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/mpage.h>
18 #include <linux/swap.h>
19 #include <linux/writeback.h>
20 #include <linux/compat.h>
21 #include <linux/bit_spinlock.h>
22 #include <linux/xattr.h>
23 #include <linux/posix_acl.h>
24 #include <linux/falloc.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/mount.h>
28 #include <linux/btrfs.h>
29 #include <linux/blkdev.h>
30 #include <linux/posix_acl_xattr.h>
31 #include <linux/uio.h>
32 #include <linux/magic.h>
33 #include <linux/iversion.h>
34 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "print-tree.h"
40 #include "ordered-data.h"
44 #include "compression.h"
46 #include "free-space-cache.h"
47 #include "inode-map.h"
53 struct btrfs_iget_args {
54 struct btrfs_key *location;
55 struct btrfs_root *root;
58 struct btrfs_dio_data {
60 u64 unsubmitted_oe_range_start;
61 u64 unsubmitted_oe_range_end;
65 static const struct inode_operations btrfs_dir_inode_operations;
66 static const struct inode_operations btrfs_symlink_inode_operations;
67 static const struct inode_operations btrfs_dir_ro_inode_operations;
68 static const struct inode_operations btrfs_special_inode_operations;
69 static const struct inode_operations btrfs_file_inode_operations;
70 static const struct address_space_operations btrfs_aops;
71 static const struct address_space_operations btrfs_symlink_aops;
72 static const struct file_operations btrfs_dir_file_operations;
73 static const struct extent_io_ops btrfs_extent_io_ops;
75 static struct kmem_cache *btrfs_inode_cachep;
76 struct kmem_cache *btrfs_trans_handle_cachep;
77 struct kmem_cache *btrfs_path_cachep;
78 struct kmem_cache *btrfs_free_space_cachep;
81 static const unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
82 [S_IFREG >> S_SHIFT] = BTRFS_FT_REG_FILE,
83 [S_IFDIR >> S_SHIFT] = BTRFS_FT_DIR,
84 [S_IFCHR >> S_SHIFT] = BTRFS_FT_CHRDEV,
85 [S_IFBLK >> S_SHIFT] = BTRFS_FT_BLKDEV,
86 [S_IFIFO >> S_SHIFT] = BTRFS_FT_FIFO,
87 [S_IFSOCK >> S_SHIFT] = BTRFS_FT_SOCK,
88 [S_IFLNK >> S_SHIFT] = BTRFS_FT_SYMLINK,
91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
92 static int btrfs_truncate(struct inode *inode, bool skip_writeback);
93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
94 static noinline int cow_file_range(struct inode *inode,
95 struct page *locked_page,
96 u64 start, u64 end, u64 delalloc_end,
97 int *page_started, unsigned long *nr_written,
98 int unlock, struct btrfs_dedupe_hash *hash);
99 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
100 u64 orig_start, u64 block_start,
101 u64 block_len, u64 orig_block_len,
102 u64 ram_bytes, int compress_type,
105 static void __endio_write_update_ordered(struct inode *inode,
106 const u64 offset, const u64 bytes,
107 const bool uptodate);
110 * Cleanup all submitted ordered extents in specified range to handle errors
111 * from the fill_dellaloc() callback.
113 * NOTE: caller must ensure that when an error happens, it can not call
114 * extent_clear_unlock_delalloc() to clear both the bits EXTENT_DO_ACCOUNTING
115 * and EXTENT_DELALLOC simultaneously, because that causes the reserved metadata
116 * to be released, which we want to happen only when finishing the ordered
117 * extent (btrfs_finish_ordered_io()). Also note that the caller of the
118 * fill_delalloc() callback already does proper cleanup for the first page of
119 * the range, that is, it invokes the callback writepage_end_io_hook() for the
120 * range of the first page.
122 static inline void btrfs_cleanup_ordered_extents(struct inode *inode,
126 unsigned long index = offset >> PAGE_SHIFT;
127 unsigned long end_index = (offset + bytes - 1) >> PAGE_SHIFT;
130 while (index <= end_index) {
131 page = find_get_page(inode->i_mapping, index);
135 ClearPagePrivate2(page);
138 return __endio_write_update_ordered(inode, offset + PAGE_SIZE,
139 bytes - PAGE_SIZE, false);
142 static int btrfs_dirty_inode(struct inode *inode);
144 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
145 void btrfs_test_inode_set_ops(struct inode *inode)
147 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
151 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
152 struct inode *inode, struct inode *dir,
153 const struct qstr *qstr)
157 err = btrfs_init_acl(trans, inode, dir);
159 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
164 * this does all the hard work for inserting an inline extent into
165 * the btree. The caller should have done a btrfs_drop_extents so that
166 * no overlapping inline items exist in the btree
168 static int insert_inline_extent(struct btrfs_trans_handle *trans,
169 struct btrfs_path *path, int extent_inserted,
170 struct btrfs_root *root, struct inode *inode,
171 u64 start, size_t size, size_t compressed_size,
173 struct page **compressed_pages)
175 struct extent_buffer *leaf;
176 struct page *page = NULL;
179 struct btrfs_file_extent_item *ei;
181 size_t cur_size = size;
182 unsigned long offset;
184 if (compressed_size && compressed_pages)
185 cur_size = compressed_size;
187 inode_add_bytes(inode, size);
189 if (!extent_inserted) {
190 struct btrfs_key key;
193 key.objectid = btrfs_ino(BTRFS_I(inode));
195 key.type = BTRFS_EXTENT_DATA_KEY;
197 datasize = btrfs_file_extent_calc_inline_size(cur_size);
198 path->leave_spinning = 1;
199 ret = btrfs_insert_empty_item(trans, root, path, &key,
204 leaf = path->nodes[0];
205 ei = btrfs_item_ptr(leaf, path->slots[0],
206 struct btrfs_file_extent_item);
207 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
208 btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
209 btrfs_set_file_extent_encryption(leaf, ei, 0);
210 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
211 btrfs_set_file_extent_ram_bytes(leaf, ei, size);
212 ptr = btrfs_file_extent_inline_start(ei);
214 if (compress_type != BTRFS_COMPRESS_NONE) {
217 while (compressed_size > 0) {
218 cpage = compressed_pages[i];
219 cur_size = min_t(unsigned long, compressed_size,
222 kaddr = kmap_atomic(cpage);
223 write_extent_buffer(leaf, kaddr, ptr, cur_size);
224 kunmap_atomic(kaddr);
228 compressed_size -= cur_size;
230 btrfs_set_file_extent_compression(leaf, ei,
233 page = find_get_page(inode->i_mapping,
234 start >> PAGE_SHIFT);
235 btrfs_set_file_extent_compression(leaf, ei, 0);
236 kaddr = kmap_atomic(page);
237 offset = start & (PAGE_SIZE - 1);
238 write_extent_buffer(leaf, kaddr + offset, ptr, size);
239 kunmap_atomic(kaddr);
242 btrfs_mark_buffer_dirty(leaf);
243 btrfs_release_path(path);
246 * we're an inline extent, so nobody can
247 * extend the file past i_size without locking
248 * a page we already have locked.
250 * We must do any isize and inode updates
251 * before we unlock the pages. Otherwise we
252 * could end up racing with unlink.
254 BTRFS_I(inode)->disk_i_size = inode->i_size;
255 ret = btrfs_update_inode(trans, root, inode);
263 * conditionally insert an inline extent into the file. This
264 * does the checks required to make sure the data is small enough
265 * to fit as an inline extent.
267 static noinline int cow_file_range_inline(struct inode *inode, u64 start,
268 u64 end, size_t compressed_size,
270 struct page **compressed_pages)
272 struct btrfs_root *root = BTRFS_I(inode)->root;
273 struct btrfs_fs_info *fs_info = root->fs_info;
274 struct btrfs_trans_handle *trans;
275 u64 isize = i_size_read(inode);
276 u64 actual_end = min(end + 1, isize);
277 u64 inline_len = actual_end - start;
278 u64 aligned_end = ALIGN(end, fs_info->sectorsize);
279 u64 data_len = inline_len;
281 struct btrfs_path *path;
282 int extent_inserted = 0;
283 u32 extent_item_size;
286 data_len = compressed_size;
289 actual_end > fs_info->sectorsize ||
290 data_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info) ||
292 (actual_end & (fs_info->sectorsize - 1)) == 0) ||
294 data_len > fs_info->max_inline) {
298 path = btrfs_alloc_path();
302 trans = btrfs_join_transaction(root);
304 btrfs_free_path(path);
305 return PTR_ERR(trans);
307 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
309 if (compressed_size && compressed_pages)
310 extent_item_size = btrfs_file_extent_calc_inline_size(
313 extent_item_size = btrfs_file_extent_calc_inline_size(
316 ret = __btrfs_drop_extents(trans, root, inode, path,
317 start, aligned_end, NULL,
318 1, 1, extent_item_size, &extent_inserted);
320 btrfs_abort_transaction(trans, ret);
324 if (isize > actual_end)
325 inline_len = min_t(u64, isize, actual_end);
326 ret = insert_inline_extent(trans, path, extent_inserted,
328 inline_len, compressed_size,
329 compress_type, compressed_pages);
330 if (ret && ret != -ENOSPC) {
331 btrfs_abort_transaction(trans, ret);
333 } else if (ret == -ENOSPC) {
338 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
339 btrfs_drop_extent_cache(BTRFS_I(inode), start, aligned_end - 1, 0);
342 * Don't forget to free the reserved space, as for inlined extent
343 * it won't count as data extent, free them directly here.
344 * And at reserve time, it's always aligned to page size, so
345 * just free one page here.
347 btrfs_qgroup_free_data(inode, NULL, 0, PAGE_SIZE);
348 btrfs_free_path(path);
349 btrfs_end_transaction(trans);
353 struct async_extent {
358 unsigned long nr_pages;
360 struct list_head list;
365 struct btrfs_root *root;
366 struct page *locked_page;
369 unsigned int write_flags;
370 struct list_head extents;
371 struct btrfs_work work;
374 static noinline int add_async_extent(struct async_cow *cow,
375 u64 start, u64 ram_size,
378 unsigned long nr_pages,
381 struct async_extent *async_extent;
383 async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
384 BUG_ON(!async_extent); /* -ENOMEM */
385 async_extent->start = start;
386 async_extent->ram_size = ram_size;
387 async_extent->compressed_size = compressed_size;
388 async_extent->pages = pages;
389 async_extent->nr_pages = nr_pages;
390 async_extent->compress_type = compress_type;
391 list_add_tail(&async_extent->list, &cow->extents);
395 static inline int inode_need_compress(struct inode *inode, u64 start, u64 end)
397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
400 if (btrfs_test_opt(fs_info, FORCE_COMPRESS))
403 if (BTRFS_I(inode)->defrag_compress)
405 /* bad compression ratios */
406 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
408 if (btrfs_test_opt(fs_info, COMPRESS) ||
409 BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
410 BTRFS_I(inode)->prop_compress)
411 return btrfs_compress_heuristic(inode, start, end);
415 static inline void inode_should_defrag(struct btrfs_inode *inode,
416 u64 start, u64 end, u64 num_bytes, u64 small_write)
418 /* If this is a small write inside eof, kick off a defrag */
419 if (num_bytes < small_write &&
420 (start > 0 || end + 1 < inode->disk_i_size))
421 btrfs_add_inode_defrag(NULL, inode);
425 * we create compressed extents in two phases. The first
426 * phase compresses a range of pages that have already been
427 * locked (both pages and state bits are locked).
429 * This is done inside an ordered work queue, and the compression
430 * is spread across many cpus. The actual IO submission is step
431 * two, and the ordered work queue takes care of making sure that
432 * happens in the same order things were put onto the queue by
433 * writepages and friends.
435 * If this code finds it can't get good compression, it puts an
436 * entry onto the work queue to write the uncompressed bytes. This
437 * makes sure that both compressed inodes and uncompressed inodes
438 * are written in the same order that the flusher thread sent them
441 static noinline void compress_file_range(struct inode *inode,
442 struct page *locked_page,
444 struct async_cow *async_cow,
447 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
448 u64 blocksize = fs_info->sectorsize;
450 u64 isize = i_size_read(inode);
452 struct page **pages = NULL;
453 unsigned long nr_pages;
454 unsigned long total_compressed = 0;
455 unsigned long total_in = 0;
458 int compress_type = fs_info->compress_type;
461 inode_should_defrag(BTRFS_I(inode), start, end, end - start + 1,
464 actual_end = min_t(u64, isize, end + 1);
467 nr_pages = (end >> PAGE_SHIFT) - (start >> PAGE_SHIFT) + 1;
468 BUILD_BUG_ON((BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0);
469 nr_pages = min_t(unsigned long, nr_pages,
470 BTRFS_MAX_COMPRESSED / PAGE_SIZE);
473 * we don't want to send crud past the end of i_size through
474 * compression, that's just a waste of CPU time. So, if the
475 * end of the file is before the start of our current
476 * requested range of bytes, we bail out to the uncompressed
477 * cleanup code that can deal with all of this.
479 * It isn't really the fastest way to fix things, but this is a
480 * very uncommon corner.
482 if (actual_end <= start)
483 goto cleanup_and_bail_uncompressed;
485 total_compressed = actual_end - start;
488 * skip compression for a small file range(<=blocksize) that
489 * isn't an inline extent, since it doesn't save disk space at all.
491 if (total_compressed <= blocksize &&
492 (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
493 goto cleanup_and_bail_uncompressed;
495 total_compressed = min_t(unsigned long, total_compressed,
496 BTRFS_MAX_UNCOMPRESSED);
501 * we do compression for mount -o compress and when the
502 * inode has not been flagged as nocompress. This flag can
503 * change at any time if we discover bad compression ratios.
505 if (inode_need_compress(inode, start, end)) {
507 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_NOFS);
509 /* just bail out to the uncompressed code */
513 if (BTRFS_I(inode)->defrag_compress)
514 compress_type = BTRFS_I(inode)->defrag_compress;
515 else if (BTRFS_I(inode)->prop_compress)
516 compress_type = BTRFS_I(inode)->prop_compress;
519 * we need to call clear_page_dirty_for_io on each
520 * page in the range. Otherwise applications with the file
521 * mmap'd can wander in and change the page contents while
522 * we are compressing them.
524 * If the compression fails for any reason, we set the pages
525 * dirty again later on.
527 * Note that the remaining part is redirtied, the start pointer
528 * has moved, the end is the original one.
531 extent_range_clear_dirty_for_io(inode, start, end);
535 /* Compression level is applied here and only here */
536 ret = btrfs_compress_pages(
537 compress_type | (fs_info->compress_level << 4),
538 inode->i_mapping, start,
545 unsigned long offset = total_compressed &
547 struct page *page = pages[nr_pages - 1];
550 /* zero the tail end of the last page, we might be
551 * sending it down to disk
554 kaddr = kmap_atomic(page);
555 memset(kaddr + offset, 0,
557 kunmap_atomic(kaddr);
564 /* lets try to make an inline extent */
565 if (ret || total_in < actual_end) {
566 /* we didn't compress the entire range, try
567 * to make an uncompressed inline extent.
569 ret = cow_file_range_inline(inode, start, end, 0,
570 BTRFS_COMPRESS_NONE, NULL);
572 /* try making a compressed inline extent */
573 ret = cow_file_range_inline(inode, start, end,
575 compress_type, pages);
578 unsigned long clear_flags = EXTENT_DELALLOC |
579 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
580 EXTENT_DO_ACCOUNTING;
581 unsigned long page_error_op;
583 page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
586 * inline extent creation worked or returned error,
587 * we don't need to create any more async work items.
588 * Unlock and free up our temp pages.
590 * We use DO_ACCOUNTING here because we need the
591 * delalloc_release_metadata to be done _after_ we drop
592 * our outstanding extent for clearing delalloc for this
595 extent_clear_unlock_delalloc(inode, start, end, end,
608 * we aren't doing an inline extent round the compressed size
609 * up to a block size boundary so the allocator does sane
612 total_compressed = ALIGN(total_compressed, blocksize);
615 * one last check to make sure the compression is really a
616 * win, compare the page count read with the blocks on disk,
617 * compression must free at least one sector size
619 total_in = ALIGN(total_in, PAGE_SIZE);
620 if (total_compressed + blocksize <= total_in) {
624 * The async work queues will take care of doing actual
625 * allocation on disk for these compressed pages, and
626 * will submit them to the elevator.
628 add_async_extent(async_cow, start, total_in,
629 total_compressed, pages, nr_pages,
632 if (start + total_in < end) {
643 * the compression code ran but failed to make things smaller,
644 * free any pages it allocated and our page pointer array
646 for (i = 0; i < nr_pages; i++) {
647 WARN_ON(pages[i]->mapping);
652 total_compressed = 0;
655 /* flag the file so we don't compress in the future */
656 if (!btrfs_test_opt(fs_info, FORCE_COMPRESS) &&
657 !(BTRFS_I(inode)->prop_compress)) {
658 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
661 cleanup_and_bail_uncompressed:
663 * No compression, but we still need to write the pages in the file
664 * we've been given so far. redirty the locked page if it corresponds
665 * to our extent and set things up for the async work queue to run
666 * cow_file_range to do the normal delalloc dance.
668 if (page_offset(locked_page) >= start &&
669 page_offset(locked_page) <= end)
670 __set_page_dirty_nobuffers(locked_page);
671 /* unlocked later on in the async handlers */
674 extent_range_redirty_for_io(inode, start, end);
675 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0,
676 BTRFS_COMPRESS_NONE);
682 for (i = 0; i < nr_pages; i++) {
683 WARN_ON(pages[i]->mapping);
689 static void free_async_extent_pages(struct async_extent *async_extent)
693 if (!async_extent->pages)
696 for (i = 0; i < async_extent->nr_pages; i++) {
697 WARN_ON(async_extent->pages[i]->mapping);
698 put_page(async_extent->pages[i]);
700 kfree(async_extent->pages);
701 async_extent->nr_pages = 0;
702 async_extent->pages = NULL;
706 * phase two of compressed writeback. This is the ordered portion
707 * of the code, which only gets called in the order the work was
708 * queued. We walk all the async extents created by compress_file_range
709 * and send them down to the disk.
711 static noinline void submit_compressed_extents(struct inode *inode,
712 struct async_cow *async_cow)
714 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
715 struct async_extent *async_extent;
717 struct btrfs_key ins;
718 struct extent_map *em;
719 struct btrfs_root *root = BTRFS_I(inode)->root;
720 struct extent_io_tree *io_tree;
724 while (!list_empty(&async_cow->extents)) {
725 async_extent = list_entry(async_cow->extents.next,
726 struct async_extent, list);
727 list_del(&async_extent->list);
729 io_tree = &BTRFS_I(inode)->io_tree;
732 /* did the compression code fall back to uncompressed IO? */
733 if (!async_extent->pages) {
734 int page_started = 0;
735 unsigned long nr_written = 0;
737 lock_extent(io_tree, async_extent->start,
738 async_extent->start +
739 async_extent->ram_size - 1);
741 /* allocate blocks */
742 ret = cow_file_range(inode, async_cow->locked_page,
744 async_extent->start +
745 async_extent->ram_size - 1,
746 async_extent->start +
747 async_extent->ram_size - 1,
748 &page_started, &nr_written, 0,
754 * if page_started, cow_file_range inserted an
755 * inline extent and took care of all the unlocking
756 * and IO for us. Otherwise, we need to submit
757 * all those pages down to the drive.
759 if (!page_started && !ret)
760 extent_write_locked_range(inode,
762 async_extent->start +
763 async_extent->ram_size - 1,
766 unlock_page(async_cow->locked_page);
772 lock_extent(io_tree, async_extent->start,
773 async_extent->start + async_extent->ram_size - 1);
775 ret = btrfs_reserve_extent(root, async_extent->ram_size,
776 async_extent->compressed_size,
777 async_extent->compressed_size,
778 0, alloc_hint, &ins, 1, 1);
780 free_async_extent_pages(async_extent);
782 if (ret == -ENOSPC) {
783 unlock_extent(io_tree, async_extent->start,
784 async_extent->start +
785 async_extent->ram_size - 1);
788 * we need to redirty the pages if we decide to
789 * fallback to uncompressed IO, otherwise we
790 * will not submit these pages down to lower
793 extent_range_redirty_for_io(inode,
795 async_extent->start +
796 async_extent->ram_size - 1);
803 * here we're doing allocation and writeback of the
806 em = create_io_em(inode, async_extent->start,
807 async_extent->ram_size, /* len */
808 async_extent->start, /* orig_start */
809 ins.objectid, /* block_start */
810 ins.offset, /* block_len */
811 ins.offset, /* orig_block_len */
812 async_extent->ram_size, /* ram_bytes */
813 async_extent->compress_type,
814 BTRFS_ORDERED_COMPRESSED);
816 /* ret value is not necessary due to void function */
817 goto out_free_reserve;
820 ret = btrfs_add_ordered_extent_compress(inode,
823 async_extent->ram_size,
825 BTRFS_ORDERED_COMPRESSED,
826 async_extent->compress_type);
828 btrfs_drop_extent_cache(BTRFS_I(inode),
830 async_extent->start +
831 async_extent->ram_size - 1, 0);
832 goto out_free_reserve;
834 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
837 * clear dirty, set writeback and unlock the pages.
839 extent_clear_unlock_delalloc(inode, async_extent->start,
840 async_extent->start +
841 async_extent->ram_size - 1,
842 async_extent->start +
843 async_extent->ram_size - 1,
844 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
845 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
847 if (btrfs_submit_compressed_write(inode,
849 async_extent->ram_size,
851 ins.offset, async_extent->pages,
852 async_extent->nr_pages,
853 async_cow->write_flags)) {
854 struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
855 struct page *p = async_extent->pages[0];
856 const u64 start = async_extent->start;
857 const u64 end = start + async_extent->ram_size - 1;
859 p->mapping = inode->i_mapping;
860 tree->ops->writepage_end_io_hook(p, start, end,
863 extent_clear_unlock_delalloc(inode, start, end, end,
867 free_async_extent_pages(async_extent);
869 alloc_hint = ins.objectid + ins.offset;
875 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
876 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
878 extent_clear_unlock_delalloc(inode, async_extent->start,
879 async_extent->start +
880 async_extent->ram_size - 1,
881 async_extent->start +
882 async_extent->ram_size - 1,
883 NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
884 EXTENT_DELALLOC_NEW |
885 EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
886 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
887 PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
889 free_async_extent_pages(async_extent);
894 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
897 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
898 struct extent_map *em;
901 read_lock(&em_tree->lock);
902 em = search_extent_mapping(em_tree, start, num_bytes);
905 * if block start isn't an actual block number then find the
906 * first block in this inode and use that as a hint. If that
907 * block is also bogus then just don't worry about it.
909 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
911 em = search_extent_mapping(em_tree, 0, 0);
912 if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
913 alloc_hint = em->block_start;
917 alloc_hint = em->block_start;
921 read_unlock(&em_tree->lock);
927 * when extent_io.c finds a delayed allocation range in the file,
928 * the call backs end up in this code. The basic idea is to
929 * allocate extents on disk for the range, and create ordered data structs
930 * in ram to track those extents.
932 * locked_page is the page that writepage had locked already. We use
933 * it to make sure we don't do extra locks or unlocks.
935 * *page_started is set to one if we unlock locked_page and do everything
936 * required to start IO on it. It may be clean and already done with
939 static noinline int cow_file_range(struct inode *inode,
940 struct page *locked_page,
941 u64 start, u64 end, u64 delalloc_end,
942 int *page_started, unsigned long *nr_written,
943 int unlock, struct btrfs_dedupe_hash *hash)
945 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
946 struct btrfs_root *root = BTRFS_I(inode)->root;
949 unsigned long ram_size;
950 u64 cur_alloc_size = 0;
951 u64 blocksize = fs_info->sectorsize;
952 struct btrfs_key ins;
953 struct extent_map *em;
955 unsigned long page_ops;
956 bool extent_reserved = false;
959 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
965 num_bytes = ALIGN(end - start + 1, blocksize);
966 num_bytes = max(blocksize, num_bytes);
967 ASSERT(num_bytes <= btrfs_super_total_bytes(fs_info->super_copy));
969 inode_should_defrag(BTRFS_I(inode), start, end, num_bytes, SZ_64K);
972 /* lets try to make an inline extent */
973 ret = cow_file_range_inline(inode, start, end, 0,
974 BTRFS_COMPRESS_NONE, NULL);
977 * We use DO_ACCOUNTING here because we need the
978 * delalloc_release_metadata to be run _after_ we drop
979 * our outstanding extent for clearing delalloc for this
982 extent_clear_unlock_delalloc(inode, start, end,
984 EXTENT_LOCKED | EXTENT_DELALLOC |
985 EXTENT_DELALLOC_NEW | EXTENT_DEFRAG |
986 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
987 PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
989 *nr_written = *nr_written +
990 (end - start + PAGE_SIZE) / PAGE_SIZE;
993 } else if (ret < 0) {
998 alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
999 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1000 start + num_bytes - 1, 0);
1002 while (num_bytes > 0) {
1003 cur_alloc_size = num_bytes;
1004 ret = btrfs_reserve_extent(root, cur_alloc_size, cur_alloc_size,
1005 fs_info->sectorsize, 0, alloc_hint,
1009 cur_alloc_size = ins.offset;
1010 extent_reserved = true;
1012 ram_size = ins.offset;
1013 em = create_io_em(inode, start, ins.offset, /* len */
1014 start, /* orig_start */
1015 ins.objectid, /* block_start */
1016 ins.offset, /* block_len */
1017 ins.offset, /* orig_block_len */
1018 ram_size, /* ram_bytes */
1019 BTRFS_COMPRESS_NONE, /* compress_type */
1020 BTRFS_ORDERED_REGULAR /* type */);
1025 free_extent_map(em);
1027 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1028 ram_size, cur_alloc_size, 0);
1030 goto out_drop_extent_cache;
1032 if (root->root_key.objectid ==
1033 BTRFS_DATA_RELOC_TREE_OBJECTID) {
1034 ret = btrfs_reloc_clone_csums(inode, start,
1037 * Only drop cache here, and process as normal.
1039 * We must not allow extent_clear_unlock_delalloc()
1040 * at out_unlock label to free meta of this ordered
1041 * extent, as its meta should be freed by
1042 * btrfs_finish_ordered_io().
1044 * So we must continue until @start is increased to
1045 * skip current ordered extent.
1048 btrfs_drop_extent_cache(BTRFS_I(inode), start,
1049 start + ram_size - 1, 0);
1052 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1054 /* we're not doing compressed IO, don't unlock the first
1055 * page (which the caller expects to stay locked), don't
1056 * clear any dirty bits and don't set any writeback bits
1058 * Do set the Private2 bit so we know this page was properly
1059 * setup for writepage
1061 page_ops = unlock ? PAGE_UNLOCK : 0;
1062 page_ops |= PAGE_SET_PRIVATE2;
1064 extent_clear_unlock_delalloc(inode, start,
1065 start + ram_size - 1,
1066 delalloc_end, locked_page,
1067 EXTENT_LOCKED | EXTENT_DELALLOC,
1069 if (num_bytes < cur_alloc_size)
1072 num_bytes -= cur_alloc_size;
1073 alloc_hint = ins.objectid + ins.offset;
1074 start += cur_alloc_size;
1075 extent_reserved = false;
1078 * btrfs_reloc_clone_csums() error, since start is increased
1079 * extent_clear_unlock_delalloc() at out_unlock label won't
1080 * free metadata of current ordered extent, we're OK to exit.
1088 out_drop_extent_cache:
1089 btrfs_drop_extent_cache(BTRFS_I(inode), start, start + ram_size - 1, 0);
1091 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
1092 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 1);
1094 clear_bits = EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
1095 EXTENT_DEFRAG | EXTENT_CLEAR_META_RESV;
1096 page_ops = PAGE_UNLOCK | PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
1099 * If we reserved an extent for our delalloc range (or a subrange) and
1100 * failed to create the respective ordered extent, then it means that
1101 * when we reserved the extent we decremented the extent's size from
1102 * the data space_info's bytes_may_use counter and incremented the
1103 * space_info's bytes_reserved counter by the same amount. We must make
1104 * sure extent_clear_unlock_delalloc() does not try to decrement again
1105 * the data space_info's bytes_may_use counter, therefore we do not pass
1106 * it the flag EXTENT_CLEAR_DATA_RESV.
1108 if (extent_reserved) {
1109 extent_clear_unlock_delalloc(inode, start,
1110 start + cur_alloc_size,
1111 start + cur_alloc_size,
1115 start += cur_alloc_size;
1119 extent_clear_unlock_delalloc(inode, start, end, delalloc_end,
1121 clear_bits | EXTENT_CLEAR_DATA_RESV,
1127 * work queue call back to started compression on a file and pages
1129 static noinline void async_cow_start(struct btrfs_work *work)
1131 struct async_cow *async_cow;
1133 async_cow = container_of(work, struct async_cow, work);
1135 compress_file_range(async_cow->inode, async_cow->locked_page,
1136 async_cow->start, async_cow->end, async_cow,
1138 if (num_added == 0) {
1139 btrfs_add_delayed_iput(async_cow->inode);
1140 async_cow->inode = NULL;
1145 * work queue call back to submit previously compressed pages
1147 static noinline void async_cow_submit(struct btrfs_work *work)
1149 struct btrfs_fs_info *fs_info;
1150 struct async_cow *async_cow;
1151 struct btrfs_root *root;
1152 unsigned long nr_pages;
1154 async_cow = container_of(work, struct async_cow, work);
1156 root = async_cow->root;
1157 fs_info = root->fs_info;
1158 nr_pages = (async_cow->end - async_cow->start + PAGE_SIZE) >>
1161 /* atomic_sub_return implies a barrier */
1162 if (atomic_sub_return(nr_pages, &fs_info->async_delalloc_pages) <
1164 cond_wake_up_nomb(&fs_info->async_submit_wait);
1166 if (async_cow->inode)
1167 submit_compressed_extents(async_cow->inode, async_cow);
1170 static noinline void async_cow_free(struct btrfs_work *work)
1172 struct async_cow *async_cow;
1173 async_cow = container_of(work, struct async_cow, work);
1174 if (async_cow->inode)
1175 btrfs_add_delayed_iput(async_cow->inode);
1179 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1180 u64 start, u64 end, int *page_started,
1181 unsigned long *nr_written,
1182 unsigned int write_flags)
1184 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1185 struct async_cow *async_cow;
1186 struct btrfs_root *root = BTRFS_I(inode)->root;
1187 unsigned long nr_pages;
1190 clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1192 while (start < end) {
1193 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1194 BUG_ON(!async_cow); /* -ENOMEM */
1195 async_cow->inode = igrab(inode);
1196 async_cow->root = root;
1197 async_cow->locked_page = locked_page;
1198 async_cow->start = start;
1199 async_cow->write_flags = write_flags;
1201 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1202 !btrfs_test_opt(fs_info, FORCE_COMPRESS))
1205 cur_end = min(end, start + SZ_512K - 1);
1207 async_cow->end = cur_end;
1208 INIT_LIST_HEAD(&async_cow->extents);
1210 btrfs_init_work(&async_cow->work,
1211 btrfs_delalloc_helper,
1212 async_cow_start, async_cow_submit,
1215 nr_pages = (cur_end - start + PAGE_SIZE) >>
1217 atomic_add(nr_pages, &fs_info->async_delalloc_pages);
1219 btrfs_queue_work(fs_info->delalloc_workers, &async_cow->work);
1221 *nr_written += nr_pages;
1222 start = cur_end + 1;
1228 static noinline int csum_exist_in_range(struct btrfs_fs_info *fs_info,
1229 u64 bytenr, u64 num_bytes)
1232 struct btrfs_ordered_sum *sums;
1235 ret = btrfs_lookup_csums_range(fs_info->csum_root, bytenr,
1236 bytenr + num_bytes - 1, &list, 0);
1237 if (ret == 0 && list_empty(&list))
1240 while (!list_empty(&list)) {
1241 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1242 list_del(&sums->list);
1251 * when nowcow writeback call back. This checks for snapshots or COW copies
1252 * of the extents that exist in the file, and COWs the file as required.
1254 * If no cow copies or snapshots exist, we write directly to the existing
1257 static noinline int run_delalloc_nocow(struct inode *inode,
1258 struct page *locked_page,
1259 u64 start, u64 end, int *page_started, int force,
1260 unsigned long *nr_written)
1262 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1263 struct btrfs_root *root = BTRFS_I(inode)->root;
1264 struct extent_buffer *leaf;
1265 struct btrfs_path *path;
1266 struct btrfs_file_extent_item *fi;
1267 struct btrfs_key found_key;
1268 struct extent_map *em;
1283 u64 ino = btrfs_ino(BTRFS_I(inode));
1285 path = btrfs_alloc_path();
1287 extent_clear_unlock_delalloc(inode, start, end, end,
1289 EXTENT_LOCKED | EXTENT_DELALLOC |
1290 EXTENT_DO_ACCOUNTING |
1291 EXTENT_DEFRAG, PAGE_UNLOCK |
1293 PAGE_SET_WRITEBACK |
1294 PAGE_END_WRITEBACK);
1298 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
1300 cow_start = (u64)-1;
1303 ret = btrfs_lookup_file_extent(NULL, root, path, ino,
1307 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1308 leaf = path->nodes[0];
1309 btrfs_item_key_to_cpu(leaf, &found_key,
1310 path->slots[0] - 1);
1311 if (found_key.objectid == ino &&
1312 found_key.type == BTRFS_EXTENT_DATA_KEY)
1317 leaf = path->nodes[0];
1318 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1319 ret = btrfs_next_leaf(root, path);
1321 if (cow_start != (u64)-1)
1322 cur_offset = cow_start;
1327 leaf = path->nodes[0];
1333 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1335 if (found_key.objectid > ino)
1337 if (WARN_ON_ONCE(found_key.objectid < ino) ||
1338 found_key.type < BTRFS_EXTENT_DATA_KEY) {
1342 if (found_key.type > BTRFS_EXTENT_DATA_KEY ||
1343 found_key.offset > end)
1346 if (found_key.offset > cur_offset) {
1347 extent_end = found_key.offset;
1352 fi = btrfs_item_ptr(leaf, path->slots[0],
1353 struct btrfs_file_extent_item);
1354 extent_type = btrfs_file_extent_type(leaf, fi);
1356 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1357 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1358 extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1359 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1360 extent_offset = btrfs_file_extent_offset(leaf, fi);
1361 extent_end = found_key.offset +
1362 btrfs_file_extent_num_bytes(leaf, fi);
1364 btrfs_file_extent_disk_num_bytes(leaf, fi);
1365 if (extent_end <= start) {
1369 if (disk_bytenr == 0)
1371 if (btrfs_file_extent_compression(leaf, fi) ||
1372 btrfs_file_extent_encryption(leaf, fi) ||
1373 btrfs_file_extent_other_encoding(leaf, fi))
1376 * Do the same check as in btrfs_cross_ref_exist but
1377 * without the unnecessary search.
1379 if (btrfs_file_extent_generation(leaf, fi) <=
1380 btrfs_root_last_snapshot(&root->root_item))
1382 if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1384 if (btrfs_extent_readonly(fs_info, disk_bytenr))
1386 ret = btrfs_cross_ref_exist(root, ino,
1388 extent_offset, disk_bytenr);
1391 * ret could be -EIO if the above fails to read
1395 if (cow_start != (u64)-1)
1396 cur_offset = cow_start;
1400 WARN_ON_ONCE(nolock);
1403 disk_bytenr += extent_offset;
1404 disk_bytenr += cur_offset - found_key.offset;
1405 num_bytes = min(end + 1, extent_end) - cur_offset;
1407 * if there are pending snapshots for this root,
1408 * we fall into common COW way.
1411 err = btrfs_start_write_no_snapshotting(root);
1416 * force cow if csum exists in the range.
1417 * this ensure that csum for a given extent are
1418 * either valid or do not exist.
1420 ret = csum_exist_in_range(fs_info, disk_bytenr,
1424 btrfs_end_write_no_snapshotting(root);
1427 * ret could be -EIO if the above fails to read
1431 if (cow_start != (u64)-1)
1432 cur_offset = cow_start;
1435 WARN_ON_ONCE(nolock);
1438 if (!btrfs_inc_nocow_writers(fs_info, disk_bytenr)) {
1440 btrfs_end_write_no_snapshotting(root);
1444 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1445 extent_end = found_key.offset +
1446 btrfs_file_extent_inline_len(leaf,
1447 path->slots[0], fi);
1448 extent_end = ALIGN(extent_end,
1449 fs_info->sectorsize);
1454 if (extent_end <= start) {
1456 if (!nolock && nocow)
1457 btrfs_end_write_no_snapshotting(root);
1459 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1463 if (cow_start == (u64)-1)
1464 cow_start = cur_offset;
1465 cur_offset = extent_end;
1466 if (cur_offset > end)
1472 btrfs_release_path(path);
1473 if (cow_start != (u64)-1) {
1474 ret = cow_file_range(inode, locked_page,
1475 cow_start, found_key.offset - 1,
1476 end, page_started, nr_written, 1,
1479 if (!nolock && nocow)
1480 btrfs_end_write_no_snapshotting(root);
1482 btrfs_dec_nocow_writers(fs_info,
1486 cow_start = (u64)-1;
1489 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1490 u64 orig_start = found_key.offset - extent_offset;
1492 em = create_io_em(inode, cur_offset, num_bytes,
1494 disk_bytenr, /* block_start */
1495 num_bytes, /* block_len */
1496 disk_num_bytes, /* orig_block_len */
1497 ram_bytes, BTRFS_COMPRESS_NONE,
1498 BTRFS_ORDERED_PREALLOC);
1500 if (!nolock && nocow)
1501 btrfs_end_write_no_snapshotting(root);
1503 btrfs_dec_nocow_writers(fs_info,
1508 free_extent_map(em);
1511 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1512 type = BTRFS_ORDERED_PREALLOC;
1514 type = BTRFS_ORDERED_NOCOW;
1517 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1518 num_bytes, num_bytes, type);
1520 btrfs_dec_nocow_writers(fs_info, disk_bytenr);
1521 BUG_ON(ret); /* -ENOMEM */
1523 if (root->root_key.objectid ==
1524 BTRFS_DATA_RELOC_TREE_OBJECTID)
1526 * Error handled later, as we must prevent
1527 * extent_clear_unlock_delalloc() in error handler
1528 * from freeing metadata of created ordered extent.
1530 ret = btrfs_reloc_clone_csums(inode, cur_offset,
1533 extent_clear_unlock_delalloc(inode, cur_offset,
1534 cur_offset + num_bytes - 1, end,
1535 locked_page, EXTENT_LOCKED |
1537 EXTENT_CLEAR_DATA_RESV,
1538 PAGE_UNLOCK | PAGE_SET_PRIVATE2);
1540 if (!nolock && nocow)
1541 btrfs_end_write_no_snapshotting(root);
1542 cur_offset = extent_end;
1545 * btrfs_reloc_clone_csums() error, now we're OK to call error
1546 * handler, as metadata for created ordered extent will only
1547 * be freed by btrfs_finish_ordered_io().
1551 if (cur_offset > end)
1554 btrfs_release_path(path);
1556 if (cur_offset <= end && cow_start == (u64)-1) {
1557 cow_start = cur_offset;
1561 if (cow_start != (u64)-1) {
1562 ret = cow_file_range(inode, locked_page, cow_start, end, end,
1563 page_started, nr_written, 1, NULL);
1569 if (ret && cur_offset < end)
1570 extent_clear_unlock_delalloc(inode, cur_offset, end, end,
1571 locked_page, EXTENT_LOCKED |
1572 EXTENT_DELALLOC | EXTENT_DEFRAG |
1573 EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1575 PAGE_SET_WRITEBACK |
1576 PAGE_END_WRITEBACK);
1577 btrfs_free_path(path);
1581 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1584 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1585 !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1589 * @defrag_bytes is a hint value, no spinlock held here,
1590 * if is not zero, it means the file is defragging.
1591 * Force cow if given extent needs to be defragged.
1593 if (BTRFS_I(inode)->defrag_bytes &&
1594 test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1595 EXTENT_DEFRAG, 0, NULL))
1602 * extent_io.c call back to do delayed allocation processing
1604 static int run_delalloc_range(void *private_data, struct page *locked_page,
1605 u64 start, u64 end, int *page_started,
1606 unsigned long *nr_written,
1607 struct writeback_control *wbc)
1609 struct inode *inode = private_data;
1611 int force_cow = need_force_cow(inode, start, end);
1612 unsigned int write_flags = wbc_to_write_flags(wbc);
1614 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1615 ret = run_delalloc_nocow(inode, locked_page, start, end,
1616 page_started, 1, nr_written);
1617 } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1618 ret = run_delalloc_nocow(inode, locked_page, start, end,
1619 page_started, 0, nr_written);
1620 } else if (!inode_need_compress(inode, start, end)) {
1621 ret = cow_file_range(inode, locked_page, start, end, end,
1622 page_started, nr_written, 1, NULL);
1624 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1625 &BTRFS_I(inode)->runtime_flags);
1626 ret = cow_file_range_async(inode, locked_page, start, end,
1627 page_started, nr_written,
1631 btrfs_cleanup_ordered_extents(inode, start, end - start + 1);
1635 static void btrfs_split_extent_hook(void *private_data,
1636 struct extent_state *orig, u64 split)
1638 struct inode *inode = private_data;
1641 /* not delalloc, ignore it */
1642 if (!(orig->state & EXTENT_DELALLOC))
1645 size = orig->end - orig->start + 1;
1646 if (size > BTRFS_MAX_EXTENT_SIZE) {
1651 * See the explanation in btrfs_merge_extent_hook, the same
1652 * applies here, just in reverse.
1654 new_size = orig->end - split + 1;
1655 num_extents = count_max_extents(new_size);
1656 new_size = split - orig->start;
1657 num_extents += count_max_extents(new_size);
1658 if (count_max_extents(size) >= num_extents)
1662 spin_lock(&BTRFS_I(inode)->lock);
1663 btrfs_mod_outstanding_extents(BTRFS_I(inode), 1);
1664 spin_unlock(&BTRFS_I(inode)->lock);
1668 * extent_io.c merge_extent_hook, used to track merged delayed allocation
1669 * extents so we can keep track of new extents that are just merged onto old
1670 * extents, such as when we are doing sequential writes, so we can properly
1671 * account for the metadata space we'll need.
1673 static void btrfs_merge_extent_hook(void *private_data,
1674 struct extent_state *new,
1675 struct extent_state *other)
1677 struct inode *inode = private_data;
1678 u64 new_size, old_size;
1681 /* not delalloc, ignore it */
1682 if (!(other->state & EXTENT_DELALLOC))
1685 if (new->start > other->start)
1686 new_size = new->end - other->start + 1;
1688 new_size = other->end - new->start + 1;
1690 /* we're not bigger than the max, unreserve the space and go */
1691 if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1692 spin_lock(&BTRFS_I(inode)->lock);
1693 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1694 spin_unlock(&BTRFS_I(inode)->lock);
1699 * We have to add up either side to figure out how many extents were
1700 * accounted for before we merged into one big extent. If the number of
1701 * extents we accounted for is <= the amount we need for the new range
1702 * then we can return, otherwise drop. Think of it like this
1706 * So we've grown the extent by a MAX_SIZE extent, this would mean we
1707 * need 2 outstanding extents, on one side we have 1 and the other side
1708 * we have 1 so they are == and we can return. But in this case
1710 * [MAX_SIZE+4k][MAX_SIZE+4k]
1712 * Each range on their own accounts for 2 extents, but merged together
1713 * they are only 3 extents worth of accounting, so we need to drop in
1716 old_size = other->end - other->start + 1;
1717 num_extents = count_max_extents(old_size);
1718 old_size = new->end - new->start + 1;
1719 num_extents += count_max_extents(old_size);
1720 if (count_max_extents(new_size) >= num_extents)
1723 spin_lock(&BTRFS_I(inode)->lock);
1724 btrfs_mod_outstanding_extents(BTRFS_I(inode), -1);
1725 spin_unlock(&BTRFS_I(inode)->lock);
1728 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1729 struct inode *inode)
1731 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1733 spin_lock(&root->delalloc_lock);
1734 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1735 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1736 &root->delalloc_inodes);
1737 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1738 &BTRFS_I(inode)->runtime_flags);
1739 root->nr_delalloc_inodes++;
1740 if (root->nr_delalloc_inodes == 1) {
1741 spin_lock(&fs_info->delalloc_root_lock);
1742 BUG_ON(!list_empty(&root->delalloc_root));
1743 list_add_tail(&root->delalloc_root,
1744 &fs_info->delalloc_roots);
1745 spin_unlock(&fs_info->delalloc_root_lock);
1748 spin_unlock(&root->delalloc_lock);
1752 void __btrfs_del_delalloc_inode(struct btrfs_root *root,
1753 struct btrfs_inode *inode)
1755 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1757 if (!list_empty(&inode->delalloc_inodes)) {
1758 list_del_init(&inode->delalloc_inodes);
1759 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1760 &inode->runtime_flags);
1761 root->nr_delalloc_inodes--;
1762 if (!root->nr_delalloc_inodes) {
1763 ASSERT(list_empty(&root->delalloc_inodes));
1764 spin_lock(&fs_info->delalloc_root_lock);
1765 BUG_ON(list_empty(&root->delalloc_root));
1766 list_del_init(&root->delalloc_root);
1767 spin_unlock(&fs_info->delalloc_root_lock);
1772 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1773 struct btrfs_inode *inode)
1775 spin_lock(&root->delalloc_lock);
1776 __btrfs_del_delalloc_inode(root, inode);
1777 spin_unlock(&root->delalloc_lock);
1781 * extent_io.c set_bit_hook, used to track delayed allocation
1782 * bytes in this file, and to maintain the list of inodes that
1783 * have pending delalloc work to be done.
1785 static void btrfs_set_bit_hook(void *private_data,
1786 struct extent_state *state, unsigned *bits)
1788 struct inode *inode = private_data;
1790 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1792 if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1795 * set_bit and clear bit hooks normally require _irqsave/restore
1796 * but in this case, we are only testing for the DELALLOC
1797 * bit, which is only set or cleared with irqs on
1799 if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1800 struct btrfs_root *root = BTRFS_I(inode)->root;
1801 u64 len = state->end + 1 - state->start;
1802 u32 num_extents = count_max_extents(len);
1803 bool do_list = !btrfs_is_free_space_inode(BTRFS_I(inode));
1805 spin_lock(&BTRFS_I(inode)->lock);
1806 btrfs_mod_outstanding_extents(BTRFS_I(inode), num_extents);
1807 spin_unlock(&BTRFS_I(inode)->lock);
1809 /* For sanity tests */
1810 if (btrfs_is_testing(fs_info))
1813 percpu_counter_add_batch(&fs_info->delalloc_bytes, len,
1814 fs_info->delalloc_batch);
1815 spin_lock(&BTRFS_I(inode)->lock);
1816 BTRFS_I(inode)->delalloc_bytes += len;
1817 if (*bits & EXTENT_DEFRAG)
1818 BTRFS_I(inode)->defrag_bytes += len;
1819 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1820 &BTRFS_I(inode)->runtime_flags))
1821 btrfs_add_delalloc_inodes(root, inode);
1822 spin_unlock(&BTRFS_I(inode)->lock);
1825 if (!(state->state & EXTENT_DELALLOC_NEW) &&
1826 (*bits & EXTENT_DELALLOC_NEW)) {
1827 spin_lock(&BTRFS_I(inode)->lock);
1828 BTRFS_I(inode)->new_delalloc_bytes += state->end + 1 -
1830 spin_unlock(&BTRFS_I(inode)->lock);
1835 * extent_io.c clear_bit_hook, see set_bit_hook for why
1837 static void btrfs_clear_bit_hook(void *private_data,
1838 struct extent_state *state,
1841 struct btrfs_inode *inode = BTRFS_I((struct inode *)private_data);
1842 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1843 u64 len = state->end + 1 - state->start;
1844 u32 num_extents = count_max_extents(len);
1846 if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG)) {
1847 spin_lock(&inode->lock);
1848 inode->defrag_bytes -= len;
1849 spin_unlock(&inode->lock);
1853 * set_bit and clear bit hooks normally require _irqsave/restore
1854 * but in this case, we are only testing for the DELALLOC
1855 * bit, which is only set or cleared with irqs on
1857 if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1858 struct btrfs_root *root = inode->root;
1859 bool do_list = !btrfs_is_free_space_inode(inode);
1861 spin_lock(&inode->lock);
1862 btrfs_mod_outstanding_extents(inode, -num_extents);
1863 spin_unlock(&inode->lock);
1866 * We don't reserve metadata space for space cache inodes so we
1867 * don't need to call dellalloc_release_metadata if there is an
1870 if (*bits & EXTENT_CLEAR_META_RESV &&
1871 root != fs_info->tree_root)
1872 btrfs_delalloc_release_metadata(inode, len, false);
1874 /* For sanity tests. */
1875 if (btrfs_is_testing(fs_info))
1878 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID &&
1879 do_list && !(state->state & EXTENT_NORESERVE) &&
1880 (*bits & EXTENT_CLEAR_DATA_RESV))
1881 btrfs_free_reserved_data_space_noquota(
1885 percpu_counter_add_batch(&fs_info->delalloc_bytes, -len,
1886 fs_info->delalloc_batch);
1887 spin_lock(&inode->lock);
1888 inode->delalloc_bytes -= len;
1889 if (do_list && inode->delalloc_bytes == 0 &&
1890 test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1891 &inode->runtime_flags))
1892 btrfs_del_delalloc_inode(root, inode);
1893 spin_unlock(&inode->lock);
1896 if ((state->state & EXTENT_DELALLOC_NEW) &&
1897 (*bits & EXTENT_DELALLOC_NEW)) {
1898 spin_lock(&inode->lock);
1899 ASSERT(inode->new_delalloc_bytes >= len);
1900 inode->new_delalloc_bytes -= len;
1901 spin_unlock(&inode->lock);
1906 * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1907 * we don't create bios that span stripes or chunks
1909 * return 1 if page cannot be merged to bio
1910 * return 0 if page can be merged to bio
1911 * return error otherwise
1913 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1914 size_t size, struct bio *bio,
1915 unsigned long bio_flags)
1917 struct inode *inode = page->mapping->host;
1918 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1919 u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1924 if (bio_flags & EXTENT_BIO_COMPRESSED)
1927 length = bio->bi_iter.bi_size;
1928 map_length = length;
1929 ret = btrfs_map_block(fs_info, btrfs_op(bio), logical, &map_length,
1933 if (map_length < length + size)
1939 * in order to insert checksums into the metadata in large chunks,
1940 * we wait until bio submission time. All the pages in the bio are
1941 * checksummed and sums are attached onto the ordered extent record.
1943 * At IO completion time the cums attached on the ordered extent record
1944 * are inserted into the btree
1946 static blk_status_t btrfs_submit_bio_start(void *private_data, struct bio *bio,
1949 struct inode *inode = private_data;
1950 blk_status_t ret = 0;
1952 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
1953 BUG_ON(ret); /* -ENOMEM */
1958 * in order to insert checksums into the metadata in large chunks,
1959 * we wait until bio submission time. All the pages in the bio are
1960 * checksummed and sums are attached onto the ordered extent record.
1962 * At IO completion time the cums attached on the ordered extent record
1963 * are inserted into the btree
1965 static blk_status_t btrfs_submit_bio_done(void *private_data, struct bio *bio,
1968 struct inode *inode = private_data;
1969 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
1972 ret = btrfs_map_bio(fs_info, bio, mirror_num, 1);
1974 bio->bi_status = ret;
1981 * extent_io.c submission hook. This does the right thing for csum calculation
1982 * on write, or reading the csums from the tree before a read.
1984 * Rules about async/sync submit,
1985 * a) read: sync submit
1987 * b) write without checksum: sync submit
1989 * c) write with checksum:
1990 * c-1) if bio is issued by fsync: sync submit
1991 * (sync_writers != 0)
1993 * c-2) if root is reloc root: sync submit
1994 * (only in case of buffered IO)
1996 * c-3) otherwise: async submit
1998 static blk_status_t btrfs_submit_bio_hook(void *private_data, struct bio *bio,
1999 int mirror_num, unsigned long bio_flags,
2002 struct inode *inode = private_data;
2003 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2004 struct btrfs_root *root = BTRFS_I(inode)->root;
2005 enum btrfs_wq_endio_type metadata = BTRFS_WQ_ENDIO_DATA;
2006 blk_status_t ret = 0;
2008 int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
2010 skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
2012 if (btrfs_is_free_space_inode(BTRFS_I(inode)))
2013 metadata = BTRFS_WQ_ENDIO_FREE_SPACE;
2015 if (bio_op(bio) != REQ_OP_WRITE) {
2016 ret = btrfs_bio_wq_end_io(fs_info, bio, metadata);
2020 if (bio_flags & EXTENT_BIO_COMPRESSED) {
2021 ret = btrfs_submit_compressed_read(inode, bio,
2025 } else if (!skip_sum) {
2026 ret = btrfs_lookup_bio_sums(inode, bio, NULL);
2031 } else if (async && !skip_sum) {
2032 /* csum items have already been cloned */
2033 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
2035 /* we're doing a write, do the async checksumming */
2036 ret = btrfs_wq_submit_bio(fs_info, bio, mirror_num, bio_flags,
2038 btrfs_submit_bio_start,
2039 btrfs_submit_bio_done);
2041 } else if (!skip_sum) {
2042 ret = btrfs_csum_one_bio(inode, bio, 0, 0);
2048 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
2052 bio->bi_status = ret;
2059 * given a list of ordered sums record them in the inode. This happens
2060 * at IO completion time based on sums calculated at bio submission time.
2062 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
2063 struct inode *inode, struct list_head *list)
2065 struct btrfs_ordered_sum *sum;
2068 list_for_each_entry(sum, list, list) {
2069 trans->adding_csums = true;
2070 ret = btrfs_csum_file_blocks(trans,
2071 BTRFS_I(inode)->root->fs_info->csum_root, sum);
2072 trans->adding_csums = false;
2079 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
2080 unsigned int extra_bits,
2081 struct extent_state **cached_state, int dedupe)
2083 WARN_ON((end & (PAGE_SIZE - 1)) == 0);
2084 return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
2085 extra_bits, cached_state);
2088 /* see btrfs_writepage_start_hook for details on why this is required */
2089 struct btrfs_writepage_fixup {
2091 struct btrfs_work work;
2094 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
2096 struct btrfs_writepage_fixup *fixup;
2097 struct btrfs_ordered_extent *ordered;
2098 struct extent_state *cached_state = NULL;
2099 struct extent_changeset *data_reserved = NULL;
2101 struct inode *inode;
2106 fixup = container_of(work, struct btrfs_writepage_fixup, work);
2110 if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
2111 ClearPageChecked(page);
2115 inode = page->mapping->host;
2116 page_start = page_offset(page);
2117 page_end = page_offset(page) + PAGE_SIZE - 1;
2119 lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end,
2122 /* already ordered? We're done */
2123 if (PagePrivate2(page))
2126 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
2129 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
2130 page_end, &cached_state);
2132 btrfs_start_ordered_extent(inode, ordered, 1);
2133 btrfs_put_ordered_extent(ordered);
2137 ret = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
2140 mapping_set_error(page->mapping, ret);
2141 end_extent_writepage(page, ret, page_start, page_end);
2142 ClearPageChecked(page);
2146 ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
2149 mapping_set_error(page->mapping, ret);
2150 end_extent_writepage(page, ret, page_start, page_end);
2151 ClearPageChecked(page);
2155 ClearPageChecked(page);
2156 set_page_dirty(page);
2157 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, false);
2159 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
2165 extent_changeset_free(data_reserved);
2169 * There are a few paths in the higher layers of the kernel that directly
2170 * set the page dirty bit without asking the filesystem if it is a
2171 * good idea. This causes problems because we want to make sure COW
2172 * properly happens and the data=ordered rules are followed.
2174 * In our case any range that doesn't have the ORDERED bit set
2175 * hasn't been properly setup for IO. We kick off an async process
2176 * to fix it up. The async helper will wait for ordered extents, set
2177 * the delalloc bit and make it safe to write the page.
2179 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
2181 struct inode *inode = page->mapping->host;
2182 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2183 struct btrfs_writepage_fixup *fixup;
2185 /* this page is properly in the ordered list */
2186 if (TestClearPagePrivate2(page))
2189 if (PageChecked(page))
2192 fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2196 SetPageChecked(page);
2198 btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2199 btrfs_writepage_fixup_worker, NULL, NULL);
2201 btrfs_queue_work(fs_info->fixup_workers, &fixup->work);
2205 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2206 struct inode *inode, u64 file_pos,
2207 u64 disk_bytenr, u64 disk_num_bytes,
2208 u64 num_bytes, u64 ram_bytes,
2209 u8 compression, u8 encryption,
2210 u16 other_encoding, int extent_type)
2212 struct btrfs_root *root = BTRFS_I(inode)->root;
2213 struct btrfs_file_extent_item *fi;
2214 struct btrfs_path *path;
2215 struct extent_buffer *leaf;
2216 struct btrfs_key ins;
2218 int extent_inserted = 0;
2221 path = btrfs_alloc_path();
2226 * we may be replacing one extent in the tree with another.
2227 * The new extent is pinned in the extent map, and we don't want
2228 * to drop it from the cache until it is completely in the btree.
2230 * So, tell btrfs_drop_extents to leave this extent in the cache.
2231 * the caller is expected to unpin it and allow it to be merged
2234 ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2235 file_pos + num_bytes, NULL, 0,
2236 1, sizeof(*fi), &extent_inserted);
2240 if (!extent_inserted) {
2241 ins.objectid = btrfs_ino(BTRFS_I(inode));
2242 ins.offset = file_pos;
2243 ins.type = BTRFS_EXTENT_DATA_KEY;
2245 path->leave_spinning = 1;
2246 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2251 leaf = path->nodes[0];
2252 fi = btrfs_item_ptr(leaf, path->slots[0],
2253 struct btrfs_file_extent_item);
2254 btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2255 btrfs_set_file_extent_type(leaf, fi, extent_type);
2256 btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2257 btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2258 btrfs_set_file_extent_offset(leaf, fi, 0);
2259 btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2260 btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2261 btrfs_set_file_extent_compression(leaf, fi, compression);
2262 btrfs_set_file_extent_encryption(leaf, fi, encryption);
2263 btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2265 btrfs_mark_buffer_dirty(leaf);
2266 btrfs_release_path(path);
2268 inode_add_bytes(inode, num_bytes);
2270 ins.objectid = disk_bytenr;
2271 ins.offset = disk_num_bytes;
2272 ins.type = BTRFS_EXTENT_ITEM_KEY;
2275 * Release the reserved range from inode dirty range map, as it is
2276 * already moved into delayed_ref_head
2278 ret = btrfs_qgroup_release_data(inode, file_pos, ram_bytes);
2282 ret = btrfs_alloc_reserved_file_extent(trans, root,
2283 btrfs_ino(BTRFS_I(inode)),
2284 file_pos, qg_released, &ins);
2286 btrfs_free_path(path);
2291 /* snapshot-aware defrag */
2292 struct sa_defrag_extent_backref {
2293 struct rb_node node;
2294 struct old_sa_defrag_extent *old;
2303 struct old_sa_defrag_extent {
2304 struct list_head list;
2305 struct new_sa_defrag_extent *new;
2314 struct new_sa_defrag_extent {
2315 struct rb_root root;
2316 struct list_head head;
2317 struct btrfs_path *path;
2318 struct inode *inode;
2326 static int backref_comp(struct sa_defrag_extent_backref *b1,
2327 struct sa_defrag_extent_backref *b2)
2329 if (b1->root_id < b2->root_id)
2331 else if (b1->root_id > b2->root_id)
2334 if (b1->inum < b2->inum)
2336 else if (b1->inum > b2->inum)
2339 if (b1->file_pos < b2->file_pos)
2341 else if (b1->file_pos > b2->file_pos)
2345 * [------------------------------] ===> (a range of space)
2346 * |<--->| |<---->| =============> (fs/file tree A)
2347 * |<---------------------------->| ===> (fs/file tree B)
2349 * A range of space can refer to two file extents in one tree while
2350 * refer to only one file extent in another tree.
2352 * So we may process a disk offset more than one time(two extents in A)
2353 * and locate at the same extent(one extent in B), then insert two same
2354 * backrefs(both refer to the extent in B).
2359 static void backref_insert(struct rb_root *root,
2360 struct sa_defrag_extent_backref *backref)
2362 struct rb_node **p = &root->rb_node;
2363 struct rb_node *parent = NULL;
2364 struct sa_defrag_extent_backref *entry;
2369 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2371 ret = backref_comp(backref, entry);
2375 p = &(*p)->rb_right;
2378 rb_link_node(&backref->node, parent, p);
2379 rb_insert_color(&backref->node, root);
2383 * Note the backref might has changed, and in this case we just return 0.
2385 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2388 struct btrfs_file_extent_item *extent;
2389 struct old_sa_defrag_extent *old = ctx;
2390 struct new_sa_defrag_extent *new = old->new;
2391 struct btrfs_path *path = new->path;
2392 struct btrfs_key key;
2393 struct btrfs_root *root;
2394 struct sa_defrag_extent_backref *backref;
2395 struct extent_buffer *leaf;
2396 struct inode *inode = new->inode;
2397 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2403 if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2404 inum == btrfs_ino(BTRFS_I(inode)))
2407 key.objectid = root_id;
2408 key.type = BTRFS_ROOT_ITEM_KEY;
2409 key.offset = (u64)-1;
2411 root = btrfs_read_fs_root_no_name(fs_info, &key);
2413 if (PTR_ERR(root) == -ENOENT)
2416 btrfs_debug(fs_info, "inum=%llu, offset=%llu, root_id=%llu",
2417 inum, offset, root_id);
2418 return PTR_ERR(root);
2421 key.objectid = inum;
2422 key.type = BTRFS_EXTENT_DATA_KEY;
2423 if (offset > (u64)-1 << 32)
2426 key.offset = offset;
2428 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2429 if (WARN_ON(ret < 0))
2436 leaf = path->nodes[0];
2437 slot = path->slots[0];
2439 if (slot >= btrfs_header_nritems(leaf)) {
2440 ret = btrfs_next_leaf(root, path);
2443 } else if (ret > 0) {
2452 btrfs_item_key_to_cpu(leaf, &key, slot);
2454 if (key.objectid > inum)
2457 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2460 extent = btrfs_item_ptr(leaf, slot,
2461 struct btrfs_file_extent_item);
2463 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2467 * 'offset' refers to the exact key.offset,
2468 * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2469 * (key.offset - extent_offset).
2471 if (key.offset != offset)
2474 extent_offset = btrfs_file_extent_offset(leaf, extent);
2475 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2477 if (extent_offset >= old->extent_offset + old->offset +
2478 old->len || extent_offset + num_bytes <=
2479 old->extent_offset + old->offset)
2484 backref = kmalloc(sizeof(*backref), GFP_NOFS);
2490 backref->root_id = root_id;
2491 backref->inum = inum;
2492 backref->file_pos = offset;
2493 backref->num_bytes = num_bytes;
2494 backref->extent_offset = extent_offset;
2495 backref->generation = btrfs_file_extent_generation(leaf, extent);
2497 backref_insert(&new->root, backref);
2500 btrfs_release_path(path);
2505 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2506 struct new_sa_defrag_extent *new)
2508 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2509 struct old_sa_defrag_extent *old, *tmp;
2514 list_for_each_entry_safe(old, tmp, &new->head, list) {
2515 ret = iterate_inodes_from_logical(old->bytenr +
2516 old->extent_offset, fs_info,
2517 path, record_one_backref,
2519 if (ret < 0 && ret != -ENOENT)
2522 /* no backref to be processed for this extent */
2524 list_del(&old->list);
2529 if (list_empty(&new->head))
2535 static int relink_is_mergable(struct extent_buffer *leaf,
2536 struct btrfs_file_extent_item *fi,
2537 struct new_sa_defrag_extent *new)
2539 if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2542 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2545 if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2548 if (btrfs_file_extent_encryption(leaf, fi) ||
2549 btrfs_file_extent_other_encoding(leaf, fi))
2556 * Note the backref might has changed, and in this case we just return 0.
2558 static noinline int relink_extent_backref(struct btrfs_path *path,
2559 struct sa_defrag_extent_backref *prev,
2560 struct sa_defrag_extent_backref *backref)
2562 struct btrfs_file_extent_item *extent;
2563 struct btrfs_file_extent_item *item;
2564 struct btrfs_ordered_extent *ordered;
2565 struct btrfs_trans_handle *trans;
2566 struct btrfs_root *root;
2567 struct btrfs_key key;
2568 struct extent_buffer *leaf;
2569 struct old_sa_defrag_extent *old = backref->old;
2570 struct new_sa_defrag_extent *new = old->new;
2571 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2572 struct inode *inode;
2573 struct extent_state *cached = NULL;
2582 if (prev && prev->root_id == backref->root_id &&
2583 prev->inum == backref->inum &&
2584 prev->file_pos + prev->num_bytes == backref->file_pos)
2587 /* step 1: get root */
2588 key.objectid = backref->root_id;
2589 key.type = BTRFS_ROOT_ITEM_KEY;
2590 key.offset = (u64)-1;
2592 index = srcu_read_lock(&fs_info->subvol_srcu);
2594 root = btrfs_read_fs_root_no_name(fs_info, &key);
2596 srcu_read_unlock(&fs_info->subvol_srcu, index);
2597 if (PTR_ERR(root) == -ENOENT)
2599 return PTR_ERR(root);
2602 if (btrfs_root_readonly(root)) {
2603 srcu_read_unlock(&fs_info->subvol_srcu, index);
2607 /* step 2: get inode */
2608 key.objectid = backref->inum;
2609 key.type = BTRFS_INODE_ITEM_KEY;
2612 inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2613 if (IS_ERR(inode)) {
2614 srcu_read_unlock(&fs_info->subvol_srcu, index);
2618 srcu_read_unlock(&fs_info->subvol_srcu, index);
2620 /* step 3: relink backref */
2621 lock_start = backref->file_pos;
2622 lock_end = backref->file_pos + backref->num_bytes - 1;
2623 lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2626 ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2628 btrfs_put_ordered_extent(ordered);
2632 trans = btrfs_join_transaction(root);
2633 if (IS_ERR(trans)) {
2634 ret = PTR_ERR(trans);
2638 key.objectid = backref->inum;
2639 key.type = BTRFS_EXTENT_DATA_KEY;
2640 key.offset = backref->file_pos;
2642 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2645 } else if (ret > 0) {
2650 extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2651 struct btrfs_file_extent_item);
2653 if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2654 backref->generation)
2657 btrfs_release_path(path);
2659 start = backref->file_pos;
2660 if (backref->extent_offset < old->extent_offset + old->offset)
2661 start += old->extent_offset + old->offset -
2662 backref->extent_offset;
2664 len = min(backref->extent_offset + backref->num_bytes,
2665 old->extent_offset + old->offset + old->len);
2666 len -= max(backref->extent_offset, old->extent_offset + old->offset);
2668 ret = btrfs_drop_extents(trans, root, inode, start,
2673 key.objectid = btrfs_ino(BTRFS_I(inode));
2674 key.type = BTRFS_EXTENT_DATA_KEY;
2677 path->leave_spinning = 1;
2679 struct btrfs_file_extent_item *fi;
2681 struct btrfs_key found_key;
2683 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2688 leaf = path->nodes[0];
2689 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2691 fi = btrfs_item_ptr(leaf, path->slots[0],
2692 struct btrfs_file_extent_item);
2693 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2695 if (extent_len + found_key.offset == start &&
2696 relink_is_mergable(leaf, fi, new)) {
2697 btrfs_set_file_extent_num_bytes(leaf, fi,
2699 btrfs_mark_buffer_dirty(leaf);
2700 inode_add_bytes(inode, len);
2706 btrfs_release_path(path);
2711 ret = btrfs_insert_empty_item(trans, root, path, &key,
2714 btrfs_abort_transaction(trans, ret);
2718 leaf = path->nodes[0];
2719 item = btrfs_item_ptr(leaf, path->slots[0],
2720 struct btrfs_file_extent_item);
2721 btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2722 btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2723 btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2724 btrfs_set_file_extent_num_bytes(leaf, item, len);
2725 btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2726 btrfs_set_file_extent_generation(leaf, item, trans->transid);
2727 btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2728 btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2729 btrfs_set_file_extent_encryption(leaf, item, 0);
2730 btrfs_set_file_extent_other_encoding(leaf, item, 0);
2732 btrfs_mark_buffer_dirty(leaf);
2733 inode_add_bytes(inode, len);
2734 btrfs_release_path(path);
2736 ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2738 backref->root_id, backref->inum,
2739 new->file_pos); /* start - extent_offset */
2741 btrfs_abort_transaction(trans, ret);
2747 btrfs_release_path(path);
2748 path->leave_spinning = 0;
2749 btrfs_end_transaction(trans);
2751 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2757 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2759 struct old_sa_defrag_extent *old, *tmp;
2764 list_for_each_entry_safe(old, tmp, &new->head, list) {
2770 static void relink_file_extents(struct new_sa_defrag_extent *new)
2772 struct btrfs_fs_info *fs_info = btrfs_sb(new->inode->i_sb);
2773 struct btrfs_path *path;
2774 struct sa_defrag_extent_backref *backref;
2775 struct sa_defrag_extent_backref *prev = NULL;
2776 struct inode *inode;
2777 struct rb_node *node;
2782 path = btrfs_alloc_path();
2786 if (!record_extent_backrefs(path, new)) {
2787 btrfs_free_path(path);
2790 btrfs_release_path(path);
2793 node = rb_first(&new->root);
2796 rb_erase(node, &new->root);
2798 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2800 ret = relink_extent_backref(path, prev, backref);
2813 btrfs_free_path(path);
2815 free_sa_defrag_extent(new);
2817 atomic_dec(&fs_info->defrag_running);
2818 wake_up(&fs_info->transaction_wait);
2821 static struct new_sa_defrag_extent *
2822 record_old_file_extents(struct inode *inode,
2823 struct btrfs_ordered_extent *ordered)
2825 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2826 struct btrfs_root *root = BTRFS_I(inode)->root;
2827 struct btrfs_path *path;
2828 struct btrfs_key key;
2829 struct old_sa_defrag_extent *old;
2830 struct new_sa_defrag_extent *new;
2833 new = kmalloc(sizeof(*new), GFP_NOFS);
2838 new->file_pos = ordered->file_offset;
2839 new->len = ordered->len;
2840 new->bytenr = ordered->start;
2841 new->disk_len = ordered->disk_len;
2842 new->compress_type = ordered->compress_type;
2843 new->root = RB_ROOT;
2844 INIT_LIST_HEAD(&new->head);
2846 path = btrfs_alloc_path();
2850 key.objectid = btrfs_ino(BTRFS_I(inode));
2851 key.type = BTRFS_EXTENT_DATA_KEY;
2852 key.offset = new->file_pos;
2854 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2857 if (ret > 0 && path->slots[0] > 0)
2860 /* find out all the old extents for the file range */
2862 struct btrfs_file_extent_item *extent;
2863 struct extent_buffer *l;
2872 slot = path->slots[0];
2874 if (slot >= btrfs_header_nritems(l)) {
2875 ret = btrfs_next_leaf(root, path);
2883 btrfs_item_key_to_cpu(l, &key, slot);
2885 if (key.objectid != btrfs_ino(BTRFS_I(inode)))
2887 if (key.type != BTRFS_EXTENT_DATA_KEY)
2889 if (key.offset >= new->file_pos + new->len)
2892 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2894 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2895 if (key.offset + num_bytes < new->file_pos)
2898 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2902 extent_offset = btrfs_file_extent_offset(l, extent);
2904 old = kmalloc(sizeof(*old), GFP_NOFS);
2908 offset = max(new->file_pos, key.offset);
2909 end = min(new->file_pos + new->len, key.offset + num_bytes);
2911 old->bytenr = disk_bytenr;
2912 old->extent_offset = extent_offset;
2913 old->offset = offset - key.offset;
2914 old->len = end - offset;
2917 list_add_tail(&old->list, &new->head);
2923 btrfs_free_path(path);
2924 atomic_inc(&fs_info->defrag_running);
2929 btrfs_free_path(path);
2931 free_sa_defrag_extent(new);
2935 static void btrfs_release_delalloc_bytes(struct btrfs_fs_info *fs_info,
2938 struct btrfs_block_group_cache *cache;
2940 cache = btrfs_lookup_block_group(fs_info, start);
2943 spin_lock(&cache->lock);
2944 cache->delalloc_bytes -= len;
2945 spin_unlock(&cache->lock);
2947 btrfs_put_block_group(cache);
2950 /* as ordered data IO finishes, this gets called so we can finish
2951 * an ordered extent if the range of bytes in the file it covers are
2954 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2956 struct inode *inode = ordered_extent->inode;
2957 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2958 struct btrfs_root *root = BTRFS_I(inode)->root;
2959 struct btrfs_trans_handle *trans = NULL;
2960 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2961 struct extent_state *cached_state = NULL;
2962 struct new_sa_defrag_extent *new = NULL;
2963 int compress_type = 0;
2965 u64 logical_len = ordered_extent->len;
2967 bool truncated = false;
2968 bool range_locked = false;
2969 bool clear_new_delalloc_bytes = false;
2971 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2972 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags) &&
2973 !test_bit(BTRFS_ORDERED_DIRECT, &ordered_extent->flags))
2974 clear_new_delalloc_bytes = true;
2976 nolock = btrfs_is_free_space_inode(BTRFS_I(inode));
2978 if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2983 btrfs_free_io_failure_record(BTRFS_I(inode),
2984 ordered_extent->file_offset,
2985 ordered_extent->file_offset +
2986 ordered_extent->len - 1);
2988 if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2990 logical_len = ordered_extent->truncated_len;
2991 /* Truncated the entire extent, don't bother adding */
2996 if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2997 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
3000 * For mwrite(mmap + memset to write) case, we still reserve
3001 * space for NOCOW range.
3002 * As NOCOW won't cause a new delayed ref, just free the space
3004 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3005 ordered_extent->len);
3006 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3008 trans = btrfs_join_transaction_nolock(root);
3010 trans = btrfs_join_transaction(root);
3011 if (IS_ERR(trans)) {
3012 ret = PTR_ERR(trans);
3016 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3017 ret = btrfs_update_inode_fallback(trans, root, inode);
3018 if (ret) /* -ENOMEM or corruption */
3019 btrfs_abort_transaction(trans, ret);
3023 range_locked = true;
3024 lock_extent_bits(io_tree, ordered_extent->file_offset,
3025 ordered_extent->file_offset + ordered_extent->len - 1,
3028 ret = test_range_bit(io_tree, ordered_extent->file_offset,
3029 ordered_extent->file_offset + ordered_extent->len - 1,
3030 EXTENT_DEFRAG, 0, cached_state);
3032 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
3033 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
3034 /* the inode is shared */
3035 new = record_old_file_extents(inode, ordered_extent);
3037 clear_extent_bit(io_tree, ordered_extent->file_offset,
3038 ordered_extent->file_offset + ordered_extent->len - 1,
3039 EXTENT_DEFRAG, 0, 0, &cached_state);
3043 trans = btrfs_join_transaction_nolock(root);
3045 trans = btrfs_join_transaction(root);
3046 if (IS_ERR(trans)) {
3047 ret = PTR_ERR(trans);
3052 trans->block_rsv = &BTRFS_I(inode)->block_rsv;
3054 if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
3055 compress_type = ordered_extent->compress_type;
3056 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
3057 BUG_ON(compress_type);
3058 btrfs_qgroup_free_data(inode, NULL, ordered_extent->file_offset,
3059 ordered_extent->len);
3060 ret = btrfs_mark_extent_written(trans, BTRFS_I(inode),
3061 ordered_extent->file_offset,
3062 ordered_extent->file_offset +
3065 BUG_ON(root == fs_info->tree_root);
3066 ret = insert_reserved_file_extent(trans, inode,
3067 ordered_extent->file_offset,
3068 ordered_extent->start,
3069 ordered_extent->disk_len,
3070 logical_len, logical_len,
3071 compress_type, 0, 0,
3072 BTRFS_FILE_EXTENT_REG);
3074 btrfs_release_delalloc_bytes(fs_info,
3075 ordered_extent->start,
3076 ordered_extent->disk_len);
3078 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
3079 ordered_extent->file_offset, ordered_extent->len,
3082 btrfs_abort_transaction(trans, ret);
3086 ret = add_pending_csums(trans, inode, &ordered_extent->list);
3088 btrfs_abort_transaction(trans, ret);
3092 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
3093 ret = btrfs_update_inode_fallback(trans, root, inode);
3094 if (ret) { /* -ENOMEM or corruption */
3095 btrfs_abort_transaction(trans, ret);
3100 if (range_locked || clear_new_delalloc_bytes) {
3101 unsigned int clear_bits = 0;
3104 clear_bits |= EXTENT_LOCKED;
3105 if (clear_new_delalloc_bytes)
3106 clear_bits |= EXTENT_DELALLOC_NEW;
3107 clear_extent_bit(&BTRFS_I(inode)->io_tree,
3108 ordered_extent->file_offset,
3109 ordered_extent->file_offset +
3110 ordered_extent->len - 1,
3112 (clear_bits & EXTENT_LOCKED) ? 1 : 0,
3117 btrfs_end_transaction(trans);
3119 if (ret || truncated) {
3123 start = ordered_extent->file_offset + logical_len;
3125 start = ordered_extent->file_offset;
3126 end = ordered_extent->file_offset + ordered_extent->len - 1;
3127 clear_extent_uptodate(io_tree, start, end, NULL);
3129 /* Drop the cache for the part of the extent we didn't write. */
3130 btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3133 * If the ordered extent had an IOERR or something else went
3134 * wrong we need to return the space for this ordered extent
3135 * back to the allocator. We only free the extent in the
3136 * truncated case if we didn't write out the extent at all.
3138 if ((ret || !logical_len) &&
3139 !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
3140 !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
3141 btrfs_free_reserved_extent(fs_info,
3142 ordered_extent->start,
3143 ordered_extent->disk_len, 1);
3148 * This needs to be done to make sure anybody waiting knows we are done
3149 * updating everything for this ordered extent.
3151 btrfs_remove_ordered_extent(inode, ordered_extent);
3153 /* for snapshot-aware defrag */
3156 free_sa_defrag_extent(new);
3157 atomic_dec(&fs_info->defrag_running);
3159 relink_file_extents(new);
3164 btrfs_put_ordered_extent(ordered_extent);
3165 /* once for the tree */
3166 btrfs_put_ordered_extent(ordered_extent);
3168 /* Try to release some metadata so we don't get an OOM but don't wait */
3169 btrfs_btree_balance_dirty_nodelay(fs_info);
3174 static void finish_ordered_fn(struct btrfs_work *work)
3176 struct btrfs_ordered_extent *ordered_extent;
3177 ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
3178 btrfs_finish_ordered_io(ordered_extent);
3181 static void btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
3182 struct extent_state *state, int uptodate)
3184 struct inode *inode = page->mapping->host;
3185 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3186 struct btrfs_ordered_extent *ordered_extent = NULL;
3187 struct btrfs_workqueue *wq;
3188 btrfs_work_func_t func;
3190 trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
3192 ClearPagePrivate2(page);
3193 if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
3194 end - start + 1, uptodate))
3197 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
3198 wq = fs_info->endio_freespace_worker;
3199 func = btrfs_freespace_write_helper;
3201 wq = fs_info->endio_write_workers;
3202 func = btrfs_endio_write_helper;
3205 btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
3207 btrfs_queue_work(wq, &ordered_extent->work);
3210 static int __readpage_endio_check(struct inode *inode,
3211 struct btrfs_io_bio *io_bio,
3212 int icsum, struct page *page,
3213 int pgoff, u64 start, size_t len)
3219 csum_expected = *(((u32 *)io_bio->csum) + icsum);
3221 kaddr = kmap_atomic(page);
3222 csum = btrfs_csum_data(kaddr + pgoff, csum, len);
3223 btrfs_csum_final(csum, (u8 *)&csum);
3224 if (csum != csum_expected)
3227 kunmap_atomic(kaddr);
3230 btrfs_print_data_csum_error(BTRFS_I(inode), start, csum, csum_expected,
3231 io_bio->mirror_num);
3232 memset(kaddr + pgoff, 1, len);
3233 flush_dcache_page(page);
3234 kunmap_atomic(kaddr);
3239 * when reads are done, we need to check csums to verify the data is correct
3240 * if there's a match, we allow the bio to finish. If not, the code in
3241 * extent_io.c will try to find good copies for us.
3243 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3244 u64 phy_offset, struct page *page,
3245 u64 start, u64 end, int mirror)
3247 size_t offset = start - page_offset(page);
3248 struct inode *inode = page->mapping->host;
3249 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3250 struct btrfs_root *root = BTRFS_I(inode)->root;
3252 if (PageChecked(page)) {
3253 ClearPageChecked(page);
3257 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3260 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3261 test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3262 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM);
3266 phy_offset >>= inode->i_sb->s_blocksize_bits;
3267 return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3268 start, (size_t)(end - start + 1));
3272 * btrfs_add_delayed_iput - perform a delayed iput on @inode
3274 * @inode: The inode we want to perform iput on
3276 * This function uses the generic vfs_inode::i_count to track whether we should
3277 * just decrement it (in case it's > 1) or if this is the last iput then link
3278 * the inode to the delayed iput machinery. Delayed iputs are processed at
3279 * transaction commit time/superblock commit/cleaner kthread.
3281 void btrfs_add_delayed_iput(struct inode *inode)
3283 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3284 struct btrfs_inode *binode = BTRFS_I(inode);
3286 if (atomic_add_unless(&inode->i_count, -1, 1))
3289 spin_lock(&fs_info->delayed_iput_lock);
3290 ASSERT(list_empty(&binode->delayed_iput));
3291 list_add_tail(&binode->delayed_iput, &fs_info->delayed_iputs);
3292 spin_unlock(&fs_info->delayed_iput_lock);
3295 void btrfs_run_delayed_iputs(struct btrfs_fs_info *fs_info)
3298 spin_lock(&fs_info->delayed_iput_lock);
3299 while (!list_empty(&fs_info->delayed_iputs)) {
3300 struct btrfs_inode *inode;
3302 inode = list_first_entry(&fs_info->delayed_iputs,
3303 struct btrfs_inode, delayed_iput);
3304 list_del_init(&inode->delayed_iput);
3305 spin_unlock(&fs_info->delayed_iput_lock);
3306 iput(&inode->vfs_inode);
3307 spin_lock(&fs_info->delayed_iput_lock);
3309 spin_unlock(&fs_info->delayed_iput_lock);
3313 * This creates an orphan entry for the given inode in case something goes wrong
3314 * in the middle of an unlink.
3316 int btrfs_orphan_add(struct btrfs_trans_handle *trans,
3317 struct btrfs_inode *inode)
3321 ret = btrfs_insert_orphan_item(trans, inode->root, btrfs_ino(inode));
3322 if (ret && ret != -EEXIST) {
3323 btrfs_abort_transaction(trans, ret);
3331 * We have done the delete so we can go ahead and remove the orphan item for
3332 * this particular inode.
3334 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3335 struct btrfs_inode *inode)
3337 return btrfs_del_orphan_item(trans, inode->root, btrfs_ino(inode));
3341 * this cleans up any orphans that may be left on the list from the last use
3344 int btrfs_orphan_cleanup(struct btrfs_root *root)
3346 struct btrfs_fs_info *fs_info = root->fs_info;
3347 struct btrfs_path *path;
3348 struct extent_buffer *leaf;
3349 struct btrfs_key key, found_key;
3350 struct btrfs_trans_handle *trans;
3351 struct inode *inode;
3352 u64 last_objectid = 0;
3353 int ret = 0, nr_unlink = 0;
3355 if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3358 path = btrfs_alloc_path();
3363 path->reada = READA_BACK;
3365 key.objectid = BTRFS_ORPHAN_OBJECTID;
3366 key.type = BTRFS_ORPHAN_ITEM_KEY;
3367 key.offset = (u64)-1;
3370 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3375 * if ret == 0 means we found what we were searching for, which
3376 * is weird, but possible, so only screw with path if we didn't
3377 * find the key and see if we have stuff that matches
3381 if (path->slots[0] == 0)
3386 /* pull out the item */
3387 leaf = path->nodes[0];
3388 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3390 /* make sure the item matches what we want */
3391 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3393 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3396 /* release the path since we're done with it */
3397 btrfs_release_path(path);
3400 * this is where we are basically btrfs_lookup, without the
3401 * crossing root thing. we store the inode number in the
3402 * offset of the orphan item.
3405 if (found_key.offset == last_objectid) {
3407 "Error removing orphan entry, stopping orphan cleanup");
3412 last_objectid = found_key.offset;
3414 found_key.objectid = found_key.offset;
3415 found_key.type = BTRFS_INODE_ITEM_KEY;
3416 found_key.offset = 0;
3417 inode = btrfs_iget(fs_info->sb, &found_key, root, NULL);
3418 ret = PTR_ERR_OR_ZERO(inode);
3419 if (ret && ret != -ENOENT)
3422 if (ret == -ENOENT && root == fs_info->tree_root) {
3423 struct btrfs_root *dead_root;
3424 struct btrfs_fs_info *fs_info = root->fs_info;
3425 int is_dead_root = 0;
3428 * this is an orphan in the tree root. Currently these
3429 * could come from 2 sources:
3430 * a) a snapshot deletion in progress
3431 * b) a free space cache inode
3432 * We need to distinguish those two, as the snapshot
3433 * orphan must not get deleted.
3434 * find_dead_roots already ran before us, so if this
3435 * is a snapshot deletion, we should find the root
3436 * in the dead_roots list
3438 spin_lock(&fs_info->trans_lock);
3439 list_for_each_entry(dead_root, &fs_info->dead_roots,
3441 if (dead_root->root_key.objectid ==
3442 found_key.objectid) {
3447 spin_unlock(&fs_info->trans_lock);
3449 /* prevent this orphan from being found again */
3450 key.offset = found_key.objectid - 1;
3457 * If we have an inode with links, there are a couple of
3458 * possibilities. Old kernels (before v3.12) used to create an
3459 * orphan item for truncate indicating that there were possibly
3460 * extent items past i_size that needed to be deleted. In v3.12,
3461 * truncate was changed to update i_size in sync with the extent
3462 * items, but the (useless) orphan item was still created. Since
3463 * v4.18, we don't create the orphan item for truncate at all.
3465 * So, this item could mean that we need to do a truncate, but
3466 * only if this filesystem was last used on a pre-v3.12 kernel
3467 * and was not cleanly unmounted. The odds of that are quite
3468 * slim, and it's a pain to do the truncate now, so just delete
3471 * It's also possible that this orphan item was supposed to be
3472 * deleted but wasn't. The inode number may have been reused,
3473 * but either way, we can delete the orphan item.
3475 if (ret == -ENOENT || inode->i_nlink) {
3478 trans = btrfs_start_transaction(root, 1);
3479 if (IS_ERR(trans)) {
3480 ret = PTR_ERR(trans);
3483 btrfs_debug(fs_info, "auto deleting %Lu",
3484 found_key.objectid);
3485 ret = btrfs_del_orphan_item(trans, root,
3486 found_key.objectid);
3487 btrfs_end_transaction(trans);
3495 /* this will do delete_inode and everything for us */
3500 /* release the path since we're done with it */
3501 btrfs_release_path(path);
3503 root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3505 if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3506 trans = btrfs_join_transaction(root);
3508 btrfs_end_transaction(trans);
3512 btrfs_debug(fs_info, "unlinked %d orphans", nr_unlink);
3516 btrfs_err(fs_info, "could not do orphan cleanup %d", ret);
3517 btrfs_free_path(path);
3522 * very simple check to peek ahead in the leaf looking for xattrs. If we
3523 * don't find any xattrs, we know there can't be any acls.
3525 * slot is the slot the inode is in, objectid is the objectid of the inode
3527 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3528 int slot, u64 objectid,
3529 int *first_xattr_slot)
3531 u32 nritems = btrfs_header_nritems(leaf);
3532 struct btrfs_key found_key;
3533 static u64 xattr_access = 0;
3534 static u64 xattr_default = 0;
3537 if (!xattr_access) {
3538 xattr_access = btrfs_name_hash(XATTR_NAME_POSIX_ACL_ACCESS,
3539 strlen(XATTR_NAME_POSIX_ACL_ACCESS));
3540 xattr_default = btrfs_name_hash(XATTR_NAME_POSIX_ACL_DEFAULT,
3541 strlen(XATTR_NAME_POSIX_ACL_DEFAULT));
3545 *first_xattr_slot = -1;
3546 while (slot < nritems) {
3547 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3549 /* we found a different objectid, there must not be acls */
3550 if (found_key.objectid != objectid)
3553 /* we found an xattr, assume we've got an acl */
3554 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3555 if (*first_xattr_slot == -1)
3556 *first_xattr_slot = slot;
3557 if (found_key.offset == xattr_access ||
3558 found_key.offset == xattr_default)
3563 * we found a key greater than an xattr key, there can't
3564 * be any acls later on
3566 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3573 * it goes inode, inode backrefs, xattrs, extents,
3574 * so if there are a ton of hard links to an inode there can
3575 * be a lot of backrefs. Don't waste time searching too hard,
3576 * this is just an optimization
3581 /* we hit the end of the leaf before we found an xattr or
3582 * something larger than an xattr. We have to assume the inode
3585 if (*first_xattr_slot == -1)
3586 *first_xattr_slot = slot;
3591 * read an inode from the btree into the in-memory inode
3593 static int btrfs_read_locked_inode(struct inode *inode)
3595 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3596 struct btrfs_path *path;
3597 struct extent_buffer *leaf;
3598 struct btrfs_inode_item *inode_item;
3599 struct btrfs_root *root = BTRFS_I(inode)->root;
3600 struct btrfs_key location;
3605 bool filled = false;
3606 int first_xattr_slot;
3608 ret = btrfs_fill_inode(inode, &rdev);
3612 path = btrfs_alloc_path();
3618 memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3620 ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3627 leaf = path->nodes[0];
3632 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3633 struct btrfs_inode_item);
3634 inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3635 set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3636 i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3637 i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3638 btrfs_i_size_write(BTRFS_I(inode), btrfs_inode_size(leaf, inode_item));
3640 inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3641 inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3643 inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3644 inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3646 inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3647 inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3649 BTRFS_I(inode)->i_otime.tv_sec =
3650 btrfs_timespec_sec(leaf, &inode_item->otime);
3651 BTRFS_I(inode)->i_otime.tv_nsec =
3652 btrfs_timespec_nsec(leaf, &inode_item->otime);
3654 inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3655 BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3656 BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3658 inode_set_iversion_queried(inode,
3659 btrfs_inode_sequence(leaf, inode_item));
3660 inode->i_generation = BTRFS_I(inode)->generation;
3662 rdev = btrfs_inode_rdev(leaf, inode_item);
3664 BTRFS_I(inode)->index_cnt = (u64)-1;
3665 BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3669 * If we were modified in the current generation and evicted from memory
3670 * and then re-read we need to do a full sync since we don't have any
3671 * idea about which extents were modified before we were evicted from
3674 * This is required for both inode re-read from disk and delayed inode
3675 * in delayed_nodes_tree.
3677 if (BTRFS_I(inode)->last_trans == fs_info->generation)
3678 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3679 &BTRFS_I(inode)->runtime_flags);
3682 * We don't persist the id of the transaction where an unlink operation
3683 * against the inode was last made. So here we assume the inode might
3684 * have been evicted, and therefore the exact value of last_unlink_trans
3685 * lost, and set it to last_trans to avoid metadata inconsistencies
3686 * between the inode and its parent if the inode is fsync'ed and the log
3687 * replayed. For example, in the scenario:
3690 * ln mydir/foo mydir/bar
3693 * echo 2 > /proc/sys/vm/drop_caches # evicts inode
3694 * xfs_io -c fsync mydir/foo
3696 * mount fs, triggers fsync log replay
3698 * We must make sure that when we fsync our inode foo we also log its
3699 * parent inode, otherwise after log replay the parent still has the
3700 * dentry with the "bar" name but our inode foo has a link count of 1
3701 * and doesn't have an inode ref with the name "bar" anymore.
3703 * Setting last_unlink_trans to last_trans is a pessimistic approach,
3704 * but it guarantees correctness at the expense of occasional full
3705 * transaction commits on fsync if our inode is a directory, or if our
3706 * inode is not a directory, logging its parent unnecessarily.
3708 BTRFS_I(inode)->last_unlink_trans = BTRFS_I(inode)->last_trans;
3711 if (inode->i_nlink != 1 ||
3712 path->slots[0] >= btrfs_header_nritems(leaf))
3715 btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3716 if (location.objectid != btrfs_ino(BTRFS_I(inode)))
3719 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3720 if (location.type == BTRFS_INODE_REF_KEY) {
3721 struct btrfs_inode_ref *ref;
3723 ref = (struct btrfs_inode_ref *)ptr;
3724 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3725 } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3726 struct btrfs_inode_extref *extref;
3728 extref = (struct btrfs_inode_extref *)ptr;
3729 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3734 * try to precache a NULL acl entry for files that don't have
3735 * any xattrs or acls
3737 maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3738 btrfs_ino(BTRFS_I(inode)), &first_xattr_slot);
3739 if (first_xattr_slot != -1) {
3740 path->slots[0] = first_xattr_slot;
3741 ret = btrfs_load_inode_props(inode, path);
3744 "error loading props for ino %llu (root %llu): %d",
3745 btrfs_ino(BTRFS_I(inode)),
3746 root->root_key.objectid, ret);
3748 btrfs_free_path(path);
3751 cache_no_acl(inode);
3753 switch (inode->i_mode & S_IFMT) {
3755 inode->i_mapping->a_ops = &btrfs_aops;
3756 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3757 inode->i_fop = &btrfs_file_operations;
3758 inode->i_op = &btrfs_file_inode_operations;
3761 inode->i_fop = &btrfs_dir_file_operations;
3762 inode->i_op = &btrfs_dir_inode_operations;
3765 inode->i_op = &btrfs_symlink_inode_operations;
3766 inode_nohighmem(inode);
3767 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3770 inode->i_op = &btrfs_special_inode_operations;
3771 init_special_inode(inode, inode->i_mode, rdev);
3775 btrfs_sync_inode_flags_to_i_flags(inode);
3779 btrfs_free_path(path);
3780 make_bad_inode(inode);
3785 * given a leaf and an inode, copy the inode fields into the leaf
3787 static void fill_inode_item(struct btrfs_trans_handle *trans,
3788 struct extent_buffer *leaf,
3789 struct btrfs_inode_item *item,
3790 struct inode *inode)
3792 struct btrfs_map_token token;
3794 btrfs_init_map_token(&token);
3796 btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3797 btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3798 btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3800 btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3801 btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3803 btrfs_set_token_timespec_sec(leaf, &item->atime,
3804 inode->i_atime.tv_sec, &token);
3805 btrfs_set_token_timespec_nsec(leaf, &item->atime,
3806 inode->i_atime.tv_nsec, &token);
3808 btrfs_set_token_timespec_sec(leaf, &item->mtime,
3809 inode->i_mtime.tv_sec, &token);
3810 btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3811 inode->i_mtime.tv_nsec, &token);
3813 btrfs_set_token_timespec_sec(leaf, &item->ctime,
3814 inode->i_ctime.tv_sec, &token);
3815 btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3816 inode->i_ctime.tv_nsec, &token);
3818 btrfs_set_token_timespec_sec(leaf, &item->otime,
3819 BTRFS_I(inode)->i_otime.tv_sec, &token);
3820 btrfs_set_token_timespec_nsec(leaf, &item->otime,
3821 BTRFS_I(inode)->i_otime.tv_nsec, &token);
3823 btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3825 btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3827 btrfs_set_token_inode_sequence(leaf, item, inode_peek_iversion(inode),
3829 btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3830 btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3831 btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3832 btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3836 * copy everything in the in-memory inode into the btree.
3838 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3839 struct btrfs_root *root, struct inode *inode)
3841 struct btrfs_inode_item *inode_item;
3842 struct btrfs_path *path;
3843 struct extent_buffer *leaf;
3846 path = btrfs_alloc_path();
3850 path->leave_spinning = 1;
3851 ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3859 leaf = path->nodes[0];
3860 inode_item = btrfs_item_ptr(leaf, path->slots[0],
3861 struct btrfs_inode_item);
3863 fill_inode_item(trans, leaf, inode_item, inode);
3864 btrfs_mark_buffer_dirty(leaf);
3865 btrfs_set_inode_last_trans(trans, inode);
3868 btrfs_free_path(path);
3873 * copy everything in the in-memory inode into the btree.
3875 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3876 struct btrfs_root *root, struct inode *inode)
3878 struct btrfs_fs_info *fs_info = root->fs_info;
3882 * If the inode is a free space inode, we can deadlock during commit
3883 * if we put it into the delayed code.
3885 * The data relocation inode should also be directly updated
3888 if (!btrfs_is_free_space_inode(BTRFS_I(inode))
3889 && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3890 && !test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) {
3891 btrfs_update_root_times(trans, root);
3893 ret = btrfs_delayed_update_inode(trans, root, inode);
3895 btrfs_set_inode_last_trans(trans, inode);
3899 return btrfs_update_inode_item(trans, root, inode);
3902 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3903 struct btrfs_root *root,
3904 struct inode *inode)
3908 ret = btrfs_update_inode(trans, root, inode);
3910 return btrfs_update_inode_item(trans, root, inode);
3915 * unlink helper that gets used here in inode.c and in the tree logging
3916 * recovery code. It remove a link in a directory with a given name, and
3917 * also drops the back refs in the inode to the directory
3919 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3920 struct btrfs_root *root,
3921 struct btrfs_inode *dir,
3922 struct btrfs_inode *inode,
3923 const char *name, int name_len)
3925 struct btrfs_fs_info *fs_info = root->fs_info;
3926 struct btrfs_path *path;
3928 struct extent_buffer *leaf;
3929 struct btrfs_dir_item *di;
3930 struct btrfs_key key;
3932 u64 ino = btrfs_ino(inode);
3933 u64 dir_ino = btrfs_ino(dir);
3935 path = btrfs_alloc_path();
3941 path->leave_spinning = 1;
3942 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3943 name, name_len, -1);
3952 leaf = path->nodes[0];
3953 btrfs_dir_item_key_to_cpu(leaf, di, &key);
3954 ret = btrfs_delete_one_dir_name(trans, root, path, di);
3957 btrfs_release_path(path);
3960 * If we don't have dir index, we have to get it by looking up
3961 * the inode ref, since we get the inode ref, remove it directly,
3962 * it is unnecessary to do delayed deletion.
3964 * But if we have dir index, needn't search inode ref to get it.
3965 * Since the inode ref is close to the inode item, it is better
3966 * that we delay to delete it, and just do this deletion when
3967 * we update the inode item.
3969 if (inode->dir_index) {
3970 ret = btrfs_delayed_delete_inode_ref(inode);
3972 index = inode->dir_index;
3977 ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3981 "failed to delete reference to %.*s, inode %llu parent %llu",
3982 name_len, name, ino, dir_ino);
3983 btrfs_abort_transaction(trans, ret);
3987 ret = btrfs_delete_delayed_dir_index(trans, fs_info, dir, index);
3989 btrfs_abort_transaction(trans, ret);
3993 ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len, inode,
3995 if (ret != 0 && ret != -ENOENT) {
3996 btrfs_abort_transaction(trans, ret);
4000 ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len, dir,
4005 btrfs_abort_transaction(trans, ret);
4007 btrfs_free_path(path);
4011 btrfs_i_size_write(dir, dir->vfs_inode.i_size - name_len * 2);
4012 inode_inc_iversion(&inode->vfs_inode);
4013 inode_inc_iversion(&dir->vfs_inode);
4014 inode->vfs_inode.i_ctime = dir->vfs_inode.i_mtime =
4015 dir->vfs_inode.i_ctime = current_time(&inode->vfs_inode);
4016 ret = btrfs_update_inode(trans, root, &dir->vfs_inode);
4021 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
4022 struct btrfs_root *root,
4023 struct btrfs_inode *dir, struct btrfs_inode *inode,
4024 const char *name, int name_len)
4027 ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
4029 drop_nlink(&inode->vfs_inode);
4030 ret = btrfs_update_inode(trans, root, &inode->vfs_inode);
4036 * helper to start transaction for unlink and rmdir.
4038 * unlink and rmdir are special in btrfs, they do not always free space, so
4039 * if we cannot make our reservations the normal way try and see if there is
4040 * plenty of slack room in the global reserve to migrate, otherwise we cannot
4041 * allow the unlink to occur.
4043 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
4045 struct btrfs_root *root = BTRFS_I(dir)->root;
4048 * 1 for the possible orphan item
4049 * 1 for the dir item
4050 * 1 for the dir index
4051 * 1 for the inode ref
4054 return btrfs_start_transaction_fallback_global_rsv(root, 5, 5);
4057 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
4059 struct btrfs_root *root = BTRFS_I(dir)->root;
4060 struct btrfs_trans_handle *trans;
4061 struct inode *inode = d_inode(dentry);
4064 trans = __unlink_start_trans(dir);
4066 return PTR_ERR(trans);
4068 btrfs_record_unlink_dir(trans, BTRFS_I(dir), BTRFS_I(d_inode(dentry)),
4071 ret = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4072 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4073 dentry->d_name.len);
4077 if (inode->i_nlink == 0) {
4078 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
4084 btrfs_end_transaction(trans);
4085 btrfs_btree_balance_dirty(root->fs_info);
4089 static int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4090 struct btrfs_root *root,
4091 struct inode *dir, u64 objectid,
4092 const char *name, int name_len)
4094 struct btrfs_fs_info *fs_info = root->fs_info;
4095 struct btrfs_path *path;
4096 struct extent_buffer *leaf;
4097 struct btrfs_dir_item *di;
4098 struct btrfs_key key;
4101 u64 dir_ino = btrfs_ino(BTRFS_I(dir));
4103 path = btrfs_alloc_path();
4107 di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4108 name, name_len, -1);
4109 if (IS_ERR_OR_NULL(di)) {
4117 leaf = path->nodes[0];
4118 btrfs_dir_item_key_to_cpu(leaf, di, &key);
4119 WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4120 ret = btrfs_delete_one_dir_name(trans, root, path, di);
4122 btrfs_abort_transaction(trans, ret);
4125 btrfs_release_path(path);
4127 ret = btrfs_del_root_ref(trans, fs_info, objectid,
4128 root->root_key.objectid, dir_ino,
4129 &index, name, name_len);
4131 if (ret != -ENOENT) {
4132 btrfs_abort_transaction(trans, ret);
4135 di = btrfs_search_dir_index_item(root, path, dir_ino,
4137 if (IS_ERR_OR_NULL(di)) {
4142 btrfs_abort_transaction(trans, ret);
4146 leaf = path->nodes[0];
4147 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4148 btrfs_release_path(path);
4151 btrfs_release_path(path);
4153 ret = btrfs_delete_delayed_dir_index(trans, fs_info, BTRFS_I(dir), index);
4155 btrfs_abort_transaction(trans, ret);
4159 btrfs_i_size_write(BTRFS_I(dir), dir->i_size - name_len * 2);
4160 inode_inc_iversion(dir);
4161 dir->i_mtime = dir->i_ctime = current_time(dir);
4162 ret = btrfs_update_inode_fallback(trans, root, dir);
4164 btrfs_abort_transaction(trans, ret);
4166 btrfs_free_path(path);
4171 * Helper to check if the subvolume references other subvolumes or if it's
4174 static noinline int may_destroy_subvol(struct btrfs_root *root)
4176 struct btrfs_fs_info *fs_info = root->fs_info;
4177 struct btrfs_path *path;
4178 struct btrfs_dir_item *di;
4179 struct btrfs_key key;
4183 path = btrfs_alloc_path();
4187 /* Make sure this root isn't set as the default subvol */
4188 dir_id = btrfs_super_root_dir(fs_info->super_copy);
4189 di = btrfs_lookup_dir_item(NULL, fs_info->tree_root, path,
4190 dir_id, "default", 7, 0);
4191 if (di && !IS_ERR(di)) {
4192 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &key);
4193 if (key.objectid == root->root_key.objectid) {
4196 "deleting default subvolume %llu is not allowed",
4200 btrfs_release_path(path);
4203 key.objectid = root->root_key.objectid;
4204 key.type = BTRFS_ROOT_REF_KEY;
4205 key.offset = (u64)-1;
4207 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
4213 if (path->slots[0] > 0) {
4215 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
4216 if (key.objectid == root->root_key.objectid &&
4217 key.type == BTRFS_ROOT_REF_KEY)
4221 btrfs_free_path(path);
4225 /* Delete all dentries for inodes belonging to the root */
4226 static void btrfs_prune_dentries(struct btrfs_root *root)
4228 struct btrfs_fs_info *fs_info = root->fs_info;
4229 struct rb_node *node;
4230 struct rb_node *prev;
4231 struct btrfs_inode *entry;
4232 struct inode *inode;
4235 if (!test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
4236 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4238 spin_lock(&root->inode_lock);
4240 node = root->inode_tree.rb_node;
4244 entry = rb_entry(node, struct btrfs_inode, rb_node);
4246 if (objectid < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
4247 node = node->rb_left;
4248 else if (objectid > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
4249 node = node->rb_right;
4255 entry = rb_entry(prev, struct btrfs_inode, rb_node);
4256 if (objectid <= btrfs_ino(BTRFS_I(&entry->vfs_inode))) {
4260 prev = rb_next(prev);
4264 entry = rb_entry(node, struct btrfs_inode, rb_node);
4265 objectid = btrfs_ino(BTRFS_I(&entry->vfs_inode)) + 1;
4266 inode = igrab(&entry->vfs_inode);
4268 spin_unlock(&root->inode_lock);
4269 if (atomic_read(&inode->i_count) > 1)
4270 d_prune_aliases(inode);
4272 * btrfs_drop_inode will have it removed from the inode
4273 * cache when its usage count hits zero.
4277 spin_lock(&root->inode_lock);
4281 if (cond_resched_lock(&root->inode_lock))
4284 node = rb_next(node);
4286 spin_unlock(&root->inode_lock);
4289 int btrfs_delete_subvolume(struct inode *dir, struct dentry *dentry)
4291 struct btrfs_fs_info *fs_info = btrfs_sb(dentry->d_sb);
4292 struct btrfs_root *root = BTRFS_I(dir)->root;
4293 struct inode *inode = d_inode(dentry);
4294 struct btrfs_root *dest = BTRFS_I(inode)->root;
4295 struct btrfs_trans_handle *trans;
4296 struct btrfs_block_rsv block_rsv;
4302 * Don't allow to delete a subvolume with send in progress. This is
4303 * inside the inode lock so the error handling that has to drop the bit
4304 * again is not run concurrently.
4306 spin_lock(&dest->root_item_lock);
4307 root_flags = btrfs_root_flags(&dest->root_item);
4308 if (dest->send_in_progress == 0) {
4309 btrfs_set_root_flags(&dest->root_item,
4310 root_flags | BTRFS_ROOT_SUBVOL_DEAD);
4311 spin_unlock(&dest->root_item_lock);
4313 spin_unlock(&dest->root_item_lock);
4315 "attempt to delete subvolume %llu during send",
4316 dest->root_key.objectid);
4320 down_write(&fs_info->subvol_sem);
4322 err = may_destroy_subvol(dest);
4326 btrfs_init_block_rsv(&block_rsv, BTRFS_BLOCK_RSV_TEMP);
4328 * One for dir inode,
4329 * two for dir entries,
4330 * two for root ref/backref.
4332 err = btrfs_subvolume_reserve_metadata(root, &block_rsv, 5, true);
4336 trans = btrfs_start_transaction(root, 0);
4337 if (IS_ERR(trans)) {
4338 err = PTR_ERR(trans);
4341 trans->block_rsv = &block_rsv;
4342 trans->bytes_reserved = block_rsv.size;
4344 btrfs_record_snapshot_destroy(trans, BTRFS_I(dir));
4346 ret = btrfs_unlink_subvol(trans, root, dir,
4347 dest->root_key.objectid,
4348 dentry->d_name.name,
4349 dentry->d_name.len);
4352 btrfs_abort_transaction(trans, ret);
4356 btrfs_record_root_in_trans(trans, dest);
4358 memset(&dest->root_item.drop_progress, 0,
4359 sizeof(dest->root_item.drop_progress));
4360 dest->root_item.drop_level = 0;
4361 btrfs_set_root_refs(&dest->root_item, 0);
4363 if (!test_and_set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &dest->state)) {
4364 ret = btrfs_insert_orphan_item(trans,
4366 dest->root_key.objectid);
4368 btrfs_abort_transaction(trans, ret);
4374 ret = btrfs_uuid_tree_remove(trans, dest->root_item.uuid,
4375 BTRFS_UUID_KEY_SUBVOL,
4376 dest->root_key.objectid);
4377 if (ret && ret != -ENOENT) {
4378 btrfs_abort_transaction(trans, ret);
4382 if (!btrfs_is_empty_uuid(dest->root_item.received_uuid)) {
4383 ret = btrfs_uuid_tree_remove(trans,
4384 dest->root_item.received_uuid,
4385 BTRFS_UUID_KEY_RECEIVED_SUBVOL,
4386 dest->root_key.objectid);
4387 if (ret && ret != -ENOENT) {
4388 btrfs_abort_transaction(trans, ret);
4395 trans->block_rsv = NULL;
4396 trans->bytes_reserved = 0;
4397 ret = btrfs_end_transaction(trans);
4400 inode->i_flags |= S_DEAD;
4402 btrfs_subvolume_release_metadata(fs_info, &block_rsv);
4404 up_write(&fs_info->subvol_sem);
4406 spin_lock(&dest->root_item_lock);
4407 root_flags = btrfs_root_flags(&dest->root_item);
4408 btrfs_set_root_flags(&dest->root_item,
4409 root_flags & ~BTRFS_ROOT_SUBVOL_DEAD);
4410 spin_unlock(&dest->root_item_lock);
4412 d_invalidate(dentry);
4413 btrfs_prune_dentries(dest);
4414 ASSERT(dest->send_in_progress == 0);
4417 if (dest->ino_cache_inode) {
4418 iput(dest->ino_cache_inode);
4419 dest->ino_cache_inode = NULL;
4426 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4428 struct inode *inode = d_inode(dentry);
4430 struct btrfs_root *root = BTRFS_I(dir)->root;
4431 struct btrfs_trans_handle *trans;
4432 u64 last_unlink_trans;
4434 if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4436 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_FIRST_FREE_OBJECTID)
4437 return btrfs_delete_subvolume(dir, dentry);
4439 trans = __unlink_start_trans(dir);
4441 return PTR_ERR(trans);
4443 if (unlikely(btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4444 err = btrfs_unlink_subvol(trans, root, dir,
4445 BTRFS_I(inode)->location.objectid,
4446 dentry->d_name.name,
4447 dentry->d_name.len);
4451 err = btrfs_orphan_add(trans, BTRFS_I(inode));
4455 last_unlink_trans = BTRFS_I(inode)->last_unlink_trans;
4457 /* now the directory is empty */
4458 err = btrfs_unlink_inode(trans, root, BTRFS_I(dir),
4459 BTRFS_I(d_inode(dentry)), dentry->d_name.name,
4460 dentry->d_name.len);
4462 btrfs_i_size_write(BTRFS_I(inode), 0);
4464 * Propagate the last_unlink_trans value of the deleted dir to
4465 * its parent directory. This is to prevent an unrecoverable
4466 * log tree in the case we do something like this:
4468 * 2) create snapshot under dir foo
4469 * 3) delete the snapshot
4472 * 6) fsync foo or some file inside foo
4474 if (last_unlink_trans >= trans->transid)
4475 BTRFS_I(dir)->last_unlink_trans = last_unlink_trans;
4478 btrfs_end_transaction(trans);
4479 btrfs_btree_balance_dirty(root->fs_info);
4484 static int truncate_space_check(struct btrfs_trans_handle *trans,
4485 struct btrfs_root *root,
4488 struct btrfs_fs_info *fs_info = root->fs_info;
4492 * This is only used to apply pressure to the enospc system, we don't
4493 * intend to use this reservation at all.
4495 bytes_deleted = btrfs_csum_bytes_to_leaves(fs_info, bytes_deleted);
4496 bytes_deleted *= fs_info->nodesize;
4497 ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
4498 bytes_deleted, BTRFS_RESERVE_NO_FLUSH);
4500 trace_btrfs_space_reservation(fs_info, "transaction",
4503 trans->bytes_reserved += bytes_deleted;
4510 * Return this if we need to call truncate_block for the last bit of the
4513 #define NEED_TRUNCATE_BLOCK 1
4516 * this can truncate away extent items, csum items and directory items.
4517 * It starts at a high offset and removes keys until it can't find
4518 * any higher than new_size
4520 * csum items that cross the new i_size are truncated to the new size
4523 * min_type is the minimum key type to truncate down to. If set to 0, this
4524 * will kill all the items on this inode, including the INODE_ITEM_KEY.
4526 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4527 struct btrfs_root *root,
4528 struct inode *inode,
4529 u64 new_size, u32 min_type)
4531 struct btrfs_fs_info *fs_info = root->fs_info;
4532 struct btrfs_path *path;
4533 struct extent_buffer *leaf;
4534 struct btrfs_file_extent_item *fi;
4535 struct btrfs_key key;
4536 struct btrfs_key found_key;
4537 u64 extent_start = 0;
4538 u64 extent_num_bytes = 0;
4539 u64 extent_offset = 0;
4541 u64 last_size = new_size;
4542 u32 found_type = (u8)-1;
4545 int pending_del_nr = 0;
4546 int pending_del_slot = 0;
4547 int extent_type = -1;
4549 u64 ino = btrfs_ino(BTRFS_I(inode));
4550 u64 bytes_deleted = 0;
4551 bool be_nice = false;
4552 bool should_throttle = false;
4553 bool should_end = false;
4555 BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4558 * for non-free space inodes and ref cows, we want to back off from
4561 if (!btrfs_is_free_space_inode(BTRFS_I(inode)) &&
4562 test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4565 path = btrfs_alloc_path();
4568 path->reada = READA_BACK;
4571 * We want to drop from the next block forward in case this new size is
4572 * not block aligned since we will be keeping the last block of the
4573 * extent just the way it is.
4575 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4576 root == fs_info->tree_root)
4577 btrfs_drop_extent_cache(BTRFS_I(inode), ALIGN(new_size,
4578 fs_info->sectorsize),
4582 * This function is also used to drop the items in the log tree before
4583 * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4584 * it is used to drop the loged items. So we shouldn't kill the delayed
4587 if (min_type == 0 && root == BTRFS_I(inode)->root)
4588 btrfs_kill_delayed_inode_items(BTRFS_I(inode));
4591 key.offset = (u64)-1;
4596 * with a 16K leaf size and 128MB extents, you can actually queue
4597 * up a huge file in a single leaf. Most of the time that
4598 * bytes_deleted is > 0, it will be huge by the time we get here
4600 if (be_nice && bytes_deleted > SZ_32M &&
4601 btrfs_should_end_transaction(trans)) {
4606 path->leave_spinning = 1;
4607 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4613 /* there are no items in the tree for us to truncate, we're
4616 if (path->slots[0] == 0)
4623 leaf = path->nodes[0];
4624 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4625 found_type = found_key.type;
4627 if (found_key.objectid != ino)
4630 if (found_type < min_type)
4633 item_end = found_key.offset;
4634 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4635 fi = btrfs_item_ptr(leaf, path->slots[0],
4636 struct btrfs_file_extent_item);
4637 extent_type = btrfs_file_extent_type(leaf, fi);
4638 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4640 btrfs_file_extent_num_bytes(leaf, fi);
4642 trace_btrfs_truncate_show_fi_regular(
4643 BTRFS_I(inode), leaf, fi,
4645 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4646 item_end += btrfs_file_extent_inline_len(leaf,
4647 path->slots[0], fi);
4649 trace_btrfs_truncate_show_fi_inline(
4650 BTRFS_I(inode), leaf, fi, path->slots[0],
4655 if (found_type > min_type) {
4658 if (item_end < new_size)
4660 if (found_key.offset >= new_size)
4666 /* FIXME, shrink the extent if the ref count is only 1 */
4667 if (found_type != BTRFS_EXTENT_DATA_KEY)
4670 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4672 extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4674 u64 orig_num_bytes =
4675 btrfs_file_extent_num_bytes(leaf, fi);
4676 extent_num_bytes = ALIGN(new_size -
4678 fs_info->sectorsize);
4679 btrfs_set_file_extent_num_bytes(leaf, fi,
4681 num_dec = (orig_num_bytes -
4683 if (test_bit(BTRFS_ROOT_REF_COWS,
4686 inode_sub_bytes(inode, num_dec);
4687 btrfs_mark_buffer_dirty(leaf);
4690 btrfs_file_extent_disk_num_bytes(leaf,
4692 extent_offset = found_key.offset -
4693 btrfs_file_extent_offset(leaf, fi);
4695 /* FIXME blocksize != 4096 */
4696 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4697 if (extent_start != 0) {
4699 if (test_bit(BTRFS_ROOT_REF_COWS,
4701 inode_sub_bytes(inode, num_dec);
4704 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4706 * we can't truncate inline items that have had
4710 btrfs_file_extent_encryption(leaf, fi) == 0 &&
4711 btrfs_file_extent_other_encoding(leaf, fi) == 0 &&
4712 btrfs_file_extent_compression(leaf, fi) == 0) {
4713 u32 size = (u32)(new_size - found_key.offset);
4715 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4716 size = btrfs_file_extent_calc_inline_size(size);
4717 btrfs_truncate_item(root->fs_info, path, size, 1);
4718 } else if (!del_item) {
4720 * We have to bail so the last_size is set to
4721 * just before this extent.
4723 ret = NEED_TRUNCATE_BLOCK;
4727 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4728 inode_sub_bytes(inode, item_end + 1 - new_size);
4732 last_size = found_key.offset;
4734 last_size = new_size;
4736 if (!pending_del_nr) {
4737 /* no pending yet, add ourselves */
4738 pending_del_slot = path->slots[0];
4740 } else if (pending_del_nr &&
4741 path->slots[0] + 1 == pending_del_slot) {
4742 /* hop on the pending chunk */
4744 pending_del_slot = path->slots[0];
4751 should_throttle = false;
4754 (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4755 root == fs_info->tree_root)) {
4756 btrfs_set_path_blocking(path);
4757 bytes_deleted += extent_num_bytes;
4758 ret = btrfs_free_extent(trans, root, extent_start,
4759 extent_num_bytes, 0,
4760 btrfs_header_owner(leaf),
4761 ino, extent_offset);
4763 btrfs_abort_transaction(trans, ret);
4766 if (btrfs_should_throttle_delayed_refs(trans, fs_info))
4767 btrfs_async_run_delayed_refs(fs_info,
4768 trans->delayed_ref_updates * 2,
4771 if (truncate_space_check(trans, root,
4772 extent_num_bytes)) {
4775 if (btrfs_should_throttle_delayed_refs(trans,
4777 should_throttle = true;
4781 if (found_type == BTRFS_INODE_ITEM_KEY)
4784 if (path->slots[0] == 0 ||
4785 path->slots[0] != pending_del_slot ||
4786 should_throttle || should_end) {
4787 if (pending_del_nr) {
4788 ret = btrfs_del_items(trans, root, path,
4792 btrfs_abort_transaction(trans, ret);
4797 btrfs_release_path(path);
4798 if (should_throttle) {
4799 unsigned long updates = trans->delayed_ref_updates;
4801 trans->delayed_ref_updates = 0;
4802 ret = btrfs_run_delayed_refs(trans,
4809 * if we failed to refill our space rsv, bail out
4810 * and let the transaction restart
4822 if (ret >= 0 && pending_del_nr) {
4825 err = btrfs_del_items(trans, root, path, pending_del_slot,
4828 btrfs_abort_transaction(trans, err);
4832 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
4833 ASSERT(last_size >= new_size);
4834 if (!ret && last_size > new_size)
4835 last_size = new_size;
4836 btrfs_ordered_update_i_size(inode, last_size, NULL);
4839 btrfs_free_path(path);
4841 if (be_nice && bytes_deleted > SZ_32M && (ret >= 0 || ret == -EAGAIN)) {
4842 unsigned long updates = trans->delayed_ref_updates;
4846 trans->delayed_ref_updates = 0;
4847 err = btrfs_run_delayed_refs(trans, updates * 2);
4856 * btrfs_truncate_block - read, zero a chunk and write a block
4857 * @inode - inode that we're zeroing
4858 * @from - the offset to start zeroing
4859 * @len - the length to zero, 0 to zero the entire range respective to the
4861 * @front - zero up to the offset instead of from the offset on
4863 * This will find the block for the "from" offset and cow the block and zero the
4864 * part we want to zero. This is used with truncate and hole punching.
4866 int btrfs_truncate_block(struct inode *inode, loff_t from, loff_t len,
4869 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4870 struct address_space *mapping = inode->i_mapping;
4871 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4872 struct btrfs_ordered_extent *ordered;
4873 struct extent_state *cached_state = NULL;
4874 struct extent_changeset *data_reserved = NULL;
4876 u32 blocksize = fs_info->sectorsize;
4877 pgoff_t index = from >> PAGE_SHIFT;
4878 unsigned offset = from & (blocksize - 1);
4880 gfp_t mask = btrfs_alloc_write_mask(mapping);
4885 if (IS_ALIGNED(offset, blocksize) &&
4886 (!len || IS_ALIGNED(len, blocksize)))
4889 block_start = round_down(from, blocksize);
4890 block_end = block_start + blocksize - 1;
4892 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
4893 block_start, blocksize);
4898 page = find_or_create_page(mapping, index, mask);
4900 btrfs_delalloc_release_space(inode, data_reserved,
4901 block_start, blocksize, true);
4902 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, true);
4907 if (!PageUptodate(page)) {
4908 ret = btrfs_readpage(NULL, page);
4910 if (page->mapping != mapping) {
4915 if (!PageUptodate(page)) {
4920 wait_on_page_writeback(page);
4922 lock_extent_bits(io_tree, block_start, block_end, &cached_state);
4923 set_page_extent_mapped(page);
4925 ordered = btrfs_lookup_ordered_extent(inode, block_start);
4927 unlock_extent_cached(io_tree, block_start, block_end,
4931 btrfs_start_ordered_extent(inode, ordered, 1);
4932 btrfs_put_ordered_extent(ordered);
4936 clear_extent_bit(&BTRFS_I(inode)->io_tree, block_start, block_end,
4937 EXTENT_DIRTY | EXTENT_DELALLOC |
4938 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4939 0, 0, &cached_state);
4941 ret = btrfs_set_extent_delalloc(inode, block_start, block_end, 0,
4944 unlock_extent_cached(io_tree, block_start, block_end,
4949 if (offset != blocksize) {
4951 len = blocksize - offset;
4954 memset(kaddr + (block_start - page_offset(page)),
4957 memset(kaddr + (block_start - page_offset(page)) + offset,
4959 flush_dcache_page(page);
4962 ClearPageChecked(page);
4963 set_page_dirty(page);
4964 unlock_extent_cached(io_tree, block_start, block_end, &cached_state);
4968 btrfs_delalloc_release_space(inode, data_reserved, block_start,
4970 btrfs_delalloc_release_extents(BTRFS_I(inode), blocksize, (ret != 0));
4974 extent_changeset_free(data_reserved);
4978 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4979 u64 offset, u64 len)
4981 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4982 struct btrfs_trans_handle *trans;
4986 * Still need to make sure the inode looks like it's been updated so
4987 * that any holes get logged if we fsync.
4989 if (btrfs_fs_incompat(fs_info, NO_HOLES)) {
4990 BTRFS_I(inode)->last_trans = fs_info->generation;
4991 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4992 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4997 * 1 - for the one we're dropping
4998 * 1 - for the one we're adding
4999 * 1 - for updating the inode.
5001 trans = btrfs_start_transaction(root, 3);
5003 return PTR_ERR(trans);
5005 ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
5007 btrfs_abort_transaction(trans, ret);
5008 btrfs_end_transaction(trans);
5012 ret = btrfs_insert_file_extent(trans, root, btrfs_ino(BTRFS_I(inode)),
5013 offset, 0, 0, len, 0, len, 0, 0, 0);
5015 btrfs_abort_transaction(trans, ret);
5017 btrfs_update_inode(trans, root, inode);
5018 btrfs_end_transaction(trans);
5023 * This function puts in dummy file extents for the area we're creating a hole
5024 * for. So if we are truncating this file to a larger size we need to insert
5025 * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
5026 * the range between oldsize and size
5028 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
5030 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5031 struct btrfs_root *root = BTRFS_I(inode)->root;
5032 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5033 struct extent_map *em = NULL;
5034 struct extent_state *cached_state = NULL;
5035 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5036 u64 hole_start = ALIGN(oldsize, fs_info->sectorsize);
5037 u64 block_end = ALIGN(size, fs_info->sectorsize);
5044 * If our size started in the middle of a block we need to zero out the
5045 * rest of the block before we expand the i_size, otherwise we could
5046 * expose stale data.
5048 err = btrfs_truncate_block(inode, oldsize, 0, 0);
5052 if (size <= hole_start)
5056 struct btrfs_ordered_extent *ordered;
5058 lock_extent_bits(io_tree, hole_start, block_end - 1,
5060 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), hole_start,
5061 block_end - hole_start);
5064 unlock_extent_cached(io_tree, hole_start, block_end - 1,
5066 btrfs_start_ordered_extent(inode, ordered, 1);
5067 btrfs_put_ordered_extent(ordered);
5070 cur_offset = hole_start;
5072 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, cur_offset,
5073 block_end - cur_offset, 0);
5079 last_byte = min(extent_map_end(em), block_end);
5080 last_byte = ALIGN(last_byte, fs_info->sectorsize);
5081 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
5082 struct extent_map *hole_em;
5083 hole_size = last_byte - cur_offset;
5085 err = maybe_insert_hole(root, inode, cur_offset,
5089 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
5090 cur_offset + hole_size - 1, 0);
5091 hole_em = alloc_extent_map();
5093 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
5094 &BTRFS_I(inode)->runtime_flags);
5097 hole_em->start = cur_offset;
5098 hole_em->len = hole_size;
5099 hole_em->orig_start = cur_offset;
5101 hole_em->block_start = EXTENT_MAP_HOLE;
5102 hole_em->block_len = 0;
5103 hole_em->orig_block_len = 0;
5104 hole_em->ram_bytes = hole_size;
5105 hole_em->bdev = fs_info->fs_devices->latest_bdev;
5106 hole_em->compress_type = BTRFS_COMPRESS_NONE;
5107 hole_em->generation = fs_info->generation;
5110 write_lock(&em_tree->lock);
5111 err = add_extent_mapping(em_tree, hole_em, 1);
5112 write_unlock(&em_tree->lock);
5115 btrfs_drop_extent_cache(BTRFS_I(inode),
5120 free_extent_map(hole_em);
5123 free_extent_map(em);
5125 cur_offset = last_byte;
5126 if (cur_offset >= block_end)
5129 free_extent_map(em);
5130 unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state);
5134 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
5136 struct btrfs_root *root = BTRFS_I(inode)->root;
5137 struct btrfs_trans_handle *trans;
5138 loff_t oldsize = i_size_read(inode);
5139 loff_t newsize = attr->ia_size;
5140 int mask = attr->ia_valid;
5144 * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
5145 * special case where we need to update the times despite not having
5146 * these flags set. For all other operations the VFS set these flags
5147 * explicitly if it wants a timestamp update.
5149 if (newsize != oldsize) {
5150 inode_inc_iversion(inode);
5151 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
5152 inode->i_ctime = inode->i_mtime =
5153 current_time(inode);
5156 if (newsize > oldsize) {
5158 * Don't do an expanding truncate while snapshotting is ongoing.
5159 * This is to ensure the snapshot captures a fully consistent
5160 * state of this file - if the snapshot captures this expanding
5161 * truncation, it must capture all writes that happened before
5164 btrfs_wait_for_snapshot_creation(root);
5165 ret = btrfs_cont_expand(inode, oldsize, newsize);
5167 btrfs_end_write_no_snapshotting(root);
5171 trans = btrfs_start_transaction(root, 1);
5172 if (IS_ERR(trans)) {
5173 btrfs_end_write_no_snapshotting(root);
5174 return PTR_ERR(trans);
5177 i_size_write(inode, newsize);
5178 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
5179 pagecache_isize_extended(inode, oldsize, newsize);
5180 ret = btrfs_update_inode(trans, root, inode);
5181 btrfs_end_write_no_snapshotting(root);
5182 btrfs_end_transaction(trans);
5186 * We're truncating a file that used to have good data down to
5187 * zero. Make sure it gets into the ordered flush list so that
5188 * any new writes get down to disk quickly.
5191 set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
5192 &BTRFS_I(inode)->runtime_flags);
5194 truncate_setsize(inode, newsize);
5196 /* Disable nonlocked read DIO to avoid the end less truncate */
5197 btrfs_inode_block_unlocked_dio(BTRFS_I(inode));
5198 inode_dio_wait(inode);
5199 btrfs_inode_resume_unlocked_dio(BTRFS_I(inode));
5201 ret = btrfs_truncate(inode, newsize == oldsize);
5202 if (ret && inode->i_nlink) {
5206 * Truncate failed, so fix up the in-memory size. We
5207 * adjusted disk_i_size down as we removed extents, so
5208 * wait for disk_i_size to be stable and then update the
5209 * in-memory size to match.
5211 err = btrfs_wait_ordered_range(inode, 0, (u64)-1);
5214 i_size_write(inode, BTRFS_I(inode)->disk_i_size);
5221 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
5223 struct inode *inode = d_inode(dentry);
5224 struct btrfs_root *root = BTRFS_I(inode)->root;
5227 if (btrfs_root_readonly(root))
5230 err = setattr_prepare(dentry, attr);
5234 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
5235 err = btrfs_setsize(inode, attr);
5240 if (attr->ia_valid) {
5241 setattr_copy(inode, attr);
5242 inode_inc_iversion(inode);
5243 err = btrfs_dirty_inode(inode);
5245 if (!err && attr->ia_valid & ATTR_MODE)
5246 err = posix_acl_chmod(inode, inode->i_mode);
5253 * While truncating the inode pages during eviction, we get the VFS calling
5254 * btrfs_invalidatepage() against each page of the inode. This is slow because
5255 * the calls to btrfs_invalidatepage() result in a huge amount of calls to
5256 * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
5257 * extent_state structures over and over, wasting lots of time.
5259 * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
5260 * those expensive operations on a per page basis and do only the ordered io
5261 * finishing, while we release here the extent_map and extent_state structures,
5262 * without the excessive merging and splitting.
5264 static void evict_inode_truncate_pages(struct inode *inode)
5266 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5267 struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
5268 struct rb_node *node;
5270 ASSERT(inode->i_state & I_FREEING);
5271 truncate_inode_pages_final(&inode->i_data);
5273 write_lock(&map_tree->lock);
5274 while (!RB_EMPTY_ROOT(&map_tree->map)) {
5275 struct extent_map *em;
5277 node = rb_first(&map_tree->map);
5278 em = rb_entry(node, struct extent_map, rb_node);
5279 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
5280 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
5281 remove_extent_mapping(map_tree, em);
5282 free_extent_map(em);
5283 if (need_resched()) {
5284 write_unlock(&map_tree->lock);
5286 write_lock(&map_tree->lock);
5289 write_unlock(&map_tree->lock);
5292 * Keep looping until we have no more ranges in the io tree.
5293 * We can have ongoing bios started by readpages (called from readahead)
5294 * that have their endio callback (extent_io.c:end_bio_extent_readpage)
5295 * still in progress (unlocked the pages in the bio but did not yet
5296 * unlocked the ranges in the io tree). Therefore this means some
5297 * ranges can still be locked and eviction started because before
5298 * submitting those bios, which are executed by a separate task (work
5299 * queue kthread), inode references (inode->i_count) were not taken
5300 * (which would be dropped in the end io callback of each bio).
5301 * Therefore here we effectively end up waiting for those bios and
5302 * anyone else holding locked ranges without having bumped the inode's
5303 * reference count - if we don't do it, when they access the inode's
5304 * io_tree to unlock a range it may be too late, leading to an
5305 * use-after-free issue.
5307 spin_lock(&io_tree->lock);
5308 while (!RB_EMPTY_ROOT(&io_tree->state)) {
5309 struct extent_state *state;
5310 struct extent_state *cached_state = NULL;
5314 node = rb_first(&io_tree->state);
5315 state = rb_entry(node, struct extent_state, rb_node);
5316 start = state->start;
5318 spin_unlock(&io_tree->lock);
5320 lock_extent_bits(io_tree, start, end, &cached_state);
5323 * If still has DELALLOC flag, the extent didn't reach disk,
5324 * and its reserved space won't be freed by delayed_ref.
5325 * So we need to free its reserved space here.
5326 * (Refer to comment in btrfs_invalidatepage, case 2)
5328 * Note, end is the bytenr of last byte, so we need + 1 here.
5330 if (state->state & EXTENT_DELALLOC)
5331 btrfs_qgroup_free_data(inode, NULL, start, end - start + 1);
5333 clear_extent_bit(io_tree, start, end,
5334 EXTENT_LOCKED | EXTENT_DIRTY |
5335 EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
5336 EXTENT_DEFRAG, 1, 1, &cached_state);
5339 spin_lock(&io_tree->lock);
5341 spin_unlock(&io_tree->lock);
5344 static struct btrfs_trans_handle *evict_refill_and_join(struct btrfs_root *root,
5345 struct btrfs_block_rsv *rsv,
5348 struct btrfs_fs_info *fs_info = root->fs_info;
5349 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5353 struct btrfs_trans_handle *trans;
5356 ret = btrfs_block_rsv_refill(root, rsv, min_size,
5357 BTRFS_RESERVE_FLUSH_LIMIT);
5359 if (ret && ++failures > 2) {
5361 "could not allocate space for a delete; will truncate on mount");
5362 return ERR_PTR(-ENOSPC);
5365 trans = btrfs_join_transaction(root);
5366 if (IS_ERR(trans) || !ret)
5370 * Try to steal from the global reserve if there is space for
5373 if (!btrfs_check_space_for_delayed_refs(trans, fs_info) &&
5374 !btrfs_block_rsv_migrate(global_rsv, rsv, min_size, 0))
5377 /* If not, commit and try again. */
5378 ret = btrfs_commit_transaction(trans);
5380 return ERR_PTR(ret);
5384 void btrfs_evict_inode(struct inode *inode)
5386 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5387 struct btrfs_trans_handle *trans;
5388 struct btrfs_root *root = BTRFS_I(inode)->root;
5389 struct btrfs_block_rsv *rsv;
5393 trace_btrfs_inode_evict(inode);
5400 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
5402 evict_inode_truncate_pages(inode);
5404 if (inode->i_nlink &&
5405 ((btrfs_root_refs(&root->root_item) != 0 &&
5406 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
5407 btrfs_is_free_space_inode(BTRFS_I(inode))))
5410 if (is_bad_inode(inode))
5412 /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
5413 if (!special_file(inode->i_mode))
5414 btrfs_wait_ordered_range(inode, 0, (u64)-1);
5416 btrfs_free_io_failure_record(BTRFS_I(inode), 0, (u64)-1);
5418 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
5421 if (inode->i_nlink > 0) {
5422 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
5423 root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
5427 ret = btrfs_commit_inode_delayed_inode(BTRFS_I(inode));
5431 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
5434 rsv->size = min_size;
5437 btrfs_i_size_write(BTRFS_I(inode), 0);
5440 trans = evict_refill_and_join(root, rsv, min_size);
5444 trans->block_rsv = rsv;
5446 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
5447 trans->block_rsv = &fs_info->trans_block_rsv;
5448 btrfs_end_transaction(trans);
5449 btrfs_btree_balance_dirty(fs_info);
5450 if (ret && ret != -ENOSPC && ret != -EAGAIN)
5457 * Errors here aren't a big deal, it just means we leave orphan items in
5458 * the tree. They will be cleaned up on the next mount. If the inode
5459 * number gets reused, cleanup deletes the orphan item without doing
5460 * anything, and unlink reuses the existing orphan item.
5462 * If it turns out that we are dropping too many of these, we might want
5463 * to add a mechanism for retrying these after a commit.
5465 trans = evict_refill_and_join(root, rsv, min_size);
5466 if (!IS_ERR(trans)) {
5467 trans->block_rsv = rsv;
5468 btrfs_orphan_del(trans, BTRFS_I(inode));
5469 trans->block_rsv = &fs_info->trans_block_rsv;
5470 btrfs_end_transaction(trans);
5473 if (!(root == fs_info->tree_root ||
5474 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5475 btrfs_return_ino(root, btrfs_ino(BTRFS_I(inode)));
5478 btrfs_free_block_rsv(fs_info, rsv);
5481 * If we didn't successfully delete, the orphan item will still be in
5482 * the tree and we'll retry on the next mount. Again, we might also want
5483 * to retry these periodically in the future.
5485 btrfs_remove_delayed_node(BTRFS_I(inode));
5490 * this returns the key found in the dir entry in the location pointer.
5491 * If no dir entries were found, returns -ENOENT.
5492 * If found a corrupted location in dir entry, returns -EUCLEAN.
5494 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5495 struct btrfs_key *location)
5497 const char *name = dentry->d_name.name;
5498 int namelen = dentry->d_name.len;
5499 struct btrfs_dir_item *di;
5500 struct btrfs_path *path;
5501 struct btrfs_root *root = BTRFS_I(dir)->root;
5504 path = btrfs_alloc_path();
5508 di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(BTRFS_I(dir)),
5519 btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5520 if (location->type != BTRFS_INODE_ITEM_KEY &&
5521 location->type != BTRFS_ROOT_ITEM_KEY) {
5523 btrfs_warn(root->fs_info,
5524 "%s gets something invalid in DIR_ITEM (name %s, directory ino %llu, location(%llu %u %llu))",
5525 __func__, name, btrfs_ino(BTRFS_I(dir)),
5526 location->objectid, location->type, location->offset);
5529 btrfs_free_path(path);
5534 * when we hit a tree root in a directory, the btrfs part of the inode
5535 * needs to be changed to reflect the root directory of the tree root. This
5536 * is kind of like crossing a mount point.
5538 static int fixup_tree_root_location(struct btrfs_fs_info *fs_info,
5540 struct dentry *dentry,
5541 struct btrfs_key *location,
5542 struct btrfs_root **sub_root)
5544 struct btrfs_path *path;
5545 struct btrfs_root *new_root;
5546 struct btrfs_root_ref *ref;
5547 struct extent_buffer *leaf;
5548 struct btrfs_key key;
5552 path = btrfs_alloc_path();
5559 key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5560 key.type = BTRFS_ROOT_REF_KEY;
5561 key.offset = location->objectid;
5563 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
5570 leaf = path->nodes[0];
5571 ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5572 if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(BTRFS_I(dir)) ||
5573 btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5576 ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5577 (unsigned long)(ref + 1),
5578 dentry->d_name.len);
5582 btrfs_release_path(path);
5584 new_root = btrfs_read_fs_root_no_name(fs_info, location);
5585 if (IS_ERR(new_root)) {
5586 err = PTR_ERR(new_root);
5590 *sub_root = new_root;
5591 location->objectid = btrfs_root_dirid(&new_root->root_item);
5592 location->type = BTRFS_INODE_ITEM_KEY;
5593 location->offset = 0;
5596 btrfs_free_path(path);
5600 static void inode_tree_add(struct inode *inode)
5602 struct btrfs_root *root = BTRFS_I(inode)->root;
5603 struct btrfs_inode *entry;
5605 struct rb_node *parent;
5606 struct rb_node *new = &BTRFS_I(inode)->rb_node;
5607 u64 ino = btrfs_ino(BTRFS_I(inode));
5609 if (inode_unhashed(inode))
5612 spin_lock(&root->inode_lock);
5613 p = &root->inode_tree.rb_node;
5616 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5618 if (ino < btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5619 p = &parent->rb_left;
5620 else if (ino > btrfs_ino(BTRFS_I(&entry->vfs_inode)))
5621 p = &parent->rb_right;
5623 WARN_ON(!(entry->vfs_inode.i_state &
5624 (I_WILL_FREE | I_FREEING)));
5625 rb_replace_node(parent, new, &root->inode_tree);
5626 RB_CLEAR_NODE(parent);
5627 spin_unlock(&root->inode_lock);
5631 rb_link_node(new, parent, p);
5632 rb_insert_color(new, &root->inode_tree);
5633 spin_unlock(&root->inode_lock);
5636 static void inode_tree_del(struct inode *inode)
5638 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
5639 struct btrfs_root *root = BTRFS_I(inode)->root;
5642 spin_lock(&root->inode_lock);
5643 if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5644 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5645 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5646 empty = RB_EMPTY_ROOT(&root->inode_tree);
5648 spin_unlock(&root->inode_lock);
5650 if (empty && btrfs_root_refs(&root->root_item) == 0) {
5651 synchronize_srcu(&fs_info->subvol_srcu);
5652 spin_lock(&root->inode_lock);
5653 empty = RB_EMPTY_ROOT(&root->inode_tree);
5654 spin_unlock(&root->inode_lock);
5656 btrfs_add_dead_root(root);
5661 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5663 struct btrfs_iget_args *args = p;
5664 inode->i_ino = args->location->objectid;
5665 memcpy(&BTRFS_I(inode)->location, args->location,
5666 sizeof(*args->location));
5667 BTRFS_I(inode)->root = args->root;
5671 static int btrfs_find_actor(struct inode *inode, void *opaque)
5673 struct btrfs_iget_args *args = opaque;
5674 return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5675 args->root == BTRFS_I(inode)->root;
5678 static struct inode *btrfs_iget_locked(struct super_block *s,
5679 struct btrfs_key *location,
5680 struct btrfs_root *root)
5682 struct inode *inode;
5683 struct btrfs_iget_args args;
5684 unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5686 args.location = location;
5689 inode = iget5_locked(s, hashval, btrfs_find_actor,
5690 btrfs_init_locked_inode,
5695 /* Get an inode object given its location and corresponding root.
5696 * Returns in *is_new if the inode was read from disk
5698 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5699 struct btrfs_root *root, int *new)
5701 struct inode *inode;
5703 inode = btrfs_iget_locked(s, location, root);
5705 return ERR_PTR(-ENOMEM);
5707 if (inode->i_state & I_NEW) {
5710 ret = btrfs_read_locked_inode(inode);
5711 if (!is_bad_inode(inode)) {
5712 inode_tree_add(inode);
5713 unlock_new_inode(inode);
5717 unlock_new_inode(inode);
5720 inode = ERR_PTR(ret < 0 ? ret : -ESTALE);
5727 static struct inode *new_simple_dir(struct super_block *s,
5728 struct btrfs_key *key,
5729 struct btrfs_root *root)
5731 struct inode *inode = new_inode(s);
5734 return ERR_PTR(-ENOMEM);
5736 BTRFS_I(inode)->root = root;
5737 memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5738 set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5740 inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5741 inode->i_op = &btrfs_dir_ro_inode_operations;
5742 inode->i_opflags &= ~IOP_XATTR;
5743 inode->i_fop = &simple_dir_operations;
5744 inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5745 inode->i_mtime = current_time(inode);
5746 inode->i_atime = inode->i_mtime;
5747 inode->i_ctime = inode->i_mtime;
5748 BTRFS_I(inode)->i_otime = timespec64_to_timespec(inode->i_mtime);
5753 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5755 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
5756 struct inode *inode;
5757 struct btrfs_root *root = BTRFS_I(dir)->root;
5758 struct btrfs_root *sub_root = root;
5759 struct btrfs_key location;
5763 if (dentry->d_name.len > BTRFS_NAME_LEN)
5764 return ERR_PTR(-ENAMETOOLONG);
5766 ret = btrfs_inode_by_name(dir, dentry, &location);
5768 return ERR_PTR(ret);
5770 if (location.type == BTRFS_INODE_ITEM_KEY) {
5771 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5775 index = srcu_read_lock(&fs_info->subvol_srcu);
5776 ret = fixup_tree_root_location(fs_info, dir, dentry,
5777 &location, &sub_root);
5780 inode = ERR_PTR(ret);
5782 inode = new_simple_dir(dir->i_sb, &location, sub_root);
5784 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5786 srcu_read_unlock(&fs_info->subvol_srcu, index);
5788 if (!IS_ERR(inode) && root != sub_root) {
5789 down_read(&fs_info->cleanup_work_sem);
5790 if (!sb_rdonly(inode->i_sb))
5791 ret = btrfs_orphan_cleanup(sub_root);
5792 up_read(&fs_info->cleanup_work_sem);
5795 inode = ERR_PTR(ret);
5802 static int btrfs_dentry_delete(const struct dentry *dentry)
5804 struct btrfs_root *root;
5805 struct inode *inode = d_inode(dentry);
5807 if (!inode && !IS_ROOT(dentry))
5808 inode = d_inode(dentry->d_parent);
5811 root = BTRFS_I(inode)->root;
5812 if (btrfs_root_refs(&root->root_item) == 0)
5815 if (btrfs_ino(BTRFS_I(inode)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5821 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5824 struct inode *inode;
5826 inode = btrfs_lookup_dentry(dir, dentry);
5827 if (IS_ERR(inode)) {
5828 if (PTR_ERR(inode) == -ENOENT)
5831 return ERR_CAST(inode);
5834 return d_splice_alias(inode, dentry);
5837 unsigned char btrfs_filetype_table[] = {
5838 DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5842 * All this infrastructure exists because dir_emit can fault, and we are holding
5843 * the tree lock when doing readdir. For now just allocate a buffer and copy
5844 * our information into that, and then dir_emit from the buffer. This is
5845 * similar to what NFS does, only we don't keep the buffer around in pagecache
5846 * because I'm afraid I'll mess that up. Long term we need to make filldir do
5847 * copy_to_user_inatomic so we don't have to worry about page faulting under the
5850 static int btrfs_opendir(struct inode *inode, struct file *file)
5852 struct btrfs_file_private *private;
5854 private = kzalloc(sizeof(struct btrfs_file_private), GFP_KERNEL);
5857 private->filldir_buf = kzalloc(PAGE_SIZE, GFP_KERNEL);
5858 if (!private->filldir_buf) {
5862 file->private_data = private;
5873 static int btrfs_filldir(void *addr, int entries, struct dir_context *ctx)
5876 struct dir_entry *entry = addr;
5877 char *name = (char *)(entry + 1);
5879 ctx->pos = get_unaligned(&entry->offset);
5880 if (!dir_emit(ctx, name, get_unaligned(&entry->name_len),
5881 get_unaligned(&entry->ino),
5882 get_unaligned(&entry->type)))
5884 addr += sizeof(struct dir_entry) +
5885 get_unaligned(&entry->name_len);
5891 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5893 struct inode *inode = file_inode(file);
5894 struct btrfs_root *root = BTRFS_I(inode)->root;
5895 struct btrfs_file_private *private = file->private_data;
5896 struct btrfs_dir_item *di;
5897 struct btrfs_key key;
5898 struct btrfs_key found_key;
5899 struct btrfs_path *path;
5901 struct list_head ins_list;
5902 struct list_head del_list;
5904 struct extent_buffer *leaf;
5911 struct btrfs_key location;
5913 if (!dir_emit_dots(file, ctx))
5916 path = btrfs_alloc_path();
5920 addr = private->filldir_buf;
5921 path->reada = READA_FORWARD;
5923 INIT_LIST_HEAD(&ins_list);
5924 INIT_LIST_HEAD(&del_list);
5925 put = btrfs_readdir_get_delayed_items(inode, &ins_list, &del_list);
5928 key.type = BTRFS_DIR_INDEX_KEY;
5929 key.offset = ctx->pos;
5930 key.objectid = btrfs_ino(BTRFS_I(inode));
5932 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5937 struct dir_entry *entry;
5939 leaf = path->nodes[0];
5940 slot = path->slots[0];
5941 if (slot >= btrfs_header_nritems(leaf)) {
5942 ret = btrfs_next_leaf(root, path);
5950 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5952 if (found_key.objectid != key.objectid)
5954 if (found_key.type != BTRFS_DIR_INDEX_KEY)
5956 if (found_key.offset < ctx->pos)
5958 if (btrfs_should_delete_dir_index(&del_list, found_key.offset))
5960 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5961 name_len = btrfs_dir_name_len(leaf, di);
5962 if ((total_len + sizeof(struct dir_entry) + name_len) >=
5964 btrfs_release_path(path);
5965 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5968 addr = private->filldir_buf;
5975 put_unaligned(name_len, &entry->name_len);
5976 name_ptr = (char *)(entry + 1);
5977 read_extent_buffer(leaf, name_ptr, (unsigned long)(di + 1),
5979 put_unaligned(btrfs_filetype_table[btrfs_dir_type(leaf, di)],
5981 btrfs_dir_item_key_to_cpu(leaf, di, &location);
5982 put_unaligned(location.objectid, &entry->ino);
5983 put_unaligned(found_key.offset, &entry->offset);
5985 addr += sizeof(struct dir_entry) + name_len;
5986 total_len += sizeof(struct dir_entry) + name_len;
5990 btrfs_release_path(path);
5992 ret = btrfs_filldir(private->filldir_buf, entries, ctx);
5996 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
6001 * Stop new entries from being returned after we return the last
6004 * New directory entries are assigned a strictly increasing
6005 * offset. This means that new entries created during readdir
6006 * are *guaranteed* to be seen in the future by that readdir.
6007 * This has broken buggy programs which operate on names as
6008 * they're returned by readdir. Until we re-use freed offsets
6009 * we have this hack to stop new entries from being returned
6010 * under the assumption that they'll never reach this huge
6013 * This is being careful not to overflow 32bit loff_t unless the
6014 * last entry requires it because doing so has broken 32bit apps
6017 if (ctx->pos >= INT_MAX)
6018 ctx->pos = LLONG_MAX;
6025 btrfs_readdir_put_delayed_items(inode, &ins_list, &del_list);
6026 btrfs_free_path(path);
6030 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
6032 struct btrfs_root *root = BTRFS_I(inode)->root;
6033 struct btrfs_trans_handle *trans;
6035 bool nolock = false;
6037 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6040 if (btrfs_fs_closing(root->fs_info) &&
6041 btrfs_is_free_space_inode(BTRFS_I(inode)))
6044 if (wbc->sync_mode == WB_SYNC_ALL) {
6046 trans = btrfs_join_transaction_nolock(root);
6048 trans = btrfs_join_transaction(root);
6050 return PTR_ERR(trans);
6051 ret = btrfs_commit_transaction(trans);
6057 * This is somewhat expensive, updating the tree every time the
6058 * inode changes. But, it is most likely to find the inode in cache.
6059 * FIXME, needs more benchmarking...there are no reasons other than performance
6060 * to keep or drop this code.
6062 static int btrfs_dirty_inode(struct inode *inode)
6064 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6065 struct btrfs_root *root = BTRFS_I(inode)->root;
6066 struct btrfs_trans_handle *trans;
6069 if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
6072 trans = btrfs_join_transaction(root);
6074 return PTR_ERR(trans);
6076 ret = btrfs_update_inode(trans, root, inode);
6077 if (ret && ret == -ENOSPC) {
6078 /* whoops, lets try again with the full transaction */
6079 btrfs_end_transaction(trans);
6080 trans = btrfs_start_transaction(root, 1);
6082 return PTR_ERR(trans);
6084 ret = btrfs_update_inode(trans, root, inode);
6086 btrfs_end_transaction(trans);
6087 if (BTRFS_I(inode)->delayed_node)
6088 btrfs_balance_delayed_items(fs_info);
6094 * This is a copy of file_update_time. We need this so we can return error on
6095 * ENOSPC for updating the inode in the case of file write and mmap writes.
6097 static int btrfs_update_time(struct inode *inode, struct timespec64 *now,
6100 struct btrfs_root *root = BTRFS_I(inode)->root;
6101 bool dirty = flags & ~S_VERSION;
6103 if (btrfs_root_readonly(root))
6106 if (flags & S_VERSION)
6107 dirty |= inode_maybe_inc_iversion(inode, dirty);
6108 if (flags & S_CTIME)
6109 inode->i_ctime = *now;
6110 if (flags & S_MTIME)
6111 inode->i_mtime = *now;
6112 if (flags & S_ATIME)
6113 inode->i_atime = *now;
6114 return dirty ? btrfs_dirty_inode(inode) : 0;
6118 * find the highest existing sequence number in a directory
6119 * and then set the in-memory index_cnt variable to reflect
6120 * free sequence numbers
6122 static int btrfs_set_inode_index_count(struct btrfs_inode *inode)
6124 struct btrfs_root *root = inode->root;
6125 struct btrfs_key key, found_key;
6126 struct btrfs_path *path;
6127 struct extent_buffer *leaf;
6130 key.objectid = btrfs_ino(inode);
6131 key.type = BTRFS_DIR_INDEX_KEY;
6132 key.offset = (u64)-1;
6134 path = btrfs_alloc_path();
6138 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
6141 /* FIXME: we should be able to handle this */
6147 * MAGIC NUMBER EXPLANATION:
6148 * since we search a directory based on f_pos we have to start at 2
6149 * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
6150 * else has to start at 2
6152 if (path->slots[0] == 0) {
6153 inode->index_cnt = 2;
6159 leaf = path->nodes[0];
6160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6162 if (found_key.objectid != btrfs_ino(inode) ||
6163 found_key.type != BTRFS_DIR_INDEX_KEY) {
6164 inode->index_cnt = 2;
6168 inode->index_cnt = found_key.offset + 1;
6170 btrfs_free_path(path);
6175 * helper to find a free sequence number in a given directory. This current
6176 * code is very simple, later versions will do smarter things in the btree
6178 int btrfs_set_inode_index(struct btrfs_inode *dir, u64 *index)
6182 if (dir->index_cnt == (u64)-1) {
6183 ret = btrfs_inode_delayed_dir_index_count(dir);
6185 ret = btrfs_set_inode_index_count(dir);
6191 *index = dir->index_cnt;
6197 static int btrfs_insert_inode_locked(struct inode *inode)
6199 struct btrfs_iget_args args;
6200 args.location = &BTRFS_I(inode)->location;
6201 args.root = BTRFS_I(inode)->root;
6203 return insert_inode_locked4(inode,
6204 btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
6205 btrfs_find_actor, &args);
6209 * Inherit flags from the parent inode.
6211 * Currently only the compression flags and the cow flags are inherited.
6213 static void btrfs_inherit_iflags(struct inode *inode, struct inode *dir)
6220 flags = BTRFS_I(dir)->flags;
6222 if (flags & BTRFS_INODE_NOCOMPRESS) {
6223 BTRFS_I(inode)->flags &= ~BTRFS_INODE_COMPRESS;
6224 BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
6225 } else if (flags & BTRFS_INODE_COMPRESS) {
6226 BTRFS_I(inode)->flags &= ~BTRFS_INODE_NOCOMPRESS;
6227 BTRFS_I(inode)->flags |= BTRFS_INODE_COMPRESS;
6230 if (flags & BTRFS_INODE_NODATACOW) {
6231 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
6232 if (S_ISREG(inode->i_mode))
6233 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6236 btrfs_sync_inode_flags_to_i_flags(inode);
6239 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
6240 struct btrfs_root *root,
6242 const char *name, int name_len,
6243 u64 ref_objectid, u64 objectid,
6244 umode_t mode, u64 *index)
6246 struct btrfs_fs_info *fs_info = root->fs_info;
6247 struct inode *inode;
6248 struct btrfs_inode_item *inode_item;
6249 struct btrfs_key *location;
6250 struct btrfs_path *path;
6251 struct btrfs_inode_ref *ref;
6252 struct btrfs_key key[2];
6254 int nitems = name ? 2 : 1;
6258 path = btrfs_alloc_path();
6260 return ERR_PTR(-ENOMEM);
6262 inode = new_inode(fs_info->sb);
6264 btrfs_free_path(path);
6265 return ERR_PTR(-ENOMEM);
6269 * O_TMPFILE, set link count to 0, so that after this point,
6270 * we fill in an inode item with the correct link count.
6273 set_nlink(inode, 0);
6276 * we have to initialize this early, so we can reclaim the inode
6277 * number if we fail afterwards in this function.
6279 inode->i_ino = objectid;
6282 trace_btrfs_inode_request(dir);
6284 ret = btrfs_set_inode_index(BTRFS_I(dir), index);
6286 btrfs_free_path(path);
6288 return ERR_PTR(ret);
6294 * index_cnt is ignored for everything but a dir,
6295 * btrfs_set_inode_index_count has an explanation for the magic
6298 BTRFS_I(inode)->index_cnt = 2;
6299 BTRFS_I(inode)->dir_index = *index;
6300 BTRFS_I(inode)->root = root;
6301 BTRFS_I(inode)->generation = trans->transid;
6302 inode->i_generation = BTRFS_I(inode)->generation;
6305 * We could have gotten an inode number from somebody who was fsynced
6306 * and then removed in this same transaction, so let's just set full
6307 * sync since it will be a full sync anyway and this will blow away the
6308 * old info in the log.
6310 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6312 key[0].objectid = objectid;
6313 key[0].type = BTRFS_INODE_ITEM_KEY;
6316 sizes[0] = sizeof(struct btrfs_inode_item);
6320 * Start new inodes with an inode_ref. This is slightly more
6321 * efficient for small numbers of hard links since they will
6322 * be packed into one item. Extended refs will kick in if we
6323 * add more hard links than can fit in the ref item.
6325 key[1].objectid = objectid;
6326 key[1].type = BTRFS_INODE_REF_KEY;
6327 key[1].offset = ref_objectid;
6329 sizes[1] = name_len + sizeof(*ref);
6332 location = &BTRFS_I(inode)->location;
6333 location->objectid = objectid;
6334 location->offset = 0;
6335 location->type = BTRFS_INODE_ITEM_KEY;
6337 ret = btrfs_insert_inode_locked(inode);
6341 path->leave_spinning = 1;
6342 ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
6346 inode_init_owner(inode, dir, mode);
6347 inode_set_bytes(inode, 0);
6349 inode->i_mtime = current_time(inode);
6350 inode->i_atime = inode->i_mtime;
6351 inode->i_ctime = inode->i_mtime;
6352 BTRFS_I(inode)->i_otime = timespec64_to_timespec(inode->i_mtime);
6354 inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
6355 struct btrfs_inode_item);
6356 memzero_extent_buffer(path->nodes[0], (unsigned long)inode_item,
6357 sizeof(*inode_item));
6358 fill_inode_item(trans, path->nodes[0], inode_item, inode);
6361 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
6362 struct btrfs_inode_ref);
6363 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
6364 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
6365 ptr = (unsigned long)(ref + 1);
6366 write_extent_buffer(path->nodes[0], name, ptr, name_len);
6369 btrfs_mark_buffer_dirty(path->nodes[0]);
6370 btrfs_free_path(path);
6372 btrfs_inherit_iflags(inode, dir);
6374 if (S_ISREG(mode)) {
6375 if (btrfs_test_opt(fs_info, NODATASUM))
6376 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
6377 if (btrfs_test_opt(fs_info, NODATACOW))
6378 BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
6379 BTRFS_INODE_NODATASUM;
6382 inode_tree_add(inode);
6384 trace_btrfs_inode_new(inode);
6385 btrfs_set_inode_last_trans(trans, inode);
6387 btrfs_update_root_times(trans, root);
6389 ret = btrfs_inode_inherit_props(trans, inode, dir);
6392 "error inheriting props for ino %llu (root %llu): %d",
6393 btrfs_ino(BTRFS_I(inode)), root->root_key.objectid, ret);
6398 unlock_new_inode(inode);
6401 BTRFS_I(dir)->index_cnt--;
6402 btrfs_free_path(path);
6404 return ERR_PTR(ret);
6407 static inline u8 btrfs_inode_type(struct inode *inode)
6409 return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
6413 * utility function to add 'inode' into 'parent_inode' with
6414 * a give name and a given sequence number.
6415 * if 'add_backref' is true, also insert a backref from the
6416 * inode to the parent directory.
6418 int btrfs_add_link(struct btrfs_trans_handle *trans,
6419 struct btrfs_inode *parent_inode, struct btrfs_inode *inode,
6420 const char *name, int name_len, int add_backref, u64 index)
6422 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6424 struct btrfs_key key;
6425 struct btrfs_root *root = parent_inode->root;
6426 u64 ino = btrfs_ino(inode);
6427 u64 parent_ino = btrfs_ino(parent_inode);
6429 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6430 memcpy(&key, &inode->root->root_key, sizeof(key));
6433 key.type = BTRFS_INODE_ITEM_KEY;
6437 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6438 ret = btrfs_add_root_ref(trans, fs_info, key.objectid,
6439 root->root_key.objectid, parent_ino,
6440 index, name, name_len);
6441 } else if (add_backref) {
6442 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
6446 /* Nothing to clean up yet */
6450 ret = btrfs_insert_dir_item(trans, root, name, name_len,
6452 btrfs_inode_type(&inode->vfs_inode), index);
6453 if (ret == -EEXIST || ret == -EOVERFLOW)
6456 btrfs_abort_transaction(trans, ret);
6460 btrfs_i_size_write(parent_inode, parent_inode->vfs_inode.i_size +
6462 inode_inc_iversion(&parent_inode->vfs_inode);
6463 parent_inode->vfs_inode.i_mtime = parent_inode->vfs_inode.i_ctime =
6464 current_time(&parent_inode->vfs_inode);
6465 ret = btrfs_update_inode(trans, root, &parent_inode->vfs_inode);
6467 btrfs_abort_transaction(trans, ret);
6471 if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6474 err = btrfs_del_root_ref(trans, fs_info, key.objectid,
6475 root->root_key.objectid, parent_ino,
6476 &local_index, name, name_len);
6478 } else if (add_backref) {
6482 err = btrfs_del_inode_ref(trans, root, name, name_len,
6483 ino, parent_ino, &local_index);
6488 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6489 struct btrfs_inode *dir, struct dentry *dentry,
6490 struct btrfs_inode *inode, int backref, u64 index)
6492 int err = btrfs_add_link(trans, dir, inode,
6493 dentry->d_name.name, dentry->d_name.len,
6500 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6501 umode_t mode, dev_t rdev)
6503 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6504 struct btrfs_trans_handle *trans;
6505 struct btrfs_root *root = BTRFS_I(dir)->root;
6506 struct inode *inode = NULL;
6513 * 2 for inode item and ref
6515 * 1 for xattr if selinux is on
6517 trans = btrfs_start_transaction(root, 5);
6519 return PTR_ERR(trans);
6521 err = btrfs_find_free_ino(root, &objectid);
6525 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6526 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6528 if (IS_ERR(inode)) {
6529 err = PTR_ERR(inode);
6534 * If the active LSM wants to access the inode during
6535 * d_instantiate it needs these. Smack checks to see
6536 * if the filesystem supports xattrs by looking at the
6539 inode->i_op = &btrfs_special_inode_operations;
6540 init_special_inode(inode, inode->i_mode, rdev);
6542 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6544 goto out_unlock_inode;
6546 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6549 goto out_unlock_inode;
6551 btrfs_update_inode(trans, root, inode);
6552 d_instantiate_new(dentry, inode);
6556 btrfs_end_transaction(trans);
6557 btrfs_btree_balance_dirty(fs_info);
6559 inode_dec_link_count(inode);
6566 unlock_new_inode(inode);
6571 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6572 umode_t mode, bool excl)
6574 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6575 struct btrfs_trans_handle *trans;
6576 struct btrfs_root *root = BTRFS_I(dir)->root;
6577 struct inode *inode = NULL;
6578 int drop_inode_on_err = 0;
6584 * 2 for inode item and ref
6586 * 1 for xattr if selinux is on
6588 trans = btrfs_start_transaction(root, 5);
6590 return PTR_ERR(trans);
6592 err = btrfs_find_free_ino(root, &objectid);
6596 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6597 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6599 if (IS_ERR(inode)) {
6600 err = PTR_ERR(inode);
6603 drop_inode_on_err = 1;
6605 * If the active LSM wants to access the inode during
6606 * d_instantiate it needs these. Smack checks to see
6607 * if the filesystem supports xattrs by looking at the
6610 inode->i_fop = &btrfs_file_operations;
6611 inode->i_op = &btrfs_file_inode_operations;
6612 inode->i_mapping->a_ops = &btrfs_aops;
6614 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6616 goto out_unlock_inode;
6618 err = btrfs_update_inode(trans, root, inode);
6620 goto out_unlock_inode;
6622 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6625 goto out_unlock_inode;
6627 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6628 d_instantiate_new(dentry, inode);
6631 btrfs_end_transaction(trans);
6632 if (err && drop_inode_on_err) {
6633 inode_dec_link_count(inode);
6636 btrfs_btree_balance_dirty(fs_info);
6640 unlock_new_inode(inode);
6645 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6646 struct dentry *dentry)
6648 struct btrfs_trans_handle *trans = NULL;
6649 struct btrfs_root *root = BTRFS_I(dir)->root;
6650 struct inode *inode = d_inode(old_dentry);
6651 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
6656 /* do not allow sys_link's with other subvols of the same device */
6657 if (root->objectid != BTRFS_I(inode)->root->objectid)
6660 if (inode->i_nlink >= BTRFS_LINK_MAX)
6663 err = btrfs_set_inode_index(BTRFS_I(dir), &index);
6668 * 2 items for inode and inode ref
6669 * 2 items for dir items
6670 * 1 item for parent inode
6671 * 1 item for orphan item deletion if O_TMPFILE
6673 trans = btrfs_start_transaction(root, inode->i_nlink ? 5 : 6);
6674 if (IS_ERR(trans)) {
6675 err = PTR_ERR(trans);
6680 /* There are several dir indexes for this inode, clear the cache. */
6681 BTRFS_I(inode)->dir_index = 0ULL;
6683 inode_inc_iversion(inode);
6684 inode->i_ctime = current_time(inode);
6686 set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6688 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry, BTRFS_I(inode),
6694 struct dentry *parent = dentry->d_parent;
6695 err = btrfs_update_inode(trans, root, inode);
6698 if (inode->i_nlink == 1) {
6700 * If new hard link count is 1, it's a file created
6701 * with open(2) O_TMPFILE flag.
6703 err = btrfs_orphan_del(trans, BTRFS_I(inode));
6707 d_instantiate(dentry, inode);
6708 btrfs_log_new_name(trans, BTRFS_I(inode), NULL, parent);
6713 btrfs_end_transaction(trans);
6715 inode_dec_link_count(inode);
6718 btrfs_btree_balance_dirty(fs_info);
6722 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6724 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
6725 struct inode *inode = NULL;
6726 struct btrfs_trans_handle *trans;
6727 struct btrfs_root *root = BTRFS_I(dir)->root;
6729 int drop_on_err = 0;
6734 * 2 items for inode and ref
6735 * 2 items for dir items
6736 * 1 for xattr if selinux is on
6738 trans = btrfs_start_transaction(root, 5);
6740 return PTR_ERR(trans);
6742 err = btrfs_find_free_ino(root, &objectid);
6746 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6747 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)), objectid,
6748 S_IFDIR | mode, &index);
6749 if (IS_ERR(inode)) {
6750 err = PTR_ERR(inode);
6755 /* these must be set before we unlock the inode */
6756 inode->i_op = &btrfs_dir_inode_operations;
6757 inode->i_fop = &btrfs_dir_file_operations;
6759 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6761 goto out_fail_inode;
6763 btrfs_i_size_write(BTRFS_I(inode), 0);
6764 err = btrfs_update_inode(trans, root, inode);
6766 goto out_fail_inode;
6768 err = btrfs_add_link(trans, BTRFS_I(dir), BTRFS_I(inode),
6769 dentry->d_name.name,
6770 dentry->d_name.len, 0, index);
6772 goto out_fail_inode;
6774 d_instantiate_new(dentry, inode);
6778 btrfs_end_transaction(trans);
6780 inode_dec_link_count(inode);
6783 btrfs_btree_balance_dirty(fs_info);
6787 unlock_new_inode(inode);
6791 static noinline int uncompress_inline(struct btrfs_path *path,
6793 size_t pg_offset, u64 extent_offset,
6794 struct btrfs_file_extent_item *item)
6797 struct extent_buffer *leaf = path->nodes[0];
6800 unsigned long inline_size;
6804 WARN_ON(pg_offset != 0);
6805 compress_type = btrfs_file_extent_compression(leaf, item);
6806 max_size = btrfs_file_extent_ram_bytes(leaf, item);
6807 inline_size = btrfs_file_extent_inline_item_len(leaf,
6808 btrfs_item_nr(path->slots[0]));
6809 tmp = kmalloc(inline_size, GFP_NOFS);
6812 ptr = btrfs_file_extent_inline_start(item);
6814 read_extent_buffer(leaf, tmp, ptr, inline_size);
6816 max_size = min_t(unsigned long, PAGE_SIZE, max_size);
6817 ret = btrfs_decompress(compress_type, tmp, page,
6818 extent_offset, inline_size, max_size);
6821 * decompression code contains a memset to fill in any space between the end
6822 * of the uncompressed data and the end of max_size in case the decompressed
6823 * data ends up shorter than ram_bytes. That doesn't cover the hole between
6824 * the end of an inline extent and the beginning of the next block, so we
6825 * cover that region here.
6828 if (max_size + pg_offset < PAGE_SIZE) {
6829 char *map = kmap(page);
6830 memset(map + pg_offset + max_size, 0, PAGE_SIZE - max_size - pg_offset);
6838 * a bit scary, this does extent mapping from logical file offset to the disk.
6839 * the ugly parts come from merging extents from the disk with the in-ram
6840 * representation. This gets more complex because of the data=ordered code,
6841 * where the in-ram extents might be locked pending data=ordered completion.
6843 * This also copies inline extents directly into the page.
6845 struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
6847 size_t pg_offset, u64 start, u64 len,
6850 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6853 u64 extent_start = 0;
6855 u64 objectid = btrfs_ino(inode);
6857 struct btrfs_path *path = NULL;
6858 struct btrfs_root *root = inode->root;
6859 struct btrfs_file_extent_item *item;
6860 struct extent_buffer *leaf;
6861 struct btrfs_key found_key;
6862 struct extent_map *em = NULL;
6863 struct extent_map_tree *em_tree = &inode->extent_tree;
6864 struct extent_io_tree *io_tree = &inode->io_tree;
6865 const bool new_inline = !page || create;
6867 read_lock(&em_tree->lock);
6868 em = lookup_extent_mapping(em_tree, start, len);
6870 em->bdev = fs_info->fs_devices->latest_bdev;
6871 read_unlock(&em_tree->lock);
6874 if (em->start > start || em->start + em->len <= start)
6875 free_extent_map(em);
6876 else if (em->block_start == EXTENT_MAP_INLINE && page)
6877 free_extent_map(em);
6881 em = alloc_extent_map();
6886 em->bdev = fs_info->fs_devices->latest_bdev;
6887 em->start = EXTENT_MAP_HOLE;
6888 em->orig_start = EXTENT_MAP_HOLE;
6890 em->block_len = (u64)-1;
6893 path = btrfs_alloc_path();
6899 * Chances are we'll be called again, so go ahead and do
6902 path->reada = READA_FORWARD;
6905 ret = btrfs_lookup_file_extent(NULL, root, path, objectid, start, 0);
6912 if (path->slots[0] == 0)
6917 leaf = path->nodes[0];
6918 item = btrfs_item_ptr(leaf, path->slots[0],
6919 struct btrfs_file_extent_item);
6920 /* are we inside the extent that was found? */
6921 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6922 found_type = found_key.type;
6923 if (found_key.objectid != objectid ||
6924 found_type != BTRFS_EXTENT_DATA_KEY) {
6926 * If we backup past the first extent we want to move forward
6927 * and see if there is an extent in front of us, otherwise we'll
6928 * say there is a hole for our whole search range which can
6935 found_type = btrfs_file_extent_type(leaf, item);
6936 extent_start = found_key.offset;
6937 if (found_type == BTRFS_FILE_EXTENT_REG ||
6938 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6939 extent_end = extent_start +
6940 btrfs_file_extent_num_bytes(leaf, item);
6942 trace_btrfs_get_extent_show_fi_regular(inode, leaf, item,
6944 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6946 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6947 extent_end = ALIGN(extent_start + size,
6948 fs_info->sectorsize);
6950 trace_btrfs_get_extent_show_fi_inline(inode, leaf, item,
6955 if (start >= extent_end) {
6957 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6958 ret = btrfs_next_leaf(root, path);
6965 leaf = path->nodes[0];
6967 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6968 if (found_key.objectid != objectid ||
6969 found_key.type != BTRFS_EXTENT_DATA_KEY)
6971 if (start + len <= found_key.offset)
6973 if (start > found_key.offset)
6976 em->orig_start = start;
6977 em->len = found_key.offset - start;
6981 btrfs_extent_item_to_extent_map(inode, path, item,
6984 if (found_type == BTRFS_FILE_EXTENT_REG ||
6985 found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6987 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6991 size_t extent_offset;
6997 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6998 extent_offset = page_offset(page) + pg_offset - extent_start;
6999 copy_size = min_t(u64, PAGE_SIZE - pg_offset,
7000 size - extent_offset);
7001 em->start = extent_start + extent_offset;
7002 em->len = ALIGN(copy_size, fs_info->sectorsize);
7003 em->orig_block_len = em->len;
7004 em->orig_start = em->start;
7005 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
7006 if (!PageUptodate(page)) {
7007 if (btrfs_file_extent_compression(leaf, item) !=
7008 BTRFS_COMPRESS_NONE) {
7009 ret = uncompress_inline(path, page, pg_offset,
7010 extent_offset, item);
7017 read_extent_buffer(leaf, map + pg_offset, ptr,
7019 if (pg_offset + copy_size < PAGE_SIZE) {
7020 memset(map + pg_offset + copy_size, 0,
7021 PAGE_SIZE - pg_offset -
7026 flush_dcache_page(page);
7028 set_extent_uptodate(io_tree, em->start,
7029 extent_map_end(em) - 1, NULL, GFP_NOFS);
7034 em->orig_start = start;
7037 em->block_start = EXTENT_MAP_HOLE;
7039 btrfs_release_path(path);
7040 if (em->start > start || extent_map_end(em) <= start) {
7042 "bad extent! em: [%llu %llu] passed [%llu %llu]",
7043 em->start, em->len, start, len);
7049 write_lock(&em_tree->lock);
7050 err = btrfs_add_extent_mapping(fs_info, em_tree, &em, start, len);
7051 write_unlock(&em_tree->lock);
7054 trace_btrfs_get_extent(root, inode, em);
7056 btrfs_free_path(path);
7058 free_extent_map(em);
7059 return ERR_PTR(err);
7061 BUG_ON(!em); /* Error is always set */
7065 struct extent_map *btrfs_get_extent_fiemap(struct btrfs_inode *inode,
7067 size_t pg_offset, u64 start, u64 len,
7070 struct extent_map *em;
7071 struct extent_map *hole_em = NULL;
7072 u64 range_start = start;
7078 em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
7082 * If our em maps to:
7084 * - a pre-alloc extent,
7085 * there might actually be delalloc bytes behind it.
7087 if (em->block_start != EXTENT_MAP_HOLE &&
7088 !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7093 /* check to see if we've wrapped (len == -1 or similar) */
7102 /* ok, we didn't find anything, lets look for delalloc */
7103 found = count_range_bits(&inode->io_tree, &range_start,
7104 end, len, EXTENT_DELALLOC, 1);
7105 found_end = range_start + found;
7106 if (found_end < range_start)
7107 found_end = (u64)-1;
7110 * we didn't find anything useful, return
7111 * the original results from get_extent()
7113 if (range_start > end || found_end <= start) {
7119 /* adjust the range_start to make sure it doesn't
7120 * go backwards from the start they passed in
7122 range_start = max(start, range_start);
7123 found = found_end - range_start;
7126 u64 hole_start = start;
7129 em = alloc_extent_map();
7135 * when btrfs_get_extent can't find anything it
7136 * returns one huge hole
7138 * make sure what it found really fits our range, and
7139 * adjust to make sure it is based on the start from
7143 u64 calc_end = extent_map_end(hole_em);
7145 if (calc_end <= start || (hole_em->start > end)) {
7146 free_extent_map(hole_em);
7149 hole_start = max(hole_em->start, start);
7150 hole_len = calc_end - hole_start;
7154 if (hole_em && range_start > hole_start) {
7155 /* our hole starts before our delalloc, so we
7156 * have to return just the parts of the hole
7157 * that go until the delalloc starts
7159 em->len = min(hole_len,
7160 range_start - hole_start);
7161 em->start = hole_start;
7162 em->orig_start = hole_start;
7164 * don't adjust block start at all,
7165 * it is fixed at EXTENT_MAP_HOLE
7167 em->block_start = hole_em->block_start;
7168 em->block_len = hole_len;
7169 if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
7170 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7172 em->start = range_start;
7174 em->orig_start = range_start;
7175 em->block_start = EXTENT_MAP_DELALLOC;
7176 em->block_len = found;
7183 free_extent_map(hole_em);
7185 free_extent_map(em);
7186 return ERR_PTR(err);
7191 static struct extent_map *btrfs_create_dio_extent(struct inode *inode,
7194 const u64 orig_start,
7195 const u64 block_start,
7196 const u64 block_len,
7197 const u64 orig_block_len,
7198 const u64 ram_bytes,
7201 struct extent_map *em = NULL;
7204 if (type != BTRFS_ORDERED_NOCOW) {
7205 em = create_io_em(inode, start, len, orig_start,
7206 block_start, block_len, orig_block_len,
7208 BTRFS_COMPRESS_NONE, /* compress_type */
7213 ret = btrfs_add_ordered_extent_dio(inode, start, block_start,
7214 len, block_len, type);
7217 free_extent_map(em);
7218 btrfs_drop_extent_cache(BTRFS_I(inode), start,
7219 start + len - 1, 0);
7228 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
7231 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7232 struct btrfs_root *root = BTRFS_I(inode)->root;
7233 struct extent_map *em;
7234 struct btrfs_key ins;
7238 alloc_hint = get_extent_allocation_hint(inode, start, len);
7239 ret = btrfs_reserve_extent(root, len, len, fs_info->sectorsize,
7240 0, alloc_hint, &ins, 1, 1);
7242 return ERR_PTR(ret);
7244 em = btrfs_create_dio_extent(inode, start, ins.offset, start,
7245 ins.objectid, ins.offset, ins.offset,
7246 ins.offset, BTRFS_ORDERED_REGULAR);
7247 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
7249 btrfs_free_reserved_extent(fs_info, ins.objectid,
7256 * returns 1 when the nocow is safe, < 1 on error, 0 if the
7257 * block must be cow'd
7259 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
7260 u64 *orig_start, u64 *orig_block_len,
7263 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7264 struct btrfs_path *path;
7266 struct extent_buffer *leaf;
7267 struct btrfs_root *root = BTRFS_I(inode)->root;
7268 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7269 struct btrfs_file_extent_item *fi;
7270 struct btrfs_key key;
7277 bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
7279 path = btrfs_alloc_path();
7283 ret = btrfs_lookup_file_extent(NULL, root, path,
7284 btrfs_ino(BTRFS_I(inode)), offset, 0);
7288 slot = path->slots[0];
7291 /* can't find the item, must cow */
7298 leaf = path->nodes[0];
7299 btrfs_item_key_to_cpu(leaf, &key, slot);
7300 if (key.objectid != btrfs_ino(BTRFS_I(inode)) ||
7301 key.type != BTRFS_EXTENT_DATA_KEY) {
7302 /* not our file or wrong item type, must cow */
7306 if (key.offset > offset) {
7307 /* Wrong offset, must cow */
7311 fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
7312 found_type = btrfs_file_extent_type(leaf, fi);
7313 if (found_type != BTRFS_FILE_EXTENT_REG &&
7314 found_type != BTRFS_FILE_EXTENT_PREALLOC) {
7315 /* not a regular extent, must cow */
7319 if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
7322 extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
7323 if (extent_end <= offset)
7326 disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
7327 if (disk_bytenr == 0)
7330 if (btrfs_file_extent_compression(leaf, fi) ||
7331 btrfs_file_extent_encryption(leaf, fi) ||
7332 btrfs_file_extent_other_encoding(leaf, fi))
7336 * Do the same check as in btrfs_cross_ref_exist but without the
7337 * unnecessary search.
7339 if (btrfs_file_extent_generation(leaf, fi) <=
7340 btrfs_root_last_snapshot(&root->root_item))
7343 backref_offset = btrfs_file_extent_offset(leaf, fi);
7346 *orig_start = key.offset - backref_offset;
7347 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
7348 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
7351 if (btrfs_extent_readonly(fs_info, disk_bytenr))
7354 num_bytes = min(offset + *len, extent_end) - offset;
7355 if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
7358 range_end = round_up(offset + num_bytes,
7359 root->fs_info->sectorsize) - 1;
7360 ret = test_range_bit(io_tree, offset, range_end,
7361 EXTENT_DELALLOC, 0, NULL);
7368 btrfs_release_path(path);
7371 * look for other files referencing this extent, if we
7372 * find any we must cow
7375 ret = btrfs_cross_ref_exist(root, btrfs_ino(BTRFS_I(inode)),
7376 key.offset - backref_offset, disk_bytenr);
7383 * adjust disk_bytenr and num_bytes to cover just the bytes
7384 * in this extent we are about to write. If there
7385 * are any csums in that range we have to cow in order
7386 * to keep the csums correct
7388 disk_bytenr += backref_offset;
7389 disk_bytenr += offset - key.offset;
7390 if (csum_exist_in_range(fs_info, disk_bytenr, num_bytes))
7393 * all of the above have passed, it is safe to overwrite this extent
7399 btrfs_free_path(path);
7403 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7404 struct extent_state **cached_state, int writing)
7406 struct btrfs_ordered_extent *ordered;
7410 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7413 * We're concerned with the entire range that we're going to be
7414 * doing DIO to, so we need to make sure there's no ordered
7415 * extents in this range.
7417 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), lockstart,
7418 lockend - lockstart + 1);
7421 * We need to make sure there are no buffered pages in this
7422 * range either, we could have raced between the invalidate in
7423 * generic_file_direct_write and locking the extent. The
7424 * invalidate needs to happen so that reads after a write do not
7428 (!writing || !filemap_range_has_page(inode->i_mapping,
7429 lockstart, lockend)))
7432 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7437 * If we are doing a DIO read and the ordered extent we
7438 * found is for a buffered write, we can not wait for it
7439 * to complete and retry, because if we do so we can
7440 * deadlock with concurrent buffered writes on page
7441 * locks. This happens only if our DIO read covers more
7442 * than one extent map, if at this point has already
7443 * created an ordered extent for a previous extent map
7444 * and locked its range in the inode's io tree, and a
7445 * concurrent write against that previous extent map's
7446 * range and this range started (we unlock the ranges
7447 * in the io tree only when the bios complete and
7448 * buffered writes always lock pages before attempting
7449 * to lock range in the io tree).
7452 test_bit(BTRFS_ORDERED_DIRECT, &ordered->flags))
7453 btrfs_start_ordered_extent(inode, ordered, 1);
7456 btrfs_put_ordered_extent(ordered);
7459 * We could trigger writeback for this range (and wait
7460 * for it to complete) and then invalidate the pages for
7461 * this range (through invalidate_inode_pages2_range()),
7462 * but that can lead us to a deadlock with a concurrent
7463 * call to readpages() (a buffered read or a defrag call
7464 * triggered a readahead) on a page lock due to an
7465 * ordered dio extent we created before but did not have
7466 * yet a corresponding bio submitted (whence it can not
7467 * complete), which makes readpages() wait for that
7468 * ordered extent to complete while holding a lock on
7483 /* The callers of this must take lock_extent() */
7484 static struct extent_map *create_io_em(struct inode *inode, u64 start, u64 len,
7485 u64 orig_start, u64 block_start,
7486 u64 block_len, u64 orig_block_len,
7487 u64 ram_bytes, int compress_type,
7490 struct extent_map_tree *em_tree;
7491 struct extent_map *em;
7492 struct btrfs_root *root = BTRFS_I(inode)->root;
7495 ASSERT(type == BTRFS_ORDERED_PREALLOC ||
7496 type == BTRFS_ORDERED_COMPRESSED ||
7497 type == BTRFS_ORDERED_NOCOW ||
7498 type == BTRFS_ORDERED_REGULAR);
7500 em_tree = &BTRFS_I(inode)->extent_tree;
7501 em = alloc_extent_map();
7503 return ERR_PTR(-ENOMEM);
7506 em->orig_start = orig_start;
7508 em->block_len = block_len;
7509 em->block_start = block_start;
7510 em->bdev = root->fs_info->fs_devices->latest_bdev;
7511 em->orig_block_len = orig_block_len;
7512 em->ram_bytes = ram_bytes;
7513 em->generation = -1;
7514 set_bit(EXTENT_FLAG_PINNED, &em->flags);
7515 if (type == BTRFS_ORDERED_PREALLOC) {
7516 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7517 } else if (type == BTRFS_ORDERED_COMPRESSED) {
7518 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
7519 em->compress_type = compress_type;
7523 btrfs_drop_extent_cache(BTRFS_I(inode), em->start,
7524 em->start + em->len - 1, 0);
7525 write_lock(&em_tree->lock);
7526 ret = add_extent_mapping(em_tree, em, 1);
7527 write_unlock(&em_tree->lock);
7529 * The caller has taken lock_extent(), who could race with us
7532 } while (ret == -EEXIST);
7535 free_extent_map(em);
7536 return ERR_PTR(ret);
7539 /* em got 2 refs now, callers needs to do free_extent_map once. */
7544 static int btrfs_get_blocks_direct_read(struct extent_map *em,
7545 struct buffer_head *bh_result,
7546 struct inode *inode,
7549 if (em->block_start == EXTENT_MAP_HOLE ||
7550 test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7553 len = min(len, em->len - (start - em->start));
7555 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7557 bh_result->b_size = len;
7558 bh_result->b_bdev = em->bdev;
7559 set_buffer_mapped(bh_result);
7564 static int btrfs_get_blocks_direct_write(struct extent_map **map,
7565 struct buffer_head *bh_result,
7566 struct inode *inode,
7567 struct btrfs_dio_data *dio_data,
7570 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7571 struct extent_map *em = *map;
7575 * We don't allocate a new extent in the following cases
7577 * 1) The inode is marked as NODATACOW. In this case we'll just use the
7579 * 2) The extent is marked as PREALLOC. We're good to go here and can
7580 * just use the extent.
7583 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7584 ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7585 em->block_start != EXTENT_MAP_HOLE)) {
7587 u64 block_start, orig_start, orig_block_len, ram_bytes;
7589 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7590 type = BTRFS_ORDERED_PREALLOC;
7592 type = BTRFS_ORDERED_NOCOW;
7593 len = min(len, em->len - (start - em->start));
7594 block_start = em->block_start + (start - em->start);
7596 if (can_nocow_extent(inode, start, &len, &orig_start,
7597 &orig_block_len, &ram_bytes) == 1 &&
7598 btrfs_inc_nocow_writers(fs_info, block_start)) {
7599 struct extent_map *em2;
7601 em2 = btrfs_create_dio_extent(inode, start, len,
7602 orig_start, block_start,
7603 len, orig_block_len,
7605 btrfs_dec_nocow_writers(fs_info, block_start);
7606 if (type == BTRFS_ORDERED_PREALLOC) {
7607 free_extent_map(em);
7611 if (em2 && IS_ERR(em2)) {
7616 * For inode marked NODATACOW or extent marked PREALLOC,
7617 * use the existing or preallocated extent, so does not
7618 * need to adjust btrfs_space_info's bytes_may_use.
7620 btrfs_free_reserved_data_space_noquota(inode, start,
7626 /* this will cow the extent */
7627 len = bh_result->b_size;
7628 free_extent_map(em);
7629 *map = em = btrfs_new_extent_direct(inode, start, len);
7635 len = min(len, em->len - (start - em->start));
7638 bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7640 bh_result->b_size = len;
7641 bh_result->b_bdev = em->bdev;
7642 set_buffer_mapped(bh_result);
7644 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7645 set_buffer_new(bh_result);
7648 * Need to update the i_size under the extent lock so buffered
7649 * readers will get the updated i_size when we unlock.
7651 if (!dio_data->overwrite && start + len > i_size_read(inode))
7652 i_size_write(inode, start + len);
7654 WARN_ON(dio_data->reserve < len);
7655 dio_data->reserve -= len;
7656 dio_data->unsubmitted_oe_range_end = start + len;
7657 current->journal_info = dio_data;
7662 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7663 struct buffer_head *bh_result, int create)
7665 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7666 struct extent_map *em;
7667 struct extent_state *cached_state = NULL;
7668 struct btrfs_dio_data *dio_data = NULL;
7669 u64 start = iblock << inode->i_blkbits;
7670 u64 lockstart, lockend;
7671 u64 len = bh_result->b_size;
7672 int unlock_bits = EXTENT_LOCKED;
7676 unlock_bits |= EXTENT_DIRTY;
7678 len = min_t(u64, len, fs_info->sectorsize);
7681 lockend = start + len - 1;
7683 if (current->journal_info) {
7685 * Need to pull our outstanding extents and set journal_info to NULL so
7686 * that anything that needs to check if there's a transaction doesn't get
7689 dio_data = current->journal_info;
7690 current->journal_info = NULL;
7694 * If this errors out it's because we couldn't invalidate pagecache for
7695 * this range and we need to fallback to buffered.
7697 if (lock_extent_direct(inode, lockstart, lockend, &cached_state,
7703 em = btrfs_get_extent(BTRFS_I(inode), NULL, 0, start, len, 0);
7710 * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7711 * io. INLINE is special, and we could probably kludge it in here, but
7712 * it's still buffered so for safety lets just fall back to the generic
7715 * For COMPRESSED we _have_ to read the entire extent in so we can
7716 * decompress it, so there will be buffering required no matter what we
7717 * do, so go ahead and fallback to buffered.
7719 * We return -ENOTBLK because that's what makes DIO go ahead and go back
7720 * to buffered IO. Don't blame me, this is the price we pay for using
7723 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7724 em->block_start == EXTENT_MAP_INLINE) {
7725 free_extent_map(em);
7731 ret = btrfs_get_blocks_direct_write(&em, bh_result, inode,
7732 dio_data, start, len);
7736 /* clear and unlock the entire range */
7737 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7738 unlock_bits, 1, 0, &cached_state);
7740 ret = btrfs_get_blocks_direct_read(em, bh_result, inode,
7742 /* Can be negative only if we read from a hole */
7745 free_extent_map(em);
7749 * We need to unlock only the end area that we aren't using.
7750 * The rest is going to be unlocked by the endio routine.
7752 lockstart = start + bh_result->b_size;
7753 if (lockstart < lockend) {
7754 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7755 lockend, unlock_bits, 1, 0,
7758 free_extent_state(cached_state);
7762 free_extent_map(em);
7767 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7768 unlock_bits, 1, 0, &cached_state);
7771 current->journal_info = dio_data;
7775 static inline blk_status_t submit_dio_repair_bio(struct inode *inode,
7779 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7782 BUG_ON(bio_op(bio) == REQ_OP_WRITE);
7784 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DIO_REPAIR);
7788 ret = btrfs_map_bio(fs_info, bio, mirror_num, 0);
7793 static int btrfs_check_dio_repairable(struct inode *inode,
7794 struct bio *failed_bio,
7795 struct io_failure_record *failrec,
7798 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
7801 num_copies = btrfs_num_copies(fs_info, failrec->logical, failrec->len);
7802 if (num_copies == 1) {
7804 * we only have a single copy of the data, so don't bother with
7805 * all the retry and error correction code that follows. no
7806 * matter what the error is, it is very likely to persist.
7808 btrfs_debug(fs_info,
7809 "Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d",
7810 num_copies, failrec->this_mirror, failed_mirror);
7814 failrec->failed_mirror = failed_mirror;
7815 failrec->this_mirror++;
7816 if (failrec->this_mirror == failed_mirror)
7817 failrec->this_mirror++;
7819 if (failrec->this_mirror > num_copies) {
7820 btrfs_debug(fs_info,
7821 "Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d",
7822 num_copies, failrec->this_mirror, failed_mirror);
7829 static blk_status_t dio_read_error(struct inode *inode, struct bio *failed_bio,
7830 struct page *page, unsigned int pgoff,
7831 u64 start, u64 end, int failed_mirror,
7832 bio_end_io_t *repair_endio, void *repair_arg)
7834 struct io_failure_record *failrec;
7835 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7836 struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
7839 unsigned int read_mode = 0;
7842 blk_status_t status;
7843 struct bio_vec bvec;
7845 BUG_ON(bio_op(failed_bio) == REQ_OP_WRITE);
7847 ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7849 return errno_to_blk_status(ret);
7851 ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7854 free_io_failure(failure_tree, io_tree, failrec);
7855 return BLK_STS_IOERR;
7858 segs = bio_segments(failed_bio);
7859 bio_get_first_bvec(failed_bio, &bvec);
7861 (bvec.bv_len > btrfs_inode_sectorsize(inode)))
7862 read_mode |= REQ_FAILFAST_DEV;
7864 isector = start - btrfs_io_bio(failed_bio)->logical;
7865 isector >>= inode->i_sb->s_blocksize_bits;
7866 bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7867 pgoff, isector, repair_endio, repair_arg);
7868 bio_set_op_attrs(bio, REQ_OP_READ, read_mode);
7870 btrfs_debug(BTRFS_I(inode)->root->fs_info,
7871 "repair DIO read error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d",
7872 read_mode, failrec->this_mirror, failrec->in_validation);
7874 status = submit_dio_repair_bio(inode, bio, failrec->this_mirror);
7876 free_io_failure(failure_tree, io_tree, failrec);
7883 struct btrfs_retry_complete {
7884 struct completion done;
7885 struct inode *inode;
7890 static void btrfs_retry_endio_nocsum(struct bio *bio)
7892 struct btrfs_retry_complete *done = bio->bi_private;
7893 struct inode *inode = done->inode;
7894 struct bio_vec *bvec;
7895 struct extent_io_tree *io_tree, *failure_tree;
7901 ASSERT(bio->bi_vcnt == 1);
7902 io_tree = &BTRFS_I(inode)->io_tree;
7903 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7904 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(inode));
7907 ASSERT(!bio_flagged(bio, BIO_CLONED));
7908 bio_for_each_segment_all(bvec, bio, i)
7909 clean_io_failure(BTRFS_I(inode)->root->fs_info, failure_tree,
7910 io_tree, done->start, bvec->bv_page,
7911 btrfs_ino(BTRFS_I(inode)), 0);
7913 complete(&done->done);
7917 static blk_status_t __btrfs_correct_data_nocsum(struct inode *inode,
7918 struct btrfs_io_bio *io_bio)
7920 struct btrfs_fs_info *fs_info;
7921 struct bio_vec bvec;
7922 struct bvec_iter iter;
7923 struct btrfs_retry_complete done;
7929 blk_status_t err = BLK_STS_OK;
7931 fs_info = BTRFS_I(inode)->root->fs_info;
7932 sectorsize = fs_info->sectorsize;
7934 start = io_bio->logical;
7936 io_bio->bio.bi_iter = io_bio->iter;
7938 bio_for_each_segment(bvec, &io_bio->bio, iter) {
7939 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
7940 pgoff = bvec.bv_offset;
7942 next_block_or_try_again:
7945 init_completion(&done.done);
7947 ret = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
7948 pgoff, start, start + sectorsize - 1,
7950 btrfs_retry_endio_nocsum, &done);
7956 wait_for_completion_io(&done.done);
7958 if (!done.uptodate) {
7959 /* We might have another mirror, so try again */
7960 goto next_block_or_try_again;
7964 start += sectorsize;
7968 pgoff += sectorsize;
7969 ASSERT(pgoff < PAGE_SIZE);
7970 goto next_block_or_try_again;
7977 static void btrfs_retry_endio(struct bio *bio)
7979 struct btrfs_retry_complete *done = bio->bi_private;
7980 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7981 struct extent_io_tree *io_tree, *failure_tree;
7982 struct inode *inode = done->inode;
7983 struct bio_vec *bvec;
7993 ASSERT(bio->bi_vcnt == 1);
7994 ASSERT(bio_first_bvec_all(bio)->bv_len == btrfs_inode_sectorsize(done->inode));
7996 io_tree = &BTRFS_I(inode)->io_tree;
7997 failure_tree = &BTRFS_I(inode)->io_failure_tree;
7999 ASSERT(!bio_flagged(bio, BIO_CLONED));
8000 bio_for_each_segment_all(bvec, bio, i) {
8001 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
8002 bvec->bv_offset, done->start,
8005 clean_io_failure(BTRFS_I(inode)->root->fs_info,
8006 failure_tree, io_tree, done->start,
8008 btrfs_ino(BTRFS_I(inode)),
8014 done->uptodate = uptodate;
8016 complete(&done->done);
8020 static blk_status_t __btrfs_subio_endio_read(struct inode *inode,
8021 struct btrfs_io_bio *io_bio, blk_status_t err)
8023 struct btrfs_fs_info *fs_info;
8024 struct bio_vec bvec;
8025 struct bvec_iter iter;
8026 struct btrfs_retry_complete done;
8033 bool uptodate = (err == 0);
8035 blk_status_t status;
8037 fs_info = BTRFS_I(inode)->root->fs_info;
8038 sectorsize = fs_info->sectorsize;
8041 start = io_bio->logical;
8043 io_bio->bio.bi_iter = io_bio->iter;
8045 bio_for_each_segment(bvec, &io_bio->bio, iter) {
8046 nr_sectors = BTRFS_BYTES_TO_BLKS(fs_info, bvec.bv_len);
8048 pgoff = bvec.bv_offset;
8051 csum_pos = BTRFS_BYTES_TO_BLKS(fs_info, offset);
8052 ret = __readpage_endio_check(inode, io_bio, csum_pos,
8053 bvec.bv_page, pgoff, start, sectorsize);
8060 init_completion(&done.done);
8062 status = dio_read_error(inode, &io_bio->bio, bvec.bv_page,
8063 pgoff, start, start + sectorsize - 1,
8064 io_bio->mirror_num, btrfs_retry_endio,
8071 wait_for_completion_io(&done.done);
8073 if (!done.uptodate) {
8074 /* We might have another mirror, so try again */
8078 offset += sectorsize;
8079 start += sectorsize;
8085 pgoff += sectorsize;
8086 ASSERT(pgoff < PAGE_SIZE);
8094 static blk_status_t btrfs_subio_endio_read(struct inode *inode,
8095 struct btrfs_io_bio *io_bio, blk_status_t err)
8097 bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
8101 return __btrfs_correct_data_nocsum(inode, io_bio);
8105 return __btrfs_subio_endio_read(inode, io_bio, err);
8109 static void btrfs_endio_direct_read(struct bio *bio)
8111 struct btrfs_dio_private *dip = bio->bi_private;
8112 struct inode *inode = dip->inode;
8113 struct bio *dio_bio;
8114 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8115 blk_status_t err = bio->bi_status;
8117 if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
8118 err = btrfs_subio_endio_read(inode, io_bio, err);
8120 unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
8121 dip->logical_offset + dip->bytes - 1);
8122 dio_bio = dip->dio_bio;
8126 dio_bio->bi_status = err;
8127 dio_end_io(dio_bio);
8130 io_bio->end_io(io_bio, blk_status_to_errno(err));
8134 static void __endio_write_update_ordered(struct inode *inode,
8135 const u64 offset, const u64 bytes,
8136 const bool uptodate)
8138 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8139 struct btrfs_ordered_extent *ordered = NULL;
8140 struct btrfs_workqueue *wq;
8141 btrfs_work_func_t func;
8142 u64 ordered_offset = offset;
8143 u64 ordered_bytes = bytes;
8146 if (btrfs_is_free_space_inode(BTRFS_I(inode))) {
8147 wq = fs_info->endio_freespace_worker;
8148 func = btrfs_freespace_write_helper;
8150 wq = fs_info->endio_write_workers;
8151 func = btrfs_endio_write_helper;
8154 while (ordered_offset < offset + bytes) {
8155 last_offset = ordered_offset;
8156 if (btrfs_dec_test_first_ordered_pending(inode, &ordered,
8160 btrfs_init_work(&ordered->work, func,
8163 btrfs_queue_work(wq, &ordered->work);
8166 * If btrfs_dec_test_ordered_pending does not find any ordered
8167 * extent in the range, we can exit.
8169 if (ordered_offset == last_offset)
8172 * Our bio might span multiple ordered extents. In this case
8173 * we keep goin until we have accounted the whole dio.
8175 if (ordered_offset < offset + bytes) {
8176 ordered_bytes = offset + bytes - ordered_offset;
8182 static void btrfs_endio_direct_write(struct bio *bio)
8184 struct btrfs_dio_private *dip = bio->bi_private;
8185 struct bio *dio_bio = dip->dio_bio;
8187 __endio_write_update_ordered(dip->inode, dip->logical_offset,
8188 dip->bytes, !bio->bi_status);
8192 dio_bio->bi_status = bio->bi_status;
8193 dio_end_io(dio_bio);
8197 static blk_status_t btrfs_submit_bio_start_direct_io(void *private_data,
8198 struct bio *bio, u64 offset)
8200 struct inode *inode = private_data;
8202 ret = btrfs_csum_one_bio(inode, bio, offset, 1);
8203 BUG_ON(ret); /* -ENOMEM */
8207 static void btrfs_end_dio_bio(struct bio *bio)
8209 struct btrfs_dio_private *dip = bio->bi_private;
8210 blk_status_t err = bio->bi_status;
8213 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
8214 "direct IO failed ino %llu rw %d,%u sector %#Lx len %u err no %d",
8215 btrfs_ino(BTRFS_I(dip->inode)), bio_op(bio),
8217 (unsigned long long)bio->bi_iter.bi_sector,
8218 bio->bi_iter.bi_size, err);
8220 if (dip->subio_endio)
8221 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
8225 * We want to perceive the errors flag being set before
8226 * decrementing the reference count. We don't need a barrier
8227 * since atomic operations with a return value are fully
8228 * ordered as per atomic_t.txt
8233 /* if there are more bios still pending for this dio, just exit */
8234 if (!atomic_dec_and_test(&dip->pending_bios))
8238 bio_io_error(dip->orig_bio);
8240 dip->dio_bio->bi_status = BLK_STS_OK;
8241 bio_endio(dip->orig_bio);
8247 static inline blk_status_t btrfs_lookup_and_bind_dio_csum(struct inode *inode,
8248 struct btrfs_dio_private *dip,
8252 struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
8253 struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
8257 * We load all the csum data we need when we submit
8258 * the first bio to reduce the csum tree search and
8261 if (dip->logical_offset == file_offset) {
8262 ret = btrfs_lookup_bio_sums_dio(inode, dip->orig_bio,
8268 if (bio == dip->orig_bio)
8271 file_offset -= dip->logical_offset;
8272 file_offset >>= inode->i_sb->s_blocksize_bits;
8273 io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
8278 static inline blk_status_t btrfs_submit_dio_bio(struct bio *bio,
8279 struct inode *inode, u64 file_offset, int async_submit)
8281 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8282 struct btrfs_dio_private *dip = bio->bi_private;
8283 bool write = bio_op(bio) == REQ_OP_WRITE;
8286 /* Check btrfs_submit_bio_hook() for rules about async submit. */
8288 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
8291 ret = btrfs_bio_wq_end_io(fs_info, bio, BTRFS_WQ_ENDIO_DATA);
8296 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
8299 if (write && async_submit) {
8300 ret = btrfs_wq_submit_bio(fs_info, bio, 0, 0,
8302 btrfs_submit_bio_start_direct_io,
8303 btrfs_submit_bio_done);
8307 * If we aren't doing async submit, calculate the csum of the
8310 ret = btrfs_csum_one_bio(inode, bio, file_offset, 1);
8314 ret = btrfs_lookup_and_bind_dio_csum(inode, dip, bio,
8320 ret = btrfs_map_bio(fs_info, bio, 0, 0);
8325 static int btrfs_submit_direct_hook(struct btrfs_dio_private *dip)
8327 struct inode *inode = dip->inode;
8328 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8330 struct bio *orig_bio = dip->orig_bio;
8331 u64 start_sector = orig_bio->bi_iter.bi_sector;
8332 u64 file_offset = dip->logical_offset;
8334 int async_submit = 0;
8336 int clone_offset = 0;
8339 blk_status_t status;
8341 map_length = orig_bio->bi_iter.bi_size;
8342 submit_len = map_length;
8343 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio), start_sector << 9,
8344 &map_length, NULL, 0);
8348 if (map_length >= submit_len) {
8350 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
8354 /* async crcs make it difficult to collect full stripe writes. */
8355 if (btrfs_data_alloc_profile(fs_info) & BTRFS_BLOCK_GROUP_RAID56_MASK)
8361 ASSERT(map_length <= INT_MAX);
8362 atomic_inc(&dip->pending_bios);
8364 clone_len = min_t(int, submit_len, map_length);
8367 * This will never fail as it's passing GPF_NOFS and
8368 * the allocation is backed by btrfs_bioset.
8370 bio = btrfs_bio_clone_partial(orig_bio, clone_offset,
8372 bio->bi_private = dip;
8373 bio->bi_end_io = btrfs_end_dio_bio;
8374 btrfs_io_bio(bio)->logical = file_offset;
8376 ASSERT(submit_len >= clone_len);
8377 submit_len -= clone_len;
8378 if (submit_len == 0)
8382 * Increase the count before we submit the bio so we know
8383 * the end IO handler won't happen before we increase the
8384 * count. Otherwise, the dip might get freed before we're
8385 * done setting it up.
8387 atomic_inc(&dip->pending_bios);
8389 status = btrfs_submit_dio_bio(bio, inode, file_offset,
8393 atomic_dec(&dip->pending_bios);
8397 clone_offset += clone_len;
8398 start_sector += clone_len >> 9;
8399 file_offset += clone_len;
8401 map_length = submit_len;
8402 ret = btrfs_map_block(fs_info, btrfs_op(orig_bio),
8403 start_sector << 9, &map_length, NULL, 0);
8406 } while (submit_len > 0);
8409 status = btrfs_submit_dio_bio(bio, inode, file_offset, async_submit);
8417 * Before atomic variable goto zero, we must make sure dip->errors is
8418 * perceived to be set. This ordering is ensured by the fact that an
8419 * atomic operations with a return value are fully ordered as per
8422 if (atomic_dec_and_test(&dip->pending_bios))
8423 bio_io_error(dip->orig_bio);
8425 /* bio_end_io() will handle error, so we needn't return it */
8429 static void btrfs_submit_direct(struct bio *dio_bio, struct inode *inode,
8432 struct btrfs_dio_private *dip = NULL;
8433 struct bio *bio = NULL;
8434 struct btrfs_io_bio *io_bio;
8435 bool write = (bio_op(dio_bio) == REQ_OP_WRITE);
8438 bio = btrfs_bio_clone(dio_bio);
8440 dip = kzalloc(sizeof(*dip), GFP_NOFS);
8446 dip->private = dio_bio->bi_private;
8448 dip->logical_offset = file_offset;
8449 dip->bytes = dio_bio->bi_iter.bi_size;
8450 dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
8451 bio->bi_private = dip;
8452 dip->orig_bio = bio;
8453 dip->dio_bio = dio_bio;
8454 atomic_set(&dip->pending_bios, 0);
8455 io_bio = btrfs_io_bio(bio);
8456 io_bio->logical = file_offset;
8459 bio->bi_end_io = btrfs_endio_direct_write;
8461 bio->bi_end_io = btrfs_endio_direct_read;
8462 dip->subio_endio = btrfs_subio_endio_read;
8466 * Reset the range for unsubmitted ordered extents (to a 0 length range)
8467 * even if we fail to submit a bio, because in such case we do the
8468 * corresponding error handling below and it must not be done a second
8469 * time by btrfs_direct_IO().
8472 struct btrfs_dio_data *dio_data = current->journal_info;
8474 dio_data->unsubmitted_oe_range_end = dip->logical_offset +
8476 dio_data->unsubmitted_oe_range_start =
8477 dio_data->unsubmitted_oe_range_end;
8480 ret = btrfs_submit_direct_hook(dip);
8485 io_bio->end_io(io_bio, ret);
8489 * If we arrived here it means either we failed to submit the dip
8490 * or we either failed to clone the dio_bio or failed to allocate the
8491 * dip. If we cloned the dio_bio and allocated the dip, we can just
8492 * call bio_endio against our io_bio so that we get proper resource
8493 * cleanup if we fail to submit the dip, otherwise, we must do the
8494 * same as btrfs_endio_direct_[write|read] because we can't call these
8495 * callbacks - they require an allocated dip and a clone of dio_bio.
8500 * The end io callbacks free our dip, do the final put on bio
8501 * and all the cleanup and final put for dio_bio (through
8508 __endio_write_update_ordered(inode,
8510 dio_bio->bi_iter.bi_size,
8513 unlock_extent(&BTRFS_I(inode)->io_tree, file_offset,
8514 file_offset + dio_bio->bi_iter.bi_size - 1);
8516 dio_bio->bi_status = BLK_STS_IOERR;
8518 * Releases and cleans up our dio_bio, no need to bio_put()
8519 * nor bio_endio()/bio_io_error() against dio_bio.
8521 dio_end_io(dio_bio);
8528 static ssize_t check_direct_IO(struct btrfs_fs_info *fs_info,
8529 const struct iov_iter *iter, loff_t offset)
8533 unsigned int blocksize_mask = fs_info->sectorsize - 1;
8534 ssize_t retval = -EINVAL;
8536 if (offset & blocksize_mask)
8539 if (iov_iter_alignment(iter) & blocksize_mask)
8542 /* If this is a write we don't need to check anymore */
8543 if (iov_iter_rw(iter) != READ || !iter_is_iovec(iter))
8546 * Check to make sure we don't have duplicate iov_base's in this
8547 * iovec, if so return EINVAL, otherwise we'll get csum errors
8548 * when reading back.
8550 for (seg = 0; seg < iter->nr_segs; seg++) {
8551 for (i = seg + 1; i < iter->nr_segs; i++) {
8552 if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8561 static ssize_t btrfs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
8563 struct file *file = iocb->ki_filp;
8564 struct inode *inode = file->f_mapping->host;
8565 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8566 struct btrfs_dio_data dio_data = { 0 };
8567 struct extent_changeset *data_reserved = NULL;
8568 loff_t offset = iocb->ki_pos;
8572 bool relock = false;
8575 if (check_direct_IO(fs_info, iter, offset))
8578 inode_dio_begin(inode);
8581 * The generic stuff only does filemap_write_and_wait_range, which
8582 * isn't enough if we've written compressed pages to this area, so
8583 * we need to flush the dirty pages again to make absolutely sure
8584 * that any outstanding dirty pages are on disk.
8586 count = iov_iter_count(iter);
8587 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8588 &BTRFS_I(inode)->runtime_flags))
8589 filemap_fdatawrite_range(inode->i_mapping, offset,
8590 offset + count - 1);
8592 if (iov_iter_rw(iter) == WRITE) {
8594 * If the write DIO is beyond the EOF, we need update
8595 * the isize, but it is protected by i_mutex. So we can
8596 * not unlock the i_mutex at this case.
8598 if (offset + count <= inode->i_size) {
8599 dio_data.overwrite = 1;
8600 inode_unlock(inode);
8602 } else if (iocb->ki_flags & IOCB_NOWAIT) {
8606 ret = btrfs_delalloc_reserve_space(inode, &data_reserved,
8612 * We need to know how many extents we reserved so that we can
8613 * do the accounting properly if we go over the number we
8614 * originally calculated. Abuse current->journal_info for this.
8616 dio_data.reserve = round_up(count,
8617 fs_info->sectorsize);
8618 dio_data.unsubmitted_oe_range_start = (u64)offset;
8619 dio_data.unsubmitted_oe_range_end = (u64)offset;
8620 current->journal_info = &dio_data;
8621 down_read(&BTRFS_I(inode)->dio_sem);
8622 } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8623 &BTRFS_I(inode)->runtime_flags)) {
8624 inode_dio_end(inode);
8625 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8629 ret = __blockdev_direct_IO(iocb, inode,
8630 fs_info->fs_devices->latest_bdev,
8631 iter, btrfs_get_blocks_direct, NULL,
8632 btrfs_submit_direct, flags);
8633 if (iov_iter_rw(iter) == WRITE) {
8634 up_read(&BTRFS_I(inode)->dio_sem);
8635 current->journal_info = NULL;
8636 if (ret < 0 && ret != -EIOCBQUEUED) {
8637 if (dio_data.reserve)
8638 btrfs_delalloc_release_space(inode, data_reserved,
8639 offset, dio_data.reserve, true);
8641 * On error we might have left some ordered extents
8642 * without submitting corresponding bios for them, so
8643 * cleanup them up to avoid other tasks getting them
8644 * and waiting for them to complete forever.
8646 if (dio_data.unsubmitted_oe_range_start <
8647 dio_data.unsubmitted_oe_range_end)
8648 __endio_write_update_ordered(inode,
8649 dio_data.unsubmitted_oe_range_start,
8650 dio_data.unsubmitted_oe_range_end -
8651 dio_data.unsubmitted_oe_range_start,
8653 } else if (ret >= 0 && (size_t)ret < count)
8654 btrfs_delalloc_release_space(inode, data_reserved,
8655 offset, count - (size_t)ret, true);
8656 btrfs_delalloc_release_extents(BTRFS_I(inode), count, false);
8660 inode_dio_end(inode);
8664 extent_changeset_free(data_reserved);
8668 #define BTRFS_FIEMAP_FLAGS (FIEMAP_FLAG_SYNC)
8670 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8671 __u64 start, __u64 len)
8675 ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8679 return extent_fiemap(inode, fieinfo, start, len);
8682 int btrfs_readpage(struct file *file, struct page *page)
8684 struct extent_io_tree *tree;
8685 tree = &BTRFS_I(page->mapping->host)->io_tree;
8686 return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8689 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8691 struct inode *inode = page->mapping->host;
8694 if (current->flags & PF_MEMALLOC) {
8695 redirty_page_for_writepage(wbc, page);
8701 * If we are under memory pressure we will call this directly from the
8702 * VM, we need to make sure we have the inode referenced for the ordered
8703 * extent. If not just return like we didn't do anything.
8705 if (!igrab(inode)) {
8706 redirty_page_for_writepage(wbc, page);
8707 return AOP_WRITEPAGE_ACTIVATE;
8709 ret = extent_write_full_page(page, wbc);
8710 btrfs_add_delayed_iput(inode);
8714 static int btrfs_writepages(struct address_space *mapping,
8715 struct writeback_control *wbc)
8717 return extent_writepages(mapping, wbc);
8721 btrfs_readpages(struct file *file, struct address_space *mapping,
8722 struct list_head *pages, unsigned nr_pages)
8724 return extent_readpages(mapping, pages, nr_pages);
8727 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8729 int ret = try_release_extent_mapping(page, gfp_flags);
8731 ClearPagePrivate(page);
8732 set_page_private(page, 0);
8738 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8740 if (PageWriteback(page) || PageDirty(page))
8742 return __btrfs_releasepage(page, gfp_flags);
8745 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8746 unsigned int length)
8748 struct inode *inode = page->mapping->host;
8749 struct extent_io_tree *tree;
8750 struct btrfs_ordered_extent *ordered;
8751 struct extent_state *cached_state = NULL;
8752 u64 page_start = page_offset(page);
8753 u64 page_end = page_start + PAGE_SIZE - 1;
8756 int inode_evicting = inode->i_state & I_FREEING;
8759 * we have the page locked, so new writeback can't start,
8760 * and the dirty bit won't be cleared while we are here.
8762 * Wait for IO on this page so that we can safely clear
8763 * the PagePrivate2 bit and do ordered accounting
8765 wait_on_page_writeback(page);
8767 tree = &BTRFS_I(inode)->io_tree;
8769 btrfs_releasepage(page, GFP_NOFS);
8773 if (!inode_evicting)
8774 lock_extent_bits(tree, page_start, page_end, &cached_state);
8777 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), start,
8778 page_end - start + 1);
8780 end = min(page_end, ordered->file_offset + ordered->len - 1);
8782 * IO on this page will never be started, so we need
8783 * to account for any ordered extents now
8785 if (!inode_evicting)
8786 clear_extent_bit(tree, start, end,
8787 EXTENT_DIRTY | EXTENT_DELALLOC |
8788 EXTENT_DELALLOC_NEW |
8789 EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8790 EXTENT_DEFRAG, 1, 0, &cached_state);
8792 * whoever cleared the private bit is responsible
8793 * for the finish_ordered_io
8795 if (TestClearPagePrivate2(page)) {
8796 struct btrfs_ordered_inode_tree *tree;
8799 tree = &BTRFS_I(inode)->ordered_tree;
8801 spin_lock_irq(&tree->lock);
8802 set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8803 new_len = start - ordered->file_offset;
8804 if (new_len < ordered->truncated_len)
8805 ordered->truncated_len = new_len;
8806 spin_unlock_irq(&tree->lock);
8808 if (btrfs_dec_test_ordered_pending(inode, &ordered,
8810 end - start + 1, 1))
8811 btrfs_finish_ordered_io(ordered);
8813 btrfs_put_ordered_extent(ordered);
8814 if (!inode_evicting) {
8815 cached_state = NULL;
8816 lock_extent_bits(tree, start, end,
8821 if (start < page_end)
8826 * Qgroup reserved space handler
8827 * Page here will be either
8828 * 1) Already written to disk
8829 * In this case, its reserved space is released from data rsv map
8830 * and will be freed by delayed_ref handler finally.
8831 * So even we call qgroup_free_data(), it won't decrease reserved
8833 * 2) Not written to disk
8834 * This means the reserved space should be freed here. However,
8835 * if a truncate invalidates the page (by clearing PageDirty)
8836 * and the page is accounted for while allocating extent
8837 * in btrfs_check_data_free_space() we let delayed_ref to
8838 * free the entire extent.
8840 if (PageDirty(page))
8841 btrfs_qgroup_free_data(inode, NULL, page_start, PAGE_SIZE);
8842 if (!inode_evicting) {
8843 clear_extent_bit(tree, page_start, page_end,
8844 EXTENT_LOCKED | EXTENT_DIRTY |
8845 EXTENT_DELALLOC | EXTENT_DELALLOC_NEW |
8846 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
8849 __btrfs_releasepage(page, GFP_NOFS);
8852 ClearPageChecked(page);
8853 if (PagePrivate(page)) {
8854 ClearPagePrivate(page);
8855 set_page_private(page, 0);
8861 * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8862 * called from a page fault handler when a page is first dirtied. Hence we must
8863 * be careful to check for EOF conditions here. We set the page up correctly
8864 * for a written page which means we get ENOSPC checking when writing into
8865 * holes and correct delalloc and unwritten extent mapping on filesystems that
8866 * support these features.
8868 * We are not allowed to take the i_mutex here so we have to play games to
8869 * protect against truncate races as the page could now be beyond EOF. Because
8870 * truncate_setsize() writes the inode size before removing pages, once we have
8871 * the page lock we can determine safely if the page is beyond EOF. If it is not
8872 * beyond EOF, then the page is guaranteed safe against truncation until we
8875 vm_fault_t btrfs_page_mkwrite(struct vm_fault *vmf)
8877 struct page *page = vmf->page;
8878 struct inode *inode = file_inode(vmf->vma->vm_file);
8879 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
8880 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8881 struct btrfs_ordered_extent *ordered;
8882 struct extent_state *cached_state = NULL;
8883 struct extent_changeset *data_reserved = NULL;
8885 unsigned long zero_start;
8895 reserved_space = PAGE_SIZE;
8897 sb_start_pagefault(inode->i_sb);
8898 page_start = page_offset(page);
8899 page_end = page_start + PAGE_SIZE - 1;
8903 * Reserving delalloc space after obtaining the page lock can lead to
8904 * deadlock. For example, if a dirty page is locked by this function
8905 * and the call to btrfs_delalloc_reserve_space() ends up triggering
8906 * dirty page write out, then the btrfs_writepage() function could
8907 * end up waiting indefinitely to get a lock on the page currently
8908 * being processed by btrfs_page_mkwrite() function.
8910 ret2 = btrfs_delalloc_reserve_space(inode, &data_reserved, page_start,
8913 ret2 = file_update_time(vmf->vma->vm_file);
8917 ret = vmf_error(ret2);
8923 ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8926 size = i_size_read(inode);
8928 if ((page->mapping != inode->i_mapping) ||
8929 (page_start >= size)) {
8930 /* page got truncated out from underneath us */
8933 wait_on_page_writeback(page);
8935 lock_extent_bits(io_tree, page_start, page_end, &cached_state);
8936 set_page_extent_mapped(page);
8939 * we can't set the delalloc bits if there are pending ordered
8940 * extents. Drop our locks and wait for them to finish
8942 ordered = btrfs_lookup_ordered_range(BTRFS_I(inode), page_start,
8945 unlock_extent_cached(io_tree, page_start, page_end,
8948 btrfs_start_ordered_extent(inode, ordered, 1);
8949 btrfs_put_ordered_extent(ordered);
8953 if (page->index == ((size - 1) >> PAGE_SHIFT)) {
8954 reserved_space = round_up(size - page_start,
8955 fs_info->sectorsize);
8956 if (reserved_space < PAGE_SIZE) {
8957 end = page_start + reserved_space - 1;
8958 btrfs_delalloc_release_space(inode, data_reserved,
8959 page_start, PAGE_SIZE - reserved_space,
8965 * page_mkwrite gets called when the page is firstly dirtied after it's
8966 * faulted in, but write(2) could also dirty a page and set delalloc
8967 * bits, thus in this case for space account reason, we still need to
8968 * clear any delalloc bits within this page range since we have to
8969 * reserve data&meta space before lock_page() (see above comments).
8971 clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, end,
8972 EXTENT_DIRTY | EXTENT_DELALLOC |
8973 EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8974 0, 0, &cached_state);
8976 ret2 = btrfs_set_extent_delalloc(inode, page_start, end, 0,
8979 unlock_extent_cached(io_tree, page_start, page_end,
8981 ret = VM_FAULT_SIGBUS;
8986 /* page is wholly or partially inside EOF */
8987 if (page_start + PAGE_SIZE > size)
8988 zero_start = size & ~PAGE_MASK;
8990 zero_start = PAGE_SIZE;
8992 if (zero_start != PAGE_SIZE) {
8994 memset(kaddr + zero_start, 0, PAGE_SIZE - zero_start);
8995 flush_dcache_page(page);
8998 ClearPageChecked(page);
8999 set_page_dirty(page);
9000 SetPageUptodate(page);
9002 BTRFS_I(inode)->last_trans = fs_info->generation;
9003 BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
9004 BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
9006 unlock_extent_cached(io_tree, page_start, page_end, &cached_state);
9010 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, true);
9011 sb_end_pagefault(inode->i_sb);
9012 extent_changeset_free(data_reserved);
9013 return VM_FAULT_LOCKED;
9017 btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, (ret != 0));
9018 btrfs_delalloc_release_space(inode, data_reserved, page_start,
9019 reserved_space, (ret != 0));
9021 sb_end_pagefault(inode->i_sb);
9022 extent_changeset_free(data_reserved);
9026 static int btrfs_truncate(struct inode *inode, bool skip_writeback)
9028 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9029 struct btrfs_root *root = BTRFS_I(inode)->root;
9030 struct btrfs_block_rsv *rsv;
9032 struct btrfs_trans_handle *trans;
9033 u64 mask = fs_info->sectorsize - 1;
9034 u64 min_size = btrfs_calc_trunc_metadata_size(fs_info, 1);
9036 if (!skip_writeback) {
9037 ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
9044 * Yes ladies and gentlemen, this is indeed ugly. We have a couple of
9045 * things going on here:
9047 * 1) We need to reserve space to update our inode.
9049 * 2) We need to have something to cache all the space that is going to
9050 * be free'd up by the truncate operation, but also have some slack
9051 * space reserved in case it uses space during the truncate (thank you
9052 * very much snapshotting).
9054 * And we need these to be separate. The fact is we can use a lot of
9055 * space doing the truncate, and we have no earthly idea how much space
9056 * we will use, so we need the truncate reservation to be separate so it
9057 * doesn't end up using space reserved for updating the inode. We also
9058 * need to be able to stop the transaction and start a new one, which
9059 * means we need to be able to update the inode several times, and we
9060 * have no idea of knowing how many times that will be, so we can't just
9061 * reserve 1 item for the entirety of the operation, so that has to be
9062 * done separately as well.
9064 * So that leaves us with
9066 * 1) rsv - for the truncate reservation, which we will steal from the
9067 * transaction reservation.
9068 * 2) fs_info->trans_block_rsv - this will have 1 items worth left for
9069 * updating the inode.
9071 rsv = btrfs_alloc_block_rsv(fs_info, BTRFS_BLOCK_RSV_TEMP);
9074 rsv->size = min_size;
9078 * 1 for the truncate slack space
9079 * 1 for updating the inode.
9081 trans = btrfs_start_transaction(root, 2);
9082 if (IS_ERR(trans)) {
9083 ret = PTR_ERR(trans);
9087 /* Migrate the slack space for the truncate to our reserve */
9088 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv, rsv,
9093 * So if we truncate and then write and fsync we normally would just
9094 * write the extents that changed, which is a problem if we need to
9095 * first truncate that entire inode. So set this flag so we write out
9096 * all of the extents in the inode to the sync log so we're completely
9099 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
9100 trans->block_rsv = rsv;
9103 ret = btrfs_truncate_inode_items(trans, root, inode,
9105 BTRFS_EXTENT_DATA_KEY);
9106 trans->block_rsv = &fs_info->trans_block_rsv;
9107 if (ret != -ENOSPC && ret != -EAGAIN)
9110 ret = btrfs_update_inode(trans, root, inode);
9114 btrfs_end_transaction(trans);
9115 btrfs_btree_balance_dirty(fs_info);
9117 trans = btrfs_start_transaction(root, 2);
9118 if (IS_ERR(trans)) {
9119 ret = PTR_ERR(trans);
9124 btrfs_block_rsv_release(fs_info, rsv, -1);
9125 ret = btrfs_block_rsv_migrate(&fs_info->trans_block_rsv,
9127 BUG_ON(ret); /* shouldn't happen */
9128 trans->block_rsv = rsv;
9132 * We can't call btrfs_truncate_block inside a trans handle as we could
9133 * deadlock with freeze, if we got NEED_TRUNCATE_BLOCK then we know
9134 * we've truncated everything except the last little bit, and can do
9135 * btrfs_truncate_block and then update the disk_i_size.
9137 if (ret == NEED_TRUNCATE_BLOCK) {
9138 btrfs_end_transaction(trans);
9139 btrfs_btree_balance_dirty(fs_info);
9141 ret = btrfs_truncate_block(inode, inode->i_size, 0, 0);
9144 trans = btrfs_start_transaction(root, 1);
9145 if (IS_ERR(trans)) {
9146 ret = PTR_ERR(trans);
9149 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
9155 trans->block_rsv = &fs_info->trans_block_rsv;
9156 ret2 = btrfs_update_inode(trans, root, inode);
9160 ret2 = btrfs_end_transaction(trans);
9163 btrfs_btree_balance_dirty(fs_info);
9166 btrfs_free_block_rsv(fs_info, rsv);
9172 * create a new subvolume directory/inode (helper for the ioctl).
9174 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
9175 struct btrfs_root *new_root,
9176 struct btrfs_root *parent_root,
9179 struct inode *inode;
9183 inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
9184 new_dirid, new_dirid,
9185 S_IFDIR | (~current_umask() & S_IRWXUGO),
9188 return PTR_ERR(inode);
9189 inode->i_op = &btrfs_dir_inode_operations;
9190 inode->i_fop = &btrfs_dir_file_operations;
9192 set_nlink(inode, 1);
9193 btrfs_i_size_write(BTRFS_I(inode), 0);
9194 unlock_new_inode(inode);
9196 err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
9198 btrfs_err(new_root->fs_info,
9199 "error inheriting subvolume %llu properties: %d",
9200 new_root->root_key.objectid, err);
9202 err = btrfs_update_inode(trans, new_root, inode);
9208 struct inode *btrfs_alloc_inode(struct super_block *sb)
9210 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
9211 struct btrfs_inode *ei;
9212 struct inode *inode;
9214 ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_KERNEL);
9221 ei->last_sub_trans = 0;
9222 ei->logged_trans = 0;
9223 ei->delalloc_bytes = 0;
9224 ei->new_delalloc_bytes = 0;
9225 ei->defrag_bytes = 0;
9226 ei->disk_i_size = 0;
9229 ei->index_cnt = (u64)-1;
9231 ei->last_unlink_trans = 0;
9232 ei->last_log_commit = 0;
9234 spin_lock_init(&ei->lock);
9235 ei->outstanding_extents = 0;
9236 if (sb->s_magic != BTRFS_TEST_MAGIC)
9237 btrfs_init_metadata_block_rsv(fs_info, &ei->block_rsv,
9238 BTRFS_BLOCK_RSV_DELALLOC);
9239 ei->runtime_flags = 0;
9240 ei->prop_compress = BTRFS_COMPRESS_NONE;
9241 ei->defrag_compress = BTRFS_COMPRESS_NONE;
9243 ei->delayed_node = NULL;
9245 ei->i_otime.tv_sec = 0;
9246 ei->i_otime.tv_nsec = 0;
9248 inode = &ei->vfs_inode;
9249 extent_map_tree_init(&ei->extent_tree);
9250 extent_io_tree_init(&ei->io_tree, inode);
9251 extent_io_tree_init(&ei->io_failure_tree, inode);
9252 ei->io_tree.track_uptodate = 1;
9253 ei->io_failure_tree.track_uptodate = 1;
9254 atomic_set(&ei->sync_writers, 0);
9255 mutex_init(&ei->log_mutex);
9256 mutex_init(&ei->delalloc_mutex);
9257 btrfs_ordered_inode_tree_init(&ei->ordered_tree);
9258 INIT_LIST_HEAD(&ei->delalloc_inodes);
9259 INIT_LIST_HEAD(&ei->delayed_iput);
9260 RB_CLEAR_NODE(&ei->rb_node);
9261 init_rwsem(&ei->dio_sem);
9266 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
9267 void btrfs_test_destroy_inode(struct inode *inode)
9269 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9270 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9274 static void btrfs_i_callback(struct rcu_head *head)
9276 struct inode *inode = container_of(head, struct inode, i_rcu);
9277 kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
9280 void btrfs_destroy_inode(struct inode *inode)
9282 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
9283 struct btrfs_ordered_extent *ordered;
9284 struct btrfs_root *root = BTRFS_I(inode)->root;
9286 WARN_ON(!hlist_empty(&inode->i_dentry));
9287 WARN_ON(inode->i_data.nrpages);
9288 WARN_ON(BTRFS_I(inode)->block_rsv.reserved);
9289 WARN_ON(BTRFS_I(inode)->block_rsv.size);
9290 WARN_ON(BTRFS_I(inode)->outstanding_extents);
9291 WARN_ON(BTRFS_I(inode)->delalloc_bytes);
9292 WARN_ON(BTRFS_I(inode)->new_delalloc_bytes);
9293 WARN_ON(BTRFS_I(inode)->csum_bytes);
9294 WARN_ON(BTRFS_I(inode)->defrag_bytes);
9297 * This can happen where we create an inode, but somebody else also
9298 * created the same inode and we need to destroy the one we already
9305 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
9310 "found ordered extent %llu %llu on inode cleanup",
9311 ordered->file_offset, ordered->len);
9312 btrfs_remove_ordered_extent(inode, ordered);
9313 btrfs_put_ordered_extent(ordered);
9314 btrfs_put_ordered_extent(ordered);
9317 btrfs_qgroup_check_reserved_leak(inode);
9318 inode_tree_del(inode);
9319 btrfs_drop_extent_cache(BTRFS_I(inode), 0, (u64)-1, 0);
9321 call_rcu(&inode->i_rcu, btrfs_i_callback);
9324 int btrfs_drop_inode(struct inode *inode)
9326 struct btrfs_root *root = BTRFS_I(inode)->root;
9331 /* the snap/subvol tree is on deleting */
9332 if (btrfs_root_refs(&root->root_item) == 0)
9335 return generic_drop_inode(inode);
9338 static void init_once(void *foo)
9340 struct btrfs_inode *ei = (struct btrfs_inode *) foo;
9342 inode_init_once(&ei->vfs_inode);
9345 void __cold btrfs_destroy_cachep(void)
9348 * Make sure all delayed rcu free inodes are flushed before we
9352 kmem_cache_destroy(btrfs_inode_cachep);
9353 kmem_cache_destroy(btrfs_trans_handle_cachep);
9354 kmem_cache_destroy(btrfs_path_cachep);
9355 kmem_cache_destroy(btrfs_free_space_cachep);
9358 int __init btrfs_init_cachep(void)
9360 btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
9361 sizeof(struct btrfs_inode), 0,
9362 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD | SLAB_ACCOUNT,
9364 if (!btrfs_inode_cachep)
9367 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
9368 sizeof(struct btrfs_trans_handle), 0,
9369 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL);
9370 if (!btrfs_trans_handle_cachep)
9373 btrfs_path_cachep = kmem_cache_create("btrfs_path",
9374 sizeof(struct btrfs_path), 0,
9375 SLAB_MEM_SPREAD, NULL);
9376 if (!btrfs_path_cachep)
9379 btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
9380 sizeof(struct btrfs_free_space), 0,
9381 SLAB_MEM_SPREAD, NULL);
9382 if (!btrfs_free_space_cachep)
9387 btrfs_destroy_cachep();
9391 static int btrfs_getattr(const struct path *path, struct kstat *stat,
9392 u32 request_mask, unsigned int flags)
9395 struct inode *inode = d_inode(path->dentry);
9396 u32 blocksize = inode->i_sb->s_blocksize;
9397 u32 bi_flags = BTRFS_I(inode)->flags;
9399 stat->result_mask |= STATX_BTIME;
9400 stat->btime.tv_sec = BTRFS_I(inode)->i_otime.tv_sec;
9401 stat->btime.tv_nsec = BTRFS_I(inode)->i_otime.tv_nsec;
9402 if (bi_flags & BTRFS_INODE_APPEND)
9403 stat->attributes |= STATX_ATTR_APPEND;
9404 if (bi_flags & BTRFS_INODE_COMPRESS)
9405 stat->attributes |= STATX_ATTR_COMPRESSED;
9406 if (bi_flags & BTRFS_INODE_IMMUTABLE)
9407 stat->attributes |= STATX_ATTR_IMMUTABLE;
9408 if (bi_flags & BTRFS_INODE_NODUMP)
9409 stat->attributes |= STATX_ATTR_NODUMP;
9411 stat->attributes_mask |= (STATX_ATTR_APPEND |
9412 STATX_ATTR_COMPRESSED |
9413 STATX_ATTR_IMMUTABLE |
9416 generic_fillattr(inode, stat);
9417 stat->dev = BTRFS_I(inode)->root->anon_dev;
9419 spin_lock(&BTRFS_I(inode)->lock);
9420 delalloc_bytes = BTRFS_I(inode)->new_delalloc_bytes;
9421 spin_unlock(&BTRFS_I(inode)->lock);
9422 stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
9423 ALIGN(delalloc_bytes, blocksize)) >> 9;
9427 static int btrfs_rename_exchange(struct inode *old_dir,
9428 struct dentry *old_dentry,
9429 struct inode *new_dir,
9430 struct dentry *new_dentry)
9432 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9433 struct btrfs_trans_handle *trans;
9434 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9435 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9436 struct inode *new_inode = new_dentry->d_inode;
9437 struct inode *old_inode = old_dentry->d_inode;
9438 struct timespec64 ctime = current_time(old_inode);
9439 struct dentry *parent;
9440 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9441 u64 new_ino = btrfs_ino(BTRFS_I(new_inode));
9446 bool root_log_pinned = false;
9447 bool dest_log_pinned = false;
9449 /* we only allow rename subvolume link between subvolumes */
9450 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9453 /* close the race window with snapshot create/destroy ioctl */
9454 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9455 down_read(&fs_info->subvol_sem);
9456 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9457 down_read(&fs_info->subvol_sem);
9460 * We want to reserve the absolute worst case amount of items. So if
9461 * both inodes are subvols and we need to unlink them then that would
9462 * require 4 item modifications, but if they are both normal inodes it
9463 * would require 5 item modifications, so we'll assume their normal
9464 * inodes. So 5 * 2 is 10, plus 2 for the new links, so 12 total items
9465 * should cover the worst case number of items we'll modify.
9467 trans = btrfs_start_transaction(root, 12);
9468 if (IS_ERR(trans)) {
9469 ret = PTR_ERR(trans);
9474 * We need to find a free sequence number both in the source and
9475 * in the destination directory for the exchange.
9477 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &old_idx);
9480 ret = btrfs_set_inode_index(BTRFS_I(old_dir), &new_idx);
9484 BTRFS_I(old_inode)->dir_index = 0ULL;
9485 BTRFS_I(new_inode)->dir_index = 0ULL;
9487 /* Reference for the source. */
9488 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9489 /* force full log commit if subvolume involved. */
9490 btrfs_set_log_full_commit(fs_info, trans);
9492 btrfs_pin_log_trans(root);
9493 root_log_pinned = true;
9494 ret = btrfs_insert_inode_ref(trans, dest,
9495 new_dentry->d_name.name,
9496 new_dentry->d_name.len,
9498 btrfs_ino(BTRFS_I(new_dir)),
9504 /* And now for the dest. */
9505 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9506 /* force full log commit if subvolume involved. */
9507 btrfs_set_log_full_commit(fs_info, trans);
9509 btrfs_pin_log_trans(dest);
9510 dest_log_pinned = true;
9511 ret = btrfs_insert_inode_ref(trans, root,
9512 old_dentry->d_name.name,
9513 old_dentry->d_name.len,
9515 btrfs_ino(BTRFS_I(old_dir)),
9521 /* Update inode version and ctime/mtime. */
9522 inode_inc_iversion(old_dir);
9523 inode_inc_iversion(new_dir);
9524 inode_inc_iversion(old_inode);
9525 inode_inc_iversion(new_inode);
9526 old_dir->i_ctime = old_dir->i_mtime = ctime;
9527 new_dir->i_ctime = new_dir->i_mtime = ctime;
9528 old_inode->i_ctime = ctime;
9529 new_inode->i_ctime = ctime;
9531 if (old_dentry->d_parent != new_dentry->d_parent) {
9532 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9533 BTRFS_I(old_inode), 1);
9534 btrfs_record_unlink_dir(trans, BTRFS_I(new_dir),
9535 BTRFS_I(new_inode), 1);
9538 /* src is a subvolume */
9539 if (old_ino == BTRFS_FIRST_FREE_OBJECTID) {
9540 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9541 ret = btrfs_unlink_subvol(trans, root, old_dir,
9543 old_dentry->d_name.name,
9544 old_dentry->d_name.len);
9545 } else { /* src is an inode */
9546 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9547 BTRFS_I(old_dentry->d_inode),
9548 old_dentry->d_name.name,
9549 old_dentry->d_name.len);
9551 ret = btrfs_update_inode(trans, root, old_inode);
9554 btrfs_abort_transaction(trans, ret);
9558 /* dest is a subvolume */
9559 if (new_ino == BTRFS_FIRST_FREE_OBJECTID) {
9560 root_objectid = BTRFS_I(new_inode)->root->root_key.objectid;
9561 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9563 new_dentry->d_name.name,
9564 new_dentry->d_name.len);
9565 } else { /* dest is an inode */
9566 ret = __btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9567 BTRFS_I(new_dentry->d_inode),
9568 new_dentry->d_name.name,
9569 new_dentry->d_name.len);
9571 ret = btrfs_update_inode(trans, dest, new_inode);
9574 btrfs_abort_transaction(trans, ret);
9578 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9579 new_dentry->d_name.name,
9580 new_dentry->d_name.len, 0, old_idx);
9582 btrfs_abort_transaction(trans, ret);
9586 ret = btrfs_add_link(trans, BTRFS_I(old_dir), BTRFS_I(new_inode),
9587 old_dentry->d_name.name,
9588 old_dentry->d_name.len, 0, new_idx);
9590 btrfs_abort_transaction(trans, ret);
9594 if (old_inode->i_nlink == 1)
9595 BTRFS_I(old_inode)->dir_index = old_idx;
9596 if (new_inode->i_nlink == 1)
9597 BTRFS_I(new_inode)->dir_index = new_idx;
9599 if (root_log_pinned) {
9600 parent = new_dentry->d_parent;
9601 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9603 btrfs_end_log_trans(root);
9604 root_log_pinned = false;
9606 if (dest_log_pinned) {
9607 parent = old_dentry->d_parent;
9608 btrfs_log_new_name(trans, BTRFS_I(new_inode), BTRFS_I(new_dir),
9610 btrfs_end_log_trans(dest);
9611 dest_log_pinned = false;
9615 * If we have pinned a log and an error happened, we unpin tasks
9616 * trying to sync the log and force them to fallback to a transaction
9617 * commit if the log currently contains any of the inodes involved in
9618 * this rename operation (to ensure we do not persist a log with an
9619 * inconsistent state for any of these inodes or leading to any
9620 * inconsistencies when replayed). If the transaction was aborted, the
9621 * abortion reason is propagated to userspace when attempting to commit
9622 * the transaction. If the log does not contain any of these inodes, we
9623 * allow the tasks to sync it.
9625 if (ret && (root_log_pinned || dest_log_pinned)) {
9626 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9627 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9628 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9630 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9631 btrfs_set_log_full_commit(fs_info, trans);
9633 if (root_log_pinned) {
9634 btrfs_end_log_trans(root);
9635 root_log_pinned = false;
9637 if (dest_log_pinned) {
9638 btrfs_end_log_trans(dest);
9639 dest_log_pinned = false;
9642 ret = btrfs_end_transaction(trans);
9644 if (new_ino == BTRFS_FIRST_FREE_OBJECTID)
9645 up_read(&fs_info->subvol_sem);
9646 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9647 up_read(&fs_info->subvol_sem);
9652 static int btrfs_whiteout_for_rename(struct btrfs_trans_handle *trans,
9653 struct btrfs_root *root,
9655 struct dentry *dentry)
9658 struct inode *inode;
9662 ret = btrfs_find_free_ino(root, &objectid);
9666 inode = btrfs_new_inode(trans, root, dir,
9667 dentry->d_name.name,
9669 btrfs_ino(BTRFS_I(dir)),
9671 S_IFCHR | WHITEOUT_MODE,
9674 if (IS_ERR(inode)) {
9675 ret = PTR_ERR(inode);
9679 inode->i_op = &btrfs_special_inode_operations;
9680 init_special_inode(inode, inode->i_mode,
9683 ret = btrfs_init_inode_security(trans, inode, dir,
9688 ret = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
9689 BTRFS_I(inode), 0, index);
9693 ret = btrfs_update_inode(trans, root, inode);
9695 unlock_new_inode(inode);
9697 inode_dec_link_count(inode);
9703 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
9704 struct inode *new_dir, struct dentry *new_dentry,
9707 struct btrfs_fs_info *fs_info = btrfs_sb(old_dir->i_sb);
9708 struct btrfs_trans_handle *trans;
9709 unsigned int trans_num_items;
9710 struct btrfs_root *root = BTRFS_I(old_dir)->root;
9711 struct btrfs_root *dest = BTRFS_I(new_dir)->root;
9712 struct inode *new_inode = d_inode(new_dentry);
9713 struct inode *old_inode = d_inode(old_dentry);
9717 u64 old_ino = btrfs_ino(BTRFS_I(old_inode));
9718 bool log_pinned = false;
9720 if (btrfs_ino(BTRFS_I(new_dir)) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
9723 /* we only allow rename subvolume link between subvolumes */
9724 if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
9727 if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
9728 (new_inode && btrfs_ino(BTRFS_I(new_inode)) == BTRFS_FIRST_FREE_OBJECTID))
9731 if (S_ISDIR(old_inode->i_mode) && new_inode &&
9732 new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
9736 /* check for collisions, even if the name isn't there */
9737 ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
9738 new_dentry->d_name.name,
9739 new_dentry->d_name.len);
9742 if (ret == -EEXIST) {
9744 * eexist without a new_inode */
9745 if (WARN_ON(!new_inode)) {
9749 /* maybe -EOVERFLOW */
9756 * we're using rename to replace one file with another. Start IO on it
9757 * now so we don't add too much work to the end of the transaction
9759 if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
9760 filemap_flush(old_inode->i_mapping);
9762 /* close the racy window with snapshot create/destroy ioctl */
9763 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9764 down_read(&fs_info->subvol_sem);
9766 * We want to reserve the absolute worst case amount of items. So if
9767 * both inodes are subvols and we need to unlink them then that would
9768 * require 4 item modifications, but if they are both normal inodes it
9769 * would require 5 item modifications, so we'll assume they are normal
9770 * inodes. So 5 * 2 is 10, plus 1 for the new link, so 11 total items
9771 * should cover the worst case number of items we'll modify.
9772 * If our rename has the whiteout flag, we need more 5 units for the
9773 * new inode (1 inode item, 1 inode ref, 2 dir items and 1 xattr item
9774 * when selinux is enabled).
9776 trans_num_items = 11;
9777 if (flags & RENAME_WHITEOUT)
9778 trans_num_items += 5;
9779 trans = btrfs_start_transaction(root, trans_num_items);
9780 if (IS_ERR(trans)) {
9781 ret = PTR_ERR(trans);
9786 btrfs_record_root_in_trans(trans, dest);
9788 ret = btrfs_set_inode_index(BTRFS_I(new_dir), &index);
9792 BTRFS_I(old_inode)->dir_index = 0ULL;
9793 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9794 /* force full log commit if subvolume involved. */
9795 btrfs_set_log_full_commit(fs_info, trans);
9797 btrfs_pin_log_trans(root);
9799 ret = btrfs_insert_inode_ref(trans, dest,
9800 new_dentry->d_name.name,
9801 new_dentry->d_name.len,
9803 btrfs_ino(BTRFS_I(new_dir)), index);
9808 inode_inc_iversion(old_dir);
9809 inode_inc_iversion(new_dir);
9810 inode_inc_iversion(old_inode);
9811 old_dir->i_ctime = old_dir->i_mtime =
9812 new_dir->i_ctime = new_dir->i_mtime =
9813 old_inode->i_ctime = current_time(old_dir);
9815 if (old_dentry->d_parent != new_dentry->d_parent)
9816 btrfs_record_unlink_dir(trans, BTRFS_I(old_dir),
9817 BTRFS_I(old_inode), 1);
9819 if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
9820 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
9821 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
9822 old_dentry->d_name.name,
9823 old_dentry->d_name.len);
9825 ret = __btrfs_unlink_inode(trans, root, BTRFS_I(old_dir),
9826 BTRFS_I(d_inode(old_dentry)),
9827 old_dentry->d_name.name,
9828 old_dentry->d_name.len);
9830 ret = btrfs_update_inode(trans, root, old_inode);
9833 btrfs_abort_transaction(trans, ret);
9838 inode_inc_iversion(new_inode);
9839 new_inode->i_ctime = current_time(new_inode);
9840 if (unlikely(btrfs_ino(BTRFS_I(new_inode)) ==
9841 BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
9842 root_objectid = BTRFS_I(new_inode)->location.objectid;
9843 ret = btrfs_unlink_subvol(trans, dest, new_dir,
9845 new_dentry->d_name.name,
9846 new_dentry->d_name.len);
9847 BUG_ON(new_inode->i_nlink == 0);
9849 ret = btrfs_unlink_inode(trans, dest, BTRFS_I(new_dir),
9850 BTRFS_I(d_inode(new_dentry)),
9851 new_dentry->d_name.name,
9852 new_dentry->d_name.len);
9854 if (!ret && new_inode->i_nlink == 0)
9855 ret = btrfs_orphan_add(trans,
9856 BTRFS_I(d_inode(new_dentry)));
9858 btrfs_abort_transaction(trans, ret);
9863 ret = btrfs_add_link(trans, BTRFS_I(new_dir), BTRFS_I(old_inode),
9864 new_dentry->d_name.name,
9865 new_dentry->d_name.len, 0, index);
9867 btrfs_abort_transaction(trans, ret);
9871 if (old_inode->i_nlink == 1)
9872 BTRFS_I(old_inode)->dir_index = index;
9875 struct dentry *parent = new_dentry->d_parent;
9877 btrfs_log_new_name(trans, BTRFS_I(old_inode), BTRFS_I(old_dir),
9879 btrfs_end_log_trans(root);
9883 if (flags & RENAME_WHITEOUT) {
9884 ret = btrfs_whiteout_for_rename(trans, root, old_dir,
9888 btrfs_abort_transaction(trans, ret);
9894 * If we have pinned the log and an error happened, we unpin tasks
9895 * trying to sync the log and force them to fallback to a transaction
9896 * commit if the log currently contains any of the inodes involved in
9897 * this rename operation (to ensure we do not persist a log with an
9898 * inconsistent state for any of these inodes or leading to any
9899 * inconsistencies when replayed). If the transaction was aborted, the
9900 * abortion reason is propagated to userspace when attempting to commit
9901 * the transaction. If the log does not contain any of these inodes, we
9902 * allow the tasks to sync it.
9904 if (ret && log_pinned) {
9905 if (btrfs_inode_in_log(BTRFS_I(old_dir), fs_info->generation) ||
9906 btrfs_inode_in_log(BTRFS_I(new_dir), fs_info->generation) ||
9907 btrfs_inode_in_log(BTRFS_I(old_inode), fs_info->generation) ||
9909 btrfs_inode_in_log(BTRFS_I(new_inode), fs_info->generation)))
9910 btrfs_set_log_full_commit(fs_info, trans);
9912 btrfs_end_log_trans(root);
9915 btrfs_end_transaction(trans);
9917 if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9918 up_read(&fs_info->subvol_sem);
9923 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9924 struct inode *new_dir, struct dentry *new_dentry,
9927 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
9930 if (flags & RENAME_EXCHANGE)
9931 return btrfs_rename_exchange(old_dir, old_dentry, new_dir,
9934 return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry, flags);
9937 struct btrfs_delalloc_work {
9938 struct inode *inode;
9939 struct completion completion;
9940 struct list_head list;
9941 struct btrfs_work work;
9944 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9946 struct btrfs_delalloc_work *delalloc_work;
9947 struct inode *inode;
9949 delalloc_work = container_of(work, struct btrfs_delalloc_work,
9951 inode = delalloc_work->inode;
9952 filemap_flush(inode->i_mapping);
9953 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9954 &BTRFS_I(inode)->runtime_flags))
9955 filemap_flush(inode->i_mapping);
9958 complete(&delalloc_work->completion);
9961 static struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode)
9963 struct btrfs_delalloc_work *work;
9965 work = kmalloc(sizeof(*work), GFP_NOFS);
9969 init_completion(&work->completion);
9970 INIT_LIST_HEAD(&work->list);
9971 work->inode = inode;
9972 WARN_ON_ONCE(!inode);
9973 btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9974 btrfs_run_delalloc_work, NULL, NULL);
9980 * some fairly slow code that needs optimization. This walks the list
9981 * of all the inodes with pending delalloc and forces them to disk.
9983 static int start_delalloc_inodes(struct btrfs_root *root, int nr)
9985 struct btrfs_inode *binode;
9986 struct inode *inode;
9987 struct btrfs_delalloc_work *work, *next;
9988 struct list_head works;
9989 struct list_head splice;
9992 INIT_LIST_HEAD(&works);
9993 INIT_LIST_HEAD(&splice);
9995 mutex_lock(&root->delalloc_mutex);
9996 spin_lock(&root->delalloc_lock);
9997 list_splice_init(&root->delalloc_inodes, &splice);
9998 while (!list_empty(&splice)) {
9999 binode = list_entry(splice.next, struct btrfs_inode,
10002 list_move_tail(&binode->delalloc_inodes,
10003 &root->delalloc_inodes);
10004 inode = igrab(&binode->vfs_inode);
10006 cond_resched_lock(&root->delalloc_lock);
10009 spin_unlock(&root->delalloc_lock);
10011 work = btrfs_alloc_delalloc_work(inode);
10017 list_add_tail(&work->list, &works);
10018 btrfs_queue_work(root->fs_info->flush_workers,
10021 if (nr != -1 && ret >= nr)
10024 spin_lock(&root->delalloc_lock);
10026 spin_unlock(&root->delalloc_lock);
10029 list_for_each_entry_safe(work, next, &works, list) {
10030 list_del_init(&work->list);
10031 wait_for_completion(&work->completion);
10035 if (!list_empty(&splice)) {
10036 spin_lock(&root->delalloc_lock);
10037 list_splice_tail(&splice, &root->delalloc_inodes);
10038 spin_unlock(&root->delalloc_lock);
10040 mutex_unlock(&root->delalloc_mutex);
10044 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
10046 struct btrfs_fs_info *fs_info = root->fs_info;
10049 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10052 ret = start_delalloc_inodes(root, -1);
10058 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int nr)
10060 struct btrfs_root *root;
10061 struct list_head splice;
10064 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
10067 INIT_LIST_HEAD(&splice);
10069 mutex_lock(&fs_info->delalloc_root_mutex);
10070 spin_lock(&fs_info->delalloc_root_lock);
10071 list_splice_init(&fs_info->delalloc_roots, &splice);
10072 while (!list_empty(&splice) && nr) {
10073 root = list_first_entry(&splice, struct btrfs_root,
10075 root = btrfs_grab_fs_root(root);
10077 list_move_tail(&root->delalloc_root,
10078 &fs_info->delalloc_roots);
10079 spin_unlock(&fs_info->delalloc_root_lock);
10081 ret = start_delalloc_inodes(root, nr);
10082 btrfs_put_fs_root(root);
10090 spin_lock(&fs_info->delalloc_root_lock);
10092 spin_unlock(&fs_info->delalloc_root_lock);
10096 if (!list_empty(&splice)) {
10097 spin_lock(&fs_info->delalloc_root_lock);
10098 list_splice_tail(&splice, &fs_info->delalloc_roots);
10099 spin_unlock(&fs_info->delalloc_root_lock);
10101 mutex_unlock(&fs_info->delalloc_root_mutex);
10105 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
10106 const char *symname)
10108 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10109 struct btrfs_trans_handle *trans;
10110 struct btrfs_root *root = BTRFS_I(dir)->root;
10111 struct btrfs_path *path;
10112 struct btrfs_key key;
10113 struct inode *inode = NULL;
10115 int drop_inode = 0;
10121 struct btrfs_file_extent_item *ei;
10122 struct extent_buffer *leaf;
10124 name_len = strlen(symname);
10125 if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(fs_info))
10126 return -ENAMETOOLONG;
10129 * 2 items for inode item and ref
10130 * 2 items for dir items
10131 * 1 item for updating parent inode item
10132 * 1 item for the inline extent item
10133 * 1 item for xattr if selinux is on
10135 trans = btrfs_start_transaction(root, 7);
10137 return PTR_ERR(trans);
10139 err = btrfs_find_free_ino(root, &objectid);
10143 inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
10144 dentry->d_name.len, btrfs_ino(BTRFS_I(dir)),
10145 objectid, S_IFLNK|S_IRWXUGO, &index);
10146 if (IS_ERR(inode)) {
10147 err = PTR_ERR(inode);
10152 * If the active LSM wants to access the inode during
10153 * d_instantiate it needs these. Smack checks to see
10154 * if the filesystem supports xattrs by looking at the
10157 inode->i_fop = &btrfs_file_operations;
10158 inode->i_op = &btrfs_file_inode_operations;
10159 inode->i_mapping->a_ops = &btrfs_aops;
10160 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10162 err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
10164 goto out_unlock_inode;
10166 path = btrfs_alloc_path();
10169 goto out_unlock_inode;
10171 key.objectid = btrfs_ino(BTRFS_I(inode));
10173 key.type = BTRFS_EXTENT_DATA_KEY;
10174 datasize = btrfs_file_extent_calc_inline_size(name_len);
10175 err = btrfs_insert_empty_item(trans, root, path, &key,
10178 btrfs_free_path(path);
10179 goto out_unlock_inode;
10181 leaf = path->nodes[0];
10182 ei = btrfs_item_ptr(leaf, path->slots[0],
10183 struct btrfs_file_extent_item);
10184 btrfs_set_file_extent_generation(leaf, ei, trans->transid);
10185 btrfs_set_file_extent_type(leaf, ei,
10186 BTRFS_FILE_EXTENT_INLINE);
10187 btrfs_set_file_extent_encryption(leaf, ei, 0);
10188 btrfs_set_file_extent_compression(leaf, ei, 0);
10189 btrfs_set_file_extent_other_encoding(leaf, ei, 0);
10190 btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
10192 ptr = btrfs_file_extent_inline_start(ei);
10193 write_extent_buffer(leaf, symname, ptr, name_len);
10194 btrfs_mark_buffer_dirty(leaf);
10195 btrfs_free_path(path);
10197 inode->i_op = &btrfs_symlink_inode_operations;
10198 inode_nohighmem(inode);
10199 inode->i_mapping->a_ops = &btrfs_symlink_aops;
10200 inode_set_bytes(inode, name_len);
10201 btrfs_i_size_write(BTRFS_I(inode), name_len);
10202 err = btrfs_update_inode(trans, root, inode);
10204 * Last step, add directory indexes for our symlink inode. This is the
10205 * last step to avoid extra cleanup of these indexes if an error happens
10209 err = btrfs_add_nondir(trans, BTRFS_I(dir), dentry,
10210 BTRFS_I(inode), 0, index);
10213 goto out_unlock_inode;
10216 d_instantiate_new(dentry, inode);
10219 btrfs_end_transaction(trans);
10221 inode_dec_link_count(inode);
10224 btrfs_btree_balance_dirty(fs_info);
10229 unlock_new_inode(inode);
10233 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
10234 u64 start, u64 num_bytes, u64 min_size,
10235 loff_t actual_len, u64 *alloc_hint,
10236 struct btrfs_trans_handle *trans)
10238 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
10239 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
10240 struct extent_map *em;
10241 struct btrfs_root *root = BTRFS_I(inode)->root;
10242 struct btrfs_key ins;
10243 u64 cur_offset = start;
10246 u64 last_alloc = (u64)-1;
10248 bool own_trans = true;
10249 u64 end = start + num_bytes - 1;
10253 while (num_bytes > 0) {
10255 trans = btrfs_start_transaction(root, 3);
10256 if (IS_ERR(trans)) {
10257 ret = PTR_ERR(trans);
10262 cur_bytes = min_t(u64, num_bytes, SZ_256M);
10263 cur_bytes = max(cur_bytes, min_size);
10265 * If we are severely fragmented we could end up with really
10266 * small allocations, so if the allocator is returning small
10267 * chunks lets make its job easier by only searching for those
10270 cur_bytes = min(cur_bytes, last_alloc);
10271 ret = btrfs_reserve_extent(root, cur_bytes, cur_bytes,
10272 min_size, 0, *alloc_hint, &ins, 1, 0);
10275 btrfs_end_transaction(trans);
10278 btrfs_dec_block_group_reservations(fs_info, ins.objectid);
10280 last_alloc = ins.offset;
10281 ret = insert_reserved_file_extent(trans, inode,
10282 cur_offset, ins.objectid,
10283 ins.offset, ins.offset,
10284 ins.offset, 0, 0, 0,
10285 BTRFS_FILE_EXTENT_PREALLOC);
10287 btrfs_free_reserved_extent(fs_info, ins.objectid,
10289 btrfs_abort_transaction(trans, ret);
10291 btrfs_end_transaction(trans);
10295 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10296 cur_offset + ins.offset -1, 0);
10298 em = alloc_extent_map();
10300 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
10301 &BTRFS_I(inode)->runtime_flags);
10305 em->start = cur_offset;
10306 em->orig_start = cur_offset;
10307 em->len = ins.offset;
10308 em->block_start = ins.objectid;
10309 em->block_len = ins.offset;
10310 em->orig_block_len = ins.offset;
10311 em->ram_bytes = ins.offset;
10312 em->bdev = fs_info->fs_devices->latest_bdev;
10313 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
10314 em->generation = trans->transid;
10317 write_lock(&em_tree->lock);
10318 ret = add_extent_mapping(em_tree, em, 1);
10319 write_unlock(&em_tree->lock);
10320 if (ret != -EEXIST)
10322 btrfs_drop_extent_cache(BTRFS_I(inode), cur_offset,
10323 cur_offset + ins.offset - 1,
10326 free_extent_map(em);
10328 num_bytes -= ins.offset;
10329 cur_offset += ins.offset;
10330 *alloc_hint = ins.objectid + ins.offset;
10332 inode_inc_iversion(inode);
10333 inode->i_ctime = current_time(inode);
10334 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
10335 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
10336 (actual_len > inode->i_size) &&
10337 (cur_offset > inode->i_size)) {
10338 if (cur_offset > actual_len)
10339 i_size = actual_len;
10341 i_size = cur_offset;
10342 i_size_write(inode, i_size);
10343 btrfs_ordered_update_i_size(inode, i_size, NULL);
10346 ret = btrfs_update_inode(trans, root, inode);
10349 btrfs_abort_transaction(trans, ret);
10351 btrfs_end_transaction(trans);
10356 btrfs_end_transaction(trans);
10358 if (cur_offset < end)
10359 btrfs_free_reserved_data_space(inode, NULL, cur_offset,
10360 end - cur_offset + 1);
10364 int btrfs_prealloc_file_range(struct inode *inode, int mode,
10365 u64 start, u64 num_bytes, u64 min_size,
10366 loff_t actual_len, u64 *alloc_hint)
10368 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10369 min_size, actual_len, alloc_hint,
10373 int btrfs_prealloc_file_range_trans(struct inode *inode,
10374 struct btrfs_trans_handle *trans, int mode,
10375 u64 start, u64 num_bytes, u64 min_size,
10376 loff_t actual_len, u64 *alloc_hint)
10378 return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
10379 min_size, actual_len, alloc_hint, trans);
10382 static int btrfs_set_page_dirty(struct page *page)
10384 return __set_page_dirty_nobuffers(page);
10387 static int btrfs_permission(struct inode *inode, int mask)
10389 struct btrfs_root *root = BTRFS_I(inode)->root;
10390 umode_t mode = inode->i_mode;
10392 if (mask & MAY_WRITE &&
10393 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
10394 if (btrfs_root_readonly(root))
10396 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
10399 return generic_permission(inode, mask);
10402 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
10404 struct btrfs_fs_info *fs_info = btrfs_sb(dir->i_sb);
10405 struct btrfs_trans_handle *trans;
10406 struct btrfs_root *root = BTRFS_I(dir)->root;
10407 struct inode *inode = NULL;
10413 * 5 units required for adding orphan entry
10415 trans = btrfs_start_transaction(root, 5);
10417 return PTR_ERR(trans);
10419 ret = btrfs_find_free_ino(root, &objectid);
10423 inode = btrfs_new_inode(trans, root, dir, NULL, 0,
10424 btrfs_ino(BTRFS_I(dir)), objectid, mode, &index);
10425 if (IS_ERR(inode)) {
10426 ret = PTR_ERR(inode);
10431 inode->i_fop = &btrfs_file_operations;
10432 inode->i_op = &btrfs_file_inode_operations;
10434 inode->i_mapping->a_ops = &btrfs_aops;
10435 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
10437 ret = btrfs_init_inode_security(trans, inode, dir, NULL);
10441 ret = btrfs_update_inode(trans, root, inode);
10444 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10449 * We set number of links to 0 in btrfs_new_inode(), and here we set
10450 * it to 1 because d_tmpfile() will issue a warning if the count is 0,
10453 * d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
10455 set_nlink(inode, 1);
10456 unlock_new_inode(inode);
10457 d_tmpfile(dentry, inode);
10458 mark_inode_dirty(inode);
10461 btrfs_end_transaction(trans);
10464 btrfs_btree_balance_dirty(fs_info);
10468 unlock_new_inode(inode);
10473 __attribute__((const))
10474 static int btrfs_readpage_io_failed_hook(struct page *page, int failed_mirror)
10479 static struct btrfs_fs_info *iotree_fs_info(void *private_data)
10481 struct inode *inode = private_data;
10482 return btrfs_sb(inode->i_sb);
10485 static void btrfs_check_extent_io_range(void *private_data, const char *caller,
10486 u64 start, u64 end)
10488 struct inode *inode = private_data;
10491 isize = i_size_read(inode);
10492 if (end >= PAGE_SIZE && (end % 2) == 0 && end != isize - 1) {
10493 btrfs_debug_rl(BTRFS_I(inode)->root->fs_info,
10494 "%s: ino %llu isize %llu odd range [%llu,%llu]",
10495 caller, btrfs_ino(BTRFS_I(inode)), isize, start, end);
10499 void btrfs_set_range_writeback(void *private_data, u64 start, u64 end)
10501 struct inode *inode = private_data;
10502 unsigned long index = start >> PAGE_SHIFT;
10503 unsigned long end_index = end >> PAGE_SHIFT;
10506 while (index <= end_index) {
10507 page = find_get_page(inode->i_mapping, index);
10508 ASSERT(page); /* Pages should be in the extent_io_tree */
10509 set_page_writeback(page);
10515 static const struct inode_operations btrfs_dir_inode_operations = {
10516 .getattr = btrfs_getattr,
10517 .lookup = btrfs_lookup,
10518 .create = btrfs_create,
10519 .unlink = btrfs_unlink,
10520 .link = btrfs_link,
10521 .mkdir = btrfs_mkdir,
10522 .rmdir = btrfs_rmdir,
10523 .rename = btrfs_rename2,
10524 .symlink = btrfs_symlink,
10525 .setattr = btrfs_setattr,
10526 .mknod = btrfs_mknod,
10527 .listxattr = btrfs_listxattr,
10528 .permission = btrfs_permission,
10529 .get_acl = btrfs_get_acl,
10530 .set_acl = btrfs_set_acl,
10531 .update_time = btrfs_update_time,
10532 .tmpfile = btrfs_tmpfile,
10534 static const struct inode_operations btrfs_dir_ro_inode_operations = {
10535 .lookup = btrfs_lookup,
10536 .permission = btrfs_permission,
10537 .update_time = btrfs_update_time,
10540 static const struct file_operations btrfs_dir_file_operations = {
10541 .llseek = generic_file_llseek,
10542 .read = generic_read_dir,
10543 .iterate_shared = btrfs_real_readdir,
10544 .open = btrfs_opendir,
10545 .unlocked_ioctl = btrfs_ioctl,
10546 #ifdef CONFIG_COMPAT
10547 .compat_ioctl = btrfs_compat_ioctl,
10549 .release = btrfs_release_file,
10550 .fsync = btrfs_sync_file,
10553 static const struct extent_io_ops btrfs_extent_io_ops = {
10554 /* mandatory callbacks */
10555 .submit_bio_hook = btrfs_submit_bio_hook,
10556 .readpage_end_io_hook = btrfs_readpage_end_io_hook,
10557 .merge_bio_hook = btrfs_merge_bio_hook,
10558 .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
10559 .tree_fs_info = iotree_fs_info,
10560 .set_range_writeback = btrfs_set_range_writeback,
10562 /* optional callbacks */
10563 .fill_delalloc = run_delalloc_range,
10564 .writepage_end_io_hook = btrfs_writepage_end_io_hook,
10565 .writepage_start_hook = btrfs_writepage_start_hook,
10566 .set_bit_hook = btrfs_set_bit_hook,
10567 .clear_bit_hook = btrfs_clear_bit_hook,
10568 .merge_extent_hook = btrfs_merge_extent_hook,
10569 .split_extent_hook = btrfs_split_extent_hook,
10570 .check_extent_io_range = btrfs_check_extent_io_range,
10574 * btrfs doesn't support the bmap operation because swapfiles
10575 * use bmap to make a mapping of extents in the file. They assume
10576 * these extents won't change over the life of the file and they
10577 * use the bmap result to do IO directly to the drive.
10579 * the btrfs bmap call would return logical addresses that aren't
10580 * suitable for IO and they also will change frequently as COW
10581 * operations happen. So, swapfile + btrfs == corruption.
10583 * For now we're avoiding this by dropping bmap.
10585 static const struct address_space_operations btrfs_aops = {
10586 .readpage = btrfs_readpage,
10587 .writepage = btrfs_writepage,
10588 .writepages = btrfs_writepages,
10589 .readpages = btrfs_readpages,
10590 .direct_IO = btrfs_direct_IO,
10591 .invalidatepage = btrfs_invalidatepage,
10592 .releasepage = btrfs_releasepage,
10593 .set_page_dirty = btrfs_set_page_dirty,
10594 .error_remove_page = generic_error_remove_page,
10597 static const struct address_space_operations btrfs_symlink_aops = {
10598 .readpage = btrfs_readpage,
10599 .writepage = btrfs_writepage,
10600 .invalidatepage = btrfs_invalidatepage,
10601 .releasepage = btrfs_releasepage,
10604 static const struct inode_operations btrfs_file_inode_operations = {
10605 .getattr = btrfs_getattr,
10606 .setattr = btrfs_setattr,
10607 .listxattr = btrfs_listxattr,
10608 .permission = btrfs_permission,
10609 .fiemap = btrfs_fiemap,
10610 .get_acl = btrfs_get_acl,
10611 .set_acl = btrfs_set_acl,
10612 .update_time = btrfs_update_time,
10614 static const struct inode_operations btrfs_special_inode_operations = {
10615 .getattr = btrfs_getattr,
10616 .setattr = btrfs_setattr,
10617 .permission = btrfs_permission,
10618 .listxattr = btrfs_listxattr,
10619 .get_acl = btrfs_get_acl,
10620 .set_acl = btrfs_set_acl,
10621 .update_time = btrfs_update_time,
10623 static const struct inode_operations btrfs_symlink_inode_operations = {
10624 .get_link = page_get_link,
10625 .getattr = btrfs_getattr,
10626 .setattr = btrfs_setattr,
10627 .permission = btrfs_permission,
10628 .listxattr = btrfs_listxattr,
10629 .update_time = btrfs_update_time,
10632 const struct dentry_operations btrfs_dentry_operations = {
10633 .d_delete = btrfs_dentry_delete,