, u64 end, int *page_started,
1176 unsigned long *nr_written,
1177 unsigned int write_flags)
1179 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1180 struct async_cow *async_cow;
1181 struct btrfs_root *root = BTRFS_I(inode)->root;
1182 unsigned long nr_pages;
1185 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1187 while (start < end) {
1188 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1189 BUG_ON(!async_cow); /* -ENOMEM */
1190 async_cow->inode = igrab(inode);
1191 async_cow->root = root;
1192 async_cow->locked_page = locked_page;
1193 async_cow->start = start;
1194 async_cow->write_flags = write_flags;
1196 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1197 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1200 cur_end = min(end, start + SZ_512K - 1);
1202 async_cow->end = cur_end;
1203 INIT_LIST_HEAD(&async_cow->extents);
1205 btrfs_init_work(&async_cow->work,
1206 btrfs_delalloc_helper,
1207 async_cow_start, async_cow_submit,
1210 nr_pages = (cur_end - start + PAGE_SIZE) >>
1212 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1214 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1216 *nr_written += nr_pages;
1217 start = cur_end + 1;
1223 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1224 u64 bytenr, u64 num_bytes)
1227 struct btrfs_ordered_sum *sums;
1230 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1231 bytenr + num_bytes - 1, &list, 0);
1232 if (ret == 0 && list_empty(&list))
1235 while (!list_empty(&list)) {
1236 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1237 list_del(&sums->list);
1246 * when nowcow writeback call back. This checks for snapshots or COW copies
1247 * of the extents that exist in the file, and COWs the file as required.
1249 * If no cow copies or snapshots exist, we write directly to the existing
1252 static noinline int run_delalloc_nocow(struct inode *inode,
1253 struct page *locked_page,
1254 u64 start, u64 end, int *page_started, int force,
1255 unsigned long *nr_written)
1257 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1258 struct btrfs_root *root = BTRFS_I(inode)->root;
1259 struct extent_buffer *leaf;
1260 struct btrfs_path *path;
1261 struct btrfs_file_extent_item *fi;
1262 struct btrfs_key found_key;
1263 struct extent_map *em;
1278 u64 ino = btrfs_ino(BTRFS_I(inode));
1280 path = btrfs_alloc_path();
1282 extent_clear_unlock_delalloc(inode, start, end, end,
1284 EXTENT_LOCKED | EXTENT_DELALLOC |
1285 EXTENT_DO_ACCOUNTING |
1286 EXTENT_DEFRAG, PAGE_UNLOCK |
1288 PAGE_SET_WRITEBACK |
1289 PAGE_END_WRITEBACK);
1293 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1295 cow_start = (u64)-1;
1298 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1302 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1303 leaf = path->nodes[0];
1304 btrfs_item_key_to_cpu(leaf, &found_key,
1305 path->slots[0] - 1);
1306 if (found_key.objectid == ino &&
1307 found_key.type == BTRFS_EXTENT_DATA_KEY)
1312 leaf = path->nodes[0];
1313 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1314 ret = btrfs_next_leaf(root, path);
1316 if (cow_start != (u64)-1)
1317 cur_offset = cow_start;
1322 leaf = path->nodes[0];
1328 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1330 if (found_key.objectid > ino)
1332 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1333 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1337 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1338 found_key.offset > end)
1341 if (found_key.offset > cur_offset) {
1342 extent_end = found_key.offset;
1347 fi = btrfs_item_ptr(leaf, path->slots[0],
1348 struct btrfs_file_extent_item);
1349 extent_type = btrfs_file_extent_type(leaf, fi);
1351 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1352 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1353 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1354 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1355 extent_offset = btrfs_file_extent_offset(leaf, fi);
1356 extent_end = found_key.offset +
1357 btrfs_file_extent_num_bytes(leaf, fi);
1359 btrfs_file_extent_disk_num_bytes(leaf, fi);
1360 if (extent_end <= start) {
1364 if (disk_bytenr == 0)
1366 if (btrfs_file_extent_compression(leaf, fi) ||
1367 btrfs_file_extent_encryption(leaf, fi) ||
1368 btrfs_file_extent_other_encoding(leaf, fi))
1371 * Do the same check as in btrfs_cross_ref_exist but
1372 * without the unnecessary search.
1374 if (btrfs_file_extent_generation(leaf, fi) <=
1375 btrfs_root_last_snapshot(&root->root_item))
1377 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1379 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1381 ret = btrfs_cross_ref_exist(root, ino,
1383 extent_offset, disk_bytenr);
1386 * ret could be -EIO if the above fails to read
1390 if (cow_start != (u64)-1)
1391 cur_offset = cow_start;
1395 WARN_ON_ONCE(nolock);
1398 disk_bytenr += extent_offset;
1399 disk_bytenr += cur_offset - found_key.offset;
1400 num_bytes = min(end + 1, extent_end) - cur_offset;
1402 * if there are pending snapshots for this root,
1403 * we fall into common COW way.
1405 if (!nolock && atomic_read(&root->snapshot_force_cow))
1408 * force cow if csum exists in the range.
1409 * this ensure that csum for a given extent are
1410 * either valid or do not exist.
1412 ret = csum_exist_in_range(fs_info, disk_bytenr,
1416 * ret could be -EIO if the above fails to read
1420 if (cow_start != (u64)-1)
1421 cur_offset = cow_start;
1424 WARN_ON_ONCE(nolock);
1427 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr))
1430 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1431 extent_end = found_key.offset +
1432 btrfs_file_extent_ram_bytes(leaf, fi);
1433 extent_end = ALIGN(extent_end,
1434 fs_info->sectorsize);
1439 if (extent_end <= start) {
1442 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1446 if (cow_start == (u64)-1)
1447 cow_start = cur_offset;
1448 cur_offset = extent_end;
1449 if (cur_offset > end)
1455 btrfs_release_path(path);
1456 if (cow_start != (u64)-1) {
1457 ret = cow_file_range(inode, locked_page,
1458 cow_start, found_key.offset - 1,
1459 end, page_started, nr_written, 1,
1463 btrfs_dec_nocow_writers(fs_info,
1467 cow_start = (u64)-1;
1470 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1471 u64 orig_start = found_key.offset - extent_offset;
1473 em = create_io_em(inode, cur_offset, num_bytes,
1475 disk_bytenr, /* block_start */
1476 num_bytes, /* block_len */
1477 disk_num_bytes, /* orig_block_len */
1478 ram_bytes, BTRFS_COMPRESS_NONE,
1479 BTRFS_ORDERED_PREALLOC);
1482 btrfs_dec_nocow_writers(fs_info,
1487 free_extent_map(em);
1490 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1491 type = BTRFS_ORDERED_PREALLOC;
1493 type = BTRFS_ORDERED_NOCOW;
1496 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1497 num_bytes, num_bytes, type);
1499 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1500 BUG_ON(ret); /* -ENOMEM */
1502 if (root->root_key.objectid ==
1503 BTRFS_DATA_RELOC_TREE_OBJECTID)
1505 * Error handled later, as we must prevent
1506 * extent_clear_unlock_delalloc() in error handler
1507 * from freeing metadata of created ordered extent.
1509 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1512 extent_clear_unlock_delalloc(inode, cur_offset,
1513 cur_offset + num_bytes - 1, end,
1514 locked_page, EXTENT_LOCKED |
1516 EXTENT_CLEAR_DATA_RESV,
1517 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1519 cur_offset = extent_end;
1522 * btrfs_reloc_clone_csums() error, now we're OK to call error
1523 * handler, as metadata for created ordered extent will only
1524 * be freed by btrfs_finish_ordered_io().
1528 if (cur_offset > end)
1531 btrfs_release_path(path);
1533 if (cur_offset <= end && cow_start == (u64)-1) {
1534 cow_start = cur_offset;
1538 if (cow_start != (u64)-1) {
1539 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1540 page_started, nr_written, 1, NULL);
1546 if (ret && cur_offset < end)
1547 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1548 locked_page, EXTENT_LOCKED |
1549 EXTENT_DELALLOC | EXTENT_DEFRAG |
1550 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1552 PAGE_SET_WRITEBACK |
1553 PAGE_END_WRITEBACK);
1554 btrfs_free_path(path);
1558 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1561 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1562 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1566 * @defrag_bytes is a hint value, no spinlock held here,
1567 * if is not zero, it means the file is defragging.
1568 * Force cow if given extent needs to be defragged.
1570 if (BTRFS_I(inode)->defrag_bytes &&
1571 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1572 EXTENT_DEFRAG, 0, NULL))
1579 * extent_io.c call back to do delayed allocation processing
1581 static int run_delalloc_range(void *private_data, struct page *locked_page,
1582 u64 start, u64 end, int *page_started,
1583 unsigned long *nr_written,
1584 struct writeback_control *wbc)
1586 struct inode *inode = private_data;
1588 int force_cow = need_force_cow(inode, start, end);
1589 unsigned int write_flags = wbc_to_write_flags(wbc);
1591 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1592 ret = run_delalloc_nocow(inode, locked_page, start, end,
1593 page_started, 1, nr_written);
1594 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1595 ret = run_delalloc_nocow(inode, locked_page, start, end,
1596 page_started, 0, nr_written);
1597 } else if (!inode_need_compress(inode, start, end)) {
1598 ret = cow_file_range(inode, locked_page, start, end, end,
1599 page_started, nr_written, 1, NULL);
1601 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1602 &BTRFS_I(inode)->runtime_flags);
1603 ret = cow_file_range_async(inode, locked_page, start, end,
1604 page_started, nr_written,
1608 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1612 static void btrfs_split_extent_hook(void *private_data,
1613 struct extent_state *orig, u64 split)
1615 struct inode *inode = private_data;
1618 /* not delalloc, ignore it */
1619 if (!(orig->state & EXTENT_DELALLOC))
1622 size = orig->end - orig->start + 1;
1623 if (size > BTRFS_MAX_EXTENT_SIZE) {
1628 * See the explanation in btrfs_merge_extent_hook, the same
1629 * applies here, just in reverse.
1631 new_size = orig->end - split + 1;
1632 num_extents = count_max_extents(new_size);
1633 new_size = split - orig->start;
1634 num_extents += count_max_extents(new_size);
1635 if (count_max_extents(size) >= num_extents)
1639 spin_lock(&BTRFS_I(inode)->lock);
1640 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1641 spin_unlock(&BTRFS_I(inode)->lock);
1645 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1646 * extents so we can keep track of new extents that are just merged onto old
1647 * extents, such as when we are doing sequential writes, so we can properly
1648 * account for the metadata space we'll need.
1650 static void btrfs_merge_extent_hook(void *private_data,
1651 struct extent_state *new,
1652 struct extent_state *other)
1654 struct inode *inode = private_data;
1655 u64 new_size, old_size;
1658 /* not delalloc, ignore it */
1659 if (!(other->state & EXTENT_DELALLOC))
1662 if (new->start > other->start)
1663 new_size = new->end - other->start + 1;
1665 new_size = other->end - new->start + 1;
1667 /* we're not bigger than the max, unreserve the space and go */
1668 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1669 spin_lock(&BTRFS_I(inode)->lock);
1670 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1671 spin_unlock(&BTRFS_I(inode)->lock);
1676 * We have to add up either side to figure out how many extents were
1677 * accounted for before we merged into one big extent. If the number of
1678 * extents we accounted for is <= the amount we need for the new range
1679 * then we can return, otherwise drop. Think of it like this
1683 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1684 * need 2 outstanding extents, on one side we have 1 and the other side
1685 * we have 1 so they are == and we can return. But in this case
1687 * [MAX_SIZE+4k][MAX_SIZE+4k]
1689 * Each range on their own accounts for 2 extents, but merged together
1690 * they are only 3 extents worth of accounting, so we need to drop in
1693 old_size = other->end - other->start + 1;
1694 num_extents = count_max_extents(old_size);
1695 old_size = new->end - new->start + 1;
1696 num_extents += count_max_extents(old_size);
1697 if (count_max_extents(new_size) >= num_extents)
1700 spin_lock(&BTRFS_I(inode)->lock);
1701 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1702 spin_unlock(&BTRFS_I(inode)->lock);
1705 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1706 struct inode *inode)
1708 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1710 spin_lock(&root->delalloc_lock);
1711 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1712 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1713 &root->delalloc_inodes);
1714 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1715 &BTRFS_I(inode)->runtime_flags);
1716 root->nr_delalloc_inodes++;
1717 if (root->nr_delalloc_inodes == 1) {
1718 spin_lock(&fs_info->delalloc_root_lock);
1719 BUG_ON(!list_empty(&root->delalloc_root));
1720 list_add_tail(&root->delalloc_root,
1721 &fs_info->delalloc_roots);
1722 spin_unlock(&fs_info->delalloc_root_lock);
1725 spin_unlock(&root->delalloc_lock);
1729 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1730 struct btrfs_inode *inode)
1732 struct btrfs_fs_info *fs_info = root->fs_info;
1734 if (!list_empty(&inode->delalloc_inodes)) {
1735 list_del_init(&inode->delalloc_inodes);
1736 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1737 &inode->runtime_flags);
1738 root->nr_delalloc_inodes--;
1739 if (!root->nr_delalloc_inodes) {
1740 ASSERT(list_empty(&root->delalloc_inodes));
1741 spin_lock(&fs_info->delalloc_root_lock);
1742 BUG_ON(list_empty(&root->delalloc_root));
1743 list_del_init(&root->delalloc_root);
1744 spin_unlock(&fs_info->delalloc_root_lock);
1749 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1750 struct btrfs_inode *inode)
1752 spin_lock(&root->delalloc_lock);
1753 __btrfs_del_delalloc_inode(root, inode);
1754 spin_unlock(&root->delalloc_lock);
1758 * extent_io.c set_bit_hook, used to track delayed allocation
1759 * bytes in this file, and to maintain the list of inodes that
1760 * have pending delalloc work to be done.
1762 static void btrfs_set_bit_hook(void *private_data,
1763 struct extent_state *state, unsigned *bits)
1765 struct inode *inode = private_data;
1767 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1769 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1772 * set_bit and clear bit hooks normally require _irqsave/restore
1773 * but in this case, we are only testing for the DELALLOC
1774 * bit, which is only set or cleared with irqs on
1776 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1777 struct btrfs_root *root = BTRFS_I(inode)->root;
1778 u64 len = state->end + 1 - state->start;
1779 u32 num_extents = count_max_extents(len);
1780 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1782 spin_lock(&BTRFS_I(inode)->lock);
1783 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1784 spin_unlock(&BTRFS_I(inode)->lock);
1786 /* For sanity tests */
1787 if (btrfs_is_testing(fs_info))
1790 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1791 fs_info->delalloc_batch);
1792 spin_lock(&BTRFS_I(inode)->lock);
1793 BTRFS_I(inode)->delalloc_bytes += len;
1794 if (*bits & EXTENT_DEFRAG)
1795 BTRFS_I(inode)->defrag_bytes += len;
1796 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1797 &BTRFS_I(inode)->runtime_flags))
1798 btrfs_add_delalloc_inodes(root, inode);
1799 spin_unlock(&BTRFS_I(inode)->lock);
1802 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1803 (*bits & EXTENT_DELALLOC_NEW)) {
1804 spin_lock(&BTRFS_I(inode)->lock);
1805 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1807 spin_unlock(&BTRFS_I(inode)->lock);
1812 * extent_io.c clear_bit_hook, see set_bit_hook for why
1814 static void btrfs_clear_bit_hook(void *private_data,
1815 struct extent_state *state,
1818 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1819 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1820 u64 len = state->end + 1 - state->start;
1821 u32 num_extents = count_max_extents(len);
1823 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1824 spin_lock(&inode->lock);
1825 inode->defrag_bytes -= len;
1826 spin_unlock(&inode->lock);
1830 * set_bit and clear bit hooks normally require _irqsave/restore
1831 * but in this case, we are only testing for the DELALLOC
1832 * bit, which is only set or cleared with irqs on
1834 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1835 struct btrfs_root *root = inode->root;
1836 bool do_list = !btrfs_is_free_space_inode(inode);
1838 spin_lock(&inode->lock);
1839 btrfs_mod_outstanding_extents(inode, -num_extents);
1840 spin_unlock(&inode->lock);
1843 * We don't reserve metadata space for space cache inodes so we
1844 * don't need to call dellalloc_release_metadata if there is an
1847 if (*bits & EXTENT_CLEAR_META_RESV &&
1848 root != fs_info->tree_root)
1849 btrfs_delalloc_release_metadata(inode, len, false);
1851 /* For sanity tests. */
1852 if (btrfs_is_testing(fs_info))
1855 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1856 do_list && !(state->state & EXTENT_NORESERVE) &&
1857 (*bits & EXTENT_CLEAR_DATA_RESV))
1858 btrfs_free_reserved_data_space_noquota(
1862 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1863 fs_info->delalloc_batch);
1864 spin_lock(&inode->lock);
1865 inode->delalloc_bytes -= len;
1866 if (do_list && inode->delalloc_bytes == 0 &&
1867 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1868 &inode->runtime_flags))
1869 btrfs_del_delalloc_inode(root, inode);
1870 spin_unlock(&inode->lock);
1873 if ((state->state & EXTENT_DELALLOC_NEW) &&
1874 (*bits & EXTENT_DELALLOC_NEW)) {
1875 spin_lock(&inode->lock);
1876 ASSERT(inode->new_delalloc_bytes >= len);
1877 inode->new_delalloc_bytes -= len;
1878 spin_unlock(&inode->lock);
1883 * Merge bio hook, this must check the chunk tree to make sure we don't create
1884 * bios that span stripes or chunks
1886 * return 1 if page cannot be merged to bio
1887 * return 0 if page can be merged to bio
1888 * return error otherwise
1890 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1891 size_t size, struct bio *bio,
1892 unsigned long bio_flags)
1894 struct inode *inode = page->mapping->host;
1895 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1896 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1901 if (bio_flags & EXTENT_BIO_COMPRESSED)
1904 length = bio->bi_iter.bi_size;
1905 map_length = length;
1906 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1910 if (map_length < length + size)
1916 * in order to insert checksums into the metadata in large chunks,
1917 * we wait until bio submission time. All the pages in the bio are
1918 * checksummed and sums are attached onto the ordered extent record.
1920 * At IO completion time the cums attached on the ordered extent record
1921 * are inserted into the btree
1923 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1926 struct inode *inode = private_data;
1927 blk_status_t ret = 0;
1929 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1930 BUG_ON(ret); /* -ENOMEM */
1935 * in order to insert checksums into the metadata in large chunks,
1936 * we wait until bio submission time. All the pages in the bio are
1937 * checksummed and sums are attached onto the ordered extent record.
1939 * At IO completion time the cums attached on the ordered extent record
1940 * are inserted into the btree
1942 blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1945 struct inode *inode = private_data;
1946 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1949 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1951 bio->bi_status = ret;
1958 * extent_io.c submission hook. This does the right thing for csum calculation
1959 * on write, or reading the csums from the tree before a read.
1961 * Rules about async/sync submit,
1962 * a) read: sync submit
1964 * b) write without checksum: sync submit
1966 * c) write with checksum:
1967 * c-1) if bio is issued by fsync: sync submit
1968 * (sync_writers != 0)
1970 * c-2) if root is reloc root: sync submit
1971 * (only in case of buffered IO)
1973 * c-3) otherwise: async submit
1975 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1976 int mirror_num, unsigned long bio_flags,
1979 struct inode *inode = private_data;
1980 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1981 struct btrfs_root *root = BTRFS_I(inode)->root;
1982 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
1983 blk_status_t ret = 0;
1985 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1987 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1989 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
1990 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
1992 if (bio_op(bio) != REQ_OP_WRITE) {
1993 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
1997 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1998 ret = btrfs_submit_compressed_read(inode, bio,
2002 } else if (!skip_sum) {
2003 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2008 } else if (async && !skip_sum) {
2009 /* csum items have already been cloned */
2010 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2012 /* we're doing a write, do the async checksumming */
2013 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2015 btrfs_submit_bio_start);
2017 } else if (!skip_sum) {
2018 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2024 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2028 bio->bi_status = ret;
2035 * given a list of ordered sums record them in the inode. This happens
2036 * at IO completion time based on sums calculated at bio submission time.
2038 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2039 struct inode *inode, struct list_head *list)
2041 struct btrfs_ordered_sum *sum;
2044 list_for_each_entry(sum, list, list) {
2045 trans->adding_csums = true;
2046 ret = btrfs_csum_file_blocks(trans,
2047 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2048 trans->adding_csums = false;
2055 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2056 unsigned int extra_bits,
2057 struct extent_state **cached_state, int dedupe)
2059 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2060 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2061 extra_bits, cached_state);
2064 /* see btrfs_writepage_start_hook for details on why this is required */
2065 struct btrfs_writepage_fixup {
2067 struct btrfs_work work;
2070 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2072 struct btrfs_writepage_fixup *fixup;
2073 struct btrfs_ordered_extent *ordered;
2074 struct extent_state *cached_state = NULL;
2075 struct extent_changeset *data_reserved = NULL;
2077 struct inode *inode;
2082 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2086 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2087 ClearPageChecked(page);
2091 inode = page->mapping->host;
2092 page_start = page_offset(page);
2093 page_end = page_offset(page) + PAGE_SIZE - 1;
2095 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2098 /* already ordered? We're done */
2099 if (PagePrivate2(page))
2102 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2105 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2106 page_end, &cached_state);
2108 btrfs_start_ordered_extent(inode, ordered, 1);
2109 btrfs_put_ordered_extent(ordered);
2113 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2116 mapping_set_error(page->mapping, ret);
2117 end_extent_writepage(page, ret, page_start, page_end);
2118 ClearPageChecked(page);
2122 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2125 mapping_set_error(page->mapping, ret);
2126 end_extent_writepage(page, ret, page_start, page_end);
2127 ClearPageChecked(page);
2131 ClearPageChecked(page);
2132 set_page_dirty(page);
2133 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2135 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2141 extent_changeset_free(data_reserved);
2145 * There are a few paths in the higher layers of the kernel that directly
2146 * set the page dirty bit without asking the filesystem if it is a
2147 * good idea. This causes problems because we want to make sure COW
2148 * properly happens and the data=ordered rules are followed.
2150 * In our case any range that doesn't have the ORDERED bit set
2151 * hasn't been properly setup for IO. We kick off an async process
2152 * to fix it up. The async helper will wait for ordered extents, set
2153 * the delalloc bit and make it safe to write the page.
2155 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2157 struct inode *inode = page->mapping->host;
2158 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2159 struct btrfs_writepage_fixup *fixup;
2161 /* this page is properly in the ordered list */
2162 if (TestClearPagePrivate2(page))
2165 if (PageChecked(page))
2168 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2172 SetPageChecked(page);
2174 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2175 btrfs_writepage_fixup_worker, NULL, NULL);
2177 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2181 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2182 struct inode *inode, u64 file_pos,
2183 u64 disk_bytenr, u64 disk_num_bytes,
2184 u64 num_bytes, u64 ram_bytes,
2185 u8 compression, u8 encryption,
2186 u16 other_encoding, int extent_type)
2188 struct btrfs_root *root = BTRFS_I(inode)->root;
2189 struct btrfs_file_extent_item *fi;
2190 struct btrfs_path *path;
2191 struct extent_buffer *leaf;
2192 struct btrfs_key ins;
2194 int extent_inserted = 0;
2197 path = btrfs_alloc_path();
2202 * we may be replacing one extent in the tree with another.
2203 * The new extent is pinned in the extent map, and we don't want
2204 * to drop it from the cache until it is completely in the btree.
2206 * So, tell btrfs_drop_extents to leave this extent in the cache.
2207 * the caller is expected to unpin it and allow it to be merged
2210 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2211 file_pos + num_bytes, NULL, 0,
2212 1, sizeof(*fi), &extent_inserted);
2216 if (!extent_inserted) {
2217 ins.objectid = btrfs_ino(BTRFS_I(inode));
2218 ins.offset = file_pos;
2219 ins.type = BTRFS_EXTENT_DATA_KEY;
2221 path->leave_spinning = 1;
2222 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2227 leaf = path->nodes[0];
2228 fi = btrfs_item_ptr(leaf, path->slots[0],
2229 struct btrfs_file_extent_item);
2230 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2231 btrfs_set_file_extent_type(leaf, fi, extent_type);
2232 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2233 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2234 btrfs_set_file_extent_offset(leaf, fi, 0);
2235 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2236 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2237 btrfs_set_file_extent_compression(leaf, fi, compression);
2238 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2239 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2241 btrfs_mark_buffer_dirty(leaf);
2242 btrfs_release_path(path);
2244 inode_add_bytes(inode, num_bytes);
2246 ins.objectid = disk_bytenr;
2247 ins.offset = disk_num_bytes;
2248 ins.type = BTRFS_EXTENT_ITEM_KEY;
2251 * Release the reserved range from inode dirty range map, as it is
2252 * already moved into delayed_ref_head
2254 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2258 ret = btrfs_alloc_reserved_file_extent(trans, root,
2259 btrfs_ino(BTRFS_I(inode)),
2260 file_pos, qg_released, &ins);
2262 btrfs_free_path(path);
2267 /* snapshot-aware defrag */
2268 struct sa_defrag_extent_backref {
2269 struct rb_node node;
2270 struct old_sa_defrag_extent *old;
2279 struct old_sa_defrag_extent {
2280 struct list_head list;
2281 struct new_sa_defrag_extent *new;
2290 struct new_sa_defrag_extent {
2291 struct rb_root root;
2292 struct list_head head;
2293 struct btrfs_path *path;
2294 struct inode *inode;
2302 static int backref_comp(struct sa_defrag_extent_backref *b1,
2303 struct sa_defrag_extent_backref *b2)
2305 if (b1->root_id < b2->root_id)
2307 else if (b1->root_id > b2->root_id)
2310 if (b1->inum < b2->inum)
2312 else if (b1->inum > b2->inum)
2315 if (b1->file_pos < b2->file_pos)
2317 else if (b1->file_pos > b2->file_pos)
2321 * [------------------------------] ===> (a range of space)
2322 * |<--->| |<---->| =============> (fs/file tree A)
2323 * |<---------------------------->| ===> (fs/file tree B)
2325 * A range of space can refer to two file extents in one tree while
2326 * refer to only one file extent in another tree.
2328 * So we may process a disk offset more than one time(two extents in A)
2329 * and locate at the same extent(one extent in B), then insert two same
2330 * backrefs(both refer to the extent in B).
2335 static void backref_insert(struct rb_root *root,
2336 struct sa_defrag_extent_backref *backref)
2338 struct rb_node **p = &root->rb_node;
2339 struct rb_node *parent = NULL;
2340 struct sa_defrag_extent_backref *entry;
2345 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2347 ret = backref_comp(backref, entry);
2351 p = &(*p)->rb_right;
2354 rb_link_node(&backref->node, parent, p);
2355 rb_insert_color(&backref->node, root);
2359 * Note the backref might has changed, and in this case we just return 0.
2361 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2364 struct btrfs_file_extent_item *extent;
2365 struct old_sa_defrag_extent *old = ctx;
2366 struct new_sa_defrag_extent *new = old->new;
2367 struct btrfs_path *path = new->path;
2368 struct btrfs_key key;
2369 struct btrfs_root *root;
2370 struct sa_defrag_extent_backref *backref;
2371 struct extent_buffer *leaf;
2372 struct inode *inode = new->inode;
2373 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2379 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2380 inum == btrfs_ino(BTRFS_I(inode)))
2383 key.objectid = root_id;
2384 key.type = BTRFS_ROOT_ITEM_KEY;
2385 key.offset = (u64)-1;
2387 root = btrfs_read_fs_root_no_name(fs_info, &key);
2389 if (PTR_ERR(root) == -ENOENT)
2392 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2393 inum, offset, root_id);
2394 return PTR_ERR(root);
2397 key.objectid = inum;
2398 key.type = BTRFS_EXTENT_DATA_KEY;
2399 if (offset > (u64)-1 << 32)
2402 key.offset = offset;
2404 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2405 if (WARN_ON(ret < 0))
2412 leaf = path->nodes[0];
2413 slot = path->slots[0];
2415 if (slot >= btrfs_header_nritems(leaf)) {
2416 ret = btrfs_next_leaf(root, path);
2419 } else if (ret > 0) {
2428 btrfs_item_key_to_cpu(leaf, &key, slot);
2430 if (key.objectid > inum)
2433 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2436 extent = btrfs_item_ptr(leaf, slot,
2437 struct btrfs_file_extent_item);
2439 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2443 * 'offset' refers to the exact key.offset,
2444 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2445 * (key.offset - extent_offset).
2447 if (key.offset != offset)
2450 extent_offset = btrfs_file_extent_offset(leaf, extent);
2451 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2453 if (extent_offset >= old->extent_offset + old->offset +
2454 old->len || extent_offset + num_bytes <=
2455 old->extent_offset + old->offset)
2460 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2466 backref->root_id = root_id;
2467 backref->inum = inum;
2468 backref->file_pos = offset;
2469 backref->num_bytes = num_bytes;
2470 backref->extent_offset = extent_offset;
2471 backref->generation = btrfs_file_extent_generation(leaf, extent);
2473 backref_insert(&new->root, backref);
2476 btrfs_release_path(path);
2481 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2482 struct new_sa_defrag_extent *new)
2484 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2485 struct old_sa_defrag_extent *old, *tmp;
2490 list_for_each_entry_safe(old, tmp, &new->head, list) {
2491 ret = iterate_inodes_from_logical(old->bytenr +
2492 old->extent_offset, fs_info,
2493 path, record_one_backref,
2495 if (ret < 0 && ret != -ENOENT)
2498 /* no backref to be processed for this extent */
2500 list_del(&old->list);
2505 if (list_empty(&new->head))
2511 static int relink_is_mergable(struct extent_buffer *leaf,
2512 struct btrfs_file_extent_item *fi,
2513 struct new_sa_defrag_extent *new)
2515 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2518 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2521 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2524 if (btrfs_file_extent_encryption(leaf, fi) ||
2525 btrfs_file_extent_other_encoding(leaf, fi))
2532 * Note the backref might has changed, and in this case we just return 0.
2534 static noinline int relink_extent_backref(struct btrfs_path *path,
2535 struct sa_defrag_extent_backref *prev,
2536 struct sa_defrag_extent_backref *backref)
2538 struct btrfs_file_extent_item *extent;
2539 struct btrfs_file_extent_item *item;
2540 struct btrfs_ordered_extent *ordered;
2541 struct btrfs_trans_handle *trans;
2542 struct btrfs_root *root;
2543 struct btrfs_key key;
2544 struct extent_buffer *leaf;
2545 struct old_sa_defrag_extent *old = backref->old;
2546 struct new_sa_defrag_extent *new = old->new;
2547 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2548 struct inode *inode;
2549 struct extent_state *cached = NULL;
2558 if (prev && prev->root_id == backref->root_id &&
2559 prev->inum == backref->inum &&
2560 prev->file_pos + prev->num_bytes == backref->file_pos)
2563 /* step 1: get root */
2564 key.objectid = backref->root_id;
2565 key.type = BTRFS_ROOT_ITEM_KEY;
2566 key.offset = (u64)-1;
2568 index = srcu_read_lock(&fs_info->subvol_srcu);
2570 root = btrfs_read_fs_root_no_name(fs_info, &key);
2572 srcu_read_unlock(&fs_info->subvol_srcu, index);
2573 if (PTR_ERR(root) == -ENOENT)
2575 return PTR_ERR(root);
2578 if (btrfs_root_readonly(root)) {
2579 srcu_read_unlock(&fs_info->subvol_srcu, index);
2583 /* step 2: get inode */
2584 key.objectid = backref->inum;
2585 key.type = BTRFS_INODE_ITEM_KEY;
2588 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2589 if (IS_ERR(inode)) {
2590 srcu_read_unlock(&fs_info->subvol_srcu, index);
2594 srcu_read_unlock(&fs_info->subvol_srcu, index);
2596 /* step 3: relink backref */
2597 lock_start = backref->file_pos;
2598 lock_end = backref->file_pos + backref->num_bytes - 1;
2599 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2602 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2604 btrfs_put_ordered_extent(ordered);
2608 trans = btrfs_join_transaction(root);
2609 if (IS_ERR(trans)) {
2610 ret = PTR_ERR(trans);
2614 key.objectid = backref->inum;
2615 key.type = BTRFS_EXTENT_DATA_KEY;
2616 key.offset = backref->file_pos;
2618 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2621 } else if (ret > 0) {
2626 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2627 struct btrfs_file_extent_item);
2629 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2630 backref->generation)
2633 btrfs_release_path(path);
2635 start = backref->file_pos;
2636 if (backref->extent_offset < old->extent_offset + old->offset)
2637 start += old->extent_offset + old->offset -
2638 backref->extent_offset;
2640 len = min(backref->extent_offset + backref->num_bytes,
2641 old->extent_offset + old->offset + old->len);
2642 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2644 ret = btrfs_drop_extents(trans, root, inode, start,
2649 key.objectid = btrfs_ino(BTRFS_I(inode));
2650 key.type = BTRFS_EXTENT_DATA_KEY;
2653 path->leave_spinning = 1;
2655 struct btrfs_file_extent_item *fi;
2657 struct btrfs_key found_key;
2659 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2664 leaf = path->nodes[0];
2665 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2667 fi = btrfs_item_ptr(leaf, path->slots[0],
2668 struct btrfs_file_extent_item);
2669 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2671 if (extent_len + found_key.offset == start &&
2672 relink_is_mergable(leaf, fi, new)) {
2673 btrfs_set_file_extent_num_bytes(leaf, fi,
2675 btrfs_mark_buffer_dirty(leaf);
2676 inode_add_bytes(inode, len);
2682 btrfs_release_path(path);
2687 ret = btrfs_insert_empty_item(trans, root, path, &key,
2690 btrfs_abort_transaction(trans, ret);
2694 leaf = path->nodes[0];
2695 item = btrfs_item_ptr(leaf, path->slots[0],
2696 struct btrfs_file_extent_item);
2697 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2698 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2699 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2700 btrfs_set_file_extent_num_bytes(leaf, item, len);
2701 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2702 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2703 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2704 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2705 btrfs_set_file_extent_encryption(leaf, item, 0);
2706 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2708 btrfs_mark_buffer_dirty(leaf);
2709 inode_add_bytes(inode, len);
2710 btrfs_release_path(path);
2712 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2714 backref->root_id, backref->inum,
2715 new->file_pos); /* start - extent_offset */
2717 btrfs_abort_transaction(trans, ret);
2723 btrfs_release_path(path);
2724 path->leave_spinning = 0;
2725 btrfs_end_transaction(trans);
2727 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2733 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2735 struct old_sa_defrag_extent *old, *tmp;
2740 list_for_each_entry_safe(old, tmp, &new->head, list) {
2746 static void relink_file_extents(struct new_sa_defrag_extent *new)
2748 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2749 struct btrfs_path *path;
2750 struct sa_defrag_extent_backref *backref;
2751 struct sa_defrag_extent_backref *prev = NULL;
2752 struct rb_node *node;
2755 path = btrfs_alloc_path();
2759 if (!record_extent_backrefs(path, new)) {
2760 btrfs_free_path(path);
2763 btrfs_release_path(path);
2766 node = rb_first(&new->root);
2769 rb_erase(node, &new->root);
2771 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2773 ret = relink_extent_backref(path, prev, backref);
2786 btrfs_free_path(path);
2788 free_sa_defrag_extent(new);
2790 atomic_dec(&fs_info->defrag_running);
2791 wake_up(&fs_info->transaction_wait);
2794 static struct new_sa_defrag_extent *
2795 record_old_file_extents(struct inode *inode,
2796 struct btrfs_ordered_extent *ordered)
2798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2799 struct btrfs_root *root = BTRFS_I(inode)->root;
2800 struct btrfs_path *path;
2801 struct btrfs_key key;
2802 struct old_sa_defrag_extent *old;
2803 struct new_sa_defrag_extent *new;
2806 new = kmalloc(sizeof(*new), GFP_NOFS);
2811 new->file_pos = ordered->file_offset;
2812 new->len = ordered->len;
2813 new->bytenr = ordered->start;
2814 new->disk_len = ordered->disk_len;
2815 new->compress_type = ordered->compress_type;
2816 new->root = RB_ROOT;
2817 INIT_LIST_HEAD(&new->head);
2819 path = btrfs_alloc_path();
2823 key.objectid = btrfs_ino(BTRFS_I(inode));
2824 key.type = BTRFS_EXTENT_DATA_KEY;
2825 key.offset = new->file_pos;
2827 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2830 if (ret > 0 && path->slots[0] > 0)
2833 /* find out all the old extents for the file range */
2835 struct btrfs_file_extent_item *extent;
2836 struct extent_buffer *l;
2845 slot = path->slots[0];
2847 if (slot >= btrfs_header_nritems(l)) {
2848 ret = btrfs_next_leaf(root, path);
2856 btrfs_item_key_to_cpu(l, &key, slot);
2858 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2860 if (key.type != BTRFS_EXTENT_DATA_KEY)
2862 if (key.offset >= new->file_pos + new->len)
2865 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2867 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2868 if (key.offset + num_bytes < new->file_pos)
2871 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2875 extent_offset = btrfs_file_extent_offset(l, extent);
2877 old = kmalloc(sizeof(*old), GFP_NOFS);
2881 offset = max(new->file_pos, key.offset);
2882 end = min(new->file_pos + new->len, key.offset + num_bytes);
2884 old->bytenr = disk_bytenr;
2885 old->extent_offset = extent_offset;
2886 old->offset = offset - key.offset;
2887 old->len = end - offset;
2890 list_add_tail(&old->list, &new->head);
2896 btrfs_free_path(path);
2897 atomic_inc(&fs_info->defrag_running);
2902 btrfs_free_path(path);
2904 free_sa_defrag_extent(new);
2908 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2911 struct btrfs_block_group_cache *cache;
2913 cache = btrfs_lookup_block_group(fs_info, start);
2916 spin_lock(&cache->lock);
2917 cache->delalloc_bytes -= len;
2918 spin_unlock(&cache->lock);
2920 btrfs_put_block_group(cache);
2923 /* as ordered data IO finishes, this gets called so we can finish
2924 * an ordered extent if the range of bytes in the file it covers are
2927 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2929 struct inode *inode = ordered_extent->inode;
2930 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2931 struct btrfs_root *root = BTRFS_I(inode)->root;
2932 struct btrfs_trans_handle *trans = NULL;
2933 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2934 struct extent_state *cached_state = NULL;
2935 struct new_sa_defrag_extent *new = NULL;
2936 int compress_type = 0;
2938 u64 logical_len = ordered_extent->len;
2940 bool truncated = false;
2941 bool range_locked = false;
2942 bool clear_new_delalloc_bytes = false;
2944 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2945 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2946 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2947 clear_new_delalloc_bytes = true;
2949 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2951 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2956 btrfs_free_io_failure_record(BTRFS_I(inode),
2957 ordered_extent->file_offset,
2958 ordered_extent->file_offset +
2959 ordered_extent->len - 1);
2961 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2963 logical_len = ordered_extent->truncated_len;
2964 /* Truncated the entire extent, don't bother adding */
2969 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2970 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2973 * For mwrite(mmap + memset to write) case, we still reserve
2974 * space for NOCOW range.
2975 * As NOCOW won't cause a new delayed ref, just free the space
2977 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
2978 ordered_extent->len);
2979 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2981 trans = btrfs_join_transaction_nolock(root);
2983 trans = btrfs_join_transaction(root);
2984 if (IS_ERR(trans)) {
2985 ret = PTR_ERR(trans);
2989 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
2990 ret = btrfs_update_inode_fallback(trans, root, inode);
2991 if (ret) /* -ENOMEM or corruption */
2992 btrfs_abort_transaction(trans, ret);
2996 range_locked = true;
2997 lock_extent_bits(io_tree, ordered_extent->file_offset,
2998 ordered_extent->file_offset + ordered_extent->len - 1,
3001 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3002 ordered_extent->file_offset + ordered_extent->len - 1,
3003 EXTENT_DEFRAG, 0, cached_state);
3005 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3006 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3007 /* the inode is shared */
3008 new = record_old_file_extents(inode, ordered_extent);
3010 clear_extent_bit(io_tree, ordered_extent->file_offset,
3011 ordered_extent->file_offset + ordered_extent->len - 1,
3012 EXTENT_DEFRAG, 0, 0, &cached_state);
3016 trans = btrfs_join_transaction_nolock(root);
3018 trans = btrfs_join_transaction(root);
3019 if (IS_ERR(trans)) {
3020 ret = PTR_ERR(trans);
3025 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3027 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3028 compress_type = ordered_extent->compress_type;
3029 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3030 BUG_ON(compress_type);
3031 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3032 ordered_extent->len);
3033 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3034 ordered_extent->file_offset,
3035 ordered_extent->file_offset +
3038 BUG_ON(root == fs_info->tree_root);
3039 ret = insert_reserved_file_extent(trans, inode,
3040 ordered_extent->file_offset,
3041 ordered_extent->start,
3042 ordered_extent->disk_len,
3043 logical_len, logical_len,
3044 compress_type, 0, 0,
3045 BTRFS_FILE_EXTENT_REG);
3047 btrfs_release_delalloc_bytes(fs_info,
3048 ordered_extent->start,
3049 ordered_extent->disk_len);
3051 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3052 ordered_extent->file_offset, ordered_extent->len,
3055 btrfs_abort_transaction(trans, ret);
3059 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3061 btrfs_abort_transaction(trans, ret);
3065 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3066 ret = btrfs_update_inode_fallback(trans, root, inode);
3067 if (ret) { /* -ENOMEM or corruption */
3068 btrfs_abort_transaction(trans, ret);
3073 if (range_locked || clear_new_delalloc_bytes) {
3074 unsigned int clear_bits = 0;
3077 clear_bits |= EXTENT_LOCKED;
3078 if (clear_new_delalloc_bytes)
3079 clear_bits |= EXTENT_DELALLOC_NEW;
3080 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3081 ordered_extent->file_offset,
3082 ordered_extent->file_offset +
3083 ordered_extent->len - 1,
3085 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3090 btrfs_end_transaction(trans);
3092 if (ret || truncated) {
3096 start = ordered_extent->file_offset + logical_len;
3098 start = ordered_extent->file_offset;
3099 end = ordered_extent->file_offset + ordered_extent->len - 1;
3100 clear_extent_uptodate(io_tree, start, end, NULL);
3102 /* Drop the cache for the part of the extent we didn't write. */
3103 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3106 * If the ordered extent had an IOERR or something else went
3107 * wrong we need to return the space for this ordered extent
3108 * back to the allocator. We only free the extent in the
3109 * truncated case if we didn't write out the extent at all.
3111 if ((ret || !logical_len) &&
3112 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3113 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3114 btrfs_free_reserved_extent(fs_info,
3115 ordered_extent->start,
3116 ordered_extent->disk_len, 1);
3121 * This needs to be done to make sure anybody waiting knows we are done
3122 * updating everything for this ordered extent.
3124 btrfs_remove_ordered_extent(inode, ordered_extent);
3126 /* for snapshot-aware defrag */
3129 free_sa_defrag_extent(new);
3130 atomic_dec(&fs_info->defrag_running);
3132 relink_file_extents(new);
3137 btrfs_put_ordered_extent(ordered_extent);
3138 /* once for the tree */
3139 btrfs_put_ordered_extent(ordered_extent);
3141 /* Try to release some metadata so we don't get an OOM but don't wait */
3142 btrfs_btree_balance_dirty_nodelay(fs_info);
3147 static void finish_ordered_fn(struct btrfs_work *work)
3149 struct btrfs_ordered_extent *ordered_extent;
3150 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3151 btrfs_finish_ordered_io(ordered_extent);
3154 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3155 struct extent_state *state, int uptodate)
3157 struct inode *inode = page->mapping->host;
3158 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3159 struct btrfs_ordered_extent *ordered_extent = NULL;
3160 struct btrfs_workqueue *wq;
3161 btrfs_work_func_t func;
3163 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3165 ClearPagePrivate2(page);
3166 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3167 end - start + 1, uptodate))
3170 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3171 wq = fs_info->endio_freespace_worker;
3172 func = btrfs_freespace_write_helper;
3174 wq = fs_info->endio_write_workers;
3175 func = btrfs_endio_write_helper;
3178 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3180 btrfs_queue_work(wq, &ordered_extent->work);
3183 static int __readpage_endio_check(struct inode *inode,
3184 struct btrfs_io_bio *io_bio,
3185 int icsum, struct page *page,
3186 int pgoff, u64 start, size_t len)
3192 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3194 kaddr = kmap_atomic(page);
3195 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3196 btrfs_csum_final(csum, (u8 *)&csum);
3197 if (csum != csum_expected)
3200 kunmap_atomic(kaddr);
3203 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3204 io_bio->mirror_num);
3205 memset(kaddr + pgoff, 1, len);
3206 flush_dcache_page(page);
3207 kunmap_atomic(kaddr);
3212 * when reads are done, we need to check csums to verify the data is correct
3213 * if there's a match, we allow the bio to finish. If not, the code in
3214 * extent_io.c will try to find good copies for us.
3216 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3217 u64 phy_offset, struct page *page,
3218 u64 start, u64 end, int mirror)
3220 size_t offset = start - page_offset(page);
3221 struct inode *inode = page->mapping->host;
3222 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3223 struct btrfs_root *root = BTRFS_I(inode)->root;
3225 if (PageChecked(page)) {
3226 ClearPageChecked(page);
3230 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3233 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3234 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3235 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3239 phy_offset >>= inode->i_sb->s_blocksize_bits;
3240 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3241 start, (size_t)(end - start + 1));
3245 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3247 * @inode: The inode we want to perform iput on
3249 * This function uses the generic vfs_inode::i_count to track whether we should
3250 * just decrement it (in case it's > 1) or if this is the last iput then link
3251 * the inode to the delayed iput machinery. Delayed iputs are processed at
3252 * transaction commit time/superblock commit/cleaner kthread.
3254 void btrfs_add_delayed_iput(struct inode *inode)
3256 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3257 struct btrfs_inode *binode = BTRFS_I(inode);
3259 if (atomic_add_unless(&inode->i_count, -1, 1))
3262 spin_lock(&fs_info->delayed_iput_lock);
3263 ASSERT(list_empty(&binode->delayed_iput));
3264 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3265 spin_unlock(&fs_info->delayed_iput_lock);
3268 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3271 spin_lock(&fs_info->delayed_iput_lock);
3272 while (!list_empty(&fs_info->delayed_iputs)) {
3273 struct btrfs_inode *inode;
3275 inode = list_first_entry(&fs_info->delayed_iputs,
3276 struct btrfs_inode, delayed_iput);
3277 list_del_init(&inode->delayed_iput);
3278 spin_unlock(&fs_info->delayed_iput_lock);
3279 iput(&inode->vfs_inode);
3280 spin_lock(&fs_info->delayed_iput_lock);
3282 spin_unlock(&fs_info->delayed_iput_lock);
3286 * This creates an orphan entry for the given inode in case something goes wrong
3287 * in the middle of an unlink.
3289 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3290 struct btrfs_inode *inode)
3294 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3295 if (ret && ret != -EEXIST) {
3296 btrfs_abort_transaction(trans, ret);
3304 * We have done the delete so we can go ahead and remove the orphan item for
3305 * this particular inode.
3307 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3308 struct btrfs_inode *inode)
3310 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3314 * this cleans up any orphans that may be left on the list from the last use
3317 int btrfs_orphan_cleanup(struct btrfs_root *root)
3319 struct btrfs_fs_info *fs_info = root->fs_info;
3320 struct btrfs_path *path;
3321 struct extent_buffer *leaf;
3322 struct btrfs_key key, found_key;
3323 struct btrfs_trans_handle *trans;
3324 struct inode *inode;
3325 u64 last_objectid = 0;
3326 int ret = 0, nr_unlink = 0;
3328 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3331 path = btrfs_alloc_path();
3336 path->reada = READA_BACK;
3338 key.objectid = BTRFS_ORPHAN_OBJECTID;
3339 key.type = BTRFS_ORPHAN_ITEM_KEY;
3340 key.offset = (u64)-1;
3343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3348 * if ret == 0 means we found what we were searching for, which
3349 * is weird, but possible, so only screw with path if we didn't
3350 * find the key and see if we have stuff that matches
3354 if (path->slots[0] == 0)
3359 /* pull out the item */
3360 leaf = path->nodes[0];
3361 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3363 /* make sure the item matches what we want */
3364 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3366 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3369 /* release the path since we're done with it */
3370 btrfs_release_path(path);
3373 * this is where we are basically btrfs_lookup, without the
3374 * crossing root thing. we store the inode number in the
3375 * offset of the orphan item.
3378 if (found_key.offset == last_objectid) {
3380 "Error removing orphan entry, stopping orphan cleanup");
3385 last_objectid = found_key.offset;
3387 found_key.objectid = found_key.offset;
3388 found_key.type = BTRFS_INODE_ITEM_KEY;
3389 found_key.offset = 0;
3390 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3391 ret = PTR_ERR_OR_ZERO(inode);
3392 if (ret && ret != -ENOENT)
3395 if (ret == -ENOENT && root == fs_info->tree_root) {
3396 struct btrfs_root *dead_root;
3397 struct btrfs_fs_info *fs_info = root->fs_info;
3398 int is_dead_root = 0;
3401 * this is an orphan in the tree root. Currently these
3402 * could come from 2 sources:
3403 * a) a snapshot deletion in progress
3404 * b) a free space cache inode
3405 * We need to distinguish those two, as the snapshot
3406 * orphan must not get deleted.
3407 * find_dead_roots already ran before us, so if this
3408 * is a snapshot deletion, we should find the root
3409 * in the dead_roots list
3411 spin_lock(&fs_info->trans_lock);
3412 list_for_each_entry(dead_root, &fs_info->dead_roots,
3414 if (dead_root->root_key.objectid ==
3415 found_key.objectid) {
3420 spin_unlock(&fs_info->trans_lock);
3422 /* prevent this orphan from being found again */
3423 key.offset = found_key.objectid - 1;
3430 * If we have an inode with links, there are a couple of
3431 * possibilities. Old kernels (before v3.12) used to create an
3432 * orphan item for truncate indicating that there were possibly
3433 * extent items past i_size that needed to be deleted. In v3.12,
3434 * truncate was changed to update i_size in sync with the extent
3435 * items, but the (useless) orphan item was still created. Since
3436 * v4.18, we don't create the orphan item for truncate at all.
3438 * So, this item could mean that we need to do a truncate, but
3439 * only if this filesystem was last used on a pre-v3.12 kernel
3440 * and was not cleanly unmounted. The odds of that are quite
3441 * slim, and it's a pain to do the truncate now, so just delete
3444 * It's also possible that this orphan item was supposed to be
3445 * deleted but wasn't. The inode number may have been reused,
3446 * but either way, we can delete the orphan item.
3448 if (ret == -ENOENT || inode->i_nlink) {
3451 trans = btrfs_start_transaction(root, 1);
3452 if (IS_ERR(trans)) {
3453 ret = PTR_ERR(trans);
3456 btrfs_debug(fs_info, "auto deleting %Lu",
3457 found_key.objectid);
3458 ret = btrfs_del_orphan_item(trans, root,
3459 found_key.objectid);
3460 btrfs_end_transaction(trans);
3468 /* this will do delete_inode and everything for us */
3471 /* release the path since we're done with it */
3472 btrfs_release_path(path);
3474 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3476 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3477 trans = btrfs_join_transaction(root);
3479 btrfs_end_transaction(trans);
3483 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3487 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3488 btrfs_free_path(path);
3493 * very simple check to peek ahead in the leaf looking for xattrs. If we
3494 * don't find any xattrs, we know there can't be any acls.
3496 * slot is the slot the inode is in, objectid is the objectid of the inode
3498 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3499 int slot, u64 objectid,
3500 int *first_xattr_slot)
3502 u32 nritems = btrfs_header_nritems(leaf);
3503 struct btrfs_key found_key;
3504 static u64 xattr_access = 0;
3505 static u64 xattr_default = 0;
3508 if (!xattr_access) {
3509 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3510 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3511 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3512 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3516 *first_xattr_slot = -1;
3517 while (slot < nritems) {
3518 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3520 /* we found a different objectid, there must not be acls */
3521 if (found_key.objectid != objectid)
3524 /* we found an xattr, assume we've got an acl */
3525 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3526 if (*first_xattr_slot == -1)
3527 *first_xattr_slot = slot;
3528 if (found_key.offset == xattr_access ||
3529 found_key.offset == xattr_default)
3534 * we found a key greater than an xattr key, there can't
3535 * be any acls later on
3537 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3544 * it goes inode, inode backrefs, xattrs, extents,
3545 * so if there are a ton of hard links to an inode there can
3546 * be a lot of backrefs. Don't waste time searching too hard,
3547 * this is just an optimization
3552 /* we hit the end of the leaf before we found an xattr or
3553 * something larger than an xattr. We have to assume the inode
3556 if (*first_xattr_slot == -1)
3557 *first_xattr_slot = slot;
3562 * read an inode from the btree into the in-memory inode
3564 static int btrfs_read_locked_inode(struct inode *inode)
3566 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3567 struct btrfs_path *path;
3568 struct extent_buffer *leaf;
3569 struct btrfs_inode_item *inode_item;
3570 struct btrfs_root *root = BTRFS_I(inode)->root;
3571 struct btrfs_key location;
3576 bool filled = false;
3577 int first_xattr_slot;
3579 ret = btrfs_fill_inode(inode, &rdev);
3583 path = btrfs_alloc_path();
3587 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3589 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3591 btrfs_free_path(path);
3595 leaf = path->nodes[0];
3600 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3601 struct btrfs_inode_item);
3602 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3603 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3604 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3605 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3606 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3608 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3609 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3611 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3612 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3614 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3615 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3617 BTRFS_I(inode)->i_otime.tv_sec =
3618 btrfs_timespec_sec(leaf, &inode_item->otime);
3619 BTRFS_I(inode)->i_otime.tv_nsec =
3620 btrfs_timespec_nsec(leaf, &inode_item->otime);
3622 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3623 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3624 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3626 inode_set_iversion_queried(inode,
3627 btrfs_inode_sequence(leaf, inode_item));
3628 inode->i_generation = BTRFS_I(inode)->generation;
3630 rdev = btrfs_inode_rdev(leaf, inode_item);
3632 BTRFS_I(inode)->index_cnt = (u64)-1;
3633 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3637 * If we were modified in the current generation and evicted from memory
3638 * and then re-read we need to do a full sync since we don't have any
3639 * idea about which extents were modified before we were evicted from
3642 * This is required for both inode re-read from disk and delayed inode
3643 * in delayed_nodes_tree.
3645 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3646 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3647 &BTRFS_I(inode)->runtime_flags);
3650 * We don't persist the id of the transaction where an unlink operation
3651 * against the inode was last made. So here we assume the inode might
3652 * have been evicted, and therefore the exact value of last_unlink_trans
3653 * lost, and set it to last_trans to avoid metadata inconsistencies
3654 * between the inode and its parent if the inode is fsync'ed and the log
3655 * replayed. For example, in the scenario:
3658 * ln mydir/foo mydir/bar
3661 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3662 * xfs_io -c fsync mydir/foo
3664 * mount fs, triggers fsync log replay
3666 * We must make sure that when we fsync our inode foo we also log its
3667 * parent inode, otherwise after log replay the parent still has the
3668 * dentry with the "bar" name but our inode foo has a link count of 1
3669 * and doesn't have an inode ref with the name "bar" anymore.
3671 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3672 * but it guarantees correctness at the expense of occasional full
3673 * transaction commits on fsync if our inode is a directory, or if our
3674 * inode is not a directory, logging its parent unnecessarily.
3676 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3679 if (inode->i_nlink != 1 ||
3680 path->slots[0] >= btrfs_header_nritems(leaf))
3683 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3684 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3687 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3688 if (location.type == BTRFS_INODE_REF_KEY) {
3689 struct btrfs_inode_ref *ref;
3691 ref = (struct btrfs_inode_ref *)ptr;
3692 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3693 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3694 struct btrfs_inode_extref *extref;
3696 extref = (struct btrfs_inode_extref *)ptr;
3697 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3702 * try to precache a NULL acl entry for files that don't have
3703 * any xattrs or acls
3705 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3706 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3707 if (first_xattr_slot != -1) {
3708 path->slots[0] = first_xattr_slot;
3709 ret = btrfs_load_inode_props(inode, path);
3712 "error loading props for ino %llu (root %llu): %d",
3713 btrfs_ino(BTRFS_I(inode)),
3714 root->root_key.objectid, ret);
3716 btrfs_free_path(path);
3719 cache_no_acl(inode);
3721 switch (inode->i_mode & S_IFMT) {
3723 inode->i_mapping->a_ops = &btrfs_aops;
3724 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3725 inode->i_fop = &btrfs_file_operations;
3726 inode->i_op = &btrfs_file_inode_operations;
3729 inode->i_fop = &btrfs_dir_file_operations;
3730 inode->i_op = &btrfs_dir_inode_operations;
3733 inode->i_op = &btrfs_symlink_inode_operations;
3734 inode_nohighmem(inode);
3735 inode->i_mapping->a_ops = &btrfs_aops;
3738 inode->i_op = &btrfs_special_inode_operations;
3739 init_special_inode(inode, inode->i_mode, rdev);
3743 btrfs_sync_inode_flags_to_i_flags(inode);
3748 * given a leaf and an inode, copy the inode fields into the leaf
3750 static void fill_inode_item(struct btrfs_trans_handle *trans,
3751 struct extent_buffer *leaf,
3752 struct btrfs_inode_item *item,
3753 struct inode *inode)
3755 struct btrfs_map_token token;
3757 btrfs_init_map_token(&token);
3759 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3760 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3761 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3763 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3764 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3766 btrfs_set_token_timespec_sec(leaf, &item->atime,
3767 inode->i_atime.tv_sec, &token);
3768 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3769 inode->i_atime.tv_nsec, &token);
3771 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3772 inode->i_mtime.tv_sec, &token);
3773 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3774 inode->i_mtime.tv_nsec, &token);
3776 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3777 inode->i_ctime.tv_sec, &token);
3778 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3779 inode->i_ctime.tv_nsec, &token);
3781 btrfs_set_token_timespec_sec(leaf, &item->otime,
3782 BTRFS_I(inode)->i_otime.tv_sec, &token);
3783 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3784 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3786 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3788 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3790 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3792 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3793 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3794 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3795 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3799 * copy everything in the in-memory inode into the btree.
3801 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3802 struct btrfs_root *root, struct inode *inode)
3804 struct btrfs_inode_item *inode_item;
3805 struct btrfs_path *path;
3806 struct extent_buffer *leaf;
3809 path = btrfs_alloc_path();
3813 path->leave_spinning = 1;
3814 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3822 leaf = path->nodes[0];
3823 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3824 struct btrfs_inode_item);
3826 fill_inode_item(trans, leaf, inode_item, inode);
3827 btrfs_mark_buffer_dirty(leaf);
3828 btrfs_set_inode_last_trans(trans, inode);
3831 btrfs_free_path(path);
3836 * copy everything in the in-memory inode into the btree.
3838 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3839 struct btrfs_root *root, struct inode *inode)
3841 struct btrfs_fs_info *fs_info = root->fs_info;
3845 * If the inode is a free space inode, we can deadlock during commit
3846 * if we put it into the delayed code.
3848 * The data relocation inode should also be directly updated
3851 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3852 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3853 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3854 btrfs_update_root_times(trans, root);
3856 ret = btrfs_delayed_update_inode(trans, root, inode);
3858 btrfs_set_inode_last_trans(trans, inode);
3862 return btrfs_update_inode_item(trans, root, inode);
3865 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3866 struct btrfs_root *root,
3867 struct inode *inode)
3871 ret = btrfs_update_inode(trans, root, inode);
3873 return btrfs_update_inode_item(trans, root, inode);
3878 * unlink helper that gets used here in inode.c and in the tree logging
3879 * recovery code. It remove a link in a directory with a given name, and
3880 * also drops the back refs in the inode to the directory
3882 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3883 struct btrfs_root *root,
3884 struct btrfs_inode *dir,
3885 struct btrfs_inode *inode,
3886 const char *name, int name_len)
3888 struct btrfs_fs_info *fs_info = root->fs_info;
3889 struct btrfs_path *path;
3891 struct extent_buffer *leaf;
3892 struct btrfs_dir_item *di;
3893 struct btrfs_key key;
3895 u64 ino = btrfs_ino(inode);
3896 u64 dir_ino = btrfs_ino(dir);
3898 path = btrfs_alloc_path();
3904 path->leave_spinning = 1;
3905 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3906 name, name_len, -1);
3907 if (IS_ERR_OR_NULL(di)) {
3908 ret = di ? PTR_ERR(di) : -ENOENT;
3911 leaf = path->nodes[0];
3912 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3913 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3916 btrfs_release_path(path);
3919 * If we don't have dir index, we have to get it by looking up
3920 * the inode ref, since we get the inode ref, remove it directly,
3921 * it is unnecessary to do delayed deletion.
3923 * But if we have dir index, needn't search inode ref to get it.
3924 * Since the inode ref is close to the inode item, it is better
3925 * that we delay to delete it, and just do this deletion when
3926 * we update the inode item.
3928 if (inode->dir_index) {
3929 ret = btrfs_delayed_delete_inode_ref(inode);
3931 index = inode->dir_index;
3936 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3940 "failed to delete reference to %.*s, inode %llu parent %llu",
3941 name_len, name, ino, dir_ino);
3942 btrfs_abort_transaction(trans, ret);
3946 ret = btrfs_delete_delayed_dir_index(trans, dir, index);
3948 btrfs_abort_transaction(trans, ret);
3952 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3954 if (ret != 0 && ret != -ENOENT) {
3955 btrfs_abort_transaction(trans, ret);
3959 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
3964 btrfs_abort_transaction(trans, ret);
3966 btrfs_free_path(path);
3970 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
3971 inode_inc_iversion(&inode->vfs_inode);
3972 inode_inc_iversion(&dir->vfs_inode);
3973 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
3974 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
3975 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
3980 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3981 struct btrfs_root *root,
3982 struct btrfs_inode *dir, struct btrfs_inode *inode,
3983 const char *name, int name_len)
3986 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3988 drop_nlink(&inode->vfs_inode);
3989 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
3995 * helper to start transaction for unlink and rmdir.
3997 * unlink and rmdir are special in btrfs, they do not always free space, so
3998 * if we cannot make our reservations the normal way try and see if there is
3999 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4000 * allow the unlink to occur.
4002 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4004 struct btrfs_root *root = BTRFS_I(dir)->root;
4007 * 1 for the possible orphan item
4008 * 1 for the dir item
4009 * 1 for the dir index
4010 * 1 for the inode ref
4013 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4016 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4018 struct btrfs_root *root = BTRFS_I(dir)->root;
4019 struct btrfs_trans_handle *trans;
4020 struct inode *inode = d_inode(dentry);
4023 trans = __unlink_start_trans(dir);
4025 return PTR_ERR(trans);
4027 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4030 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4031 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4032 dentry->d_name.len);
4036 if (inode->i_nlink == 0) {
4037 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4043 btrfs_end_transaction(trans);
4044 btrfs_btree_balance_dirty(root->fs_info);
4048 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4049 struct inode *dir, u64 objectid,
4050 const char *name, int name_len)
4052 struct btrfs_root *root = BTRFS_I(dir)->root;
4053 struct btrfs_path *path;
4054 struct extent_buffer *leaf;
4055 struct btrfs_dir_item *di;
4056 struct btrfs_key key;
4059 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4061 path = btrfs_alloc_path();
4065 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4066 name, name_len, -1);
4067 if (IS_ERR_OR_NULL(di)) {
4068 ret = di ? PTR_ERR(di) : -ENOENT;
4072 leaf = path->nodes[0];
4073 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4074 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4075 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4077 btrfs_abort_transaction(trans, ret);
4080 btrfs_release_path(path);
4082 ret = btrfs_del_root_ref(trans, objectid, root->root_key.objectid,
4083 dir_ino, &index, name, name_len);
4085 if (ret != -ENOENT) {
4086 btrfs_abort_transaction(trans, ret);
4089 di = btrfs_search_dir_index_item(root, path, dir_ino,
4091 if (IS_ERR_OR_NULL(di)) {
4096 btrfs_abort_transaction(trans, ret);
4100 leaf = path->nodes[0];
4101 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4104 btrfs_release_path(path);
4106 ret = btrfs_delete_delayed_dir_index(trans, BTRFS_I(dir), index);
4108 btrfs_abort_transaction(trans, ret);
4112 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4113 inode_inc_iversion(dir);
4114 dir->i_mtime = dir->i_ctime = current_time(dir);
4115 ret = btrfs_update_inode_fallback(trans, root, dir);
4117 btrfs_abort_transaction(trans, ret);
4119 btrfs_free_path(path);
4124 * Helper to check if the subvolume references other subvolumes or if it's
4127 static noinline int may_destroy_subvol(struct btrfs_root *root)
4129 struct btrfs_fs_info *fs_info = root->fs_info;
4130 struct btrfs_path *path;
4131 struct btrfs_dir_item *di;
4132 struct btrfs_key key;
4136 path = btrfs_alloc_path();
4140 /* Make sure this root isn't set as the default subvol */
4141 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4142 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4143 dir_id, "default", 7, 0);
4144 if (di && !IS_ERR(di)) {
4145 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4146 if (key.objectid == root->root_key.objectid) {
4149 "deleting default subvolume %llu is not allowed",
4153 btrfs_release_path(path);
4156 key.objectid = root->root_key.objectid;
4157 key.type = BTRFS_ROOT_REF_KEY;
4158 key.offset = (u64)-1;
4160 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4166 if (path->slots[0] > 0) {
4168 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4169 if (key.objectid == root->root_key.objectid &&
4170 key.type == BTRFS_ROOT_REF_KEY)
4174 btrfs_free_path(path);
4178 /* Delete all dentries for inodes belonging to the root */
4179 static void btrfs_prune_dentries(struct btrfs_root *root)
4181 struct btrfs_fs_info *fs_info = root->fs_info;
4182 struct rb_node *node;
4183 struct rb_node *prev;
4184 struct btrfs_inode *entry;
4185 struct inode *inode;
4188 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4189 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4191 spin_lock(&root->inode_lock);
4193 node = root->inode_tree.rb_node;
4197 entry = rb_entry(node, struct btrfs_inode, rb_node);
4199 if (objectid < btrfs_ino(entry))
4200 node = node->rb_left;
4201 else if (objectid > btrfs_ino(entry))
4202 node = node->rb_right;
4208 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4209 if (objectid <= btrfs_ino(entry)) {
4213 prev = rb_next(prev);
4217 entry = rb_entry(node, struct btrfs_inode, rb_node);
4218 objectid = btrfs_ino(entry) + 1;
4219 inode = igrab(&entry->vfs_inode);
4221 spin_unlock(&root->inode_lock);
4222 if (atomic_read(&inode->i_count) > 1)
4223 d_prune_aliases(inode);
4225 * btrfs_drop_inode will have it removed from the inode
4226 * cache when its usage count hits zero.
4230 spin_lock(&root->inode_lock);
4234 if (cond_resched_lock(&root->inode_lock))
4237 node = rb_next(node);
4239 spin_unlock(&root->inode_lock);
4242 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4244 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4245 struct btrfs_root *root = BTRFS_I(dir)->root;
4246 struct inode *inode = d_inode(dentry);
4247 struct btrfs_root *dest = BTRFS_I(inode)->root;
4248 struct btrfs_trans_handle *trans;
4249 struct btrfs_block_rsv block_rsv;
4255 * Don't allow to delete a subvolume with send in progress. This is
4256 * inside the inode lock so the error handling that has to drop the bit
4257 * again is not run concurrently.
4259 spin_lock(&dest->root_item_lock);
4260 if (dest->send_in_progress) {
4261 spin_unlock(&dest->root_item_lock);
4263 "attempt to delete subvolume %llu during send",
4264 dest->root_key.objectid);
4267 root_flags = btrfs_root_flags(&dest->root_item);
4268 btrfs_set_root_flags(&dest->root_item,
4269 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4270 spin_unlock(&dest->root_item_lock);
4272 down_write(&fs_info->subvol_sem);
4274 err = may_destroy_subvol(dest);
4278 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4280 * One for dir inode,
4281 * two for dir entries,
4282 * two for root ref/backref.
4284 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4288 trans = btrfs_start_transaction(root, 0);
4289 if (IS_ERR(trans)) {
4290 err = PTR_ERR(trans);
4293 trans->block_rsv = &block_rsv;
4294 trans->bytes_reserved = block_rsv.size;
4296 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4298 ret = btrfs_unlink_subvol(trans, dir, dest->root_key.objectid,
4299 dentry->d_name.name, dentry->d_name.len);
4302 btrfs_abort_transaction(trans, ret);
4306 btrfs_record_root_in_trans(trans, dest);
4308 memset(&dest->root_item.drop_progress, 0,
4309 sizeof(dest->root_item.drop_progress));
4310 dest->root_item.drop_level = 0;
4311 btrfs_set_root_refs(&dest->root_item, 0);
4313 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4314 ret = btrfs_insert_orphan_item(trans,
4316 dest->root_key.objectid);
4318 btrfs_abort_transaction(trans, ret);
4324 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4325 BTRFS_UUID_KEY_SUBVOL,
4326 dest->root_key.objectid);
4327 if (ret && ret != -ENOENT) {
4328 btrfs_abort_transaction(trans, ret);
4332 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4333 ret = btrfs_uuid_tree_remove(trans,
4334 dest->root_item.received_uuid,
4335 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4336 dest->root_key.objectid);
4337 if (ret && ret != -ENOENT) {
4338 btrfs_abort_transaction(trans, ret);
4345 trans->block_rsv = NULL;
4346 trans->bytes_reserved = 0;
4347 ret = btrfs_end_transaction(trans);
4350 inode->i_flags |= S_DEAD;
4352 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4354 up_write(&fs_info->subvol_sem);
4356 spin_lock(&dest->root_item_lock);
4357 root_flags = btrfs_root_flags(&dest->root_item);
4358 btrfs_set_root_flags(&dest->root_item,
4359 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4360 spin_unlock(&dest->root_item_lock);
4362 d_invalidate(dentry);
4363 btrfs_prune_dentries(dest);
4364 ASSERT(dest->send_in_progress == 0);
4367 if (dest->ino_cache_inode) {
4368 iput(dest->ino_cache_inode);
4369 dest->ino_cache_inode = NULL;
4376 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4378 struct inode *inode = d_inode(dentry);
4380 struct btrfs_root *root = BTRFS_I(dir)->root;
4381 struct btrfs_trans_handle *trans;
4382 u64 last_unlink_trans;
4384 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4386 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4387 return btrfs_delete_subvolume(dir, dentry);
4389 trans = __unlink_start_trans(dir);
4391 return PTR_ERR(trans);
4393 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4394 err = btrfs_unlink_subvol(trans, dir,
4395 BTRFS_I(inode)->location.objectid,
4396 dentry->d_name.name,
4397 dentry->d_name.len);
4401 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4405 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4407 /* now the directory is empty */
4408 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4409 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4410 dentry->d_name.len);
4412 btrfs_i_size_write(BTRFS_I(inode), 0);
4414 * Propagate the last_unlink_trans value of the deleted dir to
4415 * its parent directory. This is to prevent an unrecoverable
4416 * log tree in the case we do something like this:
4418 * 2) create snapshot under dir foo
4419 * 3) delete the snapshot
4422 * 6) fsync foo or some file inside foo
4424 if (last_unlink_trans >= trans->transid)
4425 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4428 btrfs_end_transaction(trans);
4429 btrfs_btree_balance_dirty(root->fs_info);
4434 static int truncate_space_check(struct btrfs_trans_handle *trans,
4435 struct btrfs_root *root,
4438 struct btrfs_fs_info *fs_info = root->fs_info;
4442 * This is only used to apply pressure to the enospc system, we don't
4443 * intend to use this reservation at all.
4445 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4446 bytes_deleted *= fs_info->nodesize;
4447 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4448 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4450 trace_btrfs_space_reservation(fs_info, "transaction",
4453 trans->bytes_reserved += bytes_deleted;
4460 * Return this if we need to call truncate_block for the last bit of the
4463 #define NEED_TRUNCATE_BLOCK 1
4466 * this can truncate away extent items, csum items and directory items.
4467 * It starts at a high offset and removes keys until it can't find
4468 * any higher than new_size
4470 * csum items that cross the new i_size are truncated to the new size
4473 * min_type is the minimum key type to truncate down to. If set to 0, this
4474 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4476 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4477 struct btrfs_root *root,
4478 struct inode *inode,
4479 u64 new_size, u32 min_type)
4481 struct btrfs_fs_info *fs_info = root->fs_info;
4482 struct btrfs_path *path;
4483 struct extent_buffer *leaf;
4484 struct btrfs_file_extent_item *fi;
4485 struct btrfs_key key;
4486 struct btrfs_key found_key;
4487 u64 extent_start = 0;
4488 u64 extent_num_bytes = 0;
4489 u64 extent_offset = 0;
4491 u64 last_size = new_size;
4492 u32 found_type = (u8)-1;
4495 int pending_del_nr = 0;
4496 int pending_del_slot = 0;
4497 int extent_type = -1;
4499 u64 ino = btrfs_ino(BTRFS_I(inode));
4500 u64 bytes_deleted = 0;
4501 bool be_nice = false;
4502 bool should_throttle = false;
4503 bool should_end = false;
4505 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4508 * for non-free space inodes and ref cows, we want to back off from
4511 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4512 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4515 path = btrfs_alloc_path();
4518 path->reada = READA_BACK;
4521 * We want to drop from the next block forward in case this new size is
4522 * not block aligned since we will be keeping the last block of the
4523 * extent just the way it is.
4525 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4526 root == fs_info->tree_root)
4527 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4528 fs_info->sectorsize),
4532 * This function is also used to drop the items in the log tree before
4533 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4534 * it is used to drop the loged items. So we shouldn't kill the delayed
4537 if (min_type == 0 && root == BTRFS_I(inode)->root)
4538 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4541 key.offset = (u64)-1;
4546 * with a 16K leaf size and 128MB extents, you can actually queue
4547 * up a huge file in a single leaf. Most of the time that
4548 * bytes_deleted is > 0, it will be huge by the time we get here
4550 if (be_nice && bytes_deleted > SZ_32M &&
4551 btrfs_should_end_transaction(trans)) {
4556 path->leave_spinning = 1;
4557 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4563 /* there are no items in the tree for us to truncate, we're
4566 if (path->slots[0] == 0)
4573 leaf = path->nodes[0];
4574 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4575 found_type = found_key.type;
4577 if (found_key.objectid != ino)
4580 if (found_type < min_type)
4583 item_end = found_key.offset;
4584 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4585 fi = btrfs_item_ptr(leaf, path->slots[0],
4586 struct btrfs_file_extent_item);
4587 extent_type = btrfs_file_extent_type(leaf, fi);
4588 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4590 btrfs_file_extent_num_bytes(leaf, fi);
4592 trace_btrfs_truncate_show_fi_regular(
4593 BTRFS_I(inode), leaf, fi,
4595 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4596 item_end += btrfs_file_extent_ram_bytes(leaf,
4599 trace_btrfs_truncate_show_fi_inline(
4600 BTRFS_I(inode), leaf, fi, path->slots[0],
4605 if (found_type > min_type) {
4608 if (item_end < new_size)
4610 if (found_key.offset >= new_size)
4616 /* FIXME, shrink the extent if the ref count is only 1 */
4617 if (found_type != BTRFS_EXTENT_DATA_KEY)
4620 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4622 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4624 u64 orig_num_bytes =
4625 btrfs_file_extent_num_bytes(leaf, fi);
4626 extent_num_bytes = ALIGN(new_size -
4628 fs_info->sectorsize);
4629 btrfs_set_file_extent_num_bytes(leaf, fi,
4631 num_dec = (orig_num_bytes -
4633 if (test_bit(BTRFS_ROOT_REF_COWS,
4636 inode_sub_bytes(inode, num_dec);
4637 btrfs_mark_buffer_dirty(leaf);
4640 btrfs_file_extent_disk_num_bytes(leaf,
4642 extent_offset = found_key.offset -
4643 btrfs_file_extent_offset(leaf, fi);
4645 /* FIXME blocksize != 4096 */
4646 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4647 if (extent_start != 0) {
4649 if (test_bit(BTRFS_ROOT_REF_COWS,
4651 inode_sub_bytes(inode, num_dec);
4654 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4656 * we can't truncate inline items that have had
4660 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4661 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4662 btrfs_file_extent_compression(leaf, fi) == 0) {
4663 u32 size = (u32)(new_size - found_key.offset);
4665 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4666 size = btrfs_file_extent_calc_inline_size(size);
4667 btrfs_truncate_item(root->fs_info, path, size, 1);
4668 } else if (!del_item) {
4670 * We have to bail so the last_size is set to
4671 * just before this extent.
4673 ret = NEED_TRUNCATE_BLOCK;
4677 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4678 inode_sub_bytes(inode, item_end + 1 - new_size);
4682 last_size = found_key.offset;
4684 last_size = new_size;
4686 if (!pending_del_nr) {
4687 /* no pending yet, add ourselves */
4688 pending_del_slot = path->slots[0];
4690 } else if (pending_del_nr &&
4691 path->slots[0] + 1 == pending_del_slot) {
4692 /* hop on the pending chunk */
4694 pending_del_slot = path->slots[0];
4701 should_throttle = false;
4704 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4705 root == fs_info->tree_root)) {
4706 btrfs_set_path_blocking(path);
4707 bytes_deleted += extent_num_bytes;
4708 ret = btrfs_free_extent(trans, root, extent_start,
4709 extent_num_bytes, 0,
4710 btrfs_header_owner(leaf),
4711 ino, extent_offset);
4713 btrfs_abort_transaction(trans, ret);
4716 if (btrfs_should_throttle_delayed_refs(trans))
4717 btrfs_async_run_delayed_refs(fs_info,
4718 trans->delayed_ref_updates * 2,
4721 if (truncate_space_check(trans, root,
4722 extent_num_bytes)) {
4725 if (btrfs_should_throttle_delayed_refs(trans))
4726 should_throttle = true;
4730 if (found_type == BTRFS_INODE_ITEM_KEY)
4733 if (path->slots[0] == 0 ||
4734 path->slots[0] != pending_del_slot ||
4735 should_throttle || should_end) {
4736 if (pending_del_nr) {
4737 ret = btrfs_del_items(trans, root, path,
4741 btrfs_abort_transaction(trans, ret);
4746 btrfs_release_path(path);
4747 if (should_throttle) {
4748 unsigned long updates = trans->delayed_ref_updates;
4750 trans->delayed_ref_updates = 0;
4751 ret = btrfs_run_delayed_refs(trans,
4758 * if we failed to refill our space rsv, bail out
4759 * and let the transaction restart
4771 if (ret >= 0 && pending_del_nr) {
4774 err = btrfs_del_items(trans, root, path, pending_del_slot,
4777 btrfs_abort_transaction(trans, err);
4781 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4782 ASSERT(last_size >= new_size);
4783 if (!ret && last_size > new_size)
4784 last_size = new_size;
4785 btrfs_ordered_update_i_size(inode, last_size, NULL);
4788 btrfs_free_path(path);
4790 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4791 unsigned long updates = trans->delayed_ref_updates;
4795 trans->delayed_ref_updates = 0;
4796 err = btrfs_run_delayed_refs(trans, updates * 2);
4805 * btrfs_truncate_block - read, zero a chunk and write a block
4806 * @inode - inode that we're zeroing
4807 * @from - the offset to start zeroing
4808 * @len - the length to zero, 0 to zero the entire range respective to the
4810 * @front - zero up to the offset instead of from the offset on
4812 * This will find the block for the "from" offset and cow the block and zero the
4813 * part we want to zero. This is used with truncate and hole punching.
4815 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4818 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4819 struct address_space *mapping = inode->i_mapping;
4820 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4821 struct btrfs_ordered_extent *ordered;
4822 struct extent_state *cached_state = NULL;
4823 struct extent_changeset *data_reserved = NULL;
4825 u32 blocksize = fs_info->sectorsize;
4826 pgoff_t index = from >> PAGE_SHIFT;
4827 unsigned offset = from & (blocksize - 1);
4829 gfp_t mask = btrfs_alloc_write_mask(mapping);
4834 if (IS_ALIGNED(offset, blocksize) &&
4835 (!len || IS_ALIGNED(len, blocksize)))
4838 block_start = round_down(from, blocksize);
4839 block_end = block_start + blocksize - 1;
4841 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4842 block_start, blocksize);
4847 page = find_or_create_page(mapping, index, mask);
4849 btrfs_delalloc_release_space(inode, data_reserved,
4850 block_start, blocksize, true);
4851 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4856 if (!PageUptodate(page)) {
4857 ret = btrfs_readpage(NULL, page);
4859 if (page->mapping != mapping) {
4864 if (!PageUptodate(page)) {
4869 wait_on_page_writeback(page);
4871 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4872 set_page_extent_mapped(page);
4874 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4876 unlock_extent_cached(io_tree, block_start, block_end,
4880 btrfs_start_ordered_extent(inode, ordered, 1);
4881 btrfs_put_ordered_extent(ordered);
4885 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4886 EXTENT_DIRTY | EXTENT_DELALLOC |
4887 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4888 0, 0, &cached_state);
4890 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4893 unlock_extent_cached(io_tree, block_start, block_end,
4898 if (offset != blocksize) {
4900 len = blocksize - offset;
4903 memset(kaddr + (block_start - page_offset(page)),
4906 memset(kaddr + (block_start - page_offset(page)) + offset,
4908 flush_dcache_page(page);
4911 ClearPageChecked(page);
4912 set_page_dirty(page);
4913 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4917 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4919 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4923 extent_changeset_free(data_reserved);
4927 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4928 u64 offset, u64 len)
4930 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4931 struct btrfs_trans_handle *trans;
4935 * Still need to make sure the inode looks like it's been updated so
4936 * that any holes get logged if we fsync.
4938 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4939 BTRFS_I(inode)->last_trans = fs_info->generation;
4940 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4941 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4946 * 1 - for the one we're dropping
4947 * 1 - for the one we're adding
4948 * 1 - for updating the inode.
4950 trans = btrfs_start_transaction(root, 3);
4952 return PTR_ERR(trans);
4954 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4956 btrfs_abort_transaction(trans, ret);
4957 btrfs_end_transaction(trans);
4961 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
4962 offset, 0, 0, len, 0, len, 0, 0, 0);
4964 btrfs_abort_transaction(trans, ret);
4966 btrfs_update_inode(trans, root, inode);
4967 btrfs_end_transaction(trans);
4972 * This function puts in dummy file extents for the area we're creating a hole
4973 * for. So if we are truncating this file to a larger size we need to insert
4974 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4975 * the range between oldsize and size
4977 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4979 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4980 struct btrfs_root *root = BTRFS_I(inode)->root;
4981 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4982 struct extent_map *em = NULL;
4983 struct extent_state *cached_state = NULL;
4984 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4985 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
4986 u64 block_end = ALIGN(size, fs_info->sectorsize);
4993 * If our size started in the middle of a block we need to zero out the
4994 * rest of the block before we expand the i_size, otherwise we could
4995 * expose stale data.
4997 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5001 if (size <= hole_start)
5005 struct btrfs_ordered_extent *ordered;
5007 lock_extent_bits(io_tree, hole_start, block_end - 1,
5009 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5010 block_end - hole_start);
5013 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5015 btrfs_start_ordered_extent(inode, ordered, 1);
5016 btrfs_put_ordered_extent(ordered);
5019 cur_offset = hole_start;
5021 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5022 block_end - cur_offset, 0);
5028 last_byte = min(extent_map_end(em), block_end);
5029 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5030 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5031 struct extent_map *hole_em;
5032 hole_size = last_byte - cur_offset;
5034 err = maybe_insert_hole(root, inode, cur_offset,
5038 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5039 cur_offset + hole_size - 1, 0);
5040 hole_em = alloc_extent_map();
5042 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5043 &BTRFS_I(inode)->runtime_flags);
5046 hole_em->start = cur_offset;
5047 hole_em->len = hole_size;
5048 hole_em->orig_start = cur_offset;
5050 hole_em->block_start = EXTENT_MAP_HOLE;
5051 hole_em->block_len = 0;
5052 hole_em->orig_block_len = 0;
5053 hole_em->ram_bytes = hole_size;
5054 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5055 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5056 hole_em->generation = fs_info->generation;
5059 write_lock(&em_tree->lock);
5060 err = add_extent_mapping(em_tree, hole_em, 1);
5061 write_unlock(&em_tree->lock);
5064 btrfs_drop_extent_cache(BTRFS_I(inode),
5069 free_extent_map(hole_em);
5072 free_extent_map(em);
5074 cur_offset = last_byte;
5075 if (cur_offset >= block_end)
5078 free_extent_map(em);
5079 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5083 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5085 struct btrfs_root *root = BTRFS_I(inode)->root;
5086 struct btrfs_trans_handle *trans;
5087 loff_t oldsize = i_size_read(inode);
5088 loff_t newsize = attr->ia_size;
5089 int mask = attr->ia_valid;
5093 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5094 * special case where we need to update the times despite not having
5095 * these flags set. For all other operations the VFS set these flags
5096 * explicitly if it wants a timestamp update.
5098 if (newsize != oldsize) {
5099 inode_inc_iversion(inode);
5100 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5101 inode->i_ctime = inode->i_mtime =
5102 current_time(inode);
5105 if (newsize > oldsize) {
5107 * Don't do an expanding truncate while snapshotting is ongoing.
5108 * This is to ensure the snapshot captures a fully consistent
5109 * state of this file - if the snapshot captures this expanding
5110 * truncation, it must capture all writes that happened before
5113 btrfs_wait_for_snapshot_creation(root);
5114 ret = btrfs_cont_expand(inode, oldsize, newsize);
5116 btrfs_end_write_no_snapshotting(root);
5120 trans = btrfs_start_transaction(root, 1);
5121 if (IS_ERR(trans)) {
5122 btrfs_end_write_no_snapshotting(root);
5123 return PTR_ERR(trans);
5126 i_size_write(inode, newsize);
5127 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5128 pagecache_isize_extended(inode, oldsize, newsize);
5129 ret = btrfs_update_inode(trans, root, inode);
5130 btrfs_end_write_no_snapshotting(root);
5131 btrfs_end_transaction(trans);
5135 * We're truncating a file that used to have good data down to
5136 * zero. Make sure it gets into the ordered flush list so that
5137 * any new writes get down to disk quickly.
5140 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5141 &BTRFS_I(inode)->runtime_flags);
5143 truncate_setsize(inode, newsize);
5145 /* Disable nonlocked read DIO to avoid the end less truncate */
5146 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5147 inode_dio_wait(inode);
5148 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5150 ret = btrfs_truncate(inode, newsize == oldsize);
5151 if (ret && inode->i_nlink) {
5155 * Truncate failed, so fix up the in-memory size. We
5156 * adjusted disk_i_size down as we removed extents, so
5157 * wait for disk_i_size to be stable and then update the
5158 * in-memory size to match.
5160 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5163 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5170 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5172 struct inode *inode = d_inode(dentry);
5173 struct btrfs_root *root = BTRFS_I(inode)->root;
5176 if (btrfs_root_readonly(root))
5179 err = setattr_prepare(dentry, attr);
5183 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5184 err = btrfs_setsize(inode, attr);
5189 if (attr->ia_valid) {
5190 setattr_copy(inode, attr);
5191 inode_inc_iversion(inode);
5192 err = btrfs_dirty_inode(inode);
5194 if (!err && attr->ia_valid & ATTR_MODE)
5195 err = posix_acl_chmod(inode, inode->i_mode);
5202 * While truncating the inode pages during eviction, we get the VFS calling
5203 * btrfs_invalidatepage() against each page of the inode. This is slow because
5204 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5205 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5206 * extent_state structures over and over, wasting lots of time.
5208 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5209 * those expensive operations on a per page basis and do only the ordered io
5210 * finishing, while we release here the extent_map and extent_state structures,
5211 * without the excessive merging and splitting.
5213 static void evict_inode_truncate_pages(struct inode *inode)
5215 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5216 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5217 struct rb_node *node;
5219 ASSERT(inode->i_state & I_FREEING);
5220 truncate_inode_pages_final(&inode->i_data);
5222 write_lock(&map_tree->lock);
5223 while (!RB_EMPTY_ROOT(&map_tree->map.rb_root)) {
5224 struct extent_map *em;
5226 node = rb_first_cached(&map_tree->map);
5227 em = rb_entry(node, struct extent_map, rb_node);
5228 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5229 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5230 remove_extent_mapping(map_tree, em);
5231 free_extent_map(em);
5232 if (need_resched()) {
5233 write_unlock(&map_tree->lock);
5235 write_lock(&map_tree->lock);
5238 write_unlock(&map_tree->lock);
5241 * Keep looping until we have no more ranges in the io tree.
5242 * We can have ongoing bios started by readpages (called from readahead)
5243 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5244 * still in progress (unlocked the pages in the bio but did not yet
5245 * unlocked the ranges in the io tree). Therefore this means some
5246 * ranges can still be locked and eviction started because before
5247 * submitting those bios, which are executed by a separate task (work
5248 * queue kthread), inode references (inode->i_count) were not taken
5249 * (which would be dropped in the end io callback of each bio).
5250 * Therefore here we effectively end up waiting for those bios and
5251 * anyone else holding locked ranges without having bumped the inode's
5252 * reference count - if we don't do it, when they access the inode's
5253 * io_tree to unlock a range it may be too late, leading to an
5254 * use-after-free issue.
5256 spin_lock(&io_tree->lock);
5257 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5258 struct extent_state *state;
5259 struct extent_state *cached_state = NULL;
5263 node = rb_first(&io_tree->state);
5264 state = rb_entry(node, struct extent_state, rb_node);
5265 start = state->start;
5267 spin_unlock(&io_tree->lock);
5269 lock_extent_bits(io_tree, start, end, &cached_state);
5272 * If still has DELALLOC flag, the extent didn't reach disk,
5273 * and its reserved space won't be freed by delayed_ref.
5274 * So we need to free its reserved space here.
5275 * (Refer to comment in btrfs_invalidatepage, case 2)
5277 * Note, end is the bytenr of last byte, so we need + 1 here.
5279 if (state->state & EXTENT_DELALLOC)
5280 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5282 clear_extent_bit(io_tree, start, end,
5283 EXTENT_LOCKED | EXTENT_DIRTY |
5284 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5285 EXTENT_DEFRAG, 1, 1, &cached_state);
5288 spin_lock(&io_tree->lock);
5290 spin_unlock(&io_tree->lock);
5293 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5294 struct btrfs_block_rsv *rsv)
5296 struct btrfs_fs_info *fs_info = root->fs_info;
5297 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5301 struct btrfs_trans_handle *trans;
5304 ret = btrfs_block_rsv_refill(root, rsv, rsv->size,
5305 BTRFS_RESERVE_FLUSH_LIMIT);
5307 if (ret && ++failures > 2) {
5309 "could not allocate space for a delete; will truncate on mount");
5310 return ERR_PTR(-ENOSPC);
5313 trans = btrfs_join_transaction(root);
5314 if (IS_ERR(trans) || !ret)
5318 * Try to steal from the global reserve if there is space for
5321 if (!btrfs_check_space_for_delayed_refs(trans) &&
5322 !btrfs_block_rsv_migrate(global_rsv, rsv, rsv->size, false))
5325 /* If not, commit and try again. */
5326 ret = btrfs_commit_transaction(trans);
5328 return ERR_PTR(ret);
5332 void btrfs_evict_inode(struct inode *inode)
5334 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5335 struct btrfs_trans_handle *trans;
5336 struct btrfs_root *root = BTRFS_I(inode)->root;
5337 struct btrfs_block_rsv *rsv;
5340 trace_btrfs_inode_evict(inode);
5347 evict_inode_truncate_pages(inode);
5349 if (inode->i_nlink &&
5350 ((btrfs_root_refs(&root->root_item) != 0 &&
5351 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5352 btrfs_is_free_space_inode(BTRFS_I(inode))))
5355 if (is_bad_inode(inode))
5358 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5360 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5363 if (inode->i_nlink > 0) {
5364 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5365 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5369 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5373 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5376 rsv->size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5379 btrfs_i_size_write(BTRFS_I(inode), 0);
5382 trans = evict_refill_and_join(root, rsv);
5386 trans->block_rsv = rsv;
5388 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5389 trans->block_rsv = &fs_info->trans_block_rsv;
5390 btrfs_end_transaction(trans);
5391 btrfs_btree_balance_dirty(fs_info);
5392 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5399 * Errors here aren't a big deal, it just means we leave orphan items in
5400 * the tree. They will be cleaned up on the next mount. If the inode
5401 * number gets reused, cleanup deletes the orphan item without doing
5402 * anything, and unlink reuses the existing orphan item.
5404 * If it turns out that we are dropping too many of these, we might want
5405 * to add a mechanism for retrying these after a commit.
5407 trans = evict_refill_and_join(root, rsv);
5408 if (!IS_ERR(trans)) {
5409 trans->block_rsv = rsv;
5410 btrfs_orphan_del(trans, BTRFS_I(inode));
5411 trans->block_rsv = &fs_info->trans_block_rsv;
5412 btrfs_end_transaction(trans);
5415 if (!(root == fs_info->tree_root ||
5416 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5417 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5420 btrfs_free_block_rsv(fs_info, rsv);
5423 * If we didn't successfully delete, the orphan item will still be in
5424 * the tree and we'll retry on the next mount. Again, we might also want
5425 * to retry these periodically in the future.
5427 btrfs_remove_delayed_node(BTRFS_I(inode));
5432 * this returns the key found in the dir entry in the location pointer.
5433 * If no dir entries were found, returns -ENOENT.
5434 * If found a corrupted location in dir entry, returns -EUCLEAN.
5436 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5437 struct btrfs_key *location)
5439 const char *name = dentry->d_name.name;
5440 int namelen = dentry->d_name.len;
5441 struct btrfs_dir_item *di;
5442 struct btrfs_path *path;
5443 struct btrfs_root *root = BTRFS_I(dir)->root;
5446 path = btrfs_alloc_path();
5450 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5452 if (IS_ERR_OR_NULL(di)) {
5453 ret = di ? PTR_ERR(di) : -ENOENT;
5457 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5458 if (location->type != BTRFS_INODE_ITEM_KEY &&
5459 location->type != BTRFS_ROOT_ITEM_KEY) {
5461 btrfs_warn(root->fs_info,
5462 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5463 __func__, name, btrfs_ino(BTRFS_I(dir)),
5464 location->objectid, location->type, location->offset);
5467 btrfs_free_path(path);
5472 * when we hit a tree root in a directory, the btrfs part of the inode
5473 * needs to be changed to reflect the root directory of the tree root. This
5474 * is kind of like crossing a mount point.
5476 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5478 struct dentry *dentry,
5479 struct btrfs_key *location,
5480 struct btrfs_root **sub_root)
5482 struct btrfs_path *path;
5483 struct btrfs_root *new_root;
5484 struct btrfs_root_ref *ref;
5485 struct extent_buffer *leaf;
5486 struct btrfs_key key;
5490 path = btrfs_alloc_path();
5497 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5498 key.type = BTRFS_ROOT_REF_KEY;
5499 key.offset = location->objectid;
5501 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5508 leaf = path->nodes[0];
5509 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5510 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5511 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5514 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5515 (unsigned long)(ref + 1),
5516 dentry->d_name.len);
5520 btrfs_release_path(path);
5522 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5523 if (IS_ERR(new_root)) {
5524 err = PTR_ERR(new_root);
5528 *sub_root = new_root;
5529 location->objectid = btrfs_root_dirid(&new_root->root_item);
5530 location->type = BTRFS_INODE_ITEM_KEY;
5531 location->offset = 0;
5534 btrfs_free_path(path);
5538 static void inode_tree_add(struct inode *inode)
5540 struct btrfs_root *root = BTRFS_I(inode)->root;
5541 struct btrfs_inode *entry;
5543 struct rb_node *parent;
5544 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5545 u64 ino = btrfs_ino(BTRFS_I(inode));
5547 if (inode_unhashed(inode))
5550 spin_lock(&root->inode_lock);
5551 p = &root->inode_tree.rb_node;
5554 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5556 if (ino < btrfs_ino(entry))
5557 p = &parent->rb_left;
5558 else if (ino > btrfs_ino(entry))
5559 p = &parent->rb_right;
5561 WARN_ON(!(entry->vfs_inode.i_state &
5562 (I_WILL_FREE | I_FREEING)));
5563 rb_replace_node(parent, new, &root->inode_tree);
5564 RB_CLEAR_NODE(parent);
5565 spin_unlock(&root->inode_lock);
5569 rb_link_node(new, parent, p);
5570 rb_insert_color(new, &root->inode_tree);
5571 spin_unlock(&root->inode_lock);
5574 static void inode_tree_del(struct inode *inode)
5576 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5577 struct btrfs_root *root = BTRFS_I(inode)->root;
5580 spin_lock(&root->inode_lock);
5581 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5582 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5583 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5584 empty = RB_EMPTY_ROOT(&root->inode_tree);
5586 spin_unlock(&root->inode_lock);
5588 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5589 synchronize_srcu(&fs_info->subvol_srcu);
5590 spin_lock(&root->inode_lock);
5591 empty = RB_EMPTY_ROOT(&root->inode_tree);
5592 spin_unlock(&root->inode_lock);
5594 btrfs_add_dead_root(root);
5599 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5601 struct btrfs_iget_args *args = p;
5602 inode->i_ino = args->location->objectid;
5603 memcpy(&BTRFS_I(inode)->location, args->location,
5604 sizeof(*args->location));
5605 BTRFS_I(inode)->root = args->root;
5609 static int btrfs_find_actor(struct inode *inode, void *opaque)
5611 struct btrfs_iget_args *args = opaque;
5612 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5613 args->root == BTRFS_I(inode)->root;
5616 static struct inode *btrfs_iget_locked(struct super_block *s,
5617 struct btrfs_key *location,
5618 struct btrfs_root *root)
5620 struct inode *inode;
5621 struct btrfs_iget_args args;
5622 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5624 args.location = location;
5627 inode = iget5_locked(s, hashval, btrfs_find_actor,
5628 btrfs_init_locked_inode,
5633 /* Get an inode object given its location and corresponding root.
5634 * Returns in *is_new if the inode was read from disk
5636 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5637 struct btrfs_root *root, int *new)
5639 struct inode *inode;
5641 inode = btrfs_iget_locked(s, location, root);
5643 return ERR_PTR(-ENOMEM);
5645 if (inode->i_state & I_NEW) {
5648 ret = btrfs_read_locked_inode(inode);
5650 inode_tree_add(inode);
5651 unlock_new_inode(inode);
5657 * ret > 0 can come from btrfs_search_slot called by
5658 * btrfs_read_locked_inode, this means the inode item
5663 inode = ERR_PTR(ret);
5670 static struct inode *new_simple_dir(struct super_block *s,
5671 struct btrfs_key *key,
5672 struct btrfs_root *root)
5674 struct inode *inode = new_inode(s);
5677 return ERR_PTR(-ENOMEM);
5679 BTRFS_I(inode)->root = root;
5680 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5681 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5683 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5684 inode->i_op = &btrfs_dir_ro_inode_operations;
5685 inode->i_opflags &= ~IOP_XATTR;
5686 inode->i_fop = &simple_dir_operations;
5687 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5688 inode->i_mtime = current_time(inode);
5689 inode->i_atime = inode->i_mtime;
5690 inode->i_ctime = inode->i_mtime;
5691 BTRFS_I(inode)->i_otime = inode->i_mtime;
5696 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5698 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5699 struct inode *inode;
5700 struct btrfs_root *root = BTRFS_I(dir)->root;
5701 struct btrfs_root *sub_root = root;
5702 struct btrfs_key location;
5706 if (dentry->d_name.len > BTRFS_NAME_LEN)
5707 return ERR_PTR(-ENAMETOOLONG);
5709 ret = btrfs_inode_by_name(dir, dentry, &location);
5711 return ERR_PTR(ret);
5713 if (location.type == BTRFS_INODE_ITEM_KEY) {
5714 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5718 index = srcu_read_lock(&fs_info->subvol_srcu);
5719 ret = fixup_tree_root_location(fs_info, dir, dentry,
5720 &location, &sub_root);
5723 inode = ERR_PTR(ret);
5725 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5727 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5729 srcu_read_unlock(&fs_info->subvol_srcu, index);
5731 if (!IS_ERR(inode) && root != sub_root) {
5732 down_read(&fs_info->cleanup_work_sem);
5733 if (!sb_rdonly(inode->i_sb))
5734 ret = btrfs_orphan_cleanup(sub_root);
5735 up_read(&fs_info->cleanup_work_sem);
5738 inode = ERR_PTR(ret);
5745 static int btrfs_dentry_delete(const struct dentry *dentry)
5747 struct btrfs_root *root;
5748 struct inode *inode = d_inode(dentry);
5750 if (!inode && !IS_ROOT(dentry))
5751 inode = d_inode(dentry->d_parent);
5754 root = BTRFS_I(inode)->root;
5755 if (btrfs_root_refs(&root->root_item) == 0)
5758 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5764 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5767 struct inode *inode;
5769 inode = btrfs_lookup_dentry(dir, dentry);
5770 if (IS_ERR(inode)) {
5771 if (PTR_ERR(inode) == -ENOENT)
5774 return ERR_CAST(inode);
5777 return d_splice_alias(inode, dentry);
5780 unsigned char btrfs_filetype_table[] = {
5781 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5785 * All this infrastructure exists because dir_emit can fault, and we are holding
5786 * the tree lock when doing readdir. For now just allocate a buffer and copy
5787 * our information into that, and then dir_emit from the buffer. This is
5788 * similar to what NFS does, only we don't keep the buffer around in pagecache
5789 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5790 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5793 static int btrfs_opendir(struct inode *inode, struct file *file)
5795 struct btrfs_file_private *private;
5797 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5800 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5801 if (!private->filldir_buf) {
5805 file->private_data = private;
5816 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5819 struct dir_entry *entry = addr;
5820 char *name = (char *)(entry + 1);
5822 ctx->pos = get_unaligned(&entry->offset);
5823 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5824 get_unaligned(&entry->ino),
5825 get_unaligned(&entry->type)))
5827 addr += sizeof(struct dir_entry) +
5828 get_unaligned(&entry->name_len);
5834 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5836 struct inode *inode = file_inode(file);
5837 struct btrfs_root *root = BTRFS_I(inode)->root;
5838 struct btrfs_file_private *private = file->private_data;
5839 struct btrfs_dir_item *di;
5840 struct btrfs_key key;
5841 struct btrfs_key found_key;
5842 struct btrfs_path *path;
5844 struct list_head ins_list;
5845 struct list_head del_list;
5847 struct extent_buffer *leaf;
5854 struct btrfs_key location;
5856 if (!dir_emit_dots(file, ctx))
5859 path = btrfs_alloc_path();
5863 addr = private->filldir_buf;
5864 path->reada = READA_FORWARD;
5866 INIT_LIST_HEAD(&ins_list);
5867 INIT_LIST_HEAD(&del_list);
5868 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5871 key.type = BTRFS_DIR_INDEX_KEY;
5872 key.offset = ctx->pos;
5873 key.objectid = btrfs_ino(BTRFS_I(inode));
5875 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5880 struct dir_entry *entry;
5882 leaf = path->nodes[0];
5883 slot = path->slots[0];
5884 if (slot >= btrfs_header_nritems(leaf)) {
5885 ret = btrfs_next_leaf(root, path);
5893 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5895 if (found_key.objectid != key.objectid)
5897 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5899 if (found_key.offset < ctx->pos)
5901 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5903 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5904 name_len = btrfs_dir_name_len(leaf, di);
5905 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5907 btrfs_release_path(path);
5908 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5911 addr = private->filldir_buf;
5918 put_unaligned(name_len, &entry->name_len);
5919 name_ptr = (char *)(entry + 1);
5920 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5922 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5924 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5925 put_unaligned(location.objectid, &entry->ino);
5926 put_unaligned(found_key.offset, &entry->offset);
5928 addr += sizeof(struct dir_entry) + name_len;
5929 total_len += sizeof(struct dir_entry) + name_len;
5933 btrfs_release_path(path);
5935 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5939 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5944 * Stop new entries from being returned after we return the last
5947 * New directory entries are assigned a strictly increasing
5948 * offset. This means that new entries created during readdir
5949 * are *guaranteed* to be seen in the future by that readdir.
5950 * This has broken buggy programs which operate on names as
5951 * they're returned by readdir. Until we re-use freed offsets
5952 * we have this hack to stop new entries from being returned
5953 * under the assumption that they'll never reach this huge
5956 * This is being careful not to overflow 32bit loff_t unless the
5957 * last entry requires it because doing so has broken 32bit apps
5960 if (ctx->pos >= INT_MAX)
5961 ctx->pos = LLONG_MAX;
5968 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
5969 btrfs_free_path(path);
5974 * This is somewhat expensive, updating the tree every time the
5975 * inode changes. But, it is most likely to find the inode in cache.
5976 * FIXME, needs more benchmarking...there are no reasons other than performance
5977 * to keep or drop this code.
5979 static int btrfs_dirty_inode(struct inode *inode)
5981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5982 struct btrfs_root *root = BTRFS_I(inode)->root;
5983 struct btrfs_trans_handle *trans;
5986 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5989 trans = btrfs_join_transaction(root);
5991 return PTR_ERR(trans);
5993 ret = btrfs_update_inode(trans, root, inode);
5994 if (ret && ret == -ENOSPC) {
5995 /* whoops, lets try again with the full transaction */
5996 btrfs_end_transaction(trans);
5997 trans = btrfs_start_transaction(root, 1);
5999 return PTR_ERR(trans);
6001 ret = btrfs_update_inode(trans, root, inode);
6003 btrfs_end_transaction(trans);
6004 if (BTRFS_I(inode)->delayed_node)
6005 btrfs_balance_delayed_items(fs_info);
6011 * This is a copy of file_update_time. We need this so we can return error on
6012 * ENOSPC for updating the inode in the case of file write and mmap writes.
6014 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6017 struct btrfs_root *root = BTRFS_I(inode)->root;
6018 bool dirty = flags & ~S_VERSION;
6020 if (btrfs_root_readonly(root))
6023 if (flags & S_VERSION)
6024 dirty |= inode_maybe_inc_iversion(inode, dirty);
6025 if (flags & S_CTIME)
6026 inode->i_ctime = *now;
6027 if (flags & S_MTIME)
6028 inode->i_mtime = *now;
6029 if (flags & S_ATIME)
6030 inode->i_atime = *now;
6031 return dirty ? btrfs_dirty_inode(inode) : 0;
6035 * find the highest existing sequence number in a directory
6036 * and then set the in-memory index_cnt variable to reflect
6037 * free sequence numbers
6039 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6041 struct btrfs_root *root = inode->root;
6042 struct btrfs_key key, found_key;
6043 struct btrfs_path *path;
6044 struct extent_buffer *leaf;
6047 key.objectid = btrfs_ino(inode);
6048 key.type = BTRFS_DIR_INDEX_KEY;
6049 key.offset = (u64)-1;
6051 path = btrfs_alloc_path();
6055 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6058 /* FIXME: we should be able to handle this */
6064 * MAGIC NUMBER EXPLANATION:
6065 * since we search a directory based on f_pos we have to start at 2
6066 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6067 * else has to start at 2
6069 if (path->slots[0] == 0) {
6070 inode->index_cnt = 2;
6076 leaf = path->nodes[0];
6077 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6079 if (found_key.objectid != btrfs_ino(inode) ||
6080 found_key.type != BTRFS_DIR_INDEX_KEY) {
6081 inode->index_cnt = 2;
6085 inode->index_cnt = found_key.offset + 1;
6087 btrfs_free_path(path);
6092 * helper to find a free sequence number in a given directory. This current
6093 * code is very simple, later versions will do smarter things in the btree
6095 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6099 if (dir->index_cnt == (u64)-1) {
6100 ret = btrfs_inode_delayed_dir_index_count(dir);
6102 ret = btrfs_set_inode_index_count(dir);
6108 *index = dir->index_cnt;
6114 static int btrfs_insert_inode_locked(struct inode *inode)
6116 struct btrfs_iget_args args;
6117 args.location = &BTRFS_I(inode)->location;
6118 args.root = BTRFS_I(inode)->root;
6120 return insert_inode_locked4(inode,
6121 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6122 btrfs_find_actor, &args);
6126 * Inherit flags from the parent inode.
6128 * Currently only the compression flags and the cow flags are inherited.
6130 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6137 flags = BTRFS_I(dir)->flags;
6139 if (flags & BTRFS_INODE_NOCOMPRESS) {
6140 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6141 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6142 } else if (flags & BTRFS_INODE_COMPRESS) {
6143 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6144 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6147 if (flags & BTRFS_INODE_NODATACOW) {
6148 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6149 if (S_ISREG(inode->i_mode))
6150 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6153 btrfs_sync_inode_flags_to_i_flags(inode);
6156 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6157 struct btrfs_root *root,
6159 const char *name, int name_len,
6160 u64 ref_objectid, u64 objectid,
6161 umode_t mode, u64 *index)
6163 struct btrfs_fs_info *fs_info = root->fs_info;
6164 struct inode *inode;
6165 struct btrfs_inode_item *inode_item;
6166 struct btrfs_key *location;
6167 struct btrfs_path *path;
6168 struct btrfs_inode_ref *ref;
6169 struct btrfs_key key[2];
6171 int nitems = name ? 2 : 1;
6175 path = btrfs_alloc_path();
6177 return ERR_PTR(-ENOMEM);
6179 inode = new_inode(fs_info->sb);
6181 btrfs_free_path(path);
6182 return ERR_PTR(-ENOMEM);
6186 * O_TMPFILE, set link count to 0, so that after this point,
6187 * we fill in an inode item with the correct link count.
6190 set_nlink(inode, 0);
6193 * we have to initialize this early, so we can reclaim the inode
6194 * number if we fail afterwards in this function.
6196 inode->i_ino = objectid;
6199 trace_btrfs_inode_request(dir);
6201 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6203 btrfs_free_path(path);
6205 return ERR_PTR(ret);
6211 * index_cnt is ignored for everything but a dir,
6212 * btrfs_set_inode_index_count has an explanation for the magic
6215 BTRFS_I(inode)->index_cnt = 2;
6216 BTRFS_I(inode)->dir_index = *index;
6217 BTRFS_I(inode)->root = root;
6218 BTRFS_I(inode)->generation = trans->transid;
6219 inode->i_generation = BTRFS_I(inode)->generation;
6222 * We could have gotten an inode number from somebody who was fsynced
6223 * and then removed in this same transaction, so let's just set full
6224 * sync since it will be a full sync anyway and this will blow away the
6225 * old info in the log.
6227 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6229 key[0].objectid = objectid;
6230 key[0].type = BTRFS_INODE_ITEM_KEY;
6233 sizes[0] = sizeof(struct btrfs_inode_item);
6237 * Start new inodes with an inode_ref. This is slightly more
6238 * efficient for small numbers of hard links since they will
6239 * be packed into one item. Extended refs will kick in if we
6240 * add more hard links than can fit in the ref item.
6242 key[1].objectid = objectid;
6243 key[1].type = BTRFS_INODE_REF_KEY;
6244 key[1].offset = ref_objectid;
6246 sizes[1] = name_len + sizeof(*ref);
6249 location = &BTRFS_I(inode)->location;
6250 location->objectid = objectid;
6251 location->offset = 0;
6252 location->type = BTRFS_INODE_ITEM_KEY;
6254 ret = btrfs_insert_inode_locked(inode);
6260 path->leave_spinning = 1;
6261 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6265 inode_init_owner(inode, dir, mode);
6266 inode_set_bytes(inode, 0);
6268 inode->i_mtime = current_time(inode);
6269 inode->i_atime = inode->i_mtime;
6270 inode->i_ctime = inode->i_mtime;
6271 BTRFS_I(inode)->i_otime = inode->i_mtime;
6273 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6274 struct btrfs_inode_item);
6275 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6276 sizeof(*inode_item));
6277 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6280 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6281 struct btrfs_inode_ref);
6282 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6283 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6284 ptr = (unsigned long)(ref + 1);
6285 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6288 btrfs_mark_buffer_dirty(path->nodes[0]);
6289 btrfs_free_path(path);
6291 btrfs_inherit_iflags(inode, dir);
6293 if (S_ISREG(mode)) {
6294 if (btrfs_test_opt(fs_info, NODATASUM))
6295 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6296 if (btrfs_test_opt(fs_info, NODATACOW))
6297 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6298 BTRFS_INODE_NODATASUM;
6301 inode_tree_add(inode);
6303 trace_btrfs_inode_new(inode);
6304 btrfs_set_inode_last_trans(trans, inode);
6306 btrfs_update_root_times(trans, root);
6308 ret = btrfs_inode_inherit_props(trans, inode, dir);
6311 "error inheriting props for ino %llu (root %llu): %d",
6312 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6317 discard_new_inode(inode);
6320 BTRFS_I(dir)->index_cnt--;
6321 btrfs_free_path(path);
6322 return ERR_PTR(ret);
6325 static inline u8 btrfs_inode_type(struct inode *inode)
6327 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6331 * utility function to add 'inode' into 'parent_inode' with
6332 * a give name and a given sequence number.
6333 * if 'add_backref' is true, also insert a backref from the
6334 * inode to the parent directory.
6336 int btrfs_add_link(struct btrfs_trans_handle *trans,
6337 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6338 const char *name, int name_len, int add_backref, u64 index)
6341 struct btrfs_key key;
6342 struct btrfs_root *root = parent_inode->root;
6343 u64 ino = btrfs_ino(inode);
6344 u64 parent_ino = btrfs_ino(parent_inode);
6346 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6347 memcpy(&key, &inode->root->root_key, sizeof(key));
6350 key.type = BTRFS_INODE_ITEM_KEY;
6354 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6355 ret = btrfs_add_root_ref(trans, key.objectid,
6356 root->root_key.objectid, parent_ino,
6357 index, name, name_len);
6358 } else if (add_backref) {
6359 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6363 /* Nothing to clean up yet */
6367 ret = btrfs_insert_dir_item(trans, name, name_len, parent_inode, &key,
6368 btrfs_inode_type(&inode->vfs_inode), index);
6369 if (ret == -EEXIST || ret == -EOVERFLOW)
6372 btrfs_abort_transaction(trans, ret);
6376 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6378 inode_inc_iversion(&parent_inode->vfs_inode);
6379 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6380 current_time(&parent_inode->vfs_inode);
6381 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6383 btrfs_abort_transaction(trans, ret);
6387 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6390 err = btrfs_del_root_ref(trans, key.objectid,
6391 root->root_key.objectid, parent_ino,
6392 &local_index, name, name_len);
6394 } else if (add_backref) {
6398 err = btrfs_del_inode_ref(trans, root, name, name_len,
6399 ino, parent_ino, &local_index);
6404 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6405 struct btrfs_inode *dir, struct dentry *dentry,
6406 struct btrfs_inode *inode, int backref, u64 index)
6408 int err = btrfs_add_link(trans, dir, inode,
6409 dentry->d_name.name, dentry->d_name.len,
6416 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6417 umode_t mode, dev_t rdev)
6419 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6420 struct btrfs_trans_handle *trans;
6421 struct btrfs_root *root = BTRFS_I(dir)->root;
6422 struct inode *inode = NULL;
6428 * 2 for inode item and ref
6430 * 1 for xattr if selinux is on
6432 trans = btrfs_start_transaction(root, 5);
6434 return PTR_ERR(trans);
6436 err = btrfs_find_free_ino(root, &objectid);
6440 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6441 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6443 if (IS_ERR(inode)) {
6444 err = PTR_ERR(inode);
6450 * If the active LSM wants to access the inode during
6451 * d_instantiate it needs these. Smack checks to see
6452 * if the filesystem supports xattrs by looking at the
6455 inode->i_op = &btrfs_special_inode_operations;
6456 init_special_inode(inode, inode->i_mode, rdev);
6458 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6462 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6467 btrfs_update_inode(trans, root, inode);
6468 d_instantiate_new(dentry, inode);
6471 btrfs_end_transaction(trans);
6472 btrfs_btree_balance_dirty(fs_info);
6474 inode_dec_link_count(inode);
6475 discard_new_inode(inode);
6480 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6481 umode_t mode, bool excl)
6483 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6484 struct btrfs_trans_handle *trans;
6485 struct btrfs_root *root = BTRFS_I(dir)->root;
6486 struct inode *inode = NULL;
6492 * 2 for inode item and ref
6494 * 1 for xattr if selinux is on
6496 trans = btrfs_start_transaction(root, 5);
6498 return PTR_ERR(trans);
6500 err = btrfs_find_free_ino(root, &objectid);
6504 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6505 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6507 if (IS_ERR(inode)) {
6508 err = PTR_ERR(inode);
6513 * If the active LSM wants to access the inode during
6514 * d_instantiate it needs these. Smack checks to see
6515 * if the filesystem supports xattrs by looking at the
6518 inode->i_fop = &btrfs_file_operations;
6519 inode->i_op = &btrfs_file_inode_operations;
6520 inode->i_mapping->a_ops = &btrfs_aops;
6522 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6526 err = btrfs_update_inode(trans, root, inode);
6530 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6535 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6536 d_instantiate_new(dentry, inode);
6539 btrfs_end_transaction(trans);
6541 inode_dec_link_count(inode);
6542 discard_new_inode(inode);
6544 btrfs_btree_balance_dirty(fs_info);
6548 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6549 struct dentry *dentry)
6551 struct btrfs_trans_handle *trans = NULL;
6552 struct btrfs_root *root = BTRFS_I(dir)->root;
6553 struct inode *inode = d_inode(old_dentry);
6554 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6559 /* do not allow sys_link's with other subvols of the same device */
6560 if (root->root_key.objectid != BTRFS_I(inode)->root->root_key.objectid)
6563 if (inode->i_nlink >= BTRFS_LINK_MAX)
6566 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6571 * 2 items for inode and inode ref
6572 * 2 items for dir items
6573 * 1 item for parent inode
6574 * 1 item for orphan item deletion if O_TMPFILE
6576 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6577 if (IS_ERR(trans)) {
6578 err = PTR_ERR(trans);
6583 /* There are several dir indexes for this inode, clear the cache. */
6584 BTRFS_I(inode)->dir_index = 0ULL;
6586 inode_inc_iversion(inode);
6587 inode->i_ctime = current_time(inode);
6589 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6591 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6597 struct dentry *parent = dentry->d_parent;
6600 err = btrfs_update_inode(trans, root, inode);
6603 if (inode->i_nlink == 1) {
6605 * If new hard link count is 1, it's a file created
6606 * with open(2) O_TMPFILE flag.
6608 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6612 d_instantiate(dentry, inode);
6613 ret = btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent,
6615 if (ret == BTRFS_NEED_TRANS_COMMIT) {
6616 err = btrfs_commit_transaction(trans);
6623 btrfs_end_transaction(trans);
6625 inode_dec_link_count(inode);
6628 btrfs_btree_balance_dirty(fs_info);
6632 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6634 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6635 struct inode *inode = NULL;
6636 struct btrfs_trans_handle *trans;
6637 struct btrfs_root *root = BTRFS_I(dir)->root;
6639 int drop_on_err = 0;
6644 * 2 items for inode and ref
6645 * 2 items for dir items
6646 * 1 for xattr if selinux is on
6648 trans = btrfs_start_transaction(root, 5);
6650 return PTR_ERR(trans);
6652 err = btrfs_find_free_ino(root, &objectid);
6656 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6657 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6658 S_IFDIR | mode, &index);
6659 if (IS_ERR(inode)) {
6660 err = PTR_ERR(inode);
6666 /* these must be set before we unlock the inode */
6667 inode->i_op = &btrfs_dir_inode_operations;
6668 inode->i_fop = &btrfs_dir_file_operations;
6670 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6674 btrfs_i_size_write(BTRFS_I(inode), 0);
6675 err = btrfs_update_inode(trans, root, inode);
6679 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6680 dentry->d_name.name,
6681 dentry->d_name.len, 0, index);
6685 d_instantiate_new(dentry, inode);
6689 btrfs_end_transaction(trans);
6691 inode_dec_link_count(inode);
6692 discard_new_inode(inode);
6694 btrfs_btree_balance_dirty(fs_info);
6698 static noinline int uncompress_inline(struct btrfs_path *path,
6700 size_t pg_offset, u64 extent_offset,
6701 struct btrfs_file_extent_item *item)
6704 struct extent_buffer *leaf = path->nodes[0];
6707 unsigned long inline_size;
6711 WARN_ON(pg_offset != 0);
6712 compress_type = btrfs_file_extent_compression(leaf, item);
6713 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6714 inline_size = btrfs_file_extent_inline_item_len(leaf,
6715 btrfs_item_nr(path->slots[0]));
6716 tmp = kmalloc(inline_size, GFP_NOFS);
6719 ptr = btrfs_file_extent_inline_start(item);
6721 read_extent_buffer(leaf, tmp, ptr, inline_size);
6723 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6724 ret = btrfs_decompress(compress_type, tmp, page,
6725 extent_offset, inline_size, max_size);
6728 * decompression code contains a memset to fill in any space between the end
6729 * of the uncompressed data and the end of max_size in case the decompressed
6730 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6731 * the end of an inline extent and the beginning of the next block, so we
6732 * cover that region here.
6735 if (max_size + pg_offset < PAGE_SIZE) {
6736 char *map = kmap(page);
6737 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6745 * a bit scary, this does extent mapping from logical file offset to the disk.
6746 * the ugly parts come from merging extents from the disk with the in-ram
6747 * representation. This gets more complex because of the data=ordered code,
6748 * where the in-ram extents might be locked pending data=ordered completion.
6750 * This also copies inline extents directly into the page.
6752 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6754 size_t pg_offset, u64 start, u64 len,
6757 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6760 u64 extent_start = 0;
6762 u64 objectid = btrfs_ino(inode);
6764 struct btrfs_path *path = NULL;
6765 struct btrfs_root *root = inode->root;
6766 struct btrfs_file_extent_item *item;
6767 struct extent_buffer *leaf;
6768 struct btrfs_key found_key;
6769 struct extent_map *em = NULL;
6770 struct extent_map_tree *em_tree = &inode->extent_tree;
6771 struct extent_io_tree *io_tree = &inode->io_tree;
6772 const bool new_inline = !page || create;
6774 read_lock(&em_tree->lock);
6775 em = lookup_extent_mapping(em_tree, start, len);
6777 em->bdev = 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 = 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;
6799 path = btrfs_alloc_path();
6805 /* Chances are we'll be called again, so go ahead and do readahead */
6806 path->reada = READA_FORWARD;
6809 * Unless we're going to uncompress the inline extent, no sleep would
6812 path->leave_spinning = 1;
6814 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6821 if (path->slots[0] == 0)
6826 leaf = path->nodes[0];
6827 item = btrfs_item_ptr(leaf, path->slots[0],
6828 struct btrfs_file_extent_item);
6829 /* are we inside the extent that was found? */
6830 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6831 found_type = found_key.type;
6832 if (found_key.objectid != objectid ||
6833 found_type != BTRFS_EXTENT_DATA_KEY) {
6835 * If we backup past the first extent we want to move forward
6836 * and see if there is an extent in front of us, otherwise we'll
6837 * say there is a hole for our whole search range which can
6844 found_type = btrfs_file_extent_type(leaf, item);
6845 extent_start = found_key.offset;
6846 if (found_type == BTRFS_FILE_EXTENT_REG ||
6847 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6848 extent_end = extent_start +
6849 btrfs_file_extent_num_bytes(leaf, item);
6851 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6853 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6856 size = btrfs_file_extent_ram_bytes(leaf, item);
6857 extent_end = ALIGN(extent_start + size,
6858 fs_info->sectorsize);
6860 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6865 if (start >= extent_end) {
6867 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6868 ret = btrfs_next_leaf(root, path);
6875 leaf = path->nodes[0];
6877 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6878 if (found_key.objectid != objectid ||
6879 found_key.type != BTRFS_EXTENT_DATA_KEY)
6881 if (start + len <= found_key.offset)
6883 if (start > found_key.offset)
6886 em->orig_start = start;
6887 em->len = found_key.offset - start;
6891 btrfs_extent_item_to_extent_map(inode, path, item,
6894 if (found_type == BTRFS_FILE_EXTENT_REG ||
6895 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6897 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6901 size_t extent_offset;
6907 size = btrfs_file_extent_ram_bytes(leaf, item);
6908 extent_offset = page_offset(page) + pg_offset - extent_start;
6909 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
6910 size - extent_offset);
6911 em->start = extent_start + extent_offset;
6912 em->len = ALIGN(copy_size, fs_info->sectorsize);
6913 em->orig_block_len = em->len;
6914 em->orig_start = em->start;
6915 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6917 btrfs_set_path_blocking(path);
6918 if (!PageUptodate(page)) {
6919 if (btrfs_file_extent_compression(leaf, item) !=
6920 BTRFS_COMPRESS_NONE) {
6921 ret = uncompress_inline(path, page, pg_offset,
6922 extent_offset, item);
6929 read_extent_buffer(leaf, map + pg_offset, ptr,
6931 if (pg_offset + copy_size < PAGE_SIZE) {
6932 memset(map + pg_offset + copy_size, 0,
6933 PAGE_SIZE - pg_offset -
6938 flush_dcache_page(page);
6940 set_extent_uptodate(io_tree, em->start,
6941 extent_map_end(em) - 1, NULL, GFP_NOFS);
6946 em->orig_start = start;
6949 em->block_start = EXTENT_MAP_HOLE;
6951 btrfs_release_path(path);
6952 if (em->start > start || extent_map_end(em) <= start) {
6954 "bad extent! em: [%llu %llu] passed [%llu %llu]",
6955 em->start, em->len, start, len);
6961 write_lock(&em_tree->lock);
6962 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
6963 write_unlock(&em_tree->lock);
6965 btrfs_free_path(path);
6967 trace_btrfs_get_extent(root, inode, em);
6970 free_extent_map(em);
6971 return ERR_PTR(err);
6973 BUG_ON(!em); /* Error is always set */
6977 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
6979 size_t pg_offset, u64 start, u64 len,
6982 struct extent_map *em;
6983 struct extent_map *hole_em = NULL;
6984 u64 range_start = start;
6990 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6994 * If our em maps to:
6996 * - a pre-alloc extent,
6997 * there might actually be delalloc bytes behind it.
6999 if (em->block_start != EXTENT_MAP_HOLE &&
7000 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7005 /* check to see if we've wrapped (len == -1 or similar) */
7014 /* ok, we didn't find anything, lets look for delalloc */
7015 found = count_range_bits(&inode->io_tree, &range_start,
7016 end, len, EXTENT_DELALLOC, 1);
7017 found_end = range_start + found;
7018 if (found_end < range_start)
7019 found_end = (u64)-1;
7022 * we didn't find anything useful, return
7023 * the original results from get_extent()
7025 if (range_start > end || found_end <= start) {
7031 /* adjust the range_start to make sure it doesn't
7032 * go backwards from the start they passed in
7034 range_start = max(start, range_start);
7035 found = found_end - range_start;
7038 u64 hole_start = start;
7041 em = alloc_extent_map();
7047 * when btrfs_get_extent can't find anything it
7048 * returns one huge hole
7050 * make sure what it found really fits our range, and
7051 * adjust to make sure it is based on the start from
7055 u64 calc_end = extent_map_end(hole_em);
7057 if (calc_end <= start || (hole_em->start > end)) {
7058 free_extent_map(hole_em);
7061 hole_start = max(hole_em->start, start);
7062 hole_len = calc_end - hole_start;
7066 if (hole_em && range_start > hole_start) {
7067 /* our hole starts before our delalloc, so we
7068 * have to return just the parts of the hole
7069 * that go until the delalloc starts
7071 em->len = min(hole_len,
7072 range_start - hole_start);
7073 em->start = hole_start;
7074 em->orig_start = hole_start;
7076 * don't adjust block start at all,
7077 * it is fixed at EXTENT_MAP_HOLE
7079 em->block_start = hole_em->block_start;
7080 em->block_len = hole_len;
7081 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7082 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7084 em->start = range_start;
7086 em->orig_start = range_start;
7087 em->block_start = EXTENT_MAP_DELALLOC;
7088 em->block_len = found;
7095 free_extent_map(hole_em);
7097 free_extent_map(em);
7098 return ERR_PTR(err);
7103 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7106 const u64 orig_start,
7107 const u64 block_start,
7108 const u64 block_len,
7109 const u64 orig_block_len,
7110 const u64 ram_bytes,
7113 struct extent_map *em = NULL;
7116 if (type != BTRFS_ORDERED_NOCOW) {
7117 em = create_io_em(inode, start, len, orig_start,
7118 block_start, block_len, orig_block_len,
7120 BTRFS_COMPRESS_NONE, /* compress_type */
7125 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7126 len, block_len, type);
7129 free_extent_map(em);
7130 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7131 start + len - 1, 0);
7140 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7143 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7144 struct btrfs_root *root = BTRFS_I(inode)->root;
7145 struct extent_map *em;
7146 struct btrfs_key ins;
7150 alloc_hint = get_extent_allocation_hint(inode, start, len);
7151 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7152 0, alloc_hint, &ins, 1, 1);
7154 return ERR_PTR(ret);
7156 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7157 ins.objectid, ins.offset, ins.offset,
7158 ins.offset, BTRFS_ORDERED_REGULAR);
7159 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7161 btrfs_free_reserved_extent(fs_info, ins.objectid,
7168 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7169 * block must be cow'd
7171 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7172 u64 *orig_start, u64 *orig_block_len,
7175 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7176 struct btrfs_path *path;
7178 struct extent_buffer *leaf;
7179 struct btrfs_root *root = BTRFS_I(inode)->root;
7180 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7181 struct btrfs_file_extent_item *fi;
7182 struct btrfs_key key;
7189 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7191 path = btrfs_alloc_path();
7195 ret = btrfs_lookup_file_extent(NULL, root, path,
7196 btrfs_ino(BTRFS_I(inode)), offset, 0);
7200 slot = path->slots[0];
7203 /* can't find the item, must cow */
7210 leaf = path->nodes[0];
7211 btrfs_item_key_to_cpu(leaf, &key, slot);
7212 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7213 key.type != BTRFS_EXTENT_DATA_KEY) {
7214 /* not our file or wrong item type, must cow */
7218 if (key.offset > offset) {
7219 /* Wrong offset, must cow */
7223 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7224 found_type = btrfs_file_extent_type(leaf, fi);
7225 if (found_type != BTRFS_FILE_EXTENT_REG &&
7226 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7227 /* not a regular extent, must cow */
7231 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7234 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7235 if (extent_end <= offset)
7238 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7239 if (disk_bytenr == 0)
7242 if (btrfs_file_extent_compression(leaf, fi) ||
7243 btrfs_file_extent_encryption(leaf, fi) ||
7244 btrfs_file_extent_other_encoding(leaf, fi))
7248 * Do the same check as in btrfs_cross_ref_exist but without the
7249 * unnecessary search.
7251 if (btrfs_file_extent_generation(leaf, fi) <=
7252 btrfs_root_last_snapshot(&root->root_item))
7255 backref_offset = btrfs_file_extent_offset(leaf, fi);
7258 *orig_start = key.offset - backref_offset;
7259 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7260 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7263 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7266 num_bytes = min(offset + *len, extent_end) - offset;
7267 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7270 range_end = round_up(offset + num_bytes,
7271 root->fs_info->sectorsize) - 1;
7272 ret = test_range_bit(io_tree, offset, range_end,
7273 EXTENT_DELALLOC, 0, NULL);
7280 btrfs_release_path(path);
7283 * look for other files referencing this extent, if we
7284 * find any we must cow
7287 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7288 key.offset - backref_offset, disk_bytenr);
7295 * adjust disk_bytenr and num_bytes to cover just the bytes
7296 * in this extent we are about to write. If there
7297 * are any csums in that range we have to cow in order
7298 * to keep the csums correct
7300 disk_bytenr += backref_offset;
7301 disk_bytenr += offset - key.offset;
7302 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7305 * all of the above have passed, it is safe to overwrite this extent
7311 btrfs_free_path(path);
7315 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7316 struct extent_state **cached_state, int writing)
7318 struct btrfs_ordered_extent *ordered;
7322 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7325 * We're concerned with the entire range that we're going to be
7326 * doing DIO to, so we need to make sure there's no ordered
7327 * extents in this range.
7329 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7330 lockend - lockstart + 1);
7333 * We need to make sure there are no buffered pages in this
7334 * range either, we could have raced between the invalidate in
7335 * generic_file_direct_write and locking the extent. The
7336 * invalidate needs to happen so that reads after a write do not
7340 (!writing || !filemap_range_has_page(inode->i_mapping,
7341 lockstart, lockend)))
7344 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7349 * If we are doing a DIO read and the ordered extent we
7350 * found is for a buffered write, we can not wait for it
7351 * to complete and retry, because if we do so we can
7352 * deadlock with concurrent buffered writes on page
7353 * locks. This happens only if our DIO read covers more
7354 * than one extent map, if at this point has already
7355 * created an ordered extent for a previous extent map
7356 * and locked its range in the inode's io tree, and a
7357 * concurrent write against that previous extent map's
7358 * range and this range started (we unlock the ranges
7359 * in the io tree only when the bios complete and
7360 * buffered writes always lock pages before attempting
7361 * to lock range in the io tree).
7364 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7365 btrfs_start_ordered_extent(inode, ordered, 1);
7368 btrfs_put_ordered_extent(ordered);
7371 * We could trigger writeback for this range (and wait
7372 * for it to complete) and then invalidate the pages for
7373 * this range (through invalidate_inode_pages2_range()),
7374 * but that can lead us to a deadlock with a concurrent
7375 * call to readpages() (a buffered read or a defrag call
7376 * triggered a readahead) on a page lock due to an
7377 * ordered dio extent we created before but did not have
7378 * yet a corresponding bio submitted (whence it can not
7379 * complete), which makes readpages() wait for that
7380 * ordered extent to complete while holding a lock on
7395 /* The callers of this must take lock_extent() */
7396 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7397 u64 orig_start, u64 block_start,
7398 u64 block_len, u64 orig_block_len,
7399 u64 ram_bytes, int compress_type,
7402 struct extent_map_tree *em_tree;
7403 struct extent_map *em;
7404 struct btrfs_root *root = BTRFS_I(inode)->root;
7407 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7408 type == BTRFS_ORDERED_COMPRESSED ||
7409 type == BTRFS_ORDERED_NOCOW ||
7410 type == BTRFS_ORDERED_REGULAR);
7412 em_tree = &BTRFS_I(inode)->extent_tree;
7413 em = alloc_extent_map();
7415 return ERR_PTR(-ENOMEM);
7418 em->orig_start = orig_start;
7420 em->block_len = block_len;
7421 em->block_start = block_start;
7422 em->bdev = root->fs_info->fs_devices->latest_bdev;
7423 em->orig_block_len = orig_block_len;
7424 em->ram_bytes = ram_bytes;
7425 em->generation = -1;
7426 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7427 if (type == BTRFS_ORDERED_PREALLOC) {
7428 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7429 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7430 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7431 em->compress_type = compress_type;
7435 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7436 em->start + em->len - 1, 0);
7437 write_lock(&em_tree->lock);
7438 ret = add_extent_mapping(em_tree, em, 1);
7439 write_unlock(&em_tree->lock);
7441 * The caller has taken lock_extent(), who could race with us
7444 } while (ret == -EEXIST);
7447 free_extent_map(em);
7448 return ERR_PTR(ret);
7451 /* em got 2 refs now, callers needs to do free_extent_map once. */
7456 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7457 struct buffer_head *bh_result,
7458 struct inode *inode,
7461 if (em->block_start == EXTENT_MAP_HOLE ||
7462 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7465 len = min(len, em->len - (start - em->start));
7467 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7469 bh_result->b_size = len;
7470 bh_result->b_bdev = em->bdev;
7471 set_buffer_mapped(bh_result);
7476 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7477 struct buffer_head *bh_result,
7478 struct inode *inode,
7479 struct btrfs_dio_data *dio_data,
7482 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7483 struct extent_map *em = *map;
7487 * We don't allocate a new extent in the following cases
7489 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7491 * 2) The extent is marked as PREALLOC. We're good to go here and can
7492 * just use the extent.
7495 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7496 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7497 em->block_start != EXTENT_MAP_HOLE)) {
7499 u64 block_start, orig_start, orig_block_len, ram_bytes;
7501 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7502 type = BTRFS_ORDERED_PREALLOC;
7504 type = BTRFS_ORDERED_NOCOW;
7505 len = min(len, em->len - (start - em->start));
7506 block_start = em->block_start + (start - em->start);
7508 if (can_nocow_extent(inode, start, &len, &orig_start,
7509 &orig_block_len, &ram_bytes) == 1 &&
7510 btrfs_inc_nocow_writers(fs_info, block_start)) {
7511 struct extent_map *em2;
7513 em2 = btrfs_create_dio_extent(inode, start, len,
7514 orig_start, block_start,
7515 len, orig_block_len,
7517 btrfs_dec_nocow_writers(fs_info, block_start);
7518 if (type == BTRFS_ORDERED_PREALLOC) {
7519 free_extent_map(em);
7523 if (em2 && IS_ERR(em2)) {
7528 * For inode marked NODATACOW or extent marked PREALLOC,
7529 * use the existing or preallocated extent, so does not
7530 * need to adjust btrfs_space_info's bytes_may_use.
7532 btrfs_free_reserved_data_space_noquota(inode, start,
7538 /* this will cow the extent */
7539 len = bh_result->b_size;
7540 free_extent_map(em);
7541 *map = em = btrfs_new_extent_direct(inode, start, len);
7547 len = min(len, em->len - (start - em->start));
7550 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7552 bh_result->b_size = len;
7553 bh_result->b_bdev = em->bdev;
7554 set_buffer_mapped(bh_result);
7556 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7557 set_buffer_new(bh_result);
7560 * Need to update the i_size under the extent lock so buffered
7561 * readers will get the updated i_size when we unlock.
7563 if (!dio_data->overwrite && start + len > i_size_read(inode))
7564 i_size_write(inode, start + len);
7566 WARN_ON(dio_data->reserve < len);
7567 dio_data->reserve -= len;
7568 dio_data->unsubmitted_oe_range_end = start + len;
7569 current->journal_info = dio_data;
7574 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7575 struct buffer_head *bh_result, int create)
7577 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7578 struct extent_map *em;
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, fs_info->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 transaction 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(BTRFS_I(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 that's 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);
7643 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7644 dio_data, start, len);
7648 /* clear and unlock the entire range */
7649 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7650 unlock_bits, 1, 0, &cached_state);
7652 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7654 /* Can be negative only if we read from a hole */
7657 free_extent_map(em);
7661 * We need to unlock only the end area that we aren't using.
7662 * The rest is going to be unlocked by the endio routine.
7664 lockstart = start + bh_result->b_size;
7665 if (lockstart < lockend) {
7666 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7667 lockend, unlock_bits, 1, 0,
7670 free_extent_state(cached_state);
7674 free_extent_map(em);
7679 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7680 unlock_bits, 1, 0, &cached_state);
7683 current->journal_info = dio_data;
7687 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7691 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7694 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7696 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7700 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7705 static int btrfs_check_dio_repairable(struct inode *inode,
7706 struct bio *failed_bio,
7707 struct io_failure_record *failrec,
7710 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7713 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7714 if (num_copies == 1) {
7716 * we only have a single copy of the data, so don't bother with
7717 * all the retry and error correction code that follows. no
7718 * matter what the error is, it is very likely to persist.
7720 btrfs_debug(fs_info,
7721 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7722 num_copies, failrec->this_mirror, failed_mirror);
7726 failrec->failed_mirror = failed_mirror;
7727 failrec->this_mirror++;
7728 if (failrec->this_mirror == failed_mirror)
7729 failrec->this_mirror++;
7731 if (failrec->this_mirror > num_copies) {
7732 btrfs_debug(fs_info,
7733 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7734 num_copies, failrec->this_mirror, failed_mirror);
7741 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7742 struct page *page, unsigned int pgoff,
7743 u64 start, u64 end, int failed_mirror,
7744 bio_end_io_t *repair_endio, void *repair_arg)
7746 struct io_failure_record *failrec;
7747 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7748 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7751 unsigned int read_mode = 0;
7754 blk_status_t status;
7755 struct bio_vec bvec;
7757 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7759 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7761 return errno_to_blk_status(ret);
7763 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7766 free_io_failure(failure_tree, io_tree, failrec);
7767 return BLK_STS_IOERR;
7770 segs = bio_segments(failed_bio);
7771 bio_get_first_bvec(failed_bio, &bvec);
7773 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7774 read_mode |= REQ_FAILFAST_DEV;
7776 isector = start - btrfs_io_bio(failed_bio)->logical;
7777 isector >>= inode->i_sb->s_blocksize_bits;
7778 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7779 pgoff, isector, repair_endio, repair_arg);
7780 bio->bi_opf = REQ_OP_READ | read_mode;
7782 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7783 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7784 read_mode, failrec->this_mirror, failrec->in_validation);
7786 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7788 free_io_failure(failure_tree, io_tree, failrec);
7795 struct btrfs_retry_complete {
7796 struct completion done;
7797 struct inode *inode;
7802 static void btrfs_retry_endio_nocsum(struct bio *bio)
7804 struct btrfs_retry_complete *done = bio->bi_private;
7805 struct inode *inode = done->inode;
7806 struct bio_vec *bvec;
7807 struct extent_io_tree *io_tree, *failure_tree;
7813 ASSERT(bio->bi_vcnt == 1);
7814 io_tree = &BTRFS_I(inode)->io_tree;
7815 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7816 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7819 ASSERT(!bio_flagged(bio, BIO_CLONED));
7820 bio_for_each_segment_all(bvec, bio, i)
7821 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7822 io_tree, done->start, bvec->bv_page,
7823 btrfs_ino(BTRFS_I(inode)), 0);
7825 complete(&done->done);
7829 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7830 struct btrfs_io_bio *io_bio)
7832 struct btrfs_fs_info *fs_info;
7833 struct bio_vec bvec;
7834 struct bvec_iter iter;
7835 struct btrfs_retry_complete done;
7841 blk_status_t err = BLK_STS_OK;
7843 fs_info = BTRFS_I(inode)->root->fs_info;
7844 sectorsize = fs_info->sectorsize;
7846 start = io_bio->logical;
7848 io_bio->bio.bi_iter = io_bio->iter;
7850 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7851 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7852 pgoff = bvec.bv_offset;
7854 next_block_or_try_again:
7857 init_completion(&done.done);
7859 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7860 pgoff, start, start + sectorsize - 1,
7862 btrfs_retry_endio_nocsum, &done);
7868 wait_for_completion_io(&done.done);
7870 if (!done.uptodate) {
7871 /* We might have another mirror, so try again */
7872 goto next_block_or_try_again;
7876 start += sectorsize;
7880 pgoff += sectorsize;
7881 ASSERT(pgoff < PAGE_SIZE);
7882 goto next_block_or_try_again;
7889 static void btrfs_retry_endio(struct bio *bio)
7891 struct btrfs_retry_complete *done = bio->bi_private;
7892 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7893 struct extent_io_tree *io_tree, *failure_tree;
7894 struct inode *inode = done->inode;
7895 struct bio_vec *bvec;
7905 ASSERT(bio->bi_vcnt == 1);
7906 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7908 io_tree = &BTRFS_I(inode)->io_tree;
7909 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7911 ASSERT(!bio_flagged(bio, BIO_CLONED));
7912 bio_for_each_segment_all(bvec, bio, i) {
7913 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7914 bvec->bv_offset, done->start,
7917 clean_io_failure(BTRFS_I(inode)->root->fs_info,
7918 failure_tree, io_tree, done->start,
7920 btrfs_ino(BTRFS_I(inode)),
7926 done->uptodate = uptodate;
7928 complete(&done->done);
7932 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
7933 struct btrfs_io_bio *io_bio, blk_status_t err)
7935 struct btrfs_fs_info *fs_info;
7936 struct bio_vec bvec;
7937 struct bvec_iter iter;
7938 struct btrfs_retry_complete done;
7945 bool uptodate = (err == 0);
7947 blk_status_t status;
7949 fs_info = BTRFS_I(inode)->root->fs_info;
7950 sectorsize = fs_info->sectorsize;
7953 start = io_bio->logical;
7955 io_bio->bio.bi_iter = io_bio->iter;
7957 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7958 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7960 pgoff = bvec.bv_offset;
7963 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
7964 ret = __readpage_endio_check(inode, io_bio, csum_pos,
7965 bvec.bv_page, pgoff, start, sectorsize);
7972 init_completion(&done.done);
7974 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7975 pgoff, start, start + sectorsize - 1,
7976 io_bio->mirror_num, btrfs_retry_endio,
7983 wait_for_completion_io(&done.done);
7985 if (!done.uptodate) {
7986 /* We might have another mirror, so try again */
7990 offset += sectorsize;
7991 start += sectorsize;
7997 pgoff += sectorsize;
7998 ASSERT(pgoff < PAGE_SIZE);
8006 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8007 struct btrfs_io_bio *io_bio, blk_status_t err)
8009 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8013 return __btrfs_correct_data_nocsum(inode, io_bio);
8017 return __btrfs_subio_endio_read(inode, io_bio, err);
8021 static void btrfs_endio_direct_read(struct bio *bio)
8023 struct btrfs_dio_private *dip = bio->bi_private;
8024 struct inode *inode = dip->inode;
8025 struct bio *dio_bio;
8026 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8027 blk_status_t err = bio->bi_status;
8029 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8030 err = btrfs_subio_endio_read(inode, io_bio, err);
8032 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8033 dip->logical_offset + dip->bytes - 1);
8034 dio_bio = dip->dio_bio;
8038 dio_bio->bi_status = err;
8039 dio_end_io(dio_bio);
8042 io_bio->end_io(io_bio, blk_status_to_errno(err));
8046 static void __endio_write_update_ordered(struct inode *inode,
8047 const u64 offset, const u64 bytes,
8048 const bool uptodate)
8050 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8051 struct btrfs_ordered_extent *ordered = NULL;
8052 struct btrfs_workqueue *wq;
8053 btrfs_work_func_t func;
8054 u64 ordered_offset = offset;
8055 u64 ordered_bytes = bytes;
8058 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8059 wq = fs_info->endio_freespace_worker;
8060 func = btrfs_freespace_write_helper;
8062 wq = fs_info->endio_write_workers;
8063 func = btrfs_endio_write_helper;
8066 while (ordered_offset < offset + bytes) {
8067 last_offset = ordered_offset;
8068 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8072 btrfs_init_work(&ordered->work, func,
8075 btrfs_queue_work(wq, &ordered->work);
8078 * If btrfs_dec_test_ordered_pending does not find any ordered
8079 * extent in the range, we can exit.
8081 if (ordered_offset == last_offset)
8084 * Our bio might span multiple ordered extents. In this case
8085 * we keep goin until we have accounted the whole dio.
8087 if (ordered_offset < offset + bytes) {
8088 ordered_bytes = offset + bytes - ordered_offset;
8094 static void btrfs_endio_direct_write(struct bio *bio)
8096 struct btrfs_dio_private *dip = bio->bi_private;
8097 struct bio *dio_bio = dip->dio_bio;
8099 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8100 dip->bytes, !bio->bi_status);
8104 dio_bio->bi_status = bio->bi_status;
8105 dio_end_io(dio_bio);
8109 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8110 struct bio *bio, u64 offset)
8112 struct inode *inode = private_data;
8114 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8115 BUG_ON(ret); /* -ENOMEM */
8119 static void btrfs_end_dio_bio(struct bio *bio)
8121 struct btrfs_dio_private *dip = bio->bi_private;
8122 blk_status_t err = bio->bi_status;
8125 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8126 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8127 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8129 (unsigned long long)bio->bi_iter.bi_sector,
8130 bio->bi_iter.bi_size, err);
8132 if (dip->subio_endio)
8133 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8137 * We want to perceive the errors flag being set before
8138 * decrementing the reference count. We don't need a barrier
8139 * since atomic operations with a return value are fully
8140 * ordered as per atomic_t.txt
8145 /* if there are more bios still pending for this dio, just exit */
8146 if (!atomic_dec_and_test(&dip->pending_bios))
8150 bio_io_error(dip->orig_bio);
8152 dip->dio_bio->bi_status = BLK_STS_OK;
8153 bio_endio(dip->orig_bio);
8159 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8160 struct btrfs_dio_private *dip,
8164 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8165 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8169 * We load all the csum data we need when we submit
8170 * the first bio to reduce the csum tree search and
8173 if (dip->logical_offset == file_offset) {
8174 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8180 if (bio == dip->orig_bio)
8183 file_offset -= dip->logical_offset;
8184 file_offset >>= inode->i_sb->s_blocksize_bits;
8185 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8190 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8191 struct inode *inode, u64 file_offset, int async_submit)
8193 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8194 struct btrfs_dio_private *dip = bio->bi_private;
8195 bool write = bio_op(bio) == REQ_OP_WRITE;
8198 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8200 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8203 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8208 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8211 if (write && async_submit) {
8212 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8214 btrfs_submit_bio_start_direct_io);
8218 * If we aren't doing async submit, calculate the csum of the
8221 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8225 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8231 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8236 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8238 struct inode *inode = dip->inode;
8239 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8241 struct bio *orig_bio = dip->orig_bio;
8242 u64 start_sector = orig_bio->bi_iter.bi_sector;
8243 u64 file_offset = dip->logical_offset;
8245 int async_submit = 0;
8247 int clone_offset = 0;
8250 blk_status_t status;
8252 map_length = orig_bio->bi_iter.bi_size;
8253 submit_len = map_length;
8254 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8255 &map_length, NULL, 0);
8259 if (map_length >= submit_len) {
8261 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8265 /* async crcs make it difficult to collect full stripe writes. */
8266 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8272 ASSERT(map_length <= INT_MAX);
8273 atomic_inc(&dip->pending_bios);
8275 clone_len = min_t(int, submit_len, map_length);
8278 * This will never fail as it's passing GPF_NOFS and
8279 * the allocation is backed by btrfs_bioset.
8281 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8283 bio->bi_private = dip;
8284 bio->bi_end_io = btrfs_end_dio_bio;
8285 btrfs_io_bio(bio)->logical = file_offset;
8287 ASSERT(submit_len >= clone_len);
8288 submit_len -= clone_len;
8289 if (submit_len == 0)
8293 * Increase the count before we submit the bio so we know
8294 * the end IO handler won't happen before we increase the
8295 * count. Otherwise, the dip might get freed before we're
8296 * done setting it up.
8298 atomic_inc(&dip->pending_bios);
8300 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8304 atomic_dec(&dip->pending_bios);
8308 clone_offset += clone_len;
8309 start_sector += clone_len >> 9;
8310 file_offset += clone_len;
8312 map_length = submit_len;
8313 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8314 start_sector << 9, &map_length, NULL, 0);
8317 } while (submit_len > 0);
8320 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8328 * Before atomic variable goto zero, we must make sure dip->errors is
8329 * perceived to be set. This ordering is ensured by the fact that an
8330 * atomic operations with a return value are fully ordered as per
8333 if (atomic_dec_and_test(&dip->pending_bios))
8334 bio_io_error(dip->orig_bio);
8336 /* bio_end_io() will handle error, so we needn't return it */
8340 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8343 struct btrfs_dio_private *dip = NULL;
8344 struct bio *bio = NULL;
8345 struct btrfs_io_bio *io_bio;
8346 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8349 bio = btrfs_bio_clone(dio_bio);
8351 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8357 dip->private = dio_bio->bi_private;
8359 dip->logical_offset = file_offset;
8360 dip->bytes = dio_bio->bi_iter.bi_size;
8361 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8362 bio->bi_private = dip;
8363 dip->orig_bio = bio;
8364 dip->dio_bio = dio_bio;
8365 atomic_set(&dip->pending_bios, 0);
8366 io_bio = btrfs_io_bio(bio);
8367 io_bio->logical = file_offset;
8370 bio->bi_end_io = btrfs_endio_direct_write;
8372 bio->bi_end_io = btrfs_endio_direct_read;
8373 dip->subio_endio = btrfs_subio_endio_read;
8377 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8378 * even if we fail to submit a bio, because in such case we do the
8379 * corresponding error handling below and it must not be done a second
8380 * time by btrfs_direct_IO().
8383 struct btrfs_dio_data *dio_data = current->journal_info;
8385 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8387 dio_data->unsubmitted_oe_range_start =
8388 dio_data->unsubmitted_oe_range_end;
8391 ret = btrfs_submit_direct_hook(dip);
8396 io_bio->end_io(io_bio, ret);
8400 * If we arrived here it means either we failed to submit the dip
8401 * or we either failed to clone the dio_bio or failed to allocate the
8402 * dip. If we cloned the dio_bio and allocated the dip, we can just
8403 * call bio_endio against our io_bio so that we get proper resource
8404 * cleanup if we fail to submit the dip, otherwise, we must do the
8405 * same as btrfs_endio_direct_[write|read] because we can't call these
8406 * callbacks - they require an allocated dip and a clone of dio_bio.
8411 * The end io callbacks free our dip, do the final put on bio
8412 * and all the cleanup and final put for dio_bio (through
8419 __endio_write_update_ordered(inode,
8421 dio_bio->bi_iter.bi_size,
8424 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8425 file_offset + dio_bio->bi_iter.bi_size - 1);
8427 dio_bio->bi_status = BLK_STS_IOERR;
8429 * Releases and cleans up our dio_bio, no need to bio_put()
8430 * nor bio_endio()/bio_io_error() against dio_bio.
8432 dio_end_io(dio_bio);
8439 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8440 const struct iov_iter *iter, loff_t offset)
8444 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8445 ssize_t retval = -EINVAL;
8447 if (offset & blocksize_mask)
8450 if (iov_iter_alignment(iter) & blocksize_mask)
8453 /* If this is a write we don't need to check anymore */
8454 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8457 * Check to make sure we don't have duplicate iov_base's in this
8458 * iovec, if so return EINVAL, otherwise we'll get csum errors
8459 * when reading back.
8461 for (seg = 0; seg < iter->nr_segs; seg++) {
8462 for (i = seg + 1; i < iter->nr_segs; i++) {
8463 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8472 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8474 struct file *file = iocb->ki_filp;
8475 struct inode *inode = file->f_mapping->host;
8476 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8477 struct btrfs_dio_data dio_data = { 0 };
8478 struct extent_changeset *data_reserved = NULL;
8479 loff_t offset = iocb->ki_pos;
8483 bool relock = false;
8486 if (check_direct_IO(fs_info, iter, offset))
8489 inode_dio_begin(inode);
8492 * The generic stuff only does filemap_write_and_wait_range, which
8493 * isn't enough if we've written compressed pages to this area, so
8494 * we need to flush the dirty pages again to make absolutely sure
8495 * that any outstanding dirty pages are on disk.
8497 count = iov_iter_count(iter);
8498 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8499 &BTRFS_I(inode)->runtime_flags))
8500 filemap_fdatawrite_range(inode->i_mapping, offset,
8501 offset + count - 1);
8503 if (iov_iter_rw(iter) == WRITE) {
8505 * If the write DIO is beyond the EOF, we need update
8506 * the isize, but it is protected by i_mutex. So we can
8507 * not unlock the i_mutex at this case.
8509 if (offset + count <= inode->i_size) {
8510 dio_data.overwrite = 1;
8511 inode_unlock(inode);
8513 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8517 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8523 * We need to know how many extents we reserved so that we can
8524 * do the accounting properly if we go over the number we
8525 * originally calculated. Abuse current->journal_info for this.
8527 dio_data.reserve = round_up(count,
8528 fs_info->sectorsize);
8529 dio_data.unsubmitted_oe_range_start = (u64)offset;
8530 dio_data.unsubmitted_oe_range_end = (u64)offset;
8531 current->journal_info = &dio_data;
8532 down_read(&BTRFS_I(inode)->dio_sem);
8533 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8534 &BTRFS_I(inode)->runtime_flags)) {
8535 inode_dio_end(inode);
8536 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8540 ret = __blockdev_direct_IO(iocb, inode,
8541 fs_info->fs_devices->latest_bdev,
8542 iter, btrfs_get_blocks_direct, NULL,
8543 btrfs_submit_direct, flags);
8544 if (iov_iter_rw(iter) == WRITE) {
8545 up_read(&BTRFS_I(inode)->dio_sem);
8546 current->journal_info = NULL;
8547 if (ret < 0 && ret != -EIOCBQUEUED) {
8548 if (dio_data.reserve)
8549 btrfs_delalloc_release_space(inode, data_reserved,
8550 offset, dio_data.reserve, true);
8552 * On error we might have left some ordered extents
8553 * without submitting corresponding bios for them, so
8554 * cleanup them up to avoid other tasks getting them
8555 * and waiting for them to complete forever.
8557 if (dio_data.unsubmitted_oe_range_start <
8558 dio_data.unsubmitted_oe_range_end)
8559 __endio_write_update_ordered(inode,
8560 dio_data.unsubmitted_oe_range_start,
8561 dio_data.unsubmitted_oe_range_end -
8562 dio_data.unsubmitted_oe_range_start,
8564 } else if (ret >= 0 && (size_t)ret < count)
8565 btrfs_delalloc_release_space(inode, data_reserved,
8566 offset, count - (size_t)ret, true);
8567 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8571 inode_dio_end(inode);
8575 extent_changeset_free(data_reserved);
8579 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8581 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8582 __u64 start, __u64 len)
8586 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8590 return extent_fiemap(inode, fieinfo, start, len);
8593 int btrfs_readpage(struct file *file, struct page *page)
8595 struct extent_io_tree *tree;
8596 tree = &BTRFS_I(page->mapping->host)->io_tree;
8597 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8600 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8602 struct inode *inode = page->mapping->host;
8605 if (current->flags & PF_MEMALLOC) {
8606 redirty_page_for_writepage(wbc, page);
8612 * If we are under memory pressure we will call this directly from the
8613 * VM, we need to make sure we have the inode referenced for the ordered
8614 * extent. If not just return like we didn't do anything.
8616 if (!igrab(inode)) {
8617 redirty_page_for_writepage(wbc, page);
8618 return AOP_WRITEPAGE_ACTIVATE;
8620 ret = extent_write_full_page(page, wbc);
8621 btrfs_add_delayed_iput(inode);
8625 static int btrfs_writepages(struct address_space *mapping,
8626 struct writeback_control *wbc)
8628 return extent_writepages(mapping, wbc);
8632 btrfs_readpages(struct file *file, struct address_space *mapping,
8633 struct list_head *pages, unsigned nr_pages)
8635 return extent_readpages(mapping, pages, nr_pages);
8638 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8640 int ret = try_release_extent_mapping(page, gfp_flags);
8642 ClearPagePrivate(page);
8643 set_page_private(page, 0);
8649 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8651 if (PageWriteback(page) || PageDirty(page))
8653 return __btrfs_releasepage(page, gfp_flags);
8656 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8657 unsigned int length)
8659 struct inode *inode = page->mapping->host;
8660 struct extent_io_tree *tree;
8661 struct btrfs_ordered_extent *ordered;
8662 struct extent_state *cached_state = NULL;
8663 u64 page_start = page_offset(page);
8664 u64 page_end = page_start + PAGE_SIZE - 1;
8667 int inode_evicting = inode->i_state & I_FREEING;
8670 * we have the page locked, so new writeback can't start,
8671 * and the dirty bit won't be cleared while we are here.
8673 * Wait for IO on this page so that we can safely clear
8674 * the PagePrivate2 bit and do ordered accounting
8676 wait_on_page_writeback(page);
8678 tree = &BTRFS_I(inode)->io_tree;
8680 btrfs_releasepage(page, GFP_NOFS);
8684 if (!inode_evicting)
8685 lock_extent_bits(tree, page_start, page_end, &cached_state);
8688 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8689 page_end - start + 1);
8691 end = min(page_end, ordered->file_offset + ordered->len - 1);
8693 * IO on this page will never be started, so we need
8694 * to account for any ordered extents now
8696 if (!inode_evicting)
8697 clear_extent_bit(tree, start, end,
8698 EXTENT_DIRTY | EXTENT_DELALLOC |
8699 EXTENT_DELALLOC_NEW |
8700 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8701 EXTENT_DEFRAG, 1, 0, &cached_state);
8703 * whoever cleared the private bit is responsible
8704 * for the finish_ordered_io
8706 if (TestClearPagePrivate2(page)) {
8707 struct btrfs_ordered_inode_tree *tree;
8710 tree = &BTRFS_I(inode)->ordered_tree;
8712 spin_lock_irq(&tree->lock);
8713 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8714 new_len = start - ordered->file_offset;
8715 if (new_len < ordered->truncated_len)
8716 ordered->truncated_len = new_len;
8717 spin_unlock_irq(&tree->lock);
8719 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8721 end - start + 1, 1))
8722 btrfs_finish_ordered_io(ordered);
8724 btrfs_put_ordered_extent(ordered);
8725 if (!inode_evicting) {
8726 cached_state = NULL;
8727 lock_extent_bits(tree, start, end,
8732 if (start < page_end)
8737 * Qgroup reserved space handler
8738 * Page here will be either
8739 * 1) Already written to disk
8740 * In this case, its reserved space is released from data rsv map
8741 * and will be freed by delayed_ref handler finally.
8742 * So even we call qgroup_free_data(), it won't decrease reserved
8744 * 2) Not written to disk
8745 * This means the reserved space should be freed here. However,
8746 * if a truncate invalidates the page (by clearing PageDirty)
8747 * and the page is accounted for while allocating extent
8748 * in btrfs_check_data_free_space() we let delayed_ref to
8749 * free the entire extent.
8751 if (PageDirty(page))
8752 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8753 if (!inode_evicting) {
8754 clear_extent_bit(tree, page_start, page_end,
8755 EXTENT_LOCKED | EXTENT_DIRTY |
8756 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8757 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8760 __btrfs_releasepage(page, GFP_NOFS);
8763 ClearPageChecked(page);
8764 if (PagePrivate(page)) {
8765 ClearPagePrivate(page);
8766 set_page_private(page, 0);
8772 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8773 * called from a page fault handler when a page is first dirtied. Hence we must
8774 * be careful to check for EOF conditions here. We set the page up correctly
8775 * for a written page which means we get ENOSPC checking when writing into
8776 * holes and correct delalloc and unwritten extent mapping on filesystems that
8777 * support these features.
8779 * We are not allowed to take the i_mutex here so we have to play games to
8780 * protect against truncate races as the page could now be beyond EOF. Because
8781 * truncate_setsize() writes the inode size before removing pages, once we have
8782 * the page lock we can determine safely if the page is beyond EOF. If it is not
8783 * beyond EOF, then the page is guaranteed safe against truncation until we
8786 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8788 struct page *page = vmf->page;
8789 struct inode *inode = file_inode(vmf->vma->vm_file);
8790 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8791 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8792 struct btrfs_ordered_extent *ordered;
8793 struct extent_state *cached_state = NULL;
8794 struct extent_changeset *data_reserved = NULL;
8796 unsigned long zero_start;
8806 reserved_space = PAGE_SIZE;
8808 sb_start_pagefault(inode->i_sb);
8809 page_start = page_offset(page);
8810 page_end = page_start + PAGE_SIZE - 1;
8814 * Reserving delalloc space after obtaining the page lock can lead to
8815 * deadlock. For example, if a dirty page is locked by this function
8816 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8817 * dirty page write out, then the btrfs_writepage() function could
8818 * end up waiting indefinitely to get a lock on the page currently
8819 * being processed by btrfs_page_mkwrite() function.
8821 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8824 ret2 = file_update_time(vmf->vma->vm_file);
8828 ret = vmf_error(ret2);
8834 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8837 size = i_size_read(inode);
8839 if ((page->mapping != inode->i_mapping) ||
8840 (page_start >= size)) {
8841 /* page got truncated out from underneath us */
8844 wait_on_page_writeback(page);
8846 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8847 set_page_extent_mapped(page);
8850 * we can't set the delalloc bits if there are pending ordered
8851 * extents. Drop our locks and wait for them to finish
8853 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8856 unlock_extent_cached(io_tree, page_start, page_end,
8859 btrfs_start_ordered_extent(inode, ordered, 1);
8860 btrfs_put_ordered_extent(ordered);
8864 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8865 reserved_space = round_up(size - page_start,
8866 fs_info->sectorsize);
8867 if (reserved_space < PAGE_SIZE) {
8868 end = page_start + reserved_space - 1;
8869 btrfs_delalloc_release_space(inode, data_reserved,
8870 page_start, PAGE_SIZE - reserved_space,
8876 * page_mkwrite gets called when the page is firstly dirtied after it's
8877 * faulted in, but write(2) could also dirty a page and set delalloc
8878 * bits, thus in this case for space account reason, we still need to
8879 * clear any delalloc bits within this page range since we have to
8880 * reserve data&meta space before lock_page() (see above comments).
8882 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8883 EXTENT_DIRTY | EXTENT_DELALLOC |
8884 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8885 0, 0, &cached_state);
8887 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8890 unlock_extent_cached(io_tree, page_start, page_end,
8892 ret = VM_FAULT_SIGBUS;
8897 /* page is wholly or partially inside EOF */
8898 if (page_start + PAGE_SIZE > size)
8899 zero_start = size & ~PAGE_MASK;
8901 zero_start = PAGE_SIZE;
8903 if (zero_start != PAGE_SIZE) {
8905 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8906 flush_dcache_page(page);
8909 ClearPageChecked(page);
8910 set_page_dirty(page);
8911 SetPageUptodate(page);
8913 BTRFS_I(inode)->last_trans = fs_info->generation;
8914 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8915 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8917 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
8920 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
8921 sb_end_pagefault(inode->i_sb);
8922 extent_changeset_free(data_reserved);
8923 return VM_FAULT_LOCKED;
8929 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
8930 btrfs_delalloc_release_space(inode, data_reserved, page_start,
8931 reserved_space, (ret != 0));
8933 sb_end_pagefault(inode->i_sb);
8934 extent_changeset_free(data_reserved);
8938 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
8940 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8941 struct btrfs_root *root = BTRFS_I(inode)->root;
8942 struct btrfs_block_rsv *rsv;
8944 struct btrfs_trans_handle *trans;
8945 u64 mask = fs_info->sectorsize - 1;
8946 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
8948 if (!skip_writeback) {
8949 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8956 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
8957 * things going on here:
8959 * 1) We need to reserve space to update our inode.
8961 * 2) We need to have something to cache all the space that is going to
8962 * be free'd up by the truncate operation, but also have some slack
8963 * space reserved in case it uses space during the truncate (thank you
8964 * very much snapshotting).
8966 * And we need these to be separate. The fact is we can use a lot of
8967 * space doing the truncate, and we have no earthly idea how much space
8968 * we will use, so we need the truncate reservation to be separate so it
8969 * doesn't end up using space reserved for updating the inode. We also
8970 * need to be able to stop the transaction and start a new one, which
8971 * means we need to be able to update the inode several times, and we
8972 * have no idea of knowing how many times that will be, so we can't just
8973 * reserve 1 item for the entirety of the operation, so that has to be
8974 * done separately as well.
8976 * So that leaves us with
8978 * 1) rsv - for the truncate reservation, which we will steal from the
8979 * transaction reservation.
8980 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
8981 * updating the inode.
8983 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
8986 rsv->size = min_size;
8990 * 1 for the truncate slack space
8991 * 1 for updating the inode.
8993 trans = btrfs_start_transaction(root, 2);
8994 if (IS_ERR(trans)) {
8995 ret = PTR_ERR(trans);
8999 /* Migrate the slack space for the truncate to our reserve */
9000 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9005 * So if we truncate and then write and fsync we normally would just
9006 * write the extents that changed, which is a problem if we need to
9007 * first truncate that entire inode. So set this flag so we write out
9008 * all of the extents in the inode to the sync log so we're completely
9011 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9012 trans->block_rsv = rsv;
9015 ret = btrfs_truncate_inode_items(trans, root, inode,
9017 BTRFS_EXTENT_DATA_KEY);
9018 trans->block_rsv = &fs_info->trans_block_rsv;
9019 if (ret != -ENOSPC && ret != -EAGAIN)
9022 ret = btrfs_update_inode(trans, root, inode);
9026 btrfs_end_transaction(trans);
9027 btrfs_btree_balance_dirty(fs_info);
9029 trans = btrfs_start_transaction(root, 2);
9030 if (IS_ERR(trans)) {
9031 ret = PTR_ERR(trans);
9036 btrfs_block_rsv_release(fs_info, rsv, -1);
9037 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9038 rsv, min_size, false);
9039 BUG_ON(ret); /* shouldn't happen */
9040 trans->block_rsv = rsv;
9044 * We can't call btrfs_truncate_block inside a trans handle as we could
9045 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9046 * we've truncated everything except the last little bit, and can do
9047 * btrfs_truncate_block and then update the disk_i_size.
9049 if (ret == NEED_TRUNCATE_BLOCK) {
9050 btrfs_end_transaction(trans);
9051 btrfs_btree_balance_dirty(fs_info);
9053 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9056 trans = btrfs_start_transaction(root, 1);
9057 if (IS_ERR(trans)) {
9058 ret = PTR_ERR(trans);
9061 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9067 trans->block_rsv = &fs_info->trans_block_rsv;
9068 ret2 = btrfs_update_inode(trans, root, inode);
9072 ret2 = btrfs_end_transaction(trans);
9075 btrfs_btree_balance_dirty(fs_info);
9078 btrfs_free_block_rsv(fs_info, rsv);
9084 * create a new subvolume directory/inode (helper for the ioctl).
9086 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9087 struct btrfs_root *new_root,
9088 struct btrfs_root *parent_root,
9091 struct inode *inode;
9095 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9096 new_dirid, new_dirid,
9097 S_IFDIR | (~current_umask() & S_IRWXUGO),
9100 return PTR_ERR(inode);
9101 inode->i_op = &btrfs_dir_inode_operations;
9102 inode->i_fop = &btrfs_dir_file_operations;
9104 set_nlink(inode, 1);
9105 btrfs_i_size_write(BTRFS_I(inode), 0);
9106 unlock_new_inode(inode);
9108 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9110 btrfs_err(new_root->fs_info,
9111 "error inheriting subvolume %llu properties: %d",
9112 new_root->root_key.objectid, err);
9114 err = btrfs_update_inode(trans, new_root, inode);
9120 struct inode *btrfs_alloc_inode(struct super_block *sb)
9122 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9123 struct btrfs_inode *ei;
9124 struct inode *inode;
9126 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9133 ei->last_sub_trans = 0;
9134 ei->logged_trans = 0;
9135 ei->delalloc_bytes = 0;
9136 ei->new_delalloc_bytes = 0;
9137 ei->defrag_bytes = 0;
9138 ei->disk_i_size = 0;
9141 ei->index_cnt = (u64)-1;
9143 ei->last_unlink_trans = 0;
9144 ei->last_log_commit = 0;
9146 spin_lock_init(&ei->lock);
9147 ei->outstanding_extents = 0;
9148 if (sb->s_magic != BTRFS_TEST_MAGIC)
9149 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9150 BTRFS_BLOCK_RSV_DELALLOC);
9151 ei->runtime_flags = 0;
9152 ei->prop_compress = BTRFS_COMPRESS_NONE;
9153 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9155 ei->delayed_node = NULL;
9157 ei->i_otime.tv_sec = 0;
9158 ei->i_otime.tv_nsec = 0;
9160 inode = &ei->vfs_inode;
9161 extent_map_tree_init(&ei->extent_tree);
9162 extent_io_tree_init(&ei->io_tree, inode);
9163 extent_io_tree_init(&ei->io_failure_tree, inode);
9164 ei->io_tree.track_uptodate = 1;
9165 ei->io_failure_tree.track_uptodate = 1;
9166 atomic_set(&ei->sync_writers, 0);
9167 mutex_init(&ei->log_mutex);
9168 mutex_init(&ei->delalloc_mutex);
9169 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9170 INIT_LIST_HEAD(&ei->delalloc_inodes);
9171 INIT_LIST_HEAD(&ei->delayed_iput);
9172 RB_CLEAR_NODE(&ei->rb_node);
9173 init_rwsem(&ei->dio_sem);
9178 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9179 void btrfs_test_destroy_inode(struct inode *inode)
9181 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9182 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9186 static void btrfs_i_callback(struct rcu_head *head)
9188 struct inode *inode = container_of(head, struct inode, i_rcu);
9189 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9192 void btrfs_destroy_inode(struct inode *inode)
9194 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9195 struct btrfs_ordered_extent *ordered;
9196 struct btrfs_root *root = BTRFS_I(inode)->root;
9198 WARN_ON(!hlist_empty(&inode->i_dentry));
9199 WARN_ON(inode->i_data.nrpages);
9200 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9201 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9202 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9203 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9204 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9205 WARN_ON(BTRFS_I(inode)->csum_bytes);
9206 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9209 * This can happen where we create an inode, but somebody else also
9210 * created the same inode and we need to destroy the one we already
9217 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9222 "found ordered extent %llu %llu on inode cleanup",
9223 ordered->file_offset, ordered->len);
9224 btrfs_remove_ordered_extent(inode, ordered);
9225 btrfs_put_ordered_extent(ordered);
9226 btrfs_put_ordered_extent(ordered);
9229 btrfs_qgroup_check_reserved_leak(inode);
9230 inode_tree_del(inode);
9231 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9233 call_rcu(&inode->i_rcu, btrfs_i_callback);
9236 int btrfs_drop_inode(struct inode *inode)
9238 struct btrfs_root *root = BTRFS_I(inode)->root;
9243 /* the snap/subvol tree is on deleting */
9244 if (btrfs_root_refs(&root->root_item) == 0)
9247 return generic_drop_inode(inode);
9250 static void init_once(void *foo)
9252 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9254 inode_init_once(&ei->vfs_inode);
9257 void __cold btrfs_destroy_cachep(void)
9260 * Make sure all delayed rcu free inodes are flushed before we
9264 kmem_cache_destroy(btrfs_inode_cachep);
9265 kmem_cache_destroy(btrfs_trans_handle_cachep);
9266 kmem_cache_destroy(btrfs_path_cachep);
9267 kmem_cache_destroy(btrfs_free_space_cachep);
9270 int __init btrfs_init_cachep(void)
9272 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9273 sizeof(struct btrfs_inode), 0,
9274 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9276 if (!btrfs_inode_cachep)
9279 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9280 sizeof(struct btrfs_trans_handle), 0,
9281 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9282 if (!btrfs_trans_handle_cachep)
9285 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9286 sizeof(struct btrfs_path), 0,
9287 SLAB_MEM_SPREAD, NULL);
9288 if (!btrfs_path_cachep)
9291 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9292 sizeof(struct btrfs_free_space), 0,
9293 SLAB_MEM_SPREAD, NULL);
9294 if (!btrfs_free_space_cachep)
9299 btrfs_destroy_cachep();
9303 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9304 u32 request_mask, unsigned int flags)
9307 struct inode *inode = d_inode(path->dentry);
9308 u32 blocksize = inode->i_sb->s_blocksize;
9309 u32 bi_flags = BTRFS_I(inode)->flags;
9311 stat->result_mask |= STATX_BTIME;
9312 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9313 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9314 if (bi_flags & BTRFS_INODE_APPEND)
9315 stat->attributes |= STATX_ATTR_APPEND;
9316 if (bi_flags & BTRFS_INODE_COMPRESS)
9317 stat->attributes |= STATX_ATTR_COMPRESSED;
9318 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9319 stat->attributes |= STATX_ATTR_IMMUTABLE;
9320 if (bi_flags & BTRFS_INODE_NODUMP)
9321 stat->attributes |= STATX_ATTR_NODUMP;
9323 stat->attributes_mask |= (STATX_ATTR_APPEND |
9324 STATX_ATTR_COMPRESSED |
9325 STATX_ATTR_IMMUTABLE |
9328 generic_fillattr(inode, stat);
9329 stat->dev = BTRFS_I(inode)->root->anon_dev;
9331 spin_lock(&BTRFS_I(inode)->lock);
9332 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9333 spin_unlock(&BTRFS_I(inode)->lock);
9334 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9335 ALIGN(delalloc_bytes, blocksize)) >> 9;
9339 static int btrfs_rename_exchange(struct inode *old_dir,
9340 struct dentry *old_dentry,
9341 struct inode *new_dir,
9342 struct dentry *new_dentry)
9344 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9345 struct btrfs_trans_handle *trans;
9346 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9347 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9348 struct inode *new_inode = new_dentry->d_inode;
9349 struct inode *old_inode = old_dentry->d_inode;
9350 struct timespec64 ctime = current_time(old_inode);
9351 struct dentry *parent;
9352 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9353 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9358 bool root_log_pinned = false;
9359 bool dest_log_pinned = false;
9360 struct btrfs_log_ctx ctx_root;
9361 struct btrfs_log_ctx ctx_dest;
9362 bool sync_log_root = false;
9363 bool sync_log_dest = false;
9364 bool commit_transaction = false;
9366 /* we only allow rename subvolume link between subvolumes */
9367 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9370 btrfs_init_log_ctx(&ctx_root, old_inode);
9371 btrfs_init_log_ctx(&ctx_dest, new_inode);
9373 /* close the race window with snapshot create/destroy ioctl */
9374 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9375 down_read(&fs_info->subvol_sem);
9376 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9377 down_read(&fs_info->subvol_sem);
9380 * We want to reserve the absolute worst case amount of items. So if
9381 * both inodes are subvols and we need to unlink them then that would
9382 * require 4 item modifications, but if they are both normal inodes it
9383 * would require 5 item modifications, so we'll assume their normal
9384 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9385 * should cover the worst case number of items we'll modify.
9387 trans = btrfs_start_transaction(root, 12);
9388 if (IS_ERR(trans)) {
9389 ret = PTR_ERR(trans);
9394 * We need to find a free sequence number both in the source and
9395 * in the destination directory for the exchange.
9397 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9400 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9404 BTRFS_I(old_inode)->dir_index = 0ULL;
9405 BTRFS_I(new_inode)->dir_index = 0ULL;
9407 /* Reference for the source. */
9408 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9409 /* force full log commit if subvolume involved. */
9410 btrfs_set_log_full_commit(fs_info, trans);
9412 btrfs_pin_log_trans(root);
9413 root_log_pinned = true;
9414 ret = btrfs_insert_inode_ref(trans, dest,
9415 new_dentry->d_name.name,
9416 new_dentry->d_name.len,
9418 btrfs_ino(BTRFS_I(new_dir)),
9424 /* And now for the dest. */
9425 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9426 /* force full log commit if subvolume involved. */
9427 btrfs_set_log_full_commit(fs_info, trans);
9429 btrfs_pin_log_trans(dest);
9430 dest_log_pinned = true;
9431 ret = btrfs_insert_inode_ref(trans, root,
9432 old_dentry->d_name.name,
9433 old_dentry->d_name.len,
9435 btrfs_ino(BTRFS_I(old_dir)),
9441 /* Update inode version and ctime/mtime. */
9442 inode_inc_iversion(old_dir);
9443 inode_inc_iversion(new_dir);
9444 inode_inc_iversion(old_inode);
9445 inode_inc_iversion(new_inode);
9446 old_dir->i_ctime = old_dir->i_mtime = ctime;
9447 new_dir->i_ctime = new_dir->i_mtime = ctime;
9448 old_inode->i_ctime = ctime;
9449 new_inode->i_ctime = ctime;
9451 if (old_dentry->d_parent != new_dentry->d_parent) {
9452 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9453 BTRFS_I(old_inode), 1);
9454 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9455 BTRFS_I(new_inode), 1);
9458 /* src is a subvolume */
9459 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9460 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9461 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9462 old_dentry->d_name.name,
9463 old_dentry->d_name.len);
9464 } else { /* src is an inode */
9465 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9466 BTRFS_I(old_dentry->d_inode),
9467 old_dentry->d_name.name,
9468 old_dentry->d_name.len);
9470 ret = btrfs_update_inode(trans, root, old_inode);
9473 btrfs_abort_transaction(trans, ret);
9477 /* dest is a subvolume */
9478 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9479 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9480 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9481 new_dentry->d_name.name,
9482 new_dentry->d_name.len);
9483 } else { /* dest is an inode */
9484 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9485 BTRFS_I(new_dentry->d_inode),
9486 new_dentry->d_name.name,
9487 new_dentry->d_name.len);
9489 ret = btrfs_update_inode(trans, dest, new_inode);
9492 btrfs_abort_transaction(trans, ret);
9496 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9497 new_dentry->d_name.name,
9498 new_dentry->d_name.len, 0, old_idx);
9500 btrfs_abort_transaction(trans, ret);
9504 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9505 old_dentry->d_name.name,
9506 old_dentry->d_name.len, 0, new_idx);
9508 btrfs_abort_transaction(trans, ret);
9512 if (old_inode->i_nlink == 1)
9513 BTRFS_I(old_inode)->dir_index = old_idx;
9514 if (new_inode->i_nlink == 1)
9515 BTRFS_I(new_inode)->dir_index = new_idx;
9517 if (root_log_pinned) {
9518 parent = new_dentry->d_parent;
9519 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9520 BTRFS_I(old_dir), parent,
9522 if (ret == BTRFS_NEED_LOG_SYNC)
9523 sync_log_root = true;
9524 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9525 commit_transaction = true;
9527 btrfs_end_log_trans(root);
9528 root_log_pinned = false;
9530 if (dest_log_pinned) {
9531 if (!commit_transaction) {
9532 parent = old_dentry->d_parent;
9533 ret = btrfs_log_new_name(trans, BTRFS_I(new_inode),
9534 BTRFS_I(new_dir), parent,
9536 if (ret == BTRFS_NEED_LOG_SYNC)
9537 sync_log_dest = true;
9538 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9539 commit_transaction = true;
9542 btrfs_end_log_trans(dest);
9543 dest_log_pinned = false;
9547 * If we have pinned a log and an error happened, we unpin tasks
9548 * trying to sync the log and force them to fallback to a transaction
9549 * commit if the log currently contains any of the inodes involved in
9550 * this rename operation (to ensure we do not persist a log with an
9551 * inconsistent state for any of these inodes or leading to any
9552 * inconsistencies when replayed). If the transaction was aborted, the
9553 * abortion reason is propagated to userspace when attempting to commit
9554 * the transaction. If the log does not contain any of these inodes, we
9555 * allow the tasks to sync it.
9557 if (ret && (root_log_pinned || dest_log_pinned)) {
9558 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9559 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9560 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9562 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9563 btrfs_set_log_full_commit(fs_info, trans);
9565 if (root_log_pinned) {
9566 btrfs_end_log_trans(root);
9567 root_log_pinned = false;
9569 if (dest_log_pinned) {
9570 btrfs_end_log_trans(dest);
9571 dest_log_pinned = false;
9574 if (!ret && sync_log_root && !commit_transaction) {
9575 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root,
9578 commit_transaction = true;
9580 if (!ret && sync_log_dest && !commit_transaction) {
9581 ret = btrfs_sync_log(trans, BTRFS_I(new_inode)->root,
9584 commit_transaction = true;
9586 if (commit_transaction) {
9587 ret = btrfs_commit_transaction(trans);
9591 ret2 = btrfs_end_transaction(trans);
9592 ret = ret ? ret : ret2;
9595 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9596 up_read(&fs_info->subvol_sem);
9597 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9598 up_read(&fs_info->subvol_sem);
9603 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9604 struct btrfs_root *root,
9606 struct dentry *dentry)
9609 struct inode *inode;
9613 ret = btrfs_find_free_ino(root, &objectid);
9617 inode = btrfs_new_inode(trans, root, dir,
9618 dentry->d_name.name,
9620 btrfs_ino(BTRFS_I(dir)),
9622 S_IFCHR | WHITEOUT_MODE,
9625 if (IS_ERR(inode)) {
9626 ret = PTR_ERR(inode);
9630 inode->i_op = &btrfs_special_inode_operations;
9631 init_special_inode(inode, inode->i_mode,
9634 ret = btrfs_init_inode_security(trans, inode, dir,
9639 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9640 BTRFS_I(inode), 0, index);
9644 ret = btrfs_update_inode(trans, root, inode);
9646 unlock_new_inode(inode);
9648 inode_dec_link_count(inode);
9654 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9655 struct inode *new_dir, struct dentry *new_dentry,
9658 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9659 struct btrfs_trans_handle *trans;
9660 unsigned int trans_num_items;
9661 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9662 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9663 struct inode *new_inode = d_inode(new_dentry);
9664 struct inode *old_inode = d_inode(old_dentry);
9668 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9669 bool log_pinned = false;
9670 struct btrfs_log_ctx ctx;
9671 bool sync_log = false;
9672 bool commit_transaction = false;
9674 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9677 /* we only allow rename subvolume link between subvolumes */
9678 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9681 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9682 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9685 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9686 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9690 /* check for collisions, even if the name isn't there */
9691 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9692 new_dentry->d_name.name,
9693 new_dentry->d_name.len);
9696 if (ret == -EEXIST) {
9698 * eexist without a new_inode */
9699 if (WARN_ON(!new_inode)) {
9703 /* maybe -EOVERFLOW */
9710 * we're using rename to replace one file with another. Start IO on it
9711 * now so we don't add too much work to the end of the transaction
9713 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9714 filemap_flush(old_inode->i_mapping);
9716 /* close the racy window with snapshot create/destroy ioctl */
9717 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9718 down_read(&fs_info->subvol_sem);
9720 * We want to reserve the absolute worst case amount of items. So if
9721 * both inodes are subvols and we need to unlink them then that would
9722 * require 4 item modifications, but if they are both normal inodes it
9723 * would require 5 item modifications, so we'll assume they are normal
9724 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9725 * should cover the worst case number of items we'll modify.
9726 * If our rename has the whiteout flag, we need more 5 units for the
9727 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9728 * when selinux is enabled).
9730 trans_num_items = 11;
9731 if (flags & RENAME_WHITEOUT)
9732 trans_num_items += 5;
9733 trans = btrfs_start_transaction(root, trans_num_items);
9734 if (IS_ERR(trans)) {
9735 ret = PTR_ERR(trans);
9740 btrfs_record_root_in_trans(trans, dest);
9742 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9746 BTRFS_I(old_inode)->dir_index = 0ULL;
9747 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9748 /* force full log commit if subvolume involved. */
9749 btrfs_set_log_full_commit(fs_info, trans);
9751 btrfs_pin_log_trans(root);
9753 ret = btrfs_insert_inode_ref(trans, dest,
9754 new_dentry->d_name.name,
9755 new_dentry->d_name.len,
9757 btrfs_ino(BTRFS_I(new_dir)), index);
9762 inode_inc_iversion(old_dir);
9763 inode_inc_iversion(new_dir);
9764 inode_inc_iversion(old_inode);
9765 old_dir->i_ctime = old_dir->i_mtime =
9766 new_dir->i_ctime = new_dir->i_mtime =
9767 old_inode->i_ctime = current_time(old_dir);
9769 if (old_dentry->d_parent != new_dentry->d_parent)
9770 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9771 BTRFS_I(old_inode), 1);
9773 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9774 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9775 ret = btrfs_unlink_subvol(trans, old_dir, root_objectid,
9776 old_dentry->d_name.name,
9777 old_dentry->d_name.len);
9779 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9780 BTRFS_I(d_inode(old_dentry)),
9781 old_dentry->d_name.name,
9782 old_dentry->d_name.len);
9784 ret = btrfs_update_inode(trans, root, old_inode);
9787 btrfs_abort_transaction(trans, ret);
9792 inode_inc_iversion(new_inode);
9793 new_inode->i_ctime = current_time(new_inode);
9794 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9795 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9796 root_objectid = BTRFS_I(new_inode)->location.objectid;
9797 ret = btrfs_unlink_subvol(trans, new_dir, root_objectid,
9798 new_dentry->d_name.name,
9799 new_dentry->d_name.len);
9800 BUG_ON(new_inode->i_nlink == 0);
9802 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9803 BTRFS_I(d_inode(new_dentry)),
9804 new_dentry->d_name.name,
9805 new_dentry->d_name.len);
9807 if (!ret && new_inode->i_nlink == 0)
9808 ret = btrfs_orphan_add(trans,
9809 BTRFS_I(d_inode(new_dentry)));
9811 btrfs_abort_transaction(trans, ret);
9816 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9817 new_dentry->d_name.name,
9818 new_dentry->d_name.len, 0, index);
9820 btrfs_abort_transaction(trans, ret);
9824 if (old_inode->i_nlink == 1)
9825 BTRFS_I(old_inode)->dir_index = index;
9828 struct dentry *parent = new_dentry->d_parent;
9830 btrfs_init_log_ctx(&ctx, old_inode);
9831 ret = btrfs_log_new_name(trans, BTRFS_I(old_inode),
9832 BTRFS_I(old_dir), parent,
9834 if (ret == BTRFS_NEED_LOG_SYNC)
9836 else if (ret == BTRFS_NEED_TRANS_COMMIT)
9837 commit_transaction = true;
9839 btrfs_end_log_trans(root);
9843 if (flags & RENAME_WHITEOUT) {
9844 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9848 btrfs_abort_transaction(trans, ret);
9854 * If we have pinned the log and an error happened, we unpin tasks
9855 * trying to sync the log and force them to fallback to a transaction
9856 * commit if the log currently contains any of the inodes involved in
9857 * this rename operation (to ensure we do not persist a log with an
9858 * inconsistent state for any of these inodes or leading to any
9859 * inconsistencies when replayed). If the transaction was aborted, the
9860 * abortion reason is propagated to userspace when attempting to commit
9861 * the transaction. If the log does not contain any of these inodes, we
9862 * allow the tasks to sync it.
9864 if (ret && log_pinned) {
9865 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9866 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9867 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9869 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9870 btrfs_set_log_full_commit(fs_info, trans);
9872 btrfs_end_log_trans(root);
9875 if (!ret && sync_log) {
9876 ret = btrfs_sync_log(trans, BTRFS_I(old_inode)->root, &ctx);
9878 commit_transaction = true;
9880 if (commit_transaction) {
9881 ret = btrfs_commit_transaction(trans);
9885 ret2 = btrfs_end_transaction(trans);
9886 ret = ret ? ret : ret2;
9889 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9890 up_read(&fs_info->subvol_sem);
9895 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9896 struct inode *new_dir, struct dentry *new_dentry,
9899 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9902 if (flags & RENAME_EXCHANGE)
9903 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9906 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9909 struct btrfs_delalloc_work {
9910 struct inode *inode;
9911 struct completion completion;
9912 struct list_head list;
9913 struct btrfs_work work;
9916 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9918 struct btrfs_delalloc_work *delalloc_work;
9919 struct inode *inode;
9921 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9923 inode = delalloc_work->inode;
9924 filemap_flush(inode->i_mapping);
9925 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9926 &BTRFS_I(inode)->runtime_flags))
9927 filemap_flush(inode->i_mapping);
9930 complete(&delalloc_work->completion);
9933 static 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 WARN_ON_ONCE(!inode);
9945 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9946 btrfs_run_delalloc_work, NULL, NULL);
9952 * some fairly slow code that needs optimization. This walks the list
9953 * of all the inodes with pending delalloc and forces them to disk.
9955 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9957 struct btrfs_inode *binode;
9958 struct inode *inode;
9959 struct btrfs_delalloc_work *work, *next;
9960 struct list_head works;
9961 struct list_head splice;
9964 INIT_LIST_HEAD(&works);
9965 INIT_LIST_HEAD(&splice);
9967 mutex_lock(&root->delalloc_mutex);
9968 spin_lock(&root->delalloc_lock);
9969 list_splice_init(&root->delalloc_inodes, &splice);
9970 while (!list_empty(&splice)) {
9971 binode = list_entry(splice.next, struct btrfs_inode,
9974 list_move_tail(&binode->delalloc_inodes,
9975 &root->delalloc_inodes);
9976 inode = igrab(&binode->vfs_inode);
9978 cond_resched_lock(&root->delalloc_lock);
9981 spin_unlock(&root->delalloc_lock);
9983 work = btrfs_alloc_delalloc_work(inode);
9989 list_add_tail(&work->list, &works);
9990 btrfs_queue_work(root->fs_info->flush_workers,
9993 if (nr != -1 && ret >= nr)
9996 spin_lock(&root->delalloc_lock);
9998 spin_unlock(&root->delalloc_lock);
10001 list_for_each_entry_safe(work, next, &works, list) {
10002 list_del_init(&work->list);
10003 wait_for_completion(&work->completion);
10007 if (!list_empty(&splice)) {
10008 spin_lock(&root->delalloc_lock);
10009 list_splice_tail(&splice, &root->delalloc_inodes);
10010 spin_unlock(&root->delalloc_lock);
10012 mutex_unlock(&root->delalloc_mutex);
10016 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10018 struct btrfs_fs_info *fs_info = root->fs_info;
10021 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10024 ret = start_delalloc_inodes(root, -1);
10030 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10032 struct btrfs_root *root;
10033 struct list_head splice;
10036 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10039 INIT_LIST_HEAD(&splice);
10041 mutex_lock(&fs_info->delalloc_root_mutex);
10042 spin_lock(&fs_info->delalloc_root_lock);
10043 list_splice_init(&fs_info->delalloc_roots, &splice);
10044 while (!list_empty(&splice) && nr) {
10045 root = list_first_entry(&splice, struct btrfs_root,
10047 root = btrfs_grab_fs_root(root);
10049 list_move_tail(&root->delalloc_root,
10050 &fs_info->delalloc_roots);
10051 spin_unlock(&fs_info->delalloc_root_lock);
10053 ret = start_delalloc_inodes(root, nr);
10054 btrfs_put_fs_root(root);
10062 spin_lock(&fs_info->delalloc_root_lock);
10064 spin_unlock(&fs_info->delalloc_root_lock);
10068 if (!list_empty(&splice)) {
10069 spin_lock(&fs_info->delalloc_root_lock);
10070 list_splice_tail(&splice, &fs_info->delalloc_roots);
10071 spin_unlock(&fs_info->delalloc_root_lock);
10073 mutex_unlock(&fs_info->delalloc_root_mutex);
10077 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10078 const char *symname)
10080 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10081 struct btrfs_trans_handle *trans;
10082 struct btrfs_root *root = BTRFS_I(dir)->root;
10083 struct btrfs_path *path;
10084 struct btrfs_key key;
10085 struct inode *inode = NULL;
10092 struct btrfs_file_extent_item *ei;
10093 struct extent_buffer *leaf;
10095 name_len = strlen(symname);
10096 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10097 return -ENAMETOOLONG;
10100 * 2 items for inode item and ref
10101 * 2 items for dir items
10102 * 1 item for updating parent inode item
10103 * 1 item for the inline extent item
10104 * 1 item for xattr if selinux is on
10106 trans = btrfs_start_transaction(root, 7);
10108 return PTR_ERR(trans);
10110 err = btrfs_find_free_ino(root, &objectid);
10114 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10115 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10116 objectid, S_IFLNK|S_IRWXUGO, &index);
10117 if (IS_ERR(inode)) {
10118 err = PTR_ERR(inode);
10124 * If the active LSM wants to access the inode during
10125 * d_instantiate it needs these. Smack checks to see
10126 * if the filesystem supports xattrs by looking at the
10129 inode->i_fop = &btrfs_file_operations;
10130 inode->i_op = &btrfs_file_inode_operations;
10131 inode->i_mapping->a_ops = &btrfs_aops;
10132 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10134 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10138 path = btrfs_alloc_path();
10143 key.objectid = btrfs_ino(BTRFS_I(inode));
10145 key.type = BTRFS_EXTENT_DATA_KEY;
10146 datasize = btrfs_file_extent_calc_inline_size(name_len);
10147 err = btrfs_insert_empty_item(trans, root, path, &key,
10150 btrfs_free_path(path);
10153 leaf = path->nodes[0];
10154 ei = btrfs_item_ptr(leaf, path->slots[0],
10155 struct btrfs_file_extent_item);
10156 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10157 btrfs_set_file_extent_type(leaf, ei,
10158 BTRFS_FILE_EXTENT_INLINE);
10159 btrfs_set_file_extent_encryption(leaf, ei, 0);
10160 btrfs_set_file_extent_compression(leaf, ei, 0);
10161 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10162 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10164 ptr = btrfs_file_extent_inline_start(ei);
10165 write_extent_buffer(leaf, symname, ptr, name_len);
10166 btrfs_mark_buffer_dirty(leaf);
10167 btrfs_free_path(path);
10169 inode->i_op = &btrfs_symlink_inode_operations;
10170 inode_nohighmem(inode);
10171 inode->i_mapping->a_ops = &btrfs_aops;
10172 inode_set_bytes(inode, name_len);
10173 btrfs_i_size_write(BTRFS_I(inode), name_len);
10174 err = btrfs_update_inode(trans, root, inode);
10176 * Last step, add directory indexes for our symlink inode. This is the
10177 * last step to avoid extra cleanup of these indexes if an error happens
10181 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10182 BTRFS_I(inode), 0, index);
10186 d_instantiate_new(dentry, inode);
10189 btrfs_end_transaction(trans);
10190 if (err && inode) {
10191 inode_dec_link_count(inode);
10192 discard_new_inode(inode);
10194 btrfs_btree_balance_dirty(fs_info);
10198 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10199 u64 start, u64 num_bytes, u64 min_size,
10200 loff_t actual_len, u64 *alloc_hint,
10201 struct btrfs_trans_handle *trans)
10203 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10205 struct extent_map *em;
10206 struct btrfs_root *root = BTRFS_I(inode)->root;
10207 struct btrfs_key ins;
10208 u64 cur_offset = start;
10211 u64 last_alloc = (u64)-1;
10213 bool own_trans = true;
10214 u64 end = start + num_bytes - 1;
10218 while (num_bytes > 0) {
10220 trans = btrfs_start_transaction(root, 3);
10221 if (IS_ERR(trans)) {
10222 ret = PTR_ERR(trans);
10227 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10228 cur_bytes = max(cur_bytes, min_size);
10230 * If we are severely fragmented we could end up with really
10231 * small allocations, so if the allocator is returning small
10232 * chunks lets make its job easier by only searching for those
10235 cur_bytes = min(cur_bytes, last_alloc);
10236 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10237 min_size, 0, *alloc_hint, &ins, 1, 0);
10240 btrfs_end_transaction(trans);
10243 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10245 last_alloc = ins.offset;
10246 ret = insert_reserved_file_extent(trans, inode,
10247 cur_offset, ins.objectid,
10248 ins.offset, ins.offset,
10249 ins.offset, 0, 0, 0,
10250 BTRFS_FILE_EXTENT_PREALLOC);
10252 btrfs_free_reserved_extent(fs_info, ins.objectid,
10254 btrfs_abort_transaction(trans, ret);
10256 btrfs_end_transaction(trans);
10260 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10261 cur_offset + ins.offset -1, 0);
10263 em = alloc_extent_map();
10265 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10266 &BTRFS_I(inode)->runtime_flags);
10270 em->start = cur_offset;
10271 em->orig_start = cur_offset;
10272 em->len = ins.offset;
10273 em->block_start = ins.objectid;
10274 em->block_len = ins.offset;
10275 em->orig_block_len = ins.offset;
10276 em->ram_bytes = ins.offset;
10277 em->bdev = fs_info->fs_devices->latest_bdev;
10278 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10279 em->generation = trans->transid;
10282 write_lock(&em_tree->lock);
10283 ret = add_extent_mapping(em_tree, em, 1);
10284 write_unlock(&em_tree->lock);
10285 if (ret != -EEXIST)
10287 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10288 cur_offset + ins.offset - 1,
10291 free_extent_map(em);
10293 num_bytes -= ins.offset;
10294 cur_offset += ins.offset;
10295 *alloc_hint = ins.objectid + ins.offset;
10297 inode_inc_iversion(inode);
10298 inode->i_ctime = current_time(inode);
10299 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10300 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10301 (actual_len > inode->i_size) &&
10302 (cur_offset > inode->i_size)) {
10303 if (cur_offset > actual_len)
10304 i_size = actual_len;
10306 i_size = cur_offset;
10307 i_size_write(inode, i_size);
10308 btrfs_ordered_update_i_size(inode, i_size, NULL);
10311 ret = btrfs_update_inode(trans, root, inode);
10314 btrfs_abort_transaction(trans, ret);
10316 btrfs_end_transaction(trans);
10321 btrfs_end_transaction(trans);
10323 if (cur_offset < end)
10324 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10325 end - cur_offset + 1);
10329 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10330 u64 start, u64 num_bytes, u64 min_size,
10331 loff_t actual_len, u64 *alloc_hint)
10333 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10334 min_size, actual_len, alloc_hint,
10338 int btrfs_prealloc_file_range_trans(struct inode *inode,
10339 struct btrfs_trans_handle *trans, int mode,
10340 u64 start, u64 num_bytes, u64 min_size,
10341 loff_t actual_len, u64 *alloc_hint)
10343 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10344 min_size, actual_len, alloc_hint, trans);
10347 static int btrfs_set_page_dirty(struct page *page)
10349 return __set_page_dirty_nobuffers(page);
10352 static int btrfs_permission(struct inode *inode, int mask)
10354 struct btrfs_root *root = BTRFS_I(inode)->root;
10355 umode_t mode = inode->i_mode;
10357 if (mask & MAY_WRITE &&
10358 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10359 if (btrfs_root_readonly(root))
10361 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10364 return generic_permission(inode, mask);
10367 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10369 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10370 struct btrfs_trans_handle *trans;
10371 struct btrfs_root *root = BTRFS_I(dir)->root;
10372 struct inode *inode = NULL;
10378 * 5 units required for adding orphan entry
10380 trans = btrfs_start_transaction(root, 5);
10382 return PTR_ERR(trans);
10384 ret = btrfs_find_free_ino(root, &objectid);
10388 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10389 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10390 if (IS_ERR(inode)) {
10391 ret = PTR_ERR(inode);
10396 inode->i_fop = &btrfs_file_operations;
10397 inode->i_op = &btrfs_file_inode_operations;
10399 inode->i_mapping->a_ops = &btrfs_aops;
10400 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10402 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10406 ret = btrfs_update_inode(trans, root, inode);
10409 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10414 * We set number of links to 0 in btrfs_new_inode(), and here we set
10415 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10418 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10420 set_nlink(inode, 1);
10421 d_tmpfile(dentry, inode);
10422 unlock_new_inode(inode);
10423 mark_inode_dirty(inode);
10425 btrfs_end_transaction(trans);
10427 discard_new_inode(inode);
10428 btrfs_btree_balance_dirty(fs_info);
10432 __attribute__((const))
10433 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10438 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10439 u64 start, u64 end)
10441 struct inode *inode = private_data;
10444 isize = i_size_read(inode);
10445 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10446 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10447 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10448 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10452 void btrfs_set_range_writeback(struct extent_io_tree *tree, u64 start, u64 end)
10454 struct inode *inode = tree->private_data;
10455 unsigned long index = start >> PAGE_SHIFT;
10456 unsigned long end_index = end >> PAGE_SHIFT;
10459 while (index <= end_index) {
10460 page = find_get_page(inode->i_mapping, index);
10461 ASSERT(page); /* Pages should be in the extent_io_tree */
10462 set_page_writeback(page);
10468 static const struct inode_operations btrfs_dir_inode_operations = {
10469 .getattr = btrfs_getattr,
10470 .lookup = btrfs_lookup,
10471 .create = btrfs_create,
10472 .unlink = btrfs_unlink,
10473 .link = btrfs_link,
10474 .mkdir = btrfs_mkdir,
10475 .rmdir = btrfs_rmdir,
10476 .rename = btrfs_rename2,
10477 .symlink = btrfs_symlink,
10478 .setattr = btrfs_setattr,
10479 .mknod = btrfs_mknod,
10480 .listxattr = btrfs_listxattr,
10481 .permission = btrfs_permission,
10482 .get_acl = btrfs_get_acl,
10483 .set_acl = btrfs_set_acl,
10484 .update_time = btrfs_update_time,
10485 .tmpfile = btrfs_tmpfile,
10487 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10488 .lookup = btrfs_lookup,
10489 .permission = btrfs_permission,
10490 .update_time = btrfs_update_time,
10493 static const struct file_operations btrfs_dir_file_operations = {
10494 .llseek = generic_file_llseek,
10495 .read = generic_read_dir,
10496 .iterate_shared = btrfs_real_readdir,
10497 .open = btrfs_opendir,
10498 .unlocked_ioctl = btrfs_ioctl,
10499 #ifdef CONFIG_COMPAT
10500 .compat_ioctl = btrfs_compat_ioctl,
10502 .release = btrfs_release_file,
10503 .fsync = btrfs_sync_file,
10506 static const struct extent_io_ops btrfs_extent_io_ops = {
10507 /* mandatory callbacks */
10508 .submit_bio_hook = btrfs_submit_bio_hook,
10509 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10510 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10512 /* optional callbacks */
10513 .fill_delalloc = run_delalloc_range,
10514 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10515 .writepage_start_hook = btrfs_writepage_start_hook,
10516 .set_bit_hook = btrfs_set_bit_hook,
10517 .clear_bit_hook = btrfs_clear_bit_hook,
10518 .merge_extent_hook = btrfs_merge_extent_hook,
10519 .split_extent_hook = btrfs_split_extent_hook,
10520 .check_extent_io_range = btrfs_check_extent_io_range,
10524 * btrfs doesn't support the bmap operation because swapfiles
10525 * use bmap to make a mapping of extents in the file. They assume
10526 * these extents won't change over the life of the file and they
10527 * use the bmap result to do IO directly to the drive.
10529 * the btrfs bmap call would return logical addresses that aren't
10530 * suitable for IO and they also will change frequently as COW
10531 * operations happen. So, swapfile + btrfs == corruption.
10533 * For now we're avoiding this by dropping bmap.
10535 static const struct address_space_operations btrfs_aops = {
10536 .readpage = btrfs_readpage,
10537 .writepage = btrfs_writepage,
10538 .writepages = btrfs_writepages,
10539 .readpages = btrfs_readpages,
10540 .direct_IO = btrfs_direct_IO,
10541 .invalidatepage = btrfs_invalidatepage,
10542 .releasepage = btrfs_releasepage,
10543 .set_page_dirty = btrfs_set_page_dirty,
10544 .error_remove_page = generic_error_remove_page,
10547 static const struct inode_operations btrfs_file_inode_operations = {
10548 .getattr = btrfs_getattr,
10549 .setattr = btrfs_setattr,
10550 .listxattr = btrfs_listxattr,
10551 .permission = btrfs_permission,
10552 .fiemap = btrfs_fiemap,
10553 .get_acl = btrfs_get_acl,
10554 .set_acl = btrfs_set_acl,
10555 .update_time = btrfs_update_time,
10557 static const struct inode_operations btrfs_special_inode_operations = {
10558 .getattr = btrfs_getattr,
10559 .setattr = btrfs_setattr,
10560 .permission = btrfs_permission,
10561 .listxattr = btrfs_listxattr,
10562 .get_acl = btrfs_get_acl,
10563 .set_acl = btrfs_set_acl,
10564 .update_time = btrfs_update_time,
10566 static const struct inode_operations btrfs_symlink_inode_operations = {
10567 .get_link = page_get_link,
10568 .getattr = btrfs_getattr,
10569 .setattr = btrfs_setattr,
10570 .permission = btrfs_permission,
10571 .listxattr = btrfs_listxattr,
10572 .update_time = btrfs_update_time,
10575 const struct dentry_operations btrfs_dentry_operations = {
10576 .d_delete = btrfs_dentry_delete,